]> git.ipfire.org Git - thirdparty/mdadm.git/blame - managemon.c
Check all member devices in enough_fd
[thirdparty/mdadm.git] / managemon.c
CommitLineData
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/*
22 * The management thread for monitoring active md arrays.
23 * This thread does things which might block such as memory
24 * allocation.
25 * In particular:
26 *
27 * - Find out about new arrays in this container.
28 * Allocate the data structures and open the files.
29 *
30 * For this we watch /proc/mdstat and find new arrays with
31 * metadata type that confirms sharing. e.g. "md4"
32 * When we find a new array we slip it into the list of
33 * arrays and signal 'monitor' by writing to a pipe.
34 *
35 * - Respond to reshape requests by allocating new data structures
36 * and opening new files.
37 *
38 * These come as a change to raid_disks. We allocate a new
39 * version of the data structures and slip it into the list.
40 * 'monitor' will notice and release the old version.
41 * Changes to level, chunksize, layout.. do not need re-allocation.
42 * Reductions in raid_disks don't really either, but we handle
43 * them the same way for consistency.
44 *
45 * - When a device is added to the container, we add it to the metadata
46 * as a spare.
47 *
6c3fb95c
NB
48 * - Deal with degraded array
49 * We only do this when first noticing the array is degraded.
50 * This can be when we first see the array, when sync completes or
51 * when recovery completes.
52 *
53 * Check if number of failed devices suggests recovery is needed, and
54 * skip if not.
55 * Ask metadata to allocate a spare device
56 * Add device as not in_sync and give a role
57 * Update metadata.
58 * Open sysfs files and pass to monitor.
59 * Make sure that monitor Starts recovery....
549e9569
NB
60 *
61 * - Pass on metadata updates from external programs such as
62 * mdadm creating a new array.
63 *
64 * This is most-messy.
65 * It might involve adding a new array or changing the status of
66 * a spare, or any reconfig that the kernel doesn't get involved in.
67 *
68 * The required updates are received via a named pipe. There will
69 * be one named pipe for each container. Each message contains a
70 * sync marker: 0x5a5aa5a5, A byte count, and the message. This is
71 * passed to the metadata handler which will interpret and process it.
72 * For 'DDF' messages are internal data blocks with the leading
73 * 'magic number' signifying what sort of data it is.
74 *
75 */
76
77/*
78 * We select on /proc/mdstat and the named pipe.
79 * We create new arrays or updated version of arrays and slip
80 * them into the head of the list, then signal 'monitor' via a pipe write.
81 * 'monitor' will notice and place the old array on a return list.
82 * Metadata updates are placed on a queue just like they arrive
83 * from the named pipe.
84 *
85 * When new arrays are found based on correct metadata string, we
86 * need to identify them with an entry in the metadata. Maybe we require
87 * the metadata to be mdX/NN when NN is the index into an appropriate table.
88 *
89 */
90
91/*
92 * List of tasks:
93 * - Watch for spares to be added to the container, and write updated
94 * metadata to them.
95 * - Watch for new arrays using this container, confirm they match metadata
96 * and if so, start monitoring them
97 * - Watch for spares being added to monitored arrays. This shouldn't
98 * happen, as we should do all the adding. Just remove them.
99 * - Watch for change in raid-disks, chunk-size, etc. Update metadata and
100 * start a reshape.
101 */
102#ifndef _GNU_SOURCE
103#define _GNU_SOURCE
104#endif
105#include "mdadm.h"
106#include "mdmon.h"
4d43913c 107#include <sys/syscall.h>
549e9569 108#include <sys/socket.h>
1ed3f387 109#include <signal.h>
549e9569 110
2a0bb19e
DW
111static void close_aa(struct active_array *aa)
112{
113 struct mdinfo *d;
114
e1516be1
DW
115 for (d = aa->info.devs; d; d = d->next) {
116 close(d->recovery_fd);
2a0bb19e 117 close(d->state_fd);
e1516be1 118 }
2a0bb19e
DW
119
120 close(aa->action_fd);
121 close(aa->info.state_fd);
122 close(aa->resync_start_fd);
a9d868c3
AK
123 close(aa->metadata_fd);
124 close(aa->sync_completed_fd);
2a0bb19e
DW
125}
126
549e9569
NB
127static void free_aa(struct active_array *aa)
128{
2a0bb19e
DW
129 /* Note that this doesn't close fds if they are being used
130 * by a clone. ->container will be set for a clone
549e9569 131 */
4e6e574a 132 dprintf("%s: devnum: %d\n", __func__, aa->devnum);
2a0bb19e
DW
133 if (!aa->container)
134 close_aa(aa);
549e9569
NB
135 while (aa->info.devs) {
136 struct mdinfo *d = aa->info.devs;
137 aa->info.devs = d->next;
138 free(d);
139 }
140 free(aa);
141}
142
6c3fb95c
NB
143static struct active_array *duplicate_aa(struct active_array *aa)
144{
145 struct active_array *newa = malloc(sizeof(*newa));
146 struct mdinfo **dp1, **dp2;
147
148 *newa = *aa;
149 newa->next = NULL;
150 newa->replaces = NULL;
151 newa->info.next = NULL;
152
153 dp2 = &newa->info.devs;
154
155 for (dp1 = &aa->info.devs; *dp1; dp1 = &(*dp1)->next) {
156 struct mdinfo *d;
157 if ((*dp1)->state_fd < 0)
158 continue;
159
160 d = malloc(sizeof(*d));
161 *d = **dp1;
162 *dp2 = d;
163 dp2 = & d->next;
164 }
7e1432fb 165 *dp2 = NULL;
6c3fb95c
NB
166
167 return newa;
168}
169
4d43913c 170static void wakeup_monitor(void)
2a0bb19e 171{
4d43913c
NB
172 /* tgkill(getpid(), mon_tid, SIGUSR1); */
173 int pid = getpid();
174 syscall(SYS_tgkill, pid, mon_tid, SIGUSR1);
2a0bb19e
DW
175}
176
1ed3f387
NB
177static void remove_old(void)
178{
179 if (discard_this) {
180 discard_this->next = NULL;
181 free_aa(discard_this);
182 if (pending_discard == discard_this)
183 pending_discard = NULL;
184 discard_this = NULL;
48561b01 185 wakeup_monitor();
1ed3f387
NB
186 }
187}
188
549e9569
NB
189static void replace_array(struct supertype *container,
190 struct active_array *old,
191 struct active_array *new)
192{
193 /* To replace an array, we add it to the top of the list
194 * marked with ->replaces to point to the original.
195 * 'monitor' will take the original out of the list
196 * and put it on 'discard_this'. We take it from there
197 * and discard it.
198 */
1ed3f387 199 remove_old();
549e9569
NB
200 while (pending_discard) {
201 while (discard_this == NULL)
202 sleep(1);
1ed3f387 203 remove_old();
549e9569
NB
204 }
205 pending_discard = old;
206 new->replaces = old;
207 new->next = container->arrays;
208 container->arrays = new;
4d43913c 209 wakeup_monitor();
549e9569
NB
210}
211
2e735d19
NB
212struct metadata_update *update_queue = NULL;
213struct metadata_update *update_queue_handled = NULL;
214struct metadata_update *update_queue_pending = NULL;
215
071cfc42 216static void free_updates(struct metadata_update **update)
2e735d19 217{
071cfc42
DW
218 while (*update) {
219 struct metadata_update *this = *update;
cb23f1f4 220 void **space_list = this->space_list;
071cfc42
DW
221
222 *update = this->next;
904c1ef7 223 free(this->buf);
071cfc42 224 free(this->space);
cb23f1f4
N
225 while (space_list) {
226 void *space = space_list;
227 space_list = *space_list;
228 free(space);
229 }
2e735d19
NB
230 free(this);
231 }
071cfc42
DW
232}
233
234void check_update_queue(struct supertype *container)
235{
236 free_updates(&update_queue_handled);
237
2e735d19
NB
238 if (update_queue == NULL &&
239 update_queue_pending) {
240 update_queue = update_queue_pending;
241 update_queue_pending = NULL;
4d43913c 242 wakeup_monitor();
2e735d19
NB
243 }
244}
245
6c3fb95c 246static void queue_metadata_update(struct metadata_update *mu)
2e735d19
NB
247{
248 struct metadata_update **qp;
249
250 qp = &update_queue_pending;
251 while (*qp)
252 qp = & ((*qp)->next);
253 *qp = mu;
254}
255
43dad3d6
DW
256static void add_disk_to_container(struct supertype *st, struct mdinfo *sd)
257{
258 int dfd;
259 char nm[20];
661dce36 260 struct supertype *st2;
43dad3d6 261 struct metadata_update *update = NULL;
661dce36 262 struct mdinfo info;
43dad3d6
DW
263 mdu_disk_info_t dk = {
264 .number = -1,
265 .major = sd->disk.major,
266 .minor = sd->disk.minor,
267 .raid_disk = -1,
268 .state = 0,
269 };
270
271 dprintf("%s: add %d:%d to container\n",
272 __func__, sd->disk.major, sd->disk.minor);
273
04a8ac08
DW
274 sd->next = st->devs;
275 st->devs = sd;
276
43dad3d6
DW
277 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
278 dfd = dev_open(nm, O_RDWR);
279 if (dfd < 0)
280 return;
281
661dce36
N
282 /* Check the metadata and see if it is already part of this
283 * array
284 */
285 st2 = dup_super(st);
286 if (st2->ss->load_super(st2, dfd, NULL) == 0) {
a5d85af7 287 st2->ss->getinfo_super(st, &info, NULL);
661dce36
N
288 if (st->ss->compare_super(st, st2) == 0 &&
289 info.disk.raid_disk >= 0) {
290 /* Looks like a good member of array.
291 * Just accept it.
292 * mdadm will incorporate any parts into
293 * active arrays.
294 */
295 st2->ss->free_super(st2);
296 return;
297 }
298 }
299 st2->ss->free_super(st2);
300
43dad3d6
DW
301 st->update_tail = &update;
302 st->ss->add_to_super(st, &dk, dfd, NULL);
303 st->ss->write_init_super(st);
304 queue_metadata_update(update);
305 st->update_tail = NULL;
306}
307
1a64be56
LM
308/*
309 * Create and queue update structure about the removed disks.
310 * The update is prepared by super type handler and passed to the monitor
311 * thread.
312 */
313static void remove_disk_from_container(struct supertype *st, struct mdinfo *sd)
314{
315 struct metadata_update *update = NULL;
316 mdu_disk_info_t dk = {
317 .number = -1,
318 .major = sd->disk.major,
319 .minor = sd->disk.minor,
320 .raid_disk = -1,
321 .state = 0,
322 };
1a64be56
LM
323 dprintf("%s: remove %d:%d from container\n",
324 __func__, sd->disk.major, sd->disk.minor);
325
326 st->update_tail = &update;
327 st->ss->remove_from_super(st, &dk);
4dd968cc
N
328 /* FIXME this write_init_super shouldn't be here.
329 * We have it after add_to_super to write to new device,
330 * but with 'remove' we don't ant to write to that device!
331 */
1a64be56
LM
332 st->ss->write_init_super(st);
333 queue_metadata_update(update);
334 st->update_tail = NULL;
335}
336
549e9569
NB
337static void manage_container(struct mdstat_ent *mdstat,
338 struct supertype *container)
339{
1a64be56
LM
340 /* Of interest here are:
341 * - if a new device has been added to the container, we
342 * add it to the array ignoring any metadata on it.
343 * - if a device has been removed from the container, we
344 * remove it from the device list and update the metadata.
549e9569
NB
345 * FIXME should we look for compatible metadata and take hints
346 * about spare assignment.... probably not.
549e9569
NB
347 */
348 if (mdstat->devcnt != container->devcnt) {
7bc1962f
DW
349 struct mdinfo **cdp, *cd, *di, *mdi;
350 int found;
351
549e9569
NB
352 /* read /sys/block/NAME/md/dev-??/block/dev to find out
353 * what is there, and compare with container->info.devs
354 * To see what is removed and what is added.
355 * These need to be remove from, or added to, the array
356 */
b526e52d 357 mdi = sysfs_read(-1, mdstat->devnum, GET_DEVS);
313a4a82
DW
358 if (!mdi) {
359 /* invalidate the current count so we can try again */
360 container->devcnt = -1;
7bc1962f 361 return;
313a4a82 362 }
7bc1962f
DW
363
364 /* check for removals */
365 for (cdp = &container->devs; *cdp; ) {
366 found = 0;
367 for (di = mdi->devs; di; di = di->next)
368 if (di->disk.major == (*cdp)->disk.major &&
369 di->disk.minor == (*cdp)->disk.minor) {
370 found = 1;
371 break;
372 }
373 if (!found) {
374 cd = *cdp;
375 *cdp = (*cdp)->next;
1a64be56 376 remove_disk_from_container(container, cd);
7bc1962f
DW
377 free(cd);
378 } else
379 cdp = &(*cdp)->next;
380 }
43dad3d6
DW
381
382 /* check for additions */
383 for (di = mdi->devs; di; di = di->next) {
384 for (cd = container->devs; cd; cd = cd->next)
385 if (di->disk.major == cd->disk.major &&
386 di->disk.minor == cd->disk.minor)
387 break;
04a8ac08
DW
388 if (!cd) {
389 struct mdinfo *newd = malloc(sizeof(*newd));
390
391 if (!newd) {
392 container->devcnt = -1;
393 continue;
394 }
395 *newd = *di;
396 add_disk_to_container(container, newd);
397 }
43dad3d6 398 }
7bc1962f 399 sysfs_free(mdi);
549e9569
NB
400 container->devcnt = mdstat->devcnt;
401 }
402}
403
63b4aae3
DW
404static int disk_init_and_add(struct mdinfo *disk, struct mdinfo *clone,
405 struct active_array *aa)
406{
407 if (!disk || !clone)
408 return -1;
409
410 *disk = *clone;
411 disk->recovery_fd = sysfs_open(aa->devnum, disk->sys_name, "recovery_start");
412 disk->state_fd = sysfs_open(aa->devnum, disk->sys_name, "state");
413 disk->prev_state = read_dev_state(disk->state_fd);
414 disk->curr_state = disk->prev_state;
415 disk->next = aa->info.devs;
416 aa->info.devs = disk;
417
418 return 0;
419}
420
549e9569
NB
421static void manage_member(struct mdstat_ent *mdstat,
422 struct active_array *a)
423{
424 /* Compare mdstat info with known state of member array.
425 * We do not need to look for device state changes here, that
426 * is dealt with by the monitor.
427 *
0f99b4bd
N
428 * If a reshape is being requested, monitor will have noticed
429 * that sync_action changed and will have set check_reshape.
430 * We just need to see if new devices have appeared. All metadata
431 * updates will already have been processed.
6c3fb95c 432 *
0f99b4bd 433 * We also want to handle degraded arrays here by
6c3fb95c
NB
434 * trying to find and assign a spare.
435 * We do that whenever the monitor tells us too.
549e9569 436 */
bc77ed53
DW
437 char buf[64];
438 int frozen;
4e2c1a9a
N
439 struct supertype *container = a->container;
440
441 if (container == NULL)
442 /* Raced with something */
443 return;
bc77ed53 444
549e9569
NB
445 // FIXME
446 a->info.array.raid_disks = mdstat->raid_disks;
549e9569
NB
447 // MORE
448
bc77ed53
DW
449 /* honor 'frozen' */
450 if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0)
451 frozen = buf[9] == '-';
452 else
453 frozen = 1; /* can't read metadata_version assume the worst */
454
f54a6742
N
455 /* If sync_action is not 'idle' then don't try recovery now */
456 if (!frozen
457 && sysfs_get_str(&a->info, NULL, "sync_action", buf, sizeof(buf)) > 0
458 && strncmp(buf, "idle", 4) != 0)
459 frozen = 1;
460
57f8c769
AK
461 if (mdstat->level) {
462 int level = map_name(pers, mdstat->level);
7023e0b8
N
463 if (level == 0 || level == LEVEL_LINEAR) {
464 a->container = NULL;
84f3857f 465 wakeup_monitor();
7023e0b8
N
466 return;
467 }
468 else if (a->info.array.level != level && level > 0) {
57f8c769
AK
469 struct active_array *newa = duplicate_aa(a);
470 if (newa) {
471 newa->info.array.level = level;
4e2c1a9a 472 replace_array(container, a, newa);
57f8c769
AK
473 a = newa;
474 }
475 }
476 }
477
0c4f6e37 478 /* We don't check the array while any update is pending, as it
88b496c2 479 * might container a change (such as a spare assignment) which
0c4f6e37
N
480 * could affect our decisions.
481 */
88b496c2 482 if (a->check_degraded && !frozen &&
0c4f6e37 483 update_queue == NULL && update_queue_pending == NULL) {
6c3fb95c 484 struct metadata_update *updates = NULL;
071cfc42 485 struct mdinfo *newdev = NULL;
6c3fb95c 486 struct active_array *newa;
071cfc42 487 struct mdinfo *d;
3c00ffbe 488
6c3fb95c
NB
489 a->check_degraded = 0;
490
491 /* The array may not be degraded, this is just a good time
492 * to check.
493 */
4e2c1a9a 494 newdev = container->ss->activate_spare(a, &updates);
071cfc42
DW
495 if (!newdev)
496 return;
497
498 newa = duplicate_aa(a);
499 if (!newa)
500 goto out;
501 /* Cool, we can add a device or several. */
502
503 /* Add device to array and set offset/size/slot.
504 * and open files for each newdev */
505 for (d = newdev; d ; d = d->next) {
506 struct mdinfo *newd;
507
508 newd = malloc(sizeof(*newd));
509 if (!newd)
510 continue;
2904b26f 511 if (sysfs_add_disk(&newa->info, d, 0) < 0) {
071cfc42
DW
512 free(newd);
513 continue;
6c3fb95c 514 }
63b4aae3 515 disk_init_and_add(newd, d, newa);
071cfc42
DW
516 }
517 queue_metadata_update(updates);
518 updates = NULL;
4e2c1a9a 519 replace_array(container, a, newa);
071cfc42
DW
520 sysfs_set_str(&a->info, NULL, "sync_action", "recover");
521 out:
522 while (newdev) {
523 d = newdev->next;
524 free(newdev);
525 newdev = d;
6c3fb95c 526 }
071cfc42 527 free_updates(&updates);
6c3fb95c 528 }
0f99b4bd
N
529
530 if (a->check_reshape) {
531 /* mdadm might have added some devices to the array.
532 * We want to disk_init_and_add any such device to a
533 * duplicate_aa and replace a with that.
534 * mdstat doesn't have enough info so we sysfs_read
535 * and look for new stuff.
536 */
537 struct mdinfo *info, *d, *d2, *newd;
aad6f216 538 unsigned long long array_size;
0f99b4bd
N
539 struct active_array *newa = NULL;
540 a->check_reshape = 0;
541 info = sysfs_read(-1, mdstat->devnum,
542 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE);
543 if (!info)
544 goto out2;
545 for (d = info->devs; d; d = d->next) {
546 if (d->disk.raid_disk < 0)
547 continue;
548 for (d2 = a->info.devs; d2; d2 = d2->next)
549 if (d2->disk.raid_disk ==
550 d->disk.raid_disk)
551 break;
552 if (d2)
553 /* already have this one */
554 continue;
555 if (!newa) {
556 newa = duplicate_aa(a);
557 if (!newa)
558 break;
559 }
560 newd = malloc(sizeof(*newd));
561 if (!newd)
562 continue;
563 disk_init_and_add(newd, d, newa);
564 }
aad6f216 565 if (sysfs_get_ll(info, NULL, "array_size", &array_size) == 0
02eedb57 566 && a->info.custom_array_size > array_size*2) {
aad6f216 567 sysfs_set_num(info, NULL, "array_size",
02eedb57 568 a->info.custom_array_size/2);
aad6f216 569 }
0f99b4bd
N
570 out2:
571 sysfs_free(info);
572 if (newa)
4e2c1a9a 573 replace_array(container, a, newa);
0f99b4bd 574 }
549e9569
NB
575}
576
836759d5
DW
577static int aa_ready(struct active_array *aa)
578{
579 struct mdinfo *d;
580 int level = aa->info.array.level;
581
582 for (d = aa->info.devs; d; d = d->next)
583 if (d->state_fd < 0)
584 return 0;
585
586 if (aa->info.state_fd < 0)
587 return 0;
588
589 if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0))
590 return 0;
591
592 if (!aa->container)
593 return 0;
594
595 return 1;
596}
597
549e9569 598static void manage_new(struct mdstat_ent *mdstat,
2a0bb19e
DW
599 struct supertype *container,
600 struct active_array *victim)
549e9569
NB
601{
602 /* A new array has appeared in this container.
603 * Hopefully it is already recorded in the metadata.
604 * Check, then create the new array to report it to
605 * the monitor.
606 */
607
608 struct active_array *new;
609 struct mdinfo *mdi, *di;
cba0191b 610 char *inst;
549e9569 611 int i;
f1d26766 612 int failed = 0;
138477db 613 char buf[40];
549e9569 614
836759d5 615 /* check if array is ready to be monitored */
7023e0b8
N
616 if (!mdstat->active || !mdstat->level)
617 return;
618 if (strcmp(mdstat->level, "raid0") == 0 ||
619 strcmp(mdstat->level, "linear") == 0)
836759d5
DW
620 return;
621
622 mdi = sysfs_read(-1, mdstat->devnum,
623 GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT|
f1d26766 624 GET_DEGRADED|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE);
836759d5 625
549e9569
NB
626 new = malloc(sizeof(*new));
627
836759d5
DW
628 if (!new || !mdi) {
629 if (mdi)
630 sysfs_free(mdi);
631 if (new)
632 free(new);
633 return;
634 }
d52690ac
NB
635 memset(new, 0, sizeof(*new));
636
549e9569 637 new->devnum = mdstat->devnum;
7e1432fb 638 strcpy(new->info.sys_name, devnum2devname(new->devnum));
549e9569
NB
639
640 new->prev_state = new->curr_state = new->next_state = inactive;
641 new->prev_action= new->curr_action= new->next_action= idle;
642
643 new->container = container;
644
e5408a32 645 inst = to_subarray(mdstat, container->devname);
549e9569 646
549e9569 647 new->info.array = mdi->array;
272bcc48 648 new->info.component_size = mdi->component_size;
549e9569
NB
649
650 for (i = 0; i < new->info.array.raid_disks; i++) {
651 struct mdinfo *newd = malloc(sizeof(*newd));
652
653 for (di = mdi->devs; di; di = di->next)
654 if (i == di->disk.raid_disk)
655 break;
656
63b4aae3 657 if (disk_init_and_add(newd, di, new) != 0) {
7da80e6f
DW
658 if (newd)
659 free(newd);
660
f1d26766 661 failed++;
7da80e6f
DW
662 if (failed > new->info.array.failed_disks) {
663 /* we cannot properly monitor without all working disks */
664 new->container = NULL;
665 break;
666 }
549e9569 667 }
549e9569 668 }
836759d5 669
549e9569
NB
670 new->action_fd = sysfs_open(new->devnum, NULL, "sync_action");
671 new->info.state_fd = sysfs_open(new->devnum, NULL, "array_state");
c052ba30 672 new->resync_start_fd = sysfs_open(new->devnum, NULL, "resync_start");
e9dd1598 673 new->metadata_fd = sysfs_open(new->devnum, NULL, "metadata_version");
484240d8 674 new->sync_completed_fd = sysfs_open(new->devnum, NULL, "sync_completed");
4e6e574a
DW
675 dprintf("%s: inst: %d action: %d state: %d\n", __func__, atoi(inst),
676 new->action_fd, new->info.state_fd);
549e9569 677
138477db
AK
678 /* reshape_position is set by mdadm in sysfs
679 * read this information for new arrays only (empty victim)
680 */
681 if ((victim == NULL) &&
682 (sysfs_get_str(mdi, NULL, "sync_action", buf, 40) > 0) &&
683 (strncmp(buf, "reshape", 7) == 0)) {
684 if (sysfs_get_ll(mdi, NULL, "reshape_position",
685 &new->last_checkpoint) != 0)
686 new->last_checkpoint = 0;
0d51bfa2
AK
687 else {
688 int data_disks = mdi->array.raid_disks;
689 if (mdi->array.level == 4 || mdi->array.level == 5)
690 data_disks--;
691 if (mdi->array.level == 6)
692 data_disks -= 2;
693
694 new->last_checkpoint /= data_disks;
695 }
138477db
AK
696 dprintf("mdmon: New monitored array is under reshape.\n"
697 " Last checkpoint is: %llu\n",
698 new->last_checkpoint);
699 }
700
4fa5aef9 701 sysfs_free(mdi);
836759d5
DW
702
703 /* if everything checks out tell the metadata handler we want to
704 * manage this instance
705 */
706 if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) {
707 fprintf(stderr, "mdmon: failed to monitor %s\n",
708 mdstat->metadata_version);
549e9569 709 new->container = NULL;
836759d5 710 free_aa(new);
93f7caca 711 } else {
2a0bb19e 712 replace_array(container, victim, new);
93f7caca
DW
713 if (failed) {
714 new->check_degraded = 1;
715 manage_member(mdstat, new);
716 }
717 }
549e9569
NB
718}
719
5d19760d 720void manage(struct mdstat_ent *mdstat, struct supertype *container)
549e9569
NB
721{
722 /* We have just read mdstat and need to compare it with
723 * the known active arrays.
724 * Arrays with the wrong metadata are ignored.
725 */
726
727 for ( ; mdstat ; mdstat = mdstat->next) {
728 struct active_array *a;
729 if (mdstat->devnum == container->devnum) {
730 manage_container(mdstat, container);
731 continue;
732 }
883a6142 733 if (!is_container_member(mdstat, container->devname))
549e9569
NB
734 /* Not for this array */
735 continue;
736 /* Looks like a member of this container */
5d19760d 737 for (a = container->arrays; a; a = a->next) {
549e9569
NB
738 if (mdstat->devnum == a->devnum) {
739 if (a->container)
740 manage_member(mdstat, a);
741 break;
742 }
743 }
2a0bb19e
DW
744 if (a == NULL || !a->container)
745 manage_new(mdstat, container, a);
549e9569
NB
746 }
747}
748
edd8d13c 749static void handle_message(struct supertype *container, struct metadata_update *msg)
3e70c845 750{
edd8d13c
NB
751 /* queue this metadata update through to the monitor */
752
753 struct metadata_update *mu;
754
313a4a82 755 if (msg->len <= 0)
3c00ffbe
N
756 while (update_queue_pending || update_queue) {
757 check_update_queue(container);
758 usleep(15*1000);
759 }
760
313a4a82
DW
761 if (msg->len == 0) { /* ping_monitor */
762 int cnt;
763
3c00ffbe 764 cnt = monitor_loop_cnt;
1eb252b8
N
765 if (cnt & 1)
766 cnt += 2; /* wait until next pselect */
767 else
768 cnt += 3; /* wait for 2 pselects */
769 wakeup_monitor();
3c00ffbe 770
1eb252b8
N
771 while (monitor_loop_cnt - cnt < 0)
772 usleep(10 * 1000);
313a4a82
DW
773 } else if (msg->len == -1) { /* ping_manager */
774 struct mdstat_ent *mdstat = mdstat_read(1, 0);
775
776 manage(mdstat, container);
777 free_mdstat(mdstat);
6144ed44 778 } else if (!sigterm) {
edd8d13c
NB
779 mu = malloc(sizeof(*mu));
780 mu->len = msg->len;
781 mu->buf = msg->buf;
782 msg->buf = NULL;
783 mu->space = NULL;
cb23f1f4 784 mu->space_list = NULL;
edd8d13c
NB
785 mu->next = NULL;
786 if (container->ss->prepare_update)
787 container->ss->prepare_update(container, mu);
788 queue_metadata_update(mu);
789 }
3e70c845
DW
790}
791
792void read_sock(struct supertype *container)
549e9569
NB
793{
794 int fd;
bfa44e2e 795 struct metadata_update msg;
b109d928
DW
796 int terminate = 0;
797 long fl;
798 int tmo = 3; /* 3 second timeout before hanging up the socket */
549e9569 799
3e70c845 800 fd = accept(container->sock, NULL, NULL);
549e9569
NB
801 if (fd < 0)
802 return;
b109d928
DW
803
804 fl = fcntl(fd, F_GETFL, 0);
805 fl |= O_NONBLOCK;
806 fcntl(fd, F_SETFL, fl);
807
808 do {
809 msg.buf = NULL;
810
811 /* read and validate the message */
812 if (receive_message(fd, &msg, tmo) == 0) {
bfa44e2e 813 handle_message(container, &msg);
bc77ed53
DW
814 if (msg.len == 0) {
815 /* ping reply with version */
816 msg.buf = Version;
817 msg.len = strlen(Version) + 1;
818 if (send_message(fd, &msg, tmo) < 0)
819 terminate = 1;
820 } else if (ack(fd, tmo) < 0)
bfa44e2e
NB
821 terminate = 1;
822 } else
b109d928 823 terminate = 1;
b109d928 824
b109d928
DW
825 } while (!terminate);
826
549e9569
NB
827 close(fd);
828}
1ed3f387 829
e0d6609f
NB
830int exit_now = 0;
831int manager_ready = 0;
549e9569
NB
832void do_manager(struct supertype *container)
833{
834 struct mdstat_ent *mdstat;
4d43913c 835 sigset_t set;
1ed3f387 836
4d43913c
NB
837 sigprocmask(SIG_UNBLOCK, NULL, &set);
838 sigdelset(&set, SIGUSR1);
6144ed44 839 sigdelset(&set, SIGTERM);
549e9569
NB
840
841 do {
1ed3f387 842
e0d6609f
NB
843 if (exit_now)
844 exit(0);
845
3c00ffbe
N
846 /* Can only 'manage' things if 'monitor' is not making
847 * structural changes to metadata, so need to check
848 * update_queue
849 */
850 if (update_queue == NULL) {
851 mdstat = mdstat_read(1, 0);
549e9569 852
3c00ffbe 853 manage(mdstat, container);
549e9569 854
3c00ffbe 855 read_sock(container);
4fa5aef9 856
3c00ffbe
N
857 free_mdstat(mdstat);
858 }
1ed3f387
NB
859 remove_old();
860
2e735d19
NB
861 check_update_queue(container);
862
e0d6609f 863 manager_ready = 1;
4d43913c 864
6144ed44
DW
865 if (sigterm)
866 wakeup_monitor();
867
5d4d1b26 868 if (update_queue == NULL)
58a4ba2a 869 mdstat_wait_fd(container->sock, &set);
5d4d1b26 870 else
3c00ffbe
N
871 /* If an update is happening, just wait for signal */
872 pselect(0, NULL, NULL, NULL, NULL, &set);
549e9569
NB
873 } while(1);
874}