/* List of snapshots per Origin */
struct list_head list;
- /* You can't use a snapshot if this is 0 (e.g. if full) */
+ /*
+ * You can't use a snapshot if this is 0 (e.g. if full).
+ * A snapshot-merge target never clears this.
+ */
int valid;
/* Origin writes don't trigger exceptions until this is set */
spinlock_t tracked_chunk_lock;
struct hlist_head tracked_chunk_hash[DM_TRACKED_CHUNK_HASH_SIZE];
+ /*
+ * The merge operation failed if this flag is set.
+ * Failure modes are handled as follows:
+ * - I/O error reading the header
+ * => don't load the target; abort.
+ * - Header does not have "valid" flag set
+ * => use the origin; forget about the snapshot.
+ * - I/O error when reading exceptions
+ * => don't load the target; abort.
+ * (We can't use the intermediate origin state.)
+ * - I/O error while merging
+ * => stop merging; set merge_failed; process I/O normally.
+ */
+ int merge_failed;
+
/* Wait for events based on state_bits */
unsigned long state_bits;
static struct list_head *_origins;
static struct rw_semaphore _origins_lock;
+static DECLARE_WAIT_QUEUE_HEAD(_pending_exceptions_done);
+static DEFINE_SPINLOCK(_pending_exceptions_done_spinlock);
+static uint64_t _pending_exceptions_done_count;
+
static int init_origin_hash(void)
{
int i;
return r;
}
+static int origin_write_extent(struct dm_snapshot *merging_snap,
+ sector_t sector, unsigned chunk_size);
+
static void merge_callback(int read_err, unsigned long write_err,
void *context);
+static uint64_t read_pending_exceptions_done_count(void)
+{
+ uint64_t pending_exceptions_done;
+
+ spin_lock(&_pending_exceptions_done_spinlock);
+ pending_exceptions_done = _pending_exceptions_done_count;
+ spin_unlock(&_pending_exceptions_done_spinlock);
+
+ return pending_exceptions_done;
+}
+
+static void increment_pending_exceptions_done_count(void)
+{
+ spin_lock(&_pending_exceptions_done_spinlock);
+ _pending_exceptions_done_count++;
+ spin_unlock(&_pending_exceptions_done_spinlock);
+
+ wake_up_all(&_pending_exceptions_done);
+}
+
static void snapshot_merge_next_chunks(struct dm_snapshot *s)
{
- int r;
+ int i, linear_chunks;
chunk_t old_chunk, new_chunk;
struct dm_io_region src, dest;
+ sector_t io_size;
+ uint64_t previous_count;
BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
goto shut;
}
- r = s->store->type->prepare_merge(s->store, &old_chunk, &new_chunk);
- if (r <= 0) {
- if (r < 0)
+ linear_chunks = s->store->type->prepare_merge(s->store, &old_chunk,
+ &new_chunk);
+ if (linear_chunks <= 0) {
+ if (linear_chunks < 0) {
DMERR("Read error in exception store: "
"shutting down merge");
+ down_write(&s->lock);
+ s->merge_failed = 1;
+ up_write(&s->lock);
+ }
goto shut;
}
- /* TODO: use larger I/O size once we verify that kcopyd handles it */
+ /* Adjust old_chunk and new_chunk to reflect start of linear region */
+ old_chunk = old_chunk + 1 - linear_chunks;
+ new_chunk = new_chunk + 1 - linear_chunks;
+
+ /*
+ * Use one (potentially large) I/O to copy all 'linear_chunks'
+ * from the exception store to the origin
+ */
+ io_size = linear_chunks * s->store->chunk_size;
dest.bdev = s->origin->bdev;
dest.sector = chunk_to_sector(s->store, old_chunk);
- dest.count = min((sector_t)s->store->chunk_size,
- get_dev_size(dest.bdev) - dest.sector);
+ dest.count = min(io_size, get_dev_size(dest.bdev) - dest.sector);
src.bdev = s->cow->bdev;
src.sector = chunk_to_sector(s->store, new_chunk);
src.count = dest.count;
+ /*
+ * Reallocate any exceptions needed in other snapshots then
+ * wait for the pending exceptions to complete.
+ * Each time any pending exception (globally on the system)
+ * completes we are woken and repeat the process to find out
+ * if we can proceed. While this may not seem a particularly
+ * efficient algorithm, it is not expected to have any
+ * significant impact on performance.
+ */
+ previous_count = read_pending_exceptions_done_count();
+ while (origin_write_extent(s, dest.sector, io_size)) {
+ wait_event(_pending_exceptions_done,
+ (read_pending_exceptions_done_count() !=
+ previous_count));
+ /* Retry after the wait, until all exceptions are done. */
+ previous_count = read_pending_exceptions_done_count();
+ }
+
down_write(&s->lock);
s->first_merging_chunk = old_chunk;
- s->num_merging_chunks = 1;
+ s->num_merging_chunks = linear_chunks;
up_write(&s->lock);
- __check_for_conflicting_io(s, old_chunk);
+ /* Wait until writes to all 'linear_chunks' drain */
+ for (i = 0; i < linear_chunks; i++)
+ __check_for_conflicting_io(s, old_chunk + i);
dm_kcopyd_copy(s->kcopyd_client, &src, 1, &dest, 0, merge_callback, s);
return;
shut:
down_write(&s->lock);
+ s->merge_failed = 1;
b = __release_queued_bios_after_merge(s);
up_write(&s->lock);
error_bios(b);
INIT_LIST_HEAD(&s->list);
spin_lock_init(&s->pe_lock);
s->state_bits = 0;
+ s->merge_failed = 0;
s->first_merging_chunk = 0;
s->num_merging_chunks = 0;
bio_list_init(&s->bios_queued_during_merge);
origin_bios = bio_list_get(&pe->origin_bios);
free_pending_exception(pe);
+ increment_pending_exceptions_done_count();
+
up_write(&s->lock);
/* Submit any pending write bios */
down_write(&s->lock);
- /* Full snapshots are not usable */
- if (!s->valid) {
- r = -EIO;
- goto out_unlock;
- }
+ /* Full merging snapshots are redirected to the origin */
+ if (!s->valid)
+ goto redirect_to_origin;
/* If the block is already remapped - use that */
e = dm_lookup_exception(&s->complete, chunk);
goto out_unlock;
}
+redirect_to_origin:
bio->bi_bdev = s->origin->bdev;
if (bio_rw(bio) == WRITE) {
if (!snap->valid)
DMEMIT("Invalid");
+ else if (snap->merge_failed)
+ DMEMIT("Merge failed");
else {
if (snap->store->type->usage) {
sector_t total_sectors, sectors_allocated,
return r;
}
+/*
+ * Trigger exceptions in all non-merging snapshots.
+ *
+ * The chunk size of the merging snapshot may be larger than the chunk
+ * size of some other snapshot so we may need to reallocate multiple
+ * chunks in other snapshots.
+ *
+ * We scan all the overlapping exceptions in the other snapshots.
+ * Returns 1 if anything was reallocated and must be waited for,
+ * otherwise returns 0.
+ *
+ * size must be a multiple of merging_snap's chunk_size.
+ */
+static int origin_write_extent(struct dm_snapshot *merging_snap,
+ sector_t sector, unsigned size)
+{
+ int must_wait = 0;
+ sector_t n;
+ struct origin *o;
+
+ /*
+ * The origin's __minimum_chunk_size() got stored in split_io
+ * by snapshot_merge_resume().
+ */
+ down_read(&_origins_lock);
+ o = __lookup_origin(merging_snap->origin->bdev);
+ for (n = 0; n < size; n += merging_snap->ti->split_io)
+ if (__origin_write(&o->snapshots, sector + n, NULL) ==
+ DM_MAPIO_SUBMITTED)
+ must_wait = 1;
+ up_read(&_origins_lock);
+
+ return must_wait;
+}
+
/*
* Origin: maps a linear range of a device, with hooks for snapshotting.
*/
projects / karo-tx-linux.git / blobdiff