Skip to contentSkip to conversation

Crash in dec_zone_page_state when nfs_page req is freed

DISCUSSION Frank Filz 2 participants 5 messages Sep 18, 2006 – Oct 13, 2006

Subsystems (2)
lore
Share & tools
lore

Frank Filz

author
#0 lore raw

I am seeing a crash in dec_zone_page_state when called from

static void nfs_cancel_commit_list(struct list_head *head)

{
	struct nfs_page *req;

	while(!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_inode_remove_request(req);
		nfs_clear_page_writeback(req);
		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
	}
}

I see this was somewhat recently added. It appears that nfs_clear_page_writeback has resulted in the req being freed. We are running with CONFIG_SLAB_DEBUG on which poisons memory with repeated 0x6b bytes when freed so the subsequent reference to req results in a bad wb_page pointer.

Is it the expectation that the request will be freed here? If so, should the dec_zone_page_state call be re-ordered, the wb_page pointer saved before calling nfs_clear_page_writeback, or wb_count being incremented and then adding a call to nfs_release_request?

Frank Filz

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

#1 lore raw
Quoted 15 lines · Frank Filz

I am seeing a crash in dec_zone_page_state when called from

static void nfs_cancel_commit_list(struct list_head *head)

{
	struct nfs_page *req;

	while(!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_inode_remove_request(req);
		nfs_clear_page_writeback(req);
		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
	}
}
Quoted 5 lines

I see this was somewhat recently added. It appears that nfs_clear_page_writeback has resulted in the req being freed. We are running with CONFIG_SLAB_DEBUG on which poisons memory with repeated 0x6b bytes when freed so the subsequent reference to req results in a bad wb_page pointer.

Does the attached patch fix it for you?

Cheers,
Trond

---

 fs/nfs/write.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
fs/nfs/write.c +2 -2
index cf76f9c..38ba5c0 100644
@@ -591,8 +591,8 @@ static void nfs_cancel_commit_list(struc
 		req = nfs_list_entry(head->next);
 		nfs_list_remove_request(req);
 		nfs_inode_remove_request(req);
-		nfs_clear_page_writeback(req);
 		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
+		nfs_clear_page_writeback(req);
 	}
 }
 
@@ -1387,8 +1387,8 @@ nfs_commit_list(struct inode *inode, str
 		req = nfs_list_entry(head->next);
 		nfs_list_remove_request(req);
 		nfs_mark_request_commit(req);
-		nfs_clear_page_writeback(req);
 		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
+		nfs_clear_page_writeback(req);
 	}
 	return -ENOMEM;
 }

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV

_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

author
#2 lore raw
Quoted 16 lines · Trond Myklebust
Quoted 15 lines · Frank Filz

I am seeing a crash in dec_zone_page_state when called from

static void nfs_cancel_commit_list(struct list_head *head)

{
	struct nfs_page *req;

	while(!list_empty(head)) {
		req = nfs_list_entry(head->next);
		nfs_list_remove_request(req);
		nfs_inode_remove_request(req);
		nfs_clear_page_writeback(req);
		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
	}
}
Quoted 7 lines
Quoted 5 lines

I see this was somewhat recently added. It appears that nfs_clear_page_writeback has resulted in the req being freed. We are running with CONFIG_SLAB_DEBUG on which poisons memory with repeated 0x6b bytes when freed so the subsequent reference to req results in a bad wb_page pointer.

Does the attached patch fix it for you?

I've had a chance to do some testing of this. It turns out that nfs_inode_remove_request() will set wb_page to NULL, so it looks like the call to dec_zone_page_state needs to be moved up before nfs_inode_remove_request(). I'm testing that right now, but I start to wonder exactly what the dec_zone_page_state is doing in this case, I know that was a recent addition.

Frank Filz

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

#3 lore raw
Quoted 6 lines · Frank Filz

I've had a chance to do some testing of this. It turns out that nfs_inode_remove_request() will set wb_page to NULL, so it looks like the call to dec_zone_page_state needs to be moved up before nfs_inode_remove_request(). I'm testing that right now, but I start to wonder exactly what the dec_zone_page_state is doing in this case, I know that was a recent addition.

Yup, that makes sense. Furthermore, it looks like the call to nfs_clear_page_writeback is redundant, since nfs_inode_remove_request will cause the tags to be implicitly cleared.

How does the attached patch work for you?

Cheers,
Trond

Fix two bugs:
- nfs_inode_remove_request will call nfs_clear_request, so we cannot reference req->wb_page after it. Move the call to dec_zone_page_state so that it occurs while req->wb_page is still valid.
- Calling nfs_clear_page_writeback is unnecessary since the radix tree tags will have been cleared by the call to nfs_inode_remove_request. Replace with a simple call to nfs_unlock_request.

---

 fs/nfs/write.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
fs/nfs/write.c +2 -2
index f6675d2..d80fa8c 100644
@@ -588,10 +588,10 @@ static void nfs_cancel_commit_list(struc
 
 	while(!list_empty(head)) {
 		req = nfs_list_entry(head->next);
+		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
 		nfs_list_remove_request(req);
 		nfs_inode_remove_request(req);
-		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
-		nfs_clear_page_writeback(req);
+		nfs_unlock_request(req);
 	}
 }
 

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV

_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

author
#4 lore raw
Quoted 8 lines · Trond Myklebust
Quoted 6 lines · Frank Filz

I've had a chance to do some testing of this. It turns out that nfs_inode_remove_request() will set wb_page to NULL, so it looks like the call to dec_zone_page_state needs to be moved up before nfs_inode_remove_request(). I'm testing that right now, but I start to wonder exactly what the dec_zone_page_state is doing in this case, I know that was a recent addition.

Quoted 5 lines

Yup, that makes sense. Furthermore, it looks like the call to nfs_clear_page_writeback is redundant, since nfs_inode_remove_request will cause the tags to be implicitly cleared.

How does the attached patch work for you?

That did the trick. Thanks.

Frank

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs