2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
6 * Written by Anatoly P. Pinchuk pap@namesys.botik.ru
7 * Programm System Institute
8 * Pereslavl-Zalessky Russia
12 * This file contains functions dealing with S+tree
27 * pathrelse_and_restore
31 * search_for_position_by_key
33 * prepare_for_direct_item
34 * prepare_for_direntry_item
35 * prepare_for_delete_or_cut
36 * calc_deleted_bytes_number
39 * reiserfs_delete_item
40 * reiserfs_delete_solid_item
41 * reiserfs_delete_object
42 * maybe_indirect_to_direct
43 * indirect_to_direct_roll_back
44 * reiserfs_cut_from_item
46 * reiserfs_do_truncate
47 * reiserfs_paste_into_item
48 * reiserfs_insert_item
51 #include <linux/time.h>
52 #include <linux/string.h>
53 #include <linux/pagemap.h>
54 #include <linux/reiserfs_fs.h>
55 #include <linux/buffer_head.h>
56 #include <linux/quotaops.h>
58 /* Does the buffer contain a disk block which is in the tree. */
59 inline int B_IS_IN_TREE(const struct buffer_head
*bh
)
62 RFALSE(B_LEVEL(bh
) > MAX_HEIGHT
,
63 "PAP-1010: block (%b) has too big level (%z)", bh
, bh
);
65 return (B_LEVEL(bh
) != FREE_LEVEL
);
69 // to gets item head in le form
71 inline void copy_item_head(struct item_head
*to
,
72 const struct item_head
*from
)
74 memcpy(to
, from
, IH_SIZE
);
77 /* k1 is pointer to on-disk structure which is stored in little-endian
78 form. k2 is pointer to cpu variable. For key of items of the same
79 object this returns 0.
80 Returns: -1 if key1 < key2
83 inline int comp_short_keys(const struct reiserfs_key
*le_key
,
84 const struct cpu_key
*cpu_key
)
87 n
= le32_to_cpu(le_key
->k_dir_id
);
88 if (n
< cpu_key
->on_disk_key
.k_dir_id
)
90 if (n
> cpu_key
->on_disk_key
.k_dir_id
)
92 n
= le32_to_cpu(le_key
->k_objectid
);
93 if (n
< cpu_key
->on_disk_key
.k_objectid
)
95 if (n
> cpu_key
->on_disk_key
.k_objectid
)
100 /* k1 is pointer to on-disk structure which is stored in little-endian
101 form. k2 is pointer to cpu variable.
102 Compare keys using all 4 key fields.
103 Returns: -1 if key1 < key2 0
104 if key1 = key2 1 if key1 > key2 */
105 static inline int comp_keys(const struct reiserfs_key
*le_key
,
106 const struct cpu_key
*cpu_key
)
110 retval
= comp_short_keys(le_key
, cpu_key
);
113 if (le_key_k_offset(le_key_version(le_key
), le_key
) <
114 cpu_key_k_offset(cpu_key
))
116 if (le_key_k_offset(le_key_version(le_key
), le_key
) >
117 cpu_key_k_offset(cpu_key
))
120 if (cpu_key
->key_length
== 3)
123 /* this part is needed only when tail conversion is in progress */
124 if (le_key_k_type(le_key_version(le_key
), le_key
) <
125 cpu_key_k_type(cpu_key
))
128 if (le_key_k_type(le_key_version(le_key
), le_key
) >
129 cpu_key_k_type(cpu_key
))
135 inline int comp_short_le_keys(const struct reiserfs_key
*key1
,
136 const struct reiserfs_key
*key2
)
138 __u32
*k1_u32
, *k2_u32
;
139 int key_length
= REISERFS_SHORT_KEY_LEN
;
141 k1_u32
= (__u32
*) key1
;
142 k2_u32
= (__u32
*) key2
;
143 for (; key_length
--; ++k1_u32
, ++k2_u32
) {
144 if (le32_to_cpu(*k1_u32
) < le32_to_cpu(*k2_u32
))
146 if (le32_to_cpu(*k1_u32
) > le32_to_cpu(*k2_u32
))
152 inline void le_key2cpu_key(struct cpu_key
*to
, const struct reiserfs_key
*from
)
155 to
->on_disk_key
.k_dir_id
= le32_to_cpu(from
->k_dir_id
);
156 to
->on_disk_key
.k_objectid
= le32_to_cpu(from
->k_objectid
);
158 // find out version of the key
159 version
= le_key_version(from
);
160 to
->version
= version
;
161 to
->on_disk_key
.k_offset
= le_key_k_offset(version
, from
);
162 to
->on_disk_key
.k_type
= le_key_k_type(version
, from
);
165 // this does not say which one is bigger, it only returns 1 if keys
166 // are not equal, 0 otherwise
167 inline int comp_le_keys(const struct reiserfs_key
*k1
,
168 const struct reiserfs_key
*k2
)
170 return memcmp(k1
, k2
, sizeof(struct reiserfs_key
));
173 /**************************************************************************
174 * Binary search toolkit function *
175 * Search for an item in the array by the item key *
176 * Returns: 1 if found, 0 if not found; *
177 * *pos = number of the searched element if found, else the *
178 * number of the first element that is larger than key. *
179 **************************************************************************/
180 /* For those not familiar with binary search: lbound is the leftmost item that it
181 could be, rbound the rightmost item that it could be. We examine the item
182 halfway between lbound and rbound, and that tells us either that we can increase
183 lbound, or decrease rbound, or that we have found it, or if lbound <= rbound that
184 there are no possible items, and we have not found it. With each examination we
185 cut the number of possible items it could be by one more than half rounded down,
187 static inline int bin_search(const void *key
, /* Key to search for. */
188 const void *base
, /* First item in the array. */
189 int num
, /* Number of items in the array. */
190 int width
, /* Item size in the array.
191 searched. Lest the reader be
192 confused, note that this is crafted
193 as a general function, and when it
194 is applied specifically to the array
195 of item headers in a node, width
196 is actually the item header size not
198 int *pos
/* Number of the searched for element. */
201 int rbound
, lbound
, j
;
203 for (j
= ((rbound
= num
- 1) + (lbound
= 0)) / 2;
204 lbound
<= rbound
; j
= (rbound
+ lbound
) / 2)
206 ((struct reiserfs_key
*)((char *)base
+ j
* width
),
207 (struct cpu_key
*)key
)) {
216 return ITEM_FOUND
; /* Key found in the array. */
219 /* bin_search did not find given key, it returns position of key,
220 that is minimal and greater than the given one. */
222 return ITEM_NOT_FOUND
;
226 /* Minimal possible key. It is never in the tree. */
227 const struct reiserfs_key MIN_KEY
= { 0, 0, {{0, 0},} };
229 /* Maximal possible key. It is never in the tree. */
230 static const struct reiserfs_key MAX_KEY
= {
231 __constant_cpu_to_le32(0xffffffff),
232 __constant_cpu_to_le32(0xffffffff),
233 {{__constant_cpu_to_le32(0xffffffff),
234 __constant_cpu_to_le32(0xffffffff)},}
237 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
238 of the path, and going upwards. We must check the path's validity at each step. If the key is not in
239 the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
240 case we return a special key, either MIN_KEY or MAX_KEY. */
241 static inline const struct reiserfs_key
*get_lkey(const struct treepath
*chk_path
,
242 const struct super_block
*sb
)
244 int position
, path_offset
= chk_path
->path_length
;
245 struct buffer_head
*parent
;
247 RFALSE(path_offset
< FIRST_PATH_ELEMENT_OFFSET
,
248 "PAP-5010: invalid offset in the path");
250 /* While not higher in path than first element. */
251 while (path_offset
-- > FIRST_PATH_ELEMENT_OFFSET
) {
253 RFALSE(!buffer_uptodate
254 (PATH_OFFSET_PBUFFER(chk_path
, path_offset
)),
255 "PAP-5020: parent is not uptodate");
257 /* Parent at the path is not in the tree now. */
260 PATH_OFFSET_PBUFFER(chk_path
, path_offset
)))
262 /* Check whether position in the parent is correct. */
264 PATH_OFFSET_POSITION(chk_path
,
268 /* Check whether parent at the path really points to the child. */
269 if (B_N_CHILD_NUM(parent
, position
) !=
270 PATH_OFFSET_PBUFFER(chk_path
,
271 path_offset
+ 1)->b_blocknr
)
273 /* Return delimiting key if position in the parent is not equal to zero. */
275 return B_N_PDELIM_KEY(parent
, position
- 1);
277 /* Return MIN_KEY if we are in the root of the buffer tree. */
278 if (PATH_OFFSET_PBUFFER(chk_path
, FIRST_PATH_ELEMENT_OFFSET
)->
279 b_blocknr
== SB_ROOT_BLOCK(sb
))
284 /* Get delimiting key of the buffer at the path and its right neighbor. */
285 inline const struct reiserfs_key
*get_rkey(const struct treepath
*chk_path
,
286 const struct super_block
*sb
)
288 int position
, path_offset
= chk_path
->path_length
;
289 struct buffer_head
*parent
;
291 RFALSE(path_offset
< FIRST_PATH_ELEMENT_OFFSET
,
292 "PAP-5030: invalid offset in the path");
294 while (path_offset
-- > FIRST_PATH_ELEMENT_OFFSET
) {
296 RFALSE(!buffer_uptodate
297 (PATH_OFFSET_PBUFFER(chk_path
, path_offset
)),
298 "PAP-5040: parent is not uptodate");
300 /* Parent at the path is not in the tree now. */
303 PATH_OFFSET_PBUFFER(chk_path
, path_offset
)))
305 /* Check whether position in the parent is correct. */
307 PATH_OFFSET_POSITION(chk_path
,
311 /* Check whether parent at the path really points to the child. */
312 if (B_N_CHILD_NUM(parent
, position
) !=
313 PATH_OFFSET_PBUFFER(chk_path
,
314 path_offset
+ 1)->b_blocknr
)
316 /* Return delimiting key if position in the parent is not the last one. */
317 if (position
!= B_NR_ITEMS(parent
))
318 return B_N_PDELIM_KEY(parent
, position
);
320 /* Return MAX_KEY if we are in the root of the buffer tree. */
321 if (PATH_OFFSET_PBUFFER(chk_path
, FIRST_PATH_ELEMENT_OFFSET
)->
322 b_blocknr
== SB_ROOT_BLOCK(sb
))
327 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
328 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
329 the path. These delimiting keys are stored at least one level above that buffer in the tree. If the
330 buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
331 this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
332 static inline int key_in_buffer(struct treepath
*chk_path
, /* Path which should be checked. */
333 const struct cpu_key
*key
, /* Key which should be checked. */
334 struct super_block
*sb
338 RFALSE(!key
|| chk_path
->path_length
< FIRST_PATH_ELEMENT_OFFSET
339 || chk_path
->path_length
> MAX_HEIGHT
,
340 "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
341 key
, chk_path
->path_length
);
342 RFALSE(!PATH_PLAST_BUFFER(chk_path
)->b_bdev
,
343 "PAP-5060: device must not be NODEV");
345 if (comp_keys(get_lkey(chk_path
, sb
), key
) == 1)
346 /* left delimiting key is bigger, that the key we look for */
348 /* if ( comp_keys(key, get_rkey(chk_path, sb)) != -1 ) */
349 if (comp_keys(get_rkey(chk_path
, sb
), key
) != 1)
350 /* key must be less than right delimitiing key */
355 int reiserfs_check_path(struct treepath
*p
)
357 RFALSE(p
->path_length
!= ILLEGAL_PATH_ELEMENT_OFFSET
,
358 "path not properly relsed");
362 /* Drop the reference to each buffer in a path and restore
363 * dirty bits clean when preparing the buffer for the log.
364 * This version should only be called from fix_nodes() */
365 void pathrelse_and_restore(struct super_block
*sb
,
366 struct treepath
*search_path
)
368 int path_offset
= search_path
->path_length
;
370 RFALSE(path_offset
< ILLEGAL_PATH_ELEMENT_OFFSET
,
371 "clm-4000: invalid path offset");
373 while (path_offset
> ILLEGAL_PATH_ELEMENT_OFFSET
) {
374 struct buffer_head
*bh
;
375 bh
= PATH_OFFSET_PBUFFER(search_path
, path_offset
--);
376 reiserfs_restore_prepared_buffer(sb
, bh
);
379 search_path
->path_length
= ILLEGAL_PATH_ELEMENT_OFFSET
;
382 /* Drop the reference to each buffer in a path */
383 void pathrelse(struct treepath
*search_path
)
385 int path_offset
= search_path
->path_length
;
387 RFALSE(path_offset
< ILLEGAL_PATH_ELEMENT_OFFSET
,
388 "PAP-5090: invalid path offset");
390 while (path_offset
> ILLEGAL_PATH_ELEMENT_OFFSET
)
391 brelse(PATH_OFFSET_PBUFFER(search_path
, path_offset
--));
393 search_path
->path_length
= ILLEGAL_PATH_ELEMENT_OFFSET
;
396 static int is_leaf(char *buf
, int blocksize
, struct buffer_head
*bh
)
398 struct block_head
*blkh
;
399 struct item_head
*ih
;
405 blkh
= (struct block_head
*)buf
;
406 if (blkh_level(blkh
) != DISK_LEAF_NODE_LEVEL
) {
407 reiserfs_warning(NULL
, "reiserfs-5080",
408 "this should be caught earlier");
412 nr
= blkh_nr_item(blkh
);
413 if (nr
< 1 || nr
> ((blocksize
- BLKH_SIZE
) / (IH_SIZE
+ MIN_ITEM_LEN
))) {
414 /* item number is too big or too small */
415 reiserfs_warning(NULL
, "reiserfs-5081",
416 "nr_item seems wrong: %z", bh
);
419 ih
= (struct item_head
*)(buf
+ BLKH_SIZE
) + nr
- 1;
420 used_space
= BLKH_SIZE
+ IH_SIZE
* nr
+ (blocksize
- ih_location(ih
));
421 if (used_space
!= blocksize
- blkh_free_space(blkh
)) {
422 /* free space does not match to calculated amount of use space */
423 reiserfs_warning(NULL
, "reiserfs-5082",
424 "free space seems wrong: %z", bh
);
429 /* check tables of item heads */
430 ih
= (struct item_head
*)(buf
+ BLKH_SIZE
);
431 prev_location
= blocksize
;
432 for (i
= 0; i
< nr
; i
++, ih
++) {
433 if (le_ih_k_type(ih
) == TYPE_ANY
) {
434 reiserfs_warning(NULL
, "reiserfs-5083",
435 "wrong item type for item %h",
439 if (ih_location(ih
) >= blocksize
440 || ih_location(ih
) < IH_SIZE
* nr
) {
441 reiserfs_warning(NULL
, "reiserfs-5084",
442 "item location seems wrong: %h",
446 if (ih_item_len(ih
) < 1
447 || ih_item_len(ih
) > MAX_ITEM_LEN(blocksize
)) {
448 reiserfs_warning(NULL
, "reiserfs-5085",
449 "item length seems wrong: %h",
453 if (prev_location
- ih_location(ih
) != ih_item_len(ih
)) {
454 reiserfs_warning(NULL
, "reiserfs-5086",
455 "item location seems wrong "
456 "(second one): %h", ih
);
459 prev_location
= ih_location(ih
);
462 // one may imagine much more checks
466 /* returns 1 if buf looks like an internal node, 0 otherwise */
467 static int is_internal(char *buf
, int blocksize
, struct buffer_head
*bh
)
469 struct block_head
*blkh
;
473 blkh
= (struct block_head
*)buf
;
474 nr
= blkh_level(blkh
);
475 if (nr
<= DISK_LEAF_NODE_LEVEL
|| nr
> MAX_HEIGHT
) {
476 /* this level is not possible for internal nodes */
477 reiserfs_warning(NULL
, "reiserfs-5087",
478 "this should be caught earlier");
482 nr
= blkh_nr_item(blkh
);
483 if (nr
> (blocksize
- BLKH_SIZE
- DC_SIZE
) / (KEY_SIZE
+ DC_SIZE
)) {
484 /* for internal which is not root we might check min number of keys */
485 reiserfs_warning(NULL
, "reiserfs-5088",
486 "number of key seems wrong: %z", bh
);
490 used_space
= BLKH_SIZE
+ KEY_SIZE
* nr
+ DC_SIZE
* (nr
+ 1);
491 if (used_space
!= blocksize
- blkh_free_space(blkh
)) {
492 reiserfs_warning(NULL
, "reiserfs-5089",
493 "free space seems wrong: %z", bh
);
496 // one may imagine much more checks
500 // make sure that bh contains formatted node of reiserfs tree of
502 static int is_tree_node(struct buffer_head
*bh
, int level
)
504 if (B_LEVEL(bh
) != level
) {
505 reiserfs_warning(NULL
, "reiserfs-5090", "node level %d does "
506 "not match to the expected one %d",
510 if (level
== DISK_LEAF_NODE_LEVEL
)
511 return is_leaf(bh
->b_data
, bh
->b_size
, bh
);
513 return is_internal(bh
->b_data
, bh
->b_size
, bh
);
516 #define SEARCH_BY_KEY_READA 16
519 * The function is NOT SCHEDULE-SAFE!
520 * It might unlock the write lock if we needed to wait for a block
521 * to be read. Note that in this case it won't recover the lock to avoid
522 * high contention resulting from too much lock requests, especially
523 * the caller (search_by_key) will perform other schedule-unsafe
524 * operations just after calling this function.
526 * @return true if we have unlocked
528 static bool search_by_key_reada(struct super_block
*s
,
529 struct buffer_head
**bh
,
530 b_blocknr_t
*b
, int num
)
533 bool unlocked
= false;
535 for (i
= 0; i
< num
; i
++) {
536 bh
[i
] = sb_getblk(s
, b
[i
]);
539 * We are going to read some blocks on which we
540 * have a reference. It's safe, though we might be
541 * reading blocks concurrently changed if we release
542 * the lock. But it's still fine because we check later
543 * if the tree changed
545 for (j
= 0; j
< i
; j
++) {
547 * note, this needs attention if we are getting rid of the BKL
548 * you have to make sure the prepared bit isn't set on this buffer
550 if (!buffer_uptodate(bh
[j
])) {
552 reiserfs_write_unlock(s
);
555 ll_rw_block(READA
, 1, bh
+ j
);
562 /**************************************************************************
563 * Algorithm SearchByKey *
564 * look for item in the Disk S+Tree by its key *
565 * Input: sb - super block *
566 * key - pointer to the key to search *
567 * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR *
568 * search_path - path from the root to the needed leaf *
569 **************************************************************************/
571 /* This function fills up the path from the root to the leaf as it
572 descends the tree looking for the key. It uses reiserfs_bread to
573 try to find buffers in the cache given their block number. If it
574 does not find them in the cache it reads them from disk. For each
575 node search_by_key finds using reiserfs_bread it then uses
576 bin_search to look through that node. bin_search will find the
577 position of the block_number of the next node if it is looking
578 through an internal node. If it is looking through a leaf node
579 bin_search will find the position of the item which has key either
580 equal to given key, or which is the maximal key less than the given
581 key. search_by_key returns a path that must be checked for the
582 correctness of the top of the path but need not be checked for the
583 correctness of the bottom of the path */
584 /* The function is NOT SCHEDULE-SAFE! */
585 int search_by_key(struct super_block
*sb
, const struct cpu_key
*key
, /* Key to search. */
586 struct treepath
*search_path
,/* This structure was
587 allocated and initialized
589 function. It is filled up
591 int stop_level
/* How far down the tree to search. To
592 stop at leaf level - set to
593 DISK_LEAF_NODE_LEVEL */
596 b_blocknr_t block_number
;
598 struct buffer_head
*bh
;
599 struct path_element
*last_element
;
600 int node_level
, retval
;
601 int right_neighbor_of_leaf_node
;
603 struct buffer_head
*reada_bh
[SEARCH_BY_KEY_READA
];
604 b_blocknr_t reada_blocks
[SEARCH_BY_KEY_READA
];
607 #ifdef CONFIG_REISERFS_CHECK
608 int repeat_counter
= 0;
611 PROC_INFO_INC(sb
, search_by_key
);
613 /* As we add each node to a path we increase its count. This means that
614 we must be careful to release all nodes in a path before we either
615 discard the path struct or re-use the path struct, as we do here. */
617 pathrelse(search_path
);
619 right_neighbor_of_leaf_node
= 0;
621 /* With each iteration of this loop we search through the items in the
622 current node, and calculate the next current node(next path element)
623 for the next iteration of this loop.. */
624 block_number
= SB_ROOT_BLOCK(sb
);
628 #ifdef CONFIG_REISERFS_CHECK
629 if (!(++repeat_counter
% 50000))
630 reiserfs_warning(sb
, "PAP-5100",
631 "%s: there were %d iterations of "
632 "while loop looking for key %K",
633 current
->comm
, repeat_counter
,
637 /* prep path to have another element added to it. */
639 PATH_OFFSET_PELEMENT(search_path
,
640 ++search_path
->path_length
);
641 fs_gen
= get_generation(sb
);
643 /* Read the next tree node, and set the last element in the path to
644 have a pointer to it. */
645 if ((bh
= last_element
->pe_buffer
=
646 sb_getblk(sb
, block_number
))) {
647 bool unlocked
= false;
649 if (!buffer_uptodate(bh
) && reada_count
> 1)
650 /* may unlock the write lock */
651 unlocked
= search_by_key_reada(sb
, reada_bh
,
652 reada_blocks
, reada_count
);
654 * If we haven't already unlocked the write lock,
655 * then we need to do that here before reading
658 if (!buffer_uptodate(bh
) && !unlocked
) {
659 reiserfs_write_unlock(sb
);
662 ll_rw_block(READ
, 1, &bh
);
666 reiserfs_write_lock(sb
);
667 if (!buffer_uptodate(bh
))
671 search_path
->path_length
--;
672 pathrelse(search_path
);
676 if (expected_level
== -1)
677 expected_level
= SB_TREE_HEIGHT(sb
);
680 /* It is possible that schedule occurred. We must check whether the key
681 to search is still in the tree rooted from the current buffer. If
682 not then repeat search from the root. */
683 if (fs_changed(fs_gen
, sb
) &&
684 (!B_IS_IN_TREE(bh
) ||
685 B_LEVEL(bh
) != expected_level
||
686 !key_in_buffer(search_path
, key
, sb
))) {
687 PROC_INFO_INC(sb
, search_by_key_fs_changed
);
688 PROC_INFO_INC(sb
, search_by_key_restarted
);
690 sbk_restarted
[expected_level
- 1]);
691 pathrelse(search_path
);
693 /* Get the root block number so that we can repeat the search
694 starting from the root. */
695 block_number
= SB_ROOT_BLOCK(sb
);
697 right_neighbor_of_leaf_node
= 0;
699 /* repeat search from the root */
703 /* only check that the key is in the buffer if key is not
704 equal to the MAX_KEY. Latter case is only possible in
705 "finish_unfinished()" processing during mount. */
706 RFALSE(comp_keys(&MAX_KEY
, key
) &&
707 !key_in_buffer(search_path
, key
, sb
),
708 "PAP-5130: key is not in the buffer");
709 #ifdef CONFIG_REISERFS_CHECK
710 if (REISERFS_SB(sb
)->cur_tb
) {
711 print_cur_tb("5140");
712 reiserfs_panic(sb
, "PAP-5140",
713 "schedule occurred in do_balance!");
717 // make sure, that the node contents look like a node of
719 if (!is_tree_node(bh
, expected_level
)) {
720 reiserfs_error(sb
, "vs-5150",
721 "invalid format found in block %ld. "
722 "Fsck?", bh
->b_blocknr
);
723 pathrelse(search_path
);
727 /* ok, we have acquired next formatted node in the tree */
728 node_level
= B_LEVEL(bh
);
730 PROC_INFO_BH_STAT(sb
, bh
, node_level
- 1);
732 RFALSE(node_level
< stop_level
,
733 "vs-5152: tree level (%d) is less than stop level (%d)",
734 node_level
, stop_level
);
736 retval
= bin_search(key
, B_N_PITEM_HEAD(bh
, 0),
739 DISK_LEAF_NODE_LEVEL
) ? IH_SIZE
:
741 &(last_element
->pe_position
));
742 if (node_level
== stop_level
) {
746 /* we are not in the stop level */
747 if (retval
== ITEM_FOUND
)
748 /* item has been found, so we choose the pointer which is to the right of the found one */
749 last_element
->pe_position
++;
751 /* if item was not found we choose the position which is to
752 the left of the found item. This requires no code,
753 bin_search did it already. */
755 /* So we have chosen a position in the current node which is
756 an internal node. Now we calculate child block number by
757 position in the node. */
759 B_N_CHILD_NUM(bh
, last_element
->pe_position
);
761 /* if we are going to read leaf nodes, try for read ahead as well */
762 if ((search_path
->reada
& PATH_READA
) &&
763 node_level
== DISK_LEAF_NODE_LEVEL
+ 1) {
764 int pos
= last_element
->pe_position
;
765 int limit
= B_NR_ITEMS(bh
);
766 struct reiserfs_key
*le_key
;
768 if (search_path
->reada
& PATH_READA_BACK
)
770 while (reada_count
< SEARCH_BY_KEY_READA
) {
773 reada_blocks
[reada_count
++] =
774 B_N_CHILD_NUM(bh
, pos
);
775 if (search_path
->reada
& PATH_READA_BACK
)
781 * check to make sure we're in the same object
783 le_key
= B_N_PDELIM_KEY(bh
, pos
);
784 if (le32_to_cpu(le_key
->k_objectid
) !=
785 key
->on_disk_key
.k_objectid
) {
793 /* Form the path to an item and position in this item which contains
794 file byte defined by key. If there is no such item
795 corresponding to the key, we point the path to the item with
796 maximal key less than key, and *pos_in_item is set to one
797 past the last entry/byte in the item. If searching for entry in a
798 directory item, and it is not found, *pos_in_item is set to one
799 entry more than the entry with maximal key which is less than the
802 Note that if there is no entry in this same node which is one more,
803 then we point to an imaginary entry. for direct items, the
804 position is in units of bytes, for indirect items the position is
805 in units of blocknr entries, for directory items the position is in
806 units of directory entries. */
808 /* The function is NOT SCHEDULE-SAFE! */
809 int search_for_position_by_key(struct super_block
*sb
, /* Pointer to the super block. */
810 const struct cpu_key
*p_cpu_key
, /* Key to search (cpu variable) */
811 struct treepath
*search_path
/* Filled up by this function. */
814 struct item_head
*p_le_ih
; /* pointer to on-disk structure */
816 loff_t item_offset
, offset
;
817 struct reiserfs_dir_entry de
;
820 /* If searching for directory entry. */
821 if (is_direntry_cpu_key(p_cpu_key
))
822 return search_by_entry_key(sb
, p_cpu_key
, search_path
,
825 /* If not searching for directory entry. */
827 /* If item is found. */
828 retval
= search_item(sb
, p_cpu_key
, search_path
);
829 if (retval
== IO_ERROR
)
831 if (retval
== ITEM_FOUND
) {
835 (PATH_PLAST_BUFFER(search_path
),
836 PATH_LAST_POSITION(search_path
))),
837 "PAP-5165: item length equals zero");
839 pos_in_item(search_path
) = 0;
840 return POSITION_FOUND
;
843 RFALSE(!PATH_LAST_POSITION(search_path
),
844 "PAP-5170: position equals zero");
846 /* Item is not found. Set path to the previous item. */
848 B_N_PITEM_HEAD(PATH_PLAST_BUFFER(search_path
),
849 --PATH_LAST_POSITION(search_path
));
850 blk_size
= sb
->s_blocksize
;
852 if (comp_short_keys(&(p_le_ih
->ih_key
), p_cpu_key
)) {
853 return FILE_NOT_FOUND
;
856 item_offset
= le_ih_k_offset(p_le_ih
);
857 offset
= cpu_key_k_offset(p_cpu_key
);
859 /* Needed byte is contained in the item pointed to by the path. */
860 if (item_offset
<= offset
&&
861 item_offset
+ op_bytes_number(p_le_ih
, blk_size
) > offset
) {
862 pos_in_item(search_path
) = offset
- item_offset
;
863 if (is_indirect_le_ih(p_le_ih
)) {
864 pos_in_item(search_path
) /= blk_size
;
866 return POSITION_FOUND
;
869 /* Needed byte is not contained in the item pointed to by the
870 path. Set pos_in_item out of the item. */
871 if (is_indirect_le_ih(p_le_ih
))
872 pos_in_item(search_path
) =
873 ih_item_len(p_le_ih
) / UNFM_P_SIZE
;
875 pos_in_item(search_path
) = ih_item_len(p_le_ih
);
877 return POSITION_NOT_FOUND
;
880 /* Compare given item and item pointed to by the path. */
881 int comp_items(const struct item_head
*stored_ih
, const struct treepath
*path
)
883 struct buffer_head
*bh
= PATH_PLAST_BUFFER(path
);
884 struct item_head
*ih
;
886 /* Last buffer at the path is not in the tree. */
887 if (!B_IS_IN_TREE(bh
))
890 /* Last path position is invalid. */
891 if (PATH_LAST_POSITION(path
) >= B_NR_ITEMS(bh
))
894 /* we need only to know, whether it is the same item */
896 return memcmp(stored_ih
, ih
, IH_SIZE
);
899 /* unformatted nodes are not logged anymore, ever. This is safe
902 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
904 // block can not be forgotten as it is in I/O or held by someone
905 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
907 // prepare for delete or cut of direct item
908 static inline int prepare_for_direct_item(struct treepath
*path
,
909 struct item_head
*le_ih
,
911 loff_t new_file_length
, int *cut_size
)
915 if (new_file_length
== max_reiserfs_offset(inode
)) {
916 /* item has to be deleted */
917 *cut_size
= -(IH_SIZE
+ ih_item_len(le_ih
));
920 // new file gets truncated
921 if (get_inode_item_key_version(inode
) == KEY_FORMAT_3_6
) {
923 round_len
= ROUND_UP(new_file_length
);
924 /* this was new_file_length < le_ih ... */
925 if (round_len
< le_ih_k_offset(le_ih
)) {
926 *cut_size
= -(IH_SIZE
+ ih_item_len(le_ih
));
927 return M_DELETE
; /* Delete this item. */
929 /* Calculate first position and size for cutting from item. */
930 pos_in_item(path
) = round_len
- (le_ih_k_offset(le_ih
) - 1);
931 *cut_size
= -(ih_item_len(le_ih
) - pos_in_item(path
));
933 return M_CUT
; /* Cut from this item. */
936 // old file: items may have any length
938 if (new_file_length
< le_ih_k_offset(le_ih
)) {
939 *cut_size
= -(IH_SIZE
+ ih_item_len(le_ih
));
940 return M_DELETE
; /* Delete this item. */
942 /* Calculate first position and size for cutting from item. */
943 *cut_size
= -(ih_item_len(le_ih
) -
945 new_file_length
+ 1 - le_ih_k_offset(le_ih
)));
946 return M_CUT
; /* Cut from this item. */
949 static inline int prepare_for_direntry_item(struct treepath
*path
,
950 struct item_head
*le_ih
,
952 loff_t new_file_length
,
955 if (le_ih_k_offset(le_ih
) == DOT_OFFSET
&&
956 new_file_length
== max_reiserfs_offset(inode
)) {
957 RFALSE(ih_entry_count(le_ih
) != 2,
958 "PAP-5220: incorrect empty directory item (%h)", le_ih
);
959 *cut_size
= -(IH_SIZE
+ ih_item_len(le_ih
));
960 return M_DELETE
; /* Delete the directory item containing "." and ".." entry. */
963 if (ih_entry_count(le_ih
) == 1) {
964 /* Delete the directory item such as there is one record only
966 *cut_size
= -(IH_SIZE
+ ih_item_len(le_ih
));
970 /* Cut one record from the directory item. */
973 entry_length(get_last_bh(path
), le_ih
, pos_in_item(path
)));
977 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
979 /* If the path points to a directory or direct item, calculate mode and the size cut, for balance.
980 If the path points to an indirect item, remove some number of its unformatted nodes.
981 In case of file truncate calculate whether this item must be deleted/truncated or last
982 unformatted node of this item will be converted to a direct item.
983 This function returns a determination of what balance mode the calling function should employ. */
984 static char prepare_for_delete_or_cut(struct reiserfs_transaction_handle
*th
, struct inode
*inode
, struct treepath
*path
, const struct cpu_key
*item_key
, int *removed
, /* Number of unformatted nodes which were removed
985 from end of the file. */
986 int *cut_size
, unsigned long long new_file_length
/* MAX_KEY_OFFSET in case of delete. */
989 struct super_block
*sb
= inode
->i_sb
;
990 struct item_head
*p_le_ih
= PATH_PITEM_HEAD(path
);
991 struct buffer_head
*bh
= PATH_PLAST_BUFFER(path
);
993 BUG_ON(!th
->t_trans_id
);
995 /* Stat_data item. */
996 if (is_statdata_le_ih(p_le_ih
)) {
998 RFALSE(new_file_length
!= max_reiserfs_offset(inode
),
999 "PAP-5210: mode must be M_DELETE");
1001 *cut_size
= -(IH_SIZE
+ ih_item_len(p_le_ih
));
1005 /* Directory item. */
1006 if (is_direntry_le_ih(p_le_ih
))
1007 return prepare_for_direntry_item(path
, p_le_ih
, inode
,
1012 if (is_direct_le_ih(p_le_ih
))
1013 return prepare_for_direct_item(path
, p_le_ih
, inode
,
1014 new_file_length
, cut_size
);
1016 /* Case of an indirect item. */
1018 int blk_size
= sb
->s_blocksize
;
1019 struct item_head s_ih
;
1025 if ( new_file_length
== max_reiserfs_offset (inode
) ) {
1026 /* prepare_for_delete_or_cut() is called by
1027 * reiserfs_delete_item() */
1028 new_file_length
= 0;
1035 bh
= PATH_PLAST_BUFFER(path
);
1036 copy_item_head(&s_ih
, PATH_PITEM_HEAD(path
));
1037 pos
= I_UNFM_NUM(&s_ih
);
1039 while (le_ih_k_offset (&s_ih
) + (pos
- 1) * blk_size
> new_file_length
) {
1043 /* Each unformatted block deletion may involve one additional
1044 * bitmap block into the transaction, thereby the initial
1045 * journal space reservation might not be enough. */
1046 if (!delete && (*cut_size
) != 0 &&
1047 reiserfs_transaction_free_space(th
) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD
)
1050 unfm
= (__le32
*)B_I_PITEM(bh
, &s_ih
) + pos
- 1;
1051 block
= get_block_num(unfm
, 0);
1054 reiserfs_prepare_for_journal(sb
, bh
, 1);
1055 put_block_num(unfm
, 0, 0);
1056 journal_mark_dirty(th
, sb
, bh
);
1057 reiserfs_free_block(th
, inode
, block
, 1);
1060 reiserfs_write_unlock(sb
);
1062 reiserfs_write_lock(sb
);
1064 if (item_moved (&s_ih
, path
)) {
1071 (*cut_size
) -= UNFM_P_SIZE
;
1074 (*cut_size
) -= IH_SIZE
;
1079 /* a trick. If the buffer has been logged, this will do nothing. If
1080 ** we've broken the loop without logging it, it will restore the
1082 reiserfs_restore_prepared_buffer(sb
, bh
);
1083 } while (need_re_search
&&
1084 search_for_position_by_key(sb
, item_key
, path
) == POSITION_FOUND
);
1085 pos_in_item(path
) = pos
* UNFM_P_SIZE
;
1087 if (*cut_size
== 0) {
1088 /* Nothing were cut. maybe convert last unformatted node to the
1096 /* Calculate number of bytes which will be deleted or cut during balance */
1097 static int calc_deleted_bytes_number(struct tree_balance
*tb
, char mode
)
1100 struct item_head
*p_le_ih
= PATH_PITEM_HEAD(tb
->tb_path
);
1102 if (is_statdata_le_ih(p_le_ih
))
1107 M_DELETE
) ? ih_item_len(p_le_ih
) : -tb
->insert_size
[0];
1108 if (is_direntry_le_ih(p_le_ih
)) {
1112 if (is_indirect_le_ih(p_le_ih
))
1113 del_size
= (del_size
/ UNFM_P_SIZE
) *
1114 (PATH_PLAST_BUFFER(tb
->tb_path
)->b_size
);
1118 static void init_tb_struct(struct reiserfs_transaction_handle
*th
,
1119 struct tree_balance
*tb
,
1120 struct super_block
*sb
,
1121 struct treepath
*path
, int size
)
1124 BUG_ON(!th
->t_trans_id
);
1126 memset(tb
, '\0', sizeof(struct tree_balance
));
1127 tb
->transaction_handle
= th
;
1130 PATH_OFFSET_PBUFFER(path
, ILLEGAL_PATH_ELEMENT_OFFSET
) = NULL
;
1131 PATH_OFFSET_POSITION(path
, ILLEGAL_PATH_ELEMENT_OFFSET
) = 0;
1132 tb
->insert_size
[0] = size
;
1135 void padd_item(char *item
, int total_length
, int length
)
1139 for (i
= total_length
; i
> length
;)
1143 #ifdef REISERQUOTA_DEBUG
1144 char key2type(struct reiserfs_key
*ih
)
1146 if (is_direntry_le_key(2, ih
))
1148 if (is_direct_le_key(2, ih
))
1150 if (is_indirect_le_key(2, ih
))
1152 if (is_statdata_le_key(2, ih
))
1157 char head2type(struct item_head
*ih
)
1159 if (is_direntry_le_ih(ih
))
1161 if (is_direct_le_ih(ih
))
1163 if (is_indirect_le_ih(ih
))
1165 if (is_statdata_le_ih(ih
))
1171 /* Delete object item.
1172 * th - active transaction handle
1173 * path - path to the deleted item
1174 * item_key - key to search for the deleted item
1175 * indode - used for updating i_blocks and quotas
1176 * un_bh - NULL or unformatted node pointer
1178 int reiserfs_delete_item(struct reiserfs_transaction_handle
*th
,
1179 struct treepath
*path
, const struct cpu_key
*item_key
,
1180 struct inode
*inode
, struct buffer_head
*un_bh
)
1182 struct super_block
*sb
= inode
->i_sb
;
1183 struct tree_balance s_del_balance
;
1184 struct item_head s_ih
;
1185 struct item_head
*q_ih
;
1186 int quota_cut_bytes
;
1187 int ret_value
, del_size
, removed
;
1189 #ifdef CONFIG_REISERFS_CHECK
1194 BUG_ON(!th
->t_trans_id
);
1196 init_tb_struct(th
, &s_del_balance
, sb
, path
,
1197 0 /*size is unknown */ );
1202 #ifdef CONFIG_REISERFS_CHECK
1206 prepare_for_delete_or_cut(th
, inode
, path
,
1209 max_reiserfs_offset(inode
));
1211 RFALSE(mode
!= M_DELETE
, "PAP-5320: mode must be M_DELETE");
1213 copy_item_head(&s_ih
, PATH_PITEM_HEAD(path
));
1214 s_del_balance
.insert_size
[0] = del_size
;
1216 ret_value
= fix_nodes(M_DELETE
, &s_del_balance
, NULL
, NULL
);
1217 if (ret_value
!= REPEAT_SEARCH
)
1220 PROC_INFO_INC(sb
, delete_item_restarted
);
1222 // file system changed, repeat search
1224 search_for_position_by_key(sb
, item_key
, path
);
1225 if (ret_value
== IO_ERROR
)
1227 if (ret_value
== FILE_NOT_FOUND
) {
1228 reiserfs_warning(sb
, "vs-5340",
1229 "no items of the file %K found",
1235 if (ret_value
!= CARRY_ON
) {
1236 unfix_nodes(&s_del_balance
);
1239 // reiserfs_delete_item returns item length when success
1240 ret_value
= calc_deleted_bytes_number(&s_del_balance
, M_DELETE
);
1241 q_ih
= get_ih(path
);
1242 quota_cut_bytes
= ih_item_len(q_ih
);
1244 /* hack so the quota code doesn't have to guess if the file
1245 ** has a tail. On tail insert, we allocate quota for 1 unformatted node.
1246 ** We test the offset because the tail might have been
1247 ** split into multiple items, and we only want to decrement for
1248 ** the unfm node once
1250 if (!S_ISLNK(inode
->i_mode
) && is_direct_le_ih(q_ih
)) {
1251 if ((le_ih_k_offset(q_ih
) & (sb
->s_blocksize
- 1)) == 1) {
1252 quota_cut_bytes
= sb
->s_blocksize
+ UNFM_P_SIZE
;
1254 quota_cut_bytes
= 0;
1262 /* We are in direct2indirect conversion, so move tail contents
1263 to the unformatted node */
1264 /* note, we do the copy before preparing the buffer because we
1265 ** don't care about the contents of the unformatted node yet.
1266 ** the only thing we really care about is the direct item's data
1267 ** is in the unformatted node.
1269 ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1270 ** the unformatted node, which might schedule, meaning we'd have to
1271 ** loop all the way back up to the start of the while loop.
1273 ** The unformatted node must be dirtied later on. We can't be
1274 ** sure here if the entire tail has been deleted yet.
1276 ** un_bh is from the page cache (all unformatted nodes are
1277 ** from the page cache) and might be a highmem page. So, we
1278 ** can't use un_bh->b_data.
1282 data
= kmap_atomic(un_bh
->b_page
, KM_USER0
);
1283 off
= ((le_ih_k_offset(&s_ih
) - 1) & (PAGE_CACHE_SIZE
- 1));
1285 B_I_PITEM(PATH_PLAST_BUFFER(path
), &s_ih
),
1287 kunmap_atomic(data
, KM_USER0
);
1289 /* Perform balancing after all resources have been collected at once. */
1290 do_balance(&s_del_balance
, NULL
, NULL
, M_DELETE
);
1292 #ifdef REISERQUOTA_DEBUG
1293 reiserfs_debug(sb
, REISERFS_DEBUG_CODE
,
1294 "reiserquota delete_item(): freeing %u, id=%u type=%c",
1295 quota_cut_bytes
, inode
->i_uid
, head2type(&s_ih
));
1297 dquot_free_space_nodirty(inode
, quota_cut_bytes
);
1299 /* Return deleted body length */
1303 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1305 deletion of the body of the object is performed by iput(), with the
1306 result that if multiple processes are operating on a file, the
1307 deletion of the body of the file is deferred until the last process
1308 that has an open inode performs its iput().
1310 writes and truncates are protected from collisions by use of
1313 creates, linking, and mknod are protected from collisions with other
1314 processes by making the reiserfs_add_entry() the last step in the
1315 creation, and then rolling back all changes if there was a collision.
1319 /* this deletes item which never gets split */
1320 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle
*th
,
1321 struct inode
*inode
, struct reiserfs_key
*key
)
1323 struct tree_balance tb
;
1324 INITIALIZE_PATH(path
);
1327 struct cpu_key cpu_key
;
1329 int quota_cut_bytes
= 0;
1331 BUG_ON(!th
->t_trans_id
);
1333 le_key2cpu_key(&cpu_key
, key
);
1336 retval
= search_item(th
->t_super
, &cpu_key
, &path
);
1337 if (retval
== IO_ERROR
) {
1338 reiserfs_error(th
->t_super
, "vs-5350",
1339 "i/o failure occurred trying "
1340 "to delete %K", &cpu_key
);
1343 if (retval
!= ITEM_FOUND
) {
1345 // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1347 ((unsigned long long)
1348 GET_HASH_VALUE(le_key_k_offset
1349 (le_key_version(key
), key
)) == 0
1350 && (unsigned long long)
1351 GET_GENERATION_NUMBER(le_key_k_offset
1352 (le_key_version(key
),
1354 reiserfs_warning(th
->t_super
, "vs-5355",
1355 "%k not found", key
);
1360 item_len
= ih_item_len(PATH_PITEM_HEAD(&path
));
1361 init_tb_struct(th
, &tb
, th
->t_super
, &path
,
1362 -(IH_SIZE
+ item_len
));
1364 quota_cut_bytes
= ih_item_len(PATH_PITEM_HEAD(&path
));
1366 retval
= fix_nodes(M_DELETE
, &tb
, NULL
, NULL
);
1367 if (retval
== REPEAT_SEARCH
) {
1368 PROC_INFO_INC(th
->t_super
, delete_solid_item_restarted
);
1372 if (retval
== CARRY_ON
) {
1373 do_balance(&tb
, NULL
, NULL
, M_DELETE
);
1374 if (inode
) { /* Should we count quota for item? (we don't count quotas for save-links) */
1375 #ifdef REISERQUOTA_DEBUG
1376 reiserfs_debug(th
->t_super
, REISERFS_DEBUG_CODE
,
1377 "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1378 quota_cut_bytes
, inode
->i_uid
,
1381 dquot_free_space_nodirty(inode
,
1386 // IO_ERROR, NO_DISK_SPACE, etc
1387 reiserfs_warning(th
->t_super
, "vs-5360",
1388 "could not delete %K due to fix_nodes failure",
1394 reiserfs_check_path(&path
);
1397 int reiserfs_delete_object(struct reiserfs_transaction_handle
*th
,
1398 struct inode
*inode
)
1402 BUG_ON(!th
->t_trans_id
);
1404 /* for directory this deletes item containing "." and ".." */
1406 reiserfs_do_truncate(th
, inode
, NULL
, 0 /*no timestamp updates */ );
1410 #if defined(USE_INODE_GENERATION_COUNTER)
1411 if (!old_format_only(th
->t_super
)) {
1412 __le32
*inode_generation
;
1415 &REISERFS_SB(th
->t_super
)->s_rs
->s_inode_generation
;
1416 le32_add_cpu(inode_generation
, 1);
1418 /* USE_INODE_GENERATION_COUNTER */
1420 reiserfs_delete_solid_item(th
, inode
, INODE_PKEY(inode
));
1425 static void unmap_buffers(struct page
*page
, loff_t pos
)
1427 struct buffer_head
*bh
;
1428 struct buffer_head
*head
;
1429 struct buffer_head
*next
;
1430 unsigned long tail_index
;
1431 unsigned long cur_index
;
1434 if (page_has_buffers(page
)) {
1435 tail_index
= pos
& (PAGE_CACHE_SIZE
- 1);
1437 head
= page_buffers(page
);
1440 next
= bh
->b_this_page
;
1442 /* we want to unmap the buffers that contain the tail, and
1443 ** all the buffers after it (since the tail must be at the
1444 ** end of the file). We don't want to unmap file data
1445 ** before the tail, since it might be dirty and waiting to
1448 cur_index
+= bh
->b_size
;
1449 if (cur_index
> tail_index
) {
1450 reiserfs_unmap_buffer(bh
);
1453 } while (bh
!= head
);
1458 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle
*th
,
1459 struct inode
*inode
,
1461 struct treepath
*path
,
1462 const struct cpu_key
*item_key
,
1463 loff_t new_file_size
, char *mode
)
1465 struct super_block
*sb
= inode
->i_sb
;
1466 int block_size
= sb
->s_blocksize
;
1468 BUG_ON(!th
->t_trans_id
);
1469 BUG_ON(new_file_size
!= inode
->i_size
);
1471 /* the page being sent in could be NULL if there was an i/o error
1472 ** reading in the last block. The user will hit problems trying to
1473 ** read the file, but for now we just skip the indirect2direct
1475 if (atomic_read(&inode
->i_count
) > 1 ||
1476 !tail_has_to_be_packed(inode
) ||
1477 !page
|| (REISERFS_I(inode
)->i_flags
& i_nopack_mask
)) {
1478 /* leave tail in an unformatted node */
1479 *mode
= M_SKIP_BALANCING
;
1481 block_size
- (new_file_size
& (block_size
- 1));
1485 /* Perform the conversion to a direct_item. */
1486 /* return indirect_to_direct(inode, path, item_key,
1487 new_file_size, mode); */
1488 return indirect2direct(th
, inode
, page
, path
, item_key
,
1489 new_file_size
, mode
);
1492 /* we did indirect_to_direct conversion. And we have inserted direct
1493 item successesfully, but there were no disk space to cut unfm
1494 pointer being converted. Therefore we have to delete inserted
1496 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle
*th
,
1497 struct inode
*inode
, struct treepath
*path
)
1499 struct cpu_key tail_key
;
1502 BUG_ON(!th
->t_trans_id
);
1504 make_cpu_key(&tail_key
, inode
, inode
->i_size
+ 1, TYPE_DIRECT
, 4); // !!!!
1505 tail_key
.key_length
= 4;
1508 (cpu_key_k_offset(&tail_key
) & (inode
->i_sb
->s_blocksize
- 1)) - 1;
1510 /* look for the last byte of the tail */
1511 if (search_for_position_by_key(inode
->i_sb
, &tail_key
, path
) ==
1513 reiserfs_panic(inode
->i_sb
, "vs-5615",
1514 "found invalid item");
1515 RFALSE(path
->pos_in_item
!=
1516 ih_item_len(PATH_PITEM_HEAD(path
)) - 1,
1517 "vs-5616: appended bytes found");
1518 PATH_LAST_POSITION(path
)--;
1521 reiserfs_delete_item(th
, path
, &tail_key
, inode
,
1522 NULL
/*unbh not needed */ );
1524 || removed
> tail_len
,
1525 "vs-5617: there was tail %d bytes, removed item length %d bytes",
1527 tail_len
-= removed
;
1528 set_cpu_key_k_offset(&tail_key
,
1529 cpu_key_k_offset(&tail_key
) - removed
);
1531 reiserfs_warning(inode
->i_sb
, "reiserfs-5091", "indirect_to_direct "
1532 "conversion has been rolled back due to "
1533 "lack of disk space");
1534 //mark_file_without_tail (inode);
1535 mark_inode_dirty(inode
);
1538 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1539 int reiserfs_cut_from_item(struct reiserfs_transaction_handle
*th
,
1540 struct treepath
*path
,
1541 struct cpu_key
*item_key
,
1542 struct inode
*inode
,
1543 struct page
*page
, loff_t new_file_size
)
1545 struct super_block
*sb
= inode
->i_sb
;
1546 /* Every function which is going to call do_balance must first
1547 create a tree_balance structure. Then it must fill up this
1548 structure by using the init_tb_struct and fix_nodes functions.
1549 After that we can make tree balancing. */
1550 struct tree_balance s_cut_balance
;
1551 struct item_head
*p_le_ih
;
1552 int cut_size
= 0, /* Amount to be cut. */
1553 ret_value
= CARRY_ON
, removed
= 0, /* Number of the removed unformatted nodes. */
1554 is_inode_locked
= 0;
1555 char mode
; /* Mode of the balance. */
1557 int quota_cut_bytes
;
1558 loff_t tail_pos
= 0;
1560 BUG_ON(!th
->t_trans_id
);
1562 init_tb_struct(th
, &s_cut_balance
, inode
->i_sb
, path
,
1565 /* Repeat this loop until we either cut the item without needing
1566 to balance, or we fix_nodes without schedule occurring */
1568 /* Determine the balance mode, position of the first byte to
1569 be cut, and size to be cut. In case of the indirect item
1570 free unformatted nodes which are pointed to by the cut
1574 prepare_for_delete_or_cut(th
, inode
, path
,
1576 &cut_size
, new_file_size
);
1577 if (mode
== M_CONVERT
) {
1578 /* convert last unformatted node to direct item or leave
1579 tail in the unformatted node */
1580 RFALSE(ret_value
!= CARRY_ON
,
1581 "PAP-5570: can not convert twice");
1584 maybe_indirect_to_direct(th
, inode
, page
,
1586 new_file_size
, &mode
);
1587 if (mode
== M_SKIP_BALANCING
)
1588 /* tail has been left in the unformatted node */
1591 is_inode_locked
= 1;
1593 /* removing of last unformatted node will change value we
1594 have to return to truncate. Save it */
1595 retval2
= ret_value
;
1596 /*retval2 = sb->s_blocksize - (new_file_size & (sb->s_blocksize - 1)); */
1598 /* So, we have performed the first part of the conversion:
1599 inserting the new direct item. Now we are removing the
1600 last unformatted node pointer. Set key to search for
1602 set_cpu_key_k_type(item_key
, TYPE_INDIRECT
);
1603 item_key
->key_length
= 4;
1605 (new_file_size
& (sb
->s_blocksize
- 1));
1606 tail_pos
= new_file_size
;
1607 set_cpu_key_k_offset(item_key
, new_file_size
+ 1);
1608 if (search_for_position_by_key
1610 path
) == POSITION_NOT_FOUND
) {
1611 print_block(PATH_PLAST_BUFFER(path
), 3,
1612 PATH_LAST_POSITION(path
) - 1,
1613 PATH_LAST_POSITION(path
) + 1);
1614 reiserfs_panic(sb
, "PAP-5580", "item to "
1615 "convert does not exist (%K)",
1620 if (cut_size
== 0) {
1625 s_cut_balance
.insert_size
[0] = cut_size
;
1627 ret_value
= fix_nodes(mode
, &s_cut_balance
, NULL
, NULL
);
1628 if (ret_value
!= REPEAT_SEARCH
)
1631 PROC_INFO_INC(sb
, cut_from_item_restarted
);
1634 search_for_position_by_key(sb
, item_key
, path
);
1635 if (ret_value
== POSITION_FOUND
)
1638 reiserfs_warning(sb
, "PAP-5610", "item %K not found",
1640 unfix_nodes(&s_cut_balance
);
1641 return (ret_value
== IO_ERROR
) ? -EIO
: -ENOENT
;
1644 // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1645 if (ret_value
!= CARRY_ON
) {
1646 if (is_inode_locked
) {
1648 indirect_to_direct_roll_back(th
, inode
, path
);
1650 if (ret_value
== NO_DISK_SPACE
)
1651 reiserfs_warning(sb
, "reiserfs-5092",
1653 unfix_nodes(&s_cut_balance
);
1657 /* go ahead and perform balancing */
1659 RFALSE(mode
== M_PASTE
|| mode
== M_INSERT
, "invalid mode");
1661 /* Calculate number of bytes that need to be cut from the item. */
1664 M_DELETE
) ? ih_item_len(get_ih(path
)) : -s_cut_balance
.
1667 ret_value
= calc_deleted_bytes_number(&s_cut_balance
, mode
);
1669 ret_value
= retval2
;
1671 /* For direct items, we only change the quota when deleting the last
1674 p_le_ih
= PATH_PITEM_HEAD(s_cut_balance
.tb_path
);
1675 if (!S_ISLNK(inode
->i_mode
) && is_direct_le_ih(p_le_ih
)) {
1676 if (mode
== M_DELETE
&&
1677 (le_ih_k_offset(p_le_ih
) & (sb
->s_blocksize
- 1)) ==
1679 REISERFS_I(inode
)->i_first_direct_byte
= U32_MAX
;
1680 quota_cut_bytes
= sb
->s_blocksize
+ UNFM_P_SIZE
;
1682 quota_cut_bytes
= 0;
1685 #ifdef CONFIG_REISERFS_CHECK
1686 if (is_inode_locked
) {
1687 struct item_head
*le_ih
=
1688 PATH_PITEM_HEAD(s_cut_balance
.tb_path
);
1689 /* we are going to complete indirect2direct conversion. Make
1690 sure, that we exactly remove last unformatted node pointer
1692 if (!is_indirect_le_ih(le_ih
))
1693 reiserfs_panic(sb
, "vs-5652",
1694 "item must be indirect %h", le_ih
);
1696 if (mode
== M_DELETE
&& ih_item_len(le_ih
) != UNFM_P_SIZE
)
1697 reiserfs_panic(sb
, "vs-5653", "completing "
1698 "indirect2direct conversion indirect "
1699 "item %h being deleted must be of "
1700 "4 byte long", le_ih
);
1703 && s_cut_balance
.insert_size
[0] != -UNFM_P_SIZE
) {
1704 reiserfs_panic(sb
, "vs-5654", "can not complete "
1705 "indirect2direct conversion of %h "
1706 "(CUT, insert_size==%d)",
1707 le_ih
, s_cut_balance
.insert_size
[0]);
1709 /* it would be useful to make sure, that right neighboring
1710 item is direct item of this file */
1714 do_balance(&s_cut_balance
, NULL
, NULL
, mode
);
1715 if (is_inode_locked
) {
1716 /* we've done an indirect->direct conversion. when the data block
1717 ** was freed, it was removed from the list of blocks that must
1718 ** be flushed before the transaction commits, make sure to
1719 ** unmap and invalidate it
1721 unmap_buffers(page
, tail_pos
);
1722 REISERFS_I(inode
)->i_flags
&= ~i_pack_on_close_mask
;
1724 #ifdef REISERQUOTA_DEBUG
1725 reiserfs_debug(inode
->i_sb
, REISERFS_DEBUG_CODE
,
1726 "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1727 quota_cut_bytes
, inode
->i_uid
, '?');
1729 dquot_free_space_nodirty(inode
, quota_cut_bytes
);
1733 static void truncate_directory(struct reiserfs_transaction_handle
*th
,
1734 struct inode
*inode
)
1736 BUG_ON(!th
->t_trans_id
);
1738 reiserfs_error(inode
->i_sb
, "vs-5655", "link count != 0");
1740 set_le_key_k_offset(KEY_FORMAT_3_5
, INODE_PKEY(inode
), DOT_OFFSET
);
1741 set_le_key_k_type(KEY_FORMAT_3_5
, INODE_PKEY(inode
), TYPE_DIRENTRY
);
1742 reiserfs_delete_solid_item(th
, inode
, INODE_PKEY(inode
));
1743 reiserfs_update_sd(th
, inode
);
1744 set_le_key_k_offset(KEY_FORMAT_3_5
, INODE_PKEY(inode
), SD_OFFSET
);
1745 set_le_key_k_type(KEY_FORMAT_3_5
, INODE_PKEY(inode
), TYPE_STAT_DATA
);
1748 /* Truncate file to the new size. Note, this must be called with a transaction
1750 int reiserfs_do_truncate(struct reiserfs_transaction_handle
*th
,
1751 struct inode
*inode
, /* ->i_size contains new size */
1752 struct page
*page
, /* up to date for last block */
1753 int update_timestamps
/* when it is called by
1754 file_release to convert
1755 the tail - no timestamps
1756 should be updated */
1759 INITIALIZE_PATH(s_search_path
); /* Path to the current object item. */
1760 struct item_head
*p_le_ih
; /* Pointer to an item header. */
1761 struct cpu_key s_item_key
; /* Key to search for a previous file item. */
1762 loff_t file_size
, /* Old file size. */
1763 new_file_size
; /* New file size. */
1764 int deleted
; /* Number of deleted or truncated bytes. */
1768 BUG_ON(!th
->t_trans_id
);
1770 (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
1771 || S_ISLNK(inode
->i_mode
)))
1774 if (S_ISDIR(inode
->i_mode
)) {
1775 // deletion of directory - no need to update timestamps
1776 truncate_directory(th
, inode
);
1780 /* Get new file size. */
1781 new_file_size
= inode
->i_size
;
1783 make_cpu_key(&s_item_key
, inode
, max_reiserfs_offset(inode
),
1787 search_for_position_by_key(inode
->i_sb
, &s_item_key
,
1789 if (retval
== IO_ERROR
) {
1790 reiserfs_error(inode
->i_sb
, "vs-5657",
1791 "i/o failure occurred trying to truncate %K",
1796 if (retval
== POSITION_FOUND
|| retval
== FILE_NOT_FOUND
) {
1797 reiserfs_error(inode
->i_sb
, "PAP-5660",
1798 "wrong result %d of search for %K", retval
,
1805 s_search_path
.pos_in_item
--;
1807 /* Get real file size (total length of all file items) */
1808 p_le_ih
= PATH_PITEM_HEAD(&s_search_path
);
1809 if (is_statdata_le_ih(p_le_ih
))
1812 loff_t offset
= le_ih_k_offset(p_le_ih
);
1814 op_bytes_number(p_le_ih
, inode
->i_sb
->s_blocksize
);
1816 /* this may mismatch with real file size: if last direct item
1817 had no padding zeros and last unformatted node had no free
1818 space, this file would have this file size */
1819 file_size
= offset
+ bytes
- 1;
1822 * are we doing a full truncate or delete, if so
1823 * kick in the reada code
1825 if (new_file_size
== 0)
1826 s_search_path
.reada
= PATH_READA
| PATH_READA_BACK
;
1828 if (file_size
== 0 || file_size
< new_file_size
) {
1829 goto update_and_out
;
1832 /* Update key to search for the last file item. */
1833 set_cpu_key_k_offset(&s_item_key
, file_size
);
1836 /* Cut or delete file item. */
1838 reiserfs_cut_from_item(th
, &s_search_path
, &s_item_key
,
1839 inode
, page
, new_file_size
);
1841 reiserfs_warning(inode
->i_sb
, "vs-5665",
1842 "reiserfs_cut_from_item failed");
1843 reiserfs_check_path(&s_search_path
);
1847 RFALSE(deleted
> file_size
,
1848 "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1849 deleted
, file_size
, &s_item_key
);
1851 /* Change key to search the last file item. */
1852 file_size
-= deleted
;
1854 set_cpu_key_k_offset(&s_item_key
, file_size
);
1856 /* While there are bytes to truncate and previous file item is presented in the tree. */
1859 ** This loop could take a really long time, and could log
1860 ** many more blocks than a transaction can hold. So, we do a polite
1861 ** journal end here, and if the transaction needs ending, we make
1862 ** sure the file is consistent before ending the current trans
1863 ** and starting a new one
1865 if (journal_transaction_should_end(th
, 0) ||
1866 reiserfs_transaction_free_space(th
) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD
) {
1867 int orig_len_alloc
= th
->t_blocks_allocated
;
1868 pathrelse(&s_search_path
);
1870 if (update_timestamps
) {
1871 inode
->i_mtime
= CURRENT_TIME_SEC
;
1872 inode
->i_ctime
= CURRENT_TIME_SEC
;
1874 reiserfs_update_sd(th
, inode
);
1876 err
= journal_end(th
, inode
->i_sb
, orig_len_alloc
);
1879 err
= journal_begin(th
, inode
->i_sb
,
1880 JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD
+ JOURNAL_PER_BALANCE_CNT
* 4) ;
1883 reiserfs_update_inode_transaction(inode
);
1885 } while (file_size
> ROUND_UP(new_file_size
) &&
1886 search_for_position_by_key(inode
->i_sb
, &s_item_key
,
1887 &s_search_path
) == POSITION_FOUND
);
1889 RFALSE(file_size
> ROUND_UP(new_file_size
),
1890 "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1891 new_file_size
, file_size
, s_item_key
.on_disk_key
.k_objectid
);
1894 if (update_timestamps
) {
1895 // this is truncate, not file closing
1896 inode
->i_mtime
= CURRENT_TIME_SEC
;
1897 inode
->i_ctime
= CURRENT_TIME_SEC
;
1899 reiserfs_update_sd(th
, inode
);
1902 pathrelse(&s_search_path
);
1906 #ifdef CONFIG_REISERFS_CHECK
1907 // this makes sure, that we __append__, not overwrite or add holes
1908 static void check_research_for_paste(struct treepath
*path
,
1909 const struct cpu_key
*key
)
1911 struct item_head
*found_ih
= get_ih(path
);
1913 if (is_direct_le_ih(found_ih
)) {
1914 if (le_ih_k_offset(found_ih
) +
1915 op_bytes_number(found_ih
,
1916 get_last_bh(path
)->b_size
) !=
1917 cpu_key_k_offset(key
)
1918 || op_bytes_number(found_ih
,
1919 get_last_bh(path
)->b_size
) !=
1921 reiserfs_panic(NULL
, "PAP-5720", "found direct item "
1922 "%h or position (%d) does not match "
1923 "to key %K", found_ih
,
1924 pos_in_item(path
), key
);
1926 if (is_indirect_le_ih(found_ih
)) {
1927 if (le_ih_k_offset(found_ih
) +
1928 op_bytes_number(found_ih
,
1929 get_last_bh(path
)->b_size
) !=
1930 cpu_key_k_offset(key
)
1931 || I_UNFM_NUM(found_ih
) != pos_in_item(path
)
1932 || get_ih_free_space(found_ih
) != 0)
1933 reiserfs_panic(NULL
, "PAP-5730", "found indirect "
1934 "item (%h) or position (%d) does not "
1935 "match to key (%K)",
1936 found_ih
, pos_in_item(path
), key
);
1939 #endif /* config reiserfs check */
1941 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1942 int reiserfs_paste_into_item(struct reiserfs_transaction_handle
*th
, struct treepath
*search_path
, /* Path to the pasted item. */
1943 const struct cpu_key
*key
, /* Key to search for the needed item. */
1944 struct inode
*inode
, /* Inode item belongs to */
1945 const char *body
, /* Pointer to the bytes to paste. */
1947 { /* Size of pasted bytes. */
1948 struct tree_balance s_paste_balance
;
1952 BUG_ON(!th
->t_trans_id
);
1954 fs_gen
= get_generation(inode
->i_sb
);
1956 #ifdef REISERQUOTA_DEBUG
1957 reiserfs_debug(inode
->i_sb
, REISERFS_DEBUG_CODE
,
1958 "reiserquota paste_into_item(): allocating %u id=%u type=%c",
1959 pasted_size
, inode
->i_uid
,
1960 key2type(&(key
->on_disk_key
)));
1963 retval
= dquot_alloc_space_nodirty(inode
, pasted_size
);
1965 pathrelse(search_path
);
1968 init_tb_struct(th
, &s_paste_balance
, th
->t_super
, search_path
,
1970 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1971 s_paste_balance
.key
= key
->on_disk_key
;
1974 /* DQUOT_* can schedule, must check before the fix_nodes */
1975 if (fs_changed(fs_gen
, inode
->i_sb
)) {
1980 fix_nodes(M_PASTE
, &s_paste_balance
, NULL
,
1981 body
)) == REPEAT_SEARCH
) {
1983 /* file system changed while we were in the fix_nodes */
1984 PROC_INFO_INC(th
->t_super
, paste_into_item_restarted
);
1986 search_for_position_by_key(th
->t_super
, key
,
1988 if (retval
== IO_ERROR
) {
1992 if (retval
== POSITION_FOUND
) {
1993 reiserfs_warning(inode
->i_sb
, "PAP-5710",
1994 "entry or pasted byte (%K) exists",
1999 #ifdef CONFIG_REISERFS_CHECK
2000 check_research_for_paste(search_path
, key
);
2004 /* Perform balancing after all resources are collected by fix_nodes, and
2005 accessing them will not risk triggering schedule. */
2006 if (retval
== CARRY_ON
) {
2007 do_balance(&s_paste_balance
, NULL
/*ih */ , body
, M_PASTE
);
2010 retval
= (retval
== NO_DISK_SPACE
) ? -ENOSPC
: -EIO
;
2012 /* this also releases the path */
2013 unfix_nodes(&s_paste_balance
);
2014 #ifdef REISERQUOTA_DEBUG
2015 reiserfs_debug(inode
->i_sb
, REISERFS_DEBUG_CODE
,
2016 "reiserquota paste_into_item(): freeing %u id=%u type=%c",
2017 pasted_size
, inode
->i_uid
,
2018 key2type(&(key
->on_disk_key
)));
2020 dquot_free_space_nodirty(inode
, pasted_size
);
2024 /* Insert new item into the buffer at the path.
2025 * th - active transaction handle
2026 * path - path to the inserted item
2027 * ih - pointer to the item header to insert
2028 * body - pointer to the bytes to insert
2030 int reiserfs_insert_item(struct reiserfs_transaction_handle
*th
,
2031 struct treepath
*path
, const struct cpu_key
*key
,
2032 struct item_head
*ih
, struct inode
*inode
,
2035 struct tree_balance s_ins_balance
;
2038 int quota_bytes
= 0;
2040 BUG_ON(!th
->t_trans_id
);
2042 if (inode
) { /* Do we count quotas for item? */
2043 fs_gen
= get_generation(inode
->i_sb
);
2044 quota_bytes
= ih_item_len(ih
);
2046 /* hack so the quota code doesn't have to guess if the file has
2047 ** a tail, links are always tails, so there's no guessing needed
2049 if (!S_ISLNK(inode
->i_mode
) && is_direct_le_ih(ih
))
2050 quota_bytes
= inode
->i_sb
->s_blocksize
+ UNFM_P_SIZE
;
2051 #ifdef REISERQUOTA_DEBUG
2052 reiserfs_debug(inode
->i_sb
, REISERFS_DEBUG_CODE
,
2053 "reiserquota insert_item(): allocating %u id=%u type=%c",
2054 quota_bytes
, inode
->i_uid
, head2type(ih
));
2056 /* We can't dirty inode here. It would be immediately written but
2057 * appropriate stat item isn't inserted yet... */
2058 retval
= dquot_alloc_space_nodirty(inode
, quota_bytes
);
2064 init_tb_struct(th
, &s_ins_balance
, th
->t_super
, path
,
2065 IH_SIZE
+ ih_item_len(ih
));
2066 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2067 s_ins_balance
.key
= key
->on_disk_key
;
2069 /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2070 if (inode
&& fs_changed(fs_gen
, inode
->i_sb
)) {
2075 fix_nodes(M_INSERT
, &s_ins_balance
, ih
,
2076 body
)) == REPEAT_SEARCH
) {
2078 /* file system changed while we were in the fix_nodes */
2079 PROC_INFO_INC(th
->t_super
, insert_item_restarted
);
2080 retval
= search_item(th
->t_super
, key
, path
);
2081 if (retval
== IO_ERROR
) {
2085 if (retval
== ITEM_FOUND
) {
2086 reiserfs_warning(th
->t_super
, "PAP-5760",
2087 "key %K already exists in the tree",
2094 /* make balancing after all resources will be collected at a time */
2095 if (retval
== CARRY_ON
) {
2096 do_balance(&s_ins_balance
, ih
, body
, M_INSERT
);
2100 retval
= (retval
== NO_DISK_SPACE
) ? -ENOSPC
: -EIO
;
2102 /* also releases the path */
2103 unfix_nodes(&s_ins_balance
);
2104 #ifdef REISERQUOTA_DEBUG
2105 reiserfs_debug(th
->t_super
, REISERFS_DEBUG_CODE
,
2106 "reiserquota insert_item(): freeing %u id=%u type=%c",
2107 quota_bytes
, inode
->i_uid
, head2type(ih
));
2110 dquot_free_space_nodirty(inode
, quota_bytes
);