2 * btree.c - NILFS B-tree.
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 #include <linux/slab.h>
24 #include <linux/string.h>
25 #include <linux/errno.h>
26 #include <linux/pagevec.h>
34 static struct nilfs_btree_path *nilfs_btree_alloc_path(void)
36 struct nilfs_btree_path *path;
37 int level = NILFS_BTREE_LEVEL_DATA;
39 path = kmem_cache_alloc(nilfs_btree_path_cache, GFP_NOFS);
43 for (; level < NILFS_BTREE_LEVEL_MAX; level++) {
44 path[level].bp_bh = NULL;
45 path[level].bp_sib_bh = NULL;
46 path[level].bp_index = 0;
47 path[level].bp_oldreq.bpr_ptr = NILFS_BMAP_INVALID_PTR;
48 path[level].bp_newreq.bpr_ptr = NILFS_BMAP_INVALID_PTR;
49 path[level].bp_op = NULL;
56 static void nilfs_btree_free_path(struct nilfs_btree_path *path)
58 int level = NILFS_BTREE_LEVEL_DATA;
60 for (; level < NILFS_BTREE_LEVEL_MAX; level++)
61 brelse(path[level].bp_bh);
63 kmem_cache_free(nilfs_btree_path_cache, path);
67 * B-tree node operations
69 static int nilfs_btree_get_new_block(const struct nilfs_bmap *btree,
70 __u64 ptr, struct buffer_head **bhp)
72 struct address_space *btnc = &NILFS_BMAP_I(btree)->i_btnode_cache;
73 struct buffer_head *bh;
75 bh = nilfs_btnode_create_block(btnc, ptr);
79 set_buffer_nilfs_volatile(bh);
84 static int nilfs_btree_node_get_flags(const struct nilfs_btree_node *node)
86 return node->bn_flags;
90 nilfs_btree_node_set_flags(struct nilfs_btree_node *node, int flags)
92 node->bn_flags = flags;
95 static int nilfs_btree_node_root(const struct nilfs_btree_node *node)
97 return nilfs_btree_node_get_flags(node) & NILFS_BTREE_NODE_ROOT;
100 static int nilfs_btree_node_get_level(const struct nilfs_btree_node *node)
102 return node->bn_level;
106 nilfs_btree_node_set_level(struct nilfs_btree_node *node, int level)
108 node->bn_level = level;
111 static int nilfs_btree_node_get_nchildren(const struct nilfs_btree_node *node)
113 return le16_to_cpu(node->bn_nchildren);
117 nilfs_btree_node_set_nchildren(struct nilfs_btree_node *node, int nchildren)
119 node->bn_nchildren = cpu_to_le16(nchildren);
122 static int nilfs_btree_node_size(const struct nilfs_bmap *btree)
124 return 1 << btree->b_inode->i_blkbits;
127 static int nilfs_btree_nchildren_per_block(const struct nilfs_bmap *btree)
129 return btree->b_nchildren_per_block;
133 nilfs_btree_node_dkeys(const struct nilfs_btree_node *node)
135 return (__le64 *)((char *)(node + 1) +
136 (nilfs_btree_node_root(node) ?
137 0 : NILFS_BTREE_NODE_EXTRA_PAD_SIZE));
141 nilfs_btree_node_dptrs(const struct nilfs_btree_node *node, int ncmax)
143 return (__le64 *)(nilfs_btree_node_dkeys(node) + ncmax);
147 nilfs_btree_node_get_key(const struct nilfs_btree_node *node, int index)
149 return le64_to_cpu(*(nilfs_btree_node_dkeys(node) + index));
153 nilfs_btree_node_set_key(struct nilfs_btree_node *node, int index, __u64 key)
155 *(nilfs_btree_node_dkeys(node) + index) = cpu_to_le64(key);
159 nilfs_btree_node_get_ptr(const struct nilfs_btree_node *node, int index,
162 return le64_to_cpu(*(nilfs_btree_node_dptrs(node, ncmax) + index));
166 nilfs_btree_node_set_ptr(struct nilfs_btree_node *node, int index, __u64 ptr,
169 *(nilfs_btree_node_dptrs(node, ncmax) + index) = cpu_to_le64(ptr);
172 static void nilfs_btree_node_init(struct nilfs_btree_node *node, int flags,
173 int level, int nchildren, int ncmax,
174 const __u64 *keys, const __u64 *ptrs)
180 nilfs_btree_node_set_flags(node, flags);
181 nilfs_btree_node_set_level(node, level);
182 nilfs_btree_node_set_nchildren(node, nchildren);
184 dkeys = nilfs_btree_node_dkeys(node);
185 dptrs = nilfs_btree_node_dptrs(node, ncmax);
186 for (i = 0; i < nchildren; i++) {
187 dkeys[i] = cpu_to_le64(keys[i]);
188 dptrs[i] = cpu_to_le64(ptrs[i]);
192 /* Assume the buffer heads corresponding to left and right are locked. */
193 static void nilfs_btree_node_move_left(struct nilfs_btree_node *left,
194 struct nilfs_btree_node *right,
195 int n, int lncmax, int rncmax)
197 __le64 *ldkeys, *rdkeys;
198 __le64 *ldptrs, *rdptrs;
199 int lnchildren, rnchildren;
201 ldkeys = nilfs_btree_node_dkeys(left);
202 ldptrs = nilfs_btree_node_dptrs(left, lncmax);
203 lnchildren = nilfs_btree_node_get_nchildren(left);
205 rdkeys = nilfs_btree_node_dkeys(right);
206 rdptrs = nilfs_btree_node_dptrs(right, rncmax);
207 rnchildren = nilfs_btree_node_get_nchildren(right);
209 memcpy(ldkeys + lnchildren, rdkeys, n * sizeof(*rdkeys));
210 memcpy(ldptrs + lnchildren, rdptrs, n * sizeof(*rdptrs));
211 memmove(rdkeys, rdkeys + n, (rnchildren - n) * sizeof(*rdkeys));
212 memmove(rdptrs, rdptrs + n, (rnchildren - n) * sizeof(*rdptrs));
216 nilfs_btree_node_set_nchildren(left, lnchildren);
217 nilfs_btree_node_set_nchildren(right, rnchildren);
220 /* Assume that the buffer heads corresponding to left and right are locked. */
221 static void nilfs_btree_node_move_right(struct nilfs_btree_node *left,
222 struct nilfs_btree_node *right,
223 int n, int lncmax, int rncmax)
225 __le64 *ldkeys, *rdkeys;
226 __le64 *ldptrs, *rdptrs;
227 int lnchildren, rnchildren;
229 ldkeys = nilfs_btree_node_dkeys(left);
230 ldptrs = nilfs_btree_node_dptrs(left, lncmax);
231 lnchildren = nilfs_btree_node_get_nchildren(left);
233 rdkeys = nilfs_btree_node_dkeys(right);
234 rdptrs = nilfs_btree_node_dptrs(right, rncmax);
235 rnchildren = nilfs_btree_node_get_nchildren(right);
237 memmove(rdkeys + n, rdkeys, rnchildren * sizeof(*rdkeys));
238 memmove(rdptrs + n, rdptrs, rnchildren * sizeof(*rdptrs));
239 memcpy(rdkeys, ldkeys + lnchildren - n, n * sizeof(*rdkeys));
240 memcpy(rdptrs, ldptrs + lnchildren - n, n * sizeof(*rdptrs));
244 nilfs_btree_node_set_nchildren(left, lnchildren);
245 nilfs_btree_node_set_nchildren(right, rnchildren);
248 /* Assume that the buffer head corresponding to node is locked. */
249 static void nilfs_btree_node_insert(struct nilfs_btree_node *node, int index,
250 __u64 key, __u64 ptr, int ncmax)
256 dkeys = nilfs_btree_node_dkeys(node);
257 dptrs = nilfs_btree_node_dptrs(node, ncmax);
258 nchildren = nilfs_btree_node_get_nchildren(node);
259 if (index < nchildren) {
260 memmove(dkeys + index + 1, dkeys + index,
261 (nchildren - index) * sizeof(*dkeys));
262 memmove(dptrs + index + 1, dptrs + index,
263 (nchildren - index) * sizeof(*dptrs));
265 dkeys[index] = cpu_to_le64(key);
266 dptrs[index] = cpu_to_le64(ptr);
268 nilfs_btree_node_set_nchildren(node, nchildren);
271 /* Assume that the buffer head corresponding to node is locked. */
272 static void nilfs_btree_node_delete(struct nilfs_btree_node *node, int index,
273 __u64 *keyp, __u64 *ptrp, int ncmax)
281 dkeys = nilfs_btree_node_dkeys(node);
282 dptrs = nilfs_btree_node_dptrs(node, ncmax);
283 key = le64_to_cpu(dkeys[index]);
284 ptr = le64_to_cpu(dptrs[index]);
285 nchildren = nilfs_btree_node_get_nchildren(node);
291 if (index < nchildren - 1) {
292 memmove(dkeys + index, dkeys + index + 1,
293 (nchildren - index - 1) * sizeof(*dkeys));
294 memmove(dptrs + index, dptrs + index + 1,
295 (nchildren - index - 1) * sizeof(*dptrs));
298 nilfs_btree_node_set_nchildren(node, nchildren);
301 static int nilfs_btree_node_lookup(const struct nilfs_btree_node *node,
302 __u64 key, int *indexp)
305 int index, low, high, s;
309 high = nilfs_btree_node_get_nchildren(node) - 1;
312 while (low <= high) {
313 index = (low + high) / 2;
314 nkey = nilfs_btree_node_get_key(node, index);
318 } else if (nkey < key) {
328 if (nilfs_btree_node_get_level(node) > NILFS_BTREE_LEVEL_NODE_MIN) {
329 if (s > 0 && index > 0)
341 * nilfs_btree_node_broken - verify consistency of btree node
342 * @node: btree node block to be examined
343 * @size: node size (in bytes)
344 * @blocknr: block number
346 * Return Value: If node is broken, 1 is returned. Otherwise, 0 is returned.
348 static int nilfs_btree_node_broken(const struct nilfs_btree_node *node,
349 size_t size, sector_t blocknr)
351 int level, flags, nchildren;
354 level = nilfs_btree_node_get_level(node);
355 flags = nilfs_btree_node_get_flags(node);
356 nchildren = nilfs_btree_node_get_nchildren(node);
358 if (unlikely(level < NILFS_BTREE_LEVEL_NODE_MIN ||
359 level >= NILFS_BTREE_LEVEL_MAX ||
360 (flags & NILFS_BTREE_NODE_ROOT) ||
362 nchildren > NILFS_BTREE_NODE_NCHILDREN_MAX(size))) {
363 printk(KERN_CRIT "NILFS: bad btree node (blocknr=%llu): "
364 "level = %d, flags = 0x%x, nchildren = %d\n",
365 (unsigned long long)blocknr, level, flags, nchildren);
371 int nilfs_btree_broken_node_block(struct buffer_head *bh)
375 if (buffer_nilfs_checked(bh))
378 ret = nilfs_btree_node_broken((struct nilfs_btree_node *)bh->b_data,
379 bh->b_size, bh->b_blocknr);
381 set_buffer_nilfs_checked(bh);
385 static struct nilfs_btree_node *
386 nilfs_btree_get_root(const struct nilfs_bmap *btree)
388 return (struct nilfs_btree_node *)btree->b_u.u_data;
391 static struct nilfs_btree_node *
392 nilfs_btree_get_nonroot_node(const struct nilfs_btree_path *path, int level)
394 return (struct nilfs_btree_node *)path[level].bp_bh->b_data;
397 static struct nilfs_btree_node *
398 nilfs_btree_get_sib_node(const struct nilfs_btree_path *path, int level)
400 return (struct nilfs_btree_node *)path[level].bp_sib_bh->b_data;
403 static int nilfs_btree_height(const struct nilfs_bmap *btree)
405 return nilfs_btree_node_get_level(nilfs_btree_get_root(btree)) + 1;
408 static struct nilfs_btree_node *
409 nilfs_btree_get_node(const struct nilfs_bmap *btree,
410 const struct nilfs_btree_path *path,
411 int level, int *ncmaxp)
413 struct nilfs_btree_node *node;
415 if (level == nilfs_btree_height(btree) - 1) {
416 node = nilfs_btree_get_root(btree);
417 *ncmaxp = NILFS_BTREE_ROOT_NCHILDREN_MAX;
419 node = nilfs_btree_get_nonroot_node(path, level);
420 *ncmaxp = nilfs_btree_nchildren_per_block(btree);
426 nilfs_btree_bad_node(struct nilfs_btree_node *node, int level)
428 if (unlikely(nilfs_btree_node_get_level(node) != level)) {
430 printk(KERN_CRIT "NILFS: btree level mismatch: %d != %d\n",
431 nilfs_btree_node_get_level(node), level);
437 struct nilfs_btree_readahead_info {
438 struct nilfs_btree_node *node; /* parent node */
439 int max_ra_blocks; /* max nof blocks to read ahead */
440 int index; /* current index on the parent node */
441 int ncmax; /* nof children in the parent node */
444 static int __nilfs_btree_get_block(const struct nilfs_bmap *btree, __u64 ptr,
445 struct buffer_head **bhp,
446 const struct nilfs_btree_readahead_info *ra)
448 struct address_space *btnc = &NILFS_BMAP_I(btree)->i_btnode_cache;
449 struct buffer_head *bh, *ra_bh;
450 sector_t submit_ptr = 0;
453 ret = nilfs_btnode_submit_block(btnc, ptr, 0, READ, &bh, &submit_ptr);
464 /* read ahead sibling nodes */
465 for (n = ra->max_ra_blocks, i = ra->index + 1;
466 n > 0 && i < ra->ncmax; n--, i++) {
467 ptr2 = nilfs_btree_node_get_ptr(ra->node, i, ra->ncmax);
469 ret = nilfs_btnode_submit_block(btnc, ptr2, 0, READA,
470 &ra_bh, &submit_ptr);
471 if (likely(!ret || ret == -EEXIST))
473 else if (ret != -EBUSY)
475 if (!buffer_locked(bh))
483 if (!buffer_uptodate(bh)) {
489 if (nilfs_btree_broken_node_block(bh)) {
490 clear_buffer_uptodate(bh);
499 static int nilfs_btree_get_block(const struct nilfs_bmap *btree, __u64 ptr,
500 struct buffer_head **bhp)
502 return __nilfs_btree_get_block(btree, ptr, bhp, NULL);
505 static int nilfs_btree_do_lookup(const struct nilfs_bmap *btree,
506 struct nilfs_btree_path *path,
507 __u64 key, __u64 *ptrp, int minlevel,
510 struct nilfs_btree_node *node;
511 struct nilfs_btree_readahead_info p, *ra;
513 int level, index, found, ncmax, ret;
515 node = nilfs_btree_get_root(btree);
516 level = nilfs_btree_node_get_level(node);
517 if (level < minlevel || nilfs_btree_node_get_nchildren(node) <= 0)
520 found = nilfs_btree_node_lookup(node, key, &index);
521 ptr = nilfs_btree_node_get_ptr(node, index,
522 NILFS_BTREE_ROOT_NCHILDREN_MAX);
523 path[level].bp_bh = NULL;
524 path[level].bp_index = index;
526 ncmax = nilfs_btree_nchildren_per_block(btree);
528 while (--level >= minlevel) {
530 if (level == NILFS_BTREE_LEVEL_NODE_MIN && readahead) {
531 p.node = nilfs_btree_get_node(btree, path, level + 1,
537 ret = __nilfs_btree_get_block(btree, ptr, &path[level].bp_bh,
542 node = nilfs_btree_get_nonroot_node(path, level);
543 if (nilfs_btree_bad_node(node, level))
546 found = nilfs_btree_node_lookup(node, key, &index);
550 ptr = nilfs_btree_node_get_ptr(node, index, ncmax);
552 WARN_ON(found || level != NILFS_BTREE_LEVEL_NODE_MIN);
554 ptr = NILFS_BMAP_INVALID_PTR;
556 path[level].bp_index = index;
567 static int nilfs_btree_do_lookup_last(const struct nilfs_bmap *btree,
568 struct nilfs_btree_path *path,
569 __u64 *keyp, __u64 *ptrp)
571 struct nilfs_btree_node *node;
573 int index, level, ncmax, ret;
575 node = nilfs_btree_get_root(btree);
576 index = nilfs_btree_node_get_nchildren(node) - 1;
579 level = nilfs_btree_node_get_level(node);
580 ptr = nilfs_btree_node_get_ptr(node, index,
581 NILFS_BTREE_ROOT_NCHILDREN_MAX);
582 path[level].bp_bh = NULL;
583 path[level].bp_index = index;
584 ncmax = nilfs_btree_nchildren_per_block(btree);
586 for (level--; level > 0; level--) {
587 ret = nilfs_btree_get_block(btree, ptr, &path[level].bp_bh);
590 node = nilfs_btree_get_nonroot_node(path, level);
591 if (nilfs_btree_bad_node(node, level))
593 index = nilfs_btree_node_get_nchildren(node) - 1;
594 ptr = nilfs_btree_node_get_ptr(node, index, ncmax);
595 path[level].bp_index = index;
599 *keyp = nilfs_btree_node_get_key(node, index);
606 static int nilfs_btree_lookup(const struct nilfs_bmap *btree,
607 __u64 key, int level, __u64 *ptrp)
609 struct nilfs_btree_path *path;
612 path = nilfs_btree_alloc_path();
616 ret = nilfs_btree_do_lookup(btree, path, key, ptrp, level, 0);
618 nilfs_btree_free_path(path);
623 static int nilfs_btree_lookup_contig(const struct nilfs_bmap *btree,
624 __u64 key, __u64 *ptrp, unsigned maxblocks)
626 struct nilfs_btree_path *path;
627 struct nilfs_btree_node *node;
628 struct inode *dat = NULL;
631 int level = NILFS_BTREE_LEVEL_NODE_MIN;
632 int ret, cnt, index, maxlevel, ncmax;
633 struct nilfs_btree_readahead_info p;
635 path = nilfs_btree_alloc_path();
639 ret = nilfs_btree_do_lookup(btree, path, key, &ptr, level, 1);
643 if (NILFS_BMAP_USE_VBN(btree)) {
644 dat = nilfs_bmap_get_dat(btree);
645 ret = nilfs_dat_translate(dat, ptr, &blocknr);
651 if (cnt == maxblocks)
654 maxlevel = nilfs_btree_height(btree) - 1;
655 node = nilfs_btree_get_node(btree, path, level, &ncmax);
656 index = path[level].bp_index + 1;
658 while (index < nilfs_btree_node_get_nchildren(node)) {
659 if (nilfs_btree_node_get_key(node, index) !=
662 ptr2 = nilfs_btree_node_get_ptr(node, index, ncmax);
664 ret = nilfs_dat_translate(dat, ptr2, &blocknr);
669 if (ptr2 != ptr + cnt || ++cnt == maxblocks)
674 if (level == maxlevel)
677 /* look-up right sibling node */
678 p.node = nilfs_btree_get_node(btree, path, level + 1, &p.ncmax);
679 p.index = path[level + 1].bp_index + 1;
681 if (p.index >= nilfs_btree_node_get_nchildren(p.node) ||
682 nilfs_btree_node_get_key(p.node, p.index) != key + cnt)
684 ptr2 = nilfs_btree_node_get_ptr(p.node, p.index, p.ncmax);
685 path[level + 1].bp_index = p.index;
687 brelse(path[level].bp_bh);
688 path[level].bp_bh = NULL;
690 ret = __nilfs_btree_get_block(btree, ptr2, &path[level].bp_bh,
694 node = nilfs_btree_get_nonroot_node(path, level);
695 ncmax = nilfs_btree_nchildren_per_block(btree);
697 path[level].bp_index = index;
703 nilfs_btree_free_path(path);
707 static void nilfs_btree_promote_key(struct nilfs_bmap *btree,
708 struct nilfs_btree_path *path,
709 int level, __u64 key)
711 if (level < nilfs_btree_height(btree) - 1) {
713 nilfs_btree_node_set_key(
714 nilfs_btree_get_nonroot_node(path, level),
715 path[level].bp_index, key);
716 if (!buffer_dirty(path[level].bp_bh))
717 nilfs_btnode_mark_dirty(path[level].bp_bh);
718 } while ((path[level].bp_index == 0) &&
719 (++level < nilfs_btree_height(btree) - 1));
723 if (level == nilfs_btree_height(btree) - 1) {
724 nilfs_btree_node_set_key(nilfs_btree_get_root(btree),
725 path[level].bp_index, key);
729 static void nilfs_btree_do_insert(struct nilfs_bmap *btree,
730 struct nilfs_btree_path *path,
731 int level, __u64 *keyp, __u64 *ptrp)
733 struct nilfs_btree_node *node;
736 if (level < nilfs_btree_height(btree) - 1) {
737 node = nilfs_btree_get_nonroot_node(path, level);
738 ncblk = nilfs_btree_nchildren_per_block(btree);
739 nilfs_btree_node_insert(node, path[level].bp_index,
740 *keyp, *ptrp, ncblk);
741 if (!buffer_dirty(path[level].bp_bh))
742 nilfs_btnode_mark_dirty(path[level].bp_bh);
744 if (path[level].bp_index == 0)
745 nilfs_btree_promote_key(btree, path, level + 1,
746 nilfs_btree_node_get_key(node,
749 node = nilfs_btree_get_root(btree);
750 nilfs_btree_node_insert(node, path[level].bp_index,
752 NILFS_BTREE_ROOT_NCHILDREN_MAX);
756 static void nilfs_btree_carry_left(struct nilfs_bmap *btree,
757 struct nilfs_btree_path *path,
758 int level, __u64 *keyp, __u64 *ptrp)
760 struct nilfs_btree_node *node, *left;
761 int nchildren, lnchildren, n, move, ncblk;
763 node = nilfs_btree_get_nonroot_node(path, level);
764 left = nilfs_btree_get_sib_node(path, level);
765 nchildren = nilfs_btree_node_get_nchildren(node);
766 lnchildren = nilfs_btree_node_get_nchildren(left);
767 ncblk = nilfs_btree_nchildren_per_block(btree);
770 n = (nchildren + lnchildren + 1) / 2 - lnchildren;
771 if (n > path[level].bp_index) {
772 /* move insert point */
777 nilfs_btree_node_move_left(left, node, n, ncblk, ncblk);
779 if (!buffer_dirty(path[level].bp_bh))
780 nilfs_btnode_mark_dirty(path[level].bp_bh);
781 if (!buffer_dirty(path[level].bp_sib_bh))
782 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
784 nilfs_btree_promote_key(btree, path, level + 1,
785 nilfs_btree_node_get_key(node, 0));
788 brelse(path[level].bp_bh);
789 path[level].bp_bh = path[level].bp_sib_bh;
790 path[level].bp_sib_bh = NULL;
791 path[level].bp_index += lnchildren;
792 path[level + 1].bp_index--;
794 brelse(path[level].bp_sib_bh);
795 path[level].bp_sib_bh = NULL;
796 path[level].bp_index -= n;
799 nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
802 static void nilfs_btree_carry_right(struct nilfs_bmap *btree,
803 struct nilfs_btree_path *path,
804 int level, __u64 *keyp, __u64 *ptrp)
806 struct nilfs_btree_node *node, *right;
807 int nchildren, rnchildren, n, move, ncblk;
809 node = nilfs_btree_get_nonroot_node(path, level);
810 right = nilfs_btree_get_sib_node(path, level);
811 nchildren = nilfs_btree_node_get_nchildren(node);
812 rnchildren = nilfs_btree_node_get_nchildren(right);
813 ncblk = nilfs_btree_nchildren_per_block(btree);
816 n = (nchildren + rnchildren + 1) / 2 - rnchildren;
817 if (n > nchildren - path[level].bp_index) {
818 /* move insert point */
823 nilfs_btree_node_move_right(node, right, n, ncblk, ncblk);
825 if (!buffer_dirty(path[level].bp_bh))
826 nilfs_btnode_mark_dirty(path[level].bp_bh);
827 if (!buffer_dirty(path[level].bp_sib_bh))
828 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
830 path[level + 1].bp_index++;
831 nilfs_btree_promote_key(btree, path, level + 1,
832 nilfs_btree_node_get_key(right, 0));
833 path[level + 1].bp_index--;
836 brelse(path[level].bp_bh);
837 path[level].bp_bh = path[level].bp_sib_bh;
838 path[level].bp_sib_bh = NULL;
839 path[level].bp_index -= nilfs_btree_node_get_nchildren(node);
840 path[level + 1].bp_index++;
842 brelse(path[level].bp_sib_bh);
843 path[level].bp_sib_bh = NULL;
846 nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
849 static void nilfs_btree_split(struct nilfs_bmap *btree,
850 struct nilfs_btree_path *path,
851 int level, __u64 *keyp, __u64 *ptrp)
853 struct nilfs_btree_node *node, *right;
856 int nchildren, n, move, ncblk;
858 node = nilfs_btree_get_nonroot_node(path, level);
859 right = nilfs_btree_get_sib_node(path, level);
860 nchildren = nilfs_btree_node_get_nchildren(node);
861 ncblk = nilfs_btree_nchildren_per_block(btree);
864 n = (nchildren + 1) / 2;
865 if (n > nchildren - path[level].bp_index) {
870 nilfs_btree_node_move_right(node, right, n, ncblk, ncblk);
872 if (!buffer_dirty(path[level].bp_bh))
873 nilfs_btnode_mark_dirty(path[level].bp_bh);
874 if (!buffer_dirty(path[level].bp_sib_bh))
875 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
877 newkey = nilfs_btree_node_get_key(right, 0);
878 newptr = path[level].bp_newreq.bpr_ptr;
881 path[level].bp_index -= nilfs_btree_node_get_nchildren(node);
882 nilfs_btree_node_insert(right, path[level].bp_index,
883 *keyp, *ptrp, ncblk);
885 *keyp = nilfs_btree_node_get_key(right, 0);
886 *ptrp = path[level].bp_newreq.bpr_ptr;
888 brelse(path[level].bp_bh);
889 path[level].bp_bh = path[level].bp_sib_bh;
890 path[level].bp_sib_bh = NULL;
892 nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
894 *keyp = nilfs_btree_node_get_key(right, 0);
895 *ptrp = path[level].bp_newreq.bpr_ptr;
897 brelse(path[level].bp_sib_bh);
898 path[level].bp_sib_bh = NULL;
901 path[level + 1].bp_index++;
904 static void nilfs_btree_grow(struct nilfs_bmap *btree,
905 struct nilfs_btree_path *path,
906 int level, __u64 *keyp, __u64 *ptrp)
908 struct nilfs_btree_node *root, *child;
911 root = nilfs_btree_get_root(btree);
912 child = nilfs_btree_get_sib_node(path, level);
913 ncblk = nilfs_btree_nchildren_per_block(btree);
915 n = nilfs_btree_node_get_nchildren(root);
917 nilfs_btree_node_move_right(root, child, n,
918 NILFS_BTREE_ROOT_NCHILDREN_MAX, ncblk);
919 nilfs_btree_node_set_level(root, level + 1);
921 if (!buffer_dirty(path[level].bp_sib_bh))
922 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
924 path[level].bp_bh = path[level].bp_sib_bh;
925 path[level].bp_sib_bh = NULL;
927 nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
929 *keyp = nilfs_btree_node_get_key(child, 0);
930 *ptrp = path[level].bp_newreq.bpr_ptr;
933 static __u64 nilfs_btree_find_near(const struct nilfs_bmap *btree,
934 const struct nilfs_btree_path *path)
936 struct nilfs_btree_node *node;
940 return NILFS_BMAP_INVALID_PTR;
943 level = NILFS_BTREE_LEVEL_NODE_MIN;
944 if (path[level].bp_index > 0) {
945 node = nilfs_btree_get_node(btree, path, level, &ncmax);
946 return nilfs_btree_node_get_ptr(node,
947 path[level].bp_index - 1,
952 level = NILFS_BTREE_LEVEL_NODE_MIN + 1;
953 if (level <= nilfs_btree_height(btree) - 1) {
954 node = nilfs_btree_get_node(btree, path, level, &ncmax);
955 return nilfs_btree_node_get_ptr(node, path[level].bp_index,
959 return NILFS_BMAP_INVALID_PTR;
962 static __u64 nilfs_btree_find_target_v(const struct nilfs_bmap *btree,
963 const struct nilfs_btree_path *path,
968 ptr = nilfs_bmap_find_target_seq(btree, key);
969 if (ptr != NILFS_BMAP_INVALID_PTR)
970 /* sequential access */
973 ptr = nilfs_btree_find_near(btree, path);
974 if (ptr != NILFS_BMAP_INVALID_PTR)
979 return nilfs_bmap_find_target_in_group(btree);
982 static int nilfs_btree_prepare_insert(struct nilfs_bmap *btree,
983 struct nilfs_btree_path *path,
984 int *levelp, __u64 key, __u64 ptr,
985 struct nilfs_bmap_stats *stats)
987 struct buffer_head *bh;
988 struct nilfs_btree_node *node, *parent, *sib;
990 int pindex, level, ncmax, ncblk, ret;
991 struct inode *dat = NULL;
993 stats->bs_nblocks = 0;
994 level = NILFS_BTREE_LEVEL_DATA;
996 /* allocate a new ptr for data block */
997 if (NILFS_BMAP_USE_VBN(btree)) {
998 path[level].bp_newreq.bpr_ptr =
999 nilfs_btree_find_target_v(btree, path, key);
1000 dat = nilfs_bmap_get_dat(btree);
1003 ret = nilfs_bmap_prepare_alloc_ptr(btree, &path[level].bp_newreq, dat);
1007 ncblk = nilfs_btree_nchildren_per_block(btree);
1009 for (level = NILFS_BTREE_LEVEL_NODE_MIN;
1010 level < nilfs_btree_height(btree) - 1;
1012 node = nilfs_btree_get_nonroot_node(path, level);
1013 if (nilfs_btree_node_get_nchildren(node) < ncblk) {
1014 path[level].bp_op = nilfs_btree_do_insert;
1015 stats->bs_nblocks++;
1019 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
1020 pindex = path[level + 1].bp_index;
1024 sibptr = nilfs_btree_node_get_ptr(parent, pindex - 1,
1026 ret = nilfs_btree_get_block(btree, sibptr, &bh);
1028 goto err_out_child_node;
1029 sib = (struct nilfs_btree_node *)bh->b_data;
1030 if (nilfs_btree_node_get_nchildren(sib) < ncblk) {
1031 path[level].bp_sib_bh = bh;
1032 path[level].bp_op = nilfs_btree_carry_left;
1033 stats->bs_nblocks++;
1041 if (pindex < nilfs_btree_node_get_nchildren(parent) - 1) {
1042 sibptr = nilfs_btree_node_get_ptr(parent, pindex + 1,
1044 ret = nilfs_btree_get_block(btree, sibptr, &bh);
1046 goto err_out_child_node;
1047 sib = (struct nilfs_btree_node *)bh->b_data;
1048 if (nilfs_btree_node_get_nchildren(sib) < ncblk) {
1049 path[level].bp_sib_bh = bh;
1050 path[level].bp_op = nilfs_btree_carry_right;
1051 stats->bs_nblocks++;
1059 path[level].bp_newreq.bpr_ptr =
1060 path[level - 1].bp_newreq.bpr_ptr + 1;
1061 ret = nilfs_bmap_prepare_alloc_ptr(btree,
1062 &path[level].bp_newreq, dat);
1064 goto err_out_child_node;
1065 ret = nilfs_btree_get_new_block(btree,
1066 path[level].bp_newreq.bpr_ptr,
1069 goto err_out_curr_node;
1071 stats->bs_nblocks++;
1073 sib = (struct nilfs_btree_node *)bh->b_data;
1074 nilfs_btree_node_init(sib, 0, level, 0, ncblk, NULL, NULL);
1075 path[level].bp_sib_bh = bh;
1076 path[level].bp_op = nilfs_btree_split;
1080 node = nilfs_btree_get_root(btree);
1081 if (nilfs_btree_node_get_nchildren(node) <
1082 NILFS_BTREE_ROOT_NCHILDREN_MAX) {
1083 path[level].bp_op = nilfs_btree_do_insert;
1084 stats->bs_nblocks++;
1089 path[level].bp_newreq.bpr_ptr = path[level - 1].bp_newreq.bpr_ptr + 1;
1090 ret = nilfs_bmap_prepare_alloc_ptr(btree, &path[level].bp_newreq, dat);
1092 goto err_out_child_node;
1093 ret = nilfs_btree_get_new_block(btree, path[level].bp_newreq.bpr_ptr,
1096 goto err_out_curr_node;
1098 nilfs_btree_node_init((struct nilfs_btree_node *)bh->b_data,
1099 0, level, 0, ncblk, NULL, NULL);
1100 path[level].bp_sib_bh = bh;
1101 path[level].bp_op = nilfs_btree_grow;
1104 path[level].bp_op = nilfs_btree_do_insert;
1106 /* a newly-created node block and a data block are added */
1107 stats->bs_nblocks += 2;
1116 nilfs_bmap_abort_alloc_ptr(btree, &path[level].bp_newreq, dat);
1118 for (level--; level > NILFS_BTREE_LEVEL_DATA; level--) {
1119 nilfs_btnode_delete(path[level].bp_sib_bh);
1120 nilfs_bmap_abort_alloc_ptr(btree, &path[level].bp_newreq, dat);
1124 nilfs_bmap_abort_alloc_ptr(btree, &path[level].bp_newreq, dat);
1127 stats->bs_nblocks = 0;
1131 static void nilfs_btree_commit_insert(struct nilfs_bmap *btree,
1132 struct nilfs_btree_path *path,
1133 int maxlevel, __u64 key, __u64 ptr)
1135 struct inode *dat = NULL;
1138 set_buffer_nilfs_volatile((struct buffer_head *)((unsigned long)ptr));
1139 ptr = path[NILFS_BTREE_LEVEL_DATA].bp_newreq.bpr_ptr;
1140 if (NILFS_BMAP_USE_VBN(btree)) {
1141 nilfs_bmap_set_target_v(btree, key, ptr);
1142 dat = nilfs_bmap_get_dat(btree);
1145 for (level = NILFS_BTREE_LEVEL_NODE_MIN; level <= maxlevel; level++) {
1146 nilfs_bmap_commit_alloc_ptr(btree,
1147 &path[level - 1].bp_newreq, dat);
1148 path[level].bp_op(btree, path, level, &key, &ptr);
1151 if (!nilfs_bmap_dirty(btree))
1152 nilfs_bmap_set_dirty(btree);
1155 static int nilfs_btree_insert(struct nilfs_bmap *btree, __u64 key, __u64 ptr)
1157 struct nilfs_btree_path *path;
1158 struct nilfs_bmap_stats stats;
1161 path = nilfs_btree_alloc_path();
1165 ret = nilfs_btree_do_lookup(btree, path, key, NULL,
1166 NILFS_BTREE_LEVEL_NODE_MIN, 0);
1167 if (ret != -ENOENT) {
1173 ret = nilfs_btree_prepare_insert(btree, path, &level, key, ptr, &stats);
1176 nilfs_btree_commit_insert(btree, path, level, key, ptr);
1177 nilfs_bmap_add_blocks(btree, stats.bs_nblocks);
1180 nilfs_btree_free_path(path);
1184 static void nilfs_btree_do_delete(struct nilfs_bmap *btree,
1185 struct nilfs_btree_path *path,
1186 int level, __u64 *keyp, __u64 *ptrp)
1188 struct nilfs_btree_node *node;
1191 if (level < nilfs_btree_height(btree) - 1) {
1192 node = nilfs_btree_get_nonroot_node(path, level);
1193 ncblk = nilfs_btree_nchildren_per_block(btree);
1194 nilfs_btree_node_delete(node, path[level].bp_index,
1196 if (!buffer_dirty(path[level].bp_bh))
1197 nilfs_btnode_mark_dirty(path[level].bp_bh);
1198 if (path[level].bp_index == 0)
1199 nilfs_btree_promote_key(btree, path, level + 1,
1200 nilfs_btree_node_get_key(node, 0));
1202 node = nilfs_btree_get_root(btree);
1203 nilfs_btree_node_delete(node, path[level].bp_index,
1205 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1209 static void nilfs_btree_borrow_left(struct nilfs_bmap *btree,
1210 struct nilfs_btree_path *path,
1211 int level, __u64 *keyp, __u64 *ptrp)
1213 struct nilfs_btree_node *node, *left;
1214 int nchildren, lnchildren, n, ncblk;
1216 nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
1218 node = nilfs_btree_get_nonroot_node(path, level);
1219 left = nilfs_btree_get_sib_node(path, level);
1220 nchildren = nilfs_btree_node_get_nchildren(node);
1221 lnchildren = nilfs_btree_node_get_nchildren(left);
1222 ncblk = nilfs_btree_nchildren_per_block(btree);
1224 n = (nchildren + lnchildren) / 2 - nchildren;
1226 nilfs_btree_node_move_right(left, node, n, ncblk, ncblk);
1228 if (!buffer_dirty(path[level].bp_bh))
1229 nilfs_btnode_mark_dirty(path[level].bp_bh);
1230 if (!buffer_dirty(path[level].bp_sib_bh))
1231 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
1233 nilfs_btree_promote_key(btree, path, level + 1,
1234 nilfs_btree_node_get_key(node, 0));
1236 brelse(path[level].bp_sib_bh);
1237 path[level].bp_sib_bh = NULL;
1238 path[level].bp_index += n;
1241 static void nilfs_btree_borrow_right(struct nilfs_bmap *btree,
1242 struct nilfs_btree_path *path,
1243 int level, __u64 *keyp, __u64 *ptrp)
1245 struct nilfs_btree_node *node, *right;
1246 int nchildren, rnchildren, n, ncblk;
1248 nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
1250 node = nilfs_btree_get_nonroot_node(path, level);
1251 right = nilfs_btree_get_sib_node(path, level);
1252 nchildren = nilfs_btree_node_get_nchildren(node);
1253 rnchildren = nilfs_btree_node_get_nchildren(right);
1254 ncblk = nilfs_btree_nchildren_per_block(btree);
1256 n = (nchildren + rnchildren) / 2 - nchildren;
1258 nilfs_btree_node_move_left(node, right, n, ncblk, ncblk);
1260 if (!buffer_dirty(path[level].bp_bh))
1261 nilfs_btnode_mark_dirty(path[level].bp_bh);
1262 if (!buffer_dirty(path[level].bp_sib_bh))
1263 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
1265 path[level + 1].bp_index++;
1266 nilfs_btree_promote_key(btree, path, level + 1,
1267 nilfs_btree_node_get_key(right, 0));
1268 path[level + 1].bp_index--;
1270 brelse(path[level].bp_sib_bh);
1271 path[level].bp_sib_bh = NULL;
1274 static void nilfs_btree_concat_left(struct nilfs_bmap *btree,
1275 struct nilfs_btree_path *path,
1276 int level, __u64 *keyp, __u64 *ptrp)
1278 struct nilfs_btree_node *node, *left;
1281 nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
1283 node = nilfs_btree_get_nonroot_node(path, level);
1284 left = nilfs_btree_get_sib_node(path, level);
1285 ncblk = nilfs_btree_nchildren_per_block(btree);
1287 n = nilfs_btree_node_get_nchildren(node);
1289 nilfs_btree_node_move_left(left, node, n, ncblk, ncblk);
1291 if (!buffer_dirty(path[level].bp_sib_bh))
1292 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
1294 nilfs_btnode_delete(path[level].bp_bh);
1295 path[level].bp_bh = path[level].bp_sib_bh;
1296 path[level].bp_sib_bh = NULL;
1297 path[level].bp_index += nilfs_btree_node_get_nchildren(left);
1300 static void nilfs_btree_concat_right(struct nilfs_bmap *btree,
1301 struct nilfs_btree_path *path,
1302 int level, __u64 *keyp, __u64 *ptrp)
1304 struct nilfs_btree_node *node, *right;
1307 nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
1309 node = nilfs_btree_get_nonroot_node(path, level);
1310 right = nilfs_btree_get_sib_node(path, level);
1311 ncblk = nilfs_btree_nchildren_per_block(btree);
1313 n = nilfs_btree_node_get_nchildren(right);
1315 nilfs_btree_node_move_left(node, right, n, ncblk, ncblk);
1317 if (!buffer_dirty(path[level].bp_bh))
1318 nilfs_btnode_mark_dirty(path[level].bp_bh);
1320 nilfs_btnode_delete(path[level].bp_sib_bh);
1321 path[level].bp_sib_bh = NULL;
1322 path[level + 1].bp_index++;
1325 static void nilfs_btree_shrink(struct nilfs_bmap *btree,
1326 struct nilfs_btree_path *path,
1327 int level, __u64 *keyp, __u64 *ptrp)
1329 struct nilfs_btree_node *root, *child;
1332 nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
1334 root = nilfs_btree_get_root(btree);
1335 child = nilfs_btree_get_nonroot_node(path, level);
1336 ncblk = nilfs_btree_nchildren_per_block(btree);
1338 nilfs_btree_node_delete(root, 0, NULL, NULL,
1339 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1340 nilfs_btree_node_set_level(root, level);
1341 n = nilfs_btree_node_get_nchildren(child);
1342 nilfs_btree_node_move_left(root, child, n,
1343 NILFS_BTREE_ROOT_NCHILDREN_MAX, ncblk);
1345 nilfs_btnode_delete(path[level].bp_bh);
1346 path[level].bp_bh = NULL;
1350 static int nilfs_btree_prepare_delete(struct nilfs_bmap *btree,
1351 struct nilfs_btree_path *path,
1353 struct nilfs_bmap_stats *stats,
1356 struct buffer_head *bh;
1357 struct nilfs_btree_node *node, *parent, *sib;
1359 int pindex, level, ncmin, ncmax, ncblk, ret;
1362 stats->bs_nblocks = 0;
1363 ncmin = NILFS_BTREE_NODE_NCHILDREN_MIN(nilfs_btree_node_size(btree));
1364 ncblk = nilfs_btree_nchildren_per_block(btree);
1366 for (level = NILFS_BTREE_LEVEL_NODE_MIN;
1367 level < nilfs_btree_height(btree) - 1;
1369 node = nilfs_btree_get_nonroot_node(path, level);
1370 path[level].bp_oldreq.bpr_ptr =
1371 nilfs_btree_node_get_ptr(node, path[level].bp_index,
1373 ret = nilfs_bmap_prepare_end_ptr(btree,
1374 &path[level].bp_oldreq, dat);
1376 goto err_out_child_node;
1378 if (nilfs_btree_node_get_nchildren(node) > ncmin) {
1379 path[level].bp_op = nilfs_btree_do_delete;
1380 stats->bs_nblocks++;
1384 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
1385 pindex = path[level + 1].bp_index;
1389 sibptr = nilfs_btree_node_get_ptr(parent, pindex - 1,
1391 ret = nilfs_btree_get_block(btree, sibptr, &bh);
1393 goto err_out_curr_node;
1394 sib = (struct nilfs_btree_node *)bh->b_data;
1395 if (nilfs_btree_node_get_nchildren(sib) > ncmin) {
1396 path[level].bp_sib_bh = bh;
1397 path[level].bp_op = nilfs_btree_borrow_left;
1398 stats->bs_nblocks++;
1401 path[level].bp_sib_bh = bh;
1402 path[level].bp_op = nilfs_btree_concat_left;
1403 stats->bs_nblocks++;
1407 nilfs_btree_node_get_nchildren(parent) - 1) {
1409 sibptr = nilfs_btree_node_get_ptr(parent, pindex + 1,
1411 ret = nilfs_btree_get_block(btree, sibptr, &bh);
1413 goto err_out_curr_node;
1414 sib = (struct nilfs_btree_node *)bh->b_data;
1415 if (nilfs_btree_node_get_nchildren(sib) > ncmin) {
1416 path[level].bp_sib_bh = bh;
1417 path[level].bp_op = nilfs_btree_borrow_right;
1418 stats->bs_nblocks++;
1421 path[level].bp_sib_bh = bh;
1422 path[level].bp_op = nilfs_btree_concat_right;
1423 stats->bs_nblocks++;
1428 /* the only child of the root node */
1429 WARN_ON(level != nilfs_btree_height(btree) - 2);
1430 if (nilfs_btree_node_get_nchildren(node) - 1 <=
1431 NILFS_BTREE_ROOT_NCHILDREN_MAX) {
1432 path[level].bp_op = nilfs_btree_shrink;
1433 stats->bs_nblocks += 2;
1435 path[level].bp_op = nilfs_btree_do_delete;
1436 stats->bs_nblocks++;
1444 node = nilfs_btree_get_root(btree);
1445 path[level].bp_oldreq.bpr_ptr =
1446 nilfs_btree_node_get_ptr(node, path[level].bp_index,
1447 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1449 ret = nilfs_bmap_prepare_end_ptr(btree, &path[level].bp_oldreq, dat);
1451 goto err_out_child_node;
1453 /* child of the root node is deleted */
1454 path[level].bp_op = nilfs_btree_do_delete;
1455 stats->bs_nblocks++;
1464 nilfs_bmap_abort_end_ptr(btree, &path[level].bp_oldreq, dat);
1466 for (level--; level >= NILFS_BTREE_LEVEL_NODE_MIN; level--) {
1467 brelse(path[level].bp_sib_bh);
1468 nilfs_bmap_abort_end_ptr(btree, &path[level].bp_oldreq, dat);
1471 stats->bs_nblocks = 0;
1475 static void nilfs_btree_commit_delete(struct nilfs_bmap *btree,
1476 struct nilfs_btree_path *path,
1477 int maxlevel, struct inode *dat)
1481 for (level = NILFS_BTREE_LEVEL_NODE_MIN; level <= maxlevel; level++) {
1482 nilfs_bmap_commit_end_ptr(btree, &path[level].bp_oldreq, dat);
1483 path[level].bp_op(btree, path, level, NULL, NULL);
1486 if (!nilfs_bmap_dirty(btree))
1487 nilfs_bmap_set_dirty(btree);
1490 static int nilfs_btree_delete(struct nilfs_bmap *btree, __u64 key)
1493 struct nilfs_btree_path *path;
1494 struct nilfs_bmap_stats stats;
1498 path = nilfs_btree_alloc_path();
1502 ret = nilfs_btree_do_lookup(btree, path, key, NULL,
1503 NILFS_BTREE_LEVEL_NODE_MIN, 0);
1508 dat = NILFS_BMAP_USE_VBN(btree) ? nilfs_bmap_get_dat(btree) : NULL;
1510 ret = nilfs_btree_prepare_delete(btree, path, &level, &stats, dat);
1513 nilfs_btree_commit_delete(btree, path, level, dat);
1514 nilfs_bmap_sub_blocks(btree, stats.bs_nblocks);
1517 nilfs_btree_free_path(path);
1521 static int nilfs_btree_last_key(const struct nilfs_bmap *btree, __u64 *keyp)
1523 struct nilfs_btree_path *path;
1526 path = nilfs_btree_alloc_path();
1530 ret = nilfs_btree_do_lookup_last(btree, path, keyp, NULL);
1532 nilfs_btree_free_path(path);
1537 static int nilfs_btree_check_delete(struct nilfs_bmap *btree, __u64 key)
1539 struct buffer_head *bh;
1540 struct nilfs_btree_node *root, *node;
1541 __u64 maxkey, nextmaxkey;
1545 root = nilfs_btree_get_root(btree);
1546 switch (nilfs_btree_height(btree)) {
1552 nchildren = nilfs_btree_node_get_nchildren(root);
1555 ptr = nilfs_btree_node_get_ptr(root, nchildren - 1,
1556 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1557 ret = nilfs_btree_get_block(btree, ptr, &bh);
1560 node = (struct nilfs_btree_node *)bh->b_data;
1566 nchildren = nilfs_btree_node_get_nchildren(node);
1567 maxkey = nilfs_btree_node_get_key(node, nchildren - 1);
1568 nextmaxkey = (nchildren > 1) ?
1569 nilfs_btree_node_get_key(node, nchildren - 2) : 0;
1573 return (maxkey == key) && (nextmaxkey < NILFS_BMAP_LARGE_LOW);
1576 static int nilfs_btree_gather_data(struct nilfs_bmap *btree,
1577 __u64 *keys, __u64 *ptrs, int nitems)
1579 struct buffer_head *bh;
1580 struct nilfs_btree_node *node, *root;
1584 int nchildren, ncmax, i, ret;
1586 root = nilfs_btree_get_root(btree);
1587 switch (nilfs_btree_height(btree)) {
1591 ncmax = NILFS_BTREE_ROOT_NCHILDREN_MAX;
1594 nchildren = nilfs_btree_node_get_nchildren(root);
1595 WARN_ON(nchildren > 1);
1596 ptr = nilfs_btree_node_get_ptr(root, nchildren - 1,
1597 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1598 ret = nilfs_btree_get_block(btree, ptr, &bh);
1601 node = (struct nilfs_btree_node *)bh->b_data;
1602 ncmax = nilfs_btree_nchildren_per_block(btree);
1609 nchildren = nilfs_btree_node_get_nchildren(node);
1610 if (nchildren < nitems)
1612 dkeys = nilfs_btree_node_dkeys(node);
1613 dptrs = nilfs_btree_node_dptrs(node, ncmax);
1614 for (i = 0; i < nitems; i++) {
1615 keys[i] = le64_to_cpu(dkeys[i]);
1616 ptrs[i] = le64_to_cpu(dptrs[i]);
1626 nilfs_btree_prepare_convert_and_insert(struct nilfs_bmap *btree, __u64 key,
1627 union nilfs_bmap_ptr_req *dreq,
1628 union nilfs_bmap_ptr_req *nreq,
1629 struct buffer_head **bhp,
1630 struct nilfs_bmap_stats *stats)
1632 struct buffer_head *bh;
1633 struct inode *dat = NULL;
1636 stats->bs_nblocks = 0;
1639 /* cannot find near ptr */
1640 if (NILFS_BMAP_USE_VBN(btree)) {
1641 dreq->bpr_ptr = nilfs_btree_find_target_v(btree, NULL, key);
1642 dat = nilfs_bmap_get_dat(btree);
1645 ret = nilfs_bmap_prepare_alloc_ptr(btree, dreq, dat);
1650 stats->bs_nblocks++;
1652 nreq->bpr_ptr = dreq->bpr_ptr + 1;
1653 ret = nilfs_bmap_prepare_alloc_ptr(btree, nreq, dat);
1657 ret = nilfs_btree_get_new_block(btree, nreq->bpr_ptr, &bh);
1662 stats->bs_nblocks++;
1670 nilfs_bmap_abort_alloc_ptr(btree, nreq, dat);
1672 nilfs_bmap_abort_alloc_ptr(btree, dreq, dat);
1673 stats->bs_nblocks = 0;
1679 nilfs_btree_commit_convert_and_insert(struct nilfs_bmap *btree,
1680 __u64 key, __u64 ptr,
1681 const __u64 *keys, const __u64 *ptrs,
1683 union nilfs_bmap_ptr_req *dreq,
1684 union nilfs_bmap_ptr_req *nreq,
1685 struct buffer_head *bh)
1687 struct nilfs_btree_node *node;
1692 /* free resources */
1693 if (btree->b_ops->bop_clear != NULL)
1694 btree->b_ops->bop_clear(btree);
1696 /* ptr must be a pointer to a buffer head. */
1697 set_buffer_nilfs_volatile((struct buffer_head *)((unsigned long)ptr));
1699 /* convert and insert */
1700 dat = NILFS_BMAP_USE_VBN(btree) ? nilfs_bmap_get_dat(btree) : NULL;
1701 nilfs_btree_init(btree);
1703 nilfs_bmap_commit_alloc_ptr(btree, dreq, dat);
1704 nilfs_bmap_commit_alloc_ptr(btree, nreq, dat);
1706 /* create child node at level 1 */
1707 node = (struct nilfs_btree_node *)bh->b_data;
1708 ncblk = nilfs_btree_nchildren_per_block(btree);
1709 nilfs_btree_node_init(node, 0, 1, n, ncblk, keys, ptrs);
1710 nilfs_btree_node_insert(node, n, key, dreq->bpr_ptr, ncblk);
1711 if (!buffer_dirty(bh))
1712 nilfs_btnode_mark_dirty(bh);
1713 if (!nilfs_bmap_dirty(btree))
1714 nilfs_bmap_set_dirty(btree);
1718 /* create root node at level 2 */
1719 node = nilfs_btree_get_root(btree);
1720 tmpptr = nreq->bpr_ptr;
1721 nilfs_btree_node_init(node, NILFS_BTREE_NODE_ROOT, 2, 1,
1722 NILFS_BTREE_ROOT_NCHILDREN_MAX,
1725 nilfs_bmap_commit_alloc_ptr(btree, dreq, dat);
1727 /* create root node at level 1 */
1728 node = nilfs_btree_get_root(btree);
1729 nilfs_btree_node_init(node, NILFS_BTREE_NODE_ROOT, 1, n,
1730 NILFS_BTREE_ROOT_NCHILDREN_MAX,
1732 nilfs_btree_node_insert(node, n, key, dreq->bpr_ptr,
1733 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1734 if (!nilfs_bmap_dirty(btree))
1735 nilfs_bmap_set_dirty(btree);
1738 if (NILFS_BMAP_USE_VBN(btree))
1739 nilfs_bmap_set_target_v(btree, key, dreq->bpr_ptr);
1743 * nilfs_btree_convert_and_insert -
1751 int nilfs_btree_convert_and_insert(struct nilfs_bmap *btree,
1752 __u64 key, __u64 ptr,
1753 const __u64 *keys, const __u64 *ptrs, int n)
1755 struct buffer_head *bh;
1756 union nilfs_bmap_ptr_req dreq, nreq, *di, *ni;
1757 struct nilfs_bmap_stats stats;
1760 if (n + 1 <= NILFS_BTREE_ROOT_NCHILDREN_MAX) {
1763 } else if ((n + 1) <= NILFS_BTREE_NODE_NCHILDREN_MAX(
1764 1 << btree->b_inode->i_blkbits)) {
1773 ret = nilfs_btree_prepare_convert_and_insert(btree, key, di, ni, &bh,
1777 nilfs_btree_commit_convert_and_insert(btree, key, ptr, keys, ptrs, n,
1779 nilfs_bmap_add_blocks(btree, stats.bs_nblocks);
1783 static int nilfs_btree_propagate_p(struct nilfs_bmap *btree,
1784 struct nilfs_btree_path *path,
1786 struct buffer_head *bh)
1788 while ((++level < nilfs_btree_height(btree) - 1) &&
1789 !buffer_dirty(path[level].bp_bh))
1790 nilfs_btnode_mark_dirty(path[level].bp_bh);
1795 static int nilfs_btree_prepare_update_v(struct nilfs_bmap *btree,
1796 struct nilfs_btree_path *path,
1797 int level, struct inode *dat)
1799 struct nilfs_btree_node *parent;
1802 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
1803 path[level].bp_oldreq.bpr_ptr =
1804 nilfs_btree_node_get_ptr(parent, path[level + 1].bp_index,
1806 path[level].bp_newreq.bpr_ptr = path[level].bp_oldreq.bpr_ptr + 1;
1807 ret = nilfs_dat_prepare_update(dat, &path[level].bp_oldreq.bpr_req,
1808 &path[level].bp_newreq.bpr_req);
1812 if (buffer_nilfs_node(path[level].bp_bh)) {
1813 path[level].bp_ctxt.oldkey = path[level].bp_oldreq.bpr_ptr;
1814 path[level].bp_ctxt.newkey = path[level].bp_newreq.bpr_ptr;
1815 path[level].bp_ctxt.bh = path[level].bp_bh;
1816 ret = nilfs_btnode_prepare_change_key(
1817 &NILFS_BMAP_I(btree)->i_btnode_cache,
1818 &path[level].bp_ctxt);
1820 nilfs_dat_abort_update(dat,
1821 &path[level].bp_oldreq.bpr_req,
1822 &path[level].bp_newreq.bpr_req);
1830 static void nilfs_btree_commit_update_v(struct nilfs_bmap *btree,
1831 struct nilfs_btree_path *path,
1832 int level, struct inode *dat)
1834 struct nilfs_btree_node *parent;
1837 nilfs_dat_commit_update(dat, &path[level].bp_oldreq.bpr_req,
1838 &path[level].bp_newreq.bpr_req,
1839 btree->b_ptr_type == NILFS_BMAP_PTR_VS);
1841 if (buffer_nilfs_node(path[level].bp_bh)) {
1842 nilfs_btnode_commit_change_key(
1843 &NILFS_BMAP_I(btree)->i_btnode_cache,
1844 &path[level].bp_ctxt);
1845 path[level].bp_bh = path[level].bp_ctxt.bh;
1847 set_buffer_nilfs_volatile(path[level].bp_bh);
1849 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
1850 nilfs_btree_node_set_ptr(parent, path[level + 1].bp_index,
1851 path[level].bp_newreq.bpr_ptr, ncmax);
1854 static void nilfs_btree_abort_update_v(struct nilfs_bmap *btree,
1855 struct nilfs_btree_path *path,
1856 int level, struct inode *dat)
1858 nilfs_dat_abort_update(dat, &path[level].bp_oldreq.bpr_req,
1859 &path[level].bp_newreq.bpr_req);
1860 if (buffer_nilfs_node(path[level].bp_bh))
1861 nilfs_btnode_abort_change_key(
1862 &NILFS_BMAP_I(btree)->i_btnode_cache,
1863 &path[level].bp_ctxt);
1866 static int nilfs_btree_prepare_propagate_v(struct nilfs_bmap *btree,
1867 struct nilfs_btree_path *path,
1868 int minlevel, int *maxlevelp,
1874 if (!buffer_nilfs_volatile(path[level].bp_bh)) {
1875 ret = nilfs_btree_prepare_update_v(btree, path, level, dat);
1879 while ((++level < nilfs_btree_height(btree) - 1) &&
1880 !buffer_dirty(path[level].bp_bh)) {
1882 WARN_ON(buffer_nilfs_volatile(path[level].bp_bh));
1883 ret = nilfs_btree_prepare_update_v(btree, path, level, dat);
1889 *maxlevelp = level - 1;
1894 while (--level > minlevel)
1895 nilfs_btree_abort_update_v(btree, path, level, dat);
1896 if (!buffer_nilfs_volatile(path[level].bp_bh))
1897 nilfs_btree_abort_update_v(btree, path, level, dat);
1901 static void nilfs_btree_commit_propagate_v(struct nilfs_bmap *btree,
1902 struct nilfs_btree_path *path,
1903 int minlevel, int maxlevel,
1904 struct buffer_head *bh,
1909 if (!buffer_nilfs_volatile(path[minlevel].bp_bh))
1910 nilfs_btree_commit_update_v(btree, path, minlevel, dat);
1912 for (level = minlevel + 1; level <= maxlevel; level++)
1913 nilfs_btree_commit_update_v(btree, path, level, dat);
1916 static int nilfs_btree_propagate_v(struct nilfs_bmap *btree,
1917 struct nilfs_btree_path *path,
1918 int level, struct buffer_head *bh)
1920 int maxlevel = 0, ret;
1921 struct nilfs_btree_node *parent;
1922 struct inode *dat = nilfs_bmap_get_dat(btree);
1927 path[level].bp_bh = bh;
1928 ret = nilfs_btree_prepare_propagate_v(btree, path, level, &maxlevel,
1933 if (buffer_nilfs_volatile(path[level].bp_bh)) {
1934 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
1935 ptr = nilfs_btree_node_get_ptr(parent,
1936 path[level + 1].bp_index,
1938 ret = nilfs_dat_mark_dirty(dat, ptr);
1943 nilfs_btree_commit_propagate_v(btree, path, level, maxlevel, bh, dat);
1946 brelse(path[level].bp_bh);
1947 path[level].bp_bh = NULL;
1951 static int nilfs_btree_propagate(struct nilfs_bmap *btree,
1952 struct buffer_head *bh)
1954 struct nilfs_btree_path *path;
1955 struct nilfs_btree_node *node;
1959 WARN_ON(!buffer_dirty(bh));
1961 path = nilfs_btree_alloc_path();
1965 if (buffer_nilfs_node(bh)) {
1966 node = (struct nilfs_btree_node *)bh->b_data;
1967 key = nilfs_btree_node_get_key(node, 0);
1968 level = nilfs_btree_node_get_level(node);
1970 key = nilfs_bmap_data_get_key(btree, bh);
1971 level = NILFS_BTREE_LEVEL_DATA;
1974 ret = nilfs_btree_do_lookup(btree, path, key, NULL, level + 1, 0);
1976 if (unlikely(ret == -ENOENT))
1977 printk(KERN_CRIT "%s: key = %llu, level == %d\n",
1978 __func__, (unsigned long long)key, level);
1982 ret = NILFS_BMAP_USE_VBN(btree) ?
1983 nilfs_btree_propagate_v(btree, path, level, bh) :
1984 nilfs_btree_propagate_p(btree, path, level, bh);
1987 nilfs_btree_free_path(path);
1992 static int nilfs_btree_propagate_gc(struct nilfs_bmap *btree,
1993 struct buffer_head *bh)
1995 return nilfs_dat_mark_dirty(nilfs_bmap_get_dat(btree), bh->b_blocknr);
1998 static void nilfs_btree_add_dirty_buffer(struct nilfs_bmap *btree,
1999 struct list_head *lists,
2000 struct buffer_head *bh)
2002 struct list_head *head;
2003 struct buffer_head *cbh;
2004 struct nilfs_btree_node *node, *cnode;
2009 node = (struct nilfs_btree_node *)bh->b_data;
2010 key = nilfs_btree_node_get_key(node, 0);
2011 level = nilfs_btree_node_get_level(node);
2012 if (level < NILFS_BTREE_LEVEL_NODE_MIN ||
2013 level >= NILFS_BTREE_LEVEL_MAX) {
2016 "%s: invalid btree level: %d (key=%llu, ino=%lu, "
2018 __func__, level, (unsigned long long)key,
2019 NILFS_BMAP_I(btree)->vfs_inode.i_ino,
2020 (unsigned long long)bh->b_blocknr);
2024 list_for_each(head, &lists[level]) {
2025 cbh = list_entry(head, struct buffer_head, b_assoc_buffers);
2026 cnode = (struct nilfs_btree_node *)cbh->b_data;
2027 ckey = nilfs_btree_node_get_key(cnode, 0);
2031 list_add_tail(&bh->b_assoc_buffers, head);
2034 static void nilfs_btree_lookup_dirty_buffers(struct nilfs_bmap *btree,
2035 struct list_head *listp)
2037 struct address_space *btcache = &NILFS_BMAP_I(btree)->i_btnode_cache;
2038 struct list_head lists[NILFS_BTREE_LEVEL_MAX];
2039 struct pagevec pvec;
2040 struct buffer_head *bh, *head;
2044 for (level = NILFS_BTREE_LEVEL_NODE_MIN;
2045 level < NILFS_BTREE_LEVEL_MAX;
2047 INIT_LIST_HEAD(&lists[level]);
2049 pagevec_init(&pvec, 0);
2051 while (pagevec_lookup_tag(&pvec, btcache, &index, PAGECACHE_TAG_DIRTY,
2053 for (i = 0; i < pagevec_count(&pvec); i++) {
2054 bh = head = page_buffers(pvec.pages[i]);
2056 if (buffer_dirty(bh))
2057 nilfs_btree_add_dirty_buffer(btree,
2059 } while ((bh = bh->b_this_page) != head);
2061 pagevec_release(&pvec);
2065 for (level = NILFS_BTREE_LEVEL_NODE_MIN;
2066 level < NILFS_BTREE_LEVEL_MAX;
2068 list_splice_tail(&lists[level], listp);
2071 static int nilfs_btree_assign_p(struct nilfs_bmap *btree,
2072 struct nilfs_btree_path *path,
2074 struct buffer_head **bh,
2076 union nilfs_binfo *binfo)
2078 struct nilfs_btree_node *parent;
2083 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
2084 ptr = nilfs_btree_node_get_ptr(parent, path[level + 1].bp_index,
2086 if (buffer_nilfs_node(*bh)) {
2087 path[level].bp_ctxt.oldkey = ptr;
2088 path[level].bp_ctxt.newkey = blocknr;
2089 path[level].bp_ctxt.bh = *bh;
2090 ret = nilfs_btnode_prepare_change_key(
2091 &NILFS_BMAP_I(btree)->i_btnode_cache,
2092 &path[level].bp_ctxt);
2095 nilfs_btnode_commit_change_key(
2096 &NILFS_BMAP_I(btree)->i_btnode_cache,
2097 &path[level].bp_ctxt);
2098 *bh = path[level].bp_ctxt.bh;
2101 nilfs_btree_node_set_ptr(parent, path[level + 1].bp_index, blocknr,
2104 key = nilfs_btree_node_get_key(parent, path[level + 1].bp_index);
2105 /* on-disk format */
2106 binfo->bi_dat.bi_blkoff = cpu_to_le64(key);
2107 binfo->bi_dat.bi_level = level;
2112 static int nilfs_btree_assign_v(struct nilfs_bmap *btree,
2113 struct nilfs_btree_path *path,
2115 struct buffer_head **bh,
2117 union nilfs_binfo *binfo)
2119 struct nilfs_btree_node *parent;
2120 struct inode *dat = nilfs_bmap_get_dat(btree);
2123 union nilfs_bmap_ptr_req req;
2126 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
2127 ptr = nilfs_btree_node_get_ptr(parent, path[level + 1].bp_index,
2130 ret = nilfs_dat_prepare_start(dat, &req.bpr_req);
2133 nilfs_dat_commit_start(dat, &req.bpr_req, blocknr);
2135 key = nilfs_btree_node_get_key(parent, path[level + 1].bp_index);
2136 /* on-disk format */
2137 binfo->bi_v.bi_vblocknr = cpu_to_le64(ptr);
2138 binfo->bi_v.bi_blkoff = cpu_to_le64(key);
2143 static int nilfs_btree_assign(struct nilfs_bmap *btree,
2144 struct buffer_head **bh,
2146 union nilfs_binfo *binfo)
2148 struct nilfs_btree_path *path;
2149 struct nilfs_btree_node *node;
2153 path = nilfs_btree_alloc_path();
2157 if (buffer_nilfs_node(*bh)) {
2158 node = (struct nilfs_btree_node *)(*bh)->b_data;
2159 key = nilfs_btree_node_get_key(node, 0);
2160 level = nilfs_btree_node_get_level(node);
2162 key = nilfs_bmap_data_get_key(btree, *bh);
2163 level = NILFS_BTREE_LEVEL_DATA;
2166 ret = nilfs_btree_do_lookup(btree, path, key, NULL, level + 1, 0);
2168 WARN_ON(ret == -ENOENT);
2172 ret = NILFS_BMAP_USE_VBN(btree) ?
2173 nilfs_btree_assign_v(btree, path, level, bh, blocknr, binfo) :
2174 nilfs_btree_assign_p(btree, path, level, bh, blocknr, binfo);
2177 nilfs_btree_free_path(path);
2182 static int nilfs_btree_assign_gc(struct nilfs_bmap *btree,
2183 struct buffer_head **bh,
2185 union nilfs_binfo *binfo)
2187 struct nilfs_btree_node *node;
2191 ret = nilfs_dat_move(nilfs_bmap_get_dat(btree), (*bh)->b_blocknr,
2196 if (buffer_nilfs_node(*bh)) {
2197 node = (struct nilfs_btree_node *)(*bh)->b_data;
2198 key = nilfs_btree_node_get_key(node, 0);
2200 key = nilfs_bmap_data_get_key(btree, *bh);
2202 /* on-disk format */
2203 binfo->bi_v.bi_vblocknr = cpu_to_le64((*bh)->b_blocknr);
2204 binfo->bi_v.bi_blkoff = cpu_to_le64(key);
2209 static int nilfs_btree_mark(struct nilfs_bmap *btree, __u64 key, int level)
2211 struct buffer_head *bh;
2212 struct nilfs_btree_path *path;
2216 path = nilfs_btree_alloc_path();
2220 ret = nilfs_btree_do_lookup(btree, path, key, &ptr, level + 1, 0);
2222 WARN_ON(ret == -ENOENT);
2225 ret = nilfs_btree_get_block(btree, ptr, &bh);
2227 WARN_ON(ret == -ENOENT);
2231 if (!buffer_dirty(bh))
2232 nilfs_btnode_mark_dirty(bh);
2234 if (!nilfs_bmap_dirty(btree))
2235 nilfs_bmap_set_dirty(btree);
2238 nilfs_btree_free_path(path);
2242 static const struct nilfs_bmap_operations nilfs_btree_ops = {
2243 .bop_lookup = nilfs_btree_lookup,
2244 .bop_lookup_contig = nilfs_btree_lookup_contig,
2245 .bop_insert = nilfs_btree_insert,
2246 .bop_delete = nilfs_btree_delete,
2249 .bop_propagate = nilfs_btree_propagate,
2251 .bop_lookup_dirty_buffers = nilfs_btree_lookup_dirty_buffers,
2253 .bop_assign = nilfs_btree_assign,
2254 .bop_mark = nilfs_btree_mark,
2256 .bop_last_key = nilfs_btree_last_key,
2257 .bop_check_insert = NULL,
2258 .bop_check_delete = nilfs_btree_check_delete,
2259 .bop_gather_data = nilfs_btree_gather_data,
2262 static const struct nilfs_bmap_operations nilfs_btree_ops_gc = {
2264 .bop_lookup_contig = NULL,
2269 .bop_propagate = nilfs_btree_propagate_gc,
2271 .bop_lookup_dirty_buffers = nilfs_btree_lookup_dirty_buffers,
2273 .bop_assign = nilfs_btree_assign_gc,
2276 .bop_last_key = NULL,
2277 .bop_check_insert = NULL,
2278 .bop_check_delete = NULL,
2279 .bop_gather_data = NULL,
2282 int nilfs_btree_init(struct nilfs_bmap *bmap)
2284 bmap->b_ops = &nilfs_btree_ops;
2285 bmap->b_nchildren_per_block =
2286 NILFS_BTREE_NODE_NCHILDREN_MAX(nilfs_btree_node_size(bmap));
2290 void nilfs_btree_init_gc(struct nilfs_bmap *bmap)
2292 bmap->b_ops = &nilfs_btree_ops_gc;
2293 bmap->b_nchildren_per_block =
2294 NILFS_BTREE_NODE_NCHILDREN_MAX(nilfs_btree_node_size(bmap));