dlmlock.c

linux 内核源代码

/* -*- mode: c; c-basic-offset: 8; -*-
 * vim: noexpandtab sw=8 ts=8 sts=0:
 *
 * dlmlock.c
 *
 * underlying calls for lock creation
 *
 * Copyright (C) 2004 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 *
 */


#include <linux/module.h>
#include <linux/fs.h>
#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>
#include <linux/blkdev.h>
#include <linux/socket.h>
#include <linux/inet.h>
#include <linux/spinlock.h>
#include <linux/delay.h>


#include "cluster/heartbeat.h"
#include "cluster/nodemanager.h"
#include "cluster/tcp.h"

#include "dlmapi.h"
#include "dlmcommon.h"

#include "dlmconvert.h"

#define MLOG_MASK_PREFIX ML_DLM
#include "cluster/masklog.h"

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,
					       struct dlm_lock_resource *res,
					       struct dlm_lock *lock, int flags);
static void dlm_init_lock(struct dlm_lock *newlock, int type,
			  u8 node, u64 cookie);
static void dlm_lock_release(struct kref *kref);
static void dlm_lock_detach_lockres(struct dlm_lock *lock);

/* Tell us whether we can grant a new lock request.
 * locking:
 *   caller needs:  res->spinlock
 *   taken:         none
 *   held on exit:  none
 * returns: 1 if the lock can be granted, 0 otherwise.
 */
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);

		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);

		if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
			return 0;
	}

	return 1;
}

/* performs lock creation at the lockres master site
 * locking:
 *   caller needs:  none
 *   taken:         takes and drops res->spinlock
 *   held on exit:  none
 * returns: DLM_NORMAL, DLM_NOTQUEUED
 */
static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
				      struct dlm_lock_resource *res,
				      struct dlm_lock *lock, int flags)
{
	int call_ast = 0, kick_thread = 0;
	enum dlm_status status = DLM_NORMAL;

	mlog_entry("type=%d\n", lock->ml.type);

	spin_lock(&res->spinlock);
	/* if called from dlm_create_lock_handler, need to
	 * ensure it will not sleep in dlm_wait_on_lockres */
	status = __dlm_lockres_state_to_status(res);
	if (status != DLM_NORMAL &&
	    lock->ml.node != dlm->node_num) {
		/* erf.  state changed after lock was dropped. */
		spin_unlock(&res->spinlock);
		dlm_error(status);
		return status;
	}
	__dlm_wait_on_lockres(res);
	__dlm_lockres_reserve_ast(res);

	if (dlm_can_grant_new_lock(res, lock)) {
		mlog(0, "I can grant this lock right away\n");
		/* got it right away */
		lock->lksb->status = DLM_NORMAL;
		status = DLM_NORMAL;
		dlm_lock_get(lock);
		list_add_tail(&lock->list, &res->granted);

		/* for the recovery lock, we can't allow the ast
		 * to be queued since the dlmthread is already
		 * frozen.  but the recovery lock is always locked
		 * with LKM_NOQUEUE so we do not need the ast in
		 * this special case */
		if (!dlm_is_recovery_lock(res->lockname.name,
					  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) {
			status = DLM_NOTQUEUED;
			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 {
			dlm_lock_get(lock);
			list_add_tail(&lock->list, &res->blocked);
			kick_thread = 1;
		}
	}
	/* reduce the inflight count, this may result in the lockres
	 * being purged below during calc_usage */
	if (lock->ml.node == dlm->node_num)
		dlm_lockres_drop_inflight_ref(dlm, res);

	spin_unlock(&res->spinlock);
	wake_up(&res->wq);

	/* either queue the ast or release it */
	if (call_ast)
		dlm_queue_ast(dlm, lock);
	else
		dlm_lockres_release_ast(dlm, res);

	dlm_lockres_calc_usage(dlm, res);
	if (kick_thread)
		dlm_kick_thread(dlm, res);

	return status;
}

void dlm_revert_pending_lock(struct dlm_lock_resource *res,
			     struct dlm_lock *lock)
{
	/* remove from local queue if it failed */
	list_del_init(&lock->list);
	lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
}


/*
 * locking:
 *   caller needs:  none
 *   taken:         takes and drops res->spinlock
 *   held on exit:  none
 * returns: DLM_DENIED, DLM_RECOVERING, or net status
 */
static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
				      struct dlm_lock_resource *res,
				      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,
	     res->lockname.name, flags);

	spin_lock(&res->spinlock);

	/* will exit this call with spinlock held */
	__dlm_wait_on_lockres(res);
	res->state |= DLM_LOCK_RES_IN_PROGRESS;

	/* add lock to local (secondary) queue */
	dlm_lock_get(lock);
	list_add_tail(&lock->list, &res->blocked);
	lock->lock_pending = 1;
	spin_unlock(&res->spinlock);

	/* spec seems to say that you will get DLM_NORMAL when the lock
	 * has been queued, meaning we need to wait for a reply here. */
	status = dlm_send_remote_lock_request(dlm, res, lock, flags);

	spin_lock(&res->spinlock);
	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
	lock->lock_pending = 0;
	if (status != DLM_NORMAL) {
		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);

	if (lockres_changed)
		dlm_lockres_calc_usage(dlm, res);

	wake_up(&res->wq);
	return status;
}


/* for remote lock creation.
 * locking:
 *   caller needs:  none, but need res->state & DLM_LOCK_RES_IN_PROGRESS
 *   taken:         none
 *   held on exit:  none
 * returns: DLM_NOLOCKMGR, or net status
 */
static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
					       struct dlm_lock_resource *res,
					       struct dlm_lock *lock, int flags)
{
	struct dlm_create_lock create;
	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;
	create.cookie = lock->ml.cookie;
	create.namelen = res->lockname.len;
	create.flags = cpu_to_be32(flags);
	memcpy(create.name, res->lockname.name, create.namelen);

	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
		if (ret == DLM_REJECTED) {
			mlog(ML_ERROR, "%s:%.*s: BUG.  this is a stale lockres "
			     "no longer owned by %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)) {
			ret = DLM_RECOVERING;
			mlog(0, "node %u died so returning DLM_RECOVERING "
			     "from lock message!\n", res->owner);
		} else {
			ret = dlm_err_to_dlm_status(tmpret);
		}
	}

	return ret;
}

void dlm_lock_get(struct dlm_lock *lock)
{
	kref_get(&lock->lock_refs);
}

void dlm_lock_put(struct dlm_lock *lock)
{
	kref_put(&lock->lock_refs, dlm_lock_release);
}

static void dlm_lock_release(struct kref *kref)
{
	struct dlm_lock *lock;

	lock = container_of(kref, struct dlm_lock, lock_refs);

	BUG_ON(!list_empty(&lock->list));
	BUG_ON(!list_empty(&lock->ast_list));
	BUG_ON(!list_empty(&lock->bast_list));
	BUG_ON(lock->ast_pending);
	BUG_ON(lock->bast_pending);

	dlm_lock_detach_lockres(lock);

	if (lock->lksb_kernel_allocated) {
		mlog(0, "freeing kernel-allocated lksb\n");
		kfree(lock->lksb);
	}
	kfree(lock);
}

/* associate a lock with it's lockres, getting a ref on the lockres */
void dlm_lock_attach_lockres(struct dlm_lock *lock,
			     struct dlm_lock_resource *res)
{
	dlm_lockres_get(res);
	lock->lockres = res;
}

/* drop ref on lockres, if there is still one associated with lock */
static void dlm_lock_detach_lockres(struct dlm_lock *lock)
{
	struct dlm_lock_resource *res;

	res = lock->lockres;
	if (res) {
		lock->lockres = NULL;
		mlog(0, "removing lock's lockres reference\n");