]>
Commit | Line | Data |
---|---|---|
a54d5262 DW |
1 | /* |
2 | * mdmon - monitor external metadata arrays | |
3 | * | |
e736b623 N |
4 | * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de> |
5 | * Copyright (C) 2007-2009 Intel Corporation | |
a54d5262 DW |
6 | * |
7 | * This program is free software; you can redistribute it and/or modify it | |
8 | * under the terms and conditions of the GNU General Public License, | |
9 | * version 2, as published by the Free Software Foundation. | |
10 | * | |
11 | * This program is distributed in the hope it will be useful, but WITHOUT | |
12 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
13 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
14 | * more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License along with | |
17 | * this program; if not, write to the Free Software Foundation, Inc., | |
18 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
19 | */ | |
549e9569 NB |
20 | |
21 | #include "mdadm.h" | |
22 | #include "mdmon.h" | |
4d43913c | 23 | #include <sys/syscall.h> |
549e9569 | 24 | #include <sys/select.h> |
1ed3f387 | 25 | #include <signal.h> |
549e9569 NB |
26 | |
27 | static char *array_states[] = { | |
28 | "clear", "inactive", "suspended", "readonly", "read-auto", | |
29 | "clean", "active", "write-pending", "active-idle", NULL }; | |
30 | static char *sync_actions[] = { | |
31 | "idle", "reshape", "resync", "recover", "check", "repair", NULL | |
32 | }; | |
33 | ||
34 | static int write_attr(char *attr, int fd) | |
35 | { | |
36 | return write(fd, attr, strlen(attr)); | |
37 | } | |
38 | ||
39 | static void add_fd(fd_set *fds, int *maxfd, int fd) | |
40 | { | |
41 | if (fd < 0) | |
42 | return; | |
43 | if (fd > *maxfd) | |
44 | *maxfd = fd; | |
45 | FD_SET(fd, fds); | |
46 | } | |
47 | ||
48 | static int read_attr(char *buf, int len, int fd) | |
49 | { | |
50 | int n; | |
51 | ||
52 | if (fd < 0) { | |
53 | buf[0] = 0; | |
54 | return 0; | |
55 | } | |
56 | lseek(fd, 0, 0); | |
57 | n = read(fd, buf, len - 1); | |
58 | ||
59 | if (n <= 0) { | |
60 | buf[0] = 0; | |
61 | return 0; | |
62 | } | |
63 | buf[n] = 0; | |
64 | if (buf[n-1] == '\n') | |
65 | buf[n-1] = 0; | |
66 | return n; | |
67 | } | |
68 | ||
b7941fd6 | 69 | static unsigned long long read_resync_start(int fd) |
c052ba30 DW |
70 | { |
71 | char buf[30]; | |
72 | int n; | |
73 | ||
b7941fd6 | 74 | n = read_attr(buf, 30, fd); |
c052ba30 | 75 | if (n <= 0) |
b7941fd6 | 76 | return 0; |
7e7fffc4 | 77 | if (strncmp(buf, "none", 4) == 0) |
b7528a20 | 78 | return MaxSector; |
7e7fffc4 | 79 | else |
b7941fd6 | 80 | return strtoull(buf, NULL, 10); |
c052ba30 | 81 | } |
549e9569 | 82 | |
484240d8 DW |
83 | static unsigned long long read_sync_completed(int fd) |
84 | { | |
85 | unsigned long long val; | |
86 | char buf[50]; | |
87 | int n; | |
88 | char *ep; | |
89 | ||
90 | n = read_attr(buf, 50, fd); | |
91 | ||
92 | if (n <= 0) | |
93 | return 0; | |
94 | buf[n] = 0; | |
95 | val = strtoull(buf, &ep, 0); | |
96 | if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' ')) | |
97 | return 0; | |
98 | return val; | |
99 | } | |
100 | ||
549e9569 NB |
101 | static enum array_state read_state(int fd) |
102 | { | |
103 | char buf[20]; | |
104 | int n = read_attr(buf, 20, fd); | |
105 | ||
106 | if (n <= 0) | |
107 | return bad_word; | |
1770662b | 108 | return (enum array_state) sysfs_match_word(buf, array_states); |
549e9569 NB |
109 | } |
110 | ||
111 | static enum sync_action read_action( int fd) | |
112 | { | |
113 | char buf[20]; | |
114 | int n = read_attr(buf, 20, fd); | |
115 | ||
116 | if (n <= 0) | |
117 | return bad_action; | |
1770662b | 118 | return (enum sync_action) sysfs_match_word(buf, sync_actions); |
549e9569 NB |
119 | } |
120 | ||
549e9569 NB |
121 | int read_dev_state(int fd) |
122 | { | |
123 | char buf[60]; | |
124 | int n = read_attr(buf, 60, fd); | |
125 | char *cp; | |
126 | int rv = 0; | |
127 | ||
128 | if (n <= 0) | |
129 | return 0; | |
130 | ||
131 | cp = buf; | |
132 | while (cp) { | |
1770662b | 133 | if (sysfs_attr_match(cp, "faulty")) |
549e9569 | 134 | rv |= DS_FAULTY; |
1770662b | 135 | if (sysfs_attr_match(cp, "in_sync")) |
549e9569 | 136 | rv |= DS_INSYNC; |
1770662b | 137 | if (sysfs_attr_match(cp, "write_mostly")) |
549e9569 | 138 | rv |= DS_WRITE_MOSTLY; |
1770662b | 139 | if (sysfs_attr_match(cp, "spare")) |
549e9569 | 140 | rv |= DS_SPARE; |
1770662b | 141 | if (sysfs_attr_match(cp, "blocked")) |
8d45d196 | 142 | rv |= DS_BLOCKED; |
549e9569 NB |
143 | cp = strchr(cp, ','); |
144 | if (cp) | |
145 | cp++; | |
146 | } | |
147 | return rv; | |
148 | } | |
149 | ||
1ed3f387 NB |
150 | static void signal_manager(void) |
151 | { | |
4d43913c NB |
152 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ |
153 | int pid = getpid(); | |
154 | syscall(SYS_tgkill, pid, mgr_tid, SIGUSR1); | |
1ed3f387 | 155 | } |
549e9569 NB |
156 | |
157 | /* Monitor a set of active md arrays - all of which share the | |
158 | * same metadata - and respond to events that require | |
159 | * metadata update. | |
160 | * | |
161 | * New arrays are detected by another thread which allocates | |
162 | * required memory and attaches the data structure to our list. | |
163 | * | |
164 | * Events: | |
165 | * Array stops. | |
166 | * This is detected by array_state going to 'clear' or 'inactive'. | |
167 | * while we thought it was active. | |
168 | * Response is to mark metadata as clean and 'clear' the array(??) | |
169 | * write-pending | |
170 | * array_state if 'write-pending' | |
171 | * We mark metadata as 'dirty' then set array to 'active'. | |
172 | * active_idle | |
173 | * Either ignore, or mark clean, then mark metadata as clean. | |
174 | * | |
175 | * device fails | |
176 | * detected by rd-N/state reporting "faulty" | |
8d45d196 DW |
177 | * mark device as 'failed' in metadata, let the kernel release the |
178 | * device by writing '-blocked' to rd/state, and finally write 'remove' to | |
0af73f61 DW |
179 | * rd/state. Before a disk can be replaced it must be failed and removed |
180 | * from all container members, this will be preemptive for the other | |
181 | * arrays... safe? | |
549e9569 NB |
182 | * |
183 | * sync completes | |
184 | * sync_action was 'resync' and becomes 'idle' and resync_start becomes | |
185 | * MaxSector | |
186 | * Notify metadata that sync is complete. | |
549e9569 NB |
187 | * |
188 | * recovery completes | |
189 | * sync_action changes from 'recover' to 'idle' | |
190 | * Check each device state and mark metadata if 'faulty' or 'in_sync'. | |
549e9569 NB |
191 | * |
192 | * deal with resync | |
c052ba30 DW |
193 | * This only happens on finding a new array... mdadm will have set |
194 | * 'resync_start' to the correct value. If 'resync_start' indicates that an | |
195 | * resync needs to occur set the array to the 'active' state rather than the | |
196 | * initial read-auto state. | |
549e9569 NB |
197 | * |
198 | * | |
199 | * | |
200 | * We wait for a change (poll/select) on array_state, sync_action, and | |
201 | * each rd-X/state file. | |
202 | * When we get any change, we check everything. So read each state file, | |
203 | * then decide what to do. | |
204 | * | |
205 | * The core action is to write new metadata to all devices in the array. | |
206 | * This is done at most once on any wakeup. | |
207 | * After that we might: | |
208 | * - update the array_state | |
209 | * - set the role of some devices. | |
210 | * - request a sync_action | |
211 | * | |
212 | */ | |
213 | ||
77b3ac8c | 214 | #define ARRAY_DIRTY 1 |
68226a80 | 215 | #define ARRAY_BUSY 2 |
549e9569 NB |
216 | static int read_and_act(struct active_array *a) |
217 | { | |
484240d8 | 218 | unsigned long long sync_completed; |
6c3fb95c | 219 | int check_degraded = 0; |
0f99b4bd | 220 | int check_reshape = 0; |
2a0bb19e | 221 | int deactivate = 0; |
549e9569 | 222 | struct mdinfo *mdi; |
77b3ac8c | 223 | int ret = 0; |
3a5d0473 | 224 | int count = 0; |
549e9569 NB |
225 | |
226 | a->next_state = bad_word; | |
227 | a->next_action = bad_action; | |
228 | ||
229 | a->curr_state = read_state(a->info.state_fd); | |
230 | a->curr_action = read_action(a->action_fd); | |
b7941fd6 | 231 | a->info.resync_start = read_resync_start(a->resync_start_fd); |
484240d8 | 232 | sync_completed = read_sync_completed(a->sync_completed_fd); |
549e9569 NB |
233 | for (mdi = a->info.devs; mdi ; mdi = mdi->next) { |
234 | mdi->next_state = 0; | |
10ce1808 | 235 | mdi->curr_state = 0; |
e1516be1 DW |
236 | if (mdi->state_fd >= 0) { |
237 | mdi->recovery_start = read_resync_start(mdi->recovery_fd); | |
8d45d196 | 238 | mdi->curr_state = read_dev_state(mdi->state_fd); |
e1516be1 | 239 | } |
549e9569 NB |
240 | } |
241 | ||
4867e068 AK |
242 | if (a->curr_state > inactive && |
243 | a->prev_state == inactive) { | |
244 | /* array has been started | |
245 | * possible that container operation has to be completed | |
246 | */ | |
247 | a->container->ss->set_array_state(a, 0); | |
248 | } | |
549e9569 NB |
249 | if (a->curr_state <= inactive && |
250 | a->prev_state > inactive) { | |
251 | /* array has been stopped */ | |
ed9d66aa | 252 | a->container->ss->set_array_state(a, 1); |
549e9569 | 253 | a->next_state = clear; |
2a0bb19e | 254 | deactivate = 1; |
549e9569 NB |
255 | } |
256 | if (a->curr_state == write_pending) { | |
ed9d66aa | 257 | a->container->ss->set_array_state(a, 0); |
549e9569 | 258 | a->next_state = active; |
77b3ac8c | 259 | ret |= ARRAY_DIRTY; |
549e9569 NB |
260 | } |
261 | if (a->curr_state == active_idle) { | |
d797a062 DW |
262 | /* Set array to 'clean' FIRST, then mark clean |
263 | * in the metadata | |
549e9569 | 264 | */ |
d797a062 | 265 | a->next_state = clean; |
77b3ac8c | 266 | ret |= ARRAY_DIRTY; |
d797a062 DW |
267 | } |
268 | if (a->curr_state == clean) { | |
d797a062 | 269 | a->container->ss->set_array_state(a, 1); |
549e9569 | 270 | } |
140d3685 DW |
271 | if (a->curr_state == active || |
272 | a->curr_state == suspended || | |
273 | a->curr_state == bad_word) | |
77b3ac8c | 274 | ret |= ARRAY_DIRTY; |
549e9569 | 275 | if (a->curr_state == readonly) { |
e9dd1598 N |
276 | /* Well, I'm ready to handle things. If readonly |
277 | * wasn't requested, transition to read-auto. | |
549e9569 | 278 | */ |
e9dd1598 N |
279 | char buf[64]; |
280 | read_attr(buf, sizeof(buf), a->metadata_fd); | |
281 | if (strncmp(buf, "external:-", 10) == 0) { | |
282 | /* explicit request for readonly array. Leave it alone */ | |
283 | ; | |
284 | } else { | |
e9dd1598 N |
285 | if (a->container->ss->set_array_state(a, 2)) |
286 | a->next_state = read_auto; /* array is clean */ | |
140d3685 | 287 | else { |
e9dd1598 | 288 | a->next_state = active; /* Now active for recovery etc */ |
77b3ac8c | 289 | ret |= ARRAY_DIRTY; |
140d3685 | 290 | } |
e9dd1598 | 291 | } |
549e9569 NB |
292 | } |
293 | ||
00e02142 DW |
294 | if (!deactivate && |
295 | a->curr_action == idle && | |
549e9569 | 296 | a->prev_action == resync) { |
4e5528c6 NB |
297 | /* A resync has finished. The endpoint is recorded in |
298 | * 'sync_start'. We don't update the metadata | |
299 | * until the array goes inactive or readonly though. | |
300 | * Just check if we need to fiddle spares. | |
301 | */ | |
0c0c44db | 302 | a->container->ss->set_array_state(a, a->curr_state <= clean); |
549e9569 NB |
303 | check_degraded = 1; |
304 | } | |
305 | ||
00e02142 DW |
306 | if (!deactivate && |
307 | a->curr_action == idle && | |
549e9569 | 308 | a->prev_action == recover) { |
0a6bdbee DW |
309 | /* A recovery has finished. Some disks may be in sync now, |
310 | * and the array may no longer be degraded | |
311 | */ | |
549e9569 | 312 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) { |
8d45d196 DW |
313 | a->container->ss->set_disk(a, mdi->disk.raid_disk, |
314 | mdi->curr_state); | |
549e9569 NB |
315 | if (! (mdi->curr_state & DS_INSYNC)) |
316 | check_degraded = 1; | |
3a5d0473 | 317 | count++; |
549e9569 | 318 | } |
3a5d0473 KW |
319 | if (count != a->info.array.raid_disks) |
320 | check_degraded = 1; | |
549e9569 NB |
321 | } |
322 | ||
0f99b4bd N |
323 | if (!deactivate && |
324 | a->curr_action == reshape && | |
325 | a->prev_action != reshape) | |
326 | /* reshape was requested by mdadm. Need to see if | |
327 | * new devices have been added. Manager does that | |
328 | * when it sees check_reshape | |
329 | */ | |
330 | check_reshape = 1; | |
331 | ||
92967543 DW |
332 | /* Check for failures and if found: |
333 | * 1/ Record the failure in the metadata and unblock the device. | |
334 | * FIXME update the kernel to stop notifying on failed drives when | |
335 | * the array is readonly and we have cleared 'blocked' | |
336 | * 2/ Try to remove the device if the array is writable, or can be | |
337 | * made writable. | |
338 | */ | |
549e9569 NB |
339 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) { |
340 | if (mdi->curr_state & DS_FAULTY) { | |
8d45d196 DW |
341 | a->container->ss->set_disk(a, mdi->disk.raid_disk, |
342 | mdi->curr_state); | |
549e9569 | 343 | check_degraded = 1; |
1c278e81 N |
344 | if (mdi->curr_state & DS_BLOCKED) |
345 | mdi->next_state |= DS_UNBLOCK; | |
92967543 DW |
346 | if (a->curr_state == read_auto) { |
347 | a->container->ss->set_array_state(a, 0); | |
348 | a->next_state = active; | |
349 | } | |
350 | if (a->curr_state > readonly) | |
351 | mdi->next_state |= DS_REMOVE; | |
549e9569 NB |
352 | } |
353 | } | |
354 | ||
484240d8 DW |
355 | /* Check for recovery checkpoint notifications. We need to be a |
356 | * minimum distance away from the last checkpoint to prevent | |
aad6f216 N |
357 | * over checkpointing. Note reshape checkpointing is handled |
358 | * in the second branch. | |
484240d8 DW |
359 | */ |
360 | if (sync_completed > a->last_checkpoint && | |
361 | sync_completed - a->last_checkpoint > a->info.component_size >> 4 && | |
4f0a7acc DW |
362 | a->curr_action > reshape) { |
363 | /* A (non-reshape) sync_action has reached a checkpoint. | |
364 | * Record the updated position in the metadata | |
365 | */ | |
366 | a->last_checkpoint = sync_completed; | |
367 | a->container->ss->set_array_state(a, a->curr_state <= clean); | |
aad6f216 N |
368 | } else if ((a->curr_action == idle && a->prev_action == reshape) || |
369 | (a->curr_action == reshape | |
370 | && sync_completed > a->last_checkpoint) ) { | |
371 | /* Reshape has progressed or completed so we need to | |
372 | * update the array state - and possibly the array size | |
373 | */ | |
2a9f8409 AK |
374 | if (sync_completed != 0) |
375 | a->last_checkpoint = sync_completed; | |
6d4225a1 AK |
376 | /* We might need to update last_checkpoint depending on |
377 | * the reason that reshape finished. | |
378 | * if array reshape is really finished: | |
379 | * set check point to the end, this allows | |
380 | * set_array_state() to finalize reshape in metadata | |
381 | * if reshape if broken: do not set checkpoint to the end | |
382 | * this allows for reshape restart from checkpoint | |
383 | */ | |
384 | if ((a->curr_action != reshape) && | |
385 | (a->prev_action == reshape)) { | |
386 | char buf[40]; | |
387 | if ((sysfs_get_str(&a->info, NULL, | |
388 | "reshape_position", | |
389 | buf, | |
390 | sizeof(buf)) >= 0) && | |
391 | strncmp(buf, "none", 4) == 0) | |
392 | a->last_checkpoint = a->info.component_size; | |
393 | } | |
aad6f216 | 394 | a->container->ss->set_array_state(a, a->curr_state <= clean); |
2a9f8409 | 395 | a->last_checkpoint = sync_completed; |
aad6f216 N |
396 | } |
397 | ||
398 | if (sync_completed > a->last_checkpoint) | |
484240d8 | 399 | a->last_checkpoint = sync_completed; |
484240d8 | 400 | |
2e735d19 | 401 | a->container->ss->sync_metadata(a->container); |
4065aa81 DW |
402 | dprintf("%s(%d): state:%s action:%s next(", __func__, a->info.container_member, |
403 | array_states[a->curr_state], sync_actions[a->curr_action]); | |
549e9569 NB |
404 | |
405 | /* Effect state changes in the array */ | |
4e6e574a DW |
406 | if (a->next_state != bad_word) { |
407 | dprintf(" state:%s", array_states[a->next_state]); | |
549e9569 | 408 | write_attr(array_states[a->next_state], a->info.state_fd); |
4e6e574a DW |
409 | } |
410 | if (a->next_action != bad_action) { | |
549e9569 | 411 | write_attr(sync_actions[a->next_action], a->action_fd); |
4065aa81 | 412 | dprintf(" action:%s", sync_actions[a->next_action]); |
4e6e574a | 413 | } |
549e9569 | 414 | for (mdi = a->info.devs; mdi ; mdi = mdi->next) { |
92967543 DW |
415 | if (mdi->next_state & DS_UNBLOCK) { |
416 | dprintf(" %d:-blocked", mdi->disk.raid_disk); | |
417 | write_attr("-blocked", mdi->state_fd); | |
418 | } | |
419 | ||
420 | if ((mdi->next_state & DS_REMOVE) && mdi->state_fd >= 0) { | |
57632f4a | 421 | int remove_result; |
8d45d196 | 422 | |
68226a80 N |
423 | /* The kernel may not be able to immediately remove the |
424 | * disk. In that case we wait a little while and | |
425 | * try again. | |
8d45d196 | 426 | */ |
57632f4a NB |
427 | remove_result = write_attr("remove", mdi->state_fd); |
428 | if (remove_result > 0) { | |
4e6e574a | 429 | dprintf(" %d:removed", mdi->disk.raid_disk); |
8d45d196 | 430 | close(mdi->state_fd); |
e40512fd | 431 | close(mdi->recovery_fd); |
8d45d196 | 432 | mdi->state_fd = -1; |
68226a80 N |
433 | } else |
434 | ret |= ARRAY_BUSY; | |
8d45d196 | 435 | } |
4e6e574a | 436 | if (mdi->next_state & DS_INSYNC) { |
549e9569 | 437 | write_attr("+in_sync", mdi->state_fd); |
4e6e574a DW |
438 | dprintf(" %d:+in_sync", mdi->disk.raid_disk); |
439 | } | |
549e9569 | 440 | } |
4e6e574a | 441 | dprintf(" )\n"); |
549e9569 NB |
442 | |
443 | /* move curr_ to prev_ */ | |
444 | a->prev_state = a->curr_state; | |
445 | ||
446 | a->prev_action = a->curr_action; | |
447 | ||
448 | for (mdi = a->info.devs; mdi ; mdi = mdi->next) { | |
449 | mdi->prev_state = mdi->curr_state; | |
450 | mdi->next_state = 0; | |
451 | } | |
452 | ||
0f99b4bd | 453 | if (check_degraded || check_reshape) { |
7e1432fb | 454 | /* manager will do the actual check */ |
0f99b4bd N |
455 | if (check_degraded) |
456 | a->check_degraded = 1; | |
457 | if (check_reshape) | |
458 | a->check_reshape = 1; | |
7e1432fb NB |
459 | signal_manager(); |
460 | } | |
461 | ||
2a0bb19e DW |
462 | if (deactivate) |
463 | a->container = NULL; | |
464 | ||
77b3ac8c | 465 | return ret; |
549e9569 NB |
466 | } |
467 | ||
0af73f61 DW |
468 | static struct mdinfo * |
469 | find_device(struct active_array *a, int major, int minor) | |
470 | { | |
471 | struct mdinfo *mdi; | |
472 | ||
473 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) | |
474 | if (mdi->disk.major == major && mdi->disk.minor == minor) | |
475 | return mdi; | |
476 | ||
477 | return NULL; | |
478 | } | |
479 | ||
480 | static void reconcile_failed(struct active_array *aa, struct mdinfo *failed) | |
481 | { | |
482 | struct active_array *a; | |
483 | struct mdinfo *victim; | |
484 | ||
485 | for (a = aa; a; a = a->next) { | |
ba714450 | 486 | if (!a->container || a->to_remove) |
0af73f61 DW |
487 | continue; |
488 | victim = find_device(a, failed->disk.major, failed->disk.minor); | |
489 | if (!victim) | |
490 | continue; | |
491 | ||
492 | if (!(victim->curr_state & DS_FAULTY)) | |
493 | write_attr("faulty", victim->state_fd); | |
494 | } | |
495 | } | |
496 | ||
4e6e574a DW |
497 | #ifdef DEBUG |
498 | static void dprint_wake_reasons(fd_set *fds) | |
499 | { | |
500 | int i; | |
501 | char proc_path[256]; | |
502 | char link[256]; | |
503 | char *basename; | |
504 | int rv; | |
505 | ||
506 | fprintf(stderr, "monitor: wake ( "); | |
507 | for (i = 0; i < FD_SETSIZE; i++) { | |
508 | if (FD_ISSET(i, fds)) { | |
509 | sprintf(proc_path, "/proc/%d/fd/%d", | |
510 | (int) getpid(), i); | |
511 | ||
512 | rv = readlink(proc_path, link, sizeof(link) - 1); | |
513 | if (rv < 0) { | |
514 | fprintf(stderr, "%d:unknown ", i); | |
515 | continue; | |
516 | } | |
517 | link[rv] = '\0'; | |
518 | basename = strrchr(link, '/'); | |
519 | fprintf(stderr, "%d:%s ", | |
520 | i, basename ? ++basename : link); | |
521 | } | |
522 | } | |
523 | fprintf(stderr, ")\n"); | |
524 | } | |
525 | #endif | |
526 | ||
1eb252b8 N |
527 | int monitor_loop_cnt; |
528 | ||
4d43913c | 529 | static int wait_and_act(struct supertype *container, int nowait) |
549e9569 NB |
530 | { |
531 | fd_set rfds; | |
532 | int maxfd = 0; | |
e0d6609f | 533 | struct active_array **aap = &container->arrays; |
1ed3f387 | 534 | struct active_array *a, **ap; |
549e9569 | 535 | int rv; |
0af73f61 | 536 | struct mdinfo *mdi; |
6144ed44 | 537 | static unsigned int dirty_arrays = ~0; /* start at some non-zero value */ |
549e9569 NB |
538 | |
539 | FD_ZERO(&rfds); | |
540 | ||
1ed3f387 NB |
541 | for (ap = aap ; *ap ;) { |
542 | a = *ap; | |
543 | /* once an array has been deactivated we want to | |
544 | * ask the manager to discard it. | |
2a0bb19e | 545 | */ |
ba714450 | 546 | if (!a->container || a->to_remove) { |
1ed3f387 NB |
547 | if (discard_this) { |
548 | ap = &(*ap)->next; | |
549 | continue; | |
550 | } | |
551 | *ap = a->next; | |
552 | a->next = NULL; | |
553 | discard_this = a; | |
554 | signal_manager(); | |
2a0bb19e | 555 | continue; |
1ed3f387 | 556 | } |
2a0bb19e | 557 | |
549e9569 NB |
558 | add_fd(&rfds, &maxfd, a->info.state_fd); |
559 | add_fd(&rfds, &maxfd, a->action_fd); | |
484240d8 | 560 | add_fd(&rfds, &maxfd, a->sync_completed_fd); |
549e9569 NB |
561 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) |
562 | add_fd(&rfds, &maxfd, mdi->state_fd); | |
1ed3f387 NB |
563 | |
564 | ap = &(*ap)->next; | |
549e9569 NB |
565 | } |
566 | ||
6144ed44 DW |
567 | if (manager_ready && (*aap == NULL || (sigterm && !dirty_arrays))) { |
568 | /* No interesting arrays, or we have been told to | |
569 | * terminate and everything is clean. Lets see about | |
570 | * exiting. Note that blocking at this point is not a | |
571 | * problem as there are no active arrays, there is | |
572 | * nothing that we need to be ready to do. | |
e0d6609f | 573 | */ |
d998b738 N |
574 | int fd; |
575 | if (sigterm) | |
4dd2df09 | 576 | fd = open_dev_excl(container->devnm); |
d998b738 | 577 | else |
4dd2df09 | 578 | fd = open_dev_flags(container->devnm, O_RDONLY|O_EXCL); |
e0d6609f NB |
579 | if (fd >= 0 || errno != EBUSY) { |
580 | /* OK, we are safe to leave */ | |
6144ed44 DW |
581 | if (sigterm && !dirty_arrays) |
582 | dprintf("caught sigterm, all clean... exiting\n"); | |
583 | else | |
584 | dprintf("no arrays to monitor... exiting\n"); | |
fa716c83 N |
585 | if (!sigterm) |
586 | /* On SIGTERM, someone (the take-over mdmon) will | |
587 | * clean up | |
588 | */ | |
4dd2df09 | 589 | remove_pidfile(container->devnm); |
e0d6609f NB |
590 | exit_now = 1; |
591 | signal_manager(); | |
6f4cdfd9 | 592 | close(fd); |
e0d6609f NB |
593 | exit(0); |
594 | } | |
595 | } | |
596 | ||
549e9569 | 597 | if (!nowait) { |
4d43913c | 598 | sigset_t set; |
d998b738 N |
599 | struct timespec ts; |
600 | ts.tv_sec = 24*3600; | |
601 | ts.tv_nsec = 0; | |
68226a80 | 602 | if (*aap == NULL || container->retry_soon) { |
d998b738 N |
603 | /* just waiting to get O_EXCL access */ |
604 | ts.tv_sec = 0; | |
605 | ts.tv_nsec = 20000000ULL; | |
606 | } | |
4d43913c NB |
607 | sigprocmask(SIG_UNBLOCK, NULL, &set); |
608 | sigdelset(&set, SIGUSR1); | |
1eb252b8 | 609 | monitor_loop_cnt |= 1; |
d998b738 | 610 | rv = pselect(maxfd+1, NULL, NULL, &rfds, &ts, &set); |
1eb252b8 | 611 | monitor_loop_cnt += 1; |
bfa44e2e NB |
612 | if (rv == -1 && errno == EINTR) |
613 | rv = 0; | |
4e6e574a DW |
614 | #ifdef DEBUG |
615 | dprint_wake_reasons(&rfds); | |
616 | #endif | |
68226a80 | 617 | container->retry_soon = 0; |
549e9569 NB |
618 | } |
619 | ||
2e735d19 NB |
620 | if (update_queue) { |
621 | struct metadata_update *this; | |
622 | ||
623 | for (this = update_queue; this ; this = this->next) | |
624 | container->ss->process_update(container, this); | |
625 | ||
626 | update_queue_handled = update_queue; | |
627 | update_queue = NULL; | |
628 | signal_manager(); | |
629 | container->ss->sync_metadata(container); | |
630 | } | |
631 | ||
3d2c4fc7 | 632 | rv = 0; |
6144ed44 | 633 | dirty_arrays = 0; |
1ed3f387 | 634 | for (a = *aap; a ; a = a->next) { |
6144ed44 | 635 | |
2a0bb19e | 636 | if (a->replaces && !discard_this) { |
549e9569 NB |
637 | struct active_array **ap; |
638 | for (ap = &a->next; *ap && *ap != a->replaces; | |
639 | ap = & (*ap)->next) | |
640 | ; | |
641 | if (*ap) | |
642 | *ap = (*ap)->next; | |
643 | discard_this = a->replaces; | |
644 | a->replaces = NULL; | |
6c3fb95c | 645 | /* FIXME check if device->state_fd need to be cleared?*/ |
1ed3f387 | 646 | signal_manager(); |
549e9569 | 647 | } |
ba714450 | 648 | if (a->container && !a->to_remove) { |
77b3ac8c | 649 | int ret = read_and_act(a); |
140d3685 | 650 | rv |= 1; |
77b3ac8c | 651 | dirty_arrays += !!(ret & ARRAY_DIRTY); |
140d3685 DW |
652 | /* when terminating stop manipulating the array after it |
653 | * is clean, but make sure read_and_act() is given a | |
654 | * chance to handle 'active_idle' | |
655 | */ | |
77b3ac8c | 656 | if (sigterm && !(ret & ARRAY_DIRTY)) |
140d3685 | 657 | a->container = NULL; /* stop touching this array */ |
68226a80 N |
658 | if (ret & ARRAY_BUSY) |
659 | container->retry_soon = 1; | |
6144ed44 | 660 | } |
549e9569 | 661 | } |
0af73f61 DW |
662 | |
663 | /* propagate failures across container members */ | |
1ed3f387 | 664 | for (a = *aap; a ; a = a->next) { |
ba714450 | 665 | if (!a->container || a->to_remove) |
0af73f61 DW |
666 | continue; |
667 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) | |
668 | if (mdi->curr_state & DS_FAULTY) | |
1ed3f387 | 669 | reconcile_failed(*aap, mdi); |
0af73f61 DW |
670 | } |
671 | ||
549e9569 NB |
672 | return rv; |
673 | } | |
674 | ||
675 | void do_monitor(struct supertype *container) | |
676 | { | |
677 | int rv; | |
678 | int first = 1; | |
679 | do { | |
4d43913c | 680 | rv = wait_and_act(container, first); |
549e9569 NB |
681 | first = 0; |
682 | } while (rv >= 0); | |
683 | } |