]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Monitor.c
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
35 char devnm
[32]; /* to sync with mdstat info */
39 int active
, working
, failed
, spare
, raid
;
43 int devstate
[MAX_DISKS
];
44 dev_t devid
[MAX_DISKS
];
46 char parent_devnm
[32]; /* For subarray, devnm of parent.
49 struct supertype
*metadata
;
50 struct state
*subarray
;/* for a container it is a link to first subarray
51 * for a subarray it is a link to next subarray
52 * in the same container */
53 struct state
*parent
; /* for a subarray it is a link to its container
64 static int make_daemon(char *pidfile
);
65 static int check_one_sharer(int scan
);
66 static void write_autorebuild_pid(void);
67 static void alert(char *event
, char *dev
, char *disc
, struct alert_info
*info
);
68 static int check_array(struct state
*st
, struct mdstat_ent
*mdstat
,
69 int test
, struct alert_info
*info
,
70 int increments
, char *prefer
);
71 static int add_new_arrays(struct mdstat_ent
*mdstat
, struct state
**statelist
,
72 int test
, struct alert_info
*info
);
73 static void try_spare_migration(struct state
*statelist
, struct alert_info
*info
);
74 static void link_containers_with_subarrays(struct state
*list
);
76 int Monitor(struct mddev_dev
*devlist
,
77 char *mailaddr
, char *alert_cmd
,
79 int daemonise
, int oneshot
,
80 int dosyslog
, char *pidfile
, int increments
,
84 * Every few seconds, scan every md device looking for changes
85 * When a change is found, log it, possibly run the alert command,
86 * and possibly send Email
88 * For each array, we record:
90 * active/working/failed/spare drives
91 * State of each device.
92 * %rebuilt if rebuilding
94 * If the update time changes, check out all the data again
95 * It is possible that we cannot get the state of each device
96 * due to bugs in the md kernel module.
97 * We also read /proc/mdstat to get rebuild percent,
98 * and to get state on all active devices incase of kernel bug.
102 * An active device had Faulty set or Active/Sync removed
104 * A spare device had Faulty set
106 * An active device had a reverse transition
108 * percent went from -1 to +ve
110 * percent went from below to not-below NN%
112 * Couldn't access a device which was previously visible
114 * if we detect an array with active<raid and spare==0
115 * we look at other arrays that have same spare-group
116 * If we find one with active==raid and spare>0,
117 * and if we can get_disk_info and find a name
118 * Then we hot-remove and hot-add to the other array
120 * If devlist is NULL, then we can monitor everything because --scan
121 * was given. We get an initial list from config file and add anything
122 * that appears in /proc/mdstat
125 struct state
*statelist
= NULL
;
128 struct mdstat_ent
*mdstat
= NULL
;
130 struct alert_info info
;
131 struct mddev_ident
*mdlist
;
132 int delay_for_event
= c
->delay
;
135 mailaddr
= conf_get_mailaddr();
136 if (mailaddr
&& ! c
->scan
)
137 pr_err("Monitor using email address \"%s\" from config file\n",
140 mailfrom
= conf_get_mailfrom();
143 alert_cmd
= conf_get_program();
144 if (alert_cmd
&& !c
->scan
)
145 pr_err("Monitor using program \"%s\" from config file\n",
148 if (c
->scan
&& !mailaddr
&& !alert_cmd
&& !dosyslog
) {
149 pr_err("No mail address or alert command - not monitoring.\n");
152 info
.alert_cmd
= alert_cmd
;
153 info
.mailaddr
= mailaddr
;
154 info
.mailfrom
= mailfrom
;
155 info
.dosyslog
= dosyslog
;
158 if (check_one_sharer(c
->scan
))
163 int rv
= make_daemon(pidfile
);
169 write_autorebuild_pid();
171 if (devlist
== NULL
) {
172 mdlist
= conf_get_ident(NULL
);
173 for (; mdlist
; mdlist
= mdlist
->next
) {
176 if (mdlist
->devname
== NULL
)
178 if (strcasecmp(mdlist
->devname
, "<ignore>") == 0)
180 st
= xcalloc(1, sizeof *st
);
181 if (mdlist
->devname
[0] == '/')
182 st
->devname
= xstrdup(mdlist
->devname
);
184 st
->devname
= xmalloc(8+strlen(mdlist
->devname
)+1);
185 strcpy(strcpy(st
->devname
, "/dev/md/"),
188 st
->next
= statelist
;
190 st
->percent
= RESYNC_UNKNOWN
;
192 st
->expected_spares
= mdlist
->spare_disks
;
193 if (mdlist
->spare_group
)
194 st
->spare_group
= xstrdup(mdlist
->spare_group
);
198 struct mddev_dev
*dv
;
200 for (dv
= devlist
; dv
; dv
= dv
->next
) {
201 struct state
*st
= xcalloc(1, sizeof *st
);
202 mdlist
= conf_get_ident(dv
->devname
);
203 st
->devname
= xstrdup(dv
->devname
);
204 st
->next
= statelist
;
206 st
->percent
= RESYNC_UNKNOWN
;
207 st
->expected_spares
= -1;
209 st
->expected_spares
= mdlist
->spare_disks
;
210 if (mdlist
->spare_group
)
211 st
->spare_group
= xstrdup(mdlist
->spare_group
);
219 struct state
*st
, **stp
;
221 int anyredundant
= 0;
225 mdstat
= mdstat_read(oneshot
? 0 : 1, 0);
227 for (st
= statelist
; st
; st
= st
->next
) {
228 if (check_array(st
, mdstat
, c
->test
, &info
,
229 increments
, c
->prefer
))
231 /* for external arrays, metadata is filled for
234 if (st
->metadata
&& st
->metadata
->ss
->external
)
236 if (st
->err
== 0 && !anyredundant
)
240 /* now check if there are any new devices found in mdstat */
242 new_found
= add_new_arrays(mdstat
, &statelist
, c
->test
,
245 /* If an array has active < raid && spare == 0 && spare_group != NULL
246 * Look for another array with spare > 0 and active == raid and same spare_group
247 * if found, choose a device and hotremove/hotadd
249 if (share
&& anydegraded
)
250 try_spare_migration(statelist
, &info
);
254 else if (!anyredundant
) {
258 int wait_result
= mdstat_wait(delay_for_event
);
261 * If mdmonitor is awaken by event, set small delay once
262 * to deal with udev and mdadm.
264 if (wait_result
!= 0) {
268 delay_for_event
= c
->delay
;
275 for (stp
= &statelist
; (st
= *stp
) != NULL
; ) {
276 if (st
->from_auto
&& st
->err
> 5) {
279 free(st
->spare_group
);
285 for (st2
= statelist
; st2
; st2
= statelist
) {
286 statelist
= st2
->next
;
295 static int make_daemon(char *pidfile
)
298 * -1 in the forked daemon
301 * so a none-negative becomes the exit code.
308 FILE *pid_file
= NULL
;
309 int fd
= open(pidfile
, O_WRONLY
| O_CREAT
| O_TRUNC
,
312 pid_file
= fdopen(fd
, "w");
314 perror("cannot create pid file");
316 fprintf(pid_file
,"%d\n", pid
);
331 static int check_one_sharer(int scan
)
336 char comm_path
[PATH_MAX
];
340 sprintf(path
, "%s/autorebuild.pid", MDMON_DIR
);
341 fp
= fopen(path
, "r");
343 if (fscanf(fp
, "%d", &pid
) != 1)
345 snprintf(comm_path
, sizeof(comm_path
),
346 "/proc/%d/comm", pid
);
347 comm_fp
= fopen(comm_path
, "r");
349 if (fscanf(comm_fp
, "%s", comm
) &&
350 strncmp(basename(comm
), Name
, strlen(Name
)) == 0) {
352 pr_err("Only one autorebuild process allowed in scan mode, aborting\n");
357 pr_err("Warning: One autorebuild process already running.\n");
367 static void write_autorebuild_pid()
372 sprintf(path
, "%s/autorebuild.pid", MDMON_DIR
);
374 if (mkdir(MDMON_DIR
, 0700) < 0 && errno
!= EEXIST
) {
375 pr_err("Can't create autorebuild.pid file\n");
377 int fd
= open(path
, O_WRONLY
| O_CREAT
| O_TRUNC
, 0700);
380 fp
= fdopen(fd
, "w");
383 pr_err("Can't create autorebuild.pid file\n");
386 fprintf(fp
, "%d\n", pid
);
392 static void alert(char *event
, char *dev
, char *disc
, struct alert_info
*info
)
396 if (!info
->alert_cmd
&& !info
->mailaddr
&& !info
->dosyslog
) {
397 time_t now
= time(0);
399 printf("%1.15s: %s on %s %s\n", ctime(&now
) + 4,
400 event
, dev
, disc
?disc
:"unknown device");
402 if (info
->alert_cmd
) {
406 waitpid(pid
, NULL
, 0);
411 execl(info
->alert_cmd
, info
->alert_cmd
,
412 event
, dev
, disc
, NULL
);
416 if (info
->mailaddr
&& (strncmp(event
, "Fail", 4) == 0 ||
417 strncmp(event
, "Test", 4) == 0 ||
418 strncmp(event
, "Spares", 6) == 0 ||
419 strncmp(event
, "Degrade", 7) == 0)) {
420 FILE *mp
= popen(Sendmail
, "w");
424 gethostname(hname
, sizeof(hname
));
425 signal(SIGPIPE
, SIG_IGN
);
427 fprintf(mp
, "From: %s\n", info
->mailfrom
);
429 fprintf(mp
, "From: %s monitoring <root>\n",
431 fprintf(mp
, "To: %s\n", info
->mailaddr
);
432 fprintf(mp
, "Subject: %s event on %s:%s\n\n",
436 "This is an automatically generated mail message from %s\n", Name
);
437 fprintf(mp
, "running on %s\n\n", hname
);
440 "A %s event had been detected on md device %s.\n\n", event
, dev
);
442 if (disc
&& disc
[0] != ' ')
444 "It could be related to component device %s.\n\n", disc
);
445 if (disc
&& disc
[0] == ' ')
446 fprintf(mp
, "Extra information:%s.\n\n", disc
);
448 fprintf(mp
, "Faithfully yours, etc.\n");
450 mdstat
= fopen("/proc/mdstat", "r");
455 "\nP.S. The /proc/mdstat file currently contains the following:\n\n");
456 while ((n
= fread(buf
, 1, sizeof(buf
),
458 n
= fwrite(buf
, 1, n
, mp
);
465 /* log the event to syslog maybe */
466 if (info
->dosyslog
) {
467 /* Log at a different severity depending on the event.
469 * These are the critical events: */
470 if (strncmp(event
, "Fail", 4) == 0 ||
471 strncmp(event
, "Degrade", 7) == 0 ||
472 strncmp(event
, "DeviceDisappeared", 17) == 0)
474 /* Good to know about, but are not failures: */
475 else if (strncmp(event
, "Rebuild", 7) == 0 ||
476 strncmp(event
, "MoveSpare", 9) == 0 ||
477 strncmp(event
, "Spares", 6) != 0)
478 priority
= LOG_WARNING
;
479 /* Everything else: */
483 if (disc
&& disc
[0] != ' ')
485 "%s event detected on md device %s, component device %s", event
, dev
, disc
);
488 "%s event detected on md device %s: %s",
492 "%s event detected on md device %s",
497 static int check_array(struct state
*st
, struct mdstat_ent
*mdstat
,
498 int test
, struct alert_info
*ainfo
,
499 int increments
, char *prefer
)
501 /* Update the state 'st' to reflect any changes shown in mdstat,
502 * or found by directly examining the array, and return
503 * '1' if the array is degraded, or '0' if it is optimal (or dead).
505 struct { int state
, major
, minor
; } info
[MAX_DISKS
];
506 struct mdinfo
*sra
= NULL
;
507 mdu_array_info_t array
;
508 struct mdstat_ent
*mse
= NULL
, *mse2
;
509 char *dev
= st
->devname
;
516 int is_container
= 0;
517 unsigned long redundancy_only_flags
= 0;
520 alert("TestMessage", dev
, NULL
, ainfo
);
524 fd
= open(dev
, O_RDONLY
);
528 if (st
->devnm
[0] == 0)
529 strcpy(st
->devnm
, fd2devnm(fd
));
531 for (mse2
= mdstat
; mse2
; mse2
= mse2
->next
)
532 if (strcmp(mse2
->devnm
, st
->devnm
) == 0) {
533 mse2
->devnm
[0] = 0; /* flag it as "used" */
538 /* duplicated array in statelist
539 * or re-created after reading mdstat
545 if (mse
->level
== NULL
)
548 if (!is_container
&& !md_array_active(fd
))
551 fcntl(fd
, F_SETFD
, FD_CLOEXEC
);
552 if (md_get_array_info(fd
, &array
) < 0)
555 if (!is_container
&& map_name(pers
, mse
->level
) > 0)
556 redundancy_only_flags
|= GET_MISMATCH
;
558 sra
= sysfs_read(-1, st
->devnm
, GET_LEVEL
| GET_DISKS
| GET_DEVS
|
559 GET_STATE
| redundancy_only_flags
);
564 /* It's much easier to list what array levels can't
565 * have a device disappear than all of them that can
567 if (sra
->array
.level
== 0 || sra
->array
.level
== -1) {
568 if (!st
->err
&& !st
->from_config
)
569 alert("DeviceDisappeared", dev
, " Wrong-Level", ainfo
);
574 /* this array is in /proc/mdstat */
575 if (array
.utime
== 0)
576 /* external arrays don't update utime, so
577 * just make sure it is always different. */
578 array
.utime
= st
->utime
+ 1;;
581 /* New array appeared where previously had an error */
583 st
->percent
= RESYNC_NONE
;
586 alert("NewArray", st
->devname
, NULL
, ainfo
);
589 if (st
->utime
== array
.utime
&& st
->failed
== sra
->array
.failed_disks
&&
590 st
->working
== sra
->array
.working_disks
&&
591 st
->spare
== sra
->array
.spare_disks
&&
592 (mse
== NULL
|| (mse
->percent
== st
->percent
))) {
593 if ((st
->active
< st
->raid
) && st
->spare
== 0)
597 if (st
->utime
== 0 && /* new array */
598 mse
->pattern
&& strchr(mse
->pattern
, '_') /* degraded */)
599 alert("DegradedArray", dev
, NULL
, ainfo
);
601 if (st
->utime
== 0 && /* new array */ st
->expected_spares
> 0 &&
602 sra
->array
.spare_disks
< st
->expected_spares
)
603 alert("SparesMissing", dev
, NULL
, ainfo
);
604 if (st
->percent
< 0 && st
->percent
!= RESYNC_UNKNOWN
&&
606 alert("RebuildStarted", dev
, NULL
, ainfo
);
607 if (st
->percent
>= 0 && mse
->percent
>= 0 &&
608 (mse
->percent
/ increments
) > (st
->percent
/ increments
)) {
609 char percentalert
[18];
611 * "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
614 if((mse
->percent
/ increments
) == 0)
615 snprintf(percentalert
, sizeof(percentalert
),
618 snprintf(percentalert
, sizeof(percentalert
),
619 "Rebuild%02d", mse
->percent
);
621 alert(percentalert
, dev
, NULL
, ainfo
);
624 if (mse
->percent
== RESYNC_NONE
&& st
->percent
>= 0) {
625 /* Rebuild/sync/whatever just finished.
626 * If there is a number in /mismatch_cnt,
627 * we should report that.
629 if (sra
&& sra
->mismatch_cnt
> 0) {
631 snprintf(cnt
, sizeof(cnt
),
632 " mismatches found: %d (on raid level %d)",
633 sra
->mismatch_cnt
, sra
->array
.level
);
634 alert("RebuildFinished", dev
, cnt
, ainfo
);
636 alert("RebuildFinished", dev
, NULL
, ainfo
);
638 st
->percent
= mse
->percent
;
640 remaining_disks
= sra
->array
.nr_disks
;
641 for (i
= 0; i
< MAX_DISKS
&& remaining_disks
> 0; i
++) {
642 mdu_disk_info_t disc
;
644 if (md_get_disk_info(fd
, &disc
) >= 0) {
645 info
[i
].state
= disc
.state
;
646 info
[i
].major
= disc
.major
;
647 info
[i
].minor
= disc
.minor
;
648 if (disc
.major
|| disc
.minor
)
651 info
[i
].major
= info
[i
].minor
= 0;
655 if (mse
->metadata_version
&&
656 strncmp(mse
->metadata_version
, "external:", 9) == 0 &&
657 is_subarray(mse
->metadata_version
+9)) {
659 strcpy(st
->parent_devnm
, mse
->metadata_version
+ 10);
660 sl
= strchr(st
->parent_devnm
, '/');
664 st
->parent_devnm
[0] = 0;
665 if (st
->metadata
== NULL
&& st
->parent_devnm
[0] == 0)
666 st
->metadata
= super_by_fd(fd
, NULL
);
668 for (i
= 0; i
< MAX_DISKS
; i
++) {
669 mdu_disk_info_t disc
= {0, 0, 0, 0, 0};
674 if (i
< last_disk
&& (info
[i
].major
|| info
[i
].minor
)) {
675 newstate
= info
[i
].state
;
676 dv
= map_dev_preferred(info
[i
].major
, info
[i
].minor
, 1,
678 disc
.state
= newstate
;
679 disc
.major
= info
[i
].major
;
680 disc
.minor
= info
[i
].minor
;
682 newstate
= (1 << MD_DISK_REMOVED
);
684 if (dv
== NULL
&& st
->devid
[i
])
685 dv
= map_dev_preferred(major(st
->devid
[i
]),
686 minor(st
->devid
[i
]), 1, prefer
);
687 change
= newstate
^ st
->devstate
[i
];
688 if (st
->utime
&& change
&& !st
->err
&& !new_array
) {
689 if ((st
->devstate
[i
]&change
) & (1 << MD_DISK_SYNC
))
690 alert("Fail", dev
, dv
, ainfo
);
691 else if ((newstate
& (1 << MD_DISK_FAULTY
)) &&
692 (disc
.major
|| disc
.minor
) &&
693 st
->devid
[i
] == makedev(disc
.major
,
695 alert("FailSpare", dev
, dv
, ainfo
);
696 else if ((newstate
&change
) & (1 << MD_DISK_SYNC
))
697 alert("SpareActive", dev
, dv
, ainfo
);
699 st
->devstate
[i
] = newstate
;
700 st
->devid
[i
] = makedev(disc
.major
, disc
.minor
);
702 st
->active
= sra
->array
.active_disks
;
703 st
->working
= sra
->array
.working_disks
;
704 st
->spare
= sra
->array
.spare_disks
;
705 st
->failed
= sra
->array
.failed_disks
;
706 st
->utime
= array
.utime
;
707 st
->raid
= sra
->array
.raid_disks
;
709 if ((st
->active
< st
->raid
) && st
->spare
== 0)
720 if (!st
->err
&& !is_container
)
721 alert("DeviceDisappeared", dev
, NULL
, ainfo
);
726 static int add_new_arrays(struct mdstat_ent
*mdstat
, struct state
**statelist
,
727 int test
, struct alert_info
*info
)
729 struct mdstat_ent
*mse
;
733 for (mse
= mdstat
; mse
; mse
= mse
->next
)
734 if (mse
->devnm
[0] && (!mse
->level
|| /* retrieve containers */
735 (strcmp(mse
->level
, "raid0") != 0 &&
736 strcmp(mse
->level
, "linear") != 0))) {
737 struct state
*st
= xcalloc(1, sizeof *st
);
738 mdu_array_info_t array
;
741 name
= get_md_name(mse
->devnm
);
747 st
->devname
= xstrdup(name
);
748 if ((fd
= open(st
->devname
, O_RDONLY
)) < 0 ||
749 md_get_array_info(fd
, &array
) < 0) {
753 put_md_name(st
->devname
);
756 st
->metadata
->ss
->free_super(st
->metadata
);
763 st
->next
= *statelist
;
766 strcpy(st
->devnm
, mse
->devnm
);
767 st
->percent
= RESYNC_UNKNOWN
;
768 st
->expected_spares
= -1;
769 if (mse
->metadata_version
&&
770 strncmp(mse
->metadata_version
,
771 "external:", 9) == 0 &&
772 is_subarray(mse
->metadata_version
+9)) {
774 strcpy(st
->parent_devnm
,
775 mse
->metadata_version
+10);
776 sl
= strchr(st
->parent_devnm
, '/');
779 st
->parent_devnm
[0] = 0;
782 alert("TestMessage", st
->devname
, NULL
, info
);
788 static int get_required_spare_criteria(struct state
*st
,
789 struct spare_criteria
*sc
)
793 if (!st
->metadata
|| !st
->metadata
->ss
->get_spare_criteria
) {
799 fd
= open(st
->devname
, O_RDONLY
);
802 if (st
->metadata
->ss
->external
)
803 st
->metadata
->ss
->load_container(st
->metadata
, fd
, st
->devname
);
805 st
->metadata
->ss
->load_super(st
->metadata
, fd
, st
->devname
);
807 if (!st
->metadata
->sb
)
810 st
->metadata
->ss
->get_spare_criteria(st
->metadata
, sc
);
811 st
->metadata
->ss
->free_super(st
->metadata
);
816 static int check_donor(struct state
*from
, struct state
*to
)
823 /* Cannot move from a member */
827 for (sub
= from
->subarray
; sub
; sub
= sub
->subarray
)
828 /* If source array has degraded subarrays, don't
831 if (sub
->active
< sub
->raid
)
833 if (from
->metadata
->ss
->external
== 0)
834 if (from
->active
< from
->raid
)
836 if (from
->spare
<= 0)
841 static dev_t
choose_spare(struct state
*from
, struct state
*to
,
842 struct domainlist
*domlist
, struct spare_criteria
*sc
)
847 for (d
= from
->raid
; !dev
&& d
< MAX_DISKS
; d
++) {
848 if (from
->devid
[d
] > 0 && from
->devstate
[d
] == 0) {
849 struct dev_policy
*pol
;
850 unsigned long long dev_size
;
851 unsigned int dev_sector_size
;
853 if (to
->metadata
->ss
->external
&&
854 test_partition_from_id(from
->devid
[d
]))
858 dev_size_from_id(from
->devid
[d
], &dev_size
) &&
859 dev_size
< sc
->min_size
)
862 if (sc
->sector_size
&&
863 dev_sector_size_from_id(from
->devid
[d
],
865 sc
->sector_size
!= dev_sector_size
)
868 pol
= devid_policy(from
->devid
[d
]);
869 if (from
->spare_group
)
870 pol_add(&pol
, pol_domain
,
871 from
->spare_group
, NULL
);
872 if (domain_test(domlist
, pol
,
873 to
->metadata
->ss
->name
) == 1)
874 dev
= from
->devid
[d
];
875 dev_policy_free(pol
);
881 static dev_t
container_choose_spare(struct state
*from
, struct state
*to
,
882 struct domainlist
*domlist
,
883 struct spare_criteria
*sc
, int active
)
885 /* This is similar to choose_spare, but we cannot trust devstate,
886 * so we need to read the metadata instead
889 struct supertype
*st
= from
->metadata
;
890 int fd
= open(from
->devname
, O_RDONLY
);
896 if (!st
->ss
->getinfo_super_disks
) {
901 err
= st
->ss
->load_container(st
, fd
, NULL
);
907 /* We must check if number of active disks has not increased
908 * since ioctl in main loop. mdmon may have added spare
909 * to subarray. If so we do not need to look for more spares
910 * so return non zero value */
913 list
= st
->ss
->getinfo_super_disks(st
);
915 st
->ss
->free_super(st
);
920 if (dp
->disk
.state
& (1 << MD_DISK_SYNC
) &&
921 !(dp
->disk
.state
& (1 << MD_DISK_FAULTY
)))
926 if (active
< active_cnt
) {
927 /* Spare just activated.*/
928 st
->ss
->free_super(st
);
933 /* We only need one spare so full list not needed */
934 list
= container_choose_spares(st
, sc
, domlist
, from
->spare_group
,
935 to
->metadata
->ss
->name
, 1);
937 struct mdinfo
*disks
= list
->devs
;
939 dev
= makedev(disks
->disk
.major
, disks
->disk
.minor
);
942 st
->ss
->free_super(st
);
946 static void try_spare_migration(struct state
*statelist
, struct alert_info
*info
)
950 struct spare_criteria sc
;
952 link_containers_with_subarrays(statelist
);
953 for (st
= statelist
; st
; st
= st
->next
)
954 if (st
->active
< st
->raid
&& st
->spare
== 0 && !st
->err
) {
955 struct domainlist
*domlist
= NULL
;
957 struct state
*to
= st
;
959 if (to
->parent_devnm
[0] && !to
->parent
)
960 /* subarray monitored without parent container
961 * we can't move spares here */
965 /* member of a container */
968 if (get_required_spare_criteria(to
, &sc
))
970 if (to
->metadata
->ss
->external
) {
971 /* We must make sure there is
972 * no suitable spare in container already.
973 * If there is we don't add more */
974 dev_t devid
= container_choose_spare(
975 to
, to
, NULL
, &sc
, st
->active
);
979 for (d
= 0; d
< MAX_DISKS
; d
++)
981 domainlist_add_dev(&domlist
,
983 to
->metadata
->ss
->name
);
985 domain_add(&domlist
, to
->spare_group
);
987 * No spare migration if the destination
988 * has no domain. Skip this array.
992 for (from
=statelist
; from
; from
=from
->next
) {
994 if (!check_donor(from
, to
))
996 if (from
->metadata
->ss
->external
)
997 devid
= container_choose_spare(
998 from
, to
, domlist
, &sc
, 0);
1000 devid
= choose_spare(from
, to
, domlist
,
1003 move_spare(from
->devname
, to
->devname
,
1005 alert("MoveSpare", to
->devname
,
1006 from
->devname
, info
);
1010 domain_free(domlist
);
1014 /* search the statelist to connect external
1015 * metadata subarrays with their containers
1016 * We always completely rebuild the tree from scratch as
1017 * that is safest considering the possibility of entries
1018 * disappearing or changing.
1020 static void link_containers_with_subarrays(struct state
*list
)
1024 for (st
= list
; st
; st
= st
->next
) {
1026 st
->subarray
= NULL
;
1028 for (st
= list
; st
; st
= st
->next
)
1029 if (st
->parent_devnm
[0])
1030 for (cont
= list
; cont
; cont
= cont
->next
)
1031 if (!cont
->err
&& cont
->parent_devnm
[0] == 0 &&
1032 strcmp(cont
->devnm
, st
->parent_devnm
) == 0) {
1034 st
->subarray
= cont
->subarray
;
1035 cont
->subarray
= st
;
1040 /* Not really Monitor but ... */
1047 int frozen_remaining
= 3;
1049 if (!stat_is_blkdev(dev
, &rdev
))
1052 tmp
= devid2devnm(rdev
);
1054 pr_err("Cannot get md device name.\n");
1061 struct mdstat_ent
*ms
= mdstat_read(1, 0);
1062 struct mdstat_ent
*e
;
1064 for (e
= ms
; e
; e
= e
->next
)
1065 if (strcmp(e
->devnm
, devnm
) == 0)
1068 if (e
&& e
->percent
== RESYNC_NONE
) {
1069 /* We could be in the brief pause before something
1070 * starts. /proc/mdstat doesn't show that, but
1076 if (sysfs_init(&mdi
, -1, devnm
))
1078 if (sysfs_get_str(&mdi
, NULL
, "sync_action",
1080 strcmp(buf
,"idle\n") != 0) {
1081 e
->percent
= RESYNC_UNKNOWN
;
1082 if (strcmp(buf
, "frozen\n") == 0) {
1083 if (frozen_remaining
== 0)
1084 e
->percent
= RESYNC_NONE
;
1086 frozen_remaining
-= 1;
1090 if (!e
|| e
->percent
== RESYNC_NONE
) {
1091 if (e
&& e
->metadata_version
&&
1092 strncmp(e
->metadata_version
, "external:", 9) == 0) {
1093 if (is_subarray(&e
->metadata_version
[9]))
1094 ping_monitor(&e
->metadata_version
[9]);
1096 ping_monitor(devnm
);
1107 /* The state "broken" is used only for RAID0/LINEAR - it's the same as
1108 * "clean", but used in case the array has one or more members missing.
1110 static char *clean_states
[] = {
1111 "clear", "inactive", "readonly", "read-auto", "clean", "broken", NULL
};
1113 int WaitClean(char *dev
, int verbose
)
1120 if (!stat_is_blkdev(dev
, NULL
))
1122 fd
= open(dev
, O_RDONLY
);
1125 pr_err("Couldn't open %s: %s\n", dev
, strerror(errno
));
1129 strcpy(devnm
, fd2devnm(fd
));
1130 mdi
= sysfs_read(fd
, devnm
, GET_VERSION
|GET_LEVEL
|GET_SAFEMODE
);
1133 pr_err("Failed to read sysfs attributes for %s\n", dev
);
1138 switch(mdi
->array
.level
) {
1140 case LEVEL_MULTIPATH
:
1142 /* safemode delay is irrelevant for these levels */
1146 /* for internal metadata the kernel handles the final clean
1147 * transition, containers can never be dirty
1149 if (!is_subarray(mdi
->text_version
))
1152 /* safemode disabled ? */
1153 if (mdi
->safe_mode_delay
== 0)
1157 int state_fd
= sysfs_open(fd2devnm(fd
), NULL
, "array_state");
1161 /* minimize the safe_mode_delay and prepare to wait up to 5s
1162 * for writes to quiesce
1164 sysfs_set_safemode(mdi
, 1);
1166 /* wait for array_state to be clean */
1168 rv
= read(state_fd
, buf
, sizeof(buf
));
1171 if (sysfs_match_word(buf
, clean_states
) <
1172 (int)ARRAY_SIZE(clean_states
) - 1)
1174 rv
= sysfs_wait(state_fd
, &delay
);
1175 if (rv
< 0 && errno
!= EINTR
)
1177 lseek(state_fd
, 0, SEEK_SET
);
1181 else if (ping_monitor(mdi
->text_version
) == 0) {
1182 /* we need to ping to close the window between array
1183 * state transitioning to clean and the metadata being
1189 pr_err("Error connecting monitor with %s\n", dev
);
1192 pr_err("Error waiting for %s to be clean\n", dev
);
1194 /* restore the original safe_mode_delay */
1195 sysfs_set_safemode(mdi
, mdi
->safe_mode_delay
);