alloc.c

ocfs1.2.7 源码

bail:
	if (bh)
		brelse(bh);

	return status;
}

static int ocfs2_do_truncate(struct ocfs2_super *osb,
			     unsigned int clusters_to_del,
			     struct inode *inode,
			     struct buffer_head *fe_bh,
			     struct buffer_head *old_last_eb_bh,
			     struct ocfs2_journal_handle *handle,
			     struct ocfs2_truncate_context *tc)
{
	int status, i, depth;
	struct ocfs2_dinode *fe;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_block *last_eb = NULL;
	struct ocfs2_extent_list *el;
	struct buffer_head *eb_bh = NULL;
	struct buffer_head *last_eb_bh = NULL;
	u64 next_eb = 0;
	u64 delete_blk = 0;

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

	status = ocfs2_find_new_last_ext_blk(osb,
					     inode,
					     fe,
					     le32_to_cpu(fe->i_clusters) -
					     		clusters_to_del,
					     old_last_eb_bh,
					     &last_eb_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
	if (last_eb_bh)
		last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;

	status = ocfs2_journal_access(handle, inode, fe_bh,
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
	el = &(fe->id2.i_list);

	spin_lock(&OCFS2_I(inode)->ip_lock);
	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) -
				      clusters_to_del;
	spin_unlock(&OCFS2_I(inode)->ip_lock);
	le32_add_cpu(&fe->i_clusters, -clusters_to_del);
	fe->i_mtime = cpu_to_le64(CURRENT_TIME.tv_sec);
	fe->i_mtime_nsec = cpu_to_le32(CURRENT_TIME.tv_nsec);

	i = le16_to_cpu(el->l_next_free_rec) - 1;

	BUG_ON(le32_to_cpu(el->l_recs[i].e_clusters) < clusters_to_del);
	le32_add_cpu(&el->l_recs[i].e_clusters, -clusters_to_del);
	/* tree depth zero, we can just delete the clusters, otherwise
	 * we need to record the offset of the next level extent block
	 * as we may overwrite it. */
	if (!el->l_tree_depth)
		delete_blk = le64_to_cpu(el->l_recs[i].e_blkno)
			+ ocfs2_clusters_to_blocks(osb->sb,
					le32_to_cpu(el->l_recs[i].e_clusters));
	else
		next_eb = le64_to_cpu(el->l_recs[i].e_blkno);

	if (!el->l_recs[i].e_clusters) {
		/* if we deleted the whole extent record, then clear
		 * out the other fields and update the extent
		 * list. For depth > 0 trees, we've already recorded
		 * the extent block in 'next_eb' */
		el->l_recs[i].e_cpos = 0;
		el->l_recs[i].e_blkno = 0;
		BUG_ON(!el->l_next_free_rec);
		le16_add_cpu(&el->l_next_free_rec, -1);
	}

	depth = le16_to_cpu(el->l_tree_depth);
	if (!fe->i_clusters) {
		/* trunc to zero is a special case. */
		el->l_tree_depth = 0;
		fe->i_last_eb_blk = 0;
	} else if (last_eb)
		fe->i_last_eb_blk = last_eb->h_blkno;

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

	if (last_eb) {
		/* If there will be a new last extent block, then by
		 * definition, there cannot be any leaves to the right of
		 * him. */
		status = ocfs2_journal_access(handle, inode, last_eb_bh,
					      OCFS2_JOURNAL_ACCESS_WRITE);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
		last_eb->h_next_leaf_blk = 0;
		status = ocfs2_journal_dirty(handle, last_eb_bh);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

	/* if our tree depth > 0, update all the tree blocks below us. */
	while (depth) {
		mlog(0, "traveling tree (depth = %d, next_eb = %"MLFu64")\n",
		     depth,  next_eb);
		status = ocfs2_read_block(osb, next_eb, &eb_bh,
					  OCFS2_BH_CACHED, inode);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
		eb = (struct ocfs2_extent_block *)eb_bh->b_data;
		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
			status = -EIO;
			goto bail;
		}
		el = &(eb->h_list);

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

		BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
		BUG_ON(depth != (le16_to_cpu(el->l_tree_depth) + 1));

		i = le16_to_cpu(el->l_next_free_rec) - 1;

		mlog(0, "extent block %"MLFu64", before: record %d: "
		     "(%u, %u, %"MLFu64"), next = %u\n",
		     le64_to_cpu(eb->h_blkno), i,
		     le32_to_cpu(el->l_recs[i].e_cpos),
		     le32_to_cpu(el->l_recs[i].e_clusters),
		     le64_to_cpu(el->l_recs[i].e_blkno),
		     le16_to_cpu(el->l_next_free_rec));

		BUG_ON(le32_to_cpu(el->l_recs[i].e_clusters) < clusters_to_del);
		le32_add_cpu(&el->l_recs[i].e_clusters, -clusters_to_del);

		next_eb = le64_to_cpu(el->l_recs[i].e_blkno);
		/* bottom-most block requires us to delete data.*/
		if (!el->l_tree_depth)
			delete_blk = le64_to_cpu(el->l_recs[i].e_blkno)
				+ ocfs2_clusters_to_blocks(osb->sb,
					le32_to_cpu(el->l_recs[i].e_clusters));
		if (!el->l_recs[i].e_clusters) {
			el->l_recs[i].e_cpos = 0;
			el->l_recs[i].e_blkno = 0;
			BUG_ON(!el->l_next_free_rec);
			le16_add_cpu(&el->l_next_free_rec, -1);
		}
		mlog(0, "extent block %"MLFu64", after: record %d: "
		     "(%u, %u, %"MLFu64"), next = %u\n",
		     le64_to_cpu(eb->h_blkno), i,
		     le32_to_cpu(el->l_recs[i].e_cpos),
		     le32_to_cpu(el->l_recs[i].e_clusters),
		     le64_to_cpu(el->l_recs[i].e_blkno),
		     le16_to_cpu(el->l_next_free_rec));

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

		if (!el->l_next_free_rec) {
			mlog(0, "deleting this extent block.\n");

			ocfs2_remove_from_cache(inode, eb_bh);

			BUG_ON(eb->h_suballoc_slot);
			BUG_ON(el->l_recs[0].e_clusters);
			BUG_ON(el->l_recs[0].e_cpos);
			BUG_ON(el->l_recs[0].e_blkno);
			status = ocfs2_free_extent_block(handle,
							 tc->tc_ext_alloc_inode,
							 tc->tc_ext_alloc_bh,
							 eb);
			if (status < 0) {
				mlog_errno(status);
				goto bail;
			}
		}
		brelse(eb_bh);
		eb_bh = NULL;
		depth--;
	}

	BUG_ON(!delete_blk);
	status = ocfs2_truncate_log_append(osb, handle, delete_blk,
					   clusters_to_del);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
	status = 0;
bail:
	if (!status)
		ocfs2_extent_map_trunc(inode, le32_to_cpu(fe->i_clusters));
	else
		ocfs2_extent_map_drop(inode, 0);
	mlog_exit(status);
	return status;
}

/*
 * It is expected, that by the time you call this function,
 * inode->i_size and fe->i_size have been adjusted.
 *
 * WARNING: This will kfree the truncate context
 */
int ocfs2_commit_truncate(struct ocfs2_super *osb,
			  struct inode *inode,
			  struct buffer_head *fe_bh,
			  struct ocfs2_truncate_context *tc)
{
	int status, i, credits, tl_sem = 0;
	u32 clusters_to_del, target_i_clusters;
	u64 last_eb = 0;
	struct ocfs2_dinode *fe;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;
	struct buffer_head *last_eb_bh;
	struct ocfs2_journal_handle *handle = NULL;
	struct inode *tl_inode = osb->osb_tl_inode;

	mlog_entry_void();

	down_write(&OCFS2_I(inode)->ip_alloc_sem);

	target_i_clusters = ocfs2_clusters_for_bytes(osb->sb,
						     i_size_read(inode));

	last_eb_bh = tc->tc_last_eb_bh;
	tc->tc_last_eb_bh = NULL;

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

	if (fe->id2.i_list.l_tree_depth) {
		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
		el = &eb->h_list;
	} else
		el = &fe->id2.i_list;
	last_eb = le64_to_cpu(fe->i_last_eb_blk);
start:
	mlog(0, "ocfs2_commit_truncate: fe->i_clusters = %u, "
	     "last_eb = %"MLFu64", fe->i_last_eb_blk = %"MLFu64", "
	     "fe->id2.i_list.l_tree_depth = %u last_eb_bh = %p\n",
	     le32_to_cpu(fe->i_clusters), last_eb,
	     le64_to_cpu(fe->i_last_eb_blk),
	     le16_to_cpu(fe->id2.i_list.l_tree_depth), last_eb_bh);

	if (last_eb != le64_to_cpu(fe->i_last_eb_blk)) {
		mlog(0, "last_eb changed!\n");
		BUG_ON(!fe->id2.i_list.l_tree_depth);
		last_eb = le64_to_cpu(fe->i_last_eb_blk);
		/* i_last_eb_blk may have changed, read it if
		 * necessary. We don't have to worry about the
		 * truncate to zero case here (where there becomes no
		 * last_eb) because we never loop back after our work
		 * is done. */
		if (last_eb_bh) {
			brelse(last_eb_bh);
			last_eb_bh = NULL;
		}

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

	/* by now, el will point to the extent list on the bottom most
	 * portion of this tree. */
	i = le16_to_cpu(el->l_next_free_rec) - 1;
	if (le32_to_cpu(el->l_recs[i].e_cpos) >= target_i_clusters)
		clusters_to_del = le32_to_cpu(el->l_recs[i].e_clusters);
	else
		clusters_to_del = (le32_to_cpu(el->l_recs[i].e_clusters) +
				   le32_to_cpu(el->l_recs[i].e_cpos)) -
				  target_i_clusters;

	mlog(0, "clusters_to_del = %u in this pass\n", clusters_to_del);

	mutex_lock(&tl_inode->i_mutex);
	tl_sem = 1;
	/* ocfs2_truncate_log_needs_flush guarantees us at least one
	 * record is free for use. If there isn't any, we flush to get
	 * an empty truncate log.  */
	if (ocfs2_truncate_log_needs_flush(osb)) {
		status = __ocfs2_flush_truncate_log(osb);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

	credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del,
						fe, el);
	handle = ocfs2_start_trans(osb, NULL, credits);
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		handle = NULL;
		mlog_errno(status);
		goto bail;
	}

	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
	if (status < 0)
		mlog_errno(status);

	status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh,
				   last_eb_bh, handle, tc);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	mutex_unlock(&tl_inode->i_mutex);
	tl_sem = 0;

	ocfs2_commit_trans(handle);
	handle = NULL;

	BUG_ON(le32_to_cpu(fe->i_clusters) < target_i_clusters);
	if (le32_to_cpu(fe->i_clusters) > target_i_clusters)
		goto start;
bail:
	up_write(&OCFS2_I(inode)->ip_alloc_sem);

	ocfs2_schedule_truncate_log_flush(osb, 1);

	if (tl_sem)
		mutex_unlock(&tl_inode->i_mutex);

	if (handle)
		ocfs2_commit_trans(handle);

	if (last_eb_bh)
		brelse(last_eb_bh);

	/* This will drop the ext_alloc cluster lock for us */
	ocfs2_free_truncate_context(tc);

	mlog_exit(status);
	return status;
}


/*
 * Expects the inode to already be locked. This will figure out which
 * inodes need to be locked and will put them on the returned truncate
 * context.
 */
int ocfs2_prepare_truncate(struct ocfs2_super *osb,
			   struct inode *inode,
			   struct buffer_head *fe_bh,
			   struct ocfs2_truncate_context **tc)
{
	int status, metadata_delete;
	unsigned int new_i_clusters;
	struct ocfs2_dinode *fe;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;
	struct buffer_head *last_eb_bh = NULL;
	struct inode *ext_alloc_inode = NULL;
	struct buffer_head *ext_alloc_bh = NULL;

	mlog_entry_void();

	*tc = NULL;

	new_i_clusters = ocfs2_clusters_for_bytes(osb->sb,
						  i_size_read(inode));
	fe = (struct ocfs2_dinode *) fe_bh->b_data;

	mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size ="
	     "%"MLFu64"\n", fe->i_clusters, new_i_clusters, fe->i_size);

	if (le32_to_cpu(fe->i_clusters) <= new_i_clusters) {
		ocfs2_error(inode->i_sb, "Dinode %"MLFu64" has cluster count "
			    "%u and size %"MLFu64" whereas struct inode has "
			    "cluster count %u and size %llu which caused an "
			    "invalid truncate to %u clusters.",
			    le64_to_cpu(fe->i_blkno),
			    le32_to_cpu(fe->i_clusters),
			    le64_to_cpu(fe->i_size),
			    OCFS2_I(inode)->ip_clusters, i_size_read(inode),
			    new_i_clusters);
		mlog_meta_lvb(ML_ERROR, &OCFS2_I(inode)->ip_meta_lockres);
		status = -EIO;
		goto bail;
	}

	*tc = kcalloc(1, sizeof(struct ocfs2_truncate_context), GFP_KERNEL);
	if (!(*tc)) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}

	metadata_delete = 0;
	if (fe->id2.i_list.l_tree_depth) {
		/* If we have a tree, then the truncate may result in
		 * metadata deletes. Figure this out from the
		 * rightmost leaf block.*/
		status = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk),
					  &last_eb_bh, OCFS2_BH_CACHED, inode);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);

			brelse(last_eb_bh);
			status = -EIO;
			goto bail;
		}
		el = &(eb->h_list);
		if (le32_to_cpu(el->l_recs[0].e_cpos) >= new_i_clusters)
			metadata_delete = 1;
	}

	(*tc)->tc_last_eb_bh = last_eb_bh;

	if (metadata_delete) {
		mlog(0, "Will have to delete metadata for this trunc. "
		     "locking allocator.\n");
		ext_alloc_inode = ocfs2_get_system_file_inode(osb, EXTENT_ALLOC_SYSTEM_INODE, 0);
		if (!ext_alloc_inode) {
			status = -ENOMEM;
			mlog_errno(status);
			goto bail;
		}

		mutex_lock(&ext_alloc_inode->i_mutex);
		(*tc)->tc_ext_alloc_inode = ext_alloc_inode;

		status = ocfs2_meta_lock(ext_alloc_inode,
					 NULL,
					 &ext_alloc_bh,
					 1);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
		(*tc)->tc_ext_alloc_bh = ext_alloc_bh;
		(*tc)->tc_ext_alloc_locked = 1;
	}

	status = 0;
bail:
	if (status < 0) {
		if (*tc)
			ocfs2_free_truncate_context(*tc);
		*tc = NULL;
	}
	mlog_exit_void();
	return status;
}

void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc)
{
	if (tc->tc_ext_alloc_inode) {
		if (tc->tc_ext_alloc_locked)
			ocfs2_meta_unlock(tc->tc_ext_alloc_inode, 1);

		mutex_unlock(&tc->tc_ext_alloc_inode->i_mutex);
		iput(tc->tc_ext_alloc_inode);
	}

	if (tc->tc_ext_alloc_bh)
		brelse(tc->tc_ext_alloc_bh);

	if (tc->tc_last_eb_bh)
		brelse(tc->tc_last_eb_bh);

	kfree(tc);
}