[lkml]   [2007]   [Oct]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Patch in this message
SubjectRe: [PATCH] Fix bad data from non-direct-io read after direct-io write

Hmm. If I read this right, this bug seems to have been introduced by
commit 65b8291c4000e5f38fc94fb2ca0cb7e8683c8a1b ("dio: invalidate clean
pages before dio write") back in March.

Before that, we'd call invalidate_inode_pages2_range() unconditionally
after the call mapping->a_ops->direct_IO() if it was a write and there
were cached pages in the mapping (well, "unconditionally" in the sense
that it didn't depend on the return value of the ->direct_IO() call).

However, with both the old and the new code _and_ with your patch, the
return code - in case the invalidate failed - was corrupted. So we may
actually end up doing some IO, but then returning the "wrong" error code
from the invalidate. Hmm?

Somebody who cares about direct-IO and who - unlike me - doesn't think
it's a total and idiotic crock should think hard about this. I'm including
Karl's email, but also an alternate patch for consideration.

And maybe some day we can all agree that direct_IO is crap and should not
be done.


diff --git a/mm/filemap.c b/mm/filemap.c
index 5209e47..032371a 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -2510,22 +2510,17 @@ generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,

retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
- if (retval)
+ if (retval < 0)
goto out;

* Finally, try again to invalidate clean pages which might have been
* faulted in by get_user_pages() if the source of the write was an
* mmap()ed region of the file we're writing. That's a pretty crazy
- * thing to do, so we don't support it 100%. If this invalidation
- * fails and we have -EIOCBQUEUED we ignore the failure.
+ * thing to do, so we don't support it 100%.
- if (rw == WRITE && mapping->nrpages) {
- int err = invalidate_inode_pages2_range(mapping,
- offset >> PAGE_CACHE_SHIFT, end);
- if (err && retval >= 0)
- retval = err;
- }
+ if (rw == WRITE && mapping->nrpages)
+ invalidate_inode_pages2_range(mapping, offset >> PAGE_CACHE_SHIFT, end);
return retval;
On Fri, 26 Oct 2007, Karl Schendel wrote:
> This patch fixes a race between direct IO writes and non-direct IO
> reads on the same file. The symptom is a stale file page seen by
> any non-direct-IO reader, which persists until the page is invalidated
> somehow (e.g. page rewritten again, or memory pressure, or reboot).
> An improper return test caused direct-IO's after-write page invalidations
> to be skipped. If we're writing page N, and the reader is reading
> page N-x for small x, and the read code decides to readahead, it's
> not too hard to cause a race that leaves an old, stale copy of the
> page in the page cache. Retval is usually +nonzero after the
> mapping->a_ops->direct_IO call!
> Signed-off-by: Karl Schendel <>
> ---
> By the way, I agree that the userland situation is stupid, and I'm
> addressing that in the application (happens to be the Ingres DBMS).
> However, the kernel shouldn't compound the stupidity.
> I'll try to watch for replies, but it would be very useful to
> cc me at if any discussion is needed;
> I'm not subscribed to lkml.
> --- linux- 2007-10-12 12:43:44.000000000 -0400
> +++ linux- 2007-10-26 16:12:08.000000000 -0400
> @@ -2194,7 +2194,7 @@ generic_file_direct_IO(int rw, struct ki
> }
> retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs);
> - if (retval)
> + if (retval < 0)
> goto out;
> /*
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2007-10-26 23:37    [W:0.052 / U:0.864 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site