]>
Commit | Line | Data |
---|---|---|
9eefe2a2 SR |
1 | /* |
2 | * This file is part of UBIFS. | |
3 | * | |
4 | * Copyright (C) 2006-2008 Nokia Corporation | |
5 | * | |
ff94bc40 | 6 | * SPDX-License-Identifier: GPL-2.0+ |
9eefe2a2 SR |
7 | * |
8 | * Authors: Adrian Hunter | |
9 | * Artem Bityutskiy (Битюцкий Артём) | |
10 | */ | |
11 | ||
12 | /* | |
13 | * This file implements functions needed to recover from unclean un-mounts. | |
14 | * When UBIFS is mounted, it checks a flag on the master node to determine if | |
ff94bc40 HS |
15 | * an un-mount was completed successfully. If not, the process of mounting |
16 | * incorporates additional checking and fixing of on-flash data structures. | |
9eefe2a2 SR |
17 | * UBIFS always cleans away all remnants of an unclean un-mount, so that |
18 | * errors do not accumulate. However UBIFS defers recovery if it is mounted | |
19 | * read-only, and the flash is not modified in that case. | |
ff94bc40 HS |
20 | * |
21 | * The general UBIFS approach to the recovery is that it recovers from | |
22 | * corruptions which could be caused by power cuts, but it refuses to recover | |
23 | * from corruption caused by other reasons. And UBIFS tries to distinguish | |
24 | * between these 2 reasons of corruptions and silently recover in the former | |
25 | * case and loudly complain in the latter case. | |
26 | * | |
27 | * UBIFS writes only to erased LEBs, so it writes only to the flash space | |
28 | * containing only 0xFFs. UBIFS also always writes strictly from the beginning | |
29 | * of the LEB to the end. And UBIFS assumes that the underlying flash media | |
30 | * writes in @c->max_write_size bytes at a time. | |
31 | * | |
32 | * Hence, if UBIFS finds a corrupted node at offset X, it expects only the min. | |
33 | * I/O unit corresponding to offset X to contain corrupted data, all the | |
34 | * following min. I/O units have to contain empty space (all 0xFFs). If this is | |
35 | * not true, the corruption cannot be the result of a power cut, and UBIFS | |
36 | * refuses to mount. | |
9eefe2a2 SR |
37 | */ |
38 | ||
ff94bc40 HS |
39 | #define __UBOOT__ |
40 | #ifndef __UBOOT__ | |
41 | #include <linux/crc32.h> | |
42 | #include <linux/slab.h> | |
43 | #else | |
44 | #include <linux/err.h> | |
45 | #endif | |
9eefe2a2 SR |
46 | #include "ubifs.h" |
47 | ||
48 | /** | |
49 | * is_empty - determine whether a buffer is empty (contains all 0xff). | |
50 | * @buf: buffer to clean | |
51 | * @len: length of buffer | |
52 | * | |
53 | * This function returns %1 if the buffer is empty (contains all 0xff) otherwise | |
54 | * %0 is returned. | |
55 | */ | |
56 | static int is_empty(void *buf, int len) | |
57 | { | |
58 | uint8_t *p = buf; | |
59 | int i; | |
60 | ||
61 | for (i = 0; i < len; i++) | |
62 | if (*p++ != 0xff) | |
63 | return 0; | |
64 | return 1; | |
65 | } | |
66 | ||
ff94bc40 HS |
67 | /** |
68 | * first_non_ff - find offset of the first non-0xff byte. | |
69 | * @buf: buffer to search in | |
70 | * @len: length of buffer | |
71 | * | |
72 | * This function returns offset of the first non-0xff byte in @buf or %-1 if | |
73 | * the buffer contains only 0xff bytes. | |
74 | */ | |
75 | static int first_non_ff(void *buf, int len) | |
76 | { | |
77 | uint8_t *p = buf; | |
78 | int i; | |
79 | ||
80 | for (i = 0; i < len; i++) | |
81 | if (*p++ != 0xff) | |
82 | return i; | |
83 | return -1; | |
84 | } | |
85 | ||
9eefe2a2 SR |
86 | /** |
87 | * get_master_node - get the last valid master node allowing for corruption. | |
88 | * @c: UBIFS file-system description object | |
89 | * @lnum: LEB number | |
90 | * @pbuf: buffer containing the LEB read, is returned here | |
91 | * @mst: master node, if found, is returned here | |
92 | * @cor: corruption, if found, is returned here | |
93 | * | |
94 | * This function allocates a buffer, reads the LEB into it, and finds and | |
95 | * returns the last valid master node allowing for one area of corruption. | |
96 | * The corrupt area, if there is one, must be consistent with the assumption | |
97 | * that it is the result of an unclean unmount while the master node was being | |
98 | * written. Under those circumstances, it is valid to use the previously written | |
99 | * master node. | |
100 | * | |
101 | * This function returns %0 on success and a negative error code on failure. | |
102 | */ | |
103 | static int get_master_node(const struct ubifs_info *c, int lnum, void **pbuf, | |
104 | struct ubifs_mst_node **mst, void **cor) | |
105 | { | |
106 | const int sz = c->mst_node_alsz; | |
107 | int err, offs, len; | |
108 | void *sbuf, *buf; | |
109 | ||
110 | sbuf = vmalloc(c->leb_size); | |
111 | if (!sbuf) | |
112 | return -ENOMEM; | |
113 | ||
ff94bc40 | 114 | err = ubifs_leb_read(c, lnum, sbuf, 0, c->leb_size, 0); |
9eefe2a2 SR |
115 | if (err && err != -EBADMSG) |
116 | goto out_free; | |
117 | ||
118 | /* Find the first position that is definitely not a node */ | |
119 | offs = 0; | |
120 | buf = sbuf; | |
121 | len = c->leb_size; | |
122 | while (offs + UBIFS_MST_NODE_SZ <= c->leb_size) { | |
123 | struct ubifs_ch *ch = buf; | |
124 | ||
125 | if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC) | |
126 | break; | |
127 | offs += sz; | |
128 | buf += sz; | |
129 | len -= sz; | |
130 | } | |
131 | /* See if there was a valid master node before that */ | |
132 | if (offs) { | |
133 | int ret; | |
134 | ||
135 | offs -= sz; | |
136 | buf -= sz; | |
137 | len += sz; | |
138 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1); | |
139 | if (ret != SCANNED_A_NODE && offs) { | |
140 | /* Could have been corruption so check one place back */ | |
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) | |
146 | /* | |
147 | * We accept only one area of corruption because | |
148 | * we are assuming that it was caused while | |
149 | * trying to write a master node. | |
150 | */ | |
151 | goto out_err; | |
152 | } | |
153 | if (ret == SCANNED_A_NODE) { | |
154 | struct ubifs_ch *ch = buf; | |
155 | ||
156 | if (ch->node_type != UBIFS_MST_NODE) | |
157 | goto out_err; | |
158 | dbg_rcvry("found a master node at %d:%d", lnum, offs); | |
159 | *mst = buf; | |
160 | offs += sz; | |
161 | buf += sz; | |
162 | len -= sz; | |
163 | } | |
164 | } | |
165 | /* Check for corruption */ | |
166 | if (offs < c->leb_size) { | |
167 | if (!is_empty(buf, min_t(int, len, sz))) { | |
168 | *cor = buf; | |
169 | dbg_rcvry("found corruption at %d:%d", lnum, offs); | |
170 | } | |
171 | offs += sz; | |
172 | buf += sz; | |
173 | len -= sz; | |
174 | } | |
175 | /* Check remaining empty space */ | |
176 | if (offs < c->leb_size) | |
177 | if (!is_empty(buf, len)) | |
178 | goto out_err; | |
179 | *pbuf = sbuf; | |
180 | return 0; | |
181 | ||
182 | out_err: | |
183 | err = -EINVAL; | |
184 | out_free: | |
185 | vfree(sbuf); | |
186 | *mst = NULL; | |
187 | *cor = NULL; | |
188 | return err; | |
189 | } | |
190 | ||
191 | /** | |
192 | * write_rcvrd_mst_node - write recovered master node. | |
193 | * @c: UBIFS file-system description object | |
194 | * @mst: master node | |
195 | * | |
196 | * This function returns %0 on success and a negative error code on failure. | |
197 | */ | |
198 | static int write_rcvrd_mst_node(struct ubifs_info *c, | |
199 | struct ubifs_mst_node *mst) | |
200 | { | |
201 | int err = 0, lnum = UBIFS_MST_LNUM, sz = c->mst_node_alsz; | |
202 | __le32 save_flags; | |
203 | ||
204 | dbg_rcvry("recovery"); | |
205 | ||
206 | save_flags = mst->flags; | |
207 | mst->flags |= cpu_to_le32(UBIFS_MST_RCVRY); | |
208 | ||
209 | ubifs_prepare_node(c, mst, UBIFS_MST_NODE_SZ, 1); | |
ff94bc40 | 210 | err = ubifs_leb_change(c, lnum, mst, sz); |
9eefe2a2 SR |
211 | if (err) |
212 | goto out; | |
ff94bc40 | 213 | err = ubifs_leb_change(c, lnum + 1, mst, sz); |
9eefe2a2 SR |
214 | if (err) |
215 | goto out; | |
216 | out: | |
217 | mst->flags = save_flags; | |
218 | return err; | |
219 | } | |
220 | ||
221 | /** | |
222 | * ubifs_recover_master_node - recover the master node. | |
223 | * @c: UBIFS file-system description object | |
224 | * | |
225 | * This function recovers the master node from corruption that may occur due to | |
226 | * an unclean unmount. | |
227 | * | |
228 | * This function returns %0 on success and a negative error code on failure. | |
229 | */ | |
230 | int ubifs_recover_master_node(struct ubifs_info *c) | |
231 | { | |
232 | void *buf1 = NULL, *buf2 = NULL, *cor1 = NULL, *cor2 = NULL; | |
233 | struct ubifs_mst_node *mst1 = NULL, *mst2 = NULL, *mst; | |
234 | const int sz = c->mst_node_alsz; | |
235 | int err, offs1, offs2; | |
236 | ||
237 | dbg_rcvry("recovery"); | |
238 | ||
239 | err = get_master_node(c, UBIFS_MST_LNUM, &buf1, &mst1, &cor1); | |
240 | if (err) | |
241 | goto out_free; | |
242 | ||
243 | err = get_master_node(c, UBIFS_MST_LNUM + 1, &buf2, &mst2, &cor2); | |
244 | if (err) | |
245 | goto out_free; | |
246 | ||
247 | if (mst1) { | |
248 | offs1 = (void *)mst1 - buf1; | |
249 | if ((le32_to_cpu(mst1->flags) & UBIFS_MST_RCVRY) && | |
250 | (offs1 == 0 && !cor1)) { | |
251 | /* | |
252 | * mst1 was written by recovery at offset 0 with no | |
253 | * corruption. | |
254 | */ | |
255 | dbg_rcvry("recovery recovery"); | |
256 | mst = mst1; | |
257 | } else if (mst2) { | |
258 | offs2 = (void *)mst2 - buf2; | |
259 | if (offs1 == offs2) { | |
260 | /* Same offset, so must be the same */ | |
261 | if (memcmp((void *)mst1 + UBIFS_CH_SZ, | |
262 | (void *)mst2 + UBIFS_CH_SZ, | |
263 | UBIFS_MST_NODE_SZ - UBIFS_CH_SZ)) | |
264 | goto out_err; | |
265 | mst = mst1; | |
266 | } else if (offs2 + sz == offs1) { | |
267 | /* 1st LEB was written, 2nd was not */ | |
268 | if (cor1) | |
269 | goto out_err; | |
270 | mst = mst1; | |
ff94bc40 HS |
271 | } else if (offs1 == 0 && |
272 | c->leb_size - offs2 - sz < sz) { | |
9eefe2a2 SR |
273 | /* 1st LEB was unmapped and written, 2nd not */ |
274 | if (cor1) | |
275 | goto out_err; | |
276 | mst = mst1; | |
277 | } else | |
278 | goto out_err; | |
279 | } else { | |
280 | /* | |
281 | * 2nd LEB was unmapped and about to be written, so | |
282 | * there must be only one master node in the first LEB | |
283 | * and no corruption. | |
284 | */ | |
285 | if (offs1 != 0 || cor1) | |
286 | goto out_err; | |
287 | mst = mst1; | |
288 | } | |
289 | } else { | |
290 | if (!mst2) | |
291 | goto out_err; | |
292 | /* | |
293 | * 1st LEB was unmapped and about to be written, so there must | |
294 | * be no room left in 2nd LEB. | |
295 | */ | |
296 | offs2 = (void *)mst2 - buf2; | |
297 | if (offs2 + sz + sz <= c->leb_size) | |
298 | goto out_err; | |
299 | mst = mst2; | |
300 | } | |
301 | ||
ff94bc40 | 302 | ubifs_msg("recovered master node from LEB %d", |
9eefe2a2 SR |
303 | (mst == mst1 ? UBIFS_MST_LNUM : UBIFS_MST_LNUM + 1)); |
304 | ||
305 | memcpy(c->mst_node, mst, UBIFS_MST_NODE_SZ); | |
306 | ||
ff94bc40 | 307 | if (c->ro_mount) { |
9eefe2a2 SR |
308 | /* Read-only mode. Keep a copy for switching to rw mode */ |
309 | c->rcvrd_mst_node = kmalloc(sz, GFP_KERNEL); | |
310 | if (!c->rcvrd_mst_node) { | |
311 | err = -ENOMEM; | |
312 | goto out_free; | |
313 | } | |
314 | memcpy(c->rcvrd_mst_node, c->mst_node, UBIFS_MST_NODE_SZ); | |
ff94bc40 HS |
315 | |
316 | /* | |
317 | * We had to recover the master node, which means there was an | |
318 | * unclean reboot. However, it is possible that the master node | |
319 | * is clean at this point, i.e., %UBIFS_MST_DIRTY is not set. | |
320 | * E.g., consider the following chain of events: | |
321 | * | |
322 | * 1. UBIFS was cleanly unmounted, so the master node is clean | |
323 | * 2. UBIFS is being mounted R/W and starts changing the master | |
324 | * node in the first (%UBIFS_MST_LNUM). A power cut happens, | |
325 | * so this LEB ends up with some amount of garbage at the | |
326 | * end. | |
327 | * 3. UBIFS is being mounted R/O. We reach this place and | |
328 | * recover the master node from the second LEB | |
329 | * (%UBIFS_MST_LNUM + 1). But we cannot update the media | |
330 | * because we are being mounted R/O. We have to defer the | |
331 | * operation. | |
332 | * 4. However, this master node (@c->mst_node) is marked as | |
333 | * clean (since the step 1). And if we just return, the | |
334 | * mount code will be confused and won't recover the master | |
335 | * node when it is re-mounter R/W later. | |
336 | * | |
337 | * Thus, to force the recovery by marking the master node as | |
338 | * dirty. | |
339 | */ | |
340 | c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY); | |
341 | #ifndef __UBOOT__ | |
342 | } else { | |
343 | /* Write the recovered master node */ | |
344 | c->max_sqnum = le64_to_cpu(mst->ch.sqnum) - 1; | |
345 | err = write_rcvrd_mst_node(c, c->mst_node); | |
346 | if (err) | |
347 | goto out_free; | |
348 | #endif | |
9eefe2a2 SR |
349 | } |
350 | ||
351 | vfree(buf2); | |
352 | vfree(buf1); | |
353 | ||
354 | return 0; | |
355 | ||
356 | out_err: | |
357 | err = -EINVAL; | |
358 | out_free: | |
359 | ubifs_err("failed to recover master node"); | |
360 | if (mst1) { | |
ff94bc40 HS |
361 | ubifs_err("dumping first master node"); |
362 | ubifs_dump_node(c, mst1); | |
9eefe2a2 SR |
363 | } |
364 | if (mst2) { | |
ff94bc40 HS |
365 | ubifs_err("dumping second master node"); |
366 | ubifs_dump_node(c, mst2); | |
9eefe2a2 SR |
367 | } |
368 | vfree(buf2); | |
369 | vfree(buf1); | |
370 | return err; | |
371 | } | |
372 | ||
373 | /** | |
374 | * ubifs_write_rcvrd_mst_node - write the recovered master node. | |
375 | * @c: UBIFS file-system description object | |
376 | * | |
377 | * This function writes the master node that was recovered during mounting in | |
378 | * read-only mode and must now be written because we are remounting rw. | |
379 | * | |
380 | * This function returns %0 on success and a negative error code on failure. | |
381 | */ | |
382 | int ubifs_write_rcvrd_mst_node(struct ubifs_info *c) | |
383 | { | |
384 | int err; | |
385 | ||
386 | if (!c->rcvrd_mst_node) | |
387 | return 0; | |
388 | c->rcvrd_mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY); | |
389 | c->mst_node->flags |= cpu_to_le32(UBIFS_MST_DIRTY); | |
390 | err = write_rcvrd_mst_node(c, c->rcvrd_mst_node); | |
391 | if (err) | |
392 | return err; | |
393 | kfree(c->rcvrd_mst_node); | |
394 | c->rcvrd_mst_node = NULL; | |
395 | return 0; | |
396 | } | |
397 | ||
398 | /** | |
399 | * is_last_write - determine if an offset was in the last write to a LEB. | |
400 | * @c: UBIFS file-system description object | |
401 | * @buf: buffer to check | |
402 | * @offs: offset to check | |
403 | * | |
404 | * This function returns %1 if @offs was in the last write to the LEB whose data | |
ff94bc40 HS |
405 | * is in @buf, otherwise %0 is returned. The determination is made by checking |
406 | * for subsequent empty space starting from the next @c->max_write_size | |
407 | * boundary. | |
9eefe2a2 SR |
408 | */ |
409 | static int is_last_write(const struct ubifs_info *c, void *buf, int offs) | |
410 | { | |
ff94bc40 | 411 | int empty_offs, check_len; |
9eefe2a2 SR |
412 | uint8_t *p; |
413 | ||
9eefe2a2 | 414 | /* |
ff94bc40 HS |
415 | * Round up to the next @c->max_write_size boundary i.e. @offs is in |
416 | * the last wbuf written. After that should be empty space. | |
9eefe2a2 | 417 | */ |
ff94bc40 | 418 | empty_offs = ALIGN(offs + 1, c->max_write_size); |
9eefe2a2 SR |
419 | check_len = c->leb_size - empty_offs; |
420 | p = buf + empty_offs - offs; | |
ff94bc40 | 421 | return is_empty(p, check_len); |
9eefe2a2 SR |
422 | } |
423 | ||
424 | /** | |
425 | * clean_buf - clean the data from an LEB sitting in a buffer. | |
426 | * @c: UBIFS file-system description object | |
427 | * @buf: buffer to clean | |
428 | * @lnum: LEB number to clean | |
429 | * @offs: offset from which to clean | |
430 | * @len: length of buffer | |
431 | * | |
432 | * This function pads up to the next min_io_size boundary (if there is one) and | |
433 | * sets empty space to all 0xff. @buf, @offs and @len are updated to the next | |
ff94bc40 | 434 | * @c->min_io_size boundary. |
9eefe2a2 SR |
435 | */ |
436 | static void clean_buf(const struct ubifs_info *c, void **buf, int lnum, | |
437 | int *offs, int *len) | |
438 | { | |
439 | int empty_offs, pad_len; | |
440 | ||
441 | lnum = lnum; | |
442 | dbg_rcvry("cleaning corruption at %d:%d", lnum, *offs); | |
443 | ||
9eefe2a2 SR |
444 | ubifs_assert(!(*offs & 7)); |
445 | empty_offs = ALIGN(*offs, c->min_io_size); | |
446 | pad_len = empty_offs - *offs; | |
447 | ubifs_pad(c, *buf, pad_len); | |
448 | *offs += pad_len; | |
449 | *buf += pad_len; | |
450 | *len -= pad_len; | |
451 | memset(*buf, 0xff, c->leb_size - empty_offs); | |
452 | } | |
453 | ||
454 | /** | |
455 | * no_more_nodes - determine if there are no more nodes in a buffer. | |
456 | * @c: UBIFS file-system description object | |
457 | * @buf: buffer to check | |
458 | * @len: length of buffer | |
459 | * @lnum: LEB number of the LEB from which @buf was read | |
460 | * @offs: offset from which @buf was read | |
461 | * | |
6356daff AH |
462 | * This function ensures that the corrupted node at @offs is the last thing |
463 | * written to a LEB. This function returns %1 if more data is not found and | |
464 | * %0 if more data is found. | |
9eefe2a2 SR |
465 | */ |
466 | static int no_more_nodes(const struct ubifs_info *c, void *buf, int len, | |
467 | int lnum, int offs) | |
468 | { | |
6356daff AH |
469 | struct ubifs_ch *ch = buf; |
470 | int skip, dlen = le32_to_cpu(ch->len); | |
9eefe2a2 | 471 | |
6356daff | 472 | /* Check for empty space after the corrupt node's common header */ |
ff94bc40 | 473 | skip = ALIGN(offs + UBIFS_CH_SZ, c->max_write_size) - offs; |
6356daff AH |
474 | if (is_empty(buf + skip, len - skip)) |
475 | return 1; | |
476 | /* | |
477 | * The area after the common header size is not empty, so the common | |
478 | * header must be intact. Check it. | |
479 | */ | |
480 | if (ubifs_check_node(c, buf, lnum, offs, 1, 0) != -EUCLEAN) { | |
481 | dbg_rcvry("unexpected bad common header at %d:%d", lnum, offs); | |
482 | return 0; | |
9eefe2a2 | 483 | } |
6356daff | 484 | /* Now we know the corrupt node's length we can skip over it */ |
ff94bc40 | 485 | skip = ALIGN(offs + dlen, c->max_write_size) - offs; |
6356daff AH |
486 | /* After which there should be empty space */ |
487 | if (is_empty(buf + skip, len - skip)) | |
488 | return 1; | |
489 | dbg_rcvry("unexpected data at %d:%d", lnum, offs + skip); | |
490 | return 0; | |
9eefe2a2 SR |
491 | } |
492 | ||
493 | /** | |
494 | * fix_unclean_leb - fix an unclean LEB. | |
495 | * @c: UBIFS file-system description object | |
496 | * @sleb: scanned LEB information | |
497 | * @start: offset where scan started | |
498 | */ | |
499 | static int fix_unclean_leb(struct ubifs_info *c, struct ubifs_scan_leb *sleb, | |
500 | int start) | |
501 | { | |
502 | int lnum = sleb->lnum, endpt = start; | |
503 | ||
504 | /* Get the end offset of the last node we are keeping */ | |
505 | if (!list_empty(&sleb->nodes)) { | |
506 | struct ubifs_scan_node *snod; | |
507 | ||
508 | snod = list_entry(sleb->nodes.prev, | |
509 | struct ubifs_scan_node, list); | |
510 | endpt = snod->offs + snod->len; | |
511 | } | |
512 | ||
ff94bc40 | 513 | if (c->ro_mount && !c->remounting_rw) { |
9eefe2a2 SR |
514 | /* Add to recovery list */ |
515 | struct ubifs_unclean_leb *ucleb; | |
516 | ||
517 | dbg_rcvry("need to fix LEB %d start %d endpt %d", | |
518 | lnum, start, sleb->endpt); | |
519 | ucleb = kzalloc(sizeof(struct ubifs_unclean_leb), GFP_NOFS); | |
520 | if (!ucleb) | |
521 | return -ENOMEM; | |
522 | ucleb->lnum = lnum; | |
523 | ucleb->endpt = endpt; | |
524 | list_add_tail(&ucleb->list, &c->unclean_leb_list); | |
ff94bc40 HS |
525 | #ifndef __UBOOT__ |
526 | } else { | |
527 | /* Write the fixed LEB back to flash */ | |
528 | int err; | |
529 | ||
530 | dbg_rcvry("fixing LEB %d start %d endpt %d", | |
531 | lnum, start, sleb->endpt); | |
532 | if (endpt == 0) { | |
533 | err = ubifs_leb_unmap(c, lnum); | |
534 | if (err) | |
535 | return err; | |
536 | } else { | |
537 | int len = ALIGN(endpt, c->min_io_size); | |
538 | ||
539 | if (start) { | |
540 | err = ubifs_leb_read(c, lnum, sleb->buf, 0, | |
541 | start, 1); | |
542 | if (err) | |
543 | return err; | |
544 | } | |
545 | /* Pad to min_io_size */ | |
546 | if (len > endpt) { | |
547 | int pad_len = len - ALIGN(endpt, 8); | |
548 | ||
549 | if (pad_len > 0) { | |
550 | void *buf = sleb->buf + len - pad_len; | |
551 | ||
552 | ubifs_pad(c, buf, pad_len); | |
553 | } | |
554 | } | |
555 | err = ubifs_leb_change(c, lnum, sleb->buf, len); | |
556 | if (err) | |
557 | return err; | |
558 | } | |
559 | #endif | |
9eefe2a2 SR |
560 | } |
561 | return 0; | |
562 | } | |
563 | ||
564 | /** | |
ff94bc40 | 565 | * drop_last_group - drop the last group of nodes. |
9eefe2a2 SR |
566 | * @sleb: scanned LEB information |
567 | * @offs: offset of dropped nodes is returned here | |
568 | * | |
ff94bc40 HS |
569 | * This is a helper function for 'ubifs_recover_leb()' which drops the last |
570 | * group of nodes of the scanned LEB. | |
9eefe2a2 | 571 | */ |
ff94bc40 | 572 | static void drop_last_group(struct ubifs_scan_leb *sleb, int *offs) |
9eefe2a2 | 573 | { |
9eefe2a2 SR |
574 | while (!list_empty(&sleb->nodes)) { |
575 | struct ubifs_scan_node *snod; | |
576 | struct ubifs_ch *ch; | |
577 | ||
578 | snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node, | |
579 | list); | |
580 | ch = snod->node; | |
581 | if (ch->group_type != UBIFS_IN_NODE_GROUP) | |
ff94bc40 HS |
582 | break; |
583 | ||
584 | dbg_rcvry("dropping grouped node at %d:%d", | |
585 | sleb->lnum, snod->offs); | |
586 | *offs = snod->offs; | |
587 | list_del(&snod->list); | |
588 | kfree(snod); | |
589 | sleb->nodes_cnt -= 1; | |
590 | } | |
591 | } | |
592 | ||
593 | /** | |
594 | * drop_last_node - drop the last node. | |
595 | * @sleb: scanned LEB information | |
596 | * @offs: offset of dropped nodes is returned here | |
597 | * @grouped: non-zero if whole group of nodes have to be dropped | |
598 | * | |
599 | * This is a helper function for 'ubifs_recover_leb()' which drops the last | |
600 | * node of the scanned LEB. | |
601 | */ | |
602 | static void drop_last_node(struct ubifs_scan_leb *sleb, int *offs) | |
603 | { | |
604 | struct ubifs_scan_node *snod; | |
605 | ||
606 | if (!list_empty(&sleb->nodes)) { | |
607 | snod = list_entry(sleb->nodes.prev, struct ubifs_scan_node, | |
608 | list); | |
609 | ||
610 | dbg_rcvry("dropping last node at %d:%d", | |
611 | sleb->lnum, snod->offs); | |
9eefe2a2 SR |
612 | *offs = snod->offs; |
613 | list_del(&snod->list); | |
614 | kfree(snod); | |
615 | sleb->nodes_cnt -= 1; | |
9eefe2a2 | 616 | } |
9eefe2a2 SR |
617 | } |
618 | ||
619 | /** | |
620 | * ubifs_recover_leb - scan and recover a LEB. | |
621 | * @c: UBIFS file-system description object | |
622 | * @lnum: LEB number | |
623 | * @offs: offset | |
624 | * @sbuf: LEB-sized buffer to use | |
ff94bc40 HS |
625 | * @jhead: journal head number this LEB belongs to (%-1 if the LEB does not |
626 | * belong to any journal head) | |
9eefe2a2 SR |
627 | * |
628 | * This function does a scan of a LEB, but caters for errors that might have | |
629 | * been caused by the unclean unmount from which we are attempting to recover. | |
ff94bc40 HS |
630 | * Returns %0 in case of success, %-EUCLEAN if an unrecoverable corruption is |
631 | * found, and a negative error code in case of failure. | |
9eefe2a2 SR |
632 | */ |
633 | struct ubifs_scan_leb *ubifs_recover_leb(struct ubifs_info *c, int lnum, | |
ff94bc40 | 634 | int offs, void *sbuf, int jhead) |
9eefe2a2 | 635 | { |
ff94bc40 HS |
636 | int ret = 0, err, len = c->leb_size - offs, start = offs, min_io_unit; |
637 | int grouped = jhead == -1 ? 0 : c->jheads[jhead].grouped; | |
9eefe2a2 SR |
638 | struct ubifs_scan_leb *sleb; |
639 | void *buf = sbuf + offs; | |
640 | ||
ff94bc40 | 641 | dbg_rcvry("%d:%d, jhead %d, grouped %d", lnum, offs, jhead, grouped); |
9eefe2a2 SR |
642 | |
643 | sleb = ubifs_start_scan(c, lnum, offs, sbuf); | |
644 | if (IS_ERR(sleb)) | |
645 | return sleb; | |
646 | ||
ff94bc40 | 647 | ubifs_assert(len >= 8); |
9eefe2a2 | 648 | while (len >= 8) { |
9eefe2a2 SR |
649 | dbg_scan("look at LEB %d:%d (%d bytes left)", |
650 | lnum, offs, len); | |
651 | ||
652 | cond_resched(); | |
653 | ||
654 | /* | |
655 | * Scan quietly until there is an error from which we cannot | |
656 | * recover | |
657 | */ | |
ff94bc40 | 658 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, 1); |
9eefe2a2 SR |
659 | if (ret == SCANNED_A_NODE) { |
660 | /* A valid node, and not a padding node */ | |
661 | struct ubifs_ch *ch = buf; | |
662 | int node_len; | |
663 | ||
664 | err = ubifs_add_snod(c, sleb, buf, offs); | |
665 | if (err) | |
666 | goto error; | |
667 | node_len = ALIGN(le32_to_cpu(ch->len), 8); | |
668 | offs += node_len; | |
669 | buf += node_len; | |
670 | len -= node_len; | |
ff94bc40 | 671 | } else if (ret > 0) { |
9eefe2a2 SR |
672 | /* Padding bytes or a valid padding node */ |
673 | offs += ret; | |
674 | buf += ret; | |
675 | len -= ret; | |
ff94bc40 HS |
676 | } else if (ret == SCANNED_EMPTY_SPACE || |
677 | ret == SCANNED_GARBAGE || | |
678 | ret == SCANNED_A_BAD_PAD_NODE || | |
679 | ret == SCANNED_A_CORRUPT_NODE) { | |
680 | dbg_rcvry("found corruption (%d) at %d:%d", | |
681 | ret, lnum, offs); | |
9eefe2a2 | 682 | break; |
ff94bc40 HS |
683 | } else { |
684 | ubifs_err("unexpected return value %d", ret); | |
685 | err = -EINVAL; | |
686 | goto error; | |
9eefe2a2 | 687 | } |
ff94bc40 | 688 | } |
9eefe2a2 | 689 | |
ff94bc40 HS |
690 | if (ret == SCANNED_GARBAGE || ret == SCANNED_A_BAD_PAD_NODE) { |
691 | if (!is_last_write(c, buf, offs)) | |
692 | goto corrupted_rescan; | |
693 | } else if (ret == SCANNED_A_CORRUPT_NODE) { | |
694 | if (!no_more_nodes(c, buf, len, lnum, offs)) | |
695 | goto corrupted_rescan; | |
696 | } else if (!is_empty(buf, len)) { | |
697 | if (!is_last_write(c, buf, offs)) { | |
698 | int corruption = first_non_ff(buf, len); | |
9eefe2a2 | 699 | |
ff94bc40 HS |
700 | /* |
701 | * See header comment for this file for more | |
702 | * explanations about the reasons we have this check. | |
703 | */ | |
704 | ubifs_err("corrupt empty space LEB %d:%d, corruption starts at %d", | |
705 | lnum, offs, corruption); | |
706 | /* Make sure we dump interesting non-0xFF data */ | |
707 | offs += corruption; | |
708 | buf += corruption; | |
9eefe2a2 SR |
709 | goto corrupted; |
710 | } | |
711 | } | |
712 | ||
ff94bc40 HS |
713 | min_io_unit = round_down(offs, c->min_io_size); |
714 | if (grouped) | |
715 | /* | |
716 | * If nodes are grouped, always drop the incomplete group at | |
717 | * the end. | |
718 | */ | |
719 | drop_last_group(sleb, &offs); | |
9eefe2a2 | 720 | |
ff94bc40 HS |
721 | if (jhead == GCHD) { |
722 | /* | |
723 | * If this LEB belongs to the GC head then while we are in the | |
724 | * middle of the same min. I/O unit keep dropping nodes. So | |
725 | * basically, what we want is to make sure that the last min. | |
726 | * I/O unit where we saw the corruption is dropped completely | |
727 | * with all the uncorrupted nodes which may possibly sit there. | |
728 | * | |
729 | * In other words, let's name the min. I/O unit where the | |
730 | * corruption starts B, and the previous min. I/O unit A. The | |
731 | * below code tries to deal with a situation when half of B | |
732 | * contains valid nodes or the end of a valid node, and the | |
733 | * second half of B contains corrupted data or garbage. This | |
734 | * means that UBIFS had been writing to B just before the power | |
735 | * cut happened. I do not know how realistic is this scenario | |
736 | * that half of the min. I/O unit had been written successfully | |
737 | * and the other half not, but this is possible in our 'failure | |
738 | * mode emulation' infrastructure at least. | |
739 | * | |
740 | * So what is the problem, why we need to drop those nodes? Why | |
741 | * can't we just clean-up the second half of B by putting a | |
742 | * padding node there? We can, and this works fine with one | |
743 | * exception which was reproduced with power cut emulation | |
744 | * testing and happens extremely rarely. | |
745 | * | |
746 | * Imagine the file-system is full, we run GC which starts | |
747 | * moving valid nodes from LEB X to LEB Y (obviously, LEB Y is | |
748 | * the current GC head LEB). The @c->gc_lnum is -1, which means | |
749 | * that GC will retain LEB X and will try to continue. Imagine | |
750 | * that LEB X is currently the dirtiest LEB, and the amount of | |
751 | * used space in LEB Y is exactly the same as amount of free | |
752 | * space in LEB X. | |
753 | * | |
754 | * And a power cut happens when nodes are moved from LEB X to | |
755 | * LEB Y. We are here trying to recover LEB Y which is the GC | |
756 | * head LEB. We find the min. I/O unit B as described above. | |
757 | * Then we clean-up LEB Y by padding min. I/O unit. And later | |
758 | * 'ubifs_rcvry_gc_commit()' function fails, because it cannot | |
759 | * find a dirty LEB which could be GC'd into LEB Y! Even LEB X | |
760 | * does not match because the amount of valid nodes there does | |
761 | * not fit the free space in LEB Y any more! And this is | |
762 | * because of the padding node which we added to LEB Y. The | |
763 | * user-visible effect of this which I once observed and | |
764 | * analysed is that we cannot mount the file-system with | |
765 | * -ENOSPC error. | |
766 | * | |
767 | * So obviously, to make sure that situation does not happen we | |
768 | * should free min. I/O unit B in LEB Y completely and the last | |
769 | * used min. I/O unit in LEB Y should be A. This is basically | |
770 | * what the below code tries to do. | |
771 | */ | |
772 | while (offs > min_io_unit) | |
773 | drop_last_node(sleb, &offs); | |
9eefe2a2 SR |
774 | } |
775 | ||
ff94bc40 HS |
776 | buf = sbuf + offs; |
777 | len = c->leb_size - offs; | |
9eefe2a2 | 778 | |
ff94bc40 | 779 | clean_buf(c, &buf, lnum, &offs, &len); |
9eefe2a2 SR |
780 | ubifs_end_scan(c, sleb, lnum, offs); |
781 | ||
ff94bc40 HS |
782 | err = fix_unclean_leb(c, sleb, start); |
783 | if (err) | |
784 | goto error; | |
9eefe2a2 SR |
785 | |
786 | return sleb; | |
787 | ||
ff94bc40 HS |
788 | corrupted_rescan: |
789 | /* Re-scan the corrupted data with verbose messages */ | |
790 | ubifs_err("corruption %d", ret); | |
791 | ubifs_scan_a_node(c, buf, len, lnum, offs, 1); | |
9eefe2a2 SR |
792 | corrupted: |
793 | ubifs_scanned_corruption(c, lnum, offs, buf); | |
794 | err = -EUCLEAN; | |
795 | error: | |
796 | ubifs_err("LEB %d scanning failed", lnum); | |
797 | ubifs_scan_destroy(sleb); | |
798 | return ERR_PTR(err); | |
799 | } | |
800 | ||
801 | /** | |
802 | * get_cs_sqnum - get commit start sequence number. | |
803 | * @c: UBIFS file-system description object | |
804 | * @lnum: LEB number of commit start node | |
805 | * @offs: offset of commit start node | |
806 | * @cs_sqnum: commit start sequence number is returned here | |
807 | * | |
808 | * This function returns %0 on success and a negative error code on failure. | |
809 | */ | |
810 | static int get_cs_sqnum(struct ubifs_info *c, int lnum, int offs, | |
811 | unsigned long long *cs_sqnum) | |
812 | { | |
813 | struct ubifs_cs_node *cs_node = NULL; | |
814 | int err, ret; | |
815 | ||
816 | dbg_rcvry("at %d:%d", lnum, offs); | |
817 | cs_node = kmalloc(UBIFS_CS_NODE_SZ, GFP_KERNEL); | |
818 | if (!cs_node) | |
819 | return -ENOMEM; | |
820 | if (c->leb_size - offs < UBIFS_CS_NODE_SZ) | |
821 | goto out_err; | |
ff94bc40 HS |
822 | err = ubifs_leb_read(c, lnum, (void *)cs_node, offs, |
823 | UBIFS_CS_NODE_SZ, 0); | |
9eefe2a2 SR |
824 | if (err && err != -EBADMSG) |
825 | goto out_free; | |
826 | ret = ubifs_scan_a_node(c, cs_node, UBIFS_CS_NODE_SZ, lnum, offs, 0); | |
827 | if (ret != SCANNED_A_NODE) { | |
ff94bc40 | 828 | ubifs_err("Not a valid node"); |
9eefe2a2 SR |
829 | goto out_err; |
830 | } | |
831 | if (cs_node->ch.node_type != UBIFS_CS_NODE) { | |
ff94bc40 | 832 | ubifs_err("Node a CS node, type is %d", cs_node->ch.node_type); |
9eefe2a2 SR |
833 | goto out_err; |
834 | } | |
835 | if (le64_to_cpu(cs_node->cmt_no) != c->cmt_no) { | |
ff94bc40 HS |
836 | ubifs_err("CS node cmt_no %llu != current cmt_no %llu", |
837 | (unsigned long long)le64_to_cpu(cs_node->cmt_no), | |
838 | c->cmt_no); | |
9eefe2a2 SR |
839 | goto out_err; |
840 | } | |
841 | *cs_sqnum = le64_to_cpu(cs_node->ch.sqnum); | |
842 | dbg_rcvry("commit start sqnum %llu", *cs_sqnum); | |
843 | kfree(cs_node); | |
844 | return 0; | |
845 | ||
846 | out_err: | |
847 | err = -EINVAL; | |
848 | out_free: | |
849 | ubifs_err("failed to get CS sqnum"); | |
850 | kfree(cs_node); | |
851 | return err; | |
852 | } | |
853 | ||
854 | /** | |
855 | * ubifs_recover_log_leb - scan and recover a log LEB. | |
856 | * @c: UBIFS file-system description object | |
857 | * @lnum: LEB number | |
858 | * @offs: offset | |
859 | * @sbuf: LEB-sized buffer to use | |
860 | * | |
861 | * This function does a scan of a LEB, but caters for errors that might have | |
ff94bc40 HS |
862 | * been caused by unclean reboots from which we are attempting to recover |
863 | * (assume that only the last log LEB can be corrupted by an unclean reboot). | |
9eefe2a2 SR |
864 | * |
865 | * This function returns %0 on success and a negative error code on failure. | |
866 | */ | |
867 | struct ubifs_scan_leb *ubifs_recover_log_leb(struct ubifs_info *c, int lnum, | |
868 | int offs, void *sbuf) | |
869 | { | |
870 | struct ubifs_scan_leb *sleb; | |
871 | int next_lnum; | |
872 | ||
873 | dbg_rcvry("LEB %d", lnum); | |
874 | next_lnum = lnum + 1; | |
875 | if (next_lnum >= UBIFS_LOG_LNUM + c->log_lebs) | |
876 | next_lnum = UBIFS_LOG_LNUM; | |
877 | if (next_lnum != c->ltail_lnum) { | |
878 | /* | |
879 | * We can only recover at the end of the log, so check that the | |
880 | * next log LEB is empty or out of date. | |
881 | */ | |
ff94bc40 | 882 | sleb = ubifs_scan(c, next_lnum, 0, sbuf, 0); |
9eefe2a2 SR |
883 | if (IS_ERR(sleb)) |
884 | return sleb; | |
885 | if (sleb->nodes_cnt) { | |
886 | struct ubifs_scan_node *snod; | |
887 | unsigned long long cs_sqnum = c->cs_sqnum; | |
888 | ||
889 | snod = list_entry(sleb->nodes.next, | |
890 | struct ubifs_scan_node, list); | |
891 | if (cs_sqnum == 0) { | |
892 | int err; | |
893 | ||
894 | err = get_cs_sqnum(c, lnum, offs, &cs_sqnum); | |
895 | if (err) { | |
896 | ubifs_scan_destroy(sleb); | |
897 | return ERR_PTR(err); | |
898 | } | |
899 | } | |
900 | if (snod->sqnum > cs_sqnum) { | |
ff94bc40 HS |
901 | ubifs_err("unrecoverable log corruption in LEB %d", |
902 | lnum); | |
9eefe2a2 SR |
903 | ubifs_scan_destroy(sleb); |
904 | return ERR_PTR(-EUCLEAN); | |
905 | } | |
906 | } | |
907 | ubifs_scan_destroy(sleb); | |
908 | } | |
ff94bc40 | 909 | return ubifs_recover_leb(c, lnum, offs, sbuf, -1); |
9eefe2a2 SR |
910 | } |
911 | ||
912 | /** | |
913 | * recover_head - recover a head. | |
914 | * @c: UBIFS file-system description object | |
915 | * @lnum: LEB number of head to recover | |
916 | * @offs: offset of head to recover | |
917 | * @sbuf: LEB-sized buffer to use | |
918 | * | |
919 | * This function ensures that there is no data on the flash at a head location. | |
920 | * | |
921 | * This function returns %0 on success and a negative error code on failure. | |
922 | */ | |
ff94bc40 | 923 | static int recover_head(struct ubifs_info *c, int lnum, int offs, void *sbuf) |
9eefe2a2 | 924 | { |
ff94bc40 | 925 | int len = c->max_write_size, err; |
9eefe2a2 | 926 | |
9eefe2a2 SR |
927 | if (offs + len > c->leb_size) |
928 | len = c->leb_size - offs; | |
929 | ||
930 | if (!len) | |
931 | return 0; | |
932 | ||
933 | /* Read at the head location and check it is empty flash */ | |
ff94bc40 HS |
934 | err = ubifs_leb_read(c, lnum, sbuf, offs, len, 1); |
935 | if (err || !is_empty(sbuf, len)) { | |
9eefe2a2 SR |
936 | dbg_rcvry("cleaning head at %d:%d", lnum, offs); |
937 | if (offs == 0) | |
938 | return ubifs_leb_unmap(c, lnum); | |
ff94bc40 | 939 | err = ubifs_leb_read(c, lnum, sbuf, 0, offs, 1); |
9eefe2a2 SR |
940 | if (err) |
941 | return err; | |
ff94bc40 | 942 | return ubifs_leb_change(c, lnum, sbuf, offs); |
9eefe2a2 SR |
943 | } |
944 | ||
945 | return 0; | |
946 | } | |
947 | ||
948 | /** | |
949 | * ubifs_recover_inl_heads - recover index and LPT heads. | |
950 | * @c: UBIFS file-system description object | |
951 | * @sbuf: LEB-sized buffer to use | |
952 | * | |
953 | * This function ensures that there is no data on the flash at the index and | |
954 | * LPT head locations. | |
955 | * | |
956 | * This deals with the recovery of a half-completed journal commit. UBIFS is | |
957 | * careful never to overwrite the last version of the index or the LPT. Because | |
958 | * the index and LPT are wandering trees, data from a half-completed commit will | |
959 | * not be referenced anywhere in UBIFS. The data will be either in LEBs that are | |
960 | * assumed to be empty and will be unmapped anyway before use, or in the index | |
961 | * and LPT heads. | |
962 | * | |
963 | * This function returns %0 on success and a negative error code on failure. | |
964 | */ | |
ff94bc40 | 965 | int ubifs_recover_inl_heads(struct ubifs_info *c, void *sbuf) |
9eefe2a2 SR |
966 | { |
967 | int err; | |
968 | ||
ff94bc40 | 969 | ubifs_assert(!c->ro_mount || c->remounting_rw); |
9eefe2a2 SR |
970 | |
971 | dbg_rcvry("checking index head at %d:%d", c->ihead_lnum, c->ihead_offs); | |
972 | err = recover_head(c, c->ihead_lnum, c->ihead_offs, sbuf); | |
973 | if (err) | |
974 | return err; | |
975 | ||
976 | dbg_rcvry("checking LPT head at %d:%d", c->nhead_lnum, c->nhead_offs); | |
977 | err = recover_head(c, c->nhead_lnum, c->nhead_offs, sbuf); | |
978 | if (err) | |
979 | return err; | |
980 | ||
981 | return 0; | |
982 | } | |
983 | ||
984 | /** | |
ff94bc40 | 985 | * clean_an_unclean_leb - read and write a LEB to remove corruption. |
9eefe2a2 SR |
986 | * @c: UBIFS file-system description object |
987 | * @ucleb: unclean LEB information | |
988 | * @sbuf: LEB-sized buffer to use | |
989 | * | |
990 | * This function reads a LEB up to a point pre-determined by the mount recovery, | |
991 | * checks the nodes, and writes the result back to the flash, thereby cleaning | |
992 | * off any following corruption, or non-fatal ECC errors. | |
993 | * | |
994 | * This function returns %0 on success and a negative error code on failure. | |
995 | */ | |
ff94bc40 | 996 | static int clean_an_unclean_leb(struct ubifs_info *c, |
9eefe2a2 SR |
997 | struct ubifs_unclean_leb *ucleb, void *sbuf) |
998 | { | |
999 | int err, lnum = ucleb->lnum, offs = 0, len = ucleb->endpt, quiet = 1; | |
1000 | void *buf = sbuf; | |
1001 | ||
1002 | dbg_rcvry("LEB %d len %d", lnum, len); | |
1003 | ||
1004 | if (len == 0) { | |
1005 | /* Nothing to read, just unmap it */ | |
1006 | err = ubifs_leb_unmap(c, lnum); | |
1007 | if (err) | |
1008 | return err; | |
1009 | return 0; | |
1010 | } | |
1011 | ||
ff94bc40 | 1012 | err = ubifs_leb_read(c, lnum, buf, offs, len, 0); |
9eefe2a2 SR |
1013 | if (err && err != -EBADMSG) |
1014 | return err; | |
1015 | ||
1016 | while (len >= 8) { | |
1017 | int ret; | |
1018 | ||
1019 | cond_resched(); | |
1020 | ||
1021 | /* Scan quietly until there is an error */ | |
1022 | ret = ubifs_scan_a_node(c, buf, len, lnum, offs, quiet); | |
1023 | ||
1024 | if (ret == SCANNED_A_NODE) { | |
1025 | /* A valid node, and not a padding node */ | |
1026 | struct ubifs_ch *ch = buf; | |
1027 | int node_len; | |
1028 | ||
1029 | node_len = ALIGN(le32_to_cpu(ch->len), 8); | |
1030 | offs += node_len; | |
1031 | buf += node_len; | |
1032 | len -= node_len; | |
1033 | continue; | |
1034 | } | |
1035 | ||
1036 | if (ret > 0) { | |
1037 | /* Padding bytes or a valid padding node */ | |
1038 | offs += ret; | |
1039 | buf += ret; | |
1040 | len -= ret; | |
1041 | continue; | |
1042 | } | |
1043 | ||
1044 | if (ret == SCANNED_EMPTY_SPACE) { | |
1045 | ubifs_err("unexpected empty space at %d:%d", | |
1046 | lnum, offs); | |
1047 | return -EUCLEAN; | |
1048 | } | |
1049 | ||
1050 | if (quiet) { | |
1051 | /* Redo the last scan but noisily */ | |
1052 | quiet = 0; | |
1053 | continue; | |
1054 | } | |
1055 | ||
1056 | ubifs_scanned_corruption(c, lnum, offs, buf); | |
1057 | return -EUCLEAN; | |
1058 | } | |
1059 | ||
1060 | /* Pad to min_io_size */ | |
1061 | len = ALIGN(ucleb->endpt, c->min_io_size); | |
1062 | if (len > ucleb->endpt) { | |
1063 | int pad_len = len - ALIGN(ucleb->endpt, 8); | |
1064 | ||
1065 | if (pad_len > 0) { | |
1066 | buf = c->sbuf + len - pad_len; | |
1067 | ubifs_pad(c, buf, pad_len); | |
1068 | } | |
1069 | } | |
1070 | ||
1071 | /* Write back the LEB atomically */ | |
ff94bc40 | 1072 | err = ubifs_leb_change(c, lnum, sbuf, len); |
9eefe2a2 SR |
1073 | if (err) |
1074 | return err; | |
1075 | ||
1076 | dbg_rcvry("cleaned LEB %d", lnum); | |
1077 | ||
1078 | return 0; | |
1079 | } | |
1080 | ||
1081 | /** | |
1082 | * ubifs_clean_lebs - clean LEBs recovered during read-only mount. | |
1083 | * @c: UBIFS file-system description object | |
1084 | * @sbuf: LEB-sized buffer to use | |
1085 | * | |
1086 | * This function cleans a LEB identified during recovery that needs to be | |
1087 | * written but was not because UBIFS was mounted read-only. This happens when | |
1088 | * remounting to read-write mode. | |
1089 | * | |
1090 | * This function returns %0 on success and a negative error code on failure. | |
1091 | */ | |
ff94bc40 | 1092 | int ubifs_clean_lebs(struct ubifs_info *c, void *sbuf) |
9eefe2a2 SR |
1093 | { |
1094 | dbg_rcvry("recovery"); | |
1095 | while (!list_empty(&c->unclean_leb_list)) { | |
1096 | struct ubifs_unclean_leb *ucleb; | |
1097 | int err; | |
1098 | ||
1099 | ucleb = list_entry(c->unclean_leb_list.next, | |
1100 | struct ubifs_unclean_leb, list); | |
1101 | err = clean_an_unclean_leb(c, ucleb, sbuf); | |
1102 | if (err) | |
1103 | return err; | |
1104 | list_del(&ucleb->list); | |
1105 | kfree(ucleb); | |
1106 | } | |
1107 | return 0; | |
1108 | } | |
1109 | ||
ff94bc40 HS |
1110 | #ifndef __UBOOT__ |
1111 | /** | |
1112 | * grab_empty_leb - grab an empty LEB to use as GC LEB and run commit. | |
1113 | * @c: UBIFS file-system description object | |
1114 | * | |
1115 | * This is a helper function for 'ubifs_rcvry_gc_commit()' which grabs an empty | |
1116 | * LEB to be used as GC LEB (@c->gc_lnum), and then runs the commit. Returns | |
1117 | * zero in case of success and a negative error code in case of failure. | |
1118 | */ | |
1119 | static int grab_empty_leb(struct ubifs_info *c) | |
1120 | { | |
1121 | int lnum, err; | |
1122 | ||
1123 | /* | |
1124 | * Note, it is very important to first search for an empty LEB and then | |
1125 | * run the commit, not vice-versa. The reason is that there might be | |
1126 | * only one empty LEB at the moment, the one which has been the | |
1127 | * @c->gc_lnum just before the power cut happened. During the regular | |
1128 | * UBIFS operation (not now) @c->gc_lnum is marked as "taken", so no | |
1129 | * one but GC can grab it. But at this moment this single empty LEB is | |
1130 | * not marked as taken, so if we run commit - what happens? Right, the | |
1131 | * commit will grab it and write the index there. Remember that the | |
1132 | * index always expands as long as there is free space, and it only | |
1133 | * starts consolidating when we run out of space. | |
1134 | * | |
1135 | * IOW, if we run commit now, we might not be able to find a free LEB | |
1136 | * after this. | |
1137 | */ | |
1138 | lnum = ubifs_find_free_leb_for_idx(c); | |
1139 | if (lnum < 0) { | |
1140 | ubifs_err("could not find an empty LEB"); | |
1141 | ubifs_dump_lprops(c); | |
1142 | ubifs_dump_budg(c, &c->bi); | |
1143 | return lnum; | |
1144 | } | |
1145 | ||
1146 | /* Reset the index flag */ | |
1147 | err = ubifs_change_one_lp(c, lnum, LPROPS_NC, LPROPS_NC, 0, | |
1148 | LPROPS_INDEX, 0); | |
1149 | if (err) | |
1150 | return err; | |
1151 | ||
1152 | c->gc_lnum = lnum; | |
1153 | dbg_rcvry("found empty LEB %d, run commit", lnum); | |
1154 | ||
1155 | return ubifs_run_commit(c); | |
1156 | } | |
1157 | ||
1158 | /** | |
1159 | * ubifs_rcvry_gc_commit - recover the GC LEB number and run the commit. | |
1160 | * @c: UBIFS file-system description object | |
1161 | * | |
1162 | * Out-of-place garbage collection requires always one empty LEB with which to | |
1163 | * start garbage collection. The LEB number is recorded in c->gc_lnum and is | |
1164 | * written to the master node on unmounting. In the case of an unclean unmount | |
1165 | * the value of gc_lnum recorded in the master node is out of date and cannot | |
1166 | * be used. Instead, recovery must allocate an empty LEB for this purpose. | |
1167 | * However, there may not be enough empty space, in which case it must be | |
1168 | * possible to GC the dirtiest LEB into the GC head LEB. | |
1169 | * | |
1170 | * This function also runs the commit which causes the TNC updates from | |
1171 | * size-recovery and orphans to be written to the flash. That is important to | |
1172 | * ensure correct replay order for subsequent mounts. | |
1173 | * | |
1174 | * This function returns %0 on success and a negative error code on failure. | |
1175 | */ | |
1176 | int ubifs_rcvry_gc_commit(struct ubifs_info *c) | |
1177 | { | |
1178 | struct ubifs_wbuf *wbuf = &c->jheads[GCHD].wbuf; | |
1179 | struct ubifs_lprops lp; | |
1180 | int err; | |
1181 | ||
1182 | dbg_rcvry("GC head LEB %d, offs %d", wbuf->lnum, wbuf->offs); | |
1183 | ||
1184 | c->gc_lnum = -1; | |
1185 | if (wbuf->lnum == -1 || wbuf->offs == c->leb_size) | |
1186 | return grab_empty_leb(c); | |
1187 | ||
1188 | err = ubifs_find_dirty_leb(c, &lp, wbuf->offs, 2); | |
1189 | if (err) { | |
1190 | if (err != -ENOSPC) | |
1191 | return err; | |
1192 | ||
1193 | dbg_rcvry("could not find a dirty LEB"); | |
1194 | return grab_empty_leb(c); | |
1195 | } | |
1196 | ||
1197 | ubifs_assert(!(lp.flags & LPROPS_INDEX)); | |
1198 | ubifs_assert(lp.free + lp.dirty >= wbuf->offs); | |
1199 | ||
1200 | /* | |
1201 | * We run the commit before garbage collection otherwise subsequent | |
1202 | * mounts will see the GC and orphan deletion in a different order. | |
1203 | */ | |
1204 | dbg_rcvry("committing"); | |
1205 | err = ubifs_run_commit(c); | |
1206 | if (err) | |
1207 | return err; | |
1208 | ||
1209 | dbg_rcvry("GC'ing LEB %d", lp.lnum); | |
1210 | mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); | |
1211 | err = ubifs_garbage_collect_leb(c, &lp); | |
1212 | if (err >= 0) { | |
1213 | int err2 = ubifs_wbuf_sync_nolock(wbuf); | |
1214 | ||
1215 | if (err2) | |
1216 | err = err2; | |
1217 | } | |
1218 | mutex_unlock(&wbuf->io_mutex); | |
1219 | if (err < 0) { | |
1220 | ubifs_err("GC failed, error %d", err); | |
1221 | if (err == -EAGAIN) | |
1222 | err = -EINVAL; | |
1223 | return err; | |
1224 | } | |
1225 | ||
1226 | ubifs_assert(err == LEB_RETAINED); | |
1227 | if (err != LEB_RETAINED) | |
1228 | return -EINVAL; | |
1229 | ||
1230 | err = ubifs_leb_unmap(c, c->gc_lnum); | |
1231 | if (err) | |
1232 | return err; | |
1233 | ||
1234 | dbg_rcvry("allocated LEB %d for GC", lp.lnum); | |
1235 | return 0; | |
1236 | } | |
1237 | #else | |
1238 | int ubifs_rcvry_gc_commit(struct ubifs_info *c) | |
1239 | { | |
1240 | return 0; | |
1241 | } | |
1242 | #endif | |
1243 | ||
9eefe2a2 SR |
1244 | /** |
1245 | * struct size_entry - inode size information for recovery. | |
1246 | * @rb: link in the RB-tree of sizes | |
1247 | * @inum: inode number | |
1248 | * @i_size: size on inode | |
1249 | * @d_size: maximum size based on data nodes | |
1250 | * @exists: indicates whether the inode exists | |
1251 | * @inode: inode if pinned in memory awaiting rw mode to fix it | |
1252 | */ | |
1253 | struct size_entry { | |
1254 | struct rb_node rb; | |
1255 | ino_t inum; | |
1256 | loff_t i_size; | |
1257 | loff_t d_size; | |
1258 | int exists; | |
1259 | struct inode *inode; | |
1260 | }; | |
1261 | ||
1262 | /** | |
1263 | * add_ino - add an entry to the size tree. | |
1264 | * @c: UBIFS file-system description object | |
1265 | * @inum: inode number | |
1266 | * @i_size: size on inode | |
1267 | * @d_size: maximum size based on data nodes | |
1268 | * @exists: indicates whether the inode exists | |
1269 | */ | |
1270 | static int add_ino(struct ubifs_info *c, ino_t inum, loff_t i_size, | |
1271 | loff_t d_size, int exists) | |
1272 | { | |
1273 | struct rb_node **p = &c->size_tree.rb_node, *parent = NULL; | |
1274 | struct size_entry *e; | |
1275 | ||
1276 | while (*p) { | |
1277 | parent = *p; | |
1278 | e = rb_entry(parent, struct size_entry, rb); | |
1279 | if (inum < e->inum) | |
1280 | p = &(*p)->rb_left; | |
1281 | else | |
1282 | p = &(*p)->rb_right; | |
1283 | } | |
1284 | ||
1285 | e = kzalloc(sizeof(struct size_entry), GFP_KERNEL); | |
1286 | if (!e) | |
1287 | return -ENOMEM; | |
1288 | ||
1289 | e->inum = inum; | |
1290 | e->i_size = i_size; | |
1291 | e->d_size = d_size; | |
1292 | e->exists = exists; | |
1293 | ||
1294 | rb_link_node(&e->rb, parent, p); | |
1295 | rb_insert_color(&e->rb, &c->size_tree); | |
1296 | ||
1297 | return 0; | |
1298 | } | |
1299 | ||
1300 | /** | |
1301 | * find_ino - find an entry on the size tree. | |
1302 | * @c: UBIFS file-system description object | |
1303 | * @inum: inode number | |
1304 | */ | |
1305 | static struct size_entry *find_ino(struct ubifs_info *c, ino_t inum) | |
1306 | { | |
1307 | struct rb_node *p = c->size_tree.rb_node; | |
1308 | struct size_entry *e; | |
1309 | ||
1310 | while (p) { | |
1311 | e = rb_entry(p, struct size_entry, rb); | |
1312 | if (inum < e->inum) | |
1313 | p = p->rb_left; | |
1314 | else if (inum > e->inum) | |
1315 | p = p->rb_right; | |
1316 | else | |
1317 | return e; | |
1318 | } | |
1319 | return NULL; | |
1320 | } | |
1321 | ||
1322 | /** | |
1323 | * remove_ino - remove an entry from the size tree. | |
1324 | * @c: UBIFS file-system description object | |
1325 | * @inum: inode number | |
1326 | */ | |
1327 | static void remove_ino(struct ubifs_info *c, ino_t inum) | |
1328 | { | |
1329 | struct size_entry *e = find_ino(c, inum); | |
1330 | ||
1331 | if (!e) | |
1332 | return; | |
1333 | rb_erase(&e->rb, &c->size_tree); | |
1334 | kfree(e); | |
1335 | } | |
1336 | ||
ff94bc40 HS |
1337 | /** |
1338 | * ubifs_destroy_size_tree - free resources related to the size tree. | |
1339 | * @c: UBIFS file-system description object | |
1340 | */ | |
1341 | void ubifs_destroy_size_tree(struct ubifs_info *c) | |
1342 | { | |
1343 | struct size_entry *e, *n; | |
1344 | ||
1345 | rbtree_postorder_for_each_entry_safe(e, n, &c->size_tree, rb) { | |
1346 | if (e->inode) | |
1347 | iput(e->inode); | |
1348 | kfree(e); | |
1349 | } | |
1350 | ||
1351 | c->size_tree = RB_ROOT; | |
1352 | } | |
1353 | ||
9eefe2a2 SR |
1354 | /** |
1355 | * ubifs_recover_size_accum - accumulate inode sizes for recovery. | |
1356 | * @c: UBIFS file-system description object | |
1357 | * @key: node key | |
1358 | * @deletion: node is for a deletion | |
1359 | * @new_size: inode size | |
1360 | * | |
1361 | * This function has two purposes: | |
1362 | * 1) to ensure there are no data nodes that fall outside the inode size | |
1363 | * 2) to ensure there are no data nodes for inodes that do not exist | |
1364 | * To accomplish those purposes, a rb-tree is constructed containing an entry | |
1365 | * for each inode number in the journal that has not been deleted, and recording | |
1366 | * the size from the inode node, the maximum size of any data node (also altered | |
1367 | * by truncations) and a flag indicating a inode number for which no inode node | |
1368 | * was present in the journal. | |
1369 | * | |
1370 | * Note that there is still the possibility that there are data nodes that have | |
1371 | * been committed that are beyond the inode size, however the only way to find | |
1372 | * them would be to scan the entire index. Alternatively, some provision could | |
1373 | * be made to record the size of inodes at the start of commit, which would seem | |
1374 | * very cumbersome for a scenario that is quite unlikely and the only negative | |
1375 | * consequence of which is wasted space. | |
1376 | * | |
1377 | * This functions returns %0 on success and a negative error code on failure. | |
1378 | */ | |
1379 | int ubifs_recover_size_accum(struct ubifs_info *c, union ubifs_key *key, | |
1380 | int deletion, loff_t new_size) | |
1381 | { | |
1382 | ino_t inum = key_inum(c, key); | |
1383 | struct size_entry *e; | |
1384 | int err; | |
1385 | ||
1386 | switch (key_type(c, key)) { | |
1387 | case UBIFS_INO_KEY: | |
1388 | if (deletion) | |
1389 | remove_ino(c, inum); | |
1390 | else { | |
1391 | e = find_ino(c, inum); | |
1392 | if (e) { | |
1393 | e->i_size = new_size; | |
1394 | e->exists = 1; | |
1395 | } else { | |
1396 | err = add_ino(c, inum, new_size, 0, 1); | |
1397 | if (err) | |
1398 | return err; | |
1399 | } | |
1400 | } | |
1401 | break; | |
1402 | case UBIFS_DATA_KEY: | |
1403 | e = find_ino(c, inum); | |
1404 | if (e) { | |
1405 | if (new_size > e->d_size) | |
1406 | e->d_size = new_size; | |
1407 | } else { | |
1408 | err = add_ino(c, inum, 0, new_size, 0); | |
1409 | if (err) | |
1410 | return err; | |
1411 | } | |
1412 | break; | |
1413 | case UBIFS_TRUN_KEY: | |
1414 | e = find_ino(c, inum); | |
1415 | if (e) | |
1416 | e->d_size = new_size; | |
1417 | break; | |
1418 | } | |
1419 | return 0; | |
1420 | } | |
1421 | ||
ff94bc40 HS |
1422 | #ifndef __UBOOT__ |
1423 | /** | |
1424 | * fix_size_in_place - fix inode size in place on flash. | |
1425 | * @c: UBIFS file-system description object | |
1426 | * @e: inode size information for recovery | |
1427 | */ | |
1428 | static int fix_size_in_place(struct ubifs_info *c, struct size_entry *e) | |
1429 | { | |
1430 | struct ubifs_ino_node *ino = c->sbuf; | |
1431 | unsigned char *p; | |
1432 | union ubifs_key key; | |
1433 | int err, lnum, offs, len; | |
1434 | loff_t i_size; | |
1435 | uint32_t crc; | |
1436 | ||
1437 | /* Locate the inode node LEB number and offset */ | |
1438 | ino_key_init(c, &key, e->inum); | |
1439 | err = ubifs_tnc_locate(c, &key, ino, &lnum, &offs); | |
1440 | if (err) | |
1441 | goto out; | |
1442 | /* | |
1443 | * If the size recorded on the inode node is greater than the size that | |
1444 | * was calculated from nodes in the journal then don't change the inode. | |
1445 | */ | |
1446 | i_size = le64_to_cpu(ino->size); | |
1447 | if (i_size >= e->d_size) | |
1448 | return 0; | |
1449 | /* Read the LEB */ | |
1450 | err = ubifs_leb_read(c, lnum, c->sbuf, 0, c->leb_size, 1); | |
1451 | if (err) | |
1452 | goto out; | |
1453 | /* Change the size field and recalculate the CRC */ | |
1454 | ino = c->sbuf + offs; | |
1455 | ino->size = cpu_to_le64(e->d_size); | |
1456 | len = le32_to_cpu(ino->ch.len); | |
1457 | crc = crc32(UBIFS_CRC32_INIT, (void *)ino + 8, len - 8); | |
1458 | ino->ch.crc = cpu_to_le32(crc); | |
1459 | /* Work out where data in the LEB ends and free space begins */ | |
1460 | p = c->sbuf; | |
1461 | len = c->leb_size - 1; | |
1462 | while (p[len] == 0xff) | |
1463 | len -= 1; | |
1464 | len = ALIGN(len + 1, c->min_io_size); | |
1465 | /* Atomically write the fixed LEB back again */ | |
1466 | err = ubifs_leb_change(c, lnum, c->sbuf, len); | |
1467 | if (err) | |
1468 | goto out; | |
1469 | dbg_rcvry("inode %lu at %d:%d size %lld -> %lld", | |
1470 | (unsigned long)e->inum, lnum, offs, i_size, e->d_size); | |
1471 | return 0; | |
1472 | ||
1473 | out: | |
1474 | ubifs_warn("inode %lu failed to fix size %lld -> %lld error %d", | |
1475 | (unsigned long)e->inum, e->i_size, e->d_size, err); | |
1476 | return err; | |
1477 | } | |
1478 | #endif | |
1479 | ||
9eefe2a2 SR |
1480 | /** |
1481 | * ubifs_recover_size - recover inode size. | |
1482 | * @c: UBIFS file-system description object | |
1483 | * | |
1484 | * This function attempts to fix inode size discrepancies identified by the | |
1485 | * 'ubifs_recover_size_accum()' function. | |
1486 | * | |
1487 | * This functions returns %0 on success and a negative error code on failure. | |
1488 | */ | |
1489 | int ubifs_recover_size(struct ubifs_info *c) | |
1490 | { | |
1491 | struct rb_node *this = rb_first(&c->size_tree); | |
1492 | ||
1493 | while (this) { | |
1494 | struct size_entry *e; | |
1495 | int err; | |
1496 | ||
1497 | e = rb_entry(this, struct size_entry, rb); | |
1498 | if (!e->exists) { | |
1499 | union ubifs_key key; | |
1500 | ||
1501 | ino_key_init(c, &key, e->inum); | |
1502 | err = ubifs_tnc_lookup(c, &key, c->sbuf); | |
1503 | if (err && err != -ENOENT) | |
1504 | return err; | |
1505 | if (err == -ENOENT) { | |
1506 | /* Remove data nodes that have no inode */ | |
1507 | dbg_rcvry("removing ino %lu", | |
1508 | (unsigned long)e->inum); | |
1509 | err = ubifs_tnc_remove_ino(c, e->inum); | |
1510 | if (err) | |
1511 | return err; | |
1512 | } else { | |
1513 | struct ubifs_ino_node *ino = c->sbuf; | |
1514 | ||
1515 | e->exists = 1; | |
1516 | e->i_size = le64_to_cpu(ino->size); | |
1517 | } | |
1518 | } | |
ff94bc40 | 1519 | |
9eefe2a2 | 1520 | if (e->exists && e->i_size < e->d_size) { |
ff94bc40 | 1521 | if (c->ro_mount) { |
9eefe2a2 SR |
1522 | /* Fix the inode size and pin it in memory */ |
1523 | struct inode *inode; | |
ff94bc40 HS |
1524 | struct ubifs_inode *ui; |
1525 | ||
1526 | ubifs_assert(!e->inode); | |
9eefe2a2 SR |
1527 | |
1528 | inode = ubifs_iget(c->vfs_sb, e->inum); | |
1529 | if (IS_ERR(inode)) | |
1530 | return PTR_ERR(inode); | |
ff94bc40 HS |
1531 | |
1532 | ui = ubifs_inode(inode); | |
9eefe2a2 SR |
1533 | if (inode->i_size < e->d_size) { |
1534 | dbg_rcvry("ino %lu size %lld -> %lld", | |
1535 | (unsigned long)e->inum, | |
ff94bc40 | 1536 | inode->i_size, e->d_size); |
9eefe2a2 | 1537 | inode->i_size = e->d_size; |
ff94bc40 HS |
1538 | ui->ui_size = e->d_size; |
1539 | ui->synced_i_size = e->d_size; | |
9eefe2a2 SR |
1540 | e->inode = inode; |
1541 | this = rb_next(this); | |
1542 | continue; | |
1543 | } | |
1544 | iput(inode); | |
ff94bc40 HS |
1545 | #ifndef __UBOOT__ |
1546 | } else { | |
1547 | /* Fix the size in place */ | |
1548 | err = fix_size_in_place(c, e); | |
1549 | if (err) | |
1550 | return err; | |
1551 | if (e->inode) | |
1552 | iput(e->inode); | |
1553 | #endif | |
9eefe2a2 SR |
1554 | } |
1555 | } | |
ff94bc40 | 1556 | |
9eefe2a2 SR |
1557 | this = rb_next(this); |
1558 | rb_erase(&e->rb, &c->size_tree); | |
1559 | kfree(e); | |
1560 | } | |
ff94bc40 | 1561 | |
9eefe2a2 SR |
1562 | return 0; |
1563 | } |