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",
430 * Ending up here in the non-Fastmap case
431 * is a clear bug as the VID header had to
432 * be present at scan time to have it referenced.
433 * With fastmap the story is more complicated.
434 * Fastmap has the mapping info without the need
435 * of a full scan. So the LEB could have been
436 * unmapped, Fastmap cannot know this and keeps
437 * the LEB referenced.
438 * This is valid and works as the layer above UBI
439 * has to do bookkeeping about used/referenced
442 if (ubi->fast_attach) {
451 } else if (err == UBI_IO_BITFLIPS)
454 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
455 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
457 crc = be32_to_cpu(vid_hdr->data_crc);
458 ubi_free_vid_hdr(ubi, vid_hdr);
461 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
463 if (err == UBI_IO_BITFLIPS)
465 else if (mtd_is_eccerr(err)) {
466 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
470 ubi_msg(ubi, "force data checking");
479 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
481 ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
489 err = ubi_wl_scrub_peb(ubi, pnum);
491 leb_read_unlock(ubi, vol_id, lnum);
495 ubi_free_vid_hdr(ubi, vid_hdr);
497 leb_read_unlock(ubi, vol_id, lnum);
502 * ubi_eba_read_leb_sg - read data into a scatter gather list.
503 * @ubi: UBI device description object
504 * @vol: volume description object
505 * @lnum: logical eraseblock number
506 * @sgl: UBI scatter gather list to store the read data
507 * @offset: offset from where to read
508 * @len: how many bytes to read
509 * @check: data CRC check flag
511 * This function works exactly like ubi_eba_read_leb(). But instead of
512 * storing the read data into a buffer it writes to an UBI scatter gather
515 int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
516 struct ubi_sgl *sgl, int lnum, int offset, int len,
521 struct scatterlist *sg;
524 ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
525 sg = &sgl->sg[sgl->list_pos];
526 if (len < sg->length - sgl->page_pos)
529 to_read = sg->length - sgl->page_pos;
531 ret = ubi_eba_read_leb(ubi, vol, lnum,
532 sg_virt(sg) + sgl->page_pos, offset,
540 sgl->page_pos += to_read;
541 if (sgl->page_pos == sg->length) {
557 * recover_peb - recover from write failure.
558 * @ubi: UBI device description object
559 * @pnum: the physical eraseblock to recover
561 * @lnum: logical eraseblock number
562 * @buf: data which was not written because of the write failure
563 * @offset: offset of the failed write
564 * @len: how many bytes should have been written
566 * This function is called in case of a write failure and moves all good data
567 * from the potentially bad physical eraseblock to a good physical eraseblock.
568 * This function also writes the data which was not written due to the failure.
569 * Returns new physical eraseblock number in case of success, and a negative
570 * error code in case of failure.
572 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
573 const void *buf, int offset, int len)
575 int err, idx = vol_id2idx(ubi, vol_id), new_pnum, data_size, tries = 0;
576 struct ubi_volume *vol = ubi->volumes[idx];
577 struct ubi_vid_hdr *vid_hdr;
580 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
585 new_pnum = ubi_wl_get_peb(ubi);
587 ubi_free_vid_hdr(ubi, vid_hdr);
588 up_read(&ubi->fm_eba_sem);
592 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
595 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
596 if (err && err != UBI_IO_BITFLIPS) {
599 up_read(&ubi->fm_eba_sem);
603 ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
605 mutex_lock(&ubi->buf_mutex);
606 memset(ubi->peb_buf + offset, 0xFF, len);
608 /* Read everything before the area where the write failure happened */
610 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
611 if (err && err != UBI_IO_BITFLIPS) {
612 up_read(&ubi->fm_eba_sem);
617 memcpy(ubi->peb_buf + offset, buf, len);
619 data_size = offset + len;
620 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
621 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
622 vid_hdr->copy_flag = 1;
623 vid_hdr->data_size = cpu_to_be32(data_size);
624 vid_hdr->data_crc = cpu_to_be32(crc);
625 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
627 mutex_unlock(&ubi->buf_mutex);
628 up_read(&ubi->fm_eba_sem);
632 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
634 mutex_unlock(&ubi->buf_mutex);
635 up_read(&ubi->fm_eba_sem);
639 mutex_unlock(&ubi->buf_mutex);
640 ubi_free_vid_hdr(ubi, vid_hdr);
642 vol->eba_tbl[lnum] = new_pnum;
643 up_read(&ubi->fm_eba_sem);
644 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
646 ubi_msg(ubi, "data was successfully recovered");
650 mutex_unlock(&ubi->buf_mutex);
652 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
653 ubi_free_vid_hdr(ubi, vid_hdr);
658 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
661 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
662 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
663 if (++tries > UBI_IO_RETRIES) {
664 ubi_free_vid_hdr(ubi, vid_hdr);
667 ubi_msg(ubi, "try again");
672 * ubi_eba_write_leb - write data to dynamic volume.
673 * @ubi: UBI device description object
674 * @vol: volume description object
675 * @lnum: logical eraseblock number
676 * @buf: the data to write
677 * @offset: offset within the logical eraseblock where to write
678 * @len: how many bytes to write
680 * This function writes data to logical eraseblock @lnum of a dynamic volume
681 * @vol. Returns zero in case of success and a negative error code in case
682 * of failure. In case of error, it is possible that something was still
683 * written to the flash media, but may be some garbage.
685 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
686 const void *buf, int offset, int len)
688 int err, pnum, tries = 0, vol_id = vol->vol_id;
689 struct ubi_vid_hdr *vid_hdr;
694 err = leb_write_lock(ubi, vol_id, lnum);
698 pnum = vol->eba_tbl[lnum];
700 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
701 len, offset, vol_id, lnum, pnum);
703 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
705 ubi_warn(ubi, "failed to write data to PEB %d", pnum);
706 if (err == -EIO && ubi->bad_allowed)
707 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
712 leb_write_unlock(ubi, vol_id, lnum);
717 * The logical eraseblock is not mapped. We have to get a free physical
718 * eraseblock and write the volume identifier header there first.
720 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
722 leb_write_unlock(ubi, vol_id, lnum);
726 vid_hdr->vol_type = UBI_VID_DYNAMIC;
727 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
728 vid_hdr->vol_id = cpu_to_be32(vol_id);
729 vid_hdr->lnum = cpu_to_be32(lnum);
730 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
731 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
734 pnum = ubi_wl_get_peb(ubi);
736 ubi_free_vid_hdr(ubi, vid_hdr);
737 leb_write_unlock(ubi, vol_id, lnum);
738 up_read(&ubi->fm_eba_sem);
742 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
743 len, offset, vol_id, lnum, pnum);
745 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
747 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
749 up_read(&ubi->fm_eba_sem);
754 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
756 ubi_warn(ubi, "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
757 len, offset, vol_id, lnum, pnum);
758 up_read(&ubi->fm_eba_sem);
763 vol->eba_tbl[lnum] = pnum;
764 up_read(&ubi->fm_eba_sem);
766 leb_write_unlock(ubi, vol_id, lnum);
767 ubi_free_vid_hdr(ubi, vid_hdr);
771 if (err != -EIO || !ubi->bad_allowed) {
773 leb_write_unlock(ubi, vol_id, lnum);
774 ubi_free_vid_hdr(ubi, vid_hdr);
779 * Fortunately, this is the first write operation to this physical
780 * eraseblock, so just put it and request a new one. We assume that if
781 * this physical eraseblock went bad, the erase code will handle that.
783 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
784 if (err || ++tries > UBI_IO_RETRIES) {
786 leb_write_unlock(ubi, vol_id, lnum);
787 ubi_free_vid_hdr(ubi, vid_hdr);
791 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
792 ubi_msg(ubi, "try another PEB");
797 * ubi_eba_write_leb_st - write data to static volume.
798 * @ubi: UBI device description object
799 * @vol: volume description object
800 * @lnum: logical eraseblock number
801 * @buf: data to write
802 * @len: how many bytes to write
803 * @used_ebs: how many logical eraseblocks will this volume contain
805 * This function writes data to logical eraseblock @lnum of static volume
806 * @vol. The @used_ebs argument should contain total number of logical
807 * eraseblock in this static volume.
809 * When writing to the last logical eraseblock, the @len argument doesn't have
810 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
811 * to the real data size, although the @buf buffer has to contain the
812 * alignment. In all other cases, @len has to be aligned.
814 * It is prohibited to write more than once to logical eraseblocks of static
815 * volumes. This function returns zero in case of success and a negative error
816 * code in case of failure.
818 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
819 int lnum, const void *buf, int len, int used_ebs)
821 int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
822 struct ubi_vid_hdr *vid_hdr;
828 if (lnum == used_ebs - 1)
829 /* If this is the last LEB @len may be unaligned */
830 len = ALIGN(data_size, ubi->min_io_size);
832 ubi_assert(!(len & (ubi->min_io_size - 1)));
834 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
838 err = leb_write_lock(ubi, vol_id, lnum);
840 ubi_free_vid_hdr(ubi, vid_hdr);
844 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
845 vid_hdr->vol_id = cpu_to_be32(vol_id);
846 vid_hdr->lnum = cpu_to_be32(lnum);
847 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
848 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
850 crc = crc32(UBI_CRC32_INIT, buf, data_size);
851 vid_hdr->vol_type = UBI_VID_STATIC;
852 vid_hdr->data_size = cpu_to_be32(data_size);
853 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
854 vid_hdr->data_crc = cpu_to_be32(crc);
857 pnum = ubi_wl_get_peb(ubi);
859 ubi_free_vid_hdr(ubi, vid_hdr);
860 leb_write_unlock(ubi, vol_id, lnum);
861 up_read(&ubi->fm_eba_sem);
865 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
866 len, vol_id, lnum, pnum, used_ebs);
868 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
870 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
872 up_read(&ubi->fm_eba_sem);
876 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
878 ubi_warn(ubi, "failed to write %d bytes of data to PEB %d",
880 up_read(&ubi->fm_eba_sem);
884 ubi_assert(vol->eba_tbl[lnum] < 0);
885 vol->eba_tbl[lnum] = pnum;
886 up_read(&ubi->fm_eba_sem);
888 leb_write_unlock(ubi, vol_id, lnum);
889 ubi_free_vid_hdr(ubi, vid_hdr);
893 if (err != -EIO || !ubi->bad_allowed) {
895 * This flash device does not admit of bad eraseblocks or
896 * something nasty and unexpected happened. Switch to read-only
900 leb_write_unlock(ubi, vol_id, lnum);
901 ubi_free_vid_hdr(ubi, vid_hdr);
905 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
906 if (err || ++tries > UBI_IO_RETRIES) {
908 leb_write_unlock(ubi, vol_id, lnum);
909 ubi_free_vid_hdr(ubi, vid_hdr);
913 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
914 ubi_msg(ubi, "try another PEB");
919 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
920 * @ubi: UBI device description object
921 * @vol: volume description object
922 * @lnum: logical eraseblock number
923 * @buf: data to write
924 * @len: how many bytes to write
926 * This function changes the contents of a logical eraseblock atomically. @buf
927 * has to contain new logical eraseblock data, and @len - the length of the
928 * data, which has to be aligned. This function guarantees that in case of an
929 * unclean reboot the old contents is preserved. Returns zero in case of
930 * success and a negative error code in case of failure.
932 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
933 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
935 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
936 int lnum, const void *buf, int len)
938 int err, pnum, old_pnum, tries = 0, vol_id = vol->vol_id;
939 struct ubi_vid_hdr *vid_hdr;
947 * Special case when data length is zero. In this case the LEB
948 * has to be unmapped and mapped somewhere else.
950 err = ubi_eba_unmap_leb(ubi, vol, lnum);
953 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
956 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
960 mutex_lock(&ubi->alc_mutex);
961 err = leb_write_lock(ubi, vol_id, lnum);
965 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
966 vid_hdr->vol_id = cpu_to_be32(vol_id);
967 vid_hdr->lnum = cpu_to_be32(lnum);
968 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
969 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
971 crc = crc32(UBI_CRC32_INIT, buf, len);
972 vid_hdr->vol_type = UBI_VID_DYNAMIC;
973 vid_hdr->data_size = cpu_to_be32(len);
974 vid_hdr->copy_flag = 1;
975 vid_hdr->data_crc = cpu_to_be32(crc);
978 pnum = ubi_wl_get_peb(ubi);
981 up_read(&ubi->fm_eba_sem);
985 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
986 vol_id, lnum, vol->eba_tbl[lnum], pnum);
988 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
990 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
992 up_read(&ubi->fm_eba_sem);
996 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
998 ubi_warn(ubi, "failed to write %d bytes of data to PEB %d",
1000 up_read(&ubi->fm_eba_sem);
1004 old_pnum = vol->eba_tbl[lnum];
1005 vol->eba_tbl[lnum] = pnum;
1006 up_read(&ubi->fm_eba_sem);
1008 if (old_pnum >= 0) {
1009 err = ubi_wl_put_peb(ubi, vol_id, lnum, old_pnum, 0);
1011 goto out_leb_unlock;
1015 leb_write_unlock(ubi, vol_id, lnum);
1017 mutex_unlock(&ubi->alc_mutex);
1018 ubi_free_vid_hdr(ubi, vid_hdr);
1022 if (err != -EIO || !ubi->bad_allowed) {
1024 * This flash device does not admit of bad eraseblocks or
1025 * something nasty and unexpected happened. Switch to read-only
1026 * mode just in case.
1029 goto out_leb_unlock;
1032 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
1033 if (err || ++tries > UBI_IO_RETRIES) {
1035 goto out_leb_unlock;
1038 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1039 ubi_msg(ubi, "try another PEB");
1044 * is_error_sane - check whether a read error is sane.
1045 * @err: code of the error happened during reading
1047 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1048 * cannot read data from the target PEB (an error @err happened). If the error
1049 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1050 * fatal and UBI will be switched to R/O mode later.
1052 * The idea is that we try not to switch to R/O mode if the read error is
1053 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1054 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1055 * mode, simply because we do not know what happened at the MTD level, and we
1056 * cannot handle this. E.g., the underlying driver may have become crazy, and
1057 * it is safer to switch to R/O mode to preserve the data.
1059 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1060 * which we have just written.
1062 static int is_error_sane(int err)
1064 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1065 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1071 * ubi_eba_copy_leb - copy logical eraseblock.
1072 * @ubi: UBI device description object
1073 * @from: physical eraseblock number from where to copy
1074 * @to: physical eraseblock number where to copy
1075 * @vid_hdr: VID header of the @from physical eraseblock
1077 * This function copies logical eraseblock from physical eraseblock @from to
1078 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1079 * function. Returns:
1080 * o %0 in case of success;
1081 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1082 * o a negative error code in case of failure.
1084 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1085 struct ubi_vid_hdr *vid_hdr)
1087 int err, vol_id, lnum, data_size, aldata_size, idx;
1088 struct ubi_volume *vol;
1091 vol_id = be32_to_cpu(vid_hdr->vol_id);
1092 lnum = be32_to_cpu(vid_hdr->lnum);
1094 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1096 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1097 data_size = be32_to_cpu(vid_hdr->data_size);
1098 aldata_size = ALIGN(data_size, ubi->min_io_size);
1100 data_size = aldata_size =
1101 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1103 idx = vol_id2idx(ubi, vol_id);
1104 spin_lock(&ubi->volumes_lock);
1106 * Note, we may race with volume deletion, which means that the volume
1107 * this logical eraseblock belongs to might be being deleted. Since the
1108 * volume deletion un-maps all the volume's logical eraseblocks, it will
1109 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1111 vol = ubi->volumes[idx];
1112 spin_unlock(&ubi->volumes_lock);
1114 /* No need to do further work, cancel */
1115 dbg_wl("volume %d is being removed, cancel", vol_id);
1116 return MOVE_CANCEL_RACE;
1120 * We do not want anybody to write to this logical eraseblock while we
1121 * are moving it, so lock it.
1123 * Note, we are using non-waiting locking here, because we cannot sleep
1124 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1125 * unmapping the LEB which is mapped to the PEB we are going to move
1126 * (@from). This task locks the LEB and goes sleep in the
1127 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1128 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1129 * LEB is already locked, we just do not move it and return
1130 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1131 * we do not know the reasons of the contention - it may be just a
1132 * normal I/O on this LEB, so we want to re-try.
1134 err = leb_write_trylock(ubi, vol_id, lnum);
1136 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1141 * The LEB might have been put meanwhile, and the task which put it is
1142 * probably waiting on @ubi->move_mutex. No need to continue the work,
1145 if (vol->eba_tbl[lnum] != from) {
1146 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1147 vol_id, lnum, from, vol->eba_tbl[lnum]);
1148 err = MOVE_CANCEL_RACE;
1149 goto out_unlock_leb;
1153 * OK, now the LEB is locked and we can safely start moving it. Since
1154 * this function utilizes the @ubi->peb_buf buffer which is shared
1155 * with some other functions - we lock the buffer by taking the
1158 mutex_lock(&ubi->buf_mutex);
1159 dbg_wl("read %d bytes of data", aldata_size);
1160 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1161 if (err && err != UBI_IO_BITFLIPS) {
1162 ubi_warn(ubi, "error %d while reading data from PEB %d",
1164 err = MOVE_SOURCE_RD_ERR;
1165 goto out_unlock_buf;
1169 * Now we have got to calculate how much data we have to copy. In
1170 * case of a static volume it is fairly easy - the VID header contains
1171 * the data size. In case of a dynamic volume it is more difficult - we
1172 * have to read the contents, cut 0xFF bytes from the end and copy only
1173 * the first part. We must do this to avoid writing 0xFF bytes as it
1174 * may have some side-effects. And not only this. It is important not
1175 * to include those 0xFFs to CRC because later the they may be filled
1178 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1179 aldata_size = data_size =
1180 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1183 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1187 * It may turn out to be that the whole @from physical eraseblock
1188 * contains only 0xFF bytes. Then we have to only write the VID header
1189 * and do not write any data. This also means we should not set
1190 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1192 if (data_size > 0) {
1193 vid_hdr->copy_flag = 1;
1194 vid_hdr->data_size = cpu_to_be32(data_size);
1195 vid_hdr->data_crc = cpu_to_be32(crc);
1197 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1199 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1202 err = MOVE_TARGET_WR_ERR;
1203 goto out_unlock_buf;
1208 /* Read the VID header back and check if it was written correctly */
1209 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1211 if (err != UBI_IO_BITFLIPS) {
1212 ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1214 if (is_error_sane(err))
1215 err = MOVE_TARGET_RD_ERR;
1217 err = MOVE_TARGET_BITFLIPS;
1218 goto out_unlock_buf;
1221 if (data_size > 0) {
1222 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1225 err = MOVE_TARGET_WR_ERR;
1226 goto out_unlock_buf;
1232 ubi_assert(vol->eba_tbl[lnum] == from);
1233 down_read(&ubi->fm_eba_sem);
1234 vol->eba_tbl[lnum] = to;
1235 up_read(&ubi->fm_eba_sem);
1238 mutex_unlock(&ubi->buf_mutex);
1240 leb_write_unlock(ubi, vol_id, lnum);
1245 * print_rsvd_warning - warn about not having enough reserved PEBs.
1246 * @ubi: UBI device description object
1248 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1249 * cannot reserve enough PEBs for bad block handling. This function makes a
1250 * decision whether we have to print a warning or not. The algorithm is as
1252 * o if this is a new UBI image, then just print the warning
1253 * o if this is an UBI image which has already been used for some time, print
1254 * a warning only if we can reserve less than 10% of the expected amount of
1257 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1258 * of PEBs becomes smaller, which is normal and we do not want to scare users
1259 * with a warning every time they attach the MTD device. This was an issue
1260 * reported by real users.
1262 static void print_rsvd_warning(struct ubi_device *ubi,
1263 struct ubi_attach_info *ai)
1266 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1267 * large number to distinguish between newly flashed and used images.
1269 if (ai->max_sqnum > (1 << 18)) {
1270 int min = ubi->beb_rsvd_level / 10;
1274 if (ubi->beb_rsvd_pebs > min)
1278 ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1279 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1280 if (ubi->corr_peb_count)
1281 ubi_warn(ubi, "%d PEBs are corrupted and not used",
1282 ubi->corr_peb_count);
1286 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1287 * @ubi: UBI device description object
1288 * @ai_fastmap: UBI attach info object created by fastmap
1289 * @ai_scan: UBI attach info object created by scanning
1291 * Returns < 0 in case of an internal error, 0 otherwise.
1292 * If a bad EBA table entry was found it will be printed out and
1293 * ubi_assert() triggers.
1295 int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1296 struct ubi_attach_info *ai_scan)
1298 int i, j, num_volumes, ret = 0;
1299 int **scan_eba, **fm_eba;
1300 struct ubi_ainf_volume *av;
1301 struct ubi_volume *vol;
1302 struct ubi_ainf_peb *aeb;
1305 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1307 scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1311 fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1317 for (i = 0; i < num_volumes; i++) {
1318 vol = ubi->volumes[i];
1322 scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1329 fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1336 for (j = 0; j < vol->reserved_pebs; j++)
1337 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1339 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1343 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1344 scan_eba[i][aeb->lnum] = aeb->pnum;
1346 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1350 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1351 fm_eba[i][aeb->lnum] = aeb->pnum;
1353 for (j = 0; j < vol->reserved_pebs; j++) {
1354 if (scan_eba[i][j] != fm_eba[i][j]) {
1355 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1356 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1359 ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1360 vol->vol_id, j, fm_eba[i][j],
1368 for (i = 0; i < num_volumes; i++) {
1369 if (!ubi->volumes[i])
1382 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1383 * @ubi: UBI device description object
1384 * @ai: attaching information
1386 * This function returns zero in case of success and a negative error code in
1389 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1391 int i, j, err, num_volumes;
1392 struct ubi_ainf_volume *av;
1393 struct ubi_volume *vol;
1394 struct ubi_ainf_peb *aeb;
1397 dbg_eba("initialize EBA sub-system");
1399 spin_lock_init(&ubi->ltree_lock);
1400 mutex_init(&ubi->alc_mutex);
1401 ubi->ltree = RB_ROOT;
1403 ubi->global_sqnum = ai->max_sqnum + 1;
1404 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1406 for (i = 0; i < num_volumes; i++) {
1407 vol = ubi->volumes[i];
1413 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1415 if (!vol->eba_tbl) {
1420 for (j = 0; j < vol->reserved_pebs; j++)
1421 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1423 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1427 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1428 if (aeb->lnum >= vol->reserved_pebs)
1430 * This may happen in case of an unclean reboot
1433 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1435 vol->eba_tbl[aeb->lnum] = aeb->pnum;
1439 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1440 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1441 ubi->avail_pebs, EBA_RESERVED_PEBS);
1442 if (ubi->corr_peb_count)
1443 ubi_err(ubi, "%d PEBs are corrupted and not used",
1444 ubi->corr_peb_count);
1448 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1449 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1451 if (ubi->bad_allowed) {
1452 ubi_calculate_reserved(ubi);
1454 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1455 /* No enough free physical eraseblocks */
1456 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1457 print_rsvd_warning(ubi, ai);
1459 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1461 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1462 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1465 dbg_eba("EBA sub-system is initialized");
1469 for (i = 0; i < num_volumes; i++) {
1470 if (!ubi->volumes[i])
1472 kfree(ubi->volumes[i]->eba_tbl);
1473 ubi->volumes[i]->eba_tbl = NULL;