>From f45f97933fddce7d7fcf370e4a74e9281c7c0a38 Mon Sep 17 00:00:00 2001
From: Anna Czarnowska <anna.czarnowska@xxxxxxxxx>
Date: Tue, 28 Sep 2010 06:26:51 +0200
Subject: [AUTOREBUILD 6/8] Monitor: autorebuild functionality added
For each volume we check state, report any changes, note minimum size
of disks and link with parent container. After all information is updated
we call spare_sharing. spare_sharing searches suitable spares in other
arrays and moves them using move_spare to the arrays that need them.
move_spare removes spare from one array/container and adds to another.
If add fails we add back to original container. Manage_subdev function
is used to perform the spare relocation.
Signed-off-by: Marcin Labun <marcin.labun@xxxxxxxxx>
Signed-off-by: Anna Czarnowska <anna.czarnowska@xxxxxxxxx>
---
Monitor.c | 312 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++-----
1 files changed, 287 insertions(+), 25 deletions(-)
diff --git a/Monitor.c b/Monitor.c
index 93dd15d..62cbe98 100644
--- a/Monitor.c
+++ b/Monitor.c
@@ -30,6 +30,13 @@
#include <limits.h>
#include <syslog.h>
+/* define verbose mode for DEBUG compilation */
+#ifdef DEBUG
+#define VERBOSE 1
+#else
+#define VERBOSE (-1)
+#endif
+
static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mailfrom,
char *cmd, int dosyslog);
@@ -47,6 +54,7 @@ struct state {
int expected_spares;
int devstate[MaxDisks];
unsigned devid[MaxDisks];
+ unsigned long long min_size;
int percent;
char *metadata_version;
struct state *volumes;/* for a container it is a link its all volumes */
@@ -54,7 +62,8 @@ struct state {
struct state *next;
};
-
+static void spare_sharing(struct state *statelist, char *mailaddr,
+ char *mailfrom, char *alert_cmd, int dosyslog);
static void add_to_cont(struct state *cont, struct state *vol)
{
@@ -106,6 +115,10 @@ 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 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
@@ -113,6 +126,7 @@ int Monitor(mddev_dev_t devlist,
*/
int finished = 0;
+ int anydegraded;
struct mdstat_ent *mdstat = NULL;
char *mailfrom = NULL;
struct state *statelist = NULL;
@@ -222,6 +236,9 @@ 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));
statelist = st;
}
} else {
@@ -242,6 +259,9 @@ 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;
}
@@ -254,6 +274,7 @@ int Monitor(mddev_dev_t devlist,
int new_found = 0;
struct state *st;
+ anydegraded = 0;
if (mdstat)
free_mdstat(mdstat);
mdstat = mdstat_read(oneshot?0:1, 0);
@@ -334,18 +355,17 @@ int Monitor(mddev_dev_t devlist,
* metadata, so treat utime for external
* metadata as different
*/
- if ((st->utime == array.utime &&
- ((st->metadata_version == NULL) ||
- !is_external(st->metadata_version))) &&
+ if (st->utime == array.utime &&
+ (st->metadata_version &&
+ !is_external(st->metadata_version)) &&
st->failed == array.failed_disks &&
st->working == array.working_disks &&
st->spare == array.spare_disks &&
- (mse == NULL || (mse->percent == st->percent))) {
- close(fd);
+ (mse->percent == st->percent)) {
st->err = 0;
+ close(fd);
continue;
}
-
if (st->utime == 0 && /* new array */
mse->pattern && strchr(mse->pattern, '_') /* degraded */
)
@@ -409,6 +429,7 @@ 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;
@@ -453,6 +474,19 @@ int Monitor(mddev_dev_t devlist,
}
st->devstate[i] = newstate;
st->devid[i] = makedev(disc.major, disc.minor);
+
+ if (!share)
+ continue;
+ /* 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;
@@ -462,6 +496,8 @@ int Monitor(mddev_dev_t devlist,
st->raid = array.raid_disks;
st->total = array.raid_disks + array.nr_disks;
st->err = 0;
+ if ((st->active < st->raid) && st->spare == 0)
+ anydegraded = 1;
if (mse->metadata_version) {
if (!st->metadata_version)
st->metadata_version = strdup(mse->metadata_version);
@@ -515,27 +551,26 @@ int Monitor(mddev_dev_t devlist,
new_found = 1;
}
}
-
- /* search the statelist to connect external
- * metadata volumes with their containers
- */
- for (st = statelist; st; st = st->next) {
- if (st->metadata_version &&
- is_external(st->metadata_version) &&
- is_subarray(st->metadata_version+9)) {
- struct state *cont = NULL;
-
- for (cont = statelist; cont; cont = cont->next) {
- if (!cont->err &&
- cont->parent == NULL &&
- cont->metadata_version &&
- devname2devnum(st->metadata_version+10)
- == cont->devnum) {
- add_to_cont(cont, st);
- break;
+ if (share && anydegraded) {
+ /* parent-volume linking only needed when sharing spares */
+ for (st = statelist; st; st = st->next) {
+ if (!st->err &&
+ st->metadata_version &&
+ is_external(st->metadata_version) &&
+ is_subarray(st->metadata_version+9)) {
+ struct state *cont = NULL;
+ for (cont = statelist; cont; cont = cont->next) {
+ if (!cont->err &&
+ cont->parent == NULL &&
+ cont->metadata_version &&
+ devname2devnum(st->metadata_version+10) == cont->devnum) {
+ add_to_cont(cont, st);
+ break;
+ }
}
}
}
+ spare_sharing(statelist, mailaddr, mailfrom, alert_cmd, dosyslog);
}
if (!new_found) {
if (oneshot)
@@ -550,6 +585,233 @@ int Monitor(mddev_dev_t devlist,
return 0;
}
+
+
+/* get states of all disks in native volume or container
+ * from kernel or metadata handler
+ */
+static struct mdinfo *get_raid_disk_info(struct state *st)
+{
+ struct supertype *sty = NULL;
+ int fd = -1, i, rv = 1;
+ unsigned id = 0;
+ struct mdinfo *infolist = NULL, *info;
+
+ /* ignore arrays with error and get info for containers
+ * or native volumes
+ */
+ if (st->err || (is_external(st->metadata_version) &&
+ is_subarray(st->metadata_version+9)))
+ return NULL;
+
+ if (is_external(st->metadata_version)) {
+ fd = open(st->devname, O_RDONLY);
+ if (fd < 0)
+ return NULL;
+ sty = guess_super(fd);
+ if (!sty) {
+ close(fd);
+ return NULL;;
+ }
+ if (sty->ss->load_super(sty, fd, st->devname)) {
+ rv = 0;
+ goto cleanup;
+ }
+ infolist = sty->ss->getinfo_super_disks(sty);
+ } else
+ infolist = sysfs_read(-1, st->devnum,
+ GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
+ GET_DEGRADED|GET_COMPONENT|GET_VERSION);
+
+ if (!infolist) {
+ rv = 0;
+ goto cleanup;
+ }
+ for (i = 0; i < st->total; i++) {
+ if (st->devid[i] == 0)
+ continue;
+ for (info = infolist->devs; info; info = info->next) {
+ id = makedev(info->disk.major, info->disk.minor);
+ if (st->devid[i] == id) {
+ st->devstate[i] = info->disk.state;
+ break;
+ }
+ }
+ if (!info)
+ st->devstate[i] = 1<<MD_DISK_FAULTY;
+ }
+
+ cleanup:
+ if (fd >= 0)
+ close(fd);
+ if (sty) { \
+ sty->ss->free_super(sty);
+ free(sty);
+ }
+
+ if (!rv) {
+ if (infolist)
+ sysfs_free(infolist);
+ infolist = NULL;
+ return NULL;
+ }
+ return infolist;
+}
+
+int move_spare(struct state *st2, struct state *st1, unsigned *devid,
+ 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 || (*devid) == 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(*devid), minor(*devid), 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, VERBOSE, 0) == 0) {
+ devlist.disposition = 'a';
+ if (Manage_subdevs(st1->devname, to_fd, &devlist,
+ VERBOSE, 0) == 0) {
+ *devid = 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 if (Manage_subdevs(st2->devname, from_fd,
+ &devlist, VERBOSE, 0) != 0)
+ fprintf(stderr,
+ "Error: Adding back spare device"
+ "%s to container %s failed!\n",
+ st2->devname, dv);
+ }
+ /* Failed to add spare to new container */
+ close(from_fd);
+ close(to_fd);
+ return 0;
+}
+
+
+static int dev_suitable(unsigned devid, int devstate, unsigned long long size)
+{
+ unsigned long long ssize;
+ /* check if device not used in volumes, not failed, and big enough */
+ if ((devid > 0) && (devstate == 0) &&
+ dev_size_from_id(devid, &ssize) && (ssize >= size))
+ return 1;
+ return 0;
+}
+
+
+/* If an array has active < raid && spare == 0
+ * Look for another array/container with unused, unfailed spare
+ * and the same domain
+ * 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;
+ int i, ext, found;
+ struct mdinfo *sra = NULL;
+
+ 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;
+ }
+ /* get member device state updated */
+ sra = get_raid_disk_info(stp);
+ if (!sra) {
+ dprintf("no sra for device: %s\n", stp->devname);
+ continue;
+ }
+ sysfs_free(sra);
+ for (i = 0; i < stp->total; i++)
+ if (dev_suitable(stp->devid[i], stp->devstate[i],
+ st->min_size))
+ break;
+ if (i < stp->total)
+ /* there is a spare in array/parent container,
+ * it was probably just added
+ * but mdmon has not started recovery yet
+ * we will not add any more spares for now */
+ 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;
+ }
+ /* support for domain comparision needed */
+ for (i = 0; i < st2->total; i++) {
+ if (!dev_suitable(st2->devid[i],
+ st2->devstate[i],
+ st->min_size))
+ continue;
+ if (move_spare(st2, stp, &st2->devid[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
