]> git.ipfire.org Git - thirdparty/mdadm.git/blob - managemon.c
ddf: implement remove_from_super
[thirdparty/mdadm.git] / managemon.c
1 /*
2 * mdmon - monitor external metadata arrays
3 *
4 * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de>
5 * Copyright (C) 2007-2009 Intel Corporation
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 */
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 *
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....
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"
107 #include <sys/syscall.h>
108 #include <sys/socket.h>
109 #include <signal.h>
110
111 static void close_aa(struct active_array *aa)
112 {
113 struct mdinfo *d;
114
115 for (d = aa->info.devs; d; d = d->next) {
116 close(d->recovery_fd);
117 close(d->state_fd);
118 }
119
120 close(aa->action_fd);
121 close(aa->info.state_fd);
122 close(aa->resync_start_fd);
123 close(aa->metadata_fd);
124 close(aa->sync_completed_fd);
125 }
126
127 static void free_aa(struct active_array *aa)
128 {
129 /* Note that this doesn't close fds if they are being used
130 * by a clone. ->container will be set for a clone
131 */
132 dprintf("%s: devnum: %d\n", __func__, aa->devnum);
133 if (!aa->container)
134 close_aa(aa);
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
143 static 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 }
165 *dp2 = NULL;
166
167 return newa;
168 }
169
170 static void wakeup_monitor(void)
171 {
172 /* tgkill(getpid(), mon_tid, SIGUSR1); */
173 int pid = getpid();
174 syscall(SYS_tgkill, pid, mon_tid, SIGUSR1);
175 }
176
177 static 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;
185 wakeup_monitor();
186 }
187 }
188
189 static 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 */
199 remove_old();
200 while (pending_discard) {
201 while (discard_this == NULL)
202 sleep(1);
203 remove_old();
204 }
205 pending_discard = old;
206 new->replaces = old;
207 new->next = container->arrays;
208 container->arrays = new;
209 wakeup_monitor();
210 }
211
212 struct metadata_update *update_queue = NULL;
213 struct metadata_update *update_queue_handled = NULL;
214 struct metadata_update *update_queue_pending = NULL;
215
216 static void free_updates(struct metadata_update **update)
217 {
218 while (*update) {
219 struct metadata_update *this = *update;
220 void **space_list = this->space_list;
221
222 *update = this->next;
223 free(this->buf);
224 free(this->space);
225 while (space_list) {
226 void *space = space_list;
227 space_list = *space_list;
228 free(space);
229 }
230 free(this);
231 }
232 }
233
234 void check_update_queue(struct supertype *container)
235 {
236 free_updates(&update_queue_handled);
237
238 if (update_queue == NULL &&
239 update_queue_pending) {
240 update_queue = update_queue_pending;
241 update_queue_pending = NULL;
242 wakeup_monitor();
243 }
244 }
245
246 static void queue_metadata_update(struct metadata_update *mu)
247 {
248 struct metadata_update **qp;
249
250 qp = &update_queue_pending;
251 while (*qp)
252 qp = & ((*qp)->next);
253 *qp = mu;
254 }
255
256 static void add_disk_to_container(struct supertype *st, struct mdinfo *sd)
257 {
258 int dfd;
259 char nm[20];
260 struct supertype *st2;
261 struct metadata_update *update = NULL;
262 struct mdinfo info;
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
274 sd->next = st->devs;
275 st->devs = sd;
276
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
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) {
287 st2->ss->getinfo_super(st, &info, NULL);
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
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
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 */
313 static 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 };
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);
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 */
332 st->ss->write_init_super(st);
333 queue_metadata_update(update);
334 st->update_tail = NULL;
335 }
336
337 static void manage_container(struct mdstat_ent *mdstat,
338 struct supertype *container)
339 {
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.
345 * FIXME should we look for compatible metadata and take hints
346 * about spare assignment.... probably not.
347 */
348 if (mdstat->devcnt != container->devcnt) {
349 struct mdinfo **cdp, *cd, *di, *mdi;
350 int found;
351
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 */
357 mdi = sysfs_read(-1, mdstat->devnum, GET_DEVS);
358 if (!mdi) {
359 /* invalidate the current count so we can try again */
360 container->devcnt = -1;
361 return;
362 }
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;
376 remove_disk_from_container(container, cd);
377 free(cd);
378 } else
379 cdp = &(*cdp)->next;
380 }
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;
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 }
398 }
399 sysfs_free(mdi);
400 container->devcnt = mdstat->devcnt;
401 }
402 }
403
404 static 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
421 static 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 *
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.
432 *
433 * We also want to handle degraded arrays here by
434 * trying to find and assign a spare.
435 * We do that whenever the monitor tells us too.
436 */
437 char buf[64];
438 int frozen;
439
440 // FIXME
441 a->info.array.raid_disks = mdstat->raid_disks;
442 // MORE
443
444 /* honor 'frozen' */
445 if (sysfs_get_str(&a->info, NULL, "metadata_version", buf, sizeof(buf)) > 0)
446 frozen = buf[9] == '-';
447 else
448 frozen = 1; /* can't read metadata_version assume the worst */
449
450 /* If sync_action is not 'idle' then don't try recovery now */
451 if (!frozen
452 && sysfs_get_str(&a->info, NULL, "sync_action", buf, sizeof(buf)) > 0
453 && strncmp(buf, "idle", 4) != 0)
454 frozen = 1;
455
456 if (mdstat->level) {
457 int level = map_name(pers, mdstat->level);
458 if (a->info.array.level != level && level >= 0) {
459 struct active_array *newa = duplicate_aa(a);
460 if (newa) {
461 newa->info.array.level = level;
462 replace_array(a->container, a, newa);
463 a = newa;
464 }
465 }
466 }
467
468 if (a->check_degraded && !frozen) {
469 struct metadata_update *updates = NULL;
470 struct mdinfo *newdev = NULL;
471 struct active_array *newa;
472 struct mdinfo *d;
473
474 a->check_degraded = 0;
475
476 /* The array may not be degraded, this is just a good time
477 * to check.
478 */
479 newdev = a->container->ss->activate_spare(a, &updates);
480 if (!newdev)
481 return;
482
483 newa = duplicate_aa(a);
484 if (!newa)
485 goto out;
486 /* Cool, we can add a device or several. */
487
488 /* Add device to array and set offset/size/slot.
489 * and open files for each newdev */
490 for (d = newdev; d ; d = d->next) {
491 struct mdinfo *newd;
492
493 newd = malloc(sizeof(*newd));
494 if (!newd)
495 continue;
496 if (sysfs_add_disk(&newa->info, d, 0) < 0) {
497 free(newd);
498 continue;
499 }
500 disk_init_and_add(newd, d, newa);
501 }
502 queue_metadata_update(updates);
503 updates = NULL;
504 replace_array(a->container, a, newa);
505 sysfs_set_str(&a->info, NULL, "sync_action", "recover");
506 out:
507 while (newdev) {
508 d = newdev->next;
509 free(newdev);
510 newdev = d;
511 }
512 free_updates(&updates);
513 }
514
515 if (a->check_reshape) {
516 /* mdadm might have added some devices to the array.
517 * We want to disk_init_and_add any such device to a
518 * duplicate_aa and replace a with that.
519 * mdstat doesn't have enough info so we sysfs_read
520 * and look for new stuff.
521 */
522 struct mdinfo *info, *d, *d2, *newd;
523 unsigned long long array_size;
524 struct active_array *newa = NULL;
525 a->check_reshape = 0;
526 info = sysfs_read(-1, mdstat->devnum,
527 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE);
528 if (!info)
529 goto out2;
530 for (d = info->devs; d; d = d->next) {
531 if (d->disk.raid_disk < 0)
532 continue;
533 for (d2 = a->info.devs; d2; d2 = d2->next)
534 if (d2->disk.raid_disk ==
535 d->disk.raid_disk)
536 break;
537 if (d2)
538 /* already have this one */
539 continue;
540 if (!newa) {
541 newa = duplicate_aa(a);
542 if (!newa)
543 break;
544 }
545 newd = malloc(sizeof(*newd));
546 if (!newd)
547 continue;
548 disk_init_and_add(newd, d, newa);
549 }
550 if (sysfs_get_ll(info, NULL, "array_size", &array_size) == 0
551 && a->info.custom_array_size > array_size*2) {
552 sysfs_set_num(info, NULL, "array_size",
553 a->info.custom_array_size/2);
554 }
555 out2:
556 sysfs_free(info);
557 if (newa)
558 replace_array(a->container, a, newa);
559 }
560 }
561
562 static int aa_ready(struct active_array *aa)
563 {
564 struct mdinfo *d;
565 int level = aa->info.array.level;
566
567 for (d = aa->info.devs; d; d = d->next)
568 if (d->state_fd < 0)
569 return 0;
570
571 if (aa->info.state_fd < 0)
572 return 0;
573
574 if (level > 0 && (aa->action_fd < 0 || aa->resync_start_fd < 0))
575 return 0;
576
577 if (!aa->container)
578 return 0;
579
580 return 1;
581 }
582
583 static void manage_new(struct mdstat_ent *mdstat,
584 struct supertype *container,
585 struct active_array *victim)
586 {
587 /* A new array has appeared in this container.
588 * Hopefully it is already recorded in the metadata.
589 * Check, then create the new array to report it to
590 * the monitor.
591 */
592
593 struct active_array *new;
594 struct mdinfo *mdi, *di;
595 char *inst;
596 int i;
597 int failed = 0;
598 char buf[40];
599
600 /* check if array is ready to be monitored */
601 if (!mdstat->active)
602 return;
603
604 mdi = sysfs_read(-1, mdstat->devnum,
605 GET_LEVEL|GET_CHUNK|GET_DISKS|GET_COMPONENT|
606 GET_DEGRADED|GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE);
607
608 new = malloc(sizeof(*new));
609
610 if (!new || !mdi) {
611 if (mdi)
612 sysfs_free(mdi);
613 if (new)
614 free(new);
615 return;
616 }
617 memset(new, 0, sizeof(*new));
618
619 new->devnum = mdstat->devnum;
620 strcpy(new->info.sys_name, devnum2devname(new->devnum));
621
622 new->prev_state = new->curr_state = new->next_state = inactive;
623 new->prev_action= new->curr_action= new->next_action= idle;
624
625 new->container = container;
626
627 inst = to_subarray(mdstat, container->devname);
628
629 new->info.array = mdi->array;
630 new->info.component_size = mdi->component_size;
631
632 for (i = 0; i < new->info.array.raid_disks; i++) {
633 struct mdinfo *newd = malloc(sizeof(*newd));
634
635 for (di = mdi->devs; di; di = di->next)
636 if (i == di->disk.raid_disk)
637 break;
638
639 if (disk_init_and_add(newd, di, new) != 0) {
640 if (newd)
641 free(newd);
642
643 failed++;
644 if (failed > new->info.array.failed_disks) {
645 /* we cannot properly monitor without all working disks */
646 new->container = NULL;
647 break;
648 }
649 }
650 }
651
652 new->action_fd = sysfs_open(new->devnum, NULL, "sync_action");
653 new->info.state_fd = sysfs_open(new->devnum, NULL, "array_state");
654 new->resync_start_fd = sysfs_open(new->devnum, NULL, "resync_start");
655 new->metadata_fd = sysfs_open(new->devnum, NULL, "metadata_version");
656 new->sync_completed_fd = sysfs_open(new->devnum, NULL, "sync_completed");
657 dprintf("%s: inst: %d action: %d state: %d\n", __func__, atoi(inst),
658 new->action_fd, new->info.state_fd);
659
660 /* reshape_position is set by mdadm in sysfs
661 * read this information for new arrays only (empty victim)
662 */
663 if ((victim == NULL) &&
664 (sysfs_get_str(mdi, NULL, "sync_action", buf, 40) > 0) &&
665 (strncmp(buf, "reshape", 7) == 0)) {
666 if (sysfs_get_ll(mdi, NULL, "reshape_position",
667 &new->last_checkpoint) != 0)
668 new->last_checkpoint = 0;
669 else {
670 int data_disks = mdi->array.raid_disks;
671 if (mdi->array.level == 4 || mdi->array.level == 5)
672 data_disks--;
673 if (mdi->array.level == 6)
674 data_disks -= 2;
675
676 new->last_checkpoint /= data_disks;
677 }
678 dprintf("mdmon: New monitored array is under reshape.\n"
679 " Last checkpoint is: %llu\n",
680 new->last_checkpoint);
681 }
682
683 sysfs_free(mdi);
684
685 /* if everything checks out tell the metadata handler we want to
686 * manage this instance
687 */
688 if (!aa_ready(new) || container->ss->open_new(container, new, inst) < 0) {
689 fprintf(stderr, "mdmon: failed to monitor %s\n",
690 mdstat->metadata_version);
691 new->container = NULL;
692 free_aa(new);
693 } else {
694 replace_array(container, victim, new);
695 if (failed) {
696 new->check_degraded = 1;
697 manage_member(mdstat, new);
698 }
699 }
700 }
701
702 void manage(struct mdstat_ent *mdstat, struct supertype *container)
703 {
704 /* We have just read mdstat and need to compare it with
705 * the known active arrays.
706 * Arrays with the wrong metadata are ignored.
707 */
708
709 for ( ; mdstat ; mdstat = mdstat->next) {
710 struct active_array *a;
711 if (mdstat->devnum == container->devnum) {
712 manage_container(mdstat, container);
713 continue;
714 }
715 if (!is_container_member(mdstat, container->devname))
716 /* Not for this array */
717 continue;
718 /* Looks like a member of this container */
719 for (a = container->arrays; a; a = a->next) {
720 if (mdstat->devnum == a->devnum) {
721 if (a->container)
722 manage_member(mdstat, a);
723 break;
724 }
725 }
726 if (a == NULL || !a->container)
727 manage_new(mdstat, container, a);
728 }
729 }
730
731 static void handle_message(struct supertype *container, struct metadata_update *msg)
732 {
733 /* queue this metadata update through to the monitor */
734
735 struct metadata_update *mu;
736
737 if (msg->len <= 0)
738 while (update_queue_pending || update_queue) {
739 check_update_queue(container);
740 usleep(15*1000);
741 }
742
743 if (msg->len == 0) { /* ping_monitor */
744 int cnt;
745
746 cnt = monitor_loop_cnt;
747 if (cnt & 1)
748 cnt += 2; /* wait until next pselect */
749 else
750 cnt += 3; /* wait for 2 pselects */
751 wakeup_monitor();
752
753 while (monitor_loop_cnt - cnt < 0)
754 usleep(10 * 1000);
755 } else if (msg->len == -1) { /* ping_manager */
756 struct mdstat_ent *mdstat = mdstat_read(1, 0);
757
758 manage(mdstat, container);
759 free_mdstat(mdstat);
760 } else if (!sigterm) {
761 mu = malloc(sizeof(*mu));
762 mu->len = msg->len;
763 mu->buf = msg->buf;
764 msg->buf = NULL;
765 mu->space = NULL;
766 mu->space_list = NULL;
767 mu->next = NULL;
768 if (container->ss->prepare_update)
769 container->ss->prepare_update(container, mu);
770 queue_metadata_update(mu);
771 }
772 }
773
774 void read_sock(struct supertype *container)
775 {
776 int fd;
777 struct metadata_update msg;
778 int terminate = 0;
779 long fl;
780 int tmo = 3; /* 3 second timeout before hanging up the socket */
781
782 fd = accept(container->sock, NULL, NULL);
783 if (fd < 0)
784 return;
785
786 fl = fcntl(fd, F_GETFL, 0);
787 fl |= O_NONBLOCK;
788 fcntl(fd, F_SETFL, fl);
789
790 do {
791 msg.buf = NULL;
792
793 /* read and validate the message */
794 if (receive_message(fd, &msg, tmo) == 0) {
795 handle_message(container, &msg);
796 if (msg.len == 0) {
797 /* ping reply with version */
798 msg.buf = Version;
799 msg.len = strlen(Version) + 1;
800 if (send_message(fd, &msg, tmo) < 0)
801 terminate = 1;
802 } else if (ack(fd, tmo) < 0)
803 terminate = 1;
804 } else
805 terminate = 1;
806
807 } while (!terminate);
808
809 close(fd);
810 }
811
812 int exit_now = 0;
813 int manager_ready = 0;
814 void do_manager(struct supertype *container)
815 {
816 struct mdstat_ent *mdstat;
817 sigset_t set;
818
819 sigprocmask(SIG_UNBLOCK, NULL, &set);
820 sigdelset(&set, SIGUSR1);
821 sigdelset(&set, SIGTERM);
822
823 do {
824
825 if (exit_now)
826 exit(0);
827
828 /* Can only 'manage' things if 'monitor' is not making
829 * structural changes to metadata, so need to check
830 * update_queue
831 */
832 if (update_queue == NULL) {
833 mdstat = mdstat_read(1, 0);
834
835 manage(mdstat, container);
836
837 read_sock(container);
838
839 free_mdstat(mdstat);
840 }
841 remove_old();
842
843 check_update_queue(container);
844
845 manager_ready = 1;
846
847 if (sigterm)
848 wakeup_monitor();
849
850 if (update_queue == NULL)
851 mdstat_wait_fd(container->sock, &set);
852 else
853 /* If an update is happening, just wait for signal */
854 pselect(0, NULL, NULL, NULL, NULL, &set);
855 } while(1);
856 }