]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/core/socket.c
projects / thirdparty / systemd.git / blob
? search:


e5cff0de7610518042ad06497f0ab01c9b7cdf6c
[thirdparty/systemd.git] / src / core / socket.c
1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
2
3 #include <arpa/inet.h>
4 #include <fcntl.h>
5 #include <linux/sctp.h>
6 #include <mqueue.h>
7 #include <netinet/tcp.h>
8 #include <sys/stat.h>
9 #include <unistd.h>
10
11 #include "sd-bus.h"
12
13 #include "alloc-util.h"
14 #include "bpf-program.h"
15 #include "bus-error.h"
16 #include "copy.h"
17 #include "dbus-socket.h"
18 #include "dbus-unit.h"
19 #include "errno-list.h"
20 #include "errno-util.h"
21 #include "exit-status.h"
22 #include "extract-word.h"
23 #include "fd-util.h"
24 #include "fdset.h"
25 #include "format-util.h"
26 #include "fs-util.h"
27 #include "glyph-util.h"
28 #include "in-addr-util.h"
29 #include "io-util.h"
30 #include "ip-protocol-list.h"
31 #include "log.h"
32 #include "manager.h"
33 #include "mkdir-label.h"
34 #include "namespace-util.h"
35 #include "parse-util.h"
36 #include "path-util.h"
37 #include "pidfd-util.h"
38 #include "process-util.h"
39 #include "recurse-dir.h"
40 #include "selinux-util.h"
41 #include "serialize.h"
42 #include "service.h"
43 #include "set.h"
44 #include "siphash24.h"
45 #include "smack-util.h"
46 #include "socket.h"
47 #include "socket-netlink.h"
48 #include "special.h"
49 #include "string-table.h"
50 #include "string-util.h"
51 #include "strv.h"
52 #include "unit.h"
53 #include "unit-name.h"
54 #include "user-util.h"
55
56 typedef struct SocketPeer {
57 unsigned n_ref;
58
59 Socket *socket;
60 union sockaddr_union peer;
61 socklen_t peer_salen;
62 struct ucred peer_cred;
63 } SocketPeer;
64
65 static const UnitActiveState state_translation_table[_SOCKET_STATE_MAX] = {
66 [SOCKET_DEAD] = UNIT_INACTIVE,
67 [SOCKET_START_PRE] = UNIT_ACTIVATING,
68 [SOCKET_START_OPEN] = UNIT_ACTIVATING,
69 [SOCKET_START_CHOWN] = UNIT_ACTIVATING,
70 [SOCKET_START_POST] = UNIT_ACTIVATING,
71 [SOCKET_LISTENING] = UNIT_ACTIVE,
72 [SOCKET_RUNNING] = UNIT_ACTIVE,
73 [SOCKET_STOP_PRE] = UNIT_DEACTIVATING,
74 [SOCKET_STOP_PRE_SIGTERM] = UNIT_DEACTIVATING,
75 [SOCKET_STOP_PRE_SIGKILL] = UNIT_DEACTIVATING,
76 [SOCKET_STOP_POST] = UNIT_DEACTIVATING,
77 [SOCKET_FINAL_SIGTERM] = UNIT_DEACTIVATING,
78 [SOCKET_FINAL_SIGKILL] = UNIT_DEACTIVATING,
79 [SOCKET_FAILED] = UNIT_FAILED,
80 [SOCKET_CLEANING] = UNIT_MAINTENANCE,
81 };
82
83 static int socket_dispatch_io(sd_event_source *source, int fd, uint32_t revents, void *userdata);
84 static int socket_dispatch_timer(sd_event_source *source, usec_t usec, void *userdata);
85
86 static bool SOCKET_STATE_WITH_PROCESS(SocketState state) {
87 return IN_SET(state,
88 SOCKET_START_PRE,
89 SOCKET_START_CHOWN,
90 SOCKET_START_POST,
91 SOCKET_STOP_PRE,
92 SOCKET_STOP_PRE_SIGTERM,
93 SOCKET_STOP_PRE_SIGKILL,
94 SOCKET_STOP_POST,
95 SOCKET_FINAL_SIGTERM,
96 SOCKET_FINAL_SIGKILL,
97 SOCKET_CLEANING);
98 }
99
100 static bool SOCKET_SERVICE_IS_ACTIVE(Service *s, bool allow_finalize) {
101 assert(s);
102
103 /* If unit_active_state() reports inactive/failed then it's all good, otherwise we need to
104 * manually exclude SERVICE_AUTO_RESTART and SERVICE_AUTO_RESTART_QUEUED, in which cases
105 * the start job hasn't been enqueued/run, but are only placeholders in order to allow
106 * canceling auto restart. */
107
108 if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(UNIT(s))))
109 return false;
110
111 if (IN_SET(s->state, SERVICE_AUTO_RESTART, SERVICE_AUTO_RESTART_QUEUED))
112 return false;
113
114 if (allow_finalize && IN_SET(s->state, SERVICE_FINAL_SIGTERM, SERVICE_FINAL_SIGKILL, SERVICE_CLEANING))
115 return false;
116
117 return true;
118 }
119
120 static void socket_init(Unit *u) {
121 Socket *s = ASSERT_PTR(SOCKET(u));
122
123 assert(u->load_state == UNIT_STUB);
124
125 s->backlog = SOMAXCONN_DELUXE;
126 s->timeout_usec = u->manager->defaults.timeout_start_usec;
127 s->directory_mode = 0755;
128 s->socket_mode = 0666;
129
130 s->max_connections = 64;
131
132 s->pass_rights = true; /* defaults to enabled in kernel */
133 s->priority = -1;
134 s->ip_tos = -1;
135 s->ip_ttl = -1;
136 s->mark = -1;
137
138 s->exec_context.std_output = u->manager->defaults.std_output;
139 s->exec_context.std_error = u->manager->defaults.std_error;
140
141 s->control_pid = PIDREF_NULL;
142 s->control_command_id = _SOCKET_EXEC_COMMAND_INVALID;
143
144 s->trigger_limit = RATELIMIT_OFF;
145
146 s->poll_limit = RATELIMIT_OFF;
147 }
148
149 static void socket_unwatch_control_pid(Socket *s) {
150 assert(s);
151 unit_unwatch_pidref_done(UNIT(s), &s->control_pid);
152 }
153
154 static void socket_port_close_auxiliary_fds(SocketPort *p) {
155 assert(p);
156
157 close_many(p->auxiliary_fds, p->n_auxiliary_fds);
158 p->auxiliary_fds = mfree(p->auxiliary_fds);
159 p->n_auxiliary_fds = 0;
160 }
161
162 SocketPort* socket_port_free(SocketPort *p) {
163 if (!p)
164 return NULL;
165
166 sd_event_source_unref(p->event_source);
167
168 socket_port_close_auxiliary_fds(p);
169 safe_close(p->fd);
170 free(p->path);
171
172 return mfree(p);
173 }
174
175 void socket_free_ports(Socket *s) {
176 assert(s);
177
178 LIST_CLEAR(port, s->ports, socket_port_free);
179 }
180
181 static void socket_done(Unit *u) {
182 Socket *s = ASSERT_PTR(SOCKET(u));
183 SocketPeer *p;
184
185 socket_free_ports(s);
186
187 while ((p = set_steal_first(s->peers_by_address)))
188 p->socket = NULL;
189
190 s->peers_by_address = set_free(s->peers_by_address);
191
192 s->exec_runtime = exec_runtime_free(s->exec_runtime);
193
194 exec_command_free_array(s->exec_command, _SOCKET_EXEC_COMMAND_MAX);
195 s->control_command = NULL;
196
197 socket_unwatch_control_pid(s);
198
199 unit_ref_unset(&s->service);
200
201 s->tcp_congestion = mfree(s->tcp_congestion);
202 s->bind_to_device = mfree(s->bind_to_device);
203
204 s->smack = mfree(s->smack);
205 s->smack_ip_in = mfree(s->smack_ip_in);
206 s->smack_ip_out = mfree(s->smack_ip_out);
207
208 strv_free(s->symlinks);
209
210 s->user = mfree(s->user);
211 s->group = mfree(s->group);
212
213 s->fdname = mfree(s->fdname);
214
215 s->timer_event_source = sd_event_source_disable_unref(s->timer_event_source);
216 }
217
218 static int socket_arm_timer(Socket *s, bool relative, usec_t usec) {
219 assert(s);
220
221 return unit_arm_timer(UNIT(s), &s->timer_event_source, relative, usec, socket_dispatch_timer);
222 }
223
224 static bool have_non_accept_socket(Socket *s) {
225 assert(s);
226
227 if (!s->accept)
228 return true;
229
230 LIST_FOREACH(port, p, s->ports) {
231
232 if (p->type != SOCKET_SOCKET)
233 return true;
234
235 if (!socket_address_can_accept(&p->address))
236 return true;
237 }
238
239 return false;
240 }
241
242 static int socket_add_mount_dependencies(Socket *s) {
243 int r;
244
245 assert(s);
246
247 LIST_FOREACH(port, p, s->ports) {
248 const char *path = NULL;
249
250 if (p->type == SOCKET_SOCKET)
251 path = socket_address_get_path(&p->address);
252 else if (IN_SET(p->type, SOCKET_FIFO, SOCKET_SPECIAL, SOCKET_USB_FUNCTION))
253 path = p->path;
254
255 if (!path)
256 continue;
257
258 r = unit_add_mounts_for(UNIT(s), path, UNIT_DEPENDENCY_FILE, UNIT_MOUNT_REQUIRES);
259 if (r < 0)
260 return r;
261 }
262
263 return 0;
264 }
265
266 static int socket_add_device_dependencies(Socket *s) {
267 char *t;
268
269 assert(s);
270
271 if (!s->bind_to_device || streq(s->bind_to_device, "lo"))
272 return 0;
273
274 t = strjoina("/sys/subsystem/net/devices/", s->bind_to_device);
275 return unit_add_node_dependency(UNIT(s), t, UNIT_BINDS_TO, UNIT_DEPENDENCY_FILE);
276 }
277
278 static int socket_add_default_dependencies(Socket *s) {
279 int r;
280
281 assert(s);
282
283 if (!UNIT(s)->default_dependencies)
284 return 0;
285
286 r = unit_add_dependency_by_name(UNIT(s), UNIT_BEFORE, SPECIAL_SOCKETS_TARGET, true, UNIT_DEPENDENCY_DEFAULT);
287 if (r < 0)
288 return r;
289
290 if (MANAGER_IS_SYSTEM(UNIT(s)->manager)) {
291 r = unit_add_two_dependencies_by_name(UNIT(s), UNIT_AFTER, UNIT_REQUIRES, SPECIAL_SYSINIT_TARGET, true, UNIT_DEPENDENCY_DEFAULT);
292 if (r < 0)
293 return r;
294 }
295
296 return unit_add_two_dependencies_by_name(UNIT(s), UNIT_BEFORE, UNIT_CONFLICTS, SPECIAL_SHUTDOWN_TARGET, true, UNIT_DEPENDENCY_DEFAULT);
297 }
298
299 static bool socket_has_exec(Socket *s) {
300 assert(s);
301
302 FOREACH_ARRAY(i, s->exec_command, _SOCKET_EXEC_COMMAND_MAX)
303 if (*i)
304 return true;
305
306 return false;
307 }
308
309 static int socket_add_extras(Socket *s) {
310 Unit *u = UNIT(ASSERT_PTR(s));
311 int r;
312
313 /* Pick defaults for the trigger limit, if nothing was explicitly configured. We pick a relatively high limit
314 * in Accept=yes mode, and a lower limit for Accept=no. Reason: in Accept=yes mode we are invoking accept()
315 * ourselves before the trigger limit can hit, thus incoming connections are taken off the socket queue quickly
316 * and reliably. This is different for Accept=no, where the spawned service has to take the incoming traffic
317 * off the queues, which it might not necessarily do. Moreover, while Accept=no services are supposed to
318 * process whatever is queued in one go, and thus should normally never have to be started frequently. This is
319 * different for Accept=yes where each connection is processed by a new service instance, and thus frequent
320 * service starts are typical.
321 *
322 * For the poll limit we follow a similar rule, but use 3/4th of the trigger limit parameters, to
323 * trigger this earlier. */
324
325 if (s->trigger_limit.interval == USEC_INFINITY)
326 s->trigger_limit.interval = 2 * USEC_PER_SEC;
327 if (s->trigger_limit.burst == UINT_MAX)
328 s->trigger_limit.burst = s->accept ? 200 : 20;
329
330 if (s->poll_limit.interval == USEC_INFINITY)
331 s->poll_limit.interval = 2 * USEC_PER_SEC;
332 if (s->poll_limit.burst == UINT_MAX)
333 s->poll_limit.burst = s->accept ? 150 : 15;
334
335 if (have_non_accept_socket(s)) {
336
337 if (!UNIT_ISSET(s->service)) {
338 Unit *x;
339
340 r = unit_load_related_unit(u, ".service", &x);
341 if (r < 0)
342 return r;
343
344 unit_ref_set(&s->service, u, x);
345 }
346
347 r = unit_add_two_dependencies(u, UNIT_BEFORE, UNIT_TRIGGERS, UNIT_DEREF(s->service), true, UNIT_DEPENDENCY_IMPLICIT);
348 if (r < 0)
349 return r;
350 }
351
352 r = socket_add_mount_dependencies(s);
353 if (r < 0)
354 return r;
355
356 r = socket_add_device_dependencies(s);
357 if (r < 0)
358 return r;
359
360 r = unit_patch_contexts(u);
361 if (r < 0)
362 return r;
363
364 if (socket_has_exec(s)) {
365 r = unit_add_exec_dependencies(u, &s->exec_context);
366 if (r < 0)
367 return r;
368 }
369
370 r = unit_set_default_slice(u);
371 if (r < 0)
372 return r;
373
374 r = socket_add_default_dependencies(s);
375 if (r < 0)
376 return r;
377
378 return 0;
379 }
380
381 static const char* socket_find_symlink_target(Socket *s) {
382 const char *found = NULL;
383
384 assert(s);
385
386 LIST_FOREACH(port, p, s->ports) {
387 const char *f;
388
389 switch (p->type) {
390
391 case SOCKET_FIFO:
392 f = p->path;
393 break;
394
395 case SOCKET_SOCKET:
396 f = socket_address_get_path(&p->address);
397 break;
398
399 default:
400 f = NULL;
401 }
402
403 if (f) {
404 if (found)
405 return NULL;
406
407 found = f;
408 }
409 }
410
411 return found;
412 }
413
414 static int socket_verify(Socket *s) {
415 assert(s);
416 assert(UNIT(s)->load_state == UNIT_LOADED);
417
418 if (!s->ports)
419 return log_unit_error_errno(UNIT(s), SYNTHETIC_ERRNO(ENOEXEC), "Unit has no Listen setting (ListenStream=, ListenDatagram=, ListenFIFO=, ...). Refusing.");
420
421 if (s->max_connections <= 0)
422 return log_unit_error_errno(UNIT(s), SYNTHETIC_ERRNO(ENOEXEC), "MaxConnection= setting too small. Refusing.");
423
424 if (s->accept && have_non_accept_socket(s))
425 return log_unit_error_errno(UNIT(s), SYNTHETIC_ERRNO(ENOEXEC), "Unit configured for accepting sockets, but sockets are non-accepting. Refusing.");
426
427 if (s->accept && UNIT_ISSET(s->service))
428 return log_unit_error_errno(UNIT(s), SYNTHETIC_ERRNO(ENOEXEC), "Explicit service configuration for accepting socket units not supported. Refusing.");
429
430 if (!strv_isempty(s->symlinks) && !socket_find_symlink_target(s))
431 return log_unit_error_errno(UNIT(s), SYNTHETIC_ERRNO(ENOEXEC), "Unit has symlinks set but none or more than one node in the file system. Refusing.");
432
433 return 0;
434 }
435
436 static void peer_address_hash_func(const SocketPeer *s, struct siphash *state) {
437 assert(s);
438
439 if (s->peer.sa.sa_family == AF_INET)
440 siphash24_compress_typesafe(s->peer.in.sin_addr, state);
441 else if (s->peer.sa.sa_family == AF_INET6)
442 siphash24_compress_typesafe(s->peer.in6.sin6_addr, state);
443 else if (s->peer.sa.sa_family == AF_VSOCK)
444 siphash24_compress_typesafe(s->peer.vm.svm_cid, state);
445 else if (s->peer.sa.sa_family == AF_UNIX)
446 siphash24_compress_typesafe(s->peer_cred.uid, state);
447 else
448 assert_not_reached();
449 }
450
451 static int peer_address_compare_func(const SocketPeer *x, const SocketPeer *y) {
452 int r;
453
454 r = CMP(x->peer.sa.sa_family, y->peer.sa.sa_family);
455 if (r != 0)
456 return r;
457
458 switch (x->peer.sa.sa_family) {
459 case AF_INET:
460 return memcmp(&x->peer.in.sin_addr, &y->peer.in.sin_addr, sizeof(x->peer.in.sin_addr));
461 case AF_INET6:
462 return memcmp(&x->peer.in6.sin6_addr, &y->peer.in6.sin6_addr, sizeof(x->peer.in6.sin6_addr));
463 case AF_VSOCK:
464 return CMP(x->peer.vm.svm_cid, y->peer.vm.svm_cid);
465 case AF_UNIX:
466 return CMP(x->peer_cred.uid, y->peer_cred.uid);
467 }
468 assert_not_reached();
469 }
470
471 DEFINE_PRIVATE_HASH_OPS(peer_address_hash_ops, SocketPeer, peer_address_hash_func, peer_address_compare_func);
472
473 static int socket_load(Unit *u) {
474 Socket *s = ASSERT_PTR(SOCKET(u));
475 int r;
476
477 assert(u->load_state == UNIT_STUB);
478
479 r = unit_load_fragment_and_dropin(u, true);
480 if (r < 0)
481 return r;
482
483 if (u->load_state != UNIT_LOADED)
484 return 0;
485
486 /* This is a new unit? Then let's add in some extras */
487 r = socket_add_extras(s);
488 if (r < 0)
489 return r;
490
491 return socket_verify(s);
492 }
493
494 static SocketPeer* socket_peer_dup(const SocketPeer *q) {
495 SocketPeer *p;
496
497 assert(q);
498
499 p = new(SocketPeer, 1);
500 if (!p)
501 return NULL;
502
503 *p = (SocketPeer) {
504 .n_ref = 1,
505 .peer = q->peer,
506 .peer_salen = q->peer_salen,
507 .peer_cred = q->peer_cred,
508 };
509
510 return p;
511 }
512
513 static SocketPeer* socket_peer_free(SocketPeer *p) {
514 assert(p);
515
516 if (p->socket)
517 set_remove(p->socket->peers_by_address, p);
518
519 return mfree(p);
520 }
521
522 DEFINE_TRIVIAL_REF_UNREF_FUNC(SocketPeer, socket_peer, socket_peer_free);
523
524 int socket_acquire_peer(Socket *s, int fd, SocketPeer **ret) {
525 _cleanup_(socket_peer_unrefp) SocketPeer *remote = NULL;
526 SocketPeer key = {
527 .peer_salen = sizeof(union sockaddr_union),
528 .peer_cred = UCRED_INVALID,
529 }, *i;
530 int r;
531
532 assert(s);
533 assert(fd >= 0);
534 assert(ret);
535
536 if (getpeername(fd, &key.peer.sa, &key.peer_salen) < 0)
537 return log_unit_error_errno(UNIT(s), errno, "getpeername() failed: %m");
538
539 switch (key.peer.sa.sa_family) {
540 case AF_INET:
541 case AF_INET6:
542 case AF_VSOCK:
543 break;
544
545 case AF_UNIX:
546 r = getpeercred(fd, &key.peer_cred);
547 if (r < 0)
548 return log_unit_error_errno(UNIT(s), r, "Failed to get peer credentials of socket: %m");
549 break;
550
551 default:
552 *ret = NULL;
553 return 0;
554 }
555
556 i = set_get(s->peers_by_address, &key);
557 if (i) {
558 *ret = socket_peer_ref(i);
559 return 1;
560 }
561
562 remote = socket_peer_dup(&key);
563 if (!remote)
564 return log_oom();
565
566 r = set_ensure_put(&s->peers_by_address, &peer_address_hash_ops, remote);
567 if (r < 0)
568 return log_unit_error_errno(UNIT(s), r, "Failed to insert peer info into hash table: %m");
569
570 remote->socket = s;
571
572 *ret = TAKE_PTR(remote);
573 return 1;
574 }
575
576 static const char* listen_lookup(int family, int type) {
577
578 if (family == AF_NETLINK)
579 return "ListenNetlink";
580
581 if (type == SOCK_STREAM)
582 return "ListenStream";
583 else if (type == SOCK_DGRAM)
584 return "ListenDatagram";
585 else if (type == SOCK_SEQPACKET)
586 return "ListenSequentialPacket";
587
588 assert_not_reached();
589 }
590
591 static void socket_dump(Unit *u, FILE *f, const char *prefix) {
592 Socket *s = ASSERT_PTR(SOCKET(u));
593 const char *prefix2, *str;
594
595 assert(f);
596
597 prefix = strempty(prefix);
598 prefix2 = strjoina(prefix, "\t");
599
600 fprintf(f,
601 "%sSocket State: %s\n"
602 "%sResult: %s\n"
603 "%sClean Result: %s\n"
604 "%sBindIPv6Only: %s\n"
605 "%sBacklog: %u\n"
606 "%sSocketMode: %04o\n"
607 "%sDirectoryMode: %04o\n"
608 "%sKeepAlive: %s\n"
609 "%sNoDelay: %s\n"
610 "%sFreeBind: %s\n"
611 "%sTransparent: %s\n"
612 "%sBroadcast: %s\n"
613 "%sPassCredentials: %s\n"
614 "%sPassPIDFD: %s\n"
615 "%sPassSecurity: %s\n"
616 "%sPassPacketInfo: %s\n"
617 "%sAcceptFileDescriptors: %s\n"
618 "%sTCPCongestion: %s\n"
619 "%sRemoveOnStop: %s\n"
620 "%sWritable: %s\n"
621 "%sFileDescriptorName: %s\n"
622 "%sPassFileDescriptorsToExec: %s\n"
623 "%sSELinuxContextFromNet: %s\n",
624 prefix, socket_state_to_string(s->state),
625 prefix, socket_result_to_string(s->result),
626 prefix, socket_result_to_string(s->clean_result),
627 prefix, socket_address_bind_ipv6_only_to_string(s->bind_ipv6_only),
628 prefix, s->backlog,
629 prefix, s->socket_mode,
630 prefix, s->directory_mode,
631 prefix, yes_no(s->keep_alive),
632 prefix, yes_no(s->no_delay),
633 prefix, yes_no(s->free_bind),
634 prefix, yes_no(s->transparent),
635 prefix, yes_no(s->broadcast),
636 prefix, yes_no(s->pass_cred),
637 prefix, yes_no(s->pass_pidfd),
638 prefix, yes_no(s->pass_sec),
639 prefix, yes_no(s->pass_pktinfo),
640 prefix, yes_no(s->pass_rights),
641 prefix, strna(s->tcp_congestion),
642 prefix, yes_no(s->remove_on_stop),
643 prefix, yes_no(s->writable),
644 prefix, socket_fdname(s),
645 prefix, yes_no(s->pass_fds_to_exec),
646 prefix, yes_no(s->selinux_context_from_net));
647
648 if (s->timestamping != SOCKET_TIMESTAMPING_OFF)
649 fprintf(f,
650 "%sTimestamping: %s\n",
651 prefix, socket_timestamping_to_string(s->timestamping));
652
653 if (pidref_is_set(&s->control_pid))
654 fprintf(f,
655 "%sControl PID: "PID_FMT"\n",
656 prefix, s->control_pid.pid);
657
658 if (s->bind_to_device)
659 fprintf(f,
660 "%sBindToDevice: %s\n",
661 prefix, s->bind_to_device);
662
663 if (s->accept)
664 fprintf(f,
665 "%sAccepted: %u\n"
666 "%sNConnections: %u\n"
667 "%sMaxConnections: %u\n"
668 "%sMaxConnectionsPerSource: %u\n",
669 prefix, s->n_accepted,
670 prefix, s->n_connections,
671 prefix, s->max_connections,
672 prefix, s->max_connections_per_source);
673 else
674 fprintf(f,
675 "%sFlushPending: %s\n",
676 prefix, yes_no(s->flush_pending));
677
678 if (s->priority >= 0)
679 fprintf(f,
680 "%sPriority: %i\n",
681 prefix, s->priority);
682
683 if (s->receive_buffer > 0)
684 fprintf(f,
685 "%sReceiveBuffer: %zu\n",
686 prefix, s->receive_buffer);
687
688 if (s->send_buffer > 0)
689 fprintf(f,
690 "%sSendBuffer: %zu\n",
691 prefix, s->send_buffer);
692
693 if (s->ip_tos >= 0)
694 fprintf(f,
695 "%sIPTOS: %i\n",
696 prefix, s->ip_tos);
697
698 if (s->ip_ttl >= 0)
699 fprintf(f,
700 "%sIPTTL: %i\n",
701 prefix, s->ip_ttl);
702
703 if (s->pipe_size > 0)
704 fprintf(f,
705 "%sPipeSize: %zu\n",
706 prefix, s->pipe_size);
707
708 if (s->mark >= 0)
709 fprintf(f,
710 "%sMark: %i\n",
711 prefix, s->mark);
712
713 if (s->mq_maxmsg > 0)
714 fprintf(f,
715 "%sMessageQueueMaxMessages: %li\n",
716 prefix, s->mq_maxmsg);
717
718 if (s->mq_msgsize > 0)
719 fprintf(f,
720 "%sMessageQueueMessageSize: %li\n",
721 prefix, s->mq_msgsize);
722
723 if (s->reuse_port)
724 fprintf(f,
725 "%sReusePort: %s\n",
726 prefix, yes_no(s->reuse_port));
727
728 if (s->smack)
729 fprintf(f,
730 "%sSmackLabel: %s\n",
731 prefix, s->smack);
732
733 if (s->smack_ip_in)
734 fprintf(f,
735 "%sSmackLabelIPIn: %s\n",
736 prefix, s->smack_ip_in);
737
738 if (s->smack_ip_out)
739 fprintf(f,
740 "%sSmackLabelIPOut: %s\n",
741 prefix, s->smack_ip_out);
742
743 if (!isempty(s->user) || !isempty(s->group))
744 fprintf(f,
745 "%sSocketUser: %s\n"
746 "%sSocketGroup: %s\n",
747 prefix, strna(s->user),
748 prefix, strna(s->group));
749
750 if (timestamp_is_set(s->keep_alive_time))
751 fprintf(f,
752 "%sKeepAliveTimeSec: %s\n",
753 prefix, FORMAT_TIMESPAN(s->keep_alive_time, USEC_PER_SEC));
754
755 if (s->keep_alive_interval > 0)
756 fprintf(f,
757 "%sKeepAliveIntervalSec: %s\n",
758 prefix, FORMAT_TIMESPAN(s->keep_alive_interval, USEC_PER_SEC));
759
760 if (s->keep_alive_cnt > 0)
761 fprintf(f,
762 "%sKeepAliveProbes: %u\n",
763 prefix, s->keep_alive_cnt);
764
765 if (s->defer_accept > 0)
766 fprintf(f,
767 "%sDeferAcceptSec: %s\n",
768 prefix, FORMAT_TIMESPAN(s->defer_accept, USEC_PER_SEC));
769
770 LIST_FOREACH(port, p, s->ports) {
771
772 switch (p->type) {
773 case SOCKET_SOCKET: {
774 _cleanup_free_ char *k = NULL;
775 int r;
776
777 r = socket_address_print(&p->address, &k);
778 if (r < 0) {
779 errno = -r;
780 fprintf(f, "%s%s: %m\n", prefix, listen_lookup(socket_address_family(&p->address), p->address.type));
781 } else
782 fprintf(f, "%s%s: %s\n", prefix, listen_lookup(socket_address_family(&p->address), p->address.type), k);
783 break;
784 }
785 case SOCKET_SPECIAL:
786 fprintf(f, "%sListenSpecial: %s\n", prefix, p->path);
787 break;
788 case SOCKET_USB_FUNCTION:
789 fprintf(f, "%sListenUSBFunction: %s\n", prefix, p->path);
790 break;
791 case SOCKET_MQUEUE:
792 fprintf(f, "%sListenMessageQueue: %s\n", prefix, p->path);
793 break;
794 default:
795 fprintf(f, "%sListenFIFO: %s\n", prefix, p->path);
796 }
797 }
798
799 fprintf(f,
800 "%sTriggerLimitIntervalSec: %s\n"
801 "%sTriggerLimitBurst: %u\n"
802 "%sPollLimitIntervalSec: %s\n"
803 "%sPollLimitBurst: %u\n",
804 prefix, FORMAT_TIMESPAN(s->trigger_limit.interval, USEC_PER_SEC),
805 prefix, s->trigger_limit.burst,
806 prefix, FORMAT_TIMESPAN(s->poll_limit.interval, USEC_PER_SEC),
807 prefix, s->poll_limit.burst);
808
809 str = ip_protocol_to_name(s->socket_protocol);
810 if (str)
811 fprintf(f, "%sSocketProtocol: %s\n", prefix, str);
812
813 if (!strv_isempty(s->symlinks)) {
814 fprintf(f, "%sSymlinks:", prefix);
815 STRV_FOREACH(q, s->symlinks)
816 fprintf(f, " %s", *q);
817
818 fprintf(f, "\n");
819 }
820
821 fprintf(f,
822 "%sTimeoutSec: %s\n",
823 prefix, FORMAT_TIMESPAN(s->timeout_usec, USEC_PER_SEC));
824
825 exec_context_dump(&s->exec_context, f, prefix);
826 kill_context_dump(&s->kill_context, f, prefix);
827
828 for (SocketExecCommand c = 0; c < _SOCKET_EXEC_COMMAND_MAX; c++) {
829 if (!s->exec_command[c])
830 continue;
831
832 fprintf(f, "%s%s %s:\n",
833 prefix, glyph(GLYPH_ARROW_RIGHT), socket_exec_command_to_string(c));
834
835 exec_command_dump_list(s->exec_command[c], f, prefix2);
836 }
837
838 cgroup_context_dump(UNIT(s), f, prefix);
839 }
840
841 static int instance_from_socket(
842 int fd,
843 unsigned nr,
844 char **ret) {
845
846 union sockaddr_union local, remote;
847 socklen_t l;
848 int r;
849
850 assert(fd >= 0);
851 assert(ret);
852
853 l = sizeof(local);
854 if (getsockname(fd, &local.sa, &l) < 0)
855 return -errno;
856
857 l = sizeof(remote);
858 if (getpeername(fd, &remote.sa, &l) < 0)
859 return -errno;
860
861 uint64_t cookie;
862 r = socket_get_cookie(fd, &cookie);
863 if (r < 0)
864 return r;
865
866 char *s;
867
868 switch (local.sa.sa_family) {
869
870 case AF_INET: {
871 uint32_t
872 a = be32toh(local.in.sin_addr.s_addr),
873 b = be32toh(remote.in.sin_addr.s_addr);
874
875 if (asprintf(&s,
876 "%u-%" PRIu64 "-%u.%u.%u.%u:%u-%u.%u.%u.%u:%u",
877 nr,
878 cookie,
879 a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF,
880 be16toh(local.in.sin_port),
881 b >> 24, (b >> 16) & 0xFF, (b >> 8) & 0xFF, b & 0xFF,
882 be16toh(remote.in.sin_port)) < 0)
883 return -ENOMEM;
884
885 break;
886 }
887
888 case AF_INET6: {
889 static const unsigned char ipv4_prefix[] = {
890 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF, 0xFF
891 };
892
893 if (memcmp(&local.in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0 &&
894 memcmp(&remote.in6.sin6_addr, ipv4_prefix, sizeof(ipv4_prefix)) == 0) {
895 const uint8_t
896 *a = local.in6.sin6_addr.s6_addr+12,
897 *b = remote.in6.sin6_addr.s6_addr+12;
898
899 if (asprintf(&s,
900 "%u-%" PRIu64 "-%u.%u.%u.%u:%u-%u.%u.%u.%u:%u",
901 nr,
902 cookie,
903 a[0], a[1], a[2], a[3],
904 be16toh(local.in6.sin6_port),
905 b[0], b[1], b[2], b[3],
906 be16toh(remote.in6.sin6_port)) < 0)
907 return -ENOMEM;
908 } else {
909 if (asprintf(&s,
910 "%u-%" PRIu64 "-%s:%u-%s:%u",
911 nr,
912 cookie,
913 IN6_ADDR_TO_STRING(&local.in6.sin6_addr),
914 be16toh(local.in6.sin6_port),
915 IN6_ADDR_TO_STRING(&remote.in6.sin6_addr),
916 be16toh(remote.in6.sin6_port)) < 0)
917 return -ENOMEM;
918 }
919
920 break;
921 }
922
923 case AF_UNIX: {
924 struct ucred ucred;
925
926 r = getpeercred(fd, &ucred);
927 if (r >= 0) {
928 _cleanup_close_ int pidfd = getpeerpidfd(fd);
929 uint64_t pidfd_id;
930
931 if (pidfd >= 0 && pidfd_get_inode_id(pidfd, &pidfd_id) >= 0)
932 r = asprintf(&s, "%u-%" PRIu64 "-" PID_FMT "_%" PRIu64 "-" UID_FMT,
933 nr, cookie, ucred.pid, pidfd_id, ucred.uid);
934 else
935 r = asprintf(&s, "%u-%" PRIu64 "-" PID_FMT "-" UID_FMT,
936 nr, cookie, ucred.pid, ucred.uid);
937 if (r < 0)
938 return -ENOMEM;
939 } else if (r == -ENODATA) {
940 /* This handles the case where somebody is connecting from another pid/uid namespace
941 * (e.g. from outside of our container). */
942 if (asprintf(&s,
943 "%u-%" PRIu64 "-unknown",
944 nr,
945 cookie) < 0)
946 return -ENOMEM;
947 } else
948 return r;
949
950 break;
951 }
952
953 case AF_VSOCK:
954 if (asprintf(&s,
955 "%u-%" PRIu64 "-%u:%u-%u:%u",
956 nr,
957 cookie,
958 local.vm.svm_cid, local.vm.svm_port,
959 remote.vm.svm_cid, remote.vm.svm_port) < 0)
960 return -ENOMEM;
961
962 break;
963
964 default:
965 assert_not_reached();
966 }
967
968 *ret = s;
969 return 0;
970 }
971
972 static void socket_close_fds(Socket *s) {
973 assert(s);
974
975 LIST_FOREACH(port, p, s->ports) {
976 bool was_open = p->fd >= 0;
977
978 p->event_source = sd_event_source_disable_unref(p->event_source);
979 p->fd = safe_close(p->fd);
980 socket_port_close_auxiliary_fds(p);
981
982 /* One little note: we should normally not delete any sockets in the file system here! After all some
983 * other process we spawned might still have a reference of this fd and wants to continue to use
984 * it. Therefore we normally delete sockets in the file system before we create a new one, not after we
985 * stopped using one! That all said, if the user explicitly requested this, we'll delete them here
986 * anyway, but only then. */
987
988 if (!was_open || !s->remove_on_stop)
989 continue;
990
991 switch (p->type) {
992
993 case SOCKET_FIFO:
994 (void) unlink(p->path);
995 break;
996
997 case SOCKET_MQUEUE:
998 (void) mq_unlink(p->path);
999 break;
1000
1001 case SOCKET_SOCKET:
1002 (void) socket_address_unlink(&p->address);
1003 break;
1004
1005 default:
1006 ;
1007 }
1008 }
1009
1010 if (s->remove_on_stop)
1011 STRV_FOREACH(i, s->symlinks)
1012 (void) unlink(*i);
1013
1014 /* Note that we don't return NULL here, since s has not been freed. */
1015 }
1016
1017 DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(Socket*, socket_close_fds, NULL);
1018
1019 #define SOCKET_OPTION_WARNING_FORMAT_STR "Failed to set %s socket option, ignoring: %m"
1020 #define log_socket_option_warning_errno(s, error, option) \
1021 log_unit_warning_errno(UNIT(s), (error), SOCKET_OPTION_WARNING_FORMAT_STR, STRINGIFY(option))
1022
1023 static void socket_apply_socket_options(Socket *s, SocketPort *p, int fd) {
1024 int r;
1025
1026 assert(s);
1027 assert(p);
1028 assert(fd >= 0);
1029
1030 if (s->keep_alive) {
1031 r = setsockopt_int(fd, SOL_SOCKET, SO_KEEPALIVE, true);
1032 if (r < 0)
1033 log_socket_option_warning_errno(s, r, SO_KEEPALIVE);
1034 }
1035
1036 if (timestamp_is_set(s->keep_alive_time)) {
1037 r = setsockopt_int(fd, SOL_TCP, TCP_KEEPIDLE, s->keep_alive_time / USEC_PER_SEC);
1038 if (r < 0)
1039 log_socket_option_warning_errno(s, r, TCP_KEEPIDLE);
1040 }
1041
1042 if (s->keep_alive_interval > 0) {
1043 r = setsockopt_int(fd, SOL_TCP, TCP_KEEPINTVL, s->keep_alive_interval / USEC_PER_SEC);
1044 if (r < 0)
1045 log_socket_option_warning_errno(s, r, TCP_KEEPINTVL);
1046 }
1047
1048 if (s->keep_alive_cnt > 0) {
1049 r = setsockopt_int(fd, SOL_TCP, TCP_KEEPCNT, s->keep_alive_cnt);
1050 if (r < 0)
1051 log_socket_option_warning_errno(s, r, TCP_KEEPCNT);
1052 }
1053
1054 if (s->defer_accept > 0) {
1055 r = setsockopt_int(fd, SOL_TCP, TCP_DEFER_ACCEPT, s->defer_accept / USEC_PER_SEC);
1056 if (r < 0)
1057 log_socket_option_warning_errno(s, r, TCP_DEFER_ACCEPT);
1058 }
1059
1060 if (s->no_delay) {
1061 if (s->socket_protocol == IPPROTO_SCTP) {
1062 r = setsockopt_int(fd, SOL_SCTP, SCTP_NODELAY, true);
1063 if (r < 0)
1064 log_socket_option_warning_errno(s, r, SCTP_NODELAY);
1065 } else {
1066 r = setsockopt_int(fd, SOL_TCP, TCP_NODELAY, true);
1067 if (r < 0)
1068 log_socket_option_warning_errno(s, r, TCP_NODELAY);
1069 }
1070 }
1071
1072 if (s->broadcast) {
1073 r = setsockopt_int(fd, SOL_SOCKET, SO_BROADCAST, true);
1074 if (r < 0)
1075 log_socket_option_warning_errno(s, r, SO_BROADCAST);
1076 }
1077
1078 if (s->pass_cred) {
1079 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
1080 if (r < 0)
1081 log_socket_option_warning_errno(s, r, SO_PASSCRED);
1082 }
1083
1084 if (s->pass_pidfd) {
1085 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSPIDFD, true);
1086 if (r < 0)
1087 log_unit_full_errno(UNIT(s), ERRNO_IS_NEG_NOT_SUPPORTED(r) ? LOG_DEBUG : LOG_WARNING, r,
1088 SOCKET_OPTION_WARNING_FORMAT_STR, "SO_PASSPIDFD");
1089 }
1090
1091 if (s->pass_sec) {
1092 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSSEC, true);
1093 if (r < 0)
1094 log_unit_full_errno(UNIT(s), ERRNO_IS_NEG_NOT_SUPPORTED(r) ? LOG_DEBUG : LOG_WARNING, r,
1095 SOCKET_OPTION_WARNING_FORMAT_STR, "SO_PASSSEC");
1096 }
1097
1098 if (s->pass_pktinfo) {
1099 r = socket_set_recvpktinfo(fd, socket_address_family(&p->address), true);
1100 if (r < 0)
1101 log_unit_warning_errno(UNIT(s), r, SOCKET_OPTION_WARNING_FORMAT_STR, "packet info");
1102 }
1103
1104 if (!s->pass_rights) {
1105 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSRIGHTS, false);
1106 if (r < 0)
1107 log_unit_full_errno(UNIT(s), ERRNO_IS_NEG_NOT_SUPPORTED(r) ? LOG_DEBUG : LOG_WARNING, r,
1108 SOCKET_OPTION_WARNING_FORMAT_STR, "SO_PASSRIGHTS");
1109 }
1110
1111 if (s->timestamping != SOCKET_TIMESTAMPING_OFF) {
1112 r = setsockopt_int(fd, SOL_SOCKET,
1113 s->timestamping == SOCKET_TIMESTAMPING_NS ? SO_TIMESTAMPNS : SO_TIMESTAMP,
1114 true);
1115 if (r < 0)
1116 log_unit_warning_errno(UNIT(s), r, SOCKET_OPTION_WARNING_FORMAT_STR, "timestamping");
1117 }
1118
1119 if (s->priority >= 0) {
1120 r = setsockopt_int(fd, SOL_SOCKET, SO_PRIORITY, s->priority);
1121 if (r < 0)
1122 log_socket_option_warning_errno(s, r, SO_PRIORITY);
1123 }
1124
1125 if (s->receive_buffer > 0) {
1126 r = fd_set_rcvbuf(fd, s->receive_buffer, false);
1127 if (r < 0)
1128 log_unit_full_errno(UNIT(s), ERRNO_IS_NEG_PRIVILEGE(r) ? LOG_DEBUG : LOG_WARNING, r,
1129 SOCKET_OPTION_WARNING_FORMAT_STR, "SO_RCVBUF/SO_RCVBUFFORCE");
1130 }
1131
1132 if (s->send_buffer > 0) {
1133 r = fd_set_sndbuf(fd, s->send_buffer, false);
1134 if (r < 0)
1135 log_unit_full_errno(UNIT(s), ERRNO_IS_NEG_PRIVILEGE(r) ? LOG_DEBUG : LOG_WARNING, r,
1136 SOCKET_OPTION_WARNING_FORMAT_STR, "SO_SNDBUF/SO_SNDBUFFORCE");
1137 }
1138
1139 if (s->mark >= 0) {
1140 r = setsockopt_int(fd, SOL_SOCKET, SO_MARK, s->mark);
1141 if (r < 0)
1142 log_socket_option_warning_errno(s, r, SO_MARK);
1143 }
1144
1145 if (s->ip_tos >= 0) {
1146 r = setsockopt_int(fd, IPPROTO_IP, IP_TOS, s->ip_tos);
1147 if (r < 0)
1148 log_socket_option_warning_errno(s, r, IP_TOS);
1149 }
1150
1151 if (s->ip_ttl >= 0) {
1152 r = socket_set_ttl(fd, socket_address_family(&p->address), s->ip_ttl);
1153 if (r < 0)
1154 log_unit_warning_errno(UNIT(s), r, SOCKET_OPTION_WARNING_FORMAT_STR, "IP_TTL/IPV6_UNICAST_HOPS");
1155 }
1156
1157 if (s->tcp_congestion)
1158 if (setsockopt(fd, SOL_TCP, TCP_CONGESTION, s->tcp_congestion, strlen(s->tcp_congestion)+1) < 0)
1159 log_socket_option_warning_errno(s, errno, TCP_CONGESTION);
1160
1161 if (s->smack_ip_in) {
1162 r = mac_smack_apply_fd(fd, SMACK_ATTR_IPIN, s->smack_ip_in);
1163 if (r < 0)
1164 log_unit_error_errno(UNIT(s), r, "Failed to apply SMACK label for IP input, ignoring: %m");
1165 }
1166
1167 if (s->smack_ip_out) {
1168 r = mac_smack_apply_fd(fd, SMACK_ATTR_IPOUT, s->smack_ip_out);
1169 if (r < 0)
1170 log_unit_error_errno(UNIT(s), r, "Failed to apply SMACK label for IP output, ignoring: %m");
1171 }
1172 }
1173
1174 static void socket_apply_fifo_options(Socket *s, int fd) {
1175 int r;
1176
1177 assert(s);
1178 assert(fd >= 0);
1179
1180 if (s->pipe_size > 0)
1181 if (fcntl(fd, F_SETPIPE_SZ, s->pipe_size) < 0)
1182 log_unit_warning_errno(UNIT(s), errno, "Setting pipe size failed, ignoring: %m");
1183
1184 if (s->smack) {
1185 r = mac_smack_apply_fd(fd, SMACK_ATTR_ACCESS, s->smack);
1186 if (r < 0)
1187 log_unit_error_errno(UNIT(s), r, "SMACK relabelling failed, ignoring: %m");
1188 }
1189 }
1190
1191 static int fifo_address_create(
1192 const char *path,
1193 mode_t directory_mode,
1194 mode_t socket_mode) {
1195
1196 _cleanup_close_ int fd = -EBADF;
1197 mode_t old_mask;
1198 struct stat st;
1199 int r;
1200
1201 assert(path);
1202
1203 (void) mkdir_parents_label(path, directory_mode);
1204
1205 r = mac_selinux_create_file_prepare(path, S_IFIFO);
1206 if (r < 0)
1207 return r;
1208
1209 /* Enforce the right access mode for the fifo */
1210 old_mask = umask(~socket_mode);
1211
1212 /* Include the original umask in our mask */
1213 (void) umask(~socket_mode | old_mask);
1214
1215 r = mkfifo(path, socket_mode);
1216 (void) umask(old_mask);
1217
1218 if (r < 0 && errno != EEXIST) {
1219 r = -errno;
1220 goto fail;
1221 }
1222
1223 fd = open(path, O_RDWR | O_CLOEXEC | O_NOCTTY | O_NONBLOCK | O_NOFOLLOW);
1224 if (fd < 0) {
1225 r = -errno;
1226 goto fail;
1227 }
1228
1229 mac_selinux_create_file_clear();
1230
1231 if (fstat(fd, &st) < 0) {
1232 r = -errno;
1233 goto fail;
1234 }
1235
1236 if (!S_ISFIFO(st.st_mode) ||
1237 (st.st_mode & 0777) != (socket_mode & ~old_mask) ||
1238 st.st_uid != getuid() ||
1239 st.st_gid != getgid()) {
1240 r = -EEXIST;
1241 goto fail;
1242 }
1243
1244 return TAKE_FD(fd);
1245
1246 fail:
1247 mac_selinux_create_file_clear();
1248 return r;
1249 }
1250
1251 static int special_address_create(const char *path, bool writable) {
1252 _cleanup_close_ int fd = -EBADF;
1253 struct stat st;
1254
1255 assert(path);
1256
1257 fd = open(path, (writable ? O_RDWR : O_RDONLY)|O_CLOEXEC|O_NOCTTY|O_NONBLOCK|O_NOFOLLOW);
1258 if (fd < 0)
1259 return -errno;
1260
1261 if (fstat(fd, &st) < 0)
1262 return -errno;
1263
1264 /* Check whether this is a /proc, /sys or /dev file or char device */
1265 if (!S_ISREG(st.st_mode) && !S_ISCHR(st.st_mode))
1266 return -EEXIST;
1267
1268 return TAKE_FD(fd);
1269 }
1270
1271 static int usbffs_address_create_at(int dfd, const char *name) {
1272 _cleanup_close_ int fd = -EBADF;
1273 struct stat st;
1274
1275 assert(dfd >= 0);
1276 assert(name);
1277
1278 fd = openat(dfd, name, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK|O_NOFOLLOW);
1279 if (fd < 0)
1280 return -errno;
1281
1282 if (fstat(fd, &st) < 0)
1283 return -errno;
1284
1285 /* Check whether this is a regular file (ffs endpoint) */
1286 if (!S_ISREG(st.st_mode))
1287 return -EEXIST;
1288
1289 return TAKE_FD(fd);
1290 }
1291
1292 static int mq_address_create(
1293 const char *path,
1294 mode_t mq_mode,
1295 long maxmsg,
1296 long msgsize) {
1297
1298 _cleanup_close_ int fd = -EBADF;
1299 struct stat st;
1300 mode_t old_mask;
1301 struct mq_attr _attr, *attr = NULL;
1302
1303 assert(path);
1304
1305 if (maxmsg > 0 && msgsize > 0) {
1306 _attr = (struct mq_attr) {
1307 .mq_flags = O_NONBLOCK,
1308 .mq_maxmsg = maxmsg,
1309 .mq_msgsize = msgsize,
1310 };
1311 attr = &_attr;
1312 }
1313
1314 /* Enforce the right access mode for the mq */
1315 old_mask = umask(~mq_mode);
1316
1317 /* Include the original umask in our mask */
1318 (void) umask(~mq_mode | old_mask);
1319 fd = mq_open(path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_CREAT, mq_mode, attr);
1320 (void) umask(old_mask);
1321
1322 if (fd < 0)
1323 return -errno;
1324
1325 if (fstat(fd, &st) < 0)
1326 return -errno;
1327
1328 if ((st.st_mode & 0777) != (mq_mode & ~old_mask) ||
1329 st.st_uid != getuid() ||
1330 st.st_gid != getgid())
1331 return -EEXIST;
1332
1333 return TAKE_FD(fd);
1334 }
1335
1336 static int socket_symlink(Socket *s) {
1337 const char *p;
1338 int r;
1339
1340 assert(s);
1341
1342 p = socket_find_symlink_target(s);
1343 if (!p)
1344 return 0;
1345
1346 STRV_FOREACH(i, s->symlinks) {
1347 (void) mkdir_parents_label(*i, s->directory_mode);
1348
1349 r = symlink_idempotent(p, *i, false);
1350 if (r == -EEXIST && s->remove_on_stop) {
1351 /* If there's already something where we want to create the symlink, and the destructive
1352 * RemoveOnStop= mode is set, then we might as well try to remove what already exists and try
1353 * again. */
1354
1355 if (unlink(*i) >= 0)
1356 r = symlink_idempotent(p, *i, false);
1357 }
1358 if (r < 0)
1359 log_unit_warning_errno(UNIT(s), r, "Failed to create symlink %s %s %s, ignoring: %m",
1360 p, glyph(GLYPH_ARROW_RIGHT), *i);
1361 }
1362
1363 return 0;
1364 }
1365
1366 static int usbffs_write_descs(int fd, Service *s) {
1367 int r;
1368
1369 assert(fd >= 0);
1370 assert(s);
1371
1372 if (!s->usb_function_descriptors || !s->usb_function_strings)
1373 return -EINVAL;
1374
1375 r = copy_file_fd(s->usb_function_descriptors, fd, 0);
1376 if (r < 0)
1377 return r;
1378
1379 return copy_file_fd(s->usb_function_strings, fd, 0);
1380 }
1381
1382 static int usbffs_dispatch_eps(SocketPort *p, int dfd) {
1383 _cleanup_free_ DirectoryEntries *des = NULL;
1384 int r;
1385
1386 assert(p);
1387 assert(dfd >= 0);
1388
1389 r = readdir_all(dfd, RECURSE_DIR_SORT|RECURSE_DIR_IGNORE_DOT, &des);
1390 if (r < 0)
1391 return r;
1392
1393 p->auxiliary_fds = new(int, des->n_entries);
1394 if (!p->auxiliary_fds)
1395 return -ENOMEM;
1396
1397 FOREACH_ARRAY(i, des->entries, des->n_entries) {
1398 const struct dirent *de = *i;
1399
1400 if (streq(de->d_name, "ep0"))
1401 continue;
1402
1403 r = usbffs_address_create_at(dfd, de->d_name);
1404 if (r < 0)
1405 goto fail;
1406
1407 p->auxiliary_fds[p->n_auxiliary_fds++] = r;
1408 }
1409
1410 assert(p->n_auxiliary_fds < des->n_entries);
1411
1412 return 0;
1413
1414 fail:
1415 socket_port_close_auxiliary_fds(p);
1416 return r;
1417 }
1418
1419 int socket_load_service_unit(Socket *s, int cfd, Unit **ret) {
1420 int r;
1421
1422 /* Figure out what the unit that will be used to handle the connections on the socket looks like.
1423 *
1424 * If cfd < 0, then we don't have a connection yet. In case of Accept=yes sockets, use a fake
1425 * instance name.
1426 */
1427
1428 assert(s);
1429 assert(ret);
1430
1431 if (UNIT_ISSET(s->service)) {
1432 *ret = UNIT_DEREF(s->service);
1433 return 0;
1434 }
1435
1436 if (!s->accept)
1437 return -ENODATA;
1438
1439 /* Build the instance name and load the unit */
1440 _cleanup_free_ char *prefix = NULL, *instance = NULL, *name = NULL;
1441
1442 r = unit_name_to_prefix(UNIT(s)->id, &prefix);
1443 if (r < 0)
1444 return r;
1445
1446 if (cfd >= 0) {
1447 r = instance_from_socket(cfd, s->n_accepted, &instance);
1448 if (ERRNO_IS_NEG_DISCONNECT(r))
1449 /* ENOTCONN is legitimate if TCP RST was received. Other socket families might return
1450 * different errors. This connection is over, but the socket unit lives on. */
1451 return log_unit_debug_errno(UNIT(s), r,
1452 "Got %s on incoming socket, assuming aborted connection attempt, ignoring.",
1453 errno_to_name(r));
1454 if (r < 0)
1455 return r;
1456 }
1457
1458 /* For accepting sockets, we don't know how the instance will be called until we get a connection and
1459 * can figure out what the peer name is. So let's use "internal" as the instance to make it clear
1460 * that this is not an actual peer name. We use "unknown" when we cannot figure out the peer. */
1461 r = unit_name_build(prefix, instance ?: "internal", ".service", &name);
1462 if (r < 0)
1463 return r;
1464
1465 return manager_load_unit(UNIT(s)->manager, name, NULL, NULL, ret);
1466 }
1467
1468 static int socket_determine_selinux_label(Socket *s, char **ret) {
1469 Unit *service;
1470 int r;
1471
1472 assert(s);
1473 assert(ret);
1474
1475 r = socket_load_service_unit(s, /* cfd= */ -EBADF, &service);
1476 if (r == -ENODATA) {
1477 *ret = NULL;
1478 return 0;
1479 }
1480 if (r < 0)
1481 return r;
1482
1483 r = service_determine_exec_selinux_label(SERVICE(service), ret);
1484 if (r == -ENODATA) {
1485 *ret = NULL;
1486 return 0;
1487 }
1488 return r;
1489 }
1490
1491 static int socket_address_listen_do(
1492 Socket *s,
1493 const SocketAddress *address,
1494 const char *label) {
1495
1496 assert(s);
1497 assert(address);
1498
1499 return socket_address_listen(
1500 address,
1501 SOCK_CLOEXEC|SOCK_NONBLOCK,
1502 s->backlog,
1503 s->bind_ipv6_only,
1504 s->bind_to_device,
1505 s->reuse_port,
1506 s->free_bind,
1507 s->transparent,
1508 s->directory_mode,
1509 s->socket_mode,
1510 label);
1511 }
1512
1513 #define log_address_error_errno(u, address, error, fmt) \
1514 ({ \
1515 _cleanup_free_ char *_t = NULL; \
1516 \
1517 (void) socket_address_print(address, &_t); \
1518 log_unit_error_errno(u, error, fmt, strna(_t)); \
1519 })
1520
1521 static bool fork_needed(const SocketAddress *address, Socket *s) {
1522 assert(address);
1523 assert(s);
1524
1525 /* Check if we need to do the cgroup or netns stuff. If not we can do things much simpler. */
1526
1527 /* If there are any NFTSet= directives with cgroup source, we need the cgroup */
1528 Unit *u = UNIT(s);
1529 CGroupContext *c = unit_get_cgroup_context(u);
1530 if (c)
1531 FOREACH_ARRAY(nft_set, c->nft_set_context.sets, c->nft_set_context.n_sets)
1532 if (nft_set->source == NFT_SET_SOURCE_CGROUP)
1533 return true;
1534
1535 if (IN_SET(address->sockaddr.sa.sa_family, AF_INET, AF_INET6) &&
1536 bpf_program_supported() > 0) /* If BPF firewalling isn't supported anyway — there's no point in this forking complexity */
1537 return true;
1538
1539 return exec_needs_network_namespace(&s->exec_context);
1540 }
1541
1542 static int socket_address_listen_in_cgroup(
1543 Socket *s,
1544 const SocketAddress *address,
1545 const char *label) {
1546
1547 _cleanup_(pidref_done) PidRef pid = PIDREF_NULL;
1548 _cleanup_close_pair_ int pair[2] = EBADF_PAIR;
1549 int fd, r;
1550
1551 assert(s);
1552 assert(address);
1553
1554 /* This is a wrapper around socket_address_listen(), that forks off a helper process inside the
1555 * socket's cgroup and network namespace in which the socket is actually created. This way we ensure
1556 * the socket is actually properly attached to the unit's cgroup for the purpose of BPF filtering and
1557 * such. */
1558
1559 if (!fork_needed(address, s)) {
1560 /* Shortcut things... */
1561 fd = socket_address_listen_do(s, address, label);
1562 if (fd < 0)
1563 return log_address_error_errno(UNIT(s), address, fd, "Failed to create listening socket (%s): %m");
1564
1565 return fd;
1566 }
1567
1568 r = unit_setup_exec_runtime(UNIT(s));
1569 if (r < 0)
1570 return log_unit_error_errno(UNIT(s), r, "Failed acquire runtime: %m");
1571
1572 if (s->exec_context.network_namespace_path &&
1573 s->exec_runtime &&
1574 s->exec_runtime->shared &&
1575 s->exec_runtime->shared->netns_storage_socket[0] >= 0) {
1576 r = open_shareable_ns_path(s->exec_runtime->shared->netns_storage_socket, s->exec_context.network_namespace_path, CLONE_NEWNET);
1577 if (r < 0)
1578 return log_unit_error_errno(UNIT(s), r, "Failed to open network namespace path %s: %m", s->exec_context.network_namespace_path);
1579 }
1580
1581 if (s->exec_context.ipc_namespace_path &&
1582 s->exec_runtime &&
1583 s->exec_runtime->shared &&
1584 s->exec_runtime->shared->ipcns_storage_socket[0] >= 0) {
1585 r = open_shareable_ns_path(s->exec_runtime->shared->ipcns_storage_socket, s->exec_context.ipc_namespace_path, CLONE_NEWIPC);
1586 if (r < 0)
1587 return log_unit_error_errno(UNIT(s), r, "Failed to open IPC namespace path %s: %m", s->exec_context.ipc_namespace_path);
1588 }
1589
1590 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pair) < 0)
1591 return log_unit_error_errno(UNIT(s), errno, "Failed to create communication channel: %m");
1592
1593 r = unit_fork_helper_process(UNIT(s), "(sd-listen)", /* into_cgroup= */ true, &pid);
1594 if (r < 0)
1595 return log_unit_error_errno(UNIT(s), r, "Failed to fork off listener stub process: %m");
1596 if (r == 0) {
1597 /* Child */
1598
1599 pair[0] = safe_close(pair[0]);
1600
1601 if (exec_needs_network_namespace(&s->exec_context) &&
1602 s->exec_runtime &&
1603 s->exec_runtime->shared &&
1604 s->exec_runtime->shared->netns_storage_socket[0] >= 0) {
1605
1606 if (namespace_type_supported(NAMESPACE_NET)) {
1607 r = setup_shareable_ns(s->exec_runtime->shared->netns_storage_socket, CLONE_NEWNET);
1608 if (r < 0) {
1609 log_unit_error_errno(UNIT(s), r, "Failed to join network namespace: %m");
1610 _exit(EXIT_NETWORK);
1611 }
1612 } else if (s->exec_context.network_namespace_path) {
1613 log_unit_error(UNIT(s), "Network namespace path configured but network namespaces not supported.");
1614 _exit(EXIT_NETWORK);
1615 } else
1616 log_unit_warning(UNIT(s), "PrivateNetwork=yes is configured, but the kernel does not support network namespaces, ignoring.");
1617 }
1618
1619 fd = socket_address_listen_do(s, address, label);
1620 if (fd < 0) {
1621 log_address_error_errno(UNIT(s), address, fd, "Failed to create listening socket (%s): %m");
1622 _exit(EXIT_FAILURE);
1623 }
1624
1625 r = send_one_fd(pair[1], fd, 0);
1626 if (r < 0) {
1627 log_address_error_errno(UNIT(s), address, r, "Failed to send listening socket (%s) to parent: %m");
1628 _exit(EXIT_FAILURE);
1629 }
1630
1631 _exit(EXIT_SUCCESS);
1632 }
1633
1634 pair[1] = safe_close(pair[1]);
1635 fd = receive_one_fd(pair[0], 0);
1636
1637 /* We synchronously wait for the helper, as it shouldn't be slow */
1638 r = wait_for_terminate_and_check("(sd-listen)", pid.pid, WAIT_LOG_ABNORMAL);
1639 if (r < 0) {
1640 safe_close(fd);
1641 return r;
1642 }
1643
1644 if (fd < 0)
1645 return log_address_error_errno(UNIT(s), address, fd, "Failed to receive listening socket (%s): %m");
1646
1647 return fd;
1648 }
1649
1650 static int socket_open_fds(Socket *orig_s) {
1651 _cleanup_(socket_close_fdsp) Socket *s = orig_s;
1652 _cleanup_freecon_ char *label = NULL;
1653 bool know_label = false;
1654 int r;
1655
1656 assert(s);
1657
1658 LIST_FOREACH(port, p, s->ports) {
1659
1660 if (p->fd >= 0)
1661 continue;
1662
1663 switch (p->type) {
1664
1665 case SOCKET_SOCKET:
1666
1667 if (!know_label) {
1668 /* Figure out the label, if we don't it know yet. We do it once for the first
1669 * socket where we need this and remember it for the rest. */
1670
1671 r = socket_determine_selinux_label(s, &label);
1672 if (r < 0)
1673 return log_unit_error_errno(UNIT(s), r, "Failed to determine SELinux label: %m");
1674
1675 know_label = true;
1676 }
1677
1678 /* Apply the socket protocol */
1679 switch (p->address.type) {
1680
1681 case SOCK_STREAM:
1682 if (IN_SET(s->socket_protocol, IPPROTO_SCTP, IPPROTO_MPTCP))
1683 p->address.protocol = s->socket_protocol;
1684 break;
1685
1686 case SOCK_SEQPACKET:
1687 if (s->socket_protocol == IPPROTO_SCTP)
1688 p->address.protocol = s->socket_protocol;
1689 break;
1690
1691 case SOCK_DGRAM:
1692 if (s->socket_protocol == IPPROTO_UDPLITE)
1693 p->address.protocol = s->socket_protocol;
1694 break;
1695 }
1696
1697 p->fd = socket_address_listen_in_cgroup(s, &p->address, label);
1698 if (p->fd < 0)
1699 return p->fd;
1700
1701 socket_apply_socket_options(s, p, p->fd);
1702 socket_symlink(s);
1703 break;
1704
1705 case SOCKET_SPECIAL:
1706
1707 p->fd = special_address_create(p->path, s->writable);
1708 if (p->fd < 0)
1709 return log_unit_error_errno(UNIT(s), p->fd, "Failed to open special file '%s': %m", p->path);
1710 break;
1711
1712 case SOCKET_FIFO:
1713
1714 p->fd = fifo_address_create(
1715 p->path,
1716 s->directory_mode,
1717 s->socket_mode);
1718 if (p->fd < 0)
1719 return log_unit_error_errno(UNIT(s), p->fd, "Failed to open FIFO '%s': %m", p->path);
1720
1721 socket_apply_fifo_options(s, p->fd);
1722 socket_symlink(s);
1723 break;
1724
1725 case SOCKET_MQUEUE:
1726
1727 p->fd = mq_address_create(
1728 p->path,
1729 s->socket_mode,
1730 s->mq_maxmsg,
1731 s->mq_msgsize);
1732 if (p->fd < 0)
1733 return log_unit_error_errno(UNIT(s), p->fd, "Failed to open message queue '%s': %m", p->path);
1734 break;
1735
1736 case SOCKET_USB_FUNCTION: {
1737 _cleanup_close_ int dfd = -EBADF;
1738
1739 dfd = open(p->path, O_DIRECTORY|O_CLOEXEC);
1740 if (dfd < 0)
1741 return log_unit_error_errno(UNIT(s), errno,
1742 "Failed to open USB FunctionFS dir '%s': %m", p->path);
1743
1744 p->fd = usbffs_address_create_at(dfd, "ep0");
1745 if (p->fd < 0)
1746 return log_unit_error_errno(UNIT(s), p->fd, "Failed to open USB FunctionFS ep0: %m");
1747
1748 r = usbffs_write_descs(p->fd, SERVICE(UNIT_DEREF(s->service)));
1749 if (r < 0)
1750 return log_unit_error_errno(UNIT(s), r, "Failed to write to USB FunctionFS ep0: %m");
1751
1752 r = usbffs_dispatch_eps(p, dfd);
1753 if (r < 0)
1754 return log_unit_error_errno(UNIT(s), r, "Failed to dispatch USB FunctionFS eps: %m");
1755
1756 break;
1757 }
1758
1759 default:
1760 assert_not_reached();
1761 }
1762 }
1763
1764 TAKE_PTR(s);
1765 return 0;
1766 }
1767
1768 static void socket_unwatch_fds(Socket *s) {
1769 int r;
1770
1771 assert(s);
1772
1773 LIST_FOREACH(port, p, s->ports) {
1774 if (p->fd < 0)
1775 continue;
1776
1777 r = sd_event_source_set_enabled(p->event_source, SD_EVENT_OFF);
1778 if (r < 0)
1779 log_unit_debug_errno(UNIT(s), r, "Failed to disable event source: %m");
1780 }
1781 }
1782
1783 static int socket_watch_fds(Socket *s) {
1784 int r;
1785
1786 assert(s);
1787
1788 LIST_FOREACH(port, p, s->ports) {
1789 if (p->fd < 0)
1790 continue;
1791
1792 if (p->event_source) {
1793 r = sd_event_source_set_enabled(p->event_source, SD_EVENT_ON);
1794 if (r < 0)
1795 goto fail;
1796 } else {
1797 r = sd_event_add_io(UNIT(s)->manager->event, &p->event_source, p->fd, EPOLLIN, socket_dispatch_io, p);
1798 if (r < 0)
1799 goto fail;
1800
1801 (void) sd_event_source_set_description(p->event_source, "socket-port-io");
1802 }
1803
1804 r = sd_event_source_set_ratelimit(p->event_source, s->poll_limit.interval, s->poll_limit.burst);
1805 if (r < 0)
1806 log_unit_debug_errno(UNIT(s), r, "Failed to set poll limit on I/O event source, ignoring: %m");
1807 }
1808
1809 return 0;
1810
1811 fail:
1812 log_unit_warning_errno(UNIT(s), r, "Failed to watch listening fds: %m");
1813 socket_unwatch_fds(s);
1814 return r;
1815 }
1816
1817 enum {
1818 SOCKET_OPEN_NONE,
1819 SOCKET_OPEN_SOME,
1820 SOCKET_OPEN_ALL,
1821 };
1822
1823 static int socket_check_open(Socket *s) {
1824 bool have_open = false, have_closed = false;
1825
1826 assert(s);
1827
1828 LIST_FOREACH(port, p, s->ports) {
1829 if (p->fd < 0)
1830 have_closed = true;
1831 else
1832 have_open = true;
1833
1834 if (have_open && have_closed)
1835 return SOCKET_OPEN_SOME;
1836 }
1837
1838 if (have_open)
1839 return SOCKET_OPEN_ALL;
1840
1841 return SOCKET_OPEN_NONE;
1842 }
1843
1844 static void socket_set_state(Socket *s, SocketState state) {
1845 SocketState old_state;
1846
1847 assert(s);
1848
1849 if (s->state != state)
1850 bus_unit_send_pending_change_signal(UNIT(s), false);
1851
1852 old_state = s->state;
1853 s->state = state;
1854
1855 if (!SOCKET_STATE_WITH_PROCESS(state)) {
1856 s->timer_event_source = sd_event_source_disable_unref(s->timer_event_source);
1857 socket_unwatch_control_pid(s);
1858 s->control_command = NULL;
1859 s->control_command_id = _SOCKET_EXEC_COMMAND_INVALID;
1860 }
1861
1862 if (state != SOCKET_LISTENING)
1863 socket_unwatch_fds(s);
1864
1865 if (!IN_SET(state,
1866 SOCKET_START_OPEN,
1867 SOCKET_START_CHOWN,
1868 SOCKET_START_POST,
1869 SOCKET_LISTENING,
1870 SOCKET_RUNNING,
1871 SOCKET_STOP_PRE,
1872 SOCKET_STOP_PRE_SIGTERM,
1873 SOCKET_STOP_PRE_SIGKILL))
1874 socket_close_fds(s);
1875
1876 if (state != old_state)
1877 log_unit_debug(UNIT(s), "Changed %s -> %s", socket_state_to_string(old_state), socket_state_to_string(state));
1878
1879 unit_notify(UNIT(s), state_translation_table[old_state], state_translation_table[state], /* reload_success = */ true);
1880 }
1881
1882 static int socket_coldplug(Unit *u) {
1883 Socket *s = ASSERT_PTR(SOCKET(u));
1884 int r;
1885
1886 assert(s->state == SOCKET_DEAD);
1887
1888 if (s->deserialized_state == s->state)
1889 return 0;
1890
1891 if (pidref_is_set(&s->control_pid) &&
1892 pidref_is_unwaited(&s->control_pid) > 0 &&
1893 SOCKET_STATE_WITH_PROCESS(s->deserialized_state)) {
1894
1895 r = unit_watch_pidref(UNIT(s), &s->control_pid, /* exclusive= */ false);
1896 if (r < 0)
1897 return r;
1898
1899 r = socket_arm_timer(s, /* relative= */ false, usec_add(u->state_change_timestamp.monotonic, s->timeout_usec));
1900 if (r < 0)
1901 return r;
1902 }
1903
1904 if (IN_SET(s->deserialized_state,
1905 SOCKET_START_OPEN,
1906 SOCKET_START_CHOWN,
1907 SOCKET_START_POST,
1908 SOCKET_LISTENING,
1909 SOCKET_RUNNING)) {
1910
1911 /* Originally, we used to simply reopen all sockets here that we didn't have file descriptors
1912 * for. However, this is problematic, as we won't traverse through the SOCKET_START_CHOWN state for
1913 * them, and thus the UID/GID wouldn't be right. Hence, instead simply check if we have all fds open,
1914 * and if there's a mismatch, warn loudly.
1915 *
1916 * Note that SOCKET_START_OPEN requires no special treatment, as it's only intermediate
1917 * between SOCKET_START_PRE and SOCKET_START_CHOWN and shall otherwise not be observed.
1918 * It's listed only for consistency. */
1919
1920 r = socket_check_open(s);
1921 if (r == SOCKET_OPEN_NONE)
1922 log_unit_warning(UNIT(s),
1923 "Socket unit configuration has changed while unit has been running, "
1924 "no open socket file descriptor left. "
1925 "The socket unit is not functional until restarted.");
1926 else if (r == SOCKET_OPEN_SOME)
1927 log_unit_warning(UNIT(s),
1928 "Socket unit configuration has changed while unit has been running, "
1929 "and some socket file descriptors have not been opened yet. "
1930 "The socket unit is not fully functional until restarted.");
1931 }
1932
1933 if (s->deserialized_state == SOCKET_LISTENING) {
1934 r = socket_watch_fds(s);
1935 if (r < 0)
1936 return r;
1937 }
1938
1939 if (!IN_SET(s->deserialized_state, SOCKET_DEAD, SOCKET_FAILED, SOCKET_CLEANING))
1940 (void) unit_setup_exec_runtime(u);
1941
1942 socket_set_state(s, s->deserialized_state);
1943 return 0;
1944 }
1945
1946 static int socket_spawn(Socket *s, ExecCommand *c, PidRef *ret_pid) {
1947 _cleanup_(exec_params_shallow_clear) ExecParameters exec_params = EXEC_PARAMETERS_INIT(
1948 EXEC_APPLY_SANDBOXING|EXEC_APPLY_CHROOT|EXEC_APPLY_TTY_STDIN);
1949 _cleanup_(pidref_done) PidRef pidref = PIDREF_NULL;
1950 int r;
1951
1952 assert(s);
1953 assert(c);
1954 assert(ret_pid);
1955
1956 r = unit_prepare_exec(UNIT(s));
1957 if (r < 0)
1958 return r;
1959
1960 r = socket_arm_timer(s, /* relative= */ true, s->timeout_usec);
1961 if (r < 0)
1962 return r;
1963
1964 r = unit_set_exec_params(UNIT(s), &exec_params);
1965 if (r < 0)
1966 return r;
1967
1968 /* Note that ExecStartPre= command doesn't inherit any FDs. It runs before we open listen FDs. */
1969 if (s->pass_fds_to_exec) {
1970 _cleanup_strv_free_ char **fd_names = NULL;
1971 _cleanup_free_ int *fds = NULL;
1972 int n_fds;
1973
1974 n_fds = socket_collect_fds(s, &fds);
1975 if (n_fds < 0)
1976 return n_fds;
1977
1978 r = strv_extend_n(&fd_names, socket_fdname(s), n_fds);
1979 if (r < 0)
1980 return r;
1981
1982 exec_params.flags |= EXEC_PASS_FDS;
1983 exec_params.fds = TAKE_PTR(fds);
1984 exec_params.fd_names = TAKE_PTR(fd_names);
1985 exec_params.n_socket_fds = n_fds;
1986 }
1987
1988 r = exec_spawn(UNIT(s),
1989 c,
1990 &s->exec_context,
1991 &exec_params,
1992 s->exec_runtime,
1993 &s->cgroup_context,
1994 &pidref);
1995 if (r < 0)
1996 return r;
1997
1998 r = unit_watch_pidref(UNIT(s), &pidref, /* exclusive= */ true);
1999 if (r < 0)
2000 return r;
2001
2002 *ret_pid = TAKE_PIDREF(pidref);
2003 return 0;
2004 }
2005
2006 static int socket_chown(Socket *s, PidRef *ret_pid) {
2007 _cleanup_(pidref_done) PidRef pid = PIDREF_NULL;
2008 int r;
2009
2010 assert(s);
2011
2012 r = socket_arm_timer(s, /* relative= */ true, s->timeout_usec);
2013 if (r < 0)
2014 return r;
2015
2016 /* We have to resolve the user names out-of-process, hence
2017 * let's fork here. It's messy, but well, what can we do? */
2018
2019 r = unit_fork_helper_process(UNIT(s), "(sd-chown)", /* into_cgroup= */ true, &pid);
2020 if (r < 0)
2021 return r;
2022 if (r == 0) {
2023 uid_t uid = UID_INVALID;
2024 gid_t gid = GID_INVALID;
2025
2026 /* Child */
2027
2028 if (!isempty(s->user)) {
2029 const char *user = s->user;
2030
2031 r = get_user_creds(&user, &uid, &gid, NULL, NULL, 0);
2032 if (r < 0) {
2033 log_unit_error_errno(UNIT(s), r, "Failed to resolve user %s: %m", user);
2034 _exit(EXIT_USER);
2035 }
2036 }
2037
2038 if (!isempty(s->group)) {
2039 const char *group = s->group;
2040
2041 r = get_group_creds(&group, &gid, 0);
2042 if (r < 0) {
2043 log_unit_error_errno(UNIT(s), r, "Failed to resolve group %s: %m", group);
2044 _exit(EXIT_GROUP);
2045 }
2046 }
2047
2048 LIST_FOREACH(port, p, s->ports) {
2049 const char *path = NULL;
2050
2051 if (p->type == SOCKET_SOCKET)
2052 path = socket_address_get_path(&p->address);
2053 else if (p->type == SOCKET_FIFO)
2054 path = p->path;
2055 else if (p->type == SOCKET_MQUEUE) {
2056 /* Use fchown on the fd since /dev/mqueue might not be mounted. */
2057 if (fchown(p->fd, uid, gid) < 0) {
2058 log_unit_error_errno(UNIT(s), errno, "Failed to fchown(): %m");
2059 _exit(EXIT_CHOWN);
2060 }
2061 continue;
2062 }
2063
2064 if (!path)
2065 continue;
2066
2067 if (chown(path, uid, gid) < 0) {
2068 log_unit_error_errno(UNIT(s), errno, "Failed to chown(): %m");
2069 _exit(EXIT_CHOWN);
2070 }
2071 }
2072
2073 _exit(EXIT_SUCCESS);
2074 }
2075
2076 r = unit_watch_pidref(UNIT(s), &pid, /* exclusive= */ true);
2077 if (r < 0)
2078 return r;
2079
2080 *ret_pid = TAKE_PIDREF(pid);
2081 return 0;
2082 }
2083
2084 static void socket_enter_dead(Socket *s, SocketResult f) {
2085 assert(s);
2086
2087 if (s->result == SOCKET_SUCCESS)
2088 s->result = f;
2089
2090 if (s->result == SOCKET_SUCCESS)
2091 unit_log_success(UNIT(s));
2092 else
2093 unit_log_failure(UNIT(s), socket_result_to_string(s->result));
2094
2095 unit_warn_leftover_processes(UNIT(s), /* start = */ false);
2096
2097 socket_set_state(s, s->result != SOCKET_SUCCESS ? SOCKET_FAILED : SOCKET_DEAD);
2098
2099 s->exec_runtime = exec_runtime_destroy(s->exec_runtime);
2100
2101 unit_destroy_runtime_data(UNIT(s), &s->exec_context, /* destroy_runtime_dir = */ true);
2102
2103 unit_unref_uid_gid(UNIT(s), true);
2104 }
2105
2106 static void socket_enter_signal(Socket *s, SocketState state, SocketResult f);
2107
2108 static void socket_enter_stop_post(Socket *s, SocketResult f) {
2109 int r;
2110
2111 assert(s);
2112
2113 if (s->result == SOCKET_SUCCESS)
2114 s->result = f;
2115
2116 socket_unwatch_control_pid(s);
2117 s->control_command_id = SOCKET_EXEC_STOP_POST;
2118 s->control_command = s->exec_command[SOCKET_EXEC_STOP_POST];
2119
2120 if (s->control_command) {
2121 pidref_done(&s->control_pid);
2122
2123 r = socket_spawn(s, s->control_command, &s->control_pid);
2124 if (r < 0) {
2125 log_unit_warning_errno(UNIT(s), r, "Failed to spawn 'stop-post' task: %m");
2126 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, SOCKET_FAILURE_RESOURCES);
2127 return;
2128 }
2129
2130 socket_set_state(s, SOCKET_STOP_POST);
2131 } else
2132 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, SOCKET_SUCCESS);
2133 }
2134
2135 static int state_to_kill_operation(Socket *s, SocketState state) {
2136 assert(s);
2137
2138 if (state == SOCKET_STOP_PRE_SIGTERM)
2139 return unit_has_job_type(UNIT(s), JOB_RESTART) ? KILL_RESTART : KILL_TERMINATE;
2140
2141 if (state == SOCKET_FINAL_SIGTERM)
2142 return KILL_TERMINATE;
2143
2144 return KILL_KILL;
2145 }
2146
2147 static void socket_enter_signal(Socket *s, SocketState state, SocketResult f) {
2148 int r;
2149
2150 assert(s);
2151
2152 if (s->result == SOCKET_SUCCESS)
2153 s->result = f;
2154
2155 r = unit_kill_context(UNIT(s), state_to_kill_operation(s, state));
2156 if (r < 0) {
2157 log_unit_warning_errno(UNIT(s), r, "Failed to kill processes: %m");
2158 goto fail;
2159 }
2160 if (r > 0) {
2161 r = socket_arm_timer(s, /* relative= */ true, s->timeout_usec);
2162 if (r < 0) {
2163 log_unit_warning_errno(UNIT(s), r, "Failed to install timer: %m");
2164 goto fail;
2165 }
2166
2167 socket_set_state(s, state);
2168 } else if (state == SOCKET_STOP_PRE_SIGTERM)
2169 socket_enter_signal(s, SOCKET_STOP_PRE_SIGKILL, SOCKET_SUCCESS);
2170 else if (state == SOCKET_STOP_PRE_SIGKILL)
2171 socket_enter_stop_post(s, SOCKET_SUCCESS);
2172 else if (state == SOCKET_FINAL_SIGTERM)
2173 socket_enter_signal(s, SOCKET_FINAL_SIGKILL, SOCKET_SUCCESS);
2174 else
2175 socket_enter_dead(s, SOCKET_SUCCESS);
2176
2177 return;
2178
2179 fail:
2180 if (IN_SET(state, SOCKET_STOP_PRE_SIGTERM, SOCKET_STOP_PRE_SIGKILL))
2181 socket_enter_stop_post(s, SOCKET_FAILURE_RESOURCES);
2182 else
2183 socket_enter_dead(s, SOCKET_FAILURE_RESOURCES);
2184 }
2185
2186 static void socket_enter_stop_pre(Socket *s, SocketResult f) {
2187 int r;
2188
2189 assert(s);
2190
2191 if (s->result == SOCKET_SUCCESS)
2192 s->result = f;
2193
2194 socket_unwatch_control_pid(s);
2195 s->control_command_id = SOCKET_EXEC_STOP_PRE;
2196 s->control_command = s->exec_command[SOCKET_EXEC_STOP_PRE];
2197
2198 if (s->control_command) {
2199 pidref_done(&s->control_pid);
2200
2201 r = socket_spawn(s, s->control_command, &s->control_pid);
2202 if (r < 0) {
2203 log_unit_warning_errno(UNIT(s), r, "Failed to spawn 'stop-pre' task: %m");
2204 socket_enter_stop_post(s, SOCKET_FAILURE_RESOURCES);
2205 return;
2206 }
2207
2208 socket_set_state(s, SOCKET_STOP_PRE);
2209 } else
2210 socket_enter_stop_post(s, SOCKET_SUCCESS);
2211 }
2212
2213 static void flush_ports(Socket *s) {
2214 assert(s);
2215
2216 /* Flush all incoming traffic, regardless if actual bytes or new connections, so that this socket isn't busy
2217 * anymore */
2218
2219 LIST_FOREACH(port, p, s->ports) {
2220 if (p->fd < 0)
2221 continue;
2222
2223 if (p->type == SOCKET_MQUEUE)
2224 (void) flush_mqueue(p->fd);
2225 else {
2226 (void) flush_accept(p->fd);
2227 (void) flush_fd(p->fd);
2228 }
2229 }
2230 }
2231
2232 static void socket_enter_listening(Socket *s) {
2233 int r;
2234
2235 assert(s);
2236
2237 if (!s->accept && s->flush_pending) {
2238 log_unit_debug(UNIT(s), "Flushing socket before listening.");
2239 flush_ports(s);
2240 }
2241
2242 r = socket_watch_fds(s);
2243 if (r < 0) {
2244 log_unit_warning_errno(UNIT(s), r, "Failed to watch sockets: %m");
2245 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2246 return;
2247 }
2248
2249 socket_set_state(s, SOCKET_LISTENING);
2250 }
2251
2252 static void socket_enter_start_post(Socket *s) {
2253 int r;
2254
2255 assert(s);
2256
2257 socket_unwatch_control_pid(s);
2258 s->control_command_id = SOCKET_EXEC_START_POST;
2259 s->control_command = s->exec_command[SOCKET_EXEC_START_POST];
2260
2261 if (s->control_command) {
2262 pidref_done(&s->control_pid);
2263
2264 r = socket_spawn(s, s->control_command, &s->control_pid);
2265 if (r < 0) {
2266 log_unit_warning_errno(UNIT(s), r, "Failed to spawn 'start-post' task: %m");
2267 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2268 return;
2269 }
2270
2271 socket_set_state(s, SOCKET_START_POST);
2272 } else
2273 socket_enter_listening(s);
2274 }
2275
2276 static void socket_enter_start_chown(Socket *s) {
2277 int r;
2278
2279 assert(s);
2280 assert(s->state == SOCKET_START_OPEN);
2281
2282 if (!isempty(s->user) || !isempty(s->group)) {
2283
2284 socket_unwatch_control_pid(s);
2285 s->control_command_id = SOCKET_EXEC_START_CHOWN;
2286 s->control_command = NULL;
2287
2288 r = socket_chown(s, &s->control_pid);
2289 if (r < 0) {
2290 log_unit_warning_errno(UNIT(s), r, "Failed to spawn 'start-chown' task: %m");
2291 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2292 return;
2293 }
2294
2295 socket_set_state(s, SOCKET_START_CHOWN);
2296 } else
2297 socket_enter_start_post(s);
2298 }
2299
2300 static void socket_enter_start_open(Socket *s) {
2301 int r;
2302
2303 assert(s);
2304 assert(IN_SET(s->state, SOCKET_DEAD, SOCKET_FAILED, SOCKET_START_PRE));
2305
2306 /* We force a state transition here even though we're not spawning any process (i.e. the state is purely
2307 * intermediate), so that failure of socket_open_fds() always causes a state change in unit_notify().
2308 * Otherwise, if no Exec*= is defined, we might go from previous SOCKET_FAILED to SOCKET_FAILED,
2309 * meaning the OnFailure= deps are unexpectedly skipped (#35635). */
2310
2311 socket_set_state(s, SOCKET_START_OPEN);
2312
2313 r = socket_open_fds(s);
2314 if (r < 0) {
2315 log_unit_error_errno(UNIT(s), r, "Failed to listen on sockets: %m");
2316 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2317 return;
2318 }
2319
2320 socket_enter_start_chown(s);
2321 }
2322
2323 static void socket_enter_start_pre(Socket *s) {
2324 int r;
2325
2326 assert(s);
2327
2328 socket_unwatch_control_pid(s);
2329
2330 unit_warn_leftover_processes(UNIT(s), /* start = */ true);
2331
2332 s->control_command_id = SOCKET_EXEC_START_PRE;
2333 s->control_command = s->exec_command[SOCKET_EXEC_START_PRE];
2334
2335 if (s->control_command) {
2336 pidref_done(&s->control_pid);
2337
2338 r = socket_spawn(s, s->control_command, &s->control_pid);
2339 if (r < 0) {
2340 log_unit_warning_errno(UNIT(s), r, "Failed to spawn 'start-pre' task: %m");
2341 socket_enter_dead(s, SOCKET_FAILURE_RESOURCES);
2342 return;
2343 }
2344
2345 socket_set_state(s, SOCKET_START_PRE);
2346 } else
2347 socket_enter_start_open(s);
2348 }
2349
2350 static void socket_enter_running(Socket *s, int cfd_in) {
2351 /* Note that this call takes possession of the connection fd passed. It either has to assign it
2352 * somewhere or close it. */
2353 _cleanup_close_ int cfd = cfd_in;
2354 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
2355 int r;
2356
2357 assert(s);
2358
2359 /* We don't take connections anymore if we are supposed to shut down anyway */
2360 if (unit_stop_pending(UNIT(s))) {
2361
2362 log_unit_debug(UNIT(s), "Suppressing connection request since unit stop is scheduled.");
2363
2364 if (cfd >= 0)
2365 goto refuse;
2366
2367 flush_ports(s);
2368 return;
2369 }
2370
2371 if (!ratelimit_below(&s->trigger_limit)) {
2372 log_unit_warning(UNIT(s), "Trigger limit hit, refusing further activation.");
2373 socket_enter_stop_pre(s, SOCKET_FAILURE_TRIGGER_LIMIT_HIT);
2374 goto refuse;
2375 }
2376
2377 if (cfd < 0) { /* Accept=no case */
2378 bool pending = false;
2379 Unit *other;
2380
2381 /* If there's already a start pending don't bother to do anything */
2382 UNIT_FOREACH_DEPENDENCY(other, UNIT(s), UNIT_ATOM_TRIGGERS)
2383 if (unit_active_or_pending(other)) {
2384 pending = true;
2385 break;
2386 }
2387
2388 if (!pending) {
2389 if (!UNIT_ISSET(s->service)) {
2390 log_unit_warning(UNIT(s),
2391 "Service to activate vanished, refusing activation.");
2392 goto fail;
2393 }
2394
2395 r = manager_add_job(UNIT(s)->manager, JOB_START, UNIT_DEREF(s->service), JOB_REPLACE, &error, /* ret = */ NULL);
2396 if (r == -EDEADLK)
2397 return (void) log_unit_debug_errno(UNIT(s), r, "Failed to queue service startup job, ignoring: %s", bus_error_message(&error, r));
2398 if (r < 0)
2399 goto queue_error;
2400 }
2401
2402 socket_set_state(s, SOCKET_RUNNING);
2403 } else { /* Accept=yes case */
2404 _cleanup_(socket_peer_unrefp) SocketPeer *p = NULL;
2405 Unit *service;
2406
2407 if (s->n_connections >= s->max_connections) {
2408 log_unit_warning(UNIT(s), "Too many incoming connections (%u), dropping connection.",
2409 s->n_connections);
2410 goto refuse;
2411 }
2412
2413 if (s->max_connections_per_source > 0) {
2414 r = socket_acquire_peer(s, cfd, &p);
2415 if (r < 0) {
2416 if (ERRNO_IS_DISCONNECT(r))
2417 return;
2418 /* We didn't have enough resources to acquire peer information, let's fail. */
2419 goto fail;
2420 }
2421 if (r > 0 && p->n_ref > s->max_connections_per_source) {
2422 _cleanup_free_ char *t = NULL;
2423
2424 if (p->peer.sa.sa_family == AF_UNIX)
2425 (void) asprintf(&t, "UID " UID_FMT, p->peer_cred.uid);
2426 else
2427 (void) sockaddr_pretty(&p->peer.sa, p->peer_salen, /* translate_ipv6= */ true, /* include_port= */ false, &t);
2428
2429 log_unit_warning(UNIT(s),
2430 "Too many incoming connections (%u) from source %s, dropping connection.",
2431 p->n_ref, strnull(t));
2432 goto refuse;
2433 }
2434 }
2435
2436 r = socket_load_service_unit(s, cfd, &service);
2437 if (ERRNO_IS_NEG_DISCONNECT(r))
2438 return;
2439 if (r < 0 || UNIT_IS_LOAD_ERROR(service->load_state)) {
2440 log_unit_warning_errno(UNIT(s), r < 0 ? r : service->load_error,
2441 "Failed to load connection service unit: %m");
2442 goto fail;
2443 }
2444 if (service->load_state == UNIT_MASKED) {
2445 log_unit_warning(UNIT(s), "Connection service unit is masked, refusing.");
2446 goto fail;
2447 }
2448
2449 s->n_accepted++;
2450
2451 r = service_set_socket_fd(SERVICE(service), cfd, s, p, s->selinux_context_from_net);
2452 if (ERRNO_IS_NEG_DISCONNECT(r))
2453 return;
2454 if (r < 0) {
2455 log_unit_warning_errno(UNIT(s), r, "Failed to set socket on service: %m");
2456 goto fail;
2457 }
2458
2459 /* We passed ownership of the fd and socket peer to the service now. */
2460 TAKE_FD(cfd);
2461 TAKE_PTR(p);
2462
2463 s->n_connections++;
2464
2465 r = manager_add_job(UNIT(s)->manager, JOB_START, service, JOB_REPLACE, &error, /* ret = */ NULL);
2466 if (r < 0) {
2467 /* We failed to activate the new service, but it still exists. Let's make sure the
2468 * service closes and forgets the connection fd again, immediately. */
2469 service_release_socket_fd(SERVICE(service));
2470 goto queue_error;
2471 }
2472
2473 /* Notify clients about changed counters */
2474 unit_add_to_dbus_queue(UNIT(s));
2475 }
2476
2477 return;
2478
2479 refuse:
2480 s->n_refused++;
2481 return;
2482
2483 queue_error:
2484 log_unit_warning_errno(UNIT(s), r, "Failed to queue service startup job%s: %s",
2485 cfd >= 0 && !ERRNO_IS_RESOURCE(r) ? " (Maybe the service is missing or is a template unit?)" : "",
2486 bus_error_message(&error, r));
2487
2488 fail:
2489 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2490 }
2491
2492 static void socket_run_next(Socket *s) {
2493 int r;
2494
2495 assert(s);
2496 assert(s->control_command);
2497 assert(s->control_command->command_next);
2498
2499 socket_unwatch_control_pid(s);
2500
2501 s->control_command = s->control_command->command_next;
2502
2503 pidref_done(&s->control_pid);
2504
2505 r = socket_spawn(s, s->control_command, &s->control_pid);
2506 if (r < 0) {
2507 log_unit_warning_errno(UNIT(s), r, "Failed to spawn next task: %m");
2508
2509 if (s->state == SOCKET_START_POST)
2510 socket_enter_stop_pre(s, SOCKET_FAILURE_RESOURCES);
2511 else if (s->state == SOCKET_STOP_POST)
2512 socket_enter_dead(s, SOCKET_FAILURE_RESOURCES);
2513 else
2514 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, SOCKET_FAILURE_RESOURCES);
2515 }
2516 }
2517
2518 static int socket_start(Unit *u) {
2519 Socket *s = ASSERT_PTR(SOCKET(u));
2520 int r;
2521
2522 /* We cannot fulfill this request right now, try again later
2523 * please! */
2524 if (IN_SET(s->state,
2525 SOCKET_STOP_PRE,
2526 SOCKET_STOP_PRE_SIGKILL,
2527 SOCKET_STOP_PRE_SIGTERM,
2528 SOCKET_STOP_POST,
2529 SOCKET_FINAL_SIGTERM,
2530 SOCKET_FINAL_SIGKILL,
2531 SOCKET_CLEANING))
2532 return -EAGAIN;
2533
2534 /* Already on it! */
2535 if (IN_SET(s->state,
2536 SOCKET_START_PRE,
2537 SOCKET_START_OPEN,
2538 SOCKET_START_CHOWN,
2539 SOCKET_START_POST))
2540 return 0;
2541
2542 /* Cannot run this without the service being around */
2543 if (UNIT_ISSET(s->service)) {
2544 Service *service = ASSERT_PTR(SERVICE(UNIT_DEREF(s->service)));
2545
2546 if (UNIT(service)->load_state != UNIT_LOADED)
2547 return log_unit_error_errno(u, SYNTHETIC_ERRNO(ENOENT),
2548 "Socket service %s not loaded, refusing.", UNIT(service)->id);
2549
2550 /* If the service is already active we cannot start the socket */
2551 if (SOCKET_SERVICE_IS_ACTIVE(service, /* allow_finalize = */ false))
2552 return log_unit_error_errno(u, SYNTHETIC_ERRNO(EBUSY),
2553 "Socket service %s already active, refusing.", UNIT(service)->id);
2554 }
2555
2556 assert(IN_SET(s->state, SOCKET_DEAD, SOCKET_FAILED));
2557
2558 r = unit_acquire_invocation_id(u);
2559 if (r < 0)
2560 return r;
2561
2562 s->result = SOCKET_SUCCESS;
2563 exec_command_reset_status_list_array(s->exec_command, _SOCKET_EXEC_COMMAND_MAX);
2564
2565 if (s->cgroup_runtime)
2566 s->cgroup_runtime->reset_accounting = true;
2567
2568 socket_enter_start_pre(s);
2569 return 1;
2570 }
2571
2572 static int socket_stop(Unit *u) {
2573 Socket *s = ASSERT_PTR(SOCKET(u));
2574
2575 /* Already on it */
2576 if (IN_SET(s->state,
2577 SOCKET_STOP_PRE,
2578 SOCKET_STOP_PRE_SIGTERM,
2579 SOCKET_STOP_PRE_SIGKILL,
2580 SOCKET_STOP_POST,
2581 SOCKET_FINAL_SIGTERM,
2582 SOCKET_FINAL_SIGKILL))
2583 return 0;
2584
2585 /* If there's already something running we go directly into
2586 * kill mode. */
2587 if (IN_SET(s->state,
2588 SOCKET_START_PRE,
2589 SOCKET_START_OPEN,
2590 SOCKET_START_CHOWN,
2591 SOCKET_START_POST)) {
2592 socket_enter_signal(s, SOCKET_STOP_PRE_SIGTERM, SOCKET_SUCCESS);
2593 return -EAGAIN;
2594 }
2595
2596 /* If we are currently cleaning, then abort it, brutally. */
2597 if (s->state == SOCKET_CLEANING) {
2598 socket_enter_signal(s, SOCKET_FINAL_SIGKILL, SOCKET_SUCCESS);
2599 return 0;
2600 }
2601
2602 assert(IN_SET(s->state, SOCKET_LISTENING, SOCKET_RUNNING));
2603
2604 socket_enter_stop_pre(s, SOCKET_SUCCESS);
2605 return 1;
2606 }
2607
2608 static int socket_serialize(Unit *u, FILE *f, FDSet *fds) {
2609 Socket *s = ASSERT_PTR(SOCKET(u));
2610 int r;
2611
2612 assert(f);
2613 assert(fds);
2614
2615 (void) serialize_item(f, "state", socket_state_to_string(s->state));
2616 (void) serialize_item(f, "result", socket_result_to_string(s->result));
2617 (void) serialize_item_format(f, "n-accepted", "%u", s->n_accepted);
2618 (void) serialize_item_format(f, "n-refused", "%u", s->n_refused);
2619 (void) serialize_pidref(f, fds, "control-pid", &s->control_pid);
2620
2621 if (s->control_command_id >= 0)
2622 (void) serialize_item(f, "control-command", socket_exec_command_to_string(s->control_command_id));
2623
2624 LIST_FOREACH(port, p, s->ports) {
2625 int copy;
2626
2627 if (p->fd < 0)
2628 continue;
2629
2630 copy = fdset_put_dup(fds, p->fd);
2631 if (copy < 0)
2632 return log_unit_warning_errno(u, copy, "Failed to serialize socket fd: %m");
2633
2634 if (p->type == SOCKET_SOCKET) {
2635 _cleanup_free_ char *t = NULL;
2636
2637 r = socket_address_print(&p->address, &t);
2638 if (r < 0)
2639 return log_unit_error_errno(u, r, "Failed to format socket address: %m");
2640
2641 if (socket_address_family(&p->address) == AF_NETLINK)
2642 (void) serialize_item_format(f, "netlink", "%i %s", copy, t);
2643 else
2644 (void) serialize_item_format(f, "socket", "%i %i %s", copy, p->address.type, t);
2645 } else if (p->type == SOCKET_SPECIAL)
2646 (void) serialize_item_format(f, "special", "%i %s", copy, p->path);
2647 else if (p->type == SOCKET_MQUEUE)
2648 (void) serialize_item_format(f, "mqueue", "%i %s", copy, p->path);
2649 else if (p->type == SOCKET_USB_FUNCTION)
2650 (void) serialize_item_format(f, "ffs", "%i %s", copy, p->path);
2651 else {
2652 assert(p->type == SOCKET_FIFO);
2653 (void) serialize_item_format(f, "fifo", "%i %s", copy, p->path);
2654 }
2655 }
2656
2657 (void) serialize_ratelimit(f, "trigger-ratelimit", &s->trigger_limit);
2658
2659 return 0;
2660 }
2661
2662 static int socket_deserialize_item(Unit *u, const char *key, const char *value, FDSet *fds) {
2663 Socket *s = ASSERT_PTR(SOCKET(u));
2664 int r;
2665
2666 assert(key);
2667 assert(value);
2668
2669 if (streq(key, "state")) {
2670 SocketState state;
2671
2672 state = socket_state_from_string(value);
2673 if (state < 0)
2674 log_unit_debug(u, "Failed to parse state value: %s", value);
2675 else
2676 s->deserialized_state = state;
2677 } else if (streq(key, "result")) {
2678 SocketResult f;
2679
2680 f = socket_result_from_string(value);
2681 if (f < 0)
2682 log_unit_debug(u, "Failed to parse result value: %s", value);
2683 else if (f != SOCKET_SUCCESS)
2684 s->result = f;
2685
2686 } else if (streq(key, "n-accepted")) {
2687 unsigned k;
2688
2689 if (safe_atou(value, &k) < 0)
2690 log_unit_debug(u, "Failed to parse n-accepted value: %s", value);
2691 else
2692 s->n_accepted += k;
2693 } else if (streq(key, "n-refused")) {
2694 unsigned k;
2695
2696 if (safe_atou(value, &k) < 0)
2697 log_unit_debug(u, "Failed to parse n-refused value: %s", value);
2698 else
2699 s->n_refused += k;
2700 } else if (streq(key, "control-pid")) {
2701
2702 if (!pidref_is_set(&s->control_pid))
2703 (void) deserialize_pidref(fds, value, &s->control_pid);
2704
2705 } else if (streq(key, "control-command")) {
2706 SocketExecCommand id;
2707
2708 id = socket_exec_command_from_string(value);
2709 if (id < 0)
2710 log_unit_debug(u, "Failed to parse exec-command value: %s", value);
2711 else {
2712 s->control_command_id = id;
2713 s->control_command = s->exec_command[id];
2714 }
2715 } else if (streq(key, "fifo")) {
2716 _cleanup_free_ char *fdv = NULL;
2717 bool found = false;
2718 int fd;
2719
2720 r = extract_first_word(&value, &fdv, NULL, 0);
2721 if (r <= 0) {
2722 log_unit_debug(u, "Failed to parse fifo value: %s", value);
2723 return 0;
2724 }
2725
2726 fd = parse_fd(fdv);
2727 if (fd < 0 || !fdset_contains(fds, fd)) {
2728 log_unit_debug(u, "Invalid fifo value: %s", fdv);
2729 return 0;
2730 }
2731
2732 LIST_FOREACH(port, p, s->ports)
2733 if (p->fd < 0 &&
2734 p->type == SOCKET_FIFO &&
2735 path_equal_or_inode_same(p->path, value, 0)) {
2736 p->fd = fdset_remove(fds, fd);
2737 found = true;
2738 break;
2739 }
2740 if (!found)
2741 log_unit_debug(u, "No matching fifo socket found: %s", value);
2742
2743 } else if (streq(key, "special")) {
2744 _cleanup_free_ char *fdv = NULL;
2745 bool found = false;
2746 int fd;
2747
2748 r = extract_first_word(&value, &fdv, NULL, 0);
2749 if (r <= 0) {
2750 log_unit_debug(u, "Failed to parse special value: %s", value);
2751 return 0;
2752 }
2753
2754 fd = parse_fd(fdv);
2755 if (fd < 0 || !fdset_contains(fds, fd)) {
2756 log_unit_debug(u, "Invalid special value: %s", fdv);
2757 return 0;
2758 }
2759
2760 LIST_FOREACH(port, p, s->ports)
2761 if (p->fd < 0 &&
2762 p->type == SOCKET_SPECIAL &&
2763 path_equal_or_inode_same(p->path, value, 0)) {
2764 p->fd = fdset_remove(fds, fd);
2765 found = true;
2766 break;
2767 }
2768 if (!found)
2769 log_unit_debug(u, "No matching special socket found: %s", value);
2770
2771 } else if (streq(key, "mqueue")) {
2772 _cleanup_free_ char *fdv = NULL;
2773 bool found = false;
2774 int fd;
2775
2776 r = extract_first_word(&value, &fdv, NULL, 0);
2777 if (r <= 0) {
2778 log_unit_debug(u, "Failed to parse mqueue value: %s", value);
2779 return 0;
2780 }
2781
2782 fd = parse_fd(fdv);
2783 if (fd < 0 || !fdset_contains(fds, fd)) {
2784 log_unit_debug(u, "Invalid mqueue value: %s", fdv);
2785 return 0;
2786 }
2787
2788 LIST_FOREACH(port, p, s->ports)
2789 if (p->fd < 0 &&
2790 p->type == SOCKET_MQUEUE &&
2791 streq(p->path, value)) {
2792 p->fd = fdset_remove(fds, fd);
2793 found = true;
2794 break;
2795 }
2796 if (!found)
2797 log_unit_debug(u, "No matching mqueue socket found: %s", value);
2798
2799 } else if (streq(key, "socket")) {
2800 _cleanup_free_ char *fdv = NULL, *typev = NULL;
2801 bool found = false;
2802 int fd, type;
2803
2804 r = extract_first_word(&value, &fdv, NULL, 0);
2805 if (r <= 0) {
2806 log_unit_debug(u, "Failed to parse socket fd from value: %s", value);
2807 return 0;
2808 }
2809
2810 fd = parse_fd(fdv);
2811 if (fd < 0 || !fdset_contains(fds, fd)) {
2812 log_unit_debug(u, "Invalid socket fd: %s", fdv);
2813 return 0;
2814 }
2815
2816 r = extract_first_word(&value, &typev, NULL, 0);
2817 if (r <= 0) {
2818 log_unit_debug(u, "Failed to parse socket type from value: %s", value);
2819 return 0;
2820 }
2821
2822 if (safe_atoi(typev, &type) < 0 || type < 0) {
2823 log_unit_debug(u, "Invalid socket type: %s", typev);
2824 return 0;
2825 }
2826
2827 LIST_FOREACH(port, p, s->ports)
2828 if (p->fd < 0 &&
2829 socket_address_is(&p->address, value, type)) {
2830 p->fd = fdset_remove(fds, fd);
2831 found = true;
2832 break;
2833 }
2834 if (!found)
2835 log_unit_debug(u, "No matching %s socket found: %s",
2836 socket_address_type_to_string(type), value);
2837
2838 } else if (streq(key, "netlink")) {
2839 _cleanup_free_ char *fdv = NULL;
2840 bool found = false;
2841 int fd;
2842
2843 r = extract_first_word(&value, &fdv, NULL, 0);
2844 if (r <= 0) {
2845 log_unit_debug(u, "Failed to parse socket value: %s", value);
2846 return 0;
2847 }
2848
2849 fd = parse_fd(fdv);
2850 if (fd < 0 || !fdset_contains(fds, fd)) {
2851 log_unit_debug(u, "Invalid socket value: %s", fdv);
2852 return 0;
2853 }
2854
2855 LIST_FOREACH(port, p, s->ports)
2856 if (p->fd < 0 &&
2857 socket_address_is_netlink(&p->address, value)) {
2858 p->fd = fdset_remove(fds, fd);
2859 found = true;
2860 break;
2861 }
2862 if (!found)
2863 log_unit_debug(u, "No matching netlink socket found: %s", value);
2864
2865 } else if (streq(key, "ffs")) {
2866 _cleanup_free_ char *fdv = NULL;
2867 bool found = false;
2868 int fd;
2869
2870 r = extract_first_word(&value, &fdv, NULL, 0);
2871 if (r <= 0) {
2872 log_unit_debug(u, "Failed to parse ffs value: %s", value);
2873 return 0;
2874 }
2875
2876 fd = parse_fd(fdv);
2877 if (fd < 0 || !fdset_contains(fds, fd)) {
2878 log_unit_debug(u, "Invalid ffs value: %s", fdv);
2879 return 0;
2880 }
2881
2882 LIST_FOREACH(port, p, s->ports)
2883 if (p->fd < 0 &&
2884 p->type == SOCKET_USB_FUNCTION &&
2885 path_equal_or_inode_same(p->path, value, 0)) {
2886 p->fd = fdset_remove(fds, fd);
2887 found = true;
2888 break;
2889 }
2890 if (!found)
2891 log_unit_debug(u, "No matching ffs socket found: %s", value);
2892
2893 } else if (streq(key, "trigger-ratelimit"))
2894 (void) deserialize_ratelimit(&s->trigger_limit, key, value);
2895 else
2896 log_unit_debug(UNIT(s), "Unknown serialization key: %s", key);
2897
2898 return 0;
2899 }
2900
2901 static void socket_distribute_fds(Unit *u, FDSet *fds) {
2902 Socket *s = ASSERT_PTR(SOCKET(u));
2903
2904 LIST_FOREACH(port, p, s->ports) {
2905 int fd;
2906
2907 if (p->type != SOCKET_SOCKET)
2908 continue;
2909
2910 if (p->fd >= 0)
2911 continue;
2912
2913 FDSET_FOREACH(fd, fds) {
2914 if (socket_address_matches_fd(&p->address, fd)) {
2915 p->fd = fdset_remove(fds, fd);
2916 s->deserialized_state = SOCKET_LISTENING;
2917 break;
2918 }
2919 }
2920 }
2921 }
2922
2923 static UnitActiveState socket_active_state(Unit *u) {
2924 Socket *s = ASSERT_PTR(SOCKET(u));
2925
2926 return state_translation_table[s->state];
2927 }
2928
2929 static const char *socket_sub_state_to_string(Unit *u) {
2930 Socket *s = ASSERT_PTR(SOCKET(u));
2931
2932 return socket_state_to_string(s->state);
2933 }
2934
2935 int socket_port_to_address(const SocketPort *p, char **ret) {
2936 _cleanup_free_ char *address = NULL;
2937 int r;
2938
2939 assert(p);
2940 assert(ret);
2941
2942 switch (p->type) {
2943 case SOCKET_SOCKET: {
2944 r = socket_address_print(&p->address, &address);
2945 if (r < 0)
2946 return r;
2947
2948 break;
2949 }
2950
2951 case SOCKET_SPECIAL:
2952 case SOCKET_MQUEUE:
2953 case SOCKET_FIFO:
2954 case SOCKET_USB_FUNCTION:
2955 address = strdup(p->path);
2956 if (!address)
2957 return -ENOMEM;
2958 break;
2959
2960 default:
2961 assert_not_reached();
2962 }
2963
2964 *ret = TAKE_PTR(address);
2965
2966 return 0;
2967 }
2968
2969 const char* socket_port_type_to_string(SocketPort *p) {
2970 assert(p);
2971
2972 switch (p->type) {
2973
2974 case SOCKET_SOCKET:
2975
2976 switch (p->address.type) {
2977
2978 case SOCK_STREAM:
2979 return "Stream";
2980
2981 case SOCK_DGRAM:
2982 return "Datagram";
2983
2984 case SOCK_SEQPACKET:
2985 return "SequentialPacket";
2986
2987 case SOCK_RAW:
2988 if (socket_address_family(&p->address) == AF_NETLINK)
2989 return "Netlink";
2990
2991 _fallthrough_;
2992 default:
2993 return NULL;
2994 }
2995
2996 case SOCKET_SPECIAL:
2997 return "Special";
2998
2999 case SOCKET_MQUEUE:
3000 return "MessageQueue";
3001
3002 case SOCKET_FIFO:
3003 return "FIFO";
3004
3005 case SOCKET_USB_FUNCTION:
3006 return "USBFunction";
3007
3008 default:
3009 return NULL;
3010 }
3011 }
3012
3013 SocketType socket_port_type_from_string(const char *s) {
3014 assert(s);
3015
3016 if (STR_IN_SET(s, "Stream", "Datagram", "SequentialPacket", "Netlink"))
3017 return SOCKET_SOCKET;
3018 else if (streq(s, "Special"))
3019 return SOCKET_SPECIAL;
3020 else if (streq(s, "MessageQueue"))
3021 return SOCKET_MQUEUE;
3022 else if (streq(s, "FIFO"))
3023 return SOCKET_FIFO;
3024 else if (streq(s, "USBFunction"))
3025 return SOCKET_USB_FUNCTION;
3026 else
3027 return _SOCKET_TYPE_INVALID;
3028 }
3029
3030 static bool socket_may_gc(Unit *u) {
3031 Socket *s = ASSERT_PTR(SOCKET(u));
3032
3033 return s->n_connections == 0;
3034 }
3035
3036 static int socket_accept_do(Socket *s, int fd) {
3037 int cfd;
3038
3039 assert(s);
3040 assert(fd >= 0);
3041
3042 cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC);
3043 if (cfd < 0)
3044 /* Convert transient network errors into clean and well-defined EAGAIN */
3045 return ERRNO_IS_ACCEPT_AGAIN(errno) ? -EAGAIN : -errno;
3046
3047 return cfd;
3048 }
3049
3050 static int socket_accept_in_cgroup(Socket *s, SocketPort *p, int fd) {
3051 _cleanup_(pidref_done) PidRef pid = PIDREF_NULL;
3052 _cleanup_close_pair_ int pair[2] = EBADF_PAIR;
3053 int cfd, r;
3054
3055 assert(s);
3056 assert(p);
3057 assert(fd >= 0);
3058
3059 /* Similar to socket_address_listen_in_cgroup(), but for accept() rather than socket(): make sure that any
3060 * connection socket is also properly associated with the cgroup. */
3061
3062 if (!IN_SET(p->address.sockaddr.sa.sa_family, AF_INET, AF_INET6))
3063 goto shortcut;
3064
3065 if (bpf_program_supported() <= 0)
3066 goto shortcut;
3067
3068 if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pair) < 0)
3069 return log_unit_error_errno(UNIT(s), errno, "Failed to create communication channel: %m");
3070
3071 r = unit_fork_helper_process(UNIT(s), "(sd-accept)", /* into_cgroup= */ true, &pid);
3072 if (r < 0)
3073 return log_unit_error_errno(UNIT(s), r, "Failed to fork off accept stub process: %m");
3074 if (r == 0) {
3075 /* Child */
3076
3077 pair[0] = safe_close(pair[0]);
3078
3079 cfd = socket_accept_do(s, fd);
3080 if (cfd == -EAGAIN) /* spurious accept() */
3081 _exit(EXIT_SUCCESS);
3082 if (cfd < 0) {
3083 log_unit_error_errno(UNIT(s), cfd, "Failed to accept connection socket: %m");
3084 _exit(EXIT_FAILURE);
3085 }
3086
3087 r = send_one_fd(pair[1], cfd, 0);
3088 if (r < 0) {
3089 log_unit_error_errno(UNIT(s), r, "Failed to send connection socket to parent: %m");
3090 _exit(EXIT_FAILURE);
3091 }
3092
3093 _exit(EXIT_SUCCESS);
3094 }
3095
3096 pair[1] = safe_close(pair[1]);
3097 cfd = receive_one_fd(pair[0], 0);
3098
3099 /* We synchronously wait for the helper, as it shouldn't be slow */
3100 r = wait_for_terminate_and_check("(sd-accept)", pid.pid, WAIT_LOG_ABNORMAL);
3101 if (r < 0) {
3102 safe_close(cfd);
3103 return r;
3104 }
3105
3106 /* If we received no fd, we got EIO here. If this happens with a process exit code of EXIT_SUCCESS
3107 * this is a spurious accept(), let's convert that back to EAGAIN here. */
3108 if (cfd == -EIO)
3109 return -EAGAIN;
3110 if (cfd < 0)
3111 return log_unit_error_errno(UNIT(s), cfd, "Failed to receive connection socket: %m");
3112
3113 return cfd;
3114
3115 shortcut:
3116 cfd = socket_accept_do(s, fd);
3117 if (cfd == -EAGAIN) /* spurious accept(), skip it silently */
3118 return -EAGAIN;
3119 if (cfd < 0)
3120 return log_unit_error_errno(UNIT(s), cfd, "Failed to accept connection socket: %m");
3121
3122 return cfd;
3123 }
3124
3125 static int socket_dispatch_io(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3126 SocketPort *p = ASSERT_PTR(userdata);
3127 int cfd = -EBADF;
3128
3129 assert(fd >= 0);
3130
3131 if (p->socket->state != SOCKET_LISTENING)
3132 return 0;
3133
3134 log_unit_debug(UNIT(p->socket), "Incoming traffic");
3135
3136 if (revents != EPOLLIN) {
3137 if (revents & EPOLLHUP)
3138 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.");
3139 else
3140 log_unit_error(UNIT(p->socket), "Got unexpected poll event (0x%x) on socket.", revents);
3141 goto fail;
3142 }
3143
3144 if (p->socket->accept &&
3145 p->type == SOCKET_SOCKET &&
3146 socket_address_can_accept(&p->address)) {
3147
3148 cfd = socket_accept_in_cgroup(p->socket, p, fd);
3149 if (cfd == -EAGAIN) /* Spurious accept() */
3150 return 0;
3151 if (cfd < 0)
3152 goto fail;
3153
3154 socket_apply_socket_options(p->socket, p, cfd);
3155 }
3156
3157 socket_enter_running(p->socket, cfd);
3158 return 0;
3159
3160 fail:
3161 socket_enter_stop_pre(p->socket, SOCKET_FAILURE_RESOURCES);
3162 return 0;
3163 }
3164
3165 static void socket_sigchld_event(Unit *u, pid_t pid, int code, int status) {
3166 Socket *s = ASSERT_PTR(SOCKET(u));
3167 SocketResult f;
3168
3169 assert(pid >= 0);
3170
3171 if (pid != s->control_pid.pid)
3172 return;
3173
3174 pidref_done(&s->control_pid);
3175
3176 if (is_clean_exit(code, status, EXIT_CLEAN_COMMAND, NULL))
3177 f = SOCKET_SUCCESS;
3178 else if (code == CLD_EXITED)
3179 f = SOCKET_FAILURE_EXIT_CODE;
3180 else if (code == CLD_KILLED)
3181 f = SOCKET_FAILURE_SIGNAL;
3182 else if (code == CLD_DUMPED)
3183 f = SOCKET_FAILURE_CORE_DUMP;
3184 else
3185 assert_not_reached();
3186
3187 if (s->control_command) {
3188 exec_status_exit(&s->control_command->exec_status, &s->exec_context, pid, code, status);
3189
3190 if (s->control_command->flags & EXEC_COMMAND_IGNORE_FAILURE)
3191 f = SOCKET_SUCCESS;
3192 }
3193
3194 unit_log_process_exit(
3195 u,
3196 "Control process",
3197 socket_exec_command_to_string(s->control_command_id),
3198 f == SOCKET_SUCCESS,
3199 code, status);
3200
3201 if (s->result == SOCKET_SUCCESS)
3202 s->result = f;
3203
3204 if (s->control_command &&
3205 s->control_command->command_next &&
3206 f == SOCKET_SUCCESS) {
3207
3208 log_unit_debug(u, "Running next command for state %s", socket_state_to_string(s->state));
3209 socket_run_next(s);
3210 } else {
3211 s->control_command = NULL;
3212 s->control_command_id = _SOCKET_EXEC_COMMAND_INVALID;
3213
3214 /* No further commands for this step, so let's figure
3215 * out what to do next */
3216
3217 log_unit_debug(u, "Got final SIGCHLD for state %s", socket_state_to_string(s->state));
3218
3219 switch (s->state) {
3220
3221 case SOCKET_START_PRE:
3222 if (f == SOCKET_SUCCESS)
3223 socket_enter_start_open(s);
3224 else
3225 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, f);
3226 break;
3227
3228 case SOCKET_START_CHOWN:
3229 if (f == SOCKET_SUCCESS)
3230 socket_enter_start_post(s);
3231 else
3232 socket_enter_stop_pre(s, f);
3233 break;
3234
3235 case SOCKET_START_POST:
3236 if (f == SOCKET_SUCCESS)
3237 socket_enter_listening(s);
3238 else
3239 socket_enter_stop_pre(s, f);
3240 break;
3241
3242 case SOCKET_STOP_PRE:
3243 case SOCKET_STOP_PRE_SIGTERM:
3244 case SOCKET_STOP_PRE_SIGKILL:
3245 socket_enter_stop_post(s, f);
3246 break;
3247
3248 case SOCKET_STOP_POST:
3249 case SOCKET_FINAL_SIGTERM:
3250 case SOCKET_FINAL_SIGKILL:
3251 socket_enter_dead(s, f);
3252 break;
3253
3254 case SOCKET_CLEANING:
3255
3256 if (s->clean_result == SOCKET_SUCCESS)
3257 s->clean_result = f;
3258
3259 socket_enter_dead(s, SOCKET_SUCCESS);
3260 break;
3261
3262 default:
3263 assert_not_reached();
3264 }
3265 }
3266
3267 /* Notify clients about changed exit status */
3268 unit_add_to_dbus_queue(u);
3269 }
3270
3271 static int socket_dispatch_timer(sd_event_source *source, usec_t usec, void *userdata) {
3272 Socket *s = ASSERT_PTR(SOCKET(userdata));
3273
3274 assert(s->timer_event_source == source);
3275
3276 switch (s->state) {
3277
3278 case SOCKET_START_PRE:
3279 log_unit_warning(UNIT(s), "Starting timed out. Terminating.");
3280 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, SOCKET_FAILURE_TIMEOUT);
3281 break;
3282
3283 case SOCKET_START_CHOWN:
3284 case SOCKET_START_POST:
3285 log_unit_warning(UNIT(s), "Starting timed out. Stopping.");
3286 socket_enter_stop_pre(s, SOCKET_FAILURE_TIMEOUT);
3287 break;
3288
3289 case SOCKET_STOP_PRE:
3290 log_unit_warning(UNIT(s), "Stopping timed out. Terminating.");
3291 socket_enter_signal(s, SOCKET_STOP_PRE_SIGTERM, SOCKET_FAILURE_TIMEOUT);
3292 break;
3293
3294 case SOCKET_STOP_PRE_SIGTERM:
3295 if (s->kill_context.send_sigkill) {
3296 log_unit_warning(UNIT(s), "Stopping timed out. Killing.");
3297 socket_enter_signal(s, SOCKET_STOP_PRE_SIGKILL, SOCKET_FAILURE_TIMEOUT);
3298 } else {
3299 log_unit_warning(UNIT(s), "Stopping timed out. Skipping SIGKILL. Ignoring.");
3300 socket_enter_stop_post(s, SOCKET_FAILURE_TIMEOUT);
3301 }
3302 break;
3303
3304 case SOCKET_STOP_PRE_SIGKILL:
3305 log_unit_warning(UNIT(s), "Processes still around after SIGKILL. Ignoring.");
3306 socket_enter_stop_post(s, SOCKET_FAILURE_TIMEOUT);
3307 break;
3308
3309 case SOCKET_STOP_POST:
3310 log_unit_warning(UNIT(s), "Stopping timed out (2). Terminating.");
3311 socket_enter_signal(s, SOCKET_FINAL_SIGTERM, SOCKET_FAILURE_TIMEOUT);
3312 break;
3313
3314 case SOCKET_FINAL_SIGTERM:
3315 if (s->kill_context.send_sigkill) {
3316 log_unit_warning(UNIT(s), "Stopping timed out (2). Killing.");
3317 socket_enter_signal(s, SOCKET_FINAL_SIGKILL, SOCKET_FAILURE_TIMEOUT);
3318 } else {
3319 log_unit_warning(UNIT(s), "Stopping timed out (2). Skipping SIGKILL. Ignoring.");
3320 socket_enter_dead(s, SOCKET_FAILURE_TIMEOUT);
3321 }
3322 break;
3323
3324 case SOCKET_FINAL_SIGKILL:
3325 log_unit_warning(UNIT(s), "Still around after SIGKILL (2). Entering failed mode.");
3326 socket_enter_dead(s, SOCKET_FAILURE_TIMEOUT);
3327 break;
3328
3329 case SOCKET_CLEANING:
3330 log_unit_warning(UNIT(s), "Cleaning timed out. killing.");
3331
3332 if (s->clean_result == SOCKET_SUCCESS)
3333 s->clean_result = SOCKET_FAILURE_TIMEOUT;
3334
3335 socket_enter_signal(s, SOCKET_FINAL_SIGKILL, 0);
3336 break;
3337
3338 default:
3339 assert_not_reached();
3340 }
3341
3342 return 0;
3343 }
3344
3345 int socket_collect_fds(Socket *s, int **ret) {
3346 size_t n = 0, k = 0;
3347
3348 assert(s);
3349 assert(ret);
3350
3351 /* Called from the service code for requesting our fds */
3352
3353 LIST_FOREACH(port, p, s->ports) {
3354 if (p->fd >= 0)
3355 n++;
3356 n += p->n_auxiliary_fds;
3357 }
3358
3359 if (n == 0) {
3360 *ret = NULL;
3361 return 0;
3362 }
3363
3364 int *fds = new(int, n);
3365 if (!fds)
3366 return -ENOMEM;
3367
3368 LIST_FOREACH(port, p, s->ports) {
3369 if (p->fd >= 0)
3370 fds[k++] = p->fd;
3371 FOREACH_ARRAY(i, p->auxiliary_fds, p->n_auxiliary_fds)
3372 fds[k++] = *i;
3373 }
3374
3375 assert(k == n);
3376
3377 *ret = fds;
3378 return (int) n;
3379 }
3380
3381 static void socket_reset_failed(Unit *u) {
3382 Socket *s = SOCKET(u);
3383
3384 assert(s);
3385
3386 if (s->state == SOCKET_FAILED)
3387 socket_set_state(s, SOCKET_DEAD);
3388
3389 s->result = SOCKET_SUCCESS;
3390 s->clean_result = SOCKET_SUCCESS;
3391 }
3392
3393 void socket_connection_unref(Socket *s) {
3394 assert(s);
3395
3396 /* The service is dead. Yay!
3397 *
3398 * This is strictly for one-instance-per-connection
3399 * services. */
3400
3401 assert(s->n_connections > 0);
3402 s->n_connections--;
3403
3404 log_unit_debug(UNIT(s), "One connection closed, %u left.", s->n_connections);
3405 }
3406
3407 static void socket_trigger_notify(Unit *u, Unit *other) {
3408 Socket *s = ASSERT_PTR(SOCKET(u));
3409
3410 assert(other);
3411
3412 /* Filter out invocations with bogus state */
3413 assert(UNIT_IS_LOAD_COMPLETE(other->load_state));
3414
3415 Service *service = ASSERT_PTR(SERVICE(other));
3416
3417 /* Don't propagate state changes from the service if we are already down */
3418 if (!IN_SET(s->state, SOCKET_RUNNING, SOCKET_LISTENING))
3419 return;
3420
3421 /* We don't care for the service state if we are in Accept=yes mode */
3422 if (s->accept)
3423 return;
3424
3425 /* Propagate start limit hit state */
3426 if (other->start_limit_hit) {
3427 socket_enter_stop_pre(s, SOCKET_FAILURE_SERVICE_START_LIMIT_HIT);
3428 return;
3429 }
3430
3431 /* Don't propagate anything if there's still a job queued */
3432 if (other->job)
3433 return;
3434
3435 if (!SOCKET_SERVICE_IS_ACTIVE(service, /* allow_finalize = */ true))
3436 socket_enter_listening(s);
3437
3438 if (SERVICE(other)->state == SERVICE_RUNNING)
3439 socket_set_state(s, SOCKET_RUNNING);
3440 }
3441
3442 static void socket_handoff_timestamp(
3443 Unit *u,
3444 const struct ucred *ucred,
3445 const dual_timestamp *ts) {
3446
3447 Socket *s = ASSERT_PTR(SOCKET(u));
3448
3449 assert(ucred);
3450 assert(ts);
3451
3452 if (s->control_pid.pid == ucred->pid && s->control_command) {
3453 exec_status_handoff(&s->control_command->exec_status, ucred, ts);
3454 unit_add_to_dbus_queue(u);
3455 }
3456 }
3457
3458 static int socket_get_timeout(Unit *u, usec_t *timeout) {
3459 Socket *s = ASSERT_PTR(SOCKET(u));
3460 usec_t t;
3461 int r;
3462
3463 if (!s->timer_event_source)
3464 return 0;
3465
3466 r = sd_event_source_get_time(s->timer_event_source, &t);
3467 if (r < 0)
3468 return r;
3469 if (t == USEC_INFINITY)
3470 return 0;
3471
3472 *timeout = t;
3473 return 1;
3474 }
3475
3476 const char* socket_fdname(Socket *s) {
3477 assert(s);
3478
3479 /* Returns the name to use for $LISTEN_FDNAMES. If the user didn't specify anything specifically,
3480 * use the socket unit's name as fallback for Accept=no sockets, "connection" otherwise. */
3481
3482 if (s->fdname)
3483 return s->fdname;
3484
3485 if (s->accept)
3486 return "connection";
3487
3488 return UNIT(s)->id;
3489 }
3490
3491 static PidRef* socket_control_pid(Unit *u) {
3492 return &ASSERT_PTR(SOCKET(u))->control_pid;
3493 }
3494
3495 static int socket_clean(Unit *u, ExecCleanMask mask) {
3496 Socket *s = ASSERT_PTR(SOCKET(u));
3497 _cleanup_strv_free_ char **l = NULL;
3498 int r;
3499
3500 assert(mask != 0);
3501
3502 if (s->state != SOCKET_DEAD)
3503 return -EBUSY;
3504
3505 r = exec_context_get_clean_directories(&s->exec_context, u->manager->prefix, mask, &l);
3506 if (r < 0)
3507 return r;
3508
3509 if (strv_isempty(l))
3510 return -EUNATCH;
3511
3512 socket_unwatch_control_pid(s);
3513 s->clean_result = SOCKET_SUCCESS;
3514 s->control_command = NULL;
3515 s->control_command_id = _SOCKET_EXEC_COMMAND_INVALID;
3516
3517 r = socket_arm_timer(s, /* relative= */ true, s->exec_context.timeout_clean_usec);
3518 if (r < 0) {
3519 log_unit_warning_errno(u, r, "Failed to install timer: %m");
3520 goto fail;
3521 }
3522
3523 r = unit_fork_and_watch_rm_rf(u, l, &s->control_pid);
3524 if (r < 0) {
3525 log_unit_warning_errno(u, r, "Failed to spawn cleaning task: %m");
3526 goto fail;
3527 }
3528
3529 socket_set_state(s, SOCKET_CLEANING);
3530 return 0;
3531
3532 fail:
3533 s->clean_result = SOCKET_FAILURE_RESOURCES;
3534 s->timer_event_source = sd_event_source_disable_unref(s->timer_event_source);
3535 return r;
3536 }
3537
3538 static int socket_can_clean(Unit *u, ExecCleanMask *ret) {
3539 Socket *s = ASSERT_PTR(SOCKET(u));
3540
3541 return exec_context_get_clean_mask(&s->exec_context, ret);
3542 }
3543
3544 static int socket_can_start(Unit *u) {
3545 Socket *s = ASSERT_PTR(SOCKET(u));
3546 int r;
3547
3548 r = unit_test_start_limit(u);
3549 if (r < 0) {
3550 socket_enter_dead(s, SOCKET_FAILURE_START_LIMIT_HIT);
3551 return r;
3552 }
3553
3554 return 1;
3555 }
3556
3557 static const char* const socket_exec_command_table[_SOCKET_EXEC_COMMAND_MAX] = {
3558 [SOCKET_EXEC_START_PRE] = "ExecStartPre",
3559 [SOCKET_EXEC_START_CHOWN] = "ExecStartChown",
3560 [SOCKET_EXEC_START_POST] = "ExecStartPost",
3561 [SOCKET_EXEC_STOP_PRE] = "ExecStopPre",
3562 [SOCKET_EXEC_STOP_POST] = "ExecStopPost",
3563 };
3564
3565 DEFINE_STRING_TABLE_LOOKUP(socket_exec_command, SocketExecCommand);
3566
3567 static const char* const socket_result_table[_SOCKET_RESULT_MAX] = {
3568 [SOCKET_SUCCESS] = "success",
3569 [SOCKET_FAILURE_RESOURCES] = "resources",
3570 [SOCKET_FAILURE_TIMEOUT] = "timeout",
3571 [SOCKET_FAILURE_EXIT_CODE] = "exit-code",
3572 [SOCKET_FAILURE_SIGNAL] = "signal",
3573 [SOCKET_FAILURE_CORE_DUMP] = "core-dump",
3574 [SOCKET_FAILURE_START_LIMIT_HIT] = "start-limit-hit",
3575 [SOCKET_FAILURE_TRIGGER_LIMIT_HIT] = "trigger-limit-hit",
3576 [SOCKET_FAILURE_SERVICE_START_LIMIT_HIT] = "service-start-limit-hit",
3577 };
3578
3579 DEFINE_STRING_TABLE_LOOKUP(socket_result, SocketResult);
3580
3581 static const char* const socket_timestamping_table[_SOCKET_TIMESTAMPING_MAX] = {
3582 [SOCKET_TIMESTAMPING_OFF] = "off",
3583 [SOCKET_TIMESTAMPING_US] = "us",
3584 [SOCKET_TIMESTAMPING_NS] = "ns",
3585 };
3586
3587 DEFINE_STRING_TABLE_LOOKUP(socket_timestamping, SocketTimestamping);
3588
3589 SocketTimestamping socket_timestamping_from_string_harder(const char *p) {
3590 SocketTimestamping t;
3591 int r;
3592
3593 if (!p)
3594 return _SOCKET_TIMESTAMPING_INVALID;
3595
3596 t = socket_timestamping_from_string(p);
3597 if (t >= 0)
3598 return t;
3599
3600 /* Let's alternatively support the various other aliases parse_time() accepts for ns and µs here,
3601 * too. */
3602 if (streq(p, "nsec"))
3603 return SOCKET_TIMESTAMPING_NS;
3604 if (STR_IN_SET(p, "usec", "µs", "μs")) /* Accept both small greek letter mu + micro sign unicode codepoints */
3605 return SOCKET_TIMESTAMPING_US;
3606
3607 r = parse_boolean(p);
3608 if (r < 0)
3609 return _SOCKET_TIMESTAMPING_INVALID;
3610
3611 return r ? SOCKET_TIMESTAMPING_NS : SOCKET_TIMESTAMPING_OFF; /* If boolean yes, default to ns accuracy */
3612 }
3613
3614 const UnitVTable socket_vtable = {
3615 .object_size = sizeof(Socket),
3616 .exec_context_offset = offsetof(Socket, exec_context),
3617 .cgroup_context_offset = offsetof(Socket, cgroup_context),
3618 .kill_context_offset = offsetof(Socket, kill_context),
3619 .exec_runtime_offset = offsetof(Socket, exec_runtime),
3620 .cgroup_runtime_offset = offsetof(Socket, cgroup_runtime),
3621
3622 .sections =
3623 "Unit\0"
3624 "Socket\0"
3625 "Install\0",
3626 .private_section = "Socket",
3627
3628 .can_transient = true,
3629 .can_trigger = true,
3630 .can_fail = true,
3631
3632 .init = socket_init,
3633 .done = socket_done,
3634 .load = socket_load,
3635
3636 .coldplug = socket_coldplug,
3637
3638 .dump = socket_dump,
3639
3640 .start = socket_start,
3641 .stop = socket_stop,
3642
3643 .clean = socket_clean,
3644 .can_clean = socket_can_clean,
3645
3646 .get_timeout = socket_get_timeout,
3647
3648 .serialize = socket_serialize,
3649 .deserialize_item = socket_deserialize_item,
3650 .distribute_fds = socket_distribute_fds,
3651
3652 .active_state = socket_active_state,
3653 .sub_state_to_string = socket_sub_state_to_string,
3654
3655 .will_restart = unit_will_restart_default,
3656
3657 .may_gc = socket_may_gc,
3658
3659 .sigchld_event = socket_sigchld_event,
3660
3661 .trigger_notify = socket_trigger_notify,
3662
3663 .reset_failed = socket_reset_failed,
3664
3665 .notify_handoff_timestamp = socket_handoff_timestamp,
3666
3667 .control_pid = socket_control_pid,
3668
3669 .bus_set_property = bus_socket_set_property,
3670 .bus_commit_properties = bus_socket_commit_properties,
3671
3672 .status_message_formats = {
3673 .finished_start_job = {
3674 [JOB_DONE] = "Listening on %s.",
3675 [JOB_FAILED] = "Failed to listen on %s.",
3676 [JOB_TIMEOUT] = "Timed out starting %s.",
3677 },
3678 .finished_stop_job = {
3679 [JOB_DONE] = "Closed %s.",
3680 [JOB_FAILED] = "Failed stopping %s.",
3681 [JOB_TIMEOUT] = "Timed out stopping %s.",
3682 },
3683 },
3684
3685 .can_start = socket_can_start,
3686 };
Unnamed repository; edit this file 'description' to name the repository.