xref: /linux/fs/reiserfs/stree.c (revision c5aec4c76af1a2d89ee2f2d4d5463b2ad2d85de5)
1 /*
2  *  Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3  */
4 
5 /*
6  *  Written by Anatoly P. Pinchuk pap@namesys.botik.ru
7  *  Programm System Institute
8  *  Pereslavl-Zalessky Russia
9  */
10 
11 /*
12  *  This file contains functions dealing with S+tree
13  *
14  * B_IS_IN_TREE
15  * copy_item_head
16  * comp_short_keys
17  * comp_keys
18  * comp_short_le_keys
19  * le_key2cpu_key
20  * comp_le_keys
21  * bin_search
22  * get_lkey
23  * get_rkey
24  * key_in_buffer
25  * decrement_bcount
26  * reiserfs_check_path
27  * pathrelse_and_restore
28  * pathrelse
29  * search_by_key_reada
30  * search_by_key
31  * search_for_position_by_key
32  * comp_items
33  * prepare_for_direct_item
34  * prepare_for_direntry_item
35  * prepare_for_delete_or_cut
36  * calc_deleted_bytes_number
37  * init_tb_struct
38  * padd_item
39  * reiserfs_delete_item
40  * reiserfs_delete_solid_item
41  * reiserfs_delete_object
42  * maybe_indirect_to_direct
43  * indirect_to_direct_roll_back
44  * reiserfs_cut_from_item
45  * truncate_directory
46  * reiserfs_do_truncate
47  * reiserfs_paste_into_item
48  * reiserfs_insert_item
49  */
50 
51 #include <linux/time.h>
52 #include <linux/string.h>
53 #include <linux/pagemap.h>
54 #include "reiserfs.h"
55 #include <linux/buffer_head.h>
56 #include <linux/quotaops.h>
57 
58 /* Does the buffer contain a disk block which is in the tree. */
59 inline int B_IS_IN_TREE(const struct buffer_head *bh)
60 {
61 
62 	RFALSE(B_LEVEL(bh) > MAX_HEIGHT,
63 	       "PAP-1010: block (%b) has too big level (%z)", bh, bh);
64 
65 	return (B_LEVEL(bh) != FREE_LEVEL);
66 }
67 
68 //
69 // to gets item head in le form
70 //
71 inline void copy_item_head(struct item_head *to,
72 			   const struct item_head *from)
73 {
74 	memcpy(to, from, IH_SIZE);
75 }
76 
77 /* k1 is pointer to on-disk structure which is stored in little-endian
78    form. k2 is pointer to cpu variable. For key of items of the same
79    object this returns 0.
80    Returns: -1 if key1 < key2
81    0 if key1 == key2
82    1 if key1 > key2 */
83 inline int comp_short_keys(const struct reiserfs_key *le_key,
84 			   const struct cpu_key *cpu_key)
85 {
86 	__u32 n;
87 	n = le32_to_cpu(le_key->k_dir_id);
88 	if (n < cpu_key->on_disk_key.k_dir_id)
89 		return -1;
90 	if (n > cpu_key->on_disk_key.k_dir_id)
91 		return 1;
92 	n = le32_to_cpu(le_key->k_objectid);
93 	if (n < cpu_key->on_disk_key.k_objectid)
94 		return -1;
95 	if (n > cpu_key->on_disk_key.k_objectid)
96 		return 1;
97 	return 0;
98 }
99 
100 /* k1 is pointer to on-disk structure which is stored in little-endian
101    form. k2 is pointer to cpu variable.
102    Compare keys using all 4 key fields.
103    Returns: -1 if key1 < key2 0
104    if key1 = key2 1 if key1 > key2 */
105 static inline int comp_keys(const struct reiserfs_key *le_key,
106 			    const struct cpu_key *cpu_key)
107 {
108 	int retval;
109 
110 	retval = comp_short_keys(le_key, cpu_key);
111 	if (retval)
112 		return retval;
113 	if (le_key_k_offset(le_key_version(le_key), le_key) <
114 	    cpu_key_k_offset(cpu_key))
115 		return -1;
116 	if (le_key_k_offset(le_key_version(le_key), le_key) >
117 	    cpu_key_k_offset(cpu_key))
118 		return 1;
119 
120 	if (cpu_key->key_length == 3)
121 		return 0;
122 
123 	/* this part is needed only when tail conversion is in progress */
124 	if (le_key_k_type(le_key_version(le_key), le_key) <
125 	    cpu_key_k_type(cpu_key))
126 		return -1;
127 
128 	if (le_key_k_type(le_key_version(le_key), le_key) >
129 	    cpu_key_k_type(cpu_key))
130 		return 1;
131 
132 	return 0;
133 }
134 
135 inline int comp_short_le_keys(const struct reiserfs_key *key1,
136 			      const struct reiserfs_key *key2)
137 {
138 	__u32 *k1_u32, *k2_u32;
139 	int key_length = REISERFS_SHORT_KEY_LEN;
140 
141 	k1_u32 = (__u32 *) key1;
142 	k2_u32 = (__u32 *) key2;
143 	for (; key_length--; ++k1_u32, ++k2_u32) {
144 		if (le32_to_cpu(*k1_u32) < le32_to_cpu(*k2_u32))
145 			return -1;
146 		if (le32_to_cpu(*k1_u32) > le32_to_cpu(*k2_u32))
147 			return 1;
148 	}
149 	return 0;
150 }
151 
152 inline void le_key2cpu_key(struct cpu_key *to, const struct reiserfs_key *from)
153 {
154 	int version;
155 	to->on_disk_key.k_dir_id = le32_to_cpu(from->k_dir_id);
156 	to->on_disk_key.k_objectid = le32_to_cpu(from->k_objectid);
157 
158 	// find out version of the key
159 	version = le_key_version(from);
160 	to->version = version;
161 	to->on_disk_key.k_offset = le_key_k_offset(version, from);
162 	to->on_disk_key.k_type = le_key_k_type(version, from);
163 }
164 
165 // this does not say which one is bigger, it only returns 1 if keys
166 // are not equal, 0 otherwise
167 inline int comp_le_keys(const struct reiserfs_key *k1,
168 			const struct reiserfs_key *k2)
169 {
170 	return memcmp(k1, k2, sizeof(struct reiserfs_key));
171 }
172 
173 /**************************************************************************
174  *  Binary search toolkit function                                        *
175  *  Search for an item in the array by the item key                       *
176  *  Returns:    1 if found,  0 if not found;                              *
177  *        *pos = number of the searched element if found, else the        *
178  *        number of the first element that is larger than key.            *
179  **************************************************************************/
180 /* For those not familiar with binary search: lbound is the leftmost item that it
181  could be, rbound the rightmost item that it could be.  We examine the item
182  halfway between lbound and rbound, and that tells us either that we can increase
183  lbound, or decrease rbound, or that we have found it, or if lbound <= rbound that
184  there are no possible items, and we have not found it. With each examination we
185  cut the number of possible items it could be by one more than half rounded down,
186  or we find it. */
187 static inline int bin_search(const void *key,	/* Key to search for. */
188 			     const void *base,	/* First item in the array. */
189 			     int num,	/* Number of items in the array. */
190 			     int width,	/* Item size in the array.
191 					   searched. Lest the reader be
192 					   confused, note that this is crafted
193 					   as a general function, and when it
194 					   is applied specifically to the array
195 					   of item headers in a node, width
196 					   is actually the item header size not
197 					   the item size. */
198 			     int *pos /* Number of the searched for element. */
199     )
200 {
201 	int rbound, lbound, j;
202 
203 	for (j = ((rbound = num - 1) + (lbound = 0)) / 2;
204 	     lbound <= rbound; j = (rbound + lbound) / 2)
205 		switch (comp_keys
206 			((struct reiserfs_key *)((char *)base + j * width),
207 			 (struct cpu_key *)key)) {
208 		case -1:
209 			lbound = j + 1;
210 			continue;
211 		case 1:
212 			rbound = j - 1;
213 			continue;
214 		case 0:
215 			*pos = j;
216 			return ITEM_FOUND;	/* Key found in the array.  */
217 		}
218 
219 	/* bin_search did not find given key, it returns position of key,
220 	   that is minimal and greater than the given one. */
221 	*pos = lbound;
222 	return ITEM_NOT_FOUND;
223 }
224 
225 
226 /* Minimal possible key. It is never in the tree. */
227 const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} };
228 
229 /* Maximal possible key. It is never in the tree. */
230 static const struct reiserfs_key MAX_KEY = {
231 	cpu_to_le32(0xffffffff),
232 	cpu_to_le32(0xffffffff),
233 	{{cpu_to_le32(0xffffffff),
234 	  cpu_to_le32(0xffffffff)},}
235 };
236 
237 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
238    of the path, and going upwards.  We must check the path's validity at each step.  If the key is not in
239    the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
240    case we return a special key, either MIN_KEY or MAX_KEY. */
241 static inline const struct reiserfs_key *get_lkey(const struct treepath *chk_path,
242 						  const struct super_block *sb)
243 {
244 	int position, path_offset = chk_path->path_length;
245 	struct buffer_head *parent;
246 
247 	RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET,
248 	       "PAP-5010: invalid offset in the path");
249 
250 	/* While not higher in path than first element. */
251 	while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
252 
253 		RFALSE(!buffer_uptodate
254 		       (PATH_OFFSET_PBUFFER(chk_path, path_offset)),
255 		       "PAP-5020: parent is not uptodate");
256 
257 		/* Parent at the path is not in the tree now. */
258 		if (!B_IS_IN_TREE
259 		    (parent =
260 		     PATH_OFFSET_PBUFFER(chk_path, path_offset)))
261 			return &MAX_KEY;
262 		/* Check whether position in the parent is correct. */
263 		if ((position =
264 		     PATH_OFFSET_POSITION(chk_path,
265 					  path_offset)) >
266 		    B_NR_ITEMS(parent))
267 			return &MAX_KEY;
268 		/* Check whether parent at the path really points to the child. */
269 		if (B_N_CHILD_NUM(parent, position) !=
270 		    PATH_OFFSET_PBUFFER(chk_path,
271 					path_offset + 1)->b_blocknr)
272 			return &MAX_KEY;
273 		/* Return delimiting key if position in the parent is not equal to zero. */
274 		if (position)
275 			return B_N_PDELIM_KEY(parent, position - 1);
276 	}
277 	/* Return MIN_KEY if we are in the root of the buffer tree. */
278 	if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)->
279 	    b_blocknr == SB_ROOT_BLOCK(sb))
280 		return &MIN_KEY;
281 	return &MAX_KEY;
282 }
283 
284 /* Get delimiting key of the buffer at the path and its right neighbor. */
285 inline const struct reiserfs_key *get_rkey(const struct treepath *chk_path,
286 					   const struct super_block *sb)
287 {
288 	int position, path_offset = chk_path->path_length;
289 	struct buffer_head *parent;
290 
291 	RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET,
292 	       "PAP-5030: invalid offset in the path");
293 
294 	while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
295 
296 		RFALSE(!buffer_uptodate
297 		       (PATH_OFFSET_PBUFFER(chk_path, path_offset)),
298 		       "PAP-5040: parent is not uptodate");
299 
300 		/* Parent at the path is not in the tree now. */
301 		if (!B_IS_IN_TREE
302 		    (parent =
303 		     PATH_OFFSET_PBUFFER(chk_path, path_offset)))
304 			return &MIN_KEY;
305 		/* Check whether position in the parent is correct. */
306 		if ((position =
307 		     PATH_OFFSET_POSITION(chk_path,
308 					  path_offset)) >
309 		    B_NR_ITEMS(parent))
310 			return &MIN_KEY;
311 		/* Check whether parent at the path really points to the child. */
312 		if (B_N_CHILD_NUM(parent, position) !=
313 		    PATH_OFFSET_PBUFFER(chk_path,
314 					path_offset + 1)->b_blocknr)
315 			return &MIN_KEY;
316 		/* Return delimiting key if position in the parent is not the last one. */
317 		if (position != B_NR_ITEMS(parent))
318 			return B_N_PDELIM_KEY(parent, position);
319 	}
320 	/* Return MAX_KEY if we are in the root of the buffer tree. */
321 	if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)->
322 	    b_blocknr == SB_ROOT_BLOCK(sb))
323 		return &MAX_KEY;
324 	return &MIN_KEY;
325 }
326 
327 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
328 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
329    the path.  These delimiting keys are stored at least one level above that buffer in the tree. If the
330    buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
331    this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
332 static inline int key_in_buffer(struct treepath *chk_path,	/* Path which should be checked.  */
333 				const struct cpu_key *key,	/* Key which should be checked.   */
334 				struct super_block *sb
335     )
336 {
337 
338 	RFALSE(!key || chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET
339 	       || chk_path->path_length > MAX_HEIGHT,
340 	       "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
341 	       key, chk_path->path_length);
342 	RFALSE(!PATH_PLAST_BUFFER(chk_path)->b_bdev,
343 	       "PAP-5060: device must not be NODEV");
344 
345 	if (comp_keys(get_lkey(chk_path, sb), key) == 1)
346 		/* left delimiting key is bigger, that the key we look for */
347 		return 0;
348 	/*  if ( comp_keys(key, get_rkey(chk_path, sb)) != -1 ) */
349 	if (comp_keys(get_rkey(chk_path, sb), key) != 1)
350 		/* key must be less than right delimitiing key */
351 		return 0;
352 	return 1;
353 }
354 
355 int reiserfs_check_path(struct treepath *p)
356 {
357 	RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
358 	       "path not properly relsed");
359 	return 0;
360 }
361 
362 /* Drop the reference to each buffer in a path and restore
363  * dirty bits clean when preparing the buffer for the log.
364  * This version should only be called from fix_nodes() */
365 void pathrelse_and_restore(struct super_block *sb,
366 			   struct treepath *search_path)
367 {
368 	int path_offset = search_path->path_length;
369 
370 	RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
371 	       "clm-4000: invalid path offset");
372 
373 	while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
374 		struct buffer_head *bh;
375 		bh = PATH_OFFSET_PBUFFER(search_path, path_offset--);
376 		reiserfs_restore_prepared_buffer(sb, bh);
377 		brelse(bh);
378 	}
379 	search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
380 }
381 
382 /* Drop the reference to each buffer in a path */
383 void pathrelse(struct treepath *search_path)
384 {
385 	int path_offset = search_path->path_length;
386 
387 	RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
388 	       "PAP-5090: invalid path offset");
389 
390 	while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
391 		brelse(PATH_OFFSET_PBUFFER(search_path, path_offset--));
392 
393 	search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
394 }
395 
396 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
397 {
398 	struct block_head *blkh;
399 	struct item_head *ih;
400 	int used_space;
401 	int prev_location;
402 	int i;
403 	int nr;
404 
405 	blkh = (struct block_head *)buf;
406 	if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
407 		reiserfs_warning(NULL, "reiserfs-5080",
408 				 "this should be caught earlier");
409 		return 0;
410 	}
411 
412 	nr = blkh_nr_item(blkh);
413 	if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
414 		/* item number is too big or too small */
415 		reiserfs_warning(NULL, "reiserfs-5081",
416 				 "nr_item seems wrong: %z", bh);
417 		return 0;
418 	}
419 	ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
420 	used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih));
421 	if (used_space != blocksize - blkh_free_space(blkh)) {
422 		/* free space does not match to calculated amount of use space */
423 		reiserfs_warning(NULL, "reiserfs-5082",
424 				 "free space seems wrong: %z", bh);
425 		return 0;
426 	}
427 	// FIXME: it is_leaf will hit performance too much - we may have
428 	// return 1 here
429 
430 	/* check tables of item heads */
431 	ih = (struct item_head *)(buf + BLKH_SIZE);
432 	prev_location = blocksize;
433 	for (i = 0; i < nr; i++, ih++) {
434 		if (le_ih_k_type(ih) == TYPE_ANY) {
435 			reiserfs_warning(NULL, "reiserfs-5083",
436 					 "wrong item type for item %h",
437 					 ih);
438 			return 0;
439 		}
440 		if (ih_location(ih) >= blocksize
441 		    || ih_location(ih) < IH_SIZE * nr) {
442 			reiserfs_warning(NULL, "reiserfs-5084",
443 					 "item location seems wrong: %h",
444 					 ih);
445 			return 0;
446 		}
447 		if (ih_item_len(ih) < 1
448 		    || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
449 			reiserfs_warning(NULL, "reiserfs-5085",
450 					 "item length seems wrong: %h",
451 					 ih);
452 			return 0;
453 		}
454 		if (prev_location - ih_location(ih) != ih_item_len(ih)) {
455 			reiserfs_warning(NULL, "reiserfs-5086",
456 					 "item location seems wrong "
457 					 "(second one): %h", ih);
458 			return 0;
459 		}
460 		prev_location = ih_location(ih);
461 	}
462 
463 	// one may imagine much more checks
464 	return 1;
465 }
466 
467 /* returns 1 if buf looks like an internal node, 0 otherwise */
468 static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
469 {
470 	struct block_head *blkh;
471 	int nr;
472 	int used_space;
473 
474 	blkh = (struct block_head *)buf;
475 	nr = blkh_level(blkh);
476 	if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
477 		/* this level is not possible for internal nodes */
478 		reiserfs_warning(NULL, "reiserfs-5087",
479 				 "this should be caught earlier");
480 		return 0;
481 	}
482 
483 	nr = blkh_nr_item(blkh);
484 	if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
485 		/* for internal which is not root we might check min number of keys */
486 		reiserfs_warning(NULL, "reiserfs-5088",
487 				 "number of key seems wrong: %z", bh);
488 		return 0;
489 	}
490 
491 	used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
492 	if (used_space != blocksize - blkh_free_space(blkh)) {
493 		reiserfs_warning(NULL, "reiserfs-5089",
494 				 "free space seems wrong: %z", bh);
495 		return 0;
496 	}
497 	// one may imagine much more checks
498 	return 1;
499 }
500 
501 // make sure that bh contains formatted node of reiserfs tree of
502 // 'level'-th level
503 static int is_tree_node(struct buffer_head *bh, int level)
504 {
505 	if (B_LEVEL(bh) != level) {
506 		reiserfs_warning(NULL, "reiserfs-5090", "node level %d does "
507 				 "not match to the expected one %d",
508 				 B_LEVEL(bh), level);
509 		return 0;
510 	}
511 	if (level == DISK_LEAF_NODE_LEVEL)
512 		return is_leaf(bh->b_data, bh->b_size, bh);
513 
514 	return is_internal(bh->b_data, bh->b_size, bh);
515 }
516 
517 #define SEARCH_BY_KEY_READA 16
518 
519 /*
520  * The function is NOT SCHEDULE-SAFE!
521  * It might unlock the write lock if we needed to wait for a block
522  * to be read. Note that in this case it won't recover the lock to avoid
523  * high contention resulting from too much lock requests, especially
524  * the caller (search_by_key) will perform other schedule-unsafe
525  * operations just after calling this function.
526  *
527  * @return depth of lock to be restored after read completes
528  */
529 static int search_by_key_reada(struct super_block *s,
530 				struct buffer_head **bh,
531 				b_blocknr_t *b, int num)
532 {
533 	int i, j;
534 	int depth = -1;
535 
536 	for (i = 0; i < num; i++) {
537 		bh[i] = sb_getblk(s, b[i]);
538 	}
539 	/*
540 	 * We are going to read some blocks on which we
541 	 * have a reference. It's safe, though we might be
542 	 * reading blocks concurrently changed if we release
543 	 * the lock. But it's still fine because we check later
544 	 * if the tree changed
545 	 */
546 	for (j = 0; j < i; j++) {
547 		/*
548 		 * note, this needs attention if we are getting rid of the BKL
549 		 * you have to make sure the prepared bit isn't set on this buffer
550 		 */
551 		if (!buffer_uptodate(bh[j])) {
552 			if (depth == -1)
553 				depth = reiserfs_write_unlock_nested(s);
554 			ll_rw_block(READA, 1, bh + j);
555 		}
556 		brelse(bh[j]);
557 	}
558 	return depth;
559 }
560 
561 /**************************************************************************
562  * Algorithm   SearchByKey                                                *
563  *             look for item in the Disk S+Tree by its key                *
564  * Input:  sb   -  super block                                            *
565  *         key  - pointer to the key to search                            *
566  * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR                         *
567  *         search_path - path from the root to the needed leaf            *
568  **************************************************************************/
569 
570 /* This function fills up the path from the root to the leaf as it
571    descends the tree looking for the key.  It uses reiserfs_bread to
572    try to find buffers in the cache given their block number.  If it
573    does not find them in the cache it reads them from disk.  For each
574    node search_by_key finds using reiserfs_bread it then uses
575    bin_search to look through that node.  bin_search will find the
576    position of the block_number of the next node if it is looking
577    through an internal node.  If it is looking through a leaf node
578    bin_search will find the position of the item which has key either
579    equal to given key, or which is the maximal key less than the given
580    key.  search_by_key returns a path that must be checked for the
581    correctness of the top of the path but need not be checked for the
582    correctness of the bottom of the path */
583 /* The function is NOT SCHEDULE-SAFE! */
584 int search_by_key(struct super_block *sb, const struct cpu_key *key,	/* Key to search. */
585 		  struct treepath *search_path,/* This structure was
586 						   allocated and initialized
587 						   by the calling
588 						   function. It is filled up
589 						   by this function.  */
590 		  int stop_level	/* How far down the tree to search. To
591 					   stop at leaf level - set to
592 					   DISK_LEAF_NODE_LEVEL */
593     )
594 {
595 	b_blocknr_t block_number;
596 	int expected_level;
597 	struct buffer_head *bh;
598 	struct path_element *last_element;
599 	int node_level, retval;
600 	int right_neighbor_of_leaf_node;
601 	int fs_gen;
602 	struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
603 	b_blocknr_t reada_blocks[SEARCH_BY_KEY_READA];
604 	int reada_count = 0;
605 
606 #ifdef CONFIG_REISERFS_CHECK
607 	int repeat_counter = 0;
608 #endif
609 
610 	PROC_INFO_INC(sb, search_by_key);
611 
612 	/* As we add each node to a path we increase its count.  This means that
613 	   we must be careful to release all nodes in a path before we either
614 	   discard the path struct or re-use the path struct, as we do here. */
615 
616 	pathrelse(search_path);
617 
618 	right_neighbor_of_leaf_node = 0;
619 
620 	/* With each iteration of this loop we search through the items in the
621 	   current node, and calculate the next current node(next path element)
622 	   for the next iteration of this loop.. */
623 	block_number = SB_ROOT_BLOCK(sb);
624 	expected_level = -1;
625 	while (1) {
626 
627 #ifdef CONFIG_REISERFS_CHECK
628 		if (!(++repeat_counter % 50000))
629 			reiserfs_warning(sb, "PAP-5100",
630 					 "%s: there were %d iterations of "
631 					 "while loop looking for key %K",
632 					 current->comm, repeat_counter,
633 					 key);
634 #endif
635 
636 		/* prep path to have another element added to it. */
637 		last_element =
638 		    PATH_OFFSET_PELEMENT(search_path,
639 					 ++search_path->path_length);
640 		fs_gen = get_generation(sb);
641 
642 		/* Read the next tree node, and set the last element in the path to
643 		   have a pointer to it. */
644 		if ((bh = last_element->pe_buffer =
645 		     sb_getblk(sb, block_number))) {
646 
647 			/*
648 			 * We'll need to drop the lock if we encounter any
649 			 * buffers that need to be read. If all of them are
650 			 * already up to date, we don't need to drop the lock.
651 			 */
652 			int depth = -1;
653 
654 			if (!buffer_uptodate(bh) && reada_count > 1)
655 				depth = search_by_key_reada(sb, reada_bh,
656 						    reada_blocks, reada_count);
657 
658 			if (!buffer_uptodate(bh) && depth == -1)
659 				depth = reiserfs_write_unlock_nested(sb);
660 
661 			ll_rw_block(READ, 1, &bh);
662 			wait_on_buffer(bh);
663 
664 			if (depth != -1)
665 				reiserfs_write_lock_nested(sb, depth);
666 			if (!buffer_uptodate(bh))
667 				goto io_error;
668 		} else {
669 		      io_error:
670 			search_path->path_length--;
671 			pathrelse(search_path);
672 			return IO_ERROR;
673 		}
674 		reada_count = 0;
675 		if (expected_level == -1)
676 			expected_level = SB_TREE_HEIGHT(sb);
677 		expected_level--;
678 
679 		/* It is possible that schedule occurred. We must check whether the key
680 		   to search is still in the tree rooted from the current buffer. If
681 		   not then repeat search from the root. */
682 		if (fs_changed(fs_gen, sb) &&
683 		    (!B_IS_IN_TREE(bh) ||
684 		     B_LEVEL(bh) != expected_level ||
685 		     !key_in_buffer(search_path, key, sb))) {
686 			PROC_INFO_INC(sb, search_by_key_fs_changed);
687 			PROC_INFO_INC(sb, search_by_key_restarted);
688 			PROC_INFO_INC(sb,
689 				      sbk_restarted[expected_level - 1]);
690 			pathrelse(search_path);
691 
692 			/* Get the root block number so that we can repeat the search
693 			   starting from the root. */
694 			block_number = SB_ROOT_BLOCK(sb);
695 			expected_level = -1;
696 			right_neighbor_of_leaf_node = 0;
697 
698 			/* repeat search from the root */
699 			continue;
700 		}
701 
702 		/* only check that the key is in the buffer if key is not
703 		   equal to the MAX_KEY. Latter case is only possible in
704 		   "finish_unfinished()" processing during mount. */
705 		RFALSE(comp_keys(&MAX_KEY, key) &&
706 		       !key_in_buffer(search_path, key, sb),
707 		       "PAP-5130: key is not in the buffer");
708 #ifdef CONFIG_REISERFS_CHECK
709 		if (REISERFS_SB(sb)->cur_tb) {
710 			print_cur_tb("5140");
711 			reiserfs_panic(sb, "PAP-5140",
712 				       "schedule occurred in do_balance!");
713 		}
714 #endif
715 
716 		// make sure, that the node contents look like a node of
717 		// certain level
718 		if (!is_tree_node(bh, expected_level)) {
719 			reiserfs_error(sb, "vs-5150",
720 				       "invalid format found in block %ld. "
721 				       "Fsck?", bh->b_blocknr);
722 			pathrelse(search_path);
723 			return IO_ERROR;
724 		}
725 
726 		/* ok, we have acquired next formatted node in the tree */
727 		node_level = B_LEVEL(bh);
728 
729 		PROC_INFO_BH_STAT(sb, bh, node_level - 1);
730 
731 		RFALSE(node_level < stop_level,
732 		       "vs-5152: tree level (%d) is less than stop level (%d)",
733 		       node_level, stop_level);
734 
735 		retval = bin_search(key, B_N_PITEM_HEAD(bh, 0),
736 				      B_NR_ITEMS(bh),
737 				      (node_level ==
738 				       DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
739 				      KEY_SIZE,
740 				      &(last_element->pe_position));
741 		if (node_level == stop_level) {
742 			return retval;
743 		}
744 
745 		/* we are not in the stop level */
746 		if (retval == ITEM_FOUND)
747 			/* item has been found, so we choose the pointer which is to the right of the found one */
748 			last_element->pe_position++;
749 
750 		/* if item was not found we choose the position which is to
751 		   the left of the found item. This requires no code,
752 		   bin_search did it already. */
753 
754 		/* So we have chosen a position in the current node which is
755 		   an internal node.  Now we calculate child block number by
756 		   position in the node. */
757 		block_number =
758 		    B_N_CHILD_NUM(bh, last_element->pe_position);
759 
760 		/* if we are going to read leaf nodes, try for read ahead as well */
761 		if ((search_path->reada & PATH_READA) &&
762 		    node_level == DISK_LEAF_NODE_LEVEL + 1) {
763 			int pos = last_element->pe_position;
764 			int limit = B_NR_ITEMS(bh);
765 			struct reiserfs_key *le_key;
766 
767 			if (search_path->reada & PATH_READA_BACK)
768 				limit = 0;
769 			while (reada_count < SEARCH_BY_KEY_READA) {
770 				if (pos == limit)
771 					break;
772 				reada_blocks[reada_count++] =
773 				    B_N_CHILD_NUM(bh, pos);
774 				if (search_path->reada & PATH_READA_BACK)
775 					pos--;
776 				else
777 					pos++;
778 
779 				/*
780 				 * check to make sure we're in the same object
781 				 */
782 				le_key = B_N_PDELIM_KEY(bh, pos);
783 				if (le32_to_cpu(le_key->k_objectid) !=
784 				    key->on_disk_key.k_objectid) {
785 					break;
786 				}
787 			}
788 		}
789 	}
790 }
791 
792 /* Form the path to an item and position in this item which contains
793    file byte defined by key. If there is no such item
794    corresponding to the key, we point the path to the item with
795    maximal key less than key, and *pos_in_item is set to one
796    past the last entry/byte in the item.  If searching for entry in a
797    directory item, and it is not found, *pos_in_item is set to one
798    entry more than the entry with maximal key which is less than the
799    sought key.
800 
801    Note that if there is no entry in this same node which is one more,
802    then we point to an imaginary entry.  for direct items, the
803    position is in units of bytes, for indirect items the position is
804    in units of blocknr entries, for directory items the position is in
805    units of directory entries.  */
806 
807 /* The function is NOT SCHEDULE-SAFE! */
808 int search_for_position_by_key(struct super_block *sb,	/* Pointer to the super block.          */
809 			       const struct cpu_key *p_cpu_key,	/* Key to search (cpu variable)         */
810 			       struct treepath *search_path	/* Filled up by this function.          */
811     )
812 {
813 	struct item_head *p_le_ih;	/* pointer to on-disk structure */
814 	int blk_size;
815 	loff_t item_offset, offset;
816 	struct reiserfs_dir_entry de;
817 	int retval;
818 
819 	/* If searching for directory entry. */
820 	if (is_direntry_cpu_key(p_cpu_key))
821 		return search_by_entry_key(sb, p_cpu_key, search_path,
822 					   &de);
823 
824 	/* If not searching for directory entry. */
825 
826 	/* If item is found. */
827 	retval = search_item(sb, p_cpu_key, search_path);
828 	if (retval == IO_ERROR)
829 		return retval;
830 	if (retval == ITEM_FOUND) {
831 
832 		RFALSE(!ih_item_len
833 		       (B_N_PITEM_HEAD
834 			(PATH_PLAST_BUFFER(search_path),
835 			 PATH_LAST_POSITION(search_path))),
836 		       "PAP-5165: item length equals zero");
837 
838 		pos_in_item(search_path) = 0;
839 		return POSITION_FOUND;
840 	}
841 
842 	RFALSE(!PATH_LAST_POSITION(search_path),
843 	       "PAP-5170: position equals zero");
844 
845 	/* Item is not found. Set path to the previous item. */
846 	p_le_ih =
847 	    B_N_PITEM_HEAD(PATH_PLAST_BUFFER(search_path),
848 			   --PATH_LAST_POSITION(search_path));
849 	blk_size = sb->s_blocksize;
850 
851 	if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) {
852 		return FILE_NOT_FOUND;
853 	}
854 	// FIXME: quite ugly this far
855 
856 	item_offset = le_ih_k_offset(p_le_ih);
857 	offset = cpu_key_k_offset(p_cpu_key);
858 
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, blk_size) > offset) {
862 		pos_in_item(search_path) = offset - item_offset;
863 		if (is_indirect_le_ih(p_le_ih)) {
864 			pos_in_item(search_path) /= blk_size;
865 		}
866 		return POSITION_FOUND;
867 	}
868 
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(search_path) =
873 		    ih_item_len(p_le_ih) / UNFM_P_SIZE;
874 	else
875 		pos_in_item(search_path) = ih_item_len(p_le_ih);
876 
877 	return POSITION_NOT_FOUND;
878 }
879 
880 /* Compare given item and item pointed to by the path. */
881 int comp_items(const struct item_head *stored_ih, const struct treepath *path)
882 {
883 	struct buffer_head *bh = PATH_PLAST_BUFFER(path);
884 	struct item_head *ih;
885 
886 	/* Last buffer at the path is not in the tree. */
887 	if (!B_IS_IN_TREE(bh))
888 		return 1;
889 
890 	/* Last path position is invalid. */
891 	if (PATH_LAST_POSITION(path) >= B_NR_ITEMS(bh))
892 		return 1;
893 
894 	/* we need only to know, whether it is the same item */
895 	ih = get_ih(path);
896 	return memcmp(stored_ih, ih, IH_SIZE);
897 }
898 
899 /* unformatted nodes are not logged anymore, ever.  This is safe
900 ** now
901 */
902 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
903 
904 // block can not be forgotten as it is in I/O or held by someone
905 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
906 
907 // prepare for delete or cut of direct item
908 static inline int prepare_for_direct_item(struct treepath *path,
909 					  struct item_head *le_ih,
910 					  struct inode *inode,
911 					  loff_t new_file_length, int *cut_size)
912 {
913 	loff_t round_len;
914 
915 	if (new_file_length == max_reiserfs_offset(inode)) {
916 		/* item has to be deleted */
917 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
918 		return M_DELETE;
919 	}
920 	// new file gets truncated
921 	if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
922 		//
923 		round_len = ROUND_UP(new_file_length);
924 		/* this was new_file_length < le_ih ... */
925 		if (round_len < le_ih_k_offset(le_ih)) {
926 			*cut_size = -(IH_SIZE + ih_item_len(le_ih));
927 			return M_DELETE;	/* Delete this item. */
928 		}
929 		/* Calculate first position and size for cutting from item. */
930 		pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
931 		*cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
932 
933 		return M_CUT;	/* Cut from this item. */
934 	}
935 
936 	// old file: items may have any length
937 
938 	if (new_file_length < le_ih_k_offset(le_ih)) {
939 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
940 		return M_DELETE;	/* Delete this item. */
941 	}
942 	/* Calculate first position and size for cutting from item. */
943 	*cut_size = -(ih_item_len(le_ih) -
944 		      (pos_in_item(path) =
945 		       new_file_length + 1 - le_ih_k_offset(le_ih)));
946 	return M_CUT;		/* Cut from this item. */
947 }
948 
949 static inline int prepare_for_direntry_item(struct treepath *path,
950 					    struct item_head *le_ih,
951 					    struct inode *inode,
952 					    loff_t new_file_length,
953 					    int *cut_size)
954 {
955 	if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
956 	    new_file_length == max_reiserfs_offset(inode)) {
957 		RFALSE(ih_entry_count(le_ih) != 2,
958 		       "PAP-5220: incorrect empty directory item (%h)", le_ih);
959 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
960 		return M_DELETE;	/* Delete the directory item containing "." and ".." entry. */
961 	}
962 
963 	if (ih_entry_count(le_ih) == 1) {
964 		/* Delete the directory item such as there is one record only
965 		   in this item */
966 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
967 		return M_DELETE;
968 	}
969 
970 	/* Cut one record from the directory item. */
971 	*cut_size =
972 	    -(DEH_SIZE +
973 	      entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
974 	return M_CUT;
975 }
976 
977 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
978 
979 /*  If the path points to a directory or direct item, calculate mode and the size cut, for balance.
980     If the path points to an indirect item, remove some number of its unformatted nodes.
981     In case of file truncate calculate whether this item must be deleted/truncated or last
982     unformatted node of this item will be converted to a direct item.
983     This function returns a determination of what balance mode the calling function should employ. */
984 static char prepare_for_delete_or_cut(struct reiserfs_transaction_handle *th, struct inode *inode, struct treepath *path, const struct cpu_key *item_key, int *removed,	/* Number of unformatted nodes which were removed
985 																						   from end of the file. */
986 				      int *cut_size, unsigned long long new_file_length	/* MAX_KEY_OFFSET in case of delete. */
987     )
988 {
989 	struct super_block *sb = inode->i_sb;
990 	struct item_head *p_le_ih = PATH_PITEM_HEAD(path);
991 	struct buffer_head *bh = PATH_PLAST_BUFFER(path);
992 
993 	BUG_ON(!th->t_trans_id);
994 
995 	/* Stat_data item. */
996 	if (is_statdata_le_ih(p_le_ih)) {
997 
998 		RFALSE(new_file_length != max_reiserfs_offset(inode),
999 		       "PAP-5210: mode must be M_DELETE");
1000 
1001 		*cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
1002 		return M_DELETE;
1003 	}
1004 
1005 	/* Directory item. */
1006 	if (is_direntry_le_ih(p_le_ih))
1007 		return prepare_for_direntry_item(path, p_le_ih, inode,
1008 						 new_file_length,
1009 						 cut_size);
1010 
1011 	/* Direct item. */
1012 	if (is_direct_le_ih(p_le_ih))
1013 		return prepare_for_direct_item(path, p_le_ih, inode,
1014 					       new_file_length, cut_size);
1015 
1016 	/* Case of an indirect item. */
1017 	{
1018 	    int blk_size = sb->s_blocksize;
1019 	    struct item_head s_ih;
1020 	    int need_re_search;
1021 	    int delete = 0;
1022 	    int result = M_CUT;
1023 	    int pos = 0;
1024 
1025 	    if ( new_file_length == max_reiserfs_offset (inode) ) {
1026 		/* prepare_for_delete_or_cut() is called by
1027 		 * reiserfs_delete_item() */
1028 		new_file_length = 0;
1029 		delete = 1;
1030 	    }
1031 
1032 	    do {
1033 		need_re_search = 0;
1034 		*cut_size = 0;
1035 		bh = PATH_PLAST_BUFFER(path);
1036 		copy_item_head(&s_ih, PATH_PITEM_HEAD(path));
1037 		pos = I_UNFM_NUM(&s_ih);
1038 
1039 		while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > new_file_length) {
1040 		    __le32 *unfm;
1041 		    __u32 block;
1042 
1043 		    /* Each unformatted block deletion may involve one additional
1044 		     * bitmap block into the transaction, thereby the initial
1045 		     * journal space reservation might not be enough. */
1046 		    if (!delete && (*cut_size) != 0 &&
1047 			reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD)
1048 			break;
1049 
1050 		    unfm = (__le32 *)B_I_PITEM(bh, &s_ih) + pos - 1;
1051 		    block = get_block_num(unfm, 0);
1052 
1053 		    if (block != 0) {
1054 			reiserfs_prepare_for_journal(sb, bh, 1);
1055 			put_block_num(unfm, 0, 0);
1056 			journal_mark_dirty(th, sb, bh);
1057 			reiserfs_free_block(th, inode, block, 1);
1058 		    }
1059 
1060 		    reiserfs_cond_resched(sb);
1061 
1062 		    if (item_moved (&s_ih, path))  {
1063 			need_re_search = 1;
1064 			break;
1065 		    }
1066 
1067 		    pos --;
1068 		    (*removed)++;
1069 		    (*cut_size) -= UNFM_P_SIZE;
1070 
1071 		    if (pos == 0) {
1072 			(*cut_size) -= IH_SIZE;
1073 			result = M_DELETE;
1074 			break;
1075 		    }
1076 		}
1077 		/* a trick.  If the buffer has been logged, this will do nothing.  If
1078 		** we've broken the loop without logging it, it will restore the
1079 		** buffer */
1080 		reiserfs_restore_prepared_buffer(sb, bh);
1081 	    } while (need_re_search &&
1082 		     search_for_position_by_key(sb, item_key, path) == POSITION_FOUND);
1083 	    pos_in_item(path) = pos * UNFM_P_SIZE;
1084 
1085 	    if (*cut_size == 0) {
1086 		/* Nothing were cut. maybe convert last unformatted node to the
1087 		 * direct item? */
1088 		result = M_CONVERT;
1089 	    }
1090 	    return result;
1091 	}
1092 }
1093 
1094 /* Calculate number of bytes which will be deleted or cut during balance */
1095 static int calc_deleted_bytes_number(struct tree_balance *tb, char mode)
1096 {
1097 	int del_size;
1098 	struct item_head *p_le_ih = PATH_PITEM_HEAD(tb->tb_path);
1099 
1100 	if (is_statdata_le_ih(p_le_ih))
1101 		return 0;
1102 
1103 	del_size =
1104 	    (mode ==
1105 	     M_DELETE) ? ih_item_len(p_le_ih) : -tb->insert_size[0];
1106 	if (is_direntry_le_ih(p_le_ih)) {
1107 		/* return EMPTY_DIR_SIZE; We delete emty directoris only.
1108 		 * we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1109 		 * empty size.  ick. FIXME, is this right? */
1110 		return del_size;
1111 	}
1112 
1113 	if (is_indirect_le_ih(p_le_ih))
1114 		del_size = (del_size / UNFM_P_SIZE) *
1115 				(PATH_PLAST_BUFFER(tb->tb_path)->b_size);
1116 	return del_size;
1117 }
1118 
1119 static void init_tb_struct(struct reiserfs_transaction_handle *th,
1120 			   struct tree_balance *tb,
1121 			   struct super_block *sb,
1122 			   struct treepath *path, int size)
1123 {
1124 
1125 	BUG_ON(!th->t_trans_id);
1126 
1127 	memset(tb, '\0', sizeof(struct tree_balance));
1128 	tb->transaction_handle = th;
1129 	tb->tb_sb = sb;
1130 	tb->tb_path = path;
1131 	PATH_OFFSET_PBUFFER(path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1132 	PATH_OFFSET_POSITION(path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1133 	tb->insert_size[0] = size;
1134 }
1135 
1136 void padd_item(char *item, int total_length, int length)
1137 {
1138 	int i;
1139 
1140 	for (i = total_length; i > length;)
1141 		item[--i] = 0;
1142 }
1143 
1144 #ifdef REISERQUOTA_DEBUG
1145 char key2type(struct reiserfs_key *ih)
1146 {
1147 	if (is_direntry_le_key(2, ih))
1148 		return 'd';
1149 	if (is_direct_le_key(2, ih))
1150 		return 'D';
1151 	if (is_indirect_le_key(2, ih))
1152 		return 'i';
1153 	if (is_statdata_le_key(2, ih))
1154 		return 's';
1155 	return 'u';
1156 }
1157 
1158 char head2type(struct item_head *ih)
1159 {
1160 	if (is_direntry_le_ih(ih))
1161 		return 'd';
1162 	if (is_direct_le_ih(ih))
1163 		return 'D';
1164 	if (is_indirect_le_ih(ih))
1165 		return 'i';
1166 	if (is_statdata_le_ih(ih))
1167 		return 's';
1168 	return 'u';
1169 }
1170 #endif
1171 
1172 /* Delete object item.
1173  * th       - active transaction handle
1174  * path     - path to the deleted item
1175  * item_key - key to search for the deleted item
1176  * indode   - used for updating i_blocks and quotas
1177  * un_bh    - NULL or unformatted node pointer
1178  */
1179 int reiserfs_delete_item(struct reiserfs_transaction_handle *th,
1180 			 struct treepath *path, const struct cpu_key *item_key,
1181 			 struct inode *inode, struct buffer_head *un_bh)
1182 {
1183 	struct super_block *sb = inode->i_sb;
1184 	struct tree_balance s_del_balance;
1185 	struct item_head s_ih;
1186 	struct item_head *q_ih;
1187 	int quota_cut_bytes;
1188 	int ret_value, del_size, removed;
1189 	int depth;
1190 
1191 #ifdef CONFIG_REISERFS_CHECK
1192 	char mode;
1193 	int iter = 0;
1194 #endif
1195 
1196 	BUG_ON(!th->t_trans_id);
1197 
1198 	init_tb_struct(th, &s_del_balance, sb, path,
1199 		       0 /*size is unknown */ );
1200 
1201 	while (1) {
1202 		removed = 0;
1203 
1204 #ifdef CONFIG_REISERFS_CHECK
1205 		iter++;
1206 		mode =
1207 #endif
1208 		    prepare_for_delete_or_cut(th, inode, path,
1209 					      item_key, &removed,
1210 					      &del_size,
1211 					      max_reiserfs_offset(inode));
1212 
1213 		RFALSE(mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1214 
1215 		copy_item_head(&s_ih, PATH_PITEM_HEAD(path));
1216 		s_del_balance.insert_size[0] = del_size;
1217 
1218 		ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1219 		if (ret_value != REPEAT_SEARCH)
1220 			break;
1221 
1222 		PROC_INFO_INC(sb, delete_item_restarted);
1223 
1224 		// file system changed, repeat search
1225 		ret_value =
1226 		    search_for_position_by_key(sb, item_key, path);
1227 		if (ret_value == IO_ERROR)
1228 			break;
1229 		if (ret_value == FILE_NOT_FOUND) {
1230 			reiserfs_warning(sb, "vs-5340",
1231 					 "no items of the file %K found",
1232 					 item_key);
1233 			break;
1234 		}
1235 	}			/* while (1) */
1236 
1237 	if (ret_value != CARRY_ON) {
1238 		unfix_nodes(&s_del_balance);
1239 		return 0;
1240 	}
1241 	// reiserfs_delete_item returns item length when success
1242 	ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1243 	q_ih = get_ih(path);
1244 	quota_cut_bytes = ih_item_len(q_ih);
1245 
1246 	/* hack so the quota code doesn't have to guess if the file
1247 	 ** has a tail.  On tail insert, we allocate quota for 1 unformatted node.
1248 	 ** We test the offset because the tail might have been
1249 	 ** split into multiple items, and we only want to decrement for
1250 	 ** the unfm node once
1251 	 */
1252 	if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(q_ih)) {
1253 		if ((le_ih_k_offset(q_ih) & (sb->s_blocksize - 1)) == 1) {
1254 			quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
1255 		} else {
1256 			quota_cut_bytes = 0;
1257 		}
1258 	}
1259 
1260 	if (un_bh) {
1261 		int off;
1262 		char *data;
1263 
1264 		/* We are in direct2indirect conversion, so move tail contents
1265 		   to the unformatted node */
1266 		/* note, we do the copy before preparing the buffer because we
1267 		 ** don't care about the contents of the unformatted node yet.
1268 		 ** the only thing we really care about is the direct item's data
1269 		 ** is in the unformatted node.
1270 		 **
1271 		 ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1272 		 ** the unformatted node, which might schedule, meaning we'd have to
1273 		 ** loop all the way back up to the start of the while loop.
1274 		 **
1275 		 ** The unformatted node must be dirtied later on.  We can't be
1276 		 ** sure here if the entire tail has been deleted yet.
1277 		 **
1278 		 ** un_bh is from the page cache (all unformatted nodes are
1279 		 ** from the page cache) and might be a highmem page.  So, we
1280 		 ** can't use un_bh->b_data.
1281 		 ** -clm
1282 		 */
1283 
1284 		data = kmap_atomic(un_bh->b_page);
1285 		off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1286 		memcpy(data + off,
1287 		       B_I_PITEM(PATH_PLAST_BUFFER(path), &s_ih),
1288 		       ret_value);
1289 		kunmap_atomic(data);
1290 	}
1291 	/* Perform balancing after all resources have been collected at once. */
1292 	do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1293 
1294 #ifdef REISERQUOTA_DEBUG
1295 	reiserfs_debug(sb, REISERFS_DEBUG_CODE,
1296 		       "reiserquota delete_item(): freeing %u, id=%u type=%c",
1297 		       quota_cut_bytes, inode->i_uid, head2type(&s_ih));
1298 #endif
1299 	depth = reiserfs_write_unlock_nested(inode->i_sb);
1300 	dquot_free_space_nodirty(inode, quota_cut_bytes);
1301 	reiserfs_write_lock_nested(inode->i_sb, depth);
1302 
1303 	/* Return deleted body length */
1304 	return ret_value;
1305 }
1306 
1307 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1308 
1309  deletion of the body of the object is performed by iput(), with the
1310  result that if multiple processes are operating on a file, the
1311  deletion of the body of the file is deferred until the last process
1312  that has an open inode performs its iput().
1313 
1314  writes and truncates are protected from collisions by use of
1315  semaphores.
1316 
1317  creates, linking, and mknod are protected from collisions with other
1318  processes by making the reiserfs_add_entry() the last step in the
1319  creation, and then rolling back all changes if there was a collision.
1320  - Hans
1321 */
1322 
1323 /* this deletes item which never gets split */
1324 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1325 				struct inode *inode, struct reiserfs_key *key)
1326 {
1327 	struct super_block *sb = th->t_super;
1328 	struct tree_balance tb;
1329 	INITIALIZE_PATH(path);
1330 	int item_len = 0;
1331 	int tb_init = 0;
1332 	struct cpu_key cpu_key;
1333 	int retval;
1334 	int quota_cut_bytes = 0;
1335 
1336 	BUG_ON(!th->t_trans_id);
1337 
1338 	le_key2cpu_key(&cpu_key, key);
1339 
1340 	while (1) {
1341 		retval = search_item(th->t_super, &cpu_key, &path);
1342 		if (retval == IO_ERROR) {
1343 			reiserfs_error(th->t_super, "vs-5350",
1344 				       "i/o failure occurred trying "
1345 				       "to delete %K", &cpu_key);
1346 			break;
1347 		}
1348 		if (retval != ITEM_FOUND) {
1349 			pathrelse(&path);
1350 			// No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1351 			if (!
1352 			    ((unsigned long long)
1353 			     GET_HASH_VALUE(le_key_k_offset
1354 					    (le_key_version(key), key)) == 0
1355 			     && (unsigned long long)
1356 			     GET_GENERATION_NUMBER(le_key_k_offset
1357 						   (le_key_version(key),
1358 						    key)) == 1))
1359 				reiserfs_warning(th->t_super, "vs-5355",
1360 						 "%k not found", key);
1361 			break;
1362 		}
1363 		if (!tb_init) {
1364 			tb_init = 1;
1365 			item_len = ih_item_len(PATH_PITEM_HEAD(&path));
1366 			init_tb_struct(th, &tb, th->t_super, &path,
1367 				       -(IH_SIZE + item_len));
1368 		}
1369 		quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path));
1370 
1371 		retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1372 		if (retval == REPEAT_SEARCH) {
1373 			PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1374 			continue;
1375 		}
1376 
1377 		if (retval == CARRY_ON) {
1378 			do_balance(&tb, NULL, NULL, M_DELETE);
1379 			if (inode) {	/* Should we count quota for item? (we don't count quotas for save-links) */
1380 				int depth;
1381 #ifdef REISERQUOTA_DEBUG
1382 				reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1383 					       "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1384 					       quota_cut_bytes, inode->i_uid,
1385 					       key2type(key));
1386 #endif
1387 				depth = reiserfs_write_unlock_nested(sb);
1388 				dquot_free_space_nodirty(inode,
1389 							 quota_cut_bytes);
1390 				reiserfs_write_lock_nested(sb, depth);
1391 			}
1392 			break;
1393 		}
1394 		// IO_ERROR, NO_DISK_SPACE, etc
1395 		reiserfs_warning(th->t_super, "vs-5360",
1396 				 "could not delete %K due to fix_nodes failure",
1397 				 &cpu_key);
1398 		unfix_nodes(&tb);
1399 		break;
1400 	}
1401 
1402 	reiserfs_check_path(&path);
1403 }
1404 
1405 int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1406 			   struct inode *inode)
1407 {
1408 	int err;
1409 	inode->i_size = 0;
1410 	BUG_ON(!th->t_trans_id);
1411 
1412 	/* for directory this deletes item containing "." and ".." */
1413 	err =
1414 	    reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1415 	if (err)
1416 		return err;
1417 
1418 #if defined( USE_INODE_GENERATION_COUNTER )
1419 	if (!old_format_only(th->t_super)) {
1420 		__le32 *inode_generation;
1421 
1422 		inode_generation =
1423 		    &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
1424 		le32_add_cpu(inode_generation, 1);
1425 	}
1426 /* USE_INODE_GENERATION_COUNTER */
1427 #endif
1428 	reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1429 
1430 	return err;
1431 }
1432 
1433 static void unmap_buffers(struct page *page, loff_t pos)
1434 {
1435 	struct buffer_head *bh;
1436 	struct buffer_head *head;
1437 	struct buffer_head *next;
1438 	unsigned long tail_index;
1439 	unsigned long cur_index;
1440 
1441 	if (page) {
1442 		if (page_has_buffers(page)) {
1443 			tail_index = pos & (PAGE_CACHE_SIZE - 1);
1444 			cur_index = 0;
1445 			head = page_buffers(page);
1446 			bh = head;
1447 			do {
1448 				next = bh->b_this_page;
1449 
1450 				/* we want to unmap the buffers that contain the tail, and
1451 				 ** all the buffers after it (since the tail must be at the
1452 				 ** end of the file).  We don't want to unmap file data
1453 				 ** before the tail, since it might be dirty and waiting to
1454 				 ** reach disk
1455 				 */
1456 				cur_index += bh->b_size;
1457 				if (cur_index > tail_index) {
1458 					reiserfs_unmap_buffer(bh);
1459 				}
1460 				bh = next;
1461 			} while (bh != head);
1462 		}
1463 	}
1464 }
1465 
1466 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
1467 				    struct inode *inode,
1468 				    struct page *page,
1469 				    struct treepath *path,
1470 				    const struct cpu_key *item_key,
1471 				    loff_t new_file_size, char *mode)
1472 {
1473 	struct super_block *sb = inode->i_sb;
1474 	int block_size = sb->s_blocksize;
1475 	int cut_bytes;
1476 	BUG_ON(!th->t_trans_id);
1477 	BUG_ON(new_file_size != inode->i_size);
1478 
1479 	/* the page being sent in could be NULL if there was an i/o error
1480 	 ** reading in the last block.  The user will hit problems trying to
1481 	 ** read the file, but for now we just skip the indirect2direct
1482 	 */
1483 	if (atomic_read(&inode->i_count) > 1 ||
1484 	    !tail_has_to_be_packed(inode) ||
1485 	    !page || (REISERFS_I(inode)->i_flags & i_nopack_mask)) {
1486 		/* leave tail in an unformatted node */
1487 		*mode = M_SKIP_BALANCING;
1488 		cut_bytes =
1489 		    block_size - (new_file_size & (block_size - 1));
1490 		pathrelse(path);
1491 		return cut_bytes;
1492 	}
1493 	/* Perform the conversion to a direct_item. */
1494 	/* return indirect_to_direct(inode, path, item_key,
1495 				  new_file_size, mode); */
1496 	return indirect2direct(th, inode, page, path, item_key,
1497 			       new_file_size, mode);
1498 }
1499 
1500 /* we did indirect_to_direct conversion. And we have inserted direct
1501    item successesfully, but there were no disk space to cut unfm
1502    pointer being converted. Therefore we have to delete inserted
1503    direct item(s) */
1504 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
1505 					 struct inode *inode, struct treepath *path)
1506 {
1507 	struct cpu_key tail_key;
1508 	int tail_len;
1509 	int removed;
1510 	BUG_ON(!th->t_trans_id);
1511 
1512 	make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);	// !!!!
1513 	tail_key.key_length = 4;
1514 
1515 	tail_len =
1516 	    (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1517 	while (tail_len) {
1518 		/* look for the last byte of the tail */
1519 		if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1520 		    POSITION_NOT_FOUND)
1521 			reiserfs_panic(inode->i_sb, "vs-5615",
1522 				       "found invalid item");
1523 		RFALSE(path->pos_in_item !=
1524 		       ih_item_len(PATH_PITEM_HEAD(path)) - 1,
1525 		       "vs-5616: appended bytes found");
1526 		PATH_LAST_POSITION(path)--;
1527 
1528 		removed =
1529 		    reiserfs_delete_item(th, path, &tail_key, inode,
1530 					 NULL /*unbh not needed */ );
1531 		RFALSE(removed <= 0
1532 		       || removed > tail_len,
1533 		       "vs-5617: there was tail %d bytes, removed item length %d bytes",
1534 		       tail_len, removed);
1535 		tail_len -= removed;
1536 		set_cpu_key_k_offset(&tail_key,
1537 				     cpu_key_k_offset(&tail_key) - removed);
1538 	}
1539 	reiserfs_warning(inode->i_sb, "reiserfs-5091", "indirect_to_direct "
1540 			 "conversion has been rolled back due to "
1541 			 "lack of disk space");
1542 	//mark_file_without_tail (inode);
1543 	mark_inode_dirty(inode);
1544 }
1545 
1546 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1547 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
1548 			   struct treepath *path,
1549 			   struct cpu_key *item_key,
1550 			   struct inode *inode,
1551 			   struct page *page, loff_t new_file_size)
1552 {
1553 	struct super_block *sb = inode->i_sb;
1554 	/* Every function which is going to call do_balance must first
1555 	   create a tree_balance structure.  Then it must fill up this
1556 	   structure by using the init_tb_struct and fix_nodes functions.
1557 	   After that we can make tree balancing. */
1558 	struct tree_balance s_cut_balance;
1559 	struct item_head *p_le_ih;
1560 	int cut_size = 0,	/* Amount to be cut. */
1561 	    ret_value = CARRY_ON, removed = 0,	/* Number of the removed unformatted nodes. */
1562 	    is_inode_locked = 0;
1563 	char mode;		/* Mode of the balance. */
1564 	int retval2 = -1;
1565 	int quota_cut_bytes;
1566 	loff_t tail_pos = 0;
1567 	int depth;
1568 
1569 	BUG_ON(!th->t_trans_id);
1570 
1571 	init_tb_struct(th, &s_cut_balance, inode->i_sb, path,
1572 		       cut_size);
1573 
1574 	/* Repeat this loop until we either cut the item without needing
1575 	   to balance, or we fix_nodes without schedule occurring */
1576 	while (1) {
1577 		/* Determine the balance mode, position of the first byte to
1578 		   be cut, and size to be cut.  In case of the indirect item
1579 		   free unformatted nodes which are pointed to by the cut
1580 		   pointers. */
1581 
1582 		mode =
1583 		    prepare_for_delete_or_cut(th, inode, path,
1584 					      item_key, &removed,
1585 					      &cut_size, new_file_size);
1586 		if (mode == M_CONVERT) {
1587 			/* convert last unformatted node to direct item or leave
1588 			   tail in the unformatted node */
1589 			RFALSE(ret_value != CARRY_ON,
1590 			       "PAP-5570: can not convert twice");
1591 
1592 			ret_value =
1593 			    maybe_indirect_to_direct(th, inode, page,
1594 						     path, item_key,
1595 						     new_file_size, &mode);
1596 			if (mode == M_SKIP_BALANCING)
1597 				/* tail has been left in the unformatted node */
1598 				return ret_value;
1599 
1600 			is_inode_locked = 1;
1601 
1602 			/* removing of last unformatted node will change value we
1603 			   have to return to truncate. Save it */
1604 			retval2 = ret_value;
1605 			/*retval2 = sb->s_blocksize - (new_file_size & (sb->s_blocksize - 1)); */
1606 
1607 			/* So, we have performed the first part of the conversion:
1608 			   inserting the new direct item.  Now we are removing the
1609 			   last unformatted node pointer. Set key to search for
1610 			   it. */
1611 			set_cpu_key_k_type(item_key, TYPE_INDIRECT);
1612 			item_key->key_length = 4;
1613 			new_file_size -=
1614 			    (new_file_size & (sb->s_blocksize - 1));
1615 			tail_pos = new_file_size;
1616 			set_cpu_key_k_offset(item_key, new_file_size + 1);
1617 			if (search_for_position_by_key
1618 			    (sb, item_key,
1619 			     path) == POSITION_NOT_FOUND) {
1620 				print_block(PATH_PLAST_BUFFER(path), 3,
1621 					    PATH_LAST_POSITION(path) - 1,
1622 					    PATH_LAST_POSITION(path) + 1);
1623 				reiserfs_panic(sb, "PAP-5580", "item to "
1624 					       "convert does not exist (%K)",
1625 					       item_key);
1626 			}
1627 			continue;
1628 		}
1629 		if (cut_size == 0) {
1630 			pathrelse(path);
1631 			return 0;
1632 		}
1633 
1634 		s_cut_balance.insert_size[0] = cut_size;
1635 
1636 		ret_value = fix_nodes(mode, &s_cut_balance, NULL, NULL);
1637 		if (ret_value != REPEAT_SEARCH)
1638 			break;
1639 
1640 		PROC_INFO_INC(sb, cut_from_item_restarted);
1641 
1642 		ret_value =
1643 		    search_for_position_by_key(sb, item_key, path);
1644 		if (ret_value == POSITION_FOUND)
1645 			continue;
1646 
1647 		reiserfs_warning(sb, "PAP-5610", "item %K not found",
1648 				 item_key);
1649 		unfix_nodes(&s_cut_balance);
1650 		return (ret_value == IO_ERROR) ? -EIO : -ENOENT;
1651 	}			/* while */
1652 
1653 	// check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1654 	if (ret_value != CARRY_ON) {
1655 		if (is_inode_locked) {
1656 			// FIXME: this seems to be not needed: we are always able
1657 			// to cut item
1658 			indirect_to_direct_roll_back(th, inode, path);
1659 		}
1660 		if (ret_value == NO_DISK_SPACE)
1661 			reiserfs_warning(sb, "reiserfs-5092",
1662 					 "NO_DISK_SPACE");
1663 		unfix_nodes(&s_cut_balance);
1664 		return -EIO;
1665 	}
1666 
1667 	/* go ahead and perform balancing */
1668 
1669 	RFALSE(mode == M_PASTE || mode == M_INSERT, "invalid mode");
1670 
1671 	/* Calculate number of bytes that need to be cut from the item. */
1672 	quota_cut_bytes =
1673 	    (mode ==
1674 	     M_DELETE) ? ih_item_len(get_ih(path)) : -s_cut_balance.
1675 	    insert_size[0];
1676 	if (retval2 == -1)
1677 		ret_value = calc_deleted_bytes_number(&s_cut_balance, mode);
1678 	else
1679 		ret_value = retval2;
1680 
1681 	/* For direct items, we only change the quota when deleting the last
1682 	 ** item.
1683 	 */
1684 	p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path);
1685 	if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1686 		if (mode == M_DELETE &&
1687 		    (le_ih_k_offset(p_le_ih) & (sb->s_blocksize - 1)) ==
1688 		    1) {
1689 			// FIXME: this is to keep 3.5 happy
1690 			REISERFS_I(inode)->i_first_direct_byte = U32_MAX;
1691 			quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
1692 		} else {
1693 			quota_cut_bytes = 0;
1694 		}
1695 	}
1696 #ifdef CONFIG_REISERFS_CHECK
1697 	if (is_inode_locked) {
1698 		struct item_head *le_ih =
1699 		    PATH_PITEM_HEAD(s_cut_balance.tb_path);
1700 		/* we are going to complete indirect2direct conversion. Make
1701 		   sure, that we exactly remove last unformatted node pointer
1702 		   of the item */
1703 		if (!is_indirect_le_ih(le_ih))
1704 			reiserfs_panic(sb, "vs-5652",
1705 				       "item must be indirect %h", le_ih);
1706 
1707 		if (mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1708 			reiserfs_panic(sb, "vs-5653", "completing "
1709 				       "indirect2direct conversion indirect "
1710 				       "item %h being deleted must be of "
1711 				       "4 byte long", le_ih);
1712 
1713 		if (mode == M_CUT
1714 		    && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1715 			reiserfs_panic(sb, "vs-5654", "can not complete "
1716 				       "indirect2direct conversion of %h "
1717 				       "(CUT, insert_size==%d)",
1718 				       le_ih, s_cut_balance.insert_size[0]);
1719 		}
1720 		/* it would be useful to make sure, that right neighboring
1721 		   item is direct item of this file */
1722 	}
1723 #endif
1724 
1725 	do_balance(&s_cut_balance, NULL, NULL, mode);
1726 	if (is_inode_locked) {
1727 		/* we've done an indirect->direct conversion.  when the data block
1728 		 ** was freed, it was removed from the list of blocks that must
1729 		 ** be flushed before the transaction commits, make sure to
1730 		 ** unmap and invalidate it
1731 		 */
1732 		unmap_buffers(page, tail_pos);
1733 		REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
1734 	}
1735 #ifdef REISERQUOTA_DEBUG
1736 	reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1737 		       "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1738 		       quota_cut_bytes, inode->i_uid, '?');
1739 #endif
1740 	depth = reiserfs_write_unlock_nested(sb);
1741 	dquot_free_space_nodirty(inode, quota_cut_bytes);
1742 	reiserfs_write_lock_nested(sb, depth);
1743 	return ret_value;
1744 }
1745 
1746 static void truncate_directory(struct reiserfs_transaction_handle *th,
1747 			       struct inode *inode)
1748 {
1749 	BUG_ON(!th->t_trans_id);
1750 	if (inode->i_nlink)
1751 		reiserfs_error(inode->i_sb, "vs-5655", "link count != 0");
1752 
1753 	set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1754 	set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1755 	reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1756 	reiserfs_update_sd(th, inode);
1757 	set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1758 	set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
1759 }
1760 
1761 /* Truncate file to the new size. Note, this must be called with a transaction
1762    already started */
1763 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th,
1764 			  struct inode *inode,	/* ->i_size contains new size */
1765 			 struct page *page,	/* up to date for last block */
1766 			 int update_timestamps	/* when it is called by
1767 						   file_release to convert
1768 						   the tail - no timestamps
1769 						   should be updated */
1770     )
1771 {
1772 	INITIALIZE_PATH(s_search_path);	/* Path to the current object item. */
1773 	struct item_head *p_le_ih;	/* Pointer to an item header. */
1774 	struct cpu_key s_item_key;	/* Key to search for a previous file item. */
1775 	loff_t file_size,	/* Old file size. */
1776 	 new_file_size;	/* New file size. */
1777 	int deleted;		/* Number of deleted or truncated bytes. */
1778 	int retval;
1779 	int err = 0;
1780 
1781 	BUG_ON(!th->t_trans_id);
1782 	if (!
1783 	    (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1784 	     || S_ISLNK(inode->i_mode)))
1785 		return 0;
1786 
1787 	if (S_ISDIR(inode->i_mode)) {
1788 		// deletion of directory - no need to update timestamps
1789 		truncate_directory(th, inode);
1790 		return 0;
1791 	}
1792 
1793 	/* Get new file size. */
1794 	new_file_size = inode->i_size;
1795 
1796 	// FIXME: note, that key type is unimportant here
1797 	make_cpu_key(&s_item_key, inode, max_reiserfs_offset(inode),
1798 		     TYPE_DIRECT, 3);
1799 
1800 	retval =
1801 	    search_for_position_by_key(inode->i_sb, &s_item_key,
1802 				       &s_search_path);
1803 	if (retval == IO_ERROR) {
1804 		reiserfs_error(inode->i_sb, "vs-5657",
1805 			       "i/o failure occurred trying to truncate %K",
1806 			       &s_item_key);
1807 		err = -EIO;
1808 		goto out;
1809 	}
1810 	if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1811 		reiserfs_error(inode->i_sb, "PAP-5660",
1812 			       "wrong result %d of search for %K", retval,
1813 			       &s_item_key);
1814 
1815 		err = -EIO;
1816 		goto out;
1817 	}
1818 
1819 	s_search_path.pos_in_item--;
1820 
1821 	/* Get real file size (total length of all file items) */
1822 	p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1823 	if (is_statdata_le_ih(p_le_ih))
1824 		file_size = 0;
1825 	else {
1826 		loff_t offset = le_ih_k_offset(p_le_ih);
1827 		int bytes =
1828 		    op_bytes_number(p_le_ih, inode->i_sb->s_blocksize);
1829 
1830 		/* this may mismatch with real file size: if last direct item
1831 		   had no padding zeros and last unformatted node had no free
1832 		   space, this file would have this file size */
1833 		file_size = offset + bytes - 1;
1834 	}
1835 	/*
1836 	 * are we doing a full truncate or delete, if so
1837 	 * kick in the reada code
1838 	 */
1839 	if (new_file_size == 0)
1840 		s_search_path.reada = PATH_READA | PATH_READA_BACK;
1841 
1842 	if (file_size == 0 || file_size < new_file_size) {
1843 		goto update_and_out;
1844 	}
1845 
1846 	/* Update key to search for the last file item. */
1847 	set_cpu_key_k_offset(&s_item_key, file_size);
1848 
1849 	do {
1850 		/* Cut or delete file item. */
1851 		deleted =
1852 		    reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
1853 					   inode, page, new_file_size);
1854 		if (deleted < 0) {
1855 			reiserfs_warning(inode->i_sb, "vs-5665",
1856 					 "reiserfs_cut_from_item failed");
1857 			reiserfs_check_path(&s_search_path);
1858 			return 0;
1859 		}
1860 
1861 		RFALSE(deleted > file_size,
1862 		       "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1863 		       deleted, file_size, &s_item_key);
1864 
1865 		/* Change key to search the last file item. */
1866 		file_size -= deleted;
1867 
1868 		set_cpu_key_k_offset(&s_item_key, file_size);
1869 
1870 		/* While there are bytes to truncate and previous file item is presented in the tree. */
1871 
1872 		/*
1873 		 ** This loop could take a really long time, and could log
1874 		 ** many more blocks than a transaction can hold.  So, we do a polite
1875 		 ** journal end here, and if the transaction needs ending, we make
1876 		 ** sure the file is consistent before ending the current trans
1877 		 ** and starting a new one
1878 		 */
1879 		if (journal_transaction_should_end(th, 0) ||
1880 		    reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1881 			int orig_len_alloc = th->t_blocks_allocated;
1882 			pathrelse(&s_search_path);
1883 
1884 			if (update_timestamps) {
1885 				inode->i_mtime = CURRENT_TIME_SEC;
1886 				inode->i_ctime = CURRENT_TIME_SEC;
1887 			}
1888 			reiserfs_update_sd(th, inode);
1889 
1890 			err = journal_end(th, inode->i_sb, orig_len_alloc);
1891 			if (err)
1892 				goto out;
1893 			err = journal_begin(th, inode->i_sb,
1894 					    JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ;
1895 			if (err)
1896 				goto out;
1897 			reiserfs_update_inode_transaction(inode);
1898 		}
1899 	} while (file_size > ROUND_UP(new_file_size) &&
1900 		 search_for_position_by_key(inode->i_sb, &s_item_key,
1901 					    &s_search_path) == POSITION_FOUND);
1902 
1903 	RFALSE(file_size > ROUND_UP(new_file_size),
1904 	       "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1905 	       new_file_size, file_size, s_item_key.on_disk_key.k_objectid);
1906 
1907       update_and_out:
1908 	if (update_timestamps) {
1909 		// this is truncate, not file closing
1910 		inode->i_mtime = CURRENT_TIME_SEC;
1911 		inode->i_ctime = CURRENT_TIME_SEC;
1912 	}
1913 	reiserfs_update_sd(th, inode);
1914 
1915       out:
1916 	pathrelse(&s_search_path);
1917 	return err;
1918 }
1919 
1920 #ifdef CONFIG_REISERFS_CHECK
1921 // this makes sure, that we __append__, not overwrite or add holes
1922 static void check_research_for_paste(struct treepath *path,
1923 				     const struct cpu_key *key)
1924 {
1925 	struct item_head *found_ih = get_ih(path);
1926 
1927 	if (is_direct_le_ih(found_ih)) {
1928 		if (le_ih_k_offset(found_ih) +
1929 		    op_bytes_number(found_ih,
1930 				    get_last_bh(path)->b_size) !=
1931 		    cpu_key_k_offset(key)
1932 		    || op_bytes_number(found_ih,
1933 				       get_last_bh(path)->b_size) !=
1934 		    pos_in_item(path))
1935 			reiserfs_panic(NULL, "PAP-5720", "found direct item "
1936 				       "%h or position (%d) does not match "
1937 				       "to key %K", found_ih,
1938 				       pos_in_item(path), key);
1939 	}
1940 	if (is_indirect_le_ih(found_ih)) {
1941 		if (le_ih_k_offset(found_ih) +
1942 		    op_bytes_number(found_ih,
1943 				    get_last_bh(path)->b_size) !=
1944 		    cpu_key_k_offset(key)
1945 		    || I_UNFM_NUM(found_ih) != pos_in_item(path)
1946 		    || get_ih_free_space(found_ih) != 0)
1947 			reiserfs_panic(NULL, "PAP-5730", "found indirect "
1948 				       "item (%h) or position (%d) does not "
1949 				       "match to key (%K)",
1950 				       found_ih, pos_in_item(path), key);
1951 	}
1952 }
1953 #endif				/* config reiserfs check */
1954 
1955 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1956 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct treepath *search_path,	/* Path to the pasted item.	  */
1957 			     const struct cpu_key *key,	/* Key to search for the needed item. */
1958 			     struct inode *inode,	/* Inode item belongs to */
1959 			     const char *body,	/* Pointer to the bytes to paste.    */
1960 			     int pasted_size)
1961 {				/* Size of pasted bytes.             */
1962 	struct super_block *sb = inode->i_sb;
1963 	struct tree_balance s_paste_balance;
1964 	int retval;
1965 	int fs_gen;
1966 	int depth;
1967 
1968 	BUG_ON(!th->t_trans_id);
1969 
1970 	fs_gen = get_generation(inode->i_sb);
1971 
1972 #ifdef REISERQUOTA_DEBUG
1973 	reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1974 		       "reiserquota paste_into_item(): allocating %u id=%u type=%c",
1975 		       pasted_size, inode->i_uid,
1976 		       key2type(&(key->on_disk_key)));
1977 #endif
1978 
1979 	depth = reiserfs_write_unlock_nested(sb);
1980 	retval = dquot_alloc_space_nodirty(inode, pasted_size);
1981 	reiserfs_write_lock_nested(sb, depth);
1982 	if (retval) {
1983 		pathrelse(search_path);
1984 		return retval;
1985 	}
1986 	init_tb_struct(th, &s_paste_balance, th->t_super, search_path,
1987 		       pasted_size);
1988 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1989 	s_paste_balance.key = key->on_disk_key;
1990 #endif
1991 
1992 	/* DQUOT_* can schedule, must check before the fix_nodes */
1993 	if (fs_changed(fs_gen, inode->i_sb)) {
1994 		goto search_again;
1995 	}
1996 
1997 	while ((retval =
1998 		fix_nodes(M_PASTE, &s_paste_balance, NULL,
1999 			  body)) == REPEAT_SEARCH) {
2000 	      search_again:
2001 		/* file system changed while we were in the fix_nodes */
2002 		PROC_INFO_INC(th->t_super, paste_into_item_restarted);
2003 		retval =
2004 		    search_for_position_by_key(th->t_super, key,
2005 					       search_path);
2006 		if (retval == IO_ERROR) {
2007 			retval = -EIO;
2008 			goto error_out;
2009 		}
2010 		if (retval == POSITION_FOUND) {
2011 			reiserfs_warning(inode->i_sb, "PAP-5710",
2012 					 "entry or pasted byte (%K) exists",
2013 					 key);
2014 			retval = -EEXIST;
2015 			goto error_out;
2016 		}
2017 #ifdef CONFIG_REISERFS_CHECK
2018 		check_research_for_paste(search_path, key);
2019 #endif
2020 	}
2021 
2022 	/* Perform balancing after all resources are collected by fix_nodes, and
2023 	   accessing them will not risk triggering schedule. */
2024 	if (retval == CARRY_ON) {
2025 		do_balance(&s_paste_balance, NULL /*ih */ , body, M_PASTE);
2026 		return 0;
2027 	}
2028 	retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2029       error_out:
2030 	/* this also releases the path */
2031 	unfix_nodes(&s_paste_balance);
2032 #ifdef REISERQUOTA_DEBUG
2033 	reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2034 		       "reiserquota paste_into_item(): freeing %u id=%u type=%c",
2035 		       pasted_size, inode->i_uid,
2036 		       key2type(&(key->on_disk_key)));
2037 #endif
2038 	depth = reiserfs_write_unlock_nested(sb);
2039 	dquot_free_space_nodirty(inode, pasted_size);
2040 	reiserfs_write_lock_nested(sb, depth);
2041 	return retval;
2042 }
2043 
2044 /* Insert new item into the buffer at the path.
2045  * th   - active transaction handle
2046  * path - path to the inserted item
2047  * ih   - pointer to the item header to insert
2048  * body - pointer to the bytes to insert
2049  */
2050 int reiserfs_insert_item(struct reiserfs_transaction_handle *th,
2051 			 struct treepath *path, const struct cpu_key *key,
2052 			 struct item_head *ih, struct inode *inode,
2053 			 const char *body)
2054 {
2055 	struct tree_balance s_ins_balance;
2056 	int retval;
2057 	int fs_gen = 0;
2058 	int quota_bytes = 0;
2059 
2060 	BUG_ON(!th->t_trans_id);
2061 
2062 	if (inode) {		/* Do we count quotas for item? */
2063 		int depth;
2064 		fs_gen = get_generation(inode->i_sb);
2065 		quota_bytes = ih_item_len(ih);
2066 
2067 		/* hack so the quota code doesn't have to guess if the file has
2068 		 ** a tail, links are always tails, so there's no guessing needed
2069 		 */
2070 		if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(ih))
2071 			quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
2072 #ifdef REISERQUOTA_DEBUG
2073 		reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2074 			       "reiserquota insert_item(): allocating %u id=%u type=%c",
2075 			       quota_bytes, inode->i_uid, head2type(ih));
2076 #endif
2077 		/* We can't dirty inode here. It would be immediately written but
2078 		 * appropriate stat item isn't inserted yet... */
2079 		depth = reiserfs_write_unlock_nested(inode->i_sb);
2080 		retval = dquot_alloc_space_nodirty(inode, quota_bytes);
2081 		reiserfs_write_lock_nested(inode->i_sb, depth);
2082 		if (retval) {
2083 			pathrelse(path);
2084 			return retval;
2085 		}
2086 	}
2087 	init_tb_struct(th, &s_ins_balance, th->t_super, path,
2088 		       IH_SIZE + ih_item_len(ih));
2089 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2090 	s_ins_balance.key = key->on_disk_key;
2091 #endif
2092 	/* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2093 	if (inode && fs_changed(fs_gen, inode->i_sb)) {
2094 		goto search_again;
2095 	}
2096 
2097 	while ((retval =
2098 		fix_nodes(M_INSERT, &s_ins_balance, ih,
2099 			  body)) == REPEAT_SEARCH) {
2100 	      search_again:
2101 		/* file system changed while we were in the fix_nodes */
2102 		PROC_INFO_INC(th->t_super, insert_item_restarted);
2103 		retval = search_item(th->t_super, key, path);
2104 		if (retval == IO_ERROR) {
2105 			retval = -EIO;
2106 			goto error_out;
2107 		}
2108 		if (retval == ITEM_FOUND) {
2109 			reiserfs_warning(th->t_super, "PAP-5760",
2110 					 "key %K already exists in the tree",
2111 					 key);
2112 			retval = -EEXIST;
2113 			goto error_out;
2114 		}
2115 	}
2116 
2117 	/* make balancing after all resources will be collected at a time */
2118 	if (retval == CARRY_ON) {
2119 		do_balance(&s_ins_balance, ih, body, M_INSERT);
2120 		return 0;
2121 	}
2122 
2123 	retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2124       error_out:
2125 	/* also releases the path */
2126 	unfix_nodes(&s_ins_balance);
2127 #ifdef REISERQUOTA_DEBUG
2128 	reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2129 		       "reiserquota insert_item(): freeing %u id=%u type=%c",
2130 		       quota_bytes, inode->i_uid, head2type(ih));
2131 #endif
2132 	if (inode) {
2133 		int depth = reiserfs_write_unlock_nested(inode->i_sb);
2134 		dquot_free_space_nodirty(inode, quota_bytes);
2135 		reiserfs_write_lock_nested(inode->i_sb, depth);
2136 	}
2137 	return retval;
2138 }
2139