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))) {
370 "PAP-5070: decrement_bcount: trying to free free buffer %b",
375 /* Decrement b_count field of the all buffers in the path. */
376 void decrement_counters_in_path(struct treepath *p_s_search_path)
378 int n_path_offset = p_s_search_path->path_length;
380 RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET ||
381 n_path_offset > EXTENDED_MAX_HEIGHT - 1,
382 "PAP-5080: invalid path offset of %d", n_path_offset);
384 while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
385 struct buffer_head *bh;
387 bh = PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--);
388 decrement_bcount(bh);
390 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
393 int reiserfs_check_path(struct treepath *p)
395 RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
396 "path not properly relsed");
400 /* Release all buffers in the path. Restore dirty bits clean
401 ** when preparing the buffer for the log
403 ** only called from fix_nodes()
405 void pathrelse_and_restore(struct super_block *s, struct treepath *p_s_search_path)
407 int n_path_offset = p_s_search_path->path_length;
409 RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
410 "clm-4000: invalid path offset");
412 while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
413 reiserfs_restore_prepared_buffer(s,
417 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
419 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
422 /* Release all buffers in the path. */
423 void pathrelse(struct treepath *p_s_search_path)
425 int n_path_offset = p_s_search_path->path_length;
427 RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
428 "PAP-5090: invalid path offset");
430 while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
431 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
433 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
436 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
438 struct block_head *blkh;
439 struct item_head *ih;
445 blkh = (struct block_head *)buf;
446 if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
447 reiserfs_warning(NULL, "reiserfs-5080",
448 "this should be caught earlier");
452 nr = blkh_nr_item(blkh);
453 if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
454 /* item number is too big or too small */
455 reiserfs_warning(NULL, "reiserfs-5081",
456 "nr_item seems wrong: %z", bh);
459 ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
460 used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih));
461 if (used_space != blocksize - blkh_free_space(blkh)) {
462 /* free space does not match to calculated amount of use space */
463 reiserfs_warning(NULL, "reiserfs-5082",
464 "free space seems wrong: %z", bh);
467 // FIXME: it is_leaf will hit performance too much - we may have
470 /* check tables of item heads */
471 ih = (struct item_head *)(buf + BLKH_SIZE);
472 prev_location = blocksize;
473 for (i = 0; i < nr; i++, ih++) {
474 if (le_ih_k_type(ih) == TYPE_ANY) {
475 reiserfs_warning(NULL, "reiserfs-5083",
476 "wrong item type for item %h",
480 if (ih_location(ih) >= blocksize
481 || ih_location(ih) < IH_SIZE * nr) {
482 reiserfs_warning(NULL, "reiserfs-5084",
483 "item location seems wrong: %h",
487 if (ih_item_len(ih) < 1
488 || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
489 reiserfs_warning(NULL, "reiserfs-5085",
490 "item length seems wrong: %h",
494 if (prev_location - ih_location(ih) != ih_item_len(ih)) {
495 reiserfs_warning(NULL, "reiserfs-5086",
496 "item location seems wrong "
497 "(second one): %h", ih);
500 prev_location = ih_location(ih);
503 // one may imagine much more checks
507 /* returns 1 if buf looks like an internal node, 0 otherwise */
508 static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
510 struct block_head *blkh;
514 blkh = (struct block_head *)buf;
515 nr = blkh_level(blkh);
516 if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
517 /* this level is not possible for internal nodes */
518 reiserfs_warning(NULL, "reiserfs-5087",
519 "this should be caught earlier");
523 nr = blkh_nr_item(blkh);
524 if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
525 /* for internal which is not root we might check min number of keys */
526 reiserfs_warning(NULL, "reiserfs-5088",
527 "number of key seems wrong: %z", bh);
531 used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
532 if (used_space != blocksize - blkh_free_space(blkh)) {
533 reiserfs_warning(NULL, "reiserfs-5089",
534 "free space seems wrong: %z", bh);
537 // one may imagine much more checks
541 // make sure that bh contains formatted node of reiserfs tree of
543 static int is_tree_node(struct buffer_head *bh, int level)
545 if (B_LEVEL(bh) != level) {
546 reiserfs_warning(NULL, "reiserfs-5090", "node level %d does "
547 "not match to the expected one %d",
551 if (level == DISK_LEAF_NODE_LEVEL)
552 return is_leaf(bh->b_data, bh->b_size, bh);
554 return is_internal(bh->b_data, bh->b_size, bh);
557 #define SEARCH_BY_KEY_READA 16
559 /* The function is NOT SCHEDULE-SAFE! */
560 static void search_by_key_reada(struct super_block *s,
561 struct buffer_head **bh,
562 b_blocknr_t *b, int num)
566 for (i = 0; i < num; i++) {
567 bh[i] = sb_getblk(s, b[i]);
569 for (j = 0; j < i; j++) {
571 * note, this needs attention if we are getting rid of the BKL
572 * you have to make sure the prepared bit isn't set on this buffer
574 if (!buffer_uptodate(bh[j]))
575 ll_rw_block(READA, 1, bh + j);
580 /**************************************************************************
581 * Algorithm SearchByKey *
582 * look for item in the Disk S+Tree by its key *
583 * Input: p_s_sb - super block *
584 * p_s_key - pointer to the key to search *
585 * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR *
586 * p_s_search_path - path from the root to the needed leaf *
587 **************************************************************************/
589 /* This function fills up the path from the root to the leaf as it
590 descends the tree looking for the key. It uses reiserfs_bread to
591 try to find buffers in the cache given their block number. If it
592 does not find them in the cache it reads them from disk. For each
593 node search_by_key finds using reiserfs_bread it then uses
594 bin_search to look through that node. bin_search will find the
595 position of the block_number of the next node if it is looking
596 through an internal node. If it is looking through a leaf node
597 bin_search will find the position of the item which has key either
598 equal to given key, or which is the maximal key less than the given
599 key. search_by_key returns a path that must be checked for the
600 correctness of the top of the path but need not be checked for the
601 correctness of the bottom of the path */
602 /* The function is NOT SCHEDULE-SAFE! */
603 int search_by_key(struct super_block *p_s_sb, const struct cpu_key *p_s_key, /* Key to search. */
604 struct treepath *p_s_search_path,/* This structure was
605 allocated and initialized
607 function. It is filled up
609 int n_stop_level /* How far down the tree to search. To
610 stop at leaf level - set to
611 DISK_LEAF_NODE_LEVEL */
614 b_blocknr_t n_block_number;
616 struct buffer_head *p_s_bh;
617 struct path_element *p_s_last_element;
618 int n_node_level, n_retval;
619 int right_neighbor_of_leaf_node;
621 struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
622 b_blocknr_t reada_blocks[SEARCH_BY_KEY_READA];
625 #ifdef CONFIG_REISERFS_CHECK
626 int n_repeat_counter = 0;
629 PROC_INFO_INC(p_s_sb, search_by_key);
631 /* As we add each node to a path we increase its count. This means that
632 we must be careful to release all nodes in a path before we either
633 discard the path struct or re-use the path struct, as we do here. */
635 decrement_counters_in_path(p_s_search_path);
637 right_neighbor_of_leaf_node = 0;
639 /* With each iteration of this loop we search through the items in the
640 current node, and calculate the next current node(next path element)
641 for the next iteration of this loop.. */
642 n_block_number = SB_ROOT_BLOCK(p_s_sb);
646 #ifdef CONFIG_REISERFS_CHECK
647 if (!(++n_repeat_counter % 50000))
648 reiserfs_warning(p_s_sb, "PAP-5100",
649 "%s: there were %d iterations of "
650 "while loop looking for key %K",
651 current->comm, n_repeat_counter,
655 /* prep path to have another element added to it. */
657 PATH_OFFSET_PELEMENT(p_s_search_path,
658 ++p_s_search_path->path_length);
659 fs_gen = get_generation(p_s_sb);
661 /* Read the next tree node, and set the last element in the path to
662 have a pointer to it. */
663 if ((p_s_bh = p_s_last_element->pe_buffer =
664 sb_getblk(p_s_sb, n_block_number))) {
665 if (!buffer_uptodate(p_s_bh) && reada_count > 1) {
666 search_by_key_reada(p_s_sb, reada_bh,
667 reada_blocks, reada_count);
669 ll_rw_block(READ, 1, &p_s_bh);
670 wait_on_buffer(p_s_bh);
671 if (!buffer_uptodate(p_s_bh))
675 p_s_search_path->path_length--;
676 pathrelse(p_s_search_path);
680 if (expected_level == -1)
681 expected_level = SB_TREE_HEIGHT(p_s_sb);
684 /* It is possible that schedule occurred. We must check whether the key
685 to search is still in the tree rooted from the current buffer. If
686 not then repeat search from the root. */
687 if (fs_changed(fs_gen, p_s_sb) &&
688 (!B_IS_IN_TREE(p_s_bh) ||
689 B_LEVEL(p_s_bh) != expected_level ||
690 !key_in_buffer(p_s_search_path, p_s_key, p_s_sb))) {
691 PROC_INFO_INC(p_s_sb, search_by_key_fs_changed);
692 PROC_INFO_INC(p_s_sb, search_by_key_restarted);
693 PROC_INFO_INC(p_s_sb,
694 sbk_restarted[expected_level - 1]);
695 decrement_counters_in_path(p_s_search_path);
697 /* Get the root block number so that we can repeat the search
698 starting from the root. */
699 n_block_number = SB_ROOT_BLOCK(p_s_sb);
701 right_neighbor_of_leaf_node = 0;
703 /* repeat search from the root */
707 /* only check that the key is in the buffer if p_s_key is not
708 equal to the MAX_KEY. Latter case is only possible in
709 "finish_unfinished()" processing during mount. */
710 RFALSE(comp_keys(&MAX_KEY, p_s_key) &&
711 !key_in_buffer(p_s_search_path, p_s_key, p_s_sb),
712 "PAP-5130: key is not in the buffer");
713 #ifdef CONFIG_REISERFS_CHECK
715 print_cur_tb("5140");
716 reiserfs_panic(p_s_sb,
717 "PAP-5140: search_by_key: schedule occurred in do_balance!");
721 // make sure, that the node contents look like a node of
723 if (!is_tree_node(p_s_bh, expected_level)) {
724 reiserfs_warning(p_s_sb, "vs-5150",
725 "invalid format found in block %ld. "
726 "Fsck?", p_s_bh->b_blocknr);
727 pathrelse(p_s_search_path);
731 /* ok, we have acquired next formatted node in the tree */
732 n_node_level = B_LEVEL(p_s_bh);
734 PROC_INFO_BH_STAT(p_s_sb, p_s_bh, n_node_level - 1);
736 RFALSE(n_node_level < n_stop_level,
737 "vs-5152: tree level (%d) is less than stop level (%d)",
738 n_node_level, n_stop_level);
740 n_retval = bin_search(p_s_key, B_N_PITEM_HEAD(p_s_bh, 0),
743 DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
745 &(p_s_last_element->pe_position));
746 if (n_node_level == n_stop_level) {
750 /* we are not in the stop level */
751 if (n_retval == ITEM_FOUND)
752 /* item has been found, so we choose the pointer which is to the right of the found one */
753 p_s_last_element->pe_position++;
755 /* if item was not found we choose the position which is to
756 the left of the found item. This requires no code,
757 bin_search did it already. */
759 /* So we have chosen a position in the current node which is
760 an internal node. Now we calculate child block number by
761 position in the node. */
763 B_N_CHILD_NUM(p_s_bh, p_s_last_element->pe_position);
765 /* if we are going to read leaf nodes, try for read ahead as well */
766 if ((p_s_search_path->reada & PATH_READA) &&
767 n_node_level == DISK_LEAF_NODE_LEVEL + 1) {
768 int pos = p_s_last_element->pe_position;
769 int limit = B_NR_ITEMS(p_s_bh);
770 struct reiserfs_key *le_key;
772 if (p_s_search_path->reada & PATH_READA_BACK)
774 while (reada_count < SEARCH_BY_KEY_READA) {
777 reada_blocks[reada_count++] =
778 B_N_CHILD_NUM(p_s_bh, pos);
779 if (p_s_search_path->reada & PATH_READA_BACK)
785 * check to make sure we're in the same object
787 le_key = B_N_PDELIM_KEY(p_s_bh, pos);
788 if (le32_to_cpu(le_key->k_objectid) !=
789 p_s_key->on_disk_key.k_objectid) {
797 /* Form the path to an item and position in this item which contains
798 file byte defined by p_s_key. If there is no such item
799 corresponding to the key, we point the path to the item with
800 maximal key less than p_s_key, and *p_n_pos_in_item is set to one
801 past the last entry/byte in the item. If searching for entry in a
802 directory item, and it is not found, *p_n_pos_in_item is set to one
803 entry more than the entry with maximal key which is less than the
806 Note that if there is no entry in this same node which is one more,
807 then we point to an imaginary entry. for direct items, the
808 position is in units of bytes, for indirect items the position is
809 in units of blocknr entries, for directory items the position is in
810 units of directory entries. */
812 /* The function is NOT SCHEDULE-SAFE! */
813 int search_for_position_by_key(struct super_block *p_s_sb, /* Pointer to the super block. */
814 const struct cpu_key *p_cpu_key, /* Key to search (cpu variable) */
815 struct treepath *p_s_search_path /* Filled up by this function. */
818 struct item_head *p_le_ih; /* pointer to on-disk structure */
820 loff_t item_offset, offset;
821 struct reiserfs_dir_entry de;
824 /* If searching for directory entry. */
825 if (is_direntry_cpu_key(p_cpu_key))
826 return search_by_entry_key(p_s_sb, p_cpu_key, p_s_search_path,
829 /* If not searching for directory entry. */
831 /* If item is found. */
832 retval = search_item(p_s_sb, p_cpu_key, p_s_search_path);
833 if (retval == IO_ERROR)
835 if (retval == ITEM_FOUND) {
839 (PATH_PLAST_BUFFER(p_s_search_path),
840 PATH_LAST_POSITION(p_s_search_path))),
841 "PAP-5165: item length equals zero");
843 pos_in_item(p_s_search_path) = 0;
844 return POSITION_FOUND;
847 RFALSE(!PATH_LAST_POSITION(p_s_search_path),
848 "PAP-5170: position equals zero");
850 /* Item is not found. Set path to the previous item. */
852 B_N_PITEM_HEAD(PATH_PLAST_BUFFER(p_s_search_path),
853 --PATH_LAST_POSITION(p_s_search_path));
854 n_blk_size = p_s_sb->s_blocksize;
856 if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) {
857 return FILE_NOT_FOUND;
859 // FIXME: quite ugly this far
861 item_offset = le_ih_k_offset(p_le_ih);
862 offset = cpu_key_k_offset(p_cpu_key);
864 /* Needed byte is contained in the item pointed to by the path. */
865 if (item_offset <= offset &&
866 item_offset + op_bytes_number(p_le_ih, n_blk_size) > offset) {
867 pos_in_item(p_s_search_path) = offset - item_offset;
868 if (is_indirect_le_ih(p_le_ih)) {
869 pos_in_item(p_s_search_path) /= n_blk_size;
871 return POSITION_FOUND;
874 /* Needed byte is not contained in the item pointed to by the
875 path. Set pos_in_item out of the item. */
876 if (is_indirect_le_ih(p_le_ih))
877 pos_in_item(p_s_search_path) =
878 ih_item_len(p_le_ih) / UNFM_P_SIZE;
880 pos_in_item(p_s_search_path) = ih_item_len(p_le_ih);
882 return POSITION_NOT_FOUND;
885 /* Compare given item and item pointed to by the path. */
886 int comp_items(const struct item_head *stored_ih, const struct treepath *p_s_path)
888 struct buffer_head *p_s_bh;
889 struct item_head *ih;
891 /* Last buffer at the path is not in the tree. */
892 if (!B_IS_IN_TREE(p_s_bh = PATH_PLAST_BUFFER(p_s_path)))
895 /* Last path position is invalid. */
896 if (PATH_LAST_POSITION(p_s_path) >= B_NR_ITEMS(p_s_bh))
899 /* we need only to know, whether it is the same item */
900 ih = get_ih(p_s_path);
901 return memcmp(stored_ih, ih, IH_SIZE);
904 /* unformatted nodes are not logged anymore, ever. This is safe
907 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
909 // block can not be forgotten as it is in I/O or held by someone
910 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
912 // prepare for delete or cut of direct item
913 static inline int prepare_for_direct_item(struct treepath *path,
914 struct item_head *le_ih,
916 loff_t new_file_length, int *cut_size)
920 if (new_file_length == max_reiserfs_offset(inode)) {
921 /* item has to be deleted */
922 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
925 // new file gets truncated
926 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
928 round_len = ROUND_UP(new_file_length);
929 /* this was n_new_file_length < le_ih ... */
930 if (round_len < le_ih_k_offset(le_ih)) {
931 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
932 return M_DELETE; /* Delete this item. */
934 /* Calculate first position and size for cutting from item. */
935 pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
936 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
938 return M_CUT; /* Cut from this item. */
941 // old file: items may have any length
943 if (new_file_length < le_ih_k_offset(le_ih)) {
944 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
945 return M_DELETE; /* Delete this item. */
947 /* Calculate first position and size for cutting from item. */
948 *cut_size = -(ih_item_len(le_ih) -
950 new_file_length + 1 - le_ih_k_offset(le_ih)));
951 return M_CUT; /* Cut from this item. */
954 static inline int prepare_for_direntry_item(struct treepath *path,
955 struct item_head *le_ih,
957 loff_t new_file_length,
960 if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
961 new_file_length == max_reiserfs_offset(inode)) {
962 RFALSE(ih_entry_count(le_ih) != 2,
963 "PAP-5220: incorrect empty directory item (%h)", le_ih);
964 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
965 return M_DELETE; /* Delete the directory item containing "." and ".." entry. */
968 if (ih_entry_count(le_ih) == 1) {
969 /* Delete the directory item such as there is one record only
971 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
975 /* Cut one record from the directory item. */
978 entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
982 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
984 /* If the path points to a directory or direct item, calculate mode and the size cut, for balance.
985 If the path points to an indirect item, remove some number of its unformatted nodes.
986 In case of file truncate calculate whether this item must be deleted/truncated or last
987 unformatted node of this item will be converted to a direct item.
988 This function returns a determination of what balance mode the calling function should employ. */
989 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
990 from end of the file. */
991 int *p_n_cut_size, unsigned long long n_new_file_length /* MAX_KEY_OFFSET in case of delete. */
994 struct super_block *p_s_sb = inode->i_sb;
995 struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_path);
996 struct buffer_head *p_s_bh = PATH_PLAST_BUFFER(p_s_path);
998 BUG_ON(!th->t_trans_id);
1000 /* Stat_data item. */
1001 if (is_statdata_le_ih(p_le_ih)) {
1003 RFALSE(n_new_file_length != max_reiserfs_offset(inode),
1004 "PAP-5210: mode must be M_DELETE");
1006 *p_n_cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
1010 /* Directory item. */
1011 if (is_direntry_le_ih(p_le_ih))
1012 return prepare_for_direntry_item(p_s_path, p_le_ih, inode,
1017 if (is_direct_le_ih(p_le_ih))
1018 return prepare_for_direct_item(p_s_path, p_le_ih, inode,
1019 n_new_file_length, p_n_cut_size);
1021 /* Case of an indirect item. */
1023 int blk_size = p_s_sb->s_blocksize;
1024 struct item_head s_ih;
1030 if ( n_new_file_length == max_reiserfs_offset (inode) ) {
1031 /* prepare_for_delete_or_cut() is called by
1032 * reiserfs_delete_item() */
1033 n_new_file_length = 0;
1040 p_s_bh = PATH_PLAST_BUFFER(p_s_path);
1041 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1042 pos = I_UNFM_NUM(&s_ih);
1044 while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > n_new_file_length) {
1048 /* Each unformatted block deletion may involve one additional
1049 * bitmap block into the transaction, thereby the initial
1050 * journal space reservation might not be enough. */
1051 if (!delete && (*p_n_cut_size) != 0 &&
1052 reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1056 unfm = (__le32 *)B_I_PITEM(p_s_bh, &s_ih) + pos - 1;
1057 block = get_block_num(unfm, 0);
1060 reiserfs_prepare_for_journal(p_s_sb, p_s_bh, 1);
1061 put_block_num(unfm, 0, 0);
1062 journal_mark_dirty (th, p_s_sb, p_s_bh);
1063 reiserfs_free_block(th, inode, block, 1);
1068 if (item_moved (&s_ih, p_s_path)) {
1075 (*p_n_cut_size) -= UNFM_P_SIZE;
1078 (*p_n_cut_size) -= IH_SIZE;
1083 /* a trick. If the buffer has been logged, this will do nothing. If
1084 ** we've broken the loop without logging it, it will restore the
1086 reiserfs_restore_prepared_buffer(p_s_sb, p_s_bh);
1087 } while (need_re_search &&
1088 search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path) == POSITION_FOUND);
1089 pos_in_item(p_s_path) = pos * UNFM_P_SIZE;
1091 if (*p_n_cut_size == 0) {
1092 /* Nothing were cut. maybe convert last unformatted node to the
1100 /* Calculate number of bytes which will be deleted or cut during balance */
1101 static int calc_deleted_bytes_number(struct tree_balance *p_s_tb, char c_mode)
1104 struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_tb->tb_path);
1106 if (is_statdata_le_ih(p_le_ih))
1111 M_DELETE) ? ih_item_len(p_le_ih) : -p_s_tb->insert_size[0];
1112 if (is_direntry_le_ih(p_le_ih)) {
1113 // return EMPTY_DIR_SIZE; /* We delete emty directoris only. */
1114 // we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1115 // empty size. ick. FIXME, is this right?
1120 if (is_indirect_le_ih(p_le_ih))
1121 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);
1125 static void init_tb_struct(struct reiserfs_transaction_handle *th,
1126 struct tree_balance *p_s_tb,
1127 struct super_block *p_s_sb,
1128 struct treepath *p_s_path, int n_size)
1131 BUG_ON(!th->t_trans_id);
1133 memset(p_s_tb, '\0', sizeof(struct tree_balance));
1134 p_s_tb->transaction_handle = th;
1135 p_s_tb->tb_sb = p_s_sb;
1136 p_s_tb->tb_path = p_s_path;
1137 PATH_OFFSET_PBUFFER(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1138 PATH_OFFSET_POSITION(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1139 p_s_tb->insert_size[0] = n_size;
1142 void padd_item(char *item, int total_length, int length)
1146 for (i = total_length; i > length;)
1150 #ifdef REISERQUOTA_DEBUG
1151 char key2type(struct reiserfs_key *ih)
1153 if (is_direntry_le_key(2, ih))
1155 if (is_direct_le_key(2, ih))
1157 if (is_indirect_le_key(2, ih))
1159 if (is_statdata_le_key(2, ih))
1164 char head2type(struct item_head *ih)
1166 if (is_direntry_le_ih(ih))
1168 if (is_direct_le_ih(ih))
1170 if (is_indirect_le_ih(ih))
1172 if (is_statdata_le_ih(ih))
1178 /* Delete object item. */
1179 int reiserfs_delete_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_path, /* Path to the deleted item. */
1180 const struct cpu_key *p_s_item_key, /* Key to search for the deleted item. */
1181 struct inode *p_s_inode, /* inode is here just to update i_blocks and quotas */
1182 struct buffer_head *p_s_un_bh)
1183 { /* NULL or unformatted node pointer. */
1184 struct super_block *p_s_sb = p_s_inode->i_sb;
1185 struct tree_balance s_del_balance;
1186 struct item_head s_ih;
1187 struct item_head *q_ih;
1188 int quota_cut_bytes;
1189 int n_ret_value, n_del_size, n_removed;
1191 #ifdef CONFIG_REISERFS_CHECK
1196 BUG_ON(!th->t_trans_id);
1198 init_tb_struct(th, &s_del_balance, p_s_sb, p_s_path,
1199 0 /*size is unknown */ );
1204 #ifdef CONFIG_REISERFS_CHECK
1208 prepare_for_delete_or_cut(th, p_s_inode, p_s_path,
1209 p_s_item_key, &n_removed,
1211 max_reiserfs_offset(p_s_inode));
1213 RFALSE(c_mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1215 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1216 s_del_balance.insert_size[0] = n_del_size;
1218 n_ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1219 if (n_ret_value != REPEAT_SEARCH)
1222 PROC_INFO_INC(p_s_sb, delete_item_restarted);
1224 // file system changed, repeat search
1226 search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1227 if (n_ret_value == IO_ERROR)
1229 if (n_ret_value == FILE_NOT_FOUND) {
1230 reiserfs_warning(p_s_sb, "vs-5340",
1231 "no items of the file %K found",
1237 if (n_ret_value != CARRY_ON) {
1238 unfix_nodes(&s_del_balance);
1241 // reiserfs_delete_item returns item length when success
1242 n_ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1243 q_ih = get_ih(p_s_path);
1244 quota_cut_bytes = ih_item_len(q_ih);
1246 /* hack so the quota code doesn't have to guess if the file
1247 ** has a tail. On tail insert, we allocate quota for 1 unformatted node.
1248 ** We test the offset because the tail might have been
1249 ** split into multiple items, and we only want to decrement for
1250 ** the unfm node once
1252 if (!S_ISLNK(p_s_inode->i_mode) && is_direct_le_ih(q_ih)) {
1253 if ((le_ih_k_offset(q_ih) & (p_s_sb->s_blocksize - 1)) == 1) {
1254 quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1256 quota_cut_bytes = 0;
1264 /* We are in direct2indirect conversion, so move tail contents
1265 to the unformatted node */
1266 /* note, we do the copy before preparing the buffer because we
1267 ** don't care about the contents of the unformatted node yet.
1268 ** the only thing we really care about is the direct item's data
1269 ** is in the unformatted node.
1271 ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1272 ** the unformatted node, which might schedule, meaning we'd have to
1273 ** loop all the way back up to the start of the while loop.
1275 ** The unformatted node must be dirtied later on. We can't be
1276 ** sure here if the entire tail has been deleted yet.
1278 ** p_s_un_bh is from the page cache (all unformatted nodes are
1279 ** from the page cache) and might be a highmem page. So, we
1280 ** can't use p_s_un_bh->b_data.
1284 data = kmap_atomic(p_s_un_bh->b_page, KM_USER0);
1285 off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1287 B_I_PITEM(PATH_PLAST_BUFFER(p_s_path), &s_ih),
1289 kunmap_atomic(data, KM_USER0);
1291 /* Perform balancing after all resources have been collected at once. */
1292 do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1294 #ifdef REISERQUOTA_DEBUG
1295 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1296 "reiserquota delete_item(): freeing %u, id=%u type=%c",
1297 quota_cut_bytes, p_s_inode->i_uid, head2type(&s_ih));
1299 DQUOT_FREE_SPACE_NODIRTY(p_s_inode, quota_cut_bytes);
1301 /* Return deleted body length */
1305 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1307 deletion of the body of the object is performed by iput(), with the
1308 result that if multiple processes are operating on a file, the
1309 deletion of the body of the file is deferred until the last process
1310 that has an open inode performs its iput().
1312 writes and truncates are protected from collisions by use of
1315 creates, linking, and mknod are protected from collisions with other
1316 processes by making the reiserfs_add_entry() the last step in the
1317 creation, and then rolling back all changes if there was a collision.
1321 /* this deletes item which never gets split */
1322 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1323 struct inode *inode, struct reiserfs_key *key)
1325 struct tree_balance tb;
1326 INITIALIZE_PATH(path);
1329 struct cpu_key cpu_key;
1331 int quota_cut_bytes = 0;
1333 BUG_ON(!th->t_trans_id);
1335 le_key2cpu_key(&cpu_key, key);
1338 retval = search_item(th->t_super, &cpu_key, &path);
1339 if (retval == IO_ERROR) {
1340 reiserfs_warning(th->t_super, "vs-5350",
1341 "i/o failure occurred trying "
1342 "to delete %K", &cpu_key);
1345 if (retval != ITEM_FOUND) {
1347 // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1349 ((unsigned long long)
1350 GET_HASH_VALUE(le_key_k_offset
1351 (le_key_version(key), key)) == 0
1352 && (unsigned long long)
1353 GET_GENERATION_NUMBER(le_key_k_offset
1354 (le_key_version(key),
1356 reiserfs_warning(th->t_super, "vs-5355",
1357 "%k not found", key);
1362 item_len = ih_item_len(PATH_PITEM_HEAD(&path));
1363 init_tb_struct(th, &tb, th->t_super, &path,
1364 -(IH_SIZE + item_len));
1366 quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path));
1368 retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1369 if (retval == REPEAT_SEARCH) {
1370 PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1374 if (retval == CARRY_ON) {
1375 do_balance(&tb, NULL, NULL, M_DELETE);
1376 if (inode) { /* Should we count quota for item? (we don't count quotas for save-links) */
1377 #ifdef REISERQUOTA_DEBUG
1378 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1379 "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1380 quota_cut_bytes, inode->i_uid,
1383 DQUOT_FREE_SPACE_NODIRTY(inode,
1388 // IO_ERROR, NO_DISK_SPACE, etc
1389 reiserfs_warning(th->t_super, "vs-5360",
1390 "could not delete %K due to fix_nodes failure",
1396 reiserfs_check_path(&path);
1399 int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1400 struct inode *inode)
1404 BUG_ON(!th->t_trans_id);
1406 /* for directory this deletes item containing "." and ".." */
1408 reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1412 #if defined( USE_INODE_GENERATION_COUNTER )
1413 if (!old_format_only(th->t_super)) {
1414 __le32 *inode_generation;
1417 &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
1418 le32_add_cpu(inode_generation, 1);
1420 /* USE_INODE_GENERATION_COUNTER */
1422 reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1427 static void unmap_buffers(struct page *page, loff_t pos)
1429 struct buffer_head *bh;
1430 struct buffer_head *head;
1431 struct buffer_head *next;
1432 unsigned long tail_index;
1433 unsigned long cur_index;
1436 if (page_has_buffers(page)) {
1437 tail_index = pos & (PAGE_CACHE_SIZE - 1);
1439 head = page_buffers(page);
1442 next = bh->b_this_page;
1444 /* we want to unmap the buffers that contain the tail, and
1445 ** all the buffers after it (since the tail must be at the
1446 ** end of the file). We don't want to unmap file data
1447 ** before the tail, since it might be dirty and waiting to
1450 cur_index += bh->b_size;
1451 if (cur_index > tail_index) {
1452 reiserfs_unmap_buffer(bh);
1455 } while (bh != head);
1460 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
1461 struct inode *p_s_inode,
1463 struct treepath *p_s_path,
1464 const struct cpu_key *p_s_item_key,
1465 loff_t n_new_file_size, char *p_c_mode)
1467 struct super_block *p_s_sb = p_s_inode->i_sb;
1468 int n_block_size = p_s_sb->s_blocksize;
1470 BUG_ON(!th->t_trans_id);
1471 BUG_ON(n_new_file_size != p_s_inode->i_size);
1473 /* the page being sent in could be NULL if there was an i/o error
1474 ** reading in the last block. The user will hit problems trying to
1475 ** read the file, but for now we just skip the indirect2direct
1477 if (atomic_read(&p_s_inode->i_count) > 1 ||
1478 !tail_has_to_be_packed(p_s_inode) ||
1479 !page || (REISERFS_I(p_s_inode)->i_flags & i_nopack_mask)) {
1480 // leave tail in an unformatted node
1481 *p_c_mode = M_SKIP_BALANCING;
1483 n_block_size - (n_new_file_size & (n_block_size - 1));
1484 pathrelse(p_s_path);
1487 /* Permorm the conversion to a direct_item. */
1488 /*return indirect_to_direct (p_s_inode, p_s_path, p_s_item_key, n_new_file_size, p_c_mode); */
1489 return indirect2direct(th, p_s_inode, page, p_s_path, p_s_item_key,
1490 n_new_file_size, p_c_mode);
1493 /* we did indirect_to_direct conversion. And we have inserted direct
1494 item successesfully, but there were no disk space to cut unfm
1495 pointer being converted. Therefore we have to delete inserted
1497 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
1498 struct inode *inode, struct treepath *path)
1500 struct cpu_key tail_key;
1503 BUG_ON(!th->t_trans_id);
1505 make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4); // !!!!
1506 tail_key.key_length = 4;
1509 (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1511 /* look for the last byte of the tail */
1512 if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1514 reiserfs_panic(inode->i_sb,
1515 "vs-5615: indirect_to_direct_roll_back: found invalid item");
1516 RFALSE(path->pos_in_item !=
1517 ih_item_len(PATH_PITEM_HEAD(path)) - 1,
1518 "vs-5616: appended bytes found");
1519 PATH_LAST_POSITION(path)--;
1522 reiserfs_delete_item(th, path, &tail_key, inode,
1523 NULL /*unbh not needed */ );
1525 || removed > tail_len,
1526 "vs-5617: there was tail %d bytes, removed item length %d bytes",
1528 tail_len -= removed;
1529 set_cpu_key_k_offset(&tail_key,
1530 cpu_key_k_offset(&tail_key) - removed);
1532 reiserfs_warning(inode->i_sb, "reiserfs-5091", "indirect_to_direct "
1533 "conversion has been rolled back due to "
1534 "lack of disk space");
1535 //mark_file_without_tail (inode);
1536 mark_inode_dirty(inode);
1539 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1540 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
1541 struct treepath *p_s_path,
1542 struct cpu_key *p_s_item_key,
1543 struct inode *p_s_inode,
1544 struct page *page, loff_t n_new_file_size)
1546 struct super_block *p_s_sb = p_s_inode->i_sb;
1547 /* Every function which is going to call do_balance must first
1548 create a tree_balance structure. Then it must fill up this
1549 structure by using the init_tb_struct and fix_nodes functions.
1550 After that we can make tree balancing. */
1551 struct tree_balance s_cut_balance;
1552 struct item_head *p_le_ih;
1553 int n_cut_size = 0, /* Amount to be cut. */
1554 n_ret_value = CARRY_ON, n_removed = 0, /* Number of the removed unformatted nodes. */
1555 n_is_inode_locked = 0;
1556 char c_mode; /* Mode of the balance. */
1558 int quota_cut_bytes;
1559 loff_t tail_pos = 0;
1561 BUG_ON(!th->t_trans_id);
1563 init_tb_struct(th, &s_cut_balance, p_s_inode->i_sb, p_s_path,
1566 /* Repeat this loop until we either cut the item without needing
1567 to balance, or we fix_nodes without schedule occurring */
1569 /* Determine the balance mode, position of the first byte to
1570 be cut, and size to be cut. In case of the indirect item
1571 free unformatted nodes which are pointed to by the cut
1575 prepare_for_delete_or_cut(th, p_s_inode, p_s_path,
1576 p_s_item_key, &n_removed,
1577 &n_cut_size, n_new_file_size);
1578 if (c_mode == M_CONVERT) {
1579 /* convert last unformatted node to direct item or leave
1580 tail in the unformatted node */
1581 RFALSE(n_ret_value != CARRY_ON,
1582 "PAP-5570: can not convert twice");
1585 maybe_indirect_to_direct(th, p_s_inode, page,
1586 p_s_path, p_s_item_key,
1587 n_new_file_size, &c_mode);
1588 if (c_mode == M_SKIP_BALANCING)
1589 /* tail has been left in the unformatted node */
1592 n_is_inode_locked = 1;
1594 /* removing of last unformatted node will change value we
1595 have to return to truncate. Save it */
1596 retval2 = n_ret_value;
1597 /*retval2 = p_s_sb->s_blocksize - (n_new_file_size & (p_s_sb->s_blocksize - 1)); */
1599 /* So, we have performed the first part of the conversion:
1600 inserting the new direct item. Now we are removing the
1601 last unformatted node pointer. Set key to search for
1603 set_cpu_key_k_type(p_s_item_key, TYPE_INDIRECT);
1604 p_s_item_key->key_length = 4;
1606 (n_new_file_size & (p_s_sb->s_blocksize - 1));
1607 tail_pos = n_new_file_size;
1608 set_cpu_key_k_offset(p_s_item_key, n_new_file_size + 1);
1609 if (search_for_position_by_key
1610 (p_s_sb, p_s_item_key,
1611 p_s_path) == POSITION_NOT_FOUND) {
1612 print_block(PATH_PLAST_BUFFER(p_s_path), 3,
1613 PATH_LAST_POSITION(p_s_path) - 1,
1614 PATH_LAST_POSITION(p_s_path) + 1);
1615 reiserfs_panic(p_s_sb,
1616 "PAP-5580: reiserfs_cut_from_item: item to convert does not exist (%K)",
1621 if (n_cut_size == 0) {
1622 pathrelse(p_s_path);
1626 s_cut_balance.insert_size[0] = n_cut_size;
1628 n_ret_value = fix_nodes(c_mode, &s_cut_balance, NULL, NULL);
1629 if (n_ret_value != REPEAT_SEARCH)
1632 PROC_INFO_INC(p_s_sb, cut_from_item_restarted);
1635 search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1636 if (n_ret_value == POSITION_FOUND)
1639 reiserfs_warning(p_s_sb, "PAP-5610", "item %K not found",
1641 unfix_nodes(&s_cut_balance);
1642 return (n_ret_value == IO_ERROR) ? -EIO : -ENOENT;
1645 // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1646 if (n_ret_value != CARRY_ON) {
1647 if (n_is_inode_locked) {
1648 // FIXME: this seems to be not needed: we are always able
1650 indirect_to_direct_roll_back(th, p_s_inode, p_s_path);
1652 if (n_ret_value == NO_DISK_SPACE)
1653 reiserfs_warning(p_s_sb, "reiserfs-5092",
1655 unfix_nodes(&s_cut_balance);
1659 /* go ahead and perform balancing */
1661 RFALSE(c_mode == M_PASTE || c_mode == M_INSERT, "invalid mode");
1663 /* Calculate number of bytes that need to be cut from the item. */
1666 M_DELETE) ? ih_item_len(get_ih(p_s_path)) : -s_cut_balance.
1669 n_ret_value = calc_deleted_bytes_number(&s_cut_balance, c_mode);
1671 n_ret_value = retval2;
1673 /* For direct items, we only change the quota when deleting the last
1676 p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path);
1677 if (!S_ISLNK(p_s_inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1678 if (c_mode == M_DELETE &&
1679 (le_ih_k_offset(p_le_ih) & (p_s_sb->s_blocksize - 1)) ==
1681 // FIXME: this is to keep 3.5 happy
1682 REISERFS_I(p_s_inode)->i_first_direct_byte = U32_MAX;
1683 quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1685 quota_cut_bytes = 0;
1688 #ifdef CONFIG_REISERFS_CHECK
1689 if (n_is_inode_locked) {
1690 struct item_head *le_ih =
1691 PATH_PITEM_HEAD(s_cut_balance.tb_path);
1692 /* we are going to complete indirect2direct conversion. Make
1693 sure, that we exactly remove last unformatted node pointer
1695 if (!is_indirect_le_ih(le_ih))
1696 reiserfs_panic(p_s_sb,
1697 "vs-5652: reiserfs_cut_from_item: "
1698 "item must be indirect %h", le_ih);
1700 if (c_mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1701 reiserfs_panic(p_s_sb,
1702 "vs-5653: reiserfs_cut_from_item: "
1703 "completing indirect2direct conversion indirect item %h "
1704 "being deleted must be of 4 byte long",
1708 && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1709 reiserfs_panic(p_s_sb,
1710 "vs-5654: reiserfs_cut_from_item: "
1711 "can not complete indirect2direct conversion of %h (CUT, insert_size==%d)",
1712 le_ih, s_cut_balance.insert_size[0]);
1714 /* it would be useful to make sure, that right neighboring
1715 item is direct item of this file */
1719 do_balance(&s_cut_balance, NULL, NULL, c_mode);
1720 if (n_is_inode_locked) {
1721 /* we've done an indirect->direct conversion. when the data block
1722 ** was freed, it was removed from the list of blocks that must
1723 ** be flushed before the transaction commits, make sure to
1724 ** unmap and invalidate it
1726 unmap_buffers(page, tail_pos);
1727 REISERFS_I(p_s_inode)->i_flags &= ~i_pack_on_close_mask;
1729 #ifdef REISERQUOTA_DEBUG
1730 reiserfs_debug(p_s_inode->i_sb, REISERFS_DEBUG_CODE,
1731 "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1732 quota_cut_bytes, p_s_inode->i_uid, '?');
1734 DQUOT_FREE_SPACE_NODIRTY(p_s_inode, quota_cut_bytes);
1738 static void truncate_directory(struct reiserfs_transaction_handle *th,
1739 struct inode *inode)
1741 BUG_ON(!th->t_trans_id);
1743 reiserfs_warning(inode->i_sb, "vs-5655", "link count != 0");
1745 set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1746 set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1747 reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1748 reiserfs_update_sd(th, inode);
1749 set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1750 set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
1753 /* Truncate file to the new size. Note, this must be called with a transaction
1755 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th, struct inode *p_s_inode, /* ->i_size contains new
1757 struct page *page, /* up to date for last block */
1758 int update_timestamps /* when it is called by
1759 file_release to convert
1760 the tail - no timestamps
1761 should be updated */
1764 INITIALIZE_PATH(s_search_path); /* Path to the current object item. */
1765 struct item_head *p_le_ih; /* Pointer to an item header. */
1766 struct cpu_key s_item_key; /* Key to search for a previous file item. */
1767 loff_t n_file_size, /* Old file size. */
1768 n_new_file_size; /* New file size. */
1769 int n_deleted; /* Number of deleted or truncated bytes. */
1773 BUG_ON(!th->t_trans_id);
1775 (S_ISREG(p_s_inode->i_mode) || S_ISDIR(p_s_inode->i_mode)
1776 || S_ISLNK(p_s_inode->i_mode)))
1779 if (S_ISDIR(p_s_inode->i_mode)) {
1780 // deletion of directory - no need to update timestamps
1781 truncate_directory(th, p_s_inode);
1785 /* Get new file size. */
1786 n_new_file_size = p_s_inode->i_size;
1788 // FIXME: note, that key type is unimportant here
1789 make_cpu_key(&s_item_key, p_s_inode, max_reiserfs_offset(p_s_inode),
1793 search_for_position_by_key(p_s_inode->i_sb, &s_item_key,
1795 if (retval == IO_ERROR) {
1796 reiserfs_warning(p_s_inode->i_sb, "vs-5657",
1797 "i/o failure occurred trying to truncate %K",
1802 if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1803 reiserfs_warning(p_s_inode->i_sb, "PAP-5660",
1804 "wrong result %d of search for %K", retval,
1811 s_search_path.pos_in_item--;
1813 /* Get real file size (total length of all file items) */
1814 p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1815 if (is_statdata_le_ih(p_le_ih))
1818 loff_t offset = le_ih_k_offset(p_le_ih);
1820 op_bytes_number(p_le_ih, p_s_inode->i_sb->s_blocksize);
1822 /* this may mismatch with real file size: if last direct item
1823 had no padding zeros and last unformatted node had no free
1824 space, this file would have this file size */
1825 n_file_size = offset + bytes - 1;
1828 * are we doing a full truncate or delete, if so
1829 * kick in the reada code
1831 if (n_new_file_size == 0)
1832 s_search_path.reada = PATH_READA | PATH_READA_BACK;
1834 if (n_file_size == 0 || n_file_size < n_new_file_size) {
1835 goto update_and_out;
1838 /* Update key to search for the last file item. */
1839 set_cpu_key_k_offset(&s_item_key, n_file_size);
1842 /* Cut or delete file item. */
1844 reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
1845 p_s_inode, page, n_new_file_size);
1846 if (n_deleted < 0) {
1847 reiserfs_warning(p_s_inode->i_sb, "vs-5665",
1848 "reiserfs_cut_from_item failed");
1849 reiserfs_check_path(&s_search_path);
1853 RFALSE(n_deleted > n_file_size,
1854 "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1855 n_deleted, n_file_size, &s_item_key);
1857 /* Change key to search the last file item. */
1858 n_file_size -= n_deleted;
1860 set_cpu_key_k_offset(&s_item_key, n_file_size);
1862 /* While there are bytes to truncate and previous file item is presented in the tree. */
1865 ** This loop could take a really long time, and could log
1866 ** many more blocks than a transaction can hold. So, we do a polite
1867 ** journal end here, and if the transaction needs ending, we make
1868 ** sure the file is consistent before ending the current trans
1869 ** and starting a new one
1871 if (journal_transaction_should_end(th, 0) ||
1872 reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1873 int orig_len_alloc = th->t_blocks_allocated;
1874 decrement_counters_in_path(&s_search_path);
1876 if (update_timestamps) {
1877 p_s_inode->i_mtime = p_s_inode->i_ctime =
1880 reiserfs_update_sd(th, p_s_inode);
1882 err = journal_end(th, p_s_inode->i_sb, orig_len_alloc);
1885 err = journal_begin(th, p_s_inode->i_sb,
1886 JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ;
1889 reiserfs_update_inode_transaction(p_s_inode);
1891 } while (n_file_size > ROUND_UP(n_new_file_size) &&
1892 search_for_position_by_key(p_s_inode->i_sb, &s_item_key,
1893 &s_search_path) == POSITION_FOUND);
1895 RFALSE(n_file_size > ROUND_UP(n_new_file_size),
1896 "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1897 n_new_file_size, n_file_size, s_item_key.on_disk_key.k_objectid);
1900 if (update_timestamps) {
1901 // this is truncate, not file closing
1902 p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME_SEC;
1904 reiserfs_update_sd(th, p_s_inode);
1907 pathrelse(&s_search_path);
1911 #ifdef CONFIG_REISERFS_CHECK
1912 // this makes sure, that we __append__, not overwrite or add holes
1913 static void check_research_for_paste(struct treepath *path,
1914 const struct cpu_key *p_s_key)
1916 struct item_head *found_ih = get_ih(path);
1918 if (is_direct_le_ih(found_ih)) {
1919 if (le_ih_k_offset(found_ih) +
1920 op_bytes_number(found_ih,
1921 get_last_bh(path)->b_size) !=
1922 cpu_key_k_offset(p_s_key)
1923 || op_bytes_number(found_ih,
1924 get_last_bh(path)->b_size) !=
1926 reiserfs_panic(NULL,
1927 "PAP-5720: check_research_for_paste: "
1928 "found direct item %h or position (%d) does not match to key %K",
1929 found_ih, pos_in_item(path), p_s_key);
1931 if (is_indirect_le_ih(found_ih)) {
1932 if (le_ih_k_offset(found_ih) +
1933 op_bytes_number(found_ih,
1934 get_last_bh(path)->b_size) !=
1935 cpu_key_k_offset(p_s_key)
1936 || I_UNFM_NUM(found_ih) != pos_in_item(path)
1937 || get_ih_free_space(found_ih) != 0)
1938 reiserfs_panic(NULL,
1939 "PAP-5730: check_research_for_paste: "
1940 "found indirect item (%h) or position (%d) does not match to key (%K)",
1941 found_ih, pos_in_item(path), p_s_key);
1944 #endif /* config reiserfs check */
1946 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1947 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_search_path, /* Path to the pasted item. */
1948 const struct cpu_key *p_s_key, /* Key to search for the needed item. */
1949 struct inode *inode, /* Inode item belongs to */
1950 const char *p_c_body, /* Pointer to the bytes to paste. */
1952 { /* Size of pasted bytes. */
1953 struct tree_balance s_paste_balance;
1957 BUG_ON(!th->t_trans_id);
1959 fs_gen = get_generation(inode->i_sb);
1961 #ifdef REISERQUOTA_DEBUG
1962 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1963 "reiserquota paste_into_item(): allocating %u id=%u type=%c",
1964 n_pasted_size, inode->i_uid,
1965 key2type(&(p_s_key->on_disk_key)));
1968 if (DQUOT_ALLOC_SPACE_NODIRTY(inode, n_pasted_size)) {
1969 pathrelse(p_s_search_path);
1972 init_tb_struct(th, &s_paste_balance, th->t_super, p_s_search_path,
1974 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1975 s_paste_balance.key = p_s_key->on_disk_key;
1978 /* DQUOT_* can schedule, must check before the fix_nodes */
1979 if (fs_changed(fs_gen, inode->i_sb)) {
1984 fix_nodes(M_PASTE, &s_paste_balance, NULL,
1985 p_c_body)) == REPEAT_SEARCH) {
1987 /* file system changed while we were in the fix_nodes */
1988 PROC_INFO_INC(th->t_super, paste_into_item_restarted);
1990 search_for_position_by_key(th->t_super, p_s_key,
1992 if (retval == IO_ERROR) {
1996 if (retval == POSITION_FOUND) {
1997 reiserfs_warning(inode->i_sb, "PAP-5710",
1998 "entry or pasted byte (%K) exists",
2003 #ifdef CONFIG_REISERFS_CHECK
2004 check_research_for_paste(p_s_search_path, p_s_key);
2008 /* Perform balancing after all resources are collected by fix_nodes, and
2009 accessing them will not risk triggering schedule. */
2010 if (retval == CARRY_ON) {
2011 do_balance(&s_paste_balance, NULL /*ih */ , p_c_body, M_PASTE);
2014 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2016 /* this also releases the path */
2017 unfix_nodes(&s_paste_balance);
2018 #ifdef REISERQUOTA_DEBUG
2019 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2020 "reiserquota paste_into_item(): freeing %u id=%u type=%c",
2021 n_pasted_size, inode->i_uid,
2022 key2type(&(p_s_key->on_disk_key)));
2024 DQUOT_FREE_SPACE_NODIRTY(inode, n_pasted_size);
2028 /* Insert new item into the buffer at the path. */
2029 int reiserfs_insert_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_path, /* Path to the inserteded item. */
2030 const struct cpu_key *key, struct item_head *p_s_ih, /* Pointer to the item header to insert. */
2031 struct inode *inode, const char *p_c_body)
2032 { /* Pointer to the bytes to insert. */
2033 struct tree_balance s_ins_balance;
2036 int quota_bytes = 0;
2038 BUG_ON(!th->t_trans_id);
2040 if (inode) { /* Do we count quotas for item? */
2041 fs_gen = get_generation(inode->i_sb);
2042 quota_bytes = ih_item_len(p_s_ih);
2044 /* hack so the quota code doesn't have to guess if the file has
2045 ** a tail, links are always tails, so there's no guessing needed
2047 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_s_ih)) {
2048 quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
2050 #ifdef REISERQUOTA_DEBUG
2051 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2052 "reiserquota insert_item(): allocating %u id=%u type=%c",
2053 quota_bytes, inode->i_uid, head2type(p_s_ih));
2055 /* We can't dirty inode here. It would be immediately written but
2056 * appropriate stat item isn't inserted yet... */
2057 if (DQUOT_ALLOC_SPACE_NODIRTY(inode, quota_bytes)) {
2058 pathrelse(p_s_path);
2062 init_tb_struct(th, &s_ins_balance, th->t_super, p_s_path,
2063 IH_SIZE + ih_item_len(p_s_ih));
2064 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2065 s_ins_balance.key = key->on_disk_key;
2067 /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2068 if (inode && fs_changed(fs_gen, inode->i_sb)) {
2073 fix_nodes(M_INSERT, &s_ins_balance, p_s_ih,
2074 p_c_body)) == REPEAT_SEARCH) {
2076 /* file system changed while we were in the fix_nodes */
2077 PROC_INFO_INC(th->t_super, insert_item_restarted);
2078 retval = search_item(th->t_super, key, p_s_path);
2079 if (retval == IO_ERROR) {
2083 if (retval == ITEM_FOUND) {
2084 reiserfs_warning(th->t_super, "PAP-5760",
2085 "key %K already exists in the tree",
2092 /* make balancing after all resources will be collected at a time */
2093 if (retval == CARRY_ON) {
2094 do_balance(&s_ins_balance, p_s_ih, p_c_body, M_INSERT);
2098 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2100 /* also releases the path */
2101 unfix_nodes(&s_ins_balance);
2102 #ifdef REISERQUOTA_DEBUG
2103 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2104 "reiserquota insert_item(): freeing %u id=%u type=%c",
2105 quota_bytes, inode->i_uid, head2type(p_s_ih));
2108 DQUOT_FREE_SPACE_NODIRTY(inode, quota_bytes);