blob: 77df82f9e70a2090fbf6a04c17b6becc830d7c89 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 * reiserfs_check_path
27 * pathrelse_and_restore
28 * pathrelse
29 * search_by_key_reada
30 * search_by_key
31 * search_for_position_by_key
32 * comp_items
33 * prepare_for_direct_item
34 * prepare_for_direntry_item
35 * prepare_for_delete_or_cut
36 * calc_deleted_bytes_number
37 * init_tb_struct
38 * padd_item
39 * reiserfs_delete_item
40 * reiserfs_delete_solid_item
41 * reiserfs_delete_object
42 * maybe_indirect_to_direct
43 * indirect_to_direct_roll_back
44 * reiserfs_cut_from_item
45 * truncate_directory
46 * reiserfs_do_truncate
47 * reiserfs_paste_into_item
48 * reiserfs_insert_item
49 */
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/time.h>
52#include <linux/string.h>
53#include <linux/pagemap.h>
54#include <linux/reiserfs_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/buffer_head.h>
56#include <linux/quotaops.h>
57
58/* Does the buffer contain a disk block which is in the tree. */
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -040059inline int B_IS_IN_TREE(const struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070060{
61
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -040062 RFALSE(B_LEVEL(bh) > MAX_HEIGHT,
63 "PAP-1010: block (%b) has too big level (%z)", bh, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -040065 return (B_LEVEL(bh) != FREE_LEVEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066}
67
68//
69// to gets item head in le form
70//
Jeff Mahoneyd68caa92009-03-30 14:02:49 -040071inline void copy_item_head(struct item_head *to,
72 const struct item_head *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073{
Jeff Mahoneyd68caa92009-03-30 14:02:49 -040074 memcpy(to, from, IH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075}
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/* 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.
Jeff Mahoney0222e652009-03-30 14:02:44 -040080 Returns: -1 if key1 < key2
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 0 if key1 == key2
82 1 if key1 > key2 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070083inline int comp_short_keys(const struct reiserfs_key *le_key,
84 const struct cpu_key *cpu_key)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -070086 __u32 n;
87 n = le32_to_cpu(le_key->k_dir_id);
88 if (n < cpu_key->on_disk_key.k_dir_id)
89 return -1;
90 if (n > cpu_key->on_disk_key.k_dir_id)
91 return 1;
92 n = le32_to_cpu(le_key->k_objectid);
93 if (n < cpu_key->on_disk_key.k_objectid)
94 return -1;
95 if (n > cpu_key->on_disk_key.k_objectid)
96 return 1;
97 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098}
99
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100/* 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 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700105static inline int comp_keys(const struct reiserfs_key *le_key,
106 const struct cpu_key *cpu_key)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700108 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700110 retval = comp_short_keys(le_key, cpu_key);
111 if (retval)
112 return retval;
113 if (le_key_k_offset(le_key_version(le_key), le_key) <
114 cpu_key_k_offset(cpu_key))
115 return -1;
116 if (le_key_k_offset(le_key_version(le_key), le_key) >
117 cpu_key_k_offset(cpu_key))
118 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700120 if (cpu_key->key_length == 3)
121 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700123 /* this part is needed only when tail conversion is in progress */
124 if (le_key_k_type(le_key_version(le_key), le_key) <
125 cpu_key_k_type(cpu_key))
126 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700128 if (le_key_k_type(le_key_version(le_key), le_key) >
129 cpu_key_k_type(cpu_key))
130 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700132 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133}
134
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700135inline int comp_short_le_keys(const struct reiserfs_key *key1,
136 const struct reiserfs_key *key2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137{
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400138 __u32 *k1_u32, *k2_u32;
Jeff Mahoneyee939612009-03-30 14:02:50 -0400139 int key_length = REISERFS_SHORT_KEY_LEN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400141 k1_u32 = (__u32 *) key1;
142 k2_u32 = (__u32 *) key2;
Jeff Mahoneyee939612009-03-30 14:02:50 -0400143 for (; key_length--; ++k1_u32, ++k2_u32) {
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400144 if (le32_to_cpu(*k1_u32) < le32_to_cpu(*k2_u32))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700145 return -1;
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400146 if (le32_to_cpu(*k1_u32) > le32_to_cpu(*k2_u32))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700147 return 1;
148 }
149 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150}
151
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700152inline void le_key2cpu_key(struct cpu_key *to, const struct reiserfs_key *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700154 int version;
155 to->on_disk_key.k_dir_id = le32_to_cpu(from->k_dir_id);
156 to->on_disk_key.k_objectid = le32_to_cpu(from->k_objectid);
157
158 // find out version of the key
159 version = le_key_version(from);
160 to->version = version;
161 to->on_disk_key.k_offset = le_key_k_offset(version, from);
162 to->on_disk_key.k_type = le_key_k_type(version, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163}
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165// this does not say which one is bigger, it only returns 1 if keys
166// are not equal, 0 otherwise
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700167inline int comp_le_keys(const struct reiserfs_key *k1,
168 const struct reiserfs_key *k2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700170 return memcmp(k1, k2, sizeof(struct reiserfs_key));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171}
172
173/**************************************************************************
174 * Binary search toolkit function *
175 * Search for an item in the array by the item key *
176 * Returns: 1 if found, 0 if not found; *
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400177 * *pos = number of the searched element if found, else the *
178 * number of the first element that is larger than key. *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 **************************************************************************/
Jeff Mahoneyee939612009-03-30 14:02:50 -0400180/* 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 there are no possible items, and we have not found it. With each examination we
185 cut the number of possible items it could be by one more than half rounded down,
186 or we find it. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400187static inline int bin_search(const void *key, /* Key to search for. */
188 const void *base, /* First item in the array. */
189 int num, /* Number of items in the array. */
190 int width, /* Item size in the array.
191 searched. Lest the reader be
192 confused, note that this is crafted
193 as a general function, and when it
194 is applied specifically to the array
195 of item headers in a node, width
196 is actually the item header size not
197 the item size. */
198 int *pos /* Number of the searched for element. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700199 )
200{
Jeff Mahoneyee939612009-03-30 14:02:50 -0400201 int rbound, lbound, j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Jeff Mahoneyee939612009-03-30 14:02:50 -0400203 for (j = ((rbound = num - 1) + (lbound = 0)) / 2;
204 lbound <= rbound; j = (rbound + lbound) / 2)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700205 switch (comp_keys
Jeff Mahoneyee939612009-03-30 14:02:50 -0400206 ((struct reiserfs_key *)((char *)base + j * width),
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400207 (struct cpu_key *)key)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700208 case -1:
Jeff Mahoneyee939612009-03-30 14:02:50 -0400209 lbound = j + 1;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700210 continue;
211 case 1:
Jeff Mahoneyee939612009-03-30 14:02:50 -0400212 rbound = j - 1;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700213 continue;
214 case 0:
Jeff Mahoneyee939612009-03-30 14:02:50 -0400215 *pos = j;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700216 return ITEM_FOUND; /* Key found in the array. */
217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700219 /* bin_search did not find given key, it returns position of key,
220 that is minimal and greater than the given one. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400221 *pos = lbound;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700222 return ITEM_NOT_FOUND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226/* Minimal possible key. It is never in the tree. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700227const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
229/* Maximal possible key. It is never in the tree. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700230static const struct reiserfs_key MAX_KEY = {
Al Viro3e8962b2005-05-01 08:59:18 -0700231 __constant_cpu_to_le32(0xffffffff),
232 __constant_cpu_to_le32(0xffffffff),
233 {{__constant_cpu_to_le32(0xffffffff),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700234 __constant_cpu_to_le32(0xffffffff)},}
Al Viro3e8962b2005-05-01 08:59:18 -0700235};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237/* 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. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400241static inline const struct reiserfs_key *get_lkey(const struct treepath *chk_path,
242 const struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700243{
Jeff Mahoneyee939612009-03-30 14:02:50 -0400244 int position, path_offset = chk_path->path_length;
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400245 struct buffer_head *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Jeff Mahoneyee939612009-03-30 14:02:50 -0400247 RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700248 "PAP-5010: invalid offset in the path");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700250 /* While not higher in path than first element. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400251 while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700253 RFALSE(!buffer_uptodate
Jeff Mahoneyee939612009-03-30 14:02:50 -0400254 (PATH_OFFSET_PBUFFER(chk_path, path_offset)),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700255 "PAP-5020: parent is not uptodate");
256
257 /* Parent at the path is not in the tree now. */
258 if (!B_IS_IN_TREE
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400259 (parent =
Jeff Mahoneyee939612009-03-30 14:02:50 -0400260 PATH_OFFSET_PBUFFER(chk_path, path_offset)))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700261 return &MAX_KEY;
262 /* Check whether position in the parent is correct. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400263 if ((position =
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400264 PATH_OFFSET_POSITION(chk_path,
Jeff Mahoneyee939612009-03-30 14:02:50 -0400265 path_offset)) >
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400266 B_NR_ITEMS(parent))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700267 return &MAX_KEY;
268 /* Check whether parent at the path really points to the child. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400269 if (B_N_CHILD_NUM(parent, position) !=
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400270 PATH_OFFSET_PBUFFER(chk_path,
Jeff Mahoneyee939612009-03-30 14:02:50 -0400271 path_offset + 1)->b_blocknr)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700272 return &MAX_KEY;
273 /* Return delimiting key if position in the parent is not equal to zero. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400274 if (position)
275 return B_N_PDELIM_KEY(parent, position - 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700276 }
277 /* Return MIN_KEY if we are in the root of the buffer tree. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400278 if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)->
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400279 b_blocknr == SB_ROOT_BLOCK(sb))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700280 return &MIN_KEY;
281 return &MAX_KEY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282}
283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284/* Get delimiting key of the buffer at the path and its right neighbor. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400285inline const struct reiserfs_key *get_rkey(const struct treepath *chk_path,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400286 const struct super_block *sb)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700287{
Jeff Mahoneyee939612009-03-30 14:02:50 -0400288 int position, path_offset = chk_path->path_length;
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400289 struct buffer_head *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
Jeff Mahoneyee939612009-03-30 14:02:50 -0400291 RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700292 "PAP-5030: invalid offset in the path");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Jeff Mahoneyee939612009-03-30 14:02:50 -0400294 while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700296 RFALSE(!buffer_uptodate
Jeff Mahoneyee939612009-03-30 14:02:50 -0400297 (PATH_OFFSET_PBUFFER(chk_path, path_offset)),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700298 "PAP-5040: parent is not uptodate");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700300 /* Parent at the path is not in the tree now. */
301 if (!B_IS_IN_TREE
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400302 (parent =
Jeff Mahoneyee939612009-03-30 14:02:50 -0400303 PATH_OFFSET_PBUFFER(chk_path, path_offset)))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700304 return &MIN_KEY;
305 /* Check whether position in the parent is correct. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400306 if ((position =
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400307 PATH_OFFSET_POSITION(chk_path,
Jeff Mahoneyee939612009-03-30 14:02:50 -0400308 path_offset)) >
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400309 B_NR_ITEMS(parent))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700310 return &MIN_KEY;
311 /* Check whether parent at the path really points to the child. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400312 if (B_N_CHILD_NUM(parent, position) !=
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400313 PATH_OFFSET_PBUFFER(chk_path,
Jeff Mahoneyee939612009-03-30 14:02:50 -0400314 path_offset + 1)->b_blocknr)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700315 return &MIN_KEY;
316 /* Return delimiting key if position in the parent is not the last one. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400317 if (position != B_NR_ITEMS(parent))
318 return B_N_PDELIM_KEY(parent, position);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700319 }
320 /* Return MAX_KEY if we are in the root of the buffer tree. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400321 if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)->
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400322 b_blocknr == SB_ROOT_BLOCK(sb))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700323 return &MAX_KEY;
324 return &MIN_KEY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325}
326
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327/* 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. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400332static 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
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700335 )
336{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400338 RFALSE(!key || chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET
339 || chk_path->path_length > MAX_HEIGHT,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700340 "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400341 key, chk_path->path_length);
342 RFALSE(!PATH_PLAST_BUFFER(chk_path)->b_bdev,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700343 "PAP-5060: device must not be NODEV");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400345 if (comp_keys(get_lkey(chk_path, sb), key) == 1)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700346 /* left delimiting key is bigger, that the key we look for */
347 return 0;
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400348 /* 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 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700351 return 0;
352 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353}
354
Josef "Jeff" Sipekfec6d052006-12-08 02:36:32 -0800355int reiserfs_check_path(struct treepath *p)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700356{
357 RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
358 "path not properly relsed");
359 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360}
361
Jeff Mahoney3cd6dbe2009-03-30 14:02:43 -0400362/* 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() */
365void pathrelse_and_restore(struct super_block *sb,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400366 struct treepath *search_path)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700367{
Jeff Mahoneyee939612009-03-30 14:02:50 -0400368 int path_offset = search_path->path_length;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
Jeff Mahoneyee939612009-03-30 14:02:50 -0400370 RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700371 "clm-4000: invalid path offset");
372
Jeff Mahoneyee939612009-03-30 14:02:50 -0400373 while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
Jeff Mahoney3cd6dbe2009-03-30 14:02:43 -0400374 struct buffer_head *bh;
Jeff Mahoneyee939612009-03-30 14:02:50 -0400375 bh = PATH_OFFSET_PBUFFER(search_path, path_offset--);
Jeff Mahoney3cd6dbe2009-03-30 14:02:43 -0400376 reiserfs_restore_prepared_buffer(sb, bh);
377 brelse(bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700378 }
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400379 search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380}
381
Jeff Mahoney3cd6dbe2009-03-30 14:02:43 -0400382/* Drop the reference to each buffer in a path */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400383void pathrelse(struct treepath *search_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384{
Jeff Mahoneyee939612009-03-30 14:02:50 -0400385 int path_offset = search_path->path_length;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Jeff Mahoneyee939612009-03-30 14:02:50 -0400387 RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700388 "PAP-5090: invalid path offset");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Jeff Mahoneyee939612009-03-30 14:02:50 -0400390 while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
391 brelse(PATH_OFFSET_PBUFFER(search_path, path_offset--));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400393 search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
395
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700396static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
397{
398 struct block_head *blkh;
399 struct item_head *ih;
400 int used_space;
401 int prev_location;
402 int i;
403 int nr;
404
405 blkh = (struct block_head *)buf;
406 if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400407 reiserfs_warning(NULL, "reiserfs-5080",
408 "this should be caught earlier");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700409 return 0;
410 }
411
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 */
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400415 reiserfs_warning(NULL, "reiserfs-5081",
416 "nr_item seems wrong: %z", bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700417 return 0;
418 }
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 */
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400423 reiserfs_warning(NULL, "reiserfs-5082",
424 "free space seems wrong: %z", bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700425 return 0;
426 }
427 // FIXME: it is_leaf will hit performance too much - we may have
428 // return 1 here
429
430 /* check tables of item heads */
431 ih = (struct item_head *)(buf + BLKH_SIZE);
432 prev_location = blocksize;
433 for (i = 0; i < nr; i++, ih++) {
434 if (le_ih_k_type(ih) == TYPE_ANY) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400435 reiserfs_warning(NULL, "reiserfs-5083",
436 "wrong item type for item %h",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700437 ih);
438 return 0;
439 }
440 if (ih_location(ih) >= blocksize
441 || ih_location(ih) < IH_SIZE * nr) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400442 reiserfs_warning(NULL, "reiserfs-5084",
443 "item location seems wrong: %h",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700444 ih);
445 return 0;
446 }
447 if (ih_item_len(ih) < 1
448 || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400449 reiserfs_warning(NULL, "reiserfs-5085",
450 "item length seems wrong: %h",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700451 ih);
452 return 0;
453 }
454 if (prev_location - ih_location(ih) != ih_item_len(ih)) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400455 reiserfs_warning(NULL, "reiserfs-5086",
456 "item location seems wrong "
457 "(second one): %h", ih);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700458 return 0;
459 }
460 prev_location = ih_location(ih);
461 }
462
463 // one may imagine much more checks
464 return 1;
465}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
467/* returns 1 if buf looks like an internal node, 0 otherwise */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700468static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700470 struct block_head *blkh;
471 int nr;
472 int used_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700474 blkh = (struct block_head *)buf;
475 nr = blkh_level(blkh);
476 if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
477 /* this level is not possible for internal nodes */
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400478 reiserfs_warning(NULL, "reiserfs-5087",
479 "this should be caught earlier");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700480 return 0;
481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700483 nr = blkh_nr_item(blkh);
484 if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
485 /* for internal which is not root we might check min number of keys */
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400486 reiserfs_warning(NULL, "reiserfs-5088",
487 "number of key seems wrong: %z", bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700488 return 0;
489 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700491 used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
492 if (used_space != blocksize - blkh_free_space(blkh)) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400493 reiserfs_warning(NULL, "reiserfs-5089",
494 "free space seems wrong: %z", bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700495 return 0;
496 }
497 // one may imagine much more checks
498 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499}
500
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501// make sure that bh contains formatted node of reiserfs tree of
502// 'level'-th level
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700503static int is_tree_node(struct buffer_head *bh, int level)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700505 if (B_LEVEL(bh) != level) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400506 reiserfs_warning(NULL, "reiserfs-5090", "node level %d does "
507 "not match to the expected one %d",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700508 B_LEVEL(bh), level);
509 return 0;
510 }
511 if (level == DISK_LEAF_NODE_LEVEL)
512 return is_leaf(bh->b_data, bh->b_size, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700514 return is_internal(bh->b_data, bh->b_size, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515}
516
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517#define SEARCH_BY_KEY_READA 16
518
Frederic Weisbecker2ac62692009-05-14 02:56:39 +0200519/*
520 * The function is NOT SCHEDULE-SAFE!
521 * It might unlock the write lock if we needed to wait for a block
522 * to be read. Note that in this case it won't recover the lock to avoid
523 * high contention resulting from too much lock requests, especially
524 * the caller (search_by_key) will perform other schedule-unsafe
525 * operations just after calling this function.
526 *
527 * @return true if we have unlocked
528 */
529static bool search_by_key_reada(struct super_block *s,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700530 struct buffer_head **bh,
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700531 b_blocknr_t *b, int num)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700533 int i, j;
Frederic Weisbecker2ac62692009-05-14 02:56:39 +0200534 bool unlocked = false;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700535
536 for (i = 0; i < num; i++) {
537 bh[i] = sb_getblk(s, b[i]);
538 }
Frederic Weisbecker09eb47a2009-05-08 14:21:33 +0200539 /*
540 * We are going to read some blocks on which we
541 * have a reference. It's safe, though we might be
542 * reading blocks concurrently changed if we release
543 * the lock. But it's still fine because we check later
544 * if the tree changed
545 */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700546 for (j = 0; j < i; j++) {
547 /*
548 * note, this needs attention if we are getting rid of the BKL
549 * you have to make sure the prepared bit isn't set on this buffer
550 */
Frederic Weisbecker2ac62692009-05-14 02:56:39 +0200551 if (!buffer_uptodate(bh[j])) {
552 if (!unlocked) {
553 reiserfs_write_unlock(s);
554 unlocked = true;
555 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700556 ll_rw_block(READA, 1, bh + j);
Frederic Weisbecker2ac62692009-05-14 02:56:39 +0200557 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700558 brelse(bh[j]);
559 }
Frederic Weisbecker2ac62692009-05-14 02:56:39 +0200560 return unlocked;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561}
562
563/**************************************************************************
564 * Algorithm SearchByKey *
565 * look for item in the Disk S+Tree by its key *
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400566 * Input: sb - super block *
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400567 * key - pointer to the key to search *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR *
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400569 * search_path - path from the root to the needed leaf *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 **************************************************************************/
571
572/* This function fills up the path from the root to the leaf as it
573 descends the tree looking for the key. It uses reiserfs_bread to
574 try to find buffers in the cache given their block number. If it
575 does not find them in the cache it reads them from disk. For each
576 node search_by_key finds using reiserfs_bread it then uses
577 bin_search to look through that node. bin_search will find the
578 position of the block_number of the next node if it is looking
579 through an internal node. If it is looking through a leaf node
580 bin_search will find the position of the item which has key either
581 equal to given key, or which is the maximal key less than the given
582 key. search_by_key returns a path that must be checked for the
583 correctness of the top of the path but need not be checked for the
584 correctness of the bottom of the path */
585/* The function is NOT SCHEDULE-SAFE! */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400586int search_by_key(struct super_block *sb, const struct cpu_key *key, /* Key to search. */
587 struct treepath *search_path,/* This structure was
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700588 allocated and initialized
589 by the calling
590 function. It is filled up
591 by this function. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400592 int stop_level /* How far down the tree to search. To
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700593 stop at leaf level - set to
594 DISK_LEAF_NODE_LEVEL */
595 )
596{
Jeff Mahoneyee939612009-03-30 14:02:50 -0400597 b_blocknr_t block_number;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700598 int expected_level;
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400599 struct buffer_head *bh;
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400600 struct path_element *last_element;
Jeff Mahoneyee939612009-03-30 14:02:50 -0400601 int node_level, retval;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700602 int right_neighbor_of_leaf_node;
603 int fs_gen;
604 struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
Jeff Mahoney3ee16672007-10-18 23:39:25 -0700605 b_blocknr_t reada_blocks[SEARCH_BY_KEY_READA];
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700606 int reada_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoneyee939612009-03-30 14:02:50 -0400609 int repeat_counter = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400612 PROC_INFO_INC(sb, search_by_key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700614 /* As we add each node to a path we increase its count. This means that
615 we must be careful to release all nodes in a path before we either
616 discard the path struct or re-use the path struct, as we do here. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400618 pathrelse(search_path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700619
620 right_neighbor_of_leaf_node = 0;
621
622 /* With each iteration of this loop we search through the items in the
623 current node, and calculate the next current node(next path element)
624 for the next iteration of this loop.. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400625 block_number = SB_ROOT_BLOCK(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700626 expected_level = -1;
627 while (1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
629#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoneyee939612009-03-30 14:02:50 -0400630 if (!(++repeat_counter % 50000))
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400631 reiserfs_warning(sb, "PAP-5100",
Jeff Mahoney45b03d52009-03-30 14:02:21 -0400632 "%s: there were %d iterations of "
633 "while loop looking for key %K",
Jeff Mahoneyee939612009-03-30 14:02:50 -0400634 current->comm, repeat_counter,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400635 key);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636#endif
637
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700638 /* prep path to have another element added to it. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400639 last_element =
640 PATH_OFFSET_PELEMENT(search_path,
641 ++search_path->path_length);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400642 fs_gen = get_generation(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700644 /* Read the next tree node, and set the last element in the path to
645 have a pointer to it. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400646 if ((bh = last_element->pe_buffer =
Jeff Mahoneyee939612009-03-30 14:02:50 -0400647 sb_getblk(sb, block_number))) {
Frederic Weisbecker2ac62692009-05-14 02:56:39 +0200648 bool unlocked = false;
649
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400650 if (!buffer_uptodate(bh) && reada_count > 1)
Frederic Weisbecker2ac62692009-05-14 02:56:39 +0200651 /* may unlock the write lock */
652 unlocked = search_by_key_reada(sb, reada_bh,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700653 reada_blocks, reada_count);
Frederic Weisbecker2ac62692009-05-14 02:56:39 +0200654 /*
655 * If we haven't already unlocked the write lock,
656 * then we need to do that here before reading
657 * the current block
658 */
659 if (!buffer_uptodate(bh) && !unlocked) {
Frederic Weisbecker09eb47a2009-05-08 14:21:33 +0200660 reiserfs_write_unlock(sb);
Frederic Weisbecker2ac62692009-05-14 02:56:39 +0200661 unlocked = true;
662 }
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400663 ll_rw_block(READ, 1, &bh);
664 wait_on_buffer(bh);
Frederic Weisbecker2ac62692009-05-14 02:56:39 +0200665
666 if (unlocked)
667 reiserfs_write_lock(sb);
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400668 if (!buffer_uptodate(bh))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700669 goto io_error;
670 } else {
671 io_error:
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400672 search_path->path_length--;
673 pathrelse(search_path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700674 return IO_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700676 reada_count = 0;
677 if (expected_level == -1)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400678 expected_level = SB_TREE_HEIGHT(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700679 expected_level--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700681 /* It is possible that schedule occurred. We must check whether the key
682 to search is still in the tree rooted from the current buffer. If
683 not then repeat search from the root. */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400684 if (fs_changed(fs_gen, sb) &&
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400685 (!B_IS_IN_TREE(bh) ||
686 B_LEVEL(bh) != expected_level ||
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400687 !key_in_buffer(search_path, key, sb))) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400688 PROC_INFO_INC(sb, search_by_key_fs_changed);
689 PROC_INFO_INC(sb, search_by_key_restarted);
690 PROC_INFO_INC(sb,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700691 sbk_restarted[expected_level - 1]);
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400692 pathrelse(search_path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700693
694 /* Get the root block number so that we can repeat the search
695 starting from the root. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400696 block_number = SB_ROOT_BLOCK(sb);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700697 expected_level = -1;
698 right_neighbor_of_leaf_node = 0;
699
700 /* repeat search from the root */
701 continue;
702 }
703
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400704 /* only check that the key is in the buffer if key is not
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700705 equal to the MAX_KEY. Latter case is only possible in
706 "finish_unfinished()" processing during mount. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400707 RFALSE(comp_keys(&MAX_KEY, key) &&
708 !key_in_buffer(search_path, key, sb),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700709 "PAP-5130: key is not in the buffer");
710#ifdef CONFIG_REISERFS_CHECK
Frederic Weisbecker08f14fc2009-05-16 19:10:38 +0200711 if (REISERFS_SB(sb)->cur_tb) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700712 print_cur_tb("5140");
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400713 reiserfs_panic(sb, "PAP-5140",
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -0400714 "schedule occurred in do_balance!");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700715 }
716#endif
717
718 // make sure, that the node contents look like a node of
719 // certain level
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400720 if (!is_tree_node(bh, expected_level)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400721 reiserfs_error(sb, "vs-5150",
Jeff Mahoney0030b642009-03-30 14:02:28 -0400722 "invalid format found in block %ld. "
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400723 "Fsck?", bh->b_blocknr);
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400724 pathrelse(search_path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700725 return IO_ERROR;
726 }
727
728 /* ok, we have acquired next formatted node in the tree */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400729 node_level = B_LEVEL(bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700730
Jeff Mahoneyee939612009-03-30 14:02:50 -0400731 PROC_INFO_BH_STAT(sb, bh, node_level - 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700732
Jeff Mahoneyee939612009-03-30 14:02:50 -0400733 RFALSE(node_level < stop_level,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700734 "vs-5152: tree level (%d) is less than stop level (%d)",
Jeff Mahoneyee939612009-03-30 14:02:50 -0400735 node_level, stop_level);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700736
Jeff Mahoneyee939612009-03-30 14:02:50 -0400737 retval = bin_search(key, B_N_PITEM_HEAD(bh, 0),
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400738 B_NR_ITEMS(bh),
Jeff Mahoneyee939612009-03-30 14:02:50 -0400739 (node_level ==
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700740 DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
741 KEY_SIZE,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400742 &(last_element->pe_position));
Jeff Mahoneyee939612009-03-30 14:02:50 -0400743 if (node_level == stop_level) {
744 return retval;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700745 }
746
747 /* we are not in the stop level */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400748 if (retval == ITEM_FOUND)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700749 /* item has been found, so we choose the pointer which is to the right of the found one */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400750 last_element->pe_position++;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700751
752 /* if item was not found we choose the position which is to
753 the left of the found item. This requires no code,
754 bin_search did it already. */
755
756 /* So we have chosen a position in the current node which is
757 an internal node. Now we calculate child block number by
758 position in the node. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400759 block_number =
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400760 B_N_CHILD_NUM(bh, last_element->pe_position);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700761
762 /* if we are going to read leaf nodes, try for read ahead as well */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400763 if ((search_path->reada & PATH_READA) &&
Jeff Mahoneyee939612009-03-30 14:02:50 -0400764 node_level == DISK_LEAF_NODE_LEVEL + 1) {
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400765 int pos = last_element->pe_position;
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400766 int limit = B_NR_ITEMS(bh);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700767 struct reiserfs_key *le_key;
768
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400769 if (search_path->reada & PATH_READA_BACK)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700770 limit = 0;
771 while (reada_count < SEARCH_BY_KEY_READA) {
772 if (pos == limit)
773 break;
774 reada_blocks[reada_count++] =
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400775 B_N_CHILD_NUM(bh, pos);
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400776 if (search_path->reada & PATH_READA_BACK)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700777 pos--;
778 else
779 pos++;
780
781 /*
782 * check to make sure we're in the same object
783 */
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400784 le_key = B_N_PDELIM_KEY(bh, pos);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700785 if (le32_to_cpu(le_key->k_objectid) !=
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400786 key->on_disk_key.k_objectid) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700787 break;
788 }
789 }
790 }
791 }
792}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
794/* Form the path to an item and position in this item which contains
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400795 file byte defined by key. If there is no such item
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 corresponding to the key, we point the path to the item with
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400797 maximal key less than key, and *pos_in_item is set to one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 past the last entry/byte in the item. If searching for entry in a
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400799 directory item, and it is not found, *pos_in_item is set to one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 entry more than the entry with maximal key which is less than the
801 sought key.
802
803 Note that if there is no entry in this same node which is one more,
804 then we point to an imaginary entry. for direct items, the
805 position is in units of bytes, for indirect items the position is
806 in units of blocknr entries, for directory items the position is in
807 units of directory entries. */
808
809/* The function is NOT SCHEDULE-SAFE! */
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400810int search_for_position_by_key(struct super_block *sb, /* Pointer to the super block. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700811 const struct cpu_key *p_cpu_key, /* Key to search (cpu variable) */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400812 struct treepath *search_path /* Filled up by this function. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700813 )
814{
815 struct item_head *p_le_ih; /* pointer to on-disk structure */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400816 int blk_size;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700817 loff_t item_offset, offset;
818 struct reiserfs_dir_entry de;
819 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700821 /* If searching for directory entry. */
822 if (is_direntry_cpu_key(p_cpu_key))
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400823 return search_by_entry_key(sb, p_cpu_key, search_path,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700824 &de);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700826 /* If not searching for directory entry. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700828 /* If item is found. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400829 retval = search_item(sb, p_cpu_key, search_path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700830 if (retval == IO_ERROR)
831 return retval;
832 if (retval == ITEM_FOUND) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700834 RFALSE(!ih_item_len
835 (B_N_PITEM_HEAD
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400836 (PATH_PLAST_BUFFER(search_path),
837 PATH_LAST_POSITION(search_path))),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700838 "PAP-5165: item length equals zero");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400840 pos_in_item(search_path) = 0;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700841 return POSITION_FOUND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400844 RFALSE(!PATH_LAST_POSITION(search_path),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700845 "PAP-5170: position equals zero");
846
847 /* Item is not found. Set path to the previous item. */
848 p_le_ih =
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400849 B_N_PITEM_HEAD(PATH_PLAST_BUFFER(search_path),
850 --PATH_LAST_POSITION(search_path));
Jeff Mahoneyee939612009-03-30 14:02:50 -0400851 blk_size = sb->s_blocksize;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700852
853 if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) {
854 return FILE_NOT_FOUND;
855 }
856 // FIXME: quite ugly this far
857
858 item_offset = le_ih_k_offset(p_le_ih);
859 offset = cpu_key_k_offset(p_cpu_key);
860
861 /* Needed byte is contained in the item pointed to by the path. */
862 if (item_offset <= offset &&
Jeff Mahoneyee939612009-03-30 14:02:50 -0400863 item_offset + op_bytes_number(p_le_ih, blk_size) > offset) {
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400864 pos_in_item(search_path) = offset - item_offset;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700865 if (is_indirect_le_ih(p_le_ih)) {
Jeff Mahoneyee939612009-03-30 14:02:50 -0400866 pos_in_item(search_path) /= blk_size;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700867 }
868 return POSITION_FOUND;
869 }
870
871 /* Needed byte is not contained in the item pointed to by the
872 path. Set pos_in_item out of the item. */
873 if (is_indirect_le_ih(p_le_ih))
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400874 pos_in_item(search_path) =
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700875 ih_item_len(p_le_ih) / UNFM_P_SIZE;
876 else
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400877 pos_in_item(search_path) = ih_item_len(p_le_ih);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700878
879 return POSITION_NOT_FOUND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880}
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882/* Compare given item and item pointed to by the path. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400883int comp_items(const struct item_head *stored_ih, const struct treepath *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884{
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400885 struct buffer_head *bh = PATH_PLAST_BUFFER(path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700886 struct item_head *ih;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700888 /* Last buffer at the path is not in the tree. */
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -0400889 if (!B_IS_IN_TREE(bh))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700890 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700892 /* Last path position is invalid. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400893 if (PATH_LAST_POSITION(path) >= B_NR_ITEMS(bh))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700894 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700896 /* we need only to know, whether it is the same item */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400897 ih = get_ih(path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700898 return memcmp(stored_ih, ih, IH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899}
900
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901/* unformatted nodes are not logged anymore, ever. This is safe
902** now
903*/
904#define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
905
906// block can not be forgotten as it is in I/O or held by someone
907#define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
908
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909// prepare for delete or cut of direct item
Josef "Jeff" Sipekfec6d052006-12-08 02:36:32 -0800910static inline int prepare_for_direct_item(struct treepath *path,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700911 struct item_head *le_ih,
912 struct inode *inode,
913 loff_t new_file_length, int *cut_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700915 loff_t round_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700917 if (new_file_length == max_reiserfs_offset(inode)) {
918 /* item has to be deleted */
919 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
920 return M_DELETE;
921 }
922 // new file gets truncated
923 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
Jeff Mahoney0222e652009-03-30 14:02:44 -0400924 //
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700925 round_len = ROUND_UP(new_file_length);
Jeff Mahoneyee939612009-03-30 14:02:50 -0400926 /* this was new_file_length < le_ih ... */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700927 if (round_len < le_ih_k_offset(le_ih)) {
928 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
929 return M_DELETE; /* Delete this item. */
930 }
931 /* Calculate first position and size for cutting from item. */
932 pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
933 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700935 return M_CUT; /* Cut from this item. */
936 }
937
938 // old file: items may have any length
939
940 if (new_file_length < le_ih_k_offset(le_ih)) {
941 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
942 return M_DELETE; /* Delete this item. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 }
944 /* Calculate first position and size for cutting from item. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700945 *cut_size = -(ih_item_len(le_ih) -
946 (pos_in_item(path) =
947 new_file_length + 1 - le_ih_k_offset(le_ih)));
948 return M_CUT; /* Cut from this item. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949}
950
Josef "Jeff" Sipekfec6d052006-12-08 02:36:32 -0800951static inline int prepare_for_direntry_item(struct treepath *path,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700952 struct item_head *le_ih,
953 struct inode *inode,
954 loff_t new_file_length,
955 int *cut_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700957 if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
958 new_file_length == max_reiserfs_offset(inode)) {
959 RFALSE(ih_entry_count(le_ih) != 2,
960 "PAP-5220: incorrect empty directory item (%h)", le_ih);
961 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
962 return M_DELETE; /* Delete the directory item containing "." and ".." entry. */
963 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700965 if (ih_entry_count(le_ih) == 1) {
966 /* Delete the directory item such as there is one record only
967 in this item */
968 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
969 return M_DELETE;
970 }
971
972 /* Cut one record from the directory item. */
973 *cut_size =
974 -(DEH_SIZE +
975 entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
976 return M_CUT;
977}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -0800979#define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981/* If the path points to a directory or direct item, calculate mode and the size cut, for balance.
982 If the path points to an indirect item, remove some number of its unformatted nodes.
983 In case of file truncate calculate whether this item must be deleted/truncated or last
984 unformatted node of this item will be converted to a direct item.
985 This function returns a determination of what balance mode the calling function should employ. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400986static 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
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700987 from end of the file. */
Jeff Mahoneyee939612009-03-30 14:02:50 -0400988 int *cut_size, unsigned long long new_file_length /* MAX_KEY_OFFSET in case of delete. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700989 )
990{
Jeff Mahoneya9dd3642009-03-30 14:02:45 -0400991 struct super_block *sb = inode->i_sb;
Jeff Mahoneyd68caa92009-03-30 14:02:49 -0400992 struct item_head *p_le_ih = PATH_PITEM_HEAD(path);
993 struct buffer_head *bh = PATH_PLAST_BUFFER(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700995 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
Linus Torvaldsbd4c6252005-07-12 20:21:28 -0700997 /* Stat_data item. */
998 if (is_statdata_le_ih(p_le_ih)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999
Jeff Mahoneyee939612009-03-30 14:02:50 -04001000 RFALSE(new_file_length != max_reiserfs_offset(inode),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001001 "PAP-5210: mode must be M_DELETE");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001003 *cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001004 return M_DELETE;
1005 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001007 /* Directory item. */
1008 if (is_direntry_le_ih(p_le_ih))
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001009 return prepare_for_direntry_item(path, p_le_ih, inode,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001010 new_file_length,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001011 cut_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001013 /* Direct item. */
1014 if (is_direct_le_ih(p_le_ih))
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001015 return prepare_for_direct_item(path, p_le_ih, inode,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001016 new_file_length, cut_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001018 /* Case of an indirect item. */
1019 {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001020 int blk_size = sb->s_blocksize;
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001021 struct item_head s_ih;
1022 int need_re_search;
1023 int delete = 0;
1024 int result = M_CUT;
1025 int pos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
Jeff Mahoneyee939612009-03-30 14:02:50 -04001027 if ( new_file_length == max_reiserfs_offset (inode) ) {
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001028 /* prepare_for_delete_or_cut() is called by
1029 * reiserfs_delete_item() */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001030 new_file_length = 0;
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001031 delete = 1;
1032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001034 do {
1035 need_re_search = 0;
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001036 *cut_size = 0;
1037 bh = PATH_PLAST_BUFFER(path);
1038 copy_item_head(&s_ih, PATH_PITEM_HEAD(path));
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001039 pos = I_UNFM_NUM(&s_ih);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
Jeff Mahoneyee939612009-03-30 14:02:50 -04001041 while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > new_file_length) {
Al Viro87588dd2007-07-26 17:47:03 +01001042 __le32 *unfm;
1043 __u32 block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001045 /* Each unformatted block deletion may involve one additional
1046 * bitmap block into the transaction, thereby the initial
1047 * journal space reservation might not be enough. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001048 if (!delete && (*cut_size) != 0 &&
1049 reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD)
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001050 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -04001052 unfm = (__le32 *)B_I_PITEM(bh, &s_ih) + pos - 1;
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001053 block = get_block_num(unfm, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001055 if (block != 0) {
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -04001056 reiserfs_prepare_for_journal(sb, bh, 1);
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001057 put_block_num(unfm, 0, 0);
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -04001058 journal_mark_dirty(th, sb, bh);
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001059 reiserfs_free_block(th, inode, block, 1);
1060 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
Frederic Weisbecker5e69e3a2009-04-30 23:36:33 +02001062 reiserfs_write_unlock(sb);
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001063 cond_resched();
Frederic Weisbecker5e69e3a2009-04-30 23:36:33 +02001064 reiserfs_write_lock(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001066 if (item_moved (&s_ih, path)) {
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001067 need_re_search = 1;
1068 break;
1069 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001071 pos --;
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001072 (*removed)++;
1073 (*cut_size) -= UNFM_P_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001075 if (pos == 0) {
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001076 (*cut_size) -= IH_SIZE;
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001077 result = M_DELETE;
1078 break;
1079 }
1080 }
1081 /* a trick. If the buffer has been logged, this will do nothing. If
1082 ** we've broken the loop without logging it, it will restore the
1083 ** buffer */
Jeff Mahoneyad31a4f2009-03-30 14:02:46 -04001084 reiserfs_restore_prepared_buffer(sb, bh);
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001085 } while (need_re_search &&
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001086 search_for_position_by_key(sb, item_key, path) == POSITION_FOUND);
1087 pos_in_item(path) = pos * UNFM_P_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001089 if (*cut_size == 0) {
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001090 /* Nothing were cut. maybe convert last unformatted node to the
1091 * direct item? */
1092 result = M_CONVERT;
1093 }
1094 return result;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001095 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096}
1097
1098/* Calculate number of bytes which will be deleted or cut during balance */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001099static int calc_deleted_bytes_number(struct tree_balance *tb, char mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100{
Jeff Mahoneyee939612009-03-30 14:02:50 -04001101 int del_size;
Jeff Mahoneya063ae12009-03-30 14:02:48 -04001102 struct item_head *p_le_ih = PATH_PITEM_HEAD(tb->tb_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001104 if (is_statdata_le_ih(p_le_ih))
1105 return 0;
1106
Jeff Mahoneyee939612009-03-30 14:02:50 -04001107 del_size =
1108 (mode ==
Jeff Mahoneya063ae12009-03-30 14:02:48 -04001109 M_DELETE) ? ih_item_len(p_le_ih) : -tb->insert_size[0];
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001110 if (is_direntry_le_ih(p_le_ih)) {
Jeff Mahoneyee939612009-03-30 14:02:50 -04001111 /* return EMPTY_DIR_SIZE; We delete emty directoris only.
1112 * we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1113 * empty size. ick. FIXME, is this right? */
1114 return del_size;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001115 }
1116
1117 if (is_indirect_le_ih(p_le_ih))
Jeff Mahoneyee939612009-03-30 14:02:50 -04001118 del_size = (del_size / UNFM_P_SIZE) *
Jeff Mahoneya063ae12009-03-30 14:02:48 -04001119 (PATH_PLAST_BUFFER(tb->tb_path)->b_size);
Jeff Mahoneyee939612009-03-30 14:02:50 -04001120 return del_size;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001121}
1122
1123static void init_tb_struct(struct reiserfs_transaction_handle *th,
Jeff Mahoneya063ae12009-03-30 14:02:48 -04001124 struct tree_balance *tb,
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001125 struct super_block *sb,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001126 struct treepath *path, int size)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001127{
1128
1129 BUG_ON(!th->t_trans_id);
1130
Jeff Mahoneya063ae12009-03-30 14:02:48 -04001131 memset(tb, '\0', sizeof(struct tree_balance));
1132 tb->transaction_handle = th;
1133 tb->tb_sb = sb;
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001134 tb->tb_path = path;
1135 PATH_OFFSET_PBUFFER(path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1136 PATH_OFFSET_POSITION(path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
Jeff Mahoneyee939612009-03-30 14:02:50 -04001137 tb->insert_size[0] = size;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001138}
1139
1140void padd_item(char *item, int total_length, int length)
1141{
1142 int i;
1143
1144 for (i = total_length; i > length;)
1145 item[--i] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146}
1147
1148#ifdef REISERQUOTA_DEBUG
1149char key2type(struct reiserfs_key *ih)
1150{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001151 if (is_direntry_le_key(2, ih))
1152 return 'd';
1153 if (is_direct_le_key(2, ih))
1154 return 'D';
1155 if (is_indirect_le_key(2, ih))
1156 return 'i';
1157 if (is_statdata_le_key(2, ih))
1158 return 's';
1159 return 'u';
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160}
1161
1162char head2type(struct item_head *ih)
1163{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001164 if (is_direntry_le_ih(ih))
1165 return 'd';
1166 if (is_direct_le_ih(ih))
1167 return 'D';
1168 if (is_indirect_le_ih(ih))
1169 return 'i';
1170 if (is_statdata_le_ih(ih))
1171 return 's';
1172 return 'u';
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173}
1174#endif
1175
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001176/* Delete object item.
1177 * th - active transaction handle
1178 * path - path to the deleted item
1179 * item_key - key to search for the deleted item
1180 * indode - used for updating i_blocks and quotas
1181 * un_bh - NULL or unformatted node pointer
1182 */
1183int reiserfs_delete_item(struct reiserfs_transaction_handle *th,
1184 struct treepath *path, const struct cpu_key *item_key,
1185 struct inode *inode, struct buffer_head *un_bh)
1186{
Jeff Mahoney995c7622009-03-30 14:02:47 -04001187 struct super_block *sb = inode->i_sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001188 struct tree_balance s_del_balance;
1189 struct item_head s_ih;
1190 struct item_head *q_ih;
1191 int quota_cut_bytes;
Jeff Mahoneyee939612009-03-30 14:02:50 -04001192 int ret_value, del_size, removed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
1194#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoneyee939612009-03-30 14:02:50 -04001195 char mode;
1196 int iter = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197#endif
1198
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001199 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001201 init_tb_struct(th, &s_del_balance, sb, path,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001202 0 /*size is unknown */ );
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001204 while (1) {
Jeff Mahoneyee939612009-03-30 14:02:50 -04001205 removed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
1207#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoneyee939612009-03-30 14:02:50 -04001208 iter++;
1209 mode =
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210#endif
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001211 prepare_for_delete_or_cut(th, inode, path,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001212 item_key, &removed,
1213 &del_size,
Jeff Mahoney995c7622009-03-30 14:02:47 -04001214 max_reiserfs_offset(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215
Jeff Mahoneyee939612009-03-30 14:02:50 -04001216 RFALSE(mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001218 copy_item_head(&s_ih, PATH_PITEM_HEAD(path));
Jeff Mahoneyee939612009-03-30 14:02:50 -04001219 s_del_balance.insert_size[0] = del_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
Jeff Mahoneyee939612009-03-30 14:02:50 -04001221 ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1222 if (ret_value != REPEAT_SEARCH)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001223 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001225 PROC_INFO_INC(sb, delete_item_restarted);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001227 // file system changed, repeat search
Jeff Mahoneyee939612009-03-30 14:02:50 -04001228 ret_value =
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001229 search_for_position_by_key(sb, item_key, path);
Jeff Mahoneyee939612009-03-30 14:02:50 -04001230 if (ret_value == IO_ERROR)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001231 break;
Jeff Mahoneyee939612009-03-30 14:02:50 -04001232 if (ret_value == FILE_NOT_FOUND) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001233 reiserfs_warning(sb, "vs-5340",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001234 "no items of the file %K found",
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001235 item_key);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001236 break;
1237 }
1238 } /* while (1) */
1239
Jeff Mahoneyee939612009-03-30 14:02:50 -04001240 if (ret_value != CARRY_ON) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001241 unfix_nodes(&s_del_balance);
1242 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001244 // reiserfs_delete_item returns item length when success
Jeff Mahoneyee939612009-03-30 14:02:50 -04001245 ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001246 q_ih = get_ih(path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001247 quota_cut_bytes = ih_item_len(q_ih);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001249 /* hack so the quota code doesn't have to guess if the file
1250 ** has a tail. On tail insert, we allocate quota for 1 unformatted node.
1251 ** We test the offset because the tail might have been
1252 ** split into multiple items, and we only want to decrement for
1253 ** the unfm node once
1254 */
Jeff Mahoney995c7622009-03-30 14:02:47 -04001255 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(q_ih)) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001256 if ((le_ih_k_offset(q_ih) & (sb->s_blocksize - 1)) == 1) {
1257 quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001258 } else {
1259 quota_cut_bytes = 0;
1260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001263 if (un_bh) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001264 int off;
1265 char *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001267 /* We are in direct2indirect conversion, so move tail contents
1268 to the unformatted node */
1269 /* note, we do the copy before preparing the buffer because we
1270 ** don't care about the contents of the unformatted node yet.
1271 ** the only thing we really care about is the direct item's data
1272 ** is in the unformatted node.
1273 **
1274 ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1275 ** the unformatted node, which might schedule, meaning we'd have to
1276 ** loop all the way back up to the start of the while loop.
1277 **
1278 ** The unformatted node must be dirtied later on. We can't be
1279 ** sure here if the entire tail has been deleted yet.
1280 **
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001281 ** un_bh is from the page cache (all unformatted nodes are
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001282 ** from the page cache) and might be a highmem page. So, we
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001283 ** can't use un_bh->b_data.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001284 ** -clm
1285 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286
Cong Wang883da602011-11-25 23:14:35 +08001287 data = kmap_atomic(un_bh->b_page);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001288 off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1289 memcpy(data + off,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001290 B_I_PITEM(PATH_PLAST_BUFFER(path), &s_ih),
Jeff Mahoneyee939612009-03-30 14:02:50 -04001291 ret_value);
Cong Wang883da602011-11-25 23:14:35 +08001292 kunmap_atomic(data);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001293 }
1294 /* Perform balancing after all resources have been collected at once. */
1295 do_balance(&s_del_balance, NULL, NULL, M_DELETE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296
1297#ifdef REISERQUOTA_DEBUG
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001298 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001299 "reiserquota delete_item(): freeing %u, id=%u type=%c",
Jeff Mahoney995c7622009-03-30 14:02:47 -04001300 quota_cut_bytes, inode->i_uid, head2type(&s_ih));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301#endif
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001302 dquot_free_space_nodirty(inode, quota_cut_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001304 /* Return deleted body length */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001305 return ret_value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306}
1307
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308/* Summary Of Mechanisms For Handling Collisions Between Processes:
1309
1310 deletion of the body of the object is performed by iput(), with the
1311 result that if multiple processes are operating on a file, the
1312 deletion of the body of the file is deferred until the last process
1313 that has an open inode performs its iput().
1314
1315 writes and truncates are protected from collisions by use of
1316 semaphores.
1317
1318 creates, linking, and mknod are protected from collisions with other
1319 processes by making the reiserfs_add_entry() the last step in the
1320 creation, and then rolling back all changes if there was a collision.
1321 - Hans
1322*/
1323
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324/* this deletes item which never gets split */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001325void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1326 struct inode *inode, struct reiserfs_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001328 struct tree_balance tb;
1329 INITIALIZE_PATH(path);
1330 int item_len = 0;
1331 int tb_init = 0;
1332 struct cpu_key cpu_key;
1333 int retval;
1334 int quota_cut_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001336 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001338 le_key2cpu_key(&cpu_key, key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001340 while (1) {
1341 retval = search_item(th->t_super, &cpu_key, &path);
1342 if (retval == IO_ERROR) {
Jeff Mahoney0030b642009-03-30 14:02:28 -04001343 reiserfs_error(th->t_super, "vs-5350",
1344 "i/o failure occurred trying "
1345 "to delete %K", &cpu_key);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001346 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))
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001359 reiserfs_warning(th->t_super, "vs-5355",
1360 "%k not found", key);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001361 break;
1362 }
1363 if (!tb_init) {
1364 tb_init = 1;
1365 item_len = ih_item_len(PATH_PITEM_HEAD(&path));
1366 init_tb_struct(th, &tb, th->t_super, &path,
1367 -(IH_SIZE + item_len));
1368 }
1369 quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path));
1370
1371 retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1372 if (retval == REPEAT_SEARCH) {
1373 PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1374 continue;
1375 }
1376
1377 if (retval == CARRY_ON) {
1378 do_balance(&tb, NULL, NULL, M_DELETE);
1379 if (inode) { /* Should we count quota for item? (we don't count quotas for save-links) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380#ifdef REISERQUOTA_DEBUG
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001381 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1382 "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1383 quota_cut_bytes, inode->i_uid,
1384 key2type(key));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385#endif
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001386 dquot_free_space_nodirty(inode,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001387 quota_cut_bytes);
1388 }
1389 break;
1390 }
1391 // IO_ERROR, NO_DISK_SPACE, etc
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001392 reiserfs_warning(th->t_super, "vs-5360",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001393 "could not delete %K due to fix_nodes failure",
1394 &cpu_key);
1395 unfix_nodes(&tb);
1396 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 }
1398
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001399 reiserfs_check_path(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400}
1401
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001402int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1403 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001405 int err;
1406 inode->i_size = 0;
1407 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001409 /* for directory this deletes item containing "." and ".." */
1410 err =
1411 reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1412 if (err)
1413 return err;
1414
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415#if defined( USE_INODE_GENERATION_COUNTER )
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001416 if (!old_format_only(th->t_super)) {
1417 __le32 *inode_generation;
1418
1419 inode_generation =
1420 &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
Marcin Slusarz9e902df2008-04-28 02:16:20 -07001421 le32_add_cpu(inode_generation, 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001422 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423/* USE_INODE_GENERATION_COUNTER */
1424#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001425 reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001427 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428}
1429
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001430static void unmap_buffers(struct page *page, loff_t pos)
1431{
1432 struct buffer_head *bh;
1433 struct buffer_head *head;
1434 struct buffer_head *next;
1435 unsigned long tail_index;
1436 unsigned long cur_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001438 if (page) {
1439 if (page_has_buffers(page)) {
1440 tail_index = pos & (PAGE_CACHE_SIZE - 1);
1441 cur_index = 0;
1442 head = page_buffers(page);
1443 bh = head;
1444 do {
1445 next = bh->b_this_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001447 /* we want to unmap the buffers that contain the tail, and
1448 ** all the buffers after it (since the tail must be at the
1449 ** end of the file). We don't want to unmap file data
1450 ** before the tail, since it might be dirty and waiting to
1451 ** reach disk
1452 */
1453 cur_index += bh->b_size;
1454 if (cur_index > tail_index) {
1455 reiserfs_unmap_buffer(bh);
1456 }
1457 bh = next;
1458 } while (bh != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461}
1462
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001463static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
Jeff Mahoney995c7622009-03-30 14:02:47 -04001464 struct inode *inode,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001465 struct page *page,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001466 struct treepath *path,
1467 const struct cpu_key *item_key,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001468 loff_t new_file_size, char *mode)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001469{
Jeff Mahoney995c7622009-03-30 14:02:47 -04001470 struct super_block *sb = inode->i_sb;
Jeff Mahoneyee939612009-03-30 14:02:50 -04001471 int block_size = sb->s_blocksize;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001472 int cut_bytes;
1473 BUG_ON(!th->t_trans_id);
Jeff Mahoneyee939612009-03-30 14:02:50 -04001474 BUG_ON(new_file_size != inode->i_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001476 /* the page being sent in could be NULL if there was an i/o error
1477 ** reading in the last block. The user will hit problems trying to
1478 ** read the file, but for now we just skip the indirect2direct
1479 */
Jeff Mahoney995c7622009-03-30 14:02:47 -04001480 if (atomic_read(&inode->i_count) > 1 ||
1481 !tail_has_to_be_packed(inode) ||
1482 !page || (REISERFS_I(inode)->i_flags & i_nopack_mask)) {
Jeff Mahoney0222e652009-03-30 14:02:44 -04001483 /* leave tail in an unformatted node */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001484 *mode = M_SKIP_BALANCING;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001485 cut_bytes =
Jeff Mahoneyee939612009-03-30 14:02:50 -04001486 block_size - (new_file_size & (block_size - 1));
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001487 pathrelse(path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001488 return cut_bytes;
1489 }
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001490 /* Perform the conversion to a direct_item. */
1491 /* return indirect_to_direct(inode, path, item_key,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001492 new_file_size, mode); */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001493 return indirect2direct(th, inode, page, path, item_key,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001494 new_file_size, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495}
1496
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497/* we did indirect_to_direct conversion. And we have inserted direct
1498 item successesfully, but there were no disk space to cut unfm
1499 pointer being converted. Therefore we have to delete inserted
1500 direct item(s) */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001501static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
Josef "Jeff" Sipekfec6d052006-12-08 02:36:32 -08001502 struct inode *inode, struct treepath *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001504 struct cpu_key tail_key;
1505 int tail_len;
1506 int removed;
1507 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001509 make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4); // !!!!
1510 tail_key.key_length = 4;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001512 tail_len =
1513 (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1514 while (tail_len) {
1515 /* look for the last byte of the tail */
1516 if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1517 POSITION_NOT_FOUND)
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001518 reiserfs_panic(inode->i_sb, "vs-5615",
1519 "found invalid item");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001520 RFALSE(path->pos_in_item !=
1521 ih_item_len(PATH_PITEM_HEAD(path)) - 1,
1522 "vs-5616: appended bytes found");
1523 PATH_LAST_POSITION(path)--;
1524
1525 removed =
1526 reiserfs_delete_item(th, path, &tail_key, inode,
1527 NULL /*unbh not needed */ );
1528 RFALSE(removed <= 0
1529 || removed > tail_len,
1530 "vs-5617: there was tail %d bytes, removed item length %d bytes",
1531 tail_len, removed);
1532 tail_len -= removed;
1533 set_cpu_key_k_offset(&tail_key,
1534 cpu_key_k_offset(&tail_key) - removed);
1535 }
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001536 reiserfs_warning(inode->i_sb, "reiserfs-5091", "indirect_to_direct "
1537 "conversion has been rolled back due to "
1538 "lack of disk space");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001539 //mark_file_without_tail (inode);
1540 mark_inode_dirty(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541}
1542
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543/* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001544int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001545 struct treepath *path,
1546 struct cpu_key *item_key,
Jeff Mahoney995c7622009-03-30 14:02:47 -04001547 struct inode *inode,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001548 struct page *page, loff_t new_file_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549{
Jeff Mahoney995c7622009-03-30 14:02:47 -04001550 struct super_block *sb = inode->i_sb;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001551 /* Every function which is going to call do_balance must first
1552 create a tree_balance structure. Then it must fill up this
1553 structure by using the init_tb_struct and fix_nodes functions.
1554 After that we can make tree balancing. */
1555 struct tree_balance s_cut_balance;
1556 struct item_head *p_le_ih;
Jeff Mahoneyee939612009-03-30 14:02:50 -04001557 int cut_size = 0, /* Amount to be cut. */
1558 ret_value = CARRY_ON, removed = 0, /* Number of the removed unformatted nodes. */
1559 is_inode_locked = 0;
1560 char mode; /* Mode of the balance. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001561 int retval2 = -1;
1562 int quota_cut_bytes;
1563 loff_t tail_pos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001565 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001567 init_tb_struct(th, &s_cut_balance, inode->i_sb, path,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001568 cut_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001570 /* Repeat this loop until we either cut the item without needing
1571 to balance, or we fix_nodes without schedule occurring */
1572 while (1) {
1573 /* Determine the balance mode, position of the first byte to
1574 be cut, and size to be cut. In case of the indirect item
1575 free unformatted nodes which are pointed to by the cut
1576 pointers. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Jeff Mahoneyee939612009-03-30 14:02:50 -04001578 mode =
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001579 prepare_for_delete_or_cut(th, inode, path,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001580 item_key, &removed,
1581 &cut_size, new_file_size);
1582 if (mode == M_CONVERT) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001583 /* convert last unformatted node to direct item or leave
1584 tail in the unformatted node */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001585 RFALSE(ret_value != CARRY_ON,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001586 "PAP-5570: can not convert twice");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
Jeff Mahoneyee939612009-03-30 14:02:50 -04001588 ret_value =
Jeff Mahoney995c7622009-03-30 14:02:47 -04001589 maybe_indirect_to_direct(th, inode, page,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001590 path, item_key,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001591 new_file_size, &mode);
1592 if (mode == M_SKIP_BALANCING)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001593 /* tail has been left in the unformatted node */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001594 return ret_value;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001595
Jeff Mahoneyee939612009-03-30 14:02:50 -04001596 is_inode_locked = 1;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001597
1598 /* removing of last unformatted node will change value we
1599 have to return to truncate. Save it */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001600 retval2 = ret_value;
1601 /*retval2 = sb->s_blocksize - (new_file_size & (sb->s_blocksize - 1)); */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001602
1603 /* So, we have performed the first part of the conversion:
1604 inserting the new direct item. Now we are removing the
1605 last unformatted node pointer. Set key to search for
1606 it. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001607 set_cpu_key_k_type(item_key, TYPE_INDIRECT);
1608 item_key->key_length = 4;
Jeff Mahoneyee939612009-03-30 14:02:50 -04001609 new_file_size -=
1610 (new_file_size & (sb->s_blocksize - 1));
1611 tail_pos = new_file_size;
1612 set_cpu_key_k_offset(item_key, new_file_size + 1);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001613 if (search_for_position_by_key
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001614 (sb, item_key,
1615 path) == POSITION_NOT_FOUND) {
1616 print_block(PATH_PLAST_BUFFER(path), 3,
1617 PATH_LAST_POSITION(path) - 1,
1618 PATH_LAST_POSITION(path) + 1);
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001619 reiserfs_panic(sb, "PAP-5580", "item to "
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001620 "convert does not exist (%K)",
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001621 item_key);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001622 }
1623 continue;
1624 }
Jeff Mahoneyee939612009-03-30 14:02:50 -04001625 if (cut_size == 0) {
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001626 pathrelse(path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001627 return 0;
1628 }
1629
Jeff Mahoneyee939612009-03-30 14:02:50 -04001630 s_cut_balance.insert_size[0] = cut_size;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001631
Jeff Mahoneyee939612009-03-30 14:02:50 -04001632 ret_value = fix_nodes(mode, &s_cut_balance, NULL, NULL);
1633 if (ret_value != REPEAT_SEARCH)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001634 break;
1635
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001636 PROC_INFO_INC(sb, cut_from_item_restarted);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001637
Jeff Mahoneyee939612009-03-30 14:02:50 -04001638 ret_value =
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001639 search_for_position_by_key(sb, item_key, path);
Jeff Mahoneyee939612009-03-30 14:02:50 -04001640 if (ret_value == POSITION_FOUND)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001641 continue;
1642
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001643 reiserfs_warning(sb, "PAP-5610", "item %K not found",
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001644 item_key);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001645 unfix_nodes(&s_cut_balance);
Jeff Mahoneyee939612009-03-30 14:02:50 -04001646 return (ret_value == IO_ERROR) ? -EIO : -ENOENT;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001647 } /* while */
1648
1649 // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
Jeff Mahoneyee939612009-03-30 14:02:50 -04001650 if (ret_value != CARRY_ON) {
1651 if (is_inode_locked) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001652 // FIXME: this seems to be not needed: we are always able
1653 // to cut item
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001654 indirect_to_direct_roll_back(th, inode, path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001655 }
Jeff Mahoneyee939612009-03-30 14:02:50 -04001656 if (ret_value == NO_DISK_SPACE)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001657 reiserfs_warning(sb, "reiserfs-5092",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001658 "NO_DISK_SPACE");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001659 unfix_nodes(&s_cut_balance);
1660 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 }
1662
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001663 /* go ahead and perform balancing */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
Jeff Mahoneyee939612009-03-30 14:02:50 -04001665 RFALSE(mode == M_PASTE || mode == M_INSERT, "invalid mode");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001667 /* Calculate number of bytes that need to be cut from the item. */
1668 quota_cut_bytes =
Jeff Mahoneyee939612009-03-30 14:02:50 -04001669 (mode ==
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001670 M_DELETE) ? ih_item_len(get_ih(path)) : -s_cut_balance.
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001671 insert_size[0];
1672 if (retval2 == -1)
Jeff Mahoneyee939612009-03-30 14:02:50 -04001673 ret_value = calc_deleted_bytes_number(&s_cut_balance, mode);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001674 else
Jeff Mahoneyee939612009-03-30 14:02:50 -04001675 ret_value = retval2;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001676
1677 /* For direct items, we only change the quota when deleting the last
1678 ** item.
1679 */
1680 p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path);
Jeff Mahoney995c7622009-03-30 14:02:47 -04001681 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_le_ih)) {
Jeff Mahoneyee939612009-03-30 14:02:50 -04001682 if (mode == M_DELETE &&
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001683 (le_ih_k_offset(p_le_ih) & (sb->s_blocksize - 1)) ==
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001684 1) {
1685 // FIXME: this is to keep 3.5 happy
Jeff Mahoney995c7622009-03-30 14:02:47 -04001686 REISERFS_I(inode)->i_first_direct_byte = U32_MAX;
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001687 quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001688 } else {
1689 quota_cut_bytes = 0;
1690 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoneyee939612009-03-30 14:02:50 -04001693 if (is_inode_locked) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001694 struct item_head *le_ih =
1695 PATH_PITEM_HEAD(s_cut_balance.tb_path);
1696 /* we are going to complete indirect2direct conversion. Make
1697 sure, that we exactly remove last unformatted node pointer
1698 of the item */
1699 if (!is_indirect_le_ih(le_ih))
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001700 reiserfs_panic(sb, "vs-5652",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001701 "item must be indirect %h", le_ih);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
Jeff Mahoneyee939612009-03-30 14:02:50 -04001703 if (mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001704 reiserfs_panic(sb, "vs-5653", "completing "
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001705 "indirect2direct conversion indirect "
1706 "item %h being deleted must be of "
1707 "4 byte long", le_ih);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
Jeff Mahoneyee939612009-03-30 14:02:50 -04001709 if (mode == M_CUT
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001710 && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
Jeff Mahoneya9dd3642009-03-30 14:02:45 -04001711 reiserfs_panic(sb, "vs-5654", "can not complete "
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001712 "indirect2direct conversion of %h "
1713 "(CUT, insert_size==%d)",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001714 le_ih, s_cut_balance.insert_size[0]);
1715 }
1716 /* it would be useful to make sure, that right neighboring
1717 item is direct item of this file */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001720
Jeff Mahoneyee939612009-03-30 14:02:50 -04001721 do_balance(&s_cut_balance, NULL, NULL, mode);
1722 if (is_inode_locked) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001723 /* we've done an indirect->direct conversion. when the data block
1724 ** was freed, it was removed from the list of blocks that must
1725 ** be flushed before the transaction commits, make sure to
1726 ** unmap and invalidate it
1727 */
1728 unmap_buffers(page, tail_pos);
Jeff Mahoney995c7622009-03-30 14:02:47 -04001729 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001730 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731#ifdef REISERQUOTA_DEBUG
Jeff Mahoney995c7622009-03-30 14:02:47 -04001732 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001733 "reiserquota cut_from_item(): freeing %u id=%u type=%c",
Jeff Mahoney995c7622009-03-30 14:02:47 -04001734 quota_cut_bytes, inode->i_uid, '?');
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735#endif
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001736 dquot_free_space_nodirty(inode, quota_cut_bytes);
Jeff Mahoneyee939612009-03-30 14:02:50 -04001737 return ret_value;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738}
1739
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001740static void truncate_directory(struct reiserfs_transaction_handle *th,
1741 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001743 BUG_ON(!th->t_trans_id);
1744 if (inode->i_nlink)
Jeff Mahoney0030b642009-03-30 14:02:28 -04001745 reiserfs_error(inode->i_sb, "vs-5655", "link count != 0");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001747 set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1748 set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1749 reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1750 reiserfs_update_sd(th, inode);
1751 set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1752 set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753}
1754
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755/* Truncate file to the new size. Note, this must be called with a transaction
1756 already started */
Jeff Mahoney995c7622009-03-30 14:02:47 -04001757int reiserfs_do_truncate(struct reiserfs_transaction_handle *th,
1758 struct inode *inode, /* ->i_size contains new size */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001759 struct page *page, /* up to date for last block */
1760 int update_timestamps /* when it is called by
1761 file_release to convert
1762 the tail - no timestamps
1763 should be updated */
1764 )
1765{
1766 INITIALIZE_PATH(s_search_path); /* Path to the current object item. */
1767 struct item_head *p_le_ih; /* Pointer to an item header. */
1768 struct cpu_key s_item_key; /* Key to search for a previous file item. */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001769 loff_t file_size, /* Old file size. */
1770 new_file_size; /* New file size. */
1771 int deleted; /* Number of deleted or truncated bytes. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001772 int retval;
1773 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001775 BUG_ON(!th->t_trans_id);
1776 if (!
Jeff Mahoney995c7622009-03-30 14:02:47 -04001777 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1778 || S_ISLNK(inode->i_mode)))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001779 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
Jeff Mahoney995c7622009-03-30 14:02:47 -04001781 if (S_ISDIR(inode->i_mode)) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001782 // deletion of directory - no need to update timestamps
Jeff Mahoney995c7622009-03-30 14:02:47 -04001783 truncate_directory(th, inode);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001784 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 }
1786
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001787 /* Get new file size. */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001788 new_file_size = inode->i_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001790 // FIXME: note, that key type is unimportant here
Jeff Mahoney995c7622009-03-30 14:02:47 -04001791 make_cpu_key(&s_item_key, inode, max_reiserfs_offset(inode),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001792 TYPE_DIRECT, 3);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001794 retval =
Jeff Mahoney995c7622009-03-30 14:02:47 -04001795 search_for_position_by_key(inode->i_sb, &s_item_key,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001796 &s_search_path);
1797 if (retval == IO_ERROR) {
Jeff Mahoney995c7622009-03-30 14:02:47 -04001798 reiserfs_error(inode->i_sb, "vs-5657",
Jeff Mahoney0030b642009-03-30 14:02:28 -04001799 "i/o failure occurred trying to truncate %K",
1800 &s_item_key);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001801 err = -EIO;
1802 goto out;
1803 }
1804 if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
Jeff Mahoney995c7622009-03-30 14:02:47 -04001805 reiserfs_error(inode->i_sb, "PAP-5660",
Jeff Mahoney0030b642009-03-30 14:02:28 -04001806 "wrong result %d of search for %K", retval,
1807 &s_item_key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001809 err = -EIO;
1810 goto out;
1811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001813 s_search_path.pos_in_item--;
1814
1815 /* Get real file size (total length of all file items) */
1816 p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1817 if (is_statdata_le_ih(p_le_ih))
Jeff Mahoneyee939612009-03-30 14:02:50 -04001818 file_size = 0;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001819 else {
1820 loff_t offset = le_ih_k_offset(p_le_ih);
1821 int bytes =
Jeff Mahoney995c7622009-03-30 14:02:47 -04001822 op_bytes_number(p_le_ih, inode->i_sb->s_blocksize);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001823
1824 /* this may mismatch with real file size: if last direct item
1825 had no padding zeros and last unformatted node had no free
1826 space, this file would have this file size */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001827 file_size = offset + bytes - 1;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 /*
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001830 * are we doing a full truncate or delete, if so
1831 * kick in the reada code
1832 */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001833 if (new_file_size == 0)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001834 s_search_path.reada = PATH_READA | PATH_READA_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
Jeff Mahoneyee939612009-03-30 14:02:50 -04001836 if (file_size == 0 || file_size < new_file_size) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001837 goto update_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001840 /* Update key to search for the last file item. */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001841 set_cpu_key_k_offset(&s_item_key, file_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001843 do {
1844 /* Cut or delete file item. */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001845 deleted =
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001846 reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001847 inode, page, new_file_size);
1848 if (deleted < 0) {
Jeff Mahoney995c7622009-03-30 14:02:47 -04001849 reiserfs_warning(inode->i_sb, "vs-5665",
Jeff Mahoney45b03d52009-03-30 14:02:21 -04001850 "reiserfs_cut_from_item failed");
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001851 reiserfs_check_path(&s_search_path);
1852 return 0;
1853 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
Jeff Mahoneyee939612009-03-30 14:02:50 -04001855 RFALSE(deleted > file_size,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001856 "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
Jeff Mahoneyee939612009-03-30 14:02:50 -04001857 deleted, file_size, &s_item_key);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001858
1859 /* Change key to search the last file item. */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001860 file_size -= deleted;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001861
Jeff Mahoneyee939612009-03-30 14:02:50 -04001862 set_cpu_key_k_offset(&s_item_key, file_size);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001863
1864 /* While there are bytes to truncate and previous file item is presented in the tree. */
1865
1866 /*
Jeff Mahoney0222e652009-03-30 14:02:44 -04001867 ** This loop could take a really long time, and could log
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001868 ** many more blocks than a transaction can hold. So, we do a polite
1869 ** journal end here, and if the transaction needs ending, we make
1870 ** sure the file is consistent before ending the current trans
1871 ** and starting a new one
1872 */
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001873 if (journal_transaction_should_end(th, 0) ||
1874 reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001875 int orig_len_alloc = th->t_blocks_allocated;
Jeff Mahoney3cd6dbe2009-03-30 14:02:43 -04001876 pathrelse(&s_search_path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001877
1878 if (update_timestamps) {
Jeff Mahoney995c7622009-03-30 14:02:47 -04001879 inode->i_mtime = CURRENT_TIME_SEC;
1880 inode->i_ctime = CURRENT_TIME_SEC;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001881 }
Jeff Mahoney995c7622009-03-30 14:02:47 -04001882 reiserfs_update_sd(th, inode);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001883
Jeff Mahoney995c7622009-03-30 14:02:47 -04001884 err = journal_end(th, inode->i_sb, orig_len_alloc);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001885 if (err)
1886 goto out;
Jeff Mahoney995c7622009-03-30 14:02:47 -04001887 err = journal_begin(th, inode->i_sb,
Alexander Zarochentzev23f9e0f2006-03-25 03:06:57 -08001888 JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001889 if (err)
1890 goto out;
Jeff Mahoney995c7622009-03-30 14:02:47 -04001891 reiserfs_update_inode_transaction(inode);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001892 }
Jeff Mahoneyee939612009-03-30 14:02:50 -04001893 } while (file_size > ROUND_UP(new_file_size) &&
Jeff Mahoney995c7622009-03-30 14:02:47 -04001894 search_for_position_by_key(inode->i_sb, &s_item_key,
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001895 &s_search_path) == POSITION_FOUND);
1896
Jeff Mahoneyee939612009-03-30 14:02:50 -04001897 RFALSE(file_size > ROUND_UP(new_file_size),
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001898 "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
Jeff Mahoneyee939612009-03-30 14:02:50 -04001899 new_file_size, file_size, s_item_key.on_disk_key.k_objectid);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001900
1901 update_and_out:
1902 if (update_timestamps) {
1903 // this is truncate, not file closing
Jeff Mahoney995c7622009-03-30 14:02:47 -04001904 inode->i_mtime = CURRENT_TIME_SEC;
1905 inode->i_ctime = CURRENT_TIME_SEC;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001906 }
Jeff Mahoney995c7622009-03-30 14:02:47 -04001907 reiserfs_update_sd(th, inode);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001908
1909 out:
1910 pathrelse(&s_search_path);
1911 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912}
1913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914#ifdef CONFIG_REISERFS_CHECK
1915// this makes sure, that we __append__, not overwrite or add holes
Josef "Jeff" Sipekfec6d052006-12-08 02:36:32 -08001916static void check_research_for_paste(struct treepath *path,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001917 const struct cpu_key *key)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001919 struct item_head *found_ih = get_ih(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001921 if (is_direct_le_ih(found_ih)) {
1922 if (le_ih_k_offset(found_ih) +
1923 op_bytes_number(found_ih,
1924 get_last_bh(path)->b_size) !=
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001925 cpu_key_k_offset(key)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001926 || op_bytes_number(found_ih,
1927 get_last_bh(path)->b_size) !=
1928 pos_in_item(path))
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001929 reiserfs_panic(NULL, "PAP-5720", "found direct item "
1930 "%h or position (%d) does not match "
1931 "to key %K", found_ih,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001932 pos_in_item(path), key);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001933 }
1934 if (is_indirect_le_ih(found_ih)) {
1935 if (le_ih_k_offset(found_ih) +
1936 op_bytes_number(found_ih,
1937 get_last_bh(path)->b_size) !=
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001938 cpu_key_k_offset(key)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001939 || I_UNFM_NUM(found_ih) != pos_in_item(path)
1940 || get_ih_free_space(found_ih) != 0)
Jeff Mahoneyc3a9c212009-03-30 14:02:25 -04001941 reiserfs_panic(NULL, "PAP-5730", "found indirect "
1942 "item (%h) or position (%d) does not "
1943 "match to key (%K)",
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001944 found_ih, pos_in_item(path), key);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001945 }
1946}
1947#endif /* config reiserfs check */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948
1949/* Paste bytes to the existing item. Returns bytes number pasted into the item. */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001950int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct treepath *search_path, /* Path to the pasted item. */
1951 const struct cpu_key *key, /* Key to search for the needed item. */
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001952 struct inode *inode, /* Inode item belongs to */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001953 const char *body, /* Pointer to the bytes to paste. */
Jeff Mahoneyee939612009-03-30 14:02:50 -04001954 int pasted_size)
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001955{ /* Size of pasted bytes. */
1956 struct tree_balance s_paste_balance;
1957 int retval;
1958 int fs_gen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001960 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001962 fs_gen = get_generation(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963
1964#ifdef REISERQUOTA_DEBUG
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001965 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1966 "reiserquota paste_into_item(): allocating %u id=%u type=%c",
Jeff Mahoneyee939612009-03-30 14:02:50 -04001967 pasted_size, inode->i_uid,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001968 key2type(&(key->on_disk_key)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969#endif
1970
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001971 retval = dquot_alloc_space_nodirty(inode, pasted_size);
1972 if (retval) {
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001973 pathrelse(search_path);
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001974 return retval;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001975 }
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001976 init_tb_struct(th, &s_paste_balance, th->t_super, search_path,
Jeff Mahoneyee939612009-03-30 14:02:50 -04001977 pasted_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978#ifdef DISPLACE_NEW_PACKING_LOCALITIES
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001979 s_paste_balance.key = key->on_disk_key;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980#endif
1981
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001982 /* DQUOT_* can schedule, must check before the fix_nodes */
1983 if (fs_changed(fs_gen, inode->i_sb)) {
1984 goto search_again;
1985 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001987 while ((retval =
1988 fix_nodes(M_PASTE, &s_paste_balance, NULL,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001989 body)) == REPEAT_SEARCH) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001990 search_again:
1991 /* file system changed while we were in the fix_nodes */
1992 PROC_INFO_INC(th->t_super, paste_into_item_restarted);
1993 retval =
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04001994 search_for_position_by_key(th->t_super, key,
1995 search_path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07001996 if (retval == IO_ERROR) {
1997 retval = -EIO;
1998 goto error_out;
1999 }
2000 if (retval == POSITION_FOUND) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002001 reiserfs_warning(inode->i_sb, "PAP-5710",
2002 "entry or pasted byte (%K) exists",
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002003 key);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002004 retval = -EEXIST;
2005 goto error_out;
2006 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007#ifdef CONFIG_REISERFS_CHECK
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002008 check_research_for_paste(search_path, key);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002010 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002012 /* Perform balancing after all resources are collected by fix_nodes, and
2013 accessing them will not risk triggering schedule. */
2014 if (retval == CARRY_ON) {
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002015 do_balance(&s_paste_balance, NULL /*ih */ , body, M_PASTE);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002016 return 0;
2017 }
2018 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2019 error_out:
2020 /* this also releases the path */
2021 unfix_nodes(&s_paste_balance);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022#ifdef REISERQUOTA_DEBUG
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002023 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2024 "reiserquota paste_into_item(): freeing %u id=%u type=%c",
Jeff Mahoneyee939612009-03-30 14:02:50 -04002025 pasted_size, inode->i_uid,
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002026 key2type(&(key->on_disk_key)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027#endif
Christoph Hellwig5dd40562010-03-03 09:05:00 -05002028 dquot_free_space_nodirty(inode, pasted_size);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002029 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030}
2031
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002032/* Insert new item into the buffer at the path.
2033 * th - active transaction handle
2034 * path - path to the inserted item
2035 * ih - pointer to the item header to insert
2036 * body - pointer to the bytes to insert
2037 */
2038int reiserfs_insert_item(struct reiserfs_transaction_handle *th,
2039 struct treepath *path, const struct cpu_key *key,
2040 struct item_head *ih, struct inode *inode,
2041 const char *body)
2042{
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002043 struct tree_balance s_ins_balance;
2044 int retval;
2045 int fs_gen = 0;
2046 int quota_bytes = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002048 BUG_ON(!th->t_trans_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002050 if (inode) { /* Do we count quotas for item? */
2051 fs_gen = get_generation(inode->i_sb);
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002052 quota_bytes = ih_item_len(ih);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002054 /* hack so the quota code doesn't have to guess if the file has
2055 ** a tail, links are always tails, so there's no guessing needed
2056 */
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002057 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(ih))
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002058 quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059#ifdef REISERQUOTA_DEBUG
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002060 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2061 "reiserquota insert_item(): allocating %u id=%u type=%c",
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002062 quota_bytes, inode->i_uid, head2type(ih));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002064 /* We can't dirty inode here. It would be immediately written but
2065 * appropriate stat item isn't inserted yet... */
Christoph Hellwig5dd40562010-03-03 09:05:00 -05002066 retval = dquot_alloc_space_nodirty(inode, quota_bytes);
2067 if (retval) {
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002068 pathrelse(path);
Christoph Hellwig5dd40562010-03-03 09:05:00 -05002069 return retval;
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002070 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 }
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002072 init_tb_struct(th, &s_ins_balance, th->t_super, path,
2073 IH_SIZE + ih_item_len(ih));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074#ifdef DISPLACE_NEW_PACKING_LOCALITIES
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002075 s_ins_balance.key = key->on_disk_key;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002077 /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2078 if (inode && fs_changed(fs_gen, inode->i_sb)) {
2079 goto search_again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 }
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002081
2082 while ((retval =
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002083 fix_nodes(M_INSERT, &s_ins_balance, ih,
2084 body)) == REPEAT_SEARCH) {
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002085 search_again:
2086 /* file system changed while we were in the fix_nodes */
2087 PROC_INFO_INC(th->t_super, insert_item_restarted);
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002088 retval = search_item(th->t_super, key, path);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002089 if (retval == IO_ERROR) {
2090 retval = -EIO;
2091 goto error_out;
2092 }
2093 if (retval == ITEM_FOUND) {
Jeff Mahoney45b03d52009-03-30 14:02:21 -04002094 reiserfs_warning(th->t_super, "PAP-5760",
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002095 "key %K already exists in the tree",
2096 key);
2097 retval = -EEXIST;
2098 goto error_out;
2099 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002102 /* make balancing after all resources will be collected at a time */
2103 if (retval == CARRY_ON) {
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002104 do_balance(&s_ins_balance, ih, body, M_INSERT);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002105 return 0;
2106 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002108 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2109 error_out:
2110 /* also releases the path */
2111 unfix_nodes(&s_ins_balance);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112#ifdef REISERQUOTA_DEBUG
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002113 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2114 "reiserquota insert_item(): freeing %u id=%u type=%c",
Jeff Mahoneyd68caa92009-03-30 14:02:49 -04002115 quota_bytes, inode->i_uid, head2type(ih));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116#endif
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002117 if (inode)
Christoph Hellwig5dd40562010-03-03 09:05:00 -05002118 dquot_free_space_nodirty(inode, quota_bytes);
Linus Torvaldsbd4c6252005-07-12 20:21:28 -07002119 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120}