Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/dlm

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/dlm:
dlm: make plock operation killable
dlm: remove shared message stub for recovery
dlm: delayed reply message warning
dlm: Remove superfluous call to recalc_sigpending()

fs/dlm/config.c

···
 	unsigned int cl_log_debug;
 	unsigned int cl_protocol;
 	unsigned int cl_timewarn_cs;
+	unsigned int cl_waitwarn_us;
 };
 
 enum {
···
 	CLUSTER_ATTR_LOG_DEBUG,
 	CLUSTER_ATTR_PROTOCOL,
 	CLUSTER_ATTR_TIMEWARN_CS,
+	CLUSTER_ATTR_WAITWARN_US,
 };
 
 struct cluster_attribute {
···
 CLUSTER_ATTR(log_debug, 0);
 CLUSTER_ATTR(protocol, 0);
 CLUSTER_ATTR(timewarn_cs, 1);
+CLUSTER_ATTR(waitwarn_us, 0);
 
 static struct configfs_attribute *cluster_attrs[] = {
 	[CLUSTER_ATTR_TCP_PORT] = &cluster_attr_tcp_port.attr,
···
 	[CLUSTER_ATTR_LOG_DEBUG] = &cluster_attr_log_debug.attr,
 	[CLUSTER_ATTR_PROTOCOL] = &cluster_attr_protocol.attr,
 	[CLUSTER_ATTR_TIMEWARN_CS] = &cluster_attr_timewarn_cs.attr,
+	[CLUSTER_ATTR_WAITWARN_US] = &cluster_attr_waitwarn_us.attr,
 	NULL,
 };
 
···
 	cl->cl_log_debug = dlm_config.ci_log_debug;
 	cl->cl_protocol = dlm_config.ci_protocol;
 	cl->cl_timewarn_cs = dlm_config.ci_timewarn_cs;
+	cl->cl_waitwarn_us = dlm_config.ci_waitwarn_us;
 
 	space_list = &sps->ss_group;
 	comm_list = &cms->cs_group;
···
 #define DEFAULT_LOG_DEBUG          0
 #define DEFAULT_PROTOCOL           0
 #define DEFAULT_TIMEWARN_CS      500 /* 5 sec = 500 centiseconds */
+#define DEFAULT_WAITWARN_US	   0
 
 struct dlm_config_info dlm_config = {
 	.ci_tcp_port = DEFAULT_TCP_PORT,
···
 	.ci_scan_secs = DEFAULT_SCAN_SECS,
 	.ci_log_debug = DEFAULT_LOG_DEBUG,
 	.ci_protocol = DEFAULT_PROTOCOL,
-	.ci_timewarn_cs = DEFAULT_TIMEWARN_CS
+	.ci_timewarn_cs = DEFAULT_TIMEWARN_CS,
+	.ci_waitwarn_us = DEFAULT_WAITWARN_US
 };
 

fs/dlm/config.h

···
 	int ci_log_debug;
 	int ci_protocol;
 	int ci_timewarn_cs;
+	int ci_waitwarn_us;
 };
 
 extern struct dlm_config_info dlm_config;

fs/dlm/dlm_internal.h

···
 #define DLM_IFL_WATCH_TIMEWARN	0x00400000
 #define DLM_IFL_TIMEOUT_CANCEL	0x00800000
 #define DLM_IFL_DEADLOCK_CANCEL	0x01000000
+#define DLM_IFL_STUB_MS		0x02000000 /* magic number for m_flags */
 #define DLM_IFL_USER		0x00000001
 #define DLM_IFL_ORPHAN		0x00000002
 
···
 
 	int8_t			lkb_wait_type;	/* type of reply waiting for */
 	int8_t			lkb_wait_count;
+	int			lkb_wait_nodeid; /* for debugging */
 
 	struct list_head	lkb_idtbl_list;	/* lockspace lkbtbl */
 	struct list_head	lkb_statequeue;	/* rsb g/c/w list */
···
 	struct list_head	lkb_ownqueue;	/* list of locks for a process */
 	struct list_head	lkb_time_list;
 	ktime_t			lkb_timestamp;
+	ktime_t			lkb_wait_time;
 	unsigned long		lkb_timeout_cs;
 
 	struct dlm_callback	lkb_callbacks[DLM_CALLBACKS_SIZE];

fs/dlm/lock.c

···
 	return -1;
 }
 
+static int nodeid_warned(int nodeid, int num_nodes, int *warned)
+{
+	int i;
+
+	for (i = 0; i < num_nodes; i++) {
+		if (!warned[i]) {
+			warned[i] = nodeid;
+			return 0;
+		}
+		if (warned[i] == nodeid)
+			return 1;
+	}
+	return 0;
+}
+
+void dlm_scan_waiters(struct dlm_ls *ls)
+{
+	struct dlm_lkb *lkb;
+	ktime_t zero = ktime_set(0, 0);
+	s64 us;
+	s64 debug_maxus = 0;
+	u32 debug_scanned = 0;
+	u32 debug_expired = 0;
+	int num_nodes = 0;
+	int *warned = NULL;
+
+	if (!dlm_config.ci_waitwarn_us)
+		return;
+
+	mutex_lock(&ls->ls_waiters_mutex);
+
+	list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
+		if (ktime_equal(lkb->lkb_wait_time, zero))
+			continue;
+
+		debug_scanned++;
+
+		us = ktime_to_us(ktime_sub(ktime_get(), lkb->lkb_wait_time));
+
+		if (us < dlm_config.ci_waitwarn_us)
+			continue;
+
+		lkb->lkb_wait_time = zero;
+
+		debug_expired++;
+		if (us > debug_maxus)
+			debug_maxus = us;
+
+		if (!num_nodes) {
+			num_nodes = ls->ls_num_nodes;
+			warned = kmalloc(GFP_KERNEL, num_nodes * sizeof(int));
+			if (warned)
+				memset(warned, 0, num_nodes * sizeof(int));
+		}
+		if (!warned)
+			continue;
+		if (nodeid_warned(lkb->lkb_wait_nodeid, num_nodes, warned))
+			continue;
+
+		log_error(ls, "waitwarn %x %lld %d us check connection to "
+			  "node %d", lkb->lkb_id, (long long)us,
+			  dlm_config.ci_waitwarn_us, lkb->lkb_wait_nodeid);
+	}
+	mutex_unlock(&ls->ls_waiters_mutex);
+
+	if (warned)
+		kfree(warned);
+
+	if (debug_expired)
+		log_debug(ls, "scan_waiters %u warn %u over %d us max %lld us",
+			  debug_scanned, debug_expired,
+			  dlm_config.ci_waitwarn_us, (long long)debug_maxus);
+}
+
 /* add/remove lkb from global waiters list of lkb's waiting for
    a reply from a remote node */
 
-static int add_to_waiters(struct dlm_lkb *lkb, int mstype)
+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;
···
 
 	lkb->lkb_wait_count++;
 	lkb->lkb_wait_type = mstype;
+	lkb->lkb_wait_time = ktime_get();
+	lkb->lkb_wait_nodeid = to_nodeid; /* for debugging */
 	hold_lkb(lkb);
 	list_add(&lkb->lkb_wait_reply, &ls->ls_waiters);
  out:
···
 	struct dlm_ls *ls = lkb->lkb_resource->res_ls;
 	int error;
 
-	if (ms != &ls->ls_stub_ms)
+	if (ms->m_flags != DLM_IFL_STUB_MS)
 		mutex_lock(&ls->ls_waiters_mutex);
 	error = _remove_from_waiters(lkb, ms->m_type, ms);
-	if (ms != &ls->ls_stub_ms)
+	if (ms->m_flags != DLM_IFL_STUB_MS)
 		mutex_unlock(&ls->ls_waiters_mutex);
 	return error;
 }
···
 	list_for_each_entry(lkb, &ls->ls_timeout, lkb_time_list)
 		lkb->lkb_timestamp = ktime_add_us(lkb->lkb_timestamp, adj_us);
 	mutex_unlock(&ls->ls_timeout_mutex);
+
+	if (!dlm_config.ci_waitwarn_us)
+		return;
+
+	mutex_lock(&ls->ls_waiters_mutex);
+	list_for_each_entry(lkb, &ls->ls_waiters, lkb_wait_reply) {
+		if (ktime_to_us(lkb->lkb_wait_time))
+			lkb->lkb_wait_time = ktime_get();
+	}
+	mutex_unlock(&ls->ls_waiters_mutex);
 }
 
 /* lkb is master or local copy */
···
    ALTPR/ALTCW: our rqmode may have been changed to PR or CW to become
    compatible with other granted locks */
 
-static void munge_demoted(struct dlm_lkb *lkb, struct dlm_message *ms)
+static void munge_demoted(struct dlm_lkb *lkb)
 {
-	if (ms->m_type != DLM_MSG_CONVERT_REPLY) {
-		log_print("munge_demoted %x invalid reply type %d",
-			  lkb->lkb_id, ms->m_type);
-		return;
-	}
-
 	if (lkb->lkb_rqmode == DLM_LOCK_IV || lkb->lkb_grmode == DLM_LOCK_IV) {
 		log_print("munge_demoted %x invalid modes gr %d rq %d",
 			  lkb->lkb_id, lkb->lkb_grmode, lkb->lkb_rqmode);
···
 	struct dlm_mhandle *mh;
 	int to_nodeid, error;
 
-	error = add_to_waiters(lkb, mstype);
+	to_nodeid = r->res_nodeid;
+
+	error = add_to_waiters(lkb, mstype, to_nodeid);
 	if (error)
 		return error;
-
-	to_nodeid = r->res_nodeid;
 
 	error = create_message(r, lkb, to_nodeid, mstype, &ms, &mh);
 	if (error)
···
 	/* down conversions go without a reply from the master */
 	if (!error && down_conversion(lkb)) {
 		remove_from_waiters(lkb, DLM_MSG_CONVERT_REPLY);
+		r->res_ls->ls_stub_ms.m_flags = DLM_IFL_STUB_MS;
 		r->res_ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
 		r->res_ls->ls_stub_ms.m_result = 0;
-		r->res_ls->ls_stub_ms.m_flags = lkb->lkb_flags;
 		__receive_convert_reply(r, lkb, &r->res_ls->ls_stub_ms);
 	}
 
···
 	struct dlm_mhandle *mh;
 	int to_nodeid, error;
 
-	error = add_to_waiters(lkb, DLM_MSG_LOOKUP);
+	to_nodeid = dlm_dir_nodeid(r);
+
+	error = add_to_waiters(lkb, DLM_MSG_LOOKUP, to_nodeid);
 	if (error)
 		return error;
-
-	to_nodeid = dlm_dir_nodeid(r);
 
 	error = create_message(r, NULL, to_nodeid, DLM_MSG_LOOKUP, &ms, &mh);
 	if (error)
···
 
 static void receive_flags_reply(struct dlm_lkb *lkb, struct dlm_message *ms)
 {
+	if (ms->m_flags == DLM_IFL_STUB_MS)
+		return;
+
 	lkb->lkb_sbflags = ms->m_sbflags;
 	lkb->lkb_flags = (lkb->lkb_flags & 0xFFFF0000) |
 		         (ms->m_flags & 0x0000FFFF);
···
 		/* convert was queued on remote master */
 		receive_flags_reply(lkb, ms);
 		if (is_demoted(lkb))
-			munge_demoted(lkb, ms);
+			munge_demoted(lkb);
 		del_lkb(r, lkb);
 		add_lkb(r, lkb, DLM_LKSTS_CONVERT);
 		add_timeout(lkb);
···
 		/* convert was granted on remote master */
 		receive_flags_reply(lkb, ms);
 		if (is_demoted(lkb))
-			munge_demoted(lkb, ms);
+			munge_demoted(lkb);
 		grant_lock_pc(r, lkb, ms);
 		queue_cast(r, lkb, 0);
 		break;
···
 	dlm_put_lockspace(ls);
 }
 
-static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb)
+static void recover_convert_waiter(struct dlm_ls *ls, struct dlm_lkb *lkb,
+				   struct dlm_message *ms_stub)
 {
 	if (middle_conversion(lkb)) {
 		hold_lkb(lkb);
-		ls->ls_stub_ms.m_type = DLM_MSG_CONVERT_REPLY;
-		ls->ls_stub_ms.m_result = -EINPROGRESS;
-		ls->ls_stub_ms.m_flags = lkb->lkb_flags;
-		ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
-		_receive_convert_reply(lkb, &ls->ls_stub_ms);
+		memset(ms_stub, 0, sizeof(struct dlm_message));
+		ms_stub->m_flags = DLM_IFL_STUB_MS;
+		ms_stub->m_type = DLM_MSG_CONVERT_REPLY;
+		ms_stub->m_result = -EINPROGRESS;
+		ms_stub->m_header.h_nodeid = lkb->lkb_nodeid;
+		_receive_convert_reply(lkb, ms_stub);
 
 		/* Same special case as in receive_rcom_lock_args() */
 		lkb->lkb_grmode = DLM_LOCK_IV;
···
 void dlm_recover_waiters_pre(struct dlm_ls *ls)
 {
 	struct dlm_lkb *lkb, *safe;
+	struct dlm_message *ms_stub;
 	int wait_type, stub_unlock_result, stub_cancel_result;
+
+	ms_stub = kmalloc(GFP_KERNEL, sizeof(struct dlm_message));
+	if (!ms_stub) {
+		log_error(ls, "dlm_recover_waiters_pre no mem");
+		return;
+	}
 
 	mutex_lock(&ls->ls_waiters_mutex);
 
 	list_for_each_entry_safe(lkb, safe, &ls->ls_waiters, lkb_wait_reply) {
-		log_debug(ls, "pre recover waiter lkid %x type %d flags %x",
-			  lkb->lkb_id, lkb->lkb_wait_type, lkb->lkb_flags);
+
+		/* exclude debug messages about unlocks because there can be so
+		   many and they aren't very interesting */
+
+		if (lkb->lkb_wait_type != DLM_MSG_UNLOCK) {
+			log_debug(ls, "recover_waiter %x nodeid %d "
+				  "msg %d to %d", lkb->lkb_id, lkb->lkb_nodeid,
+				  lkb->lkb_wait_type, lkb->lkb_wait_nodeid);
+		}
 
 		/* all outstanding lookups, regardless of destination  will be
 		   resent after recovery is done */
···
 			break;
 
 		case DLM_MSG_CONVERT:
-			recover_convert_waiter(ls, lkb);
+			recover_convert_waiter(ls, lkb, ms_stub);
 			break;
 
 		case DLM_MSG_UNLOCK:
 			hold_lkb(lkb);
-			ls->ls_stub_ms.m_type = DLM_MSG_UNLOCK_REPLY;
-			ls->ls_stub_ms.m_result = stub_unlock_result;
-			ls->ls_stub_ms.m_flags = lkb->lkb_flags;
-			ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
-			_receive_unlock_reply(lkb, &ls->ls_stub_ms);
+			memset(ms_stub, 0, sizeof(struct dlm_message));
+			ms_stub->m_flags = DLM_IFL_STUB_MS;
+			ms_stub->m_type = DLM_MSG_UNLOCK_REPLY;
+			ms_stub->m_result = stub_unlock_result;
+			ms_stub->m_header.h_nodeid = lkb->lkb_nodeid;
+			_receive_unlock_reply(lkb, ms_stub);
 			dlm_put_lkb(lkb);
 			break;
 
 		case DLM_MSG_CANCEL:
 			hold_lkb(lkb);
-			ls->ls_stub_ms.m_type = DLM_MSG_CANCEL_REPLY;
-			ls->ls_stub_ms.m_result = stub_cancel_result;
-			ls->ls_stub_ms.m_flags = lkb->lkb_flags;
-			ls->ls_stub_ms.m_header.h_nodeid = lkb->lkb_nodeid;
-			_receive_cancel_reply(lkb, &ls->ls_stub_ms);
+			memset(ms_stub, 0, sizeof(struct dlm_message));
+			ms_stub->m_flags = DLM_IFL_STUB_MS;
+			ms_stub->m_type = DLM_MSG_CANCEL_REPLY;
+			ms_stub->m_result = stub_cancel_result;
+			ms_stub->m_header.h_nodeid = lkb->lkb_nodeid;
+			_receive_cancel_reply(lkb, ms_stub);
 			dlm_put_lkb(lkb);
 			break;
 
···
 		schedule();
 	}
 	mutex_unlock(&ls->ls_waiters_mutex);
+	kfree(ms_stub);
 }
 
 static struct dlm_lkb *find_resend_waiter(struct dlm_ls *ls)
···
 		ou = is_overlap_unlock(lkb);
 		err = 0;
 
-		log_debug(ls, "recover_waiters_post %x type %d flags %x %s",
-			  lkb->lkb_id, mstype, lkb->lkb_flags, r->res_name);
+		log_debug(ls, "recover_waiter %x nodeid %d msg %d r_nodeid %d",
+			  lkb->lkb_id, lkb->lkb_nodeid, mstype, r->res_nodeid);
 
 		/* At this point we assume that we won't get a reply to any
 		   previous op or overlap op on this lock.  First, do a big

fs/dlm/lock.h

···
 void dlm_scan_rsbs(struct dlm_ls *ls);
 int dlm_lock_recovery_try(struct dlm_ls *ls);
 void dlm_unlock_recovery(struct dlm_ls *ls);
+void dlm_scan_waiters(struct dlm_ls *ls);
 void dlm_scan_timeout(struct dlm_ls *ls);
 void dlm_adjust_timeouts(struct dlm_ls *ls);
 

fs/dlm/lockspace.c

···
 static int dlm_scand(void *data)
 {
 	struct dlm_ls *ls;
-	int timeout_jiffies = dlm_config.ci_scan_secs * HZ;
 
 	while (!kthread_should_stop()) {
 		ls = find_ls_to_scan();
···
 				ls->ls_scan_time = jiffies;
 				dlm_scan_rsbs(ls);
 				dlm_scan_timeout(ls);
+				dlm_scan_waiters(ls);
 				dlm_unlock_recovery(ls);
 			} else {
 				ls->ls_scan_time += HZ;
 			}
-		} else {
-			schedule_timeout_interruptible(timeout_jiffies);
+			continue;
 		}
+		schedule_timeout_interruptible(dlm_config.ci_scan_secs * HZ);
 	}
 	return 0;
 }

fs/dlm/plock.c

···
 	wake_up(&send_wq);
 }
 
+/* If a process was killed while waiting for the only plock on a file,
+   locks_remove_posix will not see any lock on the file so it won't
+   send an unlock-close to us to pass on to userspace to clean up the
+   abandoned waiter.  So, we have to insert the unlock-close when the
+   lock call is interrupted. */
+
+static void do_unlock_close(struct dlm_ls *ls, u64 number,
+			    struct file *file, struct file_lock *fl)
+{
+	struct plock_op *op;
+
+	op = kzalloc(sizeof(*op), GFP_NOFS);
+	if (!op)
+		return;
+
+	op->info.optype		= DLM_PLOCK_OP_UNLOCK;
+	op->info.pid		= fl->fl_pid;
+	op->info.fsid		= ls->ls_global_id;
+	op->info.number		= number;
+	op->info.start		= 0;
+	op->info.end		= OFFSET_MAX;
+	if (fl->fl_lmops && fl->fl_lmops->fl_grant)
+		op->info.owner	= (__u64) fl->fl_pid;
+	else
+		op->info.owner	= (__u64)(long) fl->fl_owner;
+
+	op->info.flags |= DLM_PLOCK_FL_CLOSE;
+	send_op(op);
+}
+
 int dlm_posix_lock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 		   int cmd, struct file_lock *fl)
 {
···
 
 	send_op(op);
 
-	if (xop->callback == NULL)
-		wait_event(recv_wq, (op->done != 0));
-	else {
+	if (xop->callback == NULL) {
+		rv = wait_event_killable(recv_wq, (op->done != 0));
+		if (rv == -ERESTARTSYS) {
+			log_debug(ls, "dlm_posix_lock: wait killed %llx",
+				  (unsigned long long)number);
+			spin_lock(&ops_lock);
+			list_del(&op->list);
+			spin_unlock(&ops_lock);
+			kfree(xop);
+			do_unlock_close(ls, number, file, fl);
+			goto out;
+		}
+	} else {
 		rv = FILE_LOCK_DEFERRED;
 		goto out;
 	}
···
 	else
 		op->info.owner	= (__u64)(long) fl->fl_owner;
 
+	if (fl->fl_flags & FL_CLOSE) {
+		op->info.flags |= DLM_PLOCK_FL_CLOSE;
+		send_op(op);
+		rv = 0;
+		goto out;
+	}
+
 	send_op(op);
 	wait_event(recv_wq, (op->done != 0));
 
···
 	spin_lock(&ops_lock);
 	if (!list_empty(&send_list)) {
 		op = list_entry(send_list.next, struct plock_op, list);
-		list_move(&op->list, &recv_list);
+		if (op->info.flags & DLM_PLOCK_FL_CLOSE)
+			list_del(&op->list);
+		else
+			list_move(&op->list, &recv_list);
 		memcpy(&info, &op->info, sizeof(info));
 	}
 	spin_unlock(&ops_lock);
 
 	if (!op)
 		return -EAGAIN;
+
+	/* there is no need to get a reply from userspace for unlocks
+	   that were generated by the vfs cleaning up for a close
+	   (the process did not make an unlock call). */
+
+	if (op->info.flags & DLM_PLOCK_FL_CLOSE)
+		kfree(op);
 
 	if (copy_to_user(u, &info, sizeof(info)))
 		return -EFAULT;

fs/dlm/user.c

···
 
  out_sig:
 	sigprocmask(SIG_SETMASK, &tmpsig, NULL);
-	recalc_sigpending();
  out_free:
 	kfree(kbuf);
 	return error;

include/linux/dlm_plock.h

···
 #define DLM_PLOCK_MISC_NAME		"dlm_plock"
 
 #define DLM_PLOCK_VERSION_MAJOR	1
-#define DLM_PLOCK_VERSION_MINOR	1
+#define DLM_PLOCK_VERSION_MINOR	2
 #define DLM_PLOCK_VERSION_PATCH	0
 
 enum {
···
 	DLM_PLOCK_OP_GET,
 };
 
+#define DLM_PLOCK_FL_CLOSE 1
+
 struct dlm_plock_info {
 	__u32 version[3];
 	__u8 optype;
 	__u8 ex;
 	__u8 wait;
-	__u8 pad;
+	__u8 flags;
 	__u32 pid;
 	__s32 nodeid;
 	__s32 rv;