kill-the-BKL/reiserfs: release write lock while rescheduling on prepare_for_delete_or...
[linux-2.6.git] / fs / reiserfs / stree.c
1 /*
2  *  Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3  */
4
5 /*
6  *  Written by Anatoly P. Pinchuk pap@namesys.botik.ru
7  *  Programm System Institute
8  *  Pereslavl-Zalessky Russia
9  */
10
11 /*
12  *  This file contains functions dealing with S+tree
13  *
14  * B_IS_IN_TREE
15  * copy_item_head
16  * comp_short_keys
17  * comp_keys
18  * comp_short_le_keys
19  * le_key2cpu_key
20  * comp_le_keys
21  * bin_search
22  * get_lkey
23  * get_rkey
24  * key_in_buffer
25  * decrement_bcount
26  * reiserfs_check_path
27  * pathrelse_and_restore
28  * pathrelse
29  * search_by_key_reada
30  * search_by_key
31  * search_for_position_by_key
32  * comp_items
33  * prepare_for_direct_item
34  * prepare_for_direntry_item
35  * prepare_for_delete_or_cut
36  * calc_deleted_bytes_number
37  * init_tb_struct
38  * padd_item
39  * reiserfs_delete_item
40  * reiserfs_delete_solid_item
41  * reiserfs_delete_object
42  * maybe_indirect_to_direct
43  * indirect_to_direct_roll_back
44  * reiserfs_cut_from_item
45  * truncate_directory
46  * reiserfs_do_truncate
47  * reiserfs_paste_into_item
48  * reiserfs_insert_item
49  */
50
51 #include <linux/time.h>
52 #include <linux/string.h>
53 #include <linux/pagemap.h>
54 #include <linux/reiserfs_fs.h>
55 #include <linux/buffer_head.h>
56 #include <linux/quotaops.h>
57
58 /* Does the buffer contain a disk block which is in the tree. */
59 inline int B_IS_IN_TREE(const struct buffer_head *bh)
60 {
61
62         RFALSE(B_LEVEL(bh) > MAX_HEIGHT,
63                "PAP-1010: block (%b) has too big level (%z)", bh, bh);
64
65         return (B_LEVEL(bh) != FREE_LEVEL);
66 }
67
68 //
69 // to gets item head in le form
70 //
71 inline void copy_item_head(struct item_head *to,
72                            const struct item_head *from)
73 {
74         memcpy(to, from, IH_SIZE);
75 }
76
77 /* k1 is pointer to on-disk structure which is stored in little-endian
78    form. k2 is pointer to cpu variable. For key of items of the same
79    object this returns 0.
80    Returns: -1 if key1 < key2
81    0 if key1 == key2
82    1 if key1 > key2 */
83 inline int comp_short_keys(const struct reiserfs_key *le_key,
84                            const struct cpu_key *cpu_key)
85 {
86         __u32 n;
87         n = le32_to_cpu(le_key->k_dir_id);
88         if (n < cpu_key->on_disk_key.k_dir_id)
89                 return -1;
90         if (n > cpu_key->on_disk_key.k_dir_id)
91                 return 1;
92         n = le32_to_cpu(le_key->k_objectid);
93         if (n < cpu_key->on_disk_key.k_objectid)
94                 return -1;
95         if (n > cpu_key->on_disk_key.k_objectid)
96                 return 1;
97         return 0;
98 }
99
100 /* k1 is pointer to on-disk structure which is stored in little-endian
101    form. k2 is pointer to cpu variable.
102    Compare keys using all 4 key fields.
103    Returns: -1 if key1 < key2 0
104    if key1 = key2 1 if key1 > key2 */
105 static inline int comp_keys(const struct reiserfs_key *le_key,
106                             const struct cpu_key *cpu_key)
107 {
108         int retval;
109
110         retval = comp_short_keys(le_key, cpu_key);
111         if (retval)
112                 return retval;
113         if (le_key_k_offset(le_key_version(le_key), le_key) <
114             cpu_key_k_offset(cpu_key))
115                 return -1;
116         if (le_key_k_offset(le_key_version(le_key), le_key) >
117             cpu_key_k_offset(cpu_key))
118                 return 1;
119
120         if (cpu_key->key_length == 3)
121                 return 0;
122
123         /* this part is needed only when tail conversion is in progress */
124         if (le_key_k_type(le_key_version(le_key), le_key) <
125             cpu_key_k_type(cpu_key))
126                 return -1;
127
128         if (le_key_k_type(le_key_version(le_key), le_key) >
129             cpu_key_k_type(cpu_key))
130                 return 1;
131
132         return 0;
133 }
134
135 inline int comp_short_le_keys(const struct reiserfs_key *key1,
136                               const struct reiserfs_key *key2)
137 {
138         __u32 *k1_u32, *k2_u32;
139         int key_length = REISERFS_SHORT_KEY_LEN;
140
141         k1_u32 = (__u32 *) key1;
142         k2_u32 = (__u32 *) key2;
143         for (; key_length--; ++k1_u32, ++k2_u32) {
144                 if (le32_to_cpu(*k1_u32) < le32_to_cpu(*k2_u32))
145                         return -1;
146                 if (le32_to_cpu(*k1_u32) > le32_to_cpu(*k2_u32))
147                         return 1;
148         }
149         return 0;
150 }
151
152 inline void le_key2cpu_key(struct cpu_key *to, const struct reiserfs_key *from)
153 {
154         int version;
155         to->on_disk_key.k_dir_id = le32_to_cpu(from->k_dir_id);
156         to->on_disk_key.k_objectid = le32_to_cpu(from->k_objectid);
157
158         // find out version of the key
159         version = le_key_version(from);
160         to->version = version;
161         to->on_disk_key.k_offset = le_key_k_offset(version, from);
162         to->on_disk_key.k_type = le_key_k_type(version, from);
163 }
164
165 // this does not say which one is bigger, it only returns 1 if keys
166 // are not equal, 0 otherwise
167 inline int comp_le_keys(const struct reiserfs_key *k1,
168                         const struct reiserfs_key *k2)
169 {
170         return memcmp(k1, k2, sizeof(struct reiserfs_key));
171 }
172
173 /**************************************************************************
174  *  Binary search toolkit function                                        *
175  *  Search for an item in the array by the item key                       *
176  *  Returns:    1 if found,  0 if not found;                              *
177  *        *pos = number of the searched element if found, else the        *
178  *        number of the first element that is larger than key.            *
179  **************************************************************************/
180 /* For those not familiar with binary search: lbound is the leftmost item that it
181  could be, rbound the rightmost item that it could be.  We examine the item
182  halfway between lbound and rbound, and that tells us either that we can increase
183  lbound, or decrease rbound, or that we have found it, or if lbound <= rbound that
184  there are no possible items, and we have not found it. With each examination we
185  cut the number of possible items it could be by one more than half rounded down,
186  or we find it. */
187 static inline int bin_search(const void *key,   /* Key to search for. */
188                              const void *base,  /* First item in the array. */
189                              int num,   /* Number of items in the array. */
190                              int width, /* Item size in the array.
191                                            searched. Lest the reader be
192                                            confused, note that this is crafted
193                                            as a general function, and when it
194                                            is applied specifically to the array
195                                            of item headers in a node, width
196                                            is actually the item header size not
197                                            the item size. */
198                              int *pos /* Number of the searched for element. */
199     )
200 {
201         int rbound, lbound, j;
202
203         for (j = ((rbound = num - 1) + (lbound = 0)) / 2;
204              lbound <= rbound; j = (rbound + lbound) / 2)
205                 switch (comp_keys
206                         ((struct reiserfs_key *)((char *)base + j * width),
207                          (struct cpu_key *)key)) {
208                 case -1:
209                         lbound = j + 1;
210                         continue;
211                 case 1:
212                         rbound = j - 1;
213                         continue;
214                 case 0:
215                         *pos = j;
216                         return ITEM_FOUND;      /* Key found in the array.  */
217                 }
218
219         /* bin_search did not find given key, it returns position of key,
220            that is minimal and greater than the given one. */
221         *pos = lbound;
222         return ITEM_NOT_FOUND;
223 }
224
225 #ifdef CONFIG_REISERFS_CHECK
226 extern struct tree_balance *cur_tb;
227 #endif
228
229 /* Minimal possible key. It is never in the tree. */
230 const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} };
231
232 /* Maximal possible key. It is never in the tree. */
233 static const struct reiserfs_key MAX_KEY = {
234         __constant_cpu_to_le32(0xffffffff),
235         __constant_cpu_to_le32(0xffffffff),
236         {{__constant_cpu_to_le32(0xffffffff),
237           __constant_cpu_to_le32(0xffffffff)},}
238 };
239
240 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
241    of the path, and going upwards.  We must check the path's validity at each step.  If the key is not in
242    the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
243    case we return a special key, either MIN_KEY or MAX_KEY. */
244 static inline const struct reiserfs_key *get_lkey(const struct treepath *chk_path,
245                                                   const struct super_block *sb)
246 {
247         int position, path_offset = chk_path->path_length;
248         struct buffer_head *parent;
249
250         RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET,
251                "PAP-5010: invalid offset in the path");
252
253         /* While not higher in path than first element. */
254         while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
255
256                 RFALSE(!buffer_uptodate
257                        (PATH_OFFSET_PBUFFER(chk_path, path_offset)),
258                        "PAP-5020: parent is not uptodate");
259
260                 /* Parent at the path is not in the tree now. */
261                 if (!B_IS_IN_TREE
262                     (parent =
263                      PATH_OFFSET_PBUFFER(chk_path, path_offset)))
264                         return &MAX_KEY;
265                 /* Check whether position in the parent is correct. */
266                 if ((position =
267                      PATH_OFFSET_POSITION(chk_path,
268                                           path_offset)) >
269                     B_NR_ITEMS(parent))
270                         return &MAX_KEY;
271                 /* Check whether parent at the path really points to the child. */
272                 if (B_N_CHILD_NUM(parent, position) !=
273                     PATH_OFFSET_PBUFFER(chk_path,
274                                         path_offset + 1)->b_blocknr)
275                         return &MAX_KEY;
276                 /* Return delimiting key if position in the parent is not equal to zero. */
277                 if (position)
278                         return B_N_PDELIM_KEY(parent, position - 1);
279         }
280         /* Return MIN_KEY if we are in the root of the buffer tree. */
281         if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)->
282             b_blocknr == SB_ROOT_BLOCK(sb))
283                 return &MIN_KEY;
284         return &MAX_KEY;
285 }
286
287 /* Get delimiting key of the buffer at the path and its right neighbor. */
288 inline const struct reiserfs_key *get_rkey(const struct treepath *chk_path,
289                                            const struct super_block *sb)
290 {
291         int position, path_offset = chk_path->path_length;
292         struct buffer_head *parent;
293
294         RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET,
295                "PAP-5030: invalid offset in the path");
296
297         while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
298
299                 RFALSE(!buffer_uptodate
300                        (PATH_OFFSET_PBUFFER(chk_path, path_offset)),
301                        "PAP-5040: parent is not uptodate");
302
303                 /* Parent at the path is not in the tree now. */
304                 if (!B_IS_IN_TREE
305                     (parent =
306                      PATH_OFFSET_PBUFFER(chk_path, path_offset)))
307                         return &MIN_KEY;
308                 /* Check whether position in the parent is correct. */
309                 if ((position =
310                      PATH_OFFSET_POSITION(chk_path,
311                                           path_offset)) >
312                     B_NR_ITEMS(parent))
313                         return &MIN_KEY;
314                 /* Check whether parent at the path really points to the child. */
315                 if (B_N_CHILD_NUM(parent, position) !=
316                     PATH_OFFSET_PBUFFER(chk_path,
317                                         path_offset + 1)->b_blocknr)
318                         return &MIN_KEY;
319                 /* Return delimiting key if position in the parent is not the last one. */
320                 if (position != B_NR_ITEMS(parent))
321                         return B_N_PDELIM_KEY(parent, position);
322         }
323         /* Return MAX_KEY if we are in the root of the buffer tree. */
324         if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)->
325             b_blocknr == SB_ROOT_BLOCK(sb))
326                 return &MAX_KEY;
327         return &MIN_KEY;
328 }
329
330 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
331 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
332    the path.  These delimiting keys are stored at least one level above that buffer in the tree. If the
333    buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
334    this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
335 static inline int key_in_buffer(struct treepath *chk_path,      /* Path which should be checked.  */
336                                 const struct cpu_key *key,      /* Key which should be checked.   */
337                                 struct super_block *sb
338     )
339 {
340
341         RFALSE(!key || chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET
342                || chk_path->path_length > MAX_HEIGHT,
343                "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
344                key, chk_path->path_length);
345         RFALSE(!PATH_PLAST_BUFFER(chk_path)->b_bdev,
346                "PAP-5060: device must not be NODEV");
347
348         if (comp_keys(get_lkey(chk_path, sb), key) == 1)
349                 /* left delimiting key is bigger, that the key we look for */
350                 return 0;
351         /*  if ( comp_keys(key, get_rkey(chk_path, sb)) != -1 ) */
352         if (comp_keys(get_rkey(chk_path, sb), key) != 1)
353                 /* key must be less than right delimitiing key */
354                 return 0;
355         return 1;
356 }
357
358 int reiserfs_check_path(struct treepath *p)
359 {
360         RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
361                "path not properly relsed");
362         return 0;
363 }
364
365 /* Drop the reference to each buffer in a path and restore
366  * dirty bits clean when preparing the buffer for the log.
367  * This version should only be called from fix_nodes() */
368 void pathrelse_and_restore(struct super_block *sb,
369                            struct treepath *search_path)
370 {
371         int path_offset = search_path->path_length;
372
373         RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
374                "clm-4000: invalid path offset");
375
376         while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
377                 struct buffer_head *bh;
378                 bh = PATH_OFFSET_PBUFFER(search_path, path_offset--);
379                 reiserfs_restore_prepared_buffer(sb, bh);
380                 brelse(bh);
381         }
382         search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
383 }
384
385 /* Drop the reference to each buffer in a path */
386 void pathrelse(struct treepath *search_path)
387 {
388         int path_offset = search_path->path_length;
389
390         RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
391                "PAP-5090: invalid path offset");
392
393         while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
394                 brelse(PATH_OFFSET_PBUFFER(search_path, path_offset--));
395
396         search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
397 }
398
399 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
400 {
401         struct block_head *blkh;
402         struct item_head *ih;
403         int used_space;
404         int prev_location;
405         int i;
406         int nr;
407
408         blkh = (struct block_head *)buf;
409         if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
410                 reiserfs_warning(NULL, "reiserfs-5080",
411                                  "this should be caught earlier");
412                 return 0;
413         }
414
415         nr = blkh_nr_item(blkh);
416         if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
417                 /* item number is too big or too small */
418                 reiserfs_warning(NULL, "reiserfs-5081",
419                                  "nr_item seems wrong: %z", bh);
420                 return 0;
421         }
422         ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
423         used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih));
424         if (used_space != blocksize - blkh_free_space(blkh)) {
425                 /* free space does not match to calculated amount of use space */
426                 reiserfs_warning(NULL, "reiserfs-5082",
427                                  "free space seems wrong: %z", bh);
428                 return 0;
429         }
430         // FIXME: it is_leaf will hit performance too much - we may have
431         // return 1 here
432
433         /* check tables of item heads */
434         ih = (struct item_head *)(buf + BLKH_SIZE);
435         prev_location = blocksize;
436         for (i = 0; i < nr; i++, ih++) {
437                 if (le_ih_k_type(ih) == TYPE_ANY) {
438                         reiserfs_warning(NULL, "reiserfs-5083",
439                                          "wrong item type for item %h",
440                                          ih);
441                         return 0;
442                 }
443                 if (ih_location(ih) >= blocksize
444                     || ih_location(ih) < IH_SIZE * nr) {
445                         reiserfs_warning(NULL, "reiserfs-5084",
446                                          "item location seems wrong: %h",
447                                          ih);
448                         return 0;
449                 }
450                 if (ih_item_len(ih) < 1
451                     || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
452                         reiserfs_warning(NULL, "reiserfs-5085",
453                                          "item length seems wrong: %h",
454                                          ih);
455                         return 0;
456                 }
457                 if (prev_location - ih_location(ih) != ih_item_len(ih)) {
458                         reiserfs_warning(NULL, "reiserfs-5086",
459                                          "item location seems wrong "
460                                          "(second one): %h", ih);
461                         return 0;
462                 }
463                 prev_location = ih_location(ih);
464         }
465
466         // one may imagine much more checks
467         return 1;
468 }
469
470 /* returns 1 if buf looks like an internal node, 0 otherwise */
471 static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
472 {
473         struct block_head *blkh;
474         int nr;
475         int used_space;
476
477         blkh = (struct block_head *)buf;
478         nr = blkh_level(blkh);
479         if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
480                 /* this level is not possible for internal nodes */
481                 reiserfs_warning(NULL, "reiserfs-5087",
482                                  "this should be caught earlier");
483                 return 0;
484         }
485
486         nr = blkh_nr_item(blkh);
487         if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
488                 /* for internal which is not root we might check min number of keys */
489                 reiserfs_warning(NULL, "reiserfs-5088",
490                                  "number of key seems wrong: %z", bh);
491                 return 0;
492         }
493
494         used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
495         if (used_space != blocksize - blkh_free_space(blkh)) {
496                 reiserfs_warning(NULL, "reiserfs-5089",
497                                  "free space seems wrong: %z", bh);
498                 return 0;
499         }
500         // one may imagine much more checks
501         return 1;
502 }
503
504 // make sure that bh contains formatted node of reiserfs tree of
505 // 'level'-th level
506 static int is_tree_node(struct buffer_head *bh, int level)
507 {
508         if (B_LEVEL(bh) != level) {
509                 reiserfs_warning(NULL, "reiserfs-5090", "node level %d does "
510                                  "not match to the expected one %d",
511                                  B_LEVEL(bh), level);
512                 return 0;
513         }
514         if (level == DISK_LEAF_NODE_LEVEL)
515                 return is_leaf(bh->b_data, bh->b_size, bh);
516
517         return is_internal(bh->b_data, bh->b_size, bh);
518 }
519
520 #define SEARCH_BY_KEY_READA 16
521
522 /* The function is NOT SCHEDULE-SAFE! */
523 static void search_by_key_reada(struct super_block *s,
524                                 struct buffer_head **bh,
525                                 b_blocknr_t *b, int num)
526 {
527         int i, j;
528
529         for (i = 0; i < num; i++) {
530                 bh[i] = sb_getblk(s, b[i]);
531         }
532         for (j = 0; j < i; j++) {
533                 /*
534                  * note, this needs attention if we are getting rid of the BKL
535                  * you have to make sure the prepared bit isn't set on this buffer
536                  */
537                 if (!buffer_uptodate(bh[j]))
538                         ll_rw_block(READA, 1, bh + j);
539                 brelse(bh[j]);
540         }
541 }
542
543 /**************************************************************************
544  * Algorithm   SearchByKey                                                *
545  *             look for item in the Disk S+Tree by its key                *
546  * Input:  sb   -  super block                                            *
547  *         key  - pointer to the key to search                            *
548  * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR                         *
549  *         search_path - path from the root to the needed leaf            *
550  **************************************************************************/
551
552 /* This function fills up the path from the root to the leaf as it
553    descends the tree looking for the key.  It uses reiserfs_bread to
554    try to find buffers in the cache given their block number.  If it
555    does not find them in the cache it reads them from disk.  For each
556    node search_by_key finds using reiserfs_bread it then uses
557    bin_search to look through that node.  bin_search will find the
558    position of the block_number of the next node if it is looking
559    through an internal node.  If it is looking through a leaf node
560    bin_search will find the position of the item which has key either
561    equal to given key, or which is the maximal key less than the given
562    key.  search_by_key returns a path that must be checked for the
563    correctness of the top of the path but need not be checked for the
564    correctness of the bottom of the path */
565 /* The function is NOT SCHEDULE-SAFE! */
566 int search_by_key(struct super_block *sb, const struct cpu_key *key,    /* Key to search. */
567                   struct treepath *search_path,/* This structure was
568                                                    allocated and initialized
569                                                    by the calling
570                                                    function. It is filled up
571                                                    by this function.  */
572                   int stop_level        /* How far down the tree to search. To
573                                            stop at leaf level - set to
574                                            DISK_LEAF_NODE_LEVEL */
575     )
576 {
577         b_blocknr_t block_number;
578         int expected_level;
579         struct buffer_head *bh;
580         struct path_element *last_element;
581         int node_level, retval;
582         int right_neighbor_of_leaf_node;
583         int fs_gen;
584         struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
585         b_blocknr_t reada_blocks[SEARCH_BY_KEY_READA];
586         int reada_count = 0;
587
588 #ifdef CONFIG_REISERFS_CHECK
589         int repeat_counter = 0;
590 #endif
591
592         PROC_INFO_INC(sb, search_by_key);
593
594         /* As we add each node to a path we increase its count.  This means that
595            we must be careful to release all nodes in a path before we either
596            discard the path struct or re-use the path struct, as we do here. */
597
598         pathrelse(search_path);
599
600         right_neighbor_of_leaf_node = 0;
601
602         /* With each iteration of this loop we search through the items in the
603            current node, and calculate the next current node(next path element)
604            for the next iteration of this loop.. */
605         block_number = SB_ROOT_BLOCK(sb);
606         expected_level = -1;
607         while (1) {
608
609 #ifdef CONFIG_REISERFS_CHECK
610                 if (!(++repeat_counter % 50000))
611                         reiserfs_warning(sb, "PAP-5100",
612                                          "%s: there were %d iterations of "
613                                          "while loop looking for key %K",
614                                          current->comm, repeat_counter,
615                                          key);
616 #endif
617
618                 /* prep path to have another element added to it. */
619                 last_element =
620                     PATH_OFFSET_PELEMENT(search_path,
621                                          ++search_path->path_length);
622                 fs_gen = get_generation(sb);
623
624                 /* Read the next tree node, and set the last element in the path to
625                    have a pointer to it. */
626                 if ((bh = last_element->pe_buffer =
627                      sb_getblk(sb, block_number))) {
628                         if (!buffer_uptodate(bh) && reada_count > 1)
629                                 search_by_key_reada(sb, reada_bh,
630                                                     reada_blocks, reada_count);
631                         ll_rw_block(READ, 1, &bh);
632                         reiserfs_write_unlock(sb);
633                         wait_on_buffer(bh);
634                         reiserfs_write_lock(sb);
635                         if (!buffer_uptodate(bh))
636                                 goto io_error;
637                 } else {
638                       io_error:
639                         search_path->path_length--;
640                         pathrelse(search_path);
641                         return IO_ERROR;
642                 }
643                 reada_count = 0;
644                 if (expected_level == -1)
645                         expected_level = SB_TREE_HEIGHT(sb);
646                 expected_level--;
647
648                 /* It is possible that schedule occurred. We must check whether the key
649                    to search is still in the tree rooted from the current buffer. If
650                    not then repeat search from the root. */
651                 if (fs_changed(fs_gen, sb) &&
652                     (!B_IS_IN_TREE(bh) ||
653                      B_LEVEL(bh) != expected_level ||
654                      !key_in_buffer(search_path, key, sb))) {
655                         PROC_INFO_INC(sb, search_by_key_fs_changed);
656                         PROC_INFO_INC(sb, search_by_key_restarted);
657                         PROC_INFO_INC(sb,
658                                       sbk_restarted[expected_level - 1]);
659                         pathrelse(search_path);
660
661                         /* Get the root block number so that we can repeat the search
662                            starting from the root. */
663                         block_number = SB_ROOT_BLOCK(sb);
664                         expected_level = -1;
665                         right_neighbor_of_leaf_node = 0;
666
667                         /* repeat search from the root */
668                         continue;
669                 }
670
671                 /* only check that the key is in the buffer if key is not
672                    equal to the MAX_KEY. Latter case is only possible in
673                    "finish_unfinished()" processing during mount. */
674                 RFALSE(comp_keys(&MAX_KEY, key) &&
675                        !key_in_buffer(search_path, key, sb),
676                        "PAP-5130: key is not in the buffer");
677 #ifdef CONFIG_REISERFS_CHECK
678                 if (cur_tb) {
679                         print_cur_tb("5140");
680                         reiserfs_panic(sb, "PAP-5140",
681                                        "schedule occurred in do_balance!");
682                 }
683 #endif
684
685                 // make sure, that the node contents look like a node of
686                 // certain level
687                 if (!is_tree_node(bh, expected_level)) {
688                         reiserfs_error(sb, "vs-5150",
689                                        "invalid format found in block %ld. "
690                                        "Fsck?", bh->b_blocknr);
691                         pathrelse(search_path);
692                         return IO_ERROR;
693                 }
694
695                 /* ok, we have acquired next formatted node in the tree */
696                 node_level = B_LEVEL(bh);
697
698                 PROC_INFO_BH_STAT(sb, bh, node_level - 1);
699
700                 RFALSE(node_level < stop_level,
701                        "vs-5152: tree level (%d) is less than stop level (%d)",
702                        node_level, stop_level);
703
704                 retval = bin_search(key, B_N_PITEM_HEAD(bh, 0),
705                                       B_NR_ITEMS(bh),
706                                       (node_level ==
707                                        DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
708                                       KEY_SIZE,
709                                       &(last_element->pe_position));
710                 if (node_level == stop_level) {
711                         return retval;
712                 }
713
714                 /* we are not in the stop level */
715                 if (retval == ITEM_FOUND)
716                         /* item has been found, so we choose the pointer which is to the right of the found one */
717                         last_element->pe_position++;
718
719                 /* if item was not found we choose the position which is to
720                    the left of the found item. This requires no code,
721                    bin_search did it already. */
722
723                 /* So we have chosen a position in the current node which is
724                    an internal node.  Now we calculate child block number by
725                    position in the node. */
726                 block_number =
727                     B_N_CHILD_NUM(bh, last_element->pe_position);
728
729                 /* if we are going to read leaf nodes, try for read ahead as well */
730                 if ((search_path->reada & PATH_READA) &&
731                     node_level == DISK_LEAF_NODE_LEVEL + 1) {
732                         int pos = last_element->pe_position;
733                         int limit = B_NR_ITEMS(bh);
734                         struct reiserfs_key *le_key;
735
736                         if (search_path->reada & PATH_READA_BACK)
737                                 limit = 0;
738                         while (reada_count < SEARCH_BY_KEY_READA) {
739                                 if (pos == limit)
740                                         break;
741                                 reada_blocks[reada_count++] =
742                                     B_N_CHILD_NUM(bh, pos);
743                                 if (search_path->reada & PATH_READA_BACK)
744                                         pos--;
745                                 else
746                                         pos++;
747
748                                 /*
749                                  * check to make sure we're in the same object
750                                  */
751                                 le_key = B_N_PDELIM_KEY(bh, pos);
752                                 if (le32_to_cpu(le_key->k_objectid) !=
753                                     key->on_disk_key.k_objectid) {
754                                         break;
755                                 }
756                         }
757                 }
758         }
759 }
760
761 /* Form the path to an item and position in this item which contains
762    file byte defined by key. If there is no such item
763    corresponding to the key, we point the path to the item with
764    maximal key less than key, and *pos_in_item is set to one
765    past the last entry/byte in the item.  If searching for entry in a
766    directory item, and it is not found, *pos_in_item is set to one
767    entry more than the entry with maximal key which is less than the
768    sought key.
769
770    Note that if there is no entry in this same node which is one more,
771    then we point to an imaginary entry.  for direct items, the
772    position is in units of bytes, for indirect items the position is
773    in units of blocknr entries, for directory items the position is in
774    units of directory entries.  */
775
776 /* The function is NOT SCHEDULE-SAFE! */
777 int search_for_position_by_key(struct super_block *sb,  /* Pointer to the super block.          */
778                                const struct cpu_key *p_cpu_key, /* Key to search (cpu variable)         */
779                                struct treepath *search_path     /* Filled up by this function.          */
780     )
781 {
782         struct item_head *p_le_ih;      /* pointer to on-disk structure */
783         int blk_size;
784         loff_t item_offset, offset;
785         struct reiserfs_dir_entry de;
786         int retval;
787
788         /* If searching for directory entry. */
789         if (is_direntry_cpu_key(p_cpu_key))
790                 return search_by_entry_key(sb, p_cpu_key, search_path,
791                                            &de);
792
793         /* If not searching for directory entry. */
794
795         /* If item is found. */
796         retval = search_item(sb, p_cpu_key, search_path);
797         if (retval == IO_ERROR)
798                 return retval;
799         if (retval == ITEM_FOUND) {
800
801                 RFALSE(!ih_item_len
802                        (B_N_PITEM_HEAD
803                         (PATH_PLAST_BUFFER(search_path),
804                          PATH_LAST_POSITION(search_path))),
805                        "PAP-5165: item length equals zero");
806
807                 pos_in_item(search_path) = 0;
808                 return POSITION_FOUND;
809         }
810
811         RFALSE(!PATH_LAST_POSITION(search_path),
812                "PAP-5170: position equals zero");
813
814         /* Item is not found. Set path to the previous item. */
815         p_le_ih =
816             B_N_PITEM_HEAD(PATH_PLAST_BUFFER(search_path),
817                            --PATH_LAST_POSITION(search_path));
818         blk_size = sb->s_blocksize;
819
820         if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) {
821                 return FILE_NOT_FOUND;
822         }
823         // FIXME: quite ugly this far
824
825         item_offset = le_ih_k_offset(p_le_ih);
826         offset = cpu_key_k_offset(p_cpu_key);
827
828         /* Needed byte is contained in the item pointed to by the path. */
829         if (item_offset <= offset &&
830             item_offset + op_bytes_number(p_le_ih, blk_size) > offset) {
831                 pos_in_item(search_path) = offset - item_offset;
832                 if (is_indirect_le_ih(p_le_ih)) {
833                         pos_in_item(search_path) /= blk_size;
834                 }
835                 return POSITION_FOUND;
836         }
837
838         /* Needed byte is not contained in the item pointed to by the
839            path. Set pos_in_item out of the item. */
840         if (is_indirect_le_ih(p_le_ih))
841                 pos_in_item(search_path) =
842                     ih_item_len(p_le_ih) / UNFM_P_SIZE;
843         else
844                 pos_in_item(search_path) = ih_item_len(p_le_ih);
845
846         return POSITION_NOT_FOUND;
847 }
848
849 /* Compare given item and item pointed to by the path. */
850 int comp_items(const struct item_head *stored_ih, const struct treepath *path)
851 {
852         struct buffer_head *bh = PATH_PLAST_BUFFER(path);
853         struct item_head *ih;
854
855         /* Last buffer at the path is not in the tree. */
856         if (!B_IS_IN_TREE(bh))
857                 return 1;
858
859         /* Last path position is invalid. */
860         if (PATH_LAST_POSITION(path) >= B_NR_ITEMS(bh))
861                 return 1;
862
863         /* we need only to know, whether it is the same item */
864         ih = get_ih(path);
865         return memcmp(stored_ih, ih, IH_SIZE);
866 }
867
868 /* unformatted nodes are not logged anymore, ever.  This is safe
869 ** now
870 */
871 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
872
873 // block can not be forgotten as it is in I/O or held by someone
874 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
875
876 // prepare for delete or cut of direct item
877 static inline int prepare_for_direct_item(struct treepath *path,
878                                           struct item_head *le_ih,
879                                           struct inode *inode,
880                                           loff_t new_file_length, int *cut_size)
881 {
882         loff_t round_len;
883
884         if (new_file_length == max_reiserfs_offset(inode)) {
885                 /* item has to be deleted */
886                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
887                 return M_DELETE;
888         }
889         // new file gets truncated
890         if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
891                 //
892                 round_len = ROUND_UP(new_file_length);
893                 /* this was new_file_length < le_ih ... */
894                 if (round_len < le_ih_k_offset(le_ih)) {
895                         *cut_size = -(IH_SIZE + ih_item_len(le_ih));
896                         return M_DELETE;        /* Delete this item. */
897                 }
898                 /* Calculate first position and size for cutting from item. */
899                 pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
900                 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
901
902                 return M_CUT;   /* Cut from this item. */
903         }
904
905         // old file: items may have any length
906
907         if (new_file_length < le_ih_k_offset(le_ih)) {
908                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
909                 return M_DELETE;        /* Delete this item. */
910         }
911         /* Calculate first position and size for cutting from item. */
912         *cut_size = -(ih_item_len(le_ih) -
913                       (pos_in_item(path) =
914                        new_file_length + 1 - le_ih_k_offset(le_ih)));
915         return M_CUT;           /* Cut from this item. */
916 }
917
918 static inline int prepare_for_direntry_item(struct treepath *path,
919                                             struct item_head *le_ih,
920                                             struct inode *inode,
921                                             loff_t new_file_length,
922                                             int *cut_size)
923 {
924         if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
925             new_file_length == max_reiserfs_offset(inode)) {
926                 RFALSE(ih_entry_count(le_ih) != 2,
927                        "PAP-5220: incorrect empty directory item (%h)", le_ih);
928                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
929                 return M_DELETE;        /* Delete the directory item containing "." and ".." entry. */
930         }
931
932         if (ih_entry_count(le_ih) == 1) {
933                 /* Delete the directory item such as there is one record only
934                    in this item */
935                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
936                 return M_DELETE;
937         }
938
939         /* Cut one record from the directory item. */
940         *cut_size =
941             -(DEH_SIZE +
942               entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
943         return M_CUT;
944 }
945
946 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
947
948 /*  If the path points to a directory or direct item, calculate mode and the size cut, for balance.
949     If the path points to an indirect item, remove some number of its unformatted nodes.
950     In case of file truncate calculate whether this item must be deleted/truncated or last
951     unformatted node of this item will be converted to a direct item.
952     This function returns a determination of what balance mode the calling function should employ. */
953 static char prepare_for_delete_or_cut(struct reiserfs_transaction_handle *th, struct inode *inode, struct treepath *path, const struct cpu_key *item_key, int *removed, /* Number of unformatted nodes which were removed
954                                                                                                                                                                                    from end of the file. */
955                                       int *cut_size, unsigned long long new_file_length /* MAX_KEY_OFFSET in case of delete. */
956     )
957 {
958         struct super_block *sb = inode->i_sb;
959         struct item_head *p_le_ih = PATH_PITEM_HEAD(path);
960         struct buffer_head *bh = PATH_PLAST_BUFFER(path);
961
962         BUG_ON(!th->t_trans_id);
963
964         /* Stat_data item. */
965         if (is_statdata_le_ih(p_le_ih)) {
966
967                 RFALSE(new_file_length != max_reiserfs_offset(inode),
968                        "PAP-5210: mode must be M_DELETE");
969
970                 *cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
971                 return M_DELETE;
972         }
973
974         /* Directory item. */
975         if (is_direntry_le_ih(p_le_ih))
976                 return prepare_for_direntry_item(path, p_le_ih, inode,
977                                                  new_file_length,
978                                                  cut_size);
979
980         /* Direct item. */
981         if (is_direct_le_ih(p_le_ih))
982                 return prepare_for_direct_item(path, p_le_ih, inode,
983                                                new_file_length, cut_size);
984
985         /* Case of an indirect item. */
986         {
987             int blk_size = sb->s_blocksize;
988             struct item_head s_ih;
989             int need_re_search;
990             int delete = 0;
991             int result = M_CUT;
992             int pos = 0;
993
994             if ( new_file_length == max_reiserfs_offset (inode) ) {
995                 /* prepare_for_delete_or_cut() is called by
996                  * reiserfs_delete_item() */
997                 new_file_length = 0;
998                 delete = 1;
999             }
1000
1001             do {
1002                 need_re_search = 0;
1003                 *cut_size = 0;
1004                 bh = PATH_PLAST_BUFFER(path);
1005                 copy_item_head(&s_ih, PATH_PITEM_HEAD(path));
1006                 pos = I_UNFM_NUM(&s_ih);
1007
1008                 while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > new_file_length) {
1009                     __le32 *unfm;
1010                     __u32 block;
1011
1012                     /* Each unformatted block deletion may involve one additional
1013                      * bitmap block into the transaction, thereby the initial
1014                      * journal space reservation might not be enough. */
1015                     if (!delete && (*cut_size) != 0 &&
1016                         reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD)
1017                         break;
1018
1019                     unfm = (__le32 *)B_I_PITEM(bh, &s_ih) + pos - 1;
1020                     block = get_block_num(unfm, 0);
1021
1022                     if (block != 0) {
1023                         reiserfs_prepare_for_journal(sb, bh, 1);
1024                         put_block_num(unfm, 0, 0);
1025                         journal_mark_dirty(th, sb, bh);
1026                         reiserfs_free_block(th, inode, block, 1);
1027                     }
1028
1029                     reiserfs_write_unlock(sb);
1030                     cond_resched();
1031                     reiserfs_write_lock(sb);
1032
1033                     if (item_moved (&s_ih, path))  {
1034                         need_re_search = 1;
1035                         break;
1036                     }
1037
1038                     pos --;
1039                     (*removed)++;
1040                     (*cut_size) -= UNFM_P_SIZE;
1041
1042                     if (pos == 0) {
1043                         (*cut_size) -= IH_SIZE;
1044                         result = M_DELETE;
1045                         break;
1046                     }
1047                 }
1048                 /* a trick.  If the buffer has been logged, this will do nothing.  If
1049                 ** we've broken the loop without logging it, it will restore the
1050                 ** buffer */
1051                 reiserfs_restore_prepared_buffer(sb, bh);
1052             } while (need_re_search &&
1053                      search_for_position_by_key(sb, item_key, path) == POSITION_FOUND);
1054             pos_in_item(path) = pos * UNFM_P_SIZE;
1055
1056             if (*cut_size == 0) {
1057                 /* Nothing were cut. maybe convert last unformatted node to the
1058                  * direct item? */
1059                 result = M_CONVERT;
1060             }
1061             return result;
1062         }
1063 }
1064
1065 /* Calculate number of bytes which will be deleted or cut during balance */
1066 static int calc_deleted_bytes_number(struct tree_balance *tb, char mode)
1067 {
1068         int del_size;
1069         struct item_head *p_le_ih = PATH_PITEM_HEAD(tb->tb_path);
1070
1071         if (is_statdata_le_ih(p_le_ih))
1072                 return 0;
1073
1074         del_size =
1075             (mode ==
1076              M_DELETE) ? ih_item_len(p_le_ih) : -tb->insert_size[0];
1077         if (is_direntry_le_ih(p_le_ih)) {
1078                 /* return EMPTY_DIR_SIZE; We delete emty directoris only.
1079                  * we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1080                  * empty size.  ick. FIXME, is this right? */
1081                 return del_size;
1082         }
1083
1084         if (is_indirect_le_ih(p_le_ih))
1085                 del_size = (del_size / UNFM_P_SIZE) *
1086                                 (PATH_PLAST_BUFFER(tb->tb_path)->b_size);
1087         return del_size;
1088 }
1089
1090 static void init_tb_struct(struct reiserfs_transaction_handle *th,
1091                            struct tree_balance *tb,
1092                            struct super_block *sb,
1093                            struct treepath *path, int size)
1094 {
1095
1096         BUG_ON(!th->t_trans_id);
1097
1098         memset(tb, '\0', sizeof(struct tree_balance));
1099         tb->transaction_handle = th;
1100         tb->tb_sb = sb;
1101         tb->tb_path = path;
1102         PATH_OFFSET_PBUFFER(path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1103         PATH_OFFSET_POSITION(path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1104         tb->insert_size[0] = size;
1105 }
1106
1107 void padd_item(char *item, int total_length, int length)
1108 {
1109         int i;
1110
1111         for (i = total_length; i > length;)
1112                 item[--i] = 0;
1113 }
1114
1115 #ifdef REISERQUOTA_DEBUG
1116 char key2type(struct reiserfs_key *ih)
1117 {
1118         if (is_direntry_le_key(2, ih))
1119                 return 'd';
1120         if (is_direct_le_key(2, ih))
1121                 return 'D';
1122         if (is_indirect_le_key(2, ih))
1123                 return 'i';
1124         if (is_statdata_le_key(2, ih))
1125                 return 's';
1126         return 'u';
1127 }
1128
1129 char head2type(struct item_head *ih)
1130 {
1131         if (is_direntry_le_ih(ih))
1132                 return 'd';
1133         if (is_direct_le_ih(ih))
1134                 return 'D';
1135         if (is_indirect_le_ih(ih))
1136                 return 'i';
1137         if (is_statdata_le_ih(ih))
1138                 return 's';
1139         return 'u';
1140 }
1141 #endif
1142
1143 /* Delete object item.
1144  * th       - active transaction handle
1145  * path     - path to the deleted item
1146  * item_key - key to search for the deleted item
1147  * indode   - used for updating i_blocks and quotas
1148  * un_bh    - NULL or unformatted node pointer
1149  */
1150 int reiserfs_delete_item(struct reiserfs_transaction_handle *th,
1151                          struct treepath *path, const struct cpu_key *item_key,
1152                          struct inode *inode, struct buffer_head *un_bh)
1153 {
1154         struct super_block *sb = inode->i_sb;
1155         struct tree_balance s_del_balance;
1156         struct item_head s_ih;
1157         struct item_head *q_ih;
1158         int quota_cut_bytes;
1159         int ret_value, del_size, removed;
1160
1161 #ifdef CONFIG_REISERFS_CHECK
1162         char mode;
1163         int iter = 0;
1164 #endif
1165
1166         BUG_ON(!th->t_trans_id);
1167
1168         init_tb_struct(th, &s_del_balance, sb, path,
1169                        0 /*size is unknown */ );
1170
1171         while (1) {
1172                 removed = 0;
1173
1174 #ifdef CONFIG_REISERFS_CHECK
1175                 iter++;
1176                 mode =
1177 #endif
1178                     prepare_for_delete_or_cut(th, inode, path,
1179                                               item_key, &removed,
1180                                               &del_size,
1181                                               max_reiserfs_offset(inode));
1182
1183                 RFALSE(mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1184
1185                 copy_item_head(&s_ih, PATH_PITEM_HEAD(path));
1186                 s_del_balance.insert_size[0] = del_size;
1187
1188                 ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1189                 if (ret_value != REPEAT_SEARCH)
1190                         break;
1191
1192                 PROC_INFO_INC(sb, delete_item_restarted);
1193
1194                 // file system changed, repeat search
1195                 ret_value =
1196                     search_for_position_by_key(sb, item_key, path);
1197                 if (ret_value == IO_ERROR)
1198                         break;
1199                 if (ret_value == FILE_NOT_FOUND) {
1200                         reiserfs_warning(sb, "vs-5340",
1201                                          "no items of the file %K found",
1202                                          item_key);
1203                         break;
1204                 }
1205         }                       /* while (1) */
1206
1207         if (ret_value != CARRY_ON) {
1208                 unfix_nodes(&s_del_balance);
1209                 return 0;
1210         }
1211         // reiserfs_delete_item returns item length when success
1212         ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1213         q_ih = get_ih(path);
1214         quota_cut_bytes = ih_item_len(q_ih);
1215
1216         /* hack so the quota code doesn't have to guess if the file
1217          ** has a tail.  On tail insert, we allocate quota for 1 unformatted node.
1218          ** We test the offset because the tail might have been
1219          ** split into multiple items, and we only want to decrement for
1220          ** the unfm node once
1221          */
1222         if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(q_ih)) {
1223                 if ((le_ih_k_offset(q_ih) & (sb->s_blocksize - 1)) == 1) {
1224                         quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
1225                 } else {
1226                         quota_cut_bytes = 0;
1227                 }
1228         }
1229
1230         if (un_bh) {
1231                 int off;
1232                 char *data;
1233
1234                 /* We are in direct2indirect conversion, so move tail contents
1235                    to the unformatted node */
1236                 /* note, we do the copy before preparing the buffer because we
1237                  ** don't care about the contents of the unformatted node yet.
1238                  ** the only thing we really care about is the direct item's data
1239                  ** is in the unformatted node.
1240                  **
1241                  ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1242                  ** the unformatted node, which might schedule, meaning we'd have to
1243                  ** loop all the way back up to the start of the while loop.
1244                  **
1245                  ** The unformatted node must be dirtied later on.  We can't be
1246                  ** sure here if the entire tail has been deleted yet.
1247                  **
1248                  ** un_bh is from the page cache (all unformatted nodes are
1249                  ** from the page cache) and might be a highmem page.  So, we
1250                  ** can't use un_bh->b_data.
1251                  ** -clm
1252                  */
1253
1254                 data = kmap_atomic(un_bh->b_page, KM_USER0);
1255                 off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1256                 memcpy(data + off,
1257                        B_I_PITEM(PATH_PLAST_BUFFER(path), &s_ih),
1258                        ret_value);
1259                 kunmap_atomic(data, KM_USER0);
1260         }
1261         /* Perform balancing after all resources have been collected at once. */
1262         do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1263
1264 #ifdef REISERQUOTA_DEBUG
1265         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
1266                        "reiserquota delete_item(): freeing %u, id=%u type=%c",
1267                        quota_cut_bytes, inode->i_uid, head2type(&s_ih));
1268 #endif
1269         vfs_dq_free_space_nodirty(inode, quota_cut_bytes);
1270
1271         /* Return deleted body length */
1272         return ret_value;
1273 }
1274
1275 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1276
1277  deletion of the body of the object is performed by iput(), with the
1278  result that if multiple processes are operating on a file, the
1279  deletion of the body of the file is deferred until the last process
1280  that has an open inode performs its iput().
1281
1282  writes and truncates are protected from collisions by use of
1283  semaphores.
1284
1285  creates, linking, and mknod are protected from collisions with other
1286  processes by making the reiserfs_add_entry() the last step in the
1287  creation, and then rolling back all changes if there was a collision.
1288  - Hans
1289 */
1290
1291 /* this deletes item which never gets split */
1292 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1293                                 struct inode *inode, struct reiserfs_key *key)
1294 {
1295         struct tree_balance tb;
1296         INITIALIZE_PATH(path);
1297         int item_len = 0;
1298         int tb_init = 0;
1299         struct cpu_key cpu_key;
1300         int retval;
1301         int quota_cut_bytes = 0;
1302
1303         BUG_ON(!th->t_trans_id);
1304
1305         le_key2cpu_key(&cpu_key, key);
1306
1307         while (1) {
1308                 retval = search_item(th->t_super, &cpu_key, &path);
1309                 if (retval == IO_ERROR) {
1310                         reiserfs_error(th->t_super, "vs-5350",
1311                                        "i/o failure occurred trying "
1312                                        "to delete %K", &cpu_key);
1313                         break;
1314                 }
1315                 if (retval != ITEM_FOUND) {
1316                         pathrelse(&path);
1317                         // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1318                         if (!
1319                             ((unsigned long long)
1320                              GET_HASH_VALUE(le_key_k_offset
1321                                             (le_key_version(key), key)) == 0
1322                              && (unsigned long long)
1323                              GET_GENERATION_NUMBER(le_key_k_offset
1324                                                    (le_key_version(key),
1325                                                     key)) == 1))
1326                                 reiserfs_warning(th->t_super, "vs-5355",
1327                                                  "%k not found", key);
1328                         break;
1329                 }
1330                 if (!tb_init) {
1331                         tb_init = 1;
1332                         item_len = ih_item_len(PATH_PITEM_HEAD(&path));
1333                         init_tb_struct(th, &tb, th->t_super, &path,
1334                                        -(IH_SIZE + item_len));
1335                 }
1336                 quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path));
1337
1338                 retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1339                 if (retval == REPEAT_SEARCH) {
1340                         PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1341                         continue;
1342                 }
1343
1344                 if (retval == CARRY_ON) {
1345                         do_balance(&tb, NULL, NULL, M_DELETE);
1346                         if (inode) {    /* Should we count quota for item? (we don't count quotas for save-links) */
1347 #ifdef REISERQUOTA_DEBUG
1348                                 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1349                                                "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1350                                                quota_cut_bytes, inode->i_uid,
1351                                                key2type(key));
1352 #endif
1353                                 vfs_dq_free_space_nodirty(inode,
1354                                                          quota_cut_bytes);
1355                         }
1356                         break;
1357                 }
1358                 // IO_ERROR, NO_DISK_SPACE, etc
1359                 reiserfs_warning(th->t_super, "vs-5360",
1360                                  "could not delete %K due to fix_nodes failure",
1361                                  &cpu_key);
1362                 unfix_nodes(&tb);
1363                 break;
1364         }
1365
1366         reiserfs_check_path(&path);
1367 }
1368
1369 int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1370                            struct inode *inode)
1371 {
1372         int err;
1373         inode->i_size = 0;
1374         BUG_ON(!th->t_trans_id);
1375
1376         /* for directory this deletes item containing "." and ".." */
1377         err =
1378             reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1379         if (err)
1380                 return err;
1381
1382 #if defined( USE_INODE_GENERATION_COUNTER )
1383         if (!old_format_only(th->t_super)) {
1384                 __le32 *inode_generation;
1385
1386                 inode_generation =
1387                     &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
1388                 le32_add_cpu(inode_generation, 1);
1389         }
1390 /* USE_INODE_GENERATION_COUNTER */
1391 #endif
1392         reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1393
1394         return err;
1395 }
1396
1397 static void unmap_buffers(struct page *page, loff_t pos)
1398 {
1399         struct buffer_head *bh;
1400         struct buffer_head *head;
1401         struct buffer_head *next;
1402         unsigned long tail_index;
1403         unsigned long cur_index;
1404
1405         if (page) {
1406                 if (page_has_buffers(page)) {
1407                         tail_index = pos & (PAGE_CACHE_SIZE - 1);
1408                         cur_index = 0;
1409                         head = page_buffers(page);
1410                         bh = head;
1411                         do {
1412                                 next = bh->b_this_page;
1413
1414                                 /* we want to unmap the buffers that contain the tail, and
1415                                  ** all the buffers after it (since the tail must be at the
1416                                  ** end of the file).  We don't want to unmap file data
1417                                  ** before the tail, since it might be dirty and waiting to
1418                                  ** reach disk
1419                                  */
1420                                 cur_index += bh->b_size;
1421                                 if (cur_index > tail_index) {
1422                                         reiserfs_unmap_buffer(bh);
1423                                 }
1424                                 bh = next;
1425                         } while (bh != head);
1426                 }
1427         }
1428 }
1429
1430 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
1431                                     struct inode *inode,
1432                                     struct page *page,
1433                                     struct treepath *path,
1434                                     const struct cpu_key *item_key,
1435                                     loff_t new_file_size, char *mode)
1436 {
1437         struct super_block *sb = inode->i_sb;
1438         int block_size = sb->s_blocksize;
1439         int cut_bytes;
1440         BUG_ON(!th->t_trans_id);
1441         BUG_ON(new_file_size != inode->i_size);
1442
1443         /* the page being sent in could be NULL if there was an i/o error
1444          ** reading in the last block.  The user will hit problems trying to
1445          ** read the file, but for now we just skip the indirect2direct
1446          */
1447         if (atomic_read(&inode->i_count) > 1 ||
1448             !tail_has_to_be_packed(inode) ||
1449             !page || (REISERFS_I(inode)->i_flags & i_nopack_mask)) {
1450                 /* leave tail in an unformatted node */
1451                 *mode = M_SKIP_BALANCING;
1452                 cut_bytes =
1453                     block_size - (new_file_size & (block_size - 1));
1454                 pathrelse(path);
1455                 return cut_bytes;
1456         }
1457         /* Perform the conversion to a direct_item. */
1458         /* return indirect_to_direct(inode, path, item_key,
1459                                   new_file_size, mode); */
1460         return indirect2direct(th, inode, page, path, item_key,
1461                                new_file_size, mode);
1462 }
1463
1464 /* we did indirect_to_direct conversion. And we have inserted direct
1465    item successesfully, but there were no disk space to cut unfm
1466    pointer being converted. Therefore we have to delete inserted
1467    direct item(s) */
1468 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
1469                                          struct inode *inode, struct treepath *path)
1470 {
1471         struct cpu_key tail_key;
1472         int tail_len;
1473         int removed;
1474         BUG_ON(!th->t_trans_id);
1475
1476         make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);      // !!!!
1477         tail_key.key_length = 4;
1478
1479         tail_len =
1480             (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1481         while (tail_len) {
1482                 /* look for the last byte of the tail */
1483                 if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1484                     POSITION_NOT_FOUND)
1485                         reiserfs_panic(inode->i_sb, "vs-5615",
1486                                        "found invalid item");
1487                 RFALSE(path->pos_in_item !=
1488                        ih_item_len(PATH_PITEM_HEAD(path)) - 1,
1489                        "vs-5616: appended bytes found");
1490                 PATH_LAST_POSITION(path)--;
1491
1492                 removed =
1493                     reiserfs_delete_item(th, path, &tail_key, inode,
1494                                          NULL /*unbh not needed */ );
1495                 RFALSE(removed <= 0
1496                        || removed > tail_len,
1497                        "vs-5617: there was tail %d bytes, removed item length %d bytes",
1498                        tail_len, removed);
1499                 tail_len -= removed;
1500                 set_cpu_key_k_offset(&tail_key,
1501                                      cpu_key_k_offset(&tail_key) - removed);
1502         }
1503         reiserfs_warning(inode->i_sb, "reiserfs-5091", "indirect_to_direct "
1504                          "conversion has been rolled back due to "
1505                          "lack of disk space");
1506         //mark_file_without_tail (inode);
1507         mark_inode_dirty(inode);
1508 }
1509
1510 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1511 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
1512                            struct treepath *path,
1513                            struct cpu_key *item_key,
1514                            struct inode *inode,
1515                            struct page *page, loff_t new_file_size)
1516 {
1517         struct super_block *sb = inode->i_sb;
1518         /* Every function which is going to call do_balance must first
1519            create a tree_balance structure.  Then it must fill up this
1520            structure by using the init_tb_struct and fix_nodes functions.
1521            After that we can make tree balancing. */
1522         struct tree_balance s_cut_balance;
1523         struct item_head *p_le_ih;
1524         int cut_size = 0,       /* Amount to be cut. */
1525             ret_value = CARRY_ON, removed = 0,  /* Number of the removed unformatted nodes. */
1526             is_inode_locked = 0;
1527         char mode;              /* Mode of the balance. */
1528         int retval2 = -1;
1529         int quota_cut_bytes;
1530         loff_t tail_pos = 0;
1531
1532         BUG_ON(!th->t_trans_id);
1533
1534         init_tb_struct(th, &s_cut_balance, inode->i_sb, path,
1535                        cut_size);
1536
1537         /* Repeat this loop until we either cut the item without needing
1538            to balance, or we fix_nodes without schedule occurring */
1539         while (1) {
1540                 /* Determine the balance mode, position of the first byte to
1541                    be cut, and size to be cut.  In case of the indirect item
1542                    free unformatted nodes which are pointed to by the cut
1543                    pointers. */
1544
1545                 mode =
1546                     prepare_for_delete_or_cut(th, inode, path,
1547                                               item_key, &removed,
1548                                               &cut_size, new_file_size);
1549                 if (mode == M_CONVERT) {
1550                         /* convert last unformatted node to direct item or leave
1551                            tail in the unformatted node */
1552                         RFALSE(ret_value != CARRY_ON,
1553                                "PAP-5570: can not convert twice");
1554
1555                         ret_value =
1556                             maybe_indirect_to_direct(th, inode, page,
1557                                                      path, item_key,
1558                                                      new_file_size, &mode);
1559                         if (mode == M_SKIP_BALANCING)
1560                                 /* tail has been left in the unformatted node */
1561                                 return ret_value;
1562
1563                         is_inode_locked = 1;
1564
1565                         /* removing of last unformatted node will change value we
1566                            have to return to truncate. Save it */
1567                         retval2 = ret_value;
1568                         /*retval2 = sb->s_blocksize - (new_file_size & (sb->s_blocksize - 1)); */
1569
1570                         /* So, we have performed the first part of the conversion:
1571                            inserting the new direct item.  Now we are removing the
1572                            last unformatted node pointer. Set key to search for
1573                            it. */
1574                         set_cpu_key_k_type(item_key, TYPE_INDIRECT);
1575                         item_key->key_length = 4;
1576                         new_file_size -=
1577                             (new_file_size & (sb->s_blocksize - 1));
1578                         tail_pos = new_file_size;
1579                         set_cpu_key_k_offset(item_key, new_file_size + 1);
1580                         if (search_for_position_by_key
1581                             (sb, item_key,
1582                              path) == POSITION_NOT_FOUND) {
1583                                 print_block(PATH_PLAST_BUFFER(path), 3,
1584                                             PATH_LAST_POSITION(path) - 1,
1585                                             PATH_LAST_POSITION(path) + 1);
1586                                 reiserfs_panic(sb, "PAP-5580", "item to "
1587                                                "convert does not exist (%K)",
1588                                                item_key);
1589                         }
1590                         continue;
1591                 }
1592                 if (cut_size == 0) {
1593                         pathrelse(path);
1594                         return 0;
1595                 }
1596
1597                 s_cut_balance.insert_size[0] = cut_size;
1598
1599                 ret_value = fix_nodes(mode, &s_cut_balance, NULL, NULL);
1600                 if (ret_value != REPEAT_SEARCH)
1601                         break;
1602
1603                 PROC_INFO_INC(sb, cut_from_item_restarted);
1604
1605                 ret_value =
1606                     search_for_position_by_key(sb, item_key, path);
1607                 if (ret_value == POSITION_FOUND)
1608                         continue;
1609
1610                 reiserfs_warning(sb, "PAP-5610", "item %K not found",
1611                                  item_key);
1612                 unfix_nodes(&s_cut_balance);
1613                 return (ret_value == IO_ERROR) ? -EIO : -ENOENT;
1614         }                       /* while */
1615
1616         // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1617         if (ret_value != CARRY_ON) {
1618                 if (is_inode_locked) {
1619                         // FIXME: this seems to be not needed: we are always able
1620                         // to cut item
1621                         indirect_to_direct_roll_back(th, inode, path);
1622                 }
1623                 if (ret_value == NO_DISK_SPACE)
1624                         reiserfs_warning(sb, "reiserfs-5092",
1625                                          "NO_DISK_SPACE");
1626                 unfix_nodes(&s_cut_balance);
1627                 return -EIO;
1628         }
1629
1630         /* go ahead and perform balancing */
1631
1632         RFALSE(mode == M_PASTE || mode == M_INSERT, "invalid mode");
1633
1634         /* Calculate number of bytes that need to be cut from the item. */
1635         quota_cut_bytes =
1636             (mode ==
1637              M_DELETE) ? ih_item_len(get_ih(path)) : -s_cut_balance.
1638             insert_size[0];
1639         if (retval2 == -1)
1640                 ret_value = calc_deleted_bytes_number(&s_cut_balance, mode);
1641         else
1642                 ret_value = retval2;
1643
1644         /* For direct items, we only change the quota when deleting the last
1645          ** item.
1646          */
1647         p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path);
1648         if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1649                 if (mode == M_DELETE &&
1650                     (le_ih_k_offset(p_le_ih) & (sb->s_blocksize - 1)) ==
1651                     1) {
1652                         // FIXME: this is to keep 3.5 happy
1653                         REISERFS_I(inode)->i_first_direct_byte = U32_MAX;
1654                         quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
1655                 } else {
1656                         quota_cut_bytes = 0;
1657                 }
1658         }
1659 #ifdef CONFIG_REISERFS_CHECK
1660         if (is_inode_locked) {
1661                 struct item_head *le_ih =
1662                     PATH_PITEM_HEAD(s_cut_balance.tb_path);
1663                 /* we are going to complete indirect2direct conversion. Make
1664                    sure, that we exactly remove last unformatted node pointer
1665                    of the item */
1666                 if (!is_indirect_le_ih(le_ih))
1667                         reiserfs_panic(sb, "vs-5652",
1668                                        "item must be indirect %h", le_ih);
1669
1670                 if (mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1671                         reiserfs_panic(sb, "vs-5653", "completing "
1672                                        "indirect2direct conversion indirect "
1673                                        "item %h being deleted must be of "
1674                                        "4 byte long", le_ih);
1675
1676                 if (mode == M_CUT
1677                     && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1678                         reiserfs_panic(sb, "vs-5654", "can not complete "
1679                                        "indirect2direct conversion of %h "
1680                                        "(CUT, insert_size==%d)",
1681                                        le_ih, s_cut_balance.insert_size[0]);
1682                 }
1683                 /* it would be useful to make sure, that right neighboring
1684                    item is direct item of this file */
1685         }
1686 #endif
1687
1688         do_balance(&s_cut_balance, NULL, NULL, mode);
1689         if (is_inode_locked) {
1690                 /* we've done an indirect->direct conversion.  when the data block
1691                  ** was freed, it was removed from the list of blocks that must
1692                  ** be flushed before the transaction commits, make sure to
1693                  ** unmap and invalidate it
1694                  */
1695                 unmap_buffers(page, tail_pos);
1696                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
1697         }
1698 #ifdef REISERQUOTA_DEBUG
1699         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1700                        "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1701                        quota_cut_bytes, inode->i_uid, '?');
1702 #endif
1703         vfs_dq_free_space_nodirty(inode, quota_cut_bytes);
1704         return ret_value;
1705 }
1706
1707 static void truncate_directory(struct reiserfs_transaction_handle *th,
1708                                struct inode *inode)
1709 {
1710         BUG_ON(!th->t_trans_id);
1711         if (inode->i_nlink)
1712                 reiserfs_error(inode->i_sb, "vs-5655", "link count != 0");
1713
1714         set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1715         set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1716         reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1717         reiserfs_update_sd(th, inode);
1718         set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1719         set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
1720 }
1721
1722 /* Truncate file to the new size. Note, this must be called with a transaction
1723    already started */
1724 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th,
1725                           struct inode *inode,  /* ->i_size contains new size */
1726                          struct page *page,     /* up to date for last block */
1727                          int update_timestamps  /* when it is called by
1728                                                    file_release to convert
1729                                                    the tail - no timestamps
1730                                                    should be updated */
1731     )
1732 {
1733         INITIALIZE_PATH(s_search_path); /* Path to the current object item. */
1734         struct item_head *p_le_ih;      /* Pointer to an item header. */
1735         struct cpu_key s_item_key;      /* Key to search for a previous file item. */
1736         loff_t file_size,       /* Old file size. */
1737          new_file_size; /* New file size. */
1738         int deleted;            /* Number of deleted or truncated bytes. */
1739         int retval;
1740         int err = 0;
1741
1742         BUG_ON(!th->t_trans_id);
1743         if (!
1744             (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1745              || S_ISLNK(inode->i_mode)))
1746                 return 0;
1747
1748         if (S_ISDIR(inode->i_mode)) {
1749                 // deletion of directory - no need to update timestamps
1750                 truncate_directory(th, inode);
1751                 return 0;
1752         }
1753
1754         /* Get new file size. */
1755         new_file_size = inode->i_size;
1756
1757         // FIXME: note, that key type is unimportant here
1758         make_cpu_key(&s_item_key, inode, max_reiserfs_offset(inode),
1759                      TYPE_DIRECT, 3);
1760
1761         retval =
1762             search_for_position_by_key(inode->i_sb, &s_item_key,
1763                                        &s_search_path);
1764         if (retval == IO_ERROR) {
1765                 reiserfs_error(inode->i_sb, "vs-5657",
1766                                "i/o failure occurred trying to truncate %K",
1767                                &s_item_key);
1768                 err = -EIO;
1769                 goto out;
1770         }
1771         if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1772                 reiserfs_error(inode->i_sb, "PAP-5660",
1773                                "wrong result %d of search for %K", retval,
1774                                &s_item_key);
1775
1776                 err = -EIO;
1777                 goto out;
1778         }
1779
1780         s_search_path.pos_in_item--;
1781
1782         /* Get real file size (total length of all file items) */
1783         p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1784         if (is_statdata_le_ih(p_le_ih))
1785                 file_size = 0;
1786         else {
1787                 loff_t offset = le_ih_k_offset(p_le_ih);
1788                 int bytes =
1789                     op_bytes_number(p_le_ih, inode->i_sb->s_blocksize);
1790
1791                 /* this may mismatch with real file size: if last direct item
1792                    had no padding zeros and last unformatted node had no free
1793                    space, this file would have this file size */
1794                 file_size = offset + bytes - 1;
1795         }
1796         /*
1797          * are we doing a full truncate or delete, if so
1798          * kick in the reada code
1799          */
1800         if (new_file_size == 0)
1801                 s_search_path.reada = PATH_READA | PATH_READA_BACK;
1802
1803         if (file_size == 0 || file_size < new_file_size) {
1804                 goto update_and_out;
1805         }
1806
1807         /* Update key to search for the last file item. */
1808         set_cpu_key_k_offset(&s_item_key, file_size);
1809
1810         do {
1811                 /* Cut or delete file item. */
1812                 deleted =
1813                     reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
1814                                            inode, page, new_file_size);
1815                 if (deleted < 0) {
1816                         reiserfs_warning(inode->i_sb, "vs-5665",
1817                                          "reiserfs_cut_from_item failed");
1818                         reiserfs_check_path(&s_search_path);
1819                         return 0;
1820                 }
1821
1822                 RFALSE(deleted > file_size,
1823                        "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1824                        deleted, file_size, &s_item_key);
1825
1826                 /* Change key to search the last file item. */
1827                 file_size -= deleted;
1828
1829                 set_cpu_key_k_offset(&s_item_key, file_size);
1830
1831                 /* While there are bytes to truncate and previous file item is presented in the tree. */
1832
1833                 /*
1834                  ** This loop could take a really long time, and could log
1835                  ** many more blocks than a transaction can hold.  So, we do a polite
1836                  ** journal end here, and if the transaction needs ending, we make
1837                  ** sure the file is consistent before ending the current trans
1838                  ** and starting a new one
1839                  */
1840                 if (journal_transaction_should_end(th, 0) ||
1841                     reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1842                         int orig_len_alloc = th->t_blocks_allocated;
1843                         pathrelse(&s_search_path);
1844
1845                         if (update_timestamps) {
1846                                 inode->i_mtime = CURRENT_TIME_SEC;
1847                                 inode->i_ctime = CURRENT_TIME_SEC;
1848                         }
1849                         reiserfs_update_sd(th, inode);
1850
1851                         err = journal_end(th, inode->i_sb, orig_len_alloc);
1852                         if (err)
1853                                 goto out;
1854                         err = journal_begin(th, inode->i_sb,
1855                                             JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ;
1856                         if (err)
1857                                 goto out;
1858                         reiserfs_update_inode_transaction(inode);
1859                 }
1860         } while (file_size > ROUND_UP(new_file_size) &&
1861                  search_for_position_by_key(inode->i_sb, &s_item_key,
1862                                             &s_search_path) == POSITION_FOUND);
1863
1864         RFALSE(file_size > ROUND_UP(new_file_size),
1865                "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1866                new_file_size, file_size, s_item_key.on_disk_key.k_objectid);
1867
1868       update_and_out:
1869         if (update_timestamps) {
1870                 // this is truncate, not file closing
1871                 inode->i_mtime = CURRENT_TIME_SEC;
1872                 inode->i_ctime = CURRENT_TIME_SEC;
1873         }
1874         reiserfs_update_sd(th, inode);
1875
1876       out:
1877         pathrelse(&s_search_path);
1878         return err;
1879 }
1880
1881 #ifdef CONFIG_REISERFS_CHECK
1882 // this makes sure, that we __append__, not overwrite or add holes
1883 static void check_research_for_paste(struct treepath *path,
1884                                      const struct cpu_key *key)
1885 {
1886         struct item_head *found_ih = get_ih(path);
1887
1888         if (is_direct_le_ih(found_ih)) {
1889                 if (le_ih_k_offset(found_ih) +
1890                     op_bytes_number(found_ih,
1891                                     get_last_bh(path)->b_size) !=
1892                     cpu_key_k_offset(key)
1893                     || op_bytes_number(found_ih,
1894                                        get_last_bh(path)->b_size) !=
1895                     pos_in_item(path))
1896                         reiserfs_panic(NULL, "PAP-5720", "found direct item "
1897                                        "%h or position (%d) does not match "
1898                                        "to key %K", found_ih,
1899                                        pos_in_item(path), key);
1900         }
1901         if (is_indirect_le_ih(found_ih)) {
1902                 if (le_ih_k_offset(found_ih) +
1903                     op_bytes_number(found_ih,
1904                                     get_last_bh(path)->b_size) !=
1905                     cpu_key_k_offset(key)
1906                     || I_UNFM_NUM(found_ih) != pos_in_item(path)
1907                     || get_ih_free_space(found_ih) != 0)
1908                         reiserfs_panic(NULL, "PAP-5730", "found indirect "
1909                                        "item (%h) or position (%d) does not "
1910                                        "match to key (%K)",
1911                                        found_ih, pos_in_item(path), key);
1912         }
1913 }
1914 #endif                          /* config reiserfs check */
1915
1916 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1917 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct treepath *search_path,      /* Path to the pasted item.       */
1918                              const struct cpu_key *key, /* Key to search for the needed item. */
1919                              struct inode *inode,       /* Inode item belongs to */
1920                              const char *body,  /* Pointer to the bytes to paste.    */
1921                              int pasted_size)
1922 {                               /* Size of pasted bytes.             */
1923         struct tree_balance s_paste_balance;
1924         int retval;
1925         int fs_gen;
1926
1927         BUG_ON(!th->t_trans_id);
1928
1929         fs_gen = get_generation(inode->i_sb);
1930
1931 #ifdef REISERQUOTA_DEBUG
1932         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1933                        "reiserquota paste_into_item(): allocating %u id=%u type=%c",
1934                        pasted_size, inode->i_uid,
1935                        key2type(&(key->on_disk_key)));
1936 #endif
1937
1938         if (vfs_dq_alloc_space_nodirty(inode, pasted_size)) {
1939                 pathrelse(search_path);
1940                 return -EDQUOT;
1941         }
1942         init_tb_struct(th, &s_paste_balance, th->t_super, search_path,
1943                        pasted_size);
1944 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1945         s_paste_balance.key = key->on_disk_key;
1946 #endif
1947
1948         /* DQUOT_* can schedule, must check before the fix_nodes */
1949         if (fs_changed(fs_gen, inode->i_sb)) {
1950                 goto search_again;
1951         }
1952
1953         while ((retval =
1954                 fix_nodes(M_PASTE, &s_paste_balance, NULL,
1955                           body)) == REPEAT_SEARCH) {
1956               search_again:
1957                 /* file system changed while we were in the fix_nodes */
1958                 PROC_INFO_INC(th->t_super, paste_into_item_restarted);
1959                 retval =
1960                     search_for_position_by_key(th->t_super, key,
1961                                                search_path);
1962                 if (retval == IO_ERROR) {
1963                         retval = -EIO;
1964                         goto error_out;
1965                 }
1966                 if (retval == POSITION_FOUND) {
1967                         reiserfs_warning(inode->i_sb, "PAP-5710",
1968                                          "entry or pasted byte (%K) exists",
1969                                          key);
1970                         retval = -EEXIST;
1971                         goto error_out;
1972                 }
1973 #ifdef CONFIG_REISERFS_CHECK
1974                 check_research_for_paste(search_path, key);
1975 #endif
1976         }
1977
1978         /* Perform balancing after all resources are collected by fix_nodes, and
1979            accessing them will not risk triggering schedule. */
1980         if (retval == CARRY_ON) {
1981                 do_balance(&s_paste_balance, NULL /*ih */ , body, M_PASTE);
1982                 return 0;
1983         }
1984         retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
1985       error_out:
1986         /* this also releases the path */
1987         unfix_nodes(&s_paste_balance);
1988 #ifdef REISERQUOTA_DEBUG
1989         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1990                        "reiserquota paste_into_item(): freeing %u id=%u type=%c",
1991                        pasted_size, inode->i_uid,
1992                        key2type(&(key->on_disk_key)));
1993 #endif
1994         vfs_dq_free_space_nodirty(inode, pasted_size);
1995         return retval;
1996 }
1997
1998 /* Insert new item into the buffer at the path.
1999  * th   - active transaction handle
2000  * path - path to the inserted item
2001  * ih   - pointer to the item header to insert
2002  * body - pointer to the bytes to insert
2003  */
2004 int reiserfs_insert_item(struct reiserfs_transaction_handle *th,
2005                          struct treepath *path, const struct cpu_key *key,
2006                          struct item_head *ih, struct inode *inode,
2007                          const char *body)
2008 {
2009         struct tree_balance s_ins_balance;
2010         int retval;
2011         int fs_gen = 0;
2012         int quota_bytes = 0;
2013
2014         BUG_ON(!th->t_trans_id);
2015
2016         if (inode) {            /* Do we count quotas for item? */
2017                 fs_gen = get_generation(inode->i_sb);
2018                 quota_bytes = ih_item_len(ih);
2019
2020                 /* hack so the quota code doesn't have to guess if the file has
2021                  ** a tail, links are always tails, so there's no guessing needed
2022                  */
2023                 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(ih))
2024                         quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
2025 #ifdef REISERQUOTA_DEBUG
2026                 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2027                                "reiserquota insert_item(): allocating %u id=%u type=%c",
2028                                quota_bytes, inode->i_uid, head2type(ih));
2029 #endif
2030                 /* We can't dirty inode here. It would be immediately written but
2031                  * appropriate stat item isn't inserted yet... */
2032                 if (vfs_dq_alloc_space_nodirty(inode, quota_bytes)) {
2033                         pathrelse(path);
2034                         return -EDQUOT;
2035                 }
2036         }
2037         init_tb_struct(th, &s_ins_balance, th->t_super, path,
2038                        IH_SIZE + ih_item_len(ih));
2039 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2040         s_ins_balance.key = key->on_disk_key;
2041 #endif
2042         /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2043         if (inode && fs_changed(fs_gen, inode->i_sb)) {
2044                 goto search_again;
2045         }
2046
2047         while ((retval =
2048                 fix_nodes(M_INSERT, &s_ins_balance, ih,
2049                           body)) == REPEAT_SEARCH) {
2050               search_again:
2051                 /* file system changed while we were in the fix_nodes */
2052                 PROC_INFO_INC(th->t_super, insert_item_restarted);
2053                 retval = search_item(th->t_super, key, path);
2054                 if (retval == IO_ERROR) {
2055                         retval = -EIO;
2056                         goto error_out;
2057                 }
2058                 if (retval == ITEM_FOUND) {
2059                         reiserfs_warning(th->t_super, "PAP-5760",
2060                                          "key %K already exists in the tree",
2061                                          key);
2062                         retval = -EEXIST;
2063                         goto error_out;
2064                 }
2065         }
2066
2067         /* make balancing after all resources will be collected at a time */
2068         if (retval == CARRY_ON) {
2069                 do_balance(&s_ins_balance, ih, body, M_INSERT);
2070                 return 0;
2071         }
2072
2073         retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2074       error_out:
2075         /* also releases the path */
2076         unfix_nodes(&s_ins_balance);
2077 #ifdef REISERQUOTA_DEBUG
2078         reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2079                        "reiserquota insert_item(): freeing %u id=%u type=%c",
2080                        quota_bytes, inode->i_uid, head2type(ih));
2081 #endif
2082         if (inode)
2083                 vfs_dq_free_space_nodirty(inode, quota_bytes);
2084         return retval;
2085 }