Skip to contentSkip to conversation

FAILED: patch "[PATCH] ubifs: Fix deadlock in concurrent bulk-read and writepage" failed to apply to 4.9-stable tree

PATCH [email protected] 2 participants 2 messages Feb 6, 2020 – Feb 7, 2020

Subsystems (2)
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

The patch below does not apply to the 4.9-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From f5de5b83303e61b1f3fb09bd77ce3ac2d7a475f2 Mon Sep 17 00:00:00 2001
From: Zhihao Cheng <[email protected]>
Date: Sat, 11 Jan 2020 17:50:36 +0800
Subject: [PATCH] ubifs: Fix deadlock in concurrent bulk-read and writepage
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

In ubifs, concurrent execution of writepage and bulk read on the same file may cause ABBA deadlock, for example (Reproduce method see Link):

Process A(Bulk-read starts from page4) Process B(write page4 back)

  vfs_read                                       wb_workfn or fsync
  ...                                            ...
  generic_file_buffered_read                     write_cache_pages

ubifs_readpage LOCK(page4)

ubifs_bulk_read ubifs_writepage
LOCK(ui->ui_mutex) ubifs_write_inode

ubifs_do_bulk_read LOCK(ui->ui_mutex)
find_or_create_page(alloc page4) ↑
LOCK(page4) <-- ABBA deadlock occurs!

In order to ensure the serialization execution of bulk read, we can't remove the big lock 'ui->ui_mutex' in ubifs_bulk_read(). Instead, we allow ubifs_do_bulk_read() to lock page failed by replacing find_or_create_page(FGP_LOCK) with pagecache_get_page(FGP_LOCK | FGP_NOWAIT).

fs/ubifs/file.c +3 -1
index cd52585c8f4f..c649048fcc81 100644
@@ -786,7 +786,9 @@ static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
 
 		if (page_offset > end_index)
 			break;
-		page = find_or_create_page(mapping, page_offset, ra_gfp_mask);
+		page = pagecache_get_page(mapping, page_offset,
+				 FGP_LOCK|FGP_ACCESSED|FGP_CREAT|FGP_NOWAIT,
+				 ra_gfp_mask);
 		if (!page)
 			break;
 		if (!PageUptodate(page))
maintainer
#1 lore raw
Quoted 42 lines · On Thu, Feb 06, 2020 at 08:20:48PM +0100, [email protected] wrote:

The patch below does not apply to the 4.9-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From f5de5b83303e61b1f3fb09bd77ce3ac2d7a475f2 Mon Sep 17 00:00:00 2001
From: Zhihao Cheng <[email protected]>
Date: Sat, 11 Jan 2020 17:50:36 +0800
Subject: [PATCH] ubifs: Fix deadlock in concurrent bulk-read and writepage
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

In ubifs, concurrent execution of writepage and bulk read on the same file may cause ABBA deadlock, for example (Reproduce method see Link):

Process A(Bulk-read starts from page4) Process B(write page4 back)

 vfs_read                                       wb_workfn or fsync
 ...                                            ...
 generic_file_buffered_read                     write_cache_pages

ubifs_readpage LOCK(page4)

ubifs_bulk_read ubifs_writepage
LOCK(ui->ui_mutex) ubifs_write_inode

ubifs_do_bulk_read LOCK(ui->ui_mutex)
find_or_create_page(alloc page4) ↑
LOCK(page4) <-- ABBA deadlock occurs!

In order to ensure the serialization execution of bulk read, we can't remove the big lock 'ui->ui_mutex' in ubifs_bulk_read(). Instead, we allow ubifs_do_bulk_read() to lock page failed by replacing find_or_create_page(FGP_LOCK) with pagecache_get_page(FGP_LOCK | FGP_NOWAIT).

Quoted 6 lines

I took in 480a1a6a3ef6 ("ubifs: Change gfp flags in page allocation for bulk read") as dependency and queued both for 4.9 and 4.4.

Signature · 2 lines
Thanks,
Sasha