[mirror_ubuntu-artful-kernel.git] / fs / dlm / member.c

/******************************************************************************
*******************************************************************************
**
**  Copyright (C) 2005-2011 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
**  of the GNU General Public License v.2.
**
*******************************************************************************
******************************************************************************/

#include "dlm_internal.h"
#include "lockspace.h"
#include "member.h"
#include "recoverd.h"
#include "recover.h"
#include "rcom.h"
#include "config.h"
#include "lowcomms.h"

int dlm_slots_version(struct dlm_header *h)
{
	if ((h->h_version & 0x0000FFFF) < DLM_HEADER_SLOTS)
		return 0;
	return 1;
}

void dlm_slot_save(struct dlm_ls *ls, struct dlm_rcom *rc,
		   struct dlm_member *memb)
{
	struct rcom_config *rf = (struct rcom_config *)rc->rc_buf;

	if (!dlm_slots_version(&rc->rc_header))
		return;

	memb->slot = le16_to_cpu(rf->rf_our_slot);
	memb->generation = le32_to_cpu(rf->rf_generation);
}

void dlm_slots_copy_out(struct dlm_ls *ls, struct dlm_rcom *rc)
{
	struct dlm_slot *slot;
	struct rcom_slot *ro;
	int i;

	ro = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config));

	/* ls_slots array is sparse, but not rcom_slots */

	for (i = 0; i < ls->ls_slots_size; i++) {
		slot = &ls->ls_slots[i];
		if (!slot->nodeid)
			continue;
		ro->ro_nodeid = cpu_to_le32(slot->nodeid);
		ro->ro_slot = cpu_to_le16(slot->slot);
		ro++;
	}
}

#define SLOT_DEBUG_LINE 128

static void log_slots(struct dlm_ls *ls, uint32_t gen, int num_slots,
		      struct rcom_slot *ro0, struct dlm_slot *array,
		      int array_size)
{
	char line[SLOT_DEBUG_LINE];
	int len = SLOT_DEBUG_LINE - 1;
	int pos = 0;
	int ret, i;

	memset(line, 0, sizeof(line));

	if (array) {
		for (i = 0; i < array_size; i++) {
			if (!array[i].nodeid)
				continue;

			ret = snprintf(line + pos, len - pos, " %d:%d",
				       array[i].slot, array[i].nodeid);
			if (ret >= len - pos)
				break;
			pos += ret;
		}
	} else if (ro0) {
		for (i = 0; i < num_slots; i++) {
			ret = snprintf(line + pos, len - pos, " %d:%d",
				       ro0[i].ro_slot, ro0[i].ro_nodeid);
			if (ret >= len - pos)
				break;
			pos += ret;
		}
	}

	log_rinfo(ls, "generation %u slots %d%s", gen, num_slots, line);
}

int dlm_slots_copy_in(struct dlm_ls *ls)
{
	struct dlm_member *memb;
	struct dlm_rcom *rc = ls->ls_recover_buf;
	struct rcom_config *rf = (struct rcom_config *)rc->rc_buf;
	struct rcom_slot *ro0, *ro;
	int our_nodeid = dlm_our_nodeid();
	int i, num_slots;
	uint32_t gen;

	if (!dlm_slots_version(&rc->rc_header))
		return -1;

	gen = le32_to_cpu(rf->rf_generation);
	if (gen <= ls->ls_generation) {
		log_error(ls, "dlm_slots_copy_in gen %u old %u",
			  gen, ls->ls_generation);
	}
	ls->ls_generation = gen;

	num_slots = le16_to_cpu(rf->rf_num_slots);
	if (!num_slots)
		return -1;

	ro0 = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config));

	for (i = 0, ro = ro0; i < num_slots; i++, ro++) {
		ro->ro_nodeid = le32_to_cpu(ro->ro_nodeid);
		ro->ro_slot = le16_to_cpu(ro->ro_slot);
	}

	log_slots(ls, gen, num_slots, ro0, NULL, 0);

	list_for_each_entry(memb, &ls->ls_nodes, list) {
		for (i = 0, ro = ro0; i < num_slots; i++, ro++) {
			if (ro->ro_nodeid != memb->nodeid)
				continue;
			memb->slot = ro->ro_slot;
			memb->slot_prev = memb->slot;
			break;
		}

		if (memb->nodeid == our_nodeid) {
			if (ls->ls_slot && ls->ls_slot != memb->slot) {
				log_error(ls, "dlm_slots_copy_in our slot "
					  "changed %d %d", ls->ls_slot,
					  memb->slot);
				return -1;
			}

			if (!ls->ls_slot)
				ls->ls_slot = memb->slot;
		}

		if (!memb->slot) {
			log_error(ls, "dlm_slots_copy_in nodeid %d no slot",
				   memb->nodeid);
			return -1;
		}
	}

	return 0;
}

/* for any nodes that do not support slots, we will not have set memb->slot
   in wait_status_all(), so memb->slot will remain -1, and we will not
   assign slots or set ls_num_slots here */

int dlm_slots_assign(struct dlm_ls *ls, int *num_slots, int *slots_size,
		     struct dlm_slot **slots_out, uint32_t *gen_out)
{
	struct dlm_member *memb;
	struct dlm_slot *array;
	int our_nodeid = dlm_our_nodeid();
	int array_size, max_slots, i;
	int need = 0;
	int max = 0;
	int num = 0;
	uint32_t gen = 0;

	/* our own memb struct will have slot -1 gen 0 */

	list_for_each_entry(memb, &ls->ls_nodes, list) {
		if (memb->nodeid == our_nodeid) {
			memb->slot = ls->ls_slot;
			memb->generation = ls->ls_generation;
			break;
		}
	}

	list_for_each_entry(memb, &ls->ls_nodes, list) {
		if (memb->generation > gen)
			gen = memb->generation;

		/* node doesn't support slots */

		if (memb->slot == -1)
			return -1;

		/* node needs a slot assigned */

		if (!memb->slot)
			need++;

		/* node has a slot assigned */

		num++;

		if (!max || max < memb->slot)
			max = memb->slot;

		/* sanity check, once slot is assigned it shouldn't change */

		if (memb->slot_prev && memb->slot && memb->slot_prev != memb->slot) {
			log_error(ls, "nodeid %d slot changed %d %d",
				  memb->nodeid, memb->slot_prev, memb->slot);
			return -1;
		}
		memb->slot_prev = memb->slot;
	}

	array_size = max + need;

	array = kzalloc(array_size * sizeof(struct dlm_slot), GFP_NOFS);
	if (!array)
		return -ENOMEM;

	num = 0;

	/* fill in slots (offsets) that are used */

	list_for_each_entry(memb, &ls->ls_nodes, list) {
		if (!memb->slot)
			continue;

		if (memb->slot > array_size) {
			log_error(ls, "invalid slot number %d", memb->slot);
			kfree(array);
			return -1;
		}

		array[memb->slot - 1].nodeid = memb->nodeid;
		array[memb->slot - 1].slot = memb->slot;
		num++;
	}

	/* assign new slots from unused offsets */

	list_for_each_entry(memb, &ls->ls_nodes, list) {
		if (memb->slot)
			continue;

		for (i = 0; i < array_size; i++) {
			if (array[i].nodeid)
				continue;

			memb->slot = i + 1;
			memb->slot_prev = memb->slot;
			array[i].nodeid = memb->nodeid;
			array[i].slot = memb->slot;
			num++;

			if (!ls->ls_slot && memb->nodeid == our_nodeid)
				ls->ls_slot = memb->slot;
			break;
		}

		if (!memb->slot) {
			log_error(ls, "no free slot found");
			kfree(array);
			return -1;
		}
	}

	gen++;

	log_slots(ls, gen, num, NULL, array, array_size);

	max_slots = (dlm_config.ci_buffer_size - sizeof(struct dlm_rcom) -
		     sizeof(struct rcom_config)) / sizeof(struct rcom_slot);

	if (num > max_slots) {
		log_error(ls, "num_slots %d exceeds max_slots %d",
			  num, max_slots);
		kfree(array);
		return -1;
	}

	*gen_out = gen;
	*slots_out = array;
	*slots_size = array_size;
	*num_slots = num;
	return 0;
}

static void add_ordered_member(struct dlm_ls *ls, struct dlm_member *new)
{
	struct dlm_member *memb = NULL;
	struct list_head *tmp;
	struct list_head *newlist = &new->list;
	struct list_head *head = &ls->ls_nodes;

	list_for_each(tmp, head) {
		memb = list_entry(tmp, struct dlm_member, list);
		if (new->nodeid < memb->nodeid)
			break;
	}

	if (!memb)
		list_add_tail(newlist, head);
	else {
		/* FIXME: can use list macro here */
		newlist->prev = tmp->prev;
		newlist->next = tmp;
		tmp->prev->next = newlist;
		tmp->prev = newlist;
	}
}

static int dlm_add_member(struct dlm_ls *ls, struct dlm_config_node *node)
{
	struct dlm_member *memb;
	int error;

	memb = kzalloc(sizeof(struct dlm_member), GFP_NOFS);
	if (!memb)
		return -ENOMEM;

	error = dlm_lowcomms_connect_node(node->nodeid);
	if (error < 0) {
		kfree(memb);
		return error;
	}

	memb->nodeid = node->nodeid;
	memb->weight = node->weight;
	memb->comm_seq = node->comm_seq;
	add_ordered_member(ls, memb);
	ls->ls_num_nodes++;
	return 0;
}

static struct dlm_member *find_memb(struct list_head *head, int nodeid)
{
	struct dlm_member *memb;

	list_for_each_entry(memb, head, list) {
		if (memb->nodeid == nodeid)
			return memb;
	}
	return NULL;
}

int dlm_is_member(struct dlm_ls *ls, int nodeid)
{
	if (find_memb(&ls->ls_nodes, nodeid))
		return 1;
	return 0;
}

int dlm_is_removed(struct dlm_ls *ls, int nodeid)
{
	if (find_memb(&ls->ls_nodes_gone, nodeid))
		return 1;
	return 0;
}

static void clear_memb_list(struct list_head *head)
{
	struct dlm_member *memb;

	while (!list_empty(head)) {
		memb = list_entry(head->next, struct dlm_member, list);
		list_del(&memb->list);
		kfree(memb);
	}
}

void dlm_clear_members(struct dlm_ls *ls)
{
	clear_memb_list(&ls->ls_nodes);
	ls->ls_num_nodes = 0;
}

void dlm_clear_members_gone(struct dlm_ls *ls)
{
	clear_memb_list(&ls->ls_nodes_gone);
}

static void make_member_array(struct dlm_ls *ls)
{
	struct dlm_member *memb;
	int i, w, x = 0, total = 0, all_zero = 0, *array;

	kfree(ls->ls_node_array);
	ls->ls_node_array = NULL;

	list_for_each_entry(memb, &ls->ls_nodes, list) {
		if (memb->weight)
			total += memb->weight;
	}

	/* all nodes revert to weight of 1 if all have weight 0 */

	if (!total) {
		total = ls->ls_num_nodes;
		all_zero = 1;
	}

	ls->ls_total_weight = total;

	array = kmalloc(sizeof(int) * total, GFP_NOFS);
	if (!array)
		return;

	list_for_each_entry(memb, &ls->ls_nodes, list) {
		if (!all_zero && !memb->weight)
			continue;

		if (all_zero)
			w = 1;
		else
			w = memb->weight;

		DLM_ASSERT(x < total, printk("total %d x %d\n", total, x););

		for (i = 0; i < w; i++)
			array[x++] = memb->nodeid;
	}

	ls->ls_node_array = array;
}

/* send a status request to all members just to establish comms connections */

static int ping_members(struct dlm_ls *ls)
{
	struct dlm_member *memb;
	int error = 0;

	list_for_each_entry(memb, &ls->ls_nodes, list) {
		error = dlm_recovery_stopped(ls);
		if (error)
			break;
		error = dlm_rcom_status(ls, memb->nodeid, 0);
		if (error)
			break;
	}
	if (error)
		log_rinfo(ls, "ping_members aborted %d last nodeid %d",
			  error, ls->ls_recover_nodeid);
	return error;
}

static void dlm_lsop_recover_prep(struct dlm_ls *ls)
{
	if (!ls->ls_ops || !ls->ls_ops->recover_prep)
		return;
	ls->ls_ops->recover_prep(ls->ls_ops_arg);
}

static void dlm_lsop_recover_slot(struct dlm_ls *ls, struct dlm_member *memb)
{
	struct dlm_slot slot;
	uint32_t seq;
	int error;

	if (!ls->ls_ops || !ls->ls_ops->recover_slot)
		return;

	/* if there is no comms connection with this node
	   or the present comms connection is newer
	   than the one when this member was added, then
	   we consider the node to have failed (versus
	   being removed due to dlm_release_lockspace) */

	error = dlm_comm_seq(memb->nodeid, &seq);

	if (!error && seq == memb->comm_seq)
		return;

	slot.nodeid = memb->nodeid;
	slot.slot = memb->slot;

	ls->ls_ops->recover_slot(ls->ls_ops_arg, &slot);
}

void dlm_lsop_recover_done(struct dlm_ls *ls)
{
	struct dlm_member *memb;
	struct dlm_slot *slots;
	int i, num;

	if (!ls->ls_ops || !ls->ls_ops->recover_done)
		return;

	num = ls->ls_num_nodes;

	slots = kzalloc(num * sizeof(struct dlm_slot), GFP_KERNEL);
	if (!slots)
		return;

	i = 0;
	list_for_each_entry(memb, &ls->ls_nodes, list) {
		if (i == num) {
			log_error(ls, "dlm_lsop_recover_done bad num %d", num);
			goto out;
		}
		slots[i].nodeid = memb->nodeid;
		slots[i].slot = memb->slot;
		i++;
	}

	ls->ls_ops->recover_done(ls->ls_ops_arg, slots, num,
				 ls->ls_slot, ls->ls_generation);
 out:
	kfree(slots);
}

static struct dlm_config_node *find_config_node(struct dlm_recover *rv,
						int nodeid)
{
	int i;

	for (i = 0; i < rv->nodes_count; i++) {
		if (rv->nodes[i].nodeid == nodeid)
			return &rv->nodes[i];
	}
	return NULL;
}

int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
{
	struct dlm_member *memb, *safe;
	struct dlm_config_node *node;
	int i, error, neg = 0, low = -1;

	/* previously removed members that we've not finished removing need to
	   count as a negative change so the "neg" recovery steps will happen */

	list_for_each_entry(memb, &ls->ls_nodes_gone, list) {
		log_rinfo(ls, "prev removed member %d", memb->nodeid);
		neg++;
	}

	/* move departed members from ls_nodes to ls_nodes_gone */

	list_for_each_entry_safe(memb, safe, &ls->ls_nodes, list) {
		node = find_config_node(rv, memb->nodeid);
		if (node && !node->new)
			continue;

		if (!node) {
			log_rinfo(ls, "remove member %d", memb->nodeid);
		} else {
			/* removed and re-added */
			log_rinfo(ls, "remove member %d comm_seq %u %u",
				  memb->nodeid, memb->comm_seq, node->comm_seq);
		}

		neg++;
		list_move(&memb->list, &ls->ls_nodes_gone);
		ls->ls_num_nodes--;
		dlm_lsop_recover_slot(ls, memb);
	}

	/* add new members to ls_nodes */

	for (i = 0; i < rv->nodes_count; i++) {
		node = &rv->nodes[i];
		if (dlm_is_member(ls, node->nodeid))
			continue;
		dlm_add_member(ls, node);
		log_rinfo(ls, "add member %d", node->nodeid);
	}

	list_for_each_entry(memb, &ls->ls_nodes, list) {
		if (low == -1 || memb->nodeid < low)
			low = memb->nodeid;
	}
	ls->ls_low_nodeid = low;

	make_member_array(ls);
	*neg_out = neg;

	error = ping_members(ls);
	if (!error || error == -EPROTO) {
		/* new_lockspace() may be waiting to know if the config
		   is good or bad */
		ls->ls_members_result = error;
		complete(&ls->ls_members_done);
	}

	log_rinfo(ls, "dlm_recover_members %d nodes", ls->ls_num_nodes);
	return error;
}

/* 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;

	/*
	 * 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 RECOVER_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);
	set_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
	new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags);
	ls->ls_recover_seq++;
	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 truly stopped.
	 * 2) Keeps any new requests from being processed until it's unlocked
	 *    when recovery is complete.
	 */

	if (new) {
		set_bit(LSFL_RECOVER_DOWN, &ls->ls_flags);
		wake_up_process(ls->ls_recoverd_task);
		wait_event(ls->ls_recover_lock_wait,
			   test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags));
	}

	/*
	 * The recoverd suspend/resume makes sure that dlm_recoverd (if
	 * running) has noticed RECOVER_STOP above and quit processing the
	 * previous recovery.
	 */

	dlm_recoverd_suspend(ls);

	spin_lock(&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);

	dlm_recoverd_resume(ls);

	if (!ls->ls_recover_begin)
		ls->ls_recover_begin = jiffies;

	dlm_lsop_recover_prep(ls);
	return 0;
}

int dlm_ls_start(struct dlm_ls *ls)
{
	struct dlm_recover *rv = NULL, *rv_old;
	struct dlm_config_node *nodes;
	int error, count;

	rv = kzalloc(sizeof(struct dlm_recover), GFP_NOFS);
	if (!rv)
		return -ENOMEM;

	error = dlm_config_nodes(ls->ls_name, &nodes, &count);
	if (error < 0)
		goto fail;

	spin_lock(&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);
		log_error(ls, "start ignored: lockspace running");
		error = -EINVAL;
		goto fail;
	}

	rv->nodes = nodes;
	rv->nodes_count = count;
	rv->seq = ++ls->ls_recover_seq;
	rv_old = ls->ls_recover_args;
	ls->ls_recover_args = rv;
	spin_unlock(&ls->ls_recover_lock);

	if (rv_old) {
		log_error(ls, "unused recovery %llx %d",
			  (unsigned long long)rv_old->seq, rv_old->nodes_count);
		kfree(rv_old->nodes);
		kfree(rv_old);
	}

	set_bit(LSFL_RECOVER_WORK, &ls->ls_flags);
	wake_up_process(ls->ls_recoverd_task);
	return 0;

 fail:
	kfree(rv);
	kfree(nodes);
	return error;
}
Commit	Line	Data
e7fd4179 DT	1	/******************************************************************************
	2	*******************************************************************************
	3	**
60f98d18	4	** Copyright (C) 2005-2011 Red Hat, Inc. All rights reserved.
e7fd4179 DT	5	**
	6	** This copyrighted material is made available to anyone wishing to use,
	7	** modify, copy, or redistribute it subject to the terms and conditions
	8	** of the GNU General Public License v.2.
	9	**
	10	*******************************************************************************
	11	******************************************************************************/
	12
	13	#include "dlm_internal.h"
	14	#include "lockspace.h"
	15	#include "member.h"
	16	#include "recoverd.h"
	17	#include "recover.h"
e7fd4179 DT	18	#include "rcom.h"
e7fd4179 DT	19	#include "config.h"
391fbdc5	20	#include "lowcomms.h"
e7fd4179	21
757a4271 DT	22	int dlm_slots_version(struct dlm_header *h)
	23	{
	24	if ((h->h_version & 0x0000FFFF) < DLM_HEADER_SLOTS)
	25	return 0;
	26	return 1;
	27	}
	28
	29	void dlm_slot_save(struct dlm_ls ls, struct dlm_rcom rc,
60f98d18	30	struct dlm_member *memb)
757a4271 DT	31	{
	32	struct rcom_config rf = (struct rcom_config )rc->rc_buf;
	33
	34	if (!dlm_slots_version(&rc->rc_header))
	35	return;
	36
	37	memb->slot = le16_to_cpu(rf->rf_our_slot);
	38	memb->generation = le32_to_cpu(rf->rf_generation);
	39	}
	40
	41	void dlm_slots_copy_out(struct dlm_ls ls, struct dlm_rcom rc)
	42	{
	43	struct dlm_slot *slot;
	44	struct rcom_slot *ro;
	45	int i;
	46
	47	ro = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config));
	48
	49	/* ls_slots array is sparse, but not rcom_slots */
	50
	51	for (i = 0; i < ls->ls_slots_size; i++) {
	52	slot = &ls->ls_slots[i];
	53	if (!slot->nodeid)
	54	continue;
	55	ro->ro_nodeid = cpu_to_le32(slot->nodeid);
	56	ro->ro_slot = cpu_to_le16(slot->slot);
	57	ro++;
	58	}
	59	}
	60
	61	#define SLOT_DEBUG_LINE 128
	62
075f0177 DT	63	static void log_slots(struct dlm_ls *ls, uint32_t gen, int num_slots,
	64	struct rcom_slot ro0, struct dlm_slot array,
	65	int array_size)
757a4271 DT	66	{
	67	char line[SLOT_DEBUG_LINE];
	68	int len = SLOT_DEBUG_LINE - 1;
	69	int pos = 0;
	70	int ret, i;
	71
757a4271 DT	72	memset(line, 0, sizeof(line));
	73
	74	if (array) {
	75	for (i = 0; i < array_size; i++) {
	76	if (!array[i].nodeid)
	77	continue;
	78
	79	ret = snprintf(line + pos, len - pos, " %d:%d",
	80	array[i].slot, array[i].nodeid);
	81	if (ret >= len - pos)
	82	break;
	83	pos += ret;
	84	}
	85	} else if (ro0) {
	86	for (i = 0; i < num_slots; i++) {
	87	ret = snprintf(line + pos, len - pos, " %d:%d",
	88	ro0[i].ro_slot, ro0[i].ro_nodeid);
	89	if (ret >= len - pos)
	90	break;
	91	pos += ret;
	92	}
	93	}
	94
075f0177	95	log_rinfo(ls, "generation %u slots %d%s", gen, num_slots, line);
757a4271 DT	96	}
	97
	98	int dlm_slots_copy_in(struct dlm_ls *ls)
	99	{
	100	struct dlm_member *memb;
	101	struct dlm_rcom *rc = ls->ls_recover_buf;
	102	struct rcom_config rf = (struct rcom_config )rc->rc_buf;
	103	struct rcom_slot ro0, ro;
	104	int our_nodeid = dlm_our_nodeid();
	105	int i, num_slots;
	106	uint32_t gen;
	107
	108	if (!dlm_slots_version(&rc->rc_header))
	109	return -1;
	110
	111	gen = le32_to_cpu(rf->rf_generation);
	112	if (gen <= ls->ls_generation) {
	113	log_error(ls, "dlm_slots_copy_in gen %u old %u",
	114	gen, ls->ls_generation);
	115	}
	116	ls->ls_generation = gen;
	117
	118	num_slots = le16_to_cpu(rf->rf_num_slots);
	119	if (!num_slots)
	120	return -1;
	121
	122	ro0 = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config));
	123
	124	for (i = 0, ro = ro0; i < num_slots; i++, ro++) {
	125	ro->ro_nodeid = le32_to_cpu(ro->ro_nodeid);
	126	ro->ro_slot = le16_to_cpu(ro->ro_slot);
	127	}
	128
075f0177	129	log_slots(ls, gen, num_slots, ro0, NULL, 0);
757a4271 DT	130
	131	list_for_each_entry(memb, &ls->ls_nodes, list) {
	132	for (i = 0, ro = ro0; i < num_slots; i++, ro++) {
	133	if (ro->ro_nodeid != memb->nodeid)
	134	continue;
	135	memb->slot = ro->ro_slot;
	136	memb->slot_prev = memb->slot;
	137	break;
	138	}
	139
	140	if (memb->nodeid == our_nodeid) {
	141	if (ls->ls_slot && ls->ls_slot != memb->slot) {
	142	log_error(ls, "dlm_slots_copy_in our slot "
	143	"changed %d %d", ls->ls_slot,
	144	memb->slot);
	145	return -1;
	146	}
	147
	148	if (!ls->ls_slot)
	149	ls->ls_slot = memb->slot;
	150	}
	151
	152	if (!memb->slot) {
	153	log_error(ls, "dlm_slots_copy_in nodeid %d no slot",
	154	memb->nodeid);
	155	return -1;
	156	}
	157	}
	158
	159	return 0;
	160	}
	161
	162	/* for any nodes that do not support slots, we will not have set memb->slot
	163	in wait_status_all(), so memb->slot will remain -1, and we will not
	164	assign slots or set ls_num_slots here */
	165
	166	int dlm_slots_assign(struct dlm_ls ls, int num_slots, int *slots_size,
	167	struct dlm_slot *slots_out, uint32_t gen_out)
	168	{
	169	struct dlm_member *memb;
	170	struct dlm_slot *array;
	171	int our_nodeid = dlm_our_nodeid();
	172	int array_size, max_slots, i;
	173	int need = 0;
	174	int max = 0;
	175	int num = 0;
	176	uint32_t gen = 0;
	177
	178	/* our own memb struct will have slot -1 gen 0 */
	179
	180	list_for_each_entry(memb, &ls->ls_nodes, list) {
	181	if (memb->nodeid == our_nodeid) {
	182	memb->slot = ls->ls_slot;
	183	memb->generation = ls->ls_generation;
	184	break;
	185	}
	186	}
	187
	188	list_for_each_entry(memb, &ls->ls_nodes, list) {
	189	if (memb->generation > gen)
	190	gen = memb->generation;
	191
	192	/* node doesn't support slots */
	193
194	if (memb->slot == -1)
195	return -1;
196
197	/* node needs a slot assigned */
198
199	if (!memb->slot)
200	need++;
201
202	/* node has a slot assigned */
203
204	num++;
205
206	if (!max \|\| max < memb->slot)
207	max = memb->slot;
208
209	/* sanity check, once slot is assigned it shouldn't change */
210
211	if (memb->slot_prev && memb->slot && memb->slot_prev != memb->slot) {
212	log_error(ls, "nodeid %d slot changed %d %d",
213	memb->nodeid, memb->slot_prev, memb->slot);
214	return -1;
215	}
216	memb->slot_prev = memb->slot;
217	}
218
219	array_size = max + need;
220
221	array = kzalloc(array_size * sizeof(struct dlm_slot), GFP_NOFS);
222	if (!array)
223	return -ENOMEM;
224
225	num = 0;
226
227	/* fill in slots (offsets) that are used */
228
229	list_for_each_entry(memb, &ls->ls_nodes, list) {
230	if (!memb->slot)
231	continue;
232
233	if (memb->slot > array_size) {
234	log_error(ls, "invalid slot number %d", memb->slot);
235	kfree(array);
236	return -1;
237	}
238
239	array[memb->slot - 1].nodeid = memb->nodeid;
240	array[memb->slot - 1].slot = memb->slot;
241	num++;
242	}
243
244	/* assign new slots from unused offsets */
245
246	list_for_each_entry(memb, &ls->ls_nodes, list) {
247	if (memb->slot)
248	continue;
249
250	for (i = 0; i < array_size; i++) {
251	if (array[i].nodeid)
252	continue;
253
254	memb->slot = i + 1;
255	memb->slot_prev = memb->slot;
256	array[i].nodeid = memb->nodeid;
257	array[i].slot = memb->slot;
258	num++;
259
260	if (!ls->ls_slot && memb->nodeid == our_nodeid)
261	ls->ls_slot = memb->slot;
262	break;
263	}
264
265	if (!memb->slot) {
266	log_error(ls, "no free slot found");
267	kfree(array);
268	return -1;
269	}
270	}
271
272	gen++;
273
075f0177	274	log_slots(ls, gen, num, NULL, array, array_size);
757a4271 DT	275
	276	max_slots = (dlm_config.ci_buffer_size - sizeof(struct dlm_rcom) -
	277	sizeof(struct rcom_config)) / sizeof(struct rcom_slot);
	278
	279	if (num > max_slots) {
	280	log_error(ls, "num_slots %d exceeds max_slots %d",
	281	num, max_slots);
	282	kfree(array);
	283	return -1;
	284	}
	285
	286	*gen_out = gen;
	287	*slots_out = array;
	288	*slots_size = array_size;
	289	*num_slots = num;
	290	return 0;
	291	}
	292
e7fd4179 DT	293	static void add_ordered_member(struct dlm_ls ls, struct dlm_member new)
	294	{
	295	struct dlm_member *memb = NULL;
	296	struct list_head *tmp;
	297	struct list_head *newlist = &new->list;
	298	struct list_head *head = &ls->ls_nodes;
	299
	300	list_for_each(tmp, head) {
	301	memb = list_entry(tmp, struct dlm_member, list);
	302	if (new->nodeid < memb->nodeid)
	303	break;
	304	}
	305
	306	if (!memb)
	307	list_add_tail(newlist, head);
	308	else {
	309	/* FIXME: can use list macro here */
	310	newlist->prev = tmp->prev;
	311	newlist->next = tmp;
	312	tmp->prev->next = newlist;
	313	tmp->prev = newlist;
	314	}
	315	}
	316
60f98d18	317	static int dlm_add_member(struct dlm_ls ls, struct dlm_config_node node)
e7fd4179 DT	318	{
e7fd4179 DT	319	struct dlm_member *memb;
60f98d18	320	int error;
e7fd4179	321
573c24c4	322	memb = kzalloc(sizeof(struct dlm_member), GFP_NOFS);
e7fd4179 DT	323	if (!memb)
	324	return -ENOMEM;
	325
60f98d18	326	error = dlm_lowcomms_connect_node(node->nodeid);
391fbdc5 CC	327	if (error < 0) {
	328	kfree(memb);
	329	return error;
	330	}
	331
60f98d18 DT	332	memb->nodeid = node->nodeid;
	333	memb->weight = node->weight;
	334	memb->comm_seq = node->comm_seq;
e7fd4179 DT	335	add_ordered_member(ls, memb);
	336	ls->ls_num_nodes++;
	337	return 0;
	338	}
	339
60f98d18	340	static struct dlm_member find_memb(struct list_head head, int nodeid)
e7fd4179 DT	341	{
	342	struct dlm_member *memb;
	343
60f98d18	344	list_for_each_entry(memb, head, list) {
e7fd4179	345	if (memb->nodeid == nodeid)
60f98d18	346	return memb;
e7fd4179	347	}
60f98d18 DT	348	return NULL;
	349	}
	350
	351	int dlm_is_member(struct dlm_ls *ls, int nodeid)
	352	{
	353	if (find_memb(&ls->ls_nodes, nodeid))
	354	return 1;
90135925	355	return 0;
e7fd4179 DT	356	}
	357
	358	int dlm_is_removed(struct dlm_ls *ls, int nodeid)
	359	{
60f98d18 DT	360	if (find_memb(&ls->ls_nodes_gone, nodeid))
60f98d18 DT	361	return 1;
90135925	362	return 0;
e7fd4179 DT	363	}
	364
	365	static void clear_memb_list(struct list_head *head)
	366	{
	367	struct dlm_member *memb;
	368
	369	while (!list_empty(head)) {
	370	memb = list_entry(head->next, struct dlm_member, list);
	371	list_del(&memb->list);
	372	kfree(memb);
	373	}
	374	}
	375
	376	void dlm_clear_members(struct dlm_ls *ls)
	377	{
	378	clear_memb_list(&ls->ls_nodes);
	379	ls->ls_num_nodes = 0;
	380	}
	381
	382	void dlm_clear_members_gone(struct dlm_ls *ls)
	383	{
	384	clear_memb_list(&ls->ls_nodes_gone);
	385	}
	386
	387	static void make_member_array(struct dlm_ls *ls)
	388	{
	389	struct dlm_member *memb;
	390	int i, w, x = 0, total = 0, all_zero = 0, *array;
	391
	392	kfree(ls->ls_node_array);
	393	ls->ls_node_array = NULL;
	394
	395	list_for_each_entry(memb, &ls->ls_nodes, list) {
	396	if (memb->weight)
	397	total += memb->weight;
	398	}
	399
	400	/* all nodes revert to weight of 1 if all have weight 0 */
	401
	402	if (!total) {
	403	total = ls->ls_num_nodes;
	404	all_zero = 1;
	405	}
	406
	407	ls->ls_total_weight = total;
	408
573c24c4	409	array = kmalloc(sizeof(int) * total, GFP_NOFS);
e7fd4179 DT	410	if (!array)
	411	return;
	412
	413	list_for_each_entry(memb, &ls->ls_nodes, list) {
	414	if (!all_zero && !memb->weight)
	415	continue;
	416
	417	if (all_zero)
	418	w = 1;
	419	else
	420	w = memb->weight;
	421
	422	DLM_ASSERT(x < total, printk("total %d x %d\n", total, x););
	423
	424	for (i = 0; i < w; i++)
	425	array[x++] = memb->nodeid;
	426	}
	427
	428	ls->ls_node_array = array;
	429	}
	430
	431	/* send a status request to all members just to establish comms connections */
	432
f6db1b8e	433	static int ping_members(struct dlm_ls *ls)
e7fd4179 DT	434	{
e7fd4179 DT	435	struct dlm_member *memb;
f6db1b8e DT	436	int error = 0;
	437
	438	list_for_each_entry(memb, &ls->ls_nodes, list) {
	439	error = dlm_recovery_stopped(ls);
	440	if (error)
	441	break;
757a4271	442	error = dlm_rcom_status(ls, memb->nodeid, 0);
f6db1b8e DT	443	if (error)
	444	break;
	445	}
	446	if (error)
075f0177	447	log_rinfo(ls, "ping_members aborted %d last nodeid %d",
faa0f267	448	error, ls->ls_recover_nodeid);
f6db1b8e	449	return error;
e7fd4179 DT	450	}
e7fd4179 DT	451
60f98d18 DT	452	static void dlm_lsop_recover_prep(struct dlm_ls *ls)
	453	{
	454	if (!ls->ls_ops \|\| !ls->ls_ops->recover_prep)
	455	return;
	456	ls->ls_ops->recover_prep(ls->ls_ops_arg);
	457	}
	458
	459	static void dlm_lsop_recover_slot(struct dlm_ls ls, struct dlm_member memb)
	460	{
	461	struct dlm_slot slot;
	462	uint32_t seq;
	463	int error;
	464
	465	if (!ls->ls_ops \|\| !ls->ls_ops->recover_slot)
	466	return;
	467
	468	/* if there is no comms connection with this node
	469	or the present comms connection is newer
	470	than the one when this member was added, then
	471	we consider the node to have failed (versus
	472	being removed due to dlm_release_lockspace) */
	473
	474	error = dlm_comm_seq(memb->nodeid, &seq);
	475
	476	if (!error && seq == memb->comm_seq)
	477	return;
	478
	479	slot.nodeid = memb->nodeid;
	480	slot.slot = memb->slot;
	481
	482	ls->ls_ops->recover_slot(ls->ls_ops_arg, &slot);
	483	}
	484
	485	void dlm_lsop_recover_done(struct dlm_ls *ls)
	486	{
	487	struct dlm_member *memb;
	488	struct dlm_slot *slots;
	489	int i, num;
	490
	491	if (!ls->ls_ops \|\| !ls->ls_ops->recover_done)
	492	return;
	493
	494	num = ls->ls_num_nodes;
	495
	496	slots = kzalloc(num * sizeof(struct dlm_slot), GFP_KERNEL);
	497	if (!slots)
	498	return;
	499
	500	i = 0;
	501	list_for_each_entry(memb, &ls->ls_nodes, list) {
	502	if (i == num) {
	503	log_error(ls, "dlm_lsop_recover_done bad num %d", num);
	504	goto out;
	505	}
	506	slots[i].nodeid = memb->nodeid;
	507	slots[i].slot = memb->slot;
	508	i++;
	509	}
	510
	511	ls->ls_ops->recover_done(ls->ls_ops_arg, slots, num,
	512	ls->ls_slot, ls->ls_generation);
	513	out:
	514	kfree(slots);
	515	}
516
517	static struct dlm_config_node find_config_node(struct dlm_recover rv,
518	int nodeid)
519	{
520	int i;
521
522	for (i = 0; i < rv->nodes_count; i++) {
523	if (rv->nodes[i].nodeid == nodeid)
524	return &rv->nodes[i];
525	}
526	return NULL;
527	}
528
e7fd4179 DT	529	int dlm_recover_members(struct dlm_ls ls, struct dlm_recover rv, int *neg_out)
	530	{
	531	struct dlm_member memb, safe;
60f98d18 DT	532	struct dlm_config_node *node;
60f98d18 DT	533	int i, error, neg = 0, low = -1;
e7fd4179	534
91c0dc93 DT	535	/* previously removed members that we've not finished removing need to
	536	count as a negative change so the "neg" recovery steps will happen */
	537
	538	list_for_each_entry(memb, &ls->ls_nodes_gone, list) {
075f0177	539	log_rinfo(ls, "prev removed member %d", memb->nodeid);
91c0dc93 DT	540	neg++;
	541	}
	542
e7fd4179 DT	543	/* move departed members from ls_nodes to ls_nodes_gone */
	544
	545	list_for_each_entry_safe(memb, safe, &ls->ls_nodes, list) {
60f98d18 DT	546	node = find_config_node(rv, memb->nodeid);
	547	if (node && !node->new)
	548	continue;
e7fd4179	549
60f98d18	550	if (!node) {
075f0177	551	log_rinfo(ls, "remove member %d", memb->nodeid);
60f98d18 DT	552	} else {
60f98d18 DT	553	/* removed and re-added */
075f0177	554	log_rinfo(ls, "remove member %d comm_seq %u %u",
60f98d18	555	memb->nodeid, memb->comm_seq, node->comm_seq);
e7fd4179	556	}
d44e0fc7	557
d44e0fc7	558	neg++;
60f98d18 DT	559	list_move(&memb->list, &ls->ls_nodes_gone);
	560	ls->ls_num_nodes--;
	561	dlm_lsop_recover_slot(ls, memb);
d44e0fc7 DT	562	}
d44e0fc7 DT	563
e7fd4179 DT	564	/* add new members to ls_nodes */
e7fd4179 DT	565
60f98d18 DT	566	for (i = 0; i < rv->nodes_count; i++) {
	567	node = &rv->nodes[i];
	568	if (dlm_is_member(ls, node->nodeid))
e7fd4179	569	continue;
60f98d18	570	dlm_add_member(ls, node);
075f0177	571	log_rinfo(ls, "add member %d", node->nodeid);
e7fd4179 DT	572	}
	573
	574	list_for_each_entry(memb, &ls->ls_nodes, list) {
	575	if (low == -1 \|\| memb->nodeid < low)
	576	low = memb->nodeid;
	577	}
	578	ls->ls_low_nodeid = low;
	579
	580	make_member_array(ls);
e7fd4179 DT	581	*neg_out = neg;
e7fd4179 DT	582
f6db1b8e	583	error = ping_members(ls);
8b0e7b2c DT	584	if (!error \|\| error == -EPROTO) {
	585	/* new_lockspace() may be waiting to know if the config
	586	is good or bad */
	587	ls->ls_members_result = error;
	588	complete(&ls->ls_members_done);
	589	}
e7fd4179	590
075f0177	591	log_rinfo(ls, "dlm_recover_members %d nodes", ls->ls_num_nodes);
e7fd4179 DT	592	return error;
	593	}
	594
c36258b5 DT	595	/* Userspace guarantees that dlm_ls_stop() has completed on all nodes before
c36258b5 DT	596	dlm_ls_start() is called on any of them to start the new recovery. */
e7fd4179 DT	597
	598	int dlm_ls_stop(struct dlm_ls *ls)
	599	{
	600	int new;
	601
	602	/*
c36258b5 DT	603	* Prevent dlm_recv from being in the middle of something when we do
	604	* the stop. This includes ensuring dlm_recv isn't processing a
	605	* recovery message (rcom), while dlm_recoverd is aborting and
	606	* resetting things from an in-progress recovery. i.e. we want
	607	* dlm_recoverd to abort its recovery without worrying about dlm_recv
	608	* processing an rcom at the same time. Stopping dlm_recv also makes
	609	* it easy for dlm_receive_message() to check locking stopped and add a
	610	* message to the requestqueue without races.
	611	*/
	612
	613	down_write(&ls->ls_recv_active);
	614
	615	/*
475f230c	616	* Abort any recovery that's in progress (see RECOVER_STOP,
c36258b5 DT	617	* dlm_recovery_stopped()) and tell any other threads running in the
c36258b5 DT	618	* dlm to quit any processing (see RUNNING, dlm_locking_stopped()).
e7fd4179 DT	619	*/
	620
	621	spin_lock(&ls->ls_recover_lock);
475f230c	622	set_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
e7fd4179 DT	623	new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags);
	624	ls->ls_recover_seq++;
	625	spin_unlock(&ls->ls_recover_lock);
	626
c36258b5 DT	627	/*
	628	* Let dlm_recv run again, now any normal messages will be saved on the
	629	* requestqueue for later.
	630	*/
	631
	632	up_write(&ls->ls_recv_active);
	633
e7fd4179 DT	634	/*
e7fd4179 DT	635	* This in_recovery lock does two things:
e7fd4179	636	* 1) Keeps this function from returning until all threads are out
c41b20e7	637	* of locking routines and locking is truly stopped.
e7fd4179 DT	638	* 2) Keeps any new requests from being processed until it's unlocked
	639	* when recovery is complete.
	640	*/
	641
475f230c DT	642	if (new) {
	643	set_bit(LSFL_RECOVER_DOWN, &ls->ls_flags);
	644	wake_up_process(ls->ls_recoverd_task);
	645	wait_event(ls->ls_recover_lock_wait,
	646	test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags));
	647	}
e7fd4179 DT	648
	649	/*
	650	* The recoverd suspend/resume makes sure that dlm_recoverd (if
475f230c	651	* running) has noticed RECOVER_STOP above and quit processing the
c36258b5	652	* previous recovery.
e7fd4179 DT	653	*/
	654
	655	dlm_recoverd_suspend(ls);
757a4271 DT	656
	657	spin_lock(&ls->ls_recover_lock);
	658	kfree(ls->ls_slots);
	659	ls->ls_slots = NULL;
	660	ls->ls_num_slots = 0;
	661	ls->ls_slots_size = 0;
e7fd4179	662	ls->ls_recover_status = 0;
757a4271 DT	663	spin_unlock(&ls->ls_recover_lock);
757a4271 DT	664
e7fd4179	665	dlm_recoverd_resume(ls);
3ae1acf9 DT	666
	667	if (!ls->ls_recover_begin)
	668	ls->ls_recover_begin = jiffies;
60f98d18 DT	669
60f98d18 DT	670	dlm_lsop_recover_prep(ls);
e7fd4179 DT	671	return 0;
	672	}
	673
	674	int dlm_ls_start(struct dlm_ls *ls)
	675	{
	676	struct dlm_recover rv = NULL, rv_old;
60f98d18 DT	677	struct dlm_config_node *nodes;
60f98d18 DT	678	int error, count;
e7fd4179	679
573c24c4	680	rv = kzalloc(sizeof(struct dlm_recover), GFP_NOFS);
e7fd4179 DT	681	if (!rv)
e7fd4179 DT	682	return -ENOMEM;
e7fd4179	683
60f98d18	684	error = dlm_config_nodes(ls->ls_name, &nodes, &count);
d44e0fc7	685	if (error < 0)
e7fd4179 DT	686	goto fail;
	687
	688	spin_lock(&ls->ls_recover_lock);
	689
	690	/* the lockspace needs to be stopped before it can be started */
	691
	692	if (!dlm_locking_stopped(ls)) {
	693	spin_unlock(&ls->ls_recover_lock);
	694	log_error(ls, "start ignored: lockspace running");
	695	error = -EINVAL;
	696	goto fail;
	697	}
	698
60f98d18 DT	699	rv->nodes = nodes;
60f98d18 DT	700	rv->nodes_count = count;
e7fd4179 DT	701	rv->seq = ++ls->ls_recover_seq;
	702	rv_old = ls->ls_recover_args;
	703	ls->ls_recover_args = rv;
	704	spin_unlock(&ls->ls_recover_lock);
	705
	706	if (rv_old) {
d44e0fc7	707	log_error(ls, "unused recovery %llx %d",
60f98d18 DT	708	(unsigned long long)rv_old->seq, rv_old->nodes_count);
60f98d18 DT	709	kfree(rv_old->nodes);
e7fd4179 DT	710	kfree(rv_old);
	711	}
	712
475f230c DT	713	set_bit(LSFL_RECOVER_WORK, &ls->ls_flags);
475f230c DT	714	wake_up_process(ls->ls_recoverd_task);
e7fd4179 DT	715	return 0;
	716
	717	fail:
	718	kfree(rv);
60f98d18	719	kfree(nodes);
e7fd4179 DT	720	return error;
	721	}
	722