summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/socket
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/socket')
-rw-r--r--pkg/sentry/socket/BUILD1
-rw-r--r--pkg/sentry/socket/hostinet/BUILD4
-rw-r--r--pkg/sentry/socket/hostinet/socket.go17
-rw-r--r--pkg/sentry/socket/hostinet/socket_vfs2.go22
-rw-r--r--pkg/sentry/socket/netfilter/netfilter.go156
-rw-r--r--pkg/sentry/socket/netfilter/targets.go2
-rw-r--r--pkg/sentry/socket/netlink/BUILD4
-rw-r--r--pkg/sentry/socket/netlink/socket.go14
-rw-r--r--pkg/sentry/socket/netlink/socket_vfs2.go14
-rw-r--r--pkg/sentry/socket/netstack/BUILD4
-rw-r--r--pkg/sentry/socket/netstack/netstack.go411
-rw-r--r--pkg/sentry/socket/netstack/netstack_vfs2.go30
-rw-r--r--pkg/sentry/socket/netstack/stack.go30
-rw-r--r--pkg/sentry/socket/socket.go4
-rw-r--r--pkg/sentry/socket/unix/BUILD3
-rw-r--r--pkg/sentry/socket/unix/transport/connectioned.go3
-rw-r--r--pkg/sentry/socket/unix/unix.go3
-rw-r--r--pkg/sentry/socket/unix/unix_vfs2.go19
18 files changed, 445 insertions, 296 deletions
diff --git a/pkg/sentry/socket/BUILD b/pkg/sentry/socket/BUILD
index c40c6d673..c0fd3425b 100644
--- a/pkg/sentry/socket/BUILD
+++ b/pkg/sentry/socket/BUILD
@@ -20,5 +20,6 @@ go_library(
"//pkg/syserr",
"//pkg/tcpip",
"//pkg/usermem",
+ "//tools/go_marshal/marshal",
],
)
diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD
index 60c9896fc..e76e498de 100644
--- a/pkg/sentry/socket/hostinet/BUILD
+++ b/pkg/sentry/socket/hostinet/BUILD
@@ -26,6 +26,7 @@ go_library(
"//pkg/sentry/device",
"//pkg/sentry/fs",
"//pkg/sentry/fs/fsutil",
+ "//pkg/sentry/fs/lock",
"//pkg/sentry/fsimpl/sockfs",
"//pkg/sentry/hostfd",
"//pkg/sentry/inet",
@@ -34,12 +35,13 @@ go_library(
"//pkg/sentry/socket",
"//pkg/sentry/socket/control",
"//pkg/sentry/vfs",
- "//pkg/sentry/vfs/lock",
"//pkg/syserr",
"//pkg/syserror",
"//pkg/tcpip/stack",
"//pkg/usermem",
"//pkg/waiter",
+ "//tools/go_marshal/marshal",
+ "//tools/go_marshal/primitive",
"@org_golang_x_sys//unix:go_default_library",
],
)
diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go
index c11e82c10..532a1ea5d 100644
--- a/pkg/sentry/socket/hostinet/socket.go
+++ b/pkg/sentry/socket/hostinet/socket.go
@@ -36,6 +36,8 @@ import (
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
"gvisor.dev/gvisor/pkg/waiter"
+ "gvisor.dev/gvisor/tools/go_marshal/marshal"
+ "gvisor.dev/gvisor/tools/go_marshal/primitive"
)
const (
@@ -319,12 +321,12 @@ func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error {
}
// GetSockOpt implements socket.Socket.GetSockOpt.
-func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) {
+func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) {
if outLen < 0 {
return nil, syserr.ErrInvalidArgument
}
- // Whitelist options and constrain option length.
+ // Only allow known and safe options.
optlen := getSockOptLen(t, level, name)
switch level {
case linux.SOL_IP:
@@ -364,12 +366,13 @@ func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr
if err != nil {
return nil, syserr.FromError(err)
}
- return opt, nil
+ optP := primitive.ByteSlice(opt)
+ return &optP, nil
}
// SetSockOpt implements socket.Socket.SetSockOpt.
func (s *socketOpsCommon) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error {
- // Whitelist options and constrain option length.
+ // Only allow known and safe options.
optlen := setSockOptLen(t, level, name)
switch level {
case linux.SOL_IP:
@@ -415,7 +418,7 @@ func (s *socketOpsCommon) SetSockOpt(t *kernel.Task, level int, name int, opt []
// RecvMsg implements socket.Socket.RecvMsg.
func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) {
- // Whitelist flags.
+ // Only allow known and safe flags.
//
// FIXME(jamieliu): We can't support MSG_ERRQUEUE because it uses ancillary
// messages that gvisor/pkg/tcpip/transport/unix doesn't understand. Kill the
@@ -537,7 +540,7 @@ func (s *socketOpsCommon) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags
// SendMsg implements socket.Socket.SendMsg.
func (s *socketOpsCommon) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) {
- // Whitelist flags.
+ // Only allow known and safe flags.
if flags&^(syscall.MSG_DONTWAIT|syscall.MSG_EOR|syscall.MSG_FASTOPEN|syscall.MSG_MORE|syscall.MSG_NOSIGNAL) != 0 {
return 0, syserr.ErrInvalidArgument
}
@@ -708,6 +711,6 @@ func (p *socketProvider) Pair(t *kernel.Task, stype linux.SockType, protocol int
func init() {
for _, family := range []int{syscall.AF_INET, syscall.AF_INET6} {
socket.RegisterProvider(family, &socketProvider{family})
- socket.RegisterProviderVFS2(family, &socketProviderVFS2{})
+ socket.RegisterProviderVFS2(family, &socketProviderVFS2{family})
}
}
diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go
index 027add1fd..8a1d52ebf 100644
--- a/pkg/sentry/socket/hostinet/socket_vfs2.go
+++ b/pkg/sentry/socket/hostinet/socket_vfs2.go
@@ -21,12 +21,12 @@ import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/fdnotifier"
"gvisor.dev/gvisor/pkg/sentry/arch"
+ fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock"
"gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs"
"gvisor.dev/gvisor/pkg/sentry/hostfd"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/socket"
"gvisor.dev/gvisor/pkg/sentry/vfs"
- "gvisor.dev/gvisor/pkg/sentry/vfs/lock"
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
@@ -61,7 +61,7 @@ func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol in
fd: fd,
},
}
- s.LockFD.Init(&lock.FileLocks{})
+ s.LockFD.Init(&vfs.FileLocks{})
if err := fdnotifier.AddFD(int32(fd), &s.queue); err != nil {
return nil, syserr.FromError(err)
}
@@ -71,6 +71,7 @@ func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol in
DenyPWrite: true,
UseDentryMetadata: true,
}); err != nil {
+ fdnotifier.RemoveFD(int32(s.fd))
return nil, syserr.FromError(err)
}
return vfsfd, nil
@@ -96,7 +97,12 @@ func (s *socketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.Syscal
return ioctl(ctx, s.fd, uio, args)
}
-// PRead implements vfs.FileDescriptionImpl.
+// Allocate implements vfs.FileDescriptionImpl.Allocate.
+func (s *socketVFS2) Allocate(ctx context.Context, mode, offset, length uint64) error {
+ return syserror.ENODEV
+}
+
+// PRead implements vfs.FileDescriptionImpl.PRead.
func (s *socketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
return 0, syserror.ESPIPE
}
@@ -134,6 +140,16 @@ func (s *socketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs
return int64(n), err
}
+// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX.
+func (s *socketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error {
+ return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block)
+}
+
+// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX.
+func (s *socketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error {
+ return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence)
+}
+
type socketProviderVFS2 struct {
family int
}
diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go
index 66015e2bc..a9f0604ae 100644
--- a/pkg/sentry/socket/netfilter/netfilter.go
+++ b/pkg/sentry/socket/netfilter/netfilter.go
@@ -41,19 +41,6 @@ const errorTargetName = "ERROR"
// change the destination port/destination IP for packets.
const redirectTargetName = "REDIRECT"
-// Metadata is used to verify that we are correctly serializing and
-// deserializing iptables into structs consumable by the iptables tool. We save
-// a metadata struct when the tables are written, and when they are read out we
-// verify that certain fields are the same.
-//
-// metadata is used by this serialization/deserializing code, not netstack.
-type metadata struct {
- HookEntry [linux.NF_INET_NUMHOOKS]uint32
- Underflow [linux.NF_INET_NUMHOOKS]uint32
- NumEntries uint32
- Size uint32
-}
-
// enableLogging controls whether to log the (de)serialization of netfilter
// structs between userspace and netstack. These logs are useful when
// developing iptables, but can pollute sentry logs otherwise.
@@ -79,33 +66,17 @@ func nflog(format string, args ...interface{}) {
func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPTGetinfo, *syserr.Error) {
// Read in the struct and table name.
var info linux.IPTGetinfo
- if _, err := t.CopyIn(outPtr, &info); err != nil {
+ if _, err := info.CopyIn(t, outPtr); err != nil {
return linux.IPTGetinfo{}, syserr.FromError(err)
}
- // Find the appropriate table.
- table, err := findTable(stack, info.Name)
+ _, info, err := convertNetstackToBinary(stack, info.Name)
if err != nil {
- nflog("%v", err)
+ nflog("couldn't convert iptables: %v", err)
return linux.IPTGetinfo{}, syserr.ErrInvalidArgument
}
- // Get the hooks that apply to this table.
- info.ValidHooks = table.ValidHooks()
-
- // Grab the metadata struct, which is used to store information (e.g.
- // the number of entries) that applies to the user's encoding of
- // iptables, but not netstack's.
- metadata := table.Metadata().(metadata)
-
- // Set values from metadata.
- info.HookEntry = metadata.HookEntry
- info.Underflow = metadata.Underflow
- info.NumEntries = metadata.NumEntries
- info.Size = metadata.Size
-
nflog("returning info: %+v", info)
-
return info, nil
}
@@ -113,28 +84,18 @@ func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPT
func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen int) (linux.KernelIPTGetEntries, *syserr.Error) {
// Read in the struct and table name.
var userEntries linux.IPTGetEntries
- if _, err := t.CopyIn(outPtr, &userEntries); err != nil {
+ if _, err := userEntries.CopyIn(t, outPtr); err != nil {
nflog("couldn't copy in entries %q", userEntries.Name)
return linux.KernelIPTGetEntries{}, syserr.FromError(err)
}
- // Find the appropriate table.
- table, err := findTable(stack, userEntries.Name)
- if err != nil {
- nflog("%v", err)
- return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument
- }
-
// Convert netstack's iptables rules to something that the iptables
// tool can understand.
- entries, meta, err := convertNetstackToBinary(userEntries.Name.String(), table)
+ entries, _, err := convertNetstackToBinary(stack, userEntries.Name)
if err != nil {
nflog("couldn't read entries: %v", err)
return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument
}
- if meta != table.Metadata().(metadata) {
- panic(fmt.Sprintf("Table %q metadata changed between writing and reading. Was saved as %+v, but is now %+v", userEntries.Name.String(), table.Metadata().(metadata), meta))
- }
if binary.Size(entries) > uintptr(outLen) {
nflog("insufficient GetEntries output size: %d", uintptr(outLen))
return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument
@@ -143,44 +104,26 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen
return entries, nil
}
-func findTable(stk *stack.Stack, tablename linux.TableName) (stack.Table, error) {
- table, ok := stk.IPTables().GetTable(tablename.String())
- if !ok {
- return stack.Table{}, fmt.Errorf("couldn't find table %q", tablename)
- }
- return table, nil
-}
-
-// FillIPTablesMetadata populates stack's IPTables with metadata.
-func FillIPTablesMetadata(stk *stack.Stack) {
- stk.IPTables().ModifyTables(func(tables map[string]stack.Table) {
- // In order to fill in the metadata, we have to translate ipt from its
- // netstack format to Linux's giant-binary-blob format.
- for name, table := range tables {
- _, metadata, err := convertNetstackToBinary(name, table)
- if err != nil {
- panic(fmt.Errorf("Unable to set default IP tables: %v", err))
- }
- table.SetMetadata(metadata)
- tables[name] = table
- }
- })
-}
-
// convertNetstackToBinary converts the iptables as stored in netstack to the
// format expected by the iptables tool. Linux stores each table as a binary
// blob that can only be traversed by parsing a bit, reading some offsets,
// jumping to those offsets, parsing again, etc.
-func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelIPTGetEntries, metadata, error) {
- // Return values.
+func convertNetstackToBinary(stack *stack.Stack, tablename linux.TableName) (linux.KernelIPTGetEntries, linux.IPTGetinfo, error) {
+ table, ok := stack.IPTables().GetTable(tablename.String())
+ if !ok {
+ return linux.KernelIPTGetEntries{}, linux.IPTGetinfo{}, fmt.Errorf("couldn't find table %q", tablename)
+ }
+
var entries linux.KernelIPTGetEntries
- var meta metadata
+ var info linux.IPTGetinfo
+ info.ValidHooks = table.ValidHooks()
// The table name has to fit in the struct.
if linux.XT_TABLE_MAXNAMELEN < len(tablename) {
- return linux.KernelIPTGetEntries{}, metadata{}, fmt.Errorf("table name %q too long.", tablename)
+ return linux.KernelIPTGetEntries{}, linux.IPTGetinfo{}, fmt.Errorf("table name %q too long", tablename)
}
- copy(entries.Name[:], tablename)
+ copy(info.Name[:], tablename[:])
+ copy(entries.Name[:], tablename[:])
for ruleIdx, rule := range table.Rules {
nflog("convert to binary: current offset: %d", entries.Size)
@@ -189,20 +132,20 @@ func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelI
for hook, hookRuleIdx := range table.BuiltinChains {
if hookRuleIdx == ruleIdx {
nflog("convert to binary: found hook %d at offset %d", hook, entries.Size)
- meta.HookEntry[hook] = entries.Size
+ info.HookEntry[hook] = entries.Size
}
}
// Is this a chain underflow point?
for underflow, underflowRuleIdx := range table.Underflows {
if underflowRuleIdx == ruleIdx {
nflog("convert to binary: found underflow %d at offset %d", underflow, entries.Size)
- meta.Underflow[underflow] = entries.Size
+ info.Underflow[underflow] = entries.Size
}
}
// Each rule corresponds to an entry.
entry := linux.KernelIPTEntry{
- IPTEntry: linux.IPTEntry{
+ Entry: linux.IPTEntry{
IP: linux.IPTIP{
Protocol: uint16(rule.Filter.Protocol),
},
@@ -210,20 +153,20 @@ func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelI
TargetOffset: linux.SizeOfIPTEntry,
},
}
- copy(entry.IPTEntry.IP.Dst[:], rule.Filter.Dst)
- copy(entry.IPTEntry.IP.DstMask[:], rule.Filter.DstMask)
- copy(entry.IPTEntry.IP.Src[:], rule.Filter.Src)
- copy(entry.IPTEntry.IP.SrcMask[:], rule.Filter.SrcMask)
- copy(entry.IPTEntry.IP.OutputInterface[:], rule.Filter.OutputInterface)
- copy(entry.IPTEntry.IP.OutputInterfaceMask[:], rule.Filter.OutputInterfaceMask)
+ copy(entry.Entry.IP.Dst[:], rule.Filter.Dst)
+ copy(entry.Entry.IP.DstMask[:], rule.Filter.DstMask)
+ copy(entry.Entry.IP.Src[:], rule.Filter.Src)
+ copy(entry.Entry.IP.SrcMask[:], rule.Filter.SrcMask)
+ copy(entry.Entry.IP.OutputInterface[:], rule.Filter.OutputInterface)
+ copy(entry.Entry.IP.OutputInterfaceMask[:], rule.Filter.OutputInterfaceMask)
if rule.Filter.DstInvert {
- entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_DSTIP
+ entry.Entry.IP.InverseFlags |= linux.IPT_INV_DSTIP
}
if rule.Filter.SrcInvert {
- entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_SRCIP
+ entry.Entry.IP.InverseFlags |= linux.IPT_INV_SRCIP
}
if rule.Filter.OutputInterfaceInvert {
- entry.IPTEntry.IP.InverseFlags |= linux.IPT_INV_VIA_OUT
+ entry.Entry.IP.InverseFlags |= linux.IPT_INV_VIA_OUT
}
for _, matcher := range rule.Matchers {
@@ -235,8 +178,8 @@ func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelI
panic(fmt.Sprintf("matcher %T is not 64-bit aligned", matcher))
}
entry.Elems = append(entry.Elems, serialized...)
- entry.NextOffset += uint16(len(serialized))
- entry.TargetOffset += uint16(len(serialized))
+ entry.Entry.NextOffset += uint16(len(serialized))
+ entry.Entry.TargetOffset += uint16(len(serialized))
}
// Serialize and append the target.
@@ -245,18 +188,18 @@ func convertNetstackToBinary(tablename string, table stack.Table) (linux.KernelI
panic(fmt.Sprintf("target %T is not 64-bit aligned", rule.Target))
}
entry.Elems = append(entry.Elems, serialized...)
- entry.NextOffset += uint16(len(serialized))
+ entry.Entry.NextOffset += uint16(len(serialized))
nflog("convert to binary: adding entry: %+v", entry)
- entries.Size += uint32(entry.NextOffset)
+ entries.Size += uint32(entry.Entry.NextOffset)
entries.Entrytable = append(entries.Entrytable, entry)
- meta.NumEntries++
+ info.NumEntries++
}
- nflog("convert to binary: finished with an marshalled size of %d", meta.Size)
- meta.Size = entries.Size
- return entries, meta, nil
+ nflog("convert to binary: finished with an marshalled size of %d", info.Size)
+ info.Size = entries.Size
+ return entries, info, nil
}
func marshalTarget(target stack.Target) []byte {
@@ -399,10 +342,10 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
// TODO(gvisor.dev/issue/170): Support other tables.
var table stack.Table
switch replace.Name.String() {
- case stack.TablenameFilter:
+ case stack.FilterTable:
table = stack.EmptyFilterTable()
- case stack.TablenameNat:
- table = stack.EmptyNatTable()
+ case stack.NATTable:
+ table = stack.EmptyNATTable()
default:
nflog("we don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String())
return syserr.ErrInvalidArgument
@@ -488,6 +431,8 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
for hook, _ := range replace.HookEntry {
if table.ValidHooks()&(1<<hook) != 0 {
hk := hookFromLinux(hook)
+ table.BuiltinChains[hk] = stack.HookUnset
+ table.Underflows[hk] = stack.HookUnset
for offset, ruleIdx := range offsets {
if offset == replace.HookEntry[hook] {
table.BuiltinChains[hk] = ruleIdx
@@ -513,8 +458,7 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
// Add the user chains.
for ruleIdx, rule := range table.Rules {
- target, ok := rule.Target.(stack.UserChainTarget)
- if !ok {
+ if _, ok := rule.Target.(stack.UserChainTarget); !ok {
continue
}
@@ -530,7 +474,6 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
nflog("user chain's first node must have no matchers")
return syserr.ErrInvalidArgument
}
- table.UserChains[target.Name] = ruleIdx + 1
}
// Set each jump to point to the appropriate rule. Right now they hold byte
@@ -556,7 +499,10 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
// Since we only support modifying the INPUT, PREROUTING and OUTPUT chain right now,
// make sure all other chains point to ACCEPT rules.
for hook, ruleIdx := range table.BuiltinChains {
- if hook == stack.Forward || hook == stack.Postrouting {
+ if hook := stack.Hook(hook); hook == stack.Forward || hook == stack.Postrouting {
+ if ruleIdx == stack.HookUnset {
+ continue
+ }
if !isUnconditionalAccept(table.Rules[ruleIdx]) {
nflog("hook %d is unsupported.", hook)
return syserr.ErrInvalidArgument
@@ -569,15 +515,7 @@ func SetEntries(stk *stack.Stack, optVal []byte) *syserr.Error {
// - There are no chains without an unconditional final rule.
// - There are no chains without an unconditional underflow rule.
- table.SetMetadata(metadata{
- HookEntry: replace.HookEntry,
- Underflow: replace.Underflow,
- NumEntries: replace.NumEntries,
- Size: replace.Size,
- })
- stk.IPTables().ReplaceTable(replace.Name.String(), table)
-
- return nil
+ return syserr.TranslateNetstackError(stk.IPTables().ReplaceTable(replace.Name.String(), table))
}
// parseMatchers parses 0 or more matchers from optVal. optVal should contain
diff --git a/pkg/sentry/socket/netfilter/targets.go b/pkg/sentry/socket/netfilter/targets.go
index 84abe8d29..b91ba3ab3 100644
--- a/pkg/sentry/socket/netfilter/targets.go
+++ b/pkg/sentry/socket/netfilter/targets.go
@@ -30,6 +30,6 @@ type JumpTarget struct {
}
// Action implements stack.Target.Action.
-func (jt JumpTarget) Action(*stack.PacketBuffer, *stack.ConnTrackTable, stack.Hook, *stack.GSO, *stack.Route, tcpip.Address) (stack.RuleVerdict, int) {
+func (jt JumpTarget) Action(*stack.PacketBuffer, *stack.ConnTrack, stack.Hook, *stack.GSO, *stack.Route, tcpip.Address) (stack.RuleVerdict, int) {
return stack.RuleJump, jt.RuleNum
}
diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD
index 420e573c9..0546801bf 100644
--- a/pkg/sentry/socket/netlink/BUILD
+++ b/pkg/sentry/socket/netlink/BUILD
@@ -20,6 +20,7 @@ go_library(
"//pkg/sentry/device",
"//pkg/sentry/fs",
"//pkg/sentry/fs/fsutil",
+ "//pkg/sentry/fs/lock",
"//pkg/sentry/fsimpl/sockfs",
"//pkg/sentry/kernel",
"//pkg/sentry/kernel/auth",
@@ -29,13 +30,14 @@ go_library(
"//pkg/sentry/socket/unix",
"//pkg/sentry/socket/unix/transport",
"//pkg/sentry/vfs",
- "//pkg/sentry/vfs/lock",
"//pkg/sync",
"//pkg/syserr",
"//pkg/syserror",
"//pkg/tcpip",
"//pkg/usermem",
"//pkg/waiter",
+ "//tools/go_marshal/marshal",
+ "//tools/go_marshal/primitive",
],
)
diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go
index 81f34c5a2..98ca7add0 100644
--- a/pkg/sentry/socket/netlink/socket.go
+++ b/pkg/sentry/socket/netlink/socket.go
@@ -38,6 +38,8 @@ import (
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/usermem"
"gvisor.dev/gvisor/pkg/waiter"
+ "gvisor.dev/gvisor/tools/go_marshal/marshal"
+ "gvisor.dev/gvisor/tools/go_marshal/primitive"
)
const sizeOfInt32 int = 4
@@ -330,7 +332,7 @@ func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error {
}
// GetSockOpt implements socket.Socket.GetSockOpt.
-func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) {
+func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) {
switch level {
case linux.SOL_SOCKET:
switch name {
@@ -340,24 +342,26 @@ func (s *socketOpsCommon) GetSockOpt(t *kernel.Task, level int, name int, outPtr
}
s.mu.Lock()
defer s.mu.Unlock()
- return int32(s.sendBufferSize), nil
+ sendBufferSizeP := primitive.Int32(s.sendBufferSize)
+ return &sendBufferSizeP, nil
case linux.SO_RCVBUF:
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
// We don't have limit on receiving size.
- return int32(math.MaxInt32), nil
+ recvBufferSizeP := primitive.Int32(math.MaxInt32)
+ return &recvBufferSizeP, nil
case linux.SO_PASSCRED:
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
- var passcred int32
+ var passcred primitive.Int32
if s.Passcred() {
passcred = 1
}
- return passcred, nil
+ return &passcred, nil
default:
socket.GetSockOptEmitUnimplementedEvent(t, name)
diff --git a/pkg/sentry/socket/netlink/socket_vfs2.go b/pkg/sentry/socket/netlink/socket_vfs2.go
index 8bfee5193..dbcd8b49a 100644
--- a/pkg/sentry/socket/netlink/socket_vfs2.go
+++ b/pkg/sentry/socket/netlink/socket_vfs2.go
@@ -18,12 +18,12 @@ import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/arch"
+ fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/socket"
"gvisor.dev/gvisor/pkg/sentry/socket/unix"
"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/sentry/vfs"
- "gvisor.dev/gvisor/pkg/sentry/vfs/lock"
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/tcpip"
@@ -78,7 +78,7 @@ func NewVFS2(t *kernel.Task, skType linux.SockType, protocol Protocol) (*SocketV
sendBufferSize: defaultSendBufferSize,
},
}
- fd.LockFD.Init(&lock.FileLocks{})
+ fd.LockFD.Init(&vfs.FileLocks{})
return fd, nil
}
@@ -140,3 +140,13 @@ func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs
n, err := s.sendMsg(ctx, src, nil, 0, socket.ControlMessages{})
return int64(n), err.ToError()
}
+
+// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX.
+func (s *SocketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error {
+ return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block)
+}
+
+// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX.
+func (s *SocketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error {
+ return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence)
+}
diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD
index 0f592ecc3..1fb777a6c 100644
--- a/pkg/sentry/socket/netstack/BUILD
+++ b/pkg/sentry/socket/netstack/BUILD
@@ -28,6 +28,7 @@ go_library(
"//pkg/sentry/device",
"//pkg/sentry/fs",
"//pkg/sentry/fs/fsutil",
+ "//pkg/sentry/fs/lock",
"//pkg/sentry/fsimpl/sockfs",
"//pkg/sentry/inet",
"//pkg/sentry/kernel",
@@ -37,7 +38,6 @@ go_library(
"//pkg/sentry/socket/netfilter",
"//pkg/sentry/unimpl",
"//pkg/sentry/vfs",
- "//pkg/sentry/vfs/lock",
"//pkg/sync",
"//pkg/syserr",
"//pkg/syserror",
@@ -51,6 +51,8 @@ go_library(
"//pkg/tcpip/transport/udp",
"//pkg/usermem",
"//pkg/waiter",
+ "//tools/go_marshal/marshal",
+ "//tools/go_marshal/primitive",
"@org_golang_x_sys//unix:go_default_library",
],
)
diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go
index 738277391..44b3fff46 100644
--- a/pkg/sentry/socket/netstack/netstack.go
+++ b/pkg/sentry/socket/netstack/netstack.go
@@ -26,6 +26,7 @@ package netstack
import (
"bytes"
+ "fmt"
"io"
"math"
"reflect"
@@ -61,6 +62,8 @@ import (
"gvisor.dev/gvisor/pkg/tcpip/transport/udp"
"gvisor.dev/gvisor/pkg/usermem"
"gvisor.dev/gvisor/pkg/waiter"
+ "gvisor.dev/gvisor/tools/go_marshal/marshal"
+ "gvisor.dev/gvisor/tools/go_marshal/primitive"
)
func mustCreateMetric(name, description string) *tcpip.StatCounter {
@@ -191,6 +194,8 @@ var Metrics = tcpip.Stats{
MalformedPacketsReceived: mustCreateMetric("/netstack/udp/malformed_packets_received", "Number of incoming UDP datagrams dropped due to the UDP header being in a malformed state."),
PacketsSent: mustCreateMetric("/netstack/udp/packets_sent", "Number of UDP datagrams sent."),
PacketSendErrors: mustCreateMetric("/netstack/udp/packet_send_errors", "Number of UDP datagrams failed to be sent."),
+ ChecksumErrors: mustCreateMetric("/netstack/udp/checksum_errors", "Number of UDP datagrams dropped due to bad checksums."),
+ InvalidSourceAddress: mustCreateMetric("/netstack/udp/invalid_source", "Number of UDP datagrams dropped due to invalid source address."),
},
}
@@ -295,8 +300,9 @@ type socketOpsCommon struct {
readView buffer.View
// readCM holds control message information for the last packet read
// from Endpoint.
- readCM tcpip.ControlMessages
- sender tcpip.FullAddress
+ readCM tcpip.ControlMessages
+ sender tcpip.FullAddress
+ linkPacketInfo tcpip.LinkPacketInfo
// sockOptTimestamp corresponds to SO_TIMESTAMP. When true, timestamps
// of returned messages can be returned via control messages. When
@@ -445,8 +451,21 @@ func (s *socketOpsCommon) fetchReadView() *syserr.Error {
}
s.readView = nil
s.sender = tcpip.FullAddress{}
+ s.linkPacketInfo = tcpip.LinkPacketInfo{}
- v, cms, err := s.Endpoint.Read(&s.sender)
+ var v buffer.View
+ var cms tcpip.ControlMessages
+ var err *tcpip.Error
+
+ switch e := s.Endpoint.(type) {
+ // The ordering of these interfaces matters. The most specific
+ // interfaces must be specified before the more generic Endpoint
+ // interface.
+ case tcpip.PacketEndpoint:
+ v, cms, err = e.ReadPacket(&s.sender, &s.linkPacketInfo)
+ case tcpip.Endpoint:
+ v, cms, err = e.Read(&s.sender)
+ }
if err != nil {
atomic.StoreUint32(&s.readViewHasData, 0)
return syserr.TranslateNetstackError(err)
@@ -893,7 +912,7 @@ func (s *socketOpsCommon) Shutdown(t *kernel.Task, how int) *syserr.Error {
// GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by
// tcpip.Endpoint.
-func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) {
+func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) {
// TODO(b/78348848): Unlike other socket options, SO_TIMESTAMP is
// implemented specifically for netstack.SocketOperations rather than
// commonEndpoint. commonEndpoint should be extended to support socket
@@ -903,25 +922,25 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
- val := int32(0)
+ val := primitive.Int32(0)
s.readMu.Lock()
defer s.readMu.Unlock()
if s.sockOptTimestamp {
val = 1
}
- return val, nil
+ return &val, nil
}
if level == linux.SOL_TCP && name == linux.TCP_INQ {
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
- val := int32(0)
+ val := primitive.Int32(0)
s.readMu.Lock()
defer s.readMu.Unlock()
if s.sockOptInq {
val = 1
}
- return val, nil
+ return &val, nil
}
if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP {
@@ -939,7 +958,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us
if err != nil {
return nil, err
}
- return info, nil
+ return &info, nil
case linux.IPT_SO_GET_ENTRIES:
if outLen < linux.SizeOfIPTGetEntries {
@@ -954,7 +973,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us
if err != nil {
return nil, err
}
- return entries, nil
+ return &entries, nil
}
}
@@ -964,7 +983,7 @@ func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr us
// GetSockOpt can be used to implement the linux syscall getsockopt(2) for
// sockets backed by a commonEndpoint.
-func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (interface{}, *syserr.Error) {
+func GetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, level, name, outLen int) (marshal.Marshallable, *syserr.Error) {
switch level {
case linux.SOL_SOCKET:
return getSockOptSocket(t, s, ep, family, skType, name, outLen)
@@ -997,7 +1016,7 @@ func boolToInt32(v bool) int32 {
}
// getSockOptSocket implements GetSockOpt when level is SOL_SOCKET.
-func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (interface{}, *syserr.Error) {
+func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, family int, skType linux.SockType, name, outLen int) (marshal.Marshallable, *syserr.Error) {
// TODO(b/124056281): Stop rejecting short optLen values in getsockopt.
switch name {
case linux.SO_ERROR:
@@ -1008,9 +1027,12 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
// Get the last error and convert it.
err := ep.GetSockOpt(tcpip.ErrorOption{})
if err == nil {
- return int32(0), nil
+ optP := primitive.Int32(0)
+ return &optP, nil
}
- return int32(syserr.TranslateNetstackError(err).ToLinux().Number()), nil
+
+ optP := primitive.Int32(syserr.TranslateNetstackError(err).ToLinux().Number())
+ return &optP, nil
case linux.SO_PEERCRED:
if family != linux.AF_UNIX || outLen < syscall.SizeofUcred {
@@ -1018,11 +1040,12 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
}
tcred := t.Credentials()
- return syscall.Ucred{
- Pid: int32(t.ThreadGroup().ID()),
- Uid: uint32(tcred.EffectiveKUID.In(tcred.UserNamespace).OrOverflow()),
- Gid: uint32(tcred.EffectiveKGID.In(tcred.UserNamespace).OrOverflow()),
- }, nil
+ creds := linux.ControlMessageCredentials{
+ PID: int32(t.ThreadGroup().ID()),
+ UID: uint32(tcred.EffectiveKUID.In(tcred.UserNamespace).OrOverflow()),
+ GID: uint32(tcred.EffectiveKGID.In(tcred.UserNamespace).OrOverflow()),
+ }
+ return &creds, nil
case linux.SO_PASSCRED:
if outLen < sizeOfInt32 {
@@ -1033,7 +1056,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
case linux.SO_SNDBUF:
if outLen < sizeOfInt32 {
@@ -1049,7 +1074,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
size = math.MaxInt32
}
- return int32(size), nil
+ sizeP := primitive.Int32(size)
+ return &sizeP, nil
case linux.SO_RCVBUF:
if outLen < sizeOfInt32 {
@@ -1065,7 +1091,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
size = math.MaxInt32
}
- return int32(size), nil
+ sizeP := primitive.Int32(size)
+ return &sizeP, nil
case linux.SO_REUSEADDR:
if outLen < sizeOfInt32 {
@@ -1076,7 +1103,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
case linux.SO_REUSEPORT:
if outLen < sizeOfInt32 {
@@ -1087,7 +1115,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
case linux.SO_BINDTODEVICE:
var v tcpip.BindToDeviceOption
@@ -1095,7 +1125,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
return nil, syserr.TranslateNetstackError(err)
}
if v == 0 {
- return []byte{}, nil
+ var b primitive.ByteSlice
+ return &b, nil
}
if outLen < linux.IFNAMSIZ {
return nil, syserr.ErrInvalidArgument
@@ -1110,7 +1141,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
// interface was removed.
return nil, syserr.ErrUnknownDevice
}
- return append([]byte(nic.Name), 0), nil
+
+ name := primitive.ByteSlice(append([]byte(nic.Name), 0))
+ return &name, nil
case linux.SO_BROADCAST:
if outLen < sizeOfInt32 {
@@ -1121,7 +1154,9 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
case linux.SO_KEEPALIVE:
if outLen < sizeOfInt32 {
@@ -1132,13 +1167,17 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
case linux.SO_LINGER:
if outLen < linux.SizeOfLinger {
return nil, syserr.ErrInvalidArgument
}
- return linux.Linger{}, nil
+
+ linger := linux.Linger{}
+ return &linger, nil
case linux.SO_SNDTIMEO:
// TODO(igudger): Linux allows shorter lengths for partial results.
@@ -1146,7 +1185,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
return nil, syserr.ErrInvalidArgument
}
- return linux.NsecToTimeval(s.SendTimeout()), nil
+ sendTimeout := linux.NsecToTimeval(s.SendTimeout())
+ return &sendTimeout, nil
case linux.SO_RCVTIMEO:
// TODO(igudger): Linux allows shorter lengths for partial results.
@@ -1154,7 +1194,8 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
return nil, syserr.ErrInvalidArgument
}
- return linux.NsecToTimeval(s.RecvTimeout()), nil
+ recvTimeout := linux.NsecToTimeval(s.RecvTimeout())
+ return &recvTimeout, nil
case linux.SO_OOBINLINE:
if outLen < sizeOfInt32 {
@@ -1166,7 +1207,20 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
return nil, syserr.TranslateNetstackError(err)
}
- return int32(v), nil
+ vP := primitive.Int32(v)
+ return &vP, nil
+
+ case linux.SO_NO_CHECK:
+ if outLen < sizeOfInt32 {
+ return nil, syserr.ErrInvalidArgument
+ }
+
+ v, err := ep.GetSockOptBool(tcpip.NoChecksumOption)
+ if err != nil {
+ return nil, syserr.TranslateNetstackError(err)
+ }
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
default:
socket.GetSockOptEmitUnimplementedEvent(t, name)
@@ -1175,7 +1229,7 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam
}
// getSockOptTCP implements GetSockOpt when level is SOL_TCP.
-func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interface{}, *syserr.Error) {
+func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (marshal.Marshallable, *syserr.Error) {
switch name {
case linux.TCP_NODELAY:
if outLen < sizeOfInt32 {
@@ -1186,7 +1240,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(!v), nil
+
+ vP := primitive.Int32(boolToInt32(!v))
+ return &vP, nil
case linux.TCP_CORK:
if outLen < sizeOfInt32 {
@@ -1197,7 +1253,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
case linux.TCP_QUICKACK:
if outLen < sizeOfInt32 {
@@ -1208,7 +1266,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
case linux.TCP_MAXSEG:
if outLen < sizeOfInt32 {
@@ -1219,8 +1279,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ vP := primitive.Int32(v)
+ return &vP, nil
case linux.TCP_KEEPIDLE:
if outLen < sizeOfInt32 {
@@ -1231,8 +1291,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
if err := ep.GetSockOpt(&v); err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(time.Duration(v) / time.Second), nil
+ keepAliveIdle := primitive.Int32(time.Duration(v) / time.Second)
+ return &keepAliveIdle, nil
case linux.TCP_KEEPINTVL:
if outLen < sizeOfInt32 {
@@ -1243,8 +1303,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
if err := ep.GetSockOpt(&v); err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(time.Duration(v) / time.Second), nil
+ keepAliveInterval := primitive.Int32(time.Duration(v) / time.Second)
+ return &keepAliveInterval, nil
case linux.TCP_KEEPCNT:
if outLen < sizeOfInt32 {
@@ -1255,8 +1315,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ vP := primitive.Int32(v)
+ return &vP, nil
case linux.TCP_USER_TIMEOUT:
if outLen < sizeOfInt32 {
@@ -1267,8 +1327,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
if err := ep.GetSockOpt(&v); err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(time.Duration(v) / time.Millisecond), nil
+ tcpUserTimeout := primitive.Int32(time.Duration(v) / time.Millisecond)
+ return &tcpUserTimeout, nil
case linux.TCP_INFO:
var v tcpip.TCPInfoOption
@@ -1281,12 +1341,13 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
info := linux.TCPInfo{}
// Linux truncates the output binary to outLen.
- ib := binary.Marshal(nil, usermem.ByteOrder, &info)
- if len(ib) > outLen {
- ib = ib[:outLen]
+ buf := t.CopyScratchBuffer(info.SizeBytes())
+ info.MarshalUnsafe(buf)
+ if len(buf) > outLen {
+ buf = buf[:outLen]
}
-
- return ib, nil
+ bufP := primitive.ByteSlice(buf)
+ return &bufP, nil
case linux.TCP_CC_INFO,
linux.TCP_NOTSENT_LOWAT,
@@ -1316,7 +1377,9 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
}
b := make([]byte, toCopy)
copy(b, v)
- return b, nil
+
+ bP := primitive.ByteSlice(b)
+ return &bP, nil
case linux.TCP_LINGER2:
if outLen < sizeOfInt32 {
@@ -1328,7 +1391,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
return nil, syserr.TranslateNetstackError(err)
}
- return int32(time.Duration(v) / time.Second), nil
+ lingerTimeout := primitive.Int32(time.Duration(v) / time.Second)
+ return &lingerTimeout, nil
case linux.TCP_DEFER_ACCEPT:
if outLen < sizeOfInt32 {
@@ -1340,7 +1404,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
return nil, syserr.TranslateNetstackError(err)
}
- return int32(time.Duration(v) / time.Second), nil
+ tcpDeferAccept := primitive.Int32(time.Duration(v) / time.Second)
+ return &tcpDeferAccept, nil
case linux.TCP_SYNCNT:
if outLen < sizeOfInt32 {
@@ -1351,8 +1416,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ vP := primitive.Int32(v)
+ return &vP, nil
case linux.TCP_WINDOW_CLAMP:
if outLen < sizeOfInt32 {
@@ -1363,8 +1428,8 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
-
- return int32(v), nil
+ vP := primitive.Int32(v)
+ return &vP, nil
default:
emitUnimplementedEventTCP(t, name)
}
@@ -1372,7 +1437,7 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa
}
// getSockOptIPv6 implements GetSockOpt when level is SOL_IPV6.
-func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interface{}, *syserr.Error) {
+func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (marshal.Marshallable, *syserr.Error) {
switch name {
case linux.IPV6_V6ONLY:
if outLen < sizeOfInt32 {
@@ -1383,7 +1448,9 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
case linux.IPV6_PATHMTU:
t.Kernel().EmitUnimplementedEvent(t)
@@ -1391,21 +1458,24 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf
case linux.IPV6_TCLASS:
// Length handling for parity with Linux.
if outLen == 0 {
- return make([]byte, 0), nil
+ var b primitive.ByteSlice
+ return &b, nil
}
v, err := ep.GetSockOptInt(tcpip.IPv6TrafficClassOption)
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- uintv := uint32(v)
+ uintv := primitive.Uint32(v)
// Linux truncates the output binary to outLen.
- ib := binary.Marshal(nil, usermem.ByteOrder, &uintv)
+ ib := t.CopyScratchBuffer(uintv.SizeBytes())
+ uintv.MarshalUnsafe(ib)
// Handle cases where outLen is lesser than sizeOfInt32.
if len(ib) > outLen {
ib = ib[:outLen]
}
- return ib, nil
+ ibP := primitive.ByteSlice(ib)
+ return &ibP, nil
case linux.IPV6_RECVTCLASS:
if outLen < sizeOfInt32 {
@@ -1416,7 +1486,9 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
default:
emitUnimplementedEventIPv6(t, name)
@@ -1425,7 +1497,7 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf
}
// getSockOptIP implements GetSockOpt when level is SOL_IP.
-func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family int) (interface{}, *syserr.Error) {
+func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family int) (marshal.Marshallable, *syserr.Error) {
switch name {
case linux.IP_TTL:
if outLen < sizeOfInt32 {
@@ -1438,11 +1510,12 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
}
// Fill in the default value, if needed.
- if v == 0 {
- v = DefaultTTL
+ vP := primitive.Int32(v)
+ if vP == 0 {
+ vP = DefaultTTL
}
- return int32(v), nil
+ return &vP, nil
case linux.IP_MULTICAST_TTL:
if outLen < sizeOfInt32 {
@@ -1454,7 +1527,8 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
return nil, syserr.TranslateNetstackError(err)
}
- return int32(v), nil
+ vP := primitive.Int32(v)
+ return &vP, nil
case linux.IP_MULTICAST_IF:
if outLen < len(linux.InetAddr{}) {
@@ -1468,7 +1542,7 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
a, _ := ConvertAddress(linux.AF_INET, tcpip.FullAddress{Addr: v.InterfaceAddr})
- return a.(*linux.SockAddrInet).Addr, nil
+ return &a.(*linux.SockAddrInet).Addr, nil
case linux.IP_MULTICAST_LOOP:
if outLen < sizeOfInt32 {
@@ -1479,21 +1553,26 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
case linux.IP_TOS:
// Length handling for parity with Linux.
if outLen == 0 {
- return []byte(nil), nil
+ var b primitive.ByteSlice
+ return &b, nil
}
v, err := ep.GetSockOptInt(tcpip.IPv4TOSOption)
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
if outLen < sizeOfInt32 {
- return uint8(v), nil
+ vP := primitive.Uint8(v)
+ return &vP, nil
}
- return int32(v), nil
+ vP := primitive.Int32(v)
+ return &vP, nil
case linux.IP_RECVTOS:
if outLen < sizeOfInt32 {
@@ -1504,7 +1583,9 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
case linux.IP_PKTINFO:
if outLen < sizeOfInt32 {
@@ -1515,7 +1596,9 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in
if err != nil {
return nil, syserr.TranslateNetstackError(err)
}
- return boolToInt32(v), nil
+
+ vP := primitive.Int32(boolToInt32(v))
+ return &vP, nil
default:
emitUnimplementedEventIP(t, name)
@@ -1719,6 +1802,14 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam
return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.OutOfBandInlineOption(v)))
+ case linux.SO_NO_CHECK:
+ if len(optVal) < sizeOfInt32 {
+ return syserr.ErrInvalidArgument
+ }
+
+ v := usermem.ByteOrder.Uint32(optVal)
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.NoChecksumOption, v != 0))
+
case linux.SO_LINGER:
if len(optVal) < linux.SizeOfLinger {
return syserr.ErrInvalidArgument
@@ -1733,6 +1824,11 @@ func setSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, nam
return nil
+ case linux.SO_DETACH_FILTER:
+ // optval is ignored.
+ var v tcpip.SocketDetachFilterOption
+ return syserr.TranslateNetstackError(ep.SetSockOpt(v))
+
default:
socket.SetSockOptEmitUnimplementedEvent(t, name)
}
@@ -2092,13 +2188,22 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s
}
return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveIPPacketInfoOption, v != 0))
+ case linux.IP_HDRINCL:
+ if len(optVal) == 0 {
+ return nil
+ }
+ v, err := parseIntOrChar(optVal)
+ if err != nil {
+ return err
+ }
+ return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.IPHdrIncludedOption, v != 0))
+
case linux.IP_ADD_SOURCE_MEMBERSHIP,
linux.IP_BIND_ADDRESS_NO_PORT,
linux.IP_BLOCK_SOURCE,
linux.IP_CHECKSUM,
linux.IP_DROP_SOURCE_MEMBERSHIP,
linux.IP_FREEBIND,
- linux.IP_HDRINCL,
linux.IP_IPSEC_POLICY,
linux.IP_MINTTL,
linux.IP_MSFILTER,
@@ -2419,6 +2524,23 @@ func (s *socketOpsCommon) fillCmsgInq(cmsg *socket.ControlMessages) {
cmsg.IP.Inq = int32(len(s.readView) + rcvBufUsed)
}
+func toLinuxPacketType(pktType tcpip.PacketType) uint8 {
+ switch pktType {
+ case tcpip.PacketHost:
+ return linux.PACKET_HOST
+ case tcpip.PacketOtherHost:
+ return linux.PACKET_OTHERHOST
+ case tcpip.PacketOutgoing:
+ return linux.PACKET_OUTGOING
+ case tcpip.PacketBroadcast:
+ return linux.PACKET_BROADCAST
+ case tcpip.PacketMulticast:
+ return linux.PACKET_MULTICAST
+ default:
+ panic(fmt.Sprintf("unknown packet type: %d", pktType))
+ }
+}
+
// nonBlockingRead issues a non-blocking read.
//
// TODO(b/78348848): Support timestamps for stream sockets.
@@ -2474,6 +2596,11 @@ func (s *socketOpsCommon) nonBlockingRead(ctx context.Context, dst usermem.IOSeq
var addrLen uint32
if isPacket && senderRequested {
addr, addrLen = ConvertAddress(s.family, s.sender)
+ switch v := addr.(type) {
+ case *linux.SockAddrLink:
+ v.Protocol = htons(uint16(s.linkPacketInfo.Protocol))
+ v.PacketType = toLinuxPacketType(s.linkPacketInfo.PktType)
+ }
}
if peek {
@@ -2708,11 +2835,16 @@ func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO,
}
func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.SyscallArguments) (uintptr, error) {
+ t := kernel.TaskFromContext(ctx)
+ if t == nil {
+ panic("ioctl(2) may only be called from a task goroutine")
+ }
+
// SIOCGSTAMP is implemented by netstack rather than all commonEndpoint
// sockets.
// TODO(b/78348848): Add a commonEndpoint method to support SIOCGSTAMP.
switch args[1].Int() {
- case syscall.SIOCGSTAMP:
+ case linux.SIOCGSTAMP:
s.readMu.Lock()
defer s.readMu.Unlock()
if !s.timestampValid {
@@ -2720,9 +2852,7 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy
}
tv := linux.NsecToTimeval(s.timestampNS)
- _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), &tv, usermem.IOOpts{
- AddressSpaceActive: true,
- })
+ _, err := tv.CopyOut(t, args[2].Pointer())
return 0, err
case linux.TIOCINQ:
@@ -2741,9 +2871,8 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy
}
// Copy result to userspace.
- _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{
- AddressSpaceActive: true,
- })
+ vP := primitive.Int32(v)
+ _, err := vP.CopyOut(t, args[2].Pointer())
return 0, err
}
@@ -2752,52 +2881,49 @@ func (s *socketOpsCommon) ioctl(ctx context.Context, io usermem.IO, args arch.Sy
// Ioctl performs a socket ioctl.
func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.SyscallArguments) (uintptr, error) {
+ t := kernel.TaskFromContext(ctx)
+ if t == nil {
+ panic("ioctl(2) may only be called from a task goroutine")
+ }
+
switch arg := int(args[1].Int()); arg {
- case syscall.SIOCGIFFLAGS,
- syscall.SIOCGIFADDR,
- syscall.SIOCGIFBRDADDR,
- syscall.SIOCGIFDSTADDR,
- syscall.SIOCGIFHWADDR,
- syscall.SIOCGIFINDEX,
- syscall.SIOCGIFMAP,
- syscall.SIOCGIFMETRIC,
- syscall.SIOCGIFMTU,
- syscall.SIOCGIFNAME,
- syscall.SIOCGIFNETMASK,
- syscall.SIOCGIFTXQLEN:
+ case linux.SIOCGIFFLAGS,
+ linux.SIOCGIFADDR,
+ linux.SIOCGIFBRDADDR,
+ linux.SIOCGIFDSTADDR,
+ linux.SIOCGIFHWADDR,
+ linux.SIOCGIFINDEX,
+ linux.SIOCGIFMAP,
+ linux.SIOCGIFMETRIC,
+ linux.SIOCGIFMTU,
+ linux.SIOCGIFNAME,
+ linux.SIOCGIFNETMASK,
+ linux.SIOCGIFTXQLEN,
+ linux.SIOCETHTOOL:
var ifr linux.IFReq
- if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &ifr, usermem.IOOpts{
- AddressSpaceActive: true,
- }); err != nil {
+ if _, err := ifr.CopyIn(t, args[2].Pointer()); err != nil {
return 0, err
}
if err := interfaceIoctl(ctx, io, arg, &ifr); err != nil {
return 0, err.ToError()
}
- _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), &ifr, usermem.IOOpts{
- AddressSpaceActive: true,
- })
+ _, err := ifr.CopyOut(t, args[2].Pointer())
return 0, err
- case syscall.SIOCGIFCONF:
+ case linux.SIOCGIFCONF:
// Return a list of interface addresses or the buffer size
// necessary to hold the list.
var ifc linux.IFConf
- if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &ifc, usermem.IOOpts{
- AddressSpaceActive: true,
- }); err != nil {
+ if _, err := ifc.CopyIn(t, args[2].Pointer()); err != nil {
return 0, err
}
- if err := ifconfIoctl(ctx, io, &ifc); err != nil {
+ if err := ifconfIoctl(ctx, t, io, &ifc); err != nil {
return 0, err
}
- _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), ifc, usermem.IOOpts{
- AddressSpaceActive: true,
- })
-
+ _, err := ifc.CopyOut(t, args[2].Pointer())
return 0, err
case linux.TIOCINQ:
@@ -2810,9 +2936,8 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc
v = math.MaxInt32
}
// Copy result to userspace.
- _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{
- AddressSpaceActive: true,
- })
+ vP := primitive.Int32(v)
+ _, err := vP.CopyOut(t, args[2].Pointer())
return 0, err
case linux.TIOCOUTQ:
@@ -2826,9 +2951,8 @@ func Ioctl(ctx context.Context, ep commonEndpoint, io usermem.IO, args arch.Sysc
}
// Copy result to userspace.
- _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), int32(v), usermem.IOOpts{
- AddressSpaceActive: true,
- })
+ vP := primitive.Int32(v)
+ _, err := vP.CopyOut(t, args[2].Pointer())
return 0, err
case linux.SIOCGIFMEM, linux.SIOCGIFPFLAGS, linux.SIOCGMIIPHY, linux.SIOCGMIIREG:
@@ -2854,7 +2978,7 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe
// SIOCGIFNAME uses ifr.ifr_ifindex rather than ifr.ifr_name to
// identify a device.
- if arg == syscall.SIOCGIFNAME {
+ if arg == linux.SIOCGIFNAME {
// Gets the name of the interface given the interface index
// stored in ifr_ifindex.
index = int32(usermem.ByteOrder.Uint32(ifr.Data[:4]))
@@ -2877,21 +3001,28 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe
}
switch arg {
- case syscall.SIOCGIFINDEX:
+ case linux.SIOCGIFINDEX:
// Copy out the index to the data.
usermem.ByteOrder.PutUint32(ifr.Data[:], uint32(index))
- case syscall.SIOCGIFHWADDR:
+ case linux.SIOCGIFHWADDR:
// Copy the hardware address out.
- ifr.Data[0] = 6 // IEEE802.2 arp type.
- ifr.Data[1] = 0
+ //
+ // Refer: https://linux.die.net/man/7/netdevice
+ // SIOCGIFHWADDR, SIOCSIFHWADDR
+ //
+ // Get or set the hardware address of a device using
+ // ifr_hwaddr. The hardware address is specified in a struct
+ // sockaddr. sa_family contains the ARPHRD_* device type,
+ // sa_data the L2 hardware address starting from byte 0. Setting
+ // the hardware address is a privileged operation.
+ usermem.ByteOrder.PutUint16(ifr.Data[:], iface.DeviceType)
n := copy(ifr.Data[2:], iface.Addr)
for i := 2 + n; i < len(ifr.Data); i++ {
ifr.Data[i] = 0 // Clear padding.
}
- usermem.ByteOrder.PutUint16(ifr.Data[:2], uint16(n))
- case syscall.SIOCGIFFLAGS:
+ case linux.SIOCGIFFLAGS:
f, err := interfaceStatusFlags(stack, iface.Name)
if err != nil {
return err
@@ -2900,7 +3031,7 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe
// matches Linux behavior.
usermem.ByteOrder.PutUint16(ifr.Data[:2], uint16(f))
- case syscall.SIOCGIFADDR:
+ case linux.SIOCGIFADDR:
// Copy the IPv4 address out.
for _, addr := range stack.InterfaceAddrs()[index] {
// This ioctl is only compatible with AF_INET addresses.
@@ -2911,32 +3042,32 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe
break
}
- case syscall.SIOCGIFMETRIC:
+ case linux.SIOCGIFMETRIC:
// Gets the metric of the device. As per netdevice(7), this
// always just sets ifr_metric to 0.
usermem.ByteOrder.PutUint32(ifr.Data[:4], 0)
- case syscall.SIOCGIFMTU:
+ case linux.SIOCGIFMTU:
// Gets the MTU of the device.
usermem.ByteOrder.PutUint32(ifr.Data[:4], iface.MTU)
- case syscall.SIOCGIFMAP:
+ case linux.SIOCGIFMAP:
// Gets the hardware parameters of the device.
// TODO(gvisor.dev/issue/505): Implement.
- case syscall.SIOCGIFTXQLEN:
+ case linux.SIOCGIFTXQLEN:
// Gets the transmit queue length of the device.
// TODO(gvisor.dev/issue/505): Implement.
- case syscall.SIOCGIFDSTADDR:
+ case linux.SIOCGIFDSTADDR:
// Gets the destination address of a point-to-point device.
// TODO(gvisor.dev/issue/505): Implement.
- case syscall.SIOCGIFBRDADDR:
+ case linux.SIOCGIFBRDADDR:
// Gets the broadcast address of a device.
// TODO(gvisor.dev/issue/505): Implement.
- case syscall.SIOCGIFNETMASK:
+ case linux.SIOCGIFNETMASK:
// Gets the network mask of a device.
for _, addr := range stack.InterfaceAddrs()[index] {
// This ioctl is only compatible with AF_INET addresses.
@@ -2953,6 +3084,14 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe
break
}
+ case linux.SIOCETHTOOL:
+ // Stubbed out for now, Ideally we should implement the required
+ // sub-commands for ETHTOOL
+ //
+ // See:
+ // https://github.com/torvalds/linux/blob/aa0c9086b40c17a7ad94425b3b70dd1fdd7497bf/net/core/dev_ioctl.c
+ return syserr.ErrEndpointOperation
+
default:
// Not a valid call.
return syserr.ErrInvalidArgument
@@ -2962,7 +3101,7 @@ func interfaceIoctl(ctx context.Context, io usermem.IO, arg int, ifr *linux.IFRe
}
// ifconfIoctl populates a struct ifconf for the SIOCGIFCONF ioctl.
-func ifconfIoctl(ctx context.Context, io usermem.IO, ifc *linux.IFConf) error {
+func ifconfIoctl(ctx context.Context, t *kernel.Task, io usermem.IO, ifc *linux.IFConf) error {
// If Ptr is NULL, return the necessary buffer size via Len.
// Otherwise, write up to Len bytes starting at Ptr containing ifreq
// structs.
@@ -2999,9 +3138,7 @@ func ifconfIoctl(ctx context.Context, io usermem.IO, ifc *linux.IFConf) error {
// Copy the ifr to userspace.
dst := uintptr(ifc.Ptr) + uintptr(ifc.Len)
ifc.Len += int32(linux.SizeOfIFReq)
- if _, err := usermem.CopyObjectOut(ctx, io, usermem.Addr(dst), ifr, usermem.IOOpts{
- AddressSpaceActive: true,
- }); err != nil {
+ if _, err := ifr.CopyOut(t, usermem.Addr(dst)); err != nil {
return err
}
}
diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go
index 1412a4810..a9025b0ec 100644
--- a/pkg/sentry/socket/netstack/netstack_vfs2.go
+++ b/pkg/sentry/socket/netstack/netstack_vfs2.go
@@ -19,18 +19,20 @@ import (
"gvisor.dev/gvisor/pkg/amutex"
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/arch"
+ fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock"
"gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs"
"gvisor.dev/gvisor/pkg/sentry/inet"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/socket"
"gvisor.dev/gvisor/pkg/sentry/socket/netfilter"
"gvisor.dev/gvisor/pkg/sentry/vfs"
- "gvisor.dev/gvisor/pkg/sentry/vfs/lock"
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/usermem"
"gvisor.dev/gvisor/pkg/waiter"
+ "gvisor.dev/gvisor/tools/go_marshal/marshal"
+ "gvisor.dev/gvisor/tools/go_marshal/primitive"
)
// SocketVFS2 encapsulates all the state needed to represent a network stack
@@ -66,7 +68,7 @@ func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, qu
protocol: protocol,
},
}
- s.LockFD.Init(&lock.FileLocks{})
+ s.LockFD.Init(&vfs.FileLocks{})
vfsfd := &s.vfsfd
if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{
DenyPRead: true,
@@ -200,7 +202,7 @@ func (s *SocketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.Syscal
// GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by
// tcpip.Endpoint.
-func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) {
+func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) {
// TODO(b/78348848): Unlike other socket options, SO_TIMESTAMP is
// implemented specifically for netstack.SocketVFS2 rather than
// commonEndpoint. commonEndpoint should be extended to support socket
@@ -210,25 +212,25 @@ func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
- val := int32(0)
+ val := primitive.Int32(0)
s.readMu.Lock()
defer s.readMu.Unlock()
if s.sockOptTimestamp {
val = 1
}
- return val, nil
+ return &val, nil
}
if level == linux.SOL_TCP && name == linux.TCP_INQ {
if outLen < sizeOfInt32 {
return nil, syserr.ErrInvalidArgument
}
- val := int32(0)
+ val := primitive.Int32(0)
s.readMu.Lock()
defer s.readMu.Unlock()
if s.sockOptInq {
val = 1
}
- return val, nil
+ return &val, nil
}
if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP {
@@ -246,7 +248,7 @@ func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.
if err != nil {
return nil, err
}
- return info, nil
+ return &info, nil
case linux.IPT_SO_GET_ENTRIES:
if outLen < linux.SizeOfIPTGetEntries {
@@ -261,7 +263,7 @@ func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.
if err != nil {
return nil, err
}
- return entries, nil
+ return &entries, nil
}
}
@@ -318,3 +320,13 @@ func (s *SocketVFS2) SetSockOpt(t *kernel.Task, level int, name int, optVal []by
return SetSockOpt(t, s, s.Endpoint, level, name, optVal)
}
+
+// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX.
+func (s *SocketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error {
+ return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block)
+}
+
+// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX.
+func (s *SocketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error {
+ return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence)
+}
diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go
index 9b44c2b89..67737ae87 100644
--- a/pkg/sentry/socket/netstack/stack.go
+++ b/pkg/sentry/socket/netstack/stack.go
@@ -15,10 +15,11 @@
package netstack
import (
+ "fmt"
+
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/log"
"gvisor.dev/gvisor/pkg/sentry/inet"
- "gvisor.dev/gvisor/pkg/sentry/socket/netfilter"
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/tcpip"
@@ -41,19 +42,29 @@ func (s *Stack) SupportsIPv6() bool {
return s.Stack.CheckNetworkProtocol(ipv6.ProtocolNumber)
}
+// Converts Netstack's ARPHardwareType to equivalent linux constants.
+func toLinuxARPHardwareType(t header.ARPHardwareType) uint16 {
+ switch t {
+ case header.ARPHardwareNone:
+ return linux.ARPHRD_NONE
+ case header.ARPHardwareLoopback:
+ return linux.ARPHRD_LOOPBACK
+ case header.ARPHardwareEther:
+ return linux.ARPHRD_ETHER
+ default:
+ panic(fmt.Sprintf("unknown ARPHRD type: %d", t))
+ }
+}
+
// Interfaces implements inet.Stack.Interfaces.
func (s *Stack) Interfaces() map[int32]inet.Interface {
is := make(map[int32]inet.Interface)
for id, ni := range s.Stack.NICInfo() {
- var devType uint16
- if ni.Flags.Loopback {
- devType = linux.ARPHRD_LOOPBACK
- }
is[int32(id)] = inet.Interface{
Name: ni.Name,
Addr: []byte(ni.LinkAddress),
Flags: uint32(nicStateFlagsToLinux(ni.Flags)),
- DeviceType: devType,
+ DeviceType: toLinuxARPHardwareType(ni.ARPHardwareType),
MTU: ni.MTU,
}
}
@@ -314,7 +325,7 @@ func (s *Stack) Statistics(stat interface{}, arg string) error {
udp.PacketsSent.Value(), // OutDatagrams.
udp.ReceiveBufferErrors.Value(), // RcvbufErrors.
0, // Udp/SndbufErrors.
- 0, // Udp/InCsumErrors.
+ udp.ChecksumErrors.Value(), // Udp/InCsumErrors.
0, // Udp/IgnoredMulti.
}
default:
@@ -366,11 +377,6 @@ func (s *Stack) IPTables() (*stack.IPTables, error) {
return s.Stack.IPTables(), nil
}
-// FillIPTablesMetadata populates stack's IPTables with metadata.
-func (s *Stack) FillIPTablesMetadata() {
- netfilter.FillIPTablesMetadata(s.Stack)
-}
-
// Resume implements inet.Stack.Resume.
func (s *Stack) Resume() {
s.Stack.Resume()
diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go
index 6580bd6e9..d112757fb 100644
--- a/pkg/sentry/socket/socket.go
+++ b/pkg/sentry/socket/socket.go
@@ -35,6 +35,7 @@ import (
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/usermem"
+ "gvisor.dev/gvisor/tools/go_marshal/marshal"
)
// ControlMessages represents the union of unix control messages and tcpip
@@ -86,7 +87,7 @@ type SocketOps interface {
Shutdown(t *kernel.Task, how int) *syserr.Error
// GetSockOpt implements the getsockopt(2) linux syscall.
- GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error)
+ GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error)
// SetSockOpt implements the setsockopt(2) linux syscall.
SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error
@@ -407,7 +408,6 @@ func emitUnimplementedEvent(t *kernel.Task, name int) {
linux.SO_MARK,
linux.SO_MAX_PACING_RATE,
linux.SO_NOFCS,
- linux.SO_NO_CHECK,
linux.SO_OOBINLINE,
linux.SO_PASSCRED,
linux.SO_PASSSEC,
diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD
index 7d4cc80fe..061a689a9 100644
--- a/pkg/sentry/socket/unix/BUILD
+++ b/pkg/sentry/socket/unix/BUILD
@@ -21,6 +21,7 @@ go_library(
"//pkg/sentry/device",
"//pkg/sentry/fs",
"//pkg/sentry/fs/fsutil",
+ "//pkg/sentry/fs/lock",
"//pkg/sentry/fsimpl/sockfs",
"//pkg/sentry/kernel",
"//pkg/sentry/kernel/time",
@@ -29,11 +30,11 @@ go_library(
"//pkg/sentry/socket/netstack",
"//pkg/sentry/socket/unix/transport",
"//pkg/sentry/vfs",
- "//pkg/sentry/vfs/lock",
"//pkg/syserr",
"//pkg/syserror",
"//pkg/tcpip",
"//pkg/usermem",
"//pkg/waiter",
+ "//tools/go_marshal/marshal",
],
)
diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go
index 09c6d3b27..a1e49cc57 100644
--- a/pkg/sentry/socket/unix/transport/connectioned.go
+++ b/pkg/sentry/socket/unix/transport/connectioned.go
@@ -476,6 +476,9 @@ func (e *connectionedEndpoint) Readiness(mask waiter.EventMask) waiter.EventMask
// State implements socket.Socket.State.
func (e *connectionedEndpoint) State() uint32 {
+ e.Lock()
+ defer e.Unlock()
+
if e.Connected() {
return linux.SS_CONNECTED
}
diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go
index 4bb2b6ff4..0482d33cf 100644
--- a/pkg/sentry/socket/unix/unix.go
+++ b/pkg/sentry/socket/unix/unix.go
@@ -40,6 +40,7 @@ import (
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/usermem"
"gvisor.dev/gvisor/pkg/waiter"
+ "gvisor.dev/gvisor/tools/go_marshal/marshal"
)
// SocketOperations is a Unix socket. It is similar to a netstack socket,
@@ -184,7 +185,7 @@ func (s *SocketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO,
// GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) {
+func (s *SocketOperations) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) {
return netstack.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen)
}
diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go
index 8c32371a2..05c16fcfe 100644
--- a/pkg/sentry/socket/unix/unix_vfs2.go
+++ b/pkg/sentry/socket/unix/unix_vfs2.go
@@ -19,6 +19,7 @@ import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/fspath"
"gvisor.dev/gvisor/pkg/sentry/arch"
+ fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock"
"gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/socket"
@@ -26,12 +27,12 @@ import (
"gvisor.dev/gvisor/pkg/sentry/socket/netstack"
"gvisor.dev/gvisor/pkg/sentry/socket/unix/transport"
"gvisor.dev/gvisor/pkg/sentry/vfs"
- "gvisor.dev/gvisor/pkg/sentry/vfs/lock"
"gvisor.dev/gvisor/pkg/syserr"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/tcpip"
"gvisor.dev/gvisor/pkg/usermem"
"gvisor.dev/gvisor/pkg/waiter"
+ "gvisor.dev/gvisor/tools/go_marshal/marshal"
)
// SocketVFS2 implements socket.SocketVFS2 (and by extension,
@@ -53,7 +54,7 @@ func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType)
mnt := t.Kernel().SocketMount()
d := sockfs.NewDentry(t.Credentials(), mnt)
- fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d, &lock.FileLocks{})
+ fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d, &vfs.FileLocks{})
if err != nil {
return nil, syserr.FromError(err)
}
@@ -62,7 +63,7 @@ func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType)
// NewFileDescription creates and returns a socket file description
// corresponding to the given mount and dentry.
-func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint32, mnt *vfs.Mount, d *vfs.Dentry, locks *lock.FileLocks) (*vfs.FileDescription, error) {
+func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint32, mnt *vfs.Mount, d *vfs.Dentry, locks *vfs.FileLocks) (*vfs.FileDescription, error) {
// You can create AF_UNIX, SOCK_RAW sockets. They're the same as
// SOCK_DGRAM and don't require CAP_NET_RAW.
if stype == linux.SOCK_RAW {
@@ -89,7 +90,7 @@ func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint3
// GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by
// a transport.Endpoint.
-func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) {
+func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) {
return netstack.GetSockOpt(t, s, s.ep, linux.AF_UNIX, s.ep.Type(), level, name, outLen)
}
@@ -300,6 +301,16 @@ func (s *SocketVFS2) SetSockOpt(t *kernel.Task, level int, name int, optVal []by
return netstack.SetSockOpt(t, s, s.ep, level, name, optVal)
}
+// LockPOSIX implements vfs.FileDescriptionImpl.LockPOSIX.
+func (s *SocketVFS2) LockPOSIX(ctx context.Context, uid fslock.UniqueID, t fslock.LockType, start, length uint64, whence int16, block fslock.Blocker) error {
+ return s.Locks().LockPOSIX(ctx, &s.vfsfd, uid, t, start, length, whence, block)
+}
+
+// UnlockPOSIX implements vfs.FileDescriptionImpl.UnlockPOSIX.
+func (s *SocketVFS2) UnlockPOSIX(ctx context.Context, uid fslock.UniqueID, start, length uint64, whence int16) error {
+ return s.Locks().UnlockPOSIX(ctx, &s.vfsfd, uid, start, length, whence)
+}
+
// providerVFS2 is a unix domain socket provider for VFS2.
type providerVFS2 struct{}
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-04 13:09:24 +0000