1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * This file is part of the ZFS Event Daemon (ZED).
4 *
5 * Developed at Lawrence Livermore National Laboratory (LLNL-CODE-403049).
6 * Copyright (C) 2013-2014 Lawrence Livermore National Security, LLC.
7 * Refer to the OpenZFS git commit log for authoritative copyright attribution.
8 *
9 * The contents of this file are subject to the terms of the
10 * Common Development and Distribution License Version 1.0 (CDDL-1.0).
11 * You can obtain a copy of the license from the top-level file
12 * "OPENSOLARIS.LICENSE" or at <http://opensource.org/licenses/CDDL-1.0>.
13 * You may not use this file except in compliance with the license.
14 */
15
16 #include <assert.h>
17 #include <ctype.h>
18 #include <errno.h>
19 #include <fcntl.h>
20 #include <stdlib.h>
21 #include <string.h>
22 #include <stddef.h>
23 #include <sys/avl.h>
24 #include <sys/resource.h>
25 #include <sys/stat.h>
26 #include <sys/wait.h>
27 #include <time.h>
28 #include <unistd.h>
29 #include <pthread.h>
30 #include <signal.h>
31
32 #include "zed_exec.h"
33 #include "zed_log.h"
34 #include "zed_strings.h"
35
36 #define ZEVENT_FILENO 3
37
38 struct launched_process_node {
39 avl_node_t node;
40 pid_t pid;
41 uint64_t eid;
42 char *name;
43 };
44
45 static int
_launched_process_node_compare(const void * x1,const void * x2)46 _launched_process_node_compare(const void *x1, const void *x2)
47 {
48 pid_t p1;
49 pid_t p2;
50
51 assert(x1 != NULL);
52 assert(x2 != NULL);
53
54 p1 = ((const struct launched_process_node *) x1)->pid;
55 p2 = ((const struct launched_process_node *) x2)->pid;
56
57 if (p1 < p2)
58 return (-1);
59 else if (p1 == p2)
60 return (0);
61 else
62 return (1);
63 }
64
65 static pthread_t _reap_children_tid = (pthread_t)-1;
66 static volatile boolean_t _reap_children_stop;
67 static avl_tree_t _launched_processes;
68 static pthread_mutex_t _launched_processes_lock = PTHREAD_MUTEX_INITIALIZER;
69 static int16_t _launched_processes_limit;
70
71 /*
72 * Create an environment string array for passing to execve() using the
73 * NAME=VALUE strings in container [zsp].
74 * Return a newly-allocated environment, or NULL on error.
75 */
76 static char **
_zed_exec_create_env(zed_strings_t * zsp)77 _zed_exec_create_env(zed_strings_t *zsp)
78 {
79 int num_ptrs;
80 int buflen;
81 char *buf;
82 char **pp;
83 char *p;
84 const char *q;
85 int i;
86 int len;
87
88 num_ptrs = zed_strings_count(zsp) + 1;
89 buflen = num_ptrs * sizeof (char *);
90 for (q = zed_strings_first(zsp); q; q = zed_strings_next(zsp))
91 buflen += strlen(q) + 1;
92
93 buf = calloc(1, buflen);
94 if (!buf)
95 return (NULL);
96
97 pp = (char **)buf;
98 p = buf + (num_ptrs * sizeof (char *));
99 i = 0;
100 for (q = zed_strings_first(zsp); q; q = zed_strings_next(zsp)) {
101 pp[i] = p;
102 len = strlen(q) + 1;
103 memcpy(p, q, len);
104 p += len;
105 i++;
106 }
107 pp[i] = NULL;
108 assert(buf + buflen == p);
109 return ((char **)buf);
110 }
111
112 /*
113 * Fork a child process to handle event [eid]. The program [prog]
114 * in directory [dir] is executed with the environment [env].
115 *
116 * The file descriptor [zfd] is the zevent_fd used to track the
117 * current cursor location within the zevent nvlist.
118 */
119 static void
_zed_exec_fork_child(uint64_t eid,const char * dir,const char * prog,char * env[],int zfd,boolean_t in_foreground)120 _zed_exec_fork_child(uint64_t eid, const char *dir, const char *prog,
121 char *env[], int zfd, boolean_t in_foreground)
122 {
123 char path[PATH_MAX];
124 int n;
125 pid_t pid;
126 int fd;
127 struct launched_process_node *node;
128 sigset_t mask;
129 struct timespec launch_timeout =
130 { .tv_sec = 0, .tv_nsec = 200 * 1000 * 1000, };
131
132 assert(dir != NULL);
133 assert(prog != NULL);
134 assert(env != NULL);
135 assert(zfd >= 0);
136
137 while (__atomic_load_n(&_launched_processes_limit,
138 __ATOMIC_SEQ_CST) <= 0)
139 (void) nanosleep(&launch_timeout, NULL);
140
141 n = snprintf(path, sizeof (path), "%s/%s", dir, prog);
142 if ((n < 0) || (n >= sizeof (path))) {
143 zed_log_msg(LOG_WARNING,
144 "Failed to fork \"%s\" for eid=%llu: %s",
145 prog, eid, strerror(ENAMETOOLONG));
146 return;
147 }
148 (void) pthread_mutex_lock(&_launched_processes_lock);
149 pid = fork();
150 if (pid < 0) {
151 (void) pthread_mutex_unlock(&_launched_processes_lock);
152 zed_log_msg(LOG_WARNING,
153 "Failed to fork \"%s\" for eid=%llu: %s",
154 prog, eid, strerror(errno));
155 return;
156 } else if (pid == 0) {
157 (void) sigemptyset(&mask);
158 (void) sigprocmask(SIG_SETMASK, &mask, NULL);
159
160 (void) umask(022);
161 if (in_foreground && /* we're already devnulled if daemonised */
162 (fd = open("/dev/null", O_RDWR | O_CLOEXEC)) != -1) {
163 (void) dup2(fd, STDIN_FILENO);
164 (void) dup2(fd, STDOUT_FILENO);
165 (void) dup2(fd, STDERR_FILENO);
166 }
167 (void) dup2(zfd, ZEVENT_FILENO);
168 execle(path, prog, NULL, env);
169 _exit(127);
170 }
171
172 /* parent process */
173
174 node = calloc(1, sizeof (*node));
175 if (node) {
176 node->pid = pid;
177 node->eid = eid;
178 node->name = strdup(prog);
179 if (node->name == NULL) {
180 perror("strdup");
181 exit(EXIT_FAILURE);
182 }
183
184 avl_add(&_launched_processes, node);
185 }
186 (void) pthread_mutex_unlock(&_launched_processes_lock);
187
188 __atomic_sub_fetch(&_launched_processes_limit, 1, __ATOMIC_SEQ_CST);
189 zed_log_msg(LOG_INFO, "Invoking \"%s\" eid=%llu pid=%d",
190 prog, eid, pid);
191 }
192
193 static void
_nop(int sig)194 _nop(int sig)
195 {
196 (void) sig;
197 }
198
199 static void *
_reap_children(void * arg)200 _reap_children(void *arg)
201 {
202 (void) arg;
203 struct launched_process_node node, *pnode;
204 pid_t pid;
205 int status;
206 struct rusage usage;
207 struct sigaction sa = {};
208
209 (void) sigfillset(&sa.sa_mask);
210 (void) sigdelset(&sa.sa_mask, SIGCHLD);
211 (void) pthread_sigmask(SIG_SETMASK, &sa.sa_mask, NULL);
212
213 (void) sigemptyset(&sa.sa_mask);
214 sa.sa_handler = _nop;
215 sa.sa_flags = SA_NOCLDSTOP;
216 (void) sigaction(SIGCHLD, &sa, NULL);
217
218 for (_reap_children_stop = B_FALSE; !_reap_children_stop; ) {
219 (void) pthread_mutex_lock(&_launched_processes_lock);
220 pid = wait4(0, &status, WNOHANG, &usage);
221
222 if (pid == 0 || pid == (pid_t)-1) {
223 (void) pthread_mutex_unlock(&_launched_processes_lock);
224 if (pid == 0 || errno == ECHILD)
225 pause();
226 else if (errno != EINTR)
227 zed_log_msg(LOG_WARNING,
228 "Failed to wait for children: %s",
229 strerror(errno));
230 } else {
231 memset(&node, 0, sizeof (node));
232 node.pid = pid;
233 pnode = avl_find(&_launched_processes, &node, NULL);
234 if (pnode) {
235 memcpy(&node, pnode, sizeof (node));
236
237 avl_remove(&_launched_processes, pnode);
238 free(pnode);
239 }
240 (void) pthread_mutex_unlock(&_launched_processes_lock);
241 __atomic_add_fetch(&_launched_processes_limit, 1,
242 __ATOMIC_SEQ_CST);
243
244 usage.ru_utime.tv_sec += usage.ru_stime.tv_sec;
245 usage.ru_utime.tv_usec += usage.ru_stime.tv_usec;
246 usage.ru_utime.tv_sec +=
247 usage.ru_utime.tv_usec / (1000 * 1000);
248 usage.ru_utime.tv_usec %= 1000 * 1000;
249
250 if (WIFEXITED(status)) {
251 zed_log_msg(LOG_INFO,
252 "Finished \"%s\" eid=%llu pid=%d "
253 "time=%llu.%06us exit=%d",
254 node.name, node.eid, pid,
255 (unsigned long long) usage.ru_utime.tv_sec,
256 (unsigned int) usage.ru_utime.tv_usec,
257 WEXITSTATUS(status));
258 } else if (WIFSIGNALED(status)) {
259 zed_log_msg(LOG_INFO,
260 "Finished \"%s\" eid=%llu pid=%d "
261 "time=%llu.%06us sig=%d/%s",
262 node.name, node.eid, pid,
263 (unsigned long long) usage.ru_utime.tv_sec,
264 (unsigned int) usage.ru_utime.tv_usec,
265 WTERMSIG(status),
266 strsignal(WTERMSIG(status)));
267 } else {
268 zed_log_msg(LOG_INFO,
269 "Finished \"%s\" eid=%llu pid=%d "
270 "time=%llu.%06us status=0x%X",
271 node.name, node.eid, pid,
272 (unsigned long long) usage.ru_utime.tv_sec,
273 (unsigned int) usage.ru_utime.tv_usec,
274 (unsigned int) status);
275 }
276
277 free(node.name);
278 }
279 }
280
281 return (NULL);
282 }
283
284 void
zed_exec_fini(void)285 zed_exec_fini(void)
286 {
287 struct launched_process_node *node;
288 void *ck = NULL;
289
290 if (_reap_children_tid == (pthread_t)-1)
291 return;
292
293 _reap_children_stop = B_TRUE;
294 (void) pthread_kill(_reap_children_tid, SIGCHLD);
295 (void) pthread_join(_reap_children_tid, NULL);
296
297 while ((node = avl_destroy_nodes(&_launched_processes, &ck)) != NULL) {
298 free(node->name);
299 free(node);
300 }
301 avl_destroy(&_launched_processes);
302
303 (void) pthread_mutex_destroy(&_launched_processes_lock);
304 (void) pthread_mutex_init(&_launched_processes_lock, NULL);
305
306 _reap_children_tid = (pthread_t)-1;
307 }
308
309 /*
310 * Process the event [eid] by synchronously invoking all zedlets with a
311 * matching class prefix.
312 *
313 * Each executable in [zcp->zedlets] from the directory [zcp->zedlet_dir]
314 * is matched against the event's [class], [subclass], and the "all" class
315 * (which matches all events).
316 * Every zedlet with a matching class prefix is invoked.
317 * The NAME=VALUE strings in [envs] will be passed to the zedlet as
318 * environment variables.
319 *
320 * The file descriptor [zcp->zevent_fd] is the zevent_fd used to track the
321 * current cursor location within the zevent nvlist.
322 *
323 * Return 0 on success, -1 on error.
324 */
325 int
zed_exec_process(uint64_t eid,const char * class,const char * subclass,struct zed_conf * zcp,zed_strings_t * envs)326 zed_exec_process(uint64_t eid, const char *class, const char *subclass,
327 struct zed_conf *zcp, zed_strings_t *envs)
328 {
329 const char *class_strings[4];
330 const char *allclass = "all";
331 const char **csp;
332 const char *z;
333 char **e;
334 int n;
335
336 if (!zcp->zedlet_dir || !zcp->zedlets || !envs || zcp->zevent_fd < 0)
337 return (-1);
338
339 if (_reap_children_tid == (pthread_t)-1) {
340 _launched_processes_limit = zcp->max_jobs;
341
342 if (pthread_create(&_reap_children_tid, NULL,
343 _reap_children, NULL) != 0)
344 return (-1);
345 pthread_setname_np(_reap_children_tid, "reap ZEDLETs");
346
347 avl_create(&_launched_processes, _launched_process_node_compare,
348 sizeof (struct launched_process_node),
349 offsetof(struct launched_process_node, node));
350 }
351
352 csp = class_strings;
353
354 if (class)
355 *csp++ = class;
356
357 if (subclass)
358 *csp++ = subclass;
359
360 if (allclass)
361 *csp++ = allclass;
362
363 *csp = NULL;
364
365 e = _zed_exec_create_env(envs);
366
367 for (z = zed_strings_first(zcp->zedlets); z;
368 z = zed_strings_next(zcp->zedlets)) {
369 for (csp = class_strings; *csp; csp++) {
370 n = strlen(*csp);
371 if ((strncmp(z, *csp, n) == 0) && !isalpha(z[n]))
372 _zed_exec_fork_child(eid, zcp->zedlet_dir,
373 z, e, zcp->zevent_fd, zcp->do_foreground);
374 }
375 }
376 free(e);
377 return (0);
378 }
379