1 files changed, 132 insertions, 47 deletions

diff --git a/fs/ocfs2/dlm/dlmlock.c b/fs/ocfs2/dlm/dlmlock.c
index d1a0038557a..66c2a491f68 100644
--- a/fs/ocfs2/dlm/dlmlock.c
+++ b/fs/ocfs2/dlm/dlmlock.c
@@ -30,7 +30,6 @@
 #include <linux/types.h>
 #include <linux/slab.h>
 #include <linux/highmem.h>
-#include <linux/utsname.h>
 #include <linux/init.h>
 #include <linux/sysctl.h>
 #include <linux/random.h>
@@ -53,7 +52,9 @@
 #define MLOG_MASK_PREFIX ML_DLM
 #include "cluster/masklog.h"
 
-static spinlock_t dlm_cookie_lock = SPIN_LOCK_UNLOCKED;
+static struct kmem_cache *dlm_lock_cache;
+
+static DEFINE_SPINLOCK(dlm_cookie_lock);
 static u64 dlm_next_cookie = 1;
 
 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
@@ -64,6 +65,22 @@ static void dlm_init_lock(struct dlm_lock *newlock, int type,
 static void dlm_lock_release(struct kref *kref);
 static void dlm_lock_detach_lockres(struct dlm_lock *lock);
 
+int dlm_init_lock_cache(void)
+{
+	dlm_lock_cache = kmem_cache_create("o2dlm_lock",
+					   sizeof(struct dlm_lock),
+					   0, SLAB_HWCACHE_ALIGN, NULL);
+	if (dlm_lock_cache == NULL)
+		return -ENOMEM;
+	return 0;
+}
+
+void dlm_destroy_lock_cache(void)
+{
+	if (dlm_lock_cache)
+		kmem_cache_destroy(dlm_lock_cache);
+}
+
 /* Tell us whether we can grant a new lock request.
  * locking:
  *   caller needs:  res->spinlock
@@ -74,21 +91,19 @@ static void dlm_lock_detach_lockres(struct dlm_lock *lock);
 static int dlm_can_grant_new_lock(struct dlm_lock_resource *res,
 				  struct dlm_lock *lock)
 {
-	struct list_head *iter;
 	struct dlm_lock *tmplock;
 
-	list_for_each(iter, &res->granted) {
-		tmplock = list_entry(iter, struct dlm_lock, list);
-
+	list_for_each_entry(tmplock, &res->granted, list) {
 		if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
 			return 0;
 	}
 
-	list_for_each(iter, &res->converting) {
-		tmplock = list_entry(iter, struct dlm_lock, list);
-
+	list_for_each_entry(tmplock, &res->converting, list) {
 		if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
 			return 0;
+		if (!dlm_lock_compatible(tmplock->ml.convert_type,
+					 lock->ml.type))
+			return 0;
 	}
 
 	return 1;
@@ -108,7 +123,7 @@ static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
 	int call_ast = 0, kick_thread = 0;
 	enum dlm_status status = DLM_NORMAL;
 
-	mlog_entry("type=%d\n", lock->ml.type);
+	mlog(0, "type=%d\n", lock->ml.type);
 
 	spin_lock(&res->spinlock);
 	/* if called from dlm_create_lock_handler, need to
@@ -141,13 +156,24 @@ static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
 					  res->lockname.len)) {
 			kick_thread = 1;
 			call_ast = 1;
+		} else {
+			mlog(0, "%s: returning DLM_NORMAL to "
+			     "node %u for reco lock\n", dlm->name,
+			     lock->ml.node);
 		}
 	} else {
 		/* for NOQUEUE request, unless we get the
 		 * lock right away, return DLM_NOTQUEUED */
-		if (flags & LKM_NOQUEUE)
+		if (flags & LKM_NOQUEUE) {
 			status = DLM_NOTQUEUED;
-		else {
+			if (dlm_is_recovery_lock(res->lockname.name,
+						 res->lockname.len)) {
+				mlog(0, "%s: returning NOTQUEUED to "
+				     "node %u for reco lock\n", dlm->name,
+				     lock->ml.node);
+			}
+		} else {
+			status = DLM_NORMAL;
 			dlm_lock_get(lock);
 			list_add_tail(&lock->list, &res->blocked);
 			kick_thread = 1;
@@ -191,15 +217,22 @@ static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
 				      struct dlm_lock *lock, int flags)
 {
 	enum dlm_status status = DLM_DENIED;
+	int lockres_changed = 1;
 
-	mlog_entry("type=%d\n", lock->ml.type);
-	mlog(0, "lockres %.*s, flags = 0x%x\n", res->lockname.len,
+	mlog(0, "type=%d, lockres %.*s, flags = 0x%x\n",
+	     lock->ml.type, res->lockname.len,
 	     res->lockname.name, flags);
 
+	/*
+	 * Wait if resource is getting recovered, remastered, etc.
+	 * If the resource was remastered and new owner is self, then exit.
+	 */
 	spin_lock(&res->spinlock);
-
-	/* will exit this call with spinlock held */
 	__dlm_wait_on_lockres(res);
+	if (res->owner == dlm->node_num) {
+		spin_unlock(&res->spinlock);
+		return DLM_RECOVERING;
+	}
 	res->state |= DLM_LOCK_RES_IN_PROGRESS;
 
 	/* add lock to local (secondary) queue */
@@ -216,14 +249,42 @@ static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
 	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
 	lock->lock_pending = 0;
 	if (status != DLM_NORMAL) {
-		if (status != DLM_NOTQUEUED)
+		if (status == DLM_RECOVERING &&
+		    dlm_is_recovery_lock(res->lockname.name,
+					 res->lockname.len)) {
+			/* recovery lock was mastered by dead node.
+			 * we need to have calc_usage shoot down this
+			 * lockres and completely remaster it. */
+			mlog(0, "%s: recovery lock was owned by "
+			     "dead node %u, remaster it now.\n",
+			     dlm->name, res->owner);
+		} else if (status != DLM_NOTQUEUED) {
+			/*
+			 * DO NOT call calc_usage, as this would unhash
+			 * the remote lockres before we ever get to use
+			 * it.  treat as if we never made any change to
+			 * the lockres.
+			 */
+			lockres_changed = 0;
 			dlm_error(status);
+		}
 		dlm_revert_pending_lock(res, lock);
 		dlm_lock_put(lock);
+	} else if (dlm_is_recovery_lock(res->lockname.name,
+					res->lockname.len)) {
+		/* special case for the $RECOVERY lock.
+		 * there will never be an AST delivered to put
+		 * this lock on the proper secondary queue
+		 * (granted), so do it manually. */
+		mlog(0, "%s: $RECOVERY lock for this node (%u) is "
+		     "mastered by %u; got lock, manually granting (no ast)\n",
+		     dlm->name, dlm->node_num, res->owner);
+		list_move_tail(&lock->list, &res->granted);
 	}
 	spin_unlock(&res->spinlock);
 
-	dlm_lockres_calc_usage(dlm, res);
+	if (lockres_changed)
+		dlm_lockres_calc_usage(dlm, res);
 
 	wake_up(&res->wq);
 	return status;
@@ -245,8 +306,6 @@ static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
 	int tmpret, status = 0;
 	enum dlm_status ret;
 
-	mlog_entry_void();
-
 	memset(&create, 0, sizeof(create));
 	create.node_idx = dlm->node_num;
 	create.requested_type = lock->ml.type;
@@ -258,17 +317,23 @@ static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
 	tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create,
 				    sizeof(create), res->owner, &status);
 	if (tmpret >= 0) {
-		// successfully sent and received
-		ret = status;  // this is already a dlm_status
+		ret = status;
+		if (ret == DLM_REJECTED) {
+			mlog(ML_ERROR, "%s: res %.*s, Stale lockres no longer "
+			     "owned by node %u. That node is coming back up "
+			     "currently.\n", dlm->name, create.namelen,
+			     create.name, res->owner);
+			dlm_print_one_lock_resource(res);
+			BUG();
+		}
 	} else {
-		mlog_errno(tmpret);
-		if (dlm_is_host_down(tmpret)) {
+		mlog(ML_ERROR, "%s: res %.*s, Error %d send CREATE LOCK to "
+		     "node %u\n", dlm->name, create.namelen, create.name,
+		     tmpret, res->owner);
+		if (dlm_is_host_down(tmpret))
 			ret = DLM_RECOVERING;
-			mlog(0, "node %u died so returning DLM_RECOVERING "
-			     "from lock message!\n", res->owner);
-		} else {
+		else
 			ret = dlm_err_to_dlm_status(tmpret);
-		}
 	}
 
 	return ret;
@@ -302,7 +367,7 @@ static void dlm_lock_release(struct kref *kref)
 		mlog(0, "freeing kernel-allocated lksb\n");
 		kfree(lock->lksb);
 	}
-	kfree(lock);
+	kmem_cache_free(dlm_lock_cache, lock);
 }
 
 /* associate a lock with it's lockres, getting a ref on the lockres */
@@ -361,15 +426,15 @@ struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
 	struct dlm_lock *lock;
 	int kernel_allocated = 0;
 
-	lock = kcalloc(1, sizeof(*lock), GFP_KERNEL);
+	lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS);
 	if (!lock)
 		return NULL;
 
 	if (!lksb) {
 		/* zero memory only if kernel-allocated */
-		lksb = kcalloc(1, sizeof(*lksb), GFP_KERNEL);
+		lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
 		if (!lksb) {
-			kfree(lock);
+			kmem_cache_free(dlm_lock_cache, lock);
 			return NULL;
 		}
 		kernel_allocated = 1;
@@ -390,7 +455,8 @@ struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
  *   held on exit:  none
  * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
  */
-int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data)
+int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
+			    void **ret_data)
 {
 	struct dlm_ctxt *dlm = data;
 	struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
@@ -403,16 +469,19 @@ int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data)
 
 	BUG_ON(!dlm);
 
-	mlog_entry_void();
-
 	if (!dlm_grab(dlm))
 		return DLM_REJECTED;
 
-	mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
-			"Domain %s not fully joined!\n", dlm->name);
-
 	name = create->name;
 	namelen = create->namelen;
+	status = DLM_REJECTED;
+	if (!dlm_domain_fully_joined(dlm)) {
+		mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
+		     "sending a create_lock message for lock %.*s!\n",
+		     dlm->name, create->node_idx, namelen, name);
+		dlm_error(status);
+		goto leave;
+	}
 
 	status = DLM_IVBUFLEN;
 	if (namelen > DLM_LOCKID_NAME_MAX) {
@@ -488,8 +557,8 @@ static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
 
 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
 			struct dlm_lockstatus *lksb, int flags,
-			const char *name, dlm_astlockfunc_t *ast, void *data,
-			dlm_bastlockfunc_t *bast)
+			const char *name, int namelen, dlm_astlockfunc_t *ast,
+			void *data, dlm_bastlockfunc_t *bast)
 {
 	enum dlm_status status;
 	struct dlm_lock_resource *res = NULL;
@@ -519,7 +588,7 @@ enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
 	recovery = (flags & LKM_RECOVERY);
 
 	if (recovery &&
-	    (!dlm_is_recovery_lock(name, strlen(name)) || convert) ) {
+	    (!dlm_is_recovery_lock(name, namelen) || convert) ) {
 		dlm_error(status);
 		goto error;
 	}
@@ -591,7 +660,7 @@ retry_convert:
 		}
 
 		status = DLM_IVBUFLEN;
-		if (strlen(name) > DLM_LOCKID_NAME_MAX || strlen(name) < 1) {
+		if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
 			dlm_error(status);
 			goto error;
 		}
@@ -607,7 +676,7 @@ retry_convert:
 			dlm_wait_for_recovery(dlm);
 
 		/* find or create the lock resource */
-		res = dlm_get_lock_resource(dlm, name, flags);
+		res = dlm_get_lock_resource(dlm, name, namelen, flags);
 		if (!res) {
 			status = DLM_IVLOCKID;
 			dlm_error(status);
@@ -643,13 +712,29 @@ retry_lock:
 
 		if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
 		    status == DLM_FORWARD) {
-			mlog(0, "retrying lock with migration/"
-			     "recovery/in progress\n");
 			msleep(100);
-			dlm_wait_for_recovery(dlm);
-			goto retry_lock;
+			if (recovery) {
+				if (status != DLM_RECOVERING)
+					goto retry_lock;
+				/* wait to see the node go down, then
+				 * drop down and allow the lockres to
+				 * get cleaned up.  need to remaster. */
+				dlm_wait_for_node_death(dlm, res->owner,
+						DLM_NODE_DEATH_WAIT_MAX);
+			} else {
+				dlm_wait_for_recovery(dlm);
+				goto retry_lock;
+			}
 		}
 
+		/* Inflight taken in dlm_get_lock_resource() is dropped here */
+		spin_lock(&res->spinlock);
+		dlm_lockres_drop_inflight_ref(dlm, res);
+		spin_unlock(&res->spinlock);
+
+		dlm_lockres_calc_usage(dlm, res);
+		dlm_kick_thread(dlm, res);
+
 		if (status != DLM_NORMAL) {
 			lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
 			if (status != DLM_NOTQUEUED)