GFS2: Pre-pull patch posting (merge window)

GFS2: Pre-pull patch posting (merge window)

[Steven Whitehouse]

Hi,

The most interesting "feature" in this patch set is the RCU glock
patch which has been a long time coming, but is finally here. That
patch contains most of the changes this time. The other patches ins
this set are mostly smaller bug fixes and performance improvements.

Steve.

[PATCH 01/15] GFS2: Use RCU for glock hash table

[Steven Whitehouse]

This has a number of advantages:

 - Reduces contention on the hash table lock
 - Makes the code smaller and simpler
 - Should speed up glock dumps when under load
 - Removes ref count changing in examine_bucket
 - No longer need hash chain lock in glock_put() in common case

There are some further changes which this enables and which
we may do in the future. One is to look at using SLAB_RCU,
and another is to look at using a per-cpu counter for the
per-sb glock counter, since that is touched twice in the
lifetime of each glock (but only used at umount time).

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index 08a8beb..c75d499 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -26,6 +26,9 @@
 #include <linux/freezer.h>
 #include <linux/workqueue.h>
 #include <linux/jiffies.h>
+#include <linux/rcupdate.h>
+#include <linux/rculist_bl.h>
+#include <linux/bit_spinlock.h>
 
 #include "gfs2.h"
 #include "incore.h"
@@ -41,10 +44,6 @@
 #define CREATE_TRACE_POINTS
 #include "trace_gfs2.h"
 
-struct gfs2_gl_hash_bucket {
-        struct hlist_head hb_list;
-};
-
 struct gfs2_glock_iter {
 	int hash;			/* hash bucket index         */
 	struct gfs2_sbd *sdp;		/* incore superblock         */
@@ -54,7 +53,6 @@ struct gfs2_glock_iter {
 
 typedef void (*glock_examiner) (struct gfs2_glock * gl);
 
-static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
 static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl);
 #define GLOCK_BUG_ON(gl,x) do { if (unlikely(x)) { __dump_glock(NULL, gl); BUG(); } } while(0)
 static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target);
@@ -70,57 +68,9 @@ static DEFINE_SPINLOCK(lru_lock);
 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
 #define GFS2_GL_HASH_MASK       (GFS2_GL_HASH_SIZE - 1)
 
-static struct gfs2_gl_hash_bucket gl_hash_table[GFS2_GL_HASH_SIZE];
+static struct hlist_bl_head gl_hash_table[GFS2_GL_HASH_SIZE];
 static struct dentry *gfs2_root;
 
-/*
- * Despite what you might think, the numbers below are not arbitrary :-)
- * They are taken from the ipv4 routing hash code, which is well tested
- * and thus should be nearly optimal. Later on we might tweek the numbers
- * but for now this should be fine.
- *
- * The reason for putting the locks in a separate array from the list heads
- * is that we can have fewer locks than list heads and save memory. We use
- * the same hash function for both, but with a different hash mask.
- */
-#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
-	defined(CONFIG_PROVE_LOCKING)
-
-#ifdef CONFIG_LOCKDEP
-# define GL_HASH_LOCK_SZ        256
-#else
-# if NR_CPUS >= 32
-#  define GL_HASH_LOCK_SZ       4096
-# elif NR_CPUS >= 16
-#  define GL_HASH_LOCK_SZ       2048
-# elif NR_CPUS >= 8
-#  define GL_HASH_LOCK_SZ       1024
-# elif NR_CPUS >= 4
-#  define GL_HASH_LOCK_SZ       512
-# else
-#  define GL_HASH_LOCK_SZ       256
-# endif
-#endif
-
-/* We never want more locks than chains */
-#if GFS2_GL_HASH_SIZE < GL_HASH_LOCK_SZ
-# undef GL_HASH_LOCK_SZ
-# define GL_HASH_LOCK_SZ GFS2_GL_HASH_SIZE
-#endif
-
-static rwlock_t gl_hash_locks[GL_HASH_LOCK_SZ];
-
-static inline rwlock_t *gl_lock_addr(unsigned int x)
-{
-	return &gl_hash_locks[x & (GL_HASH_LOCK_SZ-1)];
-}
-#else /* not SMP, so no spinlocks required */
-static inline rwlock_t *gl_lock_addr(unsigned int x)
-{
-	return NULL;
-}
-#endif
-
 /**
  * gl_hash() - Turn glock number into hash bucket number
  * @lock: The glock number
@@ -141,25 +91,30 @@ static unsigned int gl_hash(const struct gfs2_sbd *sdp,
 	return h;
 }
 
-/**
- * glock_free() - Perform a few checks and then release struct gfs2_glock
- * @gl: The glock to release
- *
- * Also calls lock module to release its internal structure for this glock.
- *
- */
+static inline void spin_lock_bucket(unsigned int hash)
+{
+	struct hlist_bl_head *bl = &gl_hash_table[hash];
+	bit_spin_lock(0, (unsigned long *)bl);
+}
+
+static inline void spin_unlock_bucket(unsigned int hash)
+{
+	struct hlist_bl_head *bl = &gl_hash_table[hash];
+	__bit_spin_unlock(0, (unsigned long *)bl);
+}
 
-static void glock_free(struct gfs2_glock *gl)
+void gfs2_glock_free(struct rcu_head *rcu)
 {
+	struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
 	struct gfs2_sbd *sdp = gl->gl_sbd;
-	struct address_space *mapping = gfs2_glock2aspace(gl);
-	struct kmem_cache *cachep = gfs2_glock_cachep;
 
-	GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
-	trace_gfs2_glock_put(gl);
-	if (mapping)
-		cachep = gfs2_glock_aspace_cachep;
-	sdp->sd_lockstruct.ls_ops->lm_put_lock(cachep, gl);
+	if (gl->gl_ops->go_flags & GLOF_ASPACE)
+		kmem_cache_free(gfs2_glock_aspace_cachep, gl);
+	else
+		kmem_cache_free(gfs2_glock_cachep, gl);
+
+	if (atomic_dec_and_test(&sdp->sd_glock_disposal))
+		wake_up(&sdp->sd_glock_wait);
 }
 
 /**
@@ -185,34 +140,49 @@ static int demote_ok(const struct gfs2_glock *gl)
 {
 	const struct gfs2_glock_operations *glops = gl->gl_ops;
 
+	/* assert_spin_locked(&gl->gl_spin); */
+
 	if (gl->gl_state == LM_ST_UNLOCKED)
 		return 0;
-	if (!list_empty(&gl->gl_holders))
+	if (test_bit(GLF_LFLUSH, &gl->gl_flags))
+		return 0;
+	if ((gl->gl_name.ln_type != LM_TYPE_INODE) &&
+	    !list_empty(&gl->gl_holders))
 		return 0;
 	if (glops->go_demote_ok)
 		return glops->go_demote_ok(gl);
 	return 1;
 }
 
+
 /**
- * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
+ * __gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
  * @gl: the glock
  *
+ * If the glock is demotable, then we add it (or move it) to the end
+ * of the glock LRU list.
  */
 
-static void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
+static void __gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
 {
-	int may_reclaim;
-	may_reclaim = (demote_ok(gl) &&
-		       (atomic_read(&gl->gl_ref) == 1 ||
-			(gl->gl_name.ln_type == LM_TYPE_INODE &&
-			 atomic_read(&gl->gl_ref) <= 2)));
-	spin_lock(&lru_lock);
-	if (list_empty(&gl->gl_lru) && may_reclaim) {
+	if (demote_ok(gl)) {
+		spin_lock(&lru_lock);
+
+		if (!list_empty(&gl->gl_lru))
+			list_del_init(&gl->gl_lru);
+		else
+			atomic_inc(&lru_count);
+
 		list_add_tail(&gl->gl_lru, &lru_list);
-		atomic_inc(&lru_count);
+		spin_unlock(&lru_lock);
 	}
-	spin_unlock(&lru_lock);
+}
+
+void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
+{
+	spin_lock(&gl->gl_spin);
+	__gfs2_glock_schedule_for_reclaim(gl);
+	spin_unlock(&gl->gl_spin);
 }
 
 /**
@@ -227,7 +197,6 @@ void gfs2_glock_put_nolock(struct gfs2_glock *gl)
 {
 	if (atomic_dec_and_test(&gl->gl_ref))
 		GLOCK_BUG_ON(gl, 1);
-	gfs2_glock_schedule_for_reclaim(gl);
 }
 
 /**
@@ -236,30 +205,26 @@ void gfs2_glock_put_nolock(struct gfs2_glock *gl)
  *
  */
 
-int gfs2_glock_put(struct gfs2_glock *gl)
+void gfs2_glock_put(struct gfs2_glock *gl)
 {
-	int rv = 0;
+	struct gfs2_sbd *sdp = gl->gl_sbd;
+	struct address_space *mapping = gfs2_glock2aspace(gl);
 
-	write_lock(gl_lock_addr(gl->gl_hash));
-	if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
-		hlist_del(&gl->gl_list);
+	if (atomic_dec_and_test(&gl->gl_ref)) {
+		spin_lock_bucket(gl->gl_hash);
+		hlist_bl_del_rcu(&gl->gl_list);
+		spin_unlock_bucket(gl->gl_hash);
+		spin_lock(&lru_lock);
 		if (!list_empty(&gl->gl_lru)) {
 			list_del_init(&gl->gl_lru);
 			atomic_dec(&lru_count);
 		}
 		spin_unlock(&lru_lock);
-		write_unlock(gl_lock_addr(gl->gl_hash));
 		GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
-		glock_free(gl);
-		rv = 1;
-		goto out;
+		GLOCK_BUG_ON(gl, mapping && mapping->nrpages);
+		trace_gfs2_glock_put(gl);
+		sdp->sd_lockstruct.ls_ops->lm_put_lock(gl);
 	}
-	spin_lock(&gl->gl_spin);
-	gfs2_glock_schedule_for_reclaim(gl);
-	spin_unlock(&gl->gl_spin);
-	write_unlock(gl_lock_addr(gl->gl_hash));
-out:
-	return rv;
 }
 
 /**
@@ -275,17 +240,15 @@ static struct gfs2_glock *search_bucket(unsigned int hash,
 					const struct lm_lockname *name)
 {
 	struct gfs2_glock *gl;
-	struct hlist_node *h;
+	struct hlist_bl_node *h;
 
-	hlist_for_each_entry(gl, h, &gl_hash_table[hash].hb_list, gl_list) {
+	hlist_bl_for_each_entry_rcu(gl, h, &gl_hash_table[hash], gl_list) {
 		if (!lm_name_equal(&gl->gl_name, name))
 			continue;
 		if (gl->gl_sbd != sdp)
 			continue;
-
-		atomic_inc(&gl->gl_ref);
-
-		return gl;
+		if (atomic_inc_not_zero(&gl->gl_ref))
+			return gl;
 	}
 
 	return NULL;
@@ -743,10 +706,11 @@ int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 	struct gfs2_glock *gl, *tmp;
 	unsigned int hash = gl_hash(sdp, &name);
 	struct address_space *mapping;
+	struct kmem_cache *cachep;
 
-	read_lock(gl_lock_addr(hash));
+	rcu_read_lock();
 	gl = search_bucket(hash, sdp, &name);
-	read_unlock(gl_lock_addr(hash));
+	rcu_read_unlock();
 
 	*glp = gl;
 	if (gl)
@@ -755,9 +719,10 @@ int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 		return -ENOENT;
 
 	if (glops->go_flags & GLOF_ASPACE)
-		gl = kmem_cache_alloc(gfs2_glock_aspace_cachep, GFP_KERNEL);
+		cachep = gfs2_glock_aspace_cachep;
 	else
-		gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
+		cachep = gfs2_glock_cachep;
+	gl = kmem_cache_alloc(cachep, GFP_KERNEL);
 	if (!gl)
 		return -ENOMEM;
 
@@ -790,15 +755,15 @@ int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 		mapping->writeback_index = 0;
 	}
 
-	write_lock(gl_lock_addr(hash));
+	spin_lock_bucket(hash);
 	tmp = search_bucket(hash, sdp, &name);
 	if (tmp) {
-		write_unlock(gl_lock_addr(hash));
-		glock_free(gl);
+		spin_unlock_bucket(hash);
+		kmem_cache_free(cachep, gl);
 		gl = tmp;
 	} else {
-		hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
-		write_unlock(gl_lock_addr(hash));
+		hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
+		spin_unlock_bucket(hash);
 	}
 
 	*glp = gl;
@@ -1113,6 +1078,7 @@ void gfs2_glock_dq(struct gfs2_holder *gh)
 		    !test_bit(GLF_DEMOTE, &gl->gl_flags))
 			fast_path = 1;
 	}
+	__gfs2_glock_schedule_for_reclaim(gl);
 	trace_gfs2_glock_queue(gh, 0);
 	spin_unlock(&gl->gl_spin);
 	if (likely(fast_path))
@@ -1440,42 +1406,30 @@ static struct shrinker glock_shrinker = {
  * @sdp: the filesystem
  * @bucket: the bucket
  *
- * Returns: 1 if the bucket has entries
  */
 
-static int examine_bucket(glock_examiner examiner, struct gfs2_sbd *sdp,
+static void examine_bucket(glock_examiner examiner, const struct gfs2_sbd *sdp,
 			  unsigned int hash)
 {
-	struct gfs2_glock *gl, *prev = NULL;
-	int has_entries = 0;
-	struct hlist_head *head = &gl_hash_table[hash].hb_list;
+	struct gfs2_glock *gl;
+	struct hlist_bl_head *head = &gl_hash_table[hash];
+	struct hlist_bl_node *pos;
 
-	read_lock(gl_lock_addr(hash));
-	/* Can't use hlist_for_each_entry - don't want prefetch here */
-	if (hlist_empty(head))
-		goto out;
-	gl = list_entry(head->first, struct gfs2_glock, gl_list);
-	while(1) {
-		if (!sdp || gl->gl_sbd == sdp) {
-			gfs2_glock_hold(gl);
-			read_unlock(gl_lock_addr(hash));
-			if (prev)
-				gfs2_glock_put(prev);
-			prev = gl;
+	rcu_read_lock();
+	hlist_bl_for_each_entry_rcu(gl, pos, head, gl_list) {
+		if ((gl->gl_sbd == sdp) && atomic_read(&gl->gl_ref))
 			examiner(gl);
-			has_entries = 1;
-			read_lock(gl_lock_addr(hash));
-		}
-		if (gl->gl_list.next == NULL)
-			break;
-		gl = list_entry(gl->gl_list.next, struct gfs2_glock, gl_list);
 	}
-out:
-	read_unlock(gl_lock_addr(hash));
-	if (prev)
-		gfs2_glock_put(prev);
+	rcu_read_unlock();
 	cond_resched();
-	return has_entries;
+}
+
+static void glock_hash_walk(glock_examiner examiner, const struct gfs2_sbd *sdp)
+{
+	unsigned x;
+
+	for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
+		examine_bucket(examiner, sdp, x);
 }
 
 
@@ -1529,10 +1483,21 @@ static void clear_glock(struct gfs2_glock *gl)
 
 void gfs2_glock_thaw(struct gfs2_sbd *sdp)
 {
-	unsigned x;
+	glock_hash_walk(thaw_glock, sdp);
+}
 
-	for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
-		examine_bucket(thaw_glock, sdp, x);
+static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
+{
+	int ret;
+	spin_lock(&gl->gl_spin);
+	ret = __dump_glock(seq, gl);
+	spin_unlock(&gl->gl_spin);
+	return ret;
+}
+
+static void dump_glock_func(struct gfs2_glock *gl)
+{
+	dump_glock(NULL, gl);
 }
 
 /**
@@ -1545,13 +1510,10 @@ void gfs2_glock_thaw(struct gfs2_sbd *sdp)
 
 void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
 {
-	unsigned int x;
-
-	for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
-		examine_bucket(clear_glock, sdp, x);
+	glock_hash_walk(clear_glock, sdp);
 	flush_workqueue(glock_workqueue);
 	wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
-	gfs2_dump_lockstate(sdp);
+	glock_hash_walk(dump_glock_func, sdp);
 }
 
 void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
@@ -1717,66 +1679,15 @@ out:
 	return error;
 }
 
-static int dump_glock(struct seq_file *seq, struct gfs2_glock *gl)
-{
-	int ret;
-	spin_lock(&gl->gl_spin);
-	ret = __dump_glock(seq, gl);
-	spin_unlock(&gl->gl_spin);
-	return ret;
-}
-
-/**
- * gfs2_dump_lockstate - print out the current lockstate
- * @sdp: the filesystem
- * @ub: the buffer to copy the information into
- *
- * If @ub is NULL, dump the lockstate to the console.
- *
- */
-
-static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
-{
-	struct gfs2_glock *gl;
-	struct hlist_node *h;
-	unsigned int x;
-	int error = 0;
-
-	for (x = 0; x < GFS2_GL_HASH_SIZE; x++) {
-
-		read_lock(gl_lock_addr(x));
-
-		hlist_for_each_entry(gl, h, &gl_hash_table[x].hb_list, gl_list) {
-			if (gl->gl_sbd != sdp)
-				continue;
-
-			error = dump_glock(NULL, gl);
-			if (error)
-				break;
-		}
-
-		read_unlock(gl_lock_addr(x));
-
-		if (error)
-			break;
-	}
-
 
-	return error;
-}
 
 
 int __init gfs2_glock_init(void)
 {
 	unsigned i;
 	for(i = 0; i < GFS2_GL_HASH_SIZE; i++) {
-		INIT_HLIST_HEAD(&gl_hash_table[i].hb_list);
-	}
-#ifdef GL_HASH_LOCK_SZ
-	for(i = 0; i < GL_HASH_LOCK_SZ; i++) {
-		rwlock_init(&gl_hash_locks[i]);
+		INIT_HLIST_BL_HEAD(&gl_hash_table[i]);
 	}
-#endif
 
 	glock_workqueue = alloc_workqueue("glock_workqueue", WQ_MEM_RECLAIM |
 					  WQ_HIGHPRI | WQ_FREEZEABLE, 0);
@@ -1802,62 +1713,54 @@ void gfs2_glock_exit(void)
 	destroy_workqueue(gfs2_delete_workqueue);
 }
 
+static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
+{
+	return hlist_bl_entry(hlist_bl_first_rcu(&gl_hash_table[hash]),
+			      struct gfs2_glock, gl_list);
+}
+
+static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
+{
+	return hlist_bl_entry(rcu_dereference_raw(gl->gl_list.next),
+			      struct gfs2_glock, gl_list);
+}
+
 static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
 {
 	struct gfs2_glock *gl;
 
-restart:
-	read_lock(gl_lock_addr(gi->hash));
-	gl = gi->gl;
-	if (gl) {
-		gi->gl = hlist_entry(gl->gl_list.next,
-				     struct gfs2_glock, gl_list);
-	} else {
-		gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
-				     struct gfs2_glock, gl_list);
-	}
-	if (gi->gl)
-		gfs2_glock_hold(gi->gl);
-	read_unlock(gl_lock_addr(gi->hash));
-	if (gl)
-		gfs2_glock_put(gl);
-	while (gi->gl == NULL) {
-		gi->hash++;
-		if (gi->hash >= GFS2_GL_HASH_SIZE)
-			return 1;
-		read_lock(gl_lock_addr(gi->hash));
-		gi->gl = hlist_entry(gl_hash_table[gi->hash].hb_list.first,
-				     struct gfs2_glock, gl_list);
-		if (gi->gl)
-			gfs2_glock_hold(gi->gl);
-		read_unlock(gl_lock_addr(gi->hash));
-	}
-
-	if (gi->sdp != gi->gl->gl_sbd)
-		goto restart;
+	do {
+		gl = gi->gl;
+		if (gl) {
+			gi->gl = glock_hash_next(gl);
+		} else {
+			gi->gl = glock_hash_chain(gi->hash);
+		}
+		while (gi->gl == NULL) {
+			gi->hash++;
+			if (gi->hash >= GFS2_GL_HASH_SIZE) {
+				rcu_read_unlock();
+				return 1;
+			}
+			gi->gl = glock_hash_chain(gi->hash);
+		}
+	/* Skip entries for other sb and dead entries */
+	} while (gi->sdp != gi->gl->gl_sbd || atomic_read(&gi->gl->gl_ref) == 0);
 
 	return 0;
 }
 
-static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
-{
-	if (gi->gl)
-		gfs2_glock_put(gi->gl);
-	gi->gl = NULL;
-}
-
 static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
 {
 	struct gfs2_glock_iter *gi = seq->private;
 	loff_t n = *pos;
 
 	gi->hash = 0;
+	rcu_read_lock();
 
 	do {
-		if (gfs2_glock_iter_next(gi)) {
-			gfs2_glock_iter_free(gi);
+		if (gfs2_glock_iter_next(gi))
 			return NULL;
-		}
 	} while (n--);
 
 	return gi->gl;
@@ -1870,10 +1773,8 @@ static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
 
 	(*pos)++;
 
-	if (gfs2_glock_iter_next(gi)) {
-		gfs2_glock_iter_free(gi);
+	if (gfs2_glock_iter_next(gi))
 		return NULL;
-	}
 
 	return gi->gl;
 }
@@ -1881,7 +1782,10 @@ static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
 static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
 {
 	struct gfs2_glock_iter *gi = seq->private;
-	gfs2_glock_iter_free(gi);
+
+	if (gi->gl)
+		rcu_read_unlock();
+	gi->gl = NULL;
 }
 
 static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
diff --git a/fs/gfs2/glock.h b/fs/gfs2/glock.h
index 691851c..afa8bfe 100644
--- a/fs/gfs2/glock.h
+++ b/fs/gfs2/glock.h
@@ -118,7 +118,7 @@ struct lm_lockops {
 	int (*lm_mount) (struct gfs2_sbd *sdp, const char *fsname);
  	void (*lm_unmount) (struct gfs2_sbd *sdp);
 	void (*lm_withdraw) (struct gfs2_sbd *sdp);
-	void (*lm_put_lock) (struct kmem_cache *cachep, struct gfs2_glock *gl);
+	void (*lm_put_lock) (struct gfs2_glock *gl);
 	int (*lm_lock) (struct gfs2_glock *gl, unsigned int req_state,
 			unsigned int flags);
 	void (*lm_cancel) (struct gfs2_glock *gl);
@@ -174,7 +174,7 @@ int gfs2_glock_get(struct gfs2_sbd *sdp,
 		   int create, struct gfs2_glock **glp);
 void gfs2_glock_hold(struct gfs2_glock *gl);
 void gfs2_glock_put_nolock(struct gfs2_glock *gl);
-int gfs2_glock_put(struct gfs2_glock *gl);
+void gfs2_glock_put(struct gfs2_glock *gl);
 void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
 		      struct gfs2_holder *gh);
 void gfs2_holder_reinit(unsigned int state, unsigned flags,
@@ -223,25 +223,22 @@ static inline int gfs2_glock_nq_init(struct gfs2_glock *gl,
 	return error;
 }
 
-/*  Lock Value Block functions  */
-
-int gfs2_lvb_hold(struct gfs2_glock *gl);
-void gfs2_lvb_unhold(struct gfs2_glock *gl);
-
-void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state);
-void gfs2_glock_complete(struct gfs2_glock *gl, int ret);
-void gfs2_reclaim_glock(struct gfs2_sbd *sdp);
-void gfs2_gl_hash_clear(struct gfs2_sbd *sdp);
-void gfs2_glock_finish_truncate(struct gfs2_inode *ip);
-void gfs2_glock_thaw(struct gfs2_sbd *sdp);
-
-int __init gfs2_glock_init(void);
-void gfs2_glock_exit(void);
-
-int gfs2_create_debugfs_file(struct gfs2_sbd *sdp);
-void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp);
-int gfs2_register_debugfs(void);
-void gfs2_unregister_debugfs(void);
+extern void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state);
+extern void gfs2_glock_complete(struct gfs2_glock *gl, int ret);
+extern void gfs2_reclaim_glock(struct gfs2_sbd *sdp);
+extern void gfs2_gl_hash_clear(struct gfs2_sbd *sdp);
+extern void gfs2_glock_finish_truncate(struct gfs2_inode *ip);
+extern void gfs2_glock_thaw(struct gfs2_sbd *sdp);
+extern void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl);
+extern void gfs2_glock_free(struct rcu_head *rcu);
+
+extern int __init gfs2_glock_init(void);
+extern void gfs2_glock_exit(void);
+
+extern int gfs2_create_debugfs_file(struct gfs2_sbd *sdp);
+extern void gfs2_delete_debugfs_file(struct gfs2_sbd *sdp);
+extern int gfs2_register_debugfs(void);
+extern void gfs2_unregister_debugfs(void);
 
 extern const struct lm_lockops gfs2_dlm_ops;
 
diff --git a/fs/gfs2/glops.c b/fs/gfs2/glops.c
index 263561b..ac5fac9 100644
--- a/fs/gfs2/glops.c
+++ b/fs/gfs2/glops.c
@@ -206,8 +206,17 @@ static void inode_go_inval(struct gfs2_glock *gl, int flags)
 static int inode_go_demote_ok(const struct gfs2_glock *gl)
 {
 	struct gfs2_sbd *sdp = gl->gl_sbd;
+	struct gfs2_holder *gh;
+
 	if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
 		return 0;
+
+	if (!list_empty(&gl->gl_holders)) {
+		gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
+		if (gh->gh_list.next != &gl->gl_holders)
+			return 0;
+	}
+
 	return 1;
 }
 
@@ -272,19 +281,6 @@ static int inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
 }
 
 /**
- * rgrp_go_demote_ok - Check to see if it's ok to unlock a RG's glock
- * @gl: the glock
- *
- * Returns: 1 if it's ok
- */
-
-static int rgrp_go_demote_ok(const struct gfs2_glock *gl)
-{
-	const struct address_space *mapping = (const struct address_space *)(gl + 1);
-	return !mapping->nrpages;
-}
-
-/**
  * rgrp_go_lock - operation done after an rgrp lock is locked by
  *    a first holder on this node.
  * @gl: the glock
@@ -410,7 +406,6 @@ const struct gfs2_glock_operations gfs2_inode_glops = {
 const struct gfs2_glock_operations gfs2_rgrp_glops = {
 	.go_xmote_th = rgrp_go_sync,
 	.go_inval = rgrp_go_inval,
-	.go_demote_ok = rgrp_go_demote_ok,
 	.go_lock = rgrp_go_lock,
 	.go_unlock = rgrp_go_unlock,
 	.go_dump = gfs2_rgrp_dump,
diff --git a/fs/gfs2/incore.h b/fs/gfs2/incore.h
index a79790c..720c1e6 100644
--- a/fs/gfs2/incore.h
+++ b/fs/gfs2/incore.h
@@ -15,6 +15,8 @@
 #include <linux/workqueue.h>
 #include <linux/dlm.h>
 #include <linux/buffer_head.h>
+#include <linux/rcupdate.h>
+#include <linux/rculist_bl.h>
 
 #define DIO_WAIT	0x00000010
 #define DIO_METADATA	0x00000020
@@ -201,7 +203,7 @@ enum {
 };
 
 struct gfs2_glock {
-	struct hlist_node gl_list;
+	struct hlist_bl_node gl_list;
 	unsigned long gl_flags;		/* GLF_... */
 	struct lm_lockname gl_name;
 	atomic_t gl_ref;
@@ -234,6 +236,7 @@ struct gfs2_glock {
 	atomic_t gl_ail_count;
 	struct delayed_work gl_work;
 	struct work_struct gl_delete;
+	struct rcu_head gl_rcu;
 };
 
 #define GFS2_MIN_LVB_SIZE 32	/* Min size of LVB that gfs2 supports */
diff --git a/fs/gfs2/lock_dlm.c b/fs/gfs2/lock_dlm.c
index 6e493ae..c80485c 100644
--- a/fs/gfs2/lock_dlm.c
+++ b/fs/gfs2/lock_dlm.c
@@ -22,7 +22,6 @@ static void gdlm_ast(void *arg)
 {
 	struct gfs2_glock *gl = arg;
 	unsigned ret = gl->gl_state;
-	struct gfs2_sbd *sdp = gl->gl_sbd;
 
 	BUG_ON(gl->gl_lksb.sb_flags & DLM_SBF_DEMOTED);
 
@@ -31,12 +30,7 @@ static void gdlm_ast(void *arg)
 
 	switch (gl->gl_lksb.sb_status) {
 	case -DLM_EUNLOCK: /* Unlocked, so glock can be freed */
-		if (gl->gl_ops->go_flags & GLOF_ASPACE)
-			kmem_cache_free(gfs2_glock_aspace_cachep, gl);
-		else
-			kmem_cache_free(gfs2_glock_cachep, gl);
-		if (atomic_dec_and_test(&sdp->sd_glock_disposal))
-			wake_up(&sdp->sd_glock_wait);
+		call_rcu(&gl->gl_rcu, gfs2_glock_free);
 		return;
 	case -DLM_ECANCEL: /* Cancel while getting lock */
 		ret |= LM_OUT_CANCELED;
@@ -164,16 +158,14 @@ static int gdlm_lock(struct gfs2_glock *gl, unsigned int req_state,
 			GDLM_STRNAME_BYTES - 1, 0, gdlm_ast, gl, gdlm_bast);
 }
 
-static void gdlm_put_lock(struct kmem_cache *cachep, struct gfs2_glock *gl)
+static void gdlm_put_lock(struct gfs2_glock *gl)
 {
 	struct gfs2_sbd *sdp = gl->gl_sbd;
 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
 	int error;
 
 	if (gl->gl_lksb.sb_lkid == 0) {
-		kmem_cache_free(cachep, gl);
-		if (atomic_dec_and_test(&sdp->sd_glock_disposal))
-			wake_up(&sdp->sd_glock_wait);
+		call_rcu(&gl->gl_rcu, gfs2_glock_free);
 		return;
 	}
 
diff --git a/fs/gfs2/lops.c b/fs/gfs2/lops.c
index bf33f82..11a73ef 100644
--- a/fs/gfs2/lops.c
+++ b/fs/gfs2/lops.c
@@ -91,7 +91,8 @@ static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 	}
 	bd->bd_ail = ai;
 	list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
-	clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
+	if (test_and_clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags))
+		gfs2_glock_schedule_for_reclaim(bd->bd_gl);
 	trace_gfs2_pin(bd, 0);
 	gfs2_log_unlock(sdp);
 	unlock_buffer(bh);
diff --git a/fs/gfs2/main.c b/fs/gfs2/main.c
index ebef7ab..d850004 100644
--- a/fs/gfs2/main.c
+++ b/fs/gfs2/main.c
@@ -14,6 +14,8 @@
 #include <linux/module.h>
 #include <linux/init.h>
 #include <linux/gfs2_ondisk.h>
+#include <linux/rcupdate.h>
+#include <linux/rculist_bl.h>
 #include <asm/atomic.h>
 
 #include "gfs2.h"
@@ -45,7 +47,7 @@ static void gfs2_init_glock_once(void *foo)
 {
 	struct gfs2_glock *gl = foo;
 
-	INIT_HLIST_NODE(&gl->gl_list);
+	INIT_HLIST_BL_NODE(&gl->gl_list);
 	spin_lock_init(&gl->gl_spin);
 	INIT_LIST_HEAD(&gl->gl_holders);
 	INIT_LIST_HEAD(&gl->gl_lru);
@@ -198,6 +200,8 @@ static void __exit exit_gfs2_fs(void)
 	unregister_filesystem(&gfs2meta_fs_type);
 	destroy_workqueue(gfs_recovery_wq);
 
+	rcu_barrier();
+
 	kmem_cache_destroy(gfs2_quotad_cachep);
 	kmem_cache_destroy(gfs2_rgrpd_cachep);
 	kmem_cache_destroy(gfs2_bufdata_cachep);
diff --git a/fs/gfs2/ops_fstype.c b/fs/gfs2/ops_fstype.c
index 777927c..a39c103 100644
--- a/fs/gfs2/ops_fstype.c
+++ b/fs/gfs2/ops_fstype.c
@@ -928,12 +928,9 @@ static const match_table_t nolock_tokens = {
 	{ Opt_err, NULL },
 };
 
-static void nolock_put_lock(struct kmem_cache *cachep, struct gfs2_glock *gl)
+static void nolock_put_lock(struct gfs2_glock *gl)
 {
-	struct gfs2_sbd *sdp = gl->gl_sbd;
-	kmem_cache_free(cachep, gl);
-	if (atomic_dec_and_test(&sdp->sd_glock_disposal))
-		wake_up(&sdp->sd_glock_wait);
+	call_rcu(&gl->gl_rcu, gfs2_glock_free);
 }
 
 static const struct lm_lockops nolock_ops = {
-- 
1.7.4

[PATCH 02/15] GFS2: Post-VFS scale update for RCU path walk

[Steven Whitehouse]

We can allow a few more cases to use RCU path walking than
originally allowed. It should be possible to also enable
RCU path walking when the glock is already cached. Thats
a bit more complicated though, so left for a future patch.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Cc: Nick Piggin <npiggin@gmail.com>

diff --git a/fs/gfs2/acl.c b/fs/gfs2/acl.c
index 7118f1a..cbc0715 100644
--- a/fs/gfs2/acl.c
+++ b/fs/gfs2/acl.c
@@ -80,8 +80,11 @@ int gfs2_check_acl(struct inode *inode, int mask, unsigned int flags)
 	struct posix_acl *acl;
 	int error;
 
-	if (flags & IPERM_FLAG_RCU)
-		return -ECHILD;
+	if (flags & IPERM_FLAG_RCU) {
+		if (!negative_cached_acl(inode, ACL_TYPE_ACCESS))
+			return -ECHILD;
+		return -EAGAIN;
+	}
 
 	acl = gfs2_acl_get(GFS2_I(inode), ACL_TYPE_ACCESS);
 	if (IS_ERR(acl))
diff --git a/fs/gfs2/ops_inode.c b/fs/gfs2/ops_inode.c
index d8b26ac..09e436a 100644
--- a/fs/gfs2/ops_inode.c
+++ b/fs/gfs2/ops_inode.c
@@ -1026,9 +1026,9 @@ static void gfs2_put_link(struct dentry *dentry, struct nameidata *nd, void *p)
 
 /**
  * gfs2_permission -
- * @inode:
- * @mask:
- * @nd: passed from Linux VFS, ignored by us
+ * @inode: The inode
+ * @mask: The mask to be tested
+ * @flags: Indicates whether this is an RCU path walk or not
  *
  * This may be called from the VFS directly, or from within GFS2 with the
  * inode locked, so we look to see if the glock is already locked and only
@@ -1044,11 +1044,11 @@ int gfs2_permission(struct inode *inode, int mask, unsigned int flags)
 	int error;
 	int unlock = 0;
 
-	if (flags & IPERM_FLAG_RCU)
-		return -ECHILD;
 
 	ip = GFS2_I(inode);
 	if (gfs2_glock_is_locked_by_me(ip->i_gl) == NULL) {
+		if (flags & IPERM_FLAG_RCU)
+			return -ECHILD;
 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 		if (error)
 			return error;
-- 
1.7.4

[PATCH 03/15] GFS2: Fix glock queue trace point

[Steven Whitehouse]

Somehow this tracepoint landed up in the wrong place. This moves it
to where it should be.

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

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index c75d499..ddc3e1e 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -972,13 +972,13 @@ fail:
 			insert_pt = &gh2->gh_list;
 	}
 	set_bit(GLF_QUEUED, &gl->gl_flags);
+	trace_gfs2_glock_queue(gh, 1);
 	if (likely(insert_pt == NULL)) {
 		list_add_tail(&gh->gh_list, &gl->gl_holders);
 		if (unlikely(gh->gh_flags & LM_FLAG_PRIORITY))
 			goto do_cancel;
 		return;
 	}
-	trace_gfs2_glock_queue(gh, 1);
 	list_add_tail(&gh->gh_list, insert_pt);
 do_cancel:
 	gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
-- 
1.7.4

[PATCH 04/15] GFS2: Improve cluster mmap scalability

[Steven Whitehouse]

The mmap system call grabs a glock when an update to atime maybe
required. It does this in order to ensure that the flags on the
inode are uptodate, but since it will only mark atime for a future
update, an exclusive lock is not required here (one will be taken
later when the actual update is performed).

Also, the lock can be skipped when the mount is marked noatime in
addition to the original check which only looked at the noatime
flag for the inode itself.

This should increase the scalability of the mmap call when multiple
nodes are all mmaping the same file.

Reported-by: Scooter Morris <scooter@cgl.ucsf.edu>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 7cfdcb9..216ad27 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -448,15 +448,20 @@ static int gfs2_mmap(struct file *file, struct vm_area_struct *vma)
 {
 	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
 
-	if (!(file->f_flags & O_NOATIME)) {
+	if (!(file->f_flags & O_NOATIME) &&
+	    !IS_NOATIME(&ip->i_inode)) {
 		struct gfs2_holder i_gh;
 		int error;
 
-		gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
+		gfs2_holder_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
 		error = gfs2_glock_nq(&i_gh);
-		file_accessed(file);
-		if (error == 0)
-			gfs2_glock_dq_uninit(&i_gh);
+		if (error == 0) {
+			file_accessed(file);
+			gfs2_glock_dq(&i_gh);
+		}
+		gfs2_holder_uninit(&i_gh);
+		if (error)
+			return error;
 	}
 	vma->vm_ops = &gfs2_vm_ops;
 	vma->vm_flags |= VM_CAN_NONLINEAR;
-- 
1.7.4

[PATCH 05/15] GFS2: panics on quotacheck update

[Steven Whitehouse]

From: Abhijith Das <adas@redhat.com>

Handle block allocation for forceful unstuffing of quota dinode during quota
update using quotactl(). Also fix block reservation for special cases when
quotas cross over block boundaries and update 2 blocks instead of 1.

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

diff --git a/fs/gfs2/quota.c b/fs/gfs2/quota.c
index a689901..6ec964c 100644
--- a/fs/gfs2/quota.c
+++ b/fs/gfs2/quota.c
@@ -1587,6 +1587,8 @@ static int gfs2_set_dqblk(struct super_block *sb, int type, qid_t id,
 
 	offset = qd2offset(qd);
 	alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
+	if (gfs2_is_stuffed(ip))
+		alloc_required = 1;
 	if (alloc_required) {
 		al = gfs2_alloc_get(ip);
 		if (al == NULL)
@@ -1600,7 +1602,9 @@ static int gfs2_set_dqblk(struct super_block *sb, int type, qid_t id,
 		blocks += gfs2_rg_blocks(al);
 	}
 
-	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 1, 0);
+	/* Some quotas span block boundaries and can update two blocks,
+	   adding an extra block to the transaction to handle such quotas */
+	error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
 	if (error)
 		goto out_release;
 
-- 
1.7.4

[PATCH 06/15] GFS2: deallocation performance patch

[Steven Whitehouse]

From: Bob Peterson <rpeterso@redhat.com>

This patch is a performance improvement to GFS2's dealloc code.
Rather than update the quota file and statfs file for every
single block that's stripped off in unlink function do_strip,
this patch keeps track and updates them once for every layer
that's stripped.  This is done entirely inside the existing
transaction, so there should be no risk of corruption.
The other functions that deallocate blocks will be unaffected
because they are using wrapper functions that do the same
thing that they do today.

I tested this code on my roth cluster by creating 200
files in a directory, each of which is 100MB, then on
four nodes, I simultaneously deleted the files, thus competing
for GFS2 resources (but different files).  The commands
I used were:

[root@roth-01]# time for i in `seq 1 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done
[root@roth-02]# time for i in `seq 2 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done
[root@roth-03]# time for i in `seq 3 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done
[root@roth-05]# time for i in `seq 4 4 200` ; do rm /mnt/gfs2/bigdir/gfs2.$i; done

The performance increase was significant:

             roth-01     roth-02     roth-03     roth-05
             ---------   ---------   ---------   ---------
old: real    0m34.027    0m25.021s   0m23.906s   0m35.646s
new: real    0m22.379s   0m24.362s   0m24.133s   0m18.562s

Total time spent deleting:
old: 118.6s
new:  89.4

For this particular case, this showed a 25% performance increase for
GFS2 unlinks.

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

diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c
index 3c4039d..ef3dc4b 100644
--- a/fs/gfs2/bmap.c
+++ b/fs/gfs2/bmap.c
@@ -21,6 +21,7 @@
 #include "meta_io.h"
 #include "quota.h"
 #include "rgrp.h"
+#include "super.h"
 #include "trans.h"
 #include "dir.h"
 #include "util.h"
@@ -757,7 +758,7 @@ static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 	struct gfs2_rgrp_list rlist;
 	u64 bn, bstart;
-	u32 blen;
+	u32 blen, btotal;
 	__be64 *p;
 	unsigned int rg_blocks = 0;
 	int metadata;
@@ -839,6 +840,7 @@ static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 
 	bstart = 0;
 	blen = 0;
+	btotal = 0;
 
 	for (p = top; p < bottom; p++) {
 		if (!*p)
@@ -851,9 +853,11 @@ static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 		else {
 			if (bstart) {
 				if (metadata)
-					gfs2_free_meta(ip, bstart, blen);
+					__gfs2_free_meta(ip, bstart, blen);
 				else
-					gfs2_free_data(ip, bstart, blen);
+					__gfs2_free_data(ip, bstart, blen);
+
+				btotal += blen;
 			}
 
 			bstart = bn;
@@ -865,11 +869,17 @@ static int do_strip(struct gfs2_inode *ip, struct buffer_head *dibh,
 	}
 	if (bstart) {
 		if (metadata)
-			gfs2_free_meta(ip, bstart, blen);
+			__gfs2_free_meta(ip, bstart, blen);
 		else
-			gfs2_free_data(ip, bstart, blen);
+			__gfs2_free_data(ip, bstart, blen);
+
+		btotal += blen;
 	}
 
+	gfs2_statfs_change(sdp, 0, +btotal, 0);
+	gfs2_quota_change(ip, -(s64)btotal, ip->i_inode.i_uid,
+			  ip->i_inode.i_gid);
+
 	ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
 
 	gfs2_dinode_out(ip, dibh->b_data);
diff --git a/fs/gfs2/rgrp.c b/fs/gfs2/rgrp.c
index 7293ea2..cf930cd 100644
--- a/fs/gfs2/rgrp.c
+++ b/fs/gfs2/rgrp.c
@@ -1602,7 +1602,7 @@ rgrp_error:
  *
  */
 
-void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen)
+void __gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen)
 {
 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 	struct gfs2_rgrpd *rgd;
@@ -1617,7 +1617,21 @@ void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen)
 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
 
 	gfs2_trans_add_rg(rgd);
+}
 
+/**
+ * gfs2_free_data - free a contiguous run of data block(s)
+ * @ip: the inode these blocks are being freed from
+ * @bstart: first block of a run of contiguous blocks
+ * @blen: the length of the block run
+ *
+ */
+
+void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen)
+{
+	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
+
+	__gfs2_free_data(ip, bstart, blen);
 	gfs2_statfs_change(sdp, 0, +blen, 0);
 	gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
 }
@@ -1630,7 +1644,7 @@ void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen)
  *
  */
 
-void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
+void __gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
 {
 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 	struct gfs2_rgrpd *rgd;
@@ -1645,10 +1659,24 @@ void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
 	gfs2_rgrp_out(rgd, rgd->rd_bits[0].bi_bh->b_data);
 
 	gfs2_trans_add_rg(rgd);
+	gfs2_meta_wipe(ip, bstart, blen);
+}
 
+/**
+ * gfs2_free_meta - free a contiguous run of data block(s)
+ * @ip: the inode these blocks are being freed from
+ * @bstart: first block of a run of contiguous blocks
+ * @blen: the length of the block run
+ *
+ */
+
+void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen)
+{
+	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
+
+	__gfs2_free_meta(ip, bstart, blen);
 	gfs2_statfs_change(sdp, 0, +blen, 0);
 	gfs2_quota_change(ip, -(s64)blen, ip->i_inode.i_uid, ip->i_inode.i_gid);
-	gfs2_meta_wipe(ip, bstart, blen);
 }
 
 void gfs2_unlink_di(struct inode *inode)
diff --git a/fs/gfs2/rgrp.h b/fs/gfs2/rgrp.h
index 50c2bb0..a80e303 100644
--- a/fs/gfs2/rgrp.h
+++ b/fs/gfs2/rgrp.h
@@ -52,7 +52,9 @@ extern int gfs2_ri_update(struct gfs2_inode *ip);
 extern int gfs2_alloc_block(struct gfs2_inode *ip, u64 *bn, unsigned int *n);
 extern int gfs2_alloc_di(struct gfs2_inode *ip, u64 *bn, u64 *generation);
 
+extern void __gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen);
 extern void gfs2_free_data(struct gfs2_inode *ip, u64 bstart, u32 blen);
+extern void __gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen);
 extern void gfs2_free_meta(struct gfs2_inode *ip, u64 bstart, u32 blen);
 extern void gfs2_free_di(struct gfs2_rgrpd *rgd, struct gfs2_inode *ip);
 extern void gfs2_unlink_di(struct inode *inode);
-- 
1.7.4

[PATCH 07/15] GFS2: quota allows exceeding hard limit

[Steven Whitehouse]

From: Abhijith Das <adas@redhat.com>

Immediately after being synced to disk, cached quotas are zeroed out and a
subsequent access of the cached quotas results in incorrect zero values. This
meant that gfs2 assumed the actual usage to be the zero (or near-zero) usage
values it found in the cached quotas and comparison against warn/limits never
triggered a quota violation.

This patch adds a new flag QDF_REFRESH that is set after a sync so that the
cached quotas are forcefully refreshed from disk on a subsequent access on
seeing this flag set.

Resolves: rhbz#675944
Signed-off-by: Abhi Das <adas@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

diff --git a/fs/gfs2/incore.h b/fs/gfs2/incore.h
index 720c1e6..59aaaa0 100644
--- a/fs/gfs2/incore.h
+++ b/fs/gfs2/incore.h
@@ -317,6 +317,7 @@ enum {
 	QDF_USER		= 0,
 	QDF_CHANGE		= 1,
 	QDF_LOCKED		= 2,
+	QDF_REFRESH		= 3,
 };
 
 struct gfs2_quota_data {
diff --git a/fs/gfs2/quota.c b/fs/gfs2/quota.c
index 6ec964c..e23d986 100644
--- a/fs/gfs2/quota.c
+++ b/fs/gfs2/quota.c
@@ -834,6 +834,7 @@ static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
 			goto out_end_trans;
 
 		do_qc(qd, -qd->qd_change_sync);
+		set_bit(QDF_REFRESH, &qd->qd_flags);
 	}
 
 	error = 0;
@@ -929,6 +930,7 @@ int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
 {
 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
 	struct gfs2_alloc *al = ip->i_alloc;
+	struct gfs2_quota_data *qd;
 	unsigned int x;
 	int error = 0;
 
@@ -942,7 +944,11 @@ int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
 	     sort_qd, NULL);
 
 	for (x = 0; x < al->al_qd_num; x++) {
-		error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
+		int force = NO_FORCE;
+		qd = al->al_qd[x];
+		if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
+			force = FORCE;
+		error = do_glock(qd, force, &al->al_qd_ghs[x]);
 		if (error)
 			break;
 	}
-- 
1.7.4

[PATCH 08/15] GFS2: Fix glock deallocation race

[Steven Whitehouse]

This patch fixes a race in deallocating glocks which was introduced
in the RCU glock patch. We need to ensure that the glock count is
kept correct even in the case that there is a race to add a new
glock into the hash table. Also, to avoid having to wait for an
RCU grace period, the glock counter can be decremented before
call_rcu() is called.

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

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index ddc3e1e..3f45a14 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -103,16 +103,21 @@ static inline void spin_unlock_bucket(unsigned int hash)
 	__bit_spin_unlock(0, (unsigned long *)bl);
 }
 
-void gfs2_glock_free(struct rcu_head *rcu)
+static void gfs2_glock_dealloc(struct rcu_head *rcu)
 {
 	struct gfs2_glock *gl = container_of(rcu, struct gfs2_glock, gl_rcu);
-	struct gfs2_sbd *sdp = gl->gl_sbd;
 
 	if (gl->gl_ops->go_flags & GLOF_ASPACE)
 		kmem_cache_free(gfs2_glock_aspace_cachep, gl);
 	else
 		kmem_cache_free(gfs2_glock_cachep, gl);
+}
+
+void gfs2_glock_free(struct gfs2_glock *gl)
+{
+	struct gfs2_sbd *sdp = gl->gl_sbd;
 
+	call_rcu(&gl->gl_rcu, gfs2_glock_dealloc);
 	if (atomic_dec_and_test(&sdp->sd_glock_disposal))
 		wake_up(&sdp->sd_glock_wait);
 }
@@ -760,6 +765,7 @@ int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
 	if (tmp) {
 		spin_unlock_bucket(hash);
 		kmem_cache_free(cachep, gl);
+		atomic_dec(&sdp->sd_glock_disposal);
 		gl = tmp;
 	} else {
 		hlist_bl_add_head_rcu(&gl->gl_list, &gl_hash_table[hash]);
diff --git a/fs/gfs2/glock.h b/fs/gfs2/glock.h
index afa8bfe..aea1606 100644
--- a/fs/gfs2/glock.h
+++ b/fs/gfs2/glock.h
@@ -230,7 +230,7 @@ extern void gfs2_gl_hash_clear(struct gfs2_sbd *sdp);
 extern void gfs2_glock_finish_truncate(struct gfs2_inode *ip);
 extern void gfs2_glock_thaw(struct gfs2_sbd *sdp);
 extern void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl);
-extern void gfs2_glock_free(struct rcu_head *rcu);
+extern void gfs2_glock_free(struct gfs2_glock *gl);
 
 extern int __init gfs2_glock_init(void);
 extern void gfs2_glock_exit(void);
diff --git a/fs/gfs2/lock_dlm.c b/fs/gfs2/lock_dlm.c
index c80485c..98c80d8 100644
--- a/fs/gfs2/lock_dlm.c
+++ b/fs/gfs2/lock_dlm.c
@@ -30,7 +30,7 @@ static void gdlm_ast(void *arg)
 
 	switch (gl->gl_lksb.sb_status) {
 	case -DLM_EUNLOCK: /* Unlocked, so glock can be freed */
-		call_rcu(&gl->gl_rcu, gfs2_glock_free);
+		gfs2_glock_free(gl);
 		return;
 	case -DLM_ECANCEL: /* Cancel while getting lock */
 		ret |= LM_OUT_CANCELED;
@@ -165,7 +165,7 @@ static void gdlm_put_lock(struct gfs2_glock *gl)
 	int error;
 
 	if (gl->gl_lksb.sb_lkid == 0) {
-		call_rcu(&gl->gl_rcu, gfs2_glock_free);
+		gfs2_glock_free(gl);
 		return;
 	}
 
diff --git a/fs/gfs2/ops_fstype.c b/fs/gfs2/ops_fstype.c
index a39c103..67654d0 100644
--- a/fs/gfs2/ops_fstype.c
+++ b/fs/gfs2/ops_fstype.c
@@ -928,14 +928,9 @@ static const match_table_t nolock_tokens = {
 	{ Opt_err, NULL },
 };
 
-static void nolock_put_lock(struct gfs2_glock *gl)
-{
-	call_rcu(&gl->gl_rcu, gfs2_glock_free);
-}
-
 static const struct lm_lockops nolock_ops = {
 	.lm_proto_name = "lock_nolock",
-	.lm_put_lock = nolock_put_lock,
+	.lm_put_lock = gfs2_glock_free,
 	.lm_tokens = &nolock_tokens,
 };
 
-- 
1.7.4

[PATCH 09/15] GFS2: Remove potential race in flock code

[Steven Whitehouse]

This patch ensures that we always wait for glock demotion when
dropping flocks on a file in order to prevent any race
conditions associated with further flock calls or closing
the file.

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

diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 216ad27..2878481 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -981,8 +981,10 @@ static void do_unflock(struct file *file, struct file_lock *fl)
 
 	mutex_lock(&fp->f_fl_mutex);
 	flock_lock_file_wait(file, fl);
-	if (fl_gh->gh_gl)
-		gfs2_glock_dq_uninit(fl_gh);
+	if (fl_gh->gh_gl) {
+		gfs2_glock_dq_wait(fl_gh);
+		gfs2_holder_uninit(fl_gh);
+	}
 	mutex_unlock(&fp->f_fl_mutex);
 }
 
-- 
1.7.4

[PATCH 10/15] GFS2: Optimize glock multiple-dequeue code

[Steven Whitehouse]

From: Bob Peterson <rpeterso@redhat.com>

This is a small patch that optimizes multiple glock dequeue
operations.  It changes the unlock order to be more efficient
and makes it easier for lock debugging tools to unravel.  It
also eliminates the need for the temp variable x, although
that would likely be optimized out.

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

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index 3f45a14..8648409 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -1248,10 +1248,8 @@ int gfs2_glock_nq_m(unsigned int num_gh, struct gfs2_holder *ghs)
 
 void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
 {
-	unsigned int x;
-
-	for (x = 0; x < num_gh; x++)
-		gfs2_glock_dq(&ghs[x]);
+	while (num_gh--)
+		gfs2_glock_dq(&ghs[num_gh]);
 }
 
 /**
@@ -1263,10 +1261,8 @@ void gfs2_glock_dq_m(unsigned int num_gh, struct gfs2_holder *ghs)
 
 void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
 {
-	unsigned int x;
-
-	for (x = 0; x < num_gh; x++)
-		gfs2_glock_dq_uninit(&ghs[x]);
+	while (num_gh--)
+		gfs2_glock_dq_uninit(&ghs[num_gh]);
 }
 
 void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
-- 
1.7.4

[PATCH 11/15] GFS2: fix block allocation check for fallocate

[Steven Whitehouse]

From: Benjamin Marzinski <bmarzins@redhat.com>

GFS2 fallocate wasn't properly checking if a blocks were already allocated.
In write_empty_blocks(), if a page didn't have buffer_heads attached, GFS2
was always treating it as if there were no blocks allocated for that page.
GFS2 now calls gfs2_block_map() to check if the blocks are allocated before
writing them out.

Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 2878481..4074b95 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -622,8 +622,7 @@ static void empty_write_end(struct page *page, unsigned from,
 {
 	struct gfs2_inode *ip = GFS2_I(page->mapping->host);
 
-	page_zero_new_buffers(page, from, to);
-	flush_dcache_page(page);
+	zero_user(page, from, to-from);
 	mark_page_accessed(page);
 
 	if (!gfs2_is_writeback(ip))
@@ -632,36 +631,43 @@ static void empty_write_end(struct page *page, unsigned from,
 	block_commit_write(page, from, to);
 }
 
-static int write_empty_blocks(struct page *page, unsigned from, unsigned to)
+static int needs_empty_write(sector_t block, struct inode *inode)
 {
-	unsigned start, end, next;
-	struct buffer_head *bh, *head;
 	int error;
+	struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
 
-	if (!page_has_buffers(page)) {
-		error = __block_write_begin(page, from, to - from, gfs2_block_map);
-		if (unlikely(error))
-			return error;
+	bh_map.b_size = 1 << inode->i_blkbits;
+	error = gfs2_block_map(inode, block, &bh_map, 0);
+	if (unlikely(error))
+		return error;
+	return !buffer_mapped(&bh_map);
+}
 
-		empty_write_end(page, from, to);
-		return 0;
-	}
+static int write_empty_blocks(struct page *page, unsigned from, unsigned to)
+{
+	struct inode *inode = page->mapping->host;
+	unsigned start, end, next, blksize;
+	sector_t block = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
+	int ret;
 
-	bh = head = page_buffers(page);
+	blksize = 1 << inode->i_blkbits;
 	next = end = 0;
 	while (next < from) {
-		next += bh->b_size;
-		bh = bh->b_this_page;
+		next += blksize;
+		block++;
 	}
 	start = next;
 	do {
-		next += bh->b_size;
-		if (buffer_mapped(bh)) {
+		next += blksize;
+		ret = needs_empty_write(block, inode);
+		if (unlikely(ret < 0))
+			return ret;
+		if (ret == 0) {
 			if (end) {
-				error = __block_write_begin(page, start, end - start,
-							    gfs2_block_map);
-				if (unlikely(error))
-					return error;
+				ret = __block_write_begin(page, start, end - start,
+							  gfs2_block_map);
+				if (unlikely(ret))
+					return ret;
 				empty_write_end(page, start, end);
 				end = 0;
 			}
@@ -669,13 +675,13 @@ static int write_empty_blocks(struct page *page, unsigned from, unsigned to)
 		}
 		else
 			end = next;
-		bh = bh->b_this_page;
+		block++;
 	} while (next < to);
 
 	if (end) {
-		error = __block_write_begin(page, start, end - start, gfs2_block_map);
-		if (unlikely(error))
-			return error;
+		ret = __block_write_begin(page, start, end - start, gfs2_block_map);
+		if (unlikely(ret))
+			return ret;
 		empty_write_end(page, start, end);
 	}
 
-- 
1.7.4

[PATCH 12/15] GFS2: introduce AIL lock

[Steven Whitehouse]

From: Dave Chinner <dchinner@redhat.com>

The log lock is currently used to protect the AIL lists and
the movements of buffers into and out of them. The lists
are self contained and no log specific items outside the
lists are accessed when starting or emptying the AIL lists.

Hence the operation of the AIL does not require the protection
of the log lock so split them out into a new AIL specific lock
to reduce the amount of traffic on the log lock. This will
also reduce the amount of serialisation that occurs when
the gfs2_logd pushes on the AIL to move it forward.

This reduces the impact of log pushing on sequential write
throughput.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

diff --git a/fs/gfs2/glops.c b/fs/gfs2/glops.c
index ac5fac9..3754e3c 100644
--- a/fs/gfs2/glops.c
+++ b/fs/gfs2/glops.c
@@ -56,20 +56,26 @@ static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
 	BUG_ON(current->journal_info);
 	current->journal_info = &tr;
 
-	gfs2_log_lock(sdp);
+	spin_lock(&sdp->sd_ail_lock);
 	while (!list_empty(head)) {
 		bd = list_entry(head->next, struct gfs2_bufdata,
 				bd_ail_gl_list);
 		bh = bd->bd_bh;
 		gfs2_remove_from_ail(bd);
+		spin_unlock(&sdp->sd_ail_lock);
+
 		bd->bd_bh = NULL;
 		bh->b_private = NULL;
 		bd->bd_blkno = bh->b_blocknr;
+		gfs2_log_lock(sdp);
 		gfs2_assert_withdraw(sdp, !buffer_busy(bh));
 		gfs2_trans_add_revoke(sdp, bd);
+		gfs2_log_unlock(sdp);
+
+		spin_lock(&sdp->sd_ail_lock);
 	}
 	gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
-	gfs2_log_unlock(sdp);
+	spin_unlock(&sdp->sd_ail_lock);
 
 	gfs2_trans_end(sdp);
 	gfs2_log_flush(sdp, NULL);
diff --git a/fs/gfs2/incore.h b/fs/gfs2/incore.h
index 59aaaa0..870a89d 100644
--- a/fs/gfs2/incore.h
+++ b/fs/gfs2/incore.h
@@ -651,6 +651,7 @@ struct gfs2_sbd {
 	unsigned int sd_log_flush_head;
 	u64 sd_log_flush_wrapped;
 
+	spinlock_t sd_ail_lock;
 	struct list_head sd_ail1_list;
 	struct list_head sd_ail2_list;
 	u64 sd_ail_sync_gen;
diff --git a/fs/gfs2/log.c b/fs/gfs2/log.c
index eb01f35..4e3c044 100644
--- a/fs/gfs2/log.c
+++ b/fs/gfs2/log.c
@@ -88,8 +88,8 @@ void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
  */
 
 static void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
-__releases(&sdp->sd_log_lock)
-__acquires(&sdp->sd_log_lock)
+__releases(&sdp->sd_ail_lock)
+__acquires(&sdp->sd_ail_lock)
 {
 	struct gfs2_bufdata *bd, *s;
 	struct buffer_head *bh;
@@ -117,7 +117,7 @@ __acquires(&sdp->sd_log_lock)
 			list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
 
 			get_bh(bh);
-			gfs2_log_unlock(sdp);
+			spin_unlock(&sdp->sd_ail_lock);
 			lock_buffer(bh);
 			if (test_clear_buffer_dirty(bh)) {
 				bh->b_end_io = end_buffer_write_sync;
@@ -126,7 +126,7 @@ __acquires(&sdp->sd_log_lock)
 				unlock_buffer(bh);
 				brelse(bh);
 			}
-			gfs2_log_lock(sdp);
+			spin_lock(&sdp->sd_ail_lock);
 
 			retry = 1;
 			break;
@@ -175,10 +175,10 @@ static void gfs2_ail1_start(struct gfs2_sbd *sdp)
 	struct gfs2_ail *ai;
 	int done = 0;
 
-	gfs2_log_lock(sdp);
+	spin_lock(&sdp->sd_ail_lock);
 	head = &sdp->sd_ail1_list;
 	if (list_empty(head)) {
-		gfs2_log_unlock(sdp);
+		spin_unlock(&sdp->sd_ail_lock);
 		return;
 	}
 	sync_gen = sdp->sd_ail_sync_gen++;
@@ -189,13 +189,13 @@ static void gfs2_ail1_start(struct gfs2_sbd *sdp)
 			if (ai->ai_sync_gen >= sync_gen)
 				continue;
 			ai->ai_sync_gen = sync_gen;
-			gfs2_ail1_start_one(sdp, ai); /* This may drop log lock */
+			gfs2_ail1_start_one(sdp, ai); /* This may drop ail lock */
 			done = 0;
 			break;
 		}
 	}
 
-	gfs2_log_unlock(sdp);
+	spin_unlock(&sdp->sd_ail_lock);
 }
 
 static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
@@ -203,7 +203,7 @@ static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
 	struct gfs2_ail *ai, *s;
 	int ret;
 
-	gfs2_log_lock(sdp);
+	spin_lock(&sdp->sd_ail_lock);
 
 	list_for_each_entry_safe_reverse(ai, s, &sdp->sd_ail1_list, ai_list) {
 		if (gfs2_ail1_empty_one(sdp, ai, flags))
@@ -214,7 +214,7 @@ static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int flags)
 
 	ret = list_empty(&sdp->sd_ail1_list);
 
-	gfs2_log_unlock(sdp);
+	spin_unlock(&sdp->sd_ail_lock);
 
 	return ret;
 }
@@ -247,7 +247,7 @@ static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 	int wrap = (new_tail < old_tail);
 	int a, b, rm;
 
-	gfs2_log_lock(sdp);
+	spin_lock(&sdp->sd_ail_lock);
 
 	list_for_each_entry_safe(ai, safe, &sdp->sd_ail2_list, ai_list) {
 		a = (old_tail <= ai->ai_first);
@@ -263,7 +263,7 @@ static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
 		kfree(ai);
 	}
 
-	gfs2_log_unlock(sdp);
+	spin_unlock(&sdp->sd_ail_lock);
 }
 
 /**
@@ -421,7 +421,7 @@ static unsigned int current_tail(struct gfs2_sbd *sdp)
 	struct gfs2_ail *ai;
 	unsigned int tail;
 
-	gfs2_log_lock(sdp);
+	spin_lock(&sdp->sd_ail_lock);
 
 	if (list_empty(&sdp->sd_ail1_list)) {
 		tail = sdp->sd_log_head;
@@ -430,7 +430,7 @@ static unsigned int current_tail(struct gfs2_sbd *sdp)
 		tail = ai->ai_first;
 	}
 
-	gfs2_log_unlock(sdp);
+	spin_unlock(&sdp->sd_ail_lock);
 
 	return tail;
 }
@@ -743,10 +743,12 @@ void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl)
 	sdp->sd_log_commited_databuf = 0;
 	sdp->sd_log_commited_revoke = 0;
 
+	spin_lock(&sdp->sd_ail_lock);
 	if (!list_empty(&ai->ai_ail1_list)) {
 		list_add(&ai->ai_list, &sdp->sd_ail1_list);
 		ai = NULL;
 	}
+	spin_unlock(&sdp->sd_ail_lock);
 	gfs2_log_unlock(sdp);
 	trace_gfs2_log_flush(sdp, 0);
 	up_write(&sdp->sd_log_flush_lock);
diff --git a/fs/gfs2/lops.c b/fs/gfs2/lops.c
index 11a73ef..4295a6a 100644
--- a/fs/gfs2/lops.c
+++ b/fs/gfs2/lops.c
@@ -80,7 +80,7 @@ static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 	mark_buffer_dirty(bh);
 	clear_buffer_pinned(bh);
 
-	gfs2_log_lock(sdp);
+	spin_lock(&sdp->sd_ail_lock);
 	if (bd->bd_ail) {
 		list_del(&bd->bd_ail_st_list);
 		brelse(bh);
@@ -91,10 +91,11 @@ static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
 	}
 	bd->bd_ail = ai;
 	list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
+	spin_unlock(&sdp->sd_ail_lock);
+
 	if (test_and_clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags))
 		gfs2_glock_schedule_for_reclaim(bd->bd_gl);
 	trace_gfs2_pin(bd, 0);
-	gfs2_log_unlock(sdp);
 	unlock_buffer(bh);
 	atomic_dec(&sdp->sd_log_pinned);
 }
diff --git a/fs/gfs2/ops_fstype.c b/fs/gfs2/ops_fstype.c
index 67654d0..42ef243 100644
--- a/fs/gfs2/ops_fstype.c
+++ b/fs/gfs2/ops_fstype.c
@@ -99,6 +99,7 @@ static struct gfs2_sbd *init_sbd(struct super_block *sb)
 
 	init_waitqueue_head(&sdp->sd_log_waitq);
 	init_waitqueue_head(&sdp->sd_logd_waitq);
+	spin_lock_init(&sdp->sd_ail_lock);
 	INIT_LIST_HEAD(&sdp->sd_ail1_list);
 	INIT_LIST_HEAD(&sdp->sd_ail2_list);
 
-- 
1.7.4

[PATCH 13/15] GFS2: Update to AIL list locking

[Steven Whitehouse]

The previous patch missed a couple of places where the AIL list
needed locking, so this fixes up those places, plus a comment
is corrected too.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Cc: Dave Chinner <dchinner@redhat.com>

diff --git a/fs/gfs2/log.c b/fs/gfs2/log.c
index 4e3c044..e7ed31f 100644
--- a/fs/gfs2/log.c
+++ b/fs/gfs2/log.c
@@ -67,7 +67,7 @@ unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct,
  * @mapping: The associated mapping (maybe NULL)
  * @bd: The gfs2_bufdata to remove
  *
- * The log lock _must_ be held when calling this function
+ * The ail lock _must_ be held when calling this function
  *
  */
 
diff --git a/fs/gfs2/lops.c b/fs/gfs2/lops.c
index 4295a6a..e919abf 100644
--- a/fs/gfs2/lops.c
+++ b/fs/gfs2/lops.c
@@ -51,8 +51,10 @@ static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
 	/* If this buffer is in the AIL and it has already been written
 	 * to in-place disk block, remove it from the AIL.
 	 */
+	spin_lock(&sdp->sd_ail_lock);
 	if (bd->bd_ail)
 		list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
+	spin_unlock(&sdp->sd_ail_lock);
 	get_bh(bh);
 	atomic_inc(&sdp->sd_log_pinned);
 	trace_gfs2_pin(bd, 1);
diff --git a/fs/gfs2/meta_io.c b/fs/gfs2/meta_io.c
index 939739c..01d97f4 100644
--- a/fs/gfs2/meta_io.c
+++ b/fs/gfs2/meta_io.c
@@ -326,6 +326,7 @@ void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int
 		brelse(bh);
 	}
 	if (bd) {
+		spin_lock(&sdp->sd_ail_lock);
 		if (bd->bd_ail) {
 			gfs2_remove_from_ail(bd);
 			bh->b_private = NULL;
@@ -333,6 +334,7 @@ void gfs2_remove_from_journal(struct buffer_head *bh, struct gfs2_trans *tr, int
 			bd->bd_blkno = bh->b_blocknr;
 			gfs2_trans_add_revoke(sdp, bd);
 		}
+		spin_unlock(&sdp->sd_ail_lock);
 	}
 	clear_buffer_dirty(bh);
 	clear_buffer_uptodate(bh);
-- 
1.7.4

[PATCH 14/15] GFS2: Adding missing unlock_page()

[Steven Whitehouse]

From: Maxim <maxim.patlasov@gmail.com>

gfs2_write_begin() calls grab_cache_page_write_begin() that returns *locked*
page. Correspondent error-handling path lacks for unlock_page() call:

> out:
> 	if (error == 0)
> 		return 0;
>
> 	page_cache_release(page);

The whole system hangs if gfs2_unstuff_dinode() called from gfs2_write_begin()
failed for some reason.

Reported-by: Maxim <maxim.patlasov@gmail.com>
Signed-off-by: Maxim <maxim.patlasov@gmail.com>
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>

diff --git a/fs/gfs2/aops.c b/fs/gfs2/aops.c
index 4f36f88..aad77e4 100644
--- a/fs/gfs2/aops.c
+++ b/fs/gfs2/aops.c
@@ -695,6 +695,7 @@ out:
 	if (error == 0)
 		return 0;
 
+	unlock_page(page);
 	page_cache_release(page);
 
 	gfs2_trans_end(sdp);
-- 
1.7.4

[PATCH 15/15] GFS2: Don't use _raw version of RCU dereference

[Steven Whitehouse]

As per RCU glock patch review comments, don't use the _raw
version of this function here.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>

diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
index 8648409..85044b4 100644
--- a/fs/gfs2/glock.c
+++ b/fs/gfs2/glock.c
@@ -1723,7 +1723,7 @@ static inline struct gfs2_glock *glock_hash_chain(unsigned hash)
 
 static inline struct gfs2_glock *glock_hash_next(struct gfs2_glock *gl)
 {
-	return hlist_bl_entry(rcu_dereference_raw(gl->gl_list.next),
+	return hlist_bl_entry(rcu_dereference(gl->gl_list.next),
 			      struct gfs2_glock, gl_list);
 }
 
-- 
1.7.4