nfsd4: reshuffle lease-setting code to allow reuse
[safe/jmp/linux-2.6] / fs / gfs2 / glock.c
index befcda0..f426633 100644 (file)
@@ -1,6 +1,6 @@
 /*
  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
- * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
+ * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
  *
  * This copyrighted material is made available to anyone wishing to use,
  * modify, copy, or redistribute it subject to the terms and conditions
@@ -10,7 +10,6 @@
 #include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/spinlock.h>
-#include <linux/completion.h>
 #include <linux/buffer_head.h>
 #include <linux/delay.h>
 #include <linux/sort.h>
@@ -18,7 +17,6 @@
 #include <linux/kallsyms.h>
 #include <linux/gfs2_ondisk.h>
 #include <linux/list.h>
-#include <linux/lm_interface.h>
 #include <linux/wait.h>
 #include <linux/module.h>
 #include <linux/rwsem.h>
 #include "glock.h"
 #include "glops.h"
 #include "inode.h"
-#include "lm.h"
 #include "lops.h"
 #include "meta_io.h"
 #include "quota.h"
 #include "super.h"
 #include "util.h"
+#include "bmap.h"
+#define CREATE_TRACE_POINTS
+#include "trace_gfs2.h"
 
 struct gfs2_gl_hash_bucket {
         struct hlist_head hb_list;
 };
 
-struct glock_iter {
-       int hash;                     /* hash bucket index         */
-       struct gfs2_sbd *sdp;         /* incore superblock         */
-       struct gfs2_glock *gl;        /* current glock struct      */
-       struct seq_file *seq;         /* sequence file for debugfs */
-       char string[512];             /* scratch space             */
+struct gfs2_glock_iter {
+       int hash;                       /* hash bucket index         */
+       struct gfs2_sbd *sdp;           /* incore superblock         */
+       struct gfs2_glock *gl;          /* current glock struct      */
+       char string[512];               /* scratch space             */
 };
 
 typedef void (*glock_examiner) (struct gfs2_glock * gl);
 
 static int gfs2_dump_lockstate(struct gfs2_sbd *sdp);
-static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl);
-static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh);
-static void gfs2_glock_drop_th(struct gfs2_glock *gl);
-static void run_queue(struct gfs2_glock *gl);
+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);
 
 static DECLARE_RWSEM(gfs2_umount_flush_sem);
 static struct dentry *gfs2_root;
-static struct task_struct *scand_process;
-static unsigned int scand_secs = 5;
 static struct workqueue_struct *glock_workqueue;
+struct workqueue_struct *gfs2_delete_workqueue;
+static LIST_HEAD(lru_list);
+static atomic_t lru_count = ATOMIC_INIT(0);
+static DEFINE_SPINLOCK(lru_lock);
 
 #define GFS2_GL_HASH_SHIFT      15
 #define GFS2_GL_HASH_SIZE       (1 << GFS2_GL_HASH_SHIFT)
@@ -124,33 +124,6 @@ static inline rwlock_t *gl_lock_addr(unsigned int x)
 #endif
 
 /**
- * relaxed_state_ok - is a requested lock compatible with the current lock mode?
- * @actual: the current state of the lock
- * @requested: the lock state that was requested by the caller
- * @flags: the modifier flags passed in by the caller
- *
- * Returns: 1 if the locks are compatible, 0 otherwise
- */
-
-static inline int relaxed_state_ok(unsigned int actual, unsigned requested,
-                                  int flags)
-{
-       if (actual == requested)
-               return 1;
-
-       if (flags & GL_EXACT)
-               return 0;
-
-       if (actual == LM_ST_EXCLUSIVE && requested == LM_ST_SHARED)
-               return 1;
-
-       if (actual != LM_ST_UNLOCKED && (flags & LM_FLAG_ANY))
-               return 1;
-
-       return 0;
-}
-
-/**
  * gl_hash() - Turn glock number into hash bucket number
  * @lock: The glock number
  *
@@ -183,12 +156,10 @@ static void glock_free(struct gfs2_glock *gl)
        struct gfs2_sbd *sdp = gl->gl_sbd;
        struct inode *aspace = gl->gl_aspace;
 
-       gfs2_lm_put_lock(sdp, gl->gl_lock);
-
        if (aspace)
                gfs2_aspace_put(aspace);
-
-       kmem_cache_free(gfs2_glock_cachep, gl);
+       trace_gfs2_glock_put(gl);
+       sdp->sd_lockstruct.ls_ops->lm_put_lock(gfs2_glock_cachep, gl);
 }
 
 /**
@@ -197,12 +168,69 @@ static void glock_free(struct gfs2_glock *gl)
  *
  */
 
-static void gfs2_glock_hold(struct gfs2_glock *gl)
+void gfs2_glock_hold(struct gfs2_glock *gl)
 {
+       GLOCK_BUG_ON(gl, atomic_read(&gl->gl_ref) == 0);
        atomic_inc(&gl->gl_ref);
 }
 
 /**
+ * demote_ok - Check to see if it's ok to unlock a glock
+ * @gl: the glock
+ *
+ * Returns: 1 if it's ok
+ */
+
+static int demote_ok(const struct gfs2_glock *gl)
+{
+       const struct gfs2_glock_operations *glops = gl->gl_ops;
+
+       if (gl->gl_state == LM_ST_UNLOCKED)
+               return 0;
+       if (!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
+ * @gl: the glock
+ *
+ */
+
+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) {
+               list_add_tail(&gl->gl_lru, &lru_list);
+               atomic_inc(&lru_count);
+       }
+       spin_unlock(&lru_lock);
+}
+
+/**
+ * gfs2_glock_put_nolock() - Decrement reference count on glock
+ * @gl: The glock to put
+ *
+ * This function should only be used if the caller has its own reference
+ * to the glock, in addition to the one it is dropping.
+ */
+
+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);
+}
+
+/**
  * gfs2_glock_put() - Decrement reference count on glock
  * @gl: The glock to put
  *
@@ -211,21 +239,24 @@ static void gfs2_glock_hold(struct gfs2_glock *gl)
 int gfs2_glock_put(struct gfs2_glock *gl)
 {
        int rv = 0;
-       struct gfs2_sbd *sdp = gl->gl_sbd;
 
        write_lock(gl_lock_addr(gl->gl_hash));
-       if (atomic_dec_and_test(&gl->gl_ref)) {
+       if (atomic_dec_and_lock(&gl->gl_ref, &lru_lock)) {
                hlist_del(&gl->gl_list);
+               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));
-               gfs2_assert(sdp, gl->gl_state == LM_ST_UNLOCKED);
-               gfs2_assert(sdp, list_empty(&gl->gl_reclaim));
-               gfs2_assert(sdp, list_empty(&gl->gl_holders));
-               gfs2_assert(sdp, list_empty(&gl->gl_waiters1));
-               gfs2_assert(sdp, list_empty(&gl->gl_waiters3));
+               GLOCK_BUG_ON(gl, !list_empty(&gl->gl_holders));
                glock_free(gl);
                rv = 1;
                goto out;
        }
+       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;
@@ -261,819 +292,673 @@ static struct gfs2_glock *search_bucket(unsigned int hash,
 }
 
 /**
- * gfs2_glock_find() - Find glock by lock number
- * @sdp: The GFS2 superblock
- * @name: The lock name
+ * may_grant - check if its ok to grant a new lock
+ * @gl: The glock
+ * @gh: The lock request which we wish to grant
  *
- * Returns: NULL, or the struct gfs2_glock with the requested number
+ * Returns: true if its ok to grant the lock
  */
 
-static struct gfs2_glock *gfs2_glock_find(const struct gfs2_sbd *sdp,
-                                         const struct lm_lockname *name)
+static inline int may_grant(const struct gfs2_glock *gl, const struct gfs2_holder *gh)
 {
-       unsigned int hash = gl_hash(sdp, name);
-       struct gfs2_glock *gl;
-
-       read_lock(gl_lock_addr(hash));
-       gl = search_bucket(hash, sdp, name);
-       read_unlock(gl_lock_addr(hash));
-
-       return gl;
+       const struct gfs2_holder *gh_head = list_entry(gl->gl_holders.next, const struct gfs2_holder, gh_list);
+       if ((gh->gh_state == LM_ST_EXCLUSIVE ||
+            gh_head->gh_state == LM_ST_EXCLUSIVE) && gh != gh_head)
+               return 0;
+       if (gl->gl_state == gh->gh_state)
+               return 1;
+       if (gh->gh_flags & GL_EXACT)
+               return 0;
+       if (gl->gl_state == LM_ST_EXCLUSIVE) {
+               if (gh->gh_state == LM_ST_SHARED && gh_head->gh_state == LM_ST_SHARED)
+                       return 1;
+               if (gh->gh_state == LM_ST_DEFERRED && gh_head->gh_state == LM_ST_DEFERRED)
+                       return 1;
+       }
+       if (gl->gl_state != LM_ST_UNLOCKED && (gh->gh_flags & LM_FLAG_ANY))
+               return 1;
+       return 0;
 }
 
-static void glock_work_func(struct work_struct *work)
+static void gfs2_holder_wake(struct gfs2_holder *gh)
 {
-       struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
-
-       spin_lock(&gl->gl_spin);
-       if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags))
-               set_bit(GLF_DEMOTE, &gl->gl_flags);
-       run_queue(gl);
-       spin_unlock(&gl->gl_spin);
-       gfs2_glock_put(gl);
+       clear_bit(HIF_WAIT, &gh->gh_iflags);
+       smp_mb__after_clear_bit();
+       wake_up_bit(&gh->gh_iflags, HIF_WAIT);
 }
 
 /**
- * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
- * @sdp: The GFS2 superblock
- * @number: the lock number
- * @glops: The glock_operations to use
- * @create: If 0, don't create the glock if it doesn't exist
- * @glp: the glock is returned here
- *
- * This does not lock a glock, just finds/creates structures for one.
- *
- * Returns: errno
+ * do_promote - promote as many requests as possible on the current queue
+ * @gl: The glock
+ * 
+ * Returns: 1 if there is a blocked holder at the head of the list, or 2
+ *          if a type specific operation is underway.
  */
 
-int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
-                  const struct gfs2_glock_operations *glops, int create,
-                  struct gfs2_glock **glp)
+static int do_promote(struct gfs2_glock *gl)
+__releases(&gl->gl_spin)
+__acquires(&gl->gl_spin)
 {
-       struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
-       struct gfs2_glock *gl, *tmp;
-       unsigned int hash = gl_hash(sdp, &name);
-       int error;
-
-       read_lock(gl_lock_addr(hash));
-       gl = search_bucket(hash, sdp, &name);
-       read_unlock(gl_lock_addr(hash));
-
-       if (gl || !create) {
-               *glp = gl;
-               return 0;
-       }
-
-       gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
-       if (!gl)
-               return -ENOMEM;
-
-       gl->gl_flags = 0;
-       gl->gl_name = name;
-       atomic_set(&gl->gl_ref, 1);
-       gl->gl_state = LM_ST_UNLOCKED;
-       gl->gl_demote_state = LM_ST_EXCLUSIVE;
-       gl->gl_hash = hash;
-       gl->gl_owner_pid = NULL;
-       gl->gl_ip = 0;
-       gl->gl_ops = glops;
-       gl->gl_req_gh = NULL;
-       gl->gl_req_bh = NULL;
-       gl->gl_vn = 0;
-       gl->gl_stamp = jiffies;
-       gl->gl_tchange = jiffies;
-       gl->gl_object = NULL;
-       gl->gl_sbd = sdp;
-       gl->gl_aspace = NULL;
-       INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
+       const struct gfs2_glock_operations *glops = gl->gl_ops;
+       struct gfs2_holder *gh, *tmp;
+       int ret;
 
-       /* If this glock protects actual on-disk data or metadata blocks,
-          create a VFS inode to manage the pages/buffers holding them. */
-       if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
-               gl->gl_aspace = gfs2_aspace_get(sdp);
-               if (!gl->gl_aspace) {
-                       error = -ENOMEM;
-                       goto fail;
+restart:
+       list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
+               if (test_bit(HIF_HOLDER, &gh->gh_iflags))
+                       continue;
+               if (may_grant(gl, gh)) {
+                       if (gh->gh_list.prev == &gl->gl_holders &&
+                           glops->go_lock) {
+                               spin_unlock(&gl->gl_spin);
+                               /* FIXME: eliminate this eventually */
+                               ret = glops->go_lock(gh);
+                               spin_lock(&gl->gl_spin);
+                               if (ret) {
+                                       if (ret == 1)
+                                               return 2;
+                                       gh->gh_error = ret;
+                                       list_del_init(&gh->gh_list);
+                                       trace_gfs2_glock_queue(gh, 0);
+                                       gfs2_holder_wake(gh);
+                                       goto restart;
+                               }
+                               set_bit(HIF_HOLDER, &gh->gh_iflags);
+                               trace_gfs2_promote(gh, 1);
+                               gfs2_holder_wake(gh);
+                               goto restart;
+                       }
+                       set_bit(HIF_HOLDER, &gh->gh_iflags);
+                       trace_gfs2_promote(gh, 0);
+                       gfs2_holder_wake(gh);
+                       continue;
                }
+               if (gh->gh_list.prev == &gl->gl_holders)
+                       return 1;
+               break;
        }
-
-       error = gfs2_lm_get_lock(sdp, &name, &gl->gl_lock);
-       if (error)
-               goto fail_aspace;
-
-       write_lock(gl_lock_addr(hash));
-       tmp = search_bucket(hash, sdp, &name);
-       if (tmp) {
-               write_unlock(gl_lock_addr(hash));
-               glock_free(gl);
-               gl = tmp;
-       } else {
-               hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
-               write_unlock(gl_lock_addr(hash));
-       }
-
-       *glp = gl;
-
        return 0;
-
-fail_aspace:
-       if (gl->gl_aspace)
-               gfs2_aspace_put(gl->gl_aspace);
-fail:
-       kmem_cache_free(gfs2_glock_cachep, gl);
-       return error;
 }
 
 /**
- * gfs2_holder_init - initialize a struct gfs2_holder in the default way
- * @gl: the glock
- * @state: the state we're requesting
- * @flags: the modifier flags
- * @gh: the holder structure
+ * do_error - Something unexpected has happened during a lock request
  *
  */
 
-void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
-                     struct gfs2_holder *gh)
+static inline void do_error(struct gfs2_glock *gl, const int ret)
 {
-       INIT_LIST_HEAD(&gh->gh_list);
-       gh->gh_gl = gl;
-       gh->gh_ip = (unsigned long)__builtin_return_address(0);
-       gh->gh_owner_pid = get_pid(task_pid(current));
-       gh->gh_state = state;
-       gh->gh_flags = flags;
-       gh->gh_error = 0;
-       gh->gh_iflags = 0;
-       gfs2_glock_hold(gl);
+       struct gfs2_holder *gh, *tmp;
+
+       list_for_each_entry_safe(gh, tmp, &gl->gl_holders, gh_list) {
+               if (test_bit(HIF_HOLDER, &gh->gh_iflags))
+                       continue;
+               if (ret & LM_OUT_ERROR)
+                       gh->gh_error = -EIO;
+               else if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
+                       gh->gh_error = GLR_TRYFAILED;
+               else
+                       continue;
+               list_del_init(&gh->gh_list);
+               trace_gfs2_glock_queue(gh, 0);
+               gfs2_holder_wake(gh);
+       }
 }
 
 /**
- * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
- * @state: the state we're requesting
- * @flags: the modifier flags
- * @gh: the holder structure
- *
- * Don't mess with the glock.
- *
+ * find_first_waiter - find the first gh that's waiting for the glock
+ * @gl: the glock
  */
 
-void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
+static inline struct gfs2_holder *find_first_waiter(const struct gfs2_glock *gl)
 {
-       gh->gh_state = state;
-       gh->gh_flags = flags;
-       gh->gh_iflags = 0;
-       gh->gh_ip = (unsigned long)__builtin_return_address(0);
+       struct gfs2_holder *gh;
+
+       list_for_each_entry(gh, &gl->gl_holders, gh_list) {
+               if (!test_bit(HIF_HOLDER, &gh->gh_iflags))
+                       return gh;
+       }
+       return NULL;
 }
 
 /**
- * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
- * @gh: the holder structure
+ * state_change - record that the glock is now in a different state
+ * @gl: the glock
+ * @new_state the new state
  *
  */
 
-void gfs2_holder_uninit(struct gfs2_holder *gh)
+static void state_change(struct gfs2_glock *gl, unsigned int new_state)
 {
-       put_pid(gh->gh_owner_pid);
-       gfs2_glock_put(gh->gh_gl);
-       gh->gh_gl = NULL;
-       gh->gh_ip = 0;
-}
+       int held1, held2;
 
-static void gfs2_holder_wake(struct gfs2_holder *gh)
-{
-       clear_bit(HIF_WAIT, &gh->gh_iflags);
-       smp_mb__after_clear_bit();
-       wake_up_bit(&gh->gh_iflags, HIF_WAIT);
-}
+       held1 = (gl->gl_state != LM_ST_UNLOCKED);
+       held2 = (new_state != LM_ST_UNLOCKED);
 
-static int just_schedule(void *word)
-{
-        schedule();
-        return 0;
-}
+       if (held1 != held2) {
+               if (held2)
+                       gfs2_glock_hold(gl);
+               else
+                       gfs2_glock_put_nolock(gl);
+       }
 
-static void wait_on_holder(struct gfs2_holder *gh)
-{
-       might_sleep();
-       wait_on_bit(&gh->gh_iflags, HIF_WAIT, just_schedule, TASK_UNINTERRUPTIBLE);
+       gl->gl_state = new_state;
+       gl->gl_tchange = jiffies;
 }
 
 static void gfs2_demote_wake(struct gfs2_glock *gl)
 {
        gl->gl_demote_state = LM_ST_EXCLUSIVE;
-        clear_bit(GLF_DEMOTE, &gl->gl_flags);
-        smp_mb__after_clear_bit();
-        wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
-}
-
-static void wait_on_demote(struct gfs2_glock *gl)
-{
-       might_sleep();
-       wait_on_bit(&gl->gl_flags, GLF_DEMOTE, just_schedule, TASK_UNINTERRUPTIBLE);
+       clear_bit(GLF_DEMOTE, &gl->gl_flags);
+       smp_mb__after_clear_bit();
+       wake_up_bit(&gl->gl_flags, GLF_DEMOTE);
 }
 
 /**
- * rq_mutex - process a mutex request in the queue
- * @gh: the glock holder
+ * finish_xmote - The DLM has replied to one of our lock requests
+ * @gl: The glock
+ * @ret: The status from the DLM
  *
- * Returns: 1 if the queue is blocked
  */
 
-static int rq_mutex(struct gfs2_holder *gh)
+static void finish_xmote(struct gfs2_glock *gl, unsigned int ret)
 {
-       struct gfs2_glock *gl = gh->gh_gl;
-
-       list_del_init(&gh->gh_list);
-       /*  gh->gh_error never examined.  */
-       set_bit(GLF_LOCK, &gl->gl_flags);
-       clear_bit(HIF_WAIT, &gh->gh_iflags);
-       smp_mb();
-       wake_up_bit(&gh->gh_iflags, HIF_WAIT);
+       const struct gfs2_glock_operations *glops = gl->gl_ops;
+       struct gfs2_holder *gh;
+       unsigned state = ret & LM_OUT_ST_MASK;
+       int rv;
 
-       return 1;
-}
-
-/**
- * rq_promote - process a promote request in the queue
- * @gh: the glock holder
- *
- * Acquire a new inter-node lock, or change a lock state to more restrictive.
- *
- * Returns: 1 if the queue is blocked
- */
-
-static int rq_promote(struct gfs2_holder *gh)
-{
-       struct gfs2_glock *gl = gh->gh_gl;
+       spin_lock(&gl->gl_spin);
+       trace_gfs2_glock_state_change(gl, state);
+       state_change(gl, state);
+       gh = find_first_waiter(gl);
+
+       /* Demote to UN request arrived during demote to SH or DF */
+       if (test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags) &&
+           state != LM_ST_UNLOCKED && gl->gl_demote_state == LM_ST_UNLOCKED)
+               gl->gl_target = LM_ST_UNLOCKED;
+
+       /* Check for state != intended state */
+       if (unlikely(state != gl->gl_target)) {
+               if (gh && !test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags)) {
+                       /* move to back of queue and try next entry */
+                       if (ret & LM_OUT_CANCELED) {
+                               if ((gh->gh_flags & LM_FLAG_PRIORITY) == 0)
+                                       list_move_tail(&gh->gh_list, &gl->gl_holders);
+                               gh = find_first_waiter(gl);
+                               gl->gl_target = gh->gh_state;
+                               goto retry;
+                       }
+                       /* Some error or failed "try lock" - report it */
+                       if ((ret & LM_OUT_ERROR) ||
+                           (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))) {
+                               gl->gl_target = gl->gl_state;
+                               do_error(gl, ret);
+                               goto out;
+                       }
+               }
+               switch(state) {
+               /* Unlocked due to conversion deadlock, try again */
+               case LM_ST_UNLOCKED:
+retry:
+                       do_xmote(gl, gh, gl->gl_target);
+                       break;
+               /* Conversion fails, unlock and try again */
+               case LM_ST_SHARED:
+               case LM_ST_DEFERRED:
+                       do_xmote(gl, gh, LM_ST_UNLOCKED);
+                       break;
+               default: /* Everything else */
+                       printk(KERN_ERR "GFS2: wanted %u got %u\n", gl->gl_target, state);
+                       GLOCK_BUG_ON(gl, 1);
+               }
+               spin_unlock(&gl->gl_spin);
+               return;
+       }
 
-       if (!relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
-               if (list_empty(&gl->gl_holders)) {
-                       gl->gl_req_gh = gh;
-                       set_bit(GLF_LOCK, &gl->gl_flags);
+       /* Fast path - we got what we asked for */
+       if (test_and_clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags))
+               gfs2_demote_wake(gl);
+       if (state != LM_ST_UNLOCKED) {
+               if (glops->go_xmote_bh) {
                        spin_unlock(&gl->gl_spin);
-                       gfs2_glock_xmote_th(gh->gh_gl, gh);
+                       rv = glops->go_xmote_bh(gl, gh);
                        spin_lock(&gl->gl_spin);
+                       if (rv) {
+                               do_error(gl, rv);
+                               goto out;
+                       }
                }
-               return 1;
-       }
-
-       if (list_empty(&gl->gl_holders)) {
-               set_bit(HIF_FIRST, &gh->gh_iflags);
-               set_bit(GLF_LOCK, &gl->gl_flags);
-       } else {
-               struct gfs2_holder *next_gh;
-               if (gh->gh_state == LM_ST_EXCLUSIVE)
-                       return 1;
-               next_gh = list_entry(gl->gl_holders.next, struct gfs2_holder,
-                                    gh_list);
-               if (next_gh->gh_state == LM_ST_EXCLUSIVE)
-                        return 1;
+               rv = do_promote(gl);
+               if (rv == 2)
+                       goto out_locked;
        }
+out:
+       clear_bit(GLF_LOCK, &gl->gl_flags);
+out_locked:
+       spin_unlock(&gl->gl_spin);
+}
 
-       list_move_tail(&gh->gh_list, &gl->gl_holders);
-       gh->gh_error = 0;
-       set_bit(HIF_HOLDER, &gh->gh_iflags);
+static unsigned int gfs2_lm_lock(struct gfs2_sbd *sdp, void *lock,
+                                unsigned int req_state,
+                                unsigned int flags)
+{
+       int ret = LM_OUT_ERROR;
 
-       gfs2_holder_wake(gh);
+       if (!sdp->sd_lockstruct.ls_ops->lm_lock)
+               return req_state == LM_ST_UNLOCKED ? 0 : req_state;
 
-       return 0;
+       if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
+               ret = sdp->sd_lockstruct.ls_ops->lm_lock(lock,
+                                                        req_state, flags);
+       return ret;
 }
 
 /**
- * rq_demote - process a demote request in the queue
- * @gh: the glock holder
+ * do_xmote - Calls the DLM to change the state of a lock
+ * @gl: The lock state
+ * @gh: The holder (only for promotes)
+ * @target: The target lock state
  *
- * Returns: 1 if the queue is blocked
  */
 
-static int rq_demote(struct gfs2_glock *gl)
+static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int target)
+__releases(&gl->gl_spin)
+__acquires(&gl->gl_spin)
 {
-       if (!list_empty(&gl->gl_holders))
-               return 1;
+       const struct gfs2_glock_operations *glops = gl->gl_ops;
+       struct gfs2_sbd *sdp = gl->gl_sbd;
+       unsigned int lck_flags = gh ? gh->gh_flags : 0;
+       int ret;
 
-       if (gl->gl_state == gl->gl_demote_state ||
-           gl->gl_state == LM_ST_UNLOCKED) {
-               gfs2_demote_wake(gl);
-               return 0;
+       lck_flags &= (LM_FLAG_TRY | LM_FLAG_TRY_1CB | LM_FLAG_NOEXP |
+                     LM_FLAG_PRIORITY);
+       BUG_ON(gl->gl_state == target);
+       BUG_ON(gl->gl_state == gl->gl_target);
+       if ((target == LM_ST_UNLOCKED || target == LM_ST_DEFERRED) &&
+           glops->go_inval) {
+               set_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
+               do_error(gl, 0); /* Fail queued try locks */
        }
+       spin_unlock(&gl->gl_spin);
+       if (glops->go_xmote_th)
+               glops->go_xmote_th(gl);
+       if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
+               glops->go_inval(gl, target == LM_ST_DEFERRED ? 0 : DIO_METADATA);
+       clear_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags);
 
-       set_bit(GLF_LOCK, &gl->gl_flags);
-       set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
-
-       if (gl->gl_demote_state == LM_ST_UNLOCKED ||
-           gl->gl_state != LM_ST_EXCLUSIVE) {
-               spin_unlock(&gl->gl_spin);
-               gfs2_glock_drop_th(gl);
+       gfs2_glock_hold(gl);
+       if (target != LM_ST_UNLOCKED && (gl->gl_state == LM_ST_SHARED ||
+           gl->gl_state == LM_ST_DEFERRED) &&
+           !(lck_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
+               lck_flags |= LM_FLAG_TRY_1CB;
+       ret = gfs2_lm_lock(sdp, gl, target, lck_flags);
+
+       if (!(ret & LM_OUT_ASYNC)) {
+               finish_xmote(gl, ret);
+               if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
+                       gfs2_glock_put(gl);
        } else {
-               spin_unlock(&gl->gl_spin);
-               gfs2_glock_xmote_th(gl, NULL);
+               GLOCK_BUG_ON(gl, ret != LM_OUT_ASYNC);
        }
-
        spin_lock(&gl->gl_spin);
-       clear_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
-
-       return 0;
 }
 
 /**
- * run_queue - process holder structures on a glock
+ * find_first_holder - find the first "holder" gh
  * @gl: the glock
- *
  */
-static void run_queue(struct gfs2_glock *gl)
+
+static inline struct gfs2_holder *find_first_holder(const struct gfs2_glock *gl)
 {
        struct gfs2_holder *gh;
-       int blocked = 1;
-
-       for (;;) {
-               if (test_bit(GLF_LOCK, &gl->gl_flags))
-                       break;
 
-               if (!list_empty(&gl->gl_waiters1)) {
-                       gh = list_entry(gl->gl_waiters1.next,
-                                       struct gfs2_holder, gh_list);
-                       blocked = rq_mutex(gh);
-               } else if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
-                       blocked = rq_demote(gl);
-                       if (gl->gl_waiters2 && !blocked) {
-                               set_bit(GLF_DEMOTE, &gl->gl_flags);
-                               gl->gl_demote_state = LM_ST_UNLOCKED;
-                       }
-                       gl->gl_waiters2 = 0;
-               } else if (!list_empty(&gl->gl_waiters3)) {
-                       gh = list_entry(gl->gl_waiters3.next,
-                                       struct gfs2_holder, gh_list);
-                       blocked = rq_promote(gh);
-               } else
-                       break;
-
-               if (blocked)
-                       break;
+       if (!list_empty(&gl->gl_holders)) {
+               gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
+               if (test_bit(HIF_HOLDER, &gh->gh_iflags))
+                       return gh;
        }
+       return NULL;
 }
 
 /**
- * gfs2_glmutex_lock - acquire a local lock on a glock
- * @gl: the glock
+ * run_queue - do all outstanding tasks related to a glock
+ * @gl: The glock in question
+ * @nonblock: True if we must not block in run_queue
  *
- * Gives caller exclusive access to manipulate a glock structure.
  */
 
-static void gfs2_glmutex_lock(struct gfs2_glock *gl)
+static void run_queue(struct gfs2_glock *gl, const int nonblock)
+__releases(&gl->gl_spin)
+__acquires(&gl->gl_spin)
 {
-       spin_lock(&gl->gl_spin);
-       if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
-               struct gfs2_holder gh;
-
-               gfs2_holder_init(gl, 0, 0, &gh);
-               set_bit(HIF_WAIT, &gh.gh_iflags);
-               list_add_tail(&gh.gh_list, &gl->gl_waiters1);
-               spin_unlock(&gl->gl_spin);
-               wait_on_holder(&gh);
-               gfs2_holder_uninit(&gh);
-       } else {
-               gl->gl_owner_pid = get_pid(task_pid(current));
-               gl->gl_ip = (unsigned long)__builtin_return_address(0);
-               spin_unlock(&gl->gl_spin);
-       }
-}
-
-/**
- * gfs2_glmutex_trylock - try to acquire a local lock on a glock
- * @gl: the glock
- *
- * Returns: 1 if the glock is acquired
- */
+       struct gfs2_holder *gh = NULL;
+       int ret;
 
-static int gfs2_glmutex_trylock(struct gfs2_glock *gl)
-{
-       int acquired = 1;
+       if (test_and_set_bit(GLF_LOCK, &gl->gl_flags))
+               return;
 
-       spin_lock(&gl->gl_spin);
-       if (test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
-               acquired = 0;
+       GLOCK_BUG_ON(gl, test_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags));
+
+       if (test_bit(GLF_DEMOTE, &gl->gl_flags) &&
+           gl->gl_demote_state != gl->gl_state) {
+               if (find_first_holder(gl))
+                       goto out_unlock;
+               if (nonblock)
+                       goto out_sched;
+               set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
+               GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
+               gl->gl_target = gl->gl_demote_state;
        } else {
-               gl->gl_owner_pid = get_pid(task_pid(current));
-               gl->gl_ip = (unsigned long)__builtin_return_address(0);
+               if (test_bit(GLF_DEMOTE, &gl->gl_flags))
+                       gfs2_demote_wake(gl);
+               ret = do_promote(gl);
+               if (ret == 0)
+                       goto out_unlock;
+               if (ret == 2)
+                       goto out;
+               gh = find_first_waiter(gl);
+               gl->gl_target = gh->gh_state;
+               if (!(gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)))
+                       do_error(gl, 0); /* Fail queued try locks */
        }
-       spin_unlock(&gl->gl_spin);
+       do_xmote(gl, gh, gl->gl_target);
+out:
+       return;
 
-       return acquired;
-}
+out_sched:
+       clear_bit(GLF_LOCK, &gl->gl_flags);
+       smp_mb__after_clear_bit();
+       gfs2_glock_hold(gl);
+       if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
+               gfs2_glock_put_nolock(gl);
+       return;
 
-/**
- * gfs2_glmutex_unlock - release a local lock on a glock
- * @gl: the glock
- *
- */
+out_unlock:
+       clear_bit(GLF_LOCK, &gl->gl_flags);
+       smp_mb__after_clear_bit();
+       return;
+}
 
-static void gfs2_glmutex_unlock(struct gfs2_glock *gl)
+static void delete_work_func(struct work_struct *work)
 {
-       struct pid *pid;
+       struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_delete);
+       struct gfs2_sbd *sdp = gl->gl_sbd;
+       struct gfs2_inode *ip = NULL;
+       struct inode *inode;
+       u64 no_addr = 0;
 
        spin_lock(&gl->gl_spin);
-       clear_bit(GLF_LOCK, &gl->gl_flags);
-       pid = gl->gl_owner_pid;
-       gl->gl_owner_pid = NULL;
-       gl->gl_ip = 0;
-       run_queue(gl);
+       ip = (struct gfs2_inode *)gl->gl_object;
+       if (ip)
+               no_addr = ip->i_no_addr;
        spin_unlock(&gl->gl_spin);
-
-       put_pid(pid);
+       if (ip) {
+               inode = gfs2_ilookup(sdp->sd_vfs, no_addr);
+               if (inode) {
+                       d_prune_aliases(inode);
+                       iput(inode);
+               }
+       }
+       gfs2_glock_put(gl);
 }
 
-/**
- * handle_callback - process a demote request
- * @gl: the glock
- * @state: the state the caller wants us to change to
- *
- * There are only two requests that we are going to see in actual
- * practise: LM_ST_SHARED and LM_ST_UNLOCKED
- */
-
-static void handle_callback(struct gfs2_glock *gl, unsigned int state,
-                           int remote, unsigned long delay)
+static void glock_work_func(struct work_struct *work)
 {
-       int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
+       unsigned long delay = 0;
+       struct gfs2_glock *gl = container_of(work, struct gfs2_glock, gl_work.work);
+       int drop_ref = 0;
 
+       if (test_and_clear_bit(GLF_REPLY_PENDING, &gl->gl_flags)) {
+               finish_xmote(gl, gl->gl_reply);
+               drop_ref = 1;
+       }
+       down_read(&gfs2_umount_flush_sem);
        spin_lock(&gl->gl_spin);
-       set_bit(bit, &gl->gl_flags);
-       if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
-               gl->gl_demote_state = state;
-               gl->gl_demote_time = jiffies;
-               if (remote && gl->gl_ops->go_type == LM_TYPE_IOPEN &&
-                   gl->gl_object) {
-                       gfs2_glock_schedule_for_reclaim(gl);
-                       spin_unlock(&gl->gl_spin);
-                       return;
-               }
-       } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
-                       gl->gl_demote_state != state) {
-               if (test_bit(GLF_DEMOTE_IN_PROGRESS,  &gl->gl_flags)) 
-                       gl->gl_waiters2 = 1;
-               else 
-                       gl->gl_demote_state = LM_ST_UNLOCKED;
+       if (test_and_clear_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
+           gl->gl_state != LM_ST_UNLOCKED &&
+           gl->gl_demote_state != LM_ST_EXCLUSIVE) {
+               unsigned long holdtime, now = jiffies;
+               holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
+               if (time_before(now, holdtime))
+                       delay = holdtime - now;
+               set_bit(delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE, &gl->gl_flags);
        }
+       run_queue(gl, 0);
        spin_unlock(&gl->gl_spin);
+       up_read(&gfs2_umount_flush_sem);
+       if (!delay ||
+           queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
+               gfs2_glock_put(gl);
+       if (drop_ref)
+               gfs2_glock_put(gl);
 }
 
 /**
- * state_change - record that the glock is now in a different state
- * @gl: the glock
- * @new_state the new state
+ * gfs2_glock_get() - Get a glock, or create one if one doesn't exist
+ * @sdp: The GFS2 superblock
+ * @number: the lock number
+ * @glops: The glock_operations to use
+ * @create: If 0, don't create the glock if it doesn't exist
+ * @glp: the glock is returned here
+ *
+ * This does not lock a glock, just finds/creates structures for one.
  *
+ * Returns: errno
  */
 
-static void state_change(struct gfs2_glock *gl, unsigned int new_state)
+int gfs2_glock_get(struct gfs2_sbd *sdp, u64 number,
+                  const struct gfs2_glock_operations *glops, int create,
+                  struct gfs2_glock **glp)
 {
-       int held1, held2;
+       struct lm_lockname name = { .ln_number = number, .ln_type = glops->go_type };
+       struct gfs2_glock *gl, *tmp;
+       unsigned int hash = gl_hash(sdp, &name);
+       int error;
 
-       held1 = (gl->gl_state != LM_ST_UNLOCKED);
-       held2 = (new_state != LM_ST_UNLOCKED);
+       read_lock(gl_lock_addr(hash));
+       gl = search_bucket(hash, sdp, &name);
+       read_unlock(gl_lock_addr(hash));
 
-       if (held1 != held2) {
-               if (held2)
-                       gfs2_glock_hold(gl);
-               else
-                       gfs2_glock_put(gl);
-       }
+       *glp = gl;
+       if (gl)
+               return 0;
+       if (!create)
+               return -ENOENT;
 
-       gl->gl_state = new_state;
-       gl->gl_tchange = jiffies;
-}
+       gl = kmem_cache_alloc(gfs2_glock_cachep, GFP_KERNEL);
+       if (!gl)
+               return -ENOMEM;
 
-/**
- * xmote_bh - Called after the lock module is done acquiring a lock
- * @gl: The glock in question
- * @ret: the int returned from the lock module
- *
- */
+       atomic_inc(&sdp->sd_glock_disposal);
+       gl->gl_flags = 0;
+       gl->gl_name = name;
+       atomic_set(&gl->gl_ref, 1);
+       gl->gl_state = LM_ST_UNLOCKED;
+       gl->gl_target = LM_ST_UNLOCKED;
+       gl->gl_demote_state = LM_ST_EXCLUSIVE;
+       gl->gl_hash = hash;
+       gl->gl_ops = glops;
+       snprintf(gl->gl_strname, GDLM_STRNAME_BYTES, "%8x%16llx", name.ln_type, (unsigned long long)number);
+       memset(&gl->gl_lksb, 0, sizeof(struct dlm_lksb));
+       gl->gl_lksb.sb_lvbptr = gl->gl_lvb;
+       gl->gl_tchange = jiffies;
+       gl->gl_object = NULL;
+       gl->gl_sbd = sdp;
+       gl->gl_aspace = NULL;
+       INIT_DELAYED_WORK(&gl->gl_work, glock_work_func);
+       INIT_WORK(&gl->gl_delete, delete_work_func);
 
-static void xmote_bh(struct gfs2_glock *gl, unsigned int ret)
-{
-       struct gfs2_sbd *sdp = gl->gl_sbd;
-       const struct gfs2_glock_operations *glops = gl->gl_ops;
-       struct gfs2_holder *gh = gl->gl_req_gh;
-       int prev_state = gl->gl_state;
-       int op_done = 1;
-
-       gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
-       gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
-       gfs2_assert_warn(sdp, !(ret & LM_OUT_ASYNC));
-
-       state_change(gl, ret & LM_OUT_ST_MASK);
-
-       if (prev_state != LM_ST_UNLOCKED && !(ret & LM_OUT_CACHEABLE)) {
-               if (glops->go_inval)
-                       glops->go_inval(gl, DIO_METADATA);
-       } else if (gl->gl_state == LM_ST_DEFERRED) {
-               /* We might not want to do this here.
-                  Look at moving to the inode glops. */
-               if (glops->go_inval)
-                       glops->go_inval(gl, 0);
+       /* If this glock protects actual on-disk data or metadata blocks,
+          create a VFS inode to manage the pages/buffers holding them. */
+       if (glops == &gfs2_inode_glops || glops == &gfs2_rgrp_glops) {
+               gl->gl_aspace = gfs2_aspace_get(sdp);
+               if (!gl->gl_aspace) {
+                       error = -ENOMEM;
+                       goto fail;
+               }
        }
 
-       /*  Deal with each possible exit condition  */
-
-       if (!gh) {
-               gl->gl_stamp = jiffies;
-               if (ret & LM_OUT_CANCELED) {
-                       op_done = 0;
-               } else {
-                       spin_lock(&gl->gl_spin);
-                       if (gl->gl_state != gl->gl_demote_state) {
-                               gl->gl_req_bh = NULL;
-                               spin_unlock(&gl->gl_spin);
-                               gfs2_glock_drop_th(gl);
-                               gfs2_glock_put(gl);
-                               return;
-                       }
-                       gfs2_demote_wake(gl);
-                       spin_unlock(&gl->gl_spin);
-               }
+       write_lock(gl_lock_addr(hash));
+       tmp = search_bucket(hash, sdp, &name);
+       if (tmp) {
+               write_unlock(gl_lock_addr(hash));
+               glock_free(gl);
+               gl = tmp;
        } else {
-               spin_lock(&gl->gl_spin);
-               list_del_init(&gh->gh_list);
-               gh->gh_error = -EIO;
-               if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) 
-                       goto out;
-               gh->gh_error = GLR_CANCELED;
-               if (ret & LM_OUT_CANCELED) 
-                       goto out;
-               if (relaxed_state_ok(gl->gl_state, gh->gh_state, gh->gh_flags)) {
-                       list_add_tail(&gh->gh_list, &gl->gl_holders);
-                       gh->gh_error = 0;
-                       set_bit(HIF_HOLDER, &gh->gh_iflags);
-                       set_bit(HIF_FIRST, &gh->gh_iflags);
-                       op_done = 0;
-                       goto out;
-               }
-               gh->gh_error = GLR_TRYFAILED;
-               if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB))
-                       goto out;
-               gh->gh_error = -EINVAL;
-               if (gfs2_assert_withdraw(sdp, 0) == -1)
-                       fs_err(sdp, "ret = 0x%.8X\n", ret);
-out:
-               spin_unlock(&gl->gl_spin);
+               hlist_add_head(&gl->gl_list, &gl_hash_table[hash].hb_list);
+               write_unlock(gl_lock_addr(hash));
        }
 
-       if (glops->go_xmote_bh)
-               glops->go_xmote_bh(gl);
-
-       if (op_done) {
-               spin_lock(&gl->gl_spin);
-               gl->gl_req_gh = NULL;
-               gl->gl_req_bh = NULL;
-               clear_bit(GLF_LOCK, &gl->gl_flags);
-               spin_unlock(&gl->gl_spin);
-       }
+       *glp = gl;
 
-       gfs2_glock_put(gl);
+       return 0;
 
-       if (gh)
-               gfs2_holder_wake(gh);
+fail:
+       kmem_cache_free(gfs2_glock_cachep, gl);
+       return error;
 }
 
 /**
- * gfs2_glock_xmote_th - Call into the lock module to acquire or change a glock
- * @gl: The glock in question
- * @state: the requested state
- * @flags: modifier flags to the lock call
+ * gfs2_holder_init - initialize a struct gfs2_holder in the default way
+ * @gl: the glock
+ * @state: the state we're requesting
+ * @flags: the modifier flags
+ * @gh: the holder structure
  *
  */
 
-static void gfs2_glock_xmote_th(struct gfs2_glock *gl, struct gfs2_holder *gh)
+void gfs2_holder_init(struct gfs2_glock *gl, unsigned int state, unsigned flags,
+                     struct gfs2_holder *gh)
 {
-       struct gfs2_sbd *sdp = gl->gl_sbd;
-       int flags = gh ? gh->gh_flags : 0;
-       unsigned state = gh ? gh->gh_state : gl->gl_demote_state;
-       const struct gfs2_glock_operations *glops = gl->gl_ops;
-       int lck_flags = flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB |
-                                LM_FLAG_NOEXP | LM_FLAG_ANY |
-                                LM_FLAG_PRIORITY);
-       unsigned int lck_ret;
-
-       if (glops->go_xmote_th)
-               glops->go_xmote_th(gl);
-
-       gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
-       gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
-       gfs2_assert_warn(sdp, state != LM_ST_UNLOCKED);
-       gfs2_assert_warn(sdp, state != gl->gl_state);
-
+       INIT_LIST_HEAD(&gh->gh_list);
+       gh->gh_gl = gl;
+       gh->gh_ip = (unsigned long)__builtin_return_address(0);
+       gh->gh_owner_pid = get_pid(task_pid(current));
+       gh->gh_state = state;
+       gh->gh_flags = flags;
+       gh->gh_error = 0;
+       gh->gh_iflags = 0;
        gfs2_glock_hold(gl);
-       gl->gl_req_bh = xmote_bh;
-
-       lck_ret = gfs2_lm_lock(sdp, gl->gl_lock, gl->gl_state, state, lck_flags);
-
-       if (gfs2_assert_withdraw(sdp, !(lck_ret & LM_OUT_ERROR)))
-               return;
-
-       if (lck_ret & LM_OUT_ASYNC)
-               gfs2_assert_warn(sdp, lck_ret == LM_OUT_ASYNC);
-       else
-               xmote_bh(gl, lck_ret);
 }
 
 /**
- * drop_bh - Called after a lock module unlock completes
- * @gl: the glock
- * @ret: the return status
+ * gfs2_holder_reinit - reinitialize a struct gfs2_holder so we can requeue it
+ * @state: the state we're requesting
+ * @flags: the modifier flags
+ * @gh: the holder structure
  *
- * Doesn't wake up the process waiting on the struct gfs2_holder (if any)
- * Doesn't drop the reference on the glock the top half took out
+ * Don't mess with the glock.
  *
  */
 
-static void drop_bh(struct gfs2_glock *gl, unsigned int ret)
+void gfs2_holder_reinit(unsigned int state, unsigned flags, struct gfs2_holder *gh)
 {
-       struct gfs2_sbd *sdp = gl->gl_sbd;
-       const struct gfs2_glock_operations *glops = gl->gl_ops;
-       struct gfs2_holder *gh = gl->gl_req_gh;
-
-       gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
-       gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
-       gfs2_assert_warn(sdp, !ret);
-
-       state_change(gl, LM_ST_UNLOCKED);
-
-       if (glops->go_inval)
-               glops->go_inval(gl, DIO_METADATA);
-
-       if (gh) {
-               spin_lock(&gl->gl_spin);
-               list_del_init(&gh->gh_list);
-               gh->gh_error = 0;
-               spin_unlock(&gl->gl_spin);
-       }
-
-       spin_lock(&gl->gl_spin);
-       gfs2_demote_wake(gl);
-       gl->gl_req_gh = NULL;
-       gl->gl_req_bh = NULL;
-       clear_bit(GLF_LOCK, &gl->gl_flags);
-       spin_unlock(&gl->gl_spin);
-
-       gfs2_glock_put(gl);
-
-       if (gh)
-               gfs2_holder_wake(gh);
+       gh->gh_state = state;
+       gh->gh_flags = flags;
+       gh->gh_iflags = 0;
+       gh->gh_ip = (unsigned long)__builtin_return_address(0);
 }
 
 /**
- * gfs2_glock_drop_th - call into the lock module to unlock a lock
- * @gl: the glock
+ * gfs2_holder_uninit - uninitialize a holder structure (drop glock reference)
+ * @gh: the holder structure
  *
  */
 
-static void gfs2_glock_drop_th(struct gfs2_glock *gl)
+void gfs2_holder_uninit(struct gfs2_holder *gh)
 {
-       struct gfs2_sbd *sdp = gl->gl_sbd;
-       const struct gfs2_glock_operations *glops = gl->gl_ops;
-       unsigned int ret;
-
-       if (glops->go_xmote_th)
-               glops->go_xmote_th(gl);
+       put_pid(gh->gh_owner_pid);
+       gfs2_glock_put(gh->gh_gl);
+       gh->gh_gl = NULL;
+       gh->gh_ip = 0;
+}
 
-       gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
-       gfs2_assert_warn(sdp, list_empty(&gl->gl_holders));
-       gfs2_assert_warn(sdp, gl->gl_state != LM_ST_UNLOCKED);
+/**
+ * gfs2_glock_holder_wait
+ * @word: unused
+ *
+ * This function and gfs2_glock_demote_wait both show up in the WCHAN
+ * field. Thus I've separated these otherwise identical functions in
+ * order to be more informative to the user.
+ */
 
-       gfs2_glock_hold(gl);
-       gl->gl_req_bh = drop_bh;
+static int gfs2_glock_holder_wait(void *word)
+{
+        schedule();
+        return 0;
+}
 
-       ret = gfs2_lm_unlock(sdp, gl->gl_lock, gl->gl_state);
+static int gfs2_glock_demote_wait(void *word)
+{
+       schedule();
+       return 0;
+}
 
-       if (gfs2_assert_withdraw(sdp, !(ret & LM_OUT_ERROR)))
-               return;
+static void wait_on_holder(struct gfs2_holder *gh)
+{
+       might_sleep();
+       wait_on_bit(&gh->gh_iflags, HIF_WAIT, gfs2_glock_holder_wait, TASK_UNINTERRUPTIBLE);
+}
 
-       if (!ret)
-               drop_bh(gl, ret);
-       else
-               gfs2_assert_warn(sdp, ret == LM_OUT_ASYNC);
+static void wait_on_demote(struct gfs2_glock *gl)
+{
+       might_sleep();
+       wait_on_bit(&gl->gl_flags, GLF_DEMOTE, gfs2_glock_demote_wait, TASK_UNINTERRUPTIBLE);
 }
 
 /**
- * do_cancels - cancel requests for locks stuck waiting on an expire flag
- * @gh: the LM_FLAG_PRIORITY holder waiting to acquire the lock
+ * handle_callback - process a demote request
+ * @gl: the glock
+ * @state: the state the caller wants us to change to
  *
- * Don't cancel GL_NOCANCEL requests.
+ * There are only two requests that we are going to see in actual
+ * practise: LM_ST_SHARED and LM_ST_UNLOCKED
  */
 
-static void do_cancels(struct gfs2_holder *gh)
+static void handle_callback(struct gfs2_glock *gl, unsigned int state,
+                           unsigned long delay)
 {
-       struct gfs2_glock *gl = gh->gh_gl;
-
-       spin_lock(&gl->gl_spin);
+       int bit = delay ? GLF_PENDING_DEMOTE : GLF_DEMOTE;
 
-       while (gl->gl_req_gh != gh &&
-              !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
-              !list_empty(&gh->gh_list)) {
-               if (gl->gl_req_bh && !(gl->gl_req_gh &&
-                                    (gl->gl_req_gh->gh_flags & GL_NOCANCEL))) {
-                       spin_unlock(&gl->gl_spin);
-                       gfs2_lm_cancel(gl->gl_sbd, gl->gl_lock);
-                       msleep(100);
-                       spin_lock(&gl->gl_spin);
-               } else {
-                       spin_unlock(&gl->gl_spin);
-                       msleep(100);
-                       spin_lock(&gl->gl_spin);
-               }
+       set_bit(bit, &gl->gl_flags);
+       if (gl->gl_demote_state == LM_ST_EXCLUSIVE) {
+               gl->gl_demote_state = state;
+               gl->gl_demote_time = jiffies;
+       } else if (gl->gl_demote_state != LM_ST_UNLOCKED &&
+                       gl->gl_demote_state != state) {
+               gl->gl_demote_state = LM_ST_UNLOCKED;
        }
-
-       spin_unlock(&gl->gl_spin);
+       if (gl->gl_ops->go_callback)
+               gl->gl_ops->go_callback(gl);
+       trace_gfs2_demote_rq(gl);
 }
 
 /**
- * glock_wait_internal - wait on a glock acquisition
+ * gfs2_glock_wait - wait on a glock acquisition
  * @gh: the glock holder
  *
  * Returns: 0 on success
  */
 
-static int glock_wait_internal(struct gfs2_holder *gh)
+int gfs2_glock_wait(struct gfs2_holder *gh)
 {
-       struct gfs2_glock *gl = gh->gh_gl;
-       struct gfs2_sbd *sdp = gl->gl_sbd;
-       const struct gfs2_glock_operations *glops = gl->gl_ops;
-
-       if (test_bit(HIF_ABORTED, &gh->gh_iflags))
-               return -EIO;
-
-       if (gh->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
-               spin_lock(&gl->gl_spin);
-               if (gl->gl_req_gh != gh &&
-                   !test_bit(HIF_HOLDER, &gh->gh_iflags) &&
-                   !list_empty(&gh->gh_list)) {
-                       list_del_init(&gh->gh_list);
-                       gh->gh_error = GLR_TRYFAILED;
-                       run_queue(gl);
-                       spin_unlock(&gl->gl_spin);
-                       return gh->gh_error;
-               }
-               spin_unlock(&gl->gl_spin);
-       }
-
-       if (gh->gh_flags & LM_FLAG_PRIORITY)
-               do_cancels(gh);
-
        wait_on_holder(gh);
-       if (gh->gh_error)
-               return gh->gh_error;
-
-       gfs2_assert_withdraw(sdp, test_bit(HIF_HOLDER, &gh->gh_iflags));
-       gfs2_assert_withdraw(sdp, relaxed_state_ok(gl->gl_state, gh->gh_state,
-                                                  gh->gh_flags));
-
-       if (test_bit(HIF_FIRST, &gh->gh_iflags)) {
-               gfs2_assert_warn(sdp, test_bit(GLF_LOCK, &gl->gl_flags));
-
-               if (glops->go_lock) {
-                       gh->gh_error = glops->go_lock(gh);
-                       if (gh->gh_error) {
-                               spin_lock(&gl->gl_spin);
-                               list_del_init(&gh->gh_list);
-                               spin_unlock(&gl->gl_spin);
-                       }
-               }
-
-               spin_lock(&gl->gl_spin);
-               gl->gl_req_gh = NULL;
-               gl->gl_req_bh = NULL;
-               clear_bit(GLF_LOCK, &gl->gl_flags);
-               run_queue(gl);
-               spin_unlock(&gl->gl_spin);
-       }
-
        return gh->gh_error;
 }
 
-static inline struct gfs2_holder *
-find_holder_by_owner(struct list_head *head, struct pid *pid)
-{
-       struct gfs2_holder *gh;
-
-       list_for_each_entry(gh, head, gh_list) {
-               if (gh->gh_owner_pid == pid)
-                       return gh;
-       }
-
-       return NULL;
-}
-
-static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
+void gfs2_print_dbg(struct seq_file *seq, const char *fmt, ...)
 {
        va_list args;
 
        va_start(args, fmt);
-       if (gi) {
+       if (seq) {
+               struct gfs2_glock_iter *gi = seq->private;
                vsprintf(gi->string, fmt, args);
-               seq_printf(gi->seq, gi->string);
-       }
-       else
+               seq_printf(seq, gi->string);
+       } else {
+               printk(KERN_ERR " ");
                vprintk(fmt, args);
+       }
        va_end(args);
 }
 
@@ -1081,50 +966,79 @@ static void print_dbg(struct glock_iter *gi, const char *fmt, ...)
  * add_to_queue - Add a holder to the wait queue (but look for recursion)
  * @gh: the holder structure to add
  *
+ * Eventually we should move the recursive locking trap to a
+ * debugging option or something like that. This is the fast
+ * path and needs to have the minimum number of distractions.
+ * 
  */
 
-static void add_to_queue(struct gfs2_holder *gh)
+static inline void add_to_queue(struct gfs2_holder *gh)
+__releases(&gl->gl_spin)
+__acquires(&gl->gl_spin)
 {
        struct gfs2_glock *gl = gh->gh_gl;
-       struct gfs2_holder *existing;
+       struct gfs2_sbd *sdp = gl->gl_sbd;
+       struct list_head *insert_pt = NULL;
+       struct gfs2_holder *gh2;
+       int try_lock = 0;
 
        BUG_ON(gh->gh_owner_pid == NULL);
        if (test_and_set_bit(HIF_WAIT, &gh->gh_iflags))
                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", 
-                                    existing->gh_ip);
-                       printk(KERN_INFO "pid : %d\n",
-                                       pid_nr(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",
-                                       pid_nr(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 & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) {
+               if (test_bit(GLF_LOCK, &gl->gl_flags))
+                       try_lock = 1;
+               if (test_bit(GLF_INVALIDATE_IN_PROGRESS, &gl->gl_flags))
+                       goto fail;
+       }
+
+       list_for_each_entry(gh2, &gl->gl_holders, gh_list) {
+               if (unlikely(gh2->gh_owner_pid == gh->gh_owner_pid &&
+                   (gh->gh_gl->gl_ops->go_type != LM_TYPE_FLOCK)))
+                       goto trap_recursive;
+               if (try_lock &&
+                   !(gh2->gh_flags & (LM_FLAG_TRY | LM_FLAG_TRY_1CB)) &&
+                   !may_grant(gl, gh)) {
+fail:
+                       gh->gh_error = GLR_TRYFAILED;
+                       gfs2_holder_wake(gh);
+                       return;
                }
+               if (test_bit(HIF_HOLDER, &gh2->gh_iflags))
+                       continue;
+               if (unlikely((gh->gh_flags & LM_FLAG_PRIORITY) && !insert_pt))
+                       insert_pt = &gh2->gh_list;
        }
+       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);
+       if (!(gh->gh_flags & LM_FLAG_PRIORITY)) {
+               spin_unlock(&gl->gl_spin);
+               if (sdp->sd_lockstruct.ls_ops->lm_cancel)
+                       sdp->sd_lockstruct.ls_ops->lm_cancel(gl);
+               spin_lock(&gl->gl_spin);
+       }
+       return;
 
-       if (gh->gh_flags & LM_FLAG_PRIORITY)
-               list_add(&gh->gh_list, &gl->gl_waiters3);
-       else
-               list_add_tail(&gh->gh_list, &gl->gl_waiters3);
+trap_recursive:
+       print_symbol(KERN_ERR "original: %s\n", gh2->gh_ip);
+       printk(KERN_ERR "pid: %d\n", pid_nr(gh2->gh_owner_pid));
+       printk(KERN_ERR "lock type: %d req lock state : %d\n",
+              gh2->gh_gl->gl_name.ln_type, gh2->gh_state);
+       print_symbol(KERN_ERR "new: %s\n", gh->gh_ip);
+       printk(KERN_ERR "pid: %d\n", pid_nr(gh->gh_owner_pid));
+       printk(KERN_ERR "lock type: %d req lock state : %d\n",
+              gh->gh_gl->gl_name.ln_type, gh->gh_state);
+       __dump_glock(NULL, gl);
+       BUG();
 }
 
 /**
@@ -1142,24 +1056,16 @@ int gfs2_glock_nq(struct gfs2_holder *gh)
        struct gfs2_sbd *sdp = gl->gl_sbd;
        int error = 0;
 
-restart:
-       if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) {
-               set_bit(HIF_ABORTED, &gh->gh_iflags);
+       if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
                return -EIO;
-       }
 
        spin_lock(&gl->gl_spin);
        add_to_queue(gh);
-       run_queue(gl);
+       run_queue(gl, 1);
        spin_unlock(&gl->gl_spin);
 
-       if (!(gh->gh_flags & GL_ASYNC)) {
-               error = glock_wait_internal(gh);
-               if (error == GLR_CANCELED) {
-                       msleep(100);
-                       goto restart;
-               }
-       }
+       if (!(gh->gh_flags & GL_ASYNC))
+               error = gfs2_glock_wait(gh);
 
        return error;
 }
@@ -1171,50 +1077,9 @@ restart:
  * Returns: 1 if the request is ready to be gfs2_glock_wait()ed on
  */
 
-int gfs2_glock_poll(struct gfs2_holder *gh)
-{
-       struct gfs2_glock *gl = gh->gh_gl;
-       int ready = 0;
-
-       spin_lock(&gl->gl_spin);
-
-       if (test_bit(HIF_HOLDER, &gh->gh_iflags))
-               ready = 1;
-       else if (list_empty(&gh->gh_list)) {
-               if (gh->gh_error == GLR_CANCELED) {
-                       spin_unlock(&gl->gl_spin);
-                       msleep(100);
-                       if (gfs2_glock_nq(gh))
-                               return 1;
-                       return 0;
-               } else
-                       ready = 1;
-       }
-
-       spin_unlock(&gl->gl_spin);
-
-       return ready;
-}
-
-/**
- * gfs2_glock_wait - wait for a lock acquisition that ended in a GLR_ASYNC
- * @gh: the holder structure
- *
- * Returns: 0, GLR_TRYFAILED, or errno on failure
- */
-
-int gfs2_glock_wait(struct gfs2_holder *gh)
+int gfs2_glock_poll(struct gfs2_holder *gh)
 {
-       int error;
-
-       error = glock_wait_internal(gh);
-       if (error == GLR_CANCELED) {
-               msleep(100);
-               gh->gh_flags &= ~GL_ASYNC;
-               error = gfs2_glock_nq(gh);
-       }
-
-       return error;
+       return test_bit(HIF_WAIT, &gh->gh_iflags) ? 0 : 1;
 }
 
 /**
@@ -1228,26 +1093,30 @@ void gfs2_glock_dq(struct gfs2_holder *gh)
        struct gfs2_glock *gl = gh->gh_gl;
        const struct gfs2_glock_operations *glops = gl->gl_ops;
        unsigned delay = 0;
+       int fast_path = 0;
 
+       spin_lock(&gl->gl_spin);
        if (gh->gh_flags & GL_NOCACHE)
-               handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
+               handle_callback(gl, LM_ST_UNLOCKED, 0);
 
-       gfs2_glmutex_lock(gl);
-
-       spin_lock(&gl->gl_spin);
        list_del_init(&gh->gh_list);
-
-       if (list_empty(&gl->gl_holders)) {
+       if (find_first_holder(gl) == NULL) {
                if (glops->go_unlock) {
+                       GLOCK_BUG_ON(gl, test_and_set_bit(GLF_LOCK, &gl->gl_flags));
                        spin_unlock(&gl->gl_spin);
                        glops->go_unlock(gh);
                        spin_lock(&gl->gl_spin);
+                       clear_bit(GLF_LOCK, &gl->gl_flags);
                }
-               gl->gl_stamp = jiffies;
+               if (list_empty(&gl->gl_holders) &&
+                   !test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
+                   !test_bit(GLF_DEMOTE, &gl->gl_flags))
+                       fast_path = 1;
        }
-
-       clear_bit(GLF_LOCK, &gl->gl_flags);
+       trace_gfs2_glock_queue(gh, 0);
        spin_unlock(&gl->gl_spin);
+       if (likely(fast_path))
+               return;
 
        gfs2_glock_hold(gl);
        if (test_bit(GLF_PENDING_DEMOTE, &gl->gl_flags) &&
@@ -1428,215 +1297,108 @@ void gfs2_glock_dq_uninit_m(unsigned int num_gh, struct gfs2_holder *ghs)
                gfs2_glock_dq_uninit(&ghs[x]);
 }
 
-/**
- * gfs2_lvb_hold - attach a LVB from a glock
- * @gl: The glock in question
- *
- */
-
-int gfs2_lvb_hold(struct gfs2_glock *gl)
-{
-       int error;
-
-       gfs2_glmutex_lock(gl);
-
-       if (!atomic_read(&gl->gl_lvb_count)) {
-               error = gfs2_lm_hold_lvb(gl->gl_sbd, gl->gl_lock, &gl->gl_lvb);
-               if (error) {
-                       gfs2_glmutex_unlock(gl);
-                       return error;
-               }
-               gfs2_glock_hold(gl);
-       }
-       atomic_inc(&gl->gl_lvb_count);
-
-       gfs2_glmutex_unlock(gl);
-
-       return 0;
-}
-
-/**
- * gfs2_lvb_unhold - detach a LVB from a glock
- * @gl: The glock in question
- *
- */
-
-void gfs2_lvb_unhold(struct gfs2_glock *gl)
-{
-       gfs2_glock_hold(gl);
-       gfs2_glmutex_lock(gl);
-
-       gfs2_assert(gl->gl_sbd, atomic_read(&gl->gl_lvb_count) > 0);
-       if (atomic_dec_and_test(&gl->gl_lvb_count)) {
-               gfs2_lm_unhold_lvb(gl->gl_sbd, gl->gl_lock, gl->gl_lvb);
-               gl->gl_lvb = NULL;
-               gfs2_glock_put(gl);
-       }
-
-       gfs2_glmutex_unlock(gl);
-       gfs2_glock_put(gl);
-}
-
-static void blocking_cb(struct gfs2_sbd *sdp, struct lm_lockname *name,
-                       unsigned int state)
+void gfs2_glock_cb(struct gfs2_glock *gl, unsigned int state)
 {
-       struct gfs2_glock *gl;
        unsigned long delay = 0;
        unsigned long holdtime;
        unsigned long now = jiffies;
 
-       gl = gfs2_glock_find(sdp, name);
-       if (!gl)
-               return;
-
+       gfs2_glock_hold(gl);
        holdtime = gl->gl_tchange + gl->gl_ops->go_min_hold_time;
        if (time_before(now, holdtime))
                delay = holdtime - now;
+       if (test_bit(GLF_REPLY_PENDING, &gl->gl_flags))
+               delay = gl->gl_ops->go_min_hold_time;
 
-       handle_callback(gl, state, 1, delay);
+       spin_lock(&gl->gl_spin);
+       handle_callback(gl, state, delay);
+       spin_unlock(&gl->gl_spin);
        if (queue_delayed_work(glock_workqueue, &gl->gl_work, delay) == 0)
                gfs2_glock_put(gl);
 }
 
 /**
- * gfs2_glock_cb - Callback used by locking module
- * @sdp: Pointer to the superblock
- * @type: Type of callback
- * @data: Type dependent data pointer
+ * gfs2_glock_complete - Callback used by locking
+ * @gl: Pointer to the glock
+ * @ret: The return value from the dlm
  *
- * Called by the locking module when it wants to tell us something.
- * Either we need to drop a lock, one of our ASYNC requests completed, or
- * a journal from another client needs to be recovered.
  */
 
-void gfs2_glock_cb(void *cb_data, unsigned int type, void *data)
+void gfs2_glock_complete(struct gfs2_glock *gl, int ret)
 {
-       struct gfs2_sbd *sdp = cb_data;
-
-       switch (type) {
-       case LM_CB_NEED_E:
-               blocking_cb(sdp, data, LM_ST_UNLOCKED);
-               return;
-
-       case LM_CB_NEED_D:
-               blocking_cb(sdp, data, LM_ST_DEFERRED);
-               return;
-
-       case LM_CB_NEED_S:
-               blocking_cb(sdp, data, LM_ST_SHARED);
-               return;
-
-       case LM_CB_ASYNC: {
-               struct lm_async_cb *async = data;
-               struct gfs2_glock *gl;
-
-               down_read(&gfs2_umount_flush_sem);
-               gl = gfs2_glock_find(sdp, &async->lc_name);
-               if (gfs2_assert_warn(sdp, gl))
+       struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
+       gl->gl_reply = ret;
+       if (unlikely(test_bit(DFL_BLOCK_LOCKS, &ls->ls_flags))) {
+               struct gfs2_holder *gh;
+               spin_lock(&gl->gl_spin);
+               gh = find_first_waiter(gl);
+               if ((!(gh && (gh->gh_flags & LM_FLAG_NOEXP)) &&
+                    (gl->gl_target != LM_ST_UNLOCKED)) ||
+                   ((ret & ~LM_OUT_ST_MASK) != 0))
+                       set_bit(GLF_FROZEN, &gl->gl_flags);
+               spin_unlock(&gl->gl_spin);
+               if (test_bit(GLF_FROZEN, &gl->gl_flags))
                        return;
-               if (!gfs2_assert_warn(sdp, gl->gl_req_bh))
-                       gl->gl_req_bh(gl, async->lc_ret);
-               if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
-                       gfs2_glock_put(gl);
-               up_read(&gfs2_umount_flush_sem);
-               return;
-       }
-
-       case LM_CB_NEED_RECOVERY:
-               gfs2_jdesc_make_dirty(sdp, *(unsigned int *)data);
-               if (sdp->sd_recoverd_process)
-                       wake_up_process(sdp->sd_recoverd_process);
-               return;
-
-       case LM_CB_DROPLOCKS:
-               gfs2_gl_hash_clear(sdp, NO_WAIT);
-               gfs2_quota_scan(sdp);
-               return;
-
-       default:
-               gfs2_assert_warn(sdp, 0);
-               return;
        }
+       set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
+       gfs2_glock_hold(gl);
+       if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
+               gfs2_glock_put(gl);
 }
 
-/**
- * demote_ok - Check to see if it's ok to unlock a glock
- * @gl: the glock
- *
- * Returns: 1 if it's ok
- */
-
-static int demote_ok(struct gfs2_glock *gl)
-{
-       const struct gfs2_glock_operations *glops = gl->gl_ops;
-       int demote = 1;
-
-       if (test_bit(GLF_STICKY, &gl->gl_flags))
-               demote = 0;
-       else if (glops->go_demote_ok)
-               demote = glops->go_demote_ok(gl);
-
-       return demote;
-}
-
-/**
- * gfs2_glock_schedule_for_reclaim - Add a glock to the reclaim list
- * @gl: the glock
- *
- */
-
-void gfs2_glock_schedule_for_reclaim(struct gfs2_glock *gl)
-{
-       struct gfs2_sbd *sdp = gl->gl_sbd;
-
-       spin_lock(&sdp->sd_reclaim_lock);
-       if (list_empty(&gl->gl_reclaim)) {
-               gfs2_glock_hold(gl);
-               list_add(&gl->gl_reclaim, &sdp->sd_reclaim_list);
-               atomic_inc(&sdp->sd_reclaim_count);
-               spin_unlock(&sdp->sd_reclaim_lock);
-               wake_up(&sdp->sd_reclaim_wq);
-       } else
-               spin_unlock(&sdp->sd_reclaim_lock);
-}
-
-/**
- * gfs2_reclaim_glock - process the next glock on the filesystem's reclaim list
- * @sdp: the filesystem
- *
- * Called from gfs2_glockd() glock reclaim daemon, or when promoting a
- * different glock and we notice that there are a lot of glocks in the
- * reclaim list.
- *
- */
 
-void gfs2_reclaim_glock(struct gfs2_sbd *sdp)
+static int gfs2_shrink_glock_memory(int nr, gfp_t gfp_mask)
 {
        struct gfs2_glock *gl;
+       int may_demote;
+       int nr_skipped = 0;
+       LIST_HEAD(skipped);
 
-       spin_lock(&sdp->sd_reclaim_lock);
-       if (list_empty(&sdp->sd_reclaim_list)) {
-               spin_unlock(&sdp->sd_reclaim_lock);
-               return;
-       }
-       gl = list_entry(sdp->sd_reclaim_list.next,
-                       struct gfs2_glock, gl_reclaim);
-       list_del_init(&gl->gl_reclaim);
-       spin_unlock(&sdp->sd_reclaim_lock);
+       if (nr == 0)
+               goto out;
 
-       atomic_dec(&sdp->sd_reclaim_count);
-       atomic_inc(&sdp->sd_reclaimed);
+       if (!(gfp_mask & __GFP_FS))
+               return -1;
 
-       if (gfs2_glmutex_trylock(gl)) {
-               if (list_empty(&gl->gl_holders) &&
-                   gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
-                       handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
-               gfs2_glmutex_unlock(gl);
-       }
+       spin_lock(&lru_lock);
+       while(nr && !list_empty(&lru_list)) {
+               gl = list_entry(lru_list.next, struct gfs2_glock, gl_lru);
+               list_del_init(&gl->gl_lru);
+               atomic_dec(&lru_count);
 
-       gfs2_glock_put(gl);
+               /* Test for being demotable */
+               if (!test_and_set_bit(GLF_LOCK, &gl->gl_flags)) {
+                       gfs2_glock_hold(gl);
+                       spin_unlock(&lru_lock);
+                       spin_lock(&gl->gl_spin);
+                       may_demote = demote_ok(gl);
+                       if (may_demote) {
+                               handle_callback(gl, LM_ST_UNLOCKED, 0);
+                               nr--;
+                       }
+                       clear_bit(GLF_LOCK, &gl->gl_flags);
+                       smp_mb__after_clear_bit();
+                       if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
+                               gfs2_glock_put_nolock(gl);
+                       spin_unlock(&gl->gl_spin);
+                       spin_lock(&lru_lock);
+                       continue;
+               }
+               nr_skipped++;
+               list_add(&gl->gl_lru, &skipped);
+       }
+       list_splice(&skipped, &lru_list);
+       atomic_add(nr_skipped, &lru_count);
+       spin_unlock(&lru_lock);
+out:
+       return (atomic_read(&lru_count) / 100) * sysctl_vfs_cache_pressure;
 }
 
+static struct shrinker glock_shrinker = {
+       .shrink = gfs2_shrink_glock_memory,
+       .seeks = DEFAULT_SEEKS,
+};
+
 /**
  * examine_bucket - Call a function for glock in a hash bucket
  * @examiner: the function
@@ -1681,28 +1443,23 @@ out:
        return has_entries;
 }
 
+
 /**
- * scan_glock - look at a glock and see if we can reclaim it
- * @gl: the glock to look at
+ * thaw_glock - thaw out a glock which has an unprocessed reply waiting
+ * @gl: The glock to thaw
  *
+ * N.B. When we freeze a glock, we leave a ref to the glock outstanding,
+ * so this has to result in the ref count being dropped by one.
  */
 
-static void scan_glock(struct gfs2_glock *gl)
+static void thaw_glock(struct gfs2_glock *gl)
 {
-       if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object)
+       if (!test_and_clear_bit(GLF_FROZEN, &gl->gl_flags))
                return;
-
-       if (gfs2_glmutex_trylock(gl)) {
-               if (list_empty(&gl->gl_holders) &&
-                   gl->gl_state != LM_ST_UNLOCKED && demote_ok(gl))
-                       goto out_schedule;
-               gfs2_glmutex_unlock(gl);
-       }
-       return;
-
-out_schedule:
-       gfs2_glmutex_unlock(gl);
-       gfs2_glock_schedule_for_reclaim(gl);
+       set_bit(GLF_REPLY_PENDING, &gl->gl_flags);
+       gfs2_glock_hold(gl);
+       if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
+               gfs2_glock_put(gl);
 }
 
 /**
@@ -1713,26 +1470,34 @@ out_schedule:
 
 static void clear_glock(struct gfs2_glock *gl)
 {
-       struct gfs2_sbd *sdp = gl->gl_sbd;
-       int released;
-
-       spin_lock(&sdp->sd_reclaim_lock);
-       if (!list_empty(&gl->gl_reclaim)) {
-               list_del_init(&gl->gl_reclaim);
-               atomic_dec(&sdp->sd_reclaim_count);
-               spin_unlock(&sdp->sd_reclaim_lock);
-               released = gfs2_glock_put(gl);
-               gfs2_assert(sdp, !released);
-       } else {
-               spin_unlock(&sdp->sd_reclaim_lock);
+       spin_lock(&lru_lock);
+       if (!list_empty(&gl->gl_lru)) {
+               list_del_init(&gl->gl_lru);
+               atomic_dec(&lru_count);
        }
+       spin_unlock(&lru_lock);
 
-       if (gfs2_glmutex_trylock(gl)) {
-               if (list_empty(&gl->gl_holders) &&
-                   gl->gl_state != LM_ST_UNLOCKED)
-                       handle_callback(gl, LM_ST_UNLOCKED, 0, 0);
-               gfs2_glmutex_unlock(gl);
-       }
+       spin_lock(&gl->gl_spin);
+       if (find_first_holder(gl) == NULL && gl->gl_state != LM_ST_UNLOCKED)
+               handle_callback(gl, LM_ST_UNLOCKED, 0);
+       spin_unlock(&gl->gl_spin);
+       gfs2_glock_hold(gl);
+       if (queue_delayed_work(glock_workqueue, &gl->gl_work, 0) == 0)
+               gfs2_glock_put(gl);
+}
+
+/**
+ * gfs2_glock_thaw - Thaw any frozen glocks
+ * @sdp: The super block
+ *
+ */
+
+void gfs2_glock_thaw(struct gfs2_sbd *sdp)
+{
+       unsigned x;
+
+       for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
+               examine_bucket(thaw_glock, sdp, x);
 }
 
 /**
@@ -1740,11 +1505,10 @@ static void clear_glock(struct gfs2_glock *gl)
  * @sdp: the filesystem
  * @wait: wait until it's all gone
  *
- * Called when unmounting the filesystem, or when inter-node lock manager
- * requests DROPLOCKS because it is running out of capacity.
+ * Called when unmounting the filesystem.
  */
 
-void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
+void gfs2_gl_hash_clear(struct gfs2_sbd *sdp)
 {
        unsigned long t;
        unsigned int x;
@@ -1759,7 +1523,7 @@ void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
                                cont = 1;
                }
 
-               if (!wait || !cont)
+               if (!cont)
                        break;
 
                if (time_after_eq(jiffies,
@@ -1775,183 +1539,182 @@ void gfs2_gl_hash_clear(struct gfs2_sbd *sdp, int wait)
                up_write(&gfs2_umount_flush_sem);
                msleep(10);
        }
+       flush_workqueue(glock_workqueue);
+       wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
+       gfs2_dump_lockstate(sdp);
 }
 
-/*
- *  Diagnostic routines to help debug distributed deadlock
- */
-
-static void gfs2_print_symbol(struct glock_iter *gi, const char *fmt,
-                              unsigned long address)
+void gfs2_glock_finish_truncate(struct gfs2_inode *ip)
 {
-       char buffer[KSYM_SYMBOL_LEN];
+       struct gfs2_glock *gl = ip->i_gl;
+       int ret;
+
+       ret = gfs2_truncatei_resume(ip);
+       gfs2_assert_withdraw(gl->gl_sbd, ret == 0);
+
+       spin_lock(&gl->gl_spin);
+       clear_bit(GLF_LOCK, &gl->gl_flags);
+       run_queue(gl, 1);
+       spin_unlock(&gl->gl_spin);
+}
 
-       sprint_symbol(buffer, address);
-       print_dbg(gi, fmt, buffer);
+static const char *state2str(unsigned state)
+{
+       switch(state) {
+       case LM_ST_UNLOCKED:
+               return "UN";
+       case LM_ST_SHARED:
+               return "SH";
+       case LM_ST_DEFERRED:
+               return "DF";
+       case LM_ST_EXCLUSIVE:
+               return "EX";
+       }
+       return "??";
+}
+
+static const char *hflags2str(char *buf, unsigned flags, unsigned long iflags)
+{
+       char *p = buf;
+       if (flags & LM_FLAG_TRY)
+               *p++ = 't';
+       if (flags & LM_FLAG_TRY_1CB)
+               *p++ = 'T';
+       if (flags & LM_FLAG_NOEXP)
+               *p++ = 'e';
+       if (flags & LM_FLAG_ANY)
+               *p++ = 'A';
+       if (flags & LM_FLAG_PRIORITY)
+               *p++ = 'p';
+       if (flags & GL_ASYNC)
+               *p++ = 'a';
+       if (flags & GL_EXACT)
+               *p++ = 'E';
+       if (flags & GL_NOCACHE)
+               *p++ = 'c';
+       if (test_bit(HIF_HOLDER, &iflags))
+               *p++ = 'H';
+       if (test_bit(HIF_WAIT, &iflags))
+               *p++ = 'W';
+       if (test_bit(HIF_FIRST, &iflags))
+               *p++ = 'F';
+       *p = 0;
+       return buf;
 }
 
 /**
  * dump_holder - print information about a glock holder
- * @str: a string naming the type of holder
+ * @seq: the seq_file struct
  * @gh: the glock holder
  *
  * Returns: 0 on success, -ENOBUFS when we run out of space
  */
 
-static int dump_holder(struct glock_iter *gi, char *str,
-                      struct gfs2_holder *gh)
+static int dump_holder(struct seq_file *seq, const struct gfs2_holder *gh)
 {
-       unsigned int x;
-       struct task_struct *gh_owner;
+       struct task_struct *gh_owner = NULL;
+       char buffer[KSYM_SYMBOL_LEN];
+       char flags_buf[32];
 
-       print_dbg(gi, "  %s\n", str);
-       if (gh->gh_owner_pid) {
-               print_dbg(gi, "    owner = %ld ",
-                               (long)pid_nr(gh->gh_owner_pid));
+       sprint_symbol(buffer, gh->gh_ip);
+       if (gh->gh_owner_pid)
                gh_owner = pid_task(gh->gh_owner_pid, PIDTYPE_PID);
-               if (gh_owner)
-                       print_dbg(gi, "(%s)\n", gh_owner->comm);
-               else
-                       print_dbg(gi, "(ended)\n");
-       } else
-               print_dbg(gi, "    owner = -1\n");
-       print_dbg(gi, "    gh_state = %u\n", gh->gh_state);
-       print_dbg(gi, "    gh_flags =");
-       for (x = 0; x < 32; x++)
-               if (gh->gh_flags & (1 << x))
-                       print_dbg(gi, " %u", x);
-       print_dbg(gi, " \n");
-       print_dbg(gi, "    error = %d\n", gh->gh_error);
-       print_dbg(gi, "    gh_iflags =");
-       for (x = 0; x < 32; x++)
-               if (test_bit(x, &gh->gh_iflags))
-                       print_dbg(gi, " %u", x);
-       print_dbg(gi, " \n");
-        gfs2_print_symbol(gi, "    initialized at: %s\n", gh->gh_ip);
-
+       gfs2_print_dbg(seq, " H: s:%s f:%s e:%d p:%ld [%s] %s\n",
+                 state2str(gh->gh_state),
+                 hflags2str(flags_buf, gh->gh_flags, gh->gh_iflags),
+                 gh->gh_error, 
+                 gh->gh_owner_pid ? (long)pid_nr(gh->gh_owner_pid) : -1,
+                 gh_owner ? gh_owner->comm : "(ended)", buffer);
        return 0;
 }
 
-/**
- * dump_inode - print information about an inode
- * @ip: the inode
- *
- * Returns: 0 on success, -ENOBUFS when we run out of space
- */
-
-static int dump_inode(struct glock_iter *gi, struct gfs2_inode *ip)
-{
-       unsigned int x;
-
-       print_dbg(gi, "  Inode:\n");
-       print_dbg(gi, "    num = %llu/%llu\n",
-                 (unsigned long long)ip->i_no_formal_ino,
-                 (unsigned long long)ip->i_no_addr);
-       print_dbg(gi, "    type = %u\n", IF2DT(ip->i_inode.i_mode));
-       print_dbg(gi, "    i_flags =");
-       for (x = 0; x < 32; x++)
-               if (test_bit(x, &ip->i_flags))
-                       print_dbg(gi, " %u", x);
-       print_dbg(gi, " \n");
-       return 0;
+static const char *gflags2str(char *buf, const unsigned long *gflags)
+{
+       char *p = buf;
+       if (test_bit(GLF_LOCK, gflags))
+               *p++ = 'l';
+       if (test_bit(GLF_DEMOTE, gflags))
+               *p++ = 'D';
+       if (test_bit(GLF_PENDING_DEMOTE, gflags))
+               *p++ = 'd';
+       if (test_bit(GLF_DEMOTE_IN_PROGRESS, gflags))
+               *p++ = 'p';
+       if (test_bit(GLF_DIRTY, gflags))
+               *p++ = 'y';
+       if (test_bit(GLF_LFLUSH, gflags))
+               *p++ = 'f';
+       if (test_bit(GLF_INVALIDATE_IN_PROGRESS, gflags))
+               *p++ = 'i';
+       if (test_bit(GLF_REPLY_PENDING, gflags))
+               *p++ = 'r';
+       if (test_bit(GLF_INITIAL, gflags))
+               *p++ = 'I';
+       if (test_bit(GLF_FROZEN, gflags))
+               *p++ = 'F';
+       *p = 0;
+       return buf;
 }
 
 /**
- * dump_glock - print information about a glock
+ * __dump_glock - print information about a glock
+ * @seq: The seq_file struct
  * @gl: the glock
- * @count: where we are in the buffer
+ *
+ * The file format is as follows:
+ * One line per object, capital letters are used to indicate objects
+ * G = glock, I = Inode, R = rgrp, H = holder. Glocks are not indented,
+ * other objects are indented by a single space and follow the glock to
+ * which they are related. Fields are indicated by lower case letters
+ * followed by a colon and the field value, except for strings which are in
+ * [] so that its possible to see if they are composed of spaces for
+ * example. The field's are n = number (id of the object), f = flags,
+ * t = type, s = state, r = refcount, e = error, p = pid.
  *
  * Returns: 0 on success, -ENOBUFS when we run out of space
  */
 
-static int dump_glock(struct glock_iter *gi, struct gfs2_glock *gl)
+static int __dump_glock(struct seq_file *seq, const struct gfs2_glock *gl)
 {
-       struct gfs2_holder *gh;
-       unsigned int x;
-       int error = -ENOBUFS;
-       struct task_struct *gl_owner;
+       const struct gfs2_glock_operations *glops = gl->gl_ops;
+       unsigned long long dtime;
+       const struct gfs2_holder *gh;
+       char gflags_buf[32];
+       int error = 0;
 
-       spin_lock(&gl->gl_spin);
+       dtime = jiffies - gl->gl_demote_time;
+       dtime *= 1000000/HZ; /* demote time in uSec */
+       if (!test_bit(GLF_DEMOTE, &gl->gl_flags))
+               dtime = 0;
+       gfs2_print_dbg(seq, "G:  s:%s n:%u/%llu f:%s t:%s d:%s/%llu a:%d r:%d\n",
+                 state2str(gl->gl_state),
+                 gl->gl_name.ln_type,
+                 (unsigned long long)gl->gl_name.ln_number,
+                 gflags2str(gflags_buf, &gl->gl_flags),
+                 state2str(gl->gl_target),
+                 state2str(gl->gl_demote_state), dtime,
+                 atomic_read(&gl->gl_ail_count),
+                 atomic_read(&gl->gl_ref));
 
-       print_dbg(gi, "Glock 0x%p (%u, 0x%llx)\n", gl, gl->gl_name.ln_type,
-                  (unsigned long long)gl->gl_name.ln_number);
-       print_dbg(gi, "  gl_flags =");
-       for (x = 0; x < 32; x++) {
-               if (test_bit(x, &gl->gl_flags))
-                       print_dbg(gi, " %u", x);
-       }
-       if (!test_bit(GLF_LOCK, &gl->gl_flags))
-               print_dbg(gi, " (unlocked)");
-       print_dbg(gi, " \n");
-       print_dbg(gi, "  gl_ref = %d\n", atomic_read(&gl->gl_ref));
-       print_dbg(gi, "  gl_state = %u\n", gl->gl_state);
-       if (gl->gl_owner_pid) {
-               gl_owner = pid_task(gl->gl_owner_pid, PIDTYPE_PID);
-               if (gl_owner)
-                       print_dbg(gi, "  gl_owner = pid %d (%s)\n",
-                                 pid_nr(gl->gl_owner_pid), gl_owner->comm);
-               else
-                       print_dbg(gi, "  gl_owner = %d (ended)\n",
-                                 pid_nr(gl->gl_owner_pid));
-       } else
-               print_dbg(gi, "  gl_owner = -1\n");
-       print_dbg(gi, "  gl_ip = %lu\n", gl->gl_ip);
-       print_dbg(gi, "  req_gh = %s\n", (gl->gl_req_gh) ? "yes" : "no");
-       print_dbg(gi, "  req_bh = %s\n", (gl->gl_req_bh) ? "yes" : "no");
-       print_dbg(gi, "  lvb_count = %d\n", atomic_read(&gl->gl_lvb_count));
-       print_dbg(gi, "  object = %s\n", (gl->gl_object) ? "yes" : "no");
-       print_dbg(gi, "  reclaim = %s\n",
-                  (list_empty(&gl->gl_reclaim)) ? "no" : "yes");
-       if (gl->gl_aspace)
-               print_dbg(gi, "  aspace = 0x%p nrpages = %lu\n", gl->gl_aspace,
-                          gl->gl_aspace->i_mapping->nrpages);
-       else
-               print_dbg(gi, "  aspace = no\n");
-       print_dbg(gi, "  ail = %d\n", atomic_read(&gl->gl_ail_count));
-       if (gl->gl_req_gh) {
-               error = dump_holder(gi, "Request", gl->gl_req_gh);
-               if (error)
-                       goto out;
-       }
        list_for_each_entry(gh, &gl->gl_holders, gh_list) {
-               error = dump_holder(gi, "Holder", gh);
-               if (error)
-                       goto out;
-       }
-       list_for_each_entry(gh, &gl->gl_waiters1, gh_list) {
-               error = dump_holder(gi, "Waiter1", gh);
-               if (error)
-                       goto out;
-       }
-       list_for_each_entry(gh, &gl->gl_waiters3, gh_list) {
-               error = dump_holder(gi, "Waiter3", gh);
+               error = dump_holder(seq, gh);
                if (error)
                        goto out;
        }
-       if (test_bit(GLF_DEMOTE, &gl->gl_flags)) {
-               print_dbg(gi, "  Demotion req to state %u (%llu uS ago)\n",
-                         gl->gl_demote_state, (unsigned long long)
-                         (jiffies - gl->gl_demote_time)*(1000000/HZ));
-       }
-       if (gl->gl_ops == &gfs2_inode_glops && gl->gl_object) {
-               if (!test_bit(GLF_LOCK, &gl->gl_flags) &&
-                       list_empty(&gl->gl_holders)) {
-                       error = dump_inode(gi, gl->gl_object);
-                       if (error)
-                               goto out;
-               } else {
-                       error = -ENOBUFS;
-                       print_dbg(gi, "  Inode: busy\n");
-               }
-       }
-
-       error = 0;
-
+       if (gl->gl_state != LM_ST_UNLOCKED && glops->go_dump)
+               error = glops->go_dump(seq, gl);
 out:
-       spin_unlock(&gl->gl_spin);
        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
@@ -1991,34 +1754,6 @@ static int gfs2_dump_lockstate(struct gfs2_sbd *sdp)
        return error;
 }
 
-/**
- * gfs2_scand - Look for cached glocks and inodes to toss from memory
- * @sdp: Pointer to GFS2 superblock
- *
- * One of these daemons runs, finding candidates to add to sd_reclaim_list.
- * See gfs2_glockd()
- */
-
-static int gfs2_scand(void *data)
-{
-       unsigned x;
-       unsigned delay;
-
-       while (!kthread_should_stop()) {
-               for (x = 0; x < GFS2_GL_HASH_SIZE; x++)
-                       examine_bucket(scan_glock, NULL, x);
-               if (freezing(current))
-                       refrigerator();
-               delay = scand_secs;
-               if (delay < 1)
-                       delay = 1;
-               schedule_timeout_interruptible(delay * HZ);
-       }
-
-       return 0;
-}
-
-
 
 int __init gfs2_glock_init(void)
 {
@@ -2032,29 +1767,28 @@ int __init gfs2_glock_init(void)
        }
 #endif
 
-       scand_process = kthread_run(gfs2_scand, NULL, "gfs2_scand");
-       if (IS_ERR(scand_process))
-               return PTR_ERR(scand_process);
-
        glock_workqueue = create_workqueue("glock_workqueue");
-       if (IS_ERR(glock_workqueue)) {
-               kthread_stop(scand_process);
+       if (IS_ERR(glock_workqueue))
                return PTR_ERR(glock_workqueue);
+       gfs2_delete_workqueue = create_workqueue("delete_workqueue");
+       if (IS_ERR(gfs2_delete_workqueue)) {
+               destroy_workqueue(glock_workqueue);
+               return PTR_ERR(gfs2_delete_workqueue);
        }
 
+       register_shrinker(&glock_shrinker);
+
        return 0;
 }
 
 void gfs2_glock_exit(void)
 {
+       unregister_shrinker(&glock_shrinker);
        destroy_workqueue(glock_workqueue);
-       kthread_stop(scand_process);
+       destroy_workqueue(gfs2_delete_workqueue);
 }
 
-module_param(scand_secs, uint, S_IRUGO|S_IWUSR);
-MODULE_PARM_DESC(scand_secs, "The number of seconds between scand runs");
-
-static int gfs2_glock_iter_next(struct glock_iter *gi)
+static int gfs2_glock_iter_next(struct gfs2_glock_iter *gi)
 {
        struct gfs2_glock *gl;
 
@@ -2064,15 +1798,17 @@ restart:
        if (gl) {
                gi->gl = hlist_entry(gl->gl_list.next,
                                     struct gfs2_glock, gl_list);
-               if (gi->gl)
-                       gfs2_glock_hold(gi->gl);
+       } 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);
-       if (gl && gi->gl == NULL)
+       while (gi->gl == NULL) {
                gi->hash++;
-       while(gi->gl == NULL) {
                if (gi->hash >= GFS2_GL_HASH_SIZE)
                        return 1;
                read_lock(gl_lock_addr(gi->hash));
@@ -2081,7 +1817,6 @@ restart:
                if (gi->gl)
                        gfs2_glock_hold(gi->gl);
                read_unlock(gl_lock_addr(gi->hash));
-               gi->hash++;
        }
 
        if (gi->sdp != gi->gl->gl_sbd)
@@ -2090,58 +1825,34 @@ restart:
        return 0;
 }
 
-static void gfs2_glock_iter_free(struct glock_iter *gi)
+static void gfs2_glock_iter_free(struct gfs2_glock_iter *gi)
 {
        if (gi->gl)
                gfs2_glock_put(gi->gl);
-       kfree(gi);
-}
-
-static struct glock_iter *gfs2_glock_iter_init(struct gfs2_sbd *sdp)
-{
-       struct glock_iter *gi;
-
-       gi = kmalloc(sizeof (*gi), GFP_KERNEL);
-       if (!gi)
-               return NULL;
-
-       gi->sdp = sdp;
-       gi->hash = 0;
-       gi->seq = NULL;
        gi->gl = NULL;
-       memset(gi->string, 0, sizeof(gi->string));
-
-       if (gfs2_glock_iter_next(gi)) {
-               gfs2_glock_iter_free(gi);
-               return NULL;
-       }
-
-       return gi;
 }
 
-static void *gfs2_glock_seq_start(struct seq_file *file, loff_t *pos)
+static void *gfs2_glock_seq_start(struct seq_file *seq, loff_t *pos)
 {
-       struct glock_iter *gi;
+       struct gfs2_glock_iter *gi = seq->private;
        loff_t n = *pos;
 
-       gi = gfs2_glock_iter_init(file->private);
-       if (!gi)
-               return NULL;
+       gi->hash = 0;
 
-       while(n--) {
+       do {
                if (gfs2_glock_iter_next(gi)) {
                        gfs2_glock_iter_free(gi);
                        return NULL;
                }
-       }
+       } while (n--);
 
-       return gi;
+       return gi->gl;
 }
 
-static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
+static void *gfs2_glock_seq_next(struct seq_file *seq, void *iter_ptr,
                                 loff_t *pos)
 {
-       struct glock_iter *gi = iter_ptr;
+       struct gfs2_glock_iter *gi = seq->private;
 
        (*pos)++;
 
@@ -2150,24 +1861,18 @@ static void *gfs2_glock_seq_next(struct seq_file *file, void *iter_ptr,
                return NULL;
        }
 
-       return gi;
+       return gi->gl;
 }
 
-static void gfs2_glock_seq_stop(struct seq_file *file, void *iter_ptr)
+static void gfs2_glock_seq_stop(struct seq_file *seq, void *iter_ptr)
 {
-       struct glock_iter *gi = iter_ptr;
-       if (gi)
-               gfs2_glock_iter_free(gi);
+       struct gfs2_glock_iter *gi = seq->private;
+       gfs2_glock_iter_free(gi);
 }
 
-static int gfs2_glock_seq_show(struct seq_file *file, void *iter_ptr)
+static int gfs2_glock_seq_show(struct seq_file *seq, void *iter_ptr)
 {
-       struct glock_iter *gi = iter_ptr;
-
-       gi->seq = file;
-       dump_glock(gi, gi->gl);
-
-       return 0;
+       return dump_glock(seq, iter_ptr);
 }
 
 static const struct seq_operations gfs2_glock_seq_ops = {
@@ -2179,17 +1884,14 @@ static const struct seq_operations gfs2_glock_seq_ops = {
 
 static int gfs2_debugfs_open(struct inode *inode, struct file *file)
 {
-       struct seq_file *seq;
-       int ret;
-
-       ret = seq_open(file, &gfs2_glock_seq_ops);
-       if (ret)
-               return ret;
-
-       seq = file->private_data;
-       seq->private = inode->i_private;
-
-       return 0;
+       int ret = seq_open_private(file, &gfs2_glock_seq_ops,
+                                  sizeof(struct gfs2_glock_iter));
+       if (ret == 0) {
+               struct seq_file *seq = file->private_data;
+               struct gfs2_glock_iter *gi = seq->private;
+               gi->sdp = inode->i_private;
+       }
+       return ret;
 }
 
 static const struct file_operations gfs2_debug_fops = {
@@ -2197,7 +1899,7 @@ static const struct file_operations gfs2_debug_fops = {
        .open    = gfs2_debugfs_open,
        .read    = seq_read,
        .llseek  = seq_lseek,
-       .release = seq_release
+       .release = seq_release_private,
 };
 
 int gfs2_create_debugfs_file(struct gfs2_sbd *sdp)