Merge branch 'master' of git://github.com/djbw/mdadm

12 files changed, 493 insertions, 153 deletions

diff --git a/Create.c b/Create.c
index d104c05..7714828 100644
--- a/Create.c
+++ b/Create.c
@@ -538,7 +538,7 @@ int Create(struct supertype *st, char *mddev,
 	     assume_clean
 		) {
 		info.array.state = 1; /* clean, but one+ drive will be missing*/
-		info.resync_start = ~0ULL;
+		info.resync_start = MaxSector;
 	} else {
 		info.array.state = 0; /* not clean, but no errors */
 		info.resync_start = 0;
diff --git a/Manage.c b/Manage.c
index 1ca371d..749fa7c 100644
--- a/Manage.c
+++ b/Manage.c
@@ -699,6 +699,7 @@ int Manage_subdevs(char *devname, int fd,
 				tst->ss->getinfo_super(tst, &new_mdi);
 				new_mdi.disk.major = disc.major;
 				new_mdi.disk.minor = disc.minor;
+				new_mdi.recovery_start = 0;
 				if (sysfs_add_disk(sra, &new_mdi, 0) != 0) {
 					fprintf(stderr, Name ": add new device to external metadata"
 						" failed for %s\n", dv->devname);
diff --git a/bitmap.c b/bitmap.c
index 850b0ce..088e37d 100644
--- a/bitmap.c
+++ b/bitmap.c
@@ -20,8 +20,6 @@
 
 #include "mdadm.h"
 
-#define min(a,b) (((a) < (b)) ? (a) : (b))
-
 inline void sb_le_to_cpu(bitmap_super_t *sb)
 {
 	sb->magic = __le32_to_cpu(sb->magic);
diff --git a/managemon.c b/managemon.c
index 5958e18..e335077 100644
--- a/managemon.c
+++ b/managemon.c
@@ -112,8 +112,10 @@ static void close_aa(struct active_array *aa)
 {
 	struct mdinfo *d;
 
-	for (d = aa->info.devs; d; d = d->next)
+	for (d = aa->info.devs; d; d = d->next) {
+		close(d->recovery_fd);
 		close(d->state_fd);
+	}
 
 	close(aa->action_fd);
 	close(aa->info.state_fd);
@@ -209,16 +211,22 @@ struct metadata_update *update_queue = NULL;
 struct metadata_update *update_queue_handled = NULL;
 struct metadata_update *update_queue_pending = NULL;
 
-void check_update_queue(struct supertype *container)
+static void free_updates(struct metadata_update **update)
 {
-	while (update_queue_handled) {
-		struct metadata_update *this = update_queue_handled;
-		update_queue_handled = this->next;
+	while (*update) {
+		struct metadata_update *this = *update;
+
+		*update = this->next;
 		free(this->buf);
-		if (this->space)
-			free(this->space);
+		free(this->space);
 		free(this);
 	}
+}
+
+void check_update_queue(struct supertype *container)
+{
+	free_updates(&update_queue_handled);
+
 	if (update_queue == NULL &&
 	    update_queue_pending) {
 		update_queue = update_queue_pending;
@@ -376,8 +384,9 @@ static void manage_member(struct mdstat_ent *mdstat,
 
 	if (a->check_degraded) {
 		struct metadata_update *updates = NULL;
-		struct mdinfo *newdev;
+		struct mdinfo *newdev = NULL;
 		struct active_array *newa;
+		struct mdinfo *d;
 
 		a->check_degraded = 0;
 
@@ -385,34 +394,46 @@ static void manage_member(struct mdstat_ent *mdstat,
 		 * to check.
 		 */
 		newdev = a->container->ss->activate_spare(a, &updates);
-		if (newdev) {
-			struct mdinfo *d;
-			/* Cool, we can add a device or several. */
-			newa = duplicate_aa(a);
-			/* suspend recovery - maybe not needed */
-
-			/* Add device to array and set offset/size/slot.
-			 * and open files for each newdev */
-			for (d = newdev; d ; d = d->next) {
-				struct mdinfo *newd;
-				if (sysfs_add_disk(&newa->info, d, 0) < 0)
-					continue;
-				newd = malloc(sizeof(*newd));
-				*newd = *d;
-				newd->next = newa->info.devs;
-				newa->info.devs = newd;
-
-				newd->state_fd = sysfs_open(a->devnum,
-							    newd->sys_name,
-							    "state");
-				newd->prev_state
-					= read_dev_state(newd->state_fd);
-				newd->curr_state = newd->prev_state;
+		if (!newdev)
+			return;
+
+		newa = duplicate_aa(a);
+		if (!newa)
+			goto out;
+		/* Cool, we can add a device or several. */
+
+		/* Add device to array and set offset/size/slot.
+		 * and open files for each newdev */
+		for (d = newdev; d ; d = d->next) {
+			struct mdinfo *newd;
+
+			newd = malloc(sizeof(*newd));
+			if (!newd)
+				continue;
+			if (sysfs_add_disk(&newa->info, d, 0) < 0) {
+				free(newd);
+				continue;
 			}
-			queue_metadata_update(updates);
-			replace_array(a->container, a, newa);
-			sysfs_set_str(&a->info, NULL, "sync_action", "recover");
+			*newd = *d;
+			newd->next = newa->info.devs;
+			newa->info.devs = newd;
+
+			newd->state_fd = sysfs_open(a->devnum, newd->sys_name,
+						    "state");
+			newd->prev_state = read_dev_state(newd->state_fd);
+			newd->curr_state = newd->prev_state;
+		}
+		queue_metadata_update(updates);
+		updates = NULL;
+		replace_array(a->container, a, newa);
+		sysfs_set_str(&a->info, NULL, "sync_action", "recover");
+ out:
+		while (newdev) {
+			d = newdev->next;
+			free(newdev);
+			newdev = d;
 		}
+		free_updates(&updates);
 	}
 }
 
@@ -498,6 +519,9 @@ static void manage_new(struct mdstat_ent *mdstat,
 			newd->state_fd = sysfs_open(new->devnum,
 						    newd->sys_name,
 						    "state");
+			newd->recovery_fd = sysfs_open(new->devnum,
+						      newd->sys_name,
+						      "recovery_start");
 
 			newd->prev_state = read_dev_state(newd->state_fd);
 			newd->curr_state = newd->prev_state;
@@ -522,7 +546,6 @@ static void manage_new(struct mdstat_ent *mdstat,
 	new->info.state_fd = sysfs_open(new->devnum, NULL, "array_state");
 	new->resync_start_fd = sysfs_open(new->devnum, NULL, "resync_start");
 	new->metadata_fd = sysfs_open(new->devnum, NULL, "metadata_version");
-	get_resync_start(new);
 	dprintf("%s: inst: %d action: %d state: %d\n", __func__, atoi(inst),
 		new->action_fd, new->info.state_fd);
 
diff --git a/mdadm.h b/mdadm.h
index 2bfe840..7ff63cd 100644
--- a/mdadm.h
+++ b/mdadm.h
@@ -129,6 +129,22 @@ extern __off64_t lseek64 __P ((int __fd, __off64_t __offset, int __whence));
 #endif /* __KLIBC__ */
 
 
+/*
+ * min()/max()/clamp() macros that also do
+ * strict type-checking.. See the
+ * "unnecessary" pointer comparison.
+ */
+#define min(x, y) ({                            \
+	typeof(x) _min1 = (x);                  \
+	typeof(y) _min2 = (y);                  \
+	(void) (&_min1 == &_min2);              \
+	_min1 < _min2 ? _min1 : _min2; })
+
+#define max(x, y) ({                            \
+	typeof(x) _max1 = (x);                  \
+	typeof(y) _max2 = (y);                  \
+	(void) (&_max1 == &_max2);              \
+	_max1 > _max2 ? _max1 : _max2; })
 
 /* general information that might be extracted from a superblock */
 struct mdinfo {
@@ -146,7 +162,11 @@ struct mdinfo {
 						    */
 	int			reshape_active;
 	unsigned long long	reshape_progress;
-	unsigned long long	resync_start;
+	union {
+		unsigned long long resync_start; /* per-array resync position */
+		unsigned long long recovery_start; /* per-device rebuild position */
+		#define MaxSector  (~0ULL) /* resync/recovery complete position */
+	};
 	unsigned long		safe_mode_delay; /* ms delay to mark clean */
 	int			new_level, delta_disks, new_layout, new_chunk;
 	int			errors;
@@ -168,6 +188,7 @@ struct mdinfo {
 	struct mdinfo *next;
 
 	/* Device info for mdmon: */
+	int recovery_fd;
 	int state_fd;
 	#define DS_FAULTY	1
 	#define	DS_INSYNC	2
@@ -380,8 +401,7 @@ extern int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 			 char *name, char *val, int size);
 extern int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms);
 extern int sysfs_set_array(struct mdinfo *info, int vers);
-extern int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd,
-			  int in_sync);
+extern int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume);
 extern int sysfs_disk_to_scsi_id(int fd, __u32 *id);
 extern int sysfs_unique_holder(int devnum, long rdev);
 extern int load_sys(char *path, char *buf);
@@ -839,6 +859,7 @@ extern int assemble_container_content(struct supertype *st, int mdfd,
 extern int add_disk(int mdfd, struct supertype *st,
 		    struct mdinfo *sra, struct mdinfo *info);
 extern int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info);
+unsigned long long min_recovery_start(struct mdinfo *array);
 
 extern char *human_size(long long bytes);
 extern char *human_size_brief(long long bytes);
diff --git a/mdmon.h b/mdmon.h
index 7cfee35..4494085 100644
--- a/mdmon.h
+++ b/mdmon.h
@@ -39,8 +39,6 @@ struct active_array {
 	int check_degraded; /* flag set by mon, read by manage */
 
 	int devnum;
-
-	unsigned long long resync_start;
 };
 
 /*
@@ -73,7 +71,6 @@ extern int socket_hup_requested;
 extern int sigterm;
 
 int read_dev_state(int fd);
-int get_resync_start(struct active_array *a);
 int is_container_member(struct mdstat_ent *mdstat, char *container);
 
 struct mdstat_ent *mdstat_read(int hold, int start);
@@ -85,9 +82,9 @@ extern int monitor_loop_cnt;
 /* helper routine to determine resync completion since MaxSector is a
  * moving target
  */
-static inline int is_resync_complete(struct active_array *a)
+static inline int is_resync_complete(struct mdinfo *array)
 {
-	if (a->resync_start >= a->info.component_size)
+	if (array->resync_start >= array->component_size)
 		return 1;
 	return 0;
 }
diff --git a/monitor.c b/monitor.c
index 0cafc3a..81fef49 100644
--- a/monitor.c
+++ b/monitor.c
@@ -66,23 +66,20 @@ static int read_attr(char *buf, int len, int fd)
 	return n;
 }
 
-int get_resync_start(struct active_array *a)
+static unsigned long long read_resync_start(int fd)
 {
 	char buf[30];
 	int n;
 
-	n = read_attr(buf, 30, a->resync_start_fd);
+	n = read_attr(buf, 30, fd);
 	if (n <= 0)
-		return n;
+		return 0;
 	if (strncmp(buf, "none", 4) == 0)
-		a->resync_start = ~0ULL;
+		return MaxSector;
 	else
-		a->resync_start = strtoull(buf, NULL, 10);
-
-	return 1;
+		return strtoull(buf, NULL, 10);
 }
 
-
 static enum array_state read_state(int fd)
 {
 	char buf[20];
@@ -208,22 +205,23 @@ static int read_and_act(struct active_array *a)
 
 	a->curr_state = read_state(a->info.state_fd);
 	a->curr_action = read_action(a->action_fd);
+	a->info.resync_start = read_resync_start(a->resync_start_fd);
 	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
 		mdi->next_state = 0;
-		if (mdi->state_fd >= 0)
+		if (mdi->state_fd >= 0) {
+			mdi->recovery_start = read_resync_start(mdi->recovery_fd);
 			mdi->curr_state = read_dev_state(mdi->state_fd);
+		}
 	}
 
 	if (a->curr_state <= inactive &&
 	    a->prev_state > inactive) {
 		/* array has been stopped */
-		get_resync_start(a);
 		a->container->ss->set_array_state(a, 1);
 		a->next_state = clear;
 		deactivate = 1;
 	}
 	if (a->curr_state == write_pending) {
-		get_resync_start(a);
 		a->container->ss->set_array_state(a, 0);
 		a->next_state = active;
 		dirty = 1;
@@ -236,7 +234,6 @@ static int read_and_act(struct active_array *a)
 		dirty = 1;
 	}
 	if (a->curr_state == clean) {
-		get_resync_start(a);
 		a->container->ss->set_array_state(a, 1);
 	}
 	if (a->curr_state == active ||
@@ -253,7 +250,6 @@ static int read_and_act(struct active_array *a)
 			/* explicit request for readonly array.  Leave it alone */
 			;
 		} else {
-			get_resync_start(a);
 			if (a->container->ss->set_array_state(a, 2))
 				a->next_state = read_auto; /* array is clean */
 			else {
@@ -271,7 +267,6 @@ static int read_and_act(struct active_array *a)
 		 * until the array goes inactive or readonly though.
 		 * Just check if we need to fiddle spares.
 		 */
-		get_resync_start(a);
 		a->container->ss->set_array_state(a, a->curr_state <= clean);
 		check_degraded = 1;
 	}
diff --git a/super-ddf.c b/super-ddf.c
index 1014d81..3e30229 100644
--- a/super-ddf.c
+++ b/super-ddf.c
@@ -1433,7 +1433,7 @@ static void getinfo_super_ddf_bvd(struct supertype *st, struct mdinfo *info)
 	    (ddf->virt->entries[info->container_member].init_state
 	     & DDF_initstate_mask)
 	    == DDF_init_full)
-		info->resync_start = ~0ULL;
+		info->resync_start = MaxSector;
 
 	uuid_from_super_ddf(st, info->uuid);
 
@@ -2921,7 +2921,7 @@ static struct mdinfo *container_content_ddf(struct supertype *st)
 			this->resync_start = 0;
 		} else {
 			this->array.state = 1;
-			this->resync_start = ~0ULL;
+			this->resync_start = MaxSector;
 		}
 		memcpy(this->name, ddf->virt->entries[i].name, 16);
 		this->name[16]=0;
@@ -2968,6 +2968,7 @@ static struct mdinfo *container_content_ddf(struct supertype *st)
 			dev->disk.minor = d->minor;
 			dev->disk.raid_disk = i;
 			dev->disk.state = (1<<MD_DISK_SYNC)|(1<<MD_DISK_ACTIVE);
+			dev->recovery_start = MaxSector;
 
 			dev->events = __be32_to_cpu(ddf->primary.seq);
 			dev->data_offset = __be64_to_cpu(vc->lba_offset[i]);
@@ -3066,7 +3067,7 @@ static int ddf_set_array_state(struct active_array *a, int consistent)
 	if (consistent == 2) {
 		/* Should check if a recovery should be started FIXME */
 		consistent = 1;
-		if (!is_resync_complete(a))
+		if (!is_resync_complete(&a->info))
 			consistent = 0;
 	}
 	if (consistent)
@@ -3078,9 +3079,9 @@ static int ddf_set_array_state(struct active_array *a, int consistent)
 
 	old = ddf->virt->entries[inst].init_state;
 	ddf->virt->entries[inst].init_state &= ~DDF_initstate_mask;
-	if (is_resync_complete(a))
+	if (is_resync_complete(&a->info))
 		ddf->virt->entries[inst].init_state |= DDF_init_full;
-	else if (a->resync_start == 0)
+	else if (a->info.resync_start == 0)
 		ddf->virt->entries[inst].init_state |= DDF_init_not;
 	else
 		ddf->virt->entries[inst].init_state |= DDF_init_quick;
@@ -3088,7 +3089,7 @@ static int ddf_set_array_state(struct active_array *a, int consistent)
 		ddf->updates_pending = 1;
 
 	dprintf("ddf mark %d %s %llu\n", inst, consistent?"clean":"dirty",
-		a->resync_start);
+		a->info.resync_start);
 	return consistent;
 }
 
@@ -3547,6 +3548,7 @@ static struct mdinfo *ddf_activate_spare(struct active_array *a,
 			di->disk.major = dl->major;
 			di->disk.minor = dl->minor;
 			di->disk.state = 0;
+			di->recovery_start = 0;
 			di->data_offset = pos;
 			di->component_size = a->info.component_size;
 			di->container_member = dl->pdnum;
diff --git a/super-intel.c b/super-intel.c
index 2e119f8..609aaf5 100644
--- a/super-intel.c
+++ b/super-intel.c
@@ -635,6 +635,8 @@ static int is_failed(struct imsm_disk *disk)
 }
 
 #ifndef MDASSEMBLE
+static __u64 blocks_per_migr_unit(struct imsm_dev *dev);
+
 static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
 {
 	__u64 sz;
@@ -669,27 +671,32 @@ static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
 	printf("     Chunk Size : %u KiB\n",
 		__le16_to_cpu(map->blocks_per_strip) / 2);
 	printf("       Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
-	printf("  Migrate State : %s", dev->vol.migr_state ? "migrating" : "idle\n");
+	printf("  Migrate State : ");
 	if (dev->vol.migr_state) {
 		if (migr_type(dev) == MIGR_INIT)
-			printf(": initializing\n");
+			printf("initialize\n");
 		else if (migr_type(dev) == MIGR_REBUILD)
-			printf(": rebuilding\n");
+			printf("rebuild\n");
 		else if (migr_type(dev) == MIGR_VERIFY)
-			printf(": check\n");
+			printf("check\n");
 		else if (migr_type(dev) == MIGR_GEN_MIGR)
-			printf(": general migration\n");
+			printf("general migration\n");
 		else if (migr_type(dev) == MIGR_STATE_CHANGE)
-			printf(": state change\n");
+			printf("state change\n");
 		else if (migr_type(dev) == MIGR_REPAIR)
-			printf(": repair\n");
+			printf("repair\n");
 		else
-			printf(": <unknown:%d>\n", migr_type(dev));
-	}
+			printf("<unknown:%d>\n", migr_type(dev));
+	} else
+		printf("idle\n");
 	printf("      Map State : %s", map_state_str[map->map_state]);
 	if (dev->vol.migr_state) {
 		struct imsm_map *map = get_imsm_map(dev, 1);
+
 		printf(" <-- %s", map_state_str[map->map_state]);
+		printf("\n     Checkpoint : %u (%llu)",
+		       __le32_to_cpu(dev->vol.curr_migr_unit),
+		       blocks_per_migr_unit(dev));
 	}
 	printf("\n");
 	printf("    Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
@@ -1215,6 +1222,179 @@ get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
 }
 #endif
 
+static __u32 migr_strip_blocks_resync(struct imsm_dev *dev)
+{
+	/* migr_strip_size when repairing or initializing parity */
+	struct imsm_map *map = get_imsm_map(dev, 0);
+	__u32 chunk = __le32_to_cpu(map->blocks_per_strip);
+
+	switch (get_imsm_raid_level(map)) {
+	case 5:
+	case 10:
+		return chunk;
+	default:
+		return 128*1024 >> 9;
+	}
+}
+
+static __u32 migr_strip_blocks_rebuild(struct imsm_dev *dev)
+{
+	/* migr_strip_size when rebuilding a degraded disk, no idea why
+	 * this is different than migr_strip_size_resync(), but it's good
+	 * to be compatible
+	 */
+	struct imsm_map *map = get_imsm_map(dev, 1);
+	__u32 chunk = __le32_to_cpu(map->blocks_per_strip);
+
+	switch (get_imsm_raid_level(map)) {
+	case 1:
+	case 10:
+		if (map->num_members % map->num_domains == 0)
+			return 128*1024 >> 9;
+		else
+			return chunk;
+	case 5:
+		return max((__u32) 64*1024 >> 9, chunk);
+	default:
+		return 128*1024 >> 9;
+	}
+}
+
+static __u32 num_stripes_per_unit_resync(struct imsm_dev *dev)
+{
+	struct imsm_map *lo = get_imsm_map(dev, 0);
+	struct imsm_map *hi = get_imsm_map(dev, 1);
+	__u32 lo_chunk = __le32_to_cpu(lo->blocks_per_strip);
+	__u32 hi_chunk = __le32_to_cpu(hi->blocks_per_strip);
+
+	return max((__u32) 1, hi_chunk / lo_chunk);
+}
+
+static __u32 num_stripes_per_unit_rebuild(struct imsm_dev *dev)
+{
+	struct imsm_map *lo = get_imsm_map(dev, 0);
+	int level = get_imsm_raid_level(lo);
+
+	if (level == 1 || level == 10) {
+		struct imsm_map *hi = get_imsm_map(dev, 1);
+
+		return hi->num_domains;
+	} else
+		return num_stripes_per_unit_resync(dev);
+}
+
+static __u8 imsm_num_data_members(struct imsm_dev *dev)
+{
+	/* named 'imsm_' because raid0, raid1 and raid10
+	 * counter-intuitively have the same number of data disks
+	 */
+	struct imsm_map *map = get_imsm_map(dev, 0);
+
+	switch (get_imsm_raid_level(map)) {
+	case 0:
+	case 1:
+	case 10:
+		return map->num_members;
+	case 5:
+		return map->num_members - 1;
+	default:
+		dprintf("%s: unsupported raid level\n", __func__);
+		return 0;
+	}
+}
+
+static __u32 parity_segment_depth(struct imsm_dev *dev)
+{
+	struct imsm_map *map = get_imsm_map(dev, 0);
+	__u32 chunk =  __le32_to_cpu(map->blocks_per_strip);
+
+	switch(get_imsm_raid_level(map)) {
+	case 1:
+	case 10:
+		return chunk * map->num_domains;
+	case 5:
+		return chunk * map->num_members;
+	default:
+		return chunk;
+	}
+}
+
+static __u32 map_migr_block(struct imsm_dev *dev, __u32 block)
+{
+	struct imsm_map *map = get_imsm_map(dev, 1);
+	__u32 chunk = __le32_to_cpu(map->blocks_per_strip);
+	__u32 strip = block / chunk;
+
+	switch (get_imsm_raid_level(map)) {
+	case 1:
+	case 10: {
+		__u32 vol_strip = (strip * map->num_domains) + 1;
+		__u32 vol_stripe = vol_strip / map->num_members;
+
+		return vol_stripe * chunk + block % chunk;
+	} case 5: {
+		__u32 stripe = strip / (map->num_members - 1);
+
+		return stripe * chunk + block % chunk;
+	}
+	default:
+		return 0;
+	}
+}
+
+static __u64 blocks_per_migr_unit(struct imsm_dev *dev)
+{
+	/* calculate the conversion factor between per member 'blocks'
+	 * (md/{resync,rebuild}_start) and imsm migration units, return
+	 * 0 for the 'not migrating' and 'unsupported migration' cases
+	 */
+	if (!dev->vol.migr_state)
+		return 0;
+
+	switch (migr_type(dev)) {
+	case MIGR_VERIFY:
+	case MIGR_REPAIR:
+	case MIGR_INIT: {
+		struct imsm_map *map = get_imsm_map(dev, 0);
+		__u32 stripes_per_unit;
+		__u32 blocks_per_unit;
+		__u32 parity_depth;
+		__u32 migr_chunk;
+		__u32 block_map;
+		__u32 block_rel;
+		__u32 segment;
+		__u32 stripe;
+		__u8  disks;
+
+		/* yes, this is really the translation of migr_units to
+		 * per-member blocks in the 'resync' case
+		 */
+		stripes_per_unit = num_stripes_per_unit_resync(dev);
+		migr_chunk = migr_strip_blocks_resync(dev);
+		disks = imsm_num_data_members(dev);
+		blocks_per_unit = stripes_per_unit * migr_chunk * disks;
+		stripe = __le32_to_cpu(map->blocks_per_strip) * disks;
+		segment = blocks_per_unit / stripe;
+		block_rel = blocks_per_unit - segment * stripe;
+		parity_depth = parity_segment_depth(dev);
+		block_map = map_migr_block(dev, block_rel);
+		return block_map + parity_depth * segment;
+	}
+	case MIGR_REBUILD: {
+		__u32 stripes_per_unit;
+		__u32 migr_chunk;
+
+		stripes_per_unit = num_stripes_per_unit_rebuild(dev);
+		migr_chunk = migr_strip_blocks_rebuild(dev);
+		return migr_chunk * stripes_per_unit;
+	}
+	case MIGR_GEN_MIGR:
+	case MIGR_STATE_CHANGE:
+	default:
+		return 0;
+	}
+}
+
 static int imsm_level_to_layout(int level)
 {
 	switch (level) {
@@ -1264,13 +1444,34 @@ static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
 	info->component_size	  = __le32_to_cpu(map->blocks_per_member);
 	memset(info->uuid, 0, sizeof(info->uuid));
 
-	if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
+	if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty) {
 		info->resync_start = 0;
-	else if (dev->vol.migr_state)
-		/* FIXME add curr_migr_unit to resync_start conversion */
-		info->resync_start = 0;
-	else
-		info->resync_start = ~0ULL;
+	} else if (dev->vol.migr_state) {
+		switch (migr_type(dev)) {
+		case MIGR_REPAIR:
+		case MIGR_INIT: {
+			__u64 blocks_per_unit = blocks_per_migr_unit(dev);
+			__u64 units = __le32_to_cpu(dev->vol.curr_migr_unit);
+
+			info->resync_start = blocks_per_unit * units;
+			break;
+		}
+		case MIGR_VERIFY:
+			/* we could emulate the checkpointing of
+			 * 'sync_action=check' migrations, but for now
+			 * we just immediately complete them
+			 */
+		case MIGR_REBUILD:
+			/* this is handled by container_content_imsm() */
+		case MIGR_GEN_MIGR:
+		case MIGR_STATE_CHANGE:
+			/* FIXME handle other migrations */
+		default:
+			/* we are not dirty, so... */
+			info->resync_start = MaxSector;
+		}
+	} else
+		info->resync_start = MaxSector;
 
 	strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
 	info->name[MAX_RAID_SERIAL_LEN] = 0;
@@ -1566,6 +1767,7 @@ static int compare_super_imsm(struct supertype *st, struct supertype *tst)
 		first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
 		first->anchor->orig_family_num = sec->anchor->orig_family_num;
 		first->anchor->family_num = sec->anchor->family_num;
+		memcpy(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH);
 		for (i = 0; i < sec->anchor->num_raid_devs; i++)
 			imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
 	}
@@ -2428,7 +2630,7 @@ imsm_thunderdome(struct intel_super **super_list, int len)
 			struct intel_disk *idisk;
 
 			idisk = disk_list_get(dl->serial, disk_list);
-			if (is_spare(&idisk->disk) &&
+			if (idisk && is_spare(&idisk->disk) &&
 			    !is_failed(&idisk->disk) && !is_configured(&idisk->disk))
 				dl->index = -1;
 			else {
@@ -3333,7 +3535,7 @@ static unsigned long long merge_extents(struct intel_super *super, int sum_exten
 	unsigned long reserve;
 
 	if (!e)
-		return ~0ULL; /* error */
+		return 0;
 
 	/* coalesce and sort all extents. also, check to see if we need to
 	 * reserve space between member arrays
@@ -3376,17 +3578,23 @@ static unsigned long long merge_extents(struct intel_super *super, int sum_exten
 	} while (e[i-1].size);
 	free(e);
 
+	if (maxsize == 0)
+		return 0;
+
+	/* FIXME assumes volume at offset 0 is the first volume in a
+	 * container
+	 */
 	if (start_extent > 0)
 		reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
 	else
 		reserve = 0;
 
 	if (maxsize < reserve)
-		return ~0ULL;
+		return 0;
 
 	super->create_offset = ~((__u32) 0);
 	if (start + reserve > super->create_offset)
-		return ~0ULL; /* start overflows create_offset */
+		return 0; /* start overflows create_offset */
 	super->create_offset = start + reserve;
 
 	return maxsize - reserve;
@@ -3415,6 +3623,34 @@ static int is_raid_level_supported(const struct imsm_orom *orom, int level, int
 }
 
 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
+static int
+validate_geometry_imsm_orom(struct intel_super *super, int level, int layout,
+			    int raiddisks, int chunk, int verbose)
+{
+	if (!is_raid_level_supported(super->orom, level, raiddisks)) {
+		pr_vrb(": platform does not support raid%d with %d disk%s\n",
+			level, raiddisks, raiddisks > 1 ? "s" : "");
+		return 0;
+	}
+	if (super->orom && level != 1 &&
+	    !imsm_orom_has_chunk(super->orom, chunk)) {
+		pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
+		return 0;
+	}
+	if (layout != imsm_level_to_layout(level)) {
+		if (level == 5)
+			pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
+		else if (level == 10)
+			pr_vrb(": imsm raid 10 only supports the n2 layout\n");
+		else
+			pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
+				layout, level);
+		return 0;
+	}
+
+	return 1;
+}
+
 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd 
  * FIX ME add ahci details
  */
@@ -3437,26 +3673,8 @@ static int validate_geometry_imsm_volume(struct supertype *st, int level,
 	if (!super)
 		return 0;
 
-	if (!is_raid_level_supported(super->orom, level, raiddisks)) {
-		pr_vrb(": platform does not support raid%d with %d disk%s\n",
-			level, raiddisks, raiddisks > 1 ? "s" : "");
+	if (!validate_geometry_imsm_orom(super, level, layout, raiddisks, chunk, verbose))
 		return 0;
-	}
-	if (super->orom && level != 1 &&
-	    !imsm_orom_has_chunk(super->orom, chunk)) {
-		pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
-		return 0;
-	}
-	if (layout != imsm_level_to_layout(level)) {
-		if (level == 5)
-			pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
-		else if (level == 10)
-			pr_vrb(": imsm raid 10 only supports the n2 layout\n");
-		else
-			pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
-				layout, level);
-		return 0;
-	}
 
 	if (!dev) {
 		/* General test:  make sure there is space for
@@ -3464,7 +3682,7 @@ static int validate_geometry_imsm_volume(struct supertype *st, int level,
 		 * offset
 		 */
 		unsigned long long minsize = size;
-		unsigned long long start_offset = ~0ULL;
+		unsigned long long start_offset = MaxSector;
 		int dcnt = 0;
 		if (minsize == 0)
 			minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
@@ -3480,7 +3698,7 @@ static int validate_geometry_imsm_volume(struct supertype *st, int level,
 				esize = e[i].start - pos;
 				if (esize >= minsize)
 					found = 1;
-				if (found && start_offset == ~0ULL) {
+				if (found && start_offset == MaxSector) {
 					start_offset = pos;
 					break;
 				} else if (found && pos != start_offset) {
@@ -3569,15 +3787,11 @@ static int validate_geometry_imsm_volume(struct supertype *st, int level,
 			i += dl->extent_cnt;
 
 	maxsize = merge_extents(super, i);
-	if (maxsize < size) {
+	if (maxsize < size || maxsize == 0) {
 		if (verbose)
 			fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
 				maxsize, size);
 		return 0;
-	} else if (maxsize == ~0ULL) {
-		if (verbose)
-			fprintf(stderr, Name ": failed to merge %d extents\n", i);
-		return 0;
 	}
 
 	*freesize = maxsize;
@@ -3634,7 +3848,8 @@ static int reserve_space(struct supertype *st, int raiddisks,
 
 	if (cnt < raiddisks ||
 	    (super->orom && used && used != raiddisks) ||
-	    maxsize < minsize) {
+	    maxsize < minsize ||
+	    maxsize == 0) {
 		fprintf(stderr, Name ": not enough devices with space to create array.\n");
 		return 0; /* No enough free spaces large enough */
 	}
@@ -3686,6 +3901,10 @@ static int validate_geometry_imsm(struct supertype *st, int level, int layout,
 			 * created.  add_to_super and getinfo_super
 			 * detect when autolayout is in progress.
 			 */
+			if (!validate_geometry_imsm_orom(st->sb, level, layout,
+							 raiddisks, chunk,
+							 verbose))
+				return 0;
 			return reserve_space(st, raiddisks, size, chunk, freesize);
 		}
 		return 1;
@@ -3763,6 +3982,46 @@ static int validate_geometry_imsm(struct supertype *st, int level, int layout,
 }
 #endif /* MDASSEMBLE */
 
+static int is_rebuilding(struct imsm_dev *dev)
+{
+	struct imsm_map *migr_map;
+
+	if (!dev->vol.migr_state)
+		return 0;
+
+	if (migr_type(dev) != MIGR_REBUILD)
+		return 0;
+
+	migr_map = get_imsm_map(dev, 1);
+
+	if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
+		return 1;
+	else
+		return 0;
+}
+
+static void update_recovery_start(struct imsm_dev *dev, struct mdinfo *array)
+{
+	struct mdinfo *rebuild = NULL;
+	struct mdinfo *d;
+	__u32 units;
+
+	if (!is_rebuilding(dev))
+		return;
+
+	/* Find the rebuild target, but punt on the dual rebuild case */
+	for (d = array->devs; d; d = d->next)
+		if (d->recovery_start == 0) {
+			if (rebuild)
+				return;
+			rebuild = d;
+		}
+
+	units = __le32_to_cpu(dev->vol.curr_migr_unit);
+	rebuild->recovery_start = units * blocks_per_migr_unit(dev);
+}
+
+
 static struct mdinfo *container_content_imsm(struct supertype *st)
 {
 	/* Given a container loaded by load_super_imsm_all,
@@ -3810,6 +4069,7 @@ static struct mdinfo *container_content_imsm(struct supertype *st)
 		super->current_vol = i;
 		getinfo_super_imsm_volume(st, this);
 		for (slot = 0 ; slot <  map->num_members; slot++) {
+			unsigned long long recovery_start;
 			struct mdinfo *info_d;
 			struct dl *d;
 			int idx;
@@ -3823,33 +4083,41 @@ static struct mdinfo *container_content_imsm(struct supertype *st)
 				if (d->index == idx)
                                         break;
 
+			recovery_start = MaxSector;
 			if (d == NULL)
 				skip = 1;
 			if (d && is_failed(&d->disk))
 				skip = 1;
 			if (ord & IMSM_ORD_REBUILD)
-				skip = 1;
+				recovery_start = 0;
 
 			/* 
 			 * if we skip some disks the array will be assmebled degraded;
-			 * reset resync start to avoid a dirty-degraded situation
+			 * reset resync start to avoid a dirty-degraded
+			 * situation when performing the intial sync
 			 *
 			 * FIXME handle dirty degraded
 			 */
-			if (skip && !dev->vol.dirty)
-				this->resync_start = ~0ULL;
+			if ((skip || recovery_start == 0) && !dev->vol.dirty)
+				this->resync_start = MaxSector;
 			if (skip)
 				continue;
 
-			info_d = malloc(sizeof(*info_d));
+			info_d = calloc(1, sizeof(*info_d));
 			if (!info_d) {
 				fprintf(stderr, Name ": failed to allocate disk"
 					" for volume %.16s\n", dev->volume);
+				info_d = this->devs;
+				while (info_d) {
+					struct mdinfo *d = info_d->next;
+
+					free(info_d);
+					info_d = d;
+				}
 				free(this);
 				this = rest;
 				break;
 			}
-			memset(info_d, 0, sizeof(*info_d));
 			info_d->next = this->devs;
 			this->devs = info_d;
 
@@ -3857,8 +4125,10 @@ static struct mdinfo *container_content_imsm(struct supertype *st)
 			info_d->disk.major = d->major;
 			info_d->disk.minor = d->minor;
 			info_d->disk.raid_disk = slot;
+			info_d->recovery_start = recovery_start;
 
-			this->array.working_disks++;
+			if (info_d->recovery_start == MaxSector)
+				this->array.working_disks++;
 
 			info_d->events = __le32_to_cpu(mpb->generation_num);
 			info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
@@ -3866,6 +4136,8 @@ static struct mdinfo *container_content_imsm(struct supertype *st)
 			if (d->devname)
 				strcpy(info_d->name, d->devname);
 		}
+		/* now that the disk list is up-to-date fixup recovery_start */
+		update_recovery_start(dev, this);
 		rest = this;
 	}
 
@@ -4008,24 +4280,6 @@ static int is_resyncing(struct imsm_dev *dev)
 		return 0;
 }
 
-static int is_rebuilding(struct imsm_dev *dev)
-{
-	struct imsm_map *migr_map;
-
-	if (!dev->vol.migr_state)
-		return 0;
-
-	if (migr_type(dev) != MIGR_REBUILD)
-		return 0;
-
-	migr_map = get_imsm_map(dev, 1);
-
-	if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
-		return 1;
-	else
-		return 0;
-}
-
 /* return true if we recorded new information */
 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
 {
@@ -4076,6 +4330,7 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
 	struct imsm_map *map = get_imsm_map(dev, 0);
 	int failed = imsm_count_failed(super, dev);
 	__u8 map_state = imsm_check_degraded(super, dev, failed);
+	__u32 blocks_per_unit;
 
 	/* before we activate this array handle any missing disks */
 	if (consistent == 2 && super->missing) {
@@ -4087,14 +4342,14 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
 			mark_missing(dev, &dl->disk, dl->index);
 		super->updates_pending++;
 	}
-		
+
 	if (consistent == 2 &&
-	    (!is_resync_complete(a) ||
+	    (!is_resync_complete(&a->info) ||
 	     map_state != IMSM_T_STATE_NORMAL ||
 	     dev->vol.migr_state))
 		consistent = 0;
 
-	if (is_resync_complete(a)) {
+	if (is_resync_complete(&a->info)) {
 		/* complete intialization / resync,
 		 * recovery and interrupted recovery is completed in
 		 * ->set_disk
@@ -4106,7 +4361,7 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
 		}
 	} else if (!is_resyncing(dev) && !failed) {
 		/* mark the start of the init process if nothing is failed */
-		dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
+		dprintf("imsm: mark resync start\n");
 		if (map->map_state == IMSM_T_STATE_UNINITIALIZED)
 			migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
 		else
@@ -4114,12 +4369,32 @@ static int imsm_set_array_state(struct active_array *a, int consistent)
 		super->updates_pending++;
 	}
 
-	 /* FIXME check if we can update curr_migr_unit from resync_start */
+	/* check if we can update curr_migr_unit from resync_start, recovery_start */
+	blocks_per_unit = blocks_per_migr_unit(dev);
+	if (blocks_per_unit && failed <= 1) {
+		__u32 units32;
+		__u64 units;
+
+		if (migr_type(dev) == MIGR_REBUILD)
+			units = min_recovery_start(&a->info) / blocks_per_unit;
+		else
+			units = a->info.resync_start / blocks_per_unit;
+		units32 = units;
+
+		/* check that we did not overflow 32-bits, and that
+		 * curr_migr_unit needs updating
+		 */
+		if (units32 == units &&
+		    __le32_to_cpu(dev->vol.curr_migr_unit) != units32) {
+			dprintf("imsm: mark checkpoint (%u)\n", units32);
+			dev->vol.curr_migr_unit = __cpu_to_le32(units32);
+			super->updates_pending++;
+		}
+	}
 
 	/* mark dirty / clean */
 	if (dev->vol.dirty != !consistent) {
-		dprintf("imsm: mark '%s' (%llu)\n",
-			consistent ? "clean" : "dirty", a->resync_start);
+		dprintf("imsm: mark '%s'\n", consistent ? "clean" : "dirty");
 		if (consistent)
 			dev->vol.dirty = 0;
 		else
@@ -4436,6 +4711,7 @@ static struct mdinfo *imsm_activate_spare(struct active_array *a,
 		di->disk.major = dl->major;
 		di->disk.minor = dl->minor;
 		di->disk.state = 0;
+		di->recovery_start = 0;
 		di->data_offset = __le32_to_cpu(map->pba_of_lba0);
 		di->component_size = a->info.component_size;
 		di->container_member = inst;
diff --git a/super1.c b/super1.c
index 2c992a4..85bb598 100644
--- a/super1.c
+++ b/super1.c
@@ -659,9 +659,9 @@ static int update_super1(struct supertype *st, struct mdinfo *info,
 		switch(__le32_to_cpu(sb->level)) {
 		case 5: case 4: case 6:
 			/* need to force clean */
-			if (sb->resync_offset != ~0ULL)
+			if (sb->resync_offset != MaxSector)
 				rv = 1;
-			sb->resync_offset = ~0ULL;
+			sb->resync_offset = MaxSector;
 		}
 	}
 	if (strcmp(update, "assemble")==0) {
@@ -855,7 +855,7 @@ static int init_super1(struct supertype *st, mdu_array_info_t *info,
 	sb->utime = sb->ctime;
 	sb->events = __cpu_to_le64(1);
 	if (info->state & (1<<MD_SB_CLEAN))
-		sb->resync_offset = ~0ULL;
+		sb->resync_offset = MaxSector;
 	else
 		sb->resync_offset = 0;
 	sb->max_dev = __cpu_to_le32((1024- sizeof(struct mdp_superblock_1))/
diff --git a/sysfs.c b/sysfs.c
index 35dfbd4..c3bbbe3 100644
--- a/sysfs.c
+++ b/sysfs.c
@@ -572,7 +572,7 @@ int sysfs_set_array(struct mdinfo *info, int vers)
 	return rv;
 }
 
-int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int in_sync)
+int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
 {
 	char dv[100];
 	char nm[100];
@@ -595,15 +595,24 @@ int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int in_sync)
 	strcpy(sd->sys_name, "dev-");
 	strcpy(sd->sys_name+4, dname);
 
+	/* test write to see if 'recovery_start' is available */
+	if (resume && sd->recovery_start < MaxSector &&
+	    sysfs_set_num(sra, sd, "recovery_start", 0)) {
+		sysfs_set_str(sra, sd, "state", "remove");
+		return -1;
+	}
+
 	rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 	rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 	if (sra->array.level != LEVEL_CONTAINER) {
-		if (in_sync)
+		if (sd->recovery_start == MaxSector)
 			/* This can correctly fail if array isn't started,
 			 * yet, so just ignore status for now.
 			 */
-			sysfs_set_str(sra, sd, "state", "in_sync");
+			sysfs_set_str(sra, sd, "state", "insync");
 		rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
+		if (resume)
+			sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
 	}
 	return rv;
 }
diff --git a/util.c b/util.c
index 67c3550..5feec43 100644
--- a/util.c
+++ b/util.c
@@ -1338,8 +1338,11 @@ int add_disk(int mdfd, struct supertype *st,
 	int rv;
 #ifndef MDASSEMBLE
 	if (st->ss->external) {
-		rv = sysfs_add_disk(sra, info,
-				    info->disk.state & (1<<MD_DISK_SYNC));
+		if (info->disk.state & (1<<MD_DISK_SYNC))
+			info->recovery_start = MaxSector;
+		else
+			info->recovery_start = 0;
+		rv = sysfs_add_disk(sra, info, 0);
 		if (! rv) {
 			struct mdinfo *sd2;
 			for (sd2 = sra->devs; sd2; sd2=sd2->next)
@@ -1383,10 +1386,25 @@ int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info)
 	return rv;
 }
 
+unsigned long long min_recovery_start(struct mdinfo *array)
+{
+	/* find the minimum recovery_start in an array for metadata
+	 * formats that only record per-array recovery progress instead
+	 * of per-device
+	 */
+	unsigned long long recovery_start = MaxSector;
+	struct mdinfo *d;
+
+	for (d = array->devs; d; d = d->next)
+		recovery_start = min(recovery_start, d->recovery_start);
+
+	return recovery_start;
+}
+
 char *devnum2devname(int num)
 {
 	char name[100];
-	if (num > 0)
+	if (num >= 0)
 		sprintf(name, "md%d", num);
 	else
 		sprintf(name, "md_d%d", -1-num);