2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
6 * Now we have all buffers that must be used in balancing of the tree
7 * Further calculations can not cause schedule(), and thus the buffer
8 * tree will be stable until the balancing will be finished
9 * balance the tree according to the analysis made before,
10 * and using buffers obtained after all above.
13 #include <asm/uaccess.h>
14 #include <linux/time.h>
16 #include <linux/buffer_head.h>
17 #include <linux/kernel.h>
19 static inline void buffer_info_init_left(struct tree_balance *tb,
20 struct buffer_info *bi)
24 bi->bi_parent = tb->FL[0];
25 bi->bi_position = get_left_neighbor_position(tb, 0);
28 static inline void buffer_info_init_right(struct tree_balance *tb,
29 struct buffer_info *bi)
33 bi->bi_parent = tb->FR[0];
34 bi->bi_position = get_right_neighbor_position(tb, 0);
37 static inline void buffer_info_init_tbS0(struct tree_balance *tb,
38 struct buffer_info *bi)
41 bi->bi_bh = PATH_PLAST_BUFFER(tb->tb_path);
42 bi->bi_parent = PATH_H_PPARENT(tb->tb_path, 0);
43 bi->bi_position = PATH_H_POSITION(tb->tb_path, 1);
46 static inline void buffer_info_init_bh(struct tree_balance *tb,
47 struct buffer_info *bi,
48 struct buffer_head *bh)
56 inline void do_balance_mark_leaf_dirty(struct tree_balance *tb,
57 struct buffer_head *bh, int flag)
59 journal_mark_dirty(tb->transaction_handle, bh);
62 #define do_balance_mark_internal_dirty do_balance_mark_leaf_dirty
63 #define do_balance_mark_sb_dirty do_balance_mark_leaf_dirty
67 * if deleting something ( tb->insert_size[0] < 0 )
68 * return(balance_leaf_when_delete()); (flag d handled here)
70 * if lnum is larger than 0 we put items into the left node
71 * if rnum is larger than 0 we put items into the right node
72 * if snum1 is larger than 0 we put items into the new node s1
73 * if snum2 is larger than 0 we put items into the new node s2
74 * Note that all *num* count new items being created.
76 * It would be easier to read balance_leaf() if each of these summary
77 * lines was a separate procedure rather than being inlined. I think
78 * that there are many passages here and in balance_leaf_when_delete() in
79 * which two calls to one procedure can replace two passages, and it
80 * might save cache space and improve software maintenance costs to do so.
82 * Vladimir made the perceptive comment that we should offload most of
83 * the decision making in this function into fix_nodes/check_balance, and
84 * then create some sort of structure in tb that says what actions should
85 * be performed by do_balance.
91 * Balance leaf node in case of delete or cut: insert_size[0] < 0
93 * lnum, rnum can have values >= -1
94 * -1 means that the neighbor must be joined with S
95 * 0 means that nothing should be done with the neighbor
96 * >0 means to shift entirely or partly the specified number of items
99 static int balance_leaf_when_delete(struct tree_balance *tb, int flag)
101 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
102 int item_pos = PATH_LAST_POSITION(tb->tb_path);
103 int pos_in_item = tb->tb_path->pos_in_item;
104 struct buffer_info bi;
106 struct item_head *ih;
108 RFALSE(tb->FR[0] && B_LEVEL(tb->FR[0]) != DISK_LEAF_NODE_LEVEL + 1,
109 "vs- 12000: level: wrong FR %z", tb->FR[0]);
110 RFALSE(tb->blknum[0] > 1,
111 "PAP-12005: tb->blknum == %d, can not be > 1", tb->blknum[0]);
112 RFALSE(!tb->blknum[0] && !PATH_H_PPARENT(tb->tb_path, 0),
113 "PAP-12010: tree can not be empty");
115 ih = item_head(tbS0, item_pos);
116 buffer_info_init_tbS0(tb, &bi);
118 /* Delete or truncate the item */
121 case M_DELETE: /* delete item in S[0] */
123 RFALSE(ih_item_len(ih) + IH_SIZE != -tb->insert_size[0],
124 "vs-12013: mode Delete, insert size %d, ih to be deleted %h",
125 -tb->insert_size[0], ih);
127 leaf_delete_items(&bi, 0, item_pos, 1, -1);
129 if (!item_pos && tb->CFL[0]) {
130 if (B_NR_ITEMS(tbS0)) {
131 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0,
134 if (!PATH_H_POSITION(tb->tb_path, 1))
135 replace_key(tb, tb->CFL[0], tb->lkey[0],
136 PATH_H_PPARENT(tb->tb_path,
141 RFALSE(!item_pos && !tb->CFL[0],
142 "PAP-12020: tb->CFL[0]==%p, tb->L[0]==%p", tb->CFL[0],
147 case M_CUT:{ /* cut item in S[0] */
148 if (is_direntry_le_ih(ih)) {
151 * UFS unlink semantics are such that you
152 * can only delete one directory entry at
157 * when we cut a directory tb->insert_size[0]
158 * means number of entries to be cut (always 1)
160 tb->insert_size[0] = -1;
161 leaf_cut_from_buffer(&bi, item_pos, pos_in_item,
162 -tb->insert_size[0]);
164 RFALSE(!item_pos && !pos_in_item && !tb->CFL[0],
165 "PAP-12030: can not change delimiting key. CFL[0]=%p",
168 if (!item_pos && !pos_in_item && tb->CFL[0]) {
169 replace_key(tb, tb->CFL[0], tb->lkey[0],
173 leaf_cut_from_buffer(&bi, item_pos, pos_in_item,
174 -tb->insert_size[0]);
176 RFALSE(!ih_item_len(ih),
177 "PAP-12035: cut must leave non-zero dynamic length of item");
183 print_cur_tb("12040");
184 reiserfs_panic(tb->tb_sb, "PAP-12040",
185 "unexpected mode: %s(%d)",
187 M_PASTE) ? "PASTE" : ((flag ==
188 M_INSERT) ? "INSERT" :
193 * the rule is that no shifting occurs unless by shifting
194 * a node can be freed
196 n = B_NR_ITEMS(tbS0);
197 /* L[0] takes part in balancing */
199 /* L[0] must be joined with S[0] */
200 if (tb->lnum[0] == -1) {
201 /* R[0] must be also joined with S[0] */
202 if (tb->rnum[0] == -1) {
203 if (tb->FR[0] == PATH_H_PPARENT(tb->tb_path, 0)) {
205 * all contents of all the 3 buffers
208 if (PATH_H_POSITION(tb->tb_path, 1) == 0
209 && 1 < B_NR_ITEMS(tb->FR[0]))
210 replace_key(tb, tb->CFL[0],
214 leaf_move_items(LEAF_FROM_S_TO_L, tb, n,
216 leaf_move_items(LEAF_FROM_R_TO_L, tb,
217 B_NR_ITEMS(tb->R[0]),
220 reiserfs_invalidate_buffer(tb, tbS0);
221 reiserfs_invalidate_buffer(tb,
227 * all contents of all the 3 buffers will
230 leaf_move_items(LEAF_FROM_S_TO_R, tb, n, -1,
232 leaf_move_items(LEAF_FROM_L_TO_R, tb,
233 B_NR_ITEMS(tb->L[0]), -1, NULL);
235 /* right_delimiting_key is correct in R[0] */
236 replace_key(tb, tb->CFR[0], tb->rkey[0],
239 reiserfs_invalidate_buffer(tb, tbS0);
240 reiserfs_invalidate_buffer(tb, tb->L[0]);
245 RFALSE(tb->rnum[0] != 0,
246 "PAP-12045: rnum must be 0 (%d)", tb->rnum[0]);
247 /* all contents of L[0] and S[0] will be in L[0] */
248 leaf_shift_left(tb, n, -1);
250 reiserfs_invalidate_buffer(tb, tbS0);
256 * a part of contents of S[0] will be in L[0] and the
257 * rest part of S[0] will be in R[0]
260 RFALSE((tb->lnum[0] + tb->rnum[0] < n) ||
261 (tb->lnum[0] + tb->rnum[0] > n + 1),
262 "PAP-12050: rnum(%d) and lnum(%d) and item number(%d) in S[0] are not consistent",
263 tb->rnum[0], tb->lnum[0], n);
264 RFALSE((tb->lnum[0] + tb->rnum[0] == n) &&
265 (tb->lbytes != -1 || tb->rbytes != -1),
266 "PAP-12055: bad rbytes (%d)/lbytes (%d) parameters when items are not split",
267 tb->rbytes, tb->lbytes);
268 RFALSE((tb->lnum[0] + tb->rnum[0] == n + 1) &&
269 (tb->lbytes < 1 || tb->rbytes != -1),
270 "PAP-12060: bad rbytes (%d)/lbytes (%d) parameters when items are split",
271 tb->rbytes, tb->lbytes);
273 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
274 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
276 reiserfs_invalidate_buffer(tb, tbS0);
281 if (tb->rnum[0] == -1) {
282 /* all contents of R[0] and S[0] will be in R[0] */
283 leaf_shift_right(tb, n, -1);
284 reiserfs_invalidate_buffer(tb, tbS0);
289 "PAP-12065: bad rnum parameter must be 0 (%d)", tb->rnum[0]);
293 static void balance_leaf_insert_left(struct tree_balance *tb,
294 struct item_head *ih, const char *body)
297 struct buffer_info bi;
298 int n = B_NR_ITEMS(tb->L[0]);
300 if (tb->item_pos == tb->lnum[0] - 1 && tb->lbytes != -1) {
301 /* part of new item falls into L[0] */
302 int new_item_len, shift;
305 ret = leaf_shift_left(tb, tb->lnum[0] - 1, -1);
307 /* Calculate item length to insert to S[0] */
308 new_item_len = ih_item_len(ih) - tb->lbytes;
310 /* Calculate and check item length to insert to L[0] */
311 put_ih_item_len(ih, ih_item_len(ih) - new_item_len);
313 RFALSE(ih_item_len(ih) <= 0,
314 "PAP-12080: there is nothing to insert into L[0]: "
315 "ih_item_len=%d", ih_item_len(ih));
317 /* Insert new item into L[0] */
318 buffer_info_init_left(tb, &bi);
319 leaf_insert_into_buf(&bi, n + tb->item_pos - ret, ih, body,
320 min_t(int, tb->zeroes_num, ih_item_len(ih)));
322 version = ih_version(ih);
325 * Calculate key component, item length and body to
329 if (is_indirect_le_ih(ih))
330 shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
332 add_le_ih_k_offset(ih, tb->lbytes << shift);
334 put_ih_item_len(ih, new_item_len);
335 if (tb->lbytes > tb->zeroes_num) {
336 body += (tb->lbytes - tb->zeroes_num);
339 tb->zeroes_num -= tb->lbytes;
341 RFALSE(ih_item_len(ih) <= 0,
342 "PAP-12085: there is nothing to insert into S[0]: "
343 "ih_item_len=%d", ih_item_len(ih));
345 /* new item in whole falls into L[0] */
346 /* Shift lnum[0]-1 items to L[0] */
347 ret = leaf_shift_left(tb, tb->lnum[0] - 1, tb->lbytes);
349 /* Insert new item into L[0] */
350 buffer_info_init_left(tb, &bi);
351 leaf_insert_into_buf(&bi, n + tb->item_pos - ret, ih, body,
353 tb->insert_size[0] = 0;
358 static void balance_leaf_paste_left_shift_dirent(struct tree_balance *tb,
359 struct item_head *ih,
362 int n = B_NR_ITEMS(tb->L[0]);
363 struct buffer_info bi;
365 RFALSE(tb->zeroes_num,
366 "PAP-12090: invalid parameter in case of a directory");
369 if (tb->lbytes > tb->pos_in_item) {
370 /* new directory entry falls into L[0] */
371 struct item_head *pasted;
372 int ret, l_pos_in_item = tb->pos_in_item;
375 * Shift lnum[0] - 1 items in whole.
376 * Shift lbytes - 1 entries from given directory item
378 ret = leaf_shift_left(tb, tb->lnum[0], tb->lbytes - 1);
379 if (ret && !tb->item_pos) {
380 pasted = item_head(tb->L[0], B_NR_ITEMS(tb->L[0]) - 1);
381 l_pos_in_item += ih_entry_count(pasted) -
385 /* Append given directory entry to directory item */
386 buffer_info_init_left(tb, &bi);
387 leaf_paste_in_buffer(&bi, n + tb->item_pos - ret,
388 l_pos_in_item, tb->insert_size[0],
389 body, tb->zeroes_num);
392 * previous string prepared space for pasting new entry,
393 * following string pastes this entry
397 * when we have merge directory item, pos_in_item
398 * has been changed too
401 /* paste new directory entry. 1 is entry number */
402 leaf_paste_entries(&bi, n + tb->item_pos - ret,
404 (struct reiserfs_de_head *) body,
405 body + DEH_SIZE, tb->insert_size[0]);
406 tb->insert_size[0] = 0;
408 /* new directory item doesn't fall into L[0] */
410 * Shift lnum[0]-1 items in whole. Shift lbytes
411 * directory entries from directory item number lnum[0]
413 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
416 /* Calculate new position to append in item body */
417 tb->pos_in_item -= tb->lbytes;
420 static void balance_leaf_paste_left_shift(struct tree_balance *tb,
421 struct item_head *ih,
424 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
425 int n = B_NR_ITEMS(tb->L[0]);
426 struct buffer_info bi;
428 if (is_direntry_le_ih(item_head(tbS0, tb->item_pos))) {
429 balance_leaf_paste_left_shift_dirent(tb, ih, body);
433 RFALSE(tb->lbytes <= 0,
434 "PAP-12095: there is nothing to shift to L[0]. "
435 "lbytes=%d", tb->lbytes);
436 RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)),
437 "PAP-12100: incorrect position to paste: "
438 "item_len=%d, pos_in_item=%d",
439 ih_item_len(item_head(tbS0, tb->item_pos)), tb->pos_in_item);
441 /* appended item will be in L[0] in whole */
442 if (tb->lbytes >= tb->pos_in_item) {
443 struct item_head *tbS0_pos_ih, *tbL0_ih;
444 struct item_head *tbS0_0_ih;
445 struct reiserfs_key *left_delim_key;
446 int ret, l_n, version, temp_l;
448 tbS0_pos_ih = item_head(tbS0, tb->item_pos);
449 tbS0_0_ih = item_head(tbS0, 0);
452 * this bytes number must be appended
453 * to the last item of L[h]
455 l_n = tb->lbytes - tb->pos_in_item;
457 /* Calculate new insert_size[0] */
458 tb->insert_size[0] -= l_n;
460 RFALSE(tb->insert_size[0] <= 0,
461 "PAP-12105: there is nothing to paste into "
462 "L[0]. insert_size=%d", tb->insert_size[0]);
464 ret = leaf_shift_left(tb, tb->lnum[0],
465 ih_item_len(tbS0_pos_ih));
467 tbL0_ih = item_head(tb->L[0], n + tb->item_pos - ret);
469 /* Append to body of item in L[0] */
470 buffer_info_init_left(tb, &bi);
471 leaf_paste_in_buffer(&bi, n + tb->item_pos - ret,
472 ih_item_len(tbL0_ih), l_n, body,
473 min_t(int, l_n, tb->zeroes_num));
476 * 0-th item in S0 can be only of DIRECT type
481 RFALSE(ih_item_len(tbS0_0_ih),
482 "PAP-12106: item length must be 0");
483 RFALSE(comp_short_le_keys(&tbS0_0_ih->ih_key,
484 leaf_key(tb->L[0], n + tb->item_pos - ret)),
485 "PAP-12107: items must be of the same file");
487 if (is_indirect_le_ih(tbL0_ih)) {
488 int shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
489 temp_l = l_n << shift;
491 /* update key of first item in S0 */
492 version = ih_version(tbS0_0_ih);
493 add_le_key_k_offset(version, &tbS0_0_ih->ih_key, temp_l);
495 /* update left delimiting key */
496 left_delim_key = internal_key(tb->CFL[0], tb->lkey[0]);
497 add_le_key_k_offset(version, left_delim_key, temp_l);
500 * Calculate new body, position in item and
503 if (l_n > tb->zeroes_num) {
504 body += (l_n - tb->zeroes_num);
507 tb->zeroes_num -= l_n;
510 RFALSE(comp_short_le_keys(&tbS0_0_ih->ih_key,
512 B_NR_ITEMS(tb->L[0]) - 1)) ||
513 !op_is_left_mergeable(leaf_key(tbS0, 0), tbS0->b_size) ||
514 !op_is_left_mergeable(left_delim_key, tbS0->b_size),
515 "PAP-12120: item must be merge-able with left "
518 /* only part of the appended item will be in L[0] */
520 /* Calculate position in item for append in S[0] */
521 tb->pos_in_item -= tb->lbytes;
523 RFALSE(tb->pos_in_item <= 0,
524 "PAP-12125: no place for paste. pos_in_item=%d",
528 * Shift lnum[0] - 1 items in whole.
529 * Shift lbytes - 1 byte from item number lnum[0]
531 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
536 /* appended item will be in L[0] in whole */
537 static void balance_leaf_paste_left_whole(struct tree_balance *tb,
538 struct item_head *ih,
541 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
542 int n = B_NR_ITEMS(tb->L[0]);
543 struct buffer_info bi;
544 struct item_head *pasted;
547 /* if we paste into first item of S[0] and it is left mergable */
549 op_is_left_mergeable(leaf_key(tbS0, 0), tbS0->b_size)) {
551 * then increment pos_in_item by the size of the
554 pasted = item_head(tb->L[0], n - 1);
555 if (is_direntry_le_ih(pasted))
556 tb->pos_in_item += ih_entry_count(pasted);
558 tb->pos_in_item += ih_item_len(pasted);
562 * Shift lnum[0] - 1 items in whole.
563 * Shift lbytes - 1 byte from item number lnum[0]
565 ret = leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
567 /* Append to body of item in L[0] */
568 buffer_info_init_left(tb, &bi);
569 leaf_paste_in_buffer(&bi, n + tb->item_pos - ret, tb->pos_in_item,
570 tb->insert_size[0], body, tb->zeroes_num);
572 /* if appended item is directory, paste entry */
573 pasted = item_head(tb->L[0], n + tb->item_pos - ret);
574 if (is_direntry_le_ih(pasted))
575 leaf_paste_entries(&bi, n + tb->item_pos - ret,
577 (struct reiserfs_de_head *)body,
578 body + DEH_SIZE, tb->insert_size[0]);
581 * if appended item is indirect item, put unformatted node
584 if (is_indirect_le_ih(pasted))
585 set_ih_free_space(pasted, 0);
587 tb->insert_size[0] = 0;
591 static void balance_leaf_paste_left(struct tree_balance *tb,
592 struct item_head *ih, const char *body)
594 /* we must shift the part of the appended item */
595 if (tb->item_pos == tb->lnum[0] - 1 && tb->lbytes != -1)
596 balance_leaf_paste_left_shift(tb, ih, body);
598 balance_leaf_paste_left_whole(tb, ih, body);
601 /* Shift lnum[0] items from S[0] to the left neighbor L[0] */
602 static void balance_leaf_left(struct tree_balance *tb, struct item_head *ih,
603 const char *body, int flag)
605 if (tb->lnum[0] <= 0)
608 /* new item or it part falls to L[0], shift it too */
609 if (tb->item_pos < tb->lnum[0]) {
610 BUG_ON(flag != M_INSERT && flag != M_PASTE);
612 if (flag == M_INSERT)
613 balance_leaf_insert_left(tb, ih, body);
615 balance_leaf_paste_left(tb, ih, body);
617 /* new item doesn't fall into L[0] */
618 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
622 static void balance_leaf_insert_right(struct tree_balance *tb,
623 struct item_head *ih, const char *body)
626 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
627 int n = B_NR_ITEMS(tbS0);
628 struct buffer_info bi;
631 /* new item or part of it doesn't fall into R[0] */
632 if (n - tb->rnum[0] >= tb->item_pos) {
633 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
637 /* new item or its part falls to R[0] */
639 /* part of new item falls into R[0] */
640 if (tb->item_pos == n - tb->rnum[0] + 1 && tb->rbytes != -1) {
641 loff_t old_key_comp, old_len, r_zeroes_number;
646 leaf_shift_right(tb, tb->rnum[0] - 1, -1);
648 version = ih_version(ih);
650 /* Remember key component and item length */
651 old_key_comp = le_ih_k_offset(ih);
652 old_len = ih_item_len(ih);
655 * Calculate key component and item length to insert
659 if (is_indirect_le_ih(ih))
660 shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
661 offset = le_ih_k_offset(ih) + ((old_len - tb->rbytes) << shift);
662 set_le_ih_k_offset(ih, offset);
663 put_ih_item_len(ih, tb->rbytes);
665 /* Insert part of the item into R[0] */
666 buffer_info_init_right(tb, &bi);
667 if ((old_len - tb->rbytes) > tb->zeroes_num) {
669 r_body = body + (old_len - tb->rbytes) - tb->zeroes_num;
672 r_zeroes_number = tb->zeroes_num -
673 (old_len - tb->rbytes);
674 tb->zeroes_num -= r_zeroes_number;
677 leaf_insert_into_buf(&bi, 0, ih, r_body, r_zeroes_number);
679 /* Replace right delimiting key by first key in R[0] */
680 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0);
683 * Calculate key component and item length to
686 set_le_ih_k_offset(ih, old_key_comp);
687 put_ih_item_len(ih, old_len - tb->rbytes);
689 tb->insert_size[0] -= tb->rbytes;
692 /* whole new item falls into R[0] */
694 /* Shift rnum[0]-1 items to R[0] */
695 ret = leaf_shift_right(tb, tb->rnum[0] - 1, tb->rbytes);
697 /* Insert new item into R[0] */
698 buffer_info_init_right(tb, &bi);
699 leaf_insert_into_buf(&bi, tb->item_pos - n + tb->rnum[0] - 1,
700 ih, body, tb->zeroes_num);
702 if (tb->item_pos - n + tb->rnum[0] - 1 == 0)
703 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0);
705 tb->zeroes_num = tb->insert_size[0] = 0;
710 static void balance_leaf_paste_right_shift_dirent(struct tree_balance *tb,
711 struct item_head *ih, const char *body)
713 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
714 struct buffer_info bi;
717 RFALSE(tb->zeroes_num,
718 "PAP-12145: invalid parameter in case of a directory");
719 entry_count = ih_entry_count(item_head(tbS0, tb->item_pos));
721 /* new directory entry falls into R[0] */
722 if (entry_count - tb->rbytes < tb->pos_in_item) {
723 int paste_entry_position;
725 RFALSE(tb->rbytes - 1 >= entry_count || !tb->insert_size[0],
726 "PAP-12150: no enough of entries to shift to R[0]: "
727 "rbytes=%d, entry_count=%d", tb->rbytes, entry_count);
730 * Shift rnum[0]-1 items in whole.
731 * Shift rbytes-1 directory entries from directory
732 * item number rnum[0]
734 leaf_shift_right(tb, tb->rnum[0], tb->rbytes - 1);
736 /* Paste given directory entry to directory item */
737 paste_entry_position = tb->pos_in_item - entry_count +
739 buffer_info_init_right(tb, &bi);
740 leaf_paste_in_buffer(&bi, 0, paste_entry_position,
741 tb->insert_size[0], body, tb->zeroes_num);
744 leaf_paste_entries(&bi, 0, paste_entry_position, 1,
745 (struct reiserfs_de_head *) body,
746 body + DEH_SIZE, tb->insert_size[0]);
748 /* change delimiting keys */
749 if (paste_entry_position == 0)
750 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0);
752 tb->insert_size[0] = 0;
755 /* new directory entry doesn't fall into R[0] */
756 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
760 static void balance_leaf_paste_right_shift(struct tree_balance *tb,
761 struct item_head *ih, const char *body)
763 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
764 int n_shift, n_rem, r_zeroes_number, version;
765 unsigned long temp_rem;
767 struct buffer_info bi;
769 /* we append to directory item */
770 if (is_direntry_le_ih(item_head(tbS0, tb->item_pos))) {
771 balance_leaf_paste_right_shift_dirent(tb, ih, body);
778 * Calculate number of bytes which must be shifted
781 n_shift = tb->rbytes - tb->insert_size[0];
785 RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)),
786 "PAP-12155: invalid position to paste. ih_item_len=%d, "
787 "pos_in_item=%d", tb->pos_in_item,
788 ih_item_len(item_head(tbS0, tb->item_pos)));
790 leaf_shift_right(tb, tb->rnum[0], n_shift);
793 * Calculate number of bytes which must remain in body
794 * after appending to R[0]
796 n_rem = tb->insert_size[0] - tb->rbytes;
802 version = ih_version(item_head(tb->R[0], 0));
804 if (is_indirect_le_key(version, leaf_key(tb->R[0], 0))) {
805 int shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
806 temp_rem = n_rem << shift;
809 add_le_key_k_offset(version, leaf_key(tb->R[0], 0), temp_rem);
810 add_le_key_k_offset(version, internal_key(tb->CFR[0], tb->rkey[0]),
813 do_balance_mark_internal_dirty(tb, tb->CFR[0], 0);
815 /* Append part of body into R[0] */
816 buffer_info_init_right(tb, &bi);
817 if (n_rem > tb->zeroes_num) {
819 r_body = body + n_rem - tb->zeroes_num;
822 r_zeroes_number = tb->zeroes_num - n_rem;
823 tb->zeroes_num -= r_zeroes_number;
826 leaf_paste_in_buffer(&bi, 0, n_shift, tb->insert_size[0] - n_rem,
827 r_body, r_zeroes_number);
829 if (is_indirect_le_ih(item_head(tb->R[0], 0)))
830 set_ih_free_space(item_head(tb->R[0], 0), 0);
832 tb->insert_size[0] = n_rem;
837 static void balance_leaf_paste_right_whole(struct tree_balance *tb,
838 struct item_head *ih, const char *body)
840 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
841 int n = B_NR_ITEMS(tbS0);
842 struct item_head *pasted;
843 struct buffer_info bi;
845 buffer_info_init_right(tb, &bi);
846 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
848 /* append item in R[0] */
849 if (tb->pos_in_item >= 0) {
850 buffer_info_init_right(tb, &bi);
851 leaf_paste_in_buffer(&bi, tb->item_pos - n + tb->rnum[0],
852 tb->pos_in_item, tb->insert_size[0], body,
856 /* paste new entry, if item is directory item */
857 pasted = item_head(tb->R[0], tb->item_pos - n + tb->rnum[0]);
858 if (is_direntry_le_ih(pasted) && tb->pos_in_item >= 0) {
859 leaf_paste_entries(&bi, tb->item_pos - n + tb->rnum[0],
861 (struct reiserfs_de_head *)body,
862 body + DEH_SIZE, tb->insert_size[0]);
864 if (!tb->pos_in_item) {
866 RFALSE(tb->item_pos - n + tb->rnum[0],
867 "PAP-12165: directory item must be first "
868 "item of node when pasting is in 0th position");
870 /* update delimiting keys */
871 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0);
875 if (is_indirect_le_ih(pasted))
876 set_ih_free_space(pasted, 0);
877 tb->zeroes_num = tb->insert_size[0] = 0;
880 static void balance_leaf_paste_right(struct tree_balance *tb,
881 struct item_head *ih, const char *body)
883 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
884 int n = B_NR_ITEMS(tbS0);
886 /* new item doesn't fall into R[0] */
887 if (n - tb->rnum[0] > tb->item_pos) {
888 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
892 /* pasted item or part of it falls to R[0] */
894 if (tb->item_pos == n - tb->rnum[0] && tb->rbytes != -1)
895 /* we must shift the part of the appended item */
896 balance_leaf_paste_right_shift(tb, ih, body);
898 /* pasted item in whole falls into R[0] */
899 balance_leaf_paste_right_whole(tb, ih, body);
902 /* shift rnum[0] items from S[0] to the right neighbor R[0] */
903 static void balance_leaf_right(struct tree_balance *tb, struct item_head *ih,
904 const char *body, int flag)
906 if (tb->rnum[0] <= 0)
909 BUG_ON(flag != M_INSERT && flag != M_PASTE);
911 if (flag == M_INSERT)
912 balance_leaf_insert_right(tb, ih, body);
914 balance_leaf_paste_right(tb, ih, body);
917 static void balance_leaf_new_nodes_insert(struct tree_balance *tb,
918 struct item_head *ih,
920 struct item_head *insert_key,
921 struct buffer_head **insert_ptr,
924 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
925 int n = B_NR_ITEMS(tbS0);
926 struct buffer_info bi;
929 /* new item or it part don't falls into S_new[i] */
930 if (n - tb->snum[i] >= tb->item_pos) {
931 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
932 tb->snum[i], tb->sbytes[i], tb->S_new[i]);
936 /* new item or it's part falls to first new node S_new[i] */
938 /* part of new item falls into S_new[i] */
939 if (tb->item_pos == n - tb->snum[i] + 1 && tb->sbytes[i] != -1) {
940 int old_key_comp, old_len, r_zeroes_number;
944 /* Move snum[i]-1 items from S[0] to S_new[i] */
945 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i] - 1, -1,
948 /* Remember key component and item length */
949 version = ih_version(ih);
950 old_key_comp = le_ih_k_offset(ih);
951 old_len = ih_item_len(ih);
954 * Calculate key component and item length to insert
958 if (is_indirect_le_ih(ih))
959 shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
960 set_le_ih_k_offset(ih,
962 ((old_len - tb->sbytes[i]) << shift));
964 put_ih_item_len(ih, tb->sbytes[i]);
966 /* Insert part of the item into S_new[i] before 0-th item */
967 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
969 if ((old_len - tb->sbytes[i]) > tb->zeroes_num) {
971 r_body = body + (old_len - tb->sbytes[i]) -
975 r_zeroes_number = tb->zeroes_num - (old_len -
977 tb->zeroes_num -= r_zeroes_number;
980 leaf_insert_into_buf(&bi, 0, ih, r_body, r_zeroes_number);
983 * Calculate key component and item length to
986 set_le_ih_k_offset(ih, old_key_comp);
987 put_ih_item_len(ih, old_len - tb->sbytes[i]);
988 tb->insert_size[0] -= tb->sbytes[i];
990 /* whole new item falls into S_new[i] */
993 * Shift snum[0] - 1 items to S_new[i]
994 * (sbytes[i] of split item)
996 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
997 tb->snum[i] - 1, tb->sbytes[i], tb->S_new[i]);
999 /* Insert new item into S_new[i] */
1000 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
1001 leaf_insert_into_buf(&bi, tb->item_pos - n + tb->snum[i] - 1,
1002 ih, body, tb->zeroes_num);
1004 tb->zeroes_num = tb->insert_size[0] = 0;
1008 /* we append to directory item */
1009 static void balance_leaf_new_nodes_paste_dirent(struct tree_balance *tb,
1010 struct item_head *ih,
1012 struct item_head *insert_key,
1013 struct buffer_head **insert_ptr,
1016 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1017 struct item_head *aux_ih = item_head(tbS0, tb->item_pos);
1018 int entry_count = ih_entry_count(aux_ih);
1019 struct buffer_info bi;
1021 if (entry_count - tb->sbytes[i] < tb->pos_in_item &&
1022 tb->pos_in_item <= entry_count) {
1023 /* new directory entry falls into S_new[i] */
1025 RFALSE(!tb->insert_size[0],
1026 "PAP-12215: insert_size is already 0");
1027 RFALSE(tb->sbytes[i] - 1 >= entry_count,
1028 "PAP-12220: there are no so much entries (%d), only %d",
1029 tb->sbytes[i] - 1, entry_count);
1032 * Shift snum[i]-1 items in whole.
1033 * Shift sbytes[i] directory entries
1034 * from directory item number snum[i]
1036 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i],
1037 tb->sbytes[i] - 1, tb->S_new[i]);
1040 * Paste given directory entry to
1043 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
1044 leaf_paste_in_buffer(&bi, 0, tb->pos_in_item - entry_count +
1045 tb->sbytes[i] - 1, tb->insert_size[0],
1046 body, tb->zeroes_num);
1048 /* paste new directory entry */
1049 leaf_paste_entries(&bi, 0, tb->pos_in_item - entry_count +
1050 tb->sbytes[i] - 1, 1,
1051 (struct reiserfs_de_head *) body,
1052 body + DEH_SIZE, tb->insert_size[0]);
1054 tb->insert_size[0] = 0;
1057 /* new directory entry doesn't fall into S_new[i] */
1058 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i],
1059 tb->sbytes[i], tb->S_new[i]);
1064 static void balance_leaf_new_nodes_paste_shift(struct tree_balance *tb,
1065 struct item_head *ih,
1067 struct item_head *insert_key,
1068 struct buffer_head **insert_ptr,
1071 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1072 struct item_head *aux_ih = item_head(tbS0, tb->item_pos);
1073 int n_shift, n_rem, r_zeroes_number, shift;
1075 struct item_head *tmp;
1076 struct buffer_info bi;
1078 RFALSE(ih, "PAP-12210: ih must be 0");
1080 if (is_direntry_le_ih(aux_ih)) {
1081 balance_leaf_new_nodes_paste_dirent(tb, ih, body, insert_key,
1086 /* regular object */
1089 RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)) ||
1090 tb->insert_size[0] <= 0,
1091 "PAP-12225: item too short or insert_size <= 0");
1094 * Calculate number of bytes which must be shifted from appended item
1096 n_shift = tb->sbytes[i] - tb->insert_size[0];
1099 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i], n_shift,
1103 * Calculate number of bytes which must remain in body after
1104 * append to S_new[i]
1106 n_rem = tb->insert_size[0] - tb->sbytes[i];
1110 /* Append part of body into S_new[0] */
1111 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
1112 if (n_rem > tb->zeroes_num) {
1113 r_zeroes_number = 0;
1114 r_body = body + n_rem - tb->zeroes_num;
1117 r_zeroes_number = tb->zeroes_num - n_rem;
1118 tb->zeroes_num -= r_zeroes_number;
1121 leaf_paste_in_buffer(&bi, 0, n_shift, tb->insert_size[0] - n_rem,
1122 r_body, r_zeroes_number);
1124 tmp = item_head(tb->S_new[i], 0);
1126 if (is_indirect_le_ih(tmp)) {
1127 set_ih_free_space(tmp, 0);
1128 shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
1130 add_le_ih_k_offset(tmp, n_rem << shift);
1132 tb->insert_size[0] = n_rem;
1137 static void balance_leaf_new_nodes_paste_whole(struct tree_balance *tb,
1138 struct item_head *ih,
1140 struct item_head *insert_key,
1141 struct buffer_head **insert_ptr,
1145 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1146 int n = B_NR_ITEMS(tbS0);
1148 struct item_head *pasted;
1149 struct buffer_info bi;
1151 #ifdef CONFIG_REISERFS_CHECK
1152 struct item_head *ih_check = item_head(tbS0, tb->item_pos);
1154 if (!is_direntry_le_ih(ih_check) &&
1155 (tb->pos_in_item != ih_item_len(ih_check) ||
1156 tb->insert_size[0] <= 0))
1157 reiserfs_panic(tb->tb_sb,
1159 "pos_in_item must be equal to ih_item_len");
1162 leaf_mi = leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i],
1163 tb->sbytes[i], tb->S_new[i]);
1166 "PAP-12240: unexpected value returned by leaf_move_items (%d)",
1169 /* paste into item */
1170 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
1171 leaf_paste_in_buffer(&bi, tb->item_pos - n + tb->snum[i],
1172 tb->pos_in_item, tb->insert_size[0],
1173 body, tb->zeroes_num);
1175 pasted = item_head(tb->S_new[i], tb->item_pos - n +
1177 if (is_direntry_le_ih(pasted))
1178 leaf_paste_entries(&bi, tb->item_pos - n + tb->snum[i],
1180 (struct reiserfs_de_head *)body,
1181 body + DEH_SIZE, tb->insert_size[0]);
1183 /* if we paste to indirect item update ih_free_space */
1184 if (is_indirect_le_ih(pasted))
1185 set_ih_free_space(pasted, 0);
1187 tb->zeroes_num = tb->insert_size[0] = 0;
1190 static void balance_leaf_new_nodes_paste(struct tree_balance *tb,
1191 struct item_head *ih,
1193 struct item_head *insert_key,
1194 struct buffer_head **insert_ptr,
1197 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1198 int n = B_NR_ITEMS(tbS0);
1200 /* pasted item doesn't fall into S_new[i] */
1201 if (n - tb->snum[i] > tb->item_pos) {
1202 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
1203 tb->snum[i], tb->sbytes[i], tb->S_new[i]);
1207 /* pasted item or part if it falls to S_new[i] */
1209 if (tb->item_pos == n - tb->snum[i] && tb->sbytes[i] != -1)
1210 /* we must shift part of the appended item */
1211 balance_leaf_new_nodes_paste_shift(tb, ih, body, insert_key,
1214 /* item falls wholly into S_new[i] */
1215 balance_leaf_new_nodes_paste_whole(tb, ih, body, insert_key,
1219 /* Fill new nodes that appear in place of S[0] */
1220 static void balance_leaf_new_nodes(struct tree_balance *tb,
1221 struct item_head *ih,
1223 struct item_head *insert_key,
1224 struct buffer_head **insert_ptr,
1228 for (i = tb->blknum[0] - 2; i >= 0; i--) {
1229 BUG_ON(flag != M_INSERT && flag != M_PASTE);
1231 RFALSE(!tb->snum[i],
1232 "PAP-12200: snum[%d] == %d. Must be > 0", i,
1235 /* here we shift from S to S_new nodes */
1237 tb->S_new[i] = get_FEB(tb);
1239 /* initialized block type and tree level */
1240 set_blkh_level(B_BLK_HEAD(tb->S_new[i]), DISK_LEAF_NODE_LEVEL);
1242 if (flag == M_INSERT)
1243 balance_leaf_new_nodes_insert(tb, ih, body, insert_key,
1246 balance_leaf_new_nodes_paste(tb, ih, body, insert_key,
1249 memcpy(insert_key + i, leaf_key(tb->S_new[i], 0), KEY_SIZE);
1250 insert_ptr[i] = tb->S_new[i];
1252 RFALSE(!buffer_journaled(tb->S_new[i])
1253 || buffer_journal_dirty(tb->S_new[i])
1254 || buffer_dirty(tb->S_new[i]),
1255 "PAP-12247: S_new[%d] : (%b)",
1256 i, format_bh(tb->S_new[i]));
1260 static void balance_leaf_finish_node_insert(struct tree_balance *tb,
1261 struct item_head *ih,
1264 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1265 struct buffer_info bi;
1266 buffer_info_init_tbS0(tb, &bi);
1267 leaf_insert_into_buf(&bi, tb->item_pos, ih,
1268 body, tb->zeroes_num);
1271 * If we insert the first key
1272 * change the delimiting key
1274 if (tb->item_pos == 0) {
1275 if (tb->CFL[0]) /* can be 0 in reiserfsck */
1276 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0);
1280 static void balance_leaf_finish_node_paste(struct tree_balance *tb,
1281 struct item_head *ih,
1284 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1285 struct buffer_info bi;
1286 struct item_head *pasted;
1288 pasted = item_head(tbS0, tb->item_pos);
1289 /* when directory, may be new entry already pasted */
1290 if (is_direntry_le_ih(pasted)) {
1291 if (tb->pos_in_item >= 0 && tb->pos_in_item <= ih_entry_count(pasted)) {
1293 RFALSE(!tb->insert_size[0],
1294 "PAP-12260: insert_size is 0 already");
1297 buffer_info_init_tbS0(tb, &bi);
1298 leaf_paste_in_buffer(&bi, tb->item_pos, tb->pos_in_item,
1299 tb->insert_size[0], body,
1303 leaf_paste_entries(&bi, tb->item_pos, tb->pos_in_item, 1,
1304 (struct reiserfs_de_head *)body,
1306 tb->insert_size[0]);
1307 if (!tb->item_pos && !tb->pos_in_item) {
1308 RFALSE(!tb->CFL[0] || !tb->L[0],
1309 "PAP-12270: CFL[0]/L[0] must be specified");
1311 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0);
1313 tb->insert_size[0] = 0;
1315 } else { /* regular object */
1316 if (tb->pos_in_item == ih_item_len(pasted)) {
1318 RFALSE(tb->insert_size[0] <= 0,
1319 "PAP-12275: insert size must not be %d",
1320 tb->insert_size[0]);
1321 buffer_info_init_tbS0(tb, &bi);
1322 leaf_paste_in_buffer(&bi, tb->item_pos, tb->pos_in_item,
1323 tb->insert_size[0], body, tb->zeroes_num);
1325 if (is_indirect_le_ih(pasted)) {
1330 "PAP-12280: insert_size for indirect item must be %d, not %d",
1335 set_ih_free_space(pasted, 0);
1337 tb->insert_size[0] = 0;
1339 #ifdef CONFIG_REISERFS_CHECK
1341 if (tb->insert_size[0]) {
1342 print_cur_tb("12285");
1343 reiserfs_panic(tb->tb_sb,
1348 tb->insert_size[0]);
1351 #endif /* CONFIG_REISERFS_CHECK */
1357 * if the affected item was not wholly shifted then we
1358 * perform all necessary operations on that part or whole
1359 * of the affected item which remains in S
1361 static void balance_leaf_finish_node(struct tree_balance *tb,
1362 struct item_head *ih,
1363 const char *body, int flag)
1365 /* if we must insert or append into buffer S[0] */
1366 if (0 <= tb->item_pos && tb->item_pos < tb->s0num) {
1367 if (flag == M_INSERT)
1368 balance_leaf_finish_node_insert(tb, ih, body);
1370 balance_leaf_finish_node_paste(tb, ih, body);
1375 * balance_leaf - reiserfs tree balancing algorithm
1376 * @tb: tree balance state
1377 * @ih: item header of inserted item (little endian)
1378 * @body: body of inserted item or bytes to paste
1379 * @flag: i - insert, d - delete, c - cut, p - paste (see do_balance)
1381 * @insert_key: key to insert new nodes
1382 * @insert_ptr: array of nodes to insert at the next level
1384 * In our processing of one level we sometimes determine what must be
1385 * inserted into the next higher level. This insertion consists of a
1386 * key or two keys and their corresponding pointers.
1388 static int balance_leaf(struct tree_balance *tb, struct item_head *ih,
1389 const char *body, int flag,
1390 struct item_head *insert_key,
1391 struct buffer_head **insert_ptr)
1393 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1395 PROC_INFO_INC(tb->tb_sb, balance_at[0]);
1397 /* Make balance in case insert_size[0] < 0 */
1398 if (tb->insert_size[0] < 0)
1399 return balance_leaf_when_delete(tb, flag);
1401 tb->item_pos = PATH_LAST_POSITION(tb->tb_path),
1402 tb->pos_in_item = tb->tb_path->pos_in_item,
1404 if (flag == M_INSERT && !body)
1405 tb->zeroes_num = ih_item_len(ih);
1408 * for indirect item pos_in_item is measured in unformatted node
1409 * pointers. Recalculate to bytes
1411 if (flag != M_INSERT
1412 && is_indirect_le_ih(item_head(tbS0, tb->item_pos)))
1413 tb->pos_in_item *= UNFM_P_SIZE;
1415 balance_leaf_left(tb, ih, body, flag);
1417 /* tb->lnum[0] > 0 */
1418 /* Calculate new item position */
1419 tb->item_pos -= (tb->lnum[0] - ((tb->lbytes != -1) ? 1 : 0));
1421 balance_leaf_right(tb, ih, body, flag);
1423 /* tb->rnum[0] > 0 */
1424 RFALSE(tb->blknum[0] > 3,
1425 "PAP-12180: blknum can not be %d. It must be <= 3", tb->blknum[0]);
1426 RFALSE(tb->blknum[0] < 0,
1427 "PAP-12185: blknum can not be %d. It must be >= 0", tb->blknum[0]);
1430 * if while adding to a node we discover that it is possible to split
1431 * it in two, and merge the left part into the left neighbor and the
1432 * right part into the right neighbor, eliminating the node
1434 if (tb->blknum[0] == 0) { /* node S[0] is empty now */
1436 RFALSE(!tb->lnum[0] || !tb->rnum[0],
1437 "PAP-12190: lnum and rnum must not be zero");
1439 * if insertion was done before 0-th position in R[0], right
1440 * delimiting key of the tb->L[0]'s and left delimiting key are
1445 reiserfs_panic(tb->tb_sb, "vs-12195",
1446 "CFR not initialized");
1447 copy_key(internal_key(tb->CFL[0], tb->lkey[0]),
1448 internal_key(tb->CFR[0], tb->rkey[0]));
1449 do_balance_mark_internal_dirty(tb, tb->CFL[0], 0);
1452 reiserfs_invalidate_buffer(tb, tbS0);
1456 balance_leaf_new_nodes(tb, ih, body, insert_key, insert_ptr, flag);
1458 balance_leaf_finish_node(tb, ih, body, flag);
1460 #ifdef CONFIG_REISERFS_CHECK
1461 if (flag == M_PASTE && tb->insert_size[0]) {
1462 print_cur_tb("12290");
1463 reiserfs_panic(tb->tb_sb,
1464 "PAP-12290", "insert_size is still not 0 (%d)",
1465 tb->insert_size[0]);
1469 /* Leaf level of the tree is balanced (end of balance_leaf) */
1473 /* Make empty node */
1474 void make_empty_node(struct buffer_info *bi)
1476 struct block_head *blkh;
1478 RFALSE(bi->bi_bh == NULL, "PAP-12295: pointer to the buffer is NULL");
1480 blkh = B_BLK_HEAD(bi->bi_bh);
1481 set_blkh_nr_item(blkh, 0);
1482 set_blkh_free_space(blkh, MAX_CHILD_SIZE(bi->bi_bh));
1485 B_N_CHILD(bi->bi_parent, bi->bi_position)->dc_size = 0; /* Endian safe if 0 */
1488 /* Get first empty buffer */
1489 struct buffer_head *get_FEB(struct tree_balance *tb)
1492 struct buffer_info bi;
1494 for (i = 0; i < MAX_FEB_SIZE; i++)
1495 if (tb->FEB[i] != NULL)
1498 if (i == MAX_FEB_SIZE)
1499 reiserfs_panic(tb->tb_sb, "vs-12300", "FEB list is empty");
1501 buffer_info_init_bh(tb, &bi, tb->FEB[i]);
1502 make_empty_node(&bi);
1503 set_buffer_uptodate(tb->FEB[i]);
1504 tb->used[i] = tb->FEB[i];
1510 /* This is now used because reiserfs_free_block has to be able to schedule. */
1511 static void store_thrown(struct tree_balance *tb, struct buffer_head *bh)
1515 if (buffer_dirty(bh))
1516 reiserfs_warning(tb->tb_sb, "reiserfs-12320",
1517 "called with dirty buffer");
1518 for (i = 0; i < ARRAY_SIZE(tb->thrown); i++)
1519 if (!tb->thrown[i]) {
1521 get_bh(bh); /* free_thrown puts this */
1524 reiserfs_warning(tb->tb_sb, "reiserfs-12321",
1525 "too many thrown buffers");
1528 static void free_thrown(struct tree_balance *tb)
1531 b_blocknr_t blocknr;
1532 for (i = 0; i < ARRAY_SIZE(tb->thrown); i++) {
1533 if (tb->thrown[i]) {
1534 blocknr = tb->thrown[i]->b_blocknr;
1535 if (buffer_dirty(tb->thrown[i]))
1536 reiserfs_warning(tb->tb_sb, "reiserfs-12322",
1537 "called with dirty buffer %d",
1539 brelse(tb->thrown[i]); /* incremented in store_thrown */
1540 reiserfs_free_block(tb->transaction_handle, NULL,
1546 void reiserfs_invalidate_buffer(struct tree_balance *tb, struct buffer_head *bh)
1548 struct block_head *blkh;
1549 blkh = B_BLK_HEAD(bh);
1550 set_blkh_level(blkh, FREE_LEVEL);
1551 set_blkh_nr_item(blkh, 0);
1553 clear_buffer_dirty(bh);
1554 store_thrown(tb, bh);
1557 /* Replace n_dest'th key in buffer dest by n_src'th key of buffer src.*/
1558 void replace_key(struct tree_balance *tb, struct buffer_head *dest, int n_dest,
1559 struct buffer_head *src, int n_src)
1562 RFALSE(dest == NULL || src == NULL,
1563 "vs-12305: source or destination buffer is 0 (src=%p, dest=%p)",
1565 RFALSE(!B_IS_KEYS_LEVEL(dest),
1566 "vs-12310: invalid level (%z) for destination buffer. dest must be leaf",
1568 RFALSE(n_dest < 0 || n_src < 0,
1569 "vs-12315: src(%d) or dest(%d) key number < 0", n_src, n_dest);
1570 RFALSE(n_dest >= B_NR_ITEMS(dest) || n_src >= B_NR_ITEMS(src),
1571 "vs-12320: src(%d(%d)) or dest(%d(%d)) key number is too big",
1572 n_src, B_NR_ITEMS(src), n_dest, B_NR_ITEMS(dest));
1574 if (B_IS_ITEMS_LEVEL(src))
1575 /* source buffer contains leaf node */
1576 memcpy(internal_key(dest, n_dest), item_head(src, n_src),
1579 memcpy(internal_key(dest, n_dest), internal_key(src, n_src),
1582 do_balance_mark_internal_dirty(tb, dest, 0);
1585 int get_left_neighbor_position(struct tree_balance *tb, int h)
1587 int Sh_position = PATH_H_POSITION(tb->tb_path, h + 1);
1589 RFALSE(PATH_H_PPARENT(tb->tb_path, h) == NULL || tb->FL[h] == NULL,
1590 "vs-12325: FL[%d](%p) or F[%d](%p) does not exist",
1591 h, tb->FL[h], h, PATH_H_PPARENT(tb->tb_path, h));
1593 if (Sh_position == 0)
1594 return B_NR_ITEMS(tb->FL[h]);
1596 return Sh_position - 1;
1599 int get_right_neighbor_position(struct tree_balance *tb, int h)
1601 int Sh_position = PATH_H_POSITION(tb->tb_path, h + 1);
1603 RFALSE(PATH_H_PPARENT(tb->tb_path, h) == NULL || tb->FR[h] == NULL,
1604 "vs-12330: F[%d](%p) or FR[%d](%p) does not exist",
1605 h, PATH_H_PPARENT(tb->tb_path, h), h, tb->FR[h]);
1607 if (Sh_position == B_NR_ITEMS(PATH_H_PPARENT(tb->tb_path, h)))
1610 return Sh_position + 1;
1613 #ifdef CONFIG_REISERFS_CHECK
1615 int is_reusable(struct super_block *s, b_blocknr_t block, int bit_value);
1616 static void check_internal_node(struct super_block *s, struct buffer_head *bh,
1619 struct disk_child *dc;
1622 RFALSE(!bh, "PAP-12336: bh == 0");
1624 if (!bh || !B_IS_IN_TREE(bh))
1627 RFALSE(!buffer_dirty(bh) &&
1628 !(buffer_journaled(bh) || buffer_journal_dirty(bh)),
1629 "PAP-12337: buffer (%b) must be dirty", bh);
1630 dc = B_N_CHILD(bh, 0);
1632 for (i = 0; i <= B_NR_ITEMS(bh); i++, dc++) {
1633 if (!is_reusable(s, dc_block_number(dc), 1)) {
1635 reiserfs_panic(s, "PAP-12338",
1636 "invalid child pointer %y in %b",
1642 static int locked_or_not_in_tree(struct tree_balance *tb,
1643 struct buffer_head *bh, char *which)
1645 if ((!buffer_journal_prepared(bh) && buffer_locked(bh)) ||
1646 !B_IS_IN_TREE(bh)) {
1647 reiserfs_warning(tb->tb_sb, "vs-12339", "%s (%b)", which, bh);
1653 static int check_before_balancing(struct tree_balance *tb)
1657 if (REISERFS_SB(tb->tb_sb)->cur_tb) {
1658 reiserfs_panic(tb->tb_sb, "vs-12335", "suspect that schedule "
1659 "occurred based on cur_tb not being null at "
1660 "this point in code. do_balance cannot properly "
1661 "handle concurrent tree accesses on a same "
1666 * double check that buffers that we will modify are unlocked.
1667 * (fix_nodes should already have prepped all of these for us).
1670 retval |= locked_or_not_in_tree(tb, tb->L[0], "L[0]");
1671 retval |= locked_or_not_in_tree(tb, tb->FL[0], "FL[0]");
1672 retval |= locked_or_not_in_tree(tb, tb->CFL[0], "CFL[0]");
1673 check_leaf(tb->L[0]);
1676 retval |= locked_or_not_in_tree(tb, tb->R[0], "R[0]");
1677 retval |= locked_or_not_in_tree(tb, tb->FR[0], "FR[0]");
1678 retval |= locked_or_not_in_tree(tb, tb->CFR[0], "CFR[0]");
1679 check_leaf(tb->R[0]);
1681 retval |= locked_or_not_in_tree(tb, PATH_PLAST_BUFFER(tb->tb_path),
1683 check_leaf(PATH_PLAST_BUFFER(tb->tb_path));
1688 static void check_after_balance_leaf(struct tree_balance *tb)
1691 if (B_FREE_SPACE(tb->L[0]) !=
1692 MAX_CHILD_SIZE(tb->L[0]) -
1694 (tb->FL[0], get_left_neighbor_position(tb, 0)))) {
1695 print_cur_tb("12221");
1696 reiserfs_panic(tb->tb_sb, "PAP-12355",
1697 "shift to left was incorrect");
1701 if (B_FREE_SPACE(tb->R[0]) !=
1702 MAX_CHILD_SIZE(tb->R[0]) -
1704 (tb->FR[0], get_right_neighbor_position(tb, 0)))) {
1705 print_cur_tb("12222");
1706 reiserfs_panic(tb->tb_sb, "PAP-12360",
1707 "shift to right was incorrect");
1710 if (PATH_H_PBUFFER(tb->tb_path, 1) &&
1711 (B_FREE_SPACE(PATH_H_PBUFFER(tb->tb_path, 0)) !=
1712 (MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)) -
1713 dc_size(B_N_CHILD(PATH_H_PBUFFER(tb->tb_path, 1),
1714 PATH_H_POSITION(tb->tb_path, 1)))))) {
1715 int left = B_FREE_SPACE(PATH_H_PBUFFER(tb->tb_path, 0));
1716 int right = (MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)) -
1717 dc_size(B_N_CHILD(PATH_H_PBUFFER(tb->tb_path, 1),
1718 PATH_H_POSITION(tb->tb_path,
1720 print_cur_tb("12223");
1721 reiserfs_warning(tb->tb_sb, "reiserfs-12363",
1722 "B_FREE_SPACE (PATH_H_PBUFFER(tb->tb_path,0)) = %d; "
1723 "MAX_CHILD_SIZE (%d) - dc_size( %y, %d ) [%d] = %d",
1725 MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)),
1726 PATH_H_PBUFFER(tb->tb_path, 1),
1727 PATH_H_POSITION(tb->tb_path, 1),
1729 (PATH_H_PBUFFER(tb->tb_path, 1),
1730 PATH_H_POSITION(tb->tb_path, 1))),
1732 reiserfs_panic(tb->tb_sb, "PAP-12365", "S is incorrect");
1736 static void check_leaf_level(struct tree_balance *tb)
1738 check_leaf(tb->L[0]);
1739 check_leaf(tb->R[0]);
1740 check_leaf(PATH_PLAST_BUFFER(tb->tb_path));
1743 static void check_internal_levels(struct tree_balance *tb)
1747 /* check all internal nodes */
1748 for (h = 1; tb->insert_size[h]; h++) {
1749 check_internal_node(tb->tb_sb, PATH_H_PBUFFER(tb->tb_path, h),
1750 "BAD BUFFER ON PATH");
1752 check_internal_node(tb->tb_sb, tb->L[h], "BAD L");
1754 check_internal_node(tb->tb_sb, tb->R[h], "BAD R");
1762 * Now we have all of the buffers that must be used in balancing of
1763 * the tree. We rely on the assumption that schedule() will not occur
1764 * while do_balance works. ( Only interrupt handlers are acceptable.)
1765 * We balance the tree according to the analysis made before this,
1766 * using buffers already obtained. For SMP support it will someday be
1767 * necessary to add ordered locking of tb.
1771 * Some interesting rules of balancing:
1772 * we delete a maximum of two nodes per level per balancing: we never
1773 * delete R, when we delete two of three nodes L, S, R then we move
1776 * we only delete L if we are deleting two nodes, if we delete only
1777 * one node we delete S
1779 * if we shift leaves then we shift as much as we can: this is a
1780 * deliberate policy of extremism in node packing which results in
1781 * higher average utilization after repeated random balance operations
1782 * at the cost of more memory copies and more balancing as a result of
1783 * small insertions to full nodes.
1785 * if we shift internal nodes we try to evenly balance the node
1786 * utilization, with consequent less balancing at the cost of lower
1789 * one could argue that the policy for directories in leaves should be
1790 * that of internal nodes, but we will wait until another day to
1791 * evaluate this.... It would be nice to someday measure and prove
1792 * these assumptions as to what is optimal....
1795 static inline void do_balance_starts(struct tree_balance *tb)
1797 /* use print_cur_tb() to see initial state of struct tree_balance */
1799 /* store_print_tb (tb); */
1801 /* do not delete, just comment it out */
1803 print_tb(flag, PATH_LAST_POSITION(tb->tb_path),
1804 tb->tb_path->pos_in_item, tb, "check");
1806 RFALSE(check_before_balancing(tb), "PAP-12340: locked buffers in TB");
1807 #ifdef CONFIG_REISERFS_CHECK
1808 REISERFS_SB(tb->tb_sb)->cur_tb = tb;
1812 static inline void do_balance_completed(struct tree_balance *tb)
1815 #ifdef CONFIG_REISERFS_CHECK
1816 check_leaf_level(tb);
1817 check_internal_levels(tb);
1818 REISERFS_SB(tb->tb_sb)->cur_tb = NULL;
1822 * reiserfs_free_block is no longer schedule safe. So, we need to
1823 * put the buffers we want freed on the thrown list during do_balance,
1824 * and then free them now
1827 REISERFS_SB(tb->tb_sb)->s_do_balance++;
1829 /* release all nodes hold to perform the balancing */
1836 * do_balance - balance the tree
1838 * @tb: tree_balance structure
1839 * @ih: item header of inserted item
1840 * @body: body of inserted item or bytes to paste
1841 * @flag: 'i' - insert, 'd' - delete, 'c' - cut, 'p' paste
1843 * Cut means delete part of an item (includes removing an entry from a
1846 * Delete means delete whole item.
1848 * Insert means add a new item into the tree.
1850 * Paste means to append to the end of an existing file or to
1851 * insert a directory entry.
1853 void do_balance(struct tree_balance *tb, struct item_head *ih,
1854 const char *body, int flag)
1856 int child_pos; /* position of a child node in its parent */
1857 int h; /* level of the tree being processed */
1860 * in our processing of one level we sometimes determine what
1861 * must be inserted into the next higher level. This insertion
1862 * consists of a key or two keys and their corresponding
1865 struct item_head insert_key[2];
1867 /* inserted node-ptrs for the next level */
1868 struct buffer_head *insert_ptr[2];
1871 tb->need_balance_dirty = 0;
1873 if (FILESYSTEM_CHANGED_TB(tb)) {
1874 reiserfs_panic(tb->tb_sb, "clm-6000", "fs generation has "
1877 /* if we have no real work to do */
1878 if (!tb->insert_size[0]) {
1879 reiserfs_warning(tb->tb_sb, "PAP-12350",
1880 "insert_size == 0, mode == %c", flag);
1885 atomic_inc(&fs_generation(tb->tb_sb));
1886 do_balance_starts(tb);
1889 * balance_leaf returns 0 except if combining L R and S into
1890 * one node. see balance_internal() for explanation of this
1893 child_pos = PATH_H_B_ITEM_ORDER(tb->tb_path, 0) +
1894 balance_leaf(tb, ih, body, flag, insert_key, insert_ptr);
1896 #ifdef CONFIG_REISERFS_CHECK
1897 check_after_balance_leaf(tb);
1900 /* Balance internal level of the tree. */
1901 for (h = 1; h < MAX_HEIGHT && tb->insert_size[h]; h++)
1903 balance_internal(tb, h, child_pos, insert_key, insert_ptr);
1905 do_balance_completed(tb);