mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Qu Wenruo <wqu@suse.com>
To: Cheng Yongkang <teel4res@gmail.com>, David Sterba <dsterba@suse.com>
Cc: Chris Mason <mason@kernel.org>,
	linux-btrfs@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v4] btrfs: tree-checker: reject file extents with oversized disk_num_bytes
Date: Tue, 17 Nov 2026 18:17:46 +1030	[thread overview]
Message-ID: <8857bbb8-11e2-44b7-a749-4396b88f1977@suse.com> (raw)
In-Reply-To: <20261117064613.771336-1-teel4res@gmail.com>



在 2026/11/17 17:16, Cheng Yongkang 写道:
> check_extent_data_item() only checks that disk_num_bytes of a regular or
> preallocated file extent is sector aligned. A crafted image can have a
> compressed extent far larger than anything btrfs writes, and reading it
> goes wrong in different ways:
> 
> - With disk_num_bytes == 4GiB, btrfs_submit_compressed_read() truncates
>    it to a zero compressed_len and submits an empty bio:
> 
>      assertion failed: IS_ALIGNED(logical, blocksize) &&
>      IS_ALIGNED(length, blocksize) && length != 0, in fs/btrfs/bio.c:895
> 
> - With a 4MiB zlib extent whose num_bytes and ram_bytes are above 4GiB,
>    bvec_offset + bv_len wraps in btrfs_decompress_buf2page() and memcpy()
>    reads past the zlib workspace buffer:
> 
>      BUG: KASAN: slab-out-of-bounds in btrfs_decompress_buf2page+0x3e0/0x650

With all those enhance checks, I do not think the two reports add any 
value now.

The patch is really hardening the tree-checker, and that's all what we 
need to know.
> 
> Compression and encoded writes never create a compressed extent larger
> than BTRFS_MAX_COMPRESSED on disk or BTRFS_MAX_UNCOMPRESSED in memory,
> and its offset + num_bytes always fits in ram_bytes. Uncompressed
> extents can be larger than 256MiB, for example preallocated extents
> from kernels before v3.9 or extents from direct I/O writes, but they
> should never get close to U32_MAX.
> 
> Reject a file extent whose disk_num_bytes is above BTRFS_MAX_COMPRESSED
> if it is compressed, or above U32_MAX otherwise. For an extent with data
> on disk, also reject:
> 
> - a zero disk_num_bytes, which makes btrfs_submit_compressed_read()
>    submit an empty bio the same way;
> - a ram_bytes above BTRFS_MAX_UNCOMPRESSED if it is compressed, or above
>    U32_MAX otherwise;
> - an offset + num_bytes beyond ram_bytes if it is compressed, as the
>    compressed read path takes the decompressed range from offset and
>    num_bytes.
> 
> Holes are skipped, since without NO_HOLES a hole extent has ram_bytes
> set to the size of the hole.
> 
> Suggested-by: Qu Wenruo <wqu@suse.com>
> Signed-off-by: Cheng Yongkang <teel4res@gmail.com>
> ---
> Changes in v4:
> - Also check ram_bytes of uncompressed extents against U32_MAX, with
>    max_ram_bytes picked the same way as max_disk_num_bytes.
> - Reject a zero disk_num_bytes for extents with data on disk, and an
>    offset + num_bytes beyond ram_bytes for compressed extents, as
>    Sashiko pointed out on v2 and v3.
> - Skip holes for the new checks, as their ram_bytes is the hole size
>    without NO_HOLES.
> - Correct the commit message: direct I/O writes can also create
>    uncompressed extents larger than 256MiB.
> - Rebase onto misc-next.
> 
> Changes in v3:
> - Relax the uncompressed disk_num_bytes limit from 256MiB to U32_MAX,
>    as preallocated extents created before v3.9 can be larger.
> - Also check ram_bytes of compressed extents against
>    BTRFS_MAX_UNCOMPRESSED.
> 
> Changes in v2:
> - Use SZ_256M directly instead of adding a new macro, and leave
>    inode.c alone.
> - Shorten the comment.
> 
>   fs/btrfs/tree-checker.c | 55 +++++++++++++++++++++++++++++++++++++++++
>   1 file changed, 55 insertions(+)
> 
> diff --git a/fs/btrfs/tree-checker.c b/fs/btrfs/tree-checker.c
> index f021ac14babf..e7e0bb6be76f 100644
> --- a/fs/btrfs/tree-checker.c
> +++ b/fs/btrfs/tree-checker.c
> @@ -244,7 +244,10 @@ static int check_extent_data_item(struct extent_buffer *leaf,
>   	struct btrfs_file_extent_item *fi;
>   	u32 sectorsize = fs_info->sectorsize;
>   	u32 item_size = btrfs_item_size(leaf, slot);
> +	u32 max_disk_num_bytes;
> +	u32 max_ram_bytes;
>   	u64 extent_end;
> +	u64 ram_end;
>   	u8 fi_type;
>   
>   	if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
> @@ -377,6 +380,58 @@ static int check_extent_data_item(struct extent_buffer *leaf,
>   		     CHECK_FE_ALIGNED(leaf, slot, fi, num_bytes, sectorsize)))
>   		return -EUCLEAN;
>   
> +	/*
> +	 * Compressed extents are at most 128K both on disk and uncompressed.
> +	 * Uncompressed extents should never get close to U32_MAX.
> +	 */
> +	if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
> +		max_disk_num_bytes = BTRFS_MAX_COMPRESSED;
> +		max_ram_bytes = BTRFS_MAX_UNCOMPRESSED;
> +	} else {
> +		max_disk_num_bytes = U32_MAX;
> +		max_ram_bytes = U32_MAX;
> +	}
> +	if (unlikely(btrfs_file_extent_disk_num_bytes(leaf, fi) > max_disk_num_bytes)) {
> +		file_extent_err(leaf, slot,
> +		"invalid disk_num_bytes for file extent, have %llu expect <= %u",
> +				btrfs_file_extent_disk_num_bytes(leaf, fi),
> +				max_disk_num_bytes);
> +		return -EUCLEAN;
> +	}
> +
> +	/*
> +	 * The remaining checks only apply to extents with data on disk. Holes
> +	 * are skipped, as without NO_HOLES their ram_bytes is the hole size.
> +	 */
> +	if (btrfs_file_extent_disk_bytenr(leaf, fi)) {

Sashiko mentioned a situation where a dedicated hole can by-pass all 
those checks:

https://sashiko.dev/#/patchset/20261117064613.771336-1-teel4res%40gmail.com

I think a dedicated explicit hole check would be safer, rejecting 
obvious incorrect things like an explicit hole but with compression set.
> +		if (unlikely(btrfs_file_extent_disk_num_bytes(leaf, fi) == 0)) {
> +			file_extent_err(leaf, slot,
> +		"invalid zero disk_num_bytes for file extent with disk_bytenr %llu",
> +					btrfs_file_extent_disk_bytenr(leaf, fi));
> +			return -EUCLEAN;
> +		}
> +		if (unlikely(btrfs_file_extent_ram_bytes(leaf, fi) > max_ram_bytes)) {
> +			file_extent_err(leaf, slot,
> +		"invalid ram_bytes for file extent, have %llu expect <= %u",
> +					btrfs_file_extent_ram_bytes(leaf, fi),
> +					max_ram_bytes);
> +			return -EUCLEAN;
> +		}
> +		/* Reads of compressed extents use offset and num_bytes in ram_bytes. */
> +		if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE &&
> +		    unlikely(check_add_overflow(btrfs_file_extent_offset(leaf, fi),
> +						btrfs_file_extent_num_bytes(leaf, fi),
> +						&ram_end) ||
> +			     ram_end > btrfs_file_extent_ram_bytes(leaf, fi))) {

This overflow check should apply to both compressed and uncompressed 
extents.
This is not limited to compressed extents AFAIK.

> +			file_extent_err(leaf, slot,
> +	"invalid offset + num_bytes for compressed file extent, have %llu + %llu expect <= %llu",
> +					btrfs_file_extent_offset(leaf, fi),
> +					btrfs_file_extent_num_bytes(leaf, fi),
> +					btrfs_file_extent_ram_bytes(leaf, fi));
> +			return -EUCLEAN;
> +		}
> +	}
> +
>   	/* Catch extent end overflow */
>   	if (unlikely(check_add_overflow(btrfs_file_extent_num_bytes(leaf, fi),
>   					key->offset, &extent_end))) {
> 
> base-commit: d45cbbab23d3ca231771e4bff996a6a8118162cf


      reply	other threads:[~2026-11-17  7:47 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-17  6:46 Cheng Yongkang
2026-11-17  7:47 ` Qu Wenruo [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=8857bbb8-11e2-44b7-a749-4396b88f1977@suse.com \
    --to=wqu@suse.com \
    --cc=dsterba@suse.com \
    --cc=linux-btrfs@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mason@kernel.org \
    --cc=teel4res@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®