Skip to contentSkip to conversation

ufs: reject oversized cylinder group metadata

PATCH In review David Lee 2 participants 2 messages Jul 17 – Aug 1 VFS

lore
Share & tools
lore

David Lee

author
#0 lore raw

PATCH 1/1 ufs: reject oversized cylinder group metadata

The on-disk cylinder group size becomes the fragment count used by ufs_read_cylinder(). A size larger than one filesystem block makes that count exceed UFS_MAXFRAG and drives buffer-head stores beyond the fixed eight-entry array in struct ufs_cg_private_info.

Reject zero-sized cylinder groups and sizes larger than the validated filesystem block size while reading the superblock. Since a filesystem block contains at most UFS_MAXFRAG fragments, every later cylinder cache count is then bounded by its buffer-head array.

Fixes 1da177e4c3f4 ("Linux-2.6.12-rc2")
Assisted-by Codex:gpt-5.5
---
 fs/ufs/super.c | 5 +++++
 1 file changed, 5 insertions(+)
fs/ufs/super.c +5 -0
index c4831a8b9b3f..fa2effdec7b0 100644
@@ -1140,6 +1140,11 @@ static int ufs_fill_super(struct super_block *sb, struct fs_context *fc)
 
 	uspi->s_cssize = fs32_to_cpu(sb, usb1->fs_cssize);
 	uspi->s_cgsize = fs32_to_cpu(sb, usb1->fs_cgsize);
+	if (uspi->s_cgsize == 0 || uspi->s_cgsize > uspi->s_bsize) {
+		pr_err("%s(): invalid cylinder group size %u\n",
+		       __func__, uspi->s_cgsize);
+		goto failed;
+	}
 	uspi->s_ntrak = fs32_to_cpu(sb, usb1->fs_ntrak);
 	uspi->s_nsect = fs32_to_cpu(sb, usb1->fs_nsect);
 	uspi->s_spc = fs32_to_cpu(sb, usb1->fs_spc);
Signature · 1 line
2.43.0
Reviewed-by on v1 1/1
#1 lore raw

I traced the count assignment this patch protects. ufs_read_cylinder() in fs/ufs/cylinder.c sets UCPI_UBH(ucpi)->count directly from s_cgsize >> s_blocksize_bits and then stores into bh[i] in a loop, so it does not go through the UFS_MAXFRAG clamp that _ubh_bread_() and ubh_bread_uspi() apply in fs/ufs/util.c. An oversized fs_cgsize therefore does drive buffer-head stores past the fixed eight entry array in struct ufs_cg_private_info, as the changelog says.

The new bound is sufficient. ufs_fill_super() already rejects a filesystem whose s_bsize over s_fsize ratio is larger than eight, so once s_cgsize is held at or below s_bsize the derived count stays at or under UFS_MAXFRAG and the bh[] loop is in range. Rejecting a zero sized cylinder group at the same point is a sensible guard as well.