Skip to contentSkip to conversation

smb: client: Fix refcount leak for cifs_sb_tlink

PATCH In review Shuhao Fu 3 participants 4 messages Oct 16, 2025

Subsystems (3)
lore
Share & tools
lore

Shuhao Fu

author
#0 lore raw

PATCH v2 1/1 smb: client: Fix refcount leak for cifs_sb_tlink

Fix three refcount inconsistency issues related to cifs_sb_tlink.

Comments for cifs_sb_tlink state that cifs_put_tlink() needs to be called after successful calls to cifs_sb_tlink(). Three calls fail to update refcount accordingly, leading to possible resource leaks.

Fixes 8ceb98437946 ("CIFS: Move rename to ops struct")
Fixes 2f1afe25997f ("cifs: Use smb 2 - 3 and cifsacl mount options getacl functions")
Fixes 366ed846df60 ("cifs: Use smb 2 - 3 and cifsacl mount options setacl function")

---
Change in v2:
1. improved patch wording
2. nicer goto label naming

Link to v1: https://lore.kernel.org/linux-cifs/[email protected]/

---
 fs/smb/client/inode.c   | 6 ++++--
 fs/smb/client/smb2ops.c | 8 ++++----
 2 files changed, 8 insertions(+), 6 deletions(-)
fs/smb/client/inode.c +4 -2
index 239dd84a3..098a79b7a 100644
@@ -2431,8 +2431,10 @@ cifs_do_rename(const unsigned int xid, struct dentry *from_dentry,
 	tcon = tlink_tcon(tlink);
 	server = tcon->ses->server;
 
-	if (!server->ops->rename)
-		return -ENOSYS;
+	if (!server->ops->rename) {
+		rc = -ENOSYS;
+		goto do_rename_exit;
+	}
 
 	/* try path-based rename first */
 	rc = server->ops->rename(xid, tcon, from_dentry,
fs/smb/client/smb2ops.c +4 -4
index 7c392cf59..00b3f769e 100644
@@ -3212,8 +3212,7 @@ get_smb2_acl_by_path(struct cifs_sb_info *cifs_sb,
 	utf16_path = cifs_convert_path_to_utf16(path, cifs_sb);
 	if (!utf16_path) {
 		rc = -ENOMEM;
-		free_xid(xid);
-		return ERR_PTR(rc);
+		goto put_tlink;
 	}
 
 	oparms = (struct cifs_open_parms) {
@@ -3245,6 +3244,7 @@ get_smb2_acl_by_path(struct cifs_sb_info *cifs_sb,
 		SMB2_close(xid, tcon, fid.persistent_fid, fid.volatile_fid);
 	}
 
+put_tlink:
 	cifs_put_tlink(tlink);
 	free_xid(xid);
 
@@ -3285,8 +3285,7 @@ set_smb2_acl(struct smb_ntsd *pnntsd, __u32 acllen,
 	utf16_path = cifs_convert_path_to_utf16(path, cifs_sb);
 	if (!utf16_path) {
 		rc = -ENOMEM;
-		free_xid(xid);
-		return rc;
+		goto put_tlink;
 	}
 
 	oparms = (struct cifs_open_parms) {
@@ -3307,6 +3306,7 @@ set_smb2_acl(struct smb_ntsd *pnntsd, __u32 acllen,
 		SMB2_close(xid, tcon, fid.persistent_fid, fid.volatile_fid);
 	}
 
+put_tlink:
 	cifs_put_tlink(tlink);
 	free_xid(xid);
 	return rc;
Signature · 1 line
2.39.5 (Apple Git-154)
#1 lore raw

updated the patch in cifs-2.6.git for-next

Quoted 83 lines · Shuhao Fu

Fix three refcount inconsistency issues related to cifs_sb_tlink.

Comments for cifs_sb_tlink state that cifs_put_tlink() needs to be called after successful calls to cifs_sb_tlink(). Three calls fail to update refcount accordingly, leading to possible resource leaks.

Fixes 8ceb98437946 ("CIFS: Move rename to ops struct")
Fixes 2f1afe25997f ("cifs: Use smb 2 - 3 and cifsacl mount options getacl functions")
Fixes 366ed846df60 ("cifs: Use smb 2 - 3 and cifsacl mount options setacl function")

---
Change in v2:
1. improved patch wording
2. nicer goto label naming

Link to v1: https://lore.kernel.org/linux-cifs/[email protected]/

---
 fs/smb/client/inode.c   | 6 ++++--
 fs/smb/client/smb2ops.c | 8 ++++----
 2 files changed, 8 insertions(+), 6 deletions(-)
fs/smb/client/inode.c +4 -2
index 239dd84a3..098a79b7a 100644
@@ -2431,8 +2431,10 @@ cifs_do_rename(const unsigned int xid, struct dentry *from_dentry,
        tcon = tlink_tcon(tlink);
        server = tcon->ses->server;
 
-       if (!server->ops->rename)
-               return -ENOSYS;
+       if (!server->ops->rename) {
+               rc = -ENOSYS;
+               goto do_rename_exit;
+       }
 
        /* try path-based rename first */
        rc = server->ops->rename(xid, tcon, from_dentry,
fs/smb/client/smb2ops.c +4 -5
index 7c392cf59..00b3f769e 100644
@@ -3212,8 +3212,7 @@ get_smb2_acl_by_path(struct cifs_sb_info *cifs_sb,
        utf16_path = cifs_convert_path_to_utf16(path, cifs_sb);
        if (!utf16_path) {
                rc = -ENOMEM;
-               free_xid(xid);
-               return ERR_PTR(rc);
+               goto put_tlink;
        }
 
        oparms = (struct cifs_open_parms) {
@@ -3245,6 +3244,7 @@ get_smb2_acl_by_path(struct cifs_sb_info *cifs_sb,
                SMB2_close(xid, tcon, fid.persistent_fid, fid.volatile_fid);
        }
 
+put_tlink:
        cifs_put_tlink(tlink);
        free_xid(xid);
 
@@ -3285,8 +3285,7 @@ set_smb2_acl(struct smb_ntsd *pnntsd, __u32 acllen,
        utf16_path = cifs_convert_path_to_utf16(path, cifs_sb);
        if (!utf16_path) {
                rc = -ENOMEM;
-               free_xid(xid);
-               return rc;
+               goto put_tlink;
        }
 
        oparms = (struct cifs_open_parms) {
@@ -3307,6 +3306,7 @@ set_smb2_acl(struct smb_ntsd *pnntsd, __u32 acllen,
                SMB2_close(xid, tcon, fid.persistent_fid, fid.volatile_fid);
        }
 
+put_tlink:
        cifs_put_tlink(tlink);
        free_xid(xid);
        return rc;
--

2.39.5 (Apple Git-154)

Signature · 3 lines
Thanks,

Steve
#2 lore raw
In reply to Shuhao Fu
Quoted 1 line

Fix three refcount inconsistency issues related to cifs_sb_tlink.

I find such an introduction sentence not so relevant here.

Quoted 3 lines

Comments for cifs_sb_tlink state that cifs_put_tlink() needs to be called after successful calls to cifs_sb_tlink(). Three calls fail to update refcount accordingly, leading to possible resource leaks.

* Can it be preferred to refer to the term “reference count”?

* Would you find a description of corresponding case distinctions more helpful?

* May resource leaks be indicated also in the summary phrase?

* Would it be helpful to append parentheses to function names at more places?

* Is there a need to mention change steps more individually?
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/process/submitting-patches.rst?h=v6.17#n94

* Will development interests grow for the application of scope-based resource management?

Regards,
Markus

#3 lore raw

The patch looks fine. More important is focusing on the fixes (and missing features) - minor wording of description can be helpful but MUCH more important is focusing on xfstests, new test scenarios, automated analysis to find places where use after frees possible etc, fuzzing (like the great scripts Dr. Morris created for us to show some potential security issues), fixing the various known bugs, adding the missing features etc

Quoted 27 lines · Markus Elfring
Quoted 2 lines

Fix three refcount inconsistency issues related to cifs_sb_tlink.

I find such an introduction sentence not so relevant here.

Quoted 3 lines

Comments for cifs_sb_tlink state that cifs_put_tlink() needs to be called after successful calls to cifs_sb_tlink(). Three calls fail to update refcount accordingly, leading to possible resource leaks.

* Can it be preferred to refer to the term “reference count”?

* Would you find a description of corresponding case distinctions more helpful?

* May resource leaks be indicated also in the summary phrase?

* Would it be helpful to append parentheses to function names at more places?

* Is there a need to mention change steps more individually?
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/process/submitting-patches.rst?h=v6.17#n94

* Will development interests grow for the application of scope-based resource management?

Regards,
Markus

Signature · 3 lines
Thanks,

Steve