]>
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 | /* | |
24 | * This file implements functions needed to recover from unclean un-mounts. | |
25 | * When UBIFS is mounted, it checks a flag on the master node to determine if | |
af901ca1 | 26 | * an un-mount was completed successfully. If not, the process of mounting |
6fb4374f | 27 | * incorporates additional checking and fixing of on-flash data structures. |
1e51764a AB |
28 | * UBIFS always cleans away all remnants of an unclean un-mount, so that |
29 | * errors do not accumulate. However UBIFS defers recovery if it is mounted | |
30 | * read-only, and the flash is not modified in that case. | |
be7b42a5 AB |
31 | * |
32 | * The general UBIFS approach to the recovery is that it recovers from | |
33 | * corruptions which could be caused by power cuts, but it refuses to recover | |
34 | * from corruption caused by other reasons. And UBIFS tries to distinguish | |
35 | * between these 2 reasons of corruptions and silently recover in the former | |
36 | * case and loudly complain in the latter case. | |
37 | * | |
38 | * UBIFS writes only to erased LEBs, so it writes only to the flash space | |
39 | * containing only 0xFFs. UBIFS also always writes strictly from the beginning | |
40 | * of the LEB to the end. And UBIFS assumes that the underlying flash media | |
2765df7d | 41 | * writes in @c->max_write_size bytes at a time. |
be7b42a5 AB |
42 | * |
43 | * Hence, if UBIFS finds a corrupted node at offset X, it expects only the min. | |
44 | * I/O unit corresponding to offset X to contain corrupted data, all the | |
45 | * following min. I/O units have to contain empty space (all 0xFFs). If this is | |
46 | * not true, the corruption cannot be the result of a power cut, and UBIFS | |
47 | * refuses to mount. | |
1e51764a AB |
48 | */ |
49 | ||
50 | #include <linux/crc32.h> | |
5a0e3ad6 | 51 | #include <linux/slab.h> |
1e51764a AB |
52 | #include "ubifs.h" |
53 | ||
54 | /** | |
55 | * is_empty - determine whether a buffer is empty (contains all 0xff). | |
56 | * @buf: buffer to clean | |
57 | * @len: length of buffer | |
58 | * | |
59 | * This function returns %1 if the buffer is empty (contains all 0xff) otherwise | |
60 | * %0 is returned. | |
61 | */ | |
62 | static int is_empty(void *buf, int len) | |
63 | { | |
64 | uint8_t *p = buf; | |
65 | int i; | |
66 | ||
67 | for (i = 0; i < len; i++) | |
68 | if (*p++ != 0xff) | |
69 | return 0; | |
70 | return 1; | |
71 | } | |
72 | ||
06112547 AB |
73 | /** |
74 | * first_non_ff - find offset of the first non-0xff byte. | |
75 | * @buf: buffer to search in | |
76 | * @len: length of buffer | |
77 | * | |
78 | * This function returns offset of the first non-0xff byte in @buf or %-1 if | |
79 | * the buffer contains only 0xff bytes. | |
80 | */ | |
81 | static int first_non_ff(void *buf, int len) | |
82 | { | |
83 | uint8_t *p = buf; | |
84 | int i; | |
85 | ||
86 | for (i = 0; i < len; i++) | |
87 | if (*p++ != 0xff) | |
88 | return i; | |
89 | return -1; | |
90 | } | |
91 | ||
1e51764a AB |
92 | /** |
93 | * get_master_node - get the last valid master node allowing for corruption. | |
94 | * @c: UBIFS file-system description object | |
95 | * @lnum: LEB number | |
96 | * @pbuf: buffer containing the LEB read, is returned here | |
97 | * @mst: master node, if found, is returned here | |
98 | * @cor: corruption, if found, is returned here | |
99 | * | |
100 | * This function allocates a buffer, reads the LEB into it, and finds and | |
101 | * returns the last valid master node allowing for one area of corruption. | |
102 | * The corrupt area, if there is one, must be consistent with the assumption | |
103 | * that it is the result of an unclean unmount while the master node was being | |
104 | * written. Under those circumstances, it is valid to use the previously written | |
105 | * master node. | |
106 | * | |
107 | * This function returns %0 on success and a negative error code on failure. | |
108 | */ | |
109 | static int get_master_node(const struct ubifs_info *c, int lnum, void **pbuf, | |
110 | struct ubifs_mst_node **mst, void **cor) | |
111 | { | |
112 | const int sz = c->mst_node_alsz; | |
113 | int err, offs, len; | |
114 | void *sbuf, *buf; | |
115 | ||
116 | sbuf = vmalloc(c->leb_size); | |
117 | if (!sbuf) | |
118 | return -ENOMEM; | |
119 | ||
120 | err = ubi_read(c->ubi, lnum, sbuf, 0, c->leb_size); | |
121 | if (err && err != -EBADMSG) | |
122 | goto out_free; | |
123 | ||
124 | /* Find the first position that is definitely not a node */ | |
125 | offs = 0; | |
126 | buf = sbuf; | |
127 | len = c->leb_size; | |
128 | while (offs + UBIFS_MST_NODE_SZ <= c->leb_size) { | |
129 | struct ubifs_ch *ch = buf; | |
130 | ||
131 | if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC) | |
132 | break; | |
133 | offs += sz; | |
134 | buf += sz; | |
135 | len -= sz; | |
136 | } | |
137 | /* See if there was a valid master node before that */ | |
138 | if (offs) { | |
139 | int ret; | |
140 | ||
141 | offs -= sz; | |
142 | buf -= sz; | |
143 | len += sz; | |
144 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1); | |
145 | if (ret != SCANNED_A_NODE && offs) { | |
146 | /* Could have been corruption so check one place back */ | |
147 | offs -= sz; | |
148 | buf -= sz; | |
149 | len += sz; | |
150 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1); | |
151 | if (ret != SCANNED_A_NODE) | |
152 | /* | |
153 | * We accept only one area of corruption because | |
154 | * we are assuming that it was caused while | |
155 | * trying to write a master node. | |
156 | */ | |
157 | goto out_err; | |
158 | } | |
159 | if (ret == SCANNED_A_NODE) { | |
160 | struct ubifs_ch *ch = buf; | |
161 | ||
162 | if (ch->node_type != UBIFS_MST_NODE) | |
163 | goto out_err; | |
164 | dbg_rcvry("found a master node at %d:%d", lnum, offs); | |
165 | *mst = buf; | |
166 | offs += sz; | |
167 | buf += sz; | |
168 | len -= sz; | |
169 | } | |
170 | } | |
171 | /* Check for corruption */ | |
172 | if (offs < c->leb_size) { | |
173 | if (!is_empty(buf, min_t(int, len, sz))) { | |
174 | *cor = buf; | |
175 | dbg_rcvry("found corruption at %d:%d", lnum, offs); | |
176 | } | |
177 | offs += sz; | |
178 | buf += sz; | |
179 | len -= sz; | |
180 | } | |
181 | /* Check remaining empty space */ | |
182 | if (offs < c->leb_size) | |
183 | if (!is_empty(buf, len)) | |
184 | goto out_err; | |
185 | *pbuf = sbuf; | |
186 | return 0; | |
187 | ||
188 | out_err: | |
189 | err = -EINVAL; | |
190 | out_free: | |
191 | vfree(sbuf); | |
192 | *mst = NULL; | |
193 | *cor = NULL; | |
194 | return err; | |
195 | } | |
196 | ||
197 | /** | |
198 | * write_rcvrd_mst_node - write recovered master node. | |
199 | * @c: UBIFS file-system description object | |
200 | * @mst: master node | |
201 | * | |
202 | * This function returns %0 on success and a negative error code on failure. | |
203 | */ | |
204 | static int write_rcvrd_mst_node(struct ubifs_info *c, | |
205 | struct ubifs_mst_node *mst) | |
206 | { | |
207 | int err = 0, lnum = UBIFS_MST_LNUM, sz = c->mst_node_alsz; | |
0ecb9529 | 208 | __le32 save_flags; |
1e51764a AB |
209 | |
210 | dbg_rcvry("recovery"); | |
211 | ||
212 | save_flags = mst->flags; | |
0ecb9529 | 213 | mst->flags |= cpu_to_le32(UBIFS_MST_RCVRY); |
1e51764a AB |
214 | |
215 | ubifs_prepare_node(c, mst, UBIFS_MST_NODE_SZ, 1); | |
216 | err = ubi_leb_change(c->ubi, lnum, mst, sz, UBI_SHORTTERM); | |
217 | if (err) | |
218 | goto out; | |
219 | err = ubi_leb_change(c->ubi, lnum + 1, mst, sz, UBI_SHORTTERM); | |
220 | if (err) | |
221 | goto out; | |
222 | out: | |
223 | mst->flags = save_flags; | |
224 | return err; | |
225 | } | |
226 | ||
227 | /** | |
228 | * ubifs_recover_master_node - recover the master node. | |
229 | * @c: UBIFS file-system description object | |
230 | * | |
231 | * This function recovers the master node from corruption that may occur due to | |
232 | * an unclean unmount. | |
233 | * | |
234 | * This function returns %0 on success and a negative error code on failure. | |
235 | */ | |
236 | int ubifs_recover_master_node(struct ubifs_info *c) | |
237 | { | |
238 | void *buf1 = NULL, *buf2 = NULL, *cor1 = NULL, *cor2 = NULL; | |
239 | struct ubifs_mst_node *mst1 = NULL, *mst2 = NULL, *mst; | |
240 | const int sz = c->mst_node_alsz; | |
241 | int err, offs1, offs2; | |
242 | ||
243 | dbg_rcvry("recovery"); | |
244 | ||
245 | err = get_master_node(c, UBIFS_MST_LNUM, &buf1, &mst1, &cor1); | |
246 | if (err) | |
247 | goto out_free; | |
248 | ||
249 | err = get_master_node(c, UBIFS_MST_LNUM + 1, &buf2, &mst2, &cor2); | |
250 | if (err) | |
251 | goto out_free; | |
252 | ||
253 | if (mst1) { | |
254 | offs1 = (void *)mst1 - buf1; | |
255 | if ((le32_to_cpu(mst1->flags) & UBIFS_MST_RCVRY) && | |
256 | (offs1 == 0 && !cor1)) { | |
257 | /* | |
258 | * mst1 was written by recovery at offset 0 with no | |
259 | * corruption. | |
260 | */ | |
261 | dbg_rcvry("recovery recovery"); | |
262 | mst = mst1; | |
263 | } else if (mst2) { | |
264 | offs2 = (void *)mst2 - buf2; | |
265 | if (offs1 == offs2) { | |
266 | /* Same offset, so must be the same */ | |
267 | if (memcmp((void *)mst1 + UBIFS_CH_SZ, | |
268 | (void *)mst2 + UBIFS_CH_SZ, | |
269 | UBIFS_MST_NODE_SZ - UBIFS_CH_SZ)) | |
270 | goto out_err; | |
271 | mst = mst1; | |
272 | } else if (offs2 + sz == offs1) { | |
273 | /* 1st LEB was written, 2nd was not */ | |
274 | if (cor1) | |
275 | goto out_err; | |
276 | mst = mst1; | |
277 | } else if (offs1 == 0 && offs2 + sz >= c->leb_size) { | |
278 | /* 1st LEB was unmapped and written, 2nd not */ | |
279 | if (cor1) | |
280 | goto out_err; | |
281 | mst = mst1; | |
282 | } else | |
283 | goto out_err; | |
284 | } else { | |
285 | /* | |
286 | * 2nd LEB was unmapped and about to be written, so | |
287 | * there must be only one master node in the first LEB | |
288 | * and no corruption. | |
289 | */ | |
290 | if (offs1 != 0 || cor1) | |
291 | goto out_err; | |
292 | mst = mst1; | |
293 | } | |
294 | } else { | |
295 | if (!mst2) | |
296 | goto out_err; | |
297 | /* | |
298 | * 1st LEB was unmapped and about to be written, so there must | |
299 | * be no room left in 2nd LEB. | |
300 | */ | |
301 | offs2 = (void *)mst2 - buf2; | |
302 | if (offs2 + sz + sz <= c->leb_size) | |
303 | goto out_err; | |
304 | mst = mst2; | |
305 | } | |
306 | ||
348709ba | 307 | ubifs_msg("recovered master node from LEB %d", |
1e51764a AB |
308 | (mst == mst1 ? UBIFS_MST_LNUM : UBIFS_MST_LNUM + 1)); |
309 | ||
310 | memcpy(c->mst_node, mst, UBIFS_MST_NODE_SZ); | |
311 | ||
2ef13294 | 312 | if (c->ro_mount) { |
1e51764a AB |
313 | /* Read-only mode. Keep a copy for switching to rw mode */ |
314 | c->rcvrd_mst_node = kmalloc(sz, GFP_KERNEL); | |
315 | if (!c->rcvrd_mst_node) { | |
316 | err = -ENOMEM; | |
317 | goto out_free; | |
318 | } | |
319 | memcpy(c->rcvrd_mst_node, c->mst_node, UBIFS_MST_NODE_SZ); | |
320 | } else { | |
321 | /* Write the recovered master node */ | |
322 | c->max_sqnum = le64_to_cpu(mst->ch.sqnum) - 1; | |
323 | err = write_rcvrd_mst_node(c, c->mst_node); | |
324 | if (err) | |
325 | goto out_free; | |
326 | } | |
327 | ||
328 | vfree(buf2); | |
329 | vfree(buf1); | |
330 | ||
331 | return 0; | |
332 | ||
333 | out_err: | |
334 | err = -EINVAL; | |
335 | out_free: | |
336 | ubifs_err("failed to recover master node"); | |
337 | if (mst1) { | |
338 | dbg_err("dumping first master node"); | |
339 | dbg_dump_node(c, mst1); | |
340 | } | |
341 | if (mst2) { | |
342 | dbg_err("dumping second master node"); | |
343 | dbg_dump_node(c, mst2); | |
344 | } | |
345 | vfree(buf2); | |
346 | vfree(buf1); | |
347 | return err; | |
348 | } | |
349 | ||
350 | /** | |
351 | * ubifs_write_rcvrd_mst_node - write the recovered master node. | |
352 | * @c: UBIFS file-system description object | |
353 | * | |
354 | * This function writes the master node that was recovered during mounting in | |
355 | * read-only mode and must now be written because we are remounting rw. | |
356 | * | |
357 | * This function returns %0 on success and a negative error code on failure. | |
358 | */ | |
359 | int ubifs_write_rcvrd_mst_node(struct ubifs_info *c) | |
360 | { | |
361 | int err; | |
362 | ||
363 | if (!c->rcvrd_mst_node) | |
364 | return 0; | |
365 | c->rcvrd_mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY); | |
366 | c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY); | |
367 | err = write_rcvrd_mst_node(c, c->rcvrd_mst_node); | |
368 | if (err) | |
369 | return err; | |
370 | kfree(c->rcvrd_mst_node); | |
371 | c->rcvrd_mst_node = NULL; | |
372 | return 0; | |
373 | } | |
374 | ||
375 | /** | |
376 | * is_last_write - determine if an offset was in the last write to a LEB. | |
377 | * @c: UBIFS file-system description object | |
378 | * @buf: buffer to check | |
379 | * @offs: offset to check | |
380 | * | |
381 | * This function returns %1 if @offs was in the last write to the LEB whose data | |
2765df7d AB |
382 | * is in @buf, otherwise %0 is returned. The determination is made by checking |
383 | * for subsequent empty space starting from the next @c->max_write_size | |
384 | * boundary. | |
1e51764a AB |
385 | */ |
386 | static int is_last_write(const struct ubifs_info *c, void *buf, int offs) | |
387 | { | |
428ff9d2 | 388 | int empty_offs, check_len; |
1e51764a AB |
389 | uint8_t *p; |
390 | ||
1e51764a | 391 | /* |
2765df7d AB |
392 | * Round up to the next @c->max_write_size boundary i.e. @offs is in |
393 | * the last wbuf written. After that should be empty space. | |
1e51764a | 394 | */ |
2765df7d | 395 | empty_offs = ALIGN(offs + 1, c->max_write_size); |
1e51764a AB |
396 | check_len = c->leb_size - empty_offs; |
397 | p = buf + empty_offs - offs; | |
431102fe | 398 | return is_empty(p, check_len); |
1e51764a AB |
399 | } |
400 | ||
401 | /** | |
402 | * clean_buf - clean the data from an LEB sitting in a buffer. | |
403 | * @c: UBIFS file-system description object | |
404 | * @buf: buffer to clean | |
405 | * @lnum: LEB number to clean | |
406 | * @offs: offset from which to clean | |
407 | * @len: length of buffer | |
408 | * | |
409 | * This function pads up to the next min_io_size boundary (if there is one) and | |
410 | * sets empty space to all 0xff. @buf, @offs and @len are updated to the next | |
428ff9d2 | 411 | * @c->min_io_size boundary. |
1e51764a AB |
412 | */ |
413 | static void clean_buf(const struct ubifs_info *c, void **buf, int lnum, | |
414 | int *offs, int *len) | |
415 | { | |
416 | int empty_offs, pad_len; | |
417 | ||
418 | lnum = lnum; | |
419 | dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs); | |
420 | ||
1e51764a AB |
421 | ubifs_assert(!(*offs & 7)); |
422 | empty_offs = ALIGN(*offs, c->min_io_size); | |
423 | pad_len = empty_offs - *offs; | |
424 | ubifs_pad(c, *buf, pad_len); | |
425 | *offs += pad_len; | |
426 | *buf += pad_len; | |
427 | *len -= pad_len; | |
428 | memset(*buf, 0xff, c->leb_size - empty_offs); | |
429 | } | |
430 | ||
431 | /** | |
432 | * no_more_nodes - determine if there are no more nodes in a buffer. | |
433 | * @c: UBIFS file-system description object | |
434 | * @buf: buffer to check | |
435 | * @len: length of buffer | |
436 | * @lnum: LEB number of the LEB from which @buf was read | |
437 | * @offs: offset from which @buf was read | |
438 | * | |
de097578 AH |
439 | * This function ensures that the corrupted node at @offs is the last thing |
440 | * written to a LEB. This function returns %1 if more data is not found and | |
441 | * %0 if more data is found. | |
1e51764a AB |
442 | */ |
443 | static int no_more_nodes(const struct ubifs_info *c, void *buf, int len, | |
444 | int lnum, int offs) | |
445 | { | |
de097578 AH |
446 | struct ubifs_ch *ch = buf; |
447 | int skip, dlen = le32_to_cpu(ch->len); | |
1e51764a | 448 | |
de097578 | 449 | /* Check for empty space after the corrupt node's common header */ |
2765df7d | 450 | skip = ALIGN(offs + UBIFS_CH_SZ, c->max_write_size) - offs; |
de097578 AH |
451 | if (is_empty(buf + skip, len - skip)) |
452 | return 1; | |
453 | /* | |
454 | * The area after the common header size is not empty, so the common | |
455 | * header must be intact. Check it. | |
456 | */ | |
457 | if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) { | |
458 | dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs); | |
459 | return 0; | |
1e51764a | 460 | } |
de097578 | 461 | /* Now we know the corrupt node's length we can skip over it */ |
2765df7d | 462 | skip = ALIGN(offs + dlen, c->max_write_size) - offs; |
de097578 AH |
463 | /* After which there should be empty space */ |
464 | if (is_empty(buf + skip, len - skip)) | |
465 | return 1; | |
466 | dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip); | |
467 | return 0; | |
1e51764a AB |
468 | } |
469 | ||
470 | /** | |
471 | * fix_unclean_leb - fix an unclean LEB. | |
472 | * @c: UBIFS file-system description object | |
473 | * @sleb: scanned LEB information | |
474 | * @start: offset where scan started | |
475 | */ | |
476 | static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb, | |
477 | int start) | |
478 | { | |
479 | int lnum = sleb->lnum, endpt = start; | |
480 | ||
481 | /* Get the end offset of the last node we are keeping */ | |
482 | if (!list_empty(&sleb->nodes)) { | |
483 | struct ubifs_scan_node *snod; | |
484 | ||
485 | snod = list_entry(sleb->nodes.prev, | |
486 | struct ubifs_scan_node, list); | |
487 | endpt = snod->offs + snod->len; | |
488 | } | |
489 | ||
2ef13294 | 490 | if (c->ro_mount && !c->remounting_rw) { |
1e51764a AB |
491 | /* Add to recovery list */ |
492 | struct ubifs_unclean_leb *ucleb; | |
493 | ||
494 | dbg_rcvry("need to fix LEB %d start %d endpt %d", | |
495 | lnum, start, sleb->endpt); | |
496 | ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS); | |
497 | if (!ucleb) | |
498 | return -ENOMEM; | |
499 | ucleb->lnum = lnum; | |
500 | ucleb->endpt = endpt; | |
501 | list_add_tail(&ucleb->list, &c->unclean_leb_list); | |
502 | } else { | |
503 | /* Write the fixed LEB back to flash */ | |
504 | int err; | |
505 | ||
506 | dbg_rcvry("fixing LEB %d start %d endpt %d", | |
507 | lnum, start, sleb->endpt); | |
508 | if (endpt == 0) { | |
509 | err = ubifs_leb_unmap(c, lnum); | |
510 | if (err) | |
511 | return err; | |
512 | } else { | |
513 | int len = ALIGN(endpt, c->min_io_size); | |
514 | ||
515 | if (start) { | |
516 | err = ubi_read(c->ubi, lnum, sleb->buf, 0, | |
517 | start); | |
518 | if (err) | |
519 | return err; | |
520 | } | |
521 | /* Pad to min_io_size */ | |
522 | if (len > endpt) { | |
523 | int pad_len = len - ALIGN(endpt, 8); | |
524 | ||
525 | if (pad_len > 0) { | |
526 | void *buf = sleb->buf + len - pad_len; | |
527 | ||
528 | ubifs_pad(c, buf, pad_len); | |
529 | } | |
530 | } | |
531 | err = ubi_leb_change(c->ubi, lnum, sleb->buf, len, | |
532 | UBI_UNKNOWN); | |
533 | if (err) | |
534 | return err; | |
535 | } | |
536 | } | |
537 | return 0; | |
538 | } | |
539 | ||
540 | /** | |
541 | * drop_incomplete_group - drop nodes from an incomplete group. | |
542 | * @sleb: scanned LEB information | |
543 | * @offs: offset of dropped nodes is returned here | |
544 | * | |
545 | * This function returns %1 if nodes are dropped and %0 otherwise. | |
546 | */ | |
547 | static int drop_incomplete_group(struct ubifs_scan_leb *sleb, int *offs) | |
548 | { | |
549 | int dropped = 0; | |
550 | ||
551 | while (!list_empty(&sleb->nodes)) { | |
552 | struct ubifs_scan_node *snod; | |
553 | struct ubifs_ch *ch; | |
554 | ||
555 | snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node, | |
556 | list); | |
557 | ch = snod->node; | |
558 | if (ch->group_type != UBIFS_IN_NODE_GROUP) | |
559 | return dropped; | |
560 | dbg_rcvry("dropping node at %d:%d", sleb->lnum, snod->offs); | |
561 | *offs = snod->offs; | |
562 | list_del(&snod->list); | |
563 | kfree(snod); | |
564 | sleb->nodes_cnt -= 1; | |
565 | dropped = 1; | |
566 | } | |
567 | return dropped; | |
568 | } | |
569 | ||
570 | /** | |
571 | * ubifs_recover_leb - scan and recover a LEB. | |
572 | * @c: UBIFS file-system description object | |
573 | * @lnum: LEB number | |
574 | * @offs: offset | |
575 | * @sbuf: LEB-sized buffer to use | |
576 | * @grouped: nodes may be grouped for recovery | |
577 | * | |
578 | * This function does a scan of a LEB, but caters for errors that might have | |
579 | * been caused by the unclean unmount from which we are attempting to recover. | |
ed43f2f0 AB |
580 | * Returns %0 in case of success, %-EUCLEAN if an unrecoverable corruption is |
581 | * found, and a negative error code in case of failure. | |
1e51764a AB |
582 | */ |
583 | struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum, | |
584 | int offs, void *sbuf, int grouped) | |
585 | { | |
586 | int err, len = c->leb_size - offs, need_clean = 0, quiet = 1; | |
587 | int empty_chkd = 0, start = offs; | |
588 | struct ubifs_scan_leb *sleb; | |
589 | void *buf = sbuf + offs; | |
590 | ||
591 | dbg_rcvry("%d:%d", lnum, offs); | |
592 | ||
593 | sleb = ubifs_start_scan(c, lnum, offs, sbuf); | |
594 | if (IS_ERR(sleb)) | |
595 | return sleb; | |
596 | ||
597 | if (sleb->ecc) | |
598 | need_clean = 1; | |
599 | ||
600 | while (len >= 8) { | |
601 | int ret; | |
602 | ||
603 | dbg_scan("look at LEB %d:%d (%d bytes left)", | |
604 | lnum, offs, len); | |
605 | ||
606 | cond_resched(); | |
607 | ||
608 | /* | |
609 | * Scan quietly until there is an error from which we cannot | |
610 | * recover | |
611 | */ | |
612 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet); | |
613 | ||
614 | if (ret == SCANNED_A_NODE) { | |
615 | /* A valid node, and not a padding node */ | |
616 | struct ubifs_ch *ch = buf; | |
617 | int node_len; | |
618 | ||
619 | err = ubifs_add_snod(c, sleb, buf, offs); | |
620 | if (err) | |
621 | goto error; | |
622 | node_len = ALIGN(le32_to_cpu(ch->len), 8); | |
623 | offs += node_len; | |
624 | buf += node_len; | |
625 | len -= node_len; | |
626 | continue; | |
627 | } | |
628 | ||
629 | if (ret > 0) { | |
630 | /* Padding bytes or a valid padding node */ | |
631 | offs += ret; | |
632 | buf += ret; | |
633 | len -= ret; | |
634 | continue; | |
635 | } | |
636 | ||
637 | if (ret == SCANNED_EMPTY_SPACE) { | |
638 | if (!is_empty(buf, len)) { | |
639 | if (!is_last_write(c, buf, offs)) | |
640 | break; | |
641 | clean_buf(c, &buf, lnum, &offs, &len); | |
642 | need_clean = 1; | |
643 | } | |
644 | empty_chkd = 1; | |
645 | break; | |
646 | } | |
647 | ||
648 | if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE) | |
649 | if (is_last_write(c, buf, offs)) { | |
650 | clean_buf(c, &buf, lnum, &offs, &len); | |
651 | need_clean = 1; | |
652 | empty_chkd = 1; | |
653 | break; | |
654 | } | |
655 | ||
656 | if (ret == SCANNED_A_CORRUPT_NODE) | |
657 | if (no_more_nodes(c, buf, len, lnum, offs)) { | |
658 | clean_buf(c, &buf, lnum, &offs, &len); | |
659 | need_clean = 1; | |
660 | empty_chkd = 1; | |
661 | break; | |
662 | } | |
663 | ||
664 | if (quiet) { | |
665 | /* Redo the last scan but noisily */ | |
666 | quiet = 0; | |
667 | continue; | |
668 | } | |
669 | ||
670 | switch (ret) { | |
671 | case SCANNED_GARBAGE: | |
672 | dbg_err("garbage"); | |
673 | goto corrupted; | |
674 | case SCANNED_A_CORRUPT_NODE: | |
675 | case SCANNED_A_BAD_PAD_NODE: | |
676 | dbg_err("bad node"); | |
677 | goto corrupted; | |
678 | default: | |
679 | dbg_err("unknown"); | |
ed43f2f0 AB |
680 | err = -EINVAL; |
681 | goto error; | |
1e51764a AB |
682 | } |
683 | } | |
684 | ||
685 | if (!empty_chkd && !is_empty(buf, len)) { | |
686 | if (is_last_write(c, buf, offs)) { | |
687 | clean_buf(c, &buf, lnum, &offs, &len); | |
688 | need_clean = 1; | |
689 | } else { | |
06112547 AB |
690 | int corruption = first_non_ff(buf, len); |
691 | ||
be7b42a5 AB |
692 | /* |
693 | * See header comment for this file for more | |
694 | * explanations about the reasons we have this check. | |
695 | */ | |
06112547 AB |
696 | ubifs_err("corrupt empty space LEB %d:%d, corruption " |
697 | "starts at %d", lnum, offs, corruption); | |
698 | /* Make sure we dump interesting non-0xFF data */ | |
10ac2797 | 699 | offs += corruption; |
06112547 | 700 | buf += corruption; |
1e51764a AB |
701 | goto corrupted; |
702 | } | |
703 | } | |
704 | ||
705 | /* Drop nodes from incomplete group */ | |
706 | if (grouped && drop_incomplete_group(sleb, &offs)) { | |
707 | buf = sbuf + offs; | |
708 | len = c->leb_size - offs; | |
709 | clean_buf(c, &buf, lnum, &offs, &len); | |
710 | need_clean = 1; | |
711 | } | |
712 | ||
713 | if (offs % c->min_io_size) { | |
714 | clean_buf(c, &buf, lnum, &offs, &len); | |
715 | need_clean = 1; | |
716 | } | |
717 | ||
718 | ubifs_end_scan(c, sleb, lnum, offs); | |
719 | ||
720 | if (need_clean) { | |
721 | err = fix_unclean_leb(c, sleb, start); | |
722 | if (err) | |
723 | goto error; | |
724 | } | |
725 | ||
726 | return sleb; | |
727 | ||
728 | corrupted: | |
729 | ubifs_scanned_corruption(c, lnum, offs, buf); | |
730 | err = -EUCLEAN; | |
731 | error: | |
732 | ubifs_err("LEB %d scanning failed", lnum); | |
733 | ubifs_scan_destroy(sleb); | |
734 | return ERR_PTR(err); | |
735 | } | |
736 | ||
737 | /** | |
738 | * get_cs_sqnum - get commit start sequence number. | |
739 | * @c: UBIFS file-system description object | |
740 | * @lnum: LEB number of commit start node | |
741 | * @offs: offset of commit start node | |
742 | * @cs_sqnum: commit start sequence number is returned here | |
743 | * | |
744 | * This function returns %0 on success and a negative error code on failure. | |
745 | */ | |
746 | static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs, | |
747 | unsigned long long *cs_sqnum) | |
748 | { | |
749 | struct ubifs_cs_node *cs_node = NULL; | |
750 | int err, ret; | |
751 | ||
752 | dbg_rcvry("at %d:%d", lnum, offs); | |
753 | cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL); | |
754 | if (!cs_node) | |
755 | return -ENOMEM; | |
756 | if (c->leb_size - offs < UBIFS_CS_NODE_SZ) | |
757 | goto out_err; | |
758 | err = ubi_read(c->ubi, lnum, (void *)cs_node, offs, UBIFS_CS_NODE_SZ); | |
759 | if (err && err != -EBADMSG) | |
760 | goto out_free; | |
761 | ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0); | |
762 | if (ret != SCANNED_A_NODE) { | |
763 | dbg_err("Not a valid node"); | |
764 | goto out_err; | |
765 | } | |
766 | if (cs_node->ch.node_type != UBIFS_CS_NODE) { | |
767 | dbg_err("Node a CS node, type is %d", cs_node->ch.node_type); | |
768 | goto out_err; | |
769 | } | |
770 | if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) { | |
771 | dbg_err("CS node cmt_no %llu != current cmt_no %llu", | |
772 | (unsigned long long)le64_to_cpu(cs_node->cmt_no), | |
773 | c->cmt_no); | |
774 | goto out_err; | |
775 | } | |
776 | *cs_sqnum = le64_to_cpu(cs_node->ch.sqnum); | |
777 | dbg_rcvry("commit start sqnum %llu", *cs_sqnum); | |
778 | kfree(cs_node); | |
779 | return 0; | |
780 | ||
781 | out_err: | |
782 | err = -EINVAL; | |
783 | out_free: | |
784 | ubifs_err("failed to get CS sqnum"); | |
785 | kfree(cs_node); | |
786 | return err; | |
787 | } | |
788 | ||
789 | /** | |
790 | * ubifs_recover_log_leb - scan and recover a log LEB. | |
791 | * @c: UBIFS file-system description object | |
792 | * @lnum: LEB number | |
793 | * @offs: offset | |
794 | * @sbuf: LEB-sized buffer to use | |
795 | * | |
796 | * This function does a scan of a LEB, but caters for errors that might have | |
7d08ae3c AB |
797 | * been caused by unclean reboots from which we are attempting to recover |
798 | * (assume that only the last log LEB can be corrupted by an unclean reboot). | |
1e51764a AB |
799 | * |
800 | * This function returns %0 on success and a negative error code on failure. | |
801 | */ | |
802 | struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum, | |
803 | int offs, void *sbuf) | |
804 | { | |
805 | struct ubifs_scan_leb *sleb; | |
806 | int next_lnum; | |
807 | ||
808 | dbg_rcvry("LEB %d", lnum); | |
809 | next_lnum = lnum + 1; | |
810 | if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs) | |
811 | next_lnum = UBIFS_LOG_LNUM; | |
812 | if (next_lnum != c->ltail_lnum) { | |
813 | /* | |
814 | * We can only recover at the end of the log, so check that the | |
815 | * next log LEB is empty or out of date. | |
816 | */ | |
348709ba | 817 | sleb = ubifs_scan(c, next_lnum, 0, sbuf, 0); |
1e51764a AB |
818 | if (IS_ERR(sleb)) |
819 | return sleb; | |
820 | if (sleb->nodes_cnt) { | |
821 | struct ubifs_scan_node *snod; | |
822 | unsigned long long cs_sqnum = c->cs_sqnum; | |
823 | ||
824 | snod = list_entry(sleb->nodes.next, | |
825 | struct ubifs_scan_node, list); | |
826 | if (cs_sqnum == 0) { | |
827 | int err; | |
828 | ||
829 | err = get_cs_sqnum(c, lnum, offs, &cs_sqnum); | |
830 | if (err) { | |
831 | ubifs_scan_destroy(sleb); | |
832 | return ERR_PTR(err); | |
833 | } | |
834 | } | |
835 | if (snod->sqnum > cs_sqnum) { | |
836 | ubifs_err("unrecoverable log corruption " | |
837 | "in LEB %d", lnum); | |
838 | ubifs_scan_destroy(sleb); | |
839 | return ERR_PTR(-EUCLEAN); | |
840 | } | |
841 | } | |
842 | ubifs_scan_destroy(sleb); | |
843 | } | |
844 | return ubifs_recover_leb(c, lnum, offs, sbuf, 0); | |
845 | } | |
846 | ||
847 | /** | |
848 | * recover_head - recover a head. | |
849 | * @c: UBIFS file-system description object | |
850 | * @lnum: LEB number of head to recover | |
851 | * @offs: offset of head to recover | |
852 | * @sbuf: LEB-sized buffer to use | |
853 | * | |
854 | * This function ensures that there is no data on the flash at a head location. | |
855 | * | |
856 | * This function returns %0 on success and a negative error code on failure. | |
857 | */ | |
858 | static int recover_head(const struct ubifs_info *c, int lnum, int offs, | |
859 | void *sbuf) | |
860 | { | |
2765df7d | 861 | int len = c->max_write_size, err; |
1e51764a | 862 | |
1e51764a AB |
863 | if (offs + len > c->leb_size) |
864 | len = c->leb_size - offs; | |
865 | ||
866 | if (!len) | |
867 | return 0; | |
868 | ||
869 | /* Read at the head location and check it is empty flash */ | |
870 | err = ubi_read(c->ubi, lnum, sbuf, offs, len); | |
431102fe | 871 | if (err || !is_empty(sbuf, len)) { |
1e51764a AB |
872 | dbg_rcvry("cleaning head at %d:%d", lnum, offs); |
873 | if (offs == 0) | |
874 | return ubifs_leb_unmap(c, lnum); | |
875 | err = ubi_read(c->ubi, lnum, sbuf, 0, offs); | |
876 | if (err) | |
877 | return err; | |
878 | return ubi_leb_change(c->ubi, lnum, sbuf, offs, UBI_UNKNOWN); | |
879 | } | |
880 | ||
881 | return 0; | |
882 | } | |
883 | ||
884 | /** | |
885 | * ubifs_recover_inl_heads - recover index and LPT heads. | |
886 | * @c: UBIFS file-system description object | |
887 | * @sbuf: LEB-sized buffer to use | |
888 | * | |
889 | * This function ensures that there is no data on the flash at the index and | |
890 | * LPT head locations. | |
891 | * | |
892 | * This deals with the recovery of a half-completed journal commit. UBIFS is | |
893 | * careful never to overwrite the last version of the index or the LPT. Because | |
894 | * the index and LPT are wandering trees, data from a half-completed commit will | |
895 | * not be referenced anywhere in UBIFS. The data will be either in LEBs that are | |
896 | * assumed to be empty and will be unmapped anyway before use, or in the index | |
897 | * and LPT heads. | |
898 | * | |
899 | * This function returns %0 on success and a negative error code on failure. | |
900 | */ | |
901 | int ubifs_recover_inl_heads(const struct ubifs_info *c, void *sbuf) | |
902 | { | |
903 | int err; | |
904 | ||
2ef13294 | 905 | ubifs_assert(!c->ro_mount || c->remounting_rw); |
1e51764a AB |
906 | |
907 | dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs); | |
908 | err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf); | |
909 | if (err) | |
910 | return err; | |
911 | ||
912 | dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs); | |
913 | err = recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf); | |
914 | if (err) | |
915 | return err; | |
916 | ||
917 | return 0; | |
918 | } | |
919 | ||
920 | /** | |
921 | * clean_an_unclean_leb - read and write a LEB to remove corruption. | |
922 | * @c: UBIFS file-system description object | |
923 | * @ucleb: unclean LEB information | |
924 | * @sbuf: LEB-sized buffer to use | |
925 | * | |
926 | * This function reads a LEB up to a point pre-determined by the mount recovery, | |
927 | * checks the nodes, and writes the result back to the flash, thereby cleaning | |
928 | * off any following corruption, or non-fatal ECC errors. | |
929 | * | |
930 | * This function returns %0 on success and a negative error code on failure. | |
931 | */ | |
932 | static int clean_an_unclean_leb(const struct ubifs_info *c, | |
933 | struct ubifs_unclean_leb *ucleb, void *sbuf) | |
934 | { | |
935 | int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1; | |
936 | void *buf = sbuf; | |
937 | ||
938 | dbg_rcvry("LEB %d len %d", lnum, len); | |
939 | ||
940 | if (len == 0) { | |
941 | /* Nothing to read, just unmap it */ | |
942 | err = ubifs_leb_unmap(c, lnum); | |
943 | if (err) | |
944 | return err; | |
945 | return 0; | |
946 | } | |
947 | ||
948 | err = ubi_read(c->ubi, lnum, buf, offs, len); | |
949 | if (err && err != -EBADMSG) | |
950 | return err; | |
951 | ||
952 | while (len >= 8) { | |
953 | int ret; | |
954 | ||
955 | cond_resched(); | |
956 | ||
957 | /* Scan quietly until there is an error */ | |
958 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet); | |
959 | ||
960 | if (ret == SCANNED_A_NODE) { | |
961 | /* A valid node, and not a padding node */ | |
962 | struct ubifs_ch *ch = buf; | |
963 | int node_len; | |
964 | ||
965 | node_len = ALIGN(le32_to_cpu(ch->len), 8); | |
966 | offs += node_len; | |
967 | buf += node_len; | |
968 | len -= node_len; | |
969 | continue; | |
970 | } | |
971 | ||
972 | if (ret > 0) { | |
973 | /* Padding bytes or a valid padding node */ | |
974 | offs += ret; | |
975 | buf += ret; | |
976 | len -= ret; | |
977 | continue; | |
978 | } | |
979 | ||
980 | if (ret == SCANNED_EMPTY_SPACE) { | |
981 | ubifs_err("unexpected empty space at %d:%d", | |
982 | lnum, offs); | |
983 | return -EUCLEAN; | |
984 | } | |
985 | ||
986 | if (quiet) { | |
987 | /* Redo the last scan but noisily */ | |
988 | quiet = 0; | |
989 | continue; | |
990 | } | |
991 | ||
992 | ubifs_scanned_corruption(c, lnum, offs, buf); | |
993 | return -EUCLEAN; | |
994 | } | |
995 | ||
996 | /* Pad to min_io_size */ | |
997 | len = ALIGN(ucleb->endpt, c->min_io_size); | |
998 | if (len > ucleb->endpt) { | |
999 | int pad_len = len - ALIGN(ucleb->endpt, 8); | |
1000 | ||
1001 | if (pad_len > 0) { | |
1002 | buf = c->sbuf + len - pad_len; | |
1003 | ubifs_pad(c, buf, pad_len); | |
1004 | } | |
1005 | } | |
1006 | ||
1007 | /* Write back the LEB atomically */ | |
1008 | err = ubi_leb_change(c->ubi, lnum, sbuf, len, UBI_UNKNOWN); | |
1009 | if (err) | |
1010 | return err; | |
1011 | ||
1012 | dbg_rcvry("cleaned LEB %d", lnum); | |
1013 | ||
1014 | return 0; | |
1015 | } | |
1016 | ||
1017 | /** | |
1018 | * ubifs_clean_lebs - clean LEBs recovered during read-only mount. | |
1019 | * @c: UBIFS file-system description object | |
1020 | * @sbuf: LEB-sized buffer to use | |
1021 | * | |
1022 | * This function cleans a LEB identified during recovery that needs to be | |
1023 | * written but was not because UBIFS was mounted read-only. This happens when | |
1024 | * remounting to read-write mode. | |
1025 | * | |
1026 | * This function returns %0 on success and a negative error code on failure. | |
1027 | */ | |
1028 | int ubifs_clean_lebs(const struct ubifs_info *c, void *sbuf) | |
1029 | { | |
1030 | dbg_rcvry("recovery"); | |
1031 | while (!list_empty(&c->unclean_leb_list)) { | |
1032 | struct ubifs_unclean_leb *ucleb; | |
1033 | int err; | |
1034 | ||
1035 | ucleb = list_entry(c->unclean_leb_list.next, | |
1036 | struct ubifs_unclean_leb, list); | |
1037 | err = clean_an_unclean_leb(c, ucleb, sbuf); | |
1038 | if (err) | |
1039 | return err; | |
1040 | list_del(&ucleb->list); | |
1041 | kfree(ucleb); | |
1042 | } | |
1043 | return 0; | |
1044 | } | |
1045 | ||
1046 | /** | |
1047 | * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit. | |
1048 | * @c: UBIFS file-system description object | |
1049 | * | |
1050 | * Out-of-place garbage collection requires always one empty LEB with which to | |
1051 | * start garbage collection. The LEB number is recorded in c->gc_lnum and is | |
1052 | * written to the master node on unmounting. In the case of an unclean unmount | |
1053 | * the value of gc_lnum recorded in the master node is out of date and cannot | |
1054 | * be used. Instead, recovery must allocate an empty LEB for this purpose. | |
1055 | * However, there may not be enough empty space, in which case it must be | |
1056 | * possible to GC the dirtiest LEB into the GC head LEB. | |
1057 | * | |
1058 | * This function also runs the commit which causes the TNC updates from | |
1059 | * size-recovery and orphans to be written to the flash. That is important to | |
1060 | * ensure correct replay order for subsequent mounts. | |
1061 | * | |
1062 | * This function returns %0 on success and a negative error code on failure. | |
1063 | */ | |
1064 | int ubifs_rcvry_gc_commit(struct ubifs_info *c) | |
1065 | { | |
1066 | struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf; | |
1067 | struct ubifs_lprops lp; | |
1068 | int lnum, err; | |
1069 | ||
1070 | c->gc_lnum = -1; | |
1071 | if (wbuf->lnum == -1) { | |
1072 | dbg_rcvry("no GC head LEB"); | |
1073 | goto find_free; | |
1074 | } | |
1075 | /* | |
1076 | * See whether the used space in the dirtiest LEB fits in the GC head | |
1077 | * LEB. | |
1078 | */ | |
1079 | if (wbuf->offs == c->leb_size) { | |
1080 | dbg_rcvry("no room in GC head LEB"); | |
1081 | goto find_free; | |
1082 | } | |
1083 | err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2); | |
1084 | if (err) { | |
6fb4374f AB |
1085 | /* |
1086 | * There are no dirty or empty LEBs subject to here being | |
1087 | * enough for the index. Try to use | |
1088 | * 'ubifs_find_free_leb_for_idx()', which will return any empty | |
1089 | * LEBs (ignoring index requirements). If the index then | |
1090 | * doesn't have enough LEBs the recovery commit will fail - | |
1091 | * which is the same result anyway i.e. recovery fails. So | |
1092 | * there is no problem ignoring index requirements and just | |
1093 | * grabbing a free LEB since we have already established there | |
1094 | * is not a dirty LEB we could have used instead. | |
1095 | */ | |
1096 | if (err == -ENOSPC) { | |
1097 | dbg_rcvry("could not find a dirty LEB"); | |
1098 | goto find_free; | |
1099 | } | |
1e51764a AB |
1100 | return err; |
1101 | } | |
1102 | ubifs_assert(!(lp.flags & LPROPS_INDEX)); | |
1103 | lnum = lp.lnum; | |
1104 | if (lp.free + lp.dirty == c->leb_size) { | |
1105 | /* An empty LEB was returned */ | |
1106 | if (lp.free != c->leb_size) { | |
1107 | err = ubifs_change_one_lp(c, lnum, c->leb_size, | |
1108 | 0, 0, 0, 0); | |
1109 | if (err) | |
1110 | return err; | |
1111 | } | |
1112 | err = ubifs_leb_unmap(c, lnum); | |
1113 | if (err) | |
1114 | return err; | |
1115 | c->gc_lnum = lnum; | |
1116 | dbg_rcvry("allocated LEB %d for GC", lnum); | |
1117 | /* Run the commit */ | |
1118 | dbg_rcvry("committing"); | |
1119 | return ubifs_run_commit(c); | |
1120 | } | |
1121 | /* | |
1122 | * There was no empty LEB so the used space in the dirtiest LEB must fit | |
1123 | * in the GC head LEB. | |
1124 | */ | |
1125 | if (lp.free + lp.dirty < wbuf->offs) { | |
1126 | dbg_rcvry("LEB %d doesn't fit in GC head LEB %d:%d", | |
1127 | lnum, wbuf->lnum, wbuf->offs); | |
1128 | err = ubifs_return_leb(c, lnum); | |
1129 | if (err) | |
1130 | return err; | |
1131 | goto find_free; | |
1132 | } | |
1133 | /* | |
1134 | * We run the commit before garbage collection otherwise subsequent | |
1135 | * mounts will see the GC and orphan deletion in a different order. | |
1136 | */ | |
1137 | dbg_rcvry("committing"); | |
1138 | err = ubifs_run_commit(c); | |
1139 | if (err) | |
1140 | return err; | |
1141 | /* | |
1142 | * The data in the dirtiest LEB fits in the GC head LEB, so do the GC | |
1143 | * - use locking to keep 'ubifs_assert()' happy. | |
1144 | */ | |
1145 | dbg_rcvry("GC'ing LEB %d", lnum); | |
1146 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
1147 | err = ubifs_garbage_collect_leb(c, &lp); | |
1148 | if (err >= 0) { | |
1149 | int err2 = ubifs_wbuf_sync_nolock(wbuf); | |
1150 | ||
1151 | if (err2) | |
1152 | err = err2; | |
1153 | } | |
1154 | mutex_unlock(&wbuf->io_mutex); | |
1155 | if (err < 0) { | |
1156 | dbg_err("GC failed, error %d", err); | |
1157 | if (err == -EAGAIN) | |
1158 | err = -EINVAL; | |
1159 | return err; | |
1160 | } | |
1161 | if (err != LEB_RETAINED) { | |
1162 | dbg_err("GC returned %d", err); | |
1163 | return -EINVAL; | |
1164 | } | |
1165 | err = ubifs_leb_unmap(c, c->gc_lnum); | |
1166 | if (err) | |
1167 | return err; | |
1168 | dbg_rcvry("allocated LEB %d for GC", lnum); | |
1169 | return 0; | |
1170 | ||
1171 | find_free: | |
1172 | /* | |
1173 | * There is no GC head LEB or the free space in the GC head LEB is too | |
6fb4374f AB |
1174 | * small, or there are not dirty LEBs. Allocate gc_lnum by calling |
1175 | * 'ubifs_find_free_leb_for_idx()' so GC is not run. | |
1e51764a AB |
1176 | */ |
1177 | lnum = ubifs_find_free_leb_for_idx(c); | |
1178 | if (lnum < 0) { | |
1179 | dbg_err("could not find an empty LEB"); | |
1180 | return lnum; | |
1181 | } | |
1182 | /* And reset the index flag */ | |
1183 | err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0, | |
1184 | LPROPS_INDEX, 0); | |
1185 | if (err) | |
1186 | return err; | |
1187 | c->gc_lnum = lnum; | |
1188 | dbg_rcvry("allocated LEB %d for GC", lnum); | |
1189 | /* Run the commit */ | |
1190 | dbg_rcvry("committing"); | |
1191 | return ubifs_run_commit(c); | |
1192 | } | |
1193 | ||
1194 | /** | |
1195 | * struct size_entry - inode size information for recovery. | |
1196 | * @rb: link in the RB-tree of sizes | |
1197 | * @inum: inode number | |
1198 | * @i_size: size on inode | |
1199 | * @d_size: maximum size based on data nodes | |
1200 | * @exists: indicates whether the inode exists | |
1201 | * @inode: inode if pinned in memory awaiting rw mode to fix it | |
1202 | */ | |
1203 | struct size_entry { | |
1204 | struct rb_node rb; | |
1205 | ino_t inum; | |
1206 | loff_t i_size; | |
1207 | loff_t d_size; | |
1208 | int exists; | |
1209 | struct inode *inode; | |
1210 | }; | |
1211 | ||
1212 | /** | |
1213 | * add_ino - add an entry to the size tree. | |
1214 | * @c: UBIFS file-system description object | |
1215 | * @inum: inode number | |
1216 | * @i_size: size on inode | |
1217 | * @d_size: maximum size based on data nodes | |
1218 | * @exists: indicates whether the inode exists | |
1219 | */ | |
1220 | static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size, | |
1221 | loff_t d_size, int exists) | |
1222 | { | |
1223 | struct rb_node **p = &c->size_tree.rb_node, *parent = NULL; | |
1224 | struct size_entry *e; | |
1225 | ||
1226 | while (*p) { | |
1227 | parent = *p; | |
1228 | e = rb_entry(parent, struct size_entry, rb); | |
1229 | if (inum < e->inum) | |
1230 | p = &(*p)->rb_left; | |
1231 | else | |
1232 | p = &(*p)->rb_right; | |
1233 | } | |
1234 | ||
1235 | e = kzalloc(sizeof(struct size_entry), GFP_KERNEL); | |
1236 | if (!e) | |
1237 | return -ENOMEM; | |
1238 | ||
1239 | e->inum = inum; | |
1240 | e->i_size = i_size; | |
1241 | e->d_size = d_size; | |
1242 | e->exists = exists; | |
1243 | ||
1244 | rb_link_node(&e->rb, parent, p); | |
1245 | rb_insert_color(&e->rb, &c->size_tree); | |
1246 | ||
1247 | return 0; | |
1248 | } | |
1249 | ||
1250 | /** | |
1251 | * find_ino - find an entry on the size tree. | |
1252 | * @c: UBIFS file-system description object | |
1253 | * @inum: inode number | |
1254 | */ | |
1255 | static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum) | |
1256 | { | |
1257 | struct rb_node *p = c->size_tree.rb_node; | |
1258 | struct size_entry *e; | |
1259 | ||
1260 | while (p) { | |
1261 | e = rb_entry(p, struct size_entry, rb); | |
1262 | if (inum < e->inum) | |
1263 | p = p->rb_left; | |
1264 | else if (inum > e->inum) | |
1265 | p = p->rb_right; | |
1266 | else | |
1267 | return e; | |
1268 | } | |
1269 | return NULL; | |
1270 | } | |
1271 | ||
1272 | /** | |
1273 | * remove_ino - remove an entry from the size tree. | |
1274 | * @c: UBIFS file-system description object | |
1275 | * @inum: inode number | |
1276 | */ | |
1277 | static void remove_ino(struct ubifs_info *c, ino_t inum) | |
1278 | { | |
1279 | struct size_entry *e = find_ino(c, inum); | |
1280 | ||
1281 | if (!e) | |
1282 | return; | |
1283 | rb_erase(&e->rb, &c->size_tree); | |
1284 | kfree(e); | |
1285 | } | |
1286 | ||
1287 | /** | |
1288 | * ubifs_destroy_size_tree - free resources related to the size tree. | |
1289 | * @c: UBIFS file-system description object | |
1290 | */ | |
1291 | void ubifs_destroy_size_tree(struct ubifs_info *c) | |
1292 | { | |
1293 | struct rb_node *this = c->size_tree.rb_node; | |
1294 | struct size_entry *e; | |
1295 | ||
1296 | while (this) { | |
1297 | if (this->rb_left) { | |
1298 | this = this->rb_left; | |
1299 | continue; | |
1300 | } else if (this->rb_right) { | |
1301 | this = this->rb_right; | |
1302 | continue; | |
1303 | } | |
1304 | e = rb_entry(this, struct size_entry, rb); | |
1305 | if (e->inode) | |
1306 | iput(e->inode); | |
1307 | this = rb_parent(this); | |
1308 | if (this) { | |
1309 | if (this->rb_left == &e->rb) | |
1310 | this->rb_left = NULL; | |
1311 | else | |
1312 | this->rb_right = NULL; | |
1313 | } | |
1314 | kfree(e); | |
1315 | } | |
1316 | c->size_tree = RB_ROOT; | |
1317 | } | |
1318 | ||
1319 | /** | |
1320 | * ubifs_recover_size_accum - accumulate inode sizes for recovery. | |
1321 | * @c: UBIFS file-system description object | |
1322 | * @key: node key | |
1323 | * @deletion: node is for a deletion | |
1324 | * @new_size: inode size | |
1325 | * | |
1326 | * This function has two purposes: | |
1327 | * 1) to ensure there are no data nodes that fall outside the inode size | |
1328 | * 2) to ensure there are no data nodes for inodes that do not exist | |
1329 | * To accomplish those purposes, a rb-tree is constructed containing an entry | |
1330 | * for each inode number in the journal that has not been deleted, and recording | |
1331 | * the size from the inode node, the maximum size of any data node (also altered | |
1332 | * by truncations) and a flag indicating a inode number for which no inode node | |
1333 | * was present in the journal. | |
1334 | * | |
1335 | * Note that there is still the possibility that there are data nodes that have | |
1336 | * been committed that are beyond the inode size, however the only way to find | |
1337 | * them would be to scan the entire index. Alternatively, some provision could | |
1338 | * be made to record the size of inodes at the start of commit, which would seem | |
1339 | * very cumbersome for a scenario that is quite unlikely and the only negative | |
1340 | * consequence of which is wasted space. | |
1341 | * | |
1342 | * This functions returns %0 on success and a negative error code on failure. | |
1343 | */ | |
1344 | int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key, | |
1345 | int deletion, loff_t new_size) | |
1346 | { | |
1347 | ino_t inum = key_inum(c, key); | |
1348 | struct size_entry *e; | |
1349 | int err; | |
1350 | ||
1351 | switch (key_type(c, key)) { | |
1352 | case UBIFS_INO_KEY: | |
1353 | if (deletion) | |
1354 | remove_ino(c, inum); | |
1355 | else { | |
1356 | e = find_ino(c, inum); | |
1357 | if (e) { | |
1358 | e->i_size = new_size; | |
1359 | e->exists = 1; | |
1360 | } else { | |
1361 | err = add_ino(c, inum, new_size, 0, 1); | |
1362 | if (err) | |
1363 | return err; | |
1364 | } | |
1365 | } | |
1366 | break; | |
1367 | case UBIFS_DATA_KEY: | |
1368 | e = find_ino(c, inum); | |
1369 | if (e) { | |
1370 | if (new_size > e->d_size) | |
1371 | e->d_size = new_size; | |
1372 | } else { | |
1373 | err = add_ino(c, inum, 0, new_size, 0); | |
1374 | if (err) | |
1375 | return err; | |
1376 | } | |
1377 | break; | |
1378 | case UBIFS_TRUN_KEY: | |
1379 | e = find_ino(c, inum); | |
1380 | if (e) | |
1381 | e->d_size = new_size; | |
1382 | break; | |
1383 | } | |
1384 | return 0; | |
1385 | } | |
1386 | ||
1387 | /** | |
1388 | * fix_size_in_place - fix inode size in place on flash. | |
1389 | * @c: UBIFS file-system description object | |
1390 | * @e: inode size information for recovery | |
1391 | */ | |
1392 | static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e) | |
1393 | { | |
1394 | struct ubifs_ino_node *ino = c->sbuf; | |
1395 | unsigned char *p; | |
1396 | union ubifs_key key; | |
1397 | int err, lnum, offs, len; | |
1398 | loff_t i_size; | |
1399 | uint32_t crc; | |
1400 | ||
1401 | /* Locate the inode node LEB number and offset */ | |
1402 | ino_key_init(c, &key, e->inum); | |
1403 | err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs); | |
1404 | if (err) | |
1405 | goto out; | |
1406 | /* | |
1407 | * If the size recorded on the inode node is greater than the size that | |
1408 | * was calculated from nodes in the journal then don't change the inode. | |
1409 | */ | |
1410 | i_size = le64_to_cpu(ino->size); | |
1411 | if (i_size >= e->d_size) | |
1412 | return 0; | |
1413 | /* Read the LEB */ | |
1414 | err = ubi_read(c->ubi, lnum, c->sbuf, 0, c->leb_size); | |
1415 | if (err) | |
1416 | goto out; | |
1417 | /* Change the size field and recalculate the CRC */ | |
1418 | ino = c->sbuf + offs; | |
1419 | ino->size = cpu_to_le64(e->d_size); | |
1420 | len = le32_to_cpu(ino->ch.len); | |
1421 | crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8); | |
1422 | ino->ch.crc = cpu_to_le32(crc); | |
1423 | /* Work out where data in the LEB ends and free space begins */ | |
1424 | p = c->sbuf; | |
1425 | len = c->leb_size - 1; | |
1426 | while (p[len] == 0xff) | |
1427 | len -= 1; | |
1428 | len = ALIGN(len + 1, c->min_io_size); | |
1429 | /* Atomically write the fixed LEB back again */ | |
1430 | err = ubi_leb_change(c->ubi, lnum, c->sbuf, len, UBI_UNKNOWN); | |
1431 | if (err) | |
1432 | goto out; | |
e84461ad AB |
1433 | dbg_rcvry("inode %lu at %d:%d size %lld -> %lld ", |
1434 | (unsigned long)e->inum, lnum, offs, i_size, e->d_size); | |
1e51764a AB |
1435 | return 0; |
1436 | ||
1437 | out: | |
1438 | ubifs_warn("inode %lu failed to fix size %lld -> %lld error %d", | |
e84461ad | 1439 | (unsigned long)e->inum, e->i_size, e->d_size, err); |
1e51764a AB |
1440 | return err; |
1441 | } | |
1442 | ||
1443 | /** | |
1444 | * ubifs_recover_size - recover inode size. | |
1445 | * @c: UBIFS file-system description object | |
1446 | * | |
1447 | * This function attempts to fix inode size discrepancies identified by the | |
1448 | * 'ubifs_recover_size_accum()' function. | |
1449 | * | |
1450 | * This functions returns %0 on success and a negative error code on failure. | |
1451 | */ | |
1452 | int ubifs_recover_size(struct ubifs_info *c) | |
1453 | { | |
1454 | struct rb_node *this = rb_first(&c->size_tree); | |
1455 | ||
1456 | while (this) { | |
1457 | struct size_entry *e; | |
1458 | int err; | |
1459 | ||
1460 | e = rb_entry(this, struct size_entry, rb); | |
1461 | if (!e->exists) { | |
1462 | union ubifs_key key; | |
1463 | ||
1464 | ino_key_init(c, &key, e->inum); | |
1465 | err = ubifs_tnc_lookup(c, &key, c->sbuf); | |
1466 | if (err && err != -ENOENT) | |
1467 | return err; | |
1468 | if (err == -ENOENT) { | |
1469 | /* Remove data nodes that have no inode */ | |
e84461ad AB |
1470 | dbg_rcvry("removing ino %lu", |
1471 | (unsigned long)e->inum); | |
1e51764a AB |
1472 | err = ubifs_tnc_remove_ino(c, e->inum); |
1473 | if (err) | |
1474 | return err; | |
1475 | } else { | |
1476 | struct ubifs_ino_node *ino = c->sbuf; | |
1477 | ||
1478 | e->exists = 1; | |
1479 | e->i_size = le64_to_cpu(ino->size); | |
1480 | } | |
1481 | } | |
1482 | if (e->exists && e->i_size < e->d_size) { | |
2ef13294 | 1483 | if (!e->inode && c->ro_mount) { |
1e51764a AB |
1484 | /* Fix the inode size and pin it in memory */ |
1485 | struct inode *inode; | |
1486 | ||
1487 | inode = ubifs_iget(c->vfs_sb, e->inum); | |
1488 | if (IS_ERR(inode)) | |
1489 | return PTR_ERR(inode); | |
1490 | if (inode->i_size < e->d_size) { | |
1491 | dbg_rcvry("ino %lu size %lld -> %lld", | |
e84461ad AB |
1492 | (unsigned long)e->inum, |
1493 | e->d_size, inode->i_size); | |
1e51764a AB |
1494 | inode->i_size = e->d_size; |
1495 | ubifs_inode(inode)->ui_size = e->d_size; | |
1496 | e->inode = inode; | |
1497 | this = rb_next(this); | |
1498 | continue; | |
1499 | } | |
1500 | iput(inode); | |
1501 | } else { | |
1502 | /* Fix the size in place */ | |
1503 | err = fix_size_in_place(c, e); | |
1504 | if (err) | |
1505 | return err; | |
1506 | if (e->inode) | |
1507 | iput(e->inode); | |
1508 | } | |
1509 | } | |
1510 | this = rb_next(this); | |
1511 | rb_erase(&e->rb, &c->size_tree); | |
1512 | kfree(e); | |
1513 | } | |
1514 | return 0; | |
1515 | } |