Loading...
1/*
2 * This file is part of UBIFS.
3 *
4 * Copyright (C) 2006-2008 Nokia Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 *
19 * Authors: Adrian Hunter
20 * Artem Bityutskiy (Битюцкий Артём)
21 */
22
23/*
24 * This file implements functions that manage the running of the commit process.
25 * Each affected module has its own functions to accomplish their part in the
26 * commit and those functions are called here.
27 *
28 * The commit is the process whereby all updates to the index and LEB properties
29 * are written out together and the journal becomes empty. This keeps the
30 * file system consistent - at all times the state can be recreated by reading
31 * the index and LEB properties and then replaying the journal.
32 *
33 * The commit is split into two parts named "commit start" and "commit end".
34 * During commit start, the commit process has exclusive access to the journal
35 * by holding the commit semaphore down for writing. As few I/O operations as
36 * possible are performed during commit start, instead the nodes that are to be
37 * written are merely identified. During commit end, the commit semaphore is no
38 * longer held and the journal is again in operation, allowing users to continue
39 * to use the file system while the bulk of the commit I/O is performed. The
40 * purpose of this two-step approach is to prevent the commit from causing any
41 * latency blips. Note that in any case, the commit does not prevent lookups
42 * (as permitted by the TNC mutex), or access to VFS data structures e.g. page
43 * cache.
44 */
45
46#include <linux/freezer.h>
47#include <linux/kthread.h>
48#include <linux/slab.h>
49#include "ubifs.h"
50
51/*
52 * nothing_to_commit - check if there is nothing to commit.
53 * @c: UBIFS file-system description object
54 *
55 * This is a helper function which checks if there is anything to commit. It is
56 * used as an optimization to avoid starting the commit if it is not really
57 * necessary. Indeed, the commit operation always assumes flash I/O (e.g.,
58 * writing the commit start node to the log), and it is better to avoid doing
59 * this unnecessarily. E.g., 'ubifs_sync_fs()' runs the commit, but if there is
60 * nothing to commit, it is more optimal to avoid any flash I/O.
61 *
62 * This function has to be called with @c->commit_sem locked for writing -
63 * this function does not take LPT/TNC locks because the @c->commit_sem
64 * guarantees that we have exclusive access to the TNC and LPT data structures.
65 *
66 * This function returns %1 if there is nothing to commit and %0 otherwise.
67 */
68static int nothing_to_commit(struct ubifs_info *c)
69{
70 /*
71 * During mounting or remounting from R/O mode to R/W mode we may
72 * commit for various recovery-related reasons.
73 */
74 if (c->mounting || c->remounting_rw)
75 return 0;
76
77 /*
78 * If the root TNC node is dirty, we definitely have something to
79 * commit.
80 */
81 if (c->zroot.znode && ubifs_zn_dirty(c->zroot.znode))
82 return 0;
83
84 /*
85 * Even though the TNC is clean, the LPT tree may have dirty nodes. For
86 * example, this may happen if the budgeting subsystem invoked GC to
87 * make some free space, and the GC found an LEB with only dirty and
88 * free space. In this case GC would just change the lprops of this
89 * LEB (by turning all space into free space) and unmap it.
90 */
91 if (c->nroot && test_bit(DIRTY_CNODE, &c->nroot->flags))
92 return 0;
93
94 ubifs_assert(atomic_long_read(&c->dirty_zn_cnt) == 0);
95 ubifs_assert(c->dirty_pn_cnt == 0);
96 ubifs_assert(c->dirty_nn_cnt == 0);
97
98 return 1;
99}
100
101/**
102 * do_commit - commit the journal.
103 * @c: UBIFS file-system description object
104 *
105 * This function implements UBIFS commit. It has to be called with commit lock
106 * locked. Returns zero in case of success and a negative error code in case of
107 * failure.
108 */
109static int do_commit(struct ubifs_info *c)
110{
111 int err, new_ltail_lnum, old_ltail_lnum, i;
112 struct ubifs_zbranch zroot;
113 struct ubifs_lp_stats lst;
114
115 dbg_cmt("start");
116 ubifs_assert(!c->ro_media && !c->ro_mount);
117
118 if (c->ro_error) {
119 err = -EROFS;
120 goto out_up;
121 }
122
123 if (nothing_to_commit(c)) {
124 up_write(&c->commit_sem);
125 err = 0;
126 goto out_cancel;
127 }
128
129 /* Sync all write buffers (necessary for recovery) */
130 for (i = 0; i < c->jhead_cnt; i++) {
131 err = ubifs_wbuf_sync(&c->jheads[i].wbuf);
132 if (err)
133 goto out_up;
134 }
135
136 c->cmt_no += 1;
137 err = ubifs_gc_start_commit(c);
138 if (err)
139 goto out_up;
140 err = dbg_check_lprops(c);
141 if (err)
142 goto out_up;
143 err = ubifs_log_start_commit(c, &new_ltail_lnum);
144 if (err)
145 goto out_up;
146 err = ubifs_tnc_start_commit(c, &zroot);
147 if (err)
148 goto out_up;
149 err = ubifs_lpt_start_commit(c);
150 if (err)
151 goto out_up;
152 err = ubifs_orphan_start_commit(c);
153 if (err)
154 goto out_up;
155
156 ubifs_get_lp_stats(c, &lst);
157
158 up_write(&c->commit_sem);
159
160 err = ubifs_tnc_end_commit(c);
161 if (err)
162 goto out;
163 err = ubifs_lpt_end_commit(c);
164 if (err)
165 goto out;
166 err = ubifs_orphan_end_commit(c);
167 if (err)
168 goto out;
169 err = dbg_check_old_index(c, &zroot);
170 if (err)
171 goto out;
172
173 c->mst_node->cmt_no = cpu_to_le64(c->cmt_no);
174 c->mst_node->log_lnum = cpu_to_le32(new_ltail_lnum);
175 c->mst_node->root_lnum = cpu_to_le32(zroot.lnum);
176 c->mst_node->root_offs = cpu_to_le32(zroot.offs);
177 c->mst_node->root_len = cpu_to_le32(zroot.len);
178 c->mst_node->ihead_lnum = cpu_to_le32(c->ihead_lnum);
179 c->mst_node->ihead_offs = cpu_to_le32(c->ihead_offs);
180 c->mst_node->index_size = cpu_to_le64(c->bi.old_idx_sz);
181 c->mst_node->lpt_lnum = cpu_to_le32(c->lpt_lnum);
182 c->mst_node->lpt_offs = cpu_to_le32(c->lpt_offs);
183 c->mst_node->nhead_lnum = cpu_to_le32(c->nhead_lnum);
184 c->mst_node->nhead_offs = cpu_to_le32(c->nhead_offs);
185 c->mst_node->ltab_lnum = cpu_to_le32(c->ltab_lnum);
186 c->mst_node->ltab_offs = cpu_to_le32(c->ltab_offs);
187 c->mst_node->lsave_lnum = cpu_to_le32(c->lsave_lnum);
188 c->mst_node->lsave_offs = cpu_to_le32(c->lsave_offs);
189 c->mst_node->lscan_lnum = cpu_to_le32(c->lscan_lnum);
190 c->mst_node->empty_lebs = cpu_to_le32(lst.empty_lebs);
191 c->mst_node->idx_lebs = cpu_to_le32(lst.idx_lebs);
192 c->mst_node->total_free = cpu_to_le64(lst.total_free);
193 c->mst_node->total_dirty = cpu_to_le64(lst.total_dirty);
194 c->mst_node->total_used = cpu_to_le64(lst.total_used);
195 c->mst_node->total_dead = cpu_to_le64(lst.total_dead);
196 c->mst_node->total_dark = cpu_to_le64(lst.total_dark);
197 if (c->no_orphs)
198 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
199 else
200 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_NO_ORPHS);
201
202 old_ltail_lnum = c->ltail_lnum;
203 err = ubifs_log_end_commit(c, new_ltail_lnum);
204 if (err)
205 goto out;
206
207 err = ubifs_log_post_commit(c, old_ltail_lnum);
208 if (err)
209 goto out;
210 err = ubifs_gc_end_commit(c);
211 if (err)
212 goto out;
213 err = ubifs_lpt_post_commit(c);
214 if (err)
215 goto out;
216
217out_cancel:
218 spin_lock(&c->cs_lock);
219 c->cmt_state = COMMIT_RESTING;
220 wake_up(&c->cmt_wq);
221 dbg_cmt("commit end");
222 spin_unlock(&c->cs_lock);
223 return 0;
224
225out_up:
226 up_write(&c->commit_sem);
227out:
228 ubifs_err(c, "commit failed, error %d", err);
229 spin_lock(&c->cs_lock);
230 c->cmt_state = COMMIT_BROKEN;
231 wake_up(&c->cmt_wq);
232 spin_unlock(&c->cs_lock);
233 ubifs_ro_mode(c, err);
234 return err;
235}
236
237/**
238 * run_bg_commit - run background commit if it is needed.
239 * @c: UBIFS file-system description object
240 *
241 * This function runs background commit if it is needed. Returns zero in case
242 * of success and a negative error code in case of failure.
243 */
244static int run_bg_commit(struct ubifs_info *c)
245{
246 spin_lock(&c->cs_lock);
247 /*
248 * Run background commit only if background commit was requested or if
249 * commit is required.
250 */
251 if (c->cmt_state != COMMIT_BACKGROUND &&
252 c->cmt_state != COMMIT_REQUIRED)
253 goto out;
254 spin_unlock(&c->cs_lock);
255
256 down_write(&c->commit_sem);
257 spin_lock(&c->cs_lock);
258 if (c->cmt_state == COMMIT_REQUIRED)
259 c->cmt_state = COMMIT_RUNNING_REQUIRED;
260 else if (c->cmt_state == COMMIT_BACKGROUND)
261 c->cmt_state = COMMIT_RUNNING_BACKGROUND;
262 else
263 goto out_cmt_unlock;
264 spin_unlock(&c->cs_lock);
265
266 return do_commit(c);
267
268out_cmt_unlock:
269 up_write(&c->commit_sem);
270out:
271 spin_unlock(&c->cs_lock);
272 return 0;
273}
274
275/**
276 * ubifs_bg_thread - UBIFS background thread function.
277 * @info: points to the file-system description object
278 *
279 * This function implements various file-system background activities:
280 * o when a write-buffer timer expires it synchronizes the appropriate
281 * write-buffer;
282 * o when the journal is about to be full, it starts in-advance commit.
283 *
284 * Note, other stuff like background garbage collection may be added here in
285 * future.
286 */
287int ubifs_bg_thread(void *info)
288{
289 int err;
290 struct ubifs_info *c = info;
291
292 ubifs_msg(c, "background thread \"%s\" started, PID %d",
293 c->bgt_name, current->pid);
294 set_freezable();
295
296 while (1) {
297 if (kthread_should_stop())
298 break;
299
300 if (try_to_freeze())
301 continue;
302
303 set_current_state(TASK_INTERRUPTIBLE);
304 /* Check if there is something to do */
305 if (!c->need_bgt) {
306 /*
307 * Nothing prevents us from going sleep now and
308 * be never woken up and block the task which
309 * could wait in 'kthread_stop()' forever.
310 */
311 if (kthread_should_stop())
312 break;
313 schedule();
314 continue;
315 } else
316 __set_current_state(TASK_RUNNING);
317
318 c->need_bgt = 0;
319 err = ubifs_bg_wbufs_sync(c);
320 if (err)
321 ubifs_ro_mode(c, err);
322
323 run_bg_commit(c);
324 cond_resched();
325 }
326
327 ubifs_msg(c, "background thread \"%s\" stops", c->bgt_name);
328 return 0;
329}
330
331/**
332 * ubifs_commit_required - set commit state to "required".
333 * @c: UBIFS file-system description object
334 *
335 * This function is called if a commit is required but cannot be done from the
336 * calling function, so it is just flagged instead.
337 */
338void ubifs_commit_required(struct ubifs_info *c)
339{
340 spin_lock(&c->cs_lock);
341 switch (c->cmt_state) {
342 case COMMIT_RESTING:
343 case COMMIT_BACKGROUND:
344 dbg_cmt("old: %s, new: %s", dbg_cstate(c->cmt_state),
345 dbg_cstate(COMMIT_REQUIRED));
346 c->cmt_state = COMMIT_REQUIRED;
347 break;
348 case COMMIT_RUNNING_BACKGROUND:
349 dbg_cmt("old: %s, new: %s", dbg_cstate(c->cmt_state),
350 dbg_cstate(COMMIT_RUNNING_REQUIRED));
351 c->cmt_state = COMMIT_RUNNING_REQUIRED;
352 break;
353 case COMMIT_REQUIRED:
354 case COMMIT_RUNNING_REQUIRED:
355 case COMMIT_BROKEN:
356 break;
357 }
358 spin_unlock(&c->cs_lock);
359}
360
361/**
362 * ubifs_request_bg_commit - notify the background thread to do a commit.
363 * @c: UBIFS file-system description object
364 *
365 * This function is called if the journal is full enough to make a commit
366 * worthwhile, so background thread is kicked to start it.
367 */
368void ubifs_request_bg_commit(struct ubifs_info *c)
369{
370 spin_lock(&c->cs_lock);
371 if (c->cmt_state == COMMIT_RESTING) {
372 dbg_cmt("old: %s, new: %s", dbg_cstate(c->cmt_state),
373 dbg_cstate(COMMIT_BACKGROUND));
374 c->cmt_state = COMMIT_BACKGROUND;
375 spin_unlock(&c->cs_lock);
376 ubifs_wake_up_bgt(c);
377 } else
378 spin_unlock(&c->cs_lock);
379}
380
381/**
382 * wait_for_commit - wait for commit.
383 * @c: UBIFS file-system description object
384 *
385 * This function sleeps until the commit operation is no longer running.
386 */
387static int wait_for_commit(struct ubifs_info *c)
388{
389 dbg_cmt("pid %d goes sleep", current->pid);
390
391 /*
392 * The following sleeps if the condition is false, and will be woken
393 * when the commit ends. It is possible, although very unlikely, that we
394 * will wake up and see the subsequent commit running, rather than the
395 * one we were waiting for, and go back to sleep. However, we will be
396 * woken again, so there is no danger of sleeping forever.
397 */
398 wait_event(c->cmt_wq, c->cmt_state != COMMIT_RUNNING_BACKGROUND &&
399 c->cmt_state != COMMIT_RUNNING_REQUIRED);
400 dbg_cmt("commit finished, pid %d woke up", current->pid);
401 return 0;
402}
403
404/**
405 * ubifs_run_commit - run or wait for commit.
406 * @c: UBIFS file-system description object
407 *
408 * This function runs commit and returns zero in case of success and a negative
409 * error code in case of failure.
410 */
411int ubifs_run_commit(struct ubifs_info *c)
412{
413 int err = 0;
414
415 spin_lock(&c->cs_lock);
416 if (c->cmt_state == COMMIT_BROKEN) {
417 err = -EROFS;
418 goto out;
419 }
420
421 if (c->cmt_state == COMMIT_RUNNING_BACKGROUND)
422 /*
423 * We set the commit state to 'running required' to indicate
424 * that we want it to complete as quickly as possible.
425 */
426 c->cmt_state = COMMIT_RUNNING_REQUIRED;
427
428 if (c->cmt_state == COMMIT_RUNNING_REQUIRED) {
429 spin_unlock(&c->cs_lock);
430 return wait_for_commit(c);
431 }
432 spin_unlock(&c->cs_lock);
433
434 /* Ok, the commit is indeed needed */
435
436 down_write(&c->commit_sem);
437 spin_lock(&c->cs_lock);
438 /*
439 * Since we unlocked 'c->cs_lock', the state may have changed, so
440 * re-check it.
441 */
442 if (c->cmt_state == COMMIT_BROKEN) {
443 err = -EROFS;
444 goto out_cmt_unlock;
445 }
446
447 if (c->cmt_state == COMMIT_RUNNING_BACKGROUND)
448 c->cmt_state = COMMIT_RUNNING_REQUIRED;
449
450 if (c->cmt_state == COMMIT_RUNNING_REQUIRED) {
451 up_write(&c->commit_sem);
452 spin_unlock(&c->cs_lock);
453 return wait_for_commit(c);
454 }
455 c->cmt_state = COMMIT_RUNNING_REQUIRED;
456 spin_unlock(&c->cs_lock);
457
458 err = do_commit(c);
459 return err;
460
461out_cmt_unlock:
462 up_write(&c->commit_sem);
463out:
464 spin_unlock(&c->cs_lock);
465 return err;
466}
467
468/**
469 * ubifs_gc_should_commit - determine if it is time for GC to run commit.
470 * @c: UBIFS file-system description object
471 *
472 * This function is called by garbage collection to determine if commit should
473 * be run. If commit state is @COMMIT_BACKGROUND, which means that the journal
474 * is full enough to start commit, this function returns true. It is not
475 * absolutely necessary to commit yet, but it feels like this should be better
476 * then to keep doing GC. This function returns %1 if GC has to initiate commit
477 * and %0 if not.
478 */
479int ubifs_gc_should_commit(struct ubifs_info *c)
480{
481 int ret = 0;
482
483 spin_lock(&c->cs_lock);
484 if (c->cmt_state == COMMIT_BACKGROUND) {
485 dbg_cmt("commit required now");
486 c->cmt_state = COMMIT_REQUIRED;
487 } else
488 dbg_cmt("commit not requested");
489 if (c->cmt_state == COMMIT_REQUIRED)
490 ret = 1;
491 spin_unlock(&c->cs_lock);
492 return ret;
493}
494
495/*
496 * Everything below is related to debugging.
497 */
498
499/**
500 * struct idx_node - hold index nodes during index tree traversal.
501 * @list: list
502 * @iip: index in parent (slot number of this indexing node in the parent
503 * indexing node)
504 * @upper_key: all keys in this indexing node have to be less or equivalent to
505 * this key
506 * @idx: index node (8-byte aligned because all node structures must be 8-byte
507 * aligned)
508 */
509struct idx_node {
510 struct list_head list;
511 int iip;
512 union ubifs_key upper_key;
513 struct ubifs_idx_node idx __aligned(8);
514};
515
516/**
517 * dbg_old_index_check_init - get information for the next old index check.
518 * @c: UBIFS file-system description object
519 * @zroot: root of the index
520 *
521 * This function records information about the index that will be needed for the
522 * next old index check i.e. 'dbg_check_old_index()'.
523 *
524 * This function returns %0 on success and a negative error code on failure.
525 */
526int dbg_old_index_check_init(struct ubifs_info *c, struct ubifs_zbranch *zroot)
527{
528 struct ubifs_idx_node *idx;
529 int lnum, offs, len, err = 0;
530 struct ubifs_debug_info *d = c->dbg;
531
532 d->old_zroot = *zroot;
533 lnum = d->old_zroot.lnum;
534 offs = d->old_zroot.offs;
535 len = d->old_zroot.len;
536
537 idx = kmalloc(c->max_idx_node_sz, GFP_NOFS);
538 if (!idx)
539 return -ENOMEM;
540
541 err = ubifs_read_node(c, idx, UBIFS_IDX_NODE, len, lnum, offs);
542 if (err)
543 goto out;
544
545 d->old_zroot_level = le16_to_cpu(idx->level);
546 d->old_zroot_sqnum = le64_to_cpu(idx->ch.sqnum);
547out:
548 kfree(idx);
549 return err;
550}
551
552/**
553 * dbg_check_old_index - check the old copy of the index.
554 * @c: UBIFS file-system description object
555 * @zroot: root of the new index
556 *
557 * In order to be able to recover from an unclean unmount, a complete copy of
558 * the index must exist on flash. This is the "old" index. The commit process
559 * must write the "new" index to flash without overwriting or destroying any
560 * part of the old index. This function is run at commit end in order to check
561 * that the old index does indeed exist completely intact.
562 *
563 * This function returns %0 on success and a negative error code on failure.
564 */
565int dbg_check_old_index(struct ubifs_info *c, struct ubifs_zbranch *zroot)
566{
567 int lnum, offs, len, err = 0, uninitialized_var(last_level), child_cnt;
568 int first = 1, iip;
569 struct ubifs_debug_info *d = c->dbg;
570 union ubifs_key uninitialized_var(lower_key), upper_key, l_key, u_key;
571 unsigned long long uninitialized_var(last_sqnum);
572 struct ubifs_idx_node *idx;
573 struct list_head list;
574 struct idx_node *i;
575 size_t sz;
576
577 if (!dbg_is_chk_index(c))
578 return 0;
579
580 INIT_LIST_HEAD(&list);
581
582 sz = sizeof(struct idx_node) + ubifs_idx_node_sz(c, c->fanout) -
583 UBIFS_IDX_NODE_SZ;
584
585 /* Start at the old zroot */
586 lnum = d->old_zroot.lnum;
587 offs = d->old_zroot.offs;
588 len = d->old_zroot.len;
589 iip = 0;
590
591 /*
592 * Traverse the index tree preorder depth-first i.e. do a node and then
593 * its subtrees from left to right.
594 */
595 while (1) {
596 struct ubifs_branch *br;
597
598 /* Get the next index node */
599 i = kmalloc(sz, GFP_NOFS);
600 if (!i) {
601 err = -ENOMEM;
602 goto out_free;
603 }
604 i->iip = iip;
605 /* Keep the index nodes on our path in a linked list */
606 list_add_tail(&i->list, &list);
607 /* Read the index node */
608 idx = &i->idx;
609 err = ubifs_read_node(c, idx, UBIFS_IDX_NODE, len, lnum, offs);
610 if (err)
611 goto out_free;
612 /* Validate index node */
613 child_cnt = le16_to_cpu(idx->child_cnt);
614 if (child_cnt < 1 || child_cnt > c->fanout) {
615 err = 1;
616 goto out_dump;
617 }
618 if (first) {
619 first = 0;
620 /* Check root level and sqnum */
621 if (le16_to_cpu(idx->level) != d->old_zroot_level) {
622 err = 2;
623 goto out_dump;
624 }
625 if (le64_to_cpu(idx->ch.sqnum) != d->old_zroot_sqnum) {
626 err = 3;
627 goto out_dump;
628 }
629 /* Set last values as though root had a parent */
630 last_level = le16_to_cpu(idx->level) + 1;
631 last_sqnum = le64_to_cpu(idx->ch.sqnum) + 1;
632 key_read(c, ubifs_idx_key(c, idx), &lower_key);
633 highest_ino_key(c, &upper_key, INUM_WATERMARK);
634 }
635 key_copy(c, &upper_key, &i->upper_key);
636 if (le16_to_cpu(idx->level) != last_level - 1) {
637 err = 3;
638 goto out_dump;
639 }
640 /*
641 * The index is always written bottom up hence a child's sqnum
642 * is always less than the parents.
643 */
644 if (le64_to_cpu(idx->ch.sqnum) >= last_sqnum) {
645 err = 4;
646 goto out_dump;
647 }
648 /* Check key range */
649 key_read(c, ubifs_idx_key(c, idx), &l_key);
650 br = ubifs_idx_branch(c, idx, child_cnt - 1);
651 key_read(c, &br->key, &u_key);
652 if (keys_cmp(c, &lower_key, &l_key) > 0) {
653 err = 5;
654 goto out_dump;
655 }
656 if (keys_cmp(c, &upper_key, &u_key) < 0) {
657 err = 6;
658 goto out_dump;
659 }
660 if (keys_cmp(c, &upper_key, &u_key) == 0)
661 if (!is_hash_key(c, &u_key)) {
662 err = 7;
663 goto out_dump;
664 }
665 /* Go to next index node */
666 if (le16_to_cpu(idx->level) == 0) {
667 /* At the bottom, so go up until can go right */
668 while (1) {
669 /* Drop the bottom of the list */
670 list_del(&i->list);
671 kfree(i);
672 /* No more list means we are done */
673 if (list_empty(&list))
674 goto out;
675 /* Look at the new bottom */
676 i = list_entry(list.prev, struct idx_node,
677 list);
678 idx = &i->idx;
679 /* Can we go right */
680 if (iip + 1 < le16_to_cpu(idx->child_cnt)) {
681 iip = iip + 1;
682 break;
683 } else
684 /* Nope, so go up again */
685 iip = i->iip;
686 }
687 } else
688 /* Go down left */
689 iip = 0;
690 /*
691 * We have the parent in 'idx' and now we set up for reading the
692 * child pointed to by slot 'iip'.
693 */
694 last_level = le16_to_cpu(idx->level);
695 last_sqnum = le64_to_cpu(idx->ch.sqnum);
696 br = ubifs_idx_branch(c, idx, iip);
697 lnum = le32_to_cpu(br->lnum);
698 offs = le32_to_cpu(br->offs);
699 len = le32_to_cpu(br->len);
700 key_read(c, &br->key, &lower_key);
701 if (iip + 1 < le16_to_cpu(idx->child_cnt)) {
702 br = ubifs_idx_branch(c, idx, iip + 1);
703 key_read(c, &br->key, &upper_key);
704 } else
705 key_copy(c, &i->upper_key, &upper_key);
706 }
707out:
708 err = dbg_old_index_check_init(c, zroot);
709 if (err)
710 goto out_free;
711
712 return 0;
713
714out_dump:
715 ubifs_err(c, "dumping index node (iip=%d)", i->iip);
716 ubifs_dump_node(c, idx);
717 list_del(&i->list);
718 kfree(i);
719 if (!list_empty(&list)) {
720 i = list_entry(list.prev, struct idx_node, list);
721 ubifs_err(c, "dumping parent index node");
722 ubifs_dump_node(c, &i->idx);
723 }
724out_free:
725 while (!list_empty(&list)) {
726 i = list_entry(list.next, struct idx_node, list);
727 list_del(&i->list);
728 kfree(i);
729 }
730 ubifs_err(c, "failed, error %d", err);
731 if (err > 0)
732 err = -EINVAL;
733 return err;
734}
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 * Authors: Adrian Hunter
8 * Artem Bityutskiy (Битюцкий Артём)
9 */
10
11/*
12 * This file implements functions that manage the running of the commit process.
13 * Each affected module has its own functions to accomplish their part in the
14 * commit and those functions are called here.
15 *
16 * The commit is the process whereby all updates to the index and LEB properties
17 * are written out together and the journal becomes empty. This keeps the
18 * file system consistent - at all times the state can be recreated by reading
19 * the index and LEB properties and then replaying the journal.
20 *
21 * The commit is split into two parts named "commit start" and "commit end".
22 * During commit start, the commit process has exclusive access to the journal
23 * by holding the commit semaphore down for writing. As few I/O operations as
24 * possible are performed during commit start, instead the nodes that are to be
25 * written are merely identified. During commit end, the commit semaphore is no
26 * longer held and the journal is again in operation, allowing users to continue
27 * to use the file system while the bulk of the commit I/O is performed. The
28 * purpose of this two-step approach is to prevent the commit from causing any
29 * latency blips. Note that in any case, the commit does not prevent lookups
30 * (as permitted by the TNC mutex), or access to VFS data structures e.g. page
31 * cache.
32 */
33
34#include <linux/freezer.h>
35#include <linux/kthread.h>
36#include <linux/slab.h>
37#include "ubifs.h"
38
39/*
40 * nothing_to_commit - check if there is nothing to commit.
41 * @c: UBIFS file-system description object
42 *
43 * This is a helper function which checks if there is anything to commit. It is
44 * used as an optimization to avoid starting the commit if it is not really
45 * necessary. Indeed, the commit operation always assumes flash I/O (e.g.,
46 * writing the commit start node to the log), and it is better to avoid doing
47 * this unnecessarily. E.g., 'ubifs_sync_fs()' runs the commit, but if there is
48 * nothing to commit, it is more optimal to avoid any flash I/O.
49 *
50 * This function has to be called with @c->commit_sem locked for writing -
51 * this function does not take LPT/TNC locks because the @c->commit_sem
52 * guarantees that we have exclusive access to the TNC and LPT data structures.
53 *
54 * This function returns %1 if there is nothing to commit and %0 otherwise.
55 */
56static int nothing_to_commit(struct ubifs_info *c)
57{
58 /*
59 * During mounting or remounting from R/O mode to R/W mode we may
60 * commit for various recovery-related reasons.
61 */
62 if (c->mounting || c->remounting_rw)
63 return 0;
64
65 /*
66 * If the root TNC node is dirty, we definitely have something to
67 * commit.
68 */
69 if (c->zroot.znode && ubifs_zn_dirty(c->zroot.znode))
70 return 0;
71
72 /*
73 * Increasing @c->dirty_pn_cnt/@c->dirty_nn_cnt and marking
74 * nnodes/pnodes as dirty in run_gc() could race with following
75 * checking, which leads inconsistent states between @c->nroot
76 * and @c->dirty_pn_cnt/@c->dirty_nn_cnt, holding @c->lp_mutex
77 * to avoid that.
78 */
79 mutex_lock(&c->lp_mutex);
80 /*
81 * Even though the TNC is clean, the LPT tree may have dirty nodes. For
82 * example, this may happen if the budgeting subsystem invoked GC to
83 * make some free space, and the GC found an LEB with only dirty and
84 * free space. In this case GC would just change the lprops of this
85 * LEB (by turning all space into free space) and unmap it.
86 */
87 if (c->nroot && test_bit(DIRTY_CNODE, &c->nroot->flags)) {
88 mutex_unlock(&c->lp_mutex);
89 return 0;
90 }
91
92 ubifs_assert(c, atomic_long_read(&c->dirty_zn_cnt) == 0);
93 ubifs_assert(c, c->dirty_pn_cnt == 0);
94 ubifs_assert(c, c->dirty_nn_cnt == 0);
95 mutex_unlock(&c->lp_mutex);
96
97 return 1;
98}
99
100/**
101 * do_commit - commit the journal.
102 * @c: UBIFS file-system description object
103 *
104 * This function implements UBIFS commit. It has to be called with commit lock
105 * locked. Returns zero in case of success and a negative error code in case of
106 * failure.
107 */
108static int do_commit(struct ubifs_info *c)
109{
110 int err, new_ltail_lnum, old_ltail_lnum, i;
111 struct ubifs_zbranch zroot;
112 struct ubifs_lp_stats lst;
113
114 dbg_cmt("start");
115 ubifs_assert(c, !c->ro_media && !c->ro_mount);
116
117 if (c->ro_error) {
118 err = -EROFS;
119 goto out_up;
120 }
121
122 if (nothing_to_commit(c)) {
123 up_write(&c->commit_sem);
124 err = 0;
125 goto out_cancel;
126 }
127
128 /* Sync all write buffers (necessary for recovery) */
129 for (i = 0; i < c->jhead_cnt; i++) {
130 err = ubifs_wbuf_sync(&c->jheads[i].wbuf);
131 if (err)
132 goto out_up;
133 }
134
135 c->cmt_no += 1;
136 err = ubifs_gc_start_commit(c);
137 if (err)
138 goto out_up;
139 err = dbg_check_lprops(c);
140 if (err)
141 goto out_up;
142 err = ubifs_log_start_commit(c, &new_ltail_lnum);
143 if (err)
144 goto out_up;
145 err = ubifs_tnc_start_commit(c, &zroot);
146 if (err)
147 goto out_up;
148 err = ubifs_lpt_start_commit(c);
149 if (err)
150 goto out_up;
151 err = ubifs_orphan_start_commit(c);
152 if (err)
153 goto out_up;
154
155 ubifs_get_lp_stats(c, &lst);
156
157 up_write(&c->commit_sem);
158
159 err = ubifs_tnc_end_commit(c);
160 if (err)
161 goto out;
162 err = ubifs_lpt_end_commit(c);
163 if (err)
164 goto out;
165 err = ubifs_orphan_end_commit(c);
166 if (err)
167 goto out;
168 err = dbg_check_old_index(c, &zroot);
169 if (err)
170 goto out;
171
172 c->mst_node->cmt_no = cpu_to_le64(c->cmt_no);
173 c->mst_node->log_lnum = cpu_to_le32(new_ltail_lnum);
174 c->mst_node->root_lnum = cpu_to_le32(zroot.lnum);
175 c->mst_node->root_offs = cpu_to_le32(zroot.offs);
176 c->mst_node->root_len = cpu_to_le32(zroot.len);
177 c->mst_node->ihead_lnum = cpu_to_le32(c->ihead_lnum);
178 c->mst_node->ihead_offs = cpu_to_le32(c->ihead_offs);
179 c->mst_node->index_size = cpu_to_le64(c->bi.old_idx_sz);
180 c->mst_node->lpt_lnum = cpu_to_le32(c->lpt_lnum);
181 c->mst_node->lpt_offs = cpu_to_le32(c->lpt_offs);
182 c->mst_node->nhead_lnum = cpu_to_le32(c->nhead_lnum);
183 c->mst_node->nhead_offs = cpu_to_le32(c->nhead_offs);
184 c->mst_node->ltab_lnum = cpu_to_le32(c->ltab_lnum);
185 c->mst_node->ltab_offs = cpu_to_le32(c->ltab_offs);
186 c->mst_node->lsave_lnum = cpu_to_le32(c->lsave_lnum);
187 c->mst_node->lsave_offs = cpu_to_le32(c->lsave_offs);
188 c->mst_node->lscan_lnum = cpu_to_le32(c->lscan_lnum);
189 c->mst_node->empty_lebs = cpu_to_le32(lst.empty_lebs);
190 c->mst_node->idx_lebs = cpu_to_le32(lst.idx_lebs);
191 c->mst_node->total_free = cpu_to_le64(lst.total_free);
192 c->mst_node->total_dirty = cpu_to_le64(lst.total_dirty);
193 c->mst_node->total_used = cpu_to_le64(lst.total_used);
194 c->mst_node->total_dead = cpu_to_le64(lst.total_dead);
195 c->mst_node->total_dark = cpu_to_le64(lst.total_dark);
196 if (c->no_orphs)
197 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
198 else
199 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_NO_ORPHS);
200
201 old_ltail_lnum = c->ltail_lnum;
202 err = ubifs_log_end_commit(c, new_ltail_lnum);
203 if (err)
204 goto out;
205
206 err = ubifs_log_post_commit(c, old_ltail_lnum);
207 if (err)
208 goto out;
209 err = ubifs_gc_end_commit(c);
210 if (err)
211 goto out;
212 err = ubifs_lpt_post_commit(c);
213 if (err)
214 goto out;
215
216out_cancel:
217 spin_lock(&c->cs_lock);
218 c->cmt_state = COMMIT_RESTING;
219 wake_up(&c->cmt_wq);
220 dbg_cmt("commit end");
221 spin_unlock(&c->cs_lock);
222 return 0;
223
224out_up:
225 up_write(&c->commit_sem);
226out:
227 ubifs_err(c, "commit failed, error %d", err);
228 spin_lock(&c->cs_lock);
229 c->cmt_state = COMMIT_BROKEN;
230 wake_up(&c->cmt_wq);
231 spin_unlock(&c->cs_lock);
232 ubifs_ro_mode(c, err);
233 return err;
234}
235
236/**
237 * run_bg_commit - run background commit if it is needed.
238 * @c: UBIFS file-system description object
239 *
240 * This function runs background commit if it is needed. Returns zero in case
241 * of success and a negative error code in case of failure.
242 */
243static int run_bg_commit(struct ubifs_info *c)
244{
245 spin_lock(&c->cs_lock);
246 /*
247 * Run background commit only if background commit was requested or if
248 * commit is required.
249 */
250 if (c->cmt_state != COMMIT_BACKGROUND &&
251 c->cmt_state != COMMIT_REQUIRED)
252 goto out;
253 spin_unlock(&c->cs_lock);
254
255 down_write(&c->commit_sem);
256 spin_lock(&c->cs_lock);
257 if (c->cmt_state == COMMIT_REQUIRED)
258 c->cmt_state = COMMIT_RUNNING_REQUIRED;
259 else if (c->cmt_state == COMMIT_BACKGROUND)
260 c->cmt_state = COMMIT_RUNNING_BACKGROUND;
261 else
262 goto out_cmt_unlock;
263 spin_unlock(&c->cs_lock);
264
265 return do_commit(c);
266
267out_cmt_unlock:
268 up_write(&c->commit_sem);
269out:
270 spin_unlock(&c->cs_lock);
271 return 0;
272}
273
274/**
275 * ubifs_bg_thread - UBIFS background thread function.
276 * @info: points to the file-system description object
277 *
278 * This function implements various file-system background activities:
279 * o when a write-buffer timer expires it synchronizes the appropriate
280 * write-buffer;
281 * o when the journal is about to be full, it starts in-advance commit.
282 *
283 * Note, other stuff like background garbage collection may be added here in
284 * future.
285 */
286int ubifs_bg_thread(void *info)
287{
288 int err;
289 struct ubifs_info *c = info;
290
291 ubifs_msg(c, "background thread \"%s\" started, PID %d",
292 c->bgt_name, current->pid);
293 set_freezable();
294
295 while (1) {
296 if (kthread_should_stop())
297 break;
298
299 if (try_to_freeze())
300 continue;
301
302 set_current_state(TASK_INTERRUPTIBLE);
303 /* Check if there is something to do */
304 if (!c->need_bgt) {
305 /*
306 * Nothing prevents us from going sleep now and
307 * be never woken up and block the task which
308 * could wait in 'kthread_stop()' forever.
309 */
310 if (kthread_should_stop())
311 break;
312 schedule();
313 continue;
314 } else
315 __set_current_state(TASK_RUNNING);
316
317 c->need_bgt = 0;
318 err = ubifs_bg_wbufs_sync(c);
319 if (err)
320 ubifs_ro_mode(c, err);
321
322 run_bg_commit(c);
323 cond_resched();
324 }
325
326 ubifs_msg(c, "background thread \"%s\" stops", c->bgt_name);
327 return 0;
328}
329
330/**
331 * ubifs_commit_required - set commit state to "required".
332 * @c: UBIFS file-system description object
333 *
334 * This function is called if a commit is required but cannot be done from the
335 * calling function, so it is just flagged instead.
336 */
337void ubifs_commit_required(struct ubifs_info *c)
338{
339 spin_lock(&c->cs_lock);
340 switch (c->cmt_state) {
341 case COMMIT_RESTING:
342 case COMMIT_BACKGROUND:
343 dbg_cmt("old: %s, new: %s", dbg_cstate(c->cmt_state),
344 dbg_cstate(COMMIT_REQUIRED));
345 c->cmt_state = COMMIT_REQUIRED;
346 break;
347 case COMMIT_RUNNING_BACKGROUND:
348 dbg_cmt("old: %s, new: %s", dbg_cstate(c->cmt_state),
349 dbg_cstate(COMMIT_RUNNING_REQUIRED));
350 c->cmt_state = COMMIT_RUNNING_REQUIRED;
351 break;
352 case COMMIT_REQUIRED:
353 case COMMIT_RUNNING_REQUIRED:
354 case COMMIT_BROKEN:
355 break;
356 }
357 spin_unlock(&c->cs_lock);
358}
359
360/**
361 * ubifs_request_bg_commit - notify the background thread to do a commit.
362 * @c: UBIFS file-system description object
363 *
364 * This function is called if the journal is full enough to make a commit
365 * worthwhile, so background thread is kicked to start it.
366 */
367void ubifs_request_bg_commit(struct ubifs_info *c)
368{
369 spin_lock(&c->cs_lock);
370 if (c->cmt_state == COMMIT_RESTING) {
371 dbg_cmt("old: %s, new: %s", dbg_cstate(c->cmt_state),
372 dbg_cstate(COMMIT_BACKGROUND));
373 c->cmt_state = COMMIT_BACKGROUND;
374 spin_unlock(&c->cs_lock);
375 ubifs_wake_up_bgt(c);
376 } else
377 spin_unlock(&c->cs_lock);
378}
379
380/**
381 * wait_for_commit - wait for commit.
382 * @c: UBIFS file-system description object
383 *
384 * This function sleeps until the commit operation is no longer running.
385 */
386static int wait_for_commit(struct ubifs_info *c)
387{
388 dbg_cmt("pid %d goes sleep", current->pid);
389
390 /*
391 * The following sleeps if the condition is false, and will be woken
392 * when the commit ends. It is possible, although very unlikely, that we
393 * will wake up and see the subsequent commit running, rather than the
394 * one we were waiting for, and go back to sleep. However, we will be
395 * woken again, so there is no danger of sleeping forever.
396 */
397 wait_event(c->cmt_wq, c->cmt_state != COMMIT_RUNNING_BACKGROUND &&
398 c->cmt_state != COMMIT_RUNNING_REQUIRED);
399 dbg_cmt("commit finished, pid %d woke up", current->pid);
400 return 0;
401}
402
403/**
404 * ubifs_run_commit - run or wait for commit.
405 * @c: UBIFS file-system description object
406 *
407 * This function runs commit and returns zero in case of success and a negative
408 * error code in case of failure.
409 */
410int ubifs_run_commit(struct ubifs_info *c)
411{
412 int err = 0;
413
414 spin_lock(&c->cs_lock);
415 if (c->cmt_state == COMMIT_BROKEN) {
416 err = -EROFS;
417 goto out;
418 }
419
420 if (c->cmt_state == COMMIT_RUNNING_BACKGROUND)
421 /*
422 * We set the commit state to 'running required' to indicate
423 * that we want it to complete as quickly as possible.
424 */
425 c->cmt_state = COMMIT_RUNNING_REQUIRED;
426
427 if (c->cmt_state == COMMIT_RUNNING_REQUIRED) {
428 spin_unlock(&c->cs_lock);
429 return wait_for_commit(c);
430 }
431 spin_unlock(&c->cs_lock);
432
433 /* Ok, the commit is indeed needed */
434
435 down_write(&c->commit_sem);
436 spin_lock(&c->cs_lock);
437 /*
438 * Since we unlocked 'c->cs_lock', the state may have changed, so
439 * re-check it.
440 */
441 if (c->cmt_state == COMMIT_BROKEN) {
442 err = -EROFS;
443 goto out_cmt_unlock;
444 }
445
446 if (c->cmt_state == COMMIT_RUNNING_BACKGROUND)
447 c->cmt_state = COMMIT_RUNNING_REQUIRED;
448
449 if (c->cmt_state == COMMIT_RUNNING_REQUIRED) {
450 up_write(&c->commit_sem);
451 spin_unlock(&c->cs_lock);
452 return wait_for_commit(c);
453 }
454 c->cmt_state = COMMIT_RUNNING_REQUIRED;
455 spin_unlock(&c->cs_lock);
456
457 err = do_commit(c);
458 return err;
459
460out_cmt_unlock:
461 up_write(&c->commit_sem);
462out:
463 spin_unlock(&c->cs_lock);
464 return err;
465}
466
467/**
468 * ubifs_gc_should_commit - determine if it is time for GC to run commit.
469 * @c: UBIFS file-system description object
470 *
471 * This function is called by garbage collection to determine if commit should
472 * be run. If commit state is @COMMIT_BACKGROUND, which means that the journal
473 * is full enough to start commit, this function returns true. It is not
474 * absolutely necessary to commit yet, but it feels like this should be better
475 * then to keep doing GC. This function returns %1 if GC has to initiate commit
476 * and %0 if not.
477 */
478int ubifs_gc_should_commit(struct ubifs_info *c)
479{
480 int ret = 0;
481
482 spin_lock(&c->cs_lock);
483 if (c->cmt_state == COMMIT_BACKGROUND) {
484 dbg_cmt("commit required now");
485 c->cmt_state = COMMIT_REQUIRED;
486 } else
487 dbg_cmt("commit not requested");
488 if (c->cmt_state == COMMIT_REQUIRED)
489 ret = 1;
490 spin_unlock(&c->cs_lock);
491 return ret;
492}
493
494/*
495 * Everything below is related to debugging.
496 */
497
498/**
499 * struct idx_node - hold index nodes during index tree traversal.
500 * @list: list
501 * @iip: index in parent (slot number of this indexing node in the parent
502 * indexing node)
503 * @upper_key: all keys in this indexing node have to be less or equivalent to
504 * this key
505 * @idx: index node (8-byte aligned because all node structures must be 8-byte
506 * aligned)
507 */
508struct idx_node {
509 struct list_head list;
510 int iip;
511 union ubifs_key upper_key;
512 struct ubifs_idx_node idx __aligned(8);
513};
514
515/**
516 * dbg_old_index_check_init - get information for the next old index check.
517 * @c: UBIFS file-system description object
518 * @zroot: root of the index
519 *
520 * This function records information about the index that will be needed for the
521 * next old index check i.e. 'dbg_check_old_index()'.
522 *
523 * This function returns %0 on success and a negative error code on failure.
524 */
525int dbg_old_index_check_init(struct ubifs_info *c, struct ubifs_zbranch *zroot)
526{
527 struct ubifs_idx_node *idx;
528 int lnum, offs, len, err = 0;
529 struct ubifs_debug_info *d = c->dbg;
530
531 d->old_zroot = *zroot;
532 lnum = d->old_zroot.lnum;
533 offs = d->old_zroot.offs;
534 len = d->old_zroot.len;
535
536 idx = kmalloc(c->max_idx_node_sz, GFP_NOFS);
537 if (!idx)
538 return -ENOMEM;
539
540 err = ubifs_read_node(c, idx, UBIFS_IDX_NODE, len, lnum, offs);
541 if (err)
542 goto out;
543
544 d->old_zroot_level = le16_to_cpu(idx->level);
545 d->old_zroot_sqnum = le64_to_cpu(idx->ch.sqnum);
546out:
547 kfree(idx);
548 return err;
549}
550
551/**
552 * dbg_check_old_index - check the old copy of the index.
553 * @c: UBIFS file-system description object
554 * @zroot: root of the new index
555 *
556 * In order to be able to recover from an unclean unmount, a complete copy of
557 * the index must exist on flash. This is the "old" index. The commit process
558 * must write the "new" index to flash without overwriting or destroying any
559 * part of the old index. This function is run at commit end in order to check
560 * that the old index does indeed exist completely intact.
561 *
562 * This function returns %0 on success and a negative error code on failure.
563 */
564int dbg_check_old_index(struct ubifs_info *c, struct ubifs_zbranch *zroot)
565{
566 int lnum, offs, len, err = 0, last_level, child_cnt;
567 int first = 1, iip;
568 struct ubifs_debug_info *d = c->dbg;
569 union ubifs_key lower_key, upper_key, l_key, u_key;
570 unsigned long long last_sqnum;
571 struct ubifs_idx_node *idx;
572 struct list_head list;
573 struct idx_node *i;
574 size_t sz;
575
576 if (!dbg_is_chk_index(c))
577 return 0;
578
579 INIT_LIST_HEAD(&list);
580
581 sz = sizeof(struct idx_node) + ubifs_idx_node_sz(c, c->fanout) -
582 UBIFS_IDX_NODE_SZ;
583
584 /* Start at the old zroot */
585 lnum = d->old_zroot.lnum;
586 offs = d->old_zroot.offs;
587 len = d->old_zroot.len;
588 iip = 0;
589
590 /*
591 * Traverse the index tree preorder depth-first i.e. do a node and then
592 * its subtrees from left to right.
593 */
594 while (1) {
595 struct ubifs_branch *br;
596
597 /* Get the next index node */
598 i = kmalloc(sz, GFP_NOFS);
599 if (!i) {
600 err = -ENOMEM;
601 goto out_free;
602 }
603 i->iip = iip;
604 /* Keep the index nodes on our path in a linked list */
605 list_add_tail(&i->list, &list);
606 /* Read the index node */
607 idx = &i->idx;
608 err = ubifs_read_node(c, idx, UBIFS_IDX_NODE, len, lnum, offs);
609 if (err)
610 goto out_free;
611 /* Validate index node */
612 child_cnt = le16_to_cpu(idx->child_cnt);
613 if (child_cnt < 1 || child_cnt > c->fanout) {
614 err = 1;
615 goto out_dump;
616 }
617 if (first) {
618 first = 0;
619 /* Check root level and sqnum */
620 if (le16_to_cpu(idx->level) != d->old_zroot_level) {
621 err = 2;
622 goto out_dump;
623 }
624 if (le64_to_cpu(idx->ch.sqnum) != d->old_zroot_sqnum) {
625 err = 3;
626 goto out_dump;
627 }
628 /* Set last values as though root had a parent */
629 last_level = le16_to_cpu(idx->level) + 1;
630 last_sqnum = le64_to_cpu(idx->ch.sqnum) + 1;
631 key_read(c, ubifs_idx_key(c, idx), &lower_key);
632 highest_ino_key(c, &upper_key, INUM_WATERMARK);
633 }
634 key_copy(c, &upper_key, &i->upper_key);
635 if (le16_to_cpu(idx->level) != last_level - 1) {
636 err = 3;
637 goto out_dump;
638 }
639 /*
640 * The index is always written bottom up hence a child's sqnum
641 * is always less than the parents.
642 */
643 if (le64_to_cpu(idx->ch.sqnum) >= last_sqnum) {
644 err = 4;
645 goto out_dump;
646 }
647 /* Check key range */
648 key_read(c, ubifs_idx_key(c, idx), &l_key);
649 br = ubifs_idx_branch(c, idx, child_cnt - 1);
650 key_read(c, &br->key, &u_key);
651 if (keys_cmp(c, &lower_key, &l_key) > 0) {
652 err = 5;
653 goto out_dump;
654 }
655 if (keys_cmp(c, &upper_key, &u_key) < 0) {
656 err = 6;
657 goto out_dump;
658 }
659 if (keys_cmp(c, &upper_key, &u_key) == 0)
660 if (!is_hash_key(c, &u_key)) {
661 err = 7;
662 goto out_dump;
663 }
664 /* Go to next index node */
665 if (le16_to_cpu(idx->level) == 0) {
666 /* At the bottom, so go up until can go right */
667 while (1) {
668 /* Drop the bottom of the list */
669 list_del(&i->list);
670 kfree(i);
671 /* No more list means we are done */
672 if (list_empty(&list))
673 goto out;
674 /* Look at the new bottom */
675 i = list_entry(list.prev, struct idx_node,
676 list);
677 idx = &i->idx;
678 /* Can we go right */
679 if (iip + 1 < le16_to_cpu(idx->child_cnt)) {
680 iip = iip + 1;
681 break;
682 } else
683 /* Nope, so go up again */
684 iip = i->iip;
685 }
686 } else
687 /* Go down left */
688 iip = 0;
689 /*
690 * We have the parent in 'idx' and now we set up for reading the
691 * child pointed to by slot 'iip'.
692 */
693 last_level = le16_to_cpu(idx->level);
694 last_sqnum = le64_to_cpu(idx->ch.sqnum);
695 br = ubifs_idx_branch(c, idx, iip);
696 lnum = le32_to_cpu(br->lnum);
697 offs = le32_to_cpu(br->offs);
698 len = le32_to_cpu(br->len);
699 key_read(c, &br->key, &lower_key);
700 if (iip + 1 < le16_to_cpu(idx->child_cnt)) {
701 br = ubifs_idx_branch(c, idx, iip + 1);
702 key_read(c, &br->key, &upper_key);
703 } else
704 key_copy(c, &i->upper_key, &upper_key);
705 }
706out:
707 err = dbg_old_index_check_init(c, zroot);
708 if (err)
709 goto out_free;
710
711 return 0;
712
713out_dump:
714 ubifs_err(c, "dumping index node (iip=%d)", i->iip);
715 ubifs_dump_node(c, idx, ubifs_idx_node_sz(c, c->fanout));
716 list_del(&i->list);
717 kfree(i);
718 if (!list_empty(&list)) {
719 i = list_entry(list.prev, struct idx_node, list);
720 ubifs_err(c, "dumping parent index node");
721 ubifs_dump_node(c, &i->idx, ubifs_idx_node_sz(c, c->fanout));
722 }
723out_free:
724 while (!list_empty(&list)) {
725 i = list_entry(list.next, struct idx_node, list);
726 list_del(&i->list);
727 kfree(i);
728 }
729 ubifs_err(c, "failed, error %d", err);
730 if (err > 0)
731 err = -EINVAL;
732 return err;
733}