1 files changed, 156 insertions, 39 deletions

diff --git a/drivers/md/dm-snap-persistent.c b/drivers/md/dm-snap-persistent.c
index 0c746420c008..7d08879689ac 100644
--- a/drivers/md/dm-snap-persistent.c
+++ b/drivers/md/dm-snap-persistent.c
@@ -55,6 +55,8 @@
  */
 #define SNAPSHOT_DISK_VERSION 1
 
+#define NUM_SNAPSHOT_HDR_CHUNKS 1
+
 struct disk_header {
         uint32_t magic;
 
@@ -120,7 +122,22 @@ struct pstore {
 
         /*
          * The next free chunk for an exception.
+         *
+         * When creating exceptions, all the chunks here and above are
+         * free.  It holds the next chunk to be allocated.  On rare
+         * occasions (e.g. after a system crash) holes can be left in
+         * the exception store because chunks can be committed out of
+         * order.
+         *
+         * When merging exceptions, it does not necessarily mean all the
+         * chunks here and above are free.  It holds the value it would
+         * have held if all chunks had been committed in order of
+         * allocation.  Consequently the value may occasionally be
+         * slightly too low, but since it's only used for 'status' and
+         * it can never reach its minimum value too early this doesn't
+         * matter.
          */
+
         chunk_t next_free;
 
         /*
@@ -214,7 +231,7 @@ static int chunk_io(struct pstore *ps, void *area, chunk_t chunk, int rw,
                     int metadata)
 {
         struct dm_io_region where = {
-                .bdev = ps->store->cow->bdev,
+                .bdev = dm_snap_cow(ps->store->snap)->bdev,
                 .sector = ps->store->chunk_size * chunk,
                 .count = ps->store->chunk_size,
         };
@@ -294,7 +311,8 @@ static int read_header(struct pstore *ps, int *new_snapshot)
          */
         if (!ps->store->chunk_size) {
                 ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS,
-                    bdev_logical_block_size(ps->store->cow->bdev) >> 9);
+                    bdev_logical_block_size(dm_snap_cow(ps->store->snap)->
+                                            bdev) >> 9);
                 ps->store->chunk_mask = ps->store->chunk_size - 1;
                 ps->store->chunk_shift = ffs(ps->store->chunk_size) - 1;
                 chunk_size_supplied = 0;
@@ -408,6 +426,15 @@ static void write_exception(struct pstore *ps,
         e->new_chunk = cpu_to_le64(de->new_chunk);
 }
 
+static void clear_exception(struct pstore *ps, uint32_t index)
+{
+        struct disk_exception *e = get_exception(ps, index);
+
+        /* clear it */
+        e->old_chunk = 0;
+        e->new_chunk = 0;
+}
+
 /*
  * Registers the exceptions that are present in the current area.
  * 'full' is filled in to indicate if the area has been
@@ -489,11 +516,23 @@ static struct pstore *get_info(struct dm_exception_store *store)
         return (struct pstore *) store->context;
 }
 
-static void persistent_fraction_full(struct dm_exception_store *store,
-                                     sector_t *numerator, sector_t *denominator)
+static void persistent_usage(struct dm_exception_store *store,
+                             sector_t *total_sectors,
+                             sector_t *sectors_allocated,
+                             sector_t *metadata_sectors)
 {
-        *numerator = get_info(store)->next_free * store->chunk_size;
-        *denominator = get_dev_size(store->cow->bdev);
+        struct pstore *ps = get_info(store);
+
+        *sectors_allocated = ps->next_free * store->chunk_size;
+        *total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev);
+
+        /*
+         * First chunk is the fixed header.
+         * Then there are (ps->current_area + 1) metadata chunks, each one
+         * separated from the next by ps->exceptions_per_area data chunks.
+         */
+        *metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) *
+                            store->chunk_size;
 }
 
 static void persistent_dtr(struct dm_exception_store *store)
@@ -552,44 +591,40 @@ static int persistent_read_metadata(struct dm_exception_store *store,
                 ps->current_area = 0;
                 zero_memory_area(ps);
                 r = zero_disk_area(ps, 0);
-                if (r) {
+                if (r)
                         DMWARN("zero_disk_area(0) failed");
-                        return r;
-                }
-        } else {
-                /*
-                 * Sanity checks.
-                 */
-                if (ps->version != SNAPSHOT_DISK_VERSION) {
-                        DMWARN("unable to handle snapshot disk version %d",
-                               ps->version);
-                        return -EINVAL;
-                }
+                return r;
+        }
+        /*
+         * Sanity checks.
+         */
+        if (ps->version != SNAPSHOT_DISK_VERSION) {
+                DMWARN("unable to handle snapshot disk version %d",
+                       ps->version);
+                return -EINVAL;
+        }
 
-                /*
-                 * Metadata are valid, but snapshot is invalidated
-                 */
-                if (!ps->valid)
-                        return 1;
+        /*
+         * Metadata are valid, but snapshot is invalidated
+         */
+        if (!ps->valid)
+                return 1;
 
-                /*
-                 * Read the metadata.
-                 */
-                r = read_exceptions(ps, callback, callback_context);
-                if (r)
-                        return r;
-        }
+        /*
+         * Read the metadata.
+         */
+        r = read_exceptions(ps, callback, callback_context);
 
-        return 0;
+        return r;
 }
 
 static int persistent_prepare_exception(struct dm_exception_store *store,
-                                        struct dm_snap_exception *e)
+                                        struct dm_exception *e)
 {
         struct pstore *ps = get_info(store);
         uint32_t stride;
         chunk_t next_free;
-        sector_t size = get_dev_size(store->cow->bdev);
+        sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev);
 
         /* Is there enough room ? */
         if (size < ((ps->next_free + 1) * store->chunk_size))
@@ -611,7 +646,7 @@ static int persistent_prepare_exception(struct dm_exception_store *store,
 }
 
 static void persistent_commit_exception(struct dm_exception_store *store,
-                                        struct dm_snap_exception *e,
+                                        struct dm_exception *e,
                                         void (*callback) (void *, int success),
                                         void *callback_context)
 {
@@ -672,6 +707,85 @@ static void persistent_commit_exception(struct dm_exception_store *store,
         ps->callback_count = 0;
 }
 
+static int persistent_prepare_merge(struct dm_exception_store *store,
+                                    chunk_t *last_old_chunk,
+                                    chunk_t *last_new_chunk)
+{
+        struct pstore *ps = get_info(store);
+        struct disk_exception de;
+        int nr_consecutive;
+        int r;
+
+        /*
+         * When current area is empty, move back to preceding area.
+         */
+        if (!ps->current_committed) {
+                /*
+                 * Have we finished?
+                 */
+                if (!ps->current_area)
+                        return 0;
+
+                ps->current_area--;
+                r = area_io(ps, READ);
+                if (r < 0)
+                        return r;
+                ps->current_committed = ps->exceptions_per_area;
+        }
+
+        read_exception(ps, ps->current_committed - 1, &de);
+        *last_old_chunk = de.old_chunk;
+        *last_new_chunk = de.new_chunk;
+
+        /*
+         * Find number of consecutive chunks within the current area,
+         * working backwards.
+         */
+        for (nr_consecutive = 1; nr_consecutive < ps->current_committed;
+             nr_consecutive++) {
+                read_exception(ps, ps->current_committed - 1 - nr_consecutive,
+                               &de);
+                if (de.old_chunk != *last_old_chunk - nr_consecutive ||
+                    de.new_chunk != *last_new_chunk - nr_consecutive)
+                        break;
+        }
+
+        return nr_consecutive;
+}
+
+static int persistent_commit_merge(struct dm_exception_store *store,
+                                   int nr_merged)
+{
+        int r, i;
+        struct pstore *ps = get_info(store);
+
+        BUG_ON(nr_merged > ps->current_committed);
+
+        for (i = 0; i < nr_merged; i++)
+                clear_exception(ps, ps->current_committed - 1 - i);
+
+        r = area_io(ps, WRITE);
+        if (r < 0)
+                return r;
+
+        ps->current_committed -= nr_merged;
+
+        /*
+         * At this stage, only persistent_usage() uses ps->next_free, so
+         * we make no attempt to keep ps->next_free strictly accurate
+         * as exceptions may have been committed out-of-order originally.
+         * Once a snapshot has become merging, we set it to the value it
+         * would have held had all the exceptions been committed in order.
+         *
+         * ps->current_area does not get reduced by prepare_merge() until
+         * after commit_merge() has removed the nr_merged previous exceptions.
+         */
+        ps->next_free = (area_location(ps, ps->current_area) - 1) +
+                        (ps->current_committed + 1) + NUM_SNAPSHOT_HDR_CHUNKS;
+
+        return 0;
+}
+
 static void persistent_drop_snapshot(struct dm_exception_store *store)
 {
         struct pstore *ps = get_info(store);
@@ -697,7 +811,7 @@ static int persistent_ctr(struct dm_exception_store *store,
         ps->area = NULL;
         ps->zero_area = NULL;
         ps->header_area = NULL;
-        ps->next_free = 2;      /* skipping the header and first area */
+        ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */
         ps->current_committed = 0;
 
         ps->callback_count = 0;
@@ -726,8 +840,7 @@ static unsigned persistent_status(struct dm_exception_store *store,
         case STATUSTYPE_INFO:
                 break;
         case STATUSTYPE_TABLE:
-                DMEMIT(" %s P %llu", store->cow->name,
-                       (unsigned long long)store->chunk_size);
+                DMEMIT(" P %llu", (unsigned long long)store->chunk_size);
         }
 
         return sz;
@@ -741,8 +854,10 @@ static struct dm_exception_store_type _persistent_type = {
         .read_metadata = persistent_read_metadata,
         .prepare_exception = persistent_prepare_exception,
         .commit_exception = persistent_commit_exception,
+        .prepare_merge = persistent_prepare_merge,
+        .commit_merge = persistent_commit_merge,
         .drop_snapshot = persistent_drop_snapshot,
-        .fraction_full = persistent_fraction_full,
+        .usage = persistent_usage,
         .status = persistent_status,
 };
 
@@ -754,8 +869,10 @@ static struct dm_exception_store_type _persistent_compat_type = {
         .read_metadata = persistent_read_metadata,
         .prepare_exception = persistent_prepare_exception,
         .commit_exception = persistent_commit_exception,
+        .prepare_merge = persistent_prepare_merge,
+        .commit_merge = persistent_commit_merge,
         .drop_snapshot = persistent_drop_snapshot,
-        .fraction_full = persistent_fraction_full,
+        .usage = persistent_usage,
         .status = persistent_status,
 };