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] */
305 ret_val = 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;
309 /* Calculate and check item length to insert to L[0] */
310 put_ih_item_len(ih, ih_item_len(ih) - new_item_len);
312 RFALSE(ih_item_len(ih) <= 0,
313 "PAP-12080: there is nothing to insert into L[0]: ih_item_len=%d",
316 /* Insert new item into L[0] */
317 buffer_info_init_left(tb, &bi);
318 leaf_insert_into_buf(&bi,
319 n + tb->item_pos - ret_val, ih, body,
320 tb->zeroes_num > ih_item_len(ih) ? ih_item_len(ih) : tb->zeroes_num);
322 version = ih_version(ih);
324 /* Calculate key component, item length and body to insert into S[0] */
325 set_le_ih_k_offset(ih, le_ih_k_offset(ih) +
326 (tb->lbytes << (is_indirect_le_ih(ih) ? tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT : 0)));
328 put_ih_item_len(ih, new_item_len);
329 if (tb->lbytes > tb->zeroes_num) {
330 body += (tb->lbytes - tb->zeroes_num);
333 tb->zeroes_num -= tb->lbytes;
335 RFALSE(ih_item_len(ih) <= 0,
336 "PAP-12085: there is nothing to insert into S[0]: ih_item_len=%d",
339 /* new item in whole falls into L[0] */
340 /* Shift lnum[0]-1 items to L[0] */
341 ret_val = leaf_shift_left(tb, tb->lnum[0] - 1, tb->lbytes);
342 /* Insert new item into L[0] */
343 buffer_info_init_left(tb, &bi);
344 leaf_insert_into_buf(&bi, n + tb->item_pos - ret_val, ih, body, tb->zeroes_num);
345 tb->insert_size[0] = 0;
352 * balance_leaf - reiserfs tree balancing algorithm
353 * @tb: tree balance state
354 * @ih: item header of inserted item (little endian)
355 * @body: body of inserted item or bytes to paste
356 * @flag: i - insert, d - delete, c - cut, p - paste (see do_balance)
358 * @insert_key: key to insert new nodes
359 * @insert_ptr: array of nodes to insert at the next level
361 * In our processing of one level we sometimes determine what must be
362 * inserted into the next higher level. This insertion consists of a
363 * key or two keys and their corresponding pointers.
365 static int balance_leaf(struct tree_balance *tb, struct item_head *ih,
366 const char *body, int flag,
367 struct item_head *insert_key,
368 struct buffer_head **insert_ptr)
370 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
371 struct buffer_info bi;
375 PROC_INFO_INC(tb->tb_sb, balance_at[0]);
377 /* Make balance in case insert_size[0] < 0 */
378 if (tb->insert_size[0] < 0)
379 return balance_leaf_when_delete(tb, flag);
381 tb->item_pos = PATH_LAST_POSITION(tb->tb_path),
382 tb->pos_in_item = tb->tb_path->pos_in_item,
384 if (flag == M_INSERT && !body)
385 tb->zeroes_num = ih_item_len(ih);
388 * for indirect item pos_in_item is measured in unformatted node
389 * pointers. Recalculate to bytes
392 && is_indirect_le_ih(item_head(tbS0, tb->item_pos)))
393 tb->pos_in_item *= UNFM_P_SIZE;
395 if (tb->lnum[0] > 0) {
396 /* Shift lnum[0] items from S[0] to the left neighbor L[0] */
397 if (tb->item_pos < tb->lnum[0]) {
398 /* new item or it part falls to L[0], shift it too */
399 n = B_NR_ITEMS(tb->L[0]);
402 case M_INSERT: /* insert item into L[0] */
403 balance_leaf_insert_left(tb, ih, body);
406 case M_PASTE: /* append item in L[0] */
408 if (tb->item_pos == tb->lnum[0] - 1 && tb->lbytes != -1) {
409 /* we must shift the part of the appended item */
410 if (is_direntry_le_ih(item_head(tbS0, tb->item_pos))) {
412 RFALSE(tb->zeroes_num,
413 "PAP-12090: invalid parameter in case of a directory");
415 if (tb->lbytes > tb->pos_in_item) {
416 /* new directory entry falls into L[0] */
417 struct item_head *pasted;
418 int l_pos_in_item = tb->pos_in_item;
420 /* Shift lnum[0] - 1 items in whole. Shift lbytes - 1 entries from given directory item */
421 ret_val = leaf_shift_left(tb, tb->lnum[0], tb->lbytes-1);
422 if (ret_val && !tb->item_pos) {
423 pasted = item_head(tb->L[0], B_NR_ITEMS(tb->L[0]) - 1);
424 l_pos_in_item += ih_entry_count(pasted) - (tb->lbytes -1);
427 /* Append given directory entry to directory item */
428 buffer_info_init_left(tb, &bi);
429 leaf_paste_in_buffer(&bi, n + tb->item_pos - ret_val, l_pos_in_item, tb->insert_size[0], body, tb->zeroes_num);
431 /* previous string prepared space for pasting new entry, following string pastes this entry */
433 /* when we have merge directory item, pos_in_item has been changed too */
435 /* paste new directory entry. 1 is entry number */
436 leaf_paste_entries(&bi, n + tb->item_pos - ret_val, l_pos_in_item,
437 1, (struct reiserfs_de_head *) body,
438 body + DEH_SIZE, tb->insert_size[0]);
439 tb->insert_size[0] = 0;
441 /* new directory item doesn't fall into L[0] */
442 /* Shift lnum[0]-1 items in whole. Shift lbytes directory entries from directory item number lnum[0] */
443 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
445 /* Calculate new position to append in item body */
446 tb->pos_in_item -= tb->lbytes;
449 RFALSE(tb->lbytes <= 0, "PAP-12095: there is nothing to shift to L[0]. lbytes=%d", tb->lbytes);
450 RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)),
451 "PAP-12100: incorrect position to paste: item_len=%d, pos_in_item=%d",
452 ih_item_len(item_head(tbS0, tb->item_pos)), tb->pos_in_item);
454 if (tb->lbytes >= tb->pos_in_item) {
455 /* appended item will be in L[0] in whole */
458 /* this bytes number must be appended to the last item of L[h] */
459 l_n = tb->lbytes - tb->pos_in_item;
461 /* Calculate new insert_size[0] */
462 tb->insert_size[0] -= l_n;
464 RFALSE(tb->insert_size[0] <= 0,
465 "PAP-12105: there is nothing to paste into L[0]. insert_size=%d",
467 ret_val = leaf_shift_left(tb, tb->lnum[0], ih_item_len
468 (item_head(tbS0, tb->item_pos)));
469 /* Append to body of item in L[0] */
470 buffer_info_init_left(tb, &bi);
472 (&bi, n + tb->item_pos - ret_val, ih_item_len
473 (item_head(tb->L[0], n + tb->item_pos - ret_val)),
475 tb->zeroes_num > l_n ? l_n : tb->zeroes_num);
476 /* 0-th item in S0 can be only of DIRECT type when l_n != 0 */
481 RFALSE(ih_item_len(item_head(tbS0, 0)),
482 "PAP-12106: item length must be 0");
483 RFALSE(comp_short_le_keys(leaf_key(tbS0, 0), leaf_key
484 (tb->L[0], n + tb->item_pos - ret_val)),
485 "PAP-12107: items must be of the same file");
486 if (is_indirect_le_ih(item_head(tb->L[0], n + tb->item_pos - ret_val))) {
487 temp_l = l_n << (tb->tb_sb-> s_blocksize_bits - UNFM_P_SHIFT);
489 /* update key of first item in S0 */
490 version = ih_version(item_head(tbS0, 0));
491 set_le_key_k_offset(version, leaf_key(tbS0, 0),
492 le_key_k_offset(version,leaf_key(tbS0, 0)) + temp_l);
493 /* update left delimiting key */
494 set_le_key_k_offset(version, internal_key(tb->CFL[0], tb->lkey[0]),
495 le_key_k_offset(version, internal_key(tb->CFL[0], tb->lkey[0])) + temp_l);
498 /* Calculate new body, position in item and insert_size[0] */
499 if (l_n > tb->zeroes_num) {
500 body += (l_n - tb->zeroes_num);
503 tb->zeroes_num -= l_n;
506 RFALSE(comp_short_le_keys(leaf_key(tbS0, 0), leaf_key(tb->L[0], B_NR_ITEMS(tb->L[0]) - 1))
507 || !op_is_left_mergeable(leaf_key(tbS0, 0), tbS0->b_size)
508 || !op_is_left_mergeable(internal_key(tb->CFL[0], tb->lkey[0]), tbS0->b_size),
509 "PAP-12120: item must be merge-able with left neighboring item");
510 } else { /* only part of the appended item will be in L[0] */
512 /* Calculate position in item for append in S[0] */
513 tb->pos_in_item -= tb->lbytes;
515 RFALSE(tb->pos_in_item <= 0, "PAP-12125: no place for paste. pos_in_item=%d", tb->pos_in_item);
517 /* Shift lnum[0] - 1 items in whole. Shift lbytes - 1 byte from item number lnum[0] */
518 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
521 } else { /* appended item will be in L[0] in whole */
523 struct item_head *pasted;
525 if (!tb->item_pos && op_is_left_mergeable(leaf_key(tbS0, 0), tbS0->b_size)) { /* if we paste into first item of S[0] and it is left mergable */
526 /* then increment pos_in_item by the size of the last item in L[0] */
527 pasted = item_head(tb->L[0], n - 1);
528 if (is_direntry_le_ih(pasted))
529 tb->pos_in_item += ih_entry_count(pasted);
531 tb->pos_in_item += ih_item_len(pasted);
534 /* Shift lnum[0] - 1 items in whole. Shift lbytes - 1 byte from item number lnum[0] */
535 ret_val = leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
536 /* Append to body of item in L[0] */
537 buffer_info_init_left(tb, &bi);
538 leaf_paste_in_buffer(&bi, n + tb->item_pos - ret_val,
541 body, tb->zeroes_num);
543 /* if appended item is directory, paste entry */
544 pasted = item_head(tb->L[0], n + tb->item_pos - ret_val);
545 if (is_direntry_le_ih(pasted))
546 leaf_paste_entries(&bi, n + tb->item_pos - ret_val,
548 (struct reiserfs_de_head *) body,
551 /* if appended item is indirect item, put unformatted node into un list */
552 if (is_indirect_le_ih(pasted))
553 set_ih_free_space(pasted, 0);
554 tb->insert_size[0] = 0;
558 default: /* cases d and t */
559 reiserfs_panic(tb->tb_sb, "PAP-12130",
560 "lnum > 0: unexpected mode: "
562 (flag == M_DELETE) ? "DELETE" : ((flag == M_CUT) ? "CUT" : "UNKNOWN"), flag);
565 /* new item doesn't fall into L[0] */
566 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
570 /* tb->lnum[0] > 0 */
571 /* Calculate new item position */
572 tb->item_pos -= (tb->lnum[0] - ((tb->lbytes != -1) ? 1 : 0));
574 if (tb->rnum[0] > 0) {
575 /* shift rnum[0] items from S[0] to the right neighbor R[0] */
576 n = B_NR_ITEMS(tbS0);
579 case M_INSERT: /* insert item */
580 if (n - tb->rnum[0] < tb->item_pos) { /* new item or its part falls to R[0] */
581 if (tb->item_pos == n - tb->rnum[0] + 1 && tb->rbytes != -1) { /* part of new item falls into R[0] */
582 loff_t old_key_comp, old_len, r_zeroes_number;
587 leaf_shift_right(tb, tb->rnum[0] - 1, -1);
589 version = ih_version(ih);
590 /* Remember key component and item length */
591 old_key_comp = le_ih_k_offset(ih);
592 old_len = ih_item_len(ih);
594 /* Calculate key component and item length to insert into R[0] */
595 offset = le_ih_k_offset(ih) + ((old_len - tb->rbytes) << (is_indirect_le_ih(ih) ? tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT : 0));
596 set_le_ih_k_offset(ih, offset);
597 put_ih_item_len(ih, tb->rbytes);
598 /* Insert part of the item into R[0] */
599 buffer_info_init_right(tb, &bi);
600 if ((old_len - tb->rbytes) > tb->zeroes_num) {
602 r_body = body + (old_len - tb->rbytes) - tb->zeroes_num;
605 r_zeroes_number = tb->zeroes_num - (old_len - tb->rbytes);
606 tb->zeroes_num -= r_zeroes_number;
609 leaf_insert_into_buf(&bi, 0, ih, r_body,
612 /* Replace right delimiting key by first key in R[0] */
613 replace_key(tb, tb->CFR[0], tb->rkey[0],
616 /* Calculate key component and item length to insert into S[0] */
617 set_le_ih_k_offset(ih, old_key_comp);
618 put_ih_item_len(ih, old_len - tb->rbytes);
620 tb->insert_size[0] -= tb->rbytes;
622 } else { /* whole new item falls into R[0] */
624 /* Shift rnum[0]-1 items to R[0] */
625 ret_val = leaf_shift_right(tb, tb->rnum[0] - 1, tb->rbytes);
626 /* Insert new item into R[0] */
627 buffer_info_init_right(tb, &bi);
628 leaf_insert_into_buf(&bi, tb->item_pos - n + tb->rnum[0] - 1,
629 ih, body, tb->zeroes_num);
631 if (tb->item_pos - n + tb->rnum[0] - 1 == 0) {
632 replace_key(tb, tb->CFR[0],
637 tb->zeroes_num = tb->insert_size[0] = 0;
639 } else { /* new item or part of it doesn't fall into R[0] */
641 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
645 case M_PASTE: /* append item */
647 if (n - tb->rnum[0] <= tb->item_pos) { /* pasted item or part of it falls to R[0] */
648 if (tb->item_pos == n - tb->rnum[0] && tb->rbytes != -1) { /* we must shift the part of the appended item */
649 if (is_direntry_le_ih(item_head(tbS0, tb->item_pos))) { /* we append to directory item */
652 RFALSE(tb->zeroes_num,
653 "PAP-12145: invalid parameter in case of a directory");
654 entry_count = ih_entry_count(item_head
655 (tbS0, tb->item_pos));
656 if (entry_count - tb->rbytes <
658 /* new directory entry falls into R[0] */
660 int paste_entry_position;
662 RFALSE(tb->rbytes - 1 >= entry_count || !tb-> insert_size[0],
663 "PAP-12150: no enough of entries to shift to R[0]: rbytes=%d, entry_count=%d",
664 tb->rbytes, entry_count);
665 /* Shift rnum[0]-1 items in whole. Shift rbytes-1 directory entries from directory item number rnum[0] */
666 leaf_shift_right(tb, tb->rnum[0], tb->rbytes - 1);
667 /* Paste given directory entry to directory item */
668 paste_entry_position = tb->pos_in_item - entry_count + tb->rbytes - 1;
669 buffer_info_init_right(tb, &bi);
670 leaf_paste_in_buffer(&bi, 0, paste_entry_position, tb->insert_size[0], body, tb->zeroes_num);
672 leaf_paste_entries(&bi, 0, paste_entry_position, 1,
673 (struct reiserfs_de_head *) body,
674 body + DEH_SIZE, tb->insert_size[0]);
676 if (paste_entry_position == 0) {
677 /* change delimiting keys */
678 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0],0);
681 tb->insert_size[0] = 0;
683 } else { /* new directory entry doesn't fall into R[0] */
685 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
687 } else { /* regular object */
689 int n_shift, n_rem, r_zeroes_number;
692 /* Calculate number of bytes which must be shifted from appended item */
693 if ((n_shift = tb->rbytes - tb->insert_size[0]) < 0)
696 RFALSE(tb->pos_in_item != ih_item_len
697 (item_head(tbS0, tb->item_pos)),
698 "PAP-12155: invalid position to paste. ih_item_len=%d, pos_in_item=%d",
699 tb->pos_in_item, ih_item_len
700 (item_head(tbS0, tb->item_pos)));
702 leaf_shift_right(tb, tb->rnum[0], n_shift);
703 /* Calculate number of bytes which must remain in body after appending to R[0] */
704 if ((n_rem = tb->insert_size[0] - tb->rbytes) < 0)
709 unsigned long temp_rem = n_rem;
711 version = ih_version(item_head(tb->R[0], 0));
712 if (is_indirect_le_key(version, leaf_key(tb->R[0], 0))) {
713 temp_rem = n_rem << (tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT);
715 set_le_key_k_offset(version, leaf_key(tb->R[0], 0),
716 le_key_k_offset(version, leaf_key(tb->R[0], 0)) + temp_rem);
717 set_le_key_k_offset(version, internal_key(tb->CFR[0], tb->rkey[0]),
718 le_key_k_offset(version, internal_key(tb->CFR[0], tb->rkey[0])) + temp_rem);
720 /* k_offset (leaf_key(tb->R[0],0)) += n_rem;
721 k_offset (internal_key(tb->CFR[0],tb->rkey[0])) += n_rem;*/
722 do_balance_mark_internal_dirty(tb, tb->CFR[0], 0);
724 /* Append part of body into R[0] */
725 buffer_info_init_right(tb, &bi);
726 if (n_rem > tb->zeroes_num) {
728 r_body = body + n_rem - tb->zeroes_num;
731 r_zeroes_number = tb->zeroes_num - n_rem;
732 tb->zeroes_num -= r_zeroes_number;
735 leaf_paste_in_buffer(&bi, 0, n_shift,
736 tb->insert_size[0] - n_rem,
737 r_body, r_zeroes_number);
739 if (is_indirect_le_ih(item_head(tb->R[0], 0))) {
742 "PAP-12160: paste more than one unformatted node pointer");
744 set_ih_free_space(item_head(tb->R[0], 0), 0);
746 tb->insert_size[0] = n_rem;
750 } else { /* pasted item in whole falls into R[0] */
752 struct item_head *pasted;
754 ret_val = leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
755 /* append item in R[0] */
756 if (tb->pos_in_item >= 0) {
757 buffer_info_init_right(tb, &bi);
758 leaf_paste_in_buffer(&bi, tb->item_pos - n + tb->rnum[0], tb->pos_in_item,
759 tb->insert_size[0], body, tb->zeroes_num);
762 /* paste new entry, if item is directory item */
763 pasted = item_head(tb->R[0], tb->item_pos - n + tb->rnum[0]);
764 if (is_direntry_le_ih(pasted) && tb->pos_in_item >= 0) {
765 leaf_paste_entries(&bi, tb->item_pos - n + tb->rnum[0],
767 (struct reiserfs_de_head *) body,
768 body + DEH_SIZE, tb->insert_size[0]);
769 if (!tb->pos_in_item) {
771 RFALSE(tb->item_pos - n + tb->rnum[0],
772 "PAP-12165: directory item must be first item of node when pasting is in 0th position");
774 /* update delimiting keys */
775 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0);
779 if (is_indirect_le_ih(pasted))
780 set_ih_free_space(pasted, 0);
781 tb->zeroes_num = tb->insert_size[0] = 0;
783 } else { /* new item doesn't fall into R[0] */
785 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
788 default: /* cases d and t */
789 reiserfs_panic(tb->tb_sb, "PAP-12175",
790 "rnum > 0: unexpected mode: %s(%d)",
791 (flag == M_DELETE) ? "DELETE" : ((flag == M_CUT) ? "CUT" : "UNKNOWN"), flag);
796 /* tb->rnum[0] > 0 */
797 RFALSE(tb->blknum[0] > 3,
798 "PAP-12180: blknum can not be %d. It must be <= 3", tb->blknum[0]);
799 RFALSE(tb->blknum[0] < 0,
800 "PAP-12185: blknum can not be %d. It must be >= 0", tb->blknum[0]);
803 * if while adding to a node we discover that it is possible to split
804 * it in two, and merge the left part into the left neighbor and the
805 * right part into the right neighbor, eliminating the node
807 if (tb->blknum[0] == 0) { /* node S[0] is empty now */
809 RFALSE(!tb->lnum[0] || !tb->rnum[0],
810 "PAP-12190: lnum and rnum must not be zero");
812 * if insertion was done before 0-th position in R[0], right
813 * delimiting key of the tb->L[0]'s and left delimiting key are
818 reiserfs_panic(tb->tb_sb, "vs-12195",
819 "CFR not initialized");
820 copy_key(internal_key(tb->CFL[0], tb->lkey[0]),
821 internal_key(tb->CFR[0], tb->rkey[0]));
822 do_balance_mark_internal_dirty(tb, tb->CFL[0], 0);
825 reiserfs_invalidate_buffer(tb, tbS0);
829 /* Fill new nodes that appear in place of S[0] */
830 for (i = tb->blknum[0] - 2; i >= 0; i--) {
833 "PAP-12200: snum[%d] == %d. Must be > 0", i,
836 /* here we shift from S to S_new nodes */
838 tb->S_new[i] = get_FEB(tb);
840 /* initialized block type and tree level */
841 set_blkh_level(B_BLK_HEAD(tb->S_new[i]), DISK_LEAF_NODE_LEVEL);
843 n = B_NR_ITEMS(tbS0);
846 case M_INSERT: /* insert item */
848 if (n - tb->snum[i] < tb->item_pos) { /* new item or it's part falls to first new node S_new[i] */
849 if (tb->item_pos == n - tb->snum[i] + 1 && tb->sbytes[i] != -1) { /* part of new item falls into S_new[i] */
850 int old_key_comp, old_len, r_zeroes_number;
854 /* Move snum[i]-1 items from S[0] to S_new[i] */
855 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
858 /* Remember key component and item length */
859 version = ih_version(ih);
860 old_key_comp = le_ih_k_offset(ih);
861 old_len = ih_item_len(ih);
863 /* Calculate key component and item length to insert into S_new[i] */
864 set_le_ih_k_offset(ih, le_ih_k_offset(ih) +
865 ((old_len - tb->sbytes[i]) << (is_indirect_le_ih(ih) ? tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT : 0)));
867 put_ih_item_len(ih, tb->sbytes[i]);
869 /* Insert part of the item into S_new[i] before 0-th item */
870 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
872 if ((old_len - tb->sbytes[i]) > tb->zeroes_num) {
874 r_body = body + (old_len - tb->sbytes[i]) - tb->zeroes_num;
877 r_zeroes_number = tb->zeroes_num - (old_len - tb->sbytes[i]);
878 tb->zeroes_num -= r_zeroes_number;
881 leaf_insert_into_buf(&bi, 0, ih, r_body, r_zeroes_number);
883 /* Calculate key component and item length to insert into S[i] */
884 set_le_ih_k_offset(ih, old_key_comp);
885 put_ih_item_len(ih, old_len - tb->sbytes[i]);
886 tb->insert_size[0] -= tb->sbytes[i];
887 } else { /* whole new item falls into S_new[i] */
889 /* Shift snum[0] - 1 items to S_new[i] (sbytes[i] of split item) */
890 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
891 tb->snum[i] - 1, tb->sbytes[i], tb->S_new[i]);
893 /* Insert new item into S_new[i] */
894 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
895 leaf_insert_into_buf(&bi, tb->item_pos - n + tb->snum[i] - 1,
896 ih, body, tb->zeroes_num);
898 tb->zeroes_num = tb->insert_size[0] = 0;
902 else { /* new item or it part don't falls into S_new[i] */
904 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
905 tb->snum[i], tb->sbytes[i], tb->S_new[i]);
909 case M_PASTE: /* append item */
911 if (n - tb->snum[i] <= tb->item_pos) { /* pasted item or part if it falls to S_new[i] */
912 if (tb->item_pos == n - tb->snum[i] && tb->sbytes[i] != -1) { /* we must shift part of the appended item */
913 struct item_head *aux_ih;
915 RFALSE(ih, "PAP-12210: ih must be 0");
917 aux_ih = item_head(tbS0, tb->item_pos);
918 if (is_direntry_le_ih(aux_ih)) {
919 /* we append to directory item */
923 entry_count = ih_entry_count(aux_ih);
925 if (entry_count - tb->sbytes[i] < tb->pos_in_item && tb->pos_in_item <= entry_count) {
926 /* new directory entry falls into S_new[i] */
928 RFALSE(!tb->insert_size[0], "PAP-12215: insert_size is already 0");
929 RFALSE(tb->sbytes[i] - 1 >= entry_count,
930 "PAP-12220: there are no so much entries (%d), only %d",
931 tb->sbytes[i] - 1, entry_count);
933 /* Shift snum[i]-1 items in whole. Shift sbytes[i] directory entries from directory item number snum[i] */
934 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i], tb->sbytes[i] - 1, tb->S_new[i]);
935 /* Paste given directory entry to directory item */
936 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
937 leaf_paste_in_buffer(&bi, 0, tb->pos_in_item - entry_count + tb->sbytes[i] - 1,
938 tb->insert_size[0], body, tb->zeroes_num);
939 /* paste new directory entry */
940 leaf_paste_entries(&bi, 0, tb->pos_in_item - entry_count + tb->sbytes[i] - 1, 1,
941 (struct reiserfs_de_head *) body,
942 body + DEH_SIZE, tb->insert_size[0]);
943 tb->insert_size[0] = 0;
945 } else { /* new directory entry doesn't fall into S_new[i] */
946 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i], tb->sbytes[i], tb->S_new[i]);
948 } else { /* regular object */
950 int n_shift, n_rem, r_zeroes_number;
953 RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)) || tb->insert_size[0] <= 0,
954 "PAP-12225: item too short or insert_size <= 0");
956 /* Calculate number of bytes which must be shifted from appended item */
957 n_shift = tb->sbytes[i] - tb->insert_size[0];
960 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i], n_shift, tb->S_new[i]);
962 /* Calculate number of bytes which must remain in body after append to S_new[i] */
963 n_rem = tb->insert_size[0] - tb->sbytes[i];
966 /* Append part of body into S_new[0] */
967 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
968 if (n_rem > tb->zeroes_num) {
970 r_body = body + n_rem - tb->zeroes_num;
973 r_zeroes_number = tb->zeroes_num - n_rem;
974 tb->zeroes_num -= r_zeroes_number;
977 leaf_paste_in_buffer(&bi, 0, n_shift,
978 tb->insert_size[0] - n_rem,
979 r_body, r_zeroes_number);
981 struct item_head *tmp;
983 tmp = item_head(tb->S_new[i], 0);
984 if (is_indirect_le_ih
986 set_ih_free_space(tmp, 0);
987 set_le_ih_k_offset(tmp, le_ih_k_offset(tmp) + (n_rem << (tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT)));
989 set_le_ih_k_offset(tmp, le_ih_k_offset(tmp) + n_rem);
993 tb->insert_size[0] = n_rem;
998 /* item falls wholly into S_new[i] */
1001 struct item_head *pasted;
1003 #ifdef CONFIG_REISERFS_CHECK
1004 struct item_head *ih_check = item_head(tbS0, tb->item_pos);
1006 if (!is_direntry_le_ih(ih_check)
1007 && (tb->pos_in_item != ih_item_len(ih_check)
1008 || tb->insert_size[0] <= 0))
1009 reiserfs_panic(tb->tb_sb,
1014 #endif /* CONFIG_REISERFS_CHECK */
1016 leaf_mi = leaf_move_items(LEAF_FROM_S_TO_SNEW,
1022 "PAP-12240: unexpected value returned by leaf_move_items (%d)",
1025 /* paste into item */
1026 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
1027 leaf_paste_in_buffer(&bi,
1028 tb->item_pos - n + tb->snum[i],
1031 body, tb->zeroes_num);
1033 pasted = item_head(tb->S_new[i], tb->item_pos - n + tb->snum[i]);
1034 if (is_direntry_le_ih(pasted)) {
1035 leaf_paste_entries(&bi,
1036 tb->item_pos - n + tb->snum[i],
1038 (struct reiserfs_de_head *)body,
1044 /* if we paste to indirect item update ih_free_space */
1045 if (is_indirect_le_ih(pasted))
1046 set_ih_free_space(pasted, 0);
1047 tb->zeroes_num = tb->insert_size[0] = 0;
1051 else { /* pasted item doesn't fall into S_new[i] */
1053 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
1054 tb->snum[i], tb->sbytes[i], tb->S_new[i]);
1057 default: /* cases d and t */
1058 reiserfs_panic(tb->tb_sb, "PAP-12245",
1059 "blknum > 2: unexpected mode: %s(%d)",
1060 (flag == M_DELETE) ? "DELETE" : ((flag == M_CUT) ? "CUT" : "UNKNOWN"), flag);
1063 memcpy(insert_key + i, leaf_key(tb->S_new[i], 0), KEY_SIZE);
1064 insert_ptr[i] = tb->S_new[i];
1066 RFALSE(!buffer_journaled(tb->S_new[i])
1067 || buffer_journal_dirty(tb->S_new[i])
1068 || buffer_dirty(tb->S_new[i]),
1069 "PAP-12247: S_new[%d] : (%b)",
1073 /* if the affected item was not wholly shifted then we perform all necessary operations on that part or whole of the
1074 affected item which remains in S */
1075 if (0 <= tb->item_pos && tb->item_pos < tb->s0num) { /* if we must insert or append into buffer S[0] */
1078 case M_INSERT: /* insert item into S[0] */
1079 buffer_info_init_tbS0(tb, &bi);
1080 leaf_insert_into_buf(&bi, tb->item_pos, ih,
1081 body, tb->zeroes_num);
1083 /* If we insert the first key change the delimiting key */
1084 if (tb->item_pos == 0) {
1085 if (tb->CFL[0]) /* can be 0 in reiserfsck */
1086 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0);
1090 case M_PASTE:{ /* append item in S[0] */
1091 struct item_head *pasted;
1093 pasted = item_head(tbS0, tb->item_pos);
1094 /* when directory, may be new entry already pasted */
1095 if (is_direntry_le_ih(pasted)) {
1096 if (tb->pos_in_item >= 0 && tb->pos_in_item <= ih_entry_count(pasted)) {
1098 RFALSE(!tb->insert_size[0],
1099 "PAP-12260: insert_size is 0 already");
1102 buffer_info_init_tbS0(tb, &bi);
1103 leaf_paste_in_buffer(&bi, tb->item_pos, tb->pos_in_item,
1104 tb->insert_size[0], body,
1108 leaf_paste_entries(&bi, tb->item_pos, tb->pos_in_item, 1,
1109 (struct reiserfs_de_head *)body,
1111 tb->insert_size[0]);
1112 if (!tb->item_pos && !tb->pos_in_item) {
1113 RFALSE(!tb->CFL[0] || !tb->L[0],
1114 "PAP-12270: CFL[0]/L[0] must be specified");
1116 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0);
1118 tb->insert_size[0] = 0;
1120 } else { /* regular object */
1121 if (tb->pos_in_item == ih_item_len(pasted)) {
1123 RFALSE(tb->insert_size[0] <= 0,
1124 "PAP-12275: insert size must not be %d",
1125 tb->insert_size[0]);
1126 buffer_info_init_tbS0(tb, &bi);
1127 leaf_paste_in_buffer(&bi, tb->item_pos, tb->pos_in_item,
1128 tb->insert_size[0], body, tb->zeroes_num);
1130 if (is_indirect_le_ih(pasted)) {
1135 "PAP-12280: insert_size for indirect item must be %d, not %d",
1140 set_ih_free_space(pasted, 0);
1142 tb->insert_size[0] = 0;
1144 #ifdef CONFIG_REISERFS_CHECK
1146 if (tb->insert_size[0]) {
1147 print_cur_tb("12285");
1148 reiserfs_panic(tb->tb_sb,
1153 tb->insert_size[0]);
1156 #endif /* CONFIG_REISERFS_CHECK */
1159 } /* case M_PASTE: */
1162 #ifdef CONFIG_REISERFS_CHECK
1163 if (flag == M_PASTE && tb->insert_size[0]) {
1164 print_cur_tb("12290");
1165 reiserfs_panic(tb->tb_sb,
1166 "PAP-12290", "insert_size is still not 0 (%d)",
1167 tb->insert_size[0]);
1171 /* Leaf level of the tree is balanced (end of balance_leaf) */
1175 /* Make empty node */
1176 void make_empty_node(struct buffer_info *bi)
1178 struct block_head *blkh;
1180 RFALSE(bi->bi_bh == NULL, "PAP-12295: pointer to the buffer is NULL");
1182 blkh = B_BLK_HEAD(bi->bi_bh);
1183 set_blkh_nr_item(blkh, 0);
1184 set_blkh_free_space(blkh, MAX_CHILD_SIZE(bi->bi_bh));
1187 B_N_CHILD(bi->bi_parent, bi->bi_position)->dc_size = 0; /* Endian safe if 0 */
1190 /* Get first empty buffer */
1191 struct buffer_head *get_FEB(struct tree_balance *tb)
1194 struct buffer_info bi;
1196 for (i = 0; i < MAX_FEB_SIZE; i++)
1197 if (tb->FEB[i] != NULL)
1200 if (i == MAX_FEB_SIZE)
1201 reiserfs_panic(tb->tb_sb, "vs-12300", "FEB list is empty");
1203 buffer_info_init_bh(tb, &bi, tb->FEB[i]);
1204 make_empty_node(&bi);
1205 set_buffer_uptodate(tb->FEB[i]);
1206 tb->used[i] = tb->FEB[i];
1212 /* This is now used because reiserfs_free_block has to be able to schedule. */
1213 static void store_thrown(struct tree_balance *tb, struct buffer_head *bh)
1217 if (buffer_dirty(bh))
1218 reiserfs_warning(tb->tb_sb, "reiserfs-12320",
1219 "called with dirty buffer");
1220 for (i = 0; i < ARRAY_SIZE(tb->thrown); i++)
1221 if (!tb->thrown[i]) {
1223 get_bh(bh); /* free_thrown puts this */
1226 reiserfs_warning(tb->tb_sb, "reiserfs-12321",
1227 "too many thrown buffers");
1230 static void free_thrown(struct tree_balance *tb)
1233 b_blocknr_t blocknr;
1234 for (i = 0; i < ARRAY_SIZE(tb->thrown); i++) {
1235 if (tb->thrown[i]) {
1236 blocknr = tb->thrown[i]->b_blocknr;
1237 if (buffer_dirty(tb->thrown[i]))
1238 reiserfs_warning(tb->tb_sb, "reiserfs-12322",
1239 "called with dirty buffer %d",
1241 brelse(tb->thrown[i]); /* incremented in store_thrown */
1242 reiserfs_free_block(tb->transaction_handle, NULL,
1248 void reiserfs_invalidate_buffer(struct tree_balance *tb, struct buffer_head *bh)
1250 struct block_head *blkh;
1251 blkh = B_BLK_HEAD(bh);
1252 set_blkh_level(blkh, FREE_LEVEL);
1253 set_blkh_nr_item(blkh, 0);
1255 clear_buffer_dirty(bh);
1256 store_thrown(tb, bh);
1259 /* Replace n_dest'th key in buffer dest by n_src'th key of buffer src.*/
1260 void replace_key(struct tree_balance *tb, struct buffer_head *dest, int n_dest,
1261 struct buffer_head *src, int n_src)
1264 RFALSE(dest == NULL || src == NULL,
1265 "vs-12305: source or destination buffer is 0 (src=%p, dest=%p)",
1267 RFALSE(!B_IS_KEYS_LEVEL(dest),
1268 "vs-12310: invalid level (%z) for destination buffer. dest must be leaf",
1270 RFALSE(n_dest < 0 || n_src < 0,
1271 "vs-12315: src(%d) or dest(%d) key number < 0", n_src, n_dest);
1272 RFALSE(n_dest >= B_NR_ITEMS(dest) || n_src >= B_NR_ITEMS(src),
1273 "vs-12320: src(%d(%d)) or dest(%d(%d)) key number is too big",
1274 n_src, B_NR_ITEMS(src), n_dest, B_NR_ITEMS(dest));
1276 if (B_IS_ITEMS_LEVEL(src))
1277 /* source buffer contains leaf node */
1278 memcpy(internal_key(dest, n_dest), item_head(src, n_src),
1281 memcpy(internal_key(dest, n_dest), internal_key(src, n_src),
1284 do_balance_mark_internal_dirty(tb, dest, 0);
1287 int get_left_neighbor_position(struct tree_balance *tb, int h)
1289 int Sh_position = PATH_H_POSITION(tb->tb_path, h + 1);
1291 RFALSE(PATH_H_PPARENT(tb->tb_path, h) == NULL || tb->FL[h] == NULL,
1292 "vs-12325: FL[%d](%p) or F[%d](%p) does not exist",
1293 h, tb->FL[h], h, PATH_H_PPARENT(tb->tb_path, h));
1295 if (Sh_position == 0)
1296 return B_NR_ITEMS(tb->FL[h]);
1298 return Sh_position - 1;
1301 int get_right_neighbor_position(struct tree_balance *tb, int h)
1303 int Sh_position = PATH_H_POSITION(tb->tb_path, h + 1);
1305 RFALSE(PATH_H_PPARENT(tb->tb_path, h) == NULL || tb->FR[h] == NULL,
1306 "vs-12330: F[%d](%p) or FR[%d](%p) does not exist",
1307 h, PATH_H_PPARENT(tb->tb_path, h), h, tb->FR[h]);
1309 if (Sh_position == B_NR_ITEMS(PATH_H_PPARENT(tb->tb_path, h)))
1312 return Sh_position + 1;
1315 #ifdef CONFIG_REISERFS_CHECK
1317 int is_reusable(struct super_block *s, b_blocknr_t block, int bit_value);
1318 static void check_internal_node(struct super_block *s, struct buffer_head *bh,
1321 struct disk_child *dc;
1324 RFALSE(!bh, "PAP-12336: bh == 0");
1326 if (!bh || !B_IS_IN_TREE(bh))
1329 RFALSE(!buffer_dirty(bh) &&
1330 !(buffer_journaled(bh) || buffer_journal_dirty(bh)),
1331 "PAP-12337: buffer (%b) must be dirty", bh);
1332 dc = B_N_CHILD(bh, 0);
1334 for (i = 0; i <= B_NR_ITEMS(bh); i++, dc++) {
1335 if (!is_reusable(s, dc_block_number(dc), 1)) {
1337 reiserfs_panic(s, "PAP-12338",
1338 "invalid child pointer %y in %b",
1344 static int locked_or_not_in_tree(struct tree_balance *tb,
1345 struct buffer_head *bh, char *which)
1347 if ((!buffer_journal_prepared(bh) && buffer_locked(bh)) ||
1348 !B_IS_IN_TREE(bh)) {
1349 reiserfs_warning(tb->tb_sb, "vs-12339", "%s (%b)", which, bh);
1355 static int check_before_balancing(struct tree_balance *tb)
1359 if (REISERFS_SB(tb->tb_sb)->cur_tb) {
1360 reiserfs_panic(tb->tb_sb, "vs-12335", "suspect that schedule "
1361 "occurred based on cur_tb not being null at "
1362 "this point in code. do_balance cannot properly "
1363 "handle concurrent tree accesses on a same "
1368 * double check that buffers that we will modify are unlocked.
1369 * (fix_nodes should already have prepped all of these for us).
1372 retval |= locked_or_not_in_tree(tb, tb->L[0], "L[0]");
1373 retval |= locked_or_not_in_tree(tb, tb->FL[0], "FL[0]");
1374 retval |= locked_or_not_in_tree(tb, tb->CFL[0], "CFL[0]");
1375 check_leaf(tb->L[0]);
1378 retval |= locked_or_not_in_tree(tb, tb->R[0], "R[0]");
1379 retval |= locked_or_not_in_tree(tb, tb->FR[0], "FR[0]");
1380 retval |= locked_or_not_in_tree(tb, tb->CFR[0], "CFR[0]");
1381 check_leaf(tb->R[0]);
1383 retval |= locked_or_not_in_tree(tb, PATH_PLAST_BUFFER(tb->tb_path),
1385 check_leaf(PATH_PLAST_BUFFER(tb->tb_path));
1390 static void check_after_balance_leaf(struct tree_balance *tb)
1393 if (B_FREE_SPACE(tb->L[0]) !=
1394 MAX_CHILD_SIZE(tb->L[0]) -
1396 (tb->FL[0], get_left_neighbor_position(tb, 0)))) {
1397 print_cur_tb("12221");
1398 reiserfs_panic(tb->tb_sb, "PAP-12355",
1399 "shift to left was incorrect");
1403 if (B_FREE_SPACE(tb->R[0]) !=
1404 MAX_CHILD_SIZE(tb->R[0]) -
1406 (tb->FR[0], get_right_neighbor_position(tb, 0)))) {
1407 print_cur_tb("12222");
1408 reiserfs_panic(tb->tb_sb, "PAP-12360",
1409 "shift to right was incorrect");
1412 if (PATH_H_PBUFFER(tb->tb_path, 1) &&
1413 (B_FREE_SPACE(PATH_H_PBUFFER(tb->tb_path, 0)) !=
1414 (MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)) -
1415 dc_size(B_N_CHILD(PATH_H_PBUFFER(tb->tb_path, 1),
1416 PATH_H_POSITION(tb->tb_path, 1)))))) {
1417 int left = B_FREE_SPACE(PATH_H_PBUFFER(tb->tb_path, 0));
1418 int right = (MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)) -
1419 dc_size(B_N_CHILD(PATH_H_PBUFFER(tb->tb_path, 1),
1420 PATH_H_POSITION(tb->tb_path,
1422 print_cur_tb("12223");
1423 reiserfs_warning(tb->tb_sb, "reiserfs-12363",
1424 "B_FREE_SPACE (PATH_H_PBUFFER(tb->tb_path,0)) = %d; "
1425 "MAX_CHILD_SIZE (%d) - dc_size( %y, %d ) [%d] = %d",
1427 MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)),
1428 PATH_H_PBUFFER(tb->tb_path, 1),
1429 PATH_H_POSITION(tb->tb_path, 1),
1431 (PATH_H_PBUFFER(tb->tb_path, 1),
1432 PATH_H_POSITION(tb->tb_path, 1))),
1434 reiserfs_panic(tb->tb_sb, "PAP-12365", "S is incorrect");
1438 static void check_leaf_level(struct tree_balance *tb)
1440 check_leaf(tb->L[0]);
1441 check_leaf(tb->R[0]);
1442 check_leaf(PATH_PLAST_BUFFER(tb->tb_path));
1445 static void check_internal_levels(struct tree_balance *tb)
1449 /* check all internal nodes */
1450 for (h = 1; tb->insert_size[h]; h++) {
1451 check_internal_node(tb->tb_sb, PATH_H_PBUFFER(tb->tb_path, h),
1452 "BAD BUFFER ON PATH");
1454 check_internal_node(tb->tb_sb, tb->L[h], "BAD L");
1456 check_internal_node(tb->tb_sb, tb->R[h], "BAD R");
1464 * Now we have all of the buffers that must be used in balancing of
1465 * the tree. We rely on the assumption that schedule() will not occur
1466 * while do_balance works. ( Only interrupt handlers are acceptable.)
1467 * We balance the tree according to the analysis made before this,
1468 * using buffers already obtained. For SMP support it will someday be
1469 * necessary to add ordered locking of tb.
1473 * Some interesting rules of balancing:
1474 * we delete a maximum of two nodes per level per balancing: we never
1475 * delete R, when we delete two of three nodes L, S, R then we move
1478 * we only delete L if we are deleting two nodes, if we delete only
1479 * one node we delete S
1481 * if we shift leaves then we shift as much as we can: this is a
1482 * deliberate policy of extremism in node packing which results in
1483 * higher average utilization after repeated random balance operations
1484 * at the cost of more memory copies and more balancing as a result of
1485 * small insertions to full nodes.
1487 * if we shift internal nodes we try to evenly balance the node
1488 * utilization, with consequent less balancing at the cost of lower
1491 * one could argue that the policy for directories in leaves should be
1492 * that of internal nodes, but we will wait until another day to
1493 * evaluate this.... It would be nice to someday measure and prove
1494 * these assumptions as to what is optimal....
1497 static inline void do_balance_starts(struct tree_balance *tb)
1499 /* use print_cur_tb() to see initial state of struct tree_balance */
1501 /* store_print_tb (tb); */
1503 /* do not delete, just comment it out */
1505 print_tb(flag, PATH_LAST_POSITION(tb->tb_path),
1506 tb->tb_path->pos_in_item, tb, "check");
1508 RFALSE(check_before_balancing(tb), "PAP-12340: locked buffers in TB");
1509 #ifdef CONFIG_REISERFS_CHECK
1510 REISERFS_SB(tb->tb_sb)->cur_tb = tb;
1514 static inline void do_balance_completed(struct tree_balance *tb)
1517 #ifdef CONFIG_REISERFS_CHECK
1518 check_leaf_level(tb);
1519 check_internal_levels(tb);
1520 REISERFS_SB(tb->tb_sb)->cur_tb = NULL;
1524 * reiserfs_free_block is no longer schedule safe. So, we need to
1525 * put the buffers we want freed on the thrown list during do_balance,
1526 * and then free them now
1529 REISERFS_SB(tb->tb_sb)->s_do_balance++;
1531 /* release all nodes hold to perform the balancing */
1538 * do_balance - balance the tree
1540 * @tb: tree_balance structure
1541 * @ih: item header of inserted item
1542 * @body: body of inserted item or bytes to paste
1543 * @flag: 'i' - insert, 'd' - delete, 'c' - cut, 'p' paste
1545 * Cut means delete part of an item (includes removing an entry from a
1548 * Delete means delete whole item.
1550 * Insert means add a new item into the tree.
1552 * Paste means to append to the end of an existing file or to
1553 * insert a directory entry.
1555 void do_balance(struct tree_balance *tb, struct item_head *ih,
1556 const char *body, int flag)
1558 int child_pos; /* position of a child node in its parent */
1559 int h; /* level of the tree being processed */
1562 * in our processing of one level we sometimes determine what
1563 * must be inserted into the next higher level. This insertion
1564 * consists of a key or two keys and their corresponding
1567 struct item_head insert_key[2];
1569 /* inserted node-ptrs for the next level */
1570 struct buffer_head *insert_ptr[2];
1573 tb->need_balance_dirty = 0;
1575 if (FILESYSTEM_CHANGED_TB(tb)) {
1576 reiserfs_panic(tb->tb_sb, "clm-6000", "fs generation has "
1579 /* if we have no real work to do */
1580 if (!tb->insert_size[0]) {
1581 reiserfs_warning(tb->tb_sb, "PAP-12350",
1582 "insert_size == 0, mode == %c", flag);
1587 atomic_inc(&fs_generation(tb->tb_sb));
1588 do_balance_starts(tb);
1591 * balance_leaf returns 0 except if combining L R and S into
1592 * one node. see balance_internal() for explanation of this
1595 child_pos = PATH_H_B_ITEM_ORDER(tb->tb_path, 0) +
1596 balance_leaf(tb, ih, body, flag, insert_key, insert_ptr);
1598 #ifdef CONFIG_REISERFS_CHECK
1599 check_after_balance_leaf(tb);
1602 /* Balance internal level of the tree. */
1603 for (h = 1; h < MAX_HEIGHT && tb->insert_size[h]; h++)
1605 balance_internal(tb, h, child_pos, insert_key, insert_ptr);
1607 do_balance_completed(tb);