Merge master.kernel.org:/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
[linux-2.6] / 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  * decrement_counters_in_path
27  * reiserfs_check_path
28  * pathrelse_and_restore
29  * pathrelse
30  * search_by_key_reada
31  * search_by_key
32  * search_for_position_by_key
33  * comp_items
34  * prepare_for_direct_item
35  * prepare_for_direntry_item
36  * prepare_for_delete_or_cut
37  * calc_deleted_bytes_number
38  * init_tb_struct
39  * padd_item
40  * reiserfs_delete_item
41  * reiserfs_delete_solid_item
42  * reiserfs_delete_object
43  * maybe_indirect_to_direct
44  * indirect_to_direct_roll_back
45  * reiserfs_cut_from_item
46  * truncate_directory
47  * reiserfs_do_truncate
48  * reiserfs_paste_into_item
49  * reiserfs_insert_item
50  */
51
52 #include <linux/time.h>
53 #include <linux/string.h>
54 #include <linux/pagemap.h>
55 #include <linux/reiserfs_fs.h>
56 #include <linux/buffer_head.h>
57 #include <linux/quotaops.h>
58
59 /* Does the buffer contain a disk block which is in the tree. */
60 inline int B_IS_IN_TREE(const struct buffer_head *p_s_bh)
61 {
62
63         RFALSE(B_LEVEL(p_s_bh) > MAX_HEIGHT,
64                "PAP-1010: block (%b) has too big level (%z)", p_s_bh, p_s_bh);
65
66         return (B_LEVEL(p_s_bh) != FREE_LEVEL);
67 }
68
69 //
70 // to gets item head in le form
71 //
72 inline void copy_item_head(struct item_head *p_v_to,
73                            const struct item_head *p_v_from)
74 {
75         memcpy(p_v_to, p_v_from, IH_SIZE);
76 }
77
78 /* k1 is pointer to on-disk structure which is stored in little-endian
79    form. k2 is pointer to cpu variable. For key of items of the same
80    object this returns 0.
81    Returns: -1 if key1 < key2 
82    0 if key1 == key2
83    1 if key1 > key2 */
84 inline int comp_short_keys(const struct reiserfs_key *le_key,
85                            const struct cpu_key *cpu_key)
86 {
87         __u32 n;
88         n = le32_to_cpu(le_key->k_dir_id);
89         if (n < cpu_key->on_disk_key.k_dir_id)
90                 return -1;
91         if (n > cpu_key->on_disk_key.k_dir_id)
92                 return 1;
93         n = le32_to_cpu(le_key->k_objectid);
94         if (n < cpu_key->on_disk_key.k_objectid)
95                 return -1;
96         if (n > cpu_key->on_disk_key.k_objectid)
97                 return 1;
98         return 0;
99 }
100
101 /* k1 is pointer to on-disk structure which is stored in little-endian
102    form. k2 is pointer to cpu variable.
103    Compare keys using all 4 key fields.
104    Returns: -1 if key1 < key2 0
105    if key1 = key2 1 if key1 > key2 */
106 static inline int comp_keys(const struct reiserfs_key *le_key,
107                             const struct cpu_key *cpu_key)
108 {
109         int retval;
110
111         retval = comp_short_keys(le_key, cpu_key);
112         if (retval)
113                 return retval;
114         if (le_key_k_offset(le_key_version(le_key), le_key) <
115             cpu_key_k_offset(cpu_key))
116                 return -1;
117         if (le_key_k_offset(le_key_version(le_key), le_key) >
118             cpu_key_k_offset(cpu_key))
119                 return 1;
120
121         if (cpu_key->key_length == 3)
122                 return 0;
123
124         /* this part is needed only when tail conversion is in progress */
125         if (le_key_k_type(le_key_version(le_key), le_key) <
126             cpu_key_k_type(cpu_key))
127                 return -1;
128
129         if (le_key_k_type(le_key_version(le_key), le_key) >
130             cpu_key_k_type(cpu_key))
131                 return 1;
132
133         return 0;
134 }
135
136 inline int comp_short_le_keys(const struct reiserfs_key *key1,
137                               const struct reiserfs_key *key2)
138 {
139         __u32 *p_s_1_u32, *p_s_2_u32;
140         int n_key_length = REISERFS_SHORT_KEY_LEN;
141
142         p_s_1_u32 = (__u32 *) key1;
143         p_s_2_u32 = (__u32 *) key2;
144         for (; n_key_length--; ++p_s_1_u32, ++p_s_2_u32) {
145                 if (le32_to_cpu(*p_s_1_u32) < le32_to_cpu(*p_s_2_u32))
146                         return -1;
147                 if (le32_to_cpu(*p_s_1_u32) > le32_to_cpu(*p_s_2_u32))
148                         return 1;
149         }
150         return 0;
151 }
152
153 inline void le_key2cpu_key(struct cpu_key *to, const struct reiserfs_key *from)
154 {
155         int version;
156         to->on_disk_key.k_dir_id = le32_to_cpu(from->k_dir_id);
157         to->on_disk_key.k_objectid = le32_to_cpu(from->k_objectid);
158
159         // find out version of the key
160         version = le_key_version(from);
161         to->version = version;
162         to->on_disk_key.k_offset = le_key_k_offset(version, from);
163         to->on_disk_key.k_type = le_key_k_type(version, from);
164 }
165
166 // this does not say which one is bigger, it only returns 1 if keys
167 // are not equal, 0 otherwise
168 inline int comp_le_keys(const struct reiserfs_key *k1,
169                         const struct reiserfs_key *k2)
170 {
171         return memcmp(k1, k2, sizeof(struct reiserfs_key));
172 }
173
174 /**************************************************************************
175  *  Binary search toolkit function                                        *
176  *  Search for an item in the array by the item key                       *
177  *  Returns:    1 if found,  0 if not found;                              *
178  *        *p_n_pos = number of the searched element if found, else the    *
179  *        number of the first element that is larger than p_v_key.        *
180  **************************************************************************/
181 /* For those not familiar with binary search: n_lbound is the leftmost item that it
182  could be, n_rbound the rightmost item that it could be.  We examine the item
183  halfway between n_lbound and n_rbound, and that tells us either that we can increase
184  n_lbound, or decrease n_rbound, or that we have found it, or if n_lbound <= n_rbound that
185  there are no possible items, and we have not found it. With each examination we
186  cut the number of possible items it could be by one more than half rounded down,
187  or we find it. */
188 static inline int bin_search(const void *p_v_key,       /* Key to search for.                   */
189                              const void *p_v_base,      /* First item in the array.             */
190                              int p_n_num,       /* Number of items in the array.        */
191                              int p_n_width,     /* Item size in the array.
192                                                    searched. Lest the reader be
193                                                    confused, note that this is crafted
194                                                    as a general function, and when it
195                                                    is applied specifically to the array
196                                                    of item headers in a node, p_n_width
197                                                    is actually the item header size not
198                                                    the item size.                      */
199                              int *p_n_pos       /* Number of the searched for element. */
200     )
201 {
202         int n_rbound, n_lbound, n_j;
203
204         for (n_j = ((n_rbound = p_n_num - 1) + (n_lbound = 0)) / 2;
205              n_lbound <= n_rbound; n_j = (n_rbound + n_lbound) / 2)
206                 switch (comp_keys
207                         ((struct reiserfs_key *)((char *)p_v_base +
208                                                  n_j * p_n_width),
209                          (struct cpu_key *)p_v_key)) {
210                 case -1:
211                         n_lbound = n_j + 1;
212                         continue;
213                 case 1:
214                         n_rbound = n_j - 1;
215                         continue;
216                 case 0:
217                         *p_n_pos = n_j;
218                         return ITEM_FOUND;      /* Key found in the array.  */
219                 }
220
221         /* bin_search did not find given key, it returns position of key,
222            that is minimal and greater than the given one. */
223         *p_n_pos = n_lbound;
224         return ITEM_NOT_FOUND;
225 }
226
227 #ifdef CONFIG_REISERFS_CHECK
228 extern struct tree_balance *cur_tb;
229 #endif
230
231 /* Minimal possible key. It is never in the tree. */
232 const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} };
233
234 /* Maximal possible key. It is never in the tree. */
235 static const struct reiserfs_key MAX_KEY = {
236         __constant_cpu_to_le32(0xffffffff),
237         __constant_cpu_to_le32(0xffffffff),
238         {{__constant_cpu_to_le32(0xffffffff),
239           __constant_cpu_to_le32(0xffffffff)},}
240 };
241
242 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
243    of the path, and going upwards.  We must check the path's validity at each step.  If the key is not in
244    the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
245    case we return a special key, either MIN_KEY or MAX_KEY. */
246 static inline const struct reiserfs_key *get_lkey(const struct treepath
247                                                   *p_s_chk_path,
248                                                   const struct super_block
249                                                   *p_s_sb)
250 {
251         int n_position, n_path_offset = p_s_chk_path->path_length;
252         struct buffer_head *p_s_parent;
253
254         RFALSE(n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
255                "PAP-5010: invalid offset in the path");
256
257         /* While not higher in path than first element. */
258         while (n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
259
260                 RFALSE(!buffer_uptodate
261                        (PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
262                        "PAP-5020: parent is not uptodate");
263
264                 /* Parent at the path is not in the tree now. */
265                 if (!B_IS_IN_TREE
266                     (p_s_parent =
267                      PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)))
268                         return &MAX_KEY;
269                 /* Check whether position in the parent is correct. */
270                 if ((n_position =
271                      PATH_OFFSET_POSITION(p_s_chk_path,
272                                           n_path_offset)) >
273                     B_NR_ITEMS(p_s_parent))
274                         return &MAX_KEY;
275                 /* Check whether parent at the path really points to the child. */
276                 if (B_N_CHILD_NUM(p_s_parent, n_position) !=
277                     PATH_OFFSET_PBUFFER(p_s_chk_path,
278                                         n_path_offset + 1)->b_blocknr)
279                         return &MAX_KEY;
280                 /* Return delimiting key if position in the parent is not equal to zero. */
281                 if (n_position)
282                         return B_N_PDELIM_KEY(p_s_parent, n_position - 1);
283         }
284         /* Return MIN_KEY if we are in the root of the buffer tree. */
285         if (PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->
286             b_blocknr == SB_ROOT_BLOCK(p_s_sb))
287                 return &MIN_KEY;
288         return &MAX_KEY;
289 }
290
291 /* Get delimiting key of the buffer at the path and its right neighbor. */
292 inline const struct reiserfs_key *get_rkey(const struct treepath *p_s_chk_path,
293                                            const struct super_block *p_s_sb)
294 {
295         int n_position, n_path_offset = p_s_chk_path->path_length;
296         struct buffer_head *p_s_parent;
297
298         RFALSE(n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
299                "PAP-5030: invalid offset in the path");
300
301         while (n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
302
303                 RFALSE(!buffer_uptodate
304                        (PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
305                        "PAP-5040: parent is not uptodate");
306
307                 /* Parent at the path is not in the tree now. */
308                 if (!B_IS_IN_TREE
309                     (p_s_parent =
310                      PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)))
311                         return &MIN_KEY;
312                 /* Check whether position in the parent is correct. */
313                 if ((n_position =
314                      PATH_OFFSET_POSITION(p_s_chk_path,
315                                           n_path_offset)) >
316                     B_NR_ITEMS(p_s_parent))
317                         return &MIN_KEY;
318                 /* Check whether parent at the path really points to the child. */
319                 if (B_N_CHILD_NUM(p_s_parent, n_position) !=
320                     PATH_OFFSET_PBUFFER(p_s_chk_path,
321                                         n_path_offset + 1)->b_blocknr)
322                         return &MIN_KEY;
323                 /* Return delimiting key if position in the parent is not the last one. */
324                 if (n_position != B_NR_ITEMS(p_s_parent))
325                         return B_N_PDELIM_KEY(p_s_parent, n_position);
326         }
327         /* Return MAX_KEY if we are in the root of the buffer tree. */
328         if (PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->
329             b_blocknr == SB_ROOT_BLOCK(p_s_sb))
330                 return &MAX_KEY;
331         return &MIN_KEY;
332 }
333
334 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
335 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
336    the path.  These delimiting keys are stored at least one level above that buffer in the tree. If the
337    buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
338    this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
339 static inline int key_in_buffer(struct treepath *p_s_chk_path,  /* Path which should be checked.  */
340                                 const struct cpu_key *p_s_key,  /* Key which should be checked.   */
341                                 struct super_block *p_s_sb      /* Super block pointer.           */
342     )
343 {
344
345         RFALSE(!p_s_key || p_s_chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET
346                || p_s_chk_path->path_length > MAX_HEIGHT,
347                "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
348                p_s_key, p_s_chk_path->path_length);
349         RFALSE(!PATH_PLAST_BUFFER(p_s_chk_path)->b_bdev,
350                "PAP-5060: device must not be NODEV");
351
352         if (comp_keys(get_lkey(p_s_chk_path, p_s_sb), p_s_key) == 1)
353                 /* left delimiting key is bigger, that the key we look for */
354                 return 0;
355         //  if ( comp_keys(p_s_key, get_rkey(p_s_chk_path, p_s_sb)) != -1 )
356         if (comp_keys(get_rkey(p_s_chk_path, p_s_sb), p_s_key) != 1)
357                 /* p_s_key must be less than right delimitiing key */
358                 return 0;
359         return 1;
360 }
361
362 inline void decrement_bcount(struct buffer_head *p_s_bh)
363 {
364         if (p_s_bh) {
365                 if (atomic_read(&(p_s_bh->b_count))) {
366                         put_bh(p_s_bh);
367                         return;
368                 }
369                 reiserfs_panic(NULL,
370                                "PAP-5070: decrement_bcount: trying to free free buffer %b",
371                                p_s_bh);
372         }
373 }
374
375 /* Decrement b_count field of the all buffers in the path. */
376 void decrement_counters_in_path(struct treepath *p_s_search_path)
377 {
378         int n_path_offset = p_s_search_path->path_length;
379
380         RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET ||
381                n_path_offset > EXTENDED_MAX_HEIGHT - 1,
382                "PAP-5080: invalid path offset of %d", n_path_offset);
383
384         while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
385                 struct buffer_head *bh;
386
387                 bh = PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--);
388                 decrement_bcount(bh);
389         }
390         p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
391 }
392
393 int reiserfs_check_path(struct treepath *p)
394 {
395         RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
396                "path not properly relsed");
397         return 0;
398 }
399
400 /* Release all buffers in the path. Restore dirty bits clean
401 ** when preparing the buffer for the log
402 **
403 ** only called from fix_nodes()
404 */
405 void pathrelse_and_restore(struct super_block *s, struct treepath *p_s_search_path)
406 {
407         int n_path_offset = p_s_search_path->path_length;
408
409         RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
410                "clm-4000: invalid path offset");
411
412         while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
413                 reiserfs_restore_prepared_buffer(s,
414                                                  PATH_OFFSET_PBUFFER
415                                                  (p_s_search_path,
416                                                   n_path_offset));
417                 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
418         }
419         p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
420 }
421
422 /* Release all buffers in the path. */
423 void pathrelse(struct treepath *p_s_search_path)
424 {
425         int n_path_offset = p_s_search_path->path_length;
426
427         RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
428                "PAP-5090: invalid path offset");
429
430         while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
431                 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
432
433         p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
434 }
435
436 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
437 {
438         struct block_head *blkh;
439         struct item_head *ih;
440         int used_space;
441         int prev_location;
442         int i;
443         int nr;
444
445         blkh = (struct block_head *)buf;
446         if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
447                 reiserfs_warning(NULL,
448                                  "is_leaf: this should be caught earlier");
449                 return 0;
450         }
451
452         nr = blkh_nr_item(blkh);
453         if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
454                 /* item number is too big or too small */
455                 reiserfs_warning(NULL, "is_leaf: nr_item seems wrong: %z", bh);
456                 return 0;
457         }
458         ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
459         used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih));
460         if (used_space != blocksize - blkh_free_space(blkh)) {
461                 /* free space does not match to calculated amount of use space */
462                 reiserfs_warning(NULL, "is_leaf: free space seems wrong: %z",
463                                  bh);
464                 return 0;
465         }
466         // FIXME: it is_leaf will hit performance too much - we may have
467         // return 1 here
468
469         /* check tables of item heads */
470         ih = (struct item_head *)(buf + BLKH_SIZE);
471         prev_location = blocksize;
472         for (i = 0; i < nr; i++, ih++) {
473                 if (le_ih_k_type(ih) == TYPE_ANY) {
474                         reiserfs_warning(NULL,
475                                          "is_leaf: wrong item type for item %h",
476                                          ih);
477                         return 0;
478                 }
479                 if (ih_location(ih) >= blocksize
480                     || ih_location(ih) < IH_SIZE * nr) {
481                         reiserfs_warning(NULL,
482                                          "is_leaf: item location seems wrong: %h",
483                                          ih);
484                         return 0;
485                 }
486                 if (ih_item_len(ih) < 1
487                     || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
488                         reiserfs_warning(NULL,
489                                          "is_leaf: item length seems wrong: %h",
490                                          ih);
491                         return 0;
492                 }
493                 if (prev_location - ih_location(ih) != ih_item_len(ih)) {
494                         reiserfs_warning(NULL,
495                                          "is_leaf: item location seems wrong (second one): %h",
496                                          ih);
497                         return 0;
498                 }
499                 prev_location = ih_location(ih);
500         }
501
502         // one may imagine much more checks
503         return 1;
504 }
505
506 /* returns 1 if buf looks like an internal node, 0 otherwise */
507 static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
508 {
509         struct block_head *blkh;
510         int nr;
511         int used_space;
512
513         blkh = (struct block_head *)buf;
514         nr = blkh_level(blkh);
515         if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
516                 /* this level is not possible for internal nodes */
517                 reiserfs_warning(NULL,
518                                  "is_internal: this should be caught earlier");
519                 return 0;
520         }
521
522         nr = blkh_nr_item(blkh);
523         if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
524                 /* for internal which is not root we might check min number of keys */
525                 reiserfs_warning(NULL,
526                                  "is_internal: number of key seems wrong: %z",
527                                  bh);
528                 return 0;
529         }
530
531         used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
532         if (used_space != blocksize - blkh_free_space(blkh)) {
533                 reiserfs_warning(NULL,
534                                  "is_internal: free space seems wrong: %z", bh);
535                 return 0;
536         }
537         // one may imagine much more checks
538         return 1;
539 }
540
541 // make sure that bh contains formatted node of reiserfs tree of
542 // 'level'-th level
543 static int is_tree_node(struct buffer_head *bh, int level)
544 {
545         if (B_LEVEL(bh) != level) {
546                 reiserfs_warning(NULL,
547                                  "is_tree_node: node level %d does not match to the expected one %d",
548                                  B_LEVEL(bh), level);
549                 return 0;
550         }
551         if (level == DISK_LEAF_NODE_LEVEL)
552                 return is_leaf(bh->b_data, bh->b_size, bh);
553
554         return is_internal(bh->b_data, bh->b_size, bh);
555 }
556
557 #define SEARCH_BY_KEY_READA 16
558
559 /* The function is NOT SCHEDULE-SAFE! */
560 static void search_by_key_reada(struct super_block *s,
561                                 struct buffer_head **bh,
562                                 unsigned long *b, int num)
563 {
564         int i, j;
565
566         for (i = 0; i < num; i++) {
567                 bh[i] = sb_getblk(s, b[i]);
568         }
569         for (j = 0; j < i; j++) {
570                 /*
571                  * note, this needs attention if we are getting rid of the BKL
572                  * you have to make sure the prepared bit isn't set on this buffer
573                  */
574                 if (!buffer_uptodate(bh[j]))
575                         ll_rw_block(READA, 1, bh + j);
576                 brelse(bh[j]);
577         }
578 }
579
580 /**************************************************************************
581  * Algorithm   SearchByKey                                                *
582  *             look for item in the Disk S+Tree by its key                *
583  * Input:  p_s_sb   -  super block                                        *
584  *         p_s_key  - pointer to the key to search                        *
585  * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR                         *
586  *         p_s_search_path - path from the root to the needed leaf        *
587  **************************************************************************/
588
589 /* This function fills up the path from the root to the leaf as it
590    descends the tree looking for the key.  It uses reiserfs_bread to
591    try to find buffers in the cache given their block number.  If it
592    does not find them in the cache it reads them from disk.  For each
593    node search_by_key finds using reiserfs_bread it then uses
594    bin_search to look through that node.  bin_search will find the
595    position of the block_number of the next node if it is looking
596    through an internal node.  If it is looking through a leaf node
597    bin_search will find the position of the item which has key either
598    equal to given key, or which is the maximal key less than the given
599    key.  search_by_key returns a path that must be checked for the
600    correctness of the top of the path but need not be checked for the
601    correctness of the bottom of the path */
602 /* The function is NOT SCHEDULE-SAFE! */
603 int search_by_key(struct super_block *p_s_sb, const struct cpu_key *p_s_key,    /* Key to search. */
604                   struct treepath *p_s_search_path,/* This structure was
605                                                    allocated and initialized
606                                                    by the calling
607                                                    function. It is filled up
608                                                    by this function.  */
609                   int n_stop_level      /* How far down the tree to search. To
610                                            stop at leaf level - set to
611                                            DISK_LEAF_NODE_LEVEL */
612     )
613 {
614         int n_block_number;
615         int expected_level;
616         struct buffer_head *p_s_bh;
617         struct path_element *p_s_last_element;
618         int n_node_level, n_retval;
619         int right_neighbor_of_leaf_node;
620         int fs_gen;
621         struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
622         unsigned long reada_blocks[SEARCH_BY_KEY_READA];
623         int reada_count = 0;
624
625 #ifdef CONFIG_REISERFS_CHECK
626         int n_repeat_counter = 0;
627 #endif
628
629         PROC_INFO_INC(p_s_sb, search_by_key);
630
631         /* As we add each node to a path we increase its count.  This means that
632            we must be careful to release all nodes in a path before we either
633            discard the path struct or re-use the path struct, as we do here. */
634
635         decrement_counters_in_path(p_s_search_path);
636
637         right_neighbor_of_leaf_node = 0;
638
639         /* With each iteration of this loop we search through the items in the
640            current node, and calculate the next current node(next path element)
641            for the next iteration of this loop.. */
642         n_block_number = SB_ROOT_BLOCK(p_s_sb);
643         expected_level = -1;
644         while (1) {
645
646 #ifdef CONFIG_REISERFS_CHECK
647                 if (!(++n_repeat_counter % 50000))
648                         reiserfs_warning(p_s_sb, "PAP-5100: search_by_key: %s:"
649                                          "there were %d iterations of while loop "
650                                          "looking for key %K",
651                                          current->comm, n_repeat_counter,
652                                          p_s_key);
653 #endif
654
655                 /* prep path to have another element added to it. */
656                 p_s_last_element =
657                     PATH_OFFSET_PELEMENT(p_s_search_path,
658                                          ++p_s_search_path->path_length);
659                 fs_gen = get_generation(p_s_sb);
660
661                 /* Read the next tree node, and set the last element in the path to
662                    have a pointer to it. */
663                 if ((p_s_bh = p_s_last_element->pe_buffer =
664                      sb_getblk(p_s_sb, n_block_number))) {
665                         if (!buffer_uptodate(p_s_bh) && reada_count > 1) {
666                                 search_by_key_reada(p_s_sb, reada_bh,
667                                                     reada_blocks, reada_count);
668                         }
669                         ll_rw_block(READ, 1, &p_s_bh);
670                         wait_on_buffer(p_s_bh);
671                         if (!buffer_uptodate(p_s_bh))
672                                 goto io_error;
673                 } else {
674                       io_error:
675                         p_s_search_path->path_length--;
676                         pathrelse(p_s_search_path);
677                         return IO_ERROR;
678                 }
679                 reada_count = 0;
680                 if (expected_level == -1)
681                         expected_level = SB_TREE_HEIGHT(p_s_sb);
682                 expected_level--;
683
684                 /* It is possible that schedule occurred. We must check whether the key
685                    to search is still in the tree rooted from the current buffer. If
686                    not then repeat search from the root. */
687                 if (fs_changed(fs_gen, p_s_sb) &&
688                     (!B_IS_IN_TREE(p_s_bh) ||
689                      B_LEVEL(p_s_bh) != expected_level ||
690                      !key_in_buffer(p_s_search_path, p_s_key, p_s_sb))) {
691                         PROC_INFO_INC(p_s_sb, search_by_key_fs_changed);
692                         PROC_INFO_INC(p_s_sb, search_by_key_restarted);
693                         PROC_INFO_INC(p_s_sb,
694                                       sbk_restarted[expected_level - 1]);
695                         decrement_counters_in_path(p_s_search_path);
696
697                         /* Get the root block number so that we can repeat the search
698                            starting from the root. */
699                         n_block_number = SB_ROOT_BLOCK(p_s_sb);
700                         expected_level = -1;
701                         right_neighbor_of_leaf_node = 0;
702
703                         /* repeat search from the root */
704                         continue;
705                 }
706
707                 /* only check that the key is in the buffer if p_s_key is not
708                    equal to the MAX_KEY. Latter case is only possible in
709                    "finish_unfinished()" processing during mount. */
710                 RFALSE(comp_keys(&MAX_KEY, p_s_key) &&
711                        !key_in_buffer(p_s_search_path, p_s_key, p_s_sb),
712                        "PAP-5130: key is not in the buffer");
713 #ifdef CONFIG_REISERFS_CHECK
714                 if (cur_tb) {
715                         print_cur_tb("5140");
716                         reiserfs_panic(p_s_sb,
717                                        "PAP-5140: search_by_key: schedule occurred in do_balance!");
718                 }
719 #endif
720
721                 // make sure, that the node contents look like a node of
722                 // certain level
723                 if (!is_tree_node(p_s_bh, expected_level)) {
724                         reiserfs_warning(p_s_sb, "vs-5150: search_by_key: "
725                                          "invalid format found in block %ld. Fsck?",
726                                          p_s_bh->b_blocknr);
727                         pathrelse(p_s_search_path);
728                         return IO_ERROR;
729                 }
730
731                 /* ok, we have acquired next formatted node in the tree */
732                 n_node_level = B_LEVEL(p_s_bh);
733
734                 PROC_INFO_BH_STAT(p_s_sb, p_s_bh, n_node_level - 1);
735
736                 RFALSE(n_node_level < n_stop_level,
737                        "vs-5152: tree level (%d) is less than stop level (%d)",
738                        n_node_level, n_stop_level);
739
740                 n_retval = bin_search(p_s_key, B_N_PITEM_HEAD(p_s_bh, 0),
741                                       B_NR_ITEMS(p_s_bh),
742                                       (n_node_level ==
743                                        DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
744                                       KEY_SIZE,
745                                       &(p_s_last_element->pe_position));
746                 if (n_node_level == n_stop_level) {
747                         return n_retval;
748                 }
749
750                 /* we are not in the stop level */
751                 if (n_retval == ITEM_FOUND)
752                         /* item has been found, so we choose the pointer which is to the right of the found one */
753                         p_s_last_element->pe_position++;
754
755                 /* if item was not found we choose the position which is to
756                    the left of the found item. This requires no code,
757                    bin_search did it already. */
758
759                 /* So we have chosen a position in the current node which is
760                    an internal node.  Now we calculate child block number by
761                    position in the node. */
762                 n_block_number =
763                     B_N_CHILD_NUM(p_s_bh, p_s_last_element->pe_position);
764
765                 /* if we are going to read leaf nodes, try for read ahead as well */
766                 if ((p_s_search_path->reada & PATH_READA) &&
767                     n_node_level == DISK_LEAF_NODE_LEVEL + 1) {
768                         int pos = p_s_last_element->pe_position;
769                         int limit = B_NR_ITEMS(p_s_bh);
770                         struct reiserfs_key *le_key;
771
772                         if (p_s_search_path->reada & PATH_READA_BACK)
773                                 limit = 0;
774                         while (reada_count < SEARCH_BY_KEY_READA) {
775                                 if (pos == limit)
776                                         break;
777                                 reada_blocks[reada_count++] =
778                                     B_N_CHILD_NUM(p_s_bh, pos);
779                                 if (p_s_search_path->reada & PATH_READA_BACK)
780                                         pos--;
781                                 else
782                                         pos++;
783
784                                 /*
785                                  * check to make sure we're in the same object
786                                  */
787                                 le_key = B_N_PDELIM_KEY(p_s_bh, pos);
788                                 if (le32_to_cpu(le_key->k_objectid) !=
789                                     p_s_key->on_disk_key.k_objectid) {
790                                         break;
791                                 }
792                         }
793                 }
794         }
795 }
796
797 /* Form the path to an item and position in this item which contains
798    file byte defined by p_s_key. If there is no such item
799    corresponding to the key, we point the path to the item with
800    maximal key less than p_s_key, and *p_n_pos_in_item is set to one
801    past the last entry/byte in the item.  If searching for entry in a
802    directory item, and it is not found, *p_n_pos_in_item is set to one
803    entry more than the entry with maximal key which is less than the
804    sought key.
805
806    Note that if there is no entry in this same node which is one more,
807    then we point to an imaginary entry.  for direct items, the
808    position is in units of bytes, for indirect items the position is
809    in units of blocknr entries, for directory items the position is in
810    units of directory entries.  */
811
812 /* The function is NOT SCHEDULE-SAFE! */
813 int search_for_position_by_key(struct super_block *p_s_sb,      /* Pointer to the super block.          */
814                                const struct cpu_key *p_cpu_key, /* Key to search (cpu variable)         */
815                                struct treepath *p_s_search_path /* Filled up by this function.          */
816     )
817 {
818         struct item_head *p_le_ih;      /* pointer to on-disk structure */
819         int n_blk_size;
820         loff_t item_offset, offset;
821         struct reiserfs_dir_entry de;
822         int retval;
823
824         /* If searching for directory entry. */
825         if (is_direntry_cpu_key(p_cpu_key))
826                 return search_by_entry_key(p_s_sb, p_cpu_key, p_s_search_path,
827                                            &de);
828
829         /* If not searching for directory entry. */
830
831         /* If item is found. */
832         retval = search_item(p_s_sb, p_cpu_key, p_s_search_path);
833         if (retval == IO_ERROR)
834                 return retval;
835         if (retval == ITEM_FOUND) {
836
837                 RFALSE(!ih_item_len
838                        (B_N_PITEM_HEAD
839                         (PATH_PLAST_BUFFER(p_s_search_path),
840                          PATH_LAST_POSITION(p_s_search_path))),
841                        "PAP-5165: item length equals zero");
842
843                 pos_in_item(p_s_search_path) = 0;
844                 return POSITION_FOUND;
845         }
846
847         RFALSE(!PATH_LAST_POSITION(p_s_search_path),
848                "PAP-5170: position equals zero");
849
850         /* Item is not found. Set path to the previous item. */
851         p_le_ih =
852             B_N_PITEM_HEAD(PATH_PLAST_BUFFER(p_s_search_path),
853                            --PATH_LAST_POSITION(p_s_search_path));
854         n_blk_size = p_s_sb->s_blocksize;
855
856         if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) {
857                 return FILE_NOT_FOUND;
858         }
859         // FIXME: quite ugly this far
860
861         item_offset = le_ih_k_offset(p_le_ih);
862         offset = cpu_key_k_offset(p_cpu_key);
863
864         /* Needed byte is contained in the item pointed to by the path. */
865         if (item_offset <= offset &&
866             item_offset + op_bytes_number(p_le_ih, n_blk_size) > offset) {
867                 pos_in_item(p_s_search_path) = offset - item_offset;
868                 if (is_indirect_le_ih(p_le_ih)) {
869                         pos_in_item(p_s_search_path) /= n_blk_size;
870                 }
871                 return POSITION_FOUND;
872         }
873
874         /* Needed byte is not contained in the item pointed to by the
875            path. Set pos_in_item out of the item. */
876         if (is_indirect_le_ih(p_le_ih))
877                 pos_in_item(p_s_search_path) =
878                     ih_item_len(p_le_ih) / UNFM_P_SIZE;
879         else
880                 pos_in_item(p_s_search_path) = ih_item_len(p_le_ih);
881
882         return POSITION_NOT_FOUND;
883 }
884
885 /* Compare given item and item pointed to by the path. */
886 int comp_items(const struct item_head *stored_ih, const struct treepath *p_s_path)
887 {
888         struct buffer_head *p_s_bh;
889         struct item_head *ih;
890
891         /* Last buffer at the path is not in the tree. */
892         if (!B_IS_IN_TREE(p_s_bh = PATH_PLAST_BUFFER(p_s_path)))
893                 return 1;
894
895         /* Last path position is invalid. */
896         if (PATH_LAST_POSITION(p_s_path) >= B_NR_ITEMS(p_s_bh))
897                 return 1;
898
899         /* we need only to know, whether it is the same item */
900         ih = get_ih(p_s_path);
901         return memcmp(stored_ih, ih, IH_SIZE);
902 }
903
904 /* unformatted nodes are not logged anymore, ever.  This is safe
905 ** now
906 */
907 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
908
909 // block can not be forgotten as it is in I/O or held by someone
910 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
911
912 // prepare for delete or cut of direct item
913 static inline int prepare_for_direct_item(struct treepath *path,
914                                           struct item_head *le_ih,
915                                           struct inode *inode,
916                                           loff_t new_file_length, int *cut_size)
917 {
918         loff_t round_len;
919
920         if (new_file_length == max_reiserfs_offset(inode)) {
921                 /* item has to be deleted */
922                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
923                 return M_DELETE;
924         }
925         // new file gets truncated
926         if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
927                 // 
928                 round_len = ROUND_UP(new_file_length);
929                 /* this was n_new_file_length < le_ih ... */
930                 if (round_len < le_ih_k_offset(le_ih)) {
931                         *cut_size = -(IH_SIZE + ih_item_len(le_ih));
932                         return M_DELETE;        /* Delete this item. */
933                 }
934                 /* Calculate first position and size for cutting from item. */
935                 pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
936                 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
937
938                 return M_CUT;   /* Cut from this item. */
939         }
940
941         // old file: items may have any length
942
943         if (new_file_length < le_ih_k_offset(le_ih)) {
944                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
945                 return M_DELETE;        /* Delete this item. */
946         }
947         /* Calculate first position and size for cutting from item. */
948         *cut_size = -(ih_item_len(le_ih) -
949                       (pos_in_item(path) =
950                        new_file_length + 1 - le_ih_k_offset(le_ih)));
951         return M_CUT;           /* Cut from this item. */
952 }
953
954 static inline int prepare_for_direntry_item(struct treepath *path,
955                                             struct item_head *le_ih,
956                                             struct inode *inode,
957                                             loff_t new_file_length,
958                                             int *cut_size)
959 {
960         if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
961             new_file_length == max_reiserfs_offset(inode)) {
962                 RFALSE(ih_entry_count(le_ih) != 2,
963                        "PAP-5220: incorrect empty directory item (%h)", le_ih);
964                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
965                 return M_DELETE;        /* Delete the directory item containing "." and ".." entry. */
966         }
967
968         if (ih_entry_count(le_ih) == 1) {
969                 /* Delete the directory item such as there is one record only
970                    in this item */
971                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
972                 return M_DELETE;
973         }
974
975         /* Cut one record from the directory item. */
976         *cut_size =
977             -(DEH_SIZE +
978               entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
979         return M_CUT;
980 }
981
982 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
983
984 /*  If the path points to a directory or direct item, calculate mode and the size cut, for balance.
985     If the path points to an indirect item, remove some number of its unformatted nodes.
986     In case of file truncate calculate whether this item must be deleted/truncated or last
987     unformatted node of this item will be converted to a direct item.
988     This function returns a determination of what balance mode the calling function should employ. */
989 static char prepare_for_delete_or_cut(struct reiserfs_transaction_handle *th, struct inode *inode, struct treepath *p_s_path, const struct cpu_key *p_s_item_key, int *p_n_removed,     /* Number of unformatted nodes which were removed
990                                                                                                                                                                                    from end of the file. */
991                                       int *p_n_cut_size, unsigned long long n_new_file_length   /* MAX_KEY_OFFSET in case of delete. */
992     )
993 {
994         struct super_block *p_s_sb = inode->i_sb;
995         struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_path);
996         struct buffer_head *p_s_bh = PATH_PLAST_BUFFER(p_s_path);
997
998         BUG_ON(!th->t_trans_id);
999
1000         /* Stat_data item. */
1001         if (is_statdata_le_ih(p_le_ih)) {
1002
1003                 RFALSE(n_new_file_length != max_reiserfs_offset(inode),
1004                        "PAP-5210: mode must be M_DELETE");
1005
1006                 *p_n_cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
1007                 return M_DELETE;
1008         }
1009
1010         /* Directory item. */
1011         if (is_direntry_le_ih(p_le_ih))
1012                 return prepare_for_direntry_item(p_s_path, p_le_ih, inode,
1013                                                  n_new_file_length,
1014                                                  p_n_cut_size);
1015
1016         /* Direct item. */
1017         if (is_direct_le_ih(p_le_ih))
1018                 return prepare_for_direct_item(p_s_path, p_le_ih, inode,
1019                                                n_new_file_length, p_n_cut_size);
1020
1021         /* Case of an indirect item. */
1022         {
1023             int blk_size = p_s_sb->s_blocksize;
1024             struct item_head s_ih;
1025             int need_re_search;
1026             int delete = 0;
1027             int result = M_CUT;
1028             int pos = 0;
1029
1030             if ( n_new_file_length == max_reiserfs_offset (inode) ) {
1031                 /* prepare_for_delete_or_cut() is called by
1032                  * reiserfs_delete_item() */
1033                 n_new_file_length = 0;
1034                 delete = 1;
1035             }
1036
1037             do {
1038                 need_re_search = 0;
1039                 *p_n_cut_size = 0;
1040                 p_s_bh = PATH_PLAST_BUFFER(p_s_path);
1041                 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1042                 pos = I_UNFM_NUM(&s_ih);
1043
1044                 while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > n_new_file_length) {
1045                     __u32 *unfm, block;
1046
1047                     /* Each unformatted block deletion may involve one additional
1048                      * bitmap block into the transaction, thereby the initial
1049                      * journal space reservation might not be enough. */
1050                     if (!delete && (*p_n_cut_size) != 0 &&
1051                         reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1052                         break;
1053                     }
1054
1055                     unfm = (__u32 *)B_I_PITEM(p_s_bh, &s_ih) + pos - 1;
1056                     block = get_block_num(unfm, 0);
1057
1058                     if (block != 0) {
1059                         reiserfs_prepare_for_journal(p_s_sb, p_s_bh, 1);
1060                         put_block_num(unfm, 0, 0);
1061                         journal_mark_dirty (th, p_s_sb, p_s_bh);
1062                         reiserfs_free_block(th, inode, block, 1);
1063                     }
1064
1065                     cond_resched();
1066
1067                     if (item_moved (&s_ih, p_s_path))  {
1068                         need_re_search = 1;
1069                         break;
1070                     }
1071
1072                     pos --;
1073                     (*p_n_removed) ++;
1074                     (*p_n_cut_size) -= UNFM_P_SIZE;
1075
1076                     if (pos == 0) {
1077                         (*p_n_cut_size) -= IH_SIZE;
1078                         result = M_DELETE;
1079                         break;
1080                     }
1081                 }
1082                 /* a trick.  If the buffer has been logged, this will do nothing.  If
1083                 ** we've broken the loop without logging it, it will restore the
1084                 ** buffer */
1085                 reiserfs_restore_prepared_buffer(p_s_sb, p_s_bh);
1086             } while (need_re_search &&
1087                      search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path) == POSITION_FOUND);
1088             pos_in_item(p_s_path) = pos * UNFM_P_SIZE;
1089
1090             if (*p_n_cut_size == 0) {
1091                 /* Nothing were cut. maybe convert last unformatted node to the
1092                  * direct item? */
1093                 result = M_CONVERT;
1094             }
1095             return result;
1096         }
1097 }
1098
1099 /* Calculate number of bytes which will be deleted or cut during balance */
1100 static int calc_deleted_bytes_number(struct tree_balance *p_s_tb, char c_mode)
1101 {
1102         int n_del_size;
1103         struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_tb->tb_path);
1104
1105         if (is_statdata_le_ih(p_le_ih))
1106                 return 0;
1107
1108         n_del_size =
1109             (c_mode ==
1110              M_DELETE) ? ih_item_len(p_le_ih) : -p_s_tb->insert_size[0];
1111         if (is_direntry_le_ih(p_le_ih)) {
1112                 // return EMPTY_DIR_SIZE; /* We delete emty directoris only. */
1113                 // we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1114                 // empty size.  ick. FIXME, is this right?
1115                 //
1116                 return n_del_size;
1117         }
1118
1119         if (is_indirect_le_ih(p_le_ih))
1120                 n_del_size = (n_del_size / UNFM_P_SIZE) * (PATH_PLAST_BUFFER(p_s_tb->tb_path)->b_size); // - get_ih_free_space (p_le_ih);
1121         return n_del_size;
1122 }
1123
1124 static void init_tb_struct(struct reiserfs_transaction_handle *th,
1125                            struct tree_balance *p_s_tb,
1126                            struct super_block *p_s_sb,
1127                            struct treepath *p_s_path, int n_size)
1128 {
1129
1130         BUG_ON(!th->t_trans_id);
1131
1132         memset(p_s_tb, '\0', sizeof(struct tree_balance));
1133         p_s_tb->transaction_handle = th;
1134         p_s_tb->tb_sb = p_s_sb;
1135         p_s_tb->tb_path = p_s_path;
1136         PATH_OFFSET_PBUFFER(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1137         PATH_OFFSET_POSITION(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1138         p_s_tb->insert_size[0] = n_size;
1139 }
1140
1141 void padd_item(char *item, int total_length, int length)
1142 {
1143         int i;
1144
1145         for (i = total_length; i > length;)
1146                 item[--i] = 0;
1147 }
1148
1149 #ifdef REISERQUOTA_DEBUG
1150 char key2type(struct reiserfs_key *ih)
1151 {
1152         if (is_direntry_le_key(2, ih))
1153                 return 'd';
1154         if (is_direct_le_key(2, ih))
1155                 return 'D';
1156         if (is_indirect_le_key(2, ih))
1157                 return 'i';
1158         if (is_statdata_le_key(2, ih))
1159                 return 's';
1160         return 'u';
1161 }
1162
1163 char head2type(struct item_head *ih)
1164 {
1165         if (is_direntry_le_ih(ih))
1166                 return 'd';
1167         if (is_direct_le_ih(ih))
1168                 return 'D';
1169         if (is_indirect_le_ih(ih))
1170                 return 'i';
1171         if (is_statdata_le_ih(ih))
1172                 return 's';
1173         return 'u';
1174 }
1175 #endif
1176
1177 /* Delete object item. */
1178 int reiserfs_delete_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_path,     /* Path to the deleted item. */
1179                          const struct cpu_key *p_s_item_key,    /* Key to search for the deleted item.  */
1180                          struct inode *p_s_inode,       /* inode is here just to update i_blocks and quotas */
1181                          struct buffer_head *p_s_un_bh)
1182 {                               /* NULL or unformatted node pointer.    */
1183         struct super_block *p_s_sb = p_s_inode->i_sb;
1184         struct tree_balance s_del_balance;
1185         struct item_head s_ih;
1186         struct item_head *q_ih;
1187         int quota_cut_bytes;
1188         int n_ret_value, n_del_size, n_removed;
1189
1190 #ifdef CONFIG_REISERFS_CHECK
1191         char c_mode;
1192         int n_iter = 0;
1193 #endif
1194
1195         BUG_ON(!th->t_trans_id);
1196
1197         init_tb_struct(th, &s_del_balance, p_s_sb, p_s_path,
1198                        0 /*size is unknown */ );
1199
1200         while (1) {
1201                 n_removed = 0;
1202
1203 #ifdef CONFIG_REISERFS_CHECK
1204                 n_iter++;
1205                 c_mode =
1206 #endif
1207                     prepare_for_delete_or_cut(th, p_s_inode, p_s_path,
1208                                               p_s_item_key, &n_removed,
1209                                               &n_del_size,
1210                                               max_reiserfs_offset(p_s_inode));
1211
1212                 RFALSE(c_mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1213
1214                 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1215                 s_del_balance.insert_size[0] = n_del_size;
1216
1217                 n_ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1218                 if (n_ret_value != REPEAT_SEARCH)
1219                         break;
1220
1221                 PROC_INFO_INC(p_s_sb, delete_item_restarted);
1222
1223                 // file system changed, repeat search
1224                 n_ret_value =
1225                     search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1226                 if (n_ret_value == IO_ERROR)
1227                         break;
1228                 if (n_ret_value == FILE_NOT_FOUND) {
1229                         reiserfs_warning(p_s_sb,
1230                                          "vs-5340: reiserfs_delete_item: "
1231                                          "no items of the file %K found",
1232                                          p_s_item_key);
1233                         break;
1234                 }
1235         }                       /* while (1) */
1236
1237         if (n_ret_value != CARRY_ON) {
1238                 unfix_nodes(&s_del_balance);
1239                 return 0;
1240         }
1241         // reiserfs_delete_item returns item length when success
1242         n_ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1243         q_ih = get_ih(p_s_path);
1244         quota_cut_bytes = ih_item_len(q_ih);
1245
1246         /* hack so the quota code doesn't have to guess if the file
1247          ** has a tail.  On tail insert, we allocate quota for 1 unformatted node.
1248          ** We test the offset because the tail might have been
1249          ** split into multiple items, and we only want to decrement for
1250          ** the unfm node once
1251          */
1252         if (!S_ISLNK(p_s_inode->i_mode) && is_direct_le_ih(q_ih)) {
1253                 if ((le_ih_k_offset(q_ih) & (p_s_sb->s_blocksize - 1)) == 1) {
1254                         quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1255                 } else {
1256                         quota_cut_bytes = 0;
1257                 }
1258         }
1259
1260         if (p_s_un_bh) {
1261                 int off;
1262                 char *data;
1263
1264                 /* We are in direct2indirect conversion, so move tail contents
1265                    to the unformatted node */
1266                 /* note, we do the copy before preparing the buffer because we
1267                  ** don't care about the contents of the unformatted node yet.
1268                  ** the only thing we really care about is the direct item's data
1269                  ** is in the unformatted node.
1270                  **
1271                  ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1272                  ** the unformatted node, which might schedule, meaning we'd have to
1273                  ** loop all the way back up to the start of the while loop.
1274                  **
1275                  ** The unformatted node must be dirtied later on.  We can't be
1276                  ** sure here if the entire tail has been deleted yet.
1277                  **
1278                  ** p_s_un_bh is from the page cache (all unformatted nodes are
1279                  ** from the page cache) and might be a highmem page.  So, we
1280                  ** can't use p_s_un_bh->b_data.
1281                  ** -clm
1282                  */
1283
1284                 data = kmap_atomic(p_s_un_bh->b_page, KM_USER0);
1285                 off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1286                 memcpy(data + off,
1287                        B_I_PITEM(PATH_PLAST_BUFFER(p_s_path), &s_ih),
1288                        n_ret_value);
1289                 kunmap_atomic(data, KM_USER0);
1290         }
1291         /* Perform balancing after all resources have been collected at once. */
1292         do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1293
1294 #ifdef REISERQUOTA_DEBUG
1295         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1296                        "reiserquota delete_item(): freeing %u, id=%u type=%c",
1297                        quota_cut_bytes, p_s_inode->i_uid, head2type(&s_ih));
1298 #endif
1299         DQUOT_FREE_SPACE_NODIRTY(p_s_inode, quota_cut_bytes);
1300
1301         /* Return deleted body length */
1302         return n_ret_value;
1303 }
1304
1305 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1306
1307  deletion of the body of the object is performed by iput(), with the
1308  result that if multiple processes are operating on a file, the
1309  deletion of the body of the file is deferred until the last process
1310  that has an open inode performs its iput().
1311
1312  writes and truncates are protected from collisions by use of
1313  semaphores.
1314
1315  creates, linking, and mknod are protected from collisions with other
1316  processes by making the reiserfs_add_entry() the last step in the
1317  creation, and then rolling back all changes if there was a collision.
1318  - Hans
1319 */
1320
1321 /* this deletes item which never gets split */
1322 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1323                                 struct inode *inode, struct reiserfs_key *key)
1324 {
1325         struct tree_balance tb;
1326         INITIALIZE_PATH(path);
1327         int item_len = 0;
1328         int tb_init = 0;
1329         struct cpu_key cpu_key;
1330         int retval;
1331         int quota_cut_bytes = 0;
1332
1333         BUG_ON(!th->t_trans_id);
1334
1335         le_key2cpu_key(&cpu_key, key);
1336
1337         while (1) {
1338                 retval = search_item(th->t_super, &cpu_key, &path);
1339                 if (retval == IO_ERROR) {
1340                         reiserfs_warning(th->t_super,
1341                                          "vs-5350: reiserfs_delete_solid_item: "
1342                                          "i/o failure occurred trying to delete %K",
1343                                          &cpu_key);
1344                         break;
1345                 }
1346                 if (retval != ITEM_FOUND) {
1347                         pathrelse(&path);
1348                         // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1349                         if (!
1350                             ((unsigned long long)
1351                              GET_HASH_VALUE(le_key_k_offset
1352                                             (le_key_version(key), key)) == 0
1353                              && (unsigned long long)
1354                              GET_GENERATION_NUMBER(le_key_k_offset
1355                                                    (le_key_version(key),
1356                                                     key)) == 1))
1357                                 reiserfs_warning(th->t_super,
1358                                                  "vs-5355: reiserfs_delete_solid_item: %k not found",
1359                                                  key);
1360                         break;
1361                 }
1362                 if (!tb_init) {
1363                         tb_init = 1;
1364                         item_len = ih_item_len(PATH_PITEM_HEAD(&path));
1365                         init_tb_struct(th, &tb, th->t_super, &path,
1366                                        -(IH_SIZE + item_len));
1367                 }
1368                 quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path));
1369
1370                 retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1371                 if (retval == REPEAT_SEARCH) {
1372                         PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1373                         continue;
1374                 }
1375
1376                 if (retval == CARRY_ON) {
1377                         do_balance(&tb, NULL, NULL, M_DELETE);
1378                         if (inode) {    /* Should we count quota for item? (we don't count quotas for save-links) */
1379 #ifdef REISERQUOTA_DEBUG
1380                                 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1381                                                "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1382                                                quota_cut_bytes, inode->i_uid,
1383                                                key2type(key));
1384 #endif
1385                                 DQUOT_FREE_SPACE_NODIRTY(inode,
1386                                                          quota_cut_bytes);
1387                         }
1388                         break;
1389                 }
1390                 // IO_ERROR, NO_DISK_SPACE, etc
1391                 reiserfs_warning(th->t_super,
1392                                  "vs-5360: reiserfs_delete_solid_item: "
1393                                  "could not delete %K due to fix_nodes failure",
1394                                  &cpu_key);
1395                 unfix_nodes(&tb);
1396                 break;
1397         }
1398
1399         reiserfs_check_path(&path);
1400 }
1401
1402 int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1403                            struct inode *inode)
1404 {
1405         int err;
1406         inode->i_size = 0;
1407         BUG_ON(!th->t_trans_id);
1408
1409         /* for directory this deletes item containing "." and ".." */
1410         err =
1411             reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1412         if (err)
1413                 return err;
1414
1415 #if defined( USE_INODE_GENERATION_COUNTER )
1416         if (!old_format_only(th->t_super)) {
1417                 __le32 *inode_generation;
1418
1419                 inode_generation =
1420                     &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
1421                 *inode_generation =
1422                     cpu_to_le32(le32_to_cpu(*inode_generation) + 1);
1423         }
1424 /* USE_INODE_GENERATION_COUNTER */
1425 #endif
1426         reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1427
1428         return err;
1429 }
1430
1431 static void unmap_buffers(struct page *page, loff_t pos)
1432 {
1433         struct buffer_head *bh;
1434         struct buffer_head *head;
1435         struct buffer_head *next;
1436         unsigned long tail_index;
1437         unsigned long cur_index;
1438
1439         if (page) {
1440                 if (page_has_buffers(page)) {
1441                         tail_index = pos & (PAGE_CACHE_SIZE - 1);
1442                         cur_index = 0;
1443                         head = page_buffers(page);
1444                         bh = head;
1445                         do {
1446                                 next = bh->b_this_page;
1447
1448                                 /* we want to unmap the buffers that contain the tail, and
1449                                  ** all the buffers after it (since the tail must be at the
1450                                  ** end of the file).  We don't want to unmap file data
1451                                  ** before the tail, since it might be dirty and waiting to
1452                                  ** reach disk
1453                                  */
1454                                 cur_index += bh->b_size;
1455                                 if (cur_index > tail_index) {
1456                                         reiserfs_unmap_buffer(bh);
1457                                 }
1458                                 bh = next;
1459                         } while (bh != head);
1460                         if (PAGE_SIZE == bh->b_size) {
1461                                 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1462                         }
1463                 }
1464         }
1465 }
1466
1467 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
1468                                     struct inode *p_s_inode,
1469                                     struct page *page,
1470                                     struct treepath *p_s_path,
1471                                     const struct cpu_key *p_s_item_key,
1472                                     loff_t n_new_file_size, char *p_c_mode)
1473 {
1474         struct super_block *p_s_sb = p_s_inode->i_sb;
1475         int n_block_size = p_s_sb->s_blocksize;
1476         int cut_bytes;
1477         BUG_ON(!th->t_trans_id);
1478         BUG_ON(n_new_file_size != p_s_inode->i_size);
1479
1480         /* the page being sent in could be NULL if there was an i/o error
1481          ** reading in the last block.  The user will hit problems trying to
1482          ** read the file, but for now we just skip the indirect2direct
1483          */
1484         if (atomic_read(&p_s_inode->i_count) > 1 ||
1485             !tail_has_to_be_packed(p_s_inode) ||
1486             !page || (REISERFS_I(p_s_inode)->i_flags & i_nopack_mask)) {
1487                 // leave tail in an unformatted node    
1488                 *p_c_mode = M_SKIP_BALANCING;
1489                 cut_bytes =
1490                     n_block_size - (n_new_file_size & (n_block_size - 1));
1491                 pathrelse(p_s_path);
1492                 return cut_bytes;
1493         }
1494         /* Permorm the conversion to a direct_item. */
1495         /*return indirect_to_direct (p_s_inode, p_s_path, p_s_item_key, n_new_file_size, p_c_mode); */
1496         return indirect2direct(th, p_s_inode, page, p_s_path, p_s_item_key,
1497                                n_new_file_size, p_c_mode);
1498 }
1499
1500 /* we did indirect_to_direct conversion. And we have inserted direct
1501    item successesfully, but there were no disk space to cut unfm
1502    pointer being converted. Therefore we have to delete inserted
1503    direct item(s) */
1504 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
1505                                          struct inode *inode, struct treepath *path)
1506 {
1507         struct cpu_key tail_key;
1508         int tail_len;
1509         int removed;
1510         BUG_ON(!th->t_trans_id);
1511
1512         make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);      // !!!!
1513         tail_key.key_length = 4;
1514
1515         tail_len =
1516             (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1517         while (tail_len) {
1518                 /* look for the last byte of the tail */
1519                 if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1520                     POSITION_NOT_FOUND)
1521                         reiserfs_panic(inode->i_sb,
1522                                        "vs-5615: indirect_to_direct_roll_back: found invalid item");
1523                 RFALSE(path->pos_in_item !=
1524                        ih_item_len(PATH_PITEM_HEAD(path)) - 1,
1525                        "vs-5616: appended bytes found");
1526                 PATH_LAST_POSITION(path)--;
1527
1528                 removed =
1529                     reiserfs_delete_item(th, path, &tail_key, inode,
1530                                          NULL /*unbh not needed */ );
1531                 RFALSE(removed <= 0
1532                        || removed > tail_len,
1533                        "vs-5617: there was tail %d bytes, removed item length %d bytes",
1534                        tail_len, removed);
1535                 tail_len -= removed;
1536                 set_cpu_key_k_offset(&tail_key,
1537                                      cpu_key_k_offset(&tail_key) - removed);
1538         }
1539         reiserfs_warning(inode->i_sb,
1540                          "indirect_to_direct_roll_back: indirect_to_direct conversion has been rolled back due to lack of disk space");
1541         //mark_file_without_tail (inode);
1542         mark_inode_dirty(inode);
1543 }
1544
1545 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1546 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
1547                            struct treepath *p_s_path,
1548                            struct cpu_key *p_s_item_key,
1549                            struct inode *p_s_inode,
1550                            struct page *page, loff_t n_new_file_size)
1551 {
1552         struct super_block *p_s_sb = p_s_inode->i_sb;
1553         /* Every function which is going to call do_balance must first
1554            create a tree_balance structure.  Then it must fill up this
1555            structure by using the init_tb_struct and fix_nodes functions.
1556            After that we can make tree balancing. */
1557         struct tree_balance s_cut_balance;
1558         struct item_head *p_le_ih;
1559         int n_cut_size = 0,     /* Amount to be cut. */
1560             n_ret_value = CARRY_ON, n_removed = 0,      /* Number of the removed unformatted nodes. */
1561             n_is_inode_locked = 0;
1562         char c_mode;            /* Mode of the balance. */
1563         int retval2 = -1;
1564         int quota_cut_bytes;
1565         loff_t tail_pos = 0;
1566
1567         BUG_ON(!th->t_trans_id);
1568
1569         init_tb_struct(th, &s_cut_balance, p_s_inode->i_sb, p_s_path,
1570                        n_cut_size);
1571
1572         /* Repeat this loop until we either cut the item without needing
1573            to balance, or we fix_nodes without schedule occurring */
1574         while (1) {
1575                 /* Determine the balance mode, position of the first byte to
1576                    be cut, and size to be cut.  In case of the indirect item
1577                    free unformatted nodes which are pointed to by the cut
1578                    pointers. */
1579
1580                 c_mode =
1581                     prepare_for_delete_or_cut(th, p_s_inode, p_s_path,
1582                                               p_s_item_key, &n_removed,
1583                                               &n_cut_size, n_new_file_size);
1584                 if (c_mode == M_CONVERT) {
1585                         /* convert last unformatted node to direct item or leave
1586                            tail in the unformatted node */
1587                         RFALSE(n_ret_value != CARRY_ON,
1588                                "PAP-5570: can not convert twice");
1589
1590                         n_ret_value =
1591                             maybe_indirect_to_direct(th, p_s_inode, page,
1592                                                      p_s_path, p_s_item_key,
1593                                                      n_new_file_size, &c_mode);
1594                         if (c_mode == M_SKIP_BALANCING)
1595                                 /* tail has been left in the unformatted node */
1596                                 return n_ret_value;
1597
1598                         n_is_inode_locked = 1;
1599
1600                         /* removing of last unformatted node will change value we
1601                            have to return to truncate. Save it */
1602                         retval2 = n_ret_value;
1603                         /*retval2 = p_s_sb->s_blocksize - (n_new_file_size & (p_s_sb->s_blocksize - 1)); */
1604
1605                         /* So, we have performed the first part of the conversion:
1606                            inserting the new direct item.  Now we are removing the
1607                            last unformatted node pointer. Set key to search for
1608                            it. */
1609                         set_cpu_key_k_type(p_s_item_key, TYPE_INDIRECT);
1610                         p_s_item_key->key_length = 4;
1611                         n_new_file_size -=
1612                             (n_new_file_size & (p_s_sb->s_blocksize - 1));
1613                         tail_pos = n_new_file_size;
1614                         set_cpu_key_k_offset(p_s_item_key, n_new_file_size + 1);
1615                         if (search_for_position_by_key
1616                             (p_s_sb, p_s_item_key,
1617                              p_s_path) == POSITION_NOT_FOUND) {
1618                                 print_block(PATH_PLAST_BUFFER(p_s_path), 3,
1619                                             PATH_LAST_POSITION(p_s_path) - 1,
1620                                             PATH_LAST_POSITION(p_s_path) + 1);
1621                                 reiserfs_panic(p_s_sb,
1622                                                "PAP-5580: reiserfs_cut_from_item: item to convert does not exist (%K)",
1623                                                p_s_item_key);
1624                         }
1625                         continue;
1626                 }
1627                 if (n_cut_size == 0) {
1628                         pathrelse(p_s_path);
1629                         return 0;
1630                 }
1631
1632                 s_cut_balance.insert_size[0] = n_cut_size;
1633
1634                 n_ret_value = fix_nodes(c_mode, &s_cut_balance, NULL, NULL);
1635                 if (n_ret_value != REPEAT_SEARCH)
1636                         break;
1637
1638                 PROC_INFO_INC(p_s_sb, cut_from_item_restarted);
1639
1640                 n_ret_value =
1641                     search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1642                 if (n_ret_value == POSITION_FOUND)
1643                         continue;
1644
1645                 reiserfs_warning(p_s_sb,
1646                                  "PAP-5610: reiserfs_cut_from_item: item %K not found",
1647                                  p_s_item_key);
1648                 unfix_nodes(&s_cut_balance);
1649                 return (n_ret_value == IO_ERROR) ? -EIO : -ENOENT;
1650         }                       /* while */
1651
1652         // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1653         if (n_ret_value != CARRY_ON) {
1654                 if (n_is_inode_locked) {
1655                         // FIXME: this seems to be not needed: we are always able
1656                         // to cut item
1657                         indirect_to_direct_roll_back(th, p_s_inode, p_s_path);
1658                 }
1659                 if (n_ret_value == NO_DISK_SPACE)
1660                         reiserfs_warning(p_s_sb, "NO_DISK_SPACE");
1661                 unfix_nodes(&s_cut_balance);
1662                 return -EIO;
1663         }
1664
1665         /* go ahead and perform balancing */
1666
1667         RFALSE(c_mode == M_PASTE || c_mode == M_INSERT, "invalid mode");
1668
1669         /* Calculate number of bytes that need to be cut from the item. */
1670         quota_cut_bytes =
1671             (c_mode ==
1672              M_DELETE) ? ih_item_len(get_ih(p_s_path)) : -s_cut_balance.
1673             insert_size[0];
1674         if (retval2 == -1)
1675                 n_ret_value = calc_deleted_bytes_number(&s_cut_balance, c_mode);
1676         else
1677                 n_ret_value = retval2;
1678
1679         /* For direct items, we only change the quota when deleting the last
1680          ** item.
1681          */
1682         p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path);
1683         if (!S_ISLNK(p_s_inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1684                 if (c_mode == M_DELETE &&
1685                     (le_ih_k_offset(p_le_ih) & (p_s_sb->s_blocksize - 1)) ==
1686                     1) {
1687                         // FIXME: this is to keep 3.5 happy
1688                         REISERFS_I(p_s_inode)->i_first_direct_byte = U32_MAX;
1689                         quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1690                 } else {
1691                         quota_cut_bytes = 0;
1692                 }
1693         }
1694 #ifdef CONFIG_REISERFS_CHECK
1695         if (n_is_inode_locked) {
1696                 struct item_head *le_ih =
1697                     PATH_PITEM_HEAD(s_cut_balance.tb_path);
1698                 /* we are going to complete indirect2direct conversion. Make
1699                    sure, that we exactly remove last unformatted node pointer
1700                    of the item */
1701                 if (!is_indirect_le_ih(le_ih))
1702                         reiserfs_panic(p_s_sb,
1703                                        "vs-5652: reiserfs_cut_from_item: "
1704                                        "item must be indirect %h", le_ih);
1705
1706                 if (c_mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1707                         reiserfs_panic(p_s_sb,
1708                                        "vs-5653: reiserfs_cut_from_item: "
1709                                        "completing indirect2direct conversion indirect item %h "
1710                                        "being deleted must be of 4 byte long",
1711                                        le_ih);
1712
1713                 if (c_mode == M_CUT
1714                     && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1715                         reiserfs_panic(p_s_sb,
1716                                        "vs-5654: reiserfs_cut_from_item: "
1717                                        "can not complete indirect2direct conversion of %h (CUT, insert_size==%d)",
1718                                        le_ih, s_cut_balance.insert_size[0]);
1719                 }
1720                 /* it would be useful to make sure, that right neighboring
1721                    item is direct item of this file */
1722         }
1723 #endif
1724
1725         do_balance(&s_cut_balance, NULL, NULL, c_mode);
1726         if (n_is_inode_locked) {
1727                 /* we've done an indirect->direct conversion.  when the data block
1728                  ** was freed, it was removed from the list of blocks that must
1729                  ** be flushed before the transaction commits, make sure to
1730                  ** unmap and invalidate it
1731                  */
1732                 unmap_buffers(page, tail_pos);
1733                 REISERFS_I(p_s_inode)->i_flags &= ~i_pack_on_close_mask;
1734         }
1735 #ifdef REISERQUOTA_DEBUG
1736         reiserfs_debug(p_s_inode->i_sb, REISERFS_DEBUG_CODE,
1737                        "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1738                        quota_cut_bytes, p_s_inode->i_uid, '?');
1739 #endif
1740         DQUOT_FREE_SPACE_NODIRTY(p_s_inode, quota_cut_bytes);
1741         return n_ret_value;
1742 }
1743
1744 static void truncate_directory(struct reiserfs_transaction_handle *th,
1745                                struct inode *inode)
1746 {
1747         BUG_ON(!th->t_trans_id);
1748         if (inode->i_nlink)
1749                 reiserfs_warning(inode->i_sb,
1750                                  "vs-5655: truncate_directory: link count != 0");
1751
1752         set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1753         set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1754         reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1755         reiserfs_update_sd(th, inode);
1756         set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1757         set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
1758 }
1759
1760 /* Truncate file to the new size. Note, this must be called with a transaction
1761    already started */
1762 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th, struct inode *p_s_inode,       /* ->i_size contains new
1763                                                                                                    size */
1764                          struct page *page,     /* up to date for last block */
1765                          int update_timestamps  /* when it is called by
1766                                                    file_release to convert
1767                                                    the tail - no timestamps
1768                                                    should be updated */
1769     )
1770 {
1771         INITIALIZE_PATH(s_search_path); /* Path to the current object item. */
1772         struct item_head *p_le_ih;      /* Pointer to an item header. */
1773         struct cpu_key s_item_key;      /* Key to search for a previous file item. */
1774         loff_t n_file_size,     /* Old file size. */
1775          n_new_file_size;       /* New file size. */
1776         int n_deleted;          /* Number of deleted or truncated bytes. */
1777         int retval;
1778         int err = 0;
1779
1780         BUG_ON(!th->t_trans_id);
1781         if (!
1782             (S_ISREG(p_s_inode->i_mode) || S_ISDIR(p_s_inode->i_mode)
1783              || S_ISLNK(p_s_inode->i_mode)))
1784                 return 0;
1785
1786         if (S_ISDIR(p_s_inode->i_mode)) {
1787                 // deletion of directory - no need to update timestamps
1788                 truncate_directory(th, p_s_inode);
1789                 return 0;
1790         }
1791
1792         /* Get new file size. */
1793         n_new_file_size = p_s_inode->i_size;
1794
1795         // FIXME: note, that key type is unimportant here
1796         make_cpu_key(&s_item_key, p_s_inode, max_reiserfs_offset(p_s_inode),
1797                      TYPE_DIRECT, 3);
1798
1799         retval =
1800             search_for_position_by_key(p_s_inode->i_sb, &s_item_key,
1801                                        &s_search_path);
1802         if (retval == IO_ERROR) {
1803                 reiserfs_warning(p_s_inode->i_sb,
1804                                  "vs-5657: reiserfs_do_truncate: "
1805                                  "i/o failure occurred trying to truncate %K",
1806                                  &s_item_key);
1807                 err = -EIO;
1808                 goto out;
1809         }
1810         if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1811                 reiserfs_warning(p_s_inode->i_sb,
1812                                  "PAP-5660: reiserfs_do_truncate: "
1813                                  "wrong result %d of search for %K", retval,
1814                                  &s_item_key);
1815
1816                 err = -EIO;
1817                 goto out;
1818         }
1819
1820         s_search_path.pos_in_item--;
1821
1822         /* Get real file size (total length of all file items) */
1823         p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1824         if (is_statdata_le_ih(p_le_ih))
1825                 n_file_size = 0;
1826         else {
1827                 loff_t offset = le_ih_k_offset(p_le_ih);
1828                 int bytes =
1829                     op_bytes_number(p_le_ih, p_s_inode->i_sb->s_blocksize);
1830
1831                 /* this may mismatch with real file size: if last direct item
1832                    had no padding zeros and last unformatted node had no free
1833                    space, this file would have this file size */
1834                 n_file_size = offset + bytes - 1;
1835         }
1836         /*
1837          * are we doing a full truncate or delete, if so
1838          * kick in the reada code
1839          */
1840         if (n_new_file_size == 0)
1841                 s_search_path.reada = PATH_READA | PATH_READA_BACK;
1842
1843         if (n_file_size == 0 || n_file_size < n_new_file_size) {
1844                 goto update_and_out;
1845         }
1846
1847         /* Update key to search for the last file item. */
1848         set_cpu_key_k_offset(&s_item_key, n_file_size);
1849
1850         do {
1851                 /* Cut or delete file item. */
1852                 n_deleted =
1853                     reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
1854                                            p_s_inode, page, n_new_file_size);
1855                 if (n_deleted < 0) {
1856                         reiserfs_warning(p_s_inode->i_sb,
1857                                          "vs-5665: reiserfs_do_truncate: reiserfs_cut_from_item failed");
1858                         reiserfs_check_path(&s_search_path);
1859                         return 0;
1860                 }
1861
1862                 RFALSE(n_deleted > n_file_size,
1863                        "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1864                        n_deleted, n_file_size, &s_item_key);
1865
1866                 /* Change key to search the last file item. */
1867                 n_file_size -= n_deleted;
1868
1869                 set_cpu_key_k_offset(&s_item_key, n_file_size);
1870
1871                 /* While there are bytes to truncate and previous file item is presented in the tree. */
1872
1873                 /*
1874                  ** This loop could take a really long time, and could log 
1875                  ** many more blocks than a transaction can hold.  So, we do a polite
1876                  ** journal end here, and if the transaction needs ending, we make
1877                  ** sure the file is consistent before ending the current trans
1878                  ** and starting a new one
1879                  */
1880                 if (journal_transaction_should_end(th, 0) ||
1881                     reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1882                         int orig_len_alloc = th->t_blocks_allocated;
1883                         decrement_counters_in_path(&s_search_path);
1884
1885                         if (update_timestamps) {
1886                                 p_s_inode->i_mtime = p_s_inode->i_ctime =
1887                                     CURRENT_TIME_SEC;
1888                         }
1889                         reiserfs_update_sd(th, p_s_inode);
1890
1891                         err = journal_end(th, p_s_inode->i_sb, orig_len_alloc);
1892                         if (err)
1893                                 goto out;
1894                         err = journal_begin(th, p_s_inode->i_sb,
1895                                             JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ;
1896                         if (err)
1897                                 goto out;
1898                         reiserfs_update_inode_transaction(p_s_inode);
1899                 }
1900         } while (n_file_size > ROUND_UP(n_new_file_size) &&
1901                  search_for_position_by_key(p_s_inode->i_sb, &s_item_key,
1902                                             &s_search_path) == POSITION_FOUND);
1903
1904         RFALSE(n_file_size > ROUND_UP(n_new_file_size),
1905                "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1906                n_new_file_size, n_file_size, s_item_key.on_disk_key.k_objectid);
1907
1908       update_and_out:
1909         if (update_timestamps) {
1910                 // this is truncate, not file closing
1911                 p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME_SEC;
1912         }
1913         reiserfs_update_sd(th, p_s_inode);
1914
1915       out:
1916         pathrelse(&s_search_path);
1917         return err;
1918 }
1919
1920 #ifdef CONFIG_REISERFS_CHECK
1921 // this makes sure, that we __append__, not overwrite or add holes
1922 static void check_research_for_paste(struct treepath *path,
1923                                      const struct cpu_key *p_s_key)
1924 {
1925         struct item_head *found_ih = get_ih(path);
1926
1927         if (is_direct_le_ih(found_ih)) {
1928                 if (le_ih_k_offset(found_ih) +
1929                     op_bytes_number(found_ih,
1930                                     get_last_bh(path)->b_size) !=
1931                     cpu_key_k_offset(p_s_key)
1932                     || op_bytes_number(found_ih,
1933                                        get_last_bh(path)->b_size) !=
1934                     pos_in_item(path))
1935                         reiserfs_panic(NULL,
1936                                        "PAP-5720: check_research_for_paste: "
1937                                        "found direct item %h or position (%d) does not match to key %K",
1938                                        found_ih, pos_in_item(path), p_s_key);
1939         }
1940         if (is_indirect_le_ih(found_ih)) {
1941                 if (le_ih_k_offset(found_ih) +
1942                     op_bytes_number(found_ih,
1943                                     get_last_bh(path)->b_size) !=
1944                     cpu_key_k_offset(p_s_key)
1945                     || I_UNFM_NUM(found_ih) != pos_in_item(path)
1946                     || get_ih_free_space(found_ih) != 0)
1947                         reiserfs_panic(NULL,
1948                                        "PAP-5730: check_research_for_paste: "
1949                                        "found indirect item (%h) or position (%d) does not match to key (%K)",
1950                                        found_ih, pos_in_item(path), p_s_key);
1951         }
1952 }
1953 #endif                          /* config reiserfs check */
1954
1955 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1956 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_search_path,  /* Path to the pasted item.          */
1957                              const struct cpu_key *p_s_key,     /* Key to search for the needed item. */
1958                              struct inode *inode,       /* Inode item belongs to */
1959                              const char *p_c_body,      /* Pointer to the bytes to paste.    */
1960                              int n_pasted_size)
1961 {                               /* Size of pasted bytes.             */
1962         struct tree_balance s_paste_balance;
1963         int retval;
1964         int fs_gen;
1965
1966         BUG_ON(!th->t_trans_id);
1967
1968         fs_gen = get_generation(inode->i_sb);
1969
1970 #ifdef REISERQUOTA_DEBUG
1971         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1972                        "reiserquota paste_into_item(): allocating %u id=%u type=%c",
1973                        n_pasted_size, inode->i_uid,
1974                        key2type(&(p_s_key->on_disk_key)));
1975 #endif
1976
1977         if (DQUOT_ALLOC_SPACE_NODIRTY(inode, n_pasted_size)) {
1978                 pathrelse(p_s_search_path);
1979                 return -EDQUOT;
1980         }
1981         init_tb_struct(th, &s_paste_balance, th->t_super, p_s_search_path,
1982                        n_pasted_size);
1983 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1984         s_paste_balance.key = p_s_key->on_disk_key;
1985 #endif
1986
1987         /* DQUOT_* can schedule, must check before the fix_nodes */
1988         if (fs_changed(fs_gen, inode->i_sb)) {
1989                 goto search_again;
1990         }
1991
1992         while ((retval =
1993                 fix_nodes(M_PASTE, &s_paste_balance, NULL,
1994                           p_c_body)) == REPEAT_SEARCH) {
1995               search_again:
1996                 /* file system changed while we were in the fix_nodes */
1997                 PROC_INFO_INC(th->t_super, paste_into_item_restarted);
1998                 retval =
1999                     search_for_position_by_key(th->t_super, p_s_key,
2000                                                p_s_search_path);
2001                 if (retval == IO_ERROR) {
2002                         retval = -EIO;
2003                         goto error_out;
2004                 }
2005                 if (retval == POSITION_FOUND) {
2006                         reiserfs_warning(inode->i_sb,
2007                                          "PAP-5710: reiserfs_paste_into_item: entry or pasted byte (%K) exists",
2008                                          p_s_key);
2009                         retval = -EEXIST;
2010                         goto error_out;
2011                 }
2012 #ifdef CONFIG_REISERFS_CHECK
2013                 check_research_for_paste(p_s_search_path, p_s_key);
2014 #endif
2015         }
2016
2017         /* Perform balancing after all resources are collected by fix_nodes, and
2018            accessing them will not risk triggering schedule. */
2019         if (retval == CARRY_ON) {
2020                 do_balance(&s_paste_balance, NULL /*ih */ , p_c_body, M_PASTE);
2021                 return 0;
2022         }
2023         retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2024       error_out:
2025         /* this also releases the path */
2026         unfix_nodes(&s_paste_balance);
2027 #ifdef REISERQUOTA_DEBUG
2028         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2029                        "reiserquota paste_into_item(): freeing %u id=%u type=%c",
2030                        n_pasted_size, inode->i_uid,
2031                        key2type(&(p_s_key->on_disk_key)));
2032 #endif
2033         DQUOT_FREE_SPACE_NODIRTY(inode, n_pasted_size);
2034         return retval;
2035 }
2036
2037 /* Insert new item into the buffer at the path. */
2038 int reiserfs_insert_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_path,     /* Path to the inserteded item.         */
2039                          const struct cpu_key *key, struct item_head *p_s_ih,   /* Pointer to the item header to insert. */
2040                          struct inode *inode, const char *p_c_body)
2041 {                               /* Pointer to the bytes to insert.      */
2042         struct tree_balance s_ins_balance;
2043         int retval;
2044         int fs_gen = 0;
2045         int quota_bytes = 0;
2046
2047         BUG_ON(!th->t_trans_id);
2048
2049         if (inode) {            /* Do we count quotas for item? */
2050                 fs_gen = get_generation(inode->i_sb);
2051                 quota_bytes = ih_item_len(p_s_ih);
2052
2053                 /* hack so the quota code doesn't have to guess if the file has
2054                  ** a tail, links are always tails, so there's no guessing needed
2055                  */
2056                 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_s_ih)) {
2057                         quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
2058                 }
2059 #ifdef REISERQUOTA_DEBUG
2060                 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2061                                "reiserquota insert_item(): allocating %u id=%u type=%c",
2062                                quota_bytes, inode->i_uid, head2type(p_s_ih));
2063 #endif
2064                 /* We can't dirty inode here. It would be immediately written but
2065                  * appropriate stat item isn't inserted yet... */
2066                 if (DQUOT_ALLOC_SPACE_NODIRTY(inode, quota_bytes)) {
2067                         pathrelse(p_s_path);
2068                         return -EDQUOT;
2069                 }
2070         }
2071         init_tb_struct(th, &s_ins_balance, th->t_super, p_s_path,
2072                        IH_SIZE + ih_item_len(p_s_ih));
2073 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2074         s_ins_balance.key = key->on_disk_key;
2075 #endif
2076         /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2077         if (inode && fs_changed(fs_gen, inode->i_sb)) {
2078                 goto search_again;
2079         }
2080
2081         while ((retval =
2082                 fix_nodes(M_INSERT, &s_ins_balance, p_s_ih,
2083                           p_c_body)) == REPEAT_SEARCH) {
2084               search_again:
2085                 /* file system changed while we were in the fix_nodes */
2086                 PROC_INFO_INC(th->t_super, insert_item_restarted);
2087                 retval = search_item(th->t_super, key, p_s_path);
2088                 if (retval == IO_ERROR) {
2089                         retval = -EIO;
2090                         goto error_out;
2091                 }
2092                 if (retval == ITEM_FOUND) {
2093                         reiserfs_warning(th->t_super,
2094                                          "PAP-5760: reiserfs_insert_item: "
2095                                          "key %K already exists in the tree",
2096                                          key);
2097                         retval = -EEXIST;
2098                         goto error_out;
2099                 }
2100         }
2101
2102         /* make balancing after all resources will be collected at a time */
2103         if (retval == CARRY_ON) {
2104                 do_balance(&s_ins_balance, p_s_ih, p_c_body, M_INSERT);
2105                 return 0;
2106         }
2107
2108         retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2109       error_out:
2110         /* also releases the path */
2111         unfix_nodes(&s_ins_balance);
2112 #ifdef REISERQUOTA_DEBUG
2113         reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2114                        "reiserquota insert_item(): freeing %u id=%u type=%c",
2115                        quota_bytes, inode->i_uid, head2type(p_s_ih));
2116 #endif
2117         if (inode)
2118                 DQUOT_FREE_SPACE_NODIRTY(inode, quota_bytes);
2119         return retval;
2120 }