lkml.org 
[lkml]   [2008]   [Jun]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
Subject[patch 1/2] mm: dont clear PG_uptodate in invalidate_complete_page2()
From: Miklos Szeredi <mszeredi@suse.cz>

Clearing the uptodate page flag will cause page_cache_pipe_buf_confirm()
to return -ENODATA if that page was in the buffer. This in turn will cause
splice() to return a short or zero count.

This manifested itself in rare I/O errors seen on nfs exported fuse
filesystems. This is because nfsd uses splice_direct_to_actor() to
read files, and fuse uses invalidate_inode_pages2() to invalidate
stale data on open.

Fix this by not clearing PG_uptodate on page invalidation. This will
result in the old, invalid page contents being copied. But that's OK,
the contents were valid at splice-in time (which is when the the
"copy" was conceptually done).

I haven't done an audit of all code that checks the PG_uptodate flags,
but I suspect, that this change won't have any harmful effects. Most
code checks page->mapping to see if the page was truncated or
invalidated, before using it, and retries the find/read on the page if
it wasn't. The page_cache_pipe_buf_confirm() code is an exception in
this regard.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
---
mm/truncate.c | 1 -
1 file changed, 1 deletion(-)
Index: linux-2.6/mm/truncate.c
===================================================================
--- linux-2.6.orig/mm/truncate.c 2008-06-24 20:49:25.000000000 +0200
+++ linux-2.6/mm/truncate.c 2008-06-24 23:28:32.000000000 +0200
@@ -356,7 +356,6 @@ invalidate_complete_page2(struct address
BUG_ON(PagePrivate(page));
__remove_from_page_cache(page);
write_unlock_irq(&mapping->tree_lock);
- ClearPageUptodate(page);
page_cache_release(page); /* pagecache ref */
return 1;
failed:
--


\
 
 \ /
  Last update: 2008-06-25 14:45    [from the cache]
©2003-2011 Jasper Spaans