suballoc.c

ocfs1.4.1 oracle分布式文件系统

				mlog_errno(status);
			goto bail;
		}
		atomic_inc(&osb->alloc_stats.bg_extends);

		/* You should never ask for this much metadata */
		BUG_ON(bits_wanted >
		       (le32_to_cpu(fe->id1.bitmap1.i_total)
			- le32_to_cpu(fe->id1.bitmap1.i_used)));
	}

	get_bh(bh);
	ac->ac_bh = bh;
bail:
	if (bh)
		brelse(bh);

	mlog_exit(status);
	return status;
}

int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
			       struct ocfs2_dinode *fe,
			       struct ocfs2_alloc_context **ac)
{
	int status;
	u32 slot;

	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
	if (!(*ac)) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}

	(*ac)->ac_bits_wanted = ocfs2_extend_meta_needed(fe);
	(*ac)->ac_which = OCFS2_AC_USE_META;
	slot = osb->slot_num;
	(*ac)->ac_group_search = ocfs2_block_group_search;

	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
					     EXTENT_ALLOC_SYSTEM_INODE,
					     slot, ALLOC_NEW_GROUP);
	if (status < 0) {
		if (status != -ENOSPC)
			mlog_errno(status);
		goto bail;
	}

	status = 0;
bail:
	if ((status < 0) && *ac) {
		ocfs2_free_alloc_context(*ac);
		*ac = NULL;
	}

	mlog_exit(status);
	return status;
}

static int ocfs2_steal_inode_from_other_nodes(struct ocfs2_super *osb,
					      struct ocfs2_alloc_context *ac)
{
	int i, status = -ENOSPC;
	s16 slot = ocfs2_get_inode_steal_slot(osb);

	/* Start to steal inodes from the first slot after ours. */
	if (slot == OCFS2_INVALID_SLOT)
		slot = osb->slot_num + 1;

	for (i = 0; i < osb->max_slots; i++, slot++) {
		if (slot == osb->max_slots)
			slot = 0;

		if (slot == osb->slot_num)
			continue;

		status = ocfs2_reserve_suballoc_bits(osb, ac,
						     INODE_ALLOC_SYSTEM_INODE,
						     slot, NOT_ALLOC_NEW_GROUP);
		if (status >= 0) {
			ocfs2_set_inode_steal_slot(osb, slot);
			break;
		}

		ocfs2_free_ac_resource(ac);
	}

	return status;
}

int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
			    struct ocfs2_alloc_context **ac)
{
	int status;
	s16 slot = ocfs2_get_inode_steal_slot(osb);

	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
	if (!(*ac)) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}

	(*ac)->ac_bits_wanted = 1;
	(*ac)->ac_which = OCFS2_AC_USE_INODE;

	(*ac)->ac_group_search = ocfs2_block_group_search;

	/*
	 * slot is set when we successfully steal inode from other nodes.
	 * It is reset in 3 places:
	 * 1. when we flush the truncate log
	 * 2. when we complete local alloc recovery.
	 * 3. when we successfully allocate from our own slot.
	 * After it is set, we will go on stealing inodes until we find the
	 * need to check our slots to see whether there is some space for us.
	 */
	if (slot != OCFS2_INVALID_SLOT &&
	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_INODES_TO_STEAL)
		goto inode_steal;

	atomic_set(&osb->s_num_inodes_stolen, 0);
	status = ocfs2_reserve_suballoc_bits(osb, *ac,
					     INODE_ALLOC_SYSTEM_INODE,
					     osb->slot_num, ALLOC_NEW_GROUP);
	if (status >= 0) {
		status = 0;

		/*
		 * Some inodes must be freed by us, so try to allocate
		 * from our own next time.
		 */
		if (slot != OCFS2_INVALID_SLOT)
			ocfs2_init_inode_steal_slot(osb);
		goto bail;
	} else if (status < 0 && status != -ENOSPC) {
		mlog_errno(status);
		goto bail;
	}

	ocfs2_free_ac_resource(*ac);

inode_steal:
	status = ocfs2_steal_inode_from_other_nodes(osb, *ac);
	atomic_inc(&osb->s_num_inodes_stolen);
	if (status < 0) {
		if (status != -ENOSPC)
			mlog_errno(status);
		goto bail;
	}

	status = 0;
bail:
	if ((status < 0) && *ac) {
		ocfs2_free_alloc_context(*ac);
		*ac = NULL;
	}

	mlog_exit(status);
	return status;
}

/* local alloc code has to do the same thing, so rather than do this
 * twice.. */
int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
				      struct ocfs2_alloc_context *ac)
{
	int status;

	ac->ac_which = OCFS2_AC_USE_MAIN;
	ac->ac_group_search = ocfs2_cluster_group_search;

	status = ocfs2_reserve_suballoc_bits(osb, ac,
					     GLOBAL_BITMAP_SYSTEM_INODE,
					     OCFS2_INVALID_SLOT,
					     ALLOC_NEW_GROUP);
	if (status < 0 && status != -ENOSPC) {
		mlog_errno(status);
		goto bail;
	}

bail:
	return status;
}

/* Callers don't need to care which bitmap (local alloc or main) to
 * use so we figure it out for them, but unfortunately this clutters
 * things a bit. */
int ocfs2_reserve_clusters(struct ocfs2_super *osb,
			   u32 bits_wanted,
			   struct ocfs2_alloc_context **ac)
{
	int status;

	mlog_entry_void();

	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
	if (!(*ac)) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}

	(*ac)->ac_bits_wanted = bits_wanted;

	status = -ENOSPC;
	if (ocfs2_alloc_should_use_local(osb, bits_wanted)) {
		status = ocfs2_reserve_local_alloc_bits(osb,
							bits_wanted,
							*ac);
		if ((status < 0) && (status != -ENOSPC)) {
			mlog_errno(status);
			goto bail;
		} else if (status == -ENOSPC) {
			/* reserve_local_bits will return enospc with
			 * the local alloc inode still locked, so we
			 * can change this safely here. */
			mlog(0, "Disabling local alloc\n");
			/* We set to OCFS2_LA_DISABLED so that umount
			 * can clean up what's left of the local
			 * allocation */
			osb->local_alloc_state = OCFS2_LA_DISABLED;
		}
	}

	if (status == -ENOSPC) {
		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
		if (status < 0) {
			if (status != -ENOSPC)
				mlog_errno(status);
			goto bail;
		}
	}

	status = 0;
bail:
	if ((status < 0) && *ac) {
		ocfs2_free_alloc_context(*ac);
		*ac = NULL;
	}

	mlog_exit(status);
	return status;
}

/*
 * More or less lifted from ext3. I'll leave their description below:
 *
 * "For ext3 allocations, we must not reuse any blocks which are
 * allocated in the bitmap buffer's "last committed data" copy.  This
 * prevents deletes from freeing up the page for reuse until we have
 * committed the delete transaction.
 *
 * If we didn't do this, then deleting something and reallocating it as
 * data would allow the old block to be overwritten before the
 * transaction committed (because we force data to disk before commit).
 * This would lead to corruption if we crashed between overwriting the
 * data and committing the delete.
 *
 * @@@ We may want to make this allocation behaviour conditional on
 * data-writes at some point, and disable it for metadata allocations or
 * sync-data inodes."
 *
 * Note: OCFS2 already does this differently for metadata vs data
 * allocations, as those bitmaps are separate and undo access is never
 * called on a metadata group descriptor.
 */
static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
					 int nr)
{
	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;

	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
		return 0;
	if (!buffer_jbd(bg_bh) || !bh2jh(bg_bh)->b_committed_data)
		return 1;

	bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
	return !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
}

static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
					     struct buffer_head *bg_bh,
					     unsigned int bits_wanted,
					     unsigned int total_bits,
					     u16 *bit_off,
					     u16 *bits_found)
{
	void *bitmap;
	u16 best_offset, best_size;
	int offset, start, found, status = 0;
	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;

	if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
		OCFS2_RO_ON_INVALID_GROUP_DESC(osb->sb, bg);
		return -EIO;
	}

	found = start = best_offset = best_size = 0;
	bitmap = bg->bg_bitmap;

	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
		if (offset == total_bits)
			break;

		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
			/* We found a zero, but we can't use it as it
			 * hasn't been put to disk yet! */
			found = 0;
			start = offset + 1;
		} else if (offset == start) {
			/* we found a zero */
			found++;
			/* move start to the next bit to test */
			start++;
		} else {
			/* got a zero after some ones */
			found = 1;
			start = offset + 1;
		}
		if (found > best_size) {
			best_size = found;
			best_offset = start - found;
		}
		/* we got everything we needed */
		if (found == bits_wanted) {
			/* mlog(0, "Found it all!\n"); */
			break;
		}
	}

	/* XXX: I think the first clause is equivalent to the second
	 * 	- jlbec */
	if (found == bits_wanted) {
		*bit_off = start - found;
		*bits_found = found;
	} else if (best_size) {
		*bit_off = best_offset;
		*bits_found = best_size;
	} else {
		status = -ENOSPC;
		/* No error log here -- see the comment above
		 * ocfs2_test_bg_bit_allocatable */
	}

	return status;
}

static inline int ocfs2_block_group_set_bits(handle_t *handle,
					     struct inode *alloc_inode,
					     struct ocfs2_group_desc *bg,
					     struct buffer_head *group_bh,
					     unsigned int bit_off,
					     unsigned int num_bits)
{
	int status;
	void *bitmap = bg->bg_bitmap;
	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;

	mlog_entry_void();

	if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
		OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
		status = -EIO;
		goto bail;
	}
	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);

	mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
	     num_bits);

	if (ocfs2_is_cluster_bitmap(alloc_inode))
		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;

	status = ocfs2_journal_access(handle,
				      alloc_inode,
				      group_bh,
				      journal_type);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);

	while(num_bits--)
		ocfs2_set_bit(bit_off++, bitmap);

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

bail:
	mlog_exit(status);
	return status;
}

/* find the one with the most empty bits */
static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
{
	u16 curr, best;

	BUG_ON(!cl->cl_next_free_rec);

	best = curr = 0;
	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
		    le32_to_cpu(cl->cl_recs[best].c_free))
			best = curr;
		curr++;
	}

	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
	return best;
}

static int ocfs2_relink_block_group(handle_t *handle,
				    struct inode *alloc_inode,
				    struct buffer_head *fe_bh,
				    struct buffer_head *bg_bh,
				    struct buffer_head *prev_bg_bh,
				    u16 chain)
{
	int status;
	/* there is a really tiny chance the journal calls could fail,
	 * but we wouldn't want inconsistent blocks in *any* case. */
	u64 fe_ptr, bg_ptr, prev_bg_ptr;
	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;

	if (!OCFS2_IS_VALID_DINODE(fe)) {
		OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
		status = -EIO;
		goto out;
	}
	if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
		OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
		status = -EIO;
		goto out;
	}
	if (!OCFS2_IS_VALID_GROUP_DESC(prev_bg)) {
		OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, prev_bg);
		status = -EIO;
		goto out;
	}

	mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
	     (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
	     (unsigned long long)le64_to_cpu(bg->bg_blkno),
	     (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));

	fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
	bg_ptr = le64_to_cpu(bg->bg_next_group);
	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);

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

	prev_bg->bg_next_group = bg->bg_next_group;

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