dlmglue.c

ocfs1.2.7 源码

	}

	status = ocfs2_read_block(OCFS2_SB(inode->i_sb),
				  OCFS2_I(inode)->ip_blkno,
				  ret_bh,
				  OCFS2_BH_CACHED,
				  inode);
	if (status < 0)
		mlog_errno(status);

	return status;
}

/*
 * returns < 0 error if the callback will never be called, otherwise
 * the result of the lock will be communicated via the callback.
 */
int ocfs2_meta_lock_full(struct inode *inode,
			 struct ocfs2_journal_handle *handle,
			 struct buffer_head **ret_bh,
			 int ex,
			 int flags,
			 ocfs2_lock_callback cb,
			 unsigned long cb_data)
{
	int status, level, dlm_flags, acquired;
	struct ocfs2_lock_res *lockres = NULL;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
	struct buffer_head *local_bh = NULL;

	BUG_ON(!inode);

	mlog_entry_void();

	mlog(0, "inode %"MLFu64", take %s META lock\n",
	     OCFS2_I(inode)->ip_blkno,
	     ex ? "EXMODE" : "PRMODE");

	status = 0;
	acquired = 0;
	/* We'll allow faking a readonly metadata lock for
	 * rodevices. */
	if (ocfs2_is_hard_readonly(osb)) {
		if (ex)
			status = -EROFS;
		goto bail;
	}

	if (ocfs2_mount_local(osb))
		goto local;

	if (!(flags & OCFS2_META_LOCK_RECOVERY))
		wait_event(osb->recovery_event,
			   ocfs2_node_map_is_empty(osb, &osb->recovery_map));

	lockres = &OCFS2_I(inode)->ip_meta_lockres;
	level = ex ? LKM_EXMODE : LKM_PRMODE;
	dlm_flags = 0;
	if (flags & OCFS2_META_LOCK_NOQUEUE)
		dlm_flags |= LKM_NOQUEUE;

	status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, cb,
				    cb_data);
	if (status < 0) {
		if (status != -EAGAIN && status != -EIOCBRETRY)
			mlog_errno(status);
		goto bail;
	}

	/* Notify the error cleanup path to drop the cluster lock. */
	acquired = 1;

	/* We wait twice because a node may have died while we were in
	 * the lower dlm layers. The second time though, we've
	 * committed to owning this lock so we don't allow signals to
	 * abort the operation. */
	if (!(flags & OCFS2_META_LOCK_RECOVERY))
		wait_event(osb->recovery_event,
			   ocfs2_node_map_is_empty(osb, &osb->recovery_map));

local:
	/* This is fun. The caller may want a bh back, or it may
	 * not. ocfs2_meta_lock_update definitely wants one in, but
	 * may or may not read one, depending on what's in the
	 * LVB. The result of all of this is that we've *only* gone to
	 * disk if we have to, so the complexity is worthwhile. */
	status = ocfs2_meta_lock_update(inode, &local_bh);
	if (status < 0) {
		if (status != -ENOENT)
			mlog_errno(status);
		goto bail;
	}

	if (ret_bh) {
		status = ocfs2_assign_bh(inode, ret_bh, local_bh);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

	if (handle) {
		status = ocfs2_handle_add_lock(handle, inode);
		if (status < 0)
			mlog_errno(status);
	}

bail:
	if (status < 0) {
		if (ret_bh && (*ret_bh)) {
			brelse(*ret_bh);
			*ret_bh = NULL;
		}
		if (acquired)
			ocfs2_meta_unlock(inode, ex);
	}

	if (local_bh)
		brelse(local_bh);

	mlog_exit(status);
	return status;
}

void ocfs2_meta_unlock(struct inode *inode,
		       int ex)
{
	int level = ex ? LKM_EXMODE : LKM_PRMODE;
	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_meta_lockres;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	mlog_entry_void();

	mlog(0, "inode %"MLFu64" drop %s META lock\n",
	     OCFS2_I(inode)->ip_blkno,
	     ex ? "EXMODE" : "PRMODE");

	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);

	mlog_exit_void();
}

int ocfs2_super_lock(struct ocfs2_super *osb,
		     int ex)
{
	int status;
	int level = ex ? LKM_EXMODE : LKM_PRMODE;
	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
	struct buffer_head *bh;
	struct ocfs2_slot_info *si = osb->slot_info;

	mlog_entry_void();

	if (ocfs2_is_hard_readonly(osb))
		return -EROFS;

	if (ocfs2_mount_local(osb))
		return 0;

	status = ocfs2_cluster_lock(osb, lockres, level, 0, NULL, 0);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	/* The super block lock path is really in the best position to
	 * know when resources covered by the lock need to be
	 * refreshed, so we do it here. Of course, making sense of
	 * everything is up to the caller :) */
	status = ocfs2_should_refresh_lock_res(lockres);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
	if (status) {
		bh = si->si_bh;
		status = ocfs2_read_block(osb, bh->b_blocknr, &bh, 0,
					  si->si_inode);
		if (status == 0)
			ocfs2_update_slot_info(si);

		ocfs2_complete_lock_res_refresh(lockres, status);

		if (status < 0)
			mlog_errno(status);
	}
bail:
	mlog_exit(status);
	return status;
}

void ocfs2_super_unlock(struct ocfs2_super *osb,
			int ex)
{
	int level = ex ? LKM_EXMODE : LKM_PRMODE;
	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;

	if (!ocfs2_mount_local(osb))
		ocfs2_cluster_unlock(osb, lockres, level);
}

int ocfs2_rename_lock(struct ocfs2_super *osb)
{
	int status;
	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;

	if (ocfs2_is_hard_readonly(osb))
		return -EROFS;

	if (ocfs2_mount_local(osb))
		return 0;

	status = ocfs2_cluster_lock(osb, lockres, LKM_EXMODE, 0, NULL, 0);
	if (status < 0)
		mlog_errno(status);

	return status;
}

void ocfs2_rename_unlock(struct ocfs2_super *osb)
{
	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;

	if (!ocfs2_mount_local(osb))
		ocfs2_cluster_unlock(osb, lockres, LKM_EXMODE);
}

/* Reference counting of the dlm debug structure. We want this because
 * open references on the debug inodes can live on after a mount, so
 * we can't rely on the ocfs2_super to always exist. */
static void ocfs2_dlm_debug_free(struct kref *kref)
{
	struct ocfs2_dlm_debug *dlm_debug;

	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);

	kfree(dlm_debug);
}

void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
{
	if (dlm_debug)
		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
}

static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
{
	kref_get(&debug->d_refcnt);
}

struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
{
	struct ocfs2_dlm_debug *dlm_debug;

	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
	if (!dlm_debug) {
		mlog_errno(-ENOMEM);
		goto out;
	}

	kref_init(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
	dlm_debug->d_locking_state = NULL;
out:
	return dlm_debug;
}

/* Access to this is arbitrated for us via seq_file->sem. */
struct ocfs2_dlm_seq_priv {
	struct ocfs2_dlm_debug *p_dlm_debug;
	struct ocfs2_lock_res p_iter_res;
	struct ocfs2_lock_res p_tmp_res;
};

static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
						 struct ocfs2_dlm_seq_priv *priv)
{
	struct ocfs2_lock_res *iter, *ret = NULL;
	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;

	assert_spin_locked(&ocfs2_dlm_tracking_lock);

	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
		/* discover the head of the list */
		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
			mlog(0, "End of list found, %p\n", ret);
			break;
		}

		/* We track our "dummy" iteration lockres' by a NULL
		 * l_ops field. */
		if (iter->l_ops != NULL) {
			ret = iter;
			break;
		}
	}

	return ret;
}

static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
{
	struct ocfs2_dlm_seq_priv *priv = m->private;
	struct ocfs2_lock_res *iter;

	spin_lock(&ocfs2_dlm_tracking_lock);
	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
	if (iter) {
		/* Since lockres' have the lifetime of their container
		 * (which can be inodes, ocfs2_supers, etc) we want to
		 * copy this out to a temporary lockres while still
		 * under the spinlock. Obviously after this we can't
		 * trust any pointers on the copy returned, but that's
		 * ok as the information we want isn't typically held
		 * in them. */
		priv->p_tmp_res = *iter;
		iter = &priv->p_tmp_res;
	}
	spin_unlock(&ocfs2_dlm_tracking_lock);

	return iter;
}

static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
{
}

static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
{
	struct ocfs2_dlm_seq_priv *priv = m->private;
	struct ocfs2_lock_res *iter = v;
	struct ocfs2_lock_res *dummy = &priv->p_iter_res;

	spin_lock(&ocfs2_dlm_tracking_lock);
	iter = ocfs2_dlm_next_res(iter, priv);
	list_del_init(&dummy->l_debug_list);
	if (iter) {
		list_add(&dummy->l_debug_list, &iter->l_debug_list);
		priv->p_tmp_res = *iter;
		iter = &priv->p_tmp_res;
	}
	spin_unlock(&ocfs2_dlm_tracking_lock);

	return iter;
}

/* So that debugfs.ocfs2 can determine which format is being used */
#define OCFS2_DLM_DEBUG_STR_VERSION 1
static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
{
	int i;
	char *lvb;
	struct ocfs2_lock_res *lockres = v;

	if (!lockres)
		return -EINVAL;

	seq_printf(m, "0x%x\t"
		   "%.*s\t"
		   "%d\t"
		   "0x%lx\t"
		   "0x%x\t"
		   "0x%x\t"
		   "%u\t"
		   "%u\t"
		   "%d\t"
		   "%d\t",
		   OCFS2_DLM_DEBUG_STR_VERSION,
		   OCFS2_LOCK_ID_MAX_LEN, lockres->l_name,
		   lockres->l_level,
		   lockres->l_flags,
		   lockres->l_action,
		   lockres->l_unlock_action,
		   lockres->l_ro_holders,
		   lockres->l_ex_holders,
		   lockres->l_requested,
		   lockres->l_blocking);

	/* Dump the raw LVB */
	lvb = lockres->l_lksb.lvb;
	for(i = 0; i < DLM_LVB_LEN; i++)
		seq_printf(m, "0x%x\t", lvb[i]);

	/* End the line */
	seq_printf(m, "\n");
	return 0;
}

static struct seq_operations ocfs2_dlm_seq_ops = {
	.start =	ocfs2_dlm_seq_start,
	.stop =		ocfs2_dlm_seq_stop,
	.next =		ocfs2_dlm_seq_next,
	.show =		ocfs2_dlm_seq_show,
};

static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
{
	struct seq_file *seq = (struct seq_file *) file->private_data;
	struct ocfs2_dlm_seq_priv *priv = seq->private;
	struct ocfs2_lock_res *res = &priv->p_iter_res;

	ocfs2_remove_lockres_tracking(res);
	ocfs2_put_dlm_debug(priv->p_dlm_debug);
	return seq_release_private(inode, file);
}

static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
{
	int ret;
	struct ocfs2_dlm_seq_priv *priv;
	struct seq_file *seq;
	struct ocfs2_super *osb;

	priv = kzalloc(sizeof(struct ocfs2_dlm_seq_priv), GFP_KERNEL);
	if (!priv) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}
#ifdef INODE_HAS_PRIVATE
	osb = inode->i_private;
#else
	osb = (struct ocfs2_super *) inode->u.generic_ip;
#endif
	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
	priv->p_dlm_debug = osb->osb_dlm_debug;
	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);

	ret = seq_open(file, &ocfs2_dlm_seq_ops);
	if (ret) {
		kfree(priv);
		mlog_errno(ret);
		goto out;
	}

	seq = (struct seq_file *) file->private_data;
	seq->private = priv;

	ocfs2_add_lockres_tracking(&priv->p_iter_res,
				   priv->p_dlm_debug);

out:
	return ret;
}

static struct file_operations ocfs2_dlm_debug_fops = {
	.open =		ocfs2_dlm_debug_open,
	.release =	ocfs2_dlm_debug_release,
	.read =		seq_read,
	.llseek =	seq_lseek,
};

static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
{
	int ret = 0;
	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;

	dlm_debug->d_locking_state = debugfs_create_file("locking_state",
							 S_IFREG|S_IRUSR,
							 osb->osb_debug_root,
							 osb,
							 &ocfs2_dlm_debug_fops);
	if (!dlm_debug->d_locking_state) {
		ret = -EINVAL;
		mlog(ML_ERROR,
		     "Unable to create locking state debugfs file.\n");
		goto out;
	}

	ocfs2_get_dlm_debug(dlm_debug);
out:
	return ret;
}

static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
{
	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;

	if (dlm_debug) {
		debugfs_remove(dlm_debug->d_locking_state);
		ocfs2_put_dlm_debug(dlm_debug);
	}
}

int ocfs2_dlm_init(struct ocfs2_super *osb)
{
	int status;
	u32 dlm_key;
	struct dlm_ctxt *dlm = NULL;

	mlog_entry_void();

	if (ocfs2_mount_local(osb))
		goto local;

	status = ocfs2_dlm_init_debug(osb);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}