btrfs: handle case when repair happens with dev-replace

[BUG]
There is a bug report that a BUG_ON() in btrfs_repair_io_failure()
(originally repair_io_failure() in v6.0 kernel) got triggered when
replacing a unreliable disk:

  BTRFS warning (device sda1): csum failed root 257 ino 2397453 off 39624704 csum 0xb0d18c75 expected csum 0x4dae9c5e mirror 3
  kernel BUG at fs/btrfs/extent_io.c:2380!
  invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
  CPU: 9 PID: 3614331 Comm: kworker/u257:2 Tainted: G           OE      6.0.0-5-amd64 #1  Debian 6.0.10-2
  Hardware name: Micro-Star International Co., Ltd. MS-7C60/TRX40 PRO WIFI (MS-7C60), BIOS 2.70 07/01/2021
  Workqueue: btrfs-endio btrfs_end_bio_work [btrfs]
  RIP: 0010:repair_io_failure+0x24a/0x260 [btrfs]
  Call Trace:
   <TASK>
   clean_io_failure+0x14d/0x180 [btrfs]
   end_bio_extent_readpage+0x412/0x6e0 [btrfs]
   ? __switch_to+0x106/0x420
   process_one_work+0x1c7/0x380
   worker_thread+0x4d/0x380
   ? rescuer_thread+0x3a0/0x3a0
   kthread+0xe9/0x110
   ? kthread_complete_and_exit+0x20/0x20
   ret_from_fork+0x22/0x30

[CAUSE]

Before the BUG_ON(), we got some read errors from the replace target
first, note the mirror number (3, which is beyond RAID1 duplication,
thus it's read from the replace target device).

Then at the BUG_ON() location, we are trying to writeback the repaired
sectors back the failed device.

The check looks like this:

		ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical,
				      &map_length, &bioc, mirror_num);
		if (ret)
			goto out_counter_dec;
		BUG_ON(mirror_num != bioc->mirror_num);

But inside btrfs_map_block(), we can modify bioc->mirror_num especially
for dev-replace:

	if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
	    !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
		ret = get_extra_mirror_from_replace(fs_info, logical, *length,
						    dev_replace->srcdev->devid,
						    &mirror_num,
					    &physical_to_patch_in_first_stripe);
		patch_the_first_stripe_for_dev_replace = 1;
	}

Thus if we're repairing the replace target device, we're going to
trigger that BUG_ON().

But in reality, the read failure from the replace target device may be
that, our replace hasn't reached the range we're reading, thus we're
reading garbage, but with replace running, the range would be properly
filled later.

Thus in that case, we don't need to do anything but let the replace
routine to handle it.

[FIX]
Instead of a BUG_ON(), just skip the repair if we're repairing the
device replace target device.

Reported-by: 小太 <nospam@kota.moe>
Link: https://lore.kernel.org/linux-btrfs/CACsxjPYyJGQZ+yvjzxA1Nn2LuqkYqTCcUH43S=+wXhyf8S00Ag@mail.gmail.com/
CC: stable@vger.kernel.org # 6.0+
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

1 files changed, 10 insertions, 1 deletions

diff --git a/fs/btrfs/bio.c b/fs/btrfs/bio.c
index b8fb7ef6b52060..8affc88b0e0a4b 100644
--- a/fs/btrfs/bio.c
+++ b/fs/btrfs/bio.c
@@ -329,7 +329,16 @@ int btrfs_repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start,
 				      &map_length, &bioc, mirror_num);
 		if (ret)
 			goto out_counter_dec;
-		BUG_ON(mirror_num != bioc->mirror_num);
+		/*
+		 * This happens when dev-replace is also running, and the
+		 * mirror_num indicates the dev-replace target.
+		 *
+		 * In this case, we don't need to do anything, as the read
+		 * error just means the replace progress hasn't reached our
+		 * read range, and later replace routine would handle it well.
+		 */
+		if (mirror_num != bioc->mirror_num)
+			goto out_counter_dec;
 	}
 
 	sector = bioc->stripes[bioc->mirror_num - 1].physical >> 9;