[PATCH] btrfs: relax dev_replace rwsem usage on scrub with rst

[Johannes Thumshirn <jth@kernel.org>]

From: Johannes Thumshin <johannes.thumshirn@wdc.com>

Running fstests btrfs/011 with MKFS_OPTIONS="-O rst" to force the usage of
the RAID stripe-tree, we get the following splat from lockdep:

 BTRFS info (device sdd): dev_replace from /dev/sdd (devid 1) to /dev/sdb started

 ============================================
 WARNING: possible recursive locking detected
 6.11.0-rc3+ #592 Not tainted
 --------------------------------------------
 btrfs/4203 is trying to acquire lock:
 ffff888103f35c98 (&fs_info->dev_replace.rwsem){++++}-{3:3}, at: btrfs_map_block+0x39f/0x2250

 but task is already holding lock:
 ffff888103f35c98 (&fs_info->dev_replace.rwsem){++++}-{3:3}, at: btrfs_map_block+0x39f/0x2250

 other info that might help us debug this:
  Possible unsafe locking scenario:

        CPU0
        ----
   lock(&fs_info->dev_replace.rwsem);
   lock(&fs_info->dev_replace.rwsem);

  *** DEADLOCK ***

  May be due to missing lock nesting notation

 1 lock held by btrfs/4203:
  #0: ffff888103f35c98 (&fs_info->dev_replace.rwsem){++++}-{3:3}, at: btrfs_map_block+0x39f/0x2250

This fixes a deadlock on RAID stripe-tree where device replace performs a
scrub operation, which in turn calls into btrfs_map_block() to find the
physical location of the block.

Cc: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
---


Signed-off-by: Johannes Thumshirn <jth@kernel.org>
---
 fs/btrfs/volumes.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 4b9b647a7e29..e5bd2bee912d 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -6459,6 +6459,7 @@ int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
 	int dev_replace_is_ongoing = 0;
 	u16 num_alloc_stripes;
 	u64 max_len;
+	bool rst;
 
 	ASSERT(bioc_ret);
 
@@ -6475,6 +6476,8 @@ int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
 	if (io_geom.mirror_num > num_copies)
 		return -EINVAL;
 
+	rst = btrfs_need_stripe_tree_update(fs_info, map->type);
+
 	map_offset = logical - map->start;
 	io_geom.raid56_full_stripe_start = (u64)-1;
 	max_len = btrfs_max_io_len(map, map_offset, &io_geom);
@@ -6597,13 +6600,19 @@ int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
 		 * For all other non-RAID56 profiles, just copy the target
 		 * stripe into the bioc.
 		 */
+		if (rst && dev_replace_is_ongoing)
+			up_read(&dev_replace->rwsem);
 		for (int i = 0; i < io_geom.num_stripes; i++) {
 			ret = set_io_stripe(fs_info, logical, length,
 					    &bioc->stripes[i], map, &io_geom);
+
 			if (ret < 0)
 				break;
 			io_geom.stripe_index++;
 		}
+		if (rst && dev_replace_is_ongoing)
+			down_read(&dev_replace->rwsem);
+
 	}
 
 	if (ret) {

---
base-commit: 4ce21d87ae81a86b488e0d326682883485317dcb
change-id: 20240814-dev_replace_rwsem-new-91c160579246

Best regards,
-- 
Johannes Thumshirn <jth@kernel.org>