imsm: set map state depending on sync state
[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);
549e9569
NB
284
285 /* Effect state changes in the array */
286 if (a->next_state != bad_word)
287 write_attr(array_states[a->next_state], a->info.state_fd);
288 if (a->next_action != bad_action)
289 write_attr(sync_actions[a->next_action], a->action_fd);
290 for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
57632f4a
NB
291 if (mdi->next_state == DS_REMOVE && mdi->state_fd >= 0) {
292 int remove_result;
8d45d196
DW
293
294 write_attr("-blocked", mdi->state_fd);
295 /* the kernel may not be able to immediately remove the
296 * disk, we can simply wait until the next event to try
297 * again.
298 */
57632f4a
NB
299 remove_result = write_attr("remove", mdi->state_fd);
300 if (remove_result > 0) {
8d45d196
DW
301 close(mdi->state_fd);
302 mdi->state_fd = -1;
303 }
304 }
549e9569
NB
305 if (mdi->next_state & DS_INSYNC)
306 write_attr("+in_sync", mdi->state_fd);
307 }
308
309 /* move curr_ to prev_ */
310 a->prev_state = a->curr_state;
311
312 a->prev_action = a->curr_action;
313
314 for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
315 mdi->prev_state = mdi->curr_state;
316 mdi->next_state = 0;
317 }
318
7e1432fb
NB
319 if (check_degraded) {
320 /* manager will do the actual check */
321 a->check_degraded = 1;
322 signal_manager();
323 }
324
2a0bb19e
DW
325 if (deactivate)
326 a->container = NULL;
327
549e9569
NB
328 return 1;
329}
330
0af73f61
DW
331static struct mdinfo *
332find_device(struct active_array *a, int major, int minor)
333{
334 struct mdinfo *mdi;
335
336 for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
337 if (mdi->disk.major == major && mdi->disk.minor == minor)
338 return mdi;
339
340 return NULL;
341}
342
343static void reconcile_failed(struct active_array *aa, struct mdinfo *failed)
344{
345 struct active_array *a;
346 struct mdinfo *victim;
347
348 for (a = aa; a; a = a->next) {
349 if (!a->container)
350 continue;
351 victim = find_device(a, failed->disk.major, failed->disk.minor);
352 if (!victim)
353 continue;
354
355 if (!(victim->curr_state & DS_FAULTY))
356 write_attr("faulty", victim->state_fd);
357 }
358}
359
3e70c845
DW
360static int handle_remove_device(struct md_remove_device_cmd *cmd, struct active_array *aa)
361{
362 struct active_array *a;
363 struct mdinfo *victim;
364 int rv;
365
366 /* scan all arrays for the given device, if ->state_fd is closed (-1)
367 * in all cases then mark the disk as removed in the metadata.
368 * Otherwise reply that it is busy.
369 */
370
371 /* pass1 check that it is not in use anywhere */
372 /* note: we are safe from re-adds as long as the device exists in the
373 * container
374 */
375 for (a = aa; a; a = a->next) {
376 if (!a->container)
377 continue;
378 victim = find_device(a, major(cmd->rdev), minor(cmd->rdev));
379 if (!victim)
380 continue;
381 if (victim->state_fd > 0)
382 return -EBUSY;
383 }
384
385 /* pass2 schedule and process removal per array */
386 for (a = aa; a; a = a->next) {
387 if (!a->container)
388 continue;
389 victim = find_device(a, major(cmd->rdev), minor(cmd->rdev));
390 if (!victim)
391 continue;
392 victim->curr_state |= DS_REMOVE;
393 rv = read_and_act(a);
394 if (rv < 0)
395 return rv;
396 }
397
398 return 0;
399}
400
401static int handle_pipe(struct md_generic_cmd *cmd, struct active_array *aa)
402{
403 switch (cmd->action) {
404 case md_action_ping_monitor:
405 return 0;
406 case md_action_remove_device:
407 return handle_remove_device((void *) cmd, aa);
408 }
409
410 return -1;
411}
412
e0d6609f 413static int wait_and_act(struct supertype *container, int pfd,
1ed3f387 414 int monfd, int nowait)
549e9569
NB
415{
416 fd_set rfds;
417 int maxfd = 0;
e0d6609f 418 struct active_array **aap = &container->arrays;
1ed3f387 419 struct active_array *a, **ap;
549e9569 420 int rv;
0af73f61 421 struct mdinfo *mdi;
549e9569
NB
422
423 FD_ZERO(&rfds);
424
425 add_fd(&rfds, &maxfd, pfd);
1ed3f387
NB
426 for (ap = aap ; *ap ;) {
427 a = *ap;
428 /* once an array has been deactivated we want to
429 * ask the manager to discard it.
2a0bb19e 430 */
1ed3f387
NB
431 if (!a->container) {
432 if (discard_this) {
433 ap = &(*ap)->next;
434 continue;
435 }
436 *ap = a->next;
437 a->next = NULL;
438 discard_this = a;
439 signal_manager();
2a0bb19e 440 continue;
1ed3f387 441 }
2a0bb19e 442
549e9569
NB
443 add_fd(&rfds, &maxfd, a->info.state_fd);
444 add_fd(&rfds, &maxfd, a->action_fd);
445 for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
446 add_fd(&rfds, &maxfd, mdi->state_fd);
1ed3f387
NB
447
448 ap = &(*ap)->next;
549e9569
NB
449 }
450
e0d6609f
NB
451 if (manager_ready && *aap == NULL) {
452 /* No interesting arrays. Lets see about exiting.
453 * Note that blocking at this point is not a problem
454 * as there are no active arrays, there is nothing that
455 * we need to be ready to do.
456 */
457 int fd = open(container->device_name, O_RDONLY|O_EXCL);
458 if (fd >= 0 || errno != EBUSY) {
459 /* OK, we are safe to leave */
460 exit_now = 1;
461 signal_manager();
462 remove_pidfile(container->devname);
463 exit(0);
464 }
465 }
466
549e9569
NB
467 if (!nowait) {
468 rv = select(maxfd+1, &rfds, NULL, NULL, NULL);
469
470 if (rv <= 0)
471 return rv;
472
473 if (FD_ISSET(pfd, &rfds)) {
3e70c845
DW
474 int err = -1;
475
476 if (read(pfd, &err, 1) > 0)
1ed3f387 477 err = handle_pipe(active_cmd, *aap);
3e70c845 478 write(monfd, &err, 1);
549e9569
NB
479 }
480 }
481
2e735d19
NB
482 if (update_queue) {
483 struct metadata_update *this;
484
485 for (this = update_queue; this ; this = this->next)
486 container->ss->process_update(container, this);
487
488 update_queue_handled = update_queue;
489 update_queue = NULL;
490 signal_manager();
491 container->ss->sync_metadata(container);
492 }
493
1ed3f387 494 for (a = *aap; a ; a = a->next) {
2a0bb19e 495 if (a->replaces && !discard_this) {
549e9569
NB
496 struct active_array **ap;
497 for (ap = &a->next; *ap && *ap != a->replaces;
498 ap = & (*ap)->next)
499 ;
500 if (*ap)
501 *ap = (*ap)->next;
502 discard_this = a->replaces;
503 a->replaces = NULL;
6c3fb95c 504 /* FIXME check if device->state_fd need to be cleared?*/
1ed3f387 505 signal_manager();
549e9569 506 }
2a0bb19e
DW
507 if (a->container)
508 rv += read_and_act(a);
549e9569 509 }
0af73f61
DW
510
511 /* propagate failures across container members */
1ed3f387 512 for (a = *aap; a ; a = a->next) {
0af73f61
DW
513 if (!a->container)
514 continue;
515 for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
516 if (mdi->curr_state & DS_FAULTY)
1ed3f387 517 reconcile_failed(*aap, mdi);
0af73f61
DW
518 }
519
549e9569
NB
520 return rv;
521}
522
523void do_monitor(struct supertype *container)
524{
525 int rv;
526 int first = 1;
527 do {
e0d6609f 528 rv = wait_and_act(container, container->mgr_pipe[0],
3e70c845 529 container->mon_pipe[1], first);
549e9569
NB
530 first = 0;
531 } while (rv >= 0);
532}