2007-05-16 03:53:53

by Dave Kleikamp

[permalink] [raw]
Subject: [PATCH -stable] JFS: Fix race waking up jfsIO kernel thread

This patch fixes a deadlock that has been reported by three different
users on recent kernels. The problem exists on mainline kernels since
2.6.18. It is included in linux-2.6.22-rc1.

JFS: Fix race waking up jfsIO kernel thread

It's possible for a journal I/O request to be added to the log_redrive
queue and the jfsIO thread to be awakened after the thread releases
log_redrive_lock but before it sets its state to TASK_INTERRUPTIBLE.

The jfsIO thread should set the state before giving up the spinlock, so
the waking thread will really wake it.

Signed-off-by: Dave Kleikamp <[email protected]>

diff -Nurp linux-2.6.21.1/fs/jfs/jfs_logmgr.c linux/fs/jfs/jfs_logmgr.c
--- linux-2.6.21.1/fs/jfs/jfs_logmgr.c 2007-04-27 16:49:26.000000000 -0500
+++ linux/fs/jfs/jfs_logmgr.c 2007-05-15 22:39:31.000000000 -0500
@@ -2354,12 +2354,13 @@ int jfsIOWait(void *arg)
lbmStartIO(bp);
spin_lock_irq(&log_redrive_lock);
}
- spin_unlock_irq(&log_redrive_lock);

if (freezing(current)) {
+ spin_unlock_irq(&log_redrive_lock);
refrigerator();
} else {
set_current_state(TASK_INTERRUPTIBLE);
+ spin_unlock_irq(&log_redrive_lock);
schedule();
current->state = TASK_RUNNING;
}

--
David Kleikamp
IBM Linux Technology Center


2007-05-17 01:55:51

by Chris Wright

[permalink] [raw]
Subject: patch jfs-fix-race-waking-up-jfsio-kernel-thread.patch queued to 2.6.21-stable tree


This is a note to let you know that we have just queued up the patch titled

Subject: JFS: Fix race waking up jfsIO kernel thread

to the 2.6.21-stable tree. Its filename is

jfs-fix-race-waking-up-jfsio-kernel-thread.patch

A git repo of this tree can be found at
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary


>From [email protected] Tue May 15 20:55:43 2007
From: Dave Kleikamp <[email protected]>
To: [email protected]
Date: Tue, 15 May 2007 22:53:36 -0500
Message-Id: <[email protected]>
Cc: linux-kernel <[email protected]>
Subject: JFS: Fix race waking up jfsIO kernel thread

It's possible for a journal I/O request to be added to the log_redrive
queue and the jfsIO thread to be awakened after the thread releases
log_redrive_lock but before it sets its state to TASK_INTERRUPTIBLE.

The jfsIO thread should set the state before giving up the spinlock, so
the waking thread will really wake it.

Signed-off-by: Dave Kleikamp <[email protected]>
Signed-off-by: Chris Wright <[email protected]>
---
fs/jfs/jfs_logmgr.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

--- linux-2.6.21.1.orig/fs/jfs/jfs_logmgr.c
+++ linux-2.6.21.1/fs/jfs/jfs_logmgr.c
@@ -2354,12 +2354,13 @@ int jfsIOWait(void *arg)
lbmStartIO(bp);
spin_lock_irq(&log_redrive_lock);
}
- spin_unlock_irq(&log_redrive_lock);

if (freezing(current)) {
+ spin_unlock_irq(&log_redrive_lock);
refrigerator();
} else {
set_current_state(TASK_INTERRUPTIBLE);
+ spin_unlock_irq(&log_redrive_lock);
schedule();
current->state = TASK_RUNNING;
}


Patches currently in stable-queue which might be from [email protected] are

queue-2.6.21/jfs-fix-race-waking-up-jfsio-kernel-thread.patch