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 <linux/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.
77 static void balance_leaf_when_delete_del(struct tree_balance *tb)
79 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
80 int item_pos = PATH_LAST_POSITION(tb->tb_path);
81 struct buffer_info bi;
82 #ifdef CONFIG_REISERFS_CHECK
83 struct item_head *ih = item_head(tbS0, item_pos);
86 RFALSE(ih_item_len(ih) + IH_SIZE != -tb->insert_size[0],
87 "vs-12013: mode Delete, insert size %d, ih to be deleted %h",
88 -tb->insert_size[0], ih);
90 buffer_info_init_tbS0(tb, &bi);
91 leaf_delete_items(&bi, 0, item_pos, 1, -1);
93 if (!item_pos && tb->CFL[0]) {
94 if (B_NR_ITEMS(tbS0)) {
95 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0);
97 if (!PATH_H_POSITION(tb->tb_path, 1))
98 replace_key(tb, tb->CFL[0], tb->lkey[0],
99 PATH_H_PPARENT(tb->tb_path, 0), 0);
103 RFALSE(!item_pos && !tb->CFL[0],
104 "PAP-12020: tb->CFL[0]==%p, tb->L[0]==%p", tb->CFL[0],
108 /* cut item in S[0] */
109 static void balance_leaf_when_delete_cut(struct tree_balance *tb)
111 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
112 int item_pos = PATH_LAST_POSITION(tb->tb_path);
113 struct item_head *ih = item_head(tbS0, item_pos);
114 int pos_in_item = tb->tb_path->pos_in_item;
115 struct buffer_info bi;
116 buffer_info_init_tbS0(tb, &bi);
118 if (is_direntry_le_ih(ih)) {
120 * UFS unlink semantics are such that you can only
121 * delete one directory entry at a time.
123 * when we cut a directory tb->insert_size[0] means
124 * number of entries to be cut (always 1)
126 tb->insert_size[0] = -1;
127 leaf_cut_from_buffer(&bi, item_pos, pos_in_item,
128 -tb->insert_size[0]);
130 RFALSE(!item_pos && !pos_in_item && !tb->CFL[0],
131 "PAP-12030: can not change delimiting key. CFL[0]=%p",
134 if (!item_pos && !pos_in_item && tb->CFL[0])
135 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0);
137 leaf_cut_from_buffer(&bi, item_pos, pos_in_item,
138 -tb->insert_size[0]);
140 RFALSE(!ih_item_len(ih),
141 "PAP-12035: cut must leave non-zero dynamic "
146 static int balance_leaf_when_delete_left(struct tree_balance *tb)
148 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
149 int n = B_NR_ITEMS(tbS0);
151 /* L[0] must be joined with S[0] */
152 if (tb->lnum[0] == -1) {
153 /* R[0] must be also joined with S[0] */
154 if (tb->rnum[0] == -1) {
155 if (tb->FR[0] == PATH_H_PPARENT(tb->tb_path, 0)) {
157 * all contents of all the
158 * 3 buffers will be in L[0]
160 if (PATH_H_POSITION(tb->tb_path, 1) == 0 &&
161 1 < B_NR_ITEMS(tb->FR[0]))
162 replace_key(tb, tb->CFL[0],
163 tb->lkey[0], tb->FR[0], 1);
165 leaf_move_items(LEAF_FROM_S_TO_L, tb, n, -1,
167 leaf_move_items(LEAF_FROM_R_TO_L, tb,
168 B_NR_ITEMS(tb->R[0]), -1,
171 reiserfs_invalidate_buffer(tb, tbS0);
172 reiserfs_invalidate_buffer(tb, tb->R[0]);
177 /* all contents of all the 3 buffers will be in R[0] */
178 leaf_move_items(LEAF_FROM_S_TO_R, tb, n, -1, NULL);
179 leaf_move_items(LEAF_FROM_L_TO_R, tb,
180 B_NR_ITEMS(tb->L[0]), -1, NULL);
182 /* right_delimiting_key is correct in R[0] */
183 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0);
185 reiserfs_invalidate_buffer(tb, tbS0);
186 reiserfs_invalidate_buffer(tb, tb->L[0]);
191 RFALSE(tb->rnum[0] != 0,
192 "PAP-12045: rnum must be 0 (%d)", tb->rnum[0]);
193 /* all contents of L[0] and S[0] will be in L[0] */
194 leaf_shift_left(tb, n, -1);
196 reiserfs_invalidate_buffer(tb, tbS0);
202 * a part of contents of S[0] will be in L[0] and
203 * the rest part of S[0] will be in R[0]
206 RFALSE((tb->lnum[0] + tb->rnum[0] < n) ||
207 (tb->lnum[0] + tb->rnum[0] > n + 1),
208 "PAP-12050: rnum(%d) and lnum(%d) and item "
209 "number(%d) in S[0] are not consistent",
210 tb->rnum[0], tb->lnum[0], n);
211 RFALSE((tb->lnum[0] + tb->rnum[0] == n) &&
212 (tb->lbytes != -1 || tb->rbytes != -1),
213 "PAP-12055: bad rbytes (%d)/lbytes (%d) "
214 "parameters when items are not split",
215 tb->rbytes, tb->lbytes);
216 RFALSE((tb->lnum[0] + tb->rnum[0] == n + 1) &&
217 (tb->lbytes < 1 || tb->rbytes != -1),
218 "PAP-12060: bad rbytes (%d)/lbytes (%d) "
219 "parameters when items are split",
220 tb->rbytes, tb->lbytes);
222 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
223 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
225 reiserfs_invalidate_buffer(tb, tbS0);
231 * Balance leaf node in case of delete or cut: insert_size[0] < 0
233 * lnum, rnum can have values >= -1
234 * -1 means that the neighbor must be joined with S
235 * 0 means that nothing should be done with the neighbor
236 * >0 means to shift entirely or partly the specified number of items
239 static int balance_leaf_when_delete(struct tree_balance *tb, int flag)
241 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
242 struct buffer_info bi;
245 RFALSE(tb->FR[0] && B_LEVEL(tb->FR[0]) != DISK_LEAF_NODE_LEVEL + 1,
246 "vs- 12000: level: wrong FR %z", tb->FR[0]);
247 RFALSE(tb->blknum[0] > 1,
248 "PAP-12005: tb->blknum == %d, can not be > 1", tb->blknum[0]);
249 RFALSE(!tb->blknum[0] && !PATH_H_PPARENT(tb->tb_path, 0),
250 "PAP-12010: tree can not be empty");
252 buffer_info_init_tbS0(tb, &bi);
254 /* Delete or truncate the item */
256 BUG_ON(flag != M_DELETE && flag != M_CUT);
257 if (flag == M_DELETE)
258 balance_leaf_when_delete_del(tb);
260 balance_leaf_when_delete_cut(tb);
264 * the rule is that no shifting occurs unless by shifting
265 * a node can be freed
267 n = B_NR_ITEMS(tbS0);
270 /* L[0] takes part in balancing */
272 return balance_leaf_when_delete_left(tb);
274 if (tb->rnum[0] == -1) {
275 /* all contents of R[0] and S[0] will be in R[0] */
276 leaf_shift_right(tb, n, -1);
277 reiserfs_invalidate_buffer(tb, tbS0);
282 "PAP-12065: bad rnum parameter must be 0 (%d)", tb->rnum[0]);
286 static unsigned int balance_leaf_insert_left(struct tree_balance *tb,
287 struct item_head *const ih,
288 const char * const body)
291 struct buffer_info bi;
292 int n = B_NR_ITEMS(tb->L[0]);
293 unsigned body_shift_bytes = 0;
295 if (tb->item_pos == tb->lnum[0] - 1 && tb->lbytes != -1) {
296 /* part of new item falls into L[0] */
297 int new_item_len, shift;
299 ret = leaf_shift_left(tb, tb->lnum[0] - 1, -1);
301 /* Calculate item length to insert to S[0] */
302 new_item_len = ih_item_len(ih) - tb->lbytes;
304 /* Calculate and check item length to insert to L[0] */
305 put_ih_item_len(ih, ih_item_len(ih) - new_item_len);
307 RFALSE(ih_item_len(ih) <= 0,
308 "PAP-12080: there is nothing to insert into L[0]: "
309 "ih_item_len=%d", ih_item_len(ih));
311 /* Insert new item into L[0] */
312 buffer_info_init_left(tb, &bi);
313 leaf_insert_into_buf(&bi, n + tb->item_pos - ret, ih, body,
314 min_t(int, tb->zeroes_num, ih_item_len(ih)));
317 * Calculate key component, item length and body to
321 if (is_indirect_le_ih(ih))
322 shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
324 add_le_ih_k_offset(ih, tb->lbytes << shift);
326 put_ih_item_len(ih, new_item_len);
327 if (tb->lbytes > tb->zeroes_num) {
328 body_shift_bytes = tb->lbytes - tb->zeroes_num;
331 tb->zeroes_num -= tb->lbytes;
333 RFALSE(ih_item_len(ih) <= 0,
334 "PAP-12085: there is nothing to insert into S[0]: "
335 "ih_item_len=%d", ih_item_len(ih));
337 /* new item in whole falls into L[0] */
338 /* Shift lnum[0]-1 items to L[0] */
339 ret = leaf_shift_left(tb, tb->lnum[0] - 1, tb->lbytes);
341 /* Insert new item into L[0] */
342 buffer_info_init_left(tb, &bi);
343 leaf_insert_into_buf(&bi, n + tb->item_pos - ret, ih, body,
345 tb->insert_size[0] = 0;
348 return body_shift_bytes;
351 static void balance_leaf_paste_left_shift_dirent(struct tree_balance *tb,
352 struct item_head * const ih,
353 const char * const body)
355 int n = B_NR_ITEMS(tb->L[0]);
356 struct buffer_info bi;
358 RFALSE(tb->zeroes_num,
359 "PAP-12090: invalid parameter in case of a directory");
362 if (tb->lbytes > tb->pos_in_item) {
363 /* new directory entry falls into L[0] */
364 struct item_head *pasted;
365 int ret, l_pos_in_item = tb->pos_in_item;
368 * Shift lnum[0] - 1 items in whole.
369 * Shift lbytes - 1 entries from given directory item
371 ret = leaf_shift_left(tb, tb->lnum[0], tb->lbytes - 1);
372 if (ret && !tb->item_pos) {
373 pasted = item_head(tb->L[0], B_NR_ITEMS(tb->L[0]) - 1);
374 l_pos_in_item += ih_entry_count(pasted) -
378 /* Append given directory entry to directory item */
379 buffer_info_init_left(tb, &bi);
380 leaf_paste_in_buffer(&bi, n + tb->item_pos - ret,
381 l_pos_in_item, tb->insert_size[0],
382 body, tb->zeroes_num);
385 * previous string prepared space for pasting new entry,
386 * following string pastes this entry
390 * when we have merge directory item, pos_in_item
391 * has been changed too
394 /* paste new directory entry. 1 is entry number */
395 leaf_paste_entries(&bi, n + tb->item_pos - ret,
397 (struct reiserfs_de_head *) body,
398 body + DEH_SIZE, tb->insert_size[0]);
399 tb->insert_size[0] = 0;
401 /* new directory item doesn't fall into L[0] */
403 * Shift lnum[0]-1 items in whole. Shift lbytes
404 * directory entries from directory item number lnum[0]
406 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
409 /* Calculate new position to append in item body */
410 tb->pos_in_item -= tb->lbytes;
413 static unsigned int balance_leaf_paste_left_shift(struct tree_balance *tb,
414 struct item_head * const ih,
415 const char * const body)
417 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
418 int n = B_NR_ITEMS(tb->L[0]);
419 struct buffer_info bi;
420 int body_shift_bytes = 0;
422 if (is_direntry_le_ih(item_head(tbS0, tb->item_pos))) {
423 balance_leaf_paste_left_shift_dirent(tb, ih, body);
427 RFALSE(tb->lbytes <= 0,
428 "PAP-12095: there is nothing to shift to L[0]. "
429 "lbytes=%d", tb->lbytes);
430 RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)),
431 "PAP-12100: incorrect position to paste: "
432 "item_len=%d, pos_in_item=%d",
433 ih_item_len(item_head(tbS0, tb->item_pos)), tb->pos_in_item);
435 /* appended item will be in L[0] in whole */
436 if (tb->lbytes >= tb->pos_in_item) {
437 struct item_head *tbS0_pos_ih, *tbL0_ih;
438 struct item_head *tbS0_0_ih;
439 struct reiserfs_key *left_delim_key;
440 int ret, l_n, version, temp_l;
442 tbS0_pos_ih = item_head(tbS0, tb->item_pos);
443 tbS0_0_ih = item_head(tbS0, 0);
446 * this bytes number must be appended
447 * to the last item of L[h]
449 l_n = tb->lbytes - tb->pos_in_item;
451 /* Calculate new insert_size[0] */
452 tb->insert_size[0] -= l_n;
454 RFALSE(tb->insert_size[0] <= 0,
455 "PAP-12105: there is nothing to paste into "
456 "L[0]. insert_size=%d", tb->insert_size[0]);
458 ret = leaf_shift_left(tb, tb->lnum[0],
459 ih_item_len(tbS0_pos_ih));
461 tbL0_ih = item_head(tb->L[0], n + tb->item_pos - ret);
463 /* Append to body of item in L[0] */
464 buffer_info_init_left(tb, &bi);
465 leaf_paste_in_buffer(&bi, n + tb->item_pos - ret,
466 ih_item_len(tbL0_ih), l_n, body,
467 min_t(int, l_n, tb->zeroes_num));
470 * 0-th item in S0 can be only of DIRECT type
475 RFALSE(ih_item_len(tbS0_0_ih),
476 "PAP-12106: item length must be 0");
477 RFALSE(comp_short_le_keys(&tbS0_0_ih->ih_key,
478 leaf_key(tb->L[0], n + tb->item_pos - ret)),
479 "PAP-12107: items must be of the same file");
481 if (is_indirect_le_ih(tbL0_ih)) {
482 int shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
483 temp_l = l_n << shift;
485 /* update key of first item in S0 */
486 version = ih_version(tbS0_0_ih);
487 add_le_key_k_offset(version, &tbS0_0_ih->ih_key, temp_l);
489 /* update left delimiting key */
490 left_delim_key = internal_key(tb->CFL[0], tb->lkey[0]);
491 add_le_key_k_offset(version, left_delim_key, temp_l);
494 * Calculate new body, position in item and
497 if (l_n > tb->zeroes_num) {
498 body_shift_bytes = l_n - tb->zeroes_num;
501 tb->zeroes_num -= l_n;
504 RFALSE(comp_short_le_keys(&tbS0_0_ih->ih_key,
506 B_NR_ITEMS(tb->L[0]) - 1)) ||
507 !op_is_left_mergeable(leaf_key(tbS0, 0), tbS0->b_size) ||
508 !op_is_left_mergeable(left_delim_key, tbS0->b_size),
509 "PAP-12120: item must be merge-able with left "
512 /* only part of the appended item will be in L[0] */
514 /* Calculate position in item for append in S[0] */
515 tb->pos_in_item -= tb->lbytes;
517 RFALSE(tb->pos_in_item <= 0,
518 "PAP-12125: no place for paste. pos_in_item=%d",
522 * Shift lnum[0] - 1 items in whole.
523 * Shift lbytes - 1 byte from item number lnum[0]
525 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
527 return body_shift_bytes;
531 /* appended item will be in L[0] in whole */
532 static void balance_leaf_paste_left_whole(struct tree_balance *tb,
533 struct item_head * const ih,
534 const char * const body)
536 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
537 int n = B_NR_ITEMS(tb->L[0]);
538 struct buffer_info bi;
539 struct item_head *pasted;
542 /* if we paste into first item of S[0] and it is left mergable */
544 op_is_left_mergeable(leaf_key(tbS0, 0), tbS0->b_size)) {
546 * then increment pos_in_item by the size of the
549 pasted = item_head(tb->L[0], n - 1);
550 if (is_direntry_le_ih(pasted))
551 tb->pos_in_item += ih_entry_count(pasted);
553 tb->pos_in_item += ih_item_len(pasted);
557 * Shift lnum[0] - 1 items in whole.
558 * Shift lbytes - 1 byte from item number lnum[0]
560 ret = leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
562 /* Append to body of item in L[0] */
563 buffer_info_init_left(tb, &bi);
564 leaf_paste_in_buffer(&bi, n + tb->item_pos - ret, tb->pos_in_item,
565 tb->insert_size[0], body, tb->zeroes_num);
567 /* if appended item is directory, paste entry */
568 pasted = item_head(tb->L[0], n + tb->item_pos - ret);
569 if (is_direntry_le_ih(pasted))
570 leaf_paste_entries(&bi, n + tb->item_pos - ret,
572 (struct reiserfs_de_head *)body,
573 body + DEH_SIZE, tb->insert_size[0]);
576 * if appended item is indirect item, put unformatted node
579 if (is_indirect_le_ih(pasted))
580 set_ih_free_space(pasted, 0);
582 tb->insert_size[0] = 0;
586 static unsigned int balance_leaf_paste_left(struct tree_balance *tb,
587 struct item_head * const ih,
588 const char * const body)
590 /* we must shift the part of the appended item */
591 if (tb->item_pos == tb->lnum[0] - 1 && tb->lbytes != -1)
592 return balance_leaf_paste_left_shift(tb, ih, body);
594 balance_leaf_paste_left_whole(tb, ih, body);
598 /* Shift lnum[0] items from S[0] to the left neighbor L[0] */
599 static unsigned int balance_leaf_left(struct tree_balance *tb,
600 struct item_head * const ih,
601 const char * const body, int flag)
603 if (tb->lnum[0] <= 0)
606 /* new item or it part falls to L[0], shift it too */
607 if (tb->item_pos < tb->lnum[0]) {
608 BUG_ON(flag != M_INSERT && flag != M_PASTE);
610 if (flag == M_INSERT)
611 return balance_leaf_insert_left(tb, ih, body);
613 return balance_leaf_paste_left(tb, ih, body);
615 /* new item doesn't fall into L[0] */
616 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
621 static void balance_leaf_insert_right(struct tree_balance *tb,
622 struct item_head * const ih,
623 const char * const body)
626 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
627 int n = B_NR_ITEMS(tbS0);
628 struct buffer_info bi;
630 /* new item or part of it doesn't fall into R[0] */
631 if (n - tb->rnum[0] >= tb->item_pos) {
632 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
636 /* new item or its part falls to R[0] */
638 /* part of new item falls into R[0] */
639 if (tb->item_pos == n - tb->rnum[0] + 1 && tb->rbytes != -1) {
640 loff_t old_key_comp, old_len, r_zeroes_number;
645 leaf_shift_right(tb, tb->rnum[0] - 1, -1);
647 /* Remember key component and item length */
648 old_key_comp = le_ih_k_offset(ih);
649 old_len = ih_item_len(ih);
652 * Calculate key component and item length to insert
656 if (is_indirect_le_ih(ih))
657 shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
658 offset = le_ih_k_offset(ih) + ((old_len - tb->rbytes) << shift);
659 set_le_ih_k_offset(ih, offset);
660 put_ih_item_len(ih, tb->rbytes);
662 /* Insert part of the item into R[0] */
663 buffer_info_init_right(tb, &bi);
664 if ((old_len - tb->rbytes) > tb->zeroes_num) {
666 r_body = body + (old_len - tb->rbytes) - tb->zeroes_num;
669 r_zeroes_number = tb->zeroes_num -
670 (old_len - tb->rbytes);
671 tb->zeroes_num -= r_zeroes_number;
674 leaf_insert_into_buf(&bi, 0, ih, r_body, r_zeroes_number);
676 /* Replace right delimiting key by first key in R[0] */
677 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0);
680 * Calculate key component and item length to
683 set_le_ih_k_offset(ih, old_key_comp);
684 put_ih_item_len(ih, old_len - tb->rbytes);
686 tb->insert_size[0] -= tb->rbytes;
689 /* whole new item falls into R[0] */
691 /* Shift rnum[0]-1 items to R[0] */
692 leaf_shift_right(tb, tb->rnum[0] - 1, tb->rbytes);
694 /* Insert new item into R[0] */
695 buffer_info_init_right(tb, &bi);
696 leaf_insert_into_buf(&bi, tb->item_pos - n + tb->rnum[0] - 1,
697 ih, body, tb->zeroes_num);
699 if (tb->item_pos - n + tb->rnum[0] - 1 == 0)
700 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0);
702 tb->zeroes_num = tb->insert_size[0] = 0;
707 static void balance_leaf_paste_right_shift_dirent(struct tree_balance *tb,
708 struct item_head * const ih,
709 const char * const body)
711 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
712 struct buffer_info bi;
715 RFALSE(tb->zeroes_num,
716 "PAP-12145: invalid parameter in case of a directory");
717 entry_count = ih_entry_count(item_head(tbS0, tb->item_pos));
719 /* new directory entry falls into R[0] */
720 if (entry_count - tb->rbytes < tb->pos_in_item) {
721 int paste_entry_position;
723 RFALSE(tb->rbytes - 1 >= entry_count || !tb->insert_size[0],
724 "PAP-12150: no enough of entries to shift to R[0]: "
725 "rbytes=%d, entry_count=%d", tb->rbytes, entry_count);
728 * Shift rnum[0]-1 items in whole.
729 * Shift rbytes-1 directory entries from directory
730 * item number rnum[0]
732 leaf_shift_right(tb, tb->rnum[0], tb->rbytes - 1);
734 /* Paste given directory entry to directory item */
735 paste_entry_position = tb->pos_in_item - entry_count +
737 buffer_info_init_right(tb, &bi);
738 leaf_paste_in_buffer(&bi, 0, paste_entry_position,
739 tb->insert_size[0], body, tb->zeroes_num);
742 leaf_paste_entries(&bi, 0, paste_entry_position, 1,
743 (struct reiserfs_de_head *) body,
744 body + DEH_SIZE, tb->insert_size[0]);
746 /* change delimiting keys */
747 if (paste_entry_position == 0)
748 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0);
750 tb->insert_size[0] = 0;
753 /* new directory entry doesn't fall into R[0] */
754 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
758 static void balance_leaf_paste_right_shift(struct tree_balance *tb,
759 struct item_head * const ih,
760 const char * const body)
762 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
763 int n_shift, n_rem, r_zeroes_number, version;
764 unsigned long temp_rem;
766 struct buffer_info bi;
768 /* we append to directory item */
769 if (is_direntry_le_ih(item_head(tbS0, tb->item_pos))) {
770 balance_leaf_paste_right_shift_dirent(tb, ih, body);
777 * Calculate number of bytes which must be shifted
780 n_shift = tb->rbytes - tb->insert_size[0];
784 RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)),
785 "PAP-12155: invalid position to paste. ih_item_len=%d, "
786 "pos_in_item=%d", tb->pos_in_item,
787 ih_item_len(item_head(tbS0, tb->item_pos)));
789 leaf_shift_right(tb, tb->rnum[0], n_shift);
792 * Calculate number of bytes which must remain in body
793 * after appending to R[0]
795 n_rem = tb->insert_size[0] - tb->rbytes;
801 version = ih_version(item_head(tb->R[0], 0));
803 if (is_indirect_le_key(version, leaf_key(tb->R[0], 0))) {
804 int shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
805 temp_rem = n_rem << shift;
808 add_le_key_k_offset(version, leaf_key(tb->R[0], 0), temp_rem);
809 add_le_key_k_offset(version, internal_key(tb->CFR[0], tb->rkey[0]),
812 do_balance_mark_internal_dirty(tb, tb->CFR[0], 0);
814 /* Append part of body into R[0] */
815 buffer_info_init_right(tb, &bi);
816 if (n_rem > tb->zeroes_num) {
818 r_body = body + n_rem - tb->zeroes_num;
821 r_zeroes_number = tb->zeroes_num - n_rem;
822 tb->zeroes_num -= r_zeroes_number;
825 leaf_paste_in_buffer(&bi, 0, n_shift, tb->insert_size[0] - n_rem,
826 r_body, r_zeroes_number);
828 if (is_indirect_le_ih(item_head(tb->R[0], 0)))
829 set_ih_free_space(item_head(tb->R[0], 0), 0);
831 tb->insert_size[0] = n_rem;
836 static void balance_leaf_paste_right_whole(struct tree_balance *tb,
837 struct item_head * const ih,
838 const char * const 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 * const ih,
882 const char * const body)
884 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
885 int n = B_NR_ITEMS(tbS0);
887 /* new item doesn't fall into R[0] */
888 if (n - tb->rnum[0] > tb->item_pos) {
889 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
893 /* pasted item or part of it falls to R[0] */
895 if (tb->item_pos == n - tb->rnum[0] && tb->rbytes != -1)
896 /* we must shift the part of the appended item */
897 balance_leaf_paste_right_shift(tb, ih, body);
899 /* pasted item in whole falls into R[0] */
900 balance_leaf_paste_right_whole(tb, ih, body);
903 /* shift rnum[0] items from S[0] to the right neighbor R[0] */
904 static void balance_leaf_right(struct tree_balance *tb,
905 struct item_head * const ih,
906 const char * const body, int flag)
908 if (tb->rnum[0] <= 0)
911 BUG_ON(flag != M_INSERT && flag != M_PASTE);
913 if (flag == M_INSERT)
914 balance_leaf_insert_right(tb, ih, body);
916 balance_leaf_paste_right(tb, ih, body);
919 static void balance_leaf_new_nodes_insert(struct tree_balance *tb,
920 struct item_head * const ih,
921 const char * const body,
922 struct item_head *insert_key,
923 struct buffer_head **insert_ptr,
926 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
927 int n = B_NR_ITEMS(tbS0);
928 struct buffer_info bi;
931 /* new item or it part don't falls into S_new[i] */
932 if (n - tb->snum[i] >= tb->item_pos) {
933 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
934 tb->snum[i], tb->sbytes[i], tb->S_new[i]);
938 /* new item or it's part falls to first new node S_new[i] */
940 /* part of new item falls into S_new[i] */
941 if (tb->item_pos == n - tb->snum[i] + 1 && tb->sbytes[i] != -1) {
942 int old_key_comp, old_len, r_zeroes_number;
945 /* Move snum[i]-1 items from S[0] to S_new[i] */
946 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i] - 1, -1,
949 /* Remember key component and item length */
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 * const ih,
1011 const char * const body,
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 * const ih,
1066 const char * const body,
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 * const ih,
1139 const char * const body,
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 * const ih,
1192 const char * const body,
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 * const ih,
1222 const char * const body,
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)",
1260 static void balance_leaf_finish_node_insert(struct tree_balance *tb,
1261 struct item_head * const ih,
1262 const char * const body)
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, body, tb->zeroes_num);
1269 /* If we insert the first key change the delimiting key */
1270 if (tb->item_pos == 0) {
1271 if (tb->CFL[0]) /* can be 0 in reiserfsck */
1272 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0);
1277 static void balance_leaf_finish_node_paste_dirent(struct tree_balance *tb,
1278 struct item_head * const ih,
1279 const char * const body)
1281 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1282 struct item_head *pasted = item_head(tbS0, tb->item_pos);
1283 struct buffer_info bi;
1285 if (tb->pos_in_item >= 0 && tb->pos_in_item <= ih_entry_count(pasted)) {
1286 RFALSE(!tb->insert_size[0],
1287 "PAP-12260: insert_size is 0 already");
1290 buffer_info_init_tbS0(tb, &bi);
1291 leaf_paste_in_buffer(&bi, tb->item_pos, tb->pos_in_item,
1292 tb->insert_size[0], body, tb->zeroes_num);
1295 leaf_paste_entries(&bi, tb->item_pos, tb->pos_in_item, 1,
1296 (struct reiserfs_de_head *)body,
1297 body + DEH_SIZE, tb->insert_size[0]);
1299 if (!tb->item_pos && !tb->pos_in_item) {
1300 RFALSE(!tb->CFL[0] || !tb->L[0],
1301 "PAP-12270: CFL[0]/L[0] must be specified");
1303 replace_key(tb, tb->CFL[0], tb->lkey[0],
1307 tb->insert_size[0] = 0;
1311 static void balance_leaf_finish_node_paste(struct tree_balance *tb,
1312 struct item_head * const ih,
1313 const char * const body)
1315 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1316 struct buffer_info bi;
1317 struct item_head *pasted = item_head(tbS0, tb->item_pos);
1319 /* when directory, may be new entry already pasted */
1320 if (is_direntry_le_ih(pasted)) {
1321 balance_leaf_finish_node_paste_dirent(tb, ih, body);
1325 /* regular object */
1327 if (tb->pos_in_item == ih_item_len(pasted)) {
1328 RFALSE(tb->insert_size[0] <= 0,
1329 "PAP-12275: insert size must not be %d",
1330 tb->insert_size[0]);
1331 buffer_info_init_tbS0(tb, &bi);
1332 leaf_paste_in_buffer(&bi, tb->item_pos,
1333 tb->pos_in_item, tb->insert_size[0], body,
1336 if (is_indirect_le_ih(pasted))
1337 set_ih_free_space(pasted, 0);
1339 tb->insert_size[0] = 0;
1341 #ifdef CONFIG_REISERFS_CHECK
1342 else if (tb->insert_size[0]) {
1343 print_cur_tb("12285");
1344 reiserfs_panic(tb->tb_sb, "PAP-12285",
1345 "insert_size must be 0 (%d)", tb->insert_size[0]);
1351 * if the affected item was not wholly shifted then we
1352 * perform all necessary operations on that part or whole
1353 * of the affected item which remains in S
1355 static void balance_leaf_finish_node(struct tree_balance *tb,
1356 struct item_head * const ih,
1357 const char * const body, int flag)
1359 /* if we must insert or append into buffer S[0] */
1360 if (0 <= tb->item_pos && tb->item_pos < tb->s0num) {
1361 if (flag == M_INSERT)
1362 balance_leaf_finish_node_insert(tb, ih, body);
1364 balance_leaf_finish_node_paste(tb, ih, body);
1369 * balance_leaf - reiserfs tree balancing algorithm
1370 * @tb: tree balance state
1371 * @ih: item header of inserted item (little endian)
1372 * @body: body of inserted item or bytes to paste
1373 * @flag: i - insert, d - delete, c - cut, p - paste (see do_balance)
1375 * @insert_key: key to insert new nodes
1376 * @insert_ptr: array of nodes to insert at the next level
1378 * In our processing of one level we sometimes determine what must be
1379 * inserted into the next higher level. This insertion consists of a
1380 * key or two keys and their corresponding pointers.
1382 static int balance_leaf(struct tree_balance *tb, struct item_head *ih,
1383 const char *body, int flag,
1384 struct item_head *insert_key,
1385 struct buffer_head **insert_ptr)
1387 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1389 PROC_INFO_INC(tb->tb_sb, balance_at[0]);
1391 /* Make balance in case insert_size[0] < 0 */
1392 if (tb->insert_size[0] < 0)
1393 return balance_leaf_when_delete(tb, flag);
1395 tb->item_pos = PATH_LAST_POSITION(tb->tb_path),
1396 tb->pos_in_item = tb->tb_path->pos_in_item,
1398 if (flag == M_INSERT && !body)
1399 tb->zeroes_num = ih_item_len(ih);
1402 * for indirect item pos_in_item is measured in unformatted node
1403 * pointers. Recalculate to bytes
1405 if (flag != M_INSERT
1406 && is_indirect_le_ih(item_head(tbS0, tb->item_pos)))
1407 tb->pos_in_item *= UNFM_P_SIZE;
1409 body += balance_leaf_left(tb, ih, body, flag);
1411 /* tb->lnum[0] > 0 */
1412 /* Calculate new item position */
1413 tb->item_pos -= (tb->lnum[0] - ((tb->lbytes != -1) ? 1 : 0));
1415 balance_leaf_right(tb, ih, body, flag);
1417 /* tb->rnum[0] > 0 */
1418 RFALSE(tb->blknum[0] > 3,
1419 "PAP-12180: blknum can not be %d. It must be <= 3", tb->blknum[0]);
1420 RFALSE(tb->blknum[0] < 0,
1421 "PAP-12185: blknum can not be %d. It must be >= 0", tb->blknum[0]);
1424 * if while adding to a node we discover that it is possible to split
1425 * it in two, and merge the left part into the left neighbor and the
1426 * right part into the right neighbor, eliminating the node
1428 if (tb->blknum[0] == 0) { /* node S[0] is empty now */
1430 RFALSE(!tb->lnum[0] || !tb->rnum[0],
1431 "PAP-12190: lnum and rnum must not be zero");
1433 * if insertion was done before 0-th position in R[0], right
1434 * delimiting key of the tb->L[0]'s and left delimiting key are
1439 reiserfs_panic(tb->tb_sb, "vs-12195",
1440 "CFR not initialized");
1441 copy_key(internal_key(tb->CFL[0], tb->lkey[0]),
1442 internal_key(tb->CFR[0], tb->rkey[0]));
1443 do_balance_mark_internal_dirty(tb, tb->CFL[0], 0);
1446 reiserfs_invalidate_buffer(tb, tbS0);
1450 balance_leaf_new_nodes(tb, ih, body, insert_key, insert_ptr, flag);
1452 balance_leaf_finish_node(tb, ih, body, flag);
1454 #ifdef CONFIG_REISERFS_CHECK
1455 if (flag == M_PASTE && tb->insert_size[0]) {
1456 print_cur_tb("12290");
1457 reiserfs_panic(tb->tb_sb,
1458 "PAP-12290", "insert_size is still not 0 (%d)",
1459 tb->insert_size[0]);
1463 /* Leaf level of the tree is balanced (end of balance_leaf) */
1467 /* Make empty node */
1468 void make_empty_node(struct buffer_info *bi)
1470 struct block_head *blkh;
1472 RFALSE(bi->bi_bh == NULL, "PAP-12295: pointer to the buffer is NULL");
1474 blkh = B_BLK_HEAD(bi->bi_bh);
1475 set_blkh_nr_item(blkh, 0);
1476 set_blkh_free_space(blkh, MAX_CHILD_SIZE(bi->bi_bh));
1479 B_N_CHILD(bi->bi_parent, bi->bi_position)->dc_size = 0; /* Endian safe if 0 */
1482 /* Get first empty buffer */
1483 struct buffer_head *get_FEB(struct tree_balance *tb)
1486 struct buffer_info bi;
1488 for (i = 0; i < MAX_FEB_SIZE; i++)
1489 if (tb->FEB[i] != NULL)
1492 if (i == MAX_FEB_SIZE)
1493 reiserfs_panic(tb->tb_sb, "vs-12300", "FEB list is empty");
1495 buffer_info_init_bh(tb, &bi, tb->FEB[i]);
1496 make_empty_node(&bi);
1497 set_buffer_uptodate(tb->FEB[i]);
1498 tb->used[i] = tb->FEB[i];
1504 /* This is now used because reiserfs_free_block has to be able to schedule. */
1505 static void store_thrown(struct tree_balance *tb, struct buffer_head *bh)
1509 if (buffer_dirty(bh))
1510 reiserfs_warning(tb->tb_sb, "reiserfs-12320",
1511 "called with dirty buffer");
1512 for (i = 0; i < ARRAY_SIZE(tb->thrown); i++)
1513 if (!tb->thrown[i]) {
1515 get_bh(bh); /* free_thrown puts this */
1518 reiserfs_warning(tb->tb_sb, "reiserfs-12321",
1519 "too many thrown buffers");
1522 static void free_thrown(struct tree_balance *tb)
1525 b_blocknr_t blocknr;
1526 for (i = 0; i < ARRAY_SIZE(tb->thrown); i++) {
1527 if (tb->thrown[i]) {
1528 blocknr = tb->thrown[i]->b_blocknr;
1529 if (buffer_dirty(tb->thrown[i]))
1530 reiserfs_warning(tb->tb_sb, "reiserfs-12322",
1531 "called with dirty buffer %d",
1533 brelse(tb->thrown[i]); /* incremented in store_thrown */
1534 reiserfs_free_block(tb->transaction_handle, NULL,
1540 void reiserfs_invalidate_buffer(struct tree_balance *tb, struct buffer_head *bh)
1542 struct block_head *blkh;
1543 blkh = B_BLK_HEAD(bh);
1544 set_blkh_level(blkh, FREE_LEVEL);
1545 set_blkh_nr_item(blkh, 0);
1547 clear_buffer_dirty(bh);
1548 store_thrown(tb, bh);
1551 /* Replace n_dest'th key in buffer dest by n_src'th key of buffer src.*/
1552 void replace_key(struct tree_balance *tb, struct buffer_head *dest, int n_dest,
1553 struct buffer_head *src, int n_src)
1556 RFALSE(dest == NULL || src == NULL,
1557 "vs-12305: source or destination buffer is 0 (src=%p, dest=%p)",
1559 RFALSE(!B_IS_KEYS_LEVEL(dest),
1560 "vs-12310: invalid level (%z) for destination buffer. dest must be leaf",
1562 RFALSE(n_dest < 0 || n_src < 0,
1563 "vs-12315: src(%d) or dest(%d) key number < 0", n_src, n_dest);
1564 RFALSE(n_dest >= B_NR_ITEMS(dest) || n_src >= B_NR_ITEMS(src),
1565 "vs-12320: src(%d(%d)) or dest(%d(%d)) key number is too big",
1566 n_src, B_NR_ITEMS(src), n_dest, B_NR_ITEMS(dest));
1568 if (B_IS_ITEMS_LEVEL(src))
1569 /* source buffer contains leaf node */
1570 memcpy(internal_key(dest, n_dest), item_head(src, n_src),
1573 memcpy(internal_key(dest, n_dest), internal_key(src, n_src),
1576 do_balance_mark_internal_dirty(tb, dest, 0);
1579 int get_left_neighbor_position(struct tree_balance *tb, int h)
1581 int Sh_position = PATH_H_POSITION(tb->tb_path, h + 1);
1583 RFALSE(PATH_H_PPARENT(tb->tb_path, h) == NULL || tb->FL[h] == NULL,
1584 "vs-12325: FL[%d](%p) or F[%d](%p) does not exist",
1585 h, tb->FL[h], h, PATH_H_PPARENT(tb->tb_path, h));
1587 if (Sh_position == 0)
1588 return B_NR_ITEMS(tb->FL[h]);
1590 return Sh_position - 1;
1593 int get_right_neighbor_position(struct tree_balance *tb, int h)
1595 int Sh_position = PATH_H_POSITION(tb->tb_path, h + 1);
1597 RFALSE(PATH_H_PPARENT(tb->tb_path, h) == NULL || tb->FR[h] == NULL,
1598 "vs-12330: F[%d](%p) or FR[%d](%p) does not exist",
1599 h, PATH_H_PPARENT(tb->tb_path, h), h, tb->FR[h]);
1601 if (Sh_position == B_NR_ITEMS(PATH_H_PPARENT(tb->tb_path, h)))
1604 return Sh_position + 1;
1607 #ifdef CONFIG_REISERFS_CHECK
1609 int is_reusable(struct super_block *s, b_blocknr_t block, int bit_value);
1610 static void check_internal_node(struct super_block *s, struct buffer_head *bh,
1613 struct disk_child *dc;
1616 RFALSE(!bh, "PAP-12336: bh == 0");
1618 if (!bh || !B_IS_IN_TREE(bh))
1621 RFALSE(!buffer_dirty(bh) &&
1622 !(buffer_journaled(bh) || buffer_journal_dirty(bh)),
1623 "PAP-12337: buffer (%b) must be dirty", bh);
1624 dc = B_N_CHILD(bh, 0);
1626 for (i = 0; i <= B_NR_ITEMS(bh); i++, dc++) {
1627 if (!is_reusable(s, dc_block_number(dc), 1)) {
1629 reiserfs_panic(s, "PAP-12338",
1630 "invalid child pointer %y in %b",
1636 static int locked_or_not_in_tree(struct tree_balance *tb,
1637 struct buffer_head *bh, char *which)
1639 if ((!buffer_journal_prepared(bh) && buffer_locked(bh)) ||
1640 !B_IS_IN_TREE(bh)) {
1641 reiserfs_warning(tb->tb_sb, "vs-12339", "%s (%b)", which, bh);
1647 static int check_before_balancing(struct tree_balance *tb)
1651 if (REISERFS_SB(tb->tb_sb)->cur_tb) {
1652 reiserfs_panic(tb->tb_sb, "vs-12335", "suspect that schedule "
1653 "occurred based on cur_tb not being null at "
1654 "this point in code. do_balance cannot properly "
1655 "handle concurrent tree accesses on a same "
1660 * double check that buffers that we will modify are unlocked.
1661 * (fix_nodes should already have prepped all of these for us).
1664 retval |= locked_or_not_in_tree(tb, tb->L[0], "L[0]");
1665 retval |= locked_or_not_in_tree(tb, tb->FL[0], "FL[0]");
1666 retval |= locked_or_not_in_tree(tb, tb->CFL[0], "CFL[0]");
1667 check_leaf(tb->L[0]);
1670 retval |= locked_or_not_in_tree(tb, tb->R[0], "R[0]");
1671 retval |= locked_or_not_in_tree(tb, tb->FR[0], "FR[0]");
1672 retval |= locked_or_not_in_tree(tb, tb->CFR[0], "CFR[0]");
1673 check_leaf(tb->R[0]);
1675 retval |= locked_or_not_in_tree(tb, PATH_PLAST_BUFFER(tb->tb_path),
1677 check_leaf(PATH_PLAST_BUFFER(tb->tb_path));
1682 static void check_after_balance_leaf(struct tree_balance *tb)
1685 if (B_FREE_SPACE(tb->L[0]) !=
1686 MAX_CHILD_SIZE(tb->L[0]) -
1688 (tb->FL[0], get_left_neighbor_position(tb, 0)))) {
1689 print_cur_tb("12221");
1690 reiserfs_panic(tb->tb_sb, "PAP-12355",
1691 "shift to left was incorrect");
1695 if (B_FREE_SPACE(tb->R[0]) !=
1696 MAX_CHILD_SIZE(tb->R[0]) -
1698 (tb->FR[0], get_right_neighbor_position(tb, 0)))) {
1699 print_cur_tb("12222");
1700 reiserfs_panic(tb->tb_sb, "PAP-12360",
1701 "shift to right was incorrect");
1704 if (PATH_H_PBUFFER(tb->tb_path, 1) &&
1705 (B_FREE_SPACE(PATH_H_PBUFFER(tb->tb_path, 0)) !=
1706 (MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)) -
1707 dc_size(B_N_CHILD(PATH_H_PBUFFER(tb->tb_path, 1),
1708 PATH_H_POSITION(tb->tb_path, 1)))))) {
1709 int left = B_FREE_SPACE(PATH_H_PBUFFER(tb->tb_path, 0));
1710 int right = (MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)) -
1711 dc_size(B_N_CHILD(PATH_H_PBUFFER(tb->tb_path, 1),
1712 PATH_H_POSITION(tb->tb_path,
1714 print_cur_tb("12223");
1715 reiserfs_warning(tb->tb_sb, "reiserfs-12363",
1716 "B_FREE_SPACE (PATH_H_PBUFFER(tb->tb_path,0)) = %d; "
1717 "MAX_CHILD_SIZE (%d) - dc_size( %y, %d ) [%d] = %d",
1719 MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)),
1720 PATH_H_PBUFFER(tb->tb_path, 1),
1721 PATH_H_POSITION(tb->tb_path, 1),
1723 (PATH_H_PBUFFER(tb->tb_path, 1),
1724 PATH_H_POSITION(tb->tb_path, 1))),
1726 reiserfs_panic(tb->tb_sb, "PAP-12365", "S is incorrect");
1730 static void check_leaf_level(struct tree_balance *tb)
1732 check_leaf(tb->L[0]);
1733 check_leaf(tb->R[0]);
1734 check_leaf(PATH_PLAST_BUFFER(tb->tb_path));
1737 static void check_internal_levels(struct tree_balance *tb)
1741 /* check all internal nodes */
1742 for (h = 1; tb->insert_size[h]; h++) {
1743 check_internal_node(tb->tb_sb, PATH_H_PBUFFER(tb->tb_path, h),
1744 "BAD BUFFER ON PATH");
1746 check_internal_node(tb->tb_sb, tb->L[h], "BAD L");
1748 check_internal_node(tb->tb_sb, tb->R[h], "BAD R");
1756 * Now we have all of the buffers that must be used in balancing of
1757 * the tree. We rely on the assumption that schedule() will not occur
1758 * while do_balance works. ( Only interrupt handlers are acceptable.)
1759 * We balance the tree according to the analysis made before this,
1760 * using buffers already obtained. For SMP support it will someday be
1761 * necessary to add ordered locking of tb.
1765 * Some interesting rules of balancing:
1766 * we delete a maximum of two nodes per level per balancing: we never
1767 * delete R, when we delete two of three nodes L, S, R then we move
1770 * we only delete L if we are deleting two nodes, if we delete only
1771 * one node we delete S
1773 * if we shift leaves then we shift as much as we can: this is a
1774 * deliberate policy of extremism in node packing which results in
1775 * higher average utilization after repeated random balance operations
1776 * at the cost of more memory copies and more balancing as a result of
1777 * small insertions to full nodes.
1779 * if we shift internal nodes we try to evenly balance the node
1780 * utilization, with consequent less balancing at the cost of lower
1783 * one could argue that the policy for directories in leaves should be
1784 * that of internal nodes, but we will wait until another day to
1785 * evaluate this.... It would be nice to someday measure and prove
1786 * these assumptions as to what is optimal....
1789 static inline void do_balance_starts(struct tree_balance *tb)
1791 /* use print_cur_tb() to see initial state of struct tree_balance */
1793 /* store_print_tb (tb); */
1795 /* do not delete, just comment it out */
1797 print_tb(flag, PATH_LAST_POSITION(tb->tb_path),
1798 tb->tb_path->pos_in_item, tb, "check");
1800 RFALSE(check_before_balancing(tb), "PAP-12340: locked buffers in TB");
1801 #ifdef CONFIG_REISERFS_CHECK
1802 REISERFS_SB(tb->tb_sb)->cur_tb = tb;
1806 static inline void do_balance_completed(struct tree_balance *tb)
1809 #ifdef CONFIG_REISERFS_CHECK
1810 check_leaf_level(tb);
1811 check_internal_levels(tb);
1812 REISERFS_SB(tb->tb_sb)->cur_tb = NULL;
1816 * reiserfs_free_block is no longer schedule safe. So, we need to
1817 * put the buffers we want freed on the thrown list during do_balance,
1818 * and then free them now
1821 REISERFS_SB(tb->tb_sb)->s_do_balance++;
1823 /* release all nodes hold to perform the balancing */
1830 * do_balance - balance the tree
1832 * @tb: tree_balance structure
1833 * @ih: item header of inserted item
1834 * @body: body of inserted item or bytes to paste
1835 * @flag: 'i' - insert, 'd' - delete, 'c' - cut, 'p' paste
1837 * Cut means delete part of an item (includes removing an entry from a
1840 * Delete means delete whole item.
1842 * Insert means add a new item into the tree.
1844 * Paste means to append to the end of an existing file or to
1845 * insert a directory entry.
1847 void do_balance(struct tree_balance *tb, struct item_head *ih,
1848 const char *body, int flag)
1850 int child_pos; /* position of a child node in its parent */
1851 int h; /* level of the tree being processed */
1854 * in our processing of one level we sometimes determine what
1855 * must be inserted into the next higher level. This insertion
1856 * consists of a key or two keys and their corresponding
1859 struct item_head insert_key[2];
1861 /* inserted node-ptrs for the next level */
1862 struct buffer_head *insert_ptr[2];
1865 tb->need_balance_dirty = 0;
1867 if (FILESYSTEM_CHANGED_TB(tb)) {
1868 reiserfs_panic(tb->tb_sb, "clm-6000", "fs generation has "
1871 /* if we have no real work to do */
1872 if (!tb->insert_size[0]) {
1873 reiserfs_warning(tb->tb_sb, "PAP-12350",
1874 "insert_size == 0, mode == %c", flag);
1879 atomic_inc(&fs_generation(tb->tb_sb));
1880 do_balance_starts(tb);
1883 * balance_leaf returns 0 except if combining L R and S into
1884 * one node. see balance_internal() for explanation of this
1887 child_pos = PATH_H_B_ITEM_ORDER(tb->tb_path, 0) +
1888 balance_leaf(tb, ih, body, flag, insert_key, insert_ptr);
1890 #ifdef CONFIG_REISERFS_CHECK
1891 check_after_balance_leaf(tb);
1894 /* Balance internal level of the tree. */
1895 for (h = 1; h < MAX_HEIGHT && tb->insert_size[h]; h++)
1896 child_pos = balance_internal(tb, h, child_pos, insert_key,
1899 do_balance_completed(tb);