2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
5 #include <linux/config.h>
6 #include <asm/uaccess.h>
7 #include <linux/string.h>
8 #include <linux/time.h>
9 #include <linux/reiserfs_fs.h>
10 #include <linux/buffer_head.h>
12 /* this is one and only function that is used outside (do_balance.c) */
13 int balance_internal (
14 struct tree_balance * ,
21 /* modes of internal_shift_left, internal_shift_right and internal_insert_childs */
22 #define INTERNAL_SHIFT_FROM_S_TO_L 0
23 #define INTERNAL_SHIFT_FROM_R_TO_S 1
24 #define INTERNAL_SHIFT_FROM_L_TO_S 2
25 #define INTERNAL_SHIFT_FROM_S_TO_R 3
26 #define INTERNAL_INSERT_TO_S 4
27 #define INTERNAL_INSERT_TO_L 5
28 #define INTERNAL_INSERT_TO_R 6
30 static void internal_define_dest_src_infos (
32 struct tree_balance * tb,
34 struct buffer_info * dest_bi,
35 struct buffer_info * src_bi,
37 struct buffer_head ** cf
40 memset (dest_bi, 0, sizeof (struct buffer_info));
41 memset (src_bi, 0, sizeof (struct buffer_info));
42 /* define dest, src, dest parent, dest position */
44 case INTERNAL_SHIFT_FROM_S_TO_L: /* used in internal_shift_left */
46 src_bi->bi_bh = PATH_H_PBUFFER (tb->tb_path, h);
47 src_bi->bi_parent = PATH_H_PPARENT (tb->tb_path, h);
48 src_bi->bi_position = PATH_H_POSITION (tb->tb_path, h + 1);
50 dest_bi->bi_bh = tb->L[h];
51 dest_bi->bi_parent = tb->FL[h];
52 dest_bi->bi_position = get_left_neighbor_position (tb, h);
56 case INTERNAL_SHIFT_FROM_L_TO_S:
58 src_bi->bi_bh = tb->L[h];
59 src_bi->bi_parent = tb->FL[h];
60 src_bi->bi_position = get_left_neighbor_position (tb, h);
62 dest_bi->bi_bh = PATH_H_PBUFFER (tb->tb_path, h);
63 dest_bi->bi_parent = PATH_H_PPARENT (tb->tb_path, h);
64 dest_bi->bi_position = PATH_H_POSITION (tb->tb_path, h + 1); /* dest position is analog of dest->b_item_order */
69 case INTERNAL_SHIFT_FROM_R_TO_S: /* used in internal_shift_left */
71 src_bi->bi_bh = tb->R[h];
72 src_bi->bi_parent = tb->FR[h];
73 src_bi->bi_position = get_right_neighbor_position (tb, h);
75 dest_bi->bi_bh = PATH_H_PBUFFER (tb->tb_path, h);
76 dest_bi->bi_parent = PATH_H_PPARENT (tb->tb_path, h);
77 dest_bi->bi_position = PATH_H_POSITION (tb->tb_path, h + 1);
82 case INTERNAL_SHIFT_FROM_S_TO_R:
84 src_bi->bi_bh = PATH_H_PBUFFER (tb->tb_path, h);
85 src_bi->bi_parent = PATH_H_PPARENT (tb->tb_path, h);
86 src_bi->bi_position = PATH_H_POSITION (tb->tb_path, h + 1);
88 dest_bi->bi_bh = tb->R[h];
89 dest_bi->bi_parent = tb->FR[h];
90 dest_bi->bi_position = get_right_neighbor_position (tb, h);
95 case INTERNAL_INSERT_TO_L:
97 dest_bi->bi_bh = tb->L[h];
98 dest_bi->bi_parent = tb->FL[h];
99 dest_bi->bi_position = get_left_neighbor_position (tb, h);
102 case INTERNAL_INSERT_TO_S:
104 dest_bi->bi_bh = PATH_H_PBUFFER (tb->tb_path, h);
105 dest_bi->bi_parent = PATH_H_PPARENT (tb->tb_path, h);
106 dest_bi->bi_position = PATH_H_POSITION (tb->tb_path, h + 1);
109 case INTERNAL_INSERT_TO_R:
111 dest_bi->bi_bh = tb->R[h];
112 dest_bi->bi_parent = tb->FR[h];
113 dest_bi->bi_position = get_right_neighbor_position (tb, h);
117 reiserfs_panic (tb->tb_sb, "internal_define_dest_src_infos: shift type is unknown (%d)", shift_mode);
123 /* Insert count node pointers into buffer cur before position to + 1.
124 * Insert count items into buffer cur before position to.
125 * Items and node pointers are specified by inserted and bh respectively.
127 static void internal_insert_childs (struct buffer_info * cur_bi,
129 struct item_head * inserted,
130 struct buffer_head ** bh
133 struct buffer_head * cur = cur_bi->bi_bh;
134 struct block_head * blkh;
136 struct reiserfs_key * ih;
137 struct disk_child new_dc[2];
138 struct disk_child * dc;
144 blkh = B_BLK_HEAD(cur);
145 nr = blkh_nr_item(blkh);
148 "too many children (%d) are to be inserted", count);
149 RFALSE( B_FREE_SPACE (cur) < count * (KEY_SIZE + DC_SIZE),
150 "no enough free space (%d), needed %d bytes",
151 B_FREE_SPACE (cur), count * (KEY_SIZE + DC_SIZE));
153 /* prepare space for count disk_child */
154 dc = B_N_CHILD(cur,to+1);
156 memmove (dc + count, dc, (nr+1-(to+1)) * DC_SIZE);
158 /* copy to_be_insert disk children */
159 for (i = 0; i < count; i ++) {
160 put_dc_size( &(new_dc[i]), MAX_CHILD_SIZE(bh[i]) - B_FREE_SPACE(bh[i]));
161 put_dc_block_number( &(new_dc[i]), bh[i]->b_blocknr );
163 memcpy (dc, new_dc, DC_SIZE * count);
166 /* prepare space for count items */
167 ih = B_N_PDELIM_KEY (cur, ((to == -1) ? 0 : to));
169 memmove (ih + count, ih, (nr - to) * KEY_SIZE + (nr + 1 + count) * DC_SIZE);
171 /* copy item headers (keys) */
172 memcpy (ih, inserted, KEY_SIZE);
174 memcpy (ih + 1, inserted + 1, KEY_SIZE);
176 /* sizes, item number */
177 set_blkh_nr_item( blkh, blkh_nr_item(blkh) + count );
178 set_blkh_free_space( blkh,
179 blkh_free_space(blkh) - count * (DC_SIZE + KEY_SIZE ) );
181 do_balance_mark_internal_dirty (cur_bi->tb, cur,0);
183 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
184 check_internal (cur);
185 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
187 if (cur_bi->bi_parent) {
188 struct disk_child *t_dc = B_N_CHILD (cur_bi->bi_parent,cur_bi->bi_position);
189 put_dc_size( t_dc, dc_size(t_dc) + (count * (DC_SIZE + KEY_SIZE)));
190 do_balance_mark_internal_dirty(cur_bi->tb, cur_bi->bi_parent, 0);
192 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
193 check_internal (cur_bi->bi_parent);
194 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
200 /* Delete del_num items and node pointers from buffer cur starting from *
201 * the first_i'th item and first_p'th pointers respectively. */
202 static void internal_delete_pointers_items (
203 struct buffer_info * cur_bi,
209 struct buffer_head * cur = cur_bi->bi_bh;
211 struct block_head * blkh;
212 struct reiserfs_key * key;
213 struct disk_child * dc;
215 RFALSE( cur == NULL, "buffer is 0");
217 "negative number of items (%d) can not be deleted", del_num);
218 RFALSE( first_p < 0 || first_p + del_num > B_NR_ITEMS (cur) + 1 || first_i < 0,
219 "first pointer order (%d) < 0 or "
220 "no so many pointers (%d), only (%d) or "
221 "first key order %d < 0", first_p,
222 first_p + del_num, B_NR_ITEMS (cur) + 1, first_i);
226 blkh = B_BLK_HEAD(cur);
227 nr = blkh_nr_item(blkh);
229 if ( first_p == 0 && del_num == nr + 1 ) {
230 RFALSE( first_i != 0, "1st deleted key must have order 0, not %d", first_i);
231 make_empty_node (cur_bi);
235 RFALSE( first_i + del_num > B_NR_ITEMS (cur),
236 "first_i = %d del_num = %d "
237 "no so many keys (%d) in the node (%b)(%z)",
238 first_i, del_num, first_i + del_num, cur, cur);
242 dc = B_N_CHILD (cur, first_p);
244 memmove (dc, dc + del_num, (nr + 1 - first_p - del_num) * DC_SIZE);
245 key = B_N_PDELIM_KEY (cur, first_i);
246 memmove (key, key + del_num, (nr - first_i - del_num) * KEY_SIZE + (nr + 1 - del_num) * DC_SIZE);
249 /* sizes, item number */
250 set_blkh_nr_item( blkh, blkh_nr_item(blkh) - del_num );
251 set_blkh_free_space( blkh,
252 blkh_free_space(blkh) + (del_num * (KEY_SIZE + DC_SIZE) ) );
254 do_balance_mark_internal_dirty (cur_bi->tb, cur, 0);
255 /*&&&&&&&&&&&&&&&&&&&&&&&*/
256 check_internal (cur);
257 /*&&&&&&&&&&&&&&&&&&&&&&&*/
259 if (cur_bi->bi_parent) {
260 struct disk_child *t_dc;
261 t_dc = B_N_CHILD (cur_bi->bi_parent, cur_bi->bi_position);
262 put_dc_size( t_dc, dc_size(t_dc) - (del_num * (KEY_SIZE + DC_SIZE) ) );
264 do_balance_mark_internal_dirty (cur_bi->tb, cur_bi->bi_parent,0);
265 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
266 check_internal (cur_bi->bi_parent);
267 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
272 /* delete n node pointers and items starting from given position */
273 static void internal_delete_childs (struct buffer_info * cur_bi,
278 i_from = (from == 0) ? from : from - 1;
280 /* delete n pointers starting from `from' position in CUR;
281 delete n keys starting from 'i_from' position in CUR;
283 internal_delete_pointers_items (cur_bi, from, i_from, n);
287 /* copy cpy_num node pointers and cpy_num - 1 items from buffer src to buffer dest
288 * last_first == FIRST_TO_LAST means, that we copy first items from src to tail of dest
289 * last_first == LAST_TO_FIRST means, that we copy last items from src to head of dest
291 static void internal_copy_pointers_items (
292 struct buffer_info * dest_bi,
293 struct buffer_head * src,
294 int last_first, int cpy_num
297 /* ATTENTION! Number of node pointers in DEST is equal to number of items in DEST *
298 * as delimiting key have already inserted to buffer dest.*/
299 struct buffer_head * dest = dest_bi->bi_bh;
301 int dest_order, src_order;
302 struct block_head * blkh;
303 struct reiserfs_key * key;
304 struct disk_child * dc;
306 nr_src = B_NR_ITEMS (src);
308 RFALSE( dest == NULL || src == NULL,
309 "src (%p) or dest (%p) buffer is 0", src, dest);
310 RFALSE( last_first != FIRST_TO_LAST && last_first != LAST_TO_FIRST,
311 "invalid last_first parameter (%d)", last_first);
312 RFALSE( nr_src < cpy_num - 1,
313 "no so many items (%d) in src (%d)", cpy_num, nr_src);
314 RFALSE( cpy_num < 0, "cpy_num less than 0 (%d)", cpy_num);
315 RFALSE( cpy_num - 1 + B_NR_ITEMS(dest) > (int)MAX_NR_KEY(dest),
316 "cpy_num (%d) + item number in dest (%d) can not be > MAX_NR_KEY(%d)",
317 cpy_num, B_NR_ITEMS(dest), MAX_NR_KEY(dest));
323 blkh = B_BLK_HEAD(dest);
324 nr_dest = blkh_nr_item(blkh);
326 /*dest_order = (last_first == LAST_TO_FIRST) ? 0 : nr_dest;*/
327 /*src_order = (last_first == LAST_TO_FIRST) ? (nr_src - cpy_num + 1) : 0;*/
328 (last_first == LAST_TO_FIRST) ? (dest_order = 0, src_order = nr_src - cpy_num + 1) :
329 (dest_order = nr_dest, src_order = 0);
331 /* prepare space for cpy_num pointers */
332 dc = B_N_CHILD (dest, dest_order);
334 memmove (dc + cpy_num, dc, (nr_dest - dest_order) * DC_SIZE);
336 /* insert pointers */
337 memcpy (dc, B_N_CHILD (src, src_order), DC_SIZE * cpy_num);
340 /* prepare space for cpy_num - 1 item headers */
341 key = B_N_PDELIM_KEY(dest, dest_order);
342 memmove (key + cpy_num - 1, key,
343 KEY_SIZE * (nr_dest - dest_order) + DC_SIZE * (nr_dest + cpy_num));
347 memcpy (key, B_N_PDELIM_KEY (src, src_order), KEY_SIZE * (cpy_num - 1));
349 /* sizes, item number */
350 set_blkh_nr_item( blkh, blkh_nr_item(blkh) + (cpy_num - 1 ) );
351 set_blkh_free_space( blkh,
352 blkh_free_space(blkh) - (KEY_SIZE * (cpy_num - 1) + DC_SIZE * cpy_num ) );
354 do_balance_mark_internal_dirty (dest_bi->tb, dest, 0);
356 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
357 check_internal (dest);
358 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
360 if (dest_bi->bi_parent) {
361 struct disk_child *t_dc;
362 t_dc = B_N_CHILD(dest_bi->bi_parent,dest_bi->bi_position);
363 put_dc_size( t_dc, dc_size(t_dc) + (KEY_SIZE * (cpy_num - 1) + DC_SIZE * cpy_num) );
365 do_balance_mark_internal_dirty (dest_bi->tb, dest_bi->bi_parent,0);
366 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
367 check_internal (dest_bi->bi_parent);
368 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
374 /* Copy cpy_num node pointers and cpy_num - 1 items from buffer src to buffer dest.
375 * Delete cpy_num - del_par items and node pointers from buffer src.
376 * last_first == FIRST_TO_LAST means, that we copy/delete first items from src.
377 * last_first == LAST_TO_FIRST means, that we copy/delete last items from src.
379 static void internal_move_pointers_items (struct buffer_info * dest_bi,
380 struct buffer_info * src_bi,
381 int last_first, int cpy_num, int del_par)
386 internal_copy_pointers_items (dest_bi, src_bi->bi_bh, last_first, cpy_num);
388 if (last_first == FIRST_TO_LAST) { /* shift_left occurs */
391 /* delete cpy_num - del_par pointers and keys starting for pointers with first_pointer,
392 for key - with first_item */
393 internal_delete_pointers_items (src_bi, first_pointer, first_item, cpy_num - del_par);
394 } else { /* shift_right occurs */
397 i = ( cpy_num - del_par == ( j = B_NR_ITEMS(src_bi->bi_bh)) + 1 ) ? 0 : j - cpy_num + del_par;
399 internal_delete_pointers_items (src_bi, j + 1 - cpy_num + del_par, i, cpy_num - del_par);
403 /* Insert n_src'th key of buffer src before n_dest'th key of buffer dest. */
404 static void internal_insert_key (struct buffer_info * dest_bi,
405 int dest_position_before, /* insert key before key with n_dest number */
406 struct buffer_head * src,
409 struct buffer_head * dest = dest_bi->bi_bh;
411 struct block_head * blkh;
412 struct reiserfs_key * key;
414 RFALSE( dest == NULL || src == NULL,
415 "source(%p) or dest(%p) buffer is 0", src, dest);
416 RFALSE( dest_position_before < 0 || src_position < 0,
417 "source(%d) or dest(%d) key number less than 0",
418 src_position, dest_position_before);
419 RFALSE( dest_position_before > B_NR_ITEMS (dest) ||
420 src_position >= B_NR_ITEMS(src),
421 "invalid position in dest (%d (key number %d)) or in src (%d (key number %d))",
422 dest_position_before, B_NR_ITEMS (dest),
423 src_position, B_NR_ITEMS(src));
424 RFALSE( B_FREE_SPACE (dest) < KEY_SIZE,
425 "no enough free space (%d) in dest buffer", B_FREE_SPACE (dest));
427 blkh = B_BLK_HEAD(dest);
428 nr = blkh_nr_item(blkh);
430 /* prepare space for inserting key */
431 key = B_N_PDELIM_KEY (dest, dest_position_before);
432 memmove (key + 1, key, (nr - dest_position_before) * KEY_SIZE + (nr + 1) * DC_SIZE);
435 memcpy (key, B_N_PDELIM_KEY(src, src_position), KEY_SIZE);
437 /* Change dirt, free space, item number fields. */
439 set_blkh_nr_item( blkh, blkh_nr_item(blkh) + 1 );
440 set_blkh_free_space( blkh, blkh_free_space(blkh) - KEY_SIZE );
442 do_balance_mark_internal_dirty (dest_bi->tb, dest, 0);
444 if (dest_bi->bi_parent) {
445 struct disk_child *t_dc;
446 t_dc = B_N_CHILD(dest_bi->bi_parent,dest_bi->bi_position);
447 put_dc_size( t_dc, dc_size(t_dc) + KEY_SIZE );
449 do_balance_mark_internal_dirty (dest_bi->tb, dest_bi->bi_parent,0);
455 /* Insert d_key'th (delimiting) key from buffer cfl to tail of dest.
456 * Copy pointer_amount node pointers and pointer_amount - 1 items from buffer src to buffer dest.
457 * Replace d_key'th key in buffer cfl.
458 * Delete pointer_amount items and node pointers from buffer src.
460 /* this can be invoked both to shift from S to L and from R to S */
461 static void internal_shift_left (
462 int mode, /* INTERNAL_FROM_S_TO_L | INTERNAL_FROM_R_TO_S */
463 struct tree_balance * tb,
468 struct buffer_info dest_bi, src_bi;
469 struct buffer_head * cf;
472 internal_define_dest_src_infos (mode, tb, h, &dest_bi, &src_bi, &d_key_position, &cf);
474 /*printk("pointer_amount = %d\n",pointer_amount);*/
476 if (pointer_amount) {
477 /* insert delimiting key from common father of dest and src to node dest into position B_NR_ITEM(dest) */
478 internal_insert_key (&dest_bi, B_NR_ITEMS(dest_bi.bi_bh), cf, d_key_position);
480 if (B_NR_ITEMS(src_bi.bi_bh) == pointer_amount - 1) {
481 if (src_bi.bi_position/*src->b_item_order*/ == 0)
482 replace_key (tb, cf, d_key_position, src_bi.bi_parent/*src->b_parent*/, 0);
484 replace_key (tb, cf, d_key_position, src_bi.bi_bh, pointer_amount - 1);
486 /* last parameter is del_parameter */
487 internal_move_pointers_items (&dest_bi, &src_bi, FIRST_TO_LAST, pointer_amount, 0);
491 /* Insert delimiting key to L[h].
492 * Copy n node pointers and n - 1 items from buffer S[h] to L[h].
493 * Delete n - 1 items and node pointers from buffer S[h].
495 /* it always shifts from S[h] to L[h] */
496 static void internal_shift1_left (
497 struct tree_balance * tb,
502 struct buffer_info dest_bi, src_bi;
503 struct buffer_head * cf;
506 internal_define_dest_src_infos (INTERNAL_SHIFT_FROM_S_TO_L, tb, h, &dest_bi, &src_bi, &d_key_position, &cf);
508 if ( pointer_amount > 0 ) /* insert lkey[h]-th key from CFL[h] to left neighbor L[h] */
509 internal_insert_key (&dest_bi, B_NR_ITEMS(dest_bi.bi_bh), cf, d_key_position);
510 /* internal_insert_key (tb->L[h], B_NR_ITEM(tb->L[h]), tb->CFL[h], tb->lkey[h]);*/
512 /* last parameter is del_parameter */
513 internal_move_pointers_items (&dest_bi, &src_bi, FIRST_TO_LAST, pointer_amount, 1);
514 /* internal_move_pointers_items (tb->L[h], tb->S[h], FIRST_TO_LAST, pointer_amount, 1);*/
518 /* Insert d_key'th (delimiting) key from buffer cfr to head of dest.
519 * Copy n node pointers and n - 1 items from buffer src to buffer dest.
520 * Replace d_key'th key in buffer cfr.
521 * Delete n items and node pointers from buffer src.
523 static void internal_shift_right (
524 int mode, /* INTERNAL_FROM_S_TO_R | INTERNAL_FROM_L_TO_S */
525 struct tree_balance * tb,
530 struct buffer_info dest_bi, src_bi;
531 struct buffer_head * cf;
536 internal_define_dest_src_infos (mode, tb, h, &dest_bi, &src_bi, &d_key_position, &cf);
538 nr = B_NR_ITEMS (src_bi.bi_bh);
540 if (pointer_amount > 0) {
541 /* insert delimiting key from common father of dest and src to dest node into position 0 */
542 internal_insert_key (&dest_bi, 0, cf, d_key_position);
543 if (nr == pointer_amount - 1) {
544 RFALSE( src_bi.bi_bh != PATH_H_PBUFFER (tb->tb_path, h)/*tb->S[h]*/ ||
545 dest_bi.bi_bh != tb->R[h],
546 "src (%p) must be == tb->S[h](%p) when it disappears",
547 src_bi.bi_bh, PATH_H_PBUFFER (tb->tb_path, h));
548 /* when S[h] disappers replace left delemiting key as well */
550 replace_key (tb, cf, d_key_position, tb->CFL[h], tb->lkey[h]);
552 replace_key (tb, cf, d_key_position, src_bi.bi_bh, nr - pointer_amount);
555 /* last parameter is del_parameter */
556 internal_move_pointers_items (&dest_bi, &src_bi, LAST_TO_FIRST, pointer_amount, 0);
559 /* Insert delimiting key to R[h].
560 * Copy n node pointers and n - 1 items from buffer S[h] to R[h].
561 * Delete n - 1 items and node pointers from buffer S[h].
563 /* it always shift from S[h] to R[h] */
564 static void internal_shift1_right (
565 struct tree_balance * tb,
570 struct buffer_info dest_bi, src_bi;
571 struct buffer_head * cf;
574 internal_define_dest_src_infos (INTERNAL_SHIFT_FROM_S_TO_R, tb, h, &dest_bi, &src_bi, &d_key_position, &cf);
576 if (pointer_amount > 0) /* insert rkey from CFR[h] to right neighbor R[h] */
577 internal_insert_key (&dest_bi, 0, cf, d_key_position);
578 /* internal_insert_key (tb->R[h], 0, tb->CFR[h], tb->rkey[h]);*/
580 /* last parameter is del_parameter */
581 internal_move_pointers_items (&dest_bi, &src_bi, LAST_TO_FIRST, pointer_amount, 1);
582 /* internal_move_pointers_items (tb->R[h], tb->S[h], LAST_TO_FIRST, pointer_amount, 1);*/
586 /* Delete insert_num node pointers together with their left items
587 * and balance current node.*/
588 static void balance_internal_when_delete (struct tree_balance * tb,
589 int h, int child_pos)
593 struct buffer_head * tbSh = PATH_H_PBUFFER (tb->tb_path, h);
594 struct buffer_info bi;
596 insert_num = tb->insert_size[h] / ((int)(DC_SIZE + KEY_SIZE));
598 /* delete child-node-pointer(s) together with their left item(s) */
601 bi.bi_parent = PATH_H_PPARENT (tb->tb_path, h);
602 bi.bi_position = PATH_H_POSITION (tb->tb_path, h + 1);
604 internal_delete_childs (&bi, child_pos, -insert_num);
606 RFALSE( tb->blknum[h] > 1,
607 "tb->blknum[%d]=%d when insert_size < 0", h, tb->blknum[h]);
609 n = B_NR_ITEMS(tbSh);
611 if ( tb->lnum[h] == 0 && tb->rnum[h] == 0 ) {
612 if ( tb->blknum[h] == 0 ) {
613 /* node S[h] (root of the tree) is empty now */
614 struct buffer_head *new_root;
616 RFALSE( n || B_FREE_SPACE (tbSh) != MAX_CHILD_SIZE(tbSh) - DC_SIZE,
617 "buffer must have only 0 keys (%d)", n);
618 RFALSE( bi.bi_parent, "root has parent (%p)", bi.bi_parent);
620 /* choose a new root */
621 if ( ! tb->L[h-1] || ! B_NR_ITEMS(tb->L[h-1]) )
622 new_root = tb->R[h-1];
624 new_root = tb->L[h-1];
625 /* switch super block's tree root block number to the new value */
626 PUT_SB_ROOT_BLOCK( tb->tb_sb, new_root->b_blocknr );
627 //REISERFS_SB(tb->tb_sb)->s_rs->s_tree_height --;
628 PUT_SB_TREE_HEIGHT( tb->tb_sb, SB_TREE_HEIGHT(tb->tb_sb) - 1 );
630 do_balance_mark_sb_dirty (tb, REISERFS_SB(tb->tb_sb)->s_sbh, 1);
631 /*&&&&&&&&&&&&&&&&&&&&&&*/
633 /* use check_internal if new root is an internal node */
634 check_internal (new_root);
635 /*&&&&&&&&&&&&&&&&&&&&&&*/
637 /* do what is needed for buffer thrown from tree */
638 reiserfs_invalidate_buffer(tb, tbSh);
644 if ( tb->L[h] && tb->lnum[h] == -B_NR_ITEMS(tb->L[h]) - 1 ) { /* join S[h] with L[h] */
646 RFALSE( tb->rnum[h] != 0,
647 "invalid tb->rnum[%d]==%d when joining S[h] with L[h]",
650 internal_shift_left (INTERNAL_SHIFT_FROM_S_TO_L, tb, h, n + 1);
651 reiserfs_invalidate_buffer(tb, tbSh);
656 if ( tb->R[h] && tb->rnum[h] == -B_NR_ITEMS(tb->R[h]) - 1 ) { /* join S[h] with R[h] */
657 RFALSE( tb->lnum[h] != 0,
658 "invalid tb->lnum[%d]==%d when joining S[h] with R[h]",
661 internal_shift_right (INTERNAL_SHIFT_FROM_S_TO_R, tb, h, n + 1);
663 reiserfs_invalidate_buffer(tb,tbSh);
667 if ( tb->lnum[h] < 0 ) { /* borrow from left neighbor L[h] */
668 RFALSE( tb->rnum[h] != 0,
669 "wrong tb->rnum[%d]==%d when borrow from L[h]", h, tb->rnum[h]);
670 /*internal_shift_right (tb, h, tb->L[h], tb->CFL[h], tb->lkey[h], tb->S[h], -tb->lnum[h]);*/
671 internal_shift_right (INTERNAL_SHIFT_FROM_L_TO_S, tb, h, -tb->lnum[h]);
675 if ( tb->rnum[h] < 0 ) { /* borrow from right neighbor R[h] */
676 RFALSE( tb->lnum[h] != 0,
677 "invalid tb->lnum[%d]==%d when borrow from R[h]",
679 internal_shift_left (INTERNAL_SHIFT_FROM_R_TO_S, tb, h, -tb->rnum[h]);/*tb->S[h], tb->CFR[h], tb->rkey[h], tb->R[h], -tb->rnum[h]);*/
683 if ( tb->lnum[h] > 0 ) { /* split S[h] into two parts and put them into neighbors */
684 RFALSE( tb->rnum[h] == 0 || tb->lnum[h] + tb->rnum[h] != n + 1,
685 "invalid tb->lnum[%d]==%d or tb->rnum[%d]==%d when S[h](item number == %d) is split between them",
686 h, tb->lnum[h], h, tb->rnum[h], n);
688 internal_shift_left (INTERNAL_SHIFT_FROM_S_TO_L, tb, h, tb->lnum[h]);/*tb->L[h], tb->CFL[h], tb->lkey[h], tb->S[h], tb->lnum[h]);*/
689 internal_shift_right (INTERNAL_SHIFT_FROM_S_TO_R, tb, h, tb->rnum[h]);
691 reiserfs_invalidate_buffer (tb, tbSh);
695 reiserfs_panic (tb->tb_sb, "balance_internal_when_delete: unexpected tb->lnum[%d]==%d or tb->rnum[%d]==%d",
696 h, tb->lnum[h], h, tb->rnum[h]);
700 /* Replace delimiting key of buffers L[h] and S[h] by the given key.*/
701 static void replace_lkey (
702 struct tree_balance * tb,
704 struct item_head * key
707 RFALSE( tb->L[h] == NULL || tb->CFL[h] == NULL,
708 "L[h](%p) and CFL[h](%p) must exist in replace_lkey",
709 tb->L[h], tb->CFL[h]);
711 if (B_NR_ITEMS(PATH_H_PBUFFER(tb->tb_path, h)) == 0)
714 memcpy (B_N_PDELIM_KEY(tb->CFL[h],tb->lkey[h]), key, KEY_SIZE);
716 do_balance_mark_internal_dirty (tb, tb->CFL[h],0);
720 /* Replace delimiting key of buffers S[h] and R[h] by the given key.*/
721 static void replace_rkey (
722 struct tree_balance * tb,
724 struct item_head * key
727 RFALSE( tb->R[h] == NULL || tb->CFR[h] == NULL,
728 "R[h](%p) and CFR[h](%p) must exist in replace_rkey",
729 tb->R[h], tb->CFR[h]);
730 RFALSE( B_NR_ITEMS(tb->R[h]) == 0,
731 "R[h] can not be empty if it exists (item number=%d)",
732 B_NR_ITEMS(tb->R[h]));
734 memcpy (B_N_PDELIM_KEY(tb->CFR[h],tb->rkey[h]), key, KEY_SIZE);
736 do_balance_mark_internal_dirty (tb, tb->CFR[h], 0);
740 int balance_internal (struct tree_balance * tb, /* tree_balance structure */
741 int h, /* level of the tree */
743 struct item_head * insert_key, /* key for insertion on higher level */
744 struct buffer_head ** insert_ptr /* node for insertion on higher level*/
746 /* if inserting/pasting
748 child_pos is the position of the node-pointer in S[h] that *
749 pointed to S[h-1] before balancing of the h-1 level; *
750 this means that new pointers and items must be inserted AFTER *
755 it is the position of the leftmost pointer that must be deleted (together with
756 its corresponding key to the left of the pointer)
757 as a result of the previous level's balancing.
761 struct buffer_head * tbSh = PATH_H_PBUFFER (tb->tb_path, h);
762 struct buffer_info bi;
763 int order; /* we return this: it is 0 if there is no S[h], else it is tb->S[h]->b_item_order */
764 int insert_num, n, k;
765 struct buffer_head * S_new;
766 struct item_head new_insert_key;
767 struct buffer_head * new_insert_ptr = NULL;
768 struct item_head * new_insert_key_addr = insert_key;
770 RFALSE( h < 1, "h (%d) can not be < 1 on internal level", h);
772 PROC_INFO_INC( tb -> tb_sb, balance_at[ h ] );
774 order = ( tbSh ) ? PATH_H_POSITION (tb->tb_path, h + 1)/*tb->S[h]->b_item_order*/ : 0;
776 /* Using insert_size[h] calculate the number insert_num of items
777 that must be inserted to or deleted from S[h]. */
778 insert_num = tb->insert_size[h]/((int)(KEY_SIZE + DC_SIZE));
780 /* Check whether insert_num is proper **/
781 RFALSE( insert_num < -2 || insert_num > 2,
782 "incorrect number of items inserted to the internal node (%d)",
784 RFALSE( h > 1 && (insert_num > 1 || insert_num < -1),
785 "incorrect number of items (%d) inserted to the internal node on a level (h=%d) higher than last internal level",
788 /* Make balance in case insert_num < 0 */
789 if ( insert_num < 0 ) {
790 balance_internal_when_delete (tb, h, child_pos);
795 if ( tb->lnum[h] > 0 ) {
796 /* shift lnum[h] items from S[h] to the left neighbor L[h].
797 check how many of new items fall into L[h] or CFL[h] after
799 n = B_NR_ITEMS (tb->L[h]); /* number of items in L[h] */
800 if ( tb->lnum[h] <= child_pos ) {
801 /* new items don't fall into L[h] or CFL[h] */
802 internal_shift_left (INTERNAL_SHIFT_FROM_S_TO_L, tb, h, tb->lnum[h]);
803 /*internal_shift_left (tb->L[h],tb->CFL[h],tb->lkey[h],tbSh,tb->lnum[h]);*/
804 child_pos -= tb->lnum[h];
805 } else if ( tb->lnum[h] > child_pos + insert_num ) {
806 /* all new items fall into L[h] */
807 internal_shift_left (INTERNAL_SHIFT_FROM_S_TO_L, tb, h, tb->lnum[h] - insert_num);
808 /* internal_shift_left(tb->L[h],tb->CFL[h],tb->lkey[h],tbSh,
809 tb->lnum[h]-insert_num);
811 /* insert insert_num keys and node-pointers into L[h] */
814 bi.bi_parent = tb->FL[h];
815 bi.bi_position = get_left_neighbor_position (tb, h);
816 internal_insert_childs (&bi,/*tb->L[h], tb->S[h-1]->b_next*/ n + child_pos + 1,
817 insert_num,insert_key,insert_ptr);
821 struct disk_child * dc;
823 /* some items fall into L[h] or CFL[h], but some don't fall */
824 internal_shift1_left(tb,h,child_pos+1);
825 /* calculate number of new items that fall into L[h] */
826 k = tb->lnum[h] - child_pos - 1;
829 bi.bi_parent = tb->FL[h];
830 bi.bi_position = get_left_neighbor_position (tb, h);
831 internal_insert_childs (&bi,/*tb->L[h], tb->S[h-1]->b_next,*/ n + child_pos + 1,k,
832 insert_key,insert_ptr);
834 replace_lkey(tb,h,insert_key + k);
836 /* replace the first node-ptr in S[h] by node-ptr to insert_ptr[k] */
837 dc = B_N_CHILD(tbSh, 0);
838 put_dc_size( dc, MAX_CHILD_SIZE(insert_ptr[k]) - B_FREE_SPACE (insert_ptr[k]));
839 put_dc_block_number( dc, insert_ptr[k]->b_blocknr );
841 do_balance_mark_internal_dirty (tb, tbSh, 0);
849 } /* tb->lnum[h] > 0 */
851 if ( tb->rnum[h] > 0 ) {
852 /*shift rnum[h] items from S[h] to the right neighbor R[h]*/
853 /* check how many of new items fall into R or CFR after shifting */
854 n = B_NR_ITEMS (tbSh); /* number of items in S[h] */
855 if ( n - tb->rnum[h] >= child_pos )
856 /* new items fall into S[h] */
857 /*internal_shift_right(tb,h,tbSh,tb->CFR[h],tb->rkey[h],tb->R[h],tb->rnum[h]);*/
858 internal_shift_right (INTERNAL_SHIFT_FROM_S_TO_R, tb, h, tb->rnum[h]);
860 if ( n + insert_num - tb->rnum[h] < child_pos )
862 /* all new items fall into R[h] */
863 /*internal_shift_right(tb,h,tbSh,tb->CFR[h],tb->rkey[h],tb->R[h],
864 tb->rnum[h] - insert_num);*/
865 internal_shift_right (INTERNAL_SHIFT_FROM_S_TO_R, tb, h, tb->rnum[h] - insert_num);
867 /* insert insert_num keys and node-pointers into R[h] */
870 bi.bi_parent = tb->FR[h];
871 bi.bi_position = get_right_neighbor_position (tb, h);
872 internal_insert_childs (&bi, /*tb->R[h],tb->S[h-1]->b_next*/ child_pos - n - insert_num + tb->rnum[h] - 1,
873 insert_num,insert_key,insert_ptr);
878 struct disk_child * dc;
880 /* one of the items falls into CFR[h] */
881 internal_shift1_right(tb,h,n - child_pos + 1);
882 /* calculate number of new items that fall into R[h] */
883 k = tb->rnum[h] - n + child_pos - 1;
886 bi.bi_parent = tb->FR[h];
887 bi.bi_position = get_right_neighbor_position (tb, h);
888 internal_insert_childs (&bi, /*tb->R[h], tb->R[h]->b_child,*/ 0, k, insert_key + 1, insert_ptr + 1);
890 replace_rkey(tb,h,insert_key + insert_num - k - 1);
892 /* replace the first node-ptr in R[h] by node-ptr insert_ptr[insert_num-k-1]*/
893 dc = B_N_CHILD(tb->R[h], 0);
894 put_dc_size( dc, MAX_CHILD_SIZE(insert_ptr[insert_num-k-1]) -
895 B_FREE_SPACE (insert_ptr[insert_num-k-1]));
896 put_dc_block_number( dc, insert_ptr[insert_num-k-1]->b_blocknr );
898 do_balance_mark_internal_dirty (tb, tb->R[h],0);
900 insert_num -= (k + 1);
904 /** Fill new node that appears instead of S[h] **/
905 RFALSE( tb->blknum[h] > 2, "blknum can not be > 2 for internal level");
906 RFALSE( tb->blknum[h] < 0, "blknum can not be < 0");
908 if ( ! tb->blknum[h] )
909 { /* node S[h] is empty now */
910 RFALSE( ! tbSh, "S[h] is equal NULL");
912 /* do what is needed for buffer thrown from tree */
913 reiserfs_invalidate_buffer(tb,tbSh);
918 /* create new root */
919 struct disk_child * dc;
920 struct buffer_head * tbSh_1 = PATH_H_PBUFFER (tb->tb_path, h - 1);
921 struct block_head * blkh;
924 if ( tb->blknum[h] != 1 )
925 reiserfs_panic(NULL, "balance_internal: One new node required for creating the new root");
926 /* S[h] = empty buffer from the list FEB. */
928 blkh = B_BLK_HEAD(tbSh);
929 set_blkh_level( blkh, h + 1 );
931 /* Put the unique node-pointer to S[h] that points to S[h-1]. */
933 dc = B_N_CHILD(tbSh, 0);
934 put_dc_block_number( dc, tbSh_1->b_blocknr );
935 put_dc_size( dc, (MAX_CHILD_SIZE (tbSh_1) - B_FREE_SPACE (tbSh_1)));
937 tb->insert_size[h] -= DC_SIZE;
938 set_blkh_free_space( blkh, blkh_free_space(blkh) - DC_SIZE );
940 do_balance_mark_internal_dirty (tb, tbSh, 0);
942 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
943 check_internal (tbSh);
944 /*&&&&&&&&&&&&&&&&&&&&&&&&*/
946 /* put new root into path structure */
947 PATH_OFFSET_PBUFFER(tb->tb_path, ILLEGAL_PATH_ELEMENT_OFFSET) = tbSh;
949 /* Change root in structure super block. */
950 PUT_SB_ROOT_BLOCK( tb->tb_sb, tbSh->b_blocknr );
951 PUT_SB_TREE_HEIGHT( tb->tb_sb, SB_TREE_HEIGHT(tb->tb_sb) + 1 );
952 do_balance_mark_sb_dirty (tb, REISERFS_SB(tb->tb_sb)->s_sbh, 1);
955 if ( tb->blknum[h] == 2 ) {
957 struct buffer_info dest_bi, src_bi;
960 /* S_new = free buffer from list FEB */
963 set_blkh_level( B_BLK_HEAD(S_new), h + 1 );
966 dest_bi.bi_bh = S_new;
967 dest_bi.bi_parent = NULL;
968 dest_bi.bi_position = 0;
971 src_bi.bi_parent = PATH_H_PPARENT (tb->tb_path, h);
972 src_bi.bi_position = PATH_H_POSITION (tb->tb_path, h + 1);
974 n = B_NR_ITEMS (tbSh); /* number of items in S[h] */
975 snum = (insert_num + n + 1)/2;
976 if ( n - snum >= child_pos ) {
977 /* new items don't fall into S_new */
978 /* store the delimiting key for the next level */
979 /* new_insert_key = (n - snum)'th key in S[h] */
980 memcpy (&new_insert_key,B_N_PDELIM_KEY(tbSh,n - snum),
982 /* last parameter is del_par */
983 internal_move_pointers_items (&dest_bi, &src_bi, LAST_TO_FIRST, snum, 0);
984 /* internal_move_pointers_items(S_new, tbSh, LAST_TO_FIRST, snum, 0);*/
985 } else if ( n + insert_num - snum < child_pos ) {
986 /* all new items fall into S_new */
987 /* store the delimiting key for the next level */
988 /* new_insert_key = (n + insert_item - snum)'th key in S[h] */
989 memcpy(&new_insert_key,B_N_PDELIM_KEY(tbSh,n + insert_num - snum),
991 /* last parameter is del_par */
992 internal_move_pointers_items (&dest_bi, &src_bi, LAST_TO_FIRST, snum - insert_num, 0);
993 /* internal_move_pointers_items(S_new,tbSh,1,snum - insert_num,0);*/
995 /* insert insert_num keys and node-pointers into S_new */
996 internal_insert_childs (&dest_bi, /*S_new,tb->S[h-1]->b_next,*/child_pos - n - insert_num + snum - 1,
997 insert_num,insert_key,insert_ptr);
1001 struct disk_child * dc;
1003 /* some items fall into S_new, but some don't fall */
1004 /* last parameter is del_par */
1005 internal_move_pointers_items (&dest_bi, &src_bi, LAST_TO_FIRST, n - child_pos + 1, 1);
1006 /* internal_move_pointers_items(S_new,tbSh,1,n - child_pos + 1,1);*/
1007 /* calculate number of new items that fall into S_new */
1008 k = snum - n + child_pos - 1;
1010 internal_insert_childs (&dest_bi, /*S_new,*/ 0, k, insert_key + 1, insert_ptr+1);
1012 /* new_insert_key = insert_key[insert_num - k - 1] */
1013 memcpy(&new_insert_key,insert_key + insert_num - k - 1,
1015 /* replace first node-ptr in S_new by node-ptr to insert_ptr[insert_num-k-1] */
1017 dc = B_N_CHILD(S_new,0);
1018 put_dc_size( dc, (MAX_CHILD_SIZE(insert_ptr[insert_num-k-1]) -
1019 B_FREE_SPACE(insert_ptr[insert_num-k-1])) );
1020 put_dc_block_number( dc, insert_ptr[insert_num-k-1]->b_blocknr );
1022 do_balance_mark_internal_dirty (tb, S_new,0);
1024 insert_num -= (k + 1);
1026 /* new_insert_ptr = node_pointer to S_new */
1027 new_insert_ptr = S_new;
1029 RFALSE (!buffer_journaled(S_new) || buffer_journal_dirty(S_new) ||
1030 buffer_dirty (S_new),
1031 "cm-00001: bad S_new (%b)", S_new);
1033 // S_new is released in unfix_nodes
1036 n = B_NR_ITEMS (tbSh); /*number of items in S[h] */
1038 if ( 0 <= child_pos && child_pos <= n && insert_num > 0 ) {
1041 bi.bi_parent = PATH_H_PPARENT (tb->tb_path, h);
1042 bi.bi_position = PATH_H_POSITION (tb->tb_path, h + 1);
1043 internal_insert_childs (
1045 /* ( tb->S[h-1]->b_parent == tb->S[h] ) ? tb->S[h-1]->b_next : tb->S[h]->b_child->b_next,*/
1046 child_pos,insert_num,insert_key,insert_ptr
1051 memcpy (new_insert_key_addr,&new_insert_key,KEY_SIZE);
1052 insert_ptr[0] = new_insert_ptr;