]>
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 | { | |
2b60d289 | 41 | struct stat st; |
549e9569 NB |
42 | if (fd < 0) |
43 | return; | |
2b60d289 | 44 | if (fstat(fd, &st) == -1) { |
1ade5cc1 | 45 | dprintf("Invalid fd %d\n", fd); |
2b60d289 | 46 | return; |
47 | } | |
48 | if (st.st_nlink == 0) { | |
1ade5cc1 | 49 | dprintf("fd %d was deleted\n", fd); |
2b60d289 | 50 | return; |
51 | } | |
549e9569 NB |
52 | if (fd > *maxfd) |
53 | *maxfd = fd; | |
54 | FD_SET(fd, fds); | |
55 | } | |
56 | ||
57 | static int read_attr(char *buf, int len, int fd) | |
58 | { | |
59 | int n; | |
60 | ||
61 | if (fd < 0) { | |
62 | buf[0] = 0; | |
63 | return 0; | |
64 | } | |
65 | lseek(fd, 0, 0); | |
66 | n = read(fd, buf, len - 1); | |
67 | ||
68 | if (n <= 0) { | |
69 | buf[0] = 0; | |
70 | return 0; | |
71 | } | |
72 | buf[n] = 0; | |
73 | if (buf[n-1] == '\n') | |
74 | buf[n-1] = 0; | |
75 | return n; | |
76 | } | |
77 | ||
40ae6f5f | 78 | static void read_resync_start(int fd, unsigned long long *v) |
c052ba30 DW |
79 | { |
80 | char buf[30]; | |
81 | int n; | |
82 | ||
b7941fd6 | 83 | n = read_attr(buf, 30, fd); |
40ae6f5f | 84 | if (n <= 0) { |
1ade5cc1 | 85 | dprintf("Failed to read resync_start (%d)\n", fd); |
40ae6f5f | 86 | return; |
87 | } | |
7e7fffc4 | 88 | if (strncmp(buf, "none", 4) == 0) |
40ae6f5f | 89 | *v = MaxSector; |
7e7fffc4 | 90 | else |
40ae6f5f | 91 | *v = strtoull(buf, NULL, 10); |
c052ba30 | 92 | } |
549e9569 | 93 | |
484240d8 DW |
94 | static unsigned long long read_sync_completed(int fd) |
95 | { | |
96 | unsigned long long val; | |
97 | char buf[50]; | |
98 | int n; | |
99 | char *ep; | |
100 | ||
101 | n = read_attr(buf, 50, fd); | |
102 | ||
103 | if (n <= 0) | |
104 | return 0; | |
105 | buf[n] = 0; | |
106 | val = strtoull(buf, &ep, 0); | |
107 | if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' ')) | |
108 | return 0; | |
109 | return val; | |
110 | } | |
111 | ||
549e9569 NB |
112 | static enum array_state read_state(int fd) |
113 | { | |
114 | char buf[20]; | |
115 | int n = read_attr(buf, 20, fd); | |
116 | ||
117 | if (n <= 0) | |
118 | return bad_word; | |
1770662b | 119 | return (enum array_state) sysfs_match_word(buf, array_states); |
549e9569 NB |
120 | } |
121 | ||
122 | static enum sync_action read_action( int fd) | |
123 | { | |
124 | char buf[20]; | |
125 | int n = read_attr(buf, 20, fd); | |
126 | ||
127 | if (n <= 0) | |
128 | return bad_action; | |
1770662b | 129 | return (enum sync_action) sysfs_match_word(buf, sync_actions); |
549e9569 NB |
130 | } |
131 | ||
549e9569 NB |
132 | int read_dev_state(int fd) |
133 | { | |
134 | char buf[60]; | |
135 | int n = read_attr(buf, 60, fd); | |
136 | char *cp; | |
137 | int rv = 0; | |
138 | ||
139 | if (n <= 0) | |
140 | return 0; | |
141 | ||
142 | cp = buf; | |
143 | while (cp) { | |
1770662b | 144 | if (sysfs_attr_match(cp, "faulty")) |
549e9569 | 145 | rv |= DS_FAULTY; |
1770662b | 146 | if (sysfs_attr_match(cp, "in_sync")) |
549e9569 | 147 | rv |= DS_INSYNC; |
1770662b | 148 | if (sysfs_attr_match(cp, "write_mostly")) |
549e9569 | 149 | rv |= DS_WRITE_MOSTLY; |
1770662b | 150 | if (sysfs_attr_match(cp, "spare")) |
549e9569 | 151 | rv |= DS_SPARE; |
1770662b | 152 | if (sysfs_attr_match(cp, "blocked")) |
8d45d196 | 153 | rv |= DS_BLOCKED; |
549e9569 NB |
154 | cp = strchr(cp, ','); |
155 | if (cp) | |
156 | cp++; | |
157 | } | |
158 | return rv; | |
159 | } | |
160 | ||
1ed3f387 NB |
161 | static void signal_manager(void) |
162 | { | |
4d43913c NB |
163 | /* tgkill(getpid(), mon_tid, SIGUSR1); */ |
164 | int pid = getpid(); | |
165 | syscall(SYS_tgkill, pid, mgr_tid, SIGUSR1); | |
1ed3f387 | 166 | } |
549e9569 NB |
167 | |
168 | /* Monitor a set of active md arrays - all of which share the | |
169 | * same metadata - and respond to events that require | |
170 | * metadata update. | |
171 | * | |
172 | * New arrays are detected by another thread which allocates | |
173 | * required memory and attaches the data structure to our list. | |
174 | * | |
175 | * Events: | |
176 | * Array stops. | |
177 | * This is detected by array_state going to 'clear' or 'inactive'. | |
178 | * while we thought it was active. | |
179 | * Response is to mark metadata as clean and 'clear' the array(??) | |
180 | * write-pending | |
181 | * array_state if 'write-pending' | |
182 | * We mark metadata as 'dirty' then set array to 'active'. | |
183 | * active_idle | |
184 | * Either ignore, or mark clean, then mark metadata as clean. | |
185 | * | |
186 | * device fails | |
187 | * detected by rd-N/state reporting "faulty" | |
8d45d196 DW |
188 | * mark device as 'failed' in metadata, let the kernel release the |
189 | * device by writing '-blocked' to rd/state, and finally write 'remove' to | |
0af73f61 DW |
190 | * rd/state. Before a disk can be replaced it must be failed and removed |
191 | * from all container members, this will be preemptive for the other | |
192 | * arrays... safe? | |
549e9569 NB |
193 | * |
194 | * sync completes | |
195 | * sync_action was 'resync' and becomes 'idle' and resync_start becomes | |
196 | * MaxSector | |
197 | * Notify metadata that sync is complete. | |
549e9569 NB |
198 | * |
199 | * recovery completes | |
200 | * sync_action changes from 'recover' to 'idle' | |
201 | * Check each device state and mark metadata if 'faulty' or 'in_sync'. | |
549e9569 NB |
202 | * |
203 | * deal with resync | |
c052ba30 DW |
204 | * This only happens on finding a new array... mdadm will have set |
205 | * 'resync_start' to the correct value. If 'resync_start' indicates that an | |
206 | * resync needs to occur set the array to the 'active' state rather than the | |
207 | * initial read-auto state. | |
549e9569 NB |
208 | * |
209 | * | |
210 | * | |
211 | * We wait for a change (poll/select) on array_state, sync_action, and | |
212 | * each rd-X/state file. | |
213 | * When we get any change, we check everything. So read each state file, | |
214 | * then decide what to do. | |
215 | * | |
216 | * The core action is to write new metadata to all devices in the array. | |
217 | * This is done at most once on any wakeup. | |
218 | * After that we might: | |
219 | * - update the array_state | |
220 | * - set the role of some devices. | |
221 | * - request a sync_action | |
222 | * | |
223 | */ | |
224 | ||
77b3ac8c | 225 | #define ARRAY_DIRTY 1 |
68226a80 | 226 | #define ARRAY_BUSY 2 |
549e9569 NB |
227 | static int read_and_act(struct active_array *a) |
228 | { | |
484240d8 | 229 | unsigned long long sync_completed; |
6c3fb95c | 230 | int check_degraded = 0; |
0f99b4bd | 231 | int check_reshape = 0; |
2a0bb19e | 232 | int deactivate = 0; |
549e9569 | 233 | struct mdinfo *mdi; |
77b3ac8c | 234 | int ret = 0; |
3a5d0473 | 235 | int count = 0; |
39da26ec | 236 | struct timeval tv; |
549e9569 NB |
237 | |
238 | a->next_state = bad_word; | |
239 | a->next_action = bad_action; | |
240 | ||
241 | a->curr_state = read_state(a->info.state_fd); | |
242 | a->curr_action = read_action(a->action_fd); | |
40ae6f5f | 243 | if (a->curr_state != clear) |
244 | /* | |
245 | * In "clear" state, resync_start may wrongly be set to "0" | |
246 | * when the kernel called md_clean but didn't remove the | |
247 | * sysfs attributes yet | |
248 | */ | |
249 | read_resync_start(a->resync_start_fd, &a->info.resync_start); | |
484240d8 | 250 | sync_completed = read_sync_completed(a->sync_completed_fd); |
549e9569 NB |
251 | for (mdi = a->info.devs; mdi ; mdi = mdi->next) { |
252 | mdi->next_state = 0; | |
10ce1808 | 253 | mdi->curr_state = 0; |
e1516be1 | 254 | if (mdi->state_fd >= 0) { |
40ae6f5f | 255 | read_resync_start(mdi->recovery_fd, |
256 | &mdi->recovery_start); | |
8d45d196 | 257 | mdi->curr_state = read_dev_state(mdi->state_fd); |
e1516be1 | 258 | } |
549e9569 NB |
259 | } |
260 | ||
39da26ec | 261 | gettimeofday(&tv, NULL); |
1ade5cc1 N |
262 | dprintf("(%d): %ld.%06ld state:%s prev:%s action:%s prev: %s start:%llu\n", |
263 | a->info.container_member, | |
39da26ec MW |
264 | tv.tv_sec, tv.tv_usec, |
265 | array_states[a->curr_state], | |
266 | array_states[a->prev_state], | |
267 | sync_actions[a->curr_action], | |
268 | sync_actions[a->prev_action], | |
269 | a->info.resync_start | |
270 | ); | |
271 | ||
6b374ba3 | 272 | if ((a->curr_state == bad_word || a->curr_state <= inactive) && |
549e9569 NB |
273 | a->prev_state > inactive) { |
274 | /* array has been stopped */ | |
ed9d66aa | 275 | a->container->ss->set_array_state(a, 1); |
549e9569 | 276 | a->next_state = clear; |
2a0bb19e | 277 | deactivate = 1; |
549e9569 NB |
278 | } |
279 | if (a->curr_state == write_pending) { | |
ed9d66aa | 280 | a->container->ss->set_array_state(a, 0); |
549e9569 | 281 | a->next_state = active; |
77b3ac8c | 282 | ret |= ARRAY_DIRTY; |
549e9569 NB |
283 | } |
284 | if (a->curr_state == active_idle) { | |
d797a062 DW |
285 | /* Set array to 'clean' FIRST, then mark clean |
286 | * in the metadata | |
549e9569 | 287 | */ |
d797a062 | 288 | a->next_state = clean; |
77b3ac8c | 289 | ret |= ARRAY_DIRTY; |
d797a062 DW |
290 | } |
291 | if (a->curr_state == clean) { | |
d797a062 | 292 | a->container->ss->set_array_state(a, 1); |
549e9569 | 293 | } |
140d3685 | 294 | if (a->curr_state == active || |
6b374ba3 | 295 | a->curr_state == suspended) |
77b3ac8c | 296 | ret |= ARRAY_DIRTY; |
549e9569 | 297 | if (a->curr_state == readonly) { |
e9dd1598 N |
298 | /* Well, I'm ready to handle things. If readonly |
299 | * wasn't requested, transition to read-auto. | |
549e9569 | 300 | */ |
e9dd1598 N |
301 | char buf[64]; |
302 | read_attr(buf, sizeof(buf), a->metadata_fd); | |
303 | if (strncmp(buf, "external:-", 10) == 0) { | |
304 | /* explicit request for readonly array. Leave it alone */ | |
305 | ; | |
306 | } else { | |
e9dd1598 N |
307 | if (a->container->ss->set_array_state(a, 2)) |
308 | a->next_state = read_auto; /* array is clean */ | |
140d3685 | 309 | else { |
e9dd1598 | 310 | a->next_state = active; /* Now active for recovery etc */ |
77b3ac8c | 311 | ret |= ARRAY_DIRTY; |
140d3685 | 312 | } |
e9dd1598 | 313 | } |
549e9569 NB |
314 | } |
315 | ||
00e02142 DW |
316 | if (!deactivate && |
317 | a->curr_action == idle && | |
549e9569 | 318 | a->prev_action == resync) { |
4e5528c6 NB |
319 | /* A resync has finished. The endpoint is recorded in |
320 | * 'sync_start'. We don't update the metadata | |
321 | * until the array goes inactive or readonly though. | |
322 | * Just check if we need to fiddle spares. | |
323 | */ | |
0c0c44db | 324 | a->container->ss->set_array_state(a, a->curr_state <= clean); |
549e9569 NB |
325 | check_degraded = 1; |
326 | } | |
327 | ||
00e02142 DW |
328 | if (!deactivate && |
329 | a->curr_action == idle && | |
549e9569 | 330 | a->prev_action == recover) { |
0a6bdbee DW |
331 | /* A recovery has finished. Some disks may be in sync now, |
332 | * and the array may no longer be degraded | |
333 | */ | |
549e9569 | 334 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) { |
8d45d196 DW |
335 | a->container->ss->set_disk(a, mdi->disk.raid_disk, |
336 | mdi->curr_state); | |
549e9569 NB |
337 | if (! (mdi->curr_state & DS_INSYNC)) |
338 | check_degraded = 1; | |
3a5d0473 | 339 | count++; |
549e9569 | 340 | } |
3a5d0473 KW |
341 | if (count != a->info.array.raid_disks) |
342 | check_degraded = 1; | |
549e9569 NB |
343 | } |
344 | ||
0f99b4bd N |
345 | if (!deactivate && |
346 | a->curr_action == reshape && | |
347 | a->prev_action != reshape) | |
348 | /* reshape was requested by mdadm. Need to see if | |
349 | * new devices have been added. Manager does that | |
350 | * when it sees check_reshape | |
351 | */ | |
352 | check_reshape = 1; | |
353 | ||
92967543 DW |
354 | /* Check for failures and if found: |
355 | * 1/ Record the failure in the metadata and unblock the device. | |
356 | * FIXME update the kernel to stop notifying on failed drives when | |
357 | * the array is readonly and we have cleared 'blocked' | |
358 | * 2/ Try to remove the device if the array is writable, or can be | |
359 | * made writable. | |
360 | */ | |
549e9569 NB |
361 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) { |
362 | if (mdi->curr_state & DS_FAULTY) { | |
8d45d196 DW |
363 | a->container->ss->set_disk(a, mdi->disk.raid_disk, |
364 | mdi->curr_state); | |
549e9569 | 365 | check_degraded = 1; |
1c278e81 N |
366 | if (mdi->curr_state & DS_BLOCKED) |
367 | mdi->next_state |= DS_UNBLOCK; | |
92967543 DW |
368 | if (a->curr_state == read_auto) { |
369 | a->container->ss->set_array_state(a, 0); | |
370 | a->next_state = active; | |
371 | } | |
372 | if (a->curr_state > readonly) | |
373 | mdi->next_state |= DS_REMOVE; | |
549e9569 NB |
374 | } |
375 | } | |
376 | ||
484240d8 DW |
377 | /* Check for recovery checkpoint notifications. We need to be a |
378 | * minimum distance away from the last checkpoint to prevent | |
aad6f216 N |
379 | * over checkpointing. Note reshape checkpointing is handled |
380 | * in the second branch. | |
484240d8 DW |
381 | */ |
382 | if (sync_completed > a->last_checkpoint && | |
383 | sync_completed - a->last_checkpoint > a->info.component_size >> 4 && | |
4f0a7acc DW |
384 | a->curr_action > reshape) { |
385 | /* A (non-reshape) sync_action has reached a checkpoint. | |
386 | * Record the updated position in the metadata | |
387 | */ | |
388 | a->last_checkpoint = sync_completed; | |
389 | a->container->ss->set_array_state(a, a->curr_state <= clean); | |
aad6f216 N |
390 | } else if ((a->curr_action == idle && a->prev_action == reshape) || |
391 | (a->curr_action == reshape | |
392 | && sync_completed > a->last_checkpoint) ) { | |
393 | /* Reshape has progressed or completed so we need to | |
394 | * update the array state - and possibly the array size | |
395 | */ | |
2a9f8409 AK |
396 | if (sync_completed != 0) |
397 | a->last_checkpoint = sync_completed; | |
6d4225a1 AK |
398 | /* We might need to update last_checkpoint depending on |
399 | * the reason that reshape finished. | |
400 | * if array reshape is really finished: | |
401 | * set check point to the end, this allows | |
402 | * set_array_state() to finalize reshape in metadata | |
403 | * if reshape if broken: do not set checkpoint to the end | |
404 | * this allows for reshape restart from checkpoint | |
405 | */ | |
406 | if ((a->curr_action != reshape) && | |
407 | (a->prev_action == reshape)) { | |
408 | char buf[40]; | |
409 | if ((sysfs_get_str(&a->info, NULL, | |
410 | "reshape_position", | |
411 | buf, | |
412 | sizeof(buf)) >= 0) && | |
413 | strncmp(buf, "none", 4) == 0) | |
414 | a->last_checkpoint = a->info.component_size; | |
415 | } | |
aad6f216 | 416 | a->container->ss->set_array_state(a, a->curr_state <= clean); |
2a9f8409 | 417 | a->last_checkpoint = sync_completed; |
aad6f216 N |
418 | } |
419 | ||
420 | if (sync_completed > a->last_checkpoint) | |
484240d8 | 421 | a->last_checkpoint = sync_completed; |
484240d8 | 422 | |
a4921f30 | 423 | a->container->ss->sync_metadata(a->container); |
1ade5cc1 | 424 | dprintf("(%d): state:%s action:%s next(", a->info.container_member, |
4065aa81 | 425 | array_states[a->curr_state], sync_actions[a->curr_action]); |
549e9569 NB |
426 | |
427 | /* Effect state changes in the array */ | |
4e6e574a | 428 | if (a->next_state != bad_word) { |
1ade5cc1 | 429 | dprintf_cont(" state:%s", array_states[a->next_state]); |
549e9569 | 430 | write_attr(array_states[a->next_state], a->info.state_fd); |
4e6e574a DW |
431 | } |
432 | if (a->next_action != bad_action) { | |
549e9569 | 433 | write_attr(sync_actions[a->next_action], a->action_fd); |
1ade5cc1 | 434 | dprintf_cont(" action:%s", sync_actions[a->next_action]); |
4e6e574a | 435 | } |
549e9569 | 436 | for (mdi = a->info.devs; mdi ; mdi = mdi->next) { |
92967543 | 437 | if (mdi->next_state & DS_UNBLOCK) { |
1ade5cc1 | 438 | dprintf_cont(" %d:-blocked", mdi->disk.raid_disk); |
92967543 DW |
439 | write_attr("-blocked", mdi->state_fd); |
440 | } | |
441 | ||
442 | if ((mdi->next_state & DS_REMOVE) && mdi->state_fd >= 0) { | |
57632f4a | 443 | int remove_result; |
8d45d196 | 444 | |
68226a80 N |
445 | /* The kernel may not be able to immediately remove the |
446 | * disk. In that case we wait a little while and | |
447 | * try again. | |
8d45d196 | 448 | */ |
57632f4a NB |
449 | remove_result = write_attr("remove", mdi->state_fd); |
450 | if (remove_result > 0) { | |
1ade5cc1 | 451 | dprintf_cont(" %d:removed", mdi->disk.raid_disk); |
8d45d196 | 452 | close(mdi->state_fd); |
e40512fd | 453 | close(mdi->recovery_fd); |
8d45d196 | 454 | mdi->state_fd = -1; |
68226a80 N |
455 | } else |
456 | ret |= ARRAY_BUSY; | |
8d45d196 | 457 | } |
4e6e574a | 458 | if (mdi->next_state & DS_INSYNC) { |
549e9569 | 459 | write_attr("+in_sync", mdi->state_fd); |
1ade5cc1 | 460 | dprintf_cont(" %d:+in_sync", mdi->disk.raid_disk); |
4e6e574a | 461 | } |
549e9569 | 462 | } |
1ade5cc1 | 463 | dprintf_cont(" )\n"); |
549e9569 NB |
464 | |
465 | /* move curr_ to prev_ */ | |
466 | a->prev_state = a->curr_state; | |
467 | ||
468 | a->prev_action = a->curr_action; | |
469 | ||
470 | for (mdi = a->info.devs; mdi ; mdi = mdi->next) { | |
471 | mdi->prev_state = mdi->curr_state; | |
472 | mdi->next_state = 0; | |
473 | } | |
474 | ||
0f99b4bd | 475 | if (check_degraded || check_reshape) { |
7e1432fb | 476 | /* manager will do the actual check */ |
0f99b4bd N |
477 | if (check_degraded) |
478 | a->check_degraded = 1; | |
479 | if (check_reshape) | |
480 | a->check_reshape = 1; | |
7e1432fb NB |
481 | signal_manager(); |
482 | } | |
483 | ||
2a0bb19e DW |
484 | if (deactivate) |
485 | a->container = NULL; | |
486 | ||
77b3ac8c | 487 | return ret; |
549e9569 NB |
488 | } |
489 | ||
0af73f61 DW |
490 | static struct mdinfo * |
491 | find_device(struct active_array *a, int major, int minor) | |
492 | { | |
493 | struct mdinfo *mdi; | |
494 | ||
495 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) | |
496 | if (mdi->disk.major == major && mdi->disk.minor == minor) | |
497 | return mdi; | |
498 | ||
499 | return NULL; | |
500 | } | |
501 | ||
502 | static void reconcile_failed(struct active_array *aa, struct mdinfo *failed) | |
503 | { | |
504 | struct active_array *a; | |
505 | struct mdinfo *victim; | |
506 | ||
507 | for (a = aa; a; a = a->next) { | |
ba714450 | 508 | if (!a->container || a->to_remove) |
0af73f61 DW |
509 | continue; |
510 | victim = find_device(a, failed->disk.major, failed->disk.minor); | |
511 | if (!victim) | |
512 | continue; | |
513 | ||
514 | if (!(victim->curr_state & DS_FAULTY)) | |
515 | write_attr("faulty", victim->state_fd); | |
516 | } | |
517 | } | |
518 | ||
4e6e574a DW |
519 | #ifdef DEBUG |
520 | static void dprint_wake_reasons(fd_set *fds) | |
521 | { | |
522 | int i; | |
523 | char proc_path[256]; | |
524 | char link[256]; | |
525 | char *basename; | |
526 | int rv; | |
527 | ||
528 | fprintf(stderr, "monitor: wake ( "); | |
529 | for (i = 0; i < FD_SETSIZE; i++) { | |
530 | if (FD_ISSET(i, fds)) { | |
531 | sprintf(proc_path, "/proc/%d/fd/%d", | |
532 | (int) getpid(), i); | |
533 | ||
534 | rv = readlink(proc_path, link, sizeof(link) - 1); | |
535 | if (rv < 0) { | |
536 | fprintf(stderr, "%d:unknown ", i); | |
537 | continue; | |
538 | } | |
539 | link[rv] = '\0'; | |
540 | basename = strrchr(link, '/'); | |
541 | fprintf(stderr, "%d:%s ", | |
542 | i, basename ? ++basename : link); | |
543 | } | |
544 | } | |
545 | fprintf(stderr, ")\n"); | |
546 | } | |
547 | #endif | |
548 | ||
1eb252b8 N |
549 | int monitor_loop_cnt; |
550 | ||
4d43913c | 551 | static int wait_and_act(struct supertype *container, int nowait) |
549e9569 NB |
552 | { |
553 | fd_set rfds; | |
554 | int maxfd = 0; | |
e0d6609f | 555 | struct active_array **aap = &container->arrays; |
1ed3f387 | 556 | struct active_array *a, **ap; |
549e9569 | 557 | int rv; |
0af73f61 | 558 | struct mdinfo *mdi; |
6144ed44 | 559 | static unsigned int dirty_arrays = ~0; /* start at some non-zero value */ |
549e9569 NB |
560 | |
561 | FD_ZERO(&rfds); | |
562 | ||
1ed3f387 NB |
563 | for (ap = aap ; *ap ;) { |
564 | a = *ap; | |
565 | /* once an array has been deactivated we want to | |
566 | * ask the manager to discard it. | |
2a0bb19e | 567 | */ |
ba714450 | 568 | if (!a->container || a->to_remove) { |
1ed3f387 NB |
569 | if (discard_this) { |
570 | ap = &(*ap)->next; | |
571 | continue; | |
572 | } | |
573 | *ap = a->next; | |
574 | a->next = NULL; | |
575 | discard_this = a; | |
576 | signal_manager(); | |
2a0bb19e | 577 | continue; |
1ed3f387 | 578 | } |
2a0bb19e | 579 | |
549e9569 NB |
580 | add_fd(&rfds, &maxfd, a->info.state_fd); |
581 | add_fd(&rfds, &maxfd, a->action_fd); | |
484240d8 | 582 | add_fd(&rfds, &maxfd, a->sync_completed_fd); |
549e9569 NB |
583 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) |
584 | add_fd(&rfds, &maxfd, mdi->state_fd); | |
1ed3f387 NB |
585 | |
586 | ap = &(*ap)->next; | |
549e9569 NB |
587 | } |
588 | ||
6144ed44 DW |
589 | if (manager_ready && (*aap == NULL || (sigterm && !dirty_arrays))) { |
590 | /* No interesting arrays, or we have been told to | |
591 | * terminate and everything is clean. Lets see about | |
592 | * exiting. Note that blocking at this point is not a | |
593 | * problem as there are no active arrays, there is | |
594 | * nothing that we need to be ready to do. | |
e0d6609f | 595 | */ |
d998b738 N |
596 | int fd; |
597 | if (sigterm) | |
4dd2df09 | 598 | fd = open_dev_excl(container->devnm); |
d998b738 | 599 | else |
4dd2df09 | 600 | fd = open_dev_flags(container->devnm, O_RDONLY|O_EXCL); |
e0d6609f NB |
601 | if (fd >= 0 || errno != EBUSY) { |
602 | /* OK, we are safe to leave */ | |
6144ed44 DW |
603 | if (sigterm && !dirty_arrays) |
604 | dprintf("caught sigterm, all clean... exiting\n"); | |
605 | else | |
606 | dprintf("no arrays to monitor... exiting\n"); | |
fa716c83 N |
607 | if (!sigterm) |
608 | /* On SIGTERM, someone (the take-over mdmon) will | |
609 | * clean up | |
610 | */ | |
4dd2df09 | 611 | remove_pidfile(container->devnm); |
e0d6609f NB |
612 | exit_now = 1; |
613 | signal_manager(); | |
6f4cdfd9 | 614 | close(fd); |
e0d6609f NB |
615 | exit(0); |
616 | } | |
617 | } | |
618 | ||
549e9569 | 619 | if (!nowait) { |
4d43913c | 620 | sigset_t set; |
d998b738 N |
621 | struct timespec ts; |
622 | ts.tv_sec = 24*3600; | |
623 | ts.tv_nsec = 0; | |
68226a80 | 624 | if (*aap == NULL || container->retry_soon) { |
d998b738 N |
625 | /* just waiting to get O_EXCL access */ |
626 | ts.tv_sec = 0; | |
627 | ts.tv_nsec = 20000000ULL; | |
628 | } | |
4d43913c NB |
629 | sigprocmask(SIG_UNBLOCK, NULL, &set); |
630 | sigdelset(&set, SIGUSR1); | |
1eb252b8 | 631 | monitor_loop_cnt |= 1; |
d998b738 | 632 | rv = pselect(maxfd+1, NULL, NULL, &rfds, &ts, &set); |
1eb252b8 | 633 | monitor_loop_cnt += 1; |
c3719360 MW |
634 | if (rv == -1) { |
635 | if (errno == EINTR) { | |
636 | rv = 0; | |
637 | dprintf("monitor: caught signal\n"); | |
638 | } else | |
639 | dprintf("monitor: error %d in pselect\n", | |
640 | errno); | |
641 | } | |
4e6e574a | 642 | #ifdef DEBUG |
c3719360 MW |
643 | else |
644 | dprint_wake_reasons(&rfds); | |
4e6e574a | 645 | #endif |
68226a80 | 646 | container->retry_soon = 0; |
549e9569 NB |
647 | } |
648 | ||
2e735d19 NB |
649 | if (update_queue) { |
650 | struct metadata_update *this; | |
651 | ||
652 | for (this = update_queue; this ; this = this->next) | |
653 | container->ss->process_update(container, this); | |
654 | ||
655 | update_queue_handled = update_queue; | |
656 | update_queue = NULL; | |
657 | signal_manager(); | |
658 | container->ss->sync_metadata(container); | |
659 | } | |
660 | ||
3d2c4fc7 | 661 | rv = 0; |
6144ed44 | 662 | dirty_arrays = 0; |
1ed3f387 | 663 | for (a = *aap; a ; a = a->next) { |
6144ed44 | 664 | |
2a0bb19e | 665 | if (a->replaces && !discard_this) { |
549e9569 NB |
666 | struct active_array **ap; |
667 | for (ap = &a->next; *ap && *ap != a->replaces; | |
668 | ap = & (*ap)->next) | |
669 | ; | |
670 | if (*ap) | |
671 | *ap = (*ap)->next; | |
672 | discard_this = a->replaces; | |
673 | a->replaces = NULL; | |
6c3fb95c | 674 | /* FIXME check if device->state_fd need to be cleared?*/ |
1ed3f387 | 675 | signal_manager(); |
549e9569 | 676 | } |
ba714450 | 677 | if (a->container && !a->to_remove) { |
77b3ac8c | 678 | int ret = read_and_act(a); |
140d3685 | 679 | rv |= 1; |
77b3ac8c | 680 | dirty_arrays += !!(ret & ARRAY_DIRTY); |
140d3685 DW |
681 | /* when terminating stop manipulating the array after it |
682 | * is clean, but make sure read_and_act() is given a | |
683 | * chance to handle 'active_idle' | |
684 | */ | |
77b3ac8c | 685 | if (sigterm && !(ret & ARRAY_DIRTY)) |
140d3685 | 686 | a->container = NULL; /* stop touching this array */ |
68226a80 N |
687 | if (ret & ARRAY_BUSY) |
688 | container->retry_soon = 1; | |
6144ed44 | 689 | } |
549e9569 | 690 | } |
0af73f61 DW |
691 | |
692 | /* propagate failures across container members */ | |
1ed3f387 | 693 | for (a = *aap; a ; a = a->next) { |
ba714450 | 694 | if (!a->container || a->to_remove) |
0af73f61 DW |
695 | continue; |
696 | for (mdi = a->info.devs ; mdi ; mdi = mdi->next) | |
697 | if (mdi->curr_state & DS_FAULTY) | |
1ed3f387 | 698 | reconcile_failed(*aap, mdi); |
0af73f61 DW |
699 | } |
700 | ||
549e9569 NB |
701 | return rv; |
702 | } | |
703 | ||
704 | void do_monitor(struct supertype *container) | |
705 | { | |
706 | int rv; | |
707 | int first = 1; | |
708 | do { | |
4d43913c | 709 | rv = wait_and_act(container, first); |
549e9569 NB |
710 | first = 0; |
711 | } while (rv >= 0); | |
712 | } |