2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
6 * Written by Anatoly P. Pinchuk pap@namesys.botik.ru
7 * Programm System Institute
8 * Pereslavl-Zalessky Russia
12 * This file contains functions dealing with S+tree
26 * decrement_counters_in_path
28 * pathrelse_and_restore
32 * search_for_position_by_key
34 * prepare_for_direct_item
35 * prepare_for_direntry_item
36 * prepare_for_delete_or_cut
37 * calc_deleted_bytes_number
40 * reiserfs_delete_item
41 * reiserfs_delete_solid_item
42 * reiserfs_delete_object
43 * maybe_indirect_to_direct
44 * indirect_to_direct_roll_back
45 * reiserfs_cut_from_item
47 * reiserfs_do_truncate
48 * reiserfs_paste_into_item
49 * reiserfs_insert_item
52 #include <linux/time.h>
53 #include <linux/string.h>
54 #include <linux/pagemap.h>
55 #include <linux/reiserfs_fs.h>
56 #include <linux/buffer_head.h>
57 #include <linux/quotaops.h>
59 /* Does the buffer contain a disk block which is in the tree. */
60 inline int B_IS_IN_TREE(const struct buffer_head *p_s_bh)
63 RFALSE(B_LEVEL(p_s_bh) > MAX_HEIGHT,
64 "PAP-1010: block (%b) has too big level (%z)", p_s_bh, p_s_bh);
66 return (B_LEVEL(p_s_bh) != FREE_LEVEL);
70 // to gets item head in le form
72 inline void copy_item_head(struct item_head *p_v_to,
73 const struct item_head *p_v_from)
75 memcpy(p_v_to, p_v_from, IH_SIZE);
78 /* k1 is pointer to on-disk structure which is stored in little-endian
79 form. k2 is pointer to cpu variable. For key of items of the same
80 object this returns 0.
81 Returns: -1 if key1 < key2
84 inline int comp_short_keys(const struct reiserfs_key *le_key,
85 const struct cpu_key *cpu_key)
88 n = le32_to_cpu(le_key->k_dir_id);
89 if (n < cpu_key->on_disk_key.k_dir_id)
91 if (n > cpu_key->on_disk_key.k_dir_id)
93 n = le32_to_cpu(le_key->k_objectid);
94 if (n < cpu_key->on_disk_key.k_objectid)
96 if (n > cpu_key->on_disk_key.k_objectid)
101 /* k1 is pointer to on-disk structure which is stored in little-endian
102 form. k2 is pointer to cpu variable.
103 Compare keys using all 4 key fields.
104 Returns: -1 if key1 < key2 0
105 if key1 = key2 1 if key1 > key2 */
106 static inline int comp_keys(const struct reiserfs_key *le_key,
107 const struct cpu_key *cpu_key)
111 retval = comp_short_keys(le_key, cpu_key);
114 if (le_key_k_offset(le_key_version(le_key), le_key) <
115 cpu_key_k_offset(cpu_key))
117 if (le_key_k_offset(le_key_version(le_key), le_key) >
118 cpu_key_k_offset(cpu_key))
121 if (cpu_key->key_length == 3)
124 /* this part is needed only when tail conversion is in progress */
125 if (le_key_k_type(le_key_version(le_key), le_key) <
126 cpu_key_k_type(cpu_key))
129 if (le_key_k_type(le_key_version(le_key), le_key) >
130 cpu_key_k_type(cpu_key))
136 inline int comp_short_le_keys(const struct reiserfs_key *key1,
137 const struct reiserfs_key *key2)
139 __u32 *p_s_1_u32, *p_s_2_u32;
140 int n_key_length = REISERFS_SHORT_KEY_LEN;
142 p_s_1_u32 = (__u32 *) key1;
143 p_s_2_u32 = (__u32 *) key2;
144 for (; n_key_length--; ++p_s_1_u32, ++p_s_2_u32) {
145 if (le32_to_cpu(*p_s_1_u32) < le32_to_cpu(*p_s_2_u32))
147 if (le32_to_cpu(*p_s_1_u32) > le32_to_cpu(*p_s_2_u32))
153 inline void le_key2cpu_key(struct cpu_key *to, const struct reiserfs_key *from)
156 to->on_disk_key.k_dir_id = le32_to_cpu(from->k_dir_id);
157 to->on_disk_key.k_objectid = le32_to_cpu(from->k_objectid);
159 // find out version of the key
160 version = le_key_version(from);
161 to->version = version;
162 to->on_disk_key.k_offset = le_key_k_offset(version, from);
163 to->on_disk_key.k_type = le_key_k_type(version, from);
166 // this does not say which one is bigger, it only returns 1 if keys
167 // are not equal, 0 otherwise
168 inline int comp_le_keys(const struct reiserfs_key *k1,
169 const struct reiserfs_key *k2)
171 return memcmp(k1, k2, sizeof(struct reiserfs_key));
174 /**************************************************************************
175 * Binary search toolkit function *
176 * Search for an item in the array by the item key *
177 * Returns: 1 if found, 0 if not found; *
178 * *p_n_pos = number of the searched element if found, else the *
179 * number of the first element that is larger than p_v_key. *
180 **************************************************************************/
181 /* For those not familiar with binary search: n_lbound is the leftmost item that it
182 could be, n_rbound the rightmost item that it could be. We examine the item
183 halfway between n_lbound and n_rbound, and that tells us either that we can increase
184 n_lbound, or decrease n_rbound, or that we have found it, or if n_lbound <= n_rbound that
185 there are no possible items, and we have not found it. With each examination we
186 cut the number of possible items it could be by one more than half rounded down,
188 static inline int bin_search(const void *p_v_key, /* Key to search for. */
189 const void *p_v_base, /* First item in the array. */
190 int p_n_num, /* Number of items in the array. */
191 int p_n_width, /* Item size in the array.
192 searched. Lest the reader be
193 confused, note that this is crafted
194 as a general function, and when it
195 is applied specifically to the array
196 of item headers in a node, p_n_width
197 is actually the item header size not
199 int *p_n_pos /* Number of the searched for element. */
202 int n_rbound, n_lbound, n_j;
204 for (n_j = ((n_rbound = p_n_num - 1) + (n_lbound = 0)) / 2;
205 n_lbound <= n_rbound; n_j = (n_rbound + n_lbound) / 2)
207 ((struct reiserfs_key *)((char *)p_v_base +
209 (struct cpu_key *)p_v_key)) {
218 return ITEM_FOUND; /* Key found in the array. */
221 /* bin_search did not find given key, it returns position of key,
222 that is minimal and greater than the given one. */
224 return ITEM_NOT_FOUND;
227 #ifdef CONFIG_REISERFS_CHECK
228 extern struct tree_balance *cur_tb;
231 /* Minimal possible key. It is never in the tree. */
232 const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} };
234 /* Maximal possible key. It is never in the tree. */
235 static const struct reiserfs_key MAX_KEY = {
236 __constant_cpu_to_le32(0xffffffff),
237 __constant_cpu_to_le32(0xffffffff),
238 {{__constant_cpu_to_le32(0xffffffff),
239 __constant_cpu_to_le32(0xffffffff)},}
242 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
243 of the path, and going upwards. We must check the path's validity at each step. If the key is not in
244 the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
245 case we return a special key, either MIN_KEY or MAX_KEY. */
246 static inline const struct reiserfs_key *get_lkey(const struct treepath
248 const struct super_block
251 int n_position, n_path_offset = p_s_chk_path->path_length;
252 struct buffer_head *p_s_parent;
254 RFALSE(n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
255 "PAP-5010: invalid offset in the path");
257 /* While not higher in path than first element. */
258 while (n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
260 RFALSE(!buffer_uptodate
261 (PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
262 "PAP-5020: parent is not uptodate");
264 /* Parent at the path is not in the tree now. */
267 PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)))
269 /* Check whether position in the parent is correct. */
271 PATH_OFFSET_POSITION(p_s_chk_path,
273 B_NR_ITEMS(p_s_parent))
275 /* Check whether parent at the path really points to the child. */
276 if (B_N_CHILD_NUM(p_s_parent, n_position) !=
277 PATH_OFFSET_PBUFFER(p_s_chk_path,
278 n_path_offset + 1)->b_blocknr)
280 /* Return delimiting key if position in the parent is not equal to zero. */
282 return B_N_PDELIM_KEY(p_s_parent, n_position - 1);
284 /* Return MIN_KEY if we are in the root of the buffer tree. */
285 if (PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->
286 b_blocknr == SB_ROOT_BLOCK(p_s_sb))
291 /* Get delimiting key of the buffer at the path and its right neighbor. */
292 inline const struct reiserfs_key *get_rkey(const struct treepath *p_s_chk_path,
293 const struct super_block *p_s_sb)
295 int n_position, n_path_offset = p_s_chk_path->path_length;
296 struct buffer_head *p_s_parent;
298 RFALSE(n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
299 "PAP-5030: invalid offset in the path");
301 while (n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
303 RFALSE(!buffer_uptodate
304 (PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
305 "PAP-5040: parent is not uptodate");
307 /* Parent at the path is not in the tree now. */
310 PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)))
312 /* Check whether position in the parent is correct. */
314 PATH_OFFSET_POSITION(p_s_chk_path,
316 B_NR_ITEMS(p_s_parent))
318 /* Check whether parent at the path really points to the child. */
319 if (B_N_CHILD_NUM(p_s_parent, n_position) !=
320 PATH_OFFSET_PBUFFER(p_s_chk_path,
321 n_path_offset + 1)->b_blocknr)
323 /* Return delimiting key if position in the parent is not the last one. */
324 if (n_position != B_NR_ITEMS(p_s_parent))
325 return B_N_PDELIM_KEY(p_s_parent, n_position);
327 /* Return MAX_KEY if we are in the root of the buffer tree. */
328 if (PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->
329 b_blocknr == SB_ROOT_BLOCK(p_s_sb))
334 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
335 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
336 the path. These delimiting keys are stored at least one level above that buffer in the tree. If the
337 buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
338 this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
339 static inline int key_in_buffer(struct treepath *p_s_chk_path, /* Path which should be checked. */
340 const struct cpu_key *p_s_key, /* Key which should be checked. */
341 struct super_block *p_s_sb /* Super block pointer. */
345 RFALSE(!p_s_key || p_s_chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET
346 || p_s_chk_path->path_length > MAX_HEIGHT,
347 "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
348 p_s_key, p_s_chk_path->path_length);
349 RFALSE(!PATH_PLAST_BUFFER(p_s_chk_path)->b_bdev,
350 "PAP-5060: device must not be NODEV");
352 if (comp_keys(get_lkey(p_s_chk_path, p_s_sb), p_s_key) == 1)
353 /* left delimiting key is bigger, that the key we look for */
355 // if ( comp_keys(p_s_key, get_rkey(p_s_chk_path, p_s_sb)) != -1 )
356 if (comp_keys(get_rkey(p_s_chk_path, p_s_sb), p_s_key) != 1)
357 /* p_s_key must be less than right delimitiing key */
362 inline void decrement_bcount(struct buffer_head *p_s_bh)
365 if (atomic_read(&(p_s_bh->b_count))) {
369 reiserfs_panic(NULL, "PAP-5070",
370 "trying to free free buffer %b", p_s_bh);
374 /* Decrement b_count field of the all buffers in the path. */
375 void decrement_counters_in_path(struct treepath *p_s_search_path)
377 int n_path_offset = p_s_search_path->path_length;
379 RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET ||
380 n_path_offset > EXTENDED_MAX_HEIGHT - 1,
381 "PAP-5080: invalid path offset of %d", n_path_offset);
383 while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
384 struct buffer_head *bh;
386 bh = PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--);
387 decrement_bcount(bh);
389 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
392 int reiserfs_check_path(struct treepath *p)
394 RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
395 "path not properly relsed");
399 /* Release all buffers in the path. Restore dirty bits clean
400 ** when preparing the buffer for the log
402 ** only called from fix_nodes()
404 void pathrelse_and_restore(struct super_block *s, struct treepath *p_s_search_path)
406 int n_path_offset = p_s_search_path->path_length;
408 RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
409 "clm-4000: invalid path offset");
411 while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
412 reiserfs_restore_prepared_buffer(s,
416 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
418 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
421 /* Release all buffers in the path. */
422 void pathrelse(struct treepath *p_s_search_path)
424 int n_path_offset = p_s_search_path->path_length;
426 RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
427 "PAP-5090: invalid path offset");
429 while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
430 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
432 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
435 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
437 struct block_head *blkh;
438 struct item_head *ih;
444 blkh = (struct block_head *)buf;
445 if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
446 reiserfs_warning(NULL, "reiserfs-5080",
447 "this should be caught earlier");
451 nr = blkh_nr_item(blkh);
452 if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
453 /* item number is too big or too small */
454 reiserfs_warning(NULL, "reiserfs-5081",
455 "nr_item seems wrong: %z", bh);
458 ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
459 used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih));
460 if (used_space != blocksize - blkh_free_space(blkh)) {
461 /* free space does not match to calculated amount of use space */
462 reiserfs_warning(NULL, "reiserfs-5082",
463 "free space seems wrong: %z", bh);
466 // FIXME: it is_leaf will hit performance too much - we may have
469 /* check tables of item heads */
470 ih = (struct item_head *)(buf + BLKH_SIZE);
471 prev_location = blocksize;
472 for (i = 0; i < nr; i++, ih++) {
473 if (le_ih_k_type(ih) == TYPE_ANY) {
474 reiserfs_warning(NULL, "reiserfs-5083",
475 "wrong item type for item %h",
479 if (ih_location(ih) >= blocksize
480 || ih_location(ih) < IH_SIZE * nr) {
481 reiserfs_warning(NULL, "reiserfs-5084",
482 "item location seems wrong: %h",
486 if (ih_item_len(ih) < 1
487 || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
488 reiserfs_warning(NULL, "reiserfs-5085",
489 "item length seems wrong: %h",
493 if (prev_location - ih_location(ih) != ih_item_len(ih)) {
494 reiserfs_warning(NULL, "reiserfs-5086",
495 "item location seems wrong "
496 "(second one): %h", ih);
499 prev_location = ih_location(ih);
502 // one may imagine much more checks
506 /* returns 1 if buf looks like an internal node, 0 otherwise */
507 static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
509 struct block_head *blkh;
513 blkh = (struct block_head *)buf;
514 nr = blkh_level(blkh);
515 if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
516 /* this level is not possible for internal nodes */
517 reiserfs_warning(NULL, "reiserfs-5087",
518 "this should be caught earlier");
522 nr = blkh_nr_item(blkh);
523 if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
524 /* for internal which is not root we might check min number of keys */
525 reiserfs_warning(NULL, "reiserfs-5088",
526 "number of key seems wrong: %z", bh);
530 used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
531 if (used_space != blocksize - blkh_free_space(blkh)) {
532 reiserfs_warning(NULL, "reiserfs-5089",
533 "free space seems wrong: %z", bh);
536 // one may imagine much more checks
540 // make sure that bh contains formatted node of reiserfs tree of
542 static int is_tree_node(struct buffer_head *bh, int level)
544 if (B_LEVEL(bh) != level) {
545 reiserfs_warning(NULL, "reiserfs-5090", "node level %d does "
546 "not match to the expected one %d",
550 if (level == DISK_LEAF_NODE_LEVEL)
551 return is_leaf(bh->b_data, bh->b_size, bh);
553 return is_internal(bh->b_data, bh->b_size, bh);
556 #define SEARCH_BY_KEY_READA 16
558 /* The function is NOT SCHEDULE-SAFE! */
559 static void search_by_key_reada(struct super_block *s,
560 struct buffer_head **bh,
561 b_blocknr_t *b, int num)
565 for (i = 0; i < num; i++) {
566 bh[i] = sb_getblk(s, b[i]);
568 for (j = 0; j < i; j++) {
570 * note, this needs attention if we are getting rid of the BKL
571 * you have to make sure the prepared bit isn't set on this buffer
573 if (!buffer_uptodate(bh[j]))
574 ll_rw_block(READA, 1, bh + j);
579 /**************************************************************************
580 * Algorithm SearchByKey *
581 * look for item in the Disk S+Tree by its key *
582 * Input: p_s_sb - super block *
583 * p_s_key - pointer to the key to search *
584 * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR *
585 * p_s_search_path - path from the root to the needed leaf *
586 **************************************************************************/
588 /* This function fills up the path from the root to the leaf as it
589 descends the tree looking for the key. It uses reiserfs_bread to
590 try to find buffers in the cache given their block number. If it
591 does not find them in the cache it reads them from disk. For each
592 node search_by_key finds using reiserfs_bread it then uses
593 bin_search to look through that node. bin_search will find the
594 position of the block_number of the next node if it is looking
595 through an internal node. If it is looking through a leaf node
596 bin_search will find the position of the item which has key either
597 equal to given key, or which is the maximal key less than the given
598 key. search_by_key returns a path that must be checked for the
599 correctness of the top of the path but need not be checked for the
600 correctness of the bottom of the path */
601 /* The function is NOT SCHEDULE-SAFE! */
602 int search_by_key(struct super_block *p_s_sb, const struct cpu_key *p_s_key, /* Key to search. */
603 struct treepath *p_s_search_path,/* This structure was
604 allocated and initialized
606 function. It is filled up
608 int n_stop_level /* How far down the tree to search. To
609 stop at leaf level - set to
610 DISK_LEAF_NODE_LEVEL */
613 b_blocknr_t n_block_number;
615 struct buffer_head *p_s_bh;
616 struct path_element *p_s_last_element;
617 int n_node_level, n_retval;
618 int right_neighbor_of_leaf_node;
620 struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
621 b_blocknr_t reada_blocks[SEARCH_BY_KEY_READA];
624 #ifdef CONFIG_REISERFS_CHECK
625 int n_repeat_counter = 0;
628 PROC_INFO_INC(p_s_sb, search_by_key);
630 /* As we add each node to a path we increase its count. This means that
631 we must be careful to release all nodes in a path before we either
632 discard the path struct or re-use the path struct, as we do here. */
634 decrement_counters_in_path(p_s_search_path);
636 right_neighbor_of_leaf_node = 0;
638 /* With each iteration of this loop we search through the items in the
639 current node, and calculate the next current node(next path element)
640 for the next iteration of this loop.. */
641 n_block_number = SB_ROOT_BLOCK(p_s_sb);
645 #ifdef CONFIG_REISERFS_CHECK
646 if (!(++n_repeat_counter % 50000))
647 reiserfs_warning(p_s_sb, "PAP-5100",
648 "%s: there were %d iterations of "
649 "while loop looking for key %K",
650 current->comm, n_repeat_counter,
654 /* prep path to have another element added to it. */
656 PATH_OFFSET_PELEMENT(p_s_search_path,
657 ++p_s_search_path->path_length);
658 fs_gen = get_generation(p_s_sb);
660 /* Read the next tree node, and set the last element in the path to
661 have a pointer to it. */
662 if ((p_s_bh = p_s_last_element->pe_buffer =
663 sb_getblk(p_s_sb, n_block_number))) {
664 if (!buffer_uptodate(p_s_bh) && reada_count > 1) {
665 search_by_key_reada(p_s_sb, reada_bh,
666 reada_blocks, reada_count);
668 ll_rw_block(READ, 1, &p_s_bh);
669 wait_on_buffer(p_s_bh);
670 if (!buffer_uptodate(p_s_bh))
674 p_s_search_path->path_length--;
675 pathrelse(p_s_search_path);
679 if (expected_level == -1)
680 expected_level = SB_TREE_HEIGHT(p_s_sb);
683 /* It is possible that schedule occurred. We must check whether the key
684 to search is still in the tree rooted from the current buffer. If
685 not then repeat search from the root. */
686 if (fs_changed(fs_gen, p_s_sb) &&
687 (!B_IS_IN_TREE(p_s_bh) ||
688 B_LEVEL(p_s_bh) != expected_level ||
689 !key_in_buffer(p_s_search_path, p_s_key, p_s_sb))) {
690 PROC_INFO_INC(p_s_sb, search_by_key_fs_changed);
691 PROC_INFO_INC(p_s_sb, search_by_key_restarted);
692 PROC_INFO_INC(p_s_sb,
693 sbk_restarted[expected_level - 1]);
694 decrement_counters_in_path(p_s_search_path);
696 /* Get the root block number so that we can repeat the search
697 starting from the root. */
698 n_block_number = SB_ROOT_BLOCK(p_s_sb);
700 right_neighbor_of_leaf_node = 0;
702 /* repeat search from the root */
706 /* only check that the key is in the buffer if p_s_key is not
707 equal to the MAX_KEY. Latter case is only possible in
708 "finish_unfinished()" processing during mount. */
709 RFALSE(comp_keys(&MAX_KEY, p_s_key) &&
710 !key_in_buffer(p_s_search_path, p_s_key, p_s_sb),
711 "PAP-5130: key is not in the buffer");
712 #ifdef CONFIG_REISERFS_CHECK
714 print_cur_tb("5140");
715 reiserfs_panic(p_s_sb, "PAP-5140",
716 "schedule occurred in do_balance!");
720 // make sure, that the node contents look like a node of
722 if (!is_tree_node(p_s_bh, expected_level)) {
723 reiserfs_error(p_s_sb, "vs-5150",
724 "invalid format found in block %ld. "
725 "Fsck?", p_s_bh->b_blocknr);
726 pathrelse(p_s_search_path);
730 /* ok, we have acquired next formatted node in the tree */
731 n_node_level = B_LEVEL(p_s_bh);
733 PROC_INFO_BH_STAT(p_s_sb, p_s_bh, n_node_level - 1);
735 RFALSE(n_node_level < n_stop_level,
736 "vs-5152: tree level (%d) is less than stop level (%d)",
737 n_node_level, n_stop_level);
739 n_retval = bin_search(p_s_key, B_N_PITEM_HEAD(p_s_bh, 0),
742 DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
744 &(p_s_last_element->pe_position));
745 if (n_node_level == n_stop_level) {
749 /* we are not in the stop level */
750 if (n_retval == ITEM_FOUND)
751 /* item has been found, so we choose the pointer which is to the right of the found one */
752 p_s_last_element->pe_position++;
754 /* if item was not found we choose the position which is to
755 the left of the found item. This requires no code,
756 bin_search did it already. */
758 /* So we have chosen a position in the current node which is
759 an internal node. Now we calculate child block number by
760 position in the node. */
762 B_N_CHILD_NUM(p_s_bh, p_s_last_element->pe_position);
764 /* if we are going to read leaf nodes, try for read ahead as well */
765 if ((p_s_search_path->reada & PATH_READA) &&
766 n_node_level == DISK_LEAF_NODE_LEVEL + 1) {
767 int pos = p_s_last_element->pe_position;
768 int limit = B_NR_ITEMS(p_s_bh);
769 struct reiserfs_key *le_key;
771 if (p_s_search_path->reada & PATH_READA_BACK)
773 while (reada_count < SEARCH_BY_KEY_READA) {
776 reada_blocks[reada_count++] =
777 B_N_CHILD_NUM(p_s_bh, pos);
778 if (p_s_search_path->reada & PATH_READA_BACK)
784 * check to make sure we're in the same object
786 le_key = B_N_PDELIM_KEY(p_s_bh, pos);
787 if (le32_to_cpu(le_key->k_objectid) !=
788 p_s_key->on_disk_key.k_objectid) {
796 /* Form the path to an item and position in this item which contains
797 file byte defined by p_s_key. If there is no such item
798 corresponding to the key, we point the path to the item with
799 maximal key less than p_s_key, and *p_n_pos_in_item is set to one
800 past the last entry/byte in the item. If searching for entry in a
801 directory item, and it is not found, *p_n_pos_in_item is set to one
802 entry more than the entry with maximal key which is less than the
805 Note that if there is no entry in this same node which is one more,
806 then we point to an imaginary entry. for direct items, the
807 position is in units of bytes, for indirect items the position is
808 in units of blocknr entries, for directory items the position is in
809 units of directory entries. */
811 /* The function is NOT SCHEDULE-SAFE! */
812 int search_for_position_by_key(struct super_block *p_s_sb, /* Pointer to the super block. */
813 const struct cpu_key *p_cpu_key, /* Key to search (cpu variable) */
814 struct treepath *p_s_search_path /* Filled up by this function. */
817 struct item_head *p_le_ih; /* pointer to on-disk structure */
819 loff_t item_offset, offset;
820 struct reiserfs_dir_entry de;
823 /* If searching for directory entry. */
824 if (is_direntry_cpu_key(p_cpu_key))
825 return search_by_entry_key(p_s_sb, p_cpu_key, p_s_search_path,
828 /* If not searching for directory entry. */
830 /* If item is found. */
831 retval = search_item(p_s_sb, p_cpu_key, p_s_search_path);
832 if (retval == IO_ERROR)
834 if (retval == ITEM_FOUND) {
838 (PATH_PLAST_BUFFER(p_s_search_path),
839 PATH_LAST_POSITION(p_s_search_path))),
840 "PAP-5165: item length equals zero");
842 pos_in_item(p_s_search_path) = 0;
843 return POSITION_FOUND;
846 RFALSE(!PATH_LAST_POSITION(p_s_search_path),
847 "PAP-5170: position equals zero");
849 /* Item is not found. Set path to the previous item. */
851 B_N_PITEM_HEAD(PATH_PLAST_BUFFER(p_s_search_path),
852 --PATH_LAST_POSITION(p_s_search_path));
853 n_blk_size = p_s_sb->s_blocksize;
855 if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) {
856 return FILE_NOT_FOUND;
858 // FIXME: quite ugly this far
860 item_offset = le_ih_k_offset(p_le_ih);
861 offset = cpu_key_k_offset(p_cpu_key);
863 /* Needed byte is contained in the item pointed to by the path. */
864 if (item_offset <= offset &&
865 item_offset + op_bytes_number(p_le_ih, n_blk_size) > offset) {
866 pos_in_item(p_s_search_path) = offset - item_offset;
867 if (is_indirect_le_ih(p_le_ih)) {
868 pos_in_item(p_s_search_path) /= n_blk_size;
870 return POSITION_FOUND;
873 /* Needed byte is not contained in the item pointed to by the
874 path. Set pos_in_item out of the item. */
875 if (is_indirect_le_ih(p_le_ih))
876 pos_in_item(p_s_search_path) =
877 ih_item_len(p_le_ih) / UNFM_P_SIZE;
879 pos_in_item(p_s_search_path) = ih_item_len(p_le_ih);
881 return POSITION_NOT_FOUND;
884 /* Compare given item and item pointed to by the path. */
885 int comp_items(const struct item_head *stored_ih, const struct treepath *p_s_path)
887 struct buffer_head *p_s_bh;
888 struct item_head *ih;
890 /* Last buffer at the path is not in the tree. */
891 if (!B_IS_IN_TREE(p_s_bh = PATH_PLAST_BUFFER(p_s_path)))
894 /* Last path position is invalid. */
895 if (PATH_LAST_POSITION(p_s_path) >= B_NR_ITEMS(p_s_bh))
898 /* we need only to know, whether it is the same item */
899 ih = get_ih(p_s_path);
900 return memcmp(stored_ih, ih, IH_SIZE);
903 /* unformatted nodes are not logged anymore, ever. This is safe
906 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
908 // block can not be forgotten as it is in I/O or held by someone
909 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
911 // prepare for delete or cut of direct item
912 static inline int prepare_for_direct_item(struct treepath *path,
913 struct item_head *le_ih,
915 loff_t new_file_length, int *cut_size)
919 if (new_file_length == max_reiserfs_offset(inode)) {
920 /* item has to be deleted */
921 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
924 // new file gets truncated
925 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
927 round_len = ROUND_UP(new_file_length);
928 /* this was n_new_file_length < le_ih ... */
929 if (round_len < le_ih_k_offset(le_ih)) {
930 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
931 return M_DELETE; /* Delete this item. */
933 /* Calculate first position and size for cutting from item. */
934 pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
935 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
937 return M_CUT; /* Cut from this item. */
940 // old file: items may have any length
942 if (new_file_length < le_ih_k_offset(le_ih)) {
943 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
944 return M_DELETE; /* Delete this item. */
946 /* Calculate first position and size for cutting from item. */
947 *cut_size = -(ih_item_len(le_ih) -
949 new_file_length + 1 - le_ih_k_offset(le_ih)));
950 return M_CUT; /* Cut from this item. */
953 static inline int prepare_for_direntry_item(struct treepath *path,
954 struct item_head *le_ih,
956 loff_t new_file_length,
959 if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
960 new_file_length == max_reiserfs_offset(inode)) {
961 RFALSE(ih_entry_count(le_ih) != 2,
962 "PAP-5220: incorrect empty directory item (%h)", le_ih);
963 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
964 return M_DELETE; /* Delete the directory item containing "." and ".." entry. */
967 if (ih_entry_count(le_ih) == 1) {
968 /* Delete the directory item such as there is one record only
970 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
974 /* Cut one record from the directory item. */
977 entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
981 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
983 /* If the path points to a directory or direct item, calculate mode and the size cut, for balance.
984 If the path points to an indirect item, remove some number of its unformatted nodes.
985 In case of file truncate calculate whether this item must be deleted/truncated or last
986 unformatted node of this item will be converted to a direct item.
987 This function returns a determination of what balance mode the calling function should employ. */
988 static char prepare_for_delete_or_cut(struct reiserfs_transaction_handle *th, struct inode *inode, struct treepath *p_s_path, const struct cpu_key *p_s_item_key, int *p_n_removed, /* Number of unformatted nodes which were removed
989 from end of the file. */
990 int *p_n_cut_size, unsigned long long n_new_file_length /* MAX_KEY_OFFSET in case of delete. */
993 struct super_block *p_s_sb = inode->i_sb;
994 struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_path);
995 struct buffer_head *p_s_bh = PATH_PLAST_BUFFER(p_s_path);
997 BUG_ON(!th->t_trans_id);
999 /* Stat_data item. */
1000 if (is_statdata_le_ih(p_le_ih)) {
1002 RFALSE(n_new_file_length != max_reiserfs_offset(inode),
1003 "PAP-5210: mode must be M_DELETE");
1005 *p_n_cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
1009 /* Directory item. */
1010 if (is_direntry_le_ih(p_le_ih))
1011 return prepare_for_direntry_item(p_s_path, p_le_ih, inode,
1016 if (is_direct_le_ih(p_le_ih))
1017 return prepare_for_direct_item(p_s_path, p_le_ih, inode,
1018 n_new_file_length, p_n_cut_size);
1020 /* Case of an indirect item. */
1022 int blk_size = p_s_sb->s_blocksize;
1023 struct item_head s_ih;
1029 if ( n_new_file_length == max_reiserfs_offset (inode) ) {
1030 /* prepare_for_delete_or_cut() is called by
1031 * reiserfs_delete_item() */
1032 n_new_file_length = 0;
1039 p_s_bh = PATH_PLAST_BUFFER(p_s_path);
1040 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1041 pos = I_UNFM_NUM(&s_ih);
1043 while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > n_new_file_length) {
1047 /* Each unformatted block deletion may involve one additional
1048 * bitmap block into the transaction, thereby the initial
1049 * journal space reservation might not be enough. */
1050 if (!delete && (*p_n_cut_size) != 0 &&
1051 reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1055 unfm = (__le32 *)B_I_PITEM(p_s_bh, &s_ih) + pos - 1;
1056 block = get_block_num(unfm, 0);
1059 reiserfs_prepare_for_journal(p_s_sb, p_s_bh, 1);
1060 put_block_num(unfm, 0, 0);
1061 journal_mark_dirty (th, p_s_sb, p_s_bh);
1062 reiserfs_free_block(th, inode, block, 1);
1067 if (item_moved (&s_ih, p_s_path)) {
1074 (*p_n_cut_size) -= UNFM_P_SIZE;
1077 (*p_n_cut_size) -= IH_SIZE;
1082 /* a trick. If the buffer has been logged, this will do nothing. If
1083 ** we've broken the loop without logging it, it will restore the
1085 reiserfs_restore_prepared_buffer(p_s_sb, p_s_bh);
1086 } while (need_re_search &&
1087 search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path) == POSITION_FOUND);
1088 pos_in_item(p_s_path) = pos * UNFM_P_SIZE;
1090 if (*p_n_cut_size == 0) {
1091 /* Nothing were cut. maybe convert last unformatted node to the
1099 /* Calculate number of bytes which will be deleted or cut during balance */
1100 static int calc_deleted_bytes_number(struct tree_balance *p_s_tb, char c_mode)
1103 struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_tb->tb_path);
1105 if (is_statdata_le_ih(p_le_ih))
1110 M_DELETE) ? ih_item_len(p_le_ih) : -p_s_tb->insert_size[0];
1111 if (is_direntry_le_ih(p_le_ih)) {
1112 // return EMPTY_DIR_SIZE; /* We delete emty directoris only. */
1113 // we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1114 // empty size. ick. FIXME, is this right?
1119 if (is_indirect_le_ih(p_le_ih))
1120 n_del_size = (n_del_size / UNFM_P_SIZE) * (PATH_PLAST_BUFFER(p_s_tb->tb_path)->b_size); // - get_ih_free_space (p_le_ih);
1124 static void init_tb_struct(struct reiserfs_transaction_handle *th,
1125 struct tree_balance *p_s_tb,
1126 struct super_block *p_s_sb,
1127 struct treepath *p_s_path, int n_size)
1130 BUG_ON(!th->t_trans_id);
1132 memset(p_s_tb, '\0', sizeof(struct tree_balance));
1133 p_s_tb->transaction_handle = th;
1134 p_s_tb->tb_sb = p_s_sb;
1135 p_s_tb->tb_path = p_s_path;
1136 PATH_OFFSET_PBUFFER(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1137 PATH_OFFSET_POSITION(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1138 p_s_tb->insert_size[0] = n_size;
1141 void padd_item(char *item, int total_length, int length)
1145 for (i = total_length; i > length;)
1149 #ifdef REISERQUOTA_DEBUG
1150 char key2type(struct reiserfs_key *ih)
1152 if (is_direntry_le_key(2, ih))
1154 if (is_direct_le_key(2, ih))
1156 if (is_indirect_le_key(2, ih))
1158 if (is_statdata_le_key(2, ih))
1163 char head2type(struct item_head *ih)
1165 if (is_direntry_le_ih(ih))
1167 if (is_direct_le_ih(ih))
1169 if (is_indirect_le_ih(ih))
1171 if (is_statdata_le_ih(ih))
1177 /* Delete object item. */
1178 int reiserfs_delete_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_path, /* Path to the deleted item. */
1179 const struct cpu_key *p_s_item_key, /* Key to search for the deleted item. */
1180 struct inode *p_s_inode, /* inode is here just to update i_blocks and quotas */
1181 struct buffer_head *p_s_un_bh)
1182 { /* NULL or unformatted node pointer. */
1183 struct super_block *p_s_sb = p_s_inode->i_sb;
1184 struct tree_balance s_del_balance;
1185 struct item_head s_ih;
1186 struct item_head *q_ih;
1187 int quota_cut_bytes;
1188 int n_ret_value, n_del_size, n_removed;
1190 #ifdef CONFIG_REISERFS_CHECK
1195 BUG_ON(!th->t_trans_id);
1197 init_tb_struct(th, &s_del_balance, p_s_sb, p_s_path,
1198 0 /*size is unknown */ );
1203 #ifdef CONFIG_REISERFS_CHECK
1207 prepare_for_delete_or_cut(th, p_s_inode, p_s_path,
1208 p_s_item_key, &n_removed,
1210 max_reiserfs_offset(p_s_inode));
1212 RFALSE(c_mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1214 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1215 s_del_balance.insert_size[0] = n_del_size;
1217 n_ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1218 if (n_ret_value != REPEAT_SEARCH)
1221 PROC_INFO_INC(p_s_sb, delete_item_restarted);
1223 // file system changed, repeat search
1225 search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1226 if (n_ret_value == IO_ERROR)
1228 if (n_ret_value == FILE_NOT_FOUND) {
1229 reiserfs_warning(p_s_sb, "vs-5340",
1230 "no items of the file %K found",
1236 if (n_ret_value != CARRY_ON) {
1237 unfix_nodes(&s_del_balance);
1240 // reiserfs_delete_item returns item length when success
1241 n_ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1242 q_ih = get_ih(p_s_path);
1243 quota_cut_bytes = ih_item_len(q_ih);
1245 /* hack so the quota code doesn't have to guess if the file
1246 ** has a tail. On tail insert, we allocate quota for 1 unformatted node.
1247 ** We test the offset because the tail might have been
1248 ** split into multiple items, and we only want to decrement for
1249 ** the unfm node once
1251 if (!S_ISLNK(p_s_inode->i_mode) && is_direct_le_ih(q_ih)) {
1252 if ((le_ih_k_offset(q_ih) & (p_s_sb->s_blocksize - 1)) == 1) {
1253 quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1255 quota_cut_bytes = 0;
1263 /* We are in direct2indirect conversion, so move tail contents
1264 to the unformatted node */
1265 /* note, we do the copy before preparing the buffer because we
1266 ** don't care about the contents of the unformatted node yet.
1267 ** the only thing we really care about is the direct item's data
1268 ** is in the unformatted node.
1270 ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1271 ** the unformatted node, which might schedule, meaning we'd have to
1272 ** loop all the way back up to the start of the while loop.
1274 ** The unformatted node must be dirtied later on. We can't be
1275 ** sure here if the entire tail has been deleted yet.
1277 ** p_s_un_bh is from the page cache (all unformatted nodes are
1278 ** from the page cache) and might be a highmem page. So, we
1279 ** can't use p_s_un_bh->b_data.
1283 data = kmap_atomic(p_s_un_bh->b_page, KM_USER0);
1284 off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1286 B_I_PITEM(PATH_PLAST_BUFFER(p_s_path), &s_ih),
1288 kunmap_atomic(data, KM_USER0);
1290 /* Perform balancing after all resources have been collected at once. */
1291 do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1293 #ifdef REISERQUOTA_DEBUG
1294 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1295 "reiserquota delete_item(): freeing %u, id=%u type=%c",
1296 quota_cut_bytes, p_s_inode->i_uid, head2type(&s_ih));
1298 DQUOT_FREE_SPACE_NODIRTY(p_s_inode, quota_cut_bytes);
1300 /* Return deleted body length */
1304 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1306 deletion of the body of the object is performed by iput(), with the
1307 result that if multiple processes are operating on a file, the
1308 deletion of the body of the file is deferred until the last process
1309 that has an open inode performs its iput().
1311 writes and truncates are protected from collisions by use of
1314 creates, linking, and mknod are protected from collisions with other
1315 processes by making the reiserfs_add_entry() the last step in the
1316 creation, and then rolling back all changes if there was a collision.
1320 /* this deletes item which never gets split */
1321 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1322 struct inode *inode, struct reiserfs_key *key)
1324 struct tree_balance tb;
1325 INITIALIZE_PATH(path);
1328 struct cpu_key cpu_key;
1330 int quota_cut_bytes = 0;
1332 BUG_ON(!th->t_trans_id);
1334 le_key2cpu_key(&cpu_key, key);
1337 retval = search_item(th->t_super, &cpu_key, &path);
1338 if (retval == IO_ERROR) {
1339 reiserfs_error(th->t_super, "vs-5350",
1340 "i/o failure occurred trying "
1341 "to delete %K", &cpu_key);
1344 if (retval != ITEM_FOUND) {
1346 // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1348 ((unsigned long long)
1349 GET_HASH_VALUE(le_key_k_offset
1350 (le_key_version(key), key)) == 0
1351 && (unsigned long long)
1352 GET_GENERATION_NUMBER(le_key_k_offset
1353 (le_key_version(key),
1355 reiserfs_warning(th->t_super, "vs-5355",
1356 "%k not found", key);
1361 item_len = ih_item_len(PATH_PITEM_HEAD(&path));
1362 init_tb_struct(th, &tb, th->t_super, &path,
1363 -(IH_SIZE + item_len));
1365 quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path));
1367 retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1368 if (retval == REPEAT_SEARCH) {
1369 PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1373 if (retval == CARRY_ON) {
1374 do_balance(&tb, NULL, NULL, M_DELETE);
1375 if (inode) { /* Should we count quota for item? (we don't count quotas for save-links) */
1376 #ifdef REISERQUOTA_DEBUG
1377 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1378 "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1379 quota_cut_bytes, inode->i_uid,
1382 DQUOT_FREE_SPACE_NODIRTY(inode,
1387 // IO_ERROR, NO_DISK_SPACE, etc
1388 reiserfs_warning(th->t_super, "vs-5360",
1389 "could not delete %K due to fix_nodes failure",
1395 reiserfs_check_path(&path);
1398 int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1399 struct inode *inode)
1403 BUG_ON(!th->t_trans_id);
1405 /* for directory this deletes item containing "." and ".." */
1407 reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1411 #if defined( USE_INODE_GENERATION_COUNTER )
1412 if (!old_format_only(th->t_super)) {
1413 __le32 *inode_generation;
1416 &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
1417 le32_add_cpu(inode_generation, 1);
1419 /* USE_INODE_GENERATION_COUNTER */
1421 reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1426 static void unmap_buffers(struct page *page, loff_t pos)
1428 struct buffer_head *bh;
1429 struct buffer_head *head;
1430 struct buffer_head *next;
1431 unsigned long tail_index;
1432 unsigned long cur_index;
1435 if (page_has_buffers(page)) {
1436 tail_index = pos & (PAGE_CACHE_SIZE - 1);
1438 head = page_buffers(page);
1441 next = bh->b_this_page;
1443 /* we want to unmap the buffers that contain the tail, and
1444 ** all the buffers after it (since the tail must be at the
1445 ** end of the file). We don't want to unmap file data
1446 ** before the tail, since it might be dirty and waiting to
1449 cur_index += bh->b_size;
1450 if (cur_index > tail_index) {
1451 reiserfs_unmap_buffer(bh);
1454 } while (bh != head);
1459 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
1460 struct inode *p_s_inode,
1462 struct treepath *p_s_path,
1463 const struct cpu_key *p_s_item_key,
1464 loff_t n_new_file_size, char *p_c_mode)
1466 struct super_block *p_s_sb = p_s_inode->i_sb;
1467 int n_block_size = p_s_sb->s_blocksize;
1469 BUG_ON(!th->t_trans_id);
1470 BUG_ON(n_new_file_size != p_s_inode->i_size);
1472 /* the page being sent in could be NULL if there was an i/o error
1473 ** reading in the last block. The user will hit problems trying to
1474 ** read the file, but for now we just skip the indirect2direct
1476 if (atomic_read(&p_s_inode->i_count) > 1 ||
1477 !tail_has_to_be_packed(p_s_inode) ||
1478 !page || (REISERFS_I(p_s_inode)->i_flags & i_nopack_mask)) {
1479 // leave tail in an unformatted node
1480 *p_c_mode = M_SKIP_BALANCING;
1482 n_block_size - (n_new_file_size & (n_block_size - 1));
1483 pathrelse(p_s_path);
1486 /* Permorm the conversion to a direct_item. */
1487 /*return indirect_to_direct (p_s_inode, p_s_path, p_s_item_key, n_new_file_size, p_c_mode); */
1488 return indirect2direct(th, p_s_inode, page, p_s_path, p_s_item_key,
1489 n_new_file_size, p_c_mode);
1492 /* we did indirect_to_direct conversion. And we have inserted direct
1493 item successesfully, but there were no disk space to cut unfm
1494 pointer being converted. Therefore we have to delete inserted
1496 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
1497 struct inode *inode, struct treepath *path)
1499 struct cpu_key tail_key;
1502 BUG_ON(!th->t_trans_id);
1504 make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4); // !!!!
1505 tail_key.key_length = 4;
1508 (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1510 /* look for the last byte of the tail */
1511 if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1513 reiserfs_panic(inode->i_sb, "vs-5615",
1514 "found invalid item");
1515 RFALSE(path->pos_in_item !=
1516 ih_item_len(PATH_PITEM_HEAD(path)) - 1,
1517 "vs-5616: appended bytes found");
1518 PATH_LAST_POSITION(path)--;
1521 reiserfs_delete_item(th, path, &tail_key, inode,
1522 NULL /*unbh not needed */ );
1524 || removed > tail_len,
1525 "vs-5617: there was tail %d bytes, removed item length %d bytes",
1527 tail_len -= removed;
1528 set_cpu_key_k_offset(&tail_key,
1529 cpu_key_k_offset(&tail_key) - removed);
1531 reiserfs_warning(inode->i_sb, "reiserfs-5091", "indirect_to_direct "
1532 "conversion has been rolled back due to "
1533 "lack of disk space");
1534 //mark_file_without_tail (inode);
1535 mark_inode_dirty(inode);
1538 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1539 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
1540 struct treepath *p_s_path,
1541 struct cpu_key *p_s_item_key,
1542 struct inode *p_s_inode,
1543 struct page *page, loff_t n_new_file_size)
1545 struct super_block *p_s_sb = p_s_inode->i_sb;
1546 /* Every function which is going to call do_balance must first
1547 create a tree_balance structure. Then it must fill up this
1548 structure by using the init_tb_struct and fix_nodes functions.
1549 After that we can make tree balancing. */
1550 struct tree_balance s_cut_balance;
1551 struct item_head *p_le_ih;
1552 int n_cut_size = 0, /* Amount to be cut. */
1553 n_ret_value = CARRY_ON, n_removed = 0, /* Number of the removed unformatted nodes. */
1554 n_is_inode_locked = 0;
1555 char c_mode; /* Mode of the balance. */
1557 int quota_cut_bytes;
1558 loff_t tail_pos = 0;
1560 BUG_ON(!th->t_trans_id);
1562 init_tb_struct(th, &s_cut_balance, p_s_inode->i_sb, p_s_path,
1565 /* Repeat this loop until we either cut the item without needing
1566 to balance, or we fix_nodes without schedule occurring */
1568 /* Determine the balance mode, position of the first byte to
1569 be cut, and size to be cut. In case of the indirect item
1570 free unformatted nodes which are pointed to by the cut
1574 prepare_for_delete_or_cut(th, p_s_inode, p_s_path,
1575 p_s_item_key, &n_removed,
1576 &n_cut_size, n_new_file_size);
1577 if (c_mode == M_CONVERT) {
1578 /* convert last unformatted node to direct item or leave
1579 tail in the unformatted node */
1580 RFALSE(n_ret_value != CARRY_ON,
1581 "PAP-5570: can not convert twice");
1584 maybe_indirect_to_direct(th, p_s_inode, page,
1585 p_s_path, p_s_item_key,
1586 n_new_file_size, &c_mode);
1587 if (c_mode == M_SKIP_BALANCING)
1588 /* tail has been left in the unformatted node */
1591 n_is_inode_locked = 1;
1593 /* removing of last unformatted node will change value we
1594 have to return to truncate. Save it */
1595 retval2 = n_ret_value;
1596 /*retval2 = p_s_sb->s_blocksize - (n_new_file_size & (p_s_sb->s_blocksize - 1)); */
1598 /* So, we have performed the first part of the conversion:
1599 inserting the new direct item. Now we are removing the
1600 last unformatted node pointer. Set key to search for
1602 set_cpu_key_k_type(p_s_item_key, TYPE_INDIRECT);
1603 p_s_item_key->key_length = 4;
1605 (n_new_file_size & (p_s_sb->s_blocksize - 1));
1606 tail_pos = n_new_file_size;
1607 set_cpu_key_k_offset(p_s_item_key, n_new_file_size + 1);
1608 if (search_for_position_by_key
1609 (p_s_sb, p_s_item_key,
1610 p_s_path) == POSITION_NOT_FOUND) {
1611 print_block(PATH_PLAST_BUFFER(p_s_path), 3,
1612 PATH_LAST_POSITION(p_s_path) - 1,
1613 PATH_LAST_POSITION(p_s_path) + 1);
1614 reiserfs_panic(p_s_sb, "PAP-5580", "item to "
1615 "convert does not exist (%K)",
1620 if (n_cut_size == 0) {
1621 pathrelse(p_s_path);
1625 s_cut_balance.insert_size[0] = n_cut_size;
1627 n_ret_value = fix_nodes(c_mode, &s_cut_balance, NULL, NULL);
1628 if (n_ret_value != REPEAT_SEARCH)
1631 PROC_INFO_INC(p_s_sb, cut_from_item_restarted);
1634 search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1635 if (n_ret_value == POSITION_FOUND)
1638 reiserfs_warning(p_s_sb, "PAP-5610", "item %K not found",
1640 unfix_nodes(&s_cut_balance);
1641 return (n_ret_value == IO_ERROR) ? -EIO : -ENOENT;
1644 // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1645 if (n_ret_value != CARRY_ON) {
1646 if (n_is_inode_locked) {
1647 // FIXME: this seems to be not needed: we are always able
1649 indirect_to_direct_roll_back(th, p_s_inode, p_s_path);
1651 if (n_ret_value == NO_DISK_SPACE)
1652 reiserfs_warning(p_s_sb, "reiserfs-5092",
1654 unfix_nodes(&s_cut_balance);
1658 /* go ahead and perform balancing */
1660 RFALSE(c_mode == M_PASTE || c_mode == M_INSERT, "invalid mode");
1662 /* Calculate number of bytes that need to be cut from the item. */
1665 M_DELETE) ? ih_item_len(get_ih(p_s_path)) : -s_cut_balance.
1668 n_ret_value = calc_deleted_bytes_number(&s_cut_balance, c_mode);
1670 n_ret_value = retval2;
1672 /* For direct items, we only change the quota when deleting the last
1675 p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path);
1676 if (!S_ISLNK(p_s_inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1677 if (c_mode == M_DELETE &&
1678 (le_ih_k_offset(p_le_ih) & (p_s_sb->s_blocksize - 1)) ==
1680 // FIXME: this is to keep 3.5 happy
1681 REISERFS_I(p_s_inode)->i_first_direct_byte = U32_MAX;
1682 quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1684 quota_cut_bytes = 0;
1687 #ifdef CONFIG_REISERFS_CHECK
1688 if (n_is_inode_locked) {
1689 struct item_head *le_ih =
1690 PATH_PITEM_HEAD(s_cut_balance.tb_path);
1691 /* we are going to complete indirect2direct conversion. Make
1692 sure, that we exactly remove last unformatted node pointer
1694 if (!is_indirect_le_ih(le_ih))
1695 reiserfs_panic(p_s_sb, "vs-5652",
1696 "item must be indirect %h", le_ih);
1698 if (c_mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1699 reiserfs_panic(p_s_sb, "vs-5653", "completing "
1700 "indirect2direct conversion indirect "
1701 "item %h being deleted must be of "
1702 "4 byte long", le_ih);
1705 && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1706 reiserfs_panic(p_s_sb, "vs-5654", "can not complete "
1707 "indirect2direct conversion of %h "
1708 "(CUT, insert_size==%d)",
1709 le_ih, s_cut_balance.insert_size[0]);
1711 /* it would be useful to make sure, that right neighboring
1712 item is direct item of this file */
1716 do_balance(&s_cut_balance, NULL, NULL, c_mode);
1717 if (n_is_inode_locked) {
1718 /* we've done an indirect->direct conversion. when the data block
1719 ** was freed, it was removed from the list of blocks that must
1720 ** be flushed before the transaction commits, make sure to
1721 ** unmap and invalidate it
1723 unmap_buffers(page, tail_pos);
1724 REISERFS_I(p_s_inode)->i_flags &= ~i_pack_on_close_mask;
1726 #ifdef REISERQUOTA_DEBUG
1727 reiserfs_debug(p_s_inode->i_sb, REISERFS_DEBUG_CODE,
1728 "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1729 quota_cut_bytes, p_s_inode->i_uid, '?');
1731 DQUOT_FREE_SPACE_NODIRTY(p_s_inode, quota_cut_bytes);
1735 static void truncate_directory(struct reiserfs_transaction_handle *th,
1736 struct inode *inode)
1738 BUG_ON(!th->t_trans_id);
1740 reiserfs_error(inode->i_sb, "vs-5655", "link count != 0");
1742 set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1743 set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1744 reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1745 reiserfs_update_sd(th, inode);
1746 set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1747 set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
1750 /* Truncate file to the new size. Note, this must be called with a transaction
1752 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th, struct inode *p_s_inode, /* ->i_size contains new
1754 struct page *page, /* up to date for last block */
1755 int update_timestamps /* when it is called by
1756 file_release to convert
1757 the tail - no timestamps
1758 should be updated */
1761 INITIALIZE_PATH(s_search_path); /* Path to the current object item. */
1762 struct item_head *p_le_ih; /* Pointer to an item header. */
1763 struct cpu_key s_item_key; /* Key to search for a previous file item. */
1764 loff_t n_file_size, /* Old file size. */
1765 n_new_file_size; /* New file size. */
1766 int n_deleted; /* Number of deleted or truncated bytes. */
1770 BUG_ON(!th->t_trans_id);
1772 (S_ISREG(p_s_inode->i_mode) || S_ISDIR(p_s_inode->i_mode)
1773 || S_ISLNK(p_s_inode->i_mode)))
1776 if (S_ISDIR(p_s_inode->i_mode)) {
1777 // deletion of directory - no need to update timestamps
1778 truncate_directory(th, p_s_inode);
1782 /* Get new file size. */
1783 n_new_file_size = p_s_inode->i_size;
1785 // FIXME: note, that key type is unimportant here
1786 make_cpu_key(&s_item_key, p_s_inode, max_reiserfs_offset(p_s_inode),
1790 search_for_position_by_key(p_s_inode->i_sb, &s_item_key,
1792 if (retval == IO_ERROR) {
1793 reiserfs_error(p_s_inode->i_sb, "vs-5657",
1794 "i/o failure occurred trying to truncate %K",
1799 if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1800 reiserfs_error(p_s_inode->i_sb, "PAP-5660",
1801 "wrong result %d of search for %K", retval,
1808 s_search_path.pos_in_item--;
1810 /* Get real file size (total length of all file items) */
1811 p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1812 if (is_statdata_le_ih(p_le_ih))
1815 loff_t offset = le_ih_k_offset(p_le_ih);
1817 op_bytes_number(p_le_ih, p_s_inode->i_sb->s_blocksize);
1819 /* this may mismatch with real file size: if last direct item
1820 had no padding zeros and last unformatted node had no free
1821 space, this file would have this file size */
1822 n_file_size = offset + bytes - 1;
1825 * are we doing a full truncate or delete, if so
1826 * kick in the reada code
1828 if (n_new_file_size == 0)
1829 s_search_path.reada = PATH_READA | PATH_READA_BACK;
1831 if (n_file_size == 0 || n_file_size < n_new_file_size) {
1832 goto update_and_out;
1835 /* Update key to search for the last file item. */
1836 set_cpu_key_k_offset(&s_item_key, n_file_size);
1839 /* Cut or delete file item. */
1841 reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
1842 p_s_inode, page, n_new_file_size);
1843 if (n_deleted < 0) {
1844 reiserfs_warning(p_s_inode->i_sb, "vs-5665",
1845 "reiserfs_cut_from_item failed");
1846 reiserfs_check_path(&s_search_path);
1850 RFALSE(n_deleted > n_file_size,
1851 "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1852 n_deleted, n_file_size, &s_item_key);
1854 /* Change key to search the last file item. */
1855 n_file_size -= n_deleted;
1857 set_cpu_key_k_offset(&s_item_key, n_file_size);
1859 /* While there are bytes to truncate and previous file item is presented in the tree. */
1862 ** This loop could take a really long time, and could log
1863 ** many more blocks than a transaction can hold. So, we do a polite
1864 ** journal end here, and if the transaction needs ending, we make
1865 ** sure the file is consistent before ending the current trans
1866 ** and starting a new one
1868 if (journal_transaction_should_end(th, 0) ||
1869 reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1870 int orig_len_alloc = th->t_blocks_allocated;
1871 decrement_counters_in_path(&s_search_path);
1873 if (update_timestamps) {
1874 p_s_inode->i_mtime = p_s_inode->i_ctime =
1877 reiserfs_update_sd(th, p_s_inode);
1879 err = journal_end(th, p_s_inode->i_sb, orig_len_alloc);
1882 err = journal_begin(th, p_s_inode->i_sb,
1883 JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ;
1886 reiserfs_update_inode_transaction(p_s_inode);
1888 } while (n_file_size > ROUND_UP(n_new_file_size) &&
1889 search_for_position_by_key(p_s_inode->i_sb, &s_item_key,
1890 &s_search_path) == POSITION_FOUND);
1892 RFALSE(n_file_size > ROUND_UP(n_new_file_size),
1893 "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1894 n_new_file_size, n_file_size, s_item_key.on_disk_key.k_objectid);
1897 if (update_timestamps) {
1898 // this is truncate, not file closing
1899 p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME_SEC;
1901 reiserfs_update_sd(th, p_s_inode);
1904 pathrelse(&s_search_path);
1908 #ifdef CONFIG_REISERFS_CHECK
1909 // this makes sure, that we __append__, not overwrite or add holes
1910 static void check_research_for_paste(struct treepath *path,
1911 const struct cpu_key *p_s_key)
1913 struct item_head *found_ih = get_ih(path);
1915 if (is_direct_le_ih(found_ih)) {
1916 if (le_ih_k_offset(found_ih) +
1917 op_bytes_number(found_ih,
1918 get_last_bh(path)->b_size) !=
1919 cpu_key_k_offset(p_s_key)
1920 || op_bytes_number(found_ih,
1921 get_last_bh(path)->b_size) !=
1923 reiserfs_panic(NULL, "PAP-5720", "found direct item "
1924 "%h or position (%d) does not match "
1925 "to key %K", found_ih,
1926 pos_in_item(path), p_s_key);
1928 if (is_indirect_le_ih(found_ih)) {
1929 if (le_ih_k_offset(found_ih) +
1930 op_bytes_number(found_ih,
1931 get_last_bh(path)->b_size) !=
1932 cpu_key_k_offset(p_s_key)
1933 || I_UNFM_NUM(found_ih) != pos_in_item(path)
1934 || get_ih_free_space(found_ih) != 0)
1935 reiserfs_panic(NULL, "PAP-5730", "found indirect "
1936 "item (%h) or position (%d) does not "
1937 "match to key (%K)",
1938 found_ih, pos_in_item(path), p_s_key);
1941 #endif /* config reiserfs check */
1943 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1944 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_search_path, /* Path to the pasted item. */
1945 const struct cpu_key *p_s_key, /* Key to search for the needed item. */
1946 struct inode *inode, /* Inode item belongs to */
1947 const char *p_c_body, /* Pointer to the bytes to paste. */
1949 { /* Size of pasted bytes. */
1950 struct tree_balance s_paste_balance;
1954 BUG_ON(!th->t_trans_id);
1956 fs_gen = get_generation(inode->i_sb);
1958 #ifdef REISERQUOTA_DEBUG
1959 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1960 "reiserquota paste_into_item(): allocating %u id=%u type=%c",
1961 n_pasted_size, inode->i_uid,
1962 key2type(&(p_s_key->on_disk_key)));
1965 if (DQUOT_ALLOC_SPACE_NODIRTY(inode, n_pasted_size)) {
1966 pathrelse(p_s_search_path);
1969 init_tb_struct(th, &s_paste_balance, th->t_super, p_s_search_path,
1971 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1972 s_paste_balance.key = p_s_key->on_disk_key;
1975 /* DQUOT_* can schedule, must check before the fix_nodes */
1976 if (fs_changed(fs_gen, inode->i_sb)) {
1981 fix_nodes(M_PASTE, &s_paste_balance, NULL,
1982 p_c_body)) == REPEAT_SEARCH) {
1984 /* file system changed while we were in the fix_nodes */
1985 PROC_INFO_INC(th->t_super, paste_into_item_restarted);
1987 search_for_position_by_key(th->t_super, p_s_key,
1989 if (retval == IO_ERROR) {
1993 if (retval == POSITION_FOUND) {
1994 reiserfs_warning(inode->i_sb, "PAP-5710",
1995 "entry or pasted byte (%K) exists",
2000 #ifdef CONFIG_REISERFS_CHECK
2001 check_research_for_paste(p_s_search_path, p_s_key);
2005 /* Perform balancing after all resources are collected by fix_nodes, and
2006 accessing them will not risk triggering schedule. */
2007 if (retval == CARRY_ON) {
2008 do_balance(&s_paste_balance, NULL /*ih */ , p_c_body, M_PASTE);
2011 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2013 /* this also releases the path */
2014 unfix_nodes(&s_paste_balance);
2015 #ifdef REISERQUOTA_DEBUG
2016 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2017 "reiserquota paste_into_item(): freeing %u id=%u type=%c",
2018 n_pasted_size, inode->i_uid,
2019 key2type(&(p_s_key->on_disk_key)));
2021 DQUOT_FREE_SPACE_NODIRTY(inode, n_pasted_size);
2025 /* Insert new item into the buffer at the path. */
2026 int reiserfs_insert_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_path, /* Path to the inserteded item. */
2027 const struct cpu_key *key, struct item_head *p_s_ih, /* Pointer to the item header to insert. */
2028 struct inode *inode, const char *p_c_body)
2029 { /* Pointer to the bytes to insert. */
2030 struct tree_balance s_ins_balance;
2033 int quota_bytes = 0;
2035 BUG_ON(!th->t_trans_id);
2037 if (inode) { /* Do we count quotas for item? */
2038 fs_gen = get_generation(inode->i_sb);
2039 quota_bytes = ih_item_len(p_s_ih);
2041 /* hack so the quota code doesn't have to guess if the file has
2042 ** a tail, links are always tails, so there's no guessing needed
2044 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_s_ih)) {
2045 quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
2047 #ifdef REISERQUOTA_DEBUG
2048 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2049 "reiserquota insert_item(): allocating %u id=%u type=%c",
2050 quota_bytes, inode->i_uid, head2type(p_s_ih));
2052 /* We can't dirty inode here. It would be immediately written but
2053 * appropriate stat item isn't inserted yet... */
2054 if (DQUOT_ALLOC_SPACE_NODIRTY(inode, quota_bytes)) {
2055 pathrelse(p_s_path);
2059 init_tb_struct(th, &s_ins_balance, th->t_super, p_s_path,
2060 IH_SIZE + ih_item_len(p_s_ih));
2061 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2062 s_ins_balance.key = key->on_disk_key;
2064 /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2065 if (inode && fs_changed(fs_gen, inode->i_sb)) {
2070 fix_nodes(M_INSERT, &s_ins_balance, p_s_ih,
2071 p_c_body)) == REPEAT_SEARCH) {
2073 /* file system changed while we were in the fix_nodes */
2074 PROC_INFO_INC(th->t_super, insert_item_restarted);
2075 retval = search_item(th->t_super, key, p_s_path);
2076 if (retval == IO_ERROR) {
2080 if (retval == ITEM_FOUND) {
2081 reiserfs_warning(th->t_super, "PAP-5760",
2082 "key %K already exists in the tree",
2089 /* make balancing after all resources will be collected at a time */
2090 if (retval == CARRY_ON) {
2091 do_balance(&s_ins_balance, p_s_ih, p_c_body, M_INSERT);
2095 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2097 /* also releases the path */
2098 unfix_nodes(&s_ins_balance);
2099 #ifdef REISERQUOTA_DEBUG
2100 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2101 "reiserquota insert_item(): freeing %u id=%u type=%c",
2102 quota_bytes, inode->i_uid, head2type(p_s_ih));
2105 DQUOT_FREE_SPACE_NODIRTY(inode, quota_bytes);