*/
#include "mdadm.h"
+#include <dirent.h>
+#include <ctype.h>
static int count_active(struct supertype *st, int mdfd, char **availp,
struct mdinfo *info);
static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
int number, __u64 events, int verbose,
char *array_name);
+static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
+ struct supertype *st, int verbose);
int Incremental(char *devname, int verbose, int runstop,
struct supertype *st, char *homehost, int require_homehost,
* start the array (auto-readonly).
*/
struct stat stb;
- struct mdinfo info;
+ struct mdinfo info, dinfo;
struct mddev_ident_s *array_list, *match;
char chosen_name[1024];
- int rv;
+ int rv = 1;
struct map_ent *mp, *map = NULL;
- int dfd, mdfd;
+ int dfd = -1, mdfd = -1;
char *avail;
int active_disks;
int trustworthy = FOREIGN;
char *name_to_use;
mdu_array_info_t ainf;
+ struct dev_policy *policy = NULL;
struct createinfo *ci = conf_get_create_info();
fprintf(stderr, Name
": %s not permitted by mdadm.conf.\n",
devname);
- return 1;
+ goto out;
}
/* 2/ Find metadata, reject if none appropriate (check
if (verbose >= 0)
fprintf(stderr, Name ": cannot open %s: %s.\n",
devname, strerror(errno));
- return 1;
+ goto out;
}
if (fstat(dfd, &stb) < 0) {
if (verbose >= 0)
fprintf(stderr, Name ": fstat failed for %s: %s.\n",
devname, strerror(errno));
- close(dfd);
- return 1;
+ goto out;
}
if ((stb.st_mode & S_IFMT) != S_IFBLK) {
if (verbose >= 0)
fprintf(stderr, Name ": %s is not a block device.\n",
devname);
- close(dfd);
- return 1;
+ goto out;
}
+ dinfo.disk.major = major(stb.st_rdev);
+ dinfo.disk.minor = minor(stb.st_rdev);
+
+ policy = disk_policy(&dinfo);
+
if (st == NULL && (st = guess_super(dfd)) == NULL) {
if (verbose >= 0)
fprintf(stderr, Name
": no recognisable superblock on %s.\n",
devname);
- close(dfd);
- return 1;
+ rv = try_spare(devname, &dfd, policy, st, verbose);
+ goto out;
}
- if (st->ss->load_super(st, dfd, NULL)) {
+ if (st->ss->compare_super == NULL ||
+ st->ss->load_super(st, dfd, NULL)) {
if (verbose >= 0)
fprintf(stderr, Name ": no RAID superblock on %s.\n",
devname);
- close(dfd);
- return 1;
+ rv = try_spare(devname, &dfd, policy, st, verbose);
+ free(st);
+ goto out;
}
- close (dfd);
+ close (dfd); dfd = -1;
memset(&info, 0, sizeof(info));
st->ss->getinfo_super(st, &info);
fprintf(stderr, Name
": multiple lines in mdadm.conf match\n");
}
- return 2;
+ rv = 2;
+ goto out;
}
match = array_list;
}
fprintf(stderr, Name ": array containing %s is explicitly"
" ignored by mdadm.conf\n",
devname);
- return 1;
+ goto out;
}
/* 3a/ if not, check for homehost match. If no match, continue
": %s has metadata type %s for which "
"auto-assembly is disabled\n",
devname, st->ss->name);
- return 1;
+ goto out;
}
if (trustworthy == LOCAL_ANY)
trustworthy = LOCAL;
else {
if (verbose)
fprintf(stderr, Name ": not enough devices to start the container\n");
- return 1;
+ rv = 0;
+ goto out;
}
/* This is a pre-built container array, so we do something
* rather different.
*/
- return Incremental_container(st, devname, verbose, runstop,
+ rv = Incremental_container(st, devname, verbose, runstop,
autof, trustworthy);
+ goto out;
}
name_to_use = info.name;
/* 4/ Check if array exists.
*/
- map_lock(&map);
+ if (map_lock(&map))
+ fprintf(stderr, Name ": failed to get exclusive lock on "
+ "mapfile\n");
mp = map_by_uuid(&map, info.uuid);
if (mp)
mdfd = open_dev(mp->devnum);
if (mdfd < 0) {
struct mdinfo *sra;
- struct mdinfo dinfo;
/* Couldn't find an existing array, maybe make a new one */
mdfd = create_mddev(match ? match->devname : NULL,
name_to_use, autof, trustworthy, chosen_name);
if (mdfd < 0)
- return 1;
+ goto out;
sysfs_init(&info, mdfd, 0);
if (set_array_info(mdfd, st, &info) != 0) {
fprintf(stderr, Name ": failed to set array info for %s: %s\n",
chosen_name, strerror(errno));
- close(mdfd);
- return 2;
+ rv = 2;
+ goto out;
}
dinfo = info;
fprintf(stderr, Name ": failed to add %s to %s: %s.\n",
devname, chosen_name, strerror(errno));
ioctl(mdfd, STOP_ARRAY, 0);
- close(mdfd);
- return 2;
+ rv = 2;
+ goto out;
}
sra = sysfs_read(mdfd, fd2devnum(mdfd), GET_DEVS);
if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
fprintf(stderr, Name
": You have an old buggy kernel which cannot support\n"
" --incremental reliably. Aborting.\n");
- close(mdfd);
sysfs_free(sra);
- return 2;
+ rv = 2;
+ goto out;
}
info.array.working_disks = 1;
sysfs_free(sra);
else
strcpy(chosen_name, devnum2devname(mp->devnum));
- /* It is generally not OK to add drives to a running array
- * as they are probably missing because they failed.
- * However if runstop is 1, then the array was possibly
- * started early and our best be is to add this anyway.
- * It would probably be good to allow explicit policy
- * statement about this.
+ /* It is generally not OK to add non-spare drives to a
+ * running array as they are probably missing because
+ * they failed. However if runstop is 1, then the
+ * array was possibly started early and our best bet is
+ * to add this anyway.
+ * Also if action policy is re-add or better we allow
+ * re-add
*/
- if (runstop < 1) {
- if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
+ if ((info.disk.state & (1<<MD_DISK_SYNC)) != 0
+ && ! policy_action_allows(policy, st->ss->name,
+ act_re_add)
+ && runstop < 1) {
+ int active = 0;
+
+ if (st->ss->external) {
+ char *devname = devnum2devname(fd2devnum(mdfd));
+
+ active = devname && is_container_active(devname);
+ free(devname);
+ } else if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0)
+ active = 1;
+ if (active) {
fprintf(stderr, Name
": not adding %s to active array (without --run) %s\n",
devname, chosen_name);
- close(mdfd);
- return 2;
+ rv = 2;
+ goto out;
}
}
sra = sysfs_read(mdfd, fd2devnum(mdfd), (GET_DEVS | GET_STATE));
-
+ if (!sra) {
+ rv = 2;
+ goto out;
+ }
if (sra->devs) {
sprintf(dn, "%d:%d", sra->devs->disk.major,
sra->devs->disk.minor);
": metadata mismatch between %s and "
"chosen array %s\n",
devname, chosen_name);
- close(mdfd);
close(dfd2);
- return 2;
+ rv = 2;
+ goto out;
}
close(dfd2);
memset(&info2, 0, sizeof(info2));
fprintf(stderr, Name
": unexpected difference between %s and %s.\n",
chosen_name, devname);
- close(mdfd);
- return 2;
+ rv = 2;
+ goto out;
}
}
info2.disk.major = major(stb.st_rdev);
if (err < 0) {
fprintf(stderr, Name ": failed to add %s to %s: %s.\n",
devname, chosen_name, strerror(errno));
- close(mdfd);
- return 2;
+ rv = 2;
+ goto out;
}
info.array.working_disks = 0;
for (d = sra->devs; d; d=d->next)
": %s attached to %s, not enough to start (%d).\n",
devname, chosen_name, active_disks);
map_unlock(&map);
- close(mdfd);
- return 0;
+ rv = 0;
+ goto out;
}
free(avail);
fprintf(stderr, Name
": %s attached to %s which is already active.\n",
devname, chosen_name);
- close(mdfd);
map_unlock(&map);
- return 0;
+ rv = 0;
+ goto out;
}
map_unlock(&map);
if (runstop > 0 || active_disks >= info.array.working_disks) {
- struct mdinfo *sra;
+ struct mdinfo *sra, *dsk;
/* Let's try to start it */
if (match && match->bitmap_file) {
int bmfd = open(match->bitmap_file, O_RDWR);
fprintf(stderr, Name
": Could not open bitmap file %s.\n",
match->bitmap_file);
- close(mdfd);
- return 1;
+ goto out;
}
if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
close(bmfd);
fprintf(stderr, Name
": Failed to set bitmapfile for %s.\n",
chosen_name);
- close(mdfd);
- return 1;
+ goto out;
}
close(bmfd);
}
- sra = sysfs_read(mdfd, fd2devnum(mdfd), 0);
+ /* GET_* needed so add_disk works below */
+ sra = sysfs_read(mdfd, fd2devnum(mdfd),
+ GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE);
if ((sra == NULL || active_disks >= info.array.working_disks)
&& trustworthy != FOREIGN)
rv = ioctl(mdfd, RUN_ARRAY, NULL);
if (rv == 0) {
if (verbose >= 0)
fprintf(stderr, Name
- ": %s attached to %s, which has been started.\n",
+ ": %s attached to %s, which has been started.\n",
devname, chosen_name);
rv = 0;
wait_for(chosen_name, mdfd);
+ /* We just started the array, so some devices
+ * might have been evicted from the array
+ * because their event counts were too old.
+ * If the action=re-add policy is in-force for
+ * those devices we should re-add them now.
+ */
+ for (dsk = sra->devs; dsk ; dsk = dsk->next) {
+ if (disk_action_allows(dsk, st->ss->name, act_re_add) &&
+ add_disk(mdfd, st, sra, dsk) == 0)
+ fprintf(stderr, Name
+ ": %s re-added to %s\n",
+ dsk->sys_name, chosen_name);
+ }
} else {
fprintf(stderr, Name
": %s attached to %s, but failed to start: %s.\n",
devname, chosen_name);
rv = 0;
}
- close(mdfd);
+out:
+ if (dfd >= 0)
+ close(dfd);
+ if (mdfd >= 0)
+ close(mdfd);
+ if (policy)
+ dev_policy_free(policy);
return rv;
}
struct mdinfo *sra = sysfs_read(mdfd, -1, GET_DEVS | GET_STATE);
char *avail = NULL;
+ if (!sra)
+ return 0;
+
for (d = sra->devs ; d ; d = d->next) {
char dn[30];
int dfd;
return cnt + cnt1;
}
+static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
+ struct supertype *st, int verbose)
+{
+ /* This device doesn't have any md metadata
+ * If it is 'bare' and theh device policy allows 'spare' look for
+ * an array or container to attach it to.
+ * If st is set, then only arrays of that type are considered
+ * Return 0 on success, or some exit code on failure, probably 1.
+ */
+ int rv = -1;
+ struct stat stb;
+ struct map_ent *mp, *map = NULL;
+ struct mdinfo *chosen = NULL;
+ int dfd = *dfdp;
+
+ if (fstat(dfd, &stb) != 0)
+ return 1;
+
+ /*
+ * Now we need to find a suitable array to add this to.
+ * We only accept arrays that:
+ * - match 'st'
+ * - are in the same domains as the device
+ * - are of an size for which the device will be useful
+ * and we choose the one that is the most degraded
+ */
+
+ if (map_lock(&map)) {
+ fprintf(stderr, Name ": failed to get exclusive lock on "
+ "mapfile\n");
+ return 1;
+ }
+ for (mp = map ; mp ; mp = mp->next) {
+ struct supertype *st2;
+ struct domainlist *dl = NULL;
+ struct mdinfo *sra;
+ unsigned long long devsize;
+
+ if (is_subarray(mp->metadata))
+ continue;
+ if (st) {
+ st2 = st->ss->match_metadata_desc(mp->metadata);
+ if (!st2 ||
+ (st->minor_version >= 0 &&
+ st->minor_version != st2->minor_version)) {
+ if (verbose > 1)
+ fprintf(stderr, Name ": not adding %s to %s as metadata type doesn't match\n",
+ devname, mp->path);
+ free(st2);
+ continue;
+ }
+ free(st2);
+ }
+ sra = sysfs_read(-1, mp->devnum,
+ GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
+ GET_DEGRADED|GET_COMPONENT|GET_VERSION);
+ if (!sra) {
+ /* Probably a container - no degraded info */
+ sra = sysfs_read(-1, mp->devnum,
+ GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
+ GET_COMPONENT|GET_VERSION);
+ if (sra)
+ sra->array.failed_disks = 0;
+ }
+ if (!sra)
+ continue;
+ if (st == NULL) {
+ int i;
+ st2 = NULL;
+ for(i=0; !st2 && superlist[i]; i++)
+ st2 = superlist[i]->match_metadata_desc(
+ sra->text_version);
+ } else
+ st2 = st;
+ get_dev_size(dfd, NULL, &devsize);
+ if (st2->ss->avail_size(st2, devsize) < sra->component_size) {
+ if (verbose > 1)
+ fprintf(stderr, Name ": not adding %s to %s as it is too small\n",
+ devname, mp->path);
+ goto next;
+ }
+ dl = domain_from_array(sra, st2->ss->name);
+ if (!domain_test(dl, pol, st2->ss->name)) {
+ /* domain test fails */
+ if (verbose > 1)
+ fprintf(stderr, Name ": not adding %s to %s as it is not in a compatible domain\n",
+ devname, mp->path);
+
+ goto next;
+ }
+ /* all tests passed, OK to add to this array */
+ if (!chosen) {
+ chosen = sra;
+ sra = NULL;
+ } else if (chosen->array.failed_disks < sra->array.failed_disks) {
+ sysfs_free(chosen);
+ chosen = sra;
+ sra = NULL;
+ }
+ next:
+ if (sra)
+ sysfs_free(sra);
+ if (st != st2)
+ free(st2);
+ if (dl)
+ domain_free(dl);
+ }
+ if (chosen) {
+ /* add current device to chosen array as a spare */
+ int mdfd = open_dev(devname2devnum(chosen->sys_name));
+ if (mdfd >= 0) {
+ struct mddev_dev_s devlist;
+ char devname[20];
+ devlist.next = NULL;
+ devlist.used = 0;
+ devlist.re_add = 0;
+ devlist.writemostly = 0;
+ devlist.devname = devname;
+ sprintf(devname, "%d:%d", major(stb.st_rdev),
+ minor(stb.st_rdev));
+ devlist.disposition = 'a';
+ close(dfd);
+ *dfdp = -1;
+ rv = Manage_subdevs(chosen->sys_name, mdfd, &devlist,
+ -1, 0);
+ close(mdfd);
+ }
+ if (verbose > 0) {
+ if (rv == 0)
+ fprintf(stderr, Name ": added %s as spare for %s\n",
+ devname, chosen->sys_name);
+ else
+ fprintf(stderr, Name ": failed to add %s as spare for %s\n",
+ devname, chosen->sys_name);
+ }
+ sysfs_free(chosen);
+ }
+ return rv ? 0 : 1;
+}
+
+static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
+ struct supertype *st, int verbose)
+{
+ /* we know that at least one partition virtual-metadata is
+ * allowed to incorporate spares like this device. We need to
+ * find a suitable device to copy partition information from.
+ *
+ * Getting a list of all disk (not partition) devices is
+ * slightly non-trivial. We could look at /sys/block, but
+ * that is theoretically due to be removed. Maybe best to use
+ * /dev/disk/by-path/?* and ignore names ending '-partNN' as
+ * we depend on this directory of 'path' info. But that fails
+ * to find loop devices and probably others. Maybe don't
+ * worry about that, they aren't the real target.
+ *
+ * So: check things in /dev/disk/by-path to see if they are in
+ * a compatible domain, then load the partition table and see
+ * if it is OK for the new device, and choose the largest
+ * partition table that fits.
+ */
+ DIR *dir;
+ struct dirent *de;
+ char *chosen = NULL;
+ unsigned long long chosen_size;
+ struct supertype *chosen_st = NULL;
+ int fd;
+
+ dir = opendir("/dev/disk/by-path");
+ if (!dir)
+ return 1;
+ while ((de = readdir(dir)) != NULL) {
+ char *ep;
+ struct dev_policy *pol2 = NULL;
+ struct domainlist *domlist = NULL;
+ int fd = -1;
+ struct mdinfo info;
+ struct supertype *st2 = NULL;
+ char *devname = NULL;
+ unsigned long long devsectors;
+
+ if (de->d_ino == 0 ||
+ de->d_name[0] == '.' ||
+ (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
+ goto next;
+
+ ep = de->d_name + strlen(de->d_name);
+ while (ep > de->d_name &&
+ isdigit(ep[-1]))
+ ep--;
+ if (ep > de->d_name + 5 &&
+ strncmp(ep-5, "-part", 5) == 0)
+ /* This is a partition - skip it */
+ goto next;
+
+ pol2 = path_policy(de->d_name, type_disk);
+
+ domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
+ if (domain_test(domlist, pol, st ? st->ss->name : NULL) == 0)
+ /* new device is incompatible with this device. */
+ goto next;
+
+ domain_free(domlist);
+ domlist = NULL;
+
+ asprintf(&devname, "/dev/disk/by-path/%s", de->d_name);
+ fd = open(devname, O_RDONLY);
+ if (fd < 0)
+ goto next;
+ if (get_dev_size(fd, devname, &devsectors) == 0)
+ goto next;
+ devsectors >>= 9;
+
+ if (st)
+ st2 = dup_super(st);
+ else
+ st2 = guess_super_type(fd, guess_partitions);
+ if (st2 == NULL ||
+ st2->ss->load_super(st2, fd, NULL) < 0)
+ goto next;
+
+ if (!st) {
+ /* Check domain policy again, this time referring to metadata */
+ domain_merge(&domlist, pol2, st2->ss->name);
+ if (domain_test(domlist, pol, st2->ss->name) == 0)
+ /* Incompatible devices for this metadata type */
+ goto next;
+ }
+
+ st2->ss->getinfo_super(st2, &info);
+ if (info.component_size > devsectors)
+ /* This partitioning doesn't fit in the device */
+ goto next;
+
+ /* This is an acceptable device to copy partition
+ * metadata from. We could just stop here, but I
+ * think I want to keep looking incase a larger
+ * metadata which makes better use of the device can
+ * be found.
+ */
+ if (chosen == NULL ||
+ chosen_size < info.component_size) {
+ chosen_size = info.component_size;
+ free(chosen);
+ chosen = devname;
+ devname = NULL;
+ if (chosen_st) {
+ chosen_st->ss->free_super(chosen_st);
+ free(chosen_st);
+ }
+ chosen_st = st2;
+ st2 = NULL;
+ }
+
+ next:
+ free(devname);
+ domain_free(domlist);
+ dev_policy_free(pol2);
+ if (st2)
+ st2->ss->free_super(st2);
+ free(st2);
+
+ if (fd >= 0)
+ close(fd);
+ }
+
+ if (!chosen)
+ return 1;
+
+ /* 'chosen' is the best device we can find. Let's write its
+ * metadata to devname dfd is read-only so don't use that
+ */
+ fd = open(devname, O_RDWR);
+ if (fd >= 0) {
+ chosen_st->ss->store_super(chosen_st, fd);
+ close(fd);
+ }
+ free(chosen);
+ chosen_st->ss->free_super(chosen_st);
+ free(chosen_st);
+ return 0;
+}
+
+
+/* adding a spare to a regular array is quite different from adding one to
+ * a set-of-partitions virtual array.
+ * This function determines which is worth trying and tries as appropriate.
+ * Arrays are given priority over partitions.
+ */
+static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
+ struct supertype *st, int verbose)
+{
+ int i;
+ int rv;
+ int arrays_ok = 0;
+ int partitions_ok = 0;
+ char bufpad[4096 + 4096];
+ char *buf = (char*)(((long)bufpad + 4096) & ~4095);
+ int dfd = *dfdp;
+
+ /* Can only add a spare if device has at least one domains */
+ if (pol_find(pol, pol_domain) == NULL)
+ return 1;
+ /* And only if some action allows spares */
+ if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
+ return 1;
+
+ /* Now check if the device is bare - we don't add non-bare devices
+ * yet even if action=-spare
+ */
+
+ if (lseek(dfd, 0, SEEK_SET) != 0 ||
+ read(dfd, buf, 4096) != 4096) {
+ not_bare:
+ if (verbose > 1)
+ fprintf(stderr, Name ": %s is not bare, so not considering as a spare\n",
+ devname);
+ return 1;
+ }
+ if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
+ goto not_bare;
+ if (memcmp(buf, buf+1, 4095) != 0)
+ goto not_bare;
+
+ /* OK, first 4K appear blank, try the end. */
+ if (lseek(dfd, -4096, SEEK_END) < 0 ||
+ read(dfd, buf, 4096) != 4096)
+ goto not_bare;
+
+ if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
+ goto not_bare;
+ if (memcmp(buf, buf+1, 4095) != 0)
+ goto not_bare;
+
+ /* This device passes our test for 'is bare'.
+ * Let's see what policy allows for such things.
+ */
+ if (st) {
+ /* just try try 'array' or 'partition' based on this metadata */
+ if (st->ss->add_to_super)
+ return array_try_spare(devname, dfdp, pol,
+ st, verbose);
+ else
+ return partition_try_spare(devname, dfdp, pol,
+ st, verbose);
+ }
+ /* Now see which metadata type support spare */
+ for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
+ if (superlist[i]->add_to_super && !arrays_ok &&
+ policy_action_allows(pol, superlist[i]->name, act_spare))
+ arrays_ok = 1;
+ if (superlist[i]->add_to_super == NULL && !partitions_ok &&
+ policy_action_allows(pol, superlist[i]->name, act_spare))
+ partitions_ok = 1;
+ }
+ rv = 0;
+ if (arrays_ok)
+ rv = array_try_spare(devname, dfdp, pol, st, verbose);
+ if (rv == 0 && partitions_ok)
+ rv = partition_try_spare(devname, dfdp, pol, st, verbose);
+ return rv;
+}
+
int IncrementalScan(int verbose)
{
/* look at every device listed in the 'map' file.
struct mdinfo *ra;
struct map_ent *map = NULL;
- map_lock(&map);
+ if (map_lock(&map))
+ fprintf(stderr, Name ": failed to get exclusive lock on "
+ "mapfile\n");
for (ra = list ; ra ; ra = ra->next) {
int mdfd;
int IncrementalRemove(char *devname, int verbose)
{
int mdfd;
+ int rv;
struct mdstat_ent *ent;
struct mddev_dev_s devlist;
memset(&devlist, 0, sizeof(devlist));
devlist.devname = devname;
devlist.disposition = 'f';
- Manage_subdevs(ent->dev, mdfd, &devlist, verbose);
+ Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0);
devlist.disposition = 'r';
- return Manage_subdevs(ent->dev, mdfd, &devlist, verbose);
+ rv = Manage_subdevs(ent->dev, mdfd, &devlist, verbose, 0);
+ close(mdfd);
+ return rv;
}
projects / thirdparty / mdadm.git / blobdiff