static int reshape_super(struct supertype *st, long long size, int level,
int layout, int chunksize, int raid_disks,
- char *backup_file, char *dev, int verbose)
+ int delta_disks, char *backup_file, char *dev,
+ int verbose)
{
/* nothing extra to check in the native case */
if (!st->ss->external)
}
return st->ss->reshape_super(st, size, level, layout, chunksize,
- raid_disks, backup_file, dev, verbose);
+ raid_disks, delta_disks, backup_file, dev,
+ verbose);
}
static void sync_metadata(struct supertype *st)
return rc;
}
-int start_reshape(struct mdinfo *sra)
+int start_reshape(struct mdinfo *sra, int already_running)
{
int err;
sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
err = sysfs_set_num(sra, NULL, "suspend_hi", 0);
err = err ?: sysfs_set_num(sra, NULL, "suspend_lo", 0);
- /* Setting sync_min can fail if the recovery is already 'running',
- * which can happen when restarting an array which is reshaping.
- * So don't worry about errors here */
- sysfs_set_num(sra, NULL, "sync_min", 0);
+ if (!already_running)
+ sysfs_set_num(sra, NULL, "sync_min", 0);
err = err ?: sysfs_set_num(sra, NULL, "sync_max", 0);
- err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
+ if (!already_running)
+ err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
return err;
}
sysfs_set_str(sra, sd, "state", "faulty");
sysfs_set_str(sra, sd, "slot", "none");
- sysfs_set_str(sra, sd, "state", "remove");
+ /* for external metadata disks should be removed in mdmon */
+ if (!st->ss->external)
+ sysfs_set_str(sra, sd, "state", "remove");
sd->disk.state |= (1<<MD_DISK_REMOVED);
sd->disk.state &= ~(1<<MD_DISK_SYNC);
sd->next = sra->devs;
}
memset(buf, 0, 512);
- for (i=0; i < blocks + 1 ; i++) {
+ for (i=0; i < blocks + 8 ; i++) {
if (write(*fdlist, buf, 512) != 512) {
fprintf(stderr, Name ": %s: cannot create"
" backup file %s: %s\n",
* when assembling an array that is undergoing reshape.
*/
int new_disks;
+ /* delta_parity records change in number of devices
+ * caused by level change
+ */
+ int delta_parity = 0;
/* If a new level not explicitly given, we assume no-change */
if (info->new_level == UnSet)
* raid0 with 1 disk
*/
if (info->new_level == 0) {
+ if (info->delta_disks != UnSet &&
+ info->delta_disks != 0)
+ return "Cannot change number of disks "
+ "with RAID1->RAID0 conversion";
re->level = 0;
re->before.data_disks = 1;
re->after.data_disks = 1;
}
if (info->array.raid_disks == 2 &&
info->new_level == 5) {
+
re->level = 5;
re->before.data_disks = 1;
- re->after.data_disks = 1;
+ if (info->delta_disks != UnSet &&
+ info->delta_disks != 0)
+ re->after.data_disks = 1 + info->delta_disks;
+ else
+ re->after.data_disks = 1;
+ if (re->after.data_disks < 1)
+ return "Number of disks too small for RAID5";
+
re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
info->array.chunk_size = 65536;
break;
return "RAID10 can only be changed to RAID0";
new_disks = (info->array.raid_disks
/ (info->array.layout & 0xff));
- if (info->delta_disks == UnSet) {
+ if (info->delta_disks == UnSet)
info->delta_disks = (new_disks
- info->array.raid_disks);
- }
+
if (info->delta_disks != new_disks - info->array.raid_disks)
return "New number of raid-devices impossible for RAID10";
if (info->new_chunk &&
* a raid4 style layout of the final level.
*/
switch (info->new_level) {
- case 0:
case 4:
+ delta_parity = 1;
+ case 0:
re->level = 4;
re->before.layout = 0;
break;
case 5:
+ delta_parity = 1;
re->level = 5;
re->before.layout = ALGORITHM_PARITY_N;
break;
case 6:
+ delta_parity = 2;
re->level = 6;
re->before.layout = ALGORITHM_PARITY_N;
break;
info->array.layout = ALGORITHM_PARITY_N;
case 5:
switch (info->new_level) {
+ case 0:
+ delta_parity = -1;
case 4:
re->level = info->array.level;
re->before.data_disks = info->array.raid_disks - 1;
re->before.layout = info->array.layout;
break;
case 6:
+ delta_parity = 1;
re->level = 6;
re->before.data_disks = info->array.raid_disks - 1;
switch (info->array.layout) {
case 1:
if (info->array.raid_disks != 2)
return "Can only convert a 2-device array to RAID1";
+ if (info->delta_disks != UnSet &&
+ info->delta_disks != 0)
+ return "Cannot set raid_disk when "
+ "converting RAID5->RAID1";
re->level = 1;
break;
default:
switch (info->new_level) {
case 4:
case 5:
+ delta_parity = -1;
case 6:
re->level = 6;
re->before.data_disks = info->array.raid_disks - 2;
*/
if (re->level != 4 && re->level != 5)
return "Cannot covert to RAID0 from this level";
- if (info->delta_disks == UnSet)
- re->after.data_disks = re->before.data_disks;
- else
- re->after.data_disks =
- info->array.raid_disks + info->delta_disks;
+
switch (re->level) {
case 4:
re->after.layout = 0 ; break;
/* We can only get to RAID4 from RAID5 */
if (re->level != 4 && re->level != 5)
return "Cannot convert to RAID4 from this level";
- if (info->delta_disks == UnSet)
- re->after.data_disks = re->before.data_disks;
- else
- re->after.data_disks =
- re->before.data_disks + info->delta_disks;
+
switch (re->level) {
case 4:
re->after.layout = 0 ; break;
/* We get to RAID5 for RAID5 or RAID6 */
if (re->level != 5 && re->level != 6)
return "Cannot convert to RAID5 from this level";
- if (info->delta_disks == UnSet)
- re->after.data_disks = re->before.data_disks;
- else if (re->level == 5)
- re->after.data_disks =
- re->before.data_disks + info->delta_disks;
- else
- re->after.data_disks =
- info->array.raid_disks + info->delta_disks - 1;
+
switch (re->level) {
case 5:
if (info->new_layout == UnSet)
/* We must already be at level 6 */
if (re->level != 6)
return "Impossible level change";
- if (info->delta_disks == UnSet)
- re->after.data_disks = re->before.data_disks;
- else
- re->after.data_disks = (info->array.raid_disks +
- info->delta_disks) - 2;
if (info->new_layout == UnSet)
re->after.layout = info->array.layout;
else
default:
return "Impossible level change requested";
}
+ if (info->delta_disks == UnSet)
+ info->delta_disks = delta_parity;
+
+ re->after.data_disks = (re->before.data_disks
+ + info->delta_disks
+ - delta_parity);
switch (re->level) {
case 6: re->parity = 2; break;
case 4:
struct supertype *st, struct mdinfo *info,
int force, char *backup_file, int quiet, int forked,
int restart);
-static int reshape_container(char *container, int cfd, char *devname,
+static int reshape_container(char *container, char *devname,
struct supertype *st,
struct mdinfo *info,
int force,
char *backup_file,
- int quiet);
+ int quiet, int restart);
int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
long long size,
if (size >= 0 && (size == 0 || size != array.size)) {
long long orig_size = array.size;
- if (reshape_super(st, size, UnSet, UnSet, 0, 0, NULL, devname, !quiet)) {
+ if (reshape_super(st, size, UnSet, UnSet, 0, 0, UnSet, NULL,
+ devname, !quiet)) {
rv = 1;
goto release;
}
/* restore metadata */
if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
- NULL, devname, !quiet) == 0)
+ UnSet, NULL, devname, !quiet) == 0)
sync_metadata(st);
fprintf(stderr, Name ": Cannot set device size for %s: %s\n",
devname, strerror(err));
info.component_size = size*2;
info.new_level = level;
info.new_chunk = chunksize * 1024;
- if (raid_disks)
+ if (info.array.level == LEVEL_CONTAINER) {
+ info.delta_disks = UnSet;
+ info.array.raid_disks = raid_disks;
+ } else if (raid_disks)
info.delta_disks = raid_disks - info.array.raid_disks;
else
info.delta_disks = UnSet;
* number of devices (On-Line Capacity Expansion) must be
* performed at the level of the container
*/
- rv = reshape_container(container, fd, devname, st, &info,
- force, backup_file, quiet);
+ rv = reshape_container(container, devname, st, &info,
+ force, backup_file, quiet, 0);
frozen = 0;
} else {
+ /* get spare devices from external metadata
+ */
+ if (st->ss->external) {
+ struct mdinfo *info2;
+
+ info2 = st->ss->container_content(st, subarray);
+ if (info2) {
+ info.array.spare_disks =
+ info2->array.spare_disks;
+ sysfs_free(info2);
+ }
+ }
+
/* Impose these changes on a single array. First
* check that the metadata is OK with the change. */
if (reshape_super(st, info.component_size, info.new_level,
info.new_layout, info.new_chunk,
- info.array.raid_disks + info.delta_disks,
+ info.array.raid_disks, info.delta_disks,
backup_file, devname, quiet)) {
rv = 1;
goto release;
int done;
struct mdinfo *sra = NULL;
- msg = analyse_change(info, &reshape);
+ /* when reshaping a RAID0, the component_size might be zero.
+ * So try to fix that up.
+ */
+ if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
+ dprintf("Cannot get array information.\n");
+ goto release;
+ }
+ if (array.level == 0 && info->component_size == 0) {
+ get_dev_size(fd, NULL, &array_size);
+ info->component_size = array_size / array.raid_disks;
+ }
+
+ if (info->reshape_active) {
+ int new_level = info->new_level;
+ info->new_level = UnSet;
+ msg = analyse_change(info, &reshape);
+ info->new_level = new_level;
+ if (!restart)
+ /* Make sure the array isn't read-only */
+ ioctl(fd, RESTART_ARRAY_RW, 0);
+ } else
+ msg = analyse_change(info, &reshape);
if (msg) {
fprintf(stderr, Name ": %s\n", msg);
goto release;
}
- if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
- dprintf("Canot get array information.\n");
+ if (restart &&
+ (reshape.level != info->array.level ||
+ reshape.before.layout != info->array.layout ||
+ reshape.before.data_disks + reshape.parity != info->array.raid_disks)) {
+ fprintf(stderr, Name ": reshape info is not in native format -"
+ " cannot continue.\n");
goto release;
}
* freeze_array and freeze_container.
*/
sysfs_freeze_array(info);
+ /* Check we have enough spares to not be degraded */
spares_needed = max(reshape.before.data_disks,
reshape.after.data_disks)
+ reshape.parity - array.raid_disks;
info->array.spare_disks);
goto release;
}
+ /* Check we have enough spares to not fail */
+ spares_needed = max(reshape.before.data_disks,
+ reshape.after.data_disks)
+ - array.raid_disks;
+ if ((info->new_level > 1 || info->new_level == 0) &&
+ spares_needed > info->array.spare_disks) {
+ fprintf(stderr,
+ Name ": Need %d spare%s to create working array,"
+ " and only have %d.\n",
+ spares_needed,
+ spares_needed == 1 ? "" : "s",
+ info->array.spare_disks);
+ goto release;
+ }
- if (reshape.level != info->array.level) {
+ if (reshape.level != array.level) {
char *c = map_num(pers, reshape.level);
int err;
if (c == NULL)
if (!quiet)
fprintf(stderr, Name ": level of %s changed to %s\n",
devname, c);
- orig_level = info->array.level;
+ orig_level = array.level;
+ sysfs_freeze_array(info);
if (reshape.level > 0 && st->ss->external) {
/* make sure mdmon is aware of the new level */
/* No restriping needed, but we might need to impose
* some more changes: layout, raid_disks, chunk_size
*/
+ /* read current array info */
+ if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
+ dprintf("Cannot get array information.\n");
+ goto release;
+ }
+ /* compare current array info with new values and if
+ * it is different update them to new */
if (info->new_layout != UnSet &&
- info->new_layout != info->array.layout) {
- info->array.layout = info->new_layout;
- if (ioctl(fd, SET_ARRAY_INFO, &info->array) != 0) {
+ info->new_layout != array.layout) {
+ array.layout = info->new_layout;
+ if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
fprintf(stderr, Name ": failed to set new layout\n");
goto release;
} else if (!quiet)
printf("layout for %s set to %d\n",
- devname, info->array.layout);
+ devname, array.layout);
}
if (info->delta_disks != UnSet &&
- info->delta_disks != 0) {
- info->array.raid_disks += info->delta_disks;
- if (ioctl(fd, SET_ARRAY_INFO, &info->array) != 0) {
+ info->delta_disks != 0 &&
+ array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
+ array.raid_disks += info->delta_disks;
+ if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
fprintf(stderr, Name ": failed to set raid disks\n");
goto release;
- } else if (!quiet)
+ } else if (!quiet) {
printf("raid_disks for %s set to %d\n",
- devname, info->array.raid_disks);
+ devname, array.raid_disks);
+ }
}
if (info->new_chunk != 0 &&
- info->new_chunk != info->array.chunk_size) {
+ info->new_chunk != array.chunk_size) {
if (sysfs_set_num(info, NULL,
"chunk_size", info->new_chunk) != 0) {
fprintf(stderr, Name ": failed to set chunk size\n");
goto release;
} else if (!quiet)
printf("chunk size for %s set to %d\n",
- devname, info->array.chunk_size);
+ devname, array.chunk_size);
}
unfreeze(st);
return 0;
/* Now we need to open all these devices so we can read/write.
*/
- nrdisks = array.raid_disks + sra->array.spare_disks;
+ nrdisks = max(reshape.before.data_disks,
+ reshape.after.data_disks) + reshape.parity
+ + sra->array.spare_disks;
fdlist = malloc((1+nrdisks) * sizeof(int));
offsets = malloc((1+nrdisks) * sizeof(offsets[0]));
if (!fdlist || !offsets) {
goto release;
}
- d = reshape_prepare_fdlist(devname, sra, array.raid_disks,
+ odisks = reshape.before.data_disks + reshape.parity;
+ d = reshape_prepare_fdlist(devname, sra, odisks,
nrdisks, blocks, backup_file,
fdlist, offsets);
if (d < 0) {
sra->new_chunk = info->new_chunk;
- if (info->reshape_active)
+ if (restart)
sra->reshape_progress = info->reshape_progress;
else {
sra->reshape_progress = 0;
reshape.before.layout == reshape.after.layout &&
st->ss->external == 0) {
/* use SET_ARRAY_INFO but only if reshape hasn't started */
+ ioctl(fd, GET_ARRAY_INFO, &array);
array.raid_disks = reshape.after.data_disks + reshape.parity;
- if (!info->reshape_active &&
+ if (!restart &&
ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
int err = errno;
goto release;
}
- } else {
+ } else if (!restart) {
/* set them all just in case some old 'new_*' value
* persists from some earlier problem.
- * We even set them when restarting in the middle. They will
- * already be set in that case so this will be a no-op,
- * but it is hard to tell the difference.
*/
int err = 0;
if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
}
}
- start_reshape(sra);
+ err = start_reshape(sra, restart);
+ if (err) {
+ fprintf(stderr,
+ Name ": Cannot %s reshape for %s\n",
+ restart ? "continue" : "start",
+ devname);
+ goto release;
+ }
if (restart)
sysfs_set_str(sra, NULL, "array_state", "active");
fd = -1;
mlockall(MCL_FUTURE);
- odisks = reshape.before.data_disks + reshape.parity;
-
if (st->ss->external) {
/* metadata handler takes it from here */
done = st->ss->manage_reshape(
out:
if (forked)
return 0;
+ unfreeze(st);
exit(0);
release:
return 1;
}
-int reshape_container(char *container, int cfd, char *devname,
+int reshape_container(char *container, char *devname,
struct supertype *st,
struct mdinfo *info,
int force,
char *backup_file,
- int quiet)
+ int quiet, int restart)
{
struct mdinfo *cc = NULL;
/* component_size is not meaningful for a container,
* so pass '-1' meaning 'no change'
*/
- if (reshape_super(st, -1, info->new_level,
+ if (!restart &&
+ reshape_super(st, -1, info->new_level,
info->new_layout, info->new_chunk,
- info->array.raid_disks + info->delta_disks,
- backup_file, devname, quiet))
+ info->array.raid_disks, info->delta_disks,
+ backup_file, devname, quiet)) {
+ unfreeze(st);
return 1;
+ }
sync_metadata(st);
switch (fork()) {
case -1: /* error */
perror("Cannot fork to complete reshape\n");
+ unfreeze(st);
return 1;
default: /* parent */
printf(Name ": multi-array reshape continues in background\n");
* reshape it. reshape_array() will re-read the metadata
* so the next time through a different array should be
* ready for reshape.
+ * It is possible that the 'different' array will not
+ * be assembled yet. In that case we simple exit.
+ * When it is assembled, the mdadm which assembles it
+ * will take over the reshape.
*/
struct mdinfo *content;
int rv;
rv = reshape_array(container, fd, adev, st,
content, force,
- backup_file, quiet, 1, 0);
+ backup_file, quiet, 1, restart);
close(fd);
+ restart = 0;
if (rv)
break;
}
* 1 if more data from backup_point - but only as far as suspend_point,
* should be backed up
* 0 if things are progressing smoothly
- * -1 if the reshape is finished, either because it is all done,
- * or due to an error.
+ * -1 if the reshape is finished because it is all done,
+ * -2 if the reshape is finished due to an error.
*/
int advancing = (reshape->after.data_disks
/* Round to chunk size as some kernels give an erroneously high number */
max_progress /= info->new_chunk/512;
max_progress *= info->new_chunk/512;
+ /* And round to old chunk size as the kernel wants that */
+ max_progress /= info->array.chunk_size/512;
+ max_progress *= info->array.chunk_size/512;
/* Limit progress to the whole device */
if (max_progress > info->component_size)
max_progress = info->component_size;
goto check_progress;
}
}
+ /* Some kernels reset 'sync_completed' to zero,
+ * we need to have real point we are in md
+ */
+ if (completed == 0)
+ completed = max_progress;
+
/* some kernels can give an incorrectly high 'completed' number */
completed /= (info->new_chunk/512);
completed *= (info->new_chunk/512);
if (offset < suspend_point/data)
break;
}
+ if (actual_stripes == 0)
+ break;
grow_backup(sra, offset, actual_stripes,
fds, offsets,
disks, chunk, level, layout,
int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
char *backup_file)
{
- int err = sysfs_set_str(info, NULL, "array_state", "readonly");
+ char buf[40];
+ char *container = NULL;
+ int err;
+
+ err = sysfs_set_str(info, NULL, "array_state", "readonly");
if (err)
return err;
- return reshape_array(NULL, mdfd, "array", st, info, 1, backup_file, 0, 0, 1);
-}
+ if (st->ss->external) {
+ fmt_devname(buf, st->container_dev);
+ container = buf;
+ freeze(st);
+ if (!mdmon_running(st->container_dev))
+ start_mdmon(st->container_dev);
+ ping_monitor(devnum2devname(st->container_dev));
+
+ if (info->reshape_active == 2) {
+ int cfd = open_dev(st->container_dev);
+ if (cfd < 0)
+ return 1;
+ st->ss->load_container(st, cfd, container);
+ close(cfd);
+ return reshape_container(container, NULL,
+ st, info, 0, backup_file,
+ 0, 1);
+ }
+ }
+ return reshape_array(container, mdfd, "array", st, info, 1,
+ backup_file, 0, 0, 1);
+}
projects / thirdparty / mdadm.git / blobdiff