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_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 * Scanned logical eraseblocks are represented by &struct ubi_ainf_peb objects.
33 * These objects are kept in per-volume RB-trees with the root at the
34 * corresponding &struct ubi_ainf_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 static int self_check_si(struct ubi_device
*ubi
, struct ubi_scan_info
*si
);
93 /* Temporary variables used during scanning */
94 static struct ubi_ec_hdr
*ech
;
95 static struct ubi_vid_hdr
*vidh
;
98 * add_to_list - add physical eraseblock to a list.
99 * @si: scanning information
100 * @pnum: physical eraseblock number to add
101 * @ec: erase counter of the physical eraseblock
102 * @to_head: if not zero, add to the head of the list
103 * @list: the list to add to
105 * This function adds physical eraseblock @pnum to free, erase, or alien lists.
106 * If @to_head is not zero, PEB will be added to the head of the list, which
107 * basically means it will be processed first later. E.g., we add corrupted
108 * PEBs (corrupted due to power cuts) to the head of the erase list to make
109 * sure we erase them first and get rid of corruptions ASAP. This function
110 * returns zero in case of success and a negative error code in case of
113 static int add_to_list(struct ubi_scan_info
*si
, int pnum
, int ec
, int to_head
,
114 struct list_head
*list
)
116 struct ubi_ainf_peb
*seb
;
118 if (list
== &si
->free
) {
119 dbg_bld("add to free: PEB %d, EC %d", pnum
, ec
);
120 } else if (list
== &si
->erase
) {
121 dbg_bld("add to erase: PEB %d, EC %d", pnum
, ec
);
122 } else if (list
== &si
->alien
) {
123 dbg_bld("add to alien: PEB %d, EC %d", pnum
, ec
);
124 si
->alien_peb_count
+= 1;
128 seb
= kmem_cache_alloc(si
->scan_leb_slab
, GFP_KERNEL
);
135 list_add(&seb
->u
.list
, list
);
137 list_add_tail(&seb
->u
.list
, list
);
142 * add_corrupted - add a corrupted physical eraseblock.
143 * @si: scanning information
144 * @pnum: physical eraseblock number to add
145 * @ec: erase counter of the physical eraseblock
147 * This function adds corrupted physical eraseblock @pnum to the 'corr' list.
148 * The corruption was presumably not caused by a power cut. Returns zero in
149 * case of success and a negative error code in case of failure.
151 static int add_corrupted(struct ubi_scan_info
*si
, int pnum
, int ec
)
153 struct ubi_ainf_peb
*seb
;
155 dbg_bld("add to corrupted: PEB %d, EC %d", pnum
, ec
);
157 seb
= kmem_cache_alloc(si
->scan_leb_slab
, GFP_KERNEL
);
161 si
->corr_peb_count
+= 1;
164 list_add(&seb
->u
.list
, &si
->corr
);
169 * validate_vid_hdr - check volume identifier header.
170 * @vid_hdr: the volume identifier header to check
171 * @sv: information about the volume this logical eraseblock belongs to
172 * @pnum: physical eraseblock number the VID header came from
174 * This function checks that data stored in @vid_hdr is consistent. Returns
175 * non-zero if an inconsistency was found and zero if not.
177 * Note, UBI does sanity check of everything it reads from the flash media.
178 * Most of the checks are done in the I/O sub-system. Here we check that the
179 * information in the VID header is consistent to the information in other VID
180 * headers of the same volume.
182 static int validate_vid_hdr(const struct ubi_vid_hdr
*vid_hdr
,
183 const struct ubi_ainf_volume
*sv
, int pnum
)
185 int vol_type
= vid_hdr
->vol_type
;
186 int vol_id
= be32_to_cpu(vid_hdr
->vol_id
);
187 int used_ebs
= be32_to_cpu(vid_hdr
->used_ebs
);
188 int data_pad
= be32_to_cpu(vid_hdr
->data_pad
);
190 if (sv
->leb_count
!= 0) {
194 * This is not the first logical eraseblock belonging to this
195 * volume. Ensure that the data in its VID header is consistent
196 * to the data in previous logical eraseblock headers.
199 if (vol_id
!= sv
->vol_id
) {
200 ubi_err("inconsistent vol_id");
204 if (sv
->vol_type
== UBI_STATIC_VOLUME
)
205 sv_vol_type
= UBI_VID_STATIC
;
207 sv_vol_type
= UBI_VID_DYNAMIC
;
209 if (vol_type
!= sv_vol_type
) {
210 ubi_err("inconsistent vol_type");
214 if (used_ebs
!= sv
->used_ebs
) {
215 ubi_err("inconsistent used_ebs");
219 if (data_pad
!= sv
->data_pad
) {
220 ubi_err("inconsistent data_pad");
228 ubi_err("inconsistent VID header at PEB %d", pnum
);
229 ubi_dump_vid_hdr(vid_hdr
);
235 * add_volume - add volume to the scanning information.
236 * @si: scanning information
237 * @vol_id: ID of the volume to add
238 * @pnum: physical eraseblock number
239 * @vid_hdr: volume identifier header
241 * If the volume corresponding to the @vid_hdr logical eraseblock is already
242 * present in the scanning information, this function does nothing. Otherwise
243 * it adds corresponding volume to the scanning information. Returns a pointer
244 * to the scanning volume object in case of success and a negative error code
245 * in case of failure.
247 static struct ubi_ainf_volume
*add_volume(struct ubi_scan_info
*si
, int vol_id
,
249 const struct ubi_vid_hdr
*vid_hdr
)
251 struct ubi_ainf_volume
*sv
;
252 struct rb_node
**p
= &si
->volumes
.rb_node
, *parent
= NULL
;
254 ubi_assert(vol_id
== be32_to_cpu(vid_hdr
->vol_id
));
256 /* Walk the volume RB-tree to look if this volume is already present */
259 sv
= rb_entry(parent
, struct ubi_ainf_volume
, rb
);
261 if (vol_id
== sv
->vol_id
)
264 if (vol_id
> sv
->vol_id
)
270 /* The volume is absent - add it */
271 sv
= kmalloc(sizeof(struct ubi_ainf_volume
), GFP_KERNEL
);
273 return ERR_PTR(-ENOMEM
);
275 sv
->highest_lnum
= sv
->leb_count
= 0;
278 sv
->used_ebs
= be32_to_cpu(vid_hdr
->used_ebs
);
279 sv
->data_pad
= be32_to_cpu(vid_hdr
->data_pad
);
280 sv
->compat
= vid_hdr
->compat
;
281 sv
->vol_type
= vid_hdr
->vol_type
== UBI_VID_DYNAMIC
? UBI_DYNAMIC_VOLUME
283 if (vol_id
> si
->highest_vol_id
)
284 si
->highest_vol_id
= vol_id
;
286 rb_link_node(&sv
->rb
, parent
, p
);
287 rb_insert_color(&sv
->rb
, &si
->volumes
);
289 dbg_bld("added volume %d", vol_id
);
294 * compare_lebs - find out which logical eraseblock is newer.
295 * @ubi: UBI device description object
296 * @seb: first logical eraseblock to compare
297 * @pnum: physical eraseblock number of the second logical eraseblock to
299 * @vid_hdr: volume identifier header of the second logical eraseblock
301 * This function compares 2 copies of a LEB and informs which one is newer. In
302 * case of success this function returns a positive value, in case of failure, a
303 * negative error code is returned. The success return codes use the following
305 * o bit 0 is cleared: the first PEB (described by @seb) is newer than the
306 * second PEB (described by @pnum and @vid_hdr);
307 * o bit 0 is set: the second PEB is newer;
308 * o bit 1 is cleared: no bit-flips were detected in the newer LEB;
309 * o bit 1 is set: bit-flips were detected in the newer LEB;
310 * o bit 2 is cleared: the older LEB is not corrupted;
311 * o bit 2 is set: the older LEB is corrupted.
313 static int compare_lebs(struct ubi_device
*ubi
, const struct ubi_ainf_peb
*seb
,
314 int pnum
, const struct ubi_vid_hdr
*vid_hdr
)
317 int len
, err
, second_is_newer
, bitflips
= 0, corrupted
= 0;
318 uint32_t data_crc
, crc
;
319 struct ubi_vid_hdr
*vh
= NULL
;
320 unsigned long long sqnum2
= be64_to_cpu(vid_hdr
->sqnum
);
322 if (sqnum2
== seb
->sqnum
) {
324 * This must be a really ancient UBI image which has been
325 * created before sequence numbers support has been added. At
326 * that times we used 32-bit LEB versions stored in logical
327 * eraseblocks. That was before UBI got into mainline. We do not
328 * support these images anymore. Well, those images still work,
329 * but only if no unclean reboots happened.
331 ubi_err("unsupported on-flash UBI format\n");
335 /* Obviously the LEB with lower sequence counter is older */
336 second_is_newer
= (sqnum2
> seb
->sqnum
);
339 * Now we know which copy is newer. If the copy flag of the PEB with
340 * newer version is not set, then we just return, otherwise we have to
341 * check data CRC. For the second PEB we already have the VID header,
342 * for the first one - we'll need to re-read it from flash.
344 * Note: this may be optimized so that we wouldn't read twice.
347 if (second_is_newer
) {
348 if (!vid_hdr
->copy_flag
) {
349 /* It is not a copy, so it is newer */
350 dbg_bld("second PEB %d is newer, copy_flag is unset",
355 if (!seb
->copy_flag
) {
356 /* It is not a copy, so it is newer */
357 dbg_bld("first PEB %d is newer, copy_flag is unset",
359 return bitflips
<< 1;
362 vh
= ubi_zalloc_vid_hdr(ubi
, GFP_KERNEL
);
367 err
= ubi_io_read_vid_hdr(ubi
, pnum
, vh
, 0);
369 if (err
== UBI_IO_BITFLIPS
)
372 ubi_err("VID of PEB %d header is bad, but it "
373 "was OK earlier, err %d", pnum
, err
);
384 /* Read the data of the copy and check the CRC */
386 len
= be32_to_cpu(vid_hdr
->data_size
);
393 err
= ubi_io_read_data(ubi
, buf
, pnum
, 0, len
);
394 if (err
&& err
!= UBI_IO_BITFLIPS
&& !mtd_is_eccerr(err
))
397 data_crc
= be32_to_cpu(vid_hdr
->data_crc
);
398 crc
= crc32(UBI_CRC32_INIT
, buf
, len
);
399 if (crc
!= data_crc
) {
400 dbg_bld("PEB %d CRC error: calculated %#08x, must be %#08x",
401 pnum
, crc
, data_crc
);
404 second_is_newer
= !second_is_newer
;
406 dbg_bld("PEB %d CRC is OK", pnum
);
411 ubi_free_vid_hdr(ubi
, vh
);
414 dbg_bld("second PEB %d is newer, copy_flag is set", pnum
);
416 dbg_bld("first PEB %d is newer, copy_flag is set", pnum
);
418 return second_is_newer
| (bitflips
<< 1) | (corrupted
<< 2);
423 ubi_free_vid_hdr(ubi
, vh
);
428 * ubi_scan_add_used - add physical eraseblock to the scanning information.
429 * @ubi: UBI device description object
430 * @si: scanning information
431 * @pnum: the physical eraseblock number
433 * @vid_hdr: the volume identifier header
434 * @bitflips: if bit-flips were detected when this physical eraseblock was read
436 * This function adds information about a used physical eraseblock to the
437 * 'used' tree of the corresponding volume. The function is rather complex
438 * because it has to handle cases when this is not the first physical
439 * eraseblock belonging to the same logical eraseblock, and the newer one has
440 * to be picked, while the older one has to be dropped. This function returns
441 * zero in case of success and a negative error code in case of failure.
443 int ubi_scan_add_used(struct ubi_device
*ubi
, struct ubi_scan_info
*si
,
444 int pnum
, int ec
, const struct ubi_vid_hdr
*vid_hdr
,
447 int err
, vol_id
, lnum
;
448 unsigned long long sqnum
;
449 struct ubi_ainf_volume
*sv
;
450 struct ubi_ainf_peb
*seb
;
451 struct rb_node
**p
, *parent
= NULL
;
453 vol_id
= be32_to_cpu(vid_hdr
->vol_id
);
454 lnum
= be32_to_cpu(vid_hdr
->lnum
);
455 sqnum
= be64_to_cpu(vid_hdr
->sqnum
);
457 dbg_bld("PEB %d, LEB %d:%d, EC %d, sqnum %llu, bitflips %d",
458 pnum
, vol_id
, lnum
, ec
, sqnum
, bitflips
);
460 sv
= add_volume(si
, vol_id
, pnum
, vid_hdr
);
464 if (si
->max_sqnum
< sqnum
)
465 si
->max_sqnum
= sqnum
;
468 * Walk the RB-tree of logical eraseblocks of volume @vol_id to look
469 * if this is the first instance of this logical eraseblock or not.
471 p
= &sv
->root
.rb_node
;
476 seb
= rb_entry(parent
, struct ubi_ainf_peb
, u
.rb
);
477 if (lnum
!= seb
->lnum
) {
478 if (lnum
< seb
->lnum
)
486 * There is already a physical eraseblock describing the same
487 * logical eraseblock present.
490 dbg_bld("this LEB already exists: PEB %d, sqnum %llu, "
491 "EC %d", seb
->pnum
, seb
->sqnum
, seb
->ec
);
494 * Make sure that the logical eraseblocks have different
495 * sequence numbers. Otherwise the image is bad.
497 * However, if the sequence number is zero, we assume it must
498 * be an ancient UBI image from the era when UBI did not have
499 * sequence numbers. We still can attach these images, unless
500 * there is a need to distinguish between old and new
501 * eraseblocks, in which case we'll refuse the image in
502 * 'compare_lebs()'. In other words, we attach old clean
503 * images, but refuse attaching old images with duplicated
504 * logical eraseblocks because there was an unclean reboot.
506 if (seb
->sqnum
== sqnum
&& sqnum
!= 0) {
507 ubi_err("two LEBs with same sequence number %llu",
509 ubi_dump_seb(seb
, 0);
510 ubi_dump_vid_hdr(vid_hdr
);
515 * Now we have to drop the older one and preserve the newer
518 cmp_res
= compare_lebs(ubi
, seb
, pnum
, vid_hdr
);
524 * This logical eraseblock is newer than the one
527 err
= validate_vid_hdr(vid_hdr
, sv
, pnum
);
531 err
= add_to_list(si
, seb
->pnum
, seb
->ec
, cmp_res
& 4,
538 seb
->scrub
= ((cmp_res
& 2) || bitflips
);
539 seb
->copy_flag
= vid_hdr
->copy_flag
;
542 if (sv
->highest_lnum
== lnum
)
544 be32_to_cpu(vid_hdr
->data_size
);
549 * This logical eraseblock is older than the one found
552 return add_to_list(si
, pnum
, ec
, cmp_res
& 4,
558 * We've met this logical eraseblock for the first time, add it to the
559 * scanning information.
562 err
= validate_vid_hdr(vid_hdr
, sv
, pnum
);
566 seb
= kmem_cache_alloc(si
->scan_leb_slab
, GFP_KERNEL
);
573 seb
->scrub
= bitflips
;
574 seb
->copy_flag
= vid_hdr
->copy_flag
;
577 if (sv
->highest_lnum
<= lnum
) {
578 sv
->highest_lnum
= lnum
;
579 sv
->last_data_size
= be32_to_cpu(vid_hdr
->data_size
);
583 rb_link_node(&seb
->u
.rb
, parent
, p
);
584 rb_insert_color(&seb
->u
.rb
, &sv
->root
);
589 * ubi_scan_find_sv - find volume in the scanning information.
590 * @si: scanning information
591 * @vol_id: the requested volume ID
593 * This function returns a pointer to the volume description or %NULL if there
594 * are no data about this volume in the scanning information.
596 struct ubi_ainf_volume
*ubi_scan_find_sv(const struct ubi_scan_info
*si
,
599 struct ubi_ainf_volume
*sv
;
600 struct rb_node
*p
= si
->volumes
.rb_node
;
603 sv
= rb_entry(p
, struct ubi_ainf_volume
, rb
);
605 if (vol_id
== sv
->vol_id
)
608 if (vol_id
> sv
->vol_id
)
618 * ubi_scan_find_seb - find LEB in the volume scanning information.
619 * @sv: a pointer to the volume scanning information
620 * @lnum: the requested logical eraseblock
622 * This function returns a pointer to the scanning logical eraseblock or %NULL
623 * if there are no data about it in the scanning volume information.
625 struct ubi_ainf_peb
*ubi_scan_find_seb(const struct ubi_ainf_volume
*sv
,
628 struct ubi_ainf_peb
*seb
;
629 struct rb_node
*p
= sv
->root
.rb_node
;
632 seb
= rb_entry(p
, struct ubi_ainf_peb
, u
.rb
);
634 if (lnum
== seb
->lnum
)
637 if (lnum
> seb
->lnum
)
647 * ubi_scan_rm_volume - delete scanning information about a volume.
648 * @si: scanning information
649 * @sv: the volume scanning information to delete
651 void ubi_scan_rm_volume(struct ubi_scan_info
*si
, struct ubi_ainf_volume
*sv
)
654 struct ubi_ainf_peb
*seb
;
656 dbg_bld("remove scanning information about volume %d", sv
->vol_id
);
658 while ((rb
= rb_first(&sv
->root
))) {
659 seb
= rb_entry(rb
, struct ubi_ainf_peb
, u
.rb
);
660 rb_erase(&seb
->u
.rb
, &sv
->root
);
661 list_add_tail(&seb
->u
.list
, &si
->erase
);
664 rb_erase(&sv
->rb
, &si
->volumes
);
670 * ubi_scan_erase_peb - erase a physical eraseblock.
671 * @ubi: UBI device description object
672 * @si: scanning information
673 * @pnum: physical eraseblock number to erase;
674 * @ec: erase counter value to write (%UBI_SCAN_UNKNOWN_EC if it is unknown)
676 * This function erases physical eraseblock 'pnum', and writes the erase
677 * counter header to it. This function should only be used on UBI device
678 * initialization stages, when the EBA sub-system had not been yet initialized.
679 * This function returns zero in case of success and a negative error code in
682 int ubi_scan_erase_peb(struct ubi_device
*ubi
, const struct ubi_scan_info
*si
,
686 struct ubi_ec_hdr
*ec_hdr
;
688 if ((long long)ec
>= UBI_MAX_ERASECOUNTER
) {
690 * Erase counter overflow. Upgrade UBI and use 64-bit
691 * erase counters internally.
693 ubi_err("erase counter overflow at PEB %d, EC %d", pnum
, ec
);
697 ec_hdr
= kzalloc(ubi
->ec_hdr_alsize
, GFP_KERNEL
);
701 ec_hdr
->ec
= cpu_to_be64(ec
);
703 err
= ubi_io_sync_erase(ubi
, pnum
, 0);
707 err
= ubi_io_write_ec_hdr(ubi
, pnum
, ec_hdr
);
715 * ubi_scan_get_free_peb - get a free physical eraseblock.
716 * @ubi: UBI device description object
717 * @si: scanning information
719 * This function returns a free physical eraseblock. It is supposed to be
720 * called on the UBI initialization stages when the wear-leveling sub-system is
721 * not initialized yet. This function picks a physical eraseblocks from one of
722 * the lists, writes the EC header if it is needed, and removes it from the
725 * This function returns scanning physical eraseblock information in case of
726 * success and an error code in case of failure.
728 struct ubi_ainf_peb
*ubi_scan_get_free_peb(struct ubi_device
*ubi
,
729 struct ubi_scan_info
*si
)
732 struct ubi_ainf_peb
*seb
, *tmp_seb
;
734 if (!list_empty(&si
->free
)) {
735 seb
= list_entry(si
->free
.next
, struct ubi_ainf_peb
, u
.list
);
736 list_del(&seb
->u
.list
);
737 dbg_bld("return free PEB %d, EC %d", seb
->pnum
, seb
->ec
);
742 * We try to erase the first physical eraseblock from the erase list
743 * and pick it if we succeed, or try to erase the next one if not. And
744 * so forth. We don't want to take care about bad eraseblocks here -
745 * they'll be handled later.
747 list_for_each_entry_safe(seb
, tmp_seb
, &si
->erase
, u
.list
) {
748 if (seb
->ec
== UBI_SCAN_UNKNOWN_EC
)
749 seb
->ec
= si
->mean_ec
;
751 err
= ubi_scan_erase_peb(ubi
, si
, seb
->pnum
, seb
->ec
+1);
756 list_del(&seb
->u
.list
);
757 dbg_bld("return PEB %d, EC %d", seb
->pnum
, seb
->ec
);
761 ubi_err("no free eraseblocks");
762 return ERR_PTR(-ENOSPC
);
766 * check_corruption - check the data area of PEB.
767 * @ubi: UBI device description object
768 * @vid_hrd: the (corrupted) VID header of this PEB
769 * @pnum: the physical eraseblock number to check
771 * This is a helper function which is used to distinguish between VID header
772 * corruptions caused by power cuts and other reasons. If the PEB contains only
773 * 0xFF bytes in the data area, the VID header is most probably corrupted
774 * because of a power cut (%0 is returned in this case). Otherwise, it was
775 * probably corrupted for some other reasons (%1 is returned in this case). A
776 * negative error code is returned if a read error occurred.
778 * If the corruption reason was a power cut, UBI can safely erase this PEB.
779 * Otherwise, it should preserve it to avoid possibly destroying important
782 static int check_corruption(struct ubi_device
*ubi
, struct ubi_vid_hdr
*vid_hdr
,
787 mutex_lock(&ubi
->buf_mutex
);
788 memset(ubi
->peb_buf
, 0x00, ubi
->leb_size
);
790 err
= ubi_io_read(ubi
, ubi
->peb_buf
, pnum
, ubi
->leb_start
,
792 if (err
== UBI_IO_BITFLIPS
|| mtd_is_eccerr(err
)) {
794 * Bit-flips or integrity errors while reading the data area.
795 * It is difficult to say for sure what type of corruption is
796 * this, but presumably a power cut happened while this PEB was
797 * erased, so it became unstable and corrupted, and should be
807 if (ubi_check_pattern(ubi
->peb_buf
, 0xFF, ubi
->leb_size
))
810 ubi_err("PEB %d contains corrupted VID header, and the data does not "
811 "contain all 0xFF, this may be a non-UBI PEB or a severe VID "
812 "header corruption which requires manual inspection", pnum
);
813 ubi_dump_vid_hdr(vid_hdr
);
814 dbg_msg("hexdump of PEB %d offset %d, length %d",
815 pnum
, ubi
->leb_start
, ubi
->leb_size
);
816 ubi_dbg_print_hex_dump(KERN_DEBUG
, "", DUMP_PREFIX_OFFSET
, 32, 1,
817 ubi
->peb_buf
, ubi
->leb_size
, 1);
821 mutex_unlock(&ubi
->buf_mutex
);
826 * process_eb - read, check UBI headers, and add them to scanning information.
827 * @ubi: UBI device description object
828 * @si: scanning information
829 * @pnum: the physical eraseblock number
831 * This function returns a zero if the physical eraseblock was successfully
832 * handled and a negative error code in case of failure.
834 static int process_eb(struct ubi_device
*ubi
, struct ubi_scan_info
*si
,
837 long long uninitialized_var(ec
);
838 int err
, bitflips
= 0, vol_id
, ec_err
= 0;
840 dbg_bld("scan PEB %d", pnum
);
842 /* Skip bad physical eraseblocks */
843 err
= ubi_io_is_bad(ubi
, pnum
);
848 * FIXME: this is actually duty of the I/O sub-system to
849 * initialize this, but MTD does not provide enough
852 si
->bad_peb_count
+= 1;
856 err
= ubi_io_read_ec_hdr(ubi
, pnum
, ech
, 0);
862 case UBI_IO_BITFLIPS
:
866 si
->empty_peb_count
+= 1;
867 return add_to_list(si
, pnum
, UBI_SCAN_UNKNOWN_EC
, 0,
869 case UBI_IO_FF_BITFLIPS
:
870 si
->empty_peb_count
+= 1;
871 return add_to_list(si
, pnum
, UBI_SCAN_UNKNOWN_EC
, 1,
873 case UBI_IO_BAD_HDR_EBADMSG
:
876 * We have to also look at the VID header, possibly it is not
877 * corrupted. Set %bitflips flag in order to make this PEB be
878 * moved and EC be re-created.
881 ec
= UBI_SCAN_UNKNOWN_EC
;
885 ubi_err("'ubi_io_read_ec_hdr()' returned unknown code %d", err
);
892 /* Make sure UBI version is OK */
893 if (ech
->version
!= UBI_VERSION
) {
894 ubi_err("this UBI version is %d, image version is %d",
895 UBI_VERSION
, (int)ech
->version
);
899 ec
= be64_to_cpu(ech
->ec
);
900 if (ec
> UBI_MAX_ERASECOUNTER
) {
902 * Erase counter overflow. The EC headers have 64 bits
903 * reserved, but we anyway make use of only 31 bit
904 * values, as this seems to be enough for any existing
905 * flash. Upgrade UBI and use 64-bit erase counters
908 ubi_err("erase counter overflow, max is %d",
909 UBI_MAX_ERASECOUNTER
);
910 ubi_dump_ec_hdr(ech
);
915 * Make sure that all PEBs have the same image sequence number.
916 * This allows us to detect situations when users flash UBI
917 * images incorrectly, so that the flash has the new UBI image
918 * and leftovers from the old one. This feature was added
919 * relatively recently, and the sequence number was always
920 * zero, because old UBI implementations always set it to zero.
921 * For this reasons, we do not panic if some PEBs have zero
922 * sequence number, while other PEBs have non-zero sequence
925 image_seq
= be32_to_cpu(ech
->image_seq
);
926 if (!ubi
->image_seq
&& image_seq
)
927 ubi
->image_seq
= image_seq
;
928 if (ubi
->image_seq
&& image_seq
&&
929 ubi
->image_seq
!= image_seq
) {
930 ubi_err("bad image sequence number %d in PEB %d, "
931 "expected %d", image_seq
, pnum
, ubi
->image_seq
);
932 ubi_dump_ec_hdr(ech
);
937 /* OK, we've done with the EC header, let's look at the VID header */
939 err
= ubi_io_read_vid_hdr(ubi
, pnum
, vidh
, 0);
945 case UBI_IO_BITFLIPS
:
948 case UBI_IO_BAD_HDR_EBADMSG
:
949 if (ec_err
== UBI_IO_BAD_HDR_EBADMSG
)
951 * Both EC and VID headers are corrupted and were read
952 * with data integrity error, probably this is a bad
953 * PEB, bit it is not marked as bad yet. This may also
954 * be a result of power cut during erasure.
956 si
->maybe_bad_peb_count
+= 1;
960 * Both headers are corrupted. There is a possibility
961 * that this a valid UBI PEB which has corresponding
962 * LEB, but the headers are corrupted. However, it is
963 * impossible to distinguish it from a PEB which just
964 * contains garbage because of a power cut during erase
965 * operation. So we just schedule this PEB for erasure.
967 * Besides, in case of NOR flash, we deliberately
968 * corrupt both headers because NOR flash erasure is
969 * slow and can start from the end.
974 * The EC was OK, but the VID header is corrupted. We
975 * have to check what is in the data area.
977 err
= check_corruption(ubi
, vidh
, pnum
);
982 /* This corruption is caused by a power cut */
983 err
= add_to_list(si
, pnum
, ec
, 1, &si
->erase
);
985 /* This is an unexpected corruption */
986 err
= add_corrupted(si
, pnum
, ec
);
990 case UBI_IO_FF_BITFLIPS
:
991 err
= add_to_list(si
, pnum
, ec
, 1, &si
->erase
);
997 err
= add_to_list(si
, pnum
, ec
, 1, &si
->erase
);
999 err
= add_to_list(si
, pnum
, ec
, 0, &si
->free
);
1002 goto adjust_mean_ec
;
1004 ubi_err("'ubi_io_read_vid_hdr()' returned unknown code %d",
1009 vol_id
= be32_to_cpu(vidh
->vol_id
);
1010 if (vol_id
> UBI_MAX_VOLUMES
&& vol_id
!= UBI_LAYOUT_VOLUME_ID
) {
1011 int lnum
= be32_to_cpu(vidh
->lnum
);
1013 /* Unsupported internal volume */
1014 switch (vidh
->compat
) {
1015 case UBI_COMPAT_DELETE
:
1016 ubi_msg("\"delete\" compatible internal volume %d:%d"
1017 " found, will remove it", vol_id
, lnum
);
1018 err
= add_to_list(si
, pnum
, ec
, 1, &si
->erase
);
1024 ubi_msg("read-only compatible internal volume %d:%d"
1025 " found, switch to read-only mode",
1030 case UBI_COMPAT_PRESERVE
:
1031 ubi_msg("\"preserve\" compatible internal volume %d:%d"
1032 " found", vol_id
, lnum
);
1033 err
= add_to_list(si
, pnum
, ec
, 0, &si
->alien
);
1038 case UBI_COMPAT_REJECT
:
1039 ubi_err("incompatible internal volume %d:%d found",
1046 ubi_warn("valid VID header but corrupted EC header at PEB %d",
1048 err
= ubi_scan_add_used(ubi
, si
, pnum
, ec
, vidh
, bitflips
);
1056 if (ec
> si
->max_ec
)
1058 if (ec
< si
->min_ec
)
1066 * check_what_we_have - check what PEB were found by scanning.
1067 * @ubi: UBI device description object
1068 * @si: scanning information
1070 * This is a helper function which takes a look what PEBs were found by
1071 * scanning, and decides whether the flash is empty and should be formatted and
1072 * whether there are too many corrupted PEBs and we should not attach this
1073 * MTD device. Returns zero if we should proceed with attaching the MTD device,
1074 * and %-EINVAL if we should not.
1076 static int check_what_we_have(struct ubi_device
*ubi
, struct ubi_scan_info
*si
)
1078 struct ubi_ainf_peb
*seb
;
1079 int max_corr
, peb_count
;
1081 peb_count
= ubi
->peb_count
- si
->bad_peb_count
- si
->alien_peb_count
;
1082 max_corr
= peb_count
/ 20 ?: 8;
1085 * Few corrupted PEBs is not a problem and may be just a result of
1086 * unclean reboots. However, many of them may indicate some problems
1087 * with the flash HW or driver.
1089 if (si
->corr_peb_count
) {
1090 ubi_err("%d PEBs are corrupted and preserved",
1091 si
->corr_peb_count
);
1092 printk(KERN_ERR
"Corrupted PEBs are:");
1093 list_for_each_entry(seb
, &si
->corr
, u
.list
)
1094 printk(KERN_CONT
" %d", seb
->pnum
);
1095 printk(KERN_CONT
"\n");
1098 * If too many PEBs are corrupted, we refuse attaching,
1099 * otherwise, only print a warning.
1101 if (si
->corr_peb_count
>= max_corr
) {
1102 ubi_err("too many corrupted PEBs, refusing");
1107 if (si
->empty_peb_count
+ si
->maybe_bad_peb_count
== peb_count
) {
1109 * All PEBs are empty, or almost all - a couple PEBs look like
1110 * they may be bad PEBs which were not marked as bad yet.
1112 * This piece of code basically tries to distinguish between
1113 * the following situations:
1115 * 1. Flash is empty, but there are few bad PEBs, which are not
1116 * marked as bad so far, and which were read with error. We
1117 * want to go ahead and format this flash. While formatting,
1118 * the faulty PEBs will probably be marked as bad.
1120 * 2. Flash contains non-UBI data and we do not want to format
1121 * it and destroy possibly important information.
1123 if (si
->maybe_bad_peb_count
<= 2) {
1125 ubi_msg("empty MTD device detected");
1126 get_random_bytes(&ubi
->image_seq
,
1127 sizeof(ubi
->image_seq
));
1129 ubi_err("MTD device is not UBI-formatted and possibly "
1130 "contains non-UBI data - refusing it");
1140 * ubi_scan - scan an MTD device.
1141 * @ubi: UBI device description object
1143 * This function does full scanning of an MTD device and returns complete
1144 * information about it. In case of failure, an error code is returned.
1146 struct ubi_scan_info
*ubi_scan(struct ubi_device
*ubi
)
1149 struct rb_node
*rb1
, *rb2
;
1150 struct ubi_ainf_volume
*sv
;
1151 struct ubi_ainf_peb
*seb
;
1152 struct ubi_scan_info
*si
;
1154 si
= kzalloc(sizeof(struct ubi_scan_info
), GFP_KERNEL
);
1156 return ERR_PTR(-ENOMEM
);
1158 INIT_LIST_HEAD(&si
->corr
);
1159 INIT_LIST_HEAD(&si
->free
);
1160 INIT_LIST_HEAD(&si
->erase
);
1161 INIT_LIST_HEAD(&si
->alien
);
1162 si
->volumes
= RB_ROOT
;
1165 si
->scan_leb_slab
= kmem_cache_create("ubi_scan_leb_slab",
1166 sizeof(struct ubi_ainf_peb
),
1168 if (!si
->scan_leb_slab
)
1171 ech
= kzalloc(ubi
->ec_hdr_alsize
, GFP_KERNEL
);
1175 vidh
= ubi_zalloc_vid_hdr(ubi
, GFP_KERNEL
);
1179 for (pnum
= 0; pnum
< ubi
->peb_count
; pnum
++) {
1182 dbg_gen("process PEB %d", pnum
);
1183 err
= process_eb(ubi
, si
, pnum
);
1188 dbg_msg("scanning is finished");
1190 /* Calculate mean erase counter */
1192 si
->mean_ec
= div_u64(si
->ec_sum
, si
->ec_count
);
1194 err
= check_what_we_have(ubi
, si
);
1199 * In case of unknown erase counter we use the mean erase counter
1202 ubi_rb_for_each_entry(rb1
, sv
, &si
->volumes
, rb
) {
1203 ubi_rb_for_each_entry(rb2
, seb
, &sv
->root
, u
.rb
)
1204 if (seb
->ec
== UBI_SCAN_UNKNOWN_EC
)
1205 seb
->ec
= si
->mean_ec
;
1208 list_for_each_entry(seb
, &si
->free
, u
.list
) {
1209 if (seb
->ec
== UBI_SCAN_UNKNOWN_EC
)
1210 seb
->ec
= si
->mean_ec
;
1213 list_for_each_entry(seb
, &si
->corr
, u
.list
)
1214 if (seb
->ec
== UBI_SCAN_UNKNOWN_EC
)
1215 seb
->ec
= si
->mean_ec
;
1217 list_for_each_entry(seb
, &si
->erase
, u
.list
)
1218 if (seb
->ec
== UBI_SCAN_UNKNOWN_EC
)
1219 seb
->ec
= si
->mean_ec
;
1221 err
= self_check_si(ubi
, si
);
1225 ubi_free_vid_hdr(ubi
, vidh
);
1231 ubi_free_vid_hdr(ubi
, vidh
);
1235 ubi_scan_destroy_si(si
);
1236 return ERR_PTR(err
);
1240 * destroy_sv - free the scanning volume information
1241 * @sv: scanning volume information
1242 * @si: scanning information
1244 * This function destroys the volume RB-tree (@sv->root) and the scanning
1245 * volume information.
1247 static void destroy_sv(struct ubi_scan_info
*si
, struct ubi_ainf_volume
*sv
)
1249 struct ubi_ainf_peb
*seb
;
1250 struct rb_node
*this = sv
->root
.rb_node
;
1254 this = this->rb_left
;
1255 else if (this->rb_right
)
1256 this = this->rb_right
;
1258 seb
= rb_entry(this, struct ubi_ainf_peb
, u
.rb
);
1259 this = rb_parent(this);
1261 if (this->rb_left
== &seb
->u
.rb
)
1262 this->rb_left
= NULL
;
1264 this->rb_right
= NULL
;
1267 kmem_cache_free(si
->scan_leb_slab
, seb
);
1274 * ubi_scan_destroy_si - destroy scanning information.
1275 * @si: scanning information
1277 void ubi_scan_destroy_si(struct ubi_scan_info
*si
)
1279 struct ubi_ainf_peb
*seb
, *seb_tmp
;
1280 struct ubi_ainf_volume
*sv
;
1283 list_for_each_entry_safe(seb
, seb_tmp
, &si
->alien
, u
.list
) {
1284 list_del(&seb
->u
.list
);
1285 kmem_cache_free(si
->scan_leb_slab
, seb
);
1287 list_for_each_entry_safe(seb
, seb_tmp
, &si
->erase
, 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
->corr
, 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
->free
, u
.list
) {
1296 list_del(&seb
->u
.list
);
1297 kmem_cache_free(si
->scan_leb_slab
, seb
);
1300 /* Destroy the volume RB-tree */
1301 rb
= si
->volumes
.rb_node
;
1305 else if (rb
->rb_right
)
1308 sv
= rb_entry(rb
, struct ubi_ainf_volume
, rb
);
1312 if (rb
->rb_left
== &sv
->rb
)
1315 rb
->rb_right
= NULL
;
1322 if (si
->scan_leb_slab
)
1323 kmem_cache_destroy(si
->scan_leb_slab
);
1329 * self_check_si - check the scanning information.
1330 * @ubi: UBI device description object
1331 * @si: scanning information
1333 * This function returns zero if the scanning information is all right, and a
1334 * negative error code if not or if an error occurred.
1336 static int self_check_si(struct ubi_device
*ubi
, struct ubi_scan_info
*si
)
1338 int pnum
, err
, vols_found
= 0;
1339 struct rb_node
*rb1
, *rb2
;
1340 struct ubi_ainf_volume
*sv
;
1341 struct ubi_ainf_peb
*seb
, *last_seb
;
1344 if (!ubi
->dbg
->chk_gen
)
1348 * At first, check that scanning information is OK.
1350 ubi_rb_for_each_entry(rb1
, sv
, &si
->volumes
, rb
) {
1358 ubi_err("bad is_empty flag");
1362 if (sv
->vol_id
< 0 || sv
->highest_lnum
< 0 ||
1363 sv
->leb_count
< 0 || sv
->vol_type
< 0 || sv
->used_ebs
< 0 ||
1364 sv
->data_pad
< 0 || sv
->last_data_size
< 0) {
1365 ubi_err("negative values");
1369 if (sv
->vol_id
>= UBI_MAX_VOLUMES
&&
1370 sv
->vol_id
< UBI_INTERNAL_VOL_START
) {
1371 ubi_err("bad vol_id");
1375 if (sv
->vol_id
> si
->highest_vol_id
) {
1376 ubi_err("highest_vol_id is %d, but vol_id %d is there",
1377 si
->highest_vol_id
, sv
->vol_id
);
1381 if (sv
->vol_type
!= UBI_DYNAMIC_VOLUME
&&
1382 sv
->vol_type
!= UBI_STATIC_VOLUME
) {
1383 ubi_err("bad vol_type");
1387 if (sv
->data_pad
> ubi
->leb_size
/ 2) {
1388 ubi_err("bad data_pad");
1393 ubi_rb_for_each_entry(rb2
, seb
, &sv
->root
, u
.rb
) {
1399 if (seb
->pnum
< 0 || seb
->ec
< 0) {
1400 ubi_err("negative values");
1404 if (seb
->ec
< si
->min_ec
) {
1405 ubi_err("bad si->min_ec (%d), %d found",
1406 si
->min_ec
, seb
->ec
);
1410 if (seb
->ec
> si
->max_ec
) {
1411 ubi_err("bad si->max_ec (%d), %d found",
1412 si
->max_ec
, seb
->ec
);
1416 if (seb
->pnum
>= ubi
->peb_count
) {
1417 ubi_err("too high PEB number %d, total PEBs %d",
1418 seb
->pnum
, ubi
->peb_count
);
1422 if (sv
->vol_type
== UBI_STATIC_VOLUME
) {
1423 if (seb
->lnum
>= sv
->used_ebs
) {
1424 ubi_err("bad lnum or used_ebs");
1428 if (sv
->used_ebs
!= 0) {
1429 ubi_err("non-zero used_ebs");
1434 if (seb
->lnum
> sv
->highest_lnum
) {
1435 ubi_err("incorrect highest_lnum or lnum");
1440 if (sv
->leb_count
!= leb_count
) {
1441 ubi_err("bad leb_count, %d objects in the tree",
1451 if (seb
->lnum
!= sv
->highest_lnum
) {
1452 ubi_err("bad highest_lnum");
1457 if (vols_found
!= si
->vols_found
) {
1458 ubi_err("bad si->vols_found %d, should be %d",
1459 si
->vols_found
, vols_found
);
1463 /* Check that scanning information is correct */
1464 ubi_rb_for_each_entry(rb1
, sv
, &si
->volumes
, rb
) {
1466 ubi_rb_for_each_entry(rb2
, seb
, &sv
->root
, u
.rb
) {
1473 err
= ubi_io_read_vid_hdr(ubi
, seb
->pnum
, vidh
, 1);
1474 if (err
&& err
!= UBI_IO_BITFLIPS
) {
1475 ubi_err("VID header is not OK (%d)", err
);
1481 vol_type
= vidh
->vol_type
== UBI_VID_DYNAMIC
?
1482 UBI_DYNAMIC_VOLUME
: UBI_STATIC_VOLUME
;
1483 if (sv
->vol_type
!= vol_type
) {
1484 ubi_err("bad vol_type");
1488 if (seb
->sqnum
!= be64_to_cpu(vidh
->sqnum
)) {
1489 ubi_err("bad sqnum %llu", seb
->sqnum
);
1493 if (sv
->vol_id
!= be32_to_cpu(vidh
->vol_id
)) {
1494 ubi_err("bad vol_id %d", sv
->vol_id
);
1498 if (sv
->compat
!= vidh
->compat
) {
1499 ubi_err("bad compat %d", vidh
->compat
);
1503 if (seb
->lnum
!= be32_to_cpu(vidh
->lnum
)) {
1504 ubi_err("bad lnum %d", seb
->lnum
);
1508 if (sv
->used_ebs
!= be32_to_cpu(vidh
->used_ebs
)) {
1509 ubi_err("bad used_ebs %d", sv
->used_ebs
);
1513 if (sv
->data_pad
!= be32_to_cpu(vidh
->data_pad
)) {
1514 ubi_err("bad data_pad %d", sv
->data_pad
);
1522 if (sv
->highest_lnum
!= be32_to_cpu(vidh
->lnum
)) {
1523 ubi_err("bad highest_lnum %d", sv
->highest_lnum
);
1527 if (sv
->last_data_size
!= be32_to_cpu(vidh
->data_size
)) {
1528 ubi_err("bad last_data_size %d", sv
->last_data_size
);
1534 * Make sure that all the physical eraseblocks are in one of the lists
1537 buf
= kzalloc(ubi
->peb_count
, GFP_KERNEL
);
1541 for (pnum
= 0; pnum
< ubi
->peb_count
; pnum
++) {
1542 err
= ubi_io_is_bad(ubi
, pnum
);
1550 ubi_rb_for_each_entry(rb1
, sv
, &si
->volumes
, rb
)
1551 ubi_rb_for_each_entry(rb2
, seb
, &sv
->root
, u
.rb
)
1554 list_for_each_entry(seb
, &si
->free
, u
.list
)
1557 list_for_each_entry(seb
, &si
->corr
, u
.list
)
1560 list_for_each_entry(seb
, &si
->erase
, u
.list
)
1563 list_for_each_entry(seb
, &si
->alien
, u
.list
)
1567 for (pnum
= 0; pnum
< ubi
->peb_count
; pnum
++)
1569 ubi_err("PEB %d is not referred", pnum
);
1579 ubi_err("bad scanning information about LEB %d", seb
->lnum
);
1580 ubi_dump_seb(seb
, 0);
1585 ubi_err("bad scanning information about volume %d", sv
->vol_id
);
1590 ubi_err("bad scanning information about volume %d", sv
->vol_id
);
1592 ubi_dump_vid_hdr(vidh
);