From: Abhijith Das <adas@redhat.com>

This patch adds a new flag to the gfs2_holder structure GL_FLOCK.
It is set on holders of glocks representing flocks. This flag is
checked in add_to_queue() and a process is permitted to queue more
than one holder onto a glock if it is set. This solves the issue
of a process not being able to do multiple flocks on the same file.
Through a single descriptor, a process can now promote and demote
flocks. Through multiple descriptors a process can now queue
multiple flocks on the same file. There's still the problem of
a process deadlocking itself (because gfs2 blocking locks are not
interruptible) by queueing incompatible deadlock.

Signed-off-by: Abhijith Das <adas@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index 931368a..d631cad 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -1106,24 +1106,31 @@ static void add_to_queue(struct gfs2_holder *gh)
 	if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
 		BUG();
 
-	existing = find_holder_by_owner(&gl->gl_holders, gh->gh_owner_pid);
-	if (existing) {
-		print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
-		printk(KERN_INFO "pid : %d\n", existing->gh_owner_pid);
-		printk(KERN_INFO "lock type : %d lock state : %d\n",
-				existing->gh_gl->gl_name.ln_type, existing->gh_gl->gl_state);
-		print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
-		printk(KERN_INFO "pid : %d\n", gh->gh_owner_pid);
-		printk(KERN_INFO "lock type : %d lock state : %d\n",
-				gl->gl_name.ln_type, gl->gl_state);
-		BUG();
-	}
-
-	existing = find_holder_by_owner(&gl->gl_waiters3, gh->gh_owner_pid);
-	if (existing) {
-		print_symbol(KERN_WARNING "original: %s\n", existing->gh_ip);
-		print_symbol(KERN_WARNING "new: %s\n", gh->gh_ip);
-		BUG();
+	if (!(gh->gh_flags & GL_FLOCK)) {
+		existing = find_holder_by_owner(&gl->gl_holders, 
+						gh->gh_owner_pid);
+		if (existing) {
+			print_symbol(KERN_WARNING "original: %s\n", 
+				   ...
[ message continues ]

From: Wendy Cheng <wcheng@redhat.com>

Move inode deletion code out of blocking_cb handle_callback route to
avoid racy conditions that end up blocking lock_dlm1 thread. Fix
bugzilla 286821.

Signed-off-by: Wendy Cheng <wcheng@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index d631cad..a37efe4 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -716,12 +716,8 @@ static void handle_callback(struct gfs2_glock *gl, unsigned int state,
 		gl->gl_demote_time = jiffies;
 		if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
 		    gl->gl_object) {
-			struct inode *inode = igrab(gl->gl_object);
+			gfs2_glock_schedule_for_reclaim(gl);
 			spin_unlock(&gl->gl_spin);
-			if (inode) {
-				d_prune_aliases(inode);
-				iput(inode);
-			}
 			return;
 		}
 	} else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
-- 
1.5.1.2

-

[ message continues ]

From: Patrick Caulfield <pcaulfie@redhat.com>

Under high recovery loads dlm_sendd can monopolise the CPU and cause soft lockups.

This one extra and one moved cond_resched() make it yield a little more during
such times keeping work moving.

Signed-Off-By: Patrick Caulfield <pcaulfie@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

diff --git a/fs/dlm/lowcomms.c b/fs/dlm/lowcomms.c
index 62a8a6c..58bf3f5 100644
--- a/fs/dlm/lowcomms.c
+++ b/fs/dlm/lowcomms.c
@@ -1264,14 +1264,15 @@ static void send_to_sock(struct connection *con)
 		if (len) {
 			ret = sendpage(con->sock, e->page, offset, len,
 				       msg_flags);
-			if (ret == -EAGAIN || ret == 0)
+			if (ret == -EAGAIN || ret == 0) {
+				cond_resched();
 				goto out;
+			}
 			if (ret <= 0)
 				goto send_error;
-		} else {
+		}
 			/* Don't starve people filling buffers */
 			cond_resched();
-		}
 
 		spin_lock(&con->writequeue_lock);
 		e->offset += ret;
-- 
1.5.1.2

-

[ message continues ]

From: Bob Peterson <rpeterso@redhat.com>

The problem boiled down to a race between the gdlm_init_threads()
function initializing thread1 and its setting of blist = 1.
Essentially, "if (current == ls->thread1)" was checked by the thread
before the thread creator set ls->thread1.

Since thread1 is the only thread who is allowed to work on the
blocking queue, and since neither thread thought it was thread1, no one
was working on the queue.  So everything just sat.

This patch reuses the ls->async_lock spin_lock to fix the race,
and it fixes the problem.  I've done more than 2000 iterations of the
loop that was recreating the failure and it seems to work.

Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

--

diff --git a/fs/gfs2/locking/dlm/thread.c b/fs/gfs2/locking/dlm/thread.c
index 1aca51e..bd938f0 100644
--- a/fs/gfs2/locking/dlm/thread.c
+++ b/fs/gfs2/locking/dlm/thread.c
@@ -268,20 +268,16 @@ static inline int check_drop(struct gdlm_ls *ls)
 	return 0;
 }
 
-static int gdlm_thread(void *data)
+static int gdlm_thread(void *data, int blist)
 {
 	struct gdlm_ls *ls = (struct gdlm_ls *) data;
 	struct gdlm_lock *lp = NULL;
-	int blist = 0;
 	uint8_t complete, blocking, submit, drop;
 	DECLARE_WAITQUEUE(wait, current);
 
 	/* Only thread1 is allowed to do blocking callbacks since gfs
 	   may wait for a completion callback within a blocking cb. */
 
-	if (current == ls->thread1)
-		blist = 1;
-
 	while (!kthread_should_stop()) {
 		set_current_state(TASK_INTERRUPTIBLE);
 		add_wait_queue(&ls->thread_wait, &wait);
@@ -333,12 +329,22 @@ static int gdlm_thread(void *data)
 	return 0;
 }
 
+static int gdlm_thread1(void *data)
+{
+	return gdlm_thread(data, 1);
+}
+
+static int gdlm_thread2(void *data)
+{
+	return gdlm_thread(data, 0);
+}
+
 int gdlm_init_threads(struct gdlm_ls *ls)
 {
 	struct task_struct *p;
 	int error;
 
-	p = kthread_run(gdlm_thread, ls, "lock_dlm1");
+	p = ...
[ message continues ]