Skip to contentSkip to conversation

jfs: fix missing bounds check in __jfs_setxattr()

PATCH In review Sebastian Alba Vives 3 participants 4 messages Apr 13 – Apr 16

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 jfs: fix missing bounds check in __jfs_setxattr()

The loop in __jfs_setxattr() iterates over EA entries without validating that NEXT_EA(ea) stays within the ealist buffer. A crafted filesystem image with a manipulated EA valuelen field can cause NEXT_EA() to land inside the buffer but past the last valid entry, leading to out-of-bounds reads in the subsequent memcmp() call.

Fix this by adding the same bounds check already present in __jfs_getxattr() and jfs_listxattr().

---
 fs/jfs/xattr.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
fs/jfs/xattr.c +7 -1
index 11d7f74d2..320659bb8 100644
@@ -688,8 +688,14 @@ int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name,
 	new_size = sizeof (struct jfs_ea_list);
 
 	if (xattr_size) {
-		for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist);
+		struct jfs_ea *ealist_end = END_EALIST(ealist);
+		for (ea = FIRST_EA(ealist); ea < ealist_end;
 		     ea = NEXT_EA(ea)) {
+			if (unlikely(ea + 1 > ealist_end) ||
+			    unlikely(NEXT_EA(ea) > ealist_end)) {
+				rc = -EUCLEAN;
+				goto release;
+			}
 			if ((namelen == ea->namelen) &&
 			    (memcmp(name, ea->name, namelen) == 0)) {
 				found = 1;
Signature · 1 line
2.43.0
#1 lore raw
Quoted 38 lines · Sebastian Alba Vives

The loop in __jfs_setxattr() iterates over EA entries without validating that NEXT_EA(ea) stays within the ealist buffer. A crafted filesystem image with a manipulated EA valuelen field can cause NEXT_EA() to land inside the buffer but past the last valid entry, leading to out-of-bounds reads in the subsequent memcmp() call.

Fix this by adding the same bounds check already present in __jfs_getxattr() and jfs_listxattr().

---
 fs/jfs/xattr.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
fs/jfs/xattr.c +7 -1
index 11d7f74d2..320659bb8 100644
@@ -688,8 +688,14 @@ int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name,
 	new_size = sizeof (struct jfs_ea_list);
 
 	if (xattr_size) {
-		for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist);
+		struct jfs_ea *ealist_end = END_EALIST(ealist);
+		for (ea = FIRST_EA(ealist); ea < ealist_end;
 		     ea = NEXT_EA(ea)) {
+			if (unlikely(ea + 1 > ealist_end) ||
+			    unlikely(NEXT_EA(ea) > ealist_end)) {
+				rc = -EUCLEAN;
+				goto release;
+			}
 			if ((namelen == ea->namelen) &&
 			    (memcmp(name, ea->name, namelen) == 0)) {
 				found = 1;
Signature · 1 line
2.43.0

Does the userspace fsck program catch this first? That would prevent this check from having to always be here, slowing down this operation.

thanks,

greg k-h

#2 lore raw
In reply to Sebastian Alba Vives
Quoted 20 lines · Sebastian Alba Vives

The loop in __jfs_setxattr() iterates over EA entries without validating that NEXT_EA(ea) stays within the ealist buffer. A crafted filesystem image with a manipulated EA valuelen field can cause NEXT_EA() to land inside the buffer but past the last valid entry, leading to out-of-bounds reads in the subsequent memcmp() call.

Fix this by adding the same bounds check already present in __jfs_getxattr() and jfs_listxattr().

---
 fs/jfs/xattr.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
fs/jfs/xattr.c +0 -0
index 11d7f74d2..320659bb8 100644

fs/jfs/xattr.c int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name, · line 694

Quoted 8 lines
fs/jfs/xattr.c +3 -1
@@ -688,8 +688,14 @@ int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name,
 	new_size = sizeof (struct jfs_ea_list);
 
 	if (xattr_size) {
-		for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist);
+		struct jfs_ea *ealist_end = END_EALIST(ealist);
+		for (ea = FIRST_EA(ealist); ea < ealist_end;
 		     ea = NEXT_EA(ea)) {
+			if (unlikely(ea + 1 > ealist_end) ||

What use is this check? How can ea + 1 > ealist_end immediately after ea < ealist_end?

Quoted 10 lines
fs/jfs/xattr.c +4 -0
+			    unlikely(NEXT_EA(ea) > ealist_end)) {
+				rc = -EUCLEAN;
+				goto release;
+			}
 			if ((namelen == ea->namelen) &&
 			    (memcmp(name, ea->name, namelen) == 0)) {
 				found = 1;
Signature · 1 line
2.43.0

You should also know that jfs is not maintained, and there are MANY of these issues (see syzbot for instance). To my knowledge the community has decided to deprioritize these kinds of issues, and perhaps eventually move to FUSE based implementations for legacy filesystems.

Also, you should cc the general list [email protected] and the jfs maintainer Dave Kleikamp <[email protected]>. Use the get_maintainer script.

That being said these fixes are still accepted. Thanks for your (first?) contribution :)

Thanks,
Jori.

#3 lore raw
Quoted 13 lines

Op 16-04-2026 12:01 CEST schreef Jori Koolstra <[email protected]>:

Quoted 11 lines
---
 fs/jfs/xattr.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
fs/jfs/xattr.c +0 -0
index 11d7f74d2..320659bb8 100644
@@ -688,8 +688,14 @@ int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name,
 	new_size = sizeof (struct jfs_ea_list);
 
Quoted 10 lines

fs/jfs/xattr.c int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name, · line 694

Quoted 6 lines
fs/jfs/xattr.c +3 -1
 	if (xattr_size) {
-		for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist);
+		struct jfs_ea *ealist_end = END_EALIST(ealist);
+		for (ea = FIRST_EA(ealist); ea < ealist_end;
 		     ea = NEXT_EA(ea)) {
+			if (unlikely(ea + 1 > ealist_end) ||

What use is this check? How can ea + 1 > ealist_end immediately after ea < ealist_end?

Maybe I should rephrase this a bit more clearly: does the

unlikely(NEXT_EA(ea) > ealist_end))

already take care of that potential scenario?