2 * Copyright (c) International Business Machines Corp., 2006
4 * SPDX-License-Identifier: GPL-2.0+
6 * Author: Artem Bityutskiy (Битюцкий Артём)
10 * UBI attaching sub-system.
12 * This sub-system is responsible for attaching MTD devices and it also
13 * implements flash media scanning.
15 * The attaching information is represented by a &struct ubi_attach_info'
16 * object. Information about volumes is represented by &struct ubi_ainf_volume
17 * objects which are kept in volume RB-tree with root at the @volumes field.
18 * The RB-tree is indexed by the volume ID.
20 * Logical eraseblocks are represented by &struct ubi_ainf_peb objects. These
21 * objects are kept in per-volume RB-trees with the root at the corresponding
22 * &struct ubi_ainf_volume object. To put it differently, we keep an RB-tree of
23 * per-volume objects and each of these objects is the root of RB-tree of
26 * Corrupted physical eraseblocks are put to the @corr list, free physical
27 * eraseblocks are put to the @free list and the physical eraseblock to be
28 * erased are put to the @erase list.
33 * UBI protects EC and VID headers with CRC-32 checksums, so it can detect
34 * whether the headers are corrupted or not. Sometimes UBI also protects the
35 * data with CRC-32, e.g., when it executes the atomic LEB change operation, or
36 * when it moves the contents of a PEB for wear-leveling purposes.
38 * UBI tries to distinguish between 2 types of corruptions.
40 * 1. Corruptions caused by power cuts. These are expected corruptions and UBI
41 * tries to handle them gracefully, without printing too many warnings and
42 * error messages. The idea is that we do not lose important data in these
43 * cases - we may lose only the data which were being written to the media just
44 * before the power cut happened, and the upper layers (e.g., UBIFS) are
45 * supposed to handle such data losses (e.g., by using the FS journal).
47 * When UBI detects a corruption (CRC-32 mismatch) in a PEB, and it looks like
48 * the reason is a power cut, UBI puts this PEB to the @erase list, and all
49 * PEBs in the @erase list are scheduled for erasure later.
51 * 2. Unexpected corruptions which are not caused by power cuts. During
52 * attaching, such PEBs are put to the @corr list and UBI preserves them.
53 * Obviously, this lessens the amount of available PEBs, and if at some point
54 * UBI runs out of free PEBs, it switches to R/O mode. UBI also loudly informs
55 * about such PEBs every time the MTD device is attached.
57 * However, it is difficult to reliably distinguish between these types of
58 * corruptions and UBI's strategy is as follows (in case of attaching by
59 * scanning). UBI assumes corruption type 2 if the VID header is corrupted and
60 * the data area does not contain all 0xFFs, and there were no bit-flips or
61 * integrity errors (e.g., ECC errors in case of NAND) while reading the data
62 * area. Otherwise UBI assumes corruption type 1. So the decision criteria
64 * o If the data area contains only 0xFFs, there are no data, and it is safe
65 * to just erase this PEB - this is corruption type 1.
66 * o If the data area has bit-flips or data integrity errors (ECC errors on
67 * NAND), it is probably a PEB which was being erased when power cut
68 * happened, so this is corruption type 1. However, this is just a guess,
69 * which might be wrong.
70 * o Otherwise this is corruption type 2.
74 #include <linux/err.h>
75 #include <linux/slab.h>
76 #include <linux/crc32.h>
77 #include <linux/random.h>
80 #include <linux/err.h>
83 #include <linux/math64.h>
85 #include <ubi_uboot.h>
88 static int self_check_ai(struct ubi_device *ubi, struct ubi_attach_info *ai);
90 /* Temporary variables used during scanning */
91 static struct ubi_ec_hdr *ech;
92 static struct ubi_vid_hdr *vidh;
95 * add_to_list - add physical eraseblock to a list.
96 * @ai: attaching information
97 * @pnum: physical eraseblock number to add
98 * @vol_id: the last used volume id for the PEB
99 * @lnum: the last used LEB number for the PEB
100 * @ec: erase counter of the physical eraseblock
101 * @to_head: if not zero, add to the head of the list
102 * @list: the list to add to
104 * This function allocates a 'struct ubi_ainf_peb' object for physical
105 * eraseblock @pnum and adds it to the "free", "erase", or "alien" lists.
106 * It stores the @lnum and @vol_id alongside, which can both be
107 * %UBI_UNKNOWN if they are not available, not readable, or not assigned.
108 * If @to_head is not zero, PEB will be added to the head of the list, which
109 * basically means it will be processed first later. E.g., we add corrupted
110 * PEBs (corrupted due to power cuts) to the head of the erase list to make
111 * sure we erase them first and get rid of corruptions ASAP. This function
112 * returns zero in case of success and a negative error code in case of
115 static int add_to_list(struct ubi_attach_info *ai, int pnum, int vol_id,
116 int lnum, int ec, int to_head, struct list_head *list)
118 struct ubi_ainf_peb *aeb;
120 if (list == &ai->free) {
121 dbg_bld("add to free: PEB %d, EC %d", pnum, ec);
122 } else if (list == &ai->erase) {
123 dbg_bld("add to erase: PEB %d, EC %d", pnum, ec);
124 } else if (list == &ai->alien) {
125 dbg_bld("add to alien: PEB %d, EC %d", pnum, ec);
126 ai->alien_peb_count += 1;
130 aeb = kmem_cache_alloc(ai->aeb_slab_cache, GFP_KERNEL);
135 aeb->vol_id = vol_id;
139 list_add(&aeb->u.list, list);
141 list_add_tail(&aeb->u.list, list);
146 * add_corrupted - add a corrupted physical eraseblock.
147 * @ai: attaching information
148 * @pnum: physical eraseblock number to add
149 * @ec: erase counter of the physical eraseblock
151 * This function allocates a 'struct ubi_ainf_peb' object for a corrupted
152 * physical eraseblock @pnum and adds it to the 'corr' list. The corruption
153 * was presumably not caused by a power cut. Returns zero in case of success
154 * and a negative error code in case of failure.
156 static int add_corrupted(struct ubi_attach_info *ai, int pnum, int ec)
158 struct ubi_ainf_peb *aeb;
160 dbg_bld("add to corrupted: PEB %d, EC %d", pnum, ec);
162 aeb = kmem_cache_alloc(ai->aeb_slab_cache, GFP_KERNEL);
166 ai->corr_peb_count += 1;
169 list_add(&aeb->u.list, &ai->corr);
174 * validate_vid_hdr - check volume identifier header.
175 * @ubi: UBI device description object
176 * @vid_hdr: the volume identifier header to check
177 * @av: information about the volume this logical eraseblock belongs to
178 * @pnum: physical eraseblock number the VID header came from
180 * This function checks that data stored in @vid_hdr is consistent. Returns
181 * non-zero if an inconsistency was found and zero if not.
183 * Note, UBI does sanity check of everything it reads from the flash media.
184 * Most of the checks are done in the I/O sub-system. Here we check that the
185 * information in the VID header is consistent to the information in other VID
186 * headers of the same volume.
188 static int validate_vid_hdr(const struct ubi_device *ubi,
189 const struct ubi_vid_hdr *vid_hdr,
190 const struct ubi_ainf_volume *av, int pnum)
192 int vol_type = vid_hdr->vol_type;
193 int vol_id = be32_to_cpu(vid_hdr->vol_id);
194 int used_ebs = be32_to_cpu(vid_hdr->used_ebs);
195 int data_pad = be32_to_cpu(vid_hdr->data_pad);
197 if (av->leb_count != 0) {
201 * This is not the first logical eraseblock belonging to this
202 * volume. Ensure that the data in its VID header is consistent
203 * to the data in previous logical eraseblock headers.
206 if (vol_id != av->vol_id) {
207 ubi_err(ubi, "inconsistent vol_id");
211 if (av->vol_type == UBI_STATIC_VOLUME)
212 av_vol_type = UBI_VID_STATIC;
214 av_vol_type = UBI_VID_DYNAMIC;
216 if (vol_type != av_vol_type) {
217 ubi_err(ubi, "inconsistent vol_type");
221 if (used_ebs != av->used_ebs) {
222 ubi_err(ubi, "inconsistent used_ebs");
226 if (data_pad != av->data_pad) {
227 ubi_err(ubi, "inconsistent data_pad");
235 ubi_err(ubi, "inconsistent VID header at PEB %d", pnum);
236 ubi_dump_vid_hdr(vid_hdr);
242 * add_volume - add volume to the attaching information.
243 * @ai: attaching information
244 * @vol_id: ID of the volume to add
245 * @pnum: physical eraseblock number
246 * @vid_hdr: volume identifier header
248 * If the volume corresponding to the @vid_hdr logical eraseblock is already
249 * present in the attaching information, this function does nothing. Otherwise
250 * it adds corresponding volume to the attaching information. Returns a pointer
251 * to the allocated "av" object in case of success and a negative error code in
254 static struct ubi_ainf_volume *add_volume(struct ubi_attach_info *ai,
255 int vol_id, int pnum,
256 const struct ubi_vid_hdr *vid_hdr)
258 struct ubi_ainf_volume *av;
259 struct rb_node **p = &ai->volumes.rb_node, *parent = NULL;
261 ubi_assert(vol_id == be32_to_cpu(vid_hdr->vol_id));
263 /* Walk the volume RB-tree to look if this volume is already present */
266 av = rb_entry(parent, struct ubi_ainf_volume, rb);
268 if (vol_id == av->vol_id)
271 if (vol_id > av->vol_id)
277 /* The volume is absent - add it */
278 av = kmalloc(sizeof(struct ubi_ainf_volume), GFP_KERNEL);
280 return ERR_PTR(-ENOMEM);
282 av->highest_lnum = av->leb_count = 0;
285 av->used_ebs = be32_to_cpu(vid_hdr->used_ebs);
286 av->data_pad = be32_to_cpu(vid_hdr->data_pad);
287 av->compat = vid_hdr->compat;
288 av->vol_type = vid_hdr->vol_type == UBI_VID_DYNAMIC ? UBI_DYNAMIC_VOLUME
290 if (vol_id > ai->highest_vol_id)
291 ai->highest_vol_id = vol_id;
293 rb_link_node(&av->rb, parent, p);
294 rb_insert_color(&av->rb, &ai->volumes);
296 dbg_bld("added volume %d", vol_id);
301 * ubi_compare_lebs - find out which logical eraseblock is newer.
302 * @ubi: UBI device description object
303 * @aeb: first logical eraseblock to compare
304 * @pnum: physical eraseblock number of the second logical eraseblock to
306 * @vid_hdr: volume identifier header of the second logical eraseblock
308 * This function compares 2 copies of a LEB and informs which one is newer. In
309 * case of success this function returns a positive value, in case of failure, a
310 * negative error code is returned. The success return codes use the following
312 * o bit 0 is cleared: the first PEB (described by @aeb) is newer than the
313 * second PEB (described by @pnum and @vid_hdr);
314 * o bit 0 is set: the second PEB is newer;
315 * o bit 1 is cleared: no bit-flips were detected in the newer LEB;
316 * o bit 1 is set: bit-flips were detected in the newer LEB;
317 * o bit 2 is cleared: the older LEB is not corrupted;
318 * o bit 2 is set: the older LEB is corrupted.
320 int ubi_compare_lebs(struct ubi_device *ubi, const struct ubi_ainf_peb *aeb,
321 int pnum, const struct ubi_vid_hdr *vid_hdr)
323 int len, err, second_is_newer, bitflips = 0, corrupted = 0;
324 uint32_t data_crc, crc;
325 struct ubi_vid_hdr *vh = NULL;
326 unsigned long long sqnum2 = be64_to_cpu(vid_hdr->sqnum);
328 if (sqnum2 == aeb->sqnum) {
330 * This must be a really ancient UBI image which has been
331 * created before sequence numbers support has been added. At
332 * that times we used 32-bit LEB versions stored in logical
333 * eraseblocks. That was before UBI got into mainline. We do not
334 * support these images anymore. Well, those images still work,
335 * but only if no unclean reboots happened.
337 ubi_err(ubi, "unsupported on-flash UBI format");
341 /* Obviously the LEB with lower sequence counter is older */
342 second_is_newer = (sqnum2 > aeb->sqnum);
345 * Now we know which copy is newer. If the copy flag of the PEB with
346 * newer version is not set, then we just return, otherwise we have to
347 * check data CRC. For the second PEB we already have the VID header,
348 * for the first one - we'll need to re-read it from flash.
350 * Note: this may be optimized so that we wouldn't read twice.
353 if (second_is_newer) {
354 if (!vid_hdr->copy_flag) {
355 /* It is not a copy, so it is newer */
356 dbg_bld("second PEB %d is newer, copy_flag is unset",
361 if (!aeb->copy_flag) {
362 /* It is not a copy, so it is newer */
363 dbg_bld("first PEB %d is newer, copy_flag is unset",
365 return bitflips << 1;
368 vh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
373 err = ubi_io_read_vid_hdr(ubi, pnum, vh, 0);
375 if (err == UBI_IO_BITFLIPS)
378 ubi_err(ubi, "VID of PEB %d header is bad, but it was OK earlier, err %d",
390 /* Read the data of the copy and check the CRC */
392 len = be32_to_cpu(vid_hdr->data_size);
394 mutex_lock(&ubi->buf_mutex);
395 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, len);
396 if (err && err != UBI_IO_BITFLIPS && !mtd_is_eccerr(err))
399 data_crc = be32_to_cpu(vid_hdr->data_crc);
400 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, len);
401 if (crc != data_crc) {
402 dbg_bld("PEB %d CRC error: calculated %#08x, must be %#08x",
403 pnum, crc, data_crc);
406 second_is_newer = !second_is_newer;
408 dbg_bld("PEB %d CRC is OK", pnum);
411 mutex_unlock(&ubi->buf_mutex);
413 ubi_free_vid_hdr(ubi, vh);
416 dbg_bld("second PEB %d is newer, copy_flag is set", pnum);
418 dbg_bld("first PEB %d is newer, copy_flag is set", pnum);
420 return second_is_newer | (bitflips << 1) | (corrupted << 2);
423 mutex_unlock(&ubi->buf_mutex);
425 ubi_free_vid_hdr(ubi, vh);
430 * ubi_add_to_av - add used physical eraseblock to the attaching information.
431 * @ubi: UBI device description object
432 * @ai: attaching information
433 * @pnum: the physical eraseblock number
435 * @vid_hdr: the volume identifier header
436 * @bitflips: if bit-flips were detected when this physical eraseblock was read
438 * This function adds information about a used physical eraseblock to the
439 * 'used' tree of the corresponding volume. The function is rather complex
440 * because it has to handle cases when this is not the first physical
441 * eraseblock belonging to the same logical eraseblock, and the newer one has
442 * to be picked, while the older one has to be dropped. This function returns
443 * zero in case of success and a negative error code in case of failure.
445 int ubi_add_to_av(struct ubi_device *ubi, struct ubi_attach_info *ai, int pnum,
446 int ec, const struct ubi_vid_hdr *vid_hdr, int bitflips)
448 int err, vol_id, lnum;
449 unsigned long long sqnum;
450 struct ubi_ainf_volume *av;
451 struct ubi_ainf_peb *aeb;
452 struct rb_node **p, *parent = NULL;
454 vol_id = be32_to_cpu(vid_hdr->vol_id);
455 lnum = be32_to_cpu(vid_hdr->lnum);
456 sqnum = be64_to_cpu(vid_hdr->sqnum);
458 dbg_bld("PEB %d, LEB %d:%d, EC %d, sqnum %llu, bitflips %d",
459 pnum, vol_id, lnum, ec, sqnum, bitflips);
461 av = add_volume(ai, vol_id, pnum, vid_hdr);
465 if (ai->max_sqnum < sqnum)
466 ai->max_sqnum = sqnum;
469 * Walk the RB-tree of logical eraseblocks of volume @vol_id to look
470 * if this is the first instance of this logical eraseblock or not.
472 p = &av->root.rb_node;
477 aeb = rb_entry(parent, struct ubi_ainf_peb, u.rb);
478 if (lnum != aeb->lnum) {
479 if (lnum < aeb->lnum)
487 * There is already a physical eraseblock describing the same
488 * logical eraseblock present.
491 dbg_bld("this LEB already exists: PEB %d, sqnum %llu, EC %d",
492 aeb->pnum, aeb->sqnum, aeb->ec);
495 * Make sure that the logical eraseblocks have different
496 * sequence numbers. Otherwise the image is bad.
498 * However, if the sequence number is zero, we assume it must
499 * be an ancient UBI image from the era when UBI did not have
500 * sequence numbers. We still can attach these images, unless
501 * there is a need to distinguish between old and new
502 * eraseblocks, in which case we'll refuse the image in
503 * 'ubi_compare_lebs()'. In other words, we attach old clean
504 * images, but refuse attaching old images with duplicated
505 * logical eraseblocks because there was an unclean reboot.
507 if (aeb->sqnum == sqnum && sqnum != 0) {
508 ubi_err(ubi, "two LEBs with same sequence number %llu",
510 ubi_dump_aeb(aeb, 0);
511 ubi_dump_vid_hdr(vid_hdr);
516 * Now we have to drop the older one and preserve the newer
519 cmp_res = ubi_compare_lebs(ubi, aeb, pnum, vid_hdr);
525 * This logical eraseblock is newer than the one
528 err = validate_vid_hdr(ubi, vid_hdr, av, pnum);
532 err = add_to_list(ai, aeb->pnum, aeb->vol_id,
533 aeb->lnum, aeb->ec, cmp_res & 4,
540 aeb->vol_id = vol_id;
542 aeb->scrub = ((cmp_res & 2) || bitflips);
543 aeb->copy_flag = vid_hdr->copy_flag;
546 if (av->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(ai, pnum, vol_id, lnum, ec,
557 cmp_res & 4, &ai->erase);
562 * We've met this logical eraseblock for the first time, add it to the
563 * attaching information.
566 err = validate_vid_hdr(ubi, vid_hdr, av, pnum);
570 aeb = kmem_cache_alloc(ai->aeb_slab_cache, GFP_KERNEL);
576 aeb->vol_id = vol_id;
578 aeb->scrub = bitflips;
579 aeb->copy_flag = vid_hdr->copy_flag;
582 if (av->highest_lnum <= lnum) {
583 av->highest_lnum = lnum;
584 av->last_data_size = be32_to_cpu(vid_hdr->data_size);
588 rb_link_node(&aeb->u.rb, parent, p);
589 rb_insert_color(&aeb->u.rb, &av->root);
594 * ubi_find_av - find volume in the attaching information.
595 * @ai: attaching information
596 * @vol_id: the requested volume ID
598 * This function returns a pointer to the volume description or %NULL if there
599 * are no data about this volume in the attaching information.
601 struct ubi_ainf_volume *ubi_find_av(const struct ubi_attach_info *ai,
604 struct ubi_ainf_volume *av;
605 struct rb_node *p = ai->volumes.rb_node;
608 av = rb_entry(p, struct ubi_ainf_volume, rb);
610 if (vol_id == av->vol_id)
613 if (vol_id > av->vol_id)
623 * ubi_remove_av - delete attaching information about a volume.
624 * @ai: attaching information
625 * @av: the volume attaching information to delete
627 void ubi_remove_av(struct ubi_attach_info *ai, struct ubi_ainf_volume *av)
630 struct ubi_ainf_peb *aeb;
632 dbg_bld("remove attaching information about volume %d", av->vol_id);
634 while ((rb = rb_first(&av->root))) {
635 aeb = rb_entry(rb, struct ubi_ainf_peb, u.rb);
636 rb_erase(&aeb->u.rb, &av->root);
637 list_add_tail(&aeb->u.list, &ai->erase);
640 rb_erase(&av->rb, &ai->volumes);
646 * early_erase_peb - erase a physical eraseblock.
647 * @ubi: UBI device description object
648 * @ai: attaching information
649 * @pnum: physical eraseblock number to erase;
650 * @ec: erase counter value to write (%UBI_UNKNOWN if it is unknown)
652 * This function erases physical eraseblock 'pnum', and writes the erase
653 * counter header to it. This function should only be used on UBI device
654 * initialization stages, when the EBA sub-system had not been yet initialized.
655 * This function returns zero in case of success and a negative error code in
658 static int early_erase_peb(struct ubi_device *ubi,
659 const struct ubi_attach_info *ai, int pnum, int ec)
662 struct ubi_ec_hdr *ec_hdr;
664 if ((long long)ec >= UBI_MAX_ERASECOUNTER) {
666 * Erase counter overflow. Upgrade UBI and use 64-bit
667 * erase counters internally.
669 ubi_err(ubi, "erase counter overflow at PEB %d, EC %d",
674 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
678 ec_hdr->ec = cpu_to_be64(ec);
680 err = ubi_io_sync_erase(ubi, pnum, 0);
684 err = ubi_io_write_ec_hdr(ubi, pnum, ec_hdr);
692 * ubi_early_get_peb - get a free physical eraseblock.
693 * @ubi: UBI device description object
694 * @ai: attaching information
696 * This function returns a free physical eraseblock. It is supposed to be
697 * called on the UBI initialization stages when the wear-leveling sub-system is
698 * not initialized yet. This function picks a physical eraseblocks from one of
699 * the lists, writes the EC header if it is needed, and removes it from the
702 * This function returns a pointer to the "aeb" of the found free PEB in case
703 * of success and an error code in case of failure.
705 struct ubi_ainf_peb *ubi_early_get_peb(struct ubi_device *ubi,
706 struct ubi_attach_info *ai)
709 struct ubi_ainf_peb *aeb, *tmp_aeb;
711 if (!list_empty(&ai->free)) {
712 aeb = list_entry(ai->free.next, struct ubi_ainf_peb, u.list);
713 list_del(&aeb->u.list);
714 dbg_bld("return free PEB %d, EC %d", aeb->pnum, aeb->ec);
719 * We try to erase the first physical eraseblock from the erase list
720 * and pick it if we succeed, or try to erase the next one if not. And
721 * so forth. We don't want to take care about bad eraseblocks here -
722 * they'll be handled later.
724 list_for_each_entry_safe(aeb, tmp_aeb, &ai->erase, u.list) {
725 if (aeb->ec == UBI_UNKNOWN)
726 aeb->ec = ai->mean_ec;
728 err = early_erase_peb(ubi, ai, aeb->pnum, aeb->ec+1);
733 list_del(&aeb->u.list);
734 dbg_bld("return PEB %d, EC %d", aeb->pnum, aeb->ec);
738 ubi_err(ubi, "no free eraseblocks");
739 return ERR_PTR(-ENOSPC);
743 * check_corruption - check the data area of PEB.
744 * @ubi: UBI device description object
745 * @vid_hdr: the (corrupted) VID header of this PEB
746 * @pnum: the physical eraseblock number to check
748 * This is a helper function which is used to distinguish between VID header
749 * corruptions caused by power cuts and other reasons. If the PEB contains only
750 * 0xFF bytes in the data area, the VID header is most probably corrupted
751 * because of a power cut (%0 is returned in this case). Otherwise, it was
752 * probably corrupted for some other reasons (%1 is returned in this case). A
753 * negative error code is returned if a read error occurred.
755 * If the corruption reason was a power cut, UBI can safely erase this PEB.
756 * Otherwise, it should preserve it to avoid possibly destroying important
759 static int check_corruption(struct ubi_device *ubi, struct ubi_vid_hdr *vid_hdr,
764 mutex_lock(&ubi->buf_mutex);
765 memset(ubi->peb_buf, 0x00, ubi->leb_size);
767 err = ubi_io_read(ubi, ubi->peb_buf, pnum, ubi->leb_start,
769 if (err == UBI_IO_BITFLIPS || mtd_is_eccerr(err)) {
771 * Bit-flips or integrity errors while reading the data area.
772 * It is difficult to say for sure what type of corruption is
773 * this, but presumably a power cut happened while this PEB was
774 * erased, so it became unstable and corrupted, and should be
784 if (ubi_check_pattern(ubi->peb_buf, 0xFF, ubi->leb_size))
787 ubi_err(ubi, "PEB %d contains corrupted VID header, and the data does not contain all 0xFF",
789 ubi_err(ubi, "this may be a non-UBI PEB or a severe VID header corruption which requires manual inspection");
790 ubi_dump_vid_hdr(vid_hdr);
791 pr_err("hexdump of PEB %d offset %d, length %d",
792 pnum, ubi->leb_start, ubi->leb_size);
793 ubi_dbg_print_hex_dump(KERN_DEBUG, "", DUMP_PREFIX_OFFSET, 32, 1,
794 ubi->peb_buf, ubi->leb_size, 1);
798 mutex_unlock(&ubi->buf_mutex);
803 * scan_peb - scan and process UBI headers of a PEB.
804 * @ubi: UBI device description object
805 * @ai: attaching information
806 * @pnum: the physical eraseblock number
807 * @vid: The volume ID of the found volume will be stored in this pointer
808 * @sqnum: The sqnum of the found volume will be stored in this pointer
810 * This function reads UBI headers of PEB @pnum, checks them, and adds
811 * information about this PEB to the corresponding list or RB-tree in the
812 * "attaching info" structure. Returns zero if the physical eraseblock was
813 * successfully handled and a negative error code in case of failure.
815 static int scan_peb(struct ubi_device *ubi, struct ubi_attach_info *ai,
816 int pnum, int *vid, unsigned long long *sqnum)
818 long long uninitialized_var(ec);
819 int err, bitflips = 0, vol_id = -1, ec_err = 0;
821 dbg_bld("scan PEB %d", pnum);
823 /* Skip bad physical eraseblocks */
824 err = ubi_io_is_bad(ubi, pnum);
828 ai->bad_peb_count += 1;
832 err = ubi_io_read_ec_hdr(ubi, pnum, ech, 0);
838 case UBI_IO_BITFLIPS:
842 ai->empty_peb_count += 1;
843 return add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
844 UBI_UNKNOWN, 0, &ai->erase);
845 case UBI_IO_FF_BITFLIPS:
846 ai->empty_peb_count += 1;
847 return add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
848 UBI_UNKNOWN, 1, &ai->erase);
849 case UBI_IO_BAD_HDR_EBADMSG:
852 * We have to also look at the VID header, possibly it is not
853 * corrupted. Set %bitflips flag in order to make this PEB be
854 * moved and EC be re-created.
861 ubi_err(ubi, "'ubi_io_read_ec_hdr()' returned unknown code %d",
869 /* Make sure UBI version is OK */
870 if (ech->version != UBI_VERSION) {
871 ubi_err(ubi, "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(ubi, "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);
904 ubi->image_seq = image_seq;
905 if (image_seq && ubi->image_seq != image_seq) {
906 ubi_err(ubi, "bad image sequence number %d in PEB %d, expected %d",
907 image_seq, pnum, ubi->image_seq);
908 ubi_dump_ec_hdr(ech);
913 /* OK, we've done with the EC header, let's look at the VID header */
915 err = ubi_io_read_vid_hdr(ubi, pnum, vidh, 0);
921 case UBI_IO_BITFLIPS:
924 case UBI_IO_BAD_HDR_EBADMSG:
925 if (ec_err == UBI_IO_BAD_HDR_EBADMSG)
927 * Both EC and VID headers are corrupted and were read
928 * with data integrity error, probably this is a bad
929 * PEB, bit it is not marked as bad yet. This may also
930 * be a result of power cut during erasure.
932 ai->maybe_bad_peb_count += 1;
936 * Both headers are corrupted. There is a possibility
937 * that this a valid UBI PEB which has corresponding
938 * LEB, but the headers are corrupted. However, it is
939 * impossible to distinguish it from a PEB which just
940 * contains garbage because of a power cut during erase
941 * operation. So we just schedule this PEB for erasure.
943 * Besides, in case of NOR flash, we deliberately
944 * corrupt both headers because NOR flash erasure is
945 * slow and can start from the end.
950 * The EC was OK, but the VID header is corrupted. We
951 * have to check what is in the data area.
953 err = check_corruption(ubi, vidh, pnum);
958 /* This corruption is caused by a power cut */
959 err = add_to_list(ai, pnum, UBI_UNKNOWN,
960 UBI_UNKNOWN, ec, 1, &ai->erase);
962 /* This is an unexpected corruption */
963 err = add_corrupted(ai, pnum, ec);
967 case UBI_IO_FF_BITFLIPS:
968 err = add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
974 if (ec_err || bitflips)
975 err = add_to_list(ai, pnum, UBI_UNKNOWN,
976 UBI_UNKNOWN, ec, 1, &ai->erase);
978 err = add_to_list(ai, pnum, UBI_UNKNOWN,
979 UBI_UNKNOWN, ec, 0, &ai->free);
984 ubi_err(ubi, "'ubi_io_read_vid_hdr()' returned unknown code %d",
989 vol_id = be32_to_cpu(vidh->vol_id);
993 *sqnum = be64_to_cpu(vidh->sqnum);
994 if (vol_id > UBI_MAX_VOLUMES && vol_id != UBI_LAYOUT_VOLUME_ID) {
995 int lnum = be32_to_cpu(vidh->lnum);
997 /* Unsupported internal volume */
998 switch (vidh->compat) {
999 case UBI_COMPAT_DELETE:
1000 if (vol_id != UBI_FM_SB_VOLUME_ID
1001 && vol_id != UBI_FM_DATA_VOLUME_ID) {
1002 ubi_msg(ubi, "\"delete\" compatible internal volume %d:%d found, will remove it",
1005 err = add_to_list(ai, pnum, vol_id, lnum,
1012 ubi_msg(ubi, "read-only compatible internal volume %d:%d found, switch to read-only mode",
1017 case UBI_COMPAT_PRESERVE:
1018 ubi_msg(ubi, "\"preserve\" compatible internal volume %d:%d found",
1020 err = add_to_list(ai, pnum, vol_id, lnum,
1026 case UBI_COMPAT_REJECT:
1027 ubi_err(ubi, "incompatible internal volume %d:%d found",
1034 ubi_warn(ubi, "valid VID header but corrupted EC header at PEB %d",
1036 err = ubi_add_to_av(ubi, ai, pnum, ec, vidh, bitflips);
1044 if (ec > ai->max_ec)
1046 if (ec < ai->min_ec)
1054 * late_analysis - analyze the overall situation with PEB.
1055 * @ubi: UBI device description object
1056 * @ai: attaching information
1058 * This is a helper function which takes a look what PEBs we have after we
1059 * gather information about all of them ("ai" is compete). It decides whether
1060 * the flash is empty and should be formatted of whether there are too many
1061 * corrupted PEBs and we should not attach this MTD device. Returns zero if we
1062 * should proceed with attaching the MTD device, and %-EINVAL if we should not.
1064 static int late_analysis(struct ubi_device *ubi, struct ubi_attach_info *ai)
1066 struct ubi_ainf_peb *aeb;
1067 int max_corr, peb_count;
1069 peb_count = ubi->peb_count - ai->bad_peb_count - ai->alien_peb_count;
1070 max_corr = peb_count / 20 ?: 8;
1073 * Few corrupted PEBs is not a problem and may be just a result of
1074 * unclean reboots. However, many of them may indicate some problems
1075 * with the flash HW or driver.
1077 if (ai->corr_peb_count) {
1078 ubi_err(ubi, "%d PEBs are corrupted and preserved",
1079 ai->corr_peb_count);
1080 pr_err("Corrupted PEBs are:");
1081 list_for_each_entry(aeb, &ai->corr, u.list)
1082 pr_cont(" %d", aeb->pnum);
1086 * If too many PEBs are corrupted, we refuse attaching,
1087 * otherwise, only print a warning.
1089 if (ai->corr_peb_count >= max_corr) {
1090 ubi_err(ubi, "too many corrupted PEBs, refusing");
1095 if (ai->empty_peb_count + ai->maybe_bad_peb_count == peb_count) {
1097 * All PEBs are empty, or almost all - a couple PEBs look like
1098 * they may be bad PEBs which were not marked as bad yet.
1100 * This piece of code basically tries to distinguish between
1101 * the following situations:
1103 * 1. Flash is empty, but there are few bad PEBs, which are not
1104 * marked as bad so far, and which were read with error. We
1105 * want to go ahead and format this flash. While formatting,
1106 * the faulty PEBs will probably be marked as bad.
1108 * 2. Flash contains non-UBI data and we do not want to format
1109 * it and destroy possibly important information.
1111 if (ai->maybe_bad_peb_count <= 2) {
1113 ubi_msg(ubi, "empty MTD device detected");
1114 get_random_bytes(&ubi->image_seq,
1115 sizeof(ubi->image_seq));
1117 ubi_err(ubi, "MTD device is not UBI-formatted and possibly contains non-UBI data - refusing it");
1127 * destroy_av - free volume attaching information.
1128 * @av: volume attaching information
1129 * @ai: attaching information
1131 * This function destroys the volume attaching information.
1133 static void destroy_av(struct ubi_attach_info *ai, struct ubi_ainf_volume *av)
1135 struct ubi_ainf_peb *aeb;
1136 struct rb_node *this = av->root.rb_node;
1140 this = this->rb_left;
1141 else if (this->rb_right)
1142 this = this->rb_right;
1144 aeb = rb_entry(this, struct ubi_ainf_peb, u.rb);
1145 this = rb_parent(this);
1147 if (this->rb_left == &aeb->u.rb)
1148 this->rb_left = NULL;
1150 this->rb_right = NULL;
1153 kmem_cache_free(ai->aeb_slab_cache, aeb);
1160 * destroy_ai - destroy attaching information.
1161 * @ai: attaching information
1163 static void destroy_ai(struct ubi_attach_info *ai)
1165 struct ubi_ainf_peb *aeb, *aeb_tmp;
1166 struct ubi_ainf_volume *av;
1169 list_for_each_entry_safe(aeb, aeb_tmp, &ai->alien, u.list) {
1170 list_del(&aeb->u.list);
1171 kmem_cache_free(ai->aeb_slab_cache, aeb);
1173 list_for_each_entry_safe(aeb, aeb_tmp, &ai->erase, u.list) {
1174 list_del(&aeb->u.list);
1175 kmem_cache_free(ai->aeb_slab_cache, aeb);
1177 list_for_each_entry_safe(aeb, aeb_tmp, &ai->corr, u.list) {
1178 list_del(&aeb->u.list);
1179 kmem_cache_free(ai->aeb_slab_cache, aeb);
1181 list_for_each_entry_safe(aeb, aeb_tmp, &ai->free, u.list) {
1182 list_del(&aeb->u.list);
1183 kmem_cache_free(ai->aeb_slab_cache, aeb);
1186 /* Destroy the volume RB-tree */
1187 rb = ai->volumes.rb_node;
1191 else if (rb->rb_right)
1194 av = rb_entry(rb, struct ubi_ainf_volume, rb);
1198 if (rb->rb_left == &av->rb)
1201 rb->rb_right = NULL;
1208 if (ai->aeb_slab_cache)
1209 kmem_cache_destroy(ai->aeb_slab_cache);
1215 * scan_all - scan entire MTD device.
1216 * @ubi: UBI device description object
1217 * @ai: attach info object
1218 * @start: start scanning at this PEB
1220 * This function does full scanning of an MTD device and returns complete
1221 * information about it in form of a "struct ubi_attach_info" object. In case
1222 * of failure, an error code is returned.
1224 static int scan_all(struct ubi_device *ubi, struct ubi_attach_info *ai,
1228 struct rb_node *rb1, *rb2;
1229 struct ubi_ainf_volume *av;
1230 struct ubi_ainf_peb *aeb;
1234 ech = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
1238 vidh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
1242 for (pnum = start; pnum < ubi->peb_count; pnum++) {
1245 dbg_gen("process PEB %d", pnum);
1246 err = scan_peb(ubi, ai, pnum, NULL, NULL);
1251 ubi_msg(ubi, "scanning is finished");
1253 /* Calculate mean erase counter */
1255 ai->mean_ec = div_u64(ai->ec_sum, ai->ec_count);
1257 err = late_analysis(ubi, ai);
1262 * In case of unknown erase counter we use the mean erase counter
1265 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1266 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb)
1267 if (aeb->ec == UBI_UNKNOWN)
1268 aeb->ec = ai->mean_ec;
1271 list_for_each_entry(aeb, &ai->free, u.list) {
1272 if (aeb->ec == UBI_UNKNOWN)
1273 aeb->ec = ai->mean_ec;
1276 list_for_each_entry(aeb, &ai->corr, u.list)
1277 if (aeb->ec == UBI_UNKNOWN)
1278 aeb->ec = ai->mean_ec;
1280 list_for_each_entry(aeb, &ai->erase, u.list)
1281 if (aeb->ec == UBI_UNKNOWN)
1282 aeb->ec = ai->mean_ec;
1284 err = self_check_ai(ubi, ai);
1288 ubi_free_vid_hdr(ubi, vidh);
1294 ubi_free_vid_hdr(ubi, vidh);
1300 static struct ubi_attach_info *alloc_ai(void)
1302 struct ubi_attach_info *ai;
1304 ai = kzalloc(sizeof(struct ubi_attach_info), GFP_KERNEL);
1308 INIT_LIST_HEAD(&ai->corr);
1309 INIT_LIST_HEAD(&ai->free);
1310 INIT_LIST_HEAD(&ai->erase);
1311 INIT_LIST_HEAD(&ai->alien);
1312 ai->volumes = RB_ROOT;
1313 ai->aeb_slab_cache = kmem_cache_create("ubi_aeb_slab_cache",
1314 sizeof(struct ubi_ainf_peb),
1316 if (!ai->aeb_slab_cache) {
1324 #ifdef CONFIG_MTD_UBI_FASTMAP
1327 * scan_fastmap - try to find a fastmap and attach from it.
1328 * @ubi: UBI device description object
1329 * @ai: attach info object
1331 * Returns 0 on success, negative return values indicate an internal
1333 * UBI_NO_FASTMAP denotes that no fastmap was found.
1334 * UBI_BAD_FASTMAP denotes that the found fastmap was invalid.
1336 static int scan_fast(struct ubi_device *ubi, struct ubi_attach_info **ai)
1338 int err, pnum, fm_anchor = -1;
1339 unsigned long long max_sqnum = 0;
1343 ech = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
1347 vidh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
1351 for (pnum = 0; pnum < UBI_FM_MAX_START; pnum++) {
1353 unsigned long long sqnum = -1;
1356 dbg_gen("process PEB %d", pnum);
1357 err = scan_peb(ubi, *ai, pnum, &vol_id, &sqnum);
1361 if (vol_id == UBI_FM_SB_VOLUME_ID && sqnum > max_sqnum) {
1367 ubi_free_vid_hdr(ubi, vidh);
1371 return UBI_NO_FASTMAP;
1378 return ubi_scan_fastmap(ubi, *ai, fm_anchor);
1381 ubi_free_vid_hdr(ubi, vidh);
1391 * ubi_attach - attach an MTD device.
1392 * @ubi: UBI device descriptor
1393 * @force_scan: if set to non-zero attach by scanning
1395 * This function returns zero in case of success and a negative error code in
1398 int ubi_attach(struct ubi_device *ubi, int force_scan)
1401 struct ubi_attach_info *ai;
1407 #ifdef CONFIG_MTD_UBI_FASTMAP
1408 /* On small flash devices we disable fastmap in any case. */
1409 if ((int)mtd_div_by_eb(ubi->mtd->size, ubi->mtd) <= UBI_FM_MAX_START) {
1410 ubi->fm_disabled = 1;
1415 err = scan_all(ubi, ai, 0);
1417 err = scan_fast(ubi, &ai);
1418 if (err > 0 || mtd_is_eccerr(err)) {
1419 if (err != UBI_NO_FASTMAP) {
1425 err = scan_all(ubi, ai, 0);
1427 err = scan_all(ubi, ai, UBI_FM_MAX_START);
1432 err = scan_all(ubi, ai, 0);
1437 ubi->bad_peb_count = ai->bad_peb_count;
1438 ubi->good_peb_count = ubi->peb_count - ubi->bad_peb_count;
1439 ubi->corr_peb_count = ai->corr_peb_count;
1440 ubi->max_ec = ai->max_ec;
1441 ubi->mean_ec = ai->mean_ec;
1442 dbg_gen("max. sequence number: %llu", ai->max_sqnum);
1444 err = ubi_read_volume_table(ubi, ai);
1448 err = ubi_wl_init(ubi, ai);
1452 err = ubi_eba_init(ubi, ai);
1456 #ifdef CONFIG_MTD_UBI_FASTMAP
1457 if (ubi->fm && ubi_dbg_chk_fastmap(ubi)) {
1458 struct ubi_attach_info *scan_ai;
1460 scan_ai = alloc_ai();
1466 err = scan_all(ubi, scan_ai, 0);
1468 destroy_ai(scan_ai);
1472 err = self_check_eba(ubi, ai, scan_ai);
1473 destroy_ai(scan_ai);
1486 ubi_free_internal_volumes(ubi);
1494 * self_check_ai - check the attaching information.
1495 * @ubi: UBI device description object
1496 * @ai: attaching information
1498 * This function returns zero if the attaching information is all right, and a
1499 * negative error code if not or if an error occurred.
1501 static int self_check_ai(struct ubi_device *ubi, struct ubi_attach_info *ai)
1503 int pnum, err, vols_found = 0;
1504 struct rb_node *rb1, *rb2;
1505 struct ubi_ainf_volume *av;
1506 struct ubi_ainf_peb *aeb, *last_aeb;
1509 if (!ubi_dbg_chk_gen(ubi))
1513 * At first, check that attaching information is OK.
1515 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1523 ubi_err(ubi, "bad is_empty flag");
1527 if (av->vol_id < 0 || av->highest_lnum < 0 ||
1528 av->leb_count < 0 || av->vol_type < 0 || av->used_ebs < 0 ||
1529 av->data_pad < 0 || av->last_data_size < 0) {
1530 ubi_err(ubi, "negative values");
1534 if (av->vol_id >= UBI_MAX_VOLUMES &&
1535 av->vol_id < UBI_INTERNAL_VOL_START) {
1536 ubi_err(ubi, "bad vol_id");
1540 if (av->vol_id > ai->highest_vol_id) {
1541 ubi_err(ubi, "highest_vol_id is %d, but vol_id %d is there",
1542 ai->highest_vol_id, av->vol_id);
1546 if (av->vol_type != UBI_DYNAMIC_VOLUME &&
1547 av->vol_type != UBI_STATIC_VOLUME) {
1548 ubi_err(ubi, "bad vol_type");
1552 if (av->data_pad > ubi->leb_size / 2) {
1553 ubi_err(ubi, "bad data_pad");
1558 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1564 if (aeb->pnum < 0 || aeb->ec < 0) {
1565 ubi_err(ubi, "negative values");
1569 if (aeb->ec < ai->min_ec) {
1570 ubi_err(ubi, "bad ai->min_ec (%d), %d found",
1571 ai->min_ec, aeb->ec);
1575 if (aeb->ec > ai->max_ec) {
1576 ubi_err(ubi, "bad ai->max_ec (%d), %d found",
1577 ai->max_ec, aeb->ec);
1581 if (aeb->pnum >= ubi->peb_count) {
1582 ubi_err(ubi, "too high PEB number %d, total PEBs %d",
1583 aeb->pnum, ubi->peb_count);
1587 if (av->vol_type == UBI_STATIC_VOLUME) {
1588 if (aeb->lnum >= av->used_ebs) {
1589 ubi_err(ubi, "bad lnum or used_ebs");
1593 if (av->used_ebs != 0) {
1594 ubi_err(ubi, "non-zero used_ebs");
1599 if (aeb->lnum > av->highest_lnum) {
1600 ubi_err(ubi, "incorrect highest_lnum or lnum");
1605 if (av->leb_count != leb_count) {
1606 ubi_err(ubi, "bad leb_count, %d objects in the tree",
1616 if (aeb->lnum != av->highest_lnum) {
1617 ubi_err(ubi, "bad highest_lnum");
1622 if (vols_found != ai->vols_found) {
1623 ubi_err(ubi, "bad ai->vols_found %d, should be %d",
1624 ai->vols_found, vols_found);
1628 /* Check that attaching information is correct */
1629 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1631 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1638 err = ubi_io_read_vid_hdr(ubi, aeb->pnum, vidh, 1);
1639 if (err && err != UBI_IO_BITFLIPS) {
1640 ubi_err(ubi, "VID header is not OK (%d)",
1647 vol_type = vidh->vol_type == UBI_VID_DYNAMIC ?
1648 UBI_DYNAMIC_VOLUME : UBI_STATIC_VOLUME;
1649 if (av->vol_type != vol_type) {
1650 ubi_err(ubi, "bad vol_type");
1654 if (aeb->sqnum != be64_to_cpu(vidh->sqnum)) {
1655 ubi_err(ubi, "bad sqnum %llu", aeb->sqnum);
1659 if (av->vol_id != be32_to_cpu(vidh->vol_id)) {
1660 ubi_err(ubi, "bad vol_id %d", av->vol_id);
1664 if (av->compat != vidh->compat) {
1665 ubi_err(ubi, "bad compat %d", vidh->compat);
1669 if (aeb->lnum != be32_to_cpu(vidh->lnum)) {
1670 ubi_err(ubi, "bad lnum %d", aeb->lnum);
1674 if (av->used_ebs != be32_to_cpu(vidh->used_ebs)) {
1675 ubi_err(ubi, "bad used_ebs %d", av->used_ebs);
1679 if (av->data_pad != be32_to_cpu(vidh->data_pad)) {
1680 ubi_err(ubi, "bad data_pad %d", av->data_pad);
1688 if (av->highest_lnum != be32_to_cpu(vidh->lnum)) {
1689 ubi_err(ubi, "bad highest_lnum %d", av->highest_lnum);
1693 if (av->last_data_size != be32_to_cpu(vidh->data_size)) {
1694 ubi_err(ubi, "bad last_data_size %d",
1695 av->last_data_size);
1701 * Make sure that all the physical eraseblocks are in one of the lists
1704 buf = kzalloc(ubi->peb_count, GFP_KERNEL);
1708 for (pnum = 0; pnum < ubi->peb_count; pnum++) {
1709 err = ubi_io_is_bad(ubi, pnum);
1717 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb)
1718 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb)
1721 list_for_each_entry(aeb, &ai->free, u.list)
1724 list_for_each_entry(aeb, &ai->corr, u.list)
1727 list_for_each_entry(aeb, &ai->erase, u.list)
1730 list_for_each_entry(aeb, &ai->alien, u.list)
1734 for (pnum = 0; pnum < ubi->peb_count; pnum++)
1736 ubi_err(ubi, "PEB %d is not referred", pnum);
1746 ubi_err(ubi, "bad attaching information about LEB %d", aeb->lnum);
1747 ubi_dump_aeb(aeb, 0);
1752 ubi_err(ubi, "bad attaching information about volume %d", av->vol_id);
1757 ubi_err(ubi, "bad attaching information about volume %d", av->vol_id);
1759 ubi_dump_vid_hdr(vidh);