Skip to contentSkip to conversation

Check for NULL fid pointers in p9_client_clunk()

PATCH In review Venkateswararao Jujjuri (JV) 2 participants 3 messages Aug 24, 2010

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

PATCH 1/1 Check for NULL fid pointers in p9_client_clunk()

NULL fid should be handled in cases where we endup calling v9fs_dir_release() before even we instantiate the fid in filp. This patch handles pasing a NULL p9_fid* to p9_client_clunk.

---
 fs/9p/vfs_dir.c |    3 ++-
 net/9p/client.c |    3 +++
 2 files changed, 5 insertions(+), 1 deletions(-)
fs/9p/vfs_dir.c +2 -1
index 16c8a2a..5f08203 100644
@@ -292,7 +292,8 @@ int v9fs_dir_release(struct inode *inode, struct file *filp)
 
 	fid = filp->private_data;
 	P9_DPRINTK(P9_DEBUG_VFS,
-			"inode: %p filp: %p fid: %d\n", inode, filp, fid->fid);
+			"JV: inode: %p filp: %p fid: %d\n", inode, filp,
+			fid ? fid->fid : -1);
 	filemap_write_and_wait(inode->i_mapping);
 	p9_client_clunk(fid);
 	return 0;
net/9p/client.c +3 -0
index dc6f2f2..9338fb3 100644
@@ -1201,6 +1201,9 @@ int p9_client_clunk(struct p9_fid *fid)
 	struct p9_client *clnt;
 	struct p9_req_t *req;
 
+	if (!fid)
+		return 0;
+
 	P9_DPRINTK(P9_DEBUG_9P, ">>> TCLUNK fid %d\n", fid->fid);
 	err = 0;
 	clnt = fid->clnt;
Signature · 1 line
1.6.5.2
#1 lore raw
Quoted 15 lines · Venkateswararao Jujjuri (JV)

NULL fid should be handled in cases where we endup calling v9fs_dir_release() before even we instantiate the fid in filp. This patch handles pasing a NULL p9_fid* to p9_client_clunk.

---
 fs/9p/vfs_dir.c |    3 ++-
 net/9p/client.c |    3 +++

 2 files changed, 5 insertions(+), 1 deletions(-)
fs/9p/vfs_dir.c +0 -0
index 16c8a2a..5f08203 100644

fs/9p/vfs_dir.c int v9fs_dir_release(struct inode *inode, struct file *filp) · line 296

Quoted 6 lines
fs/9p/vfs_dir.c +2 -1
@@ -292,7 +292,8 @@ int v9fs_dir_release(struct inode *inode, struct file *filp)
 
        fid = filp->private_data;
        P9_DPRINTK(P9_DEBUG_VFS,
-                       "inode: %p filp: %p fid: %d\n", inode, filp, fid->fid);
+                       "JV: inode: %p filp: %p fid: %d\n", inode, filp,
+                       fid ? fid->fid : -1);

Did you really mean to insert a JV: debug label in there?

Quoted 8 lines
fs/9p/vfs_dir.c +0 -0
        filemap_write_and_wait(inode->i_mapping);
        p9_client_clunk(fid);
        return 0;
net/9p/client.c +0 -0
index dc6f2f2..9338fb3 100644

net/9p/client.c int p9_client_clunk(struct p9_fid *fid) · line 1207

Quoted 7 lines
net/9p/client.c +3 -0
@@ -1201,6 +1201,9 @@ int p9_client_clunk(struct p9_fid *fid)
        struct p9_client *clnt;
        struct p9_req_t *req;
 
+       if (!fid)
+               return 0;
+
 

While this will solve the NULL pointer dereference, it will do so silently which may lead to us leaking fids/memory/resources/etc. If we were to do such a thing, I'd want warning messages. However, I wouldn't want warning messages in the generic, because now we have places we are calling p9_client_clunk from where we expect null fids to be valid.

I'd suggest keeping the fid check in v9fs_dir_release to parameterize sending the clunk since we expect to sometimes not have a fid here, and then in a separate patch adding some code to p9_client_clunk which complains loudly any time it gets called with a NULL fid. Its unclear to me whether this should be a BUG() or just a warning, a warning would probably suffice as it'll help us track down such cases during testing without breaking users.

-eric
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#2 lore raw
Quoted 18 lines · Eric Van Hensbergen
Quoted 16 lines · Venkateswararao Jujjuri (JV)

NULL fid should be handled in cases where we endup calling v9fs_dir_release() before even we instantiate the fid in filp. This patch handles pasing a NULL p9_fid* to p9_client_clunk.

---
 fs/9p/vfs_dir.c |    3 ++-
 net/9p/client.c |    3 +++
 2 files changed, 5 insertions(+), 1 deletions(-)
fs/9p/vfs_dir.c +0 -0
index 16c8a2a..5f08203 100644
@@ -292,7 +292,8 @@ int v9fs_dir_release(struct inode *inode, struct file *filp)
 
Quoted 7 lines

fs/9p/vfs_dir.c int v9fs_dir_release(struct inode *inode, struct file *filp) · line 296

Quoted 5 lines
fs/9p/vfs_dir.c +2 -1
       fid = filp->private_data;
       P9_DPRINTK(P9_DEBUG_VFS,
-                       "inode: %p filp: %p fid: %d\n", inode, filp, fid->fid);
+                       "JV: inode: %p filp: %p fid: %d\n", inode, filp,
+                       fid ? fid->fid : -1);

Did you really mean to insert a JV: debug label in there?

Oops!!

Quoted 24 lines
Quoted 9 lines
fs/9p/vfs_dir.c +0 -0
       filemap_write_and_wait(inode->i_mapping);
       p9_client_clunk(fid);
       return 0;
net/9p/client.c +0 -0
index dc6f2f2..9338fb3 100644

net/9p/client.c int p9_client_clunk(struct p9_fid *fid) · line 1207

Quoted 7 lines
net/9p/client.c +3 -0
@@ -1201,6 +1201,9 @@ int p9_client_clunk(struct p9_fid *fid)
       struct p9_client *clnt;
       struct p9_req_t *req;
 
+       if (!fid)
+               return 0;
+
 

While this will solve the NULL pointer dereference, it will do so silently which may lead to us leaking fids/memory/resources/etc. If we were to do such a thing, I'd want warning messages. However, I wouldn't want warning messages in the generic, because now we have places we are calling p9_client_clunk from where we expect null fids to be valid.

Quoted 7 lines

I'd suggest keeping the fid check in v9fs_dir_release to parameterize sending the clunk since we expect to sometimes not have a fid here, and then in a separate patch adding some code to p9_client_clunk which complains loudly any time it gets called with a NULL fid. Its unclear to me whether this should be a BUG() or just a warning, a warning would probably suffice as it'll help us track down such cases during testing without breaking users.

So basically you need a conditional call to p9_client_clunk() .

v9fs_dir_release()
{
...
if (fid)

	p9_client_clunk();
}

Do you recall any cases where we end up calling clunk w/o a valid fid? So .. may be we should go with BUG(!fid) in clunck code?

Thanks,
JV

Quoted 2 lines

-eric