Skip to contentSkip to conversation

9p: don't WARN_ONCE on racy symlink fid lookup failure

PATCH In review Deepanshu Kartikey 2 participants 4 messages Jul 25 – Sep 13 9P file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 9p: don't WARN_ONCE on racy symlink fid lookup failure

v9fs_init_request() WARN_ONCE()s when v9fs_fid_lookup() fails for a symlink read via the page cache. This is a benign TOCTOU race: a concurrent unlink()/rename() can remove the target between the walk finding the dentry and p9_client_walk() completing, so the lookup legitimately fails with -ENOENT (observed: dentry=/file0, err=-2).

With panic_on_warn=1, this WARN escalates to a full kernel panic -- an unprivileged local DoS reachable by racing mount() against a concurrent unlink/rename.

Return the error directly instead of WARN_ONCE().

---
 fs/9p/vfs_addr.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
fs/9p/vfs_addr.c +4 -1
index 1ac0b3dcc077..121679488ab9 100644
@@ -140,13 +140,16 @@ static int v9fs_init_request(struct netfs_io_request *rreq, struct file *file)
 			goto no_fid;
 		p9_fid_get(fid);
 	} else if (S_ISLNK(rreq->inode->i_mode)) {
+		/* racing unlink/rename can make this fail with -ENOENT;
+		 * that's expected, not a kernel bug, so don't WARN
+		 */
 		dentry = d_find_any_alias(rreq->inode);
 		if (!dentry)
 			goto no_fid;
 		fid = v9fs_fid_lookup(dentry);
 		dput(dentry);
 		if (IS_ERR(fid))
-			goto no_fid;
+			return PTR_ERR(fid);
 	} else {
 		fid = v9fs_fid_find_inode(rreq->inode, writing, INVALID_UID, true);
 		if (!fid)
Signature · 1 line
2.43.0
#1 lore raw
Quoted 43 lines · Deepanshu Kartikey

v9fs_init_request() WARN_ONCE()s when v9fs_fid_lookup() fails for a symlink read via the page cache. This is a benign TOCTOU race: a concurrent unlink()/rename() can remove the target between the walk finding the dentry and p9_client_walk() completing, so the lookup legitimately fails with -ENOENT (observed: dentry=/file0, err=-2).

With panic_on_warn=1, this WARN escalates to a full kernel panic -- an unprivileged local DoS reachable by racing mount() against a concurrent unlink/rename.

Return the error directly instead of WARN_ONCE().

---
 fs/9p/vfs_addr.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
fs/9p/vfs_addr.c +4 -2
index 1ac0b3dcc077..121679488ab9 100644
@@ -140,13 +140,16 @@ static int v9fs_init_request(struct netfs_io_request *rreq, struct file *file)
                        goto no_fid;
                p9_fid_get(fid);
        } else if (S_ISLNK(rreq->inode->i_mode)) {
+               /* racing unlink/rename can make this fail with -ENOENT;
+                * that's expected, not a kernel bug, so don't WARN
+                */
                dentry = d_find_any_alias(rreq->inode);
                if (!dentry)
                        goto no_fid;
                fid = v9fs_fid_lookup(dentry);
                dput(dentry);
                if (IS_ERR(fid))
-                       goto no_fid;
+                       return PTR_ERR(fid);
        } else {
                fid = v9fs_fid_find_inode(rreq->inode, writing, INVALID_UID, true);
                if (!fid)
--
Quoted 2 lines

2.43.0

Gentle Reminder. Let me know anythng else needed from my side.

Thanks

Deepanshu

#2 lore raw
In reply to Deepanshu Kartikey
Quoted 43 lines · Deepanshu Kartikey

v9fs_init_request() WARN_ONCE()s when v9fs_fid_lookup() fails for a symlink read via the page cache. This is a benign TOCTOU race: a concurrent unlink()/rename() can remove the target between the walk finding the dentry and p9_client_walk() completing, so the lookup legitimately fails with -ENOENT (observed: dentry=/file0, err=-2).

With panic_on_warn=1, this WARN escalates to a full kernel panic -- an unprivileged local DoS reachable by racing mount() against a concurrent unlink/rename.

Return the error directly instead of WARN_ONCE().

---
 fs/9p/vfs_addr.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
fs/9p/vfs_addr.c +4 -2
index 1ac0b3dcc077..121679488ab9 100644
@@ -140,13 +140,16 @@ static int v9fs_init_request(struct netfs_io_request *rreq, struct file *file)
                        goto no_fid;
                p9_fid_get(fid);
        } else if (S_ISLNK(rreq->inode->i_mode)) {
+               /* racing unlink/rename can make this fail with -ENOENT;
+                * that's expected, not a kernel bug, so don't WARN
+                */
                dentry = d_find_any_alias(rreq->inode);
                if (!dentry)
                        goto no_fid;
                fid = v9fs_fid_lookup(dentry);
                dput(dentry);
                if (IS_ERR(fid))
-                       goto no_fid;
+                       return PTR_ERR(fid);
        } else {
                fid = v9fs_fid_find_inode(rreq->inode, writing, INVALID_UID, true);
                if (!fid)
--
Quoted 2 lines

2.43.0

Gentle Reminder. Let me know anythng else needed from my side.

Thanks

Deepanshu

#3 lore raw
In reply to Deepanshu Kartikey

Deepanshu Kartikey wrote on Sat, Jul 25, 2026 at 07:14:48AM +0530:

Quoted 10 lines

v9fs_init_request() WARN_ONCE()s when v9fs_fid_lookup() fails for a symlink read via the page cache. This is a benign TOCTOU race: a concurrent unlink()/rename() can remove the target between the walk finding the dentry and p9_client_walk() completing, so the lookup legitimately fails with -ENOENT (observed: dentry=/file0, err=-2).

With panic_on_warn=1, this WARN escalates to a full kernel panic -- an unprivileged local DoS reachable by racing mount() against a concurrent unlink/rename.

Quoted 5 lines

Return the error directly instead of WARN_ONCE().

Fair enough, picked up for 7.4

Signature · 1 line
Dominique Martinet | Asmadeus