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 scanning sub-system.
24 * This sub-system is responsible for scanning the flash media, checking UBI
25 * headers and providing complete information about the UBI flash image.
27 * The scanning information is represented by a &struct ubi_scan_info' object.
28 * Information about found volumes is represented by &struct ubi_scan_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 * Scanned logical eraseblocks are represented by &struct ubi_scan_leb objects.
33 * These objects are kept in per-volume RB-trees with the root at the
34 * corresponding &struct ubi_scan_volume object. To put it differently, we keep
35 * an RB-tree of per-volume objects and each of these objects is the root of
36 * RB-tree of per-eraseblock objects.
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 case
55 * - we may lose only the data which was being written to the media just before
56 * the power cut happened, and the upper layers (e.g., UBIFS) are supposed to
57 * 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 * scanning, 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. UBI assumes corruption type 2
71 * if the VID header is corrupted and the data area does not contain all 0xFFs,
72 * and there were no bit-flips or integrity errors while reading the data area.
73 * Otherwise UBI assumes corruption type 1. So the decision criteria are as
75 * o If the data area contains only 0xFFs, there is no data, and it is safe
76 * to just erase this PEB - this is corruption type 1.
77 * o If the data area has bit-flips or data integrity errors (ECC errors on
78 * NAND), it is probably a PEB which was being erased when power cut
79 * happened, so this is corruption type 1. However, this is just a guess,
80 * which might be wrong.
81 * o Otherwise this it corruption type 2.
84 #include <linux/err.h>
85 #include <linux/slab.h>
86 #include <linux/crc32.h>
87 #include <linux/math64.h>
88 #include <linux/random.h>
91 #ifdef CONFIG_MTD_UBI_DEBUG
92 static int paranoid_check_si(struct ubi_device
*ubi
, struct ubi_scan_info
*si
);
94 #define paranoid_check_si(ubi, si) 0
97 /* Temporary variables used during scanning */
98 static struct ubi_ec_hdr
*ech
;
99 static struct ubi_vid_hdr
*vidh
;
102 * add_to_list - add physical eraseblock to a list.
103 * @si: scanning information
104 * @pnum: physical eraseblock number to add
105 * @ec: erase counter of the physical eraseblock
106 * @to_head: if not zero, add to the head of the list
107 * @list: the list to add to
109 * This function adds physical eraseblock @pnum to free, erase, or alien lists.
110 * If @to_head is not zero, PEB will be added to the head of the list, which
111 * basically means it will be processed first later. E.g., we add corrupted
112 * PEBs (corrupted due to power cuts) to the head of the erase list to make
113 * sure we erase them first and get rid of corruptions ASAP. This function
114 * returns zero in case of success and a negative error code in case of
117 static int add_to_list(struct ubi_scan_info
*si
, int pnum
, int ec
, int to_head
,
118 struct list_head
*list
)
120 struct ubi_scan_leb
*seb
;
122 if (list
== &si
->free
) {
123 dbg_bld("add to free: PEB %d, EC %d", pnum
, ec
);
124 } else if (list
== &si
->erase
) {
125 dbg_bld("add to erase: PEB %d, EC %d", pnum
, ec
);
126 } else if (list
== &si
->alien
) {
127 dbg_bld("add to alien: PEB %d, EC %d", pnum
, ec
);
128 si
->alien_peb_count
+= 1;
132 seb
= kmem_cache_alloc(si
->scan_leb_slab
, GFP_KERNEL
);
139 list_add(&seb
->u
.list
, list
);
141 list_add_tail(&seb
->u
.list
, list
);
146 * add_corrupted - add a corrupted physical eraseblock.
147 * @si: scanning information
148 * @pnum: physical eraseblock number to add
149 * @ec: erase counter of the physical eraseblock
151 * This function adds corrupted physical eraseblock @pnum to the 'corr' list.
152 * The corruption was presumably not caused by a power cut. Returns zero in
153 * case of success and a negative error code in case of failure.
155 static int add_corrupted(struct ubi_scan_info
*si
, int pnum
, int ec
)
157 struct ubi_scan_leb
*seb
;
159 dbg_bld("add to corrupted: PEB %d, EC %d", pnum
, ec
);
161 seb
= kmem_cache_alloc(si
->scan_leb_slab
, GFP_KERNEL
);
165 si
->corr_peb_count
+= 1;
168 list_add(&seb
->u
.list
, &si
->corr
);
173 * validate_vid_hdr - check volume identifier header.
174 * @vid_hdr: the volume identifier header to check
175 * @sv: information about the volume this logical eraseblock belongs to
176 * @pnum: physical eraseblock number the VID header came from
178 * This function checks that data stored in @vid_hdr is consistent. Returns
179 * non-zero if an inconsistency was found and zero if not.
181 * Note, UBI does sanity check of everything it reads from the flash media.
182 * Most of the checks are done in the I/O sub-system. Here we check that the
183 * information in the VID header is consistent to the information in other VID
184 * headers of the same volume.
186 static int validate_vid_hdr(const struct ubi_vid_hdr
*vid_hdr
,
187 const struct ubi_scan_volume
*sv
, int pnum
)
189 int vol_type
= vid_hdr
->vol_type
;
190 int vol_id
= be32_to_cpu(vid_hdr
->vol_id
);
191 int used_ebs
= be32_to_cpu(vid_hdr
->used_ebs
);
192 int data_pad
= be32_to_cpu(vid_hdr
->data_pad
);
194 if (sv
->leb_count
!= 0) {
198 * This is not the first logical eraseblock belonging to this
199 * volume. Ensure that the data in its VID header is consistent
200 * to the data in previous logical eraseblock headers.
203 if (vol_id
!= sv
->vol_id
) {
204 dbg_err("inconsistent vol_id");
208 if (sv
->vol_type
== UBI_STATIC_VOLUME
)
209 sv_vol_type
= UBI_VID_STATIC
;
211 sv_vol_type
= UBI_VID_DYNAMIC
;
213 if (vol_type
!= sv_vol_type
) {
214 dbg_err("inconsistent vol_type");
218 if (used_ebs
!= sv
->used_ebs
) {
219 dbg_err("inconsistent used_ebs");
223 if (data_pad
!= sv
->data_pad
) {
224 dbg_err("inconsistent data_pad");
232 ubi_err("inconsistent VID header at PEB %d", pnum
);
233 ubi_dbg_dump_vid_hdr(vid_hdr
);
239 * add_volume - add volume to the scanning information.
240 * @si: scanning information
241 * @vol_id: ID of the volume to add
242 * @pnum: physical eraseblock number
243 * @vid_hdr: volume identifier header
245 * If the volume corresponding to the @vid_hdr logical eraseblock is already
246 * present in the scanning information, this function does nothing. Otherwise
247 * it adds corresponding volume to the scanning information. Returns a pointer
248 * to the scanning volume object in case of success and a negative error code
249 * in case of failure.
251 static struct ubi_scan_volume
*add_volume(struct ubi_scan_info
*si
, int vol_id
,
253 const struct ubi_vid_hdr
*vid_hdr
)
255 struct ubi_scan_volume
*sv
;
256 struct rb_node
**p
= &si
->volumes
.rb_node
, *parent
= NULL
;
258 ubi_assert(vol_id
== be32_to_cpu(vid_hdr
->vol_id
));
260 /* Walk the volume RB-tree to look if this volume is already present */
263 sv
= rb_entry(parent
, struct ubi_scan_volume
, rb
);
265 if (vol_id
== sv
->vol_id
)
268 if (vol_id
> sv
->vol_id
)
274 /* The volume is absent - add it */
275 sv
= kmalloc(sizeof(struct ubi_scan_volume
), GFP_KERNEL
);
277 return ERR_PTR(-ENOMEM
);
279 sv
->highest_lnum
= sv
->leb_count
= 0;
282 sv
->used_ebs
= be32_to_cpu(vid_hdr
->used_ebs
);
283 sv
->data_pad
= be32_to_cpu(vid_hdr
->data_pad
);
284 sv
->compat
= vid_hdr
->compat
;
285 sv
->vol_type
= vid_hdr
->vol_type
== UBI_VID_DYNAMIC
? UBI_DYNAMIC_VOLUME
287 if (vol_id
> si
->highest_vol_id
)
288 si
->highest_vol_id
= vol_id
;
290 rb_link_node(&sv
->rb
, parent
, p
);
291 rb_insert_color(&sv
->rb
, &si
->volumes
);
293 dbg_bld("added volume %d", vol_id
);
298 * compare_lebs - find out which logical eraseblock is newer.
299 * @ubi: UBI device description object
300 * @seb: first logical eraseblock to compare
301 * @pnum: physical eraseblock number of the second logical eraseblock to
303 * @vid_hdr: volume identifier header of the second logical eraseblock
305 * This function compares 2 copies of a LEB and informs which one is newer. In
306 * case of success this function returns a positive value, in case of failure, a
307 * negative error code is returned. The success return codes use the following
309 * o bit 0 is cleared: the first PEB (described by @seb) is newer than the
310 * second PEB (described by @pnum and @vid_hdr);
311 * o bit 0 is set: the second PEB is newer;
312 * o bit 1 is cleared: no bit-flips were detected in the newer LEB;
313 * o bit 1 is set: bit-flips were detected in the newer LEB;
314 * o bit 2 is cleared: the older LEB is not corrupted;
315 * o bit 2 is set: the older LEB is corrupted.
317 static int compare_lebs(struct ubi_device
*ubi
, const struct ubi_scan_leb
*seb
,
318 int pnum
, const struct ubi_vid_hdr
*vid_hdr
)
321 int len
, err
, second_is_newer
, bitflips
= 0, corrupted
= 0;
322 uint32_t data_crc
, crc
;
323 struct ubi_vid_hdr
*vh
= NULL
;
324 unsigned long long sqnum2
= be64_to_cpu(vid_hdr
->sqnum
);
326 if (sqnum2
== seb
->sqnum
) {
328 * This must be a really ancient UBI image which has been
329 * created before sequence numbers support has been added. At
330 * that times we used 32-bit LEB versions stored in logical
331 * eraseblocks. That was before UBI got into mainline. We do not
332 * support these images anymore. Well, those images still work,
333 * but only if no unclean reboots happened.
335 ubi_err("unsupported on-flash UBI format\n");
339 /* Obviously the LEB with lower sequence counter is older */
340 second_is_newer
= !!(sqnum2
> seb
->sqnum
);
343 * Now we know which copy is newer. If the copy flag of the PEB with
344 * newer version is not set, then we just return, otherwise we have to
345 * check data CRC. For the second PEB we already have the VID header,
346 * for the first one - we'll need to re-read it from flash.
348 * Note: this may be optimized so that we wouldn't read twice.
351 if (second_is_newer
) {
352 if (!vid_hdr
->copy_flag
) {
353 /* It is not a copy, so it is newer */
354 dbg_bld("second PEB %d is newer, copy_flag is unset",
359 if (!seb
->copy_flag
) {
360 /* It is not a copy, so it is newer */
361 dbg_bld("first PEB %d is newer, copy_flag is unset",
363 return bitflips
<< 1;
366 vh
= ubi_zalloc_vid_hdr(ubi
, GFP_KERNEL
);
371 err
= ubi_io_read_vid_hdr(ubi
, pnum
, vh
, 0);
373 if (err
== UBI_IO_BITFLIPS
)
376 dbg_err("VID of PEB %d header is bad, but it "
377 "was OK earlier, err %d", pnum
, err
);
388 /* Read the data of the copy and check the CRC */
390 len
= be32_to_cpu(vid_hdr
->data_size
);
397 err
= ubi_io_read_data(ubi
, buf
, pnum
, 0, len
);
398 if (err
&& err
!= UBI_IO_BITFLIPS
&& !mtd_is_eccerr(err
))
401 data_crc
= be32_to_cpu(vid_hdr
->data_crc
);
402 crc
= crc32(UBI_CRC32_INIT
, buf
, len
);
403 if (crc
!= data_crc
) {
404 dbg_bld("PEB %d CRC error: calculated %#08x, must be %#08x",
405 pnum
, crc
, data_crc
);
408 second_is_newer
= !second_is_newer
;
410 dbg_bld("PEB %d CRC is OK", pnum
);
415 ubi_free_vid_hdr(ubi
, vh
);
418 dbg_bld("second PEB %d is newer, copy_flag is set", pnum
);
420 dbg_bld("first PEB %d is newer, copy_flag is set", pnum
);
422 return second_is_newer
| (bitflips
<< 1) | (corrupted
<< 2);
427 ubi_free_vid_hdr(ubi
, vh
);
432 * ubi_scan_add_used - add physical eraseblock to the scanning information.
433 * @ubi: UBI device description object
434 * @si: scanning information
435 * @pnum: the physical eraseblock number
437 * @vid_hdr: the volume identifier header
438 * @bitflips: if bit-flips were detected when this physical eraseblock was read
440 * This function adds information about a used physical eraseblock to the
441 * 'used' tree of the corresponding volume. The function is rather complex
442 * because it has to handle cases when this is not the first physical
443 * eraseblock belonging to the same logical eraseblock, and the newer one has
444 * to be picked, while the older one has to be dropped. This function returns
445 * zero in case of success and a negative error code in case of failure.
447 int ubi_scan_add_used(struct ubi_device
*ubi
, struct ubi_scan_info
*si
,
448 int pnum
, int ec
, const struct ubi_vid_hdr
*vid_hdr
,
451 int err
, vol_id
, lnum
;
452 unsigned long long sqnum
;
453 struct ubi_scan_volume
*sv
;
454 struct ubi_scan_leb
*seb
;
455 struct rb_node
**p
, *parent
= NULL
;
457 vol_id
= be32_to_cpu(vid_hdr
->vol_id
);
458 lnum
= be32_to_cpu(vid_hdr
->lnum
);
459 sqnum
= be64_to_cpu(vid_hdr
->sqnum
);
461 dbg_bld("PEB %d, LEB %d:%d, EC %d, sqnum %llu, bitflips %d",
462 pnum
, vol_id
, lnum
, ec
, sqnum
, bitflips
);
464 sv
= add_volume(si
, vol_id
, pnum
, vid_hdr
);
468 if (si
->max_sqnum
< sqnum
)
469 si
->max_sqnum
= sqnum
;
472 * Walk the RB-tree of logical eraseblocks of volume @vol_id to look
473 * if this is the first instance of this logical eraseblock or not.
475 p
= &sv
->root
.rb_node
;
480 seb
= rb_entry(parent
, struct ubi_scan_leb
, u
.rb
);
481 if (lnum
!= seb
->lnum
) {
482 if (lnum
< seb
->lnum
)
490 * There is already a physical eraseblock describing the same
491 * logical eraseblock present.
494 dbg_bld("this LEB already exists: PEB %d, sqnum %llu, "
495 "EC %d", seb
->pnum
, seb
->sqnum
, seb
->ec
);
498 * Make sure that the logical eraseblocks have different
499 * sequence numbers. Otherwise the image is bad.
501 * However, if the sequence number is zero, we assume it must
502 * be an ancient UBI image from the era when UBI did not have
503 * sequence numbers. We still can attach these images, unless
504 * there is a need to distinguish between old and new
505 * eraseblocks, in which case we'll refuse the image in
506 * 'compare_lebs()'. In other words, we attach old clean
507 * images, but refuse attaching old images with duplicated
508 * logical eraseblocks because there was an unclean reboot.
510 if (seb
->sqnum
== sqnum
&& sqnum
!= 0) {
511 ubi_err("two LEBs with same sequence number %llu",
513 ubi_dbg_dump_seb(seb
, 0);
514 ubi_dbg_dump_vid_hdr(vid_hdr
);
519 * Now we have to drop the older one and preserve the newer
522 cmp_res
= compare_lebs(ubi
, seb
, pnum
, vid_hdr
);
528 * This logical eraseblock is newer than the one
531 err
= validate_vid_hdr(vid_hdr
, sv
, pnum
);
535 err
= add_to_list(si
, seb
->pnum
, seb
->ec
, cmp_res
& 4,
542 seb
->scrub
= ((cmp_res
& 2) || bitflips
);
543 seb
->copy_flag
= vid_hdr
->copy_flag
;
546 if (sv
->highest_lnum
== lnum
)
548 be32_to_cpu(vid_hdr
->data_size
);
553 * This logical eraseblock is older than the one found
556 return add_to_list(si
, pnum
, ec
, cmp_res
& 4,
562 * We've met this logical eraseblock for the first time, add it to the
563 * scanning information.
566 err
= validate_vid_hdr(vid_hdr
, sv
, pnum
);
570 seb
= kmem_cache_alloc(si
->scan_leb_slab
, GFP_KERNEL
);
577 seb
->scrub
= bitflips
;
578 seb
->copy_flag
= vid_hdr
->copy_flag
;
581 if (sv
->highest_lnum
<= lnum
) {
582 sv
->highest_lnum
= lnum
;
583 sv
->last_data_size
= be32_to_cpu(vid_hdr
->data_size
);
587 rb_link_node(&seb
->u
.rb
, parent
, p
);
588 rb_insert_color(&seb
->u
.rb
, &sv
->root
);
593 * ubi_scan_find_sv - find volume in the scanning information.
594 * @si: scanning information
595 * @vol_id: the requested volume ID
597 * This function returns a pointer to the volume description or %NULL if there
598 * are no data about this volume in the scanning information.
600 struct ubi_scan_volume
*ubi_scan_find_sv(const struct ubi_scan_info
*si
,
603 struct ubi_scan_volume
*sv
;
604 struct rb_node
*p
= si
->volumes
.rb_node
;
607 sv
= rb_entry(p
, struct ubi_scan_volume
, rb
);
609 if (vol_id
== sv
->vol_id
)
612 if (vol_id
> sv
->vol_id
)
622 * ubi_scan_find_seb - find LEB in the volume scanning information.
623 * @sv: a pointer to the volume scanning information
624 * @lnum: the requested logical eraseblock
626 * This function returns a pointer to the scanning logical eraseblock or %NULL
627 * if there are no data about it in the scanning volume information.
629 struct ubi_scan_leb
*ubi_scan_find_seb(const struct ubi_scan_volume
*sv
,
632 struct ubi_scan_leb
*seb
;
633 struct rb_node
*p
= sv
->root
.rb_node
;
636 seb
= rb_entry(p
, struct ubi_scan_leb
, u
.rb
);
638 if (lnum
== seb
->lnum
)
641 if (lnum
> seb
->lnum
)
651 * ubi_scan_rm_volume - delete scanning information about a volume.
652 * @si: scanning information
653 * @sv: the volume scanning information to delete
655 void ubi_scan_rm_volume(struct ubi_scan_info
*si
, struct ubi_scan_volume
*sv
)
658 struct ubi_scan_leb
*seb
;
660 dbg_bld("remove scanning information about volume %d", sv
->vol_id
);
662 while ((rb
= rb_first(&sv
->root
))) {
663 seb
= rb_entry(rb
, struct ubi_scan_leb
, u
.rb
);
664 rb_erase(&seb
->u
.rb
, &sv
->root
);
665 list_add_tail(&seb
->u
.list
, &si
->erase
);
668 rb_erase(&sv
->rb
, &si
->volumes
);
674 * ubi_scan_erase_peb - erase a physical eraseblock.
675 * @ubi: UBI device description object
676 * @si: scanning information
677 * @pnum: physical eraseblock number to erase;
678 * @ec: erase counter value to write (%UBI_SCAN_UNKNOWN_EC if it is unknown)
680 * This function erases physical eraseblock 'pnum', and writes the erase
681 * counter header to it. This function should only be used on UBI device
682 * initialization stages, when the EBA sub-system had not been yet initialized.
683 * This function returns zero in case of success and a negative error code in
686 int ubi_scan_erase_peb(struct ubi_device
*ubi
, const struct ubi_scan_info
*si
,
690 struct ubi_ec_hdr
*ec_hdr
;
692 if ((long long)ec
>= UBI_MAX_ERASECOUNTER
) {
694 * Erase counter overflow. Upgrade UBI and use 64-bit
695 * erase counters internally.
697 ubi_err("erase counter overflow at PEB %d, EC %d", pnum
, ec
);
701 ec_hdr
= kzalloc(ubi
->ec_hdr_alsize
, GFP_KERNEL
);
705 ec_hdr
->ec
= cpu_to_be64(ec
);
707 err
= ubi_io_sync_erase(ubi
, pnum
, 0);
711 err
= ubi_io_write_ec_hdr(ubi
, pnum
, ec_hdr
);
719 * ubi_scan_get_free_peb - get a free physical eraseblock.
720 * @ubi: UBI device description object
721 * @si: scanning information
723 * This function returns a free physical eraseblock. It is supposed to be
724 * called on the UBI initialization stages when the wear-leveling sub-system is
725 * not initialized yet. This function picks a physical eraseblocks from one of
726 * the lists, writes the EC header if it is needed, and removes it from the
729 * This function returns scanning physical eraseblock information in case of
730 * success and an error code in case of failure.
732 struct ubi_scan_leb
*ubi_scan_get_free_peb(struct ubi_device
*ubi
,
733 struct ubi_scan_info
*si
)
736 struct ubi_scan_leb
*seb
, *tmp_seb
;
738 if (!list_empty(&si
->free
)) {
739 seb
= list_entry(si
->free
.next
, struct ubi_scan_leb
, u
.list
);
740 list_del(&seb
->u
.list
);
741 dbg_bld("return free PEB %d, EC %d", seb
->pnum
, seb
->ec
);
746 * We try to erase the first physical eraseblock from the erase list
747 * and pick it if we succeed, or try to erase the next one if not. And
748 * so forth. We don't want to take care about bad eraseblocks here -
749 * they'll be handled later.
751 list_for_each_entry_safe(seb
, tmp_seb
, &si
->erase
, u
.list
) {
752 if (seb
->ec
== UBI_SCAN_UNKNOWN_EC
)
753 seb
->ec
= si
->mean_ec
;
755 err
= ubi_scan_erase_peb(ubi
, si
, seb
->pnum
, seb
->ec
+1);
760 list_del(&seb
->u
.list
);
761 dbg_bld("return PEB %d, EC %d", seb
->pnum
, seb
->ec
);
765 ubi_err("no free eraseblocks");
766 return ERR_PTR(-ENOSPC
);
770 * check_corruption - check the data area of PEB.
771 * @ubi: UBI device description object
772 * @vid_hrd: the (corrupted) VID header of this PEB
773 * @pnum: the physical eraseblock number to check
775 * This is a helper function which is used to distinguish between VID header
776 * corruptions caused by power cuts and other reasons. If the PEB contains only
777 * 0xFF bytes in the data area, the VID header is most probably corrupted
778 * because of a power cut (%0 is returned in this case). Otherwise, it was
779 * probably corrupted for some other reasons (%1 is returned in this case). A
780 * negative error code is returned if a read error occurred.
782 * If the corruption reason was a power cut, UBI can safely erase this PEB.
783 * Otherwise, it should preserve it to avoid possibly destroying important
786 static int check_corruption(struct ubi_device
*ubi
, struct ubi_vid_hdr
*vid_hdr
,
791 mutex_lock(&ubi
->buf_mutex
);
792 memset(ubi
->peb_buf1
, 0x00, ubi
->leb_size
);
794 err
= ubi_io_read(ubi
, ubi
->peb_buf1
, pnum
, ubi
->leb_start
,
796 if (err
== UBI_IO_BITFLIPS
|| mtd_is_eccerr(err
)) {
798 * Bit-flips or integrity errors while reading the data area.
799 * It is difficult to say for sure what type of corruption is
800 * this, but presumably a power cut happened while this PEB was
801 * erased, so it became unstable and corrupted, and should be
811 if (ubi_check_pattern(ubi
->peb_buf1
, 0xFF, ubi
->leb_size
))
814 ubi_err("PEB %d contains corrupted VID header, and the data does not "
815 "contain all 0xFF, this may be a non-UBI PEB or a severe VID "
816 "header corruption which requires manual inspection", pnum
);
817 ubi_dbg_dump_vid_hdr(vid_hdr
);
818 dbg_msg("hexdump of PEB %d offset %d, length %d",
819 pnum
, ubi
->leb_start
, ubi
->leb_size
);
820 ubi_dbg_print_hex_dump(KERN_DEBUG
, "", DUMP_PREFIX_OFFSET
, 32, 1,
821 ubi
->peb_buf1
, ubi
->leb_size
, 1);
825 mutex_unlock(&ubi
->buf_mutex
);
830 * process_eb - read, check UBI headers, and add them to scanning information.
831 * @ubi: UBI device description object
832 * @si: scanning information
833 * @pnum: the physical eraseblock number
835 * This function returns a zero if the physical eraseblock was successfully
836 * handled and a negative error code in case of failure.
838 static int process_eb(struct ubi_device
*ubi
, struct ubi_scan_info
*si
,
841 long long uninitialized_var(ec
);
842 int err
, bitflips
= 0, vol_id
, ec_err
= 0;
844 dbg_bld("scan PEB %d", pnum
);
846 /* Skip bad physical eraseblocks */
847 err
= ubi_io_is_bad(ubi
, pnum
);
852 * FIXME: this is actually duty of the I/O sub-system to
853 * initialize this, but MTD does not provide enough
856 si
->bad_peb_count
+= 1;
860 err
= ubi_io_read_ec_hdr(ubi
, pnum
, ech
, 0);
866 case UBI_IO_BITFLIPS
:
870 si
->empty_peb_count
+= 1;
871 return add_to_list(si
, pnum
, UBI_SCAN_UNKNOWN_EC
, 0,
873 case UBI_IO_FF_BITFLIPS
:
874 si
->empty_peb_count
+= 1;
875 return add_to_list(si
, pnum
, UBI_SCAN_UNKNOWN_EC
, 1,
877 case UBI_IO_BAD_HDR_EBADMSG
:
880 * We have to also look at the VID header, possibly it is not
881 * corrupted. Set %bitflips flag in order to make this PEB be
882 * moved and EC be re-created.
885 ec
= UBI_SCAN_UNKNOWN_EC
;
889 ubi_err("'ubi_io_read_ec_hdr()' returned unknown code %d", err
);
896 /* Make sure UBI version is OK */
897 if (ech
->version
!= UBI_VERSION
) {
898 ubi_err("this UBI version is %d, image version is %d",
899 UBI_VERSION
, (int)ech
->version
);
903 ec
= be64_to_cpu(ech
->ec
);
904 if (ec
> UBI_MAX_ERASECOUNTER
) {
906 * Erase counter overflow. The EC headers have 64 bits
907 * reserved, but we anyway make use of only 31 bit
908 * values, as this seems to be enough for any existing
909 * flash. Upgrade UBI and use 64-bit erase counters
912 ubi_err("erase counter overflow, max is %d",
913 UBI_MAX_ERASECOUNTER
);
914 ubi_dbg_dump_ec_hdr(ech
);
919 * Make sure that all PEBs have the same image sequence number.
920 * This allows us to detect situations when users flash UBI
921 * images incorrectly, so that the flash has the new UBI image
922 * and leftovers from the old one. This feature was added
923 * relatively recently, and the sequence number was always
924 * zero, because old UBI implementations always set it to zero.
925 * For this reasons, we do not panic if some PEBs have zero
926 * sequence number, while other PEBs have non-zero sequence
929 image_seq
= be32_to_cpu(ech
->image_seq
);
930 if (!ubi
->image_seq
&& image_seq
)
931 ubi
->image_seq
= image_seq
;
932 if (ubi
->image_seq
&& image_seq
&&
933 ubi
->image_seq
!= image_seq
) {
934 ubi_err("bad image sequence number %d in PEB %d, "
935 "expected %d", image_seq
, pnum
, ubi
->image_seq
);
936 ubi_dbg_dump_ec_hdr(ech
);
941 /* OK, we've done with the EC header, let's look at the VID header */
943 err
= ubi_io_read_vid_hdr(ubi
, pnum
, vidh
, 0);
949 case UBI_IO_BITFLIPS
:
952 case UBI_IO_BAD_HDR_EBADMSG
:
953 if (ec_err
== UBI_IO_BAD_HDR_EBADMSG
)
955 * Both EC and VID headers are corrupted and were read
956 * with data integrity error, probably this is a bad
957 * PEB, bit it is not marked as bad yet. This may also
958 * be a result of power cut during erasure.
960 si
->maybe_bad_peb_count
+= 1;
964 * Both headers are corrupted. There is a possibility
965 * that this a valid UBI PEB which has corresponding
966 * LEB, but the headers are corrupted. However, it is
967 * impossible to distinguish it from a PEB which just
968 * contains garbage because of a power cut during erase
969 * operation. So we just schedule this PEB for erasure.
971 * Besides, in case of NOR flash, we deliberately
972 * corrupt both headers because NOR flash erasure is
973 * slow and can start from the end.
978 * The EC was OK, but the VID header is corrupted. We
979 * have to check what is in the data area.
981 err
= check_corruption(ubi
, vidh
, pnum
);
986 /* This corruption is caused by a power cut */
987 err
= add_to_list(si
, pnum
, ec
, 1, &si
->erase
);
989 /* This is an unexpected corruption */
990 err
= add_corrupted(si
, pnum
, ec
);
994 case UBI_IO_FF_BITFLIPS
:
995 err
= add_to_list(si
, pnum
, ec
, 1, &si
->erase
);
1001 err
= add_to_list(si
, pnum
, ec
, 1, &si
->erase
);
1003 err
= add_to_list(si
, pnum
, ec
, 0, &si
->free
);
1006 goto adjust_mean_ec
;
1008 ubi_err("'ubi_io_read_vid_hdr()' returned unknown code %d",
1013 vol_id
= be32_to_cpu(vidh
->vol_id
);
1014 if (vol_id
> UBI_MAX_VOLUMES
&& vol_id
!= UBI_LAYOUT_VOLUME_ID
) {
1015 int lnum
= be32_to_cpu(vidh
->lnum
);
1017 /* Unsupported internal volume */
1018 switch (vidh
->compat
) {
1019 case UBI_COMPAT_DELETE
:
1020 ubi_msg("\"delete\" compatible internal volume %d:%d"
1021 " found, will remove it", vol_id
, lnum
);
1022 err
= add_to_list(si
, pnum
, ec
, 1, &si
->erase
);
1028 ubi_msg("read-only compatible internal volume %d:%d"
1029 " found, switch to read-only mode",
1034 case UBI_COMPAT_PRESERVE
:
1035 ubi_msg("\"preserve\" compatible internal volume %d:%d"
1036 " found", vol_id
, lnum
);
1037 err
= add_to_list(si
, pnum
, ec
, 0, &si
->alien
);
1042 case UBI_COMPAT_REJECT
:
1043 ubi_err("incompatible internal volume %d:%d found",
1050 ubi_warn("valid VID header but corrupted EC header at PEB %d",
1052 err
= ubi_scan_add_used(ubi
, si
, pnum
, ec
, vidh
, bitflips
);
1060 if (ec
> si
->max_ec
)
1062 if (ec
< si
->min_ec
)
1070 * check_what_we_have - check what PEB were found by scanning.
1071 * @ubi: UBI device description object
1072 * @si: scanning information
1074 * This is a helper function which takes a look what PEBs were found by
1075 * scanning, and decides whether the flash is empty and should be formatted and
1076 * whether there are too many corrupted PEBs and we should not attach this
1077 * MTD device. Returns zero if we should proceed with attaching the MTD device,
1078 * and %-EINVAL if we should not.
1080 static int check_what_we_have(struct ubi_device
*ubi
, struct ubi_scan_info
*si
)
1082 struct ubi_scan_leb
*seb
;
1083 int max_corr
, peb_count
;
1085 peb_count
= ubi
->peb_count
- si
->bad_peb_count
- si
->alien_peb_count
;
1086 max_corr
= peb_count
/ 20 ?: 8;
1089 * Few corrupted PEBs is not a problem and may be just a result of
1090 * unclean reboots. However, many of them may indicate some problems
1091 * with the flash HW or driver.
1093 if (si
->corr_peb_count
) {
1094 ubi_err("%d PEBs are corrupted and preserved",
1095 si
->corr_peb_count
);
1096 printk(KERN_ERR
"Corrupted PEBs are:");
1097 list_for_each_entry(seb
, &si
->corr
, u
.list
)
1098 printk(KERN_CONT
" %d", seb
->pnum
);
1099 printk(KERN_CONT
"\n");
1102 * If too many PEBs are corrupted, we refuse attaching,
1103 * otherwise, only print a warning.
1105 if (si
->corr_peb_count
>= max_corr
) {
1106 ubi_err("too many corrupted PEBs, refusing");
1111 if (si
->empty_peb_count
+ si
->maybe_bad_peb_count
== peb_count
) {
1113 * All PEBs are empty, or almost all - a couple PEBs look like
1114 * they may be bad PEBs which were not marked as bad yet.
1116 * This piece of code basically tries to distinguish between
1117 * the following situations:
1119 * 1. Flash is empty, but there are few bad PEBs, which are not
1120 * marked as bad so far, and which were read with error. We
1121 * want to go ahead and format this flash. While formatting,
1122 * the faulty PEBs will probably be marked as bad.
1124 * 2. Flash contains non-UBI data and we do not want to format
1125 * it and destroy possibly important information.
1127 if (si
->maybe_bad_peb_count
<= 2) {
1129 ubi_msg("empty MTD device detected");
1130 get_random_bytes(&ubi
->image_seq
,
1131 sizeof(ubi
->image_seq
));
1133 ubi_err("MTD device is not UBI-formatted and possibly "
1134 "contains non-UBI data - refusing it");
1144 * ubi_scan - scan an MTD device.
1145 * @ubi: UBI device description object
1147 * This function does full scanning of an MTD device and returns complete
1148 * information about it. In case of failure, an error code is returned.
1150 struct ubi_scan_info
*ubi_scan(struct ubi_device
*ubi
)
1153 struct rb_node
*rb1
, *rb2
;
1154 struct ubi_scan_volume
*sv
;
1155 struct ubi_scan_leb
*seb
;
1156 struct ubi_scan_info
*si
;
1158 si
= kzalloc(sizeof(struct ubi_scan_info
), GFP_KERNEL
);
1160 return ERR_PTR(-ENOMEM
);
1162 INIT_LIST_HEAD(&si
->corr
);
1163 INIT_LIST_HEAD(&si
->free
);
1164 INIT_LIST_HEAD(&si
->erase
);
1165 INIT_LIST_HEAD(&si
->alien
);
1166 si
->volumes
= RB_ROOT
;
1169 si
->scan_leb_slab
= kmem_cache_create("ubi_scan_leb_slab",
1170 sizeof(struct ubi_scan_leb
),
1172 if (!si
->scan_leb_slab
)
1175 ech
= kzalloc(ubi
->ec_hdr_alsize
, GFP_KERNEL
);
1179 vidh
= ubi_zalloc_vid_hdr(ubi
, GFP_KERNEL
);
1183 for (pnum
= 0; pnum
< ubi
->peb_count
; pnum
++) {
1186 dbg_gen("process PEB %d", pnum
);
1187 err
= process_eb(ubi
, si
, pnum
);
1192 dbg_msg("scanning is finished");
1194 /* Calculate mean erase counter */
1196 si
->mean_ec
= div_u64(si
->ec_sum
, si
->ec_count
);
1198 err
= check_what_we_have(ubi
, si
);
1203 * In case of unknown erase counter we use the mean erase counter
1206 ubi_rb_for_each_entry(rb1
, sv
, &si
->volumes
, rb
) {
1207 ubi_rb_for_each_entry(rb2
, seb
, &sv
->root
, u
.rb
)
1208 if (seb
->ec
== UBI_SCAN_UNKNOWN_EC
)
1209 seb
->ec
= si
->mean_ec
;
1212 list_for_each_entry(seb
, &si
->free
, u
.list
) {
1213 if (seb
->ec
== UBI_SCAN_UNKNOWN_EC
)
1214 seb
->ec
= si
->mean_ec
;
1217 list_for_each_entry(seb
, &si
->corr
, u
.list
)
1218 if (seb
->ec
== UBI_SCAN_UNKNOWN_EC
)
1219 seb
->ec
= si
->mean_ec
;
1221 list_for_each_entry(seb
, &si
->erase
, u
.list
)
1222 if (seb
->ec
== UBI_SCAN_UNKNOWN_EC
)
1223 seb
->ec
= si
->mean_ec
;
1225 err
= paranoid_check_si(ubi
, si
);
1229 ubi_free_vid_hdr(ubi
, vidh
);
1235 ubi_free_vid_hdr(ubi
, vidh
);
1239 ubi_scan_destroy_si(si
);
1240 return ERR_PTR(err
);
1244 * destroy_sv - free the scanning volume information
1245 * @sv: scanning volume information
1246 * @si: scanning information
1248 * This function destroys the volume RB-tree (@sv->root) and the scanning
1249 * volume information.
1251 static void destroy_sv(struct ubi_scan_info
*si
, struct ubi_scan_volume
*sv
)
1253 struct ubi_scan_leb
*seb
;
1254 struct rb_node
*this = sv
->root
.rb_node
;
1258 this = this->rb_left
;
1259 else if (this->rb_right
)
1260 this = this->rb_right
;
1262 seb
= rb_entry(this, struct ubi_scan_leb
, u
.rb
);
1263 this = rb_parent(this);
1265 if (this->rb_left
== &seb
->u
.rb
)
1266 this->rb_left
= NULL
;
1268 this->rb_right
= NULL
;
1271 kmem_cache_free(si
->scan_leb_slab
, seb
);
1278 * ubi_scan_destroy_si - destroy scanning information.
1279 * @si: scanning information
1281 void ubi_scan_destroy_si(struct ubi_scan_info
*si
)
1283 struct ubi_scan_leb
*seb
, *seb_tmp
;
1284 struct ubi_scan_volume
*sv
;
1287 list_for_each_entry_safe(seb
, seb_tmp
, &si
->alien
, u
.list
) {
1288 list_del(&seb
->u
.list
);
1289 kmem_cache_free(si
->scan_leb_slab
, seb
);
1291 list_for_each_entry_safe(seb
, seb_tmp
, &si
->erase
, u
.list
) {
1292 list_del(&seb
->u
.list
);
1293 kmem_cache_free(si
->scan_leb_slab
, seb
);
1295 list_for_each_entry_safe(seb
, seb_tmp
, &si
->corr
, u
.list
) {
1296 list_del(&seb
->u
.list
);
1297 kmem_cache_free(si
->scan_leb_slab
, seb
);
1299 list_for_each_entry_safe(seb
, seb_tmp
, &si
->free
, u
.list
) {
1300 list_del(&seb
->u
.list
);
1301 kmem_cache_free(si
->scan_leb_slab
, seb
);
1304 /* Destroy the volume RB-tree */
1305 rb
= si
->volumes
.rb_node
;
1309 else if (rb
->rb_right
)
1312 sv
= rb_entry(rb
, struct ubi_scan_volume
, rb
);
1316 if (rb
->rb_left
== &sv
->rb
)
1319 rb
->rb_right
= NULL
;
1326 if (si
->scan_leb_slab
)
1327 kmem_cache_destroy(si
->scan_leb_slab
);
1332 #ifdef CONFIG_MTD_UBI_DEBUG
1335 * paranoid_check_si - check the scanning information.
1336 * @ubi: UBI device description object
1337 * @si: scanning information
1339 * This function returns zero if the scanning information is all right, and a
1340 * negative error code if not or if an error occurred.
1342 static int paranoid_check_si(struct ubi_device
*ubi
, struct ubi_scan_info
*si
)
1344 int pnum
, err
, vols_found
= 0;
1345 struct rb_node
*rb1
, *rb2
;
1346 struct ubi_scan_volume
*sv
;
1347 struct ubi_scan_leb
*seb
, *last_seb
;
1350 if (!ubi
->dbg
->chk_gen
)
1354 * At first, check that scanning information is OK.
1356 ubi_rb_for_each_entry(rb1
, sv
, &si
->volumes
, rb
) {
1364 ubi_err("bad is_empty flag");
1368 if (sv
->vol_id
< 0 || sv
->highest_lnum
< 0 ||
1369 sv
->leb_count
< 0 || sv
->vol_type
< 0 || sv
->used_ebs
< 0 ||
1370 sv
->data_pad
< 0 || sv
->last_data_size
< 0) {
1371 ubi_err("negative values");
1375 if (sv
->vol_id
>= UBI_MAX_VOLUMES
&&
1376 sv
->vol_id
< UBI_INTERNAL_VOL_START
) {
1377 ubi_err("bad vol_id");
1381 if (sv
->vol_id
> si
->highest_vol_id
) {
1382 ubi_err("highest_vol_id is %d, but vol_id %d is there",
1383 si
->highest_vol_id
, sv
->vol_id
);
1387 if (sv
->vol_type
!= UBI_DYNAMIC_VOLUME
&&
1388 sv
->vol_type
!= UBI_STATIC_VOLUME
) {
1389 ubi_err("bad vol_type");
1393 if (sv
->data_pad
> ubi
->leb_size
/ 2) {
1394 ubi_err("bad data_pad");
1399 ubi_rb_for_each_entry(rb2
, seb
, &sv
->root
, u
.rb
) {
1405 if (seb
->pnum
< 0 || seb
->ec
< 0) {
1406 ubi_err("negative values");
1410 if (seb
->ec
< si
->min_ec
) {
1411 ubi_err("bad si->min_ec (%d), %d found",
1412 si
->min_ec
, seb
->ec
);
1416 if (seb
->ec
> si
->max_ec
) {
1417 ubi_err("bad si->max_ec (%d), %d found",
1418 si
->max_ec
, seb
->ec
);
1422 if (seb
->pnum
>= ubi
->peb_count
) {
1423 ubi_err("too high PEB number %d, total PEBs %d",
1424 seb
->pnum
, ubi
->peb_count
);
1428 if (sv
->vol_type
== UBI_STATIC_VOLUME
) {
1429 if (seb
->lnum
>= sv
->used_ebs
) {
1430 ubi_err("bad lnum or used_ebs");
1434 if (sv
->used_ebs
!= 0) {
1435 ubi_err("non-zero used_ebs");
1440 if (seb
->lnum
> sv
->highest_lnum
) {
1441 ubi_err("incorrect highest_lnum or lnum");
1446 if (sv
->leb_count
!= leb_count
) {
1447 ubi_err("bad leb_count, %d objects in the tree",
1457 if (seb
->lnum
!= sv
->highest_lnum
) {
1458 ubi_err("bad highest_lnum");
1463 if (vols_found
!= si
->vols_found
) {
1464 ubi_err("bad si->vols_found %d, should be %d",
1465 si
->vols_found
, vols_found
);
1469 /* Check that scanning information is correct */
1470 ubi_rb_for_each_entry(rb1
, sv
, &si
->volumes
, rb
) {
1472 ubi_rb_for_each_entry(rb2
, seb
, &sv
->root
, u
.rb
) {
1479 err
= ubi_io_read_vid_hdr(ubi
, seb
->pnum
, vidh
, 1);
1480 if (err
&& err
!= UBI_IO_BITFLIPS
) {
1481 ubi_err("VID header is not OK (%d)", err
);
1487 vol_type
= vidh
->vol_type
== UBI_VID_DYNAMIC
?
1488 UBI_DYNAMIC_VOLUME
: UBI_STATIC_VOLUME
;
1489 if (sv
->vol_type
!= vol_type
) {
1490 ubi_err("bad vol_type");
1494 if (seb
->sqnum
!= be64_to_cpu(vidh
->sqnum
)) {
1495 ubi_err("bad sqnum %llu", seb
->sqnum
);
1499 if (sv
->vol_id
!= be32_to_cpu(vidh
->vol_id
)) {
1500 ubi_err("bad vol_id %d", sv
->vol_id
);
1504 if (sv
->compat
!= vidh
->compat
) {
1505 ubi_err("bad compat %d", vidh
->compat
);
1509 if (seb
->lnum
!= be32_to_cpu(vidh
->lnum
)) {
1510 ubi_err("bad lnum %d", seb
->lnum
);
1514 if (sv
->used_ebs
!= be32_to_cpu(vidh
->used_ebs
)) {
1515 ubi_err("bad used_ebs %d", sv
->used_ebs
);
1519 if (sv
->data_pad
!= be32_to_cpu(vidh
->data_pad
)) {
1520 ubi_err("bad data_pad %d", sv
->data_pad
);
1528 if (sv
->highest_lnum
!= be32_to_cpu(vidh
->lnum
)) {
1529 ubi_err("bad highest_lnum %d", sv
->highest_lnum
);
1533 if (sv
->last_data_size
!= be32_to_cpu(vidh
->data_size
)) {
1534 ubi_err("bad last_data_size %d", sv
->last_data_size
);
1540 * Make sure that all the physical eraseblocks are in one of the lists
1543 buf
= kzalloc(ubi
->peb_count
, GFP_KERNEL
);
1547 for (pnum
= 0; pnum
< ubi
->peb_count
; pnum
++) {
1548 err
= ubi_io_is_bad(ubi
, pnum
);
1556 ubi_rb_for_each_entry(rb1
, sv
, &si
->volumes
, rb
)
1557 ubi_rb_for_each_entry(rb2
, seb
, &sv
->root
, u
.rb
)
1560 list_for_each_entry(seb
, &si
->free
, u
.list
)
1563 list_for_each_entry(seb
, &si
->corr
, u
.list
)
1566 list_for_each_entry(seb
, &si
->erase
, u
.list
)
1569 list_for_each_entry(seb
, &si
->alien
, u
.list
)
1573 for (pnum
= 0; pnum
< ubi
->peb_count
; pnum
++)
1575 ubi_err("PEB %d is not referred", pnum
);
1585 ubi_err("bad scanning information about LEB %d", seb
->lnum
);
1586 ubi_dbg_dump_seb(seb
, 0);
1587 ubi_dbg_dump_sv(sv
);
1591 ubi_err("bad scanning information about volume %d", sv
->vol_id
);
1592 ubi_dbg_dump_sv(sv
);
1596 ubi_err("bad scanning information about volume %d", sv
->vol_id
);
1597 ubi_dbg_dump_sv(sv
);
1598 ubi_dbg_dump_vid_hdr(vidh
);
1601 ubi_dbg_dump_stack();
1605 #endif /* CONFIG_MTD_UBI_DEBUG */