]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
6069edb0 | 2 | * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2015 Nicira, Inc. |
064af421 | 3 | * |
a14bc59f BP |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
064af421 | 7 | * |
a14bc59f BP |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
064af421 BP |
15 | */ |
16 | ||
17 | #include <config.h> | |
18 | #include "daemon.h" | |
3834bcf2 | 19 | #include "daemon-private.h" |
064af421 BP |
20 | #include <errno.h> |
21 | #include <fcntl.h> | |
e91b927d AZ |
22 | #include <grp.h> |
23 | #include <pwd.h> | |
3762274e | 24 | #include <signal.h> |
064af421 BP |
25 | #include <stdlib.h> |
26 | #include <string.h> | |
3762274e | 27 | #include <sys/resource.h> |
95440284 | 28 | #include <sys/wait.h> |
309eaa2b | 29 | #include <sys/stat.h> |
064af421 | 30 | #include <unistd.h> |
e91b927d AZ |
31 | #if HAVE_LIBCAPNG |
32 | #include <cap-ng.h> | |
33 | #endif | |
40f0707c | 34 | #include "command-line.h" |
064af421 BP |
35 | #include "fatal-signal.h" |
36 | #include "dirs.h" | |
ac718c9d | 37 | #include "lockfile.h" |
728a8b14 | 38 | #include "ovs-thread.h" |
ff8decf1 | 39 | #include "process.h" |
b8781ff0 | 40 | #include "socket-util.h" |
03fbffbd | 41 | #include "timeval.h" |
064af421 | 42 | #include "util.h" |
e6211adc | 43 | #include "openvswitch/vlog.h" |
064af421 | 44 | |
a91dc444 | 45 | VLOG_DEFINE_THIS_MODULE(daemon_unix); |
5136ce49 | 46 | |
e91b927d AZ |
47 | #ifdef __linux__ |
48 | #define LINUX 1 | |
49 | #else | |
50 | #define LINUX 0 | |
51 | #endif | |
52 | ||
53 | #if HAVE_LIBCAPNG | |
54 | #define LIBCAPNG 1 | |
55 | #else | |
56 | #define LIBCAPNG 0 | |
57 | #endif | |
58 | ||
d4db8309 | 59 | /* --detach: Should we run in the background? */ |
3834bcf2 | 60 | bool detach; /* Was --detach specified? */ |
e8087a87 | 61 | static bool detached; /* Have we already detached? */ |
064af421 | 62 | |
d4db8309 | 63 | /* --pidfile: Name of pidfile (null if none). */ |
3834bcf2 | 64 | char *pidfile; |
064af421 | 65 | |
e4bd5e2a BP |
66 | /* Device and inode of pidfile, so we can avoid reopening it. */ |
67 | static dev_t pidfile_dev; | |
68 | static ino_t pidfile_ino; | |
69 | ||
d4db8309 BP |
70 | /* --overwrite-pidfile: Create pidfile even if one already exists and is |
71 | locked? */ | |
e7bd7d78 | 72 | static bool overwrite_pidfile; |
064af421 | 73 | |
d4db8309 | 74 | /* --no-chdir: Should we chdir to "/"? */ |
91a1e24d JP |
75 | static bool chdir_ = true; |
76 | ||
7943cd51 BP |
77 | /* File descriptor used by daemonize_start() and daemonize_complete(). */ |
78 | static int daemonize_fd = -1; | |
95440284 | 79 | |
ff8decf1 BP |
80 | /* --monitor: Should a supervisory process monitor the daemon and restart it if |
81 | * it dies due to an error signal? */ | |
82 | static bool monitor; | |
83 | ||
e91b927d AZ |
84 | /* --user: Only root can use this option. Switch to new uid:gid after |
85 | * initially running as root. */ | |
86 | static bool switch_user = false; | |
e91b927d AZ |
87 | static uid_t uid; |
88 | static gid_t gid; | |
89 | static char *user = NULL; | |
90 | static void daemon_become_new_user__(bool access_datapath); | |
91 | ||
aacea8ba BP |
92 | static void check_already_running(void); |
93 | static int lock_pidfile(FILE *, int command); | |
d6056bc7 GS |
94 | static pid_t fork_and_clean_up(void); |
95 | static void daemonize_post_detach(void); | |
aacea8ba | 96 | |
064af421 BP |
97 | /* Returns the file name that would be used for a pidfile if 'name' were |
98 | * provided to set_pidfile(). The caller must free the returned string. */ | |
3834bcf2 | 99 | char * |
d295e8e9 | 100 | make_pidfile_name(const char *name) |
064af421 | 101 | { |
daf03c53 | 102 | return (!name |
b43c6fe2 BP |
103 | ? xasprintf("%s/%s.pid", ovs_rundir(), program_name) |
104 | : abs_file_name(ovs_rundir(), name)); | |
064af421 BP |
105 | } |
106 | ||
91a1e24d JP |
107 | /* Sets that we do not chdir to "/". */ |
108 | void | |
109 | set_no_chdir(void) | |
110 | { | |
111 | chdir_ = false; | |
112 | } | |
113 | ||
00c08589 BP |
114 | /* Normally, daemonize() or damonize_start() will terminate the program with a |
115 | * message if a locked pidfile already exists. If this function is called, an | |
116 | * existing pidfile will be replaced, with a warning. */ | |
064af421 BP |
117 | void |
118 | ignore_existing_pidfile(void) | |
119 | { | |
e7bd7d78 | 120 | overwrite_pidfile = true; |
064af421 BP |
121 | } |
122 | ||
123 | /* Sets up a following call to daemonize() to detach from the foreground | |
124 | * session, running this process in the background. */ | |
125 | void | |
126 | set_detach(void) | |
127 | { | |
128 | detach = true; | |
129 | } | |
130 | ||
ff8decf1 BP |
131 | /* Sets up a following call to daemonize() to fork a supervisory process to |
132 | * monitor the daemon and restart it if it dies due to an error signal. */ | |
133 | void | |
134 | daemon_set_monitor(void) | |
135 | { | |
136 | monitor = true; | |
137 | } | |
138 | ||
d4db8309 BP |
139 | /* If a pidfile has been configured, creates it and stores the running |
140 | * process's pid in it. Ensures that the pidfile will be deleted when the | |
141 | * process exits. */ | |
064af421 BP |
142 | static void |
143 | make_pidfile(void) | |
144 | { | |
aacea8ba BP |
145 | long int pid = getpid(); |
146 | struct stat s; | |
147 | char *tmpfile; | |
148 | FILE *file; | |
149 | int error; | |
150 | ||
151 | /* Create a temporary pidfile. */ | |
2388a783 EJ |
152 | if (overwrite_pidfile) { |
153 | tmpfile = xasprintf("%s.tmp%ld", pidfile, pid); | |
154 | fatal_signal_add_file_to_unlink(tmpfile); | |
155 | } else { | |
156 | /* Everyone shares the same file which will be treated as a lock. To | |
157 | * avoid some uncomfortable race conditions, we can't set up the fatal | |
158 | * signal unlink until we've acquired it. */ | |
159 | tmpfile = xasprintf("%s.tmp", pidfile); | |
160 | } | |
161 | ||
162 | file = fopen(tmpfile, "a+"); | |
aacea8ba | 163 | if (!file) { |
10a89ef0 | 164 | VLOG_FATAL("%s: create failed (%s)", tmpfile, ovs_strerror(errno)); |
aacea8ba BP |
165 | } |
166 | ||
2388a783 EJ |
167 | error = lock_pidfile(file, F_SETLK); |
168 | if (error) { | |
169 | /* Looks like we failed to acquire the lock. Note that, if we failed | |
170 | * for some other reason (and '!overwrite_pidfile'), we will have | |
171 | * left 'tmpfile' as garbage in the file system. */ | |
10a89ef0 BP |
172 | VLOG_FATAL("%s: fcntl(F_SETLK) failed (%s)", tmpfile, |
173 | ovs_strerror(error)); | |
2388a783 EJ |
174 | } |
175 | ||
176 | if (!overwrite_pidfile) { | |
177 | /* We acquired the lock. Make sure to clean up on exit, and verify | |
178 | * that we're allowed to create the actual pidfile. */ | |
179 | fatal_signal_add_file_to_unlink(tmpfile); | |
180 | check_already_running(); | |
181 | } | |
182 | ||
aacea8ba | 183 | if (fstat(fileno(file), &s) == -1) { |
10a89ef0 | 184 | VLOG_FATAL("%s: fstat failed (%s)", tmpfile, ovs_strerror(errno)); |
aacea8ba BP |
185 | } |
186 | ||
2388a783 | 187 | if (ftruncate(fileno(file), 0) == -1) { |
10a89ef0 | 188 | VLOG_FATAL("%s: truncate failed (%s)", tmpfile, ovs_strerror(errno)); |
2388a783 EJ |
189 | } |
190 | ||
aacea8ba BP |
191 | fprintf(file, "%ld\n", pid); |
192 | if (fflush(file) == EOF) { | |
10a89ef0 | 193 | VLOG_FATAL("%s: write failed (%s)", tmpfile, ovs_strerror(errno)); |
aacea8ba BP |
194 | } |
195 | ||
2388a783 | 196 | error = rename(tmpfile, pidfile); |
aacea8ba | 197 | |
2388a783 EJ |
198 | /* Due to a race, 'tmpfile' may be owned by a different process, so we |
199 | * shouldn't delete it on exit. */ | |
200 | fatal_signal_remove_file_to_unlink(tmpfile); | |
201 | ||
202 | if (error < 0) { | |
203 | VLOG_FATAL("failed to rename \"%s\" to \"%s\" (%s)", | |
10a89ef0 | 204 | tmpfile, pidfile, ovs_strerror(errno)); |
aacea8ba BP |
205 | } |
206 | ||
207 | /* Ensure that the pidfile will get deleted on exit. */ | |
208 | fatal_signal_add_file_to_unlink(pidfile); | |
209 | ||
aacea8ba BP |
210 | /* Clean up. |
211 | * | |
212 | * We don't close 'file' because its file descriptor must remain open to | |
213 | * hold the lock. */ | |
214 | pidfile_dev = s.st_dev; | |
215 | pidfile_ino = s.st_ino; | |
216 | free(tmpfile); | |
064af421 BP |
217 | } |
218 | ||
8aee05cc BP |
219 | /* Calls fork() and on success returns its return value. On failure, logs an |
220 | * error and exits unsuccessfully. | |
221 | * | |
222 | * Post-fork, but before returning, this function calls a few other functions | |
223 | * that are generally useful if the child isn't planning to exec a new | |
224 | * process. */ | |
d6056bc7 | 225 | static pid_t |
8aee05cc BP |
226 | fork_and_clean_up(void) |
227 | { | |
728a8b14 | 228 | pid_t pid = xfork(); |
8aee05cc BP |
229 | if (pid > 0) { |
230 | /* Running in parent process. */ | |
231 | fatal_signal_fork(); | |
232 | } else if (!pid) { | |
233 | /* Running in child process. */ | |
8aee05cc | 234 | lockfile_postfork(); |
8aee05cc | 235 | } |
8aee05cc BP |
236 | return pid; |
237 | } | |
238 | ||
e6c5e539 BP |
239 | /* Forks, then: |
240 | * | |
241 | * - In the parent, waits for the child to signal that it has completed its | |
b925336a AA |
242 | * startup sequence. Then stores -1 in '*fdp' and returns the child's |
243 | * pid in '*child_pid' argument. | |
e6c5e539 | 244 | * |
b925336a AA |
245 | * - In the child, stores a fd in '*fdp' and returns 0 through '*child_pid' |
246 | * argument. The caller should pass the fd to fork_notify_startup() after | |
247 | * it finishes its startup sequence. | |
e6c5e539 | 248 | * |
b925336a AA |
249 | * Returns 0 on success. If something goes wrong and child process was not |
250 | * able to signal its readiness by calling fork_notify_startup(), then this | |
251 | * function returns -1. However, even in case of failure it still sets child | |
252 | * process id in '*child_pid'. */ | |
253 | static int | |
254 | fork_and_wait_for_startup(int *fdp, pid_t *child_pid) | |
7943cd51 BP |
255 | { |
256 | int fds[2]; | |
257 | pid_t pid; | |
b925336a | 258 | int ret = 0; |
7943cd51 | 259 | |
279c9e03 | 260 | xpipe(fds); |
7943cd51 | 261 | |
8aee05cc | 262 | pid = fork_and_clean_up(); |
7943cd51 BP |
263 | if (pid > 0) { |
264 | /* Running in parent process. */ | |
af9a1442 | 265 | size_t bytes_read; |
7943cd51 BP |
266 | char c; |
267 | ||
268 | close(fds[1]); | |
af9a1442 | 269 | if (read_fully(fds[0], &c, 1, &bytes_read) != 0) { |
7943cd51 BP |
270 | int retval; |
271 | int status; | |
272 | ||
273 | do { | |
274 | retval = waitpid(pid, &status, 0); | |
275 | } while (retval == -1 && errno == EINTR); | |
276 | ||
2c8fcc9c BP |
277 | if (retval == pid) { |
278 | if (WIFEXITED(status) && WEXITSTATUS(status)) { | |
279 | /* Child exited with an error. Convey the same error | |
280 | * to our parent process as a courtesy. */ | |
281 | exit(WEXITSTATUS(status)); | |
282 | } else { | |
283 | char *status_msg = process_status_msg(status); | |
b925336a AA |
284 | VLOG_ERR("fork child died before signaling startup (%s)", |
285 | status_msg); | |
286 | ret = -1; | |
2c8fcc9c BP |
287 | } |
288 | } else if (retval < 0) { | |
10a89ef0 | 289 | VLOG_FATAL("waitpid failed (%s)", ovs_strerror(errno)); |
2c8fcc9c | 290 | } else { |
428b2edd | 291 | OVS_NOT_REACHED(); |
7943cd51 | 292 | } |
7943cd51 BP |
293 | } |
294 | close(fds[0]); | |
295 | *fdp = -1; | |
296 | } else if (!pid) { | |
297 | /* Running in child process. */ | |
298 | close(fds[0]); | |
7943cd51 | 299 | *fdp = fds[1]; |
7943cd51 | 300 | } |
b925336a AA |
301 | *child_pid = pid; |
302 | return ret; | |
7943cd51 BP |
303 | } |
304 | ||
305 | static void | |
306 | fork_notify_startup(int fd) | |
307 | { | |
308 | if (fd != -1) { | |
309 | size_t bytes_written; | |
310 | int error; | |
311 | ||
312 | error = write_fully(fd, "", 1, &bytes_written); | |
313 | if (error) { | |
10a89ef0 | 314 | VLOG_FATAL("pipe write failed (%s)", ovs_strerror(error)); |
7943cd51 BP |
315 | } |
316 | ||
317 | close(fd); | |
318 | } | |
319 | } | |
320 | ||
ff8decf1 BP |
321 | static bool |
322 | should_restart(int status) | |
323 | { | |
324 | if (WIFSIGNALED(status)) { | |
325 | static const int error_signals[] = { | |
f67c3295 BP |
326 | /* This list of signals is documented in daemon.man. If you |
327 | * change the list, update the documentation too. */ | |
ff8decf1 BP |
328 | SIGABRT, SIGALRM, SIGBUS, SIGFPE, SIGILL, SIGPIPE, SIGSEGV, |
329 | SIGXCPU, SIGXFSZ | |
330 | }; | |
331 | ||
332 | size_t i; | |
333 | ||
334 | for (i = 0; i < ARRAY_SIZE(error_signals); i++) { | |
335 | if (error_signals[i] == WTERMSIG(status)) { | |
336 | return true; | |
337 | } | |
338 | } | |
339 | } | |
340 | return false; | |
341 | } | |
342 | ||
343 | static void | |
344 | monitor_daemon(pid_t daemon_pid) | |
345 | { | |
ff8decf1 | 346 | /* XXX Should log daemon's stderr output at startup time. */ |
a9633ada | 347 | time_t last_restart; |
40f0707c | 348 | char *status_msg; |
cbbdf81c | 349 | int crashes; |
b925336a | 350 | bool child_ready = true; |
ff8decf1 | 351 | |
bc9fb3a9 | 352 | set_subprogram_name("monitor"); |
40f0707c | 353 | status_msg = xstrdup("healthy"); |
a9633ada | 354 | last_restart = TIME_MIN; |
cbbdf81c | 355 | crashes = 0; |
ff8decf1 BP |
356 | for (;;) { |
357 | int retval; | |
358 | int status; | |
359 | ||
5f383751 RB |
360 | ovs_cmdl_proctitle_set("monitoring pid %lu (%s)", |
361 | (unsigned long int) daemon_pid, status_msg); | |
40f0707c | 362 | |
b925336a AA |
363 | if (child_ready) { |
364 | do { | |
365 | retval = waitpid(daemon_pid, &status, 0); | |
366 | } while (retval == -1 && errno == EINTR); | |
367 | if (retval == -1) { | |
368 | VLOG_FATAL("waitpid failed (%s)", ovs_strerror(errno)); | |
369 | } | |
370 | } | |
ff8decf1 | 371 | |
b925336a | 372 | if (!child_ready || retval == daemon_pid) { |
40f0707c | 373 | char *s = process_status_msg(status); |
40f0707c | 374 | if (should_restart(status)) { |
2bf9d87a BP |
375 | free(status_msg); |
376 | status_msg = xasprintf("%d crashes: pid %lu died, %s", | |
377 | ++crashes, | |
378 | (unsigned long int) daemon_pid, s); | |
379 | free(s); | |
380 | ||
7c2dd4c6 BP |
381 | if (WCOREDUMP(status)) { |
382 | /* Disable further core dumps to save disk space. */ | |
383 | struct rlimit r; | |
384 | ||
385 | r.rlim_cur = 0; | |
386 | r.rlim_max = 0; | |
387 | if (setrlimit(RLIMIT_CORE, &r) == -1) { | |
388 | VLOG_WARN("failed to disable core dumps: %s", | |
10a89ef0 | 389 | ovs_strerror(errno)); |
7c2dd4c6 BP |
390 | } |
391 | } | |
392 | ||
a9633ada BP |
393 | /* Throttle restarts to no more than once every 10 seconds. */ |
394 | if (time(NULL) < last_restart + 10) { | |
395 | VLOG_WARN("%s, waiting until 10 seconds since last " | |
396 | "restart", status_msg); | |
397 | for (;;) { | |
398 | time_t now = time(NULL); | |
399 | time_t wakeup = last_restart + 10; | |
400 | if (now >= wakeup) { | |
401 | break; | |
402 | } | |
275eebb9 | 403 | xsleep(wakeup - now); |
a9633ada BP |
404 | } |
405 | } | |
406 | last_restart = time(NULL); | |
407 | ||
40f0707c | 408 | VLOG_ERR("%s, restarting", status_msg); |
b925336a AA |
409 | child_ready = !fork_and_wait_for_startup(&daemonize_fd, |
410 | &daemon_pid); | |
411 | if (child_ready && !daemon_pid) { | |
412 | /* Child process needs to break out of monitoring | |
413 | * loop. */ | |
ff8decf1 BP |
414 | break; |
415 | } | |
416 | } else { | |
2bf9d87a BP |
417 | VLOG_INFO("pid %lu died, %s, exiting", |
418 | (unsigned long int) daemon_pid, s); | |
419 | free(s); | |
ff8decf1 BP |
420 | exit(0); |
421 | } | |
422 | } | |
423 | } | |
b2d06cb8 | 424 | free(status_msg); |
ff8decf1 BP |
425 | |
426 | /* Running in new daemon process. */ | |
5f383751 | 427 | ovs_cmdl_proctitle_restore(); |
bc9fb3a9 | 428 | set_subprogram_name(""); |
ff8decf1 BP |
429 | } |
430 | ||
95440284 BP |
431 | /* If daemonization is configured, then starts daemonization, by forking and |
432 | * returning in the child process. The parent process hangs around until the | |
433 | * child lets it know either that it completed startup successfully (by calling | |
434 | * daemon_complete()) or that it failed to start up (by exiting with a nonzero | |
435 | * exit code). */ | |
436 | void | |
e91b927d | 437 | daemonize_start(bool access_datapath) |
064af421 | 438 | { |
728a8b14 | 439 | assert_single_threaded(); |
7943cd51 | 440 | daemonize_fd = -1; |
95440284 | 441 | |
e91b927d AZ |
442 | if (switch_user) { |
443 | daemon_become_new_user__(access_datapath); | |
444 | switch_user = false; | |
445 | } | |
446 | ||
7943cd51 | 447 | if (detach) { |
b925336a AA |
448 | pid_t pid; |
449 | ||
450 | if (fork_and_wait_for_startup(&daemonize_fd, &pid)) { | |
451 | VLOG_FATAL("could not detach from foreground session"); | |
452 | } | |
453 | if (pid > 0) { | |
95440284 | 454 | /* Running in parent process. */ |
064af421 | 455 | exit(0); |
064af421 | 456 | } |
066f329e | 457 | |
ff8decf1 | 458 | /* Running in daemon or monitor process. */ |
066f329e | 459 | setsid(); |
ff8decf1 BP |
460 | } |
461 | ||
462 | if (monitor) { | |
463 | int saved_daemonize_fd = daemonize_fd; | |
464 | pid_t daemon_pid; | |
465 | ||
b925336a AA |
466 | if (fork_and_wait_for_startup(&daemonize_fd, &daemon_pid)) { |
467 | VLOG_FATAL("could not initiate process monitoring"); | |
468 | } | |
ff8decf1 BP |
469 | if (daemon_pid > 0) { |
470 | /* Running in monitor process. */ | |
471 | fork_notify_startup(saved_daemonize_fd); | |
472 | close_standard_fds(); | |
473 | monitor_daemon(daemon_pid); | |
474 | } | |
7943cd51 | 475 | /* Running in daemon process. */ |
064af421 | 476 | } |
7943cd51 | 477 | |
92fa2e92 BP |
478 | forbid_forking("running in daemon process"); |
479 | ||
aacea8ba BP |
480 | if (pidfile) { |
481 | make_pidfile(); | |
482 | } | |
df5d2ed9 BP |
483 | |
484 | /* Make sure that the unixctl commands for vlog get registered in a | |
485 | * daemon, even before the first log message. */ | |
486 | vlog_init(); | |
064af421 BP |
487 | } |
488 | ||
95440284 | 489 | /* If daemonization is configured, then this function notifies the parent |
e8087a87 BP |
490 | * process that the child process has completed startup successfully. It also |
491 | * call daemonize_post_detach(). | |
a7ff9bd7 BP |
492 | * |
493 | * Calling this function more than once has no additional effect. */ | |
95440284 BP |
494 | void |
495 | daemonize_complete(void) | |
496 | { | |
7ffd3f69 GS |
497 | if (pidfile) { |
498 | free(pidfile); | |
499 | pidfile = NULL; | |
500 | } | |
501 | ||
e8087a87 BP |
502 | if (!detached) { |
503 | detached = true; | |
504 | ||
505 | fork_notify_startup(daemonize_fd); | |
506 | daemonize_fd = -1; | |
507 | daemonize_post_detach(); | |
508 | } | |
509 | } | |
95440284 | 510 | |
e8087a87 BP |
511 | /* If daemonization is configured, then this function does traditional Unix |
512 | * daemonization behavior: join a new session, chdir to the root (if not | |
513 | * disabled), and close the standard file descriptors. | |
514 | * | |
515 | * It only makes sense to call this function as part of an implementation of a | |
516 | * special daemon subprocess. A normal daemon should just call | |
517 | * daemonize_complete(). */ | |
d6056bc7 | 518 | static void |
e8087a87 BP |
519 | daemonize_post_detach(void) |
520 | { | |
7943cd51 | 521 | if (detach) { |
95440284 BP |
522 | if (chdir_) { |
523 | ignore(chdir("/")); | |
524 | } | |
7943cd51 | 525 | close_standard_fds(); |
95440284 BP |
526 | } |
527 | } | |
528 | ||
064af421 BP |
529 | void |
530 | daemon_usage(void) | |
531 | { | |
532 | printf( | |
533 | "\nDaemon options:\n" | |
e7bd7d78 | 534 | " --detach run in background as daemon\n" |
91a1e24d | 535 | " --no-chdir do not chdir to '/'\n" |
e7bd7d78 JP |
536 | " --pidfile[=FILE] create pidfile (default: %s/%s.pid)\n" |
537 | " --overwrite-pidfile with --pidfile, start even if already " | |
538 | "running\n", | |
b43c6fe2 | 539 | ovs_rundir(), program_name); |
064af421 BP |
540 | } |
541 | ||
aacea8ba BP |
542 | static int |
543 | lock_pidfile__(FILE *file, int command, struct flock *lck) | |
544 | { | |
545 | int error; | |
546 | ||
547 | lck->l_type = F_WRLCK; | |
548 | lck->l_whence = SEEK_SET; | |
549 | lck->l_start = 0; | |
550 | lck->l_len = 0; | |
551 | lck->l_pid = 0; | |
552 | ||
553 | do { | |
554 | error = fcntl(fileno(file), command, lck) == -1 ? errno : 0; | |
555 | } while (error == EINTR); | |
556 | return error; | |
557 | } | |
558 | ||
559 | static int | |
560 | lock_pidfile(FILE *file, int command) | |
561 | { | |
562 | struct flock lck; | |
563 | ||
564 | return lock_pidfile__(file, command, &lck); | |
565 | } | |
566 | ||
18e124a2 | 567 | static pid_t |
aacea8ba | 568 | read_pidfile__(const char *pidfile, bool delete_if_stale) |
064af421 | 569 | { |
aacea8ba | 570 | struct stat s, s2; |
064af421 | 571 | struct flock lck; |
aacea8ba | 572 | char line[128]; |
064af421 BP |
573 | FILE *file; |
574 | int error; | |
575 | ||
e4bd5e2a BP |
576 | if ((pidfile_ino || pidfile_dev) |
577 | && !stat(pidfile, &s) | |
578 | && s.st_ino == pidfile_ino && s.st_dev == pidfile_dev) { | |
579 | /* It's our own pidfile. We can't afford to open it, because closing | |
580 | * *any* fd for a file that a process has locked also releases all the | |
581 | * locks on that file. | |
582 | * | |
583 | * Fortunately, we know the associated pid anyhow: */ | |
584 | return getpid(); | |
585 | } | |
586 | ||
aacea8ba | 587 | file = fopen(pidfile, "r+"); |
064af421 | 588 | if (!file) { |
aacea8ba | 589 | if (errno == ENOENT && delete_if_stale) { |
18e124a2 BP |
590 | return 0; |
591 | } | |
064af421 | 592 | error = errno; |
10a89ef0 | 593 | VLOG_WARN("%s: open: %s", pidfile, ovs_strerror(error)); |
064af421 BP |
594 | goto error; |
595 | } | |
596 | ||
aacea8ba BP |
597 | error = lock_pidfile__(file, F_GETLK, &lck); |
598 | if (error) { | |
10a89ef0 | 599 | VLOG_WARN("%s: fcntl: %s", pidfile, ovs_strerror(error)); |
064af421 BP |
600 | goto error; |
601 | } | |
602 | if (lck.l_type == F_UNLCK) { | |
aacea8ba BP |
603 | /* pidfile exists but it isn't locked by anyone. We need to delete it |
604 | * so that a new pidfile can go in its place. But just calling | |
605 | * unlink(pidfile) makes a nasty race: what if someone else unlinks it | |
606 | * before we do and then replaces it by a valid pidfile? We'd unlink | |
607 | * their valid pidfile. We do a little dance to avoid the race, by | |
608 | * locking the invalid pidfile. Only one process can have the invalid | |
609 | * pidfile locked, and only that process has the right to unlink it. */ | |
610 | if (!delete_if_stale) { | |
611 | error = ESRCH; | |
0b376942 | 612 | VLOG_DBG("%s: pid file is stale", pidfile); |
aacea8ba BP |
613 | goto error; |
614 | } | |
615 | ||
616 | /* Get the lock. */ | |
617 | error = lock_pidfile(file, F_SETLK); | |
618 | if (error) { | |
619 | /* We lost a race with someone else doing the same thing. */ | |
620 | VLOG_WARN("%s: lost race to lock pidfile", pidfile); | |
621 | goto error; | |
622 | } | |
623 | ||
624 | /* Is the file we have locked still named 'pidfile'? */ | |
625 | if (stat(pidfile, &s) || fstat(fileno(file), &s2) | |
626 | || s.st_ino != s2.st_ino || s.st_dev != s2.st_dev) { | |
627 | /* No. We lost a race with someone else who got the lock before | |
628 | * us, deleted the pidfile, and closed it (releasing the lock). */ | |
629 | error = EALREADY; | |
630 | VLOG_WARN("%s: lost race to delete pidfile", pidfile); | |
631 | goto error; | |
632 | } | |
633 | ||
634 | /* We won the right to delete the stale pidfile. */ | |
635 | if (unlink(pidfile)) { | |
636 | error = errno; | |
637 | VLOG_WARN("%s: failed to delete stale pidfile (%s)", | |
10a89ef0 | 638 | pidfile, ovs_strerror(error)); |
aacea8ba BP |
639 | goto error; |
640 | } | |
641 | VLOG_DBG("%s: deleted stale pidfile", pidfile); | |
642 | fclose(file); | |
643 | return 0; | |
064af421 BP |
644 | } |
645 | ||
646 | if (!fgets(line, sizeof line, file)) { | |
647 | if (ferror(file)) { | |
648 | error = errno; | |
10a89ef0 | 649 | VLOG_WARN("%s: read: %s", pidfile, ovs_strerror(error)); |
064af421 BP |
650 | } else { |
651 | error = ESRCH; | |
652 | VLOG_WARN("%s: read: unexpected end of file", pidfile); | |
653 | } | |
654 | goto error; | |
655 | } | |
656 | ||
657 | if (lck.l_pid != strtoul(line, NULL, 10)) { | |
aacea8ba BP |
658 | /* The process that has the pidfile locked is not the process that |
659 | * created it. It must be stale, with the process that has it locked | |
660 | * preparing to delete it. */ | |
064af421 | 661 | error = ESRCH; |
aacea8ba BP |
662 | VLOG_WARN("%s: stale pidfile for pid %s being deleted by pid %ld", |
663 | pidfile, line, (long int) lck.l_pid); | |
064af421 BP |
664 | goto error; |
665 | } | |
666 | ||
667 | fclose(file); | |
668 | return lck.l_pid; | |
669 | ||
670 | error: | |
671 | if (file) { | |
672 | fclose(file); | |
673 | } | |
674 | return -error; | |
675 | } | |
18e124a2 BP |
676 | |
677 | /* Opens and reads a PID from 'pidfile'. Returns the positive PID if | |
678 | * successful, otherwise a negative errno value. */ | |
679 | pid_t | |
680 | read_pidfile(const char *pidfile) | |
681 | { | |
aacea8ba | 682 | return read_pidfile__(pidfile, false); |
18e124a2 BP |
683 | } |
684 | ||
aacea8ba BP |
685 | /* Checks whether a process with the given 'pidfile' is already running and, |
686 | * if so, aborts. If 'pidfile' is stale, deletes it. */ | |
687 | static void | |
688 | check_already_running(void) | |
18e124a2 | 689 | { |
aacea8ba BP |
690 | long int pid = read_pidfile__(pidfile, true); |
691 | if (pid > 0) { | |
692 | VLOG_FATAL("%s: already running as pid %ld, aborting", pidfile, pid); | |
693 | } else if (pid < 0) { | |
694 | VLOG_FATAL("%s: pidfile check failed (%s), aborting", | |
10a89ef0 | 695 | pidfile, ovs_strerror(-pid)); |
aacea8ba | 696 | } |
18e124a2 | 697 | } |
fda546bd GS |
698 | |
699 | \f | |
700 | /* stub functions for non-windows platform. */ | |
701 | ||
702 | void | |
703 | service_start(int *argc OVS_UNUSED, char **argv[] OVS_UNUSED) | |
704 | { | |
705 | } | |
706 | ||
707 | void | |
708 | service_stop(void) | |
709 | { | |
710 | } | |
711 | ||
712 | bool | |
713 | should_service_stop(void) | |
714 | { | |
715 | return false; | |
716 | } | |
e91b927d AZ |
717 | |
718 | \f | |
719 | static bool | |
720 | gid_matches(gid_t expected, gid_t value) | |
721 | { | |
722 | return expected == -1 || expected == value; | |
723 | } | |
724 | ||
725 | static bool | |
6a54bae1 | 726 | gid_verify(gid_t gid) |
e91b927d | 727 | { |
6a54bae1 | 728 | gid_t r, e; |
e91b927d | 729 | |
6a54bae1 YT |
730 | r = getgid(); |
731 | e = getegid(); | |
732 | return (gid_matches(gid, r) && | |
733 | gid_matches(gid, e)); | |
e91b927d AZ |
734 | } |
735 | ||
736 | static void | |
6a54bae1 | 737 | daemon_switch_group(gid_t gid) |
e91b927d | 738 | { |
6a54bae1 YT |
739 | if ((setgid(gid) == -1) || !gid_verify(gid)) { |
740 | VLOG_FATAL("%s: fail to switch group to gid as %d, aborting", | |
e91b927d AZ |
741 | pidfile, gid); |
742 | } | |
743 | } | |
744 | ||
745 | static bool | |
746 | uid_matches(uid_t expected, uid_t value) | |
747 | { | |
748 | return expected == -1 || expected == value; | |
749 | } | |
750 | ||
751 | static bool | |
6a54bae1 | 752 | uid_verify(const uid_t uid) |
e91b927d | 753 | { |
6a54bae1 | 754 | uid_t r, e; |
e91b927d | 755 | |
6a54bae1 YT |
756 | r = getuid(); |
757 | e = geteuid(); | |
758 | return (uid_matches(uid, r) && | |
759 | uid_matches(uid, e)); | |
e91b927d AZ |
760 | } |
761 | ||
762 | static void | |
6a54bae1 | 763 | daemon_switch_user(const uid_t uid, const char *user) |
e91b927d | 764 | { |
6a54bae1 | 765 | if ((setuid(uid) == -1) || !uid_verify(uid)) { |
e91b927d AZ |
766 | VLOG_FATAL("%s: fail to switch user to %s, aborting", |
767 | pidfile, user); | |
768 | } | |
769 | } | |
770 | ||
771 | /* Use portable Unix APIs to switch uid:gid, when datapath | |
772 | * access is not required. On Linux systems, all capabilities | |
773 | * will be dropped. */ | |
774 | static void | |
775 | daemon_become_new_user_unix(void) | |
776 | { | |
777 | /* "Setuid Demystified" by Hao Chen, etc outlines some caveats of | |
778 | * around unix system call setuid() and friends. This implementation | |
779 | * mostly follow the advice given by the paper. The paper is | |
780 | * published in 2002, so things could have changed. */ | |
781 | ||
782 | /* Change both real and effective uid and gid will permanently | |
783 | * drop the process' privilege. "Setuid Demystified" suggested | |
784 | * that calling getuid() after each setuid() call to verify they | |
785 | * are actually set, because checking return code alone is not | |
786 | * sufficient. */ | |
6a54bae1 | 787 | daemon_switch_group(gid); |
e91b927d AZ |
788 | if (user && initgroups(user, gid) == -1) { |
789 | VLOG_FATAL("%s: fail to add supplementary group gid %d, " | |
790 | "aborting", pidfile, gid); | |
791 | } | |
6a54bae1 | 792 | daemon_switch_user(uid, user); |
e91b927d AZ |
793 | } |
794 | ||
795 | /* Linux specific implementation of daemon_become_new_user() | |
796 | * using libcap-ng. */ | |
e91b927d | 797 | static void |
6e6271d2 | 798 | daemon_become_new_user_linux(bool access_datapath OVS_UNUSED) |
e91b927d | 799 | { |
6e6271d2 | 800 | #if defined __linux__ && HAVE_LIBCAPNG |
e91b927d AZ |
801 | int ret; |
802 | ||
803 | ret = capng_get_caps_process(); | |
804 | ||
805 | if (!ret) { | |
806 | if (capng_have_capabilities(CAPNG_SELECT_CAPS) > CAPNG_NONE) { | |
807 | const capng_type_t cap_sets = CAPNG_EFFECTIVE|CAPNG_PERMITTED; | |
808 | ||
809 | capng_clear(CAPNG_SELECT_BOTH); | |
810 | ||
811 | ret = capng_update(CAPNG_ADD, cap_sets, CAP_IPC_LOCK) | |
812 | || capng_update(CAPNG_ADD, cap_sets, CAP_NET_BIND_SERVICE); | |
813 | ||
814 | if (access_datapath && !ret) { | |
815 | ret = capng_update(CAPNG_ADD, cap_sets, CAP_NET_ADMIN) | |
816 | || capng_update(CAPNG_ADD, cap_sets, CAP_NET_RAW); | |
817 | } | |
818 | } else { | |
819 | ret = -1; | |
820 | } | |
821 | } | |
822 | ||
823 | if (!ret) { | |
824 | /* CAPNG_INIT_SUPP_GRP will be a better choice than | |
825 | * CAPNG_DROP_SUPP_GRP. However this enum value is only defined | |
826 | * with libcap-ng higher than version 0.7.4, which is not wildly | |
827 | * available on many Linux distributions yet. Taking a more | |
828 | * conservative approach to make sure OVS behaves consistently. | |
829 | * | |
830 | * XXX We may change this for future OVS releases. | |
831 | */ | |
832 | ret = capng_change_id(uid, gid, CAPNG_DROP_SUPP_GRP | |
833 | | CAPNG_CLEAR_BOUNDING); | |
834 | } | |
835 | ||
836 | if (ret) { | |
837 | VLOG_FATAL("%s: libcap-ng fail to switch to user and group " | |
838 | "%d:%d, aborting", pidfile, uid, gid); | |
839 | } | |
e91b927d | 840 | #endif |
6e6271d2 | 841 | } |
e91b927d AZ |
842 | |
843 | static void | |
844 | daemon_become_new_user__(bool access_datapath) | |
845 | { | |
3de44dd1 AZ |
846 | /* If vlog file has been created, change its owner to the non-root user |
847 | * as specifed by the --user option. */ | |
de929213 | 848 | vlog_change_owner_unix(uid, gid); |
3de44dd1 | 849 | |
e91b927d AZ |
850 | if (LINUX) { |
851 | if (LIBCAPNG) { | |
852 | daemon_become_new_user_linux(access_datapath); | |
853 | } else { | |
854 | VLOG_FATAL("%s: fail to downgrade user using libcap-ng. " | |
855 | "(libcap-ng is not configured at compile time), " | |
856 | "aborting.", pidfile); | |
857 | } | |
858 | } else { | |
859 | daemon_become_new_user_unix(); | |
860 | } | |
861 | } | |
862 | ||
863 | /* Noramlly, user switch is embedded within daemonize_start(). | |
864 | * However, there in case the user switch needs to be done | |
865 | * before daemonize_start(), the following API can be used. */ | |
866 | void | |
867 | daemon_become_new_user(bool access_datapath) | |
868 | { | |
869 | assert_single_threaded(); | |
870 | if (switch_user) { | |
871 | daemon_become_new_user__(access_datapath); | |
6069edb0 | 872 | /* daemonize_start() should not switch user again. */ |
e91b927d AZ |
873 | switch_user = false; |
874 | } | |
875 | } | |
876 | ||
877 | /* Return the maximun suggested buffer size for both getpwname_r() | |
878 | * and getgrnam_r(). | |
879 | * | |
880 | * This size may still not be big enough. in case getpwname_r() | |
881 | * and friends return ERANGE, a larger buffer should be supplied to | |
882 | * retry. (The man page did not specify the max size to stop at, we | |
883 | * will keep trying with doubling the buffer size for each round until | |
884 | * the size wrapps around size_t. */ | |
885 | static size_t | |
886 | get_sysconf_buffer_size(void) | |
887 | { | |
888 | size_t bufsize, pwd_bs = 0, grp_bs = 0; | |
889 | const size_t default_bufsize = 1024; | |
890 | ||
891 | errno = 0; | |
892 | if ((pwd_bs = sysconf(_SC_GETPW_R_SIZE_MAX)) == -1) { | |
893 | if (errno) { | |
894 | VLOG_FATAL("%s: Read initial passwordd struct size " | |
895 | "failed (%s), aborting. ", pidfile, | |
896 | ovs_strerror(errno)); | |
897 | } | |
898 | } | |
899 | ||
900 | if ((grp_bs = sysconf(_SC_GETGR_R_SIZE_MAX)) == -1) { | |
901 | if (errno) { | |
902 | VLOG_FATAL("%s: Read initial group struct size " | |
903 | "failed (%s), aborting. ", pidfile, | |
904 | ovs_strerror(errno)); | |
905 | } | |
906 | } | |
907 | ||
908 | bufsize = MAX(pwd_bs, grp_bs); | |
909 | return bufsize ? bufsize : default_bufsize; | |
910 | } | |
911 | ||
912 | /* Try to double the size of '*buf', return true | |
913 | * if successful, and '*sizep' will be updated with | |
914 | * the new size. Otherwise, return false. */ | |
915 | static bool | |
916 | enlarge_buffer(char **buf, size_t *sizep) | |
917 | { | |
918 | size_t newsize = *sizep * 2; | |
919 | ||
920 | if (newsize > *sizep) { | |
921 | *buf = xrealloc(*buf, newsize); | |
922 | *sizep = newsize; | |
923 | return true; | |
924 | } | |
925 | ||
926 | return false; | |
927 | } | |
928 | ||
929 | /* Parse and sanity check user_spec. | |
930 | * | |
931 | * If successful, set global variables 'uid' and 'gid' | |
932 | * with the parsed results. Global variable 'user' | |
933 | * will be pointing to a string that stores the name | |
934 | * of the user to be switched into. | |
935 | * | |
936 | * Also set 'switch_to_new_user' to true, The actual | |
937 | * user switching is done as soon as daemonize_start() | |
938 | * is called. I/O access before calling daemonize_start() | |
939 | * will still be with root's credential. */ | |
940 | void | |
941 | daemon_set_new_user(const char *user_spec) | |
942 | { | |
943 | char *pos = strchr(user_spec, ':'); | |
944 | size_t init_bufsize, bufsize; | |
945 | ||
946 | init_bufsize = get_sysconf_buffer_size(); | |
947 | uid = getuid(); | |
948 | gid = getgid(); | |
949 | ||
950 | if (geteuid() || uid) { | |
951 | VLOG_FATAL("%s: only root can use --user option", pidfile); | |
952 | } | |
953 | ||
954 | user_spec += strspn(user_spec, " \t\r\n"); | |
955 | size_t len = pos ? pos - user_spec : strlen(user_spec); | |
956 | char *buf; | |
957 | struct passwd pwd, *res; | |
958 | int e; | |
959 | ||
960 | bufsize = init_bufsize; | |
961 | buf = xmalloc(bufsize); | |
962 | if (len) { | |
963 | user = xmemdup0(user_spec, len); | |
964 | ||
965 | while ((e = getpwnam_r(user, &pwd, buf, bufsize, &res)) == ERANGE) { | |
966 | if (!enlarge_buffer(&buf, &bufsize)) { | |
967 | break; | |
968 | } | |
969 | } | |
970 | ||
971 | if (e != 0) { | |
972 | VLOG_FATAL("%s: Failed to retrive user %s's uid (%s), aborting.", | |
973 | pidfile, user, ovs_strerror(e)); | |
974 | } | |
eaf2aa9e CE |
975 | if (res == NULL) { |
976 | VLOG_FATAL("%s: user %s not found, aborting.", pidfile, user); | |
977 | } | |
e91b927d AZ |
978 | } else { |
979 | /* User name is not specified, use current user. */ | |
980 | while ((e = getpwuid_r(uid, &pwd, buf, bufsize, &res)) == ERANGE) { | |
981 | if (!enlarge_buffer(&buf, &bufsize)) { | |
982 | break; | |
983 | } | |
984 | } | |
985 | ||
986 | if (e != 0) { | |
987 | VLOG_FATAL("%s: Failed to retrive current user's name " | |
988 | "(%s), aborting.", pidfile, ovs_strerror(e)); | |
989 | } | |
990 | user = xstrdup(pwd.pw_name); | |
991 | } | |
992 | ||
993 | uid = pwd.pw_uid; | |
994 | gid = pwd.pw_gid; | |
995 | free(buf); | |
996 | ||
997 | if (pos) { | |
998 | char *grpstr = pos + 1; | |
999 | grpstr += strspn(grpstr, " \t\r\n"); | |
1000 | ||
1001 | if (*grpstr) { | |
1002 | struct group grp, *res; | |
1003 | ||
1004 | bufsize = init_bufsize; | |
1005 | buf = xmalloc(bufsize); | |
1006 | while ((e = getgrnam_r(grpstr, &grp, buf, bufsize, &res)) | |
1007 | == ERANGE) { | |
1008 | if (!enlarge_buffer(&buf, &bufsize)) { | |
1009 | break; | |
1010 | } | |
1011 | } | |
1012 | ||
1013 | if (e) { | |
1014 | VLOG_FATAL("%s: Failed to get group entry for %s, " | |
1015 | "(%s), aborting.", pidfile, grpstr, | |
1016 | ovs_strerror(e)); | |
1017 | } | |
eaf2aa9e CE |
1018 | if (res == NULL) { |
1019 | VLOG_FATAL("%s: group %s not found, aborting.", pidfile, | |
1020 | grpstr); | |
1021 | } | |
e91b927d AZ |
1022 | |
1023 | if (gid != grp.gr_gid) { | |
1024 | char **mem; | |
1025 | ||
1026 | for (mem = grp.gr_mem; *mem; ++mem) { | |
1027 | if (!strcmp(*mem, user)) { | |
1028 | break; | |
1029 | } | |
1030 | } | |
1031 | ||
1032 | if (!*mem) { | |
1033 | VLOG_FATAL("%s: Invalid --user option %s (user %s is " | |
1034 | "not in group %s), aborting.", pidfile, | |
1035 | user_spec, user, grpstr); | |
1036 | } | |
1037 | gid = grp.gr_gid; | |
1038 | } | |
1039 | free(buf); | |
1040 | } | |
1041 | } | |
1042 | ||
6069edb0 | 1043 | switch_user = true; |
e91b927d | 1044 | } |