/*
- * mdctl - manage Linux "md" devices aka RAID arrays.
+ * mdadm - manage Linux "md" devices aka RAID arrays.
*
* Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
*
* Australia
*/
-#include "mdctl.h"
-#include "md_u.h"
-#include "md_p.h"
+#include "mdadm.h"
-int Assemble(char *mddev, int mdfd,
+int Assemble(struct superswitch *ss, char *mddev, int mdfd,
mddev_ident_t ident, char *conffile,
mddev_dev_t devlist,
int readonly, int runstop,
+ char *update,
int verbose, int force)
{
/*
* Much of the work of Assemble is in finding and/or
* checking the disks to make sure they look right.
*
- * If mddev is not set, then scan must be and we
+ * If mddev is not set, then scan must be set and we
* read through the config file for dev+uuid mapping
* We recurse, setting mddev, for each device that
* - isn't running
- * - has a valid uuid (or any uuid if !uuidset
+ * - has a valid uuid (or any uuid if !uuidset)
*
* If mddev is set, we try to determine state of md.
* check version - must be at least 0.90.0
* For each device:
* Check superblock - discard if bad
* Check uuid (set if we don't have one) - discard if no match
- * Check superblock similarity if we have a superbloc - discard if different
- * Record events, devicenum, utime
+ * Check superblock similarity if we have a superblock - discard if different
+ * Record events, devicenum
* This should give us a list of devices for the array
- * We should collect the most recent event and utime numbers
+ * We should collect the most recent event number
*
* Count disks with recent enough event count
* While force && !enough disks
*/
int old_linux = 0;
int vers;
- mdu_array_info_t array;
- mdp_super_t first_super, super;
+ void *first_super = NULL, *super = NULL;
struct {
char *devname;
- int major, minor;
+ unsigned int major, minor;
+ unsigned int oldmajor, oldminor;
long long events;
- time_t utime;
int uptodate;
+ int state;
int raid_disk;
- } devices[MD_SB_DISKS];
- int best[MD_SB_DISKS]; /* indexed by raid_disk */
- int devcnt = 0, okcnt, sparecnt;
- int i;
+ } *devices;
+ int *best = NULL; /* indexed by raid_disk */
+ unsigned int bestcnt = 0;
+ int devcnt = 0;
+ unsigned int okcnt, sparecnt;
+ unsigned int req_cnt;
+ unsigned int i;
int most_recent = 0;
int chosen_drive;
int change = 0;
int inargv = 0;
int start_partial_ok = force || devlist==NULL;
+ unsigned int num_devs;
+ mddev_dev_t tmpdev;
+ struct mdinfo info;
vers = md_get_version(mdfd);
if (vers <= 0) {
- fprintf(stderr, Name ": %s appears not to be an md device.\n");
+ fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
return 1;
}
if (vers < 9000) {
if (get_linux_version() < 2004000)
old_linux = 1;
- if (ioctl(mdfd, GET_ARRAY_INFO, &array)>=0) {
+ if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
mddev);
return 1;
devlist = conf_get_devs(conffile);
else inargv = 1;
- first_super.md_magic = 0;
- for (i=0; i<MD_SB_DISKS; i++)
- best[i] = -1;
+ tmpdev = devlist; num_devs = 0;
+ while (tmpdev) {
+ num_devs++;
+ tmpdev = tmpdev->next;
+ }
+ devices = malloc(num_devs * sizeof(*devices));
+
+ if (!ss && ident->ss) ss = ident->ss;
if (verbose)
fprintf(stderr, Name ": looking for devices for %s\n",
while ( devlist) {
char *devname;
- int this_uuid[4];
int dfd;
struct stat stb;
- int havesuper=0;
+ struct superswitch *tss = ss;
devname = devlist->devname;
devlist = devlist->next;
if (ident->devices &&
- !match_oneof(ident->devices, devname))
+ !match_oneof(ident->devices, devname)) {
+ if (inargv || verbose)
+ fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
continue;
+ }
+
+ if (super) {
+ free(super);
+ super = NULL;
+ }
- dfd = open(devname, O_RDONLY, 0);
+ dfd = open(devname, O_RDONLY|O_EXCL, 0);
if (dfd < 0) {
if (inargv || verbose)
fprintf(stderr, Name ": cannot open device %s: %s\n",
fprintf(stderr, Name ": %s is not a block device.\n",
devname);
close(dfd);
- } else if (load_super(dfd, &super)) {
+ } else if (!tss && (tss = guess_super(dfd, NULL)) == NULL) {
+ if (inargv || verbose)
+ fprintf(stderr, Name ": no recogniseable superblock\n");
+ } else if (tss->load_super(dfd, &super, NULL)) {
if (inargv || verbose)
fprintf( stderr, Name ": no RAID superblock on %s\n",
devname);
close(dfd);
} else {
- havesuper =1;
- uuid_from_super(this_uuid, &super);
+ tss->getinfo_super(&info, super);
close(dfd);
}
if (ident->uuid_set &&
- (!havesuper || same_uuid(this_uuid, ident->uuid)==0)) {
+ (!super || same_uuid(info.uuid, ident->uuid)==0)) {
if (inargv || verbose)
fprintf(stderr, Name ": %s has wrong uuid.\n",
devname);
continue;
}
- if (ident->super_minor >= 0 &&
- (!havesuper || ident->super_minor != super.md_minor)) {
+ if (ident->super_minor != UnSet &&
+ (!super || ident->super_minor != info.array.md_minor)) {
if (inargv || verbose)
fprintf(stderr, Name ": %s has wrong super-minor.\n",
devname);
continue;
}
- if (ident->level != -10 &&
- (!havesuper|| ident->level != super.level)) {
+ if (ident->level != UnSet &&
+ (!super|| ident->level != info.array.level)) {
if (inargv || verbose)
fprintf(stderr, Name ": %s has wrong raid level.\n",
devname);
continue;
}
- if (ident->raid_disks != -1 &&
- (!havesuper || ident->raid_disks!= super.raid_disks)) {
+ if (ident->raid_disks != UnSet &&
+ (!super || ident->raid_disks!= info.array.raid_disks)) {
if (inargv || verbose)
fprintf(stderr, Name ": %s requires wrong number of drives.\n",
devname);
* then we cannot continue
*/
- if (!havesuper) {
+ if (!super) {
fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
devname);
+ free(first_super);
return 1;
}
- if (compare_super(&first_super, &super)) {
+ ss = tss; /* commit to this format, if haven't already */
+ if (ss->compare_super(&first_super, super)) {
fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
devname);
+ free(super);
+ free(first_super);
return 1;
}
- if (devcnt >= MD_SB_DISKS) {
- fprintf(stderr, Name ": ouch - too many devices appear to be in this array. Ignoring %s\n",
- devname);
- continue;
+ /* looks like a good enough match to update the super block if needed */
+ if (update) {
+ /* prepare useful information in info structures */
+ struct stat stb2;
+ fstat(mdfd, &stb2);
+ info.array.md_minor = minor(stb2.st_rdev);
+
+ ss->update_super(&info, super, update, devname, verbose);
+
+ dfd = open(devname, O_RDWR|O_EXCL, 0);
+ if (dfd < 0)
+ fprintf(stderr, Name ": Cannot open %s for superblock update\n",
+ devname);
+ else if (ss->store_super(dfd, super))
+ fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
+ devname);
+ if (dfd >= 0)
+ close(dfd);
}
+
if (verbose)
fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
- devname, mddev, super.this_disk.raid_disk);
+ devname, mddev, info.disk.raid_disk);
devices[devcnt].devname = devname;
- devices[devcnt].major = MAJOR(stb.st_rdev);
- devices[devcnt].minor = MINOR(stb.st_rdev);
- devices[devcnt].events = md_event(&super);
- devices[devcnt].utime = super.utime;
- devices[devcnt].raid_disk = super.this_disk.raid_disk;
+ devices[devcnt].major = major(stb.st_rdev);
+ devices[devcnt].minor = minor(stb.st_rdev);
+ devices[devcnt].oldmajor = info.disk.major;
+ devices[devcnt].oldminor = info.disk.minor;
+ devices[devcnt].events = info.events;
+ devices[devcnt].raid_disk = info.disk.raid_disk;
devices[devcnt].uptodate = 0;
+ devices[devcnt].state = info.disk.state;
if (most_recent < devcnt) {
if (devices[devcnt].events
> devices[most_recent].events)
most_recent = devcnt;
}
- i = devices[devcnt].raid_disk;
- if (i>=0 && i < MD_SB_DISKS)
+ if (info.array.level == -4)
+ /* with multipath, the raid_disk from the superblock is meaningless */
+ i = devcnt;
+ else
+ i = devices[devcnt].raid_disk;
+ if (i < 10000) {
+ if (i >= bestcnt) {
+ unsigned int newbestcnt = i+10;
+ int *newbest = malloc(sizeof(int)*newbestcnt);
+ unsigned int c;
+ for (c=0; c < newbestcnt; c++)
+ if (c < bestcnt)
+ newbest[c] = best[c];
+ else
+ newbest[c] = -1;
+ if (best)free(best);
+ best = newbest;
+ bestcnt = newbestcnt;
+ }
if (best[i] == -1
|| devices[best[i]].events < devices[devcnt].events)
best[i] = devcnt;
-
+ }
devcnt++;
}
+ if (super)
+ free(super);
+ super = NULL;
+
if (devcnt == 0) {
fprintf(stderr, Name ": no devices found for %s\n",
mddev);
+ free(first_super);
return 1;
}
+
+ ss->getinfo_super(&info, first_super);
+
/* now we have some devices that might be suitable.
* I wonder how many
*/
okcnt = 0;
sparecnt=0;
- for (i=0; i< MD_SB_DISKS;i++) {
+ for (i=0; i< bestcnt ;i++) {
int j = best[i];
int event_margin = !force;
if (j < 0) continue;
+ /* note: we ignore error flags in multipath arrays
+ * as they don't make sense
+ */
+ if (info.array.level != -4)
+ if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
+ if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
+ sparecnt++;
+ continue;
+ }
if (devices[j].events+event_margin >=
devices[most_recent].events) {
devices[j].uptodate = 1;
- if (i < first_super.raid_disks)
+ if (i < info.array.raid_disks)
okcnt++;
else
sparecnt++;
}
}
- while (force && !enough(first_super.level, first_super.raid_disks, okcnt)) {
+ while (force && !enough(info.array.level, info.array.raid_disks, okcnt)) {
/* Choose the newest best drive which is
* not up-to-date, update the superblock
* and add it.
*/
int fd;
chosen_drive = -1;
- for (i=0; i<first_super.raid_disks; i++) {
+ for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
int j = best[i];
if (j>=0 &&
!devices[j].uptodate &&
devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
(int)(devices[chosen_drive].events),
(int)(devices[most_recent].events));
- fd = open(devices[chosen_drive].devname, O_RDWR);
+ fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
if (fd < 0) {
fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
devices[chosen_drive].devname);
devices[chosen_drive].events = 0;
continue;
}
- if (load_super(fd, &super)) {
+ if (ss->load_super(fd, &super, NULL)) {
close(fd);
fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
devices[chosen_drive].devname);
devices[chosen_drive].events = 0;
continue;
}
- super.events_hi = (devices[most_recent].events>>32)&0xFFFFFFFF;
- super.events_lo = (devices[most_recent].events)&0xFFFFFFFF;
- super.sb_csum = calc_sb_csum(&super);
-/*DRYRUN*/ if (store_super(fd, &super)) {
+ info.events = devices[most_recent].events;
+ ss->update_super(&info, super, "force", devices[chosen_drive].devname, verbose);
+
+ if (ss->store_super(fd, super)) {
close(fd);
fprintf(stderr, Name ": Could not re-write superblock on %s\n",
devices[chosen_drive].devname);
devices[chosen_drive].events = 0;
+ free(super);
continue;
}
close(fd);
devices[chosen_drive].events = devices[most_recent].events;
devices[chosen_drive].uptodate = 1;
okcnt++;
+ free(super);
}
/* Now we want to look at the superblock which the kernel will base things on
* superblock.
*/
chosen_drive = -1;
- for (i=0; chosen_drive < 0 && i<MD_SB_DISKS; i++) {
+ super = NULL;
+ for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
int j = best[i];
int fd;
+
if (j<0)
continue;
if (!devices[j].uptodate)
continue;
chosen_drive = j;
- if ((fd=open(devices[j].devname, O_RDONLY))< 0) {
+ if ((fd=open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
fprintf(stderr, Name ": Cannot open %s: %s\n",
devices[j].devname, strerror(errno));
return 1;
}
- if (load_super(fd, &super)) {
+ if (ss->load_super(fd, &super, NULL)) {
close(fd);
fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
devices[j].devname);
}
close(fd);
}
-
- for (i=0; i<MD_SB_DISKS; i++) {
+ if (super == NULL) {
+ fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
+ return 1;
+ }
+ ss->getinfo_super(&info, super);
+ for (i=0; i<bestcnt; i++) {
int j = best[i];
- int active_sync = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
+ unsigned int desired_state;
+
+ if (i < info.array.raid_disks)
+ desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
+ else
+ desired_state = 0;
+
if (j<0)
continue;
if (!devices[j].uptodate)
continue;
- if (devices[j].major != super.disks[j].major ||
- devices[j].minor != super.disks[j].minor) {
- change |= 1;
- super.disks[j].major = devices[j].major;
- super.disks[j].minor = devices[j].minor;
- }
+ info.disk.number = i;
+ info.disk.state = desired_state;
+
if (devices[j].uptodate &&
- (super.disks[i].state != active_sync)) {
+ ss->update_super(&info, super, "assemble", NULL, 0)) {
if (force) {
fprintf(stderr, Name ": "
"clearing FAULTY flag for device %d in %s for %s\n",
j, mddev, devices[j].devname);
- super.disks[i].state = active_sync;
- change |= 2;
+ change = 1;
} else {
fprintf(stderr, Name ": "
- "device %d in %s is marked faulty in superblock, but %s seems ok\n",
+ "device %d in %s has wrong state in superblock, but %s seems ok\n",
i, mddev, devices[j].devname);
}
}
+#if 0
if (!devices[j].uptodate &&
!(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
i, mddev);
}
+#endif
+ }
+ if (force && okcnt == info.array.raid_disks-1) {
+ /* FIXME check event count */
+ change += ss->update_super(&info, super, "force",
+ devices[chosen_drive].devname, 0);
}
- if ((force && (change & 2))
- || (old_linux && (change & 1))) {
+ if (change) {
int fd;
- super.sb_csum = calc_sb_csum(&super);
- fd = open(devices[chosen_drive].devname, O_RDWR);
+ fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
if (fd < 0) {
fprintf(stderr, Name ": Could open %s for write - cannot Assemble array.\n",
devices[chosen_drive].devname);
return 1;
}
- if (store_super(fd, &super)) {
+ if (ss->store_super(fd, super)) {
close(fd);
fprintf(stderr, Name ": Could not re-write superblock on %s\n",
devices[chosen_drive].devname);
return 1;
}
close(fd);
- change = 0;
}
+ /* count number of in-sync devices according to the superblock.
+ * We must have this number to start the array without -s or -R
+ */
+ req_cnt = info.array.working_disks;
+
/* Almost ready to actually *do* something */
if (!old_linux) {
if (ioctl(mdfd, SET_ARRAY_INFO, NULL) != 0) {
return 1;
}
/* First, add the raid disks, but add the chosen one last */
- for (i=0; i<=MD_SB_DISKS; i++) {
+ for (i=0; i<= bestcnt; i++) {
int j;
- if (i < MD_SB_DISKS) {
+ if (i < bestcnt) {
j = best[i];
if (j == chosen_drive)
continue;
} else
j = chosen_drive;
- if (j >= 0 && devices[j].uptodate) {
+ if (j >= 0 /* && devices[j].uptodate */) {
mdu_disk_info_t disk;
memset(&disk, 0, sizeof(disk));
disk.major = devices[j].major;
devices[j].devname,
mddev,
strerror(errno));
- if (i < first_super.raid_disks)
+ if (i < info.array.raid_disks || i == bestcnt)
okcnt--;
else
sparecnt--;
} else if (verbose)
fprintf(stderr, Name ": added %s to %s as %d\n",
devices[j].devname, mddev, devices[j].raid_disk);
- } else if (verbose && i < first_super.raid_disks)
+ } else if (verbose && i < info.array.raid_disks)
fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
i, mddev);
}
if (runstop == 1 ||
(runstop == 0 &&
- ( first_super.raid_disks == okcnt
- || start_partial_ok && enough(first_super.level, first_super.raid_disks, okcnt))
- )) {
+ ( enough(info.array.level, info.array.raid_disks, okcnt) &&
+ (okcnt >= req_cnt || start_partial_ok)
+ ))) {
if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
fprintf(stderr, Name ": %s has been started with %d drive%s",
mddev, okcnt, okcnt==1?"":"s");
+ if (okcnt < info.array.raid_disks)
+ fprintf(stderr, " (out of %d)", info.array.raid_disks);
if (sparecnt)
fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
fprintf(stderr, ".\n");
mddev, okcnt, okcnt==1?"":"s");
return 0;
}
- fprintf(stderr, Name ": %s assembled from %d drive%s - not enough to start it (use --run to insist).\n",
- mddev, okcnt, okcnt==1?"":"s");
+ fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
+ if (sparecnt)
+ fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
+ if (!enough(info.array.level, info.array.raid_disks, okcnt))
+ fprintf(stderr, " - not enough to start the array.\n");
+ else {
+ if (req_cnt == info.array.raid_disks)
+ fprintf(stderr, " - need all %d to start it", req_cnt);
+ else
+ fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
+ fprintf(stderr, " (use --run to insist).\n");
+ }
return 1;
} else {
/* The "chosen_drive" is a good choice, and if necessary, the superblock has
* so we can just start the array
*/
unsigned long dev;
- dev = MKDEV(devices[chosen_drive].major,
+ dev = makedev(devices[chosen_drive].major,
devices[chosen_drive].minor);
if (ioctl(mdfd, START_ARRAY, dev)) {
fprintf(stderr, Name ": Cannot start array: %s\n",
}
}
+ return 0;
}
projects / thirdparty / mdadm.git / blobdiff