2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
6 * Now we have all buffers that must be used in balancing of the tree
7 * Further calculations can not cause schedule(), and thus the buffer
8 * tree will be stable until the balancing will be finished
9 * balance the tree according to the analysis made before,
10 * and using buffers obtained after all above.
13 #include <asm/uaccess.h>
14 #include <linux/time.h>
16 #include <linux/buffer_head.h>
17 #include <linux/kernel.h>
19 static inline void buffer_info_init_left(struct tree_balance *tb,
20 struct buffer_info *bi)
24 bi->bi_parent = tb->FL[0];
25 bi->bi_position = get_left_neighbor_position(tb, 0);
28 static inline void buffer_info_init_right(struct tree_balance *tb,
29 struct buffer_info *bi)
33 bi->bi_parent = tb->FR[0];
34 bi->bi_position = get_right_neighbor_position(tb, 0);
37 static inline void buffer_info_init_tbS0(struct tree_balance *tb,
38 struct buffer_info *bi)
41 bi->bi_bh = PATH_PLAST_BUFFER(tb->tb_path);
42 bi->bi_parent = PATH_H_PPARENT(tb->tb_path, 0);
43 bi->bi_position = PATH_H_POSITION(tb->tb_path, 1);
46 static inline void buffer_info_init_bh(struct tree_balance *tb,
47 struct buffer_info *bi,
48 struct buffer_head *bh)
56 inline void do_balance_mark_leaf_dirty(struct tree_balance *tb,
57 struct buffer_head *bh, int flag)
59 journal_mark_dirty(tb->transaction_handle, bh);
62 #define do_balance_mark_internal_dirty do_balance_mark_leaf_dirty
63 #define do_balance_mark_sb_dirty do_balance_mark_leaf_dirty
67 * if deleting something ( tb->insert_size[0] < 0 )
68 * return(balance_leaf_when_delete()); (flag d handled here)
70 * if lnum is larger than 0 we put items into the left node
71 * if rnum is larger than 0 we put items into the right node
72 * if snum1 is larger than 0 we put items into the new node s1
73 * if snum2 is larger than 0 we put items into the new node s2
74 * Note that all *num* count new items being created.
76 * It would be easier to read balance_leaf() if each of these summary
77 * lines was a separate procedure rather than being inlined. I think
78 * that there are many passages here and in balance_leaf_when_delete() in
79 * which two calls to one procedure can replace two passages, and it
80 * might save cache space and improve software maintenance costs to do so.
82 * Vladimir made the perceptive comment that we should offload most of
83 * the decision making in this function into fix_nodes/check_balance, and
84 * then create some sort of structure in tb that says what actions should
85 * be performed by do_balance.
91 * Balance leaf node in case of delete or cut: insert_size[0] < 0
93 * lnum, rnum can have values >= -1
94 * -1 means that the neighbor must be joined with S
95 * 0 means that nothing should be done with the neighbor
96 * >0 means to shift entirely or partly the specified number of items
99 static int balance_leaf_when_delete(struct tree_balance *tb, int flag)
101 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
102 int item_pos = PATH_LAST_POSITION(tb->tb_path);
103 int pos_in_item = tb->tb_path->pos_in_item;
104 struct buffer_info bi;
106 struct item_head *ih;
108 RFALSE(tb->FR[0] && B_LEVEL(tb->FR[0]) != DISK_LEAF_NODE_LEVEL + 1,
109 "vs- 12000: level: wrong FR %z", tb->FR[0]);
110 RFALSE(tb->blknum[0] > 1,
111 "PAP-12005: tb->blknum == %d, can not be > 1", tb->blknum[0]);
112 RFALSE(!tb->blknum[0] && !PATH_H_PPARENT(tb->tb_path, 0),
113 "PAP-12010: tree can not be empty");
115 ih = item_head(tbS0, item_pos);
116 buffer_info_init_tbS0(tb, &bi);
118 /* Delete or truncate the item */
121 case M_DELETE: /* delete item in S[0] */
123 RFALSE(ih_item_len(ih) + IH_SIZE != -tb->insert_size[0],
124 "vs-12013: mode Delete, insert size %d, ih to be deleted %h",
125 -tb->insert_size[0], ih);
127 leaf_delete_items(&bi, 0, item_pos, 1, -1);
129 if (!item_pos && tb->CFL[0]) {
130 if (B_NR_ITEMS(tbS0)) {
131 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0,
134 if (!PATH_H_POSITION(tb->tb_path, 1))
135 replace_key(tb, tb->CFL[0], tb->lkey[0],
136 PATH_H_PPARENT(tb->tb_path,
141 RFALSE(!item_pos && !tb->CFL[0],
142 "PAP-12020: tb->CFL[0]==%p, tb->L[0]==%p", tb->CFL[0],
147 case M_CUT:{ /* cut item in S[0] */
148 if (is_direntry_le_ih(ih)) {
151 * UFS unlink semantics are such that you
152 * can only delete one directory entry at
157 * when we cut a directory tb->insert_size[0]
158 * means number of entries to be cut (always 1)
160 tb->insert_size[0] = -1;
161 leaf_cut_from_buffer(&bi, item_pos, pos_in_item,
162 -tb->insert_size[0]);
164 RFALSE(!item_pos && !pos_in_item && !tb->CFL[0],
165 "PAP-12030: can not change delimiting key. CFL[0]=%p",
168 if (!item_pos && !pos_in_item && tb->CFL[0]) {
169 replace_key(tb, tb->CFL[0], tb->lkey[0],
173 leaf_cut_from_buffer(&bi, item_pos, pos_in_item,
174 -tb->insert_size[0]);
176 RFALSE(!ih_item_len(ih),
177 "PAP-12035: cut must leave non-zero dynamic length of item");
183 print_cur_tb("12040");
184 reiserfs_panic(tb->tb_sb, "PAP-12040",
185 "unexpected mode: %s(%d)",
187 M_PASTE) ? "PASTE" : ((flag ==
188 M_INSERT) ? "INSERT" :
193 * the rule is that no shifting occurs unless by shifting
194 * a node can be freed
196 n = B_NR_ITEMS(tbS0);
197 /* L[0] takes part in balancing */
199 /* L[0] must be joined with S[0] */
200 if (tb->lnum[0] == -1) {
201 /* R[0] must be also joined with S[0] */
202 if (tb->rnum[0] == -1) {
203 if (tb->FR[0] == PATH_H_PPARENT(tb->tb_path, 0)) {
205 * all contents of all the 3 buffers
208 if (PATH_H_POSITION(tb->tb_path, 1) == 0
209 && 1 < B_NR_ITEMS(tb->FR[0]))
210 replace_key(tb, tb->CFL[0],
214 leaf_move_items(LEAF_FROM_S_TO_L, tb, n,
216 leaf_move_items(LEAF_FROM_R_TO_L, tb,
217 B_NR_ITEMS(tb->R[0]),
220 reiserfs_invalidate_buffer(tb, tbS0);
221 reiserfs_invalidate_buffer(tb,
227 * all contents of all the 3 buffers will
230 leaf_move_items(LEAF_FROM_S_TO_R, tb, n, -1,
232 leaf_move_items(LEAF_FROM_L_TO_R, tb,
233 B_NR_ITEMS(tb->L[0]), -1, NULL);
235 /* right_delimiting_key is correct in R[0] */
236 replace_key(tb, tb->CFR[0], tb->rkey[0],
239 reiserfs_invalidate_buffer(tb, tbS0);
240 reiserfs_invalidate_buffer(tb, tb->L[0]);
245 RFALSE(tb->rnum[0] != 0,
246 "PAP-12045: rnum must be 0 (%d)", tb->rnum[0]);
247 /* all contents of L[0] and S[0] will be in L[0] */
248 leaf_shift_left(tb, n, -1);
250 reiserfs_invalidate_buffer(tb, tbS0);
256 * a part of contents of S[0] will be in L[0] and the
257 * rest part of S[0] will be in R[0]
260 RFALSE((tb->lnum[0] + tb->rnum[0] < n) ||
261 (tb->lnum[0] + tb->rnum[0] > n + 1),
262 "PAP-12050: rnum(%d) and lnum(%d) and item number(%d) in S[0] are not consistent",
263 tb->rnum[0], tb->lnum[0], n);
264 RFALSE((tb->lnum[0] + tb->rnum[0] == n) &&
265 (tb->lbytes != -1 || tb->rbytes != -1),
266 "PAP-12055: bad rbytes (%d)/lbytes (%d) parameters when items are not split",
267 tb->rbytes, tb->lbytes);
268 RFALSE((tb->lnum[0] + tb->rnum[0] == n + 1) &&
269 (tb->lbytes < 1 || tb->rbytes != -1),
270 "PAP-12060: bad rbytes (%d)/lbytes (%d) parameters when items are split",
271 tb->rbytes, tb->lbytes);
273 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
274 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
276 reiserfs_invalidate_buffer(tb, tbS0);
281 if (tb->rnum[0] == -1) {
282 /* all contents of R[0] and S[0] will be in R[0] */
283 leaf_shift_right(tb, n, -1);
284 reiserfs_invalidate_buffer(tb, tbS0);
289 "PAP-12065: bad rnum parameter must be 0 (%d)", tb->rnum[0]);
293 static void balance_leaf_insert_left(struct tree_balance *tb,
294 struct item_head *ih, const char *body)
297 struct buffer_info bi;
298 int n = B_NR_ITEMS(tb->L[0]);
300 if (tb->item_pos == tb->lnum[0] - 1 && tb->lbytes != -1) {
301 /* part of new item falls into L[0] */
302 int new_item_len, shift;
305 ret = leaf_shift_left(tb, tb->lnum[0] - 1, -1);
307 /* Calculate item length to insert to S[0] */
308 new_item_len = ih_item_len(ih) - tb->lbytes;
310 /* Calculate and check item length to insert to L[0] */
311 put_ih_item_len(ih, ih_item_len(ih) - new_item_len);
313 RFALSE(ih_item_len(ih) <= 0,
314 "PAP-12080: there is nothing to insert into L[0]: "
315 "ih_item_len=%d", ih_item_len(ih));
317 /* Insert new item into L[0] */
318 buffer_info_init_left(tb, &bi);
319 leaf_insert_into_buf(&bi, n + tb->item_pos - ret, ih, body,
320 min_t(int, tb->zeroes_num, ih_item_len(ih)));
322 version = ih_version(ih);
325 * Calculate key component, item length and body to
329 if (is_indirect_le_ih(ih))
330 shift = tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT;
332 add_le_ih_k_offset(ih, tb->lbytes << shift);
334 put_ih_item_len(ih, new_item_len);
335 if (tb->lbytes > tb->zeroes_num) {
336 body += (tb->lbytes - tb->zeroes_num);
339 tb->zeroes_num -= tb->lbytes;
341 RFALSE(ih_item_len(ih) <= 0,
342 "PAP-12085: there is nothing to insert into S[0]: "
343 "ih_item_len=%d", ih_item_len(ih));
345 /* new item in whole falls into L[0] */
346 /* Shift lnum[0]-1 items to L[0] */
347 ret = leaf_shift_left(tb, tb->lnum[0] - 1, tb->lbytes);
349 /* Insert new item into L[0] */
350 buffer_info_init_left(tb, &bi);
351 leaf_insert_into_buf(&bi, n + tb->item_pos - ret, ih, body,
353 tb->insert_size[0] = 0;
358 static void balance_leaf_paste_left(struct tree_balance *tb,
359 struct item_head *ih, const char *body)
361 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
363 struct buffer_info bi;
364 int n = B_NR_ITEMS(tb->L[0]);
366 if (tb->item_pos == tb->lnum[0] - 1 && tb->lbytes != -1) {
367 /* we must shift the part of the appended item */
368 if (is_direntry_le_ih(item_head(tbS0, tb->item_pos))) {
370 RFALSE(tb->zeroes_num,
371 "PAP-12090: invalid parameter in case of a directory");
373 if (tb->lbytes > tb->pos_in_item) {
374 /* new directory entry falls into L[0] */
375 struct item_head *pasted;
376 int l_pos_in_item = tb->pos_in_item;
378 /* Shift lnum[0] - 1 items in whole. Shift lbytes - 1 entries from given directory item */
379 ret_val = leaf_shift_left(tb, tb->lnum[0], tb->lbytes-1);
380 if (ret_val && !tb->item_pos) {
381 pasted = item_head(tb->L[0], B_NR_ITEMS(tb->L[0]) - 1);
382 l_pos_in_item += ih_entry_count(pasted) - (tb->lbytes -1);
385 /* Append given directory entry to directory item */
386 buffer_info_init_left(tb, &bi);
387 leaf_paste_in_buffer(&bi, n + tb->item_pos - ret_val, l_pos_in_item, tb->insert_size[0], body, tb->zeroes_num);
389 /* previous string prepared space for pasting new entry, following string pastes this entry */
391 /* when we have merge directory item, pos_in_item has been changed too */
393 /* paste new directory entry. 1 is entry number */
394 leaf_paste_entries(&bi, n + tb->item_pos - ret_val, l_pos_in_item,
395 1, (struct reiserfs_de_head *) body,
396 body + DEH_SIZE, tb->insert_size[0]);
397 tb->insert_size[0] = 0;
399 /* new directory item doesn't fall into L[0] */
400 /* Shift lnum[0]-1 items in whole. Shift lbytes directory entries from directory item number lnum[0] */
401 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
403 /* Calculate new position to append in item body */
404 tb->pos_in_item -= tb->lbytes;
407 RFALSE(tb->lbytes <= 0, "PAP-12095: there is nothing to shift to L[0]. lbytes=%d", tb->lbytes);
408 RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)),
409 "PAP-12100: incorrect position to paste: item_len=%d, pos_in_item=%d",
410 ih_item_len(item_head(tbS0, tb->item_pos)), tb->pos_in_item);
412 if (tb->lbytes >= tb->pos_in_item) {
413 /* appended item will be in L[0] in whole */
416 /* this bytes number must be appended to the last item of L[h] */
417 l_n = tb->lbytes - tb->pos_in_item;
419 /* Calculate new insert_size[0] */
420 tb->insert_size[0] -= l_n;
422 RFALSE(tb->insert_size[0] <= 0,
423 "PAP-12105: there is nothing to paste into L[0]. insert_size=%d",
425 ret_val = leaf_shift_left(tb, tb->lnum[0], ih_item_len
426 (item_head(tbS0, tb->item_pos)));
427 /* Append to body of item in L[0] */
428 buffer_info_init_left(tb, &bi);
430 (&bi, n + tb->item_pos - ret_val, ih_item_len
431 (item_head(tb->L[0], n + tb->item_pos - ret_val)),
433 tb->zeroes_num > l_n ? l_n : tb->zeroes_num);
434 /* 0-th item in S0 can be only of DIRECT type when l_n != 0 */
439 RFALSE(ih_item_len(item_head(tbS0, 0)),
440 "PAP-12106: item length must be 0");
441 RFALSE(comp_short_le_keys(leaf_key(tbS0, 0), leaf_key
442 (tb->L[0], n + tb->item_pos - ret_val)),
443 "PAP-12107: items must be of the same file");
444 if (is_indirect_le_ih(item_head(tb->L[0], n + tb->item_pos - ret_val))) {
445 temp_l = l_n << (tb->tb_sb-> s_blocksize_bits - UNFM_P_SHIFT);
447 /* update key of first item in S0 */
448 version = ih_version(item_head(tbS0, 0));
449 set_le_key_k_offset(version, leaf_key(tbS0, 0),
450 le_key_k_offset(version,leaf_key(tbS0, 0)) + temp_l);
451 /* update left delimiting key */
452 set_le_key_k_offset(version, internal_key(tb->CFL[0], tb->lkey[0]),
453 le_key_k_offset(version, internal_key(tb->CFL[0], tb->lkey[0])) + temp_l);
456 /* Calculate new body, position in item and insert_size[0] */
457 if (l_n > tb->zeroes_num) {
458 body += (l_n - tb->zeroes_num);
461 tb->zeroes_num -= l_n;
464 RFALSE(comp_short_le_keys(leaf_key(tbS0, 0), leaf_key(tb->L[0], B_NR_ITEMS(tb->L[0]) - 1))
465 || !op_is_left_mergeable(leaf_key(tbS0, 0), tbS0->b_size)
466 || !op_is_left_mergeable(internal_key(tb->CFL[0], tb->lkey[0]), tbS0->b_size),
467 "PAP-12120: item must be merge-able with left neighboring item");
468 } else { /* only part of the appended item will be in L[0] */
470 /* Calculate position in item for append in S[0] */
471 tb->pos_in_item -= tb->lbytes;
473 RFALSE(tb->pos_in_item <= 0, "PAP-12125: no place for paste. pos_in_item=%d", tb->pos_in_item);
475 /* Shift lnum[0] - 1 items in whole. Shift lbytes - 1 byte from item number lnum[0] */
476 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
479 } else { /* appended item will be in L[0] in whole */
481 struct item_head *pasted;
483 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 */
484 /* then increment pos_in_item by the size of the last item in L[0] */
485 pasted = item_head(tb->L[0], n - 1);
486 if (is_direntry_le_ih(pasted))
487 tb->pos_in_item += ih_entry_count(pasted);
489 tb->pos_in_item += ih_item_len(pasted);
492 /* Shift lnum[0] - 1 items in whole. Shift lbytes - 1 byte from item number lnum[0] */
493 ret_val = leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
494 /* Append to body of item in L[0] */
495 buffer_info_init_left(tb, &bi);
496 leaf_paste_in_buffer(&bi, n + tb->item_pos - ret_val,
499 body, tb->zeroes_num);
501 /* if appended item is directory, paste entry */
502 pasted = item_head(tb->L[0], n + tb->item_pos - ret_val);
503 if (is_direntry_le_ih(pasted))
504 leaf_paste_entries(&bi, n + tb->item_pos - ret_val,
506 (struct reiserfs_de_head *) body,
509 /* if appended item is indirect item, put unformatted node into un list */
510 if (is_indirect_le_ih(pasted))
511 set_ih_free_space(pasted, 0);
512 tb->insert_size[0] = 0;
518 /* Shift lnum[0] items from S[0] to the left neighbor L[0] */
519 static void balance_leaf_left(struct tree_balance *tb, struct item_head *ih,
520 const char *body, int flag)
522 if (tb->lnum[0] <= 0)
525 /* new item or it part falls to L[0], shift it too */
526 if (tb->item_pos < tb->lnum[0]) {
527 BUG_ON(flag != M_INSERT && flag != M_PASTE);
529 if (flag == M_INSERT)
530 balance_leaf_insert_left(tb, ih, body);
532 balance_leaf_paste_left(tb, ih, body);
534 /* new item doesn't fall into L[0] */
535 leaf_shift_left(tb, tb->lnum[0], tb->lbytes);
539 static void balance_leaf_insert_right(struct tree_balance *tb,
540 struct item_head *ih, const char *body)
543 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
544 int n = B_NR_ITEMS(tbS0);
545 struct buffer_info bi;
547 if (n - tb->rnum[0] < tb->item_pos) { /* new item or its part falls to R[0] */
548 if (tb->item_pos == n - tb->rnum[0] + 1 && tb->rbytes != -1) { /* part of new item falls into R[0] */
549 loff_t old_key_comp, old_len, r_zeroes_number;
554 leaf_shift_right(tb, tb->rnum[0] - 1, -1);
556 version = ih_version(ih);
557 /* Remember key component and item length */
558 old_key_comp = le_ih_k_offset(ih);
559 old_len = ih_item_len(ih);
561 /* Calculate key component and item length to insert into R[0] */
562 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));
563 set_le_ih_k_offset(ih, offset);
564 put_ih_item_len(ih, tb->rbytes);
565 /* Insert part of the item into R[0] */
566 buffer_info_init_right(tb, &bi);
567 if ((old_len - tb->rbytes) > tb->zeroes_num) {
569 r_body = body + (old_len - tb->rbytes) - tb->zeroes_num;
572 r_zeroes_number = tb->zeroes_num - (old_len - tb->rbytes);
573 tb->zeroes_num -= r_zeroes_number;
576 leaf_insert_into_buf(&bi, 0, ih, r_body,
579 /* Replace right delimiting key by first key in R[0] */
580 replace_key(tb, tb->CFR[0], tb->rkey[0],
583 /* Calculate key component and item length to insert into S[0] */
584 set_le_ih_k_offset(ih, old_key_comp);
585 put_ih_item_len(ih, old_len - tb->rbytes);
587 tb->insert_size[0] -= tb->rbytes;
589 } else { /* whole new item falls into R[0] */
591 /* Shift rnum[0]-1 items to R[0] */
592 ret_val = leaf_shift_right(tb, tb->rnum[0] - 1, tb->rbytes);
593 /* Insert new item into R[0] */
594 buffer_info_init_right(tb, &bi);
595 leaf_insert_into_buf(&bi, tb->item_pos - n + tb->rnum[0] - 1,
596 ih, body, tb->zeroes_num);
598 if (tb->item_pos - n + tb->rnum[0] - 1 == 0) {
599 replace_key(tb, tb->CFR[0],
604 tb->zeroes_num = tb->insert_size[0] = 0;
606 } else { /* new item or part of it doesn't fall into R[0] */
608 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
613 static void balance_leaf_paste_right(struct tree_balance *tb,
614 struct item_head *ih, const char *body)
616 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
617 int n = B_NR_ITEMS(tbS0);
618 struct buffer_info bi;
621 if (n - tb->rnum[0] <= tb->item_pos) { /* pasted item or part of it falls to R[0] */
622 if (tb->item_pos == n - tb->rnum[0] && tb->rbytes != -1) { /* we must shift the part of the appended item */
623 if (is_direntry_le_ih(item_head(tbS0, tb->item_pos))) { /* we append to directory item */
626 RFALSE(tb->zeroes_num,
627 "PAP-12145: invalid parameter in case of a directory");
628 entry_count = ih_entry_count(item_head
629 (tbS0, tb->item_pos));
630 if (entry_count - tb->rbytes <
632 /* new directory entry falls into R[0] */
634 int paste_entry_position;
636 RFALSE(tb->rbytes - 1 >= entry_count || !tb-> insert_size[0],
637 "PAP-12150: no enough of entries to shift to R[0]: rbytes=%d, entry_count=%d",
638 tb->rbytes, entry_count);
639 /* Shift rnum[0]-1 items in whole. Shift rbytes-1 directory entries from directory item number rnum[0] */
640 leaf_shift_right(tb, tb->rnum[0], tb->rbytes - 1);
641 /* Paste given directory entry to directory item */
642 paste_entry_position = tb->pos_in_item - entry_count + tb->rbytes - 1;
643 buffer_info_init_right(tb, &bi);
644 leaf_paste_in_buffer(&bi, 0, paste_entry_position, tb->insert_size[0], body, tb->zeroes_num);
646 leaf_paste_entries(&bi, 0, paste_entry_position, 1,
647 (struct reiserfs_de_head *) body,
648 body + DEH_SIZE, tb->insert_size[0]);
650 if (paste_entry_position == 0) {
651 /* change delimiting keys */
652 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0],0);
655 tb->insert_size[0] = 0;
657 } else { /* new directory entry doesn't fall into R[0] */
659 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
661 } else { /* regular object */
663 int n_shift, n_rem, r_zeroes_number;
666 /* Calculate number of bytes which must be shifted from appended item */
667 if ((n_shift = tb->rbytes - tb->insert_size[0]) < 0)
670 RFALSE(tb->pos_in_item != ih_item_len
671 (item_head(tbS0, tb->item_pos)),
672 "PAP-12155: invalid position to paste. ih_item_len=%d, pos_in_item=%d",
673 tb->pos_in_item, ih_item_len
674 (item_head(tbS0, tb->item_pos)));
676 leaf_shift_right(tb, tb->rnum[0], n_shift);
677 /* Calculate number of bytes which must remain in body after appending to R[0] */
678 if ((n_rem = tb->insert_size[0] - tb->rbytes) < 0)
683 unsigned long temp_rem = n_rem;
685 version = ih_version(item_head(tb->R[0], 0));
686 if (is_indirect_le_key(version, leaf_key(tb->R[0], 0))) {
687 temp_rem = n_rem << (tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT);
689 set_le_key_k_offset(version, leaf_key(tb->R[0], 0),
690 le_key_k_offset(version, leaf_key(tb->R[0], 0)) + temp_rem);
691 set_le_key_k_offset(version, internal_key(tb->CFR[0], tb->rkey[0]),
692 le_key_k_offset(version, internal_key(tb->CFR[0], tb->rkey[0])) + temp_rem);
694 /* k_offset (leaf_key(tb->R[0],0)) += n_rem;
695 k_offset (internal_key(tb->CFR[0],tb->rkey[0])) += n_rem;*/
696 do_balance_mark_internal_dirty(tb, tb->CFR[0], 0);
698 /* Append part of body into R[0] */
699 buffer_info_init_right(tb, &bi);
700 if (n_rem > tb->zeroes_num) {
702 r_body = body + n_rem - tb->zeroes_num;
705 r_zeroes_number = tb->zeroes_num - n_rem;
706 tb->zeroes_num -= r_zeroes_number;
709 leaf_paste_in_buffer(&bi, 0, n_shift,
710 tb->insert_size[0] - n_rem,
711 r_body, r_zeroes_number);
713 if (is_indirect_le_ih(item_head(tb->R[0], 0))) {
716 "PAP-12160: paste more than one unformatted node pointer");
718 set_ih_free_space(item_head(tb->R[0], 0), 0);
720 tb->insert_size[0] = n_rem;
724 } else { /* pasted item in whole falls into R[0] */
726 struct item_head *pasted;
728 ret_val = leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
729 /* append item in R[0] */
730 if (tb->pos_in_item >= 0) {
731 buffer_info_init_right(tb, &bi);
732 leaf_paste_in_buffer(&bi, tb->item_pos - n + tb->rnum[0], tb->pos_in_item,
733 tb->insert_size[0], body, tb->zeroes_num);
736 /* paste new entry, if item is directory item */
737 pasted = item_head(tb->R[0], tb->item_pos - n + tb->rnum[0]);
738 if (is_direntry_le_ih(pasted) && tb->pos_in_item >= 0) {
739 leaf_paste_entries(&bi, tb->item_pos - n + tb->rnum[0],
741 (struct reiserfs_de_head *) body,
742 body + DEH_SIZE, tb->insert_size[0]);
743 if (!tb->pos_in_item) {
745 RFALSE(tb->item_pos - n + tb->rnum[0],
746 "PAP-12165: directory item must be first item of node when pasting is in 0th position");
748 /* update delimiting keys */
749 replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0);
753 if (is_indirect_le_ih(pasted))
754 set_ih_free_space(pasted, 0);
755 tb->zeroes_num = tb->insert_size[0] = 0;
757 } else { /* new item doesn't fall into R[0] */
759 leaf_shift_right(tb, tb->rnum[0], tb->rbytes);
764 /* shift rnum[0] items from S[0] to the right neighbor R[0] */
765 static void balance_leaf_right(struct tree_balance *tb, struct item_head *ih,
766 const char *body, int flag)
768 if (tb->rnum[0] <= 0)
771 BUG_ON(flag != M_INSERT && flag != M_PASTE);
773 if (flag == M_INSERT)
774 balance_leaf_insert_right(tb, ih, body);
776 balance_leaf_paste_right(tb, ih, body);
780 static void balance_leaf_new_nodes_insert(struct tree_balance *tb,
781 struct item_head *ih,
783 struct item_head *insert_key,
784 struct buffer_head **insert_ptr,
787 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
788 int n = B_NR_ITEMS(tbS0);
789 struct buffer_info bi;
790 if (n - tb->snum[i] < tb->item_pos) { /* new item or it's part falls to first new node S_new[i] */
791 if (tb->item_pos == n - tb->snum[i] + 1 && tb->sbytes[i] != -1) { /* part of new item falls into S_new[i] */
792 int old_key_comp, old_len, r_zeroes_number;
796 /* Move snum[i]-1 items from S[0] to S_new[i] */
797 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
800 /* Remember key component and item length */
801 version = ih_version(ih);
802 old_key_comp = le_ih_k_offset(ih);
803 old_len = ih_item_len(ih);
805 /* Calculate key component and item length to insert into S_new[i] */
806 set_le_ih_k_offset(ih, le_ih_k_offset(ih) +
807 ((old_len - tb->sbytes[i]) << (is_indirect_le_ih(ih) ? tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT : 0)));
809 put_ih_item_len(ih, tb->sbytes[i]);
811 /* Insert part of the item into S_new[i] before 0-th item */
812 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
814 if ((old_len - tb->sbytes[i]) > tb->zeroes_num) {
816 r_body = body + (old_len - tb->sbytes[i]) - tb->zeroes_num;
819 r_zeroes_number = tb->zeroes_num - (old_len - tb->sbytes[i]);
820 tb->zeroes_num -= r_zeroes_number;
823 leaf_insert_into_buf(&bi, 0, ih, r_body, r_zeroes_number);
825 /* Calculate key component and item length to insert into S[i] */
826 set_le_ih_k_offset(ih, old_key_comp);
827 put_ih_item_len(ih, old_len - tb->sbytes[i]);
828 tb->insert_size[0] -= tb->sbytes[i];
829 } else { /* whole new item falls into S_new[i] */
831 /* Shift snum[0] - 1 items to S_new[i] (sbytes[i] of split item) */
832 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
833 tb->snum[i] - 1, tb->sbytes[i], tb->S_new[i]);
835 /* Insert new item into S_new[i] */
836 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
837 leaf_insert_into_buf(&bi, tb->item_pos - n + tb->snum[i] - 1,
838 ih, body, tb->zeroes_num);
840 tb->zeroes_num = tb->insert_size[0] = 0;
844 else { /* new item or it part don't falls into S_new[i] */
846 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
847 tb->snum[i], tb->sbytes[i], tb->S_new[i]);
851 static void balance_leaf_new_nodes_paste(struct tree_balance *tb,
852 struct item_head *ih,
854 struct item_head *insert_key,
855 struct buffer_head **insert_ptr,
858 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
859 int n = B_NR_ITEMS(tbS0);
860 struct buffer_info bi;
861 if (n - tb->snum[i] <= tb->item_pos) { /* pasted item or part if it falls to S_new[i] */
862 if (tb->item_pos == n - tb->snum[i] && tb->sbytes[i] != -1) { /* we must shift part of the appended item */
863 struct item_head *aux_ih;
865 RFALSE(ih, "PAP-12210: ih must be 0");
867 aux_ih = item_head(tbS0, tb->item_pos);
868 if (is_direntry_le_ih(aux_ih)) {
869 /* we append to directory item */
873 entry_count = ih_entry_count(aux_ih);
875 if (entry_count - tb->sbytes[i] < tb->pos_in_item && tb->pos_in_item <= entry_count) {
876 /* new directory entry falls into S_new[i] */
878 RFALSE(!tb->insert_size[0], "PAP-12215: insert_size is already 0");
879 RFALSE(tb->sbytes[i] - 1 >= entry_count,
880 "PAP-12220: there are no so much entries (%d), only %d",
881 tb->sbytes[i] - 1, entry_count);
883 /* Shift snum[i]-1 items in whole. Shift sbytes[i] directory entries from directory item number snum[i] */
884 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i], tb->sbytes[i] - 1, tb->S_new[i]);
885 /* Paste given directory entry to directory item */
886 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
887 leaf_paste_in_buffer(&bi, 0, tb->pos_in_item - entry_count + tb->sbytes[i] - 1,
888 tb->insert_size[0], body, tb->zeroes_num);
889 /* paste new directory entry */
890 leaf_paste_entries(&bi, 0, tb->pos_in_item - entry_count + tb->sbytes[i] - 1, 1,
891 (struct reiserfs_de_head *) body,
892 body + DEH_SIZE, tb->insert_size[0]);
893 tb->insert_size[0] = 0;
895 } else { /* new directory entry doesn't fall into S_new[i] */
896 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i], tb->sbytes[i], tb->S_new[i]);
898 } else { /* regular object */
900 int n_shift, n_rem, r_zeroes_number;
903 RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)) || tb->insert_size[0] <= 0,
904 "PAP-12225: item too short or insert_size <= 0");
906 /* Calculate number of bytes which must be shifted from appended item */
907 n_shift = tb->sbytes[i] - tb->insert_size[0];
910 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i], n_shift, tb->S_new[i]);
912 /* Calculate number of bytes which must remain in body after append to S_new[i] */
913 n_rem = tb->insert_size[0] - tb->sbytes[i];
916 /* Append part of body into S_new[0] */
917 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
918 if (n_rem > tb->zeroes_num) {
920 r_body = body + n_rem - tb->zeroes_num;
923 r_zeroes_number = tb->zeroes_num - n_rem;
924 tb->zeroes_num -= r_zeroes_number;
927 leaf_paste_in_buffer(&bi, 0, n_shift,
928 tb->insert_size[0] - n_rem,
929 r_body, r_zeroes_number);
931 struct item_head *tmp;
933 tmp = item_head(tb->S_new[i], 0);
934 if (is_indirect_le_ih
936 set_ih_free_space(tmp, 0);
937 set_le_ih_k_offset(tmp, le_ih_k_offset(tmp) + (n_rem << (tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT)));
939 set_le_ih_k_offset(tmp, le_ih_k_offset(tmp) + n_rem);
943 tb->insert_size[0] = n_rem;
948 /* item falls wholly into S_new[i] */
951 struct item_head *pasted;
953 #ifdef CONFIG_REISERFS_CHECK
954 struct item_head *ih_check = item_head(tbS0, tb->item_pos);
956 if (!is_direntry_le_ih(ih_check)
957 && (tb->pos_in_item != ih_item_len(ih_check)
958 || tb->insert_size[0] <= 0))
959 reiserfs_panic(tb->tb_sb,
964 #endif /* CONFIG_REISERFS_CHECK */
966 leaf_mi = leaf_move_items(LEAF_FROM_S_TO_SNEW,
972 "PAP-12240: unexpected value returned by leaf_move_items (%d)",
975 /* paste into item */
976 buffer_info_init_bh(tb, &bi, tb->S_new[i]);
977 leaf_paste_in_buffer(&bi,
978 tb->item_pos - n + tb->snum[i],
981 body, tb->zeroes_num);
983 pasted = item_head(tb->S_new[i], tb->item_pos - n + tb->snum[i]);
984 if (is_direntry_le_ih(pasted)) {
985 leaf_paste_entries(&bi,
986 tb->item_pos - n + tb->snum[i],
988 (struct reiserfs_de_head *)body,
994 /* if we paste to indirect item update ih_free_space */
995 if (is_indirect_le_ih(pasted))
996 set_ih_free_space(pasted, 0);
997 tb->zeroes_num = tb->insert_size[0] = 0;
1001 else { /* pasted item doesn't fall into S_new[i] */
1003 leaf_move_items(LEAF_FROM_S_TO_SNEW, tb,
1004 tb->snum[i], tb->sbytes[i], tb->S_new[i]);
1009 /* Fill new nodes that appear in place of S[0] */
1010 static void balance_leaf_new_nodes(struct tree_balance *tb,
1011 struct item_head *ih,
1013 struct item_head *insert_key,
1014 struct buffer_head **insert_ptr,
1018 for (i = tb->blknum[0] - 2; i >= 0; i--) {
1020 RFALSE(!tb->snum[i],
1021 "PAP-12200: snum[%d] == %d. Must be > 0", i,
1024 /* here we shift from S to S_new nodes */
1026 tb->S_new[i] = get_FEB(tb);
1028 /* initialized block type and tree level */
1029 set_blkh_level(B_BLK_HEAD(tb->S_new[i]), DISK_LEAF_NODE_LEVEL);
1032 case M_INSERT: /* insert item */
1033 balance_leaf_new_nodes_insert(tb, ih, body, insert_key,
1037 case M_PASTE: /* append item */
1038 balance_leaf_new_nodes_paste(tb, ih, body, insert_key,
1041 default: /* cases d and t */
1042 reiserfs_panic(tb->tb_sb, "PAP-12245",
1043 "blknum > 2: unexpected mode: %s(%d)",
1044 (flag == M_DELETE) ? "DELETE" : ((flag == M_CUT) ? "CUT" : "UNKNOWN"), flag);
1047 memcpy(insert_key + i, leaf_key(tb->S_new[i], 0), KEY_SIZE);
1048 insert_ptr[i] = tb->S_new[i];
1050 RFALSE(!buffer_journaled(tb->S_new[i])
1051 || buffer_journal_dirty(tb->S_new[i])
1052 || buffer_dirty(tb->S_new[i]),
1053 "PAP-12247: S_new[%d] : (%b)",
1054 i, format_bh(tb->S_new[i]));
1058 static void balance_leaf_finish_node_insert(struct tree_balance *tb,
1059 struct item_head *ih,
1062 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1063 struct buffer_info bi;
1064 buffer_info_init_tbS0(tb, &bi);
1065 leaf_insert_into_buf(&bi, tb->item_pos, ih,
1066 body, tb->zeroes_num);
1069 * If we insert the first key
1070 * change the delimiting key
1072 if (tb->item_pos == 0) {
1073 if (tb->CFL[0]) /* can be 0 in reiserfsck */
1074 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0);
1078 static void balance_leaf_finish_node_paste(struct tree_balance *tb,
1079 struct item_head *ih,
1082 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1083 struct buffer_info bi;
1084 struct item_head *pasted;
1086 pasted = item_head(tbS0, tb->item_pos);
1087 /* when directory, may be new entry already pasted */
1088 if (is_direntry_le_ih(pasted)) {
1089 if (tb->pos_in_item >= 0 && tb->pos_in_item <= ih_entry_count(pasted)) {
1091 RFALSE(!tb->insert_size[0],
1092 "PAP-12260: insert_size is 0 already");
1095 buffer_info_init_tbS0(tb, &bi);
1096 leaf_paste_in_buffer(&bi, tb->item_pos, tb->pos_in_item,
1097 tb->insert_size[0], body,
1101 leaf_paste_entries(&bi, tb->item_pos, tb->pos_in_item, 1,
1102 (struct reiserfs_de_head *)body,
1104 tb->insert_size[0]);
1105 if (!tb->item_pos && !tb->pos_in_item) {
1106 RFALSE(!tb->CFL[0] || !tb->L[0],
1107 "PAP-12270: CFL[0]/L[0] must be specified");
1109 replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0);
1111 tb->insert_size[0] = 0;
1113 } else { /* regular object */
1114 if (tb->pos_in_item == ih_item_len(pasted)) {
1116 RFALSE(tb->insert_size[0] <= 0,
1117 "PAP-12275: insert size must not be %d",
1118 tb->insert_size[0]);
1119 buffer_info_init_tbS0(tb, &bi);
1120 leaf_paste_in_buffer(&bi, tb->item_pos, tb->pos_in_item,
1121 tb->insert_size[0], body, tb->zeroes_num);
1123 if (is_indirect_le_ih(pasted)) {
1128 "PAP-12280: insert_size for indirect item must be %d, not %d",
1133 set_ih_free_space(pasted, 0);
1135 tb->insert_size[0] = 0;
1137 #ifdef CONFIG_REISERFS_CHECK
1139 if (tb->insert_size[0]) {
1140 print_cur_tb("12285");
1141 reiserfs_panic(tb->tb_sb,
1146 tb->insert_size[0]);
1149 #endif /* CONFIG_REISERFS_CHECK */
1155 * if the affected item was not wholly shifted then we
1156 * perform all necessary operations on that part or whole
1157 * of the affected item which remains in S
1159 static void balance_leaf_finish_node(struct tree_balance *tb,
1160 struct item_head *ih,
1161 const char *body, int flag)
1163 /* if we must insert or append into buffer S[0] */
1164 if (0 <= tb->item_pos && tb->item_pos < tb->s0num) {
1165 if (flag == M_INSERT)
1166 balance_leaf_finish_node_insert(tb, ih, body);
1168 balance_leaf_finish_node_paste(tb, ih, body);
1173 * balance_leaf - reiserfs tree balancing algorithm
1174 * @tb: tree balance state
1175 * @ih: item header of inserted item (little endian)
1176 * @body: body of inserted item or bytes to paste
1177 * @flag: i - insert, d - delete, c - cut, p - paste (see do_balance)
1179 * @insert_key: key to insert new nodes
1180 * @insert_ptr: array of nodes to insert at the next level
1182 * In our processing of one level we sometimes determine what must be
1183 * inserted into the next higher level. This insertion consists of a
1184 * key or two keys and their corresponding pointers.
1186 static int balance_leaf(struct tree_balance *tb, struct item_head *ih,
1187 const char *body, int flag,
1188 struct item_head *insert_key,
1189 struct buffer_head **insert_ptr)
1191 struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path);
1193 PROC_INFO_INC(tb->tb_sb, balance_at[0]);
1195 /* Make balance in case insert_size[0] < 0 */
1196 if (tb->insert_size[0] < 0)
1197 return balance_leaf_when_delete(tb, flag);
1199 tb->item_pos = PATH_LAST_POSITION(tb->tb_path),
1200 tb->pos_in_item = tb->tb_path->pos_in_item,
1202 if (flag == M_INSERT && !body)
1203 tb->zeroes_num = ih_item_len(ih);
1206 * for indirect item pos_in_item is measured in unformatted node
1207 * pointers. Recalculate to bytes
1209 if (flag != M_INSERT
1210 && is_indirect_le_ih(item_head(tbS0, tb->item_pos)))
1211 tb->pos_in_item *= UNFM_P_SIZE;
1213 balance_leaf_left(tb, ih, body, flag);
1215 /* tb->lnum[0] > 0 */
1216 /* Calculate new item position */
1217 tb->item_pos -= (tb->lnum[0] - ((tb->lbytes != -1) ? 1 : 0));
1219 balance_leaf_right(tb, ih, body, flag);
1221 /* tb->rnum[0] > 0 */
1222 RFALSE(tb->blknum[0] > 3,
1223 "PAP-12180: blknum can not be %d. It must be <= 3", tb->blknum[0]);
1224 RFALSE(tb->blknum[0] < 0,
1225 "PAP-12185: blknum can not be %d. It must be >= 0", tb->blknum[0]);
1228 * if while adding to a node we discover that it is possible to split
1229 * it in two, and merge the left part into the left neighbor and the
1230 * right part into the right neighbor, eliminating the node
1232 if (tb->blknum[0] == 0) { /* node S[0] is empty now */
1234 RFALSE(!tb->lnum[0] || !tb->rnum[0],
1235 "PAP-12190: lnum and rnum must not be zero");
1237 * if insertion was done before 0-th position in R[0], right
1238 * delimiting key of the tb->L[0]'s and left delimiting key are
1243 reiserfs_panic(tb->tb_sb, "vs-12195",
1244 "CFR not initialized");
1245 copy_key(internal_key(tb->CFL[0], tb->lkey[0]),
1246 internal_key(tb->CFR[0], tb->rkey[0]));
1247 do_balance_mark_internal_dirty(tb, tb->CFL[0], 0);
1250 reiserfs_invalidate_buffer(tb, tbS0);
1254 balance_leaf_new_nodes(tb, ih, body, insert_key, insert_ptr, flag);
1256 balance_leaf_finish_node(tb, ih, body, flag);
1258 #ifdef CONFIG_REISERFS_CHECK
1259 if (flag == M_PASTE && tb->insert_size[0]) {
1260 print_cur_tb("12290");
1261 reiserfs_panic(tb->tb_sb,
1262 "PAP-12290", "insert_size is still not 0 (%d)",
1263 tb->insert_size[0]);
1267 /* Leaf level of the tree is balanced (end of balance_leaf) */
1271 /* Make empty node */
1272 void make_empty_node(struct buffer_info *bi)
1274 struct block_head *blkh;
1276 RFALSE(bi->bi_bh == NULL, "PAP-12295: pointer to the buffer is NULL");
1278 blkh = B_BLK_HEAD(bi->bi_bh);
1279 set_blkh_nr_item(blkh, 0);
1280 set_blkh_free_space(blkh, MAX_CHILD_SIZE(bi->bi_bh));
1283 B_N_CHILD(bi->bi_parent, bi->bi_position)->dc_size = 0; /* Endian safe if 0 */
1286 /* Get first empty buffer */
1287 struct buffer_head *get_FEB(struct tree_balance *tb)
1290 struct buffer_info bi;
1292 for (i = 0; i < MAX_FEB_SIZE; i++)
1293 if (tb->FEB[i] != NULL)
1296 if (i == MAX_FEB_SIZE)
1297 reiserfs_panic(tb->tb_sb, "vs-12300", "FEB list is empty");
1299 buffer_info_init_bh(tb, &bi, tb->FEB[i]);
1300 make_empty_node(&bi);
1301 set_buffer_uptodate(tb->FEB[i]);
1302 tb->used[i] = tb->FEB[i];
1308 /* This is now used because reiserfs_free_block has to be able to schedule. */
1309 static void store_thrown(struct tree_balance *tb, struct buffer_head *bh)
1313 if (buffer_dirty(bh))
1314 reiserfs_warning(tb->tb_sb, "reiserfs-12320",
1315 "called with dirty buffer");
1316 for (i = 0; i < ARRAY_SIZE(tb->thrown); i++)
1317 if (!tb->thrown[i]) {
1319 get_bh(bh); /* free_thrown puts this */
1322 reiserfs_warning(tb->tb_sb, "reiserfs-12321",
1323 "too many thrown buffers");
1326 static void free_thrown(struct tree_balance *tb)
1329 b_blocknr_t blocknr;
1330 for (i = 0; i < ARRAY_SIZE(tb->thrown); i++) {
1331 if (tb->thrown[i]) {
1332 blocknr = tb->thrown[i]->b_blocknr;
1333 if (buffer_dirty(tb->thrown[i]))
1334 reiserfs_warning(tb->tb_sb, "reiserfs-12322",
1335 "called with dirty buffer %d",
1337 brelse(tb->thrown[i]); /* incremented in store_thrown */
1338 reiserfs_free_block(tb->transaction_handle, NULL,
1344 void reiserfs_invalidate_buffer(struct tree_balance *tb, struct buffer_head *bh)
1346 struct block_head *blkh;
1347 blkh = B_BLK_HEAD(bh);
1348 set_blkh_level(blkh, FREE_LEVEL);
1349 set_blkh_nr_item(blkh, 0);
1351 clear_buffer_dirty(bh);
1352 store_thrown(tb, bh);
1355 /* Replace n_dest'th key in buffer dest by n_src'th key of buffer src.*/
1356 void replace_key(struct tree_balance *tb, struct buffer_head *dest, int n_dest,
1357 struct buffer_head *src, int n_src)
1360 RFALSE(dest == NULL || src == NULL,
1361 "vs-12305: source or destination buffer is 0 (src=%p, dest=%p)",
1363 RFALSE(!B_IS_KEYS_LEVEL(dest),
1364 "vs-12310: invalid level (%z) for destination buffer. dest must be leaf",
1366 RFALSE(n_dest < 0 || n_src < 0,
1367 "vs-12315: src(%d) or dest(%d) key number < 0", n_src, n_dest);
1368 RFALSE(n_dest >= B_NR_ITEMS(dest) || n_src >= B_NR_ITEMS(src),
1369 "vs-12320: src(%d(%d)) or dest(%d(%d)) key number is too big",
1370 n_src, B_NR_ITEMS(src), n_dest, B_NR_ITEMS(dest));
1372 if (B_IS_ITEMS_LEVEL(src))
1373 /* source buffer contains leaf node */
1374 memcpy(internal_key(dest, n_dest), item_head(src, n_src),
1377 memcpy(internal_key(dest, n_dest), internal_key(src, n_src),
1380 do_balance_mark_internal_dirty(tb, dest, 0);
1383 int get_left_neighbor_position(struct tree_balance *tb, int h)
1385 int Sh_position = PATH_H_POSITION(tb->tb_path, h + 1);
1387 RFALSE(PATH_H_PPARENT(tb->tb_path, h) == NULL || tb->FL[h] == NULL,
1388 "vs-12325: FL[%d](%p) or F[%d](%p) does not exist",
1389 h, tb->FL[h], h, PATH_H_PPARENT(tb->tb_path, h));
1391 if (Sh_position == 0)
1392 return B_NR_ITEMS(tb->FL[h]);
1394 return Sh_position - 1;
1397 int get_right_neighbor_position(struct tree_balance *tb, int h)
1399 int Sh_position = PATH_H_POSITION(tb->tb_path, h + 1);
1401 RFALSE(PATH_H_PPARENT(tb->tb_path, h) == NULL || tb->FR[h] == NULL,
1402 "vs-12330: F[%d](%p) or FR[%d](%p) does not exist",
1403 h, PATH_H_PPARENT(tb->tb_path, h), h, tb->FR[h]);
1405 if (Sh_position == B_NR_ITEMS(PATH_H_PPARENT(tb->tb_path, h)))
1408 return Sh_position + 1;
1411 #ifdef CONFIG_REISERFS_CHECK
1413 int is_reusable(struct super_block *s, b_blocknr_t block, int bit_value);
1414 static void check_internal_node(struct super_block *s, struct buffer_head *bh,
1417 struct disk_child *dc;
1420 RFALSE(!bh, "PAP-12336: bh == 0");
1422 if (!bh || !B_IS_IN_TREE(bh))
1425 RFALSE(!buffer_dirty(bh) &&
1426 !(buffer_journaled(bh) || buffer_journal_dirty(bh)),
1427 "PAP-12337: buffer (%b) must be dirty", bh);
1428 dc = B_N_CHILD(bh, 0);
1430 for (i = 0; i <= B_NR_ITEMS(bh); i++, dc++) {
1431 if (!is_reusable(s, dc_block_number(dc), 1)) {
1433 reiserfs_panic(s, "PAP-12338",
1434 "invalid child pointer %y in %b",
1440 static int locked_or_not_in_tree(struct tree_balance *tb,
1441 struct buffer_head *bh, char *which)
1443 if ((!buffer_journal_prepared(bh) && buffer_locked(bh)) ||
1444 !B_IS_IN_TREE(bh)) {
1445 reiserfs_warning(tb->tb_sb, "vs-12339", "%s (%b)", which, bh);
1451 static int check_before_balancing(struct tree_balance *tb)
1455 if (REISERFS_SB(tb->tb_sb)->cur_tb) {
1456 reiserfs_panic(tb->tb_sb, "vs-12335", "suspect that schedule "
1457 "occurred based on cur_tb not being null at "
1458 "this point in code. do_balance cannot properly "
1459 "handle concurrent tree accesses on a same "
1464 * double check that buffers that we will modify are unlocked.
1465 * (fix_nodes should already have prepped all of these for us).
1468 retval |= locked_or_not_in_tree(tb, tb->L[0], "L[0]");
1469 retval |= locked_or_not_in_tree(tb, tb->FL[0], "FL[0]");
1470 retval |= locked_or_not_in_tree(tb, tb->CFL[0], "CFL[0]");
1471 check_leaf(tb->L[0]);
1474 retval |= locked_or_not_in_tree(tb, tb->R[0], "R[0]");
1475 retval |= locked_or_not_in_tree(tb, tb->FR[0], "FR[0]");
1476 retval |= locked_or_not_in_tree(tb, tb->CFR[0], "CFR[0]");
1477 check_leaf(tb->R[0]);
1479 retval |= locked_or_not_in_tree(tb, PATH_PLAST_BUFFER(tb->tb_path),
1481 check_leaf(PATH_PLAST_BUFFER(tb->tb_path));
1486 static void check_after_balance_leaf(struct tree_balance *tb)
1489 if (B_FREE_SPACE(tb->L[0]) !=
1490 MAX_CHILD_SIZE(tb->L[0]) -
1492 (tb->FL[0], get_left_neighbor_position(tb, 0)))) {
1493 print_cur_tb("12221");
1494 reiserfs_panic(tb->tb_sb, "PAP-12355",
1495 "shift to left was incorrect");
1499 if (B_FREE_SPACE(tb->R[0]) !=
1500 MAX_CHILD_SIZE(tb->R[0]) -
1502 (tb->FR[0], get_right_neighbor_position(tb, 0)))) {
1503 print_cur_tb("12222");
1504 reiserfs_panic(tb->tb_sb, "PAP-12360",
1505 "shift to right was incorrect");
1508 if (PATH_H_PBUFFER(tb->tb_path, 1) &&
1509 (B_FREE_SPACE(PATH_H_PBUFFER(tb->tb_path, 0)) !=
1510 (MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)) -
1511 dc_size(B_N_CHILD(PATH_H_PBUFFER(tb->tb_path, 1),
1512 PATH_H_POSITION(tb->tb_path, 1)))))) {
1513 int left = B_FREE_SPACE(PATH_H_PBUFFER(tb->tb_path, 0));
1514 int right = (MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)) -
1515 dc_size(B_N_CHILD(PATH_H_PBUFFER(tb->tb_path, 1),
1516 PATH_H_POSITION(tb->tb_path,
1518 print_cur_tb("12223");
1519 reiserfs_warning(tb->tb_sb, "reiserfs-12363",
1520 "B_FREE_SPACE (PATH_H_PBUFFER(tb->tb_path,0)) = %d; "
1521 "MAX_CHILD_SIZE (%d) - dc_size( %y, %d ) [%d] = %d",
1523 MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)),
1524 PATH_H_PBUFFER(tb->tb_path, 1),
1525 PATH_H_POSITION(tb->tb_path, 1),
1527 (PATH_H_PBUFFER(tb->tb_path, 1),
1528 PATH_H_POSITION(tb->tb_path, 1))),
1530 reiserfs_panic(tb->tb_sb, "PAP-12365", "S is incorrect");
1534 static void check_leaf_level(struct tree_balance *tb)
1536 check_leaf(tb->L[0]);
1537 check_leaf(tb->R[0]);
1538 check_leaf(PATH_PLAST_BUFFER(tb->tb_path));
1541 static void check_internal_levels(struct tree_balance *tb)
1545 /* check all internal nodes */
1546 for (h = 1; tb->insert_size[h]; h++) {
1547 check_internal_node(tb->tb_sb, PATH_H_PBUFFER(tb->tb_path, h),
1548 "BAD BUFFER ON PATH");
1550 check_internal_node(tb->tb_sb, tb->L[h], "BAD L");
1552 check_internal_node(tb->tb_sb, tb->R[h], "BAD R");
1560 * Now we have all of the buffers that must be used in balancing of
1561 * the tree. We rely on the assumption that schedule() will not occur
1562 * while do_balance works. ( Only interrupt handlers are acceptable.)
1563 * We balance the tree according to the analysis made before this,
1564 * using buffers already obtained. For SMP support it will someday be
1565 * necessary to add ordered locking of tb.
1569 * Some interesting rules of balancing:
1570 * we delete a maximum of two nodes per level per balancing: we never
1571 * delete R, when we delete two of three nodes L, S, R then we move
1574 * we only delete L if we are deleting two nodes, if we delete only
1575 * one node we delete S
1577 * if we shift leaves then we shift as much as we can: this is a
1578 * deliberate policy of extremism in node packing which results in
1579 * higher average utilization after repeated random balance operations
1580 * at the cost of more memory copies and more balancing as a result of
1581 * small insertions to full nodes.
1583 * if we shift internal nodes we try to evenly balance the node
1584 * utilization, with consequent less balancing at the cost of lower
1587 * one could argue that the policy for directories in leaves should be
1588 * that of internal nodes, but we will wait until another day to
1589 * evaluate this.... It would be nice to someday measure and prove
1590 * these assumptions as to what is optimal....
1593 static inline void do_balance_starts(struct tree_balance *tb)
1595 /* use print_cur_tb() to see initial state of struct tree_balance */
1597 /* store_print_tb (tb); */
1599 /* do not delete, just comment it out */
1601 print_tb(flag, PATH_LAST_POSITION(tb->tb_path),
1602 tb->tb_path->pos_in_item, tb, "check");
1604 RFALSE(check_before_balancing(tb), "PAP-12340: locked buffers in TB");
1605 #ifdef CONFIG_REISERFS_CHECK
1606 REISERFS_SB(tb->tb_sb)->cur_tb = tb;
1610 static inline void do_balance_completed(struct tree_balance *tb)
1613 #ifdef CONFIG_REISERFS_CHECK
1614 check_leaf_level(tb);
1615 check_internal_levels(tb);
1616 REISERFS_SB(tb->tb_sb)->cur_tb = NULL;
1620 * reiserfs_free_block is no longer schedule safe. So, we need to
1621 * put the buffers we want freed on the thrown list during do_balance,
1622 * and then free them now
1625 REISERFS_SB(tb->tb_sb)->s_do_balance++;
1627 /* release all nodes hold to perform the balancing */
1634 * do_balance - balance the tree
1636 * @tb: tree_balance structure
1637 * @ih: item header of inserted item
1638 * @body: body of inserted item or bytes to paste
1639 * @flag: 'i' - insert, 'd' - delete, 'c' - cut, 'p' paste
1641 * Cut means delete part of an item (includes removing an entry from a
1644 * Delete means delete whole item.
1646 * Insert means add a new item into the tree.
1648 * Paste means to append to the end of an existing file or to
1649 * insert a directory entry.
1651 void do_balance(struct tree_balance *tb, struct item_head *ih,
1652 const char *body, int flag)
1654 int child_pos; /* position of a child node in its parent */
1655 int h; /* level of the tree being processed */
1658 * in our processing of one level we sometimes determine what
1659 * must be inserted into the next higher level. This insertion
1660 * consists of a key or two keys and their corresponding
1663 struct item_head insert_key[2];
1665 /* inserted node-ptrs for the next level */
1666 struct buffer_head *insert_ptr[2];
1669 tb->need_balance_dirty = 0;
1671 if (FILESYSTEM_CHANGED_TB(tb)) {
1672 reiserfs_panic(tb->tb_sb, "clm-6000", "fs generation has "
1675 /* if we have no real work to do */
1676 if (!tb->insert_size[0]) {
1677 reiserfs_warning(tb->tb_sb, "PAP-12350",
1678 "insert_size == 0, mode == %c", flag);
1683 atomic_inc(&fs_generation(tb->tb_sb));
1684 do_balance_starts(tb);
1687 * balance_leaf returns 0 except if combining L R and S into
1688 * one node. see balance_internal() for explanation of this
1691 child_pos = PATH_H_B_ITEM_ORDER(tb->tb_path, 0) +
1692 balance_leaf(tb, ih, body, flag, insert_key, insert_ptr);
1694 #ifdef CONFIG_REISERFS_CHECK
1695 check_after_balance_leaf(tb);
1698 /* Balance internal level of the tree. */
1699 for (h = 1; h < MAX_HEIGHT && tb->insert_size[h]; h++)
1701 balance_internal(tb, h, child_pos, insert_key, insert_ptr);
1703 do_balance_completed(tb);