__u8 num_members; /* number of member disks */
__u8 reserved[3];
__u32 filler[7]; /* expansion area */
+#define IMSM_ORD_REBUILD (1 << 24)
__u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members],
- top byte special */
+ * top byte contains some flags
+ */
} __attribute__ ((packed));
struct imsm_vol {
struct imsm_super *anchor; /* immovable parameters */
};
size_t len; /* size of the 'buf' allocation */
+ void *next_buf; /* for realloc'ing buf from the manager */
+ size_t next_len;
int updates_pending; /* count of pending updates for mdmon */
int creating_imsm; /* flag to indicate container creation */
int current_vol; /* index of raid device undergoing creation */
struct imsm_disk disk;
int fd;
} *disks;
- struct bbm_log *bbm_log;
+ struct dl *add; /* list of disks to add while mdmon active */
+ struct bbm_log *bbm_log;
};
struct extent {
enum imsm_update_type {
update_activate_spare,
update_create_array,
+ update_add_disk,
};
struct imsm_update_activate_spare {
struct imsm_dev dev;
};
+struct imsm_update_add_disk {
+ enum imsm_update_type type;
+};
+
static int imsm_env_devname_as_serial(void)
{
char *val = getenv("IMSM_DEVNAME_AS_SERIAL");
return super->dev_tbl[index];
}
-static __u32 get_imsm_disk_idx(struct imsm_map *map, int slot)
+static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev, int slot)
+{
+ struct imsm_map *map;
+
+ if (dev->vol.migr_state)
+ map = get_imsm_map(dev, 1);
+ else
+ map = get_imsm_map(dev, 0);
+
+ /* top byte identifies disk under rebuild */
+ return __le32_to_cpu(map->disk_ord_tbl[slot]);
+}
+
+#define ord_to_idx(ord) (((ord) << 8) >> 8)
+static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot)
{
- __u32 *ord_tbl = &map->disk_ord_tbl[slot];
+ __u32 ord = get_imsm_ord_tbl_ent(dev, slot);
- /* top byte is 'special' */
- return __le32_to_cpu(*ord_tbl & ~(0xff << 24));
+ return ord_to_idx(ord);
+}
+
+static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
+{
+ map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
}
static int get_imsm_raid_level(struct imsm_map *map)
struct imsm_map *map = get_imsm_map(dev, 0);
for (j = 0; j < map->num_members; j++) {
- __u32 index = get_imsm_disk_idx(map, j);
+ __u32 index = get_imsm_disk_idx(dev, j);
if (index == dl->index)
memberships++;
struct imsm_map *map = get_imsm_map(dev, 0);
for (j = 0; j < map->num_members; j++) {
- __u32 index = get_imsm_disk_idx(map, j);
+ __u32 index = get_imsm_disk_idx(dev, j);
if (index == dl->index) {
e->start = __le32_to_cpu(map->pba_of_lba0);
__u64 sz;
int slot;
struct imsm_map *map = get_imsm_map(dev, 0);
+ __u32 ord;
printf("\n");
printf("[%s]:\n", dev->volume);
printf(" RAID Level : %d\n", get_imsm_raid_level(map));
printf(" Members : %d\n", map->num_members);
for (slot = 0; slot < map->num_members; slot++)
- if (index == get_imsm_disk_idx(map, slot))
+ if (index == get_imsm_disk_idx(dev, slot))
break;
- if (slot < map->num_members)
- printf(" This Slot : %d\n", slot);
- else
+ if (slot < map->num_members) {
+ ord = get_imsm_ord_tbl_ent(dev, slot);
+ printf(" This Slot : %d%s\n", slot,
+ ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
+ } else
printf(" This Slot : ?\n");
sz = __le32_to_cpu(dev->size_high);
sz <<= 32;
printf(" Map State : %s", map_state_str[map->map_state]);
if (dev->vol.migr_state) {
struct imsm_map *map = get_imsm_map(dev, 1);
- printf(", %s", map_state_str[map->map_state]);
+ printf(" <-- %s", map_state_str[map->map_state]);
}
printf("\n");
printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
{
- printf("%s\n", __FUNCTION__);
+ /* imsm does not track uuid's so just make sure we never return
+ * the same value twice to break uuid matching in Manage_subdevs
+ * FIXME what about the use of uuid's with bitmap's?
+ */
+ static int dummy_id = 0;
+
+ uuid[0] = dummy_id++;
}
#if 0
sec->anchor->num_raid_devs > 0) {
if (first->anchor->family_num != sec->anchor->family_num)
return 3;
- if (first->anchor->mpb_size != sec->anchor->mpb_size)
- return 3;
- if (first->anchor->check_sum != sec->anchor->check_sum)
- return 3;
+ }
+
+ /* if 'first' is a spare promote it to a populated mpb with sec's
+ * family number
+ */
+ if (first->anchor->num_raid_devs == 0 &&
+ sec->anchor->num_raid_devs > 0) {
+ first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
+ first->anchor->family_num = sec->anchor->family_num;
}
return 0;
dl->fd = keep_fd ? fd : -1;
dl->devname = devname ? strdup(devname) : NULL;
strncpy((char *) dl->serial, (char *) serial, MAX_RAID_SERIAL_LEN);
+ dl->index = -2;
} else if (keep_fd) {
close(dl->fd);
dl->fd = fd;
/* only set index on disks that are a member of a
* populated contianer, i.e. one with raid_devs
*/
- if (status & SPARE_DISK)
+ if (status & FAILED_DISK)
+ dl->index = -2;
+ else if (status & SPARE_DISK)
dl->index = -1;
else
dl->index = i;
+
break;
}
}
- if (i == super->anchor->num_disks && alloc) {
- if (devname)
- fprintf(stderr,
- Name ": failed to load disk with serial \'%s\' for %s\n",
- dl->serial, devname);
- free(dl);
- return 1;
- }
- if (i == super->anchor->num_disks && dl->index >= 0) {
- if (devname)
- fprintf(stderr,
- Name ": confused... disk %d with serial \'%s\' "
- "is not listed in the current anchor\n",
- dl->index, dl->serial);
- return 1;
- }
-
if (alloc)
super->disks = dl;
-
+
return 0;
}
{
int i;
struct imsm_dev *dev_new;
- size_t len;
+ size_t len, len_migr;
+ size_t space_needed = 0;
+ struct imsm_super *mpb = super->anchor;
for (i = 0; i < super->anchor->num_raid_devs; i++) {
struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
- len = sizeof_imsm_dev(dev_iter, 1);
- dev_new = malloc(len);
+ len = sizeof_imsm_dev(dev_iter, 0);
+ len_migr = sizeof_imsm_dev(dev_iter, 1);
+ if (len_migr > len)
+ space_needed += len_migr - len;
+
+ dev_new = malloc(len_migr);
if (!dev_new)
return 1;
imsm_copy_dev(dev_new, dev_iter);
super->dev_tbl[i] = dev_new;
}
+ /* ensure that super->buf is large enough when all raid devices
+ * are migrating
+ */
+ if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
+ void *buf;
+
+ len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
+ if (posix_memalign(&buf, 512, len) != 0)
+ return 1;
+
+ memcpy(buf, super->buf, len);
+ free(super->buf);
+ super->buf = buf;
+ super->len = len;
+ }
+
return 0;
}
}
__free_imsm(super, 0);
- super->len = __le32_to_cpu(anchor->mpb_size);
super->len = ROUND_UP(anchor->mpb_size, 512);
if (posix_memalign(&super->buf, 512, super->len) != 0) {
if (devname)
rc = load_imsm_disk(fd, super, devname, 0);
if (rc == 0)
rc = parse_raid_devices(super);
+
return rc;
}
+static void __free_imsm_disk(struct dl *d)
+{
+ if (d->fd >= 0)
+ close(d->fd);
+ if (d->devname)
+ free(d->devname);
+ free(d);
+
+}
static void free_imsm_disks(struct intel_super *super)
{
while (super->disks) {
struct dl *d = super->disks;
super->disks = d->next;
- if (d->fd >= 0)
- close(d->fd);
- if (d->devname)
- free(d->devname);
- free(d);
+ __free_imsm_disk(d);
}
}
}
*sbp = super;
+ st->container_dev = fd2devnum(fd);
if (st->ss == NULL) {
st->ss = &super_imsm;
st->minor_version = 0;
st->max_devs = IMSM_MAX_DEVICES;
- st->container_dev = fd2devnum(fd);
}
return 0;
map->num_members = info->raid_disks;
for (i = 0; i < map->num_members; i++) {
/* initialized in add_to_super */
- map->disk_ord_tbl[i] = __cpu_to_le32(0);
+ set_imsm_ord_tbl_ent(map, i, 0);
}
mpb->num_raid_devs++;
super->dev_tbl[super->current_vol] = dev;
dl->index = super->anchor->num_disks;
super->anchor->num_disks++;
}
- map->disk_ord_tbl[dk->number] = __cpu_to_le32(dl->index);
+ set_imsm_ord_tbl_ent(map, dk->number, dl->index);
status = CONFIGURED_DISK | USABLE_DISK;
dl->disk.status = __cpu_to_le32(status);
dd->minor = minor(stb.st_rdev);
dd->index = -1;
dd->devname = devname ? strdup(devname) : NULL;
- dd->next = super->disks;
dd->fd = fd;
rv = imsm_read_serial(fd, devname, dd->serial);
if (rv) {
dd->disk.scsi_id = __cpu_to_le32(id);
else
dd->disk.scsi_id = __cpu_to_le32(0);
- super->disks = dd;
+
+ if (st->update_tail) {
+ dd->next = super->add;
+ super->add = dd;
+ } else {
+ dd->next = super->disks;
+ super->disks = dd;
+ }
}
static int store_imsm_mpb(int fd, struct intel_super *super);
mpb->generation_num = __cpu_to_le32(1UL);
for (d = super->disks; d; d = d->next) {
- if (d->index >= 0)
+ if (d->index != -1)
continue;
mpb->disk[0] = d->disk;
__u32 generation;
__u32 sum;
int spares = 0;
- int raid_disks = 0;
int i;
__u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
mpb->generation_num = __cpu_to_le32(generation);
for (d = super->disks; d; d = d->next) {
- if (d->index < 0)
+ if (d->index == -1)
spares++;
else {
- raid_disks++;
mpb->disk[d->index] = d->disk;
mpb_size += sizeof(struct imsm_disk);
}
}
- if (raid_disks != mpb->num_disks) {
- fprintf(stderr, "%s: expected %d disks only found %d\n",
- __func__, mpb->num_disks, raid_disks);
- return 1;
- }
for (i = 0; i < mpb->num_raid_devs; i++) {
struct imsm_dev *dev = __get_imsm_dev(mpb, i);
for (d = super->disks; d ; d = d->next) {
if (d->index < 0)
continue;
- if (store_imsm_mpb(d->fd, super)) {
+ if (store_imsm_mpb(d->fd, super))
fprintf(stderr, "%s: failed for device %d:%d %s\n",
__func__, d->major, d->minor, strerror(errno));
- return 1;
- }
if (doclose) {
close(d->fd);
d->fd = -1;
return 0;
}
+static int create_array(struct supertype *st)
+{
+ size_t len;
+ struct imsm_update_create_array *u;
+ struct intel_super *super = st->sb;
+ struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
+
+ len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0);
+ u = malloc(len);
+ if (!u) {
+ fprintf(stderr, "%s: failed to allocate update buffer\n",
+ __func__);
+ return 1;
+ }
+
+ u->type = update_create_array;
+ u->dev_idx = super->current_vol;
+ imsm_copy_dev(&u->dev, dev);
+ append_metadata_update(st, u, len);
+
+ return 0;
+}
+
+static int add_disk(struct supertype *st)
+{
+ struct intel_super *super = st->sb;
+ size_t len;
+ struct imsm_update_add_disk *u;
+
+ if (!super->add)
+ return 0;
+
+ len = sizeof(*u);
+ u = malloc(len);
+ if (!u) {
+ fprintf(stderr, "%s: failed to allocate update buffer\n",
+ __func__);
+ return 1;
+ }
+
+ u->type = update_add_disk;
+ append_metadata_update(st, u, len);
+
+ return 0;
+}
+
static int write_init_super_imsm(struct supertype *st)
{
if (st->update_tail) {
- /* queue the recently created array as a metadata update */
- size_t len;
- struct imsm_update_create_array *u;
+ /* queue the recently created array / added disk
+ * as a metadata update */
struct intel_super *super = st->sb;
- struct imsm_dev *dev;
struct dl *d;
+ int rv;
- if (super->current_vol < 0 ||
- !(dev = get_imsm_dev(super, super->current_vol))) {
- fprintf(stderr, "%s: could not determine sub-array\n",
- __func__);
- return 1;
- }
-
-
- len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0);
- u = malloc(len);
- if (!u) {
- fprintf(stderr, "%s: failed to allocate update buffer\n",
- __func__);
- return 1;
- }
-
- u->type = update_create_array;
- u->dev_idx = super->current_vol;
- imsm_copy_dev(&u->dev, dev);
- append_metadata_update(st, u, len);
+ /* determine if we are creating a volume or adding a disk */
+ if (super->current_vol < 0) {
+ /* in the add disk case we are running in mdmon
+ * context, so don't close fd's
+ */
+ return add_disk(st);
+ } else
+ rv = create_array(st);
for (d = super->disks; d ; d = d->next) {
close(d->fd);
d->fd = -1;
}
- return 0;
+ return rv;
} else
return write_super_imsm(st->sb, 1);
}
this = malloc(sizeof(*this));
memset(this, 0, sizeof(*this));
this->next = rest;
- rest = this;
this->array.level = get_imsm_raid_level(map);
this->array.raid_disks = map->num_members;
this->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9;
this->array.state = !vol->dirty;
this->container_member = i;
- if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
+ if (map->map_state == IMSM_T_STATE_UNINITIALIZED ||
+ dev->vol.dirty || dev->vol.migr_state)
this->resync_start = 0;
else
this->resync_start = ~0ULL;
struct mdinfo *info_d;
struct dl *d;
int idx;
+ int skip;
__u32 s;
+ __u32 ord;
- idx = get_imsm_disk_idx(map, slot);
+ skip = 0;
+ idx = get_imsm_disk_idx(dev, slot);
+ ord = get_imsm_ord_tbl_ent(dev, slot);
for (d = super->disks; d ; d = d->next)
if (d->index == idx)
break;
if (d == NULL)
- break; /* shouldn't this be continue ?? */
+ skip = 1;
+
+ s = d ? __le32_to_cpu(d->disk.status) : 0;
+ if (s & FAILED_DISK)
+ skip = 1;
+ if (!(s & USABLE_DISK))
+ skip = 1;
+ if (ord & IMSM_ORD_REBUILD)
+ skip = 1;
+
+ /*
+ * if we skip some disks the array will be assmebled degraded;
+ * reset resync start to avoid a dirty-degraded situation
+ *
+ * FIXME handle dirty degraded
+ */
+ if (skip && !dev->vol.dirty)
+ this->resync_start = ~0ULL;
+ if (skip)
+ continue;
info_d = malloc(sizeof(*info_d));
- if (!info_d)
- break; /* ditto ?? */
+ if (!info_d) {
+ fprintf(stderr, Name ": failed to allocate disk"
+ " for volume %s\n", (char *) dev->volume);
+ free(this);
+ this = rest;
+ break;
+ }
memset(info_d, 0, sizeof(*info_d));
info_d->next = this->devs;
this->devs = info_d;
- s = __le32_to_cpu(d->disk.status);
-
info_d->disk.number = d->index;
info_d->disk.major = d->major;
info_d->disk.minor = d->minor;
info_d->disk.raid_disk = slot;
- info_d->disk.state = s & CONFIGURED_DISK ? (1 << MD_DISK_ACTIVE) : 0;
- info_d->disk.state |= s & FAILED_DISK ? (1 << MD_DISK_FAULTY) : 0;
- info_d->disk.state |= s & USABLE_DISK ? (1 << MD_DISK_SYNC) : 0;
this->array.working_disks++;
if (d->devname)
strcpy(info_d->name, d->devname);
}
+ rest = this;
}
return rest;
return 0;
}
-static __u8 imsm_check_degraded(struct intel_super *super, int n, int failed)
+static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
{
- struct imsm_dev *dev = get_imsm_dev(super, n);
struct imsm_map *map = get_imsm_map(dev, 0);
if (!failed)
int device_per_mirror = 2; /* FIXME is this always the case?
* and are they always adjacent?
*/
- int failed = 0;
+ int r10fail = 0;
int i;
for (i = 0; i < map->num_members; i++) {
- int idx = get_imsm_disk_idx(map, i);
+ int idx = get_imsm_disk_idx(dev, i);
struct imsm_disk *disk = get_imsm_disk(super, idx);
- if (__le32_to_cpu(disk->status) & FAILED_DISK)
- failed++;
+ if (!disk)
+ r10fail++;
+ else if (__le32_to_cpu(disk->status) & FAILED_DISK)
+ r10fail++;
- if (failed >= device_per_mirror)
+ if (r10fail >= device_per_mirror)
return IMSM_T_STATE_FAILED;
- /* reset 'failed' for next mirror set */
+ /* reset 'r10fail' for next mirror set */
if (!((i + 1) % device_per_mirror))
- failed = 0;
+ r10fail = 0;
}
return IMSM_T_STATE_DEGRADED;
return map->map_state;
}
-static int imsm_count_failed(struct intel_super *super, struct imsm_map *map)
+static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
{
int i;
int failed = 0;
struct imsm_disk *disk;
+ struct imsm_map *map = get_imsm_map(dev, 0);
for (i = 0; i < map->num_members; i++) {
- int idx = get_imsm_disk_idx(map, i);
+ __u32 ord = get_imsm_ord_tbl_ent(dev, i);
+ int idx = ord_to_idx(ord);
disk = get_imsm_disk(super, idx);
- if (__le32_to_cpu(disk->status) & FAILED_DISK)
+ if (!disk ||
+ __le32_to_cpu(disk->status) & FAILED_DISK ||
+ ord & IMSM_ORD_REBUILD)
failed++;
}
return failed;
}
-static void imsm_set_array_state(struct active_array *a, int consistent)
+static int imsm_set_array_state(struct active_array *a, int consistent)
{
int inst = a->info.container_member;
struct intel_super *super = a->container->sb;
int failed;
__u8 map_state;
- failed = imsm_count_failed(super, map);
- map_state = imsm_check_degraded(super, inst, failed);
+ failed = imsm_count_failed(super, dev);
+ map_state = imsm_check_degraded(super, dev, failed);
if (consistent && !dev->vol.dirty &&
(dev->vol.migr_state || map_state != IMSM_T_STATE_NORMAL))
a->resync_start = 0ULL;
+ if (consistent == 2 && a->resync_start != ~0ULL)
+ consistent = 0;
if (a->resync_start == ~0ULL) {
/* complete recovery or initial resync */
dev->vol.dirty = dirty;
super->updates_pending++;
}
+ return consistent;
}
static void imsm_set_disk(struct active_array *a, int n, int state)
__u32 status;
int failed = 0;
int new_failure = 0;
+ __u32 ord;
if (n > map->num_members)
fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
dprintf("imsm: set_disk %d:%x\n", n, state);
- disk = get_imsm_disk(super, get_imsm_disk_idx(map, n));
+ ord = get_imsm_ord_tbl_ent(dev, n);
+ disk = get_imsm_disk(super, ord_to_idx(ord));
/* check for new failures */
status = __le32_to_cpu(disk->status);
if ((state & DS_FAULTY) && !(status & FAILED_DISK)) {
status |= FAILED_DISK;
disk->status = __cpu_to_le32(status);
+ disk->scsi_id = __cpu_to_le32(~0UL);
+ memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
new_failure = 1;
super->updates_pending++;
}
/* check if in_sync */
- if ((state & DS_INSYNC) && !(status & USABLE_DISK)) {
- status |= USABLE_DISK;
- disk->status = __cpu_to_le32(status);
+ if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD) {
+ struct imsm_map *migr_map = get_imsm_map(dev, 1);
+
+ set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
super->updates_pending++;
}
* degraded / failed status
*/
if (new_failure && map->map_state != IMSM_T_STATE_FAILED)
- failed = imsm_count_failed(super, map);
+ failed = imsm_count_failed(super, dev);
/* determine map_state based on failed or in_sync count */
if (failed)
- map->map_state = imsm_check_degraded(super, inst, failed);
+ map->map_state = imsm_check_degraded(super, dev, failed);
else if (map->map_state == IMSM_T_STATE_DEGRADED) {
struct mdinfo *d;
int working = 0;
static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
{
struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
- struct imsm_map *map = get_imsm_map(dev, 0);
- int i = get_imsm_disk_idx(map, idx);
+ int i = get_imsm_disk_idx(dev, idx);
struct dl *dl;
for (dl = super->disks; dl; dl = dl->next)
if (dl->index == i)
break;
- if (__le32_to_cpu(dl->disk.status) & FAILED_DISK)
+ if (dl && __le32_to_cpu(dl->disk.status) & FAILED_DISK)
dl = NULL;
if (dl)
int j;
int found;
__u32 array_start;
+ __u32 status;
struct dl *dl;
for (dl = super->disks; dl; dl = dl->next) {
if (d)
continue;
+ /* skip marked in use or failed drives */
+ status = __le32_to_cpu(dl->disk.status);
+ if (status & FAILED_DISK || status & CONFIGURED_DISK) {
+ dprintf("%x:%x status ( %s%s)\n",
+ dl->major, dl->minor,
+ status & FAILED_DISK ? "failed " : "",
+ status & CONFIGURED_DISK ? "configured " : "");
+ continue;
+ }
+
/* Does this unused device have the requisite free space?
* We need a->info.component_size sectors
*/
dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
inst, failed, a->info.array.raid_disks, a->info.array.level);
- if (imsm_check_degraded(super, inst, failed) != IMSM_T_STATE_DEGRADED)
+ if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
return NULL;
/* For each slot, if it is not working, find a spare */
return rv;
}
-static int disks_overlap(struct imsm_map *m1, struct imsm_map *m2)
+static int disks_overlap(struct imsm_dev *d1, struct imsm_dev *d2)
{
+ struct imsm_map *m1 = get_imsm_map(d1, 0);
+ struct imsm_map *m2 = get_imsm_map(d2, 0);
int i;
int j;
int idx;
for (i = 0; i < m1->num_members; i++) {
- idx = get_imsm_disk_idx(m1, i);
+ idx = get_imsm_disk_idx(d1, i);
for (j = 0; j < m2->num_members; j++)
- if (idx == get_imsm_disk_idx(m2, j))
+ if (idx == get_imsm_disk_idx(d2, j))
return 1;
}
return 0;
}
+static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
+
static void imsm_process_update(struct supertype *st,
struct metadata_update *update)
{
* flag
*/
struct intel_super *super = st->sb;
- struct imsm_super *mpb = super->anchor;
+ struct imsm_super *mpb;
enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
+ /* update requires a larger buf but the allocation failed */
+ if (super->next_len && !super->next_buf) {
+ super->next_len = 0;
+ return;
+ }
+
+ if (super->next_buf) {
+ memcpy(super->next_buf, super->buf, super->len);
+ free(super->buf);
+ super->len = super->next_len;
+ super->buf = super->next_buf;
+
+ super->next_len = 0;
+ super->next_buf = NULL;
+ }
+
+ mpb = super->anchor;
+
switch (type) {
case update_activate_spare: {
struct imsm_update_activate_spare *u = (void *) update->buf;
dl->index = super->anchor->num_disks;
super->anchor->num_disks++;
}
- victim = get_imsm_disk_idx(map, u->slot);
- map->disk_ord_tbl[u->slot] = __cpu_to_le32(dl->index);
+ victim = get_imsm_disk_idx(dev, u->slot);
+ set_imsm_ord_tbl_ent(map, u->slot, dl->index);
disk = &dl->disk;
status = __le32_to_cpu(disk->status);
status |= CONFIGURED_DISK;
- status &= ~(SPARE_DISK | USABLE_DISK);
+ status &= ~SPARE_DISK;
disk->status = __cpu_to_le32(status);
/* count arrays using the victim in the metadata */
found = 0;
for (a = st->arrays; a ; a = a->next) {
dev = get_imsm_dev(super, a->info.container_member);
- map = get_imsm_map(dev, 0);
for (i = 0; i < map->num_members; i++)
- if (victim == get_imsm_disk_idx(map, i))
+ if (victim == get_imsm_disk_idx(dev, i))
found++;
}
- /* clear some flags if the victim is no longer being
+ /* delete the victim if it is no longer being
* utilized anywhere
*/
if (!found) {
- disk = get_imsm_disk(super, victim);
- status = __le32_to_cpu(disk->status);
- status &= ~(CONFIGURED_DISK | USABLE_DISK);
- disk->status = __cpu_to_le32(status);
- /* at this point the disk can be removed from the
- * metadata, however we need to guarantee that we do
- * not race with any manager thread routine that relies
- * on dl->index or map->disk_ord_tbl
- */
+ struct dl **dlp;
+
+ for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
+ if ((*dlp)->index == victim)
+ break;
+ /* We know that 'manager' isn't touching anything,
+ * so it is safe to:
+ */
+ imsm_delete(super, dlp, victim);
}
break;
}
if ((new_start >= start && new_start <= end) ||
(start >= new_start && start <= new_end))
overlap = 1;
- if (overlap && disks_overlap(map, new_map)) {
+ if (overlap && disks_overlap(dev, &u->dev)) {
dprintf("%s: arrays overlap\n", __func__);
return;
}
super->updates_pending++;
dev = update->space;
+ map = get_imsm_map(dev, 0);
update->space = NULL;
imsm_copy_dev(dev, &u->dev);
+ map = get_imsm_map(dev, 0);
super->dev_tbl[u->dev_idx] = dev;
mpb->num_raid_devs++;
- /* fix up flags, if arrays overlap then the drives can not be
- * spares
- */
+ /* fix up flags */
for (i = 0; i < map->num_members; i++) {
struct imsm_disk *disk;
__u32 status;
- disk = get_imsm_disk(super, get_imsm_disk_idx(map, i));
+ disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
status = __le32_to_cpu(disk->status);
status |= CONFIGURED_DISK;
- if (overlap)
- status &= ~SPARE_DISK;
+ status &= ~SPARE_DISK;
disk->status = __cpu_to_le32(status);
}
break;
}
+ case update_add_disk:
+
+ /* we may be able to repair some arrays if disks are
+ * being added */
+ if (super->add) {
+ struct active_array *a;
+ for (a = st->arrays; a; a = a->next)
+ a->check_degraded = 1;
+ }
+ /* check if we can add / replace some disks in the
+ * metadata */
+ while (super->add) {
+ struct dl **dlp, *dl, *al;
+ al = super->add;
+ super->add = al->next;
+ for (dlp = &super->disks; *dlp ; ) {
+ if (memcmp(al->serial, (*dlp)->serial,
+ MAX_RAID_SERIAL_LEN) == 0) {
+ dl = *dlp;
+ *dlp = (*dlp)->next;
+ __free_imsm_disk(dl);
+ break;
+ } else
+ dlp = &(*dlp)->next;
+ }
+ al->next = super->disks;
+ super->disks = al;
+ }
+
+ break;
}
}
struct metadata_update *update)
{
/**
- * Allocate space to hold new disk entries, raid-device entries or a
- * new mpb if necessary. We currently maintain an mpb large enough to
- * hold 2 subarrays for the given number of disks. This may not be
- * sufficient when reshaping.
- *
- * FIX ME handle the reshape case.
- *
- * The monitor will be able to safely change super->mpb by arranging
- * for it to be freed in check_update_queue(). I.e. the monitor thread
- * will start using the new pointer and the manager can continue to use
- * the old value until check_update_queue() runs.
+ * Allocate space to hold new disk entries, raid-device entries or a new
+ * mpb if necessary. The manager synchronously waits for updates to
+ * complete in the monitor, so new mpb buffers allocated here can be
+ * integrated by the monitor thread without worrying about live pointers
+ * in the manager thread.
*/
enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
+ struct intel_super *super = st->sb;
+ struct imsm_super *mpb = super->anchor;
+ size_t buf_len;
+ size_t len = 0;
switch (type) {
case update_create_array: {
struct imsm_update_create_array *u = (void *) update->buf;
- size_t len = sizeof_imsm_dev(&u->dev, 1);
+ len = sizeof_imsm_dev(&u->dev, 1);
update->space = malloc(len);
break;
default:
}
}
- return;
+ /* check if we need a larger metadata buffer */
+ if (super->next_buf)
+ buf_len = super->next_len;
+ else
+ buf_len = super->len;
+
+ if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
+ /* ok we need a larger buf than what is currently allocated
+ * if this allocation fails process_update will notice that
+ * ->next_len is set and ->next_buf is NULL
+ */
+ buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
+ if (super->next_buf)
+ free(super->next_buf);
+
+ super->next_len = buf_len;
+ if (posix_memalign(&super->next_buf, buf_len, 512) != 0)
+ super->next_buf = NULL;
+ }
+}
+
+/* must be called while manager is quiesced */
+static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
+{
+ struct imsm_super *mpb = super->anchor;
+ struct dl *iter;
+ struct imsm_dev *dev;
+ struct imsm_map *map;
+ int i, j, num_members;
+ __u32 ord;
+
+ dprintf("%s: deleting device[%d] from imsm_super\n",
+ __func__, index);
+
+ /* shift all indexes down one */
+ for (iter = super->disks; iter; iter = iter->next)
+ if (iter->index > index)
+ iter->index--;
+
+ for (i = 0; i < mpb->num_raid_devs; i++) {
+ dev = get_imsm_dev(super, i);
+ map = get_imsm_map(dev, 0);
+ num_members = map->num_members;
+ for (j = 0; j < num_members; j++) {
+ /* update ord entries being careful not to propagate
+ * ord-flags to the first map
+ */
+ ord = get_imsm_ord_tbl_ent(dev, j);
+
+ if (ord_to_idx(ord) <= index)
+ continue;
+
+ map = get_imsm_map(dev, 0);
+ set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
+ map = get_imsm_map(dev, 1);
+ if (map)
+ set_imsm_ord_tbl_ent(map, j, ord - 1);
+ }
+ }
+
+ mpb->num_disks--;
+ super->updates_pending++;
+ if (*dlp) {
+ struct dl *dl = *dlp;
+
+ *dlp = (*dlp)->next;
+ __free_imsm_disk(dl);
+ }
}
struct superswitch super_imsm = {
projects / thirdparty / mdadm.git / blobdiff