alloc.c

ocfs1.2.7 源码

	tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);

	status = ocfs2_journal_dirty(handle, tl_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

bail:
	mlog_exit(status);
	return status;
}

static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
					 struct ocfs2_journal_handle *handle,
					 struct inode *data_alloc_inode,
					 struct buffer_head *data_alloc_bh)
{
	int status = 0;
	int i;
	unsigned int num_clusters;
	u64 start_blk;
	struct ocfs2_truncate_rec rec;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;
	struct inode *tl_inode = osb->osb_tl_inode;
	struct buffer_head *tl_bh = osb->osb_tl_bh;

	mlog_entry_void();

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	i = le16_to_cpu(tl->tl_used) - 1;
	while (i >= 0) {
		/* Caller has given us at least enough credits to
		 * update the truncate log dinode */
		status = ocfs2_journal_access(handle, tl_inode, tl_bh,
					      OCFS2_JOURNAL_ACCESS_WRITE);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		tl->tl_used = cpu_to_le16(i);

		status = ocfs2_journal_dirty(handle, tl_bh);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		/* TODO: Perhaps we can calculate the bulk of the
		 * credits up front rather than extending like
		 * this. */
		status = ocfs2_extend_trans(handle,
					    OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		rec = tl->tl_recs[i];
		start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
						    le32_to_cpu(rec.t_start));
		num_clusters = le32_to_cpu(rec.t_clusters);

		/* if start_blk is not set, we ignore the record as
		 * invalid. */
		if (start_blk) {
			mlog(0, "free record %d, start = %u, clusters = %u\n",
			     i, le32_to_cpu(rec.t_start), num_clusters);

			status = ocfs2_free_clusters(handle, data_alloc_inode,
						     data_alloc_bh, start_blk,
						     num_clusters);
			if (status < 0) {
				mlog_errno(status);
				goto bail;
			}
		}
		i--;
	}

bail:
	mlog_exit(status);
	return status;
}

/* Expects you to already be holding tl_inode->i_mutex */
static int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
{
	int status;
	unsigned int num_to_flush;
	struct ocfs2_journal_handle *handle = NULL;
	struct inode *tl_inode = osb->osb_tl_inode;
	struct inode *data_alloc_inode = NULL;
	struct buffer_head *tl_bh = osb->osb_tl_bh;
	struct buffer_head *data_alloc_bh = NULL;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;

	mlog_entry_void();

	BUG_ON(mutex_trylock(&tl_inode->i_mutex));

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	if (!OCFS2_IS_VALID_DINODE(di)) {
		OCFS2_RO_ON_INVALID_DINODE(osb->sb, di);
		status = -EIO;
		goto bail;
	}

	num_to_flush = le16_to_cpu(tl->tl_used);
	mlog(0, "Flush %u records from truncate log #%"MLFu64"\n",
	     num_to_flush, OCFS2_I(tl_inode)->ip_blkno);
	if (!num_to_flush) {
		status = 0;
		goto bail;
	}

	handle = ocfs2_alloc_handle(osb);
	if (!handle) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}

	data_alloc_inode = ocfs2_get_system_file_inode(osb,
						       GLOBAL_BITMAP_SYSTEM_INODE,
						       OCFS2_INVALID_SLOT);
	if (!data_alloc_inode) {
		status = -EINVAL;
		mlog(ML_ERROR, "Could not get bitmap inode!\n");
		goto bail;
	}

	ocfs2_handle_add_inode(handle, data_alloc_inode);
	status = ocfs2_meta_lock(data_alloc_inode, handle, &data_alloc_bh, 1);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	handle = ocfs2_start_trans(osb, handle, OCFS2_TRUNCATE_LOG_UPDATE);
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		handle = NULL;
		mlog_errno(status);
		goto bail;
	}

	status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
					       data_alloc_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

bail:
	if (handle)
		ocfs2_commit_trans(handle);

	if (data_alloc_inode)
		iput(data_alloc_inode);

	if (data_alloc_bh)
		brelse(data_alloc_bh);

	mlog_exit(status);
	return status;
}

int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
{
	int status;
	struct inode *tl_inode = osb->osb_tl_inode;

	mutex_lock(&tl_inode->i_mutex);
	status = __ocfs2_flush_truncate_log(osb);
	mutex_unlock(&tl_inode->i_mutex);

	return status;
}

static void ocfs2_truncate_log_worker(void *data)
{
	int status;
	struct ocfs2_super *osb = data;

	mlog_entry_void();

	status = ocfs2_flush_truncate_log(osb);
	if (status < 0)
		mlog_errno(status);

	mlog_exit(status);
}

#define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
				       int cancel)
{
	if (osb->osb_tl_inode) {
		/* We want to push off log flushes while truncates are
		 * still running. */
		if (cancel)
			cancel_delayed_work(&osb->osb_truncate_log_wq);

		queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq,
				   OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
	}
}

static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
				       int slot_num,
				       struct inode **tl_inode,
				       struct buffer_head **tl_bh)
{
	int status;
	struct inode *inode = NULL;
	struct buffer_head *bh = NULL;

	inode = ocfs2_get_system_file_inode(osb,
					   TRUNCATE_LOG_SYSTEM_INODE,
					   slot_num);
	if (!inode) {
		status = -EINVAL;
		mlog(ML_ERROR, "Could not get load truncate log inode!\n");
		goto bail;
	}

	status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
				  OCFS2_BH_CACHED, inode);
	if (status < 0) {
		iput(inode);
		mlog_errno(status);
		goto bail;
	}

	*tl_inode = inode;
	*tl_bh    = bh;
bail:
	mlog_exit(status);
	return status;
}

/* called during the 1st stage of node recovery. we stamp a clean
 * truncate log and pass back a copy for processing later. if the
 * truncate log does not require processing, a *tl_copy is set to
 * NULL. */
int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
				      int slot_num,
				      struct ocfs2_dinode **tl_copy)
{
	int status;
	struct inode *tl_inode = NULL;
	struct buffer_head *tl_bh = NULL;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;

	*tl_copy = NULL;

	mlog(0, "recover truncate log from slot %d\n", slot_num);

	status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	if (!OCFS2_IS_VALID_DINODE(di)) {
		OCFS2_RO_ON_INVALID_DINODE(tl_inode->i_sb, di);
		status = -EIO;
		goto bail;
	}

	if (le16_to_cpu(tl->tl_used)) {
		mlog(0, "We'll have %u logs to recover\n",
		     le16_to_cpu(tl->tl_used));

		*tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
		if (!(*tl_copy)) {
			status = -ENOMEM;
			mlog_errno(status);
			goto bail;
		}

		/* Assuming the write-out below goes well, this copy
		 * will be passed back to recovery for processing. */
		memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);

		/* All we need to do to clear the truncate log is set
		 * tl_used. */
		tl->tl_used = 0;

		status = ocfs2_write_block(osb, tl_bh, tl_inode);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

bail:
	if (tl_inode)
		iput(tl_inode);
	if (tl_bh)
		brelse(tl_bh);

	if (status < 0 && (*tl_copy)) {
		kfree(*tl_copy);
		*tl_copy = NULL;
	}

	mlog_exit(status);
	return status;
}

int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
					 struct ocfs2_dinode *tl_copy)
{
	int status = 0;
	int i;
	unsigned int clusters, num_recs, start_cluster;
	u64 start_blk;
	struct ocfs2_journal_handle *handle;
	struct inode *tl_inode = osb->osb_tl_inode;
	struct ocfs2_truncate_log *tl;

	mlog_entry_void();

	if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
		mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
		return -EINVAL;
	}

	tl = &tl_copy->id2.i_dealloc;
	num_recs = le16_to_cpu(tl->tl_used);
	mlog(0, "cleanup %u records from %"MLFu64"\n", num_recs,
	     tl_copy->i_blkno);

	mutex_lock(&tl_inode->i_mutex);
	for(i = 0; i < num_recs; i++) {
		if (ocfs2_truncate_log_needs_flush(osb)) {
			status = __ocfs2_flush_truncate_log(osb);
			if (status < 0) {
				mlog_errno(status);
				goto bail_up;
			}
		}

		handle = ocfs2_start_trans(osb, NULL,
					   OCFS2_TRUNCATE_LOG_UPDATE);
		if (IS_ERR(handle)) {
			status = PTR_ERR(handle);
			mlog_errno(status);
			goto bail_up;
		}

		clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
		start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
		start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);

		status = ocfs2_truncate_log_append(osb, handle,
						   start_blk, clusters);
		ocfs2_commit_trans(handle);
		if (status < 0) {
			mlog_errno(status);
			goto bail_up;
		}
	}

bail_up:
	mutex_unlock(&tl_inode->i_mutex);

	mlog_exit(status);
	return status;
}

void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
{
	int status;
	struct inode *tl_inode = osb->osb_tl_inode;

	mlog_entry_void();

	if (tl_inode) {
		cancel_delayed_work(&osb->osb_truncate_log_wq);
		flush_workqueue(ocfs2_wq);

		status = ocfs2_flush_truncate_log(osb);
		if (status < 0)
			mlog_errno(status);

		brelse(osb->osb_tl_bh);
		iput(osb->osb_tl_inode);
	}

	mlog_exit_void();
}

int ocfs2_truncate_log_init(struct ocfs2_super *osb)
{
	int status;
	struct inode *tl_inode = NULL;
	struct buffer_head *tl_bh = NULL;

	mlog_entry_void();

	status = ocfs2_get_truncate_log_info(osb,
					     osb->slot_num,
					     &tl_inode,
					     &tl_bh);
	if (status < 0)
		mlog_errno(status);

	/* ocfs2_truncate_log_shutdown keys on the existence of
	 * osb->osb_tl_inode so we don't set any of the osb variables
	 * until we're sure all is well. */
	INIT_WORK(&osb->osb_truncate_log_wq, ocfs2_truncate_log_worker, osb);
	osb->osb_tl_bh    = tl_bh;
	osb->osb_tl_inode = tl_inode;

	mlog_exit(status);
	return status;
}

/* This function will figure out whether the currently last extent
 * block will be deleted, and if it will, what the new last extent
 * block will be so we can update his h_next_leaf_blk field, as well
 * as the dinodes i_last_eb_blk */
static int ocfs2_find_new_last_ext_blk(struct ocfs2_super *osb,
				       struct inode *inode,
				       struct ocfs2_dinode *fe,
				       u32 new_i_clusters,
				       struct buffer_head *old_last_eb,
				       struct buffer_head **new_last_eb)
{
	int i, status = 0;
	u64 block = 0;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;
	struct buffer_head *bh = NULL;

	*new_last_eb = NULL;

	if (!OCFS2_IS_VALID_DINODE(fe)) {
		OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
		status = -EIO;
		goto bail;
	}

	/* we have no tree, so of course, no last_eb. */
	if (!fe->id2.i_list.l_tree_depth)
		goto bail;

	/* trunc to zero special case - this makes tree_depth = 0
	 * regardless of what it is.  */
	if (!new_i_clusters)
		goto bail;

	eb = (struct ocfs2_extent_block *) old_last_eb->b_data;
	el = &(eb->h_list);
	BUG_ON(!el->l_next_free_rec);

	/* Make sure that this guy will actually be empty after we
	 * clear away the data. */
	if (le32_to_cpu(el->l_recs[0].e_cpos) < new_i_clusters)
		goto bail;

	/* Ok, at this point, we know that last_eb will definitely
	 * change, so lets traverse the tree and find the second to
	 * last extent block. */
	el = &(fe->id2.i_list);
	/* go down the tree, */
	do {
		for(i = (le16_to_cpu(el->l_next_free_rec) - 1); i >= 0; i--) {
			if (le32_to_cpu(el->l_recs[i].e_cpos) <
			    new_i_clusters) {
				block = le64_to_cpu(el->l_recs[i].e_blkno);
				break;
			}
		}
		BUG_ON(i < 0);

		if (bh) {
			brelse(bh);
			bh = NULL;
		}

		status = ocfs2_read_block(osb, block, &bh, OCFS2_BH_CACHED,
					 inode);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
		eb = (struct ocfs2_extent_block *) bh->b_data;
		el = &eb->h_list;
		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
			status = -EIO;
			goto bail;
		}
	} while (el->l_tree_depth);

	*new_last_eb = bh;
	get_bh(*new_last_eb);
	mlog(0, "returning block %"MLFu64"\n", le64_to_cpu(eb->h_blkno));