Re: [PATCH] btrfs: do not check support for an invalid sectorsize

[Qu Wenruo]

在 2026/11/3 10:00, Siddharth Mishra 写道:
> From: Siddharth Mishra <brightprogrammer@xxxxxxx>
>
> [BUG]
> Mounting an image whose super block has an invalid sectorsize (for
> example 4097) on a kernel with CONFIG_BTRFS_ASSERT=y triggers a BUG:
>
>    BTRFS error (device loop0): invalid sectorsize 4097
>    assertion failed: is_power_of_2(blocksize) &&
>    blocksize >= BTRFS_MIN_BLOCKSIZE && blocksize <= BTRFS_MAX_BLOCKSIZE,
>    in fs/btrfs/fs.c:137
>    kernel BUG at fs/btrfs/fs.c:137!
>    Oops: invalid opcode: 0000 [#1] SMP KASAN NOPTI
>    RIP: 0010:btrfs_supported_blocksize.cold (fs/btrfs/fs.c:137)
>    Call Trace:
>     btrfs_validate_super (fs/btrfs/disk-io.c:2525)
>     open_ctree (fs/btrfs/disk-io.c:3507)
>     btrfs_get_tree.cold (fs/btrfs/super.c:2133)
>     vfs_get_tree (fs/super.c:1943)
>     fc_mount (fs/namespace.c:1282)
>     path_mount (fs/namespace.c:4337)
>     __x64_sys_mount (fs/namespace.c:4542)
>
> [CAUSE]
> btrfs_validate_super() rejects the sectorsize because it is not a power
> of 2, but it does not stop there: it goes on to report all invalid
> fields.  The next check passes the same sectorsize to
> btrfs_supported_blocksize(), which requires a validated value and
> asserts it since commit d728f2e5f8a0 ("btrfs: simplify support block
> size check").
>
> [FIX]
> Only check whether a sectorsize is supported after it passed the
> validity check.  An invalid sectorsize is still rejected with -EINVAL,
> and the misleading second message "sectorsize 4097 not yet supported"
> is no longer printed.
>
> This bug was found by fuzzing. The cause was analysed and the fix was
> written with the help of an AI tool. Tested on btrfs for-next
> (d6f76ca0a70d) and on v7.4-rc1, with KASAN and CONFIG_BTRFS_ASSERT=y:
> without the fix, mounting the image hits the BUG above; with the fix,
> the mount fails with -EINVAL and only "invalid sectorsize 4097" is
> printed.
> A crafted image (22 KiB, xz) and a C reproducer are available on request.
>
> Fixes: d728f2e5f8a0 ("btrfs: simplify support block size check")
> Cc: stable@xxxxxxxxxxxxxxx # 6.18+
> Signed-off-by: Siddharth Mishra <brightprogrammer@xxxxxxx>
> ---
>   fs/btrfs/disk-io.c | 4 +---
>   1 file changed, 1 insertion(+), 3 deletions(-)
>
> diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
> index 8e8fb942d..b348591f8 100644
> --- a/fs/btrfs/disk-io.c
> +++ b/fs/btrfs/disk-io.c
> @@ -2520,9 +2520,7 @@ int btrfs_validate_super(const struct btrfs_fs_info *fs_info,
>   		     sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE)) {
>   		btrfs_err(fs_info, "invalid sectorsize %u", sectorsize);
>   		ret = -EINVAL;
> -	}
> -
> -	if (unlikely(!btrfs_supported_blocksize(sectorsize))) {
> +	} else if (unlikely(!btrfs_supported_blocksize(sectorsize))) {

If you want to change this, you should touch all callsites.
Not only this one, but every unlikely() branch to return immediately.

I do not like the ad-hoc fix just be cause you can trigger an ASSERT().
>   		btrfs_err(fs_info,
>   			"sectorsize %u not yet supported for page size %lu",
>   			sectorsize, PAGE_SIZE);
>
> base-commit: d6f76ca0a70d0c814731394a625ff0dc1feb390e