Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README | |
3 | */ | |
4 | ||
098297b2 JM |
5 | /* |
6 | * Now we have all buffers that must be used in balancing of the tree | |
7 | * Further calculations can not cause schedule(), and thus the buffer | |
8 | * tree will be stable until the balancing will be finished | |
9 | * balance the tree according to the analysis made before, | |
10 | * and using buffers obtained after all above. | |
11 | */ | |
1da177e4 | 12 | |
1da177e4 LT |
13 | #include <asm/uaccess.h> |
14 | #include <linux/time.h> | |
f466c6fd | 15 | #include "reiserfs.h" |
1da177e4 | 16 | #include <linux/buffer_head.h> |
79a81aef | 17 | #include <linux/kernel.h> |
1da177e4 | 18 | |
fba4ebb5 JM |
19 | static inline void buffer_info_init_left(struct tree_balance *tb, |
20 | struct buffer_info *bi) | |
21 | { | |
22 | bi->tb = tb; | |
23 | bi->bi_bh = tb->L[0]; | |
24 | bi->bi_parent = tb->FL[0]; | |
25 | bi->bi_position = get_left_neighbor_position(tb, 0); | |
26 | } | |
27 | ||
28 | static inline void buffer_info_init_right(struct tree_balance *tb, | |
29 | struct buffer_info *bi) | |
30 | { | |
31 | bi->tb = tb; | |
32 | bi->bi_bh = tb->R[0]; | |
33 | bi->bi_parent = tb->FR[0]; | |
34 | bi->bi_position = get_right_neighbor_position(tb, 0); | |
35 | } | |
36 | ||
37 | static inline void buffer_info_init_tbS0(struct tree_balance *tb, | |
38 | struct buffer_info *bi) | |
39 | { | |
40 | bi->tb = tb; | |
41 | bi->bi_bh = PATH_PLAST_BUFFER(tb->tb_path); | |
42 | bi->bi_parent = PATH_H_PPARENT(tb->tb_path, 0); | |
43 | bi->bi_position = PATH_H_POSITION(tb->tb_path, 1); | |
44 | } | |
45 | ||
46 | static inline void buffer_info_init_bh(struct tree_balance *tb, | |
47 | struct buffer_info *bi, | |
48 | struct buffer_head *bh) | |
49 | { | |
50 | bi->tb = tb; | |
51 | bi->bi_bh = bh; | |
52 | bi->bi_parent = NULL; | |
53 | bi->bi_position = 0; | |
54 | } | |
55 | ||
bd4c625c LT |
56 | inline void do_balance_mark_leaf_dirty(struct tree_balance *tb, |
57 | struct buffer_head *bh, int flag) | |
1da177e4 | 58 | { |
09f1b80b | 59 | journal_mark_dirty(tb->transaction_handle, bh); |
1da177e4 LT |
60 | } |
61 | ||
62 | #define do_balance_mark_internal_dirty do_balance_mark_leaf_dirty | |
63 | #define do_balance_mark_sb_dirty do_balance_mark_leaf_dirty | |
64 | ||
098297b2 JM |
65 | /* |
66 | * summary: | |
67 | * if deleting something ( tb->insert_size[0] < 0 ) | |
68 | * return(balance_leaf_when_delete()); (flag d handled here) | |
69 | * else | |
70 | * if lnum is larger than 0 we put items into the left node | |
71 | * if rnum is larger than 0 we put items into the right node | |
72 | * if snum1 is larger than 0 we put items into the new node s1 | |
73 | * if snum2 is larger than 0 we put items into the new node s2 | |
74 | * Note that all *num* count new items being created. | |
75 | * | |
76 | * It would be easier to read balance_leaf() if each of these summary | |
77 | * lines was a separate procedure rather than being inlined. I think | |
78 | * that there are many passages here and in balance_leaf_when_delete() in | |
79 | * which two calls to one procedure can replace two passages, and it | |
80 | * might save cache space and improve software maintenance costs to do so. | |
81 | * | |
82 | * Vladimir made the perceptive comment that we should offload most of | |
83 | * the decision making in this function into fix_nodes/check_balance, and | |
84 | * then create some sort of structure in tb that says what actions should | |
85 | * be performed by do_balance. | |
86 | * | |
87 | * -Hans | |
88 | */ | |
89 | ||
90 | /* | |
91 | * Balance leaf node in case of delete or cut: insert_size[0] < 0 | |
1da177e4 LT |
92 | * |
93 | * lnum, rnum can have values >= -1 | |
94 | * -1 means that the neighbor must be joined with S | |
95 | * 0 means that nothing should be done with the neighbor | |
098297b2 JM |
96 | * >0 means to shift entirely or partly the specified number of items |
97 | * to the neighbor | |
1da177e4 | 98 | */ |
bd4c625c | 99 | static int balance_leaf_when_delete(struct tree_balance *tb, int flag) |
1da177e4 | 100 | { |
bd4c625c LT |
101 | struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path); |
102 | int item_pos = PATH_LAST_POSITION(tb->tb_path); | |
103 | int pos_in_item = tb->tb_path->pos_in_item; | |
104 | struct buffer_info bi; | |
105 | int n; | |
106 | struct item_head *ih; | |
1da177e4 | 107 | |
bd4c625c LT |
108 | RFALSE(tb->FR[0] && B_LEVEL(tb->FR[0]) != DISK_LEAF_NODE_LEVEL + 1, |
109 | "vs- 12000: level: wrong FR %z", tb->FR[0]); | |
110 | RFALSE(tb->blknum[0] > 1, | |
111 | "PAP-12005: tb->blknum == %d, can not be > 1", tb->blknum[0]); | |
112 | RFALSE(!tb->blknum[0] && !PATH_H_PPARENT(tb->tb_path, 0), | |
113 | "PAP-12010: tree can not be empty"); | |
1da177e4 | 114 | |
4cf5f7ad | 115 | ih = item_head(tbS0, item_pos); |
fba4ebb5 | 116 | buffer_info_init_tbS0(tb, &bi); |
1da177e4 | 117 | |
bd4c625c | 118 | /* Delete or truncate the item */ |
1da177e4 | 119 | |
bd4c625c LT |
120 | switch (flag) { |
121 | case M_DELETE: /* delete item in S[0] */ | |
122 | ||
123 | RFALSE(ih_item_len(ih) + IH_SIZE != -tb->insert_size[0], | |
124 | "vs-12013: mode Delete, insert size %d, ih to be deleted %h", | |
125 | -tb->insert_size[0], ih); | |
126 | ||
bd4c625c LT |
127 | leaf_delete_items(&bi, 0, item_pos, 1, -1); |
128 | ||
129 | if (!item_pos && tb->CFL[0]) { | |
130 | if (B_NR_ITEMS(tbS0)) { | |
131 | replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, | |
132 | 0); | |
133 | } else { | |
134 | if (!PATH_H_POSITION(tb->tb_path, 1)) | |
135 | replace_key(tb, tb->CFL[0], tb->lkey[0], | |
136 | PATH_H_PPARENT(tb->tb_path, | |
137 | 0), 0); | |
138 | } | |
139 | } | |
1da177e4 | 140 | |
bd4c625c LT |
141 | RFALSE(!item_pos && !tb->CFL[0], |
142 | "PAP-12020: tb->CFL[0]==%p, tb->L[0]==%p", tb->CFL[0], | |
143 | tb->L[0]); | |
1da177e4 | 144 | |
bd4c625c LT |
145 | break; |
146 | ||
147 | case M_CUT:{ /* cut item in S[0] */ | |
bd4c625c LT |
148 | if (is_direntry_le_ih(ih)) { |
149 | ||
098297b2 JM |
150 | /* |
151 | * UFS unlink semantics are such that you | |
152 | * can only delete one directory entry at | |
153 | * a time. | |
154 | */ | |
155 | ||
156 | /* | |
157 | * when we cut a directory tb->insert_size[0] | |
158 | * means number of entries to be cut (always 1) | |
159 | */ | |
bd4c625c LT |
160 | tb->insert_size[0] = -1; |
161 | leaf_cut_from_buffer(&bi, item_pos, pos_in_item, | |
162 | -tb->insert_size[0]); | |
163 | ||
164 | RFALSE(!item_pos && !pos_in_item && !tb->CFL[0], | |
165 | "PAP-12030: can not change delimiting key. CFL[0]=%p", | |
166 | tb->CFL[0]); | |
167 | ||
168 | if (!item_pos && !pos_in_item && tb->CFL[0]) { | |
169 | replace_key(tb, tb->CFL[0], tb->lkey[0], | |
170 | tbS0, 0); | |
171 | } | |
172 | } else { | |
173 | leaf_cut_from_buffer(&bi, item_pos, pos_in_item, | |
174 | -tb->insert_size[0]); | |
175 | ||
176 | RFALSE(!ih_item_len(ih), | |
177 | "PAP-12035: cut must leave non-zero dynamic length of item"); | |
178 | } | |
179 | break; | |
1da177e4 | 180 | } |
1da177e4 | 181 | |
bd4c625c LT |
182 | default: |
183 | print_cur_tb("12040"); | |
c3a9c210 JM |
184 | reiserfs_panic(tb->tb_sb, "PAP-12040", |
185 | "unexpected mode: %s(%d)", | |
bd4c625c LT |
186 | (flag == |
187 | M_PASTE) ? "PASTE" : ((flag == | |
188 | M_INSERT) ? "INSERT" : | |
189 | "UNKNOWN"), flag); | |
190 | } | |
1da177e4 | 191 | |
098297b2 JM |
192 | /* |
193 | * the rule is that no shifting occurs unless by shifting | |
194 | * a node can be freed | |
195 | */ | |
bd4c625c | 196 | n = B_NR_ITEMS(tbS0); |
098297b2 JM |
197 | /* L[0] takes part in balancing */ |
198 | if (tb->lnum[0]) { | |
199 | /* L[0] must be joined with S[0] */ | |
200 | if (tb->lnum[0] == -1) { | |
201 | /* R[0] must be also joined with S[0] */ | |
202 | if (tb->rnum[0] == -1) { | |
bd4c625c | 203 | if (tb->FR[0] == PATH_H_PPARENT(tb->tb_path, 0)) { |
098297b2 JM |
204 | /* |
205 | * all contents of all the 3 buffers | |
206 | * will be in L[0] | |
207 | */ | |
bd4c625c LT |
208 | if (PATH_H_POSITION(tb->tb_path, 1) == 0 |
209 | && 1 < B_NR_ITEMS(tb->FR[0])) | |
210 | replace_key(tb, tb->CFL[0], | |
211 | tb->lkey[0], | |
212 | tb->FR[0], 1); | |
213 | ||
214 | leaf_move_items(LEAF_FROM_S_TO_L, tb, n, | |
215 | -1, NULL); | |
216 | leaf_move_items(LEAF_FROM_R_TO_L, tb, | |
217 | B_NR_ITEMS(tb->R[0]), | |
218 | -1, NULL); | |
219 | ||
220 | reiserfs_invalidate_buffer(tb, tbS0); | |
221 | reiserfs_invalidate_buffer(tb, | |
222 | tb->R[0]); | |
223 | ||
224 | return 0; | |
225 | } | |
098297b2 JM |
226 | /* |
227 | * all contents of all the 3 buffers will | |
228 | * be in R[0] | |
229 | */ | |
bd4c625c LT |
230 | leaf_move_items(LEAF_FROM_S_TO_R, tb, n, -1, |
231 | NULL); | |
232 | leaf_move_items(LEAF_FROM_L_TO_R, tb, | |
233 | B_NR_ITEMS(tb->L[0]), -1, NULL); | |
234 | ||
235 | /* right_delimiting_key is correct in R[0] */ | |
236 | replace_key(tb, tb->CFR[0], tb->rkey[0], | |
237 | tb->R[0], 0); | |
1da177e4 | 238 | |
bd4c625c LT |
239 | reiserfs_invalidate_buffer(tb, tbS0); |
240 | reiserfs_invalidate_buffer(tb, tb->L[0]); | |
1da177e4 | 241 | |
bd4c625c LT |
242 | return -1; |
243 | } | |
1da177e4 | 244 | |
bd4c625c LT |
245 | RFALSE(tb->rnum[0] != 0, |
246 | "PAP-12045: rnum must be 0 (%d)", tb->rnum[0]); | |
247 | /* all contents of L[0] and S[0] will be in L[0] */ | |
248 | leaf_shift_left(tb, n, -1); | |
1da177e4 | 249 | |
bd4c625c LT |
250 | reiserfs_invalidate_buffer(tb, tbS0); |
251 | ||
252 | return 0; | |
253 | } | |
098297b2 JM |
254 | |
255 | /* | |
256 | * a part of contents of S[0] will be in L[0] and the | |
257 | * rest part of S[0] will be in R[0] | |
258 | */ | |
bd4c625c LT |
259 | |
260 | RFALSE((tb->lnum[0] + tb->rnum[0] < n) || | |
261 | (tb->lnum[0] + tb->rnum[0] > n + 1), | |
262 | "PAP-12050: rnum(%d) and lnum(%d) and item number(%d) in S[0] are not consistent", | |
263 | tb->rnum[0], tb->lnum[0], n); | |
264 | RFALSE((tb->lnum[0] + tb->rnum[0] == n) && | |
265 | (tb->lbytes != -1 || tb->rbytes != -1), | |
266 | "PAP-12055: bad rbytes (%d)/lbytes (%d) parameters when items are not split", | |
267 | tb->rbytes, tb->lbytes); | |
268 | RFALSE((tb->lnum[0] + tb->rnum[0] == n + 1) && | |
269 | (tb->lbytes < 1 || tb->rbytes != -1), | |
270 | "PAP-12060: bad rbytes (%d)/lbytes (%d) parameters when items are split", | |
271 | tb->rbytes, tb->lbytes); | |
272 | ||
273 | leaf_shift_left(tb, tb->lnum[0], tb->lbytes); | |
274 | leaf_shift_right(tb, tb->rnum[0], tb->rbytes); | |
275 | ||
276 | reiserfs_invalidate_buffer(tb, tbS0); | |
277 | ||
278 | return 0; | |
1da177e4 | 279 | } |
1da177e4 | 280 | |
bd4c625c LT |
281 | if (tb->rnum[0] == -1) { |
282 | /* all contents of R[0] and S[0] will be in R[0] */ | |
283 | leaf_shift_right(tb, n, -1); | |
284 | reiserfs_invalidate_buffer(tb, tbS0); | |
285 | return 0; | |
286 | } | |
1da177e4 | 287 | |
bd4c625c LT |
288 | RFALSE(tb->rnum[0], |
289 | "PAP-12065: bad rnum parameter must be 0 (%d)", tb->rnum[0]); | |
1da177e4 | 290 | return 0; |
1da177e4 LT |
291 | } |
292 | ||
f1f007c3 JM |
293 | static void balance_leaf_insert_left(struct tree_balance *tb, |
294 | struct item_head *ih, const char *body) | |
295 | { | |
296 | int ret_val; | |
297 | struct buffer_info bi; | |
298 | int n = B_NR_ITEMS(tb->L[0]); | |
299 | ||
300 | if (tb->item_pos == tb->lnum[0] - 1 && tb->lbytes != -1) { | |
301 | /* part of new item falls into L[0] */ | |
302 | int new_item_len; | |
303 | int version; | |
304 | ||
305 | ret_val = leaf_shift_left(tb, tb->lnum[0] - 1, -1); | |
306 | ||
307 | /* Calculate item length to insert to S[0] */ | |
308 | new_item_len = ih_item_len(ih) - tb->lbytes; | |
309 | /* Calculate and check item length to insert to L[0] */ | |
310 | put_ih_item_len(ih, ih_item_len(ih) - new_item_len); | |
311 | ||
312 | RFALSE(ih_item_len(ih) <= 0, | |
313 | "PAP-12080: there is nothing to insert into L[0]: ih_item_len=%d", | |
314 | ih_item_len(ih)); | |
315 | ||
316 | /* Insert new item into L[0] */ | |
317 | buffer_info_init_left(tb, &bi); | |
318 | leaf_insert_into_buf(&bi, | |
319 | n + tb->item_pos - ret_val, ih, body, | |
320 | tb->zeroes_num > ih_item_len(ih) ? ih_item_len(ih) : tb->zeroes_num); | |
321 | ||
322 | version = ih_version(ih); | |
323 | ||
324 | /* Calculate key component, item length and body to insert into S[0] */ | |
325 | set_le_ih_k_offset(ih, le_ih_k_offset(ih) + | |
326 | (tb->lbytes << (is_indirect_le_ih(ih) ? tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT : 0))); | |
327 | ||
328 | put_ih_item_len(ih, new_item_len); | |
329 | if (tb->lbytes > tb->zeroes_num) { | |
330 | body += (tb->lbytes - tb->zeroes_num); | |
331 | tb->zeroes_num = 0; | |
332 | } else | |
333 | tb->zeroes_num -= tb->lbytes; | |
334 | ||
335 | RFALSE(ih_item_len(ih) <= 0, | |
336 | "PAP-12085: there is nothing to insert into S[0]: ih_item_len=%d", | |
337 | ih_item_len(ih)); | |
338 | } else { | |
339 | /* new item in whole falls into L[0] */ | |
340 | /* Shift lnum[0]-1 items to L[0] */ | |
341 | ret_val = leaf_shift_left(tb, tb->lnum[0] - 1, tb->lbytes); | |
342 | /* Insert new item into L[0] */ | |
343 | buffer_info_init_left(tb, &bi); | |
344 | leaf_insert_into_buf(&bi, n + tb->item_pos - ret_val, ih, body, tb->zeroes_num); | |
345 | tb->insert_size[0] = 0; | |
346 | tb->zeroes_num = 0; | |
347 | } | |
348 | ||
349 | } | |
350 | ||
cf22df18 JM |
351 | static void balance_leaf_paste_left(struct tree_balance *tb, |
352 | struct item_head *ih, const char *body) | |
1da177e4 | 353 | { |
bd4c625c | 354 | struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path); |
bd4c625c | 355 | int ret_val; |
cf22df18 JM |
356 | struct buffer_info bi; |
357 | int n = B_NR_ITEMS(tb->L[0]); | |
bd4c625c | 358 | |
b49fb112 | 359 | if (tb->item_pos == tb->lnum[0] - 1 && tb->lbytes != -1) { |
bd4c625c | 360 | /* we must shift the part of the appended item */ |
b49fb112 | 361 | if (is_direntry_le_ih(item_head(tbS0, tb->item_pos))) { |
bd4c625c | 362 | |
b49fb112 | 363 | RFALSE(tb->zeroes_num, |
bd4c625c LT |
364 | "PAP-12090: invalid parameter in case of a directory"); |
365 | /* directory item */ | |
b49fb112 | 366 | if (tb->lbytes > tb->pos_in_item) { |
bd4c625c | 367 | /* new directory entry falls into L[0] */ |
416e2abd | 368 | struct item_head *pasted; |
b49fb112 | 369 | int l_pos_in_item = tb->pos_in_item; |
bd4c625c LT |
370 | |
371 | /* Shift lnum[0] - 1 items in whole. Shift lbytes - 1 entries from given directory item */ | |
416e2abd | 372 | ret_val = leaf_shift_left(tb, tb->lnum[0], tb->lbytes-1); |
b49fb112 | 373 | if (ret_val && !tb->item_pos) { |
4cf5f7ad JM |
374 | pasted = item_head(tb->L[0], B_NR_ITEMS(tb->L[0]) - 1); |
375 | l_pos_in_item += ih_entry_count(pasted) - (tb->lbytes -1); | |
bd4c625c LT |
376 | } |
377 | ||
378 | /* Append given directory entry to directory item */ | |
fba4ebb5 | 379 | buffer_info_init_left(tb, &bi); |
b49fb112 | 380 | leaf_paste_in_buffer(&bi, n + tb->item_pos - ret_val, l_pos_in_item, tb->insert_size[0], body, tb->zeroes_num); |
bd4c625c LT |
381 | |
382 | /* previous string prepared space for pasting new entry, following string pastes this entry */ | |
383 | ||
384 | /* when we have merge directory item, pos_in_item has been changed too */ | |
385 | ||
386 | /* paste new directory entry. 1 is entry number */ | |
b49fb112 | 387 | leaf_paste_entries(&bi, n + tb->item_pos - ret_val, l_pos_in_item, |
416e2abd DJ |
388 | 1, (struct reiserfs_de_head *) body, |
389 | body + DEH_SIZE, tb->insert_size[0]); | |
bd4c625c LT |
390 | tb->insert_size[0] = 0; |
391 | } else { | |
392 | /* new directory item doesn't fall into L[0] */ | |
393 | /* Shift lnum[0]-1 items in whole. Shift lbytes directory entries from directory item number lnum[0] */ | |
416e2abd | 394 | leaf_shift_left(tb, tb->lnum[0], tb->lbytes); |
bd4c625c LT |
395 | } |
396 | /* Calculate new position to append in item body */ | |
b49fb112 | 397 | tb->pos_in_item -= tb->lbytes; |
bd4c625c LT |
398 | } else { |
399 | /* regular object */ | |
416e2abd | 400 | RFALSE(tb->lbytes <= 0, "PAP-12095: there is nothing to shift to L[0]. lbytes=%d", tb->lbytes); |
b49fb112 | 401 | RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)), |
bd4c625c | 402 | "PAP-12100: incorrect position to paste: item_len=%d, pos_in_item=%d", |
b49fb112 | 403 | ih_item_len(item_head(tbS0, tb->item_pos)), tb->pos_in_item); |
bd4c625c | 404 | |
b49fb112 | 405 | if (tb->lbytes >= tb->pos_in_item) { |
bd4c625c LT |
406 | /* appended item will be in L[0] in whole */ |
407 | int l_n; | |
408 | ||
409 | /* this bytes number must be appended to the last item of L[h] */ | |
b49fb112 | 410 | l_n = tb->lbytes - tb->pos_in_item; |
bd4c625c LT |
411 | |
412 | /* Calculate new insert_size[0] */ | |
416e2abd | 413 | tb->insert_size[0] -= l_n; |
bd4c625c | 414 | |
416e2abd | 415 | RFALSE(tb->insert_size[0] <= 0, |
bd4c625c | 416 | "PAP-12105: there is nothing to paste into L[0]. insert_size=%d", |
416e2abd DJ |
417 | tb->insert_size[0]); |
418 | ret_val = leaf_shift_left(tb, tb->lnum[0], ih_item_len | |
b49fb112 | 419 | (item_head(tbS0, tb->item_pos))); |
bd4c625c | 420 | /* Append to body of item in L[0] */ |
fba4ebb5 | 421 | buffer_info_init_left(tb, &bi); |
bd4c625c | 422 | leaf_paste_in_buffer |
b49fb112 JM |
423 | (&bi, n + tb->item_pos - ret_val, ih_item_len |
424 | (item_head(tb->L[0], n + tb->item_pos - ret_val)), | |
416e2abd | 425 | l_n, body, |
b49fb112 | 426 | tb->zeroes_num > l_n ? l_n : tb->zeroes_num); |
bd4c625c LT |
427 | /* 0-th item in S0 can be only of DIRECT type when l_n != 0 */ |
428 | { | |
429 | int version; | |
416e2abd DJ |
430 | int temp_l = l_n; |
431 | ||
4cf5f7ad | 432 | RFALSE(ih_item_len(item_head(tbS0, 0)), |
bd4c625c | 433 | "PAP-12106: item length must be 0"); |
4cf5f7ad | 434 | RFALSE(comp_short_le_keys(leaf_key(tbS0, 0), leaf_key |
b49fb112 | 435 | (tb->L[0], n + tb->item_pos - ret_val)), |
bd4c625c | 436 | "PAP-12107: items must be of the same file"); |
b49fb112 | 437 | if (is_indirect_le_ih(item_head(tb->L[0], n + tb->item_pos - ret_val))) { |
416e2abd | 438 | temp_l = l_n << (tb->tb_sb-> s_blocksize_bits - UNFM_P_SHIFT); |
bd4c625c LT |
439 | } |
440 | /* update key of first item in S0 */ | |
4cf5f7ad JM |
441 | version = ih_version(item_head(tbS0, 0)); |
442 | set_le_key_k_offset(version, leaf_key(tbS0, 0), | |
443 | le_key_k_offset(version,leaf_key(tbS0, 0)) + temp_l); | |
bd4c625c | 444 | /* update left delimiting key */ |
4cf5f7ad JM |
445 | set_le_key_k_offset(version, internal_key(tb->CFL[0], tb->lkey[0]), |
446 | le_key_k_offset(version, internal_key(tb->CFL[0], tb->lkey[0])) + temp_l); | |
bd4c625c LT |
447 | } |
448 | ||
449 | /* Calculate new body, position in item and insert_size[0] */ | |
b49fb112 JM |
450 | if (l_n > tb->zeroes_num) { |
451 | body += (l_n - tb->zeroes_num); | |
452 | tb->zeroes_num = 0; | |
bd4c625c | 453 | } else |
b49fb112 JM |
454 | tb->zeroes_num -= l_n; |
455 | tb->pos_in_item = 0; | |
bd4c625c | 456 | |
4cf5f7ad JM |
457 | RFALSE(comp_short_le_keys(leaf_key(tbS0, 0), leaf_key(tb->L[0], B_NR_ITEMS(tb->L[0]) - 1)) |
458 | || !op_is_left_mergeable(leaf_key(tbS0, 0), tbS0->b_size) | |
459 | || !op_is_left_mergeable(internal_key(tb->CFL[0], tb->lkey[0]), tbS0->b_size), | |
bd4c625c LT |
460 | "PAP-12120: item must be merge-able with left neighboring item"); |
461 | } else { /* only part of the appended item will be in L[0] */ | |
462 | ||
463 | /* Calculate position in item for append in S[0] */ | |
b49fb112 | 464 | tb->pos_in_item -= tb->lbytes; |
bd4c625c | 465 | |
b49fb112 | 466 | RFALSE(tb->pos_in_item <= 0, "PAP-12125: no place for paste. pos_in_item=%d", tb->pos_in_item); |
bd4c625c LT |
467 | |
468 | /* Shift lnum[0] - 1 items in whole. Shift lbytes - 1 byte from item number lnum[0] */ | |
416e2abd | 469 | leaf_shift_left(tb, tb->lnum[0], tb->lbytes); |
bd4c625c LT |
470 | } |
471 | } | |
472 | } else { /* appended item will be in L[0] in whole */ | |
473 | ||
474 | struct item_head *pasted; | |
475 | ||
b49fb112 | 476 | if (!tb->item_pos && op_is_left_mergeable(leaf_key(tbS0, 0), tbS0->b_size)) { /* if we paste into first item of S[0] and it is left mergable */ |
bd4c625c | 477 | /* then increment pos_in_item by the size of the last item in L[0] */ |
4cf5f7ad | 478 | pasted = item_head(tb->L[0], n - 1); |
bd4c625c | 479 | if (is_direntry_le_ih(pasted)) |
b49fb112 | 480 | tb->pos_in_item += ih_entry_count(pasted); |
bd4c625c | 481 | else |
b49fb112 | 482 | tb->pos_in_item += ih_item_len(pasted); |
bd4c625c LT |
483 | } |
484 | ||
485 | /* Shift lnum[0] - 1 items in whole. Shift lbytes - 1 byte from item number lnum[0] */ | |
416e2abd | 486 | ret_val = leaf_shift_left(tb, tb->lnum[0], tb->lbytes); |
bd4c625c | 487 | /* Append to body of item in L[0] */ |
fba4ebb5 | 488 | buffer_info_init_left(tb, &bi); |
b49fb112 JM |
489 | leaf_paste_in_buffer(&bi, n + tb->item_pos - ret_val, |
490 | tb->pos_in_item, | |
bd4c625c | 491 | tb->insert_size[0], |
b49fb112 | 492 | body, tb->zeroes_num); |
bd4c625c LT |
493 | |
494 | /* if appended item is directory, paste entry */ | |
b49fb112 | 495 | pasted = item_head(tb->L[0], n + tb->item_pos - ret_val); |
bd4c625c | 496 | if (is_direntry_le_ih(pasted)) |
b49fb112 JM |
497 | leaf_paste_entries(&bi, n + tb->item_pos - ret_val, |
498 | tb->pos_in_item, 1, | |
416e2abd DJ |
499 | (struct reiserfs_de_head *) body, |
500 | body + DEH_SIZE, | |
501 | tb->insert_size[0]); | |
bd4c625c LT |
502 | /* if appended item is indirect item, put unformatted node into un list */ |
503 | if (is_indirect_le_ih(pasted)) | |
504 | set_ih_free_space(pasted, 0); | |
505 | tb->insert_size[0] = 0; | |
b49fb112 | 506 | tb->zeroes_num = 0; |
bd4c625c | 507 | } |
cf22df18 JM |
508 | |
509 | } | |
510 | ||
e80ef3d1 JM |
511 | static void balance_leaf_insert_right(struct tree_balance *tb, |
512 | struct item_head *ih, const char *body) | |
513 | { | |
514 | ||
515 | struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path); | |
516 | int n = B_NR_ITEMS(tbS0); | |
517 | struct buffer_info bi; | |
518 | int ret_val; | |
519 | if (n - tb->rnum[0] < tb->item_pos) { /* new item or its part falls to R[0] */ | |
520 | if (tb->item_pos == n - tb->rnum[0] + 1 && tb->rbytes != -1) { /* part of new item falls into R[0] */ | |
521 | loff_t old_key_comp, old_len, r_zeroes_number; | |
522 | const char *r_body; | |
523 | int version; | |
524 | loff_t offset; | |
525 | ||
526 | leaf_shift_right(tb, tb->rnum[0] - 1, -1); | |
527 | ||
528 | version = ih_version(ih); | |
529 | /* Remember key component and item length */ | |
530 | old_key_comp = le_ih_k_offset(ih); | |
531 | old_len = ih_item_len(ih); | |
532 | ||
533 | /* Calculate key component and item length to insert into R[0] */ | |
534 | offset = le_ih_k_offset(ih) + ((old_len - tb->rbytes) << (is_indirect_le_ih(ih) ? tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT : 0)); | |
535 | set_le_ih_k_offset(ih, offset); | |
536 | put_ih_item_len(ih, tb->rbytes); | |
537 | /* Insert part of the item into R[0] */ | |
538 | buffer_info_init_right(tb, &bi); | |
539 | if ((old_len - tb->rbytes) > tb->zeroes_num) { | |
540 | r_zeroes_number = 0; | |
541 | r_body = body + (old_len - tb->rbytes) - tb->zeroes_num; | |
542 | } else { | |
543 | r_body = body; | |
544 | r_zeroes_number = tb->zeroes_num - (old_len - tb->rbytes); | |
545 | tb->zeroes_num -= r_zeroes_number; | |
546 | } | |
547 | ||
548 | leaf_insert_into_buf(&bi, 0, ih, r_body, | |
549 | r_zeroes_number); | |
550 | ||
551 | /* Replace right delimiting key by first key in R[0] */ | |
552 | replace_key(tb, tb->CFR[0], tb->rkey[0], | |
553 | tb->R[0], 0); | |
554 | ||
555 | /* Calculate key component and item length to insert into S[0] */ | |
556 | set_le_ih_k_offset(ih, old_key_comp); | |
557 | put_ih_item_len(ih, old_len - tb->rbytes); | |
558 | ||
559 | tb->insert_size[0] -= tb->rbytes; | |
560 | ||
561 | } else { /* whole new item falls into R[0] */ | |
562 | ||
563 | /* Shift rnum[0]-1 items to R[0] */ | |
564 | ret_val = leaf_shift_right(tb, tb->rnum[0] - 1, tb->rbytes); | |
565 | /* Insert new item into R[0] */ | |
566 | buffer_info_init_right(tb, &bi); | |
567 | leaf_insert_into_buf(&bi, tb->item_pos - n + tb->rnum[0] - 1, | |
568 | ih, body, tb->zeroes_num); | |
569 | ||
570 | if (tb->item_pos - n + tb->rnum[0] - 1 == 0) { | |
571 | replace_key(tb, tb->CFR[0], | |
572 | tb->rkey[0], | |
573 | tb->R[0], 0); | |
574 | ||
575 | } | |
576 | tb->zeroes_num = tb->insert_size[0] = 0; | |
577 | } | |
578 | } else { /* new item or part of it doesn't fall into R[0] */ | |
579 | ||
580 | leaf_shift_right(tb, tb->rnum[0], tb->rbytes); | |
581 | } | |
582 | ||
583 | } | |
584 | ||
3f0eb276 JM |
585 | static void balance_leaf_paste_right(struct tree_balance *tb, |
586 | struct item_head *ih, const char *body) | |
cf22df18 JM |
587 | { |
588 | struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path); | |
3f0eb276 | 589 | int n = B_NR_ITEMS(tbS0); |
cf22df18 | 590 | struct buffer_info bi; |
cf22df18 JM |
591 | int ret_val; |
592 | ||
b49fb112 JM |
593 | if (n - tb->rnum[0] <= tb->item_pos) { /* pasted item or part of it falls to R[0] */ |
594 | if (tb->item_pos == n - tb->rnum[0] && tb->rbytes != -1) { /* we must shift the part of the appended item */ | |
595 | if (is_direntry_le_ih(item_head(tbS0, tb->item_pos))) { /* we append to directory item */ | |
bd4c625c LT |
596 | int entry_count; |
597 | ||
b49fb112 | 598 | RFALSE(tb->zeroes_num, |
bd4c625c | 599 | "PAP-12145: invalid parameter in case of a directory"); |
4cf5f7ad | 600 | entry_count = ih_entry_count(item_head |
b49fb112 | 601 | (tbS0, tb->item_pos)); |
bd4c625c | 602 | if (entry_count - tb->rbytes < |
b49fb112 | 603 | tb->pos_in_item) |
bd4c625c LT |
604 | /* new directory entry falls into R[0] */ |
605 | { | |
606 | int paste_entry_position; | |
607 | ||
416e2abd | 608 | RFALSE(tb->rbytes - 1 >= entry_count || !tb-> insert_size[0], |
bd4c625c | 609 | "PAP-12150: no enough of entries to shift to R[0]: rbytes=%d, entry_count=%d", |
416e2abd | 610 | tb->rbytes, entry_count); |
bd4c625c | 611 | /* Shift rnum[0]-1 items in whole. Shift rbytes-1 directory entries from directory item number rnum[0] */ |
416e2abd | 612 | leaf_shift_right(tb, tb->rnum[0], tb->rbytes - 1); |
bd4c625c | 613 | /* Paste given directory entry to directory item */ |
b49fb112 | 614 | paste_entry_position = tb->pos_in_item - entry_count + tb->rbytes - 1; |
fba4ebb5 | 615 | buffer_info_init_right(tb, &bi); |
b49fb112 | 616 | leaf_paste_in_buffer(&bi, 0, paste_entry_position, tb->insert_size[0], body, tb->zeroes_num); |
bd4c625c | 617 | /* paste entry */ |
416e2abd DJ |
618 | leaf_paste_entries(&bi, 0, paste_entry_position, 1, |
619 | (struct reiserfs_de_head *) body, | |
620 | body + DEH_SIZE, tb->insert_size[0]); | |
621 | ||
622 | if (paste_entry_position == 0) { | |
bd4c625c | 623 | /* change delimiting keys */ |
416e2abd | 624 | replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0],0); |
bd4c625c LT |
625 | } |
626 | ||
627 | tb->insert_size[0] = 0; | |
b49fb112 | 628 | tb->pos_in_item++; |
bd4c625c LT |
629 | } else { /* new directory entry doesn't fall into R[0] */ |
630 | ||
416e2abd | 631 | leaf_shift_right(tb, tb->rnum[0], tb->rbytes); |
bd4c625c LT |
632 | } |
633 | } else { /* regular object */ | |
634 | ||
b49fb112 | 635 | int n_shift, n_rem, r_zeroes_number; |
bd4c625c LT |
636 | const char *r_body; |
637 | ||
638 | /* Calculate number of bytes which must be shifted from appended item */ | |
416e2abd | 639 | if ((n_shift = tb->rbytes - tb->insert_size[0]) < 0) |
bd4c625c LT |
640 | n_shift = 0; |
641 | ||
b49fb112 JM |
642 | RFALSE(tb->pos_in_item != ih_item_len |
643 | (item_head(tbS0, tb->item_pos)), | |
bd4c625c | 644 | "PAP-12155: invalid position to paste. ih_item_len=%d, pos_in_item=%d", |
b49fb112 JM |
645 | tb->pos_in_item, ih_item_len |
646 | (item_head(tbS0, tb->item_pos))); | |
416e2abd DJ |
647 | |
648 | leaf_shift_right(tb, tb->rnum[0], n_shift); | |
bd4c625c | 649 | /* Calculate number of bytes which must remain in body after appending to R[0] */ |
416e2abd | 650 | if ((n_rem = tb->insert_size[0] - tb->rbytes) < 0) |
bd4c625c LT |
651 | n_rem = 0; |
652 | ||
653 | { | |
654 | int version; | |
416e2abd DJ |
655 | unsigned long temp_rem = n_rem; |
656 | ||
4cf5f7ad JM |
657 | version = ih_version(item_head(tb->R[0], 0)); |
658 | if (is_indirect_le_key(version, leaf_key(tb->R[0], 0))) { | |
416e2abd | 659 | temp_rem = n_rem << (tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT); |
bd4c625c | 660 | } |
4cf5f7ad JM |
661 | set_le_key_k_offset(version, leaf_key(tb->R[0], 0), |
662 | le_key_k_offset(version, leaf_key(tb->R[0], 0)) + temp_rem); | |
663 | set_le_key_k_offset(version, internal_key(tb->CFR[0], tb->rkey[0]), | |
664 | le_key_k_offset(version, internal_key(tb->CFR[0], tb->rkey[0])) + temp_rem); | |
bd4c625c | 665 | } |
4cf5f7ad JM |
666 | /* k_offset (leaf_key(tb->R[0],0)) += n_rem; |
667 | k_offset (internal_key(tb->CFR[0],tb->rkey[0])) += n_rem;*/ | |
416e2abd | 668 | do_balance_mark_internal_dirty(tb, tb->CFR[0], 0); |
bd4c625c LT |
669 | |
670 | /* Append part of body into R[0] */ | |
fba4ebb5 | 671 | buffer_info_init_right(tb, &bi); |
b49fb112 JM |
672 | if (n_rem > tb->zeroes_num) { |
673 | r_zeroes_number = 0; | |
674 | r_body = body + n_rem - tb->zeroes_num; | |
bd4c625c LT |
675 | } else { |
676 | r_body = body; | |
b49fb112 JM |
677 | r_zeroes_number = tb->zeroes_num - n_rem; |
678 | tb->zeroes_num -= r_zeroes_number; | |
bd4c625c LT |
679 | } |
680 | ||
416e2abd DJ |
681 | leaf_paste_in_buffer(&bi, 0, n_shift, |
682 | tb->insert_size[0] - n_rem, | |
b49fb112 | 683 | r_body, r_zeroes_number); |
416e2abd | 684 | |
4cf5f7ad | 685 | if (is_indirect_le_ih(item_head(tb->R[0], 0))) { |
1da177e4 | 686 | #if 0 |
bd4c625c LT |
687 | RFALSE(n_rem, |
688 | "PAP-12160: paste more than one unformatted node pointer"); | |
1da177e4 | 689 | #endif |
4cf5f7ad | 690 | set_ih_free_space(item_head(tb->R[0], 0), 0); |
bd4c625c LT |
691 | } |
692 | tb->insert_size[0] = n_rem; | |
693 | if (!n_rem) | |
b49fb112 | 694 | tb->pos_in_item++; |
bd4c625c LT |
695 | } |
696 | } else { /* pasted item in whole falls into R[0] */ | |
697 | ||
698 | struct item_head *pasted; | |
699 | ||
416e2abd | 700 | ret_val = leaf_shift_right(tb, tb->rnum[0], tb->rbytes); |
bd4c625c | 701 | /* append item in R[0] */ |
b49fb112 | 702 | if (tb->pos_in_item >= 0) { |
fba4ebb5 | 703 | buffer_info_init_right(tb, &bi); |
b49fb112 JM |
704 | leaf_paste_in_buffer(&bi, tb->item_pos - n + tb->rnum[0], tb->pos_in_item, |
705 | tb->insert_size[0], body, tb->zeroes_num); | |
bd4c625c LT |
706 | } |
707 | ||
708 | /* paste new entry, if item is directory item */ | |
b49fb112 JM |
709 | pasted = item_head(tb->R[0], tb->item_pos - n + tb->rnum[0]); |
710 | if (is_direntry_le_ih(pasted) && tb->pos_in_item >= 0) { | |
711 | leaf_paste_entries(&bi, tb->item_pos - n + tb->rnum[0], | |
712 | tb->pos_in_item, 1, | |
416e2abd DJ |
713 | (struct reiserfs_de_head *) body, |
714 | body + DEH_SIZE, tb->insert_size[0]); | |
b49fb112 | 715 | if (!tb->pos_in_item) { |
bd4c625c | 716 | |
b49fb112 | 717 | RFALSE(tb->item_pos - n + tb->rnum[0], |
bd4c625c LT |
718 | "PAP-12165: directory item must be first item of node when pasting is in 0th position"); |
719 | ||
720 | /* update delimiting keys */ | |
416e2abd | 721 | replace_key(tb, tb->CFR[0], tb->rkey[0], tb->R[0], 0); |
bd4c625c LT |
722 | } |
723 | } | |
724 | ||
725 | if (is_indirect_le_ih(pasted)) | |
726 | set_ih_free_space(pasted, 0); | |
b49fb112 | 727 | tb->zeroes_num = tb->insert_size[0] = 0; |
bd4c625c LT |
728 | } |
729 | } else { /* new item doesn't fall into R[0] */ | |
1da177e4 | 730 | |
bd4c625c | 731 | leaf_shift_right(tb, tb->rnum[0], tb->rbytes); |
1da177e4 | 732 | } |
3f0eb276 JM |
733 | |
734 | } | |
735 | ||
65ab18cb JM |
736 | static void balance_leaf_new_nodes_insert(struct tree_balance *tb, |
737 | struct item_head *ih, | |
738 | const char *body, | |
739 | struct item_head *insert_key, | |
740 | struct buffer_head **insert_ptr, | |
741 | int i) | |
742 | { | |
743 | struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path); | |
744 | int n = B_NR_ITEMS(tbS0); | |
745 | struct buffer_info bi; | |
746 | if (n - tb->snum[i] < tb->item_pos) { /* new item or it's part falls to first new node S_new[i] */ | |
747 | if (tb->item_pos == n - tb->snum[i] + 1 && tb->sbytes[i] != -1) { /* part of new item falls into S_new[i] */ | |
748 | int old_key_comp, old_len, r_zeroes_number; | |
749 | const char *r_body; | |
750 | int version; | |
751 | ||
752 | /* Move snum[i]-1 items from S[0] to S_new[i] */ | |
753 | leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, | |
754 | tb->snum[i] - 1, -1, | |
755 | tb->S_new[i]); | |
756 | /* Remember key component and item length */ | |
757 | version = ih_version(ih); | |
758 | old_key_comp = le_ih_k_offset(ih); | |
759 | old_len = ih_item_len(ih); | |
760 | ||
761 | /* Calculate key component and item length to insert into S_new[i] */ | |
762 | set_le_ih_k_offset(ih, le_ih_k_offset(ih) + | |
763 | ((old_len - tb->sbytes[i]) << (is_indirect_le_ih(ih) ? tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT : 0))); | |
764 | ||
765 | put_ih_item_len(ih, tb->sbytes[i]); | |
766 | ||
767 | /* Insert part of the item into S_new[i] before 0-th item */ | |
768 | buffer_info_init_bh(tb, &bi, tb->S_new[i]); | |
769 | ||
770 | if ((old_len - tb->sbytes[i]) > tb->zeroes_num) { | |
771 | r_zeroes_number = 0; | |
772 | r_body = body + (old_len - tb->sbytes[i]) - tb->zeroes_num; | |
773 | } else { | |
774 | r_body = body; | |
775 | r_zeroes_number = tb->zeroes_num - (old_len - tb->sbytes[i]); | |
776 | tb->zeroes_num -= r_zeroes_number; | |
777 | } | |
778 | ||
779 | leaf_insert_into_buf(&bi, 0, ih, r_body, r_zeroes_number); | |
780 | ||
781 | /* Calculate key component and item length to insert into S[i] */ | |
782 | set_le_ih_k_offset(ih, old_key_comp); | |
783 | put_ih_item_len(ih, old_len - tb->sbytes[i]); | |
784 | tb->insert_size[0] -= tb->sbytes[i]; | |
785 | } else { /* whole new item falls into S_new[i] */ | |
786 | ||
787 | /* Shift snum[0] - 1 items to S_new[i] (sbytes[i] of split item) */ | |
788 | leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, | |
789 | tb->snum[i] - 1, tb->sbytes[i], tb->S_new[i]); | |
790 | ||
791 | /* Insert new item into S_new[i] */ | |
792 | buffer_info_init_bh(tb, &bi, tb->S_new[i]); | |
793 | leaf_insert_into_buf(&bi, tb->item_pos - n + tb->snum[i] - 1, | |
794 | ih, body, tb->zeroes_num); | |
795 | ||
796 | tb->zeroes_num = tb->insert_size[0] = 0; | |
797 | } | |
798 | } | |
799 | ||
800 | else { /* new item or it part don't falls into S_new[i] */ | |
801 | ||
802 | leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, | |
803 | tb->snum[i], tb->sbytes[i], tb->S_new[i]); | |
804 | } | |
805 | } | |
806 | ||
9d496552 JM |
807 | static void balance_leaf_new_nodes_paste(struct tree_balance *tb, |
808 | struct item_head *ih, | |
809 | const char *body, | |
810 | struct item_head *insert_key, | |
811 | struct buffer_head **insert_ptr, | |
812 | int i) | |
813 | { | |
814 | struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path); | |
815 | int n = B_NR_ITEMS(tbS0); | |
816 | struct buffer_info bi; | |
817 | if (n - tb->snum[i] <= tb->item_pos) { /* pasted item or part if it falls to S_new[i] */ | |
818 | if (tb->item_pos == n - tb->snum[i] && tb->sbytes[i] != -1) { /* we must shift part of the appended item */ | |
819 | struct item_head *aux_ih; | |
820 | ||
821 | RFALSE(ih, "PAP-12210: ih must be 0"); | |
822 | ||
823 | aux_ih = item_head(tbS0, tb->item_pos); | |
824 | if (is_direntry_le_ih(aux_ih)) { | |
825 | /* we append to directory item */ | |
826 | ||
827 | int entry_count; | |
828 | ||
829 | entry_count = ih_entry_count(aux_ih); | |
830 | ||
831 | if (entry_count - tb->sbytes[i] < tb->pos_in_item && tb->pos_in_item <= entry_count) { | |
832 | /* new directory entry falls into S_new[i] */ | |
833 | ||
834 | RFALSE(!tb->insert_size[0], "PAP-12215: insert_size is already 0"); | |
835 | RFALSE(tb->sbytes[i] - 1 >= entry_count, | |
836 | "PAP-12220: there are no so much entries (%d), only %d", | |
837 | tb->sbytes[i] - 1, entry_count); | |
838 | ||
839 | /* Shift snum[i]-1 items in whole. Shift sbytes[i] directory entries from directory item number snum[i] */ | |
840 | leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i], tb->sbytes[i] - 1, tb->S_new[i]); | |
841 | /* Paste given directory entry to directory item */ | |
842 | buffer_info_init_bh(tb, &bi, tb->S_new[i]); | |
843 | leaf_paste_in_buffer(&bi, 0, tb->pos_in_item - entry_count + tb->sbytes[i] - 1, | |
844 | tb->insert_size[0], body, tb->zeroes_num); | |
845 | /* paste new directory entry */ | |
846 | leaf_paste_entries(&bi, 0, tb->pos_in_item - entry_count + tb->sbytes[i] - 1, 1, | |
847 | (struct reiserfs_de_head *) body, | |
848 | body + DEH_SIZE, tb->insert_size[0]); | |
849 | tb->insert_size[0] = 0; | |
850 | tb->pos_in_item++; | |
851 | } else { /* new directory entry doesn't fall into S_new[i] */ | |
852 | leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i], tb->sbytes[i], tb->S_new[i]); | |
853 | } | |
854 | } else { /* regular object */ | |
855 | ||
856 | int n_shift, n_rem, r_zeroes_number; | |
857 | const char *r_body; | |
858 | ||
859 | RFALSE(tb->pos_in_item != ih_item_len(item_head(tbS0, tb->item_pos)) || tb->insert_size[0] <= 0, | |
860 | "PAP-12225: item too short or insert_size <= 0"); | |
861 | ||
862 | /* Calculate number of bytes which must be shifted from appended item */ | |
863 | n_shift = tb->sbytes[i] - tb->insert_size[0]; | |
864 | if (n_shift < 0) | |
865 | n_shift = 0; | |
866 | leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, tb->snum[i], n_shift, tb->S_new[i]); | |
867 | ||
868 | /* Calculate number of bytes which must remain in body after append to S_new[i] */ | |
869 | n_rem = tb->insert_size[0] - tb->sbytes[i]; | |
870 | if (n_rem < 0) | |
871 | n_rem = 0; | |
872 | /* Append part of body into S_new[0] */ | |
873 | buffer_info_init_bh(tb, &bi, tb->S_new[i]); | |
874 | if (n_rem > tb->zeroes_num) { | |
875 | r_zeroes_number = 0; | |
876 | r_body = body + n_rem - tb->zeroes_num; | |
877 | } else { | |
878 | r_body = body; | |
879 | r_zeroes_number = tb->zeroes_num - n_rem; | |
880 | tb->zeroes_num -= r_zeroes_number; | |
881 | } | |
882 | ||
883 | leaf_paste_in_buffer(&bi, 0, n_shift, | |
884 | tb->insert_size[0] - n_rem, | |
885 | r_body, r_zeroes_number); | |
886 | { | |
887 | struct item_head *tmp; | |
888 | ||
889 | tmp = item_head(tb->S_new[i], 0); | |
890 | if (is_indirect_le_ih | |
891 | (tmp)) { | |
892 | set_ih_free_space(tmp, 0); | |
893 | set_le_ih_k_offset(tmp, le_ih_k_offset(tmp) + (n_rem << (tb->tb_sb->s_blocksize_bits - UNFM_P_SHIFT))); | |
894 | } else { | |
895 | set_le_ih_k_offset(tmp, le_ih_k_offset(tmp) + n_rem); | |
896 | } | |
897 | } | |
898 | ||
899 | tb->insert_size[0] = n_rem; | |
900 | if (!n_rem) | |
901 | tb->pos_in_item++; | |
902 | } | |
903 | } else | |
904 | /* item falls wholly into S_new[i] */ | |
905 | { | |
906 | int leaf_mi; | |
907 | struct item_head *pasted; | |
908 | ||
909 | #ifdef CONFIG_REISERFS_CHECK | |
910 | struct item_head *ih_check = item_head(tbS0, tb->item_pos); | |
911 | ||
912 | if (!is_direntry_le_ih(ih_check) | |
913 | && (tb->pos_in_item != ih_item_len(ih_check) | |
914 | || tb->insert_size[0] <= 0)) | |
915 | reiserfs_panic(tb->tb_sb, | |
916 | "PAP-12235", | |
917 | "pos_in_item " | |
918 | "must be equal " | |
919 | "to ih_item_len"); | |
920 | #endif /* CONFIG_REISERFS_CHECK */ | |
921 | ||
922 | leaf_mi = leaf_move_items(LEAF_FROM_S_TO_SNEW, | |
923 | tb, tb->snum[i], | |
924 | tb->sbytes[i], | |
925 | tb->S_new[i]); | |
926 | ||
927 | RFALSE(leaf_mi, | |
928 | "PAP-12240: unexpected value returned by leaf_move_items (%d)", | |
929 | leaf_mi); | |
930 | ||
931 | /* paste into item */ | |
932 | buffer_info_init_bh(tb, &bi, tb->S_new[i]); | |
933 | leaf_paste_in_buffer(&bi, | |
934 | tb->item_pos - n + tb->snum[i], | |
935 | tb->pos_in_item, | |
936 | tb->insert_size[0], | |
937 | body, tb->zeroes_num); | |
938 | ||
939 | pasted = item_head(tb->S_new[i], tb->item_pos - n + tb->snum[i]); | |
940 | if (is_direntry_le_ih(pasted)) { | |
941 | leaf_paste_entries(&bi, | |
942 | tb->item_pos - n + tb->snum[i], | |
943 | tb->pos_in_item, 1, | |
944 | (struct reiserfs_de_head *)body, | |
945 | body + DEH_SIZE, | |
946 | tb->insert_size[0] | |
947 | ); | |
948 | } | |
949 | ||
950 | /* if we paste to indirect item update ih_free_space */ | |
951 | if (is_indirect_le_ih(pasted)) | |
952 | set_ih_free_space(pasted, 0); | |
953 | tb->zeroes_num = tb->insert_size[0] = 0; | |
954 | } | |
955 | } | |
956 | ||
957 | else { /* pasted item doesn't fall into S_new[i] */ | |
958 | ||
959 | leaf_move_items(LEAF_FROM_S_TO_SNEW, tb, | |
960 | tb->snum[i], tb->sbytes[i], tb->S_new[i]); | |
961 | } | |
962 | ||
963 | } | |
964 | ||
3f0eb276 JM |
965 | /** |
966 | * balance_leaf - reiserfs tree balancing algorithm | |
967 | * @tb: tree balance state | |
968 | * @ih: item header of inserted item (little endian) | |
969 | * @body: body of inserted item or bytes to paste | |
970 | * @flag: i - insert, d - delete, c - cut, p - paste (see do_balance) | |
971 | * passed back: | |
972 | * @insert_key: key to insert new nodes | |
973 | * @insert_ptr: array of nodes to insert at the next level | |
974 | * | |
975 | * In our processing of one level we sometimes determine what must be | |
976 | * inserted into the next higher level. This insertion consists of a | |
977 | * key or two keys and their corresponding pointers. | |
978 | */ | |
979 | static int balance_leaf(struct tree_balance *tb, struct item_head *ih, | |
980 | const char *body, int flag, | |
981 | struct item_head *insert_key, | |
982 | struct buffer_head **insert_ptr) | |
983 | { | |
984 | struct buffer_head *tbS0 = PATH_PLAST_BUFFER(tb->tb_path); | |
985 | struct buffer_info bi; | |
986 | int n, i; | |
987 | ||
988 | PROC_INFO_INC(tb->tb_sb, balance_at[0]); | |
989 | ||
990 | /* Make balance in case insert_size[0] < 0 */ | |
991 | if (tb->insert_size[0] < 0) | |
992 | return balance_leaf_when_delete(tb, flag); | |
993 | ||
994 | tb->item_pos = PATH_LAST_POSITION(tb->tb_path), | |
995 | tb->pos_in_item = tb->tb_path->pos_in_item, | |
996 | tb->zeroes_num = 0; | |
997 | if (flag == M_INSERT && !body) | |
998 | tb->zeroes_num = ih_item_len(ih); | |
999 | ||
1000 | /* | |
1001 | * for indirect item pos_in_item is measured in unformatted node | |
1002 | * pointers. Recalculate to bytes | |
1003 | */ | |
1004 | if (flag != M_INSERT | |
1005 | && is_indirect_le_ih(item_head(tbS0, tb->item_pos))) | |
1006 | tb->pos_in_item *= UNFM_P_SIZE; | |
1007 | ||
1008 | if (tb->lnum[0] > 0) { | |
1009 | /* Shift lnum[0] items from S[0] to the left neighbor L[0] */ | |
1010 | if (tb->item_pos < tb->lnum[0]) { | |
1011 | /* new item or it part falls to L[0], shift it too */ | |
1012 | n = B_NR_ITEMS(tb->L[0]); | |
1013 | ||
1014 | switch (flag) { | |
1015 | case M_INSERT: /* insert item into L[0] */ | |
1016 | balance_leaf_insert_left(tb, ih, body); | |
1017 | break; | |
1018 | ||
1019 | case M_PASTE: /* append item in L[0] */ | |
1020 | balance_leaf_paste_left(tb, ih, body); | |
1021 | break; | |
1022 | default: /* cases d and t */ | |
1023 | reiserfs_panic(tb->tb_sb, "PAP-12130", | |
1024 | "lnum > 0: unexpected mode: " | |
1025 | " %s(%d)", | |
1026 | (flag == M_DELETE) ? "DELETE" : ((flag == M_CUT) ? "CUT" : "UNKNOWN"), flag); | |
1027 | } | |
1028 | } else { | |
1029 | /* new item doesn't fall into L[0] */ | |
1030 | leaf_shift_left(tb, tb->lnum[0], tb->lbytes); | |
1031 | } | |
1032 | } | |
1033 | ||
1034 | /* tb->lnum[0] > 0 */ | |
1035 | /* Calculate new item position */ | |
1036 | tb->item_pos -= (tb->lnum[0] - ((tb->lbytes != -1) ? 1 : 0)); | |
1037 | ||
1038 | if (tb->rnum[0] > 0) { | |
1039 | /* shift rnum[0] items from S[0] to the right neighbor R[0] */ | |
1040 | n = B_NR_ITEMS(tbS0); | |
1041 | switch (flag) { | |
1042 | ||
1043 | case M_INSERT: /* insert item */ | |
1044 | balance_leaf_insert_right(tb, ih, body); | |
1045 | break; | |
1046 | ||
1047 | case M_PASTE: /* append item */ | |
1048 | balance_leaf_paste_right(tb, ih, body); | |
bd4c625c LT |
1049 | break; |
1050 | default: /* cases d and t */ | |
c3a9c210 JM |
1051 | reiserfs_panic(tb->tb_sb, "PAP-12175", |
1052 | "rnum > 0: unexpected mode: %s(%d)", | |
416e2abd | 1053 | (flag == M_DELETE) ? "DELETE" : ((flag == M_CUT) ? "CUT" : "UNKNOWN"), flag); |
1da177e4 | 1054 | } |
1da177e4 | 1055 | |
bd4c625c | 1056 | } |
1da177e4 | 1057 | |
bd4c625c LT |
1058 | /* tb->rnum[0] > 0 */ |
1059 | RFALSE(tb->blknum[0] > 3, | |
416e2abd | 1060 | "PAP-12180: blknum can not be %d. It must be <= 3", tb->blknum[0]); |
bd4c625c | 1061 | RFALSE(tb->blknum[0] < 0, |
416e2abd | 1062 | "PAP-12185: blknum can not be %d. It must be >= 0", tb->blknum[0]); |
bd4c625c | 1063 | |
97fd4b97 JM |
1064 | /* |
1065 | * if while adding to a node we discover that it is possible to split | |
1066 | * it in two, and merge the left part into the left neighbor and the | |
1067 | * right part into the right neighbor, eliminating the node | |
1068 | */ | |
bd4c625c LT |
1069 | if (tb->blknum[0] == 0) { /* node S[0] is empty now */ |
1070 | ||
1071 | RFALSE(!tb->lnum[0] || !tb->rnum[0], | |
1072 | "PAP-12190: lnum and rnum must not be zero"); | |
97fd4b97 JM |
1073 | /* |
1074 | * if insertion was done before 0-th position in R[0], right | |
1075 | * delimiting key of the tb->L[0]'s and left delimiting key are | |
1076 | * not set correctly | |
1077 | */ | |
bd4c625c LT |
1078 | if (tb->CFL[0]) { |
1079 | if (!tb->CFR[0]) | |
c3a9c210 JM |
1080 | reiserfs_panic(tb->tb_sb, "vs-12195", |
1081 | "CFR not initialized"); | |
4cf5f7ad JM |
1082 | copy_key(internal_key(tb->CFL[0], tb->lkey[0]), |
1083 | internal_key(tb->CFR[0], tb->rkey[0])); | |
bd4c625c LT |
1084 | do_balance_mark_internal_dirty(tb, tb->CFL[0], 0); |
1085 | } | |
1da177e4 | 1086 | |
bd4c625c LT |
1087 | reiserfs_invalidate_buffer(tb, tbS0); |
1088 | return 0; | |
1089 | } | |
1da177e4 | 1090 | |
bd4c625c | 1091 | /* Fill new nodes that appear in place of S[0] */ |
bd4c625c LT |
1092 | for (i = tb->blknum[0] - 2; i >= 0; i--) { |
1093 | ||
b49fb112 JM |
1094 | RFALSE(!tb->snum[i], |
1095 | "PAP-12200: snum[%d] == %d. Must be > 0", i, | |
1096 | tb->snum[i]); | |
bd4c625c LT |
1097 | |
1098 | /* here we shift from S to S_new nodes */ | |
1099 | ||
b49fb112 | 1100 | tb->S_new[i] = get_FEB(tb); |
bd4c625c LT |
1101 | |
1102 | /* initialized block type and tree level */ | |
b49fb112 | 1103 | set_blkh_level(B_BLK_HEAD(tb->S_new[i]), DISK_LEAF_NODE_LEVEL); |
bd4c625c LT |
1104 | |
1105 | n = B_NR_ITEMS(tbS0); | |
1106 | ||
1107 | switch (flag) { | |
1108 | case M_INSERT: /* insert item */ | |
65ab18cb JM |
1109 | balance_leaf_new_nodes_insert(tb, ih, body, insert_key, |
1110 | insert_ptr, i); | |
bd4c625c LT |
1111 | break; |
1112 | ||
1113 | case M_PASTE: /* append item */ | |
9d496552 JM |
1114 | balance_leaf_new_nodes_paste(tb, ih, body, insert_key, |
1115 | insert_ptr, i); | |
bd4c625c LT |
1116 | break; |
1117 | default: /* cases d and t */ | |
c3a9c210 JM |
1118 | reiserfs_panic(tb->tb_sb, "PAP-12245", |
1119 | "blknum > 2: unexpected mode: %s(%d)", | |
416e2abd | 1120 | (flag == M_DELETE) ? "DELETE" : ((flag == M_CUT) ? "CUT" : "UNKNOWN"), flag); |
1da177e4 | 1121 | } |
1da177e4 | 1122 | |
b49fb112 JM |
1123 | memcpy(insert_key + i, leaf_key(tb->S_new[i], 0), KEY_SIZE); |
1124 | insert_ptr[i] = tb->S_new[i]; | |
bd4c625c | 1125 | |
b49fb112 JM |
1126 | RFALSE(!buffer_journaled(tb->S_new[i]) |
1127 | || buffer_journal_dirty(tb->S_new[i]) | |
1128 | || buffer_dirty(tb->S_new[i]), | |
1129 | "PAP-12247: S_new[%d] : (%b)", | |
1130 | i, tb->S_new[i]); | |
1da177e4 LT |
1131 | } |
1132 | ||
bd4c625c LT |
1133 | /* if the affected item was not wholly shifted then we perform all necessary operations on that part or whole of the |
1134 | affected item which remains in S */ | |
b49fb112 | 1135 | if (0 <= tb->item_pos && tb->item_pos < tb->s0num) { /* if we must insert or append into buffer S[0] */ |
bd4c625c LT |
1136 | |
1137 | switch (flag) { | |
1138 | case M_INSERT: /* insert item into S[0] */ | |
fba4ebb5 | 1139 | buffer_info_init_tbS0(tb, &bi); |
b49fb112 JM |
1140 | leaf_insert_into_buf(&bi, tb->item_pos, ih, |
1141 | body, tb->zeroes_num); | |
bd4c625c LT |
1142 | |
1143 | /* If we insert the first key change the delimiting key */ | |
b49fb112 | 1144 | if (tb->item_pos == 0) { |
bd4c625c | 1145 | if (tb->CFL[0]) /* can be 0 in reiserfsck */ |
416e2abd | 1146 | replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0); |
1da177e4 | 1147 | } |
bd4c625c LT |
1148 | break; |
1149 | ||
1150 | case M_PASTE:{ /* append item in S[0] */ | |
1151 | struct item_head *pasted; | |
1152 | ||
b49fb112 | 1153 | pasted = item_head(tbS0, tb->item_pos); |
bd4c625c LT |
1154 | /* when directory, may be new entry already pasted */ |
1155 | if (is_direntry_le_ih(pasted)) { | |
b49fb112 | 1156 | if (tb->pos_in_item >= 0 && tb->pos_in_item <= ih_entry_count(pasted)) { |
bd4c625c LT |
1157 | |
1158 | RFALSE(!tb->insert_size[0], | |
1159 | "PAP-12260: insert_size is 0 already"); | |
1160 | ||
1161 | /* prepare space */ | |
fba4ebb5 | 1162 | buffer_info_init_tbS0(tb, &bi); |
b49fb112 | 1163 | leaf_paste_in_buffer(&bi, tb->item_pos, tb->pos_in_item, |
416e2abd | 1164 | tb->insert_size[0], body, |
b49fb112 | 1165 | tb->zeroes_num); |
bd4c625c LT |
1166 | |
1167 | /* paste entry */ | |
b49fb112 | 1168 | leaf_paste_entries(&bi, tb->item_pos, tb->pos_in_item, 1, |
416e2abd DJ |
1169 | (struct reiserfs_de_head *)body, |
1170 | body + DEH_SIZE, | |
1171 | tb->insert_size[0]); | |
b49fb112 | 1172 | if (!tb->item_pos && !tb->pos_in_item) { |
416e2abd | 1173 | RFALSE(!tb->CFL[0] || !tb->L[0], |
bd4c625c | 1174 | "PAP-12270: CFL[0]/L[0] must be specified"); |
416e2abd DJ |
1175 | if (tb->CFL[0]) |
1176 | replace_key(tb, tb->CFL[0], tb->lkey[0], tbS0, 0); | |
bd4c625c LT |
1177 | } |
1178 | tb->insert_size[0] = 0; | |
1179 | } | |
1180 | } else { /* regular object */ | |
b49fb112 | 1181 | if (tb->pos_in_item == ih_item_len(pasted)) { |
bd4c625c LT |
1182 | |
1183 | RFALSE(tb->insert_size[0] <= 0, | |
1184 | "PAP-12275: insert size must not be %d", | |
1185 | tb->insert_size[0]); | |
fba4ebb5 | 1186 | buffer_info_init_tbS0(tb, &bi); |
b49fb112 JM |
1187 | leaf_paste_in_buffer(&bi, tb->item_pos, tb->pos_in_item, |
1188 | tb->insert_size[0], body, tb->zeroes_num); | |
bd4c625c LT |
1189 | |
1190 | if (is_indirect_le_ih(pasted)) { | |
1da177e4 | 1191 | #if 0 |
bd4c625c LT |
1192 | RFALSE(tb-> |
1193 | insert_size[0] != | |
1194 | UNFM_P_SIZE, | |
1195 | "PAP-12280: insert_size for indirect item must be %d, not %d", | |
1196 | UNFM_P_SIZE, | |
1197 | tb-> | |
1198 | insert_size[0]); | |
1da177e4 | 1199 | #endif |
416e2abd | 1200 | set_ih_free_space(pasted, 0); |
bd4c625c LT |
1201 | } |
1202 | tb->insert_size[0] = 0; | |
1203 | } | |
1da177e4 | 1204 | #ifdef CONFIG_REISERFS_CHECK |
bd4c625c LT |
1205 | else { |
1206 | if (tb->insert_size[0]) { | |
1207 | print_cur_tb("12285"); | |
416e2abd | 1208 | reiserfs_panic(tb->tb_sb, |
c3a9c210 JM |
1209 | "PAP-12285", |
1210 | "insert_size " | |
1211 | "must be 0 " | |
1212 | "(%d)", | |
1213 | tb->insert_size[0]); | |
bd4c625c LT |
1214 | } |
1215 | } | |
1216 | #endif /* CONFIG_REISERFS_CHECK */ | |
1217 | ||
1218 | } | |
1219 | } /* case M_PASTE: */ | |
1da177e4 | 1220 | } |
1da177e4 | 1221 | } |
1da177e4 | 1222 | #ifdef CONFIG_REISERFS_CHECK |
bd4c625c LT |
1223 | if (flag == M_PASTE && tb->insert_size[0]) { |
1224 | print_cur_tb("12290"); | |
1225 | reiserfs_panic(tb->tb_sb, | |
c3a9c210 | 1226 | "PAP-12290", "insert_size is still not 0 (%d)", |
bd4c625c LT |
1227 | tb->insert_size[0]); |
1228 | } | |
97fd4b97 JM |
1229 | #endif |
1230 | ||
1231 | /* Leaf level of the tree is balanced (end of balance_leaf) */ | |
bd4c625c | 1232 | return 0; |
97fd4b97 | 1233 | } |
1da177e4 LT |
1234 | |
1235 | /* Make empty node */ | |
bd4c625c | 1236 | void make_empty_node(struct buffer_info *bi) |
1da177e4 | 1237 | { |
bd4c625c | 1238 | struct block_head *blkh; |
1da177e4 | 1239 | |
bd4c625c | 1240 | RFALSE(bi->bi_bh == NULL, "PAP-12295: pointer to the buffer is NULL"); |
1da177e4 | 1241 | |
bd4c625c LT |
1242 | blkh = B_BLK_HEAD(bi->bi_bh); |
1243 | set_blkh_nr_item(blkh, 0); | |
1244 | set_blkh_free_space(blkh, MAX_CHILD_SIZE(bi->bi_bh)); | |
1da177e4 | 1245 | |
bd4c625c LT |
1246 | if (bi->bi_parent) |
1247 | B_N_CHILD(bi->bi_parent, bi->bi_position)->dc_size = 0; /* Endian safe if 0 */ | |
1da177e4 LT |
1248 | } |
1249 | ||
1da177e4 | 1250 | /* Get first empty buffer */ |
bd4c625c | 1251 | struct buffer_head *get_FEB(struct tree_balance *tb) |
1da177e4 | 1252 | { |
bd4c625c | 1253 | int i; |
bd4c625c | 1254 | struct buffer_info bi; |
1da177e4 | 1255 | |
bd4c625c | 1256 | for (i = 0; i < MAX_FEB_SIZE; i++) |
9dce07f1 | 1257 | if (tb->FEB[i] != NULL) |
bd4c625c LT |
1258 | break; |
1259 | ||
1260 | if (i == MAX_FEB_SIZE) | |
c3a9c210 | 1261 | reiserfs_panic(tb->tb_sb, "vs-12300", "FEB list is empty"); |
bd4c625c | 1262 | |
fba4ebb5 | 1263 | buffer_info_init_bh(tb, &bi, tb->FEB[i]); |
bd4c625c | 1264 | make_empty_node(&bi); |
fba4ebb5 JM |
1265 | set_buffer_uptodate(tb->FEB[i]); |
1266 | tb->used[i] = tb->FEB[i]; | |
bd4c625c | 1267 | tb->FEB[i] = NULL; |
bd4c625c | 1268 | |
fba4ebb5 | 1269 | return tb->used[i]; |
bd4c625c | 1270 | } |
1da177e4 | 1271 | |
098297b2 | 1272 | /* This is now used because reiserfs_free_block has to be able to schedule. */ |
bd4c625c | 1273 | static void store_thrown(struct tree_balance *tb, struct buffer_head *bh) |
1da177e4 | 1274 | { |
bd4c625c LT |
1275 | int i; |
1276 | ||
1277 | if (buffer_dirty(bh)) | |
45b03d5e JM |
1278 | reiserfs_warning(tb->tb_sb, "reiserfs-12320", |
1279 | "called with dirty buffer"); | |
79a81aef | 1280 | for (i = 0; i < ARRAY_SIZE(tb->thrown); i++) |
bd4c625c LT |
1281 | if (!tb->thrown[i]) { |
1282 | tb->thrown[i] = bh; | |
1283 | get_bh(bh); /* free_thrown puts this */ | |
1284 | return; | |
1285 | } | |
45b03d5e JM |
1286 | reiserfs_warning(tb->tb_sb, "reiserfs-12321", |
1287 | "too many thrown buffers"); | |
1da177e4 LT |
1288 | } |
1289 | ||
bd4c625c LT |
1290 | static void free_thrown(struct tree_balance *tb) |
1291 | { | |
1292 | int i; | |
1293 | b_blocknr_t blocknr; | |
79a81aef | 1294 | for (i = 0; i < ARRAY_SIZE(tb->thrown); i++) { |
bd4c625c LT |
1295 | if (tb->thrown[i]) { |
1296 | blocknr = tb->thrown[i]->b_blocknr; | |
1297 | if (buffer_dirty(tb->thrown[i])) | |
45b03d5e JM |
1298 | reiserfs_warning(tb->tb_sb, "reiserfs-12322", |
1299 | "called with dirty buffer %d", | |
bd4c625c LT |
1300 | blocknr); |
1301 | brelse(tb->thrown[i]); /* incremented in store_thrown */ | |
1302 | reiserfs_free_block(tb->transaction_handle, NULL, | |
1303 | blocknr, 0); | |
1304 | } | |
1da177e4 | 1305 | } |
1da177e4 LT |
1306 | } |
1307 | ||
bd4c625c | 1308 | void reiserfs_invalidate_buffer(struct tree_balance *tb, struct buffer_head *bh) |
1da177e4 | 1309 | { |
bd4c625c LT |
1310 | struct block_head *blkh; |
1311 | blkh = B_BLK_HEAD(bh); | |
1312 | set_blkh_level(blkh, FREE_LEVEL); | |
1313 | set_blkh_nr_item(blkh, 0); | |
1314 | ||
1315 | clear_buffer_dirty(bh); | |
1316 | store_thrown(tb, bh); | |
1da177e4 LT |
1317 | } |
1318 | ||
1319 | /* Replace n_dest'th key in buffer dest by n_src'th key of buffer src.*/ | |
bd4c625c LT |
1320 | void replace_key(struct tree_balance *tb, struct buffer_head *dest, int n_dest, |
1321 | struct buffer_head *src, int n_src) | |
1da177e4 LT |
1322 | { |
1323 | ||
bd4c625c LT |
1324 | RFALSE(dest == NULL || src == NULL, |
1325 | "vs-12305: source or destination buffer is 0 (src=%p, dest=%p)", | |
1326 | src, dest); | |
1327 | RFALSE(!B_IS_KEYS_LEVEL(dest), | |
1328 | "vs-12310: invalid level (%z) for destination buffer. dest must be leaf", | |
1329 | dest); | |
1330 | RFALSE(n_dest < 0 || n_src < 0, | |
1331 | "vs-12315: src(%d) or dest(%d) key number < 0", n_src, n_dest); | |
1332 | RFALSE(n_dest >= B_NR_ITEMS(dest) || n_src >= B_NR_ITEMS(src), | |
1333 | "vs-12320: src(%d(%d)) or dest(%d(%d)) key number is too big", | |
1334 | n_src, B_NR_ITEMS(src), n_dest, B_NR_ITEMS(dest)); | |
1335 | ||
1336 | if (B_IS_ITEMS_LEVEL(src)) | |
1337 | /* source buffer contains leaf node */ | |
4cf5f7ad | 1338 | memcpy(internal_key(dest, n_dest), item_head(src, n_src), |
bd4c625c LT |
1339 | KEY_SIZE); |
1340 | else | |
4cf5f7ad | 1341 | memcpy(internal_key(dest, n_dest), internal_key(src, n_src), |
bd4c625c LT |
1342 | KEY_SIZE); |
1343 | ||
1344 | do_balance_mark_internal_dirty(tb, dest, 0); | |
1da177e4 LT |
1345 | } |
1346 | ||
bd4c625c | 1347 | int get_left_neighbor_position(struct tree_balance *tb, int h) |
1da177e4 | 1348 | { |
bd4c625c | 1349 | int Sh_position = PATH_H_POSITION(tb->tb_path, h + 1); |
1da177e4 | 1350 | |
9dce07f1 | 1351 | RFALSE(PATH_H_PPARENT(tb->tb_path, h) == NULL || tb->FL[h] == NULL, |
bd4c625c LT |
1352 | "vs-12325: FL[%d](%p) or F[%d](%p) does not exist", |
1353 | h, tb->FL[h], h, PATH_H_PPARENT(tb->tb_path, h)); | |
1da177e4 | 1354 | |
bd4c625c LT |
1355 | if (Sh_position == 0) |
1356 | return B_NR_ITEMS(tb->FL[h]); | |
1357 | else | |
1358 | return Sh_position - 1; | |
1da177e4 LT |
1359 | } |
1360 | ||
bd4c625c | 1361 | int get_right_neighbor_position(struct tree_balance *tb, int h) |
1da177e4 | 1362 | { |
bd4c625c | 1363 | int Sh_position = PATH_H_POSITION(tb->tb_path, h + 1); |
1da177e4 | 1364 | |
9dce07f1 | 1365 | RFALSE(PATH_H_PPARENT(tb->tb_path, h) == NULL || tb->FR[h] == NULL, |
bd4c625c LT |
1366 | "vs-12330: F[%d](%p) or FR[%d](%p) does not exist", |
1367 | h, PATH_H_PPARENT(tb->tb_path, h), h, tb->FR[h]); | |
1da177e4 | 1368 | |
bd4c625c LT |
1369 | if (Sh_position == B_NR_ITEMS(PATH_H_PPARENT(tb->tb_path, h))) |
1370 | return 0; | |
1371 | else | |
1372 | return Sh_position + 1; | |
1da177e4 LT |
1373 | } |
1374 | ||
1da177e4 LT |
1375 | #ifdef CONFIG_REISERFS_CHECK |
1376 | ||
bd4c625c LT |
1377 | int is_reusable(struct super_block *s, b_blocknr_t block, int bit_value); |
1378 | static void check_internal_node(struct super_block *s, struct buffer_head *bh, | |
1379 | char *mes) | |
1da177e4 | 1380 | { |
bd4c625c LT |
1381 | struct disk_child *dc; |
1382 | int i; | |
1da177e4 | 1383 | |
bd4c625c | 1384 | RFALSE(!bh, "PAP-12336: bh == 0"); |
1da177e4 | 1385 | |
bd4c625c LT |
1386 | if (!bh || !B_IS_IN_TREE(bh)) |
1387 | return; | |
1da177e4 | 1388 | |
bd4c625c LT |
1389 | RFALSE(!buffer_dirty(bh) && |
1390 | !(buffer_journaled(bh) || buffer_journal_dirty(bh)), | |
1391 | "PAP-12337: buffer (%b) must be dirty", bh); | |
1392 | dc = B_N_CHILD(bh, 0); | |
1da177e4 | 1393 | |
bd4c625c LT |
1394 | for (i = 0; i <= B_NR_ITEMS(bh); i++, dc++) { |
1395 | if (!is_reusable(s, dc_block_number(dc), 1)) { | |
1396 | print_cur_tb(mes); | |
c3a9c210 JM |
1397 | reiserfs_panic(s, "PAP-12338", |
1398 | "invalid child pointer %y in %b", | |
bd4c625c LT |
1399 | dc, bh); |
1400 | } | |
1401 | } | |
1da177e4 LT |
1402 | } |
1403 | ||
45b03d5e JM |
1404 | static int locked_or_not_in_tree(struct tree_balance *tb, |
1405 | struct buffer_head *bh, char *which) | |
bd4c625c LT |
1406 | { |
1407 | if ((!buffer_journal_prepared(bh) && buffer_locked(bh)) || | |
1408 | !B_IS_IN_TREE(bh)) { | |
45b03d5e | 1409 | reiserfs_warning(tb->tb_sb, "vs-12339", "%s (%b)", which, bh); |
bd4c625c LT |
1410 | return 1; |
1411 | } | |
1412 | return 0; | |
1413 | } | |
1da177e4 | 1414 | |
bd4c625c | 1415 | static int check_before_balancing(struct tree_balance *tb) |
1da177e4 | 1416 | { |
bd4c625c LT |
1417 | int retval = 0; |
1418 | ||
08f14fc8 | 1419 | if (REISERFS_SB(tb->tb_sb)->cur_tb) { |
c3a9c210 JM |
1420 | reiserfs_panic(tb->tb_sb, "vs-12335", "suspect that schedule " |
1421 | "occurred based on cur_tb not being null at " | |
1422 | "this point in code. do_balance cannot properly " | |
08f14fc8 FW |
1423 | "handle concurrent tree accesses on a same " |
1424 | "mount point."); | |
1da177e4 | 1425 | } |
bd4c625c | 1426 | |
098297b2 JM |
1427 | /* |
1428 | * double check that buffers that we will modify are unlocked. | |
1429 | * (fix_nodes should already have prepped all of these for us). | |
1430 | */ | |
bd4c625c | 1431 | if (tb->lnum[0]) { |
45b03d5e JM |
1432 | retval |= locked_or_not_in_tree(tb, tb->L[0], "L[0]"); |
1433 | retval |= locked_or_not_in_tree(tb, tb->FL[0], "FL[0]"); | |
1434 | retval |= locked_or_not_in_tree(tb, tb->CFL[0], "CFL[0]"); | |
bd4c625c | 1435 | check_leaf(tb->L[0]); |
1da177e4 | 1436 | } |
bd4c625c | 1437 | if (tb->rnum[0]) { |
45b03d5e JM |
1438 | retval |= locked_or_not_in_tree(tb, tb->R[0], "R[0]"); |
1439 | retval |= locked_or_not_in_tree(tb, tb->FR[0], "FR[0]"); | |
1440 | retval |= locked_or_not_in_tree(tb, tb->CFR[0], "CFR[0]"); | |
bd4c625c LT |
1441 | check_leaf(tb->R[0]); |
1442 | } | |
45b03d5e JM |
1443 | retval |= locked_or_not_in_tree(tb, PATH_PLAST_BUFFER(tb->tb_path), |
1444 | "S[0]"); | |
bd4c625c | 1445 | check_leaf(PATH_PLAST_BUFFER(tb->tb_path)); |
1da177e4 | 1446 | |
bd4c625c LT |
1447 | return retval; |
1448 | } | |
1da177e4 | 1449 | |
bd4c625c | 1450 | static void check_after_balance_leaf(struct tree_balance *tb) |
1da177e4 | 1451 | { |
bd4c625c LT |
1452 | if (tb->lnum[0]) { |
1453 | if (B_FREE_SPACE(tb->L[0]) != | |
1454 | MAX_CHILD_SIZE(tb->L[0]) - | |
1455 | dc_size(B_N_CHILD | |
1456 | (tb->FL[0], get_left_neighbor_position(tb, 0)))) { | |
1457 | print_cur_tb("12221"); | |
c3a9c210 JM |
1458 | reiserfs_panic(tb->tb_sb, "PAP-12355", |
1459 | "shift to left was incorrect"); | |
bd4c625c LT |
1460 | } |
1461 | } | |
1462 | if (tb->rnum[0]) { | |
1463 | if (B_FREE_SPACE(tb->R[0]) != | |
1464 | MAX_CHILD_SIZE(tb->R[0]) - | |
1465 | dc_size(B_N_CHILD | |
1466 | (tb->FR[0], get_right_neighbor_position(tb, 0)))) { | |
1467 | print_cur_tb("12222"); | |
c3a9c210 JM |
1468 | reiserfs_panic(tb->tb_sb, "PAP-12360", |
1469 | "shift to right was incorrect"); | |
bd4c625c LT |
1470 | } |
1471 | } | |
1472 | if (PATH_H_PBUFFER(tb->tb_path, 1) && | |
1473 | (B_FREE_SPACE(PATH_H_PBUFFER(tb->tb_path, 0)) != | |
1474 | (MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)) - | |
1475 | dc_size(B_N_CHILD(PATH_H_PBUFFER(tb->tb_path, 1), | |
1476 | PATH_H_POSITION(tb->tb_path, 1)))))) { | |
1477 | int left = B_FREE_SPACE(PATH_H_PBUFFER(tb->tb_path, 0)); | |
1478 | int right = (MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)) - | |
1479 | dc_size(B_N_CHILD(PATH_H_PBUFFER(tb->tb_path, 1), | |
1480 | PATH_H_POSITION(tb->tb_path, | |
1481 | 1)))); | |
1482 | print_cur_tb("12223"); | |
45b03d5e | 1483 | reiserfs_warning(tb->tb_sb, "reiserfs-12363", |
bd4c625c LT |
1484 | "B_FREE_SPACE (PATH_H_PBUFFER(tb->tb_path,0)) = %d; " |
1485 | "MAX_CHILD_SIZE (%d) - dc_size( %y, %d ) [%d] = %d", | |
1486 | left, | |
1487 | MAX_CHILD_SIZE(PATH_H_PBUFFER(tb->tb_path, 0)), | |
1488 | PATH_H_PBUFFER(tb->tb_path, 1), | |
1489 | PATH_H_POSITION(tb->tb_path, 1), | |
1490 | dc_size(B_N_CHILD | |
1491 | (PATH_H_PBUFFER(tb->tb_path, 1), | |
1492 | PATH_H_POSITION(tb->tb_path, 1))), | |
1493 | right); | |
c3a9c210 | 1494 | reiserfs_panic(tb->tb_sb, "PAP-12365", "S is incorrect"); |
bd4c625c | 1495 | } |
1da177e4 LT |
1496 | } |
1497 | ||
bd4c625c | 1498 | static void check_leaf_level(struct tree_balance *tb) |
1da177e4 | 1499 | { |
bd4c625c LT |
1500 | check_leaf(tb->L[0]); |
1501 | check_leaf(tb->R[0]); | |
1502 | check_leaf(PATH_PLAST_BUFFER(tb->tb_path)); | |
1503 | } | |
1da177e4 | 1504 | |
bd4c625c LT |
1505 | static void check_internal_levels(struct tree_balance *tb) |
1506 | { | |
1507 | int h; | |
1508 | ||
1509 | /* check all internal nodes */ | |
1510 | for (h = 1; tb->insert_size[h]; h++) { | |
1511 | check_internal_node(tb->tb_sb, PATH_H_PBUFFER(tb->tb_path, h), | |
1512 | "BAD BUFFER ON PATH"); | |
1513 | if (tb->lnum[h]) | |
1514 | check_internal_node(tb->tb_sb, tb->L[h], "BAD L"); | |
1515 | if (tb->rnum[h]) | |
1516 | check_internal_node(tb->tb_sb, tb->R[h], "BAD R"); | |
1517 | } | |
1da177e4 LT |
1518 | |
1519 | } | |
1520 | ||
1521 | #endif | |
1522 | ||
098297b2 JM |
1523 | /* |
1524 | * Now we have all of the buffers that must be used in balancing of | |
1525 | * the tree. We rely on the assumption that schedule() will not occur | |
1526 | * while do_balance works. ( Only interrupt handlers are acceptable.) | |
1527 | * We balance the tree according to the analysis made before this, | |
1528 | * using buffers already obtained. For SMP support it will someday be | |
1529 | * necessary to add ordered locking of tb. | |
1530 | */ | |
1da177e4 | 1531 | |
098297b2 JM |
1532 | /* |
1533 | * Some interesting rules of balancing: | |
1534 | * we delete a maximum of two nodes per level per balancing: we never | |
1535 | * delete R, when we delete two of three nodes L, S, R then we move | |
1536 | * them into R. | |
1537 | * | |
1538 | * we only delete L if we are deleting two nodes, if we delete only | |
1539 | * one node we delete S | |
1540 | * | |
1541 | * if we shift leaves then we shift as much as we can: this is a | |
1542 | * deliberate policy of extremism in node packing which results in | |
1543 | * higher average utilization after repeated random balance operations | |
1544 | * at the cost of more memory copies and more balancing as a result of | |
1545 | * small insertions to full nodes. | |
1546 | * | |
1547 | * if we shift internal nodes we try to evenly balance the node | |
1548 | * utilization, with consequent less balancing at the cost of lower | |
1549 | * utilization. | |
1550 | * | |
1551 | * one could argue that the policy for directories in leaves should be | |
1552 | * that of internal nodes, but we will wait until another day to | |
1553 | * evaluate this.... It would be nice to someday measure and prove | |
1554 | * these assumptions as to what is optimal.... | |
1555 | */ | |
1da177e4 | 1556 | |
bd4c625c | 1557 | static inline void do_balance_starts(struct tree_balance *tb) |
1da177e4 | 1558 | { |
098297b2 | 1559 | /* use print_cur_tb() to see initial state of struct tree_balance */ |
1da177e4 | 1560 | |
bd4c625c | 1561 | /* store_print_tb (tb); */ |
1da177e4 | 1562 | |
bd4c625c | 1563 | /* do not delete, just comment it out */ |
098297b2 JM |
1564 | /* |
1565 | print_tb(flag, PATH_LAST_POSITION(tb->tb_path), | |
1566 | tb->tb_path->pos_in_item, tb, "check"); | |
1567 | */ | |
bd4c625c | 1568 | RFALSE(check_before_balancing(tb), "PAP-12340: locked buffers in TB"); |
1da177e4 | 1569 | #ifdef CONFIG_REISERFS_CHECK |
08f14fc8 | 1570 | REISERFS_SB(tb->tb_sb)->cur_tb = tb; |
1da177e4 LT |
1571 | #endif |
1572 | } | |
1573 | ||
bd4c625c | 1574 | static inline void do_balance_completed(struct tree_balance *tb) |
1da177e4 | 1575 | { |
bd4c625c | 1576 | |
1da177e4 | 1577 | #ifdef CONFIG_REISERFS_CHECK |
bd4c625c LT |
1578 | check_leaf_level(tb); |
1579 | check_internal_levels(tb); | |
08f14fc8 | 1580 | REISERFS_SB(tb->tb_sb)->cur_tb = NULL; |
1da177e4 LT |
1581 | #endif |
1582 | ||
098297b2 JM |
1583 | /* |
1584 | * reiserfs_free_block is no longer schedule safe. So, we need to | |
1585 | * put the buffers we want freed on the thrown list during do_balance, | |
1586 | * and then free them now | |
bd4c625c | 1587 | */ |
1da177e4 | 1588 | |
bd4c625c | 1589 | REISERFS_SB(tb->tb_sb)->s_do_balance++; |
1da177e4 | 1590 | |
bd4c625c LT |
1591 | /* release all nodes hold to perform the balancing */ |
1592 | unfix_nodes(tb); | |
1da177e4 | 1593 | |
bd4c625c | 1594 | free_thrown(tb); |
1da177e4 LT |
1595 | } |
1596 | ||
098297b2 JM |
1597 | /* |
1598 | * do_balance - balance the tree | |
1599 | * | |
1600 | * @tb: tree_balance structure | |
1601 | * @ih: item header of inserted item | |
1602 | * @body: body of inserted item or bytes to paste | |
1603 | * @flag: 'i' - insert, 'd' - delete, 'c' - cut, 'p' paste | |
1604 | * | |
1605 | * Cut means delete part of an item (includes removing an entry from a | |
1606 | * directory). | |
1607 | * | |
1608 | * Delete means delete whole item. | |
1609 | * | |
1610 | * Insert means add a new item into the tree. | |
1611 | * | |
1612 | * Paste means to append to the end of an existing file or to | |
1613 | * insert a directory entry. | |
1614 | */ | |
1615 | void do_balance(struct tree_balance *tb, struct item_head *ih, | |
1616 | const char *body, int flag) | |
1617 | { | |
1618 | int child_pos; /* position of a child node in its parent */ | |
1619 | int h; /* level of the tree being processed */ | |
1620 | ||
1621 | /* | |
1622 | * in our processing of one level we sometimes determine what | |
1623 | * must be inserted into the next higher level. This insertion | |
1624 | * consists of a key or two keys and their corresponding | |
1625 | * pointers | |
1626 | */ | |
1627 | struct item_head insert_key[2]; | |
1628 | ||
1629 | /* inserted node-ptrs for the next level */ | |
1630 | struct buffer_head *insert_ptr[2]; | |
bd4c625c LT |
1631 | |
1632 | tb->tb_mode = flag; | |
1633 | tb->need_balance_dirty = 0; | |
1634 | ||
1635 | if (FILESYSTEM_CHANGED_TB(tb)) { | |
c3a9c210 JM |
1636 | reiserfs_panic(tb->tb_sb, "clm-6000", "fs generation has " |
1637 | "changed"); | |
bd4c625c LT |
1638 | } |
1639 | /* if we have no real work to do */ | |
1640 | if (!tb->insert_size[0]) { | |
45b03d5e JM |
1641 | reiserfs_warning(tb->tb_sb, "PAP-12350", |
1642 | "insert_size == 0, mode == %c", flag); | |
bd4c625c LT |
1643 | unfix_nodes(tb); |
1644 | return; | |
1645 | } | |
1da177e4 | 1646 | |
a228bf8f | 1647 | atomic_inc(&fs_generation(tb->tb_sb)); |
bd4c625c | 1648 | do_balance_starts(tb); |
1da177e4 | 1649 | |
098297b2 JM |
1650 | /* |
1651 | * balance_leaf returns 0 except if combining L R and S into | |
1652 | * one node. see balance_internal() for explanation of this | |
1653 | * line of code. | |
1654 | */ | |
bd4c625c LT |
1655 | child_pos = PATH_H_B_ITEM_ORDER(tb->tb_path, 0) + |
1656 | balance_leaf(tb, ih, body, flag, insert_key, insert_ptr); | |
1da177e4 LT |
1657 | |
1658 | #ifdef CONFIG_REISERFS_CHECK | |
bd4c625c | 1659 | check_after_balance_leaf(tb); |
1da177e4 LT |
1660 | #endif |
1661 | ||
bd4c625c LT |
1662 | /* Balance internal level of the tree. */ |
1663 | for (h = 1; h < MAX_HEIGHT && tb->insert_size[h]; h++) | |
1664 | child_pos = | |
1665 | balance_internal(tb, h, child_pos, insert_key, insert_ptr); | |
1da177e4 | 1666 | |
bd4c625c | 1667 | do_balance_completed(tb); |
1da177e4 LT |
1668 | |
1669 | } |