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 is 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 * ubi_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 int ubi_compare_lebs(struct ubi_device *ubi, const struct ubi_ainf_peb *aeb,
323 int pnum, const struct ubi_vid_hdr *vid_hdr)
325 int len, err, second_is_newer, bitflips = 0, corrupted = 0;
326 uint32_t data_crc, crc;
327 struct ubi_vid_hdr *vh = NULL;
328 unsigned long long sqnum2 = be64_to_cpu(vid_hdr->sqnum);
330 if (sqnum2 == aeb->sqnum) {
332 * This must be a really ancient UBI image which has been
333 * created before sequence numbers support has been added. At
334 * that times we used 32-bit LEB versions stored in logical
335 * eraseblocks. That was before UBI got into mainline. We do not
336 * support these images anymore. Well, those images still work,
337 * but only if no unclean reboots happened.
339 ubi_err("unsupported on-flash UBI format");
343 /* Obviously the LEB with lower sequence counter is older */
344 second_is_newer = (sqnum2 > aeb->sqnum);
347 * Now we know which copy is newer. If the copy flag of the PEB with
348 * newer version is not set, then we just return, otherwise we have to
349 * check data CRC. For the second PEB we already have the VID header,
350 * for the first one - we'll need to re-read it from flash.
352 * Note: this may be optimized so that we wouldn't read twice.
355 if (second_is_newer) {
356 if (!vid_hdr->copy_flag) {
357 /* It is not a copy, so it is newer */
358 dbg_bld("second PEB %d is newer, copy_flag is unset",
363 if (!aeb->copy_flag) {
364 /* It is not a copy, so it is newer */
365 dbg_bld("first PEB %d is newer, copy_flag is unset",
367 return bitflips << 1;
370 vh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
375 err = ubi_io_read_vid_hdr(ubi, pnum, vh, 0);
377 if (err == UBI_IO_BITFLIPS)
380 ubi_err("VID of PEB %d header is bad, but it was OK earlier, err %d",
392 /* Read the data of the copy and check the CRC */
394 len = be32_to_cpu(vid_hdr->data_size);
396 mutex_lock(&ubi->buf_mutex);
397 err = ubi_io_read_data(ubi, ubi->peb_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, ubi->peb_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);
413 mutex_unlock(&ubi->buf_mutex);
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);
425 mutex_unlock(&ubi->buf_mutex);
427 ubi_free_vid_hdr(ubi, vh);
432 * ubi_add_to_av - add used physical eraseblock to the attaching information.
433 * @ubi: UBI device description object
434 * @ai: attaching 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_add_to_av(struct ubi_device *ubi, struct ubi_attach_info *ai, int pnum,
448 int ec, const struct ubi_vid_hdr *vid_hdr, int bitflips)
450 int err, vol_id, lnum;
451 unsigned long long sqnum;
452 struct ubi_ainf_volume *av;
453 struct ubi_ainf_peb *aeb;
454 struct rb_node **p, *parent = NULL;
456 vol_id = be32_to_cpu(vid_hdr->vol_id);
457 lnum = be32_to_cpu(vid_hdr->lnum);
458 sqnum = be64_to_cpu(vid_hdr->sqnum);
460 dbg_bld("PEB %d, LEB %d:%d, EC %d, sqnum %llu, bitflips %d",
461 pnum, vol_id, lnum, ec, sqnum, bitflips);
463 av = add_volume(ai, vol_id, pnum, vid_hdr);
467 if (ai->max_sqnum < sqnum)
468 ai->max_sqnum = sqnum;
471 * Walk the RB-tree of logical eraseblocks of volume @vol_id to look
472 * if this is the first instance of this logical eraseblock or not.
474 p = &av->root.rb_node;
479 aeb = rb_entry(parent, struct ubi_ainf_peb, u.rb);
480 if (lnum != aeb->lnum) {
481 if (lnum < aeb->lnum)
489 * There is already a physical eraseblock describing the same
490 * logical eraseblock present.
493 dbg_bld("this LEB already exists: PEB %d, sqnum %llu, EC %d",
494 aeb->pnum, aeb->sqnum, aeb->ec);
497 * Make sure that the logical eraseblocks have different
498 * sequence numbers. Otherwise the image is bad.
500 * However, if the sequence number is zero, we assume it must
501 * be an ancient UBI image from the era when UBI did not have
502 * sequence numbers. We still can attach these images, unless
503 * there is a need to distinguish between old and new
504 * eraseblocks, in which case we'll refuse the image in
505 * 'ubi_compare_lebs()'. In other words, we attach old clean
506 * images, but refuse attaching old images with duplicated
507 * logical eraseblocks because there was an unclean reboot.
509 if (aeb->sqnum == sqnum && sqnum != 0) {
510 ubi_err("two LEBs with same sequence number %llu",
512 ubi_dump_aeb(aeb, 0);
513 ubi_dump_vid_hdr(vid_hdr);
518 * Now we have to drop the older one and preserve the newer
521 cmp_res = ubi_compare_lebs(ubi, aeb, pnum, vid_hdr);
527 * This logical eraseblock is newer than the one
530 err = validate_vid_hdr(vid_hdr, av, pnum);
534 err = add_to_list(ai, aeb->pnum, aeb->vol_id,
535 aeb->lnum, aeb->ec, cmp_res & 4,
542 aeb->vol_id = vol_id;
544 aeb->scrub = ((cmp_res & 2) || bitflips);
545 aeb->copy_flag = vid_hdr->copy_flag;
548 if (av->highest_lnum == lnum)
550 be32_to_cpu(vid_hdr->data_size);
555 * This logical eraseblock is older than the one found
558 return add_to_list(ai, pnum, vol_id, lnum, ec,
559 cmp_res & 4, &ai->erase);
564 * We've met this logical eraseblock for the first time, add it to the
565 * attaching information.
568 err = validate_vid_hdr(vid_hdr, av, pnum);
572 aeb = kmem_cache_alloc(ai->aeb_slab_cache, GFP_KERNEL);
578 aeb->vol_id = vol_id;
580 aeb->scrub = bitflips;
581 aeb->copy_flag = vid_hdr->copy_flag;
584 if (av->highest_lnum <= lnum) {
585 av->highest_lnum = lnum;
586 av->last_data_size = be32_to_cpu(vid_hdr->data_size);
590 rb_link_node(&aeb->u.rb, parent, p);
591 rb_insert_color(&aeb->u.rb, &av->root);
596 * ubi_find_av - find volume in the attaching information.
597 * @ai: attaching information
598 * @vol_id: the requested volume ID
600 * This function returns a pointer to the volume description or %NULL if there
601 * are no data about this volume in the attaching information.
603 struct ubi_ainf_volume *ubi_find_av(const struct ubi_attach_info *ai,
606 struct ubi_ainf_volume *av;
607 struct rb_node *p = ai->volumes.rb_node;
610 av = rb_entry(p, struct ubi_ainf_volume, rb);
612 if (vol_id == av->vol_id)
615 if (vol_id > av->vol_id)
625 * ubi_remove_av - delete attaching information about a volume.
626 * @ai: attaching information
627 * @av: the volume attaching information to delete
629 void ubi_remove_av(struct ubi_attach_info *ai, struct ubi_ainf_volume *av)
632 struct ubi_ainf_peb *aeb;
634 dbg_bld("remove attaching information about volume %d", av->vol_id);
636 while ((rb = rb_first(&av->root))) {
637 aeb = rb_entry(rb, struct ubi_ainf_peb, u.rb);
638 rb_erase(&aeb->u.rb, &av->root);
639 list_add_tail(&aeb->u.list, &ai->erase);
642 rb_erase(&av->rb, &ai->volumes);
648 * early_erase_peb - erase a physical eraseblock.
649 * @ubi: UBI device description object
650 * @ai: attaching information
651 * @pnum: physical eraseblock number to erase;
652 * @ec: erase counter value to write (%UBI_UNKNOWN if it is unknown)
654 * This function erases physical eraseblock 'pnum', and writes the erase
655 * counter header to it. This function should only be used on UBI device
656 * initialization stages, when the EBA sub-system had not been yet initialized.
657 * This function returns zero in case of success and a negative error code in
660 static int early_erase_peb(struct ubi_device *ubi,
661 const struct ubi_attach_info *ai, int pnum, int ec)
664 struct ubi_ec_hdr *ec_hdr;
666 if ((long long)ec >= UBI_MAX_ERASECOUNTER) {
668 * Erase counter overflow. Upgrade UBI and use 64-bit
669 * erase counters internally.
671 ubi_err("erase counter overflow at PEB %d, EC %d", pnum, ec);
675 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
679 ec_hdr->ec = cpu_to_be64(ec);
681 err = ubi_io_sync_erase(ubi, pnum, 0);
685 err = ubi_io_write_ec_hdr(ubi, pnum, ec_hdr);
693 * ubi_early_get_peb - get a free physical eraseblock.
694 * @ubi: UBI device description object
695 * @ai: attaching information
697 * This function returns a free physical eraseblock. It is supposed to be
698 * called on the UBI initialization stages when the wear-leveling sub-system is
699 * not initialized yet. This function picks a physical eraseblocks from one of
700 * the lists, writes the EC header if it is needed, and removes it from the
703 * This function returns a pointer to the "aeb" of the found free PEB in case
704 * of success and an error code in case of failure.
706 struct ubi_ainf_peb *ubi_early_get_peb(struct ubi_device *ubi,
707 struct ubi_attach_info *ai)
710 struct ubi_ainf_peb *aeb, *tmp_aeb;
712 if (!list_empty(&ai->free)) {
713 aeb = list_entry(ai->free.next, struct ubi_ainf_peb, u.list);
714 list_del(&aeb->u.list);
715 dbg_bld("return free PEB %d, EC %d", aeb->pnum, aeb->ec);
720 * We try to erase the first physical eraseblock from the erase list
721 * and pick it if we succeed, or try to erase the next one if not. And
722 * so forth. We don't want to take care about bad eraseblocks here -
723 * they'll be handled later.
725 list_for_each_entry_safe(aeb, tmp_aeb, &ai->erase, u.list) {
726 if (aeb->ec == UBI_UNKNOWN)
727 aeb->ec = ai->mean_ec;
729 err = early_erase_peb(ubi, ai, aeb->pnum, aeb->ec+1);
734 list_del(&aeb->u.list);
735 dbg_bld("return PEB %d, EC %d", aeb->pnum, aeb->ec);
739 ubi_err("no free eraseblocks");
740 return ERR_PTR(-ENOSPC);
744 * check_corruption - check the data area of PEB.
745 * @ubi: UBI device description object
746 * @vid_hdr: the (corrupted) VID header of this PEB
747 * @pnum: the physical eraseblock number to check
749 * This is a helper function which is used to distinguish between VID header
750 * corruptions caused by power cuts and other reasons. If the PEB contains only
751 * 0xFF bytes in the data area, the VID header is most probably corrupted
752 * because of a power cut (%0 is returned in this case). Otherwise, it was
753 * probably corrupted for some other reasons (%1 is returned in this case). A
754 * negative error code is returned if a read error occurred.
756 * If the corruption reason was a power cut, UBI can safely erase this PEB.
757 * Otherwise, it should preserve it to avoid possibly destroying important
760 static int check_corruption(struct ubi_device *ubi, struct ubi_vid_hdr *vid_hdr,
765 mutex_lock(&ubi->buf_mutex);
766 memset(ubi->peb_buf, 0x00, ubi->leb_size);
768 err = ubi_io_read(ubi, ubi->peb_buf, pnum, ubi->leb_start,
770 if (err == UBI_IO_BITFLIPS || mtd_is_eccerr(err)) {
772 * Bit-flips or integrity errors while reading the data area.
773 * It is difficult to say for sure what type of corruption is
774 * this, but presumably a power cut happened while this PEB was
775 * erased, so it became unstable and corrupted, and should be
785 if (ubi_check_pattern(ubi->peb_buf, 0xFF, ubi->leb_size))
788 ubi_err("PEB %d contains corrupted VID header, and the data does not contain all 0xFF",
790 ubi_err("this may be a non-UBI PEB or a severe VID header corruption which requires manual inspection");
791 ubi_dump_vid_hdr(vid_hdr);
792 pr_err("hexdump of PEB %d offset %d, length %d",
793 pnum, ubi->leb_start, ubi->leb_size);
794 ubi_dbg_print_hex_dump(KERN_DEBUG, "", DUMP_PREFIX_OFFSET, 32, 1,
795 ubi->peb_buf, ubi->leb_size, 1);
799 mutex_unlock(&ubi->buf_mutex);
804 * scan_peb - scan and process UBI headers of a PEB.
805 * @ubi: UBI device description object
806 * @ai: attaching information
807 * @pnum: the physical eraseblock number
808 * @vid: The volume ID of the found volume will be stored in this pointer
809 * @sqnum: The sqnum of the found volume will be stored in this pointer
811 * This function reads UBI headers of PEB @pnum, checks them, and adds
812 * information about this PEB to the corresponding list or RB-tree in the
813 * "attaching info" structure. Returns zero if the physical eraseblock was
814 * successfully handled and a negative error code in case of failure.
816 static int scan_peb(struct ubi_device *ubi, struct ubi_attach_info *ai,
817 int pnum, int *vid, unsigned long long *sqnum)
819 long long uninitialized_var(ec);
820 int err, bitflips = 0, vol_id = -1, ec_err = 0;
822 dbg_bld("scan PEB %d", pnum);
824 /* Skip bad physical eraseblocks */
825 err = ubi_io_is_bad(ubi, pnum);
829 ai->bad_peb_count += 1;
833 err = ubi_io_read_ec_hdr(ubi, pnum, ech, 0);
839 case UBI_IO_BITFLIPS:
843 ai->empty_peb_count += 1;
844 return add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
845 UBI_UNKNOWN, 0, &ai->erase);
846 case UBI_IO_FF_BITFLIPS:
847 ai->empty_peb_count += 1;
848 return add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
849 UBI_UNKNOWN, 1, &ai->erase);
850 case UBI_IO_BAD_HDR_EBADMSG:
853 * We have to also look at the VID header, possibly it is not
854 * corrupted. Set %bitflips flag in order to make this PEB be
855 * moved and EC be re-created.
862 ubi_err("'ubi_io_read_ec_hdr()' returned unknown code %d", err);
869 /* Make sure UBI version is OK */
870 if (ech->version != UBI_VERSION) {
871 ubi_err("this UBI version is %d, image version is %d",
872 UBI_VERSION, (int)ech->version);
876 ec = be64_to_cpu(ech->ec);
877 if (ec > UBI_MAX_ERASECOUNTER) {
879 * Erase counter overflow. The EC headers have 64 bits
880 * reserved, but we anyway make use of only 31 bit
881 * values, as this seems to be enough for any existing
882 * flash. Upgrade UBI and use 64-bit erase counters
885 ubi_err("erase counter overflow, max is %d",
886 UBI_MAX_ERASECOUNTER);
887 ubi_dump_ec_hdr(ech);
892 * Make sure that all PEBs have the same image sequence number.
893 * This allows us to detect situations when users flash UBI
894 * images incorrectly, so that the flash has the new UBI image
895 * and leftovers from the old one. This feature was added
896 * relatively recently, and the sequence number was always
897 * zero, because old UBI implementations always set it to zero.
898 * For this reasons, we do not panic if some PEBs have zero
899 * sequence number, while other PEBs have non-zero sequence
902 image_seq = be32_to_cpu(ech->image_seq);
903 if (!ubi->image_seq && image_seq)
904 ubi->image_seq = image_seq;
905 if (ubi->image_seq && image_seq &&
906 ubi->image_seq != image_seq) {
907 ubi_err("bad image sequence number %d in PEB %d, expected %d",
908 image_seq, pnum, ubi->image_seq);
909 ubi_dump_ec_hdr(ech);
914 /* OK, we've done with the EC header, let's look at the VID header */
916 err = ubi_io_read_vid_hdr(ubi, pnum, vidh, 0);
922 case UBI_IO_BITFLIPS:
925 case UBI_IO_BAD_HDR_EBADMSG:
926 if (ec_err == UBI_IO_BAD_HDR_EBADMSG)
928 * Both EC and VID headers are corrupted and were read
929 * with data integrity error, probably this is a bad
930 * PEB, bit it is not marked as bad yet. This may also
931 * be a result of power cut during erasure.
933 ai->maybe_bad_peb_count += 1;
937 * Both headers are corrupted. There is a possibility
938 * that this a valid UBI PEB which has corresponding
939 * LEB, but the headers are corrupted. However, it is
940 * impossible to distinguish it from a PEB which just
941 * contains garbage because of a power cut during erase
942 * operation. So we just schedule this PEB for erasure.
944 * Besides, in case of NOR flash, we deliberately
945 * corrupt both headers because NOR flash erasure is
946 * slow and can start from the end.
951 * The EC was OK, but the VID header is corrupted. We
952 * have to check what is in the data area.
954 err = check_corruption(ubi, vidh, pnum);
959 /* This corruption is caused by a power cut */
960 err = add_to_list(ai, pnum, UBI_UNKNOWN,
961 UBI_UNKNOWN, ec, 1, &ai->erase);
963 /* This is an unexpected corruption */
964 err = add_corrupted(ai, pnum, ec);
968 case UBI_IO_FF_BITFLIPS:
969 err = add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
975 if (ec_err || bitflips)
976 err = add_to_list(ai, pnum, UBI_UNKNOWN,
977 UBI_UNKNOWN, ec, 1, &ai->erase);
979 err = add_to_list(ai, pnum, UBI_UNKNOWN,
980 UBI_UNKNOWN, ec, 0, &ai->free);
985 ubi_err("'ubi_io_read_vid_hdr()' returned unknown code %d",
990 vol_id = be32_to_cpu(vidh->vol_id);
994 *sqnum = be64_to_cpu(vidh->sqnum);
995 if (vol_id > UBI_MAX_VOLUMES && vol_id != UBI_LAYOUT_VOLUME_ID) {
996 int lnum = be32_to_cpu(vidh->lnum);
998 /* Unsupported internal volume */
999 switch (vidh->compat) {
1000 case UBI_COMPAT_DELETE:
1001 if (vol_id != UBI_FM_SB_VOLUME_ID
1002 && vol_id != UBI_FM_DATA_VOLUME_ID) {
1003 ubi_msg("\"delete\" compatible internal volume %d:%d found, will remove it",
1006 err = add_to_list(ai, pnum, vol_id, lnum,
1013 ubi_msg("read-only compatible internal volume %d:%d found, switch to read-only mode",
1018 case UBI_COMPAT_PRESERVE:
1019 ubi_msg("\"preserve\" compatible internal volume %d:%d found",
1021 err = add_to_list(ai, pnum, vol_id, lnum,
1027 case UBI_COMPAT_REJECT:
1028 ubi_err("incompatible internal volume %d:%d found",
1035 ubi_warn("valid VID header but corrupted EC header at PEB %d",
1037 err = ubi_add_to_av(ubi, ai, pnum, ec, vidh, bitflips);
1045 if (ec > ai->max_ec)
1047 if (ec < ai->min_ec)
1055 * late_analysis - analyze the overall situation with PEB.
1056 * @ubi: UBI device description object
1057 * @ai: attaching information
1059 * This is a helper function which takes a look what PEBs we have after we
1060 * gather information about all of them ("ai" is compete). It decides whether
1061 * the flash is empty and should be formatted of whether there are too many
1062 * corrupted PEBs and we should not attach this MTD device. Returns zero if we
1063 * should proceed with attaching the MTD device, and %-EINVAL if we should not.
1065 static int late_analysis(struct ubi_device *ubi, struct ubi_attach_info *ai)
1067 struct ubi_ainf_peb *aeb;
1068 int max_corr, peb_count;
1070 peb_count = ubi->peb_count - ai->bad_peb_count - ai->alien_peb_count;
1071 max_corr = peb_count / 20 ?: 8;
1074 * Few corrupted PEBs is not a problem and may be just a result of
1075 * unclean reboots. However, many of them may indicate some problems
1076 * with the flash HW or driver.
1078 if (ai->corr_peb_count) {
1079 ubi_err("%d PEBs are corrupted and preserved",
1080 ai->corr_peb_count);
1081 pr_err("Corrupted PEBs are:");
1082 list_for_each_entry(aeb, &ai->corr, u.list)
1083 pr_cont(" %d", aeb->pnum);
1087 * If too many PEBs are corrupted, we refuse attaching,
1088 * otherwise, only print a warning.
1090 if (ai->corr_peb_count >= max_corr) {
1091 ubi_err("too many corrupted PEBs, refusing");
1096 if (ai->empty_peb_count + ai->maybe_bad_peb_count == peb_count) {
1098 * All PEBs are empty, or almost all - a couple PEBs look like
1099 * they may be bad PEBs which were not marked as bad yet.
1101 * This piece of code basically tries to distinguish between
1102 * the following situations:
1104 * 1. Flash is empty, but there are few bad PEBs, which are not
1105 * marked as bad so far, and which were read with error. We
1106 * want to go ahead and format this flash. While formatting,
1107 * the faulty PEBs will probably be marked as bad.
1109 * 2. Flash contains non-UBI data and we do not want to format
1110 * it and destroy possibly important information.
1112 if (ai->maybe_bad_peb_count <= 2) {
1114 ubi_msg("empty MTD device detected");
1115 get_random_bytes(&ubi->image_seq,
1116 sizeof(ubi->image_seq));
1118 ubi_err("MTD device is not UBI-formatted and possibly contains non-UBI data - refusing it");
1128 * destroy_av - free volume attaching information.
1129 * @av: volume attaching information
1130 * @ai: attaching information
1132 * This function destroys the volume attaching information.
1134 static void destroy_av(struct ubi_attach_info *ai, struct ubi_ainf_volume *av)
1136 struct ubi_ainf_peb *aeb;
1137 struct rb_node *this = av->root.rb_node;
1141 this = this->rb_left;
1142 else if (this->rb_right)
1143 this = this->rb_right;
1145 aeb = rb_entry(this, struct ubi_ainf_peb, u.rb);
1146 this = rb_parent(this);
1148 if (this->rb_left == &aeb->u.rb)
1149 this->rb_left = NULL;
1151 this->rb_right = NULL;
1154 kmem_cache_free(ai->aeb_slab_cache, aeb);
1161 * destroy_ai - destroy attaching information.
1162 * @ai: attaching information
1164 static void destroy_ai(struct ubi_attach_info *ai)
1166 struct ubi_ainf_peb *aeb, *aeb_tmp;
1167 struct ubi_ainf_volume *av;
1170 list_for_each_entry_safe(aeb, aeb_tmp, &ai->alien, u.list) {
1171 list_del(&aeb->u.list);
1172 kmem_cache_free(ai->aeb_slab_cache, aeb);
1174 list_for_each_entry_safe(aeb, aeb_tmp, &ai->erase, u.list) {
1175 list_del(&aeb->u.list);
1176 kmem_cache_free(ai->aeb_slab_cache, aeb);
1178 list_for_each_entry_safe(aeb, aeb_tmp, &ai->corr, u.list) {
1179 list_del(&aeb->u.list);
1180 kmem_cache_free(ai->aeb_slab_cache, aeb);
1182 list_for_each_entry_safe(aeb, aeb_tmp, &ai->free, u.list) {
1183 list_del(&aeb->u.list);
1184 kmem_cache_free(ai->aeb_slab_cache, aeb);
1187 /* Destroy the volume RB-tree */
1188 rb = ai->volumes.rb_node;
1192 else if (rb->rb_right)
1195 av = rb_entry(rb, struct ubi_ainf_volume, rb);
1199 if (rb->rb_left == &av->rb)
1202 rb->rb_right = NULL;
1209 if (ai->aeb_slab_cache)
1210 kmem_cache_destroy(ai->aeb_slab_cache);
1216 * scan_all - scan entire MTD device.
1217 * @ubi: UBI device description object
1218 * @ai: attach info object
1219 * @start: start scanning at this PEB
1221 * This function does full scanning of an MTD device and returns complete
1222 * information about it in form of a "struct ubi_attach_info" object. In case
1223 * of failure, an error code is returned.
1225 static int scan_all(struct ubi_device *ubi, struct ubi_attach_info *ai,
1229 struct rb_node *rb1, *rb2;
1230 struct ubi_ainf_volume *av;
1231 struct ubi_ainf_peb *aeb;
1235 ech = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
1239 vidh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
1243 for (pnum = start; pnum < ubi->peb_count; pnum++) {
1246 dbg_gen("process PEB %d", pnum);
1247 err = scan_peb(ubi, ai, pnum, NULL, NULL);
1252 ubi_msg("scanning is finished");
1254 /* Calculate mean erase counter */
1256 ai->mean_ec = div_u64(ai->ec_sum, ai->ec_count);
1258 err = late_analysis(ubi, ai);
1263 * In case of unknown erase counter we use the mean erase counter
1266 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1267 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb)
1268 if (aeb->ec == UBI_UNKNOWN)
1269 aeb->ec = ai->mean_ec;
1272 list_for_each_entry(aeb, &ai->free, u.list) {
1273 if (aeb->ec == UBI_UNKNOWN)
1274 aeb->ec = ai->mean_ec;
1277 list_for_each_entry(aeb, &ai->corr, u.list)
1278 if (aeb->ec == UBI_UNKNOWN)
1279 aeb->ec = ai->mean_ec;
1281 list_for_each_entry(aeb, &ai->erase, u.list)
1282 if (aeb->ec == UBI_UNKNOWN)
1283 aeb->ec = ai->mean_ec;
1285 err = self_check_ai(ubi, ai);
1289 ubi_free_vid_hdr(ubi, vidh);
1295 ubi_free_vid_hdr(ubi, vidh);
1301 #ifdef CONFIG_MTD_UBI_FASTMAP
1304 * scan_fastmap - try to find a fastmap and attach from it.
1305 * @ubi: UBI device description object
1306 * @ai: attach info object
1308 * Returns 0 on success, negative return values indicate an internal
1310 * UBI_NO_FASTMAP denotes that no fastmap was found.
1311 * UBI_BAD_FASTMAP denotes that the found fastmap was invalid.
1313 static int scan_fast(struct ubi_device *ubi, struct ubi_attach_info *ai)
1315 int err, pnum, fm_anchor = -1;
1316 unsigned long long max_sqnum = 0;
1320 ech = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
1324 vidh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
1328 for (pnum = 0; pnum < UBI_FM_MAX_START; pnum++) {
1330 unsigned long long sqnum = -1;
1333 dbg_gen("process PEB %d", pnum);
1334 err = scan_peb(ubi, ai, pnum, &vol_id, &sqnum);
1338 if (vol_id == UBI_FM_SB_VOLUME_ID && sqnum > max_sqnum) {
1344 ubi_free_vid_hdr(ubi, vidh);
1348 return UBI_NO_FASTMAP;
1350 return ubi_scan_fastmap(ubi, ai, fm_anchor);
1353 ubi_free_vid_hdr(ubi, vidh);
1362 static struct ubi_attach_info *alloc_ai(const char *slab_name)
1364 struct ubi_attach_info *ai;
1366 ai = kzalloc(sizeof(struct ubi_attach_info), GFP_KERNEL);
1370 INIT_LIST_HEAD(&ai->corr);
1371 INIT_LIST_HEAD(&ai->free);
1372 INIT_LIST_HEAD(&ai->erase);
1373 INIT_LIST_HEAD(&ai->alien);
1374 ai->volumes = RB_ROOT;
1375 ai->aeb_slab_cache = kmem_cache_create(slab_name,
1376 sizeof(struct ubi_ainf_peb),
1378 if (!ai->aeb_slab_cache) {
1387 * ubi_attach - attach an MTD device.
1388 * @ubi: UBI device descriptor
1389 * @force_scan: if set to non-zero attach by scanning
1391 * This function returns zero in case of success and a negative error code in
1394 int ubi_attach(struct ubi_device *ubi, int force_scan)
1397 struct ubi_attach_info *ai;
1399 ai = alloc_ai("ubi_aeb_slab_cache");
1403 #ifdef CONFIG_MTD_UBI_FASTMAP
1404 /* On small flash devices we disable fastmap in any case. */
1405 if ((int)mtd_div_by_eb(ubi->mtd->size, ubi->mtd) <= UBI_FM_MAX_START) {
1406 ubi->fm_disabled = 1;
1411 err = scan_all(ubi, ai, 0);
1413 err = scan_fast(ubi, ai);
1415 if (err != UBI_NO_FASTMAP) {
1417 ai = alloc_ai("ubi_aeb_slab_cache2");
1422 err = scan_all(ubi, ai, UBI_FM_MAX_START);
1426 err = scan_all(ubi, ai, 0);
1431 ubi->bad_peb_count = ai->bad_peb_count;
1432 ubi->good_peb_count = ubi->peb_count - ubi->bad_peb_count;
1433 ubi->corr_peb_count = ai->corr_peb_count;
1434 ubi->max_ec = ai->max_ec;
1435 ubi->mean_ec = ai->mean_ec;
1436 dbg_gen("max. sequence number: %llu", ai->max_sqnum);
1438 err = ubi_read_volume_table(ubi, ai);
1442 err = ubi_wl_init(ubi, ai);
1446 err = ubi_eba_init(ubi, ai);
1450 #ifdef CONFIG_MTD_UBI_FASTMAP
1451 if (ubi->fm && ubi_dbg_chk_gen(ubi)) {
1452 struct ubi_attach_info *scan_ai;
1454 scan_ai = alloc_ai("ubi_ckh_aeb_slab_cache");
1458 err = scan_all(ubi, scan_ai, 0);
1460 destroy_ai(scan_ai);
1464 err = self_check_eba(ubi, ai, scan_ai);
1465 destroy_ai(scan_ai);
1478 ubi_free_internal_volumes(ubi);
1486 * self_check_ai - check the attaching information.
1487 * @ubi: UBI device description object
1488 * @ai: attaching information
1490 * This function returns zero if the attaching information is all right, and a
1491 * negative error code if not or if an error occurred.
1493 static int self_check_ai(struct ubi_device *ubi, struct ubi_attach_info *ai)
1495 int pnum, err, vols_found = 0;
1496 struct rb_node *rb1, *rb2;
1497 struct ubi_ainf_volume *av;
1498 struct ubi_ainf_peb *aeb, *last_aeb;
1501 if (!ubi_dbg_chk_gen(ubi))
1505 * At first, check that attaching information is OK.
1507 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1515 ubi_err("bad is_empty flag");
1519 if (av->vol_id < 0 || av->highest_lnum < 0 ||
1520 av->leb_count < 0 || av->vol_type < 0 || av->used_ebs < 0 ||
1521 av->data_pad < 0 || av->last_data_size < 0) {
1522 ubi_err("negative values");
1526 if (av->vol_id >= UBI_MAX_VOLUMES &&
1527 av->vol_id < UBI_INTERNAL_VOL_START) {
1528 ubi_err("bad vol_id");
1532 if (av->vol_id > ai->highest_vol_id) {
1533 ubi_err("highest_vol_id is %d, but vol_id %d is there",
1534 ai->highest_vol_id, av->vol_id);
1538 if (av->vol_type != UBI_DYNAMIC_VOLUME &&
1539 av->vol_type != UBI_STATIC_VOLUME) {
1540 ubi_err("bad vol_type");
1544 if (av->data_pad > ubi->leb_size / 2) {
1545 ubi_err("bad data_pad");
1550 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1556 if (aeb->pnum < 0 || aeb->ec < 0) {
1557 ubi_err("negative values");
1561 if (aeb->ec < ai->min_ec) {
1562 ubi_err("bad ai->min_ec (%d), %d found",
1563 ai->min_ec, aeb->ec);
1567 if (aeb->ec > ai->max_ec) {
1568 ubi_err("bad ai->max_ec (%d), %d found",
1569 ai->max_ec, aeb->ec);
1573 if (aeb->pnum >= ubi->peb_count) {
1574 ubi_err("too high PEB number %d, total PEBs %d",
1575 aeb->pnum, ubi->peb_count);
1579 if (av->vol_type == UBI_STATIC_VOLUME) {
1580 if (aeb->lnum >= av->used_ebs) {
1581 ubi_err("bad lnum or used_ebs");
1585 if (av->used_ebs != 0) {
1586 ubi_err("non-zero used_ebs");
1591 if (aeb->lnum > av->highest_lnum) {
1592 ubi_err("incorrect highest_lnum or lnum");
1597 if (av->leb_count != leb_count) {
1598 ubi_err("bad leb_count, %d objects in the tree",
1608 if (aeb->lnum != av->highest_lnum) {
1609 ubi_err("bad highest_lnum");
1614 if (vols_found != ai->vols_found) {
1615 ubi_err("bad ai->vols_found %d, should be %d",
1616 ai->vols_found, vols_found);
1620 /* Check that attaching information is correct */
1621 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1623 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1630 err = ubi_io_read_vid_hdr(ubi, aeb->pnum, vidh, 1);
1631 if (err && err != UBI_IO_BITFLIPS) {
1632 ubi_err("VID header is not OK (%d)", err);
1638 vol_type = vidh->vol_type == UBI_VID_DYNAMIC ?
1639 UBI_DYNAMIC_VOLUME : UBI_STATIC_VOLUME;
1640 if (av->vol_type != vol_type) {
1641 ubi_err("bad vol_type");
1645 if (aeb->sqnum != be64_to_cpu(vidh->sqnum)) {
1646 ubi_err("bad sqnum %llu", aeb->sqnum);
1650 if (av->vol_id != be32_to_cpu(vidh->vol_id)) {
1651 ubi_err("bad vol_id %d", av->vol_id);
1655 if (av->compat != vidh->compat) {
1656 ubi_err("bad compat %d", vidh->compat);
1660 if (aeb->lnum != be32_to_cpu(vidh->lnum)) {
1661 ubi_err("bad lnum %d", aeb->lnum);
1665 if (av->used_ebs != be32_to_cpu(vidh->used_ebs)) {
1666 ubi_err("bad used_ebs %d", av->used_ebs);
1670 if (av->data_pad != be32_to_cpu(vidh->data_pad)) {
1671 ubi_err("bad data_pad %d", av->data_pad);
1679 if (av->highest_lnum != be32_to_cpu(vidh->lnum)) {
1680 ubi_err("bad highest_lnum %d", av->highest_lnum);
1684 if (av->last_data_size != be32_to_cpu(vidh->data_size)) {
1685 ubi_err("bad last_data_size %d", av->last_data_size);
1691 * Make sure that all the physical eraseblocks are in one of the lists
1694 buf = kzalloc(ubi->peb_count, GFP_KERNEL);
1698 for (pnum = 0; pnum < ubi->peb_count; pnum++) {
1699 err = ubi_io_is_bad(ubi, pnum);
1707 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb)
1708 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb)
1711 list_for_each_entry(aeb, &ai->free, u.list)
1714 list_for_each_entry(aeb, &ai->corr, u.list)
1717 list_for_each_entry(aeb, &ai->erase, u.list)
1720 list_for_each_entry(aeb, &ai->alien, u.list)
1724 for (pnum = 0; pnum < ubi->peb_count; pnum++)
1726 ubi_err("PEB %d is not referred", pnum);
1736 ubi_err("bad attaching information about LEB %d", aeb->lnum);
1737 ubi_dump_aeb(aeb, 0);
1742 ubi_err("bad attaching information about volume %d", av->vol_id);
1747 ubi_err("bad attaching information about volume %d", av->vol_id);
1749 ubi_dump_vid_hdr(vidh);