diff options
108 files changed, 7108 insertions, 4493 deletions
diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml index 65a0a05b..39202098 100644 --- a/.gitlab-ci.yml +++ b/.gitlab-ci.yml @@ -134,21 +134,11 @@ docker_fedora-34-amd64: IMG_NAME: "fedora-34-amd64" <<: *docker_build -docker_centos-7-amd64: - variables: - IMG_NAME: "centos-7-amd64" - <<: *docker_build - docker_centos-8-amd64: variables: IMG_NAME: "centos-8-amd64" <<: *docker_build -docker_ubuntu-14_04-amd64: - variables: - IMG_NAME: "ubuntu-14.04-amd64" - <<: *docker_build - docker_ubuntu-16_04-amd64: variables: IMG_NAME: "ubuntu-16.04-amd64" @@ -312,18 +302,10 @@ build-fedora-34-amd64: <<: *build-linux image: registry.labs.nic.cz/labs/bird:fedora-33-amd64 -build-centos-7-amd64: - <<: *build-linux - image: registry.labs.nic.cz/labs/bird:centos-7-amd64 - build-centos-8-amd64: <<: *build-linux image: registry.labs.nic.cz/labs/bird:centos-8-amd64 -build-ubuntu-14_04-amd64: - <<: *build-linux - image: registry.labs.nic.cz/labs/bird:ubuntu-14.04-amd64 - build-ubuntu-16_04-amd64: <<: *build-linux image: registry.labs.nic.cz/labs/bird:ubuntu-16.04-amd64 @@ -468,13 +450,6 @@ pkg-fedora-34-amd64: needs: [build-fedora-34-amd64] image: registry.labs.nic.cz/labs/bird:fedora-34-amd64 -pkg-centos-7-amd64: - <<: *pkg-rpm-wa - variables: - LC_ALL: en_US.UTF-8 - needs: [build-centos-7-amd64] - image: registry.labs.nic.cz/labs/bird:centos-7-amd64 - pkg-centos-8-amd64: <<: *pkg-rpm-wa needs: [build-centos-8-amd64] @@ -27,9 +27,9 @@ Requirements For compiling BIRD you need these programs and libraries: - - GNU C Compiler (or LLVM Clang) + - GNU C Compiler (or LLVM Clang) capable of compiling C11 code - GNU Make - - GNU Bison + - GNU Bison (at least 3.0) - GNU M4 - Flex @@ -1,5 +1,6 @@ dnl ** Additional Autoconf tests for BIRD configure script dnl ** (c) 1999 Martin Mares <mj@ucw.cz> +dnl ** (c) 2021 Maria Matejka <mq@jmq.cz> AC_DEFUN([BIRD_CHECK_THREAD_LOCAL], [ @@ -9,14 +10,23 @@ AC_DEFUN([BIRD_CHECK_THREAD_LOCAL], AC_COMPILE_IFELSE([ AC_LANG_PROGRAM( [ - _Thread_local static int x = 42; + static _Thread_local int x = 42; ], [] ) ], [bird_cv_thread_local=yes], + [AC_COMPILE_IFELSE([ + AC_LANG_PROGRAM( + [ + static __thread int x = 42; + ], + [] + ) + ], + [bird_cv_thread_local=__thread], [bird_cv_thread_local=no] - ) + )]) ) ]) diff --git a/bird-gdb.py b/bird-gdb.py index 3cf65a9c..fe7827fe 100644 --- a/bird-gdb.py +++ b/bird-gdb.py @@ -4,9 +4,10 @@ class BIRDPrinter: @classmethod def lookup(cls, val): - if val.type.code != cls.typeCode: + t = val.type.strip_typedefs() + if t.code != cls.typeCode: return None - if val.type.tag != cls.typeTag: + if t.tag != cls.typeTag: return None return cls(val) @@ -123,7 +124,7 @@ class BIRDFLinePrinter(BIRDPrinter): "n": n, "code": str(self.val['items'][n]['fi_code']), } if n % 8 == 0 else str(self.val['items'][n]['fi_code']) for n in range(cnt)])) - + class BIRDFExecStackPrinter(BIRDPrinter): "Print BIRD's struct f_exec_stack" @@ -141,6 +142,323 @@ class BIRDFExecStackPrinter(BIRDPrinter): "n": n } for n in range(cnt-1, -1, -1) ]) + +class BIRD: + def skip_back(t, i, v): + if isinstance(t, str): + t = gdb.lookup_type(t) + elif isinstance(t, gdb.Value): + t = gdb.lookup_type(t.string()) + elif not isinstance(t, gdb.Type): + raise Exception(f"First argument of skip_back(t, i, v) must be a type, got {type(t)}") + + t = t.strip_typedefs() + nullptr = gdb.Value(0).cast(t.pointer()) + + if isinstance(i, gdb.Value): + i = i.string() + elif not isinstance(i, str): + raise Exception(f"Second argument of skip_back(t, i, v) must be a item name, got {type(i)}") + + if not isinstance(v, gdb.Value): + raise Exception(f"Third argument of skip_back(t, i, v) must be a value, got {type(v)}") + if v.type.code != gdb.TYPE_CODE_PTR and v.type.code != gdb.TYPE_CODE_REF: + raise Exception(f"Third argument of skip_back(t, i, v) must be a pointer, is {v.type} ({v.type.code})") + if v.type.target().strip_typedefs() != nullptr[i].type: + raise Exception(f"Third argument of skip_back(t, i, v) points to type {v.type.target().strip_typedefs()}, should be {nullptr[i].type}") + + uintptr_t = gdb.lookup_type("uintptr_t") + taddr = v.dereference().address.cast(uintptr_t) - nullptr[i].address.cast(uintptr_t) + return gdb.Value(taddr).cast(t.pointer()) + + class skip_back_gdb(gdb.Function): + "Given address of a structure item, returns address of the structure, as the SKIP_BACK macro does" + def __init__(self): + gdb.Function.__init__(self, "SKIP_BACK") + + def invoke(self, t, i, v): + return BIRD.skip_back(t, i, v) + + +BIRD.skip_back_gdb() + + +class BIRDList: + def __init__(self, val): + ltype = val.type.strip_typedefs() + if ltype.code != gdb.TYPE_CODE_UNION or ltype.tag != "list": + raise Exception(f"Not a list, is type {ltype}") + + self.head = val["head"] + self.tail_node = val["tail_node"] + + if str(self.head.address) == '0x0': + raise Exception("List head is NULL") + + if str(self.tail_node["prev"].address) == '0x0': + raise Exception("List tail is NULL") + + def walk(self, do): + cur = self.head + while cur.dereference() != self.tail_node: + do(cur) + cur = cur.dereference()["next"] + + +class BIRDListLength(gdb.Function): + """Returns length of the list, as in + print $list_length(routing_tables)""" + def __init__(self): + super(BIRDListLength, self).__init__("list_length") + + def count(self, _): + self.cnt += 1 + + def invoke(self, l): + self.cnt = 0 + BIRDList(l).walk(self.count) + return self.cnt + +BIRDListLength() + +class BIRDListItem(gdb.Function): + """Returns n-th item of the list.""" + def __init__(self): + super(BIRDListItem, self).__init__("list_item") + + class BLException(Exception): + def __init__(self, node, msg): + Exception.__init__(self, msg) + self.node = node + + def count(self, node): + if self.cnt == self.pos: + raise self.BLException(node, "Node found") + + self.cnt += 1 + + def invoke(self, l, n, t=None, item="n"): + self.cnt = 0 + self.pos = n + bl = BIRDList(l) + try: + bl.walk(self.count) + except self.BLException as e: + if t is None: + return e.node + else: + return BIRD.skip_back(t, item, e.node) + + raise Exception("List too short") + +BIRDListItem() + +class BIRDResourceSize(): + def __init__(self, netto, overhead, free): + self.netto = netto + self.overhead = overhead + self.free = free + + def __str__(self): + ns = str(self.netto) + os = str(self.overhead) + fs = str(self.free) + + return "{: >12s} | {: >12s} | {: >12s}".format(ns, os, fs) + + def __add__(self, val): + return BIRDResourceSize(self.netto + val.netto, self.overhead + val.overhead, self.free + val.free) + +class BIRDResource(): + def __init__(self, val): + self.val = val + + def __str__(self): + return f"Item {self.val.address} of class \"{self.val['class']['name'].string()}\"" + + def memsize(self): + if str(self.val["class"]["memsize"]) == '0x0': + size = self.val["class"]["size"] + ressize = gdb.lookup_type("struct resource").sizeof + return BIRDResourceSize(size - ressize, ressize, 0) + else: + raise Exception(f"Resource class {self.val['class']['name']} with defined memsize() not known by Python") + + def parse(self): + pass + +class BIRDMBResource(BIRDResource): + def __init__(self, val): + self.mbtype = gdb.lookup_type("struct mblock") + self.val = val.cast(self.mbtype) + + def memsize(self): + return BIRDResourceSize(self.val["size"], 8 + self.mbtype.sizeof, 0) + + def __str__(self): + return f"Standalone memory block {self.val.address} of size {self.val['size']}, data at {self.val['data'].address}" + +class BIRDLinPoolResource(BIRDResource): + def __init__(self, val): + self.lptype = gdb.lookup_type("struct linpool") + self.val = val.cast(self.lptype) + self.info = None + + def count_chunk(self, which): + cnt = 0 + chunk = self.val[which] + while str(chunk) != '0x0': + cnt += 1 + chunk = chunk.dereference()["next"] + return cnt + + def parse(self): + self.info = { + "std_chunks": self.count_chunk("first"), + "large_chunks": self.count_chunk("first_large"), + } + + def memsize(self): + if self.info is None: + self.parse() + + overhead = (8 - 8*self.val["use_pages"]) + gdb.lookup_type("struct lp_chunk").sizeof + return BIRDResourceSize( + self.val["total"] + self.val["total_large"], + (self.info["std_chunks"] + self.info["large_chunks"]) * overhead, + 0) + + def __str__(self): + if self.info is None: + self.parse() + + return f"Linpool {self.val.address} with {self.info['std_chunks']} standard chunks of size {self.val['chunk_size']} and {self.info['large_chunks']} large chunks" + +class BIRDSlabResource(BIRDResource): + def __init__(self, val): + self.slabtype = gdb.lookup_type("struct slab") + self.val = val.cast(self.slabtype) + self.info = None + + def count_heads_item(self, item): + self.hcnt += 1 + self.used += item.dereference().cast(self.slheadtype)["num_full"] + + def count_heads(self, which): + self.hcnt = 0 + self.used = 0 + BIRDList(self.val[which + "_heads"]).walk(self.count_heads_item) + self.info[which + "_heads"] = self.hcnt + self.info[which + "_used"] = self.used + return (self.hcnt, self.used) + + def parse(self): + self.page_size = gdb.lookup_symbol("page_size")[0].value() + self.slheadtype = gdb.lookup_type("struct sl_head") + self.info = {} + self.count_heads("empty") + self.count_heads("partial") + self.count_heads("full") + + def memsize(self): + if self.info is None: + self.parse() + + total_used = self.info["empty_used"] + self.info["partial_used"] + self.info["full_used"] + total_heads = self.info["empty_heads"] + self.info["partial_heads"] + self.info["full_heads"] + + eff_size = total_used * self.val["obj_size"] + free_size = self.info["empty_heads"] * self.page_size + total_size = total_heads * self.page_size + self.slabtype.sizeof + + return BIRDResourceSize( eff_size, total_size - free_size - eff_size, free_size) + + def __str__(self): + if self.info is None: + self.parse() + + return f"Slab {self.val.address} " + ", ".join([ + f"{self.info[x + '_heads']} {x} heads" for x in [ "empty", "partial", "full" ]]) + \ + f", {self.val['objs_per_slab']} objects of size {self.val['obj_size']} per head" + + +class BIRDPoolResource(BIRDResource): + def __init__(self, val): + self.pooltype = gdb.lookup_type("struct pool") + self.resptrtype = gdb.lookup_type("struct resource").pointer() + self.page_size = gdb.lookup_symbol("page_size")[0].value() + self.val = val.cast(self.pooltype) + self.items = None + + def parse_inside(self, val): + self.items.append(BIRDNewResource(val.cast(self.resptrtype).dereference())) + + def parse(self): + self.items = [] + BIRDList(self.val["inside"]).walk(self.parse_inside) + + def free_pages(self): + if str(self.val['pages']) == '0x0': + return 0 + else: + return self.val['pages'].dereference()['free'] + + def memsize(self): + if self.items is None: + self.parse() + + sum = BIRDResourceSize(0, self.pooltype.sizeof, self.free_pages() * self.page_size) +# for i in self.items: +# sum += i.memsize() + + return sum + + def __str__(self): + if self.items is None: + self.parse() + +# for i in self.items: +# print(i) + + return f"Resource pool {self.val.address} \"{self.val['name'].string()}\" containing {len(self.items)} items and {self.free_pages()} free pages" + +BIRDResourceMap = { + "mbl_memsize": BIRDMBResource, + "pool_memsize": BIRDPoolResource, + "lp_memsize": BIRDLinPoolResource, + "slab_memsize": BIRDSlabResource, + } + +def BIRDNewResource(res): + cms = res["class"].dereference()["memsize"] + for cx in BIRDResourceMap: + if cms == gdb.lookup_symbol(cx)[0].value(): + return BIRDResourceMap[cx](res) + + return BIRDResource(res) + + +class BIRDResourcePrinter(BIRDPrinter): + "Print BIRD's resource" + typeCode = gdb.TYPE_CODE_STRUCT + typeTag = "resource" + + def __init__(self, val): + super(BIRDResourcePrinter, self).__init__(val) + self.resource = BIRDNewResource(val) + self.resource.parse() + self.resourcetype = gdb.lookup_type("struct resource") + + if type(self.resource) == BIRDPoolResource: + self.children = self.pool_children + + def pool_children(self): + return iter([ ("\n", i.val.cast(self.resourcetype)) for i in self.resource.items ]) + + def to_string(self): + return f"[ {str(self.resource.memsize())} ] {str(self.resource)}" + + def register_printers(objfile): objfile.pretty_printers.append(BIRDFInstPrinter.lookup) objfile.pretty_printers.append(BIRDFValPrinter.lookup) @@ -148,6 +466,7 @@ def register_printers(objfile): objfile.pretty_printers.append(BIRDFLineItemPrinter.lookup) objfile.pretty_printers.append(BIRDFLinePrinter.lookup) objfile.pretty_printers.append(BIRDFExecStackPrinter.lookup) + objfile.pretty_printers.append(BIRDResourcePrinter.lookup) register_printers(gdb.current_objfile()) diff --git a/conf/conf.c b/conf/conf.c index a2b01667..e8f1559a 100644 --- a/conf/conf.c +++ b/conf/conf.c @@ -89,7 +89,7 @@ int undo_available; /* Undo was not requested from last reconfiguration */ struct config * config_alloc(const char *name) { - pool *p = rp_new(&root_pool, "Config"); + pool *p = rp_new(&root_pool, &main_birdloop, "Config"); linpool *l = lp_new_default(p); struct config *c = lp_allocz(l, sizeof(struct config)); @@ -196,7 +196,7 @@ void config_free(struct config *c) { if (c) - rfree(c->pool); + rp_free(c->pool, &root_pool); } void diff --git a/conf/conf.h b/conf/conf.h index 3bc37959..4f6aa6eb 100644 --- a/conf/conf.h +++ b/conf/conf.h @@ -35,6 +35,7 @@ struct config { u32 proto_default_debug; /* Default protocol debug mask */ u32 proto_default_mrtdump; /* Default protocol mrtdump mask */ u32 channel_default_debug; /* Default channel debug mask */ + u16 filter_vstk, filter_estk; /* Filter stack depth */ struct timeformat tf_route; /* Time format for 'show route' */ struct timeformat tf_proto; /* Time format for 'show protocol' */ struct timeformat tf_log; /* Time format for the logfile */ @@ -44,6 +45,7 @@ struct config { int cli_debug; /* Tracing of CLI connections and commands */ int latency_debug; /* I/O loop tracks duration of each event */ + int table_debug; /* Track route propagation through tables */ u32 latency_limit; /* Events with longer duration are logged (us) */ u32 watchdog_warning; /* I/O loop watchdog limit for warning (us) */ u32 watchdog_timeout; /* Watchdog timeout (in seconds, 0 = disabled) */ diff --git a/configure.ac b/configure.ac index 64181d29..5c0cf002 100644 --- a/configure.ac +++ b/configure.ac @@ -36,12 +36,6 @@ AC_ARG_ENABLE([memcheck], [enable_memcheck=yes] ) -AC_ARG_ENABLE([pthreads], - [AS_HELP_STRING([--enable-pthreads], [enable POSIX threads support @<:@try@:>@])], - [], - [enable_pthreads=try] -) - AC_ARG_ENABLE([libssh], [AS_HELP_STRING([--enable-libssh], [enable LibSSH support in RPKI @<:@try@:>@])], [], @@ -125,25 +119,19 @@ if test -z "$GCC" ; then fi BIRD_CHECK_THREAD_LOCAL -if test "$bird_cv_thread_local" = yes ; then - AC_DEFINE([HAVE_THREAD_LOCAL], [1], [Define to 1 if _Thread_local is available]) +if test "$bird_cv_thread_local" = no ; then + AC_MSG_ERROR([This program requires thread local storage.]) +elif test "$bird_cv_thread_local" != yes ; then + AC_DEFINE_UNQUOTED([_Thread_local], [$bird_cv_thread_local], [Legacy _Thread_local]) fi -if test "$enable_pthreads" != no ; then - BIRD_CHECK_PTHREADS +BIRD_CHECK_PTHREADS - if test "$bird_cv_lib_pthreads" = yes ; then - AC_DEFINE([USE_PTHREADS], [1], [Define to 1 if pthreads are enabled]) - CFLAGS="$CFLAGS -pthread" - LDFLAGS="$LDFLAGS -pthread" - proto_bfd=bfd - elif test "$enable_pthreads" = yes ; then - AC_MSG_ERROR([POSIX threads not available.]) - fi - - if test "$enable_pthreads" = try ; then - enable_pthreads="$bird_cv_lib_pthreads" - fi +if test "$bird_cv_lib_pthreads" = yes ; then + CFLAGS="$CFLAGS -pthread" + LDFLAGS="$LDFLAGS -pthread" +else + AC_MSG_ERROR([POSIX threads not available.]) fi # This is assumed to be necessary for proper BIRD build @@ -304,8 +292,7 @@ if test "$enable_mpls_kernel" != no ; then fi fi -all_protocols="$proto_bfd babel bgp mrt ospf perf pipe radv rip rpki static" - +all_protocols="bfd babel bgp mrt ospf perf pipe radv rip rpki static" all_protocols=`echo $all_protocols | sed 's/ /,/g'` if test "$with_protocols" = all ; then @@ -351,9 +338,15 @@ case $sysdesc in esac AC_CHECK_HEADERS_ONCE([alloca.h syslog.h]) -AC_CHECK_HEADER([sys/mman.h], [AC_DEFINE([HAVE_MMAP], [1], [Define to 1 if mmap() is available.])]) +AC_CHECK_HEADER([sys/mman.h], [AC_DEFINE([HAVE_MMAP], [1], [Define to 1 if mmap() is available.])], have_mman=no) +AC_CHECK_FUNC([aligned_alloc], [AC_DEFINE([HAVE_ALIGNED_ALLOC], [1], [Define to 1 if aligned_alloc() is available.])], have_aligned_alloc=no) AC_CHECK_MEMBERS([struct sockaddr.sa_len], [], [], [#include <sys/socket.h>]) +if test "$have_aligned_alloc" = "no" && test "$have_mman" = "no" ; then + AC_MSG_ERROR([No means of aligned alloc found. Need mmap() or aligned_alloc().]) +fi + + AC_C_BIGENDIAN( [AC_DEFINE([CPU_BIG_ENDIAN], [1], [Define to 1 if cpu is big endian])], [AC_DEFINE([CPU_LITTLE_ENDIAN], [1], [Define to 1 if cpu is little endian])], @@ -402,7 +395,7 @@ if test "$enable_debug" = yes ; then fi fi - if test "enable_debug_expensive" = yes ; then + if test "$enable_debug_expensive" = yes ; then AC_DEFINE([ENABLE_EXPENSIVE_CHECKS], [1], [Define to 1 if you want to run expensive consistency checks.]) fi fi @@ -467,7 +460,6 @@ AC_MSG_RESULT([ Object directory: $objdir]) AC_MSG_RESULT([ Iproute2 directory: $iproutedir]) AC_MSG_RESULT([ System configuration: $sysdesc]) AC_MSG_RESULT([ Debugging: $enable_debug]) -AC_MSG_RESULT([ POSIX threads: $enable_pthreads]) AC_MSG_RESULT([ Routing protocols: $protocols]) AC_MSG_RESULT([ LibSSH support in RPKI: $enable_libssh]) AC_MSG_RESULT([ Kernel MPLS support: $enable_mpls_kernel]) diff --git a/doc/bird.sgml b/doc/bird.sgml index 39dadaf2..d1a3b70f 100644 --- a/doc/bird.sgml +++ b/doc/bird.sgml @@ -145,6 +145,13 @@ BIRD executable by configuring out routing protocols you don't use, and <p>You can pass several command-line options to bird: <descrip> + <tag><label id="argv-block">-B <m/exp/</tag> + allocate memory using 2^<cf/exp/ byte sized blocks; + if you're expecting high memory load, raise this to + reduce number of allocated memory pages. For a million routes + in one table, the recommended setting is 18. + Default is your system page size, typically 12 for 4096 bytes. + <tag><label id="argv-config">-c <m/config name/</tag> use given configuration file instead of <it/prefix/<file>/etc/bird.conf</file>. @@ -1684,7 +1691,7 @@ Common route attributes are: <tag><label id="rta-source"><m/enum/ source</tag> what protocol has told me about this route. Possible values: - <cf/RTS_DUMMY/, <cf/RTS_STATIC/, <cf/RTS_INHERIT/, <cf/RTS_DEVICE/, + <cf/RTS_STATIC/, <cf/RTS_INHERIT/, <cf/RTS_DEVICE/, <cf/RTS_RIP/, <cf/RTS_OSPF/, <cf/RTS_OSPF_IA/, <cf/RTS_OSPF_EXT1/, <cf/RTS_OSPF_EXT2/, <cf/RTS_BGP/, <cf/RTS_PIPE/, <cf/RTS_BABEL/. @@ -4117,6 +4124,14 @@ include standard channel config options; see the example below. <tag><label id="pipe-peer-table">peer table <m/table/</tag> Defines secondary routing table to connect to. The primary one is selected by the <cf/table/ keyword. + + <tag><label id="pipe-max-generation">max generation <m/expr/</tag> + Sets maximal generation of route that may pass through this pipe. + The generation value is increased by one by each pipe on its path. + Not meeting this requirement causes an error message complaining about + an overpiped route. If you have long chains of pipes, you probably want + to raise this value; anyway the default of 16 should be enough for even + most strange uses. Maximum is 254. </descrip> <sect1>Attributes diff --git a/filter/config.Y b/filter/config.Y index 7820e719..8034b790 100644 --- a/filter/config.Y +++ b/filter/config.Y @@ -248,10 +248,6 @@ assert_assign(struct f_lval *lval, struct f_inst *expr, const char *start, const setter = f_new_inst(FI_VAR_SET, expr, lval->sym); getter = f_new_inst(FI_VAR_GET, lval->sym); break; - case F_LVAL_PREFERENCE: - setter = f_new_inst(FI_PREF_SET, expr); - getter = f_new_inst(FI_PREF_GET); - break; case F_LVAL_SA: setter = f_new_inst(FI_RTA_SET, expr, lval->sa); getter = f_new_inst(FI_RTA_GET, lval->sa); @@ -288,7 +284,7 @@ CF_KEYWORDS(FUNCTION, PRINT, PRINTN, UNSET, RETURN, PREPEND, FIRST, LAST, LAST_NONAGGREGATED, MATCH, EMPTY, FILTER, WHERE, EVAL, ATTRIBUTE, - BT_ASSERT, BT_TEST_SUITE, BT_CHECK_ASSIGN, BT_TEST_SAME, FORMAT) + BT_ASSERT, BT_TEST_SUITE, BT_CHECK_ASSIGN, BT_TEST_SAME, FORMAT, STACKS) %nonassoc THEN %nonassoc ELSE @@ -312,6 +308,12 @@ CF_KEYWORDS(FUNCTION, PRINT, PRINTN, UNSET, RETURN, CF_GRAMMAR +conf: FILTER STACKS expr expr ';' { + new_config->filter_vstk = $3; + new_config->filter_estk = $4; + } + ; + conf: filter_def ; filter_def: FILTER symbol { $2 = cf_define_symbol($2, SYM_FILTER, filter, NULL); cf_push_scope( $2 ); } @@ -751,6 +753,7 @@ static_attr: | IFNAME { $$ = f_new_static_attr(T_STRING, SA_IFNAME, 0); } | IFINDEX { $$ = f_new_static_attr(T_INT, SA_IFINDEX, 1); } | WEIGHT { $$ = f_new_static_attr(T_INT, SA_WEIGHT, 0); } + | PREFERENCE { $$ = f_new_static_attr(T_INT, SA_PREF, 0); } | GW_MPLS { $$ = f_new_static_attr(T_INT, SA_GW_MPLS, 0); } ; @@ -777,8 +780,6 @@ term: | constant { $$ = $1; } | constructor { $$ = $1; } - | PREFERENCE { $$ = f_new_inst(FI_PREF_GET); } - | static_attr { $$ = f_new_inst(FI_RTA_GET, $1); } | dynamic_attr { $$ = f_new_inst(FI_EA_GET, $1); } @@ -871,9 +872,6 @@ cmd: cf_error( "This static attribute is read-only."); $$ = f_new_inst(FI_RTA_SET, $3, $1); } - | PREFERENCE '=' term ';' { - $$ = f_new_inst(FI_PREF_SET, $3); - } | UNSET '(' dynamic_attr ')' ';' { $$ = f_new_inst(FI_EA_UNSET, $3); } @@ -916,7 +914,6 @@ get_cf_position: lvalue: CF_SYM_KNOWN { cf_assert_symbol($1, SYM_VARIABLE); $$ = (struct f_lval) { .type = F_LVAL_VARIABLE, .sym = $1 }; } - | PREFERENCE { $$ = (struct f_lval) { .type = F_LVAL_PREFERENCE }; } | static_attr { $$ = (struct f_lval) { .type = F_LVAL_SA, .sa = $1 }; } | dynamic_attr { $$ = (struct f_lval) { .type = F_LVAL_EA, .da = $1 }; }; diff --git a/filter/data.h b/filter/data.h index d296776d..45246f9f 100644 --- a/filter/data.h +++ b/filter/data.h @@ -100,6 +100,7 @@ enum f_sa_code { SA_IFNAME, SA_IFINDEX, SA_WEIGHT, + SA_PREF, SA_GW_MPLS, } PACKED; diff --git a/filter/decl.m4 b/filter/decl.m4 index 5242c04c..44537aaa 100644 --- a/filter/decl.m4 +++ b/filter/decl.m4 @@ -32,6 +32,7 @@ m4_divert(-1)m4_dnl # # 101 content of per-inst struct # 102 constructor arguments +# 110 constructor attributes # 103 constructor body # 104 dump line item content # (there may be nothing in dump-line content and @@ -45,6 +46,7 @@ m4_divert(-1)m4_dnl # Here are macros to allow you to _divert to the right directions. m4_define(FID_STRUCT_IN, `m4_divert(101)') m4_define(FID_NEW_ARGS, `m4_divert(102)') +m4_define(FID_NEW_ATTRIBUTES, `m4_divert(110)') m4_define(FID_NEW_BODY, `m4_divert(103)') m4_define(FID_DUMP_BODY, `m4_divert(104)m4_define([[FID_DUMP_BODY_EXISTS]])') m4_define(FID_LINEARIZE_BODY, `m4_divert(105)') @@ -106,15 +108,18 @@ FID_STRUCT_IN()m4_dnl struct f_inst * f$1; FID_NEW_ARGS()m4_dnl , struct f_inst * f$1 +FID_NEW_ATTRIBUTES()m4_dnl +NONNULL(m4_eval($1+1)) FID_NEW_BODY()m4_dnl whati->f$1 = f$1; -for (const struct f_inst *child = f$1; child; child = child->next) { - what->size += child->size; +const struct f_inst *child$1 = f$1; +do { + what->size += child$1->size; FID_IFCONST([[ - if (child->fi_code != FI_CONSTANT) + if (child$1->fi_code != FI_CONSTANT) constargs = 0; ]]) -} +} while (child$1 = child$1->next); FID_LINEARIZE_BODY pos = linearize(dest, whati->f$1, pos); FID_INTERPRET_BODY()') @@ -190,6 +195,7 @@ FID_INTERPRET_BODY()') # that was needed in the former implementation. m4_define(LINEX, `FID_INTERPRET_EXEC()LINEX_($1)FID_INTERPRET_NEW()return $1 FID_INTERPRET_BODY()') m4_define(LINEX_, `do { + if (fstk->ecnt + 1 >= fstk->elen) runtime("Filter execution stack overflow"); fstk->estk[fstk->ecnt].pos = 0; fstk->estk[fstk->ecnt].line = $1; fstk->estk[fstk->ecnt].ventry = fstk->vcnt; @@ -227,7 +233,7 @@ FID_INTERPRET_BODY()') # state the result and put it to the right place. m4_define(RESULT, `RESULT_TYPE([[$1]]) RESULT_([[$1]],[[$2]],[[$3]])') m4_define(RESULT_, `RESULT_VAL([[ (struct f_val) { .type = $1, .val.$2 = $3 } ]])') -m4_define(RESULT_VAL, `FID_HIC(, [[do { res = $1; fstk->vcnt++; } while (0)]], +m4_define(RESULT_VAL, `FID_HIC(, [[do { res = $1; f_vcnt_check_overflow(1); fstk->vcnt++; } while (0)]], [[return fi_constant(what, $1)]])') m4_define(RESULT_VOID, `RESULT_VAL([[ (struct f_val) { .type = T_VOID } ]])') @@ -309,7 +315,9 @@ m4_undivert(107)m4_dnl FID_NEW()m4_dnl Constructor and interpreter code together FID_HIC( [[m4_dnl Public declaration of constructor in H file -struct f_inst *f_new_inst_]]INST_NAME()[[(enum f_instruction_code fi_code +struct f_inst * +m4_undivert(110)m4_dnl +f_new_inst_]]INST_NAME()[[(enum f_instruction_code fi_code m4_undivert(102)m4_dnl );]], [[m4_dnl The one case in The Big Switch inside interpreter @@ -321,7 +329,9 @@ m4_undivert(102)m4_dnl break; ]], [[m4_dnl Constructor itself -struct f_inst *f_new_inst_]]INST_NAME()[[(enum f_instruction_code fi_code +struct f_inst * +m4_undivert(110)m4_dnl +f_new_inst_]]INST_NAME()[[(enum f_instruction_code fi_code m4_undivert(102)m4_dnl ) { diff --git a/filter/f-inst.c b/filter/f-inst.c index b876a937..0341a2f1 100644 --- a/filter/f-inst.c +++ b/filter/f-inst.c @@ -519,20 +519,21 @@ { STATIC_ATTR; ACCESS_RTE; - struct rta *rta = (*fs->rte)->attrs; + struct rta *rta = fs->rte->attrs; switch (sa.sa_code) { case SA_FROM: RESULT(sa.f_type, ip, rta->from); break; case SA_GW: RESULT(sa.f_type, ip, rta->nh.gw); break; - case SA_NET: RESULT(sa.f_type, net, (*fs->rte)->net->n.addr); break; - case SA_PROTO: RESULT(sa.f_type, s, rta->src->proto->name); break; + case SA_NET: RESULT(sa.f_type, net, fs->rte->net); break; + case SA_PROTO: RESULT(sa.f_type, s, fs->rte->src->owner->name); break; case SA_SOURCE: RESULT(sa.f_type, i, rta->source); break; case SA_SCOPE: RESULT(sa.f_type, i, rta->scope); break; case SA_DEST: RESULT(sa.f_type, i, rta->dest); break; case SA_IFNAME: RESULT(sa.f_type, s, rta->nh.iface ? rta->nh.iface->name : ""); break; case SA_IFINDEX: RESULT(sa.f_type, i, rta->nh.iface ? rta->nh.iface->index : 0); break; case SA_WEIGHT: RESULT(sa.f_type, i, rta->nh.weight + 1); break; + case SA_PREF: RESULT(sa.f_type, i, rta->pref); break; case SA_GW_MPLS: RESULT(sa.f_type, i, rta->nh.labels ? rta->nh.label[0] : MPLS_NULL); break; default: @@ -549,7 +550,7 @@ f_rta_cow(fs); { - struct rta *rta = (*fs->rte)->attrs; + struct rta *rta = fs->rte->attrs; switch (sa.sa_code) { @@ -561,7 +562,8 @@ { ip_addr ip = v1.val.ip; struct iface *ifa = ipa_is_link_local(ip) ? rta->nh.iface : NULL; - neighbor *n = neigh_find(rta->src->proto, ip, ifa, 0); + /* XXX this code supposes that every owner is a protocol XXX */ + neighbor *n = neigh_find(SKIP_BACK(struct proto, sources, fs->rte->src->owner), ip, ifa, 0); if (!n || (n->scope == SCOPE_HOST)) runtime( "Invalid gw address" ); @@ -637,6 +639,10 @@ } break; + case SA_PREF: + rta->pref = v1.val.i; + break; + default: bug("Invalid static attribute access (%u/%u)", sa.f_type, sa.sa_code); } @@ -804,20 +810,6 @@ } } - INST(FI_PREF_GET, 0, 1) { - ACCESS_RTE; - RESULT(T_INT, i, (*fs->rte)->pref); - } - - INST(FI_PREF_SET, 1, 0) { - ACCESS_RTE; - ARG(1,T_INT); - if (v1.val.i > 0xFFFF) - runtime( "Setting preference value out of bounds" ); - f_rte_cow(fs); - (*fs->rte)->pref = v1.val.i; - } - INST(FI_LENGTH, 1, 1) { /* Get length of */ ARG_ANY(1); switch(v1.type) { @@ -1003,6 +995,7 @@ curline.vbase = curline.ventry; /* Storage for local variables */ + f_vcnt_check_overflow(sym->function->vars); memset(&(fstk->vstk[fstk->vcnt]), 0, sizeof(struct f_val) * sym->function->vars); fstk->vcnt += sym->function->vars; } @@ -1219,10 +1212,10 @@ INST(FI_ROA_CHECK_IMPLICIT, 0, 1) { /* ROA Check */ NEVER_CONSTANT; RTC(1); - struct rtable *table = rtc->table; + rtable *table = rtc->table; ACCESS_RTE; ACCESS_EATTRS; - const net_addr *net = (*fs->rte)->net->n.addr; + const net_addr *net = fs->rte->net; /* We ignore temporary attributes, probably not a problem here */ /* 0x02 is a value of BA_AS_PATH, we don't want to include BGP headers */ @@ -1251,7 +1244,7 @@ ARG(1, T_NET); ARG(2, T_INT); RTC(3); - struct rtable *table = rtc->table; + rtable *table = rtc->table; u32 as = v2.val.i; diff --git a/filter/filter.c b/filter/filter.c index e505d570..625b3ade 100644 --- a/filter/filter.c +++ b/filter/filter.c @@ -50,36 +50,31 @@ enum f_exception { FE_RETURN = 0x1, }; - -struct filter_stack { - /* Value stack for execution */ -#define F_VAL_STACK_MAX 4096 - uint vcnt; /* Current value stack size; 0 for empty */ - uint ecnt; /* Current execute stack size; 0 for empty */ - - struct f_val vstk[F_VAL_STACK_MAX]; /* The stack itself */ - - /* Instruction stack for execution */ -#define F_EXEC_STACK_MAX 4096 - struct { - const struct f_line *line; /* The line that is being executed */ - uint pos; /* Instruction index in the line */ - uint ventry; /* Value stack depth on entry */ - uint vbase; /* Where to index variable positions from */ - enum f_exception emask; /* Exception mask */ - } estk[F_EXEC_STACK_MAX]; +struct filter_exec_stack { + const struct f_line *line; /* The line that is being executed */ + uint pos; /* Instruction index in the line */ + uint ventry; /* Value stack depth on entry */ + uint vbase; /* Where to index variable positions from */ + enum f_exception emask; /* Exception mask */ }; /* Internal filter state, to be allocated on stack when executing filters */ struct filter_state { /* Stacks needed for execution */ - struct filter_stack *stack; + struct filter_stack { + /* Current filter stack depth */ - /* The route we are processing. This may be NULL to indicate no route available. */ - struct rte **rte; + /* Value stack */ + uint vcnt, vlen; + struct f_val *vstk; - /* The old rta to be freed after filters are done. */ - struct rta *old_rta; + /* Instruction stack for execution */ + uint ecnt, elen; + struct filter_exec_stack *estk; + } stack; + + /* The route we are processing. This may be NULL to indicate no route available. */ + struct rte *rte; /* Cached pointer to ea_list */ struct ea_list **eattrs; @@ -95,21 +90,16 @@ struct filter_state { }; _Thread_local static struct filter_state filter_state; -_Thread_local static struct filter_stack filter_stack; void (*bt_assert_hook)(int result, const struct f_line_item *assert); -static inline void f_cache_eattrs(struct filter_state *fs) -{ - fs->eattrs = &((*fs->rte)->attrs->eattrs); -} +#define _f_stack_init(fs, px, def) ((fs).stack.px##stk = alloca(sizeof(*(fs).stack.px##stk) * ((fs).stack.px##len = (config && config->filter_##px##stk) ? config->filter_##px##stk : (def)))) -static inline void f_rte_cow(struct filter_state *fs) -{ - if (!((*fs->rte)->flags & REF_COW)) - return; +#define f_stack_init(fs) ( _f_stack_init(fs, v, 128), _f_stack_init(fs, e, 128) ) - *fs->rte = rte_cow(*fs->rte); +static inline void f_cache_eattrs(struct filter_state *fs) +{ + fs->eattrs = &(fs->rte->attrs->eattrs); } /* @@ -118,22 +108,16 @@ static inline void f_rte_cow(struct filter_state *fs) static void f_rta_cow(struct filter_state *fs) { - if (!rta_is_cached((*fs->rte)->attrs)) + if (!rta_is_cached(fs->rte->attrs)) return; - /* Prepare to modify rte */ - f_rte_cow(fs); - - /* Store old rta to free it later, it stores reference from rte_cow() */ - fs->old_rta = (*fs->rte)->attrs; - /* * Get shallow copy of rta. Fields eattrs and nexthops of rta are shared * with fs->old_rta (they will be copied when the cached rta will be obtained * at the end of f_run()), also the lock of hostentry is inherited (we * suppose hostentry is not changed by filters). */ - (*fs->rte)->attrs = rta_do_cow((*fs->rte)->attrs, fs->pool); + fs->rte->attrs = rta_do_cow(fs->rte->attrs, fs->pool); /* Re-cache the ea_list */ f_cache_eattrs(fs); @@ -163,15 +147,17 @@ interpret(struct filter_state *fs, const struct f_line *line, struct f_val *val) ASSERT(line->args == 0); /* Initialize the filter stack */ - struct filter_stack *fstk = fs->stack; + struct filter_stack *fstk = &fs->stack; fstk->vcnt = line->vars; memset(fstk->vstk, 0, sizeof(struct f_val) * line->vars); /* The same as with the value stack. Not resetting the stack for performance reasons. */ fstk->ecnt = 1; - fstk->estk[0].line = line; - fstk->estk[0].pos = 0; + fstk->estk[0] = (struct filter_exec_stack) { + .line = line, + .pos = 0, + }; #define curline fstk->estk[fstk->ecnt-1] @@ -191,6 +177,8 @@ interpret(struct filter_state *fs, const struct f_line *line, struct f_val *val) #define v2 vv(1) #define v3 vv(2) +#define f_vcnt_check_overflow(n) do { if (fstk->vcnt + n >= fstk->vlen) runtime("Filter execution stack overflow"); } while (0) + #define runtime(fmt, ...) do { \ if (!(fs->flags & FF_SILENT)) \ log_rl(&rl_runtime_err, L_ERR "filters, line %d: " fmt, what->lineno, ##__VA_ARGS__); \ @@ -241,29 +229,15 @@ interpret(struct filter_state *fs, const struct f_line *line, struct f_val *val) /** * f_run - run a filter for a route * @filter: filter to run - * @rte: route being filtered, may be modified + * @rte: route being filtered, must be write-able * @tmp_pool: all filter allocations go from this pool * @flags: flags * - * If filter needs to modify the route, there are several - * posibilities. @rte might be read-only (with REF_COW flag), in that - * case rw copy is obtained by rte_cow() and @rte is replaced. If - * @rte is originally rw, it may be directly modified (and it is never - * copied). - * - * The returned rte may reuse the (possibly cached, cloned) rta, or - * (if rta was modified) contains a modified uncached rta, which - * uses parts allocated from @tmp_pool and parts shared from original - * rta. There is one exception - if @rte is rw but contains a cached - * rta and that is modified, rta in returned rte is also cached. - * - * Ownership of cached rtas is consistent with rte, i.e. - * if a new rte is returned, it has its own clone of cached rta - * (and cached rta of read-only source rte is intact), if rte is - * modified in place, old cached rta is possibly freed. + * If @rte->attrs is cached, the returned rte allocates a new rta on + * tmp_pool, otherwise the filters may modify it. */ enum filter_return -f_run(const struct filter *filter, struct rte **rte, struct linpool *tmp_pool, int flags) +f_run(const struct filter *filter, struct rte *rte, struct linpool *tmp_pool, int flags) { if (filter == FILTER_ACCEPT) return F_ACCEPT; @@ -271,48 +245,22 @@ f_run(const struct filter *filter, struct rte **rte, struct linpool *tmp_pool, i if (filter == FILTER_REJECT) return F_REJECT; - int rte_cow = ((*rte)->flags & REF_COW); DBG( "Running filter `%s'...", filter->name ); /* Initialize the filter state */ filter_state = (struct filter_state) { - .stack = &filter_stack, .rte = rte, .pool = tmp_pool, .flags = flags, }; + f_stack_init(filter_state); + LOG_BUFFER_INIT(filter_state.buf); /* Run the interpreter itself */ enum filter_return fret = interpret(&filter_state, filter->root, NULL); - if (filter_state.old_rta) { - /* - * Cached rta was modified and filter_state->rte contains now an uncached one, - * sharing some part with the cached one. The cached rta should - * be freed (if rte was originally COW, filter_state->old_rta is a clone - * obtained during rte_cow()). - * - * This also implements the exception mentioned in f_run() - * description. The reason for this is that rta reuses parts of - * filter_state->old_rta, and these may be freed during rta_free(filter_state->old_rta). - * This is not the problem if rte was COW, because original rte - * also holds the same rta. - */ - if (!rte_cow) { - /* Cache the new attrs */ - (*filter_state.rte)->attrs = rta_lookup((*filter_state.rte)->attrs); - - /* Drop cached ea_list pointer */ - filter_state.eattrs = NULL; - } - - /* Uncache the old attrs and drop the pointer as it is invalid now. */ - rta_free(filter_state.old_rta); - filter_state.old_rta = NULL; - } - /* Process the filter output, log it and return */ if (fret < F_ACCEPT) { if (!(filter_state.flags & FF_SILENT)) @@ -337,18 +285,18 @@ f_run(const struct filter *filter, struct rte **rte, struct linpool *tmp_pool, i */ enum filter_return -f_eval_rte(const struct f_line *expr, struct rte **rte, struct linpool *tmp_pool) +f_eval_rte(const struct f_line *expr, struct rte *rte, struct linpool *tmp_pool) { filter_state = (struct filter_state) { - .stack = &filter_stack, .rte = rte, .pool = tmp_pool, }; + f_stack_init(filter_state); + LOG_BUFFER_INIT(filter_state.buf); - ASSERT(!((*rte)->flags & REF_COW)); - ASSERT(!rta_is_cached((*rte)->attrs)); + ASSERT(!rta_is_cached(rte->attrs)); return interpret(&filter_state, expr, NULL); } @@ -363,10 +311,11 @@ enum filter_return f_eval(const struct f_line *expr, struct linpool *tmp_pool, struct f_val *pres) { filter_state = (struct filter_state) { - .stack = &filter_stack, .pool = tmp_pool, }; + f_stack_init(filter_state); + LOG_BUFFER_INIT(filter_state.buf); enum filter_return fret = interpret(&filter_state, expr, pres); @@ -383,10 +332,11 @@ f_eval_int(const struct f_line *expr) { /* Called independently in parse-time to eval expressions */ filter_state = (struct filter_state) { - .stack = &filter_stack, .pool = cfg_mem, }; + f_stack_init(filter_state); + struct f_val val; LOG_BUFFER_INIT(filter_state.buf); @@ -475,6 +425,23 @@ filter_commit(struct config *new, struct config *old) } } +void channel_filter_dump(const struct filter *f) +{ + if (f == FILTER_ACCEPT) + debug(" ALL"); + else if (f == FILTER_REJECT) + debug(" NONE"); + else if (f == FILTER_UNDEF) + debug(" UNDEF"); + else if (f->sym) { + ASSERT(f->sym->filter == f); + debug(" named filter %s", f->sym->name); + } else { + debug("\n"); + f_dump_line(f->root, 2); + } +} + void filters_dump_all(void) { struct symbol *sym; @@ -494,19 +461,10 @@ void filters_dump_all(void) struct channel *c; WALK_LIST(c, sym->proto->proto->channels) { debug(" Channel %s (%s) IMPORT", c->name, net_label[c->net_type]); - if (c->in_filter == FILTER_ACCEPT) - debug(" ALL\n"); - else if (c->in_filter == FILTER_REJECT) - debug(" NONE\n"); - else if (c->in_filter == FILTER_UNDEF) - debug(" UNDEF\n"); - else if (c->in_filter->sym) { - ASSERT(c->in_filter->sym->filter == c->in_filter); - debug(" named filter %s\n", c->in_filter->sym->name); - } else { - debug("\n"); - f_dump_line(c->in_filter->root, 2); - } + channel_filter_dump(c->in_filter); + debug(" EXPORT", c->name, net_label[c->net_type]); + channel_filter_dump(c->out_filter); + debug("\n"); } } } diff --git a/filter/filter.h b/filter/filter.h index 26c1037b..9964831c 100644 --- a/filter/filter.h +++ b/filter/filter.h @@ -51,8 +51,8 @@ struct filter { struct rte; -enum filter_return f_run(const struct filter *filter, struct rte **rte, struct linpool *tmp_pool, int flags); -enum filter_return f_eval_rte(const struct f_line *expr, struct rte **rte, struct linpool *tmp_pool); +enum filter_return f_run(const struct filter *filter, struct rte *rte, struct linpool *tmp_pool, int flags); +enum filter_return f_eval_rte(const struct f_line *expr, struct rte *rte, struct linpool *tmp_pool); uint f_eval_int(const struct f_line *expr); enum filter_return f_eval_buf(const struct f_line *expr, struct linpool *tmp_pool, buffer *buf); diff --git a/filter/filter_test.c b/filter/filter_test.c index 7e4af092..2a0b5431 100644 --- a/filter/filter_test.c +++ b/filter/filter_test.c @@ -72,6 +72,7 @@ int main(int argc, char *argv[]) { bt_init(argc, argv); + bt_bird_init(); bt_assert_hook = bt_assert_filter; diff --git a/filter/test.conf b/filter/test.conf index 3a8804a1..93e7a770 100644 --- a/filter/test.conf +++ b/filter/test.conf @@ -378,7 +378,6 @@ bt_test_suite(t_ip_set, "Testing sets of ip address"); function t_enum() { - bt_assert(format(RTS_DUMMY) = "(enum 30)0"); bt_assert(format(RTS_STATIC) = "(enum 30)1"); bt_assert(format(NET_IP4) = "(enum 36)1"); bt_assert(format(NET_VPN6) = "(enum 36)4"); diff --git a/filter/tree_test.c b/filter/tree_test.c index 6472d17e..883d13bf 100644 --- a/filter/tree_test.c +++ b/filter/tree_test.c @@ -20,7 +20,7 @@ start_conf_env(void) { bt_bird_init(); - pool *p = rp_new(&root_pool, "helper_pool"); + pool *p = rp_new(&root_pool, &main_birdloop, "helper_pool"); linpool *l = lp_new_default(p); cfg_mem = l; } diff --git a/lib/Makefile b/lib/Makefile index 4378a7bd..98c5db3c 100644 --- a/lib/Makefile +++ b/lib/Makefile @@ -1,4 +1,4 @@ -src := bitmap.c bitops.c blake2s.c blake2b.c checksum.c event.c flowspec.c idm.c ip.c lists.c mac.c md5.c mempool.c net.c patmatch.c printf.c resource.c sha1.c sha256.c sha512.c slab.c slists.c strtoul.c tbf.c timer.c xmalloc.c +src := bitmap.c bitops.c blake2s.c blake2b.c checksum.c event.c flowspec.c idm.c ip.c lists.c mac.c md5.c mempool.c net.c patmatch.c printf.c resource.c rcu.c sha1.c sha256.c sha512.c slab.c slists.c strtoul.c tbf.c timer.c xmalloc.c obj := $(src-o-files) $(all-daemon) diff --git a/lib/birdlib.h b/lib/birdlib.h index 431b7c0d..385bf75c 100644 --- a/lib/birdlib.h +++ b/lib/birdlib.h @@ -9,6 +9,7 @@ #ifndef _BIRD_BIRDLIB_H_ #define _BIRD_BIRDLIB_H_ +#include "sysdep/config.h" #include "lib/alloca.h" /* Ugly structure offset handling macros */ @@ -16,7 +17,7 @@ struct align_probe { char x; long int y; }; #define OFFSETOF(s, i) ((size_t) &((s *)0)->i) -#define SKIP_BACK(s, i, p) ((s *)((char *)p - OFFSETOF(s, i))) +#define SKIP_BACK(s, i, p) ({ s *_ptr = ((s *)((char *)p - OFFSETOF(s, i))); ASSERT_DIE(&_ptr->i == p); _ptr; }) #define BIRD_ALIGN(s, a) (((s)+a-1)&~(a-1)) #define CPU_STRUCT_ALIGN (sizeof(struct align_probe)) @@ -70,6 +71,7 @@ static inline int u64_cmp(u64 i1, u64 i2) /* Macros for gcc attributes */ #define NORET __attribute__((noreturn)) +#define USE_RESULT __atribute__((warn_unused_result)) #define UNUSED __attribute__((unused)) #define PACKED __attribute__((packed)) #define NONNULL(...) __attribute__((nonnull((__VA_ARGS__)))) @@ -77,10 +79,6 @@ static inline int u64_cmp(u64 i1, u64 i2) #define STATIC_ASSERT(EXP) _Static_assert(EXP, #EXP) #define STATIC_ASSERT_MSG(EXP,MSG) _Static_assert(EXP, MSG) -#ifndef HAVE_THREAD_LOCAL -#define _Thread_local -#endif - /* Microsecond time */ typedef s64 btime; diff --git a/lib/coro.h b/lib/coro.h new file mode 100644 index 00000000..b36f1d2c --- /dev/null +++ b/lib/coro.h @@ -0,0 +1,31 @@ +/* + * BIRD Coroutines + * + * (c) 2017 Martin Mares <mj@ucw.cz> + * (c) 2020-2021 Maria Matejka <mq@jmq.cz> + * + * Can be freely distributed and used under the terms of the GNU GPL. + */ + +#ifndef _BIRD_CORO_H_ +#define _BIRD_CORO_H_ + +#include "lib/resource.h" + +/* A completely opaque coroutine handle. */ +struct coroutine; + +/* Coroutines are independent threads bound to pools. + * You request a coroutine by calling coro_run(). + * It is forbidden to free a running coroutine from outside. + * The running coroutine must free itself by rfree() before returning. + */ +struct coroutine *coro_run(pool *, void (*entry)(void *), void *data); + +/* Get self. */ +extern _Thread_local struct coroutine *this_coro; + +/* Just wait for a little while. Not intended for general use; use events if possible. */ +void coro_yield(void); + +#endif diff --git a/lib/event.c b/lib/event.c index 273447e0..5031f314 100644 --- a/lib/event.c +++ b/lib/event.c @@ -19,8 +19,14 @@ * events in them and explicitly ask to run them. */ +#undef LOCAL_DEBUG + #include "nest/bird.h" #include "lib/event.h" +#include "lib/locking.h" +#include "lib/io-loop.h" + +extern _Thread_local struct coroutine *this_coro; event_list global_event_list; event_list global_work_list; @@ -28,11 +34,16 @@ event_list global_work_list; inline void ev_postpone(event *e) { + event_list *el = e->list; + if (!el) + return; + + ASSERT_DIE(birdloop_inside(el->loop)); + + LOCK_DOMAIN(event, el->lock); if (ev_active(e)) - { - rem_node(&e->n); - e->n.next = NULL; - } + rem_node(&e->n); + UNLOCK_DOMAIN(event, el->lock); } static void @@ -95,40 +106,50 @@ ev_run(event *e) * list @l which can be run by calling ev_run_list(). */ inline void -ev_enqueue(event_list *l, event *e) +ev_send(event_list *l, event *e) { - ev_postpone(e); - add_tail(l, &e->n); -} + DBG("ev_send(%p, %p)\n", l, e); + ASSERT_DIE(e->hook); + ASSERT_DIE(!e->list || (e->list == l) || (e->list->loop == l->loop)); -/** - * ev_schedule - schedule an event - * @e: an event - * - * This function schedules an event by enqueueing it to a system-wide - * event list which is run by the platform dependent code whenever - * appropriate. - */ -void -ev_schedule(event *e) -{ - ev_enqueue(&global_event_list, e); -} + e->list = l; -/** - * ev_schedule_work - schedule a work-event. - * @e: an event - * - * This function schedules an event by enqueueing it to a system-wide work-event - * list which is run by the platform dependent code whenever appropriate. This - * is designated for work-events instead of regular events. They are executed - * less often in order to not clog I/O loop. - */ -void -ev_schedule_work(event *e) -{ - if (!ev_active(e)) - add_tail(&global_work_list, &e->n); + struct event_cork *ec = e->cork; + + uint ping = 0; + + if (ec) + { + LOCK_DOMAIN(cork, ec->lock); + LOCK_DOMAIN(event, l->lock); + + if (!enlisted(&e->n)) + if (ec->count) + add_tail(&ec->events, &e->n); + else + { + add_tail(&l->events, &e->n); + ping = 1; + } + + UNLOCK_DOMAIN(event, l->lock); + UNLOCK_DOMAIN(cork, ec->lock); + } + else + { + LOCK_DOMAIN(event, l->lock); + + if (!enlisted(&e->n)) + { + add_tail(&l->events, &e->n); + ping = 1; + } + + UNLOCK_DOMAIN(event, l->lock); + } + + if (ping) + birdloop_ping(l->loop); } void io_log_event(void *hook, void *data); @@ -142,35 +163,64 @@ void io_log_event(void *hook, void *data); int ev_run_list(event_list *l) { + const _Bool legacy = LEGACY_EVENT_LIST(l); + + if (legacy) + ASSERT_THE_BIRD_LOCKED; + node *n; - list tmp_list; + list tmp_list; init_list(&tmp_list); - add_tail_list(&tmp_list, l); - init_list(l); + + /* Move the event list contents to a local list to avoid executing repeatedly added events */ + LOCK_DOMAIN(event, l->lock); + add_tail_list(&tmp_list, &l->events); + init_list(&l->events); + UNLOCK_DOMAIN(event, l->lock); + WALK_LIST_FIRST(n, tmp_list) { event *e = SKIP_BACK(event, n, n); - /* This is ugly hack, we want to log just events executed from the main I/O loop */ - if ((l == &global_event_list) || (l == &global_work_list)) + if (legacy) + { + /* The legacy way of event execution */ io_log_event(e->hook, e->data); - - ev_run(e); + ev_postpone(e); + e->hook(e->data); + } + else + { + // io_log_event(e->hook, e->data); /* TODO: add support for event logging in other io loops */ + ASSERT_DIE(e->list == l); + LOCK_DOMAIN(event, l->lock); + rem_node(&e->n); + UNLOCK_DOMAIN(event, l->lock); + e->hook(e->data); + } } - return !EMPTY_LIST(*l); + LOCK_DOMAIN(event, l->lock); + int repeat = ! EMPTY_LIST(l->events); + UNLOCK_DOMAIN(event, l->lock); + return repeat; } int ev_run_list_limited(event_list *l, uint limit) { + ASSERT_DIE(LEGACY_EVENT_LIST(l)); + ASSERT_THE_BIRD_LOCKED; + node *n; list tmp_list; + LOCK_DOMAIN(event, l->lock); init_list(&tmp_list); - add_tail_list(&tmp_list, l); - init_list(l); + add_tail_list(&tmp_list, &l->events); + init_list(&l->events); + UNLOCK_DOMAIN(event, l->lock); WALK_LIST_FIRST(n, tmp_list) { @@ -179,21 +229,60 @@ ev_run_list_limited(event_list *l, uint limit) if (!limit) break; - /* This is ugly hack, we want to log just events executed from the main I/O loop */ - if ((l == &global_event_list) || (l == &global_work_list)) - io_log_event(e->hook, e->data); + io_log_event(e->hook, e->data); ev_run(e); limit--; } + LOCK_DOMAIN(event, l->lock); if (!EMPTY_LIST(tmp_list)) { /* Attach new items after the unprocessed old items */ - add_tail_list(&tmp_list, l); - init_list(l); - add_tail_list(l, &tmp_list); + add_tail_list(&tmp_list, &l->events); + init_list(&l->events); + add_tail_list(&l->events, &tmp_list); + } + + int repeat = ! EMPTY_LIST(l->events); + UNLOCK_DOMAIN(event, l->lock); + + return repeat; +} + +void ev_cork(struct event_cork *ec) +{ + LOCK_DOMAIN(cork, ec->lock); + ec->count++; + UNLOCK_DOMAIN(cork, ec->lock); +} + +void ev_uncork(struct event_cork *ec) +{ + LOCK_DOMAIN(cork, ec->lock); + + if (--ec->count) + { + UNLOCK_DOMAIN(cork, ec->lock); + return; } - return !EMPTY_LIST(*l); + node *n; + WALK_LIST_FIRST(n, ec->events) + { + event *e = SKIP_BACK(event, n, n); + event_list *el = e->list; + + rem_node(&e->n); + + LOCK_DOMAIN(event, el->lock); + add_tail(&el->events, &e->n); + UNLOCK_DOMAIN(event, el->lock); + + birdloop_ping(el->loop); + } + + UNLOCK_DOMAIN(cork, ec->lock); + + birdloop_ping(&main_birdloop); } diff --git a/lib/event.h b/lib/event.h index 5f3b78d8..cd85bf78 100644 --- a/lib/event.h +++ b/lib/event.h @@ -10,33 +10,92 @@ #define _BIRD_EVENT_H_ #include "lib/resource.h" +#include "lib/locking.h" + +#include <stdatomic.h> + +DEFINE_DOMAIN(event); +DEFINE_DOMAIN(cork); typedef struct event { resource r; void (*hook)(void *); void *data; node n; /* Internal link */ + struct event_list *list; /* List where this event is put in */ + struct event_cork *cork; /* Event execution limiter */ + node cork_node; } event; -typedef list event_list; +typedef struct event_list { + list events; + pool *pool; + struct birdloop *loop; + DOMAIN(event) lock; +} event_list; + +struct event_cork { + DOMAIN(cork) lock; + u32 count; + list events; +}; extern event_list global_event_list; extern event_list global_work_list; event *ev_new(pool *); void ev_run(event *); -#define ev_init_list(el) init_list(el) -void ev_enqueue(event_list *, event *); -void ev_schedule(event *); -void ev_schedule_work(event *); + +static inline void ev_init_list(event_list *el, struct birdloop *loop, const char *name) +{ + init_list(&el->events); + el->loop = loop; + el->lock = DOMAIN_NEW(event, name); +} + +static inline void ev_init_cork(struct event_cork *ec, const char *name) +{ + init_list(&ec->events); + ec->lock = DOMAIN_NEW(cork, name); + ec->count = 0; +}; + +void ev_send(event_list *, event *); +#define ev_send_loop(l, e) ev_send(birdloop_event_list((l)), (e)) + +#define ev_schedule(e) ({ ASSERT_THE_BIRD_LOCKED; if (!ev_active((e))) ev_send(&global_event_list, (e)); }) +#define ev_schedule_work(e) ({ ASSERT_THE_BIRD_LOCKED; if (!ev_active((e))) ev_send(&global_work_list, (e)); }) + void ev_postpone(event *); int ev_run_list(event_list *); int ev_run_list_limited(event_list *, uint); +#define LEGACY_EVENT_LIST(l) (((l) == &global_event_list) || ((l) == &global_work_list)) + +void ev_cork(struct event_cork *); +void ev_uncork(struct event_cork *); + +static inline u32 ev_corked(struct event_cork *ec) +{ + if (!ec) + return 0; + + LOCK_DOMAIN(cork, ec->lock); + u32 out = ec->count; + UNLOCK_DOMAIN(cork, ec->lock); + return out; +} + +_Bool birdloop_inside(struct birdloop *loop); + static inline int ev_active(event *e) { - return e->n.next != NULL; + if (e->list == NULL) + return 0; + + ASSERT_DIE(birdloop_inside(e->list->loop)); + return enlisted(&e->n); } static inline event* diff --git a/lib/event_test.c b/lib/event_test.c index e1215bba..f8bb303b 100644 --- a/lib/event_test.c +++ b/lib/event_test.c @@ -48,14 +48,17 @@ init_event_check_points(void) event_check_points[i] = 0; } +void resource_sys_init(void); + static int t_ev_run_list(void) { int i; + resource_sys_init(); resource_init(); + birdloop_init(); olock_init(); - timer_init(); io_init(); rt_init(); if_init(); @@ -82,7 +85,9 @@ main(int argc, char *argv[]) { bt_init(argc, argv); + the_bird_lock(); bt_test_suite(t_ev_run_list, "Schedule and run 3 events in right order."); + the_bird_unlock(); return bt_exit_value(); } diff --git a/lib/flowspec_test.c b/lib/flowspec_test.c index ed4afe51..518b8fc7 100644 --- a/lib/flowspec_test.c +++ b/lib/flowspec_test.c @@ -8,6 +8,7 @@ #include "test/birdtest.h" #include "lib/flowspec.h" +#include "lib/io-loop.h" #define NET_ADDR_FLOW4_(what,prefix,pxlen,data_) \ do \ @@ -446,8 +447,6 @@ t_validation6(void) static int t_builder4(void) { - resource_init(); - struct flow_builder *fb = flow_builder_init(&root_pool); linpool *lp = lp_new_default(&root_pool); @@ -529,7 +528,6 @@ t_builder6(void) { net_addr_ip6 ip; - resource_init(); linpool *lp = lp_new_default(&root_pool); struct flow_builder *fb = flow_builder_init(&root_pool); fb->ipv6 = 1; @@ -666,10 +664,16 @@ t_formatting6(void) return 1; } +void resource_sys_init(void); + int main(int argc, char *argv[]) { bt_init(argc, argv); + resource_sys_init(); + resource_init(); + the_bird_lock(); + birdloop_init(); bt_test_suite(t_read_length, "Testing get NLRI length"); bt_test_suite(t_write_length, "Testing set NLRI length"); @@ -685,5 +689,6 @@ main(int argc, char *argv[]) bt_test_suite(t_formatting4, "Formatting Flow Specification (IPv4) into text representation"); bt_test_suite(t_formatting6, "Formatting Flow Specification (IPv6) into text representation"); + the_bird_unlock(); return bt_exit_value(); } @@ -215,6 +215,12 @@ mem_hash_mix(u64 *h, const void *p, uint s) *h = *h * multiplier + pp[i]; } +static inline void +mem_hash_mix_num(u64 *h, u64 val) +{ + mem_hash_mix(h, &val, sizeof(val)); +} + static inline uint mem_hash_value(u64 *h) { diff --git a/lib/hash_test.c b/lib/hash_test.c index 59beb7c0..7ef54662 100644 --- a/lib/hash_test.c +++ b/lib/hash_test.c @@ -9,7 +9,9 @@ #undef LOCAL_DEBUG #include "test/birdtest.h" +#include "test/bt-utils.h" +#include "lib/io-loop.h" #include "lib/hash.h" struct test_node { @@ -61,8 +63,7 @@ dump_nodes(void) static void init_hash_(uint order) { - resource_init(); - my_pool = rp_new(&root_pool, "Test pool"); + my_pool = rp_new(&root_pool, &main_birdloop, "Test pool"); HASH_INIT(hash, my_pool, order); @@ -290,6 +291,7 @@ int main(int argc, char *argv[]) { bt_init(argc, argv); + bt_bird_init(); bt_test_suite(t_insert_find, "HASH_INSERT and HASH_FIND"); bt_test_suite(t_insert_find_random, "HASH_INSERT pseudo-random keys and HASH_FIND"); diff --git a/lib/io-loop.h b/lib/io-loop.h new file mode 100644 index 00000000..386a31d5 --- /dev/null +++ b/lib/io-loop.h @@ -0,0 +1,57 @@ +/* + * BIRD -- I/O and event loop + * + * Can be freely distributed and used under the terms of the GNU GPL. + */ + +#ifndef _BIRD_IO_LOOP_H_ +#define _BIRD_IO_LOOP_H_ + +#include "nest/bird.h" +#include "lib/lists.h" +#include "lib/locking.h" +#include "lib/resource.h" +#include "lib/event.h" +#include "lib/socket.h" + +void sk_start(sock *s); +void sk_stop(sock *s); +void sk_reloop(sock *s, struct birdloop *loop); + +extern struct birdloop main_birdloop; + +/* Start a new birdloop owned by given pool and domain. + * The loop allocates its internal pool for local allocations + * which is freed when the loop itself is stopped. */ +struct birdloop *birdloop_new(pool *p, uint order, const char *name); + +/* Stop the loop. At the end, the @stopped callback is called with locked + * parent to finish cleanup. The loop then frees itself together with its pool. */ +void birdloop_stop(struct birdloop *loop, void (*stopped)(void *data), void *data); +void birdloop_stop_self(struct birdloop *loop, void (*stopped)(void *data), void *data); + +/* Get birdloop's event list */ +event_list *birdloop_event_list(struct birdloop *loop); + +/* Get birdloop's time heap */ +struct timeloop *birdloop_time_loop(struct birdloop *loop); + +/* Get birdloop's resource pool */ +pool *birdloop_pool(struct birdloop *loop); + +/* Enter and exit the birdloop */ +void birdloop_enter(struct birdloop *loop); +void birdloop_leave(struct birdloop *loop); + +_Bool birdloop_inside(struct birdloop *loop); + +void birdloop_mask_wakeups(struct birdloop *loop); +void birdloop_unmask_wakeups(struct birdloop *loop); + +void birdloop_link(struct birdloop *loop); +void birdloop_unlink(struct birdloop *loop); + +void birdloop_ping(struct birdloop *loop); + +void birdloop_init(void); +#endif /* _BIRD_IO_LOOP_H_ */ diff --git a/lib/lists.c b/lib/lists.c index 200576cf..dc2e4cbb 100644 --- a/lib/lists.c +++ b/lib/lists.c @@ -26,7 +26,7 @@ #define _BIRD_LISTS_C_ -#include "nest/bird.h" +#include "lib/birdlib.h" #include "lib/lists.h" LIST_INLINE int @@ -35,11 +35,12 @@ check_list(list *l, node *n) if (!l) { ASSERT_DIE(n); - ASSERT_DIE(n->prev); - do { n = n->prev; } while (n->prev); + node *nn = n; + while (nn->prev) + nn = nn->prev; - l = SKIP_BACK(list, head_node, n); + l = SKIP_BACK(list, head_node, nn); } int seen = 0; @@ -60,7 +61,7 @@ check_list(list *l, node *n) } ASSERT_DIE(cur == &(l->tail_node)); - ASSERT_DIE(!n || (seen == 1)); + ASSERT_DIE(!n || (seen == 1) || (n == &l->head_node) || (n == &l->tail_node)); return 1; } @@ -109,6 +110,15 @@ add_head(list *l, node *n) l->head = n; } +LIST_INLINE void +self_link(node *n) +{ + ASSUME(n->prev == NULL); + ASSUME(n->next == NULL); + + n->prev = n->next = n; +} + /** * insert_node - insert a node to a list * @n: a new list node @@ -120,7 +130,7 @@ add_head(list *l, node *n) LIST_INLINE void insert_node(node *n, node *after) { - EXPENSIVE_CHECK(check_list(l, after)); + EXPENSIVE_CHECK(check_list(NULL, after)); ASSUME(n->prev == NULL); ASSUME(n->next == NULL); @@ -141,7 +151,7 @@ insert_node(node *n, node *after) LIST_INLINE void rem_node(node *n) { - EXPENSIVE_CHECK(check_list(NULL, n)); + EXPENSIVE_CHECK((n == n->prev) && (n == n->next) || check_list(NULL, n)); node *z = n->prev; node *x = n->next; diff --git a/lib/lists.h b/lib/lists.h index 479f4ed1..dc49ec8a 100644 --- a/lib/lists.h +++ b/lib/lists.h @@ -68,6 +68,18 @@ typedef union list { /* In fact two overlayed nodes */ #define EMPTY_LIST(list) (!(list).head->next) +static inline _Bool +enlisted(node *n) +{ + switch ((!!n->next) + (!!n->prev)) + { + case 0: return 0; + case 2: return 1; + case 1: bug("Garbled event list node"); + } + + bug("Maths is broken. And you should see a new heaven and a new earth: for the first heaven and the first earth had been passed away."); +} #ifndef _BIRD_LISTS_C_ #define LIST_INLINE static inline @@ -78,6 +90,7 @@ typedef union list { /* In fact two overlayed nodes */ #define LIST_INLINE void add_tail(list *, node *); void add_head(list *, node *); +void self_link(node *); void rem_node(node *); void add_tail_list(list *, list *); void init_list(list *); diff --git a/lib/locking.h b/lib/locking.h new file mode 100644 index 00000000..1a8bdcd4 --- /dev/null +++ b/lib/locking.h @@ -0,0 +1,66 @@ +/* + * BIRD Library -- Locking + * + * (c) 2020--2021 Maria Matejka <mq@jmq.cz> + * + * Can be freely distributed and used under the terms of the GNU GPL. + */ + +#ifndef _BIRD_LOCKING_H_ +#define _BIRD_LOCKING_H_ + +struct domain_generic; + +/* Here define the global lock order; first to last. */ +struct lock_order { + struct domain_generic *the_bird; + struct domain_generic *proto; + struct domain_generic *rtable; + struct domain_generic *attrs; + struct domain_generic *cork; + struct domain_generic *event; + struct domain_generic *resource; +}; + +extern _Thread_local struct lock_order locking_stack; +extern _Thread_local struct domain_generic **last_locked; + +#define DOMAIN(type) struct domain__##type +#define DEFINE_DOMAIN(type) DOMAIN(type) { struct domain_generic *type; } +#define DOMAIN_ORDER(type) OFFSETOF(struct lock_order, type) + +#define DOMAIN_NEW(type, name) (DOMAIN(type)) { .type = domain_new(name, DOMAIN_ORDER(type)) } +struct domain_generic *domain_new(const char *name, uint order); + +#define DOMAIN_FREE(type, d) domain_free((d).type) +void domain_free(struct domain_generic *); + +#define DOMAIN_NULL(type) (DOMAIN(type)) {} + +#define LOCK_DOMAIN(type, d) do_lock(((d).type), &(locking_stack.type)) +#define UNLOCK_DOMAIN(type, d) do_unlock(((d).type), &(locking_stack.type)) + +#define DOMAIN_IS_LOCKED(type, d) (((d).type) == (locking_stack.type)) +#define DG_IS_LOCKED(d) ((d) == *(DG_LSP(d))) + +/* Internal for locking */ +void do_lock(struct domain_generic *dg, struct domain_generic **lsp); +void do_unlock(struct domain_generic *dg, struct domain_generic **lsp); + +uint dg_order(struct domain_generic *dg); + +#define DG_LSP(d) ((struct domain_generic **) (((void *) &locking_stack) + dg_order(d))) +#define DG_LOCK(d) do_lock(d, DG_LSP(d)) +#define DG_UNLOCK(d) do_unlock(d, DG_LSP(d)) + +/* Use with care. To be removed in near future. */ +DEFINE_DOMAIN(the_bird); +extern DOMAIN(the_bird) the_bird_domain; + +#define the_bird_lock() LOCK_DOMAIN(the_bird, the_bird_domain) +#define the_bird_unlock() UNLOCK_DOMAIN(the_bird, the_bird_domain) +#define the_bird_locked() DOMAIN_IS_LOCKED(the_bird, the_bird_domain) + +#define ASSERT_THE_BIRD_LOCKED ({ if (!the_bird_locked()) bug("The BIRD lock must be locked here: %s:%d", __FILE__, __LINE__); }) + +#endif diff --git a/lib/mempool.c b/lib/mempool.c index 90d7c774..6d157e5c 100644 --- a/lib/mempool.c +++ b/lib/mempool.c @@ -37,9 +37,10 @@ const int lp_chunk_size = sizeof(struct lp_chunk); struct linpool { resource r; byte *ptr, *end; + pool *p; struct lp_chunk *first, *current; /* Normal (reusable) chunks */ struct lp_chunk *first_large; /* Large chunks */ - uint chunk_size, threshold, total, total_large; + uint chunk_size, threshold, total:31, use_pages:1, total_large; }; static void lp_free(resource *); @@ -69,6 +70,13 @@ linpool *lp_new(pool *p, uint blk) { linpool *m = ralloc(p, &lp_class); + m->p = p; + if (!blk) + { + m->use_pages = 1; + blk = page_size - lp_chunk_size; + } + m->chunk_size = blk; m->threshold = 3*blk/4; return m; @@ -121,7 +129,11 @@ lp_alloc(linpool *m, uint size) else { /* Need to allocate a new chunk */ - c = xmalloc(sizeof(struct lp_chunk) + m->chunk_size); + if (m->use_pages) + c = alloc_page(); + else + c = xmalloc(sizeof(struct lp_chunk) + m->chunk_size); + m->total += m->chunk_size; c->next = NULL; c->size = m->chunk_size; @@ -258,7 +270,10 @@ lp_free(resource *r) for(d=m->first; d; d = c) { c = d->next; - xfree(d); + if (m->use_pages) + free_page(d); + else + xfree(d); } for(d=m->first_large; d; d = c) { diff --git a/lib/rcu.c b/lib/rcu.c new file mode 100644 index 00000000..69f3442f --- /dev/null +++ b/lib/rcu.c @@ -0,0 +1,79 @@ +/* + * BIRD Library -- Read-Copy-Update Basic Operations + * + * (c) 2021 Maria Matejka <mq@jmq.cz> + * (c) 2021 CZ.NIC z.s.p.o. + * + * Can be freely distributed and used under the terms of the GNU GPL. + * Note: all the relevant patents shall be expired. + * + * Using the Supplementary Material for User-Level Implementations of Read-Copy-Update + * by Matthieu Desnoyers, Paul E. McKenney, Alan S. Stern, Michel R. Dagenais and Jonathan Walpole + * obtained from https://www.efficios.com/pub/rcu/urcu-supp-accepted.pdf + */ + +#include "lib/rcu.h" +#include "lib/coro.h" +#include "lib/locking.h" + +_Atomic uint rcu_gp_ctl = RCU_NEST_CNT; +_Thread_local struct rcu_coro *this_rcu_coro = NULL; + +static list rcu_coro_list; + +static struct rcu_coro main_rcu_coro; + +DEFINE_DOMAIN(resource); +static DOMAIN(resource) rcu_domain; + +static int +rcu_gp_ongoing(_Atomic uint *ctl) +{ + uint val = atomic_load(ctl); + return (val & RCU_NEST_CNT) && ((val ^ rcu_gp_ctl) & RCU_GP_PHASE); +} + +static void +update_counter_and_wait(void) +{ + atomic_fetch_xor(&rcu_gp_ctl, RCU_GP_PHASE); + struct rcu_coro *rc; + WALK_LIST(rc, rcu_coro_list) + while (rcu_gp_ongoing(&rc->ctl)) + coro_yield(); +} + +void +synchronize_rcu(void) +{ + LOCK_DOMAIN(resource, rcu_domain); + update_counter_and_wait(); + update_counter_and_wait(); + UNLOCK_DOMAIN(resource, rcu_domain); +} + +void +rcu_coro_start(struct rcu_coro *rc) +{ + LOCK_DOMAIN(resource, rcu_domain); + add_tail(&rcu_coro_list, &rc->n); + this_rcu_coro = rc; + UNLOCK_DOMAIN(resource, rcu_domain); +} + +void +rcu_coro_stop(struct rcu_coro *rc) +{ + LOCK_DOMAIN(resource, rcu_domain); + this_rcu_coro = NULL; + rem_node(&rc->n); + UNLOCK_DOMAIN(resource, rcu_domain); +} + +void +rcu_init(void) +{ + rcu_domain = DOMAIN_NEW(resource, "Read-Copy-Update"); + init_list(&rcu_coro_list); + rcu_coro_start(&main_rcu_coro); +} diff --git a/lib/rcu.h b/lib/rcu.h new file mode 100644 index 00000000..ac8fc9ce --- /dev/null +++ b/lib/rcu.h @@ -0,0 +1,55 @@ +/* + * BIRD Library -- Read-Copy-Update Basic Operations + * + * (c) 2021 Maria Matejka <mq@jmq.cz> + * (c) 2021 CZ.NIC z.s.p.o. + * + * Can be freely distributed and used under the terms of the GNU GPL. + * Note: all the relevant patents shall be expired. + */ + +#ifndef _BIRD_RCU_H_ +#define _BIRD_RCU_H_ + +#include "lib/birdlib.h" +#include "lib/lists.h" +#include <stdatomic.h> + +#define RCU_GP_PHASE 0x100000 +#define RCU_NEST_MASK 0x0fffff +#define RCU_NEST_CNT 0x000001 + +extern _Atomic uint rcu_gp_ctl; + +struct rcu_coro { + node n; + _Atomic uint ctl; +}; + +extern _Thread_local struct rcu_coro *this_rcu_coro; + +static inline void rcu_read_lock(void) +{ + uint cmp = atomic_load_explicit(&this_rcu_coro->ctl, memory_order_acquire); + + if (cmp & RCU_NEST_MASK) + atomic_store_explicit(&this_rcu_coro->ctl, cmp + RCU_NEST_CNT, memory_order_relaxed); + else + atomic_store(&this_rcu_coro->ctl, atomic_load_explicit(&rcu_gp_ctl, memory_order_acquire)); +} + +static inline void rcu_read_unlock(void) +{ + atomic_fetch_sub(&this_rcu_coro->ctl, RCU_NEST_CNT); +} + +void synchronize_rcu(void); + +/* Registering and unregistering a coroutine. To be called from coroutine implementation */ +void rcu_coro_start(struct rcu_coro *); +void rcu_coro_stop(struct rcu_coro *); + +/* Run this from resource init */ +void rcu_init(void); + +#endif diff --git a/lib/resource.c b/lib/resource.c index 5d4c7780..9a4c7717 100644 --- a/lib/resource.c +++ b/lib/resource.c @@ -14,6 +14,8 @@ #include "nest/bird.h" #include "lib/resource.h" #include "lib/string.h" +#include "lib/rcu.h" +#include "lib/io-loop.h" /** * DOC: Resource pools @@ -29,12 +31,6 @@ * is freed upon shutdown of the module. */ -struct pool { - resource r; - list inside; - const char *name; -}; - static void pool_dump(resource *); static void pool_free(resource *); static resource *pool_lookup(resource *, unsigned long); @@ -56,26 +52,39 @@ static int indent; /** * rp_new - create a resource pool * @p: parent pool + * @l: loop to assign * @name: pool name (to be included in debugging dumps) * * rp_new() creates a new resource pool inside the specified * parent pool. */ pool * -rp_new(pool *p, const char *name) +rp_new(pool *p, struct birdloop *loop, const char *name) { + ASSERT_DIE(birdloop_inside(p->loop)); + ASSERT_DIE(birdloop_inside(loop)); + pool *z = ralloc(p, &pool_class); + z->loop = loop; z->name = name; init_list(&z->inside); return z; } +_Thread_local static pool *pool_parent = NULL; + static void pool_free(resource *P) { + ASSERT_DIE(pool_parent); + pool *p = (pool *) P; - resource *r, *rr; + ASSERT_DIE(birdloop_inside(p->loop)); + + pool *parent = pool_parent; + pool_parent = p; + resource *r, *rr; r = HEAD(p->inside); while (rr = (resource *) r->n.next) { @@ -83,14 +92,25 @@ pool_free(resource *P) xfree(r); r = rr; } + + pool_parent = parent; +} + +void +rp_free(pool *p, pool *parent) +{ + ASSERT_DIE(pool_parent == NULL); + pool_parent = parent; + rfree(p); + ASSERT_DIE(pool_parent == parent); + pool_parent = NULL; } static void -pool_dump(resource *P) +pool_dump_locked(pool *p) { - pool *p = (pool *) P; resource *r; - + debug("%s\n", p->name); indent += 3; WALK_LIST(r, p->inside) @@ -98,10 +118,47 @@ pool_dump(resource *P) indent -= 3; } -static struct resmem -pool_memsize(resource *P) +static void +pool_dump(resource *P) { pool *p = (pool *) P; + + if (p->loop != pool_parent->loop) + birdloop_enter(p->loop); + + pool *parent = pool_parent; + pool_parent = p; + + pool_dump_locked(p); + + pool_parent = parent; + + if (p->loop != pool_parent->loop) + birdloop_leave(p->loop); +} + +void +rp_dump(pool *p) +{ + int inside = birdloop_inside(p->loop); + if (!inside) + birdloop_enter(p->loop); + + ASSERT_DIE(pool_parent == NULL); + pool_parent = p; + + pool_dump_locked(p); + + ASSERT_DIE(pool_parent == p); + pool_parent = NULL; + + if (!inside) + birdloop_leave(p->loop); +} + +static struct resmem +pool_memsize_locked(pool *p) +{ resource *r; struct resmem sum = { .effective = 0, @@ -118,6 +175,46 @@ pool_memsize(resource *P) return sum; } +static struct resmem +pool_memsize(resource *P) +{ + pool *p = (pool *) P; + + pool *parent = pool_parent; + pool_parent = p; + + if (p->loop != parent->loop) + birdloop_enter(p->loop); + + struct resmem sum = pool_memsize_locked(p); + + if (p->loop != parent->loop) + birdloop_leave(p->loop); + + pool_parent = parent; + + return sum; +} + +struct resmem +rp_memsize(pool *p) +{ + int inside = birdloop_inside(p->loop); + if (!inside) + birdloop_enter(p->loop); + + ASSERT_DIE(pool_parent == NULL); + pool_parent = p; + struct resmem sum = pool_memsize_locked(p); + ASSERT_DIE(pool_parent == p); + pool_parent = NULL; + + if (!inside) + birdloop_leave(p->loop); + + return sum; +} + static resource * pool_lookup(resource *P, unsigned long a) { @@ -228,12 +325,15 @@ rmemsize(void *res) void * ralloc(pool *p, struct resclass *c) { + ASSERT_DIE(p); + ASSERT_DIE(birdloop_inside(p->loop)); + resource *r = xmalloc(c->size); bzero(r, c->size); r->class = c; - if (p) - add_tail(&p->inside, &r->n); + add_tail(&p->inside, &r->n); + return r; } @@ -270,6 +370,8 @@ rlookup(unsigned long a) void resource_init(void) { + rcu_init(); + root_pool.r.class = &pool_class; root_pool.name = "Root"; init_list(&root_pool.inside); @@ -440,7 +542,6 @@ mb_free(void *m) } - #define STEP_UP(x) ((x) + (x)/2 + 4) void diff --git a/lib/resource.h b/lib/resource.h index 9ec41ed8..b80ff8c6 100644 --- a/lib/resource.h +++ b/lib/resource.h @@ -40,10 +40,16 @@ struct resclass { /* Generic resource manipulation */ -typedef struct pool pool; +typedef struct pool { + resource r; + list inside; + struct pool_pages *pages; + struct birdloop *loop; + const char *name; +} pool; void resource_init(void); -pool *rp_new(pool *, const char *); /* Create new pool */ + void rfree(void *); /* Free single resource */ void rdump(void *); /* Dump to debug output */ struct resmem rmemsize(void *res); /* Return size of memory used by the resource */ @@ -52,6 +58,11 @@ void rmove(void *, pool *); /* Move to a different pool */ void *ralloc(pool *, struct resclass *); +pool *rp_new(pool *, struct birdloop *loop, const char *); /* Create new pool */ +void rp_free(pool *p, pool *parent); /* Free parent pool */ +struct resmem rp_memsize(pool *p); /* Return size of memory used by the pool */ +void rp_dump(pool *p); /* Dump pool to debug output */ + extern pool root_pool; /* Normal memory blocks */ @@ -82,7 +93,7 @@ void lp_restore(linpool *m, lp_state *p); /* Restore state */ extern const int lp_chunk_size; #define LP_GAS 1024 #define LP_GOOD_SIZE(x) (((x + LP_GAS - 1) & (~(LP_GAS - 1))) - lp_chunk_size) -#define lp_new_default(p) lp_new(p, LP_GOOD_SIZE(LP_GAS*4)) +#define lp_new_default(p) lp_new(p, 0) /* Slabs */ @@ -100,11 +111,12 @@ void sl_free(slab *, void *); void buffer_realloc(void **buf, unsigned *size, unsigned need, unsigned item_size); +extern long page_size; + /* Allocator of whole pages; for use in slabs and other high-level allocators. */ -u64 get_page_size(void); void *alloc_page(void); void free_page(void *); -extern uint pages_kept; +#define PAGE_HEAD(x) ((void *) (((intptr_t) (x)) & ~(page_size-1))) #ifdef HAVE_LIBDMALLOC /* @@ -155,6 +155,7 @@ slab_memsize(resource *r) struct slab { resource r; + pool *p; uint obj_size, head_size, head_bitfield_len; uint objs_per_slab, num_empty_heads, data_size; list empty_heads, partial_heads, full_heads; @@ -180,7 +181,7 @@ struct sl_alignment { /* Magic structure for testing of alignment */ int x[0]; }; -#define SL_GET_HEAD(x) ((struct sl_head *) (((uintptr_t) (x)) & ~(get_page_size()-1))) +#define SL_GET_HEAD(x) ((struct sl_head *) PAGE_HEAD(x)) /** * sl_new - create a new Slab @@ -194,6 +195,7 @@ slab * sl_new(pool *p, uint size) { slab *s = ralloc(p, &sl_class); + s->p = p; uint align = sizeof(struct sl_alignment); if (align < sizeof(int)) align = sizeof(int); @@ -202,7 +204,6 @@ sl_new(pool *p, uint size) s->obj_size = size; s->head_size = sizeof(struct sl_head); - u64 page_size = get_page_size(); do { s->objs_per_slab = (page_size - s->head_size) / size; @@ -272,6 +273,9 @@ no_partial: goto okay; } h = alloc_page(); +#ifdef POISON + memset(h, 0xba, page_size); +#endif ASSERT_DIE(SL_GET_HEAD(h) == h); memset(h, 0, s->head_size); add_head(&s->partial_heads, &h->n); @@ -327,7 +331,12 @@ sl_free(slab *s, void *oo) { rem_node(&h->n); if (s->num_empty_heads >= MAX_EMPTY_HEADS) + { +#ifdef POISON + memset(h, 0xde, page_size); +#endif free_page(h); + } else { add_head(&s->empty_heads, &h->n); @@ -391,7 +400,7 @@ slab_memsize(resource *r) return (struct resmem) { .effective = eff, - .overhead = ALLOC_OVERHEAD + sizeof(struct slab) + heads * get_page_size() - eff, + .overhead = ALLOC_OVERHEAD + sizeof(struct slab) + heads * page_size - eff, }; } @@ -402,10 +411,10 @@ slab_lookup(resource *r, unsigned long a) struct sl_head *h; WALK_LIST(h, s->partial_heads) - if ((unsigned long) h < a && (unsigned long) h + get_page_size() < a) + if ((unsigned long) h < a && (unsigned long) h + page_size < a) return r; WALK_LIST(h, s->full_heads) - if ((unsigned long) h < a && (unsigned long) h + get_page_size() < a) + if ((unsigned long) h < a && (unsigned long) h + page_size < a) return r; return NULL; } diff --git a/lib/socket.h b/lib/socket.h index 96fedeeb..ff07660f 100644 --- a/lib/socket.h +++ b/lib/socket.h @@ -12,6 +12,7 @@ #include <errno.h> #include "lib/resource.h" +#include "lib/event.h" #ifdef HAVE_LIBSSH #define LIBSSH_LEGACY_0_4 #include <libssh/libssh.h> @@ -56,6 +57,7 @@ typedef struct birdsock { uint fast_rx; /* RX has higher priority in event loop */ uint rbsize; int (*rx_hook)(struct birdsock *, uint size); /* NULL=receiving turned off, returns 1 to clear rx buffer */ + struct event_cork *cork; /* Cork to temporarily stop receiving data */ byte *tbuf, *tpos; /* NULL=allocate automatically */ byte *ttx; /* Internal */ @@ -79,6 +81,7 @@ typedef struct birdsock { const char *password; /* Password for MD5 authentication */ const char *err; /* Error message */ struct ssh_sock *ssh; /* Used in SK_SSH */ + struct event reloop; /* Reloop event */ } sock; sock *sock_new(pool *); /* Allocate new socket */ @@ -128,6 +131,7 @@ extern int sk_priority_control; /* Suggested priority for control traffic, shou #define SKF_TRUNCATED 0x200 /* Received packet was truncated, set by IO layer */ #define SKF_HDRINCL 0x400 /* Used internally */ #define SKF_PKTINFO 0x800 /* Used internally */ +#define SKF_PASSIVE_THREAD 0x1000 /* Child sockets used in thread, do not add to main loop */ /* * Socket types SA SP DA DP IF TTL SendTo (?=may, -=must not, *=must) diff --git a/lib/timer.c b/lib/timer.c index 381163d0..eb7ea690 100644 --- a/lib/timer.c +++ b/lib/timer.c @@ -32,61 +32,18 @@ #include "nest/bird.h" +#include "lib/coro.h" #include "lib/heap.h" #include "lib/resource.h" #include "lib/timer.h" - -struct timeloop main_timeloop; - - -#ifdef USE_PTHREADS - #include <pthread.h> -/* Data accessed and modified from proto/bfd/io.c */ -pthread_key_t current_time_key; - -static inline struct timeloop * -timeloop_current(void) -{ - return pthread_getspecific(current_time_key); -} - -static inline void -timeloop_init_current(void) -{ - pthread_key_create(¤t_time_key, NULL); - pthread_setspecific(current_time_key, &main_timeloop); -} +_Atomic btime last_time; +_Atomic btime real_time; void wakeup_kick_current(void); -#else - -/* Just use main timelooop */ -static inline struct timeloop * timeloop_current(void) { return &main_timeloop; } -static inline void timeloop_init_current(void) { } - -#endif - -btime -current_time(void) -{ - return timeloop_current()->last_time; -} - -btime -current_real_time(void) -{ - struct timeloop *loop = timeloop_current(); - - if (!loop->real_time) - times_update_real_time(loop); - - return loop->real_time; -} - #define TIMER_LESS(a,b) ((a)->expires < (b)->expires) #define TIMER_SWAP(heap,a,b,t) (t = heap[a], heap[a] = heap[b], heap[b] = t, \ @@ -112,7 +69,7 @@ tm_dump(resource *r) if (t->recurrent) debug("recur %d, ", t->recurrent); if (t->expires) - debug("expires in %d ms)\n", (t->expires - current_time()) TO_MS); + debug("in loop %p expires in %d ms)\n", t->loop, (t->expires - current_time()) TO_MS); else debug("inactive)\n"); } @@ -135,41 +92,40 @@ tm_new(pool *p) return t; } -void -tm_set(timer *t, btime when) +static void +tm_set_in_tl(timer *t, btime when, struct timeloop *local_timeloop) { - struct timeloop *loop = timeloop_current(); - uint tc = timers_count(loop); + uint tc = timers_count(local_timeloop); if (!t->expires) { t->index = ++tc; t->expires = when; - BUFFER_PUSH(loop->timers) = t; - HEAP_INSERT(loop->timers.data, tc, timer *, TIMER_LESS, TIMER_SWAP); + BUFFER_PUSH(local_timeloop->timers) = t; + HEAP_INSERT(local_timeloop->timers.data, tc, timer *, TIMER_LESS, TIMER_SWAP); } else if (t->expires < when) { t->expires = when; - HEAP_INCREASE(loop->timers.data, tc, timer *, TIMER_LESS, TIMER_SWAP, t->index); + HEAP_INCREASE(local_timeloop->timers.data, tc, timer *, TIMER_LESS, TIMER_SWAP, t->index); } else if (t->expires > when) { t->expires = when; - HEAP_DECREASE(loop->timers.data, tc, timer *, TIMER_LESS, TIMER_SWAP, t->index); + HEAP_DECREASE(local_timeloop->timers.data, tc, timer *, TIMER_LESS, TIMER_SWAP, t->index); } -#ifdef CONFIG_BFD - /* Hack to notify BFD loops */ - if ((loop != &main_timeloop) && (t->index == 1)) - wakeup_kick_current(); -#endif + t->loop = local_timeloop; + + if ((t->index == 1) && (local_timeloop->coro != this_coro)) + birdloop_ping(local_timeloop->loop); } void -tm_start(timer *t, btime after) +tm_set_in(timer *t, btime when, struct birdloop *loop) { - tm_set(t, current_time() + MAX(after, 0)); + ASSERT_DIE(birdloop_inside(loop)); + tm_set_in_tl(t, when, birdloop_time_loop(loop)); } void @@ -178,20 +134,22 @@ tm_stop(timer *t) if (!t->expires) return; - struct timeloop *loop = timeloop_current(); - uint tc = timers_count(loop); + TLOCK_TIMER_ASSERT(t->loop); - HEAP_DELETE(loop->timers.data, tc, timer *, TIMER_LESS, TIMER_SWAP, t->index); - BUFFER_POP(loop->timers); + uint tc = timers_count(t->loop); + + HEAP_DELETE(t->loop->timers.data, tc, timer *, TIMER_LESS, TIMER_SWAP, t->index); + BUFFER_POP(t->loop->timers); t->index = -1; t->expires = 0; + t->loop = NULL; } void timers_init(struct timeloop *loop, pool *p) { - times_init(loop); + TLOCK_TIMER_ASSERT(loop); BUFFER_INIT(loop->timers, p, 4); BUFFER_PUSH(loop->timers) = NULL; @@ -200,13 +158,15 @@ timers_init(struct timeloop *loop, pool *p) void io_log_event(void *hook, void *data); void -timers_fire(struct timeloop *loop) +timers_fire(struct timeloop *loop, int io_log) { + TLOCK_TIMER_ASSERT(loop); + btime base_time; timer *t; - times_update(loop); - base_time = loop->last_time; + times_update(); + base_time = current_time(); while (t = timers_first(loop)) { @@ -217,32 +177,25 @@ timers_fire(struct timeloop *loop) { btime when = t->expires + t->recurrent; - if (when <= loop->last_time) - when = loop->last_time + t->recurrent; + if (when <= base_time) + when = base_time + t->recurrent; if (t->randomize) when += random() % (t->randomize + 1); - tm_set(t, when); + tm_set_in_tl(t, when, loop); } else tm_stop(t); /* This is ugly hack, we want to log just timers executed from the main I/O loop */ - if (loop == &main_timeloop) + if (io_log) io_log_event(t->hook, t->data); t->hook(t); } } -void -timer_init(void) -{ - timers_init(&main_timeloop, &root_pool); - timeloop_init_current(); -} - /** * tm_parse_time - parse a date and time diff --git a/lib/timer.h b/lib/timer.h index c5ea430c..04544ace 100644 --- a/lib/timer.h +++ b/lib/timer.h @@ -12,8 +12,14 @@ #include "nest/bird.h" #include "lib/buffer.h" +#include "lib/io-loop.h" +#include "lib/locking.h" #include "lib/resource.h" +#include <stdatomic.h> + +extern _Atomic btime last_time; +extern _Atomic btime real_time; typedef struct timer { @@ -25,36 +31,43 @@ typedef struct timer uint randomize; /* Amount of randomization */ uint recurrent; /* Timer recurrence */ + struct timeloop *loop; /* Loop where the timer is active */ + int index; } timer; struct timeloop { BUFFER_(timer *) timers; - btime last_time; - btime real_time; + struct domain_generic *domain; + struct birdloop *loop; + struct coroutine *coro; }; +#define TLOCK_TIMER_ASSERT(loop) ASSERT_DIE((loop)->domain && DG_IS_LOCKED((loop)->domain)) +#define TLOCK_LOCAL_ASSERT(loop) ASSERT_DIE(!(loop)->domain || DG_IS_LOCKED((loop)->domain)) + static inline uint timers_count(struct timeloop *loop) -{ return loop->timers.used - 1; } +{ TLOCK_TIMER_ASSERT(loop); return loop->timers.used - 1; } static inline timer *timers_first(struct timeloop *loop) -{ return (loop->timers.used > 1) ? loop->timers.data[1] : NULL; } +{ TLOCK_TIMER_ASSERT(loop); return (loop->timers.used > 1) ? loop->timers.data[1] : NULL; } -extern struct timeloop main_timeloop; - -btime current_time(void); -btime current_real_time(void); +#define current_time() atomic_load_explicit(&last_time, memory_order_acquire) +#define current_real_time() atomic_load_explicit(&real_time, memory_order_acquire) //#define now (current_time() TO_S) //#define now_real (current_real_time() TO_S) extern btime boot_time; timer *tm_new(pool *p); -void tm_set(timer *t, btime when); -void tm_start(timer *t, btime after); +#define tm_set(t, when) tm_set_in((t), (when), &main_birdloop) +#define tm_start(t, after) tm_start_in((t), (after), &main_birdloop) void tm_stop(timer *t); +void tm_set_in(timer *t, btime when, struct birdloop *loop); +#define tm_start_in(t, after, loop) tm_set_in((t), (current_time() + MAX_((after), 0)), loop) + static inline int tm_active(timer *t) { @@ -94,15 +107,11 @@ tm_start_max(timer *t, btime after) } /* In sysdep code */ -void times_init(struct timeloop *loop); -void times_update(struct timeloop *loop); -void times_update_real_time(struct timeloop *loop); +void times_update(void); /* For I/O loop */ void timers_init(struct timeloop *loop, pool *p); -void timers_fire(struct timeloop *loop); - -void timer_init(void); +void timers_fire(struct timeloop *loop, int io_log); struct timeformat { diff --git a/nest/a-path_test.c b/nest/a-path_test.c index 9ed0a786..2533dbae 100644 --- a/nest/a-path_test.c +++ b/nest/a-path_test.c @@ -12,6 +12,7 @@ #include "nest/route.h" #include "nest/attrs.h" #include "lib/resource.h" +#include "lib/io-loop.h" #define TESTS_NUM 30 #define AS_PATH_LENGTH 1000 @@ -23,8 +24,6 @@ static int t_as_path_match(void) { - resource_init(); - int round; for (round = 0; round < TESTS_NUM; round++) { @@ -70,8 +69,6 @@ t_as_path_match(void) static int t_path_format(void) { - resource_init(); - struct adata empty_as_path = {}; struct adata *as_path = &empty_as_path; struct linpool *lp = lp_new_default(&root_pool); @@ -116,8 +113,6 @@ count_asn_in_array(const u32 *array, u32 asn) static int t_path_include(void) { - resource_init(); - struct adata empty_as_path = {}; struct adata *as_path = &empty_as_path; struct linpool *lp = lp_new_default(&root_pool); @@ -161,8 +156,6 @@ t_path_include(void) static int t_as_path_converting(void) { - resource_init(); - struct adata empty_as_path = {}; struct adata *as_path = &empty_as_path; struct linpool *lp = lp_new_default(&root_pool); @@ -204,10 +197,18 @@ t_as_path_converting(void) } #endif +void resource_sys_init(void); +void io_init(void); + int main(int argc, char *argv[]) { bt_init(argc, argv); + resource_sys_init(); + resource_init(); + the_bird_lock(); + birdloop_init(); + io_init(); bt_test_suite(t_as_path_match, "Testing AS path matching and some a-path utilities."); bt_test_suite(t_path_format, "Testing formating as path into byte buffer"); diff --git a/nest/a-set_test.c b/nest/a-set_test.c index 96b6a727..f8f6e781 100644 --- a/nest/a-set_test.c +++ b/nest/a-set_test.c @@ -13,6 +13,7 @@ #include "nest/route.h" #include "nest/attrs.h" #include "lib/resource.h" +#include "lib/io-loop.h" #define SET_SIZE 10 static const struct adata *set_sequence; /* <0; SET_SIZE) */ @@ -71,7 +72,6 @@ t_set_int_contains(void) { int i; - resource_init(); generate_set_sequence(SET_TYPE_INT, SET_SIZE); bt_assert(int_set_get_size(set_sequence) == SET_SIZE); @@ -92,7 +92,6 @@ t_set_int_contains(void) static int t_set_int_union(void) { - resource_init(); generate_set_sequence(SET_TYPE_INT, SET_SIZE); const struct adata *set_union; @@ -111,7 +110,6 @@ t_set_int_union(void) static int t_set_int_format(void) { - resource_init(); generate_set_sequence(SET_TYPE_INT, SET_SIZE_FOR_FORMAT_OUTPUT); bt_assert(int_set_format(set_sequence, 0, 0, buf, BUFFER_SIZE) == 0); @@ -132,7 +130,6 @@ t_set_int_format(void) static int t_set_int_delete(void) { - resource_init(); generate_set_sequence(SET_TYPE_INT, SET_SIZE); const struct adata *deleting_sequence = set_sequence; @@ -160,7 +157,6 @@ t_set_ec_contains(void) { u32 i; - resource_init(); generate_set_sequence(SET_TYPE_EC, SET_SIZE); bt_assert(ec_set_get_size(set_sequence) == SET_SIZE); @@ -181,7 +177,6 @@ t_set_ec_contains(void) static int t_set_ec_union(void) { - resource_init(); generate_set_sequence(SET_TYPE_EC, SET_SIZE); const struct adata *set_union; @@ -200,8 +195,6 @@ t_set_ec_union(void) static int t_set_ec_format(void) { - resource_init(); - const struct adata empty_as_path = {}; set_sequence = set_sequence_same = set_sequence_higher = set_random = &empty_as_path; lp = lp_new_default(&root_pool); @@ -222,7 +215,6 @@ t_set_ec_format(void) static int t_set_ec_delete(void) { - resource_init(); generate_set_sequence(SET_TYPE_EC, SET_SIZE); const struct adata *deleting_sequence = set_sequence; @@ -240,10 +232,17 @@ t_set_ec_delete(void) return 1; } + +void resource_sys_init(void); + int main(int argc, char *argv[]) { bt_init(argc, argv); + resource_sys_init(); + resource_init(); + the_bird_lock(); + birdloop_init(); bt_test_suite(t_set_int_contains, "Testing sets of integers: contains, get_data"); bt_test_suite(t_set_int_format, "Testing sets of integers: format"); @@ -262,7 +262,7 @@ cli_command(struct cli *c) log(L_TRACE "CLI: %s", c->rx_buf); bzero(&f, sizeof(f)); f.mem = c->parser_pool; - f.pool = rp_new(c->pool, "Config"); + f.pool = rp_new(c->pool, &main_birdloop, "Config"); init_list(&f.symbols); cf_read_hook = cli_cmd_read_hook; cli_rh_pos = c->rx_buf; @@ -308,7 +308,7 @@ cli_event(void *data) cli * cli_new(void *priv) { - pool *p = rp_new(cli_pool, "CLI"); + pool *p = rp_new(cli_pool, &main_birdloop, "CLI"); cli *c = mb_alloc(p, sizeof(cli)); bzero(c, sizeof(cli)); @@ -413,7 +413,7 @@ cli_free(cli *c) c->cleanup(c); if (c == cmd_reconfig_stored_cli) cmd_reconfig_stored_cli = NULL; - rfree(c->pool); + rp_free(c->pool, &root_pool); } /** @@ -425,7 +425,7 @@ cli_free(cli *c) void cli_init(void) { - cli_pool = rp_new(&root_pool, "CLI"); + cli_pool = rp_new(&root_pool, &main_birdloop, "CLI"); init_list(&cli_log_hooks); cli_log_inited = 1; } diff --git a/nest/cmds.c b/nest/cmds.c index 1a16f9c7..77c92077 100644 --- a/nest/cmds.c +++ b/nest/cmds.c @@ -114,15 +114,10 @@ cmd_show_memory(void) { cli_msg(-1018, "BIRD memory usage"); cli_msg(-1018, "%-17s Effective Overhead", ""); - print_size("Routing tables:", rmemsize(rt_table_pool)); - print_size("Route attributes:", rmemsize(rta_pool)); - print_size("Protocols:", rmemsize(proto_pool)); - struct resmem total = rmemsize(&root_pool); -#ifdef HAVE_MMAP - print_size("Standby memory:", (struct resmem) { .overhead = get_page_size() * pages_kept }); - total.overhead += get_page_size() * pages_kept; -#endif - print_size("Total:", total); + print_size("Routing tables:", rp_memsize(rt_table_pool)); + print_size("Route attributes:", rp_memsize(rta_pool)); + print_size("Protocols:", rp_memsize(proto_pool)); + print_size("Total:", rp_memsize(&root_pool)); cli_msg(0, ""); } diff --git a/nest/config.Y b/nest/config.Y index 7ead8589..f9ed0e69 100644 --- a/nest/config.Y +++ b/nest/config.Y @@ -111,7 +111,7 @@ proto_postconfig(void) CF_DECLS -CF_KEYWORDS(ROUTER, ID, HOSTNAME, PROTOCOL, TEMPLATE, PREFERENCE, DISABLED, DEBUG, ALL, OFF, DIRECT) +CF_KEYWORDS(ROUTER, ID, HOSTNAME, PROTOCOL, TEMPLATE, PREFERENCE, DISABLED, DEBUG, ALL, OFF, DIRECT, PIPE) CF_KEYWORDS(INTERFACE, IMPORT, EXPORT, FILTER, NONE, VRF, DEFAULT, TABLE, STATES, ROUTES, FILTERS) CF_KEYWORDS(IPV4, IPV6, VPN4, VPN6, ROA4, ROA6, FLOW4, FLOW6, SADR, MPLS) CF_KEYWORDS(RECEIVE, LIMIT, ACTION, WARN, BLOCK, RESTART, DISABLE, KEEP, FILTERED, RPKI) @@ -128,7 +128,7 @@ CF_KEYWORDS(CHECK, LINK) /* For r_args_channel */ CF_KEYWORDS(IPV4, IPV4_MC, IPV4_MPLS, IPV6, IPV6_MC, IPV6_MPLS, IPV6_SADR, VPN4, VPN4_MC, VPN4_MPLS, VPN6, VPN6_MC, VPN6_MPLS, ROA4, ROA6, FLOW4, FLOW6, MPLS, PRI, SEC) -CF_ENUM(T_ENUM_RTS, RTS_, DUMMY, STATIC, INHERIT, DEVICE, STATIC_DEVICE, REDIRECT, +CF_ENUM(T_ENUM_RTS, RTS_, STATIC, INHERIT, DEVICE, STATIC_DEVICE, REDIRECT, RIP, OSPF, OSPF_IA, OSPF_EXT1, OSPF_EXT2, BGP, PIPE, BABEL) CF_ENUM(T_ENUM_SCOPE, SCOPE_, HOST, LINK, SITE, ORGANIZATION, UNIVERSE, UNDEFINED) CF_ENUM(T_ENUM_RTD, RTD_, UNICAST, BLACKHOLE, UNREACHABLE, PROHIBIT) @@ -266,8 +266,8 @@ proto_item: | MRTDUMP mrtdump_mask { this_proto->mrtdump = $2; } | ROUTER ID idval { this_proto->router_id = $3; } | DESCRIPTION text { this_proto->dsc = $2; } - | VRF text { this_proto->vrf = if_get_by_name($2); this_proto->vrf_set = 1; } - | VRF DEFAULT { this_proto->vrf = NULL; this_proto->vrf_set = 1; } + | VRF text { this_proto->vrf = if_get_by_name($2); } + | VRF DEFAULT { this_proto->vrf = &default_vrf; } ; @@ -348,6 +348,7 @@ debug_default: DEBUG PROTOCOLS debug_mask { new_config->proto_default_debug = $3; } | DEBUG CHANNELS debug_mask { new_config->channel_default_debug = $3; } | DEBUG COMMANDS expr { new_config->cli_debug = $3; } + | DEBUG TABLES bool { new_config->table_debug = $3; } ; /* MRTDUMP PROTOCOLS is in systep/unix/config.Y */ @@ -433,6 +434,7 @@ proto: dev_proto '}' ; dev_proto_start: proto_start DIRECT { this_proto = proto_config_new(&proto_device, $1); init_list(&DIRECT_CFG->iface_list); + this_proto->late_if_feed = 1; } ; @@ -643,12 +645,12 @@ r_args: } | r_args IMPORT TABLE channel_arg { if (!$4->in_table) cf_error("No import table in channel %s.%s", $4->proto->name, $4->name); - rt_show_add_table($$, $4->in_table); + rt_show_add_table($$, $4->in_table->tab); $$->tables_defined_by = RSD_TDB_DIRECT; } | r_args EXPORT TABLE channel_arg { if (!$4->out_table) cf_error("No export table in channel %s.%s", $4->proto->name, $4->name); - rt_show_add_table($$, $4->out_table); + rt_show_add_table($$, $4->out_table->tab); $$->tables_defined_by = RSD_TDB_DIRECT; } | r_args FILTER filter { @@ -808,7 +810,7 @@ sym_args: CF_CLI_HELP(DUMP, ..., [[Dump debugging information]]) CF_CLI(DUMP RESOURCES,,, [[Dump all allocated resource]]) -{ rdump(&root_pool); cli_msg(0, ""); } ; +{ rp_dump(&root_pool); cli_msg(0, ""); } ; CF_CLI(DUMP SOCKETS,,, [[Dump open sockets]]) { sk_dump_all(); cli_msg(0, ""); } ; CF_CLI(DUMP EVENTS,,, [[Dump event log]]) @@ -819,8 +821,10 @@ CF_CLI(DUMP NEIGHBORS,,, [[Dump neighbor cache]]) { neigh_dump_all(); cli_msg(0, ""); } ; CF_CLI(DUMP ATTRIBUTES,,, [[Dump attribute cache]]) { rta_dump_all(); cli_msg(0, ""); } ; -CF_CLI(DUMP ROUTES,,, [[Dump routing table]]) +CF_CLI(DUMP ROUTES,,, [[Dump routes]]) { rt_dump_all(); cli_msg(0, ""); } ; +CF_CLI(DUMP TABLES,,, [[Dump table connections]]) +{ rt_dump_hooks_all(); cli_msg(0, ""); } ; CF_CLI(DUMP PROTOCOLS,,, [[Dump protocol information]]) { protos_dump_all(); cli_msg(0, ""); } ; CF_CLI(DUMP FILTER ALL,,, [[Dump all filters in linearized form]]) diff --git a/nest/iface.c b/nest/iface.c index 83a633a3..dd4af243 100644 --- a/nest/iface.c +++ b/nest/iface.c @@ -37,6 +37,7 @@ static pool *if_pool; list iface_list; +struct iface default_vrf; static void if_recalc_preferred(struct iface *i); @@ -147,7 +148,7 @@ ifa_send_notify(struct proto *p, unsigned c, struct ifa *a) { if (p->ifa_notify && (p->proto_state != PS_DOWN) && - (!p->vrf_set || p->vrf == a->iface->master)) + (!p->vrf || p->vrf == a->iface->master)) { if (p->debug & D_IFACES) log(L_TRACE "%s < address %N on interface %s %s", @@ -185,7 +186,7 @@ if_send_notify(struct proto *p, unsigned c, struct iface *i) { if (p->if_notify && (p->proto_state != PS_DOWN) && - (!p->vrf_set || p->vrf == i->master)) + (!p->vrf || p->vrf == i->master)) { if (p->debug & D_IFACES) log(L_TRACE "%s < interface %s %s", p->name, i->name, @@ -243,7 +244,7 @@ if_recalc_flags(struct iface *i UNUSED, uint flags) { if ((flags & IF_ADMIN_UP) && !(flags & (IF_SHUTDOWN | IF_TMP_DOWN)) && - !(i->master_index && !i->master)) + !(i->master_index && i->master == &default_vrf)) flags |= IF_UP; else flags &= ~IF_UP; @@ -301,6 +302,9 @@ if_update(struct iface *new) struct iface *i; unsigned c; + if (!new->master) + new->master = &default_vrf; + WALK_LIST(i, iface_list) if (!strcmp(new->name, i->name)) { @@ -709,8 +713,9 @@ if_choose_router_id(struct iface_patt *mask, u32 old_id) void if_init(void) { - if_pool = rp_new(&root_pool, "Interfaces"); + if_pool = rp_new(&root_pool, &main_birdloop, "Interfaces"); init_list(&iface_list); + strcpy(default_vrf.name, "default"); neigh_init(if_pool); } @@ -843,7 +848,7 @@ if_show(void) continue; char mbuf[16 + sizeof(i->name)] = {}; - if (i->master) + if (i->master != &default_vrf) bsprintf(mbuf, " master=%s", i->master->name); else if (i->master_index) bsprintf(mbuf, " master=#%u", i->master_index); diff --git a/nest/iface.h b/nest/iface.h index 1189cdd4..13f3bd12 100644 --- a/nest/iface.h +++ b/nest/iface.h @@ -28,6 +28,8 @@ struct ifa { /* Interface address */ unsigned flags; /* Analogous to iface->flags */ }; +extern struct iface default_vrf; + struct iface { node n; char name[16]; diff --git a/nest/limit.h b/nest/limit.h new file mode 100644 index 00000000..5838ad3b --- /dev/null +++ b/nest/limit.h @@ -0,0 +1,49 @@ +/* + * BIRD Internet Routing Daemon -- Limits + * + * (c) 1998--2000 Martin Mares <mj@ucw.cz> + * (c) 2021 Maria Matejka <mq@jmq.cz> + * + * Can be freely distributed and used under the terms of the GNU GPL. + */ + +#ifndef _BIRD_LIMIT_H_ +#define _BIRD_LIMIT_H_ + +struct limit { + u32 max; + u32 count; + int (*action)(struct limit *, void *data); +}; + +static inline int limit_do_action(struct limit *l, void *data) +{ + return l->action ? l->action(l, data) : 1; +} + +static inline int limit_push(struct limit *l, void *data) +{ + if ((l->count >= l->max) && limit_do_action(l, data)) + return 1; + + l->count++; + return 0; +} + +static inline void limit_pop(struct limit *l) +{ + --l->count; +} + +static inline void limit_reset(struct limit *l) +{ + l->count = 0; +} + +static inline void limit_update(struct limit *l, void *data, u32 max) +{ + if (l->count > (l->max = max)) + limit_do_action(l, data); +} + +#endif diff --git a/nest/neighbor.c b/nest/neighbor.c index 1a31fb79..cb2d1b2b 100644 --- a/nest/neighbor.c +++ b/nest/neighbor.c @@ -142,7 +142,7 @@ if_connected(ip_addr a, struct iface *i, struct ifa **ap, uint flags) } static inline int -if_connected_any(ip_addr a, struct iface *vrf, uint vrf_set, struct iface **iface, struct ifa **addr, uint flags) +if_connected_any(ip_addr a, struct iface *vrf, struct iface **iface, struct ifa **addr, uint flags) { struct iface *i; struct ifa *b; @@ -153,7 +153,7 @@ if_connected_any(ip_addr a, struct iface *vrf, uint vrf_set, struct iface **ifac /* Prefer SCOPE_HOST or longer prefix */ WALK_LIST(i, iface_list) - if ((!vrf_set || vrf == i->master) && ((s = if_connected(a, i, &b, flags)) >= 0)) + if ((!vrf || vrf == i->master) && ((s = if_connected(a, i, &b, flags)) >= 0)) if (scope_better(s, scope) || (scope_remote(s, scope) && ifa_better(b, *addr))) { *iface = i; @@ -245,7 +245,7 @@ neigh_find(struct proto *p, ip_addr a, struct iface *iface, uint flags) iface = (scope < 0) ? NULL : iface; } else - scope = if_connected_any(a, p->vrf, p->vrf_set, &iface, &addr, flags); + scope = if_connected_any(a, p->vrf, &iface, &addr, flags); /* scope < 0 means i don't know neighbor */ /* scope >= 0 <=> iface != NULL */ @@ -369,7 +369,7 @@ neigh_update(neighbor *n, struct iface *iface) return; /* VRF-bound neighbors ignore changes in other VRFs */ - if (p->vrf_set && (p->vrf != iface->master)) + if (p->vrf && (p->vrf != iface->master)) return; scope = if_connected(n->addr, iface, &ifa, n->flags); @@ -379,7 +379,7 @@ neigh_update(neighbor *n, struct iface *iface) { /* When neighbor is going down, try to respawn it on other ifaces */ if ((scope < 0) && (n->scope >= 0) && !n->ifreq && (n->flags & NEF_STICKY)) - scope = if_connected_any(n->addr, p->vrf, p->vrf_set, &iface, &ifa, n->flags); + scope = if_connected_any(n->addr, p->vrf, &iface, &ifa, n->flags); } else { diff --git a/nest/proto-hooks.c b/nest/proto-hooks.c index bc88b4b4..716ce86c 100644 --- a/nest/proto-hooks.c +++ b/nest/proto-hooks.c @@ -76,16 +76,6 @@ void dump(struct proto *p) { DUMMY; } /** - * dump_attrs - dump protocol-dependent attributes - * @e: a route entry - * - * This hook dumps all attributes in the &rte which belong to this - * protocol to the debug output. - */ -void dump_attrs(rte *e) -{ DUMMY; } - -/** * start - request instance startup * @p: protocol instance * @@ -228,36 +218,6 @@ void neigh_notify(neighbor *neigh) { DUMMY; } /** - * make_tmp_attrs - convert embedded attributes to temporary ones - * @e: route entry - * @pool: linear pool to allocate attribute memory in - * - * This hook is called by the routing table functions if they need - * to convert the protocol attributes embedded directly in the &rte - * to temporary extended attributes in order to distribute them - * to other protocols or to filters. make_tmp_attrs() creates - * an &ea_list in the linear pool @pool, fills it with values of the - * temporary attributes and returns a pointer to it. - */ -ea_list *make_tmp_attrs(rte *e, struct linpool *pool) -{ DUMMY; } - -/** - * store_tmp_attrs - convert temporary attributes to embedded ones - * @e: route entry - * @attrs: temporary attributes to be converted - * - * This hook is an exact opposite of make_tmp_attrs() -- it takes - * a list of extended attributes and converts them to attributes - * embedded in the &rte corresponding to this protocol. - * - * You must be prepared for any of the attributes being missing - * from the list and use default values instead. - */ -void store_tmp_attrs(rte *e, ea_list *attrs) -{ DUMMY; } - -/** * preexport - pre-filtering decisions before route export * @p: protocol instance the route is going to be exported to * @e: the route in question diff --git a/nest/proto.c b/nest/proto.c index 31ee1fa1..a18d9332 100644 --- a/nest/proto.c +++ b/nest/proto.c @@ -15,6 +15,7 @@ #include "lib/event.h" #include "lib/timer.h" #include "lib/string.h" +#include "lib/coro.h" #include "conf/conf.h" #include "nest/route.h" #include "nest/iface.h" @@ -43,24 +44,46 @@ static int graceful_restart_state; static u32 graceful_restart_locks; static char *p_states[] = { "DOWN", "START", "UP", "STOP" }; -static char *c_states[] = { "DOWN", "START", "UP", "FLUSHING" }; -static char *e_states[] = { "DOWN", "FEEDING", "READY" }; +static char *c_states[] = { "DOWN", "START", "UP", "STOP", "RESTART" }; extern struct protocol proto_unix_iface; -static void channel_request_reload(struct channel *c); +static void channel_aux_request_refeed(struct channel_aux_table *cat); static void proto_shutdown_loop(timer *); static void proto_rethink_goal(struct proto *p); static char *proto_state_name(struct proto *p); -static void channel_verify_limits(struct channel *c); -static inline void channel_reset_limit(struct channel_limit *l); - +static void channel_init_limit(struct channel *c, struct limit *l, int dir, struct channel_limit *cf); +static void channel_update_limit(struct channel *c, struct limit *l, int dir, struct channel_limit *cf); +static void channel_reset_limit(struct channel *c, struct limit *l, int dir); +static void channel_feed_end(struct channel *c); +static void channel_export_stopped(struct rt_export_request *req); static inline int proto_is_done(struct proto *p) -{ return (p->proto_state == PS_DOWN) && (p->active_channels == 0); } +{ return (p->proto_state == PS_DOWN) && proto_is_inactive(p); } + +static inline event_list *proto_event_list(struct proto *p) +{ return p->loop == &main_birdloop ? &global_event_list : birdloop_event_list(p->loop); } + +static inline event_list *proto_work_list(struct proto *p) +{ return p->loop == &main_birdloop ? &global_work_list : birdloop_event_list(p->loop); } + +static inline void proto_send_event(struct proto *p) +{ ev_send(proto_event_list(p), p->event); } + +#define PROTO_ENTER_FROM_MAIN(p) ({ \ + ASSERT_DIE(birdloop_inside(&main_birdloop)); \ + struct birdloop *_loop = (p)->loop; \ + if (_loop != &main_birdloop) birdloop_enter(_loop); \ + _loop; \ + }) + +#define PROTO_LEAVE_FROM_MAIN(loop) ({ if (loop != &main_birdloop) birdloop_leave(loop); }) + +#define PROTO_LOCKED_FROM_MAIN(p) for (struct birdloop *_proto_loop = PROTO_ENTER_FROM_MAIN(p); _proto_loop; PROTO_LEAVE_FROM_MAIN(_proto_loop), (_proto_loop = NULL)) + static inline int channel_is_active(struct channel *c) -{ return (c->channel_state == CS_START) || (c->channel_state == CS_UP); } +{ return (c->channel_state != CS_DOWN); } static inline int channel_reloadable(struct channel *c) { return c->proto->reload_routes && c->reloadable; } @@ -68,10 +91,48 @@ static inline int channel_reloadable(struct channel *c) static inline void channel_log_state_change(struct channel *c) { - if (c->export_state) - CD(c, "State changed to %s/%s", c_states[c->channel_state], e_states[c->export_state]); - else - CD(c, "State changed to %s", c_states[c->channel_state]); + CD(c, "State changed to %s", c_states[c->channel_state]); +} + +void +channel_import_log_state_change(struct rt_import_request *req, u8 state) +{ + struct channel *c = SKIP_BACK(struct channel, in_req, req); + CD(c, "Channel import state changed to %s", rt_import_state_name(state)); +} + +void +channel_export_log_state_change(struct rt_export_request *req, u8 state) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + CD(c, "Channel export state changed to %s", rt_export_state_name(state)); + + switch (state) + { + case TES_FEEDING: + if (c->out_table) + rt_refresh_begin(&c->out_table->push); + else if (c->proto->feed_begin) + c->proto->feed_begin(c, !c->refeeding); + break; + case TES_READY: + channel_feed_end(c); + break; + } +} + +static void +channel_dump_import_req(struct rt_import_request *req) +{ + struct channel *c = SKIP_BACK(struct channel, in_req, req); + debug(" Channel %s.%s import request %p\n", c->proto->name, c->name, req); +} + +static void +channel_dump_export_req(struct rt_export_request *req) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + debug(" Channel %s.%s export request %p\n", c->proto->name, c->name, req); } static void @@ -111,7 +172,7 @@ proto_cf_find_channel(struct proto_config *pc, uint net_type) * Returns pointer to channel or NULL */ struct channel * -proto_find_channel_by_table(struct proto *p, struct rtable *t) +proto_find_channel_by_table(struct proto *p, rtable *t) { struct channel *c; @@ -141,6 +202,16 @@ proto_find_channel_by_name(struct proto *p, const char *n) return NULL; } +rte * channel_preimport(struct rt_import_request *req, rte *new, rte *old); +rte * channel_in_preimport(struct rt_import_request *req, rte *new, rte *old); + +void rt_notify_optimal(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe); +void rt_notify_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe); +void rt_feed_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe, rte **feed, uint count); +void rt_notify_accepted(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe, rte **feed, uint count); +void rt_notify_merged(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe, rte **feed, uint count); + + /** * proto_add_channel - connect protocol to a routing table * @p: protocol instance @@ -166,11 +237,17 @@ proto_add_channel(struct proto *p, struct channel_config *cf) c->proto = p; c->table = cf->table->table; + RT_LOCKED(c->table, t) + rt_lock_table(t); + c->in_filter = cf->in_filter; c->out_filter = cf->out_filter; - c->rx_limit = cf->rx_limit; - c->in_limit = cf->in_limit; - c->out_limit = cf->out_limit; + + channel_init_limit(c, &c->rx_limit, PLD_RX, &cf->rx_limit); + channel_init_limit(c, &c->in_limit, PLD_IN, &cf->in_limit); + channel_init_limit(c, &c->out_limit, PLD_OUT, &cf->out_limit); + + c->rte_update_pool = lp_new_default(proto_pool); c->net_type = cf->net_type; c->ra_mode = cf->ra_mode; @@ -181,7 +258,6 @@ proto_add_channel(struct proto *p, struct channel_config *cf) c->rpki_reload = cf->rpki_reload; c->channel_state = CS_DOWN; - c->export_state = ES_DOWN; c->last_state_change = current_time(); c->reloadable = 1; @@ -203,6 +279,9 @@ proto_remove_channel(struct proto *p UNUSED, struct channel *c) CD(c, "Removed", c->name); + RT_LOCKED(c->table, t) + rt_unlock_table(t); + rem_node(&c->n); mb_free(c); } @@ -223,7 +302,7 @@ proto_pause_channels(struct proto *p) struct channel *c; WALK_LIST(c, p->channels) if (!c->disabled && channel_is_active(c)) - channel_set_state(c, CS_START); + channel_set_state(c, CS_PAUSE); } static void @@ -232,7 +311,7 @@ proto_stop_channels(struct proto *p) struct channel *c; WALK_LIST(c, p->channels) if (!c->disabled && channel_is_active(c)) - channel_set_state(c, CS_FLUSHING); + channel_set_state(c, CS_STOP); } static void @@ -244,96 +323,25 @@ proto_remove_channels(struct proto *p) } static void -channel_schedule_feed(struct channel *c, int initial) -{ - // DBG("%s: Scheduling meal\n", p->name); - ASSERT(c->channel_state == CS_UP); - - c->export_state = ES_FEEDING; - c->refeeding = !initial; - - ev_schedule_work(c->feed_event); -} - -static void -channel_feed_loop(void *ptr) -{ - struct channel *c = ptr; - - if (c->export_state != ES_FEEDING) - return; - - /* Start feeding */ - if (!c->feed_active) - { - if (c->proto->feed_begin) - c->proto->feed_begin(c, !c->refeeding); - - c->refeed_pending = 0; - } - - // DBG("Feeding protocol %s continued\n", p->name); - if (!rt_feed_channel(c)) - { - ev_schedule_work(c->feed_event); - return; - } - - /* Reset export limit if the feed ended with acceptable number of exported routes */ - struct channel_limit *l = &c->out_limit; - if (c->refeeding && - (l->state == PLS_BLOCKED) && - (c->refeed_count <= l->limit) && - (c->stats.exp_routes <= l->limit)) - { - log(L_INFO "Protocol %s resets route export limit (%u)", c->proto->name, l->limit); - channel_reset_limit(&c->out_limit); - - /* Continue in feed - it will process routing table again from beginning */ - c->refeed_count = 0; - ev_schedule_work(c->feed_event); - return; - } - - // DBG("Feeding protocol %s finished\n", p->name); - c->export_state = ES_READY; - channel_log_state_change(c); - - if (c->proto->feed_end) - c->proto->feed_end(c); - - /* Restart feeding */ - if (c->refeed_pending) - channel_request_feeding(c); -} - - -static void -channel_roa_in_changed(struct rt_subscription *s) +channel_roa_in_changed(void *_data) { - struct channel *c = s->data; - int active = c->reload_event && ev_active(c->reload_event); + struct channel *c = _data; - CD(c, "Reload triggered by RPKI change%s", active ? " - already active" : ""); + CD(c, "Reload triggered by RPKI change"); - if (!active) - channel_request_reload(c); - else - c->reload_pending = 1; + channel_request_reload(c); } static void -channel_roa_out_changed(struct rt_subscription *s) +channel_roa_out_changed(void *_data) { - struct channel *c = s->data; - int active = (c->export_state == ES_FEEDING); + struct channel *c = _data; + CD(c, "Feeding triggered by RPKI change"); - CD(c, "Feeding triggered by RPKI change%s", active ? " - already active" : ""); + c->refeed_pending = 1; - if (!active) - channel_request_feeding(c); - else - c->refeed_pending = 1; + if (c->out_req.hook) + rt_stop_export(&c->out_req, channel_export_stopped); } /* Temporary code, subscriptions should be changed to resources */ @@ -345,14 +353,14 @@ struct roa_subscription { static int channel_roa_is_subscribed(struct channel *c, rtable *tab, int dir) { - void (*hook)(struct rt_subscription *) = + void (*hook)(void *) = dir ? channel_roa_in_changed : channel_roa_out_changed; struct roa_subscription *s; node *n; WALK_LIST2(s, n, c->roa_subscriptions, roa_node) - if ((s->s.tab == tab) && (s->s.hook == hook)) + if ((s->s.tab == tab) && (s->s.event->hook == hook)) return 1; return 0; @@ -366,9 +374,9 @@ channel_roa_subscribe(struct channel *c, rtable *tab, int dir) return; struct roa_subscription *s = mb_allocz(c->proto->pool, sizeof(struct roa_subscription)); + s->s.event = ev_new_init(c->proto->pool, dir ? channel_roa_in_changed : channel_roa_out_changed, c); + s->s.event->list = proto_work_list(c->proto); - s->s.hook = dir ? channel_roa_in_changed : channel_roa_out_changed; - s->s.data = c; rt_subscribe(tab, &s->s); add_tail(&c->roa_subscriptions, &s->roa_node); @@ -379,6 +387,7 @@ channel_roa_unsubscribe(struct roa_subscription *s) { rt_unsubscribe(&s->s); rem_node(&s->roa_node); + rfree(s->s.event); mb_free(s); } @@ -386,7 +395,7 @@ static void channel_roa_subscribe_filter(struct channel *c, int dir) { const struct filter *f = dir ? c->in_filter : c->out_filter; - struct rtable *tab; + rtable *tab; int valid = 1, found = 0; if ((f == FILTER_ACCEPT) || (f == FILTER_REJECT)) @@ -445,119 +454,530 @@ channel_roa_unsubscribe_all(struct channel *c) } static void -channel_start_export(struct channel *c) +channel_start_import(struct channel *c) { + if (c->in_req.hook) + { + log(L_WARN "%s.%s: Attempted to start channel's already started import", c->proto->name, c->name); + return; + } + + int nlen = strlen(c->name) + strlen(c->proto->name) + 2; + char *rn = mb_allocz(c->proto->pool, nlen); + bsprintf(rn, "%s.%s", c->proto->name, c->name); + + c->in_req = (struct rt_import_request) { + .name = rn, + .list = proto_work_list(c->proto), + .trace_routes = c->debug | c->proto->debug, + .dump_req = channel_dump_import_req, + .log_state_change = channel_import_log_state_change, + .preimport = channel_preimport, + }; + ASSERT(c->channel_state == CS_UP); - ASSERT(c->export_state == ES_DOWN); - channel_schedule_feed(c, 1); /* Sets ES_FEEDING */ + channel_reset_limit(c, &c->rx_limit, PLD_RX); + channel_reset_limit(c, &c->in_limit, PLD_IN); + + memset(&c->import_stats, 0, sizeof(struct channel_import_stats)); + + DBG("%s.%s: Channel start import req=%p\n", c->proto->name, c->name, &c->in_req); + rt_request_import(c->table, &c->in_req); } static void -channel_stop_export(struct channel *c) +channel_start_export(struct channel *c) { - /* Need to abort feeding */ - if (c->export_state == ES_FEEDING) - rt_feed_channel_abort(c); + if (c->out_req.hook) + { + c->restart_export = 1; + log(L_WARN "%s.%s: Fast channel export restart", c->proto->name, c->name); + return; + } + + ASSERT(c->channel_state == CS_UP); + int nlen = strlen(c->name) + strlen(c->proto->name) + 2; + char *rn = mb_allocz(c->proto->pool, nlen); + bsprintf(rn, "%s.%s", c->proto->name, c->name); + + c->out_req = (struct rt_export_request) { + .name = rn, + .list = proto_work_list(c->proto), + .trace_routes = c->debug | c->proto->debug, + .dump_req = channel_dump_export_req, + .log_state_change = channel_export_log_state_change, + }; + + bmap_init(&c->export_map, c->proto->pool, 1024); + bmap_init(&c->export_reject_map, c->proto->pool, 1024); + + channel_reset_limit(c, &c->out_limit, PLD_OUT); - c->export_state = ES_DOWN; - c->stats.exp_routes = 0; - bmap_reset(&c->export_map, 1024); + memset(&c->export_stats, 0, sizeof(struct channel_export_stats)); + + switch (c->ra_mode) { + case RA_OPTIMAL: + c->out_req.export_one = rt_notify_optimal; + break; + case RA_ANY: + c->out_req.export_one = rt_notify_any; + c->out_req.export_bulk = rt_feed_any; + break; + case RA_ACCEPTED: + c->out_req.export_bulk = rt_notify_accepted; + break; + case RA_MERGED: + c->out_req.export_bulk = rt_notify_merged; + break; + default: + bug("Unknown route announcement mode"); + } + + DBG("%s.%s: Channel start export req=%p\n", c->proto->name, c->name, &c->out_req); + rt_request_export(c->table, &c->out_req); } +static void +channel_check_stopped(struct channel *c) +{ + switch (c->channel_state) + { + case CS_STOP: + if (c->out_req.hook || c->in_req.hook || c->out_table || c->in_table) + return; + + channel_set_state(c, CS_DOWN); + proto_send_event(c->proto); + + break; + case CS_PAUSE: + if (c->out_req.hook) + return; + + channel_set_state(c, CS_START); + break; + default: + bug("Stopped channel in a bad state: %d", c->channel_state); + } + + DBG("%s.%s: Channel requests/hooks stopped (in state %s)\n", c->proto->name, c->name, c_states[c->channel_state]); +} -/* Called by protocol for reload from in_table */ void -channel_schedule_reload(struct channel *c) +channel_import_stopped(void *_c) { - ASSERT(c->channel_state == CS_UP); + struct channel *c = _c; + + c->in_req.hook = NULL; - rt_reload_channel_abort(c); - ev_schedule_work(c->reload_event); + mb_free(c->in_req.name); + c->in_req.name = NULL; + + channel_check_stopped(c); } static void -channel_reload_loop(void *ptr) +channel_export_stopped(struct rt_export_request *req) { - struct channel *c = ptr; + struct channel *c = SKIP_BACK(struct channel, out_req, req); - /* Start reload */ - if (!c->reload_active) - c->reload_pending = 0; + /* The hook has already stopped */ + req->hook = NULL; - if (!rt_reload_channel(c)) + if (c->refeed_pending) { - ev_schedule_work(c->reload_event); + c->refeeding = 1; + c->refeed_pending = 0; + + bmap_reset(&c->export_map, 1024); + bmap_reset(&c->export_reject_map, 1024); + + rt_request_export(c->table, req); return; } - /* Restart reload */ - if (c->reload_pending) - channel_request_reload(c); + mb_free(c->out_req.name); + c->out_req.name = NULL; + + bmap_free(&c->export_map); + bmap_free(&c->export_reject_map); + + if (c->restart_export) + { + c->restart_export = 0; + channel_start_export(c); + } + else + channel_check_stopped(c); +} + +static void +channel_feed_end(struct channel *c) +{ + struct rt_export_request *req = &c->out_req; + + /* Reset export limit if the feed ended with acceptable number of exported routes */ + struct limit *l = &c->out_limit; + if (c->refeeding && + (c->limit_active & (1 << PLD_OUT)) && + (c->refeed_count <= l->max) && + (l->count <= l->max)) + { + log(L_INFO "Protocol %s resets route export limit (%u)", c->proto->name, l->max); + channel_reset_limit(c, &c->out_limit, PLD_OUT); + + c->refeed_pending = 1; + rt_stop_export(req, channel_export_stopped); + return; + } + + if (c->out_table) + rt_refresh_end(&c->out_table->push); + else if (c->proto->feed_end) + c->proto->feed_end(c); + + if (c->refeed_pending) + rt_stop_export(req, channel_export_stopped); +} + +#define CHANNEL_AUX_TABLE_DUMP_REQ(inout, imex, pgimex, pushget) static void \ + channel_##inout##_##pushget##_dump_req(struct rt_##pgimex##_request *req) { \ + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, pushget, req); \ + debug(" Channel %s.%s " #imex " table " #pushget " request %p\n", cat->c->proto->name, cat->c->name, req); } + +CHANNEL_AUX_TABLE_DUMP_REQ(in, import, import, push) +CHANNEL_AUX_TABLE_DUMP_REQ(in, import, export, get) +CHANNEL_AUX_TABLE_DUMP_REQ(out, export, import, push) +CHANNEL_AUX_TABLE_DUMP_REQ(out, export, export, get) + +#undef CHANNEL_AUX_TABLE_DUMP_REQ + +static uint channel_aux_imex(struct channel_aux_table *cat) +{ + if (cat->c->in_table == cat) + return 0; + else if (cat->c->out_table == cat) + return 1; + else + bug("Channel aux table must be in_table or out_table"); } static void -channel_reset_import(struct channel *c) +channel_aux_stopped(void *data) { - /* Need to abort feeding */ - ev_postpone(c->reload_event); - rt_reload_channel_abort(c); + struct channel_aux_table *cat; + RT_LOCKED((rtable *) data, t) + cat = t->config->owner; + + struct channel *c = cat->c; + + if (channel_aux_imex(cat)) + c->out_table = NULL; + else + c->in_table = NULL; - rt_prune_sync(c->in_table, 1); + mb_free(cat); + channel_check_stopped(c); } static void -channel_reset_export(struct channel *c) +channel_aux_import_stopped(void *_cat) { - /* Just free the routes */ - rt_prune_sync(c->out_table, 1); + struct channel_aux_table *cat = _cat; + cat->push.hook = NULL; } -/* Called by protocol to activate in_table */ -void -channel_setup_in_table(struct channel *c) +static void +channel_aux_export_stopped(struct rt_export_request *req) +{ + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req); + req->hook = NULL; + + int del; + RT_LOCKED(cat->tab, t) + del = !!t->delete; + + if (del) + return; + + ASSERT_DIE(cat->refeed_pending); + cat->refeed_pending = 0; + rt_request_export(cat->tab, req); +} + +static void +channel_aux_stop(struct channel_aux_table *cat) +{ + RT_LOCKED(cat->tab, t) + t->delete = channel_aux_stopped; + + cat->push_stopped = (event) { + .hook = channel_aux_import_stopped, + .data = cat, + .list = proto_event_list(cat->c->proto), + }; + + rt_stop_import(&cat->push, &cat->push_stopped); + rt_stop_export(&cat->get, channel_aux_export_stopped); +} + +static void +channel_push_log_state_change(struct rt_import_request *req, u8 state) +{ + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, push, req); + const char *imex = channel_aux_imex(cat) ? "export" : "import"; + CD(cat->c, "Channel %s table import state changed to %s", imex, rt_import_state_name(state)); +} + +static void +channel_get_log_state_change(struct rt_export_request *req, u8 state) +{ + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req); + const char *imex = channel_aux_imex(cat) ? "export" : "import"; + CD(cat->c, "Channel %s table export state changed to %s", imex, rt_export_state_name(state)); + + switch (state) + { + case TES_FEEDING: + if (imex && cat->c->proto->feed_begin) + cat->c->proto->feed_begin(cat->c, !cat->c->refeeding); + else if (!imex) + rt_refresh_begin(&cat->c->in_req); + break; + + case TES_READY: + if (imex && cat->c->proto->feed_end) + cat->c->proto->feed_end(cat->c); + else if (!imex) + rt_refresh_end(&cat->c->in_req); + + if (cat->refeed_pending) + rt_stop_export(&cat->get, channel_aux_export_stopped); + + break; + } +} + +void rte_update_direct(struct channel *c, const net_addr *n, rte *new, struct rte_src *src); + +static int +channel_aux_export_one_any(struct rt_export_request *req, struct rt_pending_export *rpe, rte **new, rte **old) +{ + struct rte_src *src = rpe->new ? rpe->new->rte.src : rpe->old->rte.src; + *old = RTES_OR_NULL(rpe->old); + struct rte_storage *new_stored; + + while (rpe) + { + new_stored = rpe->new; + rpe_mark_seen(req->hook, rpe); + rpe = rpe_next(rpe, src); + } + + *new = RTES_CLONE(new_stored, *new); + + return (*new || *old) && (&new_stored->rte != *old); +} + +static int +channel_aux_export_one_best(struct rt_export_request *req, struct rt_pending_export *rpe, rte **new, rte **old) +{ + *old = RTES_OR_NULL(rpe->old_best); + struct rte_storage *new_stored; + + while (rpe) + { + new_stored = rpe->new_best; + rpe_mark_seen(req->hook, rpe); + rpe = rpe_next(rpe, NULL); + } + + *new = RTES_CLONE(new_stored, *new); + + return (*new || *old) && (&new_stored->rte != *old); +} + +static void +channel_in_export_one_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe) +{ + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req); + + rte n0, *new = &n0, *old; + if (channel_aux_export_one_any(req, rpe, &new, &old)) + rte_update_direct(cat->c, net, new, old ? old->src : new->src); +} + +static void +channel_in_export_one_best(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe) +{ + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req); + + rte n0, *new = &n0, *old; + if (channel_aux_export_one_best(req, rpe, &new, &old)) + rte_update_direct(cat->c, net, new, old ? old->src : new->src); +} + +static void +channel_in_export_bulk_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe UNUSED, rte **feed, uint count) +{ + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req); + for (uint i=0; i<count; i++) + { + rte n0 = *feed[i]; + rte_update_direct(cat->c, net, &n0, n0.src); + } +} + +static void +channel_in_export_bulk_best(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe UNUSED, rte **feed, uint count) +{ + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req); + if (!count) + return; + + rte n0 = *feed[0]; + rte_update_direct(cat->c, net, &n0, n0.src); +} + +void do_rt_notify_direct(struct channel *c, const net_addr *net, rte *new, const rte *old); + +static void +channel_out_export_one_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe) +{ + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req); + rte n0, *new = &n0, *old; + if (channel_aux_export_one_any(req, rpe, &new, &old)) + do_rt_notify_direct(cat->c, net, new, old); +} + +static void +channel_out_export_one_best(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe) +{ + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req); + rte n0, *new = &n0, *old; + if (channel_aux_export_one_best(req, rpe, &new, &old)) + do_rt_notify_direct(cat->c, net, new, old); +} + +static void +channel_out_export_bulk(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe UNUSED, rte **feed, uint count) { - struct rtable_config *cf = mb_allocz(c->proto->pool, sizeof(struct rtable_config)); + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, get, req); + if (cat->c->ra_mode != RA_ANY) + ASSERT_DIE(count <= 1); + + for (uint i=0; i<count; i++) + { + rte n0 = *feed[i]; + do_rt_notify_direct(cat->c, net, &n0, NULL); + } +} - cf->name = "import"; - cf->addr_type = c->net_type; - cf->internal = 1; +/* Called by protocol to activate in_table */ +void +channel_setup_in_table(struct channel *c, int best) +{ + int nlen = sizeof("import") + strlen(c->name) + strlen(c->proto->name) + 3; + + struct { + struct channel_aux_table cat; + struct rtable_config tab_cf; + char name[0]; + } *cat = mb_allocz(c->proto->pool, sizeof(*cat) + nlen); + + bsprintf(cat->name, "%s.%s.import", c->proto->name, c->name); + + cat->tab_cf.owner = cat; + cat->tab_cf.name = cat->name; + cat->tab_cf.addr_type = c->net_type; + cat->tab_cf.cork_limit = 4 * page_size / sizeof(struct rt_pending_export); + + c->in_table = &cat->cat; + c->in_table->push = (struct rt_import_request) { + .name = cat->name, + .list = proto_work_list(c->proto), + .trace_routes = c->debug | c->proto->debug, + .dump_req = channel_in_push_dump_req, + .log_state_change = channel_push_log_state_change, + .preimport = channel_in_preimport, + }; + c->in_table->get = (struct rt_export_request) { + .name = cat->name, + .list = proto_work_list(c->proto), + .trace_routes = c->debug | c->proto->debug, + .dump_req = channel_in_get_dump_req, + .log_state_change = channel_get_log_state_change, + .export_one = best ? channel_in_export_one_best : channel_in_export_one_any, + .export_bulk = best ? channel_in_export_bulk_best : channel_in_export_bulk_any, + }; - c->in_table = rt_setup(c->proto->pool, cf); + c->in_table->c = c; + c->in_table->tab = rt_setup(c->proto->pool, &cat->tab_cf); - c->reload_event = ev_new_init(c->proto->pool, channel_reload_loop, c); + rt_request_import(c->in_table->tab, &c->in_table->push); + rt_request_export(c->in_table->tab, &c->in_table->get); } /* Called by protocol to activate out_table */ void channel_setup_out_table(struct channel *c) { - struct rtable_config *cf = mb_allocz(c->proto->pool, sizeof(struct rtable_config)); - cf->name = "export"; - cf->addr_type = c->net_type; - cf->internal = 1; + int nlen = sizeof("export") + strlen(c->name) + strlen(c->proto->name) + 3; + + struct { + struct channel_aux_table cat; + struct rtable_config tab_cf; + char name[0]; + } *cat = mb_allocz(c->proto->pool, sizeof(*cat) + nlen); + + bsprintf(cat->name, "%s.%s.export", c->proto->name, c->name); + + cat->tab_cf.owner = cat; + cat->tab_cf.name = cat->name; + cat->tab_cf.addr_type = c->net_type; + cat->tab_cf.cork_limit = 4 * page_size / sizeof(struct rt_pending_export); - c->out_table = rt_setup(c->proto->pool, cf); + c->out_table = &cat->cat; + c->out_table->push = (struct rt_import_request) { + .name = cat->name, + .list = proto_work_list(c->proto), + .trace_routes = c->debug | c->proto->debug, + .dump_req = channel_out_push_dump_req, + .log_state_change = channel_push_log_state_change, + }; + c->out_table->get = (struct rt_export_request) { + .name = cat->name, + .list = proto_work_list(c->proto), + .trace_routes = c->debug | c->proto->debug, + .dump_req = channel_out_get_dump_req, + .log_state_change = channel_get_log_state_change, + .export_one = (c->ra_mode == RA_ANY) ? channel_out_export_one_any : channel_out_export_one_best, + .export_bulk = channel_out_export_bulk, + }; + + c->out_table->c = c; + c->out_table->tab = rt_setup(c->proto->pool, &cat->tab_cf); + + rt_request_import(c->out_table->tab, &c->out_table->push); + rt_request_export(c->out_table->tab, &c->out_table->get); } +static void +channel_aux_request_refeed(struct channel_aux_table *cat) +{ + cat->refeed_pending = 1; + rt_stop_export(&cat->get, channel_aux_export_stopped); +} static void channel_do_start(struct channel *c) { - rt_lock_table(c->table); - add_tail(&c->table->channels, &c->table_node); c->proto->active_channels++; - c->feed_event = ev_new_init(c->proto->pool, channel_feed_loop, c); - - bmap_init(&c->export_map, c->proto->pool, 1024); - memset(&c->stats, 0, sizeof(struct proto_stats)); - - channel_reset_limit(&c->rx_limit); - channel_reset_limit(&c->in_limit); - channel_reset_limit(&c->out_limit); - CALL(c->channel->start, c); + + channel_start_import(c); } static void @@ -572,9 +992,38 @@ channel_do_up(struct channel *c) } static void -channel_do_flush(struct channel *c) +channel_do_pause(struct channel *c) +{ + /* Stop export */ + if (c->out_req.hook) + { + rt_stop_export(&c->out_req, channel_export_stopped); + c->refeeding = 0; + } + + channel_roa_unsubscribe_all(c); +} + +static void +channel_do_stop(struct channel *c) { - rt_schedule_prune(c->table); + /* Drop auxiliary tables */ + if (c->in_table) + channel_aux_stop(c->in_table); + + if (c->out_table) + channel_aux_stop(c->out_table); + + /* Stop import */ + if (c->in_req.hook) + { + c->in_stopped = (event) { + .hook = channel_import_stopped, + .data = c, + .list = proto_event_list(c->proto), + }; + rt_stop_import(&c->in_req, &c->in_stopped); + } c->gr_wait = 0; if (c->gr_lock) @@ -582,48 +1031,30 @@ channel_do_flush(struct channel *c) CALL(c->channel->shutdown, c); - /* This have to be done in here, as channel pool is freed before channel_do_down() */ - bmap_free(&c->export_map); - c->in_table = NULL; - c->reload_event = NULL; - c->out_table = NULL; - channel_roa_unsubscribe_all(c); } static void channel_do_down(struct channel *c) { - ASSERT(!c->feed_active && !c->reload_active); + ASSERT(!c->out_req.hook && !c->in_req.hook && !c->out_table && !c->in_table); - rem_node(&c->table_node); - rt_unlock_table(c->table); c->proto->active_channels--; - if ((c->stats.imp_routes + c->stats.filt_routes) != 0) - log(L_ERR "%s: Channel %s is down but still has some routes", c->proto->name, c->name); - - // bmap_free(&c->export_map); - memset(&c->stats, 0, sizeof(struct proto_stats)); - - c->in_table = NULL; - c->reload_event = NULL; - c->out_table = NULL; - - /* The in_table and out_table are going to be freed by freeing their resource pools. */ + memset(&c->import_stats, 0, sizeof(struct channel_import_stats)); + memset(&c->export_stats, 0, sizeof(struct channel_export_stats)); CALL(c->channel->cleanup, c); /* Schedule protocol shutddown */ if (proto_is_done(c->proto)) - ev_schedule(c->proto->event); + proto_send_event(c->proto); } void channel_set_state(struct channel *c, uint state) { uint cs = c->channel_state; - uint es = c->export_state; DBG("%s reporting channel %s state transition %s -> %s\n", c->proto->name, c->name, c_states[cs], c_states[state]); if (state == cs) @@ -635,24 +1066,15 @@ channel_set_state(struct channel *c, uint state) switch (state) { case CS_START: - ASSERT(cs == CS_DOWN || cs == CS_UP); + ASSERT(cs == CS_DOWN || cs == CS_PAUSE); if (cs == CS_DOWN) channel_do_start(c); - if (es != ES_DOWN) - channel_stop_export(c); - - if (c->in_table && (cs == CS_UP)) - channel_reset_import(c); - - if (c->out_table && (cs == CS_UP)) - channel_reset_export(c); - break; case CS_UP: - ASSERT(cs == CS_DOWN || cs == CS_START); + ASSERT(cs == CS_DOWN || cs == CS_START || cs == CS_PAUSE); if (cs == CS_DOWN) channel_do_start(c); @@ -663,23 +1085,24 @@ channel_set_state(struct channel *c, uint state) channel_do_up(c); break; - case CS_FLUSHING: - ASSERT(cs == CS_START || cs == CS_UP); + case CS_PAUSE: + ASSERT(cs == CS_UP); - if (es != ES_DOWN) - channel_stop_export(c); + if (cs == CS_UP) + channel_do_pause(c); + break; - if (c->in_table && (cs == CS_UP)) - channel_reset_import(c); + case CS_STOP: + ASSERT(cs == CS_UP || cs == CS_START || cs == CS_PAUSE); - if (c->out_table && (cs == CS_UP)) - channel_reset_export(c); + if (cs == CS_UP) + channel_do_pause(c); - channel_do_flush(c); + channel_do_stop(c); break; case CS_DOWN: - ASSERT(cs == CS_FLUSHING); + ASSERT(cs == CS_STOP); channel_do_down(c); break; @@ -701,50 +1124,62 @@ channel_set_state(struct channel *c, uint state) * completed, it will switch back to ES_READY. This function can be called * even when feeding is already running, in that case it is restarted. */ -void -channel_request_feeding(struct channel *c) +static void +channel_request_table_feeding(struct channel *c) { - ASSERT(c->channel_state == CS_UP); + ASSERT(c->out_req.hook); - CD(c, "Feeding requested"); + c->refeed_pending = 1; + rt_stop_export(&c->out_req, channel_export_stopped); +} - /* Do nothing if we are still waiting for feeding */ - if (c->export_state == ES_DOWN) +void +channel_request_feeding(struct channel *c) +{ + if (c->gr_wait || !c->proto->rt_notify) return; - /* If we are already feeding, we want to restart it */ - if (c->export_state == ES_FEEDING) - { - /* Unless feeding is in initial state */ - if (!c->feed_active) - return; - - rt_feed_channel_abort(c); - } + CD(c, "Refeed requested"); - /* Track number of exported routes during refeed */ - c->refeed_count = 0; + ASSERT_DIE(c->out_req.hook); - channel_schedule_feed(c, 0); /* Sets ES_FEEDING */ - channel_log_state_change(c); + if (c->out_table) + channel_aux_request_refeed(c->out_table); + else + channel_request_table_feeding(c); } -static void +void channel_request_reload(struct channel *c) { - ASSERT(c->channel_state == CS_UP); + ASSERT(c->in_req.hook); ASSERT(channel_reloadable(c)); CD(c, "Reload requested"); - c->proto->reload_routes(c); + if (c->in_table) + channel_aux_request_refeed(c->in_table); + else + c->proto->reload_routes(c); +} - /* - * Should this be done before reload_routes() hook? - * Perhaps, but routes are updated asynchronously. - */ - channel_reset_limit(&c->rx_limit); - channel_reset_limit(&c->in_limit); +void +channel_refresh_begin(struct channel *c) +{ + CD(c, "Channel route refresh begin"); + if (c->in_table) + rt_refresh_begin(&c->in_table->push); + else + rt_refresh_begin(&c->in_req); +} + +void +channel_refresh_end(struct channel *c) +{ + if (c->in_table) + rt_refresh_end(&c->in_table->push); + else + rt_refresh_end(&c->in_req); } const struct channel_class channel_basic = { @@ -847,19 +1282,19 @@ channel_reconfigure(struct channel *c, struct channel_config *cf) /* Reconfigure channel fields */ c->in_filter = cf->in_filter; c->out_filter = cf->out_filter; - c->rx_limit = cf->rx_limit; - c->in_limit = cf->in_limit; - c->out_limit = cf->out_limit; + + channel_update_limit(c, &c->rx_limit, PLD_RX, &cf->rx_limit); + channel_update_limit(c, &c->in_limit, PLD_IN, &cf->in_limit); + channel_update_limit(c, &c->out_limit, PLD_OUT, &cf->out_limit); // c->ra_mode = cf->ra_mode; c->merge_limit = cf->merge_limit; c->preference = cf->preference; c->debug = cf->debug; + c->in_req.trace_routes = c->out_req.trace_routes = c->debug | c->proto->debug; c->in_keep_filtered = cf->in_keep_filtered; c->rpki_reload = cf->rpki_reload; - channel_verify_limits(c); - /* Execute channel-specific reconfigure hook */ if (c->channel->reconfigure && !c->channel->reconfigure(c, cf, &import_changed, &export_changed)) return 0; @@ -902,7 +1337,7 @@ channel_reconfigure(struct channel *c, struct channel_config *cf) channel_request_reload(c); if (export_changed) - channel_request_feeding(c); + channel_request_table_feeding(c); done: CD(c, "Reconfigured"); @@ -950,34 +1385,50 @@ proto_configure_channel(struct proto *p, struct channel **pc, struct channel_con return 1; } +static void +proto_cleanup(struct proto *p) +{ + p->active = 0; + proto_log_state_change(p); + proto_rethink_goal(p); +} static void -proto_event(void *ptr) +proto_loop_stopped(void *ptr) { struct proto *p = ptr; - if (p->do_start) - { - if_feed_baby(p); - p->do_start = 0; - } + ASSERT_DIE(birdloop_inside(&main_birdloop)); + + p->loop = &main_birdloop; + p->pool = NULL; + p->event->list = NULL; + + proto_cleanup(p); +} + +static void +proto_event(void *ptr) +{ + struct proto *p = ptr; if (p->do_stop) { if (p->proto == &proto_unix_iface) if_flush_ifaces(p); + p->do_stop = 0; } if (proto_is_done(p)) - { - if (p->proto->cleanup) - p->proto->cleanup(p); - - p->active = 0; - proto_log_state_change(p); - proto_rethink_goal(p); - } + if (p->loop != &main_birdloop) + birdloop_stop_self(p->loop, proto_loop_stopped, p); + else + { + rp_free(p->pool, proto_pool); + p->pool = NULL; + proto_cleanup(p); + } } @@ -1018,10 +1469,10 @@ proto_init(struct proto_config *c, node *n) struct protocol *pr = c->protocol; struct proto *p = pr->init(c); + p->loop = &main_birdloop; p->proto_state = PS_DOWN; p->last_state_change = current_time(); p->vrf = c->vrf; - p->vrf_set = c->vrf_set; insert_node(&p->n, n); p->event = ev_new_init(proto_pool, proto_event, p); @@ -1034,11 +1485,30 @@ proto_init(struct proto_config *c, node *n) static void proto_start(struct proto *p) { - /* Here we cannot use p->cf->name since it won't survive reconfiguration */ - p->pool = rp_new(proto_pool, p->proto->name); + DBG("Kicking %s up\n", p->name); + PD(p, "Starting"); + + int ns = strlen("Protocol ") + strlen(p->cf->name) + 1; + void *nb = mb_alloc(proto_pool, ns); + ASSERT_DIE(ns - 1 == bsnprintf(nb, ns, "Protocol %s", p->cf->name)); if (graceful_restart_state == GRS_INIT) p->gr_recovery = 1; + + if (p->cf->loop_order == DOMAIN_ORDER(the_bird)) + p->pool = rp_new(proto_pool, &main_birdloop, nb); + else + { + p->loop = birdloop_new(proto_pool, p->cf->loop_order, nb); + p->pool = birdloop_pool(p->loop); + } + + p->event->list = proto_event_list(p); + + mb_move(nb, p->pool); + + PROTO_LOCKED_FROM_MAIN(p) + proto_notify_state(p, (p->proto->start ? p->proto->start(p) : PS_UP)); } @@ -1074,6 +1544,7 @@ proto_config_new(struct protocol *pr, int class) cf->class = class; cf->debug = new_config->proto_default_debug; cf->mrtdump = new_config->proto_default_mrtdump; + cf->loop_order = DOMAIN_ORDER(the_bird); init_list(&cf->channels); @@ -1189,8 +1660,7 @@ proto_reconfigure(struct proto *p, struct proto_config *oc, struct proto_config if ((nc->protocol != oc->protocol) || (nc->net_type != oc->net_type) || (nc->disabled != p->disabled) || - (nc->vrf != oc->vrf) || - (nc->vrf_set != oc->vrf_set)) + (nc->vrf != oc->vrf)) return 0; p->name = nc->name; @@ -1363,11 +1833,20 @@ protos_commit(struct config *new, struct config *old, int force_reconfig, int ty } static void -proto_rethink_goal(struct proto *p) +proto_shutdown(struct proto *p) { - struct protocol *q; - byte goal; + if (p->proto_state == PS_START || p->proto_state == PS_UP) + { + /* Going down */ + DBG("Kicking %s down\n", p->name); + PD(p, "Shutting down"); + proto_notify_state(p, (p->proto->shutdown ? p->proto->shutdown(p) : PS_DOWN)); + } +} +static void +proto_rethink_goal(struct proto *p) +{ if (p->reconfiguring && !p->active) { struct proto_config *nc = p->cf_new; @@ -1387,32 +1866,12 @@ proto_rethink_goal(struct proto *p) /* Determine what state we want to reach */ if (p->disabled || p->reconfiguring) - goal = PS_DOWN; - else - goal = PS_UP; - - q = p->proto; - if (goal == PS_UP) - { - if (!p->active) - { - /* Going up */ - DBG("Kicking %s up\n", p->name); - PD(p, "Starting"); - proto_start(p); - proto_notify_state(p, (q->start ? q->start(p) : PS_UP)); - } - } - else { - if (p->proto_state == PS_START || p->proto_state == PS_UP) - { - /* Going down */ - DBG("Kicking %s down\n", p->name); - PD(p, "Shutting down"); - proto_notify_state(p, (q->shutdown ? q->shutdown(p) : PS_DOWN)); - } + PROTO_LOCKED_FROM_MAIN(p) + proto_shutdown(p); } + else if (!p->active) + proto_start(p); } struct proto * @@ -1524,7 +1983,7 @@ graceful_restart_done(timer *t UNUSED) WALK_LIST(c, p->channels) { /* Resume postponed export of routes */ - if ((c->channel_state == CS_UP) && c->gr_wait && c->proto->rt_notify) + if ((c->channel_state == CS_UP) && c->gr_wait && p->rt_notify) channel_start_export(c); /* Cleanup */ @@ -1614,7 +2073,11 @@ protos_dump_all(void) struct proto *p; WALK_LIST(p, proto_list) { - debug(" protocol %s state %s\n", p->name, p_states[p->proto_state]); +#define DPF(x) (p->x ? " " #x : "") + debug(" protocol %s (%p) state %s with %d active channels flags: %s%s%s%s\n", + p->name, p, p_states[p->proto_state], p->active_channels, + DPF(disabled), DPF(active), DPF(do_stop), DPF(reconfiguring)); +#undef DPF struct channel *c; WALK_LIST(c, p->channels) @@ -1624,6 +2087,23 @@ protos_dump_all(void) debug("\tInput filter: %s\n", filter_name(c->in_filter)); if (c->out_filter) debug("\tOutput filter: %s\n", filter_name(c->out_filter)); + debug("\tChannel state: %s/%s/%s\n", c_states[c->channel_state], + c->in_req.hook ? rt_import_state_name(rt_import_get_state(c->in_req.hook)) : "-", + c->out_req.hook ? rt_export_state_name(rt_export_get_state(c->out_req.hook)) : "-"); + if (c->in_table) + { + debug("\tInput aux table:\n"); + rt_dump_hooks(c->in_table->tab); + rt_dump(c->in_table->tab); + debug("\tEnd of input aux table.\n"); + } + if (c->out_table) + { + debug("\tOutput aux table:\n"); + rt_dump_hooks(c->in_table->tab); + rt_dump(c->in_table->tab); + debug("\tEnd of output aux table.\n"); + } } if (p->proto->dump && (p->proto_state != PS_DOWN)) @@ -1702,7 +2182,7 @@ protos_build(void) proto_build(&proto_perf); #endif - proto_pool = rp_new(&root_pool, "Protocols"); + proto_pool = rp_new(&root_pool, &main_birdloop, "Protocols"); proto_shutdown_timer = tm_new(proto_pool); proto_shutdown_timer->hook = proto_shutdown_loop; } @@ -1780,108 +2260,132 @@ proto_set_message(struct proto *p, char *msg, int len) } -static const char * -channel_limit_name(struct channel_limit *l) -{ - const char *actions[] = { - [PLA_WARN] = "warn", - [PLA_BLOCK] = "block", - [PLA_RESTART] = "restart", - [PLA_DISABLE] = "disable", - }; +static const char * channel_limit_name[] = { + [PLA_WARN] = "warn", + [PLA_BLOCK] = "block", + [PLA_RESTART] = "restart", + [PLA_DISABLE] = "disable", +}; - return actions[l->action]; -} -/** - * channel_notify_limit: notify about limit hit and take appropriate action - * @c: channel - * @l: limit being hit - * @dir: limit direction (PLD_*) - * @rt_count: the number of routes - * - * The function is called by the route processing core when limit @l - * is breached. It activates the limit and tooks appropriate action - * according to @l->action. - */ -void -channel_notify_limit(struct channel *c, struct channel_limit *l, int dir, u32 rt_count) +static void +channel_log_limit(struct channel *c, struct limit *l, int dir) { const char *dir_name[PLD_MAX] = { "receive", "import" , "export" }; - const byte dir_down[PLD_MAX] = { PDC_RX_LIMIT_HIT, PDC_IN_LIMIT_HIT, PDC_OUT_LIMIT_HIT }; - struct proto *p = c->proto; + log(L_WARN "Channel %s.%s hits route %s limit (%d), action: %s", + c->proto->name, c->name, dir_name[dir], l->max, channel_limit_name[c->limit_actions[dir]]); +} - if (l->state == PLS_BLOCKED) +static void +channel_activate_limit(struct channel *c, struct limit *l, int dir) +{ + if (c->limit_active & (1 << dir)) return; - /* For warning action, we want the log message every time we hit the limit */ - if (!l->state || ((l->action == PLA_WARN) && (rt_count == l->limit))) - log(L_WARN "Protocol %s hits route %s limit (%d), action: %s", - p->name, dir_name[dir], l->limit, channel_limit_name(l)); + c->limit_active |= (1 << dir); + channel_log_limit(c, l, dir); +} - switch (l->action) - { - case PLA_WARN: - l->state = PLS_ACTIVE; - break; +static int +channel_limit_warn(struct limit *l, void *data) +{ + struct channel_limit_data *cld = data; + struct channel *c = cld->c; + int dir = cld->dir; - case PLA_BLOCK: - l->state = PLS_BLOCKED; - break; + channel_log_limit(c, l, dir); - case PLA_RESTART: - case PLA_DISABLE: - l->state = PLS_BLOCKED; - if (p->proto_state == PS_UP) - proto_schedule_down(p, l->action == PLA_RESTART, dir_down[dir]); - break; - } + return 0; } -static void -channel_verify_limits(struct channel *c) +static int +channel_limit_block(struct limit *l, void *data) +{ + struct channel_limit_data *cld = data; + struct channel *c = cld->c; + int dir = cld->dir; + + channel_activate_limit(c, l, dir); + + return 1; +} + +static const byte chl_dir_down[PLD_MAX] = { PDC_RX_LIMIT_HIT, PDC_IN_LIMIT_HIT, PDC_OUT_LIMIT_HIT }; + +static int +channel_limit_down(struct limit *l, void *data) { - struct channel_limit *l; - u32 all_routes = c->stats.imp_routes + c->stats.filt_routes; + struct channel_limit_data *cld = data; + struct channel *c = cld->c; + struct proto *p = c->proto; + int dir = cld->dir; - l = &c->rx_limit; - if (l->action && (all_routes > l->limit)) - channel_notify_limit(c, l, PLD_RX, all_routes); + channel_activate_limit(c, l, dir); - l = &c->in_limit; - if (l->action && (c->stats.imp_routes > l->limit)) - channel_notify_limit(c, l, PLD_IN, c->stats.imp_routes); + if (p->proto_state == PS_UP) + proto_schedule_down(p, c->limit_actions[dir] == PLA_RESTART, chl_dir_down[dir]); - l = &c->out_limit; - if (l->action && (c->stats.exp_routes > l->limit)) - channel_notify_limit(c, l, PLD_OUT, c->stats.exp_routes); + return 1; } -static inline void -channel_reset_limit(struct channel_limit *l) +static int (*channel_limit_action[])(struct limit *, void *) = { + [PLA_NONE] = NULL, + [PLA_WARN] = channel_limit_warn, + [PLA_BLOCK] = channel_limit_block, + [PLA_RESTART] = channel_limit_down, + [PLA_DISABLE] = channel_limit_down, +}; + +static void +channel_update_limit(struct channel *c, struct limit *l, int dir, struct channel_limit *cf) +{ + l->action = channel_limit_action[cf->action]; + c->limit_actions[dir] = cf->action; + + struct channel_limit_data cld = { .c = c, .dir = dir }; + limit_update(l, &cld, cf->action ? cf->limit : ~((u32) 0)); +} + +static void +channel_init_limit(struct channel *c, struct limit *l, int dir, struct channel_limit *cf) { - if (l->action) - l->state = PLS_INITIAL; + channel_reset_limit(c, l, dir); + channel_update_limit(c, l, dir, cf); } +static void +channel_reset_limit(struct channel *c, struct limit *l, int dir) +{ + limit_reset(l); + c->limit_active &= ~(1 << dir); +} + +static struct rte_owner_class default_rte_owner_class; + static inline void proto_do_start(struct proto *p) { p->active = 1; - p->do_start = 1; - ev_schedule(p->event); + + rt_init_sources(&p->sources, p->name, proto_work_list(p)); + if (!p->sources.class) + p->sources.class = &default_rte_owner_class; + + if (!p->cf->late_if_feed) + if_feed_baby(p); } static void proto_do_up(struct proto *p) { if (!p->main_source) - { p->main_source = rt_get_source(p, 0); - rt_lock_source(p->main_source); - } + // Locked automaticaly proto_start_channels(p); + + if (p->cf->late_if_feed) + if_feed_baby(p); } static inline void @@ -1896,9 +2400,6 @@ proto_do_stop(struct proto *p) p->down_sched = 0; p->gr_recovery = 0; - p->do_stop = 1; - ev_schedule(p->event); - if (p->main_source) { rt_unlock_source(p->main_source); @@ -1906,6 +2407,10 @@ proto_do_stop(struct proto *p) } proto_stop_channels(p); + rt_destroy_sources(&p->sources, p->event); + + p->do_stop = 1; + proto_send_event(p); } static void @@ -1913,12 +2418,10 @@ proto_do_down(struct proto *p) { p->down_code = 0; neigh_prune(); - rfree(p->pool); - p->pool = NULL; /* Shutdown is finished in the protocol event */ if (proto_is_done(p)) - ev_schedule(p->event); + proto_send_event(p); } @@ -2009,38 +2512,58 @@ proto_state_name(struct proto *p) static void channel_show_stats(struct channel *c) { - struct proto_stats *s = &c->stats; + struct channel_import_stats *ch_is = &c->import_stats; + struct channel_export_stats *ch_es = &c->export_stats; + struct rt_import_stats *rt_is = c->in_req.hook ? &c->in_req.hook->stats : NULL; + struct rt_export_stats *rt_es = c->out_req.hook ? &c->out_req.hook->stats : NULL; + +#define SON(ie, item) ((ie) ? (ie)->item : 0) +#define SCI(item) SON(ch_is, item) +#define SCE(item) SON(ch_es, item) +#define SRI(item) SON(rt_is, item) +#define SRE(item) SON(rt_es, item) + + u32 rx_routes = c->rx_limit.count; + u32 in_routes = c->in_limit.count; + u32 out_routes = c->out_limit.count; if (c->in_keep_filtered) cli_msg(-1006, " Routes: %u imported, %u filtered, %u exported, %u preferred", - s->imp_routes, s->filt_routes, s->exp_routes, s->pref_routes); + in_routes, (rx_routes - in_routes), out_routes, SRI(pref)); else cli_msg(-1006, " Routes: %u imported, %u exported, %u preferred", - s->imp_routes, s->exp_routes, s->pref_routes); - - cli_msg(-1006, " Route change stats: received rejected filtered ignored accepted"); - cli_msg(-1006, " Import updates: %10u %10u %10u %10u %10u", - s->imp_updates_received, s->imp_updates_invalid, - s->imp_updates_filtered, s->imp_updates_ignored, - s->imp_updates_accepted); - cli_msg(-1006, " Import withdraws: %10u %10u --- %10u %10u", - s->imp_withdraws_received, s->imp_withdraws_invalid, - s->imp_withdraws_ignored, s->imp_withdraws_accepted); - cli_msg(-1006, " Export updates: %10u %10u %10u --- %10u", - s->exp_updates_received, s->exp_updates_rejected, - s->exp_updates_filtered, s->exp_updates_accepted); - cli_msg(-1006, " Export withdraws: %10u --- --- --- %10u", - s->exp_withdraws_received, s->exp_withdraws_accepted); + in_routes, out_routes, SRI(pref)); + + cli_msg(-1006, " Route change stats: received rejected filtered ignored limited accepted"); + cli_msg(-1006, " Import updates: %10u %10u %10u %10u %10u %10u", + SCI(updates_received), SCI(updates_invalid), + SCI(updates_filtered), SRI(updates_ignored), + SCI(updates_limited_rx) + SCI(updates_limited_in), + SRI(updates_accepted)); + cli_msg(-1006, " Import withdraws: %10u %10u --- %10u --- %10u", + SCI(withdraws_received), SCI(withdraws_invalid), + SRI(withdraws_ignored), SRI(withdraws_accepted)); + cli_msg(-1006, " Export updates: %10u %10u %10u --- %10u %10u", + SRE(updates_received), SCE(updates_rejected), + SCE(updates_filtered), SCE(updates_limited), SCE(updates_accepted)); + cli_msg(-1006, " Export withdraws: %10u --- --- --- ---%10u", + SRE(withdraws_received), SCE(withdraws_accepted)); + +#undef SRI +#undef SRE +#undef SCI +#undef SCE +#undef SON } void -channel_show_limit(struct channel_limit *l, const char *dsc) +channel_show_limit(struct limit *l, const char *dsc, int active, int action) { if (!l->action) return; - cli_msg(-1006, " %-16s%d%s", dsc, l->limit, l->state ? " [HIT]" : ""); - cli_msg(-1006, " Action: %s", channel_limit_name(l)); + cli_msg(-1006, " %-16s%d%s", dsc, l->max, active ? " [HIT]" : ""); + cli_msg(-1006, " Action: %s", channel_limit_name[action]); } void @@ -2048,6 +2571,8 @@ channel_show_info(struct channel *c) { cli_msg(-1006, " Channel %s", c->name); cli_msg(-1006, " State: %s", c_states[c->channel_state]); + cli_msg(-1006, " Import state: %s", rt_import_state_name(rt_import_get_state(c->in_req.hook))); + cli_msg(-1006, " Export state: %s", rt_export_state_name(rt_export_get_state(c->out_req.hook))); cli_msg(-1006, " Table: %s", c->table->name); cli_msg(-1006, " Preference: %d", c->preference); cli_msg(-1006, " Input filter: %s", filter_name(c->in_filter)); @@ -2058,9 +2583,9 @@ channel_show_info(struct channel *c) c->gr_lock ? " pending" : "", c->gr_wait ? " waiting" : ""); - channel_show_limit(&c->rx_limit, "Receive limit:"); - channel_show_limit(&c->in_limit, "Import limit:"); - channel_show_limit(&c->out_limit, "Export limit:"); + channel_show_limit(&c->rx_limit, "Receive limit:", c->limit_active & (1 << PLD_RX), c->limit_actions[PLD_RX]); + channel_show_limit(&c->in_limit, "Import limit:", c->limit_active & (1 << PLD_IN), c->limit_actions[PLD_IN]); + channel_show_limit(&c->out_limit, "Export limit:", c->limit_active & (1 << PLD_OUT), c->limit_actions[PLD_OUT]); if (c->channel_state != CS_DOWN) channel_show_stats(c); @@ -2106,8 +2631,8 @@ proto_cmd_show(struct proto *p, uintptr_t verbose, int cnt) cli_msg(-1006, " Message: %s", p->message); if (p->cf->router_id) cli_msg(-1006, " Router ID: %R", p->cf->router_id); - if (p->vrf_set) - cli_msg(-1006, " VRF: %s", p->vrf ? p->vrf->name : "default"); + if (p->vrf) + cli_msg(-1006, " VRF: %s", p->vrf->name); if (p->proto->show_proto_info) p->proto->show_proto_info(p); @@ -2135,7 +2660,7 @@ proto_cmd_disable(struct proto *p, uintptr_t arg, int cnt UNUSED) p->disabled = 1; p->down_code = PDC_CMD_DISABLE; proto_set_message(p, (char *) arg, -1); - proto_rethink_goal(p); + proto_shutdown(p); cli_msg(-9, "%s: disabled", p->name); } @@ -2168,9 +2693,9 @@ proto_cmd_restart(struct proto *p, uintptr_t arg, int cnt UNUSED) p->disabled = 1; p->down_code = PDC_CMD_RESTART; proto_set_message(p, (char *) arg, -1); - proto_rethink_goal(p); + proto_shutdown(p); p->disabled = 0; - proto_rethink_goal(p); + /* After the protocol shuts down, proto_rethink_goal() is run from proto_event. */ cli_msg(-12, "%s: restarted", p->name); } @@ -2243,8 +2768,15 @@ proto_apply_cmd_symbol(const struct symbol *s, void (* cmd)(struct proto *, uint return; } - cmd(s->proto->proto, arg, 0); - cli_msg(0, ""); + if (s->proto->proto) + { + struct proto *p = s->proto->proto; + PROTO_LOCKED_FROM_MAIN(p) + cmd(p, arg, 0); + cli_msg(0, ""); + } + else + cli_msg(9002, "%s does not exist", s->name); } static void @@ -2255,7 +2787,8 @@ proto_apply_cmd_patt(const char *patt, void (* cmd)(struct proto *, uintptr_t, i WALK_LIST(p, proto_list) if (!patt || patmatch(patt, p->name)) - cmd(p, arg, cnt++); + PROTO_LOCKED_FROM_MAIN(p) + cmd(p, arg, cnt++); if (!cnt) cli_msg(8003, "No protocols match"); diff --git a/nest/protocol.h b/nest/protocol.h index abcc505d..1d4f2059 100644 --- a/nest/protocol.h +++ b/nest/protocol.h @@ -13,11 +13,11 @@ #include "lib/resource.h" #include "lib/event.h" #include "nest/route.h" +#include "nest/limit.h" #include "conf/conf.h" struct iface; struct ifa; -struct rtable; struct rte; struct neighbor; struct rta; @@ -74,12 +74,9 @@ struct protocol { struct proto * (*init)(struct proto_config *); /* Create new instance */ int (*reconfigure)(struct proto *, struct proto_config *); /* Try to reconfigure instance, returns success */ void (*dump)(struct proto *); /* Debugging dump */ - void (*dump_attrs)(struct rte *); /* Dump protocol-dependent attributes */ int (*start)(struct proto *); /* Start the instance */ int (*shutdown)(struct proto *); /* Stop the instance */ - void (*cleanup)(struct proto *); /* Called after shutdown when protocol became hungry/down */ void (*get_status)(struct proto *, byte *buf); /* Get instance status (for `show protocols' command) */ - void (*get_route_info)(struct rte *, byte *buf); /* Get route information (for `show route' command) */ int (*get_attr)(const struct eattr *, byte *buf, int buflen); /* ASCIIfy dynamic attribute (returns GA_*) */ void (*show_proto_info)(struct proto *); /* Show protocol info (for `show protocols all' command) */ void (*copy_config)(struct proto_config *, struct proto_config *); /* Copy config from given protocol instance */ @@ -120,9 +117,10 @@ struct proto_config { int class; /* SYM_PROTO or SYM_TEMPLATE */ u8 net_type; /* Protocol network type (NET_*), 0 for undefined */ u8 disabled; /* Protocol enabled/disabled by default */ - u8 vrf_set; /* Related VRF instance (below) is defined */ + u8 late_if_feed; /* Delay interface feed after channels are up */ u32 debug, mrtdump; /* Debugging bitfields, both use D_* constants */ u32 router_id; /* Protocol specific router ID */ + uint loop_order; /* Launch a birdloop on this locking level; use DOMAIN_ORDER(the_bird) for mainloop */ list channels; /* List of channel configs (struct channel_config) */ struct iface *vrf; /* Related VRF instance, NULL if global */ @@ -133,31 +131,6 @@ struct proto_config { }; /* Protocol statistics */ -struct proto_stats { - /* Import - from protocol to core */ - u32 imp_routes; /* Number of routes successfully imported to the (adjacent) routing table */ - u32 filt_routes; /* Number of routes rejected in import filter but kept in the routing table */ - u32 pref_routes; /* Number of routes selected as best in the (adjacent) routing table */ - u32 imp_updates_received; /* Number of route updates received */ - u32 imp_updates_invalid; /* Number of route updates rejected as invalid */ - u32 imp_updates_filtered; /* Number of route updates rejected by filters */ - u32 imp_updates_ignored; /* Number of route updates rejected as already in route table */ - u32 imp_updates_accepted; /* Number of route updates accepted and imported */ - u32 imp_withdraws_received; /* Number of route withdraws received */ - u32 imp_withdraws_invalid; /* Number of route withdraws rejected as invalid */ - u32 imp_withdraws_ignored; /* Number of route withdraws rejected as already not in route table */ - u32 imp_withdraws_accepted; /* Number of route withdraws accepted and processed */ - - /* Export - from core to protocol */ - u32 exp_routes; /* Number of routes successfully exported to the protocol */ - u32 exp_updates_received; /* Number of route updates received */ - u32 exp_updates_rejected; /* Number of route updates rejected by protocol */ - u32 exp_updates_filtered; /* Number of route updates rejected by filters */ - u32 exp_updates_accepted; /* Number of route updates accepted and exported */ - u32 exp_withdraws_received; /* Number of route withdraws received */ - u32 exp_withdraws_accepted; /* Number of route withdraws accepted and processed */ -}; - struct proto { node n; /* Node in global proto_list */ struct protocol *proto; /* Protocol */ @@ -165,22 +138,23 @@ struct proto { struct proto_config *cf_new; /* Configuration we want to switch to after shutdown (NULL=delete) */ pool *pool; /* Pool containing local objects */ event *event; /* Protocol event */ + struct birdloop *loop; /* BIRDloop running this protocol */ list channels; /* List of channels to rtables (struct channel) */ struct channel *main_channel; /* Primary channel */ struct rte_src *main_source; /* Primary route source */ + struct rte_owner sources; /* Route source owner structure */ struct iface *vrf; /* Related VRF instance, NULL if global */ const char *name; /* Name of this instance (== cf->name) */ u32 debug; /* Debugging flags */ u32 mrtdump; /* MRTDump flags */ uint active_channels; /* Number of active channels */ + uint active_coroutines; /* Number of active coroutines */ byte net_type; /* Protocol network type (NET_*), 0 for undefined */ byte disabled; /* Manually disabled */ - byte vrf_set; /* Related VRF instance (above) is defined */ byte proto_state; /* Protocol state machine (PS_*, see below) */ byte active; /* From PS_START to cleanup after PS_STOP */ - byte do_start; /* Start actions are scheduled */ byte do_stop; /* Stop actions are scheduled */ byte reconfiguring; /* We're shutting down due to reconfiguration */ byte gr_recovery; /* Protocol should participate in graceful restart recovery */ @@ -198,12 +172,11 @@ struct proto { * ifa_notify Notify protocol about interface address changes. * rt_notify Notify protocol about routing table updates. * neigh_notify Notify protocol about neighbor cache events. - * make_tmp_attrs Add attributes to rta from from private attrs stored in rte. The route and rta MUST NOT be cached. - * store_tmp_attrs Store private attrs back to rte and undef added attributes. The route and rta MUST NOT be cached. - * preexport Called as the first step of the route exporting process. - * It can construct a new rte, add private attributes and - * decide whether the route shall be exported: 1=yes, -1=no, - * 0=process it through the export filter set by the user. + * preexport Called as the first step of the route exporting process. + * It can decide whether the route shall be exported: + * -1 = reject, + * 0 = continue to export filter + * 1 = accept immediately * reload_routes Request channel to reload all its routes to the core * (using rte_update()). Returns: 0=reload cannot be done, * 1= reload is scheduled and will happen (asynchronously). @@ -213,11 +186,9 @@ struct proto { void (*if_notify)(struct proto *, unsigned flags, struct iface *i); void (*ifa_notify)(struct proto *, unsigned flags, struct ifa *a); - void (*rt_notify)(struct proto *, struct channel *, struct network *net, struct rte *new, struct rte *old); + void (*rt_notify)(struct proto *, struct channel *, const net_addr *net, struct rte *new, const struct rte *old); void (*neigh_notify)(struct neighbor *neigh); - void (*make_tmp_attrs)(struct rte *rt, struct linpool *pool); - void (*store_tmp_attrs)(struct rte *rt, struct linpool *pool); - int (*preexport)(struct proto *, struct rte **rt, struct linpool *pool); + int (*preexport)(struct channel *, struct rte *rt); void (*reload_routes)(struct channel *); void (*feed_begin)(struct channel *, int initial); void (*feed_end)(struct channel *); @@ -233,13 +204,12 @@ struct proto { * rte_remove Called whenever a rte is removed from the routing table. */ - int (*rte_recalculate)(struct rtable *, struct network *, struct rte *, struct rte *, struct rte *); + int (*rte_recalculate)(rtable *, struct network *, struct rte *, struct rte *, struct rte *); int (*rte_better)(struct rte *, struct rte *); - int (*rte_same)(struct rte *, struct rte *); int (*rte_mergable)(struct rte *, struct rte *); - struct rte * (*rte_modify)(struct rte *, struct linpool *); void (*rte_insert)(struct network *, struct rte *); void (*rte_remove)(struct network *, struct rte *); + u32 (*rte_igp_metric)(struct rte *); /* Hic sunt protocol-specific data */ }; @@ -279,7 +249,7 @@ void channel_graceful_restart_unlock(struct channel *c); #define DEFAULT_GR_WAIT 240 -void channel_show_limit(struct channel_limit *l, const char *dsc); +void channel_show_limit(struct limit *l, const char *dsc, int active, int action); void channel_show_info(struct channel *c); void channel_cmd_debug(struct channel *c, uint mask); @@ -386,6 +356,8 @@ void proto_notify_state(struct proto *p, unsigned state); * as a result of received ROUTE-REFRESH request). */ +static inline int proto_is_inactive(struct proto *p) +{ return (p->active_channels == 0) && (p->active_coroutines == 0) && (p->sources.uc == 0); } /* @@ -434,19 +406,30 @@ extern struct proto_config *cf_dev_proto; #define PLA_RESTART 4 /* Force protocol restart */ #define PLA_DISABLE 5 /* Shutdown and disable protocol */ -#define PLS_INITIAL 0 /* Initial limit state after protocol start */ -#define PLS_ACTIVE 1 /* Limit was hit */ -#define PLS_BLOCKED 2 /* Limit is active and blocking new routes */ - struct channel_limit { u32 limit; /* Maximum number of prefixes */ u8 action; /* Action to take (PLA_*) */ - u8 state; /* State of limit (PLS_*) */ }; -void channel_notify_limit(struct channel *c, struct channel_limit *l, int dir, u32 rt_count); +struct channel_limit_data { + struct channel *c; + int dir; +}; + +#define CLP__RX(_c) (&(_c)->rx_limit) +#define CLP__IN(_c) (&(_c)->in_limit) +#define CLP__OUT(_c) (&(_c)->out_limit) +#if 0 +#define CHANNEL_LIMIT_LOG(_c, _dir, _op) log(L_TRACE "%s.%s: %s limit %s %u", (_c)->proto->name, (_c)->name, #_dir, _op, (CLP__##_dir(_c))->count) +#else +#define CHANNEL_LIMIT_LOG(_c, _dir, _op) +#endif + +#define CHANNEL_LIMIT_PUSH(_c, _dir) ({ CHANNEL_LIMIT_LOG(_c, _dir, "push from"); struct channel_limit_data cld = { .c = (_c), .dir = PLD_##_dir }; limit_push(CLP__##_dir(_c), &cld); }) +#define CHANNEL_LIMIT_POP(_c, _dir) ({ limit_pop(CLP__##_dir(_c)); CHANNEL_LIMIT_LOG(_c, _dir, "pop to"); }) + /* * Channels */ @@ -469,7 +452,6 @@ struct channel_class { void (*dump)(struct proto *); /* Debugging dump */ - void (*dump_attrs)(struct rte *); /* Dump protocol-dependent attributes */ void (*get_status)(struct proto *, byte *buf); /* Get instance status (for `show protocols' command) */ void (*get_route_info)(struct rte *, byte *buf); /* Get route information (for `show route' command) */ @@ -489,6 +471,7 @@ struct channel_config { struct proto_config *parent; /* Where channel is defined (proto or template) */ struct rtable_config *table; /* Table we're attached to */ const struct filter *in_filter, *out_filter; /* Attached filters */ + struct channel_limit rx_limit; /* Limit for receiving routes from protocol (relevant when in_keep_filtered is active) */ struct channel_limit in_limit; /* Limit for importing routes from protocol */ @@ -505,23 +488,50 @@ struct channel_config { struct channel { node n; /* Node in proto->channels */ - node table_node; /* Node in table->channels */ const char *name; /* Channel name (may be NULL) */ const struct channel_class *channel; struct proto *proto; - struct rtable *table; + rtable *table; const struct filter *in_filter; /* Input filter */ const struct filter *out_filter; /* Output filter */ - struct bmap export_map; /* Keeps track which routes passed export filter */ - struct channel_limit rx_limit; /* Receive limit (for in_keep_filtered) */ - struct channel_limit in_limit; /* Input limit */ - struct channel_limit out_limit; /* Output limit */ - - struct event *feed_event; /* Event responsible for feeding */ - struct fib_iterator feed_fit; /* Routing table iterator used during feeding */ - struct proto_stats stats; /* Per-channel protocol statistics */ + struct bmap export_map; /* Keeps track which routes were really exported */ + struct bmap export_reject_map; /* Keeps track which routes were rejected by export filter */ + + struct limit rx_limit; /* Receive limit (for in_keep_filtered) */ + struct limit in_limit; /* Input limit */ + struct limit out_limit; /* Output limit */ + + u8 limit_actions[PLD_MAX]; /* Limit actions enum */ + u8 limit_active; /* Flags for active limits */ + + linpool *rte_update_pool; + uint rte_update_nest_cnt; + + struct channel_import_stats { + /* Import - from protocol to core */ + u32 updates_received; /* Number of route updates received */ + u32 updates_invalid; /* Number of route updates rejected as invalid */ + u32 updates_filtered; /* Number of route updates rejected by filters */ + u32 updates_limited_rx; /* Number of route updates exceeding the rx_limit */ + u32 updates_limited_in; /* Number of route updates exceeding the in_limit */ + u32 withdraws_received; /* Number of route withdraws received */ + u32 withdraws_invalid; /* Number of route withdraws rejected as invalid */ + } import_stats; + + struct channel_export_stats { + /* Export - from core to protocol */ + u32 updates_rejected; /* Number of route updates rejected by protocol */ + u32 updates_filtered; /* Number of route updates rejected by filters */ + u32 updates_accepted; /* Number of route updates accepted and exported */ + u32 updates_limited; /* Number of route updates exceeding the out_limit */ + u32 withdraws_accepted; /* Number of route withdraws accepted and processed */ + } export_stats; + + struct rt_import_request in_req; /* Table import connection */ + struct rt_export_request out_req; /* Table export connection */ + u32 refeed_count; /* Number of routes exported during refeed regardless of out_limit */ u8 net_type; /* Routing table network type (NET_*), 0 for undefined */ @@ -534,31 +544,35 @@ struct channel { u8 stale; /* Used in reconfiguration */ u8 channel_state; - u8 export_state; /* Route export state (ES_*, see below) */ - u8 feed_active; - u8 flush_active; - u8 refeeding; /* We are refeeding (valid only if export_state == ES_FEEDING) */ + u8 refeeding; /* Refeeding the channel. */ u8 reloadable; /* Hook reload_routes() is allowed on the channel */ u8 gr_lock; /* Graceful restart mechanism should wait for this channel */ u8 gr_wait; /* Route export to channel is postponed until graceful restart */ + u8 restart_export; /* Route export should restart as soon as it stops */ btime last_state_change; /* Time of last state transition */ - struct rtable *in_table; /* Internal table for received routes */ - struct event *reload_event; /* Event responsible for reloading from in_table */ - struct fib_iterator reload_fit; /* FIB iterator in in_table used during reloading */ - struct rte *reload_next_rte; /* Route iterator in in_table used during reloading */ - u8 reload_active; /* Iterator reload_fit is linked */ + struct channel_aux_table *in_table; /* Internal table for received routes */ + struct event in_stopped; /* Import stop callback */ u8 reload_pending; /* Reloading and another reload is scheduled */ u8 refeed_pending; /* Refeeding and another refeed is scheduled */ u8 rpki_reload; /* RPKI changes trigger channel reload */ - struct rtable *out_table; /* Internal table for exported routes */ + struct channel_aux_table *out_table; /* Internal table for exported routes */ list roa_subscriptions; /* List of active ROA table subscriptions based on filters roa_check() */ }; +struct channel_aux_table { + struct channel *c; + struct rt_import_request push; + struct rt_export_request get; + event push_stopped; + rtable *tab; + event *stop; + u8 refeed_pending; +}; /* * Channel states @@ -585,70 +599,59 @@ struct channel { * restricted by that and is on volition of the protocol. Generally, channels * are opened in protocols' start() hooks when going to PS_UP. * - * CS_FLUSHING - The transitional state between initialized channel and closed + * CS_STOP - The transitional state between initialized channel and closed * channel. The channel is still initialized, but no route exchange is allowed. * Instead, the associated table is running flush loop to remove routes imported * through the channel. After that, the channel changes state to CS_DOWN and * is detached from the table (the table is unlocked and the channel is unlinked - * from it). Unlike other states, the CS_FLUSHING state is not explicitly + * from it). Unlike other states, the CS_STOP state is not explicitly * entered or left by the protocol. A protocol may request to close a channel * (by calling channel_close()), which causes the channel to change state to - * CS_FLUSHING and later to CS_DOWN. Also note that channels are closed + * CS_STOP and later to CS_DOWN. Also note that channels are closed * automatically by the core when the protocol is going down. * + * CS_PAUSE - Almost the same as CS_STOP, just the table import is kept and + * the table export is stopped before transitioning to CS_START. + * * Allowed transitions: * * CS_DOWN -> CS_START / CS_UP - * CS_START -> CS_UP / CS_FLUSHING - * CS_UP -> CS_START / CS_FLUSHING - * CS_FLUSHING -> CS_DOWN (automatic) + * CS_START -> CS_UP / CS_STOP + * CS_UP -> CS_PAUSE / CS_STOP + * CS_PAUSE -> CS_START (automatic) + * CS_STOP -> CS_DOWN (automatic) */ #define CS_DOWN 0 #define CS_START 1 #define CS_UP 2 -#define CS_FLUSHING 3 - -#define ES_DOWN 0 -#define ES_FEEDING 1 -#define ES_READY 2 - +#define CS_STOP 3 +#define CS_PAUSE 4 struct channel_config *proto_cf_find_channel(struct proto_config *p, uint net_type); static inline struct channel_config *proto_cf_main_channel(struct proto_config *pc) { return proto_cf_find_channel(pc, pc->net_type); } -struct channel *proto_find_channel_by_table(struct proto *p, struct rtable *t); +struct channel *proto_find_channel_by_table(struct proto *p, rtable *t); struct channel *proto_find_channel_by_name(struct proto *p, const char *n); struct channel *proto_add_channel(struct proto *p, struct channel_config *cf); int proto_configure_channel(struct proto *p, struct channel **c, struct channel_config *cf); void channel_set_state(struct channel *c, uint state); -void channel_setup_in_table(struct channel *c); +void channel_setup_in_table(struct channel *c, int best); void channel_setup_out_table(struct channel *c); void channel_schedule_reload(struct channel *c); static inline void channel_init(struct channel *c) { channel_set_state(c, CS_START); } static inline void channel_open(struct channel *c) { channel_set_state(c, CS_UP); } -static inline void channel_close(struct channel *c) { channel_set_state(c, CS_FLUSHING); } +static inline void channel_close(struct channel *c) { channel_set_state(c, CS_STOP); } void channel_request_feeding(struct channel *c); +void channel_request_reload(struct channel *c); +void channel_refresh_begin(struct channel *c); +void channel_refresh_end(struct channel *c); void *channel_config_new(const struct channel_class *cc, const char *name, uint net_type, struct proto_config *proto); void *channel_config_get(const struct channel_class *cc, const char *name, uint net_type, struct proto_config *proto); int channel_reconfigure(struct channel *c, struct channel_config *cf); - -/* Moved from route.h to avoid dependency conflicts */ -static inline void rte_update(struct proto *p, const net_addr *n, rte *new) { rte_update2(p->main_channel, n, new, p->main_source); } - -static inline void -rte_update3(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) -{ - if (c->in_table && !rte_update_in(c, n, new, src)) - return; - - rte_update2(c, n, new, src); -} - - #endif diff --git a/nest/route.h b/nest/route.h index f5fc9e31..f060ad63 100644 --- a/nest/route.h +++ b/nest/route.h @@ -2,6 +2,7 @@ * BIRD Internet Routing Daemon -- Routing Table * * (c) 1998--2000 Martin Mares <mj@ucw.cz> + * (c) 2019--2021 Maria Matejka <mq@jmq.cz> * * Can be freely distributed and used under the terms of the GNU GPL. */ @@ -10,13 +11,19 @@ #define _BIRD_ROUTE_H_ #include "lib/lists.h" +#include "lib/event.h" #include "lib/bitmap.h" #include "lib/resource.h" #include "lib/net.h" +#include "lib/hash.h" +#include "lib/event.h" + +#include <stdatomic.h> struct ea_list; struct protocol; struct proto; +struct channel; struct rte_src; struct symbol; struct timer; @@ -139,61 +146,90 @@ void fit_copy(struct fib *f, struct fib_iterator *dst, struct fib_iterator *src) * It's guaranteed that there is at most one RTE for every (prefix,proto) pair. */ -struct rtable_config { - node n; - char *name; - struct rtable *table; - struct proto_config *krt_attached; /* Kernel syncer attached to this table */ - uint addr_type; /* Type of address data stored in table (NET_*) */ - int gc_max_ops; /* Maximum number of operations before GC is run */ - int gc_min_time; /* Minimum time between two consecutive GC runs */ - byte sorted; /* Routes of network are sorted according to rte_better() */ - byte internal; /* Internal table of a protocol */ - btime min_settle_time; /* Minimum settle time for notifications */ - btime max_settle_time; /* Maximum settle time for notifications */ -}; - -typedef struct rtable { - resource r; - node n; /* Node in list of all tables */ +typedef struct rtable_private { +#define RTABLE_PUBLIC \ + resource r; \ + node n; /* Node in list of all tables */ \ + struct birdloop *loop; /* This loop runs the table */ \ + char *name; /* Name of this table */ \ + uint addr_type; /* Type of address data stored in table (NET_*) */ \ + struct rtable_config *config; /* Configuration of this table */ \ + struct event *nhu_event; /* Event to update next hops */ \ + _Atomic byte nhu_state; /* Next Hop Update state */ \ + + RTABLE_PUBLIC; pool *rp; /* Resource pool to allocate everything from, including itself */ + struct slab *rte_slab; /* Slab to allocate route objects */ struct fib fib; - char *name; /* Name of this table */ - list channels; /* List of attached channels (struct channel) */ - uint addr_type; /* Type of address data stored in table (NET_*) */ - int pipe_busy; /* Pipe loop detection */ int use_count; /* Number of protocols using this table */ u32 rt_count; /* Number of routes in the table */ + u32 rr_count; /* Number of running route refresh requests */ - byte internal; /* Internal table of a protocol */ + list imports; /* Registered route importers */ + list exports; /* Registered route exporters */ struct hmap id_map; struct hostcache *hostcache; - struct rtable_config *config; /* Configuration of this table */ - struct config *deleted; /* Table doesn't exist in current configuration, - * delete as soon as use_count becomes 0 and remove - * obstacle from this routing table. - */ - struct event *rt_event; /* Routing table event */ + struct event *prune_event; /* Event to prune abandoned routes */ + struct event *announce_event; /* Event to announce pending exports */ + struct event *ec_event; /* Event to prune finished exports */ + struct event *hcu_event; /* Event to update host cache */ + void (*delete)(void *); /* Delete callback (in parent loop context) */ btime last_rt_change; /* Last time when route changed */ btime base_settle_time; /* Start time of rtable settling interval */ btime gc_time; /* Time of last GC */ int gc_counter; /* Number of operations since last GC */ byte prune_state; /* Table prune state, 1 -> scheduled, 2-> running */ - byte hcu_scheduled; /* Hostcache update is scheduled */ - byte nhu_state; /* Next Hop Update state */ + + byte cork_active; /* Congestion control activated */ + struct fib_iterator prune_fit; /* Rtable prune FIB iterator */ struct fib_iterator nhu_fit; /* Next Hop Update FIB iterator */ + struct tbf rl_pipe; /* Rate limiting token buffer for pipe collisions */ + + linpool *nhu_lp; /* Linpool used for NHU */ list subscribers; /* Subscribers for notifications */ struct timer *settle_timer; /* Settle time for notifications */ + + list pending_exports; /* List of packed struct rt_pending_export */ + + struct rt_pending_export *first_export; /* First export to announce */ + u64 next_export_seq; /* The next export will have this ID */ +} rtable_private; + +typedef union { + struct { RTABLE_PUBLIC }; + rtable_private priv; } rtable; +#define RT_LOCK(tab) ({ birdloop_enter((tab)->loop); &(tab)->priv; }) +#define RT_UNLOCK(tab) birdloop_leave((tab)->loop) +#define RT_PRIV(tab) ({ ASSERT_DIE(birdloop_inside((tab)->loop)); &(tab)->priv; }) + +#define RT_LOCKED(tpub, tpriv) for (rtable_private *tpriv = RT_LOCK(tpub); tpriv; RT_UNLOCK(tpriv), (tpriv = NULL)) + +struct rtable_config { + node n; + char *name; + void *owner; /* Main config if global table, channel_aux_table if channel table */ + rtable *table; + struct proto_config *krt_attached; /* Kernel syncer attached to this table */ + uint addr_type; /* Type of address data stored in table (NET_*) */ + int gc_max_ops; /* Maximum number of operations before GC is run */ + int gc_min_time; /* Minimum time between two consecutive GC runs */ + byte sorted; /* Routes of network are sorted according to rte_better() */ + btime min_settle_time; /* Minimum settle time for notifications */ + btime max_settle_time; /* Maximum settle time for notifications */ + btime min_rr_settle_time; /* Minimum settle time for notifications when route refresh is running */ + btime max_rr_settle_time; /* Maximum settle time for notifications when route refresh is running */ + uint cork_limit; /* Amount of routes to be pending on export to cork imports */ +}; + struct rt_subscription { node n; rtable *tab; - void (*hook)(struct rt_subscription *b); - void *data; + event *event; }; #define NHU_CLEAN 0 @@ -202,7 +238,8 @@ struct rt_subscription { #define NHU_DIRTY 3 typedef struct network { - struct rte *routes; /* Available routes for this network */ + struct rte_storage *routes; /* Available routes for this network */ + struct rt_pending_export *last, *first; /* Routes with unfinished exports */ struct fib_node n; /* FIB flags reserved for kernel syncer */ } net; @@ -223,7 +260,7 @@ struct hostentry { ip_addr addr; /* IP address of host, part of key */ ip_addr link; /* (link-local) IP address of host, used as gw if host is directly attached */ - struct rtable *tab; /* Dependent table, part of key */ + rtable *tab; /* Dependent table, part of key */ struct hostentry *next; /* Next in hash chain */ unsigned hash_key; /* Hash key */ unsigned uc; /* Use count */ @@ -234,64 +271,178 @@ struct hostentry { }; typedef struct rte { - struct rte *next; - net *net; /* Network this RTE belongs to */ - struct channel *sender; /* Channel used to send the route to the routing table */ struct rta *attrs; /* Attributes of this route */ + const net_addr *net; /* Network this RTE belongs to */ + struct rte_src *src; /* Route source that created the route */ + struct rt_import_hook *sender; /* Import hook used to send the route to the routing table */ + btime lastmod; /* Last modified (set by table) */ u32 id; /* Table specific route id */ - byte flags; /* Flags (REF_...) */ + byte flags; /* Table-specific flags */ byte pflags; /* Protocol-specific flags */ - word pref; /* Route preference */ - btime lastmod; /* Last modified */ - union { /* Protocol-dependent data (metrics etc.) */ -#ifdef CONFIG_RIP - struct { - struct iface *from; /* Incoming iface */ - u8 metric; /* RIP metric */ - u16 tag; /* External route tag */ - } rip; -#endif -#ifdef CONFIG_OSPF - struct { - u32 metric1, metric2; /* OSPF Type 1 and Type 2 metrics */ - u32 tag; /* External route tag */ - u32 router_id; /* Router that originated this route */ - } ospf; -#endif -#ifdef CONFIG_BGP - struct { - u8 suppressed; /* Used for deterministic MED comparison */ - s8 stale; /* Route is LLGR_STALE, -1 if unknown */ - } bgp; -#endif -#ifdef CONFIG_BABEL - struct { - u16 seqno; /* Babel seqno */ - u16 metric; /* Babel metric */ - u64 router_id; /* Babel router id */ - } babel; -#endif - struct { /* Routes generated by krt sync (both temporary and inherited ones) */ - s8 src; /* Alleged route source (see krt.h) */ - u8 proto; /* Kernel source protocol ID */ - u8 seen; /* Seen during last scan */ - u8 best; /* Best route in network, propagated to core */ - u32 metric; /* Kernel metric */ - } krt; - } u; + u8 generation; /* If this route import is based on other previously exported route, + this value should be 1 + MAX(generation of the parent routes). + Otherwise the route is independent and this value is zero. */ + u8 stale_cycle; /* Auxiliary value for route refresh */ } rte; -#define REF_COW 1 /* Copy this rte on write */ +struct rte_storage { + struct rte_storage *next; /* Next in chain */ + struct rte rte; /* Route data */ +}; + +#define RTES_CLONE(r, l) ((r) ? (((*(l)) = (r)->rte), (l)) : NULL) +#define RTES_OR_NULL(r) ((r) ? &((r)->rte) : NULL) + #define REF_FILTERED 2 /* Route is rejected by import filter */ -#define REF_STALE 4 /* Route is stale in a refresh cycle */ -#define REF_DISCARD 8 /* Route is scheduled for discard */ -#define REF_MODIFY 16 /* Route is scheduled for modify */ +#define REF_USE_STALE 4 /* Do not reset route's stale_cycle to the actual value */ /* Route is valid for propagation (may depend on other flags in the future), accepts NULL */ -static inline int rte_is_valid(rte *r) { return r && !(r->flags & REF_FILTERED); } +static inline int rte_is_valid(const rte *r) { return r && !(r->flags & REF_FILTERED); } /* Route just has REF_FILTERED flag */ -static inline int rte_is_filtered(rte *r) { return !!(r->flags & REF_FILTERED); } +static inline int rte_is_filtered(const rte *r) { return !!(r->flags & REF_FILTERED); } + + +/* Table-channel connections */ + +struct rt_import_request { + struct rt_import_hook *hook; /* The table part of importer */ + char *name; + u8 trace_routes; + + event_list *list; /* Where to schedule import events */ + + void (*dump_req)(struct rt_import_request *req); + void (*log_state_change)(struct rt_import_request *req, u8 state); + /* Preimport is called when the @new route is just-to-be inserted, replacing @old. + * Return a route (may be different or modified in-place) to continue or NULL to withdraw. */ + struct rte *(*preimport)(struct rt_import_request *req, struct rte *new, struct rte *old); +}; + +struct rt_import_hook { + node n; + rtable *table; /* The connected table */ + struct rt_import_request *req; /* The requestor */ + + struct rt_import_stats { + /* Import - from protocol to core */ + u32 pref; /* Number of routes selected as best in the (adjacent) routing table */ + u32 updates_ignored; /* Number of route updates rejected as already in route table */ + u32 updates_accepted; /* Number of route updates accepted and imported */ + u32 withdraws_ignored; /* Number of route withdraws rejected as already not in route table */ + u32 withdraws_accepted; /* Number of route withdraws accepted and processed */ + } stats; + + u64 flush_seq; /* Table export seq when the channel announced flushing */ + btime last_state_change; /* Time of last state transition */ + + u8 import_state; /* IS_* */ + u8 stale_set; /* Set this stale_cycle to imported routes */ + u8 stale_valid; /* Routes with this stale_cycle and bigger are considered valid */ + u8 stale_pruned; /* Last prune finished when this value was set at stale_valid */ + u8 stale_pruning; /* Last prune started when this value was set at stale_valid */ + + struct event *export_announce_event; /* Event to run to announce new exports */ + struct event *stopped; /* Event to run when import is stopped */ +}; + +struct rt_pending_export { + struct rt_pending_export * _Atomic next; /* Next export for the same destination */ + struct rte_storage *new, *new_best, *old, *old_best; + u64 seq; /* Sequential ID (table-local) of the pending export */ +}; + +struct rt_export_request { + struct rt_export_hook *hook; /* Table part of the export */ + char *name; + u8 trace_routes; + + event_list *list; /* Where to schedule export events */ + + /* There are two methods of export. You can either request feeding every single change + * or feeding the whole route feed. In case of regular export, &export_one is preferred. + * Anyway, when feeding, &export_bulk is preferred, falling back to &export_one. + * Thus, for RA_OPTIMAL, &export_one is only set, + * for RA_MERGED and RA_ACCEPTED, &export_bulk is only set + * and for RA_ANY, both are set to accomodate for feeding all routes but receiving single changes + */ + void (*export_one)(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe); + void (*export_bulk)(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe, rte **feed, uint count); + + void (*dump_req)(struct rt_export_request *req); + void (*log_state_change)(struct rt_export_request *req, u8); +}; + +struct rt_export_hook { + node n; + rtable *table; /* The connected table */ + + pool *pool; + + struct rt_export_request *req; /* The requestor */ + + struct rt_export_stats { + /* Export - from core to protocol */ + u32 updates_received; /* Number of route updates received */ + u32 withdraws_received; /* Number of route withdraws received */ + } stats; + + struct fib_iterator feed_fit; /* Routing table iterator used during feeding */ + + struct bmap seq_map; /* Keep track which exports were already procesed */ + + struct rt_pending_export * _Atomic last_export;/* Last export processed */ + struct rt_pending_export *rpe_next; /* Next pending export to process */ + + btime last_state_change; /* Time of last state transition */ + + u8 refeed_pending; /* Refeeding and another refeed is scheduled */ + _Atomic u8 export_state; /* Route export state (TES_*, see below) */ + + struct event *event; /* Event running all the export operations */ + + void (*stopped)(struct rt_export_request *); /* Stored callback when export is stopped */ +}; + +extern struct event_cork rt_cork; + +#define TIS_DOWN 0 +#define TIS_UP 1 +#define TIS_STOP 2 +#define TIS_FLUSHING 3 +#define TIS_WAITING 4 +#define TIS_CLEARED 5 +#define TIS_MAX 6 + +#define TES_DOWN 0 +#define TES_HUNGRY 1 +#define TES_FEEDING 2 +#define TES_READY 3 +#define TES_STOP 4 +#define TES_MAX 5 + +void rt_request_import(rtable *tab, struct rt_import_request *req); +void rt_request_export(rtable *tab, struct rt_export_request *req); + +void rt_stop_import(struct rt_import_request *, struct event *stopped); +void rt_stop_export(struct rt_export_request *, void (*stopped)(struct rt_export_request *)); + +const char *rt_import_state_name(u8 state); +const char *rt_export_state_name(u8 state); + +static inline u8 rt_import_get_state(struct rt_import_hook *ih) { return ih ? ih->import_state : TIS_DOWN; } +static inline u8 rt_export_get_state(struct rt_export_hook *eh) { return eh ? eh->export_state : TES_DOWN; } + +void rte_import(struct rt_import_request *req, const net_addr *net, rte *new, struct rte_src *src); + +/* Get next rpe. If src is given, it must match. */ +struct rt_pending_export *rpe_next(struct rt_pending_export *rpe, struct rte_src *src); + +/* Mark the pending export processed */ +void rpe_mark_seen(struct rt_export_hook *hook, struct rt_pending_export *rpe); + +/* Get pending export seen status */ +int rpe_get_seen(struct rt_export_hook *hook, struct rt_pending_export *rpe); /* Types of route announcement, also used as flags */ @@ -307,56 +458,75 @@ static inline int rte_is_filtered(rte *r) { return !!(r->flags & REF_FILTERED); #define RIC_REJECT -1 /* Rejected by protocol */ #define RIC_DROP -2 /* Silently dropped by protocol */ +#define rte_update channel_rte_import +/** + * rte_update - enter a new update to a routing table + * @c: channel doing the update + * @net: network address + * @rte: a &rte representing the new route + * @src: old route source identifier + * + * This function imports a new route to the appropriate table (via the channel). + * Table keys are @net (obligatory) and @rte->attrs->src. + * Both the @net and @rte pointers can be local. + * + * The route attributes (@rte->attrs) are obligatory. They can be also allocated + * locally. Anyway, if you use an already-cached attribute object, you shall + * call rta_clone() on that object yourself. (This semantics may change in future.) + * + * If the route attributes are local, you may set @rte->attrs->src to NULL, then + * the protocol's default route source will be supplied. + * + * When rte_update() gets a route, it automatically validates it. This includes + * checking for validity of the given network and next hop addresses and also + * checking for host-scope or link-scope routes. Then the import filters are + * processed and if accepted, the route is passed to route table recalculation. + * + * The accepted routes are then inserted into the table, replacing the old route + * for the same @net identified by @src. Then the route is announced + * to all the channels connected to the table using the standard export mechanism. + * Setting @rte to NULL makes this a withdraw, otherwise @rte->src must be the same + * as @src. + * + * All memory used for temporary allocations is taken from a special linpool + * @rte_update_pool and freed when rte_update() finishes. + */ +void rte_update(struct channel *c, const net_addr *net, struct rte *rte, struct rte_src *src); + extern list routing_tables; struct config; void rt_init(void); void rt_preconfig(struct config *); void rt_commit(struct config *new, struct config *old); -void rt_lock_table(rtable *); -void rt_unlock_table(rtable *); +void rt_lock_table(rtable_private *); +void rt_unlock_table(rtable_private *); void rt_subscribe(rtable *tab, struct rt_subscription *s); void rt_unsubscribe(struct rt_subscription *s); rtable *rt_setup(pool *, struct rtable_config *); -static inline void rt_shutdown(rtable *r) { rfree(r->rp); } -static inline net *net_find(rtable *tab, const net_addr *addr) { return (net *) fib_find(&tab->fib, addr); } -static inline net *net_find_valid(rtable *tab, const net_addr *addr) -{ net *n = net_find(tab, addr); return (n && rte_is_valid(n->routes)) ? n : NULL; } -static inline net *net_get(rtable *tab, const net_addr *addr) { return (net *) fib_get(&tab->fib, addr); } -void *net_route(rtable *tab, const net_addr *n); +static inline net *net_find(rtable_private *tab, const net_addr *addr) { return (net *) fib_find(&tab->fib, addr); } +static inline net *net_find_valid(rtable_private *tab, const net_addr *addr) +{ net *n = net_find(tab, addr); return (n && n->routes && rte_is_valid(&n->routes->rte)) ? n : NULL; } +static inline net *net_get(rtable_private *tab, const net_addr *addr) { return (net *) fib_get(&tab->fib, addr); } +void *net_route(rtable_private *tab, const net_addr *n); int net_roa_check(rtable *tab, const net_addr *n, u32 asn); -rte *rte_find(net *net, struct rte_src *src); -rte *rte_get_temp(struct rta *); -void rte_update2(struct channel *c, const net_addr *n, rte *new, struct rte_src *src); -/* rte_update() moved to protocol.h to avoid dependency conflicts */ -int rt_examine(rtable *t, net_addr *a, struct proto *p, const struct filter *filter); -rte *rt_export_merged(struct channel *c, net *net, rte **rt_free, linpool *pool, int silent); -void rt_refresh_begin(rtable *t, struct channel *c); -void rt_refresh_end(rtable *t, struct channel *c); -void rt_modify_stale(rtable *t, struct channel *c); -void rt_schedule_prune(rtable *t); -void rte_dump(rte *); -void rte_free(rte *); -rte *rte_do_cow(rte *); -static inline rte * rte_cow(rte *r) { return (r->flags & REF_COW) ? rte_do_cow(r) : r; } -rte *rte_cow_rta(rte *r, linpool *lp); -void rte_init_tmp_attrs(struct rte *r, linpool *lp, uint max); -void rte_make_tmp_attr(struct rte *r, uint id, uint type, uintptr_t val); -void rte_make_tmp_attrs(struct rte **r, struct linpool *pool, struct rta **old_attrs); -uintptr_t rte_store_tmp_attr(struct rte *r, uint id); +int rt_examine(rtable_private *t, net_addr *a, struct channel *c, const struct filter *filter); +rte *rt_export_merged(struct channel *c, rte ** feed, uint count, linpool *pool, int silent); + +void rt_refresh_begin(struct rt_import_request *); +void rt_refresh_end(struct rt_import_request *); +void rt_schedule_prune(rtable_private *t); +void rte_dump(struct rte_storage *); +void rte_free(struct rte_storage *, rtable_private *); +struct rte_storage *rte_store(const rte *, net *net, rtable_private *); void rt_dump(rtable *); void rt_dump_all(void); -int rt_feed_channel(struct channel *c); -void rt_feed_channel_abort(struct channel *c); -int rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src); -int rt_reload_channel(struct channel *c); -void rt_reload_channel_abort(struct channel *c); +void rt_dump_hooks(rtable *); +void rt_dump_hooks_all(void); void rt_prune_sync(rtable *t, int all); -int rte_update_out(struct channel *c, const net_addr *n, rte *new, rte *old0, int refeed); struct rtable_config *rt_new_table(struct symbol *s, uint addr_type); - /* Default limit for ECMP next hops, defined in sysdep code */ extern const int rt_default_ecmp; @@ -379,6 +549,7 @@ struct rt_show_data { struct channel *export_channel; struct config *running_on_config; struct krt_proto *kernel; + struct rt_export_hook *kernel_export_hook; int export_mode, primary_only, filtered, stats, show_for; int table_open; /* Iteration (fit) is open */ @@ -430,30 +601,29 @@ struct nexthop { struct rte_src { struct rte_src *next; /* Hash chain */ - struct proto *proto; /* Protocol the source is based on */ + struct rte_owner *owner; /* Route source owner */ u32 private_id; /* Private ID, assigned by the protocol */ u32 global_id; /* Globally unique ID of the source */ - unsigned uc; /* Use count */ + _Atomic u64 uc; /* Use count */ }; typedef struct rta { - struct rta *next, **pprev; /* Hash chain */ - u32 uc; /* Use count */ + struct rta * _Atomic next, * _Atomic *pprev; /* Hash chain */ + _Atomic u32 uc; /* Use count */ u32 hash_key; /* Hash over important fields */ struct ea_list *eattrs; /* Extended Attribute chain */ - struct rte_src *src; /* Route source that created the route */ struct hostentry *hostentry; /* Hostentry for recursive next-hops */ ip_addr from; /* Advertising router */ u32 igp_metric; /* IGP metric to next hop (for iBGP routes) */ - u8 source; /* Route source (RTS_...) */ - u8 scope; /* Route scope (SCOPE_... -- see ip.h) */ - u8 dest; /* Route destination type (RTD_...) */ - u8 aflags; + u16 cached:1; /* Are attributes cached? */ + u16 source:7; /* Route source (RTS_...) */ + u16 scope:4; /* Route scope (SCOPE_... -- see ip.h) */ + u16 dest:4; /* Route destination type (RTD_...) */ + word pref; struct nexthop nh; /* Next hop */ } rta; -#define RTS_DUMMY 0 /* Dummy route to be removed soon */ #define RTS_STATIC 1 /* Normal static route */ #define RTS_INHERIT 2 /* Route inherited from kernel */ #define RTS_DEVICE 3 /* Device route */ @@ -471,11 +641,6 @@ typedef struct rta { #define RTS_PERF 15 /* Perf checker */ #define RTS_MAX 16 -#define RTC_UNICAST 0 -#define RTC_BROADCAST 1 -#define RTC_MULTICAST 2 -#define RTC_ANYCAST 3 /* IPv6 Anycast */ - #define RTD_NONE 0 /* Undefined next hop */ #define RTD_UNICAST 1 /* Next hop is neighbor router */ #define RTD_BLACKHOLE 2 /* Silently drop packets */ @@ -483,8 +648,6 @@ typedef struct rta { #define RTD_PROHIBIT 4 /* Administratively prohibited */ #define RTD_MAX 5 -#define RTAF_CACHED 1 /* This is a cached rta */ - #define IGP_METRIC_UNKNOWN 0x80000000 /* Default igp_metric used when no other protocol-specific metric is availabe */ @@ -508,8 +671,8 @@ typedef struct eattr { byte flags; /* Protocol-dependent flags */ byte type; /* Attribute type and several flags (EAF_...) */ union { - u32 data; - const struct adata *ptr; /* Attribute data elsewhere */ + uintptr_t data; + const struct adata *ptr; /* Attribute data elsewhere */ } u; } eattr; @@ -517,7 +680,6 @@ typedef struct eattr { #define EA_CODE(proto,id) (((proto) << 8) | (id)) #define EA_ID(ea) ((ea) & 0xff) #define EA_PROTO(ea) ((ea) >> 8) -#define EA_ID_FLAG(ea) (1 << EA_ID(ea)) #define EA_CUSTOM(id) ((id) | EA_CUSTOM_BIT) #define EA_IS_CUSTOM(ea) ((ea) & EA_CUSTOM_BIT) #define EA_CUSTOM_ID(ea) ((ea) & ~EA_CUSTOM_BIT) @@ -540,6 +702,7 @@ const char *ea_custom_name(uint ea); #define EAF_TYPE_AS_PATH 0x06 /* BGP AS path (encoding per RFC 1771:4.3) */ #define EAF_TYPE_BITFIELD 0x09 /* 32-bit embedded bitfield */ #define EAF_TYPE_INT_SET 0x0a /* Set of u32's (e.g., a community list) */ +#define EAF_TYPE_PTR 0x0d /* Pointer to an object */ #define EAF_TYPE_EC_SET 0x0e /* Set of pairs of u32's - ext. community list */ #define EAF_TYPE_LC_SET 0x12 /* Set of triplets of u32's - large community list */ #define EAF_TYPE_UNDEF 0x1f /* `force undefined' entry */ @@ -578,13 +741,52 @@ typedef struct ea_list { #define EALF_SORTED 1 /* Attributes are sorted by code */ #define EALF_BISECT 2 /* Use interval bisection for searching */ #define EALF_CACHED 4 /* Attributes belonging to cached rta */ -#define EALF_TEMP 8 /* Temporary ea_list added by make_tmp_attrs hooks */ -struct rte_src *rt_find_source(struct proto *p, u32 id); -struct rte_src *rt_get_source(struct proto *p, u32 id); -static inline void rt_lock_source(struct rte_src *src) { src->uc++; } -static inline void rt_unlock_source(struct rte_src *src) { src->uc--; } -void rt_prune_sources(void); +struct rte_owner_class { + void (*get_route_info)(struct rte *, byte *buf); /* Get route information (for `show route' command) */ + int (*rte_better)(struct rte *, struct rte *); + int (*rte_mergable)(struct rte *, struct rte *); + u32 (*rte_igp_metric)(struct rte *); +}; + +struct rte_owner { + struct rte_owner_class *class; + int (*rte_recalculate)(rtable_private *, struct network *, struct rte *, struct rte *, struct rte *); + HASH(struct rte_src) hash; + const char *name; + u32 hash_key; + u32 uc; + event_list *list; + event *prune; + event *stop; +}; + +#define RTE_SRC_PU_SHIFT 44 +#define RTE_SRC_IN_PROGRESS (1ULL << RTE_SRC_PU_SHIFT) + +struct rte_src *rt_get_source_o(struct rte_owner *o, u32 id); +#define rt_get_source(p, id) rt_get_source_o(&(p)->sources, (id)) +static inline void rt_lock_source(struct rte_src *src) +{ + u64 uc = atomic_fetch_add_explicit(&src->uc, 1, memory_order_acq_rel); + ASSERT_DIE(uc > 0); +} + +static inline void rt_unlock_source(struct rte_src *src) +{ + u64 uc = atomic_fetch_add_explicit(&src->uc, RTE_SRC_IN_PROGRESS, memory_order_acq_rel); + u64 pending = uc >> RTE_SRC_PU_SHIFT; + uc &= RTE_SRC_IN_PROGRESS - 1; + + ASSERT_DIE(uc > pending); + if (uc == pending + 1) + ev_send(src->owner->list, src->owner->prune); + + atomic_fetch_sub_explicit(&src->uc, RTE_SRC_IN_PROGRESS + 1, memory_order_acq_rel); +} + +void rt_init_sources(struct rte_owner *, const char *name, event_list *list); +void rt_destroy_sources(struct rte_owner *, event *); struct ea_walk_state { ea_list *eattrs; /* Ccurrent ea_list, initially set by caller */ @@ -594,7 +796,7 @@ struct ea_walk_state { eattr *ea_find(ea_list *, unsigned ea); eattr *ea_walk(struct ea_walk_state *s, uint id, uint max); -int ea_get_int(ea_list *, unsigned ea, int def); +uintptr_t ea_get_int(ea_list *, unsigned ea, uintptr_t def); void ea_dump(ea_list *); void ea_sort(ea_list *); /* Sort entries in all sub-lists */ unsigned ea_scan(ea_list *); /* How many bytes do we need for merged ea_list */ @@ -673,24 +875,39 @@ void rta_init(void); static inline size_t rta_size(const rta *a) { return sizeof(rta) + sizeof(u32)*a->nh.labels; } #define RTA_MAX_SIZE (sizeof(rta) + sizeof(u32)*MPLS_MAX_LABEL_STACK) rta *rta_lookup(rta *); /* Get rta equivalent to this one, uc++ */ -static inline int rta_is_cached(rta *r) { return r->aflags & RTAF_CACHED; } -static inline rta *rta_clone(rta *r) { r->uc++; return r; } +static inline int rta_is_cached(rta *r) { return r->cached; } + +static inline rta *rta_clone(rta *r) { + u32 uc = atomic_fetch_add_explicit(&r->uc, 1, memory_order_acq_rel); + ASSERT_DIE(uc > 0); + return r; +} + void rta__free(rta *r); -static inline void rta_free(rta *r) { if (r && !--r->uc) rta__free(r); } +static inline void rta_free(rta *r) { + if (!r) + return; + + u32 uc = atomic_fetch_sub_explicit(&r->uc, 1, memory_order_acq_rel); + if (uc == 1) + rta__free(r); +} + rta *rta_do_cow(rta *o, linpool *lp); static inline rta * rta_cow(rta *r, linpool *lp) { return rta_is_cached(r) ? rta_do_cow(r, lp) : r; } -void rta_dump(rta *); +static inline void rta_uncache(rta *r) { r->cached = 0; r->uc = 0; } +void rta_dump(const rta *); void rta_dump_all(void); -void rta_show(struct cli *, rta *); +void rta_show(struct cli *, const rta *); -u32 rt_get_igp_metric(rte *rt); +u32 rt_get_igp_metric(rte *); struct hostentry * rt_get_hostentry(rtable *tab, ip_addr a, ip_addr ll, rtable *dep); -void rta_apply_hostentry(rta *a, struct hostentry *he, mpls_label_stack *mls); +void rta_apply_hostentry(rta *a, struct hostentry *he, mpls_label_stack *mls, linpool *lp); static inline void -rta_set_recursive_next_hop(rtable *dep, rta *a, rtable *tab, ip_addr gw, ip_addr ll, mpls_label_stack *mls) +rta_set_recursive_next_hop(rtable *dep, rta *a, rtable *tab, ip_addr gw, ip_addr ll, mpls_label_stack *mls, linpool *lp) { - rta_apply_hostentry(a, rt_get_hostentry(tab, gw, ll, dep), mls); + rta_apply_hostentry(a, rt_get_hostentry(tab, gw, ll, dep), mls, lp); } /* diff --git a/nest/rt-attr.c b/nest/rt-attr.c index c630aa95..357cd216 100644 --- a/nest/rt-attr.c +++ b/nest/rt-attr.c @@ -54,6 +54,7 @@ #include "lib/hash.h" #include "lib/idm.h" #include "lib/resource.h" +#include "lib/rcu.h" #include "lib/string.h" #include <stddef.h> @@ -61,7 +62,6 @@ const adata null_adata; /* adata of length 0 */ const char * const rta_src_names[RTS_MAX] = { - [RTS_DUMMY] = "", [RTS_STATIC] = "static", [RTS_INHERIT] = "inherit", [RTS_DEVICE] = "device", @@ -86,7 +86,14 @@ const char * rta_dest_names[RTD_MAX] = { [RTD_PROHIBIT] = "prohibited", }; +DEFINE_DOMAIN(attrs); +static DOMAIN(attrs) src_domain; + +#define SRC_LOCK LOCK_DOMAIN(attrs, src_domain) +#define SRC_UNLOCK UNLOCK_DOMAIN(attrs, src_domain) + pool *rta_pool; +pool *src_pool; static slab *rta_slab_[4]; static slab *nexthop_slab_[4]; @@ -97,72 +104,151 @@ static struct idm src_ids; /* rte source hash */ -#define RSH_KEY(n) n->proto, n->private_id +#define RSH_KEY(n) n->private_id #define RSH_NEXT(n) n->next -#define RSH_EQ(p1,n1,p2,n2) p1 == p2 && n1 == n2 -#define RSH_FN(p,n) p->hash_key ^ u32_hash(n) +#define RSH_EQ(n1,n2) n1 == n2 +#define RSH_FN(n) u32_hash(n) #define RSH_REHASH rte_src_rehash #define RSH_PARAMS /2, *2, 1, 1, 8, 20 -#define RSH_INIT_ORDER 6 - -static HASH(struct rte_src) src_hash; +#define RSH_INIT_ORDER 2 static void rte_src_init(void) { - rte_src_slab = sl_new(rta_pool, sizeof(struct rte_src)); - - idm_init(&src_ids, rta_pool, SRC_ID_INIT_SIZE); + src_domain = DOMAIN_NEW(attrs, "Route sources"); + src_pool = rp_new(&root_pool, &main_birdloop, "Route sources"); + rte_src_slab = sl_new(src_pool, sizeof(struct rte_src)); - HASH_INIT(src_hash, rta_pool, RSH_INIT_ORDER); + idm_init(&src_ids, src_pool, SRC_ID_INIT_SIZE); } - HASH_DEFINE_REHASH_FN(RSH, struct rte_src) -struct rte_src * -rt_find_source(struct proto *p, u32 id) +static struct rte_src * +rt_find_source(struct rte_owner *p, u32 id) { - return HASH_FIND(src_hash, RSH, p, id); + return HASH_FIND(p->hash, RSH, id); } struct rte_src * -rt_get_source(struct proto *p, u32 id) +rt_get_source_o(struct rte_owner *p, u32 id) { + if (p->stop) + bug("Stopping route owner asked for another source."); + struct rte_src *src = rt_find_source(p, id); if (src) + { + UNUSED u64 uc = atomic_fetch_add_explicit(&src->uc, 1, memory_order_acq_rel); return src; + } + SRC_LOCK; src = sl_allocz(rte_src_slab); - src->proto = p; + src->owner = p; src->private_id = id; src->global_id = idm_alloc(&src_ids); - src->uc = 0; - HASH_INSERT2(src_hash, RSH, rta_pool, src); + atomic_store_explicit(&src->uc, 1, memory_order_release); + p->uc++; + + HASH_INSERT2(p->hash, RSH, src_pool, src); + if (config->table_debug) + log(L_TRACE "Allocated new rte_src for %s, ID %uL %uG, have %u sources now", + p->name, src->private_id, src->global_id, p->uc); + + SRC_UNLOCK; return src; } +static inline void +rt_done_sources(struct rte_owner *o) +{ + if (o->stop->list) + ev_send(o->stop->list, o->stop); + else + ev_send(o->list, o->stop); +} + void -rt_prune_sources(void) +rt_prune_sources(void *data) { - HASH_WALK_FILTER(src_hash, next, src, sp) + struct rte_owner *o = data; + + HASH_WALK_FILTER(o->hash, next, src, sp) { - if (src->uc == 0) + u64 uc; + while ((uc = atomic_load_explicit(&src->uc, memory_order_acquire)) >> RTE_SRC_PU_SHIFT) + ; + + if (uc == 0) { - HASH_DO_REMOVE(src_hash, RSH, sp); + o->uc--; + + HASH_DO_REMOVE(o->hash, RSH, sp); + + SRC_LOCK; idm_free(&src_ids, src->global_id); sl_free(rte_src_slab, src); + SRC_UNLOCK; } } HASH_WALK_FILTER_END; - HASH_MAY_RESIZE_DOWN(src_hash, RSH, rta_pool); + SRC_LOCK; + HASH_MAY_RESIZE_DOWN(o->hash, RSH, src_pool); + + if (o->stop && !o->uc) + { + rfree(o->prune); + SRC_UNLOCK; + + if (config->table_debug) + log(L_TRACE "All rte_src's for %s pruned, scheduling stop event", o->name); + + rt_done_sources(o); + } + else + SRC_UNLOCK; +} + +void +rt_init_sources(struct rte_owner *o, const char *name, event_list *list) +{ + SRC_LOCK; + HASH_INIT(o->hash, src_pool, RSH_INIT_ORDER); + o->hash_key = random_u32(); + o->uc = 0; + o->name = name; + o->prune = ev_new_init(src_pool, rt_prune_sources, o); + o->stop = NULL; + o->list = list; + SRC_UNLOCK; } +void +rt_destroy_sources(struct rte_owner *o, event *done) +{ + o->stop = done; + + if (!o->uc) + { + if (config->table_debug) + log(L_TRACE "Source owner %s destroy requested. All rte_src's already pruned, scheduling stop event", o->name); + + SRC_LOCK; + rfree(o->prune); + SRC_UNLOCK; + + rt_done_sources(o); + } + else + if (config->table_debug) + log(L_TRACE "Source owner %s destroy requested. Remaining %u rte_src's to prune.", o->name, o->uc); +} /* * Multipath Next Hop @@ -541,8 +627,8 @@ ea_walk(struct ea_walk_state *s, uint id, uint max) * by calling ea_find() to find the attribute, extracting its value or returning * a provided default if no such attribute is present. */ -int -ea_get_int(ea_list *e, unsigned id, int def) +uintptr_t +ea_get_int(ea_list *e, unsigned id, uintptr_t def) { eattr *a = ea_find(e, id); if (!a) @@ -1081,21 +1167,28 @@ ea_append(ea_list *to, ea_list *what) * rta's */ -static uint rta_cache_count; -static uint rta_cache_size = 32; -static uint rta_cache_limit; -static uint rta_cache_mask; -static rta **rta_hash_table; +static DOMAIN(attrs) attrs_domain; -static void -rta_alloc_hash(void) +#define RTA_LOCK LOCK_DOMAIN(attrs, attrs_domain) +#define RTA_UNLOCK UNLOCK_DOMAIN(attrs, attrs_domain) + +struct rta_cache { + u32 count; + u32 size; + u32 limit; + u32 mask; + rta * _Atomic table[0]; +} * _Atomic rta_cache; +// rta_aux, rta_cache = { .size = ATOMIC_VAR_INIT(32), }; + +static struct rta_cache * +rta_alloc_hash(u32 size) { - rta_hash_table = mb_allocz(rta_pool, sizeof(rta *) * rta_cache_size); - if (rta_cache_size < 32768) - rta_cache_limit = rta_cache_size * 2; - else - rta_cache_limit = ~0; - rta_cache_mask = rta_cache_size - 1; + struct rta_cache *c = mb_allocz(rta_pool, sizeof(struct rta_cache) + sizeof(rta * _Atomic) * size); + c->size = size; + c->limit = (size >> 20) ? (~0U) : (size * 2); + c->mask = size - 1; + return c; } static inline uint @@ -1104,13 +1197,14 @@ rta_hash(rta *a) u64 h; mem_hash_init(&h); #define MIX(f) mem_hash_mix(&h, &(a->f), sizeof(a->f)); - MIX(src); +#define BMIX(f) mem_hash_mix_num(&h, a->f); MIX(hostentry); MIX(from); MIX(igp_metric); - MIX(source); - MIX(scope); - MIX(dest); + BMIX(source); + BMIX(scope); + BMIX(dest); + MIX(pref); #undef MIX return mem_hash_value(&h) ^ nexthop_hash(&(a->nh)) ^ ea_hash(a->eattrs); @@ -1119,8 +1213,7 @@ rta_hash(rta *a) static inline int rta_same(rta *x, rta *y) { - return (x->src == y->src && - x->source == y->source && + return (x->source == y->source && x->scope == y->scope && x->dest == y->dest && x->igp_metric == y->igp_metric && @@ -1149,34 +1242,88 @@ rta_copy(rta *o) } static inline void -rta_insert(rta *r) +rta_insert(rta *r, struct rta_cache *c) { - uint h = r->hash_key & rta_cache_mask; - r->next = rta_hash_table[h]; - if (r->next) - r->next->pprev = &r->next; - r->pprev = &rta_hash_table[h]; - rta_hash_table[h] = r; + uint h = r->hash_key & c->mask; + rta *next = atomic_load_explicit(&c->table[h], memory_order_relaxed); + + atomic_store_explicit(&r->next, next, memory_order_relaxed); + r->pprev = &c->table[h]; + + if (next) + next->pprev = &r->next; + + /* This store MUST be the last and MUST have release order for thread-safety */ + atomic_store_explicit(&c->table[h], r, memory_order_release); } static void -rta_rehash(void) +rta_rehash(struct rta_cache *c) { - uint ohs = rta_cache_size; - uint h; - rta *r, *n; - rta **oht = rta_hash_table; - - rta_cache_size = 2*rta_cache_size; - DBG("Rehashing rta cache from %d to %d entries.\n", ohs, rta_cache_size); - rta_alloc_hash(); - for(h=0; h<ohs; h++) - for(r=oht[h]; r; r=n) + u32 os = c->size; + + struct rta_cache *nc = rta_alloc_hash(os * 2); + nc->count = c->count; + + /* First we simply copy every chain to both new locations */ + for (u32 h = 0; h < os; h++) + { + rta *r = atomic_load_explicit(&c->table[h], memory_order_relaxed); + atomic_store_explicit(&nc->table[h], r, memory_order_relaxed); + atomic_store_explicit(&nc->table[h + os], r, memory_order_relaxed); + } + + /* Then we exchange the hashes; release semantics forces the previous code to be already done */ + atomic_store_explicit(&rta_cache, nc, memory_order_release); + + /* And now we pass through both chains and filter them */ + for (u32 h = 0; h < c->size; h++) + { + rta * _Atomic * ap = &nc->table[h]; + rta * _Atomic * bp = &nc->table[h + os]; + + rta *r = atomic_load_explicit(ap, memory_order_relaxed); + ASSERT_DIE(r == atomic_load_explicit(bp, memory_order_relaxed)); + + while (r) + { + if (r->hash_key & os) { - n = r->next; - rta_insert(r); + r->pprev = bp; + atomic_store_explicit(bp, r, memory_order_release); + bp = &r->next; } - mb_free(oht); + else + { + r->pprev = ap; + atomic_store_explicit(ap, r, memory_order_release); + ap = &r->next; + } + + r = atomic_load_explicit(&r->next, memory_order_acquire); + } + + atomic_store_explicit(ap, NULL, memory_order_release); + atomic_store_explicit(bp, NULL, memory_order_release); + } + + synchronize_rcu(); + mb_free(c); +} + +static rta * +rta_find(rta *o, u32 h, struct rta_cache *c) +{ + rta *r = NULL; + + for (r = atomic_load_explicit(&c->table[h & c->mask], memory_order_acquire); r; r = atomic_load_explicit(&r->next, memory_order_acquire)) + if (r->hash_key == h && rta_same(r, o)) + { + atomic_fetch_add_explicit(&r->uc, 1, memory_order_acq_rel); + return r; + } + + return NULL; } /** @@ -1198,43 +1345,89 @@ rta_lookup(rta *o) rta *r; uint h; - ASSERT(!(o->aflags & RTAF_CACHED)); + ASSERT(!o->cached); if (o->eattrs) ea_normalize(o->eattrs); h = rta_hash(o); - for(r=rta_hash_table[h & rta_cache_mask]; r; r=r->next) - if (r->hash_key == h && rta_same(r, o)) - return rta_clone(r); + /* Lockless lookup */ + rcu_read_lock(); + r = rta_find(o, h, atomic_load_explicit(&rta_cache, memory_order_acquire)); + rcu_read_unlock(); + + if (r) + return r; + + RTA_LOCK; + + /* Locked lookup to avoid duplicates if possible */ + struct rta_cache *c = atomic_load_explicit(&rta_cache, memory_order_acquire); + r = rta_find(o, h, c); + if (r) + { + RTA_UNLOCK; + return r; + } + + /* Store the rta */ r = rta_copy(o); r->hash_key = h; - r->aflags = RTAF_CACHED; - rt_lock_source(r->src); + r->cached = 1; rt_lock_hostentry(r->hostentry); - rta_insert(r); + rta_insert(r, c); - if (++rta_cache_count > rta_cache_limit) - rta_rehash(); + if (++c->count > c->limit) + rta_rehash(c); + RTA_UNLOCK; return r; } void rta__free(rta *a) { - ASSERT(rta_cache_count && (a->aflags & RTAF_CACHED)); - rta_cache_count--; - *a->pprev = a->next; - if (a->next) - a->next->pprev = a->pprev; + ASSERT(a->cached); + + RTA_LOCK; + struct rta_cache *c = atomic_load_explicit(&rta_cache, memory_order_acquire); + + if (atomic_load_explicit(&a->uc, memory_order_acquire)) + { + /* Acquired inbetween */ + RTA_UNLOCK; + return; + } + + /* Relink the forward pointer */ + rta *next = atomic_load_explicit(&a->next, memory_order_acquire); + atomic_store_explicit(a->pprev, next, memory_order_release); + + /* Relink the backwards pointer */ + if (next) + next->pprev = a->pprev; + + /* Wait until nobody knows about us */ + synchronize_rcu(); + + if (atomic_load_explicit(&a->uc, memory_order_acquire)) + { + /* Acquired inbetween, relink back */ + rta_insert(a, c); + RTA_UNLOCK; + return; + } + + /* Cleared to free the memory */ rt_unlock_hostentry(a->hostentry); - rt_unlock_source(a->src); if (a->nh.next) nexthop_free(a->nh.next); ea_free(a->eattrs); - a->aflags = 0; /* Poison the entry */ + a->cached = 0; + c->count--; sl_free(rta_slab(a), a); + + RTA_UNLOCK; } rta * @@ -1248,8 +1441,7 @@ rta_do_cow(rta *o, linpool *lp) memcpy(*nhn, nho, nexthop_size(nho)); nhn = &((*nhn)->next); } - r->aflags = 0; - r->uc = 0; + rta_uncache(r); return r; } @@ -1260,22 +1452,22 @@ rta_do_cow(rta *o, linpool *lp) * This function takes a &rta and dumps its contents to the debug output. */ void -rta_dump(rta *a) +rta_dump(const rta *a) { - static char *rts[] = { "RTS_DUMMY", "RTS_STATIC", "RTS_INHERIT", "RTS_DEVICE", + static char *rts[] = { "", "RTS_STATIC", "RTS_INHERIT", "RTS_DEVICE", "RTS_STAT_DEV", "RTS_REDIR", "RTS_RIP", "RTS_OSPF", "RTS_OSPF_IA", "RTS_OSPF_EXT1", "RTS_OSPF_EXT2", "RTS_BGP", "RTS_PIPE", "RTS_BABEL" }; static char *rtd[] = { "", " DEV", " HOLE", " UNREACH", " PROHIBIT" }; - debug("p=%s uc=%d %s %s%s h=%04x", - a->src->proto->name, a->uc, rts[a->source], ip_scope_text(a->scope), + debug("pref=%d uc=%d %s %s%s h=%04x", + a->pref, a->uc, rts[a->source], ip_scope_text(a->scope), rtd[a->dest], a->hash_key); - if (!(a->aflags & RTAF_CACHED)) + if (!a->cached) debug(" !CACHED"); debug(" <-%I", a->from); if (a->dest == RTD_UNICAST) - for (struct nexthop *nh = &(a->nh); nh; nh = nh->next) + for (const struct nexthop *nh = &(a->nh); nh; nh = nh->next) { if (ipa_nonzero(nh->gw)) debug(" ->%I", nh->gw); if (nh->labels) debug(" L %d", nh->label[0]); @@ -1302,19 +1494,27 @@ rta_dump_all(void) rta *a; uint h; - debug("Route attribute cache (%d entries, rehash at %d):\n", rta_cache_count, rta_cache_limit); - for(h=0; h<rta_cache_size; h++) - for(a=rta_hash_table[h]; a; a=a->next) + RTA_LOCK; + + struct rta_cache *c = atomic_load_explicit(&rta_cache, memory_order_acquire); + + debug("Route attribute cache (%d entries, rehash at %d):\n", c->count, c->limit); + for(h=0; h<c->size; h++) + for(a = atomic_load_explicit(&c->table[h], memory_order_acquire); + a; + a = atomic_load_explicit(&a->next, memory_order_acquire)) { debug("%p ", a); rta_dump(a); debug("\n"); } debug("\n"); + + RTA_UNLOCK; } void -rta_show(struct cli *c, rta *a) +rta_show(struct cli *c, const rta *a) { cli_printf(c, -1008, "\tType: %s %s", rta_src_names[a->source], ip_scope_text(a->scope)); @@ -1332,7 +1532,9 @@ rta_show(struct cli *c, rta *a) void rta_init(void) { - rta_pool = rp_new(&root_pool, "Attributes"); + attrs_domain = DOMAIN_NEW(attrs, "Attributes"); + + rta_pool = rp_new(&root_pool, &main_birdloop, "Attributes"); rta_slab_[0] = sl_new(rta_pool, sizeof(rta)); rta_slab_[1] = sl_new(rta_pool, sizeof(rta) + sizeof(u32)); @@ -1344,7 +1546,7 @@ rta_init(void) nexthop_slab_[2] = sl_new(rta_pool, sizeof(struct nexthop) + sizeof(u32)*2); nexthop_slab_[3] = sl_new(rta_pool, sizeof(struct nexthop) + sizeof(u32)*MPLS_MAX_LABEL_STACK); - rta_alloc_hash(); + atomic_store_explicit(&rta_cache, rta_alloc_hash(32), memory_order_relaxed); rte_src_init(); } diff --git a/nest/rt-dev.c b/nest/rt-dev.c index 61f025ce..c1251675 100644 --- a/nest/rt-dev.c +++ b/nest/rt-dev.c @@ -67,13 +67,11 @@ dev_ifa_notify(struct proto *P, uint flags, struct ifa *ad) /* Use iface ID as local source ID */ struct rte_src *src = rt_get_source(P, ad->iface->index); - rte_update2(c, net, NULL, src); + rte_update(c, net, NULL, src); + rt_unlock_source(src); } else if (flags & IF_CHANGE_UP) { - rta *a; - rte *e; - DBG("dev_if_notify: %s:%I going up\n", ad->iface->name, ad->ip); if (cf->check_link && !(ad->iface->flags & IF_LINK_UP)) @@ -83,17 +81,20 @@ dev_ifa_notify(struct proto *P, uint flags, struct ifa *ad) struct rte_src *src = rt_get_source(P, ad->iface->index); rta a0 = { - .src = src, + .pref = c->preference, .source = RTS_DEVICE, .scope = SCOPE_UNIVERSE, .dest = RTD_UNICAST, .nh.iface = ad->iface, }; - a = rta_lookup(&a0); - e = rte_get_temp(a); - e->pflags = 0; - rte_update2(c, net, e, src); + rte e0 = { + .attrs = rta_lookup(&a0), + .src = src, + }; + + rte_update(c, net, &e0, src); + rt_unlock_source(src); } } diff --git a/nest/rt-show.c b/nest/rt-show.c index 7691878d..65b59af4 100644 --- a/nest/rt-show.c +++ b/nest/rt-show.c @@ -56,17 +56,17 @@ rt_show_rte(struct cli *c, byte *ia, rte *e, struct rt_show_data *d, int primary if (d->verbose && !rta_is_cached(a) && a->eattrs) ea_normalize(a->eattrs); - get_route_info = a->src->proto->proto->get_route_info; + get_route_info = e->src->owner->class ? e->src->owner->class->get_route_info : NULL; if (get_route_info) get_route_info(e, info); else - bsprintf(info, " (%d)", e->pref); + bsprintf(info, " (%d)", a->pref); if (d->last_table != d->tab) rt_show_table(c, d); cli_printf(c, -1007, "%-20s %s [%s %s%s]%s%s", ia, rta_dest_name(a->dest), - a->src->proto->name, tm, from, primary ? (sync_error ? " !" : " *") : "", info); + e->src->owner->name, tm, from, primary ? (sync_error ? " !" : " *") : "", info); if (a->dest == RTD_UNICAST) for (nh = &(a->nh); nh; nh = nh->next) @@ -95,13 +95,38 @@ rt_show_rte(struct cli *c, byte *ia, rte *e, struct rt_show_data *d, int primary } if (d->verbose) + { + cli_printf(c, -1008, "\tInternal route ID: %uL %uG %uS", e->src->private_id, e->src->global_id, e->stale_cycle); rta_show(c, a); + } +} + +static uint +rte_feed_count(net *n) +{ + uint count = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTES_OR_NULL(e))) + count++; + return count; +} + +static void +rte_feed_obtain(net *n, rte **feed, uint count) +{ + uint i = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTES_OR_NULL(e))) + { + ASSERT_DIE(i < count); + feed[i++] = &e->rte; + } + ASSERT_DIE(i == count); } static void rt_show_net(struct cli *c, net *n, struct rt_show_data *d) { - rte *e, *ee; byte ia[NET_MAX_TEXT_LENGTH+1]; struct channel *ec = d->tab->export_channel; @@ -114,9 +139,9 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) bsnprintf(ia, sizeof(ia), "%N", n->n.addr); - for (e = n->routes; e; e = e->next) + for (struct rte_storage *er = n->routes; er; er = er->next) { - if (rte_is_filtered(e) != d->filtered) + if (rte_is_filtered(&er->rte) != d->filtered) continue; d->rt_counter++; @@ -126,16 +151,15 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) if (pass) continue; - ee = e; - rte_make_tmp_attrs(&e, c->show_pool, NULL); + struct rte e = er->rte; /* Export channel is down, do not try to export routes to it */ - if (ec && (ec->export_state == ES_DOWN)) + if (ec && !ec->out_req.hook) goto skip; if (d->export_mode == RSEM_EXPORTED) { - if (!bmap_test(&ec->export_map, ee->id)) + if (!bmap_test(&ec->export_map, e.id)) goto skip; // if (ec->ra_mode != RA_ANY) @@ -144,17 +168,24 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) else if ((d->export_mode == RSEM_EXPORT) && (ec->ra_mode == RA_MERGED)) { /* Special case for merged export */ - rte *rt_free; - e = rt_export_merged(ec, n, &rt_free, c->show_pool, 1); pass = 1; + uint count = rte_feed_count(n); + if (!count) + goto skip; - if (!e) - { e = ee; goto skip; } + rte **feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(n, feed, count); + rte *em = rt_export_merged(ec, feed, count, c->show_pool, 1); + + if (em) + e = *em; + else + goto skip; } else if (d->export_mode) { struct proto *ep = ec->proto; - int ic = ep->preexport ? ep->preexport(ep, &e, c->show_pool) : 0; + int ic = ep->preexport ? ep->preexport(ec, &e) : 0; if (ec->ra_mode == RA_OPTIMAL || ec->ra_mode == RA_MERGED) pass = 1; @@ -180,24 +211,19 @@ rt_show_net(struct cli *c, net *n, struct rt_show_data *d) } } - if (d->show_protocol && (d->show_protocol != e->attrs->src->proto)) + if (d->show_protocol && (&d->show_protocol->sources != e.src->owner)) goto skip; if (f_run(d->filter, &e, c->show_pool, 0) > F_ACCEPT) goto skip; if (d->stats < 2) - rt_show_rte(c, ia, e, d, (e->net->routes == ee)); + rt_show_rte(c, ia, &e, d, (n->routes == er)); d->show_counter++; ia[0] = 0; skip: - if (e != ee) - { - rte_free(e); - e = ee; - } lp_flush(c->show_pool); if (d->primary_only) @@ -213,11 +239,13 @@ rt_show_cleanup(struct cli *c) /* Unlink the iterator */ if (d->table_open) - fit_get(&d->tab->table->fib, &d->fit); + RT_LOCKED(d->tab->table, t) + fit_get(&t->fib, &d->fit); /* Unlock referenced tables */ WALK_LIST(tab, d->tables) - rt_unlock_table(tab->table); + RT_LOCKED(tab->table, t) + rt_unlock_table(t); } static void @@ -229,8 +257,6 @@ rt_show_cont(struct cli *c) #else unsigned max = 64; #endif - struct fib *fib = &d->tab->table->fib; - struct fib_iterator *it = &d->fit; if (d->running_on_config && (d->running_on_config != config)) { @@ -238,9 +264,14 @@ rt_show_cont(struct cli *c) goto done; } + rtable_private *t = RT_LOCK(d->tab->table); + + struct fib *fib = &t->fib; + struct fib_iterator *it = &d->fit; + if (!d->table_open) { - FIB_ITERATE_INIT(&d->fit, &d->tab->table->fib); + FIB_ITERATE_INIT(&d->fit, fib); d->table_open = 1; d->table_counter++; d->kernel = rt_show_get_kernel(d); @@ -258,6 +289,7 @@ rt_show_cont(struct cli *c) if (!max--) { FIB_ITERATE_PUT(it); + RT_UNLOCK(d->tab->table); return; } rt_show_net(c, n, d); @@ -274,6 +306,8 @@ rt_show_cont(struct cli *c) d->net_counter - d->net_counter_last, d->tab->table->name); } + RT_UNLOCK(d->tab->table); + d->kernel = NULL; d->table_open = 0; d->tab = NODE_NEXT(d->tab); @@ -322,7 +356,7 @@ rt_show_get_default_tables(struct rt_show_data *d) { WALK_LIST(c, d->export_protocol->channels) { - if (c->export_state == ES_DOWN) + if (!c->out_req.hook) continue; tab = rt_show_add_table(d, c->table); @@ -339,7 +373,7 @@ rt_show_get_default_tables(struct rt_show_data *d) } for (int i=1; i<NET_MAX; i++) - if (config->def_tables[i]) + if (config->def_tables[i] && config->def_tables[i]->table) rt_show_add_table(d, config->def_tables[i]->table); } @@ -405,7 +439,8 @@ rt_show(struct rt_show_data *d) if (!d->addr) { WALK_LIST(tab, d->tables) - rt_lock_table(tab->table); + RT_LOCKED(tab->table, t) + rt_lock_table(t); /* There is at least one table */ d->tab = HEAD(d->tables); @@ -420,13 +455,17 @@ rt_show(struct rt_show_data *d) d->tab = tab; d->kernel = rt_show_get_kernel(d); + RT_LOCK(tab->table); + if (d->show_for) - n = net_route(tab->table, d->addr); + n = net_route(RT_PRIV(tab->table), d->addr); else - n = net_find(tab->table, d->addr); + n = net_find(RT_PRIV(tab->table), d->addr); if (n) rt_show_net(this_cli, n, d); + + RT_UNLOCK(tab->table); } if (d->rt_counter) diff --git a/nest/rt-table.c b/nest/rt-table.c index 390b3277..ada54396 100644 --- a/nest/rt-table.c +++ b/nest/rt-table.c @@ -45,28 +45,84 @@ #include "lib/string.h" #include "lib/alloca.h" -#ifdef CONFIG_BGP -#include "proto/bgp/bgp.h" -#endif +#include <stdatomic.h> pool *rt_table_pool; -static slab *rte_slab; -static linpool *rte_update_pool; - list routing_tables; -static void rt_free_hostcache(rtable *tab); -static void rt_notify_hostcache(rtable *tab, net *net); -static void rt_update_hostcache(rtable *tab); -static void rt_next_hop_update(rtable *tab); -static inline void rt_prune_table(rtable *tab); -static inline void rt_schedule_notify(rtable *tab); +/* Data structures for export journal */ +#define RT_PENDING_EXPORT_ITEMS (page_size - sizeof(struct rt_export_block)) / sizeof(struct rt_pending_export) + +struct rt_export_block { + node n; + _Atomic u32 end; + _Atomic _Bool not_last; + struct rt_pending_export export[]; +}; + +static void rt_free_hostcache(rtable_private *tab); +static void rt_notify_hostcache(rtable_private *tab, net *net); +static void rt_update_hostcache(void *tab); +static void rt_next_hop_update(void *tab); +static inline void rt_prune_table(void *tab); +static inline void rt_schedule_notify(rtable_private *tab); +static void rt_feed_channel(void *); + +static inline void rt_export_used(rtable_private *tab); +static void rt_export_cleanup(void *tab); + +const char *rt_import_state_name_array[TIS_MAX] = { + [TIS_DOWN] = "DOWN", + [TIS_UP] = "UP", + [TIS_STOP] = "STOP", + [TIS_FLUSHING] = "FLUSHING", + [TIS_WAITING] = "WAITING", + [TIS_CLEARED] = "CLEARED", +}; + +const char *rt_export_state_name_array[TES_MAX] = { + [TES_DOWN] = "DOWN", + [TES_HUNGRY] = "HUNGRY", + [TES_FEEDING] = "FEEDING", + [TES_READY] = "READY", + [TES_STOP] = "STOP" +}; + +const char *rt_import_state_name(u8 state) +{ + if (state >= TIS_MAX) + return "!! INVALID !!"; + else + return rt_import_state_name_array[state]; +} + +const char *rt_export_state_name(u8 state) +{ + if (state >= TES_MAX) + return "!! INVALID !!"; + else + return rt_export_state_name_array[state]; +} + +struct event_cork rt_cork; + +static inline void +rte_update_lock(struct channel *c) +{ + c->rte_update_nest_cnt++; +} +static inline void +rte_update_unlock(struct channel *c) +{ + if (!--c->rte_update_nest_cnt) + lp_flush(c->rte_update_pool); +} /* Like fib_route(), but skips empty net entries */ static inline void * -net_route_ip4(rtable *t, net_addr_ip4 *n) +net_route_ip4(rtable_private *t, net_addr_ip4 *n) { net *r; @@ -80,7 +136,7 @@ net_route_ip4(rtable *t, net_addr_ip4 *n) } static inline void * -net_route_ip6(rtable *t, net_addr_ip6 *n) +net_route_ip6(rtable_private *t, net_addr_ip6 *n) { net *r; @@ -94,7 +150,7 @@ net_route_ip6(rtable *t, net_addr_ip6 *n) } static inline void * -net_route_ip6_sadr(rtable *t, net_addr_ip6_sadr *n) +net_route_ip6_sadr(rtable_private *t, net_addr_ip6_sadr *n) { struct fib_node *fn; @@ -133,7 +189,7 @@ net_route_ip6_sadr(rtable *t, net_addr_ip6_sadr *n) } void * -net_route(rtable *tab, const net_addr *n) +net_route(rtable_private *tab, const net_addr *n) { ASSERT(tab->addr_type == n->type); @@ -162,12 +218,15 @@ net_route(rtable *tab, const net_addr *n) static int -net_roa_check_ip4(rtable *tab, const net_addr_ip4 *px, u32 asn) +net_roa_check_ip4(rtable *t, const net_addr_ip4 *px, u32 asn) { struct net_addr_roa4 n = NET_ADDR_ROA4(px->prefix, px->pxlen, 0, 0); struct fib_node *fn; int anything = 0; + RT_LOCK(t); + rtable_private *tab = RT_PRIV(t); + while (1) { for (fn = fib_get_chain(&tab->fib, (net_addr *) &n); fn; fn = fn->next) @@ -175,11 +234,14 @@ net_roa_check_ip4(rtable *tab, const net_addr_ip4 *px, u32 asn) net_addr_roa4 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); - if (net_equal_prefix_roa4(roa, &n) && rte_is_valid(r->routes)) + if (net_equal_prefix_roa4(roa, &n) && r->routes && rte_is_valid(&r->routes->rte)) { anything = 1; if (asn && (roa->asn == asn) && (roa->max_pxlen >= px->pxlen)) + { + RT_UNLOCK(tab); return ROA_VALID; + } } } @@ -190,16 +252,20 @@ net_roa_check_ip4(rtable *tab, const net_addr_ip4 *px, u32 asn) ip4_clrbit(&n.prefix, n.pxlen); } + RT_UNLOCK(tab); return anything ? ROA_INVALID : ROA_UNKNOWN; } static int -net_roa_check_ip6(rtable *tab, const net_addr_ip6 *px, u32 asn) +net_roa_check_ip6(rtable *t, const net_addr_ip6 *px, u32 asn) { struct net_addr_roa6 n = NET_ADDR_ROA6(px->prefix, px->pxlen, 0, 0); struct fib_node *fn; int anything = 0; + RT_LOCK(t); + rtable_private *tab = RT_PRIV(t); + while (1) { for (fn = fib_get_chain(&tab->fib, (net_addr *) &n); fn; fn = fn->next) @@ -207,11 +273,14 @@ net_roa_check_ip6(rtable *tab, const net_addr_ip6 *px, u32 asn) net_addr_roa6 *roa = (void *) fn->addr; net *r = fib_node_to_user(&tab->fib, fn); - if (net_equal_prefix_roa6(roa, &n) && rte_is_valid(r->routes)) + if (net_equal_prefix_roa6(roa, &n) && r->routes && rte_is_valid(&r->routes->rte)) { anything = 1; if (asn && (roa->asn == asn) && (roa->max_pxlen >= px->pxlen)) + { + RT_UNLOCK(tab); return ROA_VALID; + } } } @@ -222,6 +291,7 @@ net_roa_check_ip6(rtable *tab, const net_addr_ip6 *px, u32 asn) ip6_clrbit(&n.prefix, n.pxlen); } + RT_UNLOCK(tab); return anything ? ROA_INVALID : ROA_UNKNOWN; } @@ -256,253 +326,55 @@ net_roa_check(rtable *tab, const net_addr *n, u32 asn) * @net: network node * @src: route source * - * The rte_find() function returns a route for destination @net - * which is from route source @src. + * The rte_find() function returns a pointer to a route for destination @net + * which is from route source @src. List end pointer is returned if no route is found. */ -rte * +static struct rte_storage ** rte_find(net *net, struct rte_src *src) { - rte *e = net->routes; - - while (e && e->attrs->src != src) - e = e->next; - return e; -} + struct rte_storage **e = &net->routes; -/** - * rte_get_temp - get a temporary &rte - * @a: attributes to assign to the new route (a &rta; in case it's - * un-cached, rte_update() will create a cached copy automatically) - * - * Create a temporary &rte and bind it with the attributes @a. - * Also set route preference to the default preference set for - * the protocol. - */ -rte * -rte_get_temp(rta *a) -{ - rte *e = sl_alloc(rte_slab); + while ((*e) && (*e)->rte.src != src) + e = &(*e)->next; - e->attrs = a; - e->id = 0; - e->flags = 0; - e->pref = 0; return e; } -rte * -rte_do_cow(rte *r) -{ - rte *e = sl_alloc(rte_slab); - - memcpy(e, r, sizeof(rte)); - e->attrs = rta_clone(r->attrs); - e->flags = 0; - return e; -} -/** - * rte_cow_rta - get a private writable copy of &rte with writable &rta - * @r: a route entry to be copied - * @lp: a linpool from which to allocate &rta - * - * rte_cow_rta() takes a &rte and prepares it and associated &rta for - * modification. There are three possibilities: First, both &rte and &rta are - * private copies, in that case they are returned unchanged. Second, &rte is - * private copy, but &rta is cached, in that case &rta is duplicated using - * rta_do_cow(). Third, both &rte is shared and &rta is cached, in that case - * both structures are duplicated by rte_do_cow() and rta_do_cow(). - * - * Note that in the second case, cached &rta loses one reference, while private - * copy created by rta_do_cow() is a shallow copy sharing indirect data (eattrs, - * nexthops, ...) with it. To work properly, original shared &rta should have - * another reference during the life of created private copy. - * - * Result: a pointer to the new writable &rte with writable &rta. - */ -rte * -rte_cow_rta(rte *r, linpool *lp) +struct rte_storage * +rte_store(const rte *r, net *net, rtable_private *tab) { - if (!rta_is_cached(r->attrs)) - return r; - - r = rte_cow(r); - rta *a = rta_do_cow(r->attrs, lp); - rta_free(r->attrs); - r->attrs = a; - return r; -} + struct rte_storage *e = sl_alloc(tab->rte_slab); + e->rte = *r; + e->rte.net = net->n.addr; -/** - * rte_init_tmp_attrs - initialize temporary ea_list for route - * @r: route entry to be modified - * @lp: linpool from which to allocate attributes - * @max: maximum number of added temporary attribus - * - * This function is supposed to be called from make_tmp_attrs() and - * store_tmp_attrs() hooks before rte_make_tmp_attr() / rte_store_tmp_attr() - * functions. It allocates &ea_list with length for @max items for temporary - * attributes and puts it on top of eattrs stack. - */ -void -rte_init_tmp_attrs(rte *r, linpool *lp, uint max) -{ - struct ea_list *e = lp_alloc(lp, sizeof(struct ea_list) + max * sizeof(eattr)); - - e->next = r->attrs->eattrs; - e->flags = EALF_SORTED | EALF_TEMP; - e->count = 0; + rt_lock_source(e->rte.src); - r->attrs->eattrs = e; -} + if (e->rte.attrs->cached) + e->rte.attrs = rta_clone(e->rte.attrs); + else + e->rte.attrs = rta_lookup(e->rte.attrs); -/** - * rte_make_tmp_attr - make temporary eattr from private route fields - * @r: route entry to be modified - * @id: attribute ID - * @type: attribute type - * @val: attribute value (u32 or adata ptr) - * - * This function is supposed to be called from make_tmp_attrs() hook for - * each temporary attribute, after temporary &ea_list was initialized by - * rte_init_tmp_attrs(). It checks whether temporary attribute is supposed to - * be defined (based on route pflags) and if so then it fills &eattr field in - * preallocated temporary &ea_list on top of route @r eattrs stack. - * - * Note that it may require free &eattr in temporary &ea_list, so it must not be - * called more times than @max argument of rte_init_tmp_attrs(). - */ -void -rte_make_tmp_attr(rte *r, uint id, uint type, uintptr_t val) -{ - if (r->pflags & EA_ID_FLAG(id)) - { - ea_list *e = r->attrs->eattrs; - eattr *a = &e->attrs[e->count++]; - a->id = id; - a->type = type; - a->flags = 0; - - if (type & EAF_EMBEDDED) - a->u.data = (u32) val; - else - a->u.ptr = (struct adata *) val; - } + return e; } /** - * rte_store_tmp_attr - store temporary eattr to private route fields - * @r: route entry to be modified - * @id: attribute ID - * - * This function is supposed to be called from store_tmp_attrs() hook for - * each temporary attribute, after temporary &ea_list was initialized by - * rte_init_tmp_attrs(). It checks whether temporary attribute is defined in - * route @r eattrs stack, updates route pflags accordingly, undefines it by - * filling &eattr field in preallocated temporary &ea_list on top of the eattrs - * stack, and returns the value. Caller is supposed to store it in the - * appropriate private field. + * rte_free - delete a &rte + * @e: &struct rte_storage to be deleted + * @tab: the table which the rte belongs to * - * Note that it may require free &eattr in temporary &ea_list, so it must not be - * called more times than @max argument of rte_init_tmp_attrs() + * rte_free() deletes the given &rte from the routing table it's linked to. */ -uintptr_t -rte_store_tmp_attr(rte *r, uint id) -{ - ea_list *e = r->attrs->eattrs; - eattr *a = ea_find(e->next, id); - - if (a) - { - e->attrs[e->count++] = (struct eattr) { .id = id, .type = EAF_TYPE_UNDEF }; - r->pflags |= EA_ID_FLAG(id); - return (a->type & EAF_EMBEDDED) ? a->u.data : (uintptr_t) a->u.ptr; - } - else - { - r->pflags &= ~EA_ID_FLAG(id); - return 0; - } -} -/** - * rte_make_tmp_attrs - prepare route by adding all relevant temporary route attributes - * @r: route entry to be modified (may be replaced if COW) - * @lp: linpool from which to allocate attributes - * @old_attrs: temporary ref to old &rta (may be NULL) - * - * This function expands privately stored protocol-dependent route attributes - * to a uniform &eattr / &ea_list representation. It is essentially a wrapper - * around protocol make_tmp_attrs() hook, which does some additional work like - * ensuring that route @r is writable. - * - * The route @r may be read-only (with %REF_COW flag), in that case rw copy is - * obtained by rte_cow() and @r is replaced. If @rte is originally rw, it may be - * directly modified (and it is never copied). - * - * If the @old_attrs ptr is supplied, the function obtains another reference of - * old cached &rta, that is necessary in some cases (see rte_cow_rta() for - * details). It is freed by rte_store_tmp_attrs(), or manually by rta_free(). - * - * Generally, if caller ensures that @r is read-only (e.g. in route export) then - * it may ignore @old_attrs (and set it to NULL), but must handle replacement of - * @r. If caller ensures that @r is writable (e.g. in route import) then it may - * ignore replacement of @r, but it must handle @old_attrs. - */ void -rte_make_tmp_attrs(rte **r, linpool *lp, rta **old_attrs) +rte_free(struct rte_storage *e, rtable_private *tab) { - void (*make_tmp_attrs)(rte *r, linpool *lp); - make_tmp_attrs = (*r)->attrs->src->proto->make_tmp_attrs; - - if (!make_tmp_attrs) - return; - - /* We may need to keep ref to old attributes, will be freed in rte_store_tmp_attrs() */ - if (old_attrs) - *old_attrs = rta_is_cached((*r)->attrs) ? rta_clone((*r)->attrs) : NULL; - - *r = rte_cow_rta(*r, lp); - make_tmp_attrs(*r, lp); + rt_unlock_source(e->rte.src); + rta_free(e->rte.attrs); + sl_free(tab->rte_slab, e); } -/** - * rte_store_tmp_attrs - store temporary route attributes back to private route fields - * @r: route entry to be modified - * @lp: linpool from which to allocate attributes - * @old_attrs: temporary ref to old &rta - * - * This function stores temporary route attributes that were expanded by - * rte_make_tmp_attrs() back to private route fields and also undefines them. - * It is essentially a wrapper around protocol store_tmp_attrs() hook, which - * does some additional work like shortcut if there is no change and cleanup - * of @old_attrs reference obtained by rte_make_tmp_attrs(). - */ -static void -rte_store_tmp_attrs(rte *r, linpool *lp, rta *old_attrs) -{ - void (*store_tmp_attrs)(rte *rt, linpool *lp); - store_tmp_attrs = r->attrs->src->proto->store_tmp_attrs; - - if (!store_tmp_attrs) - return; - - ASSERT(!rta_is_cached(r->attrs)); - - /* If there is no new ea_list, we just skip the temporary ea_list */ - ea_list *ea = r->attrs->eattrs; - if (ea && (ea->flags & EALF_TEMP)) - r->attrs->eattrs = ea->next; - else - store_tmp_attrs(r, lp); - - /* Free ref we got in rte_make_tmp_attrs(), have to do rta_lookup() first */ - r->attrs = rta_lookup(r->attrs); - rta_free(old_attrs); -} - - static int /* Actually better or at least as good as */ rte_better(rte *new, rte *old) { @@ -513,20 +385,20 @@ rte_better(rte *new, rte *old) if (!rte_is_valid(new)) return 0; - if (new->pref > old->pref) + if (new->attrs->pref > old->attrs->pref) return 1; - if (new->pref < old->pref) + if (new->attrs->pref < old->attrs->pref) return 0; - if (new->attrs->src->proto->proto != old->attrs->src->proto->proto) + if (new->src->owner->class != old->src->owner->class) { /* * If the user has configured protocol preferences, so that two different protocols * have the same preference, try to break the tie by comparing addresses. Not too * useful, but keeps the ordering of routes unambiguous. */ - return new->attrs->src->proto->proto > old->attrs->src->proto->proto; + return new->src->owner->class > old->src->owner->class; } - if (better = new->attrs->src->proto->rte_better) + if (better = new->src->owner->class->rte_better) return better(new, old); return 0; } @@ -539,268 +411,300 @@ rte_mergable(rte *pri, rte *sec) if (!rte_is_valid(pri) || !rte_is_valid(sec)) return 0; - if (pri->pref != sec->pref) + if (pri->attrs->pref != sec->attrs->pref) return 0; - if (pri->attrs->src->proto->proto != sec->attrs->src->proto->proto) + if (pri->src->owner->class != sec->src->owner->class) return 0; - if (mergable = pri->attrs->src->proto->rte_mergable) + if (mergable = pri->src->owner->class->rte_mergable) return mergable(pri, sec); return 0; } static void -rte_trace(struct channel *c, rte *e, int dir, char *msg) +rte_trace(const char *name, const rte *e, int dir, const char *msg) { - log(L_TRACE "%s.%s %c %s %N %s", - c->proto->name, c->name ?: "?", dir, msg, e->net->n.addr, - rta_dest_name(e->attrs->dest)); + log(L_TRACE "%s %c %s %N src %uL %uG %uS id %u %s%s", + name, dir, msg, e->net, + e->src->private_id, e->src->global_id, e->stale_cycle, e->id, + rta_dest_name(e->attrs->dest), + rte_is_filtered(e) ? " (filtered)" : ""); } static inline void -rte_trace_in(uint flag, struct channel *c, rte *e, char *msg) +channel_rte_trace_in(uint flag, struct channel *c, const rte *e, const char *msg) { if ((c->debug & flag) || (c->proto->debug & flag)) - rte_trace(c, e, '>', msg); + rte_trace(c->in_req.name, e, '>', msg); } static inline void -rte_trace_out(uint flag, struct channel *c, rte *e, char *msg) +channel_rte_trace_out(uint flag, struct channel *c, const rte *e, const char *msg) { if ((c->debug & flag) || (c->proto->debug & flag)) - rte_trace(c, e, '<', msg); + rte_trace(c->out_req.name, e, '<', msg); +} + +static inline void +rt_rte_trace_in(uint flag, struct rt_import_request *req, const rte *e, const char *msg) +{ + if (req->trace_routes & flag) + rte_trace(req->name, e, '>', msg); +} + +#if 0 +// seems to be unused at all +static inline void +rt_rte_trace_out(uint flag, struct rt_export_request *req, const rte *e, const char *msg) +{ + if (req->trace_routes & flag) + rte_trace(req->name, e, '<', msg); +} +#endif + +static uint +rte_feed_count(net *n) +{ + uint count = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTES_OR_NULL(e))) + count++; + return count; +} + +static void +rte_feed_obtain(net *n, struct rte **feed, uint count) +{ + uint i = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTES_OR_NULL(e))) + { + ASSERT_DIE(i < count); + feed[i++] = &e->rte; + } + ASSERT_DIE(i == count); } static rte * -export_filter_(struct channel *c, rte *rt0, rte **rt_free, linpool *pool, int silent) +export_filter_(struct channel *c, rte *rt, linpool *pool, int silent) { struct proto *p = c->proto; const struct filter *filter = c->out_filter; - struct proto_stats *stats = &c->stats; - rte *rt; - int v; + struct channel_export_stats *stats = &c->export_stats; - rt = rt0; - *rt_free = NULL; + /* Do nothing if we have already rejected the route */ + if (silent && bmap_test(&c->export_reject_map, rt->id)) + goto reject_noset; - v = p->preexport ? p->preexport(p, &rt, pool) : 0; + int v = p->preexport ? p->preexport(c, rt) : 0; if (v < 0) { if (silent) - goto reject; + goto reject_noset; - stats->exp_updates_rejected++; + stats->updates_rejected++; if (v == RIC_REJECT) - rte_trace_out(D_FILTERS, c, rt, "rejected by protocol"); + channel_rte_trace_out(D_FILTERS, c, rt, "rejected by protocol"); goto reject; + } if (v > 0) { if (!silent) - rte_trace_out(D_FILTERS, c, rt, "forced accept by protocol"); + channel_rte_trace_out(D_FILTERS, c, rt, "forced accept by protocol"); goto accept; } - rte_make_tmp_attrs(&rt, pool, NULL); - v = filter && ((filter == FILTER_REJECT) || - (f_run(filter, &rt, pool, + (f_run(filter, rt, pool, (silent ? FF_SILENT : 0)) > F_ACCEPT)); if (v) { if (silent) goto reject; - stats->exp_updates_filtered++; - rte_trace_out(D_FILTERS, c, rt, "filtered out"); + stats->updates_filtered++; + channel_rte_trace_out(D_FILTERS, c, rt, "filtered out"); goto reject; } -#ifdef CONFIG_PIPE - /* Pipes need rte with stored tmpattrs, remaining protocols need expanded tmpattrs */ - if (p->proto == &proto_pipe) - rte_store_tmp_attrs(rt, pool, NULL); -#endif - accept: - if (rt != rt0) - *rt_free = rt; + /* We have accepted the route */ + bmap_clear(&c->export_reject_map, rt->id); return rt; reject: + /* We have rejected the route by filter */ + bmap_set(&c->export_reject_map, rt->id); + +reject_noset: /* Discard temporary rte */ - if (rt != rt0) - rte_free(rt); return NULL; } static inline rte * -export_filter(struct channel *c, rte *rt0, rte **rt_free, int silent) +export_filter(struct channel *c, rte *rt, int silent) { - return export_filter_(c, rt0, rt_free, rte_update_pool, silent); + return export_filter_(c, rt, c->rte_update_pool, silent); } +void do_rt_notify_direct(struct channel *c, const net_addr *net, rte *new, const rte *old); + static void -do_rt_notify(struct channel *c, net *net, rte *new, rte *old, int refeed) +do_rt_notify(struct channel *c, const net_addr *net, rte *new, const rte *old) { - struct proto *p = c->proto; - struct proto_stats *stats = &c->stats; + struct channel_export_stats *stats = &c->export_stats; - if (refeed && new) + if (c->refeeding && new) c->refeed_count++; - /* Apply export limit */ - struct channel_limit *l = &c->out_limit; - if (l->action && !old && new) - { - if (stats->exp_routes >= l->limit) - channel_notify_limit(c, l, PLD_OUT, stats->exp_routes); - - if (l->state == PLS_BLOCKED) + if (!old && new) + if (CHANNEL_LIMIT_PUSH(c, OUT)) { - stats->exp_updates_rejected++; - rte_trace_out(D_FILTERS, c, new, "rejected [limit]"); + stats->updates_rejected++; + channel_rte_trace_out(D_FILTERS, c, new, "rejected [limit]"); return; } - } - /* Apply export table */ - if (c->out_table && !rte_update_out(c, net->n.addr, new, old, refeed)) - return; - - if (new) - stats->exp_updates_accepted++; - else - stats->exp_withdraws_accepted++; + if (!new && old) + CHANNEL_LIMIT_POP(c, OUT); + /* Store route export state */ if (old) - { bmap_clear(&c->export_map, old->id); - stats->exp_routes--; - } if (new) - { bmap_set(&c->export_map, new->id); - stats->exp_routes++; - } + + /* Apply export table */ + if (c->out_table) + rte_import(&c->out_table->push, net, new, old ? old->src : new->src); + else + do_rt_notify_direct(c, net, new, old); +} + +void +do_rt_notify_direct(struct channel *c, const net_addr *net, rte *new, const rte *old) +{ + struct proto *p = c->proto; + struct channel_export_stats *stats = &c->export_stats; + + if (new) + stats->updates_accepted++; + else + stats->withdraws_accepted++; if (p->debug & D_ROUTES) { if (new && old) - rte_trace_out(D_ROUTES, c, new, "replaced"); + channel_rte_trace_out(D_ROUTES, c, new, "replaced"); else if (new) - rte_trace_out(D_ROUTES, c, new, "added"); + channel_rte_trace_out(D_ROUTES, c, new, "added"); else if (old) - rte_trace_out(D_ROUTES, c, old, "removed"); + channel_rte_trace_out(D_ROUTES, c, old, "removed"); } p->rt_notify(p, c, net, new, old); } static void -rt_notify_basic(struct channel *c, net *net, rte *new, rte *old, int refeed) +rt_notify_basic(struct channel *c, const net_addr *net, rte *new, rte *old) { - // struct proto *p = c->proto; - rte *new_free = NULL; - - if (new) - c->stats.exp_updates_received++; - else - c->stats.exp_withdraws_received++; - if (new) - new = export_filter(c, new, &new_free, 0); + new = export_filter(c, new, 0); if (old && !bmap_test(&c->export_map, old->id)) old = NULL; + if (old && (old->sender == c->in_req.hook)) + bug("bad-behaved pipe"); + if (!new && !old) return; - do_rt_notify(c, net, new, old, refeed); - - /* Discard temporary rte */ - if (new_free) - rte_free(new_free); + do_rt_notify(c, net, new, old); } static void -rt_notify_accepted(struct channel *c, net *net, rte *new_changed, rte *old_changed, int refeed) +channel_rpe_mark_seen(struct rt_export_request *req, struct rt_pending_export *rpe) { - // struct proto *p = c->proto; - rte *new_best = NULL; - rte *old_best = NULL; - rte *new_free = NULL; - int new_first = 0; - - /* - * We assume that there are no changes in net route order except (added) - * new_changed and (removed) old_changed. Therefore, the function is not - * compatible with deterministic_med (where nontrivial reordering can happen - * as a result of a route change) and with recomputation of recursive routes - * due to next hop update (where many routes can be changed in one step). - * - * Note that we need this assumption just for optimizations, we could just - * run full new_best recomputation otherwise. - * - * There are three cases: - * feed or old_best is old_changed -> we need to recompute new_best - * old_best is before new_changed -> new_best is old_best, ignore - * old_best is after new_changed -> try new_changed, otherwise old_best - */ + struct channel *c = SKIP_BACK(struct channel, out_req, req); - if (net->routes) - c->stats.exp_updates_received++; - else - c->stats.exp_withdraws_received++; + rpe_mark_seen(req->hook, rpe); + if (rpe->old) + bmap_clear(&c->export_reject_map, rpe->old->rte.id); +} - /* Find old_best - either old_changed, or route for net->routes */ - if (old_changed && bmap_test(&c->export_map, old_changed->id)) - old_best = old_changed; - else +void +rt_notify_accepted(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *rpe, + struct rte **feed, uint count) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + + rte_update_lock(c); + + rte nb0, *new_best = NULL; + const rte *old_best = NULL; + + for (uint i = 0; i < count; i++) { - for (rte *r = net->routes; rte_is_valid(r); r = r->next) + if (!rte_is_valid(feed[i])) + continue; + + /* Has been already rejected, won't bother with it */ + if (!c->refeeding && bmap_test(&c->export_reject_map, feed[i]->id)) + continue; + + /* Previously exported */ + if (!old_best && bmap_test(&c->export_map, feed[i]->id)) { - if (bmap_test(&c->export_map, r->id)) + /* is still best */ + if (!new_best) { - old_best = r; - break; + DBG("rt_notify_accepted: idempotent\n"); + goto done; } - /* Note if new_changed found before old_best */ - if (r == new_changed) - new_first = 1; + /* is superseded */ + old_best = feed[i]; + break; } - } - /* Find new_best */ - if ((new_changed == old_changed) || (old_best == old_changed)) - { - /* Feed or old_best changed -> find first accepted by filters */ - for (rte *r = net->routes; rte_is_valid(r); r = r->next) - if (new_best = export_filter(c, r, &new_free, 0)) - break; + /* Have no new best route yet */ + if (!new_best) + { + /* Try this route not seen before */ + nb0 = *feed[i]; + new_best = export_filter(c, &nb0, 0); + DBG("rt_notify_accepted: checking route id %u: %s\n", feed[i]->id, new_best ? "ok" : "no"); + } } - else + +done: + /* Check obsolete routes for previously exported */ + while (rpe) { - /* Other cases -> either new_changed, or old_best (and nothing changed) */ - if (new_first && (new_changed = export_filter(c, new_changed, &new_free, 0))) - new_best = new_changed; - else - return; + channel_rpe_mark_seen(req, rpe); + if (rpe->old) + { + if (bmap_test(&c->export_map, rpe->old->rte.id)) + { + ASSERT_DIE(old_best == NULL); + old_best = &rpe->old->rte; + } + } + rpe = rpe_next(rpe, NULL); } - if (!new_best && !old_best) - return; - - do_rt_notify(c, net, new_best, old_best, refeed); + /* Nothing to export */ + if (new_best || old_best) + do_rt_notify(c, n, new_best, old_best); + else + DBG("rt_notify_accepted: nothing to export\n"); - /* Discard temporary rte */ - if (new_free) - rte_free(new_free); + rte_update_unlock(c); } @@ -811,38 +715,45 @@ nexthop_merge_rta(struct nexthop *nhs, rta *a, linpool *pool, int max) } rte * -rt_export_merged(struct channel *c, net *net, rte **rt_free, linpool *pool, int silent) +rt_export_merged(struct channel *c, struct rte **feed, uint count, linpool *pool, int silent) { + _Thread_local static rte rloc; + // struct proto *p = c->proto; struct nexthop *nhs = NULL; - rte *best0, *best, *rt0, *rt, *tmp; - - best0 = net->routes; - *rt_free = NULL; + rte *best0 = feed[0]; + rte *best = NULL; if (!rte_is_valid(best0)) return NULL; - best = export_filter_(c, best0, rt_free, pool, silent); + /* Already rejected, no need to re-run the filter */ + if (!c->refeeding && bmap_test(&c->export_reject_map, best0->id)) + return NULL; + + rloc = *best0; + best = export_filter_(c, &rloc, pool, silent); + + if (!best) + /* Best route doesn't pass the filter */ + return NULL; - if (!best || !rte_is_reachable(best)) + if (!rte_is_reachable(best)) + /* Unreachable routes can't be merged */ return best; - for (rt0 = best0->next; rt0; rt0 = rt0->next) + for (uint i = 1; i < count; i++) { - if (!rte_mergable(best0, rt0)) + if (!rte_mergable(best0, feed[i])) continue; - rt = export_filter_(c, rt0, &tmp, pool, 1); + rte tmp0 = *feed[i]; + rte *tmp = export_filter_(c, &tmp0, pool, 1); - if (!rt) + if (!tmp || !rte_is_reachable(tmp)) continue; - if (rte_is_reachable(rt)) - nhs = nexthop_merge_rta(nhs, rt->attrs, pool, c->merge_limit); - - if (tmp) - rte_free(tmp); + nhs = nexthop_merge_rta(nhs, tmp->attrs, pool, c->merge_limit); } if (nhs) @@ -851,66 +762,208 @@ rt_export_merged(struct channel *c, net *net, rte **rt_free, linpool *pool, int if (nhs->next) { - best = rte_cow_rta(best, pool); + best->attrs = rta_cow(best->attrs, pool); nexthop_link(best->attrs, nhs); } } - if (best != best0) - *rt_free = best; - return best; } - -static void -rt_notify_merged(struct channel *c, net *net, rte *new_changed, rte *old_changed, - rte *new_best, rte *old_best, int refeed) +void +rt_notify_merged(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *rpe, + struct rte **feed, uint count) { - // struct proto *p = c->proto; - rte *new_free = NULL; + struct channel *c = SKIP_BACK(struct channel, out_req, req); - /* We assume that all rte arguments are either NULL or rte_is_valid() */ - - /* This check should be done by the caller */ - if (!new_best && !old_best) - return; + rte_update_lock(c); + // struct proto *p = c->proto; +#if 0 /* TODO: Find whether this check is possible when processing multiple changes at once. */ /* Check whether the change is relevant to the merged route */ if ((new_best == old_best) && (new_changed != old_changed) && !rte_mergable(new_best, new_changed) && !rte_mergable(old_best, old_changed)) return; +#endif - if (new_best) - c->stats.exp_updates_received++; - else - c->stats.exp_withdraws_received++; + rte *old_best = NULL; + /* Find old best route */ + for (uint i = 0; i < count; i++) + if (bmap_test(&c->export_map, feed[i]->id)) + { + old_best = feed[i]; + break; + } + + /* Check obsolete routes for previously exported */ + while (rpe) + { + channel_rpe_mark_seen(req, rpe); + if (rpe->old) + { + if (bmap_test(&c->export_map, rpe->old->rte.id)) + { + ASSERT_DIE(old_best == NULL); + old_best = &rpe->old->rte; + } + } + rpe = rpe_next(rpe, NULL); + } /* Prepare new merged route */ - if (new_best) - new_best = rt_export_merged(c, net, &new_free, rte_update_pool, 0); + rte *new_merged = count ? rt_export_merged(c, feed, count, c->rte_update_pool, 0) : NULL; - /* Check old merged route */ - if (old_best && !bmap_test(&c->export_map, old_best->id)) - old_best = NULL; + if (new_merged || old_best) + do_rt_notify(c, n, new_merged, old_best); - if (!new_best && !old_best) - return; + rte_update_unlock(c); +} - do_rt_notify(c, net, new_best, old_best, refeed); +void +rt_notify_optimal(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + rte_update_lock(c); + rte *old = RTES_OR_NULL(rpe->old_best); + struct rte_storage *new_best = rpe->new_best; - /* Discard temporary rte */ - if (new_free) - rte_free(new_free); + while (rpe) + { + channel_rpe_mark_seen(req, rpe); + new_best = rpe->new_best; + rpe = rpe_next(rpe, NULL); + } + + if (&new_best->rte != old) + { + rte n0, *new = RTES_CLONE(new_best, &n0); + rt_notify_basic(c, net, new, old); + } + + rte_update_unlock(c); +} + +void +rt_notify_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + rte_update_lock(c); + struct rte_src *src = rpe->new ? rpe->new->rte.src : rpe->old->rte.src; + rte *old = RTES_OR_NULL(rpe->old); + struct rte_storage *new_any = rpe->new; + + while (rpe) + { + channel_rpe_mark_seen(req, rpe); + new_any = rpe->new; + rpe = rpe_next(rpe, src); + } + + if (&new_any->rte != old) + { + rte n0, *new = RTES_CLONE(new_any, &n0); + rt_notify_basic(c, net, new, old); + } + + rte_update_unlock(c); +} + +void +rt_feed_any(struct rt_export_request *req, const net_addr *net, struct rt_pending_export *rpe UNUSED, rte **feed, uint count) +{ + struct channel *c = SKIP_BACK(struct channel, out_req, req); + rte_update_lock(c); + + for (uint i=0; i<count; i++) + { + rte n0 = *feed[i]; + rt_notify_basic(c, net, &n0, NULL); + } + + rte_update_unlock(c); +} + +void +rpe_mark_seen(struct rt_export_hook *hook, struct rt_pending_export *rpe) +{ + bmap_set(&hook->seq_map, rpe->seq); +} + +struct rt_pending_export * +rpe_next(struct rt_pending_export *rpe, struct rte_src *src) +{ + struct rt_pending_export *next = atomic_load_explicit(&rpe->next, memory_order_acquire); + + if (!next) + return NULL; + + if (!src) + return next; + + while (rpe = next) + if (src == (rpe->new ? rpe->new->rte.src : rpe->old->rte.src)) + return rpe; + else + next = atomic_load_explicit(&rpe->next, memory_order_acquire); + + return NULL; } +static struct rt_pending_export * rt_next_export_fast(struct rt_pending_export *last); +static void +rte_export(struct rt_export_hook *hook, struct rt_pending_export *rpe) +{ + if (bmap_test(&hook->seq_map, rpe->seq)) + goto seen; + + const net_addr *n = rpe->new_best ? rpe->new_best->rte.net : rpe->old_best->rte.net; + + if (rpe->new) + hook->stats.updates_received++; + else + hook->stats.withdraws_received++; + + if (hook->req->export_one) + hook->req->export_one(hook->req, n, rpe); + else if (hook->req->export_bulk) + { + RT_LOCK(hook->table); + net *net = SKIP_BACK(struct network, n.addr, (net_addr (*)[0]) n); + uint count = rte_feed_count(net); + rte **feed = NULL; + if (count) + { + feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(net, feed, count); + } + RT_UNLOCK(hook->table); + hook->req->export_bulk(hook->req, n, rpe, feed, count); + } + else + bug("Export request must always provide an export method"); + +seen: + /* Get the next export if exists */ + hook->rpe_next = rt_next_export_fast(rpe); + + /* The last block may be available to free */ + if (PAGE_HEAD(hook->rpe_next) != PAGE_HEAD(rpe)) + { + RT_LOCK(hook->table); + rt_export_used(RT_PRIV(hook->table)); + RT_UNLOCK(hook->table); + } + + /* Releasing this export for cleanup routine */ + DBG("store hook=%p last_export=%p seq=%lu\n", hook, rpe, rpe->seq); + atomic_store_explicit(&hook->last_export, rpe, memory_order_release); +} /** * rte_announce - announce a routing table change * @tab: table the route has been added to - * @type: type of route announcement (RA_UNDEF or RA_ANY) * @net: network in question * @new: the new or changed route * @old: the previous route replaced by the new one @@ -926,13 +979,6 @@ rt_notify_merged(struct channel *c, net *net, rte *new_changed, rte *old_changed * and @new_best and @old_best describes best routes. Other routes are not * affected, but in sorted table the order of other routes might change. * - * Second, There is a bulk change of multiple routes in @net, with shared best - * route selection. In such case separate route changes are described using - * @type of %RA_ANY, with @new and @old specifying the changed route, while - * @new_best and @old_best are NULL. After that, another notification is done - * where @new_best and @old_best are filled (may be the same), but @new and @old - * are NULL. - * * The function announces the change to all associated channels. For each * channel, an appropriate preprocessing is done according to channel &ra_mode. * For example, %RA_OPTIMAL channels receive just changes of best routes. @@ -947,130 +993,304 @@ rt_notify_merged(struct channel *c, net *net, rte *new_changed, rte *old_changed * done outside of scope of rte_announce(). */ static void -rte_announce(rtable *tab, uint type, net *net, rte *new, rte *old, - rte *new_best, rte *old_best) +rte_announce(rtable_private *tab, net *net, struct rte_storage *new, struct rte_storage *old, + struct rte_storage *new_best, struct rte_storage *old_best) { - if (!rte_is_valid(new)) - new = NULL; - - if (!rte_is_valid(old)) - old = NULL; - - if (!rte_is_valid(new_best)) + if (!new_best || !rte_is_valid(&new_best->rte)) new_best = NULL; - if (!rte_is_valid(old_best)) + if (!old_best || !rte_is_valid(&old_best->rte)) old_best = NULL; - if (!new && !old && !new_best && !old_best) + if (!new || !rte_is_valid(&new->rte)) + new = NULL; + + if (old && !rte_is_valid(&old->rte)) + { + /* Filtered old route isn't announced, should be freed immediately. */ + rte_free(old, tab); + old = NULL; + } + + if ((new == old) && (new_best == old_best)) return; if (new_best != old_best) { if (new_best) - new_best->sender->stats.pref_routes++; + new_best->rte.sender->stats.pref++; if (old_best) - old_best->sender->stats.pref_routes--; + old_best->rte.sender->stats.pref--; if (tab->hostcache) rt_notify_hostcache(tab, net); } + if (EMPTY_LIST(tab->exports) && EMPTY_LIST(tab->pending_exports)) + { + /* No export hook and no pending exports to cleanup. We may free the route immediately. */ + if (!old) + return; + + hmap_clear(&tab->id_map, old->rte.id); + rte_free(old, tab); + return; + } + + /* Get the pending export structure */ + struct rt_export_block *rpeb = NULL, *rpebsnl = NULL; + u32 end = 0; + + if (!EMPTY_LIST(tab->pending_exports)) + { + rpeb = TAIL(tab->pending_exports); + end = atomic_load_explicit(&rpeb->end, memory_order_relaxed); + if (end >= RT_PENDING_EXPORT_ITEMS) + { + ASSERT_DIE(end == RT_PENDING_EXPORT_ITEMS); + rpebsnl = rpeb; + + rpeb = NULL; + end = 0; + } + } + + if (!rpeb) + { + rpeb = alloc_page(); + *rpeb = (struct rt_export_block) {}; + add_tail(&tab->pending_exports, &rpeb->n); + } + + /* Fill the pending export */ + struct rt_pending_export *rpe = &rpeb->export[rpeb->end]; + *rpe = (struct rt_pending_export) { + .new = new, + .new_best = new_best, + .old = old, + .old_best = old_best, + .seq = tab->next_export_seq++, + }; + + DBG("rte_announce: table=%s net=%N new=%p from %p old=%p from %p new_best=%p old_best=%p seq=%lu\n", tab->name, net->n.addr, new, new ? new->sender : NULL, old, old ? old->sender : NULL, new_best, old_best, rpe->seq); + + ASSERT_DIE(atomic_fetch_add_explicit(&rpeb->end, 1, memory_order_release) == end); + + if (rpebsnl) + { + _Bool f = 0; + ASSERT_DIE(atomic_compare_exchange_strong_explicit(&rpebsnl->not_last, &f, 1, + memory_order_release, memory_order_relaxed)); + } + + /* Append to the same-network squasher list */ + if (net->last) + { + struct rt_pending_export *rpenull = NULL; + ASSERT_DIE(atomic_compare_exchange_strong_explicit( + &net->last->next, &rpenull, rpe, + memory_order_relaxed, + memory_order_relaxed)); + + } + + net->last = rpe; + + if (!net->first) + net->first = rpe; + + if (tab->first_export == NULL) + tab->first_export = rpe; + + if (!EMPTY_LIST(tab->exports) && + (tab->first_export->seq + tab->config->cork_limit <= tab->next_export_seq) && + !tab->cork_active) + { + if (config->table_debug) + log(L_TRACE "%s: cork activated", tab->name); + + ev_cork(&rt_cork); + tab->cork_active = 1; + } +} + +static struct rt_pending_export * +rt_next_export_fast(struct rt_pending_export *last) +{ + /* Get the whole export block and find our position in there. */ + struct rt_export_block *rpeb = PAGE_HEAD(last); + u32 pos = (last - &rpeb->export[0]); + u32 end = atomic_load_explicit(&rpeb->end, memory_order_acquire); + ASSERT_DIE(pos < end); + + /* Next is in the same block. */ + if (++pos < end) + return &rpeb->export[pos]; + + /* There is another block. */ + if (atomic_load_explicit(&rpeb->not_last, memory_order_acquire)) + { + /* This is OK to do non-atomically because of the not_last flag. */ + rpeb = NODE_NEXT(rpeb); + return &rpeb->export[0]; + } + + /* There is nothing more. */ + return NULL; +} + +static struct rt_pending_export * +rt_next_export(struct rt_export_hook *hook, rtable_private *tab) +{ + /* As the table is locked, it is safe to reload the last export pointer */ + struct rt_pending_export *last = atomic_load_explicit(&hook->last_export, memory_order_acquire); + + /* It is still valid, let's reuse it */ + if (last) + return rt_next_export_fast(last); + + /* No, therefore we must process the table's first pending export */ + else + return tab->first_export; +} + +static inline void +rt_send_export_event(struct rt_export_hook *hook) +{ + ev_send(hook->req->list, hook->event); +} + +static void +rt_announce_exports(void *data) +{ + rtable_private *tab = data; + ASSERT_DIE(birdloop_inside(tab->loop)); + rt_schedule_notify(tab); - struct channel *c; node *n; - WALK_LIST2(c, n, tab->channels, table_node) + struct rt_export_hook *c; node *n; + WALK_LIST2(c, n, tab->exports, n) { - if (c->export_state == ES_DOWN) + if (atomic_load_explicit(&c->export_state, memory_order_acquire) != TES_READY) continue; - if (type && (type != c->ra_mode)) - continue; + rt_send_export_event(c); + } +} - switch (c->ra_mode) +static void +rt_import_announce_exports(void *data) +{ + struct rt_import_hook *hook = data; + RT_LOCKED(hook->table, tab) + { + if (hook->import_state == TIS_CLEARED) { - case RA_OPTIMAL: - if (new_best != old_best) - rt_notify_basic(c, net, new_best, old_best, 0); - break; + rfree(hook->export_announce_event); - case RA_ANY: - if (new != old) - rt_notify_basic(c, net, new, old, 0); - break; + ev_send(hook->stopped->list, hook->stopped); + rem_node(&hook->n); + mb_free(hook); + rt_unlock_table(tab); + } + else + ev_send_loop(tab->loop, tab->announce_event); + } +} - case RA_ACCEPTED: - rt_notify_accepted(c, net, new, old, 0); - break; +static struct rt_pending_export * +rt_last_export(rtable_private *tab) +{ + struct rt_pending_export *rpe = NULL; - case RA_MERGED: - rt_notify_merged(c, net, new, old, new_best, old_best, 0); - break; + if (!EMPTY_LIST(tab->pending_exports)) + { + /* We'll continue processing exports from this export on */ + struct rt_export_block *reb = TAIL(tab->pending_exports); + ASSERT_DIE(reb->end); + rpe = &reb->export[reb->end - 1]; + } + + return rpe; +} + +#define RT_EXPORT_BULK 1024 + +static void +rt_export_hook(void *_data) +{ + struct rt_export_hook *c = _data; + + ASSERT_DIE(atomic_load_explicit(&c->export_state, memory_order_relaxed) == TES_READY); + + if (!c->rpe_next) + { + RT_LOCK(c->table); + c->rpe_next = rt_next_export(c, RT_PRIV(c->table)); + + if (!c->rpe_next) + { + rt_export_used(RT_PRIV(c->table)); + RT_UNLOCK(c->table); + return; } + + RT_UNLOCK(c->table); + } + + /* Process the export */ + for (uint i=0; i<RT_EXPORT_BULK; i++) + { + rte_export(c, c->rpe_next); + + if (!c->rpe_next) + break; } + + rt_send_export_event(c); } + static inline int -rte_validate(rte *e) +rte_validate(struct channel *ch, rte *e) { int c; - net *n = e->net; + const net_addr *n = e->net; - if (!net_validate(n->n.addr)) + if (!net_validate(n)) { log(L_WARN "Ignoring bogus prefix %N received via %s", - n->n.addr, e->sender->proto->name); + n, ch->proto->name); return 0; } /* FIXME: better handling different nettypes */ - c = !net_is_flow(n->n.addr) ? - net_classify(n->n.addr): (IADDR_HOST | SCOPE_UNIVERSE); + c = !net_is_flow(n) ? + net_classify(n): (IADDR_HOST | SCOPE_UNIVERSE); if ((c < 0) || !(c & IADDR_HOST) || ((c & IADDR_SCOPE_MASK) <= SCOPE_LINK)) { log(L_WARN "Ignoring bogus route %N received via %s", - n->n.addr, e->sender->proto->name); + n, ch->proto->name); return 0; } - if (net_type_match(n->n.addr, NB_DEST) == !e->attrs->dest) + if (net_type_match(n, NB_DEST) == !e->attrs->dest) { log(L_WARN "Ignoring route %N with invalid dest %d received via %s", - n->n.addr, e->attrs->dest, e->sender->proto->name); + n, e->attrs->dest, ch->proto->name); return 0; } if ((e->attrs->dest == RTD_UNICAST) && !nexthop_is_sorted(&(e->attrs->nh))) { log(L_WARN "Ignoring unsorted multipath route %N received via %s", - n->n.addr, e->sender->proto->name); + n, ch->proto->name); return 0; } return 1; } -/** - * rte_free - delete a &rte - * @e: &rte to be deleted - * - * rte_free() deletes the given &rte from the routing table it's linked to. - */ -void -rte_free(rte *e) -{ - if (rta_is_cached(e->attrs)) - rta_free(e->attrs); - sl_free(rte_slab, e); -} - -static inline void -rte_free_quick(rte *e) -{ - rta_free(e->attrs); - sl_free(rte_slab, e); -} - static int rte_same(rte *x, rte *y) { @@ -1078,175 +1298,105 @@ rte_same(rte *x, rte *y) return x->attrs == y->attrs && x->pflags == y->pflags && - x->pref == y->pref && - (!x->attrs->src->proto->rte_same || x->attrs->src->proto->rte_same(x, y)) && + x->src == y->src && rte_is_filtered(x) == rte_is_filtered(y); } static inline int rte_is_ok(rte *e) { return e && !rte_is_filtered(e); } static void -rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) +rte_recalculate(rtable_private *table, struct rt_import_hook *c, net *net, rte *new, struct rte_src *src) { - struct proto *p = c->proto; - struct rtable *table = c->table; - struct proto_stats *stats = &c->stats; - static struct tbf rl_pipe = TBF_DEFAULT_LOG_LIMITS; - rte *before_old = NULL; - rte *old_best = net->routes; + struct rt_import_request *req = c->req; + struct rt_import_stats *stats = &c->stats; + struct rte_storage *old_best_stored = net->routes, *old_stored = NULL; + rte *old_best = old_best_stored ? &old_best_stored->rte : NULL; rte *old = NULL; - rte **k; - k = &net->routes; /* Find and remove original route from the same protocol */ - while (old = *k) + /* Set the stale cycle unless already set */ + if (new && !(new->flags & REF_USE_STALE)) + new->stale_cycle = c->stale_set; + + /* Find and remove original route from the same protocol */ + struct rte_storage **before_old = rte_find(net, src); + + if (*before_old) { - if (old->attrs->src == src) + old = &(old_stored = (*before_old))->rte; + + /* If there is the same route in the routing table but from + * a different sender, then there are two paths from the + * source protocol to this routing table through transparent + * pipes, which is not allowed. + * We log that and ignore the route. */ + if (old->sender != c) { - /* If there is the same route in the routing table but from - * a different sender, then there are two paths from the - * source protocol to this routing table through transparent - * pipes, which is not allowed. - * - * We log that and ignore the route. If it is withdraw, we - * ignore it completely (there might be 'spurious withdraws', - * see FIXME in do_rte_announce()) - */ - if (old->sender->proto != p) - { - if (new) - { - log_rl(&rl_pipe, L_ERR "Pipe collision detected when sending %N to table %s", - net->n.addr, table->name); - rte_free_quick(new); - } - return; - } + if (!old->generation && !new->generation) + bug("Two protocols claim to author a route with the same rte_src in table %s: %N %s/%u:%u", + c->table->name, net->n.addr, old->src->owner->name, old->src->private_id, old->src->global_id); + + log_rl(&table->rl_pipe, L_ERR "Route source collision in table %s: %N %s/%u:%u", + c->table->name, net->n.addr, old->src->owner->name, old->src->private_id, old->src->global_id); + } if (new && rte_same(old, new)) { /* No changes, ignore the new route and refresh the old one */ - - old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY); + old->stale_cycle = new->stale_cycle; if (!rte_is_filtered(new)) { - stats->imp_updates_ignored++; - rte_trace_in(D_ROUTES, c, new, "ignored"); + stats->updates_ignored++; + rt_rte_trace_in(D_ROUTES, req, new, "ignored"); } - - rte_free_quick(new); return; - } - *k = old->next; - table->rt_count--; - break; - } - k = &old->next; - before_old = old; - } - - /* Save the last accessed position */ - rte **pos = k; + } - if (!old) - before_old = NULL; + *before_old = (*before_old)->next; + table->rt_count--; + } if (!old && !new) { - stats->imp_withdraws_ignored++; + stats->withdraws_ignored++; return; } + if (req->preimport) + new = req->preimport(req, new, old); + int new_ok = rte_is_ok(new); int old_ok = rte_is_ok(old); - struct channel_limit *l = &c->rx_limit; - if (l->action && !old && new && !c->in_table) - { - u32 all_routes = stats->imp_routes + stats->filt_routes; - - if (all_routes >= l->limit) - channel_notify_limit(c, l, PLD_RX, all_routes); - - if (l->state == PLS_BLOCKED) - { - /* In receive limit the situation is simple, old is NULL so - we just free new and exit like nothing happened */ - - stats->imp_updates_ignored++; - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - rte_free_quick(new); - return; - } - } - - l = &c->in_limit; - if (l->action && !old_ok && new_ok) - { - if (stats->imp_routes >= l->limit) - channel_notify_limit(c, l, PLD_IN, stats->imp_routes); - - if (l->state == PLS_BLOCKED) - { - /* In import limit the situation is more complicated. We - shouldn't just drop the route, we should handle it like - it was filtered. We also have to continue the route - processing if old or new is non-NULL, but we should exit - if both are NULL as this case is probably assumed to be - already handled. */ - - stats->imp_updates_ignored++; - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - - if (c->in_keep_filtered) - new->flags |= REF_FILTERED; - else - { rte_free_quick(new); new = NULL; } - - /* Note that old && !new could be possible when - c->in_keep_filtered changed in the recent past. */ - - if (!old && !new) - return; - - new_ok = 0; - goto skip_stats1; - } - } - if (new_ok) - stats->imp_updates_accepted++; + stats->updates_accepted++; else if (old_ok) - stats->imp_withdraws_accepted++; + stats->withdraws_accepted++; else - stats->imp_withdraws_ignored++; + stats->withdraws_ignored++; if (old_ok || new_ok) table->last_rt_change = current_time(); - skip_stats1: - - if (new) - rte_is_filtered(new) ? stats->filt_routes++ : stats->imp_routes++; - if (old) - rte_is_filtered(old) ? stats->filt_routes-- : stats->imp_routes--; + struct rte_storage *new_stored = new ? rte_store(new, net, table) : NULL; if (table->config->sorted) { /* If routes are sorted, just insert new route to appropriate position */ - if (new) + if (new_stored) { - if (before_old && !rte_better(new, before_old)) - k = &before_old->next; + struct rte_storage **k; + if ((before_old != &net->routes) && !rte_better(new, &SKIP_BACK(struct rte_storage, next, before_old)->rte)) + k = before_old; else k = &net->routes; for (; *k; k=&(*k)->next) - if (rte_better(new, *k)) + if (rte_better(new, &(*k)->rte)) break; - new->next = *k; - *k = new; + new_stored->next = *k; + *k = new_stored; table->rt_count++; } @@ -1256,16 +1406,17 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) /* If routes are not sorted, find the best route and move it on the first position. There are several optimized cases. */ - if (src->proto->rte_recalculate && src->proto->rte_recalculate(table, net, new, old, old_best)) + if (src->owner->rte_recalculate && + src->owner->rte_recalculate(table, net, new_stored ? &new_stored->rte : NULL, old, old_best)) goto do_recalculate; - if (new && rte_better(new, old_best)) + if (new_stored && rte_better(&new_stored->rte, old_best)) { /* The first case - the new route is cleary optimal, we link it at the first position */ - new->next = net->routes; - net->routes = new; + new_stored->next = net->routes; + net->routes = new_stored; table->rt_count++; } @@ -1279,10 +1430,10 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) do_recalculate: /* Add the new route to the list */ - if (new) + if (new_stored) { - new->next = *pos; - *pos = new; + new_stored->next = *before_old; + *before_old = new_stored; table->rt_count++; } @@ -1290,335 +1441,390 @@ rte_recalculate(struct channel *c, net *net, rte *new, struct rte_src *src) /* Find a new optimal route (if there is any) */ if (net->routes) { - rte **bp = &net->routes; - for (k=&(*bp)->next; *k; k=&(*k)->next) - if (rte_better(*k, *bp)) + struct rte_storage **bp = &net->routes; + for (struct rte_storage **k=&(*bp)->next; *k; k=&(*k)->next) + if (rte_better(&(*k)->rte, &(*bp)->rte)) bp = k; /* And relink it */ - rte *best = *bp; + struct rte_storage *best = *bp; *bp = best->next; best->next = net->routes; net->routes = best; } } - else if (new) + else if (new_stored) { /* The third case - the new route is not better than the old best route (therefore old_best != NULL) and the old best route was not removed (therefore old_best == net->routes). We just link the new route to the old/last position. */ - new->next = *pos; - *pos = new; + new_stored->next = *before_old; + *before_old = new_stored; table->rt_count++; } /* The fourth (empty) case - suboptimal route was removed, nothing to do */ } - if (new) + if (new_stored) { - new->lastmod = current_time(); - - if (!old) - { - new->id = hmap_first_zero(&table->id_map); - hmap_set(&table->id_map, new->id); - } - else - new->id = old->id; + new_stored->rte.lastmod = current_time(); + new_stored->rte.id = hmap_first_zero(&table->id_map); + hmap_set(&table->id_map, new_stored->rte.id); } + _Bool nb = (new_stored == net->routes); + _Bool ob = (old_best == old); + /* Log the route change */ - if ((c->debug & D_ROUTES) || (p->debug & D_ROUTES)) + if (new_ok && old_ok) { - if (new_ok) - rte_trace(c, new, '>', new == net->routes ? "added [best]" : "added"); - else if (old_ok) - { - if (old != old_best) - rte_trace(c, old, '>', "removed"); - else if (rte_is_ok(net->routes)) - rte_trace(c, old, '>', "removed [replaced]"); - else - rte_trace(c, old, '>', "removed [sole]"); - } + const char *best_indicator[2][2] = { { "updated", "updated [-best]" }, { "updated [+best]", "updated [best]" } }; + rt_rte_trace_in(D_ROUTES, req, &new_stored->rte, best_indicator[nb][ob]); } + else if (new_ok) + rt_rte_trace_in(D_ROUTES, req, &new_stored->rte, + (!net->routes->next || !rte_is_ok(&net->routes->next->rte)) ? "added [sole]" : + nb ? "added [best]" : "added"); + else if (old_ok) + rt_rte_trace_in(D_ROUTES, req, old, + (!net->routes || !rte_is_ok(&net->routes->rte)) ? "removed [sole]" : + ob ? "removed [best]" : "removed"); /* Propagate the route change */ - rte_announce(table, RA_UNDEF, net, new, old, net->routes, old_best); + rte_announce(table, net, new_stored, old_stored, + net->routes, old_best_stored); + + ev_send(req->list, c->export_announce_event); if (!net->routes && (table->gc_counter++ >= table->config->gc_max_ops) && (table->gc_time + table->config->gc_min_time <= current_time())) rt_schedule_prune(table); +#if 0 + /* Enable and reimplement these callbacks if anybody wants to use them */ if (old_ok && p->rte_remove) p->rte_remove(net, old); if (new_ok && p->rte_insert) - p->rte_insert(net, new); - - if (old) - { - if (!new) - hmap_clear(&table->id_map, old->id); + p->rte_insert(net, &new_stored->rte); +#endif - rte_free_quick(old); - } } -static int rte_update_nest_cnt; /* Nesting counter to allow recursive updates */ - -static inline void -rte_update_lock(void) -{ - rte_update_nest_cnt++; -} - -static inline void -rte_update_unlock(void) +rte * +channel_preimport(struct rt_import_request *req, rte *new, rte *old) { - if (!--rte_update_nest_cnt) - lp_flush(rte_update_pool); -} + struct channel *c = SKIP_BACK(struct channel, in_req, req); -static inline void -rte_hide_dummy_routes(net *net, rte **dummy) -{ - if (net->routes && net->routes->attrs->source == RTS_DUMMY) + if (!c->in_table) { - *dummy = net->routes; - net->routes = (*dummy)->next; + if (new && !old) + if (CHANNEL_LIMIT_PUSH(c, RX)) + return NULL; + + if (!new && old) + CHANNEL_LIMIT_POP(c, RX); } + + int new_in = new && !rte_is_filtered(new); + int old_in = old && !rte_is_filtered(old); + + if (new_in && !old_in) + if (CHANNEL_LIMIT_PUSH(c, IN)) + if (c->in_keep_filtered) + { + new->flags |= REF_FILTERED; + return new; + } + else + return NULL; + + if (!new_in && old_in) + CHANNEL_LIMIT_POP(c, IN); + + return new; } -static inline void -rte_unhide_dummy_routes(net *net, rte **dummy) +rte * +channel_in_preimport(struct rt_import_request *req, rte *new, rte *old) { - if (*dummy) - { - (*dummy)->next = net->routes; - net->routes = *dummy; - } + struct channel_aux_table *cat = SKIP_BACK(struct channel_aux_table, push, req); + + if (new && !old) + if (CHANNEL_LIMIT_PUSH(cat->c, RX)) + return NULL; + + if (!new && old) + CHANNEL_LIMIT_POP(cat->c, RX); + + return new; } -/** - * rte_update - enter a new update to a routing table - * @table: table to be updated - * @c: channel doing the update - * @net: network node - * @p: protocol submitting the update - * @src: protocol originating the update - * @new: a &rte representing the new route or %NULL for route removal. - * - * This function is called by the routing protocols whenever they discover - * a new route or wish to update/remove an existing route. The right announcement - * sequence is to build route attributes first (either un-cached with @aflags set - * to zero or a cached one using rta_lookup(); in this case please note that - * you need to increase the use count of the attributes yourself by calling - * rta_clone()), call rte_get_temp() to obtain a temporary &rte, fill in all - * the appropriate data and finally submit the new &rte by calling rte_update(). - * - * @src specifies the protocol that originally created the route and the meaning - * of protocol-dependent data of @new. If @new is not %NULL, @src have to be the - * same value as @new->attrs->proto. @p specifies the protocol that called - * rte_update(). In most cases it is the same protocol as @src. rte_update() - * stores @p in @new->sender; - * - * When rte_update() gets any route, it automatically validates it (checks, - * whether the network and next hop address are valid IP addresses and also - * whether a normal routing protocol doesn't try to smuggle a host or link - * scope route to the table), converts all protocol dependent attributes stored - * in the &rte to temporary extended attributes, consults import filters of the - * protocol to see if the route should be accepted and/or its attributes modified, - * stores the temporary attributes back to the &rte. - * - * Now, having a "public" version of the route, we - * automatically find any old route defined by the protocol @src - * for network @n, replace it by the new one (or removing it if @new is %NULL), - * recalculate the optimal route for this destination and finally broadcast - * the change (if any) to all routing protocols by calling rte_announce(). - * - * All memory used for attribute lists and other temporary allocations is taken - * from a special linear pool @rte_update_pool and freed when rte_update() - * finishes. - */ +void rte_update_direct(struct channel *c, const net_addr *n, rte *new, struct rte_src *src); void -rte_update2(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) +rte_update(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) { - // struct proto *p = c->proto; - struct proto_stats *stats = &c->stats; - const struct filter *filter = c->in_filter; - rte *dummy = NULL; - net *nn; + if (!c->in_req.hook) + return; ASSERT(c->channel_state == CS_UP); - rte_update_lock(); + if (c->in_table) + rte_import(&c->in_table->push, n, new, src); + else + rte_update_direct(c, n, new, src); +} + +void +rte_update_direct(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) +{ + const struct filter *filter = c->in_filter; + struct channel_import_stats *stats = &c->import_stats; + + rte_update_lock(c); if (new) { - /* Create a temporary table node */ - nn = alloca(sizeof(net) + n->length); - memset(nn, 0, sizeof(net) + n->length); - net_copy(nn->n.addr, n); - - new->net = nn; - new->sender = c; + new->net = n; - if (!new->pref) - new->pref = c->preference; + int fr; - stats->imp_updates_received++; - if (!rte_validate(new)) + stats->updates_received++; + if (!rte_validate(c, new)) { - rte_trace_in(D_FILTERS, c, new, "invalid"); - stats->imp_updates_invalid++; - goto drop; + channel_rte_trace_in(D_FILTERS, c, new, "invalid"); + stats->updates_invalid++; + new = NULL; } - - if (filter == FILTER_REJECT) + else if ((filter == FILTER_REJECT) || + ((fr = f_run(filter, new, c->rte_update_pool, 0)) > F_ACCEPT)) { - stats->imp_updates_filtered++; - rte_trace_in(D_FILTERS, c, new, "filtered out"); - - if (! c->in_keep_filtered) - goto drop; + stats->updates_filtered++; + channel_rte_trace_in(D_FILTERS, c, new, "filtered out"); - /* new is a private copy, i could modify it */ - new->flags |= REF_FILTERED; + if (c->in_keep_filtered) + new->flags |= REF_FILTERED; + else + new = NULL; } - else if (filter) - { - rta *old_attrs = NULL; - rte_make_tmp_attrs(&new, rte_update_pool, &old_attrs); + } + else + stats->withdraws_received++; - int fr = f_run(filter, &new, rte_update_pool, 0); - if (fr > F_ACCEPT) - { - stats->imp_updates_filtered++; - rte_trace_in(D_FILTERS, c, new, "filtered out"); + rte_import(&c->in_req, n, new, src); - if (! c->in_keep_filtered) - { - rta_free(old_attrs); - goto drop; - } + rte_update_unlock(c); +} - new->flags |= REF_FILTERED; - } +void +rte_import(struct rt_import_request *req, const net_addr *n, rte *new, struct rte_src *src) +{ + struct rt_import_hook *hook = req->hook; + if (!hook) + return; - rte_store_tmp_attrs(new, rte_update_pool, old_attrs); - } - if (!rta_is_cached(new->attrs)) /* Need to copy attributes */ - new->attrs = rta_lookup(new->attrs); - new->flags |= REF_COW; + RT_LOCK(hook->table); + rtable_private *tab = RT_PRIV(hook->table); + net *nn; + if (new) + { /* Use the actual struct network, not the dummy one */ - nn = net_get(c->table, n); - new->net = nn; + nn = net_get(tab, n); + new->net = nn->n.addr; + new->sender = hook; } - else + else if (!(nn = net_find(tab, n))) { - stats->imp_withdraws_received++; - - if (!(nn = net_find(c->table, n)) || !src) - { - stats->imp_withdraws_ignored++; - rte_update_unlock(); - return; - } + req->hook->stats.withdraws_ignored++; + RT_UNLOCK(tab); + return; } - recalc: /* And recalculate the best route */ - rte_hide_dummy_routes(nn, &dummy); - rte_recalculate(c, nn, new, src); - rte_unhide_dummy_routes(nn, &dummy); + rte_recalculate(tab, hook, nn, new, src); + RT_UNLOCK(tab); +} + +/* Check rtable for best route to given net whether it would be exported do p */ +int +rt_examine(rtable_private *t, net_addr *a, struct channel *c, const struct filter *filter) +{ + net *n = net_find(t, a); + + if (!n || !n->routes) + return 0; + + rte rt = n->routes->rte; + + if (!rte_is_valid(&rt)) + return 0; - rte_update_unlock(); - return; + rte_update_lock(c); - drop: - rte_free(new); - new = NULL; - if (nn = net_find(c->table, n)) - goto recalc; + /* Rest is stripped down export_filter() */ + int v = c->proto->preexport ? c->proto->preexport(c, &rt) : 0; + if (v == RIC_PROCESS) + v = (f_run(filter, &rt, c->rte_update_pool, FF_SILENT) <= F_ACCEPT); - rte_update_unlock(); + rte_update_unlock(c); + + return v > 0; } -/* Independent call to rte_announce(), used from next hop - recalculation, outside of rte_update(). new must be non-NULL */ -static inline void -rte_announce_i(rtable *tab, uint type, net *net, rte *new, rte *old, - rte *new_best, rte *old_best) +static void +rt_export_stopped(void *data) { - rte_update_lock(); - rte_announce(tab, type, net, new, old, new_best, old_best); - rte_update_unlock(); + struct rt_export_hook *hook = data; + + RT_LOCKED(hook->table, tab) + { + /* Drop pending exports */ + rt_export_used(tab); + + /* Unlist */ + rem_node(&hook->n); + } + + /* Report the channel as stopped. */ + hook->stopped(hook->req); + + RT_LOCKED(hook->table, tab) + { + /* Free the hook together with its coroutine. */ + rp_free(hook->pool, tab->rp); + rt_unlock_table(tab); + + DBG("Export hook %p in table %s finished uc=%u\n", hook, tab->name, tab->use_count); + } } + static inline void -rte_discard(rte *old) /* Non-filtered route deletion, used during garbage collection */ +rt_set_import_state(struct rt_import_hook *hook, u8 state) { - rte_update_lock(); - rte_recalculate(old->sender, old->net, NULL, old->attrs->src); - rte_update_unlock(); + hook->last_state_change = current_time(); + hook->import_state = state; + + if (hook->req->log_state_change) + hook->req->log_state_change(hook->req, state); } -/* Modify existing route by protocol hook, used for long-lived graceful restart */ static inline void -rte_modify(rte *old) +rt_set_export_state(struct rt_export_hook *hook, u8 state) { - rte_update_lock(); + hook->last_state_change = current_time(); + atomic_store_explicit(&hook->export_state, state, memory_order_release); - rte *new = old->sender->proto->rte_modify(old, rte_update_pool); - if (new != old) - { - if (new) - { - if (!rta_is_cached(new->attrs)) - new->attrs = rta_lookup(new->attrs); - new->flags = (old->flags & ~REF_MODIFY) | REF_COW; - } + if (hook->req->log_state_change) + hook->req->log_state_change(hook->req, state); +} - rte_recalculate(old->sender, old->net, new, old->attrs->src); - } +void +rt_request_import(rtable *t, struct rt_import_request *req) +{ + RT_LOCK(t); + rtable_private *tab = RT_PRIV(t); + rt_lock_table(tab); + + struct rt_import_hook *hook = req->hook = mb_allocz(tab->rp, sizeof(struct rt_import_hook)); + + DBG("Lock table %s for import %p req=%p uc=%u\n", tab->name, hook, req, tab->use_count); + + hook->req = req; + hook->table = t; + + hook->export_announce_event = ev_new_init(tab->rp, rt_import_announce_exports, hook); + + if (!hook->stale_set) + hook->stale_set = hook->stale_valid = hook->stale_pruning = hook->stale_pruned = 1; + + rt_set_import_state(hook, TIS_UP); + + hook->n = (node) {}; + add_tail(&tab->imports, &hook->n); - rte_update_unlock(); + RT_UNLOCK(t); } -/* Check rtable for best route to given net whether it would be exported do p */ -int -rt_examine(rtable *t, net_addr *a, struct proto *p, const struct filter *filter) +void +rt_stop_import(struct rt_import_request *req, event *stopped) { - net *n = net_find(t, a); - rte *rt = n ? n->routes : NULL; + ASSERT_DIE(req->hook); + struct rt_import_hook *hook = req->hook; - if (!rte_is_valid(rt)) - return 0; + rtable_private *tab = RT_LOCK(hook->table); - rte_update_lock(); + rt_schedule_prune(tab); - /* Rest is stripped down export_filter() */ - int v = p->preexport ? p->preexport(p, &rt, rte_update_pool) : 0; - if (v == RIC_PROCESS) - { - rte_make_tmp_attrs(&rt, rte_update_pool, NULL); - v = (f_run(filter, &rt, rte_update_pool, FF_SILENT) <= F_ACCEPT); - } + rt_set_import_state(hook, TIS_STOP); + hook->stopped = stopped; - /* Discard temporary rte */ - if (rt != n->routes) - rte_free(rt); + if (hook->stale_set < hook->stale_valid) + if (!--tab->rr_count) + rt_schedule_notify(tab); - rte_update_unlock(); + RT_UNLOCK(tab); +} - return v > 0; +void +rt_request_export(rtable *t, struct rt_export_request *req) +{ + RT_LOCK(t); + rtable_private *tab = RT_PRIV(t); + rt_lock_table(tab); + + pool *p = rp_new(tab->rp, tab->loop, "Export hook"); + struct rt_export_hook *hook = req->hook = mb_allocz(p, sizeof(struct rt_export_hook)); + hook->pool = p; + + hook->req = req; + hook->table = t; + + /* stats zeroed by mb_allocz */ + + bmap_init(&hook->seq_map, p, 1024); + + rt_set_export_state(hook, TES_HUNGRY); + + hook->n = (node) {}; + add_tail(&tab->exports, &hook->n); + + DBG("New export hook %p req %p in table %s uc=%u\n", hook, req, tab->name, tab->use_count); + + hook->event = ev_new_init(p, rt_feed_channel, hook); + RT_UNLOCK(t); + + rt_send_export_event(hook); } +void +rt_stop_export(struct rt_export_request *req, void (*stopped)(struct rt_export_request *)) +{ + ASSERT_DIE(req->hook); + struct rt_export_hook *hook = req->hook; + + RT_LOCK(hook->table); + rtable_private *tab = RT_PRIV(hook->table); + + /* Stop feeding */ + ev_postpone(hook->event); + + if (atomic_load_explicit(&hook->export_state, memory_order_relaxed) == TES_FEEDING) + fit_get(&tab->fib, &hook->feed_fit); + + hook->event->hook = rt_export_stopped; + hook->stopped = stopped; + + rt_send_export_event(hook); + + RT_UNLOCK(hook->table); + + rt_set_export_state(hook, TES_STOP); +} /** * rt_refresh_begin - start a refresh cycle @@ -1630,21 +1836,48 @@ rt_examine(rtable *t, net_addr *a, struct proto *p, const struct filter *filter) * routes to the routing table (by rte_update()). After that, all protocol * routes (more precisely routes with @c as @sender) not sent during the * refresh cycle but still in the table from the past are pruned. This is - * implemented by marking all related routes as stale by REF_STALE flag in - * rt_refresh_begin(), then marking all related stale routes with REF_DISCARD - * flag in rt_refresh_end() and then removing such routes in the prune loop. - */ + * implemented by setting rte->stale_cycle to req->stale_set in rte_update() + * and then dropping all routes with old stale_cycle values in table prune loop. */ void -rt_refresh_begin(rtable *t, struct channel *c) +rt_refresh_begin(struct rt_import_request *req) { - FIB_WALK(&t->fib, net, n) - { - rte *e; - for (e = n->routes; e; e = e->next) - if (e->sender == c) - e->flags |= REF_STALE; - } - FIB_WALK_END; + struct rt_import_hook *hook = req->hook; + ASSERT_DIE(hook); + + RT_LOCK(hook->table); + rtable_private *tab = RT_PRIV(hook->table); + + ASSERT_DIE(hook->stale_set == hook->stale_valid); + + /* If the pruning routine is too slow */ + if ((hook->stale_pruned < hook->stale_valid) && (hook->stale_pruned + 128 < hook->stale_valid) + || (hook->stale_pruned > hook->stale_valid) && (hook->stale_pruned > hook->stale_valid + 128)) + { + log(L_WARN "Route refresh flood in table %s", tab->name); + FIB_WALK(&tab->fib, net, n) + { + for (struct rte_storage *e = n->routes; e; e = e->next) + if (e->rte.sender == req->hook) + e->rte.stale_cycle = 0; + } + FIB_WALK_END; + hook->stale_set = 1; + hook->stale_valid = 0; + hook->stale_pruned = 0; + } + else if (!++hook->stale_set) + { + /* Let's reserve the stale_cycle zero value for always-invalid routes */ + hook->stale_set = 1; + hook->stale_valid = 0; + } + + tab->rr_count++; + + if (req->trace_routes & D_STATES) + log(L_TRACE "%s: route refresh begin [%u]", req->name, hook->stale_set); + + RT_UNLOCK(tab); } /** @@ -1656,45 +1889,24 @@ rt_refresh_begin(rtable *t, struct channel *c) * hook. See rt_refresh_begin() for description of refresh cycles. */ void -rt_refresh_end(rtable *t, struct channel *c) +rt_refresh_end(struct rt_import_request *req) { - int prune = 0; + struct rt_import_hook *hook = req->hook; + ASSERT_DIE(hook); - FIB_WALK(&t->fib, net, n) - { - rte *e; - for (e = n->routes; e; e = e->next) - if ((e->sender == c) && (e->flags & REF_STALE)) - { - e->flags |= REF_DISCARD; - prune = 1; - } - } - FIB_WALK_END; + rtable_private *tab = RT_LOCK(hook->table); + hook->stale_valid++; + ASSERT_DIE(hook->stale_set == hook->stale_valid); - if (prune) - rt_schedule_prune(t); -} + rt_schedule_prune(tab); -void -rt_modify_stale(rtable *t, struct channel *c) -{ - int prune = 0; + if (req->trace_routes & D_STATES) + log(L_TRACE "%s: route refresh end [%u]", req->name, hook->stale_valid); - FIB_WALK(&t->fib, net, n) - { - rte *e; - for (e = n->routes; e; e = e->next) - if ((e->sender == c) && (e->flags & REF_STALE) && !(e->flags & REF_FILTERED)) - { - e->flags |= REF_MODIFY; - prune = 1; - } - } - FIB_WALK_END; + if (!--tab->rr_count) + rt_schedule_notify(tab); - if (prune) - rt_schedule_prune(t); + RT_UNLOCK(tab); } /** @@ -1704,14 +1916,11 @@ rt_modify_stale(rtable *t, struct channel *c) * This functions dumps contents of a &rte to debug output. */ void -rte_dump(rte *e) -{ - net *n = e->net; - debug("%-1N ", n->n.addr); - debug("PF=%02x pref=%d ", e->pflags, e->pref); - rta_dump(e->attrs); - if (e->attrs->src->proto->proto->dump_attrs) - e->attrs->src->proto->proto->dump_attrs(e); +rte_dump(struct rte_storage *e) +{ + debug("%-1N ", e->rte.net); + debug("PF=%02x ", e->rte.pflags); + rta_dump(e->rte.attrs); debug("\n"); } @@ -1722,20 +1931,22 @@ rte_dump(rte *e) * This function dumps contents of a given routing table to debug output. */ void -rt_dump(rtable *t) +rt_dump(rtable *tab) { - debug("Dump of routing table <%s>\n", t->name); + RT_LOCK(tab); + rtable_private *t = RT_PRIV(tab); + debug("Dump of routing table <%s>%s\n", t->name, t->delete ? " (deleted)" : ""); #ifdef DEBUGGING fib_check(&t->fib); #endif FIB_WALK(&t->fib, net, n) { - rte *e; - for(e=n->routes; e; e=e->next) + for(struct rte_storage *e=n->routes; e; e=e->next) rte_dump(e); } FIB_WALK_END; debug("\n"); + RT_UNLOCK(tab); } /** @@ -1753,73 +1964,105 @@ rt_dump_all(void) rt_dump(t); } -static inline void -rt_schedule_hcu(rtable *tab) +void +rt_dump_hooks(rtable *t) +{ + RT_LOCK(t); + rtable_private *tab = RT_PRIV(t); + debug("Dump of hooks in routing table <%s>%s\n", tab->name, tab->delete ? " (deleted)" : ""); + debug(" nhu_state=%u hcu_scheduled=%u use_count=%d rt_count=%u\n", + atomic_load(&tab->nhu_state), ev_active(tab->hcu_event), tab->use_count, tab->rt_count); + debug(" last_rt_change=%t gc_time=%t gc_counter=%d prune_state=%u\n", + tab->last_rt_change, tab->gc_time, tab->gc_counter, tab->prune_state); + + struct rt_import_hook *ih; + WALK_LIST(ih, tab->imports) + { + ih->req->dump_req(ih->req); + debug(" Import hook %p requested by %p: pref=%u" + " last_state_change=%t import_state=%u stopped=%p\n", + ih, ih->req, ih->stats.pref, + ih->last_state_change, ih->import_state, ih->stopped); + } + + struct rt_export_hook *eh; + WALK_LIST(eh, tab->exports) + { + eh->req->dump_req(eh->req); + debug(" Export hook %p requested by %p:" + " refeed_pending=%u last_state_change=%t export_state=%u\n", + eh, eh->req, eh->refeed_pending, eh->last_state_change, atomic_load_explicit(&eh->export_state, memory_order_relaxed)); + } + debug("\n"); + RT_UNLOCK(t); +} + +void +rt_dump_hooks_all(void) { - if (tab->hcu_scheduled) - return; + rtable *t; + node *n; + + debug("Dump of all table hooks\n"); - tab->hcu_scheduled = 1; - ev_schedule(tab->rt_event); + WALK_LIST2(t, n, routing_tables, n) + rt_dump_hooks(t); } static inline void rt_schedule_nhu(rtable *tab) { - if (tab->nhu_state == NHU_CLEAN) - ev_schedule(tab->rt_event); + atomic_fetch_or_explicit(&tab->nhu_state, NHU_SCHEDULED, memory_order_acq_rel); + ev_send_loop(tab->loop, tab->nhu_event); /* state change: * NHU_CLEAN -> NHU_SCHEDULED * NHU_RUNNING -> NHU_DIRTY */ - tab->nhu_state |= NHU_SCHEDULED; } void -rt_schedule_prune(rtable *tab) +rt_schedule_prune(rtable_private *tab) { if (tab->prune_state == 0) - ev_schedule(tab->rt_event); + ev_send_loop(tab->loop, tab->prune_event); /* state change 0->1, 2->3 */ tab->prune_state |= 1; } - -static void -rt_event(void *ptr) +void +rt_export_used(rtable_private *tab) { - rtable *tab = ptr; - - rt_lock_table(tab); + if (config->table_debug) + log(L_TRACE "%s: Export cleanup requested", tab->name); - if (tab->hcu_scheduled) - rt_update_hostcache(tab); - - if (tab->nhu_state) - rt_next_hop_update(tab); - - if (tab->prune_state) - rt_prune_table(tab); - - rt_unlock_table(tab); + ev_send_loop(tab->loop, tab->ec_event); } - static inline btime -rt_settled_time(rtable *tab) +rt_settled_time(rtable_private *tab) { ASSUME(tab->base_settle_time != 0); - return MIN(tab->last_rt_change + tab->config->min_settle_time, - tab->base_settle_time + tab->config->max_settle_time); + btime min_settle_time = tab->rr_count ? tab->config->min_rr_settle_time : tab->config->min_settle_time; + btime max_settle_time = tab->rr_count ? tab->config->max_rr_settle_time : tab->config->max_settle_time; + + DBG("settled time computed from %t %t %t %t as %t / %t, now is %t\n", + tab->name, tab->last_rt_change, min_settle_time, + tab->base_settle_time, max_settle_time, + tab->last_rt_change + min_settle_time, + tab->base_settle_time + max_settle_time, current_time()); + + return MIN(tab->last_rt_change + min_settle_time, + tab->base_settle_time + max_settle_time); } static void rt_settle_timer(timer *t) { - rtable *tab = t->data; + rtable_private *tab = t->data; + ASSERT_DIE(birdloop_inside(tab->loop)); if (!tab->base_settle_time) return; @@ -1827,7 +2070,7 @@ rt_settle_timer(timer *t) btime settled_time = rt_settled_time(tab); if (current_time() < settled_time) { - tm_set(tab->settle_timer, settled_time); + tm_set_in(tab->settle_timer, settled_time, tab->loop); return; } @@ -1836,11 +2079,11 @@ rt_settle_timer(timer *t) struct rt_subscription *s; WALK_LIST(s, tab->subscribers) - s->hook(s); + ev_send(s->event->list, s->event); } static void -rt_kick_settle_timer(rtable *tab) +rt_kick_settle_timer(rtable_private *tab) { tab->base_settle_time = current_time(); @@ -1848,11 +2091,11 @@ rt_kick_settle_timer(rtable *tab) tab->settle_timer = tm_new_init(tab->rp, rt_settle_timer, tab, 0, 0); if (!tm_active(tab->settle_timer)) - tm_set(tab->settle_timer, rt_settled_time(tab)); + tm_set_in(tab->settle_timer, rt_settled_time(tab), tab->loop); } static inline void -rt_schedule_notify(rtable *tab) +rt_schedule_notify(rtable_private *tab) { if (EMPTY_LIST(tab->subscribers)) return; @@ -1864,33 +2107,40 @@ rt_schedule_notify(rtable *tab) } void -rt_subscribe(rtable *tab, struct rt_subscription *s) +rt_subscribe(rtable *t, struct rt_subscription *s) { - s->tab = tab; - rt_lock_table(tab); - add_tail(&tab->subscribers, &s->n); + s->tab = t; + RT_LOCKED(t, tab) + { + rt_lock_table(tab); + DBG("rt_subscribe(%s)\n", tab->name); + add_tail(&tab->subscribers, &s->n); + } } void rt_unsubscribe(struct rt_subscription *s) { - rem_node(&s->n); - rt_unlock_table(s->tab); + RT_LOCKED(s->tab, tab) + { + rem_node(&s->n); + if (EMPTY_LIST(tab->subscribers) && tm_active(tab->settle_timer)) + tm_stop(tab->settle_timer); + rt_unlock_table(tab); + } } static void rt_free(resource *_r) { - rtable *r = (rtable *) _r; + rtable_private *r = (rtable_private *) _r; DBG("Deleting routing table %s\n", r->name); ASSERT_DIE(r->use_count == 0); - - if (r->internal) - return; - - r->config->table = NULL; - rem_node(&r->n); + ASSERT_DIE(r->rt_count == 0); + ASSERT_DIE(!r->cork_active); + ASSERT_DIE(EMPTY_LIST(r->imports)); + ASSERT_DIE(EMPTY_LIST(r->exports)); if (r->hostcache) rt_free_hostcache(r); @@ -1907,14 +2157,14 @@ rt_free(resource *_r) static void rt_res_dump(resource *_r) { - rtable *r = (rtable *) _r; + rtable_private *r = RT_PRIV((rtable *) _r); debug("name \"%s\", addr_type=%s, rt_count=%u, use_count=%d\n", r->name, net_label[r->addr_type], r->rt_count, r->use_count); } static struct resclass rt_class = { .name = "Routing table", - .size = sizeof(struct rtable), + .size = sizeof(rtable_private), .free = rt_free, .dump = rt_res_dump, .lookup = NULL, @@ -1928,11 +2178,16 @@ rt_setup(pool *pp, struct rtable_config *cf) void *nb = mb_alloc(pp, ns); ASSERT_DIE(ns - 1 == bsnprintf(nb, ns, "Routing table %s", cf->name)); - pool *p = rp_new(pp, nb); - mb_move(nb, p); + struct birdloop *l = birdloop_new(pp, DOMAIN_ORDER(rtable), nb); + pool *p = birdloop_pool(l); - rtable *t = ralloc(p, &rt_class); + birdloop_enter(l); + + rtable_private *t = ralloc(p, &rt_class); t->rp = p; + t->loop = l; + + t->rte_slab = sl_new(p, sizeof(struct rte_storage)); t->name = cf->name; t->config = cf; @@ -1940,19 +2195,35 @@ rt_setup(pool *pp, struct rtable_config *cf) fib_init(&t->fib, p, t->addr_type, sizeof(net), OFFSETOF(net, n), 0, NULL); - if (!(t->internal = cf->internal)) - { - init_list(&t->channels); - hmap_init(&t->id_map, p, 1024); - hmap_set(&t->id_map, 0); + init_list(&t->imports); + init_list(&t->exports); - init_list(&t->subscribers); + hmap_init(&t->id_map, p, 1024); + hmap_set(&t->id_map, 0); - t->rt_event = ev_new_init(p, rt_event, t); - t->last_rt_change = t->gc_time = current_time(); - } + init_list(&t->pending_exports); + init_list(&t->subscribers); + + t->announce_event = ev_new_init(p, rt_announce_exports, t); + t->ec_event = ev_new_init(p, rt_export_cleanup, t); + t->prune_event = ev_new_init(p, rt_prune_table, t); + t->hcu_event = ev_new_init(p, rt_update_hostcache, t); + t->nhu_event = ev_new_init(p, rt_next_hop_update, t); + + t->nhu_event->cork = &rt_cork; + t->prune_event->cork = &rt_cork; + + t->last_rt_change = t->gc_time = current_time(); + t->next_export_seq = 1; - return t; + t->rl_pipe = (struct tbf) TBF_DEFAULT_LOG_LIMITS; + + t->nhu_lp = lp_new_default(p); + + mb_move(nb, p); + birdloop_leave(l); + + return (rtable *) t; } /** @@ -1965,13 +2236,11 @@ void rt_init(void) { rta_init(); - rt_table_pool = rp_new(&root_pool, "Routing tables"); - rte_update_pool = lp_new_default(rt_table_pool); - rte_slab = sl_new(rt_table_pool, sizeof(rte)); + rt_table_pool = rp_new(&root_pool, &main_birdloop, "Routing tables"); init_list(&routing_tables); + ev_init_cork(&rt_cork, "Route Table Cork"); } - /** * rt_prune_table - prune a routing table * @@ -1987,12 +2256,15 @@ rt_init(void) * iteration. */ static void -rt_prune_table(rtable *tab) +rt_prune_table(void *data) { + rtable_private *tab = data; + ASSERT_DIE(birdloop_inside(tab->loop)); + struct fib_iterator *fit = &tab->prune_fit; int limit = 512; - struct channel *c; + struct rt_import_hook *ih; node *n, *x; DBG("Pruning route table %s\n", tab->name); @@ -2003,12 +2275,21 @@ rt_prune_table(rtable *tab) if (tab->prune_state == 0) return; + rt_lock_table(tab); + if (tab->prune_state == 1) { /* Mark channels to flush */ - WALK_LIST2(c, n, tab->channels, table_node) - if (c->channel_state == CS_FLUSHING) - c->flush_active = 1; + WALK_LIST2(ih, n, tab->imports, n) + if (ih->import_state == TIS_STOP) + rt_set_import_state(ih, TIS_FLUSHING); + else if ((ih->stale_valid != ih->stale_pruning) && (ih->stale_pruning == ih->stale_pruned)) + { + ih->stale_pruning = ih->stale_valid; + + if (ih->req->trace_routes & D_STATES) + log(L_TRACE "%s: table prune after refresh begin [%u]", ih->req->name, ih->stale_pruning); + } FIB_ITERATE_INIT(fit, &tab->fib); tab->prune_state = 2; @@ -2017,43 +2298,31 @@ rt_prune_table(rtable *tab) again: FIB_ITERATE_START(&tab->fib, fit, net, n) { - rte *e; - rescan: - for (e=n->routes; e; e=e->next) + for (struct rte_storage *e=n->routes; e; e=e->next) { - if (e->sender->flush_active || (e->flags & REF_DISCARD)) - { - if (limit <= 0) - { - FIB_ITERATE_PUT(fit); - ev_schedule(tab->rt_event); - return; - } - - rte_discard(e); - limit--; + struct rt_import_hook *s = e->rte.sender; - goto rescan; - } - - if (e->flags & REF_MODIFY) + if ((s->import_state == TIS_FLUSHING) || + (e->rte.stale_cycle < s->stale_valid) || + (e->rte.stale_cycle > s->stale_set)) { if (limit <= 0) { FIB_ITERATE_PUT(fit); - ev_schedule(tab->rt_event); + ev_send_loop(tab->loop, tab->prune_event); + rt_unlock_table(tab); return; } - rte_modify(e); + rte_recalculate(tab, e->rte.sender, n, NULL, e->rte.src); limit--; goto rescan; } } - if (!n->routes) /* Orphaned FIB entry */ + if (!n->routes && !n->first) /* Orphaned FIB entry */ { FIB_ITERATE_PUT(fit); fib_delete(&tab->fib, n); @@ -2070,23 +2339,202 @@ again: tab->gc_time = current_time(); /* state change 2->0, 3->1 */ - tab->prune_state &= 1; - - if (tab->prune_state > 0) - ev_schedule(tab->rt_event); + if (tab->prune_state &= 1) + ev_send_loop(tab->loop, tab->prune_event); - /* FIXME: This should be handled in a better way */ - rt_prune_sources(); + uint flushed_channels = 0; /* Close flushed channels */ - WALK_LIST2_DELSAFE(c, n, x, tab->channels, table_node) - if (c->flush_active) + WALK_LIST2_DELSAFE(ih, n, x, tab->imports, n) + if (ih->import_state == TIS_FLUSHING) + { + ih->flush_seq = tab->next_export_seq; + rt_set_import_state(ih, TIS_WAITING); + flushed_channels++; + } + else if (ih->stale_pruning != ih->stale_pruned) + { + ih->stale_pruned = ih->stale_pruning; + + if (ih->req->trace_routes & D_STATES) + log(L_TRACE "%s: table prune after refresh end [%u]", ih->req->name, ih->stale_pruned); + } + + /* In some cases, we may want to directly proceed to export cleanup */ + if (EMPTY_LIST(tab->exports) && flushed_channels) + rt_export_cleanup(tab); + + rt_unlock_table(tab); +} + +static void +rt_export_cleanup(void *data) +{ + rtable_private *tab = data; + ASSERT_DIE(birdloop_inside(tab->loop)); + + u64 min_seq = ~((u64) 0); + struct rt_pending_export *last_export_to_free = NULL; + struct rt_pending_export *first_export = tab->first_export; + + struct rt_export_hook *eh; + node *n; + WALK_LIST2(eh, n, tab->exports, n) + { + switch (atomic_load_explicit(&eh->export_state, memory_order_acquire)) + { + case TES_DOWN: + case TES_HUNGRY: + continue; + + case TES_READY: + { + struct rt_pending_export *last = atomic_load_explicit(&eh->last_export, memory_order_acquire); + if (!last) + /* No last export means that the channel has exported nothing since last cleanup */ + goto done; + + else if (min_seq > last->seq) + { + min_seq = last->seq; + last_export_to_free = last; + } + continue; + } + + default: + /* It's only safe to cleanup when the export state is idle or regular. No feeding or stopping allowed. */ + goto done; + } + } + + tab->first_export = last_export_to_free ? rt_next_export_fast(last_export_to_free) : NULL; + + if (config->table_debug) + log(L_TRACE "%s: Export cleanup, old first_export seq %lu, new %lu, min_seq %ld", + tab->name, + first_export ? first_export->seq : 0, + tab->first_export ? tab->first_export->seq : 0, + min_seq); + + WALK_LIST2(eh, n, tab->exports, n) + { + if (atomic_load_explicit(&eh->export_state, memory_order_acquire) != TES_READY) + continue; + + struct rt_pending_export *last = atomic_load_explicit(&eh->last_export, memory_order_acquire); + if (last == last_export_to_free) + { + /* This may fail when the channel managed to export more inbetween. This is OK. */ + atomic_compare_exchange_strong_explicit( + &eh->last_export, &last, NULL, + memory_order_release, + memory_order_relaxed); + + DBG("store hook=%p last_export=NULL\n", eh); + } + } + + while (first_export && (first_export->seq <= min_seq)) + { + ASSERT_DIE(first_export->new || first_export->old); + + const net_addr *n = first_export->new ? + first_export->new->rte.net : + first_export->old->rte.net; + net *net = SKIP_BACK(struct network, n.addr, (net_addr (*)[0]) n); + + ASSERT_DIE(net->first == first_export); + + if (first_export == net->last) + /* The only export here */ + net->last = net->first = NULL; + else + /* First is now the next one */ + net->first = atomic_load_explicit(&first_export->next, memory_order_relaxed); + + /* For now, the old route may be finally freed */ + if (first_export->old) + { + rt_rte_trace_in(D_ROUTES, first_export->old->rte.sender->req, &first_export->old->rte, "freed"); + hmap_clear(&tab->id_map, first_export->old->rte.id); + rte_free(first_export->old, tab); + } + +#ifdef LOCAL_DEBUG + memset(first_export, 0xbd, sizeof(struct rt_pending_export)); +#endif + + struct rt_export_block *reb = HEAD(tab->pending_exports); + ASSERT_DIE(reb == PAGE_HEAD(first_export)); + + u32 pos = (first_export - &reb->export[0]); + u32 end = atomic_load_explicit(&reb->end, memory_order_relaxed); + ASSERT_DIE(pos < end); + + struct rt_pending_export *next = NULL; + + if (++pos < end) + next = &reb->export[pos]; + else + { + rem_node(&reb->n); + +#ifdef LOCAL_DEBUG + memset(reb, 0xbe, page_size); +#endif + + free_page(reb); + + if (EMPTY_LIST(tab->pending_exports)) { - c->flush_active = 0; - channel_set_state(c, CS_DOWN); + if (config->table_debug) + log(L_TRACE "%s: Resetting export seq", tab->name); + + node *n; + WALK_LIST2(eh, n, tab->exports, n) + { + if (atomic_load_explicit(&eh->export_state, memory_order_acquire) != TES_READY) + continue; + + ASSERT_DIE(atomic_load_explicit(&eh->last_export, memory_order_acquire) == NULL); + bmap_reset(&eh->seq_map, 1024); + } + + tab->next_export_seq = 1; + } + else + { + reb = HEAD(tab->pending_exports); + next = &reb->export[0]; + } + } + + first_export = next; + } + +done:; + struct rt_import_hook *ih; node *x; + WALK_LIST2_DELSAFE(ih, n, x, tab->imports, n) + if (ih->import_state == TIS_WAITING) + if (!first_export || (first_export->seq >= ih->flush_seq)) + { + ih->import_state = TIS_CLEARED; + ev_send(ih->req->list, ih->export_announce_event); } - return; + if (EMPTY_LIST(tab->pending_exports) && ev_active(tab->announce_event)) + ev_postpone(tab->announce_event); + + /* If reduced to at most one export block pending */ + if (tab->cork_active && + ((!tab->first_export) || (tab->first_export->seq + 128 > tab->next_export_seq))) + { + tab->cork_active = 0; + ev_uncork(&rt_cork); + if (config->table_debug) + log(L_TRACE "%s: cork released", tab->name); + } } void @@ -2120,7 +2568,7 @@ rta_next_hop_outdated(rta *a) } void -rta_apply_hostentry(rta *a, struct hostentry *he, mpls_label_stack *mls) +rta_apply_hostentry(rta *a, struct hostentry *he, mpls_label_stack *mls, linpool *lp) { a->hostentry = he; a->dest = he->dest; @@ -2155,7 +2603,7 @@ no_nexthop: else { nhr = nhp; - nhp = (nhp ? (nhp->next = lp_alloc(rte_update_pool, NEXTHOP_MAX_SIZE)) : &(a->nh)); + nhp = (nhp ? (nhp->next = lp_alloc(lp, NEXTHOP_MAX_SIZE)) : &(a->nh)); } memset(nhp, 0, NEXTHOP_MAX_SIZE); @@ -2210,8 +2658,8 @@ no_nexthop: } } -static inline rte * -rt_next_hop_update_rte(rtable *tab UNUSED, rte *old) +static inline struct rte_storage * +rt_next_hop_update_rte(rtable_private *tab, net *n, rte *old) { rta *a = alloca(RTA_MAX_SIZE); memcpy(a, old->attrs, rta_size(old->attrs)); @@ -2219,58 +2667,71 @@ rt_next_hop_update_rte(rtable *tab UNUSED, rte *old) mpls_label_stack mls = { .len = a->nh.labels_orig }; memcpy(mls.stack, &a->nh.label[a->nh.labels - mls.len], mls.len * sizeof(u32)); - rta_apply_hostentry(a, old->attrs->hostentry, &mls); - a->aflags = 0; + rta_apply_hostentry(a, old->attrs->hostentry, &mls, tab->nhu_lp); + a->cached = 0; - rte *e = sl_alloc(rte_slab); - memcpy(e, old, sizeof(rte)); - e->attrs = rta_lookup(a); + rte e0 = *old; + e0.attrs = a; - return e; + return rte_store(&e0, n, tab); } static inline int -rt_next_hop_update_net(rtable *tab, net *n) +rt_next_hop_update_net(rtable_private *tab, net *n) { - rte **k, *e, *new, *old_best, **new_best; + struct rte_storage *new; int count = 0; - int free_old_best = 0; - old_best = n->routes; + struct rte_storage *old_best = n->routes; if (!old_best) return 0; - for (k = &n->routes; e = *k; k = &e->next) - if (rta_next_hop_outdated(e->attrs)) - { - new = rt_next_hop_update_rte(tab, e); - *k = new; + for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) + if (rta_next_hop_outdated(e->rte.attrs)) + count++; + + if (!count) + return 0; + + struct rte_multiupdate { + struct rte_storage *old, *new; + } *updates = alloca(sizeof(struct rte_multiupdate) * count); - rte_trace_in(D_ROUTES, new->sender, new, "updated"); - rte_announce_i(tab, RA_ANY, n, new, e, NULL, NULL); + int pos = 0; + for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) + if (rta_next_hop_outdated(e->rte.attrs)) + { + struct rte_storage *new = rt_next_hop_update_rte(tab, n, &e->rte); /* Call a pre-comparison hook */ /* Not really an efficient way to compute this */ - if (e->attrs->src->proto->rte_recalculate) - e->attrs->src->proto->rte_recalculate(tab, n, new, e, NULL); + if (e->rte.src->owner->rte_recalculate) + e->rte.src->owner->rte_recalculate(tab, n, &new->rte, &e->rte, &old_best->rte); + + updates[pos++] = (struct rte_multiupdate) { + .old = e, + .new = new, + }; - if (e != old_best) - rte_free_quick(e); - else /* Freeing of the old best rte is postponed */ - free_old_best = 1; + /* Replace the route in the list */ + new->next = e->next; + *k = e = new; - e = new; - count++; + /* Get a new ID for the route */ + new->rte.lastmod = current_time(); + new->rte.id = hmap_first_zero(&tab->id_map); + hmap_set(&tab->id_map, new->rte.id); + + lp_flush(tab->nhu_lp); } - if (!count) - return 0; + ASSERT_DIE(pos == count); /* Find the new best route */ - new_best = NULL; - for (k = &n->routes; e = *k; k = &e->next) + struct rte_storage **new_best = NULL; + for (struct rte_storage *e, **k = &n->routes; e = *k; k = &e->next) { - if (!new_best || rte_better(e, *new_best)) + if (!new_best || rte_better(&e->rte, &(*new_best)->rte)) new_best = k; } @@ -2283,32 +2744,39 @@ rt_next_hop_update_net(rtable *tab, net *n) n->routes = new; } - /* Announce the new best route */ - if (new != old_best) - rte_trace_in(D_ROUTES, new->sender, new, "updated [best]"); - - /* Propagate changes */ - rte_announce_i(tab, RA_UNDEF, n, NULL, NULL, n->routes, old_best); - - if (free_old_best) - rte_free_quick(old_best); + /* Announce the changes */ + for (int i=0; i<count; i++) + { + _Bool nb = (new == updates[i].new), ob = (old_best == updates[i].old); + const char *best_indicator[2][2] = { + { "autoupdated", "autoupdated [-best]" }, + { "autoupdated [+best]", "autoupdated [best]" } + }; + rt_rte_trace_in(D_ROUTES, updates[i].new->rte.sender->req, &updates[i].new->rte, best_indicator[nb][ob]); + rte_announce(tab, n, updates[i].new, updates[i].old, new, old_best); + } return count; } static void -rt_next_hop_update(rtable *tab) +rt_next_hop_update(void *data) { + rtable_private *tab = data; + ASSERT_DIE(birdloop_inside(tab->loop)); + struct fib_iterator *fit = &tab->nhu_fit; int max_feed = 32; - if (tab->nhu_state == NHU_CLEAN) + if (atomic_load_explicit(&tab->nhu_state, memory_order_acquire) == NHU_CLEAN) return; - if (tab->nhu_state == NHU_SCHEDULED) + rt_lock_table(tab); + + if (atomic_load_explicit(&tab->nhu_state, memory_order_acquire) == NHU_SCHEDULED) { FIB_ITERATE_INIT(fit, &tab->fib); - tab->nhu_state = NHU_RUNNING; + ASSERT_DIE(atomic_exchange_explicit(&tab->nhu_state, NHU_RUNNING, memory_order_acq_rel) == NHU_SCHEDULED); } FIB_ITERATE_START(&tab->fib, fit, net, n) @@ -2316,7 +2784,8 @@ rt_next_hop_update(rtable *tab) if (max_feed <= 0) { FIB_ITERATE_PUT(fit); - ev_schedule(tab->rt_event); + ev_send_loop(tab->loop, tab->nhu_event); + rt_unlock_table(tab); return; } max_feed -= rt_next_hop_update_net(tab, n); @@ -2327,10 +2796,12 @@ rt_next_hop_update(rtable *tab) * NHU_DIRTY -> NHU_SCHEDULED * NHU_RUNNING -> NHU_CLEAN */ - tab->nhu_state &= 1; + if (atomic_fetch_and_explicit(&tab->nhu_state, NHU_SCHEDULED, memory_order_acq_rel) != NHU_RUNNING) + ev_send_loop(tab->loop, tab->nhu_event); - if (tab->nhu_state != NHU_CLEAN) - ev_schedule(tab->rt_event); + ev_send_loop(tab->loop, tab->announce_event); + + rt_unlock_table(tab); } @@ -2352,6 +2823,10 @@ rt_new_table(struct symbol *s, uint addr_type) c->gc_min_time = 5; c->min_settle_time = 1 S; c->max_settle_time = 20 S; + c->min_rr_settle_time = 30 S; + c->max_rr_settle_time = 90 S; + c->cork_limit = 4 * page_size / sizeof(struct rt_pending_export); + c->owner = new_config; add_tail(&new_config->tables, &c->n); @@ -2371,7 +2846,7 @@ rt_new_table(struct symbol *s, uint addr_type) * configuration. */ void -rt_lock_table(rtable *r) +rt_lock_table(rtable_private *r) { r->use_count++; } @@ -2385,16 +2860,11 @@ rt_lock_table(rtable *r) * for deletion by configuration changes. */ void -rt_unlock_table(rtable *r) +rt_unlock_table(rtable_private *r) { - if (!--r->use_count && r->deleted) - { - struct config *conf = r->deleted; - - /* Delete the routing table by freeing its pool */ - rt_shutdown(r); - config_del_obstacle(conf); - } + if (!--r->use_count && r->delete && + !r->prune_state && !atomic_load_explicit(&r->nhu_state, memory_order_acquire)) + birdloop_stop_self(r->loop, r->delete, r); } static struct rtable_config * @@ -2404,6 +2874,21 @@ rt_find_table_config(struct config *cf, char *name) return (sym && (sym->class == SYM_TABLE)) ? sym->table : NULL; } +static void +rt_done(void *data) +{ + RT_LOCKED((rtable *) data, t) + { + struct rtable_config *tc = t->config; + struct config *c = tc->owner; + + tc->table = NULL; + rem_node(&t->n); + + config_del_obstacle(c); + } +} + /** * rt_commit - commit new routing table configuration * @new: new configuration @@ -2426,14 +2911,15 @@ rt_commit(struct config *new, struct config *old) { WALK_LIST(o, old->tables) { - rtable *ot = o->table; - if (!ot->deleted) + RT_LOCK(o->table); + rtable_private *ot = RT_PRIV(o->table); + if (!ot->delete) { r = rt_find_table_config(new, o->name); if (r && (r->addr_type == o->addr_type) && !new->shutdown) { DBG("\t%s: same\n", o->name); - r->table = ot; + r->table = (rtable *) ot; ot->name = r->name; ot->config = r; if (o->sorted != r->sorted) @@ -2442,12 +2928,13 @@ rt_commit(struct config *new, struct config *old) else { DBG("\t%s: deleted\n", o->name); - ot->deleted = old; - config_add_obstacle(old); rt_lock_table(ot); + ot->delete = rt_done; + config_add_obstacle(old); rt_unlock_table(ot); } } + RT_UNLOCK(o->table); } } @@ -2461,19 +2948,6 @@ rt_commit(struct config *new, struct config *old) DBG("\tdone\n"); } -static inline void -do_feed_channel(struct channel *c, net *n, rte *e) -{ - rte_update_lock(); - if (c->ra_mode == RA_ACCEPTED) - rt_notify_accepted(c, n, NULL, NULL, c->refeeding); - else if (c->ra_mode == RA_MERGED) - rt_notify_merged(c, n, NULL, NULL, e, e, c->refeeding); - else /* RA_BASIC */ - rt_notify_basic(c, n, e, e, c->refeeding); - rte_update_unlock(); -} - /** * rt_feed_channel - advertise all routes to a channel * @c: channel to be fed @@ -2483,370 +2957,120 @@ do_feed_channel(struct channel *c, net *n, rte *e) * has something to do. (We avoid transferring all the routes in single pass in * order not to monopolize CPU time.) */ -int -rt_feed_channel(struct channel *c) +static void +rt_feed_channel(void *data) { + struct rt_export_hook *c = data; + struct fib_iterator *fit = &c->feed_fit; int max_feed = 256; - ASSERT(c->export_state == ES_FEEDING); - - if (!c->feed_active) - { - FIB_ITERATE_INIT(fit, &c->table->fib); - c->feed_active = 1; - } - - FIB_ITERATE_START(&c->table->fib, fit, net, n) - { - rte *e = n->routes; - if (max_feed <= 0) - { - FIB_ITERATE_PUT(fit); - return 0; - } - - if ((c->ra_mode == RA_OPTIMAL) || - (c->ra_mode == RA_ACCEPTED) || - (c->ra_mode == RA_MERGED)) - if (rte_is_valid(e)) - { - /* In the meantime, the protocol may fell down */ - if (c->export_state != ES_FEEDING) - goto done; - - do_feed_channel(c, n, e); - max_feed--; - } - - if (c->ra_mode == RA_ANY) - for(e = n->routes; e; e = e->next) - { - /* In the meantime, the protocol may fell down */ - if (c->export_state != ES_FEEDING) - goto done; - - if (!rte_is_valid(e)) - continue; - - do_feed_channel(c, n, e); - max_feed--; - } - } - FIB_ITERATE_END; - -done: - c->feed_active = 0; - return 1; -} - -/** - * rt_feed_baby_abort - abort protocol feeding - * @c: channel - * - * This function is called by the protocol code when the protocol stops or - * ceases to exist during the feeding. - */ -void -rt_feed_channel_abort(struct channel *c) -{ - if (c->feed_active) - { - /* Unlink the iterator */ - fit_get(&c->table->fib, &c->feed_fit); - c->feed_active = 0; - } -} - - -/* - * Import table - */ - -int -rte_update_in(struct channel *c, const net_addr *n, rte *new, struct rte_src *src) -{ - struct rtable *tab = c->in_table; - rte *old, **pos; - net *net; - - if (new) + rtable_private *tab; + if (c->export_state == TES_HUNGRY) { - net = net_get(tab, n); + rt_set_export_state(c, TES_FEEDING); + + tab = RT_LOCK(c->table); - if (!new->pref) - new->pref = c->preference; + struct rt_pending_export *rpe = rt_last_export(tab); + DBG("store hook=%p last_export=%p seq=%lu\n", c, rpe, rpe ? rpe->seq : 0); + atomic_store_explicit(&c->last_export, rpe, memory_order_relaxed); - if (!rta_is_cached(new->attrs)) - new->attrs = rta_lookup(new->attrs); + FIB_ITERATE_INIT(&c->feed_fit, &tab->fib); } else - { - net = net_find(tab, n); + tab = RT_LOCK(c->table); - if (!net) - goto drop_withdraw; - } + ASSERT_DIE(c->export_state == TES_FEEDING); - /* Find the old rte */ - for (pos = &net->routes; old = *pos; pos = &old->next) - if (old->attrs->src == src) +redo: + FIB_ITERATE_START(&tab->fib, fit, net, n) { - if (new && rte_same(old, new)) - { - /* Refresh the old rte, continue with update to main rtable */ - if (old->flags & (REF_STALE | REF_DISCARD | REF_MODIFY)) + if (max_feed <= 0) { - old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY); - return 1; - } - - goto drop_update; - } - - /* Move iterator if needed */ - if (old == c->reload_next_rte) - c->reload_next_rte = old->next; - - /* Remove the old rte */ - *pos = old->next; - rte_free_quick(old); - tab->rt_count--; - - break; - } - - if (!new) - { - if (!old) - goto drop_withdraw; - - if (!net->routes) - fib_delete(&tab->fib, net); - - return 1; - } - - struct channel_limit *l = &c->rx_limit; - if (l->action && !old) - { - if (tab->rt_count >= l->limit) - channel_notify_limit(c, l, PLD_RX, tab->rt_count); - - if (l->state == PLS_BLOCKED) - { - /* Required by rte_trace_in() */ - new->net = net; - - rte_trace_in(D_FILTERS, c, new, "ignored [limit]"); - goto drop_update; - } - } - - /* Insert the new rte */ - rte *e = rte_do_cow(new); - e->flags |= REF_COW; - e->net = net; - e->sender = c; - e->lastmod = current_time(); - e->next = *pos; - *pos = e; - tab->rt_count++; - return 1; - -drop_update: - c->stats.imp_updates_received++; - c->stats.imp_updates_ignored++; - rte_free(new); - - if (!net->routes) - fib_delete(&tab->fib, net); - - return 0; - -drop_withdraw: - c->stats.imp_withdraws_received++; - c->stats.imp_withdraws_ignored++; - return 0; -} - -int -rt_reload_channel(struct channel *c) -{ - struct rtable *tab = c->in_table; - struct fib_iterator *fit = &c->reload_fit; - int max_feed = 64; - - ASSERT(c->channel_state == CS_UP); + FIB_ITERATE_PUT(fit); + rt_send_export_event(c); - if (!c->reload_active) - { - FIB_ITERATE_INIT(fit, &tab->fib); - c->reload_active = 1; - } + RT_UNLOCK(c->table); + return; + } - do { - for (rte *e = c->reload_next_rte; e; e = e->next) - { - if (max_feed-- <= 0) + if (atomic_load_explicit(&c->export_state, memory_order_acquire) != TES_FEEDING) { - c->reload_next_rte = e; - debug("%s channel reload burst split (max_feed=%d)", c->proto->name, max_feed); - return 0; + RT_UNLOCK(c->table); + return; } - rte_update2(c, e->net->n.addr, rte_do_cow(e), e->attrs->src); - } - - c->reload_next_rte = NULL; - - FIB_ITERATE_START(&tab->fib, fit, net, n) - { - if (c->reload_next_rte = n->routes) + if (!n->routes || !rte_is_valid(&n->routes->rte)) + ; /* if no route, do nothing */ + else if (c->req->export_bulk) { - FIB_ITERATE_PUT_NEXT(fit, &tab->fib); - break; - } - } - FIB_ITERATE_END; - } - while (c->reload_next_rte); - - c->reload_active = 0; - return 1; -} - -void -rt_reload_channel_abort(struct channel *c) -{ - if (c->reload_active) - { - /* Unlink the iterator */ - fit_get(&c->in_table->fib, &c->reload_fit); - c->reload_next_rte = NULL; - c->reload_active = 0; - } -} - -void -rt_prune_sync(rtable *t, int all) -{ - struct fib_iterator fit; - - FIB_ITERATE_INIT(&fit, &t->fib); + uint count = rte_feed_count(n); + if (count) + { + rte **feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(n, feed, count); -again: - FIB_ITERATE_START(&t->fib, &fit, net, n) - { - rte *e, **ee = &n->routes; + struct rt_pending_export *rpe_last, *rpe_first = n->first; + for (struct rt_pending_export *rpe = rpe_first; rpe; rpe = rpe_next(rpe, NULL)) + rpe_last = rpe; - while (e = *ee) - { - if (all || (e->flags & (REF_STALE | REF_DISCARD))) - { - *ee = e->next; - rte_free_quick(e); - t->rt_count--; - } - else - ee = &e->next; - } + FIB_ITERATE_PUT_NEXT(fit, &tab->fib); + RT_UNLOCK(c->table); - if (all || !n->routes) - { - FIB_ITERATE_PUT(&fit); - fib_delete(&t->fib, n); - goto again; - } - } - FIB_ITERATE_END; -} + c->req->export_bulk(c->req, n->n.addr, NULL, feed, count); + RT_LOCK(c->table); -/* - * Export table - */ + for (struct rt_pending_export *rpe = rpe_first; rpe; rpe = rpe_next(rpe, NULL)) + { + rpe_mark_seen(c, rpe); + if (rpe == rpe_last) + break; + ASSERT_DIE(rpe->seq < rpe_last->seq); + } -int -rte_update_out(struct channel *c, const net_addr *n, rte *new, rte *old0, int refeed) -{ - struct rtable *tab = c->out_table; - struct rte_src *src; - rte *old, **pos; - net *net; + max_feed -= count; - if (new) - { - net = net_get(tab, n); - src = new->attrs->src; + goto redo; + } + } + else if (c->req->export_one) + { + struct rt_pending_export rpe = { .new = n->routes, .new_best = n->routes }; - rte_store_tmp_attrs(new, rte_update_pool, NULL); + struct rt_pending_export *rpe_last, *rpe_first = n->first; + for (struct rt_pending_export *rpe = rpe_first; rpe; rpe = rpe_next(rpe, NULL)) + rpe_last = rpe; - if (!rta_is_cached(new->attrs)) - new->attrs = rta_lookup(new->attrs); - } - else - { - net = net_find(tab, n); - src = old0->attrs->src; + FIB_ITERATE_PUT_NEXT(fit, &tab->fib); + RT_UNLOCK(c->table); - if (!net) - goto drop_withdraw; - } + c->req->export_one(c->req, n->n.addr, &rpe); - /* Find the old rte */ - for (pos = &net->routes; old = *pos; pos = &old->next) - if ((c->ra_mode != RA_ANY) || (old->attrs->src == src)) - { - if (new && rte_same(old, new)) - { - /* REF_STALE / REF_DISCARD not used in export table */ - /* - if (old->flags & (REF_STALE | REF_DISCARD | REF_MODIFY)) + RT_LOCK(c->table); + for (struct rt_pending_export *rpe = rpe_first; rpe; rpe = rpe_next(rpe, NULL)) { - old->flags &= ~(REF_STALE | REF_DISCARD | REF_MODIFY); - return 1; + rpe_mark_seen(c, rpe); + if (rpe == rpe_last) + break; + ASSERT_DIE(rpe->seq < rpe_last->seq); } - */ - goto drop_update; + max_feed--; + goto redo; } - - /* Remove the old rte */ - *pos = old->next; - rte_free_quick(old); - tab->rt_count--; - - break; + else + bug("Export request must always provide an export method"); } + FIB_ITERATE_END; - if (!new) - { - if (!old) - goto drop_withdraw; - - if (!net->routes) - fib_delete(&tab->fib, net); - - return 1; - } - - /* Insert the new rte */ - rte *e = rte_do_cow(new); - e->flags |= REF_COW; - e->net = net; - e->sender = c; - e->lastmod = current_time(); - e->next = *pos; - *pos = e; - tab->rt_count++; - return 1; + c->event->hook = rt_export_hook; + rt_send_export_event(c); -drop_update: - return refeed; + RT_UNLOCK(c->table); -drop_withdraw: - return 0; + rt_set_export_state(c, TES_READY); } @@ -2953,7 +3177,7 @@ hc_delete_hostentry(struct hostcache *hc, pool *p, struct hostentry *he) } static void -rt_init_hostcache(rtable *tab) +rt_init_hostcache(rtable_private *tab) { struct hostcache *hc = mb_allocz(tab->rp, sizeof(struct hostcache)); init_list(&hc->hostentries); @@ -2969,7 +3193,7 @@ rt_init_hostcache(rtable *tab) } static void -rt_free_hostcache(rtable *tab) +rt_free_hostcache(rtable_private *tab) { struct hostcache *hc = tab->hostcache; @@ -2992,13 +3216,13 @@ rt_free_hostcache(rtable *tab) } static void -rt_notify_hostcache(rtable *tab, net *net) +rt_notify_hostcache(rtable_private *tab, net *net) { - if (tab->hcu_scheduled) + if (ev_active(tab->hcu_event)) return; if (trie_match_net(tab->hostcache->trie, net->n.addr)) - rt_schedule_hcu(tab); + ev_send_loop(tab->loop, tab->hcu_event); } static int @@ -3021,41 +3245,17 @@ rt_get_igp_metric(rte *rt) if (ea) return ea->u.data; - rta *a = rt->attrs; - -#ifdef CONFIG_OSPF - if ((a->source == RTS_OSPF) || - (a->source == RTS_OSPF_IA) || - (a->source == RTS_OSPF_EXT1)) - return rt->u.ospf.metric1; -#endif - -#ifdef CONFIG_RIP - if (a->source == RTS_RIP) - return rt->u.rip.metric; -#endif - -#ifdef CONFIG_BGP - if (a->source == RTS_BGP) - { - u64 metric = bgp_total_aigp_metric(rt); - return (u32) MIN(metric, (u64) IGP_METRIC_UNKNOWN); - } -#endif - -#ifdef CONFIG_BABEL - if (a->source == RTS_BABEL) - return rt->u.babel.metric; -#endif - - if (a->source == RTS_DEVICE) + if (rt->attrs->source == RTS_DEVICE) return 0; + if (rt->src->owner->class->rte_igp_metric) + return rt->src->owner->class->rte_igp_metric(rt); + return IGP_METRIC_UNKNOWN; } static int -rt_update_hostentry(rtable *tab, struct hostentry *he) +rt_update_hostentry(rtable_private *tab, struct hostentry *he) { rta *old_src = he->src; int direct = 0; @@ -3072,11 +3272,12 @@ rt_update_hostentry(rtable *tab, struct hostentry *he) net *n = net_route(tab, &he_addr); if (n) { - rte *e = n->routes; - rta *a = e->attrs; - pxlen = n->n.addr->pxlen; + struct rte_storage *e = n->routes; + rta *a = e->rte.attrs; + word pref = a->pref; - if (a->hostentry) + for (struct rte_storage *ee = n->routes; ee; ee = ee->next) + if ((ee->rte.attrs->pref >= pref) && ee->rte.attrs->hostentry) { /* Recursive route should not depend on another recursive route */ log(L_WARN "Next hop address %I resolvable through recursive route for %N", @@ -3084,6 +3285,8 @@ rt_update_hostentry(rtable *tab, struct hostentry *he) goto done; } + pxlen = n->n.addr->pxlen; + if (a->dest == RTD_UNICAST) { for (struct nexthop *nh = &(a->nh); nh; nh = nh->next) @@ -3104,7 +3307,7 @@ rt_update_hostentry(rtable *tab, struct hostentry *he) he->src = rta_clone(a); he->dest = a->dest; he->nexthop_linkable = !direct; - he->igp_metric = rt_get_igp_metric(e); + he->igp_metric = rt_get_igp_metric(&e->rte); } done: @@ -3116,8 +3319,11 @@ done: } static void -rt_update_hostcache(rtable *tab) +rt_update_hostcache(void *data) { + rtable_private *tab = data; + ASSERT_DIE(birdloop_inside(tab->loop)); + struct hostcache *hc = tab->hostcache; struct hostentry *he; node *n, *x; @@ -3138,15 +3344,15 @@ rt_update_hostcache(rtable *tab) if (rt_update_hostentry(tab, he)) rt_schedule_nhu(he->tab); } - - tab->hcu_scheduled = 0; } struct hostentry * -rt_get_hostentry(rtable *tab, ip_addr a, ip_addr ll, rtable *dep) +rt_get_hostentry(rtable *t, ip_addr a, ip_addr ll, rtable *dep) { struct hostentry *he; + rtable_private *tab = RT_LOCK(t); + if (!tab->hostcache) rt_init_hostcache(tab); @@ -3154,10 +3360,13 @@ rt_get_hostentry(rtable *tab, ip_addr a, ip_addr ll, rtable *dep) struct hostcache *hc = tab->hostcache; for (he = hc->hash_table[k >> hc->hash_shift]; he != NULL; he = he->next) if (ipa_equal(he->addr, a) && (he->tab == dep)) - return he; + goto done; he = hc_new_hostentry(hc, tab->rp, a, ipa_zero(ll) ? a : ll, dep, k); rt_update_hostentry(tab, he); + +done: + RT_UNLOCK(t); return he; } diff --git a/proto/babel/babel.c b/proto/babel/babel.c index 1e87212c..515b27e5 100644 --- a/proto/babel/babel.c +++ b/proto/babel/babel.c @@ -641,13 +641,36 @@ babel_announce_rte(struct babel_proto *p, struct babel_entry *e) if (r) { rta a0 = { - .src = p->p.main_source, .source = RTS_BABEL, .scope = SCOPE_UNIVERSE, .dest = RTD_UNICAST, + .pref = c->preference, .from = r->neigh->addr, .nh.gw = r->next_hop, .nh.iface = r->neigh->ifa->iface, + .eattrs = alloca(sizeof(ea_list) + 3*sizeof(eattr)), + }; + + *a0.eattrs = (ea_list) { .count = 3 }; + a0.eattrs->attrs[0] = (eattr) { + .id = EA_BABEL_METRIC, + .type = EAF_TYPE_INT, + .u.data = r->metric, + }; + + struct adata *ad = alloca(sizeof(struct adata) + sizeof(u64)); + ad->length = sizeof(u64); + memcpy(ad->data, &(r->router_id), sizeof(u64)); + a0.eattrs->attrs[1] = (eattr) { + .id = EA_BABEL_ROUTER_ID, + .type = EAF_TYPE_OPAQUE, + .u.ptr = ad, + }; + + a0.eattrs->attrs[2] = (eattr) { + .id = EA_BABEL_SEQNO, + .type = EAF_TYPE_INT, + .u.data = r->seqno, }; /* @@ -658,40 +681,37 @@ babel_announce_rte(struct babel_proto *p, struct babel_entry *e) if (!neigh_find(&p->p, r->next_hop, r->neigh->ifa->iface, 0)) a0.nh.flags = RNF_ONLINK; - rta *a = rta_lookup(&a0); - rte *rte = rte_get_temp(a); - rte->u.babel.seqno = r->seqno; - rte->u.babel.metric = r->metric; - rte->u.babel.router_id = r->router_id; - rte->pflags = EA_ID_FLAG(EA_BABEL_METRIC) | EA_ID_FLAG(EA_BABEL_ROUTER_ID); + rte e0 = { + .attrs = &a0, + .src = p->p.main_source, + }; e->unreachable = 0; - rte_update2(c, e->n.addr, rte, p->p.main_source); + rte_update(c, e->n.addr, &e0, p->p.main_source); } else if (e->valid && (e->router_id != p->router_id)) { /* Unreachable */ rta a0 = { - .src = p->p.main_source, .source = RTS_BABEL, .scope = SCOPE_UNIVERSE, .dest = RTD_UNREACHABLE, + .pref = 1, }; - rta *a = rta_lookup(&a0); - rte *rte = rte_get_temp(a); - memset(&rte->u.babel, 0, sizeof(rte->u.babel)); - rte->pflags = 0; - rte->pref = 1; + rte e0 = { + .attrs = &a0, + .src = p->p.main_source, + }; e->unreachable = 1; - rte_update2(c, e->n.addr, rte, p->p.main_source); + rte_update(c, e->n.addr, &e0, p->p.main_source); } else { /* Retraction */ e->unreachable = 0; - rte_update2(c, e->n.addr, NULL, p->p.main_source); + rte_update(c, e->n.addr, NULL, p->p.main_source); } } @@ -701,7 +721,7 @@ babel_announce_retraction(struct babel_proto *p, struct babel_entry *e) { struct channel *c = (e->n.addr->type == NET_IP4) ? p->ip4_channel : p->ip6_channel; e->unreachable = 0; - rte_update2(c, e->n.addr, NULL, p->p.main_source); + rte_update(c, e->n.addr, NULL, p->p.main_source); } @@ -1727,7 +1747,7 @@ babel_add_iface(struct babel_proto *p, struct iface *new, struct babel_iface_con TRACE(D_EVENTS, "Adding interface %s", new->name); - pool *pool = rp_new(p->p.pool, new->name); + pool *pool = rp_new(p->p.pool, p->p.loop, new->name); ifa = mb_allocz(pool, sizeof(struct babel_iface)); ifa->proto = p; @@ -2010,7 +2030,13 @@ babel_dump(struct proto *P) static void babel_get_route_info(rte *rte, byte *buf) { - buf += bsprintf(buf, " (%d/%d) [%lR]", rte->pref, rte->u.babel.metric, rte->u.babel.router_id); + u64 rid = 0; + eattr *e = ea_find(rte->attrs->eattrs, EA_BABEL_ROUTER_ID); + if (e) + memcpy(&rid, e->u.ptr->data, sizeof(u64)); + + buf += bsprintf(buf, " (%d/%d) [%lR]", rte->attrs->pref, + ea_get_int(rte->attrs->eattrs, EA_BABEL_METRIC, BABEL_INFINITY), rid); } static int @@ -2018,6 +2044,9 @@ babel_get_attr(const eattr *a, byte *buf, int buflen UNUSED) { switch (a->id) { + case EA_BABEL_SEQNO: + return GA_FULL; + case EA_BABEL_METRIC: bsprintf(buf, "metric: %d", a->u.data); return GA_FULL; @@ -2231,45 +2260,23 @@ babel_kick_timer(struct babel_proto *p) static int -babel_preexport(struct proto *P, struct rte **new, struct linpool *pool UNUSED) +babel_preexport(struct channel *c, struct rte *new) { - struct rta *a = (*new)->attrs; - + struct rta *a = new->attrs; /* Reject our own unreachable routes */ - if ((a->dest == RTD_UNREACHABLE) && (a->src->proto == P)) + if ((a->dest == RTD_UNREACHABLE) && (new->src->owner == &c->proto->sources)) return -1; return 0; } -static void -babel_make_tmp_attrs(struct rte *rt, struct linpool *pool) -{ - struct adata *id = lp_alloc_adata(pool, sizeof(u64)); - memcpy(id->data, &rt->u.babel.router_id, sizeof(u64)); - - rte_init_tmp_attrs(rt, pool, 2); - rte_make_tmp_attr(rt, EA_BABEL_METRIC, EAF_TYPE_INT, rt->u.babel.metric); - rte_make_tmp_attr(rt, EA_BABEL_ROUTER_ID, EAF_TYPE_OPAQUE, (uintptr_t) id); -} - -static void -babel_store_tmp_attrs(struct rte *rt, struct linpool *pool) -{ - rte_init_tmp_attrs(rt, pool, 2); - rt->u.babel.metric = rte_store_tmp_attr(rt, EA_BABEL_METRIC); - - /* EA_BABEL_ROUTER_ID is read-only, we do not really save the value */ - rte_store_tmp_attr(rt, EA_BABEL_ROUTER_ID); -} - /* * babel_rt_notify - core tells us about new route (possibly our own), * so store it into our data structures. */ static void -babel_rt_notify(struct proto *P, struct channel *c UNUSED, struct network *net, - struct rte *new, struct rte *old UNUSED) +babel_rt_notify(struct proto *P, struct channel *c UNUSED, const net_addr *net, + struct rte *new, const struct rte *old UNUSED) { struct babel_proto *p = (void *) P; struct babel_entry *e; @@ -2277,19 +2284,31 @@ babel_rt_notify(struct proto *P, struct channel *c UNUSED, struct network *net, if (new) { /* Update */ - uint internal = (new->attrs->src->proto == P); - uint rt_seqno = internal ? new->u.babel.seqno : p->update_seqno; + uint rt_seqno; uint rt_metric = ea_get_int(new->attrs->eattrs, EA_BABEL_METRIC, 0); - u64 rt_router_id = internal ? new->u.babel.router_id : p->router_id; + u64 rt_router_id = 0; + + if (new->src->owner == &P->sources) + { + rt_seqno = ea_find(new->attrs->eattrs, EA_BABEL_SEQNO)->u.data; + eattr *e = ea_find(new->attrs->eattrs, EA_BABEL_ROUTER_ID); + if (e) + memcpy(&rt_router_id, e->u.ptr->data, sizeof(u64)); + } + else + { + rt_seqno = p->update_seqno; + rt_router_id = p->router_id; + } if (rt_metric > BABEL_INFINITY) { log(L_WARN "%s: Invalid babel_metric value %u for route %N", - p->p.name, rt_metric, net->n.addr); + p->p.name, rt_metric, net); rt_metric = BABEL_INFINITY; } - e = babel_get_entry(p, net->n.addr); + e = babel_get_entry(p, net); /* Activate triggered updates */ if ((e->valid != BABEL_ENTRY_VALID) || @@ -2307,7 +2326,7 @@ babel_rt_notify(struct proto *P, struct channel *c UNUSED, struct network *net, else { /* Withdraw */ - e = babel_find_entry(p, net->n.addr); + e = babel_find_entry(p, net); if (!e || e->valid != BABEL_ENTRY_VALID) return; @@ -2323,15 +2342,16 @@ babel_rt_notify(struct proto *P, struct channel *c UNUSED, struct network *net, static int babel_rte_better(struct rte *new, struct rte *old) { - return new->u.babel.metric < old->u.babel.metric; + uint new_metric = ea_find(new->attrs->eattrs, EA_BABEL_SEQNO)->u.data; + uint old_metric = ea_find(old->attrs->eattrs, EA_BABEL_SEQNO)->u.data; + + return new_metric < old_metric; } -static int -babel_rte_same(struct rte *new, struct rte *old) +static u32 +babel_rte_igp_metric(struct rte *rt) { - return ((new->u.babel.seqno == old->u.babel.seqno) && - (new->u.babel.metric == old->u.babel.metric) && - (new->u.babel.router_id == old->u.babel.router_id)); + return ea_get_int(rt->attrs->eattrs, EA_BABEL_METRIC, BABEL_INFINITY); } @@ -2352,6 +2372,12 @@ babel_postconfig(struct proto_config *CF) cf->ip6_channel = ip6 ?: ip6_sadr; } +static struct rte_owner_class babel_rte_owner_class = { + .get_route_info = babel_get_route_info, + .rte_better = babel_rte_better, + .rte_igp_metric = babel_rte_igp_metric, +}; + static struct proto * babel_init(struct proto_config *CF) { @@ -2365,10 +2391,8 @@ babel_init(struct proto_config *CF) P->if_notify = babel_if_notify; P->rt_notify = babel_rt_notify; P->preexport = babel_preexport; - P->make_tmp_attrs = babel_make_tmp_attrs; - P->store_tmp_attrs = babel_store_tmp_attrs; - P->rte_better = babel_rte_better; - P->rte_same = babel_rte_same; + + P->sources.class = &babel_rte_owner_class; return P; } @@ -2461,7 +2485,6 @@ babel_reconfigure(struct proto *P, struct proto_config *CF) return 1; } - struct protocol proto_babel = { .name = "Babel", .template = "babel%d", @@ -2476,6 +2499,5 @@ struct protocol proto_babel = { .start = babel_start, .shutdown = babel_shutdown, .reconfigure = babel_reconfigure, - .get_route_info = babel_get_route_info, .get_attr = babel_get_attr }; diff --git a/proto/babel/babel.h b/proto/babel/babel.h index 84feb085..8b6da3c8 100644 --- a/proto/babel/babel.h +++ b/proto/babel/babel.h @@ -28,6 +28,7 @@ #define EA_BABEL_METRIC EA_CODE(PROTOCOL_BABEL, 0) #define EA_BABEL_ROUTER_ID EA_CODE(PROTOCOL_BABEL, 1) +#define EA_BABEL_SEQNO EA_CODE(PROTOCOL_BABEL, 2) #define BABEL_MAGIC 42 #define BABEL_VERSION 2 diff --git a/proto/bfd/Makefile b/proto/bfd/Makefile index 402122fc..267dff98 100644 --- a/proto/bfd/Makefile +++ b/proto/bfd/Makefile @@ -1,6 +1,6 @@ -src := bfd.c io.c packets.c +src := bfd.c packets.c obj := $(src-o-files) $(all-daemon) $(cf-local) -tests_objs := $(tests_objs) $(src-o-files)
\ No newline at end of file +tests_objs := $(tests_objs) $(src-o-files) diff --git a/proto/bfd/bfd.c b/proto/bfd/bfd.c index dac184c5..c9b12aa1 100644 --- a/proto/bfd/bfd.c +++ b/proto/bfd/bfd.c @@ -113,8 +113,16 @@ #define HASH_IP_EQ(a1,n1,a2,n2) ipa_equal(a1, a2) && n1 == n2 #define HASH_IP_FN(a,n) ipa_hash(a) ^ u32_hash(n) -static list bfd_proto_list; -static list bfd_wait_list; +DEFINE_DOMAIN(rtable); +#define BFD_LOCK LOCK_DOMAIN(rtable, bfd_global.lock) +#define BFD_UNLOCK UNLOCK_DOMAIN(rtable, bfd_global.lock) + +static struct { + DOMAIN(rtable) lock; + list wait_list; + list pickup_list; + list proto_list; +} bfd_global; const char *bfd_state_names[] = { "AdminDown", "Down", "Init", "Up" }; @@ -188,7 +196,7 @@ bfd_session_update_tx_interval(struct bfd_session *s) return; /* Set timer relative to last tx_timer event */ - tm_set(s->tx_timer, s->last_tx + tx_int_l); + tm_set_in(s->tx_timer, s->last_tx + tx_int_l, s->ifa->bfd->p.loop); } static void @@ -202,7 +210,7 @@ bfd_session_update_detection_time(struct bfd_session *s, int kick) if (!s->last_rx) return; - tm_set(s->hold_timer, s->last_rx + timeout); + tm_set_in(s->hold_timer, s->last_rx + timeout, s->ifa->bfd->p.loop); } static void @@ -226,7 +234,7 @@ bfd_session_control_tx_timer(struct bfd_session *s, int reset) if (reset || !tm_active(s->tx_timer)) { s->last_tx = 0; - tm_start(s->tx_timer, 0); + tm_start_in(s->tx_timer, 0, s->ifa->bfd->p.loop); } return; @@ -419,7 +427,7 @@ bfd_get_free_id(struct bfd_proto *p) static struct bfd_session * bfd_add_session(struct bfd_proto *p, ip_addr addr, ip_addr local, struct iface *iface, struct bfd_options *opts) { - birdloop_enter(p->loop); + ASSERT_DIE(birdloop_inside(p->p.loop)); struct bfd_iface *ifa = bfd_get_iface(p, local, iface); @@ -444,8 +452,8 @@ bfd_add_session(struct bfd_proto *p, ip_addr addr, ip_addr local, struct iface * s->passive = s->cf.passive; s->tx_csn = random_u32(); - s->tx_timer = tm_new_init(p->tpool, bfd_tx_timer_hook, s, 0, 0); - s->hold_timer = tm_new_init(p->tpool, bfd_hold_timer_hook, s, 0, 0); + s->tx_timer = tm_new_init(p->p.pool, bfd_tx_timer_hook, s, 0, 0); + s->hold_timer = tm_new_init(p->p.pool, bfd_hold_timer_hook, s, 0, 0); bfd_session_update_tx_interval(s); bfd_session_control_tx_timer(s, 1); @@ -454,8 +462,6 @@ bfd_add_session(struct bfd_proto *p, ip_addr addr, ip_addr local, struct iface * TRACE(D_EVENTS, "Session to %I added", s->addr); - birdloop_leave(p->loop); - return s; } @@ -463,38 +469,34 @@ bfd_add_session(struct bfd_proto *p, ip_addr addr, ip_addr local, struct iface * static void bfd_open_session(struct bfd_proto *p, struct bfd_session *s, ip_addr local, struct iface *ifa) { - birdloop_enter(p->loop); + birdloop_enter(p->p.loop); s->opened = 1; bfd_session_control_tx_timer(s); - birdloop_leave(p->loop); + birdloop_leave(p->p.loop); } static void bfd_close_session(struct bfd_proto *p, struct bfd_session *s) { - birdloop_enter(p->loop); + birdloop_enter(p->p.loop); s->opened = 0; bfd_session_update_state(s, BFD_STATE_DOWN, BFD_DIAG_PATH_DOWN); bfd_session_control_tx_timer(s); - birdloop_leave(p->loop); + birdloop_leave(p->p.loop); } */ static void -bfd_remove_session(struct bfd_proto *p, struct bfd_session *s) +bfd_remove_session_locked(struct bfd_proto *p, struct bfd_session *s) { - ip_addr ip = s->addr; - /* Caller should ensure that request list is empty */ - birdloop_enter(p->loop); - /* Remove session from notify list if scheduled for notification */ /* No need for bfd_lock_sessions(), we are already protected by birdloop_enter() */ if (NODE_VALID(&s->n)) @@ -508,11 +510,17 @@ bfd_remove_session(struct bfd_proto *p, struct bfd_session *s) HASH_REMOVE(p->session_hash_id, HASH_ID, s); HASH_REMOVE(p->session_hash_ip, HASH_IP, s); - sl_free(p->session_slab, s); + TRACE(D_EVENTS, "Session to %I removed", s->addr); - TRACE(D_EVENTS, "Session to %I removed", ip); + sl_free(p->session_slab, s); +} - birdloop_leave(p->loop); +static void +bfd_remove_session(struct bfd_proto *p, struct bfd_session *s) +{ + birdloop_enter(p->p.loop); + bfd_remove_session_locked(p, s); + birdloop_leave(p->p.loop); } static void @@ -521,7 +529,7 @@ bfd_reconfigure_session(struct bfd_proto *p, struct bfd_session *s) if (EMPTY_LIST(s->request_list)) return; - birdloop_enter(p->loop); + birdloop_enter(p->p.loop); struct bfd_request *req = SKIP_BACK(struct bfd_request, n, HEAD(s->request_list)); s->cf = bfd_merge_options(s->ifa->cf, &req->opts); @@ -534,7 +542,7 @@ bfd_reconfigure_session(struct bfd_proto *p, struct bfd_session *s) bfd_session_control_tx_timer(s, 0); - birdloop_leave(p->loop); + birdloop_leave(p->p.loop); TRACE(D_EVENTS, "Session to %I reconfigured", s->addr); } @@ -573,7 +581,7 @@ bfd_get_iface(struct bfd_proto *p, ip_addr local, struct iface *iface) struct bfd_config *cf = (struct bfd_config *) (p->p.cf); struct bfd_iface_config *ic = bfd_find_iface_config(cf, iface); - ifa = mb_allocz(p->tpool, sizeof(struct bfd_iface)); + ifa = mb_allocz(p->p.pool, sizeof(struct bfd_iface)); ifa->local = local; ifa->iface = iface; ifa->cf = ic; @@ -618,9 +626,9 @@ bfd_reconfigure_iface(struct bfd_proto *p, struct bfd_iface *ifa, struct bfd_con (new->passive != old->passive); /* This should be probably changed to not access ifa->cf from the BFD thread */ - birdloop_enter(p->loop); + birdloop_enter(p->p.loop); ifa->cf = new; - birdloop_leave(p->loop); + birdloop_leave(p->p.loop); } @@ -650,7 +658,7 @@ bfd_add_request(struct bfd_proto *p, struct bfd_request *req) { struct bfd_config *cf = (struct bfd_config *) (p->p.cf); - if (p->p.vrf_set && (p->p.vrf != req->vrf)) + if (p->p.vrf && (p->p.vrf != req->vrf)) return 0; if (ipa_is_ip4(req->addr) ? !cf->accept_ipv4 : !cf->accept_ipv6) @@ -681,41 +689,68 @@ bfd_add_request(struct bfd_proto *p, struct bfd_request *req) } static void -bfd_submit_request(struct bfd_request *req) +bfd_pickup_requests(void *_data UNUSED) { node *n; + WALK_LIST(n, bfd_global.proto_list) + { + struct bfd_proto *p = SKIP_BACK(struct bfd_proto, bfd_node, n); + birdloop_enter(p->p.loop); + BFD_LOCK; - WALK_LIST(n, bfd_proto_list) - if (bfd_add_request(SKIP_BACK(struct bfd_proto, bfd_node, n), req)) - return; + node *rn, *rnxt; + WALK_LIST_DELSAFE(rn, rnxt, bfd_global.pickup_list) + bfd_add_request(p, SKIP_BACK(struct bfd_request, n, rn)); - rem_node(&req->n); - add_tail(&bfd_wait_list, &req->n); - req->session = NULL; - bfd_request_notify(req, BFD_STATE_ADMIN_DOWN, 0); + BFD_UNLOCK; + birdloop_leave(p->p.loop); + } + + BFD_LOCK; + node *rn, *rnxt; + WALK_LIST_DELSAFE(rn, rnxt, bfd_global.pickup_list) + { + rem_node(rn); + add_tail(&bfd_global.wait_list, rn); + bfd_request_notify(SKIP_BACK(struct bfd_request, n, rn), BFD_STATE_ADMIN_DOWN, 0); + } + BFD_UNLOCK; } +static event bfd_pickup_event = { .hook = bfd_pickup_requests }; + static void bfd_take_requests(struct bfd_proto *p) { node *n, *nn; - - WALK_LIST_DELSAFE(n, nn, bfd_wait_list) + BFD_LOCK; + WALK_LIST_DELSAFE(n, nn, bfd_global.wait_list) bfd_add_request(p, SKIP_BACK(struct bfd_request, n, n)); + BFD_UNLOCK; } static void bfd_drop_requests(struct bfd_proto *p) { node *n; - - HASH_WALK(p->session_hash_id, next_id, s) + BFD_LOCK; + HASH_WALK_DELSAFE(p->session_hash_id, next_id, s) { - /* We assume that p is not in bfd_proto_list */ WALK_LIST_FIRST(n, s->request_list) - bfd_submit_request(SKIP_BACK(struct bfd_request, n, n)); + { + struct bfd_request *req = SKIP_BACK(struct bfd_request, n, n); + rem_node(&req->n); + add_tail(&bfd_global.pickup_list, &req->n); + req->session = NULL; + bfd_request_notify(req, BFD_STATE_ADMIN_DOWN, 0); + } + + ev_send(&global_event_list, &bfd_pickup_event); + + bfd_remove_session_locked(p, s); } HASH_WALK_END; + BFD_UNLOCK; } static struct resclass bfd_request_class; @@ -728,9 +763,6 @@ bfd_request_session(pool *p, ip_addr addr, ip_addr local, { struct bfd_request *req = ralloc(p, &bfd_request_class); - /* Hack: self-link req->n, we will call rem_node() on it */ - req->n.prev = req->n.next = &req->n; - req->addr = addr; req->local = local; req->iface = iface; @@ -739,11 +771,16 @@ bfd_request_session(pool *p, ip_addr addr, ip_addr local, if (opts) req->opts = *opts; - bfd_submit_request(req); - req->hook = hook; req->data = data; + req->session = NULL; + + BFD_LOCK; + add_tail(&bfd_global.pickup_list, &req->n); + ev_send(&global_event_list, &bfd_pickup_event); + BFD_UNLOCK; + return req; } @@ -1001,8 +1038,10 @@ bfd_notify_init(struct bfd_proto *p) void bfd_init_all(void) { - init_list(&bfd_proto_list); - init_list(&bfd_wait_list); + bfd_global.lock = DOMAIN_NEW(rtable, "BFD Global"); + init_list(&bfd_global.wait_list); + init_list(&bfd_global.pickup_list); + init_list(&bfd_global.proto_list); } static struct proto * @@ -1021,8 +1060,6 @@ bfd_start(struct proto *P) struct bfd_proto *p = (struct bfd_proto *) P; struct bfd_config *cf = (struct bfd_config *) (P->cf); - p->loop = birdloop_new(); - p->tpool = rp_new(NULL, "BFD thread root"); pthread_spin_init(&p->lock, PTHREAD_PROCESS_PRIVATE); p->session_slab = sl_new(P->pool, sizeof(struct bfd_session)); @@ -1034,9 +1071,7 @@ bfd_start(struct proto *P) init_list(&p->notify_list); bfd_notify_init(p); - add_tail(&bfd_proto_list, &p->bfd_node); - - birdloop_enter(p->loop); + add_tail(&bfd_global.proto_list, &p->bfd_node); if (cf->accept_ipv4 && cf->accept_direct) p->rx4_1 = bfd_open_rx_sk(p, 0, SK_IPV4); @@ -1050,42 +1085,33 @@ bfd_start(struct proto *P) if (cf->accept_ipv6 && cf->accept_multihop) p->rx6_m = bfd_open_rx_sk(p, 1, SK_IPV6); - birdloop_leave(p->loop); - bfd_take_requests(p); struct bfd_neighbor *n; WALK_LIST(n, cf->neigh_list) bfd_start_neighbor(p, n); - birdloop_start(p->loop); - return PS_UP; } - static int bfd_shutdown(struct proto *P) { struct bfd_proto *p = (struct bfd_proto *) P; - struct bfd_config *cf = (struct bfd_config *) (P->cf); + struct bfd_config *cf = (struct bfd_config *) (p->p.cf); rem_node(&p->bfd_node); - birdloop_stop(p->loop); - - struct bfd_neighbor *n; - WALK_LIST(n, cf->neigh_list) - bfd_stop_neighbor(p, n); + struct bfd_neighbor *bn; + WALK_LIST(bn, cf->neigh_list) + bfd_stop_neighbor(p, bn); bfd_drop_requests(p); - /* FIXME: This is hack */ - birdloop_enter(p->loop); - rfree(p->tpool); - birdloop_leave(p->loop); - - birdloop_free(p->loop); + if (p->rx4_1) sk_stop(p->rx4_1); + if (p->rx4_m) sk_stop(p->rx4_m); + if (p->rx6_1) sk_stop(p->rx6_1); + if (p->rx6_m) sk_stop(p->rx6_m); return PS_DOWN; } @@ -1105,7 +1131,7 @@ bfd_reconfigure(struct proto *P, struct proto_config *c) (new->accept_multihop != old->accept_multihop)) return 0; - birdloop_mask_wakeups(p->loop); + birdloop_mask_wakeups(p->p.loop); WALK_LIST(ifa, p->iface_list) bfd_reconfigure_iface(p, ifa, new); @@ -1119,7 +1145,7 @@ bfd_reconfigure(struct proto *P, struct proto_config *c) bfd_reconfigure_neighbors(p, new); - birdloop_unmask_wakeups(p->loop); + birdloop_unmask_wakeups(p->p.loop); return 1; } diff --git a/proto/bfd/bfd.h b/proto/bfd/bfd.h index 91fdaa60..475b3a11 100644 --- a/proto/bfd/bfd.h +++ b/proto/bfd/bfd.h @@ -17,12 +17,12 @@ #include "nest/password.h" #include "conf/conf.h" #include "lib/hash.h" +#include "lib/io-loop.h" #include "lib/resource.h" #include "lib/socket.h" #include "lib/string.h" #include "nest/bfd.h" -#include "io.h" #define BFD_CONTROL_PORT 3784 @@ -87,9 +87,9 @@ struct bfd_neighbor struct bfd_proto { struct proto p; - struct birdloop *loop; - pool *tpool; + pthread_spinlock_t lock; + node bfd_node; slab *session_slab; diff --git a/proto/bfd/config.Y b/proto/bfd/config.Y index df1cba42..ed5479fb 100644 --- a/proto/bfd/config.Y +++ b/proto/bfd/config.Y @@ -36,6 +36,7 @@ proto: bfd_proto ; bfd_proto_start: proto_start BFD { this_proto = proto_config_new(&proto_bfd, $1); + this_proto->loop_order = DOMAIN_ORDER(proto); init_list(&BFD_CFG->patt_list); init_list(&BFD_CFG->neigh_list); BFD_CFG->accept_ipv4 = BFD_CFG->accept_ipv6 = 1; diff --git a/proto/bfd/io.c b/proto/bfd/io.c deleted file mode 100644 index 1cd9365a..00000000 --- a/proto/bfd/io.c +++ /dev/null @@ -1,535 +0,0 @@ -/* - * BIRD -- I/O and event loop - * - * Can be freely distributed and used under the terms of the GNU GPL. - */ - -#include <stdio.h> -#include <stdlib.h> -#include <unistd.h> -#include <errno.h> -#include <fcntl.h> -#include <poll.h> -#include <pthread.h> -#include <time.h> -#include <sys/time.h> - -#include "nest/bird.h" -#include "proto/bfd/io.h" - -#include "lib/buffer.h" -#include "lib/lists.h" -#include "lib/resource.h" -#include "lib/event.h" -#include "lib/timer.h" -#include "lib/socket.h" - - -struct birdloop -{ - pool *pool; - pthread_t thread; - pthread_mutex_t mutex; - - u8 stop_called; - u8 poll_active; - u8 wakeup_masked; - int wakeup_fds[2]; - - struct timeloop time; - list event_list; - list sock_list; - uint sock_num; - - BUFFER(sock *) poll_sk; - BUFFER(struct pollfd) poll_fd; - u8 poll_changed; - u8 close_scheduled; -}; - - -/* - * Current thread context - */ - -static pthread_key_t current_loop_key; -extern pthread_key_t current_time_key; - -static inline struct birdloop * -birdloop_current(void) -{ - return pthread_getspecific(current_loop_key); -} - -static inline void -birdloop_set_current(struct birdloop *loop) -{ - pthread_setspecific(current_loop_key, loop); - pthread_setspecific(current_time_key, loop ? &loop->time : &main_timeloop); -} - -static inline void -birdloop_init_current(void) -{ - pthread_key_create(¤t_loop_key, NULL); -} - - -/* - * Wakeup code for birdloop - */ - -static void -pipe_new(int *pfds) -{ - int rv = pipe(pfds); - if (rv < 0) - die("pipe: %m"); - - if (fcntl(pfds[0], F_SETFL, O_NONBLOCK) < 0) - die("fcntl(O_NONBLOCK): %m"); - - if (fcntl(pfds[1], F_SETFL, O_NONBLOCK) < 0) - die("fcntl(O_NONBLOCK): %m"); -} - -void -pipe_drain(int fd) -{ - char buf[64]; - int rv; - - try: - rv = read(fd, buf, 64); - if (rv < 0) - { - if (errno == EINTR) - goto try; - if (errno == EAGAIN) - return; - die("wakeup read: %m"); - } - if (rv == 64) - goto try; -} - -void -pipe_kick(int fd) -{ - u64 v = 1; - int rv; - - try: - rv = write(fd, &v, sizeof(u64)); - if (rv < 0) - { - if (errno == EINTR) - goto try; - if (errno == EAGAIN) - return; - die("wakeup write: %m"); - } -} - -static inline void -wakeup_init(struct birdloop *loop) -{ - pipe_new(loop->wakeup_fds); -} - -static inline void -wakeup_drain(struct birdloop *loop) -{ - pipe_drain(loop->wakeup_fds[0]); -} - -static inline void -wakeup_do_kick(struct birdloop *loop) -{ - pipe_kick(loop->wakeup_fds[1]); -} - -static inline void -wakeup_kick(struct birdloop *loop) -{ - if (!loop->wakeup_masked) - wakeup_do_kick(loop); - else - loop->wakeup_masked = 2; -} - -/* For notifications from outside */ -void -wakeup_kick_current(void) -{ - struct birdloop *loop = birdloop_current(); - - if (loop && loop->poll_active) - wakeup_kick(loop); -} - - -/* - * Events - */ - -static inline uint -events_waiting(struct birdloop *loop) -{ - return !EMPTY_LIST(loop->event_list); -} - -static inline void -events_init(struct birdloop *loop) -{ - init_list(&loop->event_list); -} - -static void -events_fire(struct birdloop *loop) -{ - times_update(&loop->time); - ev_run_list(&loop->event_list); -} - -void -ev2_schedule(event *e) -{ - struct birdloop *loop = birdloop_current(); - - if (loop->poll_active && EMPTY_LIST(loop->event_list)) - wakeup_kick(loop); - - if (e->n.next) - rem_node(&e->n); - - add_tail(&loop->event_list, &e->n); -} - - -/* - * Sockets - */ - -static void -sockets_init(struct birdloop *loop) -{ - init_list(&loop->sock_list); - loop->sock_num = 0; - - BUFFER_INIT(loop->poll_sk, loop->pool, 4); - BUFFER_INIT(loop->poll_fd, loop->pool, 4); - loop->poll_changed = 1; /* add wakeup fd */ -} - -static void -sockets_add(struct birdloop *loop, sock *s) -{ - add_tail(&loop->sock_list, &s->n); - loop->sock_num++; - - s->index = -1; - loop->poll_changed = 1; - - if (loop->poll_active) - wakeup_kick(loop); -} - -void -sk_start(sock *s) -{ - struct birdloop *loop = birdloop_current(); - - sockets_add(loop, s); -} - -static void -sockets_remove(struct birdloop *loop, sock *s) -{ - rem_node(&s->n); - loop->sock_num--; - - if (s->index >= 0) - loop->poll_sk.data[s->index] = NULL; - - s->index = -1; - loop->poll_changed = 1; - - /* Wakeup moved to sk_stop() */ -} - -void -sk_stop(sock *s) -{ - struct birdloop *loop = birdloop_current(); - - sockets_remove(loop, s); - - if (loop->poll_active) - { - loop->close_scheduled = 1; - wakeup_kick(loop); - } - else - close(s->fd); - - s->fd = -1; -} - -static inline uint sk_want_events(sock *s) -{ return (s->rx_hook ? POLLIN : 0) | ((s->ttx != s->tpos) ? POLLOUT : 0); } - -/* -FIXME: this should be called from sock code - -static void -sockets_update(struct birdloop *loop, sock *s) -{ - if (s->index >= 0) - loop->poll_fd.data[s->index].events = sk_want_events(s); -} -*/ - -static void -sockets_prepare(struct birdloop *loop) -{ - BUFFER_SET(loop->poll_sk, loop->sock_num + 1); - BUFFER_SET(loop->poll_fd, loop->sock_num + 1); - - struct pollfd *pfd = loop->poll_fd.data; - sock **psk = loop->poll_sk.data; - uint i = 0; - node *n; - - WALK_LIST(n, loop->sock_list) - { - sock *s = SKIP_BACK(sock, n, n); - - ASSERT(i < loop->sock_num); - - s->index = i; - *psk = s; - pfd->fd = s->fd; - pfd->events = sk_want_events(s); - pfd->revents = 0; - - pfd++; - psk++; - i++; - } - - ASSERT(i == loop->sock_num); - - /* Add internal wakeup fd */ - *psk = NULL; - pfd->fd = loop->wakeup_fds[0]; - pfd->events = POLLIN; - pfd->revents = 0; - - loop->poll_changed = 0; -} - -static void -sockets_close_fds(struct birdloop *loop) -{ - struct pollfd *pfd = loop->poll_fd.data; - sock **psk = loop->poll_sk.data; - int poll_num = loop->poll_fd.used - 1; - - int i; - for (i = 0; i < poll_num; i++) - if (psk[i] == NULL) - close(pfd[i].fd); - - loop->close_scheduled = 0; -} - -int sk_read(sock *s, int revents); -int sk_write(sock *s); - -static void -sockets_fire(struct birdloop *loop) -{ - struct pollfd *pfd = loop->poll_fd.data; - sock **psk = loop->poll_sk.data; - int poll_num = loop->poll_fd.used - 1; - - times_update(&loop->time); - - /* Last fd is internal wakeup fd */ - if (pfd[poll_num].revents & POLLIN) - wakeup_drain(loop); - - int i; - for (i = 0; i < poll_num; pfd++, psk++, i++) - { - int e = 1; - - if (! pfd->revents) - continue; - - if (pfd->revents & POLLNVAL) - die("poll: invalid fd %d", pfd->fd); - - if (pfd->revents & POLLIN) - while (e && *psk && (*psk)->rx_hook) - e = sk_read(*psk, 0); - - e = 1; - if (pfd->revents & POLLOUT) - while (e && *psk) - e = sk_write(*psk); - } -} - - -/* - * Birdloop - */ - -static void * birdloop_main(void *arg); - -struct birdloop * -birdloop_new(void) -{ - /* FIXME: this init should be elsewhere and thread-safe */ - static int init = 0; - if (!init) - { birdloop_init_current(); init = 1; } - - pool *p = rp_new(NULL, "Birdloop root"); - struct birdloop *loop = mb_allocz(p, sizeof(struct birdloop)); - loop->pool = p; - pthread_mutex_init(&loop->mutex, NULL); - - wakeup_init(loop); - - events_init(loop); - timers_init(&loop->time, p); - sockets_init(loop); - - return loop; -} - -void -birdloop_start(struct birdloop *loop) -{ - int rv = pthread_create(&loop->thread, NULL, birdloop_main, loop); - if (rv) - die("pthread_create(): %M", rv); -} - -void -birdloop_stop(struct birdloop *loop) -{ - pthread_mutex_lock(&loop->mutex); - loop->stop_called = 1; - wakeup_do_kick(loop); - pthread_mutex_unlock(&loop->mutex); - - int rv = pthread_join(loop->thread, NULL); - if (rv) - die("pthread_join(): %M", rv); -} - -void -birdloop_free(struct birdloop *loop) -{ - rfree(loop->pool); -} - - -void -birdloop_enter(struct birdloop *loop) -{ - /* TODO: these functions could save and restore old context */ - pthread_mutex_lock(&loop->mutex); - birdloop_set_current(loop); -} - -void -birdloop_leave(struct birdloop *loop) -{ - /* TODO: these functions could save and restore old context */ - birdloop_set_current(NULL); - pthread_mutex_unlock(&loop->mutex); -} - -void -birdloop_mask_wakeups(struct birdloop *loop) -{ - pthread_mutex_lock(&loop->mutex); - loop->wakeup_masked = 1; - pthread_mutex_unlock(&loop->mutex); -} - -void -birdloop_unmask_wakeups(struct birdloop *loop) -{ - pthread_mutex_lock(&loop->mutex); - if (loop->wakeup_masked == 2) - wakeup_do_kick(loop); - loop->wakeup_masked = 0; - pthread_mutex_unlock(&loop->mutex); -} - -static void * -birdloop_main(void *arg) -{ - struct birdloop *loop = arg; - timer *t; - int rv, timeout; - - birdloop_set_current(loop); - - pthread_mutex_lock(&loop->mutex); - while (1) - { - events_fire(loop); - timers_fire(&loop->time); - - times_update(&loop->time); - if (events_waiting(loop)) - timeout = 0; - else if (t = timers_first(&loop->time)) - timeout = (tm_remains(t) TO_MS) + 1; - else - timeout = -1; - - if (loop->poll_changed) - sockets_prepare(loop); - - loop->poll_active = 1; - pthread_mutex_unlock(&loop->mutex); - - try: - rv = poll(loop->poll_fd.data, loop->poll_fd.used, timeout); - if (rv < 0) - { - if (errno == EINTR || errno == EAGAIN) - goto try; - die("poll: %m"); - } - - pthread_mutex_lock(&loop->mutex); - loop->poll_active = 0; - - if (loop->close_scheduled) - sockets_close_fds(loop); - - if (loop->stop_called) - break; - - if (rv) - sockets_fire(loop); - - timers_fire(&loop->time); - } - - loop->stop_called = 0; - pthread_mutex_unlock(&loop->mutex); - - return NULL; -} - - diff --git a/proto/bfd/io.h b/proto/bfd/io.h deleted file mode 100644 index ec706e9a..00000000 --- a/proto/bfd/io.h +++ /dev/null @@ -1,34 +0,0 @@ -/* - * BIRD -- I/O and event loop - * - * Can be freely distributed and used under the terms of the GNU GPL. - */ - -#ifndef _BIRD_BFD_IO_H_ -#define _BIRD_BFD_IO_H_ - -#include "nest/bird.h" -#include "lib/lists.h" -#include "lib/resource.h" -#include "lib/event.h" -#include "lib/timer.h" -#include "lib/socket.h" - - -void ev2_schedule(event *e); - -void sk_start(sock *s); -void sk_stop(sock *s); - -struct birdloop *birdloop_new(void); -void birdloop_start(struct birdloop *loop); -void birdloop_stop(struct birdloop *loop); -void birdloop_free(struct birdloop *loop); - -void birdloop_enter(struct birdloop *loop); -void birdloop_leave(struct birdloop *loop); -void birdloop_mask_wakeups(struct birdloop *loop); -void birdloop_unmask_wakeups(struct birdloop *loop); - - -#endif /* _BIRD_BFD_IO_H_ */ diff --git a/proto/bfd/packets.c b/proto/bfd/packets.c index 7618e20f..37d77f37 100644 --- a/proto/bfd/packets.c +++ b/proto/bfd/packets.c @@ -410,7 +410,7 @@ bfd_err_hook(sock *sk, int err) sock * bfd_open_rx_sk(struct bfd_proto *p, int multihop, int af) { - sock *sk = sk_new(p->tpool); + sock *sk = sk_new(p->p.pool); sk->type = SK_UDP; sk->subtype = af; sk->sport = !multihop ? BFD_CONTROL_PORT : BFD_MULTI_CTL_PORT; @@ -441,7 +441,7 @@ bfd_open_rx_sk(struct bfd_proto *p, int multihop, int af) sock * bfd_open_tx_sk(struct bfd_proto *p, ip_addr local, struct iface *ifa) { - sock *sk = sk_new(p->tpool); + sock *sk = sk_new(p->p.pool); sk->type = SK_UDP; sk->saddr = local; sk->dport = ifa ? BFD_CONTROL_PORT : BFD_MULTI_CTL_PORT; diff --git a/proto/bgp/attrs.c b/proto/bgp/attrs.c index 24ba00ba..1080db77 100644 --- a/proto/bgp/attrs.c +++ b/proto/bgp/attrs.c @@ -333,26 +333,26 @@ bgp_aigp_set_metric(struct linpool *pool, const struct adata *ad, u64 metric) } int -bgp_total_aigp_metric_(rte *e, u64 *metric, const struct adata **ad) +bgp_total_aigp_metric_(struct rta *a, u64 *metric, const struct adata **ad) { - eattr *a = ea_find(e->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_AIGP)); - if (!a) + eattr *ea = ea_find(a->eattrs, EA_CODE(PROTOCOL_BGP, BA_AIGP)); + if (!ea) return 0; - const byte *b = bgp_aigp_get_tlv(a->u.ptr, BGP_AIGP_METRIC); + const byte *b = bgp_aigp_get_tlv(ea->u.ptr, BGP_AIGP_METRIC); if (!b) return 0; u64 aigp = get_u64(b + 3); - u64 step = e->attrs->igp_metric; + u64 step = a->igp_metric; - if (!rte_resolvable(e) || (step >= IGP_METRIC_UNKNOWN)) + if (!rta_resolvable(a) || (step >= IGP_METRIC_UNKNOWN)) step = BGP_AIGP_MAX; if (!step) step = 1; - *ad = a->u.ptr; + *ad = ea->u.ptr; *metric = aigp + step; if (*metric < aigp) *metric = BGP_AIGP_MAX; @@ -371,6 +371,13 @@ bgp_init_aigp_metric(rte *e, u64 *metric, const struct adata **ad) return *metric < IGP_METRIC_UNKNOWN; } +u32 +bgp_rte_igp_metric(struct rte *rt) +{ + u64 metric = bgp_total_aigp_metric(rt->attrs); + return (u32) MIN(metric, (u64) IGP_METRIC_UNKNOWN); +} + /* * Attribute hooks @@ -896,7 +903,7 @@ bgp_decode_large_community(struct bgp_parse_state *s, uint code UNUSED, uint fla static void bgp_export_mpls_label_stack(struct bgp_export_state *s, eattr *a) { - net_addr *n = s->route->net->n.addr; + const net_addr *n = s->route->net; u32 *labels = (u32 *) a->u.ptr->data; uint lnum = a->u.ptr->length / 4; @@ -1617,7 +1624,7 @@ bgp_free_prefix_table(struct bgp_channel *c) } static struct bgp_prefix * -bgp_get_prefix(struct bgp_channel *c, net_addr *net, u32 path_id) +bgp_get_prefix(struct bgp_channel *c, const net_addr *net, u32 path_id) { u32 hash = net_hash(net) ^ u32_hash(path_id); struct bgp_prefix *px = HASH_FIND(c->prefix_hash, PXH, net, path_id, hash); @@ -1661,12 +1668,10 @@ bgp_free_prefix(struct bgp_channel *c, struct bgp_prefix *px) */ int -bgp_preexport(struct proto *P, rte **new, struct linpool *pool UNUSED) +bgp_preexport(struct channel *c, rte *e) { - rte *e = *new; - struct proto *SRC = e->attrs->src->proto; - struct bgp_proto *p = (struct bgp_proto *) P; - struct bgp_proto *src = (SRC->proto == &proto_bgp) ? (struct bgp_proto *) SRC : NULL; + struct bgp_proto *p = (struct bgp_proto *) (c->proto); + struct bgp_proto *src = bgp_rte_proto(e); /* Reject our routes */ if (src == p) @@ -1690,11 +1695,11 @@ bgp_preexport(struct proto *P, rte **new, struct linpool *pool UNUSED) } /* Handle well-known communities, RFC 1997 */ - struct eattr *c; + struct eattr *com; if (p->cf->interpret_communities && - (c = ea_find(e->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY)))) + (com = ea_find(e->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY)))) { - const struct adata *d = c->u.ptr; + const struct adata *d = com->u.ptr; /* Do not export anywhere */ if (int_set_contains(d, BGP_COMM_NO_ADVERTISE)) @@ -1719,8 +1724,7 @@ bgp_preexport(struct proto *P, rte **new, struct linpool *pool UNUSED) static ea_list * bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *attrs0, struct linpool *pool) { - struct proto *SRC = e->attrs->src->proto; - struct bgp_proto *src = (SRC->proto == &proto_bgp) ? (void *) SRC : NULL; + struct bgp_proto *src = bgp_rte_proto(e); struct bgp_export_state s = { .proto = p, .channel = c, .pool = pool, .src = src, .route = e, .mpls = c->desc->mpls }; ea_list *attrs = attrs0; eattr *a; @@ -1774,7 +1778,7 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at /* AIGP attribute - accumulate local metric or originate new one */ u64 metric; if (s.local_next_hop && - (bgp_total_aigp_metric_(e, &metric, &ad) || + (bgp_total_aigp_metric_(e->attrs, &metric, &ad) || (c->cf->aigp_originate && bgp_init_aigp_metric(e, &metric, &ad)))) { ad = bgp_aigp_set_metric(pool, ad, metric); @@ -1833,7 +1837,7 @@ bgp_update_attrs(struct bgp_proto *p, struct bgp_channel *c, rte *e, ea_list *at } void -bgp_rt_notify(struct proto *P, struct channel *C, net *n, rte *new, rte *old) +bgp_rt_notify(struct proto *P, struct channel *C, const net_addr *n, rte *new, const rte *old) { struct bgp_proto *p = (void *) P; struct bgp_channel *c = (void *) C; @@ -1847,17 +1851,17 @@ bgp_rt_notify(struct proto *P, struct channel *C, net *n, rte *new, rte *old) /* If attributes are invalid, we fail back to withdraw */ buck = attrs ? bgp_get_bucket(c, attrs) : bgp_get_withdraw_bucket(c); - path = new->attrs->src->global_id; + path = new->src->global_id; lp_flush(bgp_linpool2); } else { buck = bgp_get_withdraw_bucket(c); - path = old->attrs->src->global_id; + path = old->src->global_id; } - px = bgp_get_prefix(c, n->n.addr, c->add_path_tx ? path : 0); + px = bgp_get_prefix(c, n, c->add_path_tx ? path : 0); add_tail(&buck->prefixes, &px->buck_node); bgp_schedule_packet(p->conn, c, PKT_UPDATE); @@ -1874,42 +1878,52 @@ bgp_get_neighbor(rte *r) return as; /* If AS_PATH is not defined, we treat rte as locally originated */ - struct bgp_proto *p = (void *) r->attrs->src->proto; + struct bgp_proto *p = bgp_rte_proto(r); return p->cf->confederation ?: p->local_as; } static inline int rte_stale(rte *r) { - if (r->u.bgp.stale < 0) + if (r->pflags & BGP_REF_STALE) + return 1; + + if (r->pflags & BGP_REF_NOT_STALE) + return 0; + + /* If staleness is unknown, compute and cache it */ + eattr *a = ea_find(r->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY)); + if (a && int_set_contains(a->u.ptr, BGP_COMM_LLGR_STALE)) { - /* If staleness is unknown, compute and cache it */ - eattr *a = ea_find(r->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY)); - r->u.bgp.stale = a && int_set_contains(a->u.ptr, BGP_COMM_LLGR_STALE); + r->pflags |= BGP_REF_STALE; + return 1; + } + else + { + r->pflags |= BGP_REF_NOT_STALE; + return 0; } - - return r->u.bgp.stale; } int bgp_rte_better(rte *new, rte *old) { - struct bgp_proto *new_bgp = (struct bgp_proto *) new->attrs->src->proto; - struct bgp_proto *old_bgp = (struct bgp_proto *) old->attrs->src->proto; + struct bgp_proto *new_bgp = bgp_rte_proto(new); + struct bgp_proto *old_bgp = bgp_rte_proto(old); eattr *x, *y; u32 n, o; /* Skip suppressed routes (see bgp_rte_recalculate()) */ - n = new->u.bgp.suppressed; - o = old->u.bgp.suppressed; + n = new->pflags & BGP_REF_SUPPRESSED; + o = old->pflags & BGP_REF_SUPPRESSED; if (n > o) return 0; if (n < o) return 1; /* RFC 4271 9.1.2.1. Route resolvability test */ - n = rte_resolvable(new); - o = rte_resolvable(old); + n = rta_resolvable(new->attrs); + o = rta_resolvable(old->attrs); if (n > o) return 1; if (n < o) @@ -1934,8 +1948,8 @@ bgp_rte_better(rte *new, rte *old) return 0; /* RFC 7311 4.1 - Apply AIGP metric */ - u64 n2 = bgp_total_aigp_metric(new); - u64 o2 = bgp_total_aigp_metric(old); + u64 n2 = bgp_total_aigp_metric(new->attrs); + u64 o2 = bgp_total_aigp_metric(old->attrs); if (n2 < o2) return 1; if (n2 > o2) @@ -2039,21 +2053,18 @@ bgp_rte_better(rte *new, rte *old) int bgp_rte_mergable(rte *pri, rte *sec) { - struct bgp_proto *pri_bgp = (struct bgp_proto *) pri->attrs->src->proto; - struct bgp_proto *sec_bgp = (struct bgp_proto *) sec->attrs->src->proto; + struct bgp_proto *pri_bgp = bgp_rte_proto(pri); + struct bgp_proto *sec_bgp = bgp_rte_proto(sec); eattr *x, *y; u32 p, s; /* Skip suppressed routes (see bgp_rte_recalculate()) */ - if (pri->u.bgp.suppressed != sec->u.bgp.suppressed) + /* LLGR draft - depreference stale routes */ + if (pri->pflags != sec->pflags) return 0; /* RFC 4271 9.1.2.1. Route resolvability test */ - if (rte_resolvable(pri) != rte_resolvable(sec)) - return 0; - - /* LLGR draft - depreference stale routes */ - if (rte_stale(pri) != rte_stale(sec)) + if (rta_resolvable(pri->attrs) != rta_resolvable(sec->attrs)) return 0; /* Start with local preferences */ @@ -2118,24 +2129,23 @@ bgp_rte_mergable(rte *pri, rte *sec) static inline int same_group(rte *r, u32 lpref, u32 lasn) { - return (r->pref == lpref) && (bgp_get_neighbor(r) == lasn); + return (r->attrs->pref == lpref) && (bgp_get_neighbor(r) == lasn); } static inline int -use_deterministic_med(rte *r) +use_deterministic_med(struct rte_storage *r) { - struct proto *P = r->attrs->src->proto; - return (P->proto == &proto_bgp) && ((struct bgp_proto *) P)->cf->deterministic_med; + struct bgp_proto *p = bgp_rte_proto(&r->rte); + return p && p->cf->deterministic_med; } int -bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best) +bgp_rte_recalculate(rtable_private *table, net *net, rte *new, rte *old, rte *old_best) { - rte *r, *s; rte *key = new ? new : old; - u32 lpref = key->pref; + u32 lpref = key->attrs->pref; u32 lasn = bgp_get_neighbor(key); - int old_suppressed = old ? old->u.bgp.suppressed : 0; + int old_suppressed = old ? !!(old->pflags & BGP_REF_SUPPRESSED) : 0; /* * Proper RFC 4271 path selection is a bit complicated, it cannot be @@ -2187,11 +2197,11 @@ bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best) */ if (new) - new->u.bgp.suppressed = 1; + new->pflags |= BGP_REF_SUPPRESSED; if (old) { - old->u.bgp.suppressed = 1; + old->pflags |= BGP_REF_SUPPRESSED; /* The fast case - replace not best with worse (or remove not best) */ if (old_suppressed && !(new && bgp_rte_better(new, old))) @@ -2199,13 +2209,13 @@ bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best) } /* The default case - find a new best-in-group route */ - r = new; /* new may not be in the list */ - for (s=net->routes; rte_is_valid(s); s=s->next) - if (use_deterministic_med(s) && same_group(s, lpref, lasn)) + rte *r = new; /* new may not be in the list */ + for (struct rte_storage *s = net->routes; rte_is_valid(&s->rte); s = s->next) + if (use_deterministic_med(s) && same_group(&s->rte, lpref, lasn)) { - s->u.bgp.suppressed = 1; - if (!r || bgp_rte_better(s, r)) - r = s; + s->rte.pflags |= BGP_REF_SUPPRESSED; + if (!r || bgp_rte_better(&s->rte, r)) + r = &s->rte; } /* Simple case - the last route in group disappears */ @@ -2214,16 +2224,16 @@ bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best) /* Found if new is mergable with best-in-group */ if (new && (new != r) && bgp_rte_mergable(r, new)) - new->u.bgp.suppressed = 0; + new->pflags &= ~BGP_REF_SUPPRESSED; /* Found all existing routes mergable with best-in-group */ - for (s=net->routes; rte_is_valid(s); s=s->next) - if (use_deterministic_med(s) && same_group(s, lpref, lasn)) - if ((s != r) && bgp_rte_mergable(r, s)) - s->u.bgp.suppressed = 0; + for (struct rte_storage *s = net->routes; rte_is_valid(&s->rte); s = s->next) + if (use_deterministic_med(s) && same_group(&s->rte, lpref, lasn)) + if ((&s->rte != r) && bgp_rte_mergable(r, &s->rte)) + s->rte.pflags &= ~BGP_REF_SUPPRESSED; /* Found best-in-group */ - r->u.bgp.suppressed = 0; + r->pflags &= ~BGP_REF_SUPPRESSED; /* * There are generally two reasons why we have to force @@ -2255,25 +2265,44 @@ bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best) return !old_suppressed; } -struct rte * -bgp_rte_modify_stale(struct rte *r, struct linpool *pool) +void +bgp_rte_modify_stale(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *rpe UNUSED, rte **feed, uint count) { - eattr *a = ea_find(r->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY)); - const struct adata *ad = a ? a->u.ptr : NULL; - uint flags = a ? a->flags : BAF_PARTIAL; + struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req); - if (ad && int_set_contains(ad, BGP_COMM_NO_LLGR)) - return NULL; + do { + rte *r = feed[--count]; + if (r->sender != c->c.in_req.hook) + continue; + + /* A new route, do not mark as stale */ + if (r->stale_cycle == c->c.in_req.hook->stale_set) + continue; + + eattr *ea = ea_find(r->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_COMMUNITY)); + const struct adata *ad = ea ? ea->u.ptr : NULL; + uint flags = ea ? ea->flags : BAF_PARTIAL; + + rte e0 = *r; + e0.flags |= REF_USE_STALE; - if (ad && int_set_contains(ad, BGP_COMM_LLGR_STALE)) - return r; + if (ad && int_set_contains(ad, BGP_COMM_NO_LLGR)) + rte_import(&c->c.in_req, n, NULL, r->src); - r = rte_cow_rta(r, pool); - bgp_set_attr_ptr(&(r->attrs->eattrs), pool, BA_COMMUNITY, flags, - int_set_add(pool, ad, BGP_COMM_LLGR_STALE)); - r->u.bgp.stale = 1; + else if (ad && int_set_contains(ad, BGP_COMM_LLGR_STALE)) + rte_import(&c->c.in_req, n, &e0, r->src); - return r; + else { + rta *a = e0.attrs = rta_do_cow(r->attrs, bgp_linpool); + + bgp_set_attr_ptr(&(a->eattrs), bgp_linpool, BA_COMMUNITY, flags, + int_set_add(bgp_linpool, ad, BGP_COMM_LLGR_STALE)); + e0.pflags |= BGP_REF_STALE; + + rte_import(&c->c.in_req, n, &e0, r->src); + lp_flush(bgp_linpool); + } + } while (count); } @@ -2356,22 +2385,22 @@ bgp_get_route_info(rte *e, byte *buf) eattr *o = ea_find(e->attrs->eattrs, EA_CODE(PROTOCOL_BGP, BA_ORIGIN)); u32 origas; - buf += bsprintf(buf, " (%d", e->pref); + buf += bsprintf(buf, " (%d", e->attrs->pref); - if (e->u.bgp.suppressed) + if (e->pflags & BGP_REF_SUPPRESSED) buf += bsprintf(buf, "-"); if (rte_stale(e)) buf += bsprintf(buf, "s"); - u64 metric = bgp_total_aigp_metric(e); + u64 metric = bgp_total_aigp_metric(e->attrs); if (metric < BGP_AIGP_MAX) { buf += bsprintf(buf, "/%lu", metric); } else if (e->attrs->igp_metric) { - if (!rte_resolvable(e)) + if (!rta_resolvable(e->attrs)) buf += bsprintf(buf, "/-"); else if (e->attrs->igp_metric >= IGP_METRIC_UNKNOWN) buf += bsprintf(buf, "/?"); diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c index e4d754b1..b825f778 100644 --- a/proto/bgp/bgp.c +++ b/proto/bgp/bgp.c @@ -140,6 +140,15 @@ static void bgp_update_bfd(struct bgp_proto *p, const struct bfd_options *bfd); static int bgp_incoming_connection(sock *sk, uint dummy UNUSED); static void bgp_listen_sock_err(sock *sk UNUSED, int err); +static void bgp_graceful_restart_feed(struct bgp_channel *c); +static inline void channel_refresh_end_reload(struct channel *c) +{ + channel_refresh_end(c); + + if (c->in_table) + channel_request_reload(c); +} + /** * bgp_open - open a BGP instance * @p: BGP instance @@ -575,6 +584,7 @@ bgp_conn_enter_established_state(struct bgp_conn *conn) p->link_addr = p->neigh->iface->llv6->ip; conn->sk->fast_rx = 0; + conn->sk->cork = &rt_cork; p->conn = conn; p->last_error_class = 0; @@ -775,25 +785,25 @@ bgp_handle_graceful_restart(struct bgp_proto *p) { case BGP_GRS_NONE: c->gr_active = BGP_GRS_ACTIVE; - rt_refresh_begin(c->c.table, &c->c); + channel_refresh_begin(&c->c); break; case BGP_GRS_ACTIVE: - rt_refresh_end(c->c.table, &c->c); - rt_refresh_begin(c->c.table, &c->c); + channel_refresh_end(&c->c); + channel_refresh_begin(&c->c); break; case BGP_GRS_LLGR: - rt_refresh_begin(c->c.table, &c->c); - rt_modify_stale(c->c.table, &c->c); + channel_refresh_begin(&c->c); + bgp_graceful_restart_feed(c); break; } } else { /* Just flush the routes */ - rt_refresh_begin(c->c.table, &c->c); - rt_refresh_end(c->c.table, &c->c); + channel_refresh_begin(&c->c); + channel_refresh_end(&c->c); } /* Reset bucket and prefix tables */ @@ -811,6 +821,51 @@ bgp_handle_graceful_restart(struct bgp_proto *p) tm_start(p->gr_timer, p->conn->remote_caps->gr_time S); } +static void +bgp_graceful_restart_feed_done(struct rt_export_request *req) +{ + req->hook = NULL; +} + +static void +bgp_graceful_restart_feed_dump_req(struct rt_export_request *req) +{ + struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req); + debug(" BGP-GR %s.%s export request %p\n", c->c.proto->name, c->c.name, req); +} + +static void +bgp_graceful_restart_feed_log_state_change(struct rt_export_request *req, u8 state) +{ + struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req); + struct bgp_proto *p = (void *) c->c.proto; + BGP_TRACE(D_EVENTS, "Long-lived graceful restart export state changed to %s", rt_export_state_name(state)); + + if (state == TES_READY) + rt_stop_export(req, bgp_graceful_restart_feed_done); +} + +static void +bgp_graceful_restart_drop_export(struct rt_export_request *req UNUSED, const net_addr *n UNUSED, struct rt_pending_export *rpe UNUSED) +{ /* Nothing to do */ } + +static void +bgp_graceful_restart_feed(struct bgp_channel *c) +{ + c->stale_feed = (struct rt_export_request) { + .name = "BGP-GR", + .list = &global_work_list, + .trace_routes = c->c.debug | c->c.proto->debug, + .dump_req = bgp_graceful_restart_feed_dump_req, + .log_state_change = bgp_graceful_restart_feed_log_state_change, + .export_bulk = bgp_rte_modify_stale, + .export_one = bgp_graceful_restart_drop_export, + }; + + rt_request_export(c->c.table, &c->stale_feed); +} + + /** * bgp_graceful_restart_done - finish active BGP graceful restart * @c: BGP channel @@ -833,8 +888,11 @@ bgp_graceful_restart_done(struct bgp_channel *c) if (!p->gr_active_num) BGP_TRACE(D_EVENTS, "Neighbor graceful restart done"); + if (c->stale_feed.hook) + rt_stop_export(&c->stale_feed, bgp_graceful_restart_feed_done); + tm_stop(c->stale_timer); - rt_refresh_end(c->c.table, &c->c); + channel_refresh_end_reload(&c->c); } /** @@ -876,7 +934,7 @@ bgp_graceful_restart_timeout(timer *t) /* Channel is in GR, and supports LLGR -> start LLGR */ c->gr_active = BGP_GRS_LLGR; tm_start(c->stale_timer, c->stale_time S); - rt_modify_stale(c->c.table, &c->c); + bgp_graceful_restart_feed(c); } } else @@ -913,11 +971,11 @@ bgp_refresh_begin(struct bgp_channel *c) if (c->load_state == BFS_LOADING) { log(L_WARN "%s: BEGIN-OF-RR received before END-OF-RIB, ignoring", p->p.name); return; } - c->load_state = BFS_REFRESHING; - rt_refresh_begin(c->c.table, &c->c); + if (c->load_state == BFS_REFRESHING) + channel_refresh_end(&c->c); - if (c->c.in_table) - rt_refresh_begin(c->c.in_table, &c->c); + c->load_state = BFS_REFRESHING; + channel_refresh_begin(&c->c); } /** @@ -938,10 +996,7 @@ bgp_refresh_end(struct bgp_channel *c) { log(L_WARN "%s: END-OF-RR received without prior BEGIN-OF-RR, ignoring", p->p.name); return; } c->load_state = BFS_NONE; - rt_refresh_end(c->c.table, &c->c); - - if (c->c.in_table) - rt_prune_sync(c->c.in_table, 0); + channel_refresh_end_reload(&c->c); } @@ -1408,12 +1463,9 @@ bgp_reload_routes(struct channel *C) struct bgp_proto *p = (void *) C->proto; struct bgp_channel *c = (void *) C; - ASSERT(p->conn && (p->route_refresh || c->c.in_table)); + ASSERT(p->conn && (p->route_refresh)); - if (c->c.in_table) - channel_schedule_reload(C); - else - bgp_schedule_packet(p->conn, c, PKT_ROUTE_REFRESH); + bgp_schedule_packet(p->conn, c, PKT_ROUTE_REFRESH); } static void @@ -1677,6 +1729,13 @@ done: return p->p.proto_state; } +struct rte_owner_class bgp_rte_owner_class = { + .get_route_info = bgp_get_route_info, + .rte_better = bgp_rte_better, + .rte_mergable = bgp_rte_mergable, + .rte_igp_metric = bgp_rte_igp_metric, +}; + static struct proto * bgp_init(struct proto_config *CF) { @@ -1690,10 +1749,9 @@ bgp_init(struct proto_config *CF) P->reload_routes = bgp_reload_routes; P->feed_begin = bgp_feed_begin; P->feed_end = bgp_feed_end; - P->rte_better = bgp_rte_better; - P->rte_mergable = bgp_rte_mergable; - P->rte_recalculate = cf->deterministic_med ? bgp_rte_recalculate : NULL; - P->rte_modify = bgp_rte_modify_stale; + + P->sources.class = &bgp_rte_owner_class; + P->sources.rte_recalculate = cf->deterministic_med ? bgp_rte_recalculate : NULL; p->cf = cf; p->is_internal = (cf->local_as == cf->remote_as); @@ -1745,17 +1803,19 @@ bgp_channel_start(struct channel *C) ip_addr src = p->local_ip; if (c->igp_table_ip4) - rt_lock_table(c->igp_table_ip4); + RT_LOCKED(c->igp_table_ip4, t) + rt_lock_table(t); if (c->igp_table_ip6) - rt_lock_table(c->igp_table_ip6); + RT_LOCKED(c->igp_table_ip6, t) + rt_lock_table(t); c->pool = p->p.pool; // XXXX bgp_init_bucket_table(c); bgp_init_prefix_table(c); if (c->cf->import_table) - channel_setup_in_table(C); + channel_setup_in_table(C, 0); if (c->cf->export_table) channel_setup_out_table(C); @@ -1829,10 +1889,12 @@ bgp_channel_cleanup(struct channel *C) struct bgp_channel *c = (void *) C; if (c->igp_table_ip4) - rt_unlock_table(c->igp_table_ip4); + RT_LOCKED(c->igp_table_ip4, t) + rt_unlock_table(t); if (c->igp_table_ip6) - rt_unlock_table(c->igp_table_ip6); + RT_LOCKED(c->igp_table_ip6, t) + rt_unlock_table(t); c->index = 0; @@ -2556,6 +2618,5 @@ struct protocol proto_bgp = { .copy_config = bgp_copy_config, .get_status = bgp_get_status, .get_attr = bgp_get_attr, - .get_route_info = bgp_get_route_info, .show_proto_info = bgp_show_proto_info }; diff --git a/proto/bgp/bgp.h b/proto/bgp/bgp.h index cca4b448..60f93bce 100644 --- a/proto/bgp/bgp.h +++ b/proto/bgp/bgp.h @@ -200,6 +200,10 @@ struct bgp_channel_config { #define BGP_BFD_GRACEFUL 2 /* BFD down triggers graceful restart */ +/* rte->pflags */ +#define BGP_REF_SUPPRESSED 0x1 /* Used for deterministic MED comparison */ +#define BGP_REF_STALE 0x2 /* Route is LLGR_STATE */ +#define BGP_REF_NOT_STALE 0x4 /* Route is NOT LLGR_STATE */ struct bgp_af_caps { u32 afi; @@ -362,6 +366,7 @@ struct bgp_channel { timer *stale_timer; /* Long-lived stale timer for LLGR */ u32 stale_time; /* Stored LLGR stale time from last session */ + struct rt_export_request stale_feed; /* Feeder request for stale route modification */ u8 add_path_rx; /* Session expects receive of ADD-PATH extended NLRI */ u8 add_path_tx; /* Session expects transmit of ADD-PATH extended NLRI */ @@ -513,11 +518,12 @@ struct rte_source *bgp_find_source(struct bgp_proto *p, u32 path_id); struct rte_source *bgp_get_source(struct bgp_proto *p, u32 path_id); static inline int -rte_resolvable(rte *rt) +rta_resolvable(rta *a) { - return rt->attrs->dest == RTD_UNICAST; + return a->dest == RTD_UNICAST; } +extern struct rte_owner_class bgp_rte_owner_class; #ifdef LOCAL_DEBUG #define BGP_FORCE_DEBUG 1 @@ -580,24 +586,31 @@ void bgp_free_prefix(struct bgp_channel *c, struct bgp_prefix *bp); int bgp_rte_better(struct rte *, struct rte *); int bgp_rte_mergable(rte *pri, rte *sec); -int bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best); -struct rte *bgp_rte_modify_stale(struct rte *r, struct linpool *pool); -void bgp_rt_notify(struct proto *P, struct channel *C, net *n, rte *new, rte *old); -int bgp_preexport(struct proto *, struct rte **, struct linpool *); +int bgp_rte_recalculate(rtable_private *table, net *net, rte *new, rte *old, rte *old_best); +void bgp_rte_modify_stale(struct rt_export_request *, const net_addr *, struct rt_pending_export *, rte **, uint); +u32 bgp_rte_igp_metric(struct rte *); +void bgp_rt_notify(struct proto *P, struct channel *C, const net_addr *n, rte *new, const rte *old); +int bgp_preexport(struct channel *, struct rte *); int bgp_get_attr(const struct eattr *e, byte *buf, int buflen); -void bgp_get_route_info(struct rte *, byte *buf); -int bgp_total_aigp_metric_(rte *e, u64 *metric, const struct adata **ad); +void bgp_get_route_info(struct rte *, byte *); +int bgp_total_aigp_metric_(rta *a, u64 *metric, const struct adata **ad); + +static inline struct bgp_proto *bgp_rte_proto(struct rte *rte) +{ + return (rte->src->owner->class == &bgp_rte_owner_class) ? + SKIP_BACK(struct bgp_proto, p.sources, rte->src->owner) : NULL; +} #define BGP_AIGP_METRIC 1 #define BGP_AIGP_MAX U64(0xffffffffffffffff) static inline u64 -bgp_total_aigp_metric(rte *r) +bgp_total_aigp_metric(rta *a) { u64 metric = BGP_AIGP_MAX; const struct adata *ad; - bgp_total_aigp_metric_(r, &metric, &ad); + bgp_total_aigp_metric_(a, &metric, &ad); return metric; } diff --git a/proto/bgp/packets.c b/proto/bgp/packets.c index 99b5d5b4..ea9adb4c 100644 --- a/proto/bgp/packets.c +++ b/proto/bgp/packets.c @@ -971,7 +971,7 @@ bgp_apply_next_hop(struct bgp_parse_state *s, rta *a, ip_addr gw, ip_addr ll) s->hostentry = rt_get_hostentry(tab, gw, ll, c->c.table); if (!s->mpls) - rta_apply_hostentry(a, s->hostentry, NULL); + rta_apply_hostentry(a, s->hostentry, NULL, s->pool); /* With MPLS, hostentry is applied later in bgp_apply_mpls_labels() */ } @@ -1005,7 +1005,7 @@ bgp_apply_mpls_labels(struct bgp_parse_state *s, rta *a, u32 *labels, uint lnum) ms.len = lnum; memcpy(ms.stack, labels, 4*lnum); - rta_apply_hostentry(a, s->hostentry, &ms); + rta_apply_hostentry(a, s->hostentry, &ms, s->pool); } } @@ -1339,6 +1339,8 @@ bgp_rte_update(struct bgp_parse_state *s, net_addr *n, u32 path_id, rta *a0) { if (path_id != s->last_id) { + rt_unlock_source(s->last_src); + s->last_src = rt_get_source(&s->proto->p, path_id); s->last_id = path_id; @@ -1349,28 +1351,25 @@ bgp_rte_update(struct bgp_parse_state *s, net_addr *n, u32 path_id, rta *a0) if (!a0) { /* Route withdraw */ - rte_update3(&s->channel->c, n, NULL, s->last_src); + rte_update(&s->channel->c, n, NULL, s->last_src); return; } /* Prepare cached route attributes */ if (s->cached_rta == NULL) { - a0->src = s->last_src; - /* Workaround for rta_lookup() breaking eattrs */ ea_list *ea = a0->eattrs; s->cached_rta = rta_lookup(a0); a0->eattrs = ea; } - rta *a = rta_clone(s->cached_rta); - rte *e = rte_get_temp(a); + rte e0 = { + .attrs = s->cached_rta, + .src = s->last_src, + }; - e->pflags = 0; - e->u.bgp.suppressed = 0; - e->u.bgp.stale = -1; - rte_update3(&s->channel->c, n, e, s->last_src); + rte_update(&s->channel->c, n, &e0, s->last_src); } static void @@ -2424,6 +2423,7 @@ bgp_decode_nlri(struct bgp_parse_state *s, u32 afi, byte *nlri, uint len, ea_lis s->last_id = 0; s->last_src = s->proto->p.main_source; + rt_lock_source(s->last_src); /* * IPv4 BGP and MP-BGP may be used together in one update, therefore we do not @@ -2440,6 +2440,7 @@ bgp_decode_nlri(struct bgp_parse_state *s, u32 afi, byte *nlri, uint len, ea_lis a->scope = SCOPE_UNIVERSE; a->from = s->proto->remote_ip; a->eattrs = ea; + a->pref = c->c.preference; c->desc->decode_next_hop(s, nh, nh_len, a); bgp_finish_attrs(s, a); @@ -2453,6 +2454,8 @@ bgp_decode_nlri(struct bgp_parse_state *s, u32 afi, byte *nlri, uint len, ea_lis rta_free(s->cached_rta); s->cached_rta = NULL; + + rt_unlock_source(s->last_src); } static void diff --git a/proto/mrt/mrt.c b/proto/mrt/mrt.c index 8d97c860..e12f7743 100644 --- a/proto/mrt/mrt.c +++ b/proto/mrt/mrt.c @@ -113,13 +113,13 @@ mrt_buffer_flush(buffer *b) } #define MRT_DEFINE_TYPE(S, T) \ - static inline void mrt_put_##S##_(buffer *b, T x) \ + UNUSED static inline void mrt_put_##S##_(buffer *b, T x) \ { \ put_##S(b->pos, x); \ b->pos += sizeof(T); \ } \ \ - static inline void mrt_put_##S(buffer *b, T x) \ + UNUSED static inline void mrt_put_##S(buffer *b, T x) \ { \ mrt_buffer_need(b, sizeof(T)); \ put_##S(b->pos, x); \ @@ -228,7 +228,7 @@ mrt_next_table_(rtable *tab, rtable *tab_ptr, const char *pattern) NODE_VALID(tn); tn = tn->next) { - tab = SKIP_BACK(struct rtable, n, tn); + tab = SKIP_BACK(rtable, n, tn); if (patmatch(pattern, tab->name) && ((tab->addr_type == NET_IP4) || (tab->addr_type == NET_IP6))) return tab; @@ -243,13 +243,21 @@ mrt_next_table(struct mrt_table_dump_state *s) rtable *tab = mrt_next_table_(s->table, s->table_ptr, s->table_expr); if (s->table) - rt_unlock_table(s->table); + { + RT_LOCK(s->table); + rt_unlock_table(RT_PRIV(s->table)); + RT_UNLOCK(s->table); + } s->table = tab; s->ipv4 = tab ? (tab->addr_type == NET_IP4) : 0; if (s->table) - rt_lock_table(s->table); + { + RT_LOCK(s->table); + rt_lock_table(RT_PRIV(s->table)); + RT_UNLOCK(s->table); + } return s->table; } @@ -460,7 +468,7 @@ mrt_rib_table_entry_bgp_attrs(struct mrt_table_dump_state *s, rte *r) return; fail: - mrt_log(s, "Attribute list too long for %N", r->net->n.addr); + mrt_log(s, "Attribute list too long for %N", r->net); } #endif @@ -472,9 +480,9 @@ mrt_rib_table_entry(struct mrt_table_dump_state *s, rte *r) #ifdef CONFIG_BGP /* Find peer index */ - if (r->attrs->src->proto->proto == &proto_bgp) + struct bgp_proto *p = bgp_rte_proto(r); + if (p) { - struct bgp_proto *p = (void *) r->attrs->src->proto; struct mrt_peer_entry *n = HASH_FIND(s->peer_hash, PEER, p->remote_id, p->remote_as, p->remote_ip); @@ -488,7 +496,7 @@ mrt_rib_table_entry(struct mrt_table_dump_state *s, rte *r) /* Path Identifier */ if (s->add_path) - mrt_put_u32(b, r->attrs->src->private_id); + mrt_put_u32(b, r->src->private_id); /* Route Attributes */ mrt_put_u16(b, 0); @@ -512,26 +520,21 @@ mrt_rib_table_dump(struct mrt_table_dump_state *s, net *n, int add_path) mrt_init_message(&s->buf, MRT_TABLE_DUMP_V2, subtype); mrt_rib_table_header(s, n->n.addr); - rte *rt, *rt0; - for (rt0 = n->routes; rt = rt0; rt0 = rt0->next) + for (struct rte_storage *rt, *rt0 = n->routes; rt = rt0; rt0 = rt0->next) { - if (rte_is_filtered(rt)) + if (rte_is_filtered(&rt->rte)) continue; /* Skip routes that should be reported in the other phase */ - if (!s->always_add_path && (!rt->attrs->src->private_id != !s->add_path)) + if (!s->always_add_path && (!rt->rte.src->private_id != !s->add_path)) { s->want_add_path = 1; continue; } - rte_make_tmp_attrs(&rt, s->linpool, NULL); - - if (f_run(s->filter, &rt, s->linpool, 0) <= F_ACCEPT) - mrt_rib_table_entry(s, rt); - - if (rt != rt0) - rte_free(rt); + rte e = rt->rte; + if (f_run(s->filter, &e, s->linpool, 0) <= F_ACCEPT) + mrt_rib_table_entry(s, &e); lp_flush(s->linpool); } @@ -558,10 +561,11 @@ mrt_rib_table_dump(struct mrt_table_dump_state *s, net *n, int add_path) static struct mrt_table_dump_state * mrt_table_dump_init(pool *pp) { - pool *pool = rp_new(pp, "MRT Table Dump"); + pool *pool = rp_new(pp, &main_birdloop, "MRT Table Dump"); struct mrt_table_dump_state *s = mb_allocz(pool, sizeof(struct mrt_table_dump_state)); s->pool = pool; + s->parent = pp; s->linpool = lp_new(pool, 4080); s->peer_lp = lp_new(pool, 4080); mrt_buffer_init(&s->buf, pool, 2 * MRT_ATTR_BUFFER_SIZE); @@ -578,18 +582,27 @@ mrt_table_dump_init(pool *pp) static void mrt_table_dump_free(struct mrt_table_dump_state *s) { - if (s->table_open) - FIB_ITERATE_UNLINK(&s->fit, &s->table->fib); - if (s->table) - rt_unlock_table(s->table); + { + RT_LOCK(s->table); + + if (s->table_open) + FIB_ITERATE_UNLINK(&s->fit, &RT_PRIV(s->table)->fib); + + rt_unlock_table(RT_PRIV(s->table)); + RT_UNLOCK(s->table); + } if (s->table_ptr) - rt_unlock_table(s->table_ptr); + { + RT_LOCK(s->table_ptr); + rt_unlock_table(RT_PRIV(s->table_ptr)); + RT_UNLOCK(s->table_ptr); + } config_del_obstacle(s->config); - rfree(s->pool); + rp_free(s->pool, s->parent); } @@ -601,8 +614,14 @@ mrt_table_dump_step(struct mrt_table_dump_state *s) s->max = 2048; s->bws = &bws; + rtable_private *tab; + if (s->table_open) + { + RT_LOCK(s->table); + tab = RT_PRIV(s->table); goto step; + } while (mrt_next_table(s)) { @@ -611,15 +630,18 @@ mrt_table_dump_step(struct mrt_table_dump_state *s) mrt_peer_table_dump(s); - FIB_ITERATE_INIT(&s->fit, &s->table->fib); + RT_LOCK(s->table); + tab = RT_PRIV(s->table); + FIB_ITERATE_INIT(&s->fit, &tab->fib); s->table_open = 1; step: - FIB_ITERATE_START(&s->table->fib, &s->fit, net, n) + FIB_ITERATE_START(&tab->fib, &s->fit, net, n) { if (s->max < 0) { FIB_ITERATE_PUT(&s->fit); + RT_UNLOCK(s->table); return 0; } @@ -639,6 +661,7 @@ mrt_table_dump_step(struct mrt_table_dump_state *s) mrt_peer_table_flush(s); } + RT_UNLOCK(s->table); return 1; } @@ -666,7 +689,11 @@ mrt_timer(timer *t) s->always_add_path = cf->always_add_path; if (s->table_ptr) - rt_lock_table(s->table_ptr); + { + RT_LOCK(s->table_ptr); + rt_lock_table(RT_PRIV(s->table_ptr)); + RT_UNLOCK(s->table_ptr); + } p->table_dump = s; ev_schedule(p->event); @@ -739,7 +766,11 @@ mrt_dump_cmd(struct mrt_dump_data *d) s->filename = d->filename; if (s->table_ptr) - rt_lock_table(s->table_ptr); + { + RT_LOCK(s->table_ptr); + rt_lock_table(RT_PRIV(s->table_ptr)); + RT_UNLOCK(s->table_ptr); + } this_cli->cont = mrt_dump_cont; this_cli->cleanup = mrt_dump_cleanup; diff --git a/proto/mrt/mrt.h b/proto/mrt/mrt.h index 4ff94c12..2e616f6f 100644 --- a/proto/mrt/mrt.h +++ b/proto/mrt/mrt.h @@ -40,7 +40,7 @@ struct mrt_proto { struct mrt_dump_data { const char *table_expr; - struct rtable *table_ptr; + rtable *table_ptr; const struct filter *filter; const char *filename; }; @@ -60,20 +60,21 @@ struct mrt_table_dump_state { /* Configuration information */ const char *table_expr; /* Wildcard for table name (or NULL) */ - struct rtable *table_ptr; /* Explicit table (or NULL) */ + rtable *table_ptr; /* Explicit table (or NULL) */ const struct filter *filter; /* Optional filter */ const char *filename; /* Filename pattern */ int always_add_path; /* Always use *_ADDPATH message subtypes */ /* Allocated by mrt_table_dump_init() */ pool *pool; /* Pool for table dump */ + pool *parent; /* Parent pool for cleanup */ linpool *linpool; /* Temporary linear pool */ linpool *peer_lp; /* Linear pool for peer entries in peer_hash */ buffer buf; /* Buffer for MRT messages */ HASH(struct mrt_peer_entry) peer_hash; /* Hash for peers to find the index */ - struct rtable *table; /* Processed table, NULL initially */ + rtable *table; /* Processed table, NULL initially */ struct fib_iterator fit; /* Iterator in processed table */ int table_open; /* Whether iterator is linked */ diff --git a/proto/ospf/iface.c b/proto/ospf/iface.c index f38b8210..2aae341a 100644 --- a/proto/ospf/iface.c +++ b/proto/ospf/iface.c @@ -311,7 +311,7 @@ ospf_iface_remove(struct ospf_iface *ifa) ospf_iface_sm(ifa, ISM_DOWN); rem_node(NODE ifa); - rfree(ifa->pool); + rp_free(ifa->pool, p->p.pool); } void @@ -522,7 +522,10 @@ static inline void add_nbma_node(struct ospf_iface *ifa, struct nbma_node *src, int found) { struct nbma_node *n = mb_alloc(ifa->pool, sizeof(struct nbma_node)); + + n->n = (node) {}; add_tail(&ifa->nbma_list, NODE n); + n->ip = src->ip; n->eligible = src->eligible; n->found = found; @@ -564,7 +567,7 @@ ospf_iface_new(struct ospf_area *oa, struct ifa *addr, struct ospf_iface_patt *i OSPF_TRACE(D_EVENTS, "Adding interface %s (%N) to area %R", iface->name, &addr->prefix, oa->areaid); - pool = rp_new(p->p.pool, "OSPF Interface"); + pool = rp_new(p->p.pool, p->p.loop, "OSPF Interface"); ifa = mb_allocz(pool, sizeof(struct ospf_iface)); ifa->iface = iface; ifa->addr = addr; @@ -684,7 +687,7 @@ ospf_iface_new_vlink(struct ospf_proto *p, struct ospf_iface_patt *ip) /* Vlink ifname is stored just after the ospf_iface structure */ - pool = rp_new(p->p.pool, "OSPF Vlink"); + pool = rp_new(p->p.pool, p->p.loop, "OSPF Vlink"); ifa = mb_allocz(pool, sizeof(struct ospf_iface) + 16); ifa->oa = p->backbone; ifa->cf = ip; diff --git a/proto/ospf/neighbor.c b/proto/ospf/neighbor.c index ca369819..48e38556 100644 --- a/proto/ospf/neighbor.c +++ b/proto/ospf/neighbor.c @@ -80,7 +80,7 @@ struct ospf_neighbor * ospf_neighbor_new(struct ospf_iface *ifa) { struct ospf_proto *p = ifa->oa->po; - struct pool *pool = rp_new(p->p.pool, "OSPF Neighbor"); + struct pool *pool = rp_new(p->p.pool, p->p.loop, "OSPF Neighbor"); struct ospf_neighbor *n = mb_allocz(pool, sizeof(struct ospf_neighbor)); n->pool = pool; @@ -120,7 +120,7 @@ ospf_neigh_down(struct ospf_neighbor *n) s_get(&(n->dbsi)); release_lsrtl(p, n); rem_node(NODE n); - rfree(n->pool); + rp_free(n->pool, p->p.pool); OSPF_TRACE(D_EVENTS, "Neighbor %R on %s removed", rid, ifa->ifname); } diff --git a/proto/ospf/ospf.c b/proto/ospf/ospf.c index ba8c2e2b..16774df6 100644 --- a/proto/ospf/ospf.c +++ b/proto/ospf/ospf.c @@ -107,12 +107,10 @@ #include <stdlib.h> #include "ospf.h" -static int ospf_preexport(struct proto *P, rte **new, struct linpool *pool); -static void ospf_make_tmp_attrs(struct rte *rt, struct linpool *pool); -static void ospf_store_tmp_attrs(struct rte *rt, struct linpool *pool); +static int ospf_preexport(struct channel *C, rte *new); static void ospf_reload_routes(struct channel *C); static int ospf_rte_better(struct rte *new, struct rte *old); -static int ospf_rte_same(struct rte *new, struct rte *old); +static u32 ospf_rte_igp_metric(struct rte *rt); static void ospf_disp(timer *timer); @@ -378,10 +376,8 @@ ospf_init(struct proto_config *CF) P->reload_routes = ospf_reload_routes; P->feed_begin = ospf_feed_begin; P->feed_end = ospf_feed_end; - P->make_tmp_attrs = ospf_make_tmp_attrs; - P->store_tmp_attrs = ospf_store_tmp_attrs; - P->rte_better = ospf_rte_better; - P->rte_same = ospf_rte_same; + + P->sources.class = &ospf_rte_owner_class; return P; } @@ -390,7 +386,9 @@ ospf_init(struct proto_config *CF) static int ospf_rte_better(struct rte *new, struct rte *old) { - if (new->u.ospf.metric1 == LSINFINITY) + u32 new_metric1 = ea_get_int(new->attrs->eattrs, EA_OSPF_METRIC1, LSINFINITY); + + if (new_metric1 == LSINFINITY) return 0; if(new->attrs->source < old->attrs->source) return 1; @@ -398,27 +396,27 @@ ospf_rte_better(struct rte *new, struct rte *old) if(new->attrs->source == RTS_OSPF_EXT2) { - if(new->u.ospf.metric2 < old->u.ospf.metric2) return 1; - if(new->u.ospf.metric2 > old->u.ospf.metric2) return 0; + u32 old_metric2 = ea_get_int(old->attrs->eattrs, EA_OSPF_METRIC2, LSINFINITY); + u32 new_metric2 = ea_get_int(new->attrs->eattrs, EA_OSPF_METRIC2, LSINFINITY); + if(new_metric2 < old_metric2) return 1; + if(new_metric2 > old_metric2) return 0; } - if (new->u.ospf.metric1 < old->u.ospf.metric1) + u32 old_metric1 = ea_get_int(old->attrs->eattrs, EA_OSPF_METRIC1, LSINFINITY); + if (new_metric1 < old_metric1) return 1; return 0; /* Old is shorter or same */ } -static int -ospf_rte_same(struct rte *new, struct rte *old) +static u32 +ospf_rte_igp_metric(struct rte *rt) { - /* new->attrs == old->attrs always */ - return - new->u.ospf.metric1 == old->u.ospf.metric1 && - new->u.ospf.metric2 == old->u.ospf.metric2 && - new->u.ospf.tag == old->u.ospf.tag && - new->u.ospf.router_id == old->u.ospf.router_id; -} + if (rt->attrs->source == RTS_OSPF_EXT2) + return IGP_METRIC_UNKNOWN; + return ea_get_int(rt->attrs->eattrs, EA_OSPF_METRIC1, LSINFINITY); +} void ospf_schedule_rtcalc(struct ospf_proto *p) @@ -484,14 +482,13 @@ ospf_disp(timer * timer) * import to the filters. */ static int -ospf_preexport(struct proto *P, rte **new, struct linpool *pool UNUSED) +ospf_preexport(struct channel *c, rte *e) { - struct ospf_proto *p = (struct ospf_proto *) P; + struct ospf_proto *p = (struct ospf_proto *) c->proto; struct ospf_area *oa = ospf_main_area(p); - rte *e = *new; /* Reject our own routes */ - if (e->attrs->src->proto == P) + if (e->sender == c->in_req.hook) return -1; /* Do not export routes to stub areas */ @@ -501,26 +498,6 @@ ospf_preexport(struct proto *P, rte **new, struct linpool *pool UNUSED) return 0; } -static void -ospf_make_tmp_attrs(struct rte *rt, struct linpool *pool) -{ - rte_init_tmp_attrs(rt, pool, 4); - rte_make_tmp_attr(rt, EA_OSPF_METRIC1, EAF_TYPE_INT, rt->u.ospf.metric1); - rte_make_tmp_attr(rt, EA_OSPF_METRIC2, EAF_TYPE_INT, rt->u.ospf.metric2); - rte_make_tmp_attr(rt, EA_OSPF_TAG, EAF_TYPE_INT, rt->u.ospf.tag); - rte_make_tmp_attr(rt, EA_OSPF_ROUTER_ID, EAF_TYPE_ROUTER_ID, rt->u.ospf.router_id); -} - -static void -ospf_store_tmp_attrs(struct rte *rt, struct linpool *pool) -{ - rte_init_tmp_attrs(rt, pool, 4); - rt->u.ospf.metric1 = rte_store_tmp_attr(rt, EA_OSPF_METRIC1); - rt->u.ospf.metric2 = rte_store_tmp_attr(rt, EA_OSPF_METRIC2); - rt->u.ospf.tag = rte_store_tmp_attr(rt, EA_OSPF_TAG); - rt->u.ospf.router_id = rte_store_tmp_attr(rt, EA_OSPF_ROUTER_ID); -} - /** * ospf_shutdown - Finish of OSPF instance * @P: OSPF protocol instance @@ -558,6 +535,9 @@ ospf_shutdown(struct proto *P) } FIB_WALK_END; + if (tm_active(p->disp_timer)) + tm_stop(p->disp_timer); + return PS_DOWN; } @@ -607,16 +587,20 @@ ospf_get_route_info(rte * rte, byte * buf) } buf += bsprintf(buf, " %s", type); - buf += bsprintf(buf, " (%d/%d", rte->pref, rte->u.ospf.metric1); + buf += bsprintf(buf, " (%d/%d", rte->attrs->pref, ea_get_int(rte->attrs->eattrs, EA_OSPF_METRIC1, LSINFINITY)); if (rte->attrs->source == RTS_OSPF_EXT2) - buf += bsprintf(buf, "/%d", rte->u.ospf.metric2); + buf += bsprintf(buf, "/%d", ea_get_int(rte->attrs->eattrs, EA_OSPF_METRIC2, LSINFINITY)); buf += bsprintf(buf, ")"); - if ((rte->attrs->source == RTS_OSPF_EXT1 || rte->attrs->source == RTS_OSPF_EXT2) && rte->u.ospf.tag) + if (rte->attrs->source == RTS_OSPF_EXT1 || rte->attrs->source == RTS_OSPF_EXT2) { - buf += bsprintf(buf, " [%x]", rte->u.ospf.tag); + eattr *ea = ea_find(rte->attrs->eattrs, EA_OSPF_TAG); + if (ea && (ea->u.data > 0)) + buf += bsprintf(buf, " [%x]", ea->u.data); } - if (rte->u.ospf.router_id) - buf += bsprintf(buf, " [%R]", rte->u.ospf.router_id); + + eattr *ea = ea_find(rte->attrs->eattrs, EA_OSPF_ROUTER_ID); + if (ea) + buf += bsprintf(buf, " [%R]", ea->u.data); } static int @@ -1533,6 +1517,12 @@ ospf_sh_lsadb(struct lsadb_show_data *ld) } +struct rte_owner_class ospf_rte_owner_class = { + .get_route_info = ospf_get_route_info, + .rte_better = ospf_rte_better, + .rte_igp_metric = ospf_rte_igp_metric, +}; + struct protocol proto_ospf = { .name = "OSPF", .template = "ospf%d", @@ -1548,5 +1538,4 @@ struct protocol proto_ospf = { .reconfigure = ospf_reconfigure, .get_status = ospf_get_status, .get_attr = ospf_get_attr, - .get_route_info = ospf_get_route_info }; diff --git a/proto/ospf/ospf.h b/proto/ospf/ospf.h index 3e704ae8..a5f83e79 100644 --- a/proto/ospf/ospf.h +++ b/proto/ospf/ospf.h @@ -1007,6 +1007,8 @@ void ospf_sh_state(struct proto *P, int verbose, int reachable); void ospf_sh_lsadb(struct lsadb_show_data *ld); +extern struct rte_owner_class ospf_rte_owner_class; + /* iface.c */ void ospf_iface_chstate(struct ospf_iface *ifa, u8 state); void ospf_iface_sm(struct ospf_iface *ifa, int event); diff --git a/proto/ospf/rt.c b/proto/ospf/rt.c index faee49dc..3e208023 100644 --- a/proto/ospf/rt.c +++ b/proto/ospf/rt.c @@ -144,7 +144,7 @@ orta_compare(const struct ospf_proto *p, const orta *new, const orta *old) { int r; - if (old->type == RTS_DUMMY) + if (!old->type) return 1; /* Prefer intra-area to inter-area to externals */ @@ -195,7 +195,7 @@ orta_compare_asbr(const struct ospf_proto *p, const orta *new, const orta *old) { int r; - if (old->type == RTS_DUMMY) + if (!old->type) return 1; if (!p->rfc1583) @@ -225,7 +225,7 @@ orta_compare_ext(const struct ospf_proto *p, const orta *new, const orta *old) { int r; - if (old->type == RTS_DUMMY) + if (!old->type) return 1; /* 16.4 (6a) - prefer routes with lower type */ @@ -2053,36 +2053,61 @@ again1: if (nf->n.type) /* Add the route */ { rta a0 = { - .src = p->p.main_source, .source = nf->n.type, .scope = SCOPE_UNIVERSE, .dest = RTD_UNICAST, .nh = *(nf->n.nhs), + .pref = p->p.main_channel->preference, }; if (reload || ort_changed(nf, &a0)) { - rta *a = rta_lookup(&a0); - rte *e = rte_get_temp(a); + a0.eattrs = alloca(sizeof(ea_list) + 4 * sizeof(eattr)); + memset(a0.eattrs, 0, sizeof(ea_list)); - rta_free(nf->old_rta); - nf->old_rta = rta_clone(a); - e->u.ospf.metric1 = nf->old_metric1 = nf->n.metric1; - e->u.ospf.metric2 = nf->old_metric2 = nf->n.metric2; - e->u.ospf.tag = nf->old_tag = nf->n.tag; - e->u.ospf.router_id = nf->old_rid = nf->n.rid; - e->pflags = EA_ID_FLAG(EA_OSPF_METRIC1) | EA_ID_FLAG(EA_OSPF_ROUTER_ID); + nf->old_metric1 = nf->n.metric1; + nf->old_metric2 = nf->n.metric2; + nf->old_tag = nf->n.tag; + nf->old_rid = nf->n.rid; + + a0.eattrs->attrs[a0.eattrs->count++] = (eattr) { + .id = EA_OSPF_METRIC1, + .type = EAF_TYPE_INT, + .u.data = nf->n.metric1, + }; if (nf->n.type == RTS_OSPF_EXT2) - e->pflags |= EA_ID_FLAG(EA_OSPF_METRIC2); + a0.eattrs->attrs[a0.eattrs->count++] = (eattr) { + .id = EA_OSPF_METRIC2, + .type = EAF_TYPE_INT, + .u.data = nf->n.metric2, + }; - /* Perhaps onfly if tag is non-zero? */ if ((nf->n.type == RTS_OSPF_EXT1) || (nf->n.type == RTS_OSPF_EXT2)) - e->pflags |= EA_ID_FLAG(EA_OSPF_TAG); + a0.eattrs->attrs[a0.eattrs->count++] = (eattr) { + .id = EA_OSPF_TAG, + .type = EAF_TYPE_INT, + .u.data = nf->n.tag, + }; + + a0.eattrs->attrs[a0.eattrs->count++] = (eattr) { + .id = EA_OSPF_ROUTER_ID, + .type = EAF_TYPE_ROUTER_ID, + .u.data = nf->n.rid, + }; + + rta_free(nf->old_rta); + nf->old_rta = rta_lookup(&a0); + + rte e0 = { + .attrs = nf->old_rta, + .src = p->p.main_source, + }; DBG("Mod rte type %d - %N via %I on iface %s, met %d\n", a0.source, nf->fn.addr, a0.gw, a0.iface ? a0.iface->name : "(none)", nf->n.metric1); - rte_update(&p->p, nf->fn.addr, e); + + rte_update(p->p.main_channel, nf->fn.addr, &e0, p->p.main_source); } } else if (nf->old_rta) @@ -2091,7 +2116,7 @@ again1: rta_free(nf->old_rta); nf->old_rta = NULL; - rte_update(&p->p, nf->fn.addr, NULL); + rte_update(p->p.main_channel, nf->fn.addr, NULL, p->p.main_source); } /* Remove unused rt entry, some special entries are persistent */ @@ -2107,7 +2132,6 @@ again1: } FIB_ITERATE_END; - WALK_LIST(oa, p->area_list) { /* Cleanup ASBR hash tables */ diff --git a/proto/ospf/topology.c b/proto/ospf/topology.c index 52c2a0ce..bb88d20a 100644 --- a/proto/ospf/topology.c +++ b/proto/ospf/topology.c @@ -1300,7 +1300,7 @@ find_surrogate_fwaddr(struct ospf_proto *p, struct ospf_area *oa) } void -ospf_rt_notify(struct proto *P, struct channel *ch UNUSED, net *n, rte *new, rte *old UNUSED) +ospf_rt_notify(struct proto *P, struct channel *ch UNUSED, const net_addr *n, rte *new, const rte *old UNUSED) { struct ospf_proto *p = (struct ospf_proto *) P; struct ospf_area *oa = NULL; /* non-NULL for NSSA-LSA */ @@ -1319,7 +1319,7 @@ ospf_rt_notify(struct proto *P, struct channel *ch UNUSED, net *n, rte *new, rte if (!new) { - nf = fib_find(&p->rtf, n->n.addr); + nf = fib_find(&p->rtf, n); if (!nf || !nf->external_rte) return; @@ -1346,14 +1346,14 @@ ospf_rt_notify(struct proto *P, struct channel *ch UNUSED, net *n, rte *new, rte if (m1 > LSINFINITY) { log(L_WARN "%s: Invalid ospf_metric1 value %u for route %N", - p->p.name, m1, n->n.addr); + p->p.name, m1, n); m1 = LSINFINITY; } if (m2 > LSINFINITY) { log(L_WARN "%s: Invalid ospf_metric2 value %u for route %N", - p->p.name, m2, n->n.addr); + p->p.name, m2, n); m2 = LSINFINITY; } @@ -1377,12 +1377,12 @@ ospf_rt_notify(struct proto *P, struct channel *ch UNUSED, net *n, rte *new, rte if (ipa_zero(fwd)) { log(L_ERR "%s: Cannot find forwarding address for NSSA-LSA %N", - p->p.name, n->n.addr); + p->p.name, n); return; } } - nf = fib_get(&p->rtf, n->n.addr); + nf = fib_get(&p->rtf, n); ospf_originate_ext_lsa(p, oa, nf, LSA_M_EXPORT, metric, ebit, fwd, tag, 1, p->vpn_pe); nf->external_rte = 1; } diff --git a/proto/ospf/topology.h b/proto/ospf/topology.h index 535d1f1b..c36d0b50 100644 --- a/proto/ospf/topology.h +++ b/proto/ospf/topology.h @@ -200,7 +200,7 @@ void ospf_originate_sum_rt_lsa(struct ospf_proto *p, struct ospf_area *oa, u32 d void ospf_originate_ext_lsa(struct ospf_proto *p, struct ospf_area *oa, ort *nf, u8 mode, u32 metric, u32 ebit, ip_addr fwaddr, u32 tag, int pbit, int dn); void ospf_originate_gr_lsa(struct ospf_proto *p, struct ospf_iface *ifa); -void ospf_rt_notify(struct proto *P, struct channel *ch, net *n, rte *new, rte *old); +void ospf_rt_notify(struct proto *P, struct channel *ch, const net_addr *n, rte *new, const rte *old); void ospf_update_topology(struct ospf_proto *p); struct top_hash_entry *ospf_hash_find(struct top_graph *, u32 domain, u32 lsa, u32 rtr, u32 type); diff --git a/proto/perf/perf.c b/proto/perf/perf.c index ba401a8a..aa688d88 100644 --- a/proto/perf/perf.c +++ b/proto/perf/perf.c @@ -143,10 +143,10 @@ perf_loop(void *data) if (!p->attrs_per_rte || !(i % p->attrs_per_rte)) { struct rta a0 = { - .src = p->p.main_source, .source = RTS_PERF, .scope = SCOPE_UNIVERSE, .dest = RTD_UNICAST, + .pref = p->p.main_channel->preference, .nh.iface = p->ifa->iface, .nh.gw = gw, .nh.weight = 1, @@ -160,18 +160,17 @@ perf_loop(void *data) clock_gettime(CLOCK_MONOTONIC, &ts_generated); - for (uint i=0; i<N; i++) { - rte *e = rte_get_temp(p->data[i].a); - e->pflags = 0; - - rte_update(P, &(p->data[i].net), e); + for (uint i=0; i<N; i++) + { + rte e0 = { .attrs = p->data[i].a, .src = P->main_source, }; + rte_update(P->main_channel, &(p->data[i].net), &e0, P->main_source); } clock_gettime(CLOCK_MONOTONIC, &ts_update); if (!p->keep) for (uint i=0; i<N; i++) - rte_update(P, &(p->data[i].net), NULL); + rte_update(P->main_channel, &(p->data[i].net), NULL, P->main_source); clock_gettime(CLOCK_MONOTONIC, &ts_withdraw); @@ -199,12 +198,14 @@ perf_loop(void *data) p->exp++; } - rt_schedule_prune(P->main_channel->table); + RT_LOCK(P->main_channel->table); + rt_schedule_prune(RT_PRIV(P->main_channel->table)); + RT_UNLOCK(P->main_channel->table); ev_schedule(p->loop); } static void -perf_rt_notify(struct proto *P, struct channel *c UNUSED, struct network *net UNUSED, struct rte *new UNUSED, struct rte *old UNUSED) +perf_rt_notify(struct proto *P, struct channel *c UNUSED, const net_addr *net UNUSED, struct rte *new UNUSED, const struct rte *old UNUSED) { struct perf_proto *p = (struct perf_proto *) P; p->exp++; diff --git a/proto/pipe/config.Y b/proto/pipe/config.Y index 1202c169..fc08445f 100644 --- a/proto/pipe/config.Y +++ b/proto/pipe/config.Y @@ -16,7 +16,7 @@ CF_DEFINES CF_DECLS -CF_KEYWORDS(PIPE, PEER, TABLE) +CF_KEYWORDS(PIPE, PEER, TABLE, MAX, GENERATION) CF_GRAMMAR @@ -25,6 +25,8 @@ proto: pipe_proto '}' { this_channel = NULL; } ; pipe_proto_start: proto_start PIPE { this_proto = proto_config_new(&proto_pipe, $1); + this_proto->loop_order = DOMAIN_ORDER(proto); + PIPE_CFG->max_generation = 16; } proto_name { @@ -41,6 +43,10 @@ pipe_proto: | pipe_proto proto_item ';' | pipe_proto channel_item_ ';' | pipe_proto PEER TABLE rtable ';' { PIPE_CFG->peer = $4; } + | pipe_proto MAX GENERATION expr ';' { + if (($4 < 1) || ($4 > 254)) cf_error("Max generation must be in range 1..254, got %u", $4); + PIPE_CFG->max_generation = $4; + } ; CF_CODE diff --git a/proto/pipe/pipe.c b/proto/pipe/pipe.c index 3532f114..270f7b92 100644 --- a/proto/pipe/pipe.c +++ b/proto/pipe/pipe.c @@ -43,67 +43,58 @@ #include "pipe.h" +#ifdef CONFIG_BGP +#include "proto/bgp/bgp.h" +#endif + static void -pipe_rt_notify(struct proto *P, struct channel *src_ch, net *n, rte *new, rte *old) +pipe_rt_notify(struct proto *P, struct channel *src_ch, const net_addr *n, rte *new, const rte *old) { struct pipe_proto *p = (void *) P; struct channel *dst = (src_ch == p->pri) ? p->sec : p->pri; - struct rte_src *src; - - rte *e; - rta *a; if (!new && !old) return; - if (dst->table->pipe_busy) - { - log(L_ERR "Pipe loop detected when sending %N to table %s", - n->n.addr, dst->table->name); - return; - } - if (new) { - a = alloca(rta_size(new->attrs)); + rta *a = alloca(rta_size(new->attrs)); memcpy(a, new->attrs, rta_size(new->attrs)); - a->aflags = 0; + a->cached = 0; a->hostentry = NULL; - e = rte_get_temp(a); - e->pflags = 0; - - /* Copy protocol specific embedded attributes. */ - memcpy(&(e->u), &(new->u), sizeof(e->u)); - e->pref = new->pref; - e->pflags = new->pflags; -#ifdef CONFIG_BGP - /* Hack to cleanup cached value */ - if (e->attrs->src->proto->proto == &proto_bgp) - e->u.bgp.stale = -1; -#endif + rte e0 = { + .attrs = a, + .src = new->src, + .generation = new->generation + 1, + }; - src = a->src; + rte_update(dst, n, &e0, new->src); } else - { - e = NULL; - src = old->attrs->src; - } - - src_ch->table->pipe_busy = 1; - rte_update2(dst, n->n.addr, e, src); - src_ch->table->pipe_busy = 0; + rte_update(dst, n, NULL, old->src); } static int -pipe_preexport(struct proto *P, rte **ee, struct linpool *p UNUSED) +pipe_preexport(struct channel *c, rte *e) { - struct proto *pp = (*ee)->sender->proto; + struct pipe_proto *p = (void *) c->proto; + + /* Avoid direct loopbacks */ + if (e->sender == c->in_req.hook) + return -1; - if (pp == P) - return -1; /* Avoid local loops automatically */ + /* Indirection check */ + uint max_generation = ((struct pipe_config *) p->p.cf)->max_generation; + if (e->generation >= max_generation) + { + log_rl(&p->rl_gen, L_ERR "Route overpiped (%u hops of %u configured in %s) in table %s: %N %s/%u:%u", + e->generation, max_generation, c->proto->name, + c->table->name, e->net, e->src->owner->name, e->src->private_id, e->src->global_id); + + return -1; + } return 0; } @@ -117,6 +108,23 @@ pipe_reload_routes(struct channel *C) channel_request_feeding((C == p->pri) ? p->sec : p->pri); } +static void +pipe_feed_begin(struct channel *C, int refeeding UNUSED) +{ + struct pipe_proto *p = (void *) C->proto; + struct channel *dst = (C == p->pri) ? p->sec : p->pri; + + channel_refresh_begin(dst); +} + +static void +pipe_feed_end(struct channel *C) +{ + struct pipe_proto *p = (void *) C->proto; + struct channel *dst = (C == p->pri) ? p->sec : p->pri; + + channel_refresh_end(dst); +} static void pipe_postconfig(struct proto_config *CF) @@ -187,6 +195,10 @@ pipe_init(struct proto_config *CF) P->rt_notify = pipe_rt_notify; P->preexport = pipe_preexport; P->reload_routes = pipe_reload_routes; + P->feed_begin = pipe_feed_begin; + P->feed_end = pipe_feed_end; + + p->rl_gen = (struct tbf) TBF_DEFAULT_LOG_LIMITS; pipe_configure_channels(p, cf); @@ -219,8 +231,18 @@ pipe_get_status(struct proto *P, byte *buf) static void pipe_show_stats(struct pipe_proto *p) { - struct proto_stats *s1 = &p->pri->stats; - struct proto_stats *s2 = &p->sec->stats; + struct channel_import_stats *s1i = &p->pri->import_stats; + struct channel_export_stats *s1e = &p->pri->export_stats; + struct channel_import_stats *s2i = &p->sec->import_stats; + struct channel_export_stats *s2e = &p->sec->export_stats; + + struct rt_import_stats *rs1i = p->pri->in_req.hook ? &p->pri->in_req.hook->stats : NULL; + struct rt_export_stats *rs1e = p->pri->out_req.hook ? &p->pri->out_req.hook->stats : NULL; + struct rt_import_stats *rs2i = p->sec->in_req.hook ? &p->sec->in_req.hook->stats : NULL; + struct rt_export_stats *rs2e = p->sec->out_req.hook ? &p->sec->out_req.hook->stats : NULL; + + u32 pri_routes = p->pri->in_limit.count; + u32 sec_routes = p->sec->in_limit.count; /* * Pipe stats (as anything related to pipes) are a bit tricky. There @@ -244,24 +266,22 @@ pipe_show_stats(struct pipe_proto *p) */ cli_msg(-1006, " Routes: %u imported, %u exported", - s1->imp_routes, s2->imp_routes); + pri_routes, sec_routes); cli_msg(-1006, " Route change stats: received rejected filtered ignored accepted"); cli_msg(-1006, " Import updates: %10u %10u %10u %10u %10u", - s2->exp_updates_received, s2->exp_updates_rejected + s1->imp_updates_invalid, - s2->exp_updates_filtered, s1->imp_updates_ignored, s1->imp_updates_accepted); + rs2e->updates_received, s2e->updates_rejected + s1i->updates_invalid, + s2e->updates_filtered, rs1i->updates_ignored, rs1i->updates_accepted); cli_msg(-1006, " Import withdraws: %10u %10u --- %10u %10u", - s2->exp_withdraws_received, s1->imp_withdraws_invalid, - s1->imp_withdraws_ignored, s1->imp_withdraws_accepted); + rs2e->withdraws_received, s1i->withdraws_invalid, + rs1i->withdraws_ignored, rs1i->withdraws_accepted); cli_msg(-1006, " Export updates: %10u %10u %10u %10u %10u", - s1->exp_updates_received, s1->exp_updates_rejected + s2->imp_updates_invalid, - s1->exp_updates_filtered, s2->imp_updates_ignored, s2->imp_updates_accepted); + rs1e->updates_received, s1e->updates_rejected + s2i->updates_invalid, + s1e->updates_filtered, rs2i->updates_ignored, rs2i->updates_accepted); cli_msg(-1006, " Export withdraws: %10u %10u --- %10u %10u", - s1->exp_withdraws_received, s2->imp_withdraws_invalid, - s2->imp_withdraws_ignored, s2->imp_withdraws_accepted); + rs1e->withdraws_received, s2i->withdraws_invalid, + rs2i->withdraws_ignored, rs2i->withdraws_accepted); } -static const char *pipe_feed_state[] = { [ES_DOWN] = "down", [ES_FEEDING] = "feed", [ES_READY] = "up" }; - static void pipe_show_proto_info(struct proto *P) { @@ -270,13 +290,17 @@ pipe_show_proto_info(struct proto *P) cli_msg(-1006, " Channel %s", "main"); cli_msg(-1006, " Table: %s", p->pri->table->name); cli_msg(-1006, " Peer table: %s", p->sec->table->name); - cli_msg(-1006, " Import state: %s", pipe_feed_state[p->sec->export_state]); - cli_msg(-1006, " Export state: %s", pipe_feed_state[p->pri->export_state]); + cli_msg(-1006, " Import state: %s", rt_export_state_name(rt_export_get_state(p->sec->out_req.hook))); + cli_msg(-1006, " Export state: %s", rt_export_state_name(rt_export_get_state(p->pri->out_req.hook))); cli_msg(-1006, " Import filter: %s", filter_name(p->sec->out_filter)); cli_msg(-1006, " Export filter: %s", filter_name(p->pri->out_filter)); - channel_show_limit(&p->pri->in_limit, "Import limit:"); - channel_show_limit(&p->sec->in_limit, "Export limit:"); + + + channel_show_limit(&p->pri->in_limit, "Import limit:", + (p->pri->limit_active & (1 << PLD_IN)), p->pri->limit_actions[PLD_IN]); + channel_show_limit(&p->sec->in_limit, "Export limit:", + (p->sec->limit_active & (1 << PLD_IN)), p->sec->limit_actions[PLD_IN]); if (P->proto_state != PS_DOWN) pipe_show_stats(p); diff --git a/proto/pipe/pipe.h b/proto/pipe/pipe.h index 038c6666..60c857eb 100644 --- a/proto/pipe/pipe.h +++ b/proto/pipe/pipe.h @@ -12,12 +12,14 @@ struct pipe_config { struct proto_config c; struct rtable_config *peer; /* Table we're connected to */ + u8 max_generation; }; struct pipe_proto { struct proto p; struct channel *pri; struct channel *sec; + struct tbf rl_gen; }; #endif diff --git a/proto/radv/radv.c b/proto/radv/radv.c index 66e8eb4b..b0f15514 100644 --- a/proto/radv/radv.c +++ b/proto/radv/radv.c @@ -284,7 +284,7 @@ radv_iface_new(struct radv_proto *p, struct iface *iface, struct radv_iface_conf RADV_TRACE(D_EVENTS, "Adding interface %s", iface->name); - pool *pool = rp_new(p->p.pool, iface->name); + pool *pool = rp_new(p->p.pool, p->p.loop, iface->name); ifa = mb_allocz(pool, sizeof(struct radv_iface)); ifa->pool = pool; ifa->ra = p; @@ -317,7 +317,7 @@ radv_iface_remove(struct radv_iface *ifa) rem_node(NODE ifa); - rfree(ifa->pool); + rp_free(ifa->pool, p->p.pool); } static void @@ -385,18 +385,18 @@ radv_trigger_valid(struct radv_config *cf) } static inline int -radv_net_match_trigger(struct radv_config *cf, net *n) +radv_net_match_trigger(struct radv_config *cf, const net_addr *n) { - return radv_trigger_valid(cf) && net_equal(n->n.addr, &cf->trigger); + return radv_trigger_valid(cf) && net_equal(n, &cf->trigger); } int -radv_preexport(struct proto *P, rte **new, struct linpool *pool UNUSED) +radv_preexport(struct channel *c, rte *new) { // struct radv_proto *p = (struct radv_proto *) P; - struct radv_config *cf = (struct radv_config *) (P->cf); + struct radv_config *cf = (struct radv_config *) (c->proto->cf); - if (radv_net_match_trigger(cf, (*new)->net)) + if (radv_net_match_trigger(cf, new->net)) return RIC_PROCESS; if (cf->propagate_routes) @@ -406,7 +406,7 @@ radv_preexport(struct proto *P, rte **new, struct linpool *pool UNUSED) } static void -radv_rt_notify(struct proto *P, struct channel *ch UNUSED, net *n, rte *new, rte *old UNUSED) +radv_rt_notify(struct proto *P, struct channel *ch UNUSED, const net_addr *n, rte *new, const rte *old UNUSED) { struct radv_proto *p = (struct radv_proto *) P; struct radv_config *cf = (struct radv_config *) (P->cf); @@ -457,14 +457,14 @@ radv_rt_notify(struct proto *P, struct channel *ch UNUSED, net *n, rte *new, rte (preference != RA_PREF_HIGH)) { log(L_WARN "%s: Invalid ra_preference value %u on route %N", - p->p.name, preference, n->n.addr); + p->p.name, preference, n); preference = RA_PREF_MEDIUM; preference_set = 1; lifetime = 0; lifetime_set = 1; } - rt = fib_get(&p->routes, n->n.addr); + rt = fib_get(&p->routes, n); /* Ignore update if nothing changed */ if (rt->valid && @@ -487,7 +487,7 @@ radv_rt_notify(struct proto *P, struct channel *ch UNUSED, net *n, rte *new, rte else { /* Withdraw */ - rt = fib_find(&p->routes, n->n.addr); + rt = fib_find(&p->routes, n); if (!rt || !rt->valid) return; @@ -555,7 +555,10 @@ radv_check_active(struct radv_proto *p) return 1; struct channel *c = p->p.main_channel; - return rt_examine(c->table, &cf->trigger, &p->p, c->out_filter); + RT_LOCK(c->table); + int active = rt_examine(RT_PRIV(c->table), &cf->trigger, c, c->out_filter); + RT_UNLOCK(c->table); + return active; } static void diff --git a/proto/rip/rip.c b/proto/rip/rip.c index e1a235a0..ee05f058 100644 --- a/proto/rip/rip.c +++ b/proto/rip/rip.c @@ -145,7 +145,7 @@ rip_announce_rte(struct rip_proto *p, struct rip_entry *en) { /* Update */ rta a0 = { - .src = p->p.main_source, + .pref = p->p.main_channel->preference, .source = RTS_RIP, .scope = SCOPE_UNIVERSE, .dest = RTD_UNICAST, @@ -188,21 +188,34 @@ rip_announce_rte(struct rip_proto *p, struct rip_entry *en) a0.nh.iface = rt->from->ifa->iface; } - rta *a = rta_lookup(&a0); - rte *e = rte_get_temp(a); + a0.eattrs = alloca(sizeof(ea_list) + 3*sizeof(eattr)); + memset(a0.eattrs, 0, sizeof(ea_list)); /* Zero-ing only the ea_list header */ + a0.eattrs->count = 3; + a0.eattrs->attrs[0] = (eattr) { + .id = EA_RIP_METRIC, + .type = EAF_TYPE_INT, + .u.data = rt_metric, + }; + a0.eattrs->attrs[1] = (eattr) { + .id = EA_RIP_TAG, + .type = EAF_TYPE_INT, + .u.data = rt_tag, + }; + a0.eattrs->attrs[2] = (eattr) { + .id = EA_RIP_FROM, + .type = EAF_TYPE_PTR, + .u.data = (uintptr_t) a0.nh.iface, + }; - e->u.rip.from = a0.nh.iface; - e->u.rip.metric = rt_metric; - e->u.rip.tag = rt_tag; - e->pflags = EA_ID_FLAG(EA_RIP_METRIC) | EA_ID_FLAG(EA_RIP_TAG); + rte e0 = { + .attrs = &a0, + .src = p->p.main_source, + }; - rte_update(&p->p, en->n.addr, e); + rte_update(p->p.main_channel, en->n.addr, &e0, p->p.main_source); } else - { - /* Withdraw */ - rte_update(&p->p, en->n.addr, NULL); - } + rte_update(p->p.main_channel, en->n.addr, NULL, p->p.main_source); } /** @@ -297,8 +310,8 @@ rip_withdraw_rte(struct rip_proto *p, net_addr *n, struct rip_neighbor *from) * it into our data structures. */ static void -rip_rt_notify(struct proto *P, struct channel *ch UNUSED, struct network *net, struct rte *new, - struct rte *old UNUSED) +rip_rt_notify(struct proto *P, struct channel *ch UNUSED, const net_addr *net, struct rte *new, + const struct rte *old UNUSED) { struct rip_proto *p = (struct rip_proto *) P; struct rip_entry *en; @@ -307,20 +320,21 @@ rip_rt_notify(struct proto *P, struct channel *ch UNUSED, struct network *net, s if (new) { /* Update */ - u32 rt_metric = ea_get_int(new->attrs->eattrs, EA_RIP_METRIC, 1); u32 rt_tag = ea_get_int(new->attrs->eattrs, EA_RIP_TAG, 0); + u32 rt_metric = ea_get_int(new->attrs->eattrs, EA_RIP_METRIC, 1); + struct iface *rt_from = (struct iface *) ea_get_int(new->attrs->eattrs, EA_RIP_FROM, 0); if (rt_metric > p->infinity) { log(L_WARN "%s: Invalid rip_metric value %u for route %N", - p->p.name, rt_metric, net->n.addr); + p->p.name, rt_metric, net); rt_metric = p->infinity; } if (rt_tag > 0xffff) { log(L_WARN "%s: Invalid rip_tag value %u for route %N", - p->p.name, rt_tag, net->n.addr); + p->p.name, rt_tag, net); rt_metric = p->infinity; rt_tag = 0; } @@ -332,21 +346,21 @@ rip_rt_notify(struct proto *P, struct channel *ch UNUSED, struct network *net, s * collection. */ - en = fib_get(&p->rtable, net->n.addr); + en = fib_get(&p->rtable, net); old_metric = en->valid ? en->metric : -1; en->valid = RIP_ENTRY_VALID; en->metric = rt_metric; en->tag = rt_tag; - en->from = (new->attrs->src->proto == P) ? new->u.rip.from : NULL; + en->from = (new->src->owner == &P->sources) ? rt_from : NULL; en->iface = new->attrs->nh.iface; en->next_hop = new->attrs->nh.gw; } else { /* Withdraw */ - en = fib_find(&p->rtable, net->n.addr); + en = fib_find(&p->rtable, net); if (!en || en->valid != RIP_ENTRY_VALID) return; @@ -1068,37 +1082,33 @@ rip_reload_routes(struct channel *C) rip_kick_timer(p); } -static void -rip_make_tmp_attrs(struct rte *rt, struct linpool *pool) -{ - rte_init_tmp_attrs(rt, pool, 2); - rte_make_tmp_attr(rt, EA_RIP_METRIC, EAF_TYPE_INT, rt->u.rip.metric); - rte_make_tmp_attr(rt, EA_RIP_TAG, EAF_TYPE_INT, rt->u.rip.tag); -} +static struct rte_owner_class rip_rte_owner_class; -static void -rip_store_tmp_attrs(struct rte *rt, struct linpool *pool) +static inline struct rip_proto * +rip_rte_proto(struct rte *rte) { - rte_init_tmp_attrs(rt, pool, 2); - rt->u.rip.metric = rte_store_tmp_attr(rt, EA_RIP_METRIC); - rt->u.rip.tag = rte_store_tmp_attr(rt, EA_RIP_TAG); + return (rte->src->owner->class == &rip_rte_owner_class) ? + SKIP_BACK(struct rip_proto, p.sources, rte->src->owner) : NULL; } static int rip_rte_better(struct rte *new, struct rte *old) { - return new->u.rip.metric < old->u.rip.metric; + ASSERT_DIE(new->src == old->src); + struct rip_proto *p = rip_rte_proto(new); + + u32 new_metric = ea_get_int(new->attrs->eattrs, EA_RIP_METRIC, p->infinity); + u32 old_metric = ea_get_int(old->attrs->eattrs, EA_RIP_METRIC, p->infinity); + + return new_metric < old_metric; } -static int -rip_rte_same(struct rte *new, struct rte *old) +static u32 +rip_rte_igp_metric(struct rte *rt) { - return ((new->u.rip.metric == old->u.rip.metric) && - (new->u.rip.tag == old->u.rip.tag) && - (new->u.rip.from == old->u.rip.from)); + return ea_get_int(rt->attrs->eattrs, EA_RIP_METRIC, IGP_METRIC_UNKNOWN); } - static void rip_postconfig(struct proto_config *CF) { @@ -1120,10 +1130,7 @@ rip_init(struct proto_config *CF) P->rt_notify = rip_rt_notify; P->neigh_notify = rip_neigh_notify; P->reload_routes = rip_reload_routes; - P->make_tmp_attrs = rip_make_tmp_attrs; - P->store_tmp_attrs = rip_store_tmp_attrs; - P->rte_better = rip_rte_better; - P->rte_same = rip_rte_same; + P->sources.class = &rip_rte_owner_class; return P; } @@ -1198,10 +1205,14 @@ rip_reconfigure(struct proto *P, struct proto_config *CF) static void rip_get_route_info(rte *rte, byte *buf) { - buf += bsprintf(buf, " (%d/%d)", rte->pref, rte->u.rip.metric); + struct rip_proto *p = rip_rte_proto(rte); + u32 rt_metric = ea_get_int(rte->attrs->eattrs, EA_RIP_METRIC, p->infinity); + u32 rt_tag = ea_get_int(rte->attrs->eattrs, EA_RIP_TAG, 0); + + buf += bsprintf(buf, " (%d/%d)", rte->attrs->pref, rt_metric); - if (rte->u.rip.tag) - bsprintf(buf, " [%04x]", rte->u.rip.tag); + if (rt_tag) + bsprintf(buf, " [%04x]", rt_tag); } static int @@ -1321,6 +1332,12 @@ rip_dump(struct proto *P) } +static struct rte_owner_class rip_rte_owner_class = { + .get_route_info = rip_get_route_info, + .rte_better = rip_rte_better, + .rte_igp_metric = rip_rte_igp_metric, +}; + struct protocol proto_rip = { .name = "RIP", .template = "rip%d", @@ -1335,6 +1352,5 @@ struct protocol proto_rip = { .start = rip_start, .shutdown = rip_shutdown, .reconfigure = rip_reconfigure, - .get_route_info = rip_get_route_info, .get_attr = rip_get_attr }; diff --git a/proto/rip/rip.h b/proto/rip/rip.h index 8d347000..f8713c4a 100644 --- a/proto/rip/rip.h +++ b/proto/rip/rip.h @@ -197,6 +197,7 @@ struct rip_rte #define EA_RIP_METRIC EA_CODE(PROTOCOL_RIP, 0) #define EA_RIP_TAG EA_CODE(PROTOCOL_RIP, 1) +#define EA_RIP_FROM EA_CODE(PROTOCOL_RIP, 2) static inline int rip_is_v2(struct rip_proto *p) { return p->rip2; } diff --git a/proto/rpki/config.Y b/proto/rpki/config.Y index d6d326b8..743b5b42 100644 --- a/proto/rpki/config.Y +++ b/proto/rpki/config.Y @@ -42,6 +42,7 @@ proto: rpki_proto ; rpki_proto_start: proto_start RPKI { this_proto = proto_config_new(&proto_rpki, $1); + this_proto->loop_order = DOMAIN_ORDER(proto); RPKI_CFG->retry_interval = RPKI_RETRY_INTERVAL; RPKI_CFG->refresh_interval = RPKI_REFRESH_INTERVAL; RPKI_CFG->expire_interval = RPKI_EXPIRE_INTERVAL; diff --git a/proto/rpki/packets.c b/proto/rpki/packets.c index dd11f997..abe6abfc 100644 --- a/proto/rpki/packets.c +++ b/proto/rpki/packets.c @@ -233,7 +233,12 @@ static const size_t min_pdu_size[] = { [ERROR] = 16, }; -static int rpki_send_error_pdu(struct rpki_cache *cache, const enum pdu_error_type error_code, const u32 err_pdu_len, const struct pdu_header *erroneous_pdu, const char *fmt, ...); +static int rpki_send_error_pdu_(struct rpki_cache *cache, const enum pdu_error_type error_code, const u32 err_pdu_len, const struct pdu_header *erroneous_pdu, const char *fmt, ...); + +#define rpki_send_error_pdu(cache, error_code, err_pdu_len, erroneous_pdu, fmt...) ({ \ + rpki_send_error_pdu_(cache, error_code, err_pdu_len, erroneous_pdu, #fmt); \ + CACHE_TRACE(D_PACKETS, cache, #fmt); \ + }) static void rpki_pdu_to_network_byte_order(struct pdu_header *pdu) @@ -595,6 +600,7 @@ rpki_handle_error_pdu(struct rpki_cache *cache, const struct pdu_error *pdu) case INTERNAL_ERROR: case INVALID_REQUEST: case UNSUPPORTED_PDU_TYPE: + CACHE_TRACE(D_PACKETS, cache, "Got UNSUPPORTED_PDU_TYPE"); rpki_cache_change_state(cache, RPKI_CS_ERROR_FATAL); break; @@ -652,21 +658,7 @@ rpki_handle_cache_response_pdu(struct rpki_cache *cache, const struct pdu_cache_ { if (cache->request_session_id) { - if (cache->last_update) - { - /* - * This isn't the first sync and we already received records. This point - * is after Reset Query and before importing new records from cache - * server. We need to load new ones and kick out missing ones. So start - * a refresh cycle. - */ - if (cache->p->roa4_channel) - rt_refresh_begin(cache->p->roa4_channel->table, cache->p->roa4_channel); - if (cache->p->roa6_channel) - rt_refresh_begin(cache->p->roa6_channel->table, cache->p->roa6_channel); - - cache->p->refresh_channels = 1; - } + rpki_start_refresh(cache->p); cache->session_id = pdu->session_id; cache->request_session_id = 0; } @@ -815,14 +807,7 @@ rpki_handle_end_of_data_pdu(struct rpki_cache *cache, const struct pdu_end_of_da (cf->keep_expire_interval ? "keeps " : ""), cache->expire_interval); } - if (cache->p->refresh_channels) - { - cache->p->refresh_channels = 0; - if (cache->p->roa4_channel) - rt_refresh_end(cache->p->roa4_channel->table, cache->p->roa4_channel); - if (cache->p->roa6_channel) - rt_refresh_end(cache->p->roa6_channel->table, cache->p->roa6_channel); - } + rpki_stop_refresh(cache->p); cache->last_update = current_time(); cache->serial_num = pdu->serial_num; @@ -897,6 +882,9 @@ rpki_rx_hook(struct birdsock *sk, uint size) struct rpki_cache *cache = sk->data; struct rpki_proto *p = cache->p; + if ((p->p.proto_state == PS_DOWN) || (p->cache != cache)) + return 0; + byte *pkt_start = sk->rbuf; byte *end = pkt_start + size; @@ -953,6 +941,8 @@ rpki_err_hook(struct birdsock *sk, int error_num) CACHE_TRACE(D_EVENTS, cache, "The other side closed a connection"); } + if (cache->p->cache != cache) + return; rpki_cache_change_state(cache, RPKI_CS_ERROR_TRANSPORT); } @@ -972,6 +962,9 @@ rpki_tx_hook(sock *sk) { struct rpki_cache *cache = sk->data; + if (cache->p->cache != cache) + return; + while (rpki_fire_tx(cache) > 0) ; } @@ -981,6 +974,9 @@ rpki_connected_hook(sock *sk) { struct rpki_cache *cache = sk->data; + if (cache->p->cache != cache) + return; + CACHE_TRACE(D_EVENTS, cache, "Connected"); proto_notify_state(&cache->p->p, PS_UP); @@ -1002,7 +998,7 @@ rpki_connected_hook(sock *sk) * This function prepares Error PDU and sends it to a cache server. */ static int -rpki_send_error_pdu(struct rpki_cache *cache, const enum pdu_error_type error_code, const u32 err_pdu_len, const struct pdu_header *erroneous_pdu, const char *fmt, ...) +rpki_send_error_pdu_(struct rpki_cache *cache, const enum pdu_error_type error_code, const u32 err_pdu_len, const struct pdu_header *erroneous_pdu, const char *fmt, ...) { va_list args; char msg[128]; diff --git a/proto/rpki/rpki.c b/proto/rpki/rpki.c index ab0837f3..afba2216 100644 --- a/proto/rpki/rpki.c +++ b/proto/rpki/rpki.c @@ -109,6 +109,7 @@ static void rpki_schedule_next_expire_check(struct rpki_cache *cache); static void rpki_stop_refresh_timer_event(struct rpki_cache *cache); static void rpki_stop_retry_timer_event(struct rpki_cache *cache); static void rpki_stop_expire_timer_event(struct rpki_cache *cache); +static void rpki_stop_all_timers(struct rpki_cache *cache); /* @@ -121,27 +122,48 @@ rpki_table_add_roa(struct rpki_cache *cache, struct channel *channel, const net_ struct rpki_proto *p = cache->p; rta a0 = { - .src = p->p.main_source, + .pref = channel->preference, .source = RTS_RPKI, .scope = SCOPE_UNIVERSE, .dest = RTD_NONE, }; - rta *a = rta_lookup(&a0); - rte *e = rte_get_temp(a); + rte e0 = { .attrs = &a0, .src = p->p.main_source, }; - e->pflags = 0; - - rte_update2(channel, &pfxr->n, e, a0.src); + rte_update(channel, &pfxr->n, &e0, p->p.main_source); } void rpki_table_remove_roa(struct rpki_cache *cache, struct channel *channel, const net_addr_union *pfxr) { struct rpki_proto *p = cache->p; - rte_update2(channel, &pfxr->n, NULL, p->p.main_source); + rte_update(channel, &pfxr->n, NULL, p->p.main_source); +} + +void +rpki_start_refresh(struct rpki_proto *p) +{ + if (p->roa4_channel) + rt_refresh_begin(&p->roa4_channel->in_req); + if (p->roa6_channel) + rt_refresh_begin(&p->roa6_channel->in_req); + + p->refresh_channels = 1; } +void +rpki_stop_refresh(struct rpki_proto *p) +{ + if (!p->refresh_channels) + return; + + p->refresh_channels = 0; + + if (p->roa4_channel) + rt_refresh_end(&p->roa4_channel->in_req); + if (p->roa6_channel) + rt_refresh_end(&p->roa6_channel->in_req); +} /* * RPKI Protocol Logic @@ -198,6 +220,8 @@ rpki_force_restart_proto(struct rpki_proto *p) { if (p->cache) { + rpki_tr_close(p->cache->tr_sock); + rpki_stop_all_timers(p->cache); CACHE_DBG(p->cache, "Connection object destroying"); } @@ -321,7 +345,7 @@ rpki_schedule_next_refresh(struct rpki_cache *cache) btime t = cache->refresh_interval S; CACHE_DBG(cache, "after %t s", t); - tm_start(cache->refresh_timer, t); + tm_start_in(cache->refresh_timer, t, cache->p->p.loop); } static void @@ -330,7 +354,7 @@ rpki_schedule_next_retry(struct rpki_cache *cache) btime t = cache->retry_interval S; CACHE_DBG(cache, "after %t s", t); - tm_start(cache->retry_timer, t); + tm_start_in(cache->retry_timer, t, cache->p->p.loop); } static void @@ -341,7 +365,7 @@ rpki_schedule_next_expire_check(struct rpki_cache *cache) t = MAX(t, 1 S); CACHE_DBG(cache, "after %t s", t); - tm_start(cache->expire_timer, t); + tm_start_in(cache->expire_timer, t, cache->p->p.loop); } static void @@ -358,13 +382,21 @@ rpki_stop_retry_timer_event(struct rpki_cache *cache) tm_stop(cache->retry_timer); } -static void UNUSED +static void rpki_stop_expire_timer_event(struct rpki_cache *cache) { CACHE_DBG(cache, "Stop"); tm_stop(cache->expire_timer); } +static void +rpki_stop_all_timers(struct rpki_cache *cache) +{ + rpki_stop_refresh_timer_event(cache); + rpki_stop_retry_timer_event(cache); + rpki_stop_expire_timer_event(cache); +} + static int rpki_do_we_recv_prefix_pdu_in_last_seconds(struct rpki_cache *cache) { @@ -387,6 +419,9 @@ rpki_refresh_hook(timer *tm) { struct rpki_cache *cache = tm->data; + if (cache->p->cache != cache) + return; + CACHE_DBG(cache, "%s", rpki_cache_state_to_str(cache->state)); switch (cache->state) @@ -433,6 +468,9 @@ rpki_retry_hook(timer *tm) { struct rpki_cache *cache = tm->data; + if (cache->p->cache != cache) + return; + CACHE_DBG(cache, "%s", rpki_cache_state_to_str(cache->state)); switch (cache->state) @@ -478,6 +516,9 @@ rpki_expire_hook(timer *tm) { struct rpki_cache *cache = tm->data; + if (cache->p->cache != cache) + return; + if (!cache->last_update) return; @@ -555,7 +596,7 @@ rpki_check_expire_interval(uint seconds) static struct rpki_cache * rpki_init_cache(struct rpki_proto *p, struct rpki_config *cf) { - pool *pool = rp_new(p->p.pool, cf->hostname); + pool *pool = rp_new(p->p.pool, p->p.loop, cf->hostname); struct rpki_cache *cache = mb_allocz(pool, sizeof(struct rpki_cache)); @@ -620,6 +661,7 @@ rpki_close_connection(struct rpki_cache *cache) { CACHE_TRACE(D_EVENTS, cache, "Closing a connection"); rpki_tr_close(cache->tr_sock); + rpki_stop_refresh(cache->p); proto_notify_state(&cache->p->p, PS_START); } diff --git a/proto/rpki/rpki.h b/proto/rpki/rpki.h index 8a5c38fd..a70a2027 100644 --- a/proto/rpki/rpki.h +++ b/proto/rpki/rpki.h @@ -83,6 +83,8 @@ const char *rpki_cache_state_to_str(enum rpki_cache_state state); void rpki_table_add_roa(struct rpki_cache *cache, struct channel *channel, const net_addr_union *pfxr); void rpki_table_remove_roa(struct rpki_cache *cache, struct channel *channel, const net_addr_union *pfxr); +void rpki_start_refresh(struct rpki_proto *p); +void rpki_stop_refresh(struct rpki_proto *p); /* * RPKI Protocol Logic diff --git a/proto/rpki/ssh_transport.c b/proto/rpki/ssh_transport.c index 6333f367..223afa80 100644 --- a/proto/rpki/ssh_transport.c +++ b/proto/rpki/ssh_transport.c @@ -38,6 +38,8 @@ rpki_tr_ssh_open(struct rpki_tr_sock *tr) if (sk_open(sk) != 0) return RPKI_TR_ERROR; + sk_start(sk); + return RPKI_TR_SUCCESS; } diff --git a/proto/rpki/tcp_transport.c b/proto/rpki/tcp_transport.c index 132f8e2d..4e850c44 100644 --- a/proto/rpki/tcp_transport.c +++ b/proto/rpki/tcp_transport.c @@ -31,6 +31,8 @@ rpki_tr_tcp_open(struct rpki_tr_sock *tr) if (sk_open(sk) != 0) return RPKI_TR_ERROR; + sk_start(sk); + return RPKI_TR_SUCCESS; } diff --git a/proto/rpki/transport.c b/proto/rpki/transport.c index a1ac7587..b52495dc 100644 --- a/proto/rpki/transport.c +++ b/proto/rpki/transport.c @@ -85,6 +85,7 @@ rpki_tr_open(struct rpki_tr_sock *tr) sk->rbsize = RPKI_RX_BUFFER_SIZE; sk->tbsize = RPKI_TX_BUFFER_SIZE; sk->tos = IP_PREC_INTERNET_CONTROL; + sk->flags |= SKF_THREAD; if (ipa_zero(sk->daddr) && sk->host) { @@ -119,6 +120,7 @@ rpki_tr_close(struct rpki_tr_sock *tr) if (tr->sk) { + sk_stop(tr->sk); rfree(tr->sk); tr->sk = NULL; } diff --git a/proto/static/static.c b/proto/static/static.c index 2789c1bb..bd7f3f5b 100644 --- a/proto/static/static.c +++ b/proto/static/static.c @@ -52,14 +52,18 @@ static linpool *static_lp; static inline struct rte_src * static_get_source(struct static_proto *p, uint i) { return i ? rt_get_source(&p->p, i) : p->p.main_source; } +static inline void static_free_source(struct rte_src *src, uint i) +{ if (i) rt_unlock_source(src); } + static void static_announce_rte(struct static_proto *p, struct static_route *r) { + struct rte_src *src; rta *a = allocz(RTA_MAX_SIZE); - a->src = static_get_source(p, r->index); a->source = RTS_STATIC; a->scope = SCOPE_UNIVERSE; a->dest = r->dest; + a->pref = p->p.main_channel->preference; if (r->dest == RTD_UNICAST) { @@ -94,7 +98,7 @@ static_announce_rte(struct static_proto *p, struct static_route *r) if (r->dest == RTDX_RECURSIVE) { rtable *tab = ipa_is_ip4(r->via) ? p->igp_table_ip4 : p->igp_table_ip6; - rta_set_recursive_next_hop(p->p.main_channel->table, a, tab, r->via, IPA_NONE, r->mls); + rta_set_recursive_next_hop(p->p.main_channel->table, a, tab, r->via, IPA_NONE, r->mls, static_lp); } /* Already announced */ @@ -102,24 +106,16 @@ static_announce_rte(struct static_proto *p, struct static_route *r) return; /* We skip rta_lookup() here */ - rte *e = rte_get_temp(a); - e->pflags = 0; + src = static_get_source(p, r->index); + rte e0 = { .attrs = a, .src = src, .net = r->net, }, *e = &e0; + /* Evaluate the filter */ if (r->cmds) - { - /* Create a temporary table node */ - e->net = alloca(sizeof(net) + r->net->length); - memset(e->net, 0, sizeof(net) + r->net->length); - net_copy(e->net->n.addr, r->net); + f_eval_rte(r->cmds, e, static_lp); - /* Evaluate the filter */ - f_eval_rte(r->cmds, &e, static_lp); + rte_update(p->p.main_channel, r->net, e, src); + static_free_source(src, r->index); - /* Remove the temporary node */ - e->net = NULL; - } - - rte_update2(p->p.main_channel, r->net, e, a->src); r->state = SRS_CLEAN; if (r->cmds) @@ -131,7 +127,9 @@ withdraw: if (r->state == SRS_DOWN) return; - rte_update2(p->p.main_channel, r->net, NULL, a->src); + src = static_get_source(p, r->index); + rte_update(p->p.main_channel, r->net, NULL, src); + static_free_source(src, r->index); r->state = SRS_DOWN; } @@ -297,7 +295,11 @@ static void static_remove_rte(struct static_proto *p, struct static_route *r) { if (r->state) - rte_update2(p->p.main_channel, r->net, NULL, static_get_source(p, r->index)); + { + struct rte_src *src = static_get_source(p, r->index); + rte_update(p->p.main_channel, r->net, NULL, src); + static_free_source(src, r->index); + } static_reset_rte(p, r); } @@ -454,6 +456,8 @@ static_postconfig(struct proto_config *CF) static_index_routes(cf); } +static struct rte_owner_class static_rte_owner_class; + static struct proto * static_init(struct proto_config *CF) { @@ -465,8 +469,7 @@ static_init(struct proto_config *CF) P->neigh_notify = static_neigh_notify; P->reload_routes = static_reload_routes; - P->rte_better = static_rte_better; - P->rte_mergable = static_rte_mergable; + P->sources.class = &static_rte_owner_class; if (cf->igp_table_ip4) p->igp_table_ip4 = cf->igp_table_ip4->table; @@ -488,10 +491,12 @@ static_start(struct proto *P) static_lp = lp_new(&root_pool, LP_GOOD_SIZE(1024)); if (p->igp_table_ip4) - rt_lock_table(p->igp_table_ip4); + RT_LOCKED(p->igp_table_ip4, t) + rt_lock_table(t); if (p->igp_table_ip6) - rt_lock_table(p->igp_table_ip6); + RT_LOCKED(p->igp_table_ip6, t) + rt_lock_table(t); p->event = ev_new_init(p->p.pool, static_announce_marked, p); @@ -517,19 +522,15 @@ static_shutdown(struct proto *P) WALK_LIST(r, cf->routes) static_reset_rte(p, r); - return PS_DOWN; -} - -static void -static_cleanup(struct proto *P) -{ - struct static_proto *p = (void *) P; - if (p->igp_table_ip4) - rt_unlock_table(p->igp_table_ip4); + RT_LOCKED(p->igp_table_ip4, t) + rt_unlock_table(t); if (p->igp_table_ip6) - rt_unlock_table(p->igp_table_ip6); + RT_LOCKED(p->igp_table_ip6, t) + rt_unlock_table(t); + + return PS_DOWN; } static void @@ -721,9 +722,9 @@ static_get_route_info(rte *rte, byte *buf) { eattr *a = ea_find(rte->attrs->eattrs, EA_GEN_IGP_METRIC); if (a) - buf += bsprintf(buf, " (%d/%u)", rte->pref, a->u.data); + buf += bsprintf(buf, " (%d/%u)", rte->attrs->pref, a->u.data); else - buf += bsprintf(buf, " (%d)", rte->pref); + buf += bsprintf(buf, " (%d)", rte->attrs->pref); } static void @@ -773,6 +774,11 @@ static_show(struct proto *P) static_show_rt(r); } +static struct rte_owner_class static_rte_owner_class = { + .get_route_info = static_get_route_info, + .rte_better = static_rte_better, + .rte_mergable = static_rte_mergable, +}; struct protocol proto_static = { .name = "Static", @@ -787,8 +793,6 @@ struct protocol proto_static = { .dump = static_dump, .start = static_start, .shutdown = static_shutdown, - .cleanup = static_cleanup, .reconfigure = static_reconfigure, .copy_config = static_copy_config, - .get_route_info = static_get_route_info, }; diff --git a/sysdep/bsd/krt-sock.c b/sysdep/bsd/krt-sock.c index 5c905bc9..6f788ac2 100644 --- a/sysdep/bsd/krt-sock.c +++ b/sysdep/bsd/krt-sock.c @@ -374,7 +374,6 @@ krt_read_route(struct ks_msg *msg, struct krt_proto *p, int scan) /* p is NULL iff KRT_SHARED_SOCKET and !scan */ int ipv6; - rte *e; net *net; sockaddr dst, gate, mask; ip_addr idst, igate, imask; @@ -495,7 +494,6 @@ krt_read_route(struct ks_msg *msg, struct krt_proto *p, int scan) net = net_get(p->p.main_channel->table, &ndst); rta a = { - .src = p->p.main_source, .source = RTS_INHERIT, .scope = SCOPE_UNIVERSE, }; @@ -549,19 +547,23 @@ krt_read_route(struct ks_msg *msg, struct krt_proto *p, int scan) } } - done: - e = rte_get_temp(&a); - e->net = net; - e->u.krt.src = src; - e->u.krt.proto = src2; - e->u.krt.seen = 0; - e->u.krt.best = 0; - e->u.krt.metric = 0; + done:; + rte e0 = { .attrs = &a, .net = net, }; + + ea_list *ea = alloca(sizeof(ea_list) + 1 * sizeof(eattr)); + *ea = (ea_list) { .count = 1, .next = e0.attrs->eattrs }; + e0.attrs->eattrs = ea; + + ea->attrs[0] = (eattr) { + .id = EA_KRT_SOURCE, + .type = EAF_TYPE_INT, + .u.data = src2, + }; if (scan) - krt_got_route(p, e); + krt_got_route(p, &e0, src); else - krt_got_route_async(p, e, new); + krt_got_route_async(p, &e0, new, src); } static void diff --git a/sysdep/linux/netlink.c b/sysdep/linux/netlink.c index fdf3f2db..bff2d579 100644 --- a/sysdep/linux/netlink.c +++ b/sysdep/linux/netlink.c @@ -105,7 +105,7 @@ struct nl_parse_state int scan; int merge; - net *net; + net_addr *net; rta *attrs; struct krt_proto *proto; s8 new; @@ -1233,10 +1233,9 @@ nh_bufsize(struct nexthop *nh) } static int -nl_send_route(struct krt_proto *p, rte *e, int op, int dest, struct nexthop *nh) +nl_send_route(struct krt_proto *p, const rte *e, int op, int dest, struct nexthop *nh) { eattr *ea; - net *net = e->net; rta *a = e->attrs; ea_list *eattrs = a->eattrs; int bufsize = 128 + KRT_METRICS_MAX*8 + nh_bufsize(&(a->nh)); @@ -1251,7 +1250,7 @@ nl_send_route(struct krt_proto *p, rte *e, int op, int dest, struct nexthop *nh) int rsize = sizeof(*r) + bufsize; r = alloca(rsize); - DBG("nl_send_route(%N,op=%x)\n", net->n.addr, op); + DBG("nl_send_route(%N,op=%x)\n", e->net, op); bzero(&r->h, sizeof(r->h)); bzero(&r->r, sizeof(r->r)); @@ -1260,7 +1259,7 @@ nl_send_route(struct krt_proto *p, rte *e, int op, int dest, struct nexthop *nh) r->h.nlmsg_flags = op | NLM_F_REQUEST | NLM_F_ACK; r->r.rtm_family = p->af; - r->r.rtm_dst_len = net_pxlen(net->n.addr); + r->r.rtm_dst_len = net_pxlen(e->net); r->r.rtm_protocol = RTPROT_BIRD; r->r.rtm_scope = RT_SCOPE_NOWHERE; #ifdef HAVE_MPLS_KERNEL @@ -1272,7 +1271,7 @@ nl_send_route(struct krt_proto *p, rte *e, int op, int dest, struct nexthop *nh) * 2) Never use RTA_PRIORITY */ - u32 label = net_mpls(net->n.addr); + u32 label = net_mpls(e->net); nl_add_attr_mpls(&r->h, rsize, RTA_DST, 1, &label); r->r.rtm_scope = RT_SCOPE_UNIVERSE; r->r.rtm_type = RTN_UNICAST; @@ -1280,12 +1279,12 @@ nl_send_route(struct krt_proto *p, rte *e, int op, int dest, struct nexthop *nh) else #endif { - nl_add_attr_ipa(&r->h, rsize, RTA_DST, net_prefix(net->n.addr)); + nl_add_attr_ipa(&r->h, rsize, RTA_DST, net_prefix(e->net)); /* Add source address for IPv6 SADR routes */ - if (net->n.addr->type == NET_IP6_SADR) + if (e->net->type == NET_IP6_SADR) { - net_addr_ip6_sadr *a = (void *) &net->n.addr; + net_addr_ip6_sadr *a = (void *) &e->net; nl_add_attr_ip6(&r->h, rsize, RTA_SRC, a->src_prefix); r->r.rtm_src_len = a->src_pxlen; } @@ -1305,8 +1304,6 @@ nl_send_route(struct krt_proto *p, rte *e, int op, int dest, struct nexthop *nh) if (p->af == AF_MPLS) priority = 0; - else if (a->source == RTS_DUMMY) - priority = e->u.krt.metric; else if (KRT_CF->sys.metric) priority = KRT_CF->sys.metric; else if ((op != NL_OP_DELETE) && (ea = ea_find(eattrs, EA_KRT_METRIC))) @@ -1408,7 +1405,7 @@ nl_add_rte(struct krt_proto *p, rte *e) } static inline int -nl_delete_rte(struct krt_proto *p, rte *e) +nl_delete_rte(struct krt_proto *p, const rte *e) { int err = 0; @@ -1429,7 +1426,7 @@ nl_replace_rte(struct krt_proto *p, rte *e) void -krt_replace_rte(struct krt_proto *p, net *n UNUSED, rte *new, rte *old) +krt_replace_rte(struct krt_proto *p, const net_addr *n UNUSED, rte *new, const rte *old) { int err = 0; @@ -1468,7 +1465,7 @@ krt_replace_rte(struct krt_proto *p, net *n UNUSED, rte *new, rte *old) } static int -nl_mergable_route(struct nl_parse_state *s, net *net, struct krt_proto *p, uint priority, uint krt_type, uint rtm_family) +nl_mergable_route(struct nl_parse_state *s, const net_addr *net, struct krt_proto *p, uint priority, uint krt_type, uint rtm_family) { /* Route merging is used for IPv6 scans */ if (!s->scan || (rtm_family != AF_INET6)) @@ -1488,18 +1485,30 @@ nl_mergable_route(struct nl_parse_state *s, net *net, struct krt_proto *p, uint static void nl_announce_route(struct nl_parse_state *s) { - rte *e = rte_get_temp(s->attrs); - e->net = s->net; - e->u.krt.src = s->krt_src; - e->u.krt.proto = s->krt_proto; - e->u.krt.seen = 0; - e->u.krt.best = 0; - e->u.krt.metric = s->krt_metric; + rte e0 = { + .attrs = s->attrs, + .net = s->net, + }; + + ea_list *ea = alloca(sizeof(ea_list) + 2 * sizeof(eattr)); + *ea = (ea_list) { .count = 2, .next = e0.attrs->eattrs }; + e0.attrs->eattrs = ea; + + ea->attrs[0] = (eattr) { + .id = EA_KRT_SOURCE, + .type = EAF_TYPE_INT, + .u.data = s->krt_proto, + }; + ea->attrs[1] = (eattr) { + .id = EA_KRT_METRIC, + .type = EAF_TYPE_INT, + .u.data = s->krt_metric, + }; if (s->scan) - krt_got_route(s->proto, e); + krt_got_route(s->proto, &e0, s->krt_src); else - krt_got_route_async(s->proto, e, s->new); + krt_got_route_async(s->proto, &e0, s->new, s->krt_src); s->net = NULL; s->attrs = NULL; @@ -1645,21 +1654,18 @@ nl_parse_route(struct nl_parse_state *s, struct nlmsghdr *h) krt_src = KRT_SRC_ALIEN; } - net_addr *n = &dst; + net_addr *net = &dst; if (p->p.net_type == NET_IP6_SADR) { - n = alloca(sizeof(net_addr_ip6_sadr)); - net_fill_ip6_sadr(n, net6_prefix(&dst), net6_pxlen(&dst), + net = alloca(sizeof(net_addr_ip6_sadr)); + net_fill_ip6_sadr(net, net6_prefix(&dst), net6_pxlen(&dst), net6_prefix(&src), net6_pxlen(&src)); } - net *net = net_get(p->p.main_channel->table, n); - if (s->net && !nl_mergable_route(s, net, p, priority, i->rtm_type, i->rtm_family)) nl_announce_route(s); rta *ra = lp_allocz(s->pool, RTA_MAX_SIZE); - ra->src = p->p.main_source; ra->source = RTS_INHERIT; ra->scope = SCOPE_UNIVERSE; @@ -1678,7 +1684,7 @@ nl_parse_route(struct nl_parse_state *s, struct nlmsghdr *h) struct nexthop *nh = nl_parse_multipath(s, p, a[RTA_MULTIPATH], i->rtm_family); if (!nh) { - log(L_ERR "KRT: Received strange multipath route %N", net->n.addr); + log(L_ERR "KRT: Received strange multipath route %N", net); return; } @@ -1692,7 +1698,7 @@ nl_parse_route(struct nl_parse_state *s, struct nlmsghdr *h) ra->nh.iface = if_find_by_index(oif); if (!ra->nh.iface) { - log(L_ERR "KRT: Received route %N with unknown ifindex %u", net->n.addr, oif); + log(L_ERR "KRT: Received route %N with unknown ifindex %u", net, oif); return; } @@ -1719,8 +1725,7 @@ nl_parse_route(struct nl_parse_state *s, struct nlmsghdr *h) (ra->nh.flags & RNF_ONLINK) ? NEF_ONLINK : 0); if (!nbr || (nbr->scope == SCOPE_HOST)) { - log(L_ERR "KRT: Received route %N with strange next-hop %I", net->n.addr, - ra->nh.gw); + log(L_ERR "KRT: Received route %N with strange next-hop %I", net, ra->nh.gw); return; } } @@ -1819,7 +1824,7 @@ nl_parse_route(struct nl_parse_state *s, struct nlmsghdr *h) if (nl_parse_metrics(a[RTA_METRICS], metrics, ARRAY_SIZE(metrics)) < 0) { - log(L_ERR "KRT: Received route %N with strange RTA_METRICS attribute", net->n.addr); + log(L_ERR "KRT: Received route %N with strange RTA_METRICS attribute", net); return; } @@ -1853,7 +1858,9 @@ nl_parse_route(struct nl_parse_state *s, struct nlmsghdr *h) if (!s->net) { /* Store the new route */ - s->net = net; + s->net = lp_alloc(s->pool, net->length); + net_copy(s->net, net); + s->attrs = ra; s->proto = p; s->new = new; diff --git a/sysdep/unix/Makefile b/sysdep/unix/Makefile index d0d36b5f..07f454ab 100644 --- a/sysdep/unix/Makefile +++ b/sysdep/unix/Makefile @@ -1,4 +1,4 @@ -src := alloc.c io.c krt.c log.c main.c random.c +src := alloc.c io.c io-loop.c krt.c log.c main.c random.c coroutine.c obj := $(src-o-files) $(all-daemon) $(cf-local) diff --git a/sysdep/unix/alloc.c b/sysdep/unix/alloc.c index 5dd70c99..d18f286b 100644 --- a/sysdep/unix/alloc.c +++ b/sysdep/unix/alloc.c @@ -11,69 +11,59 @@ #include "lib/lists.h" #include "lib/event.h" +#include "sysdep/unix/io-loop.h" + #include <stdlib.h> #include <unistd.h> +#include <stdatomic.h> +#include <errno.h> #ifdef HAVE_MMAP #include <sys/mman.h> #endif -#ifdef HAVE_MMAP -#define KEEP_PAGES 512 +long page_size = 0; -static u64 page_size = 0; +#ifdef HAVE_MMAP +#if DEBUGGING +#define FP_NODE_OFFSET 42 +#else +#define FP_NODE_OFFSET 1 +#endif static _Bool use_fake = 0; - -uint pages_kept = 0; -static list pages_list; - -static void cleanup_pages(void *data); -static event page_cleanup_event = { .hook = cleanup_pages }; - #else -static const u64 page_size = 4096; /* Fake page size */ +static _Bool use_fake = 1; #endif -u64 get_page_size(void) +static void * +alloc_sys_page(void) { - if (page_size) - return page_size; + void *ptr = mmap(NULL, page_size, PROT_WRITE | PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); -#ifdef HAVE_MMAP - if (page_size = sysconf(_SC_PAGESIZE)) - { - if ((u64_popcount(page_size) > 1) || (page_size > 16384)) - { - /* Too big or strange page, use the aligned allocator instead */ - page_size = 4096; - use_fake = 1; - } - return page_size; - } + if (ptr == MAP_FAILED) + bug("mmap(%lu) failed: %m", page_size); - bug("Page size must be non-zero"); -#endif + return ptr; } void * alloc_page(void) { #ifdef HAVE_MMAP - if (pages_kept) - { - node *page = TAIL(pages_list); - rem_node(page); - pages_kept--; - memset(page, 0, get_page_size()); - return page; - } - if (!use_fake) { - void *ret = mmap(NULL, get_page_size(), PROT_WRITE | PROT_READ, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); - if (ret == MAP_FAILED) - bug("mmap(%lu) failed: %m", page_size); - return ret; + struct free_pages *fp = &birdloop_current->pages; + if (!fp->cnt) + return alloc_sys_page(); + + node *n = HEAD(fp->list); + rem_node(n); + if (--fp->cnt < fp->min) + ev_send(&global_work_list, fp->cleanup); + + void *ptr = n - FP_NODE_OFFSET; + memset(ptr, 0, page_size); + return ptr; } else #endif @@ -91,14 +81,14 @@ free_page(void *ptr) #ifdef HAVE_MMAP if (!use_fake) { - if (!pages_kept) - init_list(&pages_list); - - memset(ptr, 0, sizeof(node)); - add_tail(&pages_list, ptr); - - if (++pages_kept > KEEP_PAGES) - ev_schedule(&page_cleanup_event); + struct free_pages *fp = &birdloop_current->pages; + struct node *n = ptr; + n += FP_NODE_OFFSET; + + memset(n, 0, sizeof(node)); + add_tail(&fp->list, n); + if (++fp->cnt > fp->max) + ev_send(&global_work_list, fp->cleanup); } else #endif @@ -106,24 +96,136 @@ free_page(void *ptr) } #ifdef HAVE_MMAP + +#define GFP (&main_birdloop.pages) + +void +flush_pages(struct birdloop *loop) +{ + ASSERT_DIE(birdloop_inside(&main_birdloop)); + + add_tail_list(&GFP->list, &loop->pages.list); + GFP->cnt += loop->pages.cnt; + + loop->pages.cnt = 0; + loop->pages.list = (list) {}; + loop->pages.min = 0; + loop->pages.max = 0; + + rfree(loop->pages.cleanup); + loop->pages.cleanup = NULL; +} + static void -cleanup_pages(void *data UNUSED) +cleanup_pages(void *data) { - for (uint seen = 0; (pages_kept > KEEP_PAGES) && (seen < KEEP_PAGES); seen++) + struct birdloop *loop = data; + birdloop_enter(loop); + + struct free_pages *fp = &birdloop_current->pages; + + while ((fp->cnt < fp->min) && (GFP->cnt > GFP->min)) { - void *ptr = HEAD(pages_list); - rem_node(ptr); - if (munmap(ptr, get_page_size()) == 0) - pages_kept--; -#ifdef ENOMEM + node *n = HEAD(GFP->list); + rem_node(n); + add_tail(&fp->list, n); + fp->cnt++; + GFP->cnt--; + } + + while (fp->cnt < fp->min) + { + node *n = alloc_sys_page(); + add_tail(&fp->list, n + FP_NODE_OFFSET); + fp->cnt++; + } + + while (fp->cnt > fp->max) + { + node *n = HEAD(fp->list); + rem_node(n); + add_tail(&GFP->list, n); + fp->cnt--; + GFP->cnt++; + } + + birdloop_leave(loop); + + if (GFP->cnt > GFP->max) + ev_send(&global_work_list, GFP->cleanup); +} + +static void +cleanup_global_pages(void *data UNUSED) +{ + while (GFP->cnt < GFP->max) + { + node *n = alloc_sys_page(); + add_tail(&GFP->list, n + FP_NODE_OFFSET); + GFP->cnt++; + } + + for (uint limit = GFP->cnt; (limit > 0) && (GFP->cnt > GFP->max); limit--) + { + node *n = TAIL(GFP->list); + rem_node(n); + + if (munmap(n - FP_NODE_OFFSET, page_size) == 0) + GFP->cnt--; else if (errno == ENOMEM) - add_tail(&pages_list, ptr); -#endif + add_head(&GFP->list, n); else - bug("munmap(%p) failed: %m", ptr); + bug("munmap(%p) failed: %m", n - FP_NODE_OFFSET); } +} + +void +init_pages(struct birdloop *loop) +{ + struct free_pages *fp = &loop->pages; + + init_list(&fp->list); + fp->cleanup = ev_new_init(&root_pool, cleanup_pages, loop); + fp->min = 4; + fp->max = 16; - if (pages_kept > KEEP_PAGES) - ev_schedule(&page_cleanup_event); + for (fp->cnt = 0; fp->cnt < fp->min; fp->cnt++) + { + node *n = alloc_sys_page(); + add_tail(&fp->list, n + FP_NODE_OFFSET); + } } + +static event global_free_pages_cleanup_event = { .hook = cleanup_global_pages }; + +void resource_sys_init(void) +{ + if (!(page_size = sysconf(_SC_PAGESIZE))) + die("System page size must be non-zero"); + + if (u64_popcount(page_size) == 1) + { + init_list(&GFP->list); + GFP->cleanup = &global_free_pages_cleanup_event; + GFP->min = 0; + GFP->max = 256; + return; + } + + log(L_WARN "Got strange memory page size (%lu), using the aligned allocator instead", page_size); + + /* Too big or strange page, use the aligned allocator instead */ + page_size = 4096; + use_fake = 1; +} + +#else + +void +resource_sys_init(void) +{ + page_size = 4096; + use_fake = 1; +} + #endif diff --git a/sysdep/unix/coroutine.c b/sysdep/unix/coroutine.c new file mode 100644 index 00000000..e4657157 --- /dev/null +++ b/sysdep/unix/coroutine.c @@ -0,0 +1,205 @@ +/* + * BIRD Coroutines + * + * (c) 2017 Martin Mares <mj@ucw.cz> + * (c) 2020 Maria Matejka <mq@jmq.cz> + * + * Can be freely distributed and used under the terms of the GNU GPL. + */ + +#ifndef _GNU_SOURCE +#define _GNU_SOURCE +#endif + +#undef LOCAL_DEBUG + +#undef DEBUG_LOCKING + +#include "lib/birdlib.h" +#include "lib/locking.h" +#include "lib/coro.h" +#include "lib/rcu.h" +#include "lib/resource.h" +#include "lib/timer.h" + +#include "conf/conf.h" + +#define CORO_STACK_SIZE 65536 + +/* + * Implementation of coroutines based on POSIX threads + */ + +#include <errno.h> +#include <fcntl.h> +#include <poll.h> +#include <pthread.h> +#include <semaphore.h> +#include <stdatomic.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> + +/* + * Locking subsystem + */ + +_Thread_local struct lock_order locking_stack = {}; +_Thread_local struct domain_generic **last_locked = NULL; + +#define ASSERT_NO_LOCK ASSERT_DIE(last_locked == NULL) + +struct domain_generic { + pthread_mutex_t mutex; + uint order; + struct domain_generic **prev; + struct lock_order *locked_by; + const char *name; +}; + +#define DOMAIN_INIT(_name, _order) { .mutex = PTHREAD_MUTEX_INITIALIZER, .name = _name, .order = _order } + +static struct domain_generic the_bird_domain_gen = DOMAIN_INIT("The BIRD", OFFSETOF(struct lock_order, the_bird)); + +DOMAIN(the_bird) the_bird_domain = { .the_bird = &the_bird_domain_gen }; + +struct domain_generic * +domain_new(const char *name, uint order) +{ + ASSERT_DIE(order < sizeof(struct lock_order)); + struct domain_generic *dg = xmalloc(sizeof(struct domain_generic)); + *dg = (struct domain_generic) DOMAIN_INIT(name, order); + return dg; +} + +void +domain_free(struct domain_generic *dg) +{ + pthread_mutex_destroy(&dg->mutex); + xfree(dg); +} + +uint dg_order(struct domain_generic *dg) +{ + return dg->order; +} + +void do_lock(struct domain_generic *dg, struct domain_generic **lsp) +{ + if ((char *) lsp - (char *) &locking_stack != dg->order) + bug("Trying to lock on bad position: order=%u, lsp=%p, base=%p", dg->order, lsp, &locking_stack); + + if (lsp <= last_locked) + bug("Trying to lock in a bad order"); + if (*lsp) + bug("Inconsistent locking stack state on lock"); + + btime lock_begin = current_time(); + pthread_mutex_lock(&dg->mutex); + btime duration = current_time() - lock_begin; + if (config && (duration > config->watchdog_warning)) + log(L_WARN "Locking of %s took %d ms", dg->name, (int) (duration TO_MS)); + + if (dg->prev || dg->locked_by) + bug("Previous unlock not finished correctly"); + dg->prev = last_locked; + *lsp = dg; + last_locked = lsp; + dg->locked_by = &locking_stack; +} + +void do_unlock(struct domain_generic *dg, struct domain_generic **lsp) +{ + if ((char *) lsp - (char *) &locking_stack != dg->order) + bug("Trying to unlock on bad position: order=%u, lsp=%p, base=%p", dg->order, lsp, &locking_stack); + + if (dg->locked_by != &locking_stack) + bug("Inconsistent domain state on unlock"); + if ((last_locked != lsp) || (*lsp != dg)) + bug("Inconsistent locking stack state on unlock"); + dg->locked_by = NULL; + last_locked = dg->prev; + *lsp = NULL; + dg->prev = NULL; + pthread_mutex_unlock(&dg->mutex); +} + +/* Coroutines */ +struct coroutine { + resource r; + pthread_t id; + pthread_attr_t attr; + struct rcu_coro rcu; + void (*entry)(void *); + void *data; +}; + +static _Thread_local _Bool coro_cleaned_up = 0; + +static void coro_free(resource *r) +{ + struct coroutine *c = (void *) r; + rcu_coro_stop(&c->rcu); + ASSERT_DIE(pthread_equal(pthread_self(), c->id)); + pthread_attr_destroy(&c->attr); + coro_cleaned_up = 1; +} + +static void coro_dump(resource *r UNUSED) { } + +static struct resclass coro_class = { + .name = "Coroutine", + .size = sizeof(struct coroutine), + .free = coro_free, + .dump = coro_dump, +}; + +_Thread_local struct coroutine *this_coro = NULL; + +static void *coro_entry(void *p) +{ + struct coroutine *c = p; + + ASSERT_DIE(c->entry); + + this_coro = c; + rcu_coro_start(&c->rcu); + + c->entry(c->data); + ASSERT_DIE(coro_cleaned_up); + + return NULL; +} + +struct coroutine *coro_run(pool *p, void (*entry)(void *), void *data) +{ + ASSERT_DIE(entry); + ASSERT_DIE(p); + + struct coroutine *c = ralloc(p, &coro_class); + + c->entry = entry; + c->data = data; + + int e = 0; + + if (e = pthread_attr_init(&c->attr)) + die("pthread_attr_init() failed: %M", e); + + if (e = pthread_attr_setstacksize(&c->attr, CORO_STACK_SIZE)) + die("pthread_attr_setstacksize(%u) failed: %M", CORO_STACK_SIZE, e); + + if (e = pthread_attr_setdetachstate(&c->attr, PTHREAD_CREATE_DETACHED)) + die("pthread_attr_setdetachstate(PTHREAD_CREATE_DETACHED) failed: %M", e); + + if (e = pthread_create(&c->id, &c->attr, coro_entry, c)) + die("pthread_create() failed: %M", e); + + return c; +} + +void +coro_yield(void) +{ + usleep(100); +} diff --git a/sysdep/unix/io-loop.c b/sysdep/unix/io-loop.c new file mode 100644 index 00000000..a9927b39 --- /dev/null +++ b/sysdep/unix/io-loop.c @@ -0,0 +1,583 @@ +/* + * BIRD -- I/O and event loop + * + * Can be freely distributed and used under the terms of the GNU GPL. + */ + +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <errno.h> +#include <fcntl.h> +#include <poll.h> +#include <pthread.h> +#include <time.h> +#include <sys/time.h> + +#include "nest/bird.h" + +#include "lib/buffer.h" +#include "lib/coro.h" +#include "lib/lists.h" +#include "lib/resource.h" +#include "lib/event.h" +#include "lib/timer.h" +#include "lib/socket.h" + +#include "lib/io-loop.h" +#include "sysdep/unix/io-loop.h" +#include "conf/conf.h" + +/* + * Current thread context + */ + +_Thread_local struct birdloop *birdloop_current = NULL; +static _Thread_local struct birdloop *birdloop_wakeup_masked; +static _Thread_local uint birdloop_wakeup_masked_count; + +event_list * +birdloop_event_list(struct birdloop *loop) +{ + return &loop->event_list; +} + +struct timeloop * +birdloop_time_loop(struct birdloop *loop) +{ + return &loop->time; +} + +pool * +birdloop_pool(struct birdloop *loop) +{ + return loop->pool; +} + +_Bool +birdloop_inside(struct birdloop *loop) +{ + for (struct birdloop *c = birdloop_current; c; c = c->prev_loop) + if (loop == c) + return 1; + + return 0; +} + +/* + * Wakeup code for birdloop + */ + +static void +pipe_new(int *pfds) +{ + int rv = pipe(pfds); + if (rv < 0) + die("pipe: %m"); + + if (fcntl(pfds[0], F_SETFL, O_NONBLOCK) < 0) + die("fcntl(O_NONBLOCK): %m"); + + if (fcntl(pfds[1], F_SETFL, O_NONBLOCK) < 0) + die("fcntl(O_NONBLOCK): %m"); +} + +void +pipe_drain(int fd) +{ + char buf[64]; + int rv; + + try: + rv = read(fd, buf, 64); + if (rv < 0) + { + if (errno == EINTR) + goto try; + if (errno == EAGAIN) + return; + die("wakeup read: %m"); + } + if (rv == 64) + goto try; +} + +void +pipe_kick(int fd) +{ + u64 v = 1; + int rv; + + try: + rv = write(fd, &v, sizeof(u64)); + if (rv < 0) + { + if (errno == EINTR) + goto try; + if (errno == EAGAIN) + return; + die("wakeup write: %m"); + } +} + +static inline void +wakeup_init(struct birdloop *loop) +{ + pipe_new(loop->wakeup_fds); +} + +static inline void +wakeup_drain(struct birdloop *loop) +{ + pipe_drain(loop->wakeup_fds[0]); +} + +static inline void +wakeup_do_kick(struct birdloop *loop) +{ + pipe_kick(loop->wakeup_fds[1]); +} + +void +birdloop_ping(struct birdloop *loop) +{ + u32 ping_sent = atomic_fetch_add_explicit(&loop->ping_sent, 1, memory_order_acq_rel); + if (ping_sent) + return; + + if (loop == birdloop_wakeup_masked) + birdloop_wakeup_masked_count++; + else + wakeup_do_kick(loop); +} + + +/* + * Sockets + */ + +static void +sockets_init(struct birdloop *loop) +{ + init_list(&loop->sock_list); + loop->sock_num = 0; + + BUFFER_INIT(loop->poll_sk, loop->pool, 4); + BUFFER_INIT(loop->poll_fd, loop->pool, 4); + loop->poll_changed = 1; /* add wakeup fd */ +} + +static void +sockets_add(struct birdloop *loop, sock *s) +{ + add_tail(&loop->sock_list, &s->n); + loop->sock_num++; + + s->index = -1; + loop->poll_changed = 1; + + birdloop_ping(loop); +} + +void +sk_start(sock *s) +{ + ASSERT_DIE(birdloop_current != &main_birdloop); + sockets_add(birdloop_current, s); +} + +static void +sockets_remove(struct birdloop *loop, sock *s) +{ + rem_node(&s->n); + loop->sock_num--; + + if (s->index >= 0) + { + loop->poll_sk.data[s->index] = NULL; + s->index = -1; + loop->poll_changed = 1; + loop->close_scheduled = 1; + birdloop_ping(loop); + } + else + close(s->fd); +} + +void +sk_stop(sock *s) +{ + sockets_remove(birdloop_current, s); +} + +static inline uint sk_want_events(sock *s) +{ return ((s->rx_hook && !ev_corked(s->cork)) ? POLLIN : 0) | ((s->ttx != s->tpos) ? POLLOUT : 0); } + +/* +FIXME: this should be called from sock code + +static void +sockets_update(struct birdloop *loop, sock *s) +{ + if (s->index >= 0) + loop->poll_fd.data[s->index].events = sk_want_events(s); +} +*/ + +static void +sockets_prepare(struct birdloop *loop) +{ + BUFFER_SET(loop->poll_sk, loop->sock_num + 1); + BUFFER_SET(loop->poll_fd, loop->sock_num + 1); + + struct pollfd *pfd = loop->poll_fd.data; + sock **psk = loop->poll_sk.data; + uint i = 0; + node *n; + + WALK_LIST(n, loop->sock_list) + { + sock *s = SKIP_BACK(sock, n, n); + + ASSERT(i < loop->sock_num); + + s->index = i; + *psk = s; + pfd->fd = s->fd; + pfd->events = sk_want_events(s); + pfd->revents = 0; + + pfd++; + psk++; + i++; + } + + ASSERT(i == loop->sock_num); + + /* Add internal wakeup fd */ + *psk = NULL; + pfd->fd = loop->wakeup_fds[0]; + pfd->events = POLLIN; + pfd->revents = 0; + + loop->poll_changed = 0; +} + +static void +sockets_close_fds(struct birdloop *loop) +{ + struct pollfd *pfd = loop->poll_fd.data; + sock **psk = loop->poll_sk.data; + int poll_num = loop->poll_fd.used - 1; + + int i; + for (i = 0; i < poll_num; i++) + if (psk[i] == NULL) + close(pfd[i].fd); + + loop->close_scheduled = 0; +} + +int sk_read(sock *s, int revents); +int sk_write(sock *s); + +static void +sockets_fire(struct birdloop *loop) +{ + struct pollfd *pfd = loop->poll_fd.data; + sock **psk = loop->poll_sk.data; + int poll_num = loop->poll_fd.used - 1; + + times_update(); + + /* Last fd is internal wakeup fd */ + if (pfd[poll_num].revents & POLLIN) + wakeup_drain(loop); + + int i; + for (i = 0; i < poll_num; pfd++, psk++, i++) + { + int e = 1; + + if (! pfd->revents) + continue; + + if (pfd->revents & POLLNVAL) + die("poll: invalid fd %d", pfd->fd); + + if (pfd->revents & POLLIN) + while (e && *psk && (*psk)->rx_hook) + e = sk_read(*psk, pfd->revents); + + e = 1; + if (pfd->revents & POLLOUT) + { + loop->poll_changed = 1; + while (e && *psk) + e = sk_write(*psk); + } + } +} + + +/* + * Birdloop + */ + +struct birdloop main_birdloop; + +static void birdloop_enter_locked(struct birdloop *loop); + +void +birdloop_init(void) +{ + wakeup_init(&main_birdloop); + + main_birdloop.time.domain = the_bird_domain.the_bird; + main_birdloop.time.loop = &main_birdloop; + + times_update(); + timers_init(&main_birdloop.time, &root_pool); + + root_pool.loop = &main_birdloop; + main_birdloop.pool = &root_pool; + + birdloop_enter_locked(&main_birdloop); +} + +static void birdloop_main(void *arg); + +void +birdloop_free(resource *r) +{ + struct birdloop *loop = (void *) r; + + ASSERT_DIE(loop->links == 0); + domain_free(loop->time.domain); +} + +void +birdloop_dump(resource *r) +{ + struct birdloop *loop = (void *) r; + + debug("%s\n", loop->pool->name); +} + +struct resclass birdloop_class = { + .name = "IO Loop", + .size = sizeof(struct birdloop), + .free = birdloop_free, + .dump = birdloop_dump, +}; + +struct birdloop * +birdloop_new(pool *pp, uint order, const char *name) +{ + struct domain_generic *dg = domain_new(name, order); + + struct birdloop *loop = ralloc(pp, &birdloop_class); + + loop->time.domain = dg; + loop->time.loop = loop; + + birdloop_enter(loop); + + loop->pool = rp_new(pp, loop, name); + loop->parent = pp; + rmove(&loop->r, loop->pool); + + wakeup_init(loop); + ev_init_list(&loop->event_list, loop, name); + timers_init(&loop->time, loop->pool); + sockets_init(loop); + + init_pages(loop); + + loop->time.coro = coro_run(loop->pool, birdloop_main, loop); + + birdloop_leave(loop); + + return loop; +} + +static void +birdloop_do_stop(struct birdloop *loop, void (*stopped)(void *data), void *data) +{ + loop->stopped = stopped; + loop->stop_data = data; + wakeup_do_kick(loop); +} + +void +birdloop_stop(struct birdloop *loop, void (*stopped)(void *data), void *data) +{ + DG_LOCK(loop->time.domain); + birdloop_do_stop(loop, stopped, data); + DG_UNLOCK(loop->time.domain); +} + +void +birdloop_stop_self(struct birdloop *loop, void (*stopped)(void *data), void *data) +{ + ASSERT_DIE(loop == birdloop_current); + ASSERT_DIE(DG_IS_LOCKED(loop->time.domain)); + + birdloop_do_stop(loop, stopped, data); +} + +static void +birdloop_enter_locked(struct birdloop *loop) +{ + ASSERT_DIE(DG_IS_LOCKED(loop->time.domain)); + ASSERT_DIE(!birdloop_inside(loop)); + + /* Store the old context */ + loop->prev_loop = birdloop_current; + + /* Put the new context */ + birdloop_current = loop; +} + +void +birdloop_enter(struct birdloop *loop) +{ + DG_LOCK(loop->time.domain); + return birdloop_enter_locked(loop); +} + +static void +birdloop_leave_locked(struct birdloop *loop) +{ + /* Check the current context */ + ASSERT_DIE(birdloop_current == loop); + + /* Restore the old context */ + birdloop_current = loop->prev_loop; +} + +void +birdloop_leave(struct birdloop *loop) +{ + birdloop_leave_locked(loop); + DG_UNLOCK(loop->time.domain); +} + +void +birdloop_mask_wakeups(struct birdloop *loop) +{ + ASSERT_DIE(birdloop_wakeup_masked == NULL); + birdloop_wakeup_masked = loop; +} + +void +birdloop_unmask_wakeups(struct birdloop *loop) +{ + ASSERT_DIE(birdloop_wakeup_masked == loop); + birdloop_wakeup_masked = NULL; + if (birdloop_wakeup_masked_count) + wakeup_do_kick(loop); + + birdloop_wakeup_masked_count = 0; +} + +void +birdloop_link(struct birdloop *loop) +{ + ASSERT_DIE(birdloop_inside(loop)); + loop->links++; +} + +void +birdloop_unlink(struct birdloop *loop) +{ + ASSERT_DIE(birdloop_inside(loop)); + loop->links--; +} + +static void +birdloop_main(void *arg) +{ + struct birdloop *loop = arg; + timer *t; + int rv, timeout; + + btime loop_begin = current_time(); + + birdloop_enter(loop); + while (1) + { + timers_fire(&loop->time, 0); + if (ev_run_list(&loop->event_list)) + timeout = 0; + else if (t = timers_first(&loop->time)) + timeout = (tm_remains(t) TO_MS) + 1; + else + timeout = -1; + + if (loop->poll_changed) + sockets_prepare(loop); + + btime duration = current_time() - loop_begin; + if (duration > config->watchdog_warning) + log(L_WARN "I/O loop cycle took %d ms", (int) (duration TO_MS)); + + birdloop_leave(loop); + + try: + rv = poll(loop->poll_fd.data, loop->poll_fd.used, timeout); + if (rv < 0) + { + if (errno == EINTR || errno == EAGAIN) + goto try; + die("poll: %m"); + } + + birdloop_enter(loop); + + if (loop->close_scheduled) + sockets_close_fds(loop); + + if (loop->stopped) + break; + + loop_begin = current_time(); + + if (rv) + sockets_fire(loop); + + atomic_exchange_explicit(&loop->ping_sent, 0, memory_order_acq_rel); + } + + /* Flush remaining events */ + ASSERT_DIE(!ev_run_list(&loop->event_list)); + + /* No timers allowed */ + ASSERT_DIE(timers_count(&loop->time) == 0); + ASSERT_DIE(EMPTY_LIST(loop->sock_list)); + ASSERT_DIE(loop->sock_num == 0); + + birdloop_leave(loop); + + /* Lock parent loop */ + pool *parent = loop->parent; + birdloop_enter(parent->loop); + + /* Move the loop temporarily to parent pool */ + birdloop_enter(loop); + rmove(&loop->r, parent); + birdloop_leave(loop); + + /* Announce loop stop */ + loop->stopped(loop->stop_data); + + /* Free the pool and loop */ + birdloop_enter(loop); + rp_free(loop->pool, parent); + flush_pages(loop); + birdloop_leave(loop); + rfree(&loop->r); + + /* And finally leave the parent loop before finishing */ + birdloop_leave(parent->loop); +} diff --git a/sysdep/unix/io-loop.h b/sysdep/unix/io-loop.h new file mode 100644 index 00000000..e5af52d1 --- /dev/null +++ b/sysdep/unix/io-loop.h @@ -0,0 +1,59 @@ +/* + * BIRD -- I/O and event loop + * + * Can be freely distributed and used under the terms of the GNU GPL. + */ + +#ifndef _BIRD_SYSDEP_UNIX_IO_LOOP_H_ +#define _BIRD_SYSDEP_UNIX_IO_LOOP_H_ + +#include "nest/bird.h" + +#include "lib/lists.h" +#include "lib/event.h" +#include "lib/timer.h" + +struct free_pages +{ + list list; /* List of empty pages */ + event *cleanup; /* Event to call when number of pages is outside bounds */ + u16 min, max; /* Minimal and maximal number of free pages kept */ + uint cnt; /* Number of empty pages */ +}; + +struct birdloop +{ + resource r; + + pool *pool; + pool *parent; + + struct timeloop time; + event_list event_list; + list sock_list; + uint sock_num; + + BUFFER(sock *) poll_sk; + BUFFER(struct pollfd) poll_fd; + u8 poll_changed; + u8 close_scheduled; + + _Atomic u32 ping_sent; + int wakeup_fds[2]; + + uint links; + + struct free_pages pages; + + void (*stopped)(void *data); + void *stop_data; + + struct birdloop *prev_loop; +}; + +extern _Thread_local struct birdloop *birdloop_current; + +void init_pages(struct birdloop *loop); +void flush_pages(struct birdloop *loop); + +#endif diff --git a/sysdep/unix/io.c b/sysdep/unix/io.c index 3d67d0a7..eee7b586 100644 --- a/sysdep/unix/io.c +++ b/sysdep/unix/io.c @@ -36,12 +36,14 @@ #include "lib/resource.h" #include "lib/socket.h" #include "lib/event.h" +#include "lib/locking.h" #include "lib/timer.h" #include "lib/string.h" #include "nest/iface.h" #include "conf/conf.h" #include "sysdep/unix/unix.h" +#include "sysdep/unix/io-loop.h" #include CONFIG_INCLUDE_SYSIO_H /* Maximum number of calls of tx handler for one socket in one @@ -122,55 +124,50 @@ rf_fileno(struct rfile *f) btime boot_time; + void -times_init(struct timeloop *loop) +times_update(void) { struct timespec ts; int rv; + btime old_time = current_time(); + btime old_real_time = current_real_time(); + rv = clock_gettime(CLOCK_MONOTONIC, &ts); if (rv < 0) die("Monotonic clock is missing"); if ((ts.tv_sec < 0) || (((u64) ts.tv_sec) > ((u64) 1 << 40))) log(L_WARN "Monotonic clock is crazy"); - - loop->last_time = ts.tv_sec S + ts.tv_nsec NS; - loop->real_time = 0; -} - -void -times_update(struct timeloop *loop) -{ - struct timespec ts; - int rv; - - rv = clock_gettime(CLOCK_MONOTONIC, &ts); - if (rv < 0) - die("clock_gettime: %m"); - + btime new_time = ts.tv_sec S + ts.tv_nsec NS; - if (new_time < loop->last_time) + if (new_time < old_time) log(L_ERR "Monotonic clock is broken"); - loop->last_time = new_time; - loop->real_time = 0; -} - -void -times_update_real_time(struct timeloop *loop) -{ - struct timespec ts; - int rv; - rv = clock_gettime(CLOCK_REALTIME, &ts); if (rv < 0) die("clock_gettime: %m"); - loop->real_time = ts.tv_sec S + ts.tv_nsec NS; -} + btime new_real_time = ts.tv_sec S + ts.tv_nsec NS; + if (!atomic_compare_exchange_strong_explicit( + &last_time, + &old_time, + new_time, + memory_order_acq_rel, + memory_order_relaxed)) + DBG("Time update collision: last_time"); + + if (!atomic_compare_exchange_strong_explicit( + &real_time, + &old_real_time, + new_real_time, + memory_order_acq_rel, + memory_order_relaxed)) + DBG("Time update collision: real_time"); +} /** * DOC: Sockets @@ -804,18 +801,16 @@ sk_free(resource *r) sk_ssh_free(s); #endif - if (s->fd < 0) + if ((s->fd < 0) || (s->flags & SKF_THREAD)) return; - /* FIXME: we should call sk_stop() for SKF_THREAD sockets */ - if (!(s->flags & SKF_THREAD)) - { - if (s == current_sock) - current_sock = sk_next(s); - if (s == stored_sock) - stored_sock = sk_next(s); + if (s == current_sock) + current_sock = sk_next(s); + if (s == stored_sock) + stored_sock = sk_next(s); + + if (enlisted(&s->n)) rem_node(&s->n); - } if (s->type != SK_SSH && s->type != SK_SSH_ACTIVE) close(s->fd); @@ -1108,7 +1103,11 @@ sk_passive_connected(sock *s, int type) return 1; } - sk_insert(t); + if (s->flags & SKF_PASSIVE_THREAD) + t->flags |= SKF_THREAD; + else + sk_insert(t); + sk_alloc_bufs(t); s->rx_hook(t, 0); return 1; @@ -1512,6 +1511,36 @@ sk_open_unix(sock *s, char *name) return 0; } +static void +sk_reloop_hook(void *_vs) +{ + sock *s = _vs; + if (birdloop_inside(&main_birdloop)) + { + s->flags &= ~SKF_THREAD; + sk_insert(s); + } + else + { + s->flags |= SKF_THREAD; + sk_start(s); + } +} + +void +sk_reloop(sock *s, struct birdloop *loop) +{ + if (enlisted(&s->n)) + rem_node(&s->n); + + s->reloop = (event) { + .hook = sk_reloop_hook, + .data = s, + }; + + ev_send_loop(loop, &s->reloop); +} + #define CMSG_RX_SPACE MAX(CMSG4_SPACE_PKTINFO+CMSG4_SPACE_TTL, \ CMSG6_SPACE_PKTINFO+CMSG6_SPACE_TTL) @@ -2016,30 +2045,17 @@ struct event_log_entry static struct event_log_entry event_log[EVENT_LOG_LENGTH]; static struct event_log_entry *event_open; static int event_log_pos, event_log_num, watchdog_active; -static btime last_time; +static btime last_io_time; static btime loop_time; static void io_update_time(void) { - struct timespec ts; - int rv; - - /* - * This is third time-tracking procedure (after update_times() above and - * times_update() in BFD), dedicated to internal event log and latency - * tracking. Hopefully, we consolidate these sometimes. - */ - - rv = clock_gettime(CLOCK_MONOTONIC, &ts); - if (rv < 0) - die("clock_gettime: %m"); - - last_time = ts.tv_sec S + ts.tv_nsec NS; + last_io_time = current_time(); if (event_open) { - event_open->duration = last_time - event_open->timestamp; + event_open->duration = last_io_time - event_open->timestamp; if (event_open->duration > config->latency_limit) log(L_WARN "Event 0x%p 0x%p took %d ms", @@ -2068,7 +2084,7 @@ io_log_event(void *hook, void *data) en->hook = hook; en->data = data; - en->timestamp = last_time; + en->timestamp = last_io_time; en->duration = 0; event_log_num++; @@ -2096,14 +2112,14 @@ io_log_dump(void) struct event_log_entry *en = event_log + (event_log_pos + i) % EVENT_LOG_LENGTH; if (en->hook) log(L_DEBUG " Event 0x%p 0x%p at %8d for %d ms", en->hook, en->data, - (int) ((last_time - en->timestamp) TO_MS), (int) (en->duration TO_MS)); + (int) ((last_io_time - en->timestamp) TO_MS), (int) (en->duration TO_MS)); } } void watchdog_sigalrm(int sig UNUSED) { - /* Update last_time and duration, but skip latency check */ + /* Update last_io_time and duration, but skip latency check */ config->latency_limit = 0xffffffff; io_update_time(); @@ -2116,7 +2132,7 @@ watchdog_start1(void) { io_update_time(); - loop_time = last_time; + loop_time = last_io_time; } static inline void @@ -2124,7 +2140,7 @@ watchdog_start(void) { io_update_time(); - loop_time = last_time; + loop_time = last_io_time; event_log_num = 0; if (config->watchdog_timeout) @@ -2145,7 +2161,7 @@ watchdog_stop(void) watchdog_active = 0; } - btime duration = last_time - loop_time; + btime duration = last_io_time - loop_time; if (duration > config->watchdog_warning) log(L_WARN "I/O loop cycle took %d ms for %d events", (int) (duration TO_MS), event_log_num); @@ -2160,8 +2176,9 @@ void io_init(void) { init_list(&sock_list); - init_list(&global_event_list); - init_list(&global_work_list); + ev_init_list(&global_event_list, &main_birdloop, "Global event list"); + ev_init_list(&global_work_list, &main_birdloop, "Global work list"); + ev_init_list(&main_birdloop.event_list, &main_birdloop, "Global fast event list"); krt_io_init(); // XXX init_times(); // XXX update_times(); @@ -2175,6 +2192,9 @@ static int short_loops = 0; #define SHORT_LOOP_MAX 10 #define WORK_EVENTS_MAX 10 +void pipe_drain(int fd); +void check_stored_pages(void); + void io_loop(void) { @@ -2189,27 +2209,33 @@ io_loop(void) watchdog_start1(); for(;;) { - times_update(&main_timeloop); + times_update(); events = ev_run_list(&global_event_list); events = ev_run_list_limited(&global_work_list, WORK_EVENTS_MAX) || events; - timers_fire(&main_timeloop); + events = ev_run_list(&main_birdloop.event_list) || events; + timers_fire(&main_birdloop.time, 1); io_close_event(); // FIXME poll_tout = (events ? 0 : 3000); /* Time in milliseconds */ - if (t = timers_first(&main_timeloop)) + if (t = timers_first(&main_birdloop.time)) { - times_update(&main_timeloop); + times_update(); timeout = (tm_remains(t) TO_MS) + 1; poll_tout = MIN(poll_tout, timeout); } - nfds = 0; + /* A hack to reload main io_loop() when something has changed asynchronously. */ + pfd[0].fd = main_birdloop.wakeup_fds[0]; + pfd[0].events = POLLIN; + + nfds = 1; + WALK_LIST(n, sock_list) { pfd[nfds] = (struct pollfd) { .fd = -1 }; /* everything other set to 0 by this */ s = SKIP_BACK(sock, n, n); - if (s->rx_hook) + if (s->rx_hook && !ev_corked(s->cork)) { pfd[nfds].fd = s->fd; pfd[nfds].events |= POLLIN; @@ -2263,7 +2289,9 @@ io_loop(void) /* And finally enter poll() to find active sockets */ watchdog_stop(); + birdloop_leave(&main_birdloop); pout = poll(pfd, nfds, poll_tout); + birdloop_enter(&main_birdloop); watchdog_start(); if (pout < 0) @@ -2274,7 +2302,15 @@ io_loop(void) } if (pout) { - times_update(&main_timeloop); + if (pfd[0].revents & POLLIN) + { + /* IO loop reload requested */ + pipe_drain(main_birdloop.wakeup_fds[0]); + atomic_exchange_explicit(&main_birdloop.ping_sent, 0, memory_order_acq_rel); + continue; + } + + times_update(); /* guaranteed to be non-empty */ current_sock = SKIP_BACK(sock, n, HEAD(sock_list)); diff --git a/sysdep/unix/krt.c b/sysdep/unix/krt.c index 7c2614b1..c0f2e930 100644 --- a/sysdep/unix/krt.c +++ b/sysdep/unix/krt.c @@ -74,7 +74,7 @@ static list krt_proto_list; void krt_io_init(void) { - krt_pool = rp_new(&root_pool, "Kernel Syncer"); + krt_pool = rp_new(&root_pool, &main_birdloop, "Kernel Syncer"); krt_filter_lp = lp_new_default(krt_pool); init_list(&krt_proto_list); krt_sys_io_init(); @@ -251,14 +251,14 @@ static inline void krt_trace_in(struct krt_proto *p, rte *e, char *msg) { if (p->p.debug & D_PACKETS) - log(L_TRACE "%s: %N: %s", p->p.name, e->net->n.addr, msg); + log(L_TRACE "%s: %N: %s", p->p.name, e->net, msg); } static inline void krt_trace_in_rl(struct tbf *f, struct krt_proto *p, rte *e, char *msg) { if (p->p.debug & D_PACKETS) - log_rl(f, L_TRACE "%s: %N: %s", p->p.name, e->net->n.addr, msg); + log_rl(f, L_TRACE "%s: %N: %s", p->p.name, e->net, msg); } /* @@ -277,261 +277,33 @@ static struct tbf rl_alien = TBF_DEFAULT_LOG_LIMITS; * the same key. */ -static inline int -krt_same_key(rte *a, rte *b) +static inline u32 +krt_metric(rte *a) { - return a->u.krt.metric == b->u.krt.metric; + eattr *ea = ea_find(a->attrs->eattrs, EA_KRT_METRIC); + return ea ? ea->u.data : 0; } static inline int -krt_uptodate(rte *a, rte *b) -{ - if (a->attrs != b->attrs) - return 0; - - if (a->u.krt.proto != b->u.krt.proto) - return 0; - - return 1; -} - -static void -krt_learn_announce_update(struct krt_proto *p, rte *e) -{ - net *n = e->net; - rta *aa = rta_clone(e->attrs); - rte *ee = rte_get_temp(aa); - ee->pflags = EA_ID_FLAG(EA_KRT_SOURCE) | EA_ID_FLAG(EA_KRT_METRIC); - ee->u.krt = e->u.krt; - rte_update(&p->p, n->n.addr, ee); -} - -static void -krt_learn_announce_delete(struct krt_proto *p, net *n) +krt_rte_better(rte *a, rte *b) { - rte_update(&p->p, n->n.addr, NULL); + return (krt_metric(a) > krt_metric(b)); } /* Called when alien route is discovered during scan */ static void -krt_learn_scan(struct krt_proto *p, rte *e) +krt_learn_rte(struct krt_proto *p, rte *e) { - net *n0 = e->net; - net *n = net_get(p->krt_table, n0->n.addr); - rte *m, **mm; - - e->attrs = rta_lookup(e->attrs); - - for(mm=&n->routes; m = *mm; mm=&m->next) - if (krt_same_key(m, e)) - break; - if (m) - { - if (krt_uptodate(m, e)) - { - krt_trace_in_rl(&rl_alien, p, e, "[alien] seen"); - rte_free(e); - m->u.krt.seen = 1; - } - else - { - krt_trace_in(p, e, "[alien] updated"); - *mm = m->next; - rte_free(m); - m = NULL; - } - } - else - krt_trace_in(p, e, "[alien] created"); - if (!m) - { - e->next = n->routes; - n->routes = e; - e->u.krt.seen = 1; - } -} - -static void -krt_learn_prune(struct krt_proto *p) -{ - struct fib *fib = &p->krt_table->fib; - struct fib_iterator fit; - - KRT_TRACE(p, D_EVENTS, "Pruning inherited routes"); - - FIB_ITERATE_INIT(&fit, fib); -again: - FIB_ITERATE_START(fib, &fit, net, n) - { - rte *e, **ee, *best, **pbest, *old_best; - - /* - * Note that old_best may be NULL even if there was an old best route in - * the previous step, because it might be replaced in krt_learn_scan(). - * But in that case there is a new valid best route. - */ - - old_best = NULL; - best = NULL; - pbest = NULL; - ee = &n->routes; - while (e = *ee) - { - if (e->u.krt.best) - old_best = e; - - if (!e->u.krt.seen) - { - *ee = e->next; - rte_free(e); - continue; - } - - if (!best || best->u.krt.metric > e->u.krt.metric) - { - best = e; - pbest = ee; - } - - e->u.krt.seen = 0; - e->u.krt.best = 0; - ee = &e->next; - } - if (!n->routes) - { - DBG("%I/%d: deleting\n", n->n.prefix, n->n.pxlen); - if (old_best) - krt_learn_announce_delete(p, n); - - FIB_ITERATE_PUT(&fit); - fib_delete(fib, n); - goto again; - } - - best->u.krt.best = 1; - *pbest = best->next; - best->next = n->routes; - n->routes = best; - - if ((best != old_best) || p->reload) - { - DBG("%I/%d: announcing (metric=%d)\n", n->n.prefix, n->n.pxlen, best->u.krt.metric); - krt_learn_announce_update(p, best); - } - else - DBG("%I/%d: uptodate (metric=%d)\n", n->n.prefix, n->n.pxlen, best->u.krt.metric); - } - FIB_ITERATE_END; - - p->reload = 0; -} - -static void -krt_learn_async(struct krt_proto *p, rte *e, int new) -{ - net *n0 = e->net; - net *n = net_get(p->krt_table, n0->n.addr); - rte *g, **gg, *best, **bestp, *old_best; - - e->attrs = rta_lookup(e->attrs); - - old_best = n->routes; - for(gg=&n->routes; g = *gg; gg = &g->next) - if (krt_same_key(g, e)) - break; - if (new) - { - if (g) - { - if (krt_uptodate(g, e)) - { - krt_trace_in(p, e, "[alien async] same"); - rte_free(e); - return; - } - krt_trace_in(p, e, "[alien async] updated"); - *gg = g->next; - rte_free(g); - } - else - krt_trace_in(p, e, "[alien async] created"); - - e->next = n->routes; - n->routes = e; - } - else if (!g) - { - krt_trace_in(p, e, "[alien async] delete failed"); - rte_free(e); - return; - } - else - { - krt_trace_in(p, e, "[alien async] removed"); - *gg = g->next; - rte_free(e); - rte_free(g); - } - best = n->routes; - bestp = &n->routes; - for(gg=&n->routes; g=*gg; gg=&g->next) - { - if (best->u.krt.metric > g->u.krt.metric) - { - best = g; - bestp = gg; - } - - g->u.krt.best = 0; - } - - if (best) - { - best->u.krt.best = 1; - *bestp = best->next; - best->next = n->routes; - n->routes = best; - } - - if (best != old_best) - { - DBG("krt_learn_async: distributing change\n"); - if (best) - krt_learn_announce_update(p, best); - else - krt_learn_announce_delete(p, n); - } + struct rte_src *src = e->src = rt_get_source(&p->p, krt_metric(e)); + rte_update(p->p.main_channel, e->net, e, e->src); + rt_unlock_source(src); } static void krt_learn_init(struct krt_proto *p) { if (KRT_CF->learn) - { - struct rtable_config *cf = mb_allocz(p->p.pool, sizeof(struct rtable_config)); - cf->name = "Inherited"; - cf->addr_type = p->p.net_type; - cf->internal = 1; - - p->krt_table = rt_setup(p->p.pool, cf); - } -} - -static void -krt_dump(struct proto *P) -{ - struct krt_proto *p = (struct krt_proto *) P; - - if (!KRT_CF->learn) - return; - debug("KRT: Table of inheritable routes\n"); - rt_dump(p->krt_table); -} - -static void -krt_dump_attrs(rte *e) -{ - debug(" [m=%d,p=%d]", e->u.krt.metric, e->u.krt.proto); + channel_setup_in_table(p->p.main_channel, 1); } #endif @@ -543,47 +315,58 @@ krt_dump_attrs(rte *e) static inline int krt_is_installed(struct krt_proto *p, net *n) { - return n->routes && bmap_test(&p->p.main_channel->export_map, n->routes->id); + return n->routes && bmap_test(&p->p.main_channel->export_map, n->routes->rte.id); } -static void -krt_flush_routes(struct krt_proto *p) +static uint +rte_feed_count(net *n) { - struct rtable *t = p->p.main_channel->table; + uint count = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTES_OR_NULL(e))) + count++; + return count; +} - KRT_TRACE(p, D_EVENTS, "Flushing kernel routes"); - FIB_WALK(&t->fib, net, n) +static void +rte_feed_obtain(net *n, rte **feed, uint count) +{ + uint i = 0; + for (struct rte_storage *e = n->routes; e; e = e->next) + if (rte_is_valid(RTES_OR_NULL(e))) { - if (krt_is_installed(p, n)) - { - /* FIXME: this does not work if gw is changed in export filter */ - krt_replace_rte(p, n, NULL, n->routes); - } + ASSERT_DIE(i < count); + feed[i++] = &e->rte; } - FIB_WALK_END; + ASSERT_DIE(i == count); } static struct rte * -krt_export_net(struct krt_proto *p, net *net, rte **rt_free) +krt_export_net(struct krt_proto *p, net *net) { struct channel *c = p->p.main_channel; const struct filter *filter = c->out_filter; - rte *rt; if (c->ra_mode == RA_MERGED) - return rt_export_merged(c, net, rt_free, krt_filter_lp, 1); + { + uint count = rte_feed_count(net); + if (!count) + return NULL; + + rte **feed = alloca(count * sizeof(rte *)); + rte_feed_obtain(net, feed, count); + return rt_export_merged(c, feed, count, krt_filter_lp, 1); + } - rt = net->routes; - *rt_free = NULL; + static _Thread_local rte rt; + rt = net->routes->rte; - if (!rte_is_valid(rt)) + if (!rte_is_valid(&rt)) return NULL; if (filter == FILTER_REJECT) return NULL; - rte_make_tmp_attrs(&rt, krt_filter_lp, NULL); - /* We could run krt_preexport() here, but it is already handled by krt_is_installed() */ if (filter == FILTER_ACCEPT) @@ -594,13 +377,9 @@ krt_export_net(struct krt_proto *p, net *net, rte **rt_free) accept: - if (rt != net->routes) - *rt_free = rt; - return rt; + return &rt; reject: - if (rt != net->routes) - rte_free(rt); return NULL; } @@ -624,13 +403,13 @@ krt_same_dest(rte *k, rte *e) */ void -krt_got_route(struct krt_proto *p, rte *e) +krt_got_route(struct krt_proto *p, rte *e, s8 src) { - rte *new = NULL, *rt_free = NULL; - net *n = e->net; + rte *new = NULL; + e->pflags = 0; #ifdef KRT_ALLOW_LEARN - switch (e->u.krt.src) + switch (src) { case KRT_SRC_KERNEL: goto ignore; @@ -640,26 +419,28 @@ krt_got_route(struct krt_proto *p, rte *e) case KRT_SRC_ALIEN: if (KRT_CF->learn) - krt_learn_scan(p, e); + krt_learn_rte(p, e); else - { - krt_trace_in_rl(&rl_alien, p, e, "[alien] ignored"); - rte_free(e); - } + krt_trace_in_rl(&rl_alien, p, e, "[alien] ignored"); return; } #endif /* The rest is for KRT_SRC_BIRD (or KRT_SRC_UNKNOWN) */ + RT_LOCK(p->p.main_channel->table); + /* Deleting all routes if flush is requested */ + if (p->flush_routes) + goto delete; /* We wait for the initial feed to have correct installed state */ if (!p->ready) goto ignore; - if (!krt_is_installed(p, n)) + net *net = net_find(RT_PRIV(p->p.main_channel->table), e->net); + if (!net || !krt_is_installed(p, net)) goto delete; - new = krt_export_net(p, n, &rt_free); + new = krt_export_net(p, net); /* Rejected by filters */ if (!new) @@ -692,20 +473,16 @@ ignore: update: krt_trace_in(p, new, "updating"); - krt_replace_rte(p, n, new, e); + krt_replace_rte(p, e->net, new, e); goto done; delete: krt_trace_in(p, e, "deleting"); - krt_replace_rte(p, n, NULL, e); + krt_replace_rte(p, e->net, NULL, e); goto done; done: - rte_free(e); - - if (rt_free) - rte_free(rt_free); - + RT_UNLOCK(p->p.main_channel->table); lp_flush(krt_filter_lp); } @@ -713,50 +490,65 @@ static void krt_init_scan(struct krt_proto *p) { bmap_reset(&p->seen_map, 1024); + +#ifdef KRT_ALLOW_LEARN + if (KRT_CF->learn) + channel_refresh_begin(p->p.main_channel); +#endif } static void krt_prune(struct krt_proto *p) { - struct rtable *t = p->p.main_channel->table; + RT_LOCK(p->p.main_channel->table); + rtable_private *t = RT_PRIV(p->p.main_channel->table); KRT_TRACE(p, D_EVENTS, "Pruning table %s", t->name); FIB_WALK(&t->fib, net, n) { - if (p->ready && krt_is_installed(p, n) && !bmap_test(&p->seen_map, n->routes->id)) + if (p->ready && krt_is_installed(p, n) && !bmap_test(&p->seen_map, n->routes->rte.id)) { - rte *rt_free = NULL; - rte *new = krt_export_net(p, n, &rt_free); + rte *new = krt_export_net(p, n); if (new) { krt_trace_in(p, new, "installing"); - krt_replace_rte(p, n, new, NULL); + krt_replace_rte(p, n->n.addr, new, NULL); } - if (rt_free) - rte_free(rt_free); - lp_flush(krt_filter_lp); } } FIB_WALK_END; + RT_UNLOCK(p->p.main_channel->table); + #ifdef KRT_ALLOW_LEARN if (KRT_CF->learn) - krt_learn_prune(p); + channel_refresh_end(p->p.main_channel); #endif if (p->ready) p->initialized = 1; } +static void +krt_flush_routes(struct krt_proto *p) +{ + KRT_TRACE(p, D_EVENTS, "Flushing kernel routes"); + p->flush_routes = 1; + krt_init_scan(p); + krt_do_scan(p); + /* No prune! */ + p->flush_routes = 0; +} + void -krt_got_route_async(struct krt_proto *p, rte *e, int new) +krt_got_route_async(struct krt_proto *p, rte *e, int new, s8 src) { - net *net = e->net; + e->pflags = 0; - switch (e->u.krt.src) + switch (src) { case KRT_SRC_BIRD: /* Should be filtered by the back end */ @@ -766,7 +558,7 @@ krt_got_route_async(struct krt_proto *p, rte *e, int new) if (new) { krt_trace_in(p, e, "[redirect] deleting"); - krt_replace_rte(p, net, NULL, e); + krt_replace_rte(p, e->net, NULL, e); } /* If !new, it is probably echo of our deletion */ break; @@ -775,12 +567,11 @@ krt_got_route_async(struct krt_proto *p, rte *e, int new) case KRT_SRC_ALIEN: if (KRT_CF->learn) { - krt_learn_async(p, e, new); + krt_learn_rte(p, e); return; } #endif } - rte_free(e); } /* @@ -886,29 +677,10 @@ krt_scan_timer_kick(struct krt_proto *p) * Updates */ -static void -krt_make_tmp_attrs(struct rte *rt, struct linpool *pool) -{ - rte_init_tmp_attrs(rt, pool, 2); - rte_make_tmp_attr(rt, EA_KRT_SOURCE, EAF_TYPE_INT, rt->u.krt.proto); - rte_make_tmp_attr(rt, EA_KRT_METRIC, EAF_TYPE_INT, rt->u.krt.metric); -} - -static void -krt_store_tmp_attrs(struct rte *rt, struct linpool *pool) -{ - rte_init_tmp_attrs(rt, pool, 2); - rt->u.krt.proto = rte_store_tmp_attr(rt, EA_KRT_SOURCE); - rt->u.krt.metric = rte_store_tmp_attr(rt, EA_KRT_METRIC); -} - static int -krt_preexport(struct proto *P, rte **new, struct linpool *pool UNUSED) +krt_preexport(struct channel *c, rte *e) { - // struct krt_proto *p = (struct krt_proto *) P; - rte *e = *new; - - if (e->attrs->src->proto == P) + if (e->src->owner == &c->proto->sources) return -1; if (!krt_capable(e)) @@ -918,8 +690,8 @@ krt_preexport(struct proto *P, rte **new, struct linpool *pool UNUSED) } static void -krt_rt_notify(struct proto *P, struct channel *ch UNUSED, net *net, - rte *new, rte *old) +krt_rt_notify(struct proto *P, struct channel *ch UNUSED, const net_addr *net, + rte *new, const rte *old) { struct krt_proto *p = (struct krt_proto *) P; @@ -983,14 +755,6 @@ krt_feed_end(struct channel *C) } -static int -krt_rte_same(rte *a, rte *b) -{ - /* src is always KRT_SRC_ALIEN and type is irrelevant */ - return (a->u.krt.proto == b->u.krt.proto) && (a->u.krt.metric == b->u.krt.metric); -} - - /* * Protocol glue */ @@ -1049,9 +813,7 @@ krt_init(struct proto_config *CF) p->p.if_notify = krt_if_notify; p->p.reload_routes = krt_reload_routes; p->p.feed_end = krt_feed_end; - p->p.make_tmp_attrs = krt_make_tmp_attrs; - p->p.store_tmp_attrs = krt_store_tmp_attrs; - p->p.rte_same = krt_rte_same; + p->p.rte_better = krt_rte_better; krt_sys_init(p); return &p->p; @@ -1207,8 +969,4 @@ struct protocol proto_unix_kernel = { .reconfigure = krt_reconfigure, .copy_config = krt_copy_config, .get_attr = krt_get_attr, -#ifdef KRT_ALLOW_LEARN - .dump = krt_dump, - .dump_attrs = krt_dump_attrs, -#endif }; diff --git a/sysdep/unix/krt.h b/sysdep/unix/krt.h index 62228f08..968c5b16 100644 --- a/sysdep/unix/krt.h +++ b/sysdep/unix/krt.h @@ -24,6 +24,9 @@ struct kif_proto; #define EA_KRT_SOURCE EA_CODE(PROTOCOL_KERNEL, 0) #define EA_KRT_METRIC EA_CODE(PROTOCOL_KERNEL, 1) +#define KRT_REF_SEEN 0x1 /* Seen in table */ +#define KRT_REF_BEST 0x2 /* Best in table */ + /* Whenever we recognize our own routes, we allow learing of foreign routes */ #ifdef CONFIG_SELF_CONSCIOUS @@ -48,10 +51,6 @@ struct krt_proto { struct proto p; struct krt_state sys; /* Sysdep state */ -#ifdef KRT_ALLOW_LEARN - struct rtable *krt_table; /* Internal table of inherited routes */ -#endif - #ifndef CONFIG_ALL_TABLES_AT_ONCE timer *scan_timer; #endif @@ -63,6 +62,7 @@ struct krt_proto { byte ready; /* Initial feed has been finished */ byte initialized; /* First scan has been finished */ byte reload; /* Next scan is doing reload */ + byte flush_routes; /* Scanning to flush */ }; extern pool *krt_pool; @@ -76,8 +76,8 @@ extern pool *krt_pool; struct proto_config * kif_init_config(int class); void kif_request_scan(void); -void krt_got_route(struct krt_proto *p, struct rte *e); -void krt_got_route_async(struct krt_proto *p, struct rte *e, int new); +void krt_got_route(struct krt_proto *p, struct rte *e, s8 src); +void krt_got_route_async(struct krt_proto *p, struct rte *e, int new, s8 src); static inline int krt_get_sync_error(struct krt_proto *p, struct rte *e) @@ -140,7 +140,7 @@ void krt_sys_copy_config(struct krt_config *, struct krt_config *); int krt_capable(rte *e); void krt_do_scan(struct krt_proto *); -void krt_replace_rte(struct krt_proto *p, net *n, rte *new, rte *old); +void krt_replace_rte(struct krt_proto *p, const net_addr *n, rte *new, const rte *old); int krt_sys_get_attr(const eattr *a, byte *buf, int buflen); diff --git a/sysdep/unix/log.c b/sysdep/unix/log.c index 14d18c01..68a04e78 100644 --- a/sysdep/unix/log.c +++ b/sysdep/unix/log.c @@ -15,6 +15,7 @@ * user's manual. */ +#include <stdatomic.h> #include <stdio.h> #include <stdlib.h> #include <stdarg.h> @@ -35,8 +36,10 @@ static FILE *dbgf; static list *current_log_list; static char *current_syslog_name; /* NULL -> syslog closed */ +static _Atomic uint max_coro_id = ATOMIC_VAR_INIT(1); +static _Thread_local uint this_coro_id; -#ifdef USE_PTHREADS +#define THIS_CORO_ID (this_coro_id ?: (this_coro_id = atomic_fetch_add_explicit(&max_coro_id, 1, memory_order_acq_rel))) #include <pthread.h> @@ -48,15 +51,6 @@ static pthread_t main_thread; void main_thread_init(void) { main_thread = pthread_self(); } static int main_thread_self(void) { return pthread_equal(pthread_self(), main_thread); } -#else - -static inline void log_lock(void) { } -static inline void log_unlock(void) { } -void main_thread_init(void) { } -static int main_thread_self(void) { return 1; } - -#endif - #ifdef HAVE_SYSLOG_H #include <sys/syslog.h> @@ -189,7 +183,7 @@ log_commit(int class, buffer *buf) l->pos += msg_len; } - fprintf(l->fh, "%s <%s> ", tbuf, class_names[class]); + fprintf(l->fh, "%s [%04x] <%s> ", tbuf, THIS_CORO_ID, class_names[class]); } fputs(buf->start, l->fh); fputc('\n', l->fh); @@ -299,6 +293,8 @@ die(const char *msg, ...) exit(1); } +static struct timespec dbg_time_start; + /** * debug - write to debug output * @msg: a printf-like message @@ -309,22 +305,36 @@ die(const char *msg, ...) void debug(const char *msg, ...) { -#define MAX_DEBUG_BUFSIZE 65536 +#define MAX_DEBUG_BUFSIZE 16384 va_list args; - static uint bufsize = 4096; - static char *buf = NULL; - - if (!buf) - buf = mb_alloc(&root_pool, bufsize); + char buf[MAX_DEBUG_BUFSIZE], *pos = buf; + int max = MAX_DEBUG_BUFSIZE; va_start(args, msg); if (dbgf) { - while (bvsnprintf(buf, bufsize, msg, args) < 0) - if (bufsize >= MAX_DEBUG_BUFSIZE) - bug("Extremely long debug output, split it."); - else - buf = mb_realloc(buf, (bufsize *= 2)); + struct timespec dbg_time; + clock_gettime(CLOCK_MONOTONIC, &dbg_time); + uint nsec; + uint sec; + + if (dbg_time.tv_nsec > dbg_time_start.tv_nsec) + { + nsec = dbg_time.tv_nsec - dbg_time_start.tv_nsec; + sec = dbg_time.tv_sec - dbg_time_start.tv_sec; + } + else + { + nsec = 1000000000 + dbg_time.tv_nsec - dbg_time_start.tv_nsec; + sec = dbg_time.tv_sec - dbg_time_start.tv_sec - 1; + } + + int n = bsnprintf(pos, max, "%u.%09u: [%04x] ", sec, nsec, THIS_CORO_ID); + pos += n; + max -= n; + + if (bvsnprintf(pos, max, msg, args) < 0) + bug("Extremely long debug output, split it."); fputs(buf, dbgf); } @@ -378,6 +388,21 @@ default_log_list(int initial, const char **syslog_name) } void +log_cleanup(int syslog) +{ + struct log_config *l; + + if (current_log_list) + WALK_LIST(l, *current_log_list) + if (l->rf) + log_close(l); + + if (syslog && current_syslog_name) + closelog(); +} + + +void log_switch(int initial, list *logs, const char *new_syslog_name) { struct log_config *l; @@ -389,10 +414,7 @@ log_switch(int initial, list *logs, const char *new_syslog_name) logs = default_log_list(initial, &new_syslog_name); /* Close the logs to avoid pinning them on disk when deleted */ - if (current_log_list) - WALK_LIST(l, *current_log_list) - if (l->rf) - log_close(l); + log_cleanup(0); /* Reopen the logs, needed for 'configure undo' */ if (logs) @@ -429,6 +451,8 @@ done: void log_init_debug(char *f) { + clock_gettime(CLOCK_MONOTONIC, &dbg_time_start); + if (dbgf && dbgf != stderr) fclose(dbgf); if (!f) diff --git a/sysdep/unix/main.c b/sysdep/unix/main.c index 392aff9d..c326ba2b 100644 --- a/sysdep/unix/main.c +++ b/sysdep/unix/main.c @@ -28,6 +28,7 @@ #include "lib/resource.h" #include "lib/socket.h" #include "lib/event.h" +#include "lib/locking.h" #include "lib/timer.h" #include "lib/string.h" #include "nest/route.h" @@ -51,7 +52,7 @@ async_dump(void) { debug("INTERNAL STATE DUMP\n\n"); - rdump(&root_pool); + rp_dump(&root_pool); sk_dump_all(); // XXXX tm_dump_all(); if_dump_all(); @@ -201,7 +202,9 @@ sysdep_preconfig(struct config *c) int sysdep_commit(struct config *new, struct config *old UNUSED) { - log_switch(0, &new->logfiles, new->syslog_name); + if (!new->shutdown) + log_switch(0, &new->logfiles, new->syslog_name); + return 0; } @@ -479,6 +482,14 @@ cli_err(sock *s, int err) cli_free(s->data); } +static void +cli_connect_err(sock *s UNUSED, int err) +{ + ASSERT_DIE(err); + if (config->cli_debug) + log(L_INFO "Failed to accept CLI connection: %s", strerror(err)); +} + static int cli_connect(sock *s, uint size UNUSED) { @@ -507,6 +518,7 @@ cli_init_unix(uid_t use_uid, gid_t use_gid) s = cli_sk = sk_new(cli_pool); s->type = SK_UNIX_PASSIVE; s->rx_hook = cli_connect; + s->err_hook = cli_connect_err; s->rbsize = 1024; s->fast_rx = 1; @@ -603,6 +615,7 @@ sysdep_shutdown_done(void) unlink_pid_file(); unlink(path_control_socket); log_msg(L_FATAL "Shutdown completed"); + log_cleanup(1); exit(0); } @@ -673,7 +686,7 @@ signal_init(void) * Parsing of command-line arguments */ -static char *opt_list = "bc:dD:ps:P:u:g:flRh"; +static char *opt_list = "c:dD:ps:P:u:g:flRh"; int parse_and_exit; char *bird_name; static char *use_user; @@ -852,6 +865,8 @@ parse_args(int argc, char **argv) } } +void resource_sys_init(void); + /* * Hic Est main() */ @@ -864,13 +879,17 @@ main(int argc, char **argv) dmalloc_debug(0x2f03d00); #endif + times_update(); + resource_sys_init(); parse_args(argc, argv); log_switch(1, NULL, NULL); + the_bird_lock(); + random_init(); net_init(); resource_init(); - timer_init(); + birdloop_init(); olock_init(); io_init(); rt_init(); @@ -920,6 +939,7 @@ main(int argc, char **argv) dup2(0, 2); } + main_thread_init(); write_pid_file(); diff --git a/sysdep/unix/unix.h b/sysdep/unix/unix.h index ad85d1ea..51ec1b1e 100644 --- a/sysdep/unix/unix.h +++ b/sysdep/unix/unix.h @@ -122,6 +122,7 @@ void krt_io_init(void); void main_thread_init(void); void log_init_debug(char *); /* Initialize debug dump to given file (NULL=stderr, ""=off) */ void log_switch(int initial, list *l, const char *); +void log_cleanup(int syslog); struct log_config { node n; diff --git a/test/birdtest.c b/test/birdtest.c index a1da078f..c6a09684 100644 --- a/test/birdtest.c +++ b/test/birdtest.c @@ -58,11 +58,14 @@ u64 bt_random_state[] = { 0x53d9772877c1b647, 0xab8ce3eb466de6c5, 0xad02844c8a8e865f, 0xe8cc78080295065d }; +void resource_sys_init(void); + void bt_init(int argc, char *argv[]) { int c; + resource_sys_init(); initstate(BT_RANDOM_SEED, (char *) bt_random_state, sizeof(bt_random_state)); bt_verbose = 0; @@ -240,7 +243,7 @@ bt_log_result(int result, u64 time, const char *fmt, va_list argptr) printf("%s\n", result_str); if (do_die && !result) - abort(); + exit(1); } static u64 diff --git a/test/bt-utils.c b/test/bt-utils.c index cbca3a6b..98aaed3d 100644 --- a/test/bt-utils.c +++ b/test/bt-utils.c @@ -53,16 +53,20 @@ cf_file_read(byte *dest, uint max_len, int fd) return l; } +void resource_sys_init(void); + void bt_bird_init(void) { + resource_sys_init(); if(bt_verbose) log_init_debug(""); log_switch(bt_verbose != 0, NULL, NULL); + the_bird_lock(); resource_init(); + birdloop_init(); olock_init(); - timer_init(); io_init(); rt_init(); if_init(); @@ -79,6 +83,7 @@ void bt_bird_cleanup(void) class_to_protocol[i] = NULL; config = new_config = NULL; + the_bird_unlock(); } static char * |