attach.c revision 55393ba1bdedc5ded79b34b4cc08898a7776cddb
1/*
2 * Copyright (c) International Business Machines Corp., 2006
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 *
18 * Author: Artem Bityutskiy (Битюцкий Артём)
19 */
20
21/*
22 * UBI attaching sub-system.
23 *
24 * This sub-system is responsible for attaching MTD devices and it also
25 * implements flash media scanning.
26 *
27 * The attaching information is represented by a &struct ubi_attach_info'
28 * object. Information about volumes is represented by &struct ubi_ainf_volume
29 * objects which are kept in volume RB-tree with root at the @volumes field.
30 * The RB-tree is indexed by the volume ID.
31 *
32 * Logical eraseblocks are represented by &struct ubi_ainf_peb objects. These
33 * objects are kept in per-volume RB-trees with the root at the corresponding
34 * &struct ubi_ainf_volume object. To put it differently, we keep an RB-tree of
35 * per-volume objects and each of these objects is the root of RB-tree of
36 * per-LEB objects.
37 *
38 * Corrupted physical eraseblocks are put to the @corr list, free physical
39 * eraseblocks are put to the @free list and the physical eraseblock to be
40 * erased are put to the @erase list.
41 *
42 * About corruptions
43 * ~~~~~~~~~~~~~~~~~
44 *
45 * UBI protects EC and VID headers with CRC-32 checksums, so it can detect
46 * whether the headers are corrupted or not. Sometimes UBI also protects the
47 * data with CRC-32, e.g., when it executes the atomic LEB change operation, or
48 * when it moves the contents of a PEB for wear-leveling purposes.
49 *
50 * UBI tries to distinguish between 2 types of corruptions.
51 *
52 * 1. Corruptions caused by power cuts. These are expected corruptions and UBI
53 * tries to handle them gracefully, without printing too many warnings and
54 * error messages. The idea is that we do not lose important data in these
55 * cases - we may lose only the data which were being written to the media just
56 * before the power cut happened, and the upper layers (e.g., UBIFS) are
57 * supposed to handle such data losses (e.g., by using the FS journal).
58 *
59 * When UBI detects a corruption (CRC-32 mismatch) in a PEB, and it looks like
60 * the reason is a power cut, UBI puts this PEB to the @erase list, and all
61 * PEBs in the @erase list are scheduled for erasure later.
62 *
63 * 2. Unexpected corruptions which are not caused by power cuts. During
64 * attaching, such PEBs are put to the @corr list and UBI preserves them.
65 * Obviously, this lessens the amount of available PEBs, and if at some  point
66 * UBI runs out of free PEBs, it switches to R/O mode. UBI also loudly informs
67 * about such PEBs every time the MTD device is attached.
68 *
69 * However, it is difficult to reliably distinguish between these types of
70 * corruptions and UBI's strategy is as follows (in case of attaching by
71 * scanning). UBI assumes corruption type 2 if the VID header is corrupted and
72 * the data area does not contain all 0xFFs, and there were no bit-flips or
73 * integrity errors (e.g., ECC errors in case of NAND) while reading the data
74 * area.  Otherwise UBI assumes corruption type 1. So the decision criteria
75 * are as follows.
76 *   o If the data area contains only 0xFFs, there are no data, and it is safe
77 *     to just erase this PEB - this is corruption type 1.
78 *   o If the data area has bit-flips or data integrity errors (ECC errors on
79 *     NAND), it is probably a PEB which was being erased when power cut
80 *     happened, so this is corruption type 1. However, this is just a guess,
81 *     which might be wrong.
82 *   o Otherwise this is corruption type 2.
83 */
84
85#include <linux/err.h>
86#include <linux/slab.h>
87#include <linux/crc32.h>
88#include <linux/math64.h>
89#include <linux/random.h>
90#include "ubi.h"
91
92static int self_check_ai(struct ubi_device *ubi, struct ubi_attach_info *ai);
93
94/* Temporary variables used during scanning */
95static struct ubi_ec_hdr *ech;
96static struct ubi_vid_hdr *vidh;
97
98/**
99 * add_to_list - add physical eraseblock to a list.
100 * @ai: attaching information
101 * @pnum: physical eraseblock number to add
102 * @vol_id: the last used volume id for the PEB
103 * @lnum: the last used LEB number for the PEB
104 * @ec: erase counter of the physical eraseblock
105 * @to_head: if not zero, add to the head of the list
106 * @list: the list to add to
107 *
108 * This function allocates a 'struct ubi_ainf_peb' object for physical
109 * eraseblock @pnum and adds it to the "free", "erase", or "alien" lists.
110 * It stores the @lnum and @vol_id alongside, which can both be
111 * %UBI_UNKNOWN if they are not available, not readable, or not assigned.
112 * If @to_head is not zero, PEB will be added to the head of the list, which
113 * basically means it will be processed first later. E.g., we add corrupted
114 * PEBs (corrupted due to power cuts) to the head of the erase list to make
115 * sure we erase them first and get rid of corruptions ASAP. This function
116 * returns zero in case of success and a negative error code in case of
117 * failure.
118 */
119static int add_to_list(struct ubi_attach_info *ai, int pnum, int vol_id,
120		       int lnum, int ec, int to_head, struct list_head *list)
121{
122	struct ubi_ainf_peb *aeb;
123
124	if (list == &ai->free) {
125		dbg_bld("add to free: PEB %d, EC %d", pnum, ec);
126	} else if (list == &ai->erase) {
127		dbg_bld("add to erase: PEB %d, EC %d", pnum, ec);
128	} else if (list == &ai->alien) {
129		dbg_bld("add to alien: PEB %d, EC %d", pnum, ec);
130		ai->alien_peb_count += 1;
131	} else
132		BUG();
133
134	aeb = kmem_cache_alloc(ai->aeb_slab_cache, GFP_KERNEL);
135	if (!aeb)
136		return -ENOMEM;
137
138	aeb->pnum = pnum;
139	aeb->vol_id = vol_id;
140	aeb->lnum = lnum;
141	aeb->ec = ec;
142	if (to_head)
143		list_add(&aeb->u.list, list);
144	else
145		list_add_tail(&aeb->u.list, list);
146	return 0;
147}
148
149/**
150 * add_corrupted - add a corrupted physical eraseblock.
151 * @ai: attaching information
152 * @pnum: physical eraseblock number to add
153 * @ec: erase counter of the physical eraseblock
154 *
155 * This function allocates a 'struct ubi_ainf_peb' object for a corrupted
156 * physical eraseblock @pnum and adds it to the 'corr' list.  The corruption
157 * was presumably not caused by a power cut. Returns zero in case of success
158 * and a negative error code in case of failure.
159 */
160static int add_corrupted(struct ubi_attach_info *ai, int pnum, int ec)
161{
162	struct ubi_ainf_peb *aeb;
163
164	dbg_bld("add to corrupted: PEB %d, EC %d", pnum, ec);
165
166	aeb = kmem_cache_alloc(ai->aeb_slab_cache, GFP_KERNEL);
167	if (!aeb)
168		return -ENOMEM;
169
170	ai->corr_peb_count += 1;
171	aeb->pnum = pnum;
172	aeb->ec = ec;
173	list_add(&aeb->u.list, &ai->corr);
174	return 0;
175}
176
177/**
178 * validate_vid_hdr - check volume identifier header.
179 * @vid_hdr: the volume identifier header to check
180 * @av: information about the volume this logical eraseblock belongs to
181 * @pnum: physical eraseblock number the VID header came from
182 *
183 * This function checks that data stored in @vid_hdr is consistent. Returns
184 * non-zero if an inconsistency was found and zero if not.
185 *
186 * Note, UBI does sanity check of everything it reads from the flash media.
187 * Most of the checks are done in the I/O sub-system. Here we check that the
188 * information in the VID header is consistent to the information in other VID
189 * headers of the same volume.
190 */
191static int validate_vid_hdr(const struct ubi_vid_hdr *vid_hdr,
192			    const struct ubi_ainf_volume *av, int pnum)
193{
194	int vol_type = vid_hdr->vol_type;
195	int vol_id = be32_to_cpu(vid_hdr->vol_id);
196	int used_ebs = be32_to_cpu(vid_hdr->used_ebs);
197	int data_pad = be32_to_cpu(vid_hdr->data_pad);
198
199	if (av->leb_count != 0) {
200		int av_vol_type;
201
202		/*
203		 * This is not the first logical eraseblock belonging to this
204		 * volume. Ensure that the data in its VID header is consistent
205		 * to the data in previous logical eraseblock headers.
206		 */
207
208		if (vol_id != av->vol_id) {
209			ubi_err("inconsistent vol_id");
210			goto bad;
211		}
212
213		if (av->vol_type == UBI_STATIC_VOLUME)
214			av_vol_type = UBI_VID_STATIC;
215		else
216			av_vol_type = UBI_VID_DYNAMIC;
217
218		if (vol_type != av_vol_type) {
219			ubi_err("inconsistent vol_type");
220			goto bad;
221		}
222
223		if (used_ebs != av->used_ebs) {
224			ubi_err("inconsistent used_ebs");
225			goto bad;
226		}
227
228		if (data_pad != av->data_pad) {
229			ubi_err("inconsistent data_pad");
230			goto bad;
231		}
232	}
233
234	return 0;
235
236bad:
237	ubi_err("inconsistent VID header at PEB %d", pnum);
238	ubi_dump_vid_hdr(vid_hdr);
239	ubi_dump_av(av);
240	return -EINVAL;
241}
242
243/**
244 * add_volume - add volume to the attaching information.
245 * @ai: attaching information
246 * @vol_id: ID of the volume to add
247 * @pnum: physical eraseblock number
248 * @vid_hdr: volume identifier header
249 *
250 * If the volume corresponding to the @vid_hdr logical eraseblock is already
251 * present in the attaching information, this function does nothing. Otherwise
252 * it adds corresponding volume to the attaching information. Returns a pointer
253 * to the allocated "av" object in case of success and a negative error code in
254 * case of failure.
255 */
256static struct ubi_ainf_volume *add_volume(struct ubi_attach_info *ai,
257					  int vol_id, int pnum,
258					  const struct ubi_vid_hdr *vid_hdr)
259{
260	struct ubi_ainf_volume *av;
261	struct rb_node **p = &ai->volumes.rb_node, *parent = NULL;
262
263	ubi_assert(vol_id == be32_to_cpu(vid_hdr->vol_id));
264
265	/* Walk the volume RB-tree to look if this volume is already present */
266	while (*p) {
267		parent = *p;
268		av = rb_entry(parent, struct ubi_ainf_volume, rb);
269
270		if (vol_id == av->vol_id)
271			return av;
272
273		if (vol_id > av->vol_id)
274			p = &(*p)->rb_left;
275		else
276			p = &(*p)->rb_right;
277	}
278
279	/* The volume is absent - add it */
280	av = kmalloc(sizeof(struct ubi_ainf_volume), GFP_KERNEL);
281	if (!av)
282		return ERR_PTR(-ENOMEM);
283
284	av->highest_lnum = av->leb_count = 0;
285	av->vol_id = vol_id;
286	av->root = RB_ROOT;
287	av->used_ebs = be32_to_cpu(vid_hdr->used_ebs);
288	av->data_pad = be32_to_cpu(vid_hdr->data_pad);
289	av->compat = vid_hdr->compat;
290	av->vol_type = vid_hdr->vol_type == UBI_VID_DYNAMIC ? UBI_DYNAMIC_VOLUME
291							    : UBI_STATIC_VOLUME;
292	if (vol_id > ai->highest_vol_id)
293		ai->highest_vol_id = vol_id;
294
295	rb_link_node(&av->rb, parent, p);
296	rb_insert_color(&av->rb, &ai->volumes);
297	ai->vols_found += 1;
298	dbg_bld("added volume %d", vol_id);
299	return av;
300}
301
302/**
303 * compare_lebs - find out which logical eraseblock is newer.
304 * @ubi: UBI device description object
305 * @aeb: first logical eraseblock to compare
306 * @pnum: physical eraseblock number of the second logical eraseblock to
307 * compare
308 * @vid_hdr: volume identifier header of the second logical eraseblock
309 *
310 * This function compares 2 copies of a LEB and informs which one is newer. In
311 * case of success this function returns a positive value, in case of failure, a
312 * negative error code is returned. The success return codes use the following
313 * bits:
314 *     o bit 0 is cleared: the first PEB (described by @aeb) is newer than the
315 *       second PEB (described by @pnum and @vid_hdr);
316 *     o bit 0 is set: the second PEB is newer;
317 *     o bit 1 is cleared: no bit-flips were detected in the newer LEB;
318 *     o bit 1 is set: bit-flips were detected in the newer LEB;
319 *     o bit 2 is cleared: the older LEB is not corrupted;
320 *     o bit 2 is set: the older LEB is corrupted.
321 */
322static int compare_lebs(struct ubi_device *ubi, const struct ubi_ainf_peb *aeb,
323			int pnum, const struct ubi_vid_hdr *vid_hdr)
324{
325	void *buf;
326	int len, err, second_is_newer, bitflips = 0, corrupted = 0;
327	uint32_t data_crc, crc;
328	struct ubi_vid_hdr *vh = NULL;
329	unsigned long long sqnum2 = be64_to_cpu(vid_hdr->sqnum);
330
331	if (sqnum2 == aeb->sqnum) {
332		/*
333		 * This must be a really ancient UBI image which has been
334		 * created before sequence numbers support has been added. At
335		 * that times we used 32-bit LEB versions stored in logical
336		 * eraseblocks. That was before UBI got into mainline. We do not
337		 * support these images anymore. Well, those images still work,
338		 * but only if no unclean reboots happened.
339		 */
340		ubi_err("unsupported on-flash UBI format\n");
341		return -EINVAL;
342	}
343
344	/* Obviously the LEB with lower sequence counter is older */
345	second_is_newer = (sqnum2 > aeb->sqnum);
346
347	/*
348	 * Now we know which copy is newer. If the copy flag of the PEB with
349	 * newer version is not set, then we just return, otherwise we have to
350	 * check data CRC. For the second PEB we already have the VID header,
351	 * for the first one - we'll need to re-read it from flash.
352	 *
353	 * Note: this may be optimized so that we wouldn't read twice.
354	 */
355
356	if (second_is_newer) {
357		if (!vid_hdr->copy_flag) {
358			/* It is not a copy, so it is newer */
359			dbg_bld("second PEB %d is newer, copy_flag is unset",
360				pnum);
361			return 1;
362		}
363	} else {
364		if (!aeb->copy_flag) {
365			/* It is not a copy, so it is newer */
366			dbg_bld("first PEB %d is newer, copy_flag is unset",
367				pnum);
368			return bitflips << 1;
369		}
370
371		vh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
372		if (!vh)
373			return -ENOMEM;
374
375		pnum = aeb->pnum;
376		err = ubi_io_read_vid_hdr(ubi, pnum, vh, 0);
377		if (err) {
378			if (err == UBI_IO_BITFLIPS)
379				bitflips = 1;
380			else {
381				ubi_err("VID of PEB %d header is bad, but it was OK earlier, err %d",
382					pnum, err);
383				if (err > 0)
384					err = -EIO;
385
386				goto out_free_vidh;
387			}
388		}
389
390		vid_hdr = vh;
391	}
392
393	/* Read the data of the copy and check the CRC */
394
395	len = be32_to_cpu(vid_hdr->data_size);
396	buf = vmalloc(len);
397	if (!buf) {
398		err = -ENOMEM;
399		goto out_free_vidh;
400	}
401
402	err = ubi_io_read_data(ubi, buf, pnum, 0, len);
403	if (err && err != UBI_IO_BITFLIPS && !mtd_is_eccerr(err))
404		goto out_free_buf;
405
406	data_crc = be32_to_cpu(vid_hdr->data_crc);
407	crc = crc32(UBI_CRC32_INIT, buf, len);
408	if (crc != data_crc) {
409		dbg_bld("PEB %d CRC error: calculated %#08x, must be %#08x",
410			pnum, crc, data_crc);
411		corrupted = 1;
412		bitflips = 0;
413		second_is_newer = !second_is_newer;
414	} else {
415		dbg_bld("PEB %d CRC is OK", pnum);
416		bitflips = !!err;
417	}
418
419	vfree(buf);
420	ubi_free_vid_hdr(ubi, vh);
421
422	if (second_is_newer)
423		dbg_bld("second PEB %d is newer, copy_flag is set", pnum);
424	else
425		dbg_bld("first PEB %d is newer, copy_flag is set", pnum);
426
427	return second_is_newer | (bitflips << 1) | (corrupted << 2);
428
429out_free_buf:
430	vfree(buf);
431out_free_vidh:
432	ubi_free_vid_hdr(ubi, vh);
433	return err;
434}
435
436/**
437 * ubi_add_to_av - add used physical eraseblock to the attaching information.
438 * @ubi: UBI device description object
439 * @ai: attaching information
440 * @pnum: the physical eraseblock number
441 * @ec: erase counter
442 * @vid_hdr: the volume identifier header
443 * @bitflips: if bit-flips were detected when this physical eraseblock was read
444 *
445 * This function adds information about a used physical eraseblock to the
446 * 'used' tree of the corresponding volume. The function is rather complex
447 * because it has to handle cases when this is not the first physical
448 * eraseblock belonging to the same logical eraseblock, and the newer one has
449 * to be picked, while the older one has to be dropped. This function returns
450 * zero in case of success and a negative error code in case of failure.
451 */
452int ubi_add_to_av(struct ubi_device *ubi, struct ubi_attach_info *ai, int pnum,
453		  int ec, const struct ubi_vid_hdr *vid_hdr, int bitflips)
454{
455	int err, vol_id, lnum;
456	unsigned long long sqnum;
457	struct ubi_ainf_volume *av;
458	struct ubi_ainf_peb *aeb;
459	struct rb_node **p, *parent = NULL;
460
461	vol_id = be32_to_cpu(vid_hdr->vol_id);
462	lnum = be32_to_cpu(vid_hdr->lnum);
463	sqnum = be64_to_cpu(vid_hdr->sqnum);
464
465	dbg_bld("PEB %d, LEB %d:%d, EC %d, sqnum %llu, bitflips %d",
466		pnum, vol_id, lnum, ec, sqnum, bitflips);
467
468	av = add_volume(ai, vol_id, pnum, vid_hdr);
469	if (IS_ERR(av))
470		return PTR_ERR(av);
471
472	if (ai->max_sqnum < sqnum)
473		ai->max_sqnum = sqnum;
474
475	/*
476	 * Walk the RB-tree of logical eraseblocks of volume @vol_id to look
477	 * if this is the first instance of this logical eraseblock or not.
478	 */
479	p = &av->root.rb_node;
480	while (*p) {
481		int cmp_res;
482
483		parent = *p;
484		aeb = rb_entry(parent, struct ubi_ainf_peb, u.rb);
485		if (lnum != aeb->lnum) {
486			if (lnum < aeb->lnum)
487				p = &(*p)->rb_left;
488			else
489				p = &(*p)->rb_right;
490			continue;
491		}
492
493		/*
494		 * There is already a physical eraseblock describing the same
495		 * logical eraseblock present.
496		 */
497
498		dbg_bld("this LEB already exists: PEB %d, sqnum %llu, EC %d",
499			aeb->pnum, aeb->sqnum, aeb->ec);
500
501		/*
502		 * Make sure that the logical eraseblocks have different
503		 * sequence numbers. Otherwise the image is bad.
504		 *
505		 * However, if the sequence number is zero, we assume it must
506		 * be an ancient UBI image from the era when UBI did not have
507		 * sequence numbers. We still can attach these images, unless
508		 * there is a need to distinguish between old and new
509		 * eraseblocks, in which case we'll refuse the image in
510		 * 'compare_lebs()'. In other words, we attach old clean
511		 * images, but refuse attaching old images with duplicated
512		 * logical eraseblocks because there was an unclean reboot.
513		 */
514		if (aeb->sqnum == sqnum && sqnum != 0) {
515			ubi_err("two LEBs with same sequence number %llu",
516				sqnum);
517			ubi_dump_aeb(aeb, 0);
518			ubi_dump_vid_hdr(vid_hdr);
519			return -EINVAL;
520		}
521
522		/*
523		 * Now we have to drop the older one and preserve the newer
524		 * one.
525		 */
526		cmp_res = compare_lebs(ubi, aeb, pnum, vid_hdr);
527		if (cmp_res < 0)
528			return cmp_res;
529
530		if (cmp_res & 1) {
531			/*
532			 * This logical eraseblock is newer than the one
533			 * found earlier.
534			 */
535			err = validate_vid_hdr(vid_hdr, av, pnum);
536			if (err)
537				return err;
538
539			err = add_to_list(ai, aeb->pnum, aeb->vol_id,
540					  aeb->lnum, aeb->ec, cmp_res & 4,
541					  &ai->erase);
542			if (err)
543				return err;
544
545			aeb->ec = ec;
546			aeb->pnum = pnum;
547			aeb->vol_id = vol_id;
548			aeb->lnum = lnum;
549			aeb->scrub = ((cmp_res & 2) || bitflips);
550			aeb->copy_flag = vid_hdr->copy_flag;
551			aeb->sqnum = sqnum;
552
553			if (av->highest_lnum == lnum)
554				av->last_data_size =
555					be32_to_cpu(vid_hdr->data_size);
556
557			return 0;
558		} else {
559			/*
560			 * This logical eraseblock is older than the one found
561			 * previously.
562			 */
563			return add_to_list(ai, pnum, vol_id, lnum, ec,
564					   cmp_res & 4, &ai->erase);
565		}
566	}
567
568	/*
569	 * We've met this logical eraseblock for the first time, add it to the
570	 * attaching information.
571	 */
572
573	err = validate_vid_hdr(vid_hdr, av, pnum);
574	if (err)
575		return err;
576
577	aeb = kmem_cache_alloc(ai->aeb_slab_cache, GFP_KERNEL);
578	if (!aeb)
579		return -ENOMEM;
580
581	aeb->ec = ec;
582	aeb->pnum = pnum;
583	aeb->vol_id = vol_id;
584	aeb->lnum = lnum;
585	aeb->scrub = bitflips;
586	aeb->copy_flag = vid_hdr->copy_flag;
587	aeb->sqnum = sqnum;
588
589	if (av->highest_lnum <= lnum) {
590		av->highest_lnum = lnum;
591		av->last_data_size = be32_to_cpu(vid_hdr->data_size);
592	}
593
594	av->leb_count += 1;
595	rb_link_node(&aeb->u.rb, parent, p);
596	rb_insert_color(&aeb->u.rb, &av->root);
597	return 0;
598}
599
600/**
601 * ubi_find_av - find volume in the attaching information.
602 * @ai: attaching information
603 * @vol_id: the requested volume ID
604 *
605 * This function returns a pointer to the volume description or %NULL if there
606 * are no data about this volume in the attaching information.
607 */
608struct ubi_ainf_volume *ubi_find_av(const struct ubi_attach_info *ai,
609				    int vol_id)
610{
611	struct ubi_ainf_volume *av;
612	struct rb_node *p = ai->volumes.rb_node;
613
614	while (p) {
615		av = rb_entry(p, struct ubi_ainf_volume, rb);
616
617		if (vol_id == av->vol_id)
618			return av;
619
620		if (vol_id > av->vol_id)
621			p = p->rb_left;
622		else
623			p = p->rb_right;
624	}
625
626	return NULL;
627}
628
629/**
630 * ubi_remove_av - delete attaching information about a volume.
631 * @ai: attaching information
632 * @av: the volume attaching information to delete
633 */
634void ubi_remove_av(struct ubi_attach_info *ai, struct ubi_ainf_volume *av)
635{
636	struct rb_node *rb;
637	struct ubi_ainf_peb *aeb;
638
639	dbg_bld("remove attaching information about volume %d", av->vol_id);
640
641	while ((rb = rb_first(&av->root))) {
642		aeb = rb_entry(rb, struct ubi_ainf_peb, u.rb);
643		rb_erase(&aeb->u.rb, &av->root);
644		list_add_tail(&aeb->u.list, &ai->erase);
645	}
646
647	rb_erase(&av->rb, &ai->volumes);
648	kfree(av);
649	ai->vols_found -= 1;
650}
651
652/**
653 * early_erase_peb - erase a physical eraseblock.
654 * @ubi: UBI device description object
655 * @ai: attaching information
656 * @pnum: physical eraseblock number to erase;
657 * @ec: erase counter value to write (%UBI_UNKNOWN if it is unknown)
658 *
659 * This function erases physical eraseblock 'pnum', and writes the erase
660 * counter header to it. This function should only be used on UBI device
661 * initialization stages, when the EBA sub-system had not been yet initialized.
662 * This function returns zero in case of success and a negative error code in
663 * case of failure.
664 */
665static int early_erase_peb(struct ubi_device *ubi,
666			   const struct ubi_attach_info *ai, int pnum, int ec)
667{
668	int err;
669	struct ubi_ec_hdr *ec_hdr;
670
671	if ((long long)ec >= UBI_MAX_ERASECOUNTER) {
672		/*
673		 * Erase counter overflow. Upgrade UBI and use 64-bit
674		 * erase counters internally.
675		 */
676		ubi_err("erase counter overflow at PEB %d, EC %d", pnum, ec);
677		return -EINVAL;
678	}
679
680	ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
681	if (!ec_hdr)
682		return -ENOMEM;
683
684	ec_hdr->ec = cpu_to_be64(ec);
685
686	err = ubi_io_sync_erase(ubi, pnum, 0);
687	if (err < 0)
688		goto out_free;
689
690	err = ubi_io_write_ec_hdr(ubi, pnum, ec_hdr);
691
692out_free:
693	kfree(ec_hdr);
694	return err;
695}
696
697/**
698 * ubi_early_get_peb - get a free physical eraseblock.
699 * @ubi: UBI device description object
700 * @ai: attaching information
701 *
702 * This function returns a free physical eraseblock. It is supposed to be
703 * called on the UBI initialization stages when the wear-leveling sub-system is
704 * not initialized yet. This function picks a physical eraseblocks from one of
705 * the lists, writes the EC header if it is needed, and removes it from the
706 * list.
707 *
708 * This function returns a pointer to the "aeb" of the found free PEB in case
709 * of success and an error code in case of failure.
710 */
711struct ubi_ainf_peb *ubi_early_get_peb(struct ubi_device *ubi,
712				       struct ubi_attach_info *ai)
713{
714	int err = 0;
715	struct ubi_ainf_peb *aeb, *tmp_aeb;
716
717	if (!list_empty(&ai->free)) {
718		aeb = list_entry(ai->free.next, struct ubi_ainf_peb, u.list);
719		list_del(&aeb->u.list);
720		dbg_bld("return free PEB %d, EC %d", aeb->pnum, aeb->ec);
721		return aeb;
722	}
723
724	/*
725	 * We try to erase the first physical eraseblock from the erase list
726	 * and pick it if we succeed, or try to erase the next one if not. And
727	 * so forth. We don't want to take care about bad eraseblocks here -
728	 * they'll be handled later.
729	 */
730	list_for_each_entry_safe(aeb, tmp_aeb, &ai->erase, u.list) {
731		if (aeb->ec == UBI_UNKNOWN)
732			aeb->ec = ai->mean_ec;
733
734		err = early_erase_peb(ubi, ai, aeb->pnum, aeb->ec+1);
735		if (err)
736			continue;
737
738		aeb->ec += 1;
739		list_del(&aeb->u.list);
740		dbg_bld("return PEB %d, EC %d", aeb->pnum, aeb->ec);
741		return aeb;
742	}
743
744	ubi_err("no free eraseblocks");
745	return ERR_PTR(-ENOSPC);
746}
747
748/**
749 * check_corruption - check the data area of PEB.
750 * @ubi: UBI device description object
751 * @vid_hrd: the (corrupted) VID header of this PEB
752 * @pnum: the physical eraseblock number to check
753 *
754 * This is a helper function which is used to distinguish between VID header
755 * corruptions caused by power cuts and other reasons. If the PEB contains only
756 * 0xFF bytes in the data area, the VID header is most probably corrupted
757 * because of a power cut (%0 is returned in this case). Otherwise, it was
758 * probably corrupted for some other reasons (%1 is returned in this case). A
759 * negative error code is returned if a read error occurred.
760 *
761 * If the corruption reason was a power cut, UBI can safely erase this PEB.
762 * Otherwise, it should preserve it to avoid possibly destroying important
763 * information.
764 */
765static int check_corruption(struct ubi_device *ubi, struct ubi_vid_hdr *vid_hdr,
766			    int pnum)
767{
768	int err;
769
770	mutex_lock(&ubi->buf_mutex);
771	memset(ubi->peb_buf, 0x00, ubi->leb_size);
772
773	err = ubi_io_read(ubi, ubi->peb_buf, pnum, ubi->leb_start,
774			  ubi->leb_size);
775	if (err == UBI_IO_BITFLIPS || mtd_is_eccerr(err)) {
776		/*
777		 * Bit-flips or integrity errors while reading the data area.
778		 * It is difficult to say for sure what type of corruption is
779		 * this, but presumably a power cut happened while this PEB was
780		 * erased, so it became unstable and corrupted, and should be
781		 * erased.
782		 */
783		err = 0;
784		goto out_unlock;
785	}
786
787	if (err)
788		goto out_unlock;
789
790	if (ubi_check_pattern(ubi->peb_buf, 0xFF, ubi->leb_size))
791		goto out_unlock;
792
793	ubi_err("PEB %d contains corrupted VID header, and the data does not contain all 0xFF",
794		pnum);
795	ubi_err("this may be a non-UBI PEB or a severe VID header corruption which requires manual inspection");
796	ubi_dump_vid_hdr(vid_hdr);
797	pr_err("hexdump of PEB %d offset %d, length %d",
798	       pnum, ubi->leb_start, ubi->leb_size);
799	ubi_dbg_print_hex_dump(KERN_DEBUG, "", DUMP_PREFIX_OFFSET, 32, 1,
800			       ubi->peb_buf, ubi->leb_size, 1);
801	err = 1;
802
803out_unlock:
804	mutex_unlock(&ubi->buf_mutex);
805	return err;
806}
807
808/**
809 * scan_peb - scan and process UBI headers of a PEB.
810 * @ubi: UBI device description object
811 * @ai: attaching information
812 * @pnum: the physical eraseblock number
813 *
814 * This function reads UBI headers of PEB @pnum, checks them, and adds
815 * information about this PEB to the corresponding list or RB-tree in the
816 * "attaching info" structure. Returns zero if the physical eraseblock was
817 * successfully handled and a negative error code in case of failure.
818 */
819static int scan_peb(struct ubi_device *ubi, struct ubi_attach_info *ai,
820		    int pnum)
821{
822	long long uninitialized_var(ec);
823	int err, bitflips = 0, vol_id, ec_err = 0;
824
825	dbg_bld("scan PEB %d", pnum);
826
827	/* Skip bad physical eraseblocks */
828	err = ubi_io_is_bad(ubi, pnum);
829	if (err < 0)
830		return err;
831	else if (err) {
832		ai->bad_peb_count += 1;
833		return 0;
834	}
835
836	err = ubi_io_read_ec_hdr(ubi, pnum, ech, 0);
837	if (err < 0)
838		return err;
839	switch (err) {
840	case 0:
841		break;
842	case UBI_IO_BITFLIPS:
843		bitflips = 1;
844		break;
845	case UBI_IO_FF:
846		ai->empty_peb_count += 1;
847		return add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
848				   UBI_UNKNOWN, 0, &ai->erase);
849	case UBI_IO_FF_BITFLIPS:
850		ai->empty_peb_count += 1;
851		return add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
852				   UBI_UNKNOWN, 1, &ai->erase);
853	case UBI_IO_BAD_HDR_EBADMSG:
854	case UBI_IO_BAD_HDR:
855		/*
856		 * We have to also look at the VID header, possibly it is not
857		 * corrupted. Set %bitflips flag in order to make this PEB be
858		 * moved and EC be re-created.
859		 */
860		ec_err = err;
861		ec = UBI_UNKNOWN;
862		bitflips = 1;
863		break;
864	default:
865		ubi_err("'ubi_io_read_ec_hdr()' returned unknown code %d", err);
866		return -EINVAL;
867	}
868
869	if (!ec_err) {
870		int image_seq;
871
872		/* Make sure UBI version is OK */
873		if (ech->version != UBI_VERSION) {
874			ubi_err("this UBI version is %d, image version is %d",
875				UBI_VERSION, (int)ech->version);
876			return -EINVAL;
877		}
878
879		ec = be64_to_cpu(ech->ec);
880		if (ec > UBI_MAX_ERASECOUNTER) {
881			/*
882			 * Erase counter overflow. The EC headers have 64 bits
883			 * reserved, but we anyway make use of only 31 bit
884			 * values, as this seems to be enough for any existing
885			 * flash. Upgrade UBI and use 64-bit erase counters
886			 * internally.
887			 */
888			ubi_err("erase counter overflow, max is %d",
889				UBI_MAX_ERASECOUNTER);
890			ubi_dump_ec_hdr(ech);
891			return -EINVAL;
892		}
893
894		/*
895		 * Make sure that all PEBs have the same image sequence number.
896		 * This allows us to detect situations when users flash UBI
897		 * images incorrectly, so that the flash has the new UBI image
898		 * and leftovers from the old one. This feature was added
899		 * relatively recently, and the sequence number was always
900		 * zero, because old UBI implementations always set it to zero.
901		 * For this reasons, we do not panic if some PEBs have zero
902		 * sequence number, while other PEBs have non-zero sequence
903		 * number.
904		 */
905		image_seq = be32_to_cpu(ech->image_seq);
906		if (!ubi->image_seq && image_seq)
907			ubi->image_seq = image_seq;
908		if (ubi->image_seq && image_seq &&
909		    ubi->image_seq != image_seq) {
910			ubi_err("bad image sequence number %d in PEB %d, expected %d",
911				image_seq, pnum, ubi->image_seq);
912			ubi_dump_ec_hdr(ech);
913			return -EINVAL;
914		}
915	}
916
917	/* OK, we've done with the EC header, let's look at the VID header */
918
919	err = ubi_io_read_vid_hdr(ubi, pnum, vidh, 0);
920	if (err < 0)
921		return err;
922	switch (err) {
923	case 0:
924		break;
925	case UBI_IO_BITFLIPS:
926		bitflips = 1;
927		break;
928	case UBI_IO_BAD_HDR_EBADMSG:
929		if (ec_err == UBI_IO_BAD_HDR_EBADMSG)
930			/*
931			 * Both EC and VID headers are corrupted and were read
932			 * with data integrity error, probably this is a bad
933			 * PEB, bit it is not marked as bad yet. This may also
934			 * be a result of power cut during erasure.
935			 */
936			ai->maybe_bad_peb_count += 1;
937	case UBI_IO_BAD_HDR:
938		if (ec_err)
939			/*
940			 * Both headers are corrupted. There is a possibility
941			 * that this a valid UBI PEB which has corresponding
942			 * LEB, but the headers are corrupted. However, it is
943			 * impossible to distinguish it from a PEB which just
944			 * contains garbage because of a power cut during erase
945			 * operation. So we just schedule this PEB for erasure.
946			 *
947			 * Besides, in case of NOR flash, we deliberately
948			 * corrupt both headers because NOR flash erasure is
949			 * slow and can start from the end.
950			 */
951			err = 0;
952		else
953			/*
954			 * The EC was OK, but the VID header is corrupted. We
955			 * have to check what is in the data area.
956			 */
957			err = check_corruption(ubi, vidh, pnum);
958
959		if (err < 0)
960			return err;
961		else if (!err)
962			/* This corruption is caused by a power cut */
963			err = add_to_list(ai, pnum, UBI_UNKNOWN,
964					  UBI_UNKNOWN, ec, 1, &ai->erase);
965		else
966			/* This is an unexpected corruption */
967			err = add_corrupted(ai, pnum, ec);
968		if (err)
969			return err;
970		goto adjust_mean_ec;
971	case UBI_IO_FF_BITFLIPS:
972		err = add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
973				  ec, 1, &ai->erase);
974		if (err)
975			return err;
976		goto adjust_mean_ec;
977	case UBI_IO_FF:
978		if (ec_err || bitflips)
979			err = add_to_list(ai, pnum, UBI_UNKNOWN,
980					  UBI_UNKNOWN, ec, 1, &ai->erase);
981		else
982			err = add_to_list(ai, pnum, UBI_UNKNOWN,
983					  UBI_UNKNOWN, ec, 0, &ai->free);
984		if (err)
985			return err;
986		goto adjust_mean_ec;
987	default:
988		ubi_err("'ubi_io_read_vid_hdr()' returned unknown code %d",
989			err);
990		return -EINVAL;
991	}
992
993	vol_id = be32_to_cpu(vidh->vol_id);
994	if (vol_id > UBI_MAX_VOLUMES && vol_id != UBI_LAYOUT_VOLUME_ID) {
995		int lnum = be32_to_cpu(vidh->lnum);
996
997		/* Unsupported internal volume */
998		switch (vidh->compat) {
999		case UBI_COMPAT_DELETE:
1000			ubi_msg("\"delete\" compatible internal volume %d:%d found, will remove it",
1001				vol_id, lnum);
1002			err = add_to_list(ai, pnum, vol_id, lnum,
1003					  ec, 1, &ai->erase);
1004			if (err)
1005				return err;
1006			return 0;
1007
1008		case UBI_COMPAT_RO:
1009			ubi_msg("read-only compatible internal volume %d:%d found, switch to read-only mode",
1010				vol_id, lnum);
1011			ubi->ro_mode = 1;
1012			break;
1013
1014		case UBI_COMPAT_PRESERVE:
1015			ubi_msg("\"preserve\" compatible internal volume %d:%d found",
1016				vol_id, lnum);
1017			err = add_to_list(ai, pnum, vol_id, lnum,
1018					  ec, 0, &ai->alien);
1019			if (err)
1020				return err;
1021			return 0;
1022
1023		case UBI_COMPAT_REJECT:
1024			ubi_err("incompatible internal volume %d:%d found",
1025				vol_id, lnum);
1026			return -EINVAL;
1027		}
1028	}
1029
1030	if (ec_err)
1031		ubi_warn("valid VID header but corrupted EC header at PEB %d",
1032			 pnum);
1033	err = ubi_add_to_av(ubi, ai, pnum, ec, vidh, bitflips);
1034	if (err)
1035		return err;
1036
1037adjust_mean_ec:
1038	if (!ec_err) {
1039		ai->ec_sum += ec;
1040		ai->ec_count += 1;
1041		if (ec > ai->max_ec)
1042			ai->max_ec = ec;
1043		if (ec < ai->min_ec)
1044			ai->min_ec = ec;
1045	}
1046
1047	return 0;
1048}
1049
1050/**
1051 * late_analysis - analyze the overall situation with PEB.
1052 * @ubi: UBI device description object
1053 * @ai: attaching information
1054 *
1055 * This is a helper function which takes a look what PEBs we have after we
1056 * gather information about all of them ("ai" is compete). It decides whether
1057 * the flash is empty and should be formatted of whether there are too many
1058 * corrupted PEBs and we should not attach this MTD device. Returns zero if we
1059 * should proceed with attaching the MTD device, and %-EINVAL if we should not.
1060 */
1061static int late_analysis(struct ubi_device *ubi, struct ubi_attach_info *ai)
1062{
1063	struct ubi_ainf_peb *aeb;
1064	int max_corr, peb_count;
1065
1066	peb_count = ubi->peb_count - ai->bad_peb_count - ai->alien_peb_count;
1067	max_corr = peb_count / 20 ?: 8;
1068
1069	/*
1070	 * Few corrupted PEBs is not a problem and may be just a result of
1071	 * unclean reboots. However, many of them may indicate some problems
1072	 * with the flash HW or driver.
1073	 */
1074	if (ai->corr_peb_count) {
1075		ubi_err("%d PEBs are corrupted and preserved",
1076			ai->corr_peb_count);
1077		pr_err("Corrupted PEBs are:");
1078		list_for_each_entry(aeb, &ai->corr, u.list)
1079			pr_cont(" %d", aeb->pnum);
1080		pr_cont("\n");
1081
1082		/*
1083		 * If too many PEBs are corrupted, we refuse attaching,
1084		 * otherwise, only print a warning.
1085		 */
1086		if (ai->corr_peb_count >= max_corr) {
1087			ubi_err("too many corrupted PEBs, refusing");
1088			return -EINVAL;
1089		}
1090	}
1091
1092	if (ai->empty_peb_count + ai->maybe_bad_peb_count == peb_count) {
1093		/*
1094		 * All PEBs are empty, or almost all - a couple PEBs look like
1095		 * they may be bad PEBs which were not marked as bad yet.
1096		 *
1097		 * This piece of code basically tries to distinguish between
1098		 * the following situations:
1099		 *
1100		 * 1. Flash is empty, but there are few bad PEBs, which are not
1101		 *    marked as bad so far, and which were read with error. We
1102		 *    want to go ahead and format this flash. While formatting,
1103		 *    the faulty PEBs will probably be marked as bad.
1104		 *
1105		 * 2. Flash contains non-UBI data and we do not want to format
1106		 *    it and destroy possibly important information.
1107		 */
1108		if (ai->maybe_bad_peb_count <= 2) {
1109			ai->is_empty = 1;
1110			ubi_msg("empty MTD device detected");
1111			get_random_bytes(&ubi->image_seq,
1112					 sizeof(ubi->image_seq));
1113		} else {
1114			ubi_err("MTD device is not UBI-formatted and possibly contains non-UBI data - refusing it");
1115			return -EINVAL;
1116		}
1117
1118	}
1119
1120	return 0;
1121}
1122
1123/**
1124 * scan_all - scan entire MTD device.
1125 * @ubi: UBI device description object
1126 *
1127 * This function does full scanning of an MTD device and returns complete
1128 * information about it in form of a "struct ubi_attach_info" object. In case
1129 * of failure, an error code is returned.
1130 */
1131static struct ubi_attach_info *scan_all(struct ubi_device *ubi)
1132{
1133	int err, pnum;
1134	struct rb_node *rb1, *rb2;
1135	struct ubi_ainf_volume *av;
1136	struct ubi_ainf_peb *aeb;
1137	struct ubi_attach_info *ai;
1138
1139	ai = kzalloc(sizeof(struct ubi_attach_info), GFP_KERNEL);
1140	if (!ai)
1141		return ERR_PTR(-ENOMEM);
1142
1143	INIT_LIST_HEAD(&ai->corr);
1144	INIT_LIST_HEAD(&ai->free);
1145	INIT_LIST_HEAD(&ai->erase);
1146	INIT_LIST_HEAD(&ai->alien);
1147	ai->volumes = RB_ROOT;
1148
1149	err = -ENOMEM;
1150	ai->aeb_slab_cache = kmem_cache_create("ubi_aeb_slab_cache",
1151					       sizeof(struct ubi_ainf_peb),
1152					       0, 0, NULL);
1153	if (!ai->aeb_slab_cache)
1154		goto out_ai;
1155
1156	ech = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
1157	if (!ech)
1158		goto out_ai;
1159
1160	vidh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
1161	if (!vidh)
1162		goto out_ech;
1163
1164	for (pnum = 0; pnum < ubi->peb_count; pnum++) {
1165		cond_resched();
1166
1167		dbg_gen("process PEB %d", pnum);
1168		err = scan_peb(ubi, ai, pnum);
1169		if (err < 0)
1170			goto out_vidh;
1171	}
1172
1173	ubi_msg("scanning is finished");
1174
1175	/* Calculate mean erase counter */
1176	if (ai->ec_count)
1177		ai->mean_ec = div_u64(ai->ec_sum, ai->ec_count);
1178
1179	err = late_analysis(ubi, ai);
1180	if (err)
1181		goto out_vidh;
1182
1183	/*
1184	 * In case of unknown erase counter we use the mean erase counter
1185	 * value.
1186	 */
1187	ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1188		ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb)
1189			if (aeb->ec == UBI_UNKNOWN)
1190				aeb->ec = ai->mean_ec;
1191	}
1192
1193	list_for_each_entry(aeb, &ai->free, u.list) {
1194		if (aeb->ec == UBI_UNKNOWN)
1195			aeb->ec = ai->mean_ec;
1196	}
1197
1198	list_for_each_entry(aeb, &ai->corr, u.list)
1199		if (aeb->ec == UBI_UNKNOWN)
1200			aeb->ec = ai->mean_ec;
1201
1202	list_for_each_entry(aeb, &ai->erase, u.list)
1203		if (aeb->ec == UBI_UNKNOWN)
1204			aeb->ec = ai->mean_ec;
1205
1206	err = self_check_ai(ubi, ai);
1207	if (err)
1208		goto out_vidh;
1209
1210	ubi_free_vid_hdr(ubi, vidh);
1211	kfree(ech);
1212
1213	return ai;
1214
1215out_vidh:
1216	ubi_free_vid_hdr(ubi, vidh);
1217out_ech:
1218	kfree(ech);
1219out_ai:
1220	ubi_destroy_ai(ai);
1221	return ERR_PTR(err);
1222}
1223
1224/**
1225 * ubi_attach - attach an MTD device.
1226 * @ubi: UBI device descriptor
1227 *
1228 * This function returns zero in case of success and a negative error code in
1229 * case of failure.
1230 */
1231int ubi_attach(struct ubi_device *ubi)
1232{
1233	int err;
1234	struct ubi_attach_info *ai;
1235
1236	ai = scan_all(ubi);
1237	if (IS_ERR(ai))
1238		return PTR_ERR(ai);
1239
1240	ubi->bad_peb_count = ai->bad_peb_count;
1241	ubi->good_peb_count = ubi->peb_count - ubi->bad_peb_count;
1242	ubi->corr_peb_count = ai->corr_peb_count;
1243	ubi->max_ec = ai->max_ec;
1244	ubi->mean_ec = ai->mean_ec;
1245	dbg_gen("max. sequence number:       %llu", ai->max_sqnum);
1246
1247	err = ubi_read_volume_table(ubi, ai);
1248	if (err)
1249		goto out_ai;
1250
1251	err = ubi_wl_init(ubi, ai);
1252	if (err)
1253		goto out_vtbl;
1254
1255	err = ubi_eba_init(ubi, ai);
1256	if (err)
1257		goto out_wl;
1258
1259	ubi_destroy_ai(ai);
1260	return 0;
1261
1262out_wl:
1263	ubi_wl_close(ubi);
1264out_vtbl:
1265	ubi_free_internal_volumes(ubi);
1266	vfree(ubi->vtbl);
1267out_ai:
1268	ubi_destroy_ai(ai);
1269	return err;
1270}
1271
1272/**
1273 * destroy_av - free volume attaching information.
1274 * @av: volume attaching information
1275 * @ai: attaching information
1276 *
1277 * This function destroys the volume attaching information.
1278 */
1279static void destroy_av(struct ubi_attach_info *ai, struct ubi_ainf_volume *av)
1280{
1281	struct ubi_ainf_peb *aeb;
1282	struct rb_node *this = av->root.rb_node;
1283
1284	while (this) {
1285		if (this->rb_left)
1286			this = this->rb_left;
1287		else if (this->rb_right)
1288			this = this->rb_right;
1289		else {
1290			aeb = rb_entry(this, struct ubi_ainf_peb, u.rb);
1291			this = rb_parent(this);
1292			if (this) {
1293				if (this->rb_left == &aeb->u.rb)
1294					this->rb_left = NULL;
1295				else
1296					this->rb_right = NULL;
1297			}
1298
1299			kmem_cache_free(ai->aeb_slab_cache, aeb);
1300		}
1301	}
1302	kfree(av);
1303}
1304
1305/**
1306 * ubi_destroy_ai - destroy attaching information.
1307 * @ai: attaching information
1308 */
1309void ubi_destroy_ai(struct ubi_attach_info *ai)
1310{
1311	struct ubi_ainf_peb *aeb, *aeb_tmp;
1312	struct ubi_ainf_volume *av;
1313	struct rb_node *rb;
1314
1315	list_for_each_entry_safe(aeb, aeb_tmp, &ai->alien, u.list) {
1316		list_del(&aeb->u.list);
1317		kmem_cache_free(ai->aeb_slab_cache, aeb);
1318	}
1319	list_for_each_entry_safe(aeb, aeb_tmp, &ai->erase, u.list) {
1320		list_del(&aeb->u.list);
1321		kmem_cache_free(ai->aeb_slab_cache, aeb);
1322	}
1323	list_for_each_entry_safe(aeb, aeb_tmp, &ai->corr, u.list) {
1324		list_del(&aeb->u.list);
1325		kmem_cache_free(ai->aeb_slab_cache, aeb);
1326	}
1327	list_for_each_entry_safe(aeb, aeb_tmp, &ai->free, u.list) {
1328		list_del(&aeb->u.list);
1329		kmem_cache_free(ai->aeb_slab_cache, aeb);
1330	}
1331
1332	/* Destroy the volume RB-tree */
1333	rb = ai->volumes.rb_node;
1334	while (rb) {
1335		if (rb->rb_left)
1336			rb = rb->rb_left;
1337		else if (rb->rb_right)
1338			rb = rb->rb_right;
1339		else {
1340			av = rb_entry(rb, struct ubi_ainf_volume, rb);
1341
1342			rb = rb_parent(rb);
1343			if (rb) {
1344				if (rb->rb_left == &av->rb)
1345					rb->rb_left = NULL;
1346				else
1347					rb->rb_right = NULL;
1348			}
1349
1350			destroy_av(ai, av);
1351		}
1352	}
1353
1354	if (ai->aeb_slab_cache)
1355		kmem_cache_destroy(ai->aeb_slab_cache);
1356
1357	kfree(ai);
1358}
1359
1360/**
1361 * self_check_ai - check the attaching information.
1362 * @ubi: UBI device description object
1363 * @ai: attaching information
1364 *
1365 * This function returns zero if the attaching information is all right, and a
1366 * negative error code if not or if an error occurred.
1367 */
1368static int self_check_ai(struct ubi_device *ubi, struct ubi_attach_info *ai)
1369{
1370	int pnum, err, vols_found = 0;
1371	struct rb_node *rb1, *rb2;
1372	struct ubi_ainf_volume *av;
1373	struct ubi_ainf_peb *aeb, *last_aeb;
1374	uint8_t *buf;
1375
1376	if (!ubi->dbg->chk_gen)
1377		return 0;
1378
1379	/*
1380	 * At first, check that attaching information is OK.
1381	 */
1382	ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1383		int leb_count = 0;
1384
1385		cond_resched();
1386
1387		vols_found += 1;
1388
1389		if (ai->is_empty) {
1390			ubi_err("bad is_empty flag");
1391			goto bad_av;
1392		}
1393
1394		if (av->vol_id < 0 || av->highest_lnum < 0 ||
1395		    av->leb_count < 0 || av->vol_type < 0 || av->used_ebs < 0 ||
1396		    av->data_pad < 0 || av->last_data_size < 0) {
1397			ubi_err("negative values");
1398			goto bad_av;
1399		}
1400
1401		if (av->vol_id >= UBI_MAX_VOLUMES &&
1402		    av->vol_id < UBI_INTERNAL_VOL_START) {
1403			ubi_err("bad vol_id");
1404			goto bad_av;
1405		}
1406
1407		if (av->vol_id > ai->highest_vol_id) {
1408			ubi_err("highest_vol_id is %d, but vol_id %d is there",
1409				ai->highest_vol_id, av->vol_id);
1410			goto out;
1411		}
1412
1413		if (av->vol_type != UBI_DYNAMIC_VOLUME &&
1414		    av->vol_type != UBI_STATIC_VOLUME) {
1415			ubi_err("bad vol_type");
1416			goto bad_av;
1417		}
1418
1419		if (av->data_pad > ubi->leb_size / 2) {
1420			ubi_err("bad data_pad");
1421			goto bad_av;
1422		}
1423
1424		last_aeb = NULL;
1425		ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1426			cond_resched();
1427
1428			last_aeb = aeb;
1429			leb_count += 1;
1430
1431			if (aeb->pnum < 0 || aeb->ec < 0) {
1432				ubi_err("negative values");
1433				goto bad_aeb;
1434			}
1435
1436			if (aeb->ec < ai->min_ec) {
1437				ubi_err("bad ai->min_ec (%d), %d found",
1438					ai->min_ec, aeb->ec);
1439				goto bad_aeb;
1440			}
1441
1442			if (aeb->ec > ai->max_ec) {
1443				ubi_err("bad ai->max_ec (%d), %d found",
1444					ai->max_ec, aeb->ec);
1445				goto bad_aeb;
1446			}
1447
1448			if (aeb->pnum >= ubi->peb_count) {
1449				ubi_err("too high PEB number %d, total PEBs %d",
1450					aeb->pnum, ubi->peb_count);
1451				goto bad_aeb;
1452			}
1453
1454			if (av->vol_type == UBI_STATIC_VOLUME) {
1455				if (aeb->lnum >= av->used_ebs) {
1456					ubi_err("bad lnum or used_ebs");
1457					goto bad_aeb;
1458				}
1459			} else {
1460				if (av->used_ebs != 0) {
1461					ubi_err("non-zero used_ebs");
1462					goto bad_aeb;
1463				}
1464			}
1465
1466			if (aeb->lnum > av->highest_lnum) {
1467				ubi_err("incorrect highest_lnum or lnum");
1468				goto bad_aeb;
1469			}
1470		}
1471
1472		if (av->leb_count != leb_count) {
1473			ubi_err("bad leb_count, %d objects in the tree",
1474				leb_count);
1475			goto bad_av;
1476		}
1477
1478		if (!last_aeb)
1479			continue;
1480
1481		aeb = last_aeb;
1482
1483		if (aeb->lnum != av->highest_lnum) {
1484			ubi_err("bad highest_lnum");
1485			goto bad_aeb;
1486		}
1487	}
1488
1489	if (vols_found != ai->vols_found) {
1490		ubi_err("bad ai->vols_found %d, should be %d",
1491			ai->vols_found, vols_found);
1492		goto out;
1493	}
1494
1495	/* Check that attaching information is correct */
1496	ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1497		last_aeb = NULL;
1498		ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1499			int vol_type;
1500
1501			cond_resched();
1502
1503			last_aeb = aeb;
1504
1505			err = ubi_io_read_vid_hdr(ubi, aeb->pnum, vidh, 1);
1506			if (err && err != UBI_IO_BITFLIPS) {
1507				ubi_err("VID header is not OK (%d)", err);
1508				if (err > 0)
1509					err = -EIO;
1510				return err;
1511			}
1512
1513			vol_type = vidh->vol_type == UBI_VID_DYNAMIC ?
1514				   UBI_DYNAMIC_VOLUME : UBI_STATIC_VOLUME;
1515			if (av->vol_type != vol_type) {
1516				ubi_err("bad vol_type");
1517				goto bad_vid_hdr;
1518			}
1519
1520			if (aeb->sqnum != be64_to_cpu(vidh->sqnum)) {
1521				ubi_err("bad sqnum %llu", aeb->sqnum);
1522				goto bad_vid_hdr;
1523			}
1524
1525			if (av->vol_id != be32_to_cpu(vidh->vol_id)) {
1526				ubi_err("bad vol_id %d", av->vol_id);
1527				goto bad_vid_hdr;
1528			}
1529
1530			if (av->compat != vidh->compat) {
1531				ubi_err("bad compat %d", vidh->compat);
1532				goto bad_vid_hdr;
1533			}
1534
1535			if (aeb->lnum != be32_to_cpu(vidh->lnum)) {
1536				ubi_err("bad lnum %d", aeb->lnum);
1537				goto bad_vid_hdr;
1538			}
1539
1540			if (av->used_ebs != be32_to_cpu(vidh->used_ebs)) {
1541				ubi_err("bad used_ebs %d", av->used_ebs);
1542				goto bad_vid_hdr;
1543			}
1544
1545			if (av->data_pad != be32_to_cpu(vidh->data_pad)) {
1546				ubi_err("bad data_pad %d", av->data_pad);
1547				goto bad_vid_hdr;
1548			}
1549		}
1550
1551		if (!last_aeb)
1552			continue;
1553
1554		if (av->highest_lnum != be32_to_cpu(vidh->lnum)) {
1555			ubi_err("bad highest_lnum %d", av->highest_lnum);
1556			goto bad_vid_hdr;
1557		}
1558
1559		if (av->last_data_size != be32_to_cpu(vidh->data_size)) {
1560			ubi_err("bad last_data_size %d", av->last_data_size);
1561			goto bad_vid_hdr;
1562		}
1563	}
1564
1565	/*
1566	 * Make sure that all the physical eraseblocks are in one of the lists
1567	 * or trees.
1568	 */
1569	buf = kzalloc(ubi->peb_count, GFP_KERNEL);
1570	if (!buf)
1571		return -ENOMEM;
1572
1573	for (pnum = 0; pnum < ubi->peb_count; pnum++) {
1574		err = ubi_io_is_bad(ubi, pnum);
1575		if (err < 0) {
1576			kfree(buf);
1577			return err;
1578		} else if (err)
1579			buf[pnum] = 1;
1580	}
1581
1582	ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb)
1583		ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb)
1584			buf[aeb->pnum] = 1;
1585
1586	list_for_each_entry(aeb, &ai->free, u.list)
1587		buf[aeb->pnum] = 1;
1588
1589	list_for_each_entry(aeb, &ai->corr, u.list)
1590		buf[aeb->pnum] = 1;
1591
1592	list_for_each_entry(aeb, &ai->erase, u.list)
1593		buf[aeb->pnum] = 1;
1594
1595	list_for_each_entry(aeb, &ai->alien, u.list)
1596		buf[aeb->pnum] = 1;
1597
1598	err = 0;
1599	for (pnum = 0; pnum < ubi->peb_count; pnum++)
1600		if (!buf[pnum]) {
1601			ubi_err("PEB %d is not referred", pnum);
1602			err = 1;
1603		}
1604
1605	kfree(buf);
1606	if (err)
1607		goto out;
1608	return 0;
1609
1610bad_aeb:
1611	ubi_err("bad attaching information about LEB %d", aeb->lnum);
1612	ubi_dump_aeb(aeb, 0);
1613	ubi_dump_av(av);
1614	goto out;
1615
1616bad_av:
1617	ubi_err("bad attaching information about volume %d", av->vol_id);
1618	ubi_dump_av(av);
1619	goto out;
1620
1621bad_vid_hdr:
1622	ubi_err("bad attaching information about volume %d", av->vol_id);
1623	ubi_dump_av(av);
1624	ubi_dump_vid_hdr(vidh);
1625
1626out:
1627	dump_stack();
1628	return -EINVAL;
1629}
1630