summaryrefslogtreecommitdiffstats
path: root/drivers
diff options
context:
space:
mode:
authorMikulas Patocka <mpatocka@redhat.com>2009-12-11 00:52:34 +0100
committerAlasdair G Kergon <agk@redhat.com>2009-12-11 00:52:34 +0100
commit73dfd078cf8bfee4018fb22f1e2a24f2e05b69dc (patch)
treee5cc81262315df5c44d13d5a2eb731cfe7f8ea86 /drivers
parentdm snapshot: delay merging a chunk until writes to it complete (diff)
downloadlinux-73dfd078cf8bfee4018fb22f1e2a24f2e05b69dc.tar.xz
linux-73dfd078cf8bfee4018fb22f1e2a24f2e05b69dc.zip
dm snapshot: trigger exceptions in remaining snapshots during merge
When there is one merging snapshot and other non-merging snapshots, snapshot_merge_process() must make exceptions in the non-merging snapshots. Use a sequence count to resolve the race between I/O to chunks that are about to be merged. The count increases each time an exception reallocation finishes. Use wait_event() to wait until the count changes. Signed-off-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Diffstat (limited to 'drivers')
-rw-r--r--drivers/md/dm-snap.c83
1 files changed, 83 insertions, 0 deletions
diff --git a/drivers/md/dm-snap.c b/drivers/md/dm-snap.c
index bc52776c69cc..1498704467a7 100644
--- a/drivers/md/dm-snap.c
+++ b/drivers/md/dm-snap.c
@@ -270,6 +270,10 @@ struct origin {
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;
@@ -847,14 +851,38 @@ out:
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;
chunk_t old_chunk, new_chunk;
struct dm_io_region src, dest;
+ uint64_t previous_count;
BUG_ON(!test_bit(RUNNING_MERGE, &s->state_bits));
if (unlikely(test_bit(SHUTDOWN_MERGE, &s->state_bits)))
@@ -887,6 +915,24 @@ static void snapshot_merge_next_chunks(struct dm_snapshot *s)
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, s->store->chunk_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;
@@ -1372,6 +1418,8 @@ static void pending_complete(struct dm_snap_pending_exception *pe, int success)
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 */
@@ -1963,6 +2011,41 @@ static int do_origin(struct dm_dev *origin, struct bio *bio)
}
/*
+ * 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.
*/