In the Linux kernel, the following vulnerability has been resolved:
f2fs: check discard support for conventional zones
As the helper function f2fs_bdev_support_discard() shows, f2fs checks if
the target block devices support discard by calling
bdev_max_discard_sectors() and bdev_is_zoned(). This check works well
for most cases, but it does not work for conventional zones on zoned
block devices. F2fs assumes that zoned block devices support discard,
and calls __submit_discard_cmd(). When __submit_discard_cmd() is called
for sequential write required zones, it works fine since
__submit_discard_cmd() issues zone reset commands instead of discard
commands. However, when __submit_discard_cmd() is called for
conventional zones, __blkdev_issue_discard() is called even when the
devices do not support discard.
The inappropriate __blkdev_issue_discard() call was not a problem before
the commit 30f1e7241422 ("block: move discard checks into the ioctl
handler") because __blkdev_issue_discard() checked if the target devices
support discard or not. If not, it returned EOPNOTSUPP. After the
commit, __blkdev_issue_discard() no longer checks it. It always returns
zero and sets NULL to the given bio pointer. This NULL pointer triggers
f2fs_bug_on() in __submit_discard_cmd(). The BUG is recreated with the
commands below at the umount step, where /dev/nullb0 is a zoned null_blk
with 5GB total size, 128MB zone size and 10 conventional zones.
$ mkfs.f2fs -f -m /dev/nullb0
$ mount /dev/nullb0 /mnt
$ for ((i=0;i<5;i++)); do dd if=/dev/zero of=/mnt/test bs=65536 count=1600 conv=fsync; done
$ umount /mnt
To fix the BUG, avoid the inappropriate __blkdev_issue_discard() call.
When discard is requested for conventional zones, check if the device
supports discard or not. If not, return EOPNOTSUPP.
References
In the Linux kernel, the following vulnerability has been resolved:
f2fs: check discard support for conventional zones
As the helper function f2fs_bdev_support_discard() shows, f2fs checks if
the target block devices support discard by calling
bdev_max_discard_sectors() and bdev_is_zoned(). This check works well
for most cases, but it does not work for conventional zones on zoned
block devices. F2fs assumes that zoned block devices support discard,
and calls __submit_discard_cmd(). When __submit_discard_cmd() is called
for sequential write required zones, it works fine since
__submit_discard_cmd() issues zone reset commands instead of discard
commands. However, when __submit_discard_cmd() is called for
conventional zones, __blkdev_issue_discard() is called even when the
devices do not support discard.
The inappropriate __blkdev_issue_discard() call was not a problem before
the commit 30f1e7241422 ("block: move discard checks into the ioctl
handler") because __blkdev_issue_discard() checked if the target devices
support discard or not. If not, it returned EOPNOTSUPP. After the
commit, __blkdev_issue_discard() no longer checks it. It always returns
zero and sets NULL to the given bio pointer. This NULL pointer triggers
f2fs_bug_on() in __submit_discard_cmd(). The BUG is recreated with the
commands below at the umount step, where /dev/nullb0 is a zoned null_blk
with 5GB total size, 128MB zone size and 10 conventional zones.
$ mkfs.f2fs -f -m /dev/nullb0
$ mount /dev/nullb0 /mnt
$ for ((i=0;i<5;i++)); do dd if=/dev/zero of=/mnt/test bs=65536 count=1600 conv=fsync; done
$ umount /mnt
To fix the BUG, avoid the inappropriate __blkdev_issue_discard() call.
When discard is requested for conventional zones, check if the device
supports discard or not. If not, return EOPNOTSUPP.
References