suballoc.c

ocfs1.4.1 oracle分布式文件系统

	status = ocfs2_journal_access(handle, alloc_inode, bg_bh,
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto out_rollback;
	}

	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;

	status = ocfs2_journal_dirty(handle, bg_bh);
	if (status < 0) {
		mlog_errno(status);
		goto out_rollback;
	}

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

	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;

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

	status = 0;
out_rollback:
	if (status < 0) {
		fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
		bg->bg_next_group = cpu_to_le64(bg_ptr);
		prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
	}
out:
	mlog_exit(status);
	return status;
}

static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
						     u32 wanted)
{
	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
}

/* return 0 on success, -ENOSPC to keep searching and any other < 0
 * value on error. */
static int ocfs2_cluster_group_search(struct inode *inode,
				      struct buffer_head *group_bh,
				      u32 bits_wanted, u32 min_bits,
				      u16 *bit_off, u16 *bits_found)
{
	int search = -ENOSPC;
	int ret;
	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
	u16 tmp_off, tmp_found;
	unsigned int max_bits, gd_cluster_off;

	BUG_ON(!ocfs2_is_cluster_bitmap(inode));

	if (gd->bg_free_bits_count) {
		max_bits = le16_to_cpu(gd->bg_bits);

		/* Tail groups in cluster bitmaps which aren't cpg
		 * aligned are prone to partial extention by a failed
		 * fs resize. If the file system resize never got to
		 * update the dinode cluster count, then we don't want
		 * to trust any clusters past it, regardless of what
		 * the group descriptor says. */
		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
							  le64_to_cpu(gd->bg_blkno));
		if ((gd_cluster_off + max_bits) >
		    OCFS2_I(inode)->ip_clusters) {
			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
			mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
			     (unsigned long long)le64_to_cpu(gd->bg_blkno),
			     le16_to_cpu(gd->bg_bits),
			     OCFS2_I(inode)->ip_clusters, max_bits);
		}

		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
							group_bh, bits_wanted,
							max_bits,
							&tmp_off, &tmp_found);
		if (ret)
			return ret;

		/* ocfs2_block_group_find_clear_bits() might
		 * return success, but we still want to return
		 * -ENOSPC unless it found the minimum number
		 * of bits. */
		if (min_bits <= tmp_found) {
			*bit_off = tmp_off;
			*bits_found = tmp_found;
			search = 0; /* success */
		}
	}

	return search;
}

static int ocfs2_block_group_search(struct inode *inode,
				    struct buffer_head *group_bh,
				    u32 bits_wanted, u32 min_bits,
				    u16 *bit_off, u16 *bits_found)
{
	int ret = -ENOSPC;
	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;

	BUG_ON(min_bits != 1);
	BUG_ON(ocfs2_is_cluster_bitmap(inode));

	if (bg->bg_free_bits_count)
		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
							group_bh, bits_wanted,
							le16_to_cpu(bg->bg_bits),
							bit_off, bits_found);

	return ret;
}

static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
				       handle_t *handle,
				       struct buffer_head *di_bh,
				       u32 num_bits,
				       u16 chain)
{
	int ret;
	u32 tmp_used;
	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;

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

	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);

	ret = ocfs2_journal_dirty(handle, di_bh);
	if (ret < 0)
		mlog_errno(ret);

out:
	return ret;
}

static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
				  handle_t *handle,
				  u32 bits_wanted,
				  u32 min_bits,
				  u16 *bit_off,
				  unsigned int *num_bits,
				  u64 gd_blkno,
				  u16 *bits_left)
{
	int ret;
	u16 found;
	struct buffer_head *group_bh = NULL;
	struct ocfs2_group_desc *gd;
	struct inode *alloc_inode = ac->ac_inode;

	ret = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb), gd_blkno,
			       &group_bh, OCFS2_BH_CACHED, alloc_inode);
	if (ret < 0) {
		mlog_errno(ret);
		return ret;
	}

	gd = (struct ocfs2_group_desc *) group_bh->b_data;
	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
		OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, gd);
		ret = -EIO;
		goto out;
	}

	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
				  bit_off, &found);
	if (ret < 0) {
		if (ret != -ENOSPC)
			mlog_errno(ret);
		goto out;
	}

	*num_bits = found;

	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
					       *num_bits,
					       le16_to_cpu(gd->bg_chain));
	if (ret < 0) {
		mlog_errno(ret);
		goto out;
	}

	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
					 *bit_off, *num_bits);
	if (ret < 0)
		mlog_errno(ret);

	*bits_left = le16_to_cpu(gd->bg_free_bits_count);

out:
	brelse(group_bh);

	return ret;
}

static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
			      handle_t *handle,
			      u32 bits_wanted,
			      u32 min_bits,
			      u16 *bit_off,
			      unsigned int *num_bits,
			      u64 *bg_blkno,
			      u16 *bits_left)
{
	int status;
	u16 chain, tmp_bits;
	u32 tmp_used;
	u64 next_group;
	struct inode *alloc_inode = ac->ac_inode;
	struct buffer_head *group_bh = NULL;
	struct buffer_head *prev_group_bh = NULL;
	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
	struct ocfs2_group_desc *bg;

	chain = ac->ac_chain;
	mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
	     bits_wanted, chain,
	     (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);

	status = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb),
				  le64_to_cpu(cl->cl_recs[chain].c_blkno),
				  &group_bh, OCFS2_BH_CACHED, alloc_inode);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
	bg = (struct ocfs2_group_desc *) group_bh->b_data;
	status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, bg);
	if (status) {
		mlog_errno(status);
		goto bail;
	}

	status = -ENOSPC;
	/* for now, the chain search is a bit simplistic. We just use
	 * the 1st group with any empty bits. */
	while ((status = ac->ac_group_search(alloc_inode, group_bh,
					     bits_wanted, min_bits, bit_off,
					     &tmp_bits)) == -ENOSPC) {
		if (!bg->bg_next_group)
			break;

		if (prev_group_bh) {
			brelse(prev_group_bh);
			prev_group_bh = NULL;
		}
		next_group = le64_to_cpu(bg->bg_next_group);
		prev_group_bh = group_bh;
		group_bh = NULL;
		status = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb),
					  next_group, &group_bh,
					  OCFS2_BH_CACHED, alloc_inode);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
		bg = (struct ocfs2_group_desc *) group_bh->b_data;
		status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, bg);
		if (status) {
			mlog_errno(status);
			goto bail;
		}
	}
	if (status < 0) {
		if (status != -ENOSPC)
			mlog_errno(status);
		goto bail;
	}

	mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
	     tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));

	*num_bits = tmp_bits;

	BUG_ON(*num_bits == 0);

	/*
	 * Keep track of previous block descriptor read. When
	 * we find a target, if we have read more than X
	 * number of descriptors, and the target is reasonably
	 * empty, relink him to top of his chain.
	 *
	 * We've read 0 extra blocks and only send one more to
	 * the transaction, yet the next guy to search has a
	 * much easier time.
	 *
	 * Do this *after* figuring out how many bits we're taking out
	 * of our target group.
	 */
	if (ac->ac_allow_chain_relink &&
	    (prev_group_bh) &&
	    (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
		status = ocfs2_relink_block_group(handle, alloc_inode,
						  ac->ac_bh, group_bh,
						  prev_group_bh, chain);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

	/* Ok, claim our bits now: set the info on dinode, chainlist
	 * and then the group */
	status = ocfs2_journal_access(handle,
				      alloc_inode,
				      ac->ac_bh,
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
	fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
	le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));

	status = ocfs2_journal_dirty(handle,
				     ac->ac_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	status = ocfs2_block_group_set_bits(handle,
					    alloc_inode,
					    bg,
					    group_bh,
					    *bit_off,
					    *num_bits);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
	     (unsigned long long)le64_to_cpu(fe->i_blkno));

	*bg_blkno = le64_to_cpu(bg->bg_blkno);
	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
bail:
	if (group_bh)
		brelse(group_bh);
	if (prev_group_bh)
		brelse(prev_group_bh);

	mlog_exit(status);
	return status;
}

/* will give out up to bits_wanted contiguous bits. */
static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
				     struct ocfs2_alloc_context *ac,
				     handle_t *handle,
				     u32 bits_wanted,
				     u32 min_bits,
				     u16 *bit_off,
				     unsigned int *num_bits,
				     u64 *bg_blkno)
{
	int status;
	u16 victim, i;
	u16 bits_left = 0;
	u64 hint_blkno = ac->ac_last_group;
	struct ocfs2_chain_list *cl;
	struct ocfs2_dinode *fe;

	mlog_entry_void();

	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
	BUG_ON(!ac->ac_bh);

	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
	if (!OCFS2_IS_VALID_DINODE(fe)) {
		OCFS2_RO_ON_INVALID_DINODE(osb->sb, fe);
		status = -EIO;
		goto bail;
	}
	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
		ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
			    "bits but only %u total.",
			    (unsigned long long)le64_to_cpu(fe->i_blkno),
			    le32_to_cpu(fe->id1.bitmap1.i_used),
			    le32_to_cpu(fe->id1.bitmap1.i_total));
		status = -EIO;
		goto bail;
	}

	if (hint_blkno) {
		/* Attempt to short-circuit the usual search mechanism
		 * by jumping straight to the most recently used
		 * allocation group. This helps us mantain some
		 * contiguousness across allocations. */
		status = ocfs2_search_one_group(ac, handle, bits_wanted,
						min_bits, bit_off, num_bits,
						hint_blkno, &bits_left);
		if (!status) {
			/* Be careful to update *bg_blkno here as the
			 * caller is expecting it to be filled in, and
			 * ocfs2_search_one_group() won't do that for
			 * us. */
			*bg_blkno = hint_blkno;
			goto set_hint;
		}
		if (status < 0 && status != -ENOSPC) {
			mlog_errno(status);
			goto bail;
		}
	}

	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;

	victim = ocfs2_find_victim_chain(cl);
	ac->ac_chain = victim;
	ac->ac_allow_chain_relink = 1;

	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
				    num_bits, bg_blkno, &bits_left);
	if (!status)
		goto set_hint;
	if (status < 0 && status != -ENOSPC) {
		mlog_errno(status);
		goto bail;
	}

	mlog(0, "Search of victim chain %u came up with nothing, "
	     "trying all chains now.\n", victim);

	/* If we didn't pick a good victim, then just default to
	 * searching each chain in order. Don't allow chain relinking
	 * because we only calculate enough journal credits for one
	 * relink per alloc. */
	ac->ac_allow_chain_relink = 0;
	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
		if (i == victim)
			continue;
		if (!cl->cl_recs[i].c_free)
			continue;

		ac->ac_chain = i;
		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
					    bit_off, num_bits, bg_blkno,
					    &bits_left);
		if (!status)
			break;
		if (status < 0 && status != -ENOSPC) {
			mlog_errno(status);
			goto bail;
		}
	}

set_hint: