Attachment 297648 Details for Bug 435291 – experimental patch 1 -- don't clean pages as aggressively in invalidate_inode_pages2_range

[patch] experimental patch 1 -- don't clean pages as aggressively in invalidate_inode_pages2_range

0009-BZ-435291-VM-don-t-clean-dirty-pages-in-invalidate.patch (text/plain), 2.19 KB, created by Jeff Layton on 2008-03-11 17:42:14 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Jeff Layton

Created: 2008-03-11 17:42:14 UTC

Size: 2.19 KB

patch

obsolete

>From 615246f3a1d2d0dad69a7eaadc692deba9a6b742 Mon Sep 17 00:00:00 2001
>From: Jeff Layton <jlayton@redhat.com>
>Date: Tue, 11 Mar 2008 12:36:00 -0400
>Subject: [PATCH] BZ#435291: VM: don't clean dirty pages in invalidate_inode_pages2
>
>This delta was cherry-picked from this upstream patch:
>
>	commit fba2591bf4e418b6c3f9f8794c9dd8fe40ae7bd9
>	Author: Linus Torvalds <torvalds@woody.osdl.org>
>	Date:   Wed Dec 20 13:46:42 2006 -0800
>
>	VM: Remove "clear_page_dirty()" and "test_clear_page_dirty()"
>	    functions
>
>The problem is this:
>
>nfs_revalidate_inode() will call nfs_sync_mapping() and then call
>invalidate_inode_pages2() once it returns. This is racy, however. It's
>possible for the page to be dirtied after the writeback occurs.
>
>invalidate_inode_pages2_range() will then unmap the page, and clear the
>dirty bit. Then iff invalidate_complete_page2() returns error do we
>reset the dirty bit and return error. There are many legitimate
>situations where that won't return error, and in those situations the
>data in the now dirty page just ends up being discarded.
>
>This seems wrong to me. If the page is found to be dirty at that point,
>we should be returning error if it can't be properly invalidated. This
>patch gets rid of this clean/redirty logic and just has
>invalidate_complete_page2() be called on the page with the dirty bit
>still set.
>---
> mm/truncate.c |    7 +------
> 1 files changed, 1 insertions(+), 6 deletions(-)
>
>diff --git a/mm/truncate.c b/mm/truncate.c
>index 2599770..9bd0315 100644
>--- a/mm/truncate.c
>+++ b/mm/truncate.c
>@@ -343,7 +343,6 @@ int invalidate_inode_pages2_range(struct address_space *mapping,
> 		for (i = 0; !ret && i < pagevec_count(&pvec); i++) {
> 			struct page *page = pvec.pages[i];
> 			pgoff_t page_index;
>-			int was_dirty;
> 
> 			lock_page(page);
> 			if (page->mapping != mapping) {
>@@ -379,12 +378,8 @@ int invalidate_inode_pages2_range(struct address_space *mapping,
> 					  PAGE_CACHE_SIZE, 0);
> 				}
> 			}
>-			was_dirty = test_clear_page_dirty(page);
>-			if (!invalidate_complete_page2(mapping, page)) {
>-				if (was_dirty)
>-					set_page_dirty(page);
>+			if (!invalidate_complete_page2(mapping, page))
> 				ret = -EIO;
>-			}
> 			unlock_page(page);
> 		}
> 		pagevec_release(&pvec);
>-- 
>1.5.3.6
>

Actions: View | Diff

Attachments on bug 435291: 296231 | 296232 | 297648 | 297649 | 298407 | 306161 | 309133