[lkml]   [2000]   [Dec]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Subject[found?] Re: kernel BUG at buffer.c:827 in test12-pre6 and 7
	I'm doing some massive grepping (basically, audit of page locking),
but nothing relevant so far. There was some catch (aside of documenting
the thing and finding several completely unrelated buglets):
* ramfs_writepage() doesn't UnlockPage(). Deadlock.
* udf_adinicb_writepage() does extra UnlockPage().
I don't see the fsckup in loop.c, though. On the read path it uses
do_generic_read_file() and on the write it's essentially the simplified
variant of generic_file_write(). Hell knows... It looks like we are
getting dirty buffer inheriting end_buffer_io_async from the previous

Oh, damn it. Indeed. Look:

generic_file_write() or lo_send():
->prepare_write() sets sync ->b_end_io
->commit_write() puts them on the dirty list
UnlockPage() releases the page lock
... requests are sent to driver

TryLockPage() succeeds
... goes through the bh'es and sets ->b_end_io to end_buffer_io_async()
at that point the last remaining request completes. It calls
->b_end_io(), aka. end_buffer_io_async(). And does UnlockPage().

In the meanwhile, we call ll_rw_block(). Requests are sent again.
When _they_ complete we get the second UnlockPage()

Now, I might miss some obvious reason why it could never happen. Moreover,
the real problem may be completely different - the race above is not too wide.

However, I'ld really like to hear why the scenario above is impossible. BTW,
the race isn't even that narrow - if ->prepare_write() didn't cover the
whole page we've got a get_block() to call and there's a plenty of time when
shit can hit the fan - it's a blocking operation, after all.


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
Please read the FAQ at

 \ /
  Last update: 2005-03-22 12:51    [W:0.052 / U:0.428 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site