Commit cc27b0c7 authored by NeilBrown's avatar NeilBrown Committed by Shaohua Li
Browse files

md: fix deadlock between mddev_suspend() and md_write_start() 



If mddev_suspend() races with md_write_start() we can deadlock
with mddev_suspend() waiting for the request that is currently
in md_write_start() to complete the ->make_request() call,
and md_write_start() waiting for the metadata to be updated
to mark the array as 'dirty'.
As metadata updates done by md_check_recovery() only happen then
the mddev_lock() can be claimed, and as mddev_suspend() is often
called with the lock held, these threads wait indefinitely for each
other.

We fix this by having md_write_start() abort if mddev_suspend()
is happening, and ->make_request() aborts if md_write_start()
aborted.
md_make_request() can detect this abort, decrease the ->active_io
count, and wait for mddev_suspend().

Reported-by: default avatarNix <nix@esperi.org.uk>
Fix: 68866e42(MD: no sync IO while suspended)
Cc: stable@vger.kernel.org
Signed-off-by: default avatarNeilBrown <neilb@suse.com>
Signed-off-by: default avatarShaohua Li <shli@fb.com>
parent 63f700aa

drivers/md/faulty.c

+3 −2
Original line number Diff line number Diff line
@@ -170,7 +170,7 @@ static void add_sector(struct faulty_conf *conf, sector_t start, int mode) 
		conf->nfaults = n+1;

}



static void faulty_make_request(struct mddev *mddev, struct bio *bio)

static bool faulty_make_request(struct mddev *mddev, struct bio *bio)

{

	struct faulty_conf *conf = mddev->private;

	int failit = 0;
@@ -182,7 +182,7 @@ static void faulty_make_request(struct mddev *mddev, struct bio *bio) 
			 * just fail immediately

			 */

			bio_io_error(bio);

			return;

			return true;

		}



		if (check_sector(conf, bio->bi_iter.bi_sector,
@@ -224,6 +224,7 @@ static void faulty_make_request(struct mddev *mddev, struct bio *bio) 
		bio->bi_bdev = conf->rdev->bdev;



	generic_make_request(bio);

	return true;

}



static void faulty_status(struct seq_file *seq, struct mddev *mddev)

drivers/md/linear.c

+4 −3
Original line number Diff line number Diff line
@@ -245,7 +245,7 @@ static void linear_free(struct mddev *mddev, void *priv) 
	kfree(conf);

}



static void linear_make_request(struct mddev *mddev, struct bio *bio)

static bool linear_make_request(struct mddev *mddev, struct bio *bio)

{

	char b[BDEVNAME_SIZE];

	struct dev_info *tmp_dev;
@@ -254,7 +254,7 @@ static void linear_make_request(struct mddev *mddev, struct bio *bio) 


	if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {

		md_flush_request(mddev, bio);

		return;

		return true;

	}



	tmp_dev = which_dev(mddev, bio_sector);
@@ -292,7 +292,7 @@ static void linear_make_request(struct mddev *mddev, struct bio *bio) 
		mddev_check_write_zeroes(mddev, bio);

		generic_make_request(bio);

	}

	return;

	return true;



out_of_bounds:

	pr_err("md/linear:%s: make_request: Sector %llu out of bounds on dev %s: %llu sectors, offset %llu\n",
@@ -302,6 +302,7 @@ out_of_bounds: 
	       (unsigned long long)tmp_dev->rdev->sectors,

	       (unsigned long long)start_sector);

	bio_io_error(bio);

	return true;

}



static void linear_status (struct seq_file *seq, struct mddev *mddev)

drivers/md/md.c

+17 −5
Original line number Diff line number Diff line
@@ -277,7 +277,7 @@ static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio) 
		bio_endio(bio);

		return BLK_QC_T_NONE;

	}

	smp_rmb(); /* Ensure implications of  'active' are visible */

check_suspended:

	rcu_read_lock();

	if (mddev->suspended) {

		DEFINE_WAIT(__wait);
@@ -302,7 +302,11 @@ static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio) 
	sectors = bio_sectors(bio);

	/* bio could be mergeable after passing to underlayer */

	bio->bi_opf &= ~REQ_NOMERGE;

	mddev->pers->make_request(mddev, bio);

	if (!mddev->pers->make_request(mddev, bio)) {

		atomic_dec(&mddev->active_io);

		wake_up(&mddev->sb_wait);

		goto check_suspended;

	}



	cpu = part_stat_lock();

	part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
@@ -327,6 +331,7 @@ void mddev_suspend(struct mddev *mddev) 
	if (mddev->suspended++)

		return;

	synchronize_rcu();

	wake_up(&mddev->sb_wait);

	wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);

	mddev->pers->quiesce(mddev, 1);
@@ -7950,12 +7955,14 @@ EXPORT_SYMBOL(md_done_sync); 
 * If we need to update some array metadata (e.g. 'active' flag

 * in superblock) before writing, schedule a superblock update

 * and wait for it to complete.

 * A return value of 'false' means that the write wasn't recorded

 * and cannot proceed as the array is being suspend.

 */

void md_write_start(struct mddev *mddev, struct bio *bi)

bool md_write_start(struct mddev *mddev, struct bio *bi)

{

	int did_change = 0;

	if (bio_data_dir(bi) != WRITE)

		return;

		return true;



	BUG_ON(mddev->ro == 1);

	if (mddev->ro == 2) {
@@ -7987,7 +7994,12 @@ void md_write_start(struct mddev *mddev, struct bio *bi) 
	if (did_change)

		sysfs_notify_dirent_safe(mddev->sysfs_state);

	wait_event(mddev->sb_wait,

		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));

		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) && !mddev->suspended);

	if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {

		percpu_ref_put(&mddev->writes_pending);

		return false;

	}

	return true;

}

EXPORT_SYMBOL(md_write_start);

drivers/md/md.h

+2 −2
Original line number Diff line number Diff line
@@ -510,7 +510,7 @@ struct md_personality 
	int level;

	struct list_head list;

	struct module *owner;

	void (*make_request)(struct mddev *mddev, struct bio *bio);

	bool (*make_request)(struct mddev *mddev, struct bio *bio);

	int (*run)(struct mddev *mddev);

	void (*free)(struct mddev *mddev, void *priv);

	void (*status)(struct seq_file *seq, struct mddev *mddev);
@@ -649,7 +649,7 @@ extern void md_wakeup_thread(struct md_thread *thread); 
extern void md_check_recovery(struct mddev *mddev);

extern void md_reap_sync_thread(struct mddev *mddev);

extern int mddev_init_writes_pending(struct mddev *mddev);

extern void md_write_start(struct mddev *mddev, struct bio *bi);

extern bool md_write_start(struct mddev *mddev, struct bio *bi);

extern void md_write_inc(struct mddev *mddev, struct bio *bi);

extern void md_write_end(struct mddev *mddev);

extern void md_done_sync(struct mddev *mddev, int blocks, int ok);

drivers/md/multipath.c

+4 −4
Original line number Diff line number Diff line
@@ -106,7 +106,7 @@ static void multipath_end_request(struct bio *bio) 
	rdev_dec_pending(rdev, conf->mddev);

}



static void multipath_make_request(struct mddev *mddev, struct bio * bio)

static bool multipath_make_request(struct mddev *mddev, struct bio * bio)

{

	struct mpconf *conf = mddev->private;

	struct multipath_bh * mp_bh;
@@ -114,7 +114,7 @@ static void multipath_make_request(struct mddev *mddev, struct bio * bio) 


	if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {

		md_flush_request(mddev, bio);

		return;

		return true;

	}



	mp_bh = mempool_alloc(conf->pool, GFP_NOIO);
@@ -126,7 +126,7 @@ static void multipath_make_request(struct mddev *mddev, struct bio * bio) 
	if (mp_bh->path < 0) {

		bio_io_error(bio);

		mempool_free(mp_bh, conf->pool);

		return;

		return true;

	}

	multipath = conf->multipaths + mp_bh->path;
@@ -141,7 +141,7 @@ static void multipath_make_request(struct mddev *mddev, struct bio * bio) 
	mddev_check_writesame(mddev, &mp_bh->bio);

	mddev_check_write_zeroes(mddev, &mp_bh->bio);

	generic_make_request(&mp_bh->bio);

	return;

	return true;

}



static void multipath_status(struct seq_file *seq, struct mddev *mddev)

CRA Git | Maintained and supported by SUSTech CRA and CCSE