On Mon, Feb 12, 2024 at 03:46:15PM +1030, Qu Wenruo wrote:
[BUG] There is a bug report that with zoned device and sectorsize is smaller than page size (aka, subpage), btrfs would crash with a very basic workload:
# getconfig PAGESIZE 16384 # mkfs.btrfs -f $dev -s 4k # mount $dev $mnt # $fsstress -w -n 8 -s 1707820327 -v -d $mnt # umount $mnt
The crash would look like this (with CONFIG_BTRFS_ASSERT enabled):
assertion failed: block_start != EXTENT_MAP_HOLE, in fs/btrfs/extent_io.c:1384 ------------[ cut here ]------------ kernel BUG at fs/btrfs/extent_io.c:1384! CPU: 0 PID: 872 Comm: kworker/u9:2 Tainted: G OE 6.8.0-rc3-custom+ #7 Hardware name: QEMU KVM Virtual Machine, BIOS edk2-20231122-12.fc39 11/22/2023 Workqueue: writeback wb_workfn (flush-btrfs-8) pc : __extent_writepage_io+0x404/0x460 [btrfs] lr : __extent_writepage_io+0x404/0x460 [btrfs] Call trace: __extent_writepage_io+0x404/0x460 [btrfs] extent_write_locked_range+0x16c/0x460 [btrfs] run_delalloc_cow+0x88/0x118 [btrfs] btrfs_run_delalloc_range+0x128/0x228 [btrfs] writepage_delalloc+0xb8/0x178 [btrfs] __extent_writepage+0xc8/0x3a0 [btrfs] extent_write_cache_pages+0x1cc/0x460 [btrfs] extent_writepages+0x8c/0x120 [btrfs] btrfs_writepages+0x18/0x30 [btrfs] do_writepages+0x94/0x1f8 __writeback_single_inode+0x4c/0x388 writeback_sb_inodes+0x208/0x4b0 wb_writeback+0x118/0x3c0 wb_do_writeback+0xbc/0x388 wb_workfn+0x80/0x240 process_one_work+0x154/0x3c8 worker_thread+0x2bc/0x3e0 kthread+0xf4/0x108 ret_from_fork+0x10/0x20 Code: 9102c021 90000be0 91378000 9402bf53 (d4210000) ---[ end trace 0000000000000000 ]---
[CAUSE] There are several factors causing the problem:
__extent_writepage_io() requires all dirty ranges to have delalloc executed This can be solved by adding @start and @len parameter to only submit IO for a subset of the page, and update several involved helpers to do subpage checks.
So this is not a big deal.
Subpage only accepts for full page aligned ranges for extent_write_locked_range() For zoned device, regular COW is switched to utilize extent_write_locked_range() to submit the IO.
But the caller, run_delalloc_cow() can be called to run on a subpage range, e.g.
0 4K 8K 12K 16K |/////| |/////|
Where |///| is the dirtied range.
In that case, btrfs_run_delalloc_range() would call run_delalloc_cow(), which would call extent_write_locked_range() for [0, 4K), and unlock the whole [0, 16K) page.
But btrfs_run_delalloc_range() would again be called for range [8K, 12K), as there are still dirty range left. In that case, since the whole page is already unlocked by previous iteration, and would cause different ASSERT()s inside extent_write_locked_range().
That's also why compression for subpage cases require fully page aligned range.
[WORKAROUND] A proper fix requires some big changes to delalloc workload, to allow extent_write_locked_range() to handle multiple different entries with the same @locked_page.
So for now, disable read-write support for subpage zoned btrfs.
The problem can only be solved if subpage btrfs can handle subpage compression, which need quite some work on the delalloc procedure for the @locked_page handling.
Reported-by: HAN Yuwei hrx@bupt.moe Link: https://lore.kernel.org/all/1ACD2E3643008A17+da260584-2c7f-432a-9e22-9d390aa... CC: stable@vger.kernel.org # 5.10+ Signed-off-by: Qu Wenruo wqu@suse.com
fs/btrfs/disk-io.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index c3ab268533ca..85cd23aebdd6 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c @@ -3193,7 +3193,8 @@ int btrfs_check_features(struct btrfs_fs_info *fs_info, bool is_rw_mount) * part of @locked_page. * That's also why compression for subpage only work for page aligned ranges. */
- if (fs_info->sectorsize < PAGE_SIZE && btrfs_is_zoned(fs_info) && is_rw_mount) {
- if (fs_info->sectorsize < PAGE_SIZE &&
btrfs_fs_incompat(fs_info, ZONED) && is_rw_mount) {
This does not match any code I see in usual branches and it also does not apply so it can't be picked as a fix.