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_eba_sem);
344 vol->eba_tbl[lnum] = UBI_LEB_UNMAPPED;
345 up_read(&ubi->fm_eba_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",
434 } else if (err == UBI_IO_BITFLIPS)
437 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
438 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
440 crc = be32_to_cpu(vid_hdr->data_crc);
441 ubi_free_vid_hdr(ubi, vid_hdr);
444 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
446 if (err == UBI_IO_BITFLIPS)
448 else if (mtd_is_eccerr(err)) {
449 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
453 ubi_msg(ubi, "force data checking");
462 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
464 ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
472 err = ubi_wl_scrub_peb(ubi, pnum);
474 leb_read_unlock(ubi, vol_id, lnum);
478 ubi_free_vid_hdr(ubi, vid_hdr);
480 leb_read_unlock(ubi, vol_id, lnum);
485 * ubi_eba_read_leb_sg - read data into a scatter gather list.
486 * @ubi: UBI device description object
487 * @vol: volume description object
488 * @lnum: logical eraseblock number
489 * @sgl: UBI scatter gather list to store the read data
490 * @offset: offset from where to read
491 * @len: how many bytes to read
492 * @check: data CRC check flag
494 * This function works exactly like ubi_eba_read_leb(). But instead of
495 * storing the read data into a buffer it writes to an UBI scatter gather
498 int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
499 struct ubi_sgl *sgl, int lnum, int offset, int len,
504 struct scatterlist *sg;
507 ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
508 sg = &sgl->sg[sgl->list_pos];
509 if (len < sg->length - sgl->page_pos)
512 to_read = sg->length - sgl->page_pos;
514 ret = ubi_eba_read_leb(ubi, vol, lnum,
515 sg_virt(sg) + sgl->page_pos, offset,
523 sgl->page_pos += to_read;
524 if (sgl->page_pos == sg->length) {
540 * recover_peb - recover from write failure.
541 * @ubi: UBI device description object
542 * @pnum: the physical eraseblock to recover
544 * @lnum: logical eraseblock number
545 * @buf: data which was not written because of the write failure
546 * @offset: offset of the failed write
547 * @len: how many bytes should have been written
549 * This function is called in case of a write failure and moves all good data
550 * from the potentially bad physical eraseblock to a good physical eraseblock.
551 * This function also writes the data which was not written due to the failure.
552 * Returns new physical eraseblock number in case of success, and a negative
553 * error code in case of failure.
555 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
556 const void *buf, int offset, int len)
558 int err, idx = vol_id2idx(ubi, vol_id), new_pnum, data_size, tries = 0;
559 struct ubi_volume *vol = ubi->volumes[idx];
560 struct ubi_vid_hdr *vid_hdr;
562 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
567 new_pnum = ubi_wl_get_peb(ubi);
569 ubi_free_vid_hdr(ubi, vid_hdr);
570 up_read(&ubi->fm_eba_sem);
574 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
577 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
578 if (err && err != UBI_IO_BITFLIPS) {
581 up_read(&ubi->fm_eba_sem);
585 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
586 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
588 up_read(&ubi->fm_eba_sem);
592 data_size = offset + len;
593 mutex_lock(&ubi->buf_mutex);
594 memset(ubi->peb_buf + offset, 0xFF, len);
596 /* Read everything before the area where the write failure happened */
598 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
599 if (err && err != UBI_IO_BITFLIPS) {
600 up_read(&ubi->fm_eba_sem);
605 memcpy(ubi->peb_buf + offset, buf, len);
607 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
609 mutex_unlock(&ubi->buf_mutex);
610 up_read(&ubi->fm_eba_sem);
614 mutex_unlock(&ubi->buf_mutex);
615 ubi_free_vid_hdr(ubi, vid_hdr);
617 vol->eba_tbl[lnum] = new_pnum;
618 up_read(&ubi->fm_eba_sem);
619 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
621 ubi_msg(ubi, "data was successfully recovered");
625 mutex_unlock(&ubi->buf_mutex);
627 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
628 ubi_free_vid_hdr(ubi, vid_hdr);
633 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
636 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
637 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
638 if (++tries > UBI_IO_RETRIES) {
639 ubi_free_vid_hdr(ubi, vid_hdr);
642 ubi_msg(ubi, "try again");
647 * ubi_eba_write_leb - write data to dynamic volume.
648 * @ubi: UBI device description object
649 * @vol: volume description object
650 * @lnum: logical eraseblock number
651 * @buf: the data to write
652 * @offset: offset within the logical eraseblock where to write
653 * @len: how many bytes to write
655 * This function writes data to logical eraseblock @lnum of a dynamic volume
656 * @vol. Returns zero in case of success and a negative error code in case
657 * of failure. In case of error, it is possible that something was still
658 * written to the flash media, but may be some garbage.
660 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
661 const void *buf, int offset, int len)
663 int err, pnum, tries = 0, vol_id = vol->vol_id;
664 struct ubi_vid_hdr *vid_hdr;
669 err = leb_write_lock(ubi, vol_id, lnum);
673 pnum = vol->eba_tbl[lnum];
675 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
676 len, offset, vol_id, lnum, pnum);
678 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
680 ubi_warn(ubi, "failed to write data to PEB %d", pnum);
681 if (err == -EIO && ubi->bad_allowed)
682 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
687 leb_write_unlock(ubi, vol_id, lnum);
692 * The logical eraseblock is not mapped. We have to get a free physical
693 * eraseblock and write the volume identifier header there first.
695 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
697 leb_write_unlock(ubi, vol_id, lnum);
701 vid_hdr->vol_type = UBI_VID_DYNAMIC;
702 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
703 vid_hdr->vol_id = cpu_to_be32(vol_id);
704 vid_hdr->lnum = cpu_to_be32(lnum);
705 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
706 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
709 pnum = ubi_wl_get_peb(ubi);
711 ubi_free_vid_hdr(ubi, vid_hdr);
712 leb_write_unlock(ubi, vol_id, lnum);
713 up_read(&ubi->fm_eba_sem);
717 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
718 len, offset, vol_id, lnum, pnum);
720 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
722 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
724 up_read(&ubi->fm_eba_sem);
729 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
731 ubi_warn(ubi, "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
732 len, offset, vol_id, lnum, pnum);
733 up_read(&ubi->fm_eba_sem);
738 vol->eba_tbl[lnum] = pnum;
739 up_read(&ubi->fm_eba_sem);
741 leb_write_unlock(ubi, vol_id, lnum);
742 ubi_free_vid_hdr(ubi, vid_hdr);
746 if (err != -EIO || !ubi->bad_allowed) {
748 leb_write_unlock(ubi, vol_id, lnum);
749 ubi_free_vid_hdr(ubi, vid_hdr);
754 * Fortunately, this is the first write operation to this physical
755 * eraseblock, so just put it and request a new one. We assume that if
756 * this physical eraseblock went bad, the erase code will handle that.
758 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
759 if (err || ++tries > UBI_IO_RETRIES) {
761 leb_write_unlock(ubi, vol_id, lnum);
762 ubi_free_vid_hdr(ubi, vid_hdr);
766 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
767 ubi_msg(ubi, "try another PEB");
772 * ubi_eba_write_leb_st - write data to static volume.
773 * @ubi: UBI device description object
774 * @vol: volume description object
775 * @lnum: logical eraseblock number
776 * @buf: data to write
777 * @len: how many bytes to write
778 * @used_ebs: how many logical eraseblocks will this volume contain
780 * This function writes data to logical eraseblock @lnum of static volume
781 * @vol. The @used_ebs argument should contain total number of logical
782 * eraseblock in this static volume.
784 * When writing to the last logical eraseblock, the @len argument doesn't have
785 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
786 * to the real data size, although the @buf buffer has to contain the
787 * alignment. In all other cases, @len has to be aligned.
789 * It is prohibited to write more than once to logical eraseblocks of static
790 * volumes. This function returns zero in case of success and a negative error
791 * code in case of failure.
793 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
794 int lnum, const void *buf, int len, int used_ebs)
796 int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
797 struct ubi_vid_hdr *vid_hdr;
803 if (lnum == used_ebs - 1)
804 /* If this is the last LEB @len may be unaligned */
805 len = ALIGN(data_size, ubi->min_io_size);
807 ubi_assert(!(len & (ubi->min_io_size - 1)));
809 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
813 err = leb_write_lock(ubi, vol_id, lnum);
815 ubi_free_vid_hdr(ubi, vid_hdr);
819 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
820 vid_hdr->vol_id = cpu_to_be32(vol_id);
821 vid_hdr->lnum = cpu_to_be32(lnum);
822 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
823 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
825 crc = crc32(UBI_CRC32_INIT, buf, data_size);
826 vid_hdr->vol_type = UBI_VID_STATIC;
827 vid_hdr->data_size = cpu_to_be32(data_size);
828 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
829 vid_hdr->data_crc = cpu_to_be32(crc);
832 pnum = ubi_wl_get_peb(ubi);
834 ubi_free_vid_hdr(ubi, vid_hdr);
835 leb_write_unlock(ubi, vol_id, lnum);
836 up_read(&ubi->fm_eba_sem);
840 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
841 len, vol_id, lnum, pnum, used_ebs);
843 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
845 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
847 up_read(&ubi->fm_eba_sem);
851 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
853 ubi_warn(ubi, "failed to write %d bytes of data to PEB %d",
855 up_read(&ubi->fm_eba_sem);
859 ubi_assert(vol->eba_tbl[lnum] < 0);
860 vol->eba_tbl[lnum] = pnum;
861 up_read(&ubi->fm_eba_sem);
863 leb_write_unlock(ubi, vol_id, lnum);
864 ubi_free_vid_hdr(ubi, vid_hdr);
868 if (err != -EIO || !ubi->bad_allowed) {
870 * This flash device does not admit of bad eraseblocks or
871 * something nasty and unexpected happened. Switch to read-only
875 leb_write_unlock(ubi, vol_id, lnum);
876 ubi_free_vid_hdr(ubi, vid_hdr);
880 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
881 if (err || ++tries > UBI_IO_RETRIES) {
883 leb_write_unlock(ubi, vol_id, lnum);
884 ubi_free_vid_hdr(ubi, vid_hdr);
888 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
889 ubi_msg(ubi, "try another PEB");
894 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
895 * @ubi: UBI device description object
896 * @vol: volume description object
897 * @lnum: logical eraseblock number
898 * @buf: data to write
899 * @len: how many bytes to write
901 * This function changes the contents of a logical eraseblock atomically. @buf
902 * has to contain new logical eraseblock data, and @len - the length of the
903 * data, which has to be aligned. This function guarantees that in case of an
904 * unclean reboot the old contents is preserved. Returns zero in case of
905 * success and a negative error code in case of failure.
907 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
908 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
910 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
911 int lnum, const void *buf, int len)
913 int err, pnum, old_pnum, tries = 0, vol_id = vol->vol_id;
914 struct ubi_vid_hdr *vid_hdr;
922 * Special case when data length is zero. In this case the LEB
923 * has to be unmapped and mapped somewhere else.
925 err = ubi_eba_unmap_leb(ubi, vol, lnum);
928 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
931 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
935 mutex_lock(&ubi->alc_mutex);
936 err = leb_write_lock(ubi, vol_id, lnum);
940 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
941 vid_hdr->vol_id = cpu_to_be32(vol_id);
942 vid_hdr->lnum = cpu_to_be32(lnum);
943 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
944 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
946 crc = crc32(UBI_CRC32_INIT, buf, len);
947 vid_hdr->vol_type = UBI_VID_DYNAMIC;
948 vid_hdr->data_size = cpu_to_be32(len);
949 vid_hdr->copy_flag = 1;
950 vid_hdr->data_crc = cpu_to_be32(crc);
953 pnum = ubi_wl_get_peb(ubi);
956 up_read(&ubi->fm_eba_sem);
960 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
961 vol_id, lnum, vol->eba_tbl[lnum], pnum);
963 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
965 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
967 up_read(&ubi->fm_eba_sem);
971 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
973 ubi_warn(ubi, "failed to write %d bytes of data to PEB %d",
975 up_read(&ubi->fm_eba_sem);
979 old_pnum = vol->eba_tbl[lnum];
980 vol->eba_tbl[lnum] = pnum;
981 up_read(&ubi->fm_eba_sem);
984 err = ubi_wl_put_peb(ubi, vol_id, lnum, old_pnum, 0);
990 leb_write_unlock(ubi, vol_id, lnum);
992 mutex_unlock(&ubi->alc_mutex);
993 ubi_free_vid_hdr(ubi, vid_hdr);
997 if (err != -EIO || !ubi->bad_allowed) {
999 * This flash device does not admit of bad eraseblocks or
1000 * something nasty and unexpected happened. Switch to read-only
1001 * mode just in case.
1004 goto out_leb_unlock;
1007 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
1008 if (err || ++tries > UBI_IO_RETRIES) {
1010 goto out_leb_unlock;
1013 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1014 ubi_msg(ubi, "try another PEB");
1019 * is_error_sane - check whether a read error is sane.
1020 * @err: code of the error happened during reading
1022 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1023 * cannot read data from the target PEB (an error @err happened). If the error
1024 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1025 * fatal and UBI will be switched to R/O mode later.
1027 * The idea is that we try not to switch to R/O mode if the read error is
1028 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1029 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1030 * mode, simply because we do not know what happened at the MTD level, and we
1031 * cannot handle this. E.g., the underlying driver may have become crazy, and
1032 * it is safer to switch to R/O mode to preserve the data.
1034 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1035 * which we have just written.
1037 static int is_error_sane(int err)
1039 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1040 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1046 * ubi_eba_copy_leb - copy logical eraseblock.
1047 * @ubi: UBI device description object
1048 * @from: physical eraseblock number from where to copy
1049 * @to: physical eraseblock number where to copy
1050 * @vid_hdr: VID header of the @from physical eraseblock
1052 * This function copies logical eraseblock from physical eraseblock @from to
1053 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1054 * function. Returns:
1055 * o %0 in case of success;
1056 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1057 * o a negative error code in case of failure.
1059 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1060 struct ubi_vid_hdr *vid_hdr)
1062 int err, vol_id, lnum, data_size, aldata_size, idx;
1063 struct ubi_volume *vol;
1066 vol_id = be32_to_cpu(vid_hdr->vol_id);
1067 lnum = be32_to_cpu(vid_hdr->lnum);
1069 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1071 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1072 data_size = be32_to_cpu(vid_hdr->data_size);
1073 aldata_size = ALIGN(data_size, ubi->min_io_size);
1075 data_size = aldata_size =
1076 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1078 idx = vol_id2idx(ubi, vol_id);
1079 spin_lock(&ubi->volumes_lock);
1081 * Note, we may race with volume deletion, which means that the volume
1082 * this logical eraseblock belongs to might be being deleted. Since the
1083 * volume deletion un-maps all the volume's logical eraseblocks, it will
1084 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1086 vol = ubi->volumes[idx];
1087 spin_unlock(&ubi->volumes_lock);
1089 /* No need to do further work, cancel */
1090 dbg_wl("volume %d is being removed, cancel", vol_id);
1091 return MOVE_CANCEL_RACE;
1095 * We do not want anybody to write to this logical eraseblock while we
1096 * are moving it, so lock it.
1098 * Note, we are using non-waiting locking here, because we cannot sleep
1099 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1100 * unmapping the LEB which is mapped to the PEB we are going to move
1101 * (@from). This task locks the LEB and goes sleep in the
1102 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1103 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1104 * LEB is already locked, we just do not move it and return
1105 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1106 * we do not know the reasons of the contention - it may be just a
1107 * normal I/O on this LEB, so we want to re-try.
1109 err = leb_write_trylock(ubi, vol_id, lnum);
1111 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1116 * The LEB might have been put meanwhile, and the task which put it is
1117 * probably waiting on @ubi->move_mutex. No need to continue the work,
1120 if (vol->eba_tbl[lnum] != from) {
1121 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1122 vol_id, lnum, from, vol->eba_tbl[lnum]);
1123 err = MOVE_CANCEL_RACE;
1124 goto out_unlock_leb;
1128 * OK, now the LEB is locked and we can safely start moving it. Since
1129 * this function utilizes the @ubi->peb_buf buffer which is shared
1130 * with some other functions - we lock the buffer by taking the
1133 mutex_lock(&ubi->buf_mutex);
1134 dbg_wl("read %d bytes of data", aldata_size);
1135 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1136 if (err && err != UBI_IO_BITFLIPS) {
1137 ubi_warn(ubi, "error %d while reading data from PEB %d",
1139 err = MOVE_SOURCE_RD_ERR;
1140 goto out_unlock_buf;
1144 * Now we have got to calculate how much data we have to copy. In
1145 * case of a static volume it is fairly easy - the VID header contains
1146 * the data size. In case of a dynamic volume it is more difficult - we
1147 * have to read the contents, cut 0xFF bytes from the end and copy only
1148 * the first part. We must do this to avoid writing 0xFF bytes as it
1149 * may have some side-effects. And not only this. It is important not
1150 * to include those 0xFFs to CRC because later the they may be filled
1153 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1154 aldata_size = data_size =
1155 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1158 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1162 * It may turn out to be that the whole @from physical eraseblock
1163 * contains only 0xFF bytes. Then we have to only write the VID header
1164 * and do not write any data. This also means we should not set
1165 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1167 if (data_size > 0) {
1168 vid_hdr->copy_flag = 1;
1169 vid_hdr->data_size = cpu_to_be32(data_size);
1170 vid_hdr->data_crc = cpu_to_be32(crc);
1172 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1174 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1177 err = MOVE_TARGET_WR_ERR;
1178 goto out_unlock_buf;
1183 /* Read the VID header back and check if it was written correctly */
1184 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1186 if (err != UBI_IO_BITFLIPS) {
1187 ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1189 if (is_error_sane(err))
1190 err = MOVE_TARGET_RD_ERR;
1192 err = MOVE_TARGET_BITFLIPS;
1193 goto out_unlock_buf;
1196 if (data_size > 0) {
1197 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1200 err = MOVE_TARGET_WR_ERR;
1201 goto out_unlock_buf;
1207 * We've written the data and are going to read it back to make
1208 * sure it was written correctly.
1210 memset(ubi->peb_buf, 0xFF, aldata_size);
1211 err = ubi_io_read_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1213 if (err != UBI_IO_BITFLIPS) {
1214 ubi_warn(ubi, "error %d while reading data back from PEB %d",
1216 if (is_error_sane(err))
1217 err = MOVE_TARGET_RD_ERR;
1219 err = MOVE_TARGET_BITFLIPS;
1220 goto out_unlock_buf;
1225 if (crc != crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size)) {
1226 ubi_warn(ubi, "read data back from PEB %d and it is different",
1229 goto out_unlock_buf;
1233 ubi_assert(vol->eba_tbl[lnum] == from);
1234 down_read(&ubi->fm_eba_sem);
1235 vol->eba_tbl[lnum] = to;
1236 up_read(&ubi->fm_eba_sem);
1239 mutex_unlock(&ubi->buf_mutex);
1241 leb_write_unlock(ubi, vol_id, lnum);
1246 * print_rsvd_warning - warn about not having enough reserved PEBs.
1247 * @ubi: UBI device description object
1249 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1250 * cannot reserve enough PEBs for bad block handling. This function makes a
1251 * decision whether we have to print a warning or not. The algorithm is as
1253 * o if this is a new UBI image, then just print the warning
1254 * o if this is an UBI image which has already been used for some time, print
1255 * a warning only if we can reserve less than 10% of the expected amount of
1258 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1259 * of PEBs becomes smaller, which is normal and we do not want to scare users
1260 * with a warning every time they attach the MTD device. This was an issue
1261 * reported by real users.
1263 static void print_rsvd_warning(struct ubi_device *ubi,
1264 struct ubi_attach_info *ai)
1267 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1268 * large number to distinguish between newly flashed and used images.
1270 if (ai->max_sqnum > (1 << 18)) {
1271 int min = ubi->beb_rsvd_level / 10;
1275 if (ubi->beb_rsvd_pebs > min)
1279 ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1280 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1281 if (ubi->corr_peb_count)
1282 ubi_warn(ubi, "%d PEBs are corrupted and not used",
1283 ubi->corr_peb_count);
1287 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1288 * @ubi: UBI device description object
1289 * @ai_fastmap: UBI attach info object created by fastmap
1290 * @ai_scan: UBI attach info object created by scanning
1292 * Returns < 0 in case of an internal error, 0 otherwise.
1293 * If a bad EBA table entry was found it will be printed out and
1294 * ubi_assert() triggers.
1296 int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1297 struct ubi_attach_info *ai_scan)
1299 int i, j, num_volumes, ret = 0;
1300 int **scan_eba, **fm_eba;
1301 struct ubi_ainf_volume *av;
1302 struct ubi_volume *vol;
1303 struct ubi_ainf_peb *aeb;
1306 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1308 scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1312 fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1318 for (i = 0; i < num_volumes; i++) {
1319 vol = ubi->volumes[i];
1323 scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1330 fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1337 for (j = 0; j < vol->reserved_pebs; j++)
1338 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1340 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1344 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1345 scan_eba[i][aeb->lnum] = aeb->pnum;
1347 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1351 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1352 fm_eba[i][aeb->lnum] = aeb->pnum;
1354 for (j = 0; j < vol->reserved_pebs; j++) {
1355 if (scan_eba[i][j] != fm_eba[i][j]) {
1356 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1357 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1360 ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1361 vol->vol_id, i, fm_eba[i][j],
1369 for (i = 0; i < num_volumes; i++) {
1370 if (!ubi->volumes[i])
1383 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1384 * @ubi: UBI device description object
1385 * @ai: attaching information
1387 * This function returns zero in case of success and a negative error code in
1390 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1392 int i, j, err, num_volumes;
1393 struct ubi_ainf_volume *av;
1394 struct ubi_volume *vol;
1395 struct ubi_ainf_peb *aeb;
1398 dbg_eba("initialize EBA sub-system");
1400 spin_lock_init(&ubi->ltree_lock);
1401 mutex_init(&ubi->alc_mutex);
1402 ubi->ltree = RB_ROOT;
1404 ubi->global_sqnum = ai->max_sqnum + 1;
1405 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1407 for (i = 0; i < num_volumes; i++) {
1408 vol = ubi->volumes[i];
1414 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1416 if (!vol->eba_tbl) {
1421 for (j = 0; j < vol->reserved_pebs; j++)
1422 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1424 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1428 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1429 if (aeb->lnum >= vol->reserved_pebs)
1431 * This may happen in case of an unclean reboot
1434 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1436 vol->eba_tbl[aeb->lnum] = aeb->pnum;
1440 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1441 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1442 ubi->avail_pebs, EBA_RESERVED_PEBS);
1443 if (ubi->corr_peb_count)
1444 ubi_err(ubi, "%d PEBs are corrupted and not used",
1445 ubi->corr_peb_count);
1449 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1450 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1452 if (ubi->bad_allowed) {
1453 ubi_calculate_reserved(ubi);
1455 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1456 /* No enough free physical eraseblocks */
1457 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1458 print_rsvd_warning(ubi, ai);
1460 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1462 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1463 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1466 dbg_eba("EBA sub-system is initialized");
1470 for (i = 0; i < num_volumes; i++) {
1471 if (!ubi->volumes[i])
1473 kfree(ubi->volumes[i]->eba_tbl);
1474 ubi->volumes[i]->eba_tbl = NULL;