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