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; | |
16a26b20 | 41 | u8 hash[UBIFS_HASH_ARR_SZ]; |
1e51764a AB |
42 | int i, err; |
43 | ||
44 | /* Make index node */ | |
45 | idx->ch.node_type = UBIFS_IDX_NODE; | |
46 | idx->child_cnt = cpu_to_le16(znode->child_cnt); | |
47 | idx->level = cpu_to_le16(znode->level); | |
48 | for (i = 0; i < znode->child_cnt; i++) { | |
49 | struct ubifs_branch *br = ubifs_idx_branch(c, idx, i); | |
50 | struct ubifs_zbranch *zbr = &znode->zbranch[i]; | |
51 | ||
52 | key_write_idx(c, &zbr->key, &br->key); | |
53 | br->lnum = cpu_to_le32(zbr->lnum); | |
54 | br->offs = cpu_to_le32(zbr->offs); | |
55 | br->len = cpu_to_le32(zbr->len); | |
16a26b20 | 56 | ubifs_copy_hash(c, zbr->hash, ubifs_branch_hash(c, br)); |
1e51764a | 57 | if (!zbr->lnum || !zbr->len) { |
235c362b | 58 | ubifs_err(c, "bad ref in znode"); |
edf6be24 | 59 | ubifs_dump_znode(c, znode); |
1e51764a | 60 | if (zbr->znode) |
edf6be24 | 61 | ubifs_dump_znode(c, zbr->znode); |
df71b091 RW |
62 | |
63 | return -EINVAL; | |
1e51764a AB |
64 | } |
65 | } | |
66 | ubifs_prepare_node(c, idx, len, 0); | |
16a26b20 | 67 | ubifs_node_calc_hash(c, idx, hash); |
1e51764a | 68 | |
1e51764a AB |
69 | znode->lnum = lnum; |
70 | znode->offs = offs; | |
71 | znode->len = len; | |
1e51764a AB |
72 | |
73 | err = insert_old_idx_znode(c, znode); | |
74 | ||
75 | /* Update the parent */ | |
76 | zp = znode->parent; | |
77 | if (zp) { | |
78 | struct ubifs_zbranch *zbr; | |
79 | ||
80 | zbr = &zp->zbranch[znode->iip]; | |
81 | zbr->lnum = lnum; | |
82 | zbr->offs = offs; | |
83 | zbr->len = len; | |
16a26b20 | 84 | ubifs_copy_hash(c, hash, zbr->hash); |
1e51764a AB |
85 | } else { |
86 | c->zroot.lnum = lnum; | |
87 | c->zroot.offs = offs; | |
88 | c->zroot.len = len; | |
16a26b20 | 89 | ubifs_copy_hash(c, hash, c->zroot.hash); |
1e51764a AB |
90 | } |
91 | c->calc_idx_sz += ALIGN(len, 8); | |
92 | ||
93 | atomic_long_dec(&c->dirty_zn_cnt); | |
94 | ||
6eb61d58 RW |
95 | ubifs_assert(c, ubifs_zn_dirty(znode)); |
96 | ubifs_assert(c, ubifs_zn_cow(znode)); | |
1e51764a | 97 | |
06b282a4 AB |
98 | /* |
99 | * Note, unlike 'write_index()' we do not add memory barriers here | |
100 | * because this function is called with @c->tnc_mutex locked. | |
101 | */ | |
1e51764a AB |
102 | __clear_bit(DIRTY_ZNODE, &znode->flags); |
103 | __clear_bit(COW_ZNODE, &znode->flags); | |
104 | ||
105 | return err; | |
106 | } | |
107 | ||
108 | /** | |
109 | * fill_gap - make index nodes in gaps in dirty index LEBs. | |
110 | * @c: UBIFS file-system description object | |
111 | * @lnum: LEB number that gap appears in | |
112 | * @gap_start: offset of start of gap | |
113 | * @gap_end: offset of end of gap | |
114 | * @dirt: adds dirty space to this | |
115 | * | |
116 | * This function returns the number of index nodes written into the gap. | |
117 | */ | |
118 | static int fill_gap(struct ubifs_info *c, int lnum, int gap_start, int gap_end, | |
119 | int *dirt) | |
120 | { | |
121 | int len, gap_remains, gap_pos, written, pad_len; | |
122 | ||
6eb61d58 RW |
123 | ubifs_assert(c, (gap_start & 7) == 0); |
124 | ubifs_assert(c, (gap_end & 7) == 0); | |
125 | ubifs_assert(c, gap_end >= gap_start); | |
1e51764a AB |
126 | |
127 | gap_remains = gap_end - gap_start; | |
128 | if (!gap_remains) | |
129 | return 0; | |
130 | gap_pos = gap_start; | |
131 | written = 0; | |
132 | while (c->enext) { | |
133 | len = ubifs_idx_node_sz(c, c->enext->child_cnt); | |
134 | if (len < gap_remains) { | |
135 | struct ubifs_znode *znode = c->enext; | |
136 | const int alen = ALIGN(len, 8); | |
137 | int err; | |
138 | ||
6eb61d58 | 139 | ubifs_assert(c, alen <= gap_remains); |
1e51764a AB |
140 | err = make_idx_node(c, c->ileb_buf + gap_pos, znode, |
141 | lnum, gap_pos, len); | |
142 | if (err) | |
143 | return err; | |
144 | gap_remains -= alen; | |
145 | gap_pos += alen; | |
146 | c->enext = znode->cnext; | |
147 | if (c->enext == c->cnext) | |
148 | c->enext = NULL; | |
149 | written += 1; | |
150 | } else | |
151 | break; | |
152 | } | |
153 | if (gap_end == c->leb_size) { | |
154 | c->ileb_len = ALIGN(gap_pos, c->min_io_size); | |
155 | /* Pad to end of min_io_size */ | |
156 | pad_len = c->ileb_len - gap_pos; | |
157 | } else | |
158 | /* Pad to end of gap */ | |
159 | pad_len = gap_remains; | |
160 | dbg_gc("LEB %d:%d to %d len %d nodes written %d wasted bytes %d", | |
161 | lnum, gap_start, gap_end, gap_end - gap_start, written, pad_len); | |
162 | ubifs_pad(c, c->ileb_buf + gap_pos, pad_len); | |
163 | *dirt += pad_len; | |
164 | return written; | |
165 | } | |
166 | ||
167 | /** | |
168 | * find_old_idx - find an index node obsoleted since the last commit start. | |
169 | * @c: UBIFS file-system description object | |
170 | * @lnum: LEB number of obsoleted index node | |
171 | * @offs: offset of obsoleted index node | |
172 | * | |
173 | * Returns %1 if found and %0 otherwise. | |
174 | */ | |
175 | static int find_old_idx(struct ubifs_info *c, int lnum, int offs) | |
176 | { | |
177 | struct ubifs_old_idx *o; | |
178 | struct rb_node *p; | |
179 | ||
180 | p = c->old_idx.rb_node; | |
181 | while (p) { | |
182 | o = rb_entry(p, struct ubifs_old_idx, rb); | |
183 | if (lnum < o->lnum) | |
184 | p = p->rb_left; | |
185 | else if (lnum > o->lnum) | |
186 | p = p->rb_right; | |
187 | else if (offs < o->offs) | |
188 | p = p->rb_left; | |
189 | else if (offs > o->offs) | |
190 | p = p->rb_right; | |
191 | else | |
192 | return 1; | |
193 | } | |
194 | return 0; | |
195 | } | |
196 | ||
197 | /** | |
198 | * is_idx_node_in_use - determine if an index node can be overwritten. | |
199 | * @c: UBIFS file-system description object | |
200 | * @key: key of index node | |
201 | * @level: index node level | |
202 | * @lnum: LEB number of index node | |
203 | * @offs: offset of index node | |
204 | * | |
205 | * If @key / @lnum / @offs identify an index node that was not part of the old | |
206 | * index, then this function returns %0 (obsolete). Else if the index node was | |
207 | * part of the old index but is now dirty %1 is returned, else if it is clean %2 | |
208 | * is returned. A negative error code is returned on failure. | |
209 | */ | |
210 | static int is_idx_node_in_use(struct ubifs_info *c, union ubifs_key *key, | |
211 | int level, int lnum, int offs) | |
212 | { | |
213 | int ret; | |
214 | ||
215 | ret = is_idx_node_in_tnc(c, key, level, lnum, offs); | |
216 | if (ret < 0) | |
217 | return ret; /* Error code */ | |
218 | if (ret == 0) | |
219 | if (find_old_idx(c, lnum, offs)) | |
220 | return 1; | |
221 | return ret; | |
222 | } | |
223 | ||
224 | /** | |
225 | * layout_leb_in_gaps - layout index nodes using in-the-gaps method. | |
226 | * @c: UBIFS file-system description object | |
227 | * @p: return LEB number here | |
228 | * | |
229 | * This function lays out new index nodes for dirty znodes using in-the-gaps | |
230 | * method of TNC commit. | |
231 | * This function merely puts the next znode into the next gap, making no attempt | |
232 | * to try to maximise the number of znodes that fit. | |
233 | * This function returns the number of index nodes written into the gaps, or a | |
234 | * negative error code on failure. | |
235 | */ | |
236 | static int layout_leb_in_gaps(struct ubifs_info *c, int *p) | |
237 | { | |
238 | struct ubifs_scan_leb *sleb; | |
239 | struct ubifs_scan_node *snod; | |
240 | int lnum, dirt = 0, gap_start, gap_end, err, written, tot_written; | |
241 | ||
242 | tot_written = 0; | |
243 | /* Get an index LEB with lots of obsolete index nodes */ | |
244 | lnum = ubifs_find_dirty_idx_leb(c); | |
245 | if (lnum < 0) | |
246 | /* | |
247 | * There also may be dirt in the index head that could be | |
248 | * filled, however we do not check there at present. | |
249 | */ | |
250 | return lnum; /* Error code */ | |
251 | *p = lnum; | |
252 | dbg_gc("LEB %d", lnum); | |
253 | /* | |
254 | * Scan the index LEB. We use the generic scan for this even though | |
255 | * it is more comprehensive and less efficient than is needed for this | |
256 | * purpose. | |
257 | */ | |
348709ba | 258 | sleb = ubifs_scan(c, lnum, 0, c->ileb_buf, 0); |
1e51764a AB |
259 | c->ileb_len = 0; |
260 | if (IS_ERR(sleb)) | |
261 | return PTR_ERR(sleb); | |
262 | gap_start = 0; | |
263 | list_for_each_entry(snod, &sleb->nodes, list) { | |
264 | struct ubifs_idx_node *idx; | |
265 | int in_use, level; | |
266 | ||
6eb61d58 | 267 | ubifs_assert(c, snod->type == UBIFS_IDX_NODE); |
1e51764a AB |
268 | idx = snod->node; |
269 | key_read(c, ubifs_idx_key(c, idx), &snod->key); | |
270 | level = le16_to_cpu(idx->level); | |
271 | /* Determine if the index node is in use (not obsolete) */ | |
272 | in_use = is_idx_node_in_use(c, &snod->key, level, lnum, | |
273 | snod->offs); | |
274 | if (in_use < 0) { | |
275 | ubifs_scan_destroy(sleb); | |
276 | return in_use; /* Error code */ | |
277 | } | |
278 | if (in_use) { | |
279 | if (in_use == 1) | |
280 | dirt += ALIGN(snod->len, 8); | |
281 | /* | |
282 | * The obsolete index nodes form gaps that can be | |
283 | * overwritten. This gap has ended because we have | |
284 | * found an index node that is still in use | |
285 | * i.e. not obsolete | |
286 | */ | |
287 | gap_end = snod->offs; | |
288 | /* Try to fill gap */ | |
289 | written = fill_gap(c, lnum, gap_start, gap_end, &dirt); | |
290 | if (written < 0) { | |
291 | ubifs_scan_destroy(sleb); | |
292 | return written; /* Error code */ | |
293 | } | |
294 | tot_written += written; | |
295 | gap_start = ALIGN(snod->offs + snod->len, 8); | |
296 | } | |
297 | } | |
298 | ubifs_scan_destroy(sleb); | |
299 | c->ileb_len = c->leb_size; | |
300 | gap_end = c->leb_size; | |
301 | /* Try to fill gap */ | |
302 | written = fill_gap(c, lnum, gap_start, gap_end, &dirt); | |
303 | if (written < 0) | |
304 | return written; /* Error code */ | |
305 | tot_written += written; | |
306 | if (tot_written == 0) { | |
307 | struct ubifs_lprops lp; | |
308 | ||
309 | dbg_gc("LEB %d wrote %d index nodes", lnum, tot_written); | |
310 | err = ubifs_read_one_lp(c, lnum, &lp); | |
311 | if (err) | |
312 | return err; | |
313 | if (lp.free == c->leb_size) { | |
314 | /* | |
315 | * We must have snatched this LEB from the idx_gc list | |
316 | * so we need to correct the free and dirty space. | |
317 | */ | |
318 | err = ubifs_change_one_lp(c, lnum, | |
319 | c->leb_size - c->ileb_len, | |
320 | dirt, 0, 0, 0); | |
321 | if (err) | |
322 | return err; | |
323 | } | |
324 | return 0; | |
325 | } | |
326 | err = ubifs_change_one_lp(c, lnum, c->leb_size - c->ileb_len, dirt, | |
327 | 0, 0, 0); | |
328 | if (err) | |
329 | return err; | |
b36a261e | 330 | err = ubifs_leb_change(c, lnum, c->ileb_buf, c->ileb_len); |
1e51764a AB |
331 | if (err) |
332 | return err; | |
333 | dbg_gc("LEB %d wrote %d index nodes", lnum, tot_written); | |
334 | return tot_written; | |
335 | } | |
336 | ||
337 | /** | |
338 | * get_leb_cnt - calculate the number of empty LEBs needed to commit. | |
339 | * @c: UBIFS file-system description object | |
340 | * @cnt: number of znodes to commit | |
341 | * | |
342 | * This function returns the number of empty LEBs needed to commit @cnt znodes | |
343 | * to the current index head. The number is not exact and may be more than | |
344 | * needed. | |
345 | */ | |
346 | static int get_leb_cnt(struct ubifs_info *c, int cnt) | |
347 | { | |
348 | int d; | |
349 | ||
350 | /* Assume maximum index node size (i.e. overestimate space needed) */ | |
351 | cnt -= (c->leb_size - c->ihead_offs) / c->max_idx_node_sz; | |
352 | if (cnt < 0) | |
353 | cnt = 0; | |
354 | d = c->leb_size / c->max_idx_node_sz; | |
355 | return DIV_ROUND_UP(cnt, d); | |
356 | } | |
357 | ||
358 | /** | |
359 | * layout_in_gaps - in-the-gaps method of committing TNC. | |
360 | * @c: UBIFS file-system description object | |
361 | * @cnt: number of dirty znodes to commit. | |
362 | * | |
363 | * This function lays out new index nodes for dirty znodes using in-the-gaps | |
364 | * method of TNC commit. | |
365 | * | |
366 | * This function returns %0 on success and a negative error code on failure. | |
367 | */ | |
368 | static int layout_in_gaps(struct ubifs_info *c, int cnt) | |
369 | { | |
370 | int err, leb_needed_cnt, written, *p; | |
371 | ||
372 | dbg_gc("%d znodes to write", cnt); | |
373 | ||
6da2ec56 KC |
374 | c->gap_lebs = kmalloc_array(c->lst.idx_lebs + 1, sizeof(int), |
375 | GFP_NOFS); | |
1e51764a AB |
376 | if (!c->gap_lebs) |
377 | return -ENOMEM; | |
378 | ||
379 | p = c->gap_lebs; | |
380 | do { | |
6eb61d58 | 381 | ubifs_assert(c, p < c->gap_lebs + c->lst.idx_lebs); |
1e51764a AB |
382 | written = layout_leb_in_gaps(c, p); |
383 | if (written < 0) { | |
384 | err = written; | |
0010f18a AB |
385 | if (err != -ENOSPC) { |
386 | kfree(c->gap_lebs); | |
387 | c->gap_lebs = NULL; | |
388 | return err; | |
1e51764a | 389 | } |
8d7819b4 | 390 | if (!dbg_is_chk_index(c)) { |
0010f18a AB |
391 | /* |
392 | * Do not print scary warnings if the debugging | |
393 | * option which forces in-the-gaps is enabled. | |
394 | */ | |
235c362b | 395 | ubifs_warn(c, "out of space"); |
edf6be24 AB |
396 | ubifs_dump_budg(c, &c->bi); |
397 | ubifs_dump_lprops(c); | |
0010f18a AB |
398 | } |
399 | /* Try to commit anyway */ | |
0010f18a | 400 | break; |
1e51764a AB |
401 | } |
402 | p++; | |
403 | cnt -= written; | |
404 | leb_needed_cnt = get_leb_cnt(c, cnt); | |
405 | dbg_gc("%d znodes remaining, need %d LEBs, have %d", cnt, | |
406 | leb_needed_cnt, c->ileb_cnt); | |
407 | } while (leb_needed_cnt > c->ileb_cnt); | |
408 | ||
409 | *p = -1; | |
410 | return 0; | |
411 | } | |
412 | ||
413 | /** | |
414 | * layout_in_empty_space - layout index nodes in empty space. | |
415 | * @c: UBIFS file-system description object | |
416 | * | |
417 | * This function lays out new index nodes for dirty znodes using empty LEBs. | |
418 | * | |
419 | * This function returns %0 on success and a negative error code on failure. | |
420 | */ | |
421 | static int layout_in_empty_space(struct ubifs_info *c) | |
422 | { | |
423 | struct ubifs_znode *znode, *cnext, *zp; | |
424 | int lnum, offs, len, next_len, buf_len, buf_offs, used, avail; | |
425 | int wlen, blen, err; | |
426 | ||
427 | cnext = c->enext; | |
428 | if (!cnext) | |
429 | return 0; | |
430 | ||
431 | lnum = c->ihead_lnum; | |
432 | buf_offs = c->ihead_offs; | |
433 | ||
434 | buf_len = ubifs_idx_node_sz(c, c->fanout); | |
435 | buf_len = ALIGN(buf_len, c->min_io_size); | |
436 | used = 0; | |
437 | avail = buf_len; | |
438 | ||
439 | /* Ensure there is enough room for first write */ | |
440 | next_len = ubifs_idx_node_sz(c, cnext->child_cnt); | |
441 | if (buf_offs + next_len > c->leb_size) | |
442 | lnum = -1; | |
443 | ||
444 | while (1) { | |
445 | znode = cnext; | |
446 | ||
447 | len = ubifs_idx_node_sz(c, znode->child_cnt); | |
448 | ||
449 | /* Determine the index node position */ | |
450 | if (lnum == -1) { | |
451 | if (c->ileb_nxt >= c->ileb_cnt) { | |
235c362b | 452 | ubifs_err(c, "out of space"); |
1e51764a AB |
453 | return -ENOSPC; |
454 | } | |
455 | lnum = c->ilebs[c->ileb_nxt++]; | |
456 | buf_offs = 0; | |
457 | used = 0; | |
458 | avail = buf_len; | |
459 | } | |
460 | ||
461 | offs = buf_offs + used; | |
462 | ||
1e51764a AB |
463 | znode->lnum = lnum; |
464 | znode->offs = offs; | |
465 | znode->len = len; | |
1e51764a AB |
466 | |
467 | /* Update the parent */ | |
468 | zp = znode->parent; | |
469 | if (zp) { | |
470 | struct ubifs_zbranch *zbr; | |
471 | int i; | |
472 | ||
473 | i = znode->iip; | |
474 | zbr = &zp->zbranch[i]; | |
475 | zbr->lnum = lnum; | |
476 | zbr->offs = offs; | |
477 | zbr->len = len; | |
478 | } else { | |
479 | c->zroot.lnum = lnum; | |
480 | c->zroot.offs = offs; | |
481 | c->zroot.len = len; | |
482 | } | |
483 | c->calc_idx_sz += ALIGN(len, 8); | |
484 | ||
485 | /* | |
486 | * Once lprops is updated, we can decrease the dirty znode count | |
487 | * but it is easier to just do it here. | |
488 | */ | |
489 | atomic_long_dec(&c->dirty_zn_cnt); | |
490 | ||
491 | /* | |
492 | * Calculate the next index node length to see if there is | |
493 | * enough room for it | |
494 | */ | |
495 | cnext = znode->cnext; | |
496 | if (cnext == c->cnext) | |
497 | next_len = 0; | |
498 | else | |
499 | next_len = ubifs_idx_node_sz(c, cnext->child_cnt); | |
500 | ||
1e51764a AB |
501 | /* Update buffer positions */ |
502 | wlen = used + len; | |
503 | used += ALIGN(len, 8); | |
504 | avail -= ALIGN(len, 8); | |
505 | ||
506 | if (next_len != 0 && | |
507 | buf_offs + used + next_len <= c->leb_size && | |
508 | avail > 0) | |
509 | continue; | |
510 | ||
511 | if (avail <= 0 && next_len && | |
512 | buf_offs + used + next_len <= c->leb_size) | |
513 | blen = buf_len; | |
514 | else | |
515 | blen = ALIGN(wlen, c->min_io_size); | |
516 | ||
517 | /* The buffer is full or there are no more znodes to do */ | |
518 | buf_offs += blen; | |
519 | if (next_len) { | |
520 | if (buf_offs + next_len > c->leb_size) { | |
521 | err = ubifs_update_one_lp(c, lnum, | |
522 | c->leb_size - buf_offs, blen - used, | |
523 | 0, 0); | |
524 | if (err) | |
525 | return err; | |
526 | lnum = -1; | |
527 | } | |
528 | used -= blen; | |
529 | if (used < 0) | |
530 | used = 0; | |
531 | avail = buf_len - used; | |
532 | continue; | |
533 | } | |
534 | err = ubifs_update_one_lp(c, lnum, c->leb_size - buf_offs, | |
535 | blen - used, 0, 0); | |
536 | if (err) | |
537 | return err; | |
538 | break; | |
539 | } | |
540 | ||
17c2f9f8 AB |
541 | c->dbg->new_ihead_lnum = lnum; |
542 | c->dbg->new_ihead_offs = buf_offs; | |
1e51764a AB |
543 | |
544 | return 0; | |
545 | } | |
546 | ||
547 | /** | |
548 | * layout_commit - determine positions of index nodes to commit. | |
549 | * @c: UBIFS file-system description object | |
550 | * @no_space: indicates that insufficient empty LEBs were allocated | |
551 | * @cnt: number of znodes to commit | |
552 | * | |
553 | * Calculate and update the positions of index nodes to commit. If there were | |
554 | * an insufficient number of empty LEBs allocated, then index nodes are placed | |
555 | * into the gaps created by obsolete index nodes in non-empty index LEBs. For | |
556 | * this purpose, an obsolete index node is one that was not in the index as at | |
557 | * the end of the last commit. To write "in-the-gaps" requires that those index | |
558 | * LEBs are updated atomically in-place. | |
559 | */ | |
560 | static int layout_commit(struct ubifs_info *c, int no_space, int cnt) | |
561 | { | |
562 | int err; | |
563 | ||
564 | if (no_space) { | |
565 | err = layout_in_gaps(c, cnt); | |
566 | if (err) | |
567 | return err; | |
568 | } | |
569 | err = layout_in_empty_space(c); | |
570 | return err; | |
571 | } | |
572 | ||
573 | /** | |
574 | * find_first_dirty - find first dirty znode. | |
575 | * @znode: znode to begin searching from | |
576 | */ | |
577 | static struct ubifs_znode *find_first_dirty(struct ubifs_znode *znode) | |
578 | { | |
579 | int i, cont; | |
580 | ||
581 | if (!znode) | |
582 | return NULL; | |
583 | ||
584 | while (1) { | |
585 | if (znode->level == 0) { | |
586 | if (ubifs_zn_dirty(znode)) | |
587 | return znode; | |
588 | return NULL; | |
589 | } | |
590 | cont = 0; | |
591 | for (i = 0; i < znode->child_cnt; i++) { | |
592 | struct ubifs_zbranch *zbr = &znode->zbranch[i]; | |
593 | ||
594 | if (zbr->znode && ubifs_zn_dirty(zbr->znode)) { | |
595 | znode = zbr->znode; | |
596 | cont = 1; | |
597 | break; | |
598 | } | |
599 | } | |
600 | if (!cont) { | |
601 | if (ubifs_zn_dirty(znode)) | |
602 | return znode; | |
603 | return NULL; | |
604 | } | |
605 | } | |
606 | } | |
607 | ||
608 | /** | |
609 | * find_next_dirty - find next dirty znode. | |
610 | * @znode: znode to begin searching from | |
611 | */ | |
612 | static struct ubifs_znode *find_next_dirty(struct ubifs_znode *znode) | |
613 | { | |
614 | int n = znode->iip + 1; | |
615 | ||
616 | znode = znode->parent; | |
617 | if (!znode) | |
618 | return NULL; | |
619 | for (; n < znode->child_cnt; n++) { | |
620 | struct ubifs_zbranch *zbr = &znode->zbranch[n]; | |
621 | ||
622 | if (zbr->znode && ubifs_zn_dirty(zbr->znode)) | |
623 | return find_first_dirty(zbr->znode); | |
624 | } | |
625 | return znode; | |
626 | } | |
627 | ||
628 | /** | |
629 | * get_znodes_to_commit - create list of dirty znodes to commit. | |
630 | * @c: UBIFS file-system description object | |
631 | * | |
632 | * This function returns the number of znodes to commit. | |
633 | */ | |
634 | static int get_znodes_to_commit(struct ubifs_info *c) | |
635 | { | |
636 | struct ubifs_znode *znode, *cnext; | |
637 | int cnt = 0; | |
638 | ||
639 | c->cnext = find_first_dirty(c->zroot.znode); | |
640 | znode = c->enext = c->cnext; | |
641 | if (!znode) { | |
642 | dbg_cmt("no znodes to commit"); | |
643 | return 0; | |
644 | } | |
645 | cnt += 1; | |
646 | while (1) { | |
6eb61d58 | 647 | ubifs_assert(c, !ubifs_zn_cow(znode)); |
1e51764a AB |
648 | __set_bit(COW_ZNODE, &znode->flags); |
649 | znode->alt = 0; | |
650 | cnext = find_next_dirty(znode); | |
651 | if (!cnext) { | |
652 | znode->cnext = c->cnext; | |
653 | break; | |
654 | } | |
16a26b20 SH |
655 | znode->cparent = znode->parent; |
656 | znode->ciip = znode->iip; | |
1e51764a AB |
657 | znode->cnext = cnext; |
658 | znode = cnext; | |
659 | cnt += 1; | |
660 | } | |
661 | dbg_cmt("committing %d znodes", cnt); | |
6eb61d58 | 662 | ubifs_assert(c, cnt == atomic_long_read(&c->dirty_zn_cnt)); |
1e51764a AB |
663 | return cnt; |
664 | } | |
665 | ||
666 | /** | |
667 | * alloc_idx_lebs - allocate empty LEBs to be used to commit. | |
668 | * @c: UBIFS file-system description object | |
669 | * @cnt: number of znodes to commit | |
670 | * | |
671 | * This function returns %-ENOSPC if it cannot allocate a sufficient number of | |
672 | * empty LEBs. %0 is returned on success, otherwise a negative error code | |
673 | * is returned. | |
674 | */ | |
675 | static int alloc_idx_lebs(struct ubifs_info *c, int cnt) | |
676 | { | |
677 | int i, leb_cnt, lnum; | |
678 | ||
679 | c->ileb_cnt = 0; | |
680 | c->ileb_nxt = 0; | |
681 | leb_cnt = get_leb_cnt(c, cnt); | |
682 | dbg_cmt("need about %d empty LEBS for TNC commit", leb_cnt); | |
683 | if (!leb_cnt) | |
684 | return 0; | |
6da2ec56 | 685 | c->ilebs = kmalloc_array(leb_cnt, sizeof(int), GFP_NOFS); |
1e51764a AB |
686 | if (!c->ilebs) |
687 | return -ENOMEM; | |
688 | for (i = 0; i < leb_cnt; i++) { | |
689 | lnum = ubifs_find_free_leb_for_idx(c); | |
690 | if (lnum < 0) | |
691 | return lnum; | |
692 | c->ilebs[c->ileb_cnt++] = lnum; | |
693 | dbg_cmt("LEB %d", lnum); | |
694 | } | |
3d251a5b | 695 | if (dbg_is_chk_index(c) && !(prandom_u32() & 7)) |
1e51764a AB |
696 | return -ENOSPC; |
697 | return 0; | |
698 | } | |
699 | ||
700 | /** | |
701 | * free_unused_idx_lebs - free unused LEBs that were allocated for the commit. | |
702 | * @c: UBIFS file-system description object | |
703 | * | |
704 | * It is possible that we allocate more empty LEBs for the commit than we need. | |
705 | * This functions frees the surplus. | |
706 | * | |
707 | * This function returns %0 on success and a negative error code on failure. | |
708 | */ | |
709 | static int free_unused_idx_lebs(struct ubifs_info *c) | |
710 | { | |
711 | int i, err = 0, lnum, er; | |
712 | ||
713 | for (i = c->ileb_nxt; i < c->ileb_cnt; i++) { | |
714 | lnum = c->ilebs[i]; | |
715 | dbg_cmt("LEB %d", lnum); | |
716 | er = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0, | |
717 | LPROPS_INDEX | LPROPS_TAKEN, 0); | |
718 | if (!err) | |
719 | err = er; | |
720 | } | |
721 | return err; | |
722 | } | |
723 | ||
724 | /** | |
725 | * free_idx_lebs - free unused LEBs after commit end. | |
726 | * @c: UBIFS file-system description object | |
727 | * | |
728 | * This function returns %0 on success and a negative error code on failure. | |
729 | */ | |
730 | static int free_idx_lebs(struct ubifs_info *c) | |
731 | { | |
732 | int err; | |
733 | ||
734 | err = free_unused_idx_lebs(c); | |
735 | kfree(c->ilebs); | |
736 | c->ilebs = NULL; | |
737 | return err; | |
738 | } | |
739 | ||
740 | /** | |
741 | * ubifs_tnc_start_commit - start TNC commit. | |
742 | * @c: UBIFS file-system description object | |
743 | * @zroot: new index root position is returned here | |
744 | * | |
745 | * This function prepares the list of indexing nodes to commit and lays out | |
746 | * their positions on flash. If there is not enough free space it uses the | |
747 | * in-gap commit method. Returns zero in case of success and a negative error | |
748 | * code in case of failure. | |
749 | */ | |
750 | int ubifs_tnc_start_commit(struct ubifs_info *c, struct ubifs_zbranch *zroot) | |
751 | { | |
752 | int err = 0, cnt; | |
753 | ||
754 | mutex_lock(&c->tnc_mutex); | |
755 | err = dbg_check_tnc(c, 1); | |
756 | if (err) | |
757 | goto out; | |
758 | cnt = get_znodes_to_commit(c); | |
759 | if (cnt != 0) { | |
760 | int no_space = 0; | |
761 | ||
762 | err = alloc_idx_lebs(c, cnt); | |
763 | if (err == -ENOSPC) | |
764 | no_space = 1; | |
765 | else if (err) | |
766 | goto out_free; | |
767 | err = layout_commit(c, no_space, cnt); | |
768 | if (err) | |
769 | goto out_free; | |
6eb61d58 | 770 | ubifs_assert(c, atomic_long_read(&c->dirty_zn_cnt) == 0); |
1e51764a AB |
771 | err = free_unused_idx_lebs(c); |
772 | if (err) | |
773 | goto out; | |
774 | } | |
775 | destroy_old_idx(c); | |
776 | memcpy(zroot, &c->zroot, sizeof(struct ubifs_zbranch)); | |
777 | ||
778 | err = ubifs_save_dirty_idx_lnums(c); | |
779 | if (err) | |
780 | goto out; | |
781 | ||
782 | spin_lock(&c->space_lock); | |
783 | /* | |
784 | * Although we have not finished committing yet, update size of the | |
b137545c | 785 | * committed index ('c->bi.old_idx_sz') and zero out the index growth |
1e51764a AB |
786 | * budget. It is OK to do this now, because we've reserved all the |
787 | * space which is needed to commit the index, and it is save for the | |
788 | * budgeting subsystem to assume the index is already committed, | |
789 | * even though it is not. | |
790 | */ | |
6eb61d58 | 791 | ubifs_assert(c, c->bi.min_idx_lebs == ubifs_calc_min_idx_lebs(c)); |
b137545c AB |
792 | c->bi.old_idx_sz = c->calc_idx_sz; |
793 | c->bi.uncommitted_idx = 0; | |
794 | c->bi.min_idx_lebs = ubifs_calc_min_idx_lebs(c); | |
1e51764a AB |
795 | spin_unlock(&c->space_lock); |
796 | mutex_unlock(&c->tnc_mutex); | |
797 | ||
798 | dbg_cmt("number of index LEBs %d", c->lst.idx_lebs); | |
799 | dbg_cmt("size of index %llu", c->calc_idx_sz); | |
800 | return err; | |
801 | ||
802 | out_free: | |
803 | free_idx_lebs(c); | |
804 | out: | |
805 | mutex_unlock(&c->tnc_mutex); | |
806 | return err; | |
807 | } | |
808 | ||
809 | /** | |
810 | * write_index - write index nodes. | |
811 | * @c: UBIFS file-system description object | |
812 | * | |
813 | * This function writes the index nodes whose positions were laid out in the | |
814 | * layout_in_empty_space function. | |
815 | */ | |
816 | static int write_index(struct ubifs_info *c) | |
817 | { | |
818 | struct ubifs_idx_node *idx; | |
819 | struct ubifs_znode *znode, *cnext; | |
820 | int i, lnum, offs, len, next_len, buf_len, buf_offs, used; | |
1f42596e | 821 | int avail, wlen, err, lnum_pos = 0, blen, nxt_offs; |
1e51764a AB |
822 | |
823 | cnext = c->enext; | |
824 | if (!cnext) | |
825 | return 0; | |
826 | ||
827 | /* | |
828 | * Always write index nodes to the index head so that index nodes and | |
829 | * other types of nodes are never mixed in the same erase block. | |
830 | */ | |
831 | lnum = c->ihead_lnum; | |
832 | buf_offs = c->ihead_offs; | |
833 | ||
834 | /* Allocate commit buffer */ | |
835 | buf_len = ALIGN(c->max_idx_node_sz, c->min_io_size); | |
836 | used = 0; | |
837 | avail = buf_len; | |
838 | ||
839 | /* Ensure there is enough room for first write */ | |
840 | next_len = ubifs_idx_node_sz(c, cnext->child_cnt); | |
841 | if (buf_offs + next_len > c->leb_size) { | |
842 | err = ubifs_update_one_lp(c, lnum, LPROPS_NC, 0, 0, | |
843 | LPROPS_TAKEN); | |
844 | if (err) | |
845 | return err; | |
846 | lnum = -1; | |
847 | } | |
848 | ||
849 | while (1) { | |
16a26b20 SH |
850 | u8 hash[UBIFS_HASH_ARR_SZ]; |
851 | ||
1e51764a AB |
852 | cond_resched(); |
853 | ||
854 | znode = cnext; | |
855 | idx = c->cbuf + used; | |
856 | ||
857 | /* Make index node */ | |
858 | idx->ch.node_type = UBIFS_IDX_NODE; | |
859 | idx->child_cnt = cpu_to_le16(znode->child_cnt); | |
860 | idx->level = cpu_to_le16(znode->level); | |
861 | for (i = 0; i < znode->child_cnt; i++) { | |
862 | struct ubifs_branch *br = ubifs_idx_branch(c, idx, i); | |
863 | struct ubifs_zbranch *zbr = &znode->zbranch[i]; | |
864 | ||
865 | key_write_idx(c, &zbr->key, &br->key); | |
866 | br->lnum = cpu_to_le32(zbr->lnum); | |
867 | br->offs = cpu_to_le32(zbr->offs); | |
868 | br->len = cpu_to_le32(zbr->len); | |
16a26b20 | 869 | ubifs_copy_hash(c, zbr->hash, ubifs_branch_hash(c, br)); |
1e51764a | 870 | if (!zbr->lnum || !zbr->len) { |
235c362b | 871 | ubifs_err(c, "bad ref in znode"); |
edf6be24 | 872 | ubifs_dump_znode(c, znode); |
1e51764a | 873 | if (zbr->znode) |
edf6be24 | 874 | ubifs_dump_znode(c, zbr->znode); |
df71b091 RW |
875 | |
876 | return -EINVAL; | |
1e51764a AB |
877 | } |
878 | } | |
879 | len = ubifs_idx_node_sz(c, znode->child_cnt); | |
880 | ubifs_prepare_node(c, idx, len, 0); | |
16a26b20 SH |
881 | ubifs_node_calc_hash(c, idx, hash); |
882 | ||
883 | mutex_lock(&c->tnc_mutex); | |
884 | ||
885 | if (znode->cparent) | |
886 | ubifs_copy_hash(c, hash, | |
887 | znode->cparent->zbranch[znode->ciip].hash); | |
888 | ||
889 | if (znode->parent) { | |
890 | if (!ubifs_zn_obsolete(znode)) | |
891 | ubifs_copy_hash(c, hash, | |
892 | znode->parent->zbranch[znode->iip].hash); | |
893 | } else { | |
894 | ubifs_copy_hash(c, hash, c->zroot.hash); | |
895 | } | |
896 | ||
897 | mutex_unlock(&c->tnc_mutex); | |
1e51764a AB |
898 | |
899 | /* Determine the index node position */ | |
900 | if (lnum == -1) { | |
901 | lnum = c->ilebs[lnum_pos++]; | |
902 | buf_offs = 0; | |
903 | used = 0; | |
904 | avail = buf_len; | |
905 | } | |
906 | offs = buf_offs + used; | |
907 | ||
1e51764a AB |
908 | if (lnum != znode->lnum || offs != znode->offs || |
909 | len != znode->len) { | |
235c362b | 910 | ubifs_err(c, "inconsistent znode posn"); |
1e51764a AB |
911 | return -EINVAL; |
912 | } | |
1e51764a AB |
913 | |
914 | /* Grab some stuff from znode while we still can */ | |
915 | cnext = znode->cnext; | |
916 | ||
6eb61d58 RW |
917 | ubifs_assert(c, ubifs_zn_dirty(znode)); |
918 | ubifs_assert(c, ubifs_zn_cow(znode)); | |
1e51764a AB |
919 | |
920 | /* | |
921 | * It is important that other threads should see %DIRTY_ZNODE | |
922 | * flag cleared before %COW_ZNODE. Specifically, it matters in | |
923 | * the 'dirty_cow_znode()' function. This is the reason for the | |
924 | * first barrier. Also, we want the bit changes to be seen to | |
925 | * other threads ASAP, to avoid unnecesarry copying, which is | |
926 | * the reason for the second barrier. | |
927 | */ | |
928 | clear_bit(DIRTY_ZNODE, &znode->flags); | |
4e857c58 | 929 | smp_mb__before_atomic(); |
1e51764a | 930 | clear_bit(COW_ZNODE, &znode->flags); |
4e857c58 | 931 | smp_mb__after_atomic(); |
1e51764a | 932 | |
06b282a4 AB |
933 | /* |
934 | * We have marked the znode as clean but have not updated the | |
935 | * @c->clean_zn_cnt counter. If this znode becomes dirty again | |
936 | * before 'free_obsolete_znodes()' is called, then | |
937 | * @c->clean_zn_cnt will be decremented before it gets | |
938 | * incremented (resulting in 2 decrements for the same znode). | |
939 | * This means that @c->clean_zn_cnt may become negative for a | |
940 | * while. | |
941 | * | |
942 | * Q: why we cannot increment @c->clean_zn_cnt? | |
943 | * A: because we do not have the @c->tnc_mutex locked, and the | |
944 | * following code would be racy and buggy: | |
945 | * | |
946 | * if (!ubifs_zn_obsolete(znode)) { | |
947 | * atomic_long_inc(&c->clean_zn_cnt); | |
948 | * atomic_long_inc(&ubifs_clean_zn_cnt); | |
949 | * } | |
950 | * | |
951 | * Thus, we just delay the @c->clean_zn_cnt update until we | |
952 | * have the mutex locked. | |
953 | */ | |
954 | ||
1e51764a AB |
955 | /* Do not access znode from this point on */ |
956 | ||
957 | /* Update buffer positions */ | |
958 | wlen = used + len; | |
959 | used += ALIGN(len, 8); | |
960 | avail -= ALIGN(len, 8); | |
961 | ||
962 | /* | |
963 | * Calculate the next index node length to see if there is | |
964 | * enough room for it | |
965 | */ | |
966 | if (cnext == c->cnext) | |
967 | next_len = 0; | |
968 | else | |
969 | next_len = ubifs_idx_node_sz(c, cnext->child_cnt); | |
970 | ||
1f42596e AB |
971 | nxt_offs = buf_offs + used + next_len; |
972 | if (next_len && nxt_offs <= c->leb_size) { | |
973 | if (avail > 0) | |
1e51764a | 974 | continue; |
1f42596e AB |
975 | else |
976 | blen = buf_len; | |
1e51764a | 977 | } else { |
1f42596e AB |
978 | wlen = ALIGN(wlen, 8); |
979 | blen = ALIGN(wlen, c->min_io_size); | |
980 | ubifs_pad(c, c->cbuf + wlen, blen - wlen); | |
981 | } | |
982 | ||
983 | /* The buffer is full or there are no more znodes to do */ | |
b36a261e | 984 | err = ubifs_leb_write(c, lnum, c->cbuf, buf_offs, blen); |
1f42596e AB |
985 | if (err) |
986 | return err; | |
987 | buf_offs += blen; | |
988 | if (next_len) { | |
989 | if (nxt_offs > c->leb_size) { | |
990 | err = ubifs_update_one_lp(c, lnum, LPROPS_NC, 0, | |
991 | 0, LPROPS_TAKEN); | |
992 | if (err) | |
993 | return err; | |
994 | lnum = -1; | |
1e51764a | 995 | } |
1f42596e AB |
996 | used -= blen; |
997 | if (used < 0) | |
998 | used = 0; | |
999 | avail = buf_len - used; | |
1000 | memmove(c->cbuf, c->cbuf + blen, used); | |
1001 | continue; | |
1e51764a AB |
1002 | } |
1003 | break; | |
1004 | } | |
1005 | ||
17c2f9f8 AB |
1006 | if (lnum != c->dbg->new_ihead_lnum || |
1007 | buf_offs != c->dbg->new_ihead_offs) { | |
235c362b | 1008 | ubifs_err(c, "inconsistent ihead"); |
1e51764a AB |
1009 | return -EINVAL; |
1010 | } | |
1e51764a AB |
1011 | |
1012 | c->ihead_lnum = lnum; | |
1013 | c->ihead_offs = buf_offs; | |
1014 | ||
1015 | return 0; | |
1016 | } | |
1017 | ||
1018 | /** | |
1019 | * free_obsolete_znodes - free obsolete znodes. | |
1020 | * @c: UBIFS file-system description object | |
1021 | * | |
1022 | * At the end of commit end, obsolete znodes are freed. | |
1023 | */ | |
1024 | static void free_obsolete_znodes(struct ubifs_info *c) | |
1025 | { | |
1026 | struct ubifs_znode *znode, *cnext; | |
1027 | ||
1028 | cnext = c->cnext; | |
1029 | do { | |
1030 | znode = cnext; | |
1031 | cnext = znode->cnext; | |
f42eed7c | 1032 | if (ubifs_zn_obsolete(znode)) |
1e51764a AB |
1033 | kfree(znode); |
1034 | else { | |
1035 | znode->cnext = NULL; | |
1036 | atomic_long_inc(&c->clean_zn_cnt); | |
1037 | atomic_long_inc(&ubifs_clean_zn_cnt); | |
1038 | } | |
1039 | } while (cnext != c->cnext); | |
1040 | } | |
1041 | ||
1042 | /** | |
1043 | * return_gap_lebs - return LEBs used by the in-gap commit method. | |
1044 | * @c: UBIFS file-system description object | |
1045 | * | |
1046 | * This function clears the "taken" flag for the LEBs which were used by the | |
1047 | * "commit in-the-gaps" method. | |
1048 | */ | |
1049 | static int return_gap_lebs(struct ubifs_info *c) | |
1050 | { | |
1051 | int *p, err; | |
1052 | ||
1053 | if (!c->gap_lebs) | |
1054 | return 0; | |
1055 | ||
1056 | dbg_cmt(""); | |
1057 | for (p = c->gap_lebs; *p != -1; p++) { | |
1058 | err = ubifs_change_one_lp(c, *p, LPROPS_NC, LPROPS_NC, 0, | |
1059 | LPROPS_TAKEN, 0); | |
1060 | if (err) | |
1061 | return err; | |
1062 | } | |
1063 | ||
1064 | kfree(c->gap_lebs); | |
1065 | c->gap_lebs = NULL; | |
1066 | return 0; | |
1067 | } | |
1068 | ||
1069 | /** | |
1070 | * ubifs_tnc_end_commit - update the TNC for commit end. | |
1071 | * @c: UBIFS file-system description object | |
1072 | * | |
1073 | * Write the dirty znodes. | |
1074 | */ | |
1075 | int ubifs_tnc_end_commit(struct ubifs_info *c) | |
1076 | { | |
1077 | int err; | |
1078 | ||
1079 | if (!c->cnext) | |
1080 | return 0; | |
1081 | ||
1082 | err = return_gap_lebs(c); | |
1083 | if (err) | |
1084 | return err; | |
1085 | ||
1086 | err = write_index(c); | |
1087 | if (err) | |
1088 | return err; | |
1089 | ||
1090 | mutex_lock(&c->tnc_mutex); | |
1091 | ||
1092 | dbg_cmt("TNC height is %d", c->zroot.znode->level + 1); | |
1093 | ||
1094 | free_obsolete_znodes(c); | |
1095 | ||
1096 | c->cnext = NULL; | |
1097 | kfree(c->ilebs); | |
1098 | c->ilebs = NULL; | |
1099 | ||
1100 | mutex_unlock(&c->tnc_mutex); | |
1101 | ||
1102 | return 0; | |
1103 | } |