]>
Commit | Line | Data |
---|---|---|
1e51764a AB |
1 | /* |
2 | * This file is part of UBIFS. | |
3 | * | |
4 | * Copyright (C) 2006-2008 Nokia Corporation. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify it | |
7 | * under the terms of the GNU General Public License version 2 as published by | |
8 | * the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along with | |
16 | * this program; if not, write to the Free Software Foundation, Inc., 51 | |
17 | * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
18 | * | |
19 | * Authors: Adrian Hunter | |
20 | * Artem Bityutskiy (Битюцкий Артём) | |
21 | */ | |
22 | ||
23 | /* This file implements TNC functions for committing */ | |
24 | ||
2b1844a8 | 25 | #include <linux/random.h> |
8d7819b4 | 26 | #include "ubifs.h" |
1e51764a AB |
27 | |
28 | /** | |
29 | * make_idx_node - make an index node for fill-the-gaps method of TNC commit. | |
30 | * @c: UBIFS file-system description object | |
31 | * @idx: buffer in which to place new index node | |
32 | * @znode: znode from which to make new index node | |
33 | * @lnum: LEB number where new index node will be written | |
34 | * @offs: offset where new index node will be written | |
35 | * @len: length of new index node | |
36 | */ | |
37 | static int make_idx_node(struct ubifs_info *c, struct ubifs_idx_node *idx, | |
38 | struct ubifs_znode *znode, int lnum, int offs, int len) | |
39 | { | |
40 | struct ubifs_znode *zp; | |
41 | int i, err; | |
42 | ||
43 | /* Make index node */ | |
44 | idx->ch.node_type = UBIFS_IDX_NODE; | |
45 | idx->child_cnt = cpu_to_le16(znode->child_cnt); | |
46 | idx->level = cpu_to_le16(znode->level); | |
47 | for (i = 0; i < znode->child_cnt; i++) { | |
48 | struct ubifs_branch *br = ubifs_idx_branch(c, idx, i); | |
49 | struct ubifs_zbranch *zbr = &znode->zbranch[i]; | |
50 | ||
51 | key_write_idx(c, &zbr->key, &br->key); | |
52 | br->lnum = cpu_to_le32(zbr->lnum); | |
53 | br->offs = cpu_to_le32(zbr->offs); | |
54 | br->len = cpu_to_le32(zbr->len); | |
55 | if (!zbr->lnum || !zbr->len) { | |
235c362b | 56 | ubifs_err(c, "bad ref in znode"); |
edf6be24 | 57 | ubifs_dump_znode(c, znode); |
1e51764a | 58 | if (zbr->znode) |
edf6be24 | 59 | ubifs_dump_znode(c, zbr->znode); |
df71b091 RW |
60 | |
61 | return -EINVAL; | |
1e51764a AB |
62 | } |
63 | } | |
64 | ubifs_prepare_node(c, idx, len, 0); | |
65 | ||
1e51764a AB |
66 | znode->lnum = lnum; |
67 | znode->offs = offs; | |
68 | znode->len = len; | |
1e51764a AB |
69 | |
70 | err = insert_old_idx_znode(c, znode); | |
71 | ||
72 | /* Update the parent */ | |
73 | zp = znode->parent; | |
74 | if (zp) { | |
75 | struct ubifs_zbranch *zbr; | |
76 | ||
77 | zbr = &zp->zbranch[znode->iip]; | |
78 | zbr->lnum = lnum; | |
79 | zbr->offs = offs; | |
80 | zbr->len = len; | |
81 | } else { | |
82 | c->zroot.lnum = lnum; | |
83 | c->zroot.offs = offs; | |
84 | c->zroot.len = len; | |
85 | } | |
86 | c->calc_idx_sz += ALIGN(len, 8); | |
87 | ||
88 | atomic_long_dec(&c->dirty_zn_cnt); | |
89 | ||
90 | ubifs_assert(ubifs_zn_dirty(znode)); | |
f42eed7c | 91 | ubifs_assert(ubifs_zn_cow(znode)); |
1e51764a | 92 | |
06b282a4 AB |
93 | /* |
94 | * Note, unlike 'write_index()' we do not add memory barriers here | |
95 | * because this function is called with @c->tnc_mutex locked. | |
96 | */ | |
1e51764a AB |
97 | __clear_bit(DIRTY_ZNODE, &znode->flags); |
98 | __clear_bit(COW_ZNODE, &znode->flags); | |
99 | ||
100 | return err; | |
101 | } | |
102 | ||
103 | /** | |
104 | * fill_gap - make index nodes in gaps in dirty index LEBs. | |
105 | * @c: UBIFS file-system description object | |
106 | * @lnum: LEB number that gap appears in | |
107 | * @gap_start: offset of start of gap | |
108 | * @gap_end: offset of end of gap | |
109 | * @dirt: adds dirty space to this | |
110 | * | |
111 | * This function returns the number of index nodes written into the gap. | |
112 | */ | |
113 | static int fill_gap(struct ubifs_info *c, int lnum, int gap_start, int gap_end, | |
114 | int *dirt) | |
115 | { | |
116 | int len, gap_remains, gap_pos, written, pad_len; | |
117 | ||
118 | ubifs_assert((gap_start & 7) == 0); | |
119 | ubifs_assert((gap_end & 7) == 0); | |
120 | ubifs_assert(gap_end >= gap_start); | |
121 | ||
122 | gap_remains = gap_end - gap_start; | |
123 | if (!gap_remains) | |
124 | return 0; | |
125 | gap_pos = gap_start; | |
126 | written = 0; | |
127 | while (c->enext) { | |
128 | len = ubifs_idx_node_sz(c, c->enext->child_cnt); | |
129 | if (len < gap_remains) { | |
130 | struct ubifs_znode *znode = c->enext; | |
131 | const int alen = ALIGN(len, 8); | |
132 | int err; | |
133 | ||
134 | ubifs_assert(alen <= gap_remains); | |
135 | err = make_idx_node(c, c->ileb_buf + gap_pos, znode, | |
136 | lnum, gap_pos, len); | |
137 | if (err) | |
138 | return err; | |
139 | gap_remains -= alen; | |
140 | gap_pos += alen; | |
141 | c->enext = znode->cnext; | |
142 | if (c->enext == c->cnext) | |
143 | c->enext = NULL; | |
144 | written += 1; | |
145 | } else | |
146 | break; | |
147 | } | |
148 | if (gap_end == c->leb_size) { | |
149 | c->ileb_len = ALIGN(gap_pos, c->min_io_size); | |
150 | /* Pad to end of min_io_size */ | |
151 | pad_len = c->ileb_len - gap_pos; | |
152 | } else | |
153 | /* Pad to end of gap */ | |
154 | pad_len = gap_remains; | |
155 | dbg_gc("LEB %d:%d to %d len %d nodes written %d wasted bytes %d", | |
156 | lnum, gap_start, gap_end, gap_end - gap_start, written, pad_len); | |
157 | ubifs_pad(c, c->ileb_buf + gap_pos, pad_len); | |
158 | *dirt += pad_len; | |
159 | return written; | |
160 | } | |
161 | ||
162 | /** | |
163 | * find_old_idx - find an index node obsoleted since the last commit start. | |
164 | * @c: UBIFS file-system description object | |
165 | * @lnum: LEB number of obsoleted index node | |
166 | * @offs: offset of obsoleted index node | |
167 | * | |
168 | * Returns %1 if found and %0 otherwise. | |
169 | */ | |
170 | static int find_old_idx(struct ubifs_info *c, int lnum, int offs) | |
171 | { | |
172 | struct ubifs_old_idx *o; | |
173 | struct rb_node *p; | |
174 | ||
175 | p = c->old_idx.rb_node; | |
176 | while (p) { | |
177 | o = rb_entry(p, struct ubifs_old_idx, rb); | |
178 | if (lnum < o->lnum) | |
179 | p = p->rb_left; | |
180 | else if (lnum > o->lnum) | |
181 | p = p->rb_right; | |
182 | else if (offs < o->offs) | |
183 | p = p->rb_left; | |
184 | else if (offs > o->offs) | |
185 | p = p->rb_right; | |
186 | else | |
187 | return 1; | |
188 | } | |
189 | return 0; | |
190 | } | |
191 | ||
192 | /** | |
193 | * is_idx_node_in_use - determine if an index node can be overwritten. | |
194 | * @c: UBIFS file-system description object | |
195 | * @key: key of index node | |
196 | * @level: index node level | |
197 | * @lnum: LEB number of index node | |
198 | * @offs: offset of index node | |
199 | * | |
200 | * If @key / @lnum / @offs identify an index node that was not part of the old | |
201 | * index, then this function returns %0 (obsolete). Else if the index node was | |
202 | * part of the old index but is now dirty %1 is returned, else if it is clean %2 | |
203 | * is returned. A negative error code is returned on failure. | |
204 | */ | |
205 | static int is_idx_node_in_use(struct ubifs_info *c, union ubifs_key *key, | |
206 | int level, int lnum, int offs) | |
207 | { | |
208 | int ret; | |
209 | ||
210 | ret = is_idx_node_in_tnc(c, key, level, lnum, offs); | |
211 | if (ret < 0) | |
212 | return ret; /* Error code */ | |
213 | if (ret == 0) | |
214 | if (find_old_idx(c, lnum, offs)) | |
215 | return 1; | |
216 | return ret; | |
217 | } | |
218 | ||
219 | /** | |
220 | * layout_leb_in_gaps - layout index nodes using in-the-gaps method. | |
221 | * @c: UBIFS file-system description object | |
222 | * @p: return LEB number here | |
223 | * | |
224 | * This function lays out new index nodes for dirty znodes using in-the-gaps | |
225 | * method of TNC commit. | |
226 | * This function merely puts the next znode into the next gap, making no attempt | |
227 | * to try to maximise the number of znodes that fit. | |
228 | * This function returns the number of index nodes written into the gaps, or a | |
229 | * negative error code on failure. | |
230 | */ | |
231 | static int layout_leb_in_gaps(struct ubifs_info *c, int *p) | |
232 | { | |
233 | struct ubifs_scan_leb *sleb; | |
234 | struct ubifs_scan_node *snod; | |
235 | int lnum, dirt = 0, gap_start, gap_end, err, written, tot_written; | |
236 | ||
237 | tot_written = 0; | |
238 | /* Get an index LEB with lots of obsolete index nodes */ | |
239 | lnum = ubifs_find_dirty_idx_leb(c); | |
240 | if (lnum < 0) | |
241 | /* | |
242 | * There also may be dirt in the index head that could be | |
243 | * filled, however we do not check there at present. | |
244 | */ | |
245 | return lnum; /* Error code */ | |
246 | *p = lnum; | |
247 | dbg_gc("LEB %d", lnum); | |
248 | /* | |
249 | * Scan the index LEB. We use the generic scan for this even though | |
250 | * it is more comprehensive and less efficient than is needed for this | |
251 | * purpose. | |
252 | */ | |
348709ba | 253 | sleb = ubifs_scan(c, lnum, 0, c->ileb_buf, 0); |
1e51764a AB |
254 | c->ileb_len = 0; |
255 | if (IS_ERR(sleb)) | |
256 | return PTR_ERR(sleb); | |
257 | gap_start = 0; | |
258 | list_for_each_entry(snod, &sleb->nodes, list) { | |
259 | struct ubifs_idx_node *idx; | |
260 | int in_use, level; | |
261 | ||
262 | ubifs_assert(snod->type == UBIFS_IDX_NODE); | |
263 | idx = snod->node; | |
264 | key_read(c, ubifs_idx_key(c, idx), &snod->key); | |
265 | level = le16_to_cpu(idx->level); | |
266 | /* Determine if the index node is in use (not obsolete) */ | |
267 | in_use = is_idx_node_in_use(c, &snod->key, level, lnum, | |
268 | snod->offs); | |
269 | if (in_use < 0) { | |
270 | ubifs_scan_destroy(sleb); | |
271 | return in_use; /* Error code */ | |
272 | } | |
273 | if (in_use) { | |
274 | if (in_use == 1) | |
275 | dirt += ALIGN(snod->len, 8); | |
276 | /* | |
277 | * The obsolete index nodes form gaps that can be | |
278 | * overwritten. This gap has ended because we have | |
279 | * found an index node that is still in use | |
280 | * i.e. not obsolete | |
281 | */ | |
282 | gap_end = snod->offs; | |
283 | /* Try to fill gap */ | |
284 | written = fill_gap(c, lnum, gap_start, gap_end, &dirt); | |
285 | if (written < 0) { | |
286 | ubifs_scan_destroy(sleb); | |
287 | return written; /* Error code */ | |
288 | } | |
289 | tot_written += written; | |
290 | gap_start = ALIGN(snod->offs + snod->len, 8); | |
291 | } | |
292 | } | |
293 | ubifs_scan_destroy(sleb); | |
294 | c->ileb_len = c->leb_size; | |
295 | gap_end = c->leb_size; | |
296 | /* Try to fill gap */ | |
297 | written = fill_gap(c, lnum, gap_start, gap_end, &dirt); | |
298 | if (written < 0) | |
299 | return written; /* Error code */ | |
300 | tot_written += written; | |
301 | if (tot_written == 0) { | |
302 | struct ubifs_lprops lp; | |
303 | ||
304 | dbg_gc("LEB %d wrote %d index nodes", lnum, tot_written); | |
305 | err = ubifs_read_one_lp(c, lnum, &lp); | |
306 | if (err) | |
307 | return err; | |
308 | if (lp.free == c->leb_size) { | |
309 | /* | |
310 | * We must have snatched this LEB from the idx_gc list | |
311 | * so we need to correct the free and dirty space. | |
312 | */ | |
313 | err = ubifs_change_one_lp(c, lnum, | |
314 | c->leb_size - c->ileb_len, | |
315 | dirt, 0, 0, 0); | |
316 | if (err) | |
317 | return err; | |
318 | } | |
319 | return 0; | |
320 | } | |
321 | err = ubifs_change_one_lp(c, lnum, c->leb_size - c->ileb_len, dirt, | |
322 | 0, 0, 0); | |
323 | if (err) | |
324 | return err; | |
b36a261e | 325 | err = ubifs_leb_change(c, lnum, c->ileb_buf, c->ileb_len); |
1e51764a AB |
326 | if (err) |
327 | return err; | |
328 | dbg_gc("LEB %d wrote %d index nodes", lnum, tot_written); | |
329 | return tot_written; | |
330 | } | |
331 | ||
332 | /** | |
333 | * get_leb_cnt - calculate the number of empty LEBs needed to commit. | |
334 | * @c: UBIFS file-system description object | |
335 | * @cnt: number of znodes to commit | |
336 | * | |
337 | * This function returns the number of empty LEBs needed to commit @cnt znodes | |
338 | * to the current index head. The number is not exact and may be more than | |
339 | * needed. | |
340 | */ | |
341 | static int get_leb_cnt(struct ubifs_info *c, int cnt) | |
342 | { | |
343 | int d; | |
344 | ||
345 | /* Assume maximum index node size (i.e. overestimate space needed) */ | |
346 | cnt -= (c->leb_size - c->ihead_offs) / c->max_idx_node_sz; | |
347 | if (cnt < 0) | |
348 | cnt = 0; | |
349 | d = c->leb_size / c->max_idx_node_sz; | |
350 | return DIV_ROUND_UP(cnt, d); | |
351 | } | |
352 | ||
353 | /** | |
354 | * layout_in_gaps - in-the-gaps method of committing TNC. | |
355 | * @c: UBIFS file-system description object | |
356 | * @cnt: number of dirty znodes to commit. | |
357 | * | |
358 | * This function lays out new index nodes for dirty znodes using in-the-gaps | |
359 | * method of TNC commit. | |
360 | * | |
361 | * This function returns %0 on success and a negative error code on failure. | |
362 | */ | |
363 | static int layout_in_gaps(struct ubifs_info *c, int cnt) | |
364 | { | |
365 | int err, leb_needed_cnt, written, *p; | |
366 | ||
367 | dbg_gc("%d znodes to write", cnt); | |
368 | ||
6da2ec56 KC |
369 | c->gap_lebs = kmalloc_array(c->lst.idx_lebs + 1, sizeof(int), |
370 | GFP_NOFS); | |
1e51764a AB |
371 | if (!c->gap_lebs) |
372 | return -ENOMEM; | |
373 | ||
374 | p = c->gap_lebs; | |
375 | do { | |
c0082e98 | 376 | ubifs_assert(p < c->gap_lebs + c->lst.idx_lebs); |
1e51764a AB |
377 | written = layout_leb_in_gaps(c, p); |
378 | if (written < 0) { | |
379 | err = written; | |
0010f18a AB |
380 | if (err != -ENOSPC) { |
381 | kfree(c->gap_lebs); | |
382 | c->gap_lebs = NULL; | |
383 | return err; | |
1e51764a | 384 | } |
8d7819b4 | 385 | if (!dbg_is_chk_index(c)) { |
0010f18a AB |
386 | /* |
387 | * Do not print scary warnings if the debugging | |
388 | * option which forces in-the-gaps is enabled. | |
389 | */ | |
235c362b | 390 | ubifs_warn(c, "out of space"); |
edf6be24 AB |
391 | ubifs_dump_budg(c, &c->bi); |
392 | ubifs_dump_lprops(c); | |
0010f18a AB |
393 | } |
394 | /* Try to commit anyway */ | |
0010f18a | 395 | break; |
1e51764a AB |
396 | } |
397 | p++; | |
398 | cnt -= written; | |
399 | leb_needed_cnt = get_leb_cnt(c, cnt); | |
400 | dbg_gc("%d znodes remaining, need %d LEBs, have %d", cnt, | |
401 | leb_needed_cnt, c->ileb_cnt); | |
402 | } while (leb_needed_cnt > c->ileb_cnt); | |
403 | ||
404 | *p = -1; | |
405 | return 0; | |
406 | } | |
407 | ||
408 | /** | |
409 | * layout_in_empty_space - layout index nodes in empty space. | |
410 | * @c: UBIFS file-system description object | |
411 | * | |
412 | * This function lays out new index nodes for dirty znodes using empty LEBs. | |
413 | * | |
414 | * This function returns %0 on success and a negative error code on failure. | |
415 | */ | |
416 | static int layout_in_empty_space(struct ubifs_info *c) | |
417 | { | |
418 | struct ubifs_znode *znode, *cnext, *zp; | |
419 | int lnum, offs, len, next_len, buf_len, buf_offs, used, avail; | |
420 | int wlen, blen, err; | |
421 | ||
422 | cnext = c->enext; | |
423 | if (!cnext) | |
424 | return 0; | |
425 | ||
426 | lnum = c->ihead_lnum; | |
427 | buf_offs = c->ihead_offs; | |
428 | ||
429 | buf_len = ubifs_idx_node_sz(c, c->fanout); | |
430 | buf_len = ALIGN(buf_len, c->min_io_size); | |
431 | used = 0; | |
432 | avail = buf_len; | |
433 | ||
434 | /* Ensure there is enough room for first write */ | |
435 | next_len = ubifs_idx_node_sz(c, cnext->child_cnt); | |
436 | if (buf_offs + next_len > c->leb_size) | |
437 | lnum = -1; | |
438 | ||
439 | while (1) { | |
440 | znode = cnext; | |
441 | ||
442 | len = ubifs_idx_node_sz(c, znode->child_cnt); | |
443 | ||
444 | /* Determine the index node position */ | |
445 | if (lnum == -1) { | |
446 | if (c->ileb_nxt >= c->ileb_cnt) { | |
235c362b | 447 | ubifs_err(c, "out of space"); |
1e51764a AB |
448 | return -ENOSPC; |
449 | } | |
450 | lnum = c->ilebs[c->ileb_nxt++]; | |
451 | buf_offs = 0; | |
452 | used = 0; | |
453 | avail = buf_len; | |
454 | } | |
455 | ||
456 | offs = buf_offs + used; | |
457 | ||
1e51764a AB |
458 | znode->lnum = lnum; |
459 | znode->offs = offs; | |
460 | znode->len = len; | |
1e51764a AB |
461 | |
462 | /* Update the parent */ | |
463 | zp = znode->parent; | |
464 | if (zp) { | |
465 | struct ubifs_zbranch *zbr; | |
466 | int i; | |
467 | ||
468 | i = znode->iip; | |
469 | zbr = &zp->zbranch[i]; | |
470 | zbr->lnum = lnum; | |
471 | zbr->offs = offs; | |
472 | zbr->len = len; | |
473 | } else { | |
474 | c->zroot.lnum = lnum; | |
475 | c->zroot.offs = offs; | |
476 | c->zroot.len = len; | |
477 | } | |
478 | c->calc_idx_sz += ALIGN(len, 8); | |
479 | ||
480 | /* | |
481 | * Once lprops is updated, we can decrease the dirty znode count | |
482 | * but it is easier to just do it here. | |
483 | */ | |
484 | atomic_long_dec(&c->dirty_zn_cnt); | |
485 | ||
486 | /* | |
487 | * Calculate the next index node length to see if there is | |
488 | * enough room for it | |
489 | */ | |
490 | cnext = znode->cnext; | |
491 | if (cnext == c->cnext) | |
492 | next_len = 0; | |
493 | else | |
494 | next_len = ubifs_idx_node_sz(c, cnext->child_cnt); | |
495 | ||
1e51764a AB |
496 | /* Update buffer positions */ |
497 | wlen = used + len; | |
498 | used += ALIGN(len, 8); | |
499 | avail -= ALIGN(len, 8); | |
500 | ||
501 | if (next_len != 0 && | |
502 | buf_offs + used + next_len <= c->leb_size && | |
503 | avail > 0) | |
504 | continue; | |
505 | ||
506 | if (avail <= 0 && next_len && | |
507 | buf_offs + used + next_len <= c->leb_size) | |
508 | blen = buf_len; | |
509 | else | |
510 | blen = ALIGN(wlen, c->min_io_size); | |
511 | ||
512 | /* The buffer is full or there are no more znodes to do */ | |
513 | buf_offs += blen; | |
514 | if (next_len) { | |
515 | if (buf_offs + next_len > c->leb_size) { | |
516 | err = ubifs_update_one_lp(c, lnum, | |
517 | c->leb_size - buf_offs, blen - used, | |
518 | 0, 0); | |
519 | if (err) | |
520 | return err; | |
521 | lnum = -1; | |
522 | } | |
523 | used -= blen; | |
524 | if (used < 0) | |
525 | used = 0; | |
526 | avail = buf_len - used; | |
527 | continue; | |
528 | } | |
529 | err = ubifs_update_one_lp(c, lnum, c->leb_size - buf_offs, | |
530 | blen - used, 0, 0); | |
531 | if (err) | |
532 | return err; | |
533 | break; | |
534 | } | |
535 | ||
17c2f9f8 AB |
536 | c->dbg->new_ihead_lnum = lnum; |
537 | c->dbg->new_ihead_offs = buf_offs; | |
1e51764a AB |
538 | |
539 | return 0; | |
540 | } | |
541 | ||
542 | /** | |
543 | * layout_commit - determine positions of index nodes to commit. | |
544 | * @c: UBIFS file-system description object | |
545 | * @no_space: indicates that insufficient empty LEBs were allocated | |
546 | * @cnt: number of znodes to commit | |
547 | * | |
548 | * Calculate and update the positions of index nodes to commit. If there were | |
549 | * an insufficient number of empty LEBs allocated, then index nodes are placed | |
550 | * into the gaps created by obsolete index nodes in non-empty index LEBs. For | |
551 | * this purpose, an obsolete index node is one that was not in the index as at | |
552 | * the end of the last commit. To write "in-the-gaps" requires that those index | |
553 | * LEBs are updated atomically in-place. | |
554 | */ | |
555 | static int layout_commit(struct ubifs_info *c, int no_space, int cnt) | |
556 | { | |
557 | int err; | |
558 | ||
559 | if (no_space) { | |
560 | err = layout_in_gaps(c, cnt); | |
561 | if (err) | |
562 | return err; | |
563 | } | |
564 | err = layout_in_empty_space(c); | |
565 | return err; | |
566 | } | |
567 | ||
568 | /** | |
569 | * find_first_dirty - find first dirty znode. | |
570 | * @znode: znode to begin searching from | |
571 | */ | |
572 | static struct ubifs_znode *find_first_dirty(struct ubifs_znode *znode) | |
573 | { | |
574 | int i, cont; | |
575 | ||
576 | if (!znode) | |
577 | return NULL; | |
578 | ||
579 | while (1) { | |
580 | if (znode->level == 0) { | |
581 | if (ubifs_zn_dirty(znode)) | |
582 | return znode; | |
583 | return NULL; | |
584 | } | |
585 | cont = 0; | |
586 | for (i = 0; i < znode->child_cnt; i++) { | |
587 | struct ubifs_zbranch *zbr = &znode->zbranch[i]; | |
588 | ||
589 | if (zbr->znode && ubifs_zn_dirty(zbr->znode)) { | |
590 | znode = zbr->znode; | |
591 | cont = 1; | |
592 | break; | |
593 | } | |
594 | } | |
595 | if (!cont) { | |
596 | if (ubifs_zn_dirty(znode)) | |
597 | return znode; | |
598 | return NULL; | |
599 | } | |
600 | } | |
601 | } | |
602 | ||
603 | /** | |
604 | * find_next_dirty - find next dirty znode. | |
605 | * @znode: znode to begin searching from | |
606 | */ | |
607 | static struct ubifs_znode *find_next_dirty(struct ubifs_znode *znode) | |
608 | { | |
609 | int n = znode->iip + 1; | |
610 | ||
611 | znode = znode->parent; | |
612 | if (!znode) | |
613 | return NULL; | |
614 | for (; n < znode->child_cnt; n++) { | |
615 | struct ubifs_zbranch *zbr = &znode->zbranch[n]; | |
616 | ||
617 | if (zbr->znode && ubifs_zn_dirty(zbr->znode)) | |
618 | return find_first_dirty(zbr->znode); | |
619 | } | |
620 | return znode; | |
621 | } | |
622 | ||
623 | /** | |
624 | * get_znodes_to_commit - create list of dirty znodes to commit. | |
625 | * @c: UBIFS file-system description object | |
626 | * | |
627 | * This function returns the number of znodes to commit. | |
628 | */ | |
629 | static int get_znodes_to_commit(struct ubifs_info *c) | |
630 | { | |
631 | struct ubifs_znode *znode, *cnext; | |
632 | int cnt = 0; | |
633 | ||
634 | c->cnext = find_first_dirty(c->zroot.znode); | |
635 | znode = c->enext = c->cnext; | |
636 | if (!znode) { | |
637 | dbg_cmt("no znodes to commit"); | |
638 | return 0; | |
639 | } | |
640 | cnt += 1; | |
641 | while (1) { | |
f42eed7c | 642 | ubifs_assert(!ubifs_zn_cow(znode)); |
1e51764a AB |
643 | __set_bit(COW_ZNODE, &znode->flags); |
644 | znode->alt = 0; | |
645 | cnext = find_next_dirty(znode); | |
646 | if (!cnext) { | |
647 | znode->cnext = c->cnext; | |
648 | break; | |
649 | } | |
650 | znode->cnext = cnext; | |
651 | znode = cnext; | |
652 | cnt += 1; | |
653 | } | |
654 | dbg_cmt("committing %d znodes", cnt); | |
655 | ubifs_assert(cnt == atomic_long_read(&c->dirty_zn_cnt)); | |
656 | return cnt; | |
657 | } | |
658 | ||
659 | /** | |
660 | * alloc_idx_lebs - allocate empty LEBs to be used to commit. | |
661 | * @c: UBIFS file-system description object | |
662 | * @cnt: number of znodes to commit | |
663 | * | |
664 | * This function returns %-ENOSPC if it cannot allocate a sufficient number of | |
665 | * empty LEBs. %0 is returned on success, otherwise a negative error code | |
666 | * is returned. | |
667 | */ | |
668 | static int alloc_idx_lebs(struct ubifs_info *c, int cnt) | |
669 | { | |
670 | int i, leb_cnt, lnum; | |
671 | ||
672 | c->ileb_cnt = 0; | |
673 | c->ileb_nxt = 0; | |
674 | leb_cnt = get_leb_cnt(c, cnt); | |
675 | dbg_cmt("need about %d empty LEBS for TNC commit", leb_cnt); | |
676 | if (!leb_cnt) | |
677 | return 0; | |
6da2ec56 | 678 | c->ilebs = kmalloc_array(leb_cnt, sizeof(int), GFP_NOFS); |
1e51764a AB |
679 | if (!c->ilebs) |
680 | return -ENOMEM; | |
681 | for (i = 0; i < leb_cnt; i++) { | |
682 | lnum = ubifs_find_free_leb_for_idx(c); | |
683 | if (lnum < 0) | |
684 | return lnum; | |
685 | c->ilebs[c->ileb_cnt++] = lnum; | |
686 | dbg_cmt("LEB %d", lnum); | |
687 | } | |
3d251a5b | 688 | if (dbg_is_chk_index(c) && !(prandom_u32() & 7)) |
1e51764a AB |
689 | return -ENOSPC; |
690 | return 0; | |
691 | } | |
692 | ||
693 | /** | |
694 | * free_unused_idx_lebs - free unused LEBs that were allocated for the commit. | |
695 | * @c: UBIFS file-system description object | |
696 | * | |
697 | * It is possible that we allocate more empty LEBs for the commit than we need. | |
698 | * This functions frees the surplus. | |
699 | * | |
700 | * This function returns %0 on success and a negative error code on failure. | |
701 | */ | |
702 | static int free_unused_idx_lebs(struct ubifs_info *c) | |
703 | { | |
704 | int i, err = 0, lnum, er; | |
705 | ||
706 | for (i = c->ileb_nxt; i < c->ileb_cnt; i++) { | |
707 | lnum = c->ilebs[i]; | |
708 | dbg_cmt("LEB %d", lnum); | |
709 | er = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0, | |
710 | LPROPS_INDEX | LPROPS_TAKEN, 0); | |
711 | if (!err) | |
712 | err = er; | |
713 | } | |
714 | return err; | |
715 | } | |
716 | ||
717 | /** | |
718 | * free_idx_lebs - free unused LEBs after commit end. | |
719 | * @c: UBIFS file-system description object | |
720 | * | |
721 | * This function returns %0 on success and a negative error code on failure. | |
722 | */ | |
723 | static int free_idx_lebs(struct ubifs_info *c) | |
724 | { | |
725 | int err; | |
726 | ||
727 | err = free_unused_idx_lebs(c); | |
728 | kfree(c->ilebs); | |
729 | c->ilebs = NULL; | |
730 | return err; | |
731 | } | |
732 | ||
733 | /** | |
734 | * ubifs_tnc_start_commit - start TNC commit. | |
735 | * @c: UBIFS file-system description object | |
736 | * @zroot: new index root position is returned here | |
737 | * | |
738 | * This function prepares the list of indexing nodes to commit and lays out | |
739 | * their positions on flash. If there is not enough free space it uses the | |
740 | * in-gap commit method. Returns zero in case of success and a negative error | |
741 | * code in case of failure. | |
742 | */ | |
743 | int ubifs_tnc_start_commit(struct ubifs_info *c, struct ubifs_zbranch *zroot) | |
744 | { | |
745 | int err = 0, cnt; | |
746 | ||
747 | mutex_lock(&c->tnc_mutex); | |
748 | err = dbg_check_tnc(c, 1); | |
749 | if (err) | |
750 | goto out; | |
751 | cnt = get_znodes_to_commit(c); | |
752 | if (cnt != 0) { | |
753 | int no_space = 0; | |
754 | ||
755 | err = alloc_idx_lebs(c, cnt); | |
756 | if (err == -ENOSPC) | |
757 | no_space = 1; | |
758 | else if (err) | |
759 | goto out_free; | |
760 | err = layout_commit(c, no_space, cnt); | |
761 | if (err) | |
762 | goto out_free; | |
763 | ubifs_assert(atomic_long_read(&c->dirty_zn_cnt) == 0); | |
764 | err = free_unused_idx_lebs(c); | |
765 | if (err) | |
766 | goto out; | |
767 | } | |
768 | destroy_old_idx(c); | |
769 | memcpy(zroot, &c->zroot, sizeof(struct ubifs_zbranch)); | |
770 | ||
771 | err = ubifs_save_dirty_idx_lnums(c); | |
772 | if (err) | |
773 | goto out; | |
774 | ||
775 | spin_lock(&c->space_lock); | |
776 | /* | |
777 | * Although we have not finished committing yet, update size of the | |
b137545c | 778 | * committed index ('c->bi.old_idx_sz') and zero out the index growth |
1e51764a AB |
779 | * budget. It is OK to do this now, because we've reserved all the |
780 | * space which is needed to commit the index, and it is save for the | |
781 | * budgeting subsystem to assume the index is already committed, | |
782 | * even though it is not. | |
783 | */ | |
b137545c AB |
784 | ubifs_assert(c->bi.min_idx_lebs == ubifs_calc_min_idx_lebs(c)); |
785 | c->bi.old_idx_sz = c->calc_idx_sz; | |
786 | c->bi.uncommitted_idx = 0; | |
787 | c->bi.min_idx_lebs = ubifs_calc_min_idx_lebs(c); | |
1e51764a AB |
788 | spin_unlock(&c->space_lock); |
789 | mutex_unlock(&c->tnc_mutex); | |
790 | ||
791 | dbg_cmt("number of index LEBs %d", c->lst.idx_lebs); | |
792 | dbg_cmt("size of index %llu", c->calc_idx_sz); | |
793 | return err; | |
794 | ||
795 | out_free: | |
796 | free_idx_lebs(c); | |
797 | out: | |
798 | mutex_unlock(&c->tnc_mutex); | |
799 | return err; | |
800 | } | |
801 | ||
802 | /** | |
803 | * write_index - write index nodes. | |
804 | * @c: UBIFS file-system description object | |
805 | * | |
806 | * This function writes the index nodes whose positions were laid out in the | |
807 | * layout_in_empty_space function. | |
808 | */ | |
809 | static int write_index(struct ubifs_info *c) | |
810 | { | |
811 | struct ubifs_idx_node *idx; | |
812 | struct ubifs_znode *znode, *cnext; | |
813 | int i, lnum, offs, len, next_len, buf_len, buf_offs, used; | |
1f42596e | 814 | int avail, wlen, err, lnum_pos = 0, blen, nxt_offs; |
1e51764a AB |
815 | |
816 | cnext = c->enext; | |
817 | if (!cnext) | |
818 | return 0; | |
819 | ||
820 | /* | |
821 | * Always write index nodes to the index head so that index nodes and | |
822 | * other types of nodes are never mixed in the same erase block. | |
823 | */ | |
824 | lnum = c->ihead_lnum; | |
825 | buf_offs = c->ihead_offs; | |
826 | ||
827 | /* Allocate commit buffer */ | |
828 | buf_len = ALIGN(c->max_idx_node_sz, c->min_io_size); | |
829 | used = 0; | |
830 | avail = buf_len; | |
831 | ||
832 | /* Ensure there is enough room for first write */ | |
833 | next_len = ubifs_idx_node_sz(c, cnext->child_cnt); | |
834 | if (buf_offs + next_len > c->leb_size) { | |
835 | err = ubifs_update_one_lp(c, lnum, LPROPS_NC, 0, 0, | |
836 | LPROPS_TAKEN); | |
837 | if (err) | |
838 | return err; | |
839 | lnum = -1; | |
840 | } | |
841 | ||
842 | while (1) { | |
843 | cond_resched(); | |
844 | ||
845 | znode = cnext; | |
846 | idx = c->cbuf + used; | |
847 | ||
848 | /* Make index node */ | |
849 | idx->ch.node_type = UBIFS_IDX_NODE; | |
850 | idx->child_cnt = cpu_to_le16(znode->child_cnt); | |
851 | idx->level = cpu_to_le16(znode->level); | |
852 | for (i = 0; i < znode->child_cnt; i++) { | |
853 | struct ubifs_branch *br = ubifs_idx_branch(c, idx, i); | |
854 | struct ubifs_zbranch *zbr = &znode->zbranch[i]; | |
855 | ||
856 | key_write_idx(c, &zbr->key, &br->key); | |
857 | br->lnum = cpu_to_le32(zbr->lnum); | |
858 | br->offs = cpu_to_le32(zbr->offs); | |
859 | br->len = cpu_to_le32(zbr->len); | |
860 | if (!zbr->lnum || !zbr->len) { | |
235c362b | 861 | ubifs_err(c, "bad ref in znode"); |
edf6be24 | 862 | ubifs_dump_znode(c, znode); |
1e51764a | 863 | if (zbr->znode) |
edf6be24 | 864 | ubifs_dump_znode(c, zbr->znode); |
df71b091 RW |
865 | |
866 | return -EINVAL; | |
1e51764a AB |
867 | } |
868 | } | |
869 | len = ubifs_idx_node_sz(c, znode->child_cnt); | |
870 | ubifs_prepare_node(c, idx, len, 0); | |
871 | ||
872 | /* Determine the index node position */ | |
873 | if (lnum == -1) { | |
874 | lnum = c->ilebs[lnum_pos++]; | |
875 | buf_offs = 0; | |
876 | used = 0; | |
877 | avail = buf_len; | |
878 | } | |
879 | offs = buf_offs + used; | |
880 | ||
1e51764a AB |
881 | if (lnum != znode->lnum || offs != znode->offs || |
882 | len != znode->len) { | |
235c362b | 883 | ubifs_err(c, "inconsistent znode posn"); |
1e51764a AB |
884 | return -EINVAL; |
885 | } | |
1e51764a AB |
886 | |
887 | /* Grab some stuff from znode while we still can */ | |
888 | cnext = znode->cnext; | |
889 | ||
890 | ubifs_assert(ubifs_zn_dirty(znode)); | |
f42eed7c | 891 | ubifs_assert(ubifs_zn_cow(znode)); |
1e51764a AB |
892 | |
893 | /* | |
894 | * It is important that other threads should see %DIRTY_ZNODE | |
895 | * flag cleared before %COW_ZNODE. Specifically, it matters in | |
896 | * the 'dirty_cow_znode()' function. This is the reason for the | |
897 | * first barrier. Also, we want the bit changes to be seen to | |
898 | * other threads ASAP, to avoid unnecesarry copying, which is | |
899 | * the reason for the second barrier. | |
900 | */ | |
901 | clear_bit(DIRTY_ZNODE, &znode->flags); | |
4e857c58 | 902 | smp_mb__before_atomic(); |
1e51764a | 903 | clear_bit(COW_ZNODE, &znode->flags); |
4e857c58 | 904 | smp_mb__after_atomic(); |
1e51764a | 905 | |
06b282a4 AB |
906 | /* |
907 | * We have marked the znode as clean but have not updated the | |
908 | * @c->clean_zn_cnt counter. If this znode becomes dirty again | |
909 | * before 'free_obsolete_znodes()' is called, then | |
910 | * @c->clean_zn_cnt will be decremented before it gets | |
911 | * incremented (resulting in 2 decrements for the same znode). | |
912 | * This means that @c->clean_zn_cnt may become negative for a | |
913 | * while. | |
914 | * | |
915 | * Q: why we cannot increment @c->clean_zn_cnt? | |
916 | * A: because we do not have the @c->tnc_mutex locked, and the | |
917 | * following code would be racy and buggy: | |
918 | * | |
919 | * if (!ubifs_zn_obsolete(znode)) { | |
920 | * atomic_long_inc(&c->clean_zn_cnt); | |
921 | * atomic_long_inc(&ubifs_clean_zn_cnt); | |
922 | * } | |
923 | * | |
924 | * Thus, we just delay the @c->clean_zn_cnt update until we | |
925 | * have the mutex locked. | |
926 | */ | |
927 | ||
1e51764a AB |
928 | /* Do not access znode from this point on */ |
929 | ||
930 | /* Update buffer positions */ | |
931 | wlen = used + len; | |
932 | used += ALIGN(len, 8); | |
933 | avail -= ALIGN(len, 8); | |
934 | ||
935 | /* | |
936 | * Calculate the next index node length to see if there is | |
937 | * enough room for it | |
938 | */ | |
939 | if (cnext == c->cnext) | |
940 | next_len = 0; | |
941 | else | |
942 | next_len = ubifs_idx_node_sz(c, cnext->child_cnt); | |
943 | ||
1f42596e AB |
944 | nxt_offs = buf_offs + used + next_len; |
945 | if (next_len && nxt_offs <= c->leb_size) { | |
946 | if (avail > 0) | |
1e51764a | 947 | continue; |
1f42596e AB |
948 | else |
949 | blen = buf_len; | |
1e51764a | 950 | } else { |
1f42596e AB |
951 | wlen = ALIGN(wlen, 8); |
952 | blen = ALIGN(wlen, c->min_io_size); | |
953 | ubifs_pad(c, c->cbuf + wlen, blen - wlen); | |
954 | } | |
955 | ||
956 | /* The buffer is full or there are no more znodes to do */ | |
b36a261e | 957 | err = ubifs_leb_write(c, lnum, c->cbuf, buf_offs, blen); |
1f42596e AB |
958 | if (err) |
959 | return err; | |
960 | buf_offs += blen; | |
961 | if (next_len) { | |
962 | if (nxt_offs > c->leb_size) { | |
963 | err = ubifs_update_one_lp(c, lnum, LPROPS_NC, 0, | |
964 | 0, LPROPS_TAKEN); | |
965 | if (err) | |
966 | return err; | |
967 | lnum = -1; | |
1e51764a | 968 | } |
1f42596e AB |
969 | used -= blen; |
970 | if (used < 0) | |
971 | used = 0; | |
972 | avail = buf_len - used; | |
973 | memmove(c->cbuf, c->cbuf + blen, used); | |
974 | continue; | |
1e51764a AB |
975 | } |
976 | break; | |
977 | } | |
978 | ||
17c2f9f8 AB |
979 | if (lnum != c->dbg->new_ihead_lnum || |
980 | buf_offs != c->dbg->new_ihead_offs) { | |
235c362b | 981 | ubifs_err(c, "inconsistent ihead"); |
1e51764a AB |
982 | return -EINVAL; |
983 | } | |
1e51764a AB |
984 | |
985 | c->ihead_lnum = lnum; | |
986 | c->ihead_offs = buf_offs; | |
987 | ||
988 | return 0; | |
989 | } | |
990 | ||
991 | /** | |
992 | * free_obsolete_znodes - free obsolete znodes. | |
993 | * @c: UBIFS file-system description object | |
994 | * | |
995 | * At the end of commit end, obsolete znodes are freed. | |
996 | */ | |
997 | static void free_obsolete_znodes(struct ubifs_info *c) | |
998 | { | |
999 | struct ubifs_znode *znode, *cnext; | |
1000 | ||
1001 | cnext = c->cnext; | |
1002 | do { | |
1003 | znode = cnext; | |
1004 | cnext = znode->cnext; | |
f42eed7c | 1005 | if (ubifs_zn_obsolete(znode)) |
1e51764a AB |
1006 | kfree(znode); |
1007 | else { | |
1008 | znode->cnext = NULL; | |
1009 | atomic_long_inc(&c->clean_zn_cnt); | |
1010 | atomic_long_inc(&ubifs_clean_zn_cnt); | |
1011 | } | |
1012 | } while (cnext != c->cnext); | |
1013 | } | |
1014 | ||
1015 | /** | |
1016 | * return_gap_lebs - return LEBs used by the in-gap commit method. | |
1017 | * @c: UBIFS file-system description object | |
1018 | * | |
1019 | * This function clears the "taken" flag for the LEBs which were used by the | |
1020 | * "commit in-the-gaps" method. | |
1021 | */ | |
1022 | static int return_gap_lebs(struct ubifs_info *c) | |
1023 | { | |
1024 | int *p, err; | |
1025 | ||
1026 | if (!c->gap_lebs) | |
1027 | return 0; | |
1028 | ||
1029 | dbg_cmt(""); | |
1030 | for (p = c->gap_lebs; *p != -1; p++) { | |
1031 | err = ubifs_change_one_lp(c, *p, LPROPS_NC, LPROPS_NC, 0, | |
1032 | LPROPS_TAKEN, 0); | |
1033 | if (err) | |
1034 | return err; | |
1035 | } | |
1036 | ||
1037 | kfree(c->gap_lebs); | |
1038 | c->gap_lebs = NULL; | |
1039 | return 0; | |
1040 | } | |
1041 | ||
1042 | /** | |
1043 | * ubifs_tnc_end_commit - update the TNC for commit end. | |
1044 | * @c: UBIFS file-system description object | |
1045 | * | |
1046 | * Write the dirty znodes. | |
1047 | */ | |
1048 | int ubifs_tnc_end_commit(struct ubifs_info *c) | |
1049 | { | |
1050 | int err; | |
1051 | ||
1052 | if (!c->cnext) | |
1053 | return 0; | |
1054 | ||
1055 | err = return_gap_lebs(c); | |
1056 | if (err) | |
1057 | return err; | |
1058 | ||
1059 | err = write_index(c); | |
1060 | if (err) | |
1061 | return err; | |
1062 | ||
1063 | mutex_lock(&c->tnc_mutex); | |
1064 | ||
1065 | dbg_cmt("TNC height is %d", c->zroot.znode->level + 1); | |
1066 | ||
1067 | free_obsolete_znodes(c); | |
1068 | ||
1069 | c->cnext = NULL; | |
1070 | kfree(c->ilebs); | |
1071 | c->ilebs = NULL; | |
1072 | ||
1073 | mutex_unlock(&c->tnc_mutex); | |
1074 | ||
1075 | return 0; | |
1076 | } |