Skip to contentSkip to conversation

isofs: hardening for crafted CE and NFS-handle paths

PATCH In review Michael Bommarito 2 participants 4 messages Apr 19 – Apr 20

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover isofs: hardening for crafted CE and NFS-handle paths

Two small defensive bounds checks for the ISO 9660 filesystem, one in Rock Ridge CE continuation handling and one in the NFS export path. Both surfaced while looking for missing bounds checks adjacent to recently-landed isofs fixes (0405d4b63d08, f54e18f1b831). Neither is a memory-safety bug on its own; the existing sb_bread() / isofs_iget() paths handle out-of-range blocks cleanly. These patches reject the malformed input at the earliest point it is known to be out of range.

1/2: rock_continue() reads rs->cont_extent from the Rock Ridge CE record and calls sb_bread() on it without validating the block number against ISOFS_SB(sb)->s_nzones. commit e595447e177b (2005) added the cont_offset and cont_size rejection but left the extent number unchecked; commit f54e18f1b831 ("isofs: Fix infinite looping over CE entries") later capped the CE chain at RR_MAX_CE_ENTRIES = 32 but again did not address the block number. The reachable attacker model is a crafted ISO auto-mounted via udisks2 / polkit on a desktop session; sb_bread() on an out-of-range block returns NULL cleanly, so there is no memory-safety issue, and the CE buffer is parsed as Rock Ridge records rather than returned verbatim via readlink().

2/2: isofs_fh_to_dentry() and isofs_fh_to_parent() pass attacker-controlled block numbers from the NFS file handle to isofs_export_iget(), which rejects block == 0 but not out-of-range block numbers. commit 0405d4b63d08 ("isofs: Prevent the use of too small fid", CVE-2025-37780) added fh_len checks but left the block number itself unchecked. An authenticated NFS peer with a crafted fid can drive reads of adjacent-partition data on the same block device into iso_inode_info fields reaching the client as dentry metadata. Deployment surface is narrow (isofs-over-NFS); impact is primarily EIO / ESTALE with a weak info-leak channel.

Both patches are one-line (or close to it) additions; the existing out-of-range-block check in isofs_iget() / sb_bread() handles the read-side cleanly, so these are strictly belt-and-suspenders rejection at the earliest point we know the input is out of range.

Build-tested W=1 against 7.0-rc7 with CONFIG_ISO9660_FS=y, CONFIG_JOLIET=y, CONFIG_ZISOFS=y.

Michael Bommarito (2):
isofs: validate Rock Ridge CE continuation extent against volume size
isofs: validate block number from NFS file handle in isofs_export_iget

 fs/isofs/export.c | 2 +-
 fs/isofs/rock.c   | 9 +++++++++
 2 files changed, 10 insertions(+), 1 deletion(-)
Signature · 1 line
2.53.0
#1 lore raw

PATCH 1/2 isofs: validate Rock Ridge CE continuation extent against volume size

rock_continue() reads rs->cont_extent verbatim from the Rock Ridge CE record and passes it to sb_bread() without checking that the block number is within the mounted ISO 9660 volume. commit e595447e177b ("[PATCH] rock.c: handle corrupted directories") added cont_offset and cont_size rejection for the CE continuation but did not validate the extent block number itself. commit f54e18f1b831 ("isofs: Fix infinite looping over CE entries") later capped the CE chain length at RR_MAX_CE_ENTRIES = 32 but again left the block number unchecked.

With a crafted ISO mounted via udisks2 (desktop optical auto-mount) or via CAP_SYS_ADMIN mount, rs->cont_extent can therefore point at an out-of-range block or at blocks belonging to an adjacent filesystem on the same block device. sb_bread() on an out-of-range block returns NULL cleanly via the block layer EIO path, so there is no memory-safety violation. For in-range reads of adjacent- filesystem data, the CE buffer is parsed as Rock Ridge records and only the text of SL sub-records reaches userspace through readlink(), which makes the info-leak channel narrow and difficult to exploit; still, rejecting the malformed CE outright matches the rejection shape already present in the same function for cont_offset and cont_size.

Add an ISOFS_SB(sb)->s_nzones bounds check to rock_continue() next to the existing offset/size rejection, printing the same corrupted-directory-entry notice.

Fixes f54e18f1b831 ("isofs: Fix infinite looping over CE entries")
Assisted-by Claude:claude-opus-4-7
---
 fs/isofs/rock.c | 9 +++++++++
 1 file changed, 9 insertions(+)
fs/isofs/rock.c +9 -0
index 6fe6dbd0c740..1232fab59a4e 100644
@@ -101,6 +101,15 @@ static int rock_continue(struct rock_state *rs)
 		goto out;
 	}
 
+	if ((unsigned)rs->cont_extent >= ISOFS_SB(rs->inode->i_sb)->s_nzones) {
+		printk(KERN_NOTICE "rock: corrupted directory entry. "
+			"extent=%u out of volume (nzones=%lu)\n",
+			(unsigned)rs->cont_extent,
+			ISOFS_SB(rs->inode->i_sb)->s_nzones);
+		ret = -EIO;
+		goto out;
+	}
+
 	if (rs->cont_extent) {
 		struct buffer_head *bh;
 
Signature · 1 line
2.53.0
#2 lore raw

PATCH 2/2 isofs: validate block number from NFS file handle in isofs_export_iget

In reply to Michael Bommarito

isofs_fh_to_dentry() and isofs_fh_to_parent() pass an attacker- controlled block number (ifid->block or ifid->parent_block) from the NFS file handle to isofs_export_iget(), which only rejects block == 0 before calling isofs_iget() and ultimately sb_bread(). A crafted file handle with fh_len sufficient to pass the check added by commit 0405d4b63d08 ("isofs: Prevent the use of too small fid") can still drive the server to read any in-range block on the backing device as if it were an iso_directory_record. That earlier fix was assigned CVE-2025-37780.

sb_bread() on an out-of-range block returns NULL cleanly via the EIO path, so there is no memory-safety violation. For in-range reads of adjacent-partition data on the same block device, the unrelated bytes end up in iso_inode_info fields that reach the NFS client as dentry metadata. The deployment surface (isofs exported over NFS from loop-mounted images) is narrow and requires an authenticated NFS peer, but the malformed-file-handle class is reportable as hardening next to the existing CVE-2025-37780 fix.

Reject block >= ISOFS_SB(sb)->s_nzones in isofs_export_iget() so the check covers both isofs_fh_to_dentry() and isofs_fh_to_parent() call sites with a single line.

Fixes 0405d4b63d08 ("isofs: Prevent the use of too small fid")
Assisted-by Claude:claude-opus-4-7
---
 fs/isofs/export.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/isofs/export.c +1 -1
index 421d247fae52..78f80c1a5c54 100644
@@ -24,7 +24,7 @@ isofs_export_iget(struct super_block *sb,
 {
 	struct inode *inode;
 
-	if (block == 0)
+	if (block == 0 || block >= ISOFS_SB(sb)->s_nzones)
 		return ERR_PTR(-ESTALE);
 	inode = isofs_iget(sb, block, offset);
 	if (IS_ERR(inode))
Signature · 1 line
2.53.0
maintainer
#3 lore raw
In reply to Michael Bommarito
Quoted 32 lines · Michael Bommarito

Two small defensive bounds checks for the ISO 9660 filesystem, one in Rock Ridge CE continuation handling and one in the NFS export path. Both surfaced while looking for missing bounds checks adjacent to recently-landed isofs fixes (0405d4b63d08, f54e18f1b831). Neither is a memory-safety bug on its own; the existing sb_bread() / isofs_iget() paths handle out-of-range blocks cleanly. These patches reject the malformed input at the earliest point it is known to be out of range.

1/2: rock_continue() reads rs->cont_extent from the Rock Ridge CE record and calls sb_bread() on it without validating the block number against ISOFS_SB(sb)->s_nzones. commit e595447e177b (2005) added the cont_offset and cont_size rejection but left the extent number unchecked; commit f54e18f1b831 ("isofs: Fix infinite looping over CE entries") later capped the CE chain at RR_MAX_CE_ENTRIES = 32 but again did not address the block number. The reachable attacker model is a crafted ISO auto-mounted via udisks2 / polkit on a desktop session; sb_bread() on an out-of-range block returns NULL cleanly, so there is no memory-safety issue, and the CE buffer is parsed as Rock Ridge records rather than returned verbatim via readlink().

2/2: isofs_fh_to_dentry() and isofs_fh_to_parent() pass attacker-controlled block numbers from the NFS file handle to isofs_export_iget(), which rejects block == 0 but not out-of-range block numbers. commit 0405d4b63d08 ("isofs: Prevent the use of too small fid", CVE-2025-37780) added fh_len checks but left the block number itself unchecked. An authenticated NFS peer with a crafted fid can drive reads of adjacent-partition data on the same block device into iso_inode_info fields reaching the client as dentry metadata. Deployment surface is narrow (isofs-over-NFS); impact is primarily EIO / ESTALE with a weak info-leak channel.

Quoted 7 lines

Both patches are one-line (or close to it) additions; the existing out-of-range-block check in isofs_iget() / sb_bread() handles the read-side cleanly, so these are strictly belt-and-suspenders rejection at the earliest point we know the input is out of range.

Build-tested W=1 against 7.0-rc7 with CONFIG_ISO9660_FS=y, CONFIG_JOLIET=y, CONFIG_ZISOFS=y.

Thanks! I've picked both fixes into my tree.

Honza

Quoted 12 lines

Michael Bommarito (2):
isofs: validate Rock Ridge CE continuation extent against volume size
isofs: validate block number from NFS file handle in isofs_export_iget

 fs/isofs/export.c | 2 +-
 fs/isofs/rock.c   | 9 +++++++++
 2 files changed, 10 insertions(+), 1 deletion(-)
Signature · 1 line
2.53.0
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR