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 static unsigned long long 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 vol->eba_tbl[lnum] = UBI_LEB_UNMAPPED;
344 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
347 leb_write_unlock(ubi, vol_id, lnum);
352 * ubi_eba_read_leb - read data.
353 * @ubi: UBI device description object
354 * @vol: volume description object
355 * @lnum: logical eraseblock number
356 * @buf: buffer to store the read data
357 * @offset: offset from where to read
358 * @len: how many bytes to read
359 * @check: data CRC check flag
361 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
362 * bytes. The @check flag only makes sense for static volumes and forces
363 * eraseblock data CRC checking.
365 * In case of success this function returns zero. In case of a static volume,
366 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
367 * returned for any volume type if an ECC error was detected by the MTD device
368 * driver. Other negative error cored may be returned in case of other errors.
370 int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
371 void *buf, int offset, int len, int check)
373 int err, pnum, scrub = 0, vol_id = vol->vol_id;
374 struct ubi_vid_hdr *vid_hdr;
375 uint32_t uninitialized_var(crc);
377 err = leb_read_lock(ubi, vol_id, lnum);
381 pnum = vol->eba_tbl[lnum];
384 * The logical eraseblock is not mapped, fill the whole buffer
385 * with 0xFF bytes. The exception is static volumes for which
386 * it is an error to read unmapped logical eraseblocks.
388 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
389 len, offset, vol_id, lnum);
390 leb_read_unlock(ubi, vol_id, lnum);
391 ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
392 memset(buf, 0xFF, len);
396 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
397 len, offset, vol_id, lnum, pnum);
399 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
404 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
410 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
411 if (err && err != UBI_IO_BITFLIPS) {
414 * The header is either absent or corrupted.
415 * The former case means there is a bug -
416 * switch to read-only mode just in case.
417 * The latter case means a real corruption - we
418 * may try to recover data. FIXME: but this is
421 if (err == UBI_IO_BAD_HDR_EBADMSG ||
422 err == UBI_IO_BAD_HDR) {
423 ubi_warn("corrupted VID header at PEB %d, LEB %d:%d",
430 } else if (err == UBI_IO_BITFLIPS)
433 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
434 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
436 crc = be32_to_cpu(vid_hdr->data_crc);
437 ubi_free_vid_hdr(ubi, vid_hdr);
440 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
442 if (err == UBI_IO_BITFLIPS) {
445 } else if (mtd_is_eccerr(err)) {
446 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
450 ubi_msg("force data checking");
459 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
461 ubi_warn("CRC error: calculated %#08x, must be %#08x",
469 err = ubi_wl_scrub_peb(ubi, pnum);
471 leb_read_unlock(ubi, vol_id, lnum);
475 ubi_free_vid_hdr(ubi, vid_hdr);
477 leb_read_unlock(ubi, vol_id, lnum);
482 * recover_peb - recover from write failure.
483 * @ubi: UBI device description object
484 * @pnum: the physical eraseblock to recover
486 * @lnum: logical eraseblock number
487 * @buf: data which was not written because of the write failure
488 * @offset: offset of the failed write
489 * @len: how many bytes should have been written
491 * This function is called in case of a write failure and moves all good data
492 * from the potentially bad physical eraseblock to a good physical eraseblock.
493 * This function also writes the data which was not written due to the failure.
494 * Returns new physical eraseblock number in case of success, and a negative
495 * error code in case of failure.
497 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
498 const void *buf, int offset, int len)
500 int err, idx = vol_id2idx(ubi, vol_id), new_pnum, data_size, tries = 0;
501 struct ubi_volume *vol = ubi->volumes[idx];
502 struct ubi_vid_hdr *vid_hdr;
505 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
510 new_pnum = ubi_wl_get_peb(ubi);
512 ubi_free_vid_hdr(ubi, vid_hdr);
516 ubi_msg("recover PEB %d, move data to PEB %d", pnum, new_pnum);
518 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
519 if (err && err != UBI_IO_BITFLIPS) {
525 ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
527 mutex_lock(&ubi->buf_mutex);
528 memset(ubi->peb_buf + offset, 0xFF, len);
530 /* Read everything before the area where the write failure happened */
532 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
533 if (err && err != UBI_IO_BITFLIPS)
537 memcpy(ubi->peb_buf + offset, buf, len);
539 data_size = offset + len;
540 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
541 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
542 vid_hdr->copy_flag = 1;
543 vid_hdr->data_size = cpu_to_be32(data_size);
544 vid_hdr->data_crc = cpu_to_be32(crc);
545 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
547 mutex_unlock(&ubi->buf_mutex);
551 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
553 mutex_unlock(&ubi->buf_mutex);
557 mutex_unlock(&ubi->buf_mutex);
558 ubi_free_vid_hdr(ubi, vid_hdr);
560 vol->eba_tbl[lnum] = new_pnum;
561 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
563 ubi_msg("data was successfully recovered");
567 mutex_unlock(&ubi->buf_mutex);
569 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
570 ubi_free_vid_hdr(ubi, vid_hdr);
575 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
578 ubi_warn("failed to write to PEB %d", new_pnum);
579 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
580 if (++tries > UBI_IO_RETRIES) {
581 ubi_free_vid_hdr(ubi, vid_hdr);
584 ubi_msg("try again");
589 * ubi_eba_write_leb - write data to dynamic volume.
590 * @ubi: UBI device description object
591 * @vol: volume description object
592 * @lnum: logical eraseblock number
593 * @buf: the data to write
594 * @offset: offset within the logical eraseblock where to write
595 * @len: how many bytes to write
597 * This function writes data to logical eraseblock @lnum of a dynamic volume
598 * @vol. Returns zero in case of success and a negative error code in case
599 * of failure. In case of error, it is possible that something was still
600 * written to the flash media, but may be some garbage.
602 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
603 const void *buf, int offset, int len)
605 int err, pnum, tries = 0, vol_id = vol->vol_id;
606 struct ubi_vid_hdr *vid_hdr;
611 err = leb_write_lock(ubi, vol_id, lnum);
615 pnum = vol->eba_tbl[lnum];
617 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
618 len, offset, vol_id, lnum, pnum);
620 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
622 ubi_warn("failed to write data to PEB %d", pnum);
623 if (err == -EIO && ubi->bad_allowed)
624 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
629 leb_write_unlock(ubi, vol_id, lnum);
634 * The logical eraseblock is not mapped. We have to get a free physical
635 * eraseblock and write the volume identifier header there first.
637 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
639 leb_write_unlock(ubi, vol_id, lnum);
643 vid_hdr->vol_type = UBI_VID_DYNAMIC;
644 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
645 vid_hdr->vol_id = cpu_to_be32(vol_id);
646 vid_hdr->lnum = cpu_to_be32(lnum);
647 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
648 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
651 pnum = ubi_wl_get_peb(ubi);
653 ubi_free_vid_hdr(ubi, vid_hdr);
654 leb_write_unlock(ubi, vol_id, lnum);
658 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
659 len, offset, vol_id, lnum, pnum);
661 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
663 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
669 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
671 ubi_warn("failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
672 len, offset, vol_id, lnum, pnum);
677 vol->eba_tbl[lnum] = pnum;
679 leb_write_unlock(ubi, vol_id, lnum);
680 ubi_free_vid_hdr(ubi, vid_hdr);
684 if (err != -EIO || !ubi->bad_allowed) {
686 leb_write_unlock(ubi, vol_id, lnum);
687 ubi_free_vid_hdr(ubi, vid_hdr);
692 * Fortunately, this is the first write operation to this physical
693 * eraseblock, so just put it and request a new one. We assume that if
694 * this physical eraseblock went bad, the erase code will handle that.
696 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
697 if (err || ++tries > UBI_IO_RETRIES) {
699 leb_write_unlock(ubi, vol_id, lnum);
700 ubi_free_vid_hdr(ubi, vid_hdr);
704 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
705 ubi_msg("try another PEB");
710 * ubi_eba_write_leb_st - write data to static volume.
711 * @ubi: UBI device description object
712 * @vol: volume description object
713 * @lnum: logical eraseblock number
714 * @buf: data to write
715 * @len: how many bytes to write
716 * @used_ebs: how many logical eraseblocks will this volume contain
718 * This function writes data to logical eraseblock @lnum of static volume
719 * @vol. The @used_ebs argument should contain total number of logical
720 * eraseblock in this static volume.
722 * When writing to the last logical eraseblock, the @len argument doesn't have
723 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
724 * to the real data size, although the @buf buffer has to contain the
725 * alignment. In all other cases, @len has to be aligned.
727 * It is prohibited to write more than once to logical eraseblocks of static
728 * volumes. This function returns zero in case of success and a negative error
729 * code in case of failure.
731 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
732 int lnum, const void *buf, int len, int used_ebs)
734 int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
735 struct ubi_vid_hdr *vid_hdr;
741 if (lnum == used_ebs - 1)
742 /* If this is the last LEB @len may be unaligned */
743 len = ALIGN(data_size, ubi->min_io_size);
745 ubi_assert(!(len & (ubi->min_io_size - 1)));
747 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
751 err = leb_write_lock(ubi, vol_id, lnum);
753 ubi_free_vid_hdr(ubi, vid_hdr);
757 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
758 vid_hdr->vol_id = cpu_to_be32(vol_id);
759 vid_hdr->lnum = cpu_to_be32(lnum);
760 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
761 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
763 crc = crc32(UBI_CRC32_INIT, buf, data_size);
764 vid_hdr->vol_type = UBI_VID_STATIC;
765 vid_hdr->data_size = cpu_to_be32(data_size);
766 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
767 vid_hdr->data_crc = cpu_to_be32(crc);
770 pnum = ubi_wl_get_peb(ubi);
772 ubi_free_vid_hdr(ubi, vid_hdr);
773 leb_write_unlock(ubi, vol_id, lnum);
777 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
778 len, vol_id, lnum, pnum, used_ebs);
780 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
782 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
787 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
789 ubi_warn("failed to write %d bytes of data to PEB %d",
794 ubi_assert(vol->eba_tbl[lnum] < 0);
795 vol->eba_tbl[lnum] = pnum;
797 leb_write_unlock(ubi, vol_id, lnum);
798 ubi_free_vid_hdr(ubi, vid_hdr);
802 if (err != -EIO || !ubi->bad_allowed) {
804 * This flash device does not admit of bad eraseblocks or
805 * something nasty and unexpected happened. Switch to read-only
809 leb_write_unlock(ubi, vol_id, lnum);
810 ubi_free_vid_hdr(ubi, vid_hdr);
814 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
815 if (err || ++tries > UBI_IO_RETRIES) {
817 leb_write_unlock(ubi, vol_id, lnum);
818 ubi_free_vid_hdr(ubi, vid_hdr);
822 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
823 ubi_msg("try another PEB");
828 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
829 * @ubi: UBI device description object
830 * @vol: volume description object
831 * @lnum: logical eraseblock number
832 * @buf: data to write
833 * @len: how many bytes to write
835 * This function changes the contents of a logical eraseblock atomically. @buf
836 * has to contain new logical eraseblock data, and @len - the length of the
837 * data, which has to be aligned. This function guarantees that in case of an
838 * unclean reboot the old contents is preserved. Returns zero in case of
839 * success and a negative error code in case of failure.
841 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
842 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
844 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
845 int lnum, const void *buf, int len)
847 int err, pnum, tries = 0, vol_id = vol->vol_id;
848 struct ubi_vid_hdr *vid_hdr;
856 * Special case when data length is zero. In this case the LEB
857 * has to be unmapped and mapped somewhere else.
859 err = ubi_eba_unmap_leb(ubi, vol, lnum);
862 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
865 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
869 mutex_lock(&ubi->alc_mutex);
870 err = leb_write_lock(ubi, vol_id, lnum);
874 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
875 vid_hdr->vol_id = cpu_to_be32(vol_id);
876 vid_hdr->lnum = cpu_to_be32(lnum);
877 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
878 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
880 crc = crc32(UBI_CRC32_INIT, buf, len);
881 vid_hdr->vol_type = UBI_VID_DYNAMIC;
882 vid_hdr->data_size = cpu_to_be32(len);
883 vid_hdr->copy_flag = 1;
884 vid_hdr->data_crc = cpu_to_be32(crc);
887 pnum = ubi_wl_get_peb(ubi);
893 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
894 vol_id, lnum, vol->eba_tbl[lnum], pnum);
896 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
898 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
903 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
905 ubi_warn("failed to write %d bytes of data to PEB %d",
910 if (vol->eba_tbl[lnum] >= 0) {
911 err = ubi_wl_put_peb(ubi, vol_id, lnum, vol->eba_tbl[lnum], 0);
916 vol->eba_tbl[lnum] = pnum;
919 leb_write_unlock(ubi, vol_id, lnum);
921 mutex_unlock(&ubi->alc_mutex);
922 ubi_free_vid_hdr(ubi, vid_hdr);
926 if (err != -EIO || !ubi->bad_allowed) {
928 * This flash device does not admit of bad eraseblocks or
929 * something nasty and unexpected happened. Switch to read-only
936 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
937 if (err || ++tries > UBI_IO_RETRIES) {
942 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
943 ubi_msg("try another PEB");
948 * is_error_sane - check whether a read error is sane.
949 * @err: code of the error happened during reading
951 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
952 * cannot read data from the target PEB (an error @err happened). If the error
953 * code is sane, then we treat this error as non-fatal. Otherwise the error is
954 * fatal and UBI will be switched to R/O mode later.
956 * The idea is that we try not to switch to R/O mode if the read error is
957 * something which suggests there was a real read problem. E.g., %-EIO. Or a
958 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
959 * mode, simply because we do not know what happened at the MTD level, and we
960 * cannot handle this. E.g., the underlying driver may have become crazy, and
961 * it is safer to switch to R/O mode to preserve the data.
963 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
964 * which we have just written.
966 static int is_error_sane(int err)
968 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
969 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
975 * ubi_eba_copy_leb - copy logical eraseblock.
976 * @ubi: UBI device description object
977 * @from: physical eraseblock number from where to copy
978 * @to: physical eraseblock number where to copy
979 * @vid_hdr: VID header of the @from physical eraseblock
981 * This function copies logical eraseblock from physical eraseblock @from to
982 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
984 * o %0 in case of success;
985 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
986 * o a negative error code in case of failure.
988 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
989 struct ubi_vid_hdr *vid_hdr)
991 int err, vol_id, lnum, data_size, aldata_size, idx;
992 struct ubi_volume *vol;
995 vol_id = be32_to_cpu(vid_hdr->vol_id);
996 lnum = be32_to_cpu(vid_hdr->lnum);
998 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1000 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1001 data_size = be32_to_cpu(vid_hdr->data_size);
1002 aldata_size = ALIGN(data_size, ubi->min_io_size);
1004 data_size = aldata_size =
1005 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1007 idx = vol_id2idx(ubi, vol_id);
1008 spin_lock(&ubi->volumes_lock);
1010 * Note, we may race with volume deletion, which means that the volume
1011 * this logical eraseblock belongs to might be being deleted. Since the
1012 * volume deletion un-maps all the volume's logical eraseblocks, it will
1013 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1015 vol = ubi->volumes[idx];
1016 spin_unlock(&ubi->volumes_lock);
1018 /* No need to do further work, cancel */
1019 dbg_wl("volume %d is being removed, cancel", vol_id);
1020 return MOVE_CANCEL_RACE;
1024 * We do not want anybody to write to this logical eraseblock while we
1025 * are moving it, so lock it.
1027 * Note, we are using non-waiting locking here, because we cannot sleep
1028 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1029 * unmapping the LEB which is mapped to the PEB we are going to move
1030 * (@from). This task locks the LEB and goes sleep in the
1031 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1032 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1033 * LEB is already locked, we just do not move it and return
1034 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1035 * we do not know the reasons of the contention - it may be just a
1036 * normal I/O on this LEB, so we want to re-try.
1038 err = leb_write_trylock(ubi, vol_id, lnum);
1040 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1045 * The LEB might have been put meanwhile, and the task which put it is
1046 * probably waiting on @ubi->move_mutex. No need to continue the work,
1049 if (vol->eba_tbl[lnum] != from) {
1050 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1051 vol_id, lnum, from, vol->eba_tbl[lnum]);
1052 err = MOVE_CANCEL_RACE;
1053 goto out_unlock_leb;
1057 * OK, now the LEB is locked and we can safely start moving it. Since
1058 * this function utilizes the @ubi->peb_buf buffer which is shared
1059 * with some other functions - we lock the buffer by taking the
1062 mutex_lock(&ubi->buf_mutex);
1063 dbg_wl("read %d bytes of data", aldata_size);
1064 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1065 if (err && err != UBI_IO_BITFLIPS) {
1066 ubi_warn("error %d while reading data from PEB %d",
1068 err = MOVE_SOURCE_RD_ERR;
1069 goto out_unlock_buf;
1073 * Now we have got to calculate how much data we have to copy. In
1074 * case of a static volume it is fairly easy - the VID header contains
1075 * the data size. In case of a dynamic volume it is more difficult - we
1076 * have to read the contents, cut 0xFF bytes from the end and copy only
1077 * the first part. We must do this to avoid writing 0xFF bytes as it
1078 * may have some side-effects. And not only this. It is important not
1079 * to include those 0xFFs to CRC because later the they may be filled
1082 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1083 aldata_size = data_size =
1084 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1087 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1091 * It may turn out to be that the whole @from physical eraseblock
1092 * contains only 0xFF bytes. Then we have to only write the VID header
1093 * and do not write any data. This also means we should not set
1094 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1096 if (data_size > 0) {
1097 vid_hdr->copy_flag = 1;
1098 vid_hdr->data_size = cpu_to_be32(data_size);
1099 vid_hdr->data_crc = cpu_to_be32(crc);
1101 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
1103 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1106 err = MOVE_TARGET_WR_ERR;
1107 goto out_unlock_buf;
1112 /* Read the VID header back and check if it was written correctly */
1113 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1115 if (err != UBI_IO_BITFLIPS) {
1116 ubi_warn("error %d while reading VID header back from PEB %d",
1118 if (is_error_sane(err))
1119 err = MOVE_TARGET_RD_ERR;
1121 err = MOVE_TARGET_BITFLIPS;
1122 goto out_unlock_buf;
1125 if (data_size > 0) {
1126 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1129 err = MOVE_TARGET_WR_ERR;
1130 goto out_unlock_buf;
1136 * We've written the data and are going to read it back to make
1137 * sure it was written correctly.
1139 memset(ubi->peb_buf, 0xFF, aldata_size);
1140 err = ubi_io_read_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1142 if (err != UBI_IO_BITFLIPS) {
1143 ubi_warn("error %d while reading data back from PEB %d",
1145 if (is_error_sane(err))
1146 err = MOVE_TARGET_RD_ERR;
1148 err = MOVE_TARGET_BITFLIPS;
1149 goto out_unlock_buf;
1154 if (crc != crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size)) {
1155 ubi_warn("read data back from PEB %d and it is different",
1158 goto out_unlock_buf;
1162 ubi_assert(vol->eba_tbl[lnum] == from);
1163 vol->eba_tbl[lnum] = to;
1166 mutex_unlock(&ubi->buf_mutex);
1168 leb_write_unlock(ubi, vol_id, lnum);
1173 * print_rsvd_warning - warn about not having enough reserved PEBs.
1174 * @ubi: UBI device description object
1176 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1177 * cannot reserve enough PEBs for bad block handling. This function makes a
1178 * decision whether we have to print a warning or not. The algorithm is as
1180 * o if this is a new UBI image, then just print the warning
1181 * o if this is an UBI image which has already been used for some time, print
1182 * a warning only if we can reserve less than 10% of the expected amount of
1185 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1186 * of PEBs becomes smaller, which is normal and we do not want to scare users
1187 * with a warning every time they attach the MTD device. This was an issue
1188 * reported by real users.
1190 static void print_rsvd_warning(struct ubi_device *ubi,
1191 struct ubi_attach_info *ai)
1194 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1195 * large number to distinguish between newly flashed and used images.
1197 if (ai->max_sqnum > (1 << 18)) {
1198 int min = ubi->beb_rsvd_level / 10;
1202 if (ubi->beb_rsvd_pebs > min)
1206 ubi_warn("cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1207 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1208 if (ubi->corr_peb_count)
1209 ubi_warn("%d PEBs are corrupted and not used",
1210 ubi->corr_peb_count);
1214 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1215 * @ubi: UBI device description object
1216 * @ai: attaching information
1218 * This function returns zero in case of success and a negative error code in
1221 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1223 int i, j, err, num_volumes;
1224 struct ubi_ainf_volume *av;
1225 struct ubi_volume *vol;
1226 struct ubi_ainf_peb *aeb;
1229 dbg_eba("initialize EBA sub-system");
1231 spin_lock_init(&ubi->ltree_lock);
1232 mutex_init(&ubi->alc_mutex);
1233 ubi->ltree = RB_ROOT;
1235 ubi->global_sqnum = ai->max_sqnum + 1;
1236 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1238 for (i = 0; i < num_volumes; i++) {
1239 vol = ubi->volumes[i];
1245 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1247 if (!vol->eba_tbl) {
1252 for (j = 0; j < vol->reserved_pebs; j++)
1253 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1255 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1259 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1260 if (aeb->lnum >= vol->reserved_pebs)
1262 * This may happen in case of an unclean reboot
1265 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1267 vol->eba_tbl[aeb->lnum] = aeb->pnum;
1271 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1272 ubi_err("no enough physical eraseblocks (%d, need %d)",
1273 ubi->avail_pebs, EBA_RESERVED_PEBS);
1274 if (ubi->corr_peb_count)
1275 ubi_err("%d PEBs are corrupted and not used",
1276 ubi->corr_peb_count);
1280 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1281 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1283 if (ubi->bad_allowed) {
1284 ubi_calculate_reserved(ubi);
1286 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1287 /* No enough free physical eraseblocks */
1288 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1289 print_rsvd_warning(ubi, ai);
1291 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1293 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1294 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1297 dbg_eba("EBA sub-system is initialized");
1301 for (i = 0; i < num_volumes; i++) {
1302 if (!ubi->volumes[i])
1304 kfree(ubi->volumes[i]->eba_tbl);
1305 ubi->volumes[i]->eba_tbl = NULL;