On Mon, Feb 13, 2023 at 08:37:59AM +0800, Qu Wenruo wrote:
> [BUG]
> There is a bug report that btrfs driver caused hang during file read:
>
> This breaks btrfs on the HiFive Unmatched.
>
> => pci enum
> PCIE-0: Link up (Gen1-x8, Bus0)
> => nvme scan
> => load nvme 0:2 0x8c000000 /boot/dtb/sifive/hifive-unmatched-a00.dtb
> [hangs]
>
> [CAUSE]
> The reporter provided some debug output:
>
> read_extent_data: cur=615817216, orig_len=16384, cur_len=16384
> read_extent_data: btrfs_map_block: cur_len=479944704; ret=0
> read_extent_data: ret=0
> read_extent_data: cur=615833600, orig_len=4096, cur_len=4096
> read_extent_data: btrfs_map_block: cur_len=479928320; ret=0
>
> Note the second and the last line, the @cur_len is 450+MiB, which is
> almost a chunk size.
>
> And inside __btrfs_map_block(), we limits the returned value to stripe
> length, but that's depending on the chunk type:
>
> if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
> BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4 |
> BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
> BTRFS_BLOCK_GROUP_RAID10 |
> BTRFS_BLOCK_GROUP_DUP)) {
> /* we limit the length of each bio to what fits in a stripe */
> *length = min_t(u64, ce->size - offset,
> map->stripe_len - stripe_offset);
> } else {
> *length = ce->size - offset;
> }
>
> This means, if the chunk is SINGLE profile, then we don't limit the
> returned length at all, and even for other profiles, we can still return
> a length much larger than the requested one.
>
> [FIX]
> Properly clamp the returned length, preventing it from returning a much
> larger range than expected.
>
> Reported-by: Andreas Schwab <[email protected]>
> Signed-off-by: Qu Wenruo <[email protected]>Applied to u-boot/master, thanks! -- Tom
signature.asc
Description: PGP signature
