]> git.proxmox.com Git - mirror_ovs.git/blob - lib/process.c
netdev-offload-tc: Use single 'once' variable for probing tc features
[mirror_ovs.git] / lib / process.c
1 /*
2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
3 *
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:
7 *
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.
15 */
16
17 #include <config.h>
18 #include "process.h"
19 #include <errno.h>
20 #include <fcntl.h>
21 #include <signal.h>
22 #include <stdlib.h>
23 #include <string.h>
24 #include <sys/resource.h>
25 #include <sys/stat.h>
26 #include <sys/wait.h>
27 #include <unistd.h>
28 #include "coverage.h"
29 #include "openvswitch/dynamic-string.h"
30 #include "fatal-signal.h"
31 #include "openvswitch/list.h"
32 #include "ovs-thread.h"
33 #include "openvswitch/poll-loop.h"
34 #include "signals.h"
35 #include "socket-util.h"
36 #include "timeval.h"
37 #include "util.h"
38 #include "openvswitch/vlog.h"
39
40 VLOG_DEFINE_THIS_MODULE(process);
41
42 COVERAGE_DEFINE(process_start);
43
44 #ifdef __linux__
45 #define LINUX 1
46 #include <asm/param.h>
47 #else
48 #define LINUX 0
49 #endif
50
51 struct process {
52 struct ovs_list node;
53 char *name;
54 pid_t pid;
55
56 /* State. */
57 bool exited;
58 int status;
59 };
60
61 struct raw_process_info {
62 unsigned long int vsz; /* Virtual size, in kB. */
63 unsigned long int rss; /* Resident set size, in kB. */
64 long long int uptime; /* ms since started. */
65 long long int cputime; /* ms of CPU used during 'uptime'. */
66 pid_t ppid; /* Parent. */
67 int core_id; /* Core id last executed on. */
68 char name[18]; /* Name. */
69 };
70
71 /* Pipe used to signal child termination. */
72 static int fds[2];
73
74 /* All processes. */
75 static struct ovs_list all_processes = OVS_LIST_INITIALIZER(&all_processes);
76
77 static void sigchld_handler(int signr OVS_UNUSED);
78
79 /* Initializes the process subsystem (if it is not already initialized). Calls
80 * exit() if initialization fails.
81 *
82 * This function may not be called after creating any additional threads.
83 *
84 * Calling this function is optional; it will be called automatically by
85 * process_start() if necessary. Calling it explicitly allows the client to
86 * prevent the process from exiting at an unexpected time. */
87 void
88 process_init(void)
89 {
90 #ifndef _WIN32
91 static bool inited;
92 struct sigaction sa;
93
94 assert_single_threaded();
95 if (inited) {
96 return;
97 }
98 inited = true;
99
100 /* Create notification pipe. */
101 xpipe_nonblocking(fds);
102
103 /* Set up child termination signal handler. */
104 memset(&sa, 0, sizeof sa);
105 sa.sa_handler = sigchld_handler;
106 sigemptyset(&sa.sa_mask);
107 sa.sa_flags = SA_NOCLDSTOP | SA_RESTART;
108 xsigaction(SIGCHLD, &sa, NULL);
109 #endif
110 }
111
112 char *
113 process_escape_args(char **argv)
114 {
115 struct ds ds = DS_EMPTY_INITIALIZER;
116 char **argp;
117 for (argp = argv; *argp; argp++) {
118 const char *arg = *argp;
119 const char *p;
120 if (argp != argv) {
121 ds_put_char(&ds, ' ');
122 }
123 if (arg[strcspn(arg, " \t\r\n\v\\\'\"")]) {
124 ds_put_char(&ds, '"');
125 for (p = arg; *p; p++) {
126 if (*p == '\\' || *p == '\"') {
127 ds_put_char(&ds, '\\');
128 }
129 ds_put_char(&ds, *p);
130 }
131 ds_put_char(&ds, '"');
132 } else {
133 ds_put_cstr(&ds, arg);
134 }
135 }
136 return ds_cstr(&ds);
137 }
138
139 /* Prepare to start a process whose command-line arguments are given by the
140 * null-terminated 'argv' array. Returns 0 if successful, otherwise a
141 * positive errno value. */
142 static int
143 process_prestart(char **argv)
144 {
145 char *binary;
146
147 process_init();
148
149 /* Log the process to be started. */
150 if (VLOG_IS_DBG_ENABLED()) {
151 char *args = process_escape_args(argv);
152 VLOG_DBG("starting subprocess: %s", args);
153 free(args);
154 }
155
156 /* execvp() will search PATH too, but the error in that case is more
157 * obscure, since it is only reported post-fork. */
158 binary = process_search_path(argv[0]);
159 if (!binary) {
160 VLOG_ERR("%s not found in PATH", argv[0]);
161 return ENOENT;
162 }
163 free(binary);
164
165 return 0;
166 }
167
168 /* Creates and returns a new struct process with the specified 'name' and
169 * 'pid'. */
170 static struct process *
171 process_register(const char *name, pid_t pid)
172 {
173 struct process *p;
174 const char *slash;
175
176 p = xzalloc(sizeof *p);
177 p->pid = pid;
178 slash = strrchr(name, '/');
179 p->name = xstrdup(slash ? slash + 1 : name);
180 p->exited = false;
181
182 ovs_list_push_back(&all_processes, &p->node);
183
184 return p;
185 }
186
187 #ifndef _WIN32
188 static bool
189 rlim_is_finite(rlim_t limit)
190 {
191 if (limit == RLIM_INFINITY) {
192 return false;
193 }
194
195 #ifdef RLIM_SAVED_CUR /* FreeBSD 8.0 lacks RLIM_SAVED_CUR. */
196 if (limit == RLIM_SAVED_CUR) {
197 return false;
198 }
199 #endif
200
201 #ifdef RLIM_SAVED_MAX /* FreeBSD 8.0 lacks RLIM_SAVED_MAX. */
202 if (limit == RLIM_SAVED_MAX) {
203 return false;
204 }
205 #endif
206
207 return true;
208 }
209
210 /* Returns the maximum valid FD value, plus 1. */
211 static int
212 get_max_fds(void)
213 {
214 static int max_fds;
215
216 if (!max_fds) {
217 struct rlimit r;
218 if (!getrlimit(RLIMIT_NOFILE, &r) && rlim_is_finite(r.rlim_cur)) {
219 max_fds = r.rlim_cur;
220 } else {
221 VLOG_WARN("failed to obtain fd limit, defaulting to 1024");
222 max_fds = 1024;
223 }
224 }
225
226 return max_fds;
227 }
228 #endif /* _WIN32 */
229
230 /* Starts a subprocess with the arguments in the null-terminated argv[] array.
231 * argv[0] is used as the name of the process. Searches the PATH environment
232 * variable to find the program to execute.
233 *
234 * This function may not be called after creating any additional threads.
235 *
236 * All file descriptors are closed before executing the subprocess, except for
237 * fds 0, 1, and 2.
238 *
239 * Returns 0 if successful, otherwise a positive errno value indicating the
240 * error. If successful, '*pp' is assigned a new struct process that may be
241 * used to query the process's status. On failure, '*pp' is set to NULL. */
242 int
243 process_start(char **argv, struct process **pp)
244 {
245 #ifndef _WIN32
246 pid_t pid;
247 int error;
248 sigset_t prev_mask;
249
250 assert_single_threaded();
251
252 *pp = NULL;
253 COVERAGE_INC(process_start);
254 error = process_prestart(argv);
255 if (error) {
256 return error;
257 }
258
259 fatal_signal_block(&prev_mask);
260 pid = fork();
261 if (pid < 0) {
262 VLOG_WARN("fork failed: %s", ovs_strerror(errno));
263 error = errno;
264 } else if (pid) {
265 /* Running in parent process. */
266 *pp = process_register(argv[0], pid);
267 error = 0;
268 } else {
269 /* Running in child process. */
270 int fd_max = get_max_fds();
271 int fd;
272
273 fatal_signal_fork();
274 for (fd = 3; fd < fd_max; fd++) {
275 close(fd);
276 }
277 xpthread_sigmask(SIG_SETMASK, &prev_mask, NULL);
278 execvp(argv[0], argv);
279 fprintf(stderr, "execvp(\"%s\") failed: %s\n",
280 argv[0], ovs_strerror(errno));
281 _exit(1);
282 }
283 xpthread_sigmask(SIG_SETMASK, &prev_mask, NULL);
284 return error;
285 #else
286 *pp = NULL;
287 return ENOSYS;
288 #endif
289 }
290
291 /* Destroys process 'p'. */
292 void
293 process_destroy(struct process *p)
294 {
295 if (p) {
296 ovs_list_remove(&p->node);
297 free(p->name);
298 free(p);
299 }
300 }
301
302 /* Sends signal 'signr' to process 'p'. Returns 0 if successful, otherwise a
303 * positive errno value. */
304 int
305 process_kill(const struct process *p, int signr)
306 {
307 #ifndef _WIN32
308 return (p->exited ? ESRCH
309 : !kill(p->pid, signr) ? 0
310 : errno);
311 #else
312 return ENOSYS;
313 #endif
314 }
315
316 /* Returns the pid of process 'p'. */
317 pid_t
318 process_pid(const struct process *p)
319 {
320 return p->pid;
321 }
322
323 /* Returns the name of process 'p' (the name passed to process_start() with any
324 * leading directories stripped). */
325 const char *
326 process_name(const struct process *p)
327 {
328 return p->name;
329 }
330
331 /* Returns true if process 'p' has exited, false otherwise. */
332 bool
333 process_exited(struct process *p)
334 {
335 return p->exited;
336 }
337
338 /* Returns process 'p''s exit status, as reported by waitpid(2).
339 * process_status(p) may be called only after process_exited(p) has returned
340 * true. */
341 int
342 process_status(const struct process *p)
343 {
344 ovs_assert(p->exited);
345 return p->status;
346 }
347
348 int
349 count_crashes(pid_t pid)
350 {
351 char file_name[128];
352 const char *paren;
353 char line[128];
354 int crashes = 0;
355 FILE *stream;
356
357 ovs_assert(LINUX);
358
359 sprintf(file_name, "/proc/%lu/cmdline", (unsigned long int) pid);
360 stream = fopen(file_name, "r");
361 if (!stream) {
362 VLOG_WARN_ONCE("%s: open failed (%s)", file_name, ovs_strerror(errno));
363 goto exit;
364 }
365
366 if (!fgets(line, sizeof line, stream)) {
367 VLOG_WARN_ONCE("%s: read failed (%s)", file_name,
368 feof(stream) ? "end of file" : ovs_strerror(errno));
369 goto exit_close;
370 }
371
372 paren = strchr(line, '(');
373 if (paren) {
374 int x;
375 if (ovs_scan(paren + 1, "%d", &x)) {
376 crashes = x;
377 }
378 }
379
380 exit_close:
381 fclose(stream);
382 exit:
383 return crashes;
384 }
385
386 static unsigned long long int
387 ticks_to_ms(unsigned long long int ticks)
388 {
389 ovs_assert(LINUX);
390
391 #ifndef USER_HZ
392 #define USER_HZ 100
393 #endif
394
395 #if USER_HZ == 100 /* Common case. */
396 return ticks * (1000 / USER_HZ);
397 #else /* Alpha and some other architectures. */
398 double factor = 1000.0 / USER_HZ;
399 return ticks * factor + 0.5;
400 #endif
401 }
402
403 static bool
404 get_raw_process_info(pid_t pid, struct raw_process_info *raw)
405 {
406 unsigned long long int vsize, rss, start_time, utime, stime;
407 long long int start_msec;
408 unsigned long ppid;
409 char file_name[128];
410 FILE *stream;
411 int n;
412
413 ovs_assert(LINUX);
414
415 sprintf(file_name, "/proc/%lu/stat", (unsigned long int) pid);
416 stream = fopen(file_name, "r");
417 if (!stream) {
418 VLOG_ERR_ONCE("%s: open failed (%s)",
419 file_name, ovs_strerror(errno));
420 return false;
421 }
422
423 n = fscanf(stream,
424 "%*d " /* (1. pid) */
425 "(%17[^)]) " /* 2. process name */
426 "%*c " /* (3. state) */
427 "%lu " /* 4. ppid */
428 "%*d " /* (5. pgid) */
429 "%*d " /* (6. sid) */
430 "%*d " /* (7. tty_nr) */
431 "%*d " /* (8. tty_pgrp) */
432 "%*u " /* (9. flags) */
433 "%*u " /* (10. min_flt) */
434 "%*u " /* (11. cmin_flt) */
435 "%*u " /* (12. maj_flt) */
436 "%*u " /* (13. cmaj_flt) */
437 "%llu " /* 14. utime */
438 "%llu " /* 15. stime */
439 "%*d " /* (16. cutime) */
440 "%*d " /* (17. cstime) */
441 "%*d " /* (18. priority) */
442 "%*d " /* (19. nice) */
443 "%*d " /* (20. num_threads) */
444 "%*d " /* (21. always 0) */
445 "%llu " /* 22. start_time */
446 "%llu " /* 23. vsize */
447 "%llu " /* 24. rss */
448 "%*u " /* (25. rsslim) */
449 "%*u " /* (26. start_code) */
450 "%*u " /* (27. end_code) */
451 "%*u " /* (28. start_stack) */
452 "%*u " /* (29. esp) */
453 "%*u " /* (30. eip) */
454 "%*u " /* (31. pending signals) */
455 "%*u " /* (32. blocked signals) */
456 "%*u " /* (33. ignored signals) */
457 "%*u " /* (34. caught signals) */
458 "%*u " /* (35. whcan) */
459 "%*u " /* (36. always 0) */
460 "%*u " /* (37. always 0) */
461 "%*d " /* (38. exit_signal) */
462 "%d " /* 39. task_cpu */
463 #if 0
464 /* These are here for documentation but #if'd out to save
465 * actually parsing them from the stream for no benefit. */
466 "%*u " /* (40. rt_priority) */
467 "%*u " /* (41. policy) */
468 "%*llu " /* (42. blkio_ticks) */
469 "%*lu " /* (43. gtime) */
470 "%*ld" /* (44. cgtime) */
471 #endif
472 , raw->name, &ppid, &utime, &stime, &start_time,
473 &vsize, &rss, &raw->core_id);
474 fclose(stream);
475 if (n != 8) {
476 VLOG_ERR_ONCE("%s: fscanf failed", file_name);
477 return false;
478 }
479
480 start_msec = get_boot_time() + ticks_to_ms(start_time);
481
482 raw->vsz = vsize / 1024;
483 raw->rss = rss * (get_page_size() / 1024);
484 raw->uptime = time_wall_msec() - start_msec;
485 raw->cputime = ticks_to_ms(utime + stime);
486 raw->ppid = ppid;
487
488 return true;
489 }
490
491 bool
492 get_process_info(pid_t pid, struct process_info *pinfo)
493 {
494 struct raw_process_info child;
495
496 ovs_assert(LINUX);
497 if (!get_raw_process_info(pid, &child)) {
498 return false;
499 }
500
501 ovs_strlcpy(pinfo->name, child.name, sizeof pinfo->name);
502 pinfo->vsz = child.vsz;
503 pinfo->rss = child.rss;
504 pinfo->booted = child.uptime;
505 pinfo->crashes = 0;
506 pinfo->uptime = child.uptime;
507 pinfo->cputime = child.cputime;
508 pinfo->core_id = child.core_id;
509
510 if (child.ppid) {
511 struct raw_process_info parent;
512
513 get_raw_process_info(child.ppid, &parent);
514 if (!strcmp(child.name, parent.name)) {
515 pinfo->booted = parent.uptime;
516 pinfo->crashes = count_crashes(child.ppid);
517 }
518 }
519
520 return true;
521 }
522
523 /* Given 'status', which is a process status in the form reported by waitpid(2)
524 * and returned by process_status(), returns a string describing how the
525 * process terminated. The caller is responsible for freeing the string when
526 * it is no longer needed. */
527 char *
528 process_status_msg(int status)
529 {
530 struct ds ds = DS_EMPTY_INITIALIZER;
531 #ifndef _WIN32
532 if (WIFEXITED(status)) {
533 ds_put_format(&ds, "exit status %d", WEXITSTATUS(status));
534 } else if (WIFSIGNALED(status)) {
535 char namebuf[SIGNAL_NAME_BUFSIZE];
536
537 ds_put_format(&ds, "killed (%s)",
538 signal_name(WTERMSIG(status), namebuf, sizeof namebuf));
539 } else if (WIFSTOPPED(status)) {
540 char namebuf[SIGNAL_NAME_BUFSIZE];
541
542 ds_put_format(&ds, "stopped (%s)",
543 signal_name(WSTOPSIG(status), namebuf, sizeof namebuf));
544 } else {
545 ds_put_format(&ds, "terminated abnormally (%x)", status);
546 }
547 if (WCOREDUMP(status)) {
548 ds_put_cstr(&ds, ", core dumped");
549 }
550 #else
551 ds_put_cstr(&ds, "function not supported.");
552 #endif
553 return ds_cstr(&ds);
554 }
555
556 /* Executes periodic maintenance activities required by the process module. */
557 void
558 process_run(void)
559 {
560 #ifndef _WIN32
561 char buf[_POSIX_PIPE_BUF];
562
563 if (!ovs_list_is_empty(&all_processes) && read(fds[0], buf, sizeof buf) > 0) {
564 struct process *p;
565
566 LIST_FOR_EACH (p, node, &all_processes) {
567 if (!p->exited) {
568 int retval, status;
569 do {
570 retval = waitpid(p->pid, &status, WNOHANG);
571 } while (retval == -1 && errno == EINTR);
572 if (retval == p->pid) {
573 p->exited = true;
574 p->status = status;
575 } else if (retval < 0) {
576 VLOG_WARN("waitpid: %s", ovs_strerror(errno));
577 p->exited = true;
578 p->status = -1;
579 }
580 }
581 }
582 }
583 #endif
584 }
585
586 /* Causes the next call to poll_block() to wake up when process 'p' has
587 * exited. */
588 void
589 process_wait(struct process *p)
590 {
591 #ifndef _WIN32
592 if (p->exited) {
593 poll_immediate_wake();
594 } else {
595 poll_fd_wait(fds[0], POLLIN);
596 }
597 #else
598 OVS_NOT_REACHED();
599 #endif
600 }
601
602 char *
603 process_search_path(const char *name)
604 {
605 char *save_ptr = NULL;
606 char *path, *dir;
607 struct stat s;
608
609 if (strchr(name, '/') || !getenv("PATH")) {
610 return stat(name, &s) == 0 ? xstrdup(name) : NULL;
611 }
612
613 path = xstrdup(getenv("PATH"));
614 for (dir = strtok_r(path, ":", &save_ptr); dir;
615 dir = strtok_r(NULL, ":", &save_ptr)) {
616 char *file = xasprintf("%s/%s", dir, name);
617 if (stat(file, &s) == 0) {
618 free(path);
619 return file;
620 }
621 free(file);
622 }
623 free(path);
624 return NULL;
625 }
626 \f
627 static void
628 sigchld_handler(int signr OVS_UNUSED)
629 {
630 ignore(write(fds[1], "", 1));
631 }