dlmglue.c

来自「ocfs1.4.1 oracle分布式文件系统」· C语言 代码 · 共 2,359 行 · 第 1/5 页

	int ret = 0;
	enum dlm_status status = DLM_NORMAL;
	unsigned long flags;

	mlog_entry_void();

	mlog(0, "lock %s, level = %d, flags = %d\n", lockres->l_name, level,
	     dlm_flags);

	spin_lock_irqsave(&lockres->l_lock, flags);
	if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
	    (lockres->l_flags & OCFS2_LOCK_BUSY)) {
		spin_unlock_irqrestore(&lockres->l_lock, flags);
		goto bail;
	}

	lockres->l_action = OCFS2_AST_ATTACH;
	lockres->l_requested = level;
	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	status = dlmlock(osb->dlm,
			 level,
			 &lockres->l_lksb,
			 dlm_flags,
			 lockres->l_name,
			 OCFS2_LOCK_ID_MAX_LEN - 1,
			 ocfs2_locking_ast,
			 lockres,
			 ocfs2_blocking_ast);
	if (status != DLM_NORMAL) {
		ocfs2_log_dlm_error("dlmlock", status, lockres);
		ret = -EINVAL;
		ocfs2_recover_from_dlm_error(lockres, 1);
	}

	mlog(0, "lock %s, successfull return from dlmlock\n", lockres->l_name);

bail:
	mlog_exit(ret);
	return ret;
}

static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
					int flag)
{
	unsigned long flags;
	int ret;

	spin_lock_irqsave(&lockres->l_lock, flags);
	ret = lockres->l_flags & flag;
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	return ret;
}

static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)

{
	wait_event(lockres->l_event,
		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
}

static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)

{
	wait_event(lockres->l_event,
		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
}

/* predict what lock level we'll be dropping down to on behalf
 * of another node, and return true if the currently wanted
 * level will be compatible with it. */
static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
						     int wanted)
{
	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));

	return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
}

static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
{
	INIT_LIST_HEAD(&mw->mw_item);
	init_completion(&mw->mw_complete);
	ocfs2_init_start_time(mw);
}

static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
{
	wait_for_completion(&mw->mw_complete);
	/* Re-arm the completion in case we want to wait on it again */
	INIT_COMPLETION(mw->mw_complete);
	return mw->mw_status;
}

static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
				    struct ocfs2_mask_waiter *mw,
				    unsigned long mask,
				    unsigned long goal)
{
	BUG_ON(!list_empty(&mw->mw_item));

	assert_spin_locked(&lockres->l_lock);

	list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
	mw->mw_mask = mask;
	mw->mw_goal = goal;
}

/* returns 0 if the mw that was removed was already satisfied, -EBUSY
 * if the mask still hadn't reached its goal */
static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
				      struct ocfs2_mask_waiter *mw)
{
	unsigned long flags;
	int ret = 0;

	spin_lock_irqsave(&lockres->l_lock, flags);
	if (!list_empty(&mw->mw_item)) {
		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
			ret = -EBUSY;

		list_del_init(&mw->mw_item);
		init_completion(&mw->mw_complete);
	}
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	return ret;

}

static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
					     struct ocfs2_lock_res *lockres)
{
	int ret;

	ret = wait_for_completion_interruptible(&mw->mw_complete);
	if (ret)
		lockres_remove_mask_waiter(lockres, mw);
	else
		ret = mw->mw_status;
	/* Re-arm the completion in case we want to wait on it again */
	INIT_COMPLETION(mw->mw_complete);
	return ret;
}

static int ocfs2_cluster_lock(struct ocfs2_super *osb,
			      struct ocfs2_lock_res *lockres,
			      int level,
			      int lkm_flags,
			      int arg_flags)
{
	struct ocfs2_mask_waiter mw;
	enum dlm_status status;
	int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
	int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
	unsigned long flags;

	mlog_entry_void();

	ocfs2_init_mask_waiter(&mw);

	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
		lkm_flags |= LKM_VALBLK;

again:
	wait = 0;

	if (catch_signals && signal_pending(current)) {
		ret = -ERESTARTSYS;
		goto out;
	}

	spin_lock_irqsave(&lockres->l_lock, flags);

	mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
			"Cluster lock called on freeing lockres %s! flags "
			"0x%lx\n", lockres->l_name, lockres->l_flags);

	/* We only compare against the currently granted level
	 * here. If the lock is blocked waiting on a downconvert,
	 * we'll get caught below. */
	if (lockres->l_flags & OCFS2_LOCK_BUSY &&
	    level > lockres->l_level) {
		/* is someone sitting in dlm_lock? If so, wait on
		 * them. */
		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
		wait = 1;
		goto unlock;
	}

	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
		/* is the lock is currently blocked on behalf of
		 * another node */
		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
		wait = 1;
		goto unlock;
	}

	if (level > lockres->l_level) {
		if (lockres->l_action != OCFS2_AST_INVALID)
			mlog(ML_ERROR, "lockres %s has action %u pending\n",
			     lockres->l_name, lockres->l_action);

		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
			lockres->l_action = OCFS2_AST_ATTACH;
			lkm_flags &= ~LKM_CONVERT;
		} else {
			lockres->l_action = OCFS2_AST_CONVERT;
			lkm_flags |= LKM_CONVERT;
		}

		lockres->l_requested = level;
		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
		spin_unlock_irqrestore(&lockres->l_lock, flags);

		BUG_ON(level == LKM_IVMODE);
		BUG_ON(level == LKM_NLMODE);

		mlog(0, "lock %s, convert from %d to level = %d\n",
		     lockres->l_name, lockres->l_level, level);

		/* call dlm_lock to upgrade lock now */
		status = dlmlock(osb->dlm,
				 level,
				 &lockres->l_lksb,
				 lkm_flags,
				 lockres->l_name,
				 OCFS2_LOCK_ID_MAX_LEN - 1,
				 ocfs2_locking_ast,
				 lockres,
				 ocfs2_blocking_ast);
		if (status != DLM_NORMAL) {
			if ((lkm_flags & LKM_NOQUEUE) &&
			    (status == DLM_NOTQUEUED))
				ret = -EAGAIN;
			else {
				ocfs2_log_dlm_error("dlmlock", status,
						    lockres);
				ret = -EINVAL;
			}
			ocfs2_recover_from_dlm_error(lockres, 1);
			goto out;
		}

		mlog(0, "lock %s, successfull return from dlmlock\n",
		     lockres->l_name);

		/* At this point we've gone inside the dlm and need to
		 * complete our work regardless. */
		catch_signals = 0;

		/* wait for busy to clear and carry on */
		goto again;
	}

	/* Ok, if we get here then we're good to go. */
	ocfs2_inc_holders(lockres, level);

	ret = 0;
unlock:
	spin_unlock_irqrestore(&lockres->l_lock, flags);
out:
	/*
	 * This is helping work around a lock inversion between the page lock
	 * and dlm locks.  One path holds the page lock while calling aops
	 * which block acquiring dlm locks.  The voting thread holds dlm
	 * locks while acquiring page locks while down converting data locks.
	 * This block is helping an aop path notice the inversion and back
	 * off to unlock its page lock before trying the dlm lock again.
	 */
	if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
	    mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
		wait = 0;
		if (lockres_remove_mask_waiter(lockres, &mw))
			ret = -EAGAIN;
		else
			goto again;
	}
	if (wait) {
		ret = ocfs2_wait_for_mask(&mw);
		if (ret == 0)
			goto again;
		mlog_errno(ret);
	}
	ocfs2_update_lock_stats(lockres, level, &mw, ret);

	mlog_exit(ret);
	return ret;
}

static void ocfs2_cluster_unlock(struct ocfs2_super *osb,
				 struct ocfs2_lock_res *lockres,
				 int level)
{
	unsigned long flags;

	mlog_entry_void();
	spin_lock_irqsave(&lockres->l_lock, flags);
	ocfs2_dec_holders(lockres, level);
	ocfs2_downconvert_on_unlock(osb, lockres);
	spin_unlock_irqrestore(&lockres->l_lock, flags);
	mlog_exit_void();
}

static int ocfs2_create_new_lock(struct ocfs2_super *osb,
				 struct ocfs2_lock_res *lockres,
				 int ex,
				 int local)
{
	int level =  ex ? LKM_EXMODE : LKM_PRMODE;
	unsigned long flags;
	int lkm_flags = local ? LKM_LOCAL : 0;

	spin_lock_irqsave(&lockres->l_lock, flags);
	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
	lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
	spin_unlock_irqrestore(&lockres->l_lock, flags);

	return ocfs2_lock_create(osb, lockres, level, lkm_flags);
}

/* Grants us an EX lock on the data and metadata resources, skipping
 * the normal cluster directory lookup. Use this ONLY on newly created
 * inodes which other nodes can't possibly see, and which haven't been
 * hashed in the inode hash yet. This can give us a good performance
 * increase as it'll skip the network broadcast normally associated
 * with creating a new lock resource. */
int ocfs2_create_new_inode_locks(struct inode *inode)
{
	int ret;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	BUG_ON(!inode);
	BUG_ON(!ocfs2_inode_is_new(inode));

	mlog_entry_void();

	mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);

	/* NOTE: That we don't increment any of the holder counts, nor
	 * do we add anything to a journal handle. Since this is
	 * supposed to be a new inode which the cluster doesn't know
	 * about yet, there is no need to.  As far as the LVB handling
	 * is concerned, this is basically like acquiring an EX lock
	 * on a resource which has an invalid one -- we'll set it
	 * valid when we release the EX. */

	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
	if (ret) {
		mlog_errno(ret);
		goto bail;
	}

	/*
	 * We don't want to use LKM_LOCAL on a meta data lock as they
	 * don't use a generation in their lock names.
	 */
	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
	if (ret) {
		mlog_errno(ret);
		goto bail;
	}

	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
	if (ret) {
		mlog_errno(ret);
		goto bail;
	}

bail:
	mlog_exit(ret);
	return ret;
}

int ocfs2_rw_lock(struct inode *inode, int write)
{
	int status, level;
	struct ocfs2_lock_res *lockres;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	BUG_ON(!inode);

	mlog_entry_void();

	mlog(0, "inode %llu take %s RW lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
	     write ? "EXMODE" : "PRMODE");

	if (ocfs2_mount_local(osb))
		return 0;

	lockres = &OCFS2_I(inode)->ip_rw_lockres;

	level = write ? LKM_EXMODE : LKM_PRMODE;

	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
				    0);
	if (status < 0)
		mlog_errno(status);

	mlog_exit(status);
	return status;
}

void ocfs2_rw_unlock(struct inode *inode, int write)
{
	int level = write ? LKM_EXMODE : LKM_PRMODE;
	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	mlog_entry_void();

	mlog(0, "inode %llu drop %s RW lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
	     write ? "EXMODE" : "PRMODE");

	if (!ocfs2_mount_local(osb))
		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);

	mlog_exit_void();
}

/*
 * ocfs2_open_lock always get PR mode lock.
 */
int ocfs2_open_lock(struct inode *inode)
{
	int status = 0;
	struct ocfs2_lock_res *lockres;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	BUG_ON(!inode);

	mlog_entry_void();

	mlog(0, "inode %llu take PRMODE open lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno);

	if (ocfs2_mount_local(osb))
		goto out;

	lockres = &OCFS2_I(inode)->ip_open_lockres;

	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
				    LKM_PRMODE, 0, 0);
	if (status < 0)
		mlog_errno(status);

out:
	mlog_exit(status);
	return status;
}

int ocfs2_try_open_lock(struct inode *inode, int write)
{
	int status = 0, level;
	struct ocfs2_lock_res *lockres;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	BUG_ON(!inode);

	mlog_entry_void();

	mlog(0, "inode %llu try to take %s open lock\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
	     write ? "EXMODE" : "PRMODE");

	if (ocfs2_mount_local(osb))
		goto out;