afs: Fix afs_kill_pages()
authorDavid Howells <dhowells@redhat.com>
Thu, 16 Mar 2017 16:27:48 +0000 (16:27 +0000)
committerDavid Howells <dhowells@redhat.com>
Thu, 16 Mar 2017 16:29:30 +0000 (16:29 +0000)
Fix afs_kill_pages() in two ways:

 (1) If a writeback has been partially flushed, then if we try and kill the
     pages it contains, some of them may no longer be undergoing writeback
     and end_page_writeback() will assert.

     Fix this by checking to see whether the page in question is actually
     undergoing writeback before ending that writeback.

 (2) The loop that scans for pages to kill doesn't increase the first page
     index, and so the loop may not terminate, but it will try to process
     the same pages over and over again.

     Fix this by increasing the first page index to one after the last page
     we processed.

Signed-off-by: David Howells <dhowells@redhat.com>
fs/afs/write.c

index 6e13e96..134de06 100644 (file)
@@ -321,10 +321,14 @@ static void afs_kill_pages(struct afs_vnode *vnode, bool error,
                ASSERTCMP(pv.nr, ==, count);
 
                for (loop = 0; loop < count; loop++) {
-                       ClearPageUptodate(pv.pages[loop]);
+                       struct page *page = pv.pages[loop];
+                       ClearPageUptodate(page);
                        if (error)
-                               SetPageError(pv.pages[loop]);
-                       end_page_writeback(pv.pages[loop]);
+                               SetPageError(page);
+                       if (PageWriteback(page))
+                               end_page_writeback(page);
+                       if (page->index >= first)
+                               first = page->index + 1;
                }
 
                __pagevec_release(&pv);