]>
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 | * Author: Adrian Hunter | |
20 | */ | |
21 | ||
22 | #include "ubifs.h" | |
23 | ||
24 | /* | |
25 | * An orphan is an inode number whose inode node has been committed to the index | |
26 | * with a link count of zero. That happens when an open file is deleted | |
27 | * (unlinked) and then a commit is run. In the normal course of events the inode | |
28 | * would be deleted when the file is closed. However in the case of an unclean | |
29 | * unmount, orphans need to be accounted for. After an unclean unmount, the | |
30 | * orphans' inodes must be deleted which means either scanning the entire index | |
31 | * looking for them, or keeping a list on flash somewhere. This unit implements | |
32 | * the latter approach. | |
33 | * | |
34 | * The orphan area is a fixed number of LEBs situated between the LPT area and | |
35 | * the main area. The number of orphan area LEBs is specified when the file | |
36 | * system is created. The minimum number is 1. The size of the orphan area | |
37 | * should be so that it can hold the maximum number of orphans that are expected | |
38 | * to ever exist at one time. | |
39 | * | |
40 | * The number of orphans that can fit in a LEB is: | |
41 | * | |
42 | * (c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64) | |
43 | * | |
44 | * For example: a 15872 byte LEB can fit 1980 orphans so 1 LEB may be enough. | |
45 | * | |
46 | * Orphans are accumulated in a rb-tree. When an inode's link count drops to | |
47 | * zero, the inode number is added to the rb-tree. It is removed from the tree | |
48 | * when the inode is deleted. Any new orphans that are in the orphan tree when | |
49d128aa | 49 | * the commit is run, are written to the orphan area in 1 or more orphan nodes. |
1e51764a AB |
50 | * If the orphan area is full, it is consolidated to make space. There is |
51 | * always enough space because validation prevents the user from creating more | |
52 | * than the maximum number of orphans allowed. | |
53 | */ | |
54 | ||
1e51764a | 55 | static int dbg_check_orphans(struct ubifs_info *c); |
1e51764a AB |
56 | |
57 | /** | |
58 | * ubifs_add_orphan - add an orphan. | |
59 | * @c: UBIFS file-system description object | |
60 | * @inum: orphan inode number | |
61 | * | |
62 | * Add an orphan. This function is called when an inodes link count drops to | |
63 | * zero. | |
64 | */ | |
65 | int ubifs_add_orphan(struct ubifs_info *c, ino_t inum) | |
66 | { | |
67 | struct ubifs_orphan *orphan, *o; | |
68 | struct rb_node **p, *parent = NULL; | |
69 | ||
70 | orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_NOFS); | |
71 | if (!orphan) | |
72 | return -ENOMEM; | |
73 | orphan->inum = inum; | |
74 | orphan->new = 1; | |
75 | ||
76 | spin_lock(&c->orphan_lock); | |
77 | if (c->tot_orphans >= c->max_orphans) { | |
78 | spin_unlock(&c->orphan_lock); | |
79 | kfree(orphan); | |
80 | return -ENFILE; | |
81 | } | |
82 | p = &c->orph_tree.rb_node; | |
83 | while (*p) { | |
84 | parent = *p; | |
85 | o = rb_entry(parent, struct ubifs_orphan, rb); | |
86 | if (inum < o->inum) | |
87 | p = &(*p)->rb_left; | |
88 | else if (inum > o->inum) | |
89 | p = &(*p)->rb_right; | |
90 | else { | |
a6aae4dd | 91 | ubifs_err("orphaned twice"); |
1e51764a AB |
92 | spin_unlock(&c->orphan_lock); |
93 | kfree(orphan); | |
94 | return 0; | |
95 | } | |
96 | } | |
97 | c->tot_orphans += 1; | |
98 | c->new_orphans += 1; | |
99 | rb_link_node(&orphan->rb, parent, p); | |
100 | rb_insert_color(&orphan->rb, &c->orph_tree); | |
101 | list_add_tail(&orphan->list, &c->orph_list); | |
102 | list_add_tail(&orphan->new_list, &c->orph_new); | |
103 | spin_unlock(&c->orphan_lock); | |
e84461ad | 104 | dbg_gen("ino %lu", (unsigned long)inum); |
1e51764a AB |
105 | return 0; |
106 | } | |
107 | ||
108 | /** | |
109 | * ubifs_delete_orphan - delete an orphan. | |
110 | * @c: UBIFS file-system description object | |
111 | * @inum: orphan inode number | |
112 | * | |
113 | * Delete an orphan. This function is called when an inode is deleted. | |
114 | */ | |
115 | void ubifs_delete_orphan(struct ubifs_info *c, ino_t inum) | |
116 | { | |
117 | struct ubifs_orphan *o; | |
118 | struct rb_node *p; | |
119 | ||
120 | spin_lock(&c->orphan_lock); | |
121 | p = c->orph_tree.rb_node; | |
122 | while (p) { | |
123 | o = rb_entry(p, struct ubifs_orphan, rb); | |
124 | if (inum < o->inum) | |
125 | p = p->rb_left; | |
126 | else if (inum > o->inum) | |
127 | p = p->rb_right; | |
128 | else { | |
129 | if (o->dnext) { | |
130 | spin_unlock(&c->orphan_lock); | |
e84461ad AB |
131 | dbg_gen("deleted twice ino %lu", |
132 | (unsigned long)inum); | |
1e51764a AB |
133 | return; |
134 | } | |
2928f0d0 | 135 | if (o->cmt) { |
1e51764a AB |
136 | o->dnext = c->orph_dnext; |
137 | c->orph_dnext = o; | |
138 | spin_unlock(&c->orphan_lock); | |
e84461ad AB |
139 | dbg_gen("delete later ino %lu", |
140 | (unsigned long)inum); | |
1e51764a AB |
141 | return; |
142 | } | |
143 | rb_erase(p, &c->orph_tree); | |
144 | list_del(&o->list); | |
145 | c->tot_orphans -= 1; | |
146 | if (o->new) { | |
147 | list_del(&o->new_list); | |
148 | c->new_orphans -= 1; | |
149 | } | |
150 | spin_unlock(&c->orphan_lock); | |
151 | kfree(o); | |
e84461ad | 152 | dbg_gen("inum %lu", (unsigned long)inum); |
1e51764a AB |
153 | return; |
154 | } | |
155 | } | |
156 | spin_unlock(&c->orphan_lock); | |
a6aae4dd | 157 | ubifs_err("missing orphan ino %lu", (unsigned long)inum); |
7c46d0ae | 158 | dump_stack(); |
1e51764a AB |
159 | } |
160 | ||
161 | /** | |
162 | * ubifs_orphan_start_commit - start commit of orphans. | |
163 | * @c: UBIFS file-system description object | |
164 | * | |
165 | * Start commit of orphans. | |
166 | */ | |
167 | int ubifs_orphan_start_commit(struct ubifs_info *c) | |
168 | { | |
169 | struct ubifs_orphan *orphan, **last; | |
170 | ||
171 | spin_lock(&c->orphan_lock); | |
172 | last = &c->orph_cnext; | |
173 | list_for_each_entry(orphan, &c->orph_new, new_list) { | |
174 | ubifs_assert(orphan->new); | |
2928f0d0 | 175 | ubifs_assert(!orphan->cmt); |
1e51764a | 176 | orphan->new = 0; |
2928f0d0 | 177 | orphan->cmt = 1; |
1e51764a AB |
178 | *last = orphan; |
179 | last = &orphan->cnext; | |
180 | } | |
7074e5eb | 181 | *last = NULL; |
1e51764a AB |
182 | c->cmt_orphans = c->new_orphans; |
183 | c->new_orphans = 0; | |
184 | dbg_cmt("%d orphans to commit", c->cmt_orphans); | |
185 | INIT_LIST_HEAD(&c->orph_new); | |
186 | if (c->tot_orphans == 0) | |
187 | c->no_orphs = 1; | |
188 | else | |
189 | c->no_orphs = 0; | |
190 | spin_unlock(&c->orphan_lock); | |
191 | return 0; | |
192 | } | |
193 | ||
194 | /** | |
195 | * avail_orphs - calculate available space. | |
196 | * @c: UBIFS file-system description object | |
197 | * | |
198 | * This function returns the number of orphans that can be written in the | |
199 | * available space. | |
200 | */ | |
201 | static int avail_orphs(struct ubifs_info *c) | |
202 | { | |
203 | int avail_lebs, avail, gap; | |
204 | ||
205 | avail_lebs = c->orph_lebs - (c->ohead_lnum - c->orph_first) - 1; | |
206 | avail = avail_lebs * | |
207 | ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64)); | |
208 | gap = c->leb_size - c->ohead_offs; | |
209 | if (gap >= UBIFS_ORPH_NODE_SZ + sizeof(__le64)) | |
210 | avail += (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64); | |
211 | return avail; | |
212 | } | |
213 | ||
214 | /** | |
215 | * tot_avail_orphs - calculate total space. | |
216 | * @c: UBIFS file-system description object | |
217 | * | |
218 | * This function returns the number of orphans that can be written in half | |
219 | * the total space. That leaves half the space for adding new orphans. | |
220 | */ | |
221 | static int tot_avail_orphs(struct ubifs_info *c) | |
222 | { | |
223 | int avail_lebs, avail; | |
224 | ||
225 | avail_lebs = c->orph_lebs; | |
226 | avail = avail_lebs * | |
227 | ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64)); | |
228 | return avail / 2; | |
229 | } | |
230 | ||
231 | /** | |
49d128aa | 232 | * do_write_orph_node - write a node to the orphan head. |
1e51764a AB |
233 | * @c: UBIFS file-system description object |
234 | * @len: length of node | |
235 | * @atomic: write atomically | |
236 | * | |
237 | * This function writes a node to the orphan head from the orphan buffer. If | |
238 | * %atomic is not zero, then the write is done atomically. On success, %0 is | |
239 | * returned, otherwise a negative error code is returned. | |
240 | */ | |
241 | static int do_write_orph_node(struct ubifs_info *c, int len, int atomic) | |
242 | { | |
243 | int err = 0; | |
244 | ||
245 | if (atomic) { | |
246 | ubifs_assert(c->ohead_offs == 0); | |
247 | ubifs_prepare_node(c, c->orph_buf, len, 1); | |
248 | len = ALIGN(len, c->min_io_size); | |
b36a261e | 249 | err = ubifs_leb_change(c, c->ohead_lnum, c->orph_buf, len); |
1e51764a AB |
250 | } else { |
251 | if (c->ohead_offs == 0) { | |
252 | /* Ensure LEB has been unmapped */ | |
253 | err = ubifs_leb_unmap(c, c->ohead_lnum); | |
254 | if (err) | |
255 | return err; | |
256 | } | |
257 | err = ubifs_write_node(c, c->orph_buf, len, c->ohead_lnum, | |
b36a261e | 258 | c->ohead_offs); |
1e51764a AB |
259 | } |
260 | return err; | |
261 | } | |
262 | ||
263 | /** | |
49d128aa | 264 | * write_orph_node - write an orphan node. |
1e51764a AB |
265 | * @c: UBIFS file-system description object |
266 | * @atomic: write atomically | |
267 | * | |
49d128aa | 268 | * This function builds an orphan node from the cnext list and writes it to the |
1e51764a AB |
269 | * orphan head. On success, %0 is returned, otherwise a negative error code |
270 | * is returned. | |
271 | */ | |
272 | static int write_orph_node(struct ubifs_info *c, int atomic) | |
273 | { | |
274 | struct ubifs_orphan *orphan, *cnext; | |
275 | struct ubifs_orph_node *orph; | |
276 | int gap, err, len, cnt, i; | |
277 | ||
278 | ubifs_assert(c->cmt_orphans > 0); | |
279 | gap = c->leb_size - c->ohead_offs; | |
280 | if (gap < UBIFS_ORPH_NODE_SZ + sizeof(__le64)) { | |
281 | c->ohead_lnum += 1; | |
282 | c->ohead_offs = 0; | |
283 | gap = c->leb_size; | |
284 | if (c->ohead_lnum > c->orph_last) { | |
285 | /* | |
286 | * We limit the number of orphans so that this should | |
287 | * never happen. | |
288 | */ | |
289 | ubifs_err("out of space in orphan area"); | |
290 | return -EINVAL; | |
291 | } | |
292 | } | |
293 | cnt = (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64); | |
294 | if (cnt > c->cmt_orphans) | |
295 | cnt = c->cmt_orphans; | |
296 | len = UBIFS_ORPH_NODE_SZ + cnt * sizeof(__le64); | |
297 | ubifs_assert(c->orph_buf); | |
298 | orph = c->orph_buf; | |
299 | orph->ch.node_type = UBIFS_ORPH_NODE; | |
300 | spin_lock(&c->orphan_lock); | |
301 | cnext = c->orph_cnext; | |
302 | for (i = 0; i < cnt; i++) { | |
303 | orphan = cnext; | |
2928f0d0 | 304 | ubifs_assert(orphan->cmt); |
1e51764a | 305 | orph->inos[i] = cpu_to_le64(orphan->inum); |
2928f0d0 | 306 | orphan->cmt = 0; |
1e51764a AB |
307 | cnext = orphan->cnext; |
308 | orphan->cnext = NULL; | |
309 | } | |
310 | c->orph_cnext = cnext; | |
311 | c->cmt_orphans -= cnt; | |
312 | spin_unlock(&c->orphan_lock); | |
313 | if (c->cmt_orphans) | |
014eb04b | 314 | orph->cmt_no = cpu_to_le64(c->cmt_no); |
1e51764a AB |
315 | else |
316 | /* Mark the last node of the commit */ | |
014eb04b | 317 | orph->cmt_no = cpu_to_le64((c->cmt_no) | (1ULL << 63)); |
1e51764a AB |
318 | ubifs_assert(c->ohead_offs + len <= c->leb_size); |
319 | ubifs_assert(c->ohead_lnum >= c->orph_first); | |
320 | ubifs_assert(c->ohead_lnum <= c->orph_last); | |
321 | err = do_write_orph_node(c, len, atomic); | |
322 | c->ohead_offs += ALIGN(len, c->min_io_size); | |
323 | c->ohead_offs = ALIGN(c->ohead_offs, 8); | |
324 | return err; | |
325 | } | |
326 | ||
327 | /** | |
49d128aa | 328 | * write_orph_nodes - write orphan nodes until there are no more to commit. |
1e51764a AB |
329 | * @c: UBIFS file-system description object |
330 | * @atomic: write atomically | |
331 | * | |
49d128aa | 332 | * This function writes orphan nodes for all the orphans to commit. On success, |
1e51764a AB |
333 | * %0 is returned, otherwise a negative error code is returned. |
334 | */ | |
335 | static int write_orph_nodes(struct ubifs_info *c, int atomic) | |
336 | { | |
337 | int err; | |
338 | ||
339 | while (c->cmt_orphans > 0) { | |
340 | err = write_orph_node(c, atomic); | |
341 | if (err) | |
342 | return err; | |
343 | } | |
344 | if (atomic) { | |
345 | int lnum; | |
346 | ||
347 | /* Unmap any unused LEBs after consolidation */ | |
348 | lnum = c->ohead_lnum + 1; | |
349 | for (lnum = c->ohead_lnum + 1; lnum <= c->orph_last; lnum++) { | |
350 | err = ubifs_leb_unmap(c, lnum); | |
351 | if (err) | |
352 | return err; | |
353 | } | |
354 | } | |
355 | return 0; | |
356 | } | |
357 | ||
358 | /** | |
359 | * consolidate - consolidate the orphan area. | |
360 | * @c: UBIFS file-system description object | |
361 | * | |
362 | * This function enables consolidation by putting all the orphans into the list | |
363 | * to commit. The list is in the order that the orphans were added, and the | |
364 | * LEBs are written atomically in order, so at no time can orphans be lost by | |
365 | * an unclean unmount. | |
366 | * | |
367 | * This function returns %0 on success and a negative error code on failure. | |
368 | */ | |
369 | static int consolidate(struct ubifs_info *c) | |
370 | { | |
371 | int tot_avail = tot_avail_orphs(c), err = 0; | |
372 | ||
373 | spin_lock(&c->orphan_lock); | |
374 | dbg_cmt("there is space for %d orphans and there are %d", | |
375 | tot_avail, c->tot_orphans); | |
376 | if (c->tot_orphans - c->new_orphans <= tot_avail) { | |
377 | struct ubifs_orphan *orphan, **last; | |
378 | int cnt = 0; | |
379 | ||
380 | /* Change the cnext list to include all non-new orphans */ | |
381 | last = &c->orph_cnext; | |
382 | list_for_each_entry(orphan, &c->orph_list, list) { | |
383 | if (orphan->new) | |
384 | continue; | |
2928f0d0 | 385 | orphan->cmt = 1; |
1e51764a AB |
386 | *last = orphan; |
387 | last = &orphan->cnext; | |
388 | cnt += 1; | |
389 | } | |
7074e5eb | 390 | *last = NULL; |
1e51764a AB |
391 | ubifs_assert(cnt == c->tot_orphans - c->new_orphans); |
392 | c->cmt_orphans = cnt; | |
393 | c->ohead_lnum = c->orph_first; | |
394 | c->ohead_offs = 0; | |
395 | } else { | |
396 | /* | |
397 | * We limit the number of orphans so that this should | |
398 | * never happen. | |
399 | */ | |
400 | ubifs_err("out of space in orphan area"); | |
401 | err = -EINVAL; | |
402 | } | |
403 | spin_unlock(&c->orphan_lock); | |
404 | return err; | |
405 | } | |
406 | ||
407 | /** | |
408 | * commit_orphans - commit orphans. | |
409 | * @c: UBIFS file-system description object | |
410 | * | |
411 | * This function commits orphans to flash. On success, %0 is returned, | |
412 | * otherwise a negative error code is returned. | |
413 | */ | |
414 | static int commit_orphans(struct ubifs_info *c) | |
415 | { | |
416 | int avail, atomic = 0, err; | |
417 | ||
418 | ubifs_assert(c->cmt_orphans > 0); | |
419 | avail = avail_orphs(c); | |
420 | if (avail < c->cmt_orphans) { | |
421 | /* Not enough space to write new orphans, so consolidate */ | |
422 | err = consolidate(c); | |
423 | if (err) | |
424 | return err; | |
425 | atomic = 1; | |
426 | } | |
427 | err = write_orph_nodes(c, atomic); | |
428 | return err; | |
429 | } | |
430 | ||
431 | /** | |
432 | * erase_deleted - erase the orphans marked for deletion. | |
433 | * @c: UBIFS file-system description object | |
434 | * | |
435 | * During commit, the orphans being committed cannot be deleted, so they are | |
436 | * marked for deletion and deleted by this function. Also, the recovery | |
437 | * adds killed orphans to the deletion list, and therefore they are deleted | |
438 | * here too. | |
439 | */ | |
440 | static void erase_deleted(struct ubifs_info *c) | |
441 | { | |
442 | struct ubifs_orphan *orphan, *dnext; | |
443 | ||
444 | spin_lock(&c->orphan_lock); | |
445 | dnext = c->orph_dnext; | |
446 | while (dnext) { | |
447 | orphan = dnext; | |
448 | dnext = orphan->dnext; | |
449 | ubifs_assert(!orphan->new); | |
450 | rb_erase(&orphan->rb, &c->orph_tree); | |
451 | list_del(&orphan->list); | |
452 | c->tot_orphans -= 1; | |
e84461ad | 453 | dbg_gen("deleting orphan ino %lu", (unsigned long)orphan->inum); |
1e51764a AB |
454 | kfree(orphan); |
455 | } | |
456 | c->orph_dnext = NULL; | |
457 | spin_unlock(&c->orphan_lock); | |
458 | } | |
459 | ||
460 | /** | |
461 | * ubifs_orphan_end_commit - end commit of orphans. | |
462 | * @c: UBIFS file-system description object | |
463 | * | |
464 | * End commit of orphans. | |
465 | */ | |
466 | int ubifs_orphan_end_commit(struct ubifs_info *c) | |
467 | { | |
468 | int err; | |
469 | ||
470 | if (c->cmt_orphans != 0) { | |
471 | err = commit_orphans(c); | |
472 | if (err) | |
473 | return err; | |
474 | } | |
475 | erase_deleted(c); | |
476 | err = dbg_check_orphans(c); | |
477 | return err; | |
478 | } | |
479 | ||
480 | /** | |
49d128aa | 481 | * ubifs_clear_orphans - erase all LEBs used for orphans. |
1e51764a AB |
482 | * @c: UBIFS file-system description object |
483 | * | |
484 | * If recovery is not required, then the orphans from the previous session | |
485 | * are not needed. This function locates the LEBs used to record | |
486 | * orphans, and un-maps them. | |
487 | */ | |
49d128aa | 488 | int ubifs_clear_orphans(struct ubifs_info *c) |
1e51764a AB |
489 | { |
490 | int lnum, err; | |
491 | ||
492 | for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) { | |
493 | err = ubifs_leb_unmap(c, lnum); | |
494 | if (err) | |
495 | return err; | |
496 | } | |
497 | c->ohead_lnum = c->orph_first; | |
498 | c->ohead_offs = 0; | |
499 | return 0; | |
500 | } | |
501 | ||
502 | /** | |
503 | * insert_dead_orphan - insert an orphan. | |
504 | * @c: UBIFS file-system description object | |
505 | * @inum: orphan inode number | |
506 | * | |
507 | * This function is a helper to the 'do_kill_orphans()' function. The orphan | |
508 | * must be kept until the next commit, so it is added to the rb-tree and the | |
509 | * deletion list. | |
510 | */ | |
511 | static int insert_dead_orphan(struct ubifs_info *c, ino_t inum) | |
512 | { | |
513 | struct ubifs_orphan *orphan, *o; | |
514 | struct rb_node **p, *parent = NULL; | |
515 | ||
516 | orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_KERNEL); | |
517 | if (!orphan) | |
518 | return -ENOMEM; | |
519 | orphan->inum = inum; | |
520 | ||
521 | p = &c->orph_tree.rb_node; | |
522 | while (*p) { | |
523 | parent = *p; | |
524 | o = rb_entry(parent, struct ubifs_orphan, rb); | |
525 | if (inum < o->inum) | |
526 | p = &(*p)->rb_left; | |
527 | else if (inum > o->inum) | |
528 | p = &(*p)->rb_right; | |
529 | else { | |
530 | /* Already added - no problem */ | |
531 | kfree(orphan); | |
532 | return 0; | |
533 | } | |
534 | } | |
535 | c->tot_orphans += 1; | |
536 | rb_link_node(&orphan->rb, parent, p); | |
537 | rb_insert_color(&orphan->rb, &c->orph_tree); | |
538 | list_add_tail(&orphan->list, &c->orph_list); | |
539 | orphan->dnext = c->orph_dnext; | |
540 | c->orph_dnext = orphan; | |
e84461ad AB |
541 | dbg_mnt("ino %lu, new %d, tot %d", (unsigned long)inum, |
542 | c->new_orphans, c->tot_orphans); | |
1e51764a AB |
543 | return 0; |
544 | } | |
545 | ||
546 | /** | |
547 | * do_kill_orphans - remove orphan inodes from the index. | |
548 | * @c: UBIFS file-system description object | |
549 | * @sleb: scanned LEB | |
49d128aa | 550 | * @last_cmt_no: cmt_no of last orphan node read is passed and returned here |
1e51764a | 551 | * @outofdate: whether the LEB is out of date is returned here |
49d128aa | 552 | * @last_flagged: whether the end orphan node is encountered |
1e51764a AB |
553 | * |
554 | * This function is a helper to the 'kill_orphans()' function. It goes through | |
555 | * every orphan node in a LEB and for every inode number recorded, removes | |
556 | * all keys for that inode from the TNC. | |
557 | */ | |
558 | static int do_kill_orphans(struct ubifs_info *c, struct ubifs_scan_leb *sleb, | |
559 | unsigned long long *last_cmt_no, int *outofdate, | |
560 | int *last_flagged) | |
561 | { | |
562 | struct ubifs_scan_node *snod; | |
563 | struct ubifs_orph_node *orph; | |
564 | unsigned long long cmt_no; | |
565 | ino_t inum; | |
566 | int i, n, err, first = 1; | |
567 | ||
568 | list_for_each_entry(snod, &sleb->nodes, list) { | |
569 | if (snod->type != UBIFS_ORPH_NODE) { | |
79fda517 AB |
570 | ubifs_err("invalid node type %d in orphan area at %d:%d", |
571 | snod->type, sleb->lnum, snod->offs); | |
edf6be24 | 572 | ubifs_dump_node(c, snod->node); |
1e51764a AB |
573 | return -EINVAL; |
574 | } | |
575 | ||
576 | orph = snod->node; | |
577 | ||
578 | /* Check commit number */ | |
579 | cmt_no = le64_to_cpu(orph->cmt_no) & LLONG_MAX; | |
580 | /* | |
581 | * The commit number on the master node may be less, because | |
582 | * of a failed commit. If there are several failed commits in a | |
49d128aa AH |
583 | * row, the commit number written on orphan nodes will continue |
584 | * to increase (because the commit number is adjusted here) even | |
1e51764a AB |
585 | * though the commit number on the master node stays the same |
586 | * because the master node has not been re-written. | |
587 | */ | |
588 | if (cmt_no > c->cmt_no) | |
589 | c->cmt_no = cmt_no; | |
590 | if (cmt_no < *last_cmt_no && *last_flagged) { | |
591 | /* | |
49d128aa AH |
592 | * The last orphan node had a higher commit number and |
593 | * was flagged as the last written for that commit | |
594 | * number. That makes this orphan node, out of date. | |
1e51764a AB |
595 | */ |
596 | if (!first) { | |
79fda517 | 597 | ubifs_err("out of order commit number %llu in orphan node at %d:%d", |
1e51764a | 598 | cmt_no, sleb->lnum, snod->offs); |
edf6be24 | 599 | ubifs_dump_node(c, snod->node); |
1e51764a AB |
600 | return -EINVAL; |
601 | } | |
602 | dbg_rcvry("out of date LEB %d", sleb->lnum); | |
603 | *outofdate = 1; | |
604 | return 0; | |
605 | } | |
606 | ||
607 | if (first) | |
608 | first = 0; | |
609 | ||
610 | n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3; | |
611 | for (i = 0; i < n; i++) { | |
612 | inum = le64_to_cpu(orph->inos[i]); | |
e84461ad AB |
613 | dbg_rcvry("deleting orphaned inode %lu", |
614 | (unsigned long)inum); | |
1e51764a AB |
615 | err = ubifs_tnc_remove_ino(c, inum); |
616 | if (err) | |
617 | return err; | |
618 | err = insert_dead_orphan(c, inum); | |
619 | if (err) | |
620 | return err; | |
621 | } | |
622 | ||
623 | *last_cmt_no = cmt_no; | |
624 | if (le64_to_cpu(orph->cmt_no) & (1ULL << 63)) { | |
625 | dbg_rcvry("last orph node for commit %llu at %d:%d", | |
626 | cmt_no, sleb->lnum, snod->offs); | |
627 | *last_flagged = 1; | |
628 | } else | |
629 | *last_flagged = 0; | |
630 | } | |
631 | ||
632 | return 0; | |
633 | } | |
634 | ||
635 | /** | |
636 | * kill_orphans - remove all orphan inodes from the index. | |
637 | * @c: UBIFS file-system description object | |
638 | * | |
639 | * If recovery is required, then orphan inodes recorded during the previous | |
640 | * session (which ended with an unclean unmount) must be deleted from the index. | |
641 | * This is done by updating the TNC, but since the index is not updated until | |
642 | * the next commit, the LEBs where the orphan information is recorded are not | |
643 | * erased until the next commit. | |
644 | */ | |
645 | static int kill_orphans(struct ubifs_info *c) | |
646 | { | |
647 | unsigned long long last_cmt_no = 0; | |
648 | int lnum, err = 0, outofdate = 0, last_flagged = 0; | |
649 | ||
650 | c->ohead_lnum = c->orph_first; | |
651 | c->ohead_offs = 0; | |
652 | /* Check no-orphans flag and skip this if no orphans */ | |
653 | if (c->no_orphs) { | |
654 | dbg_rcvry("no orphans"); | |
655 | return 0; | |
656 | } | |
657 | /* | |
658 | * Orph nodes always start at c->orph_first and are written to each | |
659 | * successive LEB in turn. Generally unused LEBs will have been unmapped | |
49d128aa AH |
660 | * but may contain out of date orphan nodes if the unmap didn't go |
661 | * through. In addition, the last orphan node written for each commit is | |
1e51764a | 662 | * marked (top bit of orph->cmt_no is set to 1). It is possible that |
49d128aa | 663 | * there are orphan nodes from the next commit (i.e. the commit did not |
1e51764a AB |
664 | * complete successfully). In that case, no orphans will have been lost |
665 | * due to the way that orphans are written, and any orphans added will | |
666 | * be valid orphans anyway and so can be deleted. | |
667 | */ | |
668 | for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) { | |
669 | struct ubifs_scan_leb *sleb; | |
670 | ||
671 | dbg_rcvry("LEB %d", lnum); | |
348709ba | 672 | sleb = ubifs_scan(c, lnum, 0, c->sbuf, 1); |
1e51764a | 673 | if (IS_ERR(sleb)) { |
0dcd18e4 | 674 | if (PTR_ERR(sleb) == -EUCLEAN) |
c4361570 | 675 | sleb = ubifs_recover_leb(c, lnum, 0, |
efcfde54 | 676 | c->sbuf, -1); |
1e51764a AB |
677 | if (IS_ERR(sleb)) { |
678 | err = PTR_ERR(sleb); | |
679 | break; | |
680 | } | |
681 | } | |
682 | err = do_kill_orphans(c, sleb, &last_cmt_no, &outofdate, | |
683 | &last_flagged); | |
684 | if (err || outofdate) { | |
685 | ubifs_scan_destroy(sleb); | |
686 | break; | |
687 | } | |
688 | if (sleb->endpt) { | |
689 | c->ohead_lnum = lnum; | |
690 | c->ohead_offs = sleb->endpt; | |
691 | } | |
692 | ubifs_scan_destroy(sleb); | |
693 | } | |
694 | return err; | |
695 | } | |
696 | ||
697 | /** | |
698 | * ubifs_mount_orphans - delete orphan inodes and erase LEBs that recorded them. | |
699 | * @c: UBIFS file-system description object | |
700 | * @unclean: indicates recovery from unclean unmount | |
701 | * @read_only: indicates read only mount | |
702 | * | |
703 | * This function is called when mounting to erase orphans from the previous | |
704 | * session. If UBIFS was not unmounted cleanly, then the inodes recorded as | |
705 | * orphans are deleted. | |
706 | */ | |
707 | int ubifs_mount_orphans(struct ubifs_info *c, int unclean, int read_only) | |
708 | { | |
709 | int err = 0; | |
710 | ||
711 | c->max_orphans = tot_avail_orphs(c); | |
712 | ||
713 | if (!read_only) { | |
714 | c->orph_buf = vmalloc(c->leb_size); | |
715 | if (!c->orph_buf) | |
716 | return -ENOMEM; | |
717 | } | |
718 | ||
719 | if (unclean) | |
720 | err = kill_orphans(c); | |
721 | else if (!read_only) | |
49d128aa | 722 | err = ubifs_clear_orphans(c); |
1e51764a AB |
723 | |
724 | return err; | |
725 | } | |
726 | ||
f70b7e52 AB |
727 | /* |
728 | * Everything below is related to debugging. | |
729 | */ | |
1e51764a AB |
730 | |
731 | struct check_orphan { | |
732 | struct rb_node rb; | |
733 | ino_t inum; | |
734 | }; | |
735 | ||
736 | struct check_info { | |
737 | unsigned long last_ino; | |
738 | unsigned long tot_inos; | |
739 | unsigned long missing; | |
740 | unsigned long long leaf_cnt; | |
741 | struct ubifs_ino_node *node; | |
742 | struct rb_root root; | |
743 | }; | |
744 | ||
745 | static int dbg_find_orphan(struct ubifs_info *c, ino_t inum) | |
746 | { | |
747 | struct ubifs_orphan *o; | |
748 | struct rb_node *p; | |
749 | ||
750 | spin_lock(&c->orphan_lock); | |
751 | p = c->orph_tree.rb_node; | |
752 | while (p) { | |
753 | o = rb_entry(p, struct ubifs_orphan, rb); | |
754 | if (inum < o->inum) | |
755 | p = p->rb_left; | |
756 | else if (inum > o->inum) | |
757 | p = p->rb_right; | |
758 | else { | |
759 | spin_unlock(&c->orphan_lock); | |
760 | return 1; | |
761 | } | |
762 | } | |
763 | spin_unlock(&c->orphan_lock); | |
764 | return 0; | |
765 | } | |
766 | ||
767 | static int dbg_ins_check_orphan(struct rb_root *root, ino_t inum) | |
768 | { | |
769 | struct check_orphan *orphan, *o; | |
770 | struct rb_node **p, *parent = NULL; | |
771 | ||
772 | orphan = kzalloc(sizeof(struct check_orphan), GFP_NOFS); | |
773 | if (!orphan) | |
774 | return -ENOMEM; | |
775 | orphan->inum = inum; | |
776 | ||
777 | p = &root->rb_node; | |
778 | while (*p) { | |
779 | parent = *p; | |
780 | o = rb_entry(parent, struct check_orphan, rb); | |
781 | if (inum < o->inum) | |
782 | p = &(*p)->rb_left; | |
783 | else if (inum > o->inum) | |
784 | p = &(*p)->rb_right; | |
785 | else { | |
786 | kfree(orphan); | |
787 | return 0; | |
788 | } | |
789 | } | |
790 | rb_link_node(&orphan->rb, parent, p); | |
791 | rb_insert_color(&orphan->rb, root); | |
792 | return 0; | |
793 | } | |
794 | ||
795 | static int dbg_find_check_orphan(struct rb_root *root, ino_t inum) | |
796 | { | |
797 | struct check_orphan *o; | |
798 | struct rb_node *p; | |
799 | ||
800 | p = root->rb_node; | |
801 | while (p) { | |
802 | o = rb_entry(p, struct check_orphan, rb); | |
803 | if (inum < o->inum) | |
804 | p = p->rb_left; | |
805 | else if (inum > o->inum) | |
806 | p = p->rb_right; | |
807 | else | |
808 | return 1; | |
809 | } | |
810 | return 0; | |
811 | } | |
812 | ||
813 | static void dbg_free_check_tree(struct rb_root *root) | |
814 | { | |
815 | struct rb_node *this = root->rb_node; | |
816 | struct check_orphan *o; | |
817 | ||
818 | while (this) { | |
819 | if (this->rb_left) { | |
820 | this = this->rb_left; | |
821 | continue; | |
822 | } else if (this->rb_right) { | |
823 | this = this->rb_right; | |
824 | continue; | |
825 | } | |
826 | o = rb_entry(this, struct check_orphan, rb); | |
827 | this = rb_parent(this); | |
828 | if (this) { | |
829 | if (this->rb_left == &o->rb) | |
830 | this->rb_left = NULL; | |
831 | else | |
832 | this->rb_right = NULL; | |
833 | } | |
834 | kfree(o); | |
835 | } | |
836 | } | |
837 | ||
838 | static int dbg_orphan_check(struct ubifs_info *c, struct ubifs_zbranch *zbr, | |
839 | void *priv) | |
840 | { | |
841 | struct check_info *ci = priv; | |
842 | ino_t inum; | |
843 | int err; | |
844 | ||
845 | inum = key_inum(c, &zbr->key); | |
846 | if (inum != ci->last_ino) { | |
847 | /* Lowest node type is the inode node, so it comes first */ | |
848 | if (key_type(c, &zbr->key) != UBIFS_INO_KEY) | |
e84461ad AB |
849 | ubifs_err("found orphan node ino %lu, type %d", |
850 | (unsigned long)inum, key_type(c, &zbr->key)); | |
1e51764a AB |
851 | ci->last_ino = inum; |
852 | ci->tot_inos += 1; | |
853 | err = ubifs_tnc_read_node(c, zbr, ci->node); | |
854 | if (err) { | |
855 | ubifs_err("node read failed, error %d", err); | |
856 | return err; | |
857 | } | |
858 | if (ci->node->nlink == 0) | |
859 | /* Must be recorded as an orphan */ | |
860 | if (!dbg_find_check_orphan(&ci->root, inum) && | |
861 | !dbg_find_orphan(c, inum)) { | |
e84461ad AB |
862 | ubifs_err("missing orphan, ino %lu", |
863 | (unsigned long)inum); | |
1e51764a AB |
864 | ci->missing += 1; |
865 | } | |
866 | } | |
867 | ci->leaf_cnt += 1; | |
868 | return 0; | |
869 | } | |
870 | ||
871 | static int dbg_read_orphans(struct check_info *ci, struct ubifs_scan_leb *sleb) | |
872 | { | |
873 | struct ubifs_scan_node *snod; | |
874 | struct ubifs_orph_node *orph; | |
875 | ino_t inum; | |
876 | int i, n, err; | |
877 | ||
878 | list_for_each_entry(snod, &sleb->nodes, list) { | |
879 | cond_resched(); | |
880 | if (snod->type != UBIFS_ORPH_NODE) | |
881 | continue; | |
882 | orph = snod->node; | |
883 | n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3; | |
884 | for (i = 0; i < n; i++) { | |
885 | inum = le64_to_cpu(orph->inos[i]); | |
886 | err = dbg_ins_check_orphan(&ci->root, inum); | |
887 | if (err) | |
888 | return err; | |
889 | } | |
890 | } | |
891 | return 0; | |
892 | } | |
893 | ||
894 | static int dbg_scan_orphans(struct ubifs_info *c, struct check_info *ci) | |
895 | { | |
896 | int lnum, err = 0; | |
f5cf319c | 897 | void *buf; |
1e51764a AB |
898 | |
899 | /* Check no-orphans flag and skip this if no orphans */ | |
900 | if (c->no_orphs) | |
901 | return 0; | |
902 | ||
fc5e58c0 | 903 | buf = __vmalloc(c->leb_size, GFP_NOFS, PAGE_KERNEL); |
f5cf319c AB |
904 | if (!buf) { |
905 | ubifs_err("cannot allocate memory to check orphans"); | |
906 | return 0; | |
907 | } | |
908 | ||
1e51764a AB |
909 | for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) { |
910 | struct ubifs_scan_leb *sleb; | |
911 | ||
f5cf319c | 912 | sleb = ubifs_scan(c, lnum, 0, buf, 0); |
1e51764a AB |
913 | if (IS_ERR(sleb)) { |
914 | err = PTR_ERR(sleb); | |
915 | break; | |
916 | } | |
917 | ||
918 | err = dbg_read_orphans(ci, sleb); | |
919 | ubifs_scan_destroy(sleb); | |
920 | if (err) | |
921 | break; | |
922 | } | |
923 | ||
f5cf319c | 924 | vfree(buf); |
1e51764a AB |
925 | return err; |
926 | } | |
927 | ||
928 | static int dbg_check_orphans(struct ubifs_info *c) | |
929 | { | |
930 | struct check_info ci; | |
931 | int err; | |
932 | ||
2b1844a8 | 933 | if (!dbg_is_chk_orph(c)) |
1e51764a AB |
934 | return 0; |
935 | ||
936 | ci.last_ino = 0; | |
937 | ci.tot_inos = 0; | |
938 | ci.missing = 0; | |
939 | ci.leaf_cnt = 0; | |
940 | ci.root = RB_ROOT; | |
941 | ci.node = kmalloc(UBIFS_MAX_INO_NODE_SZ, GFP_NOFS); | |
942 | if (!ci.node) { | |
943 | ubifs_err("out of memory"); | |
944 | return -ENOMEM; | |
945 | } | |
946 | ||
947 | err = dbg_scan_orphans(c, &ci); | |
948 | if (err) | |
949 | goto out; | |
950 | ||
951 | err = dbg_walk_index(c, &dbg_orphan_check, NULL, &ci); | |
952 | if (err) { | |
953 | ubifs_err("cannot scan TNC, error %d", err); | |
954 | goto out; | |
955 | } | |
956 | ||
957 | if (ci.missing) { | |
958 | ubifs_err("%lu missing orphan(s)", ci.missing); | |
959 | err = -EINVAL; | |
960 | goto out; | |
961 | } | |
962 | ||
963 | dbg_cmt("last inode number is %lu", ci.last_ino); | |
964 | dbg_cmt("total number of inodes is %lu", ci.tot_inos); | |
965 | dbg_cmt("total number of leaf nodes is %llu", ci.leaf_cnt); | |
966 | ||
967 | out: | |
968 | dbg_free_check_tree(&ci.root); | |
969 | kfree(ci.node); | |
970 | return err; | |
971 | } |