On 2019/10/21 10:32 PM, Johannes Thumshirn wrote:
On 21/10/2019 11:17, Johannes Thumshirn wrote:
[...]
-----
$ cat run-btrfs-test
modprobe -iv zram num_devices=8
udevadm trigger
sync
zramctl /dev/zram0 -s 8G && \
zramctl /dev/zram1 -s 8G && \
zramctl /dev/zram2 -s 4G && \
zramctl /dev/zram3 -s 4G && \
zramctl /dev/zram4 -s 4G && \
zramctl /dev/zram5 -s 2G && \
zramctl /dev/zram6 -s 2G && \
zramctl /dev/zram7 -s 4G && \
mkfs.btrfs /dev/zram0 && \
mkdir -p /mnt/btrfs-test && \
mount /dev/zram0 /mnt/btrfs-test && \
echo "FS Mounted" && \
btrfs dev add /dev/zram1 /mnt/btrfs-test && \
echo "Devices added" && \
for int in {1..500} ; do dd if=/dev/zero of=/mnt/btrfs-test/file${int}
bs=32M count=1 && sync ; done
btrfs dev add /dev/zram[2-7] /mnt/btrfs-test && \
btrfs fi sh /mnt/btrfs-test && \
btrfs fi df /mnt/btrfs-test && \
btrfs bal star -mconvert=raid6 /mnt/btrfs-test && \
btrfs bal star -dconvert=raid6 /mnt/btrfs-test
btrfs fi sh /mnt/btrfs-test && \
btrfs fi df /mnt/btrfs-test
I'm sorry. I ran this script in a loop for 35 iterations on 5.3.6 and
couldn't reproduce a single crash.
Interesting thing I met too. That's not reproducible on my VM but
host (Archlinux v5.3.6 same kernel config).
What's more interesting is that v5.3.7 seems to have fixed the bug.
After some bisect. The commit is
commit 417d26300214f7b593a99c6bc8badb66492ae322
Author: Qu Wenruo <wqu@xxxxxxxx>
Date: Mon Sep 23 14:56:14 2019 +0800
btrfs: relocation: fix use-after-free on dead relocation roots
commit 1fac4a54374f7ef385938f3c6cf7649c0fe4f6cd upstream.
--
Su
Is there anything else needed?
