blob: aa13ad053b14075b8b5a9b0b62067ab0b9904c2d [file] [log] [blame]
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001/*
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090048#include <linux/slab.h>
Artem Bityutskiy1e517642008-07-14 19:08:37 +030049#include "ubifs.h"
50
Artem Bityutskiy944fdef2011-01-16 19:22:02 +020051/*
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 */
Artem Bityutskiyf42eed72011-05-30 14:45:30 +030081 if (c->zroot.znode && ubifs_zn_dirty(c->zroot.znode))
Artem Bityutskiy944fdef2011-01-16 19:22:02 +020082 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
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300101/**
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");
Artem Bityutskiy2ef13292010-09-19 18:34:26 +0300116 ubifs_assert(!c->ro_media && !c->ro_mount);
Artem Bityutskiy2680d722010-09-17 16:44:28 +0300117
118 if (c->ro_error) {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300119 err = -EROFS;
120 goto out_up;
121 }
122
Artem Bityutskiy944fdef2011-01-16 19:22:02 +0200123 if (nothing_to_commit(c)) {
124 up_write(&c->commit_sem);
125 err = 0;
126 goto out_cancel;
127 }
128
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300129 /* 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
Artem Bityutskiy014eb042008-07-21 17:14:29 +0300136 c->cmt_no += 1;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300137 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 old_ltail_lnum = c->ltail_lnum;
170 err = ubifs_log_end_commit(c, new_ltail_lnum);
171 if (err)
172 goto out;
173 err = dbg_check_old_index(c, &zroot);
174 if (err)
175 goto out;
176
Artem Bityutskiy014eb042008-07-21 17:14:29 +0300177 c->mst_node->cmt_no = cpu_to_le64(c->cmt_no);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300178 c->mst_node->log_lnum = cpu_to_le32(new_ltail_lnum);
179 c->mst_node->root_lnum = cpu_to_le32(zroot.lnum);
180 c->mst_node->root_offs = cpu_to_le32(zroot.offs);
181 c->mst_node->root_len = cpu_to_le32(zroot.len);
182 c->mst_node->ihead_lnum = cpu_to_le32(c->ihead_lnum);
183 c->mst_node->ihead_offs = cpu_to_le32(c->ihead_offs);
Artem Bityutskiyb1375452011-03-29 18:04:05 +0300184 c->mst_node->index_size = cpu_to_le64(c->bi.old_idx_sz);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300185 c->mst_node->lpt_lnum = cpu_to_le32(c->lpt_lnum);
186 c->mst_node->lpt_offs = cpu_to_le32(c->lpt_offs);
187 c->mst_node->nhead_lnum = cpu_to_le32(c->nhead_lnum);
188 c->mst_node->nhead_offs = cpu_to_le32(c->nhead_offs);
189 c->mst_node->ltab_lnum = cpu_to_le32(c->ltab_lnum);
190 c->mst_node->ltab_offs = cpu_to_le32(c->ltab_offs);
191 c->mst_node->lsave_lnum = cpu_to_le32(c->lsave_lnum);
192 c->mst_node->lsave_offs = cpu_to_le32(c->lsave_offs);
193 c->mst_node->lscan_lnum = cpu_to_le32(c->lscan_lnum);
194 c->mst_node->empty_lebs = cpu_to_le32(lst.empty_lebs);
195 c->mst_node->idx_lebs = cpu_to_le32(lst.idx_lebs);
196 c->mst_node->total_free = cpu_to_le64(lst.total_free);
197 c->mst_node->total_dirty = cpu_to_le64(lst.total_dirty);
198 c->mst_node->total_used = cpu_to_le64(lst.total_used);
199 c->mst_node->total_dead = cpu_to_le64(lst.total_dead);
200 c->mst_node->total_dark = cpu_to_le64(lst.total_dark);
201 if (c->no_orphs)
202 c->mst_node->flags |= cpu_to_le32(UBIFS_MST_NO_ORPHS);
203 else
204 c->mst_node->flags &= ~cpu_to_le32(UBIFS_MST_NO_ORPHS);
205 err = ubifs_write_master(c);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300206 if (err)
207 goto out;
208
209 err = ubifs_log_post_commit(c, old_ltail_lnum);
210 if (err)
211 goto out;
212 err = ubifs_gc_end_commit(c);
213 if (err)
214 goto out;
215 err = ubifs_lpt_post_commit(c);
216 if (err)
217 goto out;
218
Artem Bityutskiy944fdef2011-01-16 19:22:02 +0200219out_cancel:
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300220 spin_lock(&c->cs_lock);
221 c->cmt_state = COMMIT_RESTING;
222 wake_up(&c->cmt_wq);
223 dbg_cmt("commit end");
224 spin_unlock(&c->cs_lock);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300225 return 0;
226
227out_up:
228 up_write(&c->commit_sem);
229out:
230 ubifs_err("commit failed, error %d", err);
231 spin_lock(&c->cs_lock);
232 c->cmt_state = COMMIT_BROKEN;
233 wake_up(&c->cmt_wq);
234 spin_unlock(&c->cs_lock);
235 ubifs_ro_mode(c, err);
236 return err;
237}
238
239/**
240 * run_bg_commit - run background commit if it is needed.
241 * @c: UBIFS file-system description object
242 *
243 * This function runs background commit if it is needed. Returns zero in case
244 * of success and a negative error code in case of failure.
245 */
246static int run_bg_commit(struct ubifs_info *c)
247{
248 spin_lock(&c->cs_lock);
249 /*
250 * Run background commit only if background commit was requested or if
251 * commit is required.
252 */
253 if (c->cmt_state != COMMIT_BACKGROUND &&
254 c->cmt_state != COMMIT_REQUIRED)
255 goto out;
256 spin_unlock(&c->cs_lock);
257
258 down_write(&c->commit_sem);
259 spin_lock(&c->cs_lock);
260 if (c->cmt_state == COMMIT_REQUIRED)
261 c->cmt_state = COMMIT_RUNNING_REQUIRED;
262 else if (c->cmt_state == COMMIT_BACKGROUND)
263 c->cmt_state = COMMIT_RUNNING_BACKGROUND;
264 else
265 goto out_cmt_unlock;
266 spin_unlock(&c->cs_lock);
267
268 return do_commit(c);
269
270out_cmt_unlock:
271 up_write(&c->commit_sem);
272out:
273 spin_unlock(&c->cs_lock);
274 return 0;
275}
276
277/**
278 * ubifs_bg_thread - UBIFS background thread function.
279 * @info: points to the file-system description object
280 *
281 * This function implements various file-system background activities:
282 * o when a write-buffer timer expires it synchronizes the appropriate
283 * write-buffer;
284 * o when the journal is about to be full, it starts in-advance commit.
285 *
286 * Note, other stuff like background garbage collection may be added here in
287 * future.
288 */
289int ubifs_bg_thread(void *info)
290{
291 int err;
292 struct ubifs_info *c = info;
293
Artem Bityutskiy3668b702012-08-27 16:56:58 +0300294 ubifs_msg("background thread \"%s\" started, PID %d",
295 c->bgt_name, current->pid);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300296 set_freezable();
297
298 while (1) {
299 if (kthread_should_stop())
300 break;
301
302 if (try_to_freeze())
303 continue;
304
305 set_current_state(TASK_INTERRUPTIBLE);
306 /* Check if there is something to do */
307 if (!c->need_bgt) {
308 /*
309 * Nothing prevents us from going sleep now and
310 * be never woken up and block the task which
311 * could wait in 'kthread_stop()' forever.
312 */
313 if (kthread_should_stop())
314 break;
315 schedule();
316 continue;
317 } else
318 __set_current_state(TASK_RUNNING);
319
320 c->need_bgt = 0;
321 err = ubifs_bg_wbufs_sync(c);
322 if (err)
323 ubifs_ro_mode(c, err);
324
325 run_bg_commit(c);
326 cond_resched();
327 }
328
Artem Bityutskiy3668b702012-08-27 16:56:58 +0300329 ubifs_msg("background thread \"%s\" stops", c->bgt_name);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300330 return 0;
331}
332
333/**
334 * ubifs_commit_required - set commit state to "required".
335 * @c: UBIFS file-system description object
336 *
337 * This function is called if a commit is required but cannot be done from the
338 * calling function, so it is just flagged instead.
339 */
340void ubifs_commit_required(struct ubifs_info *c)
341{
342 spin_lock(&c->cs_lock);
343 switch (c->cmt_state) {
344 case COMMIT_RESTING:
345 case COMMIT_BACKGROUND:
346 dbg_cmt("old: %s, new: %s", dbg_cstate(c->cmt_state),
347 dbg_cstate(COMMIT_REQUIRED));
348 c->cmt_state = COMMIT_REQUIRED;
349 break;
350 case COMMIT_RUNNING_BACKGROUND:
351 dbg_cmt("old: %s, new: %s", dbg_cstate(c->cmt_state),
352 dbg_cstate(COMMIT_RUNNING_REQUIRED));
353 c->cmt_state = COMMIT_RUNNING_REQUIRED;
354 break;
355 case COMMIT_REQUIRED:
356 case COMMIT_RUNNING_REQUIRED:
357 case COMMIT_BROKEN:
358 break;
359 }
360 spin_unlock(&c->cs_lock);
361}
362
363/**
364 * ubifs_request_bg_commit - notify the background thread to do a commit.
365 * @c: UBIFS file-system description object
366 *
367 * This function is called if the journal is full enough to make a commit
368 * worthwhile, so background thread is kicked to start it.
369 */
370void ubifs_request_bg_commit(struct ubifs_info *c)
371{
372 spin_lock(&c->cs_lock);
373 if (c->cmt_state == COMMIT_RESTING) {
374 dbg_cmt("old: %s, new: %s", dbg_cstate(c->cmt_state),
375 dbg_cstate(COMMIT_BACKGROUND));
376 c->cmt_state = COMMIT_BACKGROUND;
377 spin_unlock(&c->cs_lock);
378 ubifs_wake_up_bgt(c);
379 } else
380 spin_unlock(&c->cs_lock);
381}
382
383/**
384 * wait_for_commit - wait for commit.
385 * @c: UBIFS file-system description object
386 *
387 * This function sleeps until the commit operation is no longer running.
388 */
389static int wait_for_commit(struct ubifs_info *c)
390{
391 dbg_cmt("pid %d goes sleep", current->pid);
392
393 /*
394 * The following sleeps if the condition is false, and will be woken
395 * when the commit ends. It is possible, although very unlikely, that we
396 * will wake up and see the subsequent commit running, rather than the
397 * one we were waiting for, and go back to sleep. However, we will be
398 * woken again, so there is no danger of sleeping forever.
399 */
400 wait_event(c->cmt_wq, c->cmt_state != COMMIT_RUNNING_BACKGROUND &&
401 c->cmt_state != COMMIT_RUNNING_REQUIRED);
402 dbg_cmt("commit finished, pid %d woke up", current->pid);
403 return 0;
404}
405
406/**
407 * ubifs_run_commit - run or wait for commit.
408 * @c: UBIFS file-system description object
409 *
410 * This function runs commit and returns zero in case of success and a negative
411 * error code in case of failure.
412 */
413int ubifs_run_commit(struct ubifs_info *c)
414{
415 int err = 0;
416
417 spin_lock(&c->cs_lock);
418 if (c->cmt_state == COMMIT_BROKEN) {
Artem Bityutskiy549c9992011-05-26 16:57:14 +0300419 err = -EROFS;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300420 goto out;
421 }
422
423 if (c->cmt_state == COMMIT_RUNNING_BACKGROUND)
424 /*
425 * We set the commit state to 'running required' to indicate
426 * that we want it to complete as quickly as possible.
427 */
428 c->cmt_state = COMMIT_RUNNING_REQUIRED;
429
430 if (c->cmt_state == COMMIT_RUNNING_REQUIRED) {
431 spin_unlock(&c->cs_lock);
432 return wait_for_commit(c);
433 }
434 spin_unlock(&c->cs_lock);
435
436 /* Ok, the commit is indeed needed */
437
438 down_write(&c->commit_sem);
439 spin_lock(&c->cs_lock);
440 /*
441 * Since we unlocked 'c->cs_lock', the state may have changed, so
442 * re-check it.
443 */
444 if (c->cmt_state == COMMIT_BROKEN) {
Artem Bityutskiy549c9992011-05-26 16:57:14 +0300445 err = -EROFS;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300446 goto out_cmt_unlock;
447 }
448
449 if (c->cmt_state == COMMIT_RUNNING_BACKGROUND)
450 c->cmt_state = COMMIT_RUNNING_REQUIRED;
451
452 if (c->cmt_state == COMMIT_RUNNING_REQUIRED) {
453 up_write(&c->commit_sem);
454 spin_unlock(&c->cs_lock);
455 return wait_for_commit(c);
456 }
457 c->cmt_state = COMMIT_RUNNING_REQUIRED;
458 spin_unlock(&c->cs_lock);
459
460 err = do_commit(c);
461 return err;
462
463out_cmt_unlock:
464 up_write(&c->commit_sem);
465out:
466 spin_unlock(&c->cs_lock);
467 return err;
468}
469
470/**
471 * ubifs_gc_should_commit - determine if it is time for GC to run commit.
472 * @c: UBIFS file-system description object
473 *
474 * This function is called by garbage collection to determine if commit should
475 * be run. If commit state is @COMMIT_BACKGROUND, which means that the journal
476 * is full enough to start commit, this function returns true. It is not
477 * absolutely necessary to commit yet, but it feels like this should be better
478 * then to keep doing GC. This function returns %1 if GC has to initiate commit
479 * and %0 if not.
480 */
481int ubifs_gc_should_commit(struct ubifs_info *c)
482{
483 int ret = 0;
484
485 spin_lock(&c->cs_lock);
486 if (c->cmt_state == COMMIT_BACKGROUND) {
487 dbg_cmt("commit required now");
488 c->cmt_state = COMMIT_REQUIRED;
489 } else
490 dbg_cmt("commit not requested");
491 if (c->cmt_state == COMMIT_REQUIRED)
492 ret = 1;
493 spin_unlock(&c->cs_lock);
494 return ret;
495}
496
Artem Bityutskiyf70b7e52012-05-16 19:53:46 +0300497/*
498 * Everything below is related to debugging.
499 */
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300500
501/**
502 * struct idx_node - hold index nodes during index tree traversal.
503 * @list: list
504 * @iip: index in parent (slot number of this indexing node in the parent
505 * indexing node)
506 * @upper_key: all keys in this indexing node have to be less or equivalent to
507 * this key
508 * @idx: index node (8-byte aligned because all node structures must be 8-byte
509 * aligned)
510 */
511struct idx_node {
512 struct list_head list;
513 int iip;
514 union ubifs_key upper_key;
Artem Bityutskiy43457c62012-08-27 13:48:48 +0300515 struct ubifs_idx_node idx __aligned(8);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300516};
517
518/**
519 * dbg_old_index_check_init - get information for the next old index check.
520 * @c: UBIFS file-system description object
521 * @zroot: root of the index
522 *
523 * This function records information about the index that will be needed for the
524 * next old index check i.e. 'dbg_check_old_index()'.
525 *
526 * This function returns %0 on success and a negative error code on failure.
527 */
528int dbg_old_index_check_init(struct ubifs_info *c, struct ubifs_zbranch *zroot)
529{
530 struct ubifs_idx_node *idx;
531 int lnum, offs, len, err = 0;
Artem Bityutskiy17c2f9f2008-10-17 13:31:39 +0300532 struct ubifs_debug_info *d = c->dbg;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300533
Artem Bityutskiy17c2f9f2008-10-17 13:31:39 +0300534 d->old_zroot = *zroot;
535 lnum = d->old_zroot.lnum;
536 offs = d->old_zroot.offs;
537 len = d->old_zroot.len;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300538
539 idx = kmalloc(c->max_idx_node_sz, GFP_NOFS);
540 if (!idx)
541 return -ENOMEM;
542
543 err = ubifs_read_node(c, idx, UBIFS_IDX_NODE, len, lnum, offs);
544 if (err)
545 goto out;
546
Artem Bityutskiy17c2f9f2008-10-17 13:31:39 +0300547 d->old_zroot_level = le16_to_cpu(idx->level);
548 d->old_zroot_sqnum = le64_to_cpu(idx->ch.sqnum);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300549out:
550 kfree(idx);
551 return err;
552}
553
554/**
555 * dbg_check_old_index - check the old copy of the index.
556 * @c: UBIFS file-system description object
557 * @zroot: root of the new index
558 *
559 * In order to be able to recover from an unclean unmount, a complete copy of
560 * the index must exist on flash. This is the "old" index. The commit process
561 * must write the "new" index to flash without overwriting or destroying any
562 * part of the old index. This function is run at commit end in order to check
563 * that the old index does indeed exist completely intact.
564 *
565 * This function returns %0 on success and a negative error code on failure.
566 */
567int dbg_check_old_index(struct ubifs_info *c, struct ubifs_zbranch *zroot)
568{
569 int lnum, offs, len, err = 0, uninitialized_var(last_level), child_cnt;
570 int first = 1, iip;
Artem Bityutskiy17c2f9f2008-10-17 13:31:39 +0300571 struct ubifs_debug_info *d = c->dbg;
Subrata Modak83ef2ec2009-07-15 07:49:03 +0530572 union ubifs_key uninitialized_var(lower_key), upper_key, l_key, u_key;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300573 unsigned long long uninitialized_var(last_sqnum);
574 struct ubifs_idx_node *idx;
575 struct list_head list;
576 struct idx_node *i;
577 size_t sz;
578
Artem Bityutskiy8d7819b2011-06-03 08:53:35 +0300579 if (!dbg_is_chk_index(c))
Artem Bityutskiy8b229c72011-03-25 18:33:57 +0200580 return 0;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300581
582 INIT_LIST_HEAD(&list);
583
584 sz = sizeof(struct idx_node) + ubifs_idx_node_sz(c, c->fanout) -
585 UBIFS_IDX_NODE_SZ;
586
587 /* Start at the old zroot */
Artem Bityutskiy17c2f9f2008-10-17 13:31:39 +0300588 lnum = d->old_zroot.lnum;
589 offs = d->old_zroot.offs;
590 len = d->old_zroot.len;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300591 iip = 0;
592
593 /*
594 * Traverse the index tree preorder depth-first i.e. do a node and then
595 * its subtrees from left to right.
596 */
597 while (1) {
598 struct ubifs_branch *br;
599
600 /* Get the next index node */
601 i = kmalloc(sz, GFP_NOFS);
602 if (!i) {
603 err = -ENOMEM;
604 goto out_free;
605 }
606 i->iip = iip;
607 /* Keep the index nodes on our path in a linked list */
608 list_add_tail(&i->list, &list);
609 /* Read the index node */
610 idx = &i->idx;
611 err = ubifs_read_node(c, idx, UBIFS_IDX_NODE, len, lnum, offs);
612 if (err)
613 goto out_free;
614 /* Validate index node */
615 child_cnt = le16_to_cpu(idx->child_cnt);
616 if (child_cnt < 1 || child_cnt > c->fanout) {
617 err = 1;
618 goto out_dump;
619 }
620 if (first) {
621 first = 0;
622 /* Check root level and sqnum */
Artem Bityutskiy17c2f9f2008-10-17 13:31:39 +0300623 if (le16_to_cpu(idx->level) != d->old_zroot_level) {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300624 err = 2;
625 goto out_dump;
626 }
Artem Bityutskiy17c2f9f2008-10-17 13:31:39 +0300627 if (le64_to_cpu(idx->ch.sqnum) != d->old_zroot_sqnum) {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300628 err = 3;
629 goto out_dump;
630 }
631 /* Set last values as though root had a parent */
632 last_level = le16_to_cpu(idx->level) + 1;
633 last_sqnum = le64_to_cpu(idx->ch.sqnum) + 1;
634 key_read(c, ubifs_idx_key(c, idx), &lower_key);
635 highest_ino_key(c, &upper_key, INUM_WATERMARK);
636 }
637 key_copy(c, &upper_key, &i->upper_key);
638 if (le16_to_cpu(idx->level) != last_level - 1) {
639 err = 3;
640 goto out_dump;
641 }
642 /*
643 * The index is always written bottom up hence a child's sqnum
644 * is always less than the parents.
645 */
646 if (le64_to_cpu(idx->ch.sqnum) >= last_sqnum) {
647 err = 4;
648 goto out_dump;
649 }
650 /* Check key range */
651 key_read(c, ubifs_idx_key(c, idx), &l_key);
652 br = ubifs_idx_branch(c, idx, child_cnt - 1);
653 key_read(c, &br->key, &u_key);
654 if (keys_cmp(c, &lower_key, &l_key) > 0) {
655 err = 5;
656 goto out_dump;
657 }
658 if (keys_cmp(c, &upper_key, &u_key) < 0) {
659 err = 6;
660 goto out_dump;
661 }
662 if (keys_cmp(c, &upper_key, &u_key) == 0)
663 if (!is_hash_key(c, &u_key)) {
664 err = 7;
665 goto out_dump;
666 }
667 /* Go to next index node */
668 if (le16_to_cpu(idx->level) == 0) {
669 /* At the bottom, so go up until can go right */
670 while (1) {
671 /* Drop the bottom of the list */
672 list_del(&i->list);
673 kfree(i);
674 /* No more list means we are done */
675 if (list_empty(&list))
676 goto out;
677 /* Look at the new bottom */
678 i = list_entry(list.prev, struct idx_node,
679 list);
680 idx = &i->idx;
681 /* Can we go right */
682 if (iip + 1 < le16_to_cpu(idx->child_cnt)) {
683 iip = iip + 1;
684 break;
685 } else
686 /* Nope, so go up again */
687 iip = i->iip;
688 }
689 } else
690 /* Go down left */
691 iip = 0;
692 /*
693 * We have the parent in 'idx' and now we set up for reading the
694 * child pointed to by slot 'iip'.
695 */
696 last_level = le16_to_cpu(idx->level);
697 last_sqnum = le64_to_cpu(idx->ch.sqnum);
698 br = ubifs_idx_branch(c, idx, iip);
699 lnum = le32_to_cpu(br->lnum);
700 offs = le32_to_cpu(br->offs);
701 len = le32_to_cpu(br->len);
702 key_read(c, &br->key, &lower_key);
703 if (iip + 1 < le16_to_cpu(idx->child_cnt)) {
704 br = ubifs_idx_branch(c, idx, iip + 1);
705 key_read(c, &br->key, &upper_key);
706 } else
707 key_copy(c, &i->upper_key, &upper_key);
708 }
709out:
710 err = dbg_old_index_check_init(c, zroot);
711 if (err)
712 goto out_free;
713
714 return 0;
715
716out_dump:
Artem Bityutskiya6aae4d2012-05-16 20:11:23 +0300717 ubifs_err("dumping index node (iip=%d)", i->iip);
Artem Bityutskiyedf6be22012-05-16 19:15:56 +0300718 ubifs_dump_node(c, idx);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300719 list_del(&i->list);
720 kfree(i);
721 if (!list_empty(&list)) {
722 i = list_entry(list.prev, struct idx_node, list);
Artem Bityutskiya6aae4d2012-05-16 20:11:23 +0300723 ubifs_err("dumping parent index node");
Artem Bityutskiyedf6be22012-05-16 19:15:56 +0300724 ubifs_dump_node(c, &i->idx);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300725 }
726out_free:
727 while (!list_empty(&list)) {
728 i = list_entry(list.next, struct idx_node, list);
729 list_del(&i->list);
730 kfree(i);
731 }
732 ubifs_err("failed, error %d", err);
733 if (err > 0)
734 err = -EINVAL;
735 return err;
736}