2 * Copyright (c) International Business Machines Corp., 2006
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.
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.
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
18 * Author: Artem Bityutskiy (Битюцкий Артём)
22 * UBI attaching sub-system.
24 * This sub-system is responsible for attaching MTD devices and it also
25 * implements flash media scanning.
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.
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
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.
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.
50 * UBI tries to distinguish between 2 types of corruptions.
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).
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.
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.
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
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 it corruption type 2.
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>
92 static int self_check_ai(struct ubi_device
*ubi
, struct ubi_attach_info
*ai
);
94 /* Temporary variables used during scanning */
95 static struct ubi_ec_hdr
*ech
;
96 static struct ubi_vid_hdr
*vidh
;
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
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
119 static 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
)
122 struct ubi_ainf_peb
*aeb
;
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;
134 aeb
= kmem_cache_alloc(ai
->aeb_slab_cache
, GFP_KERNEL
);
139 aeb
->vol_id
= vol_id
;
143 list_add(&aeb
->u
.list
, list
);
145 list_add_tail(&aeb
->u
.list
, list
);
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
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.
160 static int add_corrupted(struct ubi_attach_info
*ai
, int pnum
, int ec
)
162 struct ubi_ainf_peb
*aeb
;
164 dbg_bld("add to corrupted: PEB %d, EC %d", pnum
, ec
);
166 aeb
= kmem_cache_alloc(ai
->aeb_slab_cache
, GFP_KERNEL
);
170 ai
->corr_peb_count
+= 1;
173 list_add(&aeb
->u
.list
, &ai
->corr
);
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
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.
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.
191 static int validate_vid_hdr(const struct ubi_vid_hdr
*vid_hdr
,
192 const struct ubi_ainf_volume
*av
, int pnum
)
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
);
199 if (av
->leb_count
!= 0) {
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.
208 if (vol_id
!= av
->vol_id
) {
209 ubi_err("inconsistent vol_id");
213 if (av
->vol_type
== UBI_STATIC_VOLUME
)
214 av_vol_type
= UBI_VID_STATIC
;
216 av_vol_type
= UBI_VID_DYNAMIC
;
218 if (vol_type
!= av_vol_type
) {
219 ubi_err("inconsistent vol_type");
223 if (used_ebs
!= av
->used_ebs
) {
224 ubi_err("inconsistent used_ebs");
228 if (data_pad
!= av
->data_pad
) {
229 ubi_err("inconsistent data_pad");
237 ubi_err("inconsistent VID header at PEB %d", pnum
);
238 ubi_dump_vid_hdr(vid_hdr
);
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
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
256 static 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
)
260 struct ubi_ainf_volume
*av
;
261 struct rb_node
**p
= &ai
->volumes
.rb_node
, *parent
= NULL
;
263 ubi_assert(vol_id
== be32_to_cpu(vid_hdr
->vol_id
));
265 /* Walk the volume RB-tree to look if this volume is already present */
268 av
= rb_entry(parent
, struct ubi_ainf_volume
, rb
);
270 if (vol_id
== av
->vol_id
)
273 if (vol_id
> av
->vol_id
)
279 /* The volume is absent - add it */
280 av
= kmalloc(sizeof(struct ubi_ainf_volume
), GFP_KERNEL
);
282 return ERR_PTR(-ENOMEM
);
284 av
->highest_lnum
= av
->leb_count
= 0;
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
292 if (vol_id
> ai
->highest_vol_id
)
293 ai
->highest_vol_id
= vol_id
;
295 rb_link_node(&av
->rb
, parent
, p
);
296 rb_insert_color(&av
->rb
, &ai
->volumes
);
298 dbg_bld("added volume %d", vol_id
);
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
308 * @vid_hdr: volume identifier header of the second logical eraseblock
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
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.
322 static int compare_lebs(struct ubi_device
*ubi
, const struct ubi_ainf_peb
*aeb
,
323 int pnum
, const struct ubi_vid_hdr
*vid_hdr
)
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
);
331 if (sqnum2
== aeb
->sqnum
) {
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.
340 ubi_err("unsupported on-flash UBI format\n");
344 /* Obviously the LEB with lower sequence counter is older */
345 second_is_newer
= (sqnum2
> aeb
->sqnum
);
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.
353 * Note: this may be optimized so that we wouldn't read twice.
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",
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",
368 return bitflips
<< 1;
371 vh
= ubi_zalloc_vid_hdr(ubi
, GFP_KERNEL
);
376 err
= ubi_io_read_vid_hdr(ubi
, pnum
, vh
, 0);
378 if (err
== UBI_IO_BITFLIPS
)
381 ubi_err("VID of PEB %d header is bad, but it "
382 "was OK earlier, err %d", pnum
, err
);
393 /* Read the data of the copy and check the CRC */
395 len
= be32_to_cpu(vid_hdr
->data_size
);
402 err
= ubi_io_read_data(ubi
, buf
, pnum
, 0, len
);
403 if (err
&& err
!= UBI_IO_BITFLIPS
&& !mtd_is_eccerr(err
))
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
);
413 second_is_newer
= !second_is_newer
;
415 dbg_bld("PEB %d CRC is OK", pnum
);
420 ubi_free_vid_hdr(ubi
, vh
);
423 dbg_bld("second PEB %d is newer, copy_flag is set", pnum
);
425 dbg_bld("first PEB %d is newer, copy_flag is set", pnum
);
427 return second_is_newer
| (bitflips
<< 1) | (corrupted
<< 2);
432 ubi_free_vid_hdr(ubi
, vh
);
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
442 * @vid_hdr: the volume identifier header
443 * @bitflips: if bit-flips were detected when this physical eraseblock was read
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.
452 int 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
)
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
;
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
);
465 dbg_bld("PEB %d, LEB %d:%d, EC %d, sqnum %llu, bitflips %d",
466 pnum
, vol_id
, lnum
, ec
, sqnum
, bitflips
);
468 av
= add_volume(ai
, vol_id
, pnum
, vid_hdr
);
472 if (ai
->max_sqnum
< sqnum
)
473 ai
->max_sqnum
= sqnum
;
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.
479 p
= &av
->root
.rb_node
;
484 aeb
= rb_entry(parent
, struct ubi_ainf_peb
, u
.rb
);
485 if (lnum
!= aeb
->lnum
) {
486 if (lnum
< aeb
->lnum
)
494 * There is already a physical eraseblock describing the same
495 * logical eraseblock present.
498 dbg_bld("this LEB already exists: PEB %d, sqnum %llu, EC %d",
499 aeb
->pnum
, aeb
->sqnum
, aeb
->ec
);
502 * Make sure that the logical eraseblocks have different
503 * sequence numbers. Otherwise the image is bad.
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.
514 if (aeb
->sqnum
== sqnum
&& sqnum
!= 0) {
515 ubi_err("two LEBs with same sequence number %llu",
517 ubi_dump_aeb(aeb
, 0);
518 ubi_dump_vid_hdr(vid_hdr
);
523 * Now we have to drop the older one and preserve the newer
526 cmp_res
= compare_lebs(ubi
, aeb
, pnum
, vid_hdr
);
532 * This logical eraseblock is newer than the one
535 err
= validate_vid_hdr(vid_hdr
, av
, pnum
);
539 err
= add_to_list(ai
, aeb
->pnum
, aeb
->vol_id
,
540 aeb
->lnum
, aeb
->ec
, cmp_res
& 4,
547 aeb
->vol_id
= vol_id
;
549 aeb
->scrub
= ((cmp_res
& 2) || bitflips
);
550 aeb
->copy_flag
= vid_hdr
->copy_flag
;
553 if (av
->highest_lnum
== lnum
)
555 be32_to_cpu(vid_hdr
->data_size
);
560 * This logical eraseblock is older than the one found
563 return add_to_list(ai
, pnum
, vol_id
, lnum
, ec
,
564 cmp_res
& 4, &ai
->erase
);
569 * We've met this logical eraseblock for the first time, add it to the
570 * attaching information.
573 err
= validate_vid_hdr(vid_hdr
, av
, pnum
);
577 aeb
= kmem_cache_alloc(ai
->aeb_slab_cache
, GFP_KERNEL
);
583 aeb
->vol_id
= vol_id
;
585 aeb
->scrub
= bitflips
;
586 aeb
->copy_flag
= vid_hdr
->copy_flag
;
589 if (av
->highest_lnum
<= lnum
) {
590 av
->highest_lnum
= lnum
;
591 av
->last_data_size
= be32_to_cpu(vid_hdr
->data_size
);
595 rb_link_node(&aeb
->u
.rb
, parent
, p
);
596 rb_insert_color(&aeb
->u
.rb
, &av
->root
);
601 * ubi_find_av - find volume in the attaching information.
602 * @ai: attaching information
603 * @vol_id: the requested volume ID
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.
608 struct ubi_ainf_volume
*ubi_find_av(const struct ubi_attach_info
*ai
,
611 struct ubi_ainf_volume
*av
;
612 struct rb_node
*p
= ai
->volumes
.rb_node
;
615 av
= rb_entry(p
, struct ubi_ainf_volume
, rb
);
617 if (vol_id
== av
->vol_id
)
620 if (vol_id
> av
->vol_id
)
630 * ubi_remove_av - delete attaching information about a volume.
631 * @ai: attaching information
632 * @av: the volume attaching information to delete
634 void ubi_remove_av(struct ubi_attach_info
*ai
, struct ubi_ainf_volume
*av
)
637 struct ubi_ainf_peb
*aeb
;
639 dbg_bld("remove attaching information about volume %d", av
->vol_id
);
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
);
647 rb_erase(&av
->rb
, &ai
->volumes
);
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)
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
665 static int early_erase_peb(struct ubi_device
*ubi
,
666 const struct ubi_attach_info
*ai
, int pnum
, int ec
)
669 struct ubi_ec_hdr
*ec_hdr
;
671 if ((long long)ec
>= UBI_MAX_ERASECOUNTER
) {
673 * Erase counter overflow. Upgrade UBI and use 64-bit
674 * erase counters internally.
676 ubi_err("erase counter overflow at PEB %d, EC %d", pnum
, ec
);
680 ec_hdr
= kzalloc(ubi
->ec_hdr_alsize
, GFP_KERNEL
);
684 ec_hdr
->ec
= cpu_to_be64(ec
);
686 err
= ubi_io_sync_erase(ubi
, pnum
, 0);
690 err
= ubi_io_write_ec_hdr(ubi
, pnum
, ec_hdr
);
698 * ubi_early_get_peb - get a free physical eraseblock.
699 * @ubi: UBI device description object
700 * @ai: attaching information
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
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.
711 struct ubi_ainf_peb
*ubi_early_get_peb(struct ubi_device
*ubi
,
712 struct ubi_attach_info
*ai
)
715 struct ubi_ainf_peb
*aeb
, *tmp_aeb
;
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
);
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.
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
;
734 err
= early_erase_peb(ubi
, ai
, aeb
->pnum
, aeb
->ec
+1);
739 list_del(&aeb
->u
.list
);
740 dbg_bld("return PEB %d, EC %d", aeb
->pnum
, aeb
->ec
);
744 ubi_err("no free eraseblocks");
745 return ERR_PTR(-ENOSPC
);
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
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.
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
765 static int check_corruption(struct ubi_device
*ubi
, struct ubi_vid_hdr
*vid_hdr
,
770 mutex_lock(&ubi
->buf_mutex
);
771 memset(ubi
->peb_buf
, 0x00, ubi
->leb_size
);
773 err
= ubi_io_read(ubi
, ubi
->peb_buf
, pnum
, ubi
->leb_start
,
775 if (err
== UBI_IO_BITFLIPS
|| mtd_is_eccerr(err
)) {
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
790 if (ubi_check_pattern(ubi
->peb_buf
, 0xFF, ubi
->leb_size
))
793 ubi_err("PEB %d contains corrupted VID header, and the data does not "
794 "contain all 0xFF, this may be a non-UBI PEB or a severe VID "
795 "header corruption which requires manual inspection", pnum
);
796 ubi_dump_vid_hdr(vid_hdr
);
797 dbg_msg("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);
804 mutex_unlock(&ubi
->buf_mutex
);
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
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.
819 static int scan_peb(struct ubi_device
*ubi
, struct ubi_attach_info
*ai
,
822 long long uninitialized_var(ec
);
823 int err
, bitflips
= 0, vol_id
, ec_err
= 0;
825 dbg_bld("scan PEB %d", pnum
);
827 /* Skip bad physical eraseblocks */
828 err
= ubi_io_is_bad(ubi
, pnum
);
832 ai
->bad_peb_count
+= 1;
836 err
= ubi_io_read_ec_hdr(ubi
, pnum
, ech
, 0);
842 case UBI_IO_BITFLIPS
:
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
:
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.
865 ubi_err("'ubi_io_read_ec_hdr()' returned unknown code %d", err
);
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
);
879 ec
= be64_to_cpu(ech
->ec
);
880 if (ec
> UBI_MAX_ERASECOUNTER
) {
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
888 ubi_err("erase counter overflow, max is %d",
889 UBI_MAX_ERASECOUNTER
);
890 ubi_dump_ec_hdr(ech
);
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
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, "
911 "expected %d", image_seq
, pnum
, ubi
->image_seq
);
912 ubi_dump_ec_hdr(ech
);
917 /* OK, we've done with the EC header, let's look at the VID header */
919 err
= ubi_io_read_vid_hdr(ubi
, pnum
, vidh
, 0);
925 case UBI_IO_BITFLIPS
:
928 case UBI_IO_BAD_HDR_EBADMSG
:
929 if (ec_err
== UBI_IO_BAD_HDR_EBADMSG
)
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.
936 ai
->maybe_bad_peb_count
+= 1;
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.
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.
954 * The EC was OK, but the VID header is corrupted. We
955 * have to check what is in the data area.
957 err
= check_corruption(ubi
, vidh
, pnum
);
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
);
966 /* This is an unexpected corruption */
967 err
= add_corrupted(ai
, pnum
, ec
);
971 case UBI_IO_FF_BITFLIPS
:
972 err
= add_to_list(ai
, pnum
, UBI_UNKNOWN
, UBI_UNKNOWN
,
978 if (ec_err
|| bitflips
)
979 err
= add_to_list(ai
, pnum
, UBI_UNKNOWN
,
980 UBI_UNKNOWN
, ec
, 1, &ai
->erase
);
982 err
= add_to_list(ai
, pnum
, UBI_UNKNOWN
,
983 UBI_UNKNOWN
, ec
, 0, &ai
->free
);
988 ubi_err("'ubi_io_read_vid_hdr()' returned unknown code %d",
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
);
997 /* Unsupported internal volume */
998 switch (vidh
->compat
) {
999 case UBI_COMPAT_DELETE
:
1000 ubi_msg("\"delete\" compatible internal volume %d:%d"
1001 " found, will remove it", vol_id
, lnum
);
1002 err
= add_to_list(ai
, pnum
, vol_id
, lnum
,
1009 ubi_msg("read-only compatible internal volume %d:%d"
1010 " found, switch to read-only mode",
1015 case UBI_COMPAT_PRESERVE
:
1016 ubi_msg("\"preserve\" compatible internal volume %d:%d"
1017 " found", vol_id
, lnum
);
1018 err
= add_to_list(ai
, pnum
, vol_id
, lnum
,
1024 case UBI_COMPAT_REJECT
:
1025 ubi_err("incompatible internal volume %d:%d found",
1032 ubi_warn("valid VID header but corrupted EC header at PEB %d",
1034 err
= ubi_add_to_av(ubi
, ai
, pnum
, ec
, vidh
, bitflips
);
1042 if (ec
> ai
->max_ec
)
1044 if (ec
< ai
->min_ec
)
1052 * late_analysis - analyze the overall situation with PEB.
1053 * @ubi: UBI device description object
1054 * @ai: attaching information
1056 * This is a helper function which takes a look what PEBs we have after we
1057 * gather information about all of them ("ai" is compete). It decides whether
1058 * the flash is empty and should be formatted of whether there are too many
1059 * corrupted PEBs and we should not attach this MTD device. Returns zero if we
1060 * should proceed with attaching the MTD device, and %-EINVAL if we should not.
1062 static int late_analysis(struct ubi_device
*ubi
, struct ubi_attach_info
*ai
)
1064 struct ubi_ainf_peb
*aeb
;
1065 int max_corr
, peb_count
;
1067 peb_count
= ubi
->peb_count
- ai
->bad_peb_count
- ai
->alien_peb_count
;
1068 max_corr
= peb_count
/ 20 ?: 8;
1071 * Few corrupted PEBs is not a problem and may be just a result of
1072 * unclean reboots. However, many of them may indicate some problems
1073 * with the flash HW or driver.
1075 if (ai
->corr_peb_count
) {
1076 ubi_err("%d PEBs are corrupted and preserved",
1077 ai
->corr_peb_count
);
1078 printk(KERN_ERR
"Corrupted PEBs are:");
1079 list_for_each_entry(aeb
, &ai
->corr
, u
.list
)
1080 printk(KERN_CONT
" %d", aeb
->pnum
);
1081 printk(KERN_CONT
"\n");
1084 * If too many PEBs are corrupted, we refuse attaching,
1085 * otherwise, only print a warning.
1087 if (ai
->corr_peb_count
>= max_corr
) {
1088 ubi_err("too many corrupted PEBs, refusing");
1093 if (ai
->empty_peb_count
+ ai
->maybe_bad_peb_count
== peb_count
) {
1095 * All PEBs are empty, or almost all - a couple PEBs look like
1096 * they may be bad PEBs which were not marked as bad yet.
1098 * This piece of code basically tries to distinguish between
1099 * the following situations:
1101 * 1. Flash is empty, but there are few bad PEBs, which are not
1102 * marked as bad so far, and which were read with error. We
1103 * want to go ahead and format this flash. While formatting,
1104 * the faulty PEBs will probably be marked as bad.
1106 * 2. Flash contains non-UBI data and we do not want to format
1107 * it and destroy possibly important information.
1109 if (ai
->maybe_bad_peb_count
<= 2) {
1111 ubi_msg("empty MTD device detected");
1112 get_random_bytes(&ubi
->image_seq
,
1113 sizeof(ubi
->image_seq
));
1115 ubi_err("MTD device is not UBI-formatted and possibly "
1116 "contains non-UBI data - refusing it");
1126 * scan_all - scan entire MTD device.
1127 * @ubi: UBI device description object
1129 * This function does full scanning of an MTD device and returns complete
1130 * information about it in form of a "struct ubi_attach_info" object. In case
1131 * of failure, an error code is returned.
1133 static struct ubi_attach_info
*scan_all(struct ubi_device
*ubi
)
1136 struct rb_node
*rb1
, *rb2
;
1137 struct ubi_ainf_volume
*av
;
1138 struct ubi_ainf_peb
*aeb
;
1139 struct ubi_attach_info
*ai
;
1141 ai
= kzalloc(sizeof(struct ubi_attach_info
), GFP_KERNEL
);
1143 return ERR_PTR(-ENOMEM
);
1145 INIT_LIST_HEAD(&ai
->corr
);
1146 INIT_LIST_HEAD(&ai
->free
);
1147 INIT_LIST_HEAD(&ai
->erase
);
1148 INIT_LIST_HEAD(&ai
->alien
);
1149 ai
->volumes
= RB_ROOT
;
1152 ai
->aeb_slab_cache
= kmem_cache_create("ubi_aeb_slab_cache",
1153 sizeof(struct ubi_ainf_peb
),
1155 if (!ai
->aeb_slab_cache
)
1158 ech
= kzalloc(ubi
->ec_hdr_alsize
, GFP_KERNEL
);
1162 vidh
= ubi_zalloc_vid_hdr(ubi
, GFP_KERNEL
);
1166 for (pnum
= 0; pnum
< ubi
->peb_count
; pnum
++) {
1169 dbg_gen("process PEB %d", pnum
);
1170 err
= scan_peb(ubi
, ai
, pnum
);
1175 dbg_msg("scanning is finished");
1177 /* Calculate mean erase counter */
1179 ai
->mean_ec
= div_u64(ai
->ec_sum
, ai
->ec_count
);
1181 err
= late_analysis(ubi
, ai
);
1186 * In case of unknown erase counter we use the mean erase counter
1189 ubi_rb_for_each_entry(rb1
, av
, &ai
->volumes
, rb
) {
1190 ubi_rb_for_each_entry(rb2
, aeb
, &av
->root
, u
.rb
)
1191 if (aeb
->ec
== UBI_UNKNOWN
)
1192 aeb
->ec
= ai
->mean_ec
;
1195 list_for_each_entry(aeb
, &ai
->free
, u
.list
) {
1196 if (aeb
->ec
== UBI_UNKNOWN
)
1197 aeb
->ec
= ai
->mean_ec
;
1200 list_for_each_entry(aeb
, &ai
->corr
, u
.list
)
1201 if (aeb
->ec
== UBI_UNKNOWN
)
1202 aeb
->ec
= ai
->mean_ec
;
1204 list_for_each_entry(aeb
, &ai
->erase
, u
.list
)
1205 if (aeb
->ec
== UBI_UNKNOWN
)
1206 aeb
->ec
= ai
->mean_ec
;
1208 err
= self_check_ai(ubi
, ai
);
1212 ubi_free_vid_hdr(ubi
, vidh
);
1218 ubi_free_vid_hdr(ubi
, vidh
);
1223 return ERR_PTR(err
);
1227 * ubi_attach - attach an MTD device.
1228 * @ubi: UBI device descriptor
1230 * This function returns zero in case of success and a negative error code in
1233 int ubi_attach(struct ubi_device
*ubi
)
1236 struct ubi_attach_info
*ai
;
1242 ubi
->bad_peb_count
= ai
->bad_peb_count
;
1243 ubi
->good_peb_count
= ubi
->peb_count
- ubi
->bad_peb_count
;
1244 ubi
->corr_peb_count
= ai
->corr_peb_count
;
1245 ubi
->max_ec
= ai
->max_ec
;
1246 ubi
->mean_ec
= ai
->mean_ec
;
1247 ubi_msg("max. sequence number: %llu", ai
->max_sqnum
);
1249 err
= ubi_read_volume_table(ubi
, ai
);
1253 err
= ubi_wl_init(ubi
, ai
);
1257 err
= ubi_eba_init(ubi
, ai
);
1267 ubi_free_internal_volumes(ubi
);
1275 * destroy_av - free volume attaching information.
1276 * @av: volume attaching information
1277 * @ai: attaching information
1279 * This function destroys the volume attaching information.
1281 static void destroy_av(struct ubi_attach_info
*ai
, struct ubi_ainf_volume
*av
)
1283 struct ubi_ainf_peb
*aeb
;
1284 struct rb_node
*this = av
->root
.rb_node
;
1288 this = this->rb_left
;
1289 else if (this->rb_right
)
1290 this = this->rb_right
;
1292 aeb
= rb_entry(this, struct ubi_ainf_peb
, u
.rb
);
1293 this = rb_parent(this);
1295 if (this->rb_left
== &aeb
->u
.rb
)
1296 this->rb_left
= NULL
;
1298 this->rb_right
= NULL
;
1301 kmem_cache_free(ai
->aeb_slab_cache
, aeb
);
1308 * ubi_destroy_ai - destroy attaching information.
1309 * @ai: attaching information
1311 void ubi_destroy_ai(struct ubi_attach_info
*ai
)
1313 struct ubi_ainf_peb
*aeb
, *aeb_tmp
;
1314 struct ubi_ainf_volume
*av
;
1317 list_for_each_entry_safe(aeb
, aeb_tmp
, &ai
->alien
, u
.list
) {
1318 list_del(&aeb
->u
.list
);
1319 kmem_cache_free(ai
->aeb_slab_cache
, aeb
);
1321 list_for_each_entry_safe(aeb
, aeb_tmp
, &ai
->erase
, u
.list
) {
1322 list_del(&aeb
->u
.list
);
1323 kmem_cache_free(ai
->aeb_slab_cache
, aeb
);
1325 list_for_each_entry_safe(aeb
, aeb_tmp
, &ai
->corr
, u
.list
) {
1326 list_del(&aeb
->u
.list
);
1327 kmem_cache_free(ai
->aeb_slab_cache
, aeb
);
1329 list_for_each_entry_safe(aeb
, aeb_tmp
, &ai
->free
, u
.list
) {
1330 list_del(&aeb
->u
.list
);
1331 kmem_cache_free(ai
->aeb_slab_cache
, aeb
);
1334 /* Destroy the volume RB-tree */
1335 rb
= ai
->volumes
.rb_node
;
1339 else if (rb
->rb_right
)
1342 av
= rb_entry(rb
, struct ubi_ainf_volume
, rb
);
1346 if (rb
->rb_left
== &av
->rb
)
1349 rb
->rb_right
= NULL
;
1356 if (ai
->aeb_slab_cache
)
1357 kmem_cache_destroy(ai
->aeb_slab_cache
);
1363 * self_check_ai - check the attaching information.
1364 * @ubi: UBI device description object
1365 * @ai: attaching information
1367 * This function returns zero if the attaching information is all right, and a
1368 * negative error code if not or if an error occurred.
1370 static int self_check_ai(struct ubi_device
*ubi
, struct ubi_attach_info
*ai
)
1372 int pnum
, err
, vols_found
= 0;
1373 struct rb_node
*rb1
, *rb2
;
1374 struct ubi_ainf_volume
*av
;
1375 struct ubi_ainf_peb
*aeb
, *last_aeb
;
1378 if (!ubi
->dbg
->chk_gen
)
1382 * At first, check that attaching information is OK.
1384 ubi_rb_for_each_entry(rb1
, av
, &ai
->volumes
, rb
) {
1392 ubi_err("bad is_empty flag");
1396 if (av
->vol_id
< 0 || av
->highest_lnum
< 0 ||
1397 av
->leb_count
< 0 || av
->vol_type
< 0 || av
->used_ebs
< 0 ||
1398 av
->data_pad
< 0 || av
->last_data_size
< 0) {
1399 ubi_err("negative values");
1403 if (av
->vol_id
>= UBI_MAX_VOLUMES
&&
1404 av
->vol_id
< UBI_INTERNAL_VOL_START
) {
1405 ubi_err("bad vol_id");
1409 if (av
->vol_id
> ai
->highest_vol_id
) {
1410 ubi_err("highest_vol_id is %d, but vol_id %d is there",
1411 ai
->highest_vol_id
, av
->vol_id
);
1415 if (av
->vol_type
!= UBI_DYNAMIC_VOLUME
&&
1416 av
->vol_type
!= UBI_STATIC_VOLUME
) {
1417 ubi_err("bad vol_type");
1421 if (av
->data_pad
> ubi
->leb_size
/ 2) {
1422 ubi_err("bad data_pad");
1427 ubi_rb_for_each_entry(rb2
, aeb
, &av
->root
, u
.rb
) {
1433 if (aeb
->pnum
< 0 || aeb
->ec
< 0) {
1434 ubi_err("negative values");
1438 if (aeb
->ec
< ai
->min_ec
) {
1439 ubi_err("bad ai->min_ec (%d), %d found",
1440 ai
->min_ec
, aeb
->ec
);
1444 if (aeb
->ec
> ai
->max_ec
) {
1445 ubi_err("bad ai->max_ec (%d), %d found",
1446 ai
->max_ec
, aeb
->ec
);
1450 if (aeb
->pnum
>= ubi
->peb_count
) {
1451 ubi_err("too high PEB number %d, total PEBs %d",
1452 aeb
->pnum
, ubi
->peb_count
);
1456 if (av
->vol_type
== UBI_STATIC_VOLUME
) {
1457 if (aeb
->lnum
>= av
->used_ebs
) {
1458 ubi_err("bad lnum or used_ebs");
1462 if (av
->used_ebs
!= 0) {
1463 ubi_err("non-zero used_ebs");
1468 if (aeb
->lnum
> av
->highest_lnum
) {
1469 ubi_err("incorrect highest_lnum or lnum");
1474 if (av
->leb_count
!= leb_count
) {
1475 ubi_err("bad leb_count, %d objects in the tree",
1485 if (aeb
->lnum
!= av
->highest_lnum
) {
1486 ubi_err("bad highest_lnum");
1491 if (vols_found
!= ai
->vols_found
) {
1492 ubi_err("bad ai->vols_found %d, should be %d",
1493 ai
->vols_found
, vols_found
);
1497 /* Check that attaching information is correct */
1498 ubi_rb_for_each_entry(rb1
, av
, &ai
->volumes
, rb
) {
1500 ubi_rb_for_each_entry(rb2
, aeb
, &av
->root
, u
.rb
) {
1507 err
= ubi_io_read_vid_hdr(ubi
, aeb
->pnum
, vidh
, 1);
1508 if (err
&& err
!= UBI_IO_BITFLIPS
) {
1509 ubi_err("VID header is not OK (%d)", err
);
1515 vol_type
= vidh
->vol_type
== UBI_VID_DYNAMIC
?
1516 UBI_DYNAMIC_VOLUME
: UBI_STATIC_VOLUME
;
1517 if (av
->vol_type
!= vol_type
) {
1518 ubi_err("bad vol_type");
1522 if (aeb
->sqnum
!= be64_to_cpu(vidh
->sqnum
)) {
1523 ubi_err("bad sqnum %llu", aeb
->sqnum
);
1527 if (av
->vol_id
!= be32_to_cpu(vidh
->vol_id
)) {
1528 ubi_err("bad vol_id %d", av
->vol_id
);
1532 if (av
->compat
!= vidh
->compat
) {
1533 ubi_err("bad compat %d", vidh
->compat
);
1537 if (aeb
->lnum
!= be32_to_cpu(vidh
->lnum
)) {
1538 ubi_err("bad lnum %d", aeb
->lnum
);
1542 if (av
->used_ebs
!= be32_to_cpu(vidh
->used_ebs
)) {
1543 ubi_err("bad used_ebs %d", av
->used_ebs
);
1547 if (av
->data_pad
!= be32_to_cpu(vidh
->data_pad
)) {
1548 ubi_err("bad data_pad %d", av
->data_pad
);
1556 if (av
->highest_lnum
!= be32_to_cpu(vidh
->lnum
)) {
1557 ubi_err("bad highest_lnum %d", av
->highest_lnum
);
1561 if (av
->last_data_size
!= be32_to_cpu(vidh
->data_size
)) {
1562 ubi_err("bad last_data_size %d", av
->last_data_size
);
1568 * Make sure that all the physical eraseblocks are in one of the lists
1571 buf
= kzalloc(ubi
->peb_count
, GFP_KERNEL
);
1575 for (pnum
= 0; pnum
< ubi
->peb_count
; pnum
++) {
1576 err
= ubi_io_is_bad(ubi
, pnum
);
1584 ubi_rb_for_each_entry(rb1
, av
, &ai
->volumes
, rb
)
1585 ubi_rb_for_each_entry(rb2
, aeb
, &av
->root
, u
.rb
)
1588 list_for_each_entry(aeb
, &ai
->free
, u
.list
)
1591 list_for_each_entry(aeb
, &ai
->corr
, u
.list
)
1594 list_for_each_entry(aeb
, &ai
->erase
, u
.list
)
1597 list_for_each_entry(aeb
, &ai
->alien
, u
.list
)
1601 for (pnum
= 0; pnum
< ubi
->peb_count
; pnum
++)
1603 ubi_err("PEB %d is not referred", pnum
);
1613 ubi_err("bad attaching information about LEB %d", aeb
->lnum
);
1614 ubi_dump_aeb(aeb
, 0);
1619 ubi_err("bad attaching information about volume %d", av
->vol_id
);
1624 ubi_err("bad attaching information about volume %d", av
->vol_id
);
1626 ubi_dump_vid_hdr(vidh
);