GFS2: BUG in gfs2_adjust_quota
authorAbhijith Das <adas@redhat.com>
Sun, 4 Jul 2010 05:33:24 +0000 (01:33 -0400)
committerSteven Whitehouse <swhiteho@redhat.com>
Thu, 15 Jul 2010 08:07:16 +0000 (09:07 +0100)
HighMem pages on i686 do not get mapped to the buffer_heads and this was
causing a NULL pointer dereference when we were trying to memset page buffers
to zero.
We now use zero_user() that kmaps the page and directly manipulates page data.
This patch also fixes a boundary condition that was incorrect.

Signed-off-by: Abhi Das <adas@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
fs/gfs2/quota.c

index 49667d68769ef8fe5a82920a79f1a37df216804b..b256d6f24288d6507bef949804e57d3e69c8cce2 100644 (file)
@@ -694,10 +694,8 @@ get_a_page:
                if (!buffer_mapped(bh))
                        goto unlock_out;
                /* If it's a newly allocated disk block for quota, zero it */
-               if (buffer_new(bh)) {
-                       memset(bh->b_data, 0, bh->b_size);
-                       set_buffer_uptodate(bh);
-               }
+               if (buffer_new(bh))
+                       zero_user(page, pos - blocksize, bh->b_size);
        }
 
        if (PageUptodate(page))
@@ -723,7 +721,7 @@ get_a_page:
 
        /* If quota straddles page boundary, we need to update the rest of the
         * quota at the beginning of the next page */
-       if (offset != 0) { /* first page, offset is closer to PAGE_CACHE_SIZE */
+       if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
                ptr = ptr + nbytes;
                nbytes = sizeof(struct gfs2_quota) - nbytes;
                offset = 0;