synctask: introduce synclocks for co-operative locking

This patch introduces a synclocks - co-operative locks for synctasks.
Synctasks yield themselves when a lock cannot be acquired at the time
of the lock call, and the unlocker will wake the yielded locker at
the time of unlock.

The implementation is safe in a multi-threaded syncenv framework.

It is also safe for sharing the lock between non-synctasks. i.e, the
same lock can be used for synchronization between a synctask and
a regular thread. In such a situation, waiting synctasks will yield
themselves while non-synctasks will sleep on a cond variable. The
unlocker (which could be either a synctask or a regular thread) will
wake up any type of lock waiter (synctask or regular).

Usage:

    Declaration and Initialization
    ------------------------------

    synclock_t lock;

    ret = synclock_init (&lock);
    if (ret) {
        /* lock could not be allocated */
    }

   Locking and non-blocking lock attempt
   -------------------------------------

   ret = synclock_trylock (&lock);
   if (ret && (errno == EBUSY)) {
      /* lock is held by someone else */
      return;
   }

   synclock_lock (&lock);
   {
      /* critical section */
   }
   synclock_unlock (&lock);

Change-Id: I081873edb536ddde69a20f4a7dc6558ebf19f5b2
BUG: 763820
Signed-off-by: Anand Avati <avati@redhat.com>
Reviewed-on: http://review.gluster.org/4717
Reviewed-by: Krishnan Parthasarathi <kparthas@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Raghavendra G <raghavendra@gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>

2 files changed, 167 insertions, 1 deletions

diff --git a/libglusterfs/src/syncop.c b/libglusterfs/src/syncop.c
index 115debbfbf7..876977e2a01 100644
--- a/libglusterfs/src/syncop.c
+++ b/libglusterfs/src/syncop.c
@@ -1,5 +1,5 @@
 /*
-  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  Copyright (c) 2008-2013 Red Hat, Inc. <http://www.redhat.com>
   This file is part of GlusterFS.
 
   This file is licensed to you under your choice of the GNU Lesser
@@ -507,6 +507,154 @@ syncenv_new (size_t stacksize)
 }
 
 
+int
+synclock_init (synclock_t *lock)
+{
+	if (!lock)
+		return -1;
+
+	pthread_cond_init (&lock->cond, 0);
+	lock->lock = 0;
+	INIT_LIST_HEAD (&lock->waitq);
+
+	return pthread_mutex_init (&lock->guard, 0);
+}
+
+
+int
+synclock_destroy (synclock_t *lock)
+{
+	if (!lock)
+		return -1;
+
+	pthread_cond_destroy (&lock->cond);
+	return pthread_mutex_destroy (&lock->guard);
+}
+
+
+static int
+__synclock_lock (struct synclock *lock)
+{
+	struct synctask *task = NULL;
+
+	if (!lock)
+		return -1;
+
+	task = synctask_get ();
+
+	while (lock->lock) {
+		if (task) {
+			/* called within a synctask */
+			list_add_tail (&task->waitq, &lock->waitq);
+			{
+				pthread_mutex_unlock (&lock->guard);
+				synctask_yield (task);
+				pthread_mutex_lock (&lock->guard);
+			}
+			list_del_init (&task->waitq);
+		} else {
+			/* called by a non-synctask */
+			pthread_cond_wait (&lock->cond, &lock->guard);
+		}
+	}
+
+	lock->lock = _gf_true;
+	lock->owner = task;
+
+	return 0;
+}
+
+
+int
+synclock_lock (synclock_t *lock)
+{
+	int ret = 0;
+
+	pthread_mutex_lock (&lock->guard);
+	{
+		ret = __synclock_lock (lock);
+	}
+	pthread_mutex_unlock (&lock->guard);
+
+	return ret;
+}
+
+
+int
+synclock_trylock (synclock_t *lock)
+{
+	int ret = 0;
+
+	errno = 0;
+
+	pthread_mutex_lock (&lock->guard);
+	{
+		if (lock->lock) {
+			errno = EBUSY;
+			ret = -1;
+			goto unlock;
+		}
+
+		ret = __synclock_lock (lock);
+	}
+unlock:
+	pthread_mutex_unlock (&lock->guard);
+
+	return ret;
+}
+
+
+static int
+__synclock_unlock (synclock_t *lock)
+{
+	struct synctask *task = NULL;
+	struct synctask *curr = NULL;
+
+	if (!lock)
+		return -1;
+
+	curr = synctask_get ();
+
+	if (lock->owner != curr) {
+		/* warn ? */
+	}
+
+	lock->lock = _gf_false;
+
+	/* There could be both synctasks and non synctasks
+	   waiting (or none, or either). As a mid-approach
+	   between maintaining too many waiting counters
+	   at one extreme and a thundering herd on unlock
+	   at the other, call a cond_signal (which wakes
+	   one waiter) and first synctask waiter. So at
+	   most we have two threads waking up to grab the
+	   just released lock.
+	*/
+	pthread_cond_signal (&lock->cond);
+	if (!list_empty (&lock->waitq)) {
+		task = list_entry (lock->waitq.next, struct synctask, waitq);
+		synctask_wake (task);
+	}
+
+	return 0;
+}
+
+
+int
+synclock_unlock (synclock_t *lock)
+{
+	int ret = 0;
+
+	pthread_mutex_lock (&lock->guard);
+	{
+		ret = __synclock_unlock (lock);
+	}
+	pthread_mutex_unlock (&lock->guard);
+
+	return ret;
+}
+
+
 /* FOPS */
 
 
diff --git a/libglusterfs/src/syncop.h b/libglusterfs/src/syncop.h
index aec07f36050..ee8a3277d36 100644
--- a/libglusterfs/src/syncop.h
+++ b/libglusterfs/src/syncop.h
@@ -69,6 +69,8 @@ struct synctask {
         pthread_mutex_t     mutex; /* for synchronous spawning of synctask */
         pthread_cond_t      cond;
         int                 done;
+
+	struct list_head    waitq; /* can wait only "once" at a time */
 };
 
 
@@ -96,6 +98,15 @@ struct syncenv {
 };
 
 
+struct synclock {
+	pthread_mutex_t     guard; /* guard the remaining members, pair @cond */
+	pthread_cond_t      cond;  /* waiting non-synctasks */
+	struct list_head    waitq; /* waiting synctasks */
+	gf_boolean_t        lock;  /* _gf_true or _gf_false, lock status */
+	struct synctask    *owner; /* NULL if current owner is not a synctask */
+};
+typedef struct synclock synclock_t;
+
 struct syncargs {
         int                 op_ret;
         int                 op_errno;
@@ -220,6 +231,13 @@ void synctask_waitfor (struct synctask *task, int count);
 int synctask_setid (struct synctask *task, uid_t uid, gid_t gid);
 #define SYNCTASK_SETID(uid, gid) synctask_setid (synctask_get(), uid, gid);
 
+
+int synclock_init (synclock_t *lock);
+int synclock_destory (synclock_t *lock);
+int synclock_lock (synclock_t *lock);
+int synclock_trylock (synclock_t *lock);
+int synclock_unlock (synclock_t *lock);
+
 int syncop_lookup (xlator_t *subvol, loc_t *loc, dict_t *xattr_req,
                    /* out */
                    struct iatt *iatt, dict_t **xattr_rsp, struct iatt *parent);