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