jbd: fix race between free buffer and commit transaction
authorMingming Cao <cmm@us.ibm.com>
Fri, 25 Jul 2008 08:46:22 +0000 (01:46 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 25 Jul 2008 17:53:32 +0000 (10:53 -0700)
commit3f31fddfa26b7594b44ff2b34f9a04ba409e0f91
tree88994baf22f65dc4da0bef17ce61eda09c59db2a
parent9ebfbe9f926553eabc21b4400918d1216b27ed0c
jbd: fix race between free buffer and commit transaction

journal_try_to_free_buffers() could race with jbd commit transaction when
the later is holding the buffer reference while waiting for the data
buffer to flush to disk.  If the caller of journal_try_to_free_buffers()
request tries hard to release the buffers, it will treat the failure as
error and return back to the caller.  We have seen the directo IO failed
due to this race.  Some of the caller of releasepage() also expecting the
buffer to be dropped when passed with GFP_KERNEL mask to the
releasepage()->journal_try_to_free_buffers().

With this patch, if the caller is passing the __GFP_WAIT and __GFP_FS to
indicating this call could wait, in case of try_to_free_buffers() failed,
let's waiting for journal_commit_transaction() to finish commit the
current committing transaction, then try to free those buffers again.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Mingming Cao <cmm@us.ibm.com>
Reviewed-by: Badari Pulavarty <pbadari@us.ibm.com>
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/jbd/transaction.c
mm/filemap.c