userdlm.c

ocfs1.4.1 oracle分布式文件系统

		spin_unlock(&lockres->l_lock);
		mlog(0, "can't downconvert for ex: ro = %u, ex = %u\n",
			lockres->l_ro_holders, lockres->l_ex_holders);
		goto drop_ref;
	}

	if ((lockres->l_blocking == LKM_PRMODE)
	    && lockres->l_ex_holders) {
		spin_unlock(&lockres->l_lock);
		mlog(0, "can't downconvert for pr: ex = %u\n",
			lockres->l_ex_holders);
		goto drop_ref;
	}

	/* yay, we can downconvert now. */
	new_level = user_highest_compat_lock_level(lockres->l_blocking);
	lockres->l_requested = new_level;
	lockres->l_flags |= USER_LOCK_BUSY;
	mlog(0, "Downconvert lock from %d to %d\n",
		lockres->l_level, new_level);
	spin_unlock(&lockres->l_lock);

	/* need lock downconvert request now... */
	status = dlmlock(dlm,
			 new_level,
			 &lockres->l_lksb,
			 LKM_CONVERT|LKM_VALBLK,
			 lockres->l_name,
			 lockres->l_namelen,
			 user_ast,
			 lockres,
			 user_bast);
	if (status != DLM_NORMAL) {
		user_log_dlm_error("dlmlock", status, lockres);
		user_recover_from_dlm_error(lockres);
	}

drop_ref:
	user_dlm_drop_inode_ref(lockres);
}

static inline void user_dlm_inc_holders(struct user_lock_res *lockres,
					int level)
{
	switch(level) {
	case LKM_EXMODE:
		lockres->l_ex_holders++;
		break;
	case LKM_PRMODE:
		lockres->l_ro_holders++;
		break;
	default:
		BUG();
	}
}

/* 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
user_may_continue_on_blocked_lock(struct user_lock_res *lockres,
				  int wanted)
{
	BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));

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

int user_dlm_cluster_lock(struct user_lock_res *lockres,
			  int level,
			  int lkm_flags)
{
	int status, local_flags;
	struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);

	if (level != LKM_EXMODE &&
	    level != LKM_PRMODE) {
		mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
		     lockres->l_namelen, lockres->l_name);
		status = -EINVAL;
		goto bail;
	}

	mlog(0, "lockres %.*s: asking for %s lock, passed flags = 0x%x\n",
	     lockres->l_namelen, lockres->l_name,
	     (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE",
	     lkm_flags);

again:
	if (signal_pending(current)) {
		status = -ERESTARTSYS;
		goto bail;
	}

	spin_lock(&lockres->l_lock);

	/* 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 & USER_LOCK_BUSY) &&
	    (level > lockres->l_level)) {
		/* is someone sitting in dlm_lock? If so, wait on
		 * them. */
		spin_unlock(&lockres->l_lock);

		user_wait_on_busy_lock(lockres);
		goto again;
	}

	if ((lockres->l_flags & USER_LOCK_BLOCKED) &&
	    (!user_may_continue_on_blocked_lock(lockres, level))) {
		/* is the lock is currently blocked on behalf of
		 * another node */
		spin_unlock(&lockres->l_lock);

		user_wait_on_blocked_lock(lockres);
		goto again;
	}

	if (level > lockres->l_level) {
		local_flags = lkm_flags | LKM_VALBLK;
		if (lockres->l_level != LKM_IVMODE)
			local_flags |= LKM_CONVERT;

		lockres->l_requested = level;
		lockres->l_flags |= USER_LOCK_BUSY;
		spin_unlock(&lockres->l_lock);

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

		/* call dlm_lock to upgrade lock now */
		status = dlmlock(dlm,
				 level,
				 &lockres->l_lksb,
				 local_flags,
				 lockres->l_name,
				 lockres->l_namelen,
				 user_ast,
				 lockres,
				 user_bast);
		if (status != DLM_NORMAL) {
			if ((lkm_flags & LKM_NOQUEUE) &&
			    (status == DLM_NOTQUEUED))
				status = -EAGAIN;
			else {
				user_log_dlm_error("dlmlock", status, lockres);
				status = -EINVAL;
			}
			user_recover_from_dlm_error(lockres);
			goto bail;
		}

		user_wait_on_busy_lock(lockres);
		goto again;
	}

	user_dlm_inc_holders(lockres, level);
	spin_unlock(&lockres->l_lock);

	status = 0;
bail:
	return status;
}

static inline void user_dlm_dec_holders(struct user_lock_res *lockres,
					int level)
{
	switch(level) {
	case LKM_EXMODE:
		BUG_ON(!lockres->l_ex_holders);
		lockres->l_ex_holders--;
		break;
	case LKM_PRMODE:
		BUG_ON(!lockres->l_ro_holders);
		lockres->l_ro_holders--;
		break;
	default:
		BUG();
	}
}

void user_dlm_cluster_unlock(struct user_lock_res *lockres,
			     int level)
{
	if (level != LKM_EXMODE &&
	    level != LKM_PRMODE) {
		mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
		     lockres->l_namelen, lockres->l_name);
		return;
	}

	spin_lock(&lockres->l_lock);
	user_dlm_dec_holders(lockres, level);
	__user_dlm_cond_queue_lockres(lockres);
	spin_unlock(&lockres->l_lock);
}

void user_dlm_write_lvb(struct inode *inode,
			const char *val,
			unsigned int len)
{
	struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
	char *lvb = lockres->l_lksb.lvb;

	BUG_ON(len > DLM_LVB_LEN);

	spin_lock(&lockres->l_lock);

	BUG_ON(lockres->l_level < LKM_EXMODE);
	memcpy(lvb, val, len);

	spin_unlock(&lockres->l_lock);
}

void user_dlm_read_lvb(struct inode *inode,
		       char *val,
		       unsigned int len)
{
	struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
	char *lvb = lockres->l_lksb.lvb;

	BUG_ON(len > DLM_LVB_LEN);

	spin_lock(&lockres->l_lock);

	BUG_ON(lockres->l_level < LKM_PRMODE);
	memcpy(val, lvb, len);

	spin_unlock(&lockres->l_lock);
}

void user_dlm_lock_res_init(struct user_lock_res *lockres,
			    struct dentry *dentry)
{
	memset(lockres, 0, sizeof(*lockres));

	spin_lock_init(&lockres->l_lock);
	init_waitqueue_head(&lockres->l_event);
	lockres->l_level = LKM_IVMODE;
	lockres->l_requested = LKM_IVMODE;
	lockres->l_blocking = LKM_IVMODE;

	/* should have been checked before getting here. */
	BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN);

	memcpy(lockres->l_name,
	       dentry->d_name.name,
	       dentry->d_name.len);
	lockres->l_namelen = dentry->d_name.len;
}

int user_dlm_destroy_lock(struct user_lock_res *lockres)
{
	int status = -EBUSY;
	struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);

	mlog(0, "asked to destroy %.*s\n", lockres->l_namelen, lockres->l_name);

	spin_lock(&lockres->l_lock);
	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
		spin_unlock(&lockres->l_lock);
		return 0;
	}

	lockres->l_flags |= USER_LOCK_IN_TEARDOWN;

	while (lockres->l_flags & USER_LOCK_BUSY) {
		spin_unlock(&lockres->l_lock);

		user_wait_on_busy_lock(lockres);

		spin_lock(&lockres->l_lock);
	}

	if (lockres->l_ro_holders || lockres->l_ex_holders) {
		spin_unlock(&lockres->l_lock);
		goto bail;
	}

	status = 0;
	if (!(lockres->l_flags & USER_LOCK_ATTACHED)) {
		spin_unlock(&lockres->l_lock);
		goto bail;
	}

	lockres->l_flags &= ~USER_LOCK_ATTACHED;
	lockres->l_flags |= USER_LOCK_BUSY;
	spin_unlock(&lockres->l_lock);

	status = dlmunlock(dlm,
			   &lockres->l_lksb,
			   LKM_VALBLK,
			   user_unlock_ast,
			   lockres);
	if (status != DLM_NORMAL) {
		user_log_dlm_error("dlmunlock", status, lockres);
		status = -EINVAL;
		goto bail;
	}

	user_wait_on_busy_lock(lockres);

	status = 0;
bail:
	return status;
}

struct dlm_ctxt *user_dlm_register_context(struct qstr *name,
					   struct dlm_protocol_version *proto)
{
	struct dlm_ctxt *dlm;
	u32 dlm_key;
	char *domain;

	domain = kmalloc(name->len + 1, GFP_NOFS);
	if (!domain) {
		mlog_errno(-ENOMEM);
		return ERR_PTR(-ENOMEM);
	}

	dlm_key = crc32_le(0, name->name, name->len);

	snprintf(domain, name->len + 1, "%.*s", name->len, name->name);

	dlm = dlm_register_domain(domain, dlm_key, proto);
	if (IS_ERR(dlm))
		mlog_errno(PTR_ERR(dlm));

	kfree(domain);
	return dlm;
}

void user_dlm_unregister_context(struct dlm_ctxt *dlm)
{
	dlm_unregister_domain(dlm);
}