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 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;
504 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
509 new_pnum = ubi_wl_get_peb(ubi);
511 ubi_free_vid_hdr(ubi, vid_hdr);
515 ubi_msg("recover PEB %d, move data to PEB %d", pnum, new_pnum);
517 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
518 if (err && err != UBI_IO_BITFLIPS) {
524 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
525 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
529 data_size = offset + len;
530 mutex_lock(&ubi->buf_mutex);
531 memset(ubi->peb_buf + offset, 0xFF, len);
533 /* Read everything before the area where the write failure happened */
535 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
536 if (err && err != UBI_IO_BITFLIPS)
540 memcpy(ubi->peb_buf + offset, buf, len);
542 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
544 mutex_unlock(&ubi->buf_mutex);
548 mutex_unlock(&ubi->buf_mutex);
549 ubi_free_vid_hdr(ubi, vid_hdr);
551 vol->eba_tbl[lnum] = new_pnum;
552 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
554 ubi_msg("data was successfully recovered");
558 mutex_unlock(&ubi->buf_mutex);
560 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
561 ubi_free_vid_hdr(ubi, vid_hdr);
566 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
569 ubi_warn("failed to write to PEB %d", new_pnum);
570 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
571 if (++tries > UBI_IO_RETRIES) {
572 ubi_free_vid_hdr(ubi, vid_hdr);
575 ubi_msg("try again");
580 * ubi_eba_write_leb - write data to dynamic volume.
581 * @ubi: UBI device description object
582 * @vol: volume description object
583 * @lnum: logical eraseblock number
584 * @buf: the data to write
585 * @offset: offset within the logical eraseblock where to write
586 * @len: how many bytes to write
588 * This function writes data to logical eraseblock @lnum of a dynamic volume
589 * @vol. Returns zero in case of success and a negative error code in case
590 * of failure. In case of error, it is possible that something was still
591 * written to the flash media, but may be some garbage.
593 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
594 const void *buf, int offset, int len)
596 int err, pnum, tries = 0, vol_id = vol->vol_id;
597 struct ubi_vid_hdr *vid_hdr;
602 err = leb_write_lock(ubi, vol_id, lnum);
606 pnum = vol->eba_tbl[lnum];
608 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
609 len, offset, vol_id, lnum, pnum);
611 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
613 ubi_warn("failed to write data to PEB %d", pnum);
614 if (err == -EIO && ubi->bad_allowed)
615 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
620 leb_write_unlock(ubi, vol_id, lnum);
625 * The logical eraseblock is not mapped. We have to get a free physical
626 * eraseblock and write the volume identifier header there first.
628 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
630 leb_write_unlock(ubi, vol_id, lnum);
634 vid_hdr->vol_type = UBI_VID_DYNAMIC;
635 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
636 vid_hdr->vol_id = cpu_to_be32(vol_id);
637 vid_hdr->lnum = cpu_to_be32(lnum);
638 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
639 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
642 pnum = ubi_wl_get_peb(ubi);
644 ubi_free_vid_hdr(ubi, vid_hdr);
645 leb_write_unlock(ubi, vol_id, lnum);
649 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
650 len, offset, vol_id, lnum, pnum);
652 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
654 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
660 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
662 ubi_warn("failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
663 len, offset, vol_id, lnum, pnum);
668 vol->eba_tbl[lnum] = pnum;
670 leb_write_unlock(ubi, vol_id, lnum);
671 ubi_free_vid_hdr(ubi, vid_hdr);
675 if (err != -EIO || !ubi->bad_allowed) {
677 leb_write_unlock(ubi, vol_id, lnum);
678 ubi_free_vid_hdr(ubi, vid_hdr);
683 * Fortunately, this is the first write operation to this physical
684 * eraseblock, so just put it and request a new one. We assume that if
685 * this physical eraseblock went bad, the erase code will handle that.
687 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
688 if (err || ++tries > UBI_IO_RETRIES) {
690 leb_write_unlock(ubi, vol_id, lnum);
691 ubi_free_vid_hdr(ubi, vid_hdr);
695 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
696 ubi_msg("try another PEB");
701 * ubi_eba_write_leb_st - write data to static volume.
702 * @ubi: UBI device description object
703 * @vol: volume description object
704 * @lnum: logical eraseblock number
705 * @buf: data to write
706 * @len: how many bytes to write
707 * @used_ebs: how many logical eraseblocks will this volume contain
709 * This function writes data to logical eraseblock @lnum of static volume
710 * @vol. The @used_ebs argument should contain total number of logical
711 * eraseblock in this static volume.
713 * When writing to the last logical eraseblock, the @len argument doesn't have
714 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
715 * to the real data size, although the @buf buffer has to contain the
716 * alignment. In all other cases, @len has to be aligned.
718 * It is prohibited to write more than once to logical eraseblocks of static
719 * volumes. This function returns zero in case of success and a negative error
720 * code in case of failure.
722 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
723 int lnum, const void *buf, int len, int used_ebs)
725 int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
726 struct ubi_vid_hdr *vid_hdr;
732 if (lnum == used_ebs - 1)
733 /* If this is the last LEB @len may be unaligned */
734 len = ALIGN(data_size, ubi->min_io_size);
736 ubi_assert(!(len & (ubi->min_io_size - 1)));
738 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
742 err = leb_write_lock(ubi, vol_id, lnum);
744 ubi_free_vid_hdr(ubi, vid_hdr);
748 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
749 vid_hdr->vol_id = cpu_to_be32(vol_id);
750 vid_hdr->lnum = cpu_to_be32(lnum);
751 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
752 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
754 crc = crc32(UBI_CRC32_INIT, buf, data_size);
755 vid_hdr->vol_type = UBI_VID_STATIC;
756 vid_hdr->data_size = cpu_to_be32(data_size);
757 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
758 vid_hdr->data_crc = cpu_to_be32(crc);
761 pnum = ubi_wl_get_peb(ubi);
763 ubi_free_vid_hdr(ubi, vid_hdr);
764 leb_write_unlock(ubi, vol_id, lnum);
768 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
769 len, vol_id, lnum, pnum, used_ebs);
771 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
773 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
778 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
780 ubi_warn("failed to write %d bytes of data to PEB %d",
785 ubi_assert(vol->eba_tbl[lnum] < 0);
786 vol->eba_tbl[lnum] = pnum;
788 leb_write_unlock(ubi, vol_id, lnum);
789 ubi_free_vid_hdr(ubi, vid_hdr);
793 if (err != -EIO || !ubi->bad_allowed) {
795 * This flash device does not admit of bad eraseblocks or
796 * something nasty and unexpected happened. Switch to read-only
800 leb_write_unlock(ubi, vol_id, lnum);
801 ubi_free_vid_hdr(ubi, vid_hdr);
805 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
806 if (err || ++tries > UBI_IO_RETRIES) {
808 leb_write_unlock(ubi, vol_id, lnum);
809 ubi_free_vid_hdr(ubi, vid_hdr);
813 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
814 ubi_msg("try another PEB");
819 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
820 * @ubi: UBI device description object
821 * @vol: volume description object
822 * @lnum: logical eraseblock number
823 * @buf: data to write
824 * @len: how many bytes to write
826 * This function changes the contents of a logical eraseblock atomically. @buf
827 * has to contain new logical eraseblock data, and @len - the length of the
828 * data, which has to be aligned. This function guarantees that in case of an
829 * unclean reboot the old contents is preserved. Returns zero in case of
830 * success and a negative error code in case of failure.
832 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
833 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
835 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
836 int lnum, const void *buf, int len)
838 int err, pnum, tries = 0, vol_id = vol->vol_id;
839 struct ubi_vid_hdr *vid_hdr;
847 * Special case when data length is zero. In this case the LEB
848 * has to be unmapped and mapped somewhere else.
850 err = ubi_eba_unmap_leb(ubi, vol, lnum);
853 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
856 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
860 mutex_lock(&ubi->alc_mutex);
861 err = leb_write_lock(ubi, vol_id, lnum);
865 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
866 vid_hdr->vol_id = cpu_to_be32(vol_id);
867 vid_hdr->lnum = cpu_to_be32(lnum);
868 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
869 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
871 crc = crc32(UBI_CRC32_INIT, buf, len);
872 vid_hdr->vol_type = UBI_VID_DYNAMIC;
873 vid_hdr->data_size = cpu_to_be32(len);
874 vid_hdr->copy_flag = 1;
875 vid_hdr->data_crc = cpu_to_be32(crc);
878 pnum = ubi_wl_get_peb(ubi);
884 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
885 vol_id, lnum, vol->eba_tbl[lnum], pnum);
887 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
889 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
894 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
896 ubi_warn("failed to write %d bytes of data to PEB %d",
901 if (vol->eba_tbl[lnum] >= 0) {
902 err = ubi_wl_put_peb(ubi, vol_id, lnum, vol->eba_tbl[lnum], 0);
907 vol->eba_tbl[lnum] = pnum;
910 leb_write_unlock(ubi, vol_id, lnum);
912 mutex_unlock(&ubi->alc_mutex);
913 ubi_free_vid_hdr(ubi, vid_hdr);
917 if (err != -EIO || !ubi->bad_allowed) {
919 * This flash device does not admit of bad eraseblocks or
920 * something nasty and unexpected happened. Switch to read-only
927 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
928 if (err || ++tries > UBI_IO_RETRIES) {
933 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
934 ubi_msg("try another PEB");
939 * is_error_sane - check whether a read error is sane.
940 * @err: code of the error happened during reading
942 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
943 * cannot read data from the target PEB (an error @err happened). If the error
944 * code is sane, then we treat this error as non-fatal. Otherwise the error is
945 * fatal and UBI will be switched to R/O mode later.
947 * The idea is that we try not to switch to R/O mode if the read error is
948 * something which suggests there was a real read problem. E.g., %-EIO. Or a
949 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
950 * mode, simply because we do not know what happened at the MTD level, and we
951 * cannot handle this. E.g., the underlying driver may have become crazy, and
952 * it is safer to switch to R/O mode to preserve the data.
954 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
955 * which we have just written.
957 static int is_error_sane(int err)
959 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
960 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
966 * ubi_eba_copy_leb - copy logical eraseblock.
967 * @ubi: UBI device description object
968 * @from: physical eraseblock number from where to copy
969 * @to: physical eraseblock number where to copy
970 * @vid_hdr: VID header of the @from physical eraseblock
972 * This function copies logical eraseblock from physical eraseblock @from to
973 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
975 * o %0 in case of success;
976 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
977 * o a negative error code in case of failure.
979 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
980 struct ubi_vid_hdr *vid_hdr)
982 int err, vol_id, lnum, data_size, aldata_size, idx;
983 struct ubi_volume *vol;
986 vol_id = be32_to_cpu(vid_hdr->vol_id);
987 lnum = be32_to_cpu(vid_hdr->lnum);
989 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
991 if (vid_hdr->vol_type == UBI_VID_STATIC) {
992 data_size = be32_to_cpu(vid_hdr->data_size);
993 aldata_size = ALIGN(data_size, ubi->min_io_size);
995 data_size = aldata_size =
996 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
998 idx = vol_id2idx(ubi, vol_id);
999 spin_lock(&ubi->volumes_lock);
1001 * Note, we may race with volume deletion, which means that the volume
1002 * this logical eraseblock belongs to might be being deleted. Since the
1003 * volume deletion un-maps all the volume's logical eraseblocks, it will
1004 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1006 vol = ubi->volumes[idx];
1007 spin_unlock(&ubi->volumes_lock);
1009 /* No need to do further work, cancel */
1010 dbg_wl("volume %d is being removed, cancel", vol_id);
1011 return MOVE_CANCEL_RACE;
1015 * We do not want anybody to write to this logical eraseblock while we
1016 * are moving it, so lock it.
1018 * Note, we are using non-waiting locking here, because we cannot sleep
1019 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1020 * unmapping the LEB which is mapped to the PEB we are going to move
1021 * (@from). This task locks the LEB and goes sleep in the
1022 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1023 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1024 * LEB is already locked, we just do not move it and return
1025 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1026 * we do not know the reasons of the contention - it may be just a
1027 * normal I/O on this LEB, so we want to re-try.
1029 err = leb_write_trylock(ubi, vol_id, lnum);
1031 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1036 * The LEB might have been put meanwhile, and the task which put it is
1037 * probably waiting on @ubi->move_mutex. No need to continue the work,
1040 if (vol->eba_tbl[lnum] != from) {
1041 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1042 vol_id, lnum, from, vol->eba_tbl[lnum]);
1043 err = MOVE_CANCEL_RACE;
1044 goto out_unlock_leb;
1048 * OK, now the LEB is locked and we can safely start moving it. Since
1049 * this function utilizes the @ubi->peb_buf buffer which is shared
1050 * with some other functions - we lock the buffer by taking the
1053 mutex_lock(&ubi->buf_mutex);
1054 dbg_wl("read %d bytes of data", aldata_size);
1055 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1056 if (err && err != UBI_IO_BITFLIPS) {
1057 ubi_warn("error %d while reading data from PEB %d",
1059 err = MOVE_SOURCE_RD_ERR;
1060 goto out_unlock_buf;
1064 * Now we have got to calculate how much data we have to copy. In
1065 * case of a static volume it is fairly easy - the VID header contains
1066 * the data size. In case of a dynamic volume it is more difficult - we
1067 * have to read the contents, cut 0xFF bytes from the end and copy only
1068 * the first part. We must do this to avoid writing 0xFF bytes as it
1069 * may have some side-effects. And not only this. It is important not
1070 * to include those 0xFFs to CRC because later the they may be filled
1073 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1074 aldata_size = data_size =
1075 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1078 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1082 * It may turn out to be that the whole @from physical eraseblock
1083 * contains only 0xFF bytes. Then we have to only write the VID header
1084 * and do not write any data. This also means we should not set
1085 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1087 if (data_size > 0) {
1088 vid_hdr->copy_flag = 1;
1089 vid_hdr->data_size = cpu_to_be32(data_size);
1090 vid_hdr->data_crc = cpu_to_be32(crc);
1092 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1094 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1097 err = MOVE_TARGET_WR_ERR;
1098 goto out_unlock_buf;
1103 /* Read the VID header back and check if it was written correctly */
1104 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1106 if (err != UBI_IO_BITFLIPS) {
1107 ubi_warn("error %d while reading VID header back from PEB %d",
1109 if (is_error_sane(err))
1110 err = MOVE_TARGET_RD_ERR;
1112 err = MOVE_TARGET_BITFLIPS;
1113 goto out_unlock_buf;
1116 if (data_size > 0) {
1117 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1120 err = MOVE_TARGET_WR_ERR;
1121 goto out_unlock_buf;
1127 * We've written the data and are going to read it back to make
1128 * sure it was written correctly.
1130 memset(ubi->peb_buf, 0xFF, aldata_size);
1131 err = ubi_io_read_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1133 if (err != UBI_IO_BITFLIPS) {
1134 ubi_warn("error %d while reading data back from PEB %d",
1136 if (is_error_sane(err))
1137 err = MOVE_TARGET_RD_ERR;
1139 err = MOVE_TARGET_BITFLIPS;
1140 goto out_unlock_buf;
1145 if (crc != crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size)) {
1146 ubi_warn("read data back from PEB %d and it is different",
1149 goto out_unlock_buf;
1153 ubi_assert(vol->eba_tbl[lnum] == from);
1154 vol->eba_tbl[lnum] = to;
1157 mutex_unlock(&ubi->buf_mutex);
1159 leb_write_unlock(ubi, vol_id, lnum);
1164 * print_rsvd_warning - warn about not having enough reserved PEBs.
1165 * @ubi: UBI device description object
1167 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1168 * cannot reserve enough PEBs for bad block handling. This function makes a
1169 * decision whether we have to print a warning or not. The algorithm is as
1171 * o if this is a new UBI image, then just print the warning
1172 * o if this is an UBI image which has already been used for some time, print
1173 * a warning only if we can reserve less than 10% of the expected amount of
1176 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1177 * of PEBs becomes smaller, which is normal and we do not want to scare users
1178 * with a warning every time they attach the MTD device. This was an issue
1179 * reported by real users.
1181 static void print_rsvd_warning(struct ubi_device *ubi,
1182 struct ubi_attach_info *ai)
1185 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1186 * large number to distinguish between newly flashed and used images.
1188 if (ai->max_sqnum > (1 << 18)) {
1189 int min = ubi->beb_rsvd_level / 10;
1193 if (ubi->beb_rsvd_pebs > min)
1197 ubi_warn("cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1198 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1199 if (ubi->corr_peb_count)
1200 ubi_warn("%d PEBs are corrupted and not used",
1201 ubi->corr_peb_count);
1205 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1206 * @ubi: UBI device description object
1207 * @ai: attaching information
1209 * This function returns zero in case of success and a negative error code in
1212 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1214 int i, j, err, num_volumes;
1215 struct ubi_ainf_volume *av;
1216 struct ubi_volume *vol;
1217 struct ubi_ainf_peb *aeb;
1220 dbg_eba("initialize EBA sub-system");
1222 spin_lock_init(&ubi->ltree_lock);
1223 mutex_init(&ubi->alc_mutex);
1224 ubi->ltree = RB_ROOT;
1226 ubi->global_sqnum = ai->max_sqnum + 1;
1227 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1229 for (i = 0; i < num_volumes; i++) {
1230 vol = ubi->volumes[i];
1236 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1238 if (!vol->eba_tbl) {
1243 for (j = 0; j < vol->reserved_pebs; j++)
1244 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1246 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1250 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1251 if (aeb->lnum >= vol->reserved_pebs)
1253 * This may happen in case of an unclean reboot
1256 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1257 vol->eba_tbl[aeb->lnum] = aeb->pnum;
1261 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1262 ubi_err("no enough physical eraseblocks (%d, need %d)",
1263 ubi->avail_pebs, EBA_RESERVED_PEBS);
1264 if (ubi->corr_peb_count)
1265 ubi_err("%d PEBs are corrupted and not used",
1266 ubi->corr_peb_count);
1270 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1271 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1273 if (ubi->bad_allowed) {
1274 ubi_calculate_reserved(ubi);
1276 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1277 /* No enough free physical eraseblocks */
1278 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1279 print_rsvd_warning(ubi, ai);
1281 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1283 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1284 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1287 dbg_eba("EBA sub-system is initialized");
1291 for (i = 0; i < num_volumes; i++) {
1292 if (!ubi->volumes[i])
1294 kfree(ubi->volumes[i]->eba_tbl);
1295 ubi->volumes[i]->eba_tbl = NULL;