inode.c

ocfs1.4.1 oracle分布式文件系统

	if (can_lock) {
		status = ocfs2_open_lock(inode);
		if (status) {
			make_bad_inode(inode);
			mlog_errno(status);
			return status;
		}
		status = ocfs2_inode_lock(inode, NULL, 0);
		if (status) {
			make_bad_inode(inode);
			mlog_errno(status);
			return status;
		}
	}

	if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
		status = ocfs2_try_open_lock(inode, 0);
		if (status) {
			make_bad_inode(inode);	
			return status;
		}
	}

	status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
				  can_lock ? inode : NULL);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	status = -EINVAL;
	fe = (struct ocfs2_dinode *) bh->b_data;
	if (!OCFS2_IS_VALID_DINODE(fe)) {
		mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
		     (unsigned long long)args->fi_blkno, 7,
		     fe->i_signature);
		goto bail;
	}

	/*
	 * This is a code bug. Right now the caller needs to
	 * understand whether it is asking for a system file inode or
	 * not so the proper lock names can be built.
	 */
	mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
			!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
			"Inode %llu: system file state is ambigous\n",
			(unsigned long long)args->fi_blkno);

	if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
	    S_ISBLK(le16_to_cpu(fe->i_mode)))
    		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));

	if (ocfs2_populate_inode(inode, fe, 0) < 0)
		goto bail;

	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));

	status = 0;

bail:
	if (can_lock)
		ocfs2_inode_unlock(inode, 0);

	if (status < 0)
		make_bad_inode(inode);

	if (args && bh)
		brelse(bh);

	mlog_exit(status);
	return status;
}

void ocfs2_sync_blockdev(struct super_block *sb)
{
	sync_blockdev(sb->s_bdev);
}

static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
				     struct inode *inode,
				     struct buffer_head *fe_bh)
{
	int status = 0;
	struct ocfs2_truncate_context *tc = NULL;
	struct ocfs2_dinode *fe;
	handle_t *handle = NULL;

	mlog_entry_void();

	fe = (struct ocfs2_dinode *) fe_bh->b_data;

	/*
	 * This check will also skip truncate of inodes with inline
	 * data and fast symlinks.
	 */
	if (fe->i_clusters) {
		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
		if (IS_ERR(handle)) {
			status = PTR_ERR(handle);
			mlog_errno(status);
			goto out;
		}

		status = ocfs2_journal_access(handle, inode, fe_bh,
					      OCFS2_JOURNAL_ACCESS_WRITE);
		if (status < 0) {
			mlog_errno(status);
			goto out;
		}

		i_size_write(inode, 0);

		status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
		if (status < 0) {
			mlog_errno(status);
			goto out;
		}

		ocfs2_commit_trans(osb, handle);
		handle = NULL;

		status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
		if (status < 0) {
			mlog_errno(status);
			goto out;
		}

		status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
		if (status < 0) {
			mlog_errno(status);
			goto out;
		}
	}

out:
	if (handle)
		ocfs2_commit_trans(osb, handle);
	mlog_exit(status);
	return status;
}

static int ocfs2_remove_inode(struct inode *inode,
			      struct buffer_head *di_bh,
			      struct inode *orphan_dir_inode,
			      struct buffer_head *orphan_dir_bh)
{
	int status;
	struct inode *inode_alloc_inode = NULL;
	struct buffer_head *inode_alloc_bh = NULL;
	handle_t *handle;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;

	inode_alloc_inode =
		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
					    le16_to_cpu(di->i_suballoc_slot));
	if (!inode_alloc_inode) {
		status = -EEXIST;
		mlog_errno(status);
		goto bail;
	}

	mutex_lock(&inode_alloc_inode->i_mutex);
	status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
	if (status < 0) {
		mutex_unlock(&inode_alloc_inode->i_mutex);

		mlog_errno(status);
		goto bail;
	}

	handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		mlog_errno(status);
		goto bail_unlock;
	}

	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
				  orphan_dir_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail_commit;
	}

	/* set the inodes dtime */
	status = ocfs2_journal_access(handle, inode, di_bh,
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto bail_commit;
	}

	di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
	di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));

	status = ocfs2_journal_dirty(handle, di_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail_commit;
	}

	ocfs2_remove_from_cache(inode, di_bh);

	status = ocfs2_free_dinode(handle, inode_alloc_inode,
				   inode_alloc_bh, di);
	if (status < 0)
		mlog_errno(status);

bail_commit:
	ocfs2_commit_trans(osb, handle);
bail_unlock:
	ocfs2_inode_unlock(inode_alloc_inode, 1);
	mutex_unlock(&inode_alloc_inode->i_mutex);
	brelse(inode_alloc_bh);
bail:
	iput(inode_alloc_inode);

	return status;
}

/* 
 * Serialize with orphan dir recovery. If the process doing
 * recovery on this orphan dir does an iget() with the dir
 * i_mutex held, we'll deadlock here. Instead we detect this
 * and exit early - recovery will wipe this inode for us.
 */
static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
					     int slot)
{
	int ret = 0;

	spin_lock(&osb->osb_lock);
	if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
		mlog(0, "Recovery is happening on orphan dir %d, will skip "
		     "this inode\n", slot);
		ret = -EDEADLK;
		goto out;
	}
	/* This signals to the orphan recovery process that it should
	 * wait for us to handle the wipe. */
	osb->osb_orphan_wipes[slot]++;
out:
	spin_unlock(&osb->osb_lock);
	return ret;
}

static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
					 int slot)
{
	spin_lock(&osb->osb_lock);
	osb->osb_orphan_wipes[slot]--;
	spin_unlock(&osb->osb_lock);

	wake_up(&osb->osb_wipe_event);
}

static int ocfs2_wipe_inode(struct inode *inode,
			    struct buffer_head *di_bh)
{
	int status, orphaned_slot;
	struct inode *orphan_dir_inode = NULL;
	struct buffer_head *orphan_dir_bh = NULL;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
	struct ocfs2_dinode *di;

	di = (struct ocfs2_dinode *) di_bh->b_data;
	orphaned_slot = le16_to_cpu(di->i_orphaned_slot);

	status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
	if (status)
		return status;

	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
						       ORPHAN_DIR_SYSTEM_INODE,
						       orphaned_slot);
	if (!orphan_dir_inode) {
		status = -EEXIST;
		mlog_errno(status);
		goto bail;
	}

	/* Lock the orphan dir. The lock will be held for the entire
	 * delete_inode operation. We do this now to avoid races with
	 * recovery completion on other nodes. */
	mutex_lock(&orphan_dir_inode->i_mutex);
	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
	if (status < 0) {
		mutex_unlock(&orphan_dir_inode->i_mutex);

		mlog_errno(status);
		goto bail;
	}

	/* we do this while holding the orphan dir lock because we
	 * don't want recovery being run from another node to try an
	 * inode delete underneath us -- this will result in two nodes
	 * truncating the same file! */
	status = ocfs2_truncate_for_delete(osb, inode, di_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail_unlock_dir;
	}

	status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
				    orphan_dir_bh);
	if (status < 0)
		mlog_errno(status);

bail_unlock_dir:
	ocfs2_inode_unlock(orphan_dir_inode, 1);
	mutex_unlock(&orphan_dir_inode->i_mutex);
	brelse(orphan_dir_bh);
bail:
	iput(orphan_dir_inode);
	ocfs2_signal_wipe_completion(osb, orphaned_slot);

	return status;
}

/* There is a series of simple checks that should be done before a
 * trylock is even considered. Encapsulate those in this function. */
static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
{
	int ret = 0;
	struct ocfs2_inode_info *oi = OCFS2_I(inode);
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	/* We shouldn't be getting here for the root directory
	 * inode.. */
	if (inode == osb->root_inode) {
		mlog(ML_ERROR, "Skipping delete of root inode.\n");
		goto bail;
	}

	/* If we're coming from downconvert_thread we can't go into our own
	 * voting [hello, deadlock city!], so unforuntately we just
	 * have to skip deleting this guy. That's OK though because
	 * the node who's doing the actual deleting should handle it
	 * anyway. */
	if (current == osb->dc_task) {
		mlog(0, "Skipping delete of %lu because we're currently "
		     "in downconvert\n", inode->i_ino);
		goto bail;
	}

	spin_lock(&oi->ip_lock);
	/* OCFS2 *never* deletes system files. This should technically
	 * never get here as system file inodes should always have a
	 * positive link count. */
	if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
		mlog(ML_ERROR, "Skipping delete of system file %llu\n",
		     (unsigned long long)oi->ip_blkno);
		goto bail_unlock;
	}

	/* If we have allowd wipe of this inode for another node, it
	 * will be marked here so we can safely skip it. Recovery will
	 * cleanup any inodes we might inadvertantly skip here. */
	if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
		mlog(0, "Skipping delete of %lu because another node "
		     "has done this for us.\n", inode->i_ino);
		goto bail_unlock;
	}

	ret = 1;
bail_unlock:
	spin_unlock(&oi->ip_lock);
bail:
	return ret;
}

/* Query the cluster to determine whether we should wipe an inode from
 * disk or not.
 *
 * Requires the inode to have the cluster lock. */
static int ocfs2_query_inode_wipe(struct inode *inode,
				  struct buffer_head *di_bh,
				  int *wipe)
{
	int status = 0;
	struct ocfs2_inode_info *oi = OCFS2_I(inode);
	struct ocfs2_dinode *di;

	*wipe = 0;

	/* While we were waiting for the cluster lock in
	 * ocfs2_delete_inode, another node might have asked to delete
	 * the inode. Recheck our flags to catch this. */
	if (!ocfs2_inode_is_valid_to_delete(inode)) {
		mlog(0, "Skipping delete of %llu because flags changed\n",
		     (unsigned long long)oi->ip_blkno);
		goto bail;
	}

	/* Now that we have an up to date inode, we can double check
	 * the link count. */
	if (inode->i_nlink) {
		mlog(0, "Skipping delete of %llu because nlink = %u\n",
		     (unsigned long long)oi->ip_blkno, inode->i_nlink);
		goto bail;
	}

	/* Do some basic inode verification... */
	di = (struct ocfs2_dinode *) di_bh->b_data;
	if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
		/* for lack of a better error? */
		status = -EEXIST;
		mlog(ML_ERROR,
		     "Inode %llu (on-disk %llu) not orphaned! "
		     "Disk flags  0x%x, inode flags 0x%x\n",
		     (unsigned long long)oi->ip_blkno,
		     (unsigned long long)le64_to_cpu(di->i_blkno),
		     le32_to_cpu(di->i_flags), oi->ip_flags);
		goto bail;
	}

	/* has someone already deleted us?! baaad... */
	if (di->i_dtime) {
		status = -EEXIST;
		mlog_errno(status);
		goto bail;
	}

	/*
	 * This is how ocfs2 determines whether an inode is still live
	 * within the cluster. Every node takes a shared read lock on
	 * the inode open lock in ocfs2_read_locked_inode(). When we