]>
git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - drivers/mtd/ubi/eba.c
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
;
579 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
584 new_pnum
= ubi_wl_get_peb(ubi
);
586 ubi_free_vid_hdr(ubi
, vid_hdr
);
587 up_read(&ubi
->fm_eba_sem
);
591 ubi_msg(ubi
, "recover PEB %d, move data to PEB %d",
594 err
= ubi_io_read_vid_hdr(ubi
, pnum
, vid_hdr
, 1);
595 if (err
&& err
!= UBI_IO_BITFLIPS
) {
598 up_read(&ubi
->fm_eba_sem
);
602 vid_hdr
->sqnum
= cpu_to_be64(ubi_next_sqnum(ubi
));
603 err
= ubi_io_write_vid_hdr(ubi
, new_pnum
, vid_hdr
);
605 up_read(&ubi
->fm_eba_sem
);
609 data_size
= offset
+ len
;
610 mutex_lock(&ubi
->buf_mutex
);
611 memset(ubi
->peb_buf
+ offset
, 0xFF, len
);
613 /* Read everything before the area where the write failure happened */
615 err
= ubi_io_read_data(ubi
, ubi
->peb_buf
, pnum
, 0, offset
);
616 if (err
&& err
!= UBI_IO_BITFLIPS
) {
617 up_read(&ubi
->fm_eba_sem
);
622 memcpy(ubi
->peb_buf
+ offset
, buf
, len
);
624 err
= ubi_io_write_data(ubi
, ubi
->peb_buf
, new_pnum
, 0, data_size
);
626 mutex_unlock(&ubi
->buf_mutex
);
627 up_read(&ubi
->fm_eba_sem
);
631 mutex_unlock(&ubi
->buf_mutex
);
632 ubi_free_vid_hdr(ubi
, vid_hdr
);
634 vol
->eba_tbl
[lnum
] = new_pnum
;
635 up_read(&ubi
->fm_eba_sem
);
636 ubi_wl_put_peb(ubi
, vol_id
, lnum
, pnum
, 1);
638 ubi_msg(ubi
, "data was successfully recovered");
642 mutex_unlock(&ubi
->buf_mutex
);
644 ubi_wl_put_peb(ubi
, vol_id
, lnum
, new_pnum
, 1);
645 ubi_free_vid_hdr(ubi
, vid_hdr
);
650 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
653 ubi_warn(ubi
, "failed to write to PEB %d", new_pnum
);
654 ubi_wl_put_peb(ubi
, vol_id
, lnum
, new_pnum
, 1);
655 if (++tries
> UBI_IO_RETRIES
) {
656 ubi_free_vid_hdr(ubi
, vid_hdr
);
659 ubi_msg(ubi
, "try again");
664 * ubi_eba_write_leb - write data to dynamic volume.
665 * @ubi: UBI device description object
666 * @vol: volume description object
667 * @lnum: logical eraseblock number
668 * @buf: the data to write
669 * @offset: offset within the logical eraseblock where to write
670 * @len: how many bytes to write
672 * This function writes data to logical eraseblock @lnum of a dynamic volume
673 * @vol. Returns zero in case of success and a negative error code in case
674 * of failure. In case of error, it is possible that something was still
675 * written to the flash media, but may be some garbage.
677 int ubi_eba_write_leb(struct ubi_device
*ubi
, struct ubi_volume
*vol
, int lnum
,
678 const void *buf
, int offset
, int len
)
680 int err
, pnum
, tries
= 0, vol_id
= vol
->vol_id
;
681 struct ubi_vid_hdr
*vid_hdr
;
686 err
= leb_write_lock(ubi
, vol_id
, lnum
);
690 pnum
= vol
->eba_tbl
[lnum
];
692 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
693 len
, offset
, vol_id
, lnum
, pnum
);
695 err
= ubi_io_write_data(ubi
, buf
, pnum
, offset
, len
);
697 ubi_warn(ubi
, "failed to write data to PEB %d", pnum
);
698 if (err
== -EIO
&& ubi
->bad_allowed
)
699 err
= recover_peb(ubi
, pnum
, vol_id
, lnum
, buf
,
704 leb_write_unlock(ubi
, vol_id
, lnum
);
709 * The logical eraseblock is not mapped. We have to get a free physical
710 * eraseblock and write the volume identifier header there first.
712 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
714 leb_write_unlock(ubi
, vol_id
, lnum
);
718 vid_hdr
->vol_type
= UBI_VID_DYNAMIC
;
719 vid_hdr
->sqnum
= cpu_to_be64(ubi_next_sqnum(ubi
));
720 vid_hdr
->vol_id
= cpu_to_be32(vol_id
);
721 vid_hdr
->lnum
= cpu_to_be32(lnum
);
722 vid_hdr
->compat
= ubi_get_compat(ubi
, vol_id
);
723 vid_hdr
->data_pad
= cpu_to_be32(vol
->data_pad
);
726 pnum
= ubi_wl_get_peb(ubi
);
728 ubi_free_vid_hdr(ubi
, vid_hdr
);
729 leb_write_unlock(ubi
, vol_id
, lnum
);
730 up_read(&ubi
->fm_eba_sem
);
734 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
735 len
, offset
, vol_id
, lnum
, pnum
);
737 err
= ubi_io_write_vid_hdr(ubi
, pnum
, vid_hdr
);
739 ubi_warn(ubi
, "failed to write VID header to LEB %d:%d, PEB %d",
741 up_read(&ubi
->fm_eba_sem
);
746 err
= ubi_io_write_data(ubi
, buf
, pnum
, offset
, len
);
748 ubi_warn(ubi
, "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
749 len
, offset
, vol_id
, lnum
, pnum
);
750 up_read(&ubi
->fm_eba_sem
);
755 vol
->eba_tbl
[lnum
] = pnum
;
756 up_read(&ubi
->fm_eba_sem
);
758 leb_write_unlock(ubi
, vol_id
, lnum
);
759 ubi_free_vid_hdr(ubi
, vid_hdr
);
763 if (err
!= -EIO
|| !ubi
->bad_allowed
) {
765 leb_write_unlock(ubi
, vol_id
, lnum
);
766 ubi_free_vid_hdr(ubi
, vid_hdr
);
771 * Fortunately, this is the first write operation to this physical
772 * eraseblock, so just put it and request a new one. We assume that if
773 * this physical eraseblock went bad, the erase code will handle that.
775 err
= ubi_wl_put_peb(ubi
, vol_id
, lnum
, pnum
, 1);
776 if (err
|| ++tries
> UBI_IO_RETRIES
) {
778 leb_write_unlock(ubi
, vol_id
, lnum
);
779 ubi_free_vid_hdr(ubi
, vid_hdr
);
783 vid_hdr
->sqnum
= cpu_to_be64(ubi_next_sqnum(ubi
));
784 ubi_msg(ubi
, "try another PEB");
789 * ubi_eba_write_leb_st - write data to static volume.
790 * @ubi: UBI device description object
791 * @vol: volume description object
792 * @lnum: logical eraseblock number
793 * @buf: data to write
794 * @len: how many bytes to write
795 * @used_ebs: how many logical eraseblocks will this volume contain
797 * This function writes data to logical eraseblock @lnum of static volume
798 * @vol. The @used_ebs argument should contain total number of logical
799 * eraseblock in this static volume.
801 * When writing to the last logical eraseblock, the @len argument doesn't have
802 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
803 * to the real data size, although the @buf buffer has to contain the
804 * alignment. In all other cases, @len has to be aligned.
806 * It is prohibited to write more than once to logical eraseblocks of static
807 * volumes. This function returns zero in case of success and a negative error
808 * code in case of failure.
810 int ubi_eba_write_leb_st(struct ubi_device
*ubi
, struct ubi_volume
*vol
,
811 int lnum
, const void *buf
, int len
, int used_ebs
)
813 int err
, pnum
, tries
= 0, data_size
= len
, vol_id
= vol
->vol_id
;
814 struct ubi_vid_hdr
*vid_hdr
;
820 if (lnum
== used_ebs
- 1)
821 /* If this is the last LEB @len may be unaligned */
822 len
= ALIGN(data_size
, ubi
->min_io_size
);
824 ubi_assert(!(len
& (ubi
->min_io_size
- 1)));
826 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
830 err
= leb_write_lock(ubi
, vol_id
, lnum
);
832 ubi_free_vid_hdr(ubi
, vid_hdr
);
836 vid_hdr
->sqnum
= cpu_to_be64(ubi_next_sqnum(ubi
));
837 vid_hdr
->vol_id
= cpu_to_be32(vol_id
);
838 vid_hdr
->lnum
= cpu_to_be32(lnum
);
839 vid_hdr
->compat
= ubi_get_compat(ubi
, vol_id
);
840 vid_hdr
->data_pad
= cpu_to_be32(vol
->data_pad
);
842 crc
= crc32(UBI_CRC32_INIT
, buf
, data_size
);
843 vid_hdr
->vol_type
= UBI_VID_STATIC
;
844 vid_hdr
->data_size
= cpu_to_be32(data_size
);
845 vid_hdr
->used_ebs
= cpu_to_be32(used_ebs
);
846 vid_hdr
->data_crc
= cpu_to_be32(crc
);
849 pnum
= ubi_wl_get_peb(ubi
);
851 ubi_free_vid_hdr(ubi
, vid_hdr
);
852 leb_write_unlock(ubi
, vol_id
, lnum
);
853 up_read(&ubi
->fm_eba_sem
);
857 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
858 len
, vol_id
, lnum
, pnum
, used_ebs
);
860 err
= ubi_io_write_vid_hdr(ubi
, pnum
, vid_hdr
);
862 ubi_warn(ubi
, "failed to write VID header to LEB %d:%d, PEB %d",
864 up_read(&ubi
->fm_eba_sem
);
868 err
= ubi_io_write_data(ubi
, buf
, pnum
, 0, len
);
870 ubi_warn(ubi
, "failed to write %d bytes of data to PEB %d",
872 up_read(&ubi
->fm_eba_sem
);
876 ubi_assert(vol
->eba_tbl
[lnum
] < 0);
877 vol
->eba_tbl
[lnum
] = pnum
;
878 up_read(&ubi
->fm_eba_sem
);
880 leb_write_unlock(ubi
, vol_id
, lnum
);
881 ubi_free_vid_hdr(ubi
, vid_hdr
);
885 if (err
!= -EIO
|| !ubi
->bad_allowed
) {
887 * This flash device does not admit of bad eraseblocks or
888 * something nasty and unexpected happened. Switch to read-only
892 leb_write_unlock(ubi
, vol_id
, lnum
);
893 ubi_free_vid_hdr(ubi
, vid_hdr
);
897 err
= ubi_wl_put_peb(ubi
, vol_id
, lnum
, pnum
, 1);
898 if (err
|| ++tries
> UBI_IO_RETRIES
) {
900 leb_write_unlock(ubi
, vol_id
, lnum
);
901 ubi_free_vid_hdr(ubi
, vid_hdr
);
905 vid_hdr
->sqnum
= cpu_to_be64(ubi_next_sqnum(ubi
));
906 ubi_msg(ubi
, "try another PEB");
911 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
912 * @ubi: UBI device description object
913 * @vol: volume description object
914 * @lnum: logical eraseblock number
915 * @buf: data to write
916 * @len: how many bytes to write
918 * This function changes the contents of a logical eraseblock atomically. @buf
919 * has to contain new logical eraseblock data, and @len - the length of the
920 * data, which has to be aligned. This function guarantees that in case of an
921 * unclean reboot the old contents is preserved. Returns zero in case of
922 * success and a negative error code in case of failure.
924 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
925 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
927 int ubi_eba_atomic_leb_change(struct ubi_device
*ubi
, struct ubi_volume
*vol
,
928 int lnum
, const void *buf
, int len
)
930 int err
, pnum
, old_pnum
, tries
= 0, vol_id
= vol
->vol_id
;
931 struct ubi_vid_hdr
*vid_hdr
;
939 * Special case when data length is zero. In this case the LEB
940 * has to be unmapped and mapped somewhere else.
942 err
= ubi_eba_unmap_leb(ubi
, vol
, lnum
);
945 return ubi_eba_write_leb(ubi
, vol
, lnum
, NULL
, 0, 0);
948 vid_hdr
= ubi_zalloc_vid_hdr(ubi
, GFP_NOFS
);
952 mutex_lock(&ubi
->alc_mutex
);
953 err
= leb_write_lock(ubi
, vol_id
, lnum
);
957 vid_hdr
->sqnum
= cpu_to_be64(ubi_next_sqnum(ubi
));
958 vid_hdr
->vol_id
= cpu_to_be32(vol_id
);
959 vid_hdr
->lnum
= cpu_to_be32(lnum
);
960 vid_hdr
->compat
= ubi_get_compat(ubi
, vol_id
);
961 vid_hdr
->data_pad
= cpu_to_be32(vol
->data_pad
);
963 crc
= crc32(UBI_CRC32_INIT
, buf
, len
);
964 vid_hdr
->vol_type
= UBI_VID_DYNAMIC
;
965 vid_hdr
->data_size
= cpu_to_be32(len
);
966 vid_hdr
->copy_flag
= 1;
967 vid_hdr
->data_crc
= cpu_to_be32(crc
);
970 pnum
= ubi_wl_get_peb(ubi
);
973 up_read(&ubi
->fm_eba_sem
);
977 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
978 vol_id
, lnum
, vol
->eba_tbl
[lnum
], pnum
);
980 err
= ubi_io_write_vid_hdr(ubi
, pnum
, vid_hdr
);
982 ubi_warn(ubi
, "failed to write VID header to LEB %d:%d, PEB %d",
984 up_read(&ubi
->fm_eba_sem
);
988 err
= ubi_io_write_data(ubi
, buf
, pnum
, 0, len
);
990 ubi_warn(ubi
, "failed to write %d bytes of data to PEB %d",
992 up_read(&ubi
->fm_eba_sem
);
996 old_pnum
= vol
->eba_tbl
[lnum
];
997 vol
->eba_tbl
[lnum
] = pnum
;
998 up_read(&ubi
->fm_eba_sem
);
1000 if (old_pnum
>= 0) {
1001 err
= ubi_wl_put_peb(ubi
, vol_id
, lnum
, old_pnum
, 0);
1003 goto out_leb_unlock
;
1007 leb_write_unlock(ubi
, vol_id
, lnum
);
1009 mutex_unlock(&ubi
->alc_mutex
);
1010 ubi_free_vid_hdr(ubi
, vid_hdr
);
1014 if (err
!= -EIO
|| !ubi
->bad_allowed
) {
1016 * This flash device does not admit of bad eraseblocks or
1017 * something nasty and unexpected happened. Switch to read-only
1018 * mode just in case.
1021 goto out_leb_unlock
;
1024 err
= ubi_wl_put_peb(ubi
, vol_id
, lnum
, pnum
, 1);
1025 if (err
|| ++tries
> UBI_IO_RETRIES
) {
1027 goto out_leb_unlock
;
1030 vid_hdr
->sqnum
= cpu_to_be64(ubi_next_sqnum(ubi
));
1031 ubi_msg(ubi
, "try another PEB");
1036 * is_error_sane - check whether a read error is sane.
1037 * @err: code of the error happened during reading
1039 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1040 * cannot read data from the target PEB (an error @err happened). If the error
1041 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1042 * fatal and UBI will be switched to R/O mode later.
1044 * The idea is that we try not to switch to R/O mode if the read error is
1045 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1046 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1047 * mode, simply because we do not know what happened at the MTD level, and we
1048 * cannot handle this. E.g., the underlying driver may have become crazy, and
1049 * it is safer to switch to R/O mode to preserve the data.
1051 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1052 * which we have just written.
1054 static int is_error_sane(int err
)
1056 if (err
== -EIO
|| err
== -ENOMEM
|| err
== UBI_IO_BAD_HDR
||
1057 err
== UBI_IO_BAD_HDR_EBADMSG
|| err
== -ETIMEDOUT
)
1063 * ubi_eba_copy_leb - copy logical eraseblock.
1064 * @ubi: UBI device description object
1065 * @from: physical eraseblock number from where to copy
1066 * @to: physical eraseblock number where to copy
1067 * @vid_hdr: VID header of the @from physical eraseblock
1069 * This function copies logical eraseblock from physical eraseblock @from to
1070 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1071 * function. Returns:
1072 * o %0 in case of success;
1073 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1074 * o a negative error code in case of failure.
1076 int ubi_eba_copy_leb(struct ubi_device
*ubi
, int from
, int to
,
1077 struct ubi_vid_hdr
*vid_hdr
)
1079 int err
, vol_id
, lnum
, data_size
, aldata_size
, idx
;
1080 struct ubi_volume
*vol
;
1083 vol_id
= be32_to_cpu(vid_hdr
->vol_id
);
1084 lnum
= be32_to_cpu(vid_hdr
->lnum
);
1086 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id
, lnum
, from
, to
);
1088 if (vid_hdr
->vol_type
== UBI_VID_STATIC
) {
1089 data_size
= be32_to_cpu(vid_hdr
->data_size
);
1090 aldata_size
= ALIGN(data_size
, ubi
->min_io_size
);
1092 data_size
= aldata_size
=
1093 ubi
->leb_size
- be32_to_cpu(vid_hdr
->data_pad
);
1095 idx
= vol_id2idx(ubi
, vol_id
);
1096 spin_lock(&ubi
->volumes_lock
);
1098 * Note, we may race with volume deletion, which means that the volume
1099 * this logical eraseblock belongs to might be being deleted. Since the
1100 * volume deletion un-maps all the volume's logical eraseblocks, it will
1101 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1103 vol
= ubi
->volumes
[idx
];
1104 spin_unlock(&ubi
->volumes_lock
);
1106 /* No need to do further work, cancel */
1107 dbg_wl("volume %d is being removed, cancel", vol_id
);
1108 return MOVE_CANCEL_RACE
;
1112 * We do not want anybody to write to this logical eraseblock while we
1113 * are moving it, so lock it.
1115 * Note, we are using non-waiting locking here, because we cannot sleep
1116 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1117 * unmapping the LEB which is mapped to the PEB we are going to move
1118 * (@from). This task locks the LEB and goes sleep in the
1119 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1120 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1121 * LEB is already locked, we just do not move it and return
1122 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1123 * we do not know the reasons of the contention - it may be just a
1124 * normal I/O on this LEB, so we want to re-try.
1126 err
= leb_write_trylock(ubi
, vol_id
, lnum
);
1128 dbg_wl("contention on LEB %d:%d, cancel", vol_id
, lnum
);
1133 * The LEB might have been put meanwhile, and the task which put it is
1134 * probably waiting on @ubi->move_mutex. No need to continue the work,
1137 if (vol
->eba_tbl
[lnum
] != from
) {
1138 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1139 vol_id
, lnum
, from
, vol
->eba_tbl
[lnum
]);
1140 err
= MOVE_CANCEL_RACE
;
1141 goto out_unlock_leb
;
1145 * OK, now the LEB is locked and we can safely start moving it. Since
1146 * this function utilizes the @ubi->peb_buf buffer which is shared
1147 * with some other functions - we lock the buffer by taking the
1150 mutex_lock(&ubi
->buf_mutex
);
1151 dbg_wl("read %d bytes of data", aldata_size
);
1152 err
= ubi_io_read_data(ubi
, ubi
->peb_buf
, from
, 0, aldata_size
);
1153 if (err
&& err
!= UBI_IO_BITFLIPS
) {
1154 ubi_warn(ubi
, "error %d while reading data from PEB %d",
1156 err
= MOVE_SOURCE_RD_ERR
;
1157 goto out_unlock_buf
;
1161 * Now we have got to calculate how much data we have to copy. In
1162 * case of a static volume it is fairly easy - the VID header contains
1163 * the data size. In case of a dynamic volume it is more difficult - we
1164 * have to read the contents, cut 0xFF bytes from the end and copy only
1165 * the first part. We must do this to avoid writing 0xFF bytes as it
1166 * may have some side-effects. And not only this. It is important not
1167 * to include those 0xFFs to CRC because later the they may be filled
1170 if (vid_hdr
->vol_type
== UBI_VID_DYNAMIC
)
1171 aldata_size
= data_size
=
1172 ubi_calc_data_len(ubi
, ubi
->peb_buf
, data_size
);
1175 crc
= crc32(UBI_CRC32_INIT
, ubi
->peb_buf
, data_size
);
1179 * It may turn out to be that the whole @from physical eraseblock
1180 * contains only 0xFF bytes. Then we have to only write the VID header
1181 * and do not write any data. This also means we should not set
1182 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1184 if (data_size
> 0) {
1185 vid_hdr
->copy_flag
= 1;
1186 vid_hdr
->data_size
= cpu_to_be32(data_size
);
1187 vid_hdr
->data_crc
= cpu_to_be32(crc
);
1189 vid_hdr
->sqnum
= cpu_to_be64(ubi_next_sqnum(ubi
));
1191 err
= ubi_io_write_vid_hdr(ubi
, to
, vid_hdr
);
1194 err
= MOVE_TARGET_WR_ERR
;
1195 goto out_unlock_buf
;
1200 /* Read the VID header back and check if it was written correctly */
1201 err
= ubi_io_read_vid_hdr(ubi
, to
, vid_hdr
, 1);
1203 if (err
!= UBI_IO_BITFLIPS
) {
1204 ubi_warn(ubi
, "error %d while reading VID header back from PEB %d",
1206 if (is_error_sane(err
))
1207 err
= MOVE_TARGET_RD_ERR
;
1209 err
= MOVE_TARGET_BITFLIPS
;
1210 goto out_unlock_buf
;
1213 if (data_size
> 0) {
1214 err
= ubi_io_write_data(ubi
, ubi
->peb_buf
, to
, 0, aldata_size
);
1217 err
= MOVE_TARGET_WR_ERR
;
1218 goto out_unlock_buf
;
1224 ubi_assert(vol
->eba_tbl
[lnum
] == from
);
1225 down_read(&ubi
->fm_eba_sem
);
1226 vol
->eba_tbl
[lnum
] = to
;
1227 up_read(&ubi
->fm_eba_sem
);
1230 mutex_unlock(&ubi
->buf_mutex
);
1232 leb_write_unlock(ubi
, vol_id
, lnum
);
1237 * print_rsvd_warning - warn about not having enough reserved PEBs.
1238 * @ubi: UBI device description object
1240 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1241 * cannot reserve enough PEBs for bad block handling. This function makes a
1242 * decision whether we have to print a warning or not. The algorithm is as
1244 * o if this is a new UBI image, then just print the warning
1245 * o if this is an UBI image which has already been used for some time, print
1246 * a warning only if we can reserve less than 10% of the expected amount of
1249 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1250 * of PEBs becomes smaller, which is normal and we do not want to scare users
1251 * with a warning every time they attach the MTD device. This was an issue
1252 * reported by real users.
1254 static void print_rsvd_warning(struct ubi_device
*ubi
,
1255 struct ubi_attach_info
*ai
)
1258 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1259 * large number to distinguish between newly flashed and used images.
1261 if (ai
->max_sqnum
> (1 << 18)) {
1262 int min
= ubi
->beb_rsvd_level
/ 10;
1266 if (ubi
->beb_rsvd_pebs
> min
)
1270 ubi_warn(ubi
, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1271 ubi
->beb_rsvd_pebs
, ubi
->beb_rsvd_level
);
1272 if (ubi
->corr_peb_count
)
1273 ubi_warn(ubi
, "%d PEBs are corrupted and not used",
1274 ubi
->corr_peb_count
);
1278 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1279 * @ubi: UBI device description object
1280 * @ai_fastmap: UBI attach info object created by fastmap
1281 * @ai_scan: UBI attach info object created by scanning
1283 * Returns < 0 in case of an internal error, 0 otherwise.
1284 * If a bad EBA table entry was found it will be printed out and
1285 * ubi_assert() triggers.
1287 int self_check_eba(struct ubi_device
*ubi
, struct ubi_attach_info
*ai_fastmap
,
1288 struct ubi_attach_info
*ai_scan
)
1290 int i
, j
, num_volumes
, ret
= 0;
1291 int **scan_eba
, **fm_eba
;
1292 struct ubi_ainf_volume
*av
;
1293 struct ubi_volume
*vol
;
1294 struct ubi_ainf_peb
*aeb
;
1297 num_volumes
= ubi
->vtbl_slots
+ UBI_INT_VOL_COUNT
;
1299 scan_eba
= kmalloc(sizeof(*scan_eba
) * num_volumes
, GFP_KERNEL
);
1303 fm_eba
= kmalloc(sizeof(*fm_eba
) * num_volumes
, GFP_KERNEL
);
1309 for (i
= 0; i
< num_volumes
; i
++) {
1310 vol
= ubi
->volumes
[i
];
1314 scan_eba
[i
] = kmalloc(vol
->reserved_pebs
* sizeof(**scan_eba
),
1321 fm_eba
[i
] = kmalloc(vol
->reserved_pebs
* sizeof(**fm_eba
),
1328 for (j
= 0; j
< vol
->reserved_pebs
; j
++)
1329 scan_eba
[i
][j
] = fm_eba
[i
][j
] = UBI_LEB_UNMAPPED
;
1331 av
= ubi_find_av(ai_scan
, idx2vol_id(ubi
, i
));
1335 ubi_rb_for_each_entry(rb
, aeb
, &av
->root
, u
.rb
)
1336 scan_eba
[i
][aeb
->lnum
] = aeb
->pnum
;
1338 av
= ubi_find_av(ai_fastmap
, idx2vol_id(ubi
, i
));
1342 ubi_rb_for_each_entry(rb
, aeb
, &av
->root
, u
.rb
)
1343 fm_eba
[i
][aeb
->lnum
] = aeb
->pnum
;
1345 for (j
= 0; j
< vol
->reserved_pebs
; j
++) {
1346 if (scan_eba
[i
][j
] != fm_eba
[i
][j
]) {
1347 if (scan_eba
[i
][j
] == UBI_LEB_UNMAPPED
||
1348 fm_eba
[i
][j
] == UBI_LEB_UNMAPPED
)
1351 ubi_err(ubi
, "LEB:%i:%i is PEB:%i instead of %i!",
1352 vol
->vol_id
, j
, fm_eba
[i
][j
],
1360 for (i
= 0; i
< num_volumes
; i
++) {
1361 if (!ubi
->volumes
[i
])
1374 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1375 * @ubi: UBI device description object
1376 * @ai: attaching information
1378 * This function returns zero in case of success and a negative error code in
1381 int ubi_eba_init(struct ubi_device
*ubi
, struct ubi_attach_info
*ai
)
1383 int i
, j
, err
, num_volumes
;
1384 struct ubi_ainf_volume
*av
;
1385 struct ubi_volume
*vol
;
1386 struct ubi_ainf_peb
*aeb
;
1389 dbg_eba("initialize EBA sub-system");
1391 spin_lock_init(&ubi
->ltree_lock
);
1392 mutex_init(&ubi
->alc_mutex
);
1393 ubi
->ltree
= RB_ROOT
;
1395 ubi
->global_sqnum
= ai
->max_sqnum
+ 1;
1396 num_volumes
= ubi
->vtbl_slots
+ UBI_INT_VOL_COUNT
;
1398 for (i
= 0; i
< num_volumes
; i
++) {
1399 vol
= ubi
->volumes
[i
];
1405 vol
->eba_tbl
= kmalloc(vol
->reserved_pebs
* sizeof(int),
1407 if (!vol
->eba_tbl
) {
1412 for (j
= 0; j
< vol
->reserved_pebs
; j
++)
1413 vol
->eba_tbl
[j
] = UBI_LEB_UNMAPPED
;
1415 av
= ubi_find_av(ai
, idx2vol_id(ubi
, i
));
1419 ubi_rb_for_each_entry(rb
, aeb
, &av
->root
, u
.rb
) {
1420 if (aeb
->lnum
>= vol
->reserved_pebs
)
1422 * This may happen in case of an unclean reboot
1425 ubi_move_aeb_to_list(av
, aeb
, &ai
->erase
);
1427 vol
->eba_tbl
[aeb
->lnum
] = aeb
->pnum
;
1431 if (ubi
->avail_pebs
< EBA_RESERVED_PEBS
) {
1432 ubi_err(ubi
, "no enough physical eraseblocks (%d, need %d)",
1433 ubi
->avail_pebs
, EBA_RESERVED_PEBS
);
1434 if (ubi
->corr_peb_count
)
1435 ubi_err(ubi
, "%d PEBs are corrupted and not used",
1436 ubi
->corr_peb_count
);
1440 ubi
->avail_pebs
-= EBA_RESERVED_PEBS
;
1441 ubi
->rsvd_pebs
+= EBA_RESERVED_PEBS
;
1443 if (ubi
->bad_allowed
) {
1444 ubi_calculate_reserved(ubi
);
1446 if (ubi
->avail_pebs
< ubi
->beb_rsvd_level
) {
1447 /* No enough free physical eraseblocks */
1448 ubi
->beb_rsvd_pebs
= ubi
->avail_pebs
;
1449 print_rsvd_warning(ubi
, ai
);
1451 ubi
->beb_rsvd_pebs
= ubi
->beb_rsvd_level
;
1453 ubi
->avail_pebs
-= ubi
->beb_rsvd_pebs
;
1454 ubi
->rsvd_pebs
+= ubi
->beb_rsvd_pebs
;
1457 dbg_eba("EBA sub-system is initialized");
1461 for (i
= 0; i
< num_volumes
; i
++) {
1462 if (!ubi
->volumes
[i
])
1464 kfree(ubi
->volumes
[i
]->eba_tbl
);
1465 ubi
->volumes
[i
]->eba_tbl
= NULL
;