13 files changed, 240 insertions, 192 deletions

diff --git a/fs/dlm/dlm_internal.h b/fs/dlm/dlm_internal.h
index 74901e981e1..d2fc2384c3b 100644
--- a/fs/dlm/dlm_internal.h
+++ b/fs/dlm/dlm_internal.h
@@ -491,6 +491,7 @@ 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 */
 	struct list_head	ls_requestqueue;/* queue remote requests */
 	struct mutex		ls_requestqueue_mutex;
 	char			*ls_recover_buf;
diff --git a/fs/dlm/lock.c b/fs/dlm/lock.c
index b455919c199..3915b8e1414 100644
--- a/fs/dlm/lock.c
+++ b/fs/dlm/lock.c
@@ -1670,9 +1670,10 @@ static int can_be_granted(struct dlm_rsb *r, struct dlm_lkb *lkb, int now,
    with a deadlk here, we'd have to generate something like grant_lock with
    the deadlk error.) */
 
-/* returns the highest requested mode of all blocked conversions */
+/* Returns the highest requested mode of all blocked conversions; sets
+   cw if there's a blocked conversion to DLM_LOCK_CW. */
 
-static int grant_pending_convert(struct dlm_rsb *r, int high)
+static int grant_pending_convert(struct dlm_rsb *r, int high, int *cw)
 {
 	struct dlm_lkb *lkb, *s;
 	int hi, demoted, quit, grant_restart, demote_restart;
@@ -1709,6 +1710,9 @@ static int grant_pending_convert(struct dlm_rsb *r, int high)
 		}
 
 		hi = max_t(int, lkb->lkb_rqmode, hi);
+
+		if (cw && lkb->lkb_rqmode == DLM_LOCK_CW)
+			*cw = 1;
 	}
 
 	if (grant_restart)
@@ -1721,29 +1725,52 @@ static int grant_pending_convert(struct dlm_rsb *r, int high)
 	return max_t(int, high, hi);
 }
 
-static int grant_pending_wait(struct dlm_rsb *r, int high)
+static int grant_pending_wait(struct dlm_rsb *r, int high, int *cw)
 {
 	struct dlm_lkb *lkb, *s;
 
 	list_for_each_entry_safe(lkb, s, &r->res_waitqueue, lkb_statequeue) {
 		if (can_be_granted(r, lkb, 0, NULL))
 			grant_lock_pending(r, lkb);
-                else
+                else {
 			high = max_t(int, lkb->lkb_rqmode, high);
+			if (lkb->lkb_rqmode == DLM_LOCK_CW)
+				*cw = 1;
+		}
 	}
 
 	return high;
 }
 
+/* cw of 1 means there's a lock with a rqmode of DLM_LOCK_CW that's blocked
+   on either the convert or waiting queue.
+   high is the largest rqmode of all locks blocked on the convert or
+   waiting queue. */
+
+static int lock_requires_bast(struct dlm_lkb *gr, int high, int cw)
+{
+	if (gr->lkb_grmode == DLM_LOCK_PR && cw) {
+		if (gr->lkb_highbast < DLM_LOCK_EX)
+			return 1;
+		return 0;
+	}
+
+	if (gr->lkb_highbast < high &&
+	    !__dlm_compat_matrix[gr->lkb_grmode+1][high+1])
+		return 1;
+	return 0;
+}
+
 static void grant_pending_locks(struct dlm_rsb *r)
 {
 	struct dlm_lkb *lkb, *s;
 	int high = DLM_LOCK_IV;
+	int cw = 0;
 
 	DLM_ASSERT(is_master(r), dlm_dump_rsb(r););
 
-	high = grant_pending_convert(r, high);
-	high = grant_pending_wait(r, high);
+	high = grant_pending_convert(r, high, &cw);
+	high = grant_pending_wait(r, high, &cw);
 
 	if (high == DLM_LOCK_IV)
 		return;
@@ -1751,27 +1778,41 @@ static void grant_pending_locks(struct dlm_rsb *r)
 	/*
 	 * If there are locks left on the wait/convert queue then send blocking
 	 * ASTs to granted locks based on the largest requested mode (high)
-	 * found above. FIXME: highbast < high comparison not valid for PR/CW.
+	 * found above.
 	 */
 
 	list_for_each_entry_safe(lkb, s, &r->res_grantqueue, lkb_statequeue) {
-		if (lkb->lkb_bastaddr && (lkb->lkb_highbast < high) &&
-		    !__dlm_compat_matrix[lkb->lkb_grmode+1][high+1]) {
-			queue_bast(r, lkb, high);
+		if (lkb->lkb_bastaddr && lock_requires_bast(lkb, high, cw)) {
+			if (cw && high == DLM_LOCK_PR)
+				queue_bast(r, lkb, DLM_LOCK_CW);
+			else
+				queue_bast(r, lkb, high);
 			lkb->lkb_highbast = high;
 		}
 	}
 }
 
+static int modes_require_bast(struct dlm_lkb *gr, struct dlm_lkb *rq)
+{
+	if ((gr->lkb_grmode == DLM_LOCK_PR && rq->lkb_rqmode == DLM_LOCK_CW) ||
+	    (gr->lkb_grmode == DLM_LOCK_CW && rq->lkb_rqmode == DLM_LOCK_PR)) {
+		if (gr->lkb_highbast < DLM_LOCK_EX)
+			return 1;
+		return 0;
+	}
+
+	if (gr->lkb_highbast < rq->lkb_rqmode && !modes_compat(gr, rq))
+		return 1;
+	return 0;
+}
+
 static void send_bast_queue(struct dlm_rsb *r, struct list_head *head,
 			    struct dlm_lkb *lkb)
 {
 	struct dlm_lkb *gr;
 
 	list_for_each_entry(gr, head, lkb_statequeue) {
-		if (gr->lkb_bastaddr &&
-		    gr->lkb_highbast < lkb->lkb_rqmode &&
-		    !modes_compat(gr, lkb)) {
+		if (gr->lkb_bastaddr && modes_require_bast(gr, lkb)) {
 			queue_bast(r, gr, lkb->lkb_rqmode);
 			gr->lkb_highbast = lkb->lkb_rqmode;
 		}
@@ -2235,7 +2276,7 @@ static int do_convert(struct dlm_rsb *r, struct dlm_lkb *lkb)
 	   before we try again to grant this one. */
 
 	if (is_demoted(lkb)) {
-		grant_pending_convert(r, DLM_LOCK_IV);
+		grant_pending_convert(r, DLM_LOCK_IV, NULL);
 		if (_can_be_granted(r, lkb, 1)) {
 			grant_lock(r, lkb);
 			queue_cast(r, lkb, 0);
@@ -3597,55 +3638,8 @@ static void receive_lookup_reply(struct dlm_ls *ls, struct dlm_message *ms)
 	dlm_put_lkb(lkb);
 }
 
-int dlm_receive_message(struct dlm_header *hd, int nodeid, int recovery)
+static void _receive_message(struct dlm_ls *ls, struct dlm_message *ms)
 {
-	struct dlm_message *ms = (struct dlm_message *) hd;
-	struct dlm_ls *ls;
-	int error = 0;
-
-	if (!recovery)
-		dlm_message_in(ms);
-
-	ls = dlm_find_lockspace_global(hd->h_lockspace);
-	if (!ls) {
-		log_print("drop message %d from %d for unknown lockspace %d",
-			  ms->m_type, nodeid, hd->h_lockspace);
-		return -EINVAL;
-	}
-
-	/* recovery may have just ended leaving a bunch of backed-up requests
-	   in the requestqueue; wait while dlm_recoverd clears them */
-
-	if (!recovery)
-		dlm_wait_requestqueue(ls);
-
-	/* recovery may have just started while there were a bunch of
-	   in-flight requests -- save them in requestqueue to be processed
-	   after recovery.  we can't let dlm_recvd block on the recovery
-	   lock.  if dlm_recoverd is calling this function to clear the
-	   requestqueue, it needs to be interrupted (-EINTR) if another
-	   recovery operation is starting. */
-
-	while (1) {
-		if (dlm_locking_stopped(ls)) {
-			if (recovery) {
-				error = -EINTR;
-				goto out;
-			}
-			error = dlm_add_requestqueue(ls, nodeid, hd);
-			if (error == -EAGAIN)
-				continue;
-			else {
-				error = -EINTR;
-				goto out;
-			}
-		}
-
-		if (dlm_lock_recovery_try(ls))
-			break;
-		schedule();
-	}
-
 	switch (ms->m_type) {
 
 	/* messages sent to a master node */
@@ -3720,17 +3714,90 @@ int dlm_receive_message(struct dlm_header *hd, int nodeid, int recovery)
 		log_error(ls, "unknown message type %d", ms->m_type);
 	}
 
-	dlm_unlock_recovery(ls);
- out:
-	dlm_put_lockspace(ls);
 	dlm_astd_wake();
-	return error;
 }
 
+/* If the lockspace is in recovery mode (locking stopped), then normal
+   messages are saved on the requestqueue for processing after recovery is
+   done.  When not in recovery mode, we wait for dlm_recoverd to drain saved
+   messages off the requestqueue before we process new ones. This occurs right
+   after recovery completes when we transition from saving all messages on
+   requestqueue, to processing all the saved messages, to processing new
+   messages as they arrive. */
 
-/*
- * Recovery related
- */
+static void dlm_receive_message(struct dlm_ls *ls, struct dlm_message *ms,
+				int nodeid)
+{
+	if (dlm_locking_stopped(ls)) {
+		dlm_add_requestqueue(ls, nodeid, (struct dlm_header *) ms);
+	} else {
+		dlm_wait_requestqueue(ls);
+		_receive_message(ls, ms);
+	}
+}
+
+/* This is called by dlm_recoverd to process messages that were saved on
+   the requestqueue. */
+
+void dlm_receive_message_saved(struct dlm_ls *ls, struct dlm_message *ms)
+{
+	_receive_message(ls, ms);
+}
+
+/* This is called by the midcomms layer when something is received for
+   the lockspace.  It could be either a MSG (normal message sent as part of
+   standard locking activity) or an RCOM (recovery message sent as part of
+   lockspace recovery). */
+
+void dlm_receive_buffer(struct dlm_header *hd, int nodeid)
+{
+	struct dlm_message *ms = (struct dlm_message *) hd;
+	struct dlm_rcom *rc = (struct dlm_rcom *) hd;
+	struct dlm_ls *ls;
+	int type = 0;
+
+	switch (hd->h_cmd) {
+	case DLM_MSG:
+		dlm_message_in(ms);
+		type = ms->m_type;
+		break;
+	case DLM_RCOM:
+		dlm_rcom_in(rc);
+		type = rc->rc_type;
+		break;
+	default:
+		log_print("invalid h_cmd %d from %u", hd->h_cmd, nodeid);
+		return;
+	}
+
+	if (hd->h_nodeid != nodeid) {
+		log_print("invalid h_nodeid %d from %d lockspace %x",
+			  hd->h_nodeid, nodeid, hd->h_lockspace);
+		return;
+	}
+
+	ls = dlm_find_lockspace_global(hd->h_lockspace);
+	if (!ls) {
+		log_print("invalid h_lockspace %x from %d cmd %d type %d",
+			  hd->h_lockspace, nodeid, hd->h_cmd, type);
+
+		if (hd->h_cmd == DLM_RCOM && type == DLM_RCOM_STATUS)
+			dlm_send_ls_not_ready(nodeid, rc);
+		return;
+	}
+
+	/* 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);
+	if (hd->h_cmd == DLM_MSG)
+		dlm_receive_message(ls, ms, nodeid);
+	else
+		dlm_receive_rcom(ls, rc, nodeid);
+	up_read(&ls->ls_recv_active);
+
+	dlm_put_lockspace(ls);
+}
 
 static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb)
 {
@@ -4388,7 +4455,8 @@ int dlm_user_unlock(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
 
 	if (lvb_in && ua->lksb.sb_lvbptr)
 		memcpy(ua->lksb.sb_lvbptr, lvb_in, DLM_USER_LVB_LEN);
-	ua->castparam = ua_tmp->castparam;
+	if (ua_tmp->castparam)
+		ua->castparam = ua_tmp->castparam;
 	ua->user_lksb = ua_tmp->user_lksb;
 
 	error = set_unlock_args(flags, ua, &args);
@@ -4433,7 +4501,8 @@ int dlm_user_cancel(struct dlm_ls *ls, struct dlm_user_args *ua_tmp,
 		goto out;
 
 	ua = (struct dlm_user_args *)lkb->lkb_astparam;
-	ua->castparam = ua_tmp->castparam;
+	if (ua_tmp->castparam)
+		ua->castparam = ua_tmp->castparam;
 	ua->user_lksb = ua_tmp->user_lksb;
 
 	error = set_unlock_args(flags, ua, &args);
diff --git a/fs/dlm/lock.h b/fs/dlm/lock.h
index 1720313c22d..ada04680a1e 100644
--- a/fs/dlm/lock.h
+++ b/fs/dlm/lock.h
@@ -16,7 +16,8 @@
 void dlm_print_rsb(struct dlm_rsb *r);
 void dlm_dump_rsb(struct dlm_rsb *r);
 void dlm_print_lkb(struct dlm_lkb *lkb);
-int dlm_receive_message(struct dlm_header *hd, int nodeid, int recovery);
+void dlm_receive_message_saved(struct dlm_ls *ls, struct dlm_message *ms);
+void dlm_receive_buffer(struct dlm_header *hd, int nodeid);
 int dlm_modes_compat(int mode1, int mode2);
 int dlm_find_rsb(struct dlm_ls *ls, char *name, int namelen,
 	unsigned int flags, struct dlm_rsb **r_ret);
diff --git a/fs/dlm/lockspace.c b/fs/dlm/lockspace.c
index 1dc72105ab1..6353a838452 100644
--- a/fs/dlm/lockspace.c
+++ b/fs/dlm/lockspace.c
@@ -167,7 +167,6 @@ static struct kobj_type dlm_ktype = {
 };
 
 static struct kset dlm_kset = {
-	.kobj   = {.name = "dlm",},
 	.ktype  = &dlm_ktype,
 };
 
@@ -228,6 +227,7 @@ int dlm_lockspace_init(void)
 	INIT_LIST_HEAD(&lslist);
 	spin_lock_init(&lslist_lock);
 
+	kobject_set_name(&dlm_kset.kobj, "dlm");
 	kobj_set_kset_s(&dlm_kset, kernel_subsys);
 	error = kset_register(&dlm_kset);
 	if (error)
@@ -519,6 +519,7 @@ static int new_lockspace(char *name, int namelen, void **lockspace,
 	ls->ls_recover_seq = 0;
 	ls->ls_recover_args = NULL;
 	init_rwsem(&ls->ls_in_recovery);
+	init_rwsem(&ls->ls_recv_active);
 	INIT_LIST_HEAD(&ls->ls_requestqueue);
 	mutex_init(&ls->ls_requestqueue_mutex);
 	mutex_init(&ls->ls_clear_proc_locks);
diff --git a/fs/dlm/lowcomms.c b/fs/dlm/lowcomms.c
index 0553a6158dc..58bf3f5cdbe 100644
--- a/fs/dlm/lowcomms.c
+++ b/fs/dlm/lowcomms.c
@@ -313,6 +313,7 @@ static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
 		in6_addr->sin6_port = cpu_to_be16(port);
 		*addr_len = sizeof(struct sockaddr_in6);
 	}
+	memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
 }
 
 /* Close a remote connection and tidy up */
@@ -332,6 +333,7 @@ static void close_connection(struct connection *con, bool and_other)
 		__free_page(con->rx_page);
 		con->rx_page = NULL;
 	}
+
 	con->retries = 0;
 	mutex_unlock(&con->sock_mutex);
 }
@@ -631,7 +633,7 @@ out_resched:
 
 out_close:
 	mutex_unlock(&con->sock_mutex);
-	if (ret != -EAGAIN && !test_bit(CF_IS_OTHERCON, &con->flags)) {
+	if (ret != -EAGAIN) {
 		close_connection(con, false);
 		/* Reconnect when there is something to send */
 	}
@@ -719,6 +721,8 @@ static int tcp_accept_from_sock(struct connection *con)
 			INIT_WORK(&othercon->swork, process_send_sockets);
 			INIT_WORK(&othercon->rwork, process_recv_sockets);
 			set_bit(CF_IS_OTHERCON, &othercon->flags);
+		}
+		if (!othercon->sock) {
 			newcon->othercon = othercon;
 			othercon->sock = newsock;
 			newsock->sk->sk_user_data = othercon;
@@ -1122,8 +1126,6 @@ static int tcp_listen_for_all(void)
 
 	log_print("Using TCP for communications");
 
-	set_bit(CF_IS_OTHERCON, &con->flags);
-
 	sock = tcp_create_listen_sock(con, dlm_local_addr[0]);
 	if (sock) {
 		add_sock(sock, con);
@@ -1262,14 +1264,15 @@ static void send_to_sock(struct connection *con)
 		if (len) {
 			ret = sendpage(con->sock, e->page, offset, len,
 				       msg_flags);
-			if (ret == -EAGAIN || ret == 0)
+			if (ret == -EAGAIN || ret == 0) {
+				cond_resched();
 				goto out;
+			}
 			if (ret <= 0)
 				goto send_error;
-		} else {
+		}
 			/* Don't starve people filling buffers */
 			cond_resched();
-		}
 
 		spin_lock(&con->writequeue_lock);
 		e->offset += ret;
@@ -1407,7 +1410,7 @@ void dlm_lowcomms_stop(void)
 	for (i = 0; i <= max_nodeid; i++) {
 		con = __nodeid2con(i, 0);
 		if (con) {
-			con->flags |= 0xFF;
+			con->flags |= 0x0F;
 			if (con->sock)
 				con->sock->sk->sk_user_data = NULL;
 		}
@@ -1423,8 +1426,6 @@ void dlm_lowcomms_stop(void)
 		con = __nodeid2con(i, 0);
 		if (con) {
 			close_connection(con, true);
-			if (con->othercon)
-				kmem_cache_free(con_cache, con->othercon);
 			kmem_cache_free(con_cache, con);
 		}
 	}
@@ -1449,7 +1450,7 @@ int dlm_lowcomms_start(void)
 	error = -ENOMEM;
 	con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection),
 				      __alignof__(struct connection), 0,
-				      NULL, NULL);
+				      NULL);
 	if (!con_cache)
 		goto out;
 
diff --git a/fs/dlm/member.c b/fs/dlm/member.c
index 073599dced2..e9cdcab306e 100644
--- a/fs/dlm/member.c
+++ b/fs/dlm/member.c
@@ -18,10 +18,6 @@
 #include "rcom.h"
 #include "config.h"
 
-/*
- * Following called by dlm_recoverd thread
- */
-
 static void add_ordered_member(struct dlm_ls *ls, struct dlm_member *new)
 {
 	struct dlm_member *memb = NULL;
@@ -56,8 +52,10 @@ static int dlm_add_member(struct dlm_ls *ls, int nodeid)
 		return -ENOMEM;
 
 	w = dlm_node_weight(ls->ls_name, nodeid);
-	if (w < 0)
+	if (w < 0) {
+		kfree(memb);
 		return w;
+	}
 
 	memb->nodeid = nodeid;
 	memb->weight = w;
@@ -248,18 +246,30 @@ int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
 	return error;
 }
 
-/*
- * Following called from lockspace.c
- */
+/* Userspace guarantees that dlm_ls_stop() has completed on all nodes before
+   dlm_ls_start() is called on any of them to start the new recovery. */
 
 int dlm_ls_stop(struct dlm_ls *ls)
 {
 	int new;
 
 	/*
-	 * A stop cancels any recovery that's in progress (see RECOVERY_STOP,
-	 * dlm_recovery_stopped()) and prevents any new locks from being
-	 * processed (see RUNNING, dlm_locking_stopped()).
+	 * Prevent dlm_recv from being in the middle of something when we do
+	 * the stop.  This includes ensuring dlm_recv isn't processing a
+	 * recovery message (rcom), while dlm_recoverd is aborting and
+	 * resetting things from an in-progress recovery.  i.e. we want
+	 * dlm_recoverd to abort its recovery without worrying about dlm_recv
+	 * processing an rcom at the same time.  Stopping dlm_recv also makes
+	 * it easy for dlm_receive_message() to check locking stopped and add a
+	 * message to the requestqueue without races.
+	 */
+
+	down_write(&ls->ls_recv_active);
+
+	/*
+	 * Abort any recovery that's in progress (see RECOVERY_STOP,
+	 * dlm_recovery_stopped()) and tell any other threads running in the
+	 * dlm to quit any processing (see RUNNING, dlm_locking_stopped()).
 	 */
 
 	spin_lock(&ls->ls_recover_lock);
@@ -269,8 +279,14 @@ int dlm_ls_stop(struct dlm_ls *ls)
 	spin_unlock(&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);
+
+	/*
 	 * This in_recovery lock does two things:
-	 *
 	 * 1) Keeps this function from returning until all threads are out
 	 *    of locking routines and locking is truely stopped.
 	 * 2) Keeps any new requests from being processed until it's unlocked
@@ -282,9 +298,8 @@ int dlm_ls_stop(struct dlm_ls *ls)
 
 	/*
 	 * The recoverd suspend/resume makes sure that dlm_recoverd (if
-	 * running) has noticed the clearing of RUNNING above and quit
-	 * processing the previous recovery.  This will be true for all nodes
-	 * before any nodes start the new recovery.
+	 * running) has noticed RECOVERY_STOP above and quit processing the
+	 * previous recovery.
 	 */
 
 	dlm_recoverd_suspend(ls);
diff --git a/fs/dlm/memory.c b/fs/dlm/memory.c
index f858fef6e41..ecf0e5cb203 100644
--- a/fs/dlm/memory.c
+++ b/fs/dlm/memory.c
@@ -23,7 +23,7 @@ int dlm_memory_init(void)
 	int ret = 0;
 
 	lkb_cache = kmem_cache_create("dlm_lkb", sizeof(struct dlm_lkb),
-				__alignof__(struct dlm_lkb), 0, NULL, NULL);
+				__alignof__(struct dlm_lkb), 0, NULL);
 	if (!lkb_cache)
 		ret = -ENOMEM;
 	return ret;
@@ -39,9 +39,7 @@ char *allocate_lvb(struct dlm_ls *ls)
 {
 	char *p;
 
-	p = kmalloc(ls->ls_lvblen, GFP_KERNEL);
-	if (p)
-		memset(p, 0, ls->ls_lvblen);
+	p = kzalloc(ls->ls_lvblen, GFP_KERNEL);
 	return p;
 }
 
@@ -59,9 +57,7 @@ struct dlm_rsb *allocate_rsb(struct dlm_ls *ls, int namelen)
 
 	DLM_ASSERT(namelen <= DLM_RESNAME_MAXLEN,);
 
-	r = kmalloc(sizeof(*r) + namelen, GFP_KERNEL);
-	if (r)
-		memset(r, 0, sizeof(*r) + namelen);
+	r = kzalloc(sizeof(*r) + namelen, GFP_KERNEL);
 	return r;
 }
 
@@ -101,9 +97,7 @@ struct dlm_direntry *allocate_direntry(struct dlm_ls *ls, int namelen)
 	DLM_ASSERT(namelen <= DLM_RESNAME_MAXLEN,
 		   printk("namelen = %d\n", namelen););
 
-	de = kmalloc(sizeof(*de) + namelen, GFP_KERNEL);
-	if (de)
-		memset(de, 0, sizeof(*de) + namelen);
+	de = kzalloc(sizeof(*de) + namelen, GFP_KERNEL);
 	return de;
 }
 
diff --git a/fs/dlm/midcomms.c b/fs/dlm/midcomms.c
index a5126e0c68a..f8c69dda16a 100644
--- a/fs/dlm/midcomms.c
+++ b/fs/dlm/midcomms.c
@@ -2,7 +2,7 @@
 *******************************************************************************
 **
 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
 **
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -27,7 +27,6 @@
 #include "dlm_internal.h"
 #include "lowcomms.h"
 #include "config.h"
-#include "rcom.h"
 #include "lock.h"
 #include "midcomms.h"
 
@@ -117,19 +116,7 @@ int dlm_process_incoming_buffer(int nodeid, const void *base,
 		offset &= (limit - 1);
 		len -= msglen;
 
-		switch (msg->h_cmd) {
-		case DLM_MSG:
-			dlm_receive_message(msg, nodeid, 0);
-			break;
-
-		case DLM_RCOM:
-			dlm_receive_rcom(msg, nodeid);
-			break;
-
-		default:
-			log_print("unknown msg type %x from %u: %u %u %u %u",
-				  msg->h_cmd, nodeid, msglen, len, offset, ret);
-		}
+		dlm_receive_buffer(msg, nodeid);
 	}
 
 	if (msg != (struct dlm_header *) __tmp)
diff --git a/fs/dlm/rcom.c b/fs/dlm/rcom.c
index e3a1527cbdb..ae2fd97fa4a 100644
--- a/fs/dlm/rcom.c
+++ b/fs/dlm/rcom.c
@@ -2,7 +2,7 @@
 *******************************************************************************
 **
 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2005 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2005-2007 Red Hat, Inc.  All rights reserved.
 **
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -386,8 +386,10 @@ static void receive_rcom_lock_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
 	dlm_recover_process_copy(ls, rc_in);
 }
 
-static int send_ls_not_ready(struct dlm_ls *ls, int nodeid,
-			     struct dlm_rcom *rc_in)
+/* If the lockspace doesn't exist then still send a status message
+   back; it's possible that it just doesn't have its global_id yet. */
+
+int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in)
 {
 	struct dlm_rcom *rc;
 	struct rcom_config *rf;
@@ -395,7 +397,7 @@ static int send_ls_not_ready(struct dlm_ls *ls, int nodeid,
 	char *mb;
 	int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
 
-	mh = dlm_lowcomms_get_buffer(nodeid, mb_len, ls->ls_allocation, &mb);
+	mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb);
 	if (!mh)
 		return -ENOBUFS;
 	memset(mb, 0, mb_len);
@@ -447,28 +449,11 @@ static int is_old_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
 	return rv;
 }
 
-/* Called by dlm_recvd; corresponds to dlm_receive_message() but special
+/* Called by dlm_recv; corresponds to dlm_receive_message() but special
    recovery-only comms are sent through here. */
 
-void dlm_receive_rcom(struct dlm_header *hd, int nodeid)
+void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
 {
-	struct dlm_rcom *rc = (struct dlm_rcom *) hd;
-	struct dlm_ls *ls;
-
-	dlm_rcom_in(rc);
-
-	/* If the lockspace doesn't exist then still send a status message
-	   back; it's possible that it just doesn't have its global_id yet. */
-
-	ls = dlm_find_lockspace_global(hd->h_lockspace);
-	if (!ls) {
-		log_print("lockspace %x from %d type %x not found",
-			  hd->h_lockspace, nodeid, rc->rc_type);
-		if (rc->rc_type == DLM_RCOM_STATUS)
-			send_ls_not_ready(ls, nodeid, rc);
-		return;
-	}
-
 	if (dlm_recovery_stopped(ls) && (rc->rc_type != DLM_RCOM_STATUS)) {
 		log_debug(ls, "ignoring recovery message %x from %d",
 			  rc->rc_type, nodeid);
@@ -478,12 +463,6 @@ void dlm_receive_rcom(struct dlm_header *hd, int nodeid)
 	if (is_old_reply(ls, rc))
 		goto out;
 
-	if (nodeid != rc->rc_header.h_nodeid) {
-		log_error(ls, "bad rcom nodeid %d from %d",
-			  rc->rc_header.h_nodeid, nodeid);
-		goto out;
-	}
-
 	switch (rc->rc_type) {
 	case DLM_RCOM_STATUS:
 		receive_rcom_status(ls, rc);
@@ -521,6 +500,6 @@ void dlm_receive_rcom(struct dlm_header *hd, int nodeid)
 		DLM_ASSERT(0, printk("rc_type=%x\n", rc->rc_type););
 	}
  out:
-	dlm_put_lockspace(ls);
+	return;
 }
 
diff --git a/fs/dlm/rcom.h b/fs/dlm/rcom.h
index d7984321ff4..b09abd29ba3 100644
--- a/fs/dlm/rcom.h
+++ b/fs/dlm/rcom.h
@@ -2,7 +2,7 @@
 *******************************************************************************
 **
 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2005 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2005-2007 Red Hat, Inc.  All rights reserved.
 **
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -18,7 +18,8 @@ int dlm_rcom_status(struct dlm_ls *ls, int nodeid);
 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name,int last_len);
 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid);
 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb);
-void dlm_receive_rcom(struct dlm_header *hd, int nodeid);
+void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid);
+int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in);
 
 #endif
 
diff --git a/fs/dlm/recoverd.c b/fs/dlm/recoverd.c
index 66575997861..4b89e20eebe 100644
--- a/fs/dlm/recoverd.c
+++ b/fs/dlm/recoverd.c
@@ -24,19 +24,28 @@
 
 
 /* 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. */
+   by a newer stop (ls_recover_seq), we need to leave locking disabled.
+
+   We suspend dlm_recv threads here to avoid the race where dlm_recv a) sees
+   locking stopped and b) adds a message to the requestqueue, but dlm_recoverd
+   enables locking and clears the requestqueue between a and b. */
 
 static int enable_locking(struct dlm_ls *ls, uint64_t seq)
 {
 	int error = -EINTR;
 
+	down_write(&ls->ls_recv_active);
+
 	spin_lock(&ls->ls_recover_lock);
 	if (ls->ls_recover_seq == seq) {
 		set_bit(LSFL_RUNNING, &ls->ls_flags);
+		/* unblocks processes waiting to enter the dlm */
 		up_write(&ls->ls_in_recovery);
 		error = 0;
 	}
 	spin_unlock(&ls->ls_recover_lock);
+
+	up_write(&ls->ls_recv_active);
 	return error;
 }
 
diff --git a/fs/dlm/requestqueue.c b/fs/dlm/requestqueue.c
index 65008d79c96..0de04f17cce 100644
--- a/fs/dlm/requestqueue.c
+++ b/fs/dlm/requestqueue.c
@@ -1,7 +1,7 @@
 /******************************************************************************
 *******************************************************************************
 **
-**  Copyright (C) 2005 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2005-2007 Red Hat, Inc.  All rights reserved.
 **
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -20,7 +20,7 @@
 struct rq_entry {
 	struct list_head list;
 	int nodeid;
-	char request[1];
+	char request[0];
 };
 
 /*
@@ -30,42 +30,39 @@ struct rq_entry {
  * lockspace is enabled on some while still suspended on others.
  */
 
-int dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_header *hd)
+void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_header *hd)
 {
 	struct rq_entry *e;
 	int length = hd->h_length;
-	int rv = 0;
 
 	e = kmalloc(sizeof(struct rq_entry) + length, GFP_KERNEL);
 	if (!e) {
-		log_print("dlm_add_requestqueue: out of memory\n");
-		return 0;
+		log_print("dlm_add_requestqueue: out of memory len %d", length);
+		return;
 	}
 
 	e->nodeid = nodeid;
 	memcpy(e->request, hd, length);
 
-	/* We need to check dlm_locking_stopped() after taking the mutex to
-	   avoid a race where dlm_recoverd enables locking and runs
-	   process_requestqueue between our earlier dlm_locking_stopped check
-	   and this addition to the requestqueue. */
-
 	mutex_lock(&ls->ls_requestqueue_mutex);
-	if (dlm_locking_stopped(ls))
-		list_add_tail(&e->list, &ls->ls_requestqueue);
-	else {
-		log_debug(ls, "dlm_add_requestqueue skip from %d", nodeid);
-		kfree(e);
-		rv = -EAGAIN;
-	}
+	list_add_tail(&e->list, &ls->ls_requestqueue);
 	mutex_unlock(&ls->ls_requestqueue_mutex);
-	return rv;
 }
 
+/*
+ * Called by dlm_recoverd to process normal messages saved while recovery was
+ * happening.  Normal locking has been enabled before this is called.  dlm_recv
+ * upon receiving a message, will wait for all saved messages to be drained
+ * here before processing the message it got.  If a new dlm_ls_stop() arrives
+ * while we're processing these saved messages, it may block trying to suspend
+ * dlm_recv if dlm_recv is waiting for us in dlm_wait_requestqueue.  In that
+ * case, we don't abort since locking_stopped is still 0.  If dlm_recv is not
+ * waiting for us, then this processing may be aborted due to locking_stopped.
+ */
+
 int dlm_process_requestqueue(struct dlm_ls *ls)
 {
 	struct rq_entry *e;
-	struct dlm_header *hd;
 	int error = 0;
 
 	mutex_lock(&ls->ls_requestqueue_mutex);
@@ -79,14 +76,7 @@ int dlm_process_requestqueue(struct dlm_ls *ls)
 		e = list_entry(ls->ls_requestqueue.next, struct rq_entry, list);
 		mutex_unlock(&ls->ls_requestqueue_mutex);
 
-		hd = (struct dlm_header *) e->request;
-		error = dlm_receive_message(hd, e->nodeid, 1);
-
-		if (error == -EINTR) {
-			/* entry is left on requestqueue */
-			log_debug(ls, "process_requestqueue abort eintr");
-			break;
-		}
+		dlm_receive_message_saved(ls, (struct dlm_message *)e->request);
 
 		mutex_lock(&ls->ls_requestqueue_mutex);
 		list_del(&e->list);
@@ -106,10 +96,12 @@ int dlm_process_requestqueue(struct dlm_ls *ls)
 
 /*
  * 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_recvd.  At
- * the same time, dlm_recvd will start receiving new requests from remote
- * nodes.  We want to delay dlm_recvd processing new requests until
- * dlm_recoverd has finished processing the old saved requests.
+ * 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)
@@ -118,8 +110,6 @@ void dlm_wait_requestqueue(struct dlm_ls *ls)
 		mutex_lock(&ls->ls_requestqueue_mutex);
 		if (list_empty(&ls->ls_requestqueue))
 			break;
-		if (dlm_locking_stopped(ls))
-			break;
 		mutex_unlock(&ls->ls_requestqueue_mutex);
 		schedule();
 	}
diff --git a/fs/dlm/requestqueue.h b/fs/dlm/requestqueue.h
index 6a53ea03335..aba34fc05ee 100644
--- a/fs/dlm/requestqueue.h
+++ b/fs/dlm/requestqueue.h
@@ -1,7 +1,7 @@
 /******************************************************************************
 *******************************************************************************
 **
-**  Copyright (C) 2005 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2005-2007 Red Hat, Inc.  All rights reserved.
 **
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -13,7 +13,7 @@
 #ifndef __REQUESTQUEUE_DOT_H__
 #define __REQUESTQUEUE_DOT_H__
 
-int dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_header *hd);
+void dlm_add_requestqueue(struct dlm_ls *ls, int nodeid, struct dlm_header *hd);
 int dlm_process_requestqueue(struct dlm_ls *ls);
 void dlm_wait_requestqueue(struct dlm_ls *ls);
 void dlm_purge_requestqueue(struct dlm_ls *ls);