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: invalid 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: invalid 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 invalid 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: invalid 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: invalid 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: invalid 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 */
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.. */
680 n_block_number = SB_ROOT_BLOCK (p_s_sb);
684 #ifdef CONFIG_REISERFS_CHECK
685 if ( !(++n_repeat_counter % 50000) )
686 reiserfs_warning ("PAP-5100: search_by_key: %s:"
687 "there were %d iterations of while loop "
688 "looking for key %K\n",
689 current->comm, n_repeat_counter, p_s_key);
692 /* prep path to have another element added to it. */
693 p_s_last_element = PATH_OFFSET_PELEMENT(p_s_search_path, ++p_s_search_path->path_length);
694 fs_gen = get_generation (p_s_sb);
696 #ifdef SEARCH_BY_KEY_READA
697 /* schedule read of right neighbor */
698 search_by_key_reada (p_s_sb, right_neighbor_of_leaf_node);
701 /* Read the next tree node, and set the last element in the path to
702 have a pointer to it. */
703 if ( ! (p_s_bh = p_s_last_element->pe_buffer =
704 sb_bread(p_s_sb, n_block_number)) ) {
705 p_s_search_path->path_length --;
706 pathrelse(p_s_search_path);
709 if (expected_level == -1)
710 expected_level = SB_TREE_HEIGHT (p_s_sb);
713 /* It is possible that schedule occurred. We must check whether the key
714 to search is still in the tree rooted from the current buffer. If
715 not then repeat search from the root. */
716 if ( fs_changed (fs_gen, p_s_sb) &&
717 (!B_IS_IN_TREE (p_s_bh) ||
718 B_LEVEL(p_s_bh) != expected_level ||
719 !key_in_buffer(p_s_search_path, p_s_key, p_s_sb))) {
720 PROC_INFO_INC( p_s_sb, search_by_key_fs_changed );
721 PROC_INFO_INC( p_s_sb, search_by_key_restarted );
722 PROC_INFO_INC( p_s_sb, sbk_restarted[ expected_level - 1 ] );
723 decrement_counters_in_path(p_s_search_path);
725 /* Get the root block number so that we can repeat the search
726 starting from the root. */
727 n_block_number = SB_ROOT_BLOCK (p_s_sb);
729 right_neighbor_of_leaf_node = 0;
731 /* repeat search from the root */
735 /* only check that the key is in the buffer if p_s_key is not
736 equal to the MAX_KEY. Latter case is only possible in
737 "finish_unfinished()" processing during mount. */
738 RFALSE( COMP_KEYS( &MAX_KEY, p_s_key ) &&
739 ! key_in_buffer(p_s_search_path, p_s_key, p_s_sb),
740 "PAP-5130: key is not in the buffer");
741 #ifdef CONFIG_REISERFS_CHECK
743 print_cur_tb ("5140");
744 reiserfs_panic(p_s_sb, "PAP-5140: search_by_key: schedule occurred in do_balance!");
748 // make sure, that the node contents look like a node of
750 if (!is_tree_node (p_s_bh, expected_level)) {
751 reiserfs_warning ("vs-5150: search_by_key: "
752 "invalid format found in block %ld. Fsck?\n",
754 pathrelse (p_s_search_path);
758 /* ok, we have acquired next formatted node in the tree */
759 n_node_level = B_LEVEL (p_s_bh);
761 PROC_INFO_BH_STAT( p_s_sb, p_s_bh, n_node_level - 1 );
763 RFALSE( n_node_level < n_stop_level,
764 "vs-5152: tree level (%d) is less than stop level (%d)",
765 n_node_level, n_stop_level);
767 n_retval = bin_search( p_s_key, B_N_PITEM_HEAD(p_s_bh, 0),
769 ( n_node_level == DISK_LEAF_NODE_LEVEL ) ? IH_SIZE : KEY_SIZE,
770 &(p_s_last_element->pe_position));
771 if (n_node_level == n_stop_level) {
775 /* we are not in the stop level */
776 if (n_retval == ITEM_FOUND)
777 /* item has been found, so we choose the pointer which is to the right of the found one */
778 p_s_last_element->pe_position++;
780 /* if item was not found we choose the position which is to
781 the left of the found item. This requires no code,
782 bin_search did it already.*/
784 /* So we have chosen a position in the current node which is
785 an internal node. Now we calculate child block number by
786 position in the node. */
787 n_block_number = B_N_CHILD_NUM(p_s_bh, p_s_last_element->pe_position);
789 #ifdef SEARCH_BY_KEY_READA
790 /* if we are going to read leaf node, then calculate its right neighbor if possible */
791 if (n_node_level == DISK_LEAF_NODE_LEVEL + 1 && p_s_last_element->pe_position < B_NR_ITEMS (p_s_bh))
792 right_neighbor_of_leaf_node = B_N_CHILD_NUM(p_s_bh, p_s_last_element->pe_position + 1);
798 /* Form the path to an item and position in this item which contains
799 file byte defined by p_s_key. If there is no such item
800 corresponding to the key, we point the path to the item with
801 maximal key less than p_s_key, and *p_n_pos_in_item is set to one
802 past the last entry/byte in the item. If searching for entry in a
803 directory item, and it is not found, *p_n_pos_in_item is set to one
804 entry more than the entry with maximal key which is less than the
807 Note that if there is no entry in this same node which is one more,
808 then we point to an imaginary entry. for direct items, the
809 position is in units of bytes, for indirect items the position is
810 in units of blocknr entries, for directory items the position is in
811 units of directory entries. */
813 /* The function is NOT SCHEDULE-SAFE! */
814 int search_for_position_by_key (struct super_block * p_s_sb, /* Pointer to the super block. */
815 const struct cpu_key * p_cpu_key, /* Key to search (cpu variable) */
816 struct path * p_s_search_path /* Filled up by this function. */
818 struct item_head * p_le_ih; /* pointer to on-disk structure */
820 loff_t item_offset, offset;
821 struct reiserfs_dir_entry de;
824 /* If searching for directory entry. */
825 if ( is_direntry_cpu_key (p_cpu_key) )
826 return search_by_entry_key (p_s_sb, p_cpu_key, p_s_search_path, &de);
828 /* If not searching for directory entry. */
830 /* If item is found. */
831 retval = search_item (p_s_sb, p_cpu_key, p_s_search_path);
832 if (retval == IO_ERROR)
834 if ( retval == ITEM_FOUND ) {
836 RFALSE( ! ih_item_len(
837 B_N_PITEM_HEAD(PATH_PLAST_BUFFER(p_s_search_path),
838 PATH_LAST_POSITION(p_s_search_path))),
839 "PAP-5165: item length equals zero");
841 pos_in_item(p_s_search_path) = 0;
842 return POSITION_FOUND;
845 RFALSE( ! PATH_LAST_POSITION(p_s_search_path),
846 "PAP-5170: position equals zero");
848 /* Item is not found. Set path to the previous item. */
849 p_le_ih = B_N_PITEM_HEAD(PATH_PLAST_BUFFER(p_s_search_path), --PATH_LAST_POSITION(p_s_search_path));
850 n_blk_size = p_s_sb->s_blocksize;
852 if (comp_short_keys (&(p_le_ih->ih_key), p_cpu_key)) {
853 return FILE_NOT_FOUND;
856 // FIXME: quite ugly this far
858 item_offset = le_ih_k_offset (p_le_ih);
859 offset = cpu_key_k_offset (p_cpu_key);
861 /* Needed byte is contained in the item pointed to by the path.*/
862 if (item_offset <= offset &&
863 item_offset + op_bytes_number (p_le_ih, n_blk_size) > offset) {
864 pos_in_item (p_s_search_path) = offset - item_offset;
865 if ( is_indirect_le_ih(p_le_ih) ) {
866 pos_in_item (p_s_search_path) /= n_blk_size;
868 return POSITION_FOUND;
871 /* Needed byte is not contained in the item pointed to by the
872 path. Set pos_in_item out of the item. */
873 if ( is_indirect_le_ih (p_le_ih) )
874 pos_in_item (p_s_search_path) = ih_item_len(p_le_ih) / UNFM_P_SIZE;
876 pos_in_item (p_s_search_path) = ih_item_len( p_le_ih );
878 return POSITION_NOT_FOUND;
882 /* Compare given item and item pointed to by the path. */
883 int comp_items (const struct item_head * stored_ih, const struct path * p_s_path)
885 struct buffer_head * p_s_bh;
886 struct item_head * ih;
888 /* Last buffer at the path is not in the tree. */
889 if ( ! B_IS_IN_TREE(p_s_bh = PATH_PLAST_BUFFER(p_s_path)) )
892 /* Last path position is invalid. */
893 if ( PATH_LAST_POSITION(p_s_path) >= B_NR_ITEMS(p_s_bh) )
896 /* we need only to know, whether it is the same item */
897 ih = get_ih (p_s_path);
898 return memcmp (stored_ih, ih, IH_SIZE);
902 /* unformatted nodes are not logged anymore, ever. This is safe
905 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
907 // block can not be forgotten as it is in I/O or held by someone
908 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
912 // prepare for delete or cut of direct item
913 static inline int prepare_for_direct_item (struct path * path,
914 struct item_head * le_ih,
915 struct inode * inode,
916 loff_t new_file_length,
922 if ( new_file_length == max_reiserfs_offset (inode) ) {
923 /* item has to be deleted */
924 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
928 // new file gets truncated
929 if (get_inode_item_key_version (inode) == KEY_FORMAT_3_6) {
931 round_len = ROUND_UP (new_file_length);
932 /* this was n_new_file_length < le_ih ... */
933 if ( round_len < le_ih_k_offset (le_ih) ) {
934 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
935 return M_DELETE; /* Delete this item. */
937 /* Calculate first position and size for cutting from item. */
938 pos_in_item (path) = round_len - (le_ih_k_offset (le_ih) - 1);
939 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
941 return M_CUT; /* Cut from this item. */
945 // old file: items may have any length
947 if ( new_file_length < le_ih_k_offset (le_ih) ) {
948 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
949 return M_DELETE; /* Delete this item. */
951 /* Calculate first position and size for cutting from item. */
952 *cut_size = -(ih_item_len(le_ih) -
953 (pos_in_item (path) = new_file_length + 1 - le_ih_k_offset (le_ih)));
954 return M_CUT; /* Cut from this item. */
958 static inline int prepare_for_direntry_item (struct path * path,
959 struct item_head * le_ih,
960 struct inode * inode,
961 loff_t new_file_length,
964 if (le_ih_k_offset (le_ih) == DOT_OFFSET &&
965 new_file_length == max_reiserfs_offset (inode)) {
966 RFALSE( ih_entry_count (le_ih) != 2,
967 "PAP-5220: incorrect empty directory item (%h)", le_ih);
968 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
969 return M_DELETE; /* Delete the directory item containing "." and ".." entry. */
972 if ( ih_entry_count (le_ih) == 1 ) {
973 /* Delete the directory item such as there is one record only
975 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
979 /* Cut one record from the directory item. */
980 *cut_size = -(DEH_SIZE + entry_length (get_last_bh (path), le_ih, pos_in_item (path)));
985 /* If the path points to a directory or direct item, calculate mode and the size cut, for balance.
986 If the path points to an indirect item, remove some number of its unformatted nodes.
987 In case of file truncate calculate whether this item must be deleted/truncated or last
988 unformatted node of this item will be converted to a direct item.
989 This function returns a determination of what balance mode the calling function should employ. */
990 static char prepare_for_delete_or_cut(
991 struct reiserfs_transaction_handle *th,
992 struct inode * inode,
993 struct path * p_s_path,
994 const struct cpu_key * p_s_item_key,
995 int * p_n_removed, /* Number of unformatted nodes which were removed
996 from end of the file. */
998 unsigned long long n_new_file_length /* MAX_KEY_OFFSET in case of delete. */
1000 struct super_block * p_s_sb = inode->i_sb;
1001 struct item_head * p_le_ih = PATH_PITEM_HEAD(p_s_path);
1002 struct buffer_head * p_s_bh = PATH_PLAST_BUFFER(p_s_path);
1004 /* Stat_data item. */
1005 if ( is_statdata_le_ih (p_le_ih) ) {
1007 RFALSE( n_new_file_length != max_reiserfs_offset (inode),
1008 "PAP-5210: mode must be M_DELETE");
1010 *p_n_cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
1015 /* Directory item. */
1016 if ( is_direntry_le_ih (p_le_ih) )
1017 return prepare_for_direntry_item (p_s_path, p_le_ih, inode, n_new_file_length, p_n_cut_size);
1020 if ( is_direct_le_ih (p_le_ih) )
1021 return prepare_for_direct_item (p_s_path, p_le_ih, inode, n_new_file_length, p_n_cut_size);
1024 /* Case of an indirect item. */
1026 int n_unfm_number, /* Number of the item unformatted nodes. */
1029 __u32 * p_n_unfm_pointer; /* Pointer to the unformatted node number. */
1031 struct item_head s_ih; /* Item header. */
1032 char c_mode; /* Returned mode of the balance. */
1036 n_blk_size = p_s_sb->s_blocksize;
1038 /* Search for the needed object indirect item until there are no unformatted nodes to be removed. */
1041 p_s_bh = PATH_PLAST_BUFFER(p_s_path);
1042 /* Copy indirect item header to a temp variable. */
1043 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1044 /* Calculate number of unformatted nodes in this item. */
1045 n_unfm_number = I_UNFM_NUM(&s_ih);
1047 RFALSE( ! is_indirect_le_ih(&s_ih) || ! n_unfm_number ||
1048 pos_in_item (p_s_path) + 1 != n_unfm_number,
1049 "PAP-5240: invalid item %h "
1050 "n_unfm_number = %d *p_n_pos_in_item = %d",
1051 &s_ih, n_unfm_number, pos_in_item (p_s_path));
1053 /* Calculate balance mode and position in the item to remove unformatted nodes. */
1054 if ( n_new_file_length == max_reiserfs_offset (inode) ) {/* Case of delete. */
1055 pos_in_item (p_s_path) = 0;
1056 *p_n_cut_size = -(IH_SIZE + ih_item_len(&s_ih));
1059 else { /* Case of truncate. */
1060 if ( n_new_file_length < le_ih_k_offset (&s_ih) ) {
1061 pos_in_item (p_s_path) = 0;
1062 *p_n_cut_size = -(IH_SIZE + ih_item_len(&s_ih));
1063 c_mode = M_DELETE; /* Delete this item. */
1066 /* indirect item must be truncated starting from *p_n_pos_in_item-th position */
1067 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;
1069 RFALSE( pos_in_item (p_s_path) > n_unfm_number,
1070 "PAP-5250: invalid position in the item");
1072 /* Either convert last unformatted node of indirect item to direct item or increase
1074 if ( pos_in_item (p_s_path) == n_unfm_number ) {
1075 *p_n_cut_size = 0; /* Nothing to cut. */
1076 return M_CONVERT; /* Maybe convert last unformatted node to the direct item. */
1078 /* Calculate size to cut. */
1079 *p_n_cut_size = -(ih_item_len(&s_ih) - pos_in_item(p_s_path) * UNFM_P_SIZE);
1081 c_mode = M_CUT; /* Cut from this indirect item. */
1085 RFALSE( n_unfm_number <= pos_in_item (p_s_path),
1086 "PAP-5260: invalid position in the indirect item");
1088 /* pointers to be cut */
1089 n_unfm_number -= pos_in_item (p_s_path);
1090 /* Set pointer to the last unformatted node pointer that is to be cut. */
1091 p_n_unfm_pointer = (__u32 *)B_I_PITEM(p_s_bh, &s_ih) + I_UNFM_NUM(&s_ih) - 1 - *p_n_removed;
1094 /* We go through the unformatted nodes pointers of the indirect
1095 item and look for the unformatted nodes in the cache. If we
1096 found some of them we free it, zero corresponding indirect item
1097 entry and log buffer containing that indirect item. For this we
1098 need to prepare last path element for logging. If some
1099 unformatted node has b_count > 1 we must not free this
1100 unformatted node since it is in use. */
1101 reiserfs_prepare_for_journal(p_s_sb, p_s_bh, 1);
1102 // note: path could be changed, first line in for loop takes care
1105 for (n_counter = *p_n_removed;
1106 n_counter < n_unfm_number; n_counter++, p_n_unfm_pointer-- ) {
1109 if (item_moved (&s_ih, p_s_path)) {
1113 RFALSE( p_n_unfm_pointer < (__u32 *)B_I_PITEM(p_s_bh, &s_ih) ||
1114 p_n_unfm_pointer > (__u32 *)B_I_PITEM(p_s_bh, &s_ih) + I_UNFM_NUM(&s_ih) - 1,
1115 "vs-5265: pointer out of range");
1117 /* Hole, nothing to remove. */
1118 if ( ! get_block_num(p_n_unfm_pointer,0) ) {
1125 tmp = get_block_num(p_n_unfm_pointer,0);
1126 put_block_num(p_n_unfm_pointer, 0, 0);
1127 journal_mark_dirty (th, p_s_sb, p_s_bh);
1128 inode->i_blocks -= p_s_sb->s_blocksize / 512;
1129 reiserfs_free_block(th, tmp);
1130 if ( item_moved (&s_ih, p_s_path) ) {
1136 /* a trick. If the buffer has been logged, this
1137 ** will do nothing. If we've broken the loop without
1138 ** logging it, it will restore the buffer
1141 reiserfs_restore_prepared_buffer(p_s_sb, p_s_bh);
1143 /* This loop can be optimized. */
1144 } while ( (*p_n_removed < n_unfm_number || need_research) &&
1145 search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path) == POSITION_FOUND );
1147 RFALSE( *p_n_removed < n_unfm_number,
1148 "PAP-5310: indirect item is not found");
1149 RFALSE( item_moved (&s_ih, p_s_path),
1150 "after while, comp failed, retry") ;
1152 if (c_mode == M_CUT)
1153 pos_in_item (p_s_path) *= UNFM_P_SIZE;
1159 /* Calculate bytes number which will be deleted or cutted in the balance. */
1160 int calc_deleted_bytes_number(
1161 struct tree_balance * p_s_tb,
1165 struct item_head * p_le_ih = PATH_PITEM_HEAD(p_s_tb->tb_path);
1167 if ( is_statdata_le_ih (p_le_ih) )
1170 if ( is_direntry_le_ih (p_le_ih) ) {
1171 // return EMPTY_DIR_SIZE; /* We delete emty directoris only. */
1172 // we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1173 // empty size. ick. FIXME, is this right?
1175 return ih_item_len(p_le_ih);
1177 n_del_size = ( c_mode == M_DELETE ) ? ih_item_len(p_le_ih) : -p_s_tb->insert_size[0];
1179 if ( is_indirect_le_ih (p_le_ih) )
1180 n_del_size = (n_del_size/UNFM_P_SIZE)*
1181 (PATH_PLAST_BUFFER(p_s_tb->tb_path)->b_size);// - get_ih_free_space (p_le_ih);
1185 static void init_tb_struct(
1186 struct reiserfs_transaction_handle *th,
1187 struct tree_balance * p_s_tb,
1188 struct super_block * p_s_sb,
1189 struct path * p_s_path,
1192 memset (p_s_tb,'\0',sizeof(struct tree_balance));
1193 p_s_tb->transaction_handle = th ;
1194 p_s_tb->tb_sb = p_s_sb;
1195 p_s_tb->tb_path = p_s_path;
1196 PATH_OFFSET_PBUFFER(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1197 PATH_OFFSET_POSITION(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1198 p_s_tb->insert_size[0] = n_size;
1203 void padd_item (char * item, int total_length, int length)
1207 for (i = total_length; i > length; )
1212 /* Delete object item. */
1213 int reiserfs_delete_item (struct reiserfs_transaction_handle *th,
1214 struct path * p_s_path, /* Path to the deleted item. */
1215 const struct cpu_key * p_s_item_key, /* Key to search for the deleted item. */
1216 struct inode * p_s_inode,/* inode is here just to update i_blocks */
1217 struct buffer_head * p_s_un_bh) /* NULL or unformatted node pointer. */
1219 struct super_block * p_s_sb = p_s_inode->i_sb;
1220 struct tree_balance s_del_balance;
1221 struct item_head s_ih;
1226 #ifdef CONFIG_REISERFS_CHECK
1231 init_tb_struct(th, &s_del_balance, p_s_sb, p_s_path, 0/*size is unknown*/);
1236 #ifdef CONFIG_REISERFS_CHECK
1240 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));
1242 RFALSE( c_mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1244 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1245 s_del_balance.insert_size[0] = n_del_size;
1247 n_ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, 0);
1248 if ( n_ret_value != REPEAT_SEARCH )
1251 PROC_INFO_INC( p_s_sb, delete_item_restarted );
1253 // file system changed, repeat search
1254 n_ret_value = search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1255 if (n_ret_value == IO_ERROR)
1257 if (n_ret_value == FILE_NOT_FOUND) {
1258 reiserfs_warning ("vs-5340: reiserfs_delete_item: "
1259 "no items of the file %K found\n", p_s_item_key);
1264 if ( n_ret_value != CARRY_ON ) {
1265 unfix_nodes(&s_del_balance);
1269 // reiserfs_delete_item returns item length when success
1270 n_ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1276 /* We are in direct2indirect conversion, so move tail contents
1277 to the unformatted node */
1278 /* note, we do the copy before preparing the buffer because we
1279 ** don't care about the contents of the unformatted node yet.
1280 ** the only thing we really care about is the direct item's data
1281 ** is in the unformatted node.
1283 ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1284 ** the unformatted node, which might schedule, meaning we'd have to
1285 ** loop all the way back up to the start of the while loop.
1287 ** The unformatted node must be dirtied later on. We can't be
1288 ** sure here if the entire tail has been deleted yet.
1290 ** p_s_un_bh is from the page cache (all unformatted nodes are
1291 ** from the page cache) and might be a highmem page. So, we
1292 ** can't use p_s_un_bh->b_data.
1296 data = kmap_atomic(p_s_un_bh->b_page, KM_USER0);
1297 off = ((le_ih_k_offset (&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1299 B_I_PITEM(PATH_PLAST_BUFFER(p_s_path), &s_ih), n_ret_value);
1300 kunmap_atomic(data, KM_USER0);
1303 /* Perform balancing after all resources have been collected at once. */
1304 do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1306 /* Return deleted body length */
1311 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1313 deletion of the body of the object is performed by iput(), with the
1314 result that if multiple processes are operating on a file, the
1315 deletion of the body of the file is deferred until the last process
1316 that has an open inode performs its iput().
1318 writes and truncates are protected from collisions by use of
1321 creates, linking, and mknod are protected from collisions with other
1322 processes by making the reiserfs_add_entry() the last step in the
1323 creation, and then rolling back all changes if there was a collision.
1328 /* this deletes item which never gets split */
1329 void reiserfs_delete_solid_item (struct reiserfs_transaction_handle *th,
1332 struct tree_balance tb;
1333 INITIALIZE_PATH (path);
1336 struct cpu_key cpu_key;
1339 le_key2cpu_key (&cpu_key, key);
1342 retval = search_item (th->t_super, &cpu_key, &path);
1343 if (retval == IO_ERROR) {
1344 reiserfs_warning ("vs-5350: reiserfs_delete_solid_item: "
1345 "i/o failure occurred trying to delete %K\n", &cpu_key);
1348 if (retval != ITEM_FOUND) {
1350 // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1351 if ( !( (unsigned long long) GET_HASH_VALUE (le_key_k_offset (le_key_version (key), key)) == 0 && \
1352 (unsigned long long) GET_GENERATION_NUMBER (le_key_k_offset (le_key_version (key), key)) == 1 ) )
1353 reiserfs_warning ("vs-5355: reiserfs_delete_solid_item: %k not found", key);
1358 item_len = ih_item_len( PATH_PITEM_HEAD(&path) );
1359 init_tb_struct (th, &tb, th->t_super, &path, - (IH_SIZE + item_len));
1362 retval = fix_nodes (M_DELETE, &tb, NULL, 0);
1363 if (retval == REPEAT_SEARCH) {
1364 PROC_INFO_INC( th -> t_super, delete_solid_item_restarted );
1368 if (retval == CARRY_ON) {
1369 do_balance (&tb, 0, 0, M_DELETE);
1373 // IO_ERROR, NO_DISK_SPACE, etc
1374 reiserfs_warning ("vs-5360: reiserfs_delete_solid_item: "
1375 "could not delete %K due to fix_nodes failure\n", &cpu_key);
1380 reiserfs_check_path(&path) ;
1384 void reiserfs_delete_object (struct reiserfs_transaction_handle *th, struct inode * inode)
1388 /* for directory this deletes item containing "." and ".." */
1389 reiserfs_do_truncate (th, inode, NULL, 0/*no timestamp updates*/);
1391 #if defined( USE_INODE_GENERATION_COUNTER )
1392 if( !old_format_only ( th -> t_super ) )
1394 __u32 *inode_generation;
1397 &REISERFS_SB(th -> t_super) -> s_rs -> s_inode_generation;
1398 *inode_generation = cpu_to_le32( le32_to_cpu( *inode_generation ) + 1 );
1400 /* USE_INODE_GENERATION_COUNTER */
1402 reiserfs_delete_solid_item (th, INODE_PKEY (inode));
1406 static int maybe_indirect_to_direct (struct reiserfs_transaction_handle *th,
1407 struct inode * p_s_inode,
1409 struct path * p_s_path,
1410 const struct cpu_key * p_s_item_key,
1411 loff_t n_new_file_size,
1414 struct super_block * p_s_sb = p_s_inode->i_sb;
1415 int n_block_size = p_s_sb->s_blocksize;
1418 if (n_new_file_size != p_s_inode->i_size)
1421 /* the page being sent in could be NULL if there was an i/o error
1422 ** reading in the last block. The user will hit problems trying to
1423 ** read the file, but for now we just skip the indirect2direct
1425 if (atomic_read(&p_s_inode->i_count) > 1 ||
1426 !tail_has_to_be_packed (p_s_inode) ||
1427 !page || (REISERFS_I(p_s_inode)->i_flags & i_nopack_mask)) {
1428 // leave tail in an unformatted node
1429 *p_c_mode = M_SKIP_BALANCING;
1430 cut_bytes = n_block_size - (n_new_file_size & (n_block_size - 1));
1431 pathrelse(p_s_path);
1434 /* Permorm the conversion to a direct_item. */
1435 /*return indirect_to_direct (p_s_inode, p_s_path, p_s_item_key, n_new_file_size, p_c_mode);*/
1436 return indirect2direct (th, p_s_inode, page, p_s_path, p_s_item_key, n_new_file_size, p_c_mode);
1440 /* we did indirect_to_direct conversion. And we have inserted direct
1441 item successesfully, but there were no disk space to cut unfm
1442 pointer being converted. Therefore we have to delete inserted
1444 static void indirect_to_direct_roll_back (struct reiserfs_transaction_handle *th, struct inode * inode, struct path * path)
1446 struct cpu_key tail_key;
1450 make_cpu_key (&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);// !!!!
1451 tail_key.key_length = 4;
1453 tail_len = (cpu_key_k_offset (&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1455 /* look for the last byte of the tail */
1456 if (search_for_position_by_key (inode->i_sb, &tail_key, path) == POSITION_NOT_FOUND)
1457 reiserfs_panic (inode->i_sb, "vs-5615: indirect_to_direct_roll_back: found invalid item");
1458 RFALSE( path->pos_in_item != ih_item_len(PATH_PITEM_HEAD (path)) - 1,
1459 "vs-5616: appended bytes found");
1460 PATH_LAST_POSITION (path) --;
1462 removed = reiserfs_delete_item (th, path, &tail_key, inode, 0/*unbh not needed*/);
1463 RFALSE( removed <= 0 || removed > tail_len,
1464 "vs-5617: there was tail %d bytes, removed item length %d bytes",
1466 tail_len -= removed;
1467 set_cpu_key_k_offset (&tail_key, cpu_key_k_offset (&tail_key) - removed);
1469 printk ("indirect_to_direct_roll_back: indirect_to_direct conversion has been rolled back due to lack of disk space\n");
1470 //mark_file_without_tail (inode);
1471 mark_inode_dirty (inode);
1475 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1476 int reiserfs_cut_from_item (struct reiserfs_transaction_handle *th,
1477 struct path * p_s_path,
1478 struct cpu_key * p_s_item_key,
1479 struct inode * p_s_inode,
1481 loff_t n_new_file_size)
1483 struct super_block * p_s_sb = p_s_inode->i_sb;
1484 /* Every function which is going to call do_balance must first
1485 create a tree_balance structure. Then it must fill up this
1486 structure by using the init_tb_struct and fix_nodes functions.
1487 After that we can make tree balancing. */
1488 struct tree_balance s_cut_balance;
1489 int n_cut_size = 0, /* Amount to be cut. */
1490 n_ret_value = CARRY_ON,
1491 n_removed = 0, /* Number of the removed unformatted nodes. */
1492 n_is_inode_locked = 0;
1493 char c_mode; /* Mode of the balance. */
1497 init_tb_struct(th, &s_cut_balance, p_s_inode->i_sb, p_s_path, n_cut_size);
1500 /* Repeat this loop until we either cut the item without needing
1501 to balance, or we fix_nodes without schedule occurring */
1503 /* Determine the balance mode, position of the first byte to
1504 be cut, and size to be cut. In case of the indirect item
1505 free unformatted nodes which are pointed to by the cut
1508 c_mode = prepare_for_delete_or_cut(th, p_s_inode, p_s_path, p_s_item_key, &n_removed,
1509 &n_cut_size, n_new_file_size);
1510 if ( c_mode == M_CONVERT ) {
1511 /* convert last unformatted node to direct item or leave
1512 tail in the unformatted node */
1513 RFALSE( n_ret_value != CARRY_ON, "PAP-5570: can not convert twice");
1515 n_ret_value = maybe_indirect_to_direct (th, p_s_inode, page, p_s_path, p_s_item_key,
1516 n_new_file_size, &c_mode);
1517 if ( c_mode == M_SKIP_BALANCING )
1518 /* tail has been left in the unformatted node */
1521 n_is_inode_locked = 1;
1523 /* removing of last unformatted node will change value we
1524 have to return to truncate. Save it */
1525 retval2 = n_ret_value;
1526 /*retval2 = p_s_sb->s_blocksize - (n_new_file_size & (p_s_sb->s_blocksize - 1));*/
1528 /* So, we have performed the first part of the conversion:
1529 inserting the new direct item. Now we are removing the
1530 last unformatted node pointer. Set key to search for
1532 set_cpu_key_k_type (p_s_item_key, TYPE_INDIRECT);
1533 p_s_item_key->key_length = 4;
1534 n_new_file_size -= (n_new_file_size & (p_s_sb->s_blocksize - 1));
1535 set_cpu_key_k_offset (p_s_item_key, n_new_file_size + 1);
1536 if ( search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path) == POSITION_NOT_FOUND ){
1537 print_block (PATH_PLAST_BUFFER (p_s_path), 3, PATH_LAST_POSITION (p_s_path) - 1, PATH_LAST_POSITION (p_s_path) + 1);
1538 reiserfs_panic(p_s_sb, "PAP-5580: reiserfs_cut_from_item: item to convert does not exist (%K)", p_s_item_key);
1542 if (n_cut_size == 0) {
1543 pathrelse (p_s_path);
1547 s_cut_balance.insert_size[0] = n_cut_size;
1549 n_ret_value = fix_nodes(c_mode, &s_cut_balance, NULL, 0);
1550 if ( n_ret_value != REPEAT_SEARCH )
1553 PROC_INFO_INC( p_s_sb, cut_from_item_restarted );
1555 n_ret_value = search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1556 if (n_ret_value == POSITION_FOUND)
1559 reiserfs_warning ("PAP-5610: reiserfs_cut_from_item: item %K not found\n", p_s_item_key);
1560 unfix_nodes (&s_cut_balance);
1561 return (n_ret_value == IO_ERROR) ? -EIO : -ENOENT;
1564 // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1565 if ( n_ret_value != CARRY_ON ) {
1566 if ( n_is_inode_locked ) {
1567 // FIXME: this seems to be not needed: we are always able
1569 indirect_to_direct_roll_back (th, p_s_inode, p_s_path);
1571 if (n_ret_value == NO_DISK_SPACE)
1572 reiserfs_warning ("NO_DISK_SPACE");
1573 unfix_nodes (&s_cut_balance);
1577 /* go ahead and perform balancing */
1579 RFALSE( c_mode == M_PASTE || c_mode == M_INSERT, "invalid mode");
1581 /* Calculate number of bytes that need to be cut from the item. */
1583 n_ret_value = calc_deleted_bytes_number(&s_cut_balance, c_mode);
1585 n_ret_value = retval2;
1587 if ( c_mode == M_DELETE ) {
1588 struct item_head * p_le_ih = PATH_PITEM_HEAD (s_cut_balance.tb_path);
1590 if ( is_direct_le_ih (p_le_ih) && (le_ih_k_offset (p_le_ih) & (p_s_sb->s_blocksize - 1)) == 1 ) {
1591 /* we delete first part of tail which was stored in direct
1593 // FIXME: this is to keep 3.5 happy
1594 REISERFS_I(p_s_inode)->i_first_direct_byte = U32_MAX;
1595 p_s_inode->i_blocks -= p_s_sb->s_blocksize / 512;
1599 #ifdef CONFIG_REISERFS_CHECK
1600 if (n_is_inode_locked) {
1601 struct item_head * le_ih = PATH_PITEM_HEAD (s_cut_balance.tb_path);
1602 /* we are going to complete indirect2direct conversion. Make
1603 sure, that we exactly remove last unformatted node pointer
1605 if (!is_indirect_le_ih (le_ih))
1606 reiserfs_panic (p_s_sb, "vs-5652: reiserfs_cut_from_item: "
1607 "item must be indirect %h", le_ih);
1609 if (c_mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1610 reiserfs_panic (p_s_sb, "vs-5653: reiserfs_cut_from_item: "
1611 "completing indirect2direct conversion indirect item %h "
1612 "being deleted must be of 4 byte long", le_ih);
1614 if (c_mode == M_CUT && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1615 reiserfs_panic (p_s_sb, "vs-5654: reiserfs_cut_from_item: "
1616 "can not complete indirect2direct conversion of %h (CUT, insert_size==%d)",
1617 le_ih, s_cut_balance.insert_size[0]);
1619 /* it would be useful to make sure, that right neighboring
1620 item is direct item of this file */
1624 do_balance(&s_cut_balance, NULL, NULL, c_mode);
1625 if ( n_is_inode_locked ) {
1626 /* we've done an indirect->direct conversion. when the data block
1627 ** was freed, it was removed from the list of blocks that must
1628 ** be flushed before the transaction commits, so we don't need to
1629 ** deal with it here.
1631 REISERFS_I(p_s_inode)->i_flags &= ~i_pack_on_close_mask ;
1637 static void truncate_directory (struct reiserfs_transaction_handle *th, struct inode * inode)
1640 reiserfs_warning ("vs-5655: truncate_directory: link count != 0\n");
1642 set_le_key_k_offset (KEY_FORMAT_3_5, INODE_PKEY (inode), DOT_OFFSET);
1643 set_le_key_k_type (KEY_FORMAT_3_5, INODE_PKEY (inode), TYPE_DIRENTRY);
1644 reiserfs_delete_solid_item (th, INODE_PKEY (inode));
1646 set_le_key_k_offset (KEY_FORMAT_3_5, INODE_PKEY (inode), SD_OFFSET);
1647 set_le_key_k_type (KEY_FORMAT_3_5, INODE_PKEY (inode), TYPE_STAT_DATA);
1653 /* Truncate file to the new size. Note, this must be called with a transaction
1655 void reiserfs_do_truncate (struct reiserfs_transaction_handle *th,
1656 struct inode * p_s_inode, /* ->i_size contains new
1658 struct page *page, /* up to date for last block */
1659 int update_timestamps /* when it is called by
1660 file_release to convert
1661 the tail - no timestamps
1662 should be updated */
1664 INITIALIZE_PATH (s_search_path); /* Path to the current object item. */
1665 struct item_head * p_le_ih; /* Pointer to an item header. */
1666 struct cpu_key s_item_key; /* Key to search for a previous file item. */
1667 loff_t n_file_size, /* Old file size. */
1668 n_new_file_size;/* New file size. */
1669 int n_deleted; /* Number of deleted or truncated bytes. */
1672 if ( ! (S_ISREG(p_s_inode->i_mode) || S_ISDIR(p_s_inode->i_mode) || S_ISLNK(p_s_inode->i_mode)) )
1675 if (S_ISDIR(p_s_inode->i_mode)) {
1676 // deletion of directory - no need to update timestamps
1677 truncate_directory (th, p_s_inode);
1681 /* Get new file size. */
1682 n_new_file_size = p_s_inode->i_size;
1684 // FIXME: note, that key type is unimportant here
1685 make_cpu_key (&s_item_key, p_s_inode, max_reiserfs_offset (p_s_inode), TYPE_DIRECT, 3);
1687 retval = search_for_position_by_key(p_s_inode->i_sb, &s_item_key, &s_search_path);
1688 if (retval == IO_ERROR) {
1689 reiserfs_warning ("vs-5657: reiserfs_do_truncate: "
1690 "i/o failure occurred trying to truncate %K\n", &s_item_key);
1693 if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1694 pathrelse (&s_search_path);
1695 reiserfs_warning ("PAP-5660: reiserfs_do_truncate: "
1696 "wrong result %d of search for %K\n", retval, &s_item_key);
1700 s_search_path.pos_in_item --;
1702 /* Get real file size (total length of all file items) */
1703 p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1704 if ( is_statdata_le_ih (p_le_ih) )
1707 loff_t offset = le_ih_k_offset (p_le_ih);
1708 int bytes = op_bytes_number (p_le_ih,p_s_inode->i_sb->s_blocksize);
1710 /* this may mismatch with real file size: if last direct item
1711 had no padding zeros and last unformatted node had no free
1712 space, this file would have this file size */
1713 n_file_size = offset + bytes - 1;
1716 if ( n_file_size == 0 || n_file_size < n_new_file_size ) {
1717 goto update_and_out ;
1720 /* Update key to search for the last file item. */
1721 set_cpu_key_k_offset (&s_item_key, n_file_size);
1724 /* Cut or delete file item. */
1725 n_deleted = reiserfs_cut_from_item(th, &s_search_path, &s_item_key, p_s_inode, page, n_new_file_size);
1726 if (n_deleted < 0) {
1727 reiserfs_warning ("vs-5665: reiserfs_do_truncate: reiserfs_cut_from_item failed");
1728 reiserfs_check_path(&s_search_path) ;
1732 RFALSE( n_deleted > n_file_size,
1733 "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1734 n_deleted, n_file_size, &s_item_key);
1736 /* Change key to search the last file item. */
1737 n_file_size -= n_deleted;
1739 set_cpu_key_k_offset (&s_item_key, n_file_size);
1741 /* While there are bytes to truncate and previous file item is presented in the tree. */
1744 ** This loop could take a really long time, and could log
1745 ** many more blocks than a transaction can hold. So, we do a polite
1746 ** journal end here, and if the transaction needs ending, we make
1747 ** sure the file is consistent before ending the current trans
1748 ** and starting a new one
1750 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1751 int orig_len_alloc = th->t_blocks_allocated ;
1752 decrement_counters_in_path(&s_search_path) ;
1754 if (update_timestamps) {
1755 p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME;
1757 reiserfs_update_sd(th, p_s_inode) ;
1759 journal_end(th, p_s_inode->i_sb, orig_len_alloc) ;
1760 journal_begin(th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 6) ;
1761 reiserfs_update_inode_transaction(p_s_inode) ;
1763 } while ( n_file_size > ROUND_UP (n_new_file_size) &&
1764 search_for_position_by_key(p_s_inode->i_sb, &s_item_key, &s_search_path) == POSITION_FOUND ) ;
1766 RFALSE( n_file_size > ROUND_UP (n_new_file_size),
1767 "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d\n",
1768 n_new_file_size, n_file_size, s_item_key.on_disk_key.k_objectid);
1771 if (update_timestamps) {
1772 // this is truncate, not file closing
1773 p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME;
1775 reiserfs_update_sd (th, p_s_inode);
1777 pathrelse(&s_search_path) ;
1781 #ifdef CONFIG_REISERFS_CHECK
1782 // this makes sure, that we __append__, not overwrite or add holes
1783 static void check_research_for_paste (struct path * path,
1784 const struct cpu_key * p_s_key)
1786 struct item_head * found_ih = get_ih (path);
1788 if (is_direct_le_ih (found_ih)) {
1789 if (le_ih_k_offset (found_ih) + op_bytes_number (found_ih, get_last_bh (path)->b_size) !=
1790 cpu_key_k_offset (p_s_key) ||
1791 op_bytes_number (found_ih, get_last_bh (path)->b_size) != pos_in_item (path))
1792 reiserfs_panic (0, "PAP-5720: check_research_for_paste: "
1793 "found direct item %h or position (%d) does not match to key %K",
1794 found_ih, pos_in_item (path), p_s_key);
1796 if (is_indirect_le_ih (found_ih)) {
1797 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) ||
1798 I_UNFM_NUM (found_ih) != pos_in_item (path) ||
1799 get_ih_free_space (found_ih) != 0)
1800 reiserfs_panic (0, "PAP-5730: check_research_for_paste: "
1801 "found indirect item (%h) or position (%d) does not match to key (%K)",
1802 found_ih, pos_in_item (path), p_s_key);
1805 #endif /* config reiserfs check */
1808 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1809 int reiserfs_paste_into_item (struct reiserfs_transaction_handle *th,
1810 struct path * p_s_search_path, /* Path to the pasted item. */
1811 const struct cpu_key * p_s_key, /* Key to search for the needed item.*/
1812 const char * p_c_body, /* Pointer to the bytes to paste. */
1813 int n_pasted_size) /* Size of pasted bytes. */
1815 struct tree_balance s_paste_balance;
1818 init_tb_struct(th, &s_paste_balance, th->t_super, p_s_search_path, n_pasted_size);
1819 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1820 s_paste_balance.key = p_s_key->on_disk_key;
1823 while ( (retval = fix_nodes(M_PASTE, &s_paste_balance, NULL, p_c_body)) == REPEAT_SEARCH ) {
1824 /* file system changed while we were in the fix_nodes */
1825 PROC_INFO_INC( th -> t_super, paste_into_item_restarted );
1826 retval = search_for_position_by_key (th->t_super, p_s_key, p_s_search_path);
1827 if (retval == IO_ERROR) {
1831 if (retval == POSITION_FOUND) {
1832 reiserfs_warning ("PAP-5710: reiserfs_paste_into_item: entry or pasted byte (%K) exists\n", p_s_key);
1837 #ifdef CONFIG_REISERFS_CHECK
1838 check_research_for_paste (p_s_search_path, p_s_key);
1842 /* Perform balancing after all resources are collected by fix_nodes, and
1843 accessing them will not risk triggering schedule. */
1844 if ( retval == CARRY_ON ) {
1845 do_balance(&s_paste_balance, NULL/*ih*/, p_c_body, M_PASTE);
1848 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
1850 /* this also releases the path */
1851 unfix_nodes(&s_paste_balance);
1856 /* Insert new item into the buffer at the path. */
1857 int reiserfs_insert_item(struct reiserfs_transaction_handle *th,
1858 struct path * p_s_path, /* Path to the inserteded item. */
1859 const struct cpu_key * key,
1860 struct item_head * p_s_ih, /* Pointer to the item header to insert.*/
1861 const char * p_c_body) /* Pointer to the bytes to insert. */
1863 struct tree_balance s_ins_balance;
1866 init_tb_struct(th, &s_ins_balance, th->t_super, p_s_path, IH_SIZE + ih_item_len(p_s_ih));
1867 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1868 s_ins_balance.key = key->on_disk_key;
1873 n_zeros_num = ih_item_len(p_s_ih);
1875 // le_key2cpu_key (&key, &(p_s_ih->ih_key));
1877 while ( (retval = fix_nodes(M_INSERT, &s_ins_balance, p_s_ih, p_c_body)) == REPEAT_SEARCH) {
1878 /* file system changed while we were in the fix_nodes */
1879 PROC_INFO_INC( th -> t_super, insert_item_restarted );
1880 retval = search_item (th->t_super, key, p_s_path);
1881 if (retval == IO_ERROR) {
1885 if (retval == ITEM_FOUND) {
1886 reiserfs_warning ("PAP-5760: reiserfs_insert_item: "
1887 "key %K already exists in the tree\n", key);
1893 /* make balancing after all resources will be collected at a time */
1894 if ( retval == CARRY_ON ) {
1895 do_balance (&s_ins_balance, p_s_ih, p_c_body, M_INSERT);
1899 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
1901 /* also releases the path */
1902 unfix_nodes(&s_ins_balance);