From: Marta Lewandowska <mlewando@redhat.com>
To: The development of GNU GRUB <grub-devel@gnu.org>,
Jon DeVree <nuxi@vault24.org>
Subject: Re: [PATCH v4] fs/xfs: Fix XFS directory extent parsing
Date: Wed, 1 Nov 2023 09:11:24 +0100 [thread overview]
Message-ID: <3dcf135e-d407-cbdc-3538-879ee20cc884@redhat.com> (raw)
In-Reply-To: <20231018030347.36174-1-nuxi@vault24.org>
[-- Attachment #1.1: Type: text/plain, Size: 6820 bytes --]
On 10/18/23 05:03, Jon DeVree wrote:
> The XFS directory entry parsing code has never been completely correct
> for extent based directories. The parser correctly handles the case
> where the directory is contained in a single extent, but then mistakenly
> assumes the data blocks for the multiple extent case are each identical
> to the single extent case. The difference in the format of the data
> blocks between the two cases is tiny enough that its gone unnoticed for
> a very long time.
>
> A recent change introduced some additional bounds checking into the XFS
> parser. Like GRUB's existing parser, it is correct for the single extent
> case but incorrect for the multiple extent case. When parsing a
> directory with multiple extents, this new bounds checking is sometimes
> (but not always) tripped and triggers an "invalid XFS diretory entry"
> error. This probably would have continued to go unnoticed but the
> /boot/grub/<arch> directory is large enough that it often has multiple
> extents.
>
> The difference between the two cases is that when there are multiple
> extents, the data blocks do not contain a trailer nor do they contain
> any leaf information. That information is stored in a separate set of
> extents dedicated to just the leaf information. These extents come after
> the directory entry extents and are not included in the inode size. So
> the existing parser already ignores the leaf extents.
>
> The only reason to read the trailer/leaf information at all is so that
> the parser can avoid misinterpreting that data as directory entries. So
> this updates the parser as follows:
>
> For the single extent case the parser doesn't change much:
> 1. Read the size of the leaf information from the trailer
> 2. Set the end pointer for the parser to the start of the leaf
> information. (The previous bounds checking set the end pointer to the
> start of the trailer, so this is actually a small improvement.)
> 3. Set the entries variable to the expected number of directory entries.
>
> For the multiple extent case:
> 1. Set the end pointer to the end of the block.
> 2. Do not set up the entries variable. Figuring out how many entries are
> in each individual block is complex and does not seem worth it when
> it appears to be safe to just iterate over the entire block.
>
> The bounds check itself was also dependent upon the faulty XFS parser
> because it accidentally used "filename + length - 1". Presumably this
> was able to pass the fuzzer because in the old parser there was always 8
> bytes of slack space between the tail pointer and the actual end of the
> block. Since this is no longer the case the bounds check needs to be
> updated to "filename + length + 1" in order to prevent a regressionn in
> the handling of corrupt fliesystems.
>
> Notes:
> * When there is only one extent there will only ever be one block. If
> more than one block is required then XFS will always switch to holding
> leaf information in a separate extent.
> * B-tree based directories seems to be parsed properly by the same code
> that handles multiple extents. This is unlikely to ever occur within
> /boot though because its only used when there are an extremely large
> number of directory entries.
>
> Fixes: ef7850c75 (fs/xfs: Fix issues found while fuzzing the XFS filesystem)
> Fixes: b2499b29c (Adds support for the XFS filesystem.)
> Fixes:https://savannah.gnu.org/bugs/?64376
>
> Signed-off-by: Jon DeVree<nuxi@vault24.org>
> ---
> grub-core/fs/xfs.c | 51 +++++++++++++++++++++++++++++++++-------------
> 1 file changed, 37 insertions(+), 14 deletions(-)
>
> diff --git a/grub-core/fs/xfs.c b/grub-core/fs/xfs.c
> index b91cd32b4..acdfb1a7b 100644
> --- a/grub-core/fs/xfs.c
> +++ b/grub-core/fs/xfs.c
> @@ -223,6 +223,12 @@ struct grub_xfs_inode
> /* Size of struct grub_xfs_inode v2, up to unused4 member included. */
> #define XFS_V2_INODE_SIZE (XFS_V3_INODE_SIZE - 76)
>
> +struct grub_xfs_dir_leaf_entry
> +{
> + grub_uint32_t hashval;
> + grub_uint32_t address;
> +} GRUB_PACKED;
> +
> struct grub_xfs_dirblock_tail
> {
> grub_uint32_t leaf_count;
> @@ -877,9 +883,8 @@ grub_xfs_iterate_dir (grub_fshelp_node_t dir,
> {
> struct grub_xfs_dir2_entry *direntry =
> grub_xfs_first_de(dir->data, dirblock);
> - int entries;
> - struct grub_xfs_dirblock_tail *tail =
> - grub_xfs_dir_tail(dir->data, dirblock);
> + int entries = -1;
> + char *end = dirblock + dirblk_size;
>
> numread = grub_xfs_read_file (dir, 0, 0,
> blk << dirblk_log2,
> @@ -890,14 +895,27 @@ grub_xfs_iterate_dir (grub_fshelp_node_t dir,
> return 0;
> }
>
> - entries = (grub_be_to_cpu32 (tail->leaf_count)
> - - grub_be_to_cpu32 (tail->leaf_stale));
> + /* leaf and tail information are only in the data block if the number
> + * of extents is 1 */
> + if (dir->inode.nextents == grub_cpu_to_be32_compile_time (1))
> + {
> + struct grub_xfs_dirblock_tail *tail =
> + grub_xfs_dir_tail(dir->data, dirblock);
> + end = (char *)tail;
>
> - if (!entries)
> - continue;
> + /* subtract the space used by leaf nodes */
> + end -= grub_be_to_cpu32 (tail->leaf_count) *
> + sizeof (struct grub_xfs_dir_leaf_entry);
> +
> + entries = (grub_be_to_cpu32 (tail->leaf_count)
> + - grub_be_to_cpu32 (tail->leaf_stale));
> +
> + if (!entries)
> + continue;
> + }
>
> /* Iterate over all entries within this block. */
> - while ((char *)direntry < (char *)tail)
> + while ((char *)direntry < (char *)end)
> {
> grub_uint8_t *freetag;
> char *filename;
> @@ -917,7 +935,7 @@ grub_xfs_iterate_dir (grub_fshelp_node_t dir,
> }
>
> filename = (char *)(direntry + 1);
> - if (filename + direntry->len - 1 > (char *) tail)
> + if (filename + direntry->len + 1 > (char *) end)
> return grub_error (GRUB_ERR_BAD_FS, "invalid XFS directory entry");
>
> /* The byte after the filename is for the filetype, padding, or
> @@ -931,11 +949,16 @@ grub_xfs_iterate_dir (grub_fshelp_node_t dir,
> return 1;
> }
>
> - /* Check if last direntry in this block is
> - reached. */
> - entries--;
> - if (!entries)
> - break;
> + /* the expected number of directory entries is only tracked for the
> + * single extent case */
> + if (dir->inode.nextents == grub_cpu_to_be32_compile_time (1))
> + {
> + /* Check if last direntry in this block is
> + reached. */
> + entries--;
> + if (!entries)
> + break;
> + }
>
> /* Select the next directory entry. */
> direntry = grub_xfs_next_de(dir->data, direntry);
Tested-by: Marta Lewandowska <mlewando@redhat.com>
[-- Attachment #1.2: Type: text/html, Size: 7203 bytes --]
[-- Attachment #2: Type: text/plain, Size: 141 bytes --]
_______________________________________________
Grub-devel mailing list
Grub-devel@gnu.org
https://lists.gnu.org/mailman/listinfo/grub-devel
next prev parent reply other threads:[~2023-11-01 8:11 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-10-18 3:03 [PATCH v4] fs/xfs: Fix XFS directory extent parsing Jon DeVree
2023-10-18 16:23 ` Daniel Kiper
2023-10-22 16:11 ` Sebastian Andrzej Siewior
2023-10-23 12:55 ` Daniel Kiper
2023-10-24 6:46 ` Marta Lewandowska
2023-10-25 12:04 ` Daniel Kiper
2023-11-01 8:11 ` Marta Lewandowska [this message]
2023-11-05 7:28 ` Philip Hands
2023-11-06 8:13 ` Philip Hands
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=3dcf135e-d407-cbdc-3538-879ee20cc884@redhat.com \
--to=mlewando@redhat.com \
--cc=grub-devel@gnu.org \
--cc=nuxi@vault24.org \
/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;
as well as URLs for NNTP newsgroup(s).