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
30 * decrement_counters_in_path
32 * pathrelse_and_restore
36 * search_for_position_by_key
38 * prepare_for_direct_item
39 * prepare_for_direntry_item
40 * prepare_for_delete_or_cut
41 * calc_deleted_bytes_number
44 * reiserfs_delete_item
45 * reiserfs_delete_solid_item
46 * reiserfs_delete_object
47 * maybe_indirect_to_direct
48 * indirect_to_direct_roll_back
49 * reiserfs_cut_from_item
51 * reiserfs_do_truncate
52 * reiserfs_paste_into_item
53 * reiserfs_insert_item
56 #include <linux/config.h>
57 #include <linux/time.h>
58 #include <linux/string.h>
59 #include <linux/pagemap.h>
60 #include <linux/reiserfs_fs.h>
61 #include <linux/smp_lock.h>
62 #include <linux/buffer_head.h>
63 #include <linux/quotaops.h>
65 /* Does the buffer contain a disk block which is in the tree. */
66 inline int B_IS_IN_TREE (const struct buffer_head * p_s_bh)
69 RFALSE( B_LEVEL (p_s_bh) > MAX_HEIGHT,
70 "PAP-1010: block (%b) has too big level (%z)", p_s_bh, p_s_bh);
72 return ( B_LEVEL (p_s_bh) != FREE_LEVEL );
75 inline void copy_short_key (void * to, const void * from)
77 memcpy (to, from, SHORT_KEY_SIZE);
81 // to gets item head in le form
83 inline void copy_item_head(struct item_head * p_v_to,
84 const struct item_head * p_v_from)
86 memcpy (p_v_to, p_v_from, IH_SIZE);
90 /* k1 is pointer to on-disk structure which is stored in little-endian
91 form. k2 is pointer to cpu variable. For key of items of the same
92 object this returns 0.
93 Returns: -1 if key1 < key2
96 inline int comp_short_keys (const struct key * le_key,
97 const struct cpu_key * cpu_key)
99 __u32 * p_s_le_u32, * p_s_cpu_u32;
100 int n_key_length = REISERFS_SHORT_KEY_LEN;
102 p_s_le_u32 = (__u32 *)le_key;
103 p_s_cpu_u32 = (__u32 *)&cpu_key->on_disk_key;
104 for( ; n_key_length--; ++p_s_le_u32, ++p_s_cpu_u32 ) {
105 if ( le32_to_cpu (*p_s_le_u32) < *p_s_cpu_u32 )
107 if ( le32_to_cpu (*p_s_le_u32) > *p_s_cpu_u32 )
115 /* k1 is pointer to on-disk structure which is stored in little-endian
116 form. k2 is pointer to cpu variable.
117 Compare keys using all 4 key fields.
118 Returns: -1 if key1 < key2 0
119 if key1 = key2 1 if key1 > key2 */
120 inline int comp_keys (const struct key * le_key, const struct cpu_key * cpu_key)
124 retval = comp_short_keys (le_key, cpu_key);
127 if (le_key_k_offset (le_key_version(le_key), le_key) < cpu_key_k_offset (cpu_key))
129 if (le_key_k_offset (le_key_version(le_key), le_key) > cpu_key_k_offset (cpu_key))
132 if (cpu_key->key_length == 3)
135 /* this part is needed only when tail conversion is in progress */
136 if (le_key_k_type (le_key_version(le_key), le_key) < cpu_key_k_type (cpu_key))
139 if (le_key_k_type (le_key_version(le_key), le_key) > cpu_key_k_type (cpu_key))
147 // FIXME: not used yet
149 inline int comp_cpu_keys (const struct cpu_key * key1,
150 const struct cpu_key * key2)
152 if (key1->on_disk_key.k_dir_id < key2->on_disk_key.k_dir_id)
154 if (key1->on_disk_key.k_dir_id > key2->on_disk_key.k_dir_id)
157 if (key1->on_disk_key.k_objectid < key2->on_disk_key.k_objectid)
159 if (key1->on_disk_key.k_objectid > key2->on_disk_key.k_objectid)
162 if (cpu_key_k_offset (key1) < cpu_key_k_offset (key2))
164 if (cpu_key_k_offset (key1) > cpu_key_k_offset (key2))
167 reiserfs_warning (NULL, "comp_cpu_keys: type are compared for %K and %K",
170 if (cpu_key_k_type (key1) < cpu_key_k_type (key2))
172 if (cpu_key_k_type (key1) > cpu_key_k_type (key2))
177 inline int comp_short_le_keys (const struct key * key1, const struct key * key2)
179 __u32 * p_s_1_u32, * p_s_2_u32;
180 int n_key_length = REISERFS_SHORT_KEY_LEN;
182 p_s_1_u32 = (__u32 *)key1;
183 p_s_2_u32 = (__u32 *)key2;
184 for( ; n_key_length--; ++p_s_1_u32, ++p_s_2_u32 ) {
185 if ( le32_to_cpu (*p_s_1_u32) < le32_to_cpu (*p_s_2_u32) )
187 if ( le32_to_cpu (*p_s_1_u32) > le32_to_cpu (*p_s_2_u32) )
193 inline int comp_short_cpu_keys (const struct cpu_key * key1,
194 const struct cpu_key * key2)
196 __u32 * p_s_1_u32, * p_s_2_u32;
197 int n_key_length = REISERFS_SHORT_KEY_LEN;
199 p_s_1_u32 = (__u32 *)key1;
200 p_s_2_u32 = (__u32 *)key2;
202 for( ; n_key_length--; ++p_s_1_u32, ++p_s_2_u32 ) {
203 if ( *p_s_1_u32 < *p_s_2_u32 )
205 if ( *p_s_1_u32 > *p_s_2_u32 )
213 inline void cpu_key2cpu_key (struct cpu_key * to, const struct cpu_key * from)
215 memcpy (to, from, sizeof (struct cpu_key));
219 inline void le_key2cpu_key (struct cpu_key * to, const struct key * from)
221 to->on_disk_key.k_dir_id = le32_to_cpu (from->k_dir_id);
222 to->on_disk_key.k_objectid = le32_to_cpu (from->k_objectid);
224 // find out version of the key
225 to->version = le_key_version (from);
226 if (to->version == KEY_FORMAT_3_5) {
227 to->on_disk_key.u.k_offset_v1.k_offset = le32_to_cpu (from->u.k_offset_v1.k_offset);
228 to->on_disk_key.u.k_offset_v1.k_uniqueness = le32_to_cpu (from->u.k_offset_v1.k_uniqueness);
230 to->on_disk_key.u.k_offset_v2.k_offset = offset_v2_k_offset(&from->u.k_offset_v2);
231 to->on_disk_key.u.k_offset_v2.k_type = offset_v2_k_type(&from->u.k_offset_v2);
237 // this does not say which one is bigger, it only returns 1 if keys
238 // are not equal, 0 otherwise
239 inline int comp_le_keys (const struct key * k1, const struct key * k2)
241 return memcmp (k1, k2, sizeof (struct key));
244 /**************************************************************************
245 * Binary search toolkit function *
246 * Search for an item in the array by the item key *
247 * Returns: 1 if found, 0 if not found; *
248 * *p_n_pos = number of the searched element if found, else the *
249 * number of the first element that is larger than p_v_key. *
250 **************************************************************************/
251 /* For those not familiar with binary search: n_lbound is the leftmost item that it
252 could be, n_rbound the rightmost item that it could be. We examine the item
253 halfway between n_lbound and n_rbound, and that tells us either that we can increase
254 n_lbound, or decrease n_rbound, or that we have found it, or if n_lbound <= n_rbound that
255 there are no possible items, and we have not found it. With each examination we
256 cut the number of possible items it could be by one more than half rounded down,
258 inline int bin_search (
259 const void * p_v_key, /* Key to search for. */
260 const void * p_v_base,/* First item in the array. */
261 int p_n_num, /* Number of items in the array. */
262 int p_n_width, /* Item size in the array.
263 searched. Lest the reader be
264 confused, note that this is crafted
265 as a general function, and when it
266 is applied specifically to the array
267 of item headers in a node, p_n_width
268 is actually the item header size not
270 int * p_n_pos /* Number of the searched for element. */
272 int n_rbound, n_lbound, n_j;
274 for ( n_j = ((n_rbound = p_n_num - 1) + (n_lbound = 0))/2; n_lbound <= n_rbound; n_j = (n_rbound + n_lbound)/2 )
275 switch( COMP_KEYS((struct key *)((char * )p_v_base + n_j * p_n_width), (struct cpu_key *)p_v_key) ) {
276 case -1: n_lbound = n_j + 1; continue;
277 case 1: n_rbound = n_j - 1; continue;
278 case 0: *p_n_pos = n_j; return ITEM_FOUND; /* Key found in the array. */
281 /* bin_search did not find given key, it returns position of key,
282 that is minimal and greater than the given one. */
284 return ITEM_NOT_FOUND;
287 #ifdef CONFIG_REISERFS_CHECK
288 extern struct tree_balance * cur_tb;
293 /* Minimal possible key. It is never in the tree. */
294 const struct key MIN_KEY = {0, 0, {{0, 0},}};
296 /* Maximal possible key. It is never in the tree. */
297 const struct key MAX_KEY = {0xffffffff, 0xffffffff, {{0xffffffff, 0xffffffff},}};
300 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
301 of the path, and going upwards. We must check the path's validity at each step. If the key is not in
302 the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
303 case we return a special key, either MIN_KEY or MAX_KEY. */
304 inline const struct key * get_lkey (
305 const struct path * p_s_chk_path,
306 const struct super_block * p_s_sb
308 int n_position, n_path_offset = p_s_chk_path->path_length;
309 struct buffer_head * p_s_parent;
311 RFALSE( n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
312 "PAP-5010: invalid offset in the path");
314 /* While not higher in path than first element. */
315 while ( n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET ) {
317 RFALSE( ! buffer_uptodate(PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
318 "PAP-5020: parent is not uptodate");
320 /* Parent at the path is not in the tree now. */
321 if ( ! B_IS_IN_TREE(p_s_parent = PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)) )
323 /* Check whether position in the parent is correct. */
324 if ( (n_position = PATH_OFFSET_POSITION(p_s_chk_path, n_path_offset)) > B_NR_ITEMS(p_s_parent) )
326 /* Check whether parent at the path really points to the child. */
327 if ( B_N_CHILD_NUM(p_s_parent, n_position) !=
328 PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset + 1)->b_blocknr )
330 /* Return delimiting key if position in the parent is not equal to zero. */
332 return B_N_PDELIM_KEY(p_s_parent, n_position - 1);
334 /* Return MIN_KEY if we are in the root of the buffer tree. */
335 if ( PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->b_blocknr ==
336 SB_ROOT_BLOCK (p_s_sb) )
342 /* Get delimiting key of the buffer at the path and its right neighbor. */
343 inline const struct key * get_rkey (
344 const struct path * p_s_chk_path,
345 const struct super_block * p_s_sb
348 n_path_offset = p_s_chk_path->path_length;
349 struct buffer_head * p_s_parent;
351 RFALSE( n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
352 "PAP-5030: invalid offset in the path");
354 while ( n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET ) {
356 RFALSE( ! buffer_uptodate(PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
357 "PAP-5040: parent is not uptodate");
359 /* Parent at the path is not in the tree now. */
360 if ( ! B_IS_IN_TREE(p_s_parent = PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)) )
362 /* Check whether position in the parent is correct. */
363 if ( (n_position = PATH_OFFSET_POSITION(p_s_chk_path, n_path_offset)) > B_NR_ITEMS(p_s_parent) )
365 /* Check whether parent at the path really points to the child. */
366 if ( B_N_CHILD_NUM(p_s_parent, n_position) !=
367 PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset + 1)->b_blocknr )
369 /* Return delimiting key if position in the parent is not the last one. */
370 if ( n_position != B_NR_ITEMS(p_s_parent) )
371 return B_N_PDELIM_KEY(p_s_parent, n_position);
373 /* Return MAX_KEY if we are in the root of the buffer tree. */
374 if ( PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->b_blocknr ==
375 SB_ROOT_BLOCK (p_s_sb) )
381 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
382 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
383 the path. These delimiting keys are stored at least one level above that buffer in the tree. If the
384 buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
385 this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
386 static inline int key_in_buffer (
387 struct path * p_s_chk_path, /* Path which should be checked. */
388 const struct cpu_key * p_s_key, /* Key which should be checked. */
389 struct super_block * p_s_sb /* Super block pointer. */
392 RFALSE( ! p_s_key || p_s_chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET ||
393 p_s_chk_path->path_length > MAX_HEIGHT,
394 "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
395 p_s_key, p_s_chk_path->path_length);
396 RFALSE( !PATH_PLAST_BUFFER(p_s_chk_path)->b_bdev,
397 "PAP-5060: device must not be NODEV");
399 if ( COMP_KEYS(get_lkey(p_s_chk_path, p_s_sb), p_s_key) == 1 )
400 /* left delimiting key is bigger, that the key we look for */
402 // if ( COMP_KEYS(p_s_key, get_rkey(p_s_chk_path, p_s_sb)) != -1 )
403 if ( COMP_KEYS(get_rkey(p_s_chk_path, p_s_sb), p_s_key) != 1 )
404 /* p_s_key must be less than right delimitiing key */
410 inline void decrement_bcount(
411 struct buffer_head * p_s_bh
414 if ( atomic_read (&(p_s_bh->b_count)) ) {
418 reiserfs_panic(NULL, "PAP-5070: decrement_bcount: trying to free free buffer %b", p_s_bh);
423 /* Decrement b_count field of the all buffers in the path. */
424 void decrement_counters_in_path (
425 struct path * p_s_search_path
427 int n_path_offset = p_s_search_path->path_length;
429 RFALSE( n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET ||
430 n_path_offset > EXTENDED_MAX_HEIGHT - 1,
431 "PAP-5080: invalid path offset of %d", n_path_offset);
433 while ( n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET ) {
434 struct buffer_head * bh;
436 bh = PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--);
437 decrement_bcount (bh);
439 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
443 int reiserfs_check_path(struct path *p) {
444 RFALSE( p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
445 "path not properly relsed") ;
450 /* Release all buffers in the path. Restore dirty bits clean
451 ** when preparing the buffer for the log
453 ** only called from fix_nodes()
455 void pathrelse_and_restore (
456 struct super_block *s,
457 struct path * p_s_search_path
459 int n_path_offset = p_s_search_path->path_length;
461 RFALSE( n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
462 "clm-4000: invalid path offset");
464 while ( n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET ) {
465 reiserfs_restore_prepared_buffer(s, PATH_OFFSET_PBUFFER(p_s_search_path,
467 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
469 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
472 /* Release all buffers in the path. */
474 struct path * p_s_search_path
476 int n_path_offset = p_s_search_path->path_length;
478 RFALSE( n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
479 "PAP-5090: invalid path offset");
481 while ( n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET )
482 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
484 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
489 static int is_leaf (char * buf, int blocksize, struct buffer_head * bh)
491 struct block_head * blkh;
492 struct item_head * ih;
498 blkh = (struct block_head *)buf;
499 if ( blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
500 reiserfs_warning (NULL, "is_leaf: this should be caught earlier");
504 nr = blkh_nr_item(blkh);
505 if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
506 /* item number is too big or too small */
507 reiserfs_warning (NULL, "is_leaf: nr_item seems wrong: %z", bh);
510 ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
511 used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location (ih));
512 if (used_space != blocksize - blkh_free_space(blkh)) {
513 /* free space does not match to calculated amount of use space */
514 reiserfs_warning (NULL, "is_leaf: free space seems wrong: %z", bh);
518 // FIXME: it is_leaf will hit performance too much - we may have
521 /* check tables of item heads */
522 ih = (struct item_head *)(buf + BLKH_SIZE);
523 prev_location = blocksize;
524 for (i = 0; i < nr; i ++, ih ++) {
525 if ( le_ih_k_type(ih) == TYPE_ANY) {
526 reiserfs_warning (NULL, "is_leaf: wrong item type for item %h",ih);
529 if (ih_location (ih) >= blocksize || ih_location (ih) < IH_SIZE * nr) {
530 reiserfs_warning (NULL, "is_leaf: item location seems wrong: %h", ih);
533 if (ih_item_len (ih) < 1 || ih_item_len (ih) > MAX_ITEM_LEN (blocksize)) {
534 reiserfs_warning (NULL, "is_leaf: item length seems wrong: %h", ih);
537 if (prev_location - ih_location (ih) != ih_item_len (ih)) {
538 reiserfs_warning (NULL, "is_leaf: item location seems wrong (second one): %h", ih);
541 prev_location = ih_location (ih);
544 // one may imagine much more checks
549 /* returns 1 if buf looks like an internal node, 0 otherwise */
550 static int is_internal (char * buf, int blocksize, struct buffer_head * bh)
552 struct block_head * blkh;
556 blkh = (struct block_head *)buf;
557 nr = blkh_level(blkh);
558 if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
559 /* this level is not possible for internal nodes */
560 reiserfs_warning (NULL, "is_internal: this should be caught earlier");
564 nr = blkh_nr_item(blkh);
565 if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
566 /* for internal which is not root we might check min number of keys */
567 reiserfs_warning (NULL, "is_internal: number of key seems wrong: %z", bh);
571 used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
572 if (used_space != blocksize - blkh_free_space(blkh)) {
573 reiserfs_warning (NULL, "is_internal: free space seems wrong: %z", bh);
577 // one may imagine much more checks
582 // make sure that bh contains formatted node of reiserfs tree of
584 static int is_tree_node (struct buffer_head * bh, int level)
586 if (B_LEVEL (bh) != level) {
587 reiserfs_warning (NULL, "is_tree_node: node level %d does not match to the expected one %d",
588 B_LEVEL (bh), level);
591 if (level == DISK_LEAF_NODE_LEVEL)
592 return is_leaf (bh->b_data, bh->b_size, bh);
594 return is_internal (bh->b_data, bh->b_size, bh);
599 #ifdef SEARCH_BY_KEY_READA
601 /* The function is NOT SCHEDULE-SAFE! */
602 static void search_by_key_reada (struct super_block * s, int blocknr)
604 struct buffer_head * bh;
609 bh = sb_getblk (s, blocknr);
611 if (!buffer_uptodate (bh)) {
612 ll_rw_block (READA, 1, &bh);
619 /**************************************************************************
620 * Algorithm SearchByKey *
621 * look for item in the Disk S+Tree by its key *
622 * Input: p_s_sb - super block *
623 * p_s_key - pointer to the key to search *
624 * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR *
625 * p_s_search_path - path from the root to the needed leaf *
626 **************************************************************************/
628 /* This function fills up the path from the root to the leaf as it
629 descends the tree looking for the key. It uses reiserfs_bread to
630 try to find buffers in the cache given their block number. If it
631 does not find them in the cache it reads them from disk. For each
632 node search_by_key finds using reiserfs_bread it then uses
633 bin_search to look through that node. bin_search will find the
634 position of the block_number of the next node if it is looking
635 through an internal node. If it is looking through a leaf node
636 bin_search will find the position of the item which has key either
637 equal to given key, or which is the maximal key less than the given
638 key. search_by_key returns a path that must be checked for the
639 correctness of the top of the path but need not be checked for the
640 correctness of the bottom of the path */
641 /* The function is NOT SCHEDULE-SAFE! */
642 int search_by_key (struct super_block * p_s_sb,
643 const struct cpu_key * p_s_key, /* Key to search. */
644 struct path * p_s_search_path, /* This structure was
645 allocated and initialized
647 function. It is filled up
649 int n_stop_level /* How far down the tree to search. To
650 stop at leaf level - set to
651 DISK_LEAF_NODE_LEVEL */
655 struct buffer_head * p_s_bh;
656 struct path_element * p_s_last_element;
657 int n_node_level, n_retval;
658 int right_neighbor_of_leaf_node;
661 #ifdef CONFIG_REISERFS_CHECK
662 int n_repeat_counter = 0;
665 PROC_INFO_INC( p_s_sb, search_by_key );
667 /* As we add each node to a path we increase its count. This means that
668 we must be careful to release all nodes in a path before we either
669 discard the path struct or re-use the path struct, as we do here. */
671 decrement_counters_in_path(p_s_search_path);
673 right_neighbor_of_leaf_node = 0;
675 /* With each iteration of this loop we search through the items in the
676 current node, and calculate the next current node(next path element)
677 for the next iteration of this loop.. */
678 n_block_number = SB_ROOT_BLOCK (p_s_sb);
682 #ifdef CONFIG_REISERFS_CHECK
683 if ( !(++n_repeat_counter % 50000) )
684 reiserfs_warning (p_s_sb, "PAP-5100: search_by_key: %s:"
685 "there were %d iterations of while loop "
686 "looking for key %K",
687 current->comm, n_repeat_counter, p_s_key);
690 /* prep path to have another element added to it. */
691 p_s_last_element = PATH_OFFSET_PELEMENT(p_s_search_path, ++p_s_search_path->path_length);
692 fs_gen = get_generation (p_s_sb);
694 #ifdef SEARCH_BY_KEY_READA
695 /* schedule read of right neighbor */
696 search_by_key_reada (p_s_sb, right_neighbor_of_leaf_node);
699 /* Read the next tree node, and set the last element in the path to
700 have a pointer to it. */
701 if ( ! (p_s_bh = p_s_last_element->pe_buffer =
702 sb_bread(p_s_sb, n_block_number)) ) {
703 p_s_search_path->path_length --;
704 pathrelse(p_s_search_path);
707 if (expected_level == -1)
708 expected_level = SB_TREE_HEIGHT (p_s_sb);
711 /* It is possible that schedule occurred. We must check whether the key
712 to search is still in the tree rooted from the current buffer. If
713 not then repeat search from the root. */
714 if ( fs_changed (fs_gen, p_s_sb) &&
715 (!B_IS_IN_TREE (p_s_bh) ||
716 B_LEVEL(p_s_bh) != expected_level ||
717 !key_in_buffer(p_s_search_path, p_s_key, p_s_sb))) {
718 PROC_INFO_INC( p_s_sb, search_by_key_fs_changed );
719 PROC_INFO_INC( p_s_sb, search_by_key_restarted );
720 PROC_INFO_INC( p_s_sb, sbk_restarted[ expected_level - 1 ] );
721 decrement_counters_in_path(p_s_search_path);
723 /* Get the root block number so that we can repeat the search
724 starting from the root. */
725 n_block_number = SB_ROOT_BLOCK (p_s_sb);
727 right_neighbor_of_leaf_node = 0;
729 /* repeat search from the root */
733 /* only check that the key is in the buffer if p_s_key is not
734 equal to the MAX_KEY. Latter case is only possible in
735 "finish_unfinished()" processing during mount. */
736 RFALSE( COMP_KEYS( &MAX_KEY, p_s_key ) &&
737 ! key_in_buffer(p_s_search_path, p_s_key, p_s_sb),
738 "PAP-5130: key is not in the buffer");
739 #ifdef CONFIG_REISERFS_CHECK
741 print_cur_tb ("5140");
742 reiserfs_panic(p_s_sb, "PAP-5140: search_by_key: schedule occurred in do_balance!");
746 // make sure, that the node contents look like a node of
748 if (!is_tree_node (p_s_bh, expected_level)) {
749 reiserfs_warning (p_s_sb, "vs-5150: search_by_key: "
750 "invalid format found in block %ld. Fsck?",
752 pathrelse (p_s_search_path);
756 /* ok, we have acquired next formatted node in the tree */
757 n_node_level = B_LEVEL (p_s_bh);
759 PROC_INFO_BH_STAT( p_s_sb, p_s_bh, n_node_level - 1 );
761 RFALSE( n_node_level < n_stop_level,
762 "vs-5152: tree level (%d) is less than stop level (%d)",
763 n_node_level, n_stop_level);
765 n_retval = bin_search( p_s_key, B_N_PITEM_HEAD(p_s_bh, 0),
767 ( n_node_level == DISK_LEAF_NODE_LEVEL ) ? IH_SIZE : KEY_SIZE,
768 &(p_s_last_element->pe_position));
769 if (n_node_level == n_stop_level) {
773 /* we are not in the stop level */
774 if (n_retval == ITEM_FOUND)
775 /* item has been found, so we choose the pointer which is to the right of the found one */
776 p_s_last_element->pe_position++;
778 /* if item was not found we choose the position which is to
779 the left of the found item. This requires no code,
780 bin_search did it already.*/
782 /* So we have chosen a position in the current node which is
783 an internal node. Now we calculate child block number by
784 position in the node. */
785 n_block_number = B_N_CHILD_NUM(p_s_bh, p_s_last_element->pe_position);
787 #ifdef SEARCH_BY_KEY_READA
788 /* if we are going to read leaf node, then calculate its right neighbor if possible */
789 if (n_node_level == DISK_LEAF_NODE_LEVEL + 1 && p_s_last_element->pe_position < B_NR_ITEMS (p_s_bh))
790 right_neighbor_of_leaf_node = B_N_CHILD_NUM(p_s_bh, p_s_last_element->pe_position + 1);
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 path * p_s_search_path /* Filled up by this function. */
816 struct item_head * p_le_ih; /* pointer to on-disk structure */
818 loff_t item_offset, offset;
819 struct reiserfs_dir_entry de;
822 /* If searching for directory entry. */
823 if ( is_direntry_cpu_key (p_cpu_key) )
824 return search_by_entry_key (p_s_sb, p_cpu_key, p_s_search_path, &de);
826 /* If not searching for directory entry. */
828 /* If item is found. */
829 retval = search_item (p_s_sb, p_cpu_key, p_s_search_path);
830 if (retval == IO_ERROR)
832 if ( retval == ITEM_FOUND ) {
834 RFALSE( ! ih_item_len(
835 B_N_PITEM_HEAD(PATH_PLAST_BUFFER(p_s_search_path),
836 PATH_LAST_POSITION(p_s_search_path))),
837 "PAP-5165: item length equals zero");
839 pos_in_item(p_s_search_path) = 0;
840 return POSITION_FOUND;
843 RFALSE( ! PATH_LAST_POSITION(p_s_search_path),
844 "PAP-5170: position equals zero");
846 /* Item is not found. Set path to the previous item. */
847 p_le_ih = B_N_PITEM_HEAD(PATH_PLAST_BUFFER(p_s_search_path), --PATH_LAST_POSITION(p_s_search_path));
848 n_blk_size = p_s_sb->s_blocksize;
850 if (comp_short_keys (&(p_le_ih->ih_key), p_cpu_key)) {
851 return FILE_NOT_FOUND;
854 // FIXME: quite ugly this far
856 item_offset = le_ih_k_offset (p_le_ih);
857 offset = cpu_key_k_offset (p_cpu_key);
859 /* Needed byte is contained in the item pointed to by the path.*/
860 if (item_offset <= offset &&
861 item_offset + op_bytes_number (p_le_ih, n_blk_size) > offset) {
862 pos_in_item (p_s_search_path) = offset - item_offset;
863 if ( is_indirect_le_ih(p_le_ih) ) {
864 pos_in_item (p_s_search_path) /= n_blk_size;
866 return POSITION_FOUND;
869 /* Needed byte is not contained in the item pointed to by the
870 path. Set pos_in_item out of the item. */
871 if ( is_indirect_le_ih (p_le_ih) )
872 pos_in_item (p_s_search_path) = ih_item_len(p_le_ih) / UNFM_P_SIZE;
874 pos_in_item (p_s_search_path) = ih_item_len( p_le_ih );
876 return POSITION_NOT_FOUND;
880 /* Compare given item and item pointed to by the path. */
881 int comp_items (const struct item_head * stored_ih, const struct path * p_s_path)
883 struct buffer_head * p_s_bh;
884 struct item_head * ih;
886 /* Last buffer at the path is not in the tree. */
887 if ( ! B_IS_IN_TREE(p_s_bh = PATH_PLAST_BUFFER(p_s_path)) )
890 /* Last path position is invalid. */
891 if ( PATH_LAST_POSITION(p_s_path) >= B_NR_ITEMS(p_s_bh) )
894 /* we need only to know, whether it is the same item */
895 ih = get_ih (p_s_path);
896 return memcmp (stored_ih, ih, IH_SIZE);
900 /* unformatted nodes are not logged anymore, ever. This is safe
903 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
905 // block can not be forgotten as it is in I/O or held by someone
906 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
910 // prepare for delete or cut of direct item
911 static inline int prepare_for_direct_item (struct path * path,
912 struct item_head * le_ih,
913 struct inode * inode,
914 loff_t new_file_length,
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));
926 // new file gets truncated
927 if (get_inode_item_key_version (inode) == KEY_FORMAT_3_6) {
929 round_len = ROUND_UP (new_file_length);
930 /* this was n_new_file_length < le_ih ... */
931 if ( round_len < le_ih_k_offset (le_ih) ) {
932 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
933 return M_DELETE; /* Delete this item. */
935 /* Calculate first position and size for cutting from item. */
936 pos_in_item (path) = round_len - (le_ih_k_offset (le_ih) - 1);
937 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
939 return M_CUT; /* Cut from this item. */
943 // old file: items may have any length
945 if ( new_file_length < le_ih_k_offset (le_ih) ) {
946 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
947 return M_DELETE; /* Delete this item. */
949 /* Calculate first position and size for cutting from item. */
950 *cut_size = -(ih_item_len(le_ih) -
951 (pos_in_item (path) = new_file_length + 1 - le_ih_k_offset (le_ih)));
952 return M_CUT; /* Cut from this item. */
956 static inline int prepare_for_direntry_item (struct path * path,
957 struct item_head * le_ih,
958 struct inode * inode,
959 loff_t new_file_length,
962 if (le_ih_k_offset (le_ih) == DOT_OFFSET &&
963 new_file_length == max_reiserfs_offset (inode)) {
964 RFALSE( ih_entry_count (le_ih) != 2,
965 "PAP-5220: incorrect empty directory item (%h)", le_ih);
966 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
967 return M_DELETE; /* Delete the directory item containing "." and ".." entry. */
970 if ( ih_entry_count (le_ih) == 1 ) {
971 /* Delete the directory item such as there is one record only
973 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
977 /* Cut one record from the directory item. */
978 *cut_size = -(DEH_SIZE + entry_length (get_last_bh (path), le_ih, pos_in_item (path)));
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(
989 struct reiserfs_transaction_handle *th,
990 struct inode * inode,
991 struct path * p_s_path,
992 const struct cpu_key * p_s_item_key,
993 int * p_n_removed, /* Number of unformatted nodes which were removed
994 from end of the file. */
996 unsigned long long n_new_file_length /* MAX_KEY_OFFSET in case of delete. */
998 struct super_block * p_s_sb = inode->i_sb;
999 struct item_head * p_le_ih = PATH_PITEM_HEAD(p_s_path);
1000 struct buffer_head * p_s_bh = PATH_PLAST_BUFFER(p_s_path);
1002 /* Stat_data item. */
1003 if ( is_statdata_le_ih (p_le_ih) ) {
1005 RFALSE( n_new_file_length != max_reiserfs_offset (inode),
1006 "PAP-5210: mode must be M_DELETE");
1008 *p_n_cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
1013 /* Directory item. */
1014 if ( is_direntry_le_ih (p_le_ih) )
1015 return prepare_for_direntry_item (p_s_path, p_le_ih, inode, n_new_file_length, p_n_cut_size);
1018 if ( is_direct_le_ih (p_le_ih) )
1019 return prepare_for_direct_item (p_s_path, p_le_ih, inode, n_new_file_length, p_n_cut_size);
1022 /* Case of an indirect item. */
1024 int n_unfm_number, /* Number of the item unformatted nodes. */
1027 __u32 * p_n_unfm_pointer; /* Pointer to the unformatted node number. */
1029 struct item_head s_ih; /* Item header. */
1030 char c_mode; /* Returned mode of the balance. */
1034 n_blk_size = p_s_sb->s_blocksize;
1036 /* Search for the needed object indirect item until there are no unformatted nodes to be removed. */
1039 p_s_bh = PATH_PLAST_BUFFER(p_s_path);
1040 /* Copy indirect item header to a temp variable. */
1041 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1042 /* Calculate number of unformatted nodes in this item. */
1043 n_unfm_number = I_UNFM_NUM(&s_ih);
1045 RFALSE( ! is_indirect_le_ih(&s_ih) || ! n_unfm_number ||
1046 pos_in_item (p_s_path) + 1 != n_unfm_number,
1047 "PAP-5240: invalid item %h "
1048 "n_unfm_number = %d *p_n_pos_in_item = %d",
1049 &s_ih, n_unfm_number, pos_in_item (p_s_path));
1051 /* Calculate balance mode and position in the item to remove unformatted nodes. */
1052 if ( n_new_file_length == max_reiserfs_offset (inode) ) {/* Case of delete. */
1053 pos_in_item (p_s_path) = 0;
1054 *p_n_cut_size = -(IH_SIZE + ih_item_len(&s_ih));
1057 else { /* Case of truncate. */
1058 if ( n_new_file_length < le_ih_k_offset (&s_ih) ) {
1059 pos_in_item (p_s_path) = 0;
1060 *p_n_cut_size = -(IH_SIZE + ih_item_len(&s_ih));
1061 c_mode = M_DELETE; /* Delete this item. */
1064 /* indirect item must be truncated starting from *p_n_pos_in_item-th position */
1065 pos_in_item (p_s_path) = (n_new_file_length + n_blk_size - le_ih_k_offset (&s_ih) ) >> p_s_sb->s_blocksize_bits;
1067 RFALSE( pos_in_item (p_s_path) > n_unfm_number,
1068 "PAP-5250: invalid position in the item");
1070 /* Either convert last unformatted node of indirect item to direct item or increase
1072 if ( pos_in_item (p_s_path) == n_unfm_number ) {
1073 *p_n_cut_size = 0; /* Nothing to cut. */
1074 return M_CONVERT; /* Maybe convert last unformatted node to the direct item. */
1076 /* Calculate size to cut. */
1077 *p_n_cut_size = -(ih_item_len(&s_ih) - pos_in_item(p_s_path) * UNFM_P_SIZE);
1079 c_mode = M_CUT; /* Cut from this indirect item. */
1083 RFALSE( n_unfm_number <= pos_in_item (p_s_path),
1084 "PAP-5260: invalid position in the indirect item");
1086 /* pointers to be cut */
1087 n_unfm_number -= pos_in_item (p_s_path);
1088 /* Set pointer to the last unformatted node pointer that is to be cut. */
1089 p_n_unfm_pointer = (__u32 *)B_I_PITEM(p_s_bh, &s_ih) + I_UNFM_NUM(&s_ih) - 1 - *p_n_removed;
1092 /* We go through the unformatted nodes pointers of the indirect
1093 item and look for the unformatted nodes in the cache. If we
1094 found some of them we free it, zero corresponding indirect item
1095 entry and log buffer containing that indirect item. For this we
1096 need to prepare last path element for logging. If some
1097 unformatted node has b_count > 1 we must not free this
1098 unformatted node since it is in use. */
1099 reiserfs_prepare_for_journal(p_s_sb, p_s_bh, 1);
1100 // note: path could be changed, first line in for loop takes care
1103 for (n_counter = *p_n_removed;
1104 n_counter < n_unfm_number; n_counter++, p_n_unfm_pointer-- ) {
1107 if (item_moved (&s_ih, p_s_path)) {
1111 RFALSE( p_n_unfm_pointer < (__u32 *)B_I_PITEM(p_s_bh, &s_ih) ||
1112 p_n_unfm_pointer > (__u32 *)B_I_PITEM(p_s_bh, &s_ih) + I_UNFM_NUM(&s_ih) - 1,
1113 "vs-5265: pointer out of range");
1115 /* Hole, nothing to remove. */
1116 if ( ! get_block_num(p_n_unfm_pointer,0) ) {
1123 tmp = get_block_num(p_n_unfm_pointer,0);
1124 put_block_num(p_n_unfm_pointer, 0, 0);
1125 journal_mark_dirty (th, p_s_sb, p_s_bh);
1126 reiserfs_free_block(th, inode, tmp, 1);
1127 if ( item_moved (&s_ih, p_s_path) ) {
1133 /* a trick. If the buffer has been logged, this
1134 ** will do nothing. If we've broken the loop without
1135 ** logging it, it will restore the buffer
1138 reiserfs_restore_prepared_buffer(p_s_sb, p_s_bh);
1140 /* This loop can be optimized. */
1141 } while ( (*p_n_removed < n_unfm_number || need_research) &&
1142 search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path) == POSITION_FOUND );
1144 RFALSE( *p_n_removed < n_unfm_number,
1145 "PAP-5310: indirect item is not found");
1146 RFALSE( item_moved (&s_ih, p_s_path),
1147 "after while, comp failed, retry") ;
1149 if (c_mode == M_CUT)
1150 pos_in_item (p_s_path) *= UNFM_P_SIZE;
1155 /* Calculate number of bytes which will be deleted or cut during balance */
1156 int calc_deleted_bytes_number(
1157 struct tree_balance * p_s_tb,
1161 struct item_head * p_le_ih = PATH_PITEM_HEAD(p_s_tb->tb_path);
1163 if ( is_statdata_le_ih (p_le_ih) )
1166 n_del_size = ( c_mode == M_DELETE ) ? ih_item_len(p_le_ih) : -p_s_tb->insert_size[0];
1167 if ( is_direntry_le_ih (p_le_ih) ) {
1168 // return EMPTY_DIR_SIZE; /* We delete emty directoris only. */
1169 // we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1170 // empty size. ick. FIXME, is this right?
1175 if ( is_indirect_le_ih (p_le_ih) )
1176 n_del_size = (n_del_size/UNFM_P_SIZE)*
1177 (PATH_PLAST_BUFFER(p_s_tb->tb_path)->b_size);// - get_ih_free_space (p_le_ih);
1181 static void init_tb_struct(
1182 struct reiserfs_transaction_handle *th,
1183 struct tree_balance * p_s_tb,
1184 struct super_block * p_s_sb,
1185 struct path * p_s_path,
1188 memset (p_s_tb,'\0',sizeof(struct tree_balance));
1189 p_s_tb->transaction_handle = th ;
1190 p_s_tb->tb_sb = p_s_sb;
1191 p_s_tb->tb_path = p_s_path;
1192 PATH_OFFSET_PBUFFER(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1193 PATH_OFFSET_POSITION(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1194 p_s_tb->insert_size[0] = n_size;
1199 void padd_item (char * item, int total_length, int length)
1203 for (i = total_length; i > length; )
1207 #ifdef REISERQUOTA_DEBUG
1208 char key2type(struct key *ih)
1210 if (is_direntry_le_key(2, ih))
1212 if (is_direct_le_key(2, ih))
1214 if (is_indirect_le_key(2, ih))
1216 if (is_statdata_le_key(2, ih))
1221 char head2type(struct item_head *ih)
1223 if (is_direntry_le_ih(ih))
1225 if (is_direct_le_ih(ih))
1227 if (is_indirect_le_ih(ih))
1229 if (is_statdata_le_ih(ih))
1235 /* Delete object item. */
1236 int reiserfs_delete_item (struct reiserfs_transaction_handle *th,
1237 struct path * p_s_path, /* Path to the deleted item. */
1238 const struct cpu_key * p_s_item_key, /* Key to search for the deleted item. */
1239 struct inode * p_s_inode,/* inode is here just to update i_blocks and quotas */
1240 struct buffer_head * p_s_un_bh) /* NULL or unformatted node pointer. */
1242 struct super_block * p_s_sb = p_s_inode->i_sb;
1243 struct tree_balance s_del_balance;
1244 struct item_head s_ih;
1245 struct item_head *q_ih;
1246 int quota_cut_bytes;
1251 #ifdef CONFIG_REISERFS_CHECK
1256 init_tb_struct(th, &s_del_balance, p_s_sb, p_s_path, 0/*size is unknown*/);
1261 #ifdef CONFIG_REISERFS_CHECK
1265 prepare_for_delete_or_cut(th, p_s_inode, p_s_path, p_s_item_key, &n_removed, &n_del_size, max_reiserfs_offset (p_s_inode));
1267 RFALSE( c_mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1269 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1270 s_del_balance.insert_size[0] = n_del_size;
1272 n_ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, 0);
1273 if ( n_ret_value != REPEAT_SEARCH )
1276 PROC_INFO_INC( p_s_sb, delete_item_restarted );
1278 // file system changed, repeat search
1279 n_ret_value = search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1280 if (n_ret_value == IO_ERROR)
1282 if (n_ret_value == FILE_NOT_FOUND) {
1283 reiserfs_warning (p_s_sb, "vs-5340: reiserfs_delete_item: "
1284 "no items of the file %K found", p_s_item_key);
1289 if ( n_ret_value != CARRY_ON ) {
1290 unfix_nodes(&s_del_balance);
1294 // reiserfs_delete_item returns item length when success
1295 n_ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1296 q_ih = get_ih(p_s_path) ;
1297 quota_cut_bytes = ih_item_len(q_ih) ;
1299 /* hack so the quota code doesn't have to guess if the file
1300 ** has a tail. On tail insert, we allocate quota for 1 unformatted node.
1301 ** We test the offset because the tail might have been
1302 ** split into multiple items, and we only want to decrement for
1303 ** the unfm node once
1305 if (!S_ISLNK (p_s_inode->i_mode) && is_direct_le_ih(q_ih)) {
1306 if ((le_ih_k_offset(q_ih) & (p_s_sb->s_blocksize - 1)) == 1) {
1307 quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1309 quota_cut_bytes = 0 ;
1317 /* We are in direct2indirect conversion, so move tail contents
1318 to the unformatted node */
1319 /* note, we do the copy before preparing the buffer because we
1320 ** don't care about the contents of the unformatted node yet.
1321 ** the only thing we really care about is the direct item's data
1322 ** is in the unformatted node.
1324 ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1325 ** the unformatted node, which might schedule, meaning we'd have to
1326 ** loop all the way back up to the start of the while loop.
1328 ** The unformatted node must be dirtied later on. We can't be
1329 ** sure here if the entire tail has been deleted yet.
1331 ** p_s_un_bh is from the page cache (all unformatted nodes are
1332 ** from the page cache) and might be a highmem page. So, we
1333 ** can't use p_s_un_bh->b_data.
1337 data = kmap_atomic(p_s_un_bh->b_page, KM_USER0);
1338 off = ((le_ih_k_offset (&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1340 B_I_PITEM(PATH_PLAST_BUFFER(p_s_path), &s_ih), n_ret_value);
1341 kunmap_atomic(data, KM_USER0);
1343 /* Perform balancing after all resources have been collected at once. */
1344 do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1346 #ifdef REISERQUOTA_DEBUG
1347 reiserfs_debug (p_s_sb, "reiserquota delete_item(): freeing %u, id=%u type=%c", quota_cut_bytes, p_s_inode->i_uid, head2type(&s_ih));
1349 DQUOT_FREE_SPACE_NODIRTY(p_s_inode, quota_cut_bytes);
1351 /* Return deleted body length */
1356 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1358 deletion of the body of the object is performed by iput(), with the
1359 result that if multiple processes are operating on a file, the
1360 deletion of the body of the file is deferred until the last process
1361 that has an open inode performs its iput().
1363 writes and truncates are protected from collisions by use of
1366 creates, linking, and mknod are protected from collisions with other
1367 processes by making the reiserfs_add_entry() the last step in the
1368 creation, and then rolling back all changes if there was a collision.
1373 /* this deletes item which never gets split */
1374 void reiserfs_delete_solid_item (struct reiserfs_transaction_handle *th,
1375 struct inode *inode,
1378 struct tree_balance tb;
1379 INITIALIZE_PATH (path);
1382 struct cpu_key cpu_key;
1384 int quota_cut_bytes = 0;
1386 le_key2cpu_key (&cpu_key, key);
1389 retval = search_item (th->t_super, &cpu_key, &path);
1390 if (retval == IO_ERROR) {
1391 reiserfs_warning (th->t_super,
1392 "vs-5350: reiserfs_delete_solid_item: "
1393 "i/o failure occurred trying to delete %K",
1397 if (retval != ITEM_FOUND) {
1399 // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1400 if ( !( (unsigned long long) GET_HASH_VALUE (le_key_k_offset (le_key_version (key), key)) == 0 && \
1401 (unsigned long long) GET_GENERATION_NUMBER (le_key_k_offset (le_key_version (key), key)) == 1 ) )
1402 reiserfs_warning (th->t_super, "vs-5355: reiserfs_delete_solid_item: %k not found", key);
1407 item_len = ih_item_len( PATH_PITEM_HEAD(&path) );
1408 init_tb_struct (th, &tb, th->t_super, &path, - (IH_SIZE + item_len));
1410 quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path)) ;
1412 retval = fix_nodes (M_DELETE, &tb, NULL, 0);
1413 if (retval == REPEAT_SEARCH) {
1414 PROC_INFO_INC( th -> t_super, delete_solid_item_restarted );
1418 if (retval == CARRY_ON) {
1419 do_balance (&tb, 0, 0, M_DELETE);
1420 if (inode) { /* Should we count quota for item? (we don't count quotas for save-links) */
1421 #ifdef REISERQUOTA_DEBUG
1422 reiserfs_debug (th->t_super, "reiserquota delete_solid_item(): freeing %u id=%u type=%c", quota_cut_bytes, inode->i_uid, key2type(key));
1424 DQUOT_FREE_SPACE_NODIRTY(inode, quota_cut_bytes);
1429 // IO_ERROR, NO_DISK_SPACE, etc
1430 reiserfs_warning (th->t_super, "vs-5360: reiserfs_delete_solid_item: "
1431 "could not delete %K due to fix_nodes failure", &cpu_key);
1436 reiserfs_check_path(&path) ;
1440 void reiserfs_delete_object (struct reiserfs_transaction_handle *th, struct inode * inode)
1444 /* for directory this deletes item containing "." and ".." */
1445 reiserfs_do_truncate (th, inode, NULL, 0/*no timestamp updates*/);
1447 #if defined( USE_INODE_GENERATION_COUNTER )
1448 if( !old_format_only ( th -> t_super ) )
1450 __u32 *inode_generation;
1453 &REISERFS_SB(th -> t_super) -> s_rs -> s_inode_generation;
1454 *inode_generation = cpu_to_le32( le32_to_cpu( *inode_generation ) + 1 );
1456 /* USE_INODE_GENERATION_COUNTER */
1458 reiserfs_delete_solid_item (th, inode, INODE_PKEY (inode));
1462 static int maybe_indirect_to_direct (struct reiserfs_transaction_handle *th,
1463 struct inode * p_s_inode,
1465 struct path * p_s_path,
1466 const struct cpu_key * p_s_item_key,
1467 loff_t n_new_file_size,
1470 struct super_block * p_s_sb = p_s_inode->i_sb;
1471 int n_block_size = p_s_sb->s_blocksize;
1474 if (n_new_file_size != p_s_inode->i_size)
1477 /* the page being sent in could be NULL if there was an i/o error
1478 ** reading in the last block. The user will hit problems trying to
1479 ** read the file, but for now we just skip the indirect2direct
1481 if (atomic_read(&p_s_inode->i_count) > 1 ||
1482 !tail_has_to_be_packed (p_s_inode) ||
1483 !page || (REISERFS_I(p_s_inode)->i_flags & i_nopack_mask)) {
1484 // leave tail in an unformatted node
1485 *p_c_mode = M_SKIP_BALANCING;
1486 cut_bytes = n_block_size - (n_new_file_size & (n_block_size - 1));
1487 pathrelse(p_s_path);
1490 /* Permorm the conversion to a direct_item. */
1491 /*return indirect_to_direct (p_s_inode, p_s_path, p_s_item_key, n_new_file_size, p_c_mode);*/
1492 return indirect2direct (th, p_s_inode, page, p_s_path, p_s_item_key, n_new_file_size, p_c_mode);
1496 /* we did indirect_to_direct conversion. And we have inserted direct
1497 item successesfully, but there were no disk space to cut unfm
1498 pointer being converted. Therefore we have to delete inserted
1500 static void indirect_to_direct_roll_back (struct reiserfs_transaction_handle *th, struct inode * inode, struct path * path)
1502 struct cpu_key tail_key;
1506 make_cpu_key (&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);// !!!!
1507 tail_key.key_length = 4;
1509 tail_len = (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) == POSITION_NOT_FOUND)
1513 reiserfs_panic (inode->i_sb, "vs-5615: indirect_to_direct_roll_back: found invalid item");
1514 RFALSE( path->pos_in_item != ih_item_len(PATH_PITEM_HEAD (path)) - 1,
1515 "vs-5616: appended bytes found");
1516 PATH_LAST_POSITION (path) --;
1518 removed = reiserfs_delete_item (th, path, &tail_key, inode, 0/*unbh not needed*/);
1519 RFALSE( removed <= 0 || removed > tail_len,
1520 "vs-5617: there was tail %d bytes, removed item length %d bytes",
1522 tail_len -= removed;
1523 set_cpu_key_k_offset (&tail_key, cpu_key_k_offset (&tail_key) - removed);
1525 reiserfs_warning (inode->i_sb, "indirect_to_direct_roll_back: indirect_to_direct conversion has been rolled back due to lack of disk space");
1526 //mark_file_without_tail (inode);
1527 mark_inode_dirty (inode);
1531 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1532 int reiserfs_cut_from_item (struct reiserfs_transaction_handle *th,
1533 struct path * p_s_path,
1534 struct cpu_key * p_s_item_key,
1535 struct inode * p_s_inode,
1537 loff_t n_new_file_size)
1539 struct super_block * p_s_sb = p_s_inode->i_sb;
1540 /* Every function which is going to call do_balance must first
1541 create a tree_balance structure. Then it must fill up this
1542 structure by using the init_tb_struct and fix_nodes functions.
1543 After that we can make tree balancing. */
1544 struct tree_balance s_cut_balance;
1545 struct item_head *p_le_ih;
1546 int n_cut_size = 0, /* Amount to be cut. */
1547 n_ret_value = CARRY_ON,
1548 n_removed = 0, /* Number of the removed unformatted nodes. */
1549 n_is_inode_locked = 0;
1550 char c_mode; /* Mode of the balance. */
1552 int quota_cut_bytes;
1555 init_tb_struct(th, &s_cut_balance, p_s_inode->i_sb, p_s_path, n_cut_size);
1558 /* Repeat this loop until we either cut the item without needing
1559 to balance, or we fix_nodes without schedule occurring */
1561 /* Determine the balance mode, position of the first byte to
1562 be cut, and size to be cut. In case of the indirect item
1563 free unformatted nodes which are pointed to by the cut
1566 c_mode = prepare_for_delete_or_cut(th, p_s_inode, p_s_path, p_s_item_key, &n_removed,
1567 &n_cut_size, n_new_file_size);
1568 if ( c_mode == M_CONVERT ) {
1569 /* convert last unformatted node to direct item or leave
1570 tail in the unformatted node */
1571 RFALSE( n_ret_value != CARRY_ON, "PAP-5570: can not convert twice");
1573 n_ret_value = maybe_indirect_to_direct (th, p_s_inode, page, p_s_path, p_s_item_key,
1574 n_new_file_size, &c_mode);
1575 if ( c_mode == M_SKIP_BALANCING )
1576 /* tail has been left in the unformatted node */
1579 n_is_inode_locked = 1;
1581 /* removing of last unformatted node will change value we
1582 have to return to truncate. Save it */
1583 retval2 = n_ret_value;
1584 /*retval2 = p_s_sb->s_blocksize - (n_new_file_size & (p_s_sb->s_blocksize - 1));*/
1586 /* So, we have performed the first part of the conversion:
1587 inserting the new direct item. Now we are removing the
1588 last unformatted node pointer. Set key to search for
1590 set_cpu_key_k_type (p_s_item_key, TYPE_INDIRECT);
1591 p_s_item_key->key_length = 4;
1592 n_new_file_size -= (n_new_file_size & (p_s_sb->s_blocksize - 1));
1593 set_cpu_key_k_offset (p_s_item_key, n_new_file_size + 1);
1594 if ( search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path) == POSITION_NOT_FOUND ){
1595 print_block (PATH_PLAST_BUFFER (p_s_path), 3, PATH_LAST_POSITION (p_s_path) - 1, PATH_LAST_POSITION (p_s_path) + 1);
1596 reiserfs_panic(p_s_sb, "PAP-5580: reiserfs_cut_from_item: item to convert does not exist (%K)", p_s_item_key);
1600 if (n_cut_size == 0) {
1601 pathrelse (p_s_path);
1605 s_cut_balance.insert_size[0] = n_cut_size;
1607 n_ret_value = fix_nodes(c_mode, &s_cut_balance, NULL, 0);
1608 if ( n_ret_value != REPEAT_SEARCH )
1611 PROC_INFO_INC( p_s_sb, cut_from_item_restarted );
1613 n_ret_value = search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1614 if (n_ret_value == POSITION_FOUND)
1617 reiserfs_warning (p_s_sb, "PAP-5610: reiserfs_cut_from_item: item %K not found", p_s_item_key);
1618 unfix_nodes (&s_cut_balance);
1619 return (n_ret_value == IO_ERROR) ? -EIO : -ENOENT;
1622 // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1623 if ( n_ret_value != CARRY_ON ) {
1624 if ( n_is_inode_locked ) {
1625 // FIXME: this seems to be not needed: we are always able
1627 indirect_to_direct_roll_back (th, p_s_inode, p_s_path);
1629 if (n_ret_value == NO_DISK_SPACE)
1630 reiserfs_warning (p_s_sb, "NO_DISK_SPACE");
1631 unfix_nodes (&s_cut_balance);
1635 /* go ahead and perform balancing */
1637 RFALSE( c_mode == M_PASTE || c_mode == M_INSERT, "invalid mode");
1639 /* Calculate number of bytes that need to be cut from the item. */
1640 quota_cut_bytes = ( c_mode == M_DELETE ) ? ih_item_len(get_ih(p_s_path)) : -s_cut_balance.insert_size[0];
1642 n_ret_value = calc_deleted_bytes_number(&s_cut_balance, c_mode);
1644 n_ret_value = retval2;
1647 /* For direct items, we only change the quota when deleting the last
1650 p_le_ih = PATH_PITEM_HEAD (s_cut_balance.tb_path);
1651 if (!S_ISLNK (p_s_inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1652 if (c_mode == M_DELETE &&
1653 (le_ih_k_offset (p_le_ih) & (p_s_sb->s_blocksize - 1)) == 1 ) {
1654 // FIXME: this is to keep 3.5 happy
1655 REISERFS_I(p_s_inode)->i_first_direct_byte = U32_MAX;
1656 quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE ;
1658 quota_cut_bytes = 0 ;
1661 #ifdef CONFIG_REISERFS_CHECK
1662 if (n_is_inode_locked) {
1663 struct item_head * le_ih = PATH_PITEM_HEAD (s_cut_balance.tb_path);
1664 /* we are going to complete indirect2direct conversion. Make
1665 sure, that we exactly remove last unformatted node pointer
1667 if (!is_indirect_le_ih (le_ih))
1668 reiserfs_panic (p_s_sb, "vs-5652: reiserfs_cut_from_item: "
1669 "item must be indirect %h", le_ih);
1671 if (c_mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1672 reiserfs_panic (p_s_sb, "vs-5653: reiserfs_cut_from_item: "
1673 "completing indirect2direct conversion indirect item %h "
1674 "being deleted must be of 4 byte long", le_ih);
1676 if (c_mode == M_CUT && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1677 reiserfs_panic (p_s_sb, "vs-5654: reiserfs_cut_from_item: "
1678 "can not complete indirect2direct conversion of %h (CUT, insert_size==%d)",
1679 le_ih, s_cut_balance.insert_size[0]);
1681 /* it would be useful to make sure, that right neighboring
1682 item is direct item of this file */
1686 do_balance(&s_cut_balance, NULL, NULL, c_mode);
1687 if ( n_is_inode_locked ) {
1688 /* we've done an indirect->direct conversion. when the data block
1689 ** was freed, it was removed from the list of blocks that must
1690 ** be flushed before the transaction commits, so we don't need to
1691 ** deal with it here.
1693 REISERFS_I(p_s_inode)->i_flags &= ~i_pack_on_close_mask ;
1695 #ifdef REISERQUOTA_DEBUG
1696 reiserfs_debug (p_s_inode->i_sb, "reiserquota cut_from_item(): freeing %u id=%u type=%c", quota_cut_bytes, p_s_inode->i_uid, '?');
1698 DQUOT_FREE_SPACE_NODIRTY(p_s_inode, quota_cut_bytes);
1702 static void truncate_directory (struct reiserfs_transaction_handle *th, struct inode * inode)
1705 reiserfs_warning (inode->i_sb,
1706 "vs-5655: truncate_directory: link count != 0");
1708 set_le_key_k_offset (KEY_FORMAT_3_5, INODE_PKEY (inode), DOT_OFFSET);
1709 set_le_key_k_type (KEY_FORMAT_3_5, INODE_PKEY (inode), TYPE_DIRENTRY);
1710 reiserfs_delete_solid_item (th, inode, INODE_PKEY (inode));
1711 reiserfs_update_sd(th, inode) ;
1712 set_le_key_k_offset (KEY_FORMAT_3_5, INODE_PKEY (inode), SD_OFFSET);
1713 set_le_key_k_type (KEY_FORMAT_3_5, INODE_PKEY (inode), TYPE_STAT_DATA);
1719 /* Truncate file to the new size. Note, this must be called with a transaction
1721 void reiserfs_do_truncate (struct reiserfs_transaction_handle *th,
1722 struct inode * p_s_inode, /* ->i_size contains new
1724 struct page *page, /* up to date for last block */
1725 int update_timestamps /* when it is called by
1726 file_release to convert
1727 the tail - no timestamps
1728 should be updated */
1730 INITIALIZE_PATH (s_search_path); /* Path to the current object item. */
1731 struct item_head * p_le_ih; /* Pointer to an item header. */
1732 struct cpu_key s_item_key; /* Key to search for a previous file item. */
1733 loff_t n_file_size, /* Old file size. */
1734 n_new_file_size;/* New file size. */
1735 int n_deleted; /* Number of deleted or truncated bytes. */
1738 if ( ! (S_ISREG(p_s_inode->i_mode) || S_ISDIR(p_s_inode->i_mode) || S_ISLNK(p_s_inode->i_mode)) )
1741 if (S_ISDIR(p_s_inode->i_mode)) {
1742 // deletion of directory - no need to update timestamps
1743 truncate_directory (th, p_s_inode);
1747 /* Get new file size. */
1748 n_new_file_size = p_s_inode->i_size;
1750 // FIXME: note, that key type is unimportant here
1751 make_cpu_key (&s_item_key, p_s_inode, max_reiserfs_offset (p_s_inode), TYPE_DIRECT, 3);
1753 retval = search_for_position_by_key(p_s_inode->i_sb, &s_item_key, &s_search_path);
1754 if (retval == IO_ERROR) {
1755 reiserfs_warning (p_s_inode->i_sb, "vs-5657: reiserfs_do_truncate: "
1756 "i/o failure occurred trying to truncate %K", &s_item_key);
1759 if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1760 pathrelse (&s_search_path);
1761 reiserfs_warning (p_s_inode->i_sb, "PAP-5660: reiserfs_do_truncate: "
1762 "wrong result %d of search for %K", retval, &s_item_key);
1766 s_search_path.pos_in_item --;
1768 /* Get real file size (total length of all file items) */
1769 p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1770 if ( is_statdata_le_ih (p_le_ih) )
1773 loff_t offset = le_ih_k_offset (p_le_ih);
1774 int bytes = op_bytes_number (p_le_ih,p_s_inode->i_sb->s_blocksize);
1776 /* this may mismatch with real file size: if last direct item
1777 had no padding zeros and last unformatted node had no free
1778 space, this file would have this file size */
1779 n_file_size = offset + bytes - 1;
1782 if ( n_file_size == 0 || n_file_size < n_new_file_size ) {
1783 goto update_and_out ;
1786 /* Update key to search for the last file item. */
1787 set_cpu_key_k_offset (&s_item_key, n_file_size);
1790 /* Cut or delete file item. */
1791 n_deleted = reiserfs_cut_from_item(th, &s_search_path, &s_item_key, p_s_inode, page, n_new_file_size);
1792 if (n_deleted < 0) {
1793 reiserfs_warning (p_s_inode->i_sb, "vs-5665: reiserfs_do_truncate: reiserfs_cut_from_item failed");
1794 reiserfs_check_path(&s_search_path) ;
1798 RFALSE( n_deleted > n_file_size,
1799 "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1800 n_deleted, n_file_size, &s_item_key);
1802 /* Change key to search the last file item. */
1803 n_file_size -= n_deleted;
1805 set_cpu_key_k_offset (&s_item_key, n_file_size);
1807 /* While there are bytes to truncate and previous file item is presented in the tree. */
1810 ** This loop could take a really long time, and could log
1811 ** many more blocks than a transaction can hold. So, we do a polite
1812 ** journal end here, and if the transaction needs ending, we make
1813 ** sure the file is consistent before ending the current trans
1814 ** and starting a new one
1816 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1817 int orig_len_alloc = th->t_blocks_allocated ;
1818 decrement_counters_in_path(&s_search_path) ;
1820 if (update_timestamps) {
1821 p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME;
1823 reiserfs_update_sd(th, p_s_inode) ;
1825 journal_end(th, p_s_inode->i_sb, orig_len_alloc) ;
1826 journal_begin(th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 6) ;
1827 reiserfs_update_inode_transaction(p_s_inode) ;
1829 } while ( n_file_size > ROUND_UP (n_new_file_size) &&
1830 search_for_position_by_key(p_s_inode->i_sb, &s_item_key, &s_search_path) == POSITION_FOUND ) ;
1832 RFALSE( n_file_size > ROUND_UP (n_new_file_size),
1833 "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1834 n_new_file_size, n_file_size, s_item_key.on_disk_key.k_objectid);
1837 if (update_timestamps) {
1838 // this is truncate, not file closing
1839 p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME;
1841 reiserfs_update_sd (th, p_s_inode);
1843 pathrelse(&s_search_path) ;
1847 #ifdef CONFIG_REISERFS_CHECK
1848 // this makes sure, that we __append__, not overwrite or add holes
1849 static void check_research_for_paste (struct path * path,
1850 const struct cpu_key * p_s_key)
1852 struct item_head * found_ih = get_ih (path);
1854 if (is_direct_le_ih (found_ih)) {
1855 if (le_ih_k_offset (found_ih) + op_bytes_number (found_ih, get_last_bh (path)->b_size) !=
1856 cpu_key_k_offset (p_s_key) ||
1857 op_bytes_number (found_ih, get_last_bh (path)->b_size) != pos_in_item (path))
1858 reiserfs_panic (0, "PAP-5720: check_research_for_paste: "
1859 "found direct item %h or position (%d) does not match to key %K",
1860 found_ih, pos_in_item (path), p_s_key);
1862 if (is_indirect_le_ih (found_ih)) {
1863 if (le_ih_k_offset (found_ih) + op_bytes_number (found_ih, get_last_bh (path)->b_size) != cpu_key_k_offset (p_s_key) ||
1864 I_UNFM_NUM (found_ih) != pos_in_item (path) ||
1865 get_ih_free_space (found_ih) != 0)
1866 reiserfs_panic (0, "PAP-5730: check_research_for_paste: "
1867 "found indirect item (%h) or position (%d) does not match to key (%K)",
1868 found_ih, pos_in_item (path), p_s_key);
1871 #endif /* config reiserfs check */
1874 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1875 int reiserfs_paste_into_item (struct reiserfs_transaction_handle *th,
1876 struct path * p_s_search_path, /* Path to the pasted item. */
1877 const struct cpu_key * p_s_key, /* Key to search for the needed item.*/
1878 struct inode * inode, /* Inode item belongs to */
1879 const char * p_c_body, /* Pointer to the bytes to paste. */
1880 int n_pasted_size) /* Size of pasted bytes. */
1882 struct tree_balance s_paste_balance;
1886 fs_gen = get_generation(inode->i_sb) ;
1888 #ifdef REISERQUOTA_DEBUG
1889 reiserfs_debug (inode->i_sb, "reiserquota paste_into_item(): allocating %u id=%u type=%c", n_pasted_size, inode->i_uid, key2type(&(p_s_key->on_disk_key)));
1892 if (DQUOT_ALLOC_SPACE_NODIRTY(inode, n_pasted_size)) {
1893 pathrelse(p_s_search_path);
1896 init_tb_struct(th, &s_paste_balance, th->t_super, p_s_search_path, n_pasted_size);
1897 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1898 s_paste_balance.key = p_s_key->on_disk_key;
1901 /* DQUOT_* can schedule, must check before the fix_nodes */
1902 if (fs_changed(fs_gen, inode->i_sb)) {
1906 while ((retval = fix_nodes(M_PASTE, &s_paste_balance, NULL, p_c_body)) ==
1909 /* file system changed while we were in the fix_nodes */
1910 PROC_INFO_INC( th -> t_super, paste_into_item_restarted );
1911 retval = search_for_position_by_key (th->t_super, p_s_key, p_s_search_path);
1912 if (retval == IO_ERROR) {
1916 if (retval == POSITION_FOUND) {
1917 reiserfs_warning (inode->i_sb, "PAP-5710: reiserfs_paste_into_item: entry or pasted byte (%K) exists", p_s_key);
1922 #ifdef CONFIG_REISERFS_CHECK
1923 check_research_for_paste (p_s_search_path, p_s_key);
1927 /* Perform balancing after all resources are collected by fix_nodes, and
1928 accessing them will not risk triggering schedule. */
1929 if ( retval == CARRY_ON ) {
1930 do_balance(&s_paste_balance, NULL/*ih*/, p_c_body, M_PASTE);
1933 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
1935 /* this also releases the path */
1936 unfix_nodes(&s_paste_balance);
1937 #ifdef REISERQUOTA_DEBUG
1938 reiserfs_debug (inode->i_sb, "reiserquota paste_into_item(): freeing %u id=%u type=%c", n_pasted_size, inode->i_uid, key2type(&(p_s_key->on_disk_key)));
1940 DQUOT_FREE_SPACE_NODIRTY(inode, n_pasted_size);
1945 /* Insert new item into the buffer at the path. */
1946 int reiserfs_insert_item(struct reiserfs_transaction_handle *th,
1947 struct path * p_s_path, /* Path to the inserteded item. */
1948 const struct cpu_key * key,
1949 struct item_head * p_s_ih, /* Pointer to the item header to insert.*/
1950 struct inode * inode,
1951 const char * p_c_body) /* Pointer to the bytes to insert. */
1953 struct tree_balance s_ins_balance;
1956 int quota_bytes = 0 ;
1958 if (inode) { /* Do we count quotas for item? */
1959 fs_gen = get_generation(inode->i_sb);
1960 quota_bytes = ih_item_len(p_s_ih);
1962 /* hack so the quota code doesn't have to guess if the file has
1963 ** a tail, links are always tails, so there's no guessing needed
1965 if (!S_ISLNK (inode->i_mode) && is_direct_le_ih(p_s_ih)) {
1966 quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE ;
1968 #ifdef REISERQUOTA_DEBUG
1969 reiserfs_debug (inode->i_sb, "reiserquota insert_item(): allocating %u id=%u type=%c", quota_bytes, inode->i_uid, head2type(p_s_ih));
1971 /* We can't dirty inode here. It would be immediately written but
1972 * appropriate stat item isn't inserted yet... */
1973 if (DQUOT_ALLOC_SPACE_NODIRTY(inode, quota_bytes)) {
1974 pathrelse(p_s_path);
1978 init_tb_struct(th, &s_ins_balance, th->t_super, p_s_path, IH_SIZE + ih_item_len(p_s_ih));
1979 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1980 s_ins_balance.key = key->on_disk_key;
1982 /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
1983 if (inode && fs_changed(fs_gen, inode->i_sb)) {
1987 while ( (retval = fix_nodes(M_INSERT, &s_ins_balance, p_s_ih, p_c_body)) == REPEAT_SEARCH) {
1989 /* file system changed while we were in the fix_nodes */
1990 PROC_INFO_INC( th -> t_super, insert_item_restarted );
1991 retval = search_item (th->t_super, key, p_s_path);
1992 if (retval == IO_ERROR) {
1996 if (retval == ITEM_FOUND) {
1997 reiserfs_warning (th->t_super, "PAP-5760: reiserfs_insert_item: "
1998 "key %K already exists in the tree", key);
2004 /* make balancing after all resources will be collected at a time */
2005 if ( retval == CARRY_ON ) {
2006 do_balance (&s_ins_balance, p_s_ih, p_c_body, M_INSERT);
2010 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2012 /* also releases the path */
2013 unfix_nodes(&s_ins_balance);
2014 #ifdef REISERQUOTA_DEBUG
2015 reiserfs_debug (th->t_super, "reiserquota insert_item(): freeing %u id=%u type=%c", quota_bytes, inode->i_uid, head2type(p_s_ih));
2018 DQUOT_FREE_SPACE_NODIRTY(inode, quota_bytes) ;