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, 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 "
424 "%d, LEB %d:%d", pnum, vol_id,
431 } else if (err == UBI_IO_BITFLIPS)
434 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
435 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
437 crc = be32_to_cpu(vid_hdr->data_crc);
438 ubi_free_vid_hdr(ubi, vid_hdr);
441 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
443 if (err == UBI_IO_BITFLIPS) {
446 } else if (mtd_is_eccerr(err)) {
447 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
451 ubi_msg("force data checking");
460 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
462 ubi_warn("CRC error: calculated %#08x, must be %#08x",
470 err = ubi_wl_scrub_peb(ubi, pnum);
472 leb_read_unlock(ubi, vol_id, lnum);
476 ubi_free_vid_hdr(ubi, vid_hdr);
478 leb_read_unlock(ubi, vol_id, lnum);
483 * recover_peb - recover from write failure.
484 * @ubi: UBI device description object
485 * @pnum: the physical eraseblock to recover
487 * @lnum: logical eraseblock number
488 * @buf: data which was not written because of the write failure
489 * @offset: offset of the failed write
490 * @len: how many bytes should have been written
492 * This function is called in case of a write failure and moves all good data
493 * from the potentially bad physical eraseblock to a good physical eraseblock.
494 * This function also writes the data which was not written due to the failure.
495 * Returns new physical eraseblock number in case of success, and a negative
496 * error code in case of failure.
498 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
499 const void *buf, int offset, int len)
501 int err, idx = vol_id2idx(ubi, vol_id), new_pnum, data_size, tries = 0;
502 struct ubi_volume *vol = ubi->volumes[idx];
503 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 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
526 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
530 data_size = offset + len;
531 mutex_lock(&ubi->buf_mutex);
532 memset(ubi->peb_buf + offset, 0xFF, len);
534 /* Read everything before the area where the write failure happened */
536 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
537 if (err && err != UBI_IO_BITFLIPS)
541 memcpy(ubi->peb_buf + offset, buf, len);
543 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
545 mutex_unlock(&ubi->buf_mutex);
549 mutex_unlock(&ubi->buf_mutex);
550 ubi_free_vid_hdr(ubi, vid_hdr);
552 vol->eba_tbl[lnum] = new_pnum;
553 ubi_wl_put_peb(ubi, pnum, 1);
555 ubi_msg("data was successfully recovered");
559 mutex_unlock(&ubi->buf_mutex);
561 ubi_wl_put_peb(ubi, new_pnum, 1);
562 ubi_free_vid_hdr(ubi, vid_hdr);
567 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
570 ubi_warn("failed to write to PEB %d", new_pnum);
571 ubi_wl_put_peb(ubi, new_pnum, 1);
572 if (++tries > UBI_IO_RETRIES) {
573 ubi_free_vid_hdr(ubi, vid_hdr);
576 ubi_msg("try again");
581 * ubi_eba_write_leb - write data to dynamic volume.
582 * @ubi: UBI device description object
583 * @vol: volume description object
584 * @lnum: logical eraseblock number
585 * @buf: the data to write
586 * @offset: offset within the logical eraseblock where to write
587 * @len: how many bytes to write
589 * This function writes data to logical eraseblock @lnum of a dynamic volume
590 * @vol. Returns zero in case of success and a negative error code in case
591 * of failure. In case of error, it is possible that something was still
592 * written to the flash media, but may be some garbage.
594 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
595 const void *buf, int offset, int len)
597 int err, pnum, tries = 0, vol_id = vol->vol_id;
598 struct ubi_vid_hdr *vid_hdr;
603 err = leb_write_lock(ubi, vol_id, lnum);
607 pnum = vol->eba_tbl[lnum];
609 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
610 len, offset, vol_id, lnum, pnum);
612 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
614 ubi_warn("failed to write data to PEB %d", pnum);
615 if (err == -EIO && ubi->bad_allowed)
616 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
621 leb_write_unlock(ubi, vol_id, lnum);
626 * The logical eraseblock is not mapped. We have to get a free physical
627 * eraseblock and write the volume identifier header there first.
629 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
631 leb_write_unlock(ubi, vol_id, lnum);
635 vid_hdr->vol_type = UBI_VID_DYNAMIC;
636 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
637 vid_hdr->vol_id = cpu_to_be32(vol_id);
638 vid_hdr->lnum = cpu_to_be32(lnum);
639 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
640 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
643 pnum = ubi_wl_get_peb(ubi);
645 ubi_free_vid_hdr(ubi, vid_hdr);
646 leb_write_unlock(ubi, vol_id, lnum);
650 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
651 len, offset, vol_id, lnum, pnum);
653 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
655 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
661 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
663 ubi_warn("failed to write %d bytes at offset %d of "
664 "LEB %d:%d, PEB %d", len, offset, vol_id,
670 vol->eba_tbl[lnum] = pnum;
672 leb_write_unlock(ubi, vol_id, lnum);
673 ubi_free_vid_hdr(ubi, vid_hdr);
677 if (err != -EIO || !ubi->bad_allowed) {
679 leb_write_unlock(ubi, vol_id, lnum);
680 ubi_free_vid_hdr(ubi, vid_hdr);
685 * Fortunately, this is the first write operation to this physical
686 * eraseblock, so just put it and request a new one. We assume that if
687 * this physical eraseblock went bad, the erase code will handle that.
689 err = ubi_wl_put_peb(ubi, pnum, 1);
690 if (err || ++tries > UBI_IO_RETRIES) {
692 leb_write_unlock(ubi, vol_id, lnum);
693 ubi_free_vid_hdr(ubi, vid_hdr);
697 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
698 ubi_msg("try another PEB");
703 * ubi_eba_write_leb_st - write data to static volume.
704 * @ubi: UBI device description object
705 * @vol: volume description object
706 * @lnum: logical eraseblock number
707 * @buf: data to write
708 * @len: how many bytes to write
709 * @used_ebs: how many logical eraseblocks will this volume contain
711 * This function writes data to logical eraseblock @lnum of static volume
712 * @vol. The @used_ebs argument should contain total number of logical
713 * eraseblock in this static volume.
715 * When writing to the last logical eraseblock, the @len argument doesn't have
716 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
717 * to the real data size, although the @buf buffer has to contain the
718 * alignment. In all other cases, @len has to be aligned.
720 * It is prohibited to write more than once to logical eraseblocks of static
721 * volumes. This function returns zero in case of success and a negative error
722 * code in case of failure.
724 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
725 int lnum, const void *buf, int len, int used_ebs)
727 int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
728 struct ubi_vid_hdr *vid_hdr;
734 if (lnum == used_ebs - 1)
735 /* If this is the last LEB @len may be unaligned */
736 len = ALIGN(data_size, ubi->min_io_size);
738 ubi_assert(!(len & (ubi->min_io_size - 1)));
740 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
744 err = leb_write_lock(ubi, vol_id, lnum);
746 ubi_free_vid_hdr(ubi, vid_hdr);
750 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
751 vid_hdr->vol_id = cpu_to_be32(vol_id);
752 vid_hdr->lnum = cpu_to_be32(lnum);
753 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
754 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
756 crc = crc32(UBI_CRC32_INIT, buf, data_size);
757 vid_hdr->vol_type = UBI_VID_STATIC;
758 vid_hdr->data_size = cpu_to_be32(data_size);
759 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
760 vid_hdr->data_crc = cpu_to_be32(crc);
763 pnum = ubi_wl_get_peb(ubi);
765 ubi_free_vid_hdr(ubi, vid_hdr);
766 leb_write_unlock(ubi, vol_id, lnum);
770 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
771 len, vol_id, lnum, pnum, used_ebs);
773 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
775 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
780 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
782 ubi_warn("failed to write %d bytes of data to PEB %d",
787 ubi_assert(vol->eba_tbl[lnum] < 0);
788 vol->eba_tbl[lnum] = pnum;
790 leb_write_unlock(ubi, vol_id, lnum);
791 ubi_free_vid_hdr(ubi, vid_hdr);
795 if (err != -EIO || !ubi->bad_allowed) {
797 * This flash device does not admit of bad eraseblocks or
798 * something nasty and unexpected happened. Switch to read-only
802 leb_write_unlock(ubi, vol_id, lnum);
803 ubi_free_vid_hdr(ubi, vid_hdr);
807 err = ubi_wl_put_peb(ubi, pnum, 1);
808 if (err || ++tries > UBI_IO_RETRIES) {
810 leb_write_unlock(ubi, vol_id, lnum);
811 ubi_free_vid_hdr(ubi, vid_hdr);
815 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
816 ubi_msg("try another PEB");
821 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
822 * @ubi: UBI device description object
823 * @vol: volume description object
824 * @lnum: logical eraseblock number
825 * @buf: data to write
826 * @len: how many bytes to write
828 * This function changes the contents of a logical eraseblock atomically. @buf
829 * has to contain new logical eraseblock data, and @len - the length of the
830 * data, which has to be aligned. This function guarantees that in case of an
831 * unclean reboot the old contents is preserved. Returns zero in case of
832 * success and a negative error code in case of failure.
834 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
835 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
837 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
838 int lnum, const void *buf, int len)
840 int err, pnum, tries = 0, vol_id = vol->vol_id;
841 struct ubi_vid_hdr *vid_hdr;
849 * Special case when data length is zero. In this case the LEB
850 * has to be unmapped and mapped somewhere else.
852 err = ubi_eba_unmap_leb(ubi, vol, lnum);
855 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
858 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
862 mutex_lock(&ubi->alc_mutex);
863 err = leb_write_lock(ubi, vol_id, lnum);
867 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
868 vid_hdr->vol_id = cpu_to_be32(vol_id);
869 vid_hdr->lnum = cpu_to_be32(lnum);
870 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
871 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
873 crc = crc32(UBI_CRC32_INIT, buf, len);
874 vid_hdr->vol_type = UBI_VID_DYNAMIC;
875 vid_hdr->data_size = cpu_to_be32(len);
876 vid_hdr->copy_flag = 1;
877 vid_hdr->data_crc = cpu_to_be32(crc);
880 pnum = ubi_wl_get_peb(ubi);
886 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
887 vol_id, lnum, vol->eba_tbl[lnum], pnum);
889 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
891 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
896 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
898 ubi_warn("failed to write %d bytes of data to PEB %d",
903 if (vol->eba_tbl[lnum] >= 0) {
904 err = ubi_wl_put_peb(ubi, vol->eba_tbl[lnum], 0);
909 vol->eba_tbl[lnum] = pnum;
912 leb_write_unlock(ubi, vol_id, lnum);
914 mutex_unlock(&ubi->alc_mutex);
915 ubi_free_vid_hdr(ubi, vid_hdr);
919 if (err != -EIO || !ubi->bad_allowed) {
921 * This flash device does not admit of bad eraseblocks or
922 * something nasty and unexpected happened. Switch to read-only
929 err = ubi_wl_put_peb(ubi, pnum, 1);
930 if (err || ++tries > UBI_IO_RETRIES) {
935 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
936 ubi_msg("try another PEB");
941 * is_error_sane - check whether a read error is sane.
942 * @err: code of the error happened during reading
944 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
945 * cannot read data from the target PEB (an error @err happened). If the error
946 * code is sane, then we treat this error as non-fatal. Otherwise the error is
947 * fatal and UBI will be switched to R/O mode later.
949 * The idea is that we try not to switch to R/O mode if the read error is
950 * something which suggests there was a real read problem. E.g., %-EIO. Or a
951 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
952 * mode, simply because we do not know what happened at the MTD level, and we
953 * cannot handle this. E.g., the underlying driver may have become crazy, and
954 * it is safer to switch to R/O mode to preserve the data.
956 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
957 * which we have just written.
959 static int is_error_sane(int err)
961 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
962 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
968 * ubi_eba_copy_leb - copy logical eraseblock.
969 * @ubi: UBI device description object
970 * @from: physical eraseblock number from where to copy
971 * @to: physical eraseblock number where to copy
972 * @vid_hdr: VID header of the @from physical eraseblock
974 * This function copies logical eraseblock from physical eraseblock @from to
975 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
977 * o %0 in case of success;
978 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
979 * o a negative error code in case of failure.
981 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
982 struct ubi_vid_hdr *vid_hdr)
984 int err, vol_id, lnum, data_size, aldata_size, idx;
985 struct ubi_volume *vol;
988 vol_id = be32_to_cpu(vid_hdr->vol_id);
989 lnum = be32_to_cpu(vid_hdr->lnum);
991 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
993 if (vid_hdr->vol_type == UBI_VID_STATIC) {
994 data_size = be32_to_cpu(vid_hdr->data_size);
995 aldata_size = ALIGN(data_size, ubi->min_io_size);
997 data_size = aldata_size =
998 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1000 idx = vol_id2idx(ubi, vol_id);
1001 spin_lock(&ubi->volumes_lock);
1003 * Note, we may race with volume deletion, which means that the volume
1004 * this logical eraseblock belongs to might be being deleted. Since the
1005 * volume deletion un-maps all the volume's logical eraseblocks, it will
1006 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1008 vol = ubi->volumes[idx];
1009 spin_unlock(&ubi->volumes_lock);
1011 /* No need to do further work, cancel */
1012 dbg_wl("volume %d is being removed, cancel", vol_id);
1013 return MOVE_CANCEL_RACE;
1017 * We do not want anybody to write to this logical eraseblock while we
1018 * are moving it, so lock it.
1020 * Note, we are using non-waiting locking here, because we cannot sleep
1021 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1022 * unmapping the LEB which is mapped to the PEB we are going to move
1023 * (@from). This task locks the LEB and goes sleep in the
1024 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1025 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1026 * LEB is already locked, we just do not move it and return
1027 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1028 * we do not know the reasons of the contention - it may be just a
1029 * normal I/O on this LEB, so we want to re-try.
1031 err = leb_write_trylock(ubi, vol_id, lnum);
1033 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1038 * The LEB might have been put meanwhile, and the task which put it is
1039 * probably waiting on @ubi->move_mutex. No need to continue the work,
1042 if (vol->eba_tbl[lnum] != from) {
1043 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to "
1044 "PEB %d, cancel", vol_id, lnum, from,
1045 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(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 "
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 "
1138 "from PEB %d", err, to);
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 "
1152 goto out_unlock_buf;
1156 ubi_assert(vol->eba_tbl[lnum] == from);
1157 vol->eba_tbl[lnum] = to;
1160 mutex_unlock(&ubi->buf_mutex);
1162 leb_write_unlock(ubi, vol_id, lnum);
1167 * print_rsvd_warning - warn about not having enough reserved PEBs.
1168 * @ubi: UBI device description object
1170 * This is a helper function for 'ubi_eba_init_scan()' which is called when UBI
1171 * cannot reserve enough PEBs for bad block handling. This function makes a
1172 * decision whether we have to print a warning or not. The algorithm is as
1174 * o if this is a new UBI image, then just print the warning
1175 * o if this is an UBI image which has already been used for some time, print
1176 * a warning only if we can reserve less than 10% of the expected amount of
1179 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1180 * of PEBs becomes smaller, which is normal and we do not want to scare users
1181 * with a warning every time they attach the MTD device. This was an issue
1182 * reported by real users.
1184 static void print_rsvd_warning(struct ubi_device *ubi,
1185 struct ubi_attach_info *si)
1188 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1189 * large number to distinguish between newly flashed and used images.
1191 if (si->max_sqnum > (1 << 18)) {
1192 int min = ubi->beb_rsvd_level / 10;
1196 if (ubi->beb_rsvd_pebs > min)
1200 ubi_warn("cannot reserve enough PEBs for bad PEB handling, reserved %d,"
1201 " need %d", ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1202 if (ubi->corr_peb_count)
1203 ubi_warn("%d PEBs are corrupted and not used",
1204 ubi->corr_peb_count);
1208 * ubi_eba_init_scan - initialize the EBA sub-system using scanning information.
1209 * @ubi: UBI device description object
1210 * @si: scanning information
1212 * This function returns zero in case of success and a negative error code in
1215 int ubi_eba_init_scan(struct ubi_device *ubi, struct ubi_attach_info *si)
1217 int i, j, err, num_volumes;
1218 struct ubi_ainf_volume *sv;
1219 struct ubi_volume *vol;
1220 struct ubi_ainf_peb *seb;
1223 dbg_eba("initialize EBA sub-system");
1225 spin_lock_init(&ubi->ltree_lock);
1226 mutex_init(&ubi->alc_mutex);
1227 ubi->ltree = RB_ROOT;
1229 ubi->global_sqnum = si->max_sqnum + 1;
1230 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1232 for (i = 0; i < num_volumes; i++) {
1233 vol = ubi->volumes[i];
1239 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1241 if (!vol->eba_tbl) {
1246 for (j = 0; j < vol->reserved_pebs; j++)
1247 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1249 sv = ubi_scan_find_sv(si, idx2vol_id(ubi, i));
1253 ubi_rb_for_each_entry(rb, seb, &sv->root, u.rb) {
1254 if (seb->lnum >= vol->reserved_pebs)
1256 * This may happen in case of an unclean reboot
1259 ubi_scan_move_to_list(sv, seb, &si->erase);
1260 vol->eba_tbl[seb->lnum] = seb->pnum;
1264 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1265 ubi_err("no enough physical eraseblocks (%d, need %d)",
1266 ubi->avail_pebs, EBA_RESERVED_PEBS);
1267 if (ubi->corr_peb_count)
1268 ubi_err("%d PEBs are corrupted and not used",
1269 ubi->corr_peb_count);
1273 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1274 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1276 if (ubi->bad_allowed) {
1277 ubi_calculate_reserved(ubi);
1279 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1280 /* No enough free physical eraseblocks */
1281 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1282 print_rsvd_warning(ubi, si);
1284 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1286 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1287 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1290 dbg_eba("EBA sub-system is initialized");
1294 for (i = 0; i < num_volumes; i++) {
1295 if (!ubi->volumes[i])
1297 kfree(ubi->volumes[i]->eba_tbl);
1298 ubi->volumes[i]->eba_tbl = NULL;