cifs: fix writeback race with file that is growing
Jeff Layton [Mon, 26 Nov 2012 14:48:41 +0000 (09:48 -0500)]
Commit eddb079deb4 created a regression in the writepages codepath.
Previously, whenever it needed to check the size of the file, it did so
by consulting the inode->i_size field directly. With that patch, the
i_size was fetched once on entry into the writepages code and that value
was used henceforth.

If the file is changing size though (for instance, if someone is writing
to it or has truncated it), then that value is likely to be wrong. This
can lead to data corruption. Pages past the EOF at the time that the
writepages call was issued may be silently dropped and ignored because
cifs_writepages wrongly assumes that the file must have been truncated
in the interim.

Fix cifs_writepages to properly fetch the size from the inode->i_size
field instead to properly account for this possibility.

Original bug report is here:

    https://bugzilla.kernel.org/show_bug.cgi?id=50991

Reported-and-Tested-by: Maxim Britov <ungifted01@gmail.com>
Reviewed-by: Suresh Jayaraman <sjayaraman@suse.com>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <smfrench@gmail.com>

fs/cifs/file.c

index edb25b4..70b6f4c 100644 (file)
@@ -1794,7 +1794,6 @@ static int cifs_writepages(struct address_space *mapping,
        struct TCP_Server_Info *server;
        struct page *page;
        int rc = 0;
-       loff_t isize = i_size_read(mapping->host);
 
        /*
         * If wsize is smaller than the page cache size, default to writing
@@ -1899,7 +1898,7 @@ retry:
                         */
                        set_page_writeback(page);
 
-                       if (page_offset(page) >= isize) {
+                       if (page_offset(page) >= i_size_read(mapping->host)) {
                                done = true;
                                unlock_page(page);
                                end_page_writeback(page);
@@ -1932,7 +1931,8 @@ retry:
                wdata->offset = page_offset(wdata->pages[0]);
                wdata->pagesz = PAGE_CACHE_SIZE;
                wdata->tailsz =
-                       min(isize - page_offset(wdata->pages[nr_pages - 1]),
+                       min(i_size_read(mapping->host) -
+                           page_offset(wdata->pages[nr_pages - 1]),
                            (loff_t)PAGE_CACHE_SIZE);
                wdata->bytes = ((nr_pages - 1) * PAGE_CACHE_SIZE) +
                                        wdata->tailsz;