Merge tag 'dlm-6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/linux-dlm

Pull dlm updates from David Teigland:
 "This set includes some small fixes, and some big internal changes:

   - Fix a long standing race between the unlock callback for the last
     lkb struct, and removing the rsb that became unused after the final
     unlock. This could lead different nodes to inconsistent info about
     the rsb master node.

   - Remove unnecessary refcounting on callback structs, returning to
     the way things were done in the past.

   - Do message processing in softirq context. This allows dlm messages
     to be cleared more quickly and efficiently, reducing long lists of
     incomplete requests. A future change to run callbacks directly from
     this context will make this more effective.

   - The softirq message processing involved a number of patches
     changing mutexes to spinlocks and rwlocks, and a fair amount of
     code re-org in preparation.

   - Use an rhashtable for rsb structs, rather than our old internal
     hash table implementation. This also required some re-org of lists
     and locks preparation for the change.

   - Drop the dlm_scand kthread, and use timers to clear unused rsb
     structs. Scanning all rsb's periodically was a lot of wasted work.

   - Fix recent regression in logic for copying LVB data in user space
     lock requests"

* tag 'dlm-6.10' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/linux-dlm: (34 commits)
  dlm: return -ENOMEM if ls_recover_buf fails
  dlm: fix sleep in atomic context
  dlm: use rwlock for lkbidr
  dlm: use rwlock for rsb hash table
  dlm: drop dlm_scand kthread and use timers
  dlm: do not use ref counts for rsb in the toss state
  dlm: switch to use rhashtable for rsbs
  dlm: add rsb lists for iteration
  dlm: merge toss and keep hash table lists into one list
  dlm: change to single hashtable lock
  dlm: increment ls_count for dlm_scand
  dlm: do message processing in softirq context
  dlm: use spin_lock_bh for message processing
  dlm: remove schedule in receive path
  dlm: convert ls_recv_active from rw_semaphore to rwlock
  dlm: avoid blocking receive at the end of recovery
  dlm: convert res_lock to spinlock
  dlm: convert ls_waiters_mutex to spinlock
  dlm: drop mutex use in waiters recovery
  dlm: add new struct to save position in dlm_copy_master_names
  ...

fs/dlm/ast.c

@@ -12,47 +12,50 @@
 #include <trace/events/dlm.h>
 
 #include "dlm_internal.h"
+#include "lvb_table.h"
 #include "memory.h"
 #include "lock.h"
 #include "user.h"
 #include "ast.h"
 
-void dlm_release_callback(struct kref *ref)
+static void dlm_callback_work(struct work_struct *work)
 {
-	struct dlm_callback *cb = container_of(ref, struct dlm_callback, ref);
+	struct dlm_callback *cb = container_of(work, struct dlm_callback, work);
 
-	dlm_free_cb(cb);
-}
-
-void dlm_callback_set_last_ptr(struct dlm_callback **from,
-			       struct dlm_callback *to)
-{
-	if (*from)
-		kref_put(&(*from)->ref, dlm_release_callback);
-
-	if (to)
-		kref_get(&to->ref);
+	if (cb->flags & DLM_CB_BAST) {
+		trace_dlm_bast(cb->ls_id, cb->lkb_id, cb->mode, cb->res_name,
+			       cb->res_length);
+		cb->bastfn(cb->astparam, cb->mode);
+	} else if (cb->flags & DLM_CB_CAST) {
+		trace_dlm_ast(cb->ls_id, cb->lkb_id, cb->sb_status,
+			      cb->sb_flags, cb->res_name, cb->res_length);
+		cb->lkb_lksb->sb_status = cb->sb_status;
+		cb->lkb_lksb->sb_flags = cb->sb_flags;
+		cb->astfn(cb->astparam);
+	}
 
-	*from = to;
+	dlm_free_cb(cb);
 }
 
-int dlm_enqueue_lkb_callback(struct dlm_lkb *lkb, uint32_t flags, int mode,
-			     int status, uint32_t sbflags)
+int dlm_queue_lkb_callback(struct dlm_lkb *lkb, uint32_t flags, int mode,
+			   int status, uint32_t sbflags,
+			   struct dlm_callback **cb)
 {
-	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
+	struct dlm_rsb *rsb = lkb->lkb_resource;
 	int rv = DLM_ENQUEUE_CALLBACK_SUCCESS;
-	struct dlm_callback *cb;
+	struct dlm_ls *ls = rsb->res_ls;
+	int copy_lvb = 0;
 	int prev_mode;
 
 	if (flags & DLM_CB_BAST) {
 		/* if cb is a bast, it should be skipped if the blocking mode is
 		 * compatible with the last granted mode
 		 */
-		if (lkb->lkb_last_cast) {
-			if (dlm_modes_compat(mode, lkb->lkb_last_cast->mode)) {
+		if (lkb->lkb_last_cast_cb_mode != -1) {
+			if (dlm_modes_compat(mode, lkb->lkb_last_cast_cb_mode)) {
 				log_debug(ls, "skip %x bast mode %d for cast mode %d",
 					  lkb->lkb_id, mode,
-					  lkb->lkb_last_cast->mode);
+					  lkb->lkb_last_cast_cb_mode);
 				goto out;
 			}
 		}
@@ -63,8 +66,9 @@ int dlm_enqueue_lkb_callback(struct dlm_lkb *lkb, uint32_t flags, int mode,
 		 * is a bast for the same mode or a more restrictive mode.
 		 * (the addional > PR check is needed for PR/CW inversion)
 		 */
-		if (lkb->lkb_last_cb && lkb->lkb_last_cb->flags & DLM_CB_BAST) {
-			prev_mode = lkb->lkb_last_cb->mode;
+		if (lkb->lkb_last_cb_mode != -1 &&
+		    lkb->lkb_last_cb_flags & DLM_CB_BAST) {
+			prev_mode = lkb->lkb_last_cb_mode;
 
 			if ((prev_mode == mode) ||
 			    (prev_mode > mode && prev_mode > DLM_LOCK_PR)) {
@@ -73,53 +77,55 @@ int dlm_enqueue_lkb_callback(struct dlm_lkb *lkb, uint32_t flags, int mode,
 				goto out;
 			}
 		}
+
+		lkb->lkb_last_bast_time = ktime_get();
+		lkb->lkb_last_bast_cb_mode = mode;
+	} else if (flags & DLM_CB_CAST) {
+		if (test_bit(DLM_DFL_USER_BIT, &lkb->lkb_dflags)) {
+			prev_mode = lkb->lkb_last_cast_cb_mode;
+
+			if (!status && lkb->lkb_lksb->sb_lvbptr &&
+			    dlm_lvb_operations[prev_mode + 1][mode + 1])
+				copy_lvb = 1;
+		}
+
+		lkb->lkb_last_cast_cb_mode = mode;
+		lkb->lkb_last_cast_time = ktime_get();
 	}
 
-	cb = dlm_allocate_cb();
-	if (!cb) {
+	lkb->lkb_last_cb_mode = mode;
+	lkb->lkb_last_cb_flags = flags;
+
+	*cb = dlm_allocate_cb();
+	if (!*cb) {
 		rv = DLM_ENQUEUE_CALLBACK_FAILURE;
 		goto out;
 	}
 
-	cb->flags = flags;
-	cb->mode = mode;
-	cb->sb_status = status;
-	cb->sb_flags = (sbflags & 0x000000FF);
-	kref_init(&cb->ref);
-	if (!test_and_set_bit(DLM_IFL_CB_PENDING_BIT, &lkb->lkb_iflags))
-		rv = DLM_ENQUEUE_CALLBACK_NEED_SCHED;
-
-	list_add_tail(&cb->list, &lkb->lkb_callbacks);
+	/* for tracing */
+	(*cb)->lkb_id = lkb->lkb_id;
+	(*cb)->ls_id = ls->ls_global_id;
+	memcpy((*cb)->res_name, rsb->res_name, rsb->res_length);
+	(*cb)->res_length = rsb->res_length;
 
-	if (flags & DLM_CB_CAST)
-		dlm_callback_set_last_ptr(&lkb->lkb_last_cast, cb);
+	(*cb)->flags = flags;
+	(*cb)->mode = mode;
+	(*cb)->sb_status = status;
+	(*cb)->sb_flags = (sbflags & 0x000000FF);
+	(*cb)->copy_lvb = copy_lvb;
+	(*cb)->lkb_lksb = lkb->lkb_lksb;
 
-	dlm_callback_set_last_ptr(&lkb->lkb_last_cb, cb);
+	rv = DLM_ENQUEUE_CALLBACK_NEED_SCHED;
 
- out:
+out:
 	return rv;
 }
 
-int dlm_dequeue_lkb_callback(struct dlm_lkb *lkb, struct dlm_callback **cb)
-{
-	/* oldest undelivered cb is callbacks first entry */
-	*cb = list_first_entry_or_null(&lkb->lkb_callbacks,
-				       struct dlm_callback, list);
-	if (!*cb)
-		return DLM_DEQUEUE_CALLBACK_EMPTY;
-
-	/* remove it from callbacks so shift others down */
-	list_del(&(*cb)->list);
-	if (list_empty(&lkb->lkb_callbacks))
-		return DLM_DEQUEUE_CALLBACK_LAST;
-
-	return DLM_DEQUEUE_CALLBACK_SUCCESS;
-}
-
 void dlm_add_cb(struct dlm_lkb *lkb, uint32_t flags, int mode, int status,
 		  uint32_t sbflags)
 {
 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
+	struct dlm_callback *cb;
 	int rv;
 
 	if (test_bit(DLM_DFL_USER_BIT, &lkb->lkb_dflags)) {
@@ -127,88 +133,36 @@ void dlm_add_cb(struct dlm_lkb *lkb, uint32_t flags, int mode, int status,
 		return;
 	}
 
-	spin_lock(&lkb->lkb_cb_lock);
-	rv = dlm_enqueue_lkb_callback(lkb, flags, mode, status, sbflags);
+	rv = dlm_queue_lkb_callback(lkb, flags, mode, status, sbflags,
+				    &cb);
 	switch (rv) {
 	case DLM_ENQUEUE_CALLBACK_NEED_SCHED:
-		kref_get(&lkb->lkb_ref);
-
-		spin_lock(&ls->ls_cb_lock);
-		if (test_bit(LSFL_CB_DELAY, &ls->ls_flags)) {
-			list_add(&lkb->lkb_cb_list, &ls->ls_cb_delay);
-		} else {
-			queue_work(ls->ls_callback_wq, &lkb->lkb_cb_work);
-		}
-		spin_unlock(&ls->ls_cb_lock);
-		break;
-	case DLM_ENQUEUE_CALLBACK_FAILURE:
-		WARN_ON_ONCE(1);
+		cb->astfn = lkb->lkb_astfn;
+		cb->bastfn = lkb->lkb_bastfn;
+		cb->astparam = lkb->lkb_astparam;
+		INIT_WORK(&cb->work, dlm_callback_work);
+
+		spin_lock_bh(&ls->ls_cb_lock);
+		if (test_bit(LSFL_CB_DELAY, &ls->ls_flags))
+			list_add(&cb->list, &ls->ls_cb_delay);
+		else
+			queue_work(ls->ls_callback_wq, &cb->work);
+		spin_unlock_bh(&ls->ls_cb_lock);
 		break;
 	case DLM_ENQUEUE_CALLBACK_SUCCESS:
 		break;
+	case DLM_ENQUEUE_CALLBACK_FAILURE:
+		fallthrough;
 	default:
 		WARN_ON_ONCE(1);
 		break;
 	}
-	spin_unlock(&lkb->lkb_cb_lock);
-}
-
-void dlm_callback_work(struct work_struct *work)
-{
-	struct dlm_lkb *lkb = container_of(work, struct dlm_lkb, lkb_cb_work);
-	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
-	void (*castfn) (void *astparam);
-	void (*bastfn) (void *astparam, int mode);
-	struct dlm_callback *cb;
-	int rv;
-
-	spin_lock(&lkb->lkb_cb_lock);
-	rv = dlm_dequeue_lkb_callback(lkb, &cb);
-	if (WARN_ON_ONCE(rv == DLM_DEQUEUE_CALLBACK_EMPTY)) {
-		clear_bit(DLM_IFL_CB_PENDING_BIT, &lkb->lkb_iflags);
-		spin_unlock(&lkb->lkb_cb_lock);
-		goto out;
-	}
-	spin_unlock(&lkb->lkb_cb_lock);
-
-	for (;;) {
-		castfn = lkb->lkb_astfn;
-		bastfn = lkb->lkb_bastfn;
-
-		if (cb->flags & DLM_CB_BAST) {
-			trace_dlm_bast(ls, lkb, cb->mode);
-			lkb->lkb_last_bast_time = ktime_get();
-			lkb->lkb_last_bast_mode = cb->mode;
-			bastfn(lkb->lkb_astparam, cb->mode);
-		} else if (cb->flags & DLM_CB_CAST) {
-			lkb->lkb_lksb->sb_status = cb->sb_status;
-			lkb->lkb_lksb->sb_flags = cb->sb_flags;
-			trace_dlm_ast(ls, lkb);
-			lkb->lkb_last_cast_time = ktime_get();
-			castfn(lkb->lkb_astparam);
-		}
-
-		kref_put(&cb->ref, dlm_release_callback);
-
-		spin_lock(&lkb->lkb_cb_lock);
-		rv = dlm_dequeue_lkb_callback(lkb, &cb);
-		if (rv == DLM_DEQUEUE_CALLBACK_EMPTY) {
-			clear_bit(DLM_IFL_CB_PENDING_BIT, &lkb->lkb_iflags);
-			spin_unlock(&lkb->lkb_cb_lock);
-			break;
-		}
-		spin_unlock(&lkb->lkb_cb_lock);
-	}
-
-out:
-	/* undo kref_get from dlm_add_callback, may cause lkb to be freed */
-	dlm_put_lkb(lkb);
 }
 
 int dlm_callback_start(struct dlm_ls *ls)
 {
-	ls->ls_callback_wq = alloc_workqueue("dlm_callback",
-					     WQ_HIGHPRI | WQ_MEM_RECLAIM, 0);
+	ls->ls_callback_wq = alloc_ordered_workqueue("dlm_callback",
+						     WQ_HIGHPRI | WQ_MEM_RECLAIM);
 	if (!ls->ls_callback_wq) {
 		log_print("can't start dlm_callback workqueue");
 		return -ENOMEM;
@@ -225,9 +179,9 @@ void dlm_callback_stop(struct dlm_ls *ls)
 void dlm_callback_suspend(struct dlm_ls *ls)
 {
 	if (ls->ls_callback_wq) {
-		spin_lock(&ls->ls_cb_lock);
+		spin_lock_bh(&ls->ls_cb_lock);
 		set_bit(LSFL_CB_DELAY, &ls->ls_flags);
-		spin_unlock(&ls->ls_cb_lock);
+		spin_unlock_bh(&ls->ls_cb_lock);
 
 		flush_workqueue(ls->ls_callback_wq);
 	}
@@ -237,7 +191,7 @@ void dlm_callback_suspend(struct dlm_ls *ls)
 
 void dlm_callback_resume(struct dlm_ls *ls)
 {
-	struct dlm_lkb *lkb, *safe;
+	struct dlm_callback *cb, *safe;
 	int count = 0, sum = 0;
 	bool empty;
 
@@ -245,10 +199,10 @@ void dlm_callback_resume(struct dlm_ls *ls)
 		return;
 
 more:
-	spin_lock(&ls->ls_cb_lock);
-	list_for_each_entry_safe(lkb, safe, &ls->ls_cb_delay, lkb_cb_list) {
-		list_del_init(&lkb->lkb_cb_list);
-		queue_work(ls->ls_callback_wq, &lkb->lkb_cb_work);
+	spin_lock_bh(&ls->ls_cb_lock);
+	list_for_each_entry_safe(cb, safe, &ls->ls_cb_delay, list) {
+		list_del(&cb->list);
+		queue_work(ls->ls_callback_wq, &cb->work);
 		count++;
 		if (count == MAX_CB_QUEUE)
 			break;
@@ -256,7 +210,7 @@ void dlm_callback_resume(struct dlm_ls *ls)
 	empty = list_empty(&ls->ls_cb_delay);
 	if (empty)
 		clear_bit(LSFL_CB_DELAY, &ls->ls_flags);
-	spin_unlock(&ls->ls_cb_lock);
+	spin_unlock_bh(&ls->ls_cb_lock);
 
 	sum += count;
 	if (!empty) {

fs/dlm/ast.h

@@ -14,19 +14,12 @@
 #define DLM_ENQUEUE_CALLBACK_NEED_SCHED	1
 #define DLM_ENQUEUE_CALLBACK_SUCCESS	0
 #define DLM_ENQUEUE_CALLBACK_FAILURE	-1
-int dlm_enqueue_lkb_callback(struct dlm_lkb *lkb, uint32_t flags, int mode,
-			     int status, uint32_t sbflags);
-#define DLM_DEQUEUE_CALLBACK_EMPTY	2
-#define DLM_DEQUEUE_CALLBACK_LAST	1
-#define DLM_DEQUEUE_CALLBACK_SUCCESS	0
-int dlm_dequeue_lkb_callback(struct dlm_lkb *lkb, struct dlm_callback **cb);
+int dlm_queue_lkb_callback(struct dlm_lkb *lkb, uint32_t flags, int mode,
+			   int status, uint32_t sbflags,
+			   struct dlm_callback **cb);
 void dlm_add_cb(struct dlm_lkb *lkb, uint32_t flags, int mode, int status,
                 uint32_t sbflags);
-void dlm_callback_set_last_ptr(struct dlm_callback **from,
-			       struct dlm_callback *to);
 
-void dlm_release_callback(struct kref *ref);
-void dlm_callback_work(struct work_struct *work);
 int dlm_callback_start(struct dlm_ls *ls);
 void dlm_callback_stop(struct dlm_ls *ls);
 void dlm_callback_suspend(struct dlm_ls *ls);

fs/dlm/config.c

@@ -63,6 +63,14 @@ static void release_node(struct config_item *);
 static struct configfs_attribute *comm_attrs[];
 static struct configfs_attribute *node_attrs[];
 
+const struct rhashtable_params dlm_rhash_rsb_params = {
+	.nelem_hint = 3, /* start small */
+	.key_len = DLM_RESNAME_MAXLEN,
+	.key_offset = offsetof(struct dlm_rsb, res_name),
+	.head_offset = offsetof(struct dlm_rsb, res_node),
+	.automatic_shrinking = true,
+};
+
 struct dlm_cluster {
 	struct config_group group;
 	unsigned int cl_tcp_port;

fs/dlm/config.h

@@ -21,6 +21,8 @@ struct dlm_config_node {
 	uint32_t comm_seq;
 };
 
+extern const struct rhashtable_params dlm_rhash_rsb_params;
+
 #define DLM_MAX_ADDR_COUNT 3
 
 #define DLM_PROTO_TCP	0

fs/dlm/debug_fs.c

@@ -247,7 +247,7 @@ static void print_format3_lock(struct seq_file *s, struct dlm_lkb *lkb,
 		   lkb->lkb_status,
 		   lkb->lkb_grmode,
 		   lkb->lkb_rqmode,
-		   lkb->lkb_last_bast_mode,
+		   lkb->lkb_last_bast_cb_mode,
 		   rsb_lookup,
 		   lkb->lkb_wait_type,
 		   lkb->lkb_lvbseq,
@@ -366,58 +366,10 @@ static void print_format4(struct dlm_rsb *r, struct seq_file *s)
 	unlock_rsb(r);
 }
 
-static void print_format5_lock(struct seq_file *s, struct dlm_lkb *lkb)
-{
-	struct dlm_callback *cb;
-
-	/* lkb_id lkb_flags mode flags sb_status sb_flags */
-
-	spin_lock(&lkb->lkb_cb_lock);
-	list_for_each_entry(cb, &lkb->lkb_callbacks, list) {
-		seq_printf(s, "%x %x %d %x %d %x\n",
-			   lkb->lkb_id,
-			   dlm_iflags_val(lkb),
-			   cb->mode,
-			   cb->flags,
-			   cb->sb_status,
-			   cb->sb_flags);
-	}
-	spin_unlock(&lkb->lkb_cb_lock);
-}
-
-static void print_format5(struct dlm_rsb *r, struct seq_file *s)
-{
-	struct dlm_lkb *lkb;
-
-	lock_rsb(r);
-
-	list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
-		print_format5_lock(s, lkb);
-		if (seq_has_overflowed(s))
-			goto out;
-	}
-
-	list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
-		print_format5_lock(s, lkb);
-		if (seq_has_overflowed(s))
-			goto out;
-	}
-
-	list_for_each_entry(lkb, &r->res_waitqueue, lkb_statequeue) {
-		print_format5_lock(s, lkb);
-		if (seq_has_overflowed(s))
-			goto out;
-	}
- out:
-	unlock_rsb(r);
-}
-
-struct rsbtbl_iter {
-	struct dlm_rsb *rsb;
-	unsigned bucket;
-	int format;
-	int header;
-};
+static const struct seq_operations format1_seq_ops;
+static const struct seq_operations format2_seq_ops;
+static const struct seq_operations format3_seq_ops;
+static const struct seq_operations format4_seq_ops;
 
 /*
  * If the buffer is full, seq_printf can be called again, but it
@@ -428,207 +380,61 @@ struct rsbtbl_iter {
 
 static int table_seq_show(struct seq_file *seq, void *iter_ptr)
 {
-	struct rsbtbl_iter *ri = iter_ptr;
+	struct dlm_rsb *rsb = list_entry(iter_ptr, struct dlm_rsb, res_rsbs_list);
 
-	switch (ri->format) {
-	case 1:
-		print_format1(ri->rsb, seq);
-		break;
-	case 2:
-		if (ri->header) {
-			seq_puts(seq, "id nodeid remid pid xid exflags flags sts grmode rqmode time_ms r_nodeid r_len r_name\n");
-			ri->header = 0;
-		}
-		print_format2(ri->rsb, seq);
-		break;
-	case 3:
-		if (ri->header) {
-			seq_puts(seq, "rsb ptr nodeid first_lkid flags !root_list_empty !recover_list_empty recover_locks_count len\n");
-			ri->header = 0;
-		}
-		print_format3(ri->rsb, seq);
-		break;
-	case 4:
-		if (ri->header) {
-			seq_puts(seq, "rsb ptr nodeid master_nodeid dir_nodeid our_nodeid toss_time flags len str|hex name\n");
-			ri->header = 0;
-		}
-		print_format4(ri->rsb, seq);
-		break;
-	case 5:
-		if (ri->header) {
-			seq_puts(seq, "lkb_id lkb_flags mode flags sb_status sb_flags\n");
-			ri->header = 0;
-		}
-		print_format5(ri->rsb, seq);
-		break;
-	}
+	if (seq->op == &format1_seq_ops)
+		print_format1(rsb, seq);
+	else if (seq->op == &format2_seq_ops)
+		print_format2(rsb, seq);
+	else if (seq->op == &format3_seq_ops)
+		print_format3(rsb, seq);
+	else if (seq->op == &format4_seq_ops)
+		print_format4(rsb, seq);
 
 	return 0;
 }
 
-static const struct seq_operations format1_seq_ops;
-static const struct seq_operations format2_seq_ops;
-static const struct seq_operations format3_seq_ops;
-static const struct seq_operations format4_seq_ops;
-static const struct seq_operations format5_seq_ops;
-
 static void *table_seq_start(struct seq_file *seq, loff_t *pos)
 {
-	struct rb_root *tree;
-	struct rb_node *node;
 	struct dlm_ls *ls = seq->private;
-	struct rsbtbl_iter *ri;
-	struct dlm_rsb *r;
-	loff_t n = *pos;
-	unsigned bucket, entry;
-	int toss = (seq->op == &format4_seq_ops);
-
-	bucket = n >> 32;
-	entry = n & ((1LL << 32) - 1);
-
-	if (bucket >= ls->ls_rsbtbl_size)
-		return NULL;
-
-	ri = kzalloc(sizeof(*ri), GFP_NOFS);
-	if (!ri)
-		return NULL;
-	if (n == 0)
-		ri->header = 1;
-	if (seq->op == &format1_seq_ops)
-		ri->format = 1;
+	struct list_head *list;
+
+	if (!*pos) {
 		if (seq->op == &format2_seq_ops)
-		ri->format = 2;
-	if (seq->op == &format3_seq_ops)
-		ri->format = 3;
-	if (seq->op == &format4_seq_ops)
-		ri->format = 4;
-	if (seq->op == &format5_seq_ops)
-		ri->format = 5;
-
-	tree = toss ? &ls->ls_rsbtbl[bucket].toss : &ls->ls_rsbtbl[bucket].keep;
-
-	spin_lock(&ls->ls_rsbtbl[bucket].lock);
-	if (!RB_EMPTY_ROOT(tree)) {
-		for (node = rb_first(tree); node; node = rb_next(node)) {
-			r = rb_entry(node, struct dlm_rsb, res_hashnode);
-			if (!entry--) {
-				dlm_hold_rsb(r);
-				ri->rsb = r;
-				ri->bucket = bucket;
-				spin_unlock(&ls->ls_rsbtbl[bucket].lock);
-				return ri;
-			}
-		}
+			seq_puts(seq, "id nodeid remid pid xid exflags flags sts grmode rqmode time_ms r_nodeid r_len r_name\n");
+		else if (seq->op == &format3_seq_ops)
+			seq_puts(seq, "rsb ptr nodeid first_lkid flags !root_list_empty !recover_list_empty recover_locks_count len\n");
+		else if (seq->op == &format4_seq_ops)
+			seq_puts(seq, "rsb ptr nodeid master_nodeid dir_nodeid our_nodeid toss_time flags len str|hex name\n");
 	}
-	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
-
-	/*
-	 * move to the first rsb in the next non-empty bucket
-	 */
-
-	/* zero the entry */
-	n &= ~((1LL << 32) - 1);
 
-	while (1) {
-		bucket++;
-		n += 1LL << 32;
+	if (seq->op == &format4_seq_ops)
+		list = &ls->ls_toss;
+	else
+		list = &ls->ls_keep;
 
-		if (bucket >= ls->ls_rsbtbl_size) {
-			kfree(ri);
-			return NULL;
-		}
-		tree = toss ? &ls->ls_rsbtbl[bucket].toss : &ls->ls_rsbtbl[bucket].keep;
-
-		spin_lock(&ls->ls_rsbtbl[bucket].lock);
-		if (!RB_EMPTY_ROOT(tree)) {
-			node = rb_first(tree);
-			r = rb_entry(node, struct dlm_rsb, res_hashnode);
-			dlm_hold_rsb(r);
-			ri->rsb = r;
-			ri->bucket = bucket;
-			spin_unlock(&ls->ls_rsbtbl[bucket].lock);
-			*pos = n;
-			return ri;
-		}
-		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
-	}
+	read_lock_bh(&ls->ls_rsbtbl_lock);
+	return seq_list_start(list, *pos);
 }
 
 static void *table_seq_next(struct seq_file *seq, void *iter_ptr, loff_t *pos)
 {
 	struct dlm_ls *ls = seq->private;
-	struct rsbtbl_iter *ri = iter_ptr;
-	struct rb_root *tree;
-	struct rb_node *next;
-	struct dlm_rsb *r, *rp;
-	loff_t n = *pos;
-	unsigned bucket;
-	int toss = (seq->op == &format4_seq_ops);
-
-	bucket = n >> 32;
-
-	/*
-	 * move to the next rsb in the same bucket
-	 */
-
-	spin_lock(&ls->ls_rsbtbl[bucket].lock);
-	rp = ri->rsb;
-	next = rb_next(&rp->res_hashnode);
-
-	if (next) {
-		r = rb_entry(next, struct dlm_rsb, res_hashnode);
-		dlm_hold_rsb(r);
-		ri->rsb = r;
-		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
-		dlm_put_rsb(rp);
-		++*pos;
-		return ri;
-	}
-	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
-	dlm_put_rsb(rp);
-
-	/*
-	 * move to the first rsb in the next non-empty bucket
-	 */
-
-	/* zero the entry */
-	n &= ~((1LL << 32) - 1);
+	struct list_head *list;
 
-	while (1) {
-		bucket++;
-		n += 1LL << 32;
+	if (seq->op == &format4_seq_ops)
+		list = &ls->ls_toss;
+	else
+		list = &ls->ls_keep;
 
-		if (bucket >= ls->ls_rsbtbl_size) {
-			kfree(ri);
-			++*pos;
-			return NULL;
-		}
-		tree = toss ? &ls->ls_rsbtbl[bucket].toss : &ls->ls_rsbtbl[bucket].keep;
-
-		spin_lock(&ls->ls_rsbtbl[bucket].lock);
-		if (!RB_EMPTY_ROOT(tree)) {
-			next = rb_first(tree);
-			r = rb_entry(next, struct dlm_rsb, res_hashnode);
-			dlm_hold_rsb(r);
-			ri->rsb = r;
-			ri->bucket = bucket;
-			spin_unlock(&ls->ls_rsbtbl[bucket].lock);
-			*pos = n;
-			return ri;
-		}
-		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
-	}
+	return seq_list_next(iter_ptr, list, pos);
 }
 
 static void table_seq_stop(struct seq_file *seq, void *iter_ptr)
 {
-	struct rsbtbl_iter *ri = iter_ptr;
+	struct dlm_ls *ls = seq->private;
 
-	if (ri) {
-		dlm_put_rsb(ri->rsb);
-		kfree(ri);
-	}
+	read_unlock_bh(&ls->ls_rsbtbl_lock);
 }
 
 static const struct seq_operations format1_seq_ops = {
@@ -659,18 +465,10 @@ static const struct seq_operations format4_seq_ops = {
 	.show  = table_seq_show,
 };
 
-static const struct seq_operations format5_seq_ops = {
-	.start = table_seq_start,
-	.next  = table_seq_next,
-	.stop  = table_seq_stop,
-	.show  = table_seq_show,
-};
-
 static const struct file_operations format1_fops;
 static const struct file_operations format2_fops;
 static const struct file_operations format3_fops;
 static const struct file_operations format4_fops;
-static const struct file_operations format5_fops;
 
 static int table_open1(struct inode *inode, struct file *file)
 {
@@ -757,20 +555,6 @@ static int table_open4(struct inode *inode, struct file *file)
 	return 0;
 }
 
-static int table_open5(struct inode *inode, struct file *file)
-{
-	struct seq_file *seq;
-	int ret;
-
-	ret = seq_open(file, &format5_seq_ops);
-	if (ret)
-		return ret;
-
-	seq = file->private_data;
-	seq->private = inode->i_private; /* the dlm_ls */
-	return 0;
-}
-
 static const struct file_operations format1_fops = {
 	.owner   = THIS_MODULE,
 	.open    = table_open1,
@@ -804,14 +588,6 @@ static const struct file_operations format4_fops = {
 	.release = seq_release
 };
 
-static const struct file_operations format5_fops = {
-	.owner   = THIS_MODULE,
-	.open    = table_open5,
-	.read    = seq_read,
-	.llseek  = seq_lseek,
-	.release = seq_release
-};
-
 /*
  * dump lkb's on the ls_waiters list
  */
@@ -823,7 +599,13 @@ static ssize_t waiters_read(struct file *file, char __user *userbuf,
 	size_t len = DLM_DEBUG_BUF_LEN, pos = 0, ret, rv;
 
 	mutex_lock(&debug_buf_lock);
-	mutex_lock(&ls->ls_waiters_mutex);
+	ret = dlm_lock_recovery_try(ls);
+	if (!ret) {
+		rv = -EAGAIN;
+		goto out;
+	}
+
+	spin_lock_bh(&ls->ls_waiters_lock);
 	memset(debug_buf, 0, sizeof(debug_buf));
 
 	list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
@@ -834,9 +616,11 @@ static ssize_t waiters_read(struct file *file, char __user *userbuf,
 			break;
 		pos += ret;
 	}
-	mutex_unlock(&ls->ls_waiters_mutex);
+	spin_unlock_bh(&ls->ls_waiters_lock);
+	dlm_unlock_recovery(ls);
 
 	rv = simple_read_from_buffer(userbuf, count, ppos, debug_buf, pos);
+out:
 	mutex_unlock(&debug_buf_lock);
 	return rv;
 }
@@ -858,7 +642,12 @@ static ssize_t waiters_write(struct file *file, const char __user *user_buf,
 	if (n != 3)
 		return -EINVAL;
 
+	error = dlm_lock_recovery_try(ls);
+	if (!error)
+		return -EAGAIN;
+
 	error = dlm_debug_add_lkb_to_waiters(ls, lkb_id, mstype, to_nodeid);
+	dlm_unlock_recovery(ls);
 	if (error)
 		return error;
 
@@ -1021,16 +810,6 @@ void dlm_create_debug_file(struct dlm_ls *ls)
 							  dlm_root,
 							  ls,
 							  &waiters_fops);
-
-	/* format 5 */
-
-	snprintf(name, sizeof(name), "%s_queued_asts", ls->ls_name);
-
-	ls->ls_debug_queued_asts_dentry = debugfs_create_file(name,
-							      0644,
-							      dlm_root,
-							      ls,
-							      &format5_fops);
 }
 
 void __init dlm_register_debugfs(void)

fs/dlm/dir.c

@@ -47,15 +47,13 @@ int dlm_dir_nodeid(struct dlm_rsb *r)
 	return r->res_dir_nodeid;
 }
 
-void dlm_recover_dir_nodeid(struct dlm_ls *ls)
+void dlm_recover_dir_nodeid(struct dlm_ls *ls, const struct list_head *root_list)
 {
 	struct dlm_rsb *r;
 
-	down_read(&ls->ls_root_sem);
-	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+	list_for_each_entry(r, root_list, res_root_list) {
 		r->res_dir_nodeid = dlm_hash2nodeid(ls, r->res_hash);
 	}
-	up_read(&ls->ls_root_sem);
 }
 
 int dlm_recover_directory(struct dlm_ls *ls, uint64_t seq)
@@ -200,35 +198,98 @@ static struct dlm_rsb *find_rsb_root(struct dlm_ls *ls, const char *name,
 				     int len)
 {
 	struct dlm_rsb *r;
-	uint32_t hash, bucket;
 	int rv;
 
-	hash = jhash(name, len, 0);
-	bucket = hash & (ls->ls_rsbtbl_size - 1);
-
-	spin_lock(&ls->ls_rsbtbl[bucket].lock);
-	rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[bucket].keep, name, len, &r);
-	if (rv)
-		rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[bucket].toss,
-					 name, len, &r);
-	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
-
+	read_lock_bh(&ls->ls_rsbtbl_lock);
+	rv = dlm_search_rsb_tree(&ls->ls_rsbtbl, name, len, &r);
+	read_unlock_bh(&ls->ls_rsbtbl_lock);
 	if (!rv)
 		return r;
 
-	down_read(&ls->ls_root_sem);
-	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+	list_for_each_entry(r, &ls->ls_masters_list, res_masters_list) {
 		if (len == r->res_length && !memcmp(name, r->res_name, len)) {
-			up_read(&ls->ls_root_sem);
 			log_debug(ls, "find_rsb_root revert to root_list %s",
 				  r->res_name);
 			return r;
 		}
 	}
-	up_read(&ls->ls_root_sem);
 	return NULL;
 }
 
+struct dlm_dir_dump {
+	/* init values to match if whole
+	 * dump fits to one seq. Sanity check only.
+	 */
+	uint64_t seq_init;
+	uint64_t nodeid_init;
+	/* compare local pointer with last lookup,
+	 * just a sanity check.
+	 */
+	struct list_head *last;
+
+	unsigned int sent_res; /* for log info */
+	unsigned int sent_msg; /* for log info */
+
+	struct list_head list;
+};
+
+static void drop_dir_ctx(struct dlm_ls *ls, int nodeid)
+{
+	struct dlm_dir_dump *dd, *safe;
+
+	write_lock_bh(&ls->ls_dir_dump_lock);
+	list_for_each_entry_safe(dd, safe, &ls->ls_dir_dump_list, list) {
+		if (dd->nodeid_init == nodeid) {
+			log_error(ls, "drop dump seq %llu",
+				 (unsigned long long)dd->seq_init);
+			list_del(&dd->list);
+			kfree(dd);
+		}
+	}
+	write_unlock_bh(&ls->ls_dir_dump_lock);
+}
+
+static struct dlm_dir_dump *lookup_dir_dump(struct dlm_ls *ls, int nodeid)
+{
+	struct dlm_dir_dump *iter, *dd = NULL;
+
+	read_lock_bh(&ls->ls_dir_dump_lock);
+	list_for_each_entry(iter, &ls->ls_dir_dump_list, list) {
+		if (iter->nodeid_init == nodeid) {
+			dd = iter;
+			break;
+		}
+	}
+	read_unlock_bh(&ls->ls_dir_dump_lock);
+
+	return dd;
+}
+
+static struct dlm_dir_dump *init_dir_dump(struct dlm_ls *ls, int nodeid)
+{
+	struct dlm_dir_dump *dd;
+
+	dd = lookup_dir_dump(ls, nodeid);
+	if (dd) {
+		log_error(ls, "found ongoing dir dump for node %d, will drop it",
+			  nodeid);
+		drop_dir_ctx(ls, nodeid);
+	}
+
+	dd = kzalloc(sizeof(*dd), GFP_ATOMIC);
+	if (!dd)
+		return NULL;
+
+	dd->seq_init = ls->ls_recover_seq;
+	dd->nodeid_init = nodeid;
+
+	write_lock_bh(&ls->ls_dir_dump_lock);
+	list_add(&dd->list, &ls->ls_dir_dump_list);
+	write_unlock_bh(&ls->ls_dir_dump_lock);
+
+	return dd;
+}
+
 /* Find the rsb where we left off (or start again), then send rsb names
    for rsb's we're master of and whose directory node matches the requesting
    node.  inbuf is the rsb name last sent, inlen is the name's length */
@@ -239,27 +300,50 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
 	struct list_head *list;
 	struct dlm_rsb *r;
 	int offset = 0, dir_nodeid;
+	struct dlm_dir_dump *dd;
 	__be16 be_namelen;
 
-	down_read(&ls->ls_root_sem);
+	read_lock_bh(&ls->ls_masters_lock);
 
 	if (inlen > 1) {
+		dd = lookup_dir_dump(ls, nodeid);
+		if (!dd) {
+			log_error(ls, "failed to lookup dir dump context nodeid: %d",
+				  nodeid);
+			goto out;
+		}
+
+		/* next chunk in dump */
 		r = find_rsb_root(ls, inbuf, inlen);
 		if (!r) {
 			log_error(ls, "copy_master_names from %d start %d %.*s",
 				  nodeid, inlen, inlen, inbuf);
 			goto out;
 		}
-		list = r->res_root_list.next;
+		list = r->res_masters_list.next;
+
+		/* sanity checks */
+		if (dd->last != &r->res_masters_list ||
+		    dd->seq_init != ls->ls_recover_seq) {
+			log_error(ls, "failed dir dump sanity check seq_init: %llu seq: %llu",
+				  (unsigned long long)dd->seq_init,
+				  (unsigned long long)ls->ls_recover_seq);
+			goto out;
+		}
 	} else {
-		list = ls->ls_root_list.next;
+		dd = init_dir_dump(ls, nodeid);
+		if (!dd) {
+			log_error(ls, "failed to allocate dir dump context");
+			goto out;
 		}
 
-	for (offset = 0; list != &ls->ls_root_list; list = list->next) {
-		r = list_entry(list, struct dlm_rsb, res_root_list);
-		if (r->res_nodeid)
-			continue;
+		/* start dump */
+		list = ls->ls_masters_list.next;
+		dd->last = list;
+	}
 
+	for (offset = 0; list != &ls->ls_masters_list; list = list->next) {
+		r = list_entry(list, struct dlm_rsb, res_masters_list);
 		dir_nodeid = dlm_dir_nodeid(r);
 		if (dir_nodeid != nodeid)
 			continue;
@@ -277,7 +361,7 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
 			be_namelen = cpu_to_be16(0);
 			memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
 			offset += sizeof(__be16);
-			ls->ls_recover_dir_sent_msg++;
+			dd->sent_msg++;
 			goto out;
 		}
 
@@ -286,7 +370,8 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
 		offset += sizeof(__be16);
 		memcpy(outbuf + offset, r->res_name, r->res_length);
 		offset += r->res_length;
-		ls->ls_recover_dir_sent_res++;
+		dd->sent_res++;
+		dd->last = list;
 	}
 
 	/*
@@ -294,14 +379,22 @@ void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
 	 * terminating record.
 	 */
 
-	if ((list == &ls->ls_root_list) &&
+	if ((list == &ls->ls_masters_list) &&
 	    (offset + sizeof(uint16_t) <= outlen)) {
+		/* end dump */
 		be_namelen = cpu_to_be16(0xFFFF);
 		memcpy(outbuf + offset, &be_namelen, sizeof(__be16));
 		offset += sizeof(__be16);
-		ls->ls_recover_dir_sent_msg++;
+		dd->sent_msg++;
+		log_rinfo(ls, "dlm_recover_directory nodeid %d sent %u res out %u messages",
+			  nodeid, dd->sent_res, dd->sent_msg);
+
+		write_lock_bh(&ls->ls_dir_dump_lock);
+		list_del_init(&dd->list);
+		write_unlock_bh(&ls->ls_dir_dump_lock);
+		kfree(dd);
 	}
  out:
-	up_read(&ls->ls_root_sem);
+	read_unlock_bh(&ls->ls_masters_lock);
 }
 

fs/dlm/dir.h

@@ -14,7 +14,8 @@
 
 int dlm_dir_nodeid(struct dlm_rsb *rsb);
 int dlm_hash2nodeid(struct dlm_ls *ls, uint32_t hash);
-void dlm_recover_dir_nodeid(struct dlm_ls *ls);
+void dlm_recover_dir_nodeid(struct dlm_ls *ls,
+			    const struct list_head *root_list);
 int dlm_recover_directory(struct dlm_ls *ls, uint64_t seq);
 void dlm_copy_master_names(struct dlm_ls *ls, const char *inbuf, int inlen,
 			   char *outbuf, int outlen, int nodeid);

fs/dlm/dlm_internal.h

@@ -16,6 +16,7 @@
  * This is the main header file to be included in each DLM source file.
  */
 
+#include <uapi/linux/dlm_device.h>
 #include <linux/slab.h>
 #include <linux/sched.h>
 #include <linux/types.h>
@@ -33,6 +34,7 @@
 #include <linux/kernel.h>
 #include <linux/jhash.h>
 #include <linux/miscdevice.h>
+#include <linux/rhashtable.h>
 #include <linux/mutex.h>
 #include <linux/idr.h>
 #include <linux/ratelimit.h>
@@ -98,17 +100,6 @@ do { \
   } \
 }
 
-
-#define DLM_RTF_SHRINK_BIT	0
-
-struct dlm_rsbtable {
-	struct rb_root		keep;
-	struct rb_root		toss;
-	spinlock_t		lock;
-	unsigned long		flags;
-};
-
-
 /*
  * Lockspace member (per node in a ls)
  */
@@ -204,8 +195,7 @@ struct dlm_args {
 #define DLM_IFL_OVERLAP_CANCEL_BIT 20
 #define DLM_IFL_ENDOFLIFE_BIT	21
 #define DLM_IFL_DEADLOCK_CANCEL_BIT 24
-#define DLM_IFL_CB_PENDING_BIT	25
-#define __DLM_IFL_MAX_BIT	DLM_IFL_CB_PENDING_BIT
+#define __DLM_IFL_MAX_BIT	DLM_IFL_DEADLOCK_CANCEL_BIT
 
 /* lkb_dflags */
 
@@ -217,14 +207,47 @@ struct dlm_args {
 #define DLM_CB_CAST		0x00000001
 #define DLM_CB_BAST		0x00000002
 
+/* much of this is just saving user space pointers associated with the
+ * lock that we pass back to the user lib with an ast
+ */
+
+struct dlm_user_args {
+	struct dlm_user_proc	*proc; /* each process that opens the lockspace
+					* device has private data
+					* (dlm_user_proc) on the struct file,
+					* the process's locks point back to it
+					*/
+	struct dlm_lksb		lksb;
+	struct dlm_lksb __user	*user_lksb;
+	void __user		*castparam;
+	void __user		*castaddr;
+	void __user		*bastparam;
+	void __user		*bastaddr;
+	uint64_t		xid;
+};
+
 struct dlm_callback {
 	uint32_t		flags;		/* DLM_CBF_ */
 	int			sb_status;	/* copy to lksb status */
 	uint8_t			sb_flags;	/* copy to lksb flags */
 	int8_t			mode; /* rq mode of bast, gr mode of cast */
+	bool			copy_lvb;
+	struct dlm_lksb		*lkb_lksb;
+	unsigned char		lvbptr[DLM_USER_LVB_LEN];
+
+	union {
+		void			*astparam;	/* caller's ast arg */
+		struct dlm_user_args	ua;
+	};
+	struct work_struct	work;
+	void			(*bastfn)(void *astparam, int mode);
+	void			(*astfn)(void *astparam);
+	char			res_name[DLM_RESNAME_MAXLEN];
+	size_t			res_length;
+	uint32_t		ls_id;
+	uint32_t		lkb_id;
 
 	struct list_head	list;
-	struct kref		ref;
 };
 
 struct dlm_lkb {
@@ -255,13 +278,10 @@ struct dlm_lkb {
 	struct list_head	lkb_ownqueue;	/* list of locks for a process */
 	ktime_t			lkb_timestamp;
 
-	spinlock_t		lkb_cb_lock;
-	struct work_struct	lkb_cb_work;
-	struct list_head	lkb_cb_list; /* for ls_cb_delay or proc->asts */
-	struct list_head	lkb_callbacks;
-	struct dlm_callback	*lkb_last_cast;
-	struct dlm_callback	*lkb_last_cb;
-	int			lkb_last_bast_mode;
+	int8_t			lkb_last_cast_cb_mode;
+	int8_t			lkb_last_bast_cb_mode;
+	int8_t			lkb_last_cb_mode;
+	uint8_t			lkb_last_cb_flags;
 	ktime_t			lkb_last_cast_time;	/* for debugging */
 	ktime_t			lkb_last_bast_time;	/* for debugging */
 
@@ -290,7 +310,7 @@ struct dlm_lkb {
 struct dlm_rsb {
 	struct dlm_ls		*res_ls;	/* the lockspace */
 	struct kref		res_ref;
-	struct mutex		res_mutex;
+	spinlock_t		res_lock;
 	unsigned long		res_flags;
 	int			res_length;	/* length of rsb name */
 	int			res_nodeid;
@@ -299,20 +319,22 @@ struct dlm_rsb {
 	int			res_id;		/* for ls_recover_idr */
 	uint32_t                res_lvbseq;
 	uint32_t		res_hash;
-	uint32_t		res_bucket;	/* rsbtbl */
 	unsigned long		res_toss_time;
 	uint32_t		res_first_lkid;
 	struct list_head	res_lookup;	/* lkbs waiting on first */
 	union {
 		struct list_head	res_hashchain;
-		struct rb_node		res_hashnode;	/* rsbtbl */
+		struct rhash_head	res_node; /* rsbtbl */
 	};
 	struct list_head	res_grantqueue;
 	struct list_head	res_convertqueue;
 	struct list_head	res_waitqueue;
 
+	struct list_head	res_rsbs_list;
 	struct list_head	res_root_list;	    /* used for recovery */
+	struct list_head	res_masters_list;   /* used for recovery */
 	struct list_head	res_recover_list;   /* used for recovery */
+	struct list_head	res_toss_q_list;
 	int			res_recover_locks_count;
 
 	char			*res_lvbptr;
@@ -346,6 +368,7 @@ enum rsb_flags {
 	RSB_RECOVER_CONVERT,
 	RSB_RECOVER_GRANT,
 	RSB_RECOVER_LVB_INVAL,
+	RSB_TOSS,
 };
 
 static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag)
@@ -559,24 +582,33 @@ struct dlm_ls {
 	struct kobject		ls_kobj;
 
 	struct idr		ls_lkbidr;
-	spinlock_t		ls_lkbidr_spin;
+	rwlock_t		ls_lkbidr_lock;
+
+	struct rhashtable	ls_rsbtbl;
+	rwlock_t		ls_rsbtbl_lock;
 
-	struct dlm_rsbtable	*ls_rsbtbl;
-	uint32_t		ls_rsbtbl_size;
+	struct list_head	ls_toss;
+	struct list_head	ls_keep;
+
+	struct timer_list	ls_timer;
+	/* this queue is ordered according the
+	 * absolute res_toss_time jiffies time
+	 * to mod_timer() with the first element
+	 * if necessary.
+	 */
+	struct list_head	ls_toss_q;
+	spinlock_t		ls_toss_q_lock;
 
-	struct mutex		ls_waiters_mutex;
+	spinlock_t		ls_waiters_lock;
 	struct list_head	ls_waiters;	/* lkbs needing a reply */
 
-	struct mutex		ls_orphans_mutex;
+	spinlock_t		ls_orphans_lock;
 	struct list_head	ls_orphans;
 
 	spinlock_t		ls_new_rsb_spin;
 	int			ls_new_rsb_count;
 	struct list_head	ls_new_rsb;	/* new rsb structs */
 
-	char			*ls_remove_names[DLM_REMOVE_NAMES_MAX];
-	int			ls_remove_lens[DLM_REMOVE_NAMES_MAX];
-
 	struct list_head	ls_nodes;	/* current nodes in ls */
 	struct list_head	ls_nodes_gone;	/* dead node list, recovery */
 	int			ls_num_nodes;	/* number of nodes in ls */
@@ -613,7 +645,6 @@ struct dlm_ls {
 
 	spinlock_t		ls_cb_lock;
 	struct list_head	ls_cb_delay; /* save for queue_work later */
-	struct timer_list	ls_timer;
 	struct task_struct	*ls_recoverd_task;
 	struct mutex		ls_recoverd_active;
 	spinlock_t		ls_recover_lock;
@@ -622,15 +653,11 @@ struct dlm_ls {
 	uint64_t		ls_recover_seq;
 	struct dlm_recover	*ls_recover_args;
 	struct rw_semaphore	ls_in_recovery;	/* block local requests */
-	struct rw_semaphore	ls_recv_active;	/* block dlm_recv */
+	rwlock_t		ls_recv_active;	/* block dlm_recv */
 	struct list_head	ls_requestqueue;/* queue remote requests */
-	atomic_t		ls_requestqueue_cnt;
-	wait_queue_head_t	ls_requestqueue_wait;
-	struct mutex		ls_requestqueue_mutex;
+	rwlock_t		ls_requestqueue_lock;
 	struct dlm_rcom		*ls_recover_buf;
 	int			ls_recover_nodeid; /* for debugging */
-	unsigned int		ls_recover_dir_sent_res; /* for log info */
-	unsigned int		ls_recover_dir_sent_msg; /* for log info */
 	unsigned int		ls_recover_locks_in; /* for log info */
 	uint64_t		ls_rcom_seq;
 	spinlock_t		ls_rcom_spin;
@@ -643,8 +670,10 @@ struct dlm_ls {
 	wait_queue_head_t	ls_recover_lock_wait;
 	spinlock_t		ls_clear_proc_locks;
 
-	struct list_head	ls_root_list;	/* root resources */
-	struct rw_semaphore	ls_root_sem;	/* protect root_list */
+	struct list_head	ls_masters_list; /* root resources */
+	rwlock_t		ls_masters_lock; /* protect root_list */
+	struct list_head	ls_dir_dump_list; /* root resources */
+	rwlock_t		ls_dir_dump_lock; /* protect root_list */
 
 	const struct dlm_lockspace_ops *ls_ops;
 	void			*ls_ops_arg;
@@ -686,23 +715,7 @@ struct dlm_ls {
 #define LSFL_UEVENT_WAIT	7
 #define LSFL_CB_DELAY		9
 #define LSFL_NODIR		10
-
-/* much of this is just saving user space pointers associated with the
-   lock that we pass back to the user lib with an ast */
-
-struct dlm_user_args {
-	struct dlm_user_proc	*proc; /* each process that opens the lockspace
-					  device has private data
-					  (dlm_user_proc) on the struct file,
-					  the process's locks point back to it*/
-	struct dlm_lksb		lksb;
-	struct dlm_lksb __user	*user_lksb;
-	void __user		*castparam;
-	void __user		*castaddr;
-	void __user		*bastparam;
-	void __user		*bastaddr;
-	uint64_t		xid;
-};
+#define LSFL_RECV_MSG_BLOCKED	11
 
 #define DLM_PROC_FLAGS_CLOSING 1
 #define DLM_PROC_FLAGS_COMPAT  2

fs/dlm/lock.c

@@ -201,7 +201,7 @@ void dlm_dump_rsb(struct dlm_rsb *r)
 
 /* Threads cannot use the lockspace while it's being recovered */
 
-static inline void dlm_lock_recovery(struct dlm_ls *ls)
+void dlm_lock_recovery(struct dlm_ls *ls)
 {
 	down_read(&ls->ls_in_recovery);
 }
@@ -320,11 +320,18 @@ static void queue_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int rqmode)
  * Basic operations on rsb's and lkb's
  */
 
+static inline unsigned long rsb_toss_jiffies(void)
+{
+	return jiffies + (READ_ONCE(dlm_config.ci_toss_secs) * HZ);
+}
+
 /* This is only called to add a reference when the code already holds
    a valid reference to the rsb, so there's no need for locking. */
 
 static inline void hold_rsb(struct dlm_rsb *r)
 {
+	/* rsbs in toss state never get referenced */
+	WARN_ON(rsb_flag(r, RSB_TOSS));
 	kref_get(&r->res_ref);
 }
 
@@ -333,19 +340,48 @@ void dlm_hold_rsb(struct dlm_rsb *r)
 	hold_rsb(r);
 }
 
+/* TODO move this to lib/refcount.c */
+static __must_check bool
+dlm_refcount_dec_and_write_lock_bh(refcount_t *r, rwlock_t *lock)
+__cond_acquires(lock)
+{
+	if (refcount_dec_not_one(r))
+		return false;
+
+	write_lock_bh(lock);
+	if (!refcount_dec_and_test(r)) {
+		write_unlock_bh(lock);
+		return false;
+	}
+
+	return true;
+}
+
+/* TODO move this to include/linux/kref.h */
+static inline int dlm_kref_put_write_lock_bh(struct kref *kref,
+					     void (*release)(struct kref *kref),
+					     rwlock_t *lock)
+{
+	if (dlm_refcount_dec_and_write_lock_bh(&kref->refcount, lock)) {
+		release(kref);
+		return 1;
+	}
+
+	return 0;
+}
+
 /* When all references to the rsb are gone it's transferred to
    the tossed list for later disposal. */
 
 static void put_rsb(struct dlm_rsb *r)
 {
 	struct dlm_ls *ls = r->res_ls;
-	uint32_t bucket = r->res_bucket;
 	int rv;
 
-	rv = kref_put_lock(&r->res_ref, toss_rsb,
-			   &ls->ls_rsbtbl[bucket].lock);
+	rv = dlm_kref_put_write_lock_bh(&r->res_ref, toss_rsb,
+					&ls->ls_rsbtbl_lock);
 	if (rv)
-		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
 }
 
 void dlm_put_rsb(struct dlm_rsb *r)
@@ -358,17 +394,17 @@ static int pre_rsb_struct(struct dlm_ls *ls)
 	struct dlm_rsb *r1, *r2;
 	int count = 0;
 
-	spin_lock(&ls->ls_new_rsb_spin);
+	spin_lock_bh(&ls->ls_new_rsb_spin);
 	if (ls->ls_new_rsb_count > dlm_config.ci_new_rsb_count / 2) {
-		spin_unlock(&ls->ls_new_rsb_spin);
+		spin_unlock_bh(&ls->ls_new_rsb_spin);
 		return 0;
 	}
-	spin_unlock(&ls->ls_new_rsb_spin);
+	spin_unlock_bh(&ls->ls_new_rsb_spin);
 
 	r1 = dlm_allocate_rsb(ls);
 	r2 = dlm_allocate_rsb(ls);
 
-	spin_lock(&ls->ls_new_rsb_spin);
+	spin_lock_bh(&ls->ls_new_rsb_spin);
 	if (r1) {
 		list_add(&r1->res_hashchain, &ls->ls_new_rsb);
 		ls->ls_new_rsb_count++;
@@ -378,13 +414,236 @@ static int pre_rsb_struct(struct dlm_ls *ls)
 		ls->ls_new_rsb_count++;
 	}
 	count = ls->ls_new_rsb_count;
-	spin_unlock(&ls->ls_new_rsb_spin);
+	spin_unlock_bh(&ls->ls_new_rsb_spin);
 
 	if (!count)
 		return -ENOMEM;
 	return 0;
 }
 
+/* connected with timer_delete_sync() in dlm_ls_stop() to stop
+ * new timers when recovery is triggered and don't run them
+ * again until a dlm_timer_resume() tries it again.
+ */
+static void __rsb_mod_timer(struct dlm_ls *ls, unsigned long jiffies)
+{
+	if (!dlm_locking_stopped(ls))
+		mod_timer(&ls->ls_timer, jiffies);
+}
+
+/* This function tries to resume the timer callback if a rsb
+ * is on the toss list and no timer is pending. It might that
+ * the first entry is on currently executed as timer callback
+ * but we don't care if a timer queued up again and does
+ * nothing. Should be a rare case.
+ */
+void dlm_timer_resume(struct dlm_ls *ls)
+{
+	struct dlm_rsb *r;
+
+	spin_lock_bh(&ls->ls_toss_q_lock);
+	r = list_first_entry_or_null(&ls->ls_toss_q, struct dlm_rsb,
+				     res_toss_q_list);
+	if (r && !timer_pending(&ls->ls_timer))
+		__rsb_mod_timer(ls, r->res_toss_time);
+	spin_unlock_bh(&ls->ls_toss_q_lock);
+}
+
+/* ls_rsbtbl_lock must be held and being sure the rsb is in toss state */
+static void rsb_delete_toss_timer(struct dlm_ls *ls, struct dlm_rsb *r)
+{
+	struct dlm_rsb *first;
+
+	spin_lock_bh(&ls->ls_toss_q_lock);
+	r->res_toss_time = 0;
+
+	/* if the rsb is not queued do nothing */
+	if (list_empty(&r->res_toss_q_list))
+		goto out;
+
+	/* get the first element before delete */
+	first = list_first_entry(&ls->ls_toss_q, struct dlm_rsb,
+				 res_toss_q_list);
+	list_del_init(&r->res_toss_q_list);
+	/* check if the first element was the rsb we deleted */
+	if (first == r) {
+		/* try to get the new first element, if the list
+		 * is empty now try to delete the timer, if we are
+		 * too late we don't care.
+		 *
+		 * if the list isn't empty and a new first element got
+		 * in place, set the new timer expire time.
+		 */
+		first = list_first_entry_or_null(&ls->ls_toss_q, struct dlm_rsb,
+						 res_toss_q_list);
+		if (!first)
+			timer_delete(&ls->ls_timer);
+		else
+			__rsb_mod_timer(ls, first->res_toss_time);
+	}
+
+out:
+	spin_unlock_bh(&ls->ls_toss_q_lock);
+}
+
+/* Caller must held ls_rsbtbl_lock and need to be called every time
+ * when either the rsb enters toss state or the toss state changes
+ * the dir/master nodeid.
+ */
+static void rsb_mod_timer(struct dlm_ls *ls, struct dlm_rsb *r)
+{
+	int our_nodeid = dlm_our_nodeid();
+	struct dlm_rsb *first;
+
+	/* If we're the directory record for this rsb, and
+	 * we're not the master of it, then we need to wait
+	 * for the master node to send us a dir remove for
+	 * before removing the dir record.
+	 */
+	if (!dlm_no_directory(ls) &&
+	    (r->res_master_nodeid != our_nodeid) &&
+	    (dlm_dir_nodeid(r) == our_nodeid)) {
+		rsb_delete_toss_timer(ls, r);
+		return;
+	}
+
+	spin_lock_bh(&ls->ls_toss_q_lock);
+	/* set the new rsb absolute expire time in the rsb */
+	r->res_toss_time = rsb_toss_jiffies();
+	if (list_empty(&ls->ls_toss_q)) {
+		/* if the queue is empty add the element and it's
+		 * our new expire time
+		 */
+		list_add_tail(&r->res_toss_q_list, &ls->ls_toss_q);
+		__rsb_mod_timer(ls, r->res_toss_time);
+	} else {
+		/* check if the rsb was already queued, if so delete
+		 * it from the toss queue
+		 */
+		if (!list_empty(&r->res_toss_q_list))
+			list_del(&r->res_toss_q_list);
+
+		/* try to get the maybe new first element and then add
+		 * to this rsb with the oldest expire time to the end
+		 * of the queue. If the list was empty before this
+		 * rsb expire time is our next expiration if it wasn't
+		 * the now new first elemet is our new expiration time
+		 */
+		first = list_first_entry_or_null(&ls->ls_toss_q, struct dlm_rsb,
+						 res_toss_q_list);
+		list_add_tail(&r->res_toss_q_list, &ls->ls_toss_q);
+		if (!first)
+			__rsb_mod_timer(ls, r->res_toss_time);
+		else
+			__rsb_mod_timer(ls, first->res_toss_time);
+	}
+	spin_unlock_bh(&ls->ls_toss_q_lock);
+}
+
+/* if we hit contention we do in 250 ms a retry to trylock.
+ * if there is any other mod_timer in between we don't care
+ * about that it expires earlier again this is only for the
+ * unlikely case nothing happened in this time.
+ */
+#define DLM_TOSS_TIMER_RETRY	(jiffies + msecs_to_jiffies(250))
+
+void dlm_rsb_toss_timer(struct timer_list *timer)
+{
+	struct dlm_ls *ls = from_timer(ls, timer, ls_timer);
+	int our_nodeid = dlm_our_nodeid();
+	struct dlm_rsb *r;
+	int rv;
+
+	while (1) {
+		/* interrupting point to leave iteration when
+		 * recovery waits for timer_delete_sync(), recovery
+		 * will take care to delete everything in toss queue.
+		 */
+		if (dlm_locking_stopped(ls))
+			break;
+
+		rv = spin_trylock(&ls->ls_toss_q_lock);
+		if (!rv) {
+			/* rearm again try timer */
+			__rsb_mod_timer(ls, DLM_TOSS_TIMER_RETRY);
+			break;
+		}
+
+		r = list_first_entry_or_null(&ls->ls_toss_q, struct dlm_rsb,
+					     res_toss_q_list);
+		if (!r) {
+			/* nothing to do anymore next rsb queue will
+			 * set next mod_timer() expire.
+			 */
+			spin_unlock(&ls->ls_toss_q_lock);
+			break;
+		}
+
+		/* test if the first rsb isn't expired yet, if
+		 * so we stop freeing rsb from toss queue as
+		 * the order in queue is ascending to the
+		 * absolute res_toss_time jiffies
+		 */
+		if (time_before(jiffies, r->res_toss_time)) {
+			/* rearm with the next rsb to expire in the future */
+			__rsb_mod_timer(ls, r->res_toss_time);
+			spin_unlock(&ls->ls_toss_q_lock);
+			break;
+		}
+
+		/* in find_rsb_dir/nodir there is a reverse order of this
+		 * lock, however this is only a trylock if we hit some
+		 * possible contention we try it again.
+		 *
+		 * This lock synchronized while holding ls_toss_q_lock
+		 * synchronize everything that rsb_delete_toss_timer()
+		 * or rsb_mod_timer() can't run after this timer callback
+		 * deletes the rsb from the ls_toss_q. Whereas the other
+		 * holders have always a priority to run as this is only
+		 * a caching handling and the other holders might to put
+		 * this rsb out of the toss state.
+		 */
+		rv = write_trylock(&ls->ls_rsbtbl_lock);
+		if (!rv) {
+			spin_unlock(&ls->ls_toss_q_lock);
+			/* rearm again try timer */
+			__rsb_mod_timer(ls, DLM_TOSS_TIMER_RETRY);
+			break;
+		}
+
+		list_del(&r->res_rsbs_list);
+		rhashtable_remove_fast(&ls->ls_rsbtbl, &r->res_node,
+				       dlm_rhash_rsb_params);
+
+		/* not necessary to held the ls_rsbtbl_lock when
+		 * calling send_remove()
+		 */
+		write_unlock(&ls->ls_rsbtbl_lock);
+
+		/* remove the rsb out of the toss queue its gone
+		 * drom DLM now
+		 */
+		list_del_init(&r->res_toss_q_list);
+		spin_unlock(&ls->ls_toss_q_lock);
+
+		/* no rsb in this state should ever run a timer */
+		WARN_ON(!dlm_no_directory(ls) &&
+			(r->res_master_nodeid != our_nodeid) &&
+			(dlm_dir_nodeid(r) == our_nodeid));
+
+		/* We're the master of this rsb but we're not
+		 * the directory record, so we need to tell the
+		 * dir node to remove the dir record
+		 */
+		if (!dlm_no_directory(ls) &&
+		    (r->res_master_nodeid == our_nodeid) &&
+		    (dlm_dir_nodeid(r) != our_nodeid))
+			send_remove(r);
+
+		free_toss_rsb(r);
+	}
+}
+
 /* If ls->ls_new_rsb is empty, return -EAGAIN, so the caller can
    unlock any spinlocks, go back and call pre_rsb_struct again.
    Otherwise, take an rsb off the list and return it. */
@@ -395,10 +654,10 @@ static int get_rsb_struct(struct dlm_ls *ls, const void *name, int len,
 	struct dlm_rsb *r;
 	int count;
 
-	spin_lock(&ls->ls_new_rsb_spin);
+	spin_lock_bh(&ls->ls_new_rsb_spin);
 	if (list_empty(&ls->ls_new_rsb)) {
 		count = ls->ls_new_rsb_count;
-		spin_unlock(&ls->ls_new_rsb_spin);
+		spin_unlock_bh(&ls->ls_new_rsb_spin);
 		log_debug(ls, "find_rsb retry %d %d %s",
 			  count, dlm_config.ci_new_rsb_count,
 			  (const char *)name);
@@ -407,88 +666,44 @@ static int get_rsb_struct(struct dlm_ls *ls, const void *name, int len,
 
 	r = list_first_entry(&ls->ls_new_rsb, struct dlm_rsb, res_hashchain);
 	list_del(&r->res_hashchain);
-	/* Convert the empty list_head to a NULL rb_node for tree usage: */
-	memset(&r->res_hashnode, 0, sizeof(struct rb_node));
 	ls->ls_new_rsb_count--;
-	spin_unlock(&ls->ls_new_rsb_spin);
+	spin_unlock_bh(&ls->ls_new_rsb_spin);
 
 	r->res_ls = ls;
 	r->res_length = len;
 	memcpy(r->res_name, name, len);
-	mutex_init(&r->res_mutex);
+	spin_lock_init(&r->res_lock);
 
 	INIT_LIST_HEAD(&r->res_lookup);
 	INIT_LIST_HEAD(&r->res_grantqueue);
 	INIT_LIST_HEAD(&r->res_convertqueue);
 	INIT_LIST_HEAD(&r->res_waitqueue);
 	INIT_LIST_HEAD(&r->res_root_list);
+	INIT_LIST_HEAD(&r->res_toss_q_list);
 	INIT_LIST_HEAD(&r->res_recover_list);
+	INIT_LIST_HEAD(&r->res_masters_list);
 
 	*r_ret = r;
 	return 0;
 }
 
-static int rsb_cmp(struct dlm_rsb *r, const char *name, int nlen)
-{
-	char maxname[DLM_RESNAME_MAXLEN];
-
-	memset(maxname, 0, DLM_RESNAME_MAXLEN);
-	memcpy(maxname, name, nlen);
-	return memcmp(r->res_name, maxname, DLM_RESNAME_MAXLEN);
-}
-
-int dlm_search_rsb_tree(struct rb_root *tree, const void *name, int len,
+int dlm_search_rsb_tree(struct rhashtable *rhash, const void *name, int len,
 			struct dlm_rsb **r_ret)
 {
-	struct rb_node *node = tree->rb_node;
-	struct dlm_rsb *r;
-	int rc;
-
-	while (node) {
-		r = rb_entry(node, struct dlm_rsb, res_hashnode);
-		rc = rsb_cmp(r, name, len);
-		if (rc < 0)
-			node = node->rb_left;
-		else if (rc > 0)
-			node = node->rb_right;
-		else
-			goto found;
-	}
-	*r_ret = NULL;
-	return -EBADR;
+	char key[DLM_RESNAME_MAXLEN] = {};
 
- found:
-	*r_ret = r;
+	memcpy(key, name, len);
+	*r_ret = rhashtable_lookup_fast(rhash, &key, dlm_rhash_rsb_params);
+	if (*r_ret)
 		return 0;
+
+	return -EBADR;
 }
 
-static int rsb_insert(struct dlm_rsb *rsb, struct rb_root *tree)
+static int rsb_insert(struct dlm_rsb *rsb, struct rhashtable *rhash)
 {
-	struct rb_node **newn = &tree->rb_node;
-	struct rb_node *parent = NULL;
-	int rc;
-
-	while (*newn) {
-		struct dlm_rsb *cur = rb_entry(*newn, struct dlm_rsb,
-					       res_hashnode);
-
-		parent = *newn;
-		rc = rsb_cmp(cur, rsb->res_name, rsb->res_length);
-		if (rc < 0)
-			newn = &parent->rb_left;
-		else if (rc > 0)
-			newn = &parent->rb_right;
-		else {
-			log_print("rsb_insert match");
-			dlm_dump_rsb(rsb);
-			dlm_dump_rsb(cur);
-			return -EEXIST;
-		}
-	}
-
-	rb_link_node(&rsb->res_hashnode, parent, newn);
-	rb_insert_color(&rsb->res_hashnode, tree);
-	return 0;
+	return rhashtable_insert_fast(rhash, &rsb->res_node,
+				      dlm_rhash_rsb_params);
 }
 
 /*
@@ -536,8 +751,7 @@ static int rsb_insert(struct dlm_rsb *rsb, struct rb_root *tree)
  */
 
 static int find_rsb_dir(struct dlm_ls *ls, const void *name, int len,
-			uint32_t hash, uint32_t b,
-			int dir_nodeid, int from_nodeid,
+			uint32_t hash, int dir_nodeid, int from_nodeid,
 			unsigned int flags, struct dlm_rsb **r_ret)
 {
 	struct dlm_rsb *r = NULL;
@@ -584,24 +798,46 @@ static int find_rsb_dir(struct dlm_ls *ls, const void *name, int len,
 			goto out;
 	}
 
-	spin_lock(&ls->ls_rsbtbl[b].lock);
+ retry_lookup:
 
-	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
-	if (error)
-		goto do_toss;
+	/* check if the rsb is in keep state under read lock - likely path */
+	read_lock_bh(&ls->ls_rsbtbl_lock);
+	error = dlm_search_rsb_tree(&ls->ls_rsbtbl, name, len, &r);
+	if (error) {
+		read_unlock_bh(&ls->ls_rsbtbl_lock);
+		goto do_new;
+	}
 	
 	/*
 	 * rsb is active, so we can't check master_nodeid without lock_rsb.
 	 */
 
+	if (rsb_flag(r, RSB_TOSS)) {
+		read_unlock_bh(&ls->ls_rsbtbl_lock);
+		goto do_toss;
+	}
+
 	kref_get(&r->res_ref);
-	goto out_unlock;
+	read_unlock_bh(&ls->ls_rsbtbl_lock);
+	goto out;
 
 
  do_toss:
-	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
-	if (error)
+	write_lock_bh(&ls->ls_rsbtbl_lock);
+
+	/* retry lookup under write lock to see if its still in toss state
+	 * if not it's in keep state and we relookup - unlikely path.
+	 */
+	error = dlm_search_rsb_tree(&ls->ls_rsbtbl, name, len, &r);
+	if (!error) {
+		if (!rsb_flag(r, RSB_TOSS)) {
+			write_unlock_bh(&ls->ls_rsbtbl_lock);
+			goto retry_lookup;
+		}
+	} else {
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
 		goto do_new;
+	}
 
 	/*
 	 * rsb found inactive (master_nodeid may be out of date unless
@@ -616,8 +852,9 @@ static int find_rsb_dir(struct dlm_ls *ls, const void *name, int len,
 		log_debug(ls, "find_rsb toss from_other %d master %d dir %d %s",
 			  from_nodeid, r->res_master_nodeid, dir_nodeid,
 			  r->res_name);
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
 		error = -ENOTBLK;
-		goto out_unlock;
+		goto out;
 	}
 
 	if ((r->res_master_nodeid != our_nodeid) && from_dir) {
@@ -639,9 +876,17 @@ static int find_rsb_dir(struct dlm_ls *ls, const void *name, int len,
 		r->res_first_lkid = 0;
 	}
 
-	rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
-	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
-	goto out_unlock;
+	list_move(&r->res_rsbs_list, &ls->ls_keep);
+	rsb_clear_flag(r, RSB_TOSS);
+	/* rsb got out of toss state, it becomes alive again
+	 * and we reinit the reference counter that is only
+	 * valid for keep state rsbs
+	 */
+	kref_init(&r->res_ref);
+	rsb_delete_toss_timer(ls, r);
+	write_unlock_bh(&ls->ls_rsbtbl_lock);
+
+	goto out;
 
 
  do_new:
@@ -650,18 +895,15 @@ static int find_rsb_dir(struct dlm_ls *ls, const void *name, int len,
 	 */
 
 	if (error == -EBADR && !create)
-		goto out_unlock;
+		goto out;
 
 	error = get_rsb_struct(ls, name, len, &r);
-	if (error == -EAGAIN) {
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
+	if (error == -EAGAIN)
 		goto retry;
-	}
 	if (error)
-		goto out_unlock;
+		goto out;
 
 	r->res_hash = hash;
-	r->res_bucket = b;
 	r->res_dir_nodeid = dir_nodeid;
 	kref_init(&r->res_ref);
 
@@ -681,7 +923,7 @@ static int find_rsb_dir(struct dlm_ls *ls, const void *name, int len,
 		dlm_free_rsb(r);
 		r = NULL;
 		error = -ENOTBLK;
-		goto out_unlock;
+		goto out;
 	}
 
 	if (from_other) {
@@ -701,9 +943,20 @@ static int find_rsb_dir(struct dlm_ls *ls, const void *name, int len,
 	}
 
  out_add:
-	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
- out_unlock:
-	spin_unlock(&ls->ls_rsbtbl[b].lock);
+
+	write_lock_bh(&ls->ls_rsbtbl_lock);
+	error = rsb_insert(r, &ls->ls_rsbtbl);
+	if (error == -EEXIST) {
+		/* somebody else was faster and it seems the
+		 * rsb exists now, we do a whole relookup
+		 */
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
+		dlm_free_rsb(r);
+		goto retry_lookup;
+	} else if (!error) {
+		list_add(&r->res_rsbs_list, &ls->ls_keep);
+	}
+	write_unlock_bh(&ls->ls_rsbtbl_lock);
  out:
 	*r_ret = r;
 	return error;
@@ -714,8 +967,7 @@ static int find_rsb_dir(struct dlm_ls *ls, const void *name, int len,
    dlm_recover_masters). */
 
 static int find_rsb_nodir(struct dlm_ls *ls, const void *name, int len,
-			  uint32_t hash, uint32_t b,
-			  int dir_nodeid, int from_nodeid,
+			  uint32_t hash, int dir_nodeid, int from_nodeid,
 			  unsigned int flags, struct dlm_rsb **r_ret)
 {
 	struct dlm_rsb *r = NULL;
@@ -728,24 +980,48 @@ static int find_rsb_nodir(struct dlm_ls *ls, const void *name, int len,
 	if (error < 0)
 		goto out;
 
-	spin_lock(&ls->ls_rsbtbl[b].lock);
+ retry_lookup:
 
-	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
-	if (error)
+	/* check if the rsb is in keep state under read lock - likely path */
+	read_lock_bh(&ls->ls_rsbtbl_lock);
+	error = dlm_search_rsb_tree(&ls->ls_rsbtbl, name, len, &r);
+	if (error) {
+		read_unlock_bh(&ls->ls_rsbtbl_lock);
+		goto do_new;
+	}
+
+	if (rsb_flag(r, RSB_TOSS)) {
+		read_unlock_bh(&ls->ls_rsbtbl_lock);
 		goto do_toss;
+	}
 
 	/*
 	 * rsb is active, so we can't check master_nodeid without lock_rsb.
 	 */
 
 	kref_get(&r->res_ref);
-	goto out_unlock;
+	read_unlock_bh(&ls->ls_rsbtbl_lock);
+
+	goto out;
 
 
  do_toss:
-	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
-	if (error)
+	write_lock_bh(&ls->ls_rsbtbl_lock);
+
+	/* retry lookup under write lock to see if its still in toss state
+	 * if not it's in keep state and we relookup - unlikely path.
+	 */
+	error = dlm_search_rsb_tree(&ls->ls_rsbtbl, name, len, &r);
+	if (!error) {
+		if (!rsb_flag(r, RSB_TOSS)) {
+			write_unlock_bh(&ls->ls_rsbtbl_lock);
+			goto retry_lookup;
+		}
+	} else {
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
 		goto do_new;
+	}
+
 
 	/*
 	 * rsb found inactive. No other thread is using this rsb because
@@ -759,8 +1035,9 @@ static int find_rsb_nodir(struct dlm_ls *ls, const void *name, int len,
 		log_error(ls, "find_rsb toss from_nodeid %d master %d dir %d",
 			  from_nodeid, r->res_master_nodeid, dir_nodeid);
 		dlm_print_rsb(r);
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
 		error = -ENOTBLK;
-		goto out_unlock;
+		goto out;
 	}
 
 	if (!recover && (r->res_master_nodeid != our_nodeid) &&
@@ -774,9 +1051,17 @@ static int find_rsb_nodir(struct dlm_ls *ls, const void *name, int len,
 		r->res_nodeid = 0;
 	}
 
-	rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
-	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
-	goto out_unlock;
+	list_move(&r->res_rsbs_list, &ls->ls_keep);
+	rsb_clear_flag(r, RSB_TOSS);
+	/* rsb got out of toss state, it becomes alive again
+	 * and we reinit the reference counter that is only
+	 * valid for keep state rsbs
+	 */
+	kref_init(&r->res_ref);
+	rsb_delete_toss_timer(ls, r);
+	write_unlock_bh(&ls->ls_rsbtbl_lock);
+
+	goto out;
 
 
  do_new:
@@ -786,22 +1071,31 @@ static int find_rsb_nodir(struct dlm_ls *ls, const void *name, int len,
 
 	error = get_rsb_struct(ls, name, len, &r);
 	if (error == -EAGAIN) {
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
 		goto retry;
 	}
 	if (error)
-		goto out_unlock;
+		goto out;
 
 	r->res_hash = hash;
-	r->res_bucket = b;
 	r->res_dir_nodeid = dir_nodeid;
 	r->res_master_nodeid = dir_nodeid;
 	r->res_nodeid = (dir_nodeid == our_nodeid) ? 0 : dir_nodeid;
 	kref_init(&r->res_ref);
 
-	error = rsb_insert(r, &ls->ls_rsbtbl[b].keep);
- out_unlock:
-	spin_unlock(&ls->ls_rsbtbl[b].lock);
+	write_lock_bh(&ls->ls_rsbtbl_lock);
+	error = rsb_insert(r, &ls->ls_rsbtbl);
+	if (error == -EEXIST) {
+		/* somebody else was faster and it seems the
+		 * rsb exists now, we do a whole relookup
+		 */
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
+		dlm_free_rsb(r);
+		goto retry_lookup;
+	} else if (!error) {
+		list_add(&r->res_rsbs_list, &ls->ls_keep);
+	}
+	write_unlock_bh(&ls->ls_rsbtbl_lock);
+
  out:
 	*r_ret = r;
 	return error;
@@ -811,22 +1105,20 @@ static int find_rsb(struct dlm_ls *ls, const void *name, int len,
 		    int from_nodeid, unsigned int flags,
 		    struct dlm_rsb **r_ret)
 {
-	uint32_t hash, b;
 	int dir_nodeid;
+	uint32_t hash;
 
 	if (len > DLM_RESNAME_MAXLEN)
 		return -EINVAL;
 
 	hash = jhash(name, len, 0);
-	b = hash & (ls->ls_rsbtbl_size - 1);
-
 	dir_nodeid = dlm_hash2nodeid(ls, hash);
 
 	if (dlm_no_directory(ls))
-		return find_rsb_nodir(ls, name, len, hash, b, dir_nodeid,
+		return find_rsb_nodir(ls, name, len, hash, dir_nodeid,
 				      from_nodeid, flags, r_ret);
 	else
-		return find_rsb_dir(ls, name, len, hash, b, dir_nodeid,
+		return find_rsb_dir(ls, name, len, hash, dir_nodeid,
 				    from_nodeid, flags, r_ret);
 }
 
@@ -988,7 +1280,7 @@ int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, const char *name,
 		      int len, unsigned int flags, int *r_nodeid, int *result)
 {
 	struct dlm_rsb *r = NULL;
-	uint32_t hash, b;
+	uint32_t hash;
 	int our_nodeid = dlm_our_nodeid();
 	int dir_nodeid, error;
 
@@ -1002,8 +1294,6 @@ int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, const char *name,
 	}
 
 	hash = jhash(name, len, 0);
-	b = hash & (ls->ls_rsbtbl_size - 1);
-
 	dir_nodeid = dlm_hash2nodeid(ls, hash);
 	if (dir_nodeid != our_nodeid) {
 		log_error(ls, "dlm_master_lookup from %d dir %d our %d h %x %d",
@@ -1018,15 +1308,23 @@ int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, const char *name,
 	if (error < 0)
 		return error;
 
-	spin_lock(&ls->ls_rsbtbl[b].lock);
-	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
+ retry_lookup:
+
+	/* check if the rsb is in keep state under read lock - likely path */
+	read_lock_bh(&ls->ls_rsbtbl_lock);
+	error = dlm_search_rsb_tree(&ls->ls_rsbtbl, name, len, &r);
 	if (!error) {
+		if (rsb_flag(r, RSB_TOSS)) {
+			read_unlock_bh(&ls->ls_rsbtbl_lock);
+			goto do_toss;
+		}
+
 		/* because the rsb is active, we need to lock_rsb before
 		 * checking/changing re_master_nodeid
 		 */
 
 		hold_rsb(r);
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
+		read_unlock_bh(&ls->ls_rsbtbl_lock);
 		lock_rsb(r);
 
 		__dlm_master_lookup(ls, r, our_nodeid, from_nodeid, false,
@@ -1037,11 +1335,31 @@ int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, const char *name,
 		put_rsb(r);
 
 		return 0;
+	} else {
+		read_unlock_bh(&ls->ls_rsbtbl_lock);
+		goto not_found;
 	}
 
-	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
-	if (error)
+ do_toss:
+	/* unlikely path - relookup under write */
+	write_lock_bh(&ls->ls_rsbtbl_lock);
+
+	/* rsb_mod_timer() requires to held ls_rsbtbl_lock in write lock
+	 * check if the rsb is still in toss state, if not relookup
+	 */
+	error = dlm_search_rsb_tree(&ls->ls_rsbtbl, name, len, &r);
+	if (!error) {
+		if (!rsb_flag(r, RSB_TOSS)) {
+			write_unlock_bh(&ls->ls_rsbtbl_lock);
+			/* something as changed, very unlikely but
+			 * try again
+			 */
+			goto retry_lookup;
+		}
+	} else {
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
 		goto not_found;
+	}
 
 	/* because the rsb is inactive (on toss list), it's not refcounted
 	 * and lock_rsb is not used, but is protected by the rsbtbl lock
@@ -1050,83 +1368,78 @@ int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, const char *name,
 	__dlm_master_lookup(ls, r, our_nodeid, from_nodeid, true, flags,
 			    r_nodeid, result);
 
-	r->res_toss_time = jiffies;
+	rsb_mod_timer(ls, r);
 	/* the rsb was inactive (on toss list) */
-	spin_unlock(&ls->ls_rsbtbl[b].lock);
+	write_unlock_bh(&ls->ls_rsbtbl_lock);
 
 	return 0;
 
  not_found:
 	error = get_rsb_struct(ls, name, len, &r);
-	if (error == -EAGAIN) {
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
+	if (error == -EAGAIN)
 		goto retry;
-	}
 	if (error)
-		goto out_unlock;
+		goto out;
 
 	r->res_hash = hash;
-	r->res_bucket = b;
 	r->res_dir_nodeid = our_nodeid;
 	r->res_master_nodeid = from_nodeid;
 	r->res_nodeid = from_nodeid;
 	kref_init(&r->res_ref);
-	r->res_toss_time = jiffies;
+	rsb_set_flag(r, RSB_TOSS);
 
-	error = rsb_insert(r, &ls->ls_rsbtbl[b].toss);
-	if (error) {
+	write_lock_bh(&ls->ls_rsbtbl_lock);
+	error = rsb_insert(r, &ls->ls_rsbtbl);
+	if (error == -EEXIST) {
+		/* somebody else was faster and it seems the
+		 * rsb exists now, we do a whole relookup
+		 */
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
+		dlm_free_rsb(r);
+		goto retry_lookup;
+	} else if (error) {
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
 		/* should never happen */
 		dlm_free_rsb(r);
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
 		goto retry;
 	}
 
+	list_add(&r->res_rsbs_list, &ls->ls_toss);
+	rsb_mod_timer(ls, r);
+	write_unlock_bh(&ls->ls_rsbtbl_lock);
+
 	if (result)
 		*result = DLM_LU_ADD;
 	*r_nodeid = from_nodeid;
- out_unlock:
-	spin_unlock(&ls->ls_rsbtbl[b].lock);
+ out:
 	return error;
 }
 
 static void dlm_dump_rsb_hash(struct dlm_ls *ls, uint32_t hash)
 {
-	struct rb_node *n;
 	struct dlm_rsb *r;
-	int i;
 
-	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
-		spin_lock(&ls->ls_rsbtbl[i].lock);
-		for (n = rb_first(&ls->ls_rsbtbl[i].keep); n; n = rb_next(n)) {
-			r = rb_entry(n, struct dlm_rsb, res_hashnode);
+	read_lock_bh(&ls->ls_rsbtbl_lock);
+	list_for_each_entry(r, &ls->ls_keep, res_rsbs_list) {
 		if (r->res_hash == hash)
 			dlm_dump_rsb(r);
 	}
-		spin_unlock(&ls->ls_rsbtbl[i].lock);
-	}
+	read_unlock_bh(&ls->ls_rsbtbl_lock);
 }
 
 void dlm_dump_rsb_name(struct dlm_ls *ls, const char *name, int len)
 {
 	struct dlm_rsb *r = NULL;
-	uint32_t hash, b;
 	int error;
 
-	hash = jhash(name, len, 0);
-	b = hash & (ls->ls_rsbtbl_size - 1);
-
-	spin_lock(&ls->ls_rsbtbl[b].lock);
-	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
+	read_lock_bh(&ls->ls_rsbtbl_lock);
+	error = dlm_search_rsb_tree(&ls->ls_rsbtbl, name, len, &r);
 	if (!error)
-		goto out_dump;
-
-	error = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
-	if (error)
 		goto out;
- out_dump:
+
 	dlm_dump_rsb(r);
  out:
-	spin_unlock(&ls->ls_rsbtbl[b].lock);
+	read_unlock_bh(&ls->ls_rsbtbl_lock);
 }
 
 static void toss_rsb(struct kref *kref)
@@ -1135,11 +1448,10 @@ static void toss_rsb(struct kref *kref)
 	struct dlm_ls *ls = r->res_ls;
 
 	DLM_ASSERT(list_empty(&r->res_root_list), dlm_print_rsb(r););
-	kref_init(&r->res_ref);
-	rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[r->res_bucket].keep);
-	rsb_insert(r, &ls->ls_rsbtbl[r->res_bucket].toss);
-	r->res_toss_time = jiffies;
-	set_bit(DLM_RTF_SHRINK_BIT, &ls->ls_rsbtbl[r->res_bucket].flags);
+	rsb_set_flag(r, RSB_TOSS);
+	list_move(&r->res_rsbs_list, &ls->ls_toss);
+	rsb_mod_timer(ls, r);
+
 	if (r->res_lvbptr) {
 		dlm_free_lvb(r->res_lvbptr);
 		r->res_lvbptr = NULL;
@@ -1151,23 +1463,27 @@ static void toss_rsb(struct kref *kref)
 static void unhold_rsb(struct dlm_rsb *r)
 {
 	int rv;
+
+	/* rsbs in toss state never get referenced */
+	WARN_ON(rsb_flag(r, RSB_TOSS));
 	rv = kref_put(&r->res_ref, toss_rsb);
 	DLM_ASSERT(!rv, dlm_dump_rsb(r););
 }
 
-static void kill_rsb(struct kref *kref)
+void free_toss_rsb(struct dlm_rsb *r)
 {
-	struct dlm_rsb *r = container_of(kref, struct dlm_rsb, res_ref);
-
-	/* All work is done after the return from kref_put() so we
-	   can release the write_lock before the remove and free. */
+	WARN_ON_ONCE(!rsb_flag(r, RSB_TOSS));
 
 	DLM_ASSERT(list_empty(&r->res_lookup), dlm_dump_rsb(r););
 	DLM_ASSERT(list_empty(&r->res_grantqueue), dlm_dump_rsb(r););
 	DLM_ASSERT(list_empty(&r->res_convertqueue), dlm_dump_rsb(r););
 	DLM_ASSERT(list_empty(&r->res_waitqueue), dlm_dump_rsb(r););
 	DLM_ASSERT(list_empty(&r->res_root_list), dlm_dump_rsb(r););
+	DLM_ASSERT(list_empty(&r->res_toss_q_list), dlm_dump_rsb(r););
 	DLM_ASSERT(list_empty(&r->res_recover_list), dlm_dump_rsb(r););
+	DLM_ASSERT(list_empty(&r->res_masters_list), dlm_dump_rsb(r););
+
+	dlm_free_rsb(r);
 }
 
 /* Attaching/detaching lkb's from rsb's is for rsb reference counting.
@@ -1197,24 +1513,20 @@ static int _create_lkb(struct dlm_ls *ls, struct dlm_lkb **lkb_ret,
 	if (!lkb)
 		return -ENOMEM;
 
-	lkb->lkb_last_bast_mode = -1;
+	lkb->lkb_last_bast_cb_mode = DLM_LOCK_IV;
+	lkb->lkb_last_cast_cb_mode = DLM_LOCK_IV;
+	lkb->lkb_last_cb_mode = DLM_LOCK_IV;
 	lkb->lkb_nodeid = -1;
 	lkb->lkb_grmode = DLM_LOCK_IV;
 	kref_init(&lkb->lkb_ref);
 	INIT_LIST_HEAD(&lkb->lkb_ownqueue);
 	INIT_LIST_HEAD(&lkb->lkb_rsb_lookup);
-	INIT_LIST_HEAD(&lkb->lkb_cb_list);
-	INIT_LIST_HEAD(&lkb->lkb_callbacks);
-	spin_lock_init(&lkb->lkb_cb_lock);
-	INIT_WORK(&lkb->lkb_cb_work, dlm_callback_work);
 
-	idr_preload(GFP_NOFS);
-	spin_lock(&ls->ls_lkbidr_spin);
+	write_lock_bh(&ls->ls_lkbidr_lock);
 	rv = idr_alloc(&ls->ls_lkbidr, lkb, start, end, GFP_NOWAIT);
 	if (rv >= 0)
 		lkb->lkb_id = rv;
-	spin_unlock(&ls->ls_lkbidr_spin);
-	idr_preload_end();
+	write_unlock_bh(&ls->ls_lkbidr_lock);
 
 	if (rv < 0) {
 		log_error(ls, "create_lkb idr error %d", rv);
@@ -1235,11 +1547,11 @@ static int find_lkb(struct dlm_ls *ls, uint32_t lkid, struct dlm_lkb **lkb_ret)
 {
 	struct dlm_lkb *lkb;
 
-	spin_lock(&ls->ls_lkbidr_spin);
+	read_lock_bh(&ls->ls_lkbidr_lock);
 	lkb = idr_find(&ls->ls_lkbidr, lkid);
 	if (lkb)
 		kref_get(&lkb->lkb_ref);
-	spin_unlock(&ls->ls_lkbidr_spin);
+	read_unlock_bh(&ls->ls_lkbidr_lock);
 
 	*lkb_ret = lkb;
 	return lkb ? 0 : -ENOENT;
@@ -1263,11 +1575,11 @@ static int __put_lkb(struct dlm_ls *ls, struct dlm_lkb *lkb)
 	uint32_t lkid = lkb->lkb_id;
 	int rv;
 
-	rv = kref_put_lock(&lkb->lkb_ref, kill_lkb,
-			   &ls->ls_lkbidr_spin);
+	rv = dlm_kref_put_write_lock_bh(&lkb->lkb_ref, kill_lkb,
+					&ls->ls_lkbidr_lock);
 	if (rv) {
 		idr_remove(&ls->ls_lkbidr, lkid);
-		spin_unlock(&ls->ls_lkbidr_spin);
+		write_unlock_bh(&ls->ls_lkbidr_lock);
 
 		detach_lkb(lkb);
 
@@ -1408,7 +1720,7 @@ static int add_to_waiters(struct dlm_lkb *lkb, int mstype, int to_nodeid)
 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
 	int error = 0;
 
-	mutex_lock(&ls->ls_waiters_mutex);
+	spin_lock_bh(&ls->ls_waiters_lock);
 
 	if (is_overlap_unlock(lkb) ||
 	    (is_overlap_cancel(lkb) && (mstype == DLM_MSG_CANCEL))) {
@@ -1451,7 +1763,7 @@ static int add_to_waiters(struct dlm_lkb *lkb, int mstype, int to_nodeid)
 		log_error(ls, "addwait error %x %d flags %x %d %d %s",
 			  lkb->lkb_id, error, dlm_iflags_val(lkb), mstype,
 			  lkb->lkb_wait_type, lkb->lkb_resource->res_name);
-	mutex_unlock(&ls->ls_waiters_mutex);
+	spin_unlock_bh(&ls->ls_waiters_lock);
 	return error;
 }
 
@@ -1551,14 +1863,18 @@ static int remove_from_waiters(struct dlm_lkb *lkb, int mstype)
 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
 	int error;
 
-	mutex_lock(&ls->ls_waiters_mutex);
+	spin_lock_bh(&ls->ls_waiters_lock);
 	error = _remove_from_waiters(lkb, mstype, NULL);
-	mutex_unlock(&ls->ls_waiters_mutex);
+	spin_unlock_bh(&ls->ls_waiters_lock);
 	return error;
 }
 
 /* Handles situations where we might be processing a "fake" or "local" reply in
-   which we can't try to take waiters_mutex again. */
+ * the recovery context which stops any locking activity. Only debugfs might
+ * change the lockspace waiters but they will held the recovery lock to ensure
+ * remove_from_waiters_ms() in local case will be the only user manipulating the
+ * lockspace waiters in recovery context.
+ */
 
 static int remove_from_waiters_ms(struct dlm_lkb *lkb,
 				  const struct dlm_message *ms, bool local)
@@ -1567,159 +1883,16 @@ static int remove_from_waiters_ms(struct dlm_lkb *lkb,
 	int error;
 
 	if (!local)
-		mutex_lock(&ls->ls_waiters_mutex);
+		spin_lock_bh(&ls->ls_waiters_lock);
+	else
+		WARN_ON_ONCE(!rwsem_is_locked(&ls->ls_in_recovery) ||
+			     !dlm_locking_stopped(ls));
 	error = _remove_from_waiters(lkb, le32_to_cpu(ms->m_type), ms);
 	if (!local)
-		mutex_unlock(&ls->ls_waiters_mutex);
+		spin_unlock_bh(&ls->ls_waiters_lock);
 	return error;
 }
 
-static void shrink_bucket(struct dlm_ls *ls, int b)
-{
-	struct rb_node *n, *next;
-	struct dlm_rsb *r;
-	char *name;
-	int our_nodeid = dlm_our_nodeid();
-	int remote_count = 0;
-	int need_shrink = 0;
-	int i, len, rv;
-
-	memset(&ls->ls_remove_lens, 0, sizeof(int) * DLM_REMOVE_NAMES_MAX);
-
-	spin_lock(&ls->ls_rsbtbl[b].lock);
-
-	if (!test_bit(DLM_RTF_SHRINK_BIT, &ls->ls_rsbtbl[b].flags)) {
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
-		return;
-	}
-
-	for (n = rb_first(&ls->ls_rsbtbl[b].toss); n; n = next) {
-		next = rb_next(n);
-		r = rb_entry(n, struct dlm_rsb, res_hashnode);
-
-		/* If we're the directory record for this rsb, and
-		   we're not the master of it, then we need to wait
-		   for the master node to send us a dir remove for
-		   before removing the dir record. */
-
-		if (!dlm_no_directory(ls) &&
-		    (r->res_master_nodeid != our_nodeid) &&
-		    (dlm_dir_nodeid(r) == our_nodeid)) {
-			continue;
-		}
-
-		need_shrink = 1;
-
-		if (!time_after_eq(jiffies, r->res_toss_time +
-				   dlm_config.ci_toss_secs * HZ)) {
-			continue;
-		}
-
-		if (!dlm_no_directory(ls) &&
-		    (r->res_master_nodeid == our_nodeid) &&
-		    (dlm_dir_nodeid(r) != our_nodeid)) {
-
-			/* We're the master of this rsb but we're not
-			   the directory record, so we need to tell the
-			   dir node to remove the dir record. */
-
-			ls->ls_remove_lens[remote_count] = r->res_length;
-			memcpy(ls->ls_remove_names[remote_count], r->res_name,
-			       DLM_RESNAME_MAXLEN);
-			remote_count++;
-
-			if (remote_count >= DLM_REMOVE_NAMES_MAX)
-				break;
-			continue;
-		}
-
-		if (!kref_put(&r->res_ref, kill_rsb)) {
-			log_error(ls, "tossed rsb in use %s", r->res_name);
-			continue;
-		}
-
-		rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
-		dlm_free_rsb(r);
-	}
-
-	if (need_shrink)
-		set_bit(DLM_RTF_SHRINK_BIT, &ls->ls_rsbtbl[b].flags);
-	else
-		clear_bit(DLM_RTF_SHRINK_BIT, &ls->ls_rsbtbl[b].flags);
-	spin_unlock(&ls->ls_rsbtbl[b].lock);
-
-	/*
-	 * While searching for rsb's to free, we found some that require
-	 * remote removal.  We leave them in place and find them again here
-	 * so there is a very small gap between removing them from the toss
-	 * list and sending the removal.  Keeping this gap small is
-	 * important to keep us (the master node) from being out of sync
-	 * with the remote dir node for very long.
-	 */
-
-	for (i = 0; i < remote_count; i++) {
-		name = ls->ls_remove_names[i];
-		len = ls->ls_remove_lens[i];
-
-		spin_lock(&ls->ls_rsbtbl[b].lock);
-		rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
-		if (rv) {
-			spin_unlock(&ls->ls_rsbtbl[b].lock);
-			log_debug(ls, "remove_name not toss %s", name);
-			continue;
-		}
-
-		if (r->res_master_nodeid != our_nodeid) {
-			spin_unlock(&ls->ls_rsbtbl[b].lock);
-			log_debug(ls, "remove_name master %d dir %d our %d %s",
-				  r->res_master_nodeid, r->res_dir_nodeid,
-				  our_nodeid, name);
-			continue;
-		}
-
-		if (r->res_dir_nodeid == our_nodeid) {
-			/* should never happen */
-			spin_unlock(&ls->ls_rsbtbl[b].lock);
-			log_error(ls, "remove_name dir %d master %d our %d %s",
-				  r->res_dir_nodeid, r->res_master_nodeid,
-				  our_nodeid, name);
-			continue;
-		}
-
-		if (!time_after_eq(jiffies, r->res_toss_time +
-				   dlm_config.ci_toss_secs * HZ)) {
-			spin_unlock(&ls->ls_rsbtbl[b].lock);
-			log_debug(ls, "remove_name toss_time %lu now %lu %s",
-				  r->res_toss_time, jiffies, name);
-			continue;
-		}
-
-		if (!kref_put(&r->res_ref, kill_rsb)) {
-			spin_unlock(&ls->ls_rsbtbl[b].lock);
-			log_error(ls, "remove_name in use %s", name);
-			continue;
-		}
-
-		rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
-		send_remove(r);
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
-
-		dlm_free_rsb(r);
-	}
-}
-
-void dlm_scan_rsbs(struct dlm_ls *ls)
-{
-	int i;
-
-	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
-		shrink_bucket(ls, i);
-		if (dlm_locking_stopped(ls))
-			break;
-		cond_resched();
-	}
-}
-
 /* lkb is master or local copy */
 
 static void set_lvb_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
@@ -2538,7 +2711,6 @@ static void process_lookup_list(struct dlm_rsb *r)
 	list_for_each_entry_safe(lkb, safe, &r->res_lookup, lkb_rsb_lookup) {
 		list_del_init(&lkb->lkb_rsb_lookup);
 		_request_lock(r, lkb);
-		schedule();
 	}
 }
 
@@ -3332,8 +3504,7 @@ int dlm_unlock(dlm_lockspace_t *lockspace,
 static int _create_message(struct dlm_ls *ls, int mb_len,
 			   int to_nodeid, int mstype,
 			   struct dlm_message **ms_ret,
-			   struct dlm_mhandle **mh_ret,
-			   gfp_t allocation)
+			   struct dlm_mhandle **mh_ret)
 {
 	struct dlm_message *ms;
 	struct dlm_mhandle *mh;
@@ -3343,7 +3514,7 @@ static int _create_message(struct dlm_ls *ls, int mb_len,
 	   pass into midcomms_commit and a message buffer (mb) that we
 	   write our data into */
 
-	mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, allocation, &mb);
+	mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, &mb);
 	if (!mh)
 		return -ENOBUFS;
 
@@ -3365,8 +3536,7 @@ static int _create_message(struct dlm_ls *ls, int mb_len,
 static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
 			  int to_nodeid, int mstype,
 			  struct dlm_message **ms_ret,
-			  struct dlm_mhandle **mh_ret,
-			  gfp_t allocation)
+			  struct dlm_mhandle **mh_ret)
 {
 	int mb_len = sizeof(struct dlm_message);
 
@@ -3387,7 +3557,7 @@ static int create_message(struct dlm_rsb *r, struct dlm_lkb *lkb,
 	}
 
 	return _create_message(r->res_ls, mb_len, to_nodeid, mstype,
-			       ms_ret, mh_ret, allocation);
+			       ms_ret, mh_ret);
 }
 
 /* further lowcomms enhancements or alternate implementations may make
@@ -3456,7 +3626,7 @@ static int send_common(struct dlm_rsb *r, struct dlm_lkb *lkb, int mstype)
 	if (error)
 		return error;
 
-	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh, GFP_NOFS);
+	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
 	if (error)
 		goto fail;
 
@@ -3516,8 +3686,7 @@ static int send_grant(struct dlm_rsb *r, struct dlm_lkb *lkb)
 
 	to_nodeid = lkb->lkb_nodeid;
 
-	error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh,
-			       GFP_NOFS);
+	error = create_message(r, lkb, to_nodeid, DLM_MSG_GRANT, &ms, &mh);
 	if (error)
 		goto out;
 
@@ -3538,8 +3707,7 @@ static int send_bast(struct dlm_rsb *r, struct dlm_lkb *lkb, int mode)
 
 	to_nodeid = lkb->lkb_nodeid;
 
-	error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh,
-			       GFP_NOFS);
+	error = create_message(r, NULL, to_nodeid, DLM_MSG_BAST, &ms, &mh);
 	if (error)
 		goto out;
 
@@ -3564,8 +3732,7 @@ static int send_lookup(struct dlm_rsb *r, struct dlm_lkb *lkb)
 	if (error)
 		return error;
 
-	error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh,
-			       GFP_NOFS);
+	error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh);
 	if (error)
 		goto fail;
 
@@ -3589,8 +3756,7 @@ static int send_remove(struct dlm_rsb *r)
 
 	to_nodeid = dlm_dir_nodeid(r);
 
-	error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh,
-			       GFP_ATOMIC);
+	error = create_message(r, NULL, to_nodeid, DLM_MSG_REMOVE, &ms, &mh);
 	if (error)
 		goto out;
 
@@ -3611,7 +3777,7 @@ static int send_common_reply(struct dlm_rsb *r, struct dlm_lkb *lkb,
 
 	to_nodeid = lkb->lkb_nodeid;
 
-	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh, GFP_NOFS);
+	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
 	if (error)
 		goto out;
 
@@ -3653,8 +3819,7 @@ static int send_lookup_reply(struct dlm_ls *ls,
 	struct dlm_mhandle *mh;
 	int error, nodeid = le32_to_cpu(ms_in->m_header.h_nodeid);
 
-	error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh,
-			       GFP_NOFS);
+	error = create_message(r, NULL, nodeid, DLM_MSG_LOOKUP_REPLY, &ms, &mh);
 	if (error)
 		goto out;
 
@@ -4139,7 +4304,6 @@ static void receive_remove(struct dlm_ls *ls, const struct dlm_message *ms)
 {
 	char name[DLM_RESNAME_MAXLEN+1];
 	struct dlm_rsb *r;
-	uint32_t hash, b;
 	int rv, len, dir_nodeid, from_nodeid;
 
 	from_nodeid = le32_to_cpu(ms->m_header.h_nodeid);
@@ -4159,47 +4323,44 @@ static void receive_remove(struct dlm_ls *ls, const struct dlm_message *ms)
 		return;
 	}
 
-	/* Look for name on rsbtbl.toss, if it's there, kill it.
-	   If it's on rsbtbl.keep, it's being used, and we should ignore this
-	   message.  This is an expected race between the dir node sending a
-	   request to the master node at the same time as the master node sends
-	   a remove to the dir node.  The resolution to that race is for the
-	   dir node to ignore the remove message, and the master node to
-	   recreate the master rsb when it gets a request from the dir node for
-	   an rsb it doesn't have. */
+	/* Look for name in rsb toss state, if it's there, kill it.
+	 * If it's in non toss state, it's being used, and we should ignore this
+	 * message.  This is an expected race between the dir node sending a
+	 * request to the master node at the same time as the master node sends
+	 * a remove to the dir node.  The resolution to that race is for the
+	 * dir node to ignore the remove message, and the master node to
+	 * recreate the master rsb when it gets a request from the dir node for
+	 * an rsb it doesn't have.
+	 */
 
 	memset(name, 0, sizeof(name));
 	memcpy(name, ms->m_extra, len);
 
-	hash = jhash(name, len, 0);
-	b = hash & (ls->ls_rsbtbl_size - 1);
-
-	spin_lock(&ls->ls_rsbtbl[b].lock);
+	write_lock_bh(&ls->ls_rsbtbl_lock);
 
-	rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].toss, name, len, &r);
-	if (rv) {
-		/* verify the rsb is on keep list per comment above */
-		rv = dlm_search_rsb_tree(&ls->ls_rsbtbl[b].keep, name, len, &r);
+	rv = dlm_search_rsb_tree(&ls->ls_rsbtbl, name, len, &r);
 	if (rv) {
 		/* should not happen */
-			log_error(ls, "receive_remove from %d not found %s",
+		log_error(ls, "%s from %d not found %s", __func__,
 			  from_nodeid, name);
-			spin_unlock(&ls->ls_rsbtbl[b].lock);
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
 		return;
 	}
+
+	if (!rsb_flag(r, RSB_TOSS)) {
 		if (r->res_master_nodeid != from_nodeid) {
 			/* should not happen */
 			log_error(ls, "receive_remove keep from %d master %d",
 				  from_nodeid, r->res_master_nodeid);
 			dlm_print_rsb(r);
-			spin_unlock(&ls->ls_rsbtbl[b].lock);
+			write_unlock_bh(&ls->ls_rsbtbl_lock);
 			return;
 		}
 
 		log_debug(ls, "receive_remove from %d master %d first %x %s",
 			  from_nodeid, r->res_master_nodeid, r->res_first_lkid,
 			  name);
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
 		return;
 	}
 
@@ -4207,20 +4368,16 @@ static void receive_remove(struct dlm_ls *ls, const struct dlm_message *ms)
 		log_error(ls, "receive_remove toss from %d master %d",
 			  from_nodeid, r->res_master_nodeid);
 		dlm_print_rsb(r);
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
+		write_unlock_bh(&ls->ls_rsbtbl_lock);
 		return;
 	}
 
-	if (kref_put(&r->res_ref, kill_rsb)) {
-		rb_erase(&r->res_hashnode, &ls->ls_rsbtbl[b].toss);
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
-		dlm_free_rsb(r);
-	} else {
-		log_error(ls, "receive_remove from %d rsb ref error",
-			  from_nodeid);
-		dlm_print_rsb(r);
-		spin_unlock(&ls->ls_rsbtbl[b].lock);
-	}
+	list_del(&r->res_rsbs_list);
+	rhashtable_remove_fast(&ls->ls_rsbtbl, &r->res_node,
+			       dlm_rhash_rsb_params);
+	write_unlock_bh(&ls->ls_rsbtbl_lock);
+
+	free_toss_rsb(r);
 }
 
 static void receive_purge(struct dlm_ls *ls, const struct dlm_message *ms)
@@ -4407,7 +4564,6 @@ static void _receive_convert_reply(struct dlm_lkb *lkb,
 	if (error)
 		goto out;
 
-	/* local reply can happen with waiters_mutex held */
 	error = remove_from_waiters_ms(lkb, ms, local);
 	if (error)
 		goto out;
@@ -4446,7 +4602,6 @@ static void _receive_unlock_reply(struct dlm_lkb *lkb,
 	if (error)
 		goto out;
 
-	/* local reply can happen with waiters_mutex held */
 	error = remove_from_waiters_ms(lkb, ms, local);
 	if (error)
 		goto out;
@@ -4498,7 +4653,6 @@ static void _receive_cancel_reply(struct dlm_lkb *lkb,
 	if (error)
 		goto out;
 
-	/* local reply can happen with waiters_mutex held */
 	error = remove_from_waiters_ms(lkb, ms, local);
 	if (error)
 		goto out;
@@ -4757,20 +4911,32 @@ static void _receive_message(struct dlm_ls *ls, const struct dlm_message *ms,
 static void dlm_receive_message(struct dlm_ls *ls, const struct dlm_message *ms,
 				int nodeid)
 {
-	if (dlm_locking_stopped(ls)) {
+try_again:
+	read_lock_bh(&ls->ls_requestqueue_lock);
+	if (test_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags)) {
 		/* If we were a member of this lockspace, left, and rejoined,
 		   other nodes may still be sending us messages from the
 		   lockspace generation before we left. */
 		if (WARN_ON_ONCE(!ls->ls_generation)) {
+			read_unlock_bh(&ls->ls_requestqueue_lock);
 			log_limit(ls, "receive %d from %d ignore old gen",
 				  le32_to_cpu(ms->m_type), nodeid);
 			return;
 		}
 
+		read_unlock_bh(&ls->ls_requestqueue_lock);
+		write_lock_bh(&ls->ls_requestqueue_lock);
+		/* recheck because we hold writelock now */
+		if (!test_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags)) {
+			write_unlock_bh(&ls->ls_requestqueue_lock);
+			goto try_again;
+		}
+
 		dlm_add_requestqueue(ls, nodeid, ms);
+		write_unlock_bh(&ls->ls_requestqueue_lock);
 	} else {
-		dlm_wait_requestqueue(ls);
 		_receive_message(ls, ms, 0);
+		read_unlock_bh(&ls->ls_requestqueue_lock);
 	}
 }
 
@@ -4830,7 +4996,7 @@ void dlm_receive_buffer(const union dlm_packet *p, int nodeid)
 	/* this rwsem allows dlm_ls_stop() to wait for all dlm_recv threads to
 	   be inactive (in this ls) before transitioning to recovery mode */
 
-	down_read(&ls->ls_recv_active);
+	read_lock_bh(&ls->ls_recv_active);
 	if (hd->h_cmd == DLM_MSG)
 		dlm_receive_message(ls, &p->message, nodeid);
 	else if (hd->h_cmd == DLM_RCOM)
@@ -4838,7 +5004,7 @@ void dlm_receive_buffer(const union dlm_packet *p, int nodeid)
 	else
 		log_error(ls, "invalid h_cmd %d from %d lockspace %x",
 			  hd->h_cmd, nodeid, le32_to_cpu(hd->u.h_lockspace));
-	up_read(&ls->ls_recv_active);
+	read_unlock_bh(&ls->ls_recv_active);
 
 	dlm_put_lockspace(ls);
 }
@@ -4899,8 +5065,6 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls)
 	if (!ms_local)
 		return;
 
-	mutex_lock(&ls->ls_waiters_mutex);
-
 	list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
 
 		dir_nodeid = dlm_dir_nodeid(lkb->lkb_resource);
@@ -4993,7 +5157,6 @@ void dlm_recover_waiters_pre(struct dlm_ls *ls)
 		}
 		schedule();
 	}
-	mutex_unlock(&ls->ls_waiters_mutex);
 	kfree(ms_local);
 }
 
@@ -5001,7 +5164,7 @@ static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
 {
 	struct dlm_lkb *lkb = NULL, *iter;
 
-	mutex_lock(&ls->ls_waiters_mutex);
+	spin_lock_bh(&ls->ls_waiters_lock);
 	list_for_each_entry(iter, &ls->ls_waiters, lkb_wait_reply) {
 		if (test_bit(DLM_IFL_RESEND_BIT, &iter->lkb_iflags)) {
 			hold_lkb(iter);
@@ -5009,7 +5172,7 @@ static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
 			break;
 		}
 	}
-	mutex_unlock(&ls->ls_waiters_mutex);
+	spin_unlock_bh(&ls->ls_waiters_lock);
 
 	return lkb;
 }
@@ -5109,9 +5272,9 @@ int dlm_recover_waiters_post(struct dlm_ls *ls)
 		}
 
 		/* Forcibly remove from waiters list */
-		mutex_lock(&ls->ls_waiters_mutex);
+		spin_lock_bh(&ls->ls_waiters_lock);
 		list_del_init(&lkb->lkb_wait_reply);
-		mutex_unlock(&ls->ls_waiters_mutex);
+		spin_unlock_bh(&ls->ls_waiters_lock);
 
 		/*
 		 * The lkb is now clear of all prior waiters state and can be
@@ -5236,7 +5399,7 @@ static void purge_dead_list(struct dlm_ls *ls, struct dlm_rsb *r,
 
 /* Get rid of locks held by nodes that are gone. */
 
-void dlm_recover_purge(struct dlm_ls *ls)
+void dlm_recover_purge(struct dlm_ls *ls, const struct list_head *root_list)
 {
 	struct dlm_rsb *r;
 	struct dlm_member *memb;
@@ -5255,8 +5418,7 @@ void dlm_recover_purge(struct dlm_ls *ls)
 	if (!nodes_count)
 		return;
 
-	down_write(&ls->ls_root_sem);
-	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+	list_for_each_entry(r, root_list, res_root_list) {
 		hold_rsb(r);
 		lock_rsb(r);
 		if (is_master(r)) {
@@ -5271,22 +5433,18 @@ void dlm_recover_purge(struct dlm_ls *ls)
 		unhold_rsb(r);
 		cond_resched();
 	}
-	up_write(&ls->ls_root_sem);
 
 	if (lkb_count)
 		log_rinfo(ls, "dlm_recover_purge %u locks for %u nodes",
 			  lkb_count, nodes_count);
 }
 
-static struct dlm_rsb *find_grant_rsb(struct dlm_ls *ls, int bucket)
+static struct dlm_rsb *find_grant_rsb(struct dlm_ls *ls)
 {
-	struct rb_node *n;
 	struct dlm_rsb *r;
 
-	spin_lock(&ls->ls_rsbtbl[bucket].lock);
-	for (n = rb_first(&ls->ls_rsbtbl[bucket].keep); n; n = rb_next(n)) {
-		r = rb_entry(n, struct dlm_rsb, res_hashnode);
-
+	read_lock_bh(&ls->ls_rsbtbl_lock);
+	list_for_each_entry(r, &ls->ls_keep, res_rsbs_list) {
 		if (!rsb_flag(r, RSB_RECOVER_GRANT))
 			continue;
 		if (!is_master(r)) {
@@ -5294,10 +5452,10 @@ static struct dlm_rsb *find_grant_rsb(struct dlm_ls *ls, int bucket)
 			continue;
 		}
 		hold_rsb(r);
-		spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+		read_unlock_bh(&ls->ls_rsbtbl_lock);
 		return r;
 	}
-	spin_unlock(&ls->ls_rsbtbl[bucket].lock);
+	read_unlock_bh(&ls->ls_rsbtbl_lock);
 	return NULL;
 }
 
@@ -5321,19 +5479,15 @@ static struct dlm_rsb *find_grant_rsb(struct dlm_ls *ls, int bucket)
 void dlm_recover_grant(struct dlm_ls *ls)
 {
 	struct dlm_rsb *r;
-	int bucket = 0;
 	unsigned int count = 0;
 	unsigned int rsb_count = 0;
 	unsigned int lkb_count = 0;
 
 	while (1) {
-		r = find_grant_rsb(ls, bucket);
-		if (!r) {
-			if (bucket == ls->ls_rsbtbl_size - 1)
+		r = find_grant_rsb(ls);
+		if (!r)
 			break;
-			bucket++;
-			continue;
-		}
+
 		rsb_count++;
 		count = 0;
 		lock_rsb(r);
@@ -5641,10 +5795,10 @@ int dlm_user_request(struct dlm_ls *ls, struct dlm_user_args *ua,
 	}
 
 	/* add this new lkb to the per-process list of locks */
-	spin_lock(&ua->proc->locks_spin);
+	spin_lock_bh(&ua->proc->locks_spin);
 	hold_lkb(lkb);
 	list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
-	spin_unlock(&ua->proc->locks_spin);
+	spin_unlock_bh(&ua->proc->locks_spin);
 	do_put = false;
  out_put:
 	trace_dlm_lock_end(ls, lkb, name, namelen, mode, flags, error, false);
@@ -5726,7 +5880,7 @@ int dlm_user_adopt_orphan(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
 	int found_other_mode = 0;
 	int rv = 0;
 
-	mutex_lock(&ls->ls_orphans_mutex);
+	spin_lock_bh(&ls->ls_orphans_lock);
 	list_for_each_entry(iter, &ls->ls_orphans, lkb_ownqueue) {
 		if (iter->lkb_resource->res_length != namelen)
 			continue;
@@ -5743,7 +5897,7 @@ int dlm_user_adopt_orphan(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
 		*lkid = iter->lkb_id;
 		break;
 	}
-	mutex_unlock(&ls->ls_orphans_mutex);
+	spin_unlock_bh(&ls->ls_orphans_lock);
 
 	if (!lkb && found_other_mode) {
 		rv = -EAGAIN;
@@ -5774,9 +5928,9 @@ int dlm_user_adopt_orphan(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
 	 * for the proc locks list.
 	 */
 
-	spin_lock(&ua->proc->locks_spin);
+	spin_lock_bh(&ua->proc->locks_spin);
 	list_add_tail(&lkb->lkb_ownqueue, &ua->proc->locks);
-	spin_unlock(&ua->proc->locks_spin);
+	spin_unlock_bh(&ua->proc->locks_spin);
  out:
 	kfree(ua_tmp);
 	return rv;
@@ -5820,11 +5974,11 @@ int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
 	if (error)
 		goto out_put;
 
-	spin_lock(&ua->proc->locks_spin);
+	spin_lock_bh(&ua->proc->locks_spin);
 	/* dlm_user_add_cb() may have already taken lkb off the proc list */
 	if (!list_empty(&lkb->lkb_ownqueue))
 		list_move(&lkb->lkb_ownqueue, &ua->proc->unlocking);
-	spin_unlock(&ua->proc->locks_spin);
+	spin_unlock_bh(&ua->proc->locks_spin);
  out_put:
 	trace_dlm_unlock_end(ls, lkb, flags, error);
 	dlm_put_lkb(lkb);
@@ -5935,9 +6089,9 @@ static int orphan_proc_lock(struct dlm_ls *ls, struct dlm_lkb *lkb)
 	int error;
 
 	hold_lkb(lkb); /* reference for the ls_orphans list */
-	mutex_lock(&ls->ls_orphans_mutex);
+	spin_lock_bh(&ls->ls_orphans_lock);
 	list_add_tail(&lkb->lkb_ownqueue, &ls->ls_orphans);
-	mutex_unlock(&ls->ls_orphans_mutex);
+	spin_unlock_bh(&ls->ls_orphans_lock);
 
 	set_unlock_args(0, lkb->lkb_ua, &args);
 
@@ -5975,7 +6129,7 @@ static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
 {
 	struct dlm_lkb *lkb = NULL;
 
-	spin_lock(&ls->ls_clear_proc_locks);
+	spin_lock_bh(&ls->ls_clear_proc_locks);
 	if (list_empty(&proc->locks))
 		goto out;
 
@@ -5987,7 +6141,7 @@ static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
 	else
 		set_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags);
  out:
-	spin_unlock(&ls->ls_clear_proc_locks);
+	spin_unlock_bh(&ls->ls_clear_proc_locks);
 	return lkb;
 }
 
@@ -6003,6 +6157,7 @@ static struct dlm_lkb *del_proc_lock(struct dlm_ls *ls,
 
 void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
 {
+	struct dlm_callback *cb, *cb_safe;
 	struct dlm_lkb *lkb, *safe;
 
 	dlm_lock_recovery(ls);
@@ -6023,7 +6178,7 @@ void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
 		dlm_put_lkb(lkb);
 	}
 
-	spin_lock(&ls->ls_clear_proc_locks);
+	spin_lock_bh(&ls->ls_clear_proc_locks);
 
 	/* in-progress unlocks */
 	list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
@@ -6032,29 +6187,29 @@ void dlm_clear_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
 		dlm_put_lkb(lkb);
 	}
 
-	list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_cb_list) {
-		dlm_purge_lkb_callbacks(lkb);
-		list_del_init(&lkb->lkb_cb_list);
-		dlm_put_lkb(lkb);
+	list_for_each_entry_safe(cb, cb_safe, &proc->asts, list) {
+		list_del(&cb->list);
+		dlm_free_cb(cb);
 	}
 
-	spin_unlock(&ls->ls_clear_proc_locks);
+	spin_unlock_bh(&ls->ls_clear_proc_locks);
 	dlm_unlock_recovery(ls);
 }
 
 static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
 {
+	struct dlm_callback *cb, *cb_safe;
 	struct dlm_lkb *lkb, *safe;
 
 	while (1) {
 		lkb = NULL;
-		spin_lock(&proc->locks_spin);
+		spin_lock_bh(&proc->locks_spin);
 		if (!list_empty(&proc->locks)) {
 			lkb = list_entry(proc->locks.next, struct dlm_lkb,
 					 lkb_ownqueue);
 			list_del_init(&lkb->lkb_ownqueue);
 		}
-		spin_unlock(&proc->locks_spin);
+		spin_unlock_bh(&proc->locks_spin);
 
 		if (!lkb)
 			break;
@@ -6064,21 +6219,20 @@ static void purge_proc_locks(struct dlm_ls *ls, struct dlm_user_proc *proc)
 		dlm_put_lkb(lkb); /* ref from proc->locks list */
 	}
 
-	spin_lock(&proc->locks_spin);
+	spin_lock_bh(&proc->locks_spin);
 	list_for_each_entry_safe(lkb, safe, &proc->unlocking, lkb_ownqueue) {
 		list_del_init(&lkb->lkb_ownqueue);
 		set_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags);
 		dlm_put_lkb(lkb);
 	}
-	spin_unlock(&proc->locks_spin);
+	spin_unlock_bh(&proc->locks_spin);
 
-	spin_lock(&proc->asts_spin);
-	list_for_each_entry_safe(lkb, safe, &proc->asts, lkb_cb_list) {
-		dlm_purge_lkb_callbacks(lkb);
-		list_del_init(&lkb->lkb_cb_list);
-		dlm_put_lkb(lkb);
+	spin_lock_bh(&proc->asts_spin);
+	list_for_each_entry_safe(cb, cb_safe, &proc->asts, list) {
+		list_del(&cb->list);
+		dlm_free_cb(cb);
 	}
-	spin_unlock(&proc->asts_spin);
+	spin_unlock_bh(&proc->asts_spin);
 }
 
 /* pid of 0 means purge all orphans */
@@ -6087,7 +6241,7 @@ static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
 {
 	struct dlm_lkb *lkb, *safe;
 
-	mutex_lock(&ls->ls_orphans_mutex);
+	spin_lock_bh(&ls->ls_orphans_lock);
 	list_for_each_entry_safe(lkb, safe, &ls->ls_orphans, lkb_ownqueue) {
 		if (pid && lkb->lkb_ownpid != pid)
 			continue;
@@ -6095,7 +6249,7 @@ static void do_purge(struct dlm_ls *ls, int nodeid, int pid)
 		list_del_init(&lkb->lkb_ownqueue);
 		dlm_put_lkb(lkb);
 	}
-	mutex_unlock(&ls->ls_orphans_mutex);
+	spin_unlock_bh(&ls->ls_orphans_lock);
 }
 
 static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
@@ -6105,7 +6259,7 @@ static int send_purge(struct dlm_ls *ls, int nodeid, int pid)
 	int error;
 
 	error = _create_message(ls, sizeof(struct dlm_message), nodeid,
-				DLM_MSG_PURGE, &ms, &mh, GFP_NOFS);
+				DLM_MSG_PURGE, &ms, &mh);
 	if (error)
 		return error;
 	ms->m_nodeid = cpu_to_le32(nodeid);

fs/dlm/lock.h

@@ -11,6 +11,7 @@
 #ifndef __LOCK_DOT_H__
 #define __LOCK_DOT_H__
 
+void dlm_rsb_toss_timer(struct timer_list *timer);
 void dlm_dump_rsb(struct dlm_rsb *r);
 void dlm_dump_rsb_name(struct dlm_ls *ls, const char *name, int len);
 void dlm_print_lkb(struct dlm_lkb *lkb);
@@ -18,20 +19,23 @@ void dlm_receive_message_saved(struct dlm_ls *ls, const struct dlm_message *ms,
 			       uint32_t saved_seq);
 void dlm_receive_buffer(const union dlm_packet *p, int nodeid);
 int dlm_modes_compat(int mode1, int mode2);
+void free_toss_rsb(struct dlm_rsb *r);
 void dlm_put_rsb(struct dlm_rsb *r);
 void dlm_hold_rsb(struct dlm_rsb *r);
 int dlm_put_lkb(struct dlm_lkb *lkb);
 void dlm_scan_rsbs(struct dlm_ls *ls);
 int dlm_lock_recovery_try(struct dlm_ls *ls);
+void dlm_lock_recovery(struct dlm_ls *ls);
 void dlm_unlock_recovery(struct dlm_ls *ls);
+void dlm_timer_resume(struct dlm_ls *ls);
 
 int dlm_master_lookup(struct dlm_ls *ls, int from_nodeid, const char *name,
 		      int len, unsigned int flags, int *r_nodeid, int *result);
 
-int dlm_search_rsb_tree(struct rb_root *tree, const void *name, int len,
+int dlm_search_rsb_tree(struct rhashtable *rhash, const void *name, int len,
 			struct dlm_rsb **r_ret);
 
-void dlm_recover_purge(struct dlm_ls *ls);
+void dlm_recover_purge(struct dlm_ls *ls, const struct list_head *root_list);
 void dlm_purge_mstcpy_locks(struct dlm_rsb *r);
 void dlm_recover_grant(struct dlm_ls *ls);
 int dlm_recover_waiters_post(struct dlm_ls *ls);
@@ -68,12 +72,12 @@ static inline int is_master(struct dlm_rsb *r)
 
 static inline void lock_rsb(struct dlm_rsb *r)
 {
-	mutex_lock(&r->res_mutex);
+	spin_lock_bh(&r->res_lock);
 }
 
 static inline void unlock_rsb(struct dlm_rsb *r)
 {
-	mutex_unlock(&r->res_mutex);
+	spin_unlock_bh(&r->res_lock);
 }
 
 #endif

fs/dlm/lockspace.c

@@ -29,8 +29,6 @@ static int			ls_count;
 static struct mutex		ls_lock;
 static struct list_head		lslist;
 static spinlock_t		lslist_lock;
-static struct task_struct *	scand_task;
-
 
 static ssize_t dlm_control_store(struct dlm_ls *ls, const char *buf, size_t len)
 {
@@ -247,66 +245,11 @@ void dlm_lockspace_exit(void)
 	kset_unregister(dlm_kset);
 }
 
-static struct dlm_ls *find_ls_to_scan(void)
-{
-	struct dlm_ls *ls;
-
-	spin_lock(&lslist_lock);
-	list_for_each_entry(ls, &lslist, ls_list) {
-		if (time_after_eq(jiffies, ls->ls_scan_time +
-					    dlm_config.ci_scan_secs * HZ)) {
-			spin_unlock(&lslist_lock);
-			return ls;
-		}
-	}
-	spin_unlock(&lslist_lock);
-	return NULL;
-}
-
-static int dlm_scand(void *data)
-{
-	struct dlm_ls *ls;
-
-	while (!kthread_should_stop()) {
-		ls = find_ls_to_scan();
-		if (ls) {
-			if (dlm_lock_recovery_try(ls)) {
-				ls->ls_scan_time = jiffies;
-				dlm_scan_rsbs(ls);
-				dlm_unlock_recovery(ls);
-			} else {
-				ls->ls_scan_time += HZ;
-			}
-			continue;
-		}
-		schedule_timeout_interruptible(dlm_config.ci_scan_secs * HZ);
-	}
-	return 0;
-}
-
-static int dlm_scand_start(void)
-{
-	struct task_struct *p;
-	int error = 0;
-
-	p = kthread_run(dlm_scand, NULL, "dlm_scand");
-	if (IS_ERR(p))
-		error = PTR_ERR(p);
-	else
-		scand_task = p;
-	return error;
-}
-
-static void dlm_scand_stop(void)
-{
-	kthread_stop(scand_task);
-}
-
 struct dlm_ls *dlm_find_lockspace_global(uint32_t id)
 {
 	struct dlm_ls *ls;
 
-	spin_lock(&lslist_lock);
+	spin_lock_bh(&lslist_lock);
 
 	list_for_each_entry(ls, &lslist, ls_list) {
 		if (ls->ls_global_id == id) {
@@ -316,7 +259,7 @@ struct dlm_ls *dlm_find_lockspace_global(uint32_t id)
 	}
 	ls = NULL;
  out:
-	spin_unlock(&lslist_lock);
+	spin_unlock_bh(&lslist_lock);
 	return ls;
 }
 
@@ -324,7 +267,7 @@ struct dlm_ls *dlm_find_lockspace_local(dlm_lockspace_t *lockspace)
 {
 	struct dlm_ls *ls;
 
-	spin_lock(&lslist_lock);
+	spin_lock_bh(&lslist_lock);
 	list_for_each_entry(ls, &lslist, ls_list) {
 		if (ls->ls_local_handle == lockspace) {
 			atomic_inc(&ls->ls_count);
@@ -333,7 +276,7 @@ struct dlm_ls *dlm_find_lockspace_local(dlm_lockspace_t *lockspace)
 	}
 	ls = NULL;
  out:
-	spin_unlock(&lslist_lock);
+	spin_unlock_bh(&lslist_lock);
 	return ls;
 }
 
@@ -341,7 +284,7 @@ struct dlm_ls *dlm_find_lockspace_device(int minor)
 {
 	struct dlm_ls *ls;
 
-	spin_lock(&lslist_lock);
+	spin_lock_bh(&lslist_lock);
 	list_for_each_entry(ls, &lslist, ls_list) {
 		if (ls->ls_device.minor == minor) {
 			atomic_inc(&ls->ls_count);
@@ -350,7 +293,7 @@ struct dlm_ls *dlm_find_lockspace_device(int minor)
 	}
 	ls = NULL;
  out:
-	spin_unlock(&lslist_lock);
+	spin_unlock_bh(&lslist_lock);
 	return ls;
 }
 
@@ -365,15 +308,15 @@ static void remove_lockspace(struct dlm_ls *ls)
 retry:
 	wait_event(ls->ls_count_wait, atomic_read(&ls->ls_count) == 0);
 
-	spin_lock(&lslist_lock);
+	spin_lock_bh(&lslist_lock);
 	if (atomic_read(&ls->ls_count) != 0) {
-		spin_unlock(&lslist_lock);
+		spin_unlock_bh(&lslist_lock);
 		goto retry;
 	}
 
 	WARN_ON(ls->ls_create_count != 0);
 	list_del(&ls->ls_list);
-	spin_unlock(&lslist_lock);
+	spin_unlock_bh(&lslist_lock);
 }
 
 static int threads_start(void)
@@ -382,22 +325,9 @@ static int threads_start(void)
 
 	/* Thread for sending/receiving messages for all lockspace's */
 	error = dlm_midcomms_start();
-	if (error) {
+	if (error)
 		log_print("cannot start dlm midcomms %d", error);
-		goto fail;
-	}
-
-	error = dlm_scand_start();
-	if (error) {
-		log_print("cannot start dlm_scand thread %d", error);
-		goto midcomms_fail;
-	}
 
-	return 0;
-
- midcomms_fail:
-	dlm_midcomms_stop();
- fail:
 	return error;
 }
 
@@ -407,9 +337,9 @@ static int new_lockspace(const char *name, const char *cluster,
 			 int *ops_result, dlm_lockspace_t **lockspace)
 {
 	struct dlm_ls *ls;
-	int i, size, error;
 	int do_unreg = 0;
 	int namelen = strlen(name);
+	int error;
 
 	if (namelen > DLM_LOCKSPACE_LEN || namelen == 0)
 		return -EINVAL;
@@ -448,7 +378,7 @@ static int new_lockspace(const char *name, const char *cluster,
 
 	error = 0;
 
-	spin_lock(&lslist_lock);
+	spin_lock_bh(&lslist_lock);
 	list_for_each_entry(ls, &lslist, ls_list) {
 		WARN_ON(ls->ls_create_count <= 0);
 		if (ls->ls_namelen != namelen)
@@ -464,7 +394,7 @@ static int new_lockspace(const char *name, const char *cluster,
 		error = 1;
 		break;
 	}
-	spin_unlock(&lslist_lock);
+	spin_unlock_bh(&lslist_lock);
 
 	if (error)
 		goto out;
@@ -492,32 +422,21 @@ static int new_lockspace(const char *name, const char *cluster,
 	 */
 	ls->ls_exflags = (flags & ~(DLM_LSFL_FS | DLM_LSFL_NEWEXCL));
 
-	size = READ_ONCE(dlm_config.ci_rsbtbl_size);
-	ls->ls_rsbtbl_size = size;
+	INIT_LIST_HEAD(&ls->ls_toss);
+	INIT_LIST_HEAD(&ls->ls_keep);
+	rwlock_init(&ls->ls_rsbtbl_lock);
 
-	ls->ls_rsbtbl = vmalloc(array_size(size, sizeof(struct dlm_rsbtable)));
-	if (!ls->ls_rsbtbl)
+	error = rhashtable_init(&ls->ls_rsbtbl, &dlm_rhash_rsb_params);
+	if (error)
 		goto out_lsfree;
-	for (i = 0; i < size; i++) {
-		ls->ls_rsbtbl[i].keep.rb_node = NULL;
-		ls->ls_rsbtbl[i].toss.rb_node = NULL;
-		spin_lock_init(&ls->ls_rsbtbl[i].lock);
-	}
-
-	for (i = 0; i < DLM_REMOVE_NAMES_MAX; i++) {
-		ls->ls_remove_names[i] = kzalloc(DLM_RESNAME_MAXLEN+1,
-						 GFP_KERNEL);
-		if (!ls->ls_remove_names[i])
-			goto out_rsbtbl;
-	}
 
 	idr_init(&ls->ls_lkbidr);
-	spin_lock_init(&ls->ls_lkbidr_spin);
+	rwlock_init(&ls->ls_lkbidr_lock);
 
 	INIT_LIST_HEAD(&ls->ls_waiters);
-	mutex_init(&ls->ls_waiters_mutex);
+	spin_lock_init(&ls->ls_waiters_lock);
 	INIT_LIST_HEAD(&ls->ls_orphans);
-	mutex_init(&ls->ls_orphans_mutex);
+	spin_lock_init(&ls->ls_orphans_lock);
 
 	INIT_LIST_HEAD(&ls->ls_new_rsb);
 	spin_lock_init(&ls->ls_new_rsb_spin);
@@ -552,11 +471,9 @@ static int new_lockspace(const char *name, const char *cluster,
 	ls->ls_recover_seq = get_random_u64();
 	ls->ls_recover_args = NULL;
 	init_rwsem(&ls->ls_in_recovery);
-	init_rwsem(&ls->ls_recv_active);
+	rwlock_init(&ls->ls_recv_active);
 	INIT_LIST_HEAD(&ls->ls_requestqueue);
-	atomic_set(&ls->ls_requestqueue_cnt, 0);
-	init_waitqueue_head(&ls->ls_requestqueue_wait);
-	mutex_init(&ls->ls_requestqueue_mutex);
+	rwlock_init(&ls->ls_requestqueue_lock);
 	spin_lock_init(&ls->ls_clear_proc_locks);
 
 	/* Due backwards compatibility with 3.1 we need to use maximum
@@ -565,8 +482,10 @@ static int new_lockspace(const char *name, const char *cluster,
 	 * might send less.
 	 */
 	ls->ls_recover_buf = kmalloc(DLM_MAX_SOCKET_BUFSIZE, GFP_NOFS);
-	if (!ls->ls_recover_buf)
+	if (!ls->ls_recover_buf) {
+		error = -ENOMEM;
 		goto out_lkbidr;
+	}
 
 	ls->ls_slot = 0;
 	ls->ls_num_slots = 0;
@@ -580,13 +499,20 @@ static int new_lockspace(const char *name, const char *cluster,
 	ls->ls_recover_list_count = 0;
 	ls->ls_local_handle = ls;
 	init_waitqueue_head(&ls->ls_wait_general);
-	INIT_LIST_HEAD(&ls->ls_root_list);
-	init_rwsem(&ls->ls_root_sem);
+	INIT_LIST_HEAD(&ls->ls_masters_list);
+	rwlock_init(&ls->ls_masters_lock);
+	INIT_LIST_HEAD(&ls->ls_dir_dump_list);
+	rwlock_init(&ls->ls_dir_dump_lock);
 
-	spin_lock(&lslist_lock);
+	INIT_LIST_HEAD(&ls->ls_toss_q);
+	spin_lock_init(&ls->ls_toss_q_lock);
+	timer_setup(&ls->ls_timer, dlm_rsb_toss_timer,
+		    TIMER_DEFERRABLE);
+
+	spin_lock_bh(&lslist_lock);
 	ls->ls_create_count = 1;
 	list_add(&ls->ls_list, &lslist);
-	spin_unlock(&lslist_lock);
+	spin_unlock_bh(&lslist_lock);
 
 	if (flags & DLM_LSFL_FS) {
 		error = dlm_callback_start(ls);
@@ -655,17 +581,14 @@ static int new_lockspace(const char *name, const char *cluster,
  out_callback:
 	dlm_callback_stop(ls);
  out_delist:
-	spin_lock(&lslist_lock);
+	spin_lock_bh(&lslist_lock);
 	list_del(&ls->ls_list);
-	spin_unlock(&lslist_lock);
+	spin_unlock_bh(&lslist_lock);
 	idr_destroy(&ls->ls_recover_idr);
 	kfree(ls->ls_recover_buf);
  out_lkbidr:
 	idr_destroy(&ls->ls_lkbidr);
- out_rsbtbl:
-	for (i = 0; i < DLM_REMOVE_NAMES_MAX; i++)
-		kfree(ls->ls_remove_names[i]);
-	vfree(ls->ls_rsbtbl);
+	rhashtable_destroy(&ls->ls_rsbtbl);
  out_lsfree:
 	if (do_unreg)
 		kobject_put(&ls->ls_kobj);
@@ -697,7 +620,6 @@ static int __dlm_new_lockspace(const char *name, const char *cluster,
 	if (error > 0)
 		error = 0;
 	if (!ls_count) {
-		dlm_scand_stop();
 		dlm_midcomms_shutdown();
 		dlm_midcomms_stop();
 	}
@@ -756,7 +678,7 @@ static int lockspace_busy(struct dlm_ls *ls, int force)
 {
 	int rv;
 
-	spin_lock(&ls->ls_lkbidr_spin);
+	read_lock_bh(&ls->ls_lkbidr_lock);
 	if (force == 0) {
 		rv = idr_for_each(&ls->ls_lkbidr, lkb_idr_is_any, ls);
 	} else if (force == 1) {
@@ -764,19 +686,25 @@ static int lockspace_busy(struct dlm_ls *ls, int force)
 	} else {
 		rv = 0;
 	}
-	spin_unlock(&ls->ls_lkbidr_spin);
+	read_unlock_bh(&ls->ls_lkbidr_lock);
 	return rv;
 }
 
+static void rhash_free_rsb(void *ptr, void *arg)
+{
+	struct dlm_rsb *rsb = ptr;
+
+	dlm_free_rsb(rsb);
+}
+
 static int release_lockspace(struct dlm_ls *ls, int force)
 {
 	struct dlm_rsb *rsb;
-	struct rb_node *n;
-	int i, busy, rv;
+	int busy, rv;
 
 	busy = lockspace_busy(ls, force);
 
-	spin_lock(&lslist_lock);
+	spin_lock_bh(&lslist_lock);
 	if (ls->ls_create_count == 1) {
 		if (busy) {
 			rv = -EBUSY;
@@ -790,7 +718,7 @@ static int release_lockspace(struct dlm_ls *ls, int force)
 	} else {
 		rv = -EINVAL;
 	}
-	spin_unlock(&lslist_lock);
+	spin_unlock_bh(&lslist_lock);
 
 	if (rv) {
 		log_debug(ls, "release_lockspace no remove %d", rv);
@@ -807,8 +735,13 @@ static int release_lockspace(struct dlm_ls *ls, int force)
 
 	dlm_recoverd_stop(ls);
 
+	/* clear the LSFL_RUNNING flag to fast up
+	 * time_shutdown_sync(), we don't care anymore
+	 */
+	clear_bit(LSFL_RUNNING, &ls->ls_flags);
+	timer_shutdown_sync(&ls->ls_timer);
+
 	if (ls_count == 1) {
-		dlm_scand_stop();
 		dlm_clear_members(ls);
 		dlm_midcomms_shutdown();
 	}
@@ -830,27 +763,9 @@ static int release_lockspace(struct dlm_ls *ls, int force)
 	idr_destroy(&ls->ls_lkbidr);
 
 	/*
-	 * Free all rsb's on rsbtbl[] lists
+	 * Free all rsb's on rsbtbl
 	 */
-
-	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
-		while ((n = rb_first(&ls->ls_rsbtbl[i].keep))) {
-			rsb = rb_entry(n, struct dlm_rsb, res_hashnode);
-			rb_erase(n, &ls->ls_rsbtbl[i].keep);
-			dlm_free_rsb(rsb);
-		}
-
-		while ((n = rb_first(&ls->ls_rsbtbl[i].toss))) {
-			rsb = rb_entry(n, struct dlm_rsb, res_hashnode);
-			rb_erase(n, &ls->ls_rsbtbl[i].toss);
-			dlm_free_rsb(rsb);
-		}
-	}
-
-	vfree(ls->ls_rsbtbl);
-
-	for (i = 0; i < DLM_REMOVE_NAMES_MAX; i++)
-		kfree(ls->ls_remove_names[i]);
+	rhashtable_free_and_destroy(&ls->ls_rsbtbl, rhash_free_rsb, NULL);
 
 	while (!list_empty(&ls->ls_new_rsb)) {
 		rsb = list_first_entry(&ls->ls_new_rsb, struct dlm_rsb,
@@ -918,20 +833,19 @@ void dlm_stop_lockspaces(void)
 
  restart:
 	count = 0;
-	spin_lock(&lslist_lock);
+	spin_lock_bh(&lslist_lock);
 	list_for_each_entry(ls, &lslist, ls_list) {
 		if (!test_bit(LSFL_RUNNING, &ls->ls_flags)) {
 			count++;
 			continue;
 		}
-		spin_unlock(&lslist_lock);
+		spin_unlock_bh(&lslist_lock);
 		log_error(ls, "no userland control daemon, stopping lockspace");
 		dlm_ls_stop(ls);
 		goto restart;
 	}
-	spin_unlock(&lslist_lock);
+	spin_unlock_bh(&lslist_lock);
 
 	if (count)
 		log_print("dlm user daemon left %d lockspaces", count);
 }
-

fs/dlm/lowcomms.c

@@ -204,6 +204,7 @@ static void process_dlm_messages(struct work_struct *work);
 static DECLARE_WORK(process_work, process_dlm_messages);
 static DEFINE_SPINLOCK(processqueue_lock);
 static bool process_dlm_messages_pending;
+static DECLARE_WAIT_QUEUE_HEAD(processqueue_wq);
 static atomic_t processqueue_count;
 static LIST_HEAD(processqueue);
 
@@ -248,7 +249,7 @@ struct kmem_cache *dlm_lowcomms_writequeue_cache_create(void)
 
 struct kmem_cache *dlm_lowcomms_msg_cache_create(void)
 {
-	return kmem_cache_create("dlm_msg", sizeof(struct dlm_msg), 0, 0, NULL);
+	return KMEM_CACHE(dlm_msg, 0);
 }
 
 /* need to held writequeue_lock */
@@ -867,36 +868,38 @@ static void process_dlm_messages(struct work_struct *work)
 {
 	struct processqueue_entry *pentry;
 
-	spin_lock(&processqueue_lock);
+	spin_lock_bh(&processqueue_lock);
 	pentry = list_first_entry_or_null(&processqueue,
 					  struct processqueue_entry, list);
 	if (WARN_ON_ONCE(!pentry)) {
 		process_dlm_messages_pending = false;
-		spin_unlock(&processqueue_lock);
+		spin_unlock_bh(&processqueue_lock);
 		return;
 	}
 
 	list_del(&pentry->list);
-	atomic_dec(&processqueue_count);
-	spin_unlock(&processqueue_lock);
+	if (atomic_dec_and_test(&processqueue_count))
+		wake_up(&processqueue_wq);
+	spin_unlock_bh(&processqueue_lock);
 
 	for (;;) {
 		dlm_process_incoming_buffer(pentry->nodeid, pentry->buf,
 					    pentry->buflen);
 		free_processqueue_entry(pentry);
 
-		spin_lock(&processqueue_lock);
+		spin_lock_bh(&processqueue_lock);
 		pentry = list_first_entry_or_null(&processqueue,
 						  struct processqueue_entry, list);
 		if (!pentry) {
 			process_dlm_messages_pending = false;
-			spin_unlock(&processqueue_lock);
+			spin_unlock_bh(&processqueue_lock);
 			break;
 		}
 
 		list_del(&pentry->list);
-		atomic_dec(&processqueue_count);
-		spin_unlock(&processqueue_lock);
+		if (atomic_dec_and_test(&processqueue_count))
+			wake_up(&processqueue_wq);
+		spin_unlock_bh(&processqueue_lock);
 	}
 }
 
@@ -966,14 +969,14 @@ static int receive_from_sock(struct connection *con, int buflen)
 	memmove(con->rx_leftover_buf, pentry->buf + ret,
 		con->rx_leftover);
 
-	spin_lock(&processqueue_lock);
+	spin_lock_bh(&processqueue_lock);
 	ret = atomic_inc_return(&processqueue_count);
 	list_add_tail(&pentry->list, &processqueue);
 	if (!process_dlm_messages_pending) {
 		process_dlm_messages_pending = true;
 		queue_work(process_workqueue, &process_work);
 	}
-	spin_unlock(&processqueue_lock);
+	spin_unlock_bh(&processqueue_lock);
 
 	if (ret > DLM_MAX_PROCESS_BUFFERS)
 		return DLM_IO_FLUSH;
@@ -1229,14 +1232,13 @@ static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
 };
 
 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
-						gfp_t allocation, char **ppc,
-						void (*cb)(void *data),
+						char **ppc, void (*cb)(void *data),
 						void *data)
 {
 	struct writequeue_entry *e;
 	struct dlm_msg *msg;
 
-	msg = dlm_allocate_msg(allocation);
+	msg = dlm_allocate_msg();
 	if (!msg)
 		return NULL;
 
@@ -1261,9 +1263,8 @@ static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
  * dlm_lowcomms_commit_msg which is a must call if success
  */
 #ifndef __CHECKER__
-struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
-				     char **ppc, void (*cb)(void *data),
-				     void *data)
+struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, char **ppc,
+				     void (*cb)(void *data), void *data)
 {
 	struct connection *con;
 	struct dlm_msg *msg;
@@ -1284,7 +1285,7 @@ struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
 		return NULL;
 	}
 
-	msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, data);
+	msg = dlm_lowcomms_new_msg_con(con, len, ppc, cb, data);
 	if (!msg) {
 		srcu_read_unlock(&connections_srcu, idx);
 		return NULL;
@@ -1348,8 +1349,8 @@ int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
 	if (msg->retransmit)
 		return 1;
 
-	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len,
-					      GFP_ATOMIC, &ppc, NULL, NULL);
+	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len, &ppc,
+					      NULL, NULL);
 	if (!msg_resend)
 		return -ENOMEM;
 
@@ -1513,7 +1514,20 @@ static void process_recv_sockets(struct work_struct *work)
 		/* CF_RECV_PENDING cleared */
 		break;
 	case DLM_IO_FLUSH:
-		flush_workqueue(process_workqueue);
+		/* we can't flush the process_workqueue here because a
+		 * WQ_MEM_RECLAIM workequeue can occurr a deadlock for a non
+		 * WQ_MEM_RECLAIM workqueue such as process_workqueue. Instead
+		 * we have a waitqueue to wait until all messages are
+		 * processed.
+		 *
+		 * This handling is only necessary to backoff the sender and
+		 * not queue all messages from the socket layer into DLM
+		 * processqueue. When DLM is capable to parse multiple messages
+		 * on an e.g. per socket basis this handling can might be
+		 * removed. Especially in a message burst we are too slow to
+		 * process messages and the queue will fill up memory.
+		 */
+		wait_event(processqueue_wq, !atomic_read(&processqueue_count));
 		fallthrough;
 	case DLM_IO_RESCHED:
 		cond_resched();
@@ -1703,11 +1717,7 @@ static int work_start(void)
 		return -ENOMEM;
 	}
 
-	/* ordered dlm message process queue,
-	 * should be converted to a tasklet
-	 */
-	process_workqueue = alloc_ordered_workqueue("dlm_process",
-						    WQ_HIGHPRI | WQ_MEM_RECLAIM);
+	process_workqueue = alloc_workqueue("dlm_process", WQ_HIGHPRI | WQ_BH, 0);
 	if (!process_workqueue) {
 		log_print("can't start dlm_process");
 		destroy_workqueue(io_workqueue);

fs/dlm/lowcomms.h

@@ -39,9 +39,8 @@ void dlm_lowcomms_stop(void);
 void dlm_lowcomms_init(void);
 void dlm_lowcomms_exit(void);
 int dlm_lowcomms_close(int nodeid);
-struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
-				     char **ppc, void (*cb)(void *data),
-				     void *data);
+struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, char **ppc,
+				     void (*cb)(void *data), void *data);
 void dlm_lowcomms_commit_msg(struct dlm_msg *msg);
 void dlm_lowcomms_put_msg(struct dlm_msg *msg);
 int dlm_lowcomms_resend_msg(struct dlm_msg *msg);

fs/dlm/member.c

@@ -630,7 +630,7 @@ int dlm_ls_stop(struct dlm_ls *ls)
 	 * message to the requestqueue without races.
 	 */
 
-	down_write(&ls->ls_recv_active);
+	write_lock_bh(&ls->ls_recv_active);
 
 	/*
 	 * Abort any recovery that's in progress (see RECOVER_STOP,
@@ -638,18 +638,25 @@ int dlm_ls_stop(struct dlm_ls *ls)
 	 * dlm to quit any processing (see RUNNING, dlm_locking_stopped()).
 	 */
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	set_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
 	new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags);
+	if (new)
+		timer_delete_sync(&ls->ls_timer);
 	ls->ls_recover_seq++;
-	spin_unlock(&ls->ls_recover_lock);
+
+	/* activate requestqueue and stop processing */
+	write_lock_bh(&ls->ls_requestqueue_lock);
+	set_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags);
+	write_unlock_bh(&ls->ls_requestqueue_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
 	/*
 	 * Let dlm_recv run again, now any normal messages will be saved on the
 	 * requestqueue for later.
 	 */
 
-	up_write(&ls->ls_recv_active);
+	write_unlock_bh(&ls->ls_recv_active);
 
 	/*
 	 * This in_recovery lock does two things:
@@ -674,13 +681,13 @@ int dlm_ls_stop(struct dlm_ls *ls)
 
 	dlm_recoverd_suspend(ls);
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	kfree(ls->ls_slots);
 	ls->ls_slots = NULL;
 	ls->ls_num_slots = 0;
 	ls->ls_slots_size = 0;
 	ls->ls_recover_status = 0;
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
 	dlm_recoverd_resume(ls);
 
@@ -714,12 +721,12 @@ int dlm_ls_start(struct dlm_ls *ls)
 	if (error < 0)
 		goto fail_rv;
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 
 	/* the lockspace needs to be stopped before it can be started */
 
 	if (!dlm_locking_stopped(ls)) {
-		spin_unlock(&ls->ls_recover_lock);
+		spin_unlock_bh(&ls->ls_recover_lock);
 		log_error(ls, "start ignored: lockspace running");
 		error = -EINVAL;
 		goto fail;
@@ -730,7 +737,7 @@ int dlm_ls_start(struct dlm_ls *ls)
 	rv->seq = ++ls->ls_recover_seq;
 	rv_old = ls->ls_recover_args;
 	ls->ls_recover_args = rv;
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
 	if (rv_old) {
 		log_error(ls, "unused recovery %llx %d",

fs/dlm/memory.c

@@ -84,7 +84,7 @@ char *dlm_allocate_lvb(struct dlm_ls *ls)
 {
 	char *p;
 
-	p = kzalloc(ls->ls_lvblen, GFP_NOFS);
+	p = kzalloc(ls->ls_lvblen, GFP_ATOMIC);
 	return p;
 }
 
@@ -97,7 +97,7 @@ struct dlm_rsb *dlm_allocate_rsb(struct dlm_ls *ls)
 {
 	struct dlm_rsb *r;
 
-	r = kmem_cache_zalloc(rsb_cache, GFP_NOFS);
+	r = kmem_cache_zalloc(rsb_cache, GFP_ATOMIC);
 	return r;
 }
 
@@ -112,7 +112,7 @@ struct dlm_lkb *dlm_allocate_lkb(struct dlm_ls *ls)
 {
 	struct dlm_lkb *lkb;
 
-	lkb = kmem_cache_zalloc(lkb_cache, GFP_NOFS);
+	lkb = kmem_cache_zalloc(lkb_cache, GFP_ATOMIC);
 	return lkb;
 }
 
@@ -127,16 +127,12 @@ void dlm_free_lkb(struct dlm_lkb *lkb)
 		}
 	}
 
-	/* drop references if they are set */
-	dlm_callback_set_last_ptr(&lkb->lkb_last_cast, NULL);
-	dlm_callback_set_last_ptr(&lkb->lkb_last_cb, NULL);
-
 	kmem_cache_free(lkb_cache, lkb);
 }
 
-struct dlm_mhandle *dlm_allocate_mhandle(gfp_t allocation)
+struct dlm_mhandle *dlm_allocate_mhandle(void)
 {
-	return kmem_cache_alloc(mhandle_cache, allocation);
+	return kmem_cache_alloc(mhandle_cache, GFP_ATOMIC);
 }
 
 void dlm_free_mhandle(struct dlm_mhandle *mhandle)
@@ -154,9 +150,9 @@ void dlm_free_writequeue(struct writequeue_entry *writequeue)
 	kmem_cache_free(writequeue_cache, writequeue);
 }
 
-struct dlm_msg *dlm_allocate_msg(gfp_t allocation)
+struct dlm_msg *dlm_allocate_msg(void)
 {
-	return kmem_cache_alloc(msg_cache, allocation);
+	return kmem_cache_alloc(msg_cache, GFP_ATOMIC);
 }
 
 void dlm_free_msg(struct dlm_msg *msg)

fs/dlm/memory.h

@@ -20,11 +20,11 @@ struct dlm_lkb *dlm_allocate_lkb(struct dlm_ls *ls);
 void dlm_free_lkb(struct dlm_lkb *l);
 char *dlm_allocate_lvb(struct dlm_ls *ls);
 void dlm_free_lvb(char *l);
-struct dlm_mhandle *dlm_allocate_mhandle(gfp_t allocation);
+struct dlm_mhandle *dlm_allocate_mhandle(void);
 void dlm_free_mhandle(struct dlm_mhandle *mhandle);
 struct writequeue_entry *dlm_allocate_writequeue(void);
 void dlm_free_writequeue(struct writequeue_entry *writequeue);
-struct dlm_msg *dlm_allocate_msg(gfp_t allocation);
+struct dlm_msg *dlm_allocate_msg(void);
 void dlm_free_msg(struct dlm_msg *msg);
 struct dlm_callback *dlm_allocate_cb(void);
 void dlm_free_cb(struct dlm_callback *cb);

fs/dlm/midcomms.c

@@ -226,8 +226,7 @@ static DEFINE_MUTEX(close_lock);
 
 struct kmem_cache *dlm_midcomms_cache_create(void)
 {
-	return kmem_cache_create("dlm_mhandle", sizeof(struct dlm_mhandle),
-				 0, 0, NULL);
+	return KMEM_CACHE(dlm_mhandle, 0);
 }
 
 static inline const char *dlm_state_str(int state)
@@ -365,9 +364,9 @@ int dlm_midcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
 	node->users = 0;
 	midcomms_node_reset(node);
 
-	spin_lock(&nodes_lock);
+	spin_lock_bh(&nodes_lock);
 	hlist_add_head_rcu(&node->hlist, &node_hash[r]);
-	spin_unlock(&nodes_lock);
+	spin_unlock_bh(&nodes_lock);
 
 	node->debugfs = dlm_create_debug_comms_file(nodeid, node);
 	return 0;
@@ -380,8 +379,7 @@ static int dlm_send_ack(int nodeid, uint32_t seq)
 	struct dlm_msg *msg;
 	char *ppc;
 
-	msg = dlm_lowcomms_new_msg(nodeid, mb_len, GFP_ATOMIC, &ppc,
-				   NULL, NULL);
+	msg = dlm_lowcomms_new_msg(nodeid, mb_len, &ppc, NULL, NULL);
 	if (!msg)
 		return -ENOMEM;
 
@@ -429,7 +427,7 @@ static int dlm_send_fin(struct midcomms_node *node,
 	struct dlm_mhandle *mh;
 	char *ppc;
 
-	mh = dlm_midcomms_get_mhandle(node->nodeid, mb_len, GFP_ATOMIC, &ppc);
+	mh = dlm_midcomms_get_mhandle(node->nodeid, mb_len, &ppc);
 	if (!mh)
 		return -ENOMEM;
 
@@ -479,7 +477,7 @@ static void dlm_receive_ack(struct midcomms_node *node, uint32_t seq)
 
 static void dlm_pas_fin_ack_rcv(struct midcomms_node *node)
 {
-	spin_lock(&node->state_lock);
+	spin_lock_bh(&node->state_lock);
 	pr_debug("receive passive fin ack from node %d with state %s\n",
 		 node->nodeid, dlm_state_str(node->state));
 
@@ -493,13 +491,13 @@ static void dlm_pas_fin_ack_rcv(struct midcomms_node *node)
 		wake_up(&node->shutdown_wait);
 		break;
 	default:
-		spin_unlock(&node->state_lock);
+		spin_unlock_bh(&node->state_lock);
 		log_print("%s: unexpected state: %d",
 			  __func__, node->state);
 		WARN_ON_ONCE(1);
 		return;
 	}
-	spin_unlock(&node->state_lock);
+	spin_unlock_bh(&node->state_lock);
 }
 
 static void dlm_receive_buffer_3_2_trace(uint32_t seq,
@@ -536,7 +534,7 @@ static void dlm_midcomms_receive_buffer(const union dlm_packet *p,
 	if (is_expected_seq) {
 		switch (p->header.h_cmd) {
 		case DLM_FIN:
-			spin_lock(&node->state_lock);
+			spin_lock_bh(&node->state_lock);
 			pr_debug("receive fin msg from node %d with state %s\n",
 				 node->nodeid, dlm_state_str(node->state));
 
@@ -577,13 +575,13 @@ static void dlm_midcomms_receive_buffer(const union dlm_packet *p,
 				/* probably remove_member caught it, do nothing */
 				break;
 			default:
-				spin_unlock(&node->state_lock);
+				spin_unlock_bh(&node->state_lock);
 				log_print("%s: unexpected state: %d",
 					  __func__, node->state);
 				WARN_ON_ONCE(1);
 				return;
 			}
-			spin_unlock(&node->state_lock);
+			spin_unlock_bh(&node->state_lock);
 			break;
 		default:
 			WARN_ON_ONCE(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
@@ -977,13 +975,13 @@ static void midcomms_new_msg_cb(void *data)
 }
 
 static struct dlm_msg *dlm_midcomms_get_msg_3_2(struct dlm_mhandle *mh, int nodeid,
-						int len, gfp_t allocation, char **ppc)
+						int len, char **ppc)
 {
 	struct dlm_opts *opts;
 	struct dlm_msg *msg;
 
 	msg = dlm_lowcomms_new_msg(nodeid, len + DLM_MIDCOMMS_OPT_LEN,
-				   allocation, ppc, midcomms_new_msg_cb, mh);
+				   ppc, midcomms_new_msg_cb, mh);
 	if (!msg)
 		return NULL;
 
@@ -1002,8 +1000,7 @@ static struct dlm_msg *dlm_midcomms_get_msg_3_2(struct dlm_mhandle *mh, int node
  * dlm_midcomms_commit_mhandle which is a must call if success
  */
 #ifndef __CHECKER__
-struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
-					     gfp_t allocation, char **ppc)
+struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len, char **ppc)
 {
 	struct midcomms_node *node;
 	struct dlm_mhandle *mh;
@@ -1018,7 +1015,7 @@ struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
 	/* this is a bug, however we going on and hope it will be resolved */
 	WARN_ON_ONCE(test_bit(DLM_NODE_FLAG_STOP_TX, &node->flags));
 
-	mh = dlm_allocate_mhandle(allocation);
+	mh = dlm_allocate_mhandle();
 	if (!mh)
 		goto err;
 
@@ -1029,8 +1026,7 @@ struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
 
 	switch (node->version) {
 	case DLM_VERSION_3_1:
-		msg = dlm_lowcomms_new_msg(nodeid, len, allocation, ppc,
-					   NULL, NULL);
+		msg = dlm_lowcomms_new_msg(nodeid, len, ppc, NULL, NULL);
 		if (!msg) {
 			dlm_free_mhandle(mh);
 			goto err;
@@ -1041,8 +1037,7 @@ struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
 		/* send ack back if necessary */
 		dlm_send_ack_threshold(node, DLM_SEND_ACK_BACK_MSG_THRESHOLD);
 
-		msg = dlm_midcomms_get_msg_3_2(mh, nodeid, len, allocation,
-					       ppc);
+		msg = dlm_midcomms_get_msg_3_2(mh, nodeid, len, ppc);
 		if (!msg) {
 			dlm_free_mhandle(mh);
 			goto err;
@@ -1187,7 +1182,7 @@ void dlm_midcomms_exit(void)
 
 static void dlm_act_fin_ack_rcv(struct midcomms_node *node)
 {
-	spin_lock(&node->state_lock);
+	spin_lock_bh(&node->state_lock);
 	pr_debug("receive active fin ack from node %d with state %s\n",
 		 node->nodeid, dlm_state_str(node->state));
 
@@ -1207,13 +1202,13 @@ static void dlm_act_fin_ack_rcv(struct midcomms_node *node)
 		wake_up(&node->shutdown_wait);
 		break;
 	default:
-		spin_unlock(&node->state_lock);
+		spin_unlock_bh(&node->state_lock);
 		log_print("%s: unexpected state: %d",
 			  __func__, node->state);
 		WARN_ON_ONCE(1);
 		return;
 	}
-	spin_unlock(&node->state_lock);
+	spin_unlock_bh(&node->state_lock);
 }
 
 void dlm_midcomms_add_member(int nodeid)
@@ -1228,7 +1223,7 @@ void dlm_midcomms_add_member(int nodeid)
 		return;
 	}
 
-	spin_lock(&node->state_lock);
+	spin_lock_bh(&node->state_lock);
 	if (!node->users) {
 		pr_debug("receive add member from node %d with state %s\n",
 			 node->nodeid, dlm_state_str(node->state));
@@ -1256,7 +1251,7 @@ void dlm_midcomms_add_member(int nodeid)
 
 	node->users++;
 	pr_debug("node %d users inc count %d\n", nodeid, node->users);
-	spin_unlock(&node->state_lock);
+	spin_unlock_bh(&node->state_lock);
 
 	srcu_read_unlock(&nodes_srcu, idx);
 }
@@ -1274,13 +1269,13 @@ void dlm_midcomms_remove_member(int nodeid)
 		return;
 	}
 
-	spin_lock(&node->state_lock);
+	spin_lock_bh(&node->state_lock);
 	/* case of dlm_midcomms_addr() created node but
 	 * was not added before because dlm_midcomms_close()
 	 * removed the node
 	 */
 	if (!node->users) {
-		spin_unlock(&node->state_lock);
+		spin_unlock_bh(&node->state_lock);
 		srcu_read_unlock(&nodes_srcu, idx);
 		return;
 	}
@@ -1318,7 +1313,7 @@ void dlm_midcomms_remove_member(int nodeid)
 			break;
 		}
 	}
-	spin_unlock(&node->state_lock);
+	spin_unlock_bh(&node->state_lock);
 
 	srcu_read_unlock(&nodes_srcu, idx);
 }
@@ -1356,7 +1351,7 @@ static void midcomms_shutdown(struct midcomms_node *node)
 		return;
 	}
 
-	spin_lock(&node->state_lock);
+	spin_lock_bh(&node->state_lock);
 	pr_debug("receive active shutdown for node %d with state %s\n",
 		 node->nodeid, dlm_state_str(node->state));
 	switch (node->state) {
@@ -1375,7 +1370,7 @@ static void midcomms_shutdown(struct midcomms_node *node)
 		 */
 		break;
 	}
-	spin_unlock(&node->state_lock);
+	spin_unlock_bh(&node->state_lock);
 
 	if (DLM_DEBUG_FENCE_TERMINATION)
 		msleep(5000);
@@ -1446,9 +1441,9 @@ int dlm_midcomms_close(int nodeid)
 	ret = dlm_lowcomms_close(nodeid);
 	dlm_delete_debug_comms_file(node->debugfs);
 
-	spin_lock(&nodes_lock);
+	spin_lock_bh(&nodes_lock);
 	hlist_del_rcu(&node->hlist);
-	spin_unlock(&nodes_lock);
+	spin_unlock_bh(&nodes_lock);
 	srcu_read_unlock(&nodes_srcu, idx);
 
 	/* wait that all readers left until flush send queue */
@@ -1502,8 +1497,8 @@ int dlm_midcomms_rawmsg_send(struct midcomms_node *node, void *buf,
 	rd.node = node;
 	rd.buf = buf;
 
-	msg = dlm_lowcomms_new_msg(node->nodeid, buflen, GFP_NOFS,
-				   &msgbuf, midcomms_new_rawmsg_cb, &rd);
+	msg = dlm_lowcomms_new_msg(node->nodeid, buflen, &msgbuf,
+				   midcomms_new_rawmsg_cb, &rd);
 	if (!msg)
 		return -ENOMEM;
 

fs/dlm/midcomms.h

@@ -16,8 +16,7 @@ struct midcomms_node;
 
 int dlm_validate_incoming_buffer(int nodeid, unsigned char *buf, int len);
 int dlm_process_incoming_buffer(int nodeid, unsigned char *buf, int buflen);
-struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
-					     gfp_t allocation, char **ppc);
+struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len, char **ppc);
 void dlm_midcomms_commit_mhandle(struct dlm_mhandle *mh, const void *name,
 				 int namelen);
 int dlm_midcomms_addr(int nodeid, struct sockaddr_storage *addr, int len);

fs/dlm/rcom.c

@@ -55,7 +55,7 @@ static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len,
 	struct dlm_mhandle *mh;
 	char *mb;
 
-	mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, GFP_NOFS, &mb);
+	mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, &mb);
 	if (!mh) {
 		log_print("%s to %d type %d len %d ENOBUFS",
 			  __func__, to_nodeid, type, len);
@@ -75,8 +75,7 @@ static int create_rcom_stateless(struct dlm_ls *ls, int to_nodeid, int type,
 	struct dlm_msg *msg;
 	char *mb;
 
-	msg = dlm_lowcomms_new_msg(to_nodeid, mb_len, GFP_NOFS, &mb,
-				   NULL, NULL);
+	msg = dlm_lowcomms_new_msg(to_nodeid, mb_len, &mb, NULL, NULL);
 	if (!msg) {
 		log_print("create_rcom to %d type %d len %d ENOBUFS",
 			  to_nodeid, type, len);
@@ -144,18 +143,18 @@ static int check_rcom_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
 
 static void allow_sync_reply(struct dlm_ls *ls, __le64 *new_seq)
 {
-	spin_lock(&ls->ls_rcom_spin);
+	spin_lock_bh(&ls->ls_rcom_spin);
 	*new_seq = cpu_to_le64(++ls->ls_rcom_seq);
 	set_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
-	spin_unlock(&ls->ls_rcom_spin);
+	spin_unlock_bh(&ls->ls_rcom_spin);
 }
 
 static void disallow_sync_reply(struct dlm_ls *ls)
 {
-	spin_lock(&ls->ls_rcom_spin);
+	spin_lock_bh(&ls->ls_rcom_spin);
 	clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
 	clear_bit(LSFL_RCOM_READY, &ls->ls_flags);
-	spin_unlock(&ls->ls_rcom_spin);
+	spin_unlock_bh(&ls->ls_rcom_spin);
 }
 
 /*
@@ -246,10 +245,10 @@ static void receive_rcom_status(struct dlm_ls *ls,
 		goto do_create;
 	}
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	status = ls->ls_recover_status;
 	num_slots = ls->ls_num_slots;
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 	len += num_slots * sizeof(struct rcom_slot);
 
  do_create:
@@ -267,9 +266,9 @@ static void receive_rcom_status(struct dlm_ls *ls,
 	if (!num_slots)
 		goto do_send;
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	if (ls->ls_num_slots != num_slots) {
-		spin_unlock(&ls->ls_recover_lock);
+		spin_unlock_bh(&ls->ls_recover_lock);
 		log_debug(ls, "receive_rcom_status num_slots %d to %d",
 			  num_slots, ls->ls_num_slots);
 		rc->rc_result = 0;
@@ -278,7 +277,7 @@ static void receive_rcom_status(struct dlm_ls *ls,
 	}
 
 	dlm_slots_copy_out(ls, rc);
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
  do_send:
 	send_rcom_stateless(msg, rc);
@@ -286,7 +285,7 @@ static void receive_rcom_status(struct dlm_ls *ls,
 
 static void receive_sync_reply(struct dlm_ls *ls, const struct dlm_rcom *rc_in)
 {
-	spin_lock(&ls->ls_rcom_spin);
+	spin_lock_bh(&ls->ls_rcom_spin);
 	if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
 	    le64_to_cpu(rc_in->rc_id) != ls->ls_rcom_seq) {
 		log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
@@ -302,7 +301,7 @@ static void receive_sync_reply(struct dlm_ls *ls, const struct dlm_rcom *rc_in)
 	clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
 	wake_up(&ls->ls_wait_general);
  out:
-	spin_unlock(&ls->ls_rcom_spin);
+	spin_unlock_bh(&ls->ls_rcom_spin);
 }
 
 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name,
@@ -510,7 +509,7 @@ int dlm_send_ls_not_ready(int nodeid, const struct dlm_rcom *rc_in)
 	char *mb;
 	int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
 
-	mh = dlm_midcomms_get_mhandle(nodeid, mb_len, GFP_NOFS, &mb);
+	mh = dlm_midcomms_get_mhandle(nodeid, mb_len, &mb);
 	if (!mh)
 		return -ENOBUFS;
 
@@ -614,11 +613,11 @@ void dlm_receive_rcom(struct dlm_ls *ls, const struct dlm_rcom *rc, int nodeid)
 		break;
 	}
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	status = ls->ls_recover_status;
 	stop = dlm_recovery_stopped(ls);
 	seq = ls->ls_recover_seq;
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
 	if (stop && (rc->rc_type != cpu_to_le32(DLM_RCOM_STATUS)))
 		goto ignore;

fs/dlm/recover.c

@@ -74,9 +74,9 @@ int dlm_wait_function(struct dlm_ls *ls, int (*testfn) (struct dlm_ls *ls))
 uint32_t dlm_recover_status(struct dlm_ls *ls)
 {
 	uint32_t status;
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	status = ls->ls_recover_status;
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 	return status;
 }
 
@@ -87,9 +87,9 @@ static void _set_recover_status(struct dlm_ls *ls, uint32_t status)
 
 void dlm_set_recover_status(struct dlm_ls *ls, uint32_t status)
 {
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	_set_recover_status(ls, status);
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 }
 
 static int wait_status_all(struct dlm_ls *ls, uint32_t wait_status,
@@ -188,13 +188,13 @@ int dlm_recover_members_wait(struct dlm_ls *ls, uint64_t seq)
 
 		rv = dlm_slots_assign(ls, &num_slots, &slots_size, &slots, &gen);
 		if (!rv) {
-			spin_lock(&ls->ls_recover_lock);
+			spin_lock_bh(&ls->ls_recover_lock);
 			_set_recover_status(ls, DLM_RS_NODES_ALL);
 			ls->ls_num_slots = num_slots;
 			ls->ls_slots_size = slots_size;
 			ls->ls_slots = slots;
 			ls->ls_generation = gen;
-			spin_unlock(&ls->ls_recover_lock);
+			spin_unlock_bh(&ls->ls_recover_lock);
 		} else {
 			dlm_set_recover_status(ls, DLM_RS_NODES_ALL);
 		}
@@ -241,9 +241,9 @@ static int recover_list_empty(struct dlm_ls *ls)
 {
 	int empty;
 
-	spin_lock(&ls->ls_recover_list_lock);
+	spin_lock_bh(&ls->ls_recover_list_lock);
 	empty = list_empty(&ls->ls_recover_list);
-	spin_unlock(&ls->ls_recover_list_lock);
+	spin_unlock_bh(&ls->ls_recover_list_lock);
 
 	return empty;
 }
@@ -252,23 +252,23 @@ static void recover_list_add(struct dlm_rsb *r)
 {
 	struct dlm_ls *ls = r->res_ls;
 
-	spin_lock(&ls->ls_recover_list_lock);
+	spin_lock_bh(&ls->ls_recover_list_lock);
 	if (list_empty(&r->res_recover_list)) {
 		list_add_tail(&r->res_recover_list, &ls->ls_recover_list);
 		ls->ls_recover_list_count++;
 		dlm_hold_rsb(r);
 	}
-	spin_unlock(&ls->ls_recover_list_lock);
+	spin_unlock_bh(&ls->ls_recover_list_lock);
 }
 
 static void recover_list_del(struct dlm_rsb *r)
 {
 	struct dlm_ls *ls = r->res_ls;
 
-	spin_lock(&ls->ls_recover_list_lock);
+	spin_lock_bh(&ls->ls_recover_list_lock);
 	list_del_init(&r->res_recover_list);
 	ls->ls_recover_list_count--;
-	spin_unlock(&ls->ls_recover_list_lock);
+	spin_unlock_bh(&ls->ls_recover_list_lock);
 
 	dlm_put_rsb(r);
 }
@@ -277,7 +277,7 @@ static void recover_list_clear(struct dlm_ls *ls)
 {
 	struct dlm_rsb *r, *s;
 
-	spin_lock(&ls->ls_recover_list_lock);
+	spin_lock_bh(&ls->ls_recover_list_lock);
 	list_for_each_entry_safe(r, s, &ls->ls_recover_list, res_recover_list) {
 		list_del_init(&r->res_recover_list);
 		r->res_recover_locks_count = 0;
@@ -290,17 +290,17 @@ static void recover_list_clear(struct dlm_ls *ls)
 			  ls->ls_recover_list_count);
 		ls->ls_recover_list_count = 0;
 	}
-	spin_unlock(&ls->ls_recover_list_lock);
+	spin_unlock_bh(&ls->ls_recover_list_lock);
 }
 
 static int recover_idr_empty(struct dlm_ls *ls)
 {
 	int empty = 1;
 
-	spin_lock(&ls->ls_recover_idr_lock);
+	spin_lock_bh(&ls->ls_recover_idr_lock);
 	if (ls->ls_recover_list_count)
 		empty = 0;
-	spin_unlock(&ls->ls_recover_idr_lock);
+	spin_unlock_bh(&ls->ls_recover_idr_lock);
 
 	return empty;
 }
@@ -310,8 +310,7 @@ static int recover_idr_add(struct dlm_rsb *r)
 	struct dlm_ls *ls = r->res_ls;
 	int rv;
 
-	idr_preload(GFP_NOFS);
-	spin_lock(&ls->ls_recover_idr_lock);
+	spin_lock_bh(&ls->ls_recover_idr_lock);
 	if (r->res_id) {
 		rv = -1;
 		goto out_unlock;
@@ -325,8 +324,7 @@ static int recover_idr_add(struct dlm_rsb *r)
 	dlm_hold_rsb(r);
 	rv = 0;
 out_unlock:
-	spin_unlock(&ls->ls_recover_idr_lock);
-	idr_preload_end();
+	spin_unlock_bh(&ls->ls_recover_idr_lock);
 	return rv;
 }
 
@@ -334,11 +332,11 @@ static void recover_idr_del(struct dlm_rsb *r)
 {
 	struct dlm_ls *ls = r->res_ls;
 
-	spin_lock(&ls->ls_recover_idr_lock);
+	spin_lock_bh(&ls->ls_recover_idr_lock);
 	idr_remove(&ls->ls_recover_idr, r->res_id);
 	r->res_id = 0;
 	ls->ls_recover_list_count--;
-	spin_unlock(&ls->ls_recover_idr_lock);
+	spin_unlock_bh(&ls->ls_recover_idr_lock);
 
 	dlm_put_rsb(r);
 }
@@ -347,9 +345,9 @@ static struct dlm_rsb *recover_idr_find(struct dlm_ls *ls, uint64_t id)
 {
 	struct dlm_rsb *r;
 
-	spin_lock(&ls->ls_recover_idr_lock);
+	spin_lock_bh(&ls->ls_recover_idr_lock);
 	r = idr_find(&ls->ls_recover_idr, (int)id);
-	spin_unlock(&ls->ls_recover_idr_lock);
+	spin_unlock_bh(&ls->ls_recover_idr_lock);
 	return r;
 }
 
@@ -358,7 +356,7 @@ static void recover_idr_clear(struct dlm_ls *ls)
 	struct dlm_rsb *r;
 	int id;
 
-	spin_lock(&ls->ls_recover_idr_lock);
+	spin_lock_bh(&ls->ls_recover_idr_lock);
 
 	idr_for_each_entry(&ls->ls_recover_idr, r, id) {
 		idr_remove(&ls->ls_recover_idr, id);
@@ -374,7 +372,7 @@ static void recover_idr_clear(struct dlm_ls *ls)
 			  ls->ls_recover_list_count);
 		ls->ls_recover_list_count = 0;
 	}
-	spin_unlock(&ls->ls_recover_idr_lock);
+	spin_unlock_bh(&ls->ls_recover_idr_lock);
 }
 
 
@@ -521,7 +519,8 @@ static int recover_master_static(struct dlm_rsb *r, unsigned int *count)
  * the correct dir node.
  */
 
-int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq)
+int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq,
+			const struct list_head *root_list)
 {
 	struct dlm_rsb *r;
 	unsigned int total = 0;
@@ -531,10 +530,8 @@ int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq)
 
 	log_rinfo(ls, "dlm_recover_masters");
 
-	down_read(&ls->ls_root_sem);
-	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+	list_for_each_entry(r, root_list, res_root_list) {
 		if (dlm_recovery_stopped(ls)) {
-			up_read(&ls->ls_root_sem);
 			error = -EINTR;
 			goto out;
 		}
@@ -548,12 +545,9 @@ int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq)
 		cond_resched();
 		total++;
 
-		if (error) {
-			up_read(&ls->ls_root_sem);
+		if (error)
 			goto out;
 	}
-	}
-	up_read(&ls->ls_root_sem);
 
 	log_rinfo(ls, "dlm_recover_masters %u of %u", count, total);
 
@@ -658,13 +652,13 @@ static int recover_locks(struct dlm_rsb *r, uint64_t seq)
 	return error;
 }
 
-int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq)
+int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq,
+		      const struct list_head *root_list)
 {
 	struct dlm_rsb *r;
 	int error, count = 0;
 
-	down_read(&ls->ls_root_sem);
-	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+	list_for_each_entry(r, root_list, res_root_list) {
 		if (is_master(r)) {
 			rsb_clear_flag(r, RSB_NEW_MASTER);
 			continue;
@@ -675,19 +669,15 @@ int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq)
 
 		if (dlm_recovery_stopped(ls)) {
 			error = -EINTR;
-			up_read(&ls->ls_root_sem);
 			goto out;
 		}
 
 		error = recover_locks(r, seq);
-		if (error) {
-			up_read(&ls->ls_root_sem);
+		if (error)
 			goto out;
-		}
 
 		count += r->res_recover_locks_count;
 	}
-	up_read(&ls->ls_root_sem);
 
 	log_rinfo(ls, "dlm_recover_locks %d out", count);
 
@@ -856,13 +846,12 @@ static void recover_grant(struct dlm_rsb *r)
 		rsb_set_flag(r, RSB_RECOVER_GRANT);
 }
 
-void dlm_recover_rsbs(struct dlm_ls *ls)
+void dlm_recover_rsbs(struct dlm_ls *ls, const struct list_head *root_list)
 {
 	struct dlm_rsb *r;
 	unsigned int count = 0;
 
-	down_read(&ls->ls_root_sem);
-	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
+	list_for_each_entry(r, root_list, res_root_list) {
 		lock_rsb(r);
 		if (is_master(r)) {
 			if (rsb_flag(r, RSB_RECOVER_CONVERT))
@@ -883,7 +872,6 @@ void dlm_recover_rsbs(struct dlm_ls *ls)
 		rsb_clear_flag(r, RSB_NEW_MASTER2);
 		unlock_rsb(r);
 	}
-	up_read(&ls->ls_root_sem);
 
 	if (count)
 		log_rinfo(ls, "dlm_recover_rsbs %d done", count);
@@ -891,66 +879,25 @@ void dlm_recover_rsbs(struct dlm_ls *ls)
 
 /* Create a single list of all root rsb's to be used during recovery */
 
-int dlm_create_root_list(struct dlm_ls *ls)
-{
-	struct rb_node *n;
-	struct dlm_rsb *r;
-	int i, error = 0;
-
-	down_write(&ls->ls_root_sem);
-	if (!list_empty(&ls->ls_root_list)) {
-		log_error(ls, "root list not empty");
-		error = -EINVAL;
-		goto out;
-	}
-
-	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
-		spin_lock(&ls->ls_rsbtbl[i].lock);
-		for (n = rb_first(&ls->ls_rsbtbl[i].keep); n; n = rb_next(n)) {
-			r = rb_entry(n, struct dlm_rsb, res_hashnode);
-			list_add(&r->res_root_list, &ls->ls_root_list);
-			dlm_hold_rsb(r);
-		}
-
-		if (!RB_EMPTY_ROOT(&ls->ls_rsbtbl[i].toss))
-			log_error(ls, "dlm_create_root_list toss not empty");
-		spin_unlock(&ls->ls_rsbtbl[i].lock);
-	}
- out:
-	up_write(&ls->ls_root_sem);
-	return error;
-}
-
-void dlm_release_root_list(struct dlm_ls *ls)
+void dlm_clear_toss(struct dlm_ls *ls)
 {
 	struct dlm_rsb *r, *safe;
+	unsigned int count = 0;
 
-	down_write(&ls->ls_root_sem);
-	list_for_each_entry_safe(r, safe, &ls->ls_root_list, res_root_list) {
-		list_del_init(&r->res_root_list);
-		dlm_put_rsb(r);
-	}
-	up_write(&ls->ls_root_sem);
-}
+	write_lock_bh(&ls->ls_rsbtbl_lock);
+	list_for_each_entry_safe(r, safe, &ls->ls_toss, res_rsbs_list) {
+		list_del(&r->res_rsbs_list);
+		rhashtable_remove_fast(&ls->ls_rsbtbl, &r->res_node,
+				       dlm_rhash_rsb_params);
 
-void dlm_clear_toss(struct dlm_ls *ls)
-{
-	struct rb_node *n, *next;
-	struct dlm_rsb *r;
-	unsigned int count = 0;
-	int i;
-
-	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
-		spin_lock(&ls->ls_rsbtbl[i].lock);
-		for (n = rb_first(&ls->ls_rsbtbl[i].toss); n; n = next) {
-			next = rb_next(n);
-			r = rb_entry(n, struct dlm_rsb, res_hashnode);
-			rb_erase(n, &ls->ls_rsbtbl[i].toss);
-			dlm_free_rsb(r);
+		/* remove it from the toss queue if its part of it */
+		if (!list_empty(&r->res_toss_q_list))
+			list_del_init(&r->res_toss_q_list);
+
+		free_toss_rsb(r);
 		count++;
 	}
-		spin_unlock(&ls->ls_rsbtbl[i].lock);
-	}
+	write_unlock_bh(&ls->ls_rsbtbl_lock);
 
 	if (count)
 		log_rinfo(ls, "dlm_clear_toss %u done", count);

fs/dlm/recover.h

@@ -19,14 +19,14 @@ int dlm_recover_members_wait(struct dlm_ls *ls, uint64_t seq);
 int dlm_recover_directory_wait(struct dlm_ls *ls, uint64_t seq);
 int dlm_recover_locks_wait(struct dlm_ls *ls, uint64_t seq);
 int dlm_recover_done_wait(struct dlm_ls *ls, uint64_t seq);
-int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq);
+int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq,
+			const struct list_head *root_list);
 int dlm_recover_master_reply(struct dlm_ls *ls, const struct dlm_rcom *rc);
-int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq);
+int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq,
+		      const struct list_head *root_list);
 void dlm_recovered_lock(struct dlm_rsb *r);
-int dlm_create_root_list(struct dlm_ls *ls);
-void dlm_release_root_list(struct dlm_ls *ls);
 void dlm_clear_toss(struct dlm_ls *ls);
-void dlm_recover_rsbs(struct dlm_ls *ls);
+void dlm_recover_rsbs(struct dlm_ls *ls, const struct list_head *root_list);
 
 #endif				/* __RECOVER_DOT_H__ */
 

fs/dlm/recoverd.c

@@ -20,6 +20,67 @@
 #include "requestqueue.h"
 #include "recoverd.h"
 
+static int dlm_create_masters_list(struct dlm_ls *ls)
+{
+	struct dlm_rsb *r;
+	int error = 0;
+
+	write_lock_bh(&ls->ls_masters_lock);
+	if (!list_empty(&ls->ls_masters_list)) {
+		log_error(ls, "root list not empty");
+		error = -EINVAL;
+		goto out;
+	}
+
+	read_lock_bh(&ls->ls_rsbtbl_lock);
+	list_for_each_entry(r, &ls->ls_keep, res_rsbs_list) {
+		if (r->res_nodeid)
+			continue;
+
+		list_add(&r->res_masters_list, &ls->ls_masters_list);
+		dlm_hold_rsb(r);
+	}
+	read_unlock_bh(&ls->ls_rsbtbl_lock);
+ out:
+	write_unlock_bh(&ls->ls_masters_lock);
+	return error;
+}
+
+static void dlm_release_masters_list(struct dlm_ls *ls)
+{
+	struct dlm_rsb *r, *safe;
+
+	write_lock_bh(&ls->ls_masters_lock);
+	list_for_each_entry_safe(r, safe, &ls->ls_masters_list, res_masters_list) {
+		list_del_init(&r->res_masters_list);
+		dlm_put_rsb(r);
+	}
+	write_unlock_bh(&ls->ls_masters_lock);
+}
+
+static void dlm_create_root_list(struct dlm_ls *ls, struct list_head *root_list)
+{
+	struct dlm_rsb *r;
+
+	read_lock_bh(&ls->ls_rsbtbl_lock);
+	list_for_each_entry(r, &ls->ls_keep, res_rsbs_list) {
+		list_add(&r->res_root_list, root_list);
+		dlm_hold_rsb(r);
+	}
+
+	WARN_ON_ONCE(!list_empty(&ls->ls_toss));
+	read_unlock_bh(&ls->ls_rsbtbl_lock);
+}
+
+static void dlm_release_root_list(struct list_head *root_list)
+{
+	struct dlm_rsb *r, *safe;
+
+	list_for_each_entry_safe(r, safe, root_list, res_root_list) {
+		list_del_init(&r->res_root_list);
+		dlm_put_rsb(r);
+	}
+}
 
 /* If the start for which we're re-enabling locking (seq) has been superseded
    by a newer stop (ls_recover_seq), we need to leave locking disabled.
@@ -32,24 +93,35 @@ static int enable_locking(struct dlm_ls *ls, uint64_t seq)
 {
 	int error = -EINTR;
 
-	down_write(&ls->ls_recv_active);
+	write_lock_bh(&ls->ls_recv_active);
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	if (ls->ls_recover_seq == seq) {
 		set_bit(LSFL_RUNNING, &ls->ls_flags);
+		/* Schedule next timer if recovery put something on toss.
+		 *
+		 * The rsbs that was queued while recovery on toss hasn't
+		 * started yet because LSFL_RUNNING was set everything
+		 * else recovery hasn't started as well because ls_in_recovery
+		 * is still hold. So we should not run into the case that
+		 * dlm_timer_resume() queues a timer that can occur in
+		 * a no op.
+		 */
+		dlm_timer_resume(ls);
 		/* unblocks processes waiting to enter the dlm */
 		up_write(&ls->ls_in_recovery);
 		clear_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
 		error = 0;
 	}
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
-	up_write(&ls->ls_recv_active);
+	write_unlock_bh(&ls->ls_recv_active);
 	return error;
 }
 
 static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 {
+	LIST_HEAD(root_list);
 	unsigned long start;
 	int error, neg = 0;
 
@@ -66,7 +138,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 	 * routines.
 	 */
 
-	dlm_create_root_list(ls);
+	dlm_create_root_list(ls, &root_list);
 
 	/*
 	 * Add or remove nodes from the lockspace's ls_nodes list.
@@ -82,10 +154,25 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		goto fail;
 	}
 
-	dlm_recover_dir_nodeid(ls);
+	dlm_recover_dir_nodeid(ls, &root_list);
+
+	/* Create a snapshot of all active rsbs were we are the master of.
+	 * During the barrier between dlm_recover_members_wait() and
+	 * dlm_recover_directory() other nodes can dump their necessary
+	 * directory dlm_rsb (r->res_dir_nodeid == nodeid) in rcom
+	 * communication dlm_copy_master_names() handling.
+	 *
+	 * TODO We should create a per lockspace list that contains rsbs
+	 * that we are the master of. Instead of creating this list while
+	 * recovery we keep track of those rsbs while locking handling and
+	 * recovery can use it when necessary.
+	 */
+	error = dlm_create_masters_list(ls);
+	if (error) {
+		log_rinfo(ls, "dlm_create_masters_list error %d", error);
+		goto fail_root_list;
+	}
 
-	ls->ls_recover_dir_sent_res = 0;
-	ls->ls_recover_dir_sent_msg = 0;
 	ls->ls_recover_locks_in = 0;
 
 	dlm_set_recover_status(ls, DLM_RS_NODES);
@@ -93,7 +180,8 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 	error = dlm_recover_members_wait(ls, rv->seq);
 	if (error) {
 		log_rinfo(ls, "dlm_recover_members_wait error %d", error);
-		goto fail;
+		dlm_release_masters_list(ls);
+		goto fail_root_list;
 	}
 
 	start = jiffies;
@@ -106,7 +194,8 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 	error = dlm_recover_directory(ls, rv->seq);
 	if (error) {
 		log_rinfo(ls, "dlm_recover_directory error %d", error);
-		goto fail;
+		dlm_release_masters_list(ls);
+		goto fail_root_list;
 	}
 
 	dlm_set_recover_status(ls, DLM_RS_DIR);
@@ -114,11 +203,11 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 	error = dlm_recover_directory_wait(ls, rv->seq);
 	if (error) {
 		log_rinfo(ls, "dlm_recover_directory_wait error %d", error);
-		goto fail;
+		dlm_release_masters_list(ls);
+		goto fail_root_list;
 	}
 
-	log_rinfo(ls, "dlm_recover_directory %u out %u messages",
-		  ls->ls_recover_dir_sent_res, ls->ls_recover_dir_sent_msg);
+	dlm_release_masters_list(ls);
 
 	/*
 	 * We may have outstanding operations that are waiting for a reply from
@@ -130,7 +219,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 
 	if (dlm_recovery_stopped(ls)) {
 		error = -EINTR;
-		goto fail;
+		goto fail_root_list;
 	}
 
 	if (neg || dlm_no_directory(ls)) {
@@ -138,27 +227,27 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		 * Clear lkb's for departed nodes.
 		 */
 
-		dlm_recover_purge(ls);
+		dlm_recover_purge(ls, &root_list);
 
 		/*
 		 * Get new master nodeid's for rsb's that were mastered on
 		 * departed nodes.
 		 */
 
-		error = dlm_recover_masters(ls, rv->seq);
+		error = dlm_recover_masters(ls, rv->seq, &root_list);
 		if (error) {
 			log_rinfo(ls, "dlm_recover_masters error %d", error);
-			goto fail;
+			goto fail_root_list;
 		}
 
 		/*
 		 * Send our locks on remastered rsb's to the new masters.
 		 */
 
-		error = dlm_recover_locks(ls, rv->seq);
+		error = dlm_recover_locks(ls, rv->seq, &root_list);
 		if (error) {
 			log_rinfo(ls, "dlm_recover_locks error %d", error);
-			goto fail;
+			goto fail_root_list;
 		}
 
 		dlm_set_recover_status(ls, DLM_RS_LOCKS);
@@ -166,7 +255,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		error = dlm_recover_locks_wait(ls, rv->seq);
 		if (error) {
 			log_rinfo(ls, "dlm_recover_locks_wait error %d", error);
-			goto fail;
+			goto fail_root_list;
 		}
 
 		log_rinfo(ls, "dlm_recover_locks %u in",
@@ -178,7 +267,7 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		 * settings.
 		 */
 
-		dlm_recover_rsbs(ls);
+		dlm_recover_rsbs(ls, &root_list);
 	} else {
 		/*
 		 * Other lockspace members may be going through the "neg" steps
@@ -190,11 +279,11 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 		error = dlm_recover_locks_wait(ls, rv->seq);
 		if (error) {
 			log_rinfo(ls, "dlm_recover_locks_wait error %d", error);
-			goto fail;
+			goto fail_root_list;
 		}
 	}
 
-	dlm_release_root_list(ls);
+	dlm_release_root_list(&root_list);
 
 	/*
 	 * Purge directory-related requests that are saved in requestqueue.
@@ -243,8 +332,9 @@ static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
 
 	return 0;
 
+ fail_root_list:
+	dlm_release_root_list(&root_list);
  fail:
-	dlm_release_root_list(ls);
 	mutex_unlock(&ls->ls_recoverd_active);
 
 	return error;
@@ -259,12 +349,12 @@ static void do_ls_recovery(struct dlm_ls *ls)
 	struct dlm_recover *rv = NULL;
 	int error;
 
-	spin_lock(&ls->ls_recover_lock);
+	spin_lock_bh(&ls->ls_recover_lock);
 	rv = ls->ls_recover_args;
 	ls->ls_recover_args = NULL;
 	if (rv && ls->ls_recover_seq == rv->seq)
 		clear_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
-	spin_unlock(&ls->ls_recover_lock);
+	spin_unlock_bh(&ls->ls_recover_lock);
 
 	if (rv) {
 		error = ls_recover(ls, rv);

fs/dlm/requestqueue.c

@@ -37,7 +37,7 @@ void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid,
 	int length = le16_to_cpu(ms->m_header.h_length) -
 		sizeof(struct dlm_message);
 
-	e = kmalloc(sizeof(struct rq_entry) + length, GFP_NOFS);
+	e = kmalloc(sizeof(struct rq_entry) + length, GFP_ATOMIC);
 	if (!e) {
 		log_print("dlm_add_requestqueue: out of memory len %d", length);
 		return;
@@ -48,10 +48,7 @@ void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid,
 	memcpy(&e->request, ms, sizeof(*ms));
 	memcpy(&e->request.m_extra, ms->m_extra, length);
 
-	atomic_inc(&ls->ls_requestqueue_cnt);
-	mutex_lock(&ls->ls_requestqueue_mutex);
 	list_add_tail(&e->list, &ls->ls_requestqueue);
-	mutex_unlock(&ls->ls_requestqueue_mutex);
 }
 
 /*
@@ -71,16 +68,14 @@ int dlm_process_requestqueue(struct dlm_ls *ls)
 	struct dlm_message *ms;
 	int error = 0;
 
-	mutex_lock(&ls->ls_requestqueue_mutex);
-
+	write_lock_bh(&ls->ls_requestqueue_lock);
 	for (;;) {
 		if (list_empty(&ls->ls_requestqueue)) {
-			mutex_unlock(&ls->ls_requestqueue_mutex);
+			clear_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags);
 			error = 0;
 			break;
 		}
-		e = list_entry(ls->ls_requestqueue.next, struct rq_entry, list);
-		mutex_unlock(&ls->ls_requestqueue_mutex);
+		e = list_first_entry(&ls->ls_requestqueue, struct rq_entry, list);
 
 		ms = &e->request;
 
@@ -93,41 +88,23 @@ int dlm_process_requestqueue(struct dlm_ls *ls)
 			  e->recover_seq);
 
 		dlm_receive_message_saved(ls, &e->request, e->recover_seq);
-
-		mutex_lock(&ls->ls_requestqueue_mutex);
 		list_del(&e->list);
-		if (atomic_dec_and_test(&ls->ls_requestqueue_cnt))
-			wake_up(&ls->ls_requestqueue_wait);
 		kfree(e);
 
 		if (dlm_locking_stopped(ls)) {
 			log_debug(ls, "process_requestqueue abort running");
-			mutex_unlock(&ls->ls_requestqueue_mutex);
 			error = -EINTR;
 			break;
 		}
+		write_unlock_bh(&ls->ls_requestqueue_lock);
 		schedule();
+		write_lock_bh(&ls->ls_requestqueue_lock);
 	}
+	write_unlock_bh(&ls->ls_requestqueue_lock);
 
 	return error;
 }
 
-/*
- * After recovery is done, locking is resumed and dlm_recoverd takes all the
- * saved requests and processes them as they would have been by dlm_recv.  At
- * the same time, dlm_recv will start receiving new requests from remote nodes.
- * We want to delay dlm_recv processing new requests until dlm_recoverd has
- * finished processing the old saved requests.  We don't check for locking
- * stopped here because dlm_ls_stop won't stop locking until it's suspended us
- * (dlm_recv).
- */
-
-void dlm_wait_requestqueue(struct dlm_ls *ls)
-{
-	wait_event(ls->ls_requestqueue_wait,
-		   atomic_read(&ls->ls_requestqueue_cnt) == 0);
-}
-
 static int purge_request(struct dlm_ls *ls, struct dlm_message *ms, int nodeid)
 {
 	__le32 type = ms->m_type;
@@ -158,17 +135,15 @@ void dlm_purge_requestqueue(struct dlm_ls *ls)
 	struct dlm_message *ms;
 	struct rq_entry *e, *safe;
 
-	mutex_lock(&ls->ls_requestqueue_mutex);
+	write_lock_bh(&ls->ls_requestqueue_lock);
 	list_for_each_entry_safe(e, safe, &ls->ls_requestqueue, list) {
 		ms =  &e->request;
 
 		if (purge_request(ls, ms, e->nodeid)) {
 			list_del(&e->list);
-			if (atomic_dec_and_test(&ls->ls_requestqueue_cnt))
-				wake_up(&ls->ls_requestqueue_wait);
 			kfree(e);
 		}
 	}
-	mutex_unlock(&ls->ls_requestqueue_mutex);
+	write_unlock_bh(&ls->ls_requestqueue_lock);
 }
 

fs/dlm/user.c

@@ -145,24 +145,6 @@ static void compat_output(struct dlm_lock_result *res,
 }
 #endif
 
-/* should held proc->asts_spin lock */
-void dlm_purge_lkb_callbacks(struct dlm_lkb *lkb)
-{
-	struct dlm_callback *cb, *safe;
-
-	list_for_each_entry_safe(cb, safe, &lkb->lkb_callbacks, list) {
-		list_del(&cb->list);
-		kref_put(&cb->ref, dlm_release_callback);
-	}
-
-	clear_bit(DLM_IFL_CB_PENDING_BIT, &lkb->lkb_iflags);
-
-	/* invalidate */
-	dlm_callback_set_last_ptr(&lkb->lkb_last_cast, NULL);
-	dlm_callback_set_last_ptr(&lkb->lkb_last_cb, NULL);
-	lkb->lkb_last_bast_mode = -1;
-}
-
 /* Figure out if this lock is at the end of its life and no longer
    available for the application to use.  The lkb still exists until
    the final ast is read.  A lock becomes EOL in three situations:
@@ -199,6 +181,7 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
 	struct dlm_ls *ls;
 	struct dlm_user_args *ua;
 	struct dlm_user_proc *proc;
+	struct dlm_callback *cb;
 	int rv;
 
 	if (test_bit(DLM_DFL_ORPHAN_BIT, &lkb->lkb_dflags) ||
@@ -206,7 +189,7 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
 		return;
 
 	ls = lkb->lkb_resource->res_ls;
-	spin_lock(&ls->ls_clear_proc_locks);
+	spin_lock_bh(&ls->ls_clear_proc_locks);
 
 	/* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
 	   can't be delivered.  For ORPHAN's, dlm_clear_proc_locks() freed
@@ -228,38 +211,44 @@ void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
 	if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
 		set_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags);
 
-	spin_lock(&proc->asts_spin);
+	spin_lock_bh(&proc->asts_spin);
 
-	rv = dlm_enqueue_lkb_callback(lkb, flags, mode, status, sbflags);
+	rv = dlm_queue_lkb_callback(lkb, flags, mode, status, sbflags, &cb);
 	switch (rv) {
-	case DLM_ENQUEUE_CALLBACK_FAILURE:
-		spin_unlock(&proc->asts_spin);
-		WARN_ON_ONCE(1);
-		goto out;
 	case DLM_ENQUEUE_CALLBACK_NEED_SCHED:
-		kref_get(&lkb->lkb_ref);
-		list_add_tail(&lkb->lkb_cb_list, &proc->asts);
+		cb->ua = *ua;
+		cb->lkb_lksb = &cb->ua.lksb;
+		if (cb->copy_lvb) {
+			memcpy(cb->lvbptr, ua->lksb.sb_lvbptr,
+			       DLM_USER_LVB_LEN);
+			cb->lkb_lksb->sb_lvbptr = cb->lvbptr;
+		}
+
+		list_add_tail(&cb->list, &proc->asts);
 		wake_up_interruptible(&proc->wait);
 		break;
 	case DLM_ENQUEUE_CALLBACK_SUCCESS:
 		break;
+	case DLM_ENQUEUE_CALLBACK_FAILURE:
+		fallthrough;
 	default:
+		spin_unlock_bh(&proc->asts_spin);
 		WARN_ON_ONCE(1);
-		break;
+		goto out;
 	}
-	spin_unlock(&proc->asts_spin);
+	spin_unlock_bh(&proc->asts_spin);
 
 	if (test_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags)) {
 		/* N.B. spin_lock locks_spin, not asts_spin */
-		spin_lock(&proc->locks_spin);
+		spin_lock_bh(&proc->locks_spin);
 		if (!list_empty(&lkb->lkb_ownqueue)) {
 			list_del_init(&lkb->lkb_ownqueue);
 			dlm_put_lkb(lkb);
 		}
-		spin_unlock(&proc->locks_spin);
+		spin_unlock_bh(&proc->locks_spin);
 	}
  out:
-	spin_unlock(&ls->ls_clear_proc_locks);
+	spin_unlock_bh(&ls->ls_clear_proc_locks);
 }
 
 static int device_user_lock(struct dlm_user_proc *proc,
@@ -803,11 +792,9 @@ static ssize_t device_read(struct file *file, char __user *buf, size_t count,
 			   loff_t *ppos)
 {
 	struct dlm_user_proc *proc = file->private_data;
-	struct dlm_lkb *lkb;
 	DECLARE_WAITQUEUE(wait, current);
 	struct dlm_callback *cb;
-	int rv, ret, copy_lvb = 0;
-	int old_mode, new_mode;
+	int rv, ret;
 
 	if (count == sizeof(struct dlm_device_version)) {
 		rv = copy_version_to_user(buf, count);
@@ -826,16 +813,14 @@ static ssize_t device_read(struct file *file, char __user *buf, size_t count,
 #endif
 		return -EINVAL;
 
- try_another:
-
 	/* do we really need this? can a read happen after a close? */
 	if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
 		return -EINVAL;
 
-	spin_lock(&proc->asts_spin);
+	spin_lock_bh(&proc->asts_spin);
 	if (list_empty(&proc->asts)) {
 		if (file->f_flags & O_NONBLOCK) {
-			spin_unlock(&proc->asts_spin);
+			spin_unlock_bh(&proc->asts_spin);
 			return -EAGAIN;
 		}
 
@@ -844,16 +829,16 @@ static ssize_t device_read(struct file *file, char __user *buf, size_t count,
 	repeat:
 		set_current_state(TASK_INTERRUPTIBLE);
 		if (list_empty(&proc->asts) && !signal_pending(current)) {
-			spin_unlock(&proc->asts_spin);
+			spin_unlock_bh(&proc->asts_spin);
 			schedule();
-			spin_lock(&proc->asts_spin);
+			spin_lock_bh(&proc->asts_spin);
 			goto repeat;
 		}
 		set_current_state(TASK_RUNNING);
 		remove_wait_queue(&proc->wait, &wait);
 
 		if (signal_pending(current)) {
-			spin_unlock(&proc->asts_spin);
+			spin_unlock_bh(&proc->asts_spin);
 			return -ERESTARTSYS;
 		}
 	}
@@ -862,60 +847,24 @@ static ssize_t device_read(struct file *file, char __user *buf, size_t count,
 	   without removing lkb_cb_list; so empty lkb_cb_list is always
 	   consistent with empty lkb_callbacks */
 
-	lkb = list_first_entry(&proc->asts, struct dlm_lkb, lkb_cb_list);
-
-	/* rem_lkb_callback sets a new lkb_last_cast */
-	old_mode = lkb->lkb_last_cast->mode;
-
-	rv = dlm_dequeue_lkb_callback(lkb, &cb);
-	switch (rv) {
-	case DLM_DEQUEUE_CALLBACK_EMPTY:
-		/* this shouldn't happen; lkb should have been removed from
-		 * list when last item was dequeued
-		 */
-		log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
-		list_del_init(&lkb->lkb_cb_list);
-		spin_unlock(&proc->asts_spin);
-		/* removes ref for proc->asts, may cause lkb to be freed */
-		dlm_put_lkb(lkb);
-		WARN_ON_ONCE(1);
-		goto try_another;
-	case DLM_DEQUEUE_CALLBACK_LAST:
-		list_del_init(&lkb->lkb_cb_list);
-		clear_bit(DLM_IFL_CB_PENDING_BIT, &lkb->lkb_iflags);
-		break;
-	case DLM_DEQUEUE_CALLBACK_SUCCESS:
-		break;
-	default:
-		WARN_ON_ONCE(1);
-		break;
-	}
-	spin_unlock(&proc->asts_spin);
+	cb = list_first_entry(&proc->asts, struct dlm_callback, list);
+	list_del(&cb->list);
+	spin_unlock_bh(&proc->asts_spin);
 
 	if (cb->flags & DLM_CB_BAST) {
-		trace_dlm_bast(lkb->lkb_resource->res_ls, lkb, cb->mode);
+		trace_dlm_bast(cb->ls_id, cb->lkb_id, cb->mode, cb->res_name,
+			       cb->res_length);
 	} else if (cb->flags & DLM_CB_CAST) {
-		new_mode = cb->mode;
-
-		if (!cb->sb_status && lkb->lkb_lksb->sb_lvbptr &&
-		    dlm_lvb_operations[old_mode + 1][new_mode + 1])
-			copy_lvb = 1;
-
-		lkb->lkb_lksb->sb_status = cb->sb_status;
-		lkb->lkb_lksb->sb_flags = cb->sb_flags;
-		trace_dlm_ast(lkb->lkb_resource->res_ls, lkb);
+		cb->lkb_lksb->sb_status = cb->sb_status;
+		cb->lkb_lksb->sb_flags = cb->sb_flags;
+		trace_dlm_ast(cb->ls_id, cb->lkb_id, cb->sb_status,
+			      cb->sb_flags, cb->res_name, cb->res_length);
 	}
 
-	ret = copy_result_to_user(lkb->lkb_ua,
+	ret = copy_result_to_user(&cb->ua,
 				  test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
-				  cb->flags, cb->mode, copy_lvb, buf, count);
-
-	kref_put(&cb->ref, dlm_release_callback);
-
-	/* removes ref for proc->asts, may cause lkb to be freed */
-	if (rv == DLM_DEQUEUE_CALLBACK_LAST)
-		dlm_put_lkb(lkb);
-
+				  cb->flags, cb->mode, cb->copy_lvb, buf, count);
+	dlm_free_cb(cb);
 	return ret;
 }
 
@@ -925,12 +874,12 @@ static __poll_t device_poll(struct file *file, poll_table *wait)
 
 	poll_wait(file, &proc->wait, wait);
 
-	spin_lock(&proc->asts_spin);
+	spin_lock_bh(&proc->asts_spin);
 	if (!list_empty(&proc->asts)) {
-		spin_unlock(&proc->asts_spin);
+		spin_unlock_bh(&proc->asts_spin);
 		return EPOLLIN | EPOLLRDNORM;
 	}
-	spin_unlock(&proc->asts_spin);
+	spin_unlock_bh(&proc->asts_spin);
 	return 0;
 }
 

include/trace/events/dlm.h

@@ -189,28 +189,24 @@ TRACE_EVENT(dlm_lock_end,
 
 TRACE_EVENT(dlm_bast,
 
-	TP_PROTO(struct dlm_ls *ls, struct dlm_lkb *lkb, int mode),
+	TP_PROTO(__u32 ls_id, __u32 lkb_id, int mode,
+		 const char *res_name, size_t res_length),
 
-	TP_ARGS(ls, lkb, mode),
+	TP_ARGS(ls_id, lkb_id, mode, res_name, res_length),
 
 	TP_STRUCT__entry(
 		__field(__u32, ls_id)
 		__field(__u32, lkb_id)
 		__field(int, mode)
-		__dynamic_array(unsigned char, res_name,
-				lkb->lkb_resource ? lkb->lkb_resource->res_length : 0)
+		__dynamic_array(unsigned char, res_name, res_length)
 	),
 
 	TP_fast_assign(
-		struct dlm_rsb *r;
-
-		__entry->ls_id = ls->ls_global_id;
-		__entry->lkb_id = lkb->lkb_id;
+		__entry->ls_id = ls_id;
+		__entry->lkb_id = lkb_id;
 		__entry->mode = mode;
 
-		r = lkb->lkb_resource;
-		if (r)
-			memcpy(__get_dynamic_array(res_name), r->res_name,
+		memcpy(__get_dynamic_array(res_name), res_name,
 		       __get_dynamic_array_len(res_name));
 	),
 
@@ -224,30 +220,26 @@ TRACE_EVENT(dlm_bast,
 
 TRACE_EVENT(dlm_ast,
 
-	TP_PROTO(struct dlm_ls *ls, struct dlm_lkb *lkb),
+	TP_PROTO(__u32 ls_id, __u32 lkb_id, __u8 sb_flags, int sb_status,
+		 const char *res_name, size_t res_length),
 
-	TP_ARGS(ls, lkb),
+	TP_ARGS(ls_id, lkb_id, sb_flags, sb_status, res_name, res_length),
 
 	TP_STRUCT__entry(
 		__field(__u32, ls_id)
 		__field(__u32, lkb_id)
-		__field(u8, sb_flags)
+		__field(__u8, sb_flags)
 		__field(int, sb_status)
-		__dynamic_array(unsigned char, res_name,
-				lkb->lkb_resource ? lkb->lkb_resource->res_length : 0)
+		__dynamic_array(unsigned char, res_name, res_length)
 	),
 
 	TP_fast_assign(
-		struct dlm_rsb *r;
+		__entry->ls_id = ls_id;
+		__entry->lkb_id = lkb_id;
+		__entry->sb_flags = sb_flags;
+		__entry->sb_status = sb_status;
 
-		__entry->ls_id = ls->ls_global_id;
-		__entry->lkb_id = lkb->lkb_id;
-		__entry->sb_flags = lkb->lkb_lksb->sb_flags;
-		__entry->sb_status = lkb->lkb_lksb->sb_status;
-
-		r = lkb->lkb_resource;
-		if (r)
-			memcpy(__get_dynamic_array(res_name), r->res_name,
+		memcpy(__get_dynamic_array(res_name), res_name,
 		       __get_dynamic_array_len(res_name));
 	),