[PATCH] mm: bug in set_page_dirty_buffers
authorNick Piggin <npiggin@suse.de>
Tue, 10 Oct 2006 02:36:54 +0000 (04:36 +0200)
committerLinus Torvalds <torvalds@g5.osdl.org>
Tue, 10 Oct 2006 02:51:18 +0000 (19:51 -0700)
This was triggered, but not the fault of, the dirty page accounting
patches. Suitable for -stable as well, after it goes upstream.

  Unable to handle kernel NULL pointer dereference at virtual address 0000004c
  EIP is at _spin_lock+0x12/0x66
  Call Trace:
   [<401766e7>] __set_page_dirty_buffers+0x15/0xc0
   [<401401e7>] set_page_dirty+0x2c/0x51
   [<40140db2>] set_page_dirty_balance+0xb/0x3b
   [<40145d29>] __do_fault+0x1d8/0x279
   [<40147059>] __handle_mm_fault+0x125/0x951
   [<401133f1>] do_page_fault+0x440/0x59f
   [<4034d0c1>] error_code+0x39/0x40
   [<08048a33>] 0x8048a33

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
fs/buffer.c

index 16cfbcd254f15fa05eb452eb512297c439605035..eeb8ac1aa8561f241ff007fcf3c4bd943ffffdb8 100644 (file)
@@ -701,7 +701,10 @@ EXPORT_SYMBOL(mark_buffer_dirty_inode);
  */
 int __set_page_dirty_buffers(struct page *page)
 {
-       struct address_space * const mapping = page->mapping;
+       struct address_space * const mapping = page_mapping(page);
+
+       if (unlikely(!mapping))
+               return !TestSetPageDirty(page);
 
        spin_lock(&mapping->private_lock);
        if (page_has_buffers(page)) {