mm/fadvise.c: do not discard partial pages with POSIX_FADV_DONTNEED
authorOleg Drokin <green@linuxhacker.ru>
Wed, 8 Jun 2016 22:33:59 +0000 (15:33 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 9 Jun 2016 21:23:11 +0000 (14:23 -0700)
I noticed that the logic in the fadvise64_64 syscall is incorrect for
partial pages.  While first page of the region is correctly skipped if
it is partial, the last page of the region is mistakenly discarded.
This leads to problems for applications that read data in
non-page-aligned chunks discarding already processed data between the
reads.

A somewhat misguided application that does something like write(XX bytes
(non-page-alligned)); drop the data it just wrote; repeat gets a
significant penalty in performance as a result.

Link: http://lkml.kernel.org/r/1464917140-1506698-1-git-send-email-green@linuxhacker.ru
Signed-off-by: Oleg Drokin <green@linuxhacker.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/fadvise.c

index b8024fa7101d9a8ff625606de973b9b670d0ecf4..6c707bfe02fde002feae9ea2fab3fc9647c3baa0 100644 (file)
@@ -126,6 +126,17 @@ SYSCALL_DEFINE4(fadvise64_64, int, fd, loff_t, offset, loff_t, len, int, advice)
                 */
                start_index = (offset+(PAGE_SIZE-1)) >> PAGE_SHIFT;
                end_index = (endbyte >> PAGE_SHIFT);
+               if ((endbyte & ~PAGE_MASK) != ~PAGE_MASK) {
+                       /* First page is tricky as 0 - 1 = -1, but pgoff_t
+                        * is unsigned, so the end_index >= start_index
+                        * check below would be true and we'll discard the whole
+                        * file cache which is not what was asked.
+                        */
+                       if (end_index == 0)
+                               break;
+
+                       end_index--;
+               }
 
                if (end_index >= start_index) {
                        unsigned long count = invalidate_mapping_pages(mapping,