On 2023/4/20 04:23, Torstein Eide wrote:
This is for pre-6.4 kernels, as scrub code goes through a huge rework.
[BUG]
Even before the scrub rework, if we have some corrupted metadata failed
to be repaired during replace, we still continue replace and let it
finish just as there is nothing wrong:
BTRFS info (device dm-4): dev_replace from /dev/mapper/test-scratch1 (devid 1) to /dev/mapper/test-scratch2 started
BTRFS warning (device dm-4): tree block 5578752 mirror 1 has bad csum, has 0x00000000 want 0xade80ca1
BTRFS warning (device dm-4): tree block 5578752 mirror 0 has bad csum, has 0x00000000 want 0xade80ca1
BTRFS warning (device dm-4): checksum error at logical 5578752 on dev /dev/mapper/test-scratch1, physical 5578752: metadata leaf (level 0) in tree 5
BTRFS warning (device dm-4): checksum error at logical 5578752 on dev /dev/mapper/test-scratch1, physical 5578752: metadata leaf (level 0) in tree 5
BTRFS error (device dm-4): bdev /dev/mapper/test-scratch1 errs: wr 0, rd 0, flush 0, corrupt 1, gen 0
BTRFS warning (device dm-4): tree block 5578752 mirror 1 has bad bytenr, has 0 want 5578752
BTRFS error (device dm-4): unable to fixup (regular) error at logical 5578752 on dev /dev/mapper/test-scratch1
BTRFS info (device dm-4): dev_replace from /dev/mapper/test-scratch1 (devid 1) to /dev/mapper/test-scratch2 finished
This can lead to unexpected problems for the result fs.
[CAUSE]
Btrfs reuses scrub code path for dev-replace to iterate all dev extents.
But unlike scrub, dev-replace doesn't really bother to check the scrub
progress, which records all the errors found during replace.
And even if we checks the progress, we can not really determine which
errors are minor, which are critical just by the plain numbers.
(remember we don't treat metadata/data checksum error differently).
This behavior is there from the very beginning.
[FIX]
Instead of continue the replace, just error out if we hit an unrepaired
metadata sector.
Now the dev-replace would be rejected with -EIO, to inform the user.
Although it also means, the fs has some metadata error which can not be
repaired, the user would be super upset anyway.
If one sector is bad in metadata how much secondary data is damaged?
If it's metadata, it's highly possible the metadata has extra copy by
default.
(Single disk btrfs goes DUP for meta, while multi-disk one goes RAID1 by
default).
And if it's a metadata corruption, good luck it's not a critical one, or
this would be your last time to mount the fs.
As someone who recently had to remove, and currently replace a disk.
it is upsetting, if it stopped if we stopped because 0,01% of data is
unrepairable, if we can save the other 99,99%. Can we have it
continue, print an error message to standard out, and a way for the
user to delete or copy it (with som option like -force-delete or
--force-copy) with the error to the new disk?
"Metadata at block 5578752 is damaged and unrepaired. Skipping. Read
`man btrfs-replace` for more info. "
At the end if possible, list files affected by the damaged metadata blocks.
In man answer:
How can the user know what files are connected to the metadata?
How can a user decide what to do with the damaged metadata?
At minimum, can there be some useful info to the info to the error output? like
"Replace has stopped, due to reading unrepaired metadata block, was
working on block 5578752, se `dmesg` for more details. (\s Sorry but
you are currently f..k)"
I'd say, if your metadata is already corrupted, you don't want a
"generic" solution, but ask for a developer to guide you how to recover.
Thanks,
Qu
The new dmesg would look like this:
BTRFS info (device dm-4): dev_replace from /dev/mapper/test-scratch1 (devid 1) to /dev/mapper/test-scratch2 started
BTRFS warning (device dm-4): tree block 5578752 mirror 1 has bad csum, has 0x00000000 want 0xade80ca1
BTRFS warning (device dm-4): tree block 5578752 mirror 1 has bad csum, has 0x00000000 want 0xade80ca1
BTRFS error (device dm-4): unable to fixup (regular) error at logical 5570560 on dev /dev/mapper/test-scratch1 physical 5570560
BTRFS warning (device dm-4): header error at logical 5570560 on dev /dev/mapper/test-scratch1, physical 5570560: metadata leaf (level 0) in tree 5
BTRFS warning (device dm-4): header error at logical 5570560 on dev /dev/mapper/test-scratch1, physical 5570560: metadata leaf (level 0) in tree 5
BTRFS error (device dm-4): stripe 5570560 has unrepaired metadata sector at 5578752
BTRFS error (device dm-4): btrfs_scrub_dev(/dev/mapper/test-scratch1, 1, /dev/mapper/test-scratch2) failed -5
CC: stable@xxxxxxxxxxxxxxx
Signed-off-by: Qu Wenruo <wqu@xxxxxxxx>
---
I'm not sure how should we merge this patch.
The misc-next is already merging the new scrub code, but the problem is
there for all old kernels thus we need such fixes.
Maybe we can merge this fix before the scrub rework, then the rework,
and finally the better fix using reworked interface?
---
fs/btrfs/scrub.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c
index ef4046a2572c..71f64b9bcd9f 100644
--- a/fs/btrfs/scrub.c
+++ b/fs/btrfs/scrub.c
@@ -195,6 +195,7 @@ struct scrub_ctx {
struct mutex wr_lock;
struct btrfs_device *wr_tgtdev;
bool flush_all_writes;
+ bool has_meta_failed;
/*
* statistics
@@ -1380,6 +1381,8 @@ static int scrub_handle_errored_block(struct scrub_block *sblock_to_check)
btrfs_err_rl_in_rcu(fs_info,
"unable to fixup (regular) error at logical %llu on dev %s",
logical, btrfs_dev_name(dev));
+ if (is_metadata)
+ sctx->has_meta_failed = true;
}
out:
@@ -3838,6 +3841,12 @@ static noinline_for_stack int scrub_stripe(struct scrub_ctx *sctx,
blk_finish_plug(&plug);
+ /*
+ * If we have metadata unable to be repaired, we should error
+ * out the dev-replace.
+ */
+ if (sctx->is_dev_replace && sctx->has_meta_failed && ret >= 0)
+ ret = -EIO;
if (sctx->is_dev_replace && ret >= 0) {
int ret2;
