Remove mon_pipe for communicating from monitor to manager
[thirdparty/mdadm.git] / monitor.c
CommitLineData
549e9569
NB
1
2#include "mdadm.h"
3#include "mdmon.h"
4
5#include <sys/select.h>
1ed3f387 6#include <signal.h>
549e9569
NB
7
8static char *array_states[] = {
9 "clear", "inactive", "suspended", "readonly", "read-auto",
10 "clean", "active", "write-pending", "active-idle", NULL };
11static char *sync_actions[] = {
12 "idle", "reshape", "resync", "recover", "check", "repair", NULL
13};
14
15static int write_attr(char *attr, int fd)
16{
17 return write(fd, attr, strlen(attr));
18}
19
20static void add_fd(fd_set *fds, int *maxfd, int fd)
21{
22 if (fd < 0)
23 return;
24 if (fd > *maxfd)
25 *maxfd = fd;
26 FD_SET(fd, fds);
27}
28
29static int read_attr(char *buf, int len, int fd)
30{
31 int n;
32
33 if (fd < 0) {
34 buf[0] = 0;
35 return 0;
36 }
37 lseek(fd, 0, 0);
38 n = read(fd, buf, len - 1);
39
40 if (n <= 0) {
41 buf[0] = 0;
42 return 0;
43 }
44 buf[n] = 0;
45 if (buf[n-1] == '\n')
46 buf[n-1] = 0;
47 return n;
48}
49
549e9569 50
c052ba30
DW
51static int get_resync_start(struct active_array *a)
52{
53 char buf[30];
54 int n;
55
56 n = read_attr(buf, 30, a->resync_start_fd);
57 if (n <= 0)
58 return n;
59
60 a->resync_start = strtoull(buf, NULL, 10);
61
62 return 1;
63}
549e9569
NB
64
65static int attr_match(const char *attr, const char *str)
66{
67 /* See if attr, read from a sysfs file, matches
68 * str. They must either be the same, or attr can
69 * have a trailing newline or comma
70 */
71 while (*attr && *str && *attr == *str) {
72 attr++;
73 str++;
74 }
75
76 if (*str || (*attr && *attr != ',' && *attr != '\n'))
77 return 0;
78 return 1;
79}
80
81static int match_word(const char *word, char **list)
82{
83 int n;
84 for (n=0; list[n]; n++)
85 if (attr_match(word, list[n]))
86 break;
87 return n;
88}
89
90static enum array_state read_state(int fd)
91{
92 char buf[20];
93 int n = read_attr(buf, 20, fd);
94
95 if (n <= 0)
96 return bad_word;
97 return (enum array_state) match_word(buf, array_states);
98}
99
100static enum sync_action read_action( int fd)
101{
102 char buf[20];
103 int n = read_attr(buf, 20, fd);
104
105 if (n <= 0)
106 return bad_action;
107 return (enum sync_action) match_word(buf, sync_actions);
108}
109
549e9569
NB
110int read_dev_state(int fd)
111{
112 char buf[60];
113 int n = read_attr(buf, 60, fd);
114 char *cp;
115 int rv = 0;
116
117 if (n <= 0)
118 return 0;
119
120 cp = buf;
121 while (cp) {
8d45d196 122 if (attr_match(cp, "faulty"))
549e9569 123 rv |= DS_FAULTY;
8d45d196 124 if (attr_match(cp, "in_sync"))
549e9569 125 rv |= DS_INSYNC;
8d45d196 126 if (attr_match(cp, "write_mostly"))
549e9569 127 rv |= DS_WRITE_MOSTLY;
8d45d196 128 if (attr_match(cp, "spare"))
549e9569 129 rv |= DS_SPARE;
8d45d196
DW
130 if (attr_match(cp, "blocked"))
131 rv |= DS_BLOCKED;
549e9569
NB
132 cp = strchr(cp, ',');
133 if (cp)
134 cp++;
135 }
136 return rv;
137}
138
1ed3f387
NB
139static void signal_manager(void)
140{
141 kill(getpid(), SIGUSR1);
142}
549e9569
NB
143
144/* Monitor a set of active md arrays - all of which share the
145 * same metadata - and respond to events that require
146 * metadata update.
147 *
148 * New arrays are detected by another thread which allocates
149 * required memory and attaches the data structure to our list.
150 *
151 * Events:
152 * Array stops.
153 * This is detected by array_state going to 'clear' or 'inactive'.
154 * while we thought it was active.
155 * Response is to mark metadata as clean and 'clear' the array(??)
156 * write-pending
157 * array_state if 'write-pending'
158 * We mark metadata as 'dirty' then set array to 'active'.
159 * active_idle
160 * Either ignore, or mark clean, then mark metadata as clean.
161 *
162 * device fails
163 * detected by rd-N/state reporting "faulty"
8d45d196
DW
164 * mark device as 'failed' in metadata, let the kernel release the
165 * device by writing '-blocked' to rd/state, and finally write 'remove' to
0af73f61
DW
166 * rd/state. Before a disk can be replaced it must be failed and removed
167 * from all container members, this will be preemptive for the other
168 * arrays... safe?
549e9569
NB
169 *
170 * sync completes
171 * sync_action was 'resync' and becomes 'idle' and resync_start becomes
172 * MaxSector
173 * Notify metadata that sync is complete.
549e9569
NB
174 *
175 * recovery completes
176 * sync_action changes from 'recover' to 'idle'
177 * Check each device state and mark metadata if 'faulty' or 'in_sync'.
549e9569
NB
178 *
179 * deal with resync
c052ba30
DW
180 * This only happens on finding a new array... mdadm will have set
181 * 'resync_start' to the correct value. If 'resync_start' indicates that an
182 * resync needs to occur set the array to the 'active' state rather than the
183 * initial read-auto state.
549e9569
NB
184 *
185 *
186 *
187 * We wait for a change (poll/select) on array_state, sync_action, and
188 * each rd-X/state file.
189 * When we get any change, we check everything. So read each state file,
190 * then decide what to do.
191 *
192 * The core action is to write new metadata to all devices in the array.
193 * This is done at most once on any wakeup.
194 * After that we might:
195 * - update the array_state
196 * - set the role of some devices.
197 * - request a sync_action
198 *
199 */
200
201static int read_and_act(struct active_array *a)
202{
6c3fb95c 203 int check_degraded = 0;
2a0bb19e 204 int deactivate = 0;
549e9569
NB
205 struct mdinfo *mdi;
206
207 a->next_state = bad_word;
208 a->next_action = bad_action;
209
210 a->curr_state = read_state(a->info.state_fd);
211 a->curr_action = read_action(a->action_fd);
212 for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
213 mdi->next_state = 0;
57632f4a 214 if (mdi->state_fd >= 0)
8d45d196 215 mdi->curr_state = read_dev_state(mdi->state_fd);
549e9569
NB
216 }
217
218 if (a->curr_state <= inactive &&
219 a->prev_state > inactive) {
220 /* array has been stopped */
ed9d66aa 221 a->container->ss->set_array_state(a, 1);
549e9569 222 a->next_state = clear;
2a0bb19e 223 deactivate = 1;
549e9569
NB
224 }
225 if (a->curr_state == write_pending) {
ed9d66aa
NB
226 get_resync_start(a);
227 a->container->ss->set_array_state(a, 0);
549e9569
NB
228 a->next_state = active;
229 }
230 if (a->curr_state == active_idle) {
231 /* Set array to 'clean' FIRST, then
4e5528c6 232 * a->ss->mark_clean(a, ~0ULL);
549e9569
NB
233 * just ignore for now.
234 */
235 }
236
237 if (a->curr_state == readonly) {
238 /* Well, I'm ready to handle things, so
239 * read-auto is OK. FIXME what if we really want
240 * readonly ???
241 */
c052ba30 242 get_resync_start(a);
ed9d66aa 243 printf("Found a readonly array at %llu\n", a->resync_start);
c052ba30
DW
244 if (a->resync_start == ~0ULL)
245 a->next_state = read_auto; /* array is clean */
246 else {
ed9d66aa 247 a->container->ss->set_array_state(a, 0);
c052ba30
DW
248 a->next_state = active;
249 }
549e9569
NB
250 }
251
252 if (a->curr_action == idle &&
253 a->prev_action == resync) {
4e5528c6
NB
254 /* A resync has finished. The endpoint is recorded in
255 * 'sync_start'. We don't update the metadata
256 * until the array goes inactive or readonly though.
257 * Just check if we need to fiddle spares.
258 */
77402e51 259 get_resync_start(a);
ed9d66aa 260 a->container->ss->set_array_state(a, 0);
549e9569
NB
261 check_degraded = 1;
262 }
263
264 if (a->curr_action == idle &&
265 a->prev_action == recover) {
266 for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
8d45d196
DW
267 a->container->ss->set_disk(a, mdi->disk.raid_disk,
268 mdi->curr_state);
549e9569
NB
269 if (! (mdi->curr_state & DS_INSYNC))
270 check_degraded = 1;
271 }
272 }
273
549e9569
NB
274 for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
275 if (mdi->curr_state & DS_FAULTY) {
8d45d196
DW
276 a->container->ss->set_disk(a, mdi->disk.raid_disk,
277 mdi->curr_state);
549e9569
NB
278 check_degraded = 1;
279 mdi->next_state = DS_REMOVE;
280 }
281 }
282
2e735d19 283 a->container->ss->sync_metadata(a->container);
4e6e574a 284 dprintf("%s: update[%d]: (", __func__, a->info.container_member);
549e9569
NB
285
286 /* Effect state changes in the array */
4e6e574a
DW
287 if (a->next_state != bad_word) {
288 dprintf(" state:%s", array_states[a->next_state]);
549e9569 289 write_attr(array_states[a->next_state], a->info.state_fd);
4e6e574a
DW
290 }
291 if (a->next_action != bad_action) {
549e9569 292 write_attr(sync_actions[a->next_action], a->action_fd);
4e6e574a
DW
293 dprintf(" action:%s", array_states[a->next_state]);
294 }
549e9569 295 for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
57632f4a
NB
296 if (mdi->next_state == DS_REMOVE && mdi->state_fd >= 0) {
297 int remove_result;
8d45d196
DW
298
299 write_attr("-blocked", mdi->state_fd);
300 /* the kernel may not be able to immediately remove the
301 * disk, we can simply wait until the next event to try
302 * again.
303 */
4e6e574a 304 dprintf(" %d:-blocked", mdi->disk.raid_disk);
57632f4a
NB
305 remove_result = write_attr("remove", mdi->state_fd);
306 if (remove_result > 0) {
4e6e574a 307 dprintf(" %d:removed", mdi->disk.raid_disk);
8d45d196
DW
308 close(mdi->state_fd);
309 mdi->state_fd = -1;
310 }
311 }
4e6e574a 312 if (mdi->next_state & DS_INSYNC) {
549e9569 313 write_attr("+in_sync", mdi->state_fd);
4e6e574a
DW
314 dprintf(" %d:+in_sync", mdi->disk.raid_disk);
315 }
549e9569 316 }
4e6e574a 317 dprintf(" )\n");
549e9569
NB
318
319 /* move curr_ to prev_ */
320 a->prev_state = a->curr_state;
321
322 a->prev_action = a->curr_action;
323
324 for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
325 mdi->prev_state = mdi->curr_state;
326 mdi->next_state = 0;
327 }
328
7e1432fb
NB
329 if (check_degraded) {
330 /* manager will do the actual check */
331 a->check_degraded = 1;
332 signal_manager();
333 }
334
2a0bb19e
DW
335 if (deactivate)
336 a->container = NULL;
337
549e9569
NB
338 return 1;
339}
340
0af73f61
DW
341static struct mdinfo *
342find_device(struct active_array *a, int major, int minor)
343{
344 struct mdinfo *mdi;
345
346 for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
347 if (mdi->disk.major == major && mdi->disk.minor == minor)
348 return mdi;
349
350 return NULL;
351}
352
353static void reconcile_failed(struct active_array *aa, struct mdinfo *failed)
354{
355 struct active_array *a;
356 struct mdinfo *victim;
357
358 for (a = aa; a; a = a->next) {
359 if (!a->container)
360 continue;
361 victim = find_device(a, failed->disk.major, failed->disk.minor);
362 if (!victim)
363 continue;
364
365 if (!(victim->curr_state & DS_FAULTY))
366 write_attr("faulty", victim->state_fd);
367 }
368}
369
3e70c845
DW
370static int handle_pipe(struct md_generic_cmd *cmd, struct active_array *aa)
371{
372 switch (cmd->action) {
373 case md_action_ping_monitor:
374 return 0;
3e70c845
DW
375 }
376
377 return -1;
378}
379
4e6e574a
DW
380#ifdef DEBUG
381static void dprint_wake_reasons(fd_set *fds)
382{
383 int i;
384 char proc_path[256];
385 char link[256];
386 char *basename;
387 int rv;
388
389 fprintf(stderr, "monitor: wake ( ");
390 for (i = 0; i < FD_SETSIZE; i++) {
391 if (FD_ISSET(i, fds)) {
392 sprintf(proc_path, "/proc/%d/fd/%d",
393 (int) getpid(), i);
394
395 rv = readlink(proc_path, link, sizeof(link) - 1);
396 if (rv < 0) {
397 fprintf(stderr, "%d:unknown ", i);
398 continue;
399 }
400 link[rv] = '\0';
401 basename = strrchr(link, '/');
402 fprintf(stderr, "%d:%s ",
403 i, basename ? ++basename : link);
404 }
405 }
406 fprintf(stderr, ")\n");
407}
408#endif
409
e0d6609f 410static int wait_and_act(struct supertype *container, int pfd,
2f64e61a 411 int nowait)
549e9569
NB
412{
413 fd_set rfds;
414 int maxfd = 0;
e0d6609f 415 struct active_array **aap = &container->arrays;
1ed3f387 416 struct active_array *a, **ap;
549e9569 417 int rv;
0af73f61 418 struct mdinfo *mdi;
549e9569
NB
419
420 FD_ZERO(&rfds);
421
422 add_fd(&rfds, &maxfd, pfd);
1ed3f387
NB
423 for (ap = aap ; *ap ;) {
424 a = *ap;
425 /* once an array has been deactivated we want to
426 * ask the manager to discard it.
2a0bb19e 427 */
1ed3f387
NB
428 if (!a->container) {
429 if (discard_this) {
430 ap = &(*ap)->next;
431 continue;
432 }
433 *ap = a->next;
434 a->next = NULL;
435 discard_this = a;
436 signal_manager();
2a0bb19e 437 continue;
1ed3f387 438 }
2a0bb19e 439
549e9569
NB
440 add_fd(&rfds, &maxfd, a->info.state_fd);
441 add_fd(&rfds, &maxfd, a->action_fd);
442 for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
443 add_fd(&rfds, &maxfd, mdi->state_fd);
1ed3f387
NB
444
445 ap = &(*ap)->next;
549e9569
NB
446 }
447
e0d6609f
NB
448 if (manager_ready && *aap == NULL) {
449 /* No interesting arrays. Lets see about exiting.
450 * Note that blocking at this point is not a problem
451 * as there are no active arrays, there is nothing that
452 * we need to be ready to do.
453 */
454 int fd = open(container->device_name, O_RDONLY|O_EXCL);
455 if (fd >= 0 || errno != EBUSY) {
456 /* OK, we are safe to leave */
4e6e574a 457 dprintf("no arrays to monitor... exiting\n");
e0d6609f
NB
458 exit_now = 1;
459 signal_manager();
460 remove_pidfile(container->devname);
461 exit(0);
462 }
463 }
464
549e9569
NB
465 if (!nowait) {
466 rv = select(maxfd+1, &rfds, NULL, NULL, NULL);
467
4e6e574a
DW
468 #ifdef DEBUG
469 dprint_wake_reasons(&rfds);
470 #endif
471
549e9569
NB
472 if (rv <= 0)
473 return rv;
474
475 if (FD_ISSET(pfd, &rfds)) {
3e70c845
DW
476 int err = -1;
477
478 if (read(pfd, &err, 1) > 0)
1ed3f387 479 err = handle_pipe(active_cmd, *aap);
549e9569
NB
480 }
481 }
482
2e735d19
NB
483 if (update_queue) {
484 struct metadata_update *this;
485
486 for (this = update_queue; this ; this = this->next)
487 container->ss->process_update(container, this);
488
489 update_queue_handled = update_queue;
490 update_queue = NULL;
491 signal_manager();
492 container->ss->sync_metadata(container);
493 }
494
1ed3f387 495 for (a = *aap; a ; a = a->next) {
2a0bb19e 496 if (a->replaces && !discard_this) {
549e9569
NB
497 struct active_array **ap;
498 for (ap = &a->next; *ap && *ap != a->replaces;
499 ap = & (*ap)->next)
500 ;
501 if (*ap)
502 *ap = (*ap)->next;
503 discard_this = a->replaces;
504 a->replaces = NULL;
6c3fb95c 505 /* FIXME check if device->state_fd need to be cleared?*/
1ed3f387 506 signal_manager();
549e9569 507 }
2a0bb19e
DW
508 if (a->container)
509 rv += read_and_act(a);
549e9569 510 }
0af73f61
DW
511
512 /* propagate failures across container members */
1ed3f387 513 for (a = *aap; a ; a = a->next) {
0af73f61
DW
514 if (!a->container)
515 continue;
516 for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
517 if (mdi->curr_state & DS_FAULTY)
1ed3f387 518 reconcile_failed(*aap, mdi);
0af73f61
DW
519 }
520
549e9569
NB
521 return rv;
522}
523
524void do_monitor(struct supertype *container)
525{
526 int rv;
527 int first = 1;
528 do {
e0d6609f 529 rv = wait_and_act(container, container->mgr_pipe[0],
2f64e61a 530 first);
549e9569
NB
531 first = 0;
532 } while (rv >= 0);
533}