*---------------------------------------------------------------*/
enum dm_raid1_error {
DM_RAID1_WRITE_ERROR,
+ DM_RAID1_FLUSH_ERROR,
DM_RAID1_SYNC_ERROR,
DM_RAID1_READ_ERROR
};
struct bio_list reads;
struct bio_list writes;
struct bio_list failures;
+ struct bio_list holds; /* bios are waiting until suspend */
struct dm_region_hash *rh;
struct dm_kcopyd_client *kcopyd_client;
region_t nr_regions;
int in_sync;
int log_failure;
+ int leg_failure;
atomic_t suspend;
atomic_t default_mirror; /* Default mirror */
atomic_set(&ms->default_mirror, m - m0);
}
+static struct mirror *get_valid_mirror(struct mirror_set *ms)
+{
+ struct mirror *m;
+
+ for (m = ms->mirror; m < ms->mirror + ms->nr_mirrors; m++)
+ if (!atomic_read(&m->error_count))
+ return m;
+
+ return NULL;
+}
+
/* fail_mirror
* @m: mirror device to fail
* @error_type: one of the enum's, DM_RAID1_*_ERROR
struct mirror_set *ms = m->ms;
struct mirror *new;
+ ms->leg_failure = 1;
+
/*
* error_count is used for nothing more than a
* simple way to tell if a device has encountered
goto out;
}
- for (new = ms->mirror; new < ms->mirror + ms->nr_mirrors; new++)
- if (!atomic_read(&new->error_count)) {
- set_default_mirror(new);
- break;
- }
-
- if (unlikely(new == ms->mirror + ms->nr_mirrors))
+ new = get_valid_mirror(ms);
+ if (new)
+ set_default_mirror(new);
+ else
DMWARN("All sides of mirror have failed.");
out:
for (i = 0; i < ms->nr_mirrors; i++)
if (test_bit(i, &error_bits))
fail_mirror(ms->mirror + i,
- DM_RAID1_WRITE_ERROR);
+ DM_RAID1_FLUSH_ERROR);
return -EIO;
}
io->count = bio->bi_size >> 9;
}
+static void hold_bio(struct mirror_set *ms, struct bio *bio)
+{
+ /*
+ * Lock is required to avoid race condition during suspend
+ * process.
+ */
+ spin_lock_irq(&ms->lock);
+
+ if (atomic_read(&ms->suspend)) {
+ spin_unlock_irq(&ms->lock);
+
+ /*
+ * If device is suspended, complete the bio.
+ */
+ if (dm_noflush_suspending(ms->ti))
+ bio_endio(bio, DM_ENDIO_REQUEUE);
+ else
+ bio_endio(bio, -EIO);
+ return;
+ }
+
+ /*
+ * Hold bio until the suspend is complete.
+ */
+ bio_list_add(&ms->holds, bio);
+ spin_unlock_irq(&ms->lock);
+}
+
/*-----------------------------------------------------------------
* Reads
*---------------------------------------------------------------*/
unsigned i, ret = 0;
struct bio *bio = (struct bio *) context;
struct mirror_set *ms;
- int uptodate = 0;
int should_wake = 0;
unsigned long flags;
* This way we handle both writes to SYNC and NOSYNC
* regions with the same code.
*/
- if (likely(!error))
- goto out;
+ if (likely(!error)) {
+ bio_endio(bio, ret);
+ return;
+ }
for (i = 0; i < ms->nr_mirrors; i++)
if (test_bit(i, &error))
fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR);
- else
- uptodate = 1;
- if (unlikely(!uptodate)) {
- DMERR("All replicated volumes dead, failing I/O");
- /* None of the writes succeeded, fail the I/O. */
- ret = -EIO;
- } else if (errors_handled(ms)) {
- /*
- * Need to raise event. Since raising
- * events can block, we need to do it in
- * the main thread.
- */
- spin_lock_irqsave(&ms->lock, flags);
- if (!ms->failures.head)
- should_wake = 1;
- bio_list_add(&ms->failures, bio);
- spin_unlock_irqrestore(&ms->lock, flags);
- if (should_wake)
- wakeup_mirrord(ms);
- return;
- }
-out:
- bio_endio(bio, ret);
+ /*
+ * Need to raise event. Since raising
+ * events can block, we need to do it in
+ * the main thread.
+ */
+ spin_lock_irqsave(&ms->lock, flags);
+ if (!ms->failures.head)
+ should_wake = 1;
+ bio_list_add(&ms->failures, bio);
+ spin_unlock_irqrestore(&ms->lock, flags);
+ if (should_wake)
+ wakeup_mirrord(ms);
}
static void do_write(struct mirror_set *ms, struct bio *bio)
/*
* Dispatch io.
*/
- if (unlikely(ms->log_failure)) {
+ if (unlikely(ms->log_failure) && errors_handled(ms)) {
spin_lock_irq(&ms->lock);
bio_list_merge(&ms->failures, &sync);
spin_unlock_irq(&ms->lock);
dm_rh_delay(ms->rh, bio);
while ((bio = bio_list_pop(&nosync))) {
- map_bio(get_default_mirror(ms), bio);
- generic_make_request(bio);
+ if (unlikely(ms->leg_failure) && errors_handled(ms)) {
+ spin_lock_irq(&ms->lock);
+ bio_list_add(&ms->failures, bio);
+ spin_unlock_irq(&ms->lock);
+ wakeup_mirrord(ms);
+ } else {
+ map_bio(get_default_mirror(ms), bio);
+ generic_make_request(bio);
+ }
}
}
{
struct bio *bio;
- if (!failures->head)
+ if (likely(!failures->head))
return;
- if (!ms->log_failure) {
- while ((bio = bio_list_pop(failures))) {
- ms->in_sync = 0;
- dm_rh_mark_nosync(ms->rh, bio, bio->bi_size, 0);
- }
- return;
- }
-
/*
* If the log has failed, unattempted writes are being
- * put on the failures list. We can't issue those writes
+ * put on the holds list. We can't issue those writes
* until a log has been marked, so we must store them.
*
* If a 'noflush' suspend is in progress, we can requeue
* for us to treat them the same and requeue them
* as well.
*/
- if (dm_noflush_suspending(ms->ti)) {
- while ((bio = bio_list_pop(failures)))
- bio_endio(bio, DM_ENDIO_REQUEUE);
- return;
- }
+ while ((bio = bio_list_pop(failures))) {
+ if (!ms->log_failure) {
+ ms->in_sync = 0;
+ dm_rh_mark_nosync(ms->rh, bio);
+ }
- if (atomic_read(&ms->suspend)) {
- while ((bio = bio_list_pop(failures)))
+ /*
+ * If all the legs are dead, fail the I/O.
+ * If we have been told to handle errors, hold the bio
+ * and wait for userspace to deal with the problem.
+ * Otherwise pretend that the I/O succeeded. (This would
+ * be wrong if the failed leg returned after reboot and
+ * got replicated back to the good legs.)
+ */
+ if (!get_valid_mirror(ms))
bio_endio(bio, -EIO);
- return;
+ else if (errors_handled(ms))
+ hold_bio(ms, bio);
+ else
+ bio_endio(bio, 0);
}
-
- spin_lock_irq(&ms->lock);
- bio_list_merge(&ms->failures, failures);
- spin_unlock_irq(&ms->lock);
-
- delayed_wake(ms);
}
static void trigger_event(struct work_struct *work)
}
spin_lock_init(&ms->lock);
+ bio_list_init(&ms->reads);
+ bio_list_init(&ms->writes);
+ bio_list_init(&ms->failures);
+ bio_list_init(&ms->holds);
ms->ti = ti;
ms->nr_mirrors = nr_mirrors;
ms->nr_regions = dm_sector_div_up(ti->len, region_size);
ms->in_sync = 0;
ms->log_failure = 0;
+ ms->leg_failure = 0;
atomic_set(&ms->suspend, 0);
atomic_set(&ms->default_mirror, DEFAULT_MIRROR);
return -EINVAL;
}
- if (dm_get_device(ti, argv[0], offset, ti->len,
- dm_table_get_mode(ti->table),
+ if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
&ms->mirror[mirror].dev)) {
ti->error = "Device lookup failure";
return -ENXIO;
struct mirror_set *ms = (struct mirror_set *) ti->private;
struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
+ struct bio_list holds;
+ struct bio *bio;
+
atomic_set(&ms->suspend, 1);
/*
+ * Process bios in the hold list to start recovery waiting
+ * for bios in the hold list. After the process, no bio has
+ * a chance to be added in the hold list because ms->suspend
+ * is set.
+ */
+ spin_lock_irq(&ms->lock);
+ holds = ms->holds;
+ bio_list_init(&ms->holds);
+ spin_unlock_irq(&ms->lock);
+
+ while ((bio = bio_list_pop(&holds)))
+ hold_bio(ms, bio);
+
+ /*
* We must finish up all the work that we've
* generated (i.e. recovery work).
*/
if (!atomic_read(&(m->error_count)))
return 'A';
- return (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
+ return (test_bit(DM_RAID1_FLUSH_ERROR, &(m->error_type))) ? 'F' :
+ (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
(test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' :
(test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U';
}