diff options
Diffstat (limited to 'sysdep/unix/io.c')
-rw-r--r-- | sysdep/unix/io.c | 377 |
1 files changed, 134 insertions, 243 deletions
diff --git a/sysdep/unix/io.c b/sysdep/unix/io.c index 0e5adc14..88d187a4 100644 --- a/sysdep/unix/io.c +++ b/sysdep/unix/io.c @@ -76,7 +76,7 @@ rf_free(resource *r) } static void -rf_dump(resource *r) +rf_dump(resource *r, unsigned indent UNUSED) { struct rfile *a = (struct rfile *) r; @@ -140,7 +140,7 @@ times_update(void) if ((ts.tv_sec < 0) || (((u64) ts.tv_sec) > ((u64) 1 << 40))) log(L_WARN "Monotonic clock is crazy"); - + btime new_time = ts.tv_sec S + ts.tv_nsec NS; if (new_time < old_time) @@ -722,11 +722,7 @@ sk_log_error(sock *s, const char *p) * Actual struct birdsock code */ -static list sock_list; -static struct birdsock *current_sock; -static struct birdsock *stored_sock; - -static inline sock * +sock * sk_next(sock *s) { if (!s->n.next->next) @@ -774,8 +770,7 @@ sk_ssh_free(sock *s) if (ssh->channel) { - if (ssh_channel_is_open(ssh->channel)) - ssh_channel_close(ssh->channel); + ssh_channel_close(ssh->channel); ssh_channel_free(ssh->channel); ssh->channel = NULL; } @@ -789,12 +784,12 @@ sk_ssh_free(sock *s) } #endif + static void sk_free(resource *r) { sock *s = (sock *) r; - ASSERT_DIE(!s->loop || birdloop_inside(s->loop)); sk_free_bufs(s); #ifdef HAVE_LIBSSH @@ -802,30 +797,10 @@ sk_free(resource *r) sk_ssh_free(s); #endif - if (s->cork) - { - LOCK_DOMAIN(cork, s->cork->lock); - if (enlisted(&s->cork_node)) - rem_node(&s->cork_node); - UNLOCK_DOMAIN(cork, s->cork->lock); - } - - if (!s->loop) - ; - else if (s->flags & SKF_THREAD) - sk_stop(s); - else - { - 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->loop) + birdloop_remove_socket(s->loop, s); - if (s->type != SK_SSH && s->type != SK_SSH_ACTIVE && s->fd != -1) + if (s->fd >= 0 && s->type != SK_SSH && s->type != SK_SSH_ACTIVE) close(s->fd); s->fd = -1; @@ -876,7 +851,7 @@ sk_reallocate(sock *s) } static void -sk_dump(resource *r) +sk_dump(resource *r, unsigned indent UNUSED) { sock *s = (sock *) r; static char *sk_type_names[] = { "TCP<", "TCP>", "TCP", "UDP", NULL, "IP", NULL, "MAGIC", "UNIX<", "UNIX", "SSH>", "SSH", "DEL!" }; @@ -1038,12 +1013,6 @@ sk_setup(sock *s) } static void -sk_insert(sock *s) -{ - add_tail(&sock_list, &s->n); -} - -static void sk_tcp_connected(sock *s) { sockaddr sa; @@ -1116,14 +1085,7 @@ sk_passive_connected(sock *s, int type) return 1; } - if (s->flags & SKF_PASSIVE_THREAD) - t->flags |= SKF_THREAD; - else - { - ASSERT_DIE(s->loop == &main_birdloop); - t->loop = &main_birdloop; - sk_insert(t); - } + birdloop_add_socket(s->loop, t); sk_alloc_bufs(t); s->rx_hook(t, 0); @@ -1169,34 +1131,45 @@ sk_ssh_connect(sock *s) { int server_identity_is_ok = 1; +#ifdef HAVE_SSH_OLD_SERVER_VALIDATION_API +#define ssh_session_is_known_server ssh_is_server_known +#define SSH_KNOWN_HOSTS_OK SSH_SERVER_KNOWN_OK +#define SSH_KNOWN_HOSTS_UNKNOWN SSH_SERVER_NOT_KNOWN +#define SSH_KNOWN_HOSTS_CHANGED SSH_SERVER_KNOWN_CHANGED +#define SSH_KNOWN_HOSTS_NOT_FOUND SSH_SERVER_FILE_NOT_FOUND +#define SSH_KNOWN_HOSTS_ERROR SSH_SERVER_ERROR +#define SSH_KNOWN_HOSTS_OTHER SSH_SERVER_FOUND_OTHER +#endif + /* Check server identity */ - switch (ssh_is_server_known(s->ssh->session)) + switch (ssh_session_is_known_server(s->ssh->session)) { #define LOG_WARN_ABOUT_SSH_SERVER_VALIDATION(s,msg,args...) log(L_WARN "SSH Identity %s@%s:%u: " msg, (s)->ssh->username, (s)->host, (s)->dport, ## args); - case SSH_SERVER_KNOWN_OK: + case SSH_KNOWN_HOSTS_OK: /* The server is known and has not changed. */ break; - case SSH_SERVER_NOT_KNOWN: + case SSH_KNOWN_HOSTS_UNKNOWN: LOG_WARN_ABOUT_SSH_SERVER_VALIDATION(s, "The server is unknown, its public key was not found in the known host file %s", s->ssh->server_hostkey_path); + server_identity_is_ok = 0; break; - case SSH_SERVER_KNOWN_CHANGED: + case SSH_KNOWN_HOSTS_CHANGED: LOG_WARN_ABOUT_SSH_SERVER_VALIDATION(s, "The server key has changed. Either you are under attack or the administrator changed the key."); server_identity_is_ok = 0; break; - case SSH_SERVER_FILE_NOT_FOUND: + case SSH_KNOWN_HOSTS_NOT_FOUND: LOG_WARN_ABOUT_SSH_SERVER_VALIDATION(s, "The known host file %s does not exist", s->ssh->server_hostkey_path); server_identity_is_ok = 0; break; - case SSH_SERVER_ERROR: + case SSH_KNOWN_HOSTS_ERROR: LOG_WARN_ABOUT_SSH_SERVER_VALIDATION(s, "Some error happened"); server_identity_is_ok = 0; break; - case SSH_SERVER_FOUND_OTHER: + case SSH_KNOWN_HOSTS_OTHER: LOG_WARN_ABOUT_SSH_SERVER_VALIDATION(s, "The server gave use a key of a type while we had an other type recorded. " \ "It is a possible attack."); server_identity_is_ok = 0; @@ -1327,6 +1300,7 @@ sk_open_ssh(sock *s) /** * sk_open - open a socket + * @loop: loop * @s: socket * * This function takes a socket resource created by sk_new() and @@ -1336,7 +1310,7 @@ sk_open_ssh(sock *s) * Result: 0 for success, -1 for an error. */ int -sk_open(sock *s) +sk_open(sock *s, struct birdloop *loop) { int af = AF_UNSPEC; int fd = -1; @@ -1345,17 +1319,6 @@ sk_open(sock *s) ip_addr bind_addr = IPA_NONE; sockaddr sa; - if (s->flags & SKF_THREAD) - { - ASSERT_DIE(s->loop && (s->loop != &main_birdloop)); - ASSERT_DIE(birdloop_inside(s->loop)); - } - else - { - ASSERT_DIE(!s->loop); - s->loop = &main_birdloop; - } - if (s->type <= SK_IP) { /* @@ -1500,9 +1463,7 @@ sk_open(sock *s) sk_alloc_bufs(s); } - if (!(s->flags & SKF_THREAD)) - sk_insert(s); - + birdloop_add_socket(loop, s); return 0; err: @@ -1512,7 +1473,7 @@ err: } int -sk_open_unix(sock *s, char *name) +sk_open_unix(sock *s, struct birdloop *loop, char *name) { struct sockaddr_un sa; int fd; @@ -1539,41 +1500,10 @@ sk_open_unix(sock *s, char *name) return -1; s->fd = fd; - s->loop = &main_birdloop; - sk_insert(s); + birdloop_add_socket(loop, s); 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) @@ -1696,6 +1626,13 @@ sk_recvmsg(sock *s) static inline void reset_tx_buffer(sock *s) { s->ttx = s->tpos = s->tbuf; } +_Bool +sk_tx_pending(sock *s) +{ + return s->ttx != s->tpos; +} + + static int sk_maybe_write(sock *s) { @@ -1706,7 +1643,7 @@ sk_maybe_write(sock *s) case SK_TCP: case SK_MAGIC: case SK_UNIX: - while (s->ttx != s->tpos) + while (sk_tx_pending(s)) { e = write(s->fd, s->ttx, s->tpos - s->ttx); @@ -1719,7 +1656,6 @@ sk_maybe_write(sock *s) s->err_hook(s, (errno != EPIPE) ? errno : 0); return -1; } - sk_ping(s); return 0; } s->ttx += e; @@ -1729,7 +1665,7 @@ sk_maybe_write(sock *s) #ifdef HAVE_LIBSSH case SK_SSH: - while (s->ttx != s->tpos) + while (sk_tx_pending(s)) { e = ssh_channel_write(s->ssh->channel, s->ttx, s->tpos - s->ttx); @@ -1812,7 +1748,12 @@ sk_send(sock *s, unsigned len) { s->ttx = s->tbuf; s->tpos = s->tbuf + len; - return sk_maybe_write(s); + + int e = sk_maybe_write(s); + if (e == 0) /* Trigger thread poll reload to poll this socket's write. */ + socket_changed(s); + + return e; } /** @@ -1859,7 +1800,7 @@ call_rx_hook(sock *s, int size) if (s->rx_hook(s, size)) { /* We need to be careful since the socket could have been deleted by the hook */ - if (current_sock == s) + if (s->loop->sock_active == s) s->rpos = s->rbuf; } } @@ -1913,8 +1854,8 @@ sk_read_ssh(sock *s) /* sk_read() and sk_write() are called from BFD's event loop */ -int -sk_read(sock *s, int revents) +static inline int +sk_read_noflush(sock *s, int revents) { switch (s->type) { @@ -1927,25 +1868,6 @@ sk_read(sock *s, int revents) case SK_TCP: case SK_UNIX: { - if (s->cork) - { - int cont = 0; - LOCK_DOMAIN(cork, s->cork->lock); - if (!enlisted(&s->cork_node)) - if (s->cork->count) - { -// log(L_TRACE "Socket %p corked", s); - add_tail(&s->cork->sockets, &s->cork_node); - sk_ping(s); - } - else - cont = 1; - UNLOCK_DOMAIN(cork, s->cork->lock); - - if (!cont) - return 0; - } - int c = read(s->fd, s->rpos, s->rbuf + s->rbsize - s->rpos); if (c < 0) @@ -1996,7 +1918,15 @@ sk_read(sock *s, int revents) } int -sk_write(sock *s) +sk_read(sock *s, int revents) +{ + int e = sk_read_noflush(s, revents); + tmp_flush(); + return e; +} + +static inline int +sk_write_noflush(sock *s) { switch (s->type) { @@ -2034,7 +1964,7 @@ sk_write(sock *s) #endif default: - if (s->ttx != s->tpos && sk_maybe_write(s) > 0) + if (sk_tx_pending(s) && sk_maybe_write(s) > 0) { if (s->tx_hook) s->tx_hook(s); @@ -2044,6 +1974,14 @@ sk_write(sock *s) } } +int +sk_write(sock *s) +{ + int e = sk_write_noflush(s); + tmp_flush(); + return e; +} + int sk_is_ipv4(sock *s) { return s->af == AF_INET; } @@ -2062,6 +2000,7 @@ sk_err(sock *s, int revents) } s->err_hook(s, se); + tmp_flush(); } void @@ -2071,11 +2010,11 @@ sk_dump_all(void) sock *s; debug("Open sockets:\n"); - WALK_LIST(n, sock_list) + WALK_LIST(n, main_birdloop.sock_list) { s = SKIP_BACK(sock, n, n); debug("%p ", s); - sk_dump(&s->r); + sk_dump(&s->r, 3); } debug("\n"); } @@ -2104,15 +2043,15 @@ static btime loop_time; static void io_update_time(void) { - last_io_time = current_time_update(); + last_io_time = current_time(); if (event_open) { 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", - event_open->hook, event_open->data, (int) (event_open->duration TO_MS)); + log(L_WARN "Event 0x%p 0x%p took %u.%03u ms", + event_open->hook, event_open->data, (uint) (event_open->duration TO_MS), (uint) (event_open->duration % 1000)); event_open = NULL; } @@ -2176,6 +2115,8 @@ watchdog_sigalrm(int sig UNUSED) config->latency_limit = 0xffffffff; io_update_time(); + debug_safe("Watchdog timer timed out\n"); + /* We want core dump */ abort(); } @@ -2216,8 +2157,8 @@ watchdog_stop(void) 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); + log(L_WARN "I/O loop cycle took %u.%03u ms for %d events", + (uint) (duration TO_MS), (uint) (duration % 1000), event_log_num); } @@ -2228,7 +2169,7 @@ watchdog_stop(void) void io_init(void) { - init_list(&sock_list); + init_list(&main_birdloop.sock_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"); @@ -2245,20 +2186,17 @@ 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); +sock *stored_sock; void io_loop(void) { int poll_tout, timeout; - int nfds, events, pout; - int reload_requested = 0; + int events, pout; timer *t; - sock *s; - node *n; - int fdmax = 256; - struct pollfd *pfd = xmalloc(fdmax * sizeof(struct pollfd)); + struct pfd pfd; + BUFFER_INIT(pfd.pfd, &root_pool, 16); + BUFFER_INIT(pfd.loop, &root_pool, 16); watchdog_start1(); for(;;) @@ -2270,14 +2208,8 @@ io_loop(void) timers_fire(&main_birdloop.time, 1); io_close_event(); -#if DEBUGGING -#define PERIODIC_WAKEUP 86400000 -#else -#define PERIODIC_WAKEUP 3000 -#endif -restart_poll: // FIXME - poll_tout = ((reload_requested || events) ? 0 : PERIODIC_WAKEUP); /* Time in milliseconds */ + poll_tout = (events ? 0 : 3000); /* Time in milliseconds */ if (t = timers_first(&main_birdloop.time)) { times_update(); @@ -2285,40 +2217,11 @@ restart_poll: poll_tout = MIN(poll_tout, timeout); } - /* 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 && !ev_corked(s->cork)) - { - pfd[nfds].fd = s->fd; - pfd[nfds].events |= POLLIN; - } - if (s->tx_hook && s->ttx != s->tpos) - { - pfd[nfds].fd = s->fd; - pfd[nfds].events |= POLLOUT; - } - if (pfd[nfds].fd != -1) - { - s->index = nfds; - nfds++; - } - else - s->index = -1; + BUFFER_FLUSH(pfd.pfd); + BUFFER_FLUSH(pfd.loop); - if (nfds >= fdmax) - { - fdmax *= 2; - pfd = xrealloc(pfd, fdmax * sizeof(struct pollfd)); - } - } + pipe_pollin(&main_birdloop.thread->wakeup, &pfd); + sockets_prepare(&main_birdloop, &pfd); /* * Yes, this is racy. But even if the signal comes before this test @@ -2350,7 +2253,7 @@ restart_poll: /* And finally enter poll() to find active sockets */ watchdog_stop(); birdloop_leave(&main_birdloop); - pout = poll(pfd, nfds, poll_tout); + pout = poll(pfd.pfd.data, pfd.pfd.used, poll_tout); birdloop_enter(&main_birdloop); watchdog_start(); @@ -2358,111 +2261,99 @@ restart_poll: { if (errno == EINTR || errno == EAGAIN) continue; - die("poll: %m"); + bug("poll: %m"); } - - if (pout && (pfd[0].revents & POLLIN)) - { - /* IO loop reload requested */ - pipe_drain(main_birdloop.wakeup_fds[0]); - reload_requested = 1; - goto restart_poll; - } - - if (reload_requested) - { - reload_requested = 0; - atomic_exchange_explicit(&main_birdloop.ping_sent, 0, memory_order_acq_rel); - } - if (pout) { + if (pfd.pfd.data[0].revents & POLLIN) + { + /* IO loop reload requested */ + pipe_drain(&main_birdloop.thread->wakeup); + atomic_fetch_and_explicit(&main_birdloop.thread_transition, ~LTT_PING, memory_order_acq_rel); + continue; + } + times_update(); /* guaranteed to be non-empty */ - current_sock = SKIP_BACK(sock, n, HEAD(sock_list)); + main_birdloop.sock_active = SKIP_BACK(sock, n, HEAD(main_birdloop.sock_list)); - while (current_sock) + while (main_birdloop.sock_active) + { + sock *s = main_birdloop.sock_active; + if (s->index != -1) { - sock *s = current_sock; - if (s->index == -1) - { - current_sock = sk_next(s); - goto next; - } - int e; int steps; steps = MAX_STEPS; - if (s->fast_rx && (pfd[s->index].revents & POLLIN) && s->rx_hook) + if (s->fast_rx && (pfd.pfd.data[s->index].revents & POLLIN) && s->rx_hook) do { steps--; io_log_event(s->rx_hook, s->data); - e = sk_read(s, pfd[s->index].revents); - if (s != current_sock) - goto next; + e = sk_read(s, pfd.pfd.data[s->index].revents); } - while (e && s->rx_hook && steps); + while (e && (main_birdloop.sock_active == s) && s->rx_hook && steps); + + if (s != main_birdloop.sock_active) + continue; steps = MAX_STEPS; - if (pfd[s->index].revents & POLLOUT) + if (pfd.pfd.data[s->index].revents & POLLOUT) do { steps--; io_log_event(s->tx_hook, s->data); e = sk_write(s); - if (s != current_sock) - goto next; } - while (e && steps); + while (e && (main_birdloop.sock_active == s) && steps); - current_sock = sk_next(s); - next: ; + if (s != main_birdloop.sock_active) + continue; } + main_birdloop.sock_active = sk_next(s); + } + short_loops++; if (events && (short_loops < SHORT_LOOP_MAX)) continue; short_loops = 0; int count = 0; - current_sock = stored_sock; - if (current_sock == NULL) - current_sock = SKIP_BACK(sock, n, HEAD(sock_list)); + main_birdloop.sock_active = stored_sock; + if (main_birdloop.sock_active == NULL) + main_birdloop.sock_active = SKIP_BACK(sock, n, HEAD(main_birdloop.sock_list)); - while (current_sock && count < MAX_RX_STEPS) + while (main_birdloop.sock_active && count < MAX_RX_STEPS) { - sock *s = current_sock; + sock *s = main_birdloop.sock_active; if (s->index == -1) - { - current_sock = sk_next(s); - goto next2; - } + goto next2; - if (!s->fast_rx && (pfd[s->index].revents & POLLIN) && s->rx_hook) + if (!s->fast_rx && (pfd.pfd.data[s->index].revents & POLLIN) && s->rx_hook) { count++; io_log_event(s->rx_hook, s->data); - sk_read(s, pfd[s->index].revents); - if (s != current_sock) - goto next2; + sk_read(s, pfd.pfd.data[s->index].revents); + if (s != main_birdloop.sock_active) + continue; } - if (pfd[s->index].revents & (POLLHUP | POLLERR)) + if (pfd.pfd.data[s->index].revents & (POLLHUP | POLLERR)) { - sk_err(s, pfd[s->index].revents); - if (s != current_sock) - goto next2; + sk_err(s, pfd.pfd.data[s->index].revents); + if (s != main_birdloop.sock_active) + continue; } - current_sock = sk_next(s); next2: ; + main_birdloop.sock_active = sk_next(s); } - stored_sock = current_sock; + stored_sock = main_birdloop.sock_active; } } } |