xref: /openbmc/u-boot/fs/ubifs/replay.c (revision 3cbeb0f0)
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 contains journal replay code. It runs when the file-system is being
25  * mounted and requires no locking.
26  *
27  * The larger is the journal, the longer it takes to scan it, so the longer it
28  * takes to mount UBIFS. This is why the journal has limited size which may be
29  * changed depending on the system requirements. But a larger journal gives
30  * faster I/O speed because it writes the index less frequently. So this is a
31  * trade-off. Also, the journal is indexed by the in-memory index (TNC), so the
32  * larger is the journal, the more memory its index may consume.
33  */
34 
35 #include "ubifs.h"
36 
37 /*
38  * Replay flags.
39  *
40  * REPLAY_DELETION: node was deleted
41  * REPLAY_REF: node is a reference node
42  */
43 enum {
44 	REPLAY_DELETION = 1,
45 	REPLAY_REF = 2,
46 };
47 
48 /**
49  * struct replay_entry - replay tree entry.
50  * @lnum: logical eraseblock number of the node
51  * @offs: node offset
52  * @len: node length
53  * @sqnum: node sequence number
54  * @flags: replay flags
55  * @rb: links the replay tree
56  * @key: node key
57  * @nm: directory entry name
58  * @old_size: truncation old size
59  * @new_size: truncation new size
60  * @free: amount of free space in a bud
61  * @dirty: amount of dirty space in a bud from padding and deletion nodes
62  *
63  * UBIFS journal replay must compare node sequence numbers, which means it must
64  * build a tree of node information to insert into the TNC.
65  */
66 struct replay_entry {
67 	int lnum;
68 	int offs;
69 	int len;
70 	unsigned long long sqnum;
71 	int flags;
72 	struct rb_node rb;
73 	union ubifs_key key;
74 	union {
75 		struct qstr nm;
76 		struct {
77 			loff_t old_size;
78 			loff_t new_size;
79 		};
80 		struct {
81 			int free;
82 			int dirty;
83 		};
84 	};
85 };
86 
87 /**
88  * struct bud_entry - entry in the list of buds to replay.
89  * @list: next bud in the list
90  * @bud: bud description object
91  * @free: free bytes in the bud
92  * @sqnum: reference node sequence number
93  */
94 struct bud_entry {
95 	struct list_head list;
96 	struct ubifs_bud *bud;
97 	int free;
98 	unsigned long long sqnum;
99 };
100 
101 /**
102  * set_bud_lprops - set free and dirty space used by a bud.
103  * @c: UBIFS file-system description object
104  * @r: replay entry of bud
105  */
106 static int set_bud_lprops(struct ubifs_info *c, struct replay_entry *r)
107 {
108 	const struct ubifs_lprops *lp;
109 	int err = 0, dirty;
110 
111 	ubifs_get_lprops(c);
112 
113 	lp = ubifs_lpt_lookup_dirty(c, r->lnum);
114 	if (IS_ERR(lp)) {
115 		err = PTR_ERR(lp);
116 		goto out;
117 	}
118 
119 	dirty = lp->dirty;
120 	if (r->offs == 0 && (lp->free != c->leb_size || lp->dirty != 0)) {
121 		/*
122 		 * The LEB was added to the journal with a starting offset of
123 		 * zero which means the LEB must have been empty. The LEB
124 		 * property values should be lp->free == c->leb_size and
125 		 * lp->dirty == 0, but that is not the case. The reason is that
126 		 * the LEB was garbage collected. The garbage collector resets
127 		 * the free and dirty space without recording it anywhere except
128 		 * lprops, so if there is not a commit then lprops does not have
129 		 * that information next time the file system is mounted.
130 		 *
131 		 * We do not need to adjust free space because the scan has told
132 		 * us the exact value which is recorded in the replay entry as
133 		 * r->free.
134 		 *
135 		 * However we do need to subtract from the dirty space the
136 		 * amount of space that the garbage collector reclaimed, which
137 		 * is the whole LEB minus the amount of space that was free.
138 		 */
139 		dbg_mnt("bud LEB %d was GC'd (%d free, %d dirty)", r->lnum,
140 			lp->free, lp->dirty);
141 		dbg_gc("bud LEB %d was GC'd (%d free, %d dirty)", r->lnum,
142 			lp->free, lp->dirty);
143 		dirty -= c->leb_size - lp->free;
144 		/*
145 		 * If the replay order was perfect the dirty space would now be
146 		 * zero. The order is not perfect because the the journal heads
147 		 * race with each other. This is not a problem but is does mean
148 		 * that the dirty space may temporarily exceed c->leb_size
149 		 * during the replay.
150 		 */
151 		if (dirty != 0)
152 			dbg_msg("LEB %d lp: %d free %d dirty "
153 				"replay: %d free %d dirty", r->lnum, lp->free,
154 				lp->dirty, r->free, r->dirty);
155 	}
156 	lp = ubifs_change_lp(c, lp, r->free, dirty + r->dirty,
157 			     lp->flags | LPROPS_TAKEN, 0);
158 	if (IS_ERR(lp)) {
159 		err = PTR_ERR(lp);
160 		goto out;
161 	}
162 out:
163 	ubifs_release_lprops(c);
164 	return err;
165 }
166 
167 /**
168  * trun_remove_range - apply a replay entry for a truncation to the TNC.
169  * @c: UBIFS file-system description object
170  * @r: replay entry of truncation
171  */
172 static int trun_remove_range(struct ubifs_info *c, struct replay_entry *r)
173 {
174 	unsigned min_blk, max_blk;
175 	union ubifs_key min_key, max_key;
176 	ino_t ino;
177 
178 	min_blk = r->new_size / UBIFS_BLOCK_SIZE;
179 	if (r->new_size & (UBIFS_BLOCK_SIZE - 1))
180 		min_blk += 1;
181 
182 	max_blk = r->old_size / UBIFS_BLOCK_SIZE;
183 	if ((r->old_size & (UBIFS_BLOCK_SIZE - 1)) == 0)
184 		max_blk -= 1;
185 
186 	ino = key_inum(c, &r->key);
187 
188 	data_key_init(c, &min_key, ino, min_blk);
189 	data_key_init(c, &max_key, ino, max_blk);
190 
191 	return ubifs_tnc_remove_range(c, &min_key, &max_key);
192 }
193 
194 /**
195  * apply_replay_entry - apply a replay entry to the TNC.
196  * @c: UBIFS file-system description object
197  * @r: replay entry to apply
198  *
199  * Apply a replay entry to the TNC.
200  */
201 static int apply_replay_entry(struct ubifs_info *c, struct replay_entry *r)
202 {
203 	int err, deletion = ((r->flags & REPLAY_DELETION) != 0);
204 
205 	dbg_mnt("LEB %d:%d len %d flgs %d sqnum %llu %s", r->lnum,
206 		r->offs, r->len, r->flags, r->sqnum, DBGKEY(&r->key));
207 
208 	/* Set c->replay_sqnum to help deal with dangling branches. */
209 	c->replay_sqnum = r->sqnum;
210 
211 	if (r->flags & REPLAY_REF)
212 		err = set_bud_lprops(c, r);
213 	else if (is_hash_key(c, &r->key)) {
214 		if (deletion)
215 			err = ubifs_tnc_remove_nm(c, &r->key, &r->nm);
216 		else
217 			err = ubifs_tnc_add_nm(c, &r->key, r->lnum, r->offs,
218 					       r->len, &r->nm);
219 	} else {
220 		if (deletion)
221 			switch (key_type(c, &r->key)) {
222 			case UBIFS_INO_KEY:
223 			{
224 				ino_t inum = key_inum(c, &r->key);
225 
226 				err = ubifs_tnc_remove_ino(c, inum);
227 				break;
228 			}
229 			case UBIFS_TRUN_KEY:
230 				err = trun_remove_range(c, r);
231 				break;
232 			default:
233 				err = ubifs_tnc_remove(c, &r->key);
234 				break;
235 			}
236 		else
237 			err = ubifs_tnc_add(c, &r->key, r->lnum, r->offs,
238 					    r->len);
239 		if (err)
240 			return err;
241 
242 		if (c->need_recovery)
243 			err = ubifs_recover_size_accum(c, &r->key, deletion,
244 						       r->new_size);
245 	}
246 
247 	return err;
248 }
249 
250 /**
251  * destroy_replay_tree - destroy the replay.
252  * @c: UBIFS file-system description object
253  *
254  * Destroy the replay tree.
255  */
256 static void destroy_replay_tree(struct ubifs_info *c)
257 {
258 	struct rb_node *this = c->replay_tree.rb_node;
259 	struct replay_entry *r;
260 
261 	while (this) {
262 		if (this->rb_left) {
263 			this = this->rb_left;
264 			continue;
265 		} else if (this->rb_right) {
266 			this = this->rb_right;
267 			continue;
268 		}
269 		r = rb_entry(this, struct replay_entry, rb);
270 		this = rb_parent(this);
271 		if (this) {
272 			if (this->rb_left == &r->rb)
273 				this->rb_left = NULL;
274 			else
275 				this->rb_right = NULL;
276 		}
277 		if (is_hash_key(c, &r->key))
278 			kfree((void *)r->nm.name);
279 		kfree(r);
280 	}
281 	c->replay_tree = RB_ROOT;
282 }
283 
284 /**
285  * apply_replay_tree - apply the replay tree to the TNC.
286  * @c: UBIFS file-system description object
287  *
288  * Apply the replay tree.
289  * Returns zero in case of success and a negative error code in case of
290  * failure.
291  */
292 static int apply_replay_tree(struct ubifs_info *c)
293 {
294 	struct rb_node *this = rb_first(&c->replay_tree);
295 
296 	while (this) {
297 		struct replay_entry *r;
298 		int err;
299 
300 		cond_resched();
301 
302 		r = rb_entry(this, struct replay_entry, rb);
303 		err = apply_replay_entry(c, r);
304 		if (err)
305 			return err;
306 		this = rb_next(this);
307 	}
308 	return 0;
309 }
310 
311 /**
312  * insert_node - insert a node to the replay tree.
313  * @c: UBIFS file-system description object
314  * @lnum: node logical eraseblock number
315  * @offs: node offset
316  * @len: node length
317  * @key: node key
318  * @sqnum: sequence number
319  * @deletion: non-zero if this is a deletion
320  * @used: number of bytes in use in a LEB
321  * @old_size: truncation old size
322  * @new_size: truncation new size
323  *
324  * This function inserts a scanned non-direntry node to the replay tree. The
325  * replay tree is an RB-tree containing @struct replay_entry elements which are
326  * indexed by the sequence number. The replay tree is applied at the very end
327  * of the replay process. Since the tree is sorted in sequence number order,
328  * the older modifications are applied first. This function returns zero in
329  * case of success and a negative error code in case of failure.
330  */
331 static int insert_node(struct ubifs_info *c, int lnum, int offs, int len,
332 		       union ubifs_key *key, unsigned long long sqnum,
333 		       int deletion, int *used, loff_t old_size,
334 		       loff_t new_size)
335 {
336 	struct rb_node **p = &c->replay_tree.rb_node, *parent = NULL;
337 	struct replay_entry *r;
338 
339 	if (key_inum(c, key) >= c->highest_inum)
340 		c->highest_inum = key_inum(c, key);
341 
342 	dbg_mnt("add LEB %d:%d, key %s", lnum, offs, DBGKEY(key));
343 	while (*p) {
344 		parent = *p;
345 		r = rb_entry(parent, struct replay_entry, rb);
346 		if (sqnum < r->sqnum) {
347 			p = &(*p)->rb_left;
348 			continue;
349 		} else if (sqnum > r->sqnum) {
350 			p = &(*p)->rb_right;
351 			continue;
352 		}
353 		ubifs_err("duplicate sqnum in replay");
354 		return -EINVAL;
355 	}
356 
357 	r = kzalloc(sizeof(struct replay_entry), GFP_KERNEL);
358 	if (!r)
359 		return -ENOMEM;
360 
361 	if (!deletion)
362 		*used += ALIGN(len, 8);
363 	r->lnum = lnum;
364 	r->offs = offs;
365 	r->len = len;
366 	r->sqnum = sqnum;
367 	r->flags = (deletion ? REPLAY_DELETION : 0);
368 	r->old_size = old_size;
369 	r->new_size = new_size;
370 	key_copy(c, key, &r->key);
371 
372 	rb_link_node(&r->rb, parent, p);
373 	rb_insert_color(&r->rb, &c->replay_tree);
374 	return 0;
375 }
376 
377 /**
378  * insert_dent - insert a directory entry node into the replay tree.
379  * @c: UBIFS file-system description object
380  * @lnum: node logical eraseblock number
381  * @offs: node offset
382  * @len: node length
383  * @key: node key
384  * @name: directory entry name
385  * @nlen: directory entry name length
386  * @sqnum: sequence number
387  * @deletion: non-zero if this is a deletion
388  * @used: number of bytes in use in a LEB
389  *
390  * This function inserts a scanned directory entry node to the replay tree.
391  * Returns zero in case of success and a negative error code in case of
392  * failure.
393  *
394  * This function is also used for extended attribute entries because they are
395  * implemented as directory entry nodes.
396  */
397 static int insert_dent(struct ubifs_info *c, int lnum, int offs, int len,
398 		       union ubifs_key *key, const char *name, int nlen,
399 		       unsigned long long sqnum, int deletion, int *used)
400 {
401 	struct rb_node **p = &c->replay_tree.rb_node, *parent = NULL;
402 	struct replay_entry *r;
403 	char *nbuf;
404 
405 	if (key_inum(c, key) >= c->highest_inum)
406 		c->highest_inum = key_inum(c, key);
407 
408 	dbg_mnt("add LEB %d:%d, key %s", lnum, offs, DBGKEY(key));
409 	while (*p) {
410 		parent = *p;
411 		r = rb_entry(parent, struct replay_entry, rb);
412 		if (sqnum < r->sqnum) {
413 			p = &(*p)->rb_left;
414 			continue;
415 		}
416 		if (sqnum > r->sqnum) {
417 			p = &(*p)->rb_right;
418 			continue;
419 		}
420 		ubifs_err("duplicate sqnum in replay");
421 		return -EINVAL;
422 	}
423 
424 	r = kzalloc(sizeof(struct replay_entry), GFP_KERNEL);
425 	if (!r)
426 		return -ENOMEM;
427 	nbuf = kmalloc(nlen + 1, GFP_KERNEL);
428 	if (!nbuf) {
429 		kfree(r);
430 		return -ENOMEM;
431 	}
432 
433 	if (!deletion)
434 		*used += ALIGN(len, 8);
435 	r->lnum = lnum;
436 	r->offs = offs;
437 	r->len = len;
438 	r->sqnum = sqnum;
439 	r->nm.len = nlen;
440 	memcpy(nbuf, name, nlen);
441 	nbuf[nlen] = '\0';
442 	r->nm.name = nbuf;
443 	r->flags = (deletion ? REPLAY_DELETION : 0);
444 	key_copy(c, key, &r->key);
445 
446 	ubifs_assert(!*p);
447 	rb_link_node(&r->rb, parent, p);
448 	rb_insert_color(&r->rb, &c->replay_tree);
449 	return 0;
450 }
451 
452 /**
453  * ubifs_validate_entry - validate directory or extended attribute entry node.
454  * @c: UBIFS file-system description object
455  * @dent: the node to validate
456  *
457  * This function validates directory or extended attribute entry node @dent.
458  * Returns zero if the node is all right and a %-EINVAL if not.
459  */
460 int ubifs_validate_entry(struct ubifs_info *c,
461 			 const struct ubifs_dent_node *dent)
462 {
463 	int key_type = key_type_flash(c, dent->key);
464 	int nlen = le16_to_cpu(dent->nlen);
465 
466 	if (le32_to_cpu(dent->ch.len) != nlen + UBIFS_DENT_NODE_SZ + 1 ||
467 	    dent->type >= UBIFS_ITYPES_CNT ||
468 	    nlen > UBIFS_MAX_NLEN || dent->name[nlen] != 0 ||
469 	    strnlen((char *)dent->name, nlen) != nlen ||
470 	    le64_to_cpu(dent->inum) > MAX_INUM) {
471 		ubifs_err("bad %s node", key_type == UBIFS_DENT_KEY ?
472 			  "directory entry" : "extended attribute entry");
473 		return -EINVAL;
474 	}
475 
476 	if (key_type != UBIFS_DENT_KEY && key_type != UBIFS_XENT_KEY) {
477 		ubifs_err("bad key type %d", key_type);
478 		return -EINVAL;
479 	}
480 
481 	return 0;
482 }
483 
484 /**
485  * replay_bud - replay a bud logical eraseblock.
486  * @c: UBIFS file-system description object
487  * @lnum: bud logical eraseblock number to replay
488  * @offs: bud start offset
489  * @jhead: journal head to which this bud belongs
490  * @free: amount of free space in the bud is returned here
491  * @dirty: amount of dirty space from padding and deletion nodes is returned
492  * here
493  *
494  * This function returns zero in case of success and a negative error code in
495  * case of failure.
496  */
497 static int replay_bud(struct ubifs_info *c, int lnum, int offs, int jhead,
498 		      int *free, int *dirty)
499 {
500 	int err = 0, used = 0;
501 	struct ubifs_scan_leb *sleb;
502 	struct ubifs_scan_node *snod;
503 	struct ubifs_bud *bud;
504 
505 	dbg_mnt("replay bud LEB %d, head %d", lnum, jhead);
506 	if (c->need_recovery)
507 		sleb = ubifs_recover_leb(c, lnum, offs, c->sbuf, jhead != GCHD);
508 	else
509 		sleb = ubifs_scan(c, lnum, offs, c->sbuf);
510 	if (IS_ERR(sleb))
511 		return PTR_ERR(sleb);
512 
513 	/*
514 	 * The bud does not have to start from offset zero - the beginning of
515 	 * the 'lnum' LEB may contain previously committed data. One of the
516 	 * things we have to do in replay is to correctly update lprops with
517 	 * newer information about this LEB.
518 	 *
519 	 * At this point lprops thinks that this LEB has 'c->leb_size - offs'
520 	 * bytes of free space because it only contain information about
521 	 * committed data.
522 	 *
523 	 * But we know that real amount of free space is 'c->leb_size -
524 	 * sleb->endpt', and the space in the 'lnum' LEB between 'offs' and
525 	 * 'sleb->endpt' is used by bud data. We have to correctly calculate
526 	 * how much of these data are dirty and update lprops with this
527 	 * information.
528 	 *
529 	 * The dirt in that LEB region is comprised of padding nodes, deletion
530 	 * nodes, truncation nodes and nodes which are obsoleted by subsequent
531 	 * nodes in this LEB. So instead of calculating clean space, we
532 	 * calculate used space ('used' variable).
533 	 */
534 
535 	list_for_each_entry(snod, &sleb->nodes, list) {
536 		int deletion = 0;
537 
538 		cond_resched();
539 
540 		if (snod->sqnum >= SQNUM_WATERMARK) {
541 			ubifs_err("file system's life ended");
542 			goto out_dump;
543 		}
544 
545 		if (snod->sqnum > c->max_sqnum)
546 			c->max_sqnum = snod->sqnum;
547 
548 		switch (snod->type) {
549 		case UBIFS_INO_NODE:
550 		{
551 			struct ubifs_ino_node *ino = snod->node;
552 			loff_t new_size = le64_to_cpu(ino->size);
553 
554 			if (le32_to_cpu(ino->nlink) == 0)
555 				deletion = 1;
556 			err = insert_node(c, lnum, snod->offs, snod->len,
557 					  &snod->key, snod->sqnum, deletion,
558 					  &used, 0, new_size);
559 			break;
560 		}
561 		case UBIFS_DATA_NODE:
562 		{
563 			struct ubifs_data_node *dn = snod->node;
564 			loff_t new_size = le32_to_cpu(dn->size) +
565 					  key_block(c, &snod->key) *
566 					  UBIFS_BLOCK_SIZE;
567 
568 			err = insert_node(c, lnum, snod->offs, snod->len,
569 					  &snod->key, snod->sqnum, deletion,
570 					  &used, 0, new_size);
571 			break;
572 		}
573 		case UBIFS_DENT_NODE:
574 		case UBIFS_XENT_NODE:
575 		{
576 			struct ubifs_dent_node *dent = snod->node;
577 
578 			err = ubifs_validate_entry(c, dent);
579 			if (err)
580 				goto out_dump;
581 
582 			err = insert_dent(c, lnum, snod->offs, snod->len,
583 					  &snod->key, (char *)dent->name,
584 					  le16_to_cpu(dent->nlen), snod->sqnum,
585 					  !le64_to_cpu(dent->inum), &used);
586 			break;
587 		}
588 		case UBIFS_TRUN_NODE:
589 		{
590 			struct ubifs_trun_node *trun = snod->node;
591 			loff_t old_size = le64_to_cpu(trun->old_size);
592 			loff_t new_size = le64_to_cpu(trun->new_size);
593 			union ubifs_key key;
594 
595 			/* Validate truncation node */
596 			if (old_size < 0 || old_size > c->max_inode_sz ||
597 			    new_size < 0 || new_size > c->max_inode_sz ||
598 			    old_size <= new_size) {
599 				ubifs_err("bad truncation node");
600 				goto out_dump;
601 			}
602 
603 			/*
604 			 * Create a fake truncation key just to use the same
605 			 * functions which expect nodes to have keys.
606 			 */
607 			trun_key_init(c, &key, le32_to_cpu(trun->inum));
608 			err = insert_node(c, lnum, snod->offs, snod->len,
609 					  &key, snod->sqnum, 1, &used,
610 					  old_size, new_size);
611 			break;
612 		}
613 		default:
614 			ubifs_err("unexpected node type %d in bud LEB %d:%d",
615 				  snod->type, lnum, snod->offs);
616 			err = -EINVAL;
617 			goto out_dump;
618 		}
619 		if (err)
620 			goto out;
621 	}
622 
623 	bud = ubifs_search_bud(c, lnum);
624 	if (!bud)
625 		BUG();
626 
627 	ubifs_assert(sleb->endpt - offs >= used);
628 	ubifs_assert(sleb->endpt % c->min_io_size == 0);
629 
630 	*dirty = sleb->endpt - offs - used;
631 	*free = c->leb_size - sleb->endpt;
632 
633 out:
634 	ubifs_scan_destroy(sleb);
635 	return err;
636 
637 out_dump:
638 	ubifs_err("bad node is at LEB %d:%d", lnum, snod->offs);
639 	dbg_dump_node(c, snod->node);
640 	ubifs_scan_destroy(sleb);
641 	return -EINVAL;
642 }
643 
644 /**
645  * insert_ref_node - insert a reference node to the replay tree.
646  * @c: UBIFS file-system description object
647  * @lnum: node logical eraseblock number
648  * @offs: node offset
649  * @sqnum: sequence number
650  * @free: amount of free space in bud
651  * @dirty: amount of dirty space from padding and deletion nodes
652  *
653  * This function inserts a reference node to the replay tree and returns zero
654  * in case of success or a negative error code in case of failure.
655  */
656 static int insert_ref_node(struct ubifs_info *c, int lnum, int offs,
657 			   unsigned long long sqnum, int free, int dirty)
658 {
659 	struct rb_node **p = &c->replay_tree.rb_node, *parent = NULL;
660 	struct replay_entry *r;
661 
662 	dbg_mnt("add ref LEB %d:%d", lnum, offs);
663 	while (*p) {
664 		parent = *p;
665 		r = rb_entry(parent, struct replay_entry, rb);
666 		if (sqnum < r->sqnum) {
667 			p = &(*p)->rb_left;
668 			continue;
669 		} else if (sqnum > r->sqnum) {
670 			p = &(*p)->rb_right;
671 			continue;
672 		}
673 		ubifs_err("duplicate sqnum in replay tree");
674 		return -EINVAL;
675 	}
676 
677 	r = kzalloc(sizeof(struct replay_entry), GFP_KERNEL);
678 	if (!r)
679 		return -ENOMEM;
680 
681 	r->lnum = lnum;
682 	r->offs = offs;
683 	r->sqnum = sqnum;
684 	r->flags = REPLAY_REF;
685 	r->free = free;
686 	r->dirty = dirty;
687 
688 	rb_link_node(&r->rb, parent, p);
689 	rb_insert_color(&r->rb, &c->replay_tree);
690 	return 0;
691 }
692 
693 /**
694  * replay_buds - replay all buds.
695  * @c: UBIFS file-system description object
696  *
697  * This function returns zero in case of success and a negative error code in
698  * case of failure.
699  */
700 static int replay_buds(struct ubifs_info *c)
701 {
702 	struct bud_entry *b;
703 	int err, uninitialized_var(free), uninitialized_var(dirty);
704 
705 	list_for_each_entry(b, &c->replay_buds, list) {
706 		err = replay_bud(c, b->bud->lnum, b->bud->start, b->bud->jhead,
707 				 &free, &dirty);
708 		if (err)
709 			return err;
710 		err = insert_ref_node(c, b->bud->lnum, b->bud->start, b->sqnum,
711 				      free, dirty);
712 		if (err)
713 			return err;
714 	}
715 
716 	return 0;
717 }
718 
719 /**
720  * destroy_bud_list - destroy the list of buds to replay.
721  * @c: UBIFS file-system description object
722  */
723 static void destroy_bud_list(struct ubifs_info *c)
724 {
725 	struct bud_entry *b;
726 
727 	while (!list_empty(&c->replay_buds)) {
728 		b = list_entry(c->replay_buds.next, struct bud_entry, list);
729 		list_del(&b->list);
730 		kfree(b);
731 	}
732 }
733 
734 /**
735  * add_replay_bud - add a bud to the list of buds to replay.
736  * @c: UBIFS file-system description object
737  * @lnum: bud logical eraseblock number to replay
738  * @offs: bud start offset
739  * @jhead: journal head to which this bud belongs
740  * @sqnum: reference node sequence number
741  *
742  * This function returns zero in case of success and a negative error code in
743  * case of failure.
744  */
745 static int add_replay_bud(struct ubifs_info *c, int lnum, int offs, int jhead,
746 			  unsigned long long sqnum)
747 {
748 	struct ubifs_bud *bud;
749 	struct bud_entry *b;
750 
751 	dbg_mnt("add replay bud LEB %d:%d, head %d", lnum, offs, jhead);
752 
753 	bud = kmalloc(sizeof(struct ubifs_bud), GFP_KERNEL);
754 	if (!bud)
755 		return -ENOMEM;
756 
757 	b = kmalloc(sizeof(struct bud_entry), GFP_KERNEL);
758 	if (!b) {
759 		kfree(bud);
760 		return -ENOMEM;
761 	}
762 
763 	bud->lnum = lnum;
764 	bud->start = offs;
765 	bud->jhead = jhead;
766 	ubifs_add_bud(c, bud);
767 
768 	b->bud = bud;
769 	b->sqnum = sqnum;
770 	list_add_tail(&b->list, &c->replay_buds);
771 
772 	return 0;
773 }
774 
775 /**
776  * validate_ref - validate a reference node.
777  * @c: UBIFS file-system description object
778  * @ref: the reference node to validate
779  * @ref_lnum: LEB number of the reference node
780  * @ref_offs: reference node offset
781  *
782  * This function returns %1 if a bud reference already exists for the LEB. %0 is
783  * returned if the reference node is new, otherwise %-EINVAL is returned if
784  * validation failed.
785  */
786 static int validate_ref(struct ubifs_info *c, const struct ubifs_ref_node *ref)
787 {
788 	struct ubifs_bud *bud;
789 	int lnum = le32_to_cpu(ref->lnum);
790 	unsigned int offs = le32_to_cpu(ref->offs);
791 	unsigned int jhead = le32_to_cpu(ref->jhead);
792 
793 	/*
794 	 * ref->offs may point to the end of LEB when the journal head points
795 	 * to the end of LEB and we write reference node for it during commit.
796 	 * So this is why we require 'offs > c->leb_size'.
797 	 */
798 	if (jhead >= c->jhead_cnt || lnum >= c->leb_cnt ||
799 	    lnum < c->main_first || offs > c->leb_size ||
800 	    offs & (c->min_io_size - 1))
801 		return -EINVAL;
802 
803 	/* Make sure we have not already looked at this bud */
804 	bud = ubifs_search_bud(c, lnum);
805 	if (bud) {
806 		if (bud->jhead == jhead && bud->start <= offs)
807 			return 1;
808 		ubifs_err("bud at LEB %d:%d was already referred", lnum, offs);
809 		return -EINVAL;
810 	}
811 
812 	return 0;
813 }
814 
815 /**
816  * replay_log_leb - replay a log logical eraseblock.
817  * @c: UBIFS file-system description object
818  * @lnum: log logical eraseblock to replay
819  * @offs: offset to start replaying from
820  * @sbuf: scan buffer
821  *
822  * This function replays a log LEB and returns zero in case of success, %1 if
823  * this is the last LEB in the log, and a negative error code in case of
824  * failure.
825  */
826 static int replay_log_leb(struct ubifs_info *c, int lnum, int offs, void *sbuf)
827 {
828 	int err;
829 	struct ubifs_scan_leb *sleb;
830 	struct ubifs_scan_node *snod;
831 	const struct ubifs_cs_node *node;
832 
833 	dbg_mnt("replay log LEB %d:%d", lnum, offs);
834 	sleb = ubifs_scan(c, lnum, offs, sbuf);
835 	if (IS_ERR(sleb)) {
836 		if (c->need_recovery)
837 			sleb = ubifs_recover_log_leb(c, lnum, offs, sbuf);
838 		if (IS_ERR(sleb))
839 			return PTR_ERR(sleb);
840 	}
841 
842 	if (sleb->nodes_cnt == 0) {
843 		err = 1;
844 		goto out;
845 	}
846 
847 	node = sleb->buf;
848 
849 	snod = list_entry(sleb->nodes.next, struct ubifs_scan_node, list);
850 	if (c->cs_sqnum == 0) {
851 		/*
852 		 * This is the first log LEB we are looking at, make sure that
853 		 * the first node is a commit start node. Also record its
854 		 * sequence number so that UBIFS can determine where the log
855 		 * ends, because all nodes which were have higher sequence
856 		 * numbers.
857 		 */
858 		if (snod->type != UBIFS_CS_NODE) {
859 			dbg_err("first log node at LEB %d:%d is not CS node",
860 				lnum, offs);
861 			goto out_dump;
862 		}
863 		if (le64_to_cpu(node->cmt_no) != c->cmt_no) {
864 			dbg_err("first CS node at LEB %d:%d has wrong "
865 				"commit number %llu expected %llu",
866 				lnum, offs,
867 				(unsigned long long)le64_to_cpu(node->cmt_no),
868 				c->cmt_no);
869 			goto out_dump;
870 		}
871 
872 		c->cs_sqnum = le64_to_cpu(node->ch.sqnum);
873 		dbg_mnt("commit start sqnum %llu", c->cs_sqnum);
874 	}
875 
876 	if (snod->sqnum < c->cs_sqnum) {
877 		/*
878 		 * This means that we reached end of log and now
879 		 * look to the older log data, which was already
880 		 * committed but the eraseblock was not erased (UBIFS
881 		 * only un-maps it). So this basically means we have to
882 		 * exit with "end of log" code.
883 		 */
884 		err = 1;
885 		goto out;
886 	}
887 
888 	/* Make sure the first node sits at offset zero of the LEB */
889 	if (snod->offs != 0) {
890 		dbg_err("first node is not at zero offset");
891 		goto out_dump;
892 	}
893 
894 	list_for_each_entry(snod, &sleb->nodes, list) {
895 
896 		cond_resched();
897 
898 		if (snod->sqnum >= SQNUM_WATERMARK) {
899 			ubifs_err("file system's life ended");
900 			goto out_dump;
901 		}
902 
903 		if (snod->sqnum < c->cs_sqnum) {
904 			dbg_err("bad sqnum %llu, commit sqnum %llu",
905 				snod->sqnum, c->cs_sqnum);
906 			goto out_dump;
907 		}
908 
909 		if (snod->sqnum > c->max_sqnum)
910 			c->max_sqnum = snod->sqnum;
911 
912 		switch (snod->type) {
913 		case UBIFS_REF_NODE: {
914 			const struct ubifs_ref_node *ref = snod->node;
915 
916 			err = validate_ref(c, ref);
917 			if (err == 1)
918 				break; /* Already have this bud */
919 			if (err)
920 				goto out_dump;
921 
922 			err = add_replay_bud(c, le32_to_cpu(ref->lnum),
923 					     le32_to_cpu(ref->offs),
924 					     le32_to_cpu(ref->jhead),
925 					     snod->sqnum);
926 			if (err)
927 				goto out;
928 
929 			break;
930 		}
931 		case UBIFS_CS_NODE:
932 			/* Make sure it sits at the beginning of LEB */
933 			if (snod->offs != 0) {
934 				ubifs_err("unexpected node in log");
935 				goto out_dump;
936 			}
937 			break;
938 		default:
939 			ubifs_err("unexpected node in log");
940 			goto out_dump;
941 		}
942 	}
943 
944 	if (sleb->endpt || c->lhead_offs >= c->leb_size) {
945 		c->lhead_lnum = lnum;
946 		c->lhead_offs = sleb->endpt;
947 	}
948 
949 	err = !sleb->endpt;
950 out:
951 	ubifs_scan_destroy(sleb);
952 	return err;
953 
954 out_dump:
955 	ubifs_err("log error detected while replying the log at LEB %d:%d",
956 		  lnum, offs + snod->offs);
957 	dbg_dump_node(c, snod->node);
958 	ubifs_scan_destroy(sleb);
959 	return -EINVAL;
960 }
961 
962 /**
963  * take_ihead - update the status of the index head in lprops to 'taken'.
964  * @c: UBIFS file-system description object
965  *
966  * This function returns the amount of free space in the index head LEB or a
967  * negative error code.
968  */
969 static int take_ihead(struct ubifs_info *c)
970 {
971 	const struct ubifs_lprops *lp;
972 	int err, free;
973 
974 	ubifs_get_lprops(c);
975 
976 	lp = ubifs_lpt_lookup_dirty(c, c->ihead_lnum);
977 	if (IS_ERR(lp)) {
978 		err = PTR_ERR(lp);
979 		goto out;
980 	}
981 
982 	free = lp->free;
983 
984 	lp = ubifs_change_lp(c, lp, LPROPS_NC, LPROPS_NC,
985 			     lp->flags | LPROPS_TAKEN, 0);
986 	if (IS_ERR(lp)) {
987 		err = PTR_ERR(lp);
988 		goto out;
989 	}
990 
991 	err = free;
992 out:
993 	ubifs_release_lprops(c);
994 	return err;
995 }
996 
997 /**
998  * ubifs_replay_journal - replay journal.
999  * @c: UBIFS file-system description object
1000  *
1001  * This function scans the journal, replays and cleans it up. It makes sure all
1002  * memory data structures related to uncommitted journal are built (dirty TNC
1003  * tree, tree of buds, modified lprops, etc).
1004  */
1005 int ubifs_replay_journal(struct ubifs_info *c)
1006 {
1007 	int err, i, lnum, offs, _free;
1008 	void *sbuf = NULL;
1009 
1010 	BUILD_BUG_ON(UBIFS_TRUN_KEY > 5);
1011 
1012 	/* Update the status of the index head in lprops to 'taken' */
1013 	_free = take_ihead(c);
1014 	if (_free < 0)
1015 		return _free; /* Error code */
1016 
1017 	if (c->ihead_offs != c->leb_size - _free) {
1018 		ubifs_err("bad index head LEB %d:%d", c->ihead_lnum,
1019 			  c->ihead_offs);
1020 		return -EINVAL;
1021 	}
1022 
1023 	sbuf = vmalloc(c->leb_size);
1024 	if (!sbuf)
1025 		return -ENOMEM;
1026 
1027 	dbg_mnt("start replaying the journal");
1028 
1029 	c->replaying = 1;
1030 
1031 	lnum = c->ltail_lnum = c->lhead_lnum;
1032 	offs = c->lhead_offs;
1033 
1034 	for (i = 0; i < c->log_lebs; i++, lnum++) {
1035 		if (lnum >= UBIFS_LOG_LNUM + c->log_lebs) {
1036 			/*
1037 			 * The log is logically circular, we reached the last
1038 			 * LEB, switch to the first one.
1039 			 */
1040 			lnum = UBIFS_LOG_LNUM;
1041 			offs = 0;
1042 		}
1043 		err = replay_log_leb(c, lnum, offs, sbuf);
1044 		if (err == 1)
1045 			/* We hit the end of the log */
1046 			break;
1047 		if (err)
1048 			goto out;
1049 		offs = 0;
1050 	}
1051 
1052 	err = replay_buds(c);
1053 	if (err)
1054 		goto out;
1055 
1056 	err = apply_replay_tree(c);
1057 	if (err)
1058 		goto out;
1059 
1060 	ubifs_assert(c->bud_bytes <= c->max_bud_bytes || c->need_recovery);
1061 	dbg_mnt("finished, log head LEB %d:%d, max_sqnum %llu, "
1062 		"highest_inum %lu", c->lhead_lnum, c->lhead_offs, c->max_sqnum,
1063 		(unsigned long)c->highest_inum);
1064 out:
1065 	destroy_replay_tree(c);
1066 	destroy_bud_list(c);
1067 	vfree(sbuf);
1068 	c->replaying = 0;
1069 	return err;
1070 }
1071