xref: /openbmc/linux/fs/reiserfs/stree.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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  * decrement_counters_in_path
27  * reiserfs_check_path
28  * pathrelse_and_restore
29  * pathrelse
30  * search_by_key_reada
31  * search_by_key
32  * search_for_position_by_key
33  * comp_items
34  * prepare_for_direct_item
35  * prepare_for_direntry_item
36  * prepare_for_delete_or_cut
37  * calc_deleted_bytes_number
38  * init_tb_struct
39  * padd_item
40  * reiserfs_delete_item
41  * reiserfs_delete_solid_item
42  * reiserfs_delete_object
43  * maybe_indirect_to_direct
44  * indirect_to_direct_roll_back
45  * reiserfs_cut_from_item
46  * truncate_directory
47  * reiserfs_do_truncate
48  * reiserfs_paste_into_item
49  * reiserfs_insert_item
50  */
51 
52 #include <linux/config.h>
53 #include <linux/time.h>
54 #include <linux/string.h>
55 #include <linux/pagemap.h>
56 #include <linux/reiserfs_fs.h>
57 #include <linux/smp_lock.h>
58 #include <linux/buffer_head.h>
59 #include <linux/quotaops.h>
60 
61 /* Does the buffer contain a disk block which is in the tree. */
62 inline int B_IS_IN_TREE(const struct buffer_head *p_s_bh)
63 {
64 
65 	RFALSE(B_LEVEL(p_s_bh) > MAX_HEIGHT,
66 	       "PAP-1010: block (%b) has too big level (%z)", p_s_bh, p_s_bh);
67 
68 	return (B_LEVEL(p_s_bh) != FREE_LEVEL);
69 }
70 
71 //
72 // to gets item head in le form
73 //
74 inline void copy_item_head(struct item_head *p_v_to,
75 			   const struct item_head *p_v_from)
76 {
77 	memcpy(p_v_to, p_v_from, IH_SIZE);
78 }
79 
80 /* k1 is pointer to on-disk structure which is stored in little-endian
81    form. k2 is pointer to cpu variable. For key of items of the same
82    object this returns 0.
83    Returns: -1 if key1 < key2
84    0 if key1 == key2
85    1 if key1 > key2 */
86 inline int comp_short_keys(const struct reiserfs_key *le_key,
87 			   const struct cpu_key *cpu_key)
88 {
89 	__u32 n;
90 	n = le32_to_cpu(le_key->k_dir_id);
91 	if (n < cpu_key->on_disk_key.k_dir_id)
92 		return -1;
93 	if (n > cpu_key->on_disk_key.k_dir_id)
94 		return 1;
95 	n = le32_to_cpu(le_key->k_objectid);
96 	if (n < cpu_key->on_disk_key.k_objectid)
97 		return -1;
98 	if (n > cpu_key->on_disk_key.k_objectid)
99 		return 1;
100 	return 0;
101 }
102 
103 /* k1 is pointer to on-disk structure which is stored in little-endian
104    form. k2 is pointer to cpu variable.
105    Compare keys using all 4 key fields.
106    Returns: -1 if key1 < key2 0
107    if key1 = key2 1 if key1 > key2 */
108 static inline int comp_keys(const struct reiserfs_key *le_key,
109 			    const struct cpu_key *cpu_key)
110 {
111 	int retval;
112 
113 	retval = comp_short_keys(le_key, cpu_key);
114 	if (retval)
115 		return retval;
116 	if (le_key_k_offset(le_key_version(le_key), le_key) <
117 	    cpu_key_k_offset(cpu_key))
118 		return -1;
119 	if (le_key_k_offset(le_key_version(le_key), le_key) >
120 	    cpu_key_k_offset(cpu_key))
121 		return 1;
122 
123 	if (cpu_key->key_length == 3)
124 		return 0;
125 
126 	/* this part is needed only when tail conversion is in progress */
127 	if (le_key_k_type(le_key_version(le_key), le_key) <
128 	    cpu_key_k_type(cpu_key))
129 		return -1;
130 
131 	if (le_key_k_type(le_key_version(le_key), le_key) >
132 	    cpu_key_k_type(cpu_key))
133 		return 1;
134 
135 	return 0;
136 }
137 
138 inline int comp_short_le_keys(const struct reiserfs_key *key1,
139 			      const struct reiserfs_key *key2)
140 {
141 	__u32 *p_s_1_u32, *p_s_2_u32;
142 	int n_key_length = REISERFS_SHORT_KEY_LEN;
143 
144 	p_s_1_u32 = (__u32 *) key1;
145 	p_s_2_u32 = (__u32 *) key2;
146 	for (; n_key_length--; ++p_s_1_u32, ++p_s_2_u32) {
147 		if (le32_to_cpu(*p_s_1_u32) < le32_to_cpu(*p_s_2_u32))
148 			return -1;
149 		if (le32_to_cpu(*p_s_1_u32) > le32_to_cpu(*p_s_2_u32))
150 			return 1;
151 	}
152 	return 0;
153 }
154 
155 inline void le_key2cpu_key(struct cpu_key *to, const struct reiserfs_key *from)
156 {
157 	int version;
158 	to->on_disk_key.k_dir_id = le32_to_cpu(from->k_dir_id);
159 	to->on_disk_key.k_objectid = le32_to_cpu(from->k_objectid);
160 
161 	// find out version of the key
162 	version = le_key_version(from);
163 	to->version = version;
164 	to->on_disk_key.k_offset = le_key_k_offset(version, from);
165 	to->on_disk_key.k_type = le_key_k_type(version, from);
166 }
167 
168 // this does not say which one is bigger, it only returns 1 if keys
169 // are not equal, 0 otherwise
170 inline int comp_le_keys(const struct reiserfs_key *k1,
171 			const struct reiserfs_key *k2)
172 {
173 	return memcmp(k1, k2, sizeof(struct reiserfs_key));
174 }
175 
176 /**************************************************************************
177  *  Binary search toolkit function                                        *
178  *  Search for an item in the array by the item key                       *
179  *  Returns:    1 if found,  0 if not found;                              *
180  *        *p_n_pos = number of the searched element if found, else the    *
181  *        number of the first element that is larger than p_v_key.        *
182  **************************************************************************/
183 /* For those not familiar with binary search: n_lbound is the leftmost item that it
184  could be, n_rbound the rightmost item that it could be.  We examine the item
185  halfway between n_lbound and n_rbound, and that tells us either that we can increase
186  n_lbound, or decrease n_rbound, or that we have found it, or if n_lbound <= n_rbound that
187  there are no possible items, and we have not found it. With each examination we
188  cut the number of possible items it could be by one more than half rounded down,
189  or we find it. */
190 static inline int bin_search(const void *p_v_key,	/* Key to search for.                   */
191 			     const void *p_v_base,	/* First item in the array.             */
192 			     int p_n_num,	/* Number of items in the array.        */
193 			     int p_n_width,	/* Item size in the array.
194 						   searched. Lest the reader be
195 						   confused, note that this is crafted
196 						   as a general function, and when it
197 						   is applied specifically to the array
198 						   of item headers in a node, p_n_width
199 						   is actually the item header size not
200 						   the item size.                      */
201 			     int *p_n_pos	/* Number of the searched for element. */
202     )
203 {
204 	int n_rbound, n_lbound, n_j;
205 
206 	for (n_j = ((n_rbound = p_n_num - 1) + (n_lbound = 0)) / 2;
207 	     n_lbound <= n_rbound; n_j = (n_rbound + n_lbound) / 2)
208 		switch (comp_keys
209 			((struct reiserfs_key *)((char *)p_v_base +
210 						 n_j * p_n_width),
211 			 (struct cpu_key *)p_v_key)) {
212 		case -1:
213 			n_lbound = n_j + 1;
214 			continue;
215 		case 1:
216 			n_rbound = n_j - 1;
217 			continue;
218 		case 0:
219 			*p_n_pos = n_j;
220 			return ITEM_FOUND;	/* Key found in the array.  */
221 		}
222 
223 	/* bin_search did not find given key, it returns position of key,
224 	   that is minimal and greater than the given one. */
225 	*p_n_pos = n_lbound;
226 	return ITEM_NOT_FOUND;
227 }
228 
229 #ifdef CONFIG_REISERFS_CHECK
230 extern struct tree_balance *cur_tb;
231 #endif
232 
233 /* Minimal possible key. It is never in the tree. */
234 const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} };
235 
236 /* Maximal possible key. It is never in the tree. */
237 static const struct reiserfs_key MAX_KEY = {
238 	__constant_cpu_to_le32(0xffffffff),
239 	__constant_cpu_to_le32(0xffffffff),
240 	{{__constant_cpu_to_le32(0xffffffff),
241 	  __constant_cpu_to_le32(0xffffffff)},}
242 };
243 
244 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
245    of the path, and going upwards.  We must check the path's validity at each step.  If the key is not in
246    the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
247    case we return a special key, either MIN_KEY or MAX_KEY. */
248 static inline const struct reiserfs_key *get_lkey(const struct path
249 						  *p_s_chk_path,
250 						  const struct super_block
251 						  *p_s_sb)
252 {
253 	int n_position, n_path_offset = p_s_chk_path->path_length;
254 	struct buffer_head *p_s_parent;
255 
256 	RFALSE(n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
257 	       "PAP-5010: invalid offset in the path");
258 
259 	/* While not higher in path than first element. */
260 	while (n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
261 
262 		RFALSE(!buffer_uptodate
263 		       (PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
264 		       "PAP-5020: parent is not uptodate");
265 
266 		/* Parent at the path is not in the tree now. */
267 		if (!B_IS_IN_TREE
268 		    (p_s_parent =
269 		     PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)))
270 			return &MAX_KEY;
271 		/* Check whether position in the parent is correct. */
272 		if ((n_position =
273 		     PATH_OFFSET_POSITION(p_s_chk_path,
274 					  n_path_offset)) >
275 		    B_NR_ITEMS(p_s_parent))
276 			return &MAX_KEY;
277 		/* Check whether parent at the path really points to the child. */
278 		if (B_N_CHILD_NUM(p_s_parent, n_position) !=
279 		    PATH_OFFSET_PBUFFER(p_s_chk_path,
280 					n_path_offset + 1)->b_blocknr)
281 			return &MAX_KEY;
282 		/* Return delimiting key if position in the parent is not equal to zero. */
283 		if (n_position)
284 			return B_N_PDELIM_KEY(p_s_parent, n_position - 1);
285 	}
286 	/* Return MIN_KEY if we are in the root of the buffer tree. */
287 	if (PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->
288 	    b_blocknr == SB_ROOT_BLOCK(p_s_sb))
289 		return &MIN_KEY;
290 	return &MAX_KEY;
291 }
292 
293 /* Get delimiting key of the buffer at the path and its right neighbor. */
294 inline const struct reiserfs_key *get_rkey(const struct path *p_s_chk_path,
295 					   const struct super_block *p_s_sb)
296 {
297 	int n_position, n_path_offset = p_s_chk_path->path_length;
298 	struct buffer_head *p_s_parent;
299 
300 	RFALSE(n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
301 	       "PAP-5030: invalid offset in the path");
302 
303 	while (n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
304 
305 		RFALSE(!buffer_uptodate
306 		       (PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
307 		       "PAP-5040: parent is not uptodate");
308 
309 		/* Parent at the path is not in the tree now. */
310 		if (!B_IS_IN_TREE
311 		    (p_s_parent =
312 		     PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)))
313 			return &MIN_KEY;
314 		/* Check whether position in the parent is correct. */
315 		if ((n_position =
316 		     PATH_OFFSET_POSITION(p_s_chk_path,
317 					  n_path_offset)) >
318 		    B_NR_ITEMS(p_s_parent))
319 			return &MIN_KEY;
320 		/* Check whether parent at the path really points to the child. */
321 		if (B_N_CHILD_NUM(p_s_parent, n_position) !=
322 		    PATH_OFFSET_PBUFFER(p_s_chk_path,
323 					n_path_offset + 1)->b_blocknr)
324 			return &MIN_KEY;
325 		/* Return delimiting key if position in the parent is not the last one. */
326 		if (n_position != B_NR_ITEMS(p_s_parent))
327 			return B_N_PDELIM_KEY(p_s_parent, n_position);
328 	}
329 	/* Return MAX_KEY if we are in the root of the buffer tree. */
330 	if (PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->
331 	    b_blocknr == SB_ROOT_BLOCK(p_s_sb))
332 		return &MAX_KEY;
333 	return &MIN_KEY;
334 }
335 
336 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
337 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
338    the path.  These delimiting keys are stored at least one level above that buffer in the tree. If the
339    buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
340    this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
341 static inline int key_in_buffer(struct path *p_s_chk_path,	/* Path which should be checked.  */
342 				const struct cpu_key *p_s_key,	/* Key which should be checked.   */
343 				struct super_block *p_s_sb	/* Super block pointer.           */
344     )
345 {
346 
347 	RFALSE(!p_s_key || p_s_chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET
348 	       || p_s_chk_path->path_length > MAX_HEIGHT,
349 	       "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
350 	       p_s_key, p_s_chk_path->path_length);
351 	RFALSE(!PATH_PLAST_BUFFER(p_s_chk_path)->b_bdev,
352 	       "PAP-5060: device must not be NODEV");
353 
354 	if (comp_keys(get_lkey(p_s_chk_path, p_s_sb), p_s_key) == 1)
355 		/* left delimiting key is bigger, that the key we look for */
356 		return 0;
357 	//  if ( comp_keys(p_s_key, get_rkey(p_s_chk_path, p_s_sb)) != -1 )
358 	if (comp_keys(get_rkey(p_s_chk_path, p_s_sb), p_s_key) != 1)
359 		/* p_s_key must be less than right delimitiing key */
360 		return 0;
361 	return 1;
362 }
363 
364 inline void decrement_bcount(struct buffer_head *p_s_bh)
365 {
366 	if (p_s_bh) {
367 		if (atomic_read(&(p_s_bh->b_count))) {
368 			put_bh(p_s_bh);
369 			return;
370 		}
371 		reiserfs_panic(NULL,
372 			       "PAP-5070: decrement_bcount: trying to free free buffer %b",
373 			       p_s_bh);
374 	}
375 }
376 
377 /* Decrement b_count field of the all buffers in the path. */
378 void decrement_counters_in_path(struct path *p_s_search_path)
379 {
380 	int n_path_offset = p_s_search_path->path_length;
381 
382 	RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET ||
383 	       n_path_offset > EXTENDED_MAX_HEIGHT - 1,
384 	       "PAP-5080: invalid path offset of %d", n_path_offset);
385 
386 	while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
387 		struct buffer_head *bh;
388 
389 		bh = PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--);
390 		decrement_bcount(bh);
391 	}
392 	p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
393 }
394 
395 int reiserfs_check_path(struct path *p)
396 {
397 	RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
398 	       "path not properly relsed");
399 	return 0;
400 }
401 
402 /* Release all buffers in the path. Restore dirty bits clean
403 ** when preparing the buffer for the log
404 **
405 ** only called from fix_nodes()
406 */
407 void pathrelse_and_restore(struct super_block *s, struct path *p_s_search_path)
408 {
409 	int n_path_offset = p_s_search_path->path_length;
410 
411 	RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
412 	       "clm-4000: invalid path offset");
413 
414 	while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
415 		reiserfs_restore_prepared_buffer(s,
416 						 PATH_OFFSET_PBUFFER
417 						 (p_s_search_path,
418 						  n_path_offset));
419 		brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
420 	}
421 	p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
422 }
423 
424 /* Release all buffers in the path. */
425 void pathrelse(struct path *p_s_search_path)
426 {
427 	int n_path_offset = p_s_search_path->path_length;
428 
429 	RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
430 	       "PAP-5090: invalid path offset");
431 
432 	while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
433 		brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
434 
435 	p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
436 }
437 
438 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
439 {
440 	struct block_head *blkh;
441 	struct item_head *ih;
442 	int used_space;
443 	int prev_location;
444 	int i;
445 	int nr;
446 
447 	blkh = (struct block_head *)buf;
448 	if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
449 		reiserfs_warning(NULL,
450 				 "is_leaf: this should be caught earlier");
451 		return 0;
452 	}
453 
454 	nr = blkh_nr_item(blkh);
455 	if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
456 		/* item number is too big or too small */
457 		reiserfs_warning(NULL, "is_leaf: nr_item seems wrong: %z", bh);
458 		return 0;
459 	}
460 	ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
461 	used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih));
462 	if (used_space != blocksize - blkh_free_space(blkh)) {
463 		/* free space does not match to calculated amount of use space */
464 		reiserfs_warning(NULL, "is_leaf: free space seems wrong: %z",
465 				 bh);
466 		return 0;
467 	}
468 	// FIXME: it is_leaf will hit performance too much - we may have
469 	// return 1 here
470 
471 	/* check tables of item heads */
472 	ih = (struct item_head *)(buf + BLKH_SIZE);
473 	prev_location = blocksize;
474 	for (i = 0; i < nr; i++, ih++) {
475 		if (le_ih_k_type(ih) == TYPE_ANY) {
476 			reiserfs_warning(NULL,
477 					 "is_leaf: wrong item type for item %h",
478 					 ih);
479 			return 0;
480 		}
481 		if (ih_location(ih) >= blocksize
482 		    || ih_location(ih) < IH_SIZE * nr) {
483 			reiserfs_warning(NULL,
484 					 "is_leaf: item location seems wrong: %h",
485 					 ih);
486 			return 0;
487 		}
488 		if (ih_item_len(ih) < 1
489 		    || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
490 			reiserfs_warning(NULL,
491 					 "is_leaf: item length seems wrong: %h",
492 					 ih);
493 			return 0;
494 		}
495 		if (prev_location - ih_location(ih) != ih_item_len(ih)) {
496 			reiserfs_warning(NULL,
497 					 "is_leaf: item location seems wrong (second one): %h",
498 					 ih);
499 			return 0;
500 		}
501 		prev_location = ih_location(ih);
502 	}
503 
504 	// one may imagine much more checks
505 	return 1;
506 }
507 
508 /* returns 1 if buf looks like an internal node, 0 otherwise */
509 static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
510 {
511 	struct block_head *blkh;
512 	int nr;
513 	int used_space;
514 
515 	blkh = (struct block_head *)buf;
516 	nr = blkh_level(blkh);
517 	if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
518 		/* this level is not possible for internal nodes */
519 		reiserfs_warning(NULL,
520 				 "is_internal: this should be caught earlier");
521 		return 0;
522 	}
523 
524 	nr = blkh_nr_item(blkh);
525 	if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
526 		/* for internal which is not root we might check min number of keys */
527 		reiserfs_warning(NULL,
528 				 "is_internal: number of key seems wrong: %z",
529 				 bh);
530 		return 0;
531 	}
532 
533 	used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
534 	if (used_space != blocksize - blkh_free_space(blkh)) {
535 		reiserfs_warning(NULL,
536 				 "is_internal: free space seems wrong: %z", bh);
537 		return 0;
538 	}
539 	// one may imagine much more checks
540 	return 1;
541 }
542 
543 // make sure that bh contains formatted node of reiserfs tree of
544 // 'level'-th level
545 static int is_tree_node(struct buffer_head *bh, int level)
546 {
547 	if (B_LEVEL(bh) != level) {
548 		reiserfs_warning(NULL,
549 				 "is_tree_node: node level %d does not match to the expected one %d",
550 				 B_LEVEL(bh), level);
551 		return 0;
552 	}
553 	if (level == DISK_LEAF_NODE_LEVEL)
554 		return is_leaf(bh->b_data, bh->b_size, bh);
555 
556 	return is_internal(bh->b_data, bh->b_size, bh);
557 }
558 
559 #define SEARCH_BY_KEY_READA 16
560 
561 /* The function is NOT SCHEDULE-SAFE! */
562 static void search_by_key_reada(struct super_block *s,
563 				struct buffer_head **bh,
564 				unsigned long *b, int num)
565 {
566 	int i, j;
567 
568 	for (i = 0; i < num; i++) {
569 		bh[i] = sb_getblk(s, b[i]);
570 	}
571 	for (j = 0; j < i; j++) {
572 		/*
573 		 * note, this needs attention if we are getting rid of the BKL
574 		 * you have to make sure the prepared bit isn't set on this buffer
575 		 */
576 		if (!buffer_uptodate(bh[j]))
577 			ll_rw_block(READA, 1, bh + j);
578 		brelse(bh[j]);
579 	}
580 }
581 
582 /**************************************************************************
583  * Algorithm   SearchByKey                                                *
584  *             look for item in the Disk S+Tree by its key                *
585  * Input:  p_s_sb   -  super block                                        *
586  *         p_s_key  - pointer to the key to search                        *
587  * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR                         *
588  *         p_s_search_path - path from the root to the needed leaf        *
589  **************************************************************************/
590 
591 /* This function fills up the path from the root to the leaf as it
592    descends the tree looking for the key.  It uses reiserfs_bread to
593    try to find buffers in the cache given their block number.  If it
594    does not find them in the cache it reads them from disk.  For each
595    node search_by_key finds using reiserfs_bread it then uses
596    bin_search to look through that node.  bin_search will find the
597    position of the block_number of the next node if it is looking
598    through an internal node.  If it is looking through a leaf node
599    bin_search will find the position of the item which has key either
600    equal to given key, or which is the maximal key less than the given
601    key.  search_by_key returns a path that must be checked for the
602    correctness of the top of the path but need not be checked for the
603    correctness of the bottom of the path */
604 /* The function is NOT SCHEDULE-SAFE! */
605 int search_by_key(struct super_block *p_s_sb, const struct cpu_key *p_s_key,	/* Key to search. */
606 		  struct path *p_s_search_path,	/* This structure was
607 						   allocated and initialized
608 						   by the calling
609 						   function. It is filled up
610 						   by this function.  */
611 		  int n_stop_level	/* How far down the tree to search. To
612 					   stop at leaf level - set to
613 					   DISK_LEAF_NODE_LEVEL */
614     )
615 {
616 	int n_block_number;
617 	int expected_level;
618 	struct buffer_head *p_s_bh;
619 	struct path_element *p_s_last_element;
620 	int n_node_level, n_retval;
621 	int right_neighbor_of_leaf_node;
622 	int fs_gen;
623 	struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
624 	unsigned long reada_blocks[SEARCH_BY_KEY_READA];
625 	int reada_count = 0;
626 
627 #ifdef CONFIG_REISERFS_CHECK
628 	int n_repeat_counter = 0;
629 #endif
630 
631 	PROC_INFO_INC(p_s_sb, search_by_key);
632 
633 	/* As we add each node to a path we increase its count.  This means that
634 	   we must be careful to release all nodes in a path before we either
635 	   discard the path struct or re-use the path struct, as we do here. */
636 
637 	decrement_counters_in_path(p_s_search_path);
638 
639 	right_neighbor_of_leaf_node = 0;
640 
641 	/* With each iteration of this loop we search through the items in the
642 	   current node, and calculate the next current node(next path element)
643 	   for the next iteration of this loop.. */
644 	n_block_number = SB_ROOT_BLOCK(p_s_sb);
645 	expected_level = -1;
646 	while (1) {
647 
648 #ifdef CONFIG_REISERFS_CHECK
649 		if (!(++n_repeat_counter % 50000))
650 			reiserfs_warning(p_s_sb, "PAP-5100: search_by_key: %s:"
651 					 "there were %d iterations of while loop "
652 					 "looking for key %K",
653 					 current->comm, n_repeat_counter,
654 					 p_s_key);
655 #endif
656 
657 		/* prep path to have another element added to it. */
658 		p_s_last_element =
659 		    PATH_OFFSET_PELEMENT(p_s_search_path,
660 					 ++p_s_search_path->path_length);
661 		fs_gen = get_generation(p_s_sb);
662 
663 		/* Read the next tree node, and set the last element in the path to
664 		   have a pointer to it. */
665 		if ((p_s_bh = p_s_last_element->pe_buffer =
666 		     sb_getblk(p_s_sb, n_block_number))) {
667 			if (!buffer_uptodate(p_s_bh) && reada_count > 1) {
668 				search_by_key_reada(p_s_sb, reada_bh,
669 						    reada_blocks, reada_count);
670 			}
671 			ll_rw_block(READ, 1, &p_s_bh);
672 			wait_on_buffer(p_s_bh);
673 			if (!buffer_uptodate(p_s_bh))
674 				goto io_error;
675 		} else {
676 		      io_error:
677 			p_s_search_path->path_length--;
678 			pathrelse(p_s_search_path);
679 			return IO_ERROR;
680 		}
681 		reada_count = 0;
682 		if (expected_level == -1)
683 			expected_level = SB_TREE_HEIGHT(p_s_sb);
684 		expected_level--;
685 
686 		/* It is possible that schedule occurred. We must check whether the key
687 		   to search is still in the tree rooted from the current buffer. If
688 		   not then repeat search from the root. */
689 		if (fs_changed(fs_gen, p_s_sb) &&
690 		    (!B_IS_IN_TREE(p_s_bh) ||
691 		     B_LEVEL(p_s_bh) != expected_level ||
692 		     !key_in_buffer(p_s_search_path, p_s_key, p_s_sb))) {
693 			PROC_INFO_INC(p_s_sb, search_by_key_fs_changed);
694 			PROC_INFO_INC(p_s_sb, search_by_key_restarted);
695 			PROC_INFO_INC(p_s_sb,
696 				      sbk_restarted[expected_level - 1]);
697 			decrement_counters_in_path(p_s_search_path);
698 
699 			/* Get the root block number so that we can repeat the search
700 			   starting from the root. */
701 			n_block_number = SB_ROOT_BLOCK(p_s_sb);
702 			expected_level = -1;
703 			right_neighbor_of_leaf_node = 0;
704 
705 			/* repeat search from the root */
706 			continue;
707 		}
708 
709 		/* only check that the key is in the buffer if p_s_key is not
710 		   equal to the MAX_KEY. Latter case is only possible in
711 		   "finish_unfinished()" processing during mount. */
712 		RFALSE(comp_keys(&MAX_KEY, p_s_key) &&
713 		       !key_in_buffer(p_s_search_path, p_s_key, p_s_sb),
714 		       "PAP-5130: key is not in the buffer");
715 #ifdef CONFIG_REISERFS_CHECK
716 		if (cur_tb) {
717 			print_cur_tb("5140");
718 			reiserfs_panic(p_s_sb,
719 				       "PAP-5140: search_by_key: schedule occurred in do_balance!");
720 		}
721 #endif
722 
723 		// make sure, that the node contents look like a node of
724 		// certain level
725 		if (!is_tree_node(p_s_bh, expected_level)) {
726 			reiserfs_warning(p_s_sb, "vs-5150: search_by_key: "
727 					 "invalid format found in block %ld. Fsck?",
728 					 p_s_bh->b_blocknr);
729 			pathrelse(p_s_search_path);
730 			return IO_ERROR;
731 		}
732 
733 		/* ok, we have acquired next formatted node in the tree */
734 		n_node_level = B_LEVEL(p_s_bh);
735 
736 		PROC_INFO_BH_STAT(p_s_sb, p_s_bh, n_node_level - 1);
737 
738 		RFALSE(n_node_level < n_stop_level,
739 		       "vs-5152: tree level (%d) is less than stop level (%d)",
740 		       n_node_level, n_stop_level);
741 
742 		n_retval = bin_search(p_s_key, B_N_PITEM_HEAD(p_s_bh, 0),
743 				      B_NR_ITEMS(p_s_bh),
744 				      (n_node_level ==
745 				       DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
746 				      KEY_SIZE,
747 				      &(p_s_last_element->pe_position));
748 		if (n_node_level == n_stop_level) {
749 			return n_retval;
750 		}
751 
752 		/* we are not in the stop level */
753 		if (n_retval == ITEM_FOUND)
754 			/* item has been found, so we choose the pointer which is to the right of the found one */
755 			p_s_last_element->pe_position++;
756 
757 		/* if item was not found we choose the position which is to
758 		   the left of the found item. This requires no code,
759 		   bin_search did it already. */
760 
761 		/* So we have chosen a position in the current node which is
762 		   an internal node.  Now we calculate child block number by
763 		   position in the node. */
764 		n_block_number =
765 		    B_N_CHILD_NUM(p_s_bh, p_s_last_element->pe_position);
766 
767 		/* if we are going to read leaf nodes, try for read ahead as well */
768 		if ((p_s_search_path->reada & PATH_READA) &&
769 		    n_node_level == DISK_LEAF_NODE_LEVEL + 1) {
770 			int pos = p_s_last_element->pe_position;
771 			int limit = B_NR_ITEMS(p_s_bh);
772 			struct reiserfs_key *le_key;
773 
774 			if (p_s_search_path->reada & PATH_READA_BACK)
775 				limit = 0;
776 			while (reada_count < SEARCH_BY_KEY_READA) {
777 				if (pos == limit)
778 					break;
779 				reada_blocks[reada_count++] =
780 				    B_N_CHILD_NUM(p_s_bh, pos);
781 				if (p_s_search_path->reada & PATH_READA_BACK)
782 					pos--;
783 				else
784 					pos++;
785 
786 				/*
787 				 * check to make sure we're in the same object
788 				 */
789 				le_key = B_N_PDELIM_KEY(p_s_bh, pos);
790 				if (le32_to_cpu(le_key->k_objectid) !=
791 				    p_s_key->on_disk_key.k_objectid) {
792 					break;
793 				}
794 			}
795 		}
796 	}
797 }
798 
799 /* Form the path to an item and position in this item which contains
800    file byte defined by p_s_key. If there is no such item
801    corresponding to the key, we point the path to the item with
802    maximal key less than p_s_key, and *p_n_pos_in_item is set to one
803    past the last entry/byte in the item.  If searching for entry in a
804    directory item, and it is not found, *p_n_pos_in_item is set to one
805    entry more than the entry with maximal key which is less than the
806    sought key.
807 
808    Note that if there is no entry in this same node which is one more,
809    then we point to an imaginary entry.  for direct items, the
810    position is in units of bytes, for indirect items the position is
811    in units of blocknr entries, for directory items the position is in
812    units of directory entries.  */
813 
814 /* The function is NOT SCHEDULE-SAFE! */
815 int search_for_position_by_key(struct super_block *p_s_sb,	/* Pointer to the super block.          */
816 			       const struct cpu_key *p_cpu_key,	/* Key to search (cpu variable)         */
817 			       struct path *p_s_search_path	/* Filled up by this function.          */
818     )
819 {
820 	struct item_head *p_le_ih;	/* pointer to on-disk structure */
821 	int n_blk_size;
822 	loff_t item_offset, offset;
823 	struct reiserfs_dir_entry de;
824 	int retval;
825 
826 	/* If searching for directory entry. */
827 	if (is_direntry_cpu_key(p_cpu_key))
828 		return search_by_entry_key(p_s_sb, p_cpu_key, p_s_search_path,
829 					   &de);
830 
831 	/* If not searching for directory entry. */
832 
833 	/* If item is found. */
834 	retval = search_item(p_s_sb, p_cpu_key, p_s_search_path);
835 	if (retval == IO_ERROR)
836 		return retval;
837 	if (retval == ITEM_FOUND) {
838 
839 		RFALSE(!ih_item_len
840 		       (B_N_PITEM_HEAD
841 			(PATH_PLAST_BUFFER(p_s_search_path),
842 			 PATH_LAST_POSITION(p_s_search_path))),
843 		       "PAP-5165: item length equals zero");
844 
845 		pos_in_item(p_s_search_path) = 0;
846 		return POSITION_FOUND;
847 	}
848 
849 	RFALSE(!PATH_LAST_POSITION(p_s_search_path),
850 	       "PAP-5170: position equals zero");
851 
852 	/* Item is not found. Set path to the previous item. */
853 	p_le_ih =
854 	    B_N_PITEM_HEAD(PATH_PLAST_BUFFER(p_s_search_path),
855 			   --PATH_LAST_POSITION(p_s_search_path));
856 	n_blk_size = p_s_sb->s_blocksize;
857 
858 	if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) {
859 		return FILE_NOT_FOUND;
860 	}
861 	// FIXME: quite ugly this far
862 
863 	item_offset = le_ih_k_offset(p_le_ih);
864 	offset = cpu_key_k_offset(p_cpu_key);
865 
866 	/* Needed byte is contained in the item pointed to by the path. */
867 	if (item_offset <= offset &&
868 	    item_offset + op_bytes_number(p_le_ih, n_blk_size) > offset) {
869 		pos_in_item(p_s_search_path) = offset - item_offset;
870 		if (is_indirect_le_ih(p_le_ih)) {
871 			pos_in_item(p_s_search_path) /= n_blk_size;
872 		}
873 		return POSITION_FOUND;
874 	}
875 
876 	/* Needed byte is not contained in the item pointed to by the
877 	   path. Set pos_in_item out of the item. */
878 	if (is_indirect_le_ih(p_le_ih))
879 		pos_in_item(p_s_search_path) =
880 		    ih_item_len(p_le_ih) / UNFM_P_SIZE;
881 	else
882 		pos_in_item(p_s_search_path) = ih_item_len(p_le_ih);
883 
884 	return POSITION_NOT_FOUND;
885 }
886 
887 /* Compare given item and item pointed to by the path. */
888 int comp_items(const struct item_head *stored_ih, const struct path *p_s_path)
889 {
890 	struct buffer_head *p_s_bh;
891 	struct item_head *ih;
892 
893 	/* Last buffer at the path is not in the tree. */
894 	if (!B_IS_IN_TREE(p_s_bh = PATH_PLAST_BUFFER(p_s_path)))
895 		return 1;
896 
897 	/* Last path position is invalid. */
898 	if (PATH_LAST_POSITION(p_s_path) >= B_NR_ITEMS(p_s_bh))
899 		return 1;
900 
901 	/* we need only to know, whether it is the same item */
902 	ih = get_ih(p_s_path);
903 	return memcmp(stored_ih, ih, IH_SIZE);
904 }
905 
906 /* unformatted nodes are not logged anymore, ever.  This is safe
907 ** now
908 */
909 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
910 
911 // block can not be forgotten as it is in I/O or held by someone
912 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
913 
914 // prepare for delete or cut of direct item
915 static inline int prepare_for_direct_item(struct path *path,
916 					  struct item_head *le_ih,
917 					  struct inode *inode,
918 					  loff_t new_file_length, int *cut_size)
919 {
920 	loff_t round_len;
921 
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));
925 		return M_DELETE;
926 	}
927 	// new file gets truncated
928 	if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
929 		//
930 		round_len = ROUND_UP(new_file_length);
931 		/* this was n_new_file_length < le_ih ... */
932 		if (round_len < le_ih_k_offset(le_ih)) {
933 			*cut_size = -(IH_SIZE + ih_item_len(le_ih));
934 			return M_DELETE;	/* Delete this item. */
935 		}
936 		/* Calculate first position and size for cutting from item. */
937 		pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
938 		*cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
939 
940 		return M_CUT;	/* Cut from this item. */
941 	}
942 
943 	// old file: items may have any length
944 
945 	if (new_file_length < le_ih_k_offset(le_ih)) {
946 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
947 		return M_DELETE;	/* Delete this item. */
948 	}
949 	/* Calculate first position and size for cutting from item. */
950 	*cut_size = -(ih_item_len(le_ih) -
951 		      (pos_in_item(path) =
952 		       new_file_length + 1 - le_ih_k_offset(le_ih)));
953 	return M_CUT;		/* Cut from this item. */
954 }
955 
956 static inline int prepare_for_direntry_item(struct path *path,
957 					    struct item_head *le_ih,
958 					    struct inode *inode,
959 					    loff_t new_file_length,
960 					    int *cut_size)
961 {
962 	if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
963 	    new_file_length == max_reiserfs_offset(inode)) {
964 		RFALSE(ih_entry_count(le_ih) != 2,
965 		       "PAP-5220: incorrect empty directory item (%h)", le_ih);
966 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
967 		return M_DELETE;	/* Delete the directory item containing "." and ".." entry. */
968 	}
969 
970 	if (ih_entry_count(le_ih) == 1) {
971 		/* Delete the directory item such as there is one record only
972 		   in this item */
973 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
974 		return M_DELETE;
975 	}
976 
977 	/* Cut one record from the directory item. */
978 	*cut_size =
979 	    -(DEH_SIZE +
980 	      entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
981 	return M_CUT;
982 }
983 
984 /*  If the path points to a directory or direct item, calculate mode and the size cut, for balance.
985     If the path points to an indirect item, remove some number of its unformatted nodes.
986     In case of file truncate calculate whether this item must be deleted/truncated or last
987     unformatted node of this item will be converted to a direct item.
988     This function returns a determination of what balance mode the calling function should employ. */
989 static char prepare_for_delete_or_cut(struct reiserfs_transaction_handle *th, struct inode *inode, struct path *p_s_path, const struct cpu_key *p_s_item_key, int *p_n_removed,	/* Number of unformatted nodes which were removed
990 																						   from end of the file. */
991 				      int *p_n_cut_size, unsigned long long n_new_file_length	/* MAX_KEY_OFFSET in case of delete. */
992     )
993 {
994 	struct super_block *p_s_sb = inode->i_sb;
995 	struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_path);
996 	struct buffer_head *p_s_bh = PATH_PLAST_BUFFER(p_s_path);
997 
998 	BUG_ON(!th->t_trans_id);
999 
1000 	/* Stat_data item. */
1001 	if (is_statdata_le_ih(p_le_ih)) {
1002 
1003 		RFALSE(n_new_file_length != max_reiserfs_offset(inode),
1004 		       "PAP-5210: mode must be M_DELETE");
1005 
1006 		*p_n_cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
1007 		return M_DELETE;
1008 	}
1009 
1010 	/* Directory item. */
1011 	if (is_direntry_le_ih(p_le_ih))
1012 		return prepare_for_direntry_item(p_s_path, p_le_ih, inode,
1013 						 n_new_file_length,
1014 						 p_n_cut_size);
1015 
1016 	/* Direct item. */
1017 	if (is_direct_le_ih(p_le_ih))
1018 		return prepare_for_direct_item(p_s_path, p_le_ih, inode,
1019 					       n_new_file_length, p_n_cut_size);
1020 
1021 	/* Case of an indirect item. */
1022 	{
1023 		int n_unfm_number,	/* Number of the item unformatted nodes. */
1024 		 n_counter, n_blk_size;
1025 		__le32 *p_n_unfm_pointer;	/* Pointer to the unformatted node number. */
1026 		__u32 tmp;
1027 		struct item_head s_ih;	/* Item header. */
1028 		char c_mode;	/* Returned mode of the balance. */
1029 		int need_research;
1030 
1031 		n_blk_size = p_s_sb->s_blocksize;
1032 
1033 		/* Search for the needed object indirect item until there are no unformatted nodes to be removed. */
1034 		do {
1035 			need_research = 0;
1036 			p_s_bh = PATH_PLAST_BUFFER(p_s_path);
1037 			/* Copy indirect item header to a temp variable. */
1038 			copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1039 			/* Calculate number of unformatted nodes in this item. */
1040 			n_unfm_number = I_UNFM_NUM(&s_ih);
1041 
1042 			RFALSE(!is_indirect_le_ih(&s_ih) || !n_unfm_number ||
1043 			       pos_in_item(p_s_path) + 1 != n_unfm_number,
1044 			       "PAP-5240: invalid item %h "
1045 			       "n_unfm_number = %d *p_n_pos_in_item = %d",
1046 			       &s_ih, n_unfm_number, pos_in_item(p_s_path));
1047 
1048 			/* Calculate balance mode and position in the item to remove unformatted nodes. */
1049 			if (n_new_file_length == max_reiserfs_offset(inode)) {	/* Case of delete. */
1050 				pos_in_item(p_s_path) = 0;
1051 				*p_n_cut_size = -(IH_SIZE + ih_item_len(&s_ih));
1052 				c_mode = M_DELETE;
1053 			} else {	/* Case of truncate. */
1054 				if (n_new_file_length < le_ih_k_offset(&s_ih)) {
1055 					pos_in_item(p_s_path) = 0;
1056 					*p_n_cut_size =
1057 					    -(IH_SIZE + ih_item_len(&s_ih));
1058 					c_mode = M_DELETE;	/* Delete this item. */
1059 				} else {
1060 					/* indirect item must be truncated starting from *p_n_pos_in_item-th position */
1061 					pos_in_item(p_s_path) =
1062 					    (n_new_file_length + n_blk_size -
1063 					     le_ih_k_offset(&s_ih)) >> p_s_sb->
1064 					    s_blocksize_bits;
1065 
1066 					RFALSE(pos_in_item(p_s_path) >
1067 					       n_unfm_number,
1068 					       "PAP-5250: invalid position in the item");
1069 
1070 					/* Either convert last unformatted node of indirect item to direct item or increase
1071 					   its free space.  */
1072 					if (pos_in_item(p_s_path) ==
1073 					    n_unfm_number) {
1074 						*p_n_cut_size = 0;	/* Nothing to cut. */
1075 						return M_CONVERT;	/* Maybe convert last unformatted node to the direct item. */
1076 					}
1077 					/* Calculate size to cut. */
1078 					*p_n_cut_size =
1079 					    -(ih_item_len(&s_ih) -
1080 					      pos_in_item(p_s_path) *
1081 					      UNFM_P_SIZE);
1082 
1083 					c_mode = M_CUT;	/* Cut from this indirect item. */
1084 				}
1085 			}
1086 
1087 			RFALSE(n_unfm_number <= pos_in_item(p_s_path),
1088 			       "PAP-5260: invalid position in the indirect item");
1089 
1090 			/* pointers to be cut */
1091 			n_unfm_number -= pos_in_item(p_s_path);
1092 			/* Set pointer to the last unformatted node pointer that is to be cut. */
1093 			p_n_unfm_pointer =
1094 			    (__le32 *) B_I_PITEM(p_s_bh,
1095 						 &s_ih) + I_UNFM_NUM(&s_ih) -
1096 			    1 - *p_n_removed;
1097 
1098 			/* We go through the unformatted nodes pointers of the indirect
1099 			   item and look for the unformatted nodes in the cache. If we
1100 			   found some of them we free it, zero corresponding indirect item
1101 			   entry and log buffer containing that indirect item. For this we
1102 			   need to prepare last path element for logging. If some
1103 			   unformatted node has b_count > 1 we must not free this
1104 			   unformatted node since it is in use. */
1105 			reiserfs_prepare_for_journal(p_s_sb, p_s_bh, 1);
1106 			// note: path could be changed, first line in for loop takes care
1107 			// of it
1108 
1109 			for (n_counter = *p_n_removed;
1110 			     n_counter < n_unfm_number;
1111 			     n_counter++, p_n_unfm_pointer--) {
1112 
1113 				cond_resched();
1114 				if (item_moved(&s_ih, p_s_path)) {
1115 					need_research = 1;
1116 					break;
1117 				}
1118 				RFALSE(p_n_unfm_pointer <
1119 				       (__le32 *) B_I_PITEM(p_s_bh, &s_ih)
1120 				       || p_n_unfm_pointer >
1121 				       (__le32 *) B_I_PITEM(p_s_bh,
1122 							    &s_ih) +
1123 				       I_UNFM_NUM(&s_ih) - 1,
1124 				       "vs-5265: pointer out of range");
1125 
1126 				/* Hole, nothing to remove. */
1127 				if (!get_block_num(p_n_unfm_pointer, 0)) {
1128 					(*p_n_removed)++;
1129 					continue;
1130 				}
1131 
1132 				(*p_n_removed)++;
1133 
1134 				tmp = get_block_num(p_n_unfm_pointer, 0);
1135 				put_block_num(p_n_unfm_pointer, 0, 0);
1136 				journal_mark_dirty(th, p_s_sb, p_s_bh);
1137 				reiserfs_free_block(th, inode, tmp, 1);
1138 				if (item_moved(&s_ih, p_s_path)) {
1139 					need_research = 1;
1140 					break;
1141 				}
1142 			}
1143 
1144 			/* a trick.  If the buffer has been logged, this
1145 			 ** will do nothing.  If we've broken the loop without
1146 			 ** logging it, it will restore the buffer
1147 			 **
1148 			 */
1149 			reiserfs_restore_prepared_buffer(p_s_sb, p_s_bh);
1150 
1151 			/* This loop can be optimized. */
1152 		} while ((*p_n_removed < n_unfm_number || need_research) &&
1153 			 search_for_position_by_key(p_s_sb, p_s_item_key,
1154 						    p_s_path) ==
1155 			 POSITION_FOUND);
1156 
1157 		RFALSE(*p_n_removed < n_unfm_number,
1158 		       "PAP-5310: indirect item is not found");
1159 		RFALSE(item_moved(&s_ih, p_s_path),
1160 		       "after while, comp failed, retry");
1161 
1162 		if (c_mode == M_CUT)
1163 			pos_in_item(p_s_path) *= UNFM_P_SIZE;
1164 		return c_mode;
1165 	}
1166 }
1167 
1168 /* Calculate number of bytes which will be deleted or cut during balance */
1169 static int calc_deleted_bytes_number(struct tree_balance *p_s_tb, char c_mode)
1170 {
1171 	int n_del_size;
1172 	struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_tb->tb_path);
1173 
1174 	if (is_statdata_le_ih(p_le_ih))
1175 		return 0;
1176 
1177 	n_del_size =
1178 	    (c_mode ==
1179 	     M_DELETE) ? ih_item_len(p_le_ih) : -p_s_tb->insert_size[0];
1180 	if (is_direntry_le_ih(p_le_ih)) {
1181 		// return EMPTY_DIR_SIZE; /* We delete emty directoris only. */
1182 		// we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1183 		// empty size.  ick. FIXME, is this right?
1184 		//
1185 		return n_del_size;
1186 	}
1187 
1188 	if (is_indirect_le_ih(p_le_ih))
1189 		n_del_size = (n_del_size / UNFM_P_SIZE) * (PATH_PLAST_BUFFER(p_s_tb->tb_path)->b_size);	// - get_ih_free_space (p_le_ih);
1190 	return n_del_size;
1191 }
1192 
1193 static void init_tb_struct(struct reiserfs_transaction_handle *th,
1194 			   struct tree_balance *p_s_tb,
1195 			   struct super_block *p_s_sb,
1196 			   struct path *p_s_path, int n_size)
1197 {
1198 
1199 	BUG_ON(!th->t_trans_id);
1200 
1201 	memset(p_s_tb, '\0', sizeof(struct tree_balance));
1202 	p_s_tb->transaction_handle = th;
1203 	p_s_tb->tb_sb = p_s_sb;
1204 	p_s_tb->tb_path = p_s_path;
1205 	PATH_OFFSET_PBUFFER(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1206 	PATH_OFFSET_POSITION(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1207 	p_s_tb->insert_size[0] = n_size;
1208 }
1209 
1210 void padd_item(char *item, int total_length, int length)
1211 {
1212 	int i;
1213 
1214 	for (i = total_length; i > length;)
1215 		item[--i] = 0;
1216 }
1217 
1218 #ifdef REISERQUOTA_DEBUG
1219 char key2type(struct reiserfs_key *ih)
1220 {
1221 	if (is_direntry_le_key(2, ih))
1222 		return 'd';
1223 	if (is_direct_le_key(2, ih))
1224 		return 'D';
1225 	if (is_indirect_le_key(2, ih))
1226 		return 'i';
1227 	if (is_statdata_le_key(2, ih))
1228 		return 's';
1229 	return 'u';
1230 }
1231 
1232 char head2type(struct item_head *ih)
1233 {
1234 	if (is_direntry_le_ih(ih))
1235 		return 'd';
1236 	if (is_direct_le_ih(ih))
1237 		return 'D';
1238 	if (is_indirect_le_ih(ih))
1239 		return 'i';
1240 	if (is_statdata_le_ih(ih))
1241 		return 's';
1242 	return 'u';
1243 }
1244 #endif
1245 
1246 /* Delete object item. */
1247 int reiserfs_delete_item(struct reiserfs_transaction_handle *th, struct path *p_s_path,	/* Path to the deleted item. */
1248 			 const struct cpu_key *p_s_item_key,	/* Key to search for the deleted item.  */
1249 			 struct inode *p_s_inode,	/* inode is here just to update i_blocks and quotas */
1250 			 struct buffer_head *p_s_un_bh)
1251 {				/* NULL or unformatted node pointer.    */
1252 	struct super_block *p_s_sb = p_s_inode->i_sb;
1253 	struct tree_balance s_del_balance;
1254 	struct item_head s_ih;
1255 	struct item_head *q_ih;
1256 	int quota_cut_bytes;
1257 	int n_ret_value, n_del_size, n_removed;
1258 
1259 #ifdef CONFIG_REISERFS_CHECK
1260 	char c_mode;
1261 	int n_iter = 0;
1262 #endif
1263 
1264 	BUG_ON(!th->t_trans_id);
1265 
1266 	init_tb_struct(th, &s_del_balance, p_s_sb, p_s_path,
1267 		       0 /*size is unknown */ );
1268 
1269 	while (1) {
1270 		n_removed = 0;
1271 
1272 #ifdef CONFIG_REISERFS_CHECK
1273 		n_iter++;
1274 		c_mode =
1275 #endif
1276 		    prepare_for_delete_or_cut(th, p_s_inode, p_s_path,
1277 					      p_s_item_key, &n_removed,
1278 					      &n_del_size,
1279 					      max_reiserfs_offset(p_s_inode));
1280 
1281 		RFALSE(c_mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1282 
1283 		copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1284 		s_del_balance.insert_size[0] = n_del_size;
1285 
1286 		n_ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1287 		if (n_ret_value != REPEAT_SEARCH)
1288 			break;
1289 
1290 		PROC_INFO_INC(p_s_sb, delete_item_restarted);
1291 
1292 		// file system changed, repeat search
1293 		n_ret_value =
1294 		    search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1295 		if (n_ret_value == IO_ERROR)
1296 			break;
1297 		if (n_ret_value == FILE_NOT_FOUND) {
1298 			reiserfs_warning(p_s_sb,
1299 					 "vs-5340: reiserfs_delete_item: "
1300 					 "no items of the file %K found",
1301 					 p_s_item_key);
1302 			break;
1303 		}
1304 	}			/* while (1) */
1305 
1306 	if (n_ret_value != CARRY_ON) {
1307 		unfix_nodes(&s_del_balance);
1308 		return 0;
1309 	}
1310 	// reiserfs_delete_item returns item length when success
1311 	n_ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1312 	q_ih = get_ih(p_s_path);
1313 	quota_cut_bytes = ih_item_len(q_ih);
1314 
1315 	/* hack so the quota code doesn't have to guess if the file
1316 	 ** has a tail.  On tail insert, we allocate quota for 1 unformatted node.
1317 	 ** We test the offset because the tail might have been
1318 	 ** split into multiple items, and we only want to decrement for
1319 	 ** the unfm node once
1320 	 */
1321 	if (!S_ISLNK(p_s_inode->i_mode) && is_direct_le_ih(q_ih)) {
1322 		if ((le_ih_k_offset(q_ih) & (p_s_sb->s_blocksize - 1)) == 1) {
1323 			quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1324 		} else {
1325 			quota_cut_bytes = 0;
1326 		}
1327 	}
1328 
1329 	if (p_s_un_bh) {
1330 		int off;
1331 		char *data;
1332 
1333 		/* We are in direct2indirect conversion, so move tail contents
1334 		   to the unformatted node */
1335 		/* note, we do the copy before preparing the buffer because we
1336 		 ** don't care about the contents of the unformatted node yet.
1337 		 ** the only thing we really care about is the direct item's data
1338 		 ** is in the unformatted node.
1339 		 **
1340 		 ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1341 		 ** the unformatted node, which might schedule, meaning we'd have to
1342 		 ** loop all the way back up to the start of the while loop.
1343 		 **
1344 		 ** The unformatted node must be dirtied later on.  We can't be
1345 		 ** sure here if the entire tail has been deleted yet.
1346 		 **
1347 		 ** p_s_un_bh is from the page cache (all unformatted nodes are
1348 		 ** from the page cache) and might be a highmem page.  So, we
1349 		 ** can't use p_s_un_bh->b_data.
1350 		 ** -clm
1351 		 */
1352 
1353 		data = kmap_atomic(p_s_un_bh->b_page, KM_USER0);
1354 		off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1355 		memcpy(data + off,
1356 		       B_I_PITEM(PATH_PLAST_BUFFER(p_s_path), &s_ih),
1357 		       n_ret_value);
1358 		kunmap_atomic(data, KM_USER0);
1359 	}
1360 	/* Perform balancing after all resources have been collected at once. */
1361 	do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1362 
1363 #ifdef REISERQUOTA_DEBUG
1364 	reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1365 		       "reiserquota delete_item(): freeing %u, id=%u type=%c",
1366 		       quota_cut_bytes, p_s_inode->i_uid, head2type(&s_ih));
1367 #endif
1368 	DQUOT_FREE_SPACE_NODIRTY(p_s_inode, quota_cut_bytes);
1369 
1370 	/* Return deleted body length */
1371 	return n_ret_value;
1372 }
1373 
1374 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1375 
1376  deletion of the body of the object is performed by iput(), with the
1377  result that if multiple processes are operating on a file, the
1378  deletion of the body of the file is deferred until the last process
1379  that has an open inode performs its iput().
1380 
1381  writes and truncates are protected from collisions by use of
1382  semaphores.
1383 
1384  creates, linking, and mknod are protected from collisions with other
1385  processes by making the reiserfs_add_entry() the last step in the
1386  creation, and then rolling back all changes if there was a collision.
1387  - Hans
1388 */
1389 
1390 /* this deletes item which never gets split */
1391 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1392 				struct inode *inode, struct reiserfs_key *key)
1393 {
1394 	struct tree_balance tb;
1395 	INITIALIZE_PATH(path);
1396 	int item_len = 0;
1397 	int tb_init = 0;
1398 	struct cpu_key cpu_key;
1399 	int retval;
1400 	int quota_cut_bytes = 0;
1401 
1402 	BUG_ON(!th->t_trans_id);
1403 
1404 	le_key2cpu_key(&cpu_key, key);
1405 
1406 	while (1) {
1407 		retval = search_item(th->t_super, &cpu_key, &path);
1408 		if (retval == IO_ERROR) {
1409 			reiserfs_warning(th->t_super,
1410 					 "vs-5350: reiserfs_delete_solid_item: "
1411 					 "i/o failure occurred trying to delete %K",
1412 					 &cpu_key);
1413 			break;
1414 		}
1415 		if (retval != ITEM_FOUND) {
1416 			pathrelse(&path);
1417 			// No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1418 			if (!
1419 			    ((unsigned long long)
1420 			     GET_HASH_VALUE(le_key_k_offset
1421 					    (le_key_version(key), key)) == 0
1422 			     && (unsigned long long)
1423 			     GET_GENERATION_NUMBER(le_key_k_offset
1424 						   (le_key_version(key),
1425 						    key)) == 1))
1426 				reiserfs_warning(th->t_super,
1427 						 "vs-5355: reiserfs_delete_solid_item: %k not found",
1428 						 key);
1429 			break;
1430 		}
1431 		if (!tb_init) {
1432 			tb_init = 1;
1433 			item_len = ih_item_len(PATH_PITEM_HEAD(&path));
1434 			init_tb_struct(th, &tb, th->t_super, &path,
1435 				       -(IH_SIZE + item_len));
1436 		}
1437 		quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path));
1438 
1439 		retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1440 		if (retval == REPEAT_SEARCH) {
1441 			PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1442 			continue;
1443 		}
1444 
1445 		if (retval == CARRY_ON) {
1446 			do_balance(&tb, NULL, NULL, M_DELETE);
1447 			if (inode) {	/* Should we count quota for item? (we don't count quotas for save-links) */
1448 #ifdef REISERQUOTA_DEBUG
1449 				reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1450 					       "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1451 					       quota_cut_bytes, inode->i_uid,
1452 					       key2type(key));
1453 #endif
1454 				DQUOT_FREE_SPACE_NODIRTY(inode,
1455 							 quota_cut_bytes);
1456 			}
1457 			break;
1458 		}
1459 		// IO_ERROR, NO_DISK_SPACE, etc
1460 		reiserfs_warning(th->t_super,
1461 				 "vs-5360: reiserfs_delete_solid_item: "
1462 				 "could not delete %K due to fix_nodes failure",
1463 				 &cpu_key);
1464 		unfix_nodes(&tb);
1465 		break;
1466 	}
1467 
1468 	reiserfs_check_path(&path);
1469 }
1470 
1471 int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1472 			   struct inode *inode)
1473 {
1474 	int err;
1475 	inode->i_size = 0;
1476 	BUG_ON(!th->t_trans_id);
1477 
1478 	/* for directory this deletes item containing "." and ".." */
1479 	err =
1480 	    reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1481 	if (err)
1482 		return err;
1483 
1484 #if defined( USE_INODE_GENERATION_COUNTER )
1485 	if (!old_format_only(th->t_super)) {
1486 		__le32 *inode_generation;
1487 
1488 		inode_generation =
1489 		    &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
1490 		*inode_generation =
1491 		    cpu_to_le32(le32_to_cpu(*inode_generation) + 1);
1492 	}
1493 /* USE_INODE_GENERATION_COUNTER */
1494 #endif
1495 	reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1496 
1497 	return err;
1498 }
1499 
1500 static void unmap_buffers(struct page *page, loff_t pos)
1501 {
1502 	struct buffer_head *bh;
1503 	struct buffer_head *head;
1504 	struct buffer_head *next;
1505 	unsigned long tail_index;
1506 	unsigned long cur_index;
1507 
1508 	if (page) {
1509 		if (page_has_buffers(page)) {
1510 			tail_index = pos & (PAGE_CACHE_SIZE - 1);
1511 			cur_index = 0;
1512 			head = page_buffers(page);
1513 			bh = head;
1514 			do {
1515 				next = bh->b_this_page;
1516 
1517 				/* we want to unmap the buffers that contain the tail, and
1518 				 ** all the buffers after it (since the tail must be at the
1519 				 ** end of the file).  We don't want to unmap file data
1520 				 ** before the tail, since it might be dirty and waiting to
1521 				 ** reach disk
1522 				 */
1523 				cur_index += bh->b_size;
1524 				if (cur_index > tail_index) {
1525 					reiserfs_unmap_buffer(bh);
1526 				}
1527 				bh = next;
1528 			} while (bh != head);
1529 			if (PAGE_SIZE == bh->b_size) {
1530 				clear_page_dirty(page);
1531 			}
1532 		}
1533 	}
1534 }
1535 
1536 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
1537 				    struct inode *p_s_inode,
1538 				    struct page *page,
1539 				    struct path *p_s_path,
1540 				    const struct cpu_key *p_s_item_key,
1541 				    loff_t n_new_file_size, char *p_c_mode)
1542 {
1543 	struct super_block *p_s_sb = p_s_inode->i_sb;
1544 	int n_block_size = p_s_sb->s_blocksize;
1545 	int cut_bytes;
1546 	BUG_ON(!th->t_trans_id);
1547 
1548 	if (n_new_file_size != p_s_inode->i_size)
1549 		BUG();
1550 
1551 	/* the page being sent in could be NULL if there was an i/o error
1552 	 ** reading in the last block.  The user will hit problems trying to
1553 	 ** read the file, but for now we just skip the indirect2direct
1554 	 */
1555 	if (atomic_read(&p_s_inode->i_count) > 1 ||
1556 	    !tail_has_to_be_packed(p_s_inode) ||
1557 	    !page || (REISERFS_I(p_s_inode)->i_flags & i_nopack_mask)) {
1558 		// leave tail in an unformatted node
1559 		*p_c_mode = M_SKIP_BALANCING;
1560 		cut_bytes =
1561 		    n_block_size - (n_new_file_size & (n_block_size - 1));
1562 		pathrelse(p_s_path);
1563 		return cut_bytes;
1564 	}
1565 	/* Permorm the conversion to a direct_item. */
1566 	/*return indirect_to_direct (p_s_inode, p_s_path, p_s_item_key, n_new_file_size, p_c_mode); */
1567 	return indirect2direct(th, p_s_inode, page, p_s_path, p_s_item_key,
1568 			       n_new_file_size, p_c_mode);
1569 }
1570 
1571 /* we did indirect_to_direct conversion. And we have inserted direct
1572    item successesfully, but there were no disk space to cut unfm
1573    pointer being converted. Therefore we have to delete inserted
1574    direct item(s) */
1575 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
1576 					 struct inode *inode, struct path *path)
1577 {
1578 	struct cpu_key tail_key;
1579 	int tail_len;
1580 	int removed;
1581 	BUG_ON(!th->t_trans_id);
1582 
1583 	make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);	// !!!!
1584 	tail_key.key_length = 4;
1585 
1586 	tail_len =
1587 	    (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1588 	while (tail_len) {
1589 		/* look for the last byte of the tail */
1590 		if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1591 		    POSITION_NOT_FOUND)
1592 			reiserfs_panic(inode->i_sb,
1593 				       "vs-5615: indirect_to_direct_roll_back: found invalid item");
1594 		RFALSE(path->pos_in_item !=
1595 		       ih_item_len(PATH_PITEM_HEAD(path)) - 1,
1596 		       "vs-5616: appended bytes found");
1597 		PATH_LAST_POSITION(path)--;
1598 
1599 		removed =
1600 		    reiserfs_delete_item(th, path, &tail_key, inode,
1601 					 NULL /*unbh not needed */ );
1602 		RFALSE(removed <= 0
1603 		       || removed > tail_len,
1604 		       "vs-5617: there was tail %d bytes, removed item length %d bytes",
1605 		       tail_len, removed);
1606 		tail_len -= removed;
1607 		set_cpu_key_k_offset(&tail_key,
1608 				     cpu_key_k_offset(&tail_key) - removed);
1609 	}
1610 	reiserfs_warning(inode->i_sb,
1611 			 "indirect_to_direct_roll_back: indirect_to_direct conversion has been rolled back due to lack of disk space");
1612 	//mark_file_without_tail (inode);
1613 	mark_inode_dirty(inode);
1614 }
1615 
1616 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1617 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
1618 			   struct path *p_s_path,
1619 			   struct cpu_key *p_s_item_key,
1620 			   struct inode *p_s_inode,
1621 			   struct page *page, loff_t n_new_file_size)
1622 {
1623 	struct super_block *p_s_sb = p_s_inode->i_sb;
1624 	/* Every function which is going to call do_balance must first
1625 	   create a tree_balance structure.  Then it must fill up this
1626 	   structure by using the init_tb_struct and fix_nodes functions.
1627 	   After that we can make tree balancing. */
1628 	struct tree_balance s_cut_balance;
1629 	struct item_head *p_le_ih;
1630 	int n_cut_size = 0,	/* Amount to be cut. */
1631 	    n_ret_value = CARRY_ON, n_removed = 0,	/* Number of the removed unformatted nodes. */
1632 	    n_is_inode_locked = 0;
1633 	char c_mode;		/* Mode of the balance. */
1634 	int retval2 = -1;
1635 	int quota_cut_bytes;
1636 	loff_t tail_pos = 0;
1637 
1638 	BUG_ON(!th->t_trans_id);
1639 
1640 	init_tb_struct(th, &s_cut_balance, p_s_inode->i_sb, p_s_path,
1641 		       n_cut_size);
1642 
1643 	/* Repeat this loop until we either cut the item without needing
1644 	   to balance, or we fix_nodes without schedule occurring */
1645 	while (1) {
1646 		/* Determine the balance mode, position of the first byte to
1647 		   be cut, and size to be cut.  In case of the indirect item
1648 		   free unformatted nodes which are pointed to by the cut
1649 		   pointers. */
1650 
1651 		c_mode =
1652 		    prepare_for_delete_or_cut(th, p_s_inode, p_s_path,
1653 					      p_s_item_key, &n_removed,
1654 					      &n_cut_size, n_new_file_size);
1655 		if (c_mode == M_CONVERT) {
1656 			/* convert last unformatted node to direct item or leave
1657 			   tail in the unformatted node */
1658 			RFALSE(n_ret_value != CARRY_ON,
1659 			       "PAP-5570: can not convert twice");
1660 
1661 			n_ret_value =
1662 			    maybe_indirect_to_direct(th, p_s_inode, page,
1663 						     p_s_path, p_s_item_key,
1664 						     n_new_file_size, &c_mode);
1665 			if (c_mode == M_SKIP_BALANCING)
1666 				/* tail has been left in the unformatted node */
1667 				return n_ret_value;
1668 
1669 			n_is_inode_locked = 1;
1670 
1671 			/* removing of last unformatted node will change value we
1672 			   have to return to truncate. Save it */
1673 			retval2 = n_ret_value;
1674 			/*retval2 = p_s_sb->s_blocksize - (n_new_file_size & (p_s_sb->s_blocksize - 1)); */
1675 
1676 			/* So, we have performed the first part of the conversion:
1677 			   inserting the new direct item.  Now we are removing the
1678 			   last unformatted node pointer. Set key to search for
1679 			   it. */
1680 			set_cpu_key_k_type(p_s_item_key, TYPE_INDIRECT);
1681 			p_s_item_key->key_length = 4;
1682 			n_new_file_size -=
1683 			    (n_new_file_size & (p_s_sb->s_blocksize - 1));
1684 			tail_pos = n_new_file_size;
1685 			set_cpu_key_k_offset(p_s_item_key, n_new_file_size + 1);
1686 			if (search_for_position_by_key
1687 			    (p_s_sb, p_s_item_key,
1688 			     p_s_path) == POSITION_NOT_FOUND) {
1689 				print_block(PATH_PLAST_BUFFER(p_s_path), 3,
1690 					    PATH_LAST_POSITION(p_s_path) - 1,
1691 					    PATH_LAST_POSITION(p_s_path) + 1);
1692 				reiserfs_panic(p_s_sb,
1693 					       "PAP-5580: reiserfs_cut_from_item: item to convert does not exist (%K)",
1694 					       p_s_item_key);
1695 			}
1696 			continue;
1697 		}
1698 		if (n_cut_size == 0) {
1699 			pathrelse(p_s_path);
1700 			return 0;
1701 		}
1702 
1703 		s_cut_balance.insert_size[0] = n_cut_size;
1704 
1705 		n_ret_value = fix_nodes(c_mode, &s_cut_balance, NULL, NULL);
1706 		if (n_ret_value != REPEAT_SEARCH)
1707 			break;
1708 
1709 		PROC_INFO_INC(p_s_sb, cut_from_item_restarted);
1710 
1711 		n_ret_value =
1712 		    search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1713 		if (n_ret_value == POSITION_FOUND)
1714 			continue;
1715 
1716 		reiserfs_warning(p_s_sb,
1717 				 "PAP-5610: reiserfs_cut_from_item: item %K not found",
1718 				 p_s_item_key);
1719 		unfix_nodes(&s_cut_balance);
1720 		return (n_ret_value == IO_ERROR) ? -EIO : -ENOENT;
1721 	}			/* while */
1722 
1723 	// check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1724 	if (n_ret_value != CARRY_ON) {
1725 		if (n_is_inode_locked) {
1726 			// FIXME: this seems to be not needed: we are always able
1727 			// to cut item
1728 			indirect_to_direct_roll_back(th, p_s_inode, p_s_path);
1729 		}
1730 		if (n_ret_value == NO_DISK_SPACE)
1731 			reiserfs_warning(p_s_sb, "NO_DISK_SPACE");
1732 		unfix_nodes(&s_cut_balance);
1733 		return -EIO;
1734 	}
1735 
1736 	/* go ahead and perform balancing */
1737 
1738 	RFALSE(c_mode == M_PASTE || c_mode == M_INSERT, "invalid mode");
1739 
1740 	/* Calculate number of bytes that need to be cut from the item. */
1741 	quota_cut_bytes =
1742 	    (c_mode ==
1743 	     M_DELETE) ? ih_item_len(get_ih(p_s_path)) : -s_cut_balance.
1744 	    insert_size[0];
1745 	if (retval2 == -1)
1746 		n_ret_value = calc_deleted_bytes_number(&s_cut_balance, c_mode);
1747 	else
1748 		n_ret_value = retval2;
1749 
1750 	/* For direct items, we only change the quota when deleting the last
1751 	 ** item.
1752 	 */
1753 	p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path);
1754 	if (!S_ISLNK(p_s_inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1755 		if (c_mode == M_DELETE &&
1756 		    (le_ih_k_offset(p_le_ih) & (p_s_sb->s_blocksize - 1)) ==
1757 		    1) {
1758 			// FIXME: this is to keep 3.5 happy
1759 			REISERFS_I(p_s_inode)->i_first_direct_byte = U32_MAX;
1760 			quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1761 		} else {
1762 			quota_cut_bytes = 0;
1763 		}
1764 	}
1765 #ifdef CONFIG_REISERFS_CHECK
1766 	if (n_is_inode_locked) {
1767 		struct item_head *le_ih =
1768 		    PATH_PITEM_HEAD(s_cut_balance.tb_path);
1769 		/* we are going to complete indirect2direct conversion. Make
1770 		   sure, that we exactly remove last unformatted node pointer
1771 		   of the item */
1772 		if (!is_indirect_le_ih(le_ih))
1773 			reiserfs_panic(p_s_sb,
1774 				       "vs-5652: reiserfs_cut_from_item: "
1775 				       "item must be indirect %h", le_ih);
1776 
1777 		if (c_mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1778 			reiserfs_panic(p_s_sb,
1779 				       "vs-5653: reiserfs_cut_from_item: "
1780 				       "completing indirect2direct conversion indirect item %h "
1781 				       "being deleted must be of 4 byte long",
1782 				       le_ih);
1783 
1784 		if (c_mode == M_CUT
1785 		    && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1786 			reiserfs_panic(p_s_sb,
1787 				       "vs-5654: reiserfs_cut_from_item: "
1788 				       "can not complete indirect2direct conversion of %h (CUT, insert_size==%d)",
1789 				       le_ih, s_cut_balance.insert_size[0]);
1790 		}
1791 		/* it would be useful to make sure, that right neighboring
1792 		   item is direct item of this file */
1793 	}
1794 #endif
1795 
1796 	do_balance(&s_cut_balance, NULL, NULL, c_mode);
1797 	if (n_is_inode_locked) {
1798 		/* we've done an indirect->direct conversion.  when the data block
1799 		 ** was freed, it was removed from the list of blocks that must
1800 		 ** be flushed before the transaction commits, make sure to
1801 		 ** unmap and invalidate it
1802 		 */
1803 		unmap_buffers(page, tail_pos);
1804 		REISERFS_I(p_s_inode)->i_flags &= ~i_pack_on_close_mask;
1805 	}
1806 #ifdef REISERQUOTA_DEBUG
1807 	reiserfs_debug(p_s_inode->i_sb, REISERFS_DEBUG_CODE,
1808 		       "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1809 		       quota_cut_bytes, p_s_inode->i_uid, '?');
1810 #endif
1811 	DQUOT_FREE_SPACE_NODIRTY(p_s_inode, quota_cut_bytes);
1812 	return n_ret_value;
1813 }
1814 
1815 static void truncate_directory(struct reiserfs_transaction_handle *th,
1816 			       struct inode *inode)
1817 {
1818 	BUG_ON(!th->t_trans_id);
1819 	if (inode->i_nlink)
1820 		reiserfs_warning(inode->i_sb,
1821 				 "vs-5655: truncate_directory: link count != 0");
1822 
1823 	set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1824 	set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1825 	reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1826 	reiserfs_update_sd(th, inode);
1827 	set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1828 	set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
1829 }
1830 
1831 /* Truncate file to the new size. Note, this must be called with a transaction
1832    already started */
1833 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th, struct inode *p_s_inode,	/* ->i_size contains new
1834 												   size */
1835 			 struct page *page,	/* up to date for last block */
1836 			 int update_timestamps	/* when it is called by
1837 						   file_release to convert
1838 						   the tail - no timestamps
1839 						   should be updated */
1840     )
1841 {
1842 	INITIALIZE_PATH(s_search_path);	/* Path to the current object item. */
1843 	struct item_head *p_le_ih;	/* Pointer to an item header. */
1844 	struct cpu_key s_item_key;	/* Key to search for a previous file item. */
1845 	loff_t n_file_size,	/* Old file size. */
1846 	 n_new_file_size;	/* New file size. */
1847 	int n_deleted;		/* Number of deleted or truncated bytes. */
1848 	int retval;
1849 	int err = 0;
1850 
1851 	BUG_ON(!th->t_trans_id);
1852 	if (!
1853 	    (S_ISREG(p_s_inode->i_mode) || S_ISDIR(p_s_inode->i_mode)
1854 	     || S_ISLNK(p_s_inode->i_mode)))
1855 		return 0;
1856 
1857 	if (S_ISDIR(p_s_inode->i_mode)) {
1858 		// deletion of directory - no need to update timestamps
1859 		truncate_directory(th, p_s_inode);
1860 		return 0;
1861 	}
1862 
1863 	/* Get new file size. */
1864 	n_new_file_size = p_s_inode->i_size;
1865 
1866 	// FIXME: note, that key type is unimportant here
1867 	make_cpu_key(&s_item_key, p_s_inode, max_reiserfs_offset(p_s_inode),
1868 		     TYPE_DIRECT, 3);
1869 
1870 	retval =
1871 	    search_for_position_by_key(p_s_inode->i_sb, &s_item_key,
1872 				       &s_search_path);
1873 	if (retval == IO_ERROR) {
1874 		reiserfs_warning(p_s_inode->i_sb,
1875 				 "vs-5657: reiserfs_do_truncate: "
1876 				 "i/o failure occurred trying to truncate %K",
1877 				 &s_item_key);
1878 		err = -EIO;
1879 		goto out;
1880 	}
1881 	if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1882 		reiserfs_warning(p_s_inode->i_sb,
1883 				 "PAP-5660: reiserfs_do_truncate: "
1884 				 "wrong result %d of search for %K", retval,
1885 				 &s_item_key);
1886 
1887 		err = -EIO;
1888 		goto out;
1889 	}
1890 
1891 	s_search_path.pos_in_item--;
1892 
1893 	/* Get real file size (total length of all file items) */
1894 	p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1895 	if (is_statdata_le_ih(p_le_ih))
1896 		n_file_size = 0;
1897 	else {
1898 		loff_t offset = le_ih_k_offset(p_le_ih);
1899 		int bytes =
1900 		    op_bytes_number(p_le_ih, p_s_inode->i_sb->s_blocksize);
1901 
1902 		/* this may mismatch with real file size: if last direct item
1903 		   had no padding zeros and last unformatted node had no free
1904 		   space, this file would have this file size */
1905 		n_file_size = offset + bytes - 1;
1906 	}
1907 	/*
1908 	 * are we doing a full truncate or delete, if so
1909 	 * kick in the reada code
1910 	 */
1911 	if (n_new_file_size == 0)
1912 		s_search_path.reada = PATH_READA | PATH_READA_BACK;
1913 
1914 	if (n_file_size == 0 || n_file_size < n_new_file_size) {
1915 		goto update_and_out;
1916 	}
1917 
1918 	/* Update key to search for the last file item. */
1919 	set_cpu_key_k_offset(&s_item_key, n_file_size);
1920 
1921 	do {
1922 		/* Cut or delete file item. */
1923 		n_deleted =
1924 		    reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
1925 					   p_s_inode, page, n_new_file_size);
1926 		if (n_deleted < 0) {
1927 			reiserfs_warning(p_s_inode->i_sb,
1928 					 "vs-5665: reiserfs_do_truncate: reiserfs_cut_from_item failed");
1929 			reiserfs_check_path(&s_search_path);
1930 			return 0;
1931 		}
1932 
1933 		RFALSE(n_deleted > n_file_size,
1934 		       "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1935 		       n_deleted, n_file_size, &s_item_key);
1936 
1937 		/* Change key to search the last file item. */
1938 		n_file_size -= n_deleted;
1939 
1940 		set_cpu_key_k_offset(&s_item_key, n_file_size);
1941 
1942 		/* While there are bytes to truncate and previous file item is presented in the tree. */
1943 
1944 		/*
1945 		 ** This loop could take a really long time, and could log
1946 		 ** many more blocks than a transaction can hold.  So, we do a polite
1947 		 ** journal end here, and if the transaction needs ending, we make
1948 		 ** sure the file is consistent before ending the current trans
1949 		 ** and starting a new one
1950 		 */
1951 		if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1952 			int orig_len_alloc = th->t_blocks_allocated;
1953 			decrement_counters_in_path(&s_search_path);
1954 
1955 			if (update_timestamps) {
1956 				p_s_inode->i_mtime = p_s_inode->i_ctime =
1957 				    CURRENT_TIME_SEC;
1958 			}
1959 			reiserfs_update_sd(th, p_s_inode);
1960 
1961 			err = journal_end(th, p_s_inode->i_sb, orig_len_alloc);
1962 			if (err)
1963 				goto out;
1964 			err = journal_begin(th, p_s_inode->i_sb,
1965 					    JOURNAL_PER_BALANCE_CNT * 6);
1966 			if (err)
1967 				goto out;
1968 			reiserfs_update_inode_transaction(p_s_inode);
1969 		}
1970 	} while (n_file_size > ROUND_UP(n_new_file_size) &&
1971 		 search_for_position_by_key(p_s_inode->i_sb, &s_item_key,
1972 					    &s_search_path) == POSITION_FOUND);
1973 
1974 	RFALSE(n_file_size > ROUND_UP(n_new_file_size),
1975 	       "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1976 	       n_new_file_size, n_file_size, s_item_key.on_disk_key.k_objectid);
1977 
1978       update_and_out:
1979 	if (update_timestamps) {
1980 		// this is truncate, not file closing
1981 		p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME_SEC;
1982 	}
1983 	reiserfs_update_sd(th, p_s_inode);
1984 
1985       out:
1986 	pathrelse(&s_search_path);
1987 	return err;
1988 }
1989 
1990 #ifdef CONFIG_REISERFS_CHECK
1991 // this makes sure, that we __append__, not overwrite or add holes
1992 static void check_research_for_paste(struct path *path,
1993 				     const struct cpu_key *p_s_key)
1994 {
1995 	struct item_head *found_ih = get_ih(path);
1996 
1997 	if (is_direct_le_ih(found_ih)) {
1998 		if (le_ih_k_offset(found_ih) +
1999 		    op_bytes_number(found_ih,
2000 				    get_last_bh(path)->b_size) !=
2001 		    cpu_key_k_offset(p_s_key)
2002 		    || op_bytes_number(found_ih,
2003 				       get_last_bh(path)->b_size) !=
2004 		    pos_in_item(path))
2005 			reiserfs_panic(NULL,
2006 				       "PAP-5720: check_research_for_paste: "
2007 				       "found direct item %h or position (%d) does not match to key %K",
2008 				       found_ih, pos_in_item(path), p_s_key);
2009 	}
2010 	if (is_indirect_le_ih(found_ih)) {
2011 		if (le_ih_k_offset(found_ih) +
2012 		    op_bytes_number(found_ih,
2013 				    get_last_bh(path)->b_size) !=
2014 		    cpu_key_k_offset(p_s_key)
2015 		    || I_UNFM_NUM(found_ih) != pos_in_item(path)
2016 		    || get_ih_free_space(found_ih) != 0)
2017 			reiserfs_panic(NULL,
2018 				       "PAP-5730: check_research_for_paste: "
2019 				       "found indirect item (%h) or position (%d) does not match to key (%K)",
2020 				       found_ih, pos_in_item(path), p_s_key);
2021 	}
2022 }
2023 #endif				/* config reiserfs check */
2024 
2025 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
2026 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct path *p_s_search_path,	/* Path to the pasted item.          */
2027 			     const struct cpu_key *p_s_key,	/* Key to search for the needed item. */
2028 			     struct inode *inode,	/* Inode item belongs to */
2029 			     const char *p_c_body,	/* Pointer to the bytes to paste.    */
2030 			     int n_pasted_size)
2031 {				/* Size of pasted bytes.             */
2032 	struct tree_balance s_paste_balance;
2033 	int retval;
2034 	int fs_gen;
2035 
2036 	BUG_ON(!th->t_trans_id);
2037 
2038 	fs_gen = get_generation(inode->i_sb);
2039 
2040 #ifdef REISERQUOTA_DEBUG
2041 	reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2042 		       "reiserquota paste_into_item(): allocating %u id=%u type=%c",
2043 		       n_pasted_size, inode->i_uid,
2044 		       key2type(&(p_s_key->on_disk_key)));
2045 #endif
2046 
2047 	if (DQUOT_ALLOC_SPACE_NODIRTY(inode, n_pasted_size)) {
2048 		pathrelse(p_s_search_path);
2049 		return -EDQUOT;
2050 	}
2051 	init_tb_struct(th, &s_paste_balance, th->t_super, p_s_search_path,
2052 		       n_pasted_size);
2053 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2054 	s_paste_balance.key = p_s_key->on_disk_key;
2055 #endif
2056 
2057 	/* DQUOT_* can schedule, must check before the fix_nodes */
2058 	if (fs_changed(fs_gen, inode->i_sb)) {
2059 		goto search_again;
2060 	}
2061 
2062 	while ((retval =
2063 		fix_nodes(M_PASTE, &s_paste_balance, NULL,
2064 			  p_c_body)) == REPEAT_SEARCH) {
2065 	      search_again:
2066 		/* file system changed while we were in the fix_nodes */
2067 		PROC_INFO_INC(th->t_super, paste_into_item_restarted);
2068 		retval =
2069 		    search_for_position_by_key(th->t_super, p_s_key,
2070 					       p_s_search_path);
2071 		if (retval == IO_ERROR) {
2072 			retval = -EIO;
2073 			goto error_out;
2074 		}
2075 		if (retval == POSITION_FOUND) {
2076 			reiserfs_warning(inode->i_sb,
2077 					 "PAP-5710: reiserfs_paste_into_item: entry or pasted byte (%K) exists",
2078 					 p_s_key);
2079 			retval = -EEXIST;
2080 			goto error_out;
2081 		}
2082 #ifdef CONFIG_REISERFS_CHECK
2083 		check_research_for_paste(p_s_search_path, p_s_key);
2084 #endif
2085 	}
2086 
2087 	/* Perform balancing after all resources are collected by fix_nodes, and
2088 	   accessing them will not risk triggering schedule. */
2089 	if (retval == CARRY_ON) {
2090 		do_balance(&s_paste_balance, NULL /*ih */ , p_c_body, M_PASTE);
2091 		return 0;
2092 	}
2093 	retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2094       error_out:
2095 	/* this also releases the path */
2096 	unfix_nodes(&s_paste_balance);
2097 #ifdef REISERQUOTA_DEBUG
2098 	reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2099 		       "reiserquota paste_into_item(): freeing %u id=%u type=%c",
2100 		       n_pasted_size, inode->i_uid,
2101 		       key2type(&(p_s_key->on_disk_key)));
2102 #endif
2103 	DQUOT_FREE_SPACE_NODIRTY(inode, n_pasted_size);
2104 	return retval;
2105 }
2106 
2107 /* Insert new item into the buffer at the path. */
2108 int reiserfs_insert_item(struct reiserfs_transaction_handle *th, struct path *p_s_path,	/* Path to the inserteded item.         */
2109 			 const struct cpu_key *key, struct item_head *p_s_ih,	/* Pointer to the item header to insert. */
2110 			 struct inode *inode, const char *p_c_body)
2111 {				/* Pointer to the bytes to insert.      */
2112 	struct tree_balance s_ins_balance;
2113 	int retval;
2114 	int fs_gen = 0;
2115 	int quota_bytes = 0;
2116 
2117 	BUG_ON(!th->t_trans_id);
2118 
2119 	if (inode) {		/* Do we count quotas for item? */
2120 		fs_gen = get_generation(inode->i_sb);
2121 		quota_bytes = ih_item_len(p_s_ih);
2122 
2123 		/* hack so the quota code doesn't have to guess if the file has
2124 		 ** a tail, links are always tails, so there's no guessing needed
2125 		 */
2126 		if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_s_ih)) {
2127 			quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
2128 		}
2129 #ifdef REISERQUOTA_DEBUG
2130 		reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2131 			       "reiserquota insert_item(): allocating %u id=%u type=%c",
2132 			       quota_bytes, inode->i_uid, head2type(p_s_ih));
2133 #endif
2134 		/* We can't dirty inode here. It would be immediately written but
2135 		 * appropriate stat item isn't inserted yet... */
2136 		if (DQUOT_ALLOC_SPACE_NODIRTY(inode, quota_bytes)) {
2137 			pathrelse(p_s_path);
2138 			return -EDQUOT;
2139 		}
2140 	}
2141 	init_tb_struct(th, &s_ins_balance, th->t_super, p_s_path,
2142 		       IH_SIZE + ih_item_len(p_s_ih));
2143 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2144 	s_ins_balance.key = key->on_disk_key;
2145 #endif
2146 	/* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2147 	if (inode && fs_changed(fs_gen, inode->i_sb)) {
2148 		goto search_again;
2149 	}
2150 
2151 	while ((retval =
2152 		fix_nodes(M_INSERT, &s_ins_balance, p_s_ih,
2153 			  p_c_body)) == REPEAT_SEARCH) {
2154 	      search_again:
2155 		/* file system changed while we were in the fix_nodes */
2156 		PROC_INFO_INC(th->t_super, insert_item_restarted);
2157 		retval = search_item(th->t_super, key, p_s_path);
2158 		if (retval == IO_ERROR) {
2159 			retval = -EIO;
2160 			goto error_out;
2161 		}
2162 		if (retval == ITEM_FOUND) {
2163 			reiserfs_warning(th->t_super,
2164 					 "PAP-5760: reiserfs_insert_item: "
2165 					 "key %K already exists in the tree",
2166 					 key);
2167 			retval = -EEXIST;
2168 			goto error_out;
2169 		}
2170 	}
2171 
2172 	/* make balancing after all resources will be collected at a time */
2173 	if (retval == CARRY_ON) {
2174 		do_balance(&s_ins_balance, p_s_ih, p_c_body, M_INSERT);
2175 		return 0;
2176 	}
2177 
2178 	retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2179       error_out:
2180 	/* also releases the path */
2181 	unfix_nodes(&s_ins_balance);
2182 #ifdef REISERQUOTA_DEBUG
2183 	reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2184 		       "reiserquota insert_item(): freeing %u id=%u type=%c",
2185 		       quota_bytes, inode->i_uid, head2type(p_s_ih));
2186 #endif
2187 	if (inode)
2188 		DQUOT_FREE_SPACE_NODIRTY(inode, quota_bytes);
2189 	return retval;
2190 }
2191