2.6.23.1 regression: write() data lost after lseek()

2.6.23.1 regression: write() data lost after lseek()

[Frank van Maarseveen]

2.6.23.1 client (using 2.6.22.10 and 2.6.21.7 as server), mount options
according to /proc/mounts:

rw,nodev,vers=3,rsize=32768,wsize=32768,hard,intr,proto=udp,timeo=11,retrans=2,sec=sys


The data of the last write in this program gets lost when run on 2.6.23.1 client:

--------
#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <errno.h>

int main(int argc, char **argv)
{
	int d;

	errno = 0;
	d = open(argv[1], O_CREAT|O_WRONLY|O_TRUNC, 0666);
	write(d, "a", 1);
	lseek(d, 2, SEEK_SET);
	write(d, "data lost\n", 10);
	close(d);
	perror("");
	return errno ? 1 : 0;
}
--------

$ ./prog out
Success
$ od -c out
0000000   a  \0  \0  \0  \0  \0  \0  \0  \0  \0  \0  \0
0000014


-- 
Frank

-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: 2.6.23.1 regression: write() data lost after lseek()

[Trond Myklebust]

[-- Attachment #1: Type: text/plain, Size: 434 bytes --]


On Fri, 2007-10-26 at 17:50 +0200, Frank van Maarseveen wrote:
> 2.6.23.1 client (using 2.6.22.10 and 2.6.21.7 as server), mount options
> according to /proc/mounts:
> 
> rw,nodev,vers=3,rsize=32768,wsize=32768,hard,intr,proto=udp,timeo=11,retrans=2,sec=sys
> 
> 
> The data of the last write in this program gets lost when run on 2.6.23.1 client:

Known issue. I have already submitted the attached fix to
stable@kernel.org.

Trond

[-- Attachment #2: linux-2.6.23-133-fix_write_race.dif --]
[-- Type: message/rfc822, Size: 2436 bytes --]

From: Trond Myklebust <Trond.Myklebust@netapp.com>
Subject: No Subject
Date: Thu, 18 Oct 2007 17:08:05 -0400
Message-ID: <1193417629.7486.12.camel@heimdal.trondhjem.org>

This patch fixes a regression that was introduced by commit
44dd151d5c21234cc534c47d7382f5c28c3143cd

We cannot zero the user page in nfs_mark_uptodate() any more, since

  a) We'd be modifying the page without holding the page lock
  b) We can race with other updates of the page, most notably
     because of the call to nfs_wb_page() in nfs_writepage_setup().

Instead, we do the zeroing in nfs_update_request() if we see that we're
creating a request that might potentially be marked as up to date.

Thanks to Olivier Paquet for reporting the bug and providing a test-case.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---

 fs/nfs/write.c |   17 +++++++++++++----
 1 files changed, 13 insertions(+), 4 deletions(-)

diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index e2bb66c..f418f6e 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -174,8 +174,6 @@ static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int
 		return;
 	if (count != nfs_page_length(page))
 		return;
-	if (count != PAGE_CACHE_SIZE)
-		zero_user_page(page, count, PAGE_CACHE_SIZE - count, KM_USER0);
 	SetPageUptodate(page);
 }
 
@@ -626,7 +624,8 @@ static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx,
 				return ERR_PTR(error);
 			}
 			spin_unlock(&inode->i_lock);
-			return new;
+			req = new;
+			goto zero_page;
 		}
 		spin_unlock(&inode->i_lock);
 
@@ -654,13 +653,23 @@ static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx,
 	if (offset < req->wb_offset) {
 		req->wb_offset = offset;
 		req->wb_pgbase = offset;
-		req->wb_bytes = rqend - req->wb_offset;
+		req->wb_bytes = max(end, rqend) - req->wb_offset;
+		goto zero_page;
 	}
 
 	if (end > rqend)
 		req->wb_bytes = end - req->wb_offset;
 
 	return req;
+zero_page:
+	/* If this page might potentially be marked as up to date,
+	 * then we need to zero any uninitalised data. */
+	if (req->wb_pgbase == 0 && req->wb_bytes != PAGE_CACHE_SIZE
+			&& !PageUptodate(req->wb_page))
+		zero_user_page(req->wb_page, req->wb_bytes,
+				PAGE_CACHE_SIZE - req->wb_bytes,
+				KM_USER0);
+	return req;
 }
 
 int nfs_flush_incompatible(struct file *file, struct page *page)

[-- Attachment #3: Type: text/plain, Size: 314 bytes --]

-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >> http://get.splunk.com/

[-- Attachment #4: Type: text/plain, Size: 140 bytes --]

_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs