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 * @vid_hdr: the volume identifier header to check
176 * @av: information about the volume this logical eraseblock belongs to
177 * @pnum: physical eraseblock number the VID header came from
179 * This function checks that data stored in @vid_hdr is consistent. Returns
180 * non-zero if an inconsistency was found and zero if not.
182 * Note, UBI does sanity check of everything it reads from the flash media.
183 * Most of the checks are done in the I/O sub-system. Here we check that the
184 * information in the VID header is consistent to the information in other VID
185 * headers of the same volume.
187 static int validate_vid_hdr(const struct ubi_vid_hdr *vid_hdr,
188 const struct ubi_ainf_volume *av, int pnum)
190 int vol_type = vid_hdr->vol_type;
191 int vol_id = be32_to_cpu(vid_hdr->vol_id);
192 int used_ebs = be32_to_cpu(vid_hdr->used_ebs);
193 int data_pad = be32_to_cpu(vid_hdr->data_pad);
195 if (av->leb_count != 0) {
199 * This is not the first logical eraseblock belonging to this
200 * volume. Ensure that the data in its VID header is consistent
201 * to the data in previous logical eraseblock headers.
204 if (vol_id != av->vol_id) {
205 ubi_err("inconsistent vol_id");
209 if (av->vol_type == UBI_STATIC_VOLUME)
210 av_vol_type = UBI_VID_STATIC;
212 av_vol_type = UBI_VID_DYNAMIC;
214 if (vol_type != av_vol_type) {
215 ubi_err("inconsistent vol_type");
219 if (used_ebs != av->used_ebs) {
220 ubi_err("inconsistent used_ebs");
224 if (data_pad != av->data_pad) {
225 ubi_err("inconsistent data_pad");
233 ubi_err("inconsistent VID header at PEB %d", pnum);
234 ubi_dump_vid_hdr(vid_hdr);
240 * add_volume - add volume to the attaching information.
241 * @ai: attaching information
242 * @vol_id: ID of the volume to add
243 * @pnum: physical eraseblock number
244 * @vid_hdr: volume identifier header
246 * If the volume corresponding to the @vid_hdr logical eraseblock is already
247 * present in the attaching information, this function does nothing. Otherwise
248 * it adds corresponding volume to the attaching information. Returns a pointer
249 * to the allocated "av" object in case of success and a negative error code in
252 static struct ubi_ainf_volume *add_volume(struct ubi_attach_info *ai,
253 int vol_id, int pnum,
254 const struct ubi_vid_hdr *vid_hdr)
256 struct ubi_ainf_volume *av;
257 struct rb_node **p = &ai->volumes.rb_node, *parent = NULL;
259 ubi_assert(vol_id == be32_to_cpu(vid_hdr->vol_id));
261 /* Walk the volume RB-tree to look if this volume is already present */
264 av = rb_entry(parent, struct ubi_ainf_volume, rb);
266 if (vol_id == av->vol_id)
269 if (vol_id > av->vol_id)
275 /* The volume is absent - add it */
276 av = kmalloc(sizeof(struct ubi_ainf_volume), GFP_KERNEL);
278 return ERR_PTR(-ENOMEM);
280 av->highest_lnum = av->leb_count = 0;
283 av->used_ebs = be32_to_cpu(vid_hdr->used_ebs);
284 av->data_pad = be32_to_cpu(vid_hdr->data_pad);
285 av->compat = vid_hdr->compat;
286 av->vol_type = vid_hdr->vol_type == UBI_VID_DYNAMIC ? UBI_DYNAMIC_VOLUME
288 if (vol_id > ai->highest_vol_id)
289 ai->highest_vol_id = vol_id;
291 rb_link_node(&av->rb, parent, p);
292 rb_insert_color(&av->rb, &ai->volumes);
294 dbg_bld("added volume %d", vol_id);
299 * ubi_compare_lebs - find out which logical eraseblock is newer.
300 * @ubi: UBI device description object
301 * @aeb: first logical eraseblock to compare
302 * @pnum: physical eraseblock number of the second logical eraseblock to
304 * @vid_hdr: volume identifier header of the second logical eraseblock
306 * This function compares 2 copies of a LEB and informs which one is newer. In
307 * case of success this function returns a positive value, in case of failure, a
308 * negative error code is returned. The success return codes use the following
310 * o bit 0 is cleared: the first PEB (described by @aeb) is newer than the
311 * second PEB (described by @pnum and @vid_hdr);
312 * o bit 0 is set: the second PEB is newer;
313 * o bit 1 is cleared: no bit-flips were detected in the newer LEB;
314 * o bit 1 is set: bit-flips were detected in the newer LEB;
315 * o bit 2 is cleared: the older LEB is not corrupted;
316 * o bit 2 is set: the older LEB is corrupted.
318 int ubi_compare_lebs(struct ubi_device *ubi, const struct ubi_ainf_peb *aeb,
319 int pnum, const struct ubi_vid_hdr *vid_hdr)
321 int len, err, second_is_newer, bitflips = 0, corrupted = 0;
322 uint32_t data_crc, crc;
323 struct ubi_vid_hdr *vh = NULL;
324 unsigned long long sqnum2 = be64_to_cpu(vid_hdr->sqnum);
326 if (sqnum2 == aeb->sqnum) {
328 * This must be a really ancient UBI image which has been
329 * created before sequence numbers support has been added. At
330 * that times we used 32-bit LEB versions stored in logical
331 * eraseblocks. That was before UBI got into mainline. We do not
332 * support these images anymore. Well, those images still work,
333 * but only if no unclean reboots happened.
335 ubi_err("unsupported on-flash UBI format");
339 /* Obviously the LEB with lower sequence counter is older */
340 second_is_newer = (sqnum2 > aeb->sqnum);
343 * Now we know which copy is newer. If the copy flag of the PEB with
344 * newer version is not set, then we just return, otherwise we have to
345 * check data CRC. For the second PEB we already have the VID header,
346 * for the first one - we'll need to re-read it from flash.
348 * Note: this may be optimized so that we wouldn't read twice.
351 if (second_is_newer) {
352 if (!vid_hdr->copy_flag) {
353 /* It is not a copy, so it is newer */
354 dbg_bld("second PEB %d is newer, copy_flag is unset",
359 if (!aeb->copy_flag) {
360 /* It is not a copy, so it is newer */
361 dbg_bld("first PEB %d is newer, copy_flag is unset",
363 return bitflips << 1;
366 vh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
371 err = ubi_io_read_vid_hdr(ubi, pnum, vh, 0);
373 if (err == UBI_IO_BITFLIPS)
376 ubi_err("VID of PEB %d header is bad, but it was OK earlier, err %d",
388 /* Read the data of the copy and check the CRC */
390 len = be32_to_cpu(vid_hdr->data_size);
392 mutex_lock(&ubi->buf_mutex);
393 err = ubi_io_read_data(ubi, ubi->peb_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, ubi->peb_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);
409 mutex_unlock(&ubi->buf_mutex);
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);
421 mutex_unlock(&ubi->buf_mutex);
423 ubi_free_vid_hdr(ubi, vh);
428 * ubi_add_to_av - add used physical eraseblock to the attaching information.
429 * @ubi: UBI device description object
430 * @ai: attaching 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_add_to_av(struct ubi_device *ubi, struct ubi_attach_info *ai, int pnum,
444 int ec, const struct ubi_vid_hdr *vid_hdr, int bitflips)
446 int err, vol_id, lnum;
447 unsigned long long sqnum;
448 struct ubi_ainf_volume *av;
449 struct ubi_ainf_peb *aeb;
450 struct rb_node **p, *parent = NULL;
452 vol_id = be32_to_cpu(vid_hdr->vol_id);
453 lnum = be32_to_cpu(vid_hdr->lnum);
454 sqnum = be64_to_cpu(vid_hdr->sqnum);
456 dbg_bld("PEB %d, LEB %d:%d, EC %d, sqnum %llu, bitflips %d",
457 pnum, vol_id, lnum, ec, sqnum, bitflips);
459 av = add_volume(ai, vol_id, pnum, vid_hdr);
463 if (ai->max_sqnum < sqnum)
464 ai->max_sqnum = sqnum;
467 * Walk the RB-tree of logical eraseblocks of volume @vol_id to look
468 * if this is the first instance of this logical eraseblock or not.
470 p = &av->root.rb_node;
475 aeb = rb_entry(parent, struct ubi_ainf_peb, u.rb);
476 if (lnum != aeb->lnum) {
477 if (lnum < aeb->lnum)
485 * There is already a physical eraseblock describing the same
486 * logical eraseblock present.
489 dbg_bld("this LEB already exists: PEB %d, sqnum %llu, EC %d",
490 aeb->pnum, aeb->sqnum, aeb->ec);
493 * Make sure that the logical eraseblocks have different
494 * sequence numbers. Otherwise the image is bad.
496 * However, if the sequence number is zero, we assume it must
497 * be an ancient UBI image from the era when UBI did not have
498 * sequence numbers. We still can attach these images, unless
499 * there is a need to distinguish between old and new
500 * eraseblocks, in which case we'll refuse the image in
501 * 'ubi_compare_lebs()'. In other words, we attach old clean
502 * images, but refuse attaching old images with duplicated
503 * logical eraseblocks because there was an unclean reboot.
505 if (aeb->sqnum == sqnum && sqnum != 0) {
506 ubi_err("two LEBs with same sequence number %llu",
508 ubi_dump_aeb(aeb, 0);
509 ubi_dump_vid_hdr(vid_hdr);
514 * Now we have to drop the older one and preserve the newer
517 cmp_res = ubi_compare_lebs(ubi, aeb, pnum, vid_hdr);
523 * This logical eraseblock is newer than the one
526 err = validate_vid_hdr(vid_hdr, av, pnum);
530 err = add_to_list(ai, aeb->pnum, aeb->vol_id,
531 aeb->lnum, aeb->ec, cmp_res & 4,
538 aeb->vol_id = vol_id;
540 aeb->scrub = ((cmp_res & 2) || bitflips);
541 aeb->copy_flag = vid_hdr->copy_flag;
544 if (av->highest_lnum == lnum)
546 be32_to_cpu(vid_hdr->data_size);
551 * This logical eraseblock is older than the one found
554 return add_to_list(ai, pnum, vol_id, lnum, ec,
555 cmp_res & 4, &ai->erase);
560 * We've met this logical eraseblock for the first time, add it to the
561 * attaching information.
564 err = validate_vid_hdr(vid_hdr, av, pnum);
568 aeb = kmem_cache_alloc(ai->aeb_slab_cache, GFP_KERNEL);
574 aeb->vol_id = vol_id;
576 aeb->scrub = bitflips;
577 aeb->copy_flag = vid_hdr->copy_flag;
580 if (av->highest_lnum <= lnum) {
581 av->highest_lnum = lnum;
582 av->last_data_size = be32_to_cpu(vid_hdr->data_size);
586 rb_link_node(&aeb->u.rb, parent, p);
587 rb_insert_color(&aeb->u.rb, &av->root);
592 * ubi_find_av - find volume in the attaching information.
593 * @ai: attaching information
594 * @vol_id: the requested volume ID
596 * This function returns a pointer to the volume description or %NULL if there
597 * are no data about this volume in the attaching information.
599 struct ubi_ainf_volume *ubi_find_av(const struct ubi_attach_info *ai,
602 struct ubi_ainf_volume *av;
603 struct rb_node *p = ai->volumes.rb_node;
606 av = rb_entry(p, struct ubi_ainf_volume, rb);
608 if (vol_id == av->vol_id)
611 if (vol_id > av->vol_id)
621 * ubi_remove_av - delete attaching information about a volume.
622 * @ai: attaching information
623 * @av: the volume attaching information to delete
625 void ubi_remove_av(struct ubi_attach_info *ai, struct ubi_ainf_volume *av)
628 struct ubi_ainf_peb *aeb;
630 dbg_bld("remove attaching information about volume %d", av->vol_id);
632 while ((rb = rb_first(&av->root))) {
633 aeb = rb_entry(rb, struct ubi_ainf_peb, u.rb);
634 rb_erase(&aeb->u.rb, &av->root);
635 list_add_tail(&aeb->u.list, &ai->erase);
638 rb_erase(&av->rb, &ai->volumes);
644 * early_erase_peb - erase a physical eraseblock.
645 * @ubi: UBI device description object
646 * @ai: attaching information
647 * @pnum: physical eraseblock number to erase;
648 * @ec: erase counter value to write (%UBI_UNKNOWN if it is unknown)
650 * This function erases physical eraseblock 'pnum', and writes the erase
651 * counter header to it. This function should only be used on UBI device
652 * initialization stages, when the EBA sub-system had not been yet initialized.
653 * This function returns zero in case of success and a negative error code in
656 static int early_erase_peb(struct ubi_device *ubi,
657 const struct ubi_attach_info *ai, int pnum, int ec)
660 struct ubi_ec_hdr *ec_hdr;
662 if ((long long)ec >= UBI_MAX_ERASECOUNTER) {
664 * Erase counter overflow. Upgrade UBI and use 64-bit
665 * erase counters internally.
667 ubi_err("erase counter overflow at PEB %d, EC %d", pnum, ec);
671 ec_hdr = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
675 ec_hdr->ec = cpu_to_be64(ec);
677 err = ubi_io_sync_erase(ubi, pnum, 0);
681 err = ubi_io_write_ec_hdr(ubi, pnum, ec_hdr);
689 * ubi_early_get_peb - get a free physical eraseblock.
690 * @ubi: UBI device description object
691 * @ai: attaching information
693 * This function returns a free physical eraseblock. It is supposed to be
694 * called on the UBI initialization stages when the wear-leveling sub-system is
695 * not initialized yet. This function picks a physical eraseblocks from one of
696 * the lists, writes the EC header if it is needed, and removes it from the
699 * This function returns a pointer to the "aeb" of the found free PEB in case
700 * of success and an error code in case of failure.
702 struct ubi_ainf_peb *ubi_early_get_peb(struct ubi_device *ubi,
703 struct ubi_attach_info *ai)
706 struct ubi_ainf_peb *aeb, *tmp_aeb;
708 if (!list_empty(&ai->free)) {
709 aeb = list_entry(ai->free.next, struct ubi_ainf_peb, u.list);
710 list_del(&aeb->u.list);
711 dbg_bld("return free PEB %d, EC %d", aeb->pnum, aeb->ec);
716 * We try to erase the first physical eraseblock from the erase list
717 * and pick it if we succeed, or try to erase the next one if not. And
718 * so forth. We don't want to take care about bad eraseblocks here -
719 * they'll be handled later.
721 list_for_each_entry_safe(aeb, tmp_aeb, &ai->erase, u.list) {
722 if (aeb->ec == UBI_UNKNOWN)
723 aeb->ec = ai->mean_ec;
725 err = early_erase_peb(ubi, ai, aeb->pnum, aeb->ec+1);
730 list_del(&aeb->u.list);
731 dbg_bld("return PEB %d, EC %d", aeb->pnum, aeb->ec);
735 ubi_err("no free eraseblocks");
736 return ERR_PTR(-ENOSPC);
740 * check_corruption - check the data area of PEB.
741 * @ubi: UBI device description object
742 * @vid_hdr: the (corrupted) VID header of this PEB
743 * @pnum: the physical eraseblock number to check
745 * This is a helper function which is used to distinguish between VID header
746 * corruptions caused by power cuts and other reasons. If the PEB contains only
747 * 0xFF bytes in the data area, the VID header is most probably corrupted
748 * because of a power cut (%0 is returned in this case). Otherwise, it was
749 * probably corrupted for some other reasons (%1 is returned in this case). A
750 * negative error code is returned if a read error occurred.
752 * If the corruption reason was a power cut, UBI can safely erase this PEB.
753 * Otherwise, it should preserve it to avoid possibly destroying important
756 static int check_corruption(struct ubi_device *ubi, struct ubi_vid_hdr *vid_hdr,
761 mutex_lock(&ubi->buf_mutex);
762 memset(ubi->peb_buf, 0x00, ubi->leb_size);
764 err = ubi_io_read(ubi, ubi->peb_buf, pnum, ubi->leb_start,
766 if (err == UBI_IO_BITFLIPS || mtd_is_eccerr(err)) {
768 * Bit-flips or integrity errors while reading the data area.
769 * It is difficult to say for sure what type of corruption is
770 * this, but presumably a power cut happened while this PEB was
771 * erased, so it became unstable and corrupted, and should be
781 if (ubi_check_pattern(ubi->peb_buf, 0xFF, ubi->leb_size))
784 ubi_err("PEB %d contains corrupted VID header, and the data does not contain all 0xFF",
786 ubi_err("this may be a non-UBI PEB or a severe VID header corruption which requires manual inspection");
787 ubi_dump_vid_hdr(vid_hdr);
788 pr_err("hexdump of PEB %d offset %d, length %d",
789 pnum, ubi->leb_start, ubi->leb_size);
790 ubi_dbg_print_hex_dump(KERN_DEBUG, "", DUMP_PREFIX_OFFSET, 32, 1,
791 ubi->peb_buf, ubi->leb_size, 1);
795 mutex_unlock(&ubi->buf_mutex);
800 * scan_peb - scan and process UBI headers of a PEB.
801 * @ubi: UBI device description object
802 * @ai: attaching information
803 * @pnum: the physical eraseblock number
804 * @vid: The volume ID of the found volume will be stored in this pointer
805 * @sqnum: The sqnum of the found volume will be stored in this pointer
807 * This function reads UBI headers of PEB @pnum, checks them, and adds
808 * information about this PEB to the corresponding list or RB-tree in the
809 * "attaching info" structure. Returns zero if the physical eraseblock was
810 * successfully handled and a negative error code in case of failure.
812 static int scan_peb(struct ubi_device *ubi, struct ubi_attach_info *ai,
813 int pnum, int *vid, unsigned long long *sqnum)
815 long long uninitialized_var(ec);
816 int err, bitflips = 0, vol_id = -1, ec_err = 0;
818 dbg_bld("scan PEB %d", pnum);
820 /* Skip bad physical eraseblocks */
821 err = ubi_io_is_bad(ubi, pnum);
825 ai->bad_peb_count += 1;
829 err = ubi_io_read_ec_hdr(ubi, pnum, ech, 0);
835 case UBI_IO_BITFLIPS:
839 ai->empty_peb_count += 1;
840 return add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
841 UBI_UNKNOWN, 0, &ai->erase);
842 case UBI_IO_FF_BITFLIPS:
843 ai->empty_peb_count += 1;
844 return add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
845 UBI_UNKNOWN, 1, &ai->erase);
846 case UBI_IO_BAD_HDR_EBADMSG:
849 * We have to also look at the VID header, possibly it is not
850 * corrupted. Set %bitflips flag in order to make this PEB be
851 * moved and EC be re-created.
858 ubi_err("'ubi_io_read_ec_hdr()' returned unknown code %d", err);
865 /* Make sure UBI version is OK */
866 if (ech->version != UBI_VERSION) {
867 ubi_err("this UBI version is %d, image version is %d",
868 UBI_VERSION, (int)ech->version);
872 ec = be64_to_cpu(ech->ec);
873 if (ec > UBI_MAX_ERASECOUNTER) {
875 * Erase counter overflow. The EC headers have 64 bits
876 * reserved, but we anyway make use of only 31 bit
877 * values, as this seems to be enough for any existing
878 * flash. Upgrade UBI and use 64-bit erase counters
881 ubi_err("erase counter overflow, max is %d",
882 UBI_MAX_ERASECOUNTER);
883 ubi_dump_ec_hdr(ech);
888 * Make sure that all PEBs have the same image sequence number.
889 * This allows us to detect situations when users flash UBI
890 * images incorrectly, so that the flash has the new UBI image
891 * and leftovers from the old one. This feature was added
892 * relatively recently, and the sequence number was always
893 * zero, because old UBI implementations always set it to zero.
894 * For this reasons, we do not panic if some PEBs have zero
895 * sequence number, while other PEBs have non-zero sequence
898 image_seq = be32_to_cpu(ech->image_seq);
900 ubi->image_seq = image_seq;
901 if (image_seq && ubi->image_seq != image_seq) {
902 ubi_err("bad image sequence number %d in PEB %d, expected %d",
903 image_seq, pnum, ubi->image_seq);
904 ubi_dump_ec_hdr(ech);
909 /* OK, we've done with the EC header, let's look at the VID header */
911 err = ubi_io_read_vid_hdr(ubi, pnum, vidh, 0);
917 case UBI_IO_BITFLIPS:
920 case UBI_IO_BAD_HDR_EBADMSG:
921 if (ec_err == UBI_IO_BAD_HDR_EBADMSG)
923 * Both EC and VID headers are corrupted and were read
924 * with data integrity error, probably this is a bad
925 * PEB, bit it is not marked as bad yet. This may also
926 * be a result of power cut during erasure.
928 ai->maybe_bad_peb_count += 1;
932 * Both headers are corrupted. There is a possibility
933 * that this a valid UBI PEB which has corresponding
934 * LEB, but the headers are corrupted. However, it is
935 * impossible to distinguish it from a PEB which just
936 * contains garbage because of a power cut during erase
937 * operation. So we just schedule this PEB for erasure.
939 * Besides, in case of NOR flash, we deliberately
940 * corrupt both headers because NOR flash erasure is
941 * slow and can start from the end.
946 * The EC was OK, but the VID header is corrupted. We
947 * have to check what is in the data area.
949 err = check_corruption(ubi, vidh, pnum);
954 /* This corruption is caused by a power cut */
955 err = add_to_list(ai, pnum, UBI_UNKNOWN,
956 UBI_UNKNOWN, ec, 1, &ai->erase);
958 /* This is an unexpected corruption */
959 err = add_corrupted(ai, pnum, ec);
963 case UBI_IO_FF_BITFLIPS:
964 err = add_to_list(ai, pnum, UBI_UNKNOWN, UBI_UNKNOWN,
970 if (ec_err || bitflips)
971 err = add_to_list(ai, pnum, UBI_UNKNOWN,
972 UBI_UNKNOWN, ec, 1, &ai->erase);
974 err = add_to_list(ai, pnum, UBI_UNKNOWN,
975 UBI_UNKNOWN, ec, 0, &ai->free);
980 ubi_err("'ubi_io_read_vid_hdr()' returned unknown code %d",
985 vol_id = be32_to_cpu(vidh->vol_id);
989 *sqnum = be64_to_cpu(vidh->sqnum);
990 if (vol_id > UBI_MAX_VOLUMES && vol_id != UBI_LAYOUT_VOLUME_ID) {
991 int lnum = be32_to_cpu(vidh->lnum);
993 /* Unsupported internal volume */
994 switch (vidh->compat) {
995 case UBI_COMPAT_DELETE:
996 if (vol_id != UBI_FM_SB_VOLUME_ID
997 && vol_id != UBI_FM_DATA_VOLUME_ID) {
998 ubi_msg("\"delete\" compatible internal volume %d:%d found, will remove it",
1001 err = add_to_list(ai, pnum, vol_id, lnum,
1008 ubi_msg("read-only compatible internal volume %d:%d found, switch to read-only mode",
1013 case UBI_COMPAT_PRESERVE:
1014 ubi_msg("\"preserve\" compatible internal volume %d:%d found",
1016 err = add_to_list(ai, pnum, vol_id, lnum,
1022 case UBI_COMPAT_REJECT:
1023 ubi_err("incompatible internal volume %d:%d found",
1030 ubi_warn("valid VID header but corrupted EC header at PEB %d",
1032 err = ubi_add_to_av(ubi, ai, pnum, ec, vidh, bitflips);
1040 if (ec > ai->max_ec)
1042 if (ec < ai->min_ec)
1050 * late_analysis - analyze the overall situation with PEB.
1051 * @ubi: UBI device description object
1052 * @ai: attaching information
1054 * This is a helper function which takes a look what PEBs we have after we
1055 * gather information about all of them ("ai" is compete). It decides whether
1056 * the flash is empty and should be formatted of whether there are too many
1057 * corrupted PEBs and we should not attach this MTD device. Returns zero if we
1058 * should proceed with attaching the MTD device, and %-EINVAL if we should not.
1060 static int late_analysis(struct ubi_device *ubi, struct ubi_attach_info *ai)
1062 struct ubi_ainf_peb *aeb;
1063 int max_corr, peb_count;
1065 peb_count = ubi->peb_count - ai->bad_peb_count - ai->alien_peb_count;
1066 max_corr = peb_count / 20 ?: 8;
1069 * Few corrupted PEBs is not a problem and may be just a result of
1070 * unclean reboots. However, many of them may indicate some problems
1071 * with the flash HW or driver.
1073 if (ai->corr_peb_count) {
1074 ubi_err("%d PEBs are corrupted and preserved",
1075 ai->corr_peb_count);
1076 pr_err("Corrupted PEBs are:");
1077 list_for_each_entry(aeb, &ai->corr, u.list)
1078 pr_cont(" %d", aeb->pnum);
1082 * If too many PEBs are corrupted, we refuse attaching,
1083 * otherwise, only print a warning.
1085 if (ai->corr_peb_count >= max_corr) {
1086 ubi_err("too many corrupted PEBs, refusing");
1091 if (ai->empty_peb_count + ai->maybe_bad_peb_count == peb_count) {
1093 * All PEBs are empty, or almost all - a couple PEBs look like
1094 * they may be bad PEBs which were not marked as bad yet.
1096 * This piece of code basically tries to distinguish between
1097 * the following situations:
1099 * 1. Flash is empty, but there are few bad PEBs, which are not
1100 * marked as bad so far, and which were read with error. We
1101 * want to go ahead and format this flash. While formatting,
1102 * the faulty PEBs will probably be marked as bad.
1104 * 2. Flash contains non-UBI data and we do not want to format
1105 * it and destroy possibly important information.
1107 if (ai->maybe_bad_peb_count <= 2) {
1109 ubi_msg("empty MTD device detected");
1110 get_random_bytes(&ubi->image_seq,
1111 sizeof(ubi->image_seq));
1113 ubi_err("MTD device is not UBI-formatted and possibly contains non-UBI data - refusing it");
1123 * destroy_av - free volume attaching information.
1124 * @av: volume attaching information
1125 * @ai: attaching information
1127 * This function destroys the volume attaching information.
1129 static void destroy_av(struct ubi_attach_info *ai, struct ubi_ainf_volume *av)
1131 struct ubi_ainf_peb *aeb;
1132 struct rb_node *this = av->root.rb_node;
1136 this = this->rb_left;
1137 else if (this->rb_right)
1138 this = this->rb_right;
1140 aeb = rb_entry(this, struct ubi_ainf_peb, u.rb);
1141 this = rb_parent(this);
1143 if (this->rb_left == &aeb->u.rb)
1144 this->rb_left = NULL;
1146 this->rb_right = NULL;
1149 kmem_cache_free(ai->aeb_slab_cache, aeb);
1156 * destroy_ai - destroy attaching information.
1157 * @ai: attaching information
1159 static void destroy_ai(struct ubi_attach_info *ai)
1161 struct ubi_ainf_peb *aeb, *aeb_tmp;
1162 struct ubi_ainf_volume *av;
1165 list_for_each_entry_safe(aeb, aeb_tmp, &ai->alien, u.list) {
1166 list_del(&aeb->u.list);
1167 kmem_cache_free(ai->aeb_slab_cache, aeb);
1169 list_for_each_entry_safe(aeb, aeb_tmp, &ai->erase, 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->corr, 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->free, u.list) {
1178 list_del(&aeb->u.list);
1179 kmem_cache_free(ai->aeb_slab_cache, aeb);
1182 /* Destroy the volume RB-tree */
1183 rb = ai->volumes.rb_node;
1187 else if (rb->rb_right)
1190 av = rb_entry(rb, struct ubi_ainf_volume, rb);
1194 if (rb->rb_left == &av->rb)
1197 rb->rb_right = NULL;
1204 if (ai->aeb_slab_cache)
1205 kmem_cache_destroy(ai->aeb_slab_cache);
1211 * scan_all - scan entire MTD device.
1212 * @ubi: UBI device description object
1213 * @ai: attach info object
1214 * @start: start scanning at this PEB
1216 * This function does full scanning of an MTD device and returns complete
1217 * information about it in form of a "struct ubi_attach_info" object. In case
1218 * of failure, an error code is returned.
1220 static int scan_all(struct ubi_device *ubi, struct ubi_attach_info *ai,
1224 struct rb_node *rb1, *rb2;
1225 struct ubi_ainf_volume *av;
1226 struct ubi_ainf_peb *aeb;
1230 ech = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
1234 vidh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
1238 for (pnum = start; pnum < ubi->peb_count; pnum++) {
1241 dbg_gen("process PEB %d", pnum);
1242 err = scan_peb(ubi, ai, pnum, NULL, NULL);
1247 ubi_msg("scanning is finished");
1249 /* Calculate mean erase counter */
1251 ai->mean_ec = div_u64(ai->ec_sum, ai->ec_count);
1253 err = late_analysis(ubi, ai);
1258 * In case of unknown erase counter we use the mean erase counter
1261 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1262 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb)
1263 if (aeb->ec == UBI_UNKNOWN)
1264 aeb->ec = ai->mean_ec;
1267 list_for_each_entry(aeb, &ai->free, u.list) {
1268 if (aeb->ec == UBI_UNKNOWN)
1269 aeb->ec = ai->mean_ec;
1272 list_for_each_entry(aeb, &ai->corr, u.list)
1273 if (aeb->ec == UBI_UNKNOWN)
1274 aeb->ec = ai->mean_ec;
1276 list_for_each_entry(aeb, &ai->erase, u.list)
1277 if (aeb->ec == UBI_UNKNOWN)
1278 aeb->ec = ai->mean_ec;
1280 err = self_check_ai(ubi, ai);
1284 ubi_free_vid_hdr(ubi, vidh);
1290 ubi_free_vid_hdr(ubi, vidh);
1296 #ifdef CONFIG_MTD_UBI_FASTMAP
1299 * scan_fastmap - try to find a fastmap and attach from it.
1300 * @ubi: UBI device description object
1301 * @ai: attach info object
1303 * Returns 0 on success, negative return values indicate an internal
1305 * UBI_NO_FASTMAP denotes that no fastmap was found.
1306 * UBI_BAD_FASTMAP denotes that the found fastmap was invalid.
1308 static int scan_fast(struct ubi_device *ubi, struct ubi_attach_info *ai)
1310 int err, pnum, fm_anchor = -1;
1311 unsigned long long max_sqnum = 0;
1315 ech = kzalloc(ubi->ec_hdr_alsize, GFP_KERNEL);
1319 vidh = ubi_zalloc_vid_hdr(ubi, GFP_KERNEL);
1323 for (pnum = 0; pnum < UBI_FM_MAX_START; pnum++) {
1325 unsigned long long sqnum = -1;
1328 dbg_gen("process PEB %d", pnum);
1329 err = scan_peb(ubi, ai, pnum, &vol_id, &sqnum);
1333 if (vol_id == UBI_FM_SB_VOLUME_ID && sqnum > max_sqnum) {
1339 ubi_free_vid_hdr(ubi, vidh);
1343 return UBI_NO_FASTMAP;
1345 return ubi_scan_fastmap(ubi, ai, fm_anchor);
1348 ubi_free_vid_hdr(ubi, vidh);
1357 static struct ubi_attach_info *alloc_ai(const char *slab_name)
1359 struct ubi_attach_info *ai;
1361 ai = kzalloc(sizeof(struct ubi_attach_info), GFP_KERNEL);
1365 INIT_LIST_HEAD(&ai->corr);
1366 INIT_LIST_HEAD(&ai->free);
1367 INIT_LIST_HEAD(&ai->erase);
1368 INIT_LIST_HEAD(&ai->alien);
1369 ai->volumes = RB_ROOT;
1370 ai->aeb_slab_cache = kmem_cache_create(slab_name,
1371 sizeof(struct ubi_ainf_peb),
1373 if (!ai->aeb_slab_cache) {
1382 * ubi_attach - attach an MTD device.
1383 * @ubi: UBI device descriptor
1384 * @force_scan: if set to non-zero attach by scanning
1386 * This function returns zero in case of success and a negative error code in
1389 int ubi_attach(struct ubi_device *ubi, int force_scan)
1392 struct ubi_attach_info *ai;
1394 ai = alloc_ai("ubi_aeb_slab_cache");
1398 #ifdef CONFIG_MTD_UBI_FASTMAP
1399 /* On small flash devices we disable fastmap in any case. */
1400 if ((int)mtd_div_by_eb(ubi->mtd->size, ubi->mtd) <= UBI_FM_MAX_START) {
1401 ubi->fm_disabled = 1;
1406 err = scan_all(ubi, ai, 0);
1408 err = scan_fast(ubi, ai);
1410 if (err != UBI_NO_FASTMAP) {
1412 ai = alloc_ai("ubi_aeb_slab_cache2");
1416 err = scan_all(ubi, ai, 0);
1418 err = scan_all(ubi, ai, UBI_FM_MAX_START);
1423 err = scan_all(ubi, ai, 0);
1428 ubi->bad_peb_count = ai->bad_peb_count;
1429 ubi->good_peb_count = ubi->peb_count - ubi->bad_peb_count;
1430 ubi->corr_peb_count = ai->corr_peb_count;
1431 ubi->max_ec = ai->max_ec;
1432 ubi->mean_ec = ai->mean_ec;
1433 dbg_gen("max. sequence number: %llu", ai->max_sqnum);
1435 err = ubi_read_volume_table(ubi, ai);
1439 err = ubi_wl_init(ubi, ai);
1443 err = ubi_eba_init(ubi, ai);
1447 #ifdef CONFIG_MTD_UBI_FASTMAP
1448 if (ubi->fm && ubi_dbg_chk_gen(ubi)) {
1449 struct ubi_attach_info *scan_ai;
1451 scan_ai = alloc_ai("ubi_ckh_aeb_slab_cache");
1457 err = scan_all(ubi, scan_ai, 0);
1459 destroy_ai(scan_ai);
1463 err = self_check_eba(ubi, ai, scan_ai);
1464 destroy_ai(scan_ai);
1477 ubi_free_internal_volumes(ubi);
1485 * self_check_ai - check the attaching information.
1486 * @ubi: UBI device description object
1487 * @ai: attaching information
1489 * This function returns zero if the attaching information is all right, and a
1490 * negative error code if not or if an error occurred.
1492 static int self_check_ai(struct ubi_device *ubi, struct ubi_attach_info *ai)
1494 int pnum, err, vols_found = 0;
1495 struct rb_node *rb1, *rb2;
1496 struct ubi_ainf_volume *av;
1497 struct ubi_ainf_peb *aeb, *last_aeb;
1500 if (!ubi_dbg_chk_gen(ubi))
1504 * At first, check that attaching information is OK.
1506 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1514 ubi_err("bad is_empty flag");
1518 if (av->vol_id < 0 || av->highest_lnum < 0 ||
1519 av->leb_count < 0 || av->vol_type < 0 || av->used_ebs < 0 ||
1520 av->data_pad < 0 || av->last_data_size < 0) {
1521 ubi_err("negative values");
1525 if (av->vol_id >= UBI_MAX_VOLUMES &&
1526 av->vol_id < UBI_INTERNAL_VOL_START) {
1527 ubi_err("bad vol_id");
1531 if (av->vol_id > ai->highest_vol_id) {
1532 ubi_err("highest_vol_id is %d, but vol_id %d is there",
1533 ai->highest_vol_id, av->vol_id);
1537 if (av->vol_type != UBI_DYNAMIC_VOLUME &&
1538 av->vol_type != UBI_STATIC_VOLUME) {
1539 ubi_err("bad vol_type");
1543 if (av->data_pad > ubi->leb_size / 2) {
1544 ubi_err("bad data_pad");
1549 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1555 if (aeb->pnum < 0 || aeb->ec < 0) {
1556 ubi_err("negative values");
1560 if (aeb->ec < ai->min_ec) {
1561 ubi_err("bad ai->min_ec (%d), %d found",
1562 ai->min_ec, aeb->ec);
1566 if (aeb->ec > ai->max_ec) {
1567 ubi_err("bad ai->max_ec (%d), %d found",
1568 ai->max_ec, aeb->ec);
1572 if (aeb->pnum >= ubi->peb_count) {
1573 ubi_err("too high PEB number %d, total PEBs %d",
1574 aeb->pnum, ubi->peb_count);
1578 if (av->vol_type == UBI_STATIC_VOLUME) {
1579 if (aeb->lnum >= av->used_ebs) {
1580 ubi_err("bad lnum or used_ebs");
1584 if (av->used_ebs != 0) {
1585 ubi_err("non-zero used_ebs");
1590 if (aeb->lnum > av->highest_lnum) {
1591 ubi_err("incorrect highest_lnum or lnum");
1596 if (av->leb_count != leb_count) {
1597 ubi_err("bad leb_count, %d objects in the tree",
1607 if (aeb->lnum != av->highest_lnum) {
1608 ubi_err("bad highest_lnum");
1613 if (vols_found != ai->vols_found) {
1614 ubi_err("bad ai->vols_found %d, should be %d",
1615 ai->vols_found, vols_found);
1619 /* Check that attaching information is correct */
1620 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb) {
1622 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb) {
1629 err = ubi_io_read_vid_hdr(ubi, aeb->pnum, vidh, 1);
1630 if (err && err != UBI_IO_BITFLIPS) {
1631 ubi_err("VID header is not OK (%d)", err);
1637 vol_type = vidh->vol_type == UBI_VID_DYNAMIC ?
1638 UBI_DYNAMIC_VOLUME : UBI_STATIC_VOLUME;
1639 if (av->vol_type != vol_type) {
1640 ubi_err("bad vol_type");
1644 if (aeb->sqnum != be64_to_cpu(vidh->sqnum)) {
1645 ubi_err("bad sqnum %llu", aeb->sqnum);
1649 if (av->vol_id != be32_to_cpu(vidh->vol_id)) {
1650 ubi_err("bad vol_id %d", av->vol_id);
1654 if (av->compat != vidh->compat) {
1655 ubi_err("bad compat %d", vidh->compat);
1659 if (aeb->lnum != be32_to_cpu(vidh->lnum)) {
1660 ubi_err("bad lnum %d", aeb->lnum);
1664 if (av->used_ebs != be32_to_cpu(vidh->used_ebs)) {
1665 ubi_err("bad used_ebs %d", av->used_ebs);
1669 if (av->data_pad != be32_to_cpu(vidh->data_pad)) {
1670 ubi_err("bad data_pad %d", av->data_pad);
1678 if (av->highest_lnum != be32_to_cpu(vidh->lnum)) {
1679 ubi_err("bad highest_lnum %d", av->highest_lnum);
1683 if (av->last_data_size != be32_to_cpu(vidh->data_size)) {
1684 ubi_err("bad last_data_size %d", av->last_data_size);
1690 * Make sure that all the physical eraseblocks are in one of the lists
1693 buf = kzalloc(ubi->peb_count, GFP_KERNEL);
1697 for (pnum = 0; pnum < ubi->peb_count; pnum++) {
1698 err = ubi_io_is_bad(ubi, pnum);
1706 ubi_rb_for_each_entry(rb1, av, &ai->volumes, rb)
1707 ubi_rb_for_each_entry(rb2, aeb, &av->root, u.rb)
1710 list_for_each_entry(aeb, &ai->free, u.list)
1713 list_for_each_entry(aeb, &ai->corr, u.list)
1716 list_for_each_entry(aeb, &ai->erase, u.list)
1719 list_for_each_entry(aeb, &ai->alien, u.list)
1723 for (pnum = 0; pnum < ubi->peb_count; pnum++)
1725 ubi_err("PEB %d is not referred", pnum);
1735 ubi_err("bad attaching information about LEB %d", aeb->lnum);
1736 ubi_dump_aeb(aeb, 0);
1741 ubi_err("bad attaching information about volume %d", av->vol_id);
1746 ubi_err("bad attaching information about volume %d", av->vol_id);
1748 ubi_dump_vid_hdr(vidh);