int nfd, fd2;
int d, nd;
struct supertype *st = NULL;
-
+ char *subarray = NULL;
if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
fprintf(stderr, Name ": cannot get array info for %s\n", devname);
return 1;
}
- st = super_by_fd(fd);
+ if (info.array.level != -1) {
+ fprintf(stderr, Name ": can only add devices to linear arrays\n");
+ return 1;
+ }
+
+ st = super_by_fd(fd, &subarray);
if (!st) {
fprintf(stderr, Name ": cannot handle arrays with superblock version %d\n", info.array.major_version);
return 1;
}
- if (info.array.level != -1) {
- fprintf(stderr, Name ": can only add devices to linear arrays\n");
- return 1;
+ if (subarray) {
+ fprintf(stderr, Name ": Cannot grow linear sub-arrays yet\n");
+ free(subarray);
+ free(st);
}
nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
if (nfd < 0) {
fprintf(stderr, Name ": cannot open %s\n", newdev);
+ free(st);
return 1;
}
fstat(nfd, &stb);
if ((stb.st_mode & S_IFMT) != S_IFBLK) {
fprintf(stderr, Name ": %s is not a block device!\n", newdev);
close(nfd);
+ free(st);
return 1;
}
/* now check out all the devices and make sure we can read the superblock */
mdu_disk_info_t disk;
char *dv;
+ st->ss->free_super(st);
+
disk.number = d;
if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
fprintf(stderr, Name ": cannot get device detail for device %d\n",
d);
+ close(nfd);
+ free(st);
return 1;
}
dv = map_dev(disk.major, disk.minor, 1);
if (!dv) {
fprintf(stderr, Name ": cannot find device file for device %d\n",
d);
+ close(nfd);
+ free(st);
return 1;
}
fd2 = dev_open(dv, O_RDWR);
if (!fd2) {
fprintf(stderr, Name ": cannot open device file %s\n", dv);
+ close(nfd);
+ free(st);
return 1;
}
- st->ss->free_super(st);
if (st->ss->load_super(st, fd2, NULL)) {
fprintf(stderr, Name ": cannot find super block on %s\n", dv);
+ close(nfd);
close(fd2);
+ free(st);
return 1;
}
close(fd2);
mdu_bitmap_file_t bmf;
mdu_array_info_t array;
struct supertype *st;
+ char *subarray = NULL;
int major = BITMAP_MAJOR_HI;
int vers = md_get_version(fd);
unsigned long long bitmapsize, array_size;
if (vers < 9003) {
major = BITMAP_MAJOR_HOSTENDIAN;
-#ifdef __BIG_ENDIAN
- fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
- " between different architectured. Consider upgrading the Linux kernel.\n");
-#endif
+ fprintf(stderr, Name ": Warning - bitmaps created on this kernel"
+ " are not portable\n"
+ " between different architectures. Consider upgrading"
+ " the Linux kernel.\n");
}
if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
devname);
return 1;
}
+
+ if (strcmp(file, "none") == 0) {
+ fprintf(stderr, Name ": no bitmap found on %s\n", devname);
+ return 1;
+ }
if (array.level <= 0) {
fprintf(stderr, Name ": Bitmaps not meaningful with level %s\n",
map_num(pers, array.level)?:"of this array");
bitmapsize = bitmapsize * array.raid_disks / ncopies;
}
- st = super_by_fd(fd);
+ st = super_by_fd(fd, &subarray);
if (!st) {
fprintf(stderr, Name ": Cannot understand version %d.%d\n",
array.major_version, array.minor_version);
return 1;
}
- if (strcmp(file, "none") == 0) {
- fprintf(stderr, Name ": no bitmap found on %s\n", devname);
+ if (subarray) {
+ fprintf(stderr, Name ": Cannot add bitmaps to sub-arrays yet\n");
+ free(subarray);
+ free(st);
return 1;
- } else if (strcmp(file, "internal") == 0) {
+ }
+ if (strcmp(file, "internal") == 0) {
int d;
+ if (st->ss->add_internal_bitmap == NULL) {
+ fprintf(stderr, Name ": Internal bitmaps not supported "
+ "with %s metadata\n", st->ss->name);
+ return 1;
+ }
for (d=0; d< st->max_devs; d++) {
mdu_disk_info_t disk;
char *dv;
}
array.state |= (1<<MD_SB_BITMAP_PRESENT);
if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
+ if (errno == EBUSY)
+ fprintf(stderr, Name
+ ": Cannot add bitmap while array is"
+ " resyncing or reshaping etc.\n");
fprintf(stderr, Name ": failed to set internal bitmap.\n");
return 1;
}
return 1;
}
if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
+ int err = errno;
+ if (errno == EBUSY)
+ fprintf(stderr, Name
+ ": Cannot add bitmap while array is"
+ " resyncing or reshaping etc.\n");
fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
- devname, strerror(errno));
+ devname, strerror(err));
return 1;
}
}
__u8 pad[512-68-32];
} __attribute__((aligned(512))) bsb, bsb2;
-int bsb_csum(char *buf, int len)
+static __u32 bsb_csum(char *buf, int len)
{
int i;
int csum = 0;
return __cpu_to_le32(csum);
}
-static int child_grow(int afd, struct mdinfo *sra, unsigned long blocks,
- int *fds, unsigned long long *offsets,
- int disks, int chunk, int level, int layout, int data,
- int dests, int *destfd, unsigned long long *destoffsets);
-static int child_shrink(int afd, struct mdinfo *sra, unsigned long blocks,
- int *fds, unsigned long long *offsets,
- int disks, int chunk, int level, int layout, int data,
- int dests, int *destfd, unsigned long long *destoffsets);
-static int child_same_size(int afd, struct mdinfo *sra, unsigned long blocks,
- int *fds, unsigned long long *offsets,
- unsigned long long start,
- int disks, int chunk, int level, int layout, int data,
- int dests, int *destfd, unsigned long long *destoffsets);
-
-int freeze_array(struct mdinfo *sra)
+static int check_idle(struct supertype *st)
+{
+ /* Check that all member arrays for this container, or the
+ * container of this array, are idle
+ */
+ int container_dev = (st->container_dev != NoMdDev
+ ? st->container_dev : st->devnum);
+ char container[40];
+ struct mdstat_ent *ent, *e;
+ int is_idle = 1;
+
+ fmt_devname(container, container_dev);
+ ent = mdstat_read(0, 0);
+ for (e = ent ; e; e = e->next) {
+ if (!is_container_member(e, container))
+ continue;
+ if (e->percent >= 0) {
+ is_idle = 0;
+ break;
+ }
+ }
+ free_mdstat(ent);
+ return is_idle;
+}
+
+static int freeze_container(struct supertype *st)
+{
+ int container_dev = (st->container_dev != NoMdDev
+ ? st->container_dev : st->devnum);
+ char container[40];
+
+ if (!check_idle(st))
+ return -1;
+
+ fmt_devname(container, container_dev);
+
+ if (block_monitor(container, 1)) {
+ fprintf(stderr, Name ": failed to freeze container\n");
+ return -2;
+ }
+
+ return 1;
+}
+
+static void unfreeze_container(struct supertype *st)
+{
+ int container_dev = (st->container_dev != NoMdDev
+ ? st->container_dev : st->devnum);
+ char container[40];
+
+ fmt_devname(container, container_dev);
+
+ unblock_monitor(container, 1);
+}
+
+static int freeze(struct supertype *st)
{
- /* Try to freeze resync on this array.
+ /* Try to freeze resync/rebuild on this array/container.
* Return -1 if the array is busy,
+ * return -2 container cannot be frozen,
* return 0 if this kernel doesn't support 'frozen'
* return 1 if it worked.
*/
- char buf[20];
- if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
- return 0;
- if (strcmp(buf, "idle\n") != 0 &&
- strcmp(buf, "frozen\n") != 0)
- return -1;
- if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
- return 0;
- return 1;
+ if (st->ss->external)
+ return freeze_container(st);
+ else {
+ struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
+ int err;
+
+ if (!sra)
+ return -1;
+ err = sysfs_freeze_array(sra);
+ sysfs_free(sra);
+ return err;
+ }
}
-void unfreeze_array(struct mdinfo *sra, int frozen)
+static void unfreeze(struct supertype *st, int frozen)
{
/* If 'frozen' is 1, unfreeze the array */
- if (frozen > 0)
- sysfs_set_str(sra, NULL, "sync_action", "idle");
+ if (frozen <= 0)
+ return;
+
+ if (st->ss->external)
+ return unfreeze_container(st);
+ else {
+ struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
+
+ if (sra)
+ sysfs_set_str(sra, NULL, "sync_action", "idle");
+ else
+ fprintf(stderr, Name ": failed to unfreeze array\n");
+ sysfs_free(sra);
+ }
}
-void wait_reshape(struct mdinfo *sra)
+static void wait_reshape(struct mdinfo *sra)
{
int fd = sysfs_get_fd(sra, NULL, "sync_action");
char action[20];
- do {
+ if (fd < 0)
+ return;
+
+ while (sysfs_fd_get_str(fd, action, 20) > 0 &&
+ strncmp(action, "reshape", 7) == 0) {
fd_set rfds;
FD_ZERO(&rfds);
FD_SET(fd, &rfds);
select(fd+1, NULL, NULL, &rfds, NULL);
-
- if (sysfs_fd_get_str(fd, action, 20) < 0) {
- close(fd);
- return;
- }
- } while (strncmp(action, "reshape", 7) == 0);
-}
-
-
-int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
- long long size,
- int level, char *layout_str, int chunksize, int raid_disks)
-{
- /* Make some changes in the shape of an array.
- * The kernel must support the change.
- *
- * There are three different changes. Each can trigger
- * a resync or recovery so we freeze that until we have
- * requested everything (if kernel supports freezing - 2.6.30).
- * The steps are:
- * - change size (i.e. component_size)
- * - change level
- * - change layout/chunksize/ndisks
- *
- * The last can require a reshape. It is different on different
- * levels so we need to check the level before actioning it.
- * Some times the level change needs to be requested after the
- * reshape (e.g. raid6->raid5, raid5->raid0)
- *
- */
- struct mdu_array_info_s array, orig;
- char *c;
- int rv = 0;
- struct supertype *st;
-
- int nchunk, ochunk;
- int nlayout, olayout;
- int ndisks, odisks;
- int ndata, odata;
- int orig_level = UnSet;
- char alt_layout[40];
- int *fdlist;
- unsigned long long *offsets;
- int d, i;
- int nrdisks;
- int err;
- int frozen;
- unsigned long a,b, blocks, stripes;
- int cache;
- unsigned long long array_size;
- int changed = 0;
- int done;
-
- struct mdinfo *sra;
- struct mdinfo *sd;
-
- if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
- fprintf(stderr, Name ": %s is not an active md array - aborting\n",
- devname);
- return 1;
}
+ close(fd);
+}
- if (size >= 0 &&
- (chunksize || level!= UnSet || layout_str || raid_disks)) {
- fprintf(stderr, Name ": cannot change component size at the same time "
- "as other changes.\n"
- " Change size first, then check data is intact before "
- "making other changes.\n");
+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)
+{
+ /* nothing extra to check in the native case */
+ if (!st->ss->external)
+ return 0;
+ if (!st->ss->reshape_super ||
+ !st->ss->manage_reshape) {
+ fprintf(stderr, Name ": %s metadata does not support reshape\n",
+ st->ss->name);
return 1;
}
- if (raid_disks && raid_disks < array.raid_disks && array.level > 1 &&
- get_linux_version() < 2006032 &&
- !check_env("MDADM_FORCE_FEWER")) {
- fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n"
- " Please use a newer kernel\n");
- return 1;
- }
- sra = sysfs_read(fd, 0, GET_LEVEL);
- frozen = freeze_array(sra);
- if (frozen < 0) {
- fprintf(stderr, Name ": %s is performing resync/recovery and cannot"
- " be reshaped\n", devname);
- return 1;
- }
+ return st->ss->reshape_super(st, size, level, layout, chunksize,
+ raid_disks, backup_file, dev, verbose);
+}
- /* ========= set size =============== */
- if (size >= 0 && (size == 0 || size != array.size)) {
- array.size = size;
- if (array.size != size) {
- /* got truncated to 32bit, write to
- * component_size instead
- */
- if (sra)
- rv = sysfs_set_num(sra, NULL,
- "component_size", size);
- else
- rv = -1;
+static void sync_metadata(struct supertype *st)
+{
+ if (st->ss->external) {
+ if (st->update_tail) {
+ flush_metadata_updates(st);
+ st->update_tail = &st->updates;
} else
- rv = ioctl(fd, SET_ARRAY_INFO, &array);
- if (rv != 0) {
- int err = errno;
- fprintf(stderr, Name ": Cannot set device size for %s: %s\n",
- devname, strerror(err));
- if (err == EBUSY &&
- (array.state & (1<<MD_SB_BITMAP_PRESENT)))
- fprintf(stderr, " Bitmap must be removed before size can be changed\n");
- rv = 1;
- goto release;
- }
- ioctl(fd, GET_ARRAY_INFO, &array);
- size = get_component_size(fd)/2;
- if (size == 0)
- size = array.size;
- if (!quiet)
- fprintf(stderr, Name ": component size of %s has been set to %lluK\n",
- devname, size);
- changed = 1;
- } else {
- size = get_component_size(fd)/2;
- if (size == 0)
- size = array.size;
+ st->ss->sync_metadata(st);
}
+}
- /* ======= set level =========== */
- if (level != UnSet && level != array.level) {
- /* Trying to change the level.
- * We might need to change layout first and schedule a
- * level change for later.
- * Level changes that can happen immediately are:
- * 0->4,5,6 1->5 4->5,6 5->1,6
- * Level changes that need a layout change first are:
- * 6->5,4,0 : need a -6 layout, or parity-last
- * 5->4,0 : need parity-last
- */
- if ((array.level == 6 || array.level == 5) &&
- (level == 5 || level == 4 || level == 0)) {
- /* Don't change level yet, but choose intermediate
- * layout
- */
- if (level == 5) {
- if (layout_str == NULL)
- switch (array.layout) {
- case ALGORITHM_LEFT_ASYMMETRIC:
- case ALGORITHM_LEFT_ASYMMETRIC_6:
- case ALGORITHM_ROTATING_N_RESTART:
- layout_str = "left-asymmetric-6";
- break;
- case ALGORITHM_LEFT_SYMMETRIC:
- case ALGORITHM_LEFT_SYMMETRIC_6:
- case ALGORITHM_ROTATING_N_CONTINUE:
- layout_str = "left-symmetric-6";
- break;
- case ALGORITHM_RIGHT_ASYMMETRIC:
- case ALGORITHM_RIGHT_ASYMMETRIC_6:
- case ALGORITHM_ROTATING_ZERO_RESTART:
- layout_str = "right-asymmetric-6";
- break;
- case ALGORITHM_RIGHT_SYMMETRIC:
- case ALGORITHM_RIGHT_SYMMETRIC_6:
- layout_str = "right-symmetric-6";
- break;
- case ALGORITHM_PARITY_0:
- case ALGORITHM_PARITY_0_6:
- layout_str = "parity-first-6";
- break;
- case ALGORITHM_PARITY_N:
- layout_str = "parity-last";
- break;
- default:
- fprintf(stderr, Name ": %s: cannot"
- "convert layout to RAID5 equivalent\n",
- devname);
- rv = 1;
- goto release;
- }
- else {
- int l = map_name(r5layout, layout_str);
- if (l == UnSet) {
- fprintf(stderr, Name ": %s: layout '%s' not recognised\n",
- devname, layout_str);
- rv = 1;
- goto release;
- }
- if (l != ALGORITHM_PARITY_N) {
- /* need the -6 version */
- char *ls = map_num(r5layout, l);
- strcat(strcpy(alt_layout, ls),
- "-6");
- layout_str = alt_layout;
- }
- }
- if (raid_disks)
- /* The final raid6->raid5 conversion
- * will reduce the number of disks,
- * so now we need to aim higher
- */
- raid_disks++;
- } else
- layout_str = "parity-last";
- } else {
- c = map_num(pers, level);
- if (c == NULL) {
- rv = 1;/* not possible */
- goto release;
- }
- err = sysfs_set_str(sra, NULL, "level", c);
- if (err) {
- err = errno;
- fprintf(stderr, Name ": %s: could not set level to %s\n",
- devname, c);
- if (err == EBUSY &&
- (array.state & (1<<MD_SB_BITMAP_PRESENT)))
- fprintf(stderr, " Bitmap must be removed before level can be changed\n");
- rv = 1;
- goto release;
- }
- orig = array;
- orig_level = orig.level;
- ioctl(fd, GET_ARRAY_INFO, &array);
- if (layout_str == NULL &&
- orig.level == 5 && level == 6 &&
- array.layout != orig.layout)
- layout_str = map_num(r5layout, orig.layout);
- if (!quiet)
- fprintf(stderr, Name " level of %s changed to %s\n",
- devname, c);
- changed = 1;
- }
- }
-
- /* ========= set shape (chunk_size / layout / ndisks) ============== */
- /* Check if layout change is a no-op */
- if (layout_str) switch(array.level) {
- case 5:
- if (array.layout == map_name(r5layout, layout_str))
- layout_str = NULL;
- break;
- case 6:
- if (layout_str == NULL &&
- ((chunksize && chunksize * 1024 != array.chunk_size) ||
- (raid_disks && raid_disks != array.raid_disks)) &&
- array.layout >= 16) {
- fprintf(stderr, Name
- ": %s has a non-standard layout. If you wish to preserve this\n"
- " during the reshape, please specify --layout=preserve\n"
- " If you want to change it, specify a layout or use --layout=normalise\n",
- devname);
- rv = 1;
- goto release;
- }
- if (strcmp(layout_str, "normalise") == 0 ||
- strcmp(layout_str, "normalize") == 0) {
- char *hyphen;
- strcpy(alt_layout, map_num(r6layout, array.layout));
- hyphen = strrchr(alt_layout, '-');
- if (hyphen && strcmp(hyphen, "-6") == 0) {
- *hyphen = 0;
- layout_str = alt_layout;
- }
- }
+static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
+{
+ /* when dealing with external metadata subarrays we need to be
+ * prepared to handle EAGAIN. The kernel may need to wait for
+ * mdmon to mark the array active so the kernel can handle
+ * allocations/writeback when preparing the reshape action
+ * (md_allow_write()). We temporarily disable safe_mode_delay
+ * to close a race with the array_state going clean before the
+ * next write to raid_disks / stripe_cache_size
+ */
+ char safe[50];
+ int rc;
- if (array.layout == map_name(r6layout, layout_str))
- layout_str = NULL;
- if (layout_str && strcmp(layout_str, "preserve") == 0)
- layout_str = NULL;
- break;
- }
- if (layout_str == NULL
- && (chunksize == 0 || chunksize*1024 == array.chunk_size)
- && (raid_disks == 0 || raid_disks == array.raid_disks)) {
- rv = 0;
- if (level != UnSet && level != array.level) {
- /* Looks like this level change doesn't need
- * a reshape after all.
- */
- c = map_num(pers, level);
- if (c) {
- rv = sysfs_set_str(sra, NULL, "level", c);
- if (rv) {
- int err = errno;
- fprintf(stderr, Name ": %s: could not set level to %s\n",
- devname, c);
- if (err == EBUSY &&
- (array.state & (1<<MD_SB_BITMAP_PRESENT)))
- fprintf(stderr, " Bitmap must be removed before level can be changed\n");
- }
- }
- } else if (!changed && !quiet)
- fprintf(stderr, Name ": %s: no change requested\n",
- devname);
- goto release;
+ /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
+ if (!container ||
+ (strcmp(name, "raid_disks") != 0 &&
+ strcmp(name, "stripe_cache_size") != 0))
+ return sysfs_set_num(sra, NULL, name, n);
+
+ rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
+ if (rc <= 0)
+ return -1;
+ sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
+ rc = sysfs_set_num(sra, NULL, name, n);
+ if (rc < 0 && errno == EAGAIN) {
+ ping_monitor(container);
+ /* if we get EAGAIN here then the monitor is not active
+ * so stop trying
+ */
+ rc = sysfs_set_num(sra, NULL, name, n);
}
+ sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
+ return rc;
+}
- c = map_num(pers, array.level);
- if (c == NULL) c = "-unknown-";
- switch(array.level) {
- default: /* raid0, linear, multipath cannot be reconfigured */
- fprintf(stderr, Name ": %s array %s cannot be reshaped.\n",
- c, devname);
- rv = 1;
- break;
+int start_reshape(struct mdinfo *sra)
+{
+ 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);
+ err = err ?: 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");
+
+ return err;
+}
- case LEVEL_FAULTY: /* only 'layout' change is permitted */
+void abort_reshape(struct mdinfo *sra)
+{
+ sysfs_set_str(sra, NULL, "sync_action", "idle");
+ sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
+ sysfs_set_num(sra, NULL, "suspend_hi", 0);
+ sysfs_set_num(sra, NULL, "suspend_lo", 0);
+ sysfs_set_num(sra, NULL, "sync_min", 0);
+ sysfs_set_str(sra, NULL, "sync_max", "max");
+}
- if (chunksize || raid_disks) {
- fprintf(stderr, Name ": %s: Cannot change chunksize or disks of a 'faulty' array\n",
- devname);
- rv = 1;
- break;
- }
- if (layout_str == NULL)
- break; /* nothing to do.... */
+int remove_disks_on_raid10_to_raid0_takeover(struct supertype *st,
+ struct mdinfo *sra,
+ int layout)
+{
+ int nr_of_copies;
+ struct mdinfo *remaining;
+ int slot;
- array.layout = parse_layout_faulty(layout_str);
- if (array.layout < 0) {
- int rv;
- fprintf(stderr, Name ": %s: layout %s not understood for 'faulty' array\n",
- devname, layout_str);
- rv = 1;
- break;
- }
- if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
- fprintf(stderr, Name ": Cannot set layout for %s: %s\n",
- devname, strerror(errno));
- rv = 1;
- } else if (!quiet)
- printf("layout for %s set to %d\n", devname, array.layout);
- break;
+ nr_of_copies = layout & 0xff;
- case 1: /* only raid_disks can each be changed. */
+ remaining = sra->devs;
+ sra->devs = NULL;
+ /* for each 'copy', select one device and remove from the list. */
+ for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
+ struct mdinfo **diskp;
+ int found = 0;
- if (chunksize || layout_str != NULL) {
- fprintf(stderr, Name ": %s: Cannot change chunk size or layout for a RAID1 array.\n",
- devname);
- rv = 1;
+ /* Find a working device to keep */
+ for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
+ struct mdinfo *disk = *diskp;
+
+ if (disk->disk.raid_disk < slot)
+ continue;
+ if (disk->disk.raid_disk >= slot + nr_of_copies)
+ continue;
+ if (disk->disk.state & (1<<MD_DISK_REMOVED))
+ continue;
+ if (disk->disk.state & (1<<MD_DISK_FAULTY))
+ continue;
+ if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
+ continue;
+
+ /* We have found a good disk to use! */
+ *diskp = disk->next;
+ disk->next = sra->devs;
+ sra->devs = disk;
+ found = 1;
break;
}
- if (raid_disks > 0) {
- array.raid_disks = raid_disks;
- if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
- fprintf(stderr, Name ": Cannot set raid-devices for %s: %s\n",
- devname, strerror(errno));
- rv = 1;
- }
- }
- break;
+ if (!found)
+ break;
+ }
- case 4:
- case 5:
- case 6:
+ if (slot < sra->array.raid_disks) {
+ /* didn't find all slots */
+ struct mdinfo **e;
+ e = &remaining;
+ while (*e)
+ e = &(*e)->next;
+ *e = sra->devs;
+ sra->devs = remaining;
+ return 1;
+ }
- /*
- * layout/chunksize/raid_disks can be changed
- * though the kernel may not support it all.
- */
- st = super_by_fd(fd);
-
- /*
- * There are three possibilities.
- * 1/ The array will shrink.
- * We need to ensure the reshape will pause before reaching
- * the 'critical section'. We also need to fork and wait for
- * that to happen. When it does we
- * suspend/backup/complete/unfreeze
- *
- * 2/ The array will not change size.
- * This requires that we keep a backup of a sliding window
- * so that we can restore data after a crash. So we need
- * to fork and monitor progress.
- *
- * 3/ The array will grow. This is relatively easy.
- * However the kernel's restripe routines will cheerfully
- * overwrite some early data before it is safe. So we
- * need to make a backup of the early parts of the array
- * and be ready to restore it if rebuild aborts very early.
- *
- * We backup data by writing it to one spare, or to a
- * file which was given on command line.
- *
- * [FOLLOWING IS OLD AND PARTLY WRONG]
- * So: we enumerate the devices in the array and
- * make sure we can open all of them.
- * Then we freeze the early part of the array and
- * backup to the various spares.
- * Then we request changes and start the reshape.
- * Monitor progress until it has passed the danger zone.
- * and finally invalidate the copied data and unfreeze the
- * start of the array.
- *
- * In each case, we first make sure that storage is available
- * for the required backup.
- * Then we:
- * - request the shape change.
- * - for to handle backup etc.
- */
- nchunk = ochunk = array.chunk_size;
- nlayout = olayout = array.layout;
- ndisks = odisks = array.raid_disks;
-
- if (chunksize) {
- nchunk = chunksize * 1024;
- if (size % chunksize) {
- fprintf(stderr, Name ": component size %lluK is not"
- " a multiple of chunksize %dK\n",
- size, chunksize);
- break;
- }
- }
- if (layout_str != NULL)
- switch(array.level) {
- case 4: /* ignore layout */
- break;
- case 5:
- nlayout = map_name(r5layout, layout_str);
- if (nlayout == UnSet) {
- fprintf(stderr, Name ": layout %s not understood for raid5.\n",
- layout_str);
- rv = 1;
- goto release;
- }
- break;
+ /* Remove all 'remaining' devices from the array */
+ while (remaining) {
+ struct mdinfo *sd = remaining;
+ remaining = sd->next;
+
+ sysfs_set_str(sra, sd, "state", "faulty");
+ sysfs_set_str(sra, sd, "slot", "none");
+ 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;
+ sra->devs = sd;
+ }
+ return 0;
+}
+
+void reshape_free_fdlist(int *fdlist,
+ unsigned long long *offsets,
+ int size)
+{
+ int i;
+
+ for (i = 0; i < size; i++)
+ if (fdlist[i] >= 0)
+ close(fdlist[i]);
+
+ free(fdlist);
+ free(offsets);
+}
- case 6:
- nlayout = map_name(r6layout, layout_str);
- if (nlayout == UnSet) {
- fprintf(stderr, Name ": layout %s not understood for raid6.\n",
- layout_str);
- rv = 1;
+int reshape_prepare_fdlist(char *devname,
+ struct mdinfo *sra,
+ int raid_disks,
+ int nrdisks,
+ unsigned long blocks,
+ char *backup_file,
+ int *fdlist,
+ unsigned long long *offsets)
+{
+ int d = 0;
+ struct mdinfo *sd;
+
+ for (d = 0; d <= nrdisks; d++)
+ fdlist[d] = -1;
+ d = raid_disks;
+ for (sd = sra->devs; sd; sd = sd->next) {
+ if (sd->disk.state & (1<<MD_DISK_FAULTY))
+ continue;
+ if (sd->disk.state & (1<<MD_DISK_SYNC)) {
+ char *dn = map_dev(sd->disk.major,
+ sd->disk.minor, 1);
+ fdlist[sd->disk.raid_disk]
+ = dev_open(dn, O_RDONLY);
+ offsets[sd->disk.raid_disk] = sd->data_offset*512;
+ if (fdlist[sd->disk.raid_disk] < 0) {
+ fprintf(stderr,
+ Name ": %s: cannot open component %s\n",
+ devname, dn ? dn : "-unknown-");
+ d = -1;
+ goto release;
+ }
+ } else if (backup_file == NULL) {
+ /* spare */
+ char *dn = map_dev(sd->disk.major,
+ sd->disk.minor, 1);
+ fdlist[d] = dev_open(dn, O_RDWR);
+ offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
+ if (fdlist[d] < 0) {
+ fprintf(stderr, Name ": %s: cannot open component %s\n",
+ devname, dn ? dn : "-unknown-");
+ d = -1;
goto release;
}
- break;
+ d++;
}
- if (raid_disks) ndisks = raid_disks;
+ }
+release:
+ return d;
+}
+
+int reshape_open_backup_file(char *backup_file,
+ int fd,
+ char *devname,
+ long blocks,
+ int *fdlist,
+ unsigned long long *offsets)
+{
+ /* Return 1 on success, 0 on any form of failure */
+ /* need to check backup file is large enough */
+ char buf[512];
+ struct stat stb;
+ unsigned int dev;
+ int i;
+
+ *fdlist = open(backup_file, O_RDWR|O_CREAT|O_EXCL,
+ S_IRUSR | S_IWUSR);
+ *offsets = 8 * 512;
+ if (*fdlist < 0) {
+ fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
+ devname, backup_file, strerror(errno));
+ return 0;
+ }
+ /* Guard against backup file being on array device.
+ * If array is partitioned or if LVM etc is in the
+ * way this will not notice, but it is better than
+ * nothing.
+ */
+ fstat(*fdlist, &stb);
+ dev = stb.st_dev;
+ fstat(fd, &stb);
+ if (stb.st_rdev == dev) {
+ fprintf(stderr, Name ": backup file must NOT be"
+ " on the array being reshaped.\n");
+ close(*fdlist);
+ return 0;
+ }
+
+ memset(buf, 0, 512);
+ for (i=0; i < blocks + 1 ; i++) {
+ if (write(*fdlist, buf, 512) != 512) {
+ fprintf(stderr, Name ": %s: cannot create"
+ " backup file %s: %s\n",
+ devname, backup_file, strerror(errno));
+ return 0;
+ }
+ }
+ if (fsync(*fdlist) != 0) {
+ fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
+ devname, backup_file, strerror(errno));
+ return 0;
+ }
+
+ return 1;
+}
+
+unsigned long compute_backup_blocks(int nchunk, int ochunk,
+ unsigned int ndata, unsigned int odata)
+{
+ unsigned long a, b, blocks;
+ /* So how much do we need to backup.
+ * We need an amount of data which is both a whole number of
+ * old stripes and a whole number of new stripes.
+ * So LCM for (chunksize*datadisks).
+ */
+ a = (ochunk/512) * odata;
+ b = (nchunk/512) * ndata;
+ /* Find GCD */
+ while (a != b) {
+ if (a < b)
+ b -= a;
+ if (b < a)
+ a -= b;
+ }
+ /* LCM == product / GCD */
+ blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
+
+ return blocks;
+}
+
+char *analyse_change(struct mdinfo *info, struct reshape *re)
+{
+ /* Based on the current array state in info->array and
+ * the changes in info->new_* etc, determine:
+ * - whether the change is possible
+ * - Intermediate level/raid_disks/layout
+ * - whether a restriping reshape is needed
+ * - number of sectors in minimum change unit. This
+ * will cover a whole number of stripes in 'before' and
+ * 'after'.
+ *
+ * Return message if the change should be rejected
+ * NULL if the change can be achieved
+ *
+ * This can be called as part of starting a reshape, or
+ * when assembling an array that is undergoing reshape.
+ */
+ int new_disks;
- odata = odisks-1;
- ndata = ndisks-1;
- if (array.level == 6) {
- odata--; /* number of data disks */
- ndata--;
+ /* If a new level not explicitly given, we assume no-change */
+ if (info->new_level == UnSet)
+ info->new_level = info->array.level;
+
+ if (info->new_chunk)
+ switch (info->new_level) {
+ case 0:
+ case 4:
+ case 5:
+ case 6:
+ case 10:
+ /* chunk size is meaningful, must divide component_size
+ * evenly
+ */
+ if (info->component_size % (info->new_chunk/512))
+ return "New chunk size does not"
+ " divide component size";
+ break;
+ default:
+ return "chunk size not meaningful for this level";
}
+ else
+ info->new_chunk = info->array.chunk_size;
+
+ switch (info->array.level) {
+ case 1:
+ /* RAID1 can convert to RAID1 with different disks, or
+ * raid5 with 2 disks
+ */
+ if (info->new_level == 1) {
+ if (info->delta_disks == UnSet)
+ /* Don't know what to do */
+ return "no change requested for Growing RAID1";
+ re->level = 1;
+ re->before.data_disks = (info->array.raid_disks +
+ info->delta_disks);
+ re->before.layout = 0;
+ re->backup_blocks = 0;
+ re->parity = 0;
+ return NULL;
+ }
+ if (info->array.raid_disks == 2 &&
+ info->array.raid_disks == 5) {
+ /* simple in-place conversion */
+ re->level = 5;
+ re->parity = 1;
+ re->before.data_disks = 1;
+ re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
+ re->backup_blocks = 0;
+ return NULL;
+ }
+ /* Could do some multi-stage conversions, but leave that to
+ * later.
+ */
+ return "Impossibly level change request for RAID1";
- if (odata == ndata &&
- get_linux_version() < 2006032) {
- fprintf(stderr, Name ": in-place reshape is not safe before 2.6.32, sorry.\n");
+ case 10:
+ /* RAID10 can only be converted from near mode to
+ * RAID0 by removing some devices
+ */
+ if ((info->array.layout & ~0xff) != 0x100)
+ return "Cannot Grow RAID10 with far/offset layout";
+ /* number of devices must be multiple of number of copies */
+ if (info->array.raid_disks % (info->array.layout & 0xff))
+ return "RAID10 layout too complex for Grow operation";
+
+ if (info->new_level != 0)
+ return "RAID10 can only be changed to RAID0";
+ new_disks = (info->array.raid_disks
+ / (info->array.layout & 0xff));
+ 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 &&
+ info->new_chunk != info->array.chunk_size)
+ return "Cannot change chunk-size with RAID10 Grow";
+
+ /* looks good */
+ re->level = 0;
+ re->parity = 0;
+ re->before.data_disks = new_disks;
+ re->before.layout = 0;
+ re->backup_blocks = 0;
+ return NULL;
+
+ case 0:
+ /* RAID0 can be converted to RAID10, or to RAID456 */
+ if (info->new_level == 10) {
+ if (info->new_layout == UnSet && info->delta_disks == UnSet) {
+ /* Assume near=2 layout */
+ info->new_layout = 0x102;
+ info->delta_disks = info->array.raid_disks;
+ }
+ if (info->new_layout == UnSet) {
+ int copies = 1 + (info->delta_disks
+ / info->array.raid_disks);
+ if (info->array.raid_disks * (copies-1)
+ != info->delta_disks)
+ return "Impossible number of devices"
+ " for RAID0->RAID10";
+ info->new_layout = 0x100 + copies;
+ }
+ if (info->delta_disks == UnSet) {
+ int copies = info->new_layout & 0xff;
+ if (info->new_layout != 0x100 + copies)
+ return "New layout impossible"
+ " for RAID0->RAID10";;
+ info->delta_disks = (copies - 1) *
+ info->array.raid_disks;
+ }
+ if (info->new_chunk &&
+ info->new_chunk != info->array.chunk_size)
+ return "Cannot change chunk-size with RAID0->RAID10";
+ /* looks good */
+ re->level = 10;
+ re->parity = 0;
+ re->before.data_disks = (info->array.raid_disks +
+ info->delta_disks);
+ re->before.layout = info->new_layout;
+ re->backup_blocks = 0;
+ return NULL;
+ }
+
+ /* RAID0 can also covert to RAID0/4/5/6 by first converting to
+ * a raid4 style layout of the final level.
+ */
+ switch (info->new_level) {
+ case 0:
+ case 4:
+ re->level = 4;
+ re->before.layout = 0;
break;
+ case 5:
+ re->level = 5;
+ re->before.layout = ALGORITHM_PARITY_N;
+ break;
+ case 6:
+ re->level = 6;
+ re->before.layout = ALGORITHM_PARITY_N;
+ break;
+ default:
+ return "Impossible level change requested";
}
+ re->before.data_disks = info->array.raid_disks;
+ /* determining 'after' layout happens outside this 'switch' */
+ break;
- /* Check that we can hold all the data */
- get_dev_size(fd, NULL, &array_size);
- if (ndata * size < (array_size/1024)) {
- fprintf(stderr, Name ": this change will reduce the size of the array.\n"
- " use --grow --array-size first to truncate array.\n"
- " e.g. mdadm --grow %s --array-size %llu\n",
- devname, ndata * size);
- rv = 1;
+ case 4:
+ info->array.layout = ALGORITHM_PARITY_N;
+ case 5:
+ switch (info->new_level) {
+ case 4:
+ re->level = info->array.level;
+ re->before.data_disks = info->array.raid_disks - 1;
+ re->before.layout = info->array.layout;
+ break;
+ case 5:
+ re->level = 5;
+ re->before.data_disks = info->array.raid_disks - 1;
+ re->before.layout = info->array.layout;
+ break;
+ case 6:
+ re->level = 6;
+ re->before.data_disks = info->array.raid_disks - 1;
+ switch (info->array.layout) {
+ case ALGORITHM_LEFT_ASYMMETRIC:
+ re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
+ break;
+ case ALGORITHM_RIGHT_ASYMMETRIC:
+ re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
+ break;
+ case ALGORITHM_LEFT_SYMMETRIC:
+ re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
+ break;
+ case ALGORITHM_RIGHT_SYMMETRIC:
+ re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
+ break;
+ case ALGORITHM_PARITY_0:
+ re->before.layout = ALGORITHM_PARITY_0_6;
+ break;
+ case ALGORITHM_PARITY_N:
+ re->before.layout = ALGORITHM_PARITY_N_6;
+ break;
+ default:
+ return "Cannot convert an array with this layout";
+ }
+ break;
+ case 1:
+ if (info->array.raid_disks != 2)
+ return "Can only convert a 2-device array to RAID1";
+ re->level = 1;
+ re->before.data_disks = 2;
+ re->before.layout = 0;
+ break;
+ default:
+ return "Impossible level change requested";
+ }
+ break;
+ case 6:
+ switch (info->new_level) {
+ case 4:
+ case 5:
+ case 6:
+ re->level = 6;
+ re->before.data_disks = info->array.raid_disks - 2;
+ re->before.layout = info->array.layout;
break;
+ default:
+ return "Impossible level change requested";
}
+ break;
+ }
- /* So how much do we need to backup.
- * We need an amount of data which is both a whole number of
- * old stripes and a whole number of new stripes.
- * So LCM for (chunksize*datadisks).
+ /* If we reached here then it looks like a re-stripe is
+ * happening. We have determined the intermediate level
+ * and initial raid_disks/layout and stored these in 're'.
+ *
+ * We need to deduce the final layout that can be atomically
+ * converted to the end state.
+ */
+ switch (info->new_level) {
+ case 0:
+ /* We can only get to RAID0 from RAID4 or RAID5
+ * with appropriate layout and one extra device
*/
- a = ochunk/512 * odata;
- b = nchunk/512 * ndata;
- /* Find GCD */
- while (a != b) {
- if (a < b)
- b -= a;
- if (b < a)
- a -= b;
- }
- /* LCM == product / GCD */
- blocks = ochunk/512 * nchunk/512 * odata * ndata / a;
+ 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;
+ case 5:
+ re->after.layout = ALGORITHM_PARITY_N; break;
+ }
+ break;
- sysfs_free(sra);
- sra = sysfs_read(fd, 0,
- GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|
- GET_CACHE);
-
- if (ndata == odata) {
- /* Make 'blocks' bigger for better throughput, but
- * not so big that we reject it below.
- * Try for 16 megabytes
+ case 4:
+ /* 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;
+ case 5:
+ re->after.layout = ALGORITHM_PARITY_N; break;
+ }
+ break;
+
+ case 5:
+ /* 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)
+ re->after.layout = re->before.layout;
+ else
+ re->after.layout = info->new_layout;
+ break;
+ case 6:
+ if (info->new_layout == UnSet)
+ info->new_layout = re->before.layout;
+
+ /* after.layout needs to be raid6 version of new_layout */
+ if (info->new_layout == ALGORITHM_PARITY_N)
+ re->after.layout = ALGORITHM_PARITY_N;
+ else {
+ char layout[40];
+ char *ls = map_num(r5layout, info->new_layout);
+ int l;
+ strcat(strcpy(layout, ls), "-6");
+ l = map_name(r6layout, layout);
+ if (l == UnSet)
+ return "Cannot find RAID6 layout"
+ " to convert to";
+ re->after.layout = l;
+ }
+ }
+ break;
+
+ case 6:
+ /* 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
+ re->after.layout = info->new_layout;
+ break;
+ default:
+ return "Impossible level change requested";
+ }
+ switch (re->level) {
+ case 6: re->parity = 2; break;
+ case 4:
+ case 5: re->parity = 1; break;
+ default: re->parity = 0; break;
+ }
+ /* So we have a restripe operation, we need to calculate the number
+ * of blocks per reshape operation.
+ */
+ if (info->new_chunk == 0)
+ info->new_chunk = info->array.chunk_size;
+ if (re->after.data_disks == re->before.data_disks &&
+ re->after.layout == re->before.layout &&
+ info->new_chunk == info->array.chunk_size) {
+ /* Nothing to change */
+ re->backup_blocks = 0;
+ return NULL;
+ }
+ if (re->after.data_disks == 1 && re->before.data_disks == 1) {
+ /* chunks can layout changes make no difference */
+ re->backup_blocks = 0;
+ return NULL;
+ }
+
+ if (re->after.data_disks == re->before.data_disks &&
+ get_linux_version() < 2006032)
+ return "in-place reshape is not safe before 2.6.32 - sorry.";
+
+ if (re->after.data_disks < re->before.data_disks &&
+ get_linux_version() < 2006030)
+ return "reshape to fewer devices is not supported before 2.6.32 - sorry.";
+
+ re->backup_blocks = compute_backup_blocks(
+ info->new_chunk, info->array.chunk_size,
+ re->after.data_disks,
+ re->before.data_disks);
+
+ re->new_size = info->component_size * re->after.data_disks;
+ return NULL;
+}
+
+static int reshape_array(char *container, int fd, char *devname,
+ struct supertype *st, struct mdinfo *info,
+ int force, char *backup_file, int quiet, int forked);
+static int reshape_container(char *container, int cfd, char *devname,
+ struct supertype *st,
+ struct mdinfo *info,
+ int force,
+ char *backup_file,
+ int quiet);
+
+int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
+ long long size,
+ int level, char *layout_str, int chunksize, int raid_disks,
+ int force)
+{
+ /* Make some changes in the shape of an array.
+ * The kernel must support the change.
+ *
+ * There are three different changes. Each can trigger
+ * a resync or recovery so we freeze that until we have
+ * requested everything (if kernel supports freezing - 2.6.30).
+ * The steps are:
+ * - change size (i.e. component_size)
+ * - change level
+ * - change layout/chunksize/ndisks
+ *
+ * The last can require a reshape. It is different on different
+ * levels so we need to check the level before actioning it.
+ * Some times the level change needs to be requested after the
+ * reshape (e.g. raid6->raid5, raid5->raid0)
+ *
+ */
+ struct mdu_array_info_s array;
+ int rv = 0;
+ struct supertype *st;
+ char *subarray = NULL;
+
+ int frozen;
+ int changed = 0;
+ char *container = NULL;
+ char container_buf[20];
+ int cfd = -1;
+
+ struct mdinfo info;
+ struct mdinfo *sra;
+
+ if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
+ fprintf(stderr, Name ": %s is not an active md array - aborting\n",
+ devname);
+ return 1;
+ }
+
+ if (size >= 0 &&
+ (chunksize || level!= UnSet || layout_str || raid_disks)) {
+ fprintf(stderr, Name ": cannot change component size at the same time "
+ "as other changes.\n"
+ " Change size first, then check data is intact before "
+ "making other changes.\n");
+ return 1;
+ }
+
+ if (raid_disks && raid_disks < array.raid_disks && array.level > 1 &&
+ get_linux_version() < 2006032 &&
+ !check_env("MDADM_FORCE_FEWER")) {
+ fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n"
+ " Please use a newer kernel\n");
+ return 1;
+ }
+
+ st = super_by_fd(fd, &subarray);
+ if (!st) {
+ fprintf(stderr, Name ": Unable to determine metadata format for %s\n", devname);
+ return 1;
+ }
+ if (raid_disks > st->max_devs) {
+ fprintf(stderr, Name ": Cannot increase raid-disks on this array"
+ " beyond %d\n", st->max_devs);
+ return 1;
+ }
+
+ /* in the external case we need to check that the requested reshape is
+ * supported, and perform an initial check that the container holds the
+ * pre-requisite spare devices (mdmon owns final validation)
+ */
+ if (st->ss->external) {
+ int container_dev;
+ int rv;
+
+ if (subarray) {
+ container_dev = st->container_dev;
+ cfd = open_dev_excl(st->container_dev);
+ } else {
+ container_dev = st->devnum;
+ close(fd);
+ cfd = open_dev_excl(st->devnum);
+ fd = cfd;
+ }
+ if (cfd < 0) {
+ fprintf(stderr, Name ": Unable to open container for %s\n",
+ devname);
+ free(subarray);
+ return 1;
+ }
+
+ fmt_devname(container_buf, container_dev);
+ container = container_buf;
+
+ if (subarray)
+ rv = st->ss->load_container(st, cfd, NULL);
+ else
+ rv = st->ss->load_super(st, cfd, NULL);
+ if (rv) {
+ fprintf(stderr, Name ": Cannot read superblock for %s\n",
+ devname);
+ free(subarray);
+ return 1;
+ }
+
+ if (mdmon_running(container_dev))
+ st->update_tail = &st->updates;
+ }
+
+ if (raid_disks > array.raid_disks &&
+ array.spare_disks < (raid_disks - array.raid_disks) &&
+ !force) {
+ fprintf(stderr,
+ Name ": Need %d spare%s to avoid degraded array,"
+ " and only have %d.\n"
+ " Use --force to over-ride this check.\n",
+ raid_disks - array.raid_disks,
+ raid_disks - array.raid_disks == 1 ? "" : "s",
+ array.spare_disks);
+ return 1;
+ }
+
+ sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS | GET_STATE);
+ if (sra) {
+ if (st->ss->external && subarray == NULL) {
+ array.level = LEVEL_CONTAINER;
+ sra->array.level = LEVEL_CONTAINER;
+ }
+ } else {
+ fprintf(stderr, Name ": failed to read sysfs parameters for %s\n",
+ devname);
+ return 1;
+ }
+ frozen = freeze(st);
+ if (frozen < -1) {
+ /* freeze() already spewed the reason */
+ return 1;
+ } else if (frozen < 0) {
+ fprintf(stderr, Name ": %s is performing resync/recovery and cannot"
+ " be reshaped\n", devname);
+ return 1;
+ }
+
+ /* ========= set 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)) {
+ rv = 1;
+ goto release;
+ }
+ sync_metadata(st);
+ array.size = size;
+ if (array.size != size) {
+ /* got truncated to 32bit, write to
+ * component_size instead
*/
- while (blocks * 32 < sra->component_size &&
- blocks < 16*1024*2)
- blocks *= 2;
+ if (sra)
+ rv = sysfs_set_num(sra, NULL,
+ "component_size", size);
+ else
+ rv = -1;
} else
- fprintf(stderr, Name ": Need to backup %luK of critical "
- "section..\n", blocks/2);
+ rv = ioctl(fd, SET_ARRAY_INFO, &array);
+ if (rv != 0) {
+ int err = errno;
- if (!sra) {
- fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n",
- devname);
+ /* restore metadata */
+ if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
+ NULL, devname, !quiet) == 0)
+ sync_metadata(st);
+ fprintf(stderr, Name ": Cannot set device size for %s: %s\n",
+ devname, strerror(err));
+ if (err == EBUSY &&
+ (array.state & (1<<MD_SB_BITMAP_PRESENT)))
+ fprintf(stderr, " Bitmap must be removed before size can be changed\n");
rv = 1;
- break;
+ goto release;
+ }
+ ioctl(fd, GET_ARRAY_INFO, &array);
+ size = get_component_size(fd)/2;
+ if (size == 0)
+ size = array.size;
+ if (!quiet)
+ fprintf(stderr, Name ": component size of %s has been set to %lluK\n",
+ devname, size);
+ changed = 1;
+ } else if (array.level != LEVEL_CONTAINER) {
+ size = get_component_size(fd)/2;
+ if (size == 0)
+ size = array.size;
+ }
+
+ /* ========= check for Raid10 -> Raid0 conversion ===============
+ * current implementation assumes that following conditions must be met:
+ * - far_copies == 1
+ * - near_copies == 2
+ */
+ if (level == 0 && array.level == 10 && sra &&
+ array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) {
+ int err;
+ err = remove_disks_on_raid10_to_raid0_takeover(st, sra, array.layout);
+ if (err) {
+ dprintf(Name": Array cannot be reshaped\n");
+ if (cfd > -1)
+ close(cfd);
+ rv = 1;
+ goto release;
}
+ }
- if (blocks >= sra->component_size/2) {
- fprintf(stderr, Name ": %s: Something wrong - reshape aborted\n",
+ info.array = array;
+ sysfs_init(&info, fd, NoMdDev);
+ info.component_size = size*2;
+ info.new_level = level;
+ info.new_chunk = chunksize * 1024;
+ if (raid_disks)
+ info.delta_disks = raid_disks - info.array.raid_disks;
+ else
+ info.delta_disks = UnSet;
+ if (layout_str == NULL) {
+ info.new_layout = UnSet;
+ if (info.array.level == 6 &&
+ (info.new_level == 6 || info.new_level == UnSet) &&
+ info.array.layout >= 16) {
+ fprintf(stderr, Name
+ ": %s has a non-standard layout. If you"
+ " wish to preserve this\n"
+ " during the reshape, please specify"
+ " --layout=preserve\n"
+ " If you want to change it, specify a"
+ " layout or use --layout=normalise\n",
devname);
rv = 1;
+ goto release;
+ }
+ } else if (strcmp(layout_str, "normalise") == 0 ||
+ strcmp(layout_str, "normalize") == 0) {
+ /* If we have a -6 RAID6 layout, remove the '-6'. */
+ info.new_layout = UnSet;
+ if (info.array.level == 6 && info.new_level == UnSet) {
+ char l[40], *h;
+ strcpy(l, map_num(r6layout, info.array.layout));
+ h = strrchr(l, '-');
+ if (h && strcmp(h, "-6") == 0) {
+ *h = 0;
+ info.new_layout = map_name(r6layout, l);
+ }
+ }
+ } else if (strcmp(layout_str, "preserve") == 0) {
+ info.new_layout = UnSet;
+ } else {
+ int l = info.new_level;
+ if (l == UnSet)
+ l = info.array.level;
+ switch (l) {
+ case 5:
+ info.new_layout = map_name(r5layout, layout_str);
break;
+ case 6:
+ info.new_layout = map_name(r6layout, layout_str);
+ break;
+ case 10:
+ info.new_layout = parse_layout_10(layout_str);
+ break;
+ case LEVEL_FAULTY:
+ info.new_layout = parse_layout_faulty(layout_str);
+ break;
+ default:
+ fprintf(stderr, Name ": layout not meaningful"
+ " with this level\n");
+ rv = 1;
+ goto release;
+ }
+ if (info.new_layout == UnSet) {
+ fprintf(stderr, Name ": layout %s not understood"
+ " for this level\n",
+ layout_str);
+ rv = 1;
+ goto release;
}
- nrdisks = array.nr_disks + sra->array.spare_disks;
- /* Now we need to open all these devices so we can read/write.
+ }
+
+ if (array.level == LEVEL_CONTAINER) {
+ /* This change is to be applied to every array in the
+ * container. This is only needed when the metadata imposes
+ * restraints of the various arrays in the container.
+ * Currently we only know that IMSM requires all arrays
+ * to have the same number of devices so changing the
+ * number of devices (On-Line Capacity Expansion) must be
+ * performed at the level of the container
*/
- fdlist = malloc((1+nrdisks) * sizeof(int));
- offsets = malloc((1+nrdisks) * sizeof(offsets[0]));
- if (!fdlist || !offsets) {
- fprintf(stderr, Name ": malloc failed: grow aborted\n");
+ rv = reshape_container(container, fd, devname, st, &info,
+ force, backup_file, quiet);
+ } else {
+ /* 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,
+ backup_file, devname, quiet)) {
rv = 1;
- break;
+ goto release;
}
- for (d=0; d <= nrdisks; d++)
- fdlist[d] = -1;
- d = array.raid_disks;
- for (sd = sra->devs; sd; sd=sd->next) {
- if (sd->disk.state & (1<<MD_DISK_FAULTY))
- continue;
- if (sd->disk.state & (1<<MD_DISK_SYNC)) {
- char *dn = map_dev(sd->disk.major,
- sd->disk.minor, 1);
- fdlist[sd->disk.raid_disk]
- = dev_open(dn, O_RDONLY);
- offsets[sd->disk.raid_disk] = sd->data_offset*512;
- if (fdlist[sd->disk.raid_disk] < 0) {
- fprintf(stderr, Name ": %s: cannot open component %s\n",
- devname, dn?dn:"-unknown-");
- rv = 1;
- goto release;
- }
- } else if (backup_file == NULL) {
- /* spare */
- char *dn = map_dev(sd->disk.major,
- sd->disk.minor, 1);
- fdlist[d] = dev_open(dn, O_RDWR);
- offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
- if (fdlist[d]<0) {
- fprintf(stderr, Name ": %s: cannot open component %s\n",
- devname, dn?dn:"-unknown");
- rv = 1;
- goto release;
- }
- d++;
- }
+ sync_metadata(st);
+ rv = reshape_array(container, fd, devname, st, &info, force,
+ backup_file, quiet, 0);
+ }
+ /* reshape_* released the array */
+ return rv;
+release:
+ unfreeze(st, frozen);
+ return rv;
+}
+
+static int reshape_array(char *container, int fd, char *devname,
+ struct supertype *st, struct mdinfo *info,
+ int force,
+ char *backup_file, int quiet, int forked)
+{
+ struct reshape reshape;
+ int spares_needed;
+ char *msg;
+ int orig_level = UnSet;
+ int disks, odisks;
+
+ struct mdu_array_info_s array;
+ char *c;
+ int rv = 0;
+
+ int *fdlist;
+ unsigned long long *offsets;
+ int d;
+ int nrdisks;
+ int err;
+ int frozen;
+ unsigned long blocks;
+ unsigned long cache;
+ unsigned long long array_size;
+ int done;
+ struct mdinfo *sra;
+
+ msg = analyse_change(info, &reshape);
+ if (msg) {
+ fprintf(stderr, Name ": %s\n", msg);
+ return 1;
+ }
+ if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
+ dprintf("Canot get array information.\n");
+ return 1;
+ }
+ spares_needed = max(reshape.before.data_disks,
+ reshape.after.data_disks)
+ + reshape.parity - array.raid_disks;
+
+ if (!force &&
+ info->new_level > 0 &&
+ spares_needed > info->array.spare_disks) {
+ fprintf(stderr,
+ Name ": Need %d spare%s to avoid degraded array,"
+ " and only have %d.\n"
+ " Use --force to over-ride this check.\n",
+ spares_needed,
+ spares_needed == 1 ? "" : "s",
+ info->array.spare_disks);
+ return 1;
+ }
+
+ if (reshape.level != info->array.level) {
+ char *c = map_num(pers, reshape.level);
+ int err;
+ if (c == NULL)
+ return 1; /* This should not be possible */
+
+ err = sysfs_set_str(info, NULL, "level", c);
+ if (err) {
+ err = errno;
+ fprintf(stderr, Name ": %s: could not set level to %s\n",
+ devname, c);
+ if (err == EBUSY &&
+ (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
+ fprintf(stderr, " Bitmap must be removed"
+ " before level can be changed\n");
+ return 1;
}
- if (backup_file == NULL) {
- if (ndata <= odata) {
- fprintf(stderr, Name ": %s: Cannot grow - need backup-file\n",
- devname);
- rv = 1;
- break;
- } else if (sra->array.spare_disks == 0) {
- fprintf(stderr, Name ": %s: Cannot grow - need a spare or "
- "backup-file to backup critical section\n",
- devname);
- rv = 1;
- break;
+ if (!quiet)
+ fprintf(stderr, Name ": level of %s changed to %s\n",
+ devname, c);
+ orig_level = info->array.level;
+ }
+
+ if (reshape.level > 0 && st->ss->external &&
+ !mdmon_running(st->container_dev)) {
+ start_mdmon(st->container_dev);
+ ping_monitor(container);
+ }
+
+ /* ->reshape_super might have chosen some spares from the
+ * container that it wants to be part of the new array.
+ * We can collect them with ->container_content and give
+ * them to the kernel.
+ */
+ if (st->ss->reshape_super && st->ss->container_content) {
+ char *subarray = strchr(info->text_version+1, '/')+1;
+ struct mdinfo *info2 =
+ st->ss->container_content(st, subarray);
+ struct mdinfo *d;
+
+ if (info2)
+ for (d = info2->devs; d; d = d->next) {
+ if (d->disk.state == 0 &&
+ d->disk.raid_disk >= 0) {
+ /* This is a spare that wants to
+ * be part of the array.
+ */
+ add_disk(fd, st, info2, d);
+ }
}
- if (d == array.raid_disks) {
- fprintf(stderr, Name ": %s: No spare device for backup\n",
- devname);
+ sysfs_free(info2);
+ }
+
+ if (reshape.backup_blocks == 0) {
+ /* No restriping needed, but we might need to impose
+ * some more changes: layout, raid_disks, chunk_size
+ */
+ 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) {
+ fprintf(stderr, Name ": failed to set new layout\n");
rv = 1;
- break;
- }
- } else {
- /* need to check backup file is large enough */
- char buf[512];
- fdlist[d] = open(backup_file, O_RDWR|O_CREAT|O_EXCL,
- S_IRUSR | S_IWUSR);
- offsets[d] = 8 * 512;
- if (fdlist[d] < 0) {
- fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
- devname, backup_file, strerror(errno));
+ } else if (!quiet)
+ printf("layout for %s set to %d\n",
+ devname, info->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) {
+ fprintf(stderr, Name ": failed to set raid disks\n");
rv = 1;
- break;
- }
- memset(buf, 0, 512);
- for (i=0; i < blocks + 1 ; i++) {
- if (write(fdlist[d], buf, 512) != 512) {
- fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
- devname, backup_file, strerror(errno));
- rv = 1;
- break;
- }
- }
- if (fsync(fdlist[d]) != 0) {
- fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
- devname, backup_file, strerror(errno));
+ } else if (!quiet)
+ printf("raid_disks for %s set to %d\n",
+ devname, info->array.raid_disks);
+ }
+ if (info->new_chunk != 0 &&
+ info->new_chunk != info->array.chunk_size) {
+ if (sysfs_set_num(info, NULL,
+ "chunk_size", info->new_chunk) != 0) {
+ fprintf(stderr, Name ": failed to set chunk size\n");
rv = 1;
- break;
- }
- d++;
+ } else if (!quiet)
+ printf("chunk size for %s set to %d\n",
+ devname, info->array.chunk_size);
}
- /* lastly, check that the internal stripe cache is
- * large enough, or it won't work.
+ return rv;
+ }
+
+ /*
+ * There are three possibilities.
+ * 1/ The array will shrink.
+ * We need to ensure the reshape will pause before reaching
+ * the 'critical section'. We also need to fork and wait for
+ * that to happen. When it does we
+ * suspend/backup/complete/unfreeze
+ *
+ * 2/ The array will not change size.
+ * This requires that we keep a backup of a sliding window
+ * so that we can restore data after a crash. So we need
+ * to fork and monitor progress.
+ * In future we will allow the data_offset to change, so
+ * a sliding backup becomes unnecessary.
+ *
+ * 3/ The array will grow. This is relatively easy.
+ * However the kernel's restripe routines will cheerfully
+ * overwrite some early data before it is safe. So we
+ * need to make a backup of the early parts of the array
+ * and be ready to restore it if rebuild aborts very early.
+ * For externally managed metadata, we still need a forked
+ * child to monitor the reshape and suspend IO over the region
+ * that is being reshaped.
+ *
+ * We backup data by writing it to one spare, or to a
+ * file which was given on command line.
+ *
+ * In each case, we first make sure that storage is available
+ * for the required backup.
+ * Then we:
+ * - request the shape change.
+ * - fork to handle backup etc.
+ */
+
+ /* Check that we can hold all the data */
+ get_dev_size(fd, NULL, &array_size);
+ if (reshape.new_size < (array_size/512)) {
+ fprintf(stderr,
+ Name ": this change will reduce the size of the array.\n"
+ " use --grow --array-size first to truncate array.\n"
+ " e.g. mdadm --grow %s --array-size %llu\n",
+ devname, reshape.new_size/2);
+ rv = 1;
+ goto release;
+ }
+
+ sra = sysfs_read(fd, 0,
+ GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
+ GET_CACHE);
+
+ if (!sra) {
+ fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n",
+ devname);
+ rv = 1;
+ goto release;
+ }
+
+ /* Decide how many blocks (sectors) for a reshape
+ * unit. The number we have so far is just a minimum
+ */
+ blocks = reshape.backup_blocks;
+ if (reshape.before.data_disks ==
+ reshape.after.data_disks) {
+ /* Make 'blocks' bigger for better throughput, but
+ * not so big that we reject it below.
+ * Try for 16 megabytes
*/
-
- cache = (nchunk < ochunk) ? ochunk : nchunk;
- cache = cache * 4 / 4096;
- if (cache < blocks / 8 / odisks + 16)
- /* Make it big enough to hold 'blocks' */
- cache = blocks / 8 / odisks + 16;
- if (sra->cache_size < cache)
- sysfs_set_num(sra, NULL, "stripe_cache_size",
- cache+1);
- /* Right, everything seems fine. Let's kick things off.
- * If only changing raid_disks, use ioctl, else use
- * sysfs.
+ while (blocks * 32 < sra->component_size &&
+ blocks < 16*1024*2)
+ blocks *= 2;
+ } else
+ fprintf(stderr, Name ": Need to backup %luK of critical "
+ "section..\n", blocks/2);
+
+ if (blocks >= sra->component_size/2) {
+ fprintf(stderr, Name ": %s: Something wrong"
+ " - reshape aborted\n",
+ devname);
+ rv = 1;
+ goto release;
+ }
+
+ /* Now we need to open all these devices so we can read/write.
+ */
+ nrdisks = array.raid_disks + sra->array.spare_disks;
+ fdlist = malloc((1+nrdisks) * sizeof(int));
+ offsets = malloc((1+nrdisks) * sizeof(offsets[0]));
+ if (!fdlist || !offsets) {
+ fprintf(stderr, Name ": malloc failed: grow aborted\n");
+ rv = 1;
+ goto release;
+ }
+
+ d = reshape_prepare_fdlist(devname, sra, array.raid_disks,
+ nrdisks, blocks, backup_file,
+ fdlist, offsets);
+ if (d < 0) {
+ rv = 1;
+ goto release;
+ }
+ if (backup_file == NULL) {
+ if (reshape.after.data_disks <= reshape.before.data_disks) {
+ fprintf(stderr,
+ Name ": %s: Cannot grow - need backup-file\n",
+ devname);
+ rv = 1;
+ goto release;
+ } else if (sra->array.spare_disks == 0) {
+ fprintf(stderr, Name ": %s: Cannot grow - need a spare or "
+ "backup-file to backup critical section\n",
+ devname);
+ rv = 1;
+ goto release;
+ }
+ } else {
+ if (!reshape_open_backup_file(backup_file, fd, devname,
+ (signed)blocks,
+ fdlist+d, offsets+d)) {
+ rv = 1;
+ goto release;
+ }
+ d++;
+ }
+
+ /* lastly, check that the internal stripe cache is
+ * large enough, or it won't work.
+ * It must hold at least 4 stripes of the larger
+ * chunk size
+ */
+ cache = max(info->array.chunk_size, info->new_chunk);
+ cache *= 4; /* 4 stripes minimum */
+ cache /= 512; /* convert to sectors */
+ disks = min(reshape.before.data_disks, reshape.after.data_disks);
+ /* make sure there is room for 'blocks' with a bit to spare */
+ if (cache < 16 + blocks / disks)
+ cache = 16 + blocks / disks;
+ cache /= (4096/512); /* Covert from sectors to pages */
+
+ if (sra->cache_size < cache)
+ subarray_set_num(container, sra, "stripe_cache_size",
+ cache+1);
+
+ /* Right, everything seems fine. Let's kick things off.
+ * If only changing raid_disks, use ioctl, else use
+ * sysfs.
+ */
+ sync_metadata(st);
+
+ sra->new_chunk = info->new_chunk;
+
+ if (info->array.chunk_size == info->new_chunk &&
+ reshape.before.layout == reshape.after.layout &&
+ st->ss->external == 0) {
+ array.raid_disks = reshape.after.data_disks + reshape.parity;
+ if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
+ int err = errno;
+ rv = 1;
+ fprintf(stderr,
+ Name ": Cannot set device shape for %s: %s\n",
+ devname, strerror(errno));
+
+ if (err == EBUSY &&
+ (array.state & (1<<MD_SB_BITMAP_PRESENT)))
+ fprintf(stderr,
+ " Bitmap must be removed before"
+ " shape can be changed\n");
+
+ goto release;
+ }
+ } else {
+ /* set them all just in case some old 'new_*' value
+ * persists from some earlier problem
*/
- if (ochunk == nchunk && olayout == nlayout) {
- array.raid_disks = ndisks;
- if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
- int err = errno;
- rv = 1;
- fprintf(stderr, Name ": Cannot set device shape for %s: %s\n",
- devname, strerror(errno));
- if (ndisks < odisks &&
- get_linux_version() < 2006030)
- fprintf(stderr, Name ": linux 2.6.30 or later required\n");
- if (err == EBUSY &&
- (array.state & (1<<MD_SB_BITMAP_PRESENT)))
- fprintf(stderr, " Bitmap must be removed before shape can be changed\n");
+ int err = err; /* only used if rv==1, and always set if
+ * rv==1, so initialisation not needed,
+ * despite gcc warning
+ */
+ if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
+ rv = 1, err = errno;
+ if (!rv && sysfs_set_num(sra, NULL, "layout",
+ reshape.after.layout) < 0)
+ rv = 1, err = errno;
+ if (!rv && subarray_set_num(container, sra, "raid_disks",
+ reshape.after.data_disks +
+ reshape.parity) < 0)
+ rv = 1, err = errno;
+ if (rv) {
+ fprintf(stderr, Name ": Cannot set device shape for %s\n",
+ devname);
- break;
+ if (err == EBUSY &&
+ (array.state & (1<<MD_SB_BITMAP_PRESENT)))
+ fprintf(stderr,
+ " Bitmap must be removed before"
+ " shape can be changed\n");
+ goto release;
+ }
+ }
+
+ start_reshape(sra);
+
+ /* Now we just need to kick off the reshape and watch, while
+ * handling backups of the data...
+ * This is all done by a forked background process.
+ */
+ switch(forked ? 0 : fork()) {
+ case 0:
+ close(fd);
+ if (check_env("MDADM_GROW_VERIFY"))
+ fd = open(devname, O_RDONLY | O_DIRECT);
+ else
+ 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(
+ fd, sra, &reshape, st, blocks,
+ fdlist, offsets,
+ d - odisks, fdlist+odisks,
+ offsets+odisks);
+ } else
+ done = child_monitor(
+ fd, sra, &reshape, st, blocks,
+ fdlist, offsets,
+ d - odisks, fdlist+odisks,
+ offsets+odisks);
+
+ if (backup_file && done)
+ unlink(backup_file);
+ if (!done) {
+ abort_reshape(sra);
+ goto out;
+ }
+ /* set new array size if required customer_array_size is used
+ * by this metadata.
+ */
+ if (reshape.before.data_disks !=
+ reshape.after.data_disks &&
+ info->custom_array_size) {
+ struct mdinfo *info2;
+ char *subarray = strchr(info->text_version+1, '/')+1;
+
+ wait_reshape(sra);
+ ping_monitor(container);
+
+ info2 = st->ss->container_content(st, subarray);
+ if (info2) {
+ unsigned long long current_size = 0;
+ unsigned long long new_size =
+ info2->custom_array_size/2;
+
+ if (sysfs_get_ll(sra,
+ NULL,
+ "array_size",
+ ¤t_size) == 0 &&
+ new_size > current_size) {
+ if (sysfs_set_num(sra, NULL,
+ "array_size", new_size)
+ < 0)
+ dprintf("Error: Cannot"
+ " set array size");
+ else
+ dprintf("Array size "
+ "changed");
+ dprintf(" from %llu to %llu.\n",
+ current_size, new_size);
+ }
+ sysfs_free(info2);
}
- } else {
- /* set them all just in case some old 'new_*' value
- * persists from some earlier problem
+ }
+
+ if (info->new_level != reshape.level) {
+ /* We need to wait for the reshape to finish
+ * (which will have happened unless
+ * odata < ndata) and then set the level
*/
- int err = err; /* only used if rv==1, and always set if
- * rv==1, so initialisation not needed,
- * despite gcc warning
- */
- if (sysfs_set_num(sra, NULL, "chunk_size", nchunk) < 0)
- rv = 1, err = errno;
- if (!rv && sysfs_set_num(sra, NULL, "layout", nlayout) < 0)
- rv = 1, err = errno;
- if (!rv && sysfs_set_num(sra, NULL, "raid_disks", ndisks) < 0)
- rv = 1, err = errno;
- if (rv) {
- fprintf(stderr, Name ": Cannot set device shape for %s\n",
- devname);
- if (get_linux_version() < 2006030)
- fprintf(stderr, Name ": linux 2.6.30 or later required\n");
- if (err == EBUSY &&
- (array.state & (1<<MD_SB_BITMAP_PRESENT)))
- fprintf(stderr, " Bitmap must be removed before shape can be changed\n");
- break;
+
+ if (reshape.before.data_disks <
+ reshape.after.data_disks)
+ wait_reshape(sra);
+
+ c = map_num(pers, info->new_level);
+ if (c == NULL)
+ goto out;/* not possible */
+
+ err = sysfs_set_str(sra, NULL, "level", c);
+ if (err)
+ fprintf(stderr, Name\
+ ": %s: could not set level "
+ "to %s\n", devname, c);
+ }
+ out:
+ if (forked)
+ return 0;
+ exit(0);
+ case -1:
+ fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n",
+ strerror(errno));
+ rv = 1;
+ abort_reshape(sra);
+ break;
+ default:
+ /* The child will take care of unfreezing the array */
+ frozen = 0;
+ break;
+ }
+
+
+ release:
+ if (!rv) {
+ if (container)
+ ping_monitor(container);
+ if (st->ss->external) {
+ /* Re-load the metadata as much could have changed */
+ int cfd = open_dev(st->container_dev);
+ if (cfd >= 0) {
+ st->ss->free_super(st);
+ st->ss->load_container(st, cfd, container);
+ close(cfd);
}
}
+ }
+ if (rv && orig_level != UnSet && sra) {
+ c = map_num(pers, orig_level);
+ if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
+ fprintf(stderr, Name ": aborting level change\n");
+ }
+ unfreeze(st, frozen);
+ return rv;
+}
+
+int reshape_container(char *container, int cfd, char *devname,
+ struct supertype *st,
+ struct mdinfo *info,
+ int force,
+ char *backup_file,
+ int quiet)
+{
+ struct mdinfo *cc = NULL;
+
+ if (reshape_super(st, info->component_size, info->new_level,
+ info->new_layout, info->new_chunk,
+ info->array.raid_disks + info->delta_disks,
+ backup_file, devname, quiet))
+ return 1;
+
+ sync_metadata(st);
+
+ /* ping monitor to be sure that update is on disk
+ */
+ ping_monitor(container);
- if (ndisks == 2 && odisks == 2) {
- /* No reshape is needed in this trivial case */
- rv = 0;
- break;
- }
+ switch (fork()) {
+ case -1: /* error */
+ perror("Cannot fork to complete reshape\n");
+ return 1;
+ default: /* parent */
+ printf(Name ": multi-array reshape continues in background\n");
+ return 0;
+ case 0: /* child */
+ break;
+ }
- /* set up the backup-super-block. This requires the
- * uuid from the array.
+ while(1) {
+ /* For each member array with reshape_active,
+ * we need to perform the reshape.
+ * We pick the first array that needs reshaping and
+ * reshape it. reshape_array() will re-read the metadata
+ * so the next time through a different array should be
+ * ready for reshape.
*/
- /* Find a superblock */
- for (sd = sra->devs; sd; sd = sd->next) {
- char *dn;
- int devfd;
- int ok;
- if (sd->disk.state & (1<<MD_DISK_FAULTY))
+ struct mdinfo *content;
+ int rv;
+ int fd;
+ struct mdstat_ent *mdstat;
+ char *adev;
+
+ sysfs_free(cc);
+
+ cc = st->ss->container_content(st, NULL);
+
+ for (content = cc; content ; content = content->next) {
+ char *subarray;
+ if (!content->reshape_active)
continue;
- dn = map_dev(sd->disk.major, sd->disk.minor, 1);
- devfd = dev_open(dn, O_RDONLY);
- if (devfd < 0)
+
+ subarray = strchr(content->text_version+1, '/')+1;
+ mdstat = mdstat_by_subdev(subarray,
+ devname2devnum(container));
+ if (!mdstat)
continue;
- ok = st->ss->load_super(st, devfd, NULL);
- close(devfd);
- if (ok >= 0)
- break;
- }
- if (!sd) {
- fprintf(stderr, Name ": %s: Cannot find a superblock\n",
- devname);
- rv = 1;
break;
}
+ if (!content)
+ break;
- memset(&bsb, 0, 512);
- memcpy(bsb.magic, "md_backup_data-1", 16);
- st->ss->uuid_from_super(st, (int*)&bsb.set_uuid);
- bsb.mtime = __cpu_to_le64(time(0));
- bsb.devstart2 = blocks;
- stripes = blocks / (ochunk/512) / odata;
- /* Now we just need to kick off the reshape and watch, while
- * handling backups of the data...
- * This is all done by a forked background process.
- */
- switch(fork()) {
- case 0:
- close(fd);
- if (check_env("MDADM_GROW_VERIFY"))
- fd = open(devname, O_RDONLY | O_DIRECT);
- else
- fd = -1;
- mlockall(MCL_FUTURE);
-
- if (odata < ndata)
- done = child_grow(fd, sra, stripes,
- fdlist, offsets,
- odisks, ochunk, array.level, olayout, odata,
- d - odisks, fdlist+odisks, offsets+odisks);
- else if (odata > ndata)
- done = child_shrink(fd, sra, stripes,
- fdlist, offsets,
- odisks, ochunk, array.level, olayout, odata,
- d - odisks, fdlist+odisks, offsets+odisks);
- else
- done = child_same_size(fd, sra, stripes,
- fdlist, offsets,
- 0,
- odisks, ochunk, array.level, olayout, odata,
- d - odisks, fdlist+odisks, offsets+odisks);
- if (backup_file && done)
- unlink(backup_file);
- if (level != UnSet && level != array.level) {
- /* We need to wait for the reshape to finish
- * (which will have happened unless odata < ndata)
- * and then set the level
- */
-
- c = map_num(pers, level);
- if (c == NULL)
- exit(0);/* not possible */
-
- if (odata < ndata)
- wait_reshape(sra);
- err = sysfs_set_str(sra, NULL, "level", c);
- if (err)
- fprintf(stderr, Name ": %s: could not set level to %s\n",
- devname, c);
- }
- exit(0);
- case -1:
- fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n",
- strerror(errno));
- rv = 1;
+ fd = open_dev_excl(mdstat->devnum);
+ if (fd < 0)
break;
- default:
- /* The child will take care of unfreezing the array */
- frozen = 0;
+ adev = map_dev(dev2major(mdstat->devnum),
+ dev2minor(mdstat->devnum),
+ 0);
+ if (!adev)
+ adev = content->text_version;
+
+ sysfs_init(content, fd, mdstat->devnum);
+
+ rv = reshape_array(container, fd, adev, st,
+ content, force,
+ backup_file, quiet, 1);
+ close(fd);
+ if (rv)
break;
- }
- break;
-
- }
-
- release:
- if (rv && orig_level != UnSet && sra) {
- c = map_num(pers, orig_level);
- if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
- fprintf(stderr, Name ": aborting level change\n");
}
- if (sra)
- unfreeze_array(sra, frozen);
- return rv;
+ sysfs_free(cc);
+ exit(0);
}
/*
*
*/
+int progress_reshape(struct mdinfo *info, struct reshape *reshape,
+ unsigned long long backup_point,
+ unsigned long long wait_point,
+ unsigned long long *suspend_point,
+ unsigned long long *reshape_completed)
+{
+ /* This function is called repeatedly by the reshape manager.
+ * It determines how much progress can safely be made and allows
+ * that progress.
+ * - 'info' identifies the array and particularly records in
+ * ->reshape_progress the metadata's knowledge of progress
+ * This is a sector offset from the start of the array
+ * of the next array block to be relocated. This number
+ * may increase from 0 or decrease from array_size, depending
+ * on the type of reshape that is happening.
+ * Note that in contrast, 'sync_completed' is a block count of the
+ * reshape so far. It gives the distance between the start point
+ * (head or tail of device) and the next place that data will be
+ * written. It always increases.
+ * - 'reshape' is the structure created by analyse_change
+ * - 'backup_point' shows how much the metadata manager has backed-up
+ * data. For reshapes with increasing progress, it is the next address
+ * to be backed up, previous addresses have been backed-up. For
+ * decreasing progress, it is the earliest address that has been
+ * backed up - later address are also backed up.
+ * So addresses between reshape_progress and backup_point are
+ * backed up providing those are in the 'correct' order.
+ * - 'wait_point' is an array address. When reshape_completed
+ * passes this point, progress_reshape should return. It might
+ * return earlier if it determines that ->reshape_progress needs
+ * to be updated or further backup is needed.
+ * - suspend_point is maintained by progress_reshape and the caller
+ * should not touch it except to initialise to zero.
+ * It is an array address and it only increases in 2.6.37 and earlier.
+ * This makes it difficult to handle reducing reshapes with
+ * external metadata.
+ * However: it is similar to backup_point in that it records the
+ * other end of a suspended region from reshape_progress.
+ * it is moved to extend the region that is safe to backup and/or
+ * reshape
+ * - reshape_completed is read from sysfs and returned. The caller
+ * should copy this into ->reshape_progress when it has reason to
+ * believe that the metadata knows this, and any backup outside this
+ * has been erased.
+ *
+ * Return value is:
+ * 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.
+ */
+
+ int advancing = (reshape->after.data_disks
+ >= reshape->before.data_disks);
+ unsigned long long need_backup; /* need to eventually backup all the way
+ * to here
+ */
+ unsigned long long read_offset, write_offset;
+ unsigned long long write_range;
+ unsigned long long max_progress, target, completed;
+ unsigned long long array_size = (info->component_size
+ * reshape->before.data_disks);
+ int fd;
+
+ /* First, we unsuspend any region that is now known to be safe.
+ * If suspend_point is on the 'wrong' side of reshape_progress, then
+ * we don't have or need suspension at the moment. This is true for
+ * native metadata when we don't need to back-up.
+ */
+ if (advancing) {
+ if (info->reshape_progress <= *suspend_point)
+ sysfs_set_num(info, NULL, "suspend_lo",
+ info->reshape_progress);
+ } else {
+ /* Note: this won't work in 2.6.37 and before.
+ * Something somewhere should make sure we don't need it!
+ */
+ if (info->reshape_progress >= *suspend_point)
+ sysfs_set_num(info, NULL, "suspend_hi",
+ info->reshape_progress);
+ }
+
+ /* Now work out how far it is safe to progress.
+ * If the read_offset for ->reshape_progress is less than
+ * 'blocks' beyond the write_offset, we can only progress as far
+ * as a backup.
+ * Otherwise we can progress until the write_offset for the new location
+ * reaches (within 'blocks' of) the read_offset at the current location.
+ * However that region must be suspended unless we are using native
+ * metadata.
+ * If we need to suspend more, we limit it to 128M per device, which is
+ * rather arbitrary and should be some time-based calculation.
+ */
+ read_offset = info->reshape_progress / reshape->before.data_disks;
+ write_offset = info->reshape_progress / reshape->after.data_disks;
+ write_range = info->new_chunk/512;
+ if (advancing) {
+ need_backup = 0;
+ if (read_offset < write_offset + write_range) {
+ max_progress = backup_point;
+ if (reshape->before.data_disks == reshape->after.data_disks)
+ need_backup = array_size;
+ else
+ need_backup = reshape->backup_blocks;
+ } else {
+ max_progress =
+ read_offset *
+ reshape->after.data_disks;
+ }
+ } else {
+ need_backup = array_size;
+ if (read_offset > write_offset - write_range) {
+ max_progress = backup_point;
+ if (max_progress >= info->reshape_progress)
+ need_backup = 0;
+ } else {
+ max_progress =
+ read_offset *
+ reshape->after.data_disks;
+ /* If we are using internal metadata, then we can
+ * progress all the way to the suspend_point without
+ * worrying about backing-up/suspending along the
+ * way.
+ */
+ if (max_progress < *suspend_point &&
+ info->array.major_version >= 0)
+ max_progress = *suspend_point;
+ }
+ }
+
+ /* We know it is safe to progress to 'max_progress' providing
+ * it is suspended or we are using native metadata.
+ * Consider extending suspend_point 128M per device if it
+ * is less than 64M per device beyond reshape_progress.
+ * But always do a multiple of 'blocks'
+ * FIXME this is too big - it takes to long to complete
+ * this much.
+ */
+ target = 64*1024*2 * min(reshape->before.data_disks,
+ reshape->after.data_disks);
+ target /= reshape->backup_blocks;
+ if (target < 2)
+ target = 2;
+ target *= reshape->backup_blocks;
+
+ /* For externally managed metadata we always need to suspend IO to
+ * the area being reshaped so we regularly push suspend_point forward.
+ * For native metadata we only need the suspend if we are going to do
+ * a backup.
+ */
+ if (advancing) {
+ if ((need_backup > info->reshape_progress
+ || info->array.major_version < 0) &&
+ *suspend_point < info->reshape_progress + target) {
+ if (need_backup < *suspend_point + 2 * target)
+ *suspend_point = need_backup;
+ else if (*suspend_point + 2 * target < array_size)
+ *suspend_point += 2 * target;
+ else
+ *suspend_point = array_size;
+ sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
+ if (max_progress > *suspend_point)
+ max_progress = *suspend_point;
+ }
+ } else {
+ if ((need_backup < info->reshape_progress
+ || info->array.major_version < 0) &&
+ *suspend_point > info->reshape_progress - target) {
+ if (need_backup > *suspend_point - 2 * target)
+ *suspend_point = need_backup;
+ else if (*suspend_point >= 2 * target)
+ *suspend_point -= 2 * target;
+ else
+ *suspend_point = 0;
+ sysfs_set_num(info, NULL, "suspend_lo", *suspend_point);
+ if (max_progress < *suspend_point)
+ max_progress = *suspend_point;
+ }
+ }
+
+ /* now set sync_max to allow that progress. sync_max, like
+ * sync_completed is a count of sectors written per device, so
+ * we find the difference between max_progress and the start point,
+ * and divide that by after.data_disks to get a sync_max
+ * number.
+ * At the same time we convert wait_point to a similar number
+ * for comparing against sync_completed.
+ */
+ /* scale down max_progress to per_disk */
+ max_progress /= 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;
+ /* Limit progress to the whole device */
+ if (max_progress > info->component_size)
+ max_progress = info->component_size;
+ wait_point /= reshape->after.data_disks;
+ if (!advancing) {
+ /* switch from 'device offset' to 'processed block count' */
+ max_progress = info->component_size - max_progress;
+ wait_point = info->component_size - wait_point;
+ }
+
+ sysfs_set_num(info, NULL, "sync_max", max_progress);
+
+ /* Now wait. If we have already reached the point that we were
+ * asked to wait to, don't wait at all, else wait for any change.
+ * We need to select on 'sync_completed' as that is the place that
+ * notifications happen, but we are really interested in
+ * 'reshape_position'
+ */
+ fd = sysfs_get_fd(info, NULL, "sync_completed");
+ if (fd < 0)
+ return -1;
+
+ if (sysfs_fd_get_ll(fd, &completed) < 0) {
+ close(fd);
+ return -1;
+ }
+ while (completed < max_progress && completed < wait_point) {
+ /* Check that sync_action is still 'reshape' to avoid
+ * waiting forever on a dead array
+ */
+ char action[20];
+ fd_set rfds;
+ if (sysfs_get_str(info, NULL, "sync_action",
+ action, 20) <= 0 ||
+ strncmp(action, "reshape", 7) != 0)
+ break;
+ FD_ZERO(&rfds);
+ FD_SET(fd, &rfds);
+ select(fd+1, NULL, NULL, &rfds, NULL);
+ if (sysfs_fd_get_ll(fd, &completed) < 0) {
+ close(fd);
+ return -1;
+ }
+ }
+ /* some kernels can give an incorrectly high 'completed' number */
+ completed /= (info->new_chunk/512);
+ completed *= (info->new_chunk/512);
+ /* Convert 'completed' back in to a 'progress' number */
+ completed *= reshape->after.data_disks;
+ if (!advancing) {
+ completed = info->component_size * reshape->after.data_disks
+ - completed;
+ }
+ *reshape_completed = completed;
+
+ close(fd);
+
+ /* We return the need_backup flag. Caller will decide
+ * how much - a multiple of ->backup_blocks up to *suspend_point
+ */
+ return advancing
+ ? (need_backup > info->reshape_progress)
+ : (need_backup < info->reshape_progress);
+}
+
+
/* FIXME return status is never checked */
-int grow_backup(struct mdinfo *sra,
+static int grow_backup(struct mdinfo *sra,
unsigned long long offset, /* per device */
- unsigned long stripes, /* per device */
+ unsigned long stripes, /* per device, in old chunks */
int *sources, unsigned long long *offsets,
int disks, int chunk, int level, int layout,
int dests, int *destfd, unsigned long long *destoffsets,
int odata = disks;
int rv = 0;
int i;
- unsigned long long new_degraded;
+ unsigned long long ll;
+ int new_degraded;
//printf("offset %llu\n", offset);
if (level >= 4)
odata--;
if (level == 6)
odata--;
- sysfs_set_num(sra, NULL, "suspend_hi", (offset + stripes * chunk/512) * odata);
+
/* Check that array hasn't become degraded, else we might backup the wrong data */
- sysfs_get_ll(sra, NULL, "degraded", &new_degraded);
+ sysfs_get_ll(sra, NULL, "degraded", &ll);
+ new_degraded = (int)ll;
if (new_degraded != *degraded) {
/* check each device to ensure it is still working */
struct mdinfo *sd;
}
if (part) {
bsb.arraystart2 = __cpu_to_le64(offset * odata);
- bsb.length2 = __cpu_to_le64(stripes * chunk/512 * odata);
+ bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
} else {
bsb.arraystart = __cpu_to_le64(offset * odata);
- bsb.length = __cpu_to_le64(stripes * chunk/512 * odata);
+ bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
}
if (part)
bsb.magic[15] = '2';
bsb.sb_csum2 = bsb_csum((char*)&bsb,
((char*)&bsb.sb_csum2)-((char*)&bsb));
- if (lseek64(destfd[i], destoffsets[i] - 4096, 0) != destoffsets[i] - 4096)
- rv = 1;
- rv = rv ?: write(destfd[i], &bsb, 512);
+ rv = -1;
+ if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
+ != destoffsets[i] - 4096)
+ break;
+ if (write(destfd[i], &bsb, 512) != 512)
+ break;
if (destoffsets[i] > 4096) {
- if (lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
+ if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
destoffsets[i]+stripes*chunk*odata)
- rv = 1;
- rv = rv ?: write(destfd[i], &bsb, 512);
+ break;
+ if (write(destfd[i], &bsb, 512) != 512)
+ break;
}
fsync(destfd[i]);
+ rv = 0;
}
return rv;
* every works.
*/
/* FIXME return value is often ignored */
-int wait_backup(struct mdinfo *sra,
- unsigned long long offset, /* per device */
- unsigned long long blocks, /* per device */
- unsigned long long blocks2, /* per device - hack */
+static int forget_backup(
int dests, int *destfd, unsigned long long *destoffsets,
int part)
{
- /* Wait for resync to pass the section that was backed up
- * then erase the backup and allow IO
+ /*
+ * Erase backup 'part' (which is 0 or 1)
*/
- int fd = sysfs_get_fd(sra, NULL, "sync_completed");
- unsigned long long completed;
int i;
int rv;
- if (fd < 0)
- return -1;
- sysfs_set_num(sra, NULL, "sync_max", offset + blocks + blocks2);
- if (offset == 0)
- sysfs_set_str(sra, NULL, "sync_action", "reshape");
- do {
- char action[20];
- fd_set rfds;
- FD_ZERO(&rfds);
- FD_SET(fd, &rfds);
- select(fd+1, NULL, NULL, &rfds, NULL);
- if (sysfs_fd_get_ll(fd, &completed) < 0) {
- close(fd);
- return -1;
- }
- if (sysfs_get_str(sra, NULL, "sync_action",
- action, 20) > 0 &&
- strncmp(action, "reshape", 7) != 0)
- break;
- } while (completed < offset + blocks);
- close(fd);
-
if (part) {
bsb.arraystart2 = __cpu_to_le64(0);
bsb.length2 = __cpu_to_le64(0);
if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
bsb.sb_csum2 = bsb_csum((char*)&bsb,
((char*)&bsb.sb_csum2)-((char*)&bsb));
- if (lseek64(destfd[i], destoffsets[i]-4096, 0) !=
+ if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
destoffsets[i]-4096)
- rv = 1;
- rv = rv ?: write(destfd[i], &bsb, 512);
+ rv = -1;
+ if (rv == 0 &&
+ write(destfd[i], &bsb, 512) != 512)
+ rv = -1;
fsync(destfd[i]);
}
return rv;
static void fail(char *msg)
{
int rv;
- rv = write(2, msg, strlen(msg));
- rv |= write(2, "\n", 1);
+ rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
+ rv |= (write(2, "\n", 1) != 1);
exit(rv ? 1 : 2);
}
static char *abuf, *bbuf;
-static int abuflen;
+static unsigned long long abuflen;
static void validate(int afd, int bfd, unsigned long long offset)
{
/* check that the data in the backup against the array.
}
lseek64(bfd, offset, 0);
- if (read(bfd, bbuf, len) != len) {
+ if ((unsigned long long)read(bfd, bbuf, len) != len) {
//printf("len %llu\n", len);
fail("read first backup failed");
}
lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
- if (read(afd, abuf, len) != len)
+ if ((unsigned long long)read(afd, abuf, len) != len)
fail("read first from array failed");
if (memcmp(bbuf, abuf, len) != 0) {
#if 0
}
lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
- if (read(bfd, bbuf, len) != len)
+ if ((unsigned long long)read(bfd, bbuf, len) != len)
fail("read second backup failed");
lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
- if (read(afd, abuf, len) != len)
+ if ((unsigned long long)read(afd, abuf, len) != len)
fail("read second from array failed");
if (memcmp(bbuf, abuf, len) != 0)
fail("data2 compare failed");
}
}
-static int child_grow(int afd, struct mdinfo *sra, unsigned long stripes,
- int *fds, unsigned long long *offsets,
- int disks, int chunk, int level, int layout, int data,
- int dests, int *destfd, unsigned long long *destoffsets)
+int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
+ struct supertype *st, unsigned long blocks,
+ int *fds, unsigned long long *offsets,
+ int dests, int *destfd, unsigned long long *destoffsets)
{
+ /* Monitor a reshape where backup is being performed using
+ * 'native' mechanism - either to a backup file, or
+ * to some space in a spare.
+ */
char *buf;
- int degraded = 0;
+ int degraded = -1;
+ unsigned long long speed;
+ unsigned long long suspend_point, array_size;
+ unsigned long long backup_point, wait_point;
+ unsigned long long reshape_completed;
+ int done = 0;
+ int increasing = reshape->after.data_disks >= reshape->before.data_disks;
+ int part = 0; /* The next part of the backup area to fill. It may already
+ * be full, so we need to check */
+ int level = reshape->level;
+ int layout = reshape->before.layout;
+ int data = reshape->before.data_disks;
+ int disks = reshape->before.data_disks + reshape->parity;
+ int chunk = sra->array.chunk_size;
+ struct mdinfo *sd;
+ unsigned long stripes;
- if (posix_memalign((void**)&buf, 4096, disks * chunk))
- /* Don't start the 'reshape' */
+ /* set up the backup-super-block. This requires the
+ * uuid from the array.
+ */
+ /* Find a superblock */
+ for (sd = sra->devs; sd; sd = sd->next) {
+ char *dn;
+ int devfd;
+ int ok;
+ if (sd->disk.state & (1<<MD_DISK_FAULTY))
+ continue;
+ dn = map_dev(sd->disk.major, sd->disk.minor, 1);
+ devfd = dev_open(dn, O_RDONLY);
+ if (devfd < 0)
+ continue;
+ ok = st->ss->load_super(st, devfd, NULL);
+ close(devfd);
+ if (ok >= 0)
+ break;
+ }
+ if (!sd) {
+ fprintf(stderr, Name ": Cannot find a superblock\n");
return 0;
- sysfs_set_num(sra, NULL, "suspend_hi", 0);
- sysfs_set_num(sra, NULL, "suspend_lo", 0);
- grow_backup(sra, 0, stripes,
- fds, offsets, disks, chunk, level, layout,
- dests, destfd, destoffsets,
- 0, °raded, buf);
- validate(afd, destfd[0], destoffsets[0]);
- wait_backup(sra, 0, stripes * chunk / 512, stripes * chunk / 512,
- dests, destfd, destoffsets,
- 0);
- sysfs_set_num(sra, NULL, "suspend_lo", (stripes * chunk/512) * data);
- free(buf);
- /* FIXME this should probably be numeric */
- sysfs_set_str(sra, NULL, "sync_max", "max");
- return 1;
-}
+ }
-static int child_shrink(int afd, struct mdinfo *sra, unsigned long stripes,
- int *fds, unsigned long long *offsets,
- int disks, int chunk, int level, int layout, int data,
- int dests, int *destfd, unsigned long long *destoffsets)
-{
- char *buf;
- unsigned long long start;
- int rv;
- int degraded = 0;
+ memset(&bsb, 0, 512);
+ memcpy(bsb.magic, "md_backup_data-1", 16);
+ st->ss->uuid_from_super(st, (int*)&bsb.set_uuid);
+ bsb.mtime = __cpu_to_le64(time(0));
+ bsb.devstart2 = blocks;
+
+ stripes = blocks / (sra->array.chunk_size/512) /
+ reshape->before.data_disks;
if (posix_memalign((void**)&buf, 4096, disks * chunk))
+ /* Don't start the 'reshape' */
return 0;
- start = sra->component_size - stripes * chunk/512;
- sysfs_set_num(sra, NULL, "sync_max", start);
- sysfs_set_str(sra, NULL, "sync_action", "reshape");
- sysfs_set_num(sra, NULL, "suspend_lo", 0);
- sysfs_set_num(sra, NULL, "suspend_hi", 0);
- rv = wait_backup(sra, 0, start - stripes * chunk/512, stripes * chunk/512,
- dests, destfd, destoffsets, 0);
- if (rv < 0)
- return 0;
- grow_backup(sra, 0, stripes,
- fds, offsets,
- disks, chunk, level, layout,
- dests, destfd, destoffsets,
- 0, °raded, buf);
- validate(afd, destfd[0], destoffsets[0]);
- wait_backup(sra, start, stripes*chunk/512, 0,
- dests, destfd, destoffsets, 0);
- sysfs_set_num(sra, NULL, "suspend_lo", (stripes * chunk/512) * data);
- free(buf);
- /* FIXME this should probably be numeric */
- sysfs_set_str(sra, NULL, "sync_max", "max");
- return 1;
-}
+ if (reshape->before.data_disks == reshape->after.data_disks) {
+ sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
+ sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
+ }
-static int child_same_size(int afd, struct mdinfo *sra, unsigned long stripes,
- int *fds, unsigned long long *offsets,
- unsigned long long start,
- int disks, int chunk, int level, int layout, int data,
- int dests, int *destfd, unsigned long long *destoffsets)
-{
- unsigned long long size;
- unsigned long tailstripes = stripes;
- int part;
- char *buf;
- unsigned long long speed;
- int degraded = 0;
+ if (increasing) {
+ array_size = sra->component_size * reshape->after.data_disks;
+ backup_point = sra->reshape_progress;
+ suspend_point = 0;
+ } else {
+ array_size = sra->component_size * reshape->before.data_disks;
+ backup_point = array_size;
+ suspend_point = array_size;
+ }
+ while (!done) {
+ int rv;
- if (posix_memalign((void**)&buf, 4096, disks * chunk))
- return 0;
+ /* Want to return as soon the oldest backup slot can
+ * be released as that allows us to start backing up
+ * some more, providing suspend_point has been
+ * advanced, which it should have.
+ */
+ if (increasing) {
+ wait_point = array_size;
+ if (part == 0 && __le64_to_cpu(bsb.length) > 0)
+ wait_point = (__le64_to_cpu(bsb.arraystart) +
+ __le64_to_cpu(bsb.length));
+ if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
+ wait_point = (__le64_to_cpu(bsb.arraystart2) +
+ __le64_to_cpu(bsb.length2));
+ } else {
+ wait_point = 0;
+ if (part == 0 && __le64_to_cpu(bsb.length) > 0)
+ wait_point = __le64_to_cpu(bsb.arraystart);
+ if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
+ wait_point = __le64_to_cpu(bsb.arraystart2);
+ }
+
+ rv = progress_reshape(sra, reshape,
+ backup_point, wait_point,
+ &suspend_point, &reshape_completed);
+ /* external metadata would need to ping_monitor here */
+ sra->reshape_progress = reshape_completed;
+
+ /* Clear any backup region that is before 'here' */
+ if (increasing) {
+ if (reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
+ __le64_to_cpu(bsb.length)))
+ forget_backup(dests, destfd,
+ destoffsets, 0);
+ if (reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
+ __le64_to_cpu(bsb.length2)))
+ forget_backup(dests, destfd,
+ destoffsets, 1);
+ } else {
+ if (reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
+ forget_backup(dests, destfd,
+ destoffsets, 0);
+ if (reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
+ forget_backup(dests, destfd,
+ destoffsets, 1);
+ }
- sysfs_set_num(sra, NULL, "suspend_lo", 0);
- sysfs_set_num(sra, NULL, "suspend_hi", 0);
+ if (rv < 0) {
+ done = 1;
+ break;
+ }
- sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
- sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
-
- grow_backup(sra, start, stripes,
- fds, offsets,
- disks, chunk, level, layout,
- dests, destfd, destoffsets,
- 0, °raded, buf);
- grow_backup(sra, (start + stripes) * chunk/512, stripes,
- fds, offsets,
- disks, chunk, level, layout,
- dests, destfd, destoffsets,
- 1, °raded, buf);
- validate(afd, destfd[0], destoffsets[0]);
- part = 0;
- start += stripes * 2; /* where to read next */
- size = sra->component_size / (chunk/512);
- while (start < size) {
- if (wait_backup(sra, (start-stripes*2)*chunk/512,
- stripes*chunk/512, 0,
- dests, destfd, destoffsets,
- part) < 0)
- return 0;
- sysfs_set_num(sra, NULL, "suspend_lo", start*chunk/512 * data);
- if (start + stripes > size)
- tailstripes = (size - start);
-
- grow_backup(sra, start*chunk/512, tailstripes,
- fds, offsets,
- disks, chunk, level, layout,
- dests, destfd, destoffsets,
- part, °raded, buf);
- start += stripes;
- part = 1 - part;
- validate(afd, destfd[0], destoffsets[0]);
- }
- if (wait_backup(sra, (start-stripes*2) * chunk/512, stripes * chunk/512, 0,
- dests, destfd, destoffsets,
- part) < 0)
- return 0;
- sysfs_set_num(sra, NULL, "suspend_lo", ((start-stripes)*chunk/512) * data);
- wait_backup(sra, (start-stripes) * chunk/512, tailstripes * chunk/512, 0,
- dests, destfd, destoffsets,
- 1-part);
- sysfs_set_num(sra, NULL, "suspend_lo", (size*chunk/512) * data);
- sysfs_set_num(sra, NULL, "sync_speed_min", speed);
+ while (rv) {
+ unsigned long long offset;
+ unsigned long actual_stripes;
+ /* Need to backup some data.
+ * If 'part' is not used and the desired
+ * backup size is suspended, do a backup,
+ * then consider the next part.
+ */
+ /* Check that 'part' is unused */
+ if (part == 0 && __le64_to_cpu(bsb.length) != 0)
+ break;
+ if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
+ break;
+
+ offset = backup_point / data;
+ actual_stripes = stripes;
+ if (increasing) {
+ if (offset + actual_stripes * (chunk/512) >
+ sra->component_size)
+ actual_stripes = ((sra->component_size - offset)
+ / (chunk/512));
+ if (offset + actual_stripes * (chunk/512) >
+ suspend_point/data)
+ break;
+ } else {
+ if (offset < actual_stripes * (chunk/512))
+ actual_stripes = offset / (chunk/512);
+ offset -= actual_stripes * (chunk/512);
+ if (offset < suspend_point/data)
+ break;
+ }
+ grow_backup(sra, offset, actual_stripes,
+ fds, offsets,
+ disks, chunk, level, layout,
+ dests, destfd, destoffsets,
+ part, °raded, buf);
+ validate(afd, destfd[0], destoffsets[0]);
+ /* record where 'part' is up to */
+ part = !part;
+ if (increasing)
+ backup_point += actual_stripes * (chunk/512) * data;
+ else
+ backup_point -= actual_stripes * (chunk/512) * data;
+ }
+ }
+
+ /* FIXME maybe call progress_reshape one more time instead */
+ abort_reshape(sra); /* remove any remaining suspension */
+ if (reshape->before.data_disks == reshape->after.data_disks)
+ sysfs_set_num(sra, NULL, "sync_speed_min", speed);
free(buf);
- return 1;
+ return 1; /* FIXME what does this mean? */
}
/*
if (st->ss->load_super(st, fd, NULL))
continue;
- st->ss->getinfo_super(st, &dinfo);
+ st->ss->getinfo_super(st, &dinfo, NULL);
st->ss->free_super(st);
if (lseek64(fd,
* sometimes they aren't... So allow considerable flexability in matching, and allow
* this test to be overridden by an environment variable.
*/
- if (info->array.utime > __le64_to_cpu(bsb.mtime) + 2*60*60 ||
- info->array.utime < __le64_to_cpu(bsb.mtime) - 10*60) {
+ if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
+ info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
if (check_env("MDADM_GROW_ALLOW_OLD")) {
fprintf(stderr, Name ": accepting backup with timestamp %lu "
"for array with timestamp %lu\n",
}
/* There should be a duplicate backup superblock 4k before here */
if (lseek64(fd, -4096, 1) < 0 ||
- read(fd, &bsb2, 4096) != 4096)
+ read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
goto second_fail; /* Cannot find leading superblock */
if (bsb.magic[15] == '1')
bsbsize = offsetof(struct mdp_backup_super, pad1);
if (st->ss->load_super(st, fdlist[j], NULL))
/* FIXME should be this be an error */
continue;
- st->ss->getinfo_super(st, &dinfo);
+ st->ss->getinfo_super(st, &dinfo, NULL);
st->ss->free_super(st);
offsets[j] = dinfo.data_offset * 512;
}
if (fdlist[j] < 0) continue;
if (st->ss->load_super(st, fdlist[j], NULL))
continue;
- st->ss->getinfo_super(st, &dinfo);
+ st->ss->getinfo_super(st, &dinfo, NULL);
dinfo.reshape_progress = info->reshape_progress;
st->ss->update_super(st, &dinfo,
"_reshape_progress",
int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
char *backup_file)
{
- /* Array is assembled and ready to be started, but
- * monitoring is probably required.
- * So:
- * - start read-only
- * - set upper bound for resync
- * - initialise the 'suspend' boundaries
- * - switch to read-write
- * - fork and continue monitoring
- */
- int err;
- int backup_list[1];
- unsigned long long backup_offsets[1];
- int odisks, ndisks, ochunk, nchunk,odata,ndata;
- unsigned long a,b,blocks,stripes;
- int backup_fd;
- int *fds;
- unsigned long long *offsets;
- int d;
- struct mdinfo *sra, *sd;
- int rv;
- int cache;
- int done = 0;
-
- err = sysfs_set_str(info, NULL, "array_state", "readonly");
+ int err = sysfs_set_str(info, NULL, "array_state", "readonly");
if (err)
return err;
-
- /* make sure reshape doesn't progress until we are ready */
- sysfs_set_str(info, NULL, "sync_max", "0");
- sysfs_set_str(info, NULL, "array_state", "active"); /* FIXME or clean */
-
- /* ndisks is not growing, so raid_disks is old and +delta is new */
- odisks = info->array.raid_disks;
- ndisks = odisks + info->delta_disks;
- odata = odisks - 1;
- ndata = ndisks - 1;
- if (info->array.level == 6) {
- odata--;
- ndata--;
- }
- ochunk = info->array.chunk_size;
- nchunk = info->new_chunk;
-
-
- a = ochunk/512 * odata;
- b = nchunk/512 * ndata;
- /* Find GCD */
- while (a != b) {
- if (a < b)
- b -= a;
- if (b < a)
- a -= b;
- }
- /* LCM == product / GCD */
- blocks = ochunk/512 * nchunk/512 * odata * ndata / a;
-
- sra = sysfs_read(-1, devname2devnum(info->sys_name),
- GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|
- GET_CACHE);
-
-
- if (ndata == odata)
- while (blocks * 32 < sra->component_size &&
- blocks < 16*1024*2)
- blocks *= 2;
- stripes = blocks / (info->array.chunk_size/512) / odata;
-
- /* check that the internal stripe cache is
- * large enough, or it won't work.
- */
- cache = (nchunk < ochunk) ? ochunk : nchunk;
- cache = cache * 4 / 4096;
- if (cache < blocks / 8 / odisks + 16)
- /* Make it big enough to hold 'blocks' */
- cache = blocks / 8 / odisks + 16;
- if (sra->cache_size < cache)
- sysfs_set_num(sra, NULL, "stripe_cache_size",
- cache+1);
-
- memset(&bsb, 0, 512);
- memcpy(bsb.magic, "md_backup_data-1", 16);
- memcpy(&bsb.set_uuid, info->uuid, 16);
- bsb.mtime = __cpu_to_le64(time(0));
- bsb.devstart2 = blocks;
-
- backup_fd = open(backup_file, O_RDWR|O_CREAT, S_IRUSR | S_IWUSR);
- backup_list[0] = backup_fd;
- backup_offsets[0] = 8 * 512;
- fds = malloc(odisks * sizeof(fds[0]));
- offsets = malloc(odisks * sizeof(offsets[0]));
- for (d=0; d<odisks; d++)
- fds[d] = -1;
-
- for (sd = sra->devs; sd; sd = sd->next) {
- if (sd->disk.state & (1<<MD_DISK_FAULTY))
- continue;
- if (sd->disk.state & (1<<MD_DISK_SYNC)) {
- char *dn = map_dev(sd->disk.major,
- sd->disk.minor, 1);
- fds[sd->disk.raid_disk]
- = dev_open(dn, O_RDONLY);
- offsets[sd->disk.raid_disk] = sd->data_offset*512;
- if (fds[sd->disk.raid_disk] < 0) {
- fprintf(stderr, Name ": %s: cannot open component %s\n",
- info->sys_name, dn?dn:"-unknown-");
- rv = 1;
- goto release;
- }
- free(dn);
- }
- }
-
- switch(fork()) {
- case 0:
- close(mdfd);
- mlockall(MCL_FUTURE);
- if (info->delta_disks < 0)
- done = child_shrink(-1, info, stripes,
- fds, offsets,
- info->array.raid_disks,
- info->array.chunk_size,
- info->array.level, info->array.layout,
- odata,
- 1, backup_list, backup_offsets);
- else if (info->delta_disks == 0) {
- /* The 'start' is a per-device stripe number.
- * reshape_progress is a per-array sector number.
- * So divide by ndata * chunk_size
- */
- unsigned long long start = info->reshape_progress / ndata;
- start /= (info->array.chunk_size/512);
- done = child_same_size(-1, info, stripes,
- fds, offsets,
- start,
- info->array.raid_disks,
- info->array.chunk_size,
- info->array.level, info->array.layout,
- odata,
- 1, backup_list, backup_offsets);
- }
- if (backup_file && done)
- unlink(backup_file);
- /* FIXME should I intuit a level change */
- exit(0);
- case -1:
- fprintf(stderr, Name ": Cannot run child to continue monitoring reshape: %s\n",
- strerror(errno));
- return 1;
- default:
- break;
- }
-release:
- return 0;
+ return reshape_array(NULL, mdfd, "array", st, info, 1, backup_file, 0, 0);
}