Commit 8c242593 authored by Yufen Yu's avatar Yufen Yu Committed by Shaohua Li

md/raid1: exit sync request if MD_RECOVERY_INTR is set

We met a sync thread stuck as follows:

 raid1_sync_request+0x2c9/0xb50
 md_do_sync+0x983/0xfa0
 md_thread+0x11c/0x160
 kthread+0x111/0x130
 ret_from_fork+0x35/0x40
 0xffffffffffffffff

At the same time, there is a stuck mdadm thread (mdadm --manage
/dev/md2 --add /dev/sda). It is trying to stop the sync thread:

 kthread_stop+0x42/0xf0
 md_unregister_thread+0x3a/0x70
 md_reap_sync_thread+0x15/0x160
 action_store+0x142/0x2a0
 md_attr_store+0x6c/0xb0
 kernfs_fop_write+0x102/0x180
 __vfs_write+0x33/0x170
 vfs_write+0xad/0x1a0
 SyS_write+0x52/0xc0
 do_syscall_64+0x6e/0x190
 entry_SYSCALL_64_after_hwframe+0x3d/0xa2

Debug tools show that the sync thread is waiting in raise_barrier(),
until raid1d() end all normal IO bios into bio_end_io_list(introduced
in commit 55ce74d4). But, raid1d() cannot end these bios if
MD_CHANGE_PENDING bit is set. It needs to get mddev->reconfig_mutex lock
and then clear the bit in md_check_recovery().
However, the lock is holding by mdadm in action_store().

Thus, there is a loop:
mdadm waiting for sync thread to stop, sync thread waiting for
raid1d() to end bios, raid1d() waiting for mdadm to release
mddev->reconfig_mutex lock and then it can end bios.

Fix this by checking MD_RECOVERY_INTR while waiting in raise_barrier(),
so that sync thread can exit while mdadm is stoping the sync thread.

Fixes: 55ce74d4 ("md/raid1: ensure device failure recorded before write request returns.")
Signed-off-by: default avatarJason Yan <yanaijie@huawei.com>
Signed-off-by: default avatarYufen Yu <yuyufen@huawei.com>
Signed-off-by: default avatarShaohua Li <shli@fb.com>
parent 0ea9924a
......@@ -854,7 +854,7 @@ static void flush_pending_writes(struct r1conf *conf)
* there is no normal IO happeing. It must arrange to call
* lower_barrier when the particular background IO completes.
*/
static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
static sector_t raise_barrier(struct r1conf *conf, sector_t sector_nr)
{
int idx = sector_to_idx(sector_nr);
......@@ -885,13 +885,23 @@ static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
* max resync count which allowed on current I/O barrier bucket.
*/
wait_event_lock_irq(conf->wait_barrier,
!conf->array_frozen &&
(!conf->array_frozen &&
!atomic_read(&conf->nr_pending[idx]) &&
atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH,
atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) ||
test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery),
conf->resync_lock);
if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) {
atomic_dec(&conf->barrier[idx]);
spin_unlock_irq(&conf->resync_lock);
wake_up(&conf->wait_barrier);
return -EINTR;
}
atomic_inc(&conf->nr_sync_pending);
spin_unlock_irq(&conf->resync_lock);
return 0;
}
static void lower_barrier(struct r1conf *conf, sector_t sector_nr)
......@@ -2662,9 +2672,12 @@ static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
bitmap_cond_end_sync(mddev->bitmap, sector_nr,
mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
r1_bio = raid1_alloc_init_r1buf(conf);
raise_barrier(conf, sector_nr);
if (raise_barrier(conf, sector_nr))
return 0;
r1_bio = raid1_alloc_init_r1buf(conf);
rcu_read_lock();
/*
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment