1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * This file is part of UBIFS.
4 *
5 * Copyright (C) 2006-2008 Nokia Corporation.
6 *
7 * Author: Adrian Hunter
8 */
9
10 #include "ubifs.h"
11
12 /*
13 * An orphan is an inode number whose inode node has been committed to the index
14 * with a link count of zero. That happens when an open file is deleted
15 * (unlinked) and then a commit is run. In the normal course of events the inode
16 * would be deleted when the file is closed. However in the case of an unclean
17 * unmount, orphans need to be accounted for. After an unclean unmount, the
18 * orphans' inodes must be deleted which means either scanning the entire index
19 * looking for them, or keeping a list on flash somewhere. This unit implements
20 * the latter approach.
21 *
22 * The orphan area is a fixed number of LEBs situated between the LPT area and
23 * the main area. The number of orphan area LEBs is specified when the file
24 * system is created. The minimum number is 1. The size of the orphan area
25 * should be so that it can hold the maximum number of orphans that are expected
26 * to ever exist at one time.
27 *
28 * The number of orphans that can fit in a LEB is:
29 *
30 * (c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64)
31 *
32 * For example: a 15872 byte LEB can fit 1980 orphans so 1 LEB may be enough.
33 *
34 * Orphans are accumulated in a rb-tree. When an inode's link count drops to
35 * zero, the inode number is added to the rb-tree. It is removed from the tree
36 * when the inode is deleted. Any new orphans that are in the orphan tree when
37 * the commit is run, are written to the orphan area in 1 or more orphan nodes.
38 * If the orphan area is full, it is consolidated to make space. There is
39 * always enough space because validation prevents the user from creating more
40 * than the maximum number of orphans allowed.
41 */
42
43 static int dbg_check_orphans(struct ubifs_info *c);
44
orphan_add(struct ubifs_info * c,ino_t inum,struct ubifs_orphan * parent_orphan)45 static struct ubifs_orphan *orphan_add(struct ubifs_info *c, ino_t inum,
46 struct ubifs_orphan *parent_orphan)
47 {
48 struct ubifs_orphan *orphan, *o;
49 struct rb_node **p, *parent = NULL;
50
51 orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_NOFS);
52 if (!orphan)
53 return ERR_PTR(-ENOMEM);
54 orphan->inum = inum;
55 orphan->new = 1;
56 INIT_LIST_HEAD(&orphan->child_list);
57
58 spin_lock(&c->orphan_lock);
59 if (c->tot_orphans >= c->max_orphans) {
60 spin_unlock(&c->orphan_lock);
61 kfree(orphan);
62 return ERR_PTR(-ENFILE);
63 }
64 p = &c->orph_tree.rb_node;
65 while (*p) {
66 parent = *p;
67 o = rb_entry(parent, struct ubifs_orphan, rb);
68 if (inum < o->inum)
69 p = &(*p)->rb_left;
70 else if (inum > o->inum)
71 p = &(*p)->rb_right;
72 else {
73 ubifs_err(c, "orphaned twice");
74 spin_unlock(&c->orphan_lock);
75 kfree(orphan);
76 return ERR_PTR(-EINVAL);
77 }
78 }
79 c->tot_orphans += 1;
80 c->new_orphans += 1;
81 rb_link_node(&orphan->rb, parent, p);
82 rb_insert_color(&orphan->rb, &c->orph_tree);
83 list_add_tail(&orphan->list, &c->orph_list);
84 list_add_tail(&orphan->new_list, &c->orph_new);
85
86 if (parent_orphan) {
87 list_add_tail(&orphan->child_list,
88 &parent_orphan->child_list);
89 }
90
91 spin_unlock(&c->orphan_lock);
92 dbg_gen("ino %lu", (unsigned long)inum);
93 return orphan;
94 }
95
lookup_orphan(struct ubifs_info * c,ino_t inum)96 static struct ubifs_orphan *lookup_orphan(struct ubifs_info *c, ino_t inum)
97 {
98 struct ubifs_orphan *o;
99 struct rb_node *p;
100
101 p = c->orph_tree.rb_node;
102 while (p) {
103 o = rb_entry(p, struct ubifs_orphan, rb);
104 if (inum < o->inum)
105 p = p->rb_left;
106 else if (inum > o->inum)
107 p = p->rb_right;
108 else {
109 return o;
110 }
111 }
112 return NULL;
113 }
114
__orphan_drop(struct ubifs_info * c,struct ubifs_orphan * o)115 static void __orphan_drop(struct ubifs_info *c, struct ubifs_orphan *o)
116 {
117 rb_erase(&o->rb, &c->orph_tree);
118 list_del(&o->list);
119 c->tot_orphans -= 1;
120
121 if (o->new) {
122 list_del(&o->new_list);
123 c->new_orphans -= 1;
124 }
125
126 kfree(o);
127 }
128
orphan_delete(struct ubifs_info * c,struct ubifs_orphan * orph)129 static void orphan_delete(struct ubifs_info *c, struct ubifs_orphan *orph)
130 {
131 if (orph->del) {
132 dbg_gen("deleted twice ino %lu", (unsigned long)orph->inum);
133 return;
134 }
135
136 if (orph->cmt) {
137 orph->del = 1;
138 orph->dnext = c->orph_dnext;
139 c->orph_dnext = orph;
140 dbg_gen("delete later ino %lu", (unsigned long)orph->inum);
141 return;
142 }
143
144 __orphan_drop(c, orph);
145 }
146
147 /**
148 * ubifs_add_orphan - add an orphan.
149 * @c: UBIFS file-system description object
150 * @inum: orphan inode number
151 *
152 * Add an orphan. This function is called when an inodes link count drops to
153 * zero.
154 */
ubifs_add_orphan(struct ubifs_info * c,ino_t inum)155 int ubifs_add_orphan(struct ubifs_info *c, ino_t inum)
156 {
157 int err = 0;
158 ino_t xattr_inum;
159 union ubifs_key key;
160 struct ubifs_dent_node *xent, *pxent = NULL;
161 struct fscrypt_name nm = {0};
162 struct ubifs_orphan *xattr_orphan;
163 struct ubifs_orphan *orphan;
164
165 orphan = orphan_add(c, inum, NULL);
166 if (IS_ERR(orphan))
167 return PTR_ERR(orphan);
168
169 lowest_xent_key(c, &key, inum);
170 while (1) {
171 xent = ubifs_tnc_next_ent(c, &key, &nm);
172 if (IS_ERR(xent)) {
173 err = PTR_ERR(xent);
174 if (err == -ENOENT)
175 break;
176 kfree(pxent);
177 return err;
178 }
179
180 fname_name(&nm) = xent->name;
181 fname_len(&nm) = le16_to_cpu(xent->nlen);
182 xattr_inum = le64_to_cpu(xent->inum);
183
184 xattr_orphan = orphan_add(c, xattr_inum, orphan);
185 if (IS_ERR(xattr_orphan)) {
186 kfree(pxent);
187 kfree(xent);
188 return PTR_ERR(xattr_orphan);
189 }
190
191 kfree(pxent);
192 pxent = xent;
193 key_read(c, &xent->key, &key);
194 }
195 kfree(pxent);
196
197 return 0;
198 }
199
200 /**
201 * ubifs_delete_orphan - delete an orphan.
202 * @c: UBIFS file-system description object
203 * @inum: orphan inode number
204 *
205 * Delete an orphan. This function is called when an inode is deleted.
206 */
ubifs_delete_orphan(struct ubifs_info * c,ino_t inum)207 void ubifs_delete_orphan(struct ubifs_info *c, ino_t inum)
208 {
209 struct ubifs_orphan *orph, *child_orph, *tmp_o;
210
211 spin_lock(&c->orphan_lock);
212
213 orph = lookup_orphan(c, inum);
214 if (!orph) {
215 spin_unlock(&c->orphan_lock);
216 ubifs_err(c, "missing orphan ino %lu", (unsigned long)inum);
217 dump_stack();
218
219 return;
220 }
221
222 list_for_each_entry_safe(child_orph, tmp_o, &orph->child_list, child_list) {
223 list_del(&child_orph->child_list);
224 orphan_delete(c, child_orph);
225 }
226
227 orphan_delete(c, orph);
228
229 spin_unlock(&c->orphan_lock);
230 }
231
232 /**
233 * ubifs_orphan_start_commit - start commit of orphans.
234 * @c: UBIFS file-system description object
235 *
236 * Start commit of orphans.
237 */
ubifs_orphan_start_commit(struct ubifs_info * c)238 int ubifs_orphan_start_commit(struct ubifs_info *c)
239 {
240 struct ubifs_orphan *orphan, **last;
241
242 spin_lock(&c->orphan_lock);
243 last = &c->orph_cnext;
244 list_for_each_entry(orphan, &c->orph_new, new_list) {
245 ubifs_assert(c, orphan->new);
246 ubifs_assert(c, !orphan->cmt);
247 orphan->new = 0;
248 orphan->cmt = 1;
249 *last = orphan;
250 last = &orphan->cnext;
251 }
252 *last = NULL;
253 c->cmt_orphans = c->new_orphans;
254 c->new_orphans = 0;
255 dbg_cmt("%d orphans to commit", c->cmt_orphans);
256 INIT_LIST_HEAD(&c->orph_new);
257 if (c->tot_orphans == 0)
258 c->no_orphs = 1;
259 else
260 c->no_orphs = 0;
261 spin_unlock(&c->orphan_lock);
262 return 0;
263 }
264
265 /**
266 * avail_orphs - calculate available space.
267 * @c: UBIFS file-system description object
268 *
269 * This function returns the number of orphans that can be written in the
270 * available space.
271 */
avail_orphs(struct ubifs_info * c)272 static int avail_orphs(struct ubifs_info *c)
273 {
274 int avail_lebs, avail, gap;
275
276 avail_lebs = c->orph_lebs - (c->ohead_lnum - c->orph_first) - 1;
277 avail = avail_lebs *
278 ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
279 gap = c->leb_size - c->ohead_offs;
280 if (gap >= UBIFS_ORPH_NODE_SZ + sizeof(__le64))
281 avail += (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64);
282 return avail;
283 }
284
285 /**
286 * tot_avail_orphs - calculate total space.
287 * @c: UBIFS file-system description object
288 *
289 * This function returns the number of orphans that can be written in half
290 * the total space. That leaves half the space for adding new orphans.
291 */
tot_avail_orphs(struct ubifs_info * c)292 static int tot_avail_orphs(struct ubifs_info *c)
293 {
294 int avail_lebs, avail;
295
296 avail_lebs = c->orph_lebs;
297 avail = avail_lebs *
298 ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
299 return avail / 2;
300 }
301
302 /**
303 * do_write_orph_node - write a node to the orphan head.
304 * @c: UBIFS file-system description object
305 * @len: length of node
306 * @atomic: write atomically
307 *
308 * This function writes a node to the orphan head from the orphan buffer. If
309 * %atomic is not zero, then the write is done atomically. On success, %0 is
310 * returned, otherwise a negative error code is returned.
311 */
do_write_orph_node(struct ubifs_info * c,int len,int atomic)312 static int do_write_orph_node(struct ubifs_info *c, int len, int atomic)
313 {
314 int err = 0;
315
316 if (atomic) {
317 ubifs_assert(c, c->ohead_offs == 0);
318 ubifs_prepare_node(c, c->orph_buf, len, 1);
319 len = ALIGN(len, c->min_io_size);
320 err = ubifs_leb_change(c, c->ohead_lnum, c->orph_buf, len);
321 } else {
322 if (c->ohead_offs == 0) {
323 /* Ensure LEB has been unmapped */
324 err = ubifs_leb_unmap(c, c->ohead_lnum);
325 if (err)
326 return err;
327 }
328 err = ubifs_write_node(c, c->orph_buf, len, c->ohead_lnum,
329 c->ohead_offs);
330 }
331 return err;
332 }
333
334 /**
335 * write_orph_node - write an orphan node.
336 * @c: UBIFS file-system description object
337 * @atomic: write atomically
338 *
339 * This function builds an orphan node from the cnext list and writes it to the
340 * orphan head. On success, %0 is returned, otherwise a negative error code
341 * is returned.
342 */
write_orph_node(struct ubifs_info * c,int atomic)343 static int write_orph_node(struct ubifs_info *c, int atomic)
344 {
345 struct ubifs_orphan *orphan, *cnext;
346 struct ubifs_orph_node *orph;
347 int gap, err, len, cnt, i;
348
349 ubifs_assert(c, c->cmt_orphans > 0);
350 gap = c->leb_size - c->ohead_offs;
351 if (gap < UBIFS_ORPH_NODE_SZ + sizeof(__le64)) {
352 c->ohead_lnum += 1;
353 c->ohead_offs = 0;
354 gap = c->leb_size;
355 if (c->ohead_lnum > c->orph_last) {
356 /*
357 * We limit the number of orphans so that this should
358 * never happen.
359 */
360 ubifs_err(c, "out of space in orphan area");
361 return -EINVAL;
362 }
363 }
364 cnt = (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64);
365 if (cnt > c->cmt_orphans)
366 cnt = c->cmt_orphans;
367 len = UBIFS_ORPH_NODE_SZ + cnt * sizeof(__le64);
368 ubifs_assert(c, c->orph_buf);
369 orph = c->orph_buf;
370 orph->ch.node_type = UBIFS_ORPH_NODE;
371 spin_lock(&c->orphan_lock);
372 cnext = c->orph_cnext;
373 for (i = 0; i < cnt; i++) {
374 orphan = cnext;
375 ubifs_assert(c, orphan->cmt);
376 orph->inos[i] = cpu_to_le64(orphan->inum);
377 orphan->cmt = 0;
378 cnext = orphan->cnext;
379 orphan->cnext = NULL;
380 }
381 c->orph_cnext = cnext;
382 c->cmt_orphans -= cnt;
383 spin_unlock(&c->orphan_lock);
384 if (c->cmt_orphans)
385 orph->cmt_no = cpu_to_le64(c->cmt_no);
386 else
387 /* Mark the last node of the commit */
388 orph->cmt_no = cpu_to_le64((c->cmt_no) | (1ULL << 63));
389 ubifs_assert(c, c->ohead_offs + len <= c->leb_size);
390 ubifs_assert(c, c->ohead_lnum >= c->orph_first);
391 ubifs_assert(c, c->ohead_lnum <= c->orph_last);
392 err = do_write_orph_node(c, len, atomic);
393 c->ohead_offs += ALIGN(len, c->min_io_size);
394 c->ohead_offs = ALIGN(c->ohead_offs, 8);
395 return err;
396 }
397
398 /**
399 * write_orph_nodes - write orphan nodes until there are no more to commit.
400 * @c: UBIFS file-system description object
401 * @atomic: write atomically
402 *
403 * This function writes orphan nodes for all the orphans to commit. On success,
404 * %0 is returned, otherwise a negative error code is returned.
405 */
write_orph_nodes(struct ubifs_info * c,int atomic)406 static int write_orph_nodes(struct ubifs_info *c, int atomic)
407 {
408 int err;
409
410 while (c->cmt_orphans > 0) {
411 err = write_orph_node(c, atomic);
412 if (err)
413 return err;
414 }
415 if (atomic) {
416 int lnum;
417
418 /* Unmap any unused LEBs after consolidation */
419 for (lnum = c->ohead_lnum + 1; lnum <= c->orph_last; lnum++) {
420 err = ubifs_leb_unmap(c, lnum);
421 if (err)
422 return err;
423 }
424 }
425 return 0;
426 }
427
428 /**
429 * consolidate - consolidate the orphan area.
430 * @c: UBIFS file-system description object
431 *
432 * This function enables consolidation by putting all the orphans into the list
433 * to commit. The list is in the order that the orphans were added, and the
434 * LEBs are written atomically in order, so at no time can orphans be lost by
435 * an unclean unmount.
436 *
437 * This function returns %0 on success and a negative error code on failure.
438 */
consolidate(struct ubifs_info * c)439 static int consolidate(struct ubifs_info *c)
440 {
441 int tot_avail = tot_avail_orphs(c), err = 0;
442
443 spin_lock(&c->orphan_lock);
444 dbg_cmt("there is space for %d orphans and there are %d",
445 tot_avail, c->tot_orphans);
446 if (c->tot_orphans - c->new_orphans <= tot_avail) {
447 struct ubifs_orphan *orphan, **last;
448 int cnt = 0;
449
450 /* Change the cnext list to include all non-new orphans */
451 last = &c->orph_cnext;
452 list_for_each_entry(orphan, &c->orph_list, list) {
453 if (orphan->new)
454 continue;
455 orphan->cmt = 1;
456 *last = orphan;
457 last = &orphan->cnext;
458 cnt += 1;
459 }
460 *last = NULL;
461 ubifs_assert(c, cnt == c->tot_orphans - c->new_orphans);
462 c->cmt_orphans = cnt;
463 c->ohead_lnum = c->orph_first;
464 c->ohead_offs = 0;
465 } else {
466 /*
467 * We limit the number of orphans so that this should
468 * never happen.
469 */
470 ubifs_err(c, "out of space in orphan area");
471 err = -EINVAL;
472 }
473 spin_unlock(&c->orphan_lock);
474 return err;
475 }
476
477 /**
478 * commit_orphans - commit orphans.
479 * @c: UBIFS file-system description object
480 *
481 * This function commits orphans to flash. On success, %0 is returned,
482 * otherwise a negative error code is returned.
483 */
commit_orphans(struct ubifs_info * c)484 static int commit_orphans(struct ubifs_info *c)
485 {
486 int avail, atomic = 0, err;
487
488 ubifs_assert(c, c->cmt_orphans > 0);
489 avail = avail_orphs(c);
490 if (avail < c->cmt_orphans) {
491 /* Not enough space to write new orphans, so consolidate */
492 err = consolidate(c);
493 if (err)
494 return err;
495 atomic = 1;
496 }
497 err = write_orph_nodes(c, atomic);
498 return err;
499 }
500
501 /**
502 * erase_deleted - erase the orphans marked for deletion.
503 * @c: UBIFS file-system description object
504 *
505 * During commit, the orphans being committed cannot be deleted, so they are
506 * marked for deletion and deleted by this function. Also, the recovery
507 * adds killed orphans to the deletion list, and therefore they are deleted
508 * here too.
509 */
erase_deleted(struct ubifs_info * c)510 static void erase_deleted(struct ubifs_info *c)
511 {
512 struct ubifs_orphan *orphan, *dnext;
513
514 spin_lock(&c->orphan_lock);
515 dnext = c->orph_dnext;
516 while (dnext) {
517 orphan = dnext;
518 dnext = orphan->dnext;
519 ubifs_assert(c, !orphan->new);
520 ubifs_assert(c, orphan->del);
521 rb_erase(&orphan->rb, &c->orph_tree);
522 list_del(&orphan->list);
523 c->tot_orphans -= 1;
524 dbg_gen("deleting orphan ino %lu", (unsigned long)orphan->inum);
525 kfree(orphan);
526 }
527 c->orph_dnext = NULL;
528 spin_unlock(&c->orphan_lock);
529 }
530
531 /**
532 * ubifs_orphan_end_commit - end commit of orphans.
533 * @c: UBIFS file-system description object
534 *
535 * End commit of orphans.
536 */
ubifs_orphan_end_commit(struct ubifs_info * c)537 int ubifs_orphan_end_commit(struct ubifs_info *c)
538 {
539 int err;
540
541 if (c->cmt_orphans != 0) {
542 err = commit_orphans(c);
543 if (err)
544 return err;
545 }
546 erase_deleted(c);
547 err = dbg_check_orphans(c);
548 return err;
549 }
550
551 /**
552 * ubifs_clear_orphans - erase all LEBs used for orphans.
553 * @c: UBIFS file-system description object
554 *
555 * If recovery is not required, then the orphans from the previous session
556 * are not needed. This function locates the LEBs used to record
557 * orphans, and un-maps them.
558 */
ubifs_clear_orphans(struct ubifs_info * c)559 int ubifs_clear_orphans(struct ubifs_info *c)
560 {
561 int lnum, err;
562
563 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
564 err = ubifs_leb_unmap(c, lnum);
565 if (err)
566 return err;
567 }
568 c->ohead_lnum = c->orph_first;
569 c->ohead_offs = 0;
570 return 0;
571 }
572
573 /**
574 * insert_dead_orphan - insert an orphan.
575 * @c: UBIFS file-system description object
576 * @inum: orphan inode number
577 *
578 * This function is a helper to the 'do_kill_orphans()' function. The orphan
579 * must be kept until the next commit, so it is added to the rb-tree and the
580 * deletion list.
581 */
insert_dead_orphan(struct ubifs_info * c,ino_t inum)582 static int insert_dead_orphan(struct ubifs_info *c, ino_t inum)
583 {
584 struct ubifs_orphan *orphan, *o;
585 struct rb_node **p, *parent = NULL;
586
587 orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_KERNEL);
588 if (!orphan)
589 return -ENOMEM;
590 orphan->inum = inum;
591
592 p = &c->orph_tree.rb_node;
593 while (*p) {
594 parent = *p;
595 o = rb_entry(parent, struct ubifs_orphan, rb);
596 if (inum < o->inum)
597 p = &(*p)->rb_left;
598 else if (inum > o->inum)
599 p = &(*p)->rb_right;
600 else {
601 /* Already added - no problem */
602 kfree(orphan);
603 return 0;
604 }
605 }
606 c->tot_orphans += 1;
607 rb_link_node(&orphan->rb, parent, p);
608 rb_insert_color(&orphan->rb, &c->orph_tree);
609 list_add_tail(&orphan->list, &c->orph_list);
610 orphan->del = 1;
611 orphan->dnext = c->orph_dnext;
612 c->orph_dnext = orphan;
613 dbg_mnt("ino %lu, new %d, tot %d", (unsigned long)inum,
614 c->new_orphans, c->tot_orphans);
615 return 0;
616 }
617
618 /**
619 * do_kill_orphans - remove orphan inodes from the index.
620 * @c: UBIFS file-system description object
621 * @sleb: scanned LEB
622 * @last_cmt_no: cmt_no of last orphan node read is passed and returned here
623 * @outofdate: whether the LEB is out of date is returned here
624 * @last_flagged: whether the end orphan node is encountered
625 *
626 * This function is a helper to the 'kill_orphans()' function. It goes through
627 * every orphan node in a LEB and for every inode number recorded, removes
628 * all keys for that inode from the TNC.
629 */
do_kill_orphans(struct ubifs_info * c,struct ubifs_scan_leb * sleb,unsigned long long * last_cmt_no,int * outofdate,int * last_flagged)630 static int do_kill_orphans(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
631 unsigned long long *last_cmt_no, int *outofdate,
632 int *last_flagged)
633 {
634 struct ubifs_scan_node *snod;
635 struct ubifs_orph_node *orph;
636 struct ubifs_ino_node *ino = NULL;
637 unsigned long long cmt_no;
638 ino_t inum;
639 int i, n, err, first = 1;
640
641 ino = kmalloc(UBIFS_MAX_INO_NODE_SZ, GFP_NOFS);
642 if (!ino)
643 return -ENOMEM;
644
645 list_for_each_entry(snod, &sleb->nodes, list) {
646 if (snod->type != UBIFS_ORPH_NODE) {
647 ubifs_err(c, "invalid node type %d in orphan area at %d:%d",
648 snod->type, sleb->lnum, snod->offs);
649 ubifs_dump_node(c, snod->node);
650 err = -EINVAL;
651 goto out_free;
652 }
653
654 orph = snod->node;
655
656 /* Check commit number */
657 cmt_no = le64_to_cpu(orph->cmt_no) & LLONG_MAX;
658 /*
659 * The commit number on the master node may be less, because
660 * of a failed commit. If there are several failed commits in a
661 * row, the commit number written on orphan nodes will continue
662 * to increase (because the commit number is adjusted here) even
663 * though the commit number on the master node stays the same
664 * because the master node has not been re-written.
665 */
666 if (cmt_no > c->cmt_no)
667 c->cmt_no = cmt_no;
668 if (cmt_no < *last_cmt_no && *last_flagged) {
669 /*
670 * The last orphan node had a higher commit number and
671 * was flagged as the last written for that commit
672 * number. That makes this orphan node, out of date.
673 */
674 if (!first) {
675 ubifs_err(c, "out of order commit number %llu in orphan node at %d:%d",
676 cmt_no, sleb->lnum, snod->offs);
677 ubifs_dump_node(c, snod->node);
678 err = -EINVAL;
679 goto out_free;
680 }
681 dbg_rcvry("out of date LEB %d", sleb->lnum);
682 *outofdate = 1;
683 err = 0;
684 goto out_free;
685 }
686
687 if (first)
688 first = 0;
689
690 n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
691 for (i = 0; i < n; i++) {
692 union ubifs_key key1, key2;
693
694 inum = le64_to_cpu(orph->inos[i]);
695
696 ino_key_init(c, &key1, inum);
697 err = ubifs_tnc_lookup(c, &key1, ino);
698 if (err && err != -ENOENT)
699 goto out_free;
700
701 /*
702 * Check whether an inode can really get deleted.
703 * linkat() with O_TMPFILE allows rebirth of an inode.
704 */
705 if (err == 0 && ino->nlink == 0) {
706 dbg_rcvry("deleting orphaned inode %lu",
707 (unsigned long)inum);
708
709 lowest_ino_key(c, &key1, inum);
710 highest_ino_key(c, &key2, inum);
711
712 err = ubifs_tnc_remove_range(c, &key1, &key2);
713 if (err)
714 goto out_ro;
715 }
716
717 err = insert_dead_orphan(c, inum);
718 if (err)
719 goto out_free;
720 }
721
722 *last_cmt_no = cmt_no;
723 if (le64_to_cpu(orph->cmt_no) & (1ULL << 63)) {
724 dbg_rcvry("last orph node for commit %llu at %d:%d",
725 cmt_no, sleb->lnum, snod->offs);
726 *last_flagged = 1;
727 } else
728 *last_flagged = 0;
729 }
730
731 err = 0;
732 out_free:
733 kfree(ino);
734 return err;
735
736 out_ro:
737 ubifs_ro_mode(c, err);
738 kfree(ino);
739 return err;
740 }
741
742 /**
743 * kill_orphans - remove all orphan inodes from the index.
744 * @c: UBIFS file-system description object
745 *
746 * If recovery is required, then orphan inodes recorded during the previous
747 * session (which ended with an unclean unmount) must be deleted from the index.
748 * This is done by updating the TNC, but since the index is not updated until
749 * the next commit, the LEBs where the orphan information is recorded are not
750 * erased until the next commit.
751 */
kill_orphans(struct ubifs_info * c)752 static int kill_orphans(struct ubifs_info *c)
753 {
754 unsigned long long last_cmt_no = 0;
755 int lnum, err = 0, outofdate = 0, last_flagged = 0;
756
757 c->ohead_lnum = c->orph_first;
758 c->ohead_offs = 0;
759 /* Check no-orphans flag and skip this if no orphans */
760 if (c->no_orphs) {
761 dbg_rcvry("no orphans");
762 return 0;
763 }
764 /*
765 * Orph nodes always start at c->orph_first and are written to each
766 * successive LEB in turn. Generally unused LEBs will have been unmapped
767 * but may contain out of date orphan nodes if the unmap didn't go
768 * through. In addition, the last orphan node written for each commit is
769 * marked (top bit of orph->cmt_no is set to 1). It is possible that
770 * there are orphan nodes from the next commit (i.e. the commit did not
771 * complete successfully). In that case, no orphans will have been lost
772 * due to the way that orphans are written, and any orphans added will
773 * be valid orphans anyway and so can be deleted.
774 */
775 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
776 struct ubifs_scan_leb *sleb;
777
778 dbg_rcvry("LEB %d", lnum);
779 sleb = ubifs_scan(c, lnum, 0, c->sbuf, 1);
780 if (IS_ERR(sleb)) {
781 if (PTR_ERR(sleb) == -EUCLEAN)
782 sleb = ubifs_recover_leb(c, lnum, 0,
783 c->sbuf, -1);
784 if (IS_ERR(sleb)) {
785 err = PTR_ERR(sleb);
786 break;
787 }
788 }
789 err = do_kill_orphans(c, sleb, &last_cmt_no, &outofdate,
790 &last_flagged);
791 if (err || outofdate) {
792 ubifs_scan_destroy(sleb);
793 break;
794 }
795 if (sleb->endpt) {
796 c->ohead_lnum = lnum;
797 c->ohead_offs = sleb->endpt;
798 }
799 ubifs_scan_destroy(sleb);
800 }
801 return err;
802 }
803
804 /**
805 * ubifs_mount_orphans - delete orphan inodes and erase LEBs that recorded them.
806 * @c: UBIFS file-system description object
807 * @unclean: indicates recovery from unclean unmount
808 * @read_only: indicates read only mount
809 *
810 * This function is called when mounting to erase orphans from the previous
811 * session. If UBIFS was not unmounted cleanly, then the inodes recorded as
812 * orphans are deleted.
813 */
ubifs_mount_orphans(struct ubifs_info * c,int unclean,int read_only)814 int ubifs_mount_orphans(struct ubifs_info *c, int unclean, int read_only)
815 {
816 int err = 0;
817
818 c->max_orphans = tot_avail_orphs(c);
819
820 if (!read_only) {
821 c->orph_buf = vmalloc(c->leb_size);
822 if (!c->orph_buf)
823 return -ENOMEM;
824 }
825
826 if (unclean)
827 err = kill_orphans(c);
828 else if (!read_only)
829 err = ubifs_clear_orphans(c);
830
831 return err;
832 }
833
834 /*
835 * Everything below is related to debugging.
836 */
837
838 struct check_orphan {
839 struct rb_node rb;
840 ino_t inum;
841 };
842
843 struct check_info {
844 unsigned long last_ino;
845 unsigned long tot_inos;
846 unsigned long missing;
847 unsigned long long leaf_cnt;
848 struct ubifs_ino_node *node;
849 struct rb_root root;
850 };
851
dbg_find_orphan(struct ubifs_info * c,ino_t inum)852 static bool dbg_find_orphan(struct ubifs_info *c, ino_t inum)
853 {
854 bool found = false;
855
856 spin_lock(&c->orphan_lock);
857 found = !!lookup_orphan(c, inum);
858 spin_unlock(&c->orphan_lock);
859
860 return found;
861 }
862
dbg_ins_check_orphan(struct rb_root * root,ino_t inum)863 static int dbg_ins_check_orphan(struct rb_root *root, ino_t inum)
864 {
865 struct check_orphan *orphan, *o;
866 struct rb_node **p, *parent = NULL;
867
868 orphan = kzalloc(sizeof(struct check_orphan), GFP_NOFS);
869 if (!orphan)
870 return -ENOMEM;
871 orphan->inum = inum;
872
873 p = &root->rb_node;
874 while (*p) {
875 parent = *p;
876 o = rb_entry(parent, struct check_orphan, rb);
877 if (inum < o->inum)
878 p = &(*p)->rb_left;
879 else if (inum > o->inum)
880 p = &(*p)->rb_right;
881 else {
882 kfree(orphan);
883 return 0;
884 }
885 }
886 rb_link_node(&orphan->rb, parent, p);
887 rb_insert_color(&orphan->rb, root);
888 return 0;
889 }
890
dbg_find_check_orphan(struct rb_root * root,ino_t inum)891 static int dbg_find_check_orphan(struct rb_root *root, ino_t inum)
892 {
893 struct check_orphan *o;
894 struct rb_node *p;
895
896 p = root->rb_node;
897 while (p) {
898 o = rb_entry(p, struct check_orphan, rb);
899 if (inum < o->inum)
900 p = p->rb_left;
901 else if (inum > o->inum)
902 p = p->rb_right;
903 else
904 return 1;
905 }
906 return 0;
907 }
908
dbg_free_check_tree(struct rb_root * root)909 static void dbg_free_check_tree(struct rb_root *root)
910 {
911 struct check_orphan *o, *n;
912
913 rbtree_postorder_for_each_entry_safe(o, n, root, rb)
914 kfree(o);
915 }
916
dbg_orphan_check(struct ubifs_info * c,struct ubifs_zbranch * zbr,void * priv)917 static int dbg_orphan_check(struct ubifs_info *c, struct ubifs_zbranch *zbr,
918 void *priv)
919 {
920 struct check_info *ci = priv;
921 ino_t inum;
922 int err;
923
924 inum = key_inum(c, &zbr->key);
925 if (inum != ci->last_ino) {
926 /* Lowest node type is the inode node, so it comes first */
927 if (key_type(c, &zbr->key) != UBIFS_INO_KEY)
928 ubifs_err(c, "found orphan node ino %lu, type %d",
929 (unsigned long)inum, key_type(c, &zbr->key));
930 ci->last_ino = inum;
931 ci->tot_inos += 1;
932 err = ubifs_tnc_read_node(c, zbr, ci->node);
933 if (err) {
934 ubifs_err(c, "node read failed, error %d", err);
935 return err;
936 }
937 if (ci->node->nlink == 0)
938 /* Must be recorded as an orphan */
939 if (!dbg_find_check_orphan(&ci->root, inum) &&
940 !dbg_find_orphan(c, inum)) {
941 ubifs_err(c, "missing orphan, ino %lu",
942 (unsigned long)inum);
943 ci->missing += 1;
944 }
945 }
946 ci->leaf_cnt += 1;
947 return 0;
948 }
949
dbg_read_orphans(struct check_info * ci,struct ubifs_scan_leb * sleb)950 static int dbg_read_orphans(struct check_info *ci, struct ubifs_scan_leb *sleb)
951 {
952 struct ubifs_scan_node *snod;
953 struct ubifs_orph_node *orph;
954 ino_t inum;
955 int i, n, err;
956
957 list_for_each_entry(snod, &sleb->nodes, list) {
958 cond_resched();
959 if (snod->type != UBIFS_ORPH_NODE)
960 continue;
961 orph = snod->node;
962 n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
963 for (i = 0; i < n; i++) {
964 inum = le64_to_cpu(orph->inos[i]);
965 err = dbg_ins_check_orphan(&ci->root, inum);
966 if (err)
967 return err;
968 }
969 }
970 return 0;
971 }
972
dbg_scan_orphans(struct ubifs_info * c,struct check_info * ci)973 static int dbg_scan_orphans(struct ubifs_info *c, struct check_info *ci)
974 {
975 int lnum, err = 0;
976 void *buf;
977
978 /* Check no-orphans flag and skip this if no orphans */
979 if (c->no_orphs)
980 return 0;
981
982 buf = __vmalloc(c->leb_size, GFP_NOFS);
983 if (!buf) {
984 ubifs_err(c, "cannot allocate memory to check orphans");
985 return 0;
986 }
987
988 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
989 struct ubifs_scan_leb *sleb;
990
991 sleb = ubifs_scan(c, lnum, 0, buf, 0);
992 if (IS_ERR(sleb)) {
993 err = PTR_ERR(sleb);
994 break;
995 }
996
997 err = dbg_read_orphans(ci, sleb);
998 ubifs_scan_destroy(sleb);
999 if (err)
1000 break;
1001 }
1002
1003 vfree(buf);
1004 return err;
1005 }
1006
dbg_check_orphans(struct ubifs_info * c)1007 static int dbg_check_orphans(struct ubifs_info *c)
1008 {
1009 struct check_info ci;
1010 int err;
1011
1012 if (!dbg_is_chk_orph(c))
1013 return 0;
1014
1015 ci.last_ino = 0;
1016 ci.tot_inos = 0;
1017 ci.missing = 0;
1018 ci.leaf_cnt = 0;
1019 ci.root = RB_ROOT;
1020 ci.node = kmalloc(UBIFS_MAX_INO_NODE_SZ, GFP_NOFS);
1021 if (!ci.node) {
1022 ubifs_err(c, "out of memory");
1023 return -ENOMEM;
1024 }
1025
1026 err = dbg_scan_orphans(c, &ci);
1027 if (err)
1028 goto out;
1029
1030 err = dbg_walk_index(c, &dbg_orphan_check, NULL, &ci);
1031 if (err) {
1032 ubifs_err(c, "cannot scan TNC, error %d", err);
1033 goto out;
1034 }
1035
1036 if (ci.missing) {
1037 ubifs_err(c, "%lu missing orphan(s)", ci.missing);
1038 err = -EINVAL;
1039 goto out;
1040 }
1041
1042 dbg_cmt("last inode number is %lu", ci.last_ino);
1043 dbg_cmt("total number of inodes is %lu", ci.tot_inos);
1044 dbg_cmt("total number of leaf nodes is %llu", ci.leaf_cnt);
1045
1046 out:
1047 dbg_free_check_tree(&ci.root);
1048 kfree(ci.node);
1049 return err;
1050 }
1051