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 * try_recover_peb - try to recover from write failure.
558 * @vol: volume description object
559 * @pnum: the physical eraseblock to recover
560 * @lnum: logical eraseblock number
561 * @buf: data which was not written because of the write failure
562 * @offset: offset of the failed write
563 * @len: how many bytes should have been written
565 * @retry: whether the caller should retry in case of failure
567 * This function is called in case of a write failure and moves all good data
568 * from the potentially bad physical eraseblock to a good physical eraseblock.
569 * This function also writes the data which was not written due to the failure.
570 * Returns 0 in case of success, and a negative error code in case of failure.
571 * In case of failure, the %retry parameter is set to false if this is a fatal
572 * error (retrying won't help), and true otherwise.
574 static int try_recover_peb(struct ubi_volume *vol, int pnum, int lnum,
575 const void *buf, int offset, int len,
576 struct ubi_vid_hdr *vid_hdr, bool *retry)
578 struct ubi_device *ubi = vol->ubi;
579 int new_pnum, err, vol_id = vol->vol_id, data_size;
584 new_pnum = ubi_wl_get_peb(ubi);
590 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
593 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
594 if (err && err != UBI_IO_BITFLIPS) {
600 ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
602 mutex_lock(&ubi->buf_mutex);
603 memset(ubi->peb_buf + offset, 0xFF, len);
605 /* Read everything before the area where the write failure happened */
607 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
608 if (err && err != UBI_IO_BITFLIPS)
614 memcpy(ubi->peb_buf + offset, buf, len);
616 data_size = offset + len;
617 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
618 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
619 vid_hdr->copy_flag = 1;
620 vid_hdr->data_size = cpu_to_be32(data_size);
621 vid_hdr->data_crc = cpu_to_be32(crc);
622 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
626 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
629 mutex_unlock(&ubi->buf_mutex);
632 vol->eba_tbl[lnum] = new_pnum;
635 up_read(&ubi->fm_eba_sem);
638 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
639 ubi_msg(ubi, "data was successfully recovered");
640 } else if (new_pnum >= 0) {
642 * Bad luck? This physical eraseblock is bad too? Crud. Let's
643 * try to get another one.
645 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
646 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
653 * recover_peb - recover from write failure.
654 * @ubi: UBI device description object
655 * @pnum: the physical eraseblock to recover
657 * @lnum: logical eraseblock number
658 * @buf: data which was not written because of the write failure
659 * @offset: offset of the failed write
660 * @len: how many bytes should have been written
662 * This function is called in case of a write failure and moves all good data
663 * from the potentially bad physical eraseblock to a good physical eraseblock.
664 * This function also writes the data which was not written due to the failure.
665 * Returns 0 in case of success, and a negative error code in case of failure.
666 * This function tries %UBI_IO_RETRIES before giving up.
668 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
669 const void *buf, int offset, int len)
671 int err, idx = vol_id2idx(ubi, vol_id), tries;
672 struct ubi_volume *vol = ubi->volumes[idx];
673 struct ubi_vid_hdr *vid_hdr;
675 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
679 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
682 err = try_recover_peb(vol, pnum, lnum, buf, offset, len,
687 ubi_msg(ubi, "try again");
690 ubi_free_vid_hdr(ubi, vid_hdr);
696 * try_write_vid_and_data - try to write VID header and data to a new PEB.
697 * @vol: volume description object
698 * @lnum: logical eraseblock number
699 * @vid_hdr: VID header to write
700 * @buf: buffer containing the data
701 * @offset: where to start writing data
702 * @len: how many bytes should be written
704 * This function tries to write VID header and data belonging to logical
705 * eraseblock @lnum of volume @vol to a new physical eraseblock. Returns zero
706 * in case of success and a negative error code in case of failure.
707 * In case of error, it is possible that something was still written to the
708 * flash media, but may be some garbage.
710 static int try_write_vid_and_data(struct ubi_volume *vol, int lnum,
711 struct ubi_vid_hdr *vid_hdr, const void *buf,
714 struct ubi_device *ubi = vol->ubi;
715 int pnum, opnum, err, vol_id = vol->vol_id;
717 pnum = ubi_wl_get_peb(ubi);
723 opnum = vol->eba_tbl[lnum];
725 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
726 len, offset, vol_id, lnum, pnum);
728 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
730 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
736 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
739 "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
740 len, offset, vol_id, lnum, pnum);
745 vol->eba_tbl[lnum] = pnum;
748 up_read(&ubi->fm_eba_sem);
750 if (err && pnum >= 0)
751 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
752 else if (!err && opnum >= 0)
753 err = ubi_wl_put_peb(ubi, vol_id, lnum, opnum, 0);
759 * ubi_eba_write_leb - write data to dynamic volume.
760 * @ubi: UBI device description object
761 * @vol: volume description object
762 * @lnum: logical eraseblock number
763 * @buf: the data to write
764 * @offset: offset within the logical eraseblock where to write
765 * @len: how many bytes to write
767 * This function writes data to logical eraseblock @lnum of a dynamic volume
768 * @vol. Returns zero in case of success and a negative error code in case
769 * of failure. In case of error, it is possible that something was still
770 * written to the flash media, but may be some garbage.
771 * This function retries %UBI_IO_RETRIES times before giving up.
773 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
774 const void *buf, int offset, int len)
776 int err, pnum, tries, vol_id = vol->vol_id;
777 struct ubi_vid_hdr *vid_hdr;
782 err = leb_write_lock(ubi, vol_id, lnum);
786 pnum = vol->eba_tbl[lnum];
788 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
789 len, offset, vol_id, lnum, pnum);
791 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
793 ubi_warn(ubi, "failed to write data to PEB %d", pnum);
794 if (err == -EIO && ubi->bad_allowed)
795 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
803 * The logical eraseblock is not mapped. We have to get a free physical
804 * eraseblock and write the volume identifier header there first.
806 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
808 leb_write_unlock(ubi, vol_id, lnum);
812 vid_hdr->vol_type = UBI_VID_DYNAMIC;
813 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
814 vid_hdr->vol_id = cpu_to_be32(vol_id);
815 vid_hdr->lnum = cpu_to_be32(lnum);
816 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
817 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
819 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
820 err = try_write_vid_and_data(vol, lnum, vid_hdr, buf, offset,
822 if (err != -EIO || !ubi->bad_allowed)
826 * Fortunately, this is the first write operation to this
827 * physical eraseblock, so just put it and request a new one.
828 * We assume that if this physical eraseblock went bad, the
829 * erase code will handle that.
831 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
832 ubi_msg(ubi, "try another PEB");
835 ubi_free_vid_hdr(ubi, vid_hdr);
841 leb_write_unlock(ubi, vol_id, lnum);
847 * ubi_eba_write_leb_st - write data to static volume.
848 * @ubi: UBI device description object
849 * @vol: volume description object
850 * @lnum: logical eraseblock number
851 * @buf: data to write
852 * @len: how many bytes to write
853 * @used_ebs: how many logical eraseblocks will this volume contain
855 * This function writes data to logical eraseblock @lnum of static volume
856 * @vol. The @used_ebs argument should contain total number of logical
857 * eraseblock in this static volume.
859 * When writing to the last logical eraseblock, the @len argument doesn't have
860 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
861 * to the real data size, although the @buf buffer has to contain the
862 * alignment. In all other cases, @len has to be aligned.
864 * It is prohibited to write more than once to logical eraseblocks of static
865 * volumes. This function returns zero in case of success and a negative error
866 * code in case of failure.
868 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
869 int lnum, const void *buf, int len, int used_ebs)
871 int err, tries, data_size = len, vol_id = vol->vol_id;
872 struct ubi_vid_hdr *vid_hdr;
878 if (lnum == used_ebs - 1)
879 /* If this is the last LEB @len may be unaligned */
880 len = ALIGN(data_size, ubi->min_io_size);
882 ubi_assert(!(len & (ubi->min_io_size - 1)));
884 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
888 err = leb_write_lock(ubi, vol_id, lnum);
892 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
893 vid_hdr->vol_id = cpu_to_be32(vol_id);
894 vid_hdr->lnum = cpu_to_be32(lnum);
895 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
896 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
898 crc = crc32(UBI_CRC32_INIT, buf, data_size);
899 vid_hdr->vol_type = UBI_VID_STATIC;
900 vid_hdr->data_size = cpu_to_be32(data_size);
901 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
902 vid_hdr->data_crc = cpu_to_be32(crc);
904 ubi_assert(vol->eba_tbl[lnum] < 0);
906 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
907 err = try_write_vid_and_data(vol, lnum, vid_hdr, buf, 0, len);
908 if (err != -EIO || !ubi->bad_allowed)
911 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
912 ubi_msg(ubi, "try another PEB");
918 leb_write_unlock(ubi, vol_id, lnum);
921 ubi_free_vid_hdr(ubi, vid_hdr);
927 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
928 * @ubi: UBI device description object
929 * @vol: volume description object
930 * @lnum: logical eraseblock number
931 * @buf: data to write
932 * @len: how many bytes to write
934 * This function changes the contents of a logical eraseblock atomically. @buf
935 * has to contain new logical eraseblock data, and @len - the length of the
936 * data, which has to be aligned. This function guarantees that in case of an
937 * unclean reboot the old contents is preserved. Returns zero in case of
938 * success and a negative error code in case of failure.
940 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
941 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
943 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
944 int lnum, const void *buf, int len)
946 int err, tries, vol_id = vol->vol_id;
947 struct ubi_vid_hdr *vid_hdr;
955 * Special case when data length is zero. In this case the LEB
956 * has to be unmapped and mapped somewhere else.
958 err = ubi_eba_unmap_leb(ubi, vol, lnum);
961 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
964 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
968 mutex_lock(&ubi->alc_mutex);
969 err = leb_write_lock(ubi, vol_id, lnum);
973 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
974 vid_hdr->vol_id = cpu_to_be32(vol_id);
975 vid_hdr->lnum = cpu_to_be32(lnum);
976 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
977 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
979 crc = crc32(UBI_CRC32_INIT, buf, len);
980 vid_hdr->vol_type = UBI_VID_DYNAMIC;
981 vid_hdr->data_size = cpu_to_be32(len);
982 vid_hdr->copy_flag = 1;
983 vid_hdr->data_crc = cpu_to_be32(crc);
985 dbg_eba("change LEB %d:%d", vol_id, lnum);
987 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
988 err = try_write_vid_and_data(vol, lnum, vid_hdr, buf, 0, len);
989 if (err != -EIO || !ubi->bad_allowed)
992 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
993 ubi_msg(ubi, "try another PEB");
997 * This flash device does not admit of bad eraseblocks or
998 * something nasty and unexpected happened. Switch to read-only
1004 leb_write_unlock(ubi, vol_id, lnum);
1007 mutex_unlock(&ubi->alc_mutex);
1008 ubi_free_vid_hdr(ubi, vid_hdr);
1013 * is_error_sane - check whether a read error is sane.
1014 * @err: code of the error happened during reading
1016 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1017 * cannot read data from the target PEB (an error @err happened). If the error
1018 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1019 * fatal and UBI will be switched to R/O mode later.
1021 * The idea is that we try not to switch to R/O mode if the read error is
1022 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1023 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1024 * mode, simply because we do not know what happened at the MTD level, and we
1025 * cannot handle this. E.g., the underlying driver may have become crazy, and
1026 * it is safer to switch to R/O mode to preserve the data.
1028 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1029 * which we have just written.
1031 static int is_error_sane(int err)
1033 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1034 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1040 * ubi_eba_copy_leb - copy logical eraseblock.
1041 * @ubi: UBI device description object
1042 * @from: physical eraseblock number from where to copy
1043 * @to: physical eraseblock number where to copy
1044 * @vid_hdr: VID header of the @from physical eraseblock
1046 * This function copies logical eraseblock from physical eraseblock @from to
1047 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1048 * function. Returns:
1049 * o %0 in case of success;
1050 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1051 * o a negative error code in case of failure.
1053 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1054 struct ubi_vid_hdr *vid_hdr)
1056 int err, vol_id, lnum, data_size, aldata_size, idx;
1057 struct ubi_volume *vol;
1060 vol_id = be32_to_cpu(vid_hdr->vol_id);
1061 lnum = be32_to_cpu(vid_hdr->lnum);
1063 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1065 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1066 data_size = be32_to_cpu(vid_hdr->data_size);
1067 aldata_size = ALIGN(data_size, ubi->min_io_size);
1069 data_size = aldata_size =
1070 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1072 idx = vol_id2idx(ubi, vol_id);
1073 spin_lock(&ubi->volumes_lock);
1075 * Note, we may race with volume deletion, which means that the volume
1076 * this logical eraseblock belongs to might be being deleted. Since the
1077 * volume deletion un-maps all the volume's logical eraseblocks, it will
1078 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1080 vol = ubi->volumes[idx];
1081 spin_unlock(&ubi->volumes_lock);
1083 /* No need to do further work, cancel */
1084 dbg_wl("volume %d is being removed, cancel", vol_id);
1085 return MOVE_CANCEL_RACE;
1089 * We do not want anybody to write to this logical eraseblock while we
1090 * are moving it, so lock it.
1092 * Note, we are using non-waiting locking here, because we cannot sleep
1093 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1094 * unmapping the LEB which is mapped to the PEB we are going to move
1095 * (@from). This task locks the LEB and goes sleep in the
1096 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1097 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1098 * LEB is already locked, we just do not move it and return
1099 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1100 * we do not know the reasons of the contention - it may be just a
1101 * normal I/O on this LEB, so we want to re-try.
1103 err = leb_write_trylock(ubi, vol_id, lnum);
1105 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1110 * The LEB might have been put meanwhile, and the task which put it is
1111 * probably waiting on @ubi->move_mutex. No need to continue the work,
1114 if (vol->eba_tbl[lnum] != from) {
1115 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1116 vol_id, lnum, from, vol->eba_tbl[lnum]);
1117 err = MOVE_CANCEL_RACE;
1118 goto out_unlock_leb;
1122 * OK, now the LEB is locked and we can safely start moving it. Since
1123 * this function utilizes the @ubi->peb_buf buffer which is shared
1124 * with some other functions - we lock the buffer by taking the
1127 mutex_lock(&ubi->buf_mutex);
1128 dbg_wl("read %d bytes of data", aldata_size);
1129 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1130 if (err && err != UBI_IO_BITFLIPS) {
1131 ubi_warn(ubi, "error %d while reading data from PEB %d",
1133 err = MOVE_SOURCE_RD_ERR;
1134 goto out_unlock_buf;
1138 * Now we have got to calculate how much data we have to copy. In
1139 * case of a static volume it is fairly easy - the VID header contains
1140 * the data size. In case of a dynamic volume it is more difficult - we
1141 * have to read the contents, cut 0xFF bytes from the end and copy only
1142 * the first part. We must do this to avoid writing 0xFF bytes as it
1143 * may have some side-effects. And not only this. It is important not
1144 * to include those 0xFFs to CRC because later the they may be filled
1147 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1148 aldata_size = data_size =
1149 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1152 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1156 * It may turn out to be that the whole @from physical eraseblock
1157 * contains only 0xFF bytes. Then we have to only write the VID header
1158 * and do not write any data. This also means we should not set
1159 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1161 if (data_size > 0) {
1162 vid_hdr->copy_flag = 1;
1163 vid_hdr->data_size = cpu_to_be32(data_size);
1164 vid_hdr->data_crc = cpu_to_be32(crc);
1166 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1168 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1171 err = MOVE_TARGET_WR_ERR;
1172 goto out_unlock_buf;
1177 /* Read the VID header back and check if it was written correctly */
1178 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1180 if (err != UBI_IO_BITFLIPS) {
1181 ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1183 if (is_error_sane(err))
1184 err = MOVE_TARGET_RD_ERR;
1186 err = MOVE_TARGET_BITFLIPS;
1187 goto out_unlock_buf;
1190 if (data_size > 0) {
1191 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1194 err = MOVE_TARGET_WR_ERR;
1195 goto out_unlock_buf;
1201 ubi_assert(vol->eba_tbl[lnum] == from);
1202 down_read(&ubi->fm_eba_sem);
1203 vol->eba_tbl[lnum] = to;
1204 up_read(&ubi->fm_eba_sem);
1207 mutex_unlock(&ubi->buf_mutex);
1209 leb_write_unlock(ubi, vol_id, lnum);
1214 * print_rsvd_warning - warn about not having enough reserved PEBs.
1215 * @ubi: UBI device description object
1217 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1218 * cannot reserve enough PEBs for bad block handling. This function makes a
1219 * decision whether we have to print a warning or not. The algorithm is as
1221 * o if this is a new UBI image, then just print the warning
1222 * o if this is an UBI image which has already been used for some time, print
1223 * a warning only if we can reserve less than 10% of the expected amount of
1226 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1227 * of PEBs becomes smaller, which is normal and we do not want to scare users
1228 * with a warning every time they attach the MTD device. This was an issue
1229 * reported by real users.
1231 static void print_rsvd_warning(struct ubi_device *ubi,
1232 struct ubi_attach_info *ai)
1235 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1236 * large number to distinguish between newly flashed and used images.
1238 if (ai->max_sqnum > (1 << 18)) {
1239 int min = ubi->beb_rsvd_level / 10;
1243 if (ubi->beb_rsvd_pebs > min)
1247 ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1248 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1249 if (ubi->corr_peb_count)
1250 ubi_warn(ubi, "%d PEBs are corrupted and not used",
1251 ubi->corr_peb_count);
1255 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1256 * @ubi: UBI device description object
1257 * @ai_fastmap: UBI attach info object created by fastmap
1258 * @ai_scan: UBI attach info object created by scanning
1260 * Returns < 0 in case of an internal error, 0 otherwise.
1261 * If a bad EBA table entry was found it will be printed out and
1262 * ubi_assert() triggers.
1264 int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1265 struct ubi_attach_info *ai_scan)
1267 int i, j, num_volumes, ret = 0;
1268 int **scan_eba, **fm_eba;
1269 struct ubi_ainf_volume *av;
1270 struct ubi_volume *vol;
1271 struct ubi_ainf_peb *aeb;
1274 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1276 scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1280 fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1286 for (i = 0; i < num_volumes; i++) {
1287 vol = ubi->volumes[i];
1291 scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1298 fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1305 for (j = 0; j < vol->reserved_pebs; j++)
1306 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1308 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1312 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1313 scan_eba[i][aeb->lnum] = aeb->pnum;
1315 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1319 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1320 fm_eba[i][aeb->lnum] = aeb->pnum;
1322 for (j = 0; j < vol->reserved_pebs; j++) {
1323 if (scan_eba[i][j] != fm_eba[i][j]) {
1324 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1325 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1328 ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1329 vol->vol_id, j, fm_eba[i][j],
1337 for (i = 0; i < num_volumes; i++) {
1338 if (!ubi->volumes[i])
1351 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1352 * @ubi: UBI device description object
1353 * @ai: attaching information
1355 * This function returns zero in case of success and a negative error code in
1358 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1360 int i, j, err, num_volumes;
1361 struct ubi_ainf_volume *av;
1362 struct ubi_volume *vol;
1363 struct ubi_ainf_peb *aeb;
1366 dbg_eba("initialize EBA sub-system");
1368 spin_lock_init(&ubi->ltree_lock);
1369 mutex_init(&ubi->alc_mutex);
1370 ubi->ltree = RB_ROOT;
1372 ubi->global_sqnum = ai->max_sqnum + 1;
1373 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1375 for (i = 0; i < num_volumes; i++) {
1376 vol = ubi->volumes[i];
1382 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1384 if (!vol->eba_tbl) {
1389 for (j = 0; j < vol->reserved_pebs; j++)
1390 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1392 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1396 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1397 if (aeb->lnum >= vol->reserved_pebs)
1399 * This may happen in case of an unclean reboot
1402 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1404 vol->eba_tbl[aeb->lnum] = aeb->pnum;
1408 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1409 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1410 ubi->avail_pebs, EBA_RESERVED_PEBS);
1411 if (ubi->corr_peb_count)
1412 ubi_err(ubi, "%d PEBs are corrupted and not used",
1413 ubi->corr_peb_count);
1417 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1418 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1420 if (ubi->bad_allowed) {
1421 ubi_calculate_reserved(ubi);
1423 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1424 /* No enough free physical eraseblocks */
1425 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1426 print_rsvd_warning(ubi, ai);
1428 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1430 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1431 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1434 dbg_eba("EBA sub-system is initialized");
1438 for (i = 0; i < num_volumes; i++) {
1439 if (!ubi->volumes[i])
1441 kfree(ubi->volumes[i]->eba_tbl);
1442 ubi->volumes[i]->eba_tbl = NULL;