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>
64 /* Does the buffer contain a disk block which is in the tree. */
65 inline int B_IS_IN_TREE (const struct buffer_head * p_s_bh)
68 RFALSE( B_LEVEL (p_s_bh) > MAX_HEIGHT,
69 "PAP-1010: block (%b) has too big level (%z)", p_s_bh, p_s_bh);
71 return ( B_LEVEL (p_s_bh) != FREE_LEVEL );
77 inline void copy_short_key (void * to, const void * from)
79 memcpy (to, from, SHORT_KEY_SIZE);
83 // to gets item head in le form
85 inline void copy_item_head(struct item_head * p_v_to,
86 const struct item_head * p_v_from)
88 memcpy (p_v_to, p_v_from, IH_SIZE);
92 /* k1 is pointer to on-disk structure which is stored in little-endian
93 form. k2 is pointer to cpu variable. For key of items of the same
94 object this returns 0.
95 Returns: -1 if key1 < key2
98 inline int comp_short_keys (const struct key * le_key,
99 const struct cpu_key * cpu_key)
101 __u32 * p_s_le_u32, * p_s_cpu_u32;
102 int n_key_length = REISERFS_SHORT_KEY_LEN;
104 p_s_le_u32 = (__u32 *)le_key;
105 p_s_cpu_u32 = (__u32 *)&cpu_key->on_disk_key;
106 for( ; n_key_length--; ++p_s_le_u32, ++p_s_cpu_u32 ) {
107 if ( le32_to_cpu (*p_s_le_u32) < *p_s_cpu_u32 )
109 if ( le32_to_cpu (*p_s_le_u32) > *p_s_cpu_u32 )
117 /* k1 is pointer to on-disk structure which is stored in little-endian
118 form. k2 is pointer to cpu variable.
119 Compare keys using all 4 key fields.
120 Returns: -1 if key1 < key2 0
121 if key1 = key2 1 if key1 > key2 */
122 inline int comp_keys (const struct key * le_key, const struct cpu_key * cpu_key)
126 retval = comp_short_keys (le_key, cpu_key);
129 if (le_key_k_offset (le_key_version(le_key), le_key) < cpu_key_k_offset (cpu_key))
131 if (le_key_k_offset (le_key_version(le_key), le_key) > cpu_key_k_offset (cpu_key))
134 if (cpu_key->key_length == 3)
137 /* this part is needed only when tail conversion is in progress */
138 if (le_key_k_type (le_key_version(le_key), le_key) < cpu_key_k_type (cpu_key))
141 if (le_key_k_type (le_key_version(le_key), le_key) > cpu_key_k_type (cpu_key))
149 // FIXME: not used yet
151 inline int comp_cpu_keys (const struct cpu_key * key1,
152 const struct cpu_key * key2)
154 if (key1->on_disk_key.k_dir_id < key2->on_disk_key.k_dir_id)
156 if (key1->on_disk_key.k_dir_id > key2->on_disk_key.k_dir_id)
159 if (key1->on_disk_key.k_objectid < key2->on_disk_key.k_objectid)
161 if (key1->on_disk_key.k_objectid > key2->on_disk_key.k_objectid)
164 if (cpu_key_k_offset (key1) < cpu_key_k_offset (key2))
166 if (cpu_key_k_offset (key1) > cpu_key_k_offset (key2))
169 reiserfs_warning ("comp_cpu_keys: type are compared for %K and %K\n",
172 if (cpu_key_k_type (key1) < cpu_key_k_type (key2))
174 if (cpu_key_k_type (key1) > cpu_key_k_type (key2))
179 inline int comp_short_le_keys (const struct key * key1, const struct key * key2)
181 __u32 * p_s_1_u32, * p_s_2_u32;
182 int n_key_length = REISERFS_SHORT_KEY_LEN;
184 p_s_1_u32 = (__u32 *)key1;
185 p_s_2_u32 = (__u32 *)key2;
186 for( ; n_key_length--; ++p_s_1_u32, ++p_s_2_u32 ) {
187 if ( le32_to_cpu (*p_s_1_u32) < le32_to_cpu (*p_s_2_u32) )
189 if ( le32_to_cpu (*p_s_1_u32) > le32_to_cpu (*p_s_2_u32) )
195 inline int comp_short_cpu_keys (const struct cpu_key * key1,
196 const struct cpu_key * key2)
198 __u32 * p_s_1_u32, * p_s_2_u32;
199 int n_key_length = REISERFS_SHORT_KEY_LEN;
201 p_s_1_u32 = (__u32 *)key1;
202 p_s_2_u32 = (__u32 *)key2;
204 for( ; n_key_length--; ++p_s_1_u32, ++p_s_2_u32 ) {
205 if ( *p_s_1_u32 < *p_s_2_u32 )
207 if ( *p_s_1_u32 > *p_s_2_u32 )
215 inline void cpu_key2cpu_key (struct cpu_key * to, const struct cpu_key * from)
217 memcpy (to, from, sizeof (struct cpu_key));
221 inline void le_key2cpu_key (struct cpu_key * to, const struct key * from)
223 to->on_disk_key.k_dir_id = le32_to_cpu (from->k_dir_id);
224 to->on_disk_key.k_objectid = le32_to_cpu (from->k_objectid);
226 // find out version of the key
227 to->version = le_key_version (from);
228 if (to->version == KEY_FORMAT_3_5) {
229 to->on_disk_key.u.k_offset_v1.k_offset = le32_to_cpu (from->u.k_offset_v1.k_offset);
230 to->on_disk_key.u.k_offset_v1.k_uniqueness = le32_to_cpu (from->u.k_offset_v1.k_uniqueness);
232 to->on_disk_key.u.k_offset_v2.k_offset = offset_v2_k_offset(&from->u.k_offset_v2);
233 to->on_disk_key.u.k_offset_v2.k_type = offset_v2_k_type(&from->u.k_offset_v2);
239 // this does not say which one is bigger, it only returns 1 if keys
240 // are not equal, 0 otherwise
241 inline int comp_le_keys (const struct key * k1, const struct key * k2)
243 return memcmp (k1, k2, sizeof (struct key));
246 /**************************************************************************
247 * Binary search toolkit function *
248 * Search for an item in the array by the item key *
249 * Returns: 1 if found, 0 if not found; *
250 * *p_n_pos = number of the searched element if found, else the *
251 * number of the first element that is larger than p_v_key. *
252 **************************************************************************/
253 /* For those not familiar with binary search: n_lbound is the leftmost item that it
254 could be, n_rbound the rightmost item that it could be. We examine the item
255 halfway between n_lbound and n_rbound, and that tells us either that we can increase
256 n_lbound, or decrease n_rbound, or that we have found it, or if n_lbound <= n_rbound that
257 there are no possible items, and we have not found it. With each examination we
258 cut the number of possible items it could be by one more than half rounded down,
260 inline int bin_search (
261 const void * p_v_key, /* Key to search for. */
262 const void * p_v_base,/* First item in the array. */
263 int p_n_num, /* Number of items in the array. */
264 int p_n_width, /* Item size in the array.
265 searched. Lest the reader be
266 confused, note that this is crafted
267 as a general function, and when it
268 is applied specifically to the array
269 of item headers in a node, p_n_width
270 is actually the item header size not
272 int * p_n_pos /* Number of the searched for element. */
274 int n_rbound, n_lbound, n_j;
276 for ( n_j = ((n_rbound = p_n_num - 1) + (n_lbound = 0))/2; n_lbound <= n_rbound; n_j = (n_rbound + n_lbound)/2 )
277 switch( COMP_KEYS((struct key *)((char * )p_v_base + n_j * p_n_width), (struct cpu_key *)p_v_key) ) {
278 case -1: n_lbound = n_j + 1; continue;
279 case 1: n_rbound = n_j - 1; continue;
280 case 0: *p_n_pos = n_j; return ITEM_FOUND; /* Key found in the array. */
283 /* bin_search did not find given key, it returns position of key,
284 that is minimal and greater than the given one. */
286 return ITEM_NOT_FOUND;
289 #ifdef CONFIG_REISERFS_CHECK
290 extern struct tree_balance * cur_tb;
295 /* Minimal possible key. It is never in the tree. */
296 const struct key MIN_KEY = {0, 0, {{0, 0},}};
298 /* Maximal possible key. It is never in the tree. */
299 const struct key MAX_KEY = {0xffffffff, 0xffffffff, {{0xffffffff, 0xffffffff},}};
302 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
303 of the path, and going upwards. We must check the path's validity at each step. If the key is not in
304 the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
305 case we return a special key, either MIN_KEY or MAX_KEY. */
306 inline const struct key * get_lkey (
307 const struct path * p_s_chk_path,
308 const struct super_block * p_s_sb
310 int n_position, n_path_offset = p_s_chk_path->path_length;
311 struct buffer_head * p_s_parent;
313 RFALSE( n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
314 "PAP-5010: illegal offset in the path");
316 /* While not higher in path than first element. */
317 while ( n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET ) {
319 RFALSE( ! buffer_uptodate(PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
320 "PAP-5020: parent is not uptodate");
322 /* Parent at the path is not in the tree now. */
323 if ( ! B_IS_IN_TREE(p_s_parent = PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)) )
325 /* Check whether position in the parent is correct. */
326 if ( (n_position = PATH_OFFSET_POSITION(p_s_chk_path, n_path_offset)) > B_NR_ITEMS(p_s_parent) )
328 /* Check whether parent at the path really points to the child. */
329 if ( B_N_CHILD_NUM(p_s_parent, n_position) !=
330 PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset + 1)->b_blocknr )
332 /* Return delimiting key if position in the parent is not equal to zero. */
334 return B_N_PDELIM_KEY(p_s_parent, n_position - 1);
336 /* Return MIN_KEY if we are in the root of the buffer tree. */
337 if ( PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->b_blocknr ==
338 SB_ROOT_BLOCK (p_s_sb) )
344 /* Get delimiting key of the buffer at the path and its right neighbor. */
345 inline const struct key * get_rkey (
346 const struct path * p_s_chk_path,
347 const struct super_block * p_s_sb
350 n_path_offset = p_s_chk_path->path_length;
351 struct buffer_head * p_s_parent;
353 RFALSE( n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
354 "PAP-5030: illegal offset in the path");
356 while ( n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET ) {
358 RFALSE( ! buffer_uptodate(PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
359 "PAP-5040: parent is not uptodate");
361 /* Parent at the path is not in the tree now. */
362 if ( ! B_IS_IN_TREE(p_s_parent = PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)) )
364 /* Check whether position in the parent is correct. */
365 if ( (n_position = PATH_OFFSET_POSITION(p_s_chk_path, n_path_offset)) > B_NR_ITEMS(p_s_parent) )
367 /* Check whether parent at the path really points to the child. */
368 if ( B_N_CHILD_NUM(p_s_parent, n_position) !=
369 PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset + 1)->b_blocknr )
371 /* Return delimiting key if position in the parent is not the last one. */
372 if ( n_position != B_NR_ITEMS(p_s_parent) )
373 return B_N_PDELIM_KEY(p_s_parent, n_position);
375 /* Return MAX_KEY if we are in the root of the buffer tree. */
376 if ( PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->b_blocknr ==
377 SB_ROOT_BLOCK (p_s_sb) )
383 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
384 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
385 the path. These delimiting keys are stored at least one level above that buffer in the tree. If the
386 buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
387 this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
388 static inline int key_in_buffer (
389 struct path * p_s_chk_path, /* Path which should be checked. */
390 const struct cpu_key * p_s_key, /* Key which should be checked. */
391 struct super_block * p_s_sb /* Super block pointer. */
394 RFALSE( ! p_s_key || p_s_chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET ||
395 p_s_chk_path->path_length > MAX_HEIGHT,
396 "PAP-5050: pointer to the key(%p) is NULL or illegal path length(%d)",
397 p_s_key, p_s_chk_path->path_length);
398 RFALSE( !PATH_PLAST_BUFFER(p_s_chk_path)->b_bdev,
399 "PAP-5060: device must not be NODEV");
401 if ( COMP_KEYS(get_lkey(p_s_chk_path, p_s_sb), p_s_key) == 1 )
402 /* left delimiting key is bigger, that the key we look for */
404 // if ( COMP_KEYS(p_s_key, get_rkey(p_s_chk_path, p_s_sb)) != -1 )
405 if ( COMP_KEYS(get_rkey(p_s_chk_path, p_s_sb), p_s_key) != 1 )
406 /* p_s_key must be less than right delimitiing key */
412 inline void decrement_bcount(
413 struct buffer_head * p_s_bh
416 if ( atomic_read (&(p_s_bh->b_count)) ) {
420 reiserfs_panic(NULL, "PAP-5070: decrement_bcount: trying to free free buffer %b", p_s_bh);
425 /* Decrement b_count field of the all buffers in the path. */
426 void decrement_counters_in_path (
427 struct path * p_s_search_path
429 int n_path_offset = p_s_search_path->path_length;
431 RFALSE( n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET ||
432 n_path_offset > EXTENDED_MAX_HEIGHT - 1,
433 "PAP-5080: illegal path offset of %d", n_path_offset);
435 while ( n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET ) {
436 struct buffer_head * bh;
438 bh = PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--);
439 decrement_bcount (bh);
441 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
445 int reiserfs_check_path(struct path *p) {
446 RFALSE( p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
447 "path not properly relsed") ;
452 /* Release all buffers in the path. Restore dirty bits clean
453 ** when preparing the buffer for the log
455 ** only called from fix_nodes()
457 void pathrelse_and_restore (
458 struct super_block *s,
459 struct path * p_s_search_path
461 int n_path_offset = p_s_search_path->path_length;
463 RFALSE( n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
464 "clm-4000: illegal path offset");
466 while ( n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET ) {
467 reiserfs_restore_prepared_buffer(s, PATH_OFFSET_PBUFFER(p_s_search_path,
469 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
471 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
474 /* Release all buffers in the path. */
476 struct path * p_s_search_path
478 int n_path_offset = p_s_search_path->path_length;
480 RFALSE( n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
481 "PAP-5090: illegal path offset");
483 while ( n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET )
484 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
486 p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
491 static int is_leaf (char * buf, int blocksize, struct buffer_head * bh)
493 struct block_head * blkh;
494 struct item_head * ih;
500 blkh = (struct block_head *)buf;
501 if ( blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
502 printk ("is_leaf: this should be caught earlier\n");
506 nr = blkh_nr_item(blkh);
507 if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
508 /* item number is too big or too small */
509 reiserfs_warning ("is_leaf: nr_item seems wrong: %z\n", bh);
512 ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
513 used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location (ih));
514 if (used_space != blocksize - blkh_free_space(blkh)) {
515 /* free space does not match to calculated amount of use space */
516 reiserfs_warning ("is_leaf: free space seems wrong: %z\n", bh);
520 // FIXME: it is_leaf will hit performance too much - we may have
523 /* check tables of item heads */
524 ih = (struct item_head *)(buf + BLKH_SIZE);
525 prev_location = blocksize;
526 for (i = 0; i < nr; i ++, ih ++) {
527 if ( le_ih_k_type(ih) == TYPE_ANY) {
528 reiserfs_warning ("is_leaf: wrong item type for item %h\n",ih);
531 if (ih_location (ih) >= blocksize || ih_location (ih) < IH_SIZE * nr) {
532 reiserfs_warning ("is_leaf: item location seems wrong: %h\n", ih);
535 if (ih_item_len (ih) < 1 || ih_item_len (ih) > MAX_ITEM_LEN (blocksize)) {
536 reiserfs_warning ("is_leaf: item length seems wrong: %h\n", ih);
539 if (prev_location - ih_location (ih) != ih_item_len (ih)) {
540 reiserfs_warning ("is_leaf: item location seems wrong (second one): %h\n", ih);
543 prev_location = ih_location (ih);
546 // one may imagine much more checks
551 /* returns 1 if buf looks like an internal node, 0 otherwise */
552 static int is_internal (char * buf, int blocksize, struct buffer_head * bh)
554 struct block_head * blkh;
558 blkh = (struct block_head *)buf;
559 nr = blkh_level(blkh);
560 if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
561 /* this level is not possible for internal nodes */
562 printk ("is_internal: this should be caught earlier\n");
566 nr = blkh_nr_item(blkh);
567 if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
568 /* for internal which is not root we might check min number of keys */
569 reiserfs_warning ("is_internal: number of key seems wrong: %z\n", bh);
573 used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
574 if (used_space != blocksize - blkh_free_space(blkh)) {
575 reiserfs_warning ("is_internal: free space seems wrong: %z\n", bh);
579 // one may imagine much more checks
584 // make sure that bh contains formatted node of reiserfs tree of
586 static int is_tree_node (struct buffer_head * bh, int level)
588 if (B_LEVEL (bh) != level) {
589 printk ("is_tree_node: node level %d does not match to the expected one %d\n",
590 B_LEVEL (bh), level);
593 if (level == DISK_LEAF_NODE_LEVEL)
594 return is_leaf (bh->b_data, bh->b_size, bh);
596 return is_internal (bh->b_data, bh->b_size, bh);
601 #ifdef SEARCH_BY_KEY_READA
603 /* The function is NOT SCHEDULE-SAFE! */
604 static void search_by_key_reada (struct super_block * s, int blocknr)
606 struct buffer_head * bh;
611 bh = sb_getblk (s, blocknr);
613 if (!buffer_uptodate (bh)) {
614 ll_rw_block (READA, 1, &bh);
621 /**************************************************************************
622 * Algorithm SearchByKey *
623 * look for item in the Disk S+Tree by its key *
624 * Input: p_s_sb - super block *
625 * p_s_key - pointer to the key to search *
626 * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR *
627 * p_s_search_path - path from the root to the needed leaf *
628 **************************************************************************/
630 /* This function fills up the path from the root to the leaf as it
631 descends the tree looking for the key. It uses reiserfs_bread to
632 try to find buffers in the cache given their block number. If it
633 does not find them in the cache it reads them from disk. For each
634 node search_by_key finds using reiserfs_bread it then uses
635 bin_search to look through that node. bin_search will find the
636 position of the block_number of the next node if it is looking
637 through an internal node. If it is looking through a leaf node
638 bin_search will find the position of the item which has key either
639 equal to given key, or which is the maximal key less than the given
640 key. search_by_key returns a path that must be checked for the
641 correctness of the top of the path but need not be checked for the
642 correctness of the bottom of the path */
643 /* The function is NOT SCHEDULE-SAFE! */
644 int search_by_key (struct super_block * p_s_sb,
645 const struct cpu_key * p_s_key, /* Key to search. */
646 struct path * p_s_search_path, /* This structure was
647 allocated and initialized
649 function. It is filled up
651 int n_stop_level /* How far down the tree to search. To
652 stop at leaf level - set to
653 DISK_LEAF_NODE_LEVEL */
655 int n_block_number = SB_ROOT_BLOCK (p_s_sb),
656 expected_level = SB_TREE_HEIGHT (p_s_sb);
657 struct buffer_head * p_s_bh;
658 struct path_element * p_s_last_element;
659 int n_node_level, n_retval;
660 int right_neighbor_of_leaf_node;
663 #ifdef CONFIG_REISERFS_CHECK
664 int n_repeat_counter = 0;
667 PROC_INFO_INC( p_s_sb, search_by_key );
669 /* As we add each node to a path we increase its count. This means that
670 we must be careful to release all nodes in a path before we either
671 discard the path struct or re-use the path struct, as we do here. */
673 decrement_counters_in_path(p_s_search_path);
675 right_neighbor_of_leaf_node = 0;
677 /* With each iteration of this loop we search through the items in the
678 current node, and calculate the next current node(next path element)
679 for the next iteration of this loop.. */
682 #ifdef CONFIG_REISERFS_CHECK
683 if ( !(++n_repeat_counter % 50000) )
684 reiserfs_warning ("PAP-5100: search_by_key: %s:"
685 "there were %d iterations of while loop "
686 "looking for key %K\n",
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);
695 #ifdef SEARCH_BY_KEY_READA
696 /* schedule read of right neighbor */
697 search_by_key_reada (p_s_sb, right_neighbor_of_leaf_node);
700 /* Read the next tree node, and set the last element in the path to
701 have a pointer to it. */
702 if ( ! (p_s_bh = p_s_last_element->pe_buffer =
703 sb_bread(p_s_sb, n_block_number)) ) {
704 p_s_search_path->path_length --;
705 pathrelse(p_s_search_path);
709 if( fs_changed (fs_gen, p_s_sb) ) {
710 PROC_INFO_INC( p_s_sb, search_by_key_fs_changed );
711 PROC_INFO_INC( p_s_sb, sbk_fs_changed[ expected_level - 1 ] );
714 /* It is possible that schedule occurred. We must check whether the key
715 to search is still in the tree rooted from the current buffer. If
716 not then repeat search from the root. */
717 if ( fs_changed (fs_gen, p_s_sb) &&
718 (!B_IS_IN_TREE (p_s_bh) || !key_in_buffer(p_s_search_path, p_s_key, p_s_sb)) ) {
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);
726 expected_level = SB_TREE_HEIGHT (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 ("vs-5150: search_by_key: "
750 "invalid format found in block %ld. Fsck?\n",
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: illegal 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: illegal 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: illegal 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-- ) {
1106 if (item_moved (&s_ih, p_s_path)) {
1110 RFALSE( p_n_unfm_pointer < (__u32 *)B_I_PITEM(p_s_bh, &s_ih) ||
1111 p_n_unfm_pointer > (__u32 *)B_I_PITEM(p_s_bh, &s_ih) + I_UNFM_NUM(&s_ih) - 1,
1112 "vs-5265: pointer out of range");
1114 /* Hole, nothing to remove. */
1115 if ( ! get_block_num(p_n_unfm_pointer,0) ) {
1122 tmp = get_block_num(p_n_unfm_pointer,0);
1123 put_block_num(p_n_unfm_pointer, 0, 0);
1124 journal_mark_dirty (th, p_s_sb, p_s_bh);
1125 inode->i_blocks -= p_s_sb->s_blocksize / 512;
1126 reiserfs_free_block(th, tmp);
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;
1156 /* Calculate bytes number which will be deleted or cutted in the balance. */
1157 int calc_deleted_bytes_number(
1158 struct tree_balance * p_s_tb,
1162 struct item_head * p_le_ih = PATH_PITEM_HEAD(p_s_tb->tb_path);
1164 if ( is_statdata_le_ih (p_le_ih) )
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?
1172 return ih_item_len(p_le_ih);
1174 n_del_size = ( c_mode == M_DELETE ) ? ih_item_len(p_le_ih) : -p_s_tb->insert_size[0];
1176 if ( is_indirect_le_ih (p_le_ih) )
1177 n_del_size = (n_del_size/UNFM_P_SIZE)*
1178 (PATH_PLAST_BUFFER(p_s_tb->tb_path)->b_size);// - get_ih_free_space (p_le_ih);
1182 static void init_tb_struct(
1183 struct reiserfs_transaction_handle *th,
1184 struct tree_balance * p_s_tb,
1185 struct super_block * p_s_sb,
1186 struct path * p_s_path,
1189 memset (p_s_tb,'\0',sizeof(struct tree_balance));
1190 p_s_tb->transaction_handle = th ;
1191 p_s_tb->tb_sb = p_s_sb;
1192 p_s_tb->tb_path = p_s_path;
1193 PATH_OFFSET_PBUFFER(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1194 PATH_OFFSET_POSITION(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1195 p_s_tb->insert_size[0] = n_size;
1200 void padd_item (char * item, int total_length, int length)
1204 for (i = total_length; i > length; )
1209 /* Delete object item. */
1210 int reiserfs_delete_item (struct reiserfs_transaction_handle *th,
1211 struct path * p_s_path, /* Path to the deleted item. */
1212 const struct cpu_key * p_s_item_key, /* Key to search for the deleted item. */
1213 struct inode * p_s_inode,/* inode is here just to update i_blocks */
1214 struct buffer_head * p_s_un_bh) /* NULL or unformatted node pointer. */
1216 struct super_block * p_s_sb = p_s_inode->i_sb;
1217 struct tree_balance s_del_balance;
1218 struct item_head s_ih;
1223 #ifdef CONFIG_REISERFS_CHECK
1228 init_tb_struct(th, &s_del_balance, p_s_sb, p_s_path, 0/*size is unknown*/);
1233 #ifdef CONFIG_REISERFS_CHECK
1237 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));
1239 RFALSE( c_mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1241 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1242 s_del_balance.insert_size[0] = n_del_size;
1244 n_ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, 0);
1245 if ( n_ret_value != REPEAT_SEARCH )
1248 PROC_INFO_INC( p_s_sb, delete_item_restarted );
1250 // file system changed, repeat search
1251 n_ret_value = search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1252 if (n_ret_value == IO_ERROR)
1254 if (n_ret_value == FILE_NOT_FOUND) {
1255 reiserfs_warning ("vs-5340: reiserfs_delete_item: "
1256 "no items of the file %K found\n", p_s_item_key);
1261 if ( n_ret_value != CARRY_ON ) {
1262 unfix_nodes(&s_del_balance);
1266 // reiserfs_delete_item returns item length when success
1267 n_ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1273 /* We are in direct2indirect conversion, so move tail contents
1274 to the unformatted node */
1275 /* note, we do the copy before preparing the buffer because we
1276 ** don't care about the contents of the unformatted node yet.
1277 ** the only thing we really care about is the direct item's data
1278 ** is in the unformatted node.
1280 ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1281 ** the unformatted node, which might schedule, meaning we'd have to
1282 ** loop all the way back up to the start of the while loop.
1284 ** The unformatted node must be dirtied later on. We can't be
1285 ** sure here if the entire tail has been deleted yet.
1287 ** p_s_un_bh is from the page cache (all unformatted nodes are
1288 ** from the page cache) and might be a highmem page. So, we
1289 ** can't use p_s_un_bh->b_data.
1293 data = kmap_atomic(p_s_un_bh->b_page, KM_USER0);
1294 off = ((le_ih_k_offset (&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1296 B_I_PITEM(PATH_PLAST_BUFFER(p_s_path), &s_ih), n_ret_value);
1297 kunmap_atomic(data, KM_USER0);
1300 /* Perform balancing after all resources have been collected at once. */
1301 do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1303 /* Return deleted body length */
1308 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1310 deletion of the body of the object is performed by iput(), with the
1311 result that if multiple processes are operating on a file, the
1312 deletion of the body of the file is deferred until the last process
1313 that has an open inode performs its iput().
1315 writes and truncates are protected from collisions by use of
1318 creates, linking, and mknod are protected from collisions with other
1319 processes by making the reiserfs_add_entry() the last step in the
1320 creation, and then rolling back all changes if there was a collision.
1325 /* this deletes item which never gets split */
1326 void reiserfs_delete_solid_item (struct reiserfs_transaction_handle *th,
1329 struct tree_balance tb;
1330 INITIALIZE_PATH (path);
1333 struct cpu_key cpu_key;
1336 le_key2cpu_key (&cpu_key, key);
1339 retval = search_item (th->t_super, &cpu_key, &path);
1340 if (retval == IO_ERROR) {
1341 reiserfs_warning ("vs-5350: reiserfs_delete_solid_item: "
1342 "i/o failure occurred trying to delete %K\n", &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
1348 if ( !( (unsigned long long) GET_HASH_VALUE (le_key_k_offset (le_key_version (key), key)) == 0 && \
1349 (unsigned long long) GET_GENERATION_NUMBER (le_key_k_offset (le_key_version (key), key)) == 1 ) )
1350 reiserfs_warning ("vs-5355: reiserfs_delete_solid_item: %k not found", key);
1355 item_len = ih_item_len( PATH_PITEM_HEAD(&path) );
1356 init_tb_struct (th, &tb, th->t_super, &path, - (IH_SIZE + item_len));
1359 retval = fix_nodes (M_DELETE, &tb, NULL, 0);
1360 if (retval == REPEAT_SEARCH) {
1361 PROC_INFO_INC( th -> t_super, delete_solid_item_restarted );
1365 if (retval == CARRY_ON) {
1366 do_balance (&tb, 0, 0, M_DELETE);
1370 // IO_ERROR, NO_DISK_SPACE, etc
1371 reiserfs_warning ("vs-5360: reiserfs_delete_solid_item: "
1372 "could not delete %K due to fix_nodes failure\n", &cpu_key);
1377 reiserfs_check_path(&path) ;
1381 void reiserfs_delete_object (struct reiserfs_transaction_handle *th, struct inode * inode)
1385 /* for directory this deletes item containing "." and ".." */
1386 reiserfs_do_truncate (th, inode, NULL, 0/*no timestamp updates*/);
1388 #if defined( USE_INODE_GENERATION_COUNTER )
1389 if( !old_format_only ( th -> t_super ) )
1391 __u32 *inode_generation;
1394 &REISERFS_SB(th -> t_super) -> s_rs -> s_inode_generation;
1395 *inode_generation = cpu_to_le32( le32_to_cpu( *inode_generation ) + 1 );
1397 /* USE_INODE_GENERATION_COUNTER */
1399 reiserfs_delete_solid_item (th, INODE_PKEY (inode));
1403 static int maybe_indirect_to_direct (struct reiserfs_transaction_handle *th,
1404 struct inode * p_s_inode,
1406 struct path * p_s_path,
1407 const struct cpu_key * p_s_item_key,
1408 loff_t n_new_file_size,
1411 struct super_block * p_s_sb = p_s_inode->i_sb;
1412 int n_block_size = p_s_sb->s_blocksize;
1415 if (n_new_file_size != p_s_inode->i_size)
1418 /* the page being sent in could be NULL if there was an i/o error
1419 ** reading in the last block. The user will hit problems trying to
1420 ** read the file, but for now we just skip the indirect2direct
1422 if (atomic_read(&p_s_inode->i_count) > 1 ||
1423 !tail_has_to_be_packed (p_s_inode) ||
1424 !page || (REISERFS_I(p_s_inode)->i_flags & i_nopack_mask)) {
1425 // leave tail in an unformatted node
1426 *p_c_mode = M_SKIP_BALANCING;
1427 cut_bytes = n_block_size - (n_new_file_size & (n_block_size - 1));
1428 pathrelse(p_s_path);
1431 /* Permorm the conversion to a direct_item. */
1432 /*return indirect_to_direct (p_s_inode, p_s_path, p_s_item_key, n_new_file_size, p_c_mode);*/
1433 return indirect2direct (th, p_s_inode, page, p_s_path, p_s_item_key, n_new_file_size, p_c_mode);
1437 /* we did indirect_to_direct conversion. And we have inserted direct
1438 item successesfully, but there were no disk space to cut unfm
1439 pointer being converted. Therefore we have to delete inserted
1441 static void indirect_to_direct_roll_back (struct reiserfs_transaction_handle *th, struct inode * inode, struct path * path)
1443 struct cpu_key tail_key;
1447 make_cpu_key (&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);// !!!!
1448 tail_key.key_length = 4;
1450 tail_len = (cpu_key_k_offset (&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1452 /* look for the last byte of the tail */
1453 if (search_for_position_by_key (inode->i_sb, &tail_key, path) == POSITION_NOT_FOUND)
1454 reiserfs_panic (inode->i_sb, "vs-5615: indirect_to_direct_roll_back: found invalid item");
1455 RFALSE( path->pos_in_item != ih_item_len(PATH_PITEM_HEAD (path)) - 1,
1456 "vs-5616: appended bytes found");
1457 PATH_LAST_POSITION (path) --;
1459 removed = reiserfs_delete_item (th, path, &tail_key, inode, 0/*unbh not needed*/);
1460 RFALSE( removed <= 0 || removed > tail_len,
1461 "vs-5617: there was tail %d bytes, removed item length %d bytes",
1463 tail_len -= removed;
1464 set_cpu_key_k_offset (&tail_key, cpu_key_k_offset (&tail_key) - removed);
1466 printk ("indirect_to_direct_roll_back: indirect_to_direct conversion has been rolled back due to lack of disk space\n");
1467 //mark_file_without_tail (inode);
1468 mark_inode_dirty (inode);
1472 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1473 int reiserfs_cut_from_item (struct reiserfs_transaction_handle *th,
1474 struct path * p_s_path,
1475 struct cpu_key * p_s_item_key,
1476 struct inode * p_s_inode,
1478 loff_t n_new_file_size)
1480 struct super_block * p_s_sb = p_s_inode->i_sb;
1481 /* Every function which is going to call do_balance must first
1482 create a tree_balance structure. Then it must fill up this
1483 structure by using the init_tb_struct and fix_nodes functions.
1484 After that we can make tree balancing. */
1485 struct tree_balance s_cut_balance;
1486 int n_cut_size = 0, /* Amount to be cut. */
1487 n_ret_value = CARRY_ON,
1488 n_removed = 0, /* Number of the removed unformatted nodes. */
1489 n_is_inode_locked = 0;
1490 char c_mode; /* Mode of the balance. */
1494 init_tb_struct(th, &s_cut_balance, p_s_inode->i_sb, p_s_path, n_cut_size);
1497 /* Repeat this loop until we either cut the item without needing
1498 to balance, or we fix_nodes without schedule occurring */
1500 /* Determine the balance mode, position of the first byte to
1501 be cut, and size to be cut. In case of the indirect item
1502 free unformatted nodes which are pointed to by the cut
1505 c_mode = prepare_for_delete_or_cut(th, p_s_inode, p_s_path, p_s_item_key, &n_removed,
1506 &n_cut_size, n_new_file_size);
1507 if ( c_mode == M_CONVERT ) {
1508 /* convert last unformatted node to direct item or leave
1509 tail in the unformatted node */
1510 RFALSE( n_ret_value != CARRY_ON, "PAP-5570: can not convert twice");
1512 n_ret_value = maybe_indirect_to_direct (th, p_s_inode, page, p_s_path, p_s_item_key,
1513 n_new_file_size, &c_mode);
1514 if ( c_mode == M_SKIP_BALANCING )
1515 /* tail has been left in the unformatted node */
1518 n_is_inode_locked = 1;
1520 /* removing of last unformatted node will change value we
1521 have to return to truncate. Save it */
1522 retval2 = n_ret_value;
1523 /*retval2 = p_s_sb->s_blocksize - (n_new_file_size & (p_s_sb->s_blocksize - 1));*/
1525 /* So, we have performed the first part of the conversion:
1526 inserting the new direct item. Now we are removing the
1527 last unformatted node pointer. Set key to search for
1529 set_cpu_key_k_type (p_s_item_key, TYPE_INDIRECT);
1530 p_s_item_key->key_length = 4;
1531 n_new_file_size -= (n_new_file_size & (p_s_sb->s_blocksize - 1));
1532 set_cpu_key_k_offset (p_s_item_key, n_new_file_size + 1);
1533 if ( search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path) == POSITION_NOT_FOUND ){
1534 print_block (PATH_PLAST_BUFFER (p_s_path), 3, PATH_LAST_POSITION (p_s_path) - 1, PATH_LAST_POSITION (p_s_path) + 1);
1535 reiserfs_panic(p_s_sb, "PAP-5580: reiserfs_cut_from_item: item to convert does not exist (%K)", p_s_item_key);
1539 if (n_cut_size == 0) {
1540 pathrelse (p_s_path);
1544 s_cut_balance.insert_size[0] = n_cut_size;
1546 n_ret_value = fix_nodes(c_mode, &s_cut_balance, NULL, 0);
1547 if ( n_ret_value != REPEAT_SEARCH )
1550 PROC_INFO_INC( p_s_sb, cut_from_item_restarted );
1552 n_ret_value = search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1553 if (n_ret_value == POSITION_FOUND)
1556 reiserfs_warning ("PAP-5610: reiserfs_cut_from_item: item %K not found\n", p_s_item_key);
1557 unfix_nodes (&s_cut_balance);
1558 return (n_ret_value == IO_ERROR) ? -EIO : -ENOENT;
1561 // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1562 if ( n_ret_value != CARRY_ON ) {
1563 if ( n_is_inode_locked ) {
1564 // FIXME: this seems to be not needed: we are always able
1566 indirect_to_direct_roll_back (th, p_s_inode, p_s_path);
1568 if (n_ret_value == NO_DISK_SPACE)
1569 reiserfs_warning ("NO_DISK_SPACE");
1570 unfix_nodes (&s_cut_balance);
1574 /* go ahead and perform balancing */
1576 RFALSE( c_mode == M_PASTE || c_mode == M_INSERT, "illegal mode");
1578 /* Calculate number of bytes that need to be cut from the item. */
1580 n_ret_value = calc_deleted_bytes_number(&s_cut_balance, c_mode);
1582 n_ret_value = retval2;
1584 if ( c_mode == M_DELETE ) {
1585 struct item_head * p_le_ih = PATH_PITEM_HEAD (s_cut_balance.tb_path);
1587 if ( is_direct_le_ih (p_le_ih) && (le_ih_k_offset (p_le_ih) & (p_s_sb->s_blocksize - 1)) == 1 ) {
1588 /* we delete first part of tail which was stored in direct
1590 // FIXME: this is to keep 3.5 happy
1591 REISERFS_I(p_s_inode)->i_first_direct_byte = U32_MAX;
1592 p_s_inode->i_blocks -= p_s_sb->s_blocksize / 512;
1596 #ifdef CONFIG_REISERFS_CHECK
1597 if (n_is_inode_locked) {
1598 struct item_head * le_ih = PATH_PITEM_HEAD (s_cut_balance.tb_path);
1599 /* we are going to complete indirect2direct conversion. Make
1600 sure, that we exactly remove last unformatted node pointer
1602 if (!is_indirect_le_ih (le_ih))
1603 reiserfs_panic (p_s_sb, "vs-5652: reiserfs_cut_from_item: "
1604 "item must be indirect %h", le_ih);
1606 if (c_mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1607 reiserfs_panic (p_s_sb, "vs-5653: reiserfs_cut_from_item: "
1608 "completing indirect2direct conversion indirect item %h "
1609 "being deleted must be of 4 byte long", le_ih);
1611 if (c_mode == M_CUT && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1612 reiserfs_panic (p_s_sb, "vs-5654: reiserfs_cut_from_item: "
1613 "can not complete indirect2direct conversion of %h (CUT, insert_size==%d)",
1614 le_ih, s_cut_balance.insert_size[0]);
1616 /* it would be useful to make sure, that right neighboring
1617 item is direct item of this file */
1621 do_balance(&s_cut_balance, NULL, NULL, c_mode);
1622 if ( n_is_inode_locked ) {
1623 /* we've done an indirect->direct conversion. when the data block
1624 ** was freed, it was removed from the list of blocks that must
1625 ** be flushed before the transaction commits, so we don't need to
1626 ** deal with it here.
1628 REISERFS_I(p_s_inode)->i_flags &= ~i_pack_on_close_mask ;
1634 static void truncate_directory (struct reiserfs_transaction_handle *th, struct inode * inode)
1637 reiserfs_warning ("vs-5655: truncate_directory: link count != 0\n");
1639 set_le_key_k_offset (KEY_FORMAT_3_5, INODE_PKEY (inode), DOT_OFFSET);
1640 set_le_key_k_type (KEY_FORMAT_3_5, INODE_PKEY (inode), TYPE_DIRENTRY);
1641 reiserfs_delete_solid_item (th, INODE_PKEY (inode));
1643 set_le_key_k_offset (KEY_FORMAT_3_5, INODE_PKEY (inode), SD_OFFSET);
1644 set_le_key_k_type (KEY_FORMAT_3_5, INODE_PKEY (inode), TYPE_STAT_DATA);
1650 /* Truncate file to the new size. Note, this must be called with a transaction
1652 void reiserfs_do_truncate (struct reiserfs_transaction_handle *th,
1653 struct inode * p_s_inode, /* ->i_size contains new
1655 struct page *page, /* up to date for last block */
1656 int update_timestamps /* when it is called by
1657 file_release to convert
1658 the tail - no timestamps
1659 should be updated */
1661 INITIALIZE_PATH (s_search_path); /* Path to the current object item. */
1662 struct item_head * p_le_ih; /* Pointer to an item header. */
1663 struct cpu_key s_item_key; /* Key to search for a previous file item. */
1664 loff_t n_file_size, /* Old file size. */
1665 n_new_file_size;/* New file size. */
1666 int n_deleted; /* Number of deleted or truncated bytes. */
1669 if ( ! (S_ISREG(p_s_inode->i_mode) || S_ISDIR(p_s_inode->i_mode) || S_ISLNK(p_s_inode->i_mode)) )
1672 if (S_ISDIR(p_s_inode->i_mode)) {
1673 // deletion of directory - no need to update timestamps
1674 truncate_directory (th, p_s_inode);
1678 /* Get new file size. */
1679 n_new_file_size = p_s_inode->i_size;
1681 // FIXME: note, that key type is unimportant here
1682 make_cpu_key (&s_item_key, p_s_inode, max_reiserfs_offset (p_s_inode), TYPE_DIRECT, 3);
1684 retval = search_for_position_by_key(p_s_inode->i_sb, &s_item_key, &s_search_path);
1685 if (retval == IO_ERROR) {
1686 reiserfs_warning ("vs-5657: reiserfs_do_truncate: "
1687 "i/o failure occurred trying to truncate %K\n", &s_item_key);
1690 if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1691 pathrelse (&s_search_path);
1692 reiserfs_warning ("PAP-5660: reiserfs_do_truncate: "
1693 "wrong result %d of search for %K\n", retval, &s_item_key);
1697 s_search_path.pos_in_item --;
1699 /* Get real file size (total length of all file items) */
1700 p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1701 if ( is_statdata_le_ih (p_le_ih) )
1704 loff_t offset = le_ih_k_offset (p_le_ih);
1705 int bytes = op_bytes_number (p_le_ih,p_s_inode->i_sb->s_blocksize);
1707 /* this may mismatch with real file size: if last direct item
1708 had no padding zeros and last unformatted node had no free
1709 space, this file would have this file size */
1710 n_file_size = offset + bytes - 1;
1713 if ( n_file_size == 0 || n_file_size < n_new_file_size ) {
1714 goto update_and_out ;
1717 /* Update key to search for the last file item. */
1718 set_cpu_key_k_offset (&s_item_key, n_file_size);
1721 /* Cut or delete file item. */
1722 n_deleted = reiserfs_cut_from_item(th, &s_search_path, &s_item_key, p_s_inode, page, n_new_file_size);
1723 if (n_deleted < 0) {
1724 reiserfs_warning ("vs-5665: reiserfs_do_truncate: reiserfs_cut_from_item failed");
1725 reiserfs_check_path(&s_search_path) ;
1729 RFALSE( n_deleted > n_file_size,
1730 "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1731 n_deleted, n_file_size, &s_item_key);
1733 /* Change key to search the last file item. */
1734 n_file_size -= n_deleted;
1736 set_cpu_key_k_offset (&s_item_key, n_file_size);
1738 /* While there are bytes to truncate and previous file item is presented in the tree. */
1741 ** This loop could take a really long time, and could log
1742 ** many more blocks than a transaction can hold. So, we do a polite
1743 ** journal end here, and if the transaction needs ending, we make
1744 ** sure the file is consistent before ending the current trans
1745 ** and starting a new one
1747 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1748 int orig_len_alloc = th->t_blocks_allocated ;
1749 decrement_counters_in_path(&s_search_path) ;
1751 if (update_timestamps) {
1752 p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME;
1754 reiserfs_update_sd(th, p_s_inode) ;
1756 journal_end(th, p_s_inode->i_sb, orig_len_alloc) ;
1757 journal_begin(th, p_s_inode->i_sb, orig_len_alloc) ;
1758 reiserfs_update_inode_transaction(p_s_inode) ;
1760 } while ( n_file_size > ROUND_UP (n_new_file_size) &&
1761 search_for_position_by_key(p_s_inode->i_sb, &s_item_key, &s_search_path) == POSITION_FOUND ) ;
1763 RFALSE( n_file_size > ROUND_UP (n_new_file_size),
1764 "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d\n",
1765 n_new_file_size, n_file_size, s_item_key.on_disk_key.k_objectid);
1768 if (update_timestamps) {
1769 // this is truncate, not file closing
1770 p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME;
1772 reiserfs_update_sd (th, p_s_inode);
1774 pathrelse(&s_search_path) ;
1778 #ifdef CONFIG_REISERFS_CHECK
1779 // this makes sure, that we __append__, not overwrite or add holes
1780 static void check_research_for_paste (struct path * path,
1781 const struct cpu_key * p_s_key)
1783 struct item_head * found_ih = get_ih (path);
1785 if (is_direct_le_ih (found_ih)) {
1786 if (le_ih_k_offset (found_ih) + op_bytes_number (found_ih, get_last_bh (path)->b_size) !=
1787 cpu_key_k_offset (p_s_key) ||
1788 op_bytes_number (found_ih, get_last_bh (path)->b_size) != pos_in_item (path))
1789 reiserfs_panic (0, "PAP-5720: check_research_for_paste: "
1790 "found direct item %h or position (%d) does not match to key %K",
1791 found_ih, pos_in_item (path), p_s_key);
1793 if (is_indirect_le_ih (found_ih)) {
1794 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) ||
1795 I_UNFM_NUM (found_ih) != pos_in_item (path) ||
1796 get_ih_free_space (found_ih) != 0)
1797 reiserfs_panic (0, "PAP-5730: check_research_for_paste: "
1798 "found indirect item (%h) or position (%d) does not match to key (%K)",
1799 found_ih, pos_in_item (path), p_s_key);
1802 #endif /* config reiserfs check */
1805 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1806 int reiserfs_paste_into_item (struct reiserfs_transaction_handle *th,
1807 struct path * p_s_search_path, /* Path to the pasted item. */
1808 const struct cpu_key * p_s_key, /* Key to search for the needed item.*/
1809 const char * p_c_body, /* Pointer to the bytes to paste. */
1810 int n_pasted_size) /* Size of pasted bytes. */
1812 struct tree_balance s_paste_balance;
1815 init_tb_struct(th, &s_paste_balance, th->t_super, p_s_search_path, n_pasted_size);
1816 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1817 s_paste_balance.key = p_s_key->on_disk_key;
1820 while ( (retval = fix_nodes(M_PASTE, &s_paste_balance, NULL, p_c_body)) == REPEAT_SEARCH ) {
1821 /* file system changed while we were in the fix_nodes */
1822 PROC_INFO_INC( th -> t_super, paste_into_item_restarted );
1823 retval = search_for_position_by_key (th->t_super, p_s_key, p_s_search_path);
1824 if (retval == IO_ERROR) {
1828 if (retval == POSITION_FOUND) {
1829 reiserfs_warning ("PAP-5710: reiserfs_paste_into_item: entry or pasted byte (%K) exists\n", p_s_key);
1834 #ifdef CONFIG_REISERFS_CHECK
1835 check_research_for_paste (p_s_search_path, p_s_key);
1839 /* Perform balancing after all resources are collected by fix_nodes, and
1840 accessing them will not risk triggering schedule. */
1841 if ( retval == CARRY_ON ) {
1842 do_balance(&s_paste_balance, NULL/*ih*/, p_c_body, M_PASTE);
1845 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
1847 /* this also releases the path */
1848 unfix_nodes(&s_paste_balance);
1853 /* Insert new item into the buffer at the path. */
1854 int reiserfs_insert_item(struct reiserfs_transaction_handle *th,
1855 struct path * p_s_path, /* Path to the inserteded item. */
1856 const struct cpu_key * key,
1857 struct item_head * p_s_ih, /* Pointer to the item header to insert.*/
1858 const char * p_c_body) /* Pointer to the bytes to insert. */
1860 struct tree_balance s_ins_balance;
1863 init_tb_struct(th, &s_ins_balance, th->t_super, p_s_path, IH_SIZE + ih_item_len(p_s_ih));
1864 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1865 s_ins_balance.key = key->on_disk_key;
1870 n_zeros_num = ih_item_len(p_s_ih);
1872 // le_key2cpu_key (&key, &(p_s_ih->ih_key));
1874 while ( (retval = fix_nodes(M_INSERT, &s_ins_balance, p_s_ih, p_c_body)) == REPEAT_SEARCH) {
1875 /* file system changed while we were in the fix_nodes */
1876 PROC_INFO_INC( th -> t_super, insert_item_restarted );
1877 retval = search_item (th->t_super, key, p_s_path);
1878 if (retval == IO_ERROR) {
1882 if (retval == ITEM_FOUND) {
1883 reiserfs_warning ("PAP-5760: reiserfs_insert_item: "
1884 "key %K already exists in the tree\n", key);
1890 /* make balancing after all resources will be collected at a time */
1891 if ( retval == CARRY_ON ) {
1892 do_balance (&s_ins_balance, p_s_ih, p_c_body, M_INSERT);
1896 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
1898 /* also releases the path */
1899 unfix_nodes(&s_ins_balance);