__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_update_create_array {
enum imsm_update_type type;
- struct imsm_dev dev;
int dev_idx;
+ struct imsm_dev dev;
+};
+
+struct imsm_update_add_disk {
+ enum imsm_update_type type;
};
static int imsm_env_devname_as_serial(void)
return sum - __le32_to_cpu(mpb->check_sum);
}
-static size_t sizeof_imsm_dev(struct imsm_dev *dev)
+static size_t sizeof_imsm_map(struct imsm_map *map)
+{
+ return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
+}
+
+struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
{
- size_t size = sizeof(*dev);
+ struct imsm_map *map = &dev->vol.map[0];
+
+ if (second_map && !dev->vol.migr_state)
+ return NULL;
+ else if (second_map) {
+ void *ptr = map;
- /* each map has disk_ord_tbl[num_members - 1] additional space */
- size += sizeof(__u32) * (dev->vol.map[0].num_members - 1);
+ return ptr + sizeof_imsm_map(map);
+ } else
+ return map;
+
+}
+
+/* return the size of the device.
+ * migr_state increases the returned size if map[0] were to be duplicated
+ */
+static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
+{
+ size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
+ sizeof_imsm_map(get_imsm_map(dev, 0));
/* migrating means an additional map */
- if (dev->vol.migr_state) {
- size += sizeof(struct imsm_map);
- size += sizeof(__u32) * (dev->vol.map[1].num_members - 1);
- }
+ if (dev->vol.migr_state)
+ size += sizeof_imsm_map(get_imsm_map(dev, 1));
+ else if (migr_state)
+ size += sizeof_imsm_map(get_imsm_map(dev, 0));
return size;
}
if (i == index)
return _mpb + offset;
else
- offset += sizeof_imsm_dev(_mpb + offset);
+ offset += sizeof_imsm_dev(_mpb + offset, 0);
return NULL;
}
{
__u32 *ord_tbl = &map->disk_ord_tbl[slot];
- /* top byte is 'special' */
+ /* top byte identifies disk under rebuild
+ * why not just use the USABLE bit... oh well.
+ */
return __le32_to_cpu(*ord_tbl & ~(0xff << 24));
}
+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);
+
+ return map->disk_ord_tbl[slot];
+}
+
static int get_imsm_raid_level(struct imsm_map *map)
{
if (map->raid_level == 1) {
for (i = 0; i < super->anchor->num_raid_devs; i++) {
struct imsm_dev *dev = get_imsm_dev(super, i);
- struct imsm_map *map = dev->vol.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);
for (i = 0; i < super->anchor->num_raid_devs; i++) {
struct imsm_dev *dev = get_imsm_dev(super, i);
- struct imsm_map *map = dev->vol.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);
{
__u64 sz;
int slot;
- struct imsm_map *map = dev->vol.map;
+ struct imsm_map *map = get_imsm_map(dev, 0);
printf("\n");
printf("[%s]:\n", dev->volume);
printf(" Chunk Size : %u KiB\n",
__le16_to_cpu(map->blocks_per_strip) / 2);
printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
- printf(" Migrate State : %s\n", dev->vol.migr_state ? "migrating" : "idle");
+ printf(" Migrate State : %s", dev->vol.migr_state ? "migrating" : "idle");
+ if (dev->vol.migr_state)
+ printf(": %s", dev->vol.migr_type ? "rebuilding" : "initializing");
+ printf("\n");
+ 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("\n");
printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
- printf(" Map State : %s\n", map_state_str[map->map_state]);
}
static void print_imsm_disk(struct imsm_super *mpb, int index)
static void brief_examine_super_imsm(struct supertype *st)
{
- struct intel_super *super = st->sb;
-
- printf("ARRAY /dev/imsm family=%08x metadata=external:imsm\n",
- __le32_to_cpu(super->anchor->family_num));
+ printf("ARRAY /dev/imsm metadata=imsm\n");
}
static void detail_super_imsm(struct supertype *st, char *homehost)
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
{
struct intel_super *super = st->sb;
struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
- struct imsm_map *map = &dev->vol.map[0];
+ struct imsm_map *map = get_imsm_map(dev, 0);
info->container_member = super->current_vol;
info->array.raid_disks = map->num_members;
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;
}
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 = -3;
} 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;
}
}
- 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 (dl->index == -3) {
+ fprintf(stderr, Name ": device %x:%x with serial %s"
+ " does not belong to this container\n",
+ dl->major, dl->minor, (char *) serial);
+ return 2;
}
if (alloc)
super->disks = dl;
-
+
return 0;
}
static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
{
- int i;
-
- *dest = *src;
-
- for (i = 0; i < src->vol.map[0].num_members; i++)
- dest->vol.map[0].disk_ord_tbl[i] = src->vol.map[0].disk_ord_tbl[i];
+ memcpy(dest, src, sizeof_imsm_dev(src, 0));
+}
- if (!src->vol.migr_state)
- return;
+static void dup_map(struct imsm_dev *dev)
+{
+ struct imsm_map *dest = get_imsm_map(dev, 1);
+ struct imsm_map *src = get_imsm_map(dev, 0);
- dest->vol.map[1] = src->vol.map[1];
- for (i = 0; i < src->vol.map[1].num_members; i++)
- dest->vol.map[1].disk_ord_tbl[i] = src->vol.map[1].disk_ord_tbl[i];
+ memcpy(dest, src, sizeof_imsm_map(src));
}
static int parse_raid_devices(struct intel_super *super)
{
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);
- 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;
vol->dirty = 0;
for (i = 0; i < idx; i++) {
struct imsm_dev *prev = get_imsm_dev(super, i);
- struct imsm_map *pmap = &prev->vol.map[0];
+ struct imsm_map *pmap = get_imsm_map(prev, 0);
offset += __le32_to_cpu(pmap->blocks_per_member);
offset += IMSM_RESERVED_SECTORS;
}
- map = &vol->map[0];
+ map = get_imsm_map(dev, 0);
map->pba_of_lba0 = __cpu_to_le32(offset);
map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
__u32 status;
dev = get_imsm_dev(super, super->current_vol);
- map = &dev->vol.map[0];
+ map = get_imsm_map(dev, 0);
for (dl = super->disks; dl ; dl = dl->next)
if (dl->major == dk->major &&
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);
fprintf(stderr, "%s: failed for device %d:%d %s\n",
__func__, d->major, d->minor, strerror(errno));
*mpb = mpb_save;
- return 0;
+ return 1;
}
if (doclose) {
close(d->fd);
}
*mpb = mpb_save;
- return 1;
+ return 0;
}
static int write_super_imsm(struct intel_super *super, int doclose)
int spares = 0;
int raid_disks = 0;
int i;
+ __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
/* 'generation' is incremented everytime the metadata is written */
generation = __le32_to_cpu(mpb->generation_num);
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 0;
+ return 1;
}
for (i = 0; i < mpb->num_raid_devs; i++) {
struct imsm_dev *dev = __get_imsm_dev(mpb, i);
imsm_copy_dev(dev, super->dev_tbl[i]);
+ mpb_size += sizeof_imsm_dev(dev, 0);
}
+ mpb_size += __le32_to_cpu(mpb->bbm_log_size);
+ mpb->mpb_size = __cpu_to_le32(mpb_size);
/* recalculate checksum */
sum = __gen_imsm_checksum(mpb);
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 0;
+ return 1;
}
if (doclose) {
close(d->fd);
if (spares)
return write_super_imsm_spares(super, doclose);
- return 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);
- 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);
}
for (i = 0; i < mpb->num_raid_devs; i++) {
struct imsm_dev *dev = get_imsm_dev(super, i);
struct imsm_vol *vol = &dev->vol;
- struct imsm_map *map = vol->map;
+ struct imsm_map *map = get_imsm_map(dev, 0);
struct mdinfo *this;
int slot;
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;
+ skip = 0;
idx = get_imsm_disk_idx(map, 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;
static __u8 imsm_check_degraded(struct intel_super *super, int n, int failed)
{
struct imsm_dev *dev = get_imsm_dev(super, n);
- struct imsm_map *map = dev->vol.map;
+ struct imsm_map *map = get_imsm_map(dev, 0);
if (!failed)
- return map->map_state;
+ return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
+ IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
switch (get_imsm_raid_level(map)) {
case 0:
disk = get_imsm_disk(super, idx);
if (__le32_to_cpu(disk->status) & FAILED_DISK)
failed++;
+ else if (!(__le32_to_cpu(disk->status) & USABLE_DISK))
+ 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;
struct imsm_dev *dev = get_imsm_dev(super, inst);
- struct imsm_map *map = &dev->vol.map[0];
+ struct imsm_map *map = get_imsm_map(dev, 0);
int dirty = !consistent;
int failed;
__u8 map_state;
+ failed = imsm_count_failed(super, map);
+ map_state = imsm_check_degraded(super, inst, 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) {
- failed = imsm_count_failed(super, map);
- map_state = imsm_check_degraded(super, inst, failed);
/* complete recovery or initial resync */
- if (!failed)
- map_state = IMSM_T_STATE_NORMAL;
if (map->map_state != map_state) {
dprintf("imsm: map_state %d: %d\n",
inst, map_state);
map->map_state = map_state;
super->updates_pending++;
}
-
- /* complete resync */
- if (!dirty && dev->vol.dirty) {
- dprintf("imsm: mark 'clean'\n");
- dev->vol.dirty = 0;
+ if (dev->vol.migr_state) {
+ dprintf("imsm: mark resync complete\n");
+ dev->vol.migr_state = 0;
+ dev->vol.migr_type = 0;
super->updates_pending++;
-
}
+ } else if (!dev->vol.migr_state) {
+ dprintf("imsm: mark '%s' (%llu)\n",
+ failed ? "rebuild" : "initializing", a->resync_start);
+ /* mark that we are rebuilding */
+ map->map_state = failed ? map_state : IMSM_T_STATE_NORMAL;
+ dev->vol.migr_state = 1;
+ dev->vol.migr_type = failed ? 1 : 0;
+ dup_map(dev);
+ a->check_degraded = 1;
+ super->updates_pending++;
}
- /* mark dirty */
- if (dirty && !dev->vol.dirty) {
- dprintf("imsm: mark 'dirty' (%llu)\n", a->resync_start);
- dev->vol.dirty = 1;
+ /* mark dirty / clean */
+ if (dirty != dev->vol.dirty) {
+ dprintf("imsm: mark '%s' (%llu)\n",
+ dirty ? "dirty" : "clean", a->resync_start);
+ dev->vol.dirty = dirty;
super->updates_pending++;
}
+ return consistent;
}
static void imsm_set_disk(struct active_array *a, int n, int state)
int inst = a->info.container_member;
struct intel_super *super = a->container->sb;
struct imsm_dev *dev = get_imsm_dev(super, inst);
- struct imsm_map *map = dev->vol.map;
+ struct imsm_map *map = get_imsm_map(dev, 0);
struct imsm_disk *disk;
__u32 status;
int failed = 0;
if (working == a->info.array.raid_disks) {
map->map_state = IMSM_T_STATE_NORMAL;
+ dev->vol.migr_state = 0;
+ dev->vol.migr_type = 0;
super->updates_pending++;
}
}
super->updates_pending = 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);
+ 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)
+ dl = NULL;
+
+ if (dl)
+ dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
+
+ return dl;
+}
+
+static struct dl *imsm_add_spare(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);
+ unsigned long long esize;
+ unsigned long long pos;
+ struct mdinfo *d;
+ struct extent *ex;
+ int j;
+ int found;
+ __u32 array_start;
+ __u32 status;
+ struct dl *dl;
+
+ for (dl = super->disks; dl; dl = dl->next) {
+ /* If in this array, skip */
+ for (d = a->info.devs ; d ; d = d->next)
+ if (d->disk.major == dl->major &&
+ d->disk.minor == dl->minor) {
+ dprintf("%x:%x already in array\n", dl->major, dl->minor);
+ break;
+ }
+ 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
+ */
+ ex = get_extents(super, dl);
+ if (!ex) {
+ dprintf("cannot get extents\n");
+ continue;
+ }
+ found = 0;
+ j = 0;
+ pos = 0;
+ array_start = __le32_to_cpu(map->pba_of_lba0);
+
+ do {
+ /* check that we can start at pba_of_lba0 with
+ * a->info.component_size of space
+ */
+ esize = ex[j].start - pos;
+ if (array_start >= pos &&
+ array_start + a->info.component_size < ex[j].start) {
+ found = 1;
+ break;
+ }
+ pos = ex[j].start + ex[j].size;
+ j++;
+
+ } while (ex[j-1].size);
+
+ free(ex);
+ if (!found) {
+ dprintf("%x:%x does not have %llu at %d\n",
+ dl->major, dl->minor,
+ a->info.component_size,
+ __le32_to_cpu(map->pba_of_lba0));
+ /* No room */
+ continue;
+ } else
+ break;
+ }
+
+ return dl;
+}
+
static struct mdinfo *imsm_activate_spare(struct active_array *a,
struct metadata_update **updates)
{
struct intel_super *super = a->container->sb;
int inst = a->info.container_member;
struct imsm_dev *dev = get_imsm_dev(super, inst);
- struct imsm_map *map = dev->vol.map;
+ struct imsm_map *map = get_imsm_map(dev, 0);
int failed = a->info.array.raid_disks;
struct mdinfo *rv = NULL;
struct mdinfo *d;
return NULL;
/* For each slot, if it is not working, find a spare */
- dl = super->disks;
for (i = 0; i < a->info.array.raid_disks; i++) {
for (d = a->info.devs ; d ; d = d->next)
if (d->disk.raid_disk == i)
if (d && (d->state_fd >= 0))
continue;
- /* OK, this device needs recovery. Find a spare */
- for ( ; dl ; dl = dl->next) {
- unsigned long long esize;
- unsigned long long pos;
- struct mdinfo *d2;
- struct extent *ex;
- int j;
- int found;
- __u32 array_start;
-
- /* If in this array, skip */
- for (d2 = a->info.devs ; d2 ; d2 = d2->next)
- if (d2->disk.major == dl->major &&
- d2->disk.minor == dl->minor) {
- dprintf("%x:%x already in array\n", dl->major, dl->minor);
- break;
- }
- if (d2)
- continue;
-
- /* Does this unused device have the requisite free space?
- * We need a->info.component_size sectors
- */
- ex = get_extents(super, dl);
- if (!ex) {
- dprintf("cannot get extents\n");
- continue;
- }
- found = 0;
- j = 0;
- pos = 0;
- array_start = __le32_to_cpu(map->pba_of_lba0);
-
- do {
- /* check that we can start at pba_of_lba0 with
- * a->info.component_size of space
- */
- esize = ex[j].start - pos;
- if (array_start >= pos &&
- array_start + a->info.component_size < ex[j].start) {
- found = 1;
- break;
- }
- pos = ex[j].start + ex[j].size;
- j++;
-
- } while (ex[j-1].size);
-
- free(ex);
- if (!found) {
- dprintf("%x:%x does not have %llu at %d\n",
- dl->major, dl->minor,
- a->info.component_size,
- __le32_to_cpu(map->pba_of_lba0));
- /* No room */
- continue;
- }
-
- /* found a usable disk with enough space */
- di = malloc(sizeof(*di));
- memset(di, 0, sizeof(*di));
-
- /* dl->index will be -1 in the case we are activating a
- * pristine spare. imsm_process_update() will create a
- * new index in this case. Once a disk is found to be
- * failed in all member arrays it is kicked from the
- * metadata
- */
- di->disk.number = dl->index;
+ /*
+ * OK, this device needs recovery. Try to re-add the previous
+ * occupant of this slot, if this fails add a new spare
+ */
+ dl = imsm_readd(super, i, a);
+ if (!dl)
+ dl = imsm_add_spare(super, i, a);
+ if (!dl)
+ continue;
+
+ /* found a usable disk with enough space */
+ di = malloc(sizeof(*di));
+ memset(di, 0, sizeof(*di));
+
+ /* dl->index will be -1 in the case we are activating a
+ * pristine spare. imsm_process_update() will create a
+ * new index in this case. Once a disk is found to be
+ * failed in all member arrays it is kicked from the
+ * metadata
+ */
+ di->disk.number = dl->index;
- /* (ab)use di->devs to store a pointer to the device
- * we chose
- */
- di->devs = (struct mdinfo *) dl;
-
- di->disk.raid_disk = i;
- di->disk.major = dl->major;
- di->disk.minor = dl->minor;
- di->disk.state = 0;
- di->data_offset = array_start;
- di->component_size = a->info.component_size;
- di->container_member = inst;
- di->next = rv;
- rv = di;
- num_spares++;
- dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
- i, pos);
+ /* (ab)use di->devs to store a pointer to the device
+ * we chose
+ */
+ di->devs = (struct mdinfo *) dl;
+
+ di->disk.raid_disk = i;
+ di->disk.major = dl->major;
+ di->disk.minor = dl->minor;
+ di->disk.state = 0;
+ di->data_offset = __le32_to_cpu(map->pba_of_lba0);
+ di->component_size = a->info.component_size;
+ di->container_member = inst;
+ di->next = rv;
+ rv = di;
+ num_spares++;
+ dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
+ i, di->data_offset);
- break;
- }
+ break;
}
if (!rv)
return 0;
}
+static void imsm_delete(struct intel_super *super, struct dl **dlp);
+
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;
struct imsm_dev *dev = get_imsm_dev(super, u->array);
- struct imsm_map *map = &dev->vol.map[0];
+ struct imsm_map *map = get_imsm_map(dev, 0);
struct active_array *a;
struct imsm_disk *disk;
__u32 status;
found = 0;
for (a = st->arrays; a ; a = a->next) {
dev = get_imsm_dev(super, a->info.container_member);
- map = &dev->vol.map[0];
+ map = get_imsm_map(dev, 0);
for (i = 0; i < map->num_members; i++)
if (victim == get_imsm_disk_idx(map, i))
found++;
* utilized anywhere
*/
if (!found) {
- disk = get_imsm_disk(super, victim);
+ struct dl **dlp;
+ for (dlp = &super->disks; *dlp; )
+ if ((*dlp)->index == victim)
+ break;
+ disk = &(*dlp)->disk;
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
- */
+ /* We know that 'manager' isn't touching anything,
+ * so it is safe to:
+ */
+ imsm_delete(super, dlp);
}
break;
}
/* check update is next in sequence */
if (u->dev_idx != mpb->num_raid_devs) {
- dprintf("%s: can not create arrays out of sequence\n",
- __func__);
+ dprintf("%s: can not create array %d expected index %d\n",
+ __func__, u->dev_idx, mpb->num_raid_devs);
return;
}
- new_map = &u->dev.vol.map[0];
+ new_map = get_imsm_map(&u->dev, 0);
new_start = __le32_to_cpu(new_map->pba_of_lba0);
new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
*/
for (i = 0; i < mpb->num_raid_devs; i++) {
dev = get_imsm_dev(super, i);
- map = &dev->vol.map[0];
+ map = get_imsm_map(dev, 0);
start = __le32_to_cpu(map->pba_of_lba0);
end = start + __le32_to_cpu(map->blocks_per_member);
if ((new_start >= start && new_start <= end) ||
}
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);
+ 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)
+{
+ struct imsm_super *mpb = super->anchor;
+ struct dl *dl = *dlp;
+ struct dl *iter;
+ struct imsm_dev *dev;
+ struct imsm_map *map;
+ int i, j;
+
+ dprintf("%s: deleting device %x:%x from imsm_super\n",
+ __func__, dl->major, dl->minor);
+
+ /* shift all indexes down one */
+ for (iter = super->disks; iter; iter = iter->next)
+ if (iter->index > dl->index)
+ iter->index--;
+
+ for (i = 0; i < mpb->num_raid_devs; i++) {
+ dev = get_imsm_dev(super, i);
+ map = get_imsm_map(dev, 0);
+
+ for (j = 0; j < map->num_members; j++) {
+ int idx = get_imsm_disk_idx(map, j);
+
+ if (idx > dl->index)
+ map->disk_ord_tbl[j] = __cpu_to_le32(idx - 1);
+ }
+ }
+
+ mpb->num_disks--;
+ super->updates_pending++;
+ *dlp = (*dlp)->next;
+ __free_imsm_disk(dl);
}
struct superswitch super_imsm = {
projects / thirdparty / mdadm.git / blobdiff