From: Czarnowska, Anna
Sent: Monday, July 05, 2010 11:38 AM
To: Neil Brown
Cc: linux-raid@xxxxxxxxxxxxxxx; Czarnowska, Anna; Hawrylewicz Czarnowski, Przemyslaw; Labun, Marcin; Neubauer, Wojciech; Williams, Dan J; Ciechanowski, Ed; dledford@xxxxxxxxxx
Subject: [PATCH 23/33] Monitor: Spare sharing with domain/subset support
Works for both native and external metadata.
Moves spares between arrays/containers with matching domain and subset to use for rebuild.
Signed-off-by: Anna Czarnowska <anna.czarnowska@xxxxxxxxx>
---
Monitor.c | 209 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++--
1 files changed, 202 insertions(+), 7 deletions(-)
diff --git a/Monitor.c b/Monitor.c
index ed57af0..11f3758 100644
--- a/Monitor.c
+++ b/Monitor.c
@@ -49,6 +49,7 @@ struct state {
int devstate[MaxDisks];
int devid[MaxDisks];
int percent;
+ unsigned long long min_size;
char *metadata_version;
struct state *next;
struct state *volumes;
@@ -56,6 +57,9 @@ struct state {
struct state *missing;
};
+static void spare_sharing(struct state *statelist, char *mailaddr,
+ char *mailfrom, char *alert_cmd, int dosyslog);
+
int Monitor(mddev_dev_t devlist,
char *mailaddr, char *alert_cmd,
int period, int daemonise, int scan, int oneshot, @@ -92,11 +96,10 @@ int Monitor(mddev_dev_t devlist,
* DeviceDisappeared
* Couldn't access a device which was previously visible
*
- * if we detect an array with active<raid and spare==0
- * we look at other arrays that have same spare-group
- * If we find one with active==raid and spare>0,
- * and if we can get_disk_info and find a name
- * Then we hot-remove and hot-add to the other array
+ * If we detect an array with active<raid and spare==0
+ * we look at other arrays that have a spare
+ * and are in the same domain and subset
+ * Then we hot-remove and hot-add to the other array
*
* If devlist is NULL, then we can monitor everything because --scan
* was given. We get an initial list from config file and add anything @@ -185,6 +188,9 @@ int Monitor(mddev_dev_t devlist,
st->parent = NULL;
st->volumes = NULL;
st->total = 0;
+ st->min_size = 0;
+ memset(st->devid, 0, MaxDisks*sizeof(int));
+ memset(st->devstate, 0, MaxDisks*sizeof(int));
if (mdlist->spare_group)
st->spare_group = strdup(mdlist->spare_group);
else
@@ -211,6 +217,9 @@ int Monitor(mddev_dev_t devlist,
st->parent = NULL;
st->volumes = NULL;
st->total = 0;
+ st->min_size = 0;
+ memset(st->devid, 0, MaxDisks*sizeof(int));
+ memset(st->devstate, 0, MaxDisks*sizeof(int));
if (mdlist) {
st->expected_spares = mdlist->spare_disks;
if (mdlist->spare_group)
@@ -371,6 +380,7 @@ int Monitor(mddev_dev_t devlist,
int newstate=0;
int change;
char *dv = NULL;
+ unsigned long long dsize;
disc.number = i;
if (i > array.raid_disks + array.nr_disks) {
newstate = 0;
@@ -415,6 +425,17 @@ int Monitor(mddev_dev_t devlist,
}
st->devstate[i] = newstate;
st->devid[i] = makedev(disc.major, disc.minor);
+
+ /* for volumes only we get minimum disk size
+ * (only active disks) */
+ fd = open(dv, O_RDONLY);
+ if (dv && newstate & (1<<MD_DISK_ACTIVE) &&
+ array.raid_disks && fd >= 0 &&
+ get_dev_size(fd, dv, &dsize) &&
+ (st->min_size == 0 || dsize < st->min_size))
+ st->min_size = dsize;
+ if (fd >= 0)
+ close(fd);
}
st->active = array.active_disks;
st->working = array.working_disks;
@@ -506,6 +527,7 @@ int Monitor(mddev_dev_t devlist,
new_found = 1;
}
}
+ spare_sharing(statelist, mailaddr, mailfrom, alert_cmd, dosyslog);
/* If an array has active < raid && spare == 0 && spare_group != NULL
* Look for another array with spare > 0 and active == raid and same spare_group
* if found, choose a device and hotremove/hotadd @@ -577,8 +599,63 @@ int Monitor(mddev_dev_t devlist,
return 0;
}
+static int move_spare(struct state *st2, struct state *st1, int i, char *mailaddr,
+ char *mailfrom, char *alert_cmd, int dosyslog)
+
+{
+ struct mddev_dev_s devlist;
+ char devname[20];
+ int from_fd, to_fd;
+
+ if (!st1 || !st2 || st2->devid[i] == 0)
+ return 0;
+
+ from_fd = open(st2->devname, O_RDONLY);
+ if (from_fd < 0)
+ return 0;
+
+ to_fd = open(st1->devname, O_RDONLY);
+ if (to_fd < 0) {
+ close(from_fd);
+ return 0;
+ }
+
+ devlist.next = NULL;
+ devlist.used = 0;
+ devlist.re_add = 0;
+ devlist.writemostly = 0;
+ devlist.devname = devname;
+ char *dv = map_dev(major(st2->devid[i]), minor(st2->devid[i]), 1);
+ if (!dv) {
+ close(from_fd);
+ close(to_fd);
+ return 0;
+ }
+ snprintf(devname, 20, "%s", dv);
+
+ devlist.disposition = 'r';
+ if (Manage_subdevs(st2->devname, from_fd, &devlist, -1) == 0) {
+ devlist.disposition = 'a';
+ if (Manage_subdevs(st1->devname, to_fd, &devlist, -1) == 0) {
+ st2->devid[i] = 0;
+ ping_manager(st2->devname);
+ ping_manager(st1->devname);
+ alert("MoveSpare", st1->devname, st2->devname, mailaddr,
+ mailfrom, alert_cmd, dosyslog);
+ close(from_fd);
+ close(to_fd);
+ return 1;
+ } else {
+ Manage_subdevs(st2->devname, from_fd, &devlist, -1);
+ }
+ }
+ close(from_fd);
+ close(to_fd);
+ return 0;
+}
+
/* check if disk is used in donor array (native) or any volume in donor container (external)*/ -int check_disk_is_free(struct state *donor, int disk_idx, int ext)
+static int check_disk_is_free(struct state *donor, int disk_idx, int
+ext)
{
struct state *vol = NULL;
int vol_disk;
@@ -601,7 +678,7 @@ int check_disk_is_free(struct state *donor, int disk_idx, int ext)
return disk_idx;
}
-int get_disk_domain_and_subset(int devid, char *metadata_version, struct domain_ent **domain,
+static int get_disk_domain_and_subset(int devid, char
+*metadata_version, struct domain_ent **domain,
struct subset **subset, int fcheck) {
struct supertype *sty;
@@ -650,6 +727,124 @@ fail:
return 0;
}
+/* If an array has active < raid && spare == 0
+ * Look for another array/container with unused, unfailed spare
+ * and the same domain and subset
+ * if found, hotremove/hotadd the spare (to parent container in
+external) */ static void spare_sharing(struct state *statelist, char
+*mailaddr,
+ char *mailfrom, char *alert_cmd, int dosyslog) {
+ struct state *st, *stp, *vol, *st2 = NULL;
+ struct domain_ent *domain, *spare_domain;
+ struct subset *subset, *spare_subset;
+ int i, ext, found;
+ unsigned long long ssize;
+
+ for (st = statelist; st; st = st->next) {
+ if (st->err || st->active == st->raid || st->spare > 0)
+ continue;
+
+ found = 0;
+ ext = is_external(st->metadata_version);
+
+ /*
+ * for exernal metadata spare will be moved to parent container
+ */
+ if (ext) {
+ stp = st->parent;
+ if (!stp)
+ continue;
+ } else {
+ stp = st;
+ }
+ /* check if there is a spare in this array/parent container */
+ for (i = 0; i < stp->total; i++)
+ if ((check_disk_is_free(stp, i, ext) == i) &&
+ (!ext || !disk_faulty_from_id(stp->devid[i])))
+ break;
+ if (i < stp->total)
+ /* there is a spare in array/parent container,
+ * if it is big enough it was probably just added
+ * but mdmon has not started recovery yet
+ * we will not add any more spares for now */
+ if (dev_size_from_id(stp->devid[i], &ssize) &&
+ ssize >= st->min_size)
+ continue;
+
+ /* get any good disk from array to check domain and subset */
+ for (i = 0; i < st->total; i++)
+ if (st->devid[i] > 0 &&
+ (st->devstate[i] & 1<<MD_DISK_ACTIVE))
+ break;
+
+ if (i == st->total)
+ continue;
+
+ if (!get_disk_domain_and_subset(st->devid[i],
+ stp->metadata_version, &domain, &subset, 0))
+ continue;
+
+ /* search for an array/container with unused spare */
+ for (st2 = statelist; st2; st2 = st2->next) {
+ if (st2->err || st2 == stp)
+ continue;
+
+ if ((ext && st2->parent != NULL) ||
+ (strcmp(stp->metadata_version, st2->metadata_version)
+ != 0))
+ continue;
+
+ if (ext) {
+ /* if container has degraded volume
+ * we can't remove spares */
+ for (vol = st2->volumes; vol; vol = vol->volumes)
+ if (vol->active < vol->raid)
+ break;
+ if (vol)
+ continue;
+ } else {
+ if (st2->active < st2->raid)
+ continue;
+ }
+
+ for (i = 0; i < st2->total; i++) { /* find a spare */
+ if (check_disk_is_free(st2, i, ext) == INT_MAX)
+ continue;
+
+ if (!get_disk_domain_and_subset(st2->devid[i],
+ st2->metadata_version,
+ &spare_domain, &spare_subset, ext))
+ continue;
+
+ /* check if domain and subset
+ * are the same as for st */
+ if ((domain != spare_domain) ||
+ (subset != spare_subset)) {
+ /* no point looking
+ * in that array/container */
+ break;
+ } else {
+ if (!dev_size_from_id(st2->devid[i], &ssize) ||
+ ssize < st->min_size)
+ continue;
+
+ if (move_spare(st2, stp, i, mailaddr,
+ mailfrom, alert_cmd,
+ dosyslog)) {
+ found = 1;
+ /* stop searching disks */
+ break;
+ }
+ }
+ }
+ if (found)
+ break; /* stop searching arrays */
+ }
+ }
+ return;
+}
+
static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mailfrom, char *cmd,
int dosyslog)
{
--
1.6.4.2
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
