vfs: Fix vmtruncate() regression
authorOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Wed, 13 Jan 2010 12:14:09 +0000 (21:14 +0900)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 14 Jan 2010 00:09:33 +0000 (16:09 -0800)
If __block_prepare_write() was failed in block_write_begin(), the
allocated blocks can be outside of ->i_size.

But new truncate_pagecache() in vmtuncate() does nothing if new < old.
It means the above usage is not working anymore.

So, this patch fixes it by removing "new < old" check. It would need
more cleanup/change. But, now -rc and truncate working is in progress,
so, this tried to fix it minimum change.

Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/truncate.c

index 342deee22684ffdfded563506d684ef0f2b18191..e87e37244829ce03abc22673cda7eeb37f6fff0a 100644 (file)
@@ -522,22 +522,20 @@ EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
  */
 void truncate_pagecache(struct inode *inode, loff_t old, loff_t new)
 {
-       if (new < old) {
-               struct address_space *mapping = inode->i_mapping;
-
-               /*
-                * unmap_mapping_range is called twice, first simply for
-                * efficiency so that truncate_inode_pages does fewer
-                * single-page unmaps.  However after this first call, and
-                * before truncate_inode_pages finishes, it is possible for
-                * private pages to be COWed, which remain after
-                * truncate_inode_pages finishes, hence the second
-                * unmap_mapping_range call must be made for correctness.
-                */
-               unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
-               truncate_inode_pages(mapping, new);
-               unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
-       }
+       struct address_space *mapping = inode->i_mapping;
+
+       /*
+        * unmap_mapping_range is called twice, first simply for
+        * efficiency so that truncate_inode_pages does fewer
+        * single-page unmaps.  However after this first call, and
+        * before truncate_inode_pages finishes, it is possible for
+        * private pages to be COWed, which remain after
+        * truncate_inode_pages finishes, hence the second
+        * unmap_mapping_range call must be made for correctness.
+        */
+       unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
+       truncate_inode_pages(mapping, new);
+       unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
 }
 EXPORT_SYMBOL(truncate_pagecache);