file.c

ocfs1.4.1 oracle分布式文件系统

	int free_extents;
	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
	enum ocfs2_alloc_restarted reason = RESTART_NONE;
	u32 bit_off, num_bits;
	u64 block;
	u8 flags = 0;

	BUG_ON(!clusters_to_add);

	if (mark_unwritten)
		flags = OCFS2_EXT_UNWRITTEN;

	free_extents = ocfs2_num_free_extents(osb, inode, fe);
	if (free_extents < 0) {
		status = free_extents;
		mlog_errno(status);
		goto leave;
	}

	/* there are two cases which could cause us to EAGAIN in the
	 * we-need-more-metadata case:
	 * 1) we haven't reserved *any*
	 * 2) we are so fragmented, we've needed to add metadata too
	 *    many times. */
	if (!free_extents && !meta_ac) {
		mlog(0, "we haven't reserved any metadata!\n");
		status = -EAGAIN;
		reason = RESTART_META;
		goto leave;
	} else if ((!free_extents)
		   && (ocfs2_alloc_context_bits_left(meta_ac)
		       < ocfs2_extend_meta_needed(fe))) {
		mlog(0, "filesystem is really fragmented...\n");
		status = -EAGAIN;
		reason = RESTART_META;
		goto leave;
	}

	status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
					clusters_to_add, &bit_off, &num_bits);
	if (status < 0) {
		if (status != -ENOSPC)
			mlog_errno(status);
		goto leave;
	}

	BUG_ON(num_bits > clusters_to_add);

	/* reserve our write early -- insert_extent may update the inode */
	status = ocfs2_journal_access(handle, inode, fe_bh,
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto leave;
	}

	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
	mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
	     num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
	status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
				     *logical_offset, block, num_bits,
				     flags, meta_ac);
	if (status < 0) {
		mlog_errno(status);
		goto leave;
	}

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

	clusters_to_add -= num_bits;
	*logical_offset += num_bits;

	if (clusters_to_add) {
		mlog(0, "need to alloc once more, clusters = %u, wanted = "
		     "%u\n", fe->i_clusters, clusters_to_add);
		status = -EAGAIN;
		reason = RESTART_TRANS;
	}

leave:
	mlog_exit(status);
	if (reason_ret)
		*reason_ret = reason;
	return status;
}

/*
 * For a given allocation, determine which allocators will need to be
 * accessed, and lock them, reserving the appropriate number of bits.
 *
 * Sparse file systems call this from ocfs2_write_begin_nolock()
 * and ocfs2_allocate_unwritten_extents().
 *
 * File systems which don't support holes call this from
 * ocfs2_extend_allocation().
 */
int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
			  u32 clusters_to_add, u32 extents_to_split,
			  struct ocfs2_alloc_context **data_ac,
			  struct ocfs2_alloc_context **meta_ac)
{
	int ret = 0, num_free_extents;
	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	*meta_ac = NULL;
	if (data_ac)
		*data_ac = NULL;

	BUG_ON(clusters_to_add != 0 && data_ac == NULL);

	mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
	     "clusters_to_add = %u, extents_to_split = %u\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno, (long long)i_size_read(inode),
	     le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);

	num_free_extents = ocfs2_num_free_extents(osb, inode, di);
	if (num_free_extents < 0) {
		ret = num_free_extents;
		mlog_errno(ret);
		goto out;
	}

	/*
	 * Sparse allocation file systems need to be more conservative
	 * with reserving room for expansion - the actual allocation
	 * happens while we've got a journal handle open so re-taking
	 * a cluster lock (because we ran out of room for another
	 * extent) will violate ordering rules.
	 *
	 * Most of the time we'll only be seeing this 1 cluster at a time
	 * anyway.
	 *
	 * Always lock for any unwritten extents - we might want to
	 * add blocks during a split.
	 */
	if (!num_free_extents ||
	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
		ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
		if (ret < 0) {
			if (ret != -ENOSPC)
				mlog_errno(ret);
			goto out;
		}
	}

	if (clusters_to_add == 0)
		goto out;

	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
	if (ret < 0) {
		if (ret != -ENOSPC)
			mlog_errno(ret);
		goto out;
	}

out:
	if (ret) {
		if (*meta_ac) {
			ocfs2_free_alloc_context(*meta_ac);
			*meta_ac = NULL;
		}

		/*
		 * We cannot have an error and a non null *data_ac.
		 */
	}

	return ret;
}

static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
				     u32 clusters_to_add, int mark_unwritten)
{
	int status = 0;
	int restart_func = 0;
	int credits;
	u32 prev_clusters;
	struct buffer_head *bh = NULL;
	struct ocfs2_dinode *fe = NULL;
	handle_t *handle = NULL;
	struct ocfs2_alloc_context *data_ac = NULL;
	struct ocfs2_alloc_context *meta_ac = NULL;
	enum ocfs2_alloc_restarted why;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);

	mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);

	/*
	 * This function only exists for file systems which don't
	 * support holes.
	 */
	BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));

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

	fe = (struct ocfs2_dinode *) bh->b_data;
	if (!OCFS2_IS_VALID_DINODE(fe)) {
		OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
		status = -EIO;
		goto leave;
	}

restart_all:
	BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);

	status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
				       &meta_ac);
	if (status) {
		mlog_errno(status);
		goto leave;
	}

	credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
	handle = ocfs2_start_trans(osb, credits);
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		handle = NULL;
		mlog_errno(status);
		goto leave;
	}

restarted_transaction:
	/* reserve a write to the file entry early on - that we if we
	 * run out of credits in the allocation path, we can still
	 * update i_size. */
	status = ocfs2_journal_access(handle, inode, bh,
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto leave;
	}

	prev_clusters = OCFS2_I(inode)->ip_clusters;

	status = ocfs2_do_extend_allocation(osb,
					    inode,
					    &logical_start,
					    clusters_to_add,
					    mark_unwritten,
					    bh,
					    handle,
					    data_ac,
					    meta_ac,
					    &why);
	if ((status < 0) && (status != -EAGAIN)) {
		if (status != -ENOSPC)
			mlog_errno(status);
		goto leave;
	}

	status = ocfs2_journal_dirty(handle, bh);
	if (status < 0) {
		mlog_errno(status);
		goto leave;
	}

	spin_lock(&OCFS2_I(inode)->ip_lock);
	clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
	spin_unlock(&OCFS2_I(inode)->ip_lock);

	if (why != RESTART_NONE && clusters_to_add) {
		if (why == RESTART_META) {
			mlog(0, "restarting function.\n");
			restart_func = 1;
		} else {
			BUG_ON(why != RESTART_TRANS);

			mlog(0, "restarting transaction.\n");
			/* TODO: This can be more intelligent. */
			credits = ocfs2_calc_extend_credits(osb->sb,
							    fe,
							    clusters_to_add);
			status = ocfs2_extend_trans(handle, credits);
			if (status < 0) {
				/* handle still has to be committed at
				 * this point. */
				status = -ENOMEM;
				mlog_errno(status);
				goto leave;
			}
			goto restarted_transaction;
		}
	}

	mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
	     le32_to_cpu(fe->i_clusters),
	     (unsigned long long)le64_to_cpu(fe->i_size));
	mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
	     OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));

leave:
	if (handle) {
		ocfs2_commit_trans(osb, handle);
		handle = NULL;
	}
	if (data_ac) {
		ocfs2_free_alloc_context(data_ac);
		data_ac = NULL;
	}
	if (meta_ac) {
		ocfs2_free_alloc_context(meta_ac);
		meta_ac = NULL;
	}
	if ((!status) && restart_func) {
		restart_func = 0;
		goto restart_all;
	}
	if (bh) {
		brelse(bh);
		bh = NULL;
	}

	mlog_exit(status);
	return status;
}

/* Some parts of this taken from generic_cont_expand, which turned out
 * to be too fragile to do exactly what we need without us having to
 * worry about recursive locking in ->prepare_write() and
 * ->commit_write(). */
static int ocfs2_write_zero_page(struct inode *inode,
				 u64 size)
{
	struct address_space *mapping = inode->i_mapping;
	struct page *page;
	unsigned long index;
	unsigned int offset;
	handle_t *handle = NULL;
	int ret;

	offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
	/* ugh.  in prepare/commit_write, if from==to==start of block, we 
	** skip the prepare.  make sure we never send an offset for the start
	** of a block
	*/
	if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
		offset++;
	}
	index = size >> PAGE_CACHE_SHIFT;

	page = grab_cache_page(mapping, index);
	if (!page) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
	if (ret < 0) {
		mlog_errno(ret);
		goto out_unlock;
	}

	if (ocfs2_should_order_data(inode)) {
		handle = ocfs2_start_walk_page_trans(inode, page, offset,
						     offset);
		if (IS_ERR(handle)) {
			ret = PTR_ERR(handle);
			handle = NULL;
			goto out_unlock;
		}
	}

	/* must not update i_size! */
	ret = block_commit_write(page, offset, offset);
	if (ret < 0)
		mlog_errno(ret);
	else
		ret = 0;

	if (handle)
		ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
out_unlock:
	unlock_page(page);
	page_cache_release(page);
out:
	return ret;
}

static int ocfs2_zero_extend(struct inode *inode,
			     u64 zero_to_size)
{
	int ret = 0;
	u64 start_off;
	struct super_block *sb = inode->i_sb;

	start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
	while (start_off < zero_to_size) {
		ret = ocfs2_write_zero_page(inode, start_off);
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}

		start_off += sb->s_blocksize;

		/*
		 * Very large extends have the potential to lock up
		 * the cpu for extended periods of time.
		 */
		cond_resched();
	}

out:
	return ret;
}

int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
{
	int ret;
	u32 clusters_to_add;
	struct ocfs2_inode_info *oi = OCFS2_I(inode);

	clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
	if (clusters_to_add < oi->ip_clusters)
		clusters_to_add = 0;
	else
		clusters_to_add -= oi->ip_clusters;

	if (clusters_to_add) {
		ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
						clusters_to_add, 0);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

	/*
	 * Call this even if we don't add any clusters to the tree. We
	 * still need to zero the area between the old i_size and the
	 * new i_size.
	 */
	ret = ocfs2_zero_extend(inode, zero_to);
	if (ret < 0)
		mlog_errno(ret);

out:
	return ret;
}

static int ocfs2_extend_file(struct inode *inode,
			     struct buffer_head *di_bh,
			     u64 new_i_size)
{
	int ret = 0;
	struct ocfs2_inode_info *oi = OCFS2_I(inode);

	BUG_ON(!di_bh);

	/* setattr sometimes calls us like this. */
	if (new_i_size == 0)
		goto out;

	if (i_size_read(inode) == new_i_size)
  		goto out;
	BUG_ON(new_i_size < i_size_read(inode));

	/*
	 * Fall through for converting inline data, even if the fs
	 * supports sparse files.
	 *
	 * The check for inline data here is legal - nobody can add
	 * the feature since we have i_mutex. We must check it again
	 * after acquiring ip_alloc_sem though, as paths like mmap
	 * might have raced us to converting the inode to extents.
	 */
	if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
	    && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
		goto out_update_size;

	/*
	 * The alloc sem blocks people in read/write from reading our
	 * allocation until we're done changing it. We depend on
	 * i_mutex to block other extend/truncate calls while we're
	 * here.
	 */
	down_write(&oi->ip_alloc_sem);

	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
		/*
		 * We can optimize small extends by keeping the inodes
		 * inline data.
		 */
		if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
			up_write(&oi->ip_alloc_sem);
			goto out_update_size;
		}

		ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
		if (ret) {
			up_write(&oi->ip_alloc_sem);

			mlog_errno(ret);
			goto out;
		}
	}

	if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
		ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);

	up_write(&oi->ip_alloc_sem);

	if (ret < 0) {
		mlog_errno(ret);
		goto out;
	}