xref: /openbmc/linux/fs/reiserfs/stree.c (revision d47a97bd)
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 #include <linux/time.h>
12 #include <linux/string.h>
13 #include <linux/pagemap.h>
14 #include <linux/bio.h>
15 #include "reiserfs.h"
16 #include <linux/buffer_head.h>
17 #include <linux/quotaops.h>
18 
19 /* Does the buffer contain a disk block which is in the tree. */
20 inline int B_IS_IN_TREE(const struct buffer_head *bh)
21 {
22 
23 	RFALSE(B_LEVEL(bh) > MAX_HEIGHT,
24 	       "PAP-1010: block (%b) has too big level (%z)", bh, bh);
25 
26 	return (B_LEVEL(bh) != FREE_LEVEL);
27 }
28 
29 /* to get item head in le form */
30 inline void copy_item_head(struct item_head *to,
31 			   const struct item_head *from)
32 {
33 	memcpy(to, from, IH_SIZE);
34 }
35 
36 /*
37  * k1 is pointer to on-disk structure which is stored in little-endian
38  * form. k2 is pointer to cpu variable. For key of items of the same
39  * object this returns 0.
40  * Returns: -1 if key1 < key2
41  * 0 if key1 == key2
42  * 1 if key1 > key2
43  */
44 inline int comp_short_keys(const struct reiserfs_key *le_key,
45 			   const struct cpu_key *cpu_key)
46 {
47 	__u32 n;
48 	n = le32_to_cpu(le_key->k_dir_id);
49 	if (n < cpu_key->on_disk_key.k_dir_id)
50 		return -1;
51 	if (n > cpu_key->on_disk_key.k_dir_id)
52 		return 1;
53 	n = le32_to_cpu(le_key->k_objectid);
54 	if (n < cpu_key->on_disk_key.k_objectid)
55 		return -1;
56 	if (n > cpu_key->on_disk_key.k_objectid)
57 		return 1;
58 	return 0;
59 }
60 
61 /*
62  * k1 is pointer to on-disk structure which is stored in little-endian
63  * form. k2 is pointer to cpu variable.
64  * Compare keys using all 4 key fields.
65  * Returns: -1 if key1 < key2 0
66  * if key1 = key2 1 if key1 > key2
67  */
68 static inline int comp_keys(const struct reiserfs_key *le_key,
69 			    const struct cpu_key *cpu_key)
70 {
71 	int retval;
72 
73 	retval = comp_short_keys(le_key, cpu_key);
74 	if (retval)
75 		return retval;
76 	if (le_key_k_offset(le_key_version(le_key), le_key) <
77 	    cpu_key_k_offset(cpu_key))
78 		return -1;
79 	if (le_key_k_offset(le_key_version(le_key), le_key) >
80 	    cpu_key_k_offset(cpu_key))
81 		return 1;
82 
83 	if (cpu_key->key_length == 3)
84 		return 0;
85 
86 	/* this part is needed only when tail conversion is in progress */
87 	if (le_key_k_type(le_key_version(le_key), le_key) <
88 	    cpu_key_k_type(cpu_key))
89 		return -1;
90 
91 	if (le_key_k_type(le_key_version(le_key), le_key) >
92 	    cpu_key_k_type(cpu_key))
93 		return 1;
94 
95 	return 0;
96 }
97 
98 inline int comp_short_le_keys(const struct reiserfs_key *key1,
99 			      const struct reiserfs_key *key2)
100 {
101 	__u32 *k1_u32, *k2_u32;
102 	int key_length = REISERFS_SHORT_KEY_LEN;
103 
104 	k1_u32 = (__u32 *) key1;
105 	k2_u32 = (__u32 *) key2;
106 	for (; key_length--; ++k1_u32, ++k2_u32) {
107 		if (le32_to_cpu(*k1_u32) < le32_to_cpu(*k2_u32))
108 			return -1;
109 		if (le32_to_cpu(*k1_u32) > le32_to_cpu(*k2_u32))
110 			return 1;
111 	}
112 	return 0;
113 }
114 
115 inline void le_key2cpu_key(struct cpu_key *to, const struct reiserfs_key *from)
116 {
117 	int version;
118 	to->on_disk_key.k_dir_id = le32_to_cpu(from->k_dir_id);
119 	to->on_disk_key.k_objectid = le32_to_cpu(from->k_objectid);
120 
121 	/* find out version of the key */
122 	version = le_key_version(from);
123 	to->version = version;
124 	to->on_disk_key.k_offset = le_key_k_offset(version, from);
125 	to->on_disk_key.k_type = le_key_k_type(version, from);
126 }
127 
128 /*
129  * this does not say which one is bigger, it only returns 1 if keys
130  * are not equal, 0 otherwise
131  */
132 inline int comp_le_keys(const struct reiserfs_key *k1,
133 			const struct reiserfs_key *k2)
134 {
135 	return memcmp(k1, k2, sizeof(struct reiserfs_key));
136 }
137 
138 /**************************************************************************
139  *  Binary search toolkit function                                        *
140  *  Search for an item in the array by the item key                       *
141  *  Returns:    1 if found,  0 if not found;                              *
142  *        *pos = number of the searched element if found, else the        *
143  *        number of the first element that is larger than key.            *
144  **************************************************************************/
145 /*
146  * For those not familiar with binary search: lbound is the leftmost item
147  * that it could be, rbound the rightmost item that it could be.  We examine
148  * the item halfway between lbound and rbound, and that tells us either
149  * that we can increase lbound, or decrease rbound, or that we have found it,
150  * or if lbound <= rbound that there are no possible items, and we have not
151  * found it. With each examination we cut the number of possible items it
152  * could be by one more than half rounded down, or we find it.
153  */
154 static inline int bin_search(const void *key,	/* Key to search for. */
155 			     const void *base,	/* First item in the array. */
156 			     int num,	/* Number of items in the array. */
157 			     /*
158 			      * Item size in the array.  searched. Lest the
159 			      * reader be confused, note that this is crafted
160 			      * as a general function, and when it is applied
161 			      * specifically to the array of item headers in a
162 			      * node, width is actually the item header size
163 			      * not the item size.
164 			      */
165 			     int width,
166 			     int *pos /* Number of the searched for element. */
167     )
168 {
169 	int rbound, lbound, j;
170 
171 	for (j = ((rbound = num - 1) + (lbound = 0)) / 2;
172 	     lbound <= rbound; j = (rbound + lbound) / 2)
173 		switch (comp_keys
174 			((struct reiserfs_key *)((char *)base + j * width),
175 			 (struct cpu_key *)key)) {
176 		case -1:
177 			lbound = j + 1;
178 			continue;
179 		case 1:
180 			rbound = j - 1;
181 			continue;
182 		case 0:
183 			*pos = j;
184 			return ITEM_FOUND;	/* Key found in the array.  */
185 		}
186 
187 	/*
188 	 * bin_search did not find given key, it returns position of key,
189 	 * that is minimal and greater than the given one.
190 	 */
191 	*pos = lbound;
192 	return ITEM_NOT_FOUND;
193 }
194 
195 
196 /* Minimal possible key. It is never in the tree. */
197 const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} };
198 
199 /* Maximal possible key. It is never in the tree. */
200 static const struct reiserfs_key MAX_KEY = {
201 	cpu_to_le32(0xffffffff),
202 	cpu_to_le32(0xffffffff),
203 	{{cpu_to_le32(0xffffffff),
204 	  cpu_to_le32(0xffffffff)},}
205 };
206 
207 /*
208  * Get delimiting key of the buffer by looking for it in the buffers in the
209  * path, starting from the bottom of the path, and going upwards.  We must
210  * check the path's validity at each step.  If the key is not in the path,
211  * there is no delimiting key in the tree (buffer is first or last buffer
212  * in tree), and in this case we return a special key, either MIN_KEY or
213  * MAX_KEY.
214  */
215 static inline const struct reiserfs_key *get_lkey(const struct treepath *chk_path,
216 						  const struct super_block *sb)
217 {
218 	int position, path_offset = chk_path->path_length;
219 	struct buffer_head *parent;
220 
221 	RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET,
222 	       "PAP-5010: invalid offset in the path");
223 
224 	/* While not higher in path than first element. */
225 	while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
226 
227 		RFALSE(!buffer_uptodate
228 		       (PATH_OFFSET_PBUFFER(chk_path, path_offset)),
229 		       "PAP-5020: parent is not uptodate");
230 
231 		/* Parent at the path is not in the tree now. */
232 		if (!B_IS_IN_TREE
233 		    (parent =
234 		     PATH_OFFSET_PBUFFER(chk_path, path_offset)))
235 			return &MAX_KEY;
236 		/* Check whether position in the parent is correct. */
237 		if ((position =
238 		     PATH_OFFSET_POSITION(chk_path,
239 					  path_offset)) >
240 		    B_NR_ITEMS(parent))
241 			return &MAX_KEY;
242 		/* Check whether parent at the path really points to the child. */
243 		if (B_N_CHILD_NUM(parent, position) !=
244 		    PATH_OFFSET_PBUFFER(chk_path,
245 					path_offset + 1)->b_blocknr)
246 			return &MAX_KEY;
247 		/*
248 		 * Return delimiting key if position in the parent
249 		 * is not equal to zero.
250 		 */
251 		if (position)
252 			return internal_key(parent, position - 1);
253 	}
254 	/* Return MIN_KEY if we are in the root of the buffer tree. */
255 	if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)->
256 	    b_blocknr == SB_ROOT_BLOCK(sb))
257 		return &MIN_KEY;
258 	return &MAX_KEY;
259 }
260 
261 /* Get delimiting key of the buffer at the path and its right neighbor. */
262 inline const struct reiserfs_key *get_rkey(const struct treepath *chk_path,
263 					   const struct super_block *sb)
264 {
265 	int position, path_offset = chk_path->path_length;
266 	struct buffer_head *parent;
267 
268 	RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET,
269 	       "PAP-5030: invalid offset in the path");
270 
271 	while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
272 
273 		RFALSE(!buffer_uptodate
274 		       (PATH_OFFSET_PBUFFER(chk_path, path_offset)),
275 		       "PAP-5040: parent is not uptodate");
276 
277 		/* Parent at the path is not in the tree now. */
278 		if (!B_IS_IN_TREE
279 		    (parent =
280 		     PATH_OFFSET_PBUFFER(chk_path, path_offset)))
281 			return &MIN_KEY;
282 		/* Check whether position in the parent is correct. */
283 		if ((position =
284 		     PATH_OFFSET_POSITION(chk_path,
285 					  path_offset)) >
286 		    B_NR_ITEMS(parent))
287 			return &MIN_KEY;
288 		/*
289 		 * Check whether parent at the path really points
290 		 * to the child.
291 		 */
292 		if (B_N_CHILD_NUM(parent, position) !=
293 		    PATH_OFFSET_PBUFFER(chk_path,
294 					path_offset + 1)->b_blocknr)
295 			return &MIN_KEY;
296 
297 		/*
298 		 * Return delimiting key if position in the parent
299 		 * is not the last one.
300 		 */
301 		if (position != B_NR_ITEMS(parent))
302 			return internal_key(parent, position);
303 	}
304 
305 	/* Return MAX_KEY if we are in the root of the buffer tree. */
306 	if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)->
307 	    b_blocknr == SB_ROOT_BLOCK(sb))
308 		return &MAX_KEY;
309 	return &MIN_KEY;
310 }
311 
312 /*
313  * Check whether a key is contained in the tree rooted from a buffer at a path.
314  * This works by looking at the left and right delimiting keys for the buffer
315  * in the last path_element in the path.  These delimiting keys are stored
316  * at least one level above that buffer in the tree. If the buffer is the
317  * first or last node in the tree order then one of the delimiting keys may
318  * be absent, and in this case get_lkey and get_rkey return a special key
319  * which is MIN_KEY or MAX_KEY.
320  */
321 static inline int key_in_buffer(
322 				/* Path which should be checked. */
323 				struct treepath *chk_path,
324 				/* Key which should be checked. */
325 				const struct cpu_key *key,
326 				struct super_block *sb
327     )
328 {
329 
330 	RFALSE(!key || chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET
331 	       || chk_path->path_length > MAX_HEIGHT,
332 	       "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
333 	       key, chk_path->path_length);
334 	RFALSE(!PATH_PLAST_BUFFER(chk_path)->b_bdev,
335 	       "PAP-5060: device must not be NODEV");
336 
337 	if (comp_keys(get_lkey(chk_path, sb), key) == 1)
338 		/* left delimiting key is bigger, that the key we look for */
339 		return 0;
340 	/*  if ( comp_keys(key, get_rkey(chk_path, sb)) != -1 ) */
341 	if (comp_keys(get_rkey(chk_path, sb), key) != 1)
342 		/* key must be less than right delimitiing key */
343 		return 0;
344 	return 1;
345 }
346 
347 int reiserfs_check_path(struct treepath *p)
348 {
349 	RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
350 	       "path not properly relsed");
351 	return 0;
352 }
353 
354 /*
355  * Drop the reference to each buffer in a path and restore
356  * dirty bits clean when preparing the buffer for the log.
357  * This version should only be called from fix_nodes()
358  */
359 void pathrelse_and_restore(struct super_block *sb,
360 			   struct treepath *search_path)
361 {
362 	int path_offset = search_path->path_length;
363 
364 	RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
365 	       "clm-4000: invalid path offset");
366 
367 	while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
368 		struct buffer_head *bh;
369 		bh = PATH_OFFSET_PBUFFER(search_path, path_offset--);
370 		reiserfs_restore_prepared_buffer(sb, bh);
371 		brelse(bh);
372 	}
373 	search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
374 }
375 
376 /* Drop the reference to each buffer in a path */
377 void pathrelse(struct treepath *search_path)
378 {
379 	int path_offset = search_path->path_length;
380 
381 	RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
382 	       "PAP-5090: invalid path offset");
383 
384 	while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
385 		brelse(PATH_OFFSET_PBUFFER(search_path, path_offset--));
386 
387 	search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
388 }
389 
390 static int has_valid_deh_location(struct buffer_head *bh, struct item_head *ih)
391 {
392 	struct reiserfs_de_head *deh;
393 	int i;
394 
395 	deh = B_I_DEH(bh, ih);
396 	for (i = 0; i < ih_entry_count(ih); i++) {
397 		if (deh_location(&deh[i]) > ih_item_len(ih)) {
398 			reiserfs_warning(NULL, "reiserfs-5094",
399 					 "directory entry location seems wrong %h",
400 					 &deh[i]);
401 			return 0;
402 		}
403 	}
404 
405 	return 1;
406 }
407 
408 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
409 {
410 	struct block_head *blkh;
411 	struct item_head *ih;
412 	int used_space;
413 	int prev_location;
414 	int i;
415 	int nr;
416 
417 	blkh = (struct block_head *)buf;
418 	if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
419 		reiserfs_warning(NULL, "reiserfs-5080",
420 				 "this should be caught earlier");
421 		return 0;
422 	}
423 
424 	nr = blkh_nr_item(blkh);
425 	if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
426 		/* item number is too big or too small */
427 		reiserfs_warning(NULL, "reiserfs-5081",
428 				 "nr_item seems wrong: %z", bh);
429 		return 0;
430 	}
431 	ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
432 	used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih));
433 
434 	/* free space does not match to calculated amount of use space */
435 	if (used_space != blocksize - blkh_free_space(blkh)) {
436 		reiserfs_warning(NULL, "reiserfs-5082",
437 				 "free space seems wrong: %z", bh);
438 		return 0;
439 	}
440 	/*
441 	 * FIXME: it is_leaf will hit performance too much - we may have
442 	 * return 1 here
443 	 */
444 
445 	/* check tables of item heads */
446 	ih = (struct item_head *)(buf + BLKH_SIZE);
447 	prev_location = blocksize;
448 	for (i = 0; i < nr; i++, ih++) {
449 		if (le_ih_k_type(ih) == TYPE_ANY) {
450 			reiserfs_warning(NULL, "reiserfs-5083",
451 					 "wrong item type for item %h",
452 					 ih);
453 			return 0;
454 		}
455 		if (ih_location(ih) >= blocksize
456 		    || ih_location(ih) < IH_SIZE * nr) {
457 			reiserfs_warning(NULL, "reiserfs-5084",
458 					 "item location seems wrong: %h",
459 					 ih);
460 			return 0;
461 		}
462 		if (ih_item_len(ih) < 1
463 		    || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
464 			reiserfs_warning(NULL, "reiserfs-5085",
465 					 "item length seems wrong: %h",
466 					 ih);
467 			return 0;
468 		}
469 		if (prev_location - ih_location(ih) != ih_item_len(ih)) {
470 			reiserfs_warning(NULL, "reiserfs-5086",
471 					 "item location seems wrong "
472 					 "(second one): %h", ih);
473 			return 0;
474 		}
475 		if (is_direntry_le_ih(ih)) {
476 			if (ih_item_len(ih) < (ih_entry_count(ih) * IH_SIZE)) {
477 				reiserfs_warning(NULL, "reiserfs-5093",
478 						 "item entry count seems wrong %h",
479 						 ih);
480 				return 0;
481 			}
482 			return has_valid_deh_location(bh, ih);
483 		}
484 		prev_location = ih_location(ih);
485 	}
486 
487 	/* one may imagine many more checks */
488 	return 1;
489 }
490 
491 /* returns 1 if buf looks like an internal node, 0 otherwise */
492 static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
493 {
494 	struct block_head *blkh;
495 	int nr;
496 	int used_space;
497 
498 	blkh = (struct block_head *)buf;
499 	nr = blkh_level(blkh);
500 	if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
501 		/* this level is not possible for internal nodes */
502 		reiserfs_warning(NULL, "reiserfs-5087",
503 				 "this should be caught earlier");
504 		return 0;
505 	}
506 
507 	nr = blkh_nr_item(blkh);
508 	/* for internal which is not root we might check min number of keys */
509 	if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
510 		reiserfs_warning(NULL, "reiserfs-5088",
511 				 "number of key seems wrong: %z", bh);
512 		return 0;
513 	}
514 
515 	used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
516 	if (used_space != blocksize - blkh_free_space(blkh)) {
517 		reiserfs_warning(NULL, "reiserfs-5089",
518 				 "free space seems wrong: %z", bh);
519 		return 0;
520 	}
521 
522 	/* one may imagine many more checks */
523 	return 1;
524 }
525 
526 /*
527  * make sure that bh contains formatted node of reiserfs tree of
528  * 'level'-th level
529  */
530 static int is_tree_node(struct buffer_head *bh, int level)
531 {
532 	if (B_LEVEL(bh) != level) {
533 		reiserfs_warning(NULL, "reiserfs-5090", "node level %d does "
534 				 "not match to the expected one %d",
535 				 B_LEVEL(bh), level);
536 		return 0;
537 	}
538 	if (level == DISK_LEAF_NODE_LEVEL)
539 		return is_leaf(bh->b_data, bh->b_size, bh);
540 
541 	return is_internal(bh->b_data, bh->b_size, bh);
542 }
543 
544 #define SEARCH_BY_KEY_READA 16
545 
546 /*
547  * The function is NOT SCHEDULE-SAFE!
548  * It might unlock the write lock if we needed to wait for a block
549  * to be read. Note that in this case it won't recover the lock to avoid
550  * high contention resulting from too much lock requests, especially
551  * the caller (search_by_key) will perform other schedule-unsafe
552  * operations just after calling this function.
553  *
554  * @return depth of lock to be restored after read completes
555  */
556 static int search_by_key_reada(struct super_block *s,
557 				struct buffer_head **bh,
558 				b_blocknr_t *b, int num)
559 {
560 	int i, j;
561 	int depth = -1;
562 
563 	for (i = 0; i < num; i++) {
564 		bh[i] = sb_getblk(s, b[i]);
565 	}
566 	/*
567 	 * We are going to read some blocks on which we
568 	 * have a reference. It's safe, though we might be
569 	 * reading blocks concurrently changed if we release
570 	 * the lock. But it's still fine because we check later
571 	 * if the tree changed
572 	 */
573 	for (j = 0; j < i; j++) {
574 		/*
575 		 * note, this needs attention if we are getting rid of the BKL
576 		 * you have to make sure the prepared bit isn't set on this
577 		 * buffer
578 		 */
579 		if (!buffer_uptodate(bh[j])) {
580 			if (depth == -1)
581 				depth = reiserfs_write_unlock_nested(s);
582 			bh_readahead(bh[j], REQ_RAHEAD);
583 		}
584 		brelse(bh[j]);
585 	}
586 	return depth;
587 }
588 
589 /*
590  * This function fills up the path from the root to the leaf as it
591  * descends the tree looking for the key.  It uses reiserfs_bread to
592  * try to find buffers in the cache given their block number.  If it
593  * does not find them in the cache it reads them from disk.  For each
594  * node search_by_key finds using reiserfs_bread it then uses
595  * bin_search to look through that node.  bin_search will find the
596  * position of the block_number of the next node if it is looking
597  * through an internal node.  If it is looking through a leaf node
598  * bin_search will find the position of the item which has key either
599  * equal to given key, or which is the maximal key less than the given
600  * key.  search_by_key returns a path that must be checked for the
601  * correctness of the top of the path but need not be checked for the
602  * correctness of the bottom of the path
603  */
604 /*
605  * search_by_key - search for key (and item) in stree
606  * @sb: superblock
607  * @key: pointer to key to search for
608  * @search_path: Allocated and initialized struct treepath; Returned filled
609  *		 on success.
610  * @stop_level: How far down the tree to search, Use DISK_LEAF_NODE_LEVEL to
611  *		stop at leaf level.
612  *
613  * The function is NOT SCHEDULE-SAFE!
614  */
615 int search_by_key(struct super_block *sb, const struct cpu_key *key,
616 		  struct treepath *search_path, int stop_level)
617 {
618 	b_blocknr_t block_number;
619 	int expected_level;
620 	struct buffer_head *bh;
621 	struct path_element *last_element;
622 	int node_level, retval;
623 	int fs_gen;
624 	struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
625 	b_blocknr_t reada_blocks[SEARCH_BY_KEY_READA];
626 	int reada_count = 0;
627 
628 #ifdef CONFIG_REISERFS_CHECK
629 	int repeat_counter = 0;
630 #endif
631 
632 	PROC_INFO_INC(sb, search_by_key);
633 
634 	/*
635 	 * As we add each node to a path we increase its count.  This means
636 	 * that we must be careful to release all nodes in a path before we
637 	 * either discard the path struct or re-use the path struct, as we
638 	 * do here.
639 	 */
640 
641 	pathrelse(search_path);
642 
643 	/*
644 	 * With each iteration of this loop we search through the items in the
645 	 * current node, and calculate the next current node(next path element)
646 	 * for the next iteration of this loop..
647 	 */
648 	block_number = SB_ROOT_BLOCK(sb);
649 	expected_level = -1;
650 	while (1) {
651 
652 #ifdef CONFIG_REISERFS_CHECK
653 		if (!(++repeat_counter % 50000))
654 			reiserfs_warning(sb, "PAP-5100",
655 					 "%s: there were %d iterations of "
656 					 "while loop looking for key %K",
657 					 current->comm, repeat_counter,
658 					 key);
659 #endif
660 
661 		/* prep path to have another element added to it. */
662 		last_element =
663 		    PATH_OFFSET_PELEMENT(search_path,
664 					 ++search_path->path_length);
665 		fs_gen = get_generation(sb);
666 
667 		/*
668 		 * Read the next tree node, and set the last element
669 		 * in the path to have a pointer to it.
670 		 */
671 		if ((bh = last_element->pe_buffer =
672 		     sb_getblk(sb, block_number))) {
673 
674 			/*
675 			 * We'll need to drop the lock if we encounter any
676 			 * buffers that need to be read. If all of them are
677 			 * already up to date, we don't need to drop the lock.
678 			 */
679 			int depth = -1;
680 
681 			if (!buffer_uptodate(bh) && reada_count > 1)
682 				depth = search_by_key_reada(sb, reada_bh,
683 						    reada_blocks, reada_count);
684 
685 			if (!buffer_uptodate(bh) && depth == -1)
686 				depth = reiserfs_write_unlock_nested(sb);
687 
688 			bh_read_nowait(bh, 0);
689 			wait_on_buffer(bh);
690 
691 			if (depth != -1)
692 				reiserfs_write_lock_nested(sb, depth);
693 			if (!buffer_uptodate(bh))
694 				goto io_error;
695 		} else {
696 io_error:
697 			search_path->path_length--;
698 			pathrelse(search_path);
699 			return IO_ERROR;
700 		}
701 		reada_count = 0;
702 		if (expected_level == -1)
703 			expected_level = SB_TREE_HEIGHT(sb);
704 		expected_level--;
705 
706 		/*
707 		 * It is possible that schedule occurred. We must check
708 		 * whether the key to search is still in the tree rooted
709 		 * from the current buffer. If not then repeat search
710 		 * from the root.
711 		 */
712 		if (fs_changed(fs_gen, sb) &&
713 		    (!B_IS_IN_TREE(bh) ||
714 		     B_LEVEL(bh) != expected_level ||
715 		     !key_in_buffer(search_path, key, sb))) {
716 			PROC_INFO_INC(sb, search_by_key_fs_changed);
717 			PROC_INFO_INC(sb, search_by_key_restarted);
718 			PROC_INFO_INC(sb,
719 				      sbk_restarted[expected_level - 1]);
720 			pathrelse(search_path);
721 
722 			/*
723 			 * Get the root block number so that we can
724 			 * repeat the search starting from the root.
725 			 */
726 			block_number = SB_ROOT_BLOCK(sb);
727 			expected_level = -1;
728 
729 			/* repeat search from the root */
730 			continue;
731 		}
732 
733 		/*
734 		 * only check that the key is in the buffer if key is not
735 		 * equal to the MAX_KEY. Latter case is only possible in
736 		 * "finish_unfinished()" processing during mount.
737 		 */
738 		RFALSE(comp_keys(&MAX_KEY, key) &&
739 		       !key_in_buffer(search_path, key, sb),
740 		       "PAP-5130: key is not in the buffer");
741 #ifdef CONFIG_REISERFS_CHECK
742 		if (REISERFS_SB(sb)->cur_tb) {
743 			print_cur_tb("5140");
744 			reiserfs_panic(sb, "PAP-5140",
745 				       "schedule occurred in do_balance!");
746 		}
747 #endif
748 
749 		/*
750 		 * make sure, that the node contents look like a node of
751 		 * certain level
752 		 */
753 		if (!is_tree_node(bh, expected_level)) {
754 			reiserfs_error(sb, "vs-5150",
755 				       "invalid format found in block %ld. "
756 				       "Fsck?", bh->b_blocknr);
757 			pathrelse(search_path);
758 			return IO_ERROR;
759 		}
760 
761 		/* ok, we have acquired next formatted node in the tree */
762 		node_level = B_LEVEL(bh);
763 
764 		PROC_INFO_BH_STAT(sb, bh, node_level - 1);
765 
766 		RFALSE(node_level < stop_level,
767 		       "vs-5152: tree level (%d) is less than stop level (%d)",
768 		       node_level, stop_level);
769 
770 		retval = bin_search(key, item_head(bh, 0),
771 				      B_NR_ITEMS(bh),
772 				      (node_level ==
773 				       DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
774 				      KEY_SIZE,
775 				      &last_element->pe_position);
776 		if (node_level == stop_level) {
777 			return retval;
778 		}
779 
780 		/* we are not in the stop level */
781 		/*
782 		 * item has been found, so we choose the pointer which
783 		 * is to the right of the found one
784 		 */
785 		if (retval == ITEM_FOUND)
786 			last_element->pe_position++;
787 
788 		/*
789 		 * if item was not found we choose the position which is to
790 		 * the left of the found item. This requires no code,
791 		 * bin_search did it already.
792 		 */
793 
794 		/*
795 		 * So we have chosen a position in the current node which is
796 		 * an internal node.  Now we calculate child block number by
797 		 * position in the node.
798 		 */
799 		block_number =
800 		    B_N_CHILD_NUM(bh, last_element->pe_position);
801 
802 		/*
803 		 * if we are going to read leaf nodes, try for read
804 		 * ahead as well
805 		 */
806 		if ((search_path->reada & PATH_READA) &&
807 		    node_level == DISK_LEAF_NODE_LEVEL + 1) {
808 			int pos = last_element->pe_position;
809 			int limit = B_NR_ITEMS(bh);
810 			struct reiserfs_key *le_key;
811 
812 			if (search_path->reada & PATH_READA_BACK)
813 				limit = 0;
814 			while (reada_count < SEARCH_BY_KEY_READA) {
815 				if (pos == limit)
816 					break;
817 				reada_blocks[reada_count++] =
818 				    B_N_CHILD_NUM(bh, pos);
819 				if (search_path->reada & PATH_READA_BACK)
820 					pos--;
821 				else
822 					pos++;
823 
824 				/*
825 				 * check to make sure we're in the same object
826 				 */
827 				le_key = internal_key(bh, pos);
828 				if (le32_to_cpu(le_key->k_objectid) !=
829 				    key->on_disk_key.k_objectid) {
830 					break;
831 				}
832 			}
833 		}
834 	}
835 }
836 
837 /*
838  * Form the path to an item and position in this item which contains
839  * file byte defined by key. If there is no such item
840  * corresponding to the key, we point the path to the item with
841  * maximal key less than key, and *pos_in_item is set to one
842  * past the last entry/byte in the item.  If searching for entry in a
843  * directory item, and it is not found, *pos_in_item is set to one
844  * entry more than the entry with maximal key which is less than the
845  * sought key.
846  *
847  * Note that if there is no entry in this same node which is one more,
848  * then we point to an imaginary entry.  for direct items, the
849  * position is in units of bytes, for indirect items the position is
850  * in units of blocknr entries, for directory items the position is in
851  * units of directory entries.
852  */
853 /* The function is NOT SCHEDULE-SAFE! */
854 int search_for_position_by_key(struct super_block *sb,
855 			       /* Key to search (cpu variable) */
856 			       const struct cpu_key *p_cpu_key,
857 			       /* Filled up by this function. */
858 			       struct treepath *search_path)
859 {
860 	struct item_head *p_le_ih;	/* pointer to on-disk structure */
861 	int blk_size;
862 	loff_t item_offset, offset;
863 	struct reiserfs_dir_entry de;
864 	int retval;
865 
866 	/* If searching for directory entry. */
867 	if (is_direntry_cpu_key(p_cpu_key))
868 		return search_by_entry_key(sb, p_cpu_key, search_path,
869 					   &de);
870 
871 	/* If not searching for directory entry. */
872 
873 	/* If item is found. */
874 	retval = search_item(sb, p_cpu_key, search_path);
875 	if (retval == IO_ERROR)
876 		return retval;
877 	if (retval == ITEM_FOUND) {
878 
879 		RFALSE(!ih_item_len
880 		       (item_head
881 			(PATH_PLAST_BUFFER(search_path),
882 			 PATH_LAST_POSITION(search_path))),
883 		       "PAP-5165: item length equals zero");
884 
885 		pos_in_item(search_path) = 0;
886 		return POSITION_FOUND;
887 	}
888 
889 	RFALSE(!PATH_LAST_POSITION(search_path),
890 	       "PAP-5170: position equals zero");
891 
892 	/* Item is not found. Set path to the previous item. */
893 	p_le_ih =
894 	    item_head(PATH_PLAST_BUFFER(search_path),
895 			   --PATH_LAST_POSITION(search_path));
896 	blk_size = sb->s_blocksize;
897 
898 	if (comp_short_keys(&p_le_ih->ih_key, p_cpu_key))
899 		return FILE_NOT_FOUND;
900 
901 	/* FIXME: quite ugly this far */
902 
903 	item_offset = le_ih_k_offset(p_le_ih);
904 	offset = cpu_key_k_offset(p_cpu_key);
905 
906 	/* Needed byte is contained in the item pointed to by the path. */
907 	if (item_offset <= offset &&
908 	    item_offset + op_bytes_number(p_le_ih, blk_size) > offset) {
909 		pos_in_item(search_path) = offset - item_offset;
910 		if (is_indirect_le_ih(p_le_ih)) {
911 			pos_in_item(search_path) /= blk_size;
912 		}
913 		return POSITION_FOUND;
914 	}
915 
916 	/*
917 	 * Needed byte is not contained in the item pointed to by the
918 	 * path. Set pos_in_item out of the item.
919 	 */
920 	if (is_indirect_le_ih(p_le_ih))
921 		pos_in_item(search_path) =
922 		    ih_item_len(p_le_ih) / UNFM_P_SIZE;
923 	else
924 		pos_in_item(search_path) = ih_item_len(p_le_ih);
925 
926 	return POSITION_NOT_FOUND;
927 }
928 
929 /* Compare given item and item pointed to by the path. */
930 int comp_items(const struct item_head *stored_ih, const struct treepath *path)
931 {
932 	struct buffer_head *bh = PATH_PLAST_BUFFER(path);
933 	struct item_head *ih;
934 
935 	/* Last buffer at the path is not in the tree. */
936 	if (!B_IS_IN_TREE(bh))
937 		return 1;
938 
939 	/* Last path position is invalid. */
940 	if (PATH_LAST_POSITION(path) >= B_NR_ITEMS(bh))
941 		return 1;
942 
943 	/* we need only to know, whether it is the same item */
944 	ih = tp_item_head(path);
945 	return memcmp(stored_ih, ih, IH_SIZE);
946 }
947 
948 /* prepare for delete or cut of direct item */
949 static inline int prepare_for_direct_item(struct treepath *path,
950 					  struct item_head *le_ih,
951 					  struct inode *inode,
952 					  loff_t new_file_length, int *cut_size)
953 {
954 	loff_t round_len;
955 
956 	if (new_file_length == max_reiserfs_offset(inode)) {
957 		/* item has to be deleted */
958 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
959 		return M_DELETE;
960 	}
961 	/* new file gets truncated */
962 	if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
963 		round_len = ROUND_UP(new_file_length);
964 		/* this was new_file_length < le_ih ... */
965 		if (round_len < le_ih_k_offset(le_ih)) {
966 			*cut_size = -(IH_SIZE + ih_item_len(le_ih));
967 			return M_DELETE;	/* Delete this item. */
968 		}
969 		/* Calculate first position and size for cutting from item. */
970 		pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
971 		*cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
972 
973 		return M_CUT;	/* Cut from this item. */
974 	}
975 
976 	/* old file: items may have any length */
977 
978 	if (new_file_length < le_ih_k_offset(le_ih)) {
979 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
980 		return M_DELETE;	/* Delete this item. */
981 	}
982 
983 	/* Calculate first position and size for cutting from item. */
984 	*cut_size = -(ih_item_len(le_ih) -
985 		      (pos_in_item(path) =
986 		       new_file_length + 1 - le_ih_k_offset(le_ih)));
987 	return M_CUT;		/* Cut from this item. */
988 }
989 
990 static inline int prepare_for_direntry_item(struct treepath *path,
991 					    struct item_head *le_ih,
992 					    struct inode *inode,
993 					    loff_t new_file_length,
994 					    int *cut_size)
995 {
996 	if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
997 	    new_file_length == max_reiserfs_offset(inode)) {
998 		RFALSE(ih_entry_count(le_ih) != 2,
999 		       "PAP-5220: incorrect empty directory item (%h)", le_ih);
1000 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
1001 		/* Delete the directory item containing "." and ".." entry. */
1002 		return M_DELETE;
1003 	}
1004 
1005 	if (ih_entry_count(le_ih) == 1) {
1006 		/*
1007 		 * Delete the directory item such as there is one record only
1008 		 * in this item
1009 		 */
1010 		*cut_size = -(IH_SIZE + ih_item_len(le_ih));
1011 		return M_DELETE;
1012 	}
1013 
1014 	/* Cut one record from the directory item. */
1015 	*cut_size =
1016 	    -(DEH_SIZE +
1017 	      entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
1018 	return M_CUT;
1019 }
1020 
1021 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
1022 
1023 /*
1024  * If the path points to a directory or direct item, calculate mode
1025  * and the size cut, for balance.
1026  * If the path points to an indirect item, remove some number of its
1027  * unformatted nodes.
1028  * In case of file truncate calculate whether this item must be
1029  * deleted/truncated or last unformatted node of this item will be
1030  * converted to a direct item.
1031  * This function returns a determination of what balance mode the
1032  * calling function should employ.
1033  */
1034 static char prepare_for_delete_or_cut(struct reiserfs_transaction_handle *th,
1035 				      struct inode *inode,
1036 				      struct treepath *path,
1037 				      const struct cpu_key *item_key,
1038 				      /*
1039 				       * Number of unformatted nodes
1040 				       * which were removed from end
1041 				       * of the file.
1042 				       */
1043 				      int *removed,
1044 				      int *cut_size,
1045 				      /* MAX_KEY_OFFSET in case of delete. */
1046 				      unsigned long long new_file_length
1047     )
1048 {
1049 	struct super_block *sb = inode->i_sb;
1050 	struct item_head *p_le_ih = tp_item_head(path);
1051 	struct buffer_head *bh = PATH_PLAST_BUFFER(path);
1052 
1053 	BUG_ON(!th->t_trans_id);
1054 
1055 	/* Stat_data item. */
1056 	if (is_statdata_le_ih(p_le_ih)) {
1057 
1058 		RFALSE(new_file_length != max_reiserfs_offset(inode),
1059 		       "PAP-5210: mode must be M_DELETE");
1060 
1061 		*cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
1062 		return M_DELETE;
1063 	}
1064 
1065 	/* Directory item. */
1066 	if (is_direntry_le_ih(p_le_ih))
1067 		return prepare_for_direntry_item(path, p_le_ih, inode,
1068 						 new_file_length,
1069 						 cut_size);
1070 
1071 	/* Direct item. */
1072 	if (is_direct_le_ih(p_le_ih))
1073 		return prepare_for_direct_item(path, p_le_ih, inode,
1074 					       new_file_length, cut_size);
1075 
1076 	/* Case of an indirect item. */
1077 	{
1078 	    int blk_size = sb->s_blocksize;
1079 	    struct item_head s_ih;
1080 	    int need_re_search;
1081 	    int delete = 0;
1082 	    int result = M_CUT;
1083 	    int pos = 0;
1084 
1085 	    if ( new_file_length == max_reiserfs_offset (inode) ) {
1086 		/*
1087 		 * prepare_for_delete_or_cut() is called by
1088 		 * reiserfs_delete_item()
1089 		 */
1090 		new_file_length = 0;
1091 		delete = 1;
1092 	    }
1093 
1094 	    do {
1095 		need_re_search = 0;
1096 		*cut_size = 0;
1097 		bh = PATH_PLAST_BUFFER(path);
1098 		copy_item_head(&s_ih, tp_item_head(path));
1099 		pos = I_UNFM_NUM(&s_ih);
1100 
1101 		while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > new_file_length) {
1102 		    __le32 *unfm;
1103 		    __u32 block;
1104 
1105 		    /*
1106 		     * Each unformatted block deletion may involve
1107 		     * one additional bitmap block into the transaction,
1108 		     * thereby the initial journal space reservation
1109 		     * might not be enough.
1110 		     */
1111 		    if (!delete && (*cut_size) != 0 &&
1112 			reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD)
1113 			break;
1114 
1115 		    unfm = (__le32 *)ih_item_body(bh, &s_ih) + pos - 1;
1116 		    block = get_block_num(unfm, 0);
1117 
1118 		    if (block != 0) {
1119 			reiserfs_prepare_for_journal(sb, bh, 1);
1120 			put_block_num(unfm, 0, 0);
1121 			journal_mark_dirty(th, bh);
1122 			reiserfs_free_block(th, inode, block, 1);
1123 		    }
1124 
1125 		    reiserfs_cond_resched(sb);
1126 
1127 		    if (item_moved (&s_ih, path))  {
1128 			need_re_search = 1;
1129 			break;
1130 		    }
1131 
1132 		    pos --;
1133 		    (*removed)++;
1134 		    (*cut_size) -= UNFM_P_SIZE;
1135 
1136 		    if (pos == 0) {
1137 			(*cut_size) -= IH_SIZE;
1138 			result = M_DELETE;
1139 			break;
1140 		    }
1141 		}
1142 		/*
1143 		 * a trick.  If the buffer has been logged, this will
1144 		 * do nothing.  If we've broken the loop without logging
1145 		 * it, it will restore the buffer
1146 		 */
1147 		reiserfs_restore_prepared_buffer(sb, bh);
1148 	    } while (need_re_search &&
1149 		     search_for_position_by_key(sb, item_key, path) == POSITION_FOUND);
1150 	    pos_in_item(path) = pos * UNFM_P_SIZE;
1151 
1152 	    if (*cut_size == 0) {
1153 		/*
1154 		 * Nothing was cut. maybe convert last unformatted node to the
1155 		 * direct item?
1156 		 */
1157 		result = M_CONVERT;
1158 	    }
1159 	    return result;
1160 	}
1161 }
1162 
1163 /* Calculate number of bytes which will be deleted or cut during balance */
1164 static int calc_deleted_bytes_number(struct tree_balance *tb, char mode)
1165 {
1166 	int del_size;
1167 	struct item_head *p_le_ih = tp_item_head(tb->tb_path);
1168 
1169 	if (is_statdata_le_ih(p_le_ih))
1170 		return 0;
1171 
1172 	del_size =
1173 	    (mode ==
1174 	     M_DELETE) ? ih_item_len(p_le_ih) : -tb->insert_size[0];
1175 	if (is_direntry_le_ih(p_le_ih)) {
1176 		/*
1177 		 * return EMPTY_DIR_SIZE; We delete emty directories only.
1178 		 * we can't use EMPTY_DIR_SIZE, as old format dirs have a
1179 		 * different empty size.  ick. FIXME, is this right?
1180 		 */
1181 		return del_size;
1182 	}
1183 
1184 	if (is_indirect_le_ih(p_le_ih))
1185 		del_size = (del_size / UNFM_P_SIZE) *
1186 				(PATH_PLAST_BUFFER(tb->tb_path)->b_size);
1187 	return del_size;
1188 }
1189 
1190 static void init_tb_struct(struct reiserfs_transaction_handle *th,
1191 			   struct tree_balance *tb,
1192 			   struct super_block *sb,
1193 			   struct treepath *path, int size)
1194 {
1195 
1196 	BUG_ON(!th->t_trans_id);
1197 
1198 	memset(tb, '\0', sizeof(struct tree_balance));
1199 	tb->transaction_handle = th;
1200 	tb->tb_sb = sb;
1201 	tb->tb_path = path;
1202 	PATH_OFFSET_PBUFFER(path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1203 	PATH_OFFSET_POSITION(path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1204 	tb->insert_size[0] = size;
1205 }
1206 
1207 void padd_item(char *item, int total_length, int length)
1208 {
1209 	int i;
1210 
1211 	for (i = total_length; i > length;)
1212 		item[--i] = 0;
1213 }
1214 
1215 #ifdef REISERQUOTA_DEBUG
1216 char key2type(struct reiserfs_key *ih)
1217 {
1218 	if (is_direntry_le_key(2, ih))
1219 		return 'd';
1220 	if (is_direct_le_key(2, ih))
1221 		return 'D';
1222 	if (is_indirect_le_key(2, ih))
1223 		return 'i';
1224 	if (is_statdata_le_key(2, ih))
1225 		return 's';
1226 	return 'u';
1227 }
1228 
1229 char head2type(struct item_head *ih)
1230 {
1231 	if (is_direntry_le_ih(ih))
1232 		return 'd';
1233 	if (is_direct_le_ih(ih))
1234 		return 'D';
1235 	if (is_indirect_le_ih(ih))
1236 		return 'i';
1237 	if (is_statdata_le_ih(ih))
1238 		return 's';
1239 	return 'u';
1240 }
1241 #endif
1242 
1243 /*
1244  * Delete object item.
1245  * th       - active transaction handle
1246  * path     - path to the deleted item
1247  * item_key - key to search for the deleted item
1248  * indode   - used for updating i_blocks and quotas
1249  * un_bh    - NULL or unformatted node pointer
1250  */
1251 int reiserfs_delete_item(struct reiserfs_transaction_handle *th,
1252 			 struct treepath *path, const struct cpu_key *item_key,
1253 			 struct inode *inode, struct buffer_head *un_bh)
1254 {
1255 	struct super_block *sb = inode->i_sb;
1256 	struct tree_balance s_del_balance;
1257 	struct item_head s_ih;
1258 	struct item_head *q_ih;
1259 	int quota_cut_bytes;
1260 	int ret_value, del_size, removed;
1261 	int depth;
1262 
1263 #ifdef CONFIG_REISERFS_CHECK
1264 	char mode;
1265 	int iter = 0;
1266 #endif
1267 
1268 	BUG_ON(!th->t_trans_id);
1269 
1270 	init_tb_struct(th, &s_del_balance, sb, path,
1271 		       0 /*size is unknown */ );
1272 
1273 	while (1) {
1274 		removed = 0;
1275 
1276 #ifdef CONFIG_REISERFS_CHECK
1277 		iter++;
1278 		mode =
1279 #endif
1280 		    prepare_for_delete_or_cut(th, inode, path,
1281 					      item_key, &removed,
1282 					      &del_size,
1283 					      max_reiserfs_offset(inode));
1284 
1285 		RFALSE(mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1286 
1287 		copy_item_head(&s_ih, tp_item_head(path));
1288 		s_del_balance.insert_size[0] = del_size;
1289 
1290 		ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1291 		if (ret_value != REPEAT_SEARCH)
1292 			break;
1293 
1294 		PROC_INFO_INC(sb, delete_item_restarted);
1295 
1296 		/* file system changed, repeat search */
1297 		ret_value =
1298 		    search_for_position_by_key(sb, item_key, path);
1299 		if (ret_value == IO_ERROR)
1300 			break;
1301 		if (ret_value == FILE_NOT_FOUND) {
1302 			reiserfs_warning(sb, "vs-5340",
1303 					 "no items of the file %K found",
1304 					 item_key);
1305 			break;
1306 		}
1307 	}			/* while (1) */
1308 
1309 	if (ret_value != CARRY_ON) {
1310 		unfix_nodes(&s_del_balance);
1311 		return 0;
1312 	}
1313 
1314 	/* reiserfs_delete_item returns item length when success */
1315 	ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1316 	q_ih = tp_item_head(path);
1317 	quota_cut_bytes = ih_item_len(q_ih);
1318 
1319 	/*
1320 	 * hack so the quota code doesn't have to guess if the file has a
1321 	 * tail.  On tail insert, we allocate quota for 1 unformatted node.
1322 	 * We test the offset because the tail might have been
1323 	 * split into multiple items, and we only want to decrement for
1324 	 * the unfm node once
1325 	 */
1326 	if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(q_ih)) {
1327 		if ((le_ih_k_offset(q_ih) & (sb->s_blocksize - 1)) == 1) {
1328 			quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
1329 		} else {
1330 			quota_cut_bytes = 0;
1331 		}
1332 	}
1333 
1334 	if (un_bh) {
1335 		int off;
1336 		char *data;
1337 
1338 		/*
1339 		 * We are in direct2indirect conversion, so move tail contents
1340 		 * to the unformatted node
1341 		 */
1342 		/*
1343 		 * note, we do the copy before preparing the buffer because we
1344 		 * don't care about the contents of the unformatted node yet.
1345 		 * the only thing we really care about is the direct item's
1346 		 * data is in the unformatted node.
1347 		 *
1348 		 * Otherwise, we would have to call
1349 		 * reiserfs_prepare_for_journal on the unformatted node,
1350 		 * which might schedule, meaning we'd have to loop all the
1351 		 * way back up to the start of the while loop.
1352 		 *
1353 		 * The unformatted node must be dirtied later on.  We can't be
1354 		 * sure here if the entire tail has been deleted yet.
1355 		 *
1356 		 * un_bh is from the page cache (all unformatted nodes are
1357 		 * from the page cache) and might be a highmem page.  So, we
1358 		 * can't use un_bh->b_data.
1359 		 * -clm
1360 		 */
1361 
1362 		data = kmap_atomic(un_bh->b_page);
1363 		off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_SIZE - 1));
1364 		memcpy(data + off,
1365 		       ih_item_body(PATH_PLAST_BUFFER(path), &s_ih),
1366 		       ret_value);
1367 		kunmap_atomic(data);
1368 	}
1369 
1370 	/* Perform balancing after all resources have been collected at once. */
1371 	do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1372 
1373 #ifdef REISERQUOTA_DEBUG
1374 	reiserfs_debug(sb, REISERFS_DEBUG_CODE,
1375 		       "reiserquota delete_item(): freeing %u, id=%u type=%c",
1376 		       quota_cut_bytes, inode->i_uid, head2type(&s_ih));
1377 #endif
1378 	depth = reiserfs_write_unlock_nested(inode->i_sb);
1379 	dquot_free_space_nodirty(inode, quota_cut_bytes);
1380 	reiserfs_write_lock_nested(inode->i_sb, depth);
1381 
1382 	/* Return deleted body length */
1383 	return ret_value;
1384 }
1385 
1386 /*
1387  * Summary Of Mechanisms For Handling Collisions Between Processes:
1388  *
1389  *  deletion of the body of the object is performed by iput(), with the
1390  *  result that if multiple processes are operating on a file, the
1391  *  deletion of the body of the file is deferred until the last process
1392  *  that has an open inode performs its iput().
1393  *
1394  *  writes and truncates are protected from collisions by use of
1395  *  semaphores.
1396  *
1397  *  creates, linking, and mknod are protected from collisions with other
1398  *  processes by making the reiserfs_add_entry() the last step in the
1399  *  creation, and then rolling back all changes if there was a collision.
1400  *  - Hans
1401 */
1402 
1403 /* this deletes item which never gets split */
1404 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1405 				struct inode *inode, struct reiserfs_key *key)
1406 {
1407 	struct super_block *sb = th->t_super;
1408 	struct tree_balance tb;
1409 	INITIALIZE_PATH(path);
1410 	int item_len = 0;
1411 	int tb_init = 0;
1412 	struct cpu_key cpu_key;
1413 	int retval;
1414 	int quota_cut_bytes = 0;
1415 
1416 	BUG_ON(!th->t_trans_id);
1417 
1418 	le_key2cpu_key(&cpu_key, key);
1419 
1420 	while (1) {
1421 		retval = search_item(th->t_super, &cpu_key, &path);
1422 		if (retval == IO_ERROR) {
1423 			reiserfs_error(th->t_super, "vs-5350",
1424 				       "i/o failure occurred trying "
1425 				       "to delete %K", &cpu_key);
1426 			break;
1427 		}
1428 		if (retval != ITEM_FOUND) {
1429 			pathrelse(&path);
1430 			/*
1431 			 * No need for a warning, if there is just no free
1432 			 * space to insert '..' item into the
1433 			 * newly-created subdir
1434 			 */
1435 			if (!
1436 			    ((unsigned long long)
1437 			     GET_HASH_VALUE(le_key_k_offset
1438 					    (le_key_version(key), key)) == 0
1439 			     && (unsigned long long)
1440 			     GET_GENERATION_NUMBER(le_key_k_offset
1441 						   (le_key_version(key),
1442 						    key)) == 1))
1443 				reiserfs_warning(th->t_super, "vs-5355",
1444 						 "%k not found", key);
1445 			break;
1446 		}
1447 		if (!tb_init) {
1448 			tb_init = 1;
1449 			item_len = ih_item_len(tp_item_head(&path));
1450 			init_tb_struct(th, &tb, th->t_super, &path,
1451 				       -(IH_SIZE + item_len));
1452 		}
1453 		quota_cut_bytes = ih_item_len(tp_item_head(&path));
1454 
1455 		retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1456 		if (retval == REPEAT_SEARCH) {
1457 			PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1458 			continue;
1459 		}
1460 
1461 		if (retval == CARRY_ON) {
1462 			do_balance(&tb, NULL, NULL, M_DELETE);
1463 			/*
1464 			 * Should we count quota for item? (we don't
1465 			 * count quotas for save-links)
1466 			 */
1467 			if (inode) {
1468 				int depth;
1469 #ifdef REISERQUOTA_DEBUG
1470 				reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1471 					       "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1472 					       quota_cut_bytes, inode->i_uid,
1473 					       key2type(key));
1474 #endif
1475 				depth = reiserfs_write_unlock_nested(sb);
1476 				dquot_free_space_nodirty(inode,
1477 							 quota_cut_bytes);
1478 				reiserfs_write_lock_nested(sb, depth);
1479 			}
1480 			break;
1481 		}
1482 
1483 		/* IO_ERROR, NO_DISK_SPACE, etc */
1484 		reiserfs_warning(th->t_super, "vs-5360",
1485 				 "could not delete %K due to fix_nodes failure",
1486 				 &cpu_key);
1487 		unfix_nodes(&tb);
1488 		break;
1489 	}
1490 
1491 	reiserfs_check_path(&path);
1492 }
1493 
1494 int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1495 			   struct inode *inode)
1496 {
1497 	int err;
1498 	inode->i_size = 0;
1499 	BUG_ON(!th->t_trans_id);
1500 
1501 	/* for directory this deletes item containing "." and ".." */
1502 	err =
1503 	    reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1504 	if (err)
1505 		return err;
1506 
1507 #if defined( USE_INODE_GENERATION_COUNTER )
1508 	if (!old_format_only(th->t_super)) {
1509 		__le32 *inode_generation;
1510 
1511 		inode_generation =
1512 		    &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
1513 		le32_add_cpu(inode_generation, 1);
1514 	}
1515 /* USE_INODE_GENERATION_COUNTER */
1516 #endif
1517 	reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1518 
1519 	return err;
1520 }
1521 
1522 static void unmap_buffers(struct page *page, loff_t pos)
1523 {
1524 	struct buffer_head *bh;
1525 	struct buffer_head *head;
1526 	struct buffer_head *next;
1527 	unsigned long tail_index;
1528 	unsigned long cur_index;
1529 
1530 	if (page) {
1531 		if (page_has_buffers(page)) {
1532 			tail_index = pos & (PAGE_SIZE - 1);
1533 			cur_index = 0;
1534 			head = page_buffers(page);
1535 			bh = head;
1536 			do {
1537 				next = bh->b_this_page;
1538 
1539 				/*
1540 				 * we want to unmap the buffers that contain
1541 				 * the tail, and all the buffers after it
1542 				 * (since the tail must be at the end of the
1543 				 * file).  We don't want to unmap file data
1544 				 * before the tail, since it might be dirty
1545 				 * and waiting to reach disk
1546 				 */
1547 				cur_index += bh->b_size;
1548 				if (cur_index > tail_index) {
1549 					reiserfs_unmap_buffer(bh);
1550 				}
1551 				bh = next;
1552 			} while (bh != head);
1553 		}
1554 	}
1555 }
1556 
1557 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
1558 				    struct inode *inode,
1559 				    struct page *page,
1560 				    struct treepath *path,
1561 				    const struct cpu_key *item_key,
1562 				    loff_t new_file_size, char *mode)
1563 {
1564 	struct super_block *sb = inode->i_sb;
1565 	int block_size = sb->s_blocksize;
1566 	int cut_bytes;
1567 	BUG_ON(!th->t_trans_id);
1568 	BUG_ON(new_file_size != inode->i_size);
1569 
1570 	/*
1571 	 * the page being sent in could be NULL if there was an i/o error
1572 	 * reading in the last block.  The user will hit problems trying to
1573 	 * read the file, but for now we just skip the indirect2direct
1574 	 */
1575 	if (atomic_read(&inode->i_count) > 1 ||
1576 	    !tail_has_to_be_packed(inode) ||
1577 	    !page || (REISERFS_I(inode)->i_flags & i_nopack_mask)) {
1578 		/* leave tail in an unformatted node */
1579 		*mode = M_SKIP_BALANCING;
1580 		cut_bytes =
1581 		    block_size - (new_file_size & (block_size - 1));
1582 		pathrelse(path);
1583 		return cut_bytes;
1584 	}
1585 
1586 	/* Perform the conversion to a direct_item. */
1587 	return indirect2direct(th, inode, page, path, item_key,
1588 			       new_file_size, mode);
1589 }
1590 
1591 /*
1592  * we did indirect_to_direct conversion. And we have inserted direct
1593  * item successesfully, but there were no disk space to cut unfm
1594  * pointer being converted. Therefore we have to delete inserted
1595  * direct item(s)
1596  */
1597 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
1598 					 struct inode *inode, struct treepath *path)
1599 {
1600 	struct cpu_key tail_key;
1601 	int tail_len;
1602 	int removed;
1603 	BUG_ON(!th->t_trans_id);
1604 
1605 	make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);
1606 	tail_key.key_length = 4;
1607 
1608 	tail_len =
1609 	    (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1610 	while (tail_len) {
1611 		/* look for the last byte of the tail */
1612 		if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1613 		    POSITION_NOT_FOUND)
1614 			reiserfs_panic(inode->i_sb, "vs-5615",
1615 				       "found invalid item");
1616 		RFALSE(path->pos_in_item !=
1617 		       ih_item_len(tp_item_head(path)) - 1,
1618 		       "vs-5616: appended bytes found");
1619 		PATH_LAST_POSITION(path)--;
1620 
1621 		removed =
1622 		    reiserfs_delete_item(th, path, &tail_key, inode,
1623 					 NULL /*unbh not needed */ );
1624 		RFALSE(removed <= 0
1625 		       || removed > tail_len,
1626 		       "vs-5617: there was tail %d bytes, removed item length %d bytes",
1627 		       tail_len, removed);
1628 		tail_len -= removed;
1629 		set_cpu_key_k_offset(&tail_key,
1630 				     cpu_key_k_offset(&tail_key) - removed);
1631 	}
1632 	reiserfs_warning(inode->i_sb, "reiserfs-5091", "indirect_to_direct "
1633 			 "conversion has been rolled back due to "
1634 			 "lack of disk space");
1635 	mark_inode_dirty(inode);
1636 }
1637 
1638 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1639 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
1640 			   struct treepath *path,
1641 			   struct cpu_key *item_key,
1642 			   struct inode *inode,
1643 			   struct page *page, loff_t new_file_size)
1644 {
1645 	struct super_block *sb = inode->i_sb;
1646 	/*
1647 	 * Every function which is going to call do_balance must first
1648 	 * create a tree_balance structure.  Then it must fill up this
1649 	 * structure by using the init_tb_struct and fix_nodes functions.
1650 	 * After that we can make tree balancing.
1651 	 */
1652 	struct tree_balance s_cut_balance;
1653 	struct item_head *p_le_ih;
1654 	int cut_size = 0;	/* Amount to be cut. */
1655 	int ret_value = CARRY_ON;
1656 	int removed = 0;	/* Number of the removed unformatted nodes. */
1657 	int is_inode_locked = 0;
1658 	char mode;		/* Mode of the balance. */
1659 	int retval2 = -1;
1660 	int quota_cut_bytes;
1661 	loff_t tail_pos = 0;
1662 	int depth;
1663 
1664 	BUG_ON(!th->t_trans_id);
1665 
1666 	init_tb_struct(th, &s_cut_balance, inode->i_sb, path,
1667 		       cut_size);
1668 
1669 	/*
1670 	 * Repeat this loop until we either cut the item without needing
1671 	 * to balance, or we fix_nodes without schedule occurring
1672 	 */
1673 	while (1) {
1674 		/*
1675 		 * Determine the balance mode, position of the first byte to
1676 		 * be cut, and size to be cut.  In case of the indirect item
1677 		 * free unformatted nodes which are pointed to by the cut
1678 		 * pointers.
1679 		 */
1680 
1681 		mode =
1682 		    prepare_for_delete_or_cut(th, inode, path,
1683 					      item_key, &removed,
1684 					      &cut_size, new_file_size);
1685 		if (mode == M_CONVERT) {
1686 			/*
1687 			 * convert last unformatted node to direct item or
1688 			 * leave tail in the unformatted node
1689 			 */
1690 			RFALSE(ret_value != CARRY_ON,
1691 			       "PAP-5570: can not convert twice");
1692 
1693 			ret_value =
1694 			    maybe_indirect_to_direct(th, inode, page,
1695 						     path, item_key,
1696 						     new_file_size, &mode);
1697 			if (mode == M_SKIP_BALANCING)
1698 				/* tail has been left in the unformatted node */
1699 				return ret_value;
1700 
1701 			is_inode_locked = 1;
1702 
1703 			/*
1704 			 * removing of last unformatted node will
1705 			 * change value we have to return to truncate.
1706 			 * Save it
1707 			 */
1708 			retval2 = ret_value;
1709 
1710 			/*
1711 			 * So, we have performed the first part of the
1712 			 * conversion:
1713 			 * inserting the new direct item.  Now we are
1714 			 * removing the last unformatted node pointer.
1715 			 * Set key to search for it.
1716 			 */
1717 			set_cpu_key_k_type(item_key, TYPE_INDIRECT);
1718 			item_key->key_length = 4;
1719 			new_file_size -=
1720 			    (new_file_size & (sb->s_blocksize - 1));
1721 			tail_pos = new_file_size;
1722 			set_cpu_key_k_offset(item_key, new_file_size + 1);
1723 			if (search_for_position_by_key
1724 			    (sb, item_key,
1725 			     path) == POSITION_NOT_FOUND) {
1726 				print_block(PATH_PLAST_BUFFER(path), 3,
1727 					    PATH_LAST_POSITION(path) - 1,
1728 					    PATH_LAST_POSITION(path) + 1);
1729 				reiserfs_panic(sb, "PAP-5580", "item to "
1730 					       "convert does not exist (%K)",
1731 					       item_key);
1732 			}
1733 			continue;
1734 		}
1735 		if (cut_size == 0) {
1736 			pathrelse(path);
1737 			return 0;
1738 		}
1739 
1740 		s_cut_balance.insert_size[0] = cut_size;
1741 
1742 		ret_value = fix_nodes(mode, &s_cut_balance, NULL, NULL);
1743 		if (ret_value != REPEAT_SEARCH)
1744 			break;
1745 
1746 		PROC_INFO_INC(sb, cut_from_item_restarted);
1747 
1748 		ret_value =
1749 		    search_for_position_by_key(sb, item_key, path);
1750 		if (ret_value == POSITION_FOUND)
1751 			continue;
1752 
1753 		reiserfs_warning(sb, "PAP-5610", "item %K not found",
1754 				 item_key);
1755 		unfix_nodes(&s_cut_balance);
1756 		return (ret_value == IO_ERROR) ? -EIO : -ENOENT;
1757 	}			/* while */
1758 
1759 	/* check fix_nodes results (IO_ERROR or NO_DISK_SPACE) */
1760 	if (ret_value != CARRY_ON) {
1761 		if (is_inode_locked) {
1762 			/*
1763 			 * FIXME: this seems to be not needed: we are always
1764 			 * able to cut item
1765 			 */
1766 			indirect_to_direct_roll_back(th, inode, path);
1767 		}
1768 		if (ret_value == NO_DISK_SPACE)
1769 			reiserfs_warning(sb, "reiserfs-5092",
1770 					 "NO_DISK_SPACE");
1771 		unfix_nodes(&s_cut_balance);
1772 		return -EIO;
1773 	}
1774 
1775 	/* go ahead and perform balancing */
1776 
1777 	RFALSE(mode == M_PASTE || mode == M_INSERT, "invalid mode");
1778 
1779 	/* Calculate number of bytes that need to be cut from the item. */
1780 	quota_cut_bytes =
1781 	    (mode ==
1782 	     M_DELETE) ? ih_item_len(tp_item_head(path)) : -s_cut_balance.
1783 	    insert_size[0];
1784 	if (retval2 == -1)
1785 		ret_value = calc_deleted_bytes_number(&s_cut_balance, mode);
1786 	else
1787 		ret_value = retval2;
1788 
1789 	/*
1790 	 * For direct items, we only change the quota when deleting the last
1791 	 * item.
1792 	 */
1793 	p_le_ih = tp_item_head(s_cut_balance.tb_path);
1794 	if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1795 		if (mode == M_DELETE &&
1796 		    (le_ih_k_offset(p_le_ih) & (sb->s_blocksize - 1)) ==
1797 		    1) {
1798 			/* FIXME: this is to keep 3.5 happy */
1799 			REISERFS_I(inode)->i_first_direct_byte = U32_MAX;
1800 			quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
1801 		} else {
1802 			quota_cut_bytes = 0;
1803 		}
1804 	}
1805 #ifdef CONFIG_REISERFS_CHECK
1806 	if (is_inode_locked) {
1807 		struct item_head *le_ih =
1808 		    tp_item_head(s_cut_balance.tb_path);
1809 		/*
1810 		 * we are going to complete indirect2direct conversion. Make
1811 		 * sure, that we exactly remove last unformatted node pointer
1812 		 * of the item
1813 		 */
1814 		if (!is_indirect_le_ih(le_ih))
1815 			reiserfs_panic(sb, "vs-5652",
1816 				       "item must be indirect %h", le_ih);
1817 
1818 		if (mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1819 			reiserfs_panic(sb, "vs-5653", "completing "
1820 				       "indirect2direct conversion indirect "
1821 				       "item %h being deleted must be of "
1822 				       "4 byte long", le_ih);
1823 
1824 		if (mode == M_CUT
1825 		    && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1826 			reiserfs_panic(sb, "vs-5654", "can not complete "
1827 				       "indirect2direct conversion of %h "
1828 				       "(CUT, insert_size==%d)",
1829 				       le_ih, s_cut_balance.insert_size[0]);
1830 		}
1831 		/*
1832 		 * it would be useful to make sure, that right neighboring
1833 		 * item is direct item of this file
1834 		 */
1835 	}
1836 #endif
1837 
1838 	do_balance(&s_cut_balance, NULL, NULL, mode);
1839 	if (is_inode_locked) {
1840 		/*
1841 		 * we've done an indirect->direct conversion.  when the
1842 		 * data block was freed, it was removed from the list of
1843 		 * blocks that must be flushed before the transaction
1844 		 * commits, make sure to unmap and invalidate it
1845 		 */
1846 		unmap_buffers(page, tail_pos);
1847 		REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
1848 	}
1849 #ifdef REISERQUOTA_DEBUG
1850 	reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1851 		       "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1852 		       quota_cut_bytes, inode->i_uid, '?');
1853 #endif
1854 	depth = reiserfs_write_unlock_nested(sb);
1855 	dquot_free_space_nodirty(inode, quota_cut_bytes);
1856 	reiserfs_write_lock_nested(sb, depth);
1857 	return ret_value;
1858 }
1859 
1860 static void truncate_directory(struct reiserfs_transaction_handle *th,
1861 			       struct inode *inode)
1862 {
1863 	BUG_ON(!th->t_trans_id);
1864 	if (inode->i_nlink)
1865 		reiserfs_error(inode->i_sb, "vs-5655", "link count != 0");
1866 
1867 	set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1868 	set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1869 	reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1870 	reiserfs_update_sd(th, inode);
1871 	set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1872 	set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
1873 }
1874 
1875 /*
1876  * Truncate file to the new size. Note, this must be called with a
1877  * transaction already started
1878  */
1879 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th,
1880 			 struct inode *inode,	/* ->i_size contains new size */
1881 			 struct page *page,	/* up to date for last block */
1882 			 /*
1883 			  * when it is called by file_release to convert
1884 			  * the tail - no timestamps should be updated
1885 			  */
1886 			 int update_timestamps
1887     )
1888 {
1889 	INITIALIZE_PATH(s_search_path);	/* Path to the current object item. */
1890 	struct item_head *p_le_ih;	/* Pointer to an item header. */
1891 
1892 	/* Key to search for a previous file item. */
1893 	struct cpu_key s_item_key;
1894 	loff_t file_size,	/* Old file size. */
1895 	 new_file_size;	/* New file size. */
1896 	int deleted;		/* Number of deleted or truncated bytes. */
1897 	int retval;
1898 	int err = 0;
1899 
1900 	BUG_ON(!th->t_trans_id);
1901 	if (!
1902 	    (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1903 	     || S_ISLNK(inode->i_mode)))
1904 		return 0;
1905 
1906 	/* deletion of directory - no need to update timestamps */
1907 	if (S_ISDIR(inode->i_mode)) {
1908 		truncate_directory(th, inode);
1909 		return 0;
1910 	}
1911 
1912 	/* Get new file size. */
1913 	new_file_size = inode->i_size;
1914 
1915 	/* FIXME: note, that key type is unimportant here */
1916 	make_cpu_key(&s_item_key, inode, max_reiserfs_offset(inode),
1917 		     TYPE_DIRECT, 3);
1918 
1919 	retval =
1920 	    search_for_position_by_key(inode->i_sb, &s_item_key,
1921 				       &s_search_path);
1922 	if (retval == IO_ERROR) {
1923 		reiserfs_error(inode->i_sb, "vs-5657",
1924 			       "i/o failure occurred trying to truncate %K",
1925 			       &s_item_key);
1926 		err = -EIO;
1927 		goto out;
1928 	}
1929 	if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1930 		reiserfs_error(inode->i_sb, "PAP-5660",
1931 			       "wrong result %d of search for %K", retval,
1932 			       &s_item_key);
1933 
1934 		err = -EIO;
1935 		goto out;
1936 	}
1937 
1938 	s_search_path.pos_in_item--;
1939 
1940 	/* Get real file size (total length of all file items) */
1941 	p_le_ih = tp_item_head(&s_search_path);
1942 	if (is_statdata_le_ih(p_le_ih))
1943 		file_size = 0;
1944 	else {
1945 		loff_t offset = le_ih_k_offset(p_le_ih);
1946 		int bytes =
1947 		    op_bytes_number(p_le_ih, inode->i_sb->s_blocksize);
1948 
1949 		/*
1950 		 * this may mismatch with real file size: if last direct item
1951 		 * had no padding zeros and last unformatted node had no free
1952 		 * space, this file would have this file size
1953 		 */
1954 		file_size = offset + bytes - 1;
1955 	}
1956 	/*
1957 	 * are we doing a full truncate or delete, if so
1958 	 * kick in the reada code
1959 	 */
1960 	if (new_file_size == 0)
1961 		s_search_path.reada = PATH_READA | PATH_READA_BACK;
1962 
1963 	if (file_size == 0 || file_size < new_file_size) {
1964 		goto update_and_out;
1965 	}
1966 
1967 	/* Update key to search for the last file item. */
1968 	set_cpu_key_k_offset(&s_item_key, file_size);
1969 
1970 	do {
1971 		/* Cut or delete file item. */
1972 		deleted =
1973 		    reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
1974 					   inode, page, new_file_size);
1975 		if (deleted < 0) {
1976 			reiserfs_warning(inode->i_sb, "vs-5665",
1977 					 "reiserfs_cut_from_item failed");
1978 			reiserfs_check_path(&s_search_path);
1979 			return 0;
1980 		}
1981 
1982 		RFALSE(deleted > file_size,
1983 		       "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1984 		       deleted, file_size, &s_item_key);
1985 
1986 		/* Change key to search the last file item. */
1987 		file_size -= deleted;
1988 
1989 		set_cpu_key_k_offset(&s_item_key, file_size);
1990 
1991 		/*
1992 		 * While there are bytes to truncate and previous
1993 		 * file item is presented in the tree.
1994 		 */
1995 
1996 		/*
1997 		 * This loop could take a really long time, and could log
1998 		 * many more blocks than a transaction can hold.  So, we do
1999 		 * a polite journal end here, and if the transaction needs
2000 		 * ending, we make sure the file is consistent before ending
2001 		 * the current trans and starting a new one
2002 		 */
2003 		if (journal_transaction_should_end(th, 0) ||
2004 		    reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
2005 			pathrelse(&s_search_path);
2006 
2007 			if (update_timestamps) {
2008 				inode->i_mtime = current_time(inode);
2009 				inode->i_ctime = current_time(inode);
2010 			}
2011 			reiserfs_update_sd(th, inode);
2012 
2013 			err = journal_end(th);
2014 			if (err)
2015 				goto out;
2016 			err = journal_begin(th, inode->i_sb,
2017 					    JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ;
2018 			if (err)
2019 				goto out;
2020 			reiserfs_update_inode_transaction(inode);
2021 		}
2022 	} while (file_size > ROUND_UP(new_file_size) &&
2023 		 search_for_position_by_key(inode->i_sb, &s_item_key,
2024 					    &s_search_path) == POSITION_FOUND);
2025 
2026 	RFALSE(file_size > ROUND_UP(new_file_size),
2027 	       "PAP-5680: truncate did not finish: new_file_size %lld, current %lld, oid %d",
2028 	       new_file_size, file_size, s_item_key.on_disk_key.k_objectid);
2029 
2030 update_and_out:
2031 	if (update_timestamps) {
2032 		/* this is truncate, not file closing */
2033 		inode->i_mtime = current_time(inode);
2034 		inode->i_ctime = current_time(inode);
2035 	}
2036 	reiserfs_update_sd(th, inode);
2037 
2038 out:
2039 	pathrelse(&s_search_path);
2040 	return err;
2041 }
2042 
2043 #ifdef CONFIG_REISERFS_CHECK
2044 /* this makes sure, that we __append__, not overwrite or add holes */
2045 static void check_research_for_paste(struct treepath *path,
2046 				     const struct cpu_key *key)
2047 {
2048 	struct item_head *found_ih = tp_item_head(path);
2049 
2050 	if (is_direct_le_ih(found_ih)) {
2051 		if (le_ih_k_offset(found_ih) +
2052 		    op_bytes_number(found_ih,
2053 				    get_last_bh(path)->b_size) !=
2054 		    cpu_key_k_offset(key)
2055 		    || op_bytes_number(found_ih,
2056 				       get_last_bh(path)->b_size) !=
2057 		    pos_in_item(path))
2058 			reiserfs_panic(NULL, "PAP-5720", "found direct item "
2059 				       "%h or position (%d) does not match "
2060 				       "to key %K", found_ih,
2061 				       pos_in_item(path), key);
2062 	}
2063 	if (is_indirect_le_ih(found_ih)) {
2064 		if (le_ih_k_offset(found_ih) +
2065 		    op_bytes_number(found_ih,
2066 				    get_last_bh(path)->b_size) !=
2067 		    cpu_key_k_offset(key)
2068 		    || I_UNFM_NUM(found_ih) != pos_in_item(path)
2069 		    || get_ih_free_space(found_ih) != 0)
2070 			reiserfs_panic(NULL, "PAP-5730", "found indirect "
2071 				       "item (%h) or position (%d) does not "
2072 				       "match to key (%K)",
2073 				       found_ih, pos_in_item(path), key);
2074 	}
2075 }
2076 #endif				/* config reiserfs check */
2077 
2078 /*
2079  * Paste bytes to the existing item.
2080  * Returns bytes number pasted into the item.
2081  */
2082 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th,
2083 			     /* Path to the pasted item. */
2084 			     struct treepath *search_path,
2085 			     /* Key to search for the needed item. */
2086 			     const struct cpu_key *key,
2087 			     /* Inode item belongs to */
2088 			     struct inode *inode,
2089 			     /* Pointer to the bytes to paste. */
2090 			     const char *body,
2091 			     /* Size of pasted bytes. */
2092 			     int pasted_size)
2093 {
2094 	struct super_block *sb = inode->i_sb;
2095 	struct tree_balance s_paste_balance;
2096 	int retval;
2097 	int fs_gen;
2098 	int depth;
2099 
2100 	BUG_ON(!th->t_trans_id);
2101 
2102 	fs_gen = get_generation(inode->i_sb);
2103 
2104 #ifdef REISERQUOTA_DEBUG
2105 	reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2106 		       "reiserquota paste_into_item(): allocating %u id=%u type=%c",
2107 		       pasted_size, inode->i_uid,
2108 		       key2type(&key->on_disk_key));
2109 #endif
2110 
2111 	depth = reiserfs_write_unlock_nested(sb);
2112 	retval = dquot_alloc_space_nodirty(inode, pasted_size);
2113 	reiserfs_write_lock_nested(sb, depth);
2114 	if (retval) {
2115 		pathrelse(search_path);
2116 		return retval;
2117 	}
2118 	init_tb_struct(th, &s_paste_balance, th->t_super, search_path,
2119 		       pasted_size);
2120 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2121 	s_paste_balance.key = key->on_disk_key;
2122 #endif
2123 
2124 	/* DQUOT_* can schedule, must check before the fix_nodes */
2125 	if (fs_changed(fs_gen, inode->i_sb)) {
2126 		goto search_again;
2127 	}
2128 
2129 	while ((retval =
2130 		fix_nodes(M_PASTE, &s_paste_balance, NULL,
2131 			  body)) == REPEAT_SEARCH) {
2132 search_again:
2133 		/* file system changed while we were in the fix_nodes */
2134 		PROC_INFO_INC(th->t_super, paste_into_item_restarted);
2135 		retval =
2136 		    search_for_position_by_key(th->t_super, key,
2137 					       search_path);
2138 		if (retval == IO_ERROR) {
2139 			retval = -EIO;
2140 			goto error_out;
2141 		}
2142 		if (retval == POSITION_FOUND) {
2143 			reiserfs_warning(inode->i_sb, "PAP-5710",
2144 					 "entry or pasted byte (%K) exists",
2145 					 key);
2146 			retval = -EEXIST;
2147 			goto error_out;
2148 		}
2149 #ifdef CONFIG_REISERFS_CHECK
2150 		check_research_for_paste(search_path, key);
2151 #endif
2152 	}
2153 
2154 	/*
2155 	 * Perform balancing after all resources are collected by fix_nodes,
2156 	 * and accessing them will not risk triggering schedule.
2157 	 */
2158 	if (retval == CARRY_ON) {
2159 		do_balance(&s_paste_balance, NULL /*ih */ , body, M_PASTE);
2160 		return 0;
2161 	}
2162 	retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2163 error_out:
2164 	/* this also releases the path */
2165 	unfix_nodes(&s_paste_balance);
2166 #ifdef REISERQUOTA_DEBUG
2167 	reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2168 		       "reiserquota paste_into_item(): freeing %u id=%u type=%c",
2169 		       pasted_size, inode->i_uid,
2170 		       key2type(&key->on_disk_key));
2171 #endif
2172 	depth = reiserfs_write_unlock_nested(sb);
2173 	dquot_free_space_nodirty(inode, pasted_size);
2174 	reiserfs_write_lock_nested(sb, depth);
2175 	return retval;
2176 }
2177 
2178 /*
2179  * Insert new item into the buffer at the path.
2180  * th   - active transaction handle
2181  * path - path to the inserted item
2182  * ih   - pointer to the item header to insert
2183  * body - pointer to the bytes to insert
2184  */
2185 int reiserfs_insert_item(struct reiserfs_transaction_handle *th,
2186 			 struct treepath *path, const struct cpu_key *key,
2187 			 struct item_head *ih, struct inode *inode,
2188 			 const char *body)
2189 {
2190 	struct tree_balance s_ins_balance;
2191 	int retval;
2192 	int fs_gen = 0;
2193 	int quota_bytes = 0;
2194 
2195 	BUG_ON(!th->t_trans_id);
2196 
2197 	if (inode) {		/* Do we count quotas for item? */
2198 		int depth;
2199 		fs_gen = get_generation(inode->i_sb);
2200 		quota_bytes = ih_item_len(ih);
2201 
2202 		/*
2203 		 * hack so the quota code doesn't have to guess
2204 		 * if the file has a tail, links are always tails,
2205 		 * so there's no guessing needed
2206 		 */
2207 		if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(ih))
2208 			quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
2209 #ifdef REISERQUOTA_DEBUG
2210 		reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2211 			       "reiserquota insert_item(): allocating %u id=%u type=%c",
2212 			       quota_bytes, inode->i_uid, head2type(ih));
2213 #endif
2214 		/*
2215 		 * We can't dirty inode here. It would be immediately
2216 		 * written but appropriate stat item isn't inserted yet...
2217 		 */
2218 		depth = reiserfs_write_unlock_nested(inode->i_sb);
2219 		retval = dquot_alloc_space_nodirty(inode, quota_bytes);
2220 		reiserfs_write_lock_nested(inode->i_sb, depth);
2221 		if (retval) {
2222 			pathrelse(path);
2223 			return retval;
2224 		}
2225 	}
2226 	init_tb_struct(th, &s_ins_balance, th->t_super, path,
2227 		       IH_SIZE + ih_item_len(ih));
2228 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2229 	s_ins_balance.key = key->on_disk_key;
2230 #endif
2231 	/*
2232 	 * DQUOT_* can schedule, must check to be sure calling
2233 	 * fix_nodes is safe
2234 	 */
2235 	if (inode && fs_changed(fs_gen, inode->i_sb)) {
2236 		goto search_again;
2237 	}
2238 
2239 	while ((retval =
2240 		fix_nodes(M_INSERT, &s_ins_balance, ih,
2241 			  body)) == REPEAT_SEARCH) {
2242 search_again:
2243 		/* file system changed while we were in the fix_nodes */
2244 		PROC_INFO_INC(th->t_super, insert_item_restarted);
2245 		retval = search_item(th->t_super, key, path);
2246 		if (retval == IO_ERROR) {
2247 			retval = -EIO;
2248 			goto error_out;
2249 		}
2250 		if (retval == ITEM_FOUND) {
2251 			reiserfs_warning(th->t_super, "PAP-5760",
2252 					 "key %K already exists in the tree",
2253 					 key);
2254 			retval = -EEXIST;
2255 			goto error_out;
2256 		}
2257 	}
2258 
2259 	/* make balancing after all resources will be collected at a time */
2260 	if (retval == CARRY_ON) {
2261 		do_balance(&s_ins_balance, ih, body, M_INSERT);
2262 		return 0;
2263 	}
2264 
2265 	retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2266 error_out:
2267 	/* also releases the path */
2268 	unfix_nodes(&s_ins_balance);
2269 #ifdef REISERQUOTA_DEBUG
2270 	if (inode)
2271 		reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2272 		       "reiserquota insert_item(): freeing %u id=%u type=%c",
2273 		       quota_bytes, inode->i_uid, head2type(ih));
2274 #endif
2275 	if (inode) {
2276 		int depth = reiserfs_write_unlock_nested(inode->i_sb);
2277 		dquot_free_space_nodirty(inode, quota_bytes);
2278 		reiserfs_write_lock_nested(inode->i_sb, depth);
2279 	}
2280 	return retval;
2281 }
2282