2 * Copyright (c) International Business Machines Corp., 2006
4 * SPDX-License-Identifier: GPL-2.0+
6 * Author: Artem Bityutskiy (Битюцкий Артём)
10 * The UBI Eraseblock Association (EBA) sub-system.
12 * This sub-system is responsible for I/O to/from logical eraseblock.
14 * Although in this implementation the EBA table is fully kept and managed in
15 * RAM, which assumes poor scalability, it might be (partially) maintained on
16 * flash in future implementations.
18 * The EBA sub-system implements per-logical eraseblock locking. Before
19 * accessing a logical eraseblock it is locked for reading or writing. The
20 * per-logical eraseblock locking is implemented by means of the lock tree. The
21 * lock tree is an RB-tree which refers all the currently locked logical
22 * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
23 * They are indexed by (@vol_id, @lnum) pairs.
25 * EBA also maintains the global sequence counter which is incremented each
26 * time a logical eraseblock is mapped to a physical eraseblock and it is
27 * stored in the volume identifier header. This means that each VID header has
28 * a unique sequence number. The sequence number is only increased an we assume
29 * 64 bits is enough to never overflow.
33 #include <linux/slab.h>
34 #include <linux/crc32.h>
36 #include <ubi_uboot.h>
39 #include <linux/err.h>
42 /* Number of physical eraseblocks reserved for atomic LEB change operation */
43 #define EBA_RESERVED_PEBS 1
46 * next_sqnum - get next sequence number.
47 * @ubi: UBI device description object
49 * This function returns next sequence number to use, which is just the current
50 * global sequence counter value. It also increases the global sequence
53 unsigned long long ubi_next_sqnum(struct ubi_device *ubi)
55 unsigned long long sqnum;
57 spin_lock(&ubi->ltree_lock);
58 sqnum = ubi->global_sqnum++;
59 spin_unlock(&ubi->ltree_lock);
65 * ubi_get_compat - get compatibility flags of a volume.
66 * @ubi: UBI device description object
69 * This function returns compatibility flags for an internal volume. User
70 * volumes have no compatibility flags, so %0 is returned.
72 static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
74 if (vol_id == UBI_LAYOUT_VOLUME_ID)
75 return UBI_LAYOUT_VOLUME_COMPAT;
80 * ltree_lookup - look up the lock tree.
81 * @ubi: UBI device description object
83 * @lnum: logical eraseblock number
85 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
86 * object if the logical eraseblock is locked and %NULL if it is not.
87 * @ubi->ltree_lock has to be locked.
89 static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
94 p = ubi->ltree.rb_node;
96 struct ubi_ltree_entry *le;
98 le = rb_entry(p, struct ubi_ltree_entry, rb);
100 if (vol_id < le->vol_id)
102 else if (vol_id > le->vol_id)
107 else if (lnum > le->lnum)
118 * ltree_add_entry - add new entry to the lock tree.
119 * @ubi: UBI device description object
121 * @lnum: logical eraseblock number
123 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
124 * lock tree. If such entry is already there, its usage counter is increased.
125 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
128 static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
129 int vol_id, int lnum)
131 struct ubi_ltree_entry *le, *le1, *le_free;
133 le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
135 return ERR_PTR(-ENOMEM);
138 init_rwsem(&le->mutex);
142 spin_lock(&ubi->ltree_lock);
143 le1 = ltree_lookup(ubi, vol_id, lnum);
147 * This logical eraseblock is already locked. The newly
148 * allocated lock entry is not needed.
153 struct rb_node **p, *parent = NULL;
156 * No lock entry, add the newly allocated one to the
157 * @ubi->ltree RB-tree.
161 p = &ubi->ltree.rb_node;
164 le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
166 if (vol_id < le1->vol_id)
168 else if (vol_id > le1->vol_id)
171 ubi_assert(lnum != le1->lnum);
172 if (lnum < le1->lnum)
179 rb_link_node(&le->rb, parent, p);
180 rb_insert_color(&le->rb, &ubi->ltree);
183 spin_unlock(&ubi->ltree_lock);
190 * leb_read_lock - lock logical eraseblock for reading.
191 * @ubi: UBI device description object
193 * @lnum: logical eraseblock number
195 * This function locks a logical eraseblock for reading. Returns zero in case
196 * of success and a negative error code in case of failure.
198 static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
200 struct ubi_ltree_entry *le;
202 le = ltree_add_entry(ubi, vol_id, lnum);
205 down_read(&le->mutex);
210 * leb_read_unlock - unlock logical eraseblock.
211 * @ubi: UBI device description object
213 * @lnum: logical eraseblock number
215 static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
217 struct ubi_ltree_entry *le;
219 spin_lock(&ubi->ltree_lock);
220 le = ltree_lookup(ubi, vol_id, lnum);
222 ubi_assert(le->users >= 0);
224 if (le->users == 0) {
225 rb_erase(&le->rb, &ubi->ltree);
228 spin_unlock(&ubi->ltree_lock);
232 * leb_write_lock - lock logical eraseblock for writing.
233 * @ubi: UBI device description object
235 * @lnum: logical eraseblock number
237 * This function locks a logical eraseblock for writing. Returns zero in case
238 * of success and a negative error code in case of failure.
240 static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
242 struct ubi_ltree_entry *le;
244 le = ltree_add_entry(ubi, vol_id, lnum);
247 down_write(&le->mutex);
252 * leb_write_lock - lock logical eraseblock for writing.
253 * @ubi: UBI device description object
255 * @lnum: logical eraseblock number
257 * This function locks a logical eraseblock for writing if there is no
258 * contention and does nothing if there is contention. Returns %0 in case of
259 * success, %1 in case of contention, and and a negative error code in case of
262 static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
264 struct ubi_ltree_entry *le;
266 le = ltree_add_entry(ubi, vol_id, lnum);
269 if (down_write_trylock(&le->mutex))
272 /* Contention, cancel */
273 spin_lock(&ubi->ltree_lock);
275 ubi_assert(le->users >= 0);
276 if (le->users == 0) {
277 rb_erase(&le->rb, &ubi->ltree);
280 spin_unlock(&ubi->ltree_lock);
286 * leb_write_unlock - unlock logical eraseblock.
287 * @ubi: UBI device description object
289 * @lnum: logical eraseblock number
291 static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
293 struct ubi_ltree_entry *le;
295 spin_lock(&ubi->ltree_lock);
296 le = ltree_lookup(ubi, vol_id, lnum);
298 ubi_assert(le->users >= 0);
299 up_write(&le->mutex);
300 if (le->users == 0) {
301 rb_erase(&le->rb, &ubi->ltree);
304 spin_unlock(&ubi->ltree_lock);
308 * ubi_eba_unmap_leb - un-map logical eraseblock.
309 * @ubi: UBI device description object
310 * @vol: volume description object
311 * @lnum: logical eraseblock number
313 * This function un-maps logical eraseblock @lnum and schedules corresponding
314 * physical eraseblock for erasure. Returns zero in case of success and a
315 * negative error code in case of failure.
317 int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
320 int err, pnum, vol_id = vol->vol_id;
325 err = leb_write_lock(ubi, vol_id, lnum);
329 pnum = vol->eba_tbl[lnum];
331 /* This logical eraseblock is already unmapped */
334 dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
336 down_read(&ubi->fm_sem);
337 vol->eba_tbl[lnum] = UBI_LEB_UNMAPPED;
338 up_read(&ubi->fm_sem);
339 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
342 leb_write_unlock(ubi, vol_id, lnum);
347 * ubi_eba_read_leb - read data.
348 * @ubi: UBI device description object
349 * @vol: volume description object
350 * @lnum: logical eraseblock number
351 * @buf: buffer to store the read data
352 * @offset: offset from where to read
353 * @len: how many bytes to read
354 * @check: data CRC check flag
356 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
357 * bytes. The @check flag only makes sense for static volumes and forces
358 * eraseblock data CRC checking.
360 * In case of success this function returns zero. In case of a static volume,
361 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
362 * returned for any volume type if an ECC error was detected by the MTD device
363 * driver. Other negative error cored may be returned in case of other errors.
365 int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
366 void *buf, int offset, int len, int check)
368 int err, pnum, scrub = 0, vol_id = vol->vol_id;
369 struct ubi_vid_hdr *vid_hdr;
370 uint32_t uninitialized_var(crc);
372 err = leb_read_lock(ubi, vol_id, lnum);
376 pnum = vol->eba_tbl[lnum];
379 * The logical eraseblock is not mapped, fill the whole buffer
380 * with 0xFF bytes. The exception is static volumes for which
381 * it is an error to read unmapped logical eraseblocks.
383 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
384 len, offset, vol_id, lnum);
385 leb_read_unlock(ubi, vol_id, lnum);
386 ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
387 memset(buf, 0xFF, len);
391 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
392 len, offset, vol_id, lnum, pnum);
394 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
399 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
405 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
406 if (err && err != UBI_IO_BITFLIPS) {
409 * The header is either absent or corrupted.
410 * The former case means there is a bug -
411 * switch to read-only mode just in case.
412 * The latter case means a real corruption - we
413 * may try to recover data. FIXME: but this is
416 if (err == UBI_IO_BAD_HDR_EBADMSG ||
417 err == UBI_IO_BAD_HDR) {
418 ubi_warn("corrupted VID header at PEB %d, LEB %d:%d",
425 } else if (err == UBI_IO_BITFLIPS)
428 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
429 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
431 crc = be32_to_cpu(vid_hdr->data_crc);
432 ubi_free_vid_hdr(ubi, vid_hdr);
435 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
437 if (err == UBI_IO_BITFLIPS) {
440 } else if (mtd_is_eccerr(err)) {
441 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
445 ubi_msg("force data checking");
454 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
456 ubi_warn("CRC error: calculated %#08x, must be %#08x",
464 err = ubi_wl_scrub_peb(ubi, pnum);
466 leb_read_unlock(ubi, vol_id, lnum);
470 ubi_free_vid_hdr(ubi, vid_hdr);
472 leb_read_unlock(ubi, vol_id, lnum);
477 * recover_peb - recover from write failure.
478 * @ubi: UBI device description object
479 * @pnum: the physical eraseblock to recover
481 * @lnum: logical eraseblock number
482 * @buf: data which was not written because of the write failure
483 * @offset: offset of the failed write
484 * @len: how many bytes should have been written
486 * This function is called in case of a write failure and moves all good data
487 * from the potentially bad physical eraseblock to a good physical eraseblock.
488 * This function also writes the data which was not written due to the failure.
489 * Returns new physical eraseblock number in case of success, and a negative
490 * error code in case of failure.
492 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
493 const void *buf, int offset, int len)
495 int err, idx = vol_id2idx(ubi, vol_id), new_pnum, data_size, tries = 0;
496 struct ubi_volume *vol = ubi->volumes[idx];
497 struct ubi_vid_hdr *vid_hdr;
499 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
504 new_pnum = ubi_wl_get_peb(ubi);
506 ubi_free_vid_hdr(ubi, vid_hdr);
510 ubi_msg("recover PEB %d, move data to PEB %d", pnum, new_pnum);
512 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
513 if (err && err != UBI_IO_BITFLIPS) {
519 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
520 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
524 data_size = offset + len;
525 mutex_lock(&ubi->buf_mutex);
526 memset(ubi->peb_buf + offset, 0xFF, len);
528 /* Read everything before the area where the write failure happened */
530 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
531 if (err && err != UBI_IO_BITFLIPS)
535 memcpy(ubi->peb_buf + offset, buf, len);
537 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
539 mutex_unlock(&ubi->buf_mutex);
543 mutex_unlock(&ubi->buf_mutex);
544 ubi_free_vid_hdr(ubi, vid_hdr);
546 down_read(&ubi->fm_sem);
547 vol->eba_tbl[lnum] = new_pnum;
548 up_read(&ubi->fm_sem);
549 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
551 ubi_msg("data was successfully recovered");
555 mutex_unlock(&ubi->buf_mutex);
557 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
558 ubi_free_vid_hdr(ubi, vid_hdr);
563 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
566 ubi_warn("failed to write to PEB %d", new_pnum);
567 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
568 if (++tries > UBI_IO_RETRIES) {
569 ubi_free_vid_hdr(ubi, vid_hdr);
572 ubi_msg("try again");
577 * ubi_eba_write_leb - write data to dynamic volume.
578 * @ubi: UBI device description object
579 * @vol: volume description object
580 * @lnum: logical eraseblock number
581 * @buf: the data to write
582 * @offset: offset within the logical eraseblock where to write
583 * @len: how many bytes to write
585 * This function writes data to logical eraseblock @lnum of a dynamic volume
586 * @vol. Returns zero in case of success and a negative error code in case
587 * of failure. In case of error, it is possible that something was still
588 * written to the flash media, but may be some garbage.
590 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
591 const void *buf, int offset, int len)
593 int err, pnum, tries = 0, vol_id = vol->vol_id;
594 struct ubi_vid_hdr *vid_hdr;
599 err = leb_write_lock(ubi, vol_id, lnum);
603 pnum = vol->eba_tbl[lnum];
605 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
606 len, offset, vol_id, lnum, pnum);
608 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
610 ubi_warn("failed to write data to PEB %d", pnum);
611 if (err == -EIO && ubi->bad_allowed)
612 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
617 leb_write_unlock(ubi, vol_id, lnum);
622 * The logical eraseblock is not mapped. We have to get a free physical
623 * eraseblock and write the volume identifier header there first.
625 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
627 leb_write_unlock(ubi, vol_id, lnum);
631 vid_hdr->vol_type = UBI_VID_DYNAMIC;
632 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
633 vid_hdr->vol_id = cpu_to_be32(vol_id);
634 vid_hdr->lnum = cpu_to_be32(lnum);
635 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
636 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
639 pnum = ubi_wl_get_peb(ubi);
641 ubi_free_vid_hdr(ubi, vid_hdr);
642 leb_write_unlock(ubi, vol_id, lnum);
646 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
647 len, offset, vol_id, lnum, pnum);
649 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
651 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
657 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
659 ubi_warn("failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
660 len, offset, vol_id, lnum, pnum);
665 down_read(&ubi->fm_sem);
666 vol->eba_tbl[lnum] = pnum;
667 up_read(&ubi->fm_sem);
669 leb_write_unlock(ubi, vol_id, lnum);
670 ubi_free_vid_hdr(ubi, vid_hdr);
674 if (err != -EIO || !ubi->bad_allowed) {
676 leb_write_unlock(ubi, vol_id, lnum);
677 ubi_free_vid_hdr(ubi, vid_hdr);
682 * Fortunately, this is the first write operation to this physical
683 * eraseblock, so just put it and request a new one. We assume that if
684 * this physical eraseblock went bad, the erase code will handle that.
686 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
687 if (err || ++tries > UBI_IO_RETRIES) {
689 leb_write_unlock(ubi, vol_id, lnum);
690 ubi_free_vid_hdr(ubi, vid_hdr);
694 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
695 ubi_msg("try another PEB");
700 * ubi_eba_write_leb_st - write data to static volume.
701 * @ubi: UBI device description object
702 * @vol: volume description object
703 * @lnum: logical eraseblock number
704 * @buf: data to write
705 * @len: how many bytes to write
706 * @used_ebs: how many logical eraseblocks will this volume contain
708 * This function writes data to logical eraseblock @lnum of static volume
709 * @vol. The @used_ebs argument should contain total number of logical
710 * eraseblock in this static volume.
712 * When writing to the last logical eraseblock, the @len argument doesn't have
713 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
714 * to the real data size, although the @buf buffer has to contain the
715 * alignment. In all other cases, @len has to be aligned.
717 * It is prohibited to write more than once to logical eraseblocks of static
718 * volumes. This function returns zero in case of success and a negative error
719 * code in case of failure.
721 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
722 int lnum, const void *buf, int len, int used_ebs)
724 int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
725 struct ubi_vid_hdr *vid_hdr;
731 if (lnum == used_ebs - 1)
732 /* If this is the last LEB @len may be unaligned */
733 len = ALIGN(data_size, ubi->min_io_size);
735 ubi_assert(!(len & (ubi->min_io_size - 1)));
737 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
741 err = leb_write_lock(ubi, vol_id, lnum);
743 ubi_free_vid_hdr(ubi, vid_hdr);
747 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
748 vid_hdr->vol_id = cpu_to_be32(vol_id);
749 vid_hdr->lnum = cpu_to_be32(lnum);
750 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
751 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
753 crc = crc32(UBI_CRC32_INIT, buf, data_size);
754 vid_hdr->vol_type = UBI_VID_STATIC;
755 vid_hdr->data_size = cpu_to_be32(data_size);
756 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
757 vid_hdr->data_crc = cpu_to_be32(crc);
760 pnum = ubi_wl_get_peb(ubi);
762 ubi_free_vid_hdr(ubi, vid_hdr);
763 leb_write_unlock(ubi, vol_id, lnum);
767 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
768 len, vol_id, lnum, pnum, used_ebs);
770 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
772 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
777 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
779 ubi_warn("failed to write %d bytes of data to PEB %d",
784 ubi_assert(vol->eba_tbl[lnum] < 0);
785 down_read(&ubi->fm_sem);
786 vol->eba_tbl[lnum] = pnum;
787 up_read(&ubi->fm_sem);
789 leb_write_unlock(ubi, vol_id, lnum);
790 ubi_free_vid_hdr(ubi, vid_hdr);
794 if (err != -EIO || !ubi->bad_allowed) {
796 * This flash device does not admit of bad eraseblocks or
797 * something nasty and unexpected happened. Switch to read-only
801 leb_write_unlock(ubi, vol_id, lnum);
802 ubi_free_vid_hdr(ubi, vid_hdr);
806 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
807 if (err || ++tries > UBI_IO_RETRIES) {
809 leb_write_unlock(ubi, vol_id, lnum);
810 ubi_free_vid_hdr(ubi, vid_hdr);
814 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
815 ubi_msg("try another PEB");
820 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
821 * @ubi: UBI device description object
822 * @vol: volume description object
823 * @lnum: logical eraseblock number
824 * @buf: data to write
825 * @len: how many bytes to write
827 * This function changes the contents of a logical eraseblock atomically. @buf
828 * has to contain new logical eraseblock data, and @len - the length of the
829 * data, which has to be aligned. This function guarantees that in case of an
830 * unclean reboot the old contents is preserved. Returns zero in case of
831 * success and a negative error code in case of failure.
833 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
834 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
836 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
837 int lnum, const void *buf, int len)
839 int err, pnum, tries = 0, vol_id = vol->vol_id;
840 struct ubi_vid_hdr *vid_hdr;
848 * Special case when data length is zero. In this case the LEB
849 * has to be unmapped and mapped somewhere else.
851 err = ubi_eba_unmap_leb(ubi, vol, lnum);
854 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
857 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
861 mutex_lock(&ubi->alc_mutex);
862 err = leb_write_lock(ubi, vol_id, lnum);
866 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
867 vid_hdr->vol_id = cpu_to_be32(vol_id);
868 vid_hdr->lnum = cpu_to_be32(lnum);
869 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
870 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
872 crc = crc32(UBI_CRC32_INIT, buf, len);
873 vid_hdr->vol_type = UBI_VID_DYNAMIC;
874 vid_hdr->data_size = cpu_to_be32(len);
875 vid_hdr->copy_flag = 1;
876 vid_hdr->data_crc = cpu_to_be32(crc);
879 pnum = ubi_wl_get_peb(ubi);
885 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
886 vol_id, lnum, vol->eba_tbl[lnum], pnum);
888 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
890 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
895 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
897 ubi_warn("failed to write %d bytes of data to PEB %d",
902 if (vol->eba_tbl[lnum] >= 0) {
903 err = ubi_wl_put_peb(ubi, vol_id, lnum, vol->eba_tbl[lnum], 0);
908 down_read(&ubi->fm_sem);
909 vol->eba_tbl[lnum] = pnum;
910 up_read(&ubi->fm_sem);
913 leb_write_unlock(ubi, vol_id, lnum);
915 mutex_unlock(&ubi->alc_mutex);
916 ubi_free_vid_hdr(ubi, vid_hdr);
920 if (err != -EIO || !ubi->bad_allowed) {
922 * This flash device does not admit of bad eraseblocks or
923 * something nasty and unexpected happened. Switch to read-only
930 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
931 if (err || ++tries > UBI_IO_RETRIES) {
936 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
937 ubi_msg("try another PEB");
942 * is_error_sane - check whether a read error is sane.
943 * @err: code of the error happened during reading
945 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
946 * cannot read data from the target PEB (an error @err happened). If the error
947 * code is sane, then we treat this error as non-fatal. Otherwise the error is
948 * fatal and UBI will be switched to R/O mode later.
950 * The idea is that we try not to switch to R/O mode if the read error is
951 * something which suggests there was a real read problem. E.g., %-EIO. Or a
952 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
953 * mode, simply because we do not know what happened at the MTD level, and we
954 * cannot handle this. E.g., the underlying driver may have become crazy, and
955 * it is safer to switch to R/O mode to preserve the data.
957 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
958 * which we have just written.
960 static int is_error_sane(int err)
962 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
963 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
969 * ubi_eba_copy_leb - copy logical eraseblock.
970 * @ubi: UBI device description object
971 * @from: physical eraseblock number from where to copy
972 * @to: physical eraseblock number where to copy
973 * @vid_hdr: VID header of the @from physical eraseblock
975 * This function copies logical eraseblock from physical eraseblock @from to
976 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
978 * o %0 in case of success;
979 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
980 * o a negative error code in case of failure.
982 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
983 struct ubi_vid_hdr *vid_hdr)
985 int err, vol_id, lnum, data_size, aldata_size, idx;
986 struct ubi_volume *vol;
989 vol_id = be32_to_cpu(vid_hdr->vol_id);
990 lnum = be32_to_cpu(vid_hdr->lnum);
992 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
994 if (vid_hdr->vol_type == UBI_VID_STATIC) {
995 data_size = be32_to_cpu(vid_hdr->data_size);
996 aldata_size = ALIGN(data_size, ubi->min_io_size);
998 data_size = aldata_size =
999 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1001 idx = vol_id2idx(ubi, vol_id);
1002 spin_lock(&ubi->volumes_lock);
1004 * Note, we may race with volume deletion, which means that the volume
1005 * this logical eraseblock belongs to might be being deleted. Since the
1006 * volume deletion un-maps all the volume's logical eraseblocks, it will
1007 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1009 vol = ubi->volumes[idx];
1010 spin_unlock(&ubi->volumes_lock);
1012 /* No need to do further work, cancel */
1013 dbg_wl("volume %d is being removed, cancel", vol_id);
1014 return MOVE_CANCEL_RACE;
1018 * We do not want anybody to write to this logical eraseblock while we
1019 * are moving it, so lock it.
1021 * Note, we are using non-waiting locking here, because we cannot sleep
1022 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1023 * unmapping the LEB which is mapped to the PEB we are going to move
1024 * (@from). This task locks the LEB and goes sleep in the
1025 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1026 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1027 * LEB is already locked, we just do not move it and return
1028 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1029 * we do not know the reasons of the contention - it may be just a
1030 * normal I/O on this LEB, so we want to re-try.
1032 err = leb_write_trylock(ubi, vol_id, lnum);
1034 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1039 * The LEB might have been put meanwhile, and the task which put it is
1040 * probably waiting on @ubi->move_mutex. No need to continue the work,
1043 if (vol->eba_tbl[lnum] != from) {
1044 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1045 vol_id, lnum, from, vol->eba_tbl[lnum]);
1046 err = MOVE_CANCEL_RACE;
1047 goto out_unlock_leb;
1051 * OK, now the LEB is locked and we can safely start moving it. Since
1052 * this function utilizes the @ubi->peb_buf buffer which is shared
1053 * with some other functions - we lock the buffer by taking the
1056 mutex_lock(&ubi->buf_mutex);
1057 dbg_wl("read %d bytes of data", aldata_size);
1058 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1059 if (err && err != UBI_IO_BITFLIPS) {
1060 ubi_warn("error %d while reading data from PEB %d",
1062 err = MOVE_SOURCE_RD_ERR;
1063 goto out_unlock_buf;
1067 * Now we have got to calculate how much data we have to copy. In
1068 * case of a static volume it is fairly easy - the VID header contains
1069 * the data size. In case of a dynamic volume it is more difficult - we
1070 * have to read the contents, cut 0xFF bytes from the end and copy only
1071 * the first part. We must do this to avoid writing 0xFF bytes as it
1072 * may have some side-effects. And not only this. It is important not
1073 * to include those 0xFFs to CRC because later the they may be filled
1076 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1077 aldata_size = data_size =
1078 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1081 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1085 * It may turn out to be that the whole @from physical eraseblock
1086 * contains only 0xFF bytes. Then we have to only write the VID header
1087 * and do not write any data. This also means we should not set
1088 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1090 if (data_size > 0) {
1091 vid_hdr->copy_flag = 1;
1092 vid_hdr->data_size = cpu_to_be32(data_size);
1093 vid_hdr->data_crc = cpu_to_be32(crc);
1095 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1097 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1100 err = MOVE_TARGET_WR_ERR;
1101 goto out_unlock_buf;
1106 /* Read the VID header back and check if it was written correctly */
1107 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1109 if (err != UBI_IO_BITFLIPS) {
1110 ubi_warn("error %d while reading VID header back from PEB %d",
1112 if (is_error_sane(err))
1113 err = MOVE_TARGET_RD_ERR;
1115 err = MOVE_TARGET_BITFLIPS;
1116 goto out_unlock_buf;
1119 if (data_size > 0) {
1120 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1123 err = MOVE_TARGET_WR_ERR;
1124 goto out_unlock_buf;
1130 * We've written the data and are going to read it back to make
1131 * sure it was written correctly.
1133 memset(ubi->peb_buf, 0xFF, aldata_size);
1134 err = ubi_io_read_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1136 if (err != UBI_IO_BITFLIPS) {
1137 ubi_warn("error %d while reading data back from PEB %d",
1139 if (is_error_sane(err))
1140 err = MOVE_TARGET_RD_ERR;
1142 err = MOVE_TARGET_BITFLIPS;
1143 goto out_unlock_buf;
1148 if (crc != crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size)) {
1149 ubi_warn("read data back from PEB %d and it is different",
1152 goto out_unlock_buf;
1156 ubi_assert(vol->eba_tbl[lnum] == from);
1157 down_read(&ubi->fm_sem);
1158 vol->eba_tbl[lnum] = to;
1159 up_read(&ubi->fm_sem);
1162 mutex_unlock(&ubi->buf_mutex);
1164 leb_write_unlock(ubi, vol_id, lnum);
1169 * print_rsvd_warning - warn about not having enough reserved PEBs.
1170 * @ubi: UBI device description object
1172 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1173 * cannot reserve enough PEBs for bad block handling. This function makes a
1174 * decision whether we have to print a warning or not. The algorithm is as
1176 * o if this is a new UBI image, then just print the warning
1177 * o if this is an UBI image which has already been used for some time, print
1178 * a warning only if we can reserve less than 10% of the expected amount of
1181 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1182 * of PEBs becomes smaller, which is normal and we do not want to scare users
1183 * with a warning every time they attach the MTD device. This was an issue
1184 * reported by real users.
1186 static void print_rsvd_warning(struct ubi_device *ubi,
1187 struct ubi_attach_info *ai)
1190 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1191 * large number to distinguish between newly flashed and used images.
1193 if (ai->max_sqnum > (1 << 18)) {
1194 int min = ubi->beb_rsvd_level / 10;
1198 if (ubi->beb_rsvd_pebs > min)
1202 ubi_warn("cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1203 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1204 if (ubi->corr_peb_count)
1205 ubi_warn("%d PEBs are corrupted and not used",
1206 ubi->corr_peb_count);
1210 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1211 * @ubi: UBI device description object
1212 * @ai_fastmap: UBI attach info object created by fastmap
1213 * @ai_scan: UBI attach info object created by scanning
1215 * Returns < 0 in case of an internal error, 0 otherwise.
1216 * If a bad EBA table entry was found it will be printed out and
1217 * ubi_assert() triggers.
1219 int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1220 struct ubi_attach_info *ai_scan)
1222 int i, j, num_volumes, ret = 0;
1223 int **scan_eba, **fm_eba;
1224 struct ubi_ainf_volume *av;
1225 struct ubi_volume *vol;
1226 struct ubi_ainf_peb *aeb;
1229 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1231 scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1235 fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1241 for (i = 0; i < num_volumes; i++) {
1242 vol = ubi->volumes[i];
1246 scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1253 fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1260 for (j = 0; j < vol->reserved_pebs; j++)
1261 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1263 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1267 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1268 scan_eba[i][aeb->lnum] = aeb->pnum;
1270 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1274 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1275 fm_eba[i][aeb->lnum] = aeb->pnum;
1277 for (j = 0; j < vol->reserved_pebs; j++) {
1278 if (scan_eba[i][j] != fm_eba[i][j]) {
1279 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1280 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1283 ubi_err("LEB:%i:%i is PEB:%i instead of %i!",
1284 vol->vol_id, i, fm_eba[i][j],
1292 for (i = 0; i < num_volumes; i++) {
1293 if (!ubi->volumes[i])
1306 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1307 * @ubi: UBI device description object
1308 * @ai: attaching information
1310 * This function returns zero in case of success and a negative error code in
1313 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1315 int i, j, err, num_volumes;
1316 struct ubi_ainf_volume *av;
1317 struct ubi_volume *vol;
1318 struct ubi_ainf_peb *aeb;
1321 dbg_eba("initialize EBA sub-system");
1323 spin_lock_init(&ubi->ltree_lock);
1324 mutex_init(&ubi->alc_mutex);
1325 ubi->ltree = RB_ROOT;
1327 ubi->global_sqnum = ai->max_sqnum + 1;
1328 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1330 for (i = 0; i < num_volumes; i++) {
1331 vol = ubi->volumes[i];
1337 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1339 if (!vol->eba_tbl) {
1344 for (j = 0; j < vol->reserved_pebs; j++)
1345 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1347 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1351 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1352 if (aeb->lnum >= vol->reserved_pebs)
1354 * This may happen in case of an unclean reboot
1357 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1358 vol->eba_tbl[aeb->lnum] = aeb->pnum;
1362 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1363 ubi_err("no enough physical eraseblocks (%d, need %d)",
1364 ubi->avail_pebs, EBA_RESERVED_PEBS);
1365 if (ubi->corr_peb_count)
1366 ubi_err("%d PEBs are corrupted and not used",
1367 ubi->corr_peb_count);
1371 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1372 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1374 if (ubi->bad_allowed) {
1375 ubi_calculate_reserved(ubi);
1377 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1378 /* No enough free physical eraseblocks */
1379 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1380 print_rsvd_warning(ubi, ai);
1382 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1384 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1385 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1388 dbg_eba("EBA sub-system is initialized");
1392 for (i = 0; i < num_volumes; i++) {
1393 if (!ubi->volumes[i])
1395 kfree(ubi->volumes[i]->eba_tbl);
1396 ubi->volumes[i]->eba_tbl = NULL;