On 2017-05-16 05:53, Tom Hale wrote:
Hi Chris,
On 09/05/17 02:26, Chris Murphy wrote:
Read errors are fixed by overwrites. If the underlying device doesn't
report an error for the write command, it's assumed to succeed. Even
md and LVM raid's do this.
I understand assuming writes succeed in general. However, for a tool
which says (in its usage):
Outside of dm-error and/or dm-flakey, what hardware have you seen that
fails writes without reporting a write error? SSD's might do it at the
lowest level (below firmware), but the SSD firmware will notice and
should correct it itself, and it won't happen persistently unless the
device is essentially dead, but in general, short of the storage media
itself failing, it is entirely safe to assume that what you told the
device to write is what got written as long as the write commands succeeded.
"verify checksums of data and metadata"
I would expect that the tool reports the *final state* of the data:
Which is easy enough to do by hand without needing to make scrub do more
work all the time (just run a read-only scrub after the first one).
* Because the explicit function of scrub is to verify that all data can
be read.
Not exactly, it's function is to verify that checksums are correct,
verifying that there are no low-level read errors is simply a side
effect of that.
* In a RAID1 situation, the sysadmin could think that the there are two
valid copies of all data, when in fact there may only be one.
What are the complete kernel messages for the scrub event? This should
show what problem Btrfs detects and how it fixes it, and what sectors
it's fixing each time.
Attached are the kernel logs (and the `grep fixed` version) showing the
same logical blocks are "fixed" twice.
Below (and attached because of formatting) I show how to reproduce the
uncorrected errors, producing the attached logs.
Please reply-all as I'm not subscribed.
----------------------------------------------------
$ pacman -Q btrfs-progs
btrfs-progs 4.10.2-1
$ uname -r
4.9.24-1-MANJARO
# fallocate -l 500M good-disk
# fallocate -l 500M bad-disk
# losetup -f bad-disk # loop0
# losetup -f good-disk # loop1
# mdadm --create -v /dev/md0 --level=linear --force --raid-devices=1
/dev/loop0
mdadm: Defaulting to version 1.2 metadata
mdadm: array /dev/md0 started.
# mkfs.btrfs -m raid1 -d raid1 /dev/loop1 /dev/md0
btrfs-progs v4.10.2
See http://btrfs.wiki.kernel.org for more information.
Performing full device TRIM /dev/loop1 (500.00MiB) ...
Performing full device TRIM /dev/md0 (499.73MiB) ...
Label: (null)
UUID: d748537b-3fa4-47cc-9934-62cf391fb638
Node size: 16384
Sector size: 4096
Filesystem size: 999.73MiB
Block group profiles:
Data: RAID1 64.00MiB
Metadata: RAID1 49.94MiB
System: RAID1 8.00MiB
SSD detected: no
Incompat features: extref, skinny-metadata
Number of devices: 2
Devices:
ID SIZE PATH
1 500.00MiB /dev/loop1
2 499.73MiB /dev/md0
# mount /dev/md0 /mnt/tmp
# dd if=/dev/urandom of=/mnt/tmp/rand bs=1M
dd: error writing '/mnt/tmp/rand': No space left on device
441+0 records in
440+0 records out
461963264 bytes (462 MB, 441 MiB) copied, 5.74358 s, 80.4 MB/s
# umount /mnt/tmp
# mdadm --stop /dev/md0
mdadm: stopped /dev/md0
# mdadm --build /dev/md0 --level=faulty --raid-devices=1 /dev/loop0
mdadm: array /dev/md0 built and started.
├─vg_svelte-swap swap suspend
0792d327-566d-4306-93f4-810e247827e4 [SWAP]
└─vg_svelte-rootfs btrfs root
a1d2a0b4-8d4b-4c1c-9c87-5a55de61918e /mnt/btrfs-vol/rootfs
# lsblk -f | head -n4
NAME FSTYPE LABEL UUID
MOUNTPOINT
loop0 linux_raid_member svelte:0
f7019aa8-3b2d-4e4c-9106-635d0dddca78
└─md0 linux_raid_member svelte:0
f7019aa8-3b2d-4e4c-9106-635d0dddca78
└─md127 btrfs
d748537b-3fa4-47cc-9934-62cf391fb638
# mount /dev/md127 /mnt/tmp
# btrfs scrub start -Bd /mnt/tmp
scrub device /dev/loop1 (id 1) done
scrub started at Tue May 16 13:04:50 2017 and finished after
00:00:03
total bytes scrubbed: 441.22MiB with 0 errors
scrub device /dev/md127 (id 2) done
scrub started at Tue May 16 13:04:50 2017 and finished after
00:00:02
total bytes scrubbed: 441.22MiB with 0 errors
# # Introduce errors in /dev/md127:
# mdadm --grow /dev/md0 --layout=rp400
layout for /dev/md0 set to 12803
# btrfs scrub start -Bd /mnt/tmp
scrub device /dev/loop1 (id 1) done
scrub started at Tue May 16 13:07:57 2017 and finished after
00:00:02
total bytes scrubbed: 441.22MiB with 0 errors
scrub device /dev/md127 (id 2) done
scrub started at Tue May 16 13:07:57 2017 and finished after
00:00:03
total bytes scrubbed: 441.22MiB with 19 errors
error details: read=19
corrected errors: 19, uncorrectable errors: 0, unverified
errors: 589
WARNING: errors detected during scrubbing, corrected
# # Stop producing additional errors, keeping the already existing
badblocks:
# mdadm --grow /dev/md0 --layout=clear
layout for /dev/md0 set to 31
# # Run scrub for 2nd time:
# btrfs scrub start -Bd /mnt/tmp
scrub device /dev/loop1 (id 1) done
scrub started at Tue May 16 13:10:19 2017 and finished after
00:00:03
total bytes scrubbed: 441.22MiB with 0 errors
scrub device /dev/md127 (id 2) done
scrub started at Tue May 16 13:10:19 2017 and finished after
00:00:03
total bytes scrubbed: 441.22MiB with 19 errors
error details: read=19
corrected errors: 19, uncorrectable errors: 0, unverified
errors: 589
WARNING: errors detected during scrubbing, corrected
# # Clean up
# umount /mnt/tmp
# mdadm --stop /dev/md127
mdadm: stopped /dev/md127
# mdadm --stop /dev/md0
mdadm: stopped /dev/md0
# losetup -d /dev/loop[01]
#
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
