[PATCH] ocfs2: reject inconsistent local xattr entries

[PATCH] ocfs2: reject inconsistent local xattr entries

[ZhengYuan Huang]

[BUG]
A corrupt OCFS2 xattr entry can set OCFS2_XATTR_ENTRY_LOCAL while
keeping xe_value_size larger than OCFS2_XATTR_INLINE_SIZE.Once
that entry reaches namevalue_size_xe(), the filesystem trips over:

kernel BUG at fs/ocfs2/xattr.c:231!
Oops: invalid opcode: 0000 [#1] SMP KASAN NOPTI
RIP: 0010:namevalue_size_xe fs/ocfs2/xattr.c:231 [inline]
RIP: 0010:ocfs2_xa_block_wipe_namevalue+0x2e4/0x330 fs/ocfs2/xattr.c:1638
Call Trace:
 ocfs2_xa_wipe_namevalue fs/ocfs2/xattr.c:1470 [inline]
 ocfs2_xa_remove_entry+0xae/0x1d0 fs/ocfs2/xattr.c:1941
 ocfs2_xa_remove fs/ocfs2/xattr.c:2043 [inline]
 ocfs2_xa_set+0x11a8/0x30a0 fs/ocfs2/xattr.c:2247
 ocfs2_xattr_ibody_set+0x302/0xc50 fs/ocfs2/xattr.c:2795
 __ocfs2_xattr_set_handle+0x7e6/0xdb0 fs/ocfs2/xattr.c:3416
 ocfs2_xattr_set+0x1447/0x2610 fs/ocfs2/xattr.c:3650
 ocfs2_xattr_security_set+0x37/0x50 fs/ocfs2/xattr.c:7241
 __vfs_removexattr+0x14d/0x1d0 fs/xattr.c:518
 cap_inode_killpriv+0x29/0x50 security/commoncap.c:355
 security_inode_killpriv+0x105/0x220 security/security.c:2724
 setattr_prepare+0x147/0x8a0 fs/attr.c:219
 ocfs2_setattr+0x504/0x1fd0 fs/ocfs2/file.c:1148
 notify_change+0x4b5/0x1030 fs/attr.c:546
 do_truncate+0x1d2/0x230 fs/open.c:68
 handle_truncate fs/namei.c:3596 [inline]
 do_open fs/namei.c:3979 [inline]
 path_openat+0x260f/0x2ce0 fs/namei.c:4134
 do_filp_open+0x1f6/0x430 fs/namei.c:4161
 do_sys_openat2+0x117/0x1c0 fs/open.c:1437
 ...

[CAUSE]
namevalue_size_xe() assumes on-disk xattr entries keep the local flag
and value size consistent, but OCFS2 does not reject an entry that
claims local storage for a value larger than the inline limit before
handing that entry to namevalue_size_xe().

The exact-match lookup path can return such an entry through xs->here,
and bucket defragmentation and splitting can also walk the corrupt entry
directly while reorganizing metadata.

[FIX]
Add a small validator for the local/value_size invariant and call it in
the paths that feed disk entries to namevalue_size_xe(): exact-match
lookup, bucket lookup results, bucket defragmentation, and bucket
splitting.

Corrupt metadata is now reported with ocfs2_error() and -EFSCORRUPTED
instead of crashing the kernel with BUG_ON().

Signed-off-by: ZhengYuan Huang <gality369@gmail.com>
---
No Fixes tag was added because this is more like a defensive check
that is long overdue, not a regression that can be clearly attributed
to a specific introduction commit.
---
diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
index 86cfd4c2adf9..63b1e55b242f 100644
--- a/fs/ocfs2/xattr.c
+++ b/fs/ocfs2/xattr.c
@@ -237,6 +237,23 @@ static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 	return namevalue_size(xe->xe_name_len, value_len);
 }
 
+static int ocfs2_validate_xattr_entry(struct inode *inode,
+				      struct ocfs2_xattr_entry *xe)
+{
+	u64 value_len = le64_to_cpu(xe->xe_value_size);
+
+	if (value_len > OCFS2_XATTR_INLINE_SIZE &&
+	    ocfs2_xattr_is_local(xe)) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %llu has corrupt xattr entry: local value_size %llu\n",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    value_len);
+		return -EFSCORRUPTED;
+	}
+
+	return 0;
+}
+
 
 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 					     struct ocfs2_xattr_header *xh,
@@ -1099,6 +1116,7 @@ static int ocfs2_xattr_find_entry(struct inode *inode, int name_index,
 	struct ocfs2_xattr_entry *entry;
 	size_t name_len;
 	int i, name_offset, cmp = 1;
+	int ret;
 
 	if (name == NULL)
 		return -EINVAL;
@@ -1121,6 +1139,11 @@ static int ocfs2_xattr_find_entry(struct inode *inode, int name_index,
 				return -EFSCORRUPTED;
 			}
 			cmp = memcmp(name, (xs->base + name_offset), name_len);
+			if (!cmp) {
+				ret = ocfs2_validate_xattr_entry(inode, entry);
+				if (ret)
+					return ret;
+			}
 		}
 		if (cmp == 0)
 			break;
@@ -3823,6 +3846,9 @@ static int ocfs2_find_xe_in_bucket(struct inode *inode,
 
 		xe_name = bucket_block(bucket, block_off) + new_offset;
 		if (!memcmp(name, xe_name, name_len)) {
+			ret = ocfs2_validate_xattr_entry(inode, xe);
+			if (ret)
+				break;
 			*xe_index = i;
 			*found = 1;
 			ret = 0;
@@ -4463,6 +4489,10 @@ static int ocfs2_defrag_xattr_bucket(struct inode *inode,
 	xe = xh->xh_entries;
 	end = OCFS2_XATTR_BUCKET_SIZE;
 	for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
+		ret = ocfs2_validate_xattr_entry(inode, xe);
+		if (ret)
+			goto out;
+
 		offset = le16_to_cpu(xe->xe_name_offset);
 		len = namevalue_size_xe(xe);
 
@@ -4745,6 +4775,10 @@ static int ocfs2_divide_xattr_bucket(struct inode *inode,
 	name_value_len = 0;
 	for (i = 0; i < start; i++) {
 		xe = &xh->xh_entries[i];
+		ret = ocfs2_validate_xattr_entry(inode, xe);
+		if (ret)
+			goto out;
+
 		name_value_len += namevalue_size_xe(xe);
 		if (le16_to_cpu(xe->xe_name_offset) < name_offset)
 			name_offset = le16_to_cpu(xe->xe_name_offset);
-- 
2.43.0

Re: [PATCH] ocfs2: reject inconsistent local xattr entries

[Joseph Qi]

This conflicts with the following commit in linux-next:
ocfs2: fix circular locking dependency in ocfs2_init_acl()

Thanks,
Joseph

On 8/4/26 7:28 PM, ZhengYuan Huang wrote:
> [BUG]
> A corrupt OCFS2 xattr entry can set OCFS2_XATTR_ENTRY_LOCAL while
> keeping xe_value_size larger than OCFS2_XATTR_INLINE_SIZE.Once
> that entry reaches namevalue_size_xe(), the filesystem trips over:
> 
> kernel BUG at fs/ocfs2/xattr.c:231!
> Oops: invalid opcode: 0000 [#1] SMP KASAN NOPTI
> RIP: 0010:namevalue_size_xe fs/ocfs2/xattr.c:231 [inline]
> RIP: 0010:ocfs2_xa_block_wipe_namevalue+0x2e4/0x330 fs/ocfs2/xattr.c:1638
> Call Trace:
>  ocfs2_xa_wipe_namevalue fs/ocfs2/xattr.c:1470 [inline]
>  ocfs2_xa_remove_entry+0xae/0x1d0 fs/ocfs2/xattr.c:1941
>  ocfs2_xa_remove fs/ocfs2/xattr.c:2043 [inline]
>  ocfs2_xa_set+0x11a8/0x30a0 fs/ocfs2/xattr.c:2247
>  ocfs2_xattr_ibody_set+0x302/0xc50 fs/ocfs2/xattr.c:2795
>  __ocfs2_xattr_set_handle+0x7e6/0xdb0 fs/ocfs2/xattr.c:3416
>  ocfs2_xattr_set+0x1447/0x2610 fs/ocfs2/xattr.c:3650
>  ocfs2_xattr_security_set+0x37/0x50 fs/ocfs2/xattr.c:7241
>  __vfs_removexattr+0x14d/0x1d0 fs/xattr.c:518
>  cap_inode_killpriv+0x29/0x50 security/commoncap.c:355
>  security_inode_killpriv+0x105/0x220 security/security.c:2724
>  setattr_prepare+0x147/0x8a0 fs/attr.c:219
>  ocfs2_setattr+0x504/0x1fd0 fs/ocfs2/file.c:1148
>  notify_change+0x4b5/0x1030 fs/attr.c:546
>  do_truncate+0x1d2/0x230 fs/open.c:68
>  handle_truncate fs/namei.c:3596 [inline]
>  do_open fs/namei.c:3979 [inline]
>  path_openat+0x260f/0x2ce0 fs/namei.c:4134
>  do_filp_open+0x1f6/0x430 fs/namei.c:4161
>  do_sys_openat2+0x117/0x1c0 fs/open.c:1437
>  ...
> 
> [CAUSE]
> namevalue_size_xe() assumes on-disk xattr entries keep the local flag
> and value size consistent, but OCFS2 does not reject an entry that
> claims local storage for a value larger than the inline limit before
> handing that entry to namevalue_size_xe().
> 
> The exact-match lookup path can return such an entry through xs->here,
> and bucket defragmentation and splitting can also walk the corrupt entry
> directly while reorganizing metadata.
> 
> [FIX]
> Add a small validator for the local/value_size invariant and call it in
> the paths that feed disk entries to namevalue_size_xe(): exact-match
> lookup, bucket lookup results, bucket defragmentation, and bucket
> splitting.
> 
> Corrupt metadata is now reported with ocfs2_error() and -EFSCORRUPTED
> instead of crashing the kernel with BUG_ON().
> 
> Signed-off-by: ZhengYuan Huang <gality369@gmail.com>
> ---
> No Fixes tag was added because this is more like a defensive check
> that is long overdue, not a regression that can be clearly attributed
> to a specific introduction commit.
> ---
> diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
> index 86cfd4c2adf9..63b1e55b242f 100644
> --- a/fs/ocfs2/xattr.c
> +++ b/fs/ocfs2/xattr.c
> @@ -237,6 +237,23 @@ static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
>  	return namevalue_size(xe->xe_name_len, value_len);
>  }
>  
> +static int ocfs2_validate_xattr_entry(struct inode *inode,
> +				      struct ocfs2_xattr_entry *xe)
> +{
> +	u64 value_len = le64_to_cpu(xe->xe_value_size);
> +
> +	if (value_len > OCFS2_XATTR_INLINE_SIZE &&
> +	    ocfs2_xattr_is_local(xe)) {
> +		ocfs2_error(inode->i_sb,
> +			    "Inode %llu has corrupt xattr entry: local value_size %llu\n",
> +			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
> +			    value_len);
> +		return -EFSCORRUPTED;
> +	}
> +
> +	return 0;
> +}
> +
>  
>  static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
>  					     struct ocfs2_xattr_header *xh,
> @@ -1099,6 +1116,7 @@ static int ocfs2_xattr_find_entry(struct inode *inode, int name_index,
>  	struct ocfs2_xattr_entry *entry;
>  	size_t name_len;
>  	int i, name_offset, cmp = 1;
> +	int ret;
>  
>  	if (name == NULL)
>  		return -EINVAL;
> @@ -1121,6 +1139,11 @@ static int ocfs2_xattr_find_entry(struct inode *inode, int name_index,
>  				return -EFSCORRUPTED;
>  			}
>  			cmp = memcmp(name, (xs->base + name_offset), name_len);
> +			if (!cmp) {
> +				ret = ocfs2_validate_xattr_entry(inode, entry);
> +				if (ret)
> +					return ret;
> +			}
>  		}
>  		if (cmp == 0)
>  			break;
> @@ -3823,6 +3846,9 @@ static int ocfs2_find_xe_in_bucket(struct inode *inode,
>  
>  		xe_name = bucket_block(bucket, block_off) + new_offset;
>  		if (!memcmp(name, xe_name, name_len)) {
> +			ret = ocfs2_validate_xattr_entry(inode, xe);
> +			if (ret)
> +				break;
>  			*xe_index = i;
>  			*found = 1;
>  			ret = 0;
> @@ -4463,6 +4489,10 @@ static int ocfs2_defrag_xattr_bucket(struct inode *inode,
>  	xe = xh->xh_entries;
>  	end = OCFS2_XATTR_BUCKET_SIZE;
>  	for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
> +		ret = ocfs2_validate_xattr_entry(inode, xe);
> +		if (ret)
> +			goto out;
> +
>  		offset = le16_to_cpu(xe->xe_name_offset);
>  		len = namevalue_size_xe(xe);
>  
> @@ -4745,6 +4775,10 @@ static int ocfs2_divide_xattr_bucket(struct inode *inode,
>  	name_value_len = 0;
>  	for (i = 0; i < start; i++) {
>  		xe = &xh->xh_entries[i];
> +		ret = ocfs2_validate_xattr_entry(inode, xe);
> +		if (ret)
> +			goto out;
> +
>  		name_value_len += namevalue_size_xe(xe);
>  		if (le16_to_cpu(xe->xe_name_offset) < name_offset)
>  			name_offset = le16_to_cpu(xe->xe_name_offset);

[PATCH] ocfs2: reject inconsistent local xattr entries

[gality369]

[BUG]
A corrupt OCFS2 xattr entry can set OCFS2_XATTR_ENTRY_LOCAL while
keeping xe_value_size larger than OCFS2_XATTR_INLINE_SIZE.

Once that entry reaches namevalue_size_xe(), the filesystem trips over:

kernel BUG at fs/ocfs2/xattr.c:231!
Oops: invalid opcode: 0000 [#1] SMP KASAN NOPTI
RIP: 0010:namevalue_size_xe fs/ocfs2/xattr.c:231 [inline]
RIP: 0010:ocfs2_xa_block_wipe_namevalue+0x2e4/0x330 fs/ocfs2/xattr.c:1638
Call Trace:
 ocfs2_xa_wipe_namevalue fs/ocfs2/xattr.c:1470 [inline]
 ocfs2_xa_remove_entry+0xae/0x1d0 fs/ocfs2/xattr.c:1941
 ocfs2_xa_remove fs/ocfs2/xattr.c:2043 [inline]
 ocfs2_xa_set+0x11a8/0x30a0 fs/ocfs2/xattr.c:2247
 ocfs2_xattr_ibody_set+0x302/0xc50 fs/ocfs2/xattr.c:2795
 __ocfs2_xattr_set_handle+0x7e6/0xdb0 fs/ocfs2/xattr.c:3416
 ocfs2_xattr_set+0x1447/0x2610 fs/ocfs2/xattr.c:3650
 ocfs2_xattr_security_set+0x37/0x50 fs/ocfs2/xattr.c:7241
 __vfs_removexattr+0x14d/0x1d0 fs/xattr.c:518
 cap_inode_killpriv+0x29/0x50 security/commoncap.c:355
 security_inode_killpriv+0x105/0x220 security/security.c:2724
 setattr_prepare+0x147/0x8a0 fs/attr.c:219
 ocfs2_setattr+0x504/0x1fd0 fs/ocfs2/file.c:1148
 notify_change+0x4b5/0x1030 fs/attr.c:546
 do_truncate+0x1d2/0x230 fs/open.c:68
 handle_truncate fs/namei.c:3596 [inline]
 do_open fs/namei.c:3979 [inline]
 path_openat+0x260f/0x2ce0 fs/namei.c:4134
 do_filp_open+0x1f6/0x430 fs/namei.c:4161
 do_sys_openat2+0x117/0x1c0 fs/open.c:1437
 do_sys_open fs/open.c:1452 [inline]
 ...

[CAUSE]
namevalue_size_xe() assumes on-disk xattr entries keep the local flag
and value size consistent, but OCFS2 does not reject an entry that
claims local storage for a value larger than the inline limit before
handing that entry to namevalue_size_xe().

The exact-match lookup path can return such an entry through xs->here,
and bucket defragmentation and splitting can also walk the corrupt entry
directly while reorganizing metadata.

[FIX]
Add a small validator for the local/value_size invariant and call it in
the paths that feed disk entries to namevalue_size_xe(): exact-match
lookup, bucket lookup results, bucket defragmentation, and bucket
splitting.

Corrupt metadata is now reported with ocfs2_error() and -EFSCORRUPTED
instead of crashing the kernel with BUG_ON().

Signed-off-by: gality369 <gality369@example.com>
---
No Fixes tag was added because this is more like a defensive check
that is long overdue, not a regression that can be clearly attributed
to a specific introduction commit.
---
diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
index 86cfd4c2adf9..63b1e55b242f 100644
--- a/fs/ocfs2/xattr.c
+++ b/fs/ocfs2/xattr.c
@@ -237,6 +237,23 @@ static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
 	return namevalue_size(xe->xe_name_len, value_len);
 }
 
+static int ocfs2_validate_xattr_entry(struct inode *inode,
+				      struct ocfs2_xattr_entry *xe)
+{
+	u64 value_len = le64_to_cpu(xe->xe_value_size);
+
+	if (value_len > OCFS2_XATTR_INLINE_SIZE &&
+	    ocfs2_xattr_is_local(xe)) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %llu has corrupt xattr entry: local value_size %llu\n",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    value_len);
+		return -EFSCORRUPTED;
+	}
+
+	return 0;
+}
+
 
 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 					     struct ocfs2_xattr_header *xh,
@@ -1099,6 +1116,7 @@ static int ocfs2_xattr_find_entry(struct inode *inode, int name_index,
 	struct ocfs2_xattr_entry *entry;
 	size_t name_len;
 	int i, name_offset, cmp = 1;
+	int ret;
 
 	if (name == NULL)
 		return -EINVAL;
@@ -1121,6 +1139,11 @@ static int ocfs2_xattr_find_entry(struct inode *inode, int name_index,
 				return -EFSCORRUPTED;
 			}
 			cmp = memcmp(name, (xs->base + name_offset), name_len);
+			if (!cmp) {
+				ret = ocfs2_validate_xattr_entry(inode, entry);
+				if (ret)
+					return ret;
+			}
 		}
 		if (cmp == 0)
 			break;
@@ -3823,6 +3846,9 @@ static int ocfs2_find_xe_in_bucket(struct inode *inode,
 
 		xe_name = bucket_block(bucket, block_off) + new_offset;
 		if (!memcmp(name, xe_name, name_len)) {
+			ret = ocfs2_validate_xattr_entry(inode, xe);
+			if (ret)
+				break;
 			*xe_index = i;
 			*found = 1;
 			ret = 0;
@@ -4463,6 +4489,10 @@ static int ocfs2_defrag_xattr_bucket(struct inode *inode,
 	xe = xh->xh_entries;
 	end = OCFS2_XATTR_BUCKET_SIZE;
 	for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
+		ret = ocfs2_validate_xattr_entry(inode, xe);
+		if (ret)
+			goto out;
+
 		offset = le16_to_cpu(xe->xe_name_offset);
 		len = namevalue_size_xe(xe);
 
@@ -4745,6 +4775,10 @@ static int ocfs2_divide_xattr_bucket(struct inode *inode,
 	name_value_len = 0;
 	for (i = 0; i < start; i++) {
 		xe = &xh->xh_entries[i];
+		ret = ocfs2_validate_xattr_entry(inode, xe);
+		if (ret)
+			goto out;
+
 		name_value_len += namevalue_size_xe(xe);
 		if (le16_to_cpu(xe->xe_name_offset) < name_offset)
 			name_offset = le16_to_cpu(xe->xe_name_offset);
-- 
2.43.0