2 * Copyright (c) International Business Machines Corp., 2006
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 * Author: Artem Bityutskiy (Битюцкий Артём)
22 * The UBI Eraseblock Association (EBA) sub-system.
24 * This sub-system is responsible for I/O to/from logical eraseblock.
26 * Although in this implementation the EBA table is fully kept and managed in
27 * RAM, which assumes poor scalability, it might be (partially) maintained on
28 * flash in future implementations.
30 * The EBA sub-system implements per-logical eraseblock locking. Before
31 * accessing a logical eraseblock it is locked for reading or writing. The
32 * per-logical eraseblock locking is implemented by means of the lock tree. The
33 * lock tree is an RB-tree which refers all the currently locked logical
34 * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
35 * They are indexed by (@vol_id, @lnum) pairs.
37 * EBA also maintains the global sequence counter which is incremented each
38 * time a logical eraseblock is mapped to a physical eraseblock and it is
39 * stored in the volume identifier header. This means that each VID header has
40 * a unique sequence number. The sequence number is only increased an we assume
41 * 64 bits is enough to never overflow.
44 #include <linux/slab.h>
45 #include <linux/crc32.h>
46 #include <linux/err.h>
49 /* Number of physical eraseblocks reserved for atomic LEB change operation */
50 #define EBA_RESERVED_PEBS 1
53 * next_sqnum - get next sequence number.
54 * @ubi: UBI device description object
56 * This function returns next sequence number to use, which is just the current
57 * global sequence counter value. It also increases the global sequence
60 unsigned long long ubi_next_sqnum(struct ubi_device *ubi)
62 unsigned long long sqnum;
64 spin_lock(&ubi->ltree_lock);
65 sqnum = ubi->global_sqnum++;
66 spin_unlock(&ubi->ltree_lock);
72 * ubi_get_compat - get compatibility flags of a volume.
73 * @ubi: UBI device description object
76 * This function returns compatibility flags for an internal volume. User
77 * volumes have no compatibility flags, so %0 is returned.
79 static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
81 if (vol_id == UBI_LAYOUT_VOLUME_ID)
82 return UBI_LAYOUT_VOLUME_COMPAT;
87 * ltree_lookup - look up the lock tree.
88 * @ubi: UBI device description object
90 * @lnum: logical eraseblock number
92 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
93 * object if the logical eraseblock is locked and %NULL if it is not.
94 * @ubi->ltree_lock has to be locked.
96 static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
101 p = ubi->ltree.rb_node;
103 struct ubi_ltree_entry *le;
105 le = rb_entry(p, struct ubi_ltree_entry, rb);
107 if (vol_id < le->vol_id)
109 else if (vol_id > le->vol_id)
114 else if (lnum > le->lnum)
125 * ltree_add_entry - add new entry to the lock tree.
126 * @ubi: UBI device description object
128 * @lnum: logical eraseblock number
130 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
131 * lock tree. If such entry is already there, its usage counter is increased.
132 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
135 static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
136 int vol_id, int lnum)
138 struct ubi_ltree_entry *le, *le1, *le_free;
140 le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
142 return ERR_PTR(-ENOMEM);
145 init_rwsem(&le->mutex);
149 spin_lock(&ubi->ltree_lock);
150 le1 = ltree_lookup(ubi, vol_id, lnum);
154 * This logical eraseblock is already locked. The newly
155 * allocated lock entry is not needed.
160 struct rb_node **p, *parent = NULL;
163 * No lock entry, add the newly allocated one to the
164 * @ubi->ltree RB-tree.
168 p = &ubi->ltree.rb_node;
171 le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
173 if (vol_id < le1->vol_id)
175 else if (vol_id > le1->vol_id)
178 ubi_assert(lnum != le1->lnum);
179 if (lnum < le1->lnum)
186 rb_link_node(&le->rb, parent, p);
187 rb_insert_color(&le->rb, &ubi->ltree);
190 spin_unlock(&ubi->ltree_lock);
197 * leb_read_lock - lock logical eraseblock for reading.
198 * @ubi: UBI device description object
200 * @lnum: logical eraseblock number
202 * This function locks a logical eraseblock for reading. Returns zero in case
203 * of success and a negative error code in case of failure.
205 static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
207 struct ubi_ltree_entry *le;
209 le = ltree_add_entry(ubi, vol_id, lnum);
212 down_read(&le->mutex);
217 * leb_read_unlock - unlock logical eraseblock.
218 * @ubi: UBI device description object
220 * @lnum: logical eraseblock number
222 static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
224 struct ubi_ltree_entry *le;
226 spin_lock(&ubi->ltree_lock);
227 le = ltree_lookup(ubi, vol_id, lnum);
229 ubi_assert(le->users >= 0);
231 if (le->users == 0) {
232 rb_erase(&le->rb, &ubi->ltree);
235 spin_unlock(&ubi->ltree_lock);
239 * leb_write_lock - lock logical eraseblock for writing.
240 * @ubi: UBI device description object
242 * @lnum: logical eraseblock number
244 * This function locks a logical eraseblock for writing. Returns zero in case
245 * of success and a negative error code in case of failure.
247 static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
249 struct ubi_ltree_entry *le;
251 le = ltree_add_entry(ubi, vol_id, lnum);
254 down_write(&le->mutex);
259 * leb_write_lock - lock logical eraseblock for writing.
260 * @ubi: UBI device description object
262 * @lnum: logical eraseblock number
264 * This function locks a logical eraseblock for writing if there is no
265 * contention and does nothing if there is contention. Returns %0 in case of
266 * success, %1 in case of contention, and and a negative error code in case of
269 static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
271 struct ubi_ltree_entry *le;
273 le = ltree_add_entry(ubi, vol_id, lnum);
276 if (down_write_trylock(&le->mutex))
279 /* Contention, cancel */
280 spin_lock(&ubi->ltree_lock);
282 ubi_assert(le->users >= 0);
283 if (le->users == 0) {
284 rb_erase(&le->rb, &ubi->ltree);
287 spin_unlock(&ubi->ltree_lock);
293 * leb_write_unlock - unlock logical eraseblock.
294 * @ubi: UBI device description object
296 * @lnum: logical eraseblock number
298 static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
300 struct ubi_ltree_entry *le;
302 spin_lock(&ubi->ltree_lock);
303 le = ltree_lookup(ubi, vol_id, lnum);
305 ubi_assert(le->users >= 0);
306 up_write(&le->mutex);
307 if (le->users == 0) {
308 rb_erase(&le->rb, &ubi->ltree);
311 spin_unlock(&ubi->ltree_lock);
315 * ubi_eba_unmap_leb - un-map logical eraseblock.
316 * @ubi: UBI device description object
317 * @vol: volume description object
318 * @lnum: logical eraseblock number
320 * This function un-maps logical eraseblock @lnum and schedules corresponding
321 * physical eraseblock for erasure. Returns zero in case of success and a
322 * negative error code in case of failure.
324 int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
327 int err, pnum, vol_id = vol->vol_id;
332 err = leb_write_lock(ubi, vol_id, lnum);
336 pnum = vol->eba_tbl[lnum];
338 /* This logical eraseblock is already unmapped */
341 dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
343 down_read(&ubi->fm_sem);
344 vol->eba_tbl[lnum] = UBI_LEB_UNMAPPED;
345 up_read(&ubi->fm_sem);
346 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
349 leb_write_unlock(ubi, vol_id, lnum);
354 * ubi_eba_read_leb - read data.
355 * @ubi: UBI device description object
356 * @vol: volume description object
357 * @lnum: logical eraseblock number
358 * @buf: buffer to store the read data
359 * @offset: offset from where to read
360 * @len: how many bytes to read
361 * @check: data CRC check flag
363 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
364 * bytes. The @check flag only makes sense for static volumes and forces
365 * eraseblock data CRC checking.
367 * In case of success this function returns zero. In case of a static volume,
368 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
369 * returned for any volume type if an ECC error was detected by the MTD device
370 * driver. Other negative error cored may be returned in case of other errors.
372 int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
373 void *buf, int offset, int len, int check)
375 int err, pnum, scrub = 0, vol_id = vol->vol_id;
376 struct ubi_vid_hdr *vid_hdr;
377 uint32_t uninitialized_var(crc);
379 err = leb_read_lock(ubi, vol_id, lnum);
383 pnum = vol->eba_tbl[lnum];
386 * The logical eraseblock is not mapped, fill the whole buffer
387 * with 0xFF bytes. The exception is static volumes for which
388 * it is an error to read unmapped logical eraseblocks.
390 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
391 len, offset, vol_id, lnum);
392 leb_read_unlock(ubi, vol_id, lnum);
393 ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
394 memset(buf, 0xFF, len);
398 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
399 len, offset, vol_id, lnum, pnum);
401 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
406 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
412 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
413 if (err && err != UBI_IO_BITFLIPS) {
416 * The header is either absent or corrupted.
417 * The former case means there is a bug -
418 * switch to read-only mode just in case.
419 * The latter case means a real corruption - we
420 * may try to recover data. FIXME: but this is
423 if (err == UBI_IO_BAD_HDR_EBADMSG ||
424 err == UBI_IO_BAD_HDR) {
425 ubi_warn(ubi, "corrupted VID header at PEB %d, LEB %d:%d",
433 } else if (err == UBI_IO_BITFLIPS)
436 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
437 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
439 crc = be32_to_cpu(vid_hdr->data_crc);
440 ubi_free_vid_hdr(ubi, vid_hdr);
443 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
445 if (err == UBI_IO_BITFLIPS)
447 else if (mtd_is_eccerr(err)) {
448 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
452 ubi_msg(ubi, "force data checking");
461 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
463 ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
471 err = ubi_wl_scrub_peb(ubi, pnum);
473 leb_read_unlock(ubi, vol_id, lnum);
477 ubi_free_vid_hdr(ubi, vid_hdr);
479 leb_read_unlock(ubi, vol_id, lnum);
484 * ubi_eba_read_leb_sg - read data into a scatter gather list.
485 * @ubi: UBI device description object
486 * @vol: volume description object
487 * @lnum: logical eraseblock number
488 * @sgl: UBI scatter gather list to store the read data
489 * @offset: offset from where to read
490 * @len: how many bytes to read
491 * @check: data CRC check flag
493 * This function works exactly like ubi_eba_read_leb(). But instead of
494 * storing the read data into a buffer it writes to an UBI scatter gather
497 int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
498 struct ubi_sgl *sgl, int lnum, int offset, int len,
503 struct scatterlist *sg;
506 ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
507 sg = &sgl->sg[sgl->list_pos];
508 if (len < sg->length - sgl->page_pos)
511 to_read = sg->length - sgl->page_pos;
513 ret = ubi_eba_read_leb(ubi, vol, lnum,
514 sg_virt(sg) + sgl->page_pos, offset,
522 sgl->page_pos += to_read;
523 if (sgl->page_pos == sg->length) {
539 * recover_peb - recover from write failure.
540 * @ubi: UBI device description object
541 * @pnum: the physical eraseblock to recover
543 * @lnum: logical eraseblock number
544 * @buf: data which was not written because of the write failure
545 * @offset: offset of the failed write
546 * @len: how many bytes should have been written
548 * This function is called in case of a write failure and moves all good data
549 * from the potentially bad physical eraseblock to a good physical eraseblock.
550 * This function also writes the data which was not written due to the failure.
551 * Returns new physical eraseblock number in case of success, and a negative
552 * error code in case of failure.
554 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
555 const void *buf, int offset, int len)
557 int err, idx = vol_id2idx(ubi, vol_id), new_pnum, data_size, tries = 0;
558 struct ubi_volume *vol = ubi->volumes[idx];
559 struct ubi_vid_hdr *vid_hdr;
561 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
566 new_pnum = ubi_wl_get_peb(ubi);
568 ubi_free_vid_hdr(ubi, vid_hdr);
572 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
575 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
576 if (err && err != UBI_IO_BITFLIPS) {
582 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
583 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
587 data_size = offset + len;
588 mutex_lock(&ubi->buf_mutex);
589 memset(ubi->peb_buf + offset, 0xFF, len);
591 /* Read everything before the area where the write failure happened */
593 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
594 if (err && err != UBI_IO_BITFLIPS)
598 memcpy(ubi->peb_buf + offset, buf, len);
600 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
602 mutex_unlock(&ubi->buf_mutex);
606 mutex_unlock(&ubi->buf_mutex);
607 ubi_free_vid_hdr(ubi, vid_hdr);
609 down_read(&ubi->fm_sem);
610 vol->eba_tbl[lnum] = new_pnum;
611 up_read(&ubi->fm_sem);
612 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
614 ubi_msg(ubi, "data was successfully recovered");
618 mutex_unlock(&ubi->buf_mutex);
620 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
621 ubi_free_vid_hdr(ubi, vid_hdr);
626 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
629 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
630 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
631 if (++tries > UBI_IO_RETRIES) {
632 ubi_free_vid_hdr(ubi, vid_hdr);
635 ubi_msg(ubi, "try again");
640 * ubi_eba_write_leb - write data to dynamic volume.
641 * @ubi: UBI device description object
642 * @vol: volume description object
643 * @lnum: logical eraseblock number
644 * @buf: the data to write
645 * @offset: offset within the logical eraseblock where to write
646 * @len: how many bytes to write
648 * This function writes data to logical eraseblock @lnum of a dynamic volume
649 * @vol. Returns zero in case of success and a negative error code in case
650 * of failure. In case of error, it is possible that something was still
651 * written to the flash media, but may be some garbage.
653 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
654 const void *buf, int offset, int len)
656 int err, pnum, tries = 0, vol_id = vol->vol_id;
657 struct ubi_vid_hdr *vid_hdr;
662 err = leb_write_lock(ubi, vol_id, lnum);
666 pnum = vol->eba_tbl[lnum];
668 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
669 len, offset, vol_id, lnum, pnum);
671 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
673 ubi_warn(ubi, "failed to write data to PEB %d", pnum);
674 if (err == -EIO && ubi->bad_allowed)
675 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
680 leb_write_unlock(ubi, vol_id, lnum);
685 * The logical eraseblock is not mapped. We have to get a free physical
686 * eraseblock and write the volume identifier header there first.
688 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
690 leb_write_unlock(ubi, vol_id, lnum);
694 vid_hdr->vol_type = UBI_VID_DYNAMIC;
695 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
696 vid_hdr->vol_id = cpu_to_be32(vol_id);
697 vid_hdr->lnum = cpu_to_be32(lnum);
698 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
699 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
702 pnum = ubi_wl_get_peb(ubi);
704 ubi_free_vid_hdr(ubi, vid_hdr);
705 leb_write_unlock(ubi, vol_id, lnum);
709 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
710 len, offset, vol_id, lnum, pnum);
712 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
714 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
720 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
722 ubi_warn(ubi, "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
723 len, offset, vol_id, lnum, pnum);
728 down_read(&ubi->fm_sem);
729 vol->eba_tbl[lnum] = pnum;
730 up_read(&ubi->fm_sem);
732 leb_write_unlock(ubi, vol_id, lnum);
733 ubi_free_vid_hdr(ubi, vid_hdr);
737 if (err != -EIO || !ubi->bad_allowed) {
739 leb_write_unlock(ubi, vol_id, lnum);
740 ubi_free_vid_hdr(ubi, vid_hdr);
745 * Fortunately, this is the first write operation to this physical
746 * eraseblock, so just put it and request a new one. We assume that if
747 * this physical eraseblock went bad, the erase code will handle that.
749 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
750 if (err || ++tries > UBI_IO_RETRIES) {
752 leb_write_unlock(ubi, vol_id, lnum);
753 ubi_free_vid_hdr(ubi, vid_hdr);
757 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
758 ubi_msg(ubi, "try another PEB");
763 * ubi_eba_write_leb_st - write data to static volume.
764 * @ubi: UBI device description object
765 * @vol: volume description object
766 * @lnum: logical eraseblock number
767 * @buf: data to write
768 * @len: how many bytes to write
769 * @used_ebs: how many logical eraseblocks will this volume contain
771 * This function writes data to logical eraseblock @lnum of static volume
772 * @vol. The @used_ebs argument should contain total number of logical
773 * eraseblock in this static volume.
775 * When writing to the last logical eraseblock, the @len argument doesn't have
776 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
777 * to the real data size, although the @buf buffer has to contain the
778 * alignment. In all other cases, @len has to be aligned.
780 * It is prohibited to write more than once to logical eraseblocks of static
781 * volumes. This function returns zero in case of success and a negative error
782 * code in case of failure.
784 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
785 int lnum, const void *buf, int len, int used_ebs)
787 int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
788 struct ubi_vid_hdr *vid_hdr;
794 if (lnum == used_ebs - 1)
795 /* If this is the last LEB @len may be unaligned */
796 len = ALIGN(data_size, ubi->min_io_size);
798 ubi_assert(!(len & (ubi->min_io_size - 1)));
800 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
804 err = leb_write_lock(ubi, vol_id, lnum);
806 ubi_free_vid_hdr(ubi, vid_hdr);
810 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
811 vid_hdr->vol_id = cpu_to_be32(vol_id);
812 vid_hdr->lnum = cpu_to_be32(lnum);
813 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
814 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
816 crc = crc32(UBI_CRC32_INIT, buf, data_size);
817 vid_hdr->vol_type = UBI_VID_STATIC;
818 vid_hdr->data_size = cpu_to_be32(data_size);
819 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
820 vid_hdr->data_crc = cpu_to_be32(crc);
823 pnum = ubi_wl_get_peb(ubi);
825 ubi_free_vid_hdr(ubi, vid_hdr);
826 leb_write_unlock(ubi, vol_id, lnum);
830 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
831 len, vol_id, lnum, pnum, used_ebs);
833 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
835 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
840 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
842 ubi_warn(ubi, "failed to write %d bytes of data to PEB %d",
847 ubi_assert(vol->eba_tbl[lnum] < 0);
848 down_read(&ubi->fm_sem);
849 vol->eba_tbl[lnum] = pnum;
850 up_read(&ubi->fm_sem);
852 leb_write_unlock(ubi, vol_id, lnum);
853 ubi_free_vid_hdr(ubi, vid_hdr);
857 if (err != -EIO || !ubi->bad_allowed) {
859 * This flash device does not admit of bad eraseblocks or
860 * something nasty and unexpected happened. Switch to read-only
864 leb_write_unlock(ubi, vol_id, lnum);
865 ubi_free_vid_hdr(ubi, vid_hdr);
869 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
870 if (err || ++tries > UBI_IO_RETRIES) {
872 leb_write_unlock(ubi, vol_id, lnum);
873 ubi_free_vid_hdr(ubi, vid_hdr);
877 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
878 ubi_msg(ubi, "try another PEB");
883 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
884 * @ubi: UBI device description object
885 * @vol: volume description object
886 * @lnum: logical eraseblock number
887 * @buf: data to write
888 * @len: how many bytes to write
890 * This function changes the contents of a logical eraseblock atomically. @buf
891 * has to contain new logical eraseblock data, and @len - the length of the
892 * data, which has to be aligned. This function guarantees that in case of an
893 * unclean reboot the old contents is preserved. Returns zero in case of
894 * success and a negative error code in case of failure.
896 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
897 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
899 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
900 int lnum, const void *buf, int len)
902 int err, pnum, tries = 0, vol_id = vol->vol_id;
903 struct ubi_vid_hdr *vid_hdr;
911 * Special case when data length is zero. In this case the LEB
912 * has to be unmapped and mapped somewhere else.
914 err = ubi_eba_unmap_leb(ubi, vol, lnum);
917 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
920 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
924 mutex_lock(&ubi->alc_mutex);
925 err = leb_write_lock(ubi, vol_id, lnum);
929 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
930 vid_hdr->vol_id = cpu_to_be32(vol_id);
931 vid_hdr->lnum = cpu_to_be32(lnum);
932 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
933 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
935 crc = crc32(UBI_CRC32_INIT, buf, len);
936 vid_hdr->vol_type = UBI_VID_DYNAMIC;
937 vid_hdr->data_size = cpu_to_be32(len);
938 vid_hdr->copy_flag = 1;
939 vid_hdr->data_crc = cpu_to_be32(crc);
942 pnum = ubi_wl_get_peb(ubi);
948 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
949 vol_id, lnum, vol->eba_tbl[lnum], pnum);
951 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
953 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
958 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
960 ubi_warn(ubi, "failed to write %d bytes of data to PEB %d",
965 if (vol->eba_tbl[lnum] >= 0) {
966 err = ubi_wl_put_peb(ubi, vol_id, lnum, vol->eba_tbl[lnum], 0);
971 down_read(&ubi->fm_sem);
972 vol->eba_tbl[lnum] = pnum;
973 up_read(&ubi->fm_sem);
976 leb_write_unlock(ubi, vol_id, lnum);
978 mutex_unlock(&ubi->alc_mutex);
979 ubi_free_vid_hdr(ubi, vid_hdr);
983 if (err != -EIO || !ubi->bad_allowed) {
985 * This flash device does not admit of bad eraseblocks or
986 * something nasty and unexpected happened. Switch to read-only
993 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
994 if (err || ++tries > UBI_IO_RETRIES) {
999 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1000 ubi_msg(ubi, "try another PEB");
1005 * is_error_sane - check whether a read error is sane.
1006 * @err: code of the error happened during reading
1008 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1009 * cannot read data from the target PEB (an error @err happened). If the error
1010 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1011 * fatal and UBI will be switched to R/O mode later.
1013 * The idea is that we try not to switch to R/O mode if the read error is
1014 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1015 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1016 * mode, simply because we do not know what happened at the MTD level, and we
1017 * cannot handle this. E.g., the underlying driver may have become crazy, and
1018 * it is safer to switch to R/O mode to preserve the data.
1020 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1021 * which we have just written.
1023 static int is_error_sane(int err)
1025 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1026 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1032 * ubi_eba_copy_leb - copy logical eraseblock.
1033 * @ubi: UBI device description object
1034 * @from: physical eraseblock number from where to copy
1035 * @to: physical eraseblock number where to copy
1036 * @vid_hdr: VID header of the @from physical eraseblock
1038 * This function copies logical eraseblock from physical eraseblock @from to
1039 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1040 * function. Returns:
1041 * o %0 in case of success;
1042 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1043 * o a negative error code in case of failure.
1045 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1046 struct ubi_vid_hdr *vid_hdr)
1048 int err, vol_id, lnum, data_size, aldata_size, idx;
1049 struct ubi_volume *vol;
1052 vol_id = be32_to_cpu(vid_hdr->vol_id);
1053 lnum = be32_to_cpu(vid_hdr->lnum);
1055 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1057 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1058 data_size = be32_to_cpu(vid_hdr->data_size);
1059 aldata_size = ALIGN(data_size, ubi->min_io_size);
1061 data_size = aldata_size =
1062 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1064 idx = vol_id2idx(ubi, vol_id);
1065 spin_lock(&ubi->volumes_lock);
1067 * Note, we may race with volume deletion, which means that the volume
1068 * this logical eraseblock belongs to might be being deleted. Since the
1069 * volume deletion un-maps all the volume's logical eraseblocks, it will
1070 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1072 vol = ubi->volumes[idx];
1073 spin_unlock(&ubi->volumes_lock);
1075 /* No need to do further work, cancel */
1076 dbg_wl("volume %d is being removed, cancel", vol_id);
1077 return MOVE_CANCEL_RACE;
1081 * We do not want anybody to write to this logical eraseblock while we
1082 * are moving it, so lock it.
1084 * Note, we are using non-waiting locking here, because we cannot sleep
1085 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1086 * unmapping the LEB which is mapped to the PEB we are going to move
1087 * (@from). This task locks the LEB and goes sleep in the
1088 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1089 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1090 * LEB is already locked, we just do not move it and return
1091 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1092 * we do not know the reasons of the contention - it may be just a
1093 * normal I/O on this LEB, so we want to re-try.
1095 err = leb_write_trylock(ubi, vol_id, lnum);
1097 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1102 * The LEB might have been put meanwhile, and the task which put it is
1103 * probably waiting on @ubi->move_mutex. No need to continue the work,
1106 if (vol->eba_tbl[lnum] != from) {
1107 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1108 vol_id, lnum, from, vol->eba_tbl[lnum]);
1109 err = MOVE_CANCEL_RACE;
1110 goto out_unlock_leb;
1114 * OK, now the LEB is locked and we can safely start moving it. Since
1115 * this function utilizes the @ubi->peb_buf buffer which is shared
1116 * with some other functions - we lock the buffer by taking the
1119 mutex_lock(&ubi->buf_mutex);
1120 dbg_wl("read %d bytes of data", aldata_size);
1121 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1122 if (err && err != UBI_IO_BITFLIPS) {
1123 ubi_warn(ubi, "error %d while reading data from PEB %d",
1125 err = MOVE_SOURCE_RD_ERR;
1126 goto out_unlock_buf;
1130 * Now we have got to calculate how much data we have to copy. In
1131 * case of a static volume it is fairly easy - the VID header contains
1132 * the data size. In case of a dynamic volume it is more difficult - we
1133 * have to read the contents, cut 0xFF bytes from the end and copy only
1134 * the first part. We must do this to avoid writing 0xFF bytes as it
1135 * may have some side-effects. And not only this. It is important not
1136 * to include those 0xFFs to CRC because later the they may be filled
1139 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1140 aldata_size = data_size =
1141 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1144 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1148 * It may turn out to be that the whole @from physical eraseblock
1149 * contains only 0xFF bytes. Then we have to only write the VID header
1150 * and do not write any data. This also means we should not set
1151 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1153 if (data_size > 0) {
1154 vid_hdr->copy_flag = 1;
1155 vid_hdr->data_size = cpu_to_be32(data_size);
1156 vid_hdr->data_crc = cpu_to_be32(crc);
1158 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1160 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1163 err = MOVE_TARGET_WR_ERR;
1164 goto out_unlock_buf;
1169 /* Read the VID header back and check if it was written correctly */
1170 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1172 if (err != UBI_IO_BITFLIPS) {
1173 ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1175 if (is_error_sane(err))
1176 err = MOVE_TARGET_RD_ERR;
1178 err = MOVE_TARGET_BITFLIPS;
1179 goto out_unlock_buf;
1182 if (data_size > 0) {
1183 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1186 err = MOVE_TARGET_WR_ERR;
1187 goto out_unlock_buf;
1193 * We've written the data and are going to read it back to make
1194 * sure it was written correctly.
1196 memset(ubi->peb_buf, 0xFF, aldata_size);
1197 err = ubi_io_read_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1199 if (err != UBI_IO_BITFLIPS) {
1200 ubi_warn(ubi, "error %d while reading data back from PEB %d",
1202 if (is_error_sane(err))
1203 err = MOVE_TARGET_RD_ERR;
1205 err = MOVE_TARGET_BITFLIPS;
1206 goto out_unlock_buf;
1211 if (crc != crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size)) {
1212 ubi_warn(ubi, "read data back from PEB %d and it is different",
1215 goto out_unlock_buf;
1219 ubi_assert(vol->eba_tbl[lnum] == from);
1220 down_read(&ubi->fm_sem);
1221 vol->eba_tbl[lnum] = to;
1222 up_read(&ubi->fm_sem);
1225 mutex_unlock(&ubi->buf_mutex);
1227 leb_write_unlock(ubi, vol_id, lnum);
1232 * print_rsvd_warning - warn about not having enough reserved PEBs.
1233 * @ubi: UBI device description object
1235 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1236 * cannot reserve enough PEBs for bad block handling. This function makes a
1237 * decision whether we have to print a warning or not. The algorithm is as
1239 * o if this is a new UBI image, then just print the warning
1240 * o if this is an UBI image which has already been used for some time, print
1241 * a warning only if we can reserve less than 10% of the expected amount of
1244 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1245 * of PEBs becomes smaller, which is normal and we do not want to scare users
1246 * with a warning every time they attach the MTD device. This was an issue
1247 * reported by real users.
1249 static void print_rsvd_warning(struct ubi_device *ubi,
1250 struct ubi_attach_info *ai)
1253 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1254 * large number to distinguish between newly flashed and used images.
1256 if (ai->max_sqnum > (1 << 18)) {
1257 int min = ubi->beb_rsvd_level / 10;
1261 if (ubi->beb_rsvd_pebs > min)
1265 ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1266 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1267 if (ubi->corr_peb_count)
1268 ubi_warn(ubi, "%d PEBs are corrupted and not used",
1269 ubi->corr_peb_count);
1273 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1274 * @ubi: UBI device description object
1275 * @ai_fastmap: UBI attach info object created by fastmap
1276 * @ai_scan: UBI attach info object created by scanning
1278 * Returns < 0 in case of an internal error, 0 otherwise.
1279 * If a bad EBA table entry was found it will be printed out and
1280 * ubi_assert() triggers.
1282 int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1283 struct ubi_attach_info *ai_scan)
1285 int i, j, num_volumes, ret = 0;
1286 int **scan_eba, **fm_eba;
1287 struct ubi_ainf_volume *av;
1288 struct ubi_volume *vol;
1289 struct ubi_ainf_peb *aeb;
1292 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1294 scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1298 fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1304 for (i = 0; i < num_volumes; i++) {
1305 vol = ubi->volumes[i];
1309 scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1316 fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1323 for (j = 0; j < vol->reserved_pebs; j++)
1324 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1326 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1330 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1331 scan_eba[i][aeb->lnum] = aeb->pnum;
1333 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1337 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1338 fm_eba[i][aeb->lnum] = aeb->pnum;
1340 for (j = 0; j < vol->reserved_pebs; j++) {
1341 if (scan_eba[i][j] != fm_eba[i][j]) {
1342 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1343 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1346 ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1347 vol->vol_id, i, fm_eba[i][j],
1355 for (i = 0; i < num_volumes; i++) {
1356 if (!ubi->volumes[i])
1369 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1370 * @ubi: UBI device description object
1371 * @ai: attaching information
1373 * This function returns zero in case of success and a negative error code in
1376 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1378 int i, j, err, num_volumes;
1379 struct ubi_ainf_volume *av;
1380 struct ubi_volume *vol;
1381 struct ubi_ainf_peb *aeb;
1384 dbg_eba("initialize EBA sub-system");
1386 spin_lock_init(&ubi->ltree_lock);
1387 mutex_init(&ubi->alc_mutex);
1388 ubi->ltree = RB_ROOT;
1390 ubi->global_sqnum = ai->max_sqnum + 1;
1391 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1393 for (i = 0; i < num_volumes; i++) {
1394 vol = ubi->volumes[i];
1400 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1402 if (!vol->eba_tbl) {
1407 for (j = 0; j < vol->reserved_pebs; j++)
1408 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1410 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1414 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1415 if (aeb->lnum >= vol->reserved_pebs)
1417 * This may happen in case of an unclean reboot
1420 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1421 vol->eba_tbl[aeb->lnum] = aeb->pnum;
1425 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1426 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1427 ubi->avail_pebs, EBA_RESERVED_PEBS);
1428 if (ubi->corr_peb_count)
1429 ubi_err(ubi, "%d PEBs are corrupted and not used",
1430 ubi->corr_peb_count);
1434 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1435 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1437 if (ubi->bad_allowed) {
1438 ubi_calculate_reserved(ubi);
1440 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1441 /* No enough free physical eraseblocks */
1442 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1443 print_rsvd_warning(ubi, ai);
1445 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1447 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1448 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1451 dbg_eba("EBA sub-system is initialized");
1455 for (i = 0; i < num_volumes; i++) {
1456 if (!ubi->volumes[i])
1458 kfree(ubi->volumes[i]->eba_tbl);
1459 ubi->volumes[i]->eba_tbl = NULL;