]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/socket.c
projects / thirdparty / systemd.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
core: reduce the number of stalled PIDs from the watched processes list when possible
[thirdparty/systemd.git] / src / core / socket.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <arpa/inet.h>
4 #include <errno.h>
5 #include <fcntl.h>
6 #include <mqueue.h>
7 #include <netinet/tcp.h>
8 #include <signal.h>
9 #include <sys/epoll.h>
10 #include <sys/stat.h>
11 #include <unistd.h>
12 #include <linux/sctp.h>
13
14 #include "alloc-util.h"
15 #include "bpf-firewall.h"
16 #include "bus-error.h"
17 #include "bus-util.h"
18 #include "copy.h"
19 #include "dbus-socket.h"
20 #include "dbus-unit.h"
21 #include "def.h"
22 #include "exit-status.h"
23 #include "fd-util.h"
24 #include "format-util.h"
25 #include "fs-util.h"
26 #include "in-addr-util.h"
27 #include "io-util.h"
28 #include "ip-protocol-list.h"
29 #include "label.h"
30 #include "log.h"
31 #include "missing.h"
32 #include "mkdir.h"
33 #include "parse-util.h"
34 #include "path-util.h"
35 #include "process-util.h"
36 #include "selinux-util.h"
37 #include "serialize.h"
38 #include "signal-util.h"
39 #include "smack-util.h"
40 #include "socket.h"
41 #include "special.h"
42 #include "string-table.h"
43 #include "string-util.h"
44 #include "strv.h"
45 #include "unit-name.h"
46 #include "unit.h"
47 #include "user-util.h"
48
49 struct SocketPeer {
50 unsigned n_ref;
51
52 Socket *socket;
53 union sockaddr_union peer;
54 socklen_t peer_salen;
55 };
56
57 static const UnitActiveState state_translation_table[_SOCKET_STATE_MAX] = {
58 [SOCKET_DEAD] = UNIT_INACTIVE,
59 [SOCKET_START_PRE] = UNIT_ACTIVATING,
60 [SOCKET_START_CHOWN] = UNIT_ACTIVATING,
61 [SOCKET_START_POST] = UNIT_ACTIVATING,
62 [SOCKET_LISTENING] = UNIT_ACTIVE,
63 [SOCKET_RUNNING] = UNIT_ACTIVE,
64 [SOCKET_STOP_PRE] = UNIT_DEACTIVATING,
65 [SOCKET_STOP_PRE_SIGTERM] = UNIT_DEACTIVATING,
66 [SOCKET_STOP_PRE_SIGKILL] = UNIT_DEACTIVATING,
67 [SOCKET_STOP_POST] = UNIT_DEACTIVATING,
68 [SOCKET_FINAL_SIGTERM] = UNIT_DEACTIVATING,
69 [SOCKET_FINAL_SIGKILL] = UNIT_DEACTIVATING,
70 [SOCKET_FAILED] = UNIT_FAILED
71 };
72
73 static int socket_dispatch_io(sd_event_source *source, int fd, uint32_t revents, void *userdata);
74 static int socket_dispatch_timer(sd_event_source *source, usec_t usec, void *userdata);
75
76 static void socket_init(Unit *u) {
77 Socket *s = SOCKET(u);
78
79 assert(u);
80 assert(u->load_state == UNIT_STUB);
81
82 s->backlog = SOMAXCONN;
83 s->timeout_usec = u->manager->default_timeout_start_usec;
84 s->directory_mode = 0755;
85 s->socket_mode = 0666;
86
87 s->max_connections = 64;
88
89 s->priority = -1;
90 s->ip_tos = -1;
91 s->ip_ttl = -1;
92 s->mark = -1;
93
94 s->exec_context.std_output = u->manager->default_std_output;
95 s->exec_context.std_error = u->manager->default_std_error;
96
97 s->control_command_id = _SOCKET_EXEC_COMMAND_INVALID;
98
99 s->trigger_limit.interval = USEC_INFINITY;
100 s->trigger_limit.burst = (unsigned) -1;
101 }
102
103 static void socket_unwatch_control_pid(Socket *s) {
104 assert(s);
105
106 if (s->control_pid <= 0)
107 return;
108
109 unit_unwatch_pid(UNIT(s), s->control_pid);
110 s->control_pid = 0;
111 }
112
113 static void socket_cleanup_fd_list(SocketPort *p) {
114 assert(p);
115
116 close_many(p->auxiliary_fds, p->n_auxiliary_fds);
117 p->auxiliary_fds = mfree(p->auxiliary_fds);
118 p->n_auxiliary_fds = 0;
119 }
120
121 void socket_free_ports(Socket *s) {
122 SocketPort *p;
123
124 assert(s);
125
126 while ((p = s->ports)) {
127 LIST_REMOVE(port, s->ports, p);
128
129 sd_event_source_unref(p->event_source);
130
131 socket_cleanup_fd_list(p);
132 safe_close(p->fd);
133 free(p->path);
134 free(p);
135 }
136 }
137
138 static void socket_done(Unit *u) {
139 Socket *s = SOCKET(u);
140 SocketPeer *p;
141
142 assert(s);
143
144 socket_free_ports(s);
145
146 while ((p = set_steal_first(s->peers_by_address)))
147 p->socket = NULL;
148
149 s->peers_by_address = set_free(s->peers_by_address);
150
151 s->exec_runtime = exec_runtime_unref(s->exec_runtime, false);
152 exec_command_free_array(s->exec_command, _SOCKET_EXEC_COMMAND_MAX);
153 s->control_command = NULL;
154
155 dynamic_creds_unref(&s->dynamic_creds);
156
157 socket_unwatch_control_pid(s);
158
159 unit_ref_unset(&s->service);
160
161 s->tcp_congestion = mfree(s->tcp_congestion);
162 s->bind_to_device = mfree(s->bind_to_device);
163
164 s->smack = mfree(s->smack);
165 s->smack_ip_in = mfree(s->smack_ip_in);
166 s->smack_ip_out = mfree(s->smack_ip_out);
167
168 strv_free(s->symlinks);
169
170 s->user = mfree(s->user);
171 s->group = mfree(s->group);
172
173 s->fdname = mfree(s->fdname);
174
175 s->timer_event_source = sd_event_source_unref(s->timer_event_source);
176 }
177
178 static int socket_arm_timer(Socket *s, usec_t usec) {
179 int r;
180
181 assert(s);
182
183 if (s->timer_event_source) {
184 r = sd_event_source_set_time(s->timer_event_source, usec);
185 if (r < 0)
186 return r;
187
188 return sd_event_source_set_enabled(s->timer_event_source, SD_EVENT_ONESHOT);
189 }
190
191 if (usec == USEC_INFINITY)
192 return 0;
193
194 r = sd_event_add_time(
195 UNIT(s)->manager->event,
196 &s->timer_event_source,
197 CLOCK_MONOTONIC,
198 usec, 0,
199 socket_dispatch_timer, s);
200 if (r < 0)
201 return r;
202
203 (void) sd_event_source_set_description(s->timer_event_source, "socket-timer");
204
205 return 0;
206 }
207
208 int socket_instantiate_service(Socket *s) {
209 _cleanup_free_ char *prefix = NULL, *name = NULL;
210 int r;
211 Unit *u;
212
213 assert(s);
214
215 /* This fills in s->service if it isn't filled in yet. For
216 * Accept=yes sockets we create the next connection service
217 * here. For Accept=no this is mostly a NOP since the service
218 * is figured out at load time anyway. */
219
220 if (UNIT_DEREF(s->service))
221 return 0;
222
223 if (!s->accept)
224 return 0;
225
226 r = unit_name_to_prefix(UNIT(s)->id, &prefix);
227 if (r < 0)
228 return r;
229
230 if (asprintf(&name, "%s@%u.service", prefix, s->n_accepted) < 0)
231 return -ENOMEM;
232
233 r = manager_load_unit(UNIT(s)->manager, name, NULL, NULL, &u);
234 if (r < 0)
235 return r;
236
237 unit_ref_set(&s->service, UNIT(s), u);
238
239 return unit_add_two_dependencies(UNIT(s), UNIT_BEFORE, UNIT_TRIGGERS, u, false, UNIT_DEPENDENCY_IMPLICIT);
240 }
241
242 static bool have_non_accept_socket(Socket *s) {
243 SocketPort *p;
244
245 assert(s);
246
247 if (!s->accept)
248 return true;
249
250 LIST_FOREACH(port, p, s->ports) {
251
252 if (p->type != SOCKET_SOCKET)
253 return true;
254
255 if (!socket_address_can_accept(&p->address))
256 return true;
257 }
258
259 return false;
260 }
261
262 static int socket_add_mount_dependencies(Socket *s) {
263 SocketPort *p;
264 int r;
265
266 assert(s);
267
268 LIST_FOREACH(port, p, s->ports) {
269 const char *path = NULL;
270
271 if (p->type == SOCKET_SOCKET)
272 path = socket_address_get_path(&p->address);
273 else if (IN_SET(p->type, SOCKET_FIFO, SOCKET_SPECIAL, SOCKET_USB_FUNCTION))
274 path = p->path;
275
276 if (!path)
277 continue;
278
279 r = unit_require_mounts_for(UNIT(s), path, UNIT_DEPENDENCY_FILE);
280 if (r < 0)
281 return r;
282 }
283
284 return 0;
285 }
286
287 static int socket_add_device_dependencies(Socket *s) {
288 char *t;
289
290 assert(s);
291
292 if (!s->bind_to_device || streq(s->bind_to_device, "lo"))
293 return 0;
294
295 t = strjoina("/sys/subsystem/net/devices/", s->bind_to_device);
296 return unit_add_node_dependency(UNIT(s), t, false, UNIT_BINDS_TO, UNIT_DEPENDENCY_FILE);
297 }
298
299 static int socket_add_default_dependencies(Socket *s) {
300 int r;
301 assert(s);
302
303 if (!UNIT(s)->default_dependencies)
304 return 0;
305
306 r = unit_add_dependency_by_name(UNIT(s), UNIT_BEFORE, SPECIAL_SOCKETS_TARGET, true, UNIT_DEPENDENCY_DEFAULT);
307 if (r < 0)
308 return r;
309
310 if (MANAGER_IS_SYSTEM(UNIT(s)->manager)) {
311 r = unit_add_two_dependencies_by_name(UNIT(s), UNIT_AFTER, UNIT_REQUIRES, SPECIAL_SYSINIT_TARGET, true, UNIT_DEPENDENCY_DEFAULT);
312 if (r < 0)
313 return r;
314 }
315
316 return unit_add_two_dependencies_by_name(UNIT(s), UNIT_BEFORE, UNIT_CONFLICTS, SPECIAL_SHUTDOWN_TARGET, true, UNIT_DEPENDENCY_DEFAULT);
317 }
318
319 _pure_ static bool socket_has_exec(Socket *s) {
320 unsigned i;
321 assert(s);
322
323 for (i = 0; i < _SOCKET_EXEC_COMMAND_MAX; i++)
324 if (s->exec_command[i])
325 return true;
326
327 return false;
328 }
329
330 static int socket_add_extras(Socket *s) {
331 Unit *u = UNIT(s);
332 int r;
333
334 assert(s);
335
336 /* Pick defaults for the trigger limit, if nothing was explicitly configured. We pick a relatively high limit
337 * in Accept=yes mode, and a lower limit for Accept=no. Reason: in Accept=yes mode we are invoking accept()
338 * ourselves before the trigger limit can hit, thus incoming connections are taken off the socket queue quickly
339 * and reliably. This is different for Accept=no, where the spawned service has to take the incoming traffic
340 * off the queues, which it might not necessarily do. Moreover, while Accept=no services are supposed to
341 * process whatever is queued in one go, and thus should normally never have to be started frequently. This is
342 * different for Accept=yes where each connection is processed by a new service instance, and thus frequent
343 * service starts are typical. */
344
345 if (s->trigger_limit.interval == USEC_INFINITY)
346 s->trigger_limit.interval = 2 * USEC_PER_SEC;
347
348 if (s->trigger_limit.burst == (unsigned) -1) {
349 if (s->accept)
350 s->trigger_limit.burst = 200;
351 else
352 s->trigger_limit.burst = 20;
353 }
354
355 if (have_non_accept_socket(s)) {
356
357 if (!UNIT_DEREF(s->service)) {
358 Unit *x;
359
360 r = unit_load_related_unit(u, ".service", &x);
361 if (r < 0)
362 return r;
363
364 unit_ref_set(&s->service, u, x);
365 }
366
367 r = unit_add_two_dependencies(u, UNIT_BEFORE, UNIT_TRIGGERS, UNIT_DEREF(s->service), true, UNIT_DEPENDENCY_IMPLICIT);
368 if (r < 0)
369 return r;
370 }
371
372 r = socket_add_mount_dependencies(s);
373 if (r < 0)
374 return r;
375
376 r = socket_add_device_dependencies(s);
377 if (r < 0)
378 return r;
379
380 r = unit_patch_contexts(u);
381 if (r < 0)
382 return r;
383
384 if (socket_has_exec(s)) {
385 r = unit_add_exec_dependencies(u, &s->exec_context);
386 if (r < 0)
387 return r;
388 }
389
390 r = unit_set_default_slice(u);
391 if (r < 0)
392 return r;
393
394 r = socket_add_default_dependencies(s);
395 if (r < 0)
396 return r;
397
398 return 0;
399 }
400
401 static const char *socket_find_symlink_target(Socket *s) {
402 const char *found = NULL;
403 SocketPort *p;
404
405 LIST_FOREACH(port, p, s->ports) {
406 const char *f = NULL;
407
408 switch (p->type) {
409
410 case SOCKET_FIFO:
411 f = p->path;
412 break;
413
414 case SOCKET_SOCKET:
415 f = socket_address_get_path(&p->address);
416 break;
417
418 default:
419 break;
420 }
421
422 if (f) {
423 if (found)
424 return NULL;
425
426 found = f;
427 }
428 }
429
430 return found;
431 }
432
433 static int socket_verify(Socket *s) {
434 assert(s);
435
436 if (UNIT(s)->load_state != UNIT_LOADED)
437 return 0;
438
439 if (!s->ports) {
440 log_unit_error(UNIT(s), "Unit has no Listen setting (ListenStream=, ListenDatagram=, ListenFIFO=, ...). Refusing.");
441 return -ENOEXEC;
442 }
443
444 if (s->accept && have_non_accept_socket(s)) {
445 log_unit_error(UNIT(s), "Unit configured for accepting sockets, but sockets are non-accepting. Refusing.");
446 return -ENOEXEC;
447 }
448
449 if (s->accept && s->max_connections <= 0) {
450 log_unit_error(UNIT(s), "MaxConnection= setting too small. Refusing.");
451 return -ENOEXEC;
452 }
453
454 if (s->accept && UNIT_DEREF(s->service)) {
455 log_unit_error(UNIT(s), "Explicit service configuration for accepting socket units not supported. Refusing.");
456 return -ENOEXEC;
457 }
458
459 if (s->exec_context.pam_name && s->kill_context.kill_mode != KILL_CONTROL_GROUP) {
460 log_unit_error(UNIT(s), "Unit has PAM enabled. Kill mode must be set to 'control-group'. Refusing.");
461 return -ENOEXEC;
462 }
463
464 if (!strv_isempty(s->symlinks) && !socket_find_symlink_target(s)) {
465 log_unit_error(UNIT(s), "Unit has symlinks set but none or more than one node in the file system. Refusing.");
466 return -ENOEXEC;
467 }
468
469 return 0;
470 }
471
472 static void peer_address_hash_func(const SocketPeer *s, struct siphash *state) {
473 assert(s);
474
475 if (s->peer.sa.sa_family == AF_INET)
476 siphash24_compress(&s->peer.in.sin_addr, sizeof(s->peer.in.sin_addr), state);
477 else if (s->peer.sa.sa_family == AF_INET6)
478 siphash24_compress(&s->peer.in6.sin6_addr, sizeof(s->peer.in6.sin6_addr), state);
479 else if (s->peer.sa.sa_family == AF_VSOCK)
480 siphash24_compress(&s->peer.vm.svm_cid, sizeof(s->peer.vm.svm_cid), state);
481 else
482 assert_not_reached("Unknown address family.");
483 }
484
485 static int peer_address_compare_func(const SocketPeer *x, const SocketPeer *y) {
486 int r;
487
488 r = CMP(x->peer.sa.sa_family, y->peer.sa.sa_family);
489 if (r != 0)
490 return r;
491
492 switch(x->peer.sa.sa_family) {
493 case AF_INET:
494 return memcmp(&x->peer.in.sin_addr, &y->peer.in.sin_addr, sizeof(x->peer.in.sin_addr));
495 case AF_INET6:
496 return memcmp(&x->peer.in6.sin6_addr, &y->peer.in6.sin6_addr, sizeof(x->peer.in6.sin6_addr));
497 case AF_VSOCK:
498 return CMP(x->peer.vm.svm_cid, y->peer.vm.svm_cid);
499 }
500 assert_not_reached("Black sheep in the family!");
501 }
502
503 DEFINE_PRIVATE_HASH_OPS(peer_address_hash_ops, SocketPeer, peer_address_hash_func, peer_address_compare_func);
504
505 static int socket_load(Unit *u) {
506 Socket *s = SOCKET(u);
507 int r;
508
509 assert(u);
510 assert(u->load_state == UNIT_STUB);
511
512 r = set_ensure_allocated(&s->peers_by_address, &peer_address_hash_ops);
513 if (r < 0)
514 return r;
515
516 r = unit_load_fragment_and_dropin(u);
517 if (r < 0)
518 return r;
519
520 if (u->load_state == UNIT_LOADED) {
521 /* This is a new unit? Then let's add in some extras */
522 r = socket_add_extras(s);
523 if (r < 0)
524 return r;
525 }
526
527 return socket_verify(s);
528 }
529
530 static SocketPeer *socket_peer_new(void) {
531 SocketPeer *p;
532
533 p = new0(SocketPeer, 1);
534 if (!p)
535 return NULL;
536
537 p->n_ref = 1;
538
539 return p;
540 }
541
542 static SocketPeer *socket_peer_free(SocketPeer *p) {
543 assert(p);
544
545 if (p->socket)
546 set_remove(p->socket->peers_by_address, p);
547
548 return mfree(p);
549 }
550
551 DEFINE_TRIVIAL_REF_UNREF_FUNC(SocketPeer, socket_peer, socket_peer_free);
552
553 int socket_acquire_peer(Socket *s, int fd, SocketPeer **p) {
554 _cleanup_(socket_peer_unrefp) SocketPeer *remote = NULL;
555 SocketPeer sa = {}, *i;
556 socklen_t salen = sizeof(sa.peer);
557 int r;
558
559 assert(fd >= 0);
560 assert(s);
561
562 r = getpeername(fd, &sa.peer.sa, &salen);
563 if (r < 0)
564 return log_unit_error_errno(UNIT(s), errno, "getpeername failed: %m");
565
566 if (!IN_SET(sa.peer.sa.sa_family, AF_INET, AF_INET6, AF_VSOCK)) {
567 *p = NULL;
568 return 0;
569 }
570
571 i = set_get(s->peers_by_address, &sa);
572 if (i) {
573 *p = socket_peer_ref(i);
574 return 1;
575 }
576
577 remote = socket_peer_new();
578 if (!remote)
579 return log_oom();
580
581 remote->peer = sa.peer;
582 remote->peer_salen = salen;
583
584 r = set_put(s->peers_by_address, remote);
585 if (r < 0)
586 return r;
587
588 remote->socket = s;
589
590 *p = TAKE_PTR(remote);
591
592 return 1;
593 }
594
595 _const_ static const char* listen_lookup(int family, int type) {
596
597 if (family == AF_NETLINK)
598 return "ListenNetlink";
599
600 if (type == SOCK_STREAM)
601 return "ListenStream";
602 else if (type == SOCK_DGRAM)
603 return "ListenDatagram";
604 else if (type == SOCK_SEQPACKET)
605 return "ListenSequentialPacket";
606
607 assert_not_reached("Unknown socket type");
608 return NULL;
609 }
610
611 static void socket_dump(Unit *u, FILE *f, const char *prefix) {
612 char time_string[FORMAT_TIMESPAN_MAX];
613 SocketExecCommand c;
614 Socket *s = SOCKET(u);
615 SocketPort *p;
616 const char *prefix2, *str;
617
618 assert(s);
619 assert(f);
620
621 prefix = strempty(prefix);
622 prefix2 = strjoina(prefix, "\t");
623
624 fprintf(f,
625 "%sSocket State: %s\n"
626 "%sResult: %s\n"
627 "%sBindIPv6Only: %s\n"
628 "%sBacklog: %u\n"
629 "%sSocketMode: %04o\n"
630 "%sDirectoryMode: %04o\n"
631 "%sKeepAlive: %s\n"
632 "%sNoDelay: %s\n"
633 "%sFreeBind: %s\n"
634 "%sTransparent: %s\n"
635 "%sBroadcast: %s\n"
636 "%sPassCredentials: %s\n"
637 "%sPassSecurity: %s\n"
638 "%sTCPCongestion: %s\n"
639 "%sRemoveOnStop: %s\n"
640 "%sWritable: %s\n"
641 "%sFileDescriptorName: %s\n"
642 "%sSELinuxContextFromNet: %s\n",
643 prefix, socket_state_to_string(s->state),
644 prefix, socket_result_to_string(s->result),
645 prefix, socket_address_bind_ipv6_only_to_string(s->bind_ipv6_only),
646 prefix, s->backlog,
647 prefix, s->socket_mode,
648 prefix, s->directory_mode,
649 prefix, yes_no(s->keep_alive),
650 prefix, yes_no(s->no_delay),
651 prefix, yes_no(s->free_bind),
652 prefix, yes_no(s->transparent),
653 prefix, yes_no(s->broadcast),
654 prefix, yes_no(s->pass_cred),
655 prefix, yes_no(s->pass_sec),
656 prefix, strna(s->tcp_congestion),
657 prefix, yes_no(s->remove_on_stop),
658 prefix, yes_no(s->writable),
659 prefix, socket_fdname(s),
660 prefix, yes_no(s->selinux_context_from_net));
661
662 if (s->control_pid > 0)
663 fprintf(f,
664 "%sControl PID: "PID_FMT"\n",
665 prefix, s->control_pid);
666
667 if (s->bind_to_device)
668 fprintf(f,
669 "%sBindToDevice: %s\n",
670 prefix, s->bind_to_device);
671
672 if (s->accept)
673 fprintf(f,
674 "%sAccepted: %u\n"
675 "%sNConnections: %u\n"
676 "%sMaxConnections: %u\n"
677 "%sMaxConnectionsPerSource: %u\n",
678 prefix, s->n_accepted,
679 prefix, s->n_connections,
680 prefix, s->max_connections,
681 prefix, s->max_connections_per_source);
682
683 if (s->priority >= 0)
684 fprintf(f,
685 "%sPriority: %i\n",
686 prefix, s->priority);
687
688 if (s->receive_buffer > 0)
689 fprintf(f,
690 "%sReceiveBuffer: %zu\n",
691 prefix, s->receive_buffer);
692
693 if (s->send_buffer > 0)
694 fprintf(f,
695 "%sSendBuffer: %zu\n",
696 prefix, s->send_buffer);
697
698 if (s->ip_tos >= 0)
699 fprintf(f,
700 "%sIPTOS: %i\n",
701 prefix, s->ip_tos);
702
703 if (s->ip_ttl >= 0)
704 fprintf(f,
705 "%sIPTTL: %i\n",
706 prefix, s->ip_ttl);
707
708 if (s->pipe_size > 0)
709 fprintf(f,
710 "%sPipeSize: %zu\n",
711 prefix, s->pipe_size);
712
713 if (s->mark >= 0)
714 fprintf(f,
715 "%sMark: %i\n",
716 prefix, s->mark);
717
718 if (s->mq_maxmsg > 0)
719 fprintf(f,
720 "%sMessageQueueMaxMessages: %li\n",
721 prefix, s->mq_maxmsg);
722
723 if (s->mq_msgsize > 0)
724 fprintf(f,
725 "%sMessageQueueMessageSize: %li\n",
726 prefix, s->mq_msgsize);
727
728 if (s->reuse_port)
729 fprintf(f,
730 "%sReusePort: %s\n",
731 prefix, yes_no(s->reuse_port));
732
733 if (s->smack)
734 fprintf(f,
735 "%sSmackLabel: %s\n",
736 prefix, s->smack);
737
738 if (s->smack_ip_in)
739 fprintf(f,
740 "%sSmackLabelIPIn: %s\n",
741 prefix, s->smack_ip_in);
742
743 if (s->smack_ip_out)
744 fprintf(f,
745 "%sSmackLabelIPOut: %s\n",
746 prefix, s->smack_ip_out);
747
748 if (!isempty(s->user) || !isempty(s->group))
749 fprintf(f,
750 "%sSocketUser: %s\n"
751 "%sSocketGroup: %s\n",
752 prefix, strna(s->user),
753 prefix, strna(s->group));
754
755 if (s->keep_alive_time > 0)
756 fprintf(f,
757 "%sKeepAliveTimeSec: %s\n",
758 prefix, format_timespan(time_string, FORMAT_TIMESPAN_MAX, s->keep_alive_time, USEC_PER_SEC));
759
760 if (s->keep_alive_interval > 0)
761 fprintf(f,
762 "%sKeepAliveIntervalSec: %s\n",
763 prefix, format_timespan(time_string, FORMAT_TIMESPAN_MAX, s->keep_alive_interval, USEC_PER_SEC));
764
765 if (s->keep_alive_cnt > 0)
766 fprintf(f,
767 "%sKeepAliveProbes: %u\n",
768 prefix, s->keep_alive_cnt);
769
770 if (s->defer_accept > 0)
771 fprintf(f,
772 "%sDeferAcceptSec: %s\n",
773 prefix, format_timespan(time_string, FORMAT_TIMESPAN_MAX, s->defer_accept, USEC_PER_SEC));
774
775 LIST_FOREACH(port, p, s->ports) {
776
777 switch (p->type) {
778 case SOCKET_SOCKET: {
779 _cleanup_free_ char *k = NULL;
780 const char *t;
781 int r;
782
783 r = socket_address_print(&p->address, &k);
784 if (r < 0)
785 t = strerror(-r);
786 else
787 t = k;
788
789 fprintf(f, "%s%s: %s\n", prefix, listen_lookup(socket_address_family(&p->address), p->address.type), t);
790 break;
791 }
792 case SOCKET_SPECIAL:
793 fprintf(f, "%sListenSpecial: %s\n", prefix, p->path);
794 break;
795 case SOCKET_USB_FUNCTION:
796 fprintf(f, "%sListenUSBFunction: %s\n", prefix, p->path);
797 break;
798 case SOCKET_MQUEUE:
799 fprintf(f, "%sListenMessageQueue: %s\n", prefix, p->path);
800 break;
801 default:
802 fprintf(f, "%sListenFIFO: %s\n", prefix, p->path);
803 }
804 }
805
806 fprintf(f,
807 "%sTriggerLimitIntervalSec: %s\n"
808 "%sTriggerLimitBurst: %u\n",
809 prefix, format_timespan(time_string, FORMAT_TIMESPAN_MAX, s->trigger_limit.interval, USEC_PER_SEC),
810 prefix, s->trigger_limit.burst);
811
812 str = ip_protocol_to_name(s->socket_protocol);
813 if (str)
814 fprintf(f, "%sSocketProtocol: %s\n", prefix, str);
815
816 if (!strv_isempty(s->symlinks)) {
817 char **q;
818
819 fprintf(f, "%sSymlinks:", prefix);
820 STRV_FOREACH(q, s->symlinks)
821 fprintf(f, " %s", *q);
822
823 fprintf(f, "\n");
824 }
825
826 fprintf(f,
827 "%sTimeoutSec: %s\n",
828 prefix, format_timespan(time_string, FORMAT_TIMESPAN_MAX, s->timeout_usec, USEC_PER_SEC));
829
830 exec_context_dump(&s->exec_context, f, prefix);
831 kill_context_dump(&s->kill_context, f, prefix);
832
833 for (c = 0; c < _SOCKET_EXEC_COMMAND_MAX; c++) {
834 if (!s->exec_command[c])
835 continue;
836
837 fprintf(f, "%s-> %s:\n",
838 prefix, socket_exec_command_to_string(c));
839
840 exec_command_dump_list(s->exec_command[c], f, prefix2);
841 }
842
843 cgroup_context_dump(&s->cgroup_context, f, prefix);
844 }
845
846 static int instance_from_socket(int fd, unsigned nr, char **instance) {
847 socklen_t l;
848 char *r;
849 union sockaddr_union local, remote;
850
851 assert(fd >= 0);
852 assert(instance);
853
854 l = sizeof(local);
855 if (getsockname(fd, &local.sa, &l) < 0)
856 return -errno;
857
858 l = sizeof(remote);
859 if (getpeername(fd, &remote.sa, &l) < 0)
860 return -errno;
861
862 switch (local.sa.sa_family) {
863
864 case AF_INET: {
865 uint32_t
866 a = be32toh(local.in.sin_addr.s_addr),
867 b = be32toh(remote.in.sin_addr.s_addr);
868
869 if (asprintf(&r,
870 "%u-%u.%u.%u.%u:%u-%u.%u.%u.%u:%u",
871 nr,
872 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF,
873 be16toh(local.in.sin_port),
874 b >> 24, (b >> 16) & 0xFF, (b >> 8) & 0xFF, b & 0xFF,
875 be16toh(remote.in.sin_port)) < 0)
876 return -ENOMEM;
877
878 break;
879 }
880
881 case AF_INET6: {
882 static const unsigned char ipv4_prefix[] = {
883 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF
884 };
885
886 if (memcmp(&local.in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0 &&
887 memcmp(&remote.in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) {
888 const uint8_t
889 *a = local.in6.sin6_addr.s6_addr+12,
890 *b = remote.in6.sin6_addr.s6_addr+12;
891
892 if (asprintf(&r,
893 "%u-%u.%u.%u.%u:%u-%u.%u.%u.%u:%u",
894 nr,
895 a[0], a[1], a[2], a[3],
896 be16toh(local.in6.sin6_port),
897 b[0], b[1], b[2], b[3],
898 be16toh(remote.in6.sin6_port)) < 0)
899 return -ENOMEM;
900 } else {
901 char a[INET6_ADDRSTRLEN], b[INET6_ADDRSTRLEN];
902
903 if (asprintf(&r,
904 "%u-%s:%u-%s:%u",
905 nr,
906 inet_ntop(AF_INET6, &local.in6.sin6_addr, a, sizeof(a)),
907 be16toh(local.in6.sin6_port),
908 inet_ntop(AF_INET6, &remote.in6.sin6_addr, b, sizeof(b)),
909 be16toh(remote.in6.sin6_port)) < 0)
910 return -ENOMEM;
911 }
912
913 break;
914 }
915
916 case AF_UNIX: {
917 struct ucred ucred;
918 int k;
919
920 k = getpeercred(fd, &ucred);
921 if (k >= 0) {
922 if (asprintf(&r,
923 "%u-"PID_FMT"-"UID_FMT,
924 nr, ucred.pid, ucred.uid) < 0)
925 return -ENOMEM;
926 } else if (k == -ENODATA) {
927 /* This handles the case where somebody is
928 * connecting from another pid/uid namespace
929 * (e.g. from outside of our container). */
930 if (asprintf(&r,
931 "%u-unknown",
932 nr) < 0)
933 return -ENOMEM;
934 } else
935 return k;
936
937 break;
938 }
939
940 case AF_VSOCK:
941 if (asprintf(&r,
942 "%u-%u:%u-%u:%u",
943 nr,
944 local.vm.svm_cid, local.vm.svm_port,
945 remote.vm.svm_cid, remote.vm.svm_port) < 0)
946 return -ENOMEM;
947
948 break;
949
950 default:
951 assert_not_reached("Unhandled socket type.");
952 }
953
954 *instance = r;
955 return 0;
956 }
957
958 static void socket_close_fds(Socket *s) {
959 SocketPort *p;
960 char **i;
961
962 assert(s);
963
964 LIST_FOREACH(port, p, s->ports) {
965 bool was_open;
966
967 was_open = p->fd >= 0;
968
969 p->event_source = sd_event_source_unref(p->event_source);
970 p->fd = safe_close(p->fd);
971 socket_cleanup_fd_list(p);
972
973 /* One little note: we should normally not delete any sockets in the file system here! After all some
974 * other process we spawned might still have a reference of this fd and wants to continue to use
975 * it. Therefore we normally delete sockets in the file system before we create a new one, not after we
976 * stopped using one! That all said, if the user explicitly requested this, we'll delete them here
977 * anyway, but only then. */
978
979 if (!was_open || !s->remove_on_stop)
980 continue;
981
982 switch (p->type) {
983
984 case SOCKET_FIFO:
985 (void) unlink(p->path);
986 break;
987
988 case SOCKET_MQUEUE:
989 (void) mq_unlink(p->path);
990 break;
991
992 case SOCKET_SOCKET:
993 (void) socket_address_unlink(&p->address);
994 break;
995
996 default:
997 break;
998 }
999 }
1000
1001 if (s->remove_on_stop)
1002 STRV_FOREACH(i, s->symlinks)
1003 (void) unlink(*i);
1004 }
1005
1006 static void socket_apply_socket_options(Socket *s, int fd) {
1007 int r;
1008
1009 assert(s);
1010 assert(fd >= 0);
1011
1012 if (s->keep_alive) {
1013 r = setsockopt_int(fd, SOL_SOCKET, SO_KEEPALIVE, true);
1014 if (r < 0)
1015 log_unit_warning_errno(UNIT(s), r, "SO_KEEPALIVE failed: %m");
1016 }
1017
1018 if (s->keep_alive_time > 0) {
1019 r = setsockopt_int(fd, SOL_TCP, TCP_KEEPIDLE, s->keep_alive_time / USEC_PER_SEC);
1020 if (r < 0)
1021 log_unit_warning_errno(UNIT(s), r, "TCP_KEEPIDLE failed: %m");
1022 }
1023
1024 if (s->keep_alive_interval > 0) {
1025 r = setsockopt_int(fd, SOL_TCP, TCP_KEEPINTVL, s->keep_alive_interval / USEC_PER_SEC);
1026 if (r < 0)
1027 log_unit_warning_errno(UNIT(s), r, "TCP_KEEPINTVL failed: %m");
1028 }
1029
1030 if (s->keep_alive_cnt > 0) {
1031 r = setsockopt_int(fd, SOL_TCP, TCP_KEEPCNT, s->keep_alive_cnt);
1032 if (r < 0)
1033 log_unit_warning_errno(UNIT(s), r, "TCP_KEEPCNT failed: %m");
1034 }
1035
1036 if (s->defer_accept > 0) {
1037 r = setsockopt_int(fd, SOL_TCP, TCP_DEFER_ACCEPT, s->defer_accept / USEC_PER_SEC);
1038 if (r < 0)
1039 log_unit_warning_errno(UNIT(s), r, "TCP_DEFER_ACCEPT failed: %m");
1040 }
1041
1042 if (s->no_delay) {
1043 if (s->socket_protocol == IPPROTO_SCTP) {
1044 r = setsockopt_int(fd, SOL_SCTP, SCTP_NODELAY, true);
1045 if (r < 0)
1046 log_unit_warning_errno(UNIT(s), r, "SCTP_NODELAY failed: %m");
1047 } else {
1048 r = setsockopt_int(fd, SOL_TCP, TCP_NODELAY, true);
1049 if (r < 0)
1050 log_unit_warning_errno(UNIT(s), r, "TCP_NODELAY failed: %m");
1051 }
1052 }
1053
1054 if (s->broadcast) {
1055 r = setsockopt_int(fd, SOL_SOCKET, SO_BROADCAST, true);
1056 if (r < 0)
1057 log_unit_warning_errno(UNIT(s), r, "SO_BROADCAST failed: %m");
1058 }
1059
1060 if (s->pass_cred) {
1061 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
1062 if (r < 0)
1063 log_unit_warning_errno(UNIT(s), r, "SO_PASSCRED failed: %m");
1064 }
1065
1066 if (s->pass_sec) {
1067 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSSEC, true);
1068 if (r < 0)
1069 log_unit_warning_errno(UNIT(s), r, "SO_PASSSEC failed: %m");
1070 }
1071
1072 if (s->priority >= 0) {
1073 r = setsockopt_int(fd, SOL_SOCKET, SO_PRIORITY, s->priority);
1074 if (r < 0)
1075 log_unit_warning_errno(UNIT(s), r, "SO_PRIORITY failed: %m");
1076 }
1077
1078 if (s->receive_buffer > 0) {
1079 /* We first try with SO_RCVBUFFORCE, in case we have the perms for that */
1080 if (setsockopt_int(fd, SOL_SOCKET, SO_RCVBUFFORCE, s->receive_buffer) < 0) {
1081 r = setsockopt_int(fd, SOL_SOCKET, SO_RCVBUF, s->receive_buffer);
1082 if (r < 0)
1083 log_unit_warning_errno(UNIT(s), r, "SO_RCVBUF failed: %m");
1084 }
1085 }
1086
1087 if (s->send_buffer > 0) {
1088 if (setsockopt_int(fd, SOL_SOCKET, SO_SNDBUFFORCE, s->send_buffer) < 0) {
1089 r = setsockopt_int(fd, SOL_SOCKET, SO_SNDBUF, s->send_buffer);
1090 if (r < 0)
1091 log_unit_warning_errno(UNIT(s), r, "SO_SNDBUF failed: %m");
1092 }
1093 }
1094
1095 if (s->mark >= 0) {
1096 r = setsockopt_int(fd, SOL_SOCKET, SO_MARK, s->mark);
1097 if (r < 0)
1098 log_unit_warning_errno(UNIT(s), r, "SO_MARK failed: %m");
1099 }
1100
1101 if (s->ip_tos >= 0) {
1102 r = setsockopt_int(fd, IPPROTO_IP, IP_TOS, s->ip_tos);
1103 if (r < 0)
1104 log_unit_warning_errno(UNIT(s), r, "IP_TOS failed: %m");
1105 }
1106
1107 if (s->ip_ttl >= 0) {
1108 int x;
1109
1110 r = setsockopt_int(fd, IPPROTO_IP, IP_TTL, s->ip_ttl);
1111
1112 if (socket_ipv6_is_supported())
1113 x = setsockopt_int(fd, IPPROTO_IPV6, IPV6_UNICAST_HOPS, s->ip_ttl);
1114 else
1115 x = -EAFNOSUPPORT;
1116
1117 if (r < 0 && x < 0)
1118 log_unit_warning_errno(UNIT(s), r, "IP_TTL/IPV6_UNICAST_HOPS failed: %m");
1119 }
1120
1121 if (s->tcp_congestion)
1122 if (setsockopt(fd, SOL_TCP, TCP_CONGESTION, s->tcp_congestion, strlen(s->tcp_congestion)+1) < 0)
1123 log_unit_warning_errno(UNIT(s), errno, "TCP_CONGESTION failed: %m");
1124
1125 if (s->smack_ip_in) {
1126 r = mac_smack_apply_fd(fd, SMACK_ATTR_IPIN, s->smack_ip_in);
1127 if (r < 0)
1128 log_unit_error_errno(UNIT(s), r, "mac_smack_apply_ip_in_fd: %m");
1129 }
1130
1131 if (s->smack_ip_out) {
1132 r = mac_smack_apply_fd(fd, SMACK_ATTR_IPOUT, s->smack_ip_out);
1133 if (r < 0)
1134 log_unit_error_errno(UNIT(s), r, "mac_smack_apply_ip_out_fd: %m");
1135 }
1136 }
1137
1138 static void socket_apply_fifo_options(Socket *s, int fd) {
1139 int r;
1140
1141 assert(s);
1142 assert(fd >= 0);
1143
1144 if (s->pipe_size > 0)
1145 if (fcntl(fd, F_SETPIPE_SZ, s->pipe_size) < 0)
1146 log_unit_warning_errno(UNIT(s), errno, "Setting pipe size failed, ignoring: %m");
1147
1148 if (s->smack) {
1149 r = mac_smack_apply_fd(fd, SMACK_ATTR_ACCESS, s->smack);
1150 if (r < 0)
1151 log_unit_error_errno(UNIT(s), r, "SMACK relabelling failed, ignoring: %m");
1152 }
1153 }
1154
1155 static int fifo_address_create(
1156 const char *path,
1157 mode_t directory_mode,
1158 mode_t socket_mode) {
1159
1160 _cleanup_close_ int fd = -1;
1161 mode_t old_mask;
1162 struct stat st;
1163 int r;
1164
1165 assert(path);
1166
1167 (void) mkdir_parents_label(path, directory_mode);
1168
1169 r = mac_selinux_create_file_prepare(path, S_IFIFO);
1170 if (r < 0)
1171 return r;
1172
1173 /* Enforce the right access mode for the fifo */
1174 old_mask = umask(~socket_mode);
1175
1176 /* Include the original umask in our mask */
1177 (void) umask(~socket_mode | old_mask);
1178
1179 r = mkfifo(path, socket_mode);
1180 (void) umask(old_mask);
1181
1182 if (r < 0 && errno != EEXIST) {
1183 r = -errno;
1184 goto fail;
1185 }
1186
1187 fd = open(path, O_RDWR | O_CLOEXEC | O_NOCTTY | O_NONBLOCK | O_NOFOLLOW);
1188 if (fd < 0) {
1189 r = -errno;
1190 goto fail;
1191 }
1192
1193 mac_selinux_create_file_clear();
1194
1195 if (fstat(fd, &st) < 0) {
1196 r = -errno;
1197 goto fail;
1198 }
1199
1200 if (!S_ISFIFO(st.st_mode) ||
1201 (st.st_mode & 0777) != (socket_mode & ~old_mask) ||
1202 st.st_uid != getuid() ||
1203 st.st_gid != getgid()) {
1204 r = -EEXIST;
1205 goto fail;
1206 }
1207
1208 return TAKE_FD(fd);
1209
1210 fail:
1211 mac_selinux_create_file_clear();
1212 return r;
1213 }
1214
1215 static int special_address_create(const char *path, bool writable) {
1216 _cleanup_close_ int fd = -1;
1217 struct stat st;
1218
1219 assert(path);
1220
1221 fd = open(path, (writable ? O_RDWR : O_RDONLY)|O_CLOEXEC|O_NOCTTY|O_NONBLOCK|O_NOFOLLOW);
1222 if (fd < 0)
1223 return -errno;
1224
1225 if (fstat(fd, &st) < 0)
1226 return -errno;
1227
1228 /* Check whether this is a /proc, /sys or /dev file or char device */
1229 if (!S_ISREG(st.st_mode) && !S_ISCHR(st.st_mode))
1230 return -EEXIST;
1231
1232 return TAKE_FD(fd);
1233 }
1234
1235 static int usbffs_address_create(const char *path) {
1236 _cleanup_close_ int fd = -1;
1237 struct stat st;
1238
1239 assert(path);
1240
1241 fd = open(path, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK|O_NOFOLLOW);
1242 if (fd < 0)
1243 return -errno;
1244
1245 if (fstat(fd, &st) < 0)
1246 return -errno;
1247
1248 /* Check whether this is a regular file (ffs endpoint) */
1249 if (!S_ISREG(st.st_mode))
1250 return -EEXIST;
1251
1252 return TAKE_FD(fd);
1253 }
1254
1255 static int mq_address_create(
1256 const char *path,
1257 mode_t mq_mode,
1258 long maxmsg,
1259 long msgsize) {
1260
1261 _cleanup_close_ int fd = -1;
1262 struct stat st;
1263 mode_t old_mask;
1264 struct mq_attr _attr, *attr = NULL;
1265
1266 assert(path);
1267
1268 if (maxmsg > 0 && msgsize > 0) {
1269 _attr = (struct mq_attr) {
1270 .mq_flags = O_NONBLOCK,
1271 .mq_maxmsg = maxmsg,
1272 .mq_msgsize = msgsize,
1273 };
1274 attr = &_attr;
1275 }
1276
1277 /* Enforce the right access mode for the mq */
1278 old_mask = umask(~mq_mode);
1279
1280 /* Include the original umask in our mask */
1281 (void) umask(~mq_mode | old_mask);
1282 fd = mq_open(path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_CREAT, mq_mode, attr);
1283 (void) umask(old_mask);
1284
1285 if (fd < 0)
1286 return -errno;
1287
1288 if (fstat(fd, &st) < 0)
1289 return -errno;
1290
1291 if ((st.st_mode & 0777) != (mq_mode & ~old_mask) ||
1292 st.st_uid != getuid() ||
1293 st.st_gid != getgid())
1294 return -EEXIST;
1295
1296 return TAKE_FD(fd);
1297 }
1298
1299 static int socket_symlink(Socket *s) {
1300 const char *p;
1301 char **i;
1302 int r;
1303
1304 assert(s);
1305
1306 p = socket_find_symlink_target(s);
1307 if (!p)
1308 return 0;
1309
1310 STRV_FOREACH(i, s->symlinks) {
1311 (void) mkdir_parents_label(*i, s->directory_mode);
1312
1313 r = symlink_idempotent(p, *i, false);
1314
1315 if (r == -EEXIST && s->remove_on_stop) {
1316 /* If there's already something where we want to create the symlink, and the destructive
1317 * RemoveOnStop= mode is set, then we might as well try to remove what already exists and try
1318 * again. */
1319
1320 if (unlink(*i) >= 0)
1321 r = symlink_idempotent(p, *i, false);
1322 }
1323
1324 if (r < 0)
1325 log_unit_warning_errno(UNIT(s), r, "Failed to create symlink %s → %s, ignoring: %m", p, *i);
1326 }
1327
1328 return 0;
1329 }
1330
1331 static int usbffs_write_descs(int fd, Service *s) {
1332 int r;
1333
1334 if (!s->usb_function_descriptors || !s->usb_function_strings)
1335 return -EINVAL;
1336
1337 r = copy_file_fd(s->usb_function_descriptors, fd, 0);
1338 if (r < 0)
1339 return r;
1340
1341 return copy_file_fd(s->usb_function_strings, fd, 0);
1342 }
1343
1344 static int usbffs_select_ep(const struct dirent *d) {
1345 return d->d_name[0] != '.' && !streq(d->d_name, "ep0");
1346 }
1347
1348 static int usbffs_dispatch_eps(SocketPort *p) {
1349 _cleanup_free_ struct dirent **ent = NULL;
1350 size_t n, k, i;
1351 int r;
1352
1353 r = scandir(p->path, &ent, usbffs_select_ep, alphasort);
1354 if (r < 0)
1355 return -errno;
1356
1357 n = (size_t) r;
1358 p->auxiliary_fds = new(int, n);
1359 if (!p->auxiliary_fds) {
1360 r = -ENOMEM;
1361 goto clear;
1362 }
1363
1364 p->n_auxiliary_fds = n;
1365
1366 k = 0;
1367 for (i = 0; i < n; ++i) {
1368 _cleanup_free_ char *ep = NULL;
1369
1370 ep = path_make_absolute(ent[i]->d_name, p->path);
1371 if (!ep) {
1372 r = -ENOMEM;
1373 goto fail;
1374 }
1375
1376 path_simplify(ep, false);
1377
1378 r = usbffs_address_create(ep);
1379 if (r < 0)
1380 goto fail;
1381
1382 p->auxiliary_fds[k++] = r;
1383 }
1384
1385 r = 0;
1386 goto clear;
1387
1388 fail:
1389 close_many(p->auxiliary_fds, k);
1390 p->auxiliary_fds = mfree(p->auxiliary_fds);
1391 p->n_auxiliary_fds = 0;
1392
1393 clear:
1394 for (i = 0; i < n; ++i)
1395 free(ent[i]);
1396
1397 return r;
1398 }
1399
1400 static int socket_determine_selinux_label(Socket *s, char **ret) {
1401 Service *service;
1402 ExecCommand *c;
1403 _cleanup_free_ char *path = NULL;
1404 int r;
1405
1406 assert(s);
1407 assert(ret);
1408
1409 if (s->selinux_context_from_net) {
1410 /* If this is requested, get label from the network label */
1411
1412 r = mac_selinux_get_our_label(ret);
1413 if (r == -EOPNOTSUPP)
1414 goto no_label;
1415
1416 } else {
1417 /* Otherwise, get it from the executable we are about to start */
1418 r = socket_instantiate_service(s);
1419 if (r < 0)
1420 return r;
1421
1422 if (!UNIT_ISSET(s->service))
1423 goto no_label;
1424
1425 service = SERVICE(UNIT_DEREF(s->service));
1426 c = service->exec_command[SERVICE_EXEC_START];
1427 if (!c)
1428 goto no_label;
1429
1430 r = chase_symlinks(c->path, service->exec_context.root_directory, CHASE_PREFIX_ROOT, &path);
1431 if (r < 0)
1432 goto no_label;
1433
1434 r = mac_selinux_get_create_label_from_exe(path, ret);
1435 if (IN_SET(r, -EPERM, -EOPNOTSUPP))
1436 goto no_label;
1437 }
1438
1439 return r;
1440
1441 no_label:
1442 *ret = NULL;
1443 return 0;
1444 }
1445
1446 static int socket_address_listen_do(
1447 Socket *s,
1448 const SocketAddress *address,
1449 const char *label) {
1450
1451 assert(s);
1452 assert(address);
1453
1454 return socket_address_listen(
1455 address,
1456 SOCK_CLOEXEC|SOCK_NONBLOCK,
1457 s->backlog,
1458 s->bind_ipv6_only,
1459 s->bind_to_device,
1460 s->reuse_port,
1461 s->free_bind,
1462 s->transparent,
1463 s->directory_mode,
1464 s->socket_mode,
1465 label);
1466 }
1467
1468 #define log_address_error_errno(u, address, error, fmt) \
1469 ({ \
1470 _cleanup_free_ char *_t = NULL; \
1471 \
1472 (void) socket_address_print(address, &_t); \
1473 log_unit_error_errno(u, error, fmt, strna(_t)); \
1474 })
1475
1476 static int fork_needed(const SocketAddress *address, const ExecContext *context) {
1477 int r;
1478
1479 assert(address);
1480 assert(context);
1481
1482 /* Check if we need to do the cgroup or netns stuff. If not we can do things much simpler. */
1483
1484 if (IN_SET(address->sockaddr.sa.sa_family, AF_INET, AF_INET6)) {
1485 r = bpf_firewall_supported();
1486 if (r < 0)
1487 return r;
1488 if (r != BPF_FIREWALL_UNSUPPORTED) /* If BPF firewalling isn't supported anyway — there's no point in this forking complexity */
1489 return true;
1490 }
1491
1492 return context->private_network || context->network_namespace_path;
1493 }
1494
1495 static int socket_address_listen_in_cgroup(
1496 Socket *s,
1497 const SocketAddress *address,
1498 const char *label) {
1499
1500 _cleanup_close_pair_ int pair[2] = { -1, -1 };
1501 int fd, r;
1502 pid_t pid;
1503
1504 assert(s);
1505 assert(address);
1506
1507 /* This is a wrapper around socket_address_listen(), that forks off a helper process inside the
1508 * socket's cgroup and network namespace in which the socket is actually created. This way we ensure
1509 * the socket is actually properly attached to the unit's cgroup for the purpose of BPF filtering and
1510 * such. */
1511
1512 r = fork_needed(address, &s->exec_context);
1513 if (r < 0)
1514 return r;
1515 if (r == 0) {
1516 /* Shortcut things... */
1517 fd = socket_address_listen_do(s, address, label);
1518 if (fd < 0)
1519 return log_address_error_errno(UNIT(s), address, fd, "Failed to create listening socket (%s): %m");
1520
1521 return fd;
1522 }
1523
1524 r = unit_setup_exec_runtime(UNIT(s));
1525 if (r < 0)
1526 return log_unit_error_errno(UNIT(s), r, "Failed acquire runtime: %m");
1527
1528 if (s->exec_context.network_namespace_path &&
1529 s->exec_runtime &&
1530 s->exec_runtime->netns_storage_socket[0] >= 0) {
1531 r = open_netns_path(s->exec_runtime->netns_storage_socket, s->exec_context.network_namespace_path);
1532 if (r < 0)
1533 return log_unit_error_errno(UNIT(s), r, "Failed to open network namespace path %s: %m", s->exec_context.network_namespace_path);
1534 }
1535
1536 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pair) < 0)
1537 return log_unit_error_errno(UNIT(s), errno, "Failed to create communication channel: %m");
1538
1539 r = unit_fork_helper_process(UNIT(s), "(sd-listen)", &pid);
1540 if (r < 0)
1541 return log_unit_error_errno(UNIT(s), r, "Failed to fork off listener stub process: %m");
1542 if (r == 0) {
1543 /* Child */
1544
1545 pair[0] = safe_close(pair[0]);
1546
1547 if ((s->exec_context.private_network || s->exec_context.network_namespace_path) &&
1548 s->exec_runtime &&
1549 s->exec_runtime->netns_storage_socket[0] >= 0) {
1550
1551 if (ns_type_supported(NAMESPACE_NET)) {
1552 r = setup_netns(s->exec_runtime->netns_storage_socket);
1553 if (r < 0) {
1554 log_unit_error_errno(UNIT(s), r, "Failed to join network namespace: %m");
1555 _exit(EXIT_NETWORK);
1556 }
1557 } else if (s->exec_context.network_namespace_path) {
1558 log_unit_error(UNIT(s), "Network namespace path configured but network namespaces not supported.");
1559 _exit(EXIT_NETWORK);
1560 } else
1561 log_unit_warning(UNIT(s), "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
1562 }
1563
1564 fd = socket_address_listen_do(s, address, label);
1565 if (fd < 0) {
1566 log_address_error_errno(UNIT(s), address, fd, "Failed to create listening socket (%s): %m");
1567 _exit(EXIT_FAILURE);
1568 }
1569
1570 r = send_one_fd(pair[1], fd, 0);
1571 if (r < 0) {
1572 log_address_error_errno(UNIT(s), address, r, "Failed to send listening socket (%s) to parent: %m");
1573 _exit(EXIT_FAILURE);
1574 }
1575
1576 _exit(EXIT_SUCCESS);
1577 }
1578
1579 pair[1] = safe_close(pair[1]);
1580 fd = receive_one_fd(pair[0], 0);
1581
1582 /* We synchronously wait for the helper, as it shouldn't be slow */
1583 r = wait_for_terminate_and_check("(sd-listen)", pid, WAIT_LOG_ABNORMAL);
1584 if (r < 0) {
1585 safe_close(fd);
1586 return r;
1587 }
1588
1589 if (fd < 0)
1590 return log_address_error_errno(UNIT(s), address, fd, "Failed to receive listening socket (%s): %m");
1591
1592 return fd;
1593 }
1594
1595 DEFINE_TRIVIAL_CLEANUP_FUNC(Socket *, socket_close_fds);
1596
1597 static int socket_open_fds(Socket *_s) {
1598 _cleanup_(socket_close_fdsp) Socket *s = _s;
1599 _cleanup_(mac_selinux_freep) char *label = NULL;
1600 bool know_label = false;
1601 SocketPort *p;
1602 int r;
1603
1604 assert(s);
1605
1606 LIST_FOREACH(port, p, s->ports) {
1607
1608 if (p->fd >= 0)
1609 continue;
1610
1611 switch (p->type) {
1612
1613 case SOCKET_SOCKET:
1614
1615 if (!know_label) {
1616 /* Figure out label, if we don't it know yet. We do it once, for the first socket where
1617 * we need this and remember it for the rest. */
1618
1619 r = socket_determine_selinux_label(s, &label);
1620 if (r < 0)
1621 return log_unit_error_errno(UNIT(s), r, "Failed to determine SELinux label: %m");
1622
1623 know_label = true;
1624 }
1625
1626 /* Apply the socket protocol */
1627 switch (p->address.type) {
1628
1629 case SOCK_STREAM:
1630 case SOCK_SEQPACKET:
1631 if (s->socket_protocol == IPPROTO_SCTP)
1632 p->address.protocol = s->socket_protocol;
1633 break;
1634
1635 case SOCK_DGRAM:
1636 if (s->socket_protocol == IPPROTO_UDPLITE)
1637 p->address.protocol = s->socket_protocol;
1638 break;
1639 }
1640
1641 p->fd = socket_address_listen_in_cgroup(s, &p->address, label);
1642 if (p->fd < 0)
1643 return p->fd;
1644
1645 socket_apply_socket_options(s, p->fd);
1646 socket_symlink(s);
1647 break;
1648
1649 case SOCKET_SPECIAL:
1650
1651 p->fd = special_address_create(p->path, s->writable);
1652 if (p->fd < 0)
1653 return log_unit_error_errno(UNIT(s), p->fd, "Failed to open special file %s: %m", p->path);
1654 break;
1655
1656 case SOCKET_FIFO:
1657
1658 p->fd = fifo_address_create(
1659 p->path,
1660 s->directory_mode,
1661 s->socket_mode);
1662 if (p->fd < 0)
1663 return log_unit_error_errno(UNIT(s), p->fd, "Failed to open FIFO %s: %m", p->path);
1664
1665 socket_apply_fifo_options(s, p->fd);
1666 socket_symlink(s);
1667 break;
1668
1669 case SOCKET_MQUEUE:
1670
1671 p->fd = mq_address_create(
1672 p->path,
1673 s->socket_mode,
1674 s->mq_maxmsg,
1675 s->mq_msgsize);
1676 if (p->fd < 0)
1677 return log_unit_error_errno(UNIT(s), p->fd, "Failed to open message queue %s: %m", p->path);
1678 break;
1679
1680 case SOCKET_USB_FUNCTION: {
1681 _cleanup_free_ char *ep = NULL;
1682
1683 ep = path_make_absolute("ep0", p->path);
1684
1685 p->fd = usbffs_address_create(ep);
1686 if (p->fd < 0)
1687 return p->fd;
1688
1689 r = usbffs_write_descs(p->fd, SERVICE(UNIT_DEREF(s->service)));
1690 if (r < 0)
1691 return r;
1692
1693 r = usbffs_dispatch_eps(p);
1694 if (r < 0)
1695 return r;
1696
1697 break;
1698 }
1699 default:
1700 assert_not_reached("Unknown port type");
1701 }
1702 }
1703
1704 s = NULL;
1705 return 0;
1706 }
1707
1708 static void socket_unwatch_fds(Socket *s) {
1709 SocketPort *p;
1710 int r;
1711
1712 assert(s);
1713
1714 LIST_FOREACH(port, p, s->ports) {
1715 if (p->fd < 0)
1716 continue;
1717
1718 if (!p->event_source)
1719 continue;
1720
1721 r = sd_event_source_set_enabled(p->event_source, SD_EVENT_OFF);
1722 if (r < 0)
1723 log_unit_debug_errno(UNIT(s), r, "Failed to disable event source: %m");
1724 }
1725 }
1726
1727 static int socket_watch_fds(Socket *s) {
1728 SocketPort *p;
1729 int r;
1730
1731 assert(s);
1732
1733 LIST_FOREACH(port, p, s->ports) {
1734 if (p->fd < 0)
1735 continue;
1736
1737 if (p->event_source) {
1738 r = sd_event_source_set_enabled(p->event_source, SD_EVENT_ON);
1739 if (r < 0)
1740 goto fail;
1741 } else {
1742 r = sd_event_add_io(UNIT(s)->manager->event, &p->event_source, p->fd, EPOLLIN, socket_dispatch_io, p);
1743 if (r < 0)
1744 goto fail;
1745
1746 (void) sd_event_source_set_description(p->event_source, "socket-port-io");
1747 }
1748 }
1749
1750 return 0;
1751
1752 fail:
1753 log_unit_warning_errno(UNIT(s), r, "Failed to watch listening fds: %m");
1754 socket_unwatch_fds(s);
1755 return r;
1756 }
1757
1758 enum {
1759 SOCKET_OPEN_NONE,
1760 SOCKET_OPEN_SOME,
1761 SOCKET_OPEN_ALL,
1762 };
1763
1764 static int socket_check_open(Socket *s) {
1765 bool have_open = false, have_closed = false;
1766 SocketPort *p;
1767
1768 assert(s);
1769
1770 LIST_FOREACH(port, p, s->ports) {
1771 if (p->fd < 0)
1772 have_closed = true;
1773 else
1774 have_open = true;
1775
1776 if (have_open && have_closed)
1777 return SOCKET_OPEN_SOME;
1778 }
1779
1780 if (have_open)
1781 return SOCKET_OPEN_ALL;
1782
1783 return SOCKET_OPEN_NONE;
1784 }
1785
1786 static void socket_set_state(Socket *s, SocketState state) {
1787 SocketState old_state;
1788 assert(s);
1789
1790 if (s->state != state)
1791 bus_unit_send_pending_change_signal(UNIT(s), false);
1792
1793 old_state = s->state;
1794 s->state = state;
1795
1796 if (!IN_SET(state,
1797 SOCKET_START_PRE,
1798 SOCKET_START_CHOWN,
1799 SOCKET_START_POST,
1800 SOCKET_STOP_PRE,
1801 SOCKET_STOP_PRE_SIGTERM,
1802 SOCKET_STOP_PRE_SIGKILL,
1803 SOCKET_STOP_POST,
1804 SOCKET_FINAL_SIGTERM,
1805 SOCKET_FINAL_SIGKILL)) {
1806
1807 s->timer_event_source = sd_event_source_unref(s->timer_event_source);
1808 socket_unwatch_control_pid(s);
1809 s->control_command = NULL;
1810 s->control_command_id = _SOCKET_EXEC_COMMAND_INVALID;
1811 }
1812
1813 if (state != SOCKET_LISTENING)
1814 socket_unwatch_fds(s);
1815
1816 if (!IN_SET(state,
1817 SOCKET_START_CHOWN,
1818 SOCKET_START_POST,
1819 SOCKET_LISTENING,
1820 SOCKET_RUNNING,
1821 SOCKET_STOP_PRE,
1822 SOCKET_STOP_PRE_SIGTERM,
1823 SOCKET_STOP_PRE_SIGKILL))
1824 socket_close_fds(s);
1825
1826 if (state != old_state)
1827 log_unit_debug(UNIT(s), "Changed %s -> %s", socket_state_to_string(old_state), socket_state_to_string(state));
1828
1829 unit_notify(UNIT(s), state_translation_table[old_state], state_translation_table[state], 0);
1830 }
1831
1832 static int socket_coldplug(Unit *u) {
1833 Socket *s = SOCKET(u);
1834 int r;
1835
1836 assert(s);
1837 assert(s->state == SOCKET_DEAD);
1838
1839 if (s->deserialized_state == s->state)
1840 return 0;
1841
1842 if (s->control_pid > 0 &&
1843 pid_is_unwaited(s->control_pid) &&
1844 IN_SET(s->deserialized_state,
1845 SOCKET_START_PRE,
1846 SOCKET_START_CHOWN,
1847 SOCKET_START_POST,
1848 SOCKET_STOP_PRE,
1849 SOCKET_STOP_PRE_SIGTERM,
1850 SOCKET_STOP_PRE_SIGKILL,
1851 SOCKET_STOP_POST,
1852 SOCKET_FINAL_SIGTERM,
1853 SOCKET_FINAL_SIGKILL)) {
1854
1855 r = unit_watch_pid(UNIT(s), s->control_pid, false);
1856 if (r < 0)
1857 return r;
1858
1859 r = socket_arm_timer(s, usec_add(u->state_change_timestamp.monotonic, s->timeout_usec));
1860 if (r < 0)
1861 return r;
1862 }
1863
1864 if (IN_SET(s->deserialized_state,
1865 SOCKET_START_CHOWN,
1866 SOCKET_START_POST,
1867 SOCKET_LISTENING,
1868 SOCKET_RUNNING)) {
1869
1870 /* Originally, we used to simply reopen all sockets here that we didn't have file descriptors
1871 * for. However, this is problematic, as we won't traverse throught the SOCKET_START_CHOWN state for
1872 * them, and thus the UID/GID wouldn't be right. Hence, instead simply check if we have all fds open,
1873 * and if there's a mismatch, warn loudly. */
1874
1875 r = socket_check_open(s);
1876 if (r == SOCKET_OPEN_NONE)
1877 log_unit_warning(UNIT(s),
1878 "Socket unit configuration has changed while unit has been running, "
1879 "no open socket file descriptor left. "
1880 "The socket unit is not functional until restarted.");
1881 else if (r == SOCKET_OPEN_SOME)
1882 log_unit_warning(UNIT(s),
1883 "Socket unit configuration has changed while unit has been running, "
1884 "and some socket file descriptors have not been opened yet. "
1885 "The socket unit is not fully functional until restarted.");
1886 }
1887
1888 if (s->deserialized_state == SOCKET_LISTENING) {
1889 r = socket_watch_fds(s);
1890 if (r < 0)
1891 return r;
1892 }
1893
1894 if (!IN_SET(s->deserialized_state, SOCKET_DEAD, SOCKET_FAILED)) {
1895 (void) unit_setup_dynamic_creds(u);
1896 (void) unit_setup_exec_runtime(u);
1897 }
1898
1899 socket_set_state(s, s->deserialized_state);
1900 return 0;
1901 }
1902
1903 static int socket_spawn(Socket *s, ExecCommand *c, pid_t *_pid) {
1904
1905 _cleanup_(exec_params_clear) ExecParameters exec_params = {
1906 .flags = EXEC_APPLY_SANDBOXING|EXEC_APPLY_CHROOT|EXEC_APPLY_TTY_STDIN,
1907 .stdin_fd = -1,
1908 .stdout_fd = -1,
1909 .stderr_fd = -1,
1910 .exec_fd = -1,
1911 };
1912 pid_t pid;
1913 int r;
1914
1915 assert(s);
1916 assert(c);
1917 assert(_pid);
1918
1919 r = unit_prepare_exec(UNIT(s));
1920 if (r < 0)
1921 return r;
1922
1923 r = socket_arm_timer(s, usec_add(now(CLOCK_MONOTONIC), s->timeout_usec));
1924 if (r < 0)
1925 return r;
1926
1927 r = unit_set_exec_params(UNIT(s), &exec_params);
1928 if (r < 0)
1929 return r;
1930
1931 r = exec_spawn(UNIT(s),
1932 c,
1933 &s->exec_context,
1934 &exec_params,
1935 s->exec_runtime,
1936 &s->dynamic_creds,
1937 &pid);
1938 if (r < 0)
1939 return r;
1940
1941 r = unit_watch_pid(UNIT(s), pid, true);
1942 if (r < 0)
1943 return r;
1944
1945 *_pid = pid;
1946
1947 return 0;
1948 }
1949
1950 static int socket_chown(Socket *s, pid_t *_pid) {
1951 pid_t pid;
1952 int r;
1953
1954 r = socket_arm_timer(s, usec_add(now(CLOCK_MONOTONIC), s->timeout_usec));
1955 if (r < 0)
1956 goto fail;
1957
1958 /* We have to resolve the user names out-of-process, hence
1959 * let's fork here. It's messy, but well, what can we do? */
1960
1961 r = unit_fork_helper_process(UNIT(s), "(sd-chown)", &pid);
1962 if (r < 0)
1963 return r;
1964 if (r == 0) {
1965 uid_t uid = UID_INVALID;
1966 gid_t gid = GID_INVALID;
1967 SocketPort *p;
1968
1969 /* Child */
1970
1971 if (!isempty(s->user)) {
1972 const char *user = s->user;
1973
1974 r = get_user_creds(&user, &uid, &gid, NULL, NULL, 0);
1975 if (r < 0) {
1976 log_unit_error_errno(UNIT(s), r, "Failed to resolve user %s: %m", user);
1977 _exit(EXIT_USER);
1978 }
1979 }
1980
1981 if (!isempty(s->group)) {
1982 const char *group = s->group;
1983
1984 r = get_group_creds(&group, &gid, 0);
1985 if (r < 0) {
1986 log_unit_error_errno(UNIT(s), r, "Failed to resolve group %s: %m", group);
1987 _exit(EXIT_GROUP);
1988 }
1989 }
1990
1991 LIST_FOREACH(port, p, s->ports) {
1992 const char *path = NULL;
1993
1994 if (p->type == SOCKET_SOCKET)
1995 path = socket_address_get_path(&p->address);
1996 else if (p->type == SOCKET_FIFO)
1997 path = p->path;
1998
1999 if (!path)
2000 continue;
2001
2002 if (chown(path, uid, gid) < 0) {
2003 log_unit_error_errno(UNIT(s), errno, "Failed to chown(): %m");
2004 _exit(EXIT_CHOWN);
2005 }
2006 }
2007
2008 _exit(EXIT_SUCCESS);
2009 }
2010
2011 r = unit_watch_pid(UNIT(s), pid, true);
2012 if (r < 0)
2013 goto fail;
2014
2015 *_pid = pid;
2016 return 0;
2017
2018 fail:
2019 s->timer_event_source = sd_event_source_unref(s->timer_event_source);
2020 return r;
2021 }
2022
2023 static void socket_enter_dead(Socket *s, SocketResult f) {
2024 assert(s);
2025
2026 if (s->result == SOCKET_SUCCESS)
2027 s->result = f;
2028
2029 if (s->result == SOCKET_SUCCESS)
2030 unit_log_success(UNIT(s));
2031 else
2032 unit_log_failure(UNIT(s), socket_result_to_string(s->result));
2033
2034 socket_set_state(s, s->result != SOCKET_SUCCESS ? SOCKET_FAILED : SOCKET_DEAD);
2035
2036 s->exec_runtime = exec_runtime_unref(s->exec_runtime, true);
2037
2038 exec_context_destroy_runtime_directory(&s->exec_context, UNIT(s)->manager->prefix[EXEC_DIRECTORY_RUNTIME]);
2039
2040 unit_unref_uid_gid(UNIT(s), true);
2041
2042 dynamic_creds_destroy(&s->dynamic_creds);
2043 }
2044
2045 static void socket_enter_signal(Socket *s, SocketState state, SocketResult f);
2046
2047 static void socket_enter_stop_post(Socket *s, SocketResult f) {
2048 int r;
2049 assert(s);
2050
2051 if (s->result == SOCKET_SUCCESS)
2052 s->result = f;
2053
2054 socket_unwatch_control_pid(s);
2055 s->control_command_id = SOCKET_EXEC_STOP_POST;
2056 s->control_command = s->exec_command[SOCKET_EXEC_STOP_POST];
2057
2058 if (s->control_command) {
2059 r = socket_spawn(s, s->control_command, &s->control_pid);
2060 if (r < 0)
2061 goto fail;
2062
2063 socket_set_state(s, SOCKET_STOP_POST);
2064 } else
2065 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, SOCKET_SUCCESS);
2066
2067 return;
2068
2069 fail:
2070 log_unit_warning_errno(UNIT(s), r, "Failed to run 'stop-post' task: %m");
2071 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, SOCKET_FAILURE_RESOURCES);
2072 }
2073
2074 static void socket_enter_signal(Socket *s, SocketState state, SocketResult f) {
2075 int r;
2076
2077 assert(s);
2078
2079 if (s->result == SOCKET_SUCCESS)
2080 s->result = f;
2081
2082 r = unit_kill_context(
2083 UNIT(s),
2084 &s->kill_context,
2085 !IN_SET(state, SOCKET_STOP_PRE_SIGTERM, SOCKET_FINAL_SIGTERM) ?
2086 KILL_KILL : KILL_TERMINATE,
2087 -1,
2088 s->control_pid,
2089 false);
2090 if (r < 0)
2091 goto fail;
2092
2093 if (r > 0) {
2094 r = socket_arm_timer(s, usec_add(now(CLOCK_MONOTONIC), s->timeout_usec));
2095 if (r < 0)
2096 goto fail;
2097
2098 socket_set_state(s, state);
2099 } else if (state == SOCKET_STOP_PRE_SIGTERM)
2100 socket_enter_signal(s, SOCKET_STOP_PRE_SIGKILL, SOCKET_SUCCESS);
2101 else if (state == SOCKET_STOP_PRE_SIGKILL)
2102 socket_enter_stop_post(s, SOCKET_SUCCESS);
2103 else if (state == SOCKET_FINAL_SIGTERM)
2104 socket_enter_signal(s, SOCKET_FINAL_SIGKILL, SOCKET_SUCCESS);
2105 else
2106 socket_enter_dead(s, SOCKET_SUCCESS);
2107
2108 return;
2109
2110 fail:
2111 log_unit_warning_errno(UNIT(s), r, "Failed to kill processes: %m");
2112
2113 if (IN_SET(state, SOCKET_STOP_PRE_SIGTERM, SOCKET_STOP_PRE_SIGKILL))
2114 socket_enter_stop_post(s, SOCKET_FAILURE_RESOURCES);
2115 else
2116 socket_enter_dead(s, SOCKET_FAILURE_RESOURCES);
2117 }
2118
2119 static void socket_enter_stop_pre(Socket *s, SocketResult f) {
2120 int r;
2121 assert(s);
2122
2123 if (s->result == SOCKET_SUCCESS)
2124 s->result = f;
2125
2126 socket_unwatch_control_pid(s);
2127 s->control_command_id = SOCKET_EXEC_STOP_PRE;
2128 s->control_command = s->exec_command[SOCKET_EXEC_STOP_PRE];
2129
2130 if (s->control_command) {
2131 r = socket_spawn(s, s->control_command, &s->control_pid);
2132 if (r < 0)
2133 goto fail;
2134
2135 socket_set_state(s, SOCKET_STOP_PRE);
2136 } else
2137 socket_enter_stop_post(s, SOCKET_SUCCESS);
2138
2139 return;
2140
2141 fail:
2142 log_unit_warning_errno(UNIT(s), r, "Failed to run 'stop-pre' task: %m");
2143 socket_enter_stop_post(s, SOCKET_FAILURE_RESOURCES);
2144 }
2145
2146 static void socket_enter_listening(Socket *s) {
2147 int r;
2148 assert(s);
2149
2150 r = socket_watch_fds(s);
2151 if (r < 0) {
2152 log_unit_warning_errno(UNIT(s), r, "Failed to watch sockets: %m");
2153 goto fail;
2154 }
2155
2156 socket_set_state(s, SOCKET_LISTENING);
2157 return;
2158
2159 fail:
2160 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2161 }
2162
2163 static void socket_enter_start_post(Socket *s) {
2164 int r;
2165 assert(s);
2166
2167 socket_unwatch_control_pid(s);
2168 s->control_command_id = SOCKET_EXEC_START_POST;
2169 s->control_command = s->exec_command[SOCKET_EXEC_START_POST];
2170
2171 if (s->control_command) {
2172 r = socket_spawn(s, s->control_command, &s->control_pid);
2173 if (r < 0) {
2174 log_unit_warning_errno(UNIT(s), r, "Failed to run 'start-post' task: %m");
2175 goto fail;
2176 }
2177
2178 socket_set_state(s, SOCKET_START_POST);
2179 } else
2180 socket_enter_listening(s);
2181
2182 return;
2183
2184 fail:
2185 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2186 }
2187
2188 static void socket_enter_start_chown(Socket *s) {
2189 int r;
2190
2191 assert(s);
2192
2193 r = socket_open_fds(s);
2194 if (r < 0) {
2195 log_unit_warning_errno(UNIT(s), r, "Failed to listen on sockets: %m");
2196 goto fail;
2197 }
2198
2199 if (!isempty(s->user) || !isempty(s->group)) {
2200
2201 socket_unwatch_control_pid(s);
2202 s->control_command_id = SOCKET_EXEC_START_CHOWN;
2203 s->control_command = NULL;
2204
2205 r = socket_chown(s, &s->control_pid);
2206 if (r < 0) {
2207 log_unit_warning_errno(UNIT(s), r, "Failed to fork 'start-chown' task: %m");
2208 goto fail;
2209 }
2210
2211 socket_set_state(s, SOCKET_START_CHOWN);
2212 } else
2213 socket_enter_start_post(s);
2214
2215 return;
2216
2217 fail:
2218 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2219 }
2220
2221 static void socket_enter_start_pre(Socket *s) {
2222 int r;
2223 assert(s);
2224
2225 socket_unwatch_control_pid(s);
2226
2227 unit_warn_leftover_processes(UNIT(s));
2228
2229 s->control_command_id = SOCKET_EXEC_START_PRE;
2230 s->control_command = s->exec_command[SOCKET_EXEC_START_PRE];
2231
2232 if (s->control_command) {
2233 r = socket_spawn(s, s->control_command, &s->control_pid);
2234 if (r < 0) {
2235 log_unit_warning_errno(UNIT(s), r, "Failed to run 'start-pre' task: %m");
2236 goto fail;
2237 }
2238
2239 socket_set_state(s, SOCKET_START_PRE);
2240 } else
2241 socket_enter_start_chown(s);
2242
2243 return;
2244
2245 fail:
2246 socket_enter_dead(s, SOCKET_FAILURE_RESOURCES);
2247 }
2248
2249 static void flush_ports(Socket *s) {
2250 SocketPort *p;
2251
2252 /* Flush all incoming traffic, regardless if actual bytes or new connections, so that this socket isn't busy
2253 * anymore */
2254
2255 LIST_FOREACH(port, p, s->ports) {
2256 if (p->fd < 0)
2257 continue;
2258
2259 (void) flush_accept(p->fd);
2260 (void) flush_fd(p->fd);
2261 }
2262 }
2263
2264 static void socket_enter_running(Socket *s, int cfd) {
2265 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2266 int r;
2267
2268 /* Note that this call takes possession of the connection fd passed. It either has to assign it somewhere or
2269 * close it. */
2270
2271 assert(s);
2272
2273 /* We don't take connections anymore if we are supposed to shut down anyway */
2274 if (unit_stop_pending(UNIT(s))) {
2275
2276 log_unit_debug(UNIT(s), "Suppressing connection request since unit stop is scheduled.");
2277
2278 if (cfd >= 0)
2279 goto refuse;
2280 else
2281 flush_ports(s);
2282
2283 return;
2284 }
2285
2286 if (!ratelimit_below(&s->trigger_limit)) {
2287 log_unit_warning(UNIT(s), "Trigger limit hit, refusing further activation.");
2288 socket_enter_stop_pre(s, SOCKET_FAILURE_TRIGGER_LIMIT_HIT);
2289 goto refuse;
2290 }
2291
2292 if (cfd < 0) {
2293 bool pending = false;
2294 Unit *other;
2295 Iterator i;
2296 void *v;
2297
2298 /* If there's already a start pending don't bother to
2299 * do anything */
2300 HASHMAP_FOREACH_KEY(v, other, UNIT(s)->dependencies[UNIT_TRIGGERS], i)
2301 if (unit_active_or_pending(other)) {
2302 pending = true;
2303 break;
2304 }
2305
2306 if (!pending) {
2307 if (!UNIT_ISSET(s->service)) {
2308 log_unit_error(UNIT(s), "Service to activate vanished, refusing activation.");
2309 r = -ENOENT;
2310 goto fail;
2311 }
2312
2313 r = manager_add_job(UNIT(s)->manager, JOB_START, UNIT_DEREF(s->service), JOB_REPLACE, &error, NULL);
2314 if (r < 0)
2315 goto fail;
2316 }
2317
2318 socket_set_state(s, SOCKET_RUNNING);
2319 } else {
2320 _cleanup_free_ char *prefix = NULL, *instance = NULL, *name = NULL;
2321 _cleanup_(socket_peer_unrefp) SocketPeer *p = NULL;
2322 Service *service;
2323
2324 if (s->n_connections >= s->max_connections) {
2325 log_unit_warning(UNIT(s), "Too many incoming connections (%u), dropping connection.",
2326 s->n_connections);
2327 goto refuse;
2328 }
2329
2330 if (s->max_connections_per_source > 0) {
2331 r = socket_acquire_peer(s, cfd, &p);
2332 if (r < 0) {
2333 goto refuse;
2334 } else if (r > 0 && p->n_ref > s->max_connections_per_source) {
2335 _cleanup_free_ char *t = NULL;
2336
2337 (void) sockaddr_pretty(&p->peer.sa, p->peer_salen, true, false, &t);
2338
2339 log_unit_warning(UNIT(s),
2340 "Too many incoming connections (%u) from source %s, dropping connection.",
2341 p->n_ref, strnull(t));
2342 goto refuse;
2343 }
2344 }
2345
2346 r = socket_instantiate_service(s);
2347 if (r < 0)
2348 goto fail;
2349
2350 r = instance_from_socket(cfd, s->n_accepted, &instance);
2351 if (r < 0) {
2352 if (r != -ENOTCONN)
2353 goto fail;
2354
2355 /* ENOTCONN is legitimate if TCP RST was received.
2356 * This connection is over, but the socket unit lives on. */
2357 log_unit_debug(UNIT(s), "Got ENOTCONN on incoming socket, assuming aborted connection attempt, ignoring.");
2358 goto refuse;
2359 }
2360
2361 r = unit_name_to_prefix(UNIT(s)->id, &prefix);
2362 if (r < 0)
2363 goto fail;
2364
2365 r = unit_name_build(prefix, instance, ".service", &name);
2366 if (r < 0)
2367 goto fail;
2368
2369 r = unit_add_name(UNIT_DEREF(s->service), name);
2370 if (r < 0)
2371 goto fail;
2372
2373 service = SERVICE(UNIT_DEREF(s->service));
2374 unit_ref_unset(&s->service);
2375
2376 s->n_accepted++;
2377 unit_choose_id(UNIT(service), name);
2378
2379 r = service_set_socket_fd(service, cfd, s, s->selinux_context_from_net);
2380 if (r < 0)
2381 goto fail;
2382
2383 cfd = -1; /* We passed ownership of the fd to the service now. Forget it here. */
2384 s->n_connections++;
2385
2386 service->peer = TAKE_PTR(p); /* Pass ownership of the peer reference */
2387
2388 r = manager_add_job(UNIT(s)->manager, JOB_START, UNIT(service), JOB_REPLACE, &error, NULL);
2389 if (r < 0) {
2390 /* We failed to activate the new service, but it still exists. Let's make sure the service
2391 * closes and forgets the connection fd again, immediately. */
2392 service_close_socket_fd(service);
2393 goto fail;
2394 }
2395
2396 /* Notify clients about changed counters */
2397 unit_add_to_dbus_queue(UNIT(s));
2398 }
2399
2400 return;
2401
2402 refuse:
2403 s->n_refused++;
2404 safe_close(cfd);
2405 return;
2406
2407 fail:
2408 log_unit_warning(UNIT(s), "Failed to queue service startup job (Maybe the service file is missing or not a %s unit?): %s",
2409 cfd >= 0 ? "template" : "non-template",
2410 bus_error_message(&error, r));
2411
2412 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2413 safe_close(cfd);
2414 }
2415
2416 static void socket_run_next(Socket *s) {
2417 int r;
2418
2419 assert(s);
2420 assert(s->control_command);
2421 assert(s->control_command->command_next);
2422
2423 socket_unwatch_control_pid(s);
2424
2425 s->control_command = s->control_command->command_next;
2426
2427 r = socket_spawn(s, s->control_command, &s->control_pid);
2428 if (r < 0)
2429 goto fail;
2430
2431 return;
2432
2433 fail:
2434 log_unit_warning_errno(UNIT(s), r, "Failed to run next task: %m");
2435
2436 if (s->state == SOCKET_START_POST)
2437 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2438 else if (s->state == SOCKET_STOP_POST)
2439 socket_enter_dead(s, SOCKET_FAILURE_RESOURCES);
2440 else
2441 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, SOCKET_FAILURE_RESOURCES);
2442 }
2443
2444 static int socket_start(Unit *u) {
2445 Socket *s = SOCKET(u);
2446 int r;
2447
2448 assert(s);
2449
2450 /* We cannot fulfill this request right now, try again later
2451 * please! */
2452 if (IN_SET(s->state,
2453 SOCKET_STOP_PRE,
2454 SOCKET_STOP_PRE_SIGKILL,
2455 SOCKET_STOP_PRE_SIGTERM,
2456 SOCKET_STOP_POST,
2457 SOCKET_FINAL_SIGTERM,
2458 SOCKET_FINAL_SIGKILL))
2459 return -EAGAIN;
2460
2461 /* Already on it! */
2462 if (IN_SET(s->state,
2463 SOCKET_START_PRE,
2464 SOCKET_START_CHOWN,
2465 SOCKET_START_POST))
2466 return 0;
2467
2468 /* Cannot run this without the service being around */
2469 if (UNIT_ISSET(s->service)) {
2470 Service *service;
2471
2472 service = SERVICE(UNIT_DEREF(s->service));
2473
2474 if (UNIT(service)->load_state != UNIT_LOADED) {
2475 log_unit_error(u, "Socket service %s not loaded, refusing.", UNIT(service)->id);
2476 return -ENOENT;
2477 }
2478
2479 /* If the service is already active we cannot start the
2480 * socket */
2481 if (!IN_SET(service->state, SERVICE_DEAD, SERVICE_FAILED, SERVICE_AUTO_RESTART)) {
2482 log_unit_error(u, "Socket service %s already active, refusing.", UNIT(service)->id);
2483 return -EBUSY;
2484 }
2485 }
2486
2487 assert(IN_SET(s->state, SOCKET_DEAD, SOCKET_FAILED));
2488
2489 r = unit_test_start_limit(u);
2490 if (r < 0) {
2491 socket_enter_dead(s, SOCKET_FAILURE_START_LIMIT_HIT);
2492 return r;
2493 }
2494
2495 r = unit_acquire_invocation_id(u);
2496 if (r < 0)
2497 return r;
2498
2499 s->result = SOCKET_SUCCESS;
2500 exec_command_reset_status_list_array(s->exec_command, _SOCKET_EXEC_COMMAND_MAX);
2501
2502 u->reset_accounting = true;
2503
2504 socket_enter_start_pre(s);
2505 return 1;
2506 }
2507
2508 static int socket_stop(Unit *u) {
2509 Socket *s = SOCKET(u);
2510
2511 assert(s);
2512
2513 /* Already on it */
2514 if (IN_SET(s->state,
2515 SOCKET_STOP_PRE,
2516 SOCKET_STOP_PRE_SIGTERM,
2517 SOCKET_STOP_PRE_SIGKILL,
2518 SOCKET_STOP_POST,
2519 SOCKET_FINAL_SIGTERM,
2520 SOCKET_FINAL_SIGKILL))
2521 return 0;
2522
2523 /* If there's already something running we go directly into
2524 * kill mode. */
2525 if (IN_SET(s->state,
2526 SOCKET_START_PRE,
2527 SOCKET_START_CHOWN,
2528 SOCKET_START_POST)) {
2529 socket_enter_signal(s, SOCKET_STOP_PRE_SIGTERM, SOCKET_SUCCESS);
2530 return -EAGAIN;
2531 }
2532
2533 assert(IN_SET(s->state, SOCKET_LISTENING, SOCKET_RUNNING));
2534
2535 socket_enter_stop_pre(s, SOCKET_SUCCESS);
2536 return 1;
2537 }
2538
2539 static int socket_serialize(Unit *u, FILE *f, FDSet *fds) {
2540 Socket *s = SOCKET(u);
2541 SocketPort *p;
2542 int r;
2543
2544 assert(u);
2545 assert(f);
2546 assert(fds);
2547
2548 (void) serialize_item(f, "state", socket_state_to_string(s->state));
2549 (void) serialize_item(f, "result", socket_result_to_string(s->result));
2550 (void) serialize_item_format(f, "n-accepted", "%u", s->n_accepted);
2551 (void) serialize_item_format(f, "n-refused", "%u", s->n_refused);
2552
2553 if (s->control_pid > 0)
2554 (void) serialize_item_format(f, "control-pid", PID_FMT, s->control_pid);
2555
2556 if (s->control_command_id >= 0)
2557 (void) serialize_item(f, "control-command", socket_exec_command_to_string(s->control_command_id));
2558
2559 LIST_FOREACH(port, p, s->ports) {
2560 int copy;
2561
2562 if (p->fd < 0)
2563 continue;
2564
2565 copy = fdset_put_dup(fds, p->fd);
2566 if (copy < 0)
2567 return log_unit_warning_errno(u, copy, "Failed to serialize socket fd: %m");
2568
2569 if (p->type == SOCKET_SOCKET) {
2570 _cleanup_free_ char *t = NULL;
2571
2572 r = socket_address_print(&p->address, &t);
2573 if (r < 0)
2574 return log_unit_error_errno(u, r, "Failed to format socket address: %m");
2575
2576 if (socket_address_family(&p->address) == AF_NETLINK)
2577 (void) serialize_item_format(f, "netlink", "%i %s", copy, t);
2578 else
2579 (void) serialize_item_format(f, "socket", "%i %i %s", copy, p->address.type, t);
2580 } else if (p->type == SOCKET_SPECIAL)
2581 (void) serialize_item_format(f, "special", "%i %s", copy, p->path);
2582 else if (p->type == SOCKET_MQUEUE)
2583 (void) serialize_item_format(f, "mqueue", "%i %s", copy, p->path);
2584 else if (p->type == SOCKET_USB_FUNCTION)
2585 (void) serialize_item_format(f, "ffs", "%i %s", copy, p->path);
2586 else {
2587 assert(p->type == SOCKET_FIFO);
2588 (void) serialize_item_format(f, "fifo", "%i %s", copy, p->path);
2589 }
2590 }
2591
2592 return 0;
2593 }
2594
2595 static void socket_port_take_fd(SocketPort *p, FDSet *fds, int fd) {
2596 assert(p);
2597
2598 safe_close(p->fd);
2599 p->fd = fdset_remove(fds, fd);
2600 }
2601
2602 static int socket_deserialize_item(Unit *u, const char *key, const char *value, FDSet *fds) {
2603 Socket *s = SOCKET(u);
2604
2605 assert(u);
2606 assert(key);
2607 assert(value);
2608
2609 if (streq(key, "state")) {
2610 SocketState state;
2611
2612 state = socket_state_from_string(value);
2613 if (state < 0)
2614 log_unit_debug(u, "Failed to parse state value: %s", value);
2615 else
2616 s->deserialized_state = state;
2617 } else if (streq(key, "result")) {
2618 SocketResult f;
2619
2620 f = socket_result_from_string(value);
2621 if (f < 0)
2622 log_unit_debug(u, "Failed to parse result value: %s", value);
2623 else if (f != SOCKET_SUCCESS)
2624 s->result = f;
2625
2626 } else if (streq(key, "n-accepted")) {
2627 unsigned k;
2628
2629 if (safe_atou(value, &k) < 0)
2630 log_unit_debug(u, "Failed to parse n-accepted value: %s", value);
2631 else
2632 s->n_accepted += k;
2633 } else if (streq(key, "n-refused")) {
2634 unsigned k;
2635
2636 if (safe_atou(value, &k) < 0)
2637 log_unit_debug(u, "Failed to parse n-refused value: %s", value);
2638 else
2639 s->n_refused += k;
2640 } else if (streq(key, "control-pid")) {
2641 pid_t pid;
2642
2643 if (parse_pid(value, &pid) < 0)
2644 log_unit_debug(u, "Failed to parse control-pid value: %s", value);
2645 else
2646 s->control_pid = pid;
2647 } else if (streq(key, "control-command")) {
2648 SocketExecCommand id;
2649
2650 id = socket_exec_command_from_string(value);
2651 if (id < 0)
2652 log_unit_debug(u, "Failed to parse exec-command value: %s", value);
2653 else {
2654 s->control_command_id = id;
2655 s->control_command = s->exec_command[id];
2656 }
2657 } else if (streq(key, "fifo")) {
2658 int fd, skip = 0;
2659 SocketPort *p;
2660
2661 if (sscanf(value, "%i %n", &fd, &skip) < 1 || fd < 0 || !fdset_contains(fds, fd))
2662 log_unit_debug(u, "Failed to parse fifo value: %s", value);
2663 else
2664 LIST_FOREACH(port, p, s->ports)
2665 if (p->type == SOCKET_FIFO &&
2666 path_equal_or_files_same(p->path, value+skip, 0)) {
2667 socket_port_take_fd(p, fds, fd);
2668 break;
2669 }
2670
2671 } else if (streq(key, "special")) {
2672 int fd, skip = 0;
2673 SocketPort *p;
2674
2675 if (sscanf(value, "%i %n", &fd, &skip) < 1 || fd < 0 || !fdset_contains(fds, fd))
2676 log_unit_debug(u, "Failed to parse special value: %s", value);
2677 else
2678 LIST_FOREACH(port, p, s->ports)
2679 if (p->type == SOCKET_SPECIAL &&
2680 path_equal_or_files_same(p->path, value+skip, 0)) {
2681 socket_port_take_fd(p, fds, fd);
2682 break;
2683 }
2684
2685 } else if (streq(key, "mqueue")) {
2686 int fd, skip = 0;
2687 SocketPort *p;
2688
2689 if (sscanf(value, "%i %n", &fd, &skip) < 1 || fd < 0 || !fdset_contains(fds, fd))
2690 log_unit_debug(u, "Failed to parse mqueue value: %s", value);
2691 else
2692 LIST_FOREACH(port, p, s->ports)
2693 if (p->type == SOCKET_MQUEUE &&
2694 streq(p->path, value+skip)) {
2695 socket_port_take_fd(p, fds, fd);
2696 break;
2697 }
2698
2699 } else if (streq(key, "socket")) {
2700 int fd, type, skip = 0;
2701 SocketPort *p;
2702
2703 if (sscanf(value, "%i %i %n", &fd, &type, &skip) < 2 || fd < 0 || type < 0 || !fdset_contains(fds, fd))
2704 log_unit_debug(u, "Failed to parse socket value: %s", value);
2705 else
2706 LIST_FOREACH(port, p, s->ports)
2707 if (socket_address_is(&p->address, value+skip, type)) {
2708 socket_port_take_fd(p, fds, fd);
2709 break;
2710 }
2711
2712 } else if (streq(key, "netlink")) {
2713 int fd, skip = 0;
2714 SocketPort *p;
2715
2716 if (sscanf(value, "%i %n", &fd, &skip) < 1 || fd < 0 || !fdset_contains(fds, fd))
2717 log_unit_debug(u, "Failed to parse socket value: %s", value);
2718 else
2719 LIST_FOREACH(port, p, s->ports)
2720 if (socket_address_is_netlink(&p->address, value+skip)) {
2721 socket_port_take_fd(p, fds, fd);
2722 break;
2723 }
2724
2725 } else if (streq(key, "ffs")) {
2726 int fd, skip = 0;
2727 SocketPort *p;
2728
2729 if (sscanf(value, "%i %n", &fd, &skip) < 1 || fd < 0 || !fdset_contains(fds, fd))
2730 log_unit_debug(u, "Failed to parse ffs value: %s", value);
2731 else
2732 LIST_FOREACH(port, p, s->ports)
2733 if (p->type == SOCKET_USB_FUNCTION &&
2734 path_equal_or_files_same(p->path, value+skip, 0)) {
2735 socket_port_take_fd(p, fds, fd);
2736 break;
2737 }
2738
2739 } else
2740 log_unit_debug(UNIT(s), "Unknown serialization key: %s", key);
2741
2742 return 0;
2743 }
2744
2745 static void socket_distribute_fds(Unit *u, FDSet *fds) {
2746 Socket *s = SOCKET(u);
2747 SocketPort *p;
2748
2749 assert(u);
2750
2751 LIST_FOREACH(port, p, s->ports) {
2752 Iterator i;
2753 int fd;
2754
2755 if (p->type != SOCKET_SOCKET)
2756 continue;
2757
2758 if (p->fd >= 0)
2759 continue;
2760
2761 FDSET_FOREACH(fd, fds, i) {
2762 if (socket_address_matches_fd(&p->address, fd)) {
2763 p->fd = fdset_remove(fds, fd);
2764 s->deserialized_state = SOCKET_LISTENING;
2765 break;
2766 }
2767 }
2768 }
2769 }
2770
2771 _pure_ static UnitActiveState socket_active_state(Unit *u) {
2772 assert(u);
2773
2774 return state_translation_table[SOCKET(u)->state];
2775 }
2776
2777 _pure_ static const char *socket_sub_state_to_string(Unit *u) {
2778 assert(u);
2779
2780 return socket_state_to_string(SOCKET(u)->state);
2781 }
2782
2783 const char* socket_port_type_to_string(SocketPort *p) {
2784
2785 assert(p);
2786
2787 switch (p->type) {
2788
2789 case SOCKET_SOCKET:
2790
2791 switch (p->address.type) {
2792
2793 case SOCK_STREAM:
2794 return "Stream";
2795
2796 case SOCK_DGRAM:
2797 return "Datagram";
2798
2799 case SOCK_SEQPACKET:
2800 return "SequentialPacket";
2801
2802 case SOCK_RAW:
2803 if (socket_address_family(&p->address) == AF_NETLINK)
2804 return "Netlink";
2805
2806 _fallthrough_;
2807 default:
2808 return NULL;
2809 }
2810
2811 case SOCKET_SPECIAL:
2812 return "Special";
2813
2814 case SOCKET_MQUEUE:
2815 return "MessageQueue";
2816
2817 case SOCKET_FIFO:
2818 return "FIFO";
2819
2820 case SOCKET_USB_FUNCTION:
2821 return "USBFunction";
2822
2823 default:
2824 return NULL;
2825 }
2826 }
2827
2828 SocketType socket_port_type_from_string(const char *s) {
2829 assert(s);
2830
2831 if (STR_IN_SET(s, "Stream", "Datagram", "SequentialPacket", "Netlink"))
2832 return SOCKET_SOCKET;
2833 else if (streq(s, "Special"))
2834 return SOCKET_SPECIAL;
2835 else if (streq(s, "MessageQueue"))
2836 return SOCKET_MQUEUE;
2837 else if (streq(s, "FIFO"))
2838 return SOCKET_FIFO;
2839 else if (streq(s, "USBFunction"))
2840 return SOCKET_USB_FUNCTION;
2841 else
2842 return _SOCKET_TYPE_INVALID;
2843 }
2844
2845 _pure_ static bool socket_may_gc(Unit *u) {
2846 Socket *s = SOCKET(u);
2847
2848 assert(u);
2849
2850 return s->n_connections == 0;
2851 }
2852
2853 static int socket_accept_do(Socket *s, int fd) {
2854 int cfd;
2855
2856 assert(s);
2857 assert(fd >= 0);
2858
2859 for (;;) {
2860 cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK);
2861 if (cfd < 0) {
2862 if (errno == EINTR)
2863 continue;
2864
2865 return -errno;
2866 }
2867
2868 break;
2869 }
2870
2871 return cfd;
2872 }
2873
2874 static int socket_accept_in_cgroup(Socket *s, SocketPort *p, int fd) {
2875 _cleanup_close_pair_ int pair[2] = { -1, -1 };
2876 int cfd, r;
2877 pid_t pid;
2878
2879 assert(s);
2880 assert(p);
2881 assert(fd >= 0);
2882
2883 /* Similar to socket_address_listen_in_cgroup(), but for accept() rathern than socket(): make sure that any
2884 * connection socket is also properly associated with the cgroup. */
2885
2886 if (!IN_SET(p->address.sockaddr.sa.sa_family, AF_INET, AF_INET6))
2887 goto shortcut;
2888
2889 r = bpf_firewall_supported();
2890 if (r < 0)
2891 return r;
2892 if (r == BPF_FIREWALL_UNSUPPORTED)
2893 goto shortcut;
2894
2895 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pair) < 0)
2896 return log_unit_error_errno(UNIT(s), errno, "Failed to create communication channel: %m");
2897
2898 r = unit_fork_helper_process(UNIT(s), "(sd-accept)", &pid);
2899 if (r < 0)
2900 return log_unit_error_errno(UNIT(s), r, "Failed to fork off accept stub process: %m");
2901 if (r == 0) {
2902 /* Child */
2903
2904 pair[0] = safe_close(pair[0]);
2905
2906 cfd = socket_accept_do(s, fd);
2907 if (cfd < 0) {
2908 log_unit_error_errno(UNIT(s), cfd, "Failed to accept connection socket: %m");
2909 _exit(EXIT_FAILURE);
2910 }
2911
2912 r = send_one_fd(pair[1], cfd, 0);
2913 if (r < 0) {
2914 log_unit_error_errno(UNIT(s), r, "Failed to send connection socket to parent: %m");
2915 _exit(EXIT_FAILURE);
2916 }
2917
2918 _exit(EXIT_SUCCESS);
2919 }
2920
2921 pair[1] = safe_close(pair[1]);
2922 cfd = receive_one_fd(pair[0], 0);
2923
2924 /* We synchronously wait for the helper, as it shouldn't be slow */
2925 r = wait_for_terminate_and_check("(sd-accept)", pid, WAIT_LOG_ABNORMAL);
2926 if (r < 0) {
2927 safe_close(cfd);
2928 return r;
2929 }
2930
2931 if (cfd < 0)
2932 return log_unit_error_errno(UNIT(s), cfd, "Failed to receive connection socket: %m");
2933
2934 return cfd;
2935
2936 shortcut:
2937 cfd = socket_accept_do(s, fd);
2938 if (cfd < 0)
2939 return log_unit_error_errno(UNIT(s), cfd, "Failed to accept connection socket: %m");
2940
2941 return cfd;
2942 }
2943
2944 static int socket_dispatch_io(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2945 SocketPort *p = userdata;
2946 int cfd = -1;
2947
2948 assert(p);
2949 assert(fd >= 0);
2950
2951 if (p->socket->state != SOCKET_LISTENING)
2952 return 0;
2953
2954 log_unit_debug(UNIT(p->socket), "Incoming traffic");
2955
2956 if (revents != EPOLLIN) {
2957
2958 if (revents & EPOLLHUP)
2959 log_unit_error(UNIT(p->socket), "Got POLLHUP on a listening socket. The service probably invoked shutdown() on it, and should better not do that.");
2960 else
2961 log_unit_error(UNIT(p->socket), "Got unexpected poll event (0x%x) on socket.", revents);
2962 goto fail;
2963 }
2964
2965 if (p->socket->accept &&
2966 p->type == SOCKET_SOCKET &&
2967 socket_address_can_accept(&p->address)) {
2968
2969 cfd = socket_accept_in_cgroup(p->socket, p, fd);
2970 if (cfd < 0)
2971 goto fail;
2972
2973 socket_apply_socket_options(p->socket, cfd);
2974 }
2975
2976 socket_enter_running(p->socket, cfd);
2977 return 0;
2978
2979 fail:
2980 socket_enter_stop_pre(p->socket, SOCKET_FAILURE_RESOURCES);
2981 return 0;
2982 }
2983
2984 static void socket_sigchld_event(Unit *u, pid_t pid, int code, int status) {
2985 Socket *s = SOCKET(u);
2986 SocketResult f;
2987
2988 assert(s);
2989 assert(pid >= 0);
2990
2991 if (pid != s->control_pid)
2992 return;
2993
2994 s->control_pid = 0;
2995
2996 if (is_clean_exit(code, status, EXIT_CLEAN_COMMAND, NULL))
2997 f = SOCKET_SUCCESS;
2998 else if (code == CLD_EXITED)
2999 f = SOCKET_FAILURE_EXIT_CODE;
3000 else if (code == CLD_KILLED)
3001 f = SOCKET_FAILURE_SIGNAL;
3002 else if (code == CLD_DUMPED)
3003 f = SOCKET_FAILURE_CORE_DUMP;
3004 else
3005 assert_not_reached("Unknown sigchld code");
3006
3007 if (s->control_command) {
3008 exec_status_exit(&s->control_command->exec_status, &s->exec_context, pid, code, status);
3009
3010 if (s->control_command->flags & EXEC_COMMAND_IGNORE_FAILURE)
3011 f = SOCKET_SUCCESS;
3012 }
3013
3014 unit_log_process_exit(
3015 u, f == SOCKET_SUCCESS ? LOG_DEBUG : LOG_NOTICE,
3016 "Control process",
3017 socket_exec_command_to_string(s->control_command_id),
3018 code, status);
3019
3020 if (s->result == SOCKET_SUCCESS)
3021 s->result = f;
3022
3023 if (s->control_command &&
3024 s->control_command->command_next &&
3025 f == SOCKET_SUCCESS) {
3026
3027 log_unit_debug(u, "Running next command for state %s", socket_state_to_string(s->state));
3028 socket_run_next(s);
3029 } else {
3030 s->control_command = NULL;
3031 s->control_command_id = _SOCKET_EXEC_COMMAND_INVALID;
3032
3033 /* No further commands for this step, so let's figure
3034 * out what to do next */
3035
3036 log_unit_debug(u, "Got final SIGCHLD for state %s", socket_state_to_string(s->state));
3037
3038 switch (s->state) {
3039
3040 case SOCKET_START_PRE:
3041 if (f == SOCKET_SUCCESS)
3042 socket_enter_start_chown(s);
3043 else
3044 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, f);
3045 break;
3046
3047 case SOCKET_START_CHOWN:
3048 if (f == SOCKET_SUCCESS)
3049 socket_enter_start_post(s);
3050 else
3051 socket_enter_stop_pre(s, f);
3052 break;
3053
3054 case SOCKET_START_POST:
3055 if (f == SOCKET_SUCCESS)
3056 socket_enter_listening(s);
3057 else
3058 socket_enter_stop_pre(s, f);
3059 break;
3060
3061 case SOCKET_STOP_PRE:
3062 case SOCKET_STOP_PRE_SIGTERM:
3063 case SOCKET_STOP_PRE_SIGKILL:
3064 socket_enter_stop_post(s, f);
3065 break;
3066
3067 case SOCKET_STOP_POST:
3068 case SOCKET_FINAL_SIGTERM:
3069 case SOCKET_FINAL_SIGKILL:
3070 socket_enter_dead(s, f);
3071 break;
3072
3073 default:
3074 assert_not_reached("Uh, control process died at wrong time.");
3075 }
3076 }
3077
3078 /* Notify clients about changed exit status */
3079 unit_add_to_dbus_queue(u);
3080 }
3081
3082 static int socket_dispatch_timer(sd_event_source *source, usec_t usec, void *userdata) {
3083 Socket *s = SOCKET(userdata);
3084
3085 assert(s);
3086 assert(s->timer_event_source == source);
3087
3088 switch (s->state) {
3089
3090 case SOCKET_START_PRE:
3091 log_unit_warning(UNIT(s), "Starting timed out. Terminating.");
3092 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, SOCKET_FAILURE_TIMEOUT);
3093 break;
3094
3095 case SOCKET_START_CHOWN:
3096 case SOCKET_START_POST:
3097 log_unit_warning(UNIT(s), "Starting timed out. Stopping.");
3098 socket_enter_stop_pre(s, SOCKET_FAILURE_TIMEOUT);
3099 break;
3100
3101 case SOCKET_STOP_PRE:
3102 log_unit_warning(UNIT(s), "Stopping timed out. Terminating.");
3103 socket_enter_signal(s, SOCKET_STOP_PRE_SIGTERM, SOCKET_FAILURE_TIMEOUT);
3104 break;
3105
3106 case SOCKET_STOP_PRE_SIGTERM:
3107 if (s->kill_context.send_sigkill) {
3108 log_unit_warning(UNIT(s), "Stopping timed out. Killing.");
3109 socket_enter_signal(s, SOCKET_STOP_PRE_SIGKILL, SOCKET_FAILURE_TIMEOUT);
3110 } else {
3111 log_unit_warning(UNIT(s), "Stopping timed out. Skipping SIGKILL. Ignoring.");
3112 socket_enter_stop_post(s, SOCKET_FAILURE_TIMEOUT);
3113 }
3114 break;
3115
3116 case SOCKET_STOP_PRE_SIGKILL:
3117 log_unit_warning(UNIT(s), "Processes still around after SIGKILL. Ignoring.");
3118 socket_enter_stop_post(s, SOCKET_FAILURE_TIMEOUT);
3119 break;
3120
3121 case SOCKET_STOP_POST:
3122 log_unit_warning(UNIT(s), "Stopping timed out (2). Terminating.");
3123 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, SOCKET_FAILURE_TIMEOUT);
3124 break;
3125
3126 case SOCKET_FINAL_SIGTERM:
3127 if (s->kill_context.send_sigkill) {
3128 log_unit_warning(UNIT(s), "Stopping timed out (2). Killing.");
3129 socket_enter_signal(s, SOCKET_FINAL_SIGKILL, SOCKET_FAILURE_TIMEOUT);
3130 } else {
3131 log_unit_warning(UNIT(s), "Stopping timed out (2). Skipping SIGKILL. Ignoring.");
3132 socket_enter_dead(s, SOCKET_FAILURE_TIMEOUT);
3133 }
3134 break;
3135
3136 case SOCKET_FINAL_SIGKILL:
3137 log_unit_warning(UNIT(s), "Still around after SIGKILL (2). Entering failed mode.");
3138 socket_enter_dead(s, SOCKET_FAILURE_TIMEOUT);
3139 break;
3140
3141 default:
3142 assert_not_reached("Timeout at wrong time.");
3143 }
3144
3145 return 0;
3146 }
3147
3148 int socket_collect_fds(Socket *s, int **fds) {
3149 size_t k = 0, n = 0;
3150 SocketPort *p;
3151 int *rfds;
3152
3153 assert(s);
3154 assert(fds);
3155
3156 /* Called from the service code for requesting our fds */
3157
3158 LIST_FOREACH(port, p, s->ports) {
3159 if (p->fd >= 0)
3160 n++;
3161 n += p->n_auxiliary_fds;
3162 }
3163
3164 if (n <= 0) {
3165 *fds = NULL;
3166 return 0;
3167 }
3168
3169 rfds = new(int, n);
3170 if (!rfds)
3171 return -ENOMEM;
3172
3173 LIST_FOREACH(port, p, s->ports) {
3174 size_t i;
3175
3176 if (p->fd >= 0)
3177 rfds[k++] = p->fd;
3178 for (i = 0; i < p->n_auxiliary_fds; ++i)
3179 rfds[k++] = p->auxiliary_fds[i];
3180 }
3181
3182 assert(k == n);
3183
3184 *fds = rfds;
3185 return (int) n;
3186 }
3187
3188 static void socket_reset_failed(Unit *u) {
3189 Socket *s = SOCKET(u);
3190
3191 assert(s);
3192
3193 if (s->state == SOCKET_FAILED)
3194 socket_set_state(s, SOCKET_DEAD);
3195
3196 s->result = SOCKET_SUCCESS;
3197 }
3198
3199 void socket_connection_unref(Socket *s) {
3200 assert(s);
3201
3202 /* The service is dead. Yay!
3203 *
3204 * This is strictly for one-instance-per-connection
3205 * services. */
3206
3207 assert(s->n_connections > 0);
3208 s->n_connections--;
3209
3210 log_unit_debug(UNIT(s), "One connection closed, %u left.", s->n_connections);
3211 }
3212
3213 static void socket_trigger_notify(Unit *u, Unit *other) {
3214 Socket *s = SOCKET(u);
3215
3216 assert(u);
3217 assert(other);
3218
3219 /* Filter out invocations with bogus state */
3220 if (other->load_state != UNIT_LOADED || other->type != UNIT_SERVICE)
3221 return;
3222
3223 /* Don't propagate state changes from the service if we are already down */
3224 if (!IN_SET(s->state, SOCKET_RUNNING, SOCKET_LISTENING))
3225 return;
3226
3227 /* We don't care for the service state if we are in Accept=yes mode */
3228 if (s->accept)
3229 return;
3230
3231 /* Propagate start limit hit state */
3232 if (other->start_limit_hit) {
3233 socket_enter_stop_pre(s, SOCKET_FAILURE_SERVICE_START_LIMIT_HIT);
3234 return;
3235 }
3236
3237 /* Don't propagate anything if there's still a job queued */
3238 if (other->job)
3239 return;
3240
3241 if (IN_SET(SERVICE(other)->state,
3242 SERVICE_DEAD, SERVICE_FAILED,
3243 SERVICE_FINAL_SIGTERM, SERVICE_FINAL_SIGKILL,
3244 SERVICE_AUTO_RESTART))
3245 socket_enter_listening(s);
3246
3247 if (SERVICE(other)->state == SERVICE_RUNNING)
3248 socket_set_state(s, SOCKET_RUNNING);
3249 }
3250
3251 static int socket_kill(Unit *u, KillWho who, int signo, sd_bus_error *error) {
3252 return unit_kill_common(u, who, signo, -1, SOCKET(u)->control_pid, error);
3253 }
3254
3255 static int socket_get_timeout(Unit *u, usec_t *timeout) {
3256 Socket *s = SOCKET(u);
3257 usec_t t;
3258 int r;
3259
3260 if (!s->timer_event_source)
3261 return 0;
3262
3263 r = sd_event_source_get_time(s->timer_event_source, &t);
3264 if (r < 0)
3265 return r;
3266 if (t == USEC_INFINITY)
3267 return 0;
3268
3269 *timeout = t;
3270 return 1;
3271 }
3272
3273 char *socket_fdname(Socket *s) {
3274 assert(s);
3275
3276 /* Returns the name to use for $LISTEN_NAMES. If the user
3277 * didn't specify anything specifically, use the socket unit's
3278 * name as fallback. */
3279
3280 return s->fdname ?: UNIT(s)->id;
3281 }
3282
3283 static int socket_control_pid(Unit *u) {
3284 Socket *s = SOCKET(u);
3285
3286 assert(s);
3287
3288 return s->control_pid;
3289 }
3290
3291 static const char* const socket_exec_command_table[_SOCKET_EXEC_COMMAND_MAX] = {
3292 [SOCKET_EXEC_START_PRE] = "ExecStartPre",
3293 [SOCKET_EXEC_START_CHOWN] = "ExecStartChown",
3294 [SOCKET_EXEC_START_POST] = "ExecStartPost",
3295 [SOCKET_EXEC_STOP_PRE] = "ExecStopPre",
3296 [SOCKET_EXEC_STOP_POST] = "ExecStopPost"
3297 };
3298
3299 DEFINE_STRING_TABLE_LOOKUP(socket_exec_command, SocketExecCommand);
3300
3301 static const char* const socket_result_table[_SOCKET_RESULT_MAX] = {
3302 [SOCKET_SUCCESS] = "success",
3303 [SOCKET_FAILURE_RESOURCES] = "resources",
3304 [SOCKET_FAILURE_TIMEOUT] = "timeout",
3305 [SOCKET_FAILURE_EXIT_CODE] = "exit-code",
3306 [SOCKET_FAILURE_SIGNAL] = "signal",
3307 [SOCKET_FAILURE_CORE_DUMP] = "core-dump",
3308 [SOCKET_FAILURE_START_LIMIT_HIT] = "start-limit-hit",
3309 [SOCKET_FAILURE_TRIGGER_LIMIT_HIT] = "trigger-limit-hit",
3310 [SOCKET_FAILURE_SERVICE_START_LIMIT_HIT] = "service-start-limit-hit"
3311 };
3312
3313 DEFINE_STRING_TABLE_LOOKUP(socket_result, SocketResult);
3314
3315 const UnitVTable socket_vtable = {
3316 .object_size = sizeof(Socket),
3317 .exec_context_offset = offsetof(Socket, exec_context),
3318 .cgroup_context_offset = offsetof(Socket, cgroup_context),
3319 .kill_context_offset = offsetof(Socket, kill_context),
3320 .exec_runtime_offset = offsetof(Socket, exec_runtime),
3321 .dynamic_creds_offset = offsetof(Socket, dynamic_creds),
3322
3323 .sections =
3324 "Unit\0"
3325 "Socket\0"
3326 "Install\0",
3327 .private_section = "Socket",
3328
3329 .can_transient = true,
3330
3331 .init = socket_init,
3332 .done = socket_done,
3333 .load = socket_load,
3334
3335 .coldplug = socket_coldplug,
3336
3337 .dump = socket_dump,
3338
3339 .start = socket_start,
3340 .stop = socket_stop,
3341
3342 .kill = socket_kill,
3343
3344 .get_timeout = socket_get_timeout,
3345
3346 .serialize = socket_serialize,
3347 .deserialize_item = socket_deserialize_item,
3348 .distribute_fds = socket_distribute_fds,
3349
3350 .active_state = socket_active_state,
3351 .sub_state_to_string = socket_sub_state_to_string,
3352
3353 .may_gc = socket_may_gc,
3354
3355 .sigchld_event = socket_sigchld_event,
3356
3357 .trigger_notify = socket_trigger_notify,
3358
3359 .reset_failed = socket_reset_failed,
3360
3361 .control_pid = socket_control_pid,
3362
3363 .bus_vtable = bus_socket_vtable,
3364 .bus_set_property = bus_socket_set_property,
3365 .bus_commit_properties = bus_socket_commit_properties,
3366
3367 .status_message_formats = {
3368 /*.starting_stopping = {
3369 [0] = "Starting socket %s...",
3370 [1] = "Stopping socket %s...",
3371 },*/
3372 .finished_start_job = {
3373 [JOB_DONE] = "Listening on %s.",
3374 [JOB_FAILED] = "Failed to listen on %s.",
3375 [JOB_TIMEOUT] = "Timed out starting %s.",
3376 },
3377 .finished_stop_job = {
3378 [JOB_DONE] = "Closed %s.",
3379 [JOB_FAILED] = "Failed stopping %s.",
3380 [JOB_TIMEOUT] = "Timed out stopping %s.",
3381 },
3382 },
3383 };
Unnamed repository; edit this file 'description' to name the repository.