2 * Copyright (c) International Business Machines Corp., 2006
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 * Author: Artem Bityutskiy (Битюцкий Артём)
22 * The UBI Eraseblock Association (EBA) sub-system.
24 * This sub-system is responsible for I/O to/from logical eraseblock.
26 * Although in this implementation the EBA table is fully kept and managed in
27 * RAM, which assumes poor scalability, it might be (partially) maintained on
28 * flash in future implementations.
30 * The EBA sub-system implements per-logical eraseblock locking. Before
31 * accessing a logical eraseblock it is locked for reading or writing. The
32 * per-logical eraseblock locking is implemented by means of the lock tree. The
33 * lock tree is an RB-tree which refers all the currently locked logical
34 * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
35 * They are indexed by (@vol_id, @lnum) pairs.
37 * EBA also maintains the global sequence counter which is incremented each
38 * time a logical eraseblock is mapped to a physical eraseblock and it is
39 * stored in the volume identifier header. This means that each VID header has
40 * a unique sequence number. The sequence number is only increased an we assume
41 * 64 bits is enough to never overflow.
44 #include <linux/slab.h>
45 #include <linux/crc32.h>
46 #include <linux/err.h>
49 /* Number of physical eraseblocks reserved for atomic LEB change operation */
50 #define EBA_RESERVED_PEBS 1
53 * next_sqnum - get next sequence number.
54 * @ubi: UBI device description object
56 * This function returns next sequence number to use, which is just the current
57 * global sequence counter value. It also increases the global sequence
60 static unsigned long long next_sqnum(struct ubi_device
*ubi
)
62 unsigned long long sqnum
;
64 spin_lock(&ubi
->ltree_lock
);
65 sqnum
= ubi
->global_sqnum
++;
66 spin_unlock(&ubi
->ltree_lock
);
72 * ubi_get_compat - get compatibility flags of a volume.
73 * @ubi: UBI device description object
76 * This function returns compatibility flags for an internal volume. User
77 * volumes have no compatibility flags, so %0 is returned.
79 static int ubi_get_compat(const struct ubi_device
*ubi
, int vol_id
)
81 if (vol_id
== UBI_LAYOUT_VOLUME_ID
)
82 return UBI_LAYOUT_VOLUME_COMPAT
;
87 * ltree_lookup - look up the lock tree.
88 * @ubi: UBI device description object
90 * @lnum: logical eraseblock number
92 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
93 * object if the logical eraseblock is locked and %NULL if it is not.
94 * @ubi->ltree_lock has to be locked.
96 static struct ubi_ltree_entry
*ltree_lookup(struct ubi_device
*ubi
, int vol_id
,
101 p
= ubi
->ltree
.rb_node
;
103 struct ubi_ltree_entry
*le
;
105 le
= rb_entry(p
, struct ubi_ltree_entry
, rb
);
107 if (vol_id
< le
->vol_id
)
109 else if (vol_id
> le
->vol_id
)
114 else if (lnum
> le
->lnum
)
125 * ltree_add_entry - add new entry to the lock tree.
126 * @ubi: UBI device description object
128 * @lnum: logical eraseblock number
130 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
131 * lock tree. If such entry is already there, its usage counter is increased.
132 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
135 static struct ubi_ltree_entry
*ltree_add_entry(struct ubi_device
*ubi
,
136 int vol_id
, int lnum
)
138 struct ubi_ltree_entry
*le
, *le1
, *le_free
;
140 le
= kmalloc(sizeof(struct ubi_ltree_entry
), GFP_NOFS
);
142 return ERR_PTR(-ENOMEM
);
145 init_rwsem(&le
->mutex
);
149 spin_lock(&ubi
->ltree_lock
);
150 le1
= ltree_lookup(ubi
, vol_id
, lnum
);
154 * This logical eraseblock is already locked. The newly
155 * allocated lock entry is not needed.
160 struct rb_node
**p
, *parent
= NULL
;
163 * No lock entry, add the newly allocated one to the
164 * @ubi->ltree RB-tree.
168 p
= &ubi
->ltree
.rb_node
;
171 le1
= rb_entry(parent
, struct ubi_ltree_entry
, rb
);
173 if (vol_id
< le1
->vol_id
)
175 else if (vol_id
> le1
->vol_id
)
178 ubi_assert(lnum
!= le1
->lnum
);
179 if (lnum
< le1
->lnum
)
186 rb_link_node(&le
->rb
, parent
, p
);
187 rb_insert_color(&le
->rb
, &ubi
->ltree
);
190 spin_unlock(&ubi
->ltree_lock
);
197 * leb_read_lock - lock logical eraseblock for reading.
198 * @ubi: UBI device description object
200 * @lnum: logical eraseblock number
202 * This function locks a logical eraseblock for reading. Returns zero in case
203 * of success and a negative error code in case of failure.
205 static int leb_read_lock(struct ubi_device
*ubi
, int vol_id
, int lnum
)
207 struct ubi_ltree_entry
*le
;
209 le
= ltree_add_entry(ubi
, vol_id
, lnum
);
212 down_read(&le
->mutex
);
217 * leb_read_unlock - unlock logical eraseblock.
218 * @ubi: UBI device description object
220 * @lnum: logical eraseblock number
222 static void leb_read_unlock(struct ubi_device
*ubi
, int vol_id
, int lnum
)
224 struct ubi_ltree_entry
*le
;
226 spin_lock(&ubi
->ltree_lock
);
227 le
= ltree_lookup(ubi
, vol_id
, lnum
);
229 ubi_assert(le
->users
>= 0);
231 if (le
->users
== 0) {
232 rb_erase(&le
->rb
, &ubi
->ltree
);
235 spin_unlock(&ubi
->ltree_lock
);
239 * leb_write_lock - lock logical eraseblock for writing.
240 * @ubi: UBI device description object
242 * @lnum: logical eraseblock number
244 * This function locks a logical eraseblock for writing. Returns zero in case
245 * of success and a negative error code in case of failure.
247 static int leb_write_lock(struct ubi_device
*ubi
, int vol_id
, int lnum
)
249 struct ubi_ltree_entry
*le
;
251 le
= ltree_add_entry(ubi
, vol_id
, lnum
);
254 down_write(&le
->mutex
);
259 * leb_write_lock - lock logical eraseblock for writing.
260 * @ubi: UBI device description object
262 * @lnum: logical eraseblock number
264 * This function locks a logical eraseblock for writing if there is no
265 * contention and does nothing if there is contention. Returns %0 in case of
266 * success, %1 in case of contention, and and a negative error code in case of
269 static int leb_write_trylock(struct ubi_device
*ubi
, int vol_id
, int lnum
)
271 struct ubi_ltree_entry
*le
;
273 le
= ltree_add_entry(ubi
, vol_id
, lnum
);
276 if (down_write_trylock(&le
->mutex
))
279 /* Contention, cancel */
280 spin_lock(&ubi
->ltree_lock
);
282 ubi_assert(le
->users
>= 0);
283 if (le
->users
== 0) {
284 rb_erase(&le
->rb
, &ubi
->ltree
);
287 spin_unlock(&ubi
->ltree_lock
);
293 * leb_write_unlock - unlock logical eraseblock.
294 * @ubi: UBI device description object
296 * @lnum: logical eraseblock number
298 static void leb_write_unlock(struct ubi_device
*ubi
, int vol_id
, int lnum
)
300 struct ubi_ltree_entry
*le
;
302 spin_lock(&ubi
->ltree_lock
);
303 le
= ltree_lookup(ubi
, vol_id
, lnum
);
305 ubi_assert(le
->users
>= 0);
306 up_write(&le
->mutex
);
307 if (le
->users
== 0) {
308 rb_erase(&le
->rb
, &ubi
->ltree
);
311 spin_unlock(&ubi
->ltree_lock
);
315 * ubi_eba_unmap_leb - un-map logical eraseblock.
316 * @ubi: UBI device description object
317 * @vol: volume description object
318 * @lnum: logical eraseblock number
320 * This function un-maps logical eraseblock @lnum and schedules corresponding
321 * physical eraseblock for erasure. Returns zero in case of success and a
322 * negative error code in case of failure.
324 int ubi_eba_unmap_leb(struct ubi_device
*ubi
, struct ubi_volume
*vol
,
327 int err
, pnum
, vol_id
= vol
->vol_id
;
332 err
= leb_write_lock(ubi
, vol_id
, lnum
);
336 pnum
= vol
->eba_tbl
[lnum
];
338 /* This logical eraseblock is already unmapped */
341 dbg_eba("erase LEB %d:%d, PEB %d", vol_id
, lnum
, pnum
);
343 vol
->eba_tbl
[lnum
] = UBI_LEB_UNMAPPED
;
344 err
= ubi_wl_put_peb(ubi
, pnum
, 0);
347 leb_write_unlock(ubi
, vol_id
, lnum
);
352 * ubi_eba_read_leb - read data.
353 * @ubi: UBI device description object
354 * @vol: volume description object
355 * @lnum: logical eraseblock number
356 * @buf: buffer to store the read data
357 * @offset: offset from where to read
358 * @len: how many bytes to read
359 * @check: data CRC check flag
361 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
362 * bytes. The @check flag only makes sense for static volumes and forces
363 * eraseblock data CRC checking.
365 * In case of success this function returns zero. In case of a static volume,
366 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
367 * returned for any volume type if an ECC error was detected by the MTD device
368 * driver. Other negative error cored may be returned in case of other errors.
370 int ubi_eba_read_leb(struct ubi_device
*ubi
, struct ubi_volume
*vol
, int lnum
,
371 void *buf
, int offset
, int len
, int check
)
373 int err
, pnum
, scrub
= 0, vol_id
= vol
->vol_id
;
374 struct ubi_vid_hdr
*vid_hdr
;
375 uint32_t uninitialized_var(crc
);
377 err
= leb_read_lock(ubi
, vol_id
, lnum
);
381 pnum
= vol
->eba_tbl
[lnum
];
384 * The logical eraseblock is not mapped, fill the whole buffer
385 * with 0xFF bytes. The exception is static volumes for which
386 * it is an error to read unmapped logical eraseblocks.
388 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
389 len
, offset
, vol_id
, lnum
);
390 leb_read_unlock(ubi
, vol_id
, lnum
);
391 ubi_assert(vol
->vol_type
!= UBI_STATIC_VOLUME
);
392 memset(buf
, 0xFF, len
);
396 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
397 len
, offset
, vol_id
, lnum
, pnum
);
399 if (vol
->vol_type
== UBI_DYNAMIC_VOLUME
)
404 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
410 err
= ubi_io_read_vid_hdr(ubi
, pnum
, vid_hdr
, 1);
411 if (err
&& err
!= UBI_IO_BITFLIPS
) {
413 if (err
== UBI_IO_BAD_HDR_READ
||
414 err
== UBI_IO_BAD_HDR
) {
415 ubi_warn("corrupted VID header at PEB "
416 "%d, LEB %d:%d", pnum
, vol_id
,
423 } else if (err
== UBI_IO_BITFLIPS
)
426 ubi_assert(lnum
< be32_to_cpu(vid_hdr
->used_ebs
));
427 ubi_assert(len
== be32_to_cpu(vid_hdr
->data_size
));
429 crc
= be32_to_cpu(vid_hdr
->data_crc
);
430 ubi_free_vid_hdr(ubi
, vid_hdr
);
433 err
= ubi_io_read_data(ubi
, buf
, pnum
, offset
, len
);
435 if (err
== UBI_IO_BITFLIPS
) {
438 } else if (err
== -EBADMSG
) {
439 if (vol
->vol_type
== UBI_DYNAMIC_VOLUME
)
443 ubi_msg("force data checking");
452 uint32_t crc1
= crc32(UBI_CRC32_INIT
, buf
, len
);
454 ubi_warn("CRC error: calculated %#08x, must be %#08x",
462 err
= ubi_wl_scrub_peb(ubi
, pnum
);
464 leb_read_unlock(ubi
, vol_id
, lnum
);
468 ubi_free_vid_hdr(ubi
, vid_hdr
);
470 leb_read_unlock(ubi
, vol_id
, lnum
);
475 * recover_peb - recover from write failure.
476 * @ubi: UBI device description object
477 * @pnum: the physical eraseblock to recover
479 * @lnum: logical eraseblock number
480 * @buf: data which was not written because of the write failure
481 * @offset: offset of the failed write
482 * @len: how many bytes should have been written
484 * This function is called in case of a write failure and moves all good data
485 * from the potentially bad physical eraseblock to a good physical eraseblock.
486 * This function also writes the data which was not written due to the failure.
487 * Returns new physical eraseblock number in case of success, and a negative
488 * error code in case of failure.
490 static int recover_peb(struct ubi_device
*ubi
, int pnum
, int vol_id
, int lnum
,
491 const void *buf
, int offset
, int len
)
493 int err
, idx
= vol_id2idx(ubi
, vol_id
), new_pnum
, data_size
, tries
= 0;
494 struct ubi_volume
*vol
= ubi
->volumes
[idx
];
495 struct ubi_vid_hdr
*vid_hdr
;
497 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
502 new_pnum
= ubi_wl_get_peb(ubi
, UBI_UNKNOWN
);
504 ubi_free_vid_hdr(ubi
, vid_hdr
);
508 ubi_msg("recover PEB %d, move data to PEB %d", pnum
, new_pnum
);
510 err
= ubi_io_read_vid_hdr(ubi
, pnum
, vid_hdr
, 1);
511 if (err
&& err
!= UBI_IO_BITFLIPS
) {
517 vid_hdr
->sqnum
= cpu_to_be64(next_sqnum(ubi
));
518 err
= ubi_io_write_vid_hdr(ubi
, new_pnum
, vid_hdr
);
522 data_size
= offset
+ len
;
523 mutex_lock(&ubi
->buf_mutex
);
524 memset(ubi
->peb_buf1
+ offset
, 0xFF, len
);
526 /* Read everything before the area where the write failure happened */
528 err
= ubi_io_read_data(ubi
, ubi
->peb_buf1
, pnum
, 0, offset
);
529 if (err
&& err
!= UBI_IO_BITFLIPS
)
533 memcpy(ubi
->peb_buf1
+ offset
, buf
, len
);
535 err
= ubi_io_write_data(ubi
, ubi
->peb_buf1
, new_pnum
, 0, data_size
);
537 mutex_unlock(&ubi
->buf_mutex
);
541 mutex_unlock(&ubi
->buf_mutex
);
542 ubi_free_vid_hdr(ubi
, vid_hdr
);
544 vol
->eba_tbl
[lnum
] = new_pnum
;
545 ubi_wl_put_peb(ubi
, pnum
, 1);
547 ubi_msg("data was successfully recovered");
551 mutex_unlock(&ubi
->buf_mutex
);
553 ubi_wl_put_peb(ubi
, new_pnum
, 1);
554 ubi_free_vid_hdr(ubi
, vid_hdr
);
559 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
562 ubi_warn("failed to write to PEB %d", new_pnum
);
563 ubi_wl_put_peb(ubi
, new_pnum
, 1);
564 if (++tries
> UBI_IO_RETRIES
) {
565 ubi_free_vid_hdr(ubi
, vid_hdr
);
568 ubi_msg("try again");
573 * ubi_eba_write_leb - write data to dynamic volume.
574 * @ubi: UBI device description object
575 * @vol: volume description object
576 * @lnum: logical eraseblock number
577 * @buf: the data to write
578 * @offset: offset within the logical eraseblock where to write
579 * @len: how many bytes to write
582 * This function writes data to logical eraseblock @lnum of a dynamic volume
583 * @vol. Returns zero in case of success and a negative error code in case
584 * of failure. In case of error, it is possible that something was still
585 * written to the flash media, but may be some garbage.
587 int ubi_eba_write_leb(struct ubi_device
*ubi
, struct ubi_volume
*vol
, int lnum
,
588 const void *buf
, int offset
, int len
, int dtype
)
590 int err
, pnum
, tries
= 0, vol_id
= vol
->vol_id
;
591 struct ubi_vid_hdr
*vid_hdr
;
596 err
= leb_write_lock(ubi
, vol_id
, lnum
);
600 pnum
= vol
->eba_tbl
[lnum
];
602 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
603 len
, offset
, vol_id
, lnum
, pnum
);
605 err
= ubi_io_write_data(ubi
, buf
, pnum
, offset
, len
);
607 ubi_warn("failed to write data to PEB %d", pnum
);
608 if (err
== -EIO
&& ubi
->bad_allowed
)
609 err
= recover_peb(ubi
, pnum
, vol_id
, lnum
, buf
,
614 leb_write_unlock(ubi
, vol_id
, lnum
);
619 * The logical eraseblock is not mapped. We have to get a free physical
620 * eraseblock and write the volume identifier header there first.
622 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
624 leb_write_unlock(ubi
, vol_id
, lnum
);
628 vid_hdr
->vol_type
= UBI_VID_DYNAMIC
;
629 vid_hdr
->sqnum
= cpu_to_be64(next_sqnum(ubi
));
630 vid_hdr
->vol_id
= cpu_to_be32(vol_id
);
631 vid_hdr
->lnum
= cpu_to_be32(lnum
);
632 vid_hdr
->compat
= ubi_get_compat(ubi
, vol_id
);
633 vid_hdr
->data_pad
= cpu_to_be32(vol
->data_pad
);
636 pnum
= ubi_wl_get_peb(ubi
, dtype
);
638 ubi_free_vid_hdr(ubi
, vid_hdr
);
639 leb_write_unlock(ubi
, vol_id
, lnum
);
643 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
644 len
, offset
, vol_id
, lnum
, pnum
);
646 err
= ubi_io_write_vid_hdr(ubi
, pnum
, vid_hdr
);
648 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
654 err
= ubi_io_write_data(ubi
, buf
, pnum
, offset
, len
);
656 ubi_warn("failed to write %d bytes at offset %d of "
657 "LEB %d:%d, PEB %d", len
, offset
, vol_id
,
663 vol
->eba_tbl
[lnum
] = pnum
;
665 leb_write_unlock(ubi
, vol_id
, lnum
);
666 ubi_free_vid_hdr(ubi
, vid_hdr
);
670 if (err
!= -EIO
|| !ubi
->bad_allowed
) {
672 leb_write_unlock(ubi
, vol_id
, lnum
);
673 ubi_free_vid_hdr(ubi
, vid_hdr
);
678 * Fortunately, this is the first write operation to this physical
679 * eraseblock, so just put it and request a new one. We assume that if
680 * this physical eraseblock went bad, the erase code will handle that.
682 err
= ubi_wl_put_peb(ubi
, pnum
, 1);
683 if (err
|| ++tries
> UBI_IO_RETRIES
) {
685 leb_write_unlock(ubi
, vol_id
, lnum
);
686 ubi_free_vid_hdr(ubi
, vid_hdr
);
690 vid_hdr
->sqnum
= cpu_to_be64(next_sqnum(ubi
));
691 ubi_msg("try another PEB");
696 * ubi_eba_write_leb_st - write data to static volume.
697 * @ubi: UBI device description object
698 * @vol: volume description object
699 * @lnum: logical eraseblock number
700 * @buf: data to write
701 * @len: how many bytes to write
703 * @used_ebs: how many logical eraseblocks will this volume contain
705 * This function writes data to logical eraseblock @lnum of static volume
706 * @vol. The @used_ebs argument should contain total number of logical
707 * eraseblock in this static volume.
709 * When writing to the last logical eraseblock, the @len argument doesn't have
710 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
711 * to the real data size, although the @buf buffer has to contain the
712 * alignment. In all other cases, @len has to be aligned.
714 * It is prohibited to write more than once to logical eraseblocks of static
715 * volumes. This function returns zero in case of success and a negative error
716 * code in case of failure.
718 int ubi_eba_write_leb_st(struct ubi_device
*ubi
, struct ubi_volume
*vol
,
719 int lnum
, const void *buf
, int len
, int dtype
,
722 int err
, pnum
, tries
= 0, data_size
= len
, vol_id
= vol
->vol_id
;
723 struct ubi_vid_hdr
*vid_hdr
;
729 if (lnum
== used_ebs
- 1)
730 /* If this is the last LEB @len may be unaligned */
731 len
= ALIGN(data_size
, ubi
->min_io_size
);
733 ubi_assert(!(len
& (ubi
->min_io_size
- 1)));
735 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
739 err
= leb_write_lock(ubi
, vol_id
, lnum
);
741 ubi_free_vid_hdr(ubi
, vid_hdr
);
745 vid_hdr
->sqnum
= cpu_to_be64(next_sqnum(ubi
));
746 vid_hdr
->vol_id
= cpu_to_be32(vol_id
);
747 vid_hdr
->lnum
= cpu_to_be32(lnum
);
748 vid_hdr
->compat
= ubi_get_compat(ubi
, vol_id
);
749 vid_hdr
->data_pad
= cpu_to_be32(vol
->data_pad
);
751 crc
= crc32(UBI_CRC32_INIT
, buf
, data_size
);
752 vid_hdr
->vol_type
= UBI_VID_STATIC
;
753 vid_hdr
->data_size
= cpu_to_be32(data_size
);
754 vid_hdr
->used_ebs
= cpu_to_be32(used_ebs
);
755 vid_hdr
->data_crc
= cpu_to_be32(crc
);
758 pnum
= ubi_wl_get_peb(ubi
, dtype
);
760 ubi_free_vid_hdr(ubi
, vid_hdr
);
761 leb_write_unlock(ubi
, vol_id
, lnum
);
765 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
766 len
, vol_id
, lnum
, pnum
, used_ebs
);
768 err
= ubi_io_write_vid_hdr(ubi
, pnum
, vid_hdr
);
770 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
775 err
= ubi_io_write_data(ubi
, buf
, pnum
, 0, len
);
777 ubi_warn("failed to write %d bytes of data to PEB %d",
782 ubi_assert(vol
->eba_tbl
[lnum
] < 0);
783 vol
->eba_tbl
[lnum
] = pnum
;
785 leb_write_unlock(ubi
, vol_id
, lnum
);
786 ubi_free_vid_hdr(ubi
, vid_hdr
);
790 if (err
!= -EIO
|| !ubi
->bad_allowed
) {
792 * This flash device does not admit of bad eraseblocks or
793 * something nasty and unexpected happened. Switch to read-only
797 leb_write_unlock(ubi
, vol_id
, lnum
);
798 ubi_free_vid_hdr(ubi
, vid_hdr
);
802 err
= ubi_wl_put_peb(ubi
, pnum
, 1);
803 if (err
|| ++tries
> UBI_IO_RETRIES
) {
805 leb_write_unlock(ubi
, vol_id
, lnum
);
806 ubi_free_vid_hdr(ubi
, vid_hdr
);
810 vid_hdr
->sqnum
= cpu_to_be64(next_sqnum(ubi
));
811 ubi_msg("try another PEB");
816 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
817 * @ubi: UBI device description object
818 * @vol: volume description object
819 * @lnum: logical eraseblock number
820 * @buf: data to write
821 * @len: how many bytes to write
824 * This function changes the contents of a logical eraseblock atomically. @buf
825 * has to contain new logical eraseblock data, and @len - the length of the
826 * data, which has to be aligned. This function guarantees that in case of an
827 * unclean reboot the old contents is preserved. Returns zero in case of
828 * success and a negative error code in case of failure.
830 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
831 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
833 int ubi_eba_atomic_leb_change(struct ubi_device
*ubi
, struct ubi_volume
*vol
,
834 int lnum
, const void *buf
, int len
, int dtype
)
836 int err
, pnum
, tries
= 0, vol_id
= vol
->vol_id
;
837 struct ubi_vid_hdr
*vid_hdr
;
845 * Special case when data length is zero. In this case the LEB
846 * has to be unmapped and mapped somewhere else.
848 err
= ubi_eba_unmap_leb(ubi
, vol
, lnum
);
851 return ubi_eba_write_leb(ubi
, vol
, lnum
, NULL
, 0, 0, dtype
);
854 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
858 mutex_lock(&ubi
->alc_mutex
);
859 err
= leb_write_lock(ubi
, vol_id
, lnum
);
863 vid_hdr
->sqnum
= cpu_to_be64(next_sqnum(ubi
));
864 vid_hdr
->vol_id
= cpu_to_be32(vol_id
);
865 vid_hdr
->lnum
= cpu_to_be32(lnum
);
866 vid_hdr
->compat
= ubi_get_compat(ubi
, vol_id
);
867 vid_hdr
->data_pad
= cpu_to_be32(vol
->data_pad
);
869 crc
= crc32(UBI_CRC32_INIT
, buf
, len
);
870 vid_hdr
->vol_type
= UBI_VID_DYNAMIC
;
871 vid_hdr
->data_size
= cpu_to_be32(len
);
872 vid_hdr
->copy_flag
= 1;
873 vid_hdr
->data_crc
= cpu_to_be32(crc
);
876 pnum
= ubi_wl_get_peb(ubi
, dtype
);
882 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
883 vol_id
, lnum
, vol
->eba_tbl
[lnum
], pnum
);
885 err
= ubi_io_write_vid_hdr(ubi
, pnum
, vid_hdr
);
887 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
892 err
= ubi_io_write_data(ubi
, buf
, pnum
, 0, len
);
894 ubi_warn("failed to write %d bytes of data to PEB %d",
899 if (vol
->eba_tbl
[lnum
] >= 0) {
900 err
= ubi_wl_put_peb(ubi
, vol
->eba_tbl
[lnum
], 0);
905 vol
->eba_tbl
[lnum
] = pnum
;
908 leb_write_unlock(ubi
, vol_id
, lnum
);
910 mutex_unlock(&ubi
->alc_mutex
);
911 ubi_free_vid_hdr(ubi
, vid_hdr
);
915 if (err
!= -EIO
|| !ubi
->bad_allowed
) {
917 * This flash device does not admit of bad eraseblocks or
918 * something nasty and unexpected happened. Switch to read-only
925 err
= ubi_wl_put_peb(ubi
, pnum
, 1);
926 if (err
|| ++tries
> UBI_IO_RETRIES
) {
931 vid_hdr
->sqnum
= cpu_to_be64(next_sqnum(ubi
));
932 ubi_msg("try another PEB");
937 * is_error_sane - check whether a read error is sane.
938 * @err: code of the error happened during reading
940 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
941 * cannot read data from the target PEB (an error @err happened). If the error
942 * code is sane, then we treat this error as non-fatal. Otherwise the error is
943 * fatal and UBI will be switched to R/O mode later.
945 * The idea is that we try not to switch to R/O mode if the read error is
946 * something which suggests there was a real read problem. E.g., %-EIO. Or a
947 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
948 * mode, simply because we do not know what happened at the MTD level, and we
949 * cannot handle this. E.g., the underlying driver may have become crazy, and
950 * it is safer to switch to R/O mode to preserve the data.
952 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
953 * which we have just written.
955 static int is_error_sane(int err
)
957 if (err
== -EIO
|| err
== -ENOMEM
|| err
== UBI_IO_BAD_HDR
||
958 err
== UBI_IO_BAD_HDR_READ
|| err
== -ETIMEDOUT
)
964 * ubi_eba_copy_leb - copy logical eraseblock.
965 * @ubi: UBI device description object
966 * @from: physical eraseblock number from where to copy
967 * @to: physical eraseblock number where to copy
968 * @vid_hdr: VID header of the @from physical eraseblock
970 * This function copies logical eraseblock from physical eraseblock @from to
971 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
973 * o %0 in case of success;
974 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_CANCEL_BITFLIPS, etc;
975 * o a negative error code in case of failure.
977 int ubi_eba_copy_leb(struct ubi_device
*ubi
, int from
, int to
,
978 struct ubi_vid_hdr
*vid_hdr
)
980 int err
, vol_id
, lnum
, data_size
, aldata_size
, idx
;
981 struct ubi_volume
*vol
;
984 vol_id
= be32_to_cpu(vid_hdr
->vol_id
);
985 lnum
= be32_to_cpu(vid_hdr
->lnum
);
987 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id
, lnum
, from
, to
);
989 if (vid_hdr
->vol_type
== UBI_VID_STATIC
) {
990 data_size
= be32_to_cpu(vid_hdr
->data_size
);
991 aldata_size
= ALIGN(data_size
, ubi
->min_io_size
);
993 data_size
= aldata_size
=
994 ubi
->leb_size
- be32_to_cpu(vid_hdr
->data_pad
);
996 idx
= vol_id2idx(ubi
, vol_id
);
997 spin_lock(&ubi
->volumes_lock
);
999 * Note, we may race with volume deletion, which means that the volume
1000 * this logical eraseblock belongs to might be being deleted. Since the
1001 * volume deletion un-maps all the volume's logical eraseblocks, it will
1002 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1004 vol
= ubi
->volumes
[idx
];
1005 spin_unlock(&ubi
->volumes_lock
);
1007 /* No need to do further work, cancel */
1008 dbg_wl("volume %d is being removed, cancel", vol_id
);
1009 return MOVE_CANCEL_RACE
;
1013 * We do not want anybody to write to this logical eraseblock while we
1014 * are moving it, so lock it.
1016 * Note, we are using non-waiting locking here, because we cannot sleep
1017 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1018 * unmapping the LEB which is mapped to the PEB we are going to move
1019 * (@from). This task locks the LEB and goes sleep in the
1020 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1021 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1022 * LEB is already locked, we just do not move it and return
1023 * %MOVE_CANCEL_RACE, which means that UBI will re-try, but later.
1025 err
= leb_write_trylock(ubi
, vol_id
, lnum
);
1027 dbg_wl("contention on LEB %d:%d, cancel", vol_id
, lnum
);
1028 return MOVE_CANCEL_RACE
;
1032 * The LEB might have been put meanwhile, and the task which put it is
1033 * probably waiting on @ubi->move_mutex. No need to continue the work,
1036 if (vol
->eba_tbl
[lnum
] != from
) {
1037 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to "
1038 "PEB %d, cancel", vol_id
, lnum
, from
,
1039 vol
->eba_tbl
[lnum
]);
1040 err
= MOVE_CANCEL_RACE
;
1041 goto out_unlock_leb
;
1045 * OK, now the LEB is locked and we can safely start moving it. Since
1046 * this function utilizes the @ubi->peb_buf1 buffer which is shared
1047 * with some other functions - we lock the buffer by taking the
1050 mutex_lock(&ubi
->buf_mutex
);
1051 dbg_wl("read %d bytes of data", aldata_size
);
1052 err
= ubi_io_read_data(ubi
, ubi
->peb_buf1
, from
, 0, aldata_size
);
1053 if (err
&& err
!= UBI_IO_BITFLIPS
) {
1054 ubi_warn("error %d while reading data from PEB %d",
1056 err
= MOVE_SOURCE_RD_ERR
;
1057 goto out_unlock_buf
;
1061 * Now we have got to calculate how much data we have to copy. In
1062 * case of a static volume it is fairly easy - the VID header contains
1063 * the data size. In case of a dynamic volume it is more difficult - we
1064 * have to read the contents, cut 0xFF bytes from the end and copy only
1065 * the first part. We must do this to avoid writing 0xFF bytes as it
1066 * may have some side-effects. And not only this. It is important not
1067 * to include those 0xFFs to CRC because later the they may be filled
1070 if (vid_hdr
->vol_type
== UBI_VID_DYNAMIC
)
1071 aldata_size
= data_size
=
1072 ubi_calc_data_len(ubi
, ubi
->peb_buf1
, data_size
);
1075 crc
= crc32(UBI_CRC32_INIT
, ubi
->peb_buf1
, data_size
);
1079 * It may turn out to be that the whole @from physical eraseblock
1080 * contains only 0xFF bytes. Then we have to only write the VID header
1081 * and do not write any data. This also means we should not set
1082 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1084 if (data_size
> 0) {
1085 vid_hdr
->copy_flag
= 1;
1086 vid_hdr
->data_size
= cpu_to_be32(data_size
);
1087 vid_hdr
->data_crc
= cpu_to_be32(crc
);
1089 vid_hdr
->sqnum
= cpu_to_be64(next_sqnum(ubi
));
1091 err
= ubi_io_write_vid_hdr(ubi
, to
, vid_hdr
);
1094 err
= MOVE_TARGET_WR_ERR
;
1095 goto out_unlock_buf
;
1100 /* Read the VID header back and check if it was written correctly */
1101 err
= ubi_io_read_vid_hdr(ubi
, to
, vid_hdr
, 1);
1103 if (err
!= UBI_IO_BITFLIPS
) {
1104 ubi_warn("error %d while reading VID header back from "
1106 if (is_error_sane(err
))
1107 err
= MOVE_TARGET_RD_ERR
;
1109 err
= MOVE_CANCEL_BITFLIPS
;
1110 goto out_unlock_buf
;
1113 if (data_size
> 0) {
1114 err
= ubi_io_write_data(ubi
, ubi
->peb_buf1
, to
, 0, aldata_size
);
1117 err
= MOVE_TARGET_WR_ERR
;
1118 goto out_unlock_buf
;
1124 * We've written the data and are going to read it back to make
1125 * sure it was written correctly.
1128 err
= ubi_io_read_data(ubi
, ubi
->peb_buf2
, to
, 0, aldata_size
);
1130 if (err
!= UBI_IO_BITFLIPS
) {
1131 ubi_warn("error %d while reading data back "
1132 "from PEB %d", err
, to
);
1133 if (is_error_sane(err
))
1134 err
= MOVE_TARGET_RD_ERR
;
1136 err
= MOVE_CANCEL_BITFLIPS
;
1137 goto out_unlock_buf
;
1142 if (memcmp(ubi
->peb_buf1
, ubi
->peb_buf2
, aldata_size
)) {
1143 ubi_warn("read data back from PEB %d and it is "
1146 goto out_unlock_buf
;
1150 ubi_assert(vol
->eba_tbl
[lnum
] == from
);
1151 vol
->eba_tbl
[lnum
] = to
;
1154 mutex_unlock(&ubi
->buf_mutex
);
1156 leb_write_unlock(ubi
, vol_id
, lnum
);
1161 * print_rsvd_warning - warn about not having enough reserved PEBs.
1162 * @ubi: UBI device description object
1164 * This is a helper function for 'ubi_eba_init_scan()' which is called when UBI
1165 * cannot reserve enough PEBs for bad block handling. This function makes a
1166 * decision whether we have to print a warning or not. The algorithm is as
1168 * o if this is a new UBI image, then just print the warning
1169 * o if this is an UBI image which has already been used for some time, print
1170 * a warning only if we can reserve less than 10% of the expected amount of
1173 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1174 * of PEBs becomes smaller, which is normal and we do not want to scare users
1175 * with a warning every time they attach the MTD device. This was an issue
1176 * reported by real users.
1178 static void print_rsvd_warning(struct ubi_device
*ubi
,
1179 struct ubi_scan_info
*si
)
1182 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1183 * large number to distinguish between newly flashed and used images.
1185 if (si
->max_sqnum
> (1 << 18)) {
1186 int min
= ubi
->beb_rsvd_level
/ 10;
1190 if (ubi
->beb_rsvd_pebs
> min
)
1194 ubi_warn("cannot reserve enough PEBs for bad PEB handling, reserved %d,"
1195 " need %d", ubi
->beb_rsvd_pebs
, ubi
->beb_rsvd_level
);
1199 * ubi_eba_init_scan - initialize the EBA sub-system using scanning information.
1200 * @ubi: UBI device description object
1201 * @si: scanning information
1203 * This function returns zero in case of success and a negative error code in
1206 int ubi_eba_init_scan(struct ubi_device
*ubi
, struct ubi_scan_info
*si
)
1208 int i
, j
, err
, num_volumes
;
1209 struct ubi_scan_volume
*sv
;
1210 struct ubi_volume
*vol
;
1211 struct ubi_scan_leb
*seb
;
1214 dbg_eba("initialize EBA sub-system");
1216 spin_lock_init(&ubi
->ltree_lock
);
1217 mutex_init(&ubi
->alc_mutex
);
1218 ubi
->ltree
= RB_ROOT
;
1220 ubi
->global_sqnum
= si
->max_sqnum
+ 1;
1221 num_volumes
= ubi
->vtbl_slots
+ UBI_INT_VOL_COUNT
;
1223 for (i
= 0; i
< num_volumes
; i
++) {
1224 vol
= ubi
->volumes
[i
];
1230 vol
->eba_tbl
= kmalloc(vol
->reserved_pebs
* sizeof(int),
1232 if (!vol
->eba_tbl
) {
1237 for (j
= 0; j
< vol
->reserved_pebs
; j
++)
1238 vol
->eba_tbl
[j
] = UBI_LEB_UNMAPPED
;
1240 sv
= ubi_scan_find_sv(si
, idx2vol_id(ubi
, i
));
1244 ubi_rb_for_each_entry(rb
, seb
, &sv
->root
, u
.rb
) {
1245 if (seb
->lnum
>= vol
->reserved_pebs
)
1247 * This may happen in case of an unclean reboot
1250 ubi_scan_move_to_list(sv
, seb
, &si
->erase
);
1251 vol
->eba_tbl
[seb
->lnum
] = seb
->pnum
;
1255 if (ubi
->avail_pebs
< EBA_RESERVED_PEBS
) {
1256 ubi_err("no enough physical eraseblocks (%d, need %d)",
1257 ubi
->avail_pebs
, EBA_RESERVED_PEBS
);
1261 ubi
->avail_pebs
-= EBA_RESERVED_PEBS
;
1262 ubi
->rsvd_pebs
+= EBA_RESERVED_PEBS
;
1264 if (ubi
->bad_allowed
) {
1265 ubi_calculate_reserved(ubi
);
1267 if (ubi
->avail_pebs
< ubi
->beb_rsvd_level
) {
1268 /* No enough free physical eraseblocks */
1269 ubi
->beb_rsvd_pebs
= ubi
->avail_pebs
;
1270 print_rsvd_warning(ubi
, si
);
1272 ubi
->beb_rsvd_pebs
= ubi
->beb_rsvd_level
;
1274 ubi
->avail_pebs
-= ubi
->beb_rsvd_pebs
;
1275 ubi
->rsvd_pebs
+= ubi
->beb_rsvd_pebs
;
1278 dbg_eba("EBA sub-system is initialized");
1282 for (i
= 0; i
< num_volumes
; i
++) {
1283 if (!ubi
->volumes
[i
])
1285 kfree(ubi
->volumes
[i
]->eba_tbl
);
1286 ubi
->volumes
[i
]->eba_tbl
= NULL
;