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
20 * Written by Koji Sato <koji@osrg.net>.
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_block(const struct nilfs_bmap *btree, __u64 ptr,
70 struct buffer_head **bhp)
72 struct address_space *btnc = &NILFS_BMAP_I(btree)->i_btnode_cache;
73 struct buffer_head *bh;
77 err = nilfs_btnode_submit_block(btnc, ptr, pbn, READ, bhp, &pbn);
79 return err == -EEXIST ? 0 : err;
83 if (!buffer_uptodate(bh)) {
87 if (nilfs_btree_broken_node_block(bh)) {
88 clear_buffer_uptodate(bh);
95 static int nilfs_btree_get_new_block(const struct nilfs_bmap *btree,
96 __u64 ptr, struct buffer_head **bhp)
98 struct address_space *btnc = &NILFS_BMAP_I(btree)->i_btnode_cache;
99 struct buffer_head *bh;
101 bh = nilfs_btnode_create_block(btnc, ptr);
105 set_buffer_nilfs_volatile(bh);
110 static int nilfs_btree_node_get_flags(const struct nilfs_btree_node *node)
112 return node->bn_flags;
116 nilfs_btree_node_set_flags(struct nilfs_btree_node *node, int flags)
118 node->bn_flags = flags;
121 static int nilfs_btree_node_root(const struct nilfs_btree_node *node)
123 return nilfs_btree_node_get_flags(node) & NILFS_BTREE_NODE_ROOT;
126 static int nilfs_btree_node_get_level(const struct nilfs_btree_node *node)
128 return node->bn_level;
132 nilfs_btree_node_set_level(struct nilfs_btree_node *node, int level)
134 node->bn_level = level;
137 static int nilfs_btree_node_get_nchildren(const struct nilfs_btree_node *node)
139 return le16_to_cpu(node->bn_nchildren);
143 nilfs_btree_node_set_nchildren(struct nilfs_btree_node *node, int nchildren)
145 node->bn_nchildren = cpu_to_le16(nchildren);
148 static int nilfs_btree_node_size(const struct nilfs_bmap *btree)
150 return 1 << btree->b_inode->i_blkbits;
153 static int nilfs_btree_nchildren_per_block(const struct nilfs_bmap *btree)
155 return btree->b_nchildren_per_block;
159 nilfs_btree_node_dkeys(const struct nilfs_btree_node *node)
161 return (__le64 *)((char *)(node + 1) +
162 (nilfs_btree_node_root(node) ?
163 0 : NILFS_BTREE_NODE_EXTRA_PAD_SIZE));
167 nilfs_btree_node_dptrs(const struct nilfs_btree_node *node, int ncmax)
169 return (__le64 *)(nilfs_btree_node_dkeys(node) + ncmax);
173 nilfs_btree_node_get_key(const struct nilfs_btree_node *node, int index)
175 return le64_to_cpu(*(nilfs_btree_node_dkeys(node) + index));
179 nilfs_btree_node_set_key(struct nilfs_btree_node *node, int index, __u64 key)
181 *(nilfs_btree_node_dkeys(node) + index) = cpu_to_le64(key);
185 nilfs_btree_node_get_ptr(const struct nilfs_btree_node *node, int index,
188 return le64_to_cpu(*(nilfs_btree_node_dptrs(node, ncmax) + index));
192 nilfs_btree_node_set_ptr(struct nilfs_btree_node *node, int index, __u64 ptr,
195 *(nilfs_btree_node_dptrs(node, ncmax) + index) = cpu_to_le64(ptr);
198 static void nilfs_btree_node_init(struct nilfs_btree_node *node, int flags,
199 int level, int nchildren, int ncmax,
200 const __u64 *keys, const __u64 *ptrs)
206 nilfs_btree_node_set_flags(node, flags);
207 nilfs_btree_node_set_level(node, level);
208 nilfs_btree_node_set_nchildren(node, nchildren);
210 dkeys = nilfs_btree_node_dkeys(node);
211 dptrs = nilfs_btree_node_dptrs(node, ncmax);
212 for (i = 0; i < nchildren; i++) {
213 dkeys[i] = cpu_to_le64(keys[i]);
214 dptrs[i] = cpu_to_le64(ptrs[i]);
218 /* Assume the buffer heads corresponding to left and right are locked. */
219 static void nilfs_btree_node_move_left(struct nilfs_btree_node *left,
220 struct nilfs_btree_node *right,
221 int n, int lncmax, int rncmax)
223 __le64 *ldkeys, *rdkeys;
224 __le64 *ldptrs, *rdptrs;
225 int lnchildren, rnchildren;
227 ldkeys = nilfs_btree_node_dkeys(left);
228 ldptrs = nilfs_btree_node_dptrs(left, lncmax);
229 lnchildren = nilfs_btree_node_get_nchildren(left);
231 rdkeys = nilfs_btree_node_dkeys(right);
232 rdptrs = nilfs_btree_node_dptrs(right, rncmax);
233 rnchildren = nilfs_btree_node_get_nchildren(right);
235 memcpy(ldkeys + lnchildren, rdkeys, n * sizeof(*rdkeys));
236 memcpy(ldptrs + lnchildren, rdptrs, n * sizeof(*rdptrs));
237 memmove(rdkeys, rdkeys + n, (rnchildren - n) * sizeof(*rdkeys));
238 memmove(rdptrs, rdptrs + n, (rnchildren - n) * sizeof(*rdptrs));
242 nilfs_btree_node_set_nchildren(left, lnchildren);
243 nilfs_btree_node_set_nchildren(right, rnchildren);
246 /* Assume that the buffer heads corresponding to left and right are locked. */
247 static void nilfs_btree_node_move_right(struct nilfs_btree_node *left,
248 struct nilfs_btree_node *right,
249 int n, int lncmax, int rncmax)
251 __le64 *ldkeys, *rdkeys;
252 __le64 *ldptrs, *rdptrs;
253 int lnchildren, rnchildren;
255 ldkeys = nilfs_btree_node_dkeys(left);
256 ldptrs = nilfs_btree_node_dptrs(left, lncmax);
257 lnchildren = nilfs_btree_node_get_nchildren(left);
259 rdkeys = nilfs_btree_node_dkeys(right);
260 rdptrs = nilfs_btree_node_dptrs(right, rncmax);
261 rnchildren = nilfs_btree_node_get_nchildren(right);
263 memmove(rdkeys + n, rdkeys, rnchildren * sizeof(*rdkeys));
264 memmove(rdptrs + n, rdptrs, rnchildren * sizeof(*rdptrs));
265 memcpy(rdkeys, ldkeys + lnchildren - n, n * sizeof(*rdkeys));
266 memcpy(rdptrs, ldptrs + lnchildren - n, n * sizeof(*rdptrs));
270 nilfs_btree_node_set_nchildren(left, lnchildren);
271 nilfs_btree_node_set_nchildren(right, rnchildren);
274 /* Assume that the buffer head corresponding to node is locked. */
275 static void nilfs_btree_node_insert(struct nilfs_btree_node *node, int index,
276 __u64 key, __u64 ptr, int ncmax)
282 dkeys = nilfs_btree_node_dkeys(node);
283 dptrs = nilfs_btree_node_dptrs(node, ncmax);
284 nchildren = nilfs_btree_node_get_nchildren(node);
285 if (index < nchildren) {
286 memmove(dkeys + index + 1, dkeys + index,
287 (nchildren - index) * sizeof(*dkeys));
288 memmove(dptrs + index + 1, dptrs + index,
289 (nchildren - index) * sizeof(*dptrs));
291 dkeys[index] = cpu_to_le64(key);
292 dptrs[index] = cpu_to_le64(ptr);
294 nilfs_btree_node_set_nchildren(node, nchildren);
297 /* Assume that the buffer head corresponding to node is locked. */
298 static void nilfs_btree_node_delete(struct nilfs_btree_node *node, int index,
299 __u64 *keyp, __u64 *ptrp, int ncmax)
307 dkeys = nilfs_btree_node_dkeys(node);
308 dptrs = nilfs_btree_node_dptrs(node, ncmax);
309 key = le64_to_cpu(dkeys[index]);
310 ptr = le64_to_cpu(dptrs[index]);
311 nchildren = nilfs_btree_node_get_nchildren(node);
317 if (index < nchildren - 1) {
318 memmove(dkeys + index, dkeys + index + 1,
319 (nchildren - index - 1) * sizeof(*dkeys));
320 memmove(dptrs + index, dptrs + index + 1,
321 (nchildren - index - 1) * sizeof(*dptrs));
324 nilfs_btree_node_set_nchildren(node, nchildren);
327 static int nilfs_btree_node_lookup(const struct nilfs_btree_node *node,
328 __u64 key, int *indexp)
331 int index, low, high, s;
335 high = nilfs_btree_node_get_nchildren(node) - 1;
338 while (low <= high) {
339 index = (low + high) / 2;
340 nkey = nilfs_btree_node_get_key(node, index);
344 } else if (nkey < key) {
354 if (nilfs_btree_node_get_level(node) > NILFS_BTREE_LEVEL_NODE_MIN) {
355 if (s > 0 && index > 0)
367 * nilfs_btree_node_broken - verify consistency of btree node
368 * @node: btree node block to be examined
369 * @size: node size (in bytes)
370 * @blocknr: block number
372 * Return Value: If node is broken, 1 is returned. Otherwise, 0 is returned.
374 static int nilfs_btree_node_broken(const struct nilfs_btree_node *node,
375 size_t size, sector_t blocknr)
377 int level, flags, nchildren;
380 level = nilfs_btree_node_get_level(node);
381 flags = nilfs_btree_node_get_flags(node);
382 nchildren = nilfs_btree_node_get_nchildren(node);
384 if (unlikely(level < NILFS_BTREE_LEVEL_NODE_MIN ||
385 level >= NILFS_BTREE_LEVEL_MAX ||
386 (flags & NILFS_BTREE_NODE_ROOT) ||
388 nchildren > NILFS_BTREE_NODE_NCHILDREN_MAX(size))) {
389 printk(KERN_CRIT "NILFS: bad btree node (blocknr=%llu): "
390 "level = %d, flags = 0x%x, nchildren = %d\n",
391 (unsigned long long)blocknr, level, flags, nchildren);
397 int nilfs_btree_broken_node_block(struct buffer_head *bh)
399 return nilfs_btree_node_broken((struct nilfs_btree_node *)bh->b_data,
400 bh->b_size, bh->b_blocknr);
403 static struct nilfs_btree_node *
404 nilfs_btree_get_root(const struct nilfs_bmap *btree)
406 return (struct nilfs_btree_node *)btree->b_u.u_data;
409 static struct nilfs_btree_node *
410 nilfs_btree_get_nonroot_node(const struct nilfs_btree_path *path, int level)
412 return (struct nilfs_btree_node *)path[level].bp_bh->b_data;
415 static struct nilfs_btree_node *
416 nilfs_btree_get_sib_node(const struct nilfs_btree_path *path, int level)
418 return (struct nilfs_btree_node *)path[level].bp_sib_bh->b_data;
421 static int nilfs_btree_height(const struct nilfs_bmap *btree)
423 return nilfs_btree_node_get_level(nilfs_btree_get_root(btree)) + 1;
426 static struct nilfs_btree_node *
427 nilfs_btree_get_node(const struct nilfs_bmap *btree,
428 const struct nilfs_btree_path *path,
429 int level, int *ncmaxp)
431 struct nilfs_btree_node *node;
433 if (level == nilfs_btree_height(btree) - 1) {
434 node = nilfs_btree_get_root(btree);
435 *ncmaxp = NILFS_BTREE_ROOT_NCHILDREN_MAX;
437 node = nilfs_btree_get_nonroot_node(path, level);
438 *ncmaxp = nilfs_btree_nchildren_per_block(btree);
444 nilfs_btree_bad_node(struct nilfs_btree_node *node, int level)
446 if (unlikely(nilfs_btree_node_get_level(node) != level)) {
448 printk(KERN_CRIT "NILFS: btree level mismatch: %d != %d\n",
449 nilfs_btree_node_get_level(node), level);
455 static int nilfs_btree_do_lookup(const struct nilfs_bmap *btree,
456 struct nilfs_btree_path *path,
457 __u64 key, __u64 *ptrp, int minlevel)
459 struct nilfs_btree_node *node;
461 int level, index, found, ncmax, ret;
463 node = nilfs_btree_get_root(btree);
464 level = nilfs_btree_node_get_level(node);
465 if (level < minlevel || nilfs_btree_node_get_nchildren(node) <= 0)
468 found = nilfs_btree_node_lookup(node, key, &index);
469 ptr = nilfs_btree_node_get_ptr(node, index,
470 NILFS_BTREE_ROOT_NCHILDREN_MAX);
471 path[level].bp_bh = NULL;
472 path[level].bp_index = index;
474 ncmax = nilfs_btree_nchildren_per_block(btree);
476 for (level--; level >= minlevel; level--) {
477 ret = nilfs_btree_get_block(btree, ptr, &path[level].bp_bh);
480 node = nilfs_btree_get_nonroot_node(path, level);
481 if (nilfs_btree_bad_node(node, level))
484 found = nilfs_btree_node_lookup(node, key, &index);
488 ptr = nilfs_btree_node_get_ptr(node, index, ncmax);
490 WARN_ON(found || level != NILFS_BTREE_LEVEL_NODE_MIN);
492 ptr = NILFS_BMAP_INVALID_PTR;
494 path[level].bp_index = index;
505 static int nilfs_btree_do_lookup_last(const struct nilfs_bmap *btree,
506 struct nilfs_btree_path *path,
507 __u64 *keyp, __u64 *ptrp)
509 struct nilfs_btree_node *node;
511 int index, level, ncmax, ret;
513 node = nilfs_btree_get_root(btree);
514 index = nilfs_btree_node_get_nchildren(node) - 1;
517 level = nilfs_btree_node_get_level(node);
518 ptr = nilfs_btree_node_get_ptr(node, index,
519 NILFS_BTREE_ROOT_NCHILDREN_MAX);
520 path[level].bp_bh = NULL;
521 path[level].bp_index = index;
522 ncmax = nilfs_btree_nchildren_per_block(btree);
524 for (level--; level > 0; level--) {
525 ret = nilfs_btree_get_block(btree, ptr, &path[level].bp_bh);
528 node = nilfs_btree_get_nonroot_node(path, level);
529 if (nilfs_btree_bad_node(node, level))
531 index = nilfs_btree_node_get_nchildren(node) - 1;
532 ptr = nilfs_btree_node_get_ptr(node, index, ncmax);
533 path[level].bp_index = index;
537 *keyp = nilfs_btree_node_get_key(node, index);
544 static int nilfs_btree_lookup(const struct nilfs_bmap *btree,
545 __u64 key, int level, __u64 *ptrp)
547 struct nilfs_btree_path *path;
550 path = nilfs_btree_alloc_path();
554 ret = nilfs_btree_do_lookup(btree, path, key, ptrp, level);
556 nilfs_btree_free_path(path);
561 static int nilfs_btree_lookup_contig(const struct nilfs_bmap *btree,
562 __u64 key, __u64 *ptrp, unsigned maxblocks)
564 struct nilfs_btree_path *path;
565 struct nilfs_btree_node *node;
566 struct inode *dat = NULL;
569 int level = NILFS_BTREE_LEVEL_NODE_MIN;
570 int ret, cnt, index, maxlevel, ncmax;
572 path = nilfs_btree_alloc_path();
576 ret = nilfs_btree_do_lookup(btree, path, key, &ptr, level);
580 if (NILFS_BMAP_USE_VBN(btree)) {
581 dat = nilfs_bmap_get_dat(btree);
582 ret = nilfs_dat_translate(dat, ptr, &blocknr);
588 if (cnt == maxblocks)
591 maxlevel = nilfs_btree_height(btree) - 1;
592 node = nilfs_btree_get_node(btree, path, level, &ncmax);
593 index = path[level].bp_index + 1;
595 while (index < nilfs_btree_node_get_nchildren(node)) {
596 if (nilfs_btree_node_get_key(node, index) !=
599 ptr2 = nilfs_btree_node_get_ptr(node, index, ncmax);
601 ret = nilfs_dat_translate(dat, ptr2, &blocknr);
606 if (ptr2 != ptr + cnt || ++cnt == maxblocks)
611 if (level == maxlevel)
614 /* look-up right sibling node */
615 node = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
616 index = path[level + 1].bp_index + 1;
617 if (index >= nilfs_btree_node_get_nchildren(node) ||
618 nilfs_btree_node_get_key(node, index) != key + cnt)
620 ptr2 = nilfs_btree_node_get_ptr(node, index, ncmax);
621 path[level + 1].bp_index = index;
623 brelse(path[level].bp_bh);
624 path[level].bp_bh = NULL;
625 ret = nilfs_btree_get_block(btree, ptr2, &path[level].bp_bh);
628 node = nilfs_btree_get_nonroot_node(path, level);
629 ncmax = nilfs_btree_nchildren_per_block(btree);
631 path[level].bp_index = index;
637 nilfs_btree_free_path(path);
641 static void nilfs_btree_promote_key(struct nilfs_bmap *btree,
642 struct nilfs_btree_path *path,
643 int level, __u64 key)
645 if (level < nilfs_btree_height(btree) - 1) {
647 nilfs_btree_node_set_key(
648 nilfs_btree_get_nonroot_node(path, level),
649 path[level].bp_index, key);
650 if (!buffer_dirty(path[level].bp_bh))
651 nilfs_btnode_mark_dirty(path[level].bp_bh);
652 } while ((path[level].bp_index == 0) &&
653 (++level < nilfs_btree_height(btree) - 1));
657 if (level == nilfs_btree_height(btree) - 1) {
658 nilfs_btree_node_set_key(nilfs_btree_get_root(btree),
659 path[level].bp_index, key);
663 static void nilfs_btree_do_insert(struct nilfs_bmap *btree,
664 struct nilfs_btree_path *path,
665 int level, __u64 *keyp, __u64 *ptrp)
667 struct nilfs_btree_node *node;
670 if (level < nilfs_btree_height(btree) - 1) {
671 node = nilfs_btree_get_nonroot_node(path, level);
672 ncblk = nilfs_btree_nchildren_per_block(btree);
673 nilfs_btree_node_insert(node, path[level].bp_index,
674 *keyp, *ptrp, ncblk);
675 if (!buffer_dirty(path[level].bp_bh))
676 nilfs_btnode_mark_dirty(path[level].bp_bh);
678 if (path[level].bp_index == 0)
679 nilfs_btree_promote_key(btree, path, level + 1,
680 nilfs_btree_node_get_key(node,
683 node = nilfs_btree_get_root(btree);
684 nilfs_btree_node_insert(node, path[level].bp_index,
686 NILFS_BTREE_ROOT_NCHILDREN_MAX);
690 static void nilfs_btree_carry_left(struct nilfs_bmap *btree,
691 struct nilfs_btree_path *path,
692 int level, __u64 *keyp, __u64 *ptrp)
694 struct nilfs_btree_node *node, *left;
695 int nchildren, lnchildren, n, move, ncblk;
697 node = nilfs_btree_get_nonroot_node(path, level);
698 left = nilfs_btree_get_sib_node(path, level);
699 nchildren = nilfs_btree_node_get_nchildren(node);
700 lnchildren = nilfs_btree_node_get_nchildren(left);
701 ncblk = nilfs_btree_nchildren_per_block(btree);
704 n = (nchildren + lnchildren + 1) / 2 - lnchildren;
705 if (n > path[level].bp_index) {
706 /* move insert point */
711 nilfs_btree_node_move_left(left, node, n, ncblk, ncblk);
713 if (!buffer_dirty(path[level].bp_bh))
714 nilfs_btnode_mark_dirty(path[level].bp_bh);
715 if (!buffer_dirty(path[level].bp_sib_bh))
716 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
718 nilfs_btree_promote_key(btree, path, level + 1,
719 nilfs_btree_node_get_key(node, 0));
722 brelse(path[level].bp_bh);
723 path[level].bp_bh = path[level].bp_sib_bh;
724 path[level].bp_sib_bh = NULL;
725 path[level].bp_index += lnchildren;
726 path[level + 1].bp_index--;
728 brelse(path[level].bp_sib_bh);
729 path[level].bp_sib_bh = NULL;
730 path[level].bp_index -= n;
733 nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
736 static void nilfs_btree_carry_right(struct nilfs_bmap *btree,
737 struct nilfs_btree_path *path,
738 int level, __u64 *keyp, __u64 *ptrp)
740 struct nilfs_btree_node *node, *right;
741 int nchildren, rnchildren, n, move, ncblk;
743 node = nilfs_btree_get_nonroot_node(path, level);
744 right = nilfs_btree_get_sib_node(path, level);
745 nchildren = nilfs_btree_node_get_nchildren(node);
746 rnchildren = nilfs_btree_node_get_nchildren(right);
747 ncblk = nilfs_btree_nchildren_per_block(btree);
750 n = (nchildren + rnchildren + 1) / 2 - rnchildren;
751 if (n > nchildren - path[level].bp_index) {
752 /* move insert point */
757 nilfs_btree_node_move_right(node, right, n, ncblk, ncblk);
759 if (!buffer_dirty(path[level].bp_bh))
760 nilfs_btnode_mark_dirty(path[level].bp_bh);
761 if (!buffer_dirty(path[level].bp_sib_bh))
762 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
764 path[level + 1].bp_index++;
765 nilfs_btree_promote_key(btree, path, level + 1,
766 nilfs_btree_node_get_key(right, 0));
767 path[level + 1].bp_index--;
770 brelse(path[level].bp_bh);
771 path[level].bp_bh = path[level].bp_sib_bh;
772 path[level].bp_sib_bh = NULL;
773 path[level].bp_index -= nilfs_btree_node_get_nchildren(node);
774 path[level + 1].bp_index++;
776 brelse(path[level].bp_sib_bh);
777 path[level].bp_sib_bh = NULL;
780 nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
783 static void nilfs_btree_split(struct nilfs_bmap *btree,
784 struct nilfs_btree_path *path,
785 int level, __u64 *keyp, __u64 *ptrp)
787 struct nilfs_btree_node *node, *right;
790 int nchildren, n, move, ncblk;
792 node = nilfs_btree_get_nonroot_node(path, level);
793 right = nilfs_btree_get_sib_node(path, level);
794 nchildren = nilfs_btree_node_get_nchildren(node);
795 ncblk = nilfs_btree_nchildren_per_block(btree);
798 n = (nchildren + 1) / 2;
799 if (n > nchildren - path[level].bp_index) {
804 nilfs_btree_node_move_right(node, right, n, ncblk, ncblk);
806 if (!buffer_dirty(path[level].bp_bh))
807 nilfs_btnode_mark_dirty(path[level].bp_bh);
808 if (!buffer_dirty(path[level].bp_sib_bh))
809 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
811 newkey = nilfs_btree_node_get_key(right, 0);
812 newptr = path[level].bp_newreq.bpr_ptr;
815 path[level].bp_index -= nilfs_btree_node_get_nchildren(node);
816 nilfs_btree_node_insert(right, path[level].bp_index,
817 *keyp, *ptrp, ncblk);
819 *keyp = nilfs_btree_node_get_key(right, 0);
820 *ptrp = path[level].bp_newreq.bpr_ptr;
822 brelse(path[level].bp_bh);
823 path[level].bp_bh = path[level].bp_sib_bh;
824 path[level].bp_sib_bh = NULL;
826 nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
828 *keyp = nilfs_btree_node_get_key(right, 0);
829 *ptrp = path[level].bp_newreq.bpr_ptr;
831 brelse(path[level].bp_sib_bh);
832 path[level].bp_sib_bh = NULL;
835 path[level + 1].bp_index++;
838 static void nilfs_btree_grow(struct nilfs_bmap *btree,
839 struct nilfs_btree_path *path,
840 int level, __u64 *keyp, __u64 *ptrp)
842 struct nilfs_btree_node *root, *child;
845 root = nilfs_btree_get_root(btree);
846 child = nilfs_btree_get_sib_node(path, level);
847 ncblk = nilfs_btree_nchildren_per_block(btree);
849 n = nilfs_btree_node_get_nchildren(root);
851 nilfs_btree_node_move_right(root, child, n,
852 NILFS_BTREE_ROOT_NCHILDREN_MAX, ncblk);
853 nilfs_btree_node_set_level(root, level + 1);
855 if (!buffer_dirty(path[level].bp_sib_bh))
856 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
858 path[level].bp_bh = path[level].bp_sib_bh;
859 path[level].bp_sib_bh = NULL;
861 nilfs_btree_do_insert(btree, path, level, keyp, ptrp);
863 *keyp = nilfs_btree_node_get_key(child, 0);
864 *ptrp = path[level].bp_newreq.bpr_ptr;
867 static __u64 nilfs_btree_find_near(const struct nilfs_bmap *btree,
868 const struct nilfs_btree_path *path)
870 struct nilfs_btree_node *node;
874 return NILFS_BMAP_INVALID_PTR;
877 level = NILFS_BTREE_LEVEL_NODE_MIN;
878 if (path[level].bp_index > 0) {
879 node = nilfs_btree_get_node(btree, path, level, &ncmax);
880 return nilfs_btree_node_get_ptr(node,
881 path[level].bp_index - 1,
886 level = NILFS_BTREE_LEVEL_NODE_MIN + 1;
887 if (level <= nilfs_btree_height(btree) - 1) {
888 node = nilfs_btree_get_node(btree, path, level, &ncmax);
889 return nilfs_btree_node_get_ptr(node, path[level].bp_index,
893 return NILFS_BMAP_INVALID_PTR;
896 static __u64 nilfs_btree_find_target_v(const struct nilfs_bmap *btree,
897 const struct nilfs_btree_path *path,
902 ptr = nilfs_bmap_find_target_seq(btree, key);
903 if (ptr != NILFS_BMAP_INVALID_PTR)
904 /* sequential access */
907 ptr = nilfs_btree_find_near(btree, path);
908 if (ptr != NILFS_BMAP_INVALID_PTR)
913 return nilfs_bmap_find_target_in_group(btree);
916 static int nilfs_btree_prepare_insert(struct nilfs_bmap *btree,
917 struct nilfs_btree_path *path,
918 int *levelp, __u64 key, __u64 ptr,
919 struct nilfs_bmap_stats *stats)
921 struct buffer_head *bh;
922 struct nilfs_btree_node *node, *parent, *sib;
924 int pindex, level, ncmax, ncblk, ret;
925 struct inode *dat = NULL;
927 stats->bs_nblocks = 0;
928 level = NILFS_BTREE_LEVEL_DATA;
930 /* allocate a new ptr for data block */
931 if (NILFS_BMAP_USE_VBN(btree)) {
932 path[level].bp_newreq.bpr_ptr =
933 nilfs_btree_find_target_v(btree, path, key);
934 dat = nilfs_bmap_get_dat(btree);
937 ret = nilfs_bmap_prepare_alloc_ptr(btree, &path[level].bp_newreq, dat);
941 ncblk = nilfs_btree_nchildren_per_block(btree);
943 for (level = NILFS_BTREE_LEVEL_NODE_MIN;
944 level < nilfs_btree_height(btree) - 1;
946 node = nilfs_btree_get_nonroot_node(path, level);
947 if (nilfs_btree_node_get_nchildren(node) < ncblk) {
948 path[level].bp_op = nilfs_btree_do_insert;
953 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
954 pindex = path[level + 1].bp_index;
958 sibptr = nilfs_btree_node_get_ptr(parent, pindex - 1,
960 ret = nilfs_btree_get_block(btree, sibptr, &bh);
962 goto err_out_child_node;
963 sib = (struct nilfs_btree_node *)bh->b_data;
964 if (nilfs_btree_node_get_nchildren(sib) < ncblk) {
965 path[level].bp_sib_bh = bh;
966 path[level].bp_op = nilfs_btree_carry_left;
975 if (pindex < nilfs_btree_node_get_nchildren(parent) - 1) {
976 sibptr = nilfs_btree_node_get_ptr(parent, pindex + 1,
978 ret = nilfs_btree_get_block(btree, sibptr, &bh);
980 goto err_out_child_node;
981 sib = (struct nilfs_btree_node *)bh->b_data;
982 if (nilfs_btree_node_get_nchildren(sib) < ncblk) {
983 path[level].bp_sib_bh = bh;
984 path[level].bp_op = nilfs_btree_carry_right;
993 path[level].bp_newreq.bpr_ptr =
994 path[level - 1].bp_newreq.bpr_ptr + 1;
995 ret = nilfs_bmap_prepare_alloc_ptr(btree,
996 &path[level].bp_newreq, dat);
998 goto err_out_child_node;
999 ret = nilfs_btree_get_new_block(btree,
1000 path[level].bp_newreq.bpr_ptr,
1003 goto err_out_curr_node;
1005 stats->bs_nblocks++;
1007 sib = (struct nilfs_btree_node *)bh->b_data;
1008 nilfs_btree_node_init(sib, 0, level, 0, ncblk, NULL, NULL);
1009 path[level].bp_sib_bh = bh;
1010 path[level].bp_op = nilfs_btree_split;
1014 node = nilfs_btree_get_root(btree);
1015 if (nilfs_btree_node_get_nchildren(node) <
1016 NILFS_BTREE_ROOT_NCHILDREN_MAX) {
1017 path[level].bp_op = nilfs_btree_do_insert;
1018 stats->bs_nblocks++;
1023 path[level].bp_newreq.bpr_ptr = path[level - 1].bp_newreq.bpr_ptr + 1;
1024 ret = nilfs_bmap_prepare_alloc_ptr(btree, &path[level].bp_newreq, dat);
1026 goto err_out_child_node;
1027 ret = nilfs_btree_get_new_block(btree, path[level].bp_newreq.bpr_ptr,
1030 goto err_out_curr_node;
1032 nilfs_btree_node_init((struct nilfs_btree_node *)bh->b_data,
1033 0, level, 0, ncblk, NULL, NULL);
1034 path[level].bp_sib_bh = bh;
1035 path[level].bp_op = nilfs_btree_grow;
1038 path[level].bp_op = nilfs_btree_do_insert;
1040 /* a newly-created node block and a data block are added */
1041 stats->bs_nblocks += 2;
1050 nilfs_bmap_abort_alloc_ptr(btree, &path[level].bp_newreq, dat);
1052 for (level--; level > NILFS_BTREE_LEVEL_DATA; level--) {
1053 nilfs_btnode_delete(path[level].bp_sib_bh);
1054 nilfs_bmap_abort_alloc_ptr(btree, &path[level].bp_newreq, dat);
1058 nilfs_bmap_abort_alloc_ptr(btree, &path[level].bp_newreq, dat);
1061 stats->bs_nblocks = 0;
1065 static void nilfs_btree_commit_insert(struct nilfs_bmap *btree,
1066 struct nilfs_btree_path *path,
1067 int maxlevel, __u64 key, __u64 ptr)
1069 struct inode *dat = NULL;
1072 set_buffer_nilfs_volatile((struct buffer_head *)((unsigned long)ptr));
1073 ptr = path[NILFS_BTREE_LEVEL_DATA].bp_newreq.bpr_ptr;
1074 if (NILFS_BMAP_USE_VBN(btree)) {
1075 nilfs_bmap_set_target_v(btree, key, ptr);
1076 dat = nilfs_bmap_get_dat(btree);
1079 for (level = NILFS_BTREE_LEVEL_NODE_MIN; level <= maxlevel; level++) {
1080 nilfs_bmap_commit_alloc_ptr(btree,
1081 &path[level - 1].bp_newreq, dat);
1082 path[level].bp_op(btree, path, level, &key, &ptr);
1085 if (!nilfs_bmap_dirty(btree))
1086 nilfs_bmap_set_dirty(btree);
1089 static int nilfs_btree_insert(struct nilfs_bmap *btree, __u64 key, __u64 ptr)
1091 struct nilfs_btree_path *path;
1092 struct nilfs_bmap_stats stats;
1095 path = nilfs_btree_alloc_path();
1099 ret = nilfs_btree_do_lookup(btree, path, key, NULL,
1100 NILFS_BTREE_LEVEL_NODE_MIN);
1101 if (ret != -ENOENT) {
1107 ret = nilfs_btree_prepare_insert(btree, path, &level, key, ptr, &stats);
1110 nilfs_btree_commit_insert(btree, path, level, key, ptr);
1111 nilfs_bmap_add_blocks(btree, stats.bs_nblocks);
1114 nilfs_btree_free_path(path);
1118 static void nilfs_btree_do_delete(struct nilfs_bmap *btree,
1119 struct nilfs_btree_path *path,
1120 int level, __u64 *keyp, __u64 *ptrp)
1122 struct nilfs_btree_node *node;
1125 if (level < nilfs_btree_height(btree) - 1) {
1126 node = nilfs_btree_get_nonroot_node(path, level);
1127 ncblk = nilfs_btree_nchildren_per_block(btree);
1128 nilfs_btree_node_delete(node, path[level].bp_index,
1130 if (!buffer_dirty(path[level].bp_bh))
1131 nilfs_btnode_mark_dirty(path[level].bp_bh);
1132 if (path[level].bp_index == 0)
1133 nilfs_btree_promote_key(btree, path, level + 1,
1134 nilfs_btree_node_get_key(node, 0));
1136 node = nilfs_btree_get_root(btree);
1137 nilfs_btree_node_delete(node, path[level].bp_index,
1139 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1143 static void nilfs_btree_borrow_left(struct nilfs_bmap *btree,
1144 struct nilfs_btree_path *path,
1145 int level, __u64 *keyp, __u64 *ptrp)
1147 struct nilfs_btree_node *node, *left;
1148 int nchildren, lnchildren, n, ncblk;
1150 nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
1152 node = nilfs_btree_get_nonroot_node(path, level);
1153 left = nilfs_btree_get_sib_node(path, level);
1154 nchildren = nilfs_btree_node_get_nchildren(node);
1155 lnchildren = nilfs_btree_node_get_nchildren(left);
1156 ncblk = nilfs_btree_nchildren_per_block(btree);
1158 n = (nchildren + lnchildren) / 2 - nchildren;
1160 nilfs_btree_node_move_right(left, node, n, ncblk, ncblk);
1162 if (!buffer_dirty(path[level].bp_bh))
1163 nilfs_btnode_mark_dirty(path[level].bp_bh);
1164 if (!buffer_dirty(path[level].bp_sib_bh))
1165 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
1167 nilfs_btree_promote_key(btree, path, level + 1,
1168 nilfs_btree_node_get_key(node, 0));
1170 brelse(path[level].bp_sib_bh);
1171 path[level].bp_sib_bh = NULL;
1172 path[level].bp_index += n;
1175 static void nilfs_btree_borrow_right(struct nilfs_bmap *btree,
1176 struct nilfs_btree_path *path,
1177 int level, __u64 *keyp, __u64 *ptrp)
1179 struct nilfs_btree_node *node, *right;
1180 int nchildren, rnchildren, n, ncblk;
1182 nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
1184 node = nilfs_btree_get_nonroot_node(path, level);
1185 right = nilfs_btree_get_sib_node(path, level);
1186 nchildren = nilfs_btree_node_get_nchildren(node);
1187 rnchildren = nilfs_btree_node_get_nchildren(right);
1188 ncblk = nilfs_btree_nchildren_per_block(btree);
1190 n = (nchildren + rnchildren) / 2 - nchildren;
1192 nilfs_btree_node_move_left(node, right, n, ncblk, ncblk);
1194 if (!buffer_dirty(path[level].bp_bh))
1195 nilfs_btnode_mark_dirty(path[level].bp_bh);
1196 if (!buffer_dirty(path[level].bp_sib_bh))
1197 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
1199 path[level + 1].bp_index++;
1200 nilfs_btree_promote_key(btree, path, level + 1,
1201 nilfs_btree_node_get_key(right, 0));
1202 path[level + 1].bp_index--;
1204 brelse(path[level].bp_sib_bh);
1205 path[level].bp_sib_bh = NULL;
1208 static void nilfs_btree_concat_left(struct nilfs_bmap *btree,
1209 struct nilfs_btree_path *path,
1210 int level, __u64 *keyp, __u64 *ptrp)
1212 struct nilfs_btree_node *node, *left;
1215 nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
1217 node = nilfs_btree_get_nonroot_node(path, level);
1218 left = nilfs_btree_get_sib_node(path, level);
1219 ncblk = nilfs_btree_nchildren_per_block(btree);
1221 n = nilfs_btree_node_get_nchildren(node);
1223 nilfs_btree_node_move_left(left, node, n, ncblk, ncblk);
1225 if (!buffer_dirty(path[level].bp_sib_bh))
1226 nilfs_btnode_mark_dirty(path[level].bp_sib_bh);
1228 nilfs_btnode_delete(path[level].bp_bh);
1229 path[level].bp_bh = path[level].bp_sib_bh;
1230 path[level].bp_sib_bh = NULL;
1231 path[level].bp_index += nilfs_btree_node_get_nchildren(left);
1234 static void nilfs_btree_concat_right(struct nilfs_bmap *btree,
1235 struct nilfs_btree_path *path,
1236 int level, __u64 *keyp, __u64 *ptrp)
1238 struct nilfs_btree_node *node, *right;
1241 nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
1243 node = nilfs_btree_get_nonroot_node(path, level);
1244 right = nilfs_btree_get_sib_node(path, level);
1245 ncblk = nilfs_btree_nchildren_per_block(btree);
1247 n = nilfs_btree_node_get_nchildren(right);
1249 nilfs_btree_node_move_left(node, right, n, ncblk, ncblk);
1251 if (!buffer_dirty(path[level].bp_bh))
1252 nilfs_btnode_mark_dirty(path[level].bp_bh);
1254 nilfs_btnode_delete(path[level].bp_sib_bh);
1255 path[level].bp_sib_bh = NULL;
1256 path[level + 1].bp_index++;
1259 static void nilfs_btree_shrink(struct nilfs_bmap *btree,
1260 struct nilfs_btree_path *path,
1261 int level, __u64 *keyp, __u64 *ptrp)
1263 struct nilfs_btree_node *root, *child;
1266 nilfs_btree_do_delete(btree, path, level, keyp, ptrp);
1268 root = nilfs_btree_get_root(btree);
1269 child = nilfs_btree_get_nonroot_node(path, level);
1270 ncblk = nilfs_btree_nchildren_per_block(btree);
1272 nilfs_btree_node_delete(root, 0, NULL, NULL,
1273 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1274 nilfs_btree_node_set_level(root, level);
1275 n = nilfs_btree_node_get_nchildren(child);
1276 nilfs_btree_node_move_left(root, child, n,
1277 NILFS_BTREE_ROOT_NCHILDREN_MAX, ncblk);
1279 nilfs_btnode_delete(path[level].bp_bh);
1280 path[level].bp_bh = NULL;
1284 static int nilfs_btree_prepare_delete(struct nilfs_bmap *btree,
1285 struct nilfs_btree_path *path,
1287 struct nilfs_bmap_stats *stats,
1290 struct buffer_head *bh;
1291 struct nilfs_btree_node *node, *parent, *sib;
1293 int pindex, level, ncmin, ncmax, ncblk, ret;
1296 stats->bs_nblocks = 0;
1297 ncmin = NILFS_BTREE_NODE_NCHILDREN_MIN(nilfs_btree_node_size(btree));
1298 ncblk = nilfs_btree_nchildren_per_block(btree);
1300 for (level = NILFS_BTREE_LEVEL_NODE_MIN;
1301 level < nilfs_btree_height(btree) - 1;
1303 node = nilfs_btree_get_nonroot_node(path, level);
1304 path[level].bp_oldreq.bpr_ptr =
1305 nilfs_btree_node_get_ptr(node, path[level].bp_index,
1307 ret = nilfs_bmap_prepare_end_ptr(btree,
1308 &path[level].bp_oldreq, dat);
1310 goto err_out_child_node;
1312 if (nilfs_btree_node_get_nchildren(node) > ncmin) {
1313 path[level].bp_op = nilfs_btree_do_delete;
1314 stats->bs_nblocks++;
1318 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
1319 pindex = path[level + 1].bp_index;
1323 sibptr = nilfs_btree_node_get_ptr(parent, pindex - 1,
1325 ret = nilfs_btree_get_block(btree, sibptr, &bh);
1327 goto err_out_curr_node;
1328 sib = (struct nilfs_btree_node *)bh->b_data;
1329 if (nilfs_btree_node_get_nchildren(sib) > ncmin) {
1330 path[level].bp_sib_bh = bh;
1331 path[level].bp_op = nilfs_btree_borrow_left;
1332 stats->bs_nblocks++;
1335 path[level].bp_sib_bh = bh;
1336 path[level].bp_op = nilfs_btree_concat_left;
1337 stats->bs_nblocks++;
1341 nilfs_btree_node_get_nchildren(parent) - 1) {
1343 sibptr = nilfs_btree_node_get_ptr(parent, pindex + 1,
1345 ret = nilfs_btree_get_block(btree, sibptr, &bh);
1347 goto err_out_curr_node;
1348 sib = (struct nilfs_btree_node *)bh->b_data;
1349 if (nilfs_btree_node_get_nchildren(sib) > ncmin) {
1350 path[level].bp_sib_bh = bh;
1351 path[level].bp_op = nilfs_btree_borrow_right;
1352 stats->bs_nblocks++;
1355 path[level].bp_sib_bh = bh;
1356 path[level].bp_op = nilfs_btree_concat_right;
1357 stats->bs_nblocks++;
1362 /* the only child of the root node */
1363 WARN_ON(level != nilfs_btree_height(btree) - 2);
1364 if (nilfs_btree_node_get_nchildren(node) - 1 <=
1365 NILFS_BTREE_ROOT_NCHILDREN_MAX) {
1366 path[level].bp_op = nilfs_btree_shrink;
1367 stats->bs_nblocks += 2;
1369 path[level].bp_op = nilfs_btree_do_delete;
1370 stats->bs_nblocks++;
1378 node = nilfs_btree_get_root(btree);
1379 path[level].bp_oldreq.bpr_ptr =
1380 nilfs_btree_node_get_ptr(node, path[level].bp_index,
1381 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1383 ret = nilfs_bmap_prepare_end_ptr(btree, &path[level].bp_oldreq, dat);
1385 goto err_out_child_node;
1387 /* child of the root node is deleted */
1388 path[level].bp_op = nilfs_btree_do_delete;
1389 stats->bs_nblocks++;
1398 nilfs_bmap_abort_end_ptr(btree, &path[level].bp_oldreq, dat);
1400 for (level--; level >= NILFS_BTREE_LEVEL_NODE_MIN; level--) {
1401 brelse(path[level].bp_sib_bh);
1402 nilfs_bmap_abort_end_ptr(btree, &path[level].bp_oldreq, dat);
1405 stats->bs_nblocks = 0;
1409 static void nilfs_btree_commit_delete(struct nilfs_bmap *btree,
1410 struct nilfs_btree_path *path,
1411 int maxlevel, struct inode *dat)
1415 for (level = NILFS_BTREE_LEVEL_NODE_MIN; level <= maxlevel; level++) {
1416 nilfs_bmap_commit_end_ptr(btree, &path[level].bp_oldreq, dat);
1417 path[level].bp_op(btree, path, level, NULL, NULL);
1420 if (!nilfs_bmap_dirty(btree))
1421 nilfs_bmap_set_dirty(btree);
1424 static int nilfs_btree_delete(struct nilfs_bmap *btree, __u64 key)
1427 struct nilfs_btree_path *path;
1428 struct nilfs_bmap_stats stats;
1432 path = nilfs_btree_alloc_path();
1436 ret = nilfs_btree_do_lookup(btree, path, key, NULL,
1437 NILFS_BTREE_LEVEL_NODE_MIN);
1442 dat = NILFS_BMAP_USE_VBN(btree) ? nilfs_bmap_get_dat(btree) : NULL;
1444 ret = nilfs_btree_prepare_delete(btree, path, &level, &stats, dat);
1447 nilfs_btree_commit_delete(btree, path, level, dat);
1448 nilfs_bmap_sub_blocks(btree, stats.bs_nblocks);
1451 nilfs_btree_free_path(path);
1455 static int nilfs_btree_last_key(const struct nilfs_bmap *btree, __u64 *keyp)
1457 struct nilfs_btree_path *path;
1460 path = nilfs_btree_alloc_path();
1464 ret = nilfs_btree_do_lookup_last(btree, path, keyp, NULL);
1466 nilfs_btree_free_path(path);
1471 static int nilfs_btree_check_delete(struct nilfs_bmap *btree, __u64 key)
1473 struct buffer_head *bh;
1474 struct nilfs_btree_node *root, *node;
1475 __u64 maxkey, nextmaxkey;
1479 root = nilfs_btree_get_root(btree);
1480 switch (nilfs_btree_height(btree)) {
1486 nchildren = nilfs_btree_node_get_nchildren(root);
1489 ptr = nilfs_btree_node_get_ptr(root, nchildren - 1,
1490 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1491 ret = nilfs_btree_get_block(btree, ptr, &bh);
1494 node = (struct nilfs_btree_node *)bh->b_data;
1500 nchildren = nilfs_btree_node_get_nchildren(node);
1501 maxkey = nilfs_btree_node_get_key(node, nchildren - 1);
1502 nextmaxkey = (nchildren > 1) ?
1503 nilfs_btree_node_get_key(node, nchildren - 2) : 0;
1507 return (maxkey == key) && (nextmaxkey < NILFS_BMAP_LARGE_LOW);
1510 static int nilfs_btree_gather_data(struct nilfs_bmap *btree,
1511 __u64 *keys, __u64 *ptrs, int nitems)
1513 struct buffer_head *bh;
1514 struct nilfs_btree_node *node, *root;
1518 int nchildren, ncmax, i, ret;
1520 root = nilfs_btree_get_root(btree);
1521 switch (nilfs_btree_height(btree)) {
1525 ncmax = NILFS_BTREE_ROOT_NCHILDREN_MAX;
1528 nchildren = nilfs_btree_node_get_nchildren(root);
1529 WARN_ON(nchildren > 1);
1530 ptr = nilfs_btree_node_get_ptr(root, nchildren - 1,
1531 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1532 ret = nilfs_btree_get_block(btree, ptr, &bh);
1535 node = (struct nilfs_btree_node *)bh->b_data;
1536 ncmax = nilfs_btree_nchildren_per_block(btree);
1543 nchildren = nilfs_btree_node_get_nchildren(node);
1544 if (nchildren < nitems)
1546 dkeys = nilfs_btree_node_dkeys(node);
1547 dptrs = nilfs_btree_node_dptrs(node, ncmax);
1548 for (i = 0; i < nitems; i++) {
1549 keys[i] = le64_to_cpu(dkeys[i]);
1550 ptrs[i] = le64_to_cpu(dptrs[i]);
1560 nilfs_btree_prepare_convert_and_insert(struct nilfs_bmap *btree, __u64 key,
1561 union nilfs_bmap_ptr_req *dreq,
1562 union nilfs_bmap_ptr_req *nreq,
1563 struct buffer_head **bhp,
1564 struct nilfs_bmap_stats *stats)
1566 struct buffer_head *bh;
1567 struct inode *dat = NULL;
1570 stats->bs_nblocks = 0;
1573 /* cannot find near ptr */
1574 if (NILFS_BMAP_USE_VBN(btree)) {
1575 dreq->bpr_ptr = nilfs_btree_find_target_v(btree, NULL, key);
1576 dat = nilfs_bmap_get_dat(btree);
1579 ret = nilfs_bmap_prepare_alloc_ptr(btree, dreq, dat);
1584 stats->bs_nblocks++;
1586 nreq->bpr_ptr = dreq->bpr_ptr + 1;
1587 ret = nilfs_bmap_prepare_alloc_ptr(btree, nreq, dat);
1591 ret = nilfs_btree_get_new_block(btree, nreq->bpr_ptr, &bh);
1596 stats->bs_nblocks++;
1604 nilfs_bmap_abort_alloc_ptr(btree, nreq, dat);
1606 nilfs_bmap_abort_alloc_ptr(btree, dreq, dat);
1607 stats->bs_nblocks = 0;
1613 nilfs_btree_commit_convert_and_insert(struct nilfs_bmap *btree,
1614 __u64 key, __u64 ptr,
1615 const __u64 *keys, const __u64 *ptrs,
1617 union nilfs_bmap_ptr_req *dreq,
1618 union nilfs_bmap_ptr_req *nreq,
1619 struct buffer_head *bh)
1621 struct nilfs_btree_node *node;
1626 /* free resources */
1627 if (btree->b_ops->bop_clear != NULL)
1628 btree->b_ops->bop_clear(btree);
1630 /* ptr must be a pointer to a buffer head. */
1631 set_buffer_nilfs_volatile((struct buffer_head *)((unsigned long)ptr));
1633 /* convert and insert */
1634 dat = NILFS_BMAP_USE_VBN(btree) ? nilfs_bmap_get_dat(btree) : NULL;
1635 nilfs_btree_init(btree);
1637 nilfs_bmap_commit_alloc_ptr(btree, dreq, dat);
1638 nilfs_bmap_commit_alloc_ptr(btree, nreq, dat);
1640 /* create child node at level 1 */
1641 node = (struct nilfs_btree_node *)bh->b_data;
1642 ncblk = nilfs_btree_nchildren_per_block(btree);
1643 nilfs_btree_node_init(node, 0, 1, n, ncblk, keys, ptrs);
1644 nilfs_btree_node_insert(node, n, key, dreq->bpr_ptr, ncblk);
1645 if (!buffer_dirty(bh))
1646 nilfs_btnode_mark_dirty(bh);
1647 if (!nilfs_bmap_dirty(btree))
1648 nilfs_bmap_set_dirty(btree);
1652 /* create root node at level 2 */
1653 node = nilfs_btree_get_root(btree);
1654 tmpptr = nreq->bpr_ptr;
1655 nilfs_btree_node_init(node, NILFS_BTREE_NODE_ROOT, 2, 1,
1656 NILFS_BTREE_ROOT_NCHILDREN_MAX,
1659 nilfs_bmap_commit_alloc_ptr(btree, dreq, dat);
1661 /* create root node at level 1 */
1662 node = nilfs_btree_get_root(btree);
1663 nilfs_btree_node_init(node, NILFS_BTREE_NODE_ROOT, 1, n,
1664 NILFS_BTREE_ROOT_NCHILDREN_MAX,
1666 nilfs_btree_node_insert(node, n, key, dreq->bpr_ptr,
1667 NILFS_BTREE_ROOT_NCHILDREN_MAX);
1668 if (!nilfs_bmap_dirty(btree))
1669 nilfs_bmap_set_dirty(btree);
1672 if (NILFS_BMAP_USE_VBN(btree))
1673 nilfs_bmap_set_target_v(btree, key, dreq->bpr_ptr);
1677 * nilfs_btree_convert_and_insert -
1685 int nilfs_btree_convert_and_insert(struct nilfs_bmap *btree,
1686 __u64 key, __u64 ptr,
1687 const __u64 *keys, const __u64 *ptrs, int n)
1689 struct buffer_head *bh;
1690 union nilfs_bmap_ptr_req dreq, nreq, *di, *ni;
1691 struct nilfs_bmap_stats stats;
1694 if (n + 1 <= NILFS_BTREE_ROOT_NCHILDREN_MAX) {
1697 } else if ((n + 1) <= NILFS_BTREE_NODE_NCHILDREN_MAX(
1698 1 << btree->b_inode->i_blkbits)) {
1707 ret = nilfs_btree_prepare_convert_and_insert(btree, key, di, ni, &bh,
1711 nilfs_btree_commit_convert_and_insert(btree, key, ptr, keys, ptrs, n,
1713 nilfs_bmap_add_blocks(btree, stats.bs_nblocks);
1717 static int nilfs_btree_propagate_p(struct nilfs_bmap *btree,
1718 struct nilfs_btree_path *path,
1720 struct buffer_head *bh)
1722 while ((++level < nilfs_btree_height(btree) - 1) &&
1723 !buffer_dirty(path[level].bp_bh))
1724 nilfs_btnode_mark_dirty(path[level].bp_bh);
1729 static int nilfs_btree_prepare_update_v(struct nilfs_bmap *btree,
1730 struct nilfs_btree_path *path,
1731 int level, struct inode *dat)
1733 struct nilfs_btree_node *parent;
1736 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
1737 path[level].bp_oldreq.bpr_ptr =
1738 nilfs_btree_node_get_ptr(parent, path[level + 1].bp_index,
1740 path[level].bp_newreq.bpr_ptr = path[level].bp_oldreq.bpr_ptr + 1;
1741 ret = nilfs_dat_prepare_update(dat, &path[level].bp_oldreq.bpr_req,
1742 &path[level].bp_newreq.bpr_req);
1746 if (buffer_nilfs_node(path[level].bp_bh)) {
1747 path[level].bp_ctxt.oldkey = path[level].bp_oldreq.bpr_ptr;
1748 path[level].bp_ctxt.newkey = path[level].bp_newreq.bpr_ptr;
1749 path[level].bp_ctxt.bh = path[level].bp_bh;
1750 ret = nilfs_btnode_prepare_change_key(
1751 &NILFS_BMAP_I(btree)->i_btnode_cache,
1752 &path[level].bp_ctxt);
1754 nilfs_dat_abort_update(dat,
1755 &path[level].bp_oldreq.bpr_req,
1756 &path[level].bp_newreq.bpr_req);
1764 static void nilfs_btree_commit_update_v(struct nilfs_bmap *btree,
1765 struct nilfs_btree_path *path,
1766 int level, struct inode *dat)
1768 struct nilfs_btree_node *parent;
1771 nilfs_dat_commit_update(dat, &path[level].bp_oldreq.bpr_req,
1772 &path[level].bp_newreq.bpr_req,
1773 btree->b_ptr_type == NILFS_BMAP_PTR_VS);
1775 if (buffer_nilfs_node(path[level].bp_bh)) {
1776 nilfs_btnode_commit_change_key(
1777 &NILFS_BMAP_I(btree)->i_btnode_cache,
1778 &path[level].bp_ctxt);
1779 path[level].bp_bh = path[level].bp_ctxt.bh;
1781 set_buffer_nilfs_volatile(path[level].bp_bh);
1783 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
1784 nilfs_btree_node_set_ptr(parent, path[level + 1].bp_index,
1785 path[level].bp_newreq.bpr_ptr, ncmax);
1788 static void nilfs_btree_abort_update_v(struct nilfs_bmap *btree,
1789 struct nilfs_btree_path *path,
1790 int level, struct inode *dat)
1792 nilfs_dat_abort_update(dat, &path[level].bp_oldreq.bpr_req,
1793 &path[level].bp_newreq.bpr_req);
1794 if (buffer_nilfs_node(path[level].bp_bh))
1795 nilfs_btnode_abort_change_key(
1796 &NILFS_BMAP_I(btree)->i_btnode_cache,
1797 &path[level].bp_ctxt);
1800 static int nilfs_btree_prepare_propagate_v(struct nilfs_bmap *btree,
1801 struct nilfs_btree_path *path,
1802 int minlevel, int *maxlevelp,
1808 if (!buffer_nilfs_volatile(path[level].bp_bh)) {
1809 ret = nilfs_btree_prepare_update_v(btree, path, level, dat);
1813 while ((++level < nilfs_btree_height(btree) - 1) &&
1814 !buffer_dirty(path[level].bp_bh)) {
1816 WARN_ON(buffer_nilfs_volatile(path[level].bp_bh));
1817 ret = nilfs_btree_prepare_update_v(btree, path, level, dat);
1823 *maxlevelp = level - 1;
1828 while (--level > minlevel)
1829 nilfs_btree_abort_update_v(btree, path, level, dat);
1830 if (!buffer_nilfs_volatile(path[level].bp_bh))
1831 nilfs_btree_abort_update_v(btree, path, level, dat);
1835 static void nilfs_btree_commit_propagate_v(struct nilfs_bmap *btree,
1836 struct nilfs_btree_path *path,
1837 int minlevel, int maxlevel,
1838 struct buffer_head *bh,
1843 if (!buffer_nilfs_volatile(path[minlevel].bp_bh))
1844 nilfs_btree_commit_update_v(btree, path, minlevel, dat);
1846 for (level = minlevel + 1; level <= maxlevel; level++)
1847 nilfs_btree_commit_update_v(btree, path, level, dat);
1850 static int nilfs_btree_propagate_v(struct nilfs_bmap *btree,
1851 struct nilfs_btree_path *path,
1852 int level, struct buffer_head *bh)
1854 int maxlevel = 0, ret;
1855 struct nilfs_btree_node *parent;
1856 struct inode *dat = nilfs_bmap_get_dat(btree);
1861 path[level].bp_bh = bh;
1862 ret = nilfs_btree_prepare_propagate_v(btree, path, level, &maxlevel,
1867 if (buffer_nilfs_volatile(path[level].bp_bh)) {
1868 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
1869 ptr = nilfs_btree_node_get_ptr(parent,
1870 path[level + 1].bp_index,
1872 ret = nilfs_dat_mark_dirty(dat, ptr);
1877 nilfs_btree_commit_propagate_v(btree, path, level, maxlevel, bh, dat);
1880 brelse(path[level].bp_bh);
1881 path[level].bp_bh = NULL;
1885 static int nilfs_btree_propagate(struct nilfs_bmap *btree,
1886 struct buffer_head *bh)
1888 struct nilfs_btree_path *path;
1889 struct nilfs_btree_node *node;
1893 WARN_ON(!buffer_dirty(bh));
1895 path = nilfs_btree_alloc_path();
1899 if (buffer_nilfs_node(bh)) {
1900 node = (struct nilfs_btree_node *)bh->b_data;
1901 key = nilfs_btree_node_get_key(node, 0);
1902 level = nilfs_btree_node_get_level(node);
1904 key = nilfs_bmap_data_get_key(btree, bh);
1905 level = NILFS_BTREE_LEVEL_DATA;
1908 ret = nilfs_btree_do_lookup(btree, path, key, NULL, level + 1);
1910 if (unlikely(ret == -ENOENT))
1911 printk(KERN_CRIT "%s: key = %llu, level == %d\n",
1912 __func__, (unsigned long long)key, level);
1916 ret = NILFS_BMAP_USE_VBN(btree) ?
1917 nilfs_btree_propagate_v(btree, path, level, bh) :
1918 nilfs_btree_propagate_p(btree, path, level, bh);
1921 nilfs_btree_free_path(path);
1926 static int nilfs_btree_propagate_gc(struct nilfs_bmap *btree,
1927 struct buffer_head *bh)
1929 return nilfs_dat_mark_dirty(nilfs_bmap_get_dat(btree), bh->b_blocknr);
1932 static void nilfs_btree_add_dirty_buffer(struct nilfs_bmap *btree,
1933 struct list_head *lists,
1934 struct buffer_head *bh)
1936 struct list_head *head;
1937 struct buffer_head *cbh;
1938 struct nilfs_btree_node *node, *cnode;
1943 node = (struct nilfs_btree_node *)bh->b_data;
1944 key = nilfs_btree_node_get_key(node, 0);
1945 level = nilfs_btree_node_get_level(node);
1946 if (level < NILFS_BTREE_LEVEL_NODE_MIN ||
1947 level >= NILFS_BTREE_LEVEL_MAX) {
1950 "%s: invalid btree level: %d (key=%llu, ino=%lu, "
1952 __func__, level, (unsigned long long)key,
1953 NILFS_BMAP_I(btree)->vfs_inode.i_ino,
1954 (unsigned long long)bh->b_blocknr);
1958 list_for_each(head, &lists[level]) {
1959 cbh = list_entry(head, struct buffer_head, b_assoc_buffers);
1960 cnode = (struct nilfs_btree_node *)cbh->b_data;
1961 ckey = nilfs_btree_node_get_key(cnode, 0);
1965 list_add_tail(&bh->b_assoc_buffers, head);
1968 static void nilfs_btree_lookup_dirty_buffers(struct nilfs_bmap *btree,
1969 struct list_head *listp)
1971 struct address_space *btcache = &NILFS_BMAP_I(btree)->i_btnode_cache;
1972 struct list_head lists[NILFS_BTREE_LEVEL_MAX];
1973 struct pagevec pvec;
1974 struct buffer_head *bh, *head;
1978 for (level = NILFS_BTREE_LEVEL_NODE_MIN;
1979 level < NILFS_BTREE_LEVEL_MAX;
1981 INIT_LIST_HEAD(&lists[level]);
1983 pagevec_init(&pvec, 0);
1985 while (pagevec_lookup_tag(&pvec, btcache, &index, PAGECACHE_TAG_DIRTY,
1987 for (i = 0; i < pagevec_count(&pvec); i++) {
1988 bh = head = page_buffers(pvec.pages[i]);
1990 if (buffer_dirty(bh))
1991 nilfs_btree_add_dirty_buffer(btree,
1993 } while ((bh = bh->b_this_page) != head);
1995 pagevec_release(&pvec);
1999 for (level = NILFS_BTREE_LEVEL_NODE_MIN;
2000 level < NILFS_BTREE_LEVEL_MAX;
2002 list_splice_tail(&lists[level], listp);
2005 static int nilfs_btree_assign_p(struct nilfs_bmap *btree,
2006 struct nilfs_btree_path *path,
2008 struct buffer_head **bh,
2010 union nilfs_binfo *binfo)
2012 struct nilfs_btree_node *parent;
2017 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
2018 ptr = nilfs_btree_node_get_ptr(parent, path[level + 1].bp_index,
2020 if (buffer_nilfs_node(*bh)) {
2021 path[level].bp_ctxt.oldkey = ptr;
2022 path[level].bp_ctxt.newkey = blocknr;
2023 path[level].bp_ctxt.bh = *bh;
2024 ret = nilfs_btnode_prepare_change_key(
2025 &NILFS_BMAP_I(btree)->i_btnode_cache,
2026 &path[level].bp_ctxt);
2029 nilfs_btnode_commit_change_key(
2030 &NILFS_BMAP_I(btree)->i_btnode_cache,
2031 &path[level].bp_ctxt);
2032 *bh = path[level].bp_ctxt.bh;
2035 nilfs_btree_node_set_ptr(parent, path[level + 1].bp_index, blocknr,
2038 key = nilfs_btree_node_get_key(parent, path[level + 1].bp_index);
2039 /* on-disk format */
2040 binfo->bi_dat.bi_blkoff = cpu_to_le64(key);
2041 binfo->bi_dat.bi_level = level;
2046 static int nilfs_btree_assign_v(struct nilfs_bmap *btree,
2047 struct nilfs_btree_path *path,
2049 struct buffer_head **bh,
2051 union nilfs_binfo *binfo)
2053 struct nilfs_btree_node *parent;
2054 struct inode *dat = nilfs_bmap_get_dat(btree);
2057 union nilfs_bmap_ptr_req req;
2060 parent = nilfs_btree_get_node(btree, path, level + 1, &ncmax);
2061 ptr = nilfs_btree_node_get_ptr(parent, path[level + 1].bp_index,
2064 ret = nilfs_dat_prepare_start(dat, &req.bpr_req);
2067 nilfs_dat_commit_start(dat, &req.bpr_req, blocknr);
2069 key = nilfs_btree_node_get_key(parent, path[level + 1].bp_index);
2070 /* on-disk format */
2071 binfo->bi_v.bi_vblocknr = cpu_to_le64(ptr);
2072 binfo->bi_v.bi_blkoff = cpu_to_le64(key);
2077 static int nilfs_btree_assign(struct nilfs_bmap *btree,
2078 struct buffer_head **bh,
2080 union nilfs_binfo *binfo)
2082 struct nilfs_btree_path *path;
2083 struct nilfs_btree_node *node;
2087 path = nilfs_btree_alloc_path();
2091 if (buffer_nilfs_node(*bh)) {
2092 node = (struct nilfs_btree_node *)(*bh)->b_data;
2093 key = nilfs_btree_node_get_key(node, 0);
2094 level = nilfs_btree_node_get_level(node);
2096 key = nilfs_bmap_data_get_key(btree, *bh);
2097 level = NILFS_BTREE_LEVEL_DATA;
2100 ret = nilfs_btree_do_lookup(btree, path, key, NULL, level + 1);
2102 WARN_ON(ret == -ENOENT);
2106 ret = NILFS_BMAP_USE_VBN(btree) ?
2107 nilfs_btree_assign_v(btree, path, level, bh, blocknr, binfo) :
2108 nilfs_btree_assign_p(btree, path, level, bh, blocknr, binfo);
2111 nilfs_btree_free_path(path);
2116 static int nilfs_btree_assign_gc(struct nilfs_bmap *btree,
2117 struct buffer_head **bh,
2119 union nilfs_binfo *binfo)
2121 struct nilfs_btree_node *node;
2125 ret = nilfs_dat_move(nilfs_bmap_get_dat(btree), (*bh)->b_blocknr,
2130 if (buffer_nilfs_node(*bh)) {
2131 node = (struct nilfs_btree_node *)(*bh)->b_data;
2132 key = nilfs_btree_node_get_key(node, 0);
2134 key = nilfs_bmap_data_get_key(btree, *bh);
2136 /* on-disk format */
2137 binfo->bi_v.bi_vblocknr = cpu_to_le64((*bh)->b_blocknr);
2138 binfo->bi_v.bi_blkoff = cpu_to_le64(key);
2143 static int nilfs_btree_mark(struct nilfs_bmap *btree, __u64 key, int level)
2145 struct buffer_head *bh;
2146 struct nilfs_btree_path *path;
2150 path = nilfs_btree_alloc_path();
2154 ret = nilfs_btree_do_lookup(btree, path, key, &ptr, level + 1);
2156 WARN_ON(ret == -ENOENT);
2159 ret = nilfs_btree_get_block(btree, ptr, &bh);
2161 WARN_ON(ret == -ENOENT);
2165 if (!buffer_dirty(bh))
2166 nilfs_btnode_mark_dirty(bh);
2168 if (!nilfs_bmap_dirty(btree))
2169 nilfs_bmap_set_dirty(btree);
2172 nilfs_btree_free_path(path);
2176 static const struct nilfs_bmap_operations nilfs_btree_ops = {
2177 .bop_lookup = nilfs_btree_lookup,
2178 .bop_lookup_contig = nilfs_btree_lookup_contig,
2179 .bop_insert = nilfs_btree_insert,
2180 .bop_delete = nilfs_btree_delete,
2183 .bop_propagate = nilfs_btree_propagate,
2185 .bop_lookup_dirty_buffers = nilfs_btree_lookup_dirty_buffers,
2187 .bop_assign = nilfs_btree_assign,
2188 .bop_mark = nilfs_btree_mark,
2190 .bop_last_key = nilfs_btree_last_key,
2191 .bop_check_insert = NULL,
2192 .bop_check_delete = nilfs_btree_check_delete,
2193 .bop_gather_data = nilfs_btree_gather_data,
2196 static const struct nilfs_bmap_operations nilfs_btree_ops_gc = {
2198 .bop_lookup_contig = NULL,
2203 .bop_propagate = nilfs_btree_propagate_gc,
2205 .bop_lookup_dirty_buffers = nilfs_btree_lookup_dirty_buffers,
2207 .bop_assign = nilfs_btree_assign_gc,
2210 .bop_last_key = NULL,
2211 .bop_check_insert = NULL,
2212 .bop_check_delete = NULL,
2213 .bop_gather_data = NULL,
2216 int nilfs_btree_init(struct nilfs_bmap *bmap)
2218 bmap->b_ops = &nilfs_btree_ops;
2219 bmap->b_nchildren_per_block =
2220 NILFS_BTREE_NODE_NCHILDREN_MAX(nilfs_btree_node_size(bmap));
2224 void nilfs_btree_init_gc(struct nilfs_bmap *bmap)
2226 bmap->b_ops = &nilfs_btree_ops_gc;
2227 bmap->b_nchildren_per_block =
2228 NILFS_BTREE_NODE_NCHILDREN_MAX(nilfs_btree_node_size(bmap));