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_is_mapped - check if a LEB is mapped.
316 * @vol: volume description object
317 * @lnum: logical eraseblock number
319 * This function returns true if the LEB is mapped, false otherwise.
321 bool ubi_eba_is_mapped(struct ubi_volume *vol, int lnum)
323 return vol->eba_tbl[lnum] >= 0;
327 * ubi_eba_unmap_leb - un-map logical eraseblock.
328 * @ubi: UBI device description object
329 * @vol: volume description object
330 * @lnum: logical eraseblock number
332 * This function un-maps logical eraseblock @lnum and schedules corresponding
333 * physical eraseblock for erasure. Returns zero in case of success and a
334 * negative error code in case of failure.
336 int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
339 int err, pnum, vol_id = vol->vol_id;
344 err = leb_write_lock(ubi, vol_id, lnum);
348 pnum = vol->eba_tbl[lnum];
350 /* This logical eraseblock is already unmapped */
353 dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
355 down_read(&ubi->fm_eba_sem);
356 vol->eba_tbl[lnum] = UBI_LEB_UNMAPPED;
357 up_read(&ubi->fm_eba_sem);
358 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
361 leb_write_unlock(ubi, vol_id, lnum);
366 * ubi_eba_read_leb - read data.
367 * @ubi: UBI device description object
368 * @vol: volume description object
369 * @lnum: logical eraseblock number
370 * @buf: buffer to store the read data
371 * @offset: offset from where to read
372 * @len: how many bytes to read
373 * @check: data CRC check flag
375 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
376 * bytes. The @check flag only makes sense for static volumes and forces
377 * eraseblock data CRC checking.
379 * In case of success this function returns zero. In case of a static volume,
380 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
381 * returned for any volume type if an ECC error was detected by the MTD device
382 * driver. Other negative error cored may be returned in case of other errors.
384 int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
385 void *buf, int offset, int len, int check)
387 int err, pnum, scrub = 0, vol_id = vol->vol_id;
388 struct ubi_vid_hdr *vid_hdr;
389 uint32_t uninitialized_var(crc);
391 err = leb_read_lock(ubi, vol_id, lnum);
395 pnum = vol->eba_tbl[lnum];
398 * The logical eraseblock is not mapped, fill the whole buffer
399 * with 0xFF bytes. The exception is static volumes for which
400 * it is an error to read unmapped logical eraseblocks.
402 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
403 len, offset, vol_id, lnum);
404 leb_read_unlock(ubi, vol_id, lnum);
405 ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
406 memset(buf, 0xFF, len);
410 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
411 len, offset, vol_id, lnum, pnum);
413 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
418 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
424 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
425 if (err && err != UBI_IO_BITFLIPS) {
428 * The header is either absent or corrupted.
429 * The former case means there is a bug -
430 * switch to read-only mode just in case.
431 * The latter case means a real corruption - we
432 * may try to recover data. FIXME: but this is
435 if (err == UBI_IO_BAD_HDR_EBADMSG ||
436 err == UBI_IO_BAD_HDR) {
437 ubi_warn(ubi, "corrupted VID header at PEB %d, LEB %d:%d",
442 * Ending up here in the non-Fastmap case
443 * is a clear bug as the VID header had to
444 * be present at scan time to have it referenced.
445 * With fastmap the story is more complicated.
446 * Fastmap has the mapping info without the need
447 * of a full scan. So the LEB could have been
448 * unmapped, Fastmap cannot know this and keeps
449 * the LEB referenced.
450 * This is valid and works as the layer above UBI
451 * has to do bookkeeping about used/referenced
454 if (ubi->fast_attach) {
463 } else if (err == UBI_IO_BITFLIPS)
466 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
467 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
469 crc = be32_to_cpu(vid_hdr->data_crc);
470 ubi_free_vid_hdr(ubi, vid_hdr);
473 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
475 if (err == UBI_IO_BITFLIPS)
477 else if (mtd_is_eccerr(err)) {
478 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
482 ubi_msg(ubi, "force data checking");
491 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
493 ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
501 err = ubi_wl_scrub_peb(ubi, pnum);
503 leb_read_unlock(ubi, vol_id, lnum);
507 ubi_free_vid_hdr(ubi, vid_hdr);
509 leb_read_unlock(ubi, vol_id, lnum);
514 * ubi_eba_read_leb_sg - read data into a scatter gather list.
515 * @ubi: UBI device description object
516 * @vol: volume description object
517 * @lnum: logical eraseblock number
518 * @sgl: UBI scatter gather list to store the read data
519 * @offset: offset from where to read
520 * @len: how many bytes to read
521 * @check: data CRC check flag
523 * This function works exactly like ubi_eba_read_leb(). But instead of
524 * storing the read data into a buffer it writes to an UBI scatter gather
527 int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
528 struct ubi_sgl *sgl, int lnum, int offset, int len,
533 struct scatterlist *sg;
536 ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
537 sg = &sgl->sg[sgl->list_pos];
538 if (len < sg->length - sgl->page_pos)
541 to_read = sg->length - sgl->page_pos;
543 ret = ubi_eba_read_leb(ubi, vol, lnum,
544 sg_virt(sg) + sgl->page_pos, offset,
552 sgl->page_pos += to_read;
553 if (sgl->page_pos == sg->length) {
569 * try_recover_peb - try to recover from write failure.
570 * @vol: volume description object
571 * @pnum: the physical eraseblock to recover
572 * @lnum: logical eraseblock number
573 * @buf: data which was not written because of the write failure
574 * @offset: offset of the failed write
575 * @len: how many bytes should have been written
577 * @retry: whether the caller should retry in case of failure
579 * This function is called in case of a write failure and moves all good data
580 * from the potentially bad physical eraseblock to a good physical eraseblock.
581 * This function also writes the data which was not written due to the failure.
582 * Returns 0 in case of success, and a negative error code in case of failure.
583 * In case of failure, the %retry parameter is set to false if this is a fatal
584 * error (retrying won't help), and true otherwise.
586 static int try_recover_peb(struct ubi_volume *vol, int pnum, int lnum,
587 const void *buf, int offset, int len,
588 struct ubi_vid_hdr *vid_hdr, bool *retry)
590 struct ubi_device *ubi = vol->ubi;
591 int new_pnum, err, vol_id = vol->vol_id, data_size;
596 new_pnum = ubi_wl_get_peb(ubi);
602 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
605 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
606 if (err && err != UBI_IO_BITFLIPS) {
612 ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
614 mutex_lock(&ubi->buf_mutex);
615 memset(ubi->peb_buf + offset, 0xFF, len);
617 /* Read everything before the area where the write failure happened */
619 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
620 if (err && err != UBI_IO_BITFLIPS)
626 memcpy(ubi->peb_buf + offset, buf, len);
628 data_size = offset + len;
629 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
630 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
631 vid_hdr->copy_flag = 1;
632 vid_hdr->data_size = cpu_to_be32(data_size);
633 vid_hdr->data_crc = cpu_to_be32(crc);
634 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
638 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
641 mutex_unlock(&ubi->buf_mutex);
644 vol->eba_tbl[lnum] = new_pnum;
647 up_read(&ubi->fm_eba_sem);
650 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
651 ubi_msg(ubi, "data was successfully recovered");
652 } else if (new_pnum >= 0) {
654 * Bad luck? This physical eraseblock is bad too? Crud. Let's
655 * try to get another one.
657 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
658 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
665 * recover_peb - recover from write failure.
666 * @ubi: UBI device description object
667 * @pnum: the physical eraseblock to recover
669 * @lnum: logical eraseblock number
670 * @buf: data which was not written because of the write failure
671 * @offset: offset of the failed write
672 * @len: how many bytes should have been written
674 * This function is called in case of a write failure and moves all good data
675 * from the potentially bad physical eraseblock to a good physical eraseblock.
676 * This function also writes the data which was not written due to the failure.
677 * Returns 0 in case of success, and a negative error code in case of failure.
678 * This function tries %UBI_IO_RETRIES before giving up.
680 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
681 const void *buf, int offset, int len)
683 int err, idx = vol_id2idx(ubi, vol_id), tries;
684 struct ubi_volume *vol = ubi->volumes[idx];
685 struct ubi_vid_hdr *vid_hdr;
687 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
691 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
694 err = try_recover_peb(vol, pnum, lnum, buf, offset, len,
699 ubi_msg(ubi, "try again");
702 ubi_free_vid_hdr(ubi, vid_hdr);
708 * try_write_vid_and_data - try to write VID header and data to a new PEB.
709 * @vol: volume description object
710 * @lnum: logical eraseblock number
711 * @vid_hdr: VID header to write
712 * @buf: buffer containing the data
713 * @offset: where to start writing data
714 * @len: how many bytes should be written
716 * This function tries to write VID header and data belonging to logical
717 * eraseblock @lnum of volume @vol to a new physical eraseblock. Returns zero
718 * in case of success and a negative error code in case of failure.
719 * In case of error, it is possible that something was still written to the
720 * flash media, but may be some garbage.
722 static int try_write_vid_and_data(struct ubi_volume *vol, int lnum,
723 struct ubi_vid_hdr *vid_hdr, const void *buf,
726 struct ubi_device *ubi = vol->ubi;
727 int pnum, opnum, err, vol_id = vol->vol_id;
729 pnum = ubi_wl_get_peb(ubi);
735 opnum = vol->eba_tbl[lnum];
737 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
738 len, offset, vol_id, lnum, pnum);
740 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
742 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
748 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
751 "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
752 len, offset, vol_id, lnum, pnum);
757 vol->eba_tbl[lnum] = pnum;
760 up_read(&ubi->fm_eba_sem);
762 if (err && pnum >= 0)
763 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
764 else if (!err && opnum >= 0)
765 err = ubi_wl_put_peb(ubi, vol_id, lnum, opnum, 0);
771 * ubi_eba_write_leb - write data to dynamic volume.
772 * @ubi: UBI device description object
773 * @vol: volume description object
774 * @lnum: logical eraseblock number
775 * @buf: the data to write
776 * @offset: offset within the logical eraseblock where to write
777 * @len: how many bytes to write
779 * This function writes data to logical eraseblock @lnum of a dynamic volume
780 * @vol. Returns zero in case of success and a negative error code in case
781 * of failure. In case of error, it is possible that something was still
782 * written to the flash media, but may be some garbage.
783 * This function retries %UBI_IO_RETRIES times before giving up.
785 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
786 const void *buf, int offset, int len)
788 int err, pnum, tries, vol_id = vol->vol_id;
789 struct ubi_vid_hdr *vid_hdr;
794 err = leb_write_lock(ubi, vol_id, lnum);
798 pnum = vol->eba_tbl[lnum];
800 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
801 len, offset, vol_id, lnum, pnum);
803 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
805 ubi_warn(ubi, "failed to write data to PEB %d", pnum);
806 if (err == -EIO && ubi->bad_allowed)
807 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
815 * The logical eraseblock is not mapped. We have to get a free physical
816 * eraseblock and write the volume identifier header there first.
818 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
820 leb_write_unlock(ubi, vol_id, lnum);
824 vid_hdr->vol_type = UBI_VID_DYNAMIC;
825 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
826 vid_hdr->vol_id = cpu_to_be32(vol_id);
827 vid_hdr->lnum = cpu_to_be32(lnum);
828 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
829 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
831 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
832 err = try_write_vid_and_data(vol, lnum, vid_hdr, buf, offset,
834 if (err != -EIO || !ubi->bad_allowed)
838 * Fortunately, this is the first write operation to this
839 * physical eraseblock, so just put it and request a new one.
840 * We assume that if this physical eraseblock went bad, the
841 * erase code will handle that.
843 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
844 ubi_msg(ubi, "try another PEB");
847 ubi_free_vid_hdr(ubi, vid_hdr);
853 leb_write_unlock(ubi, vol_id, lnum);
859 * ubi_eba_write_leb_st - write data to static volume.
860 * @ubi: UBI device description object
861 * @vol: volume description object
862 * @lnum: logical eraseblock number
863 * @buf: data to write
864 * @len: how many bytes to write
865 * @used_ebs: how many logical eraseblocks will this volume contain
867 * This function writes data to logical eraseblock @lnum of static volume
868 * @vol. The @used_ebs argument should contain total number of logical
869 * eraseblock in this static volume.
871 * When writing to the last logical eraseblock, the @len argument doesn't have
872 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
873 * to the real data size, although the @buf buffer has to contain the
874 * alignment. In all other cases, @len has to be aligned.
876 * It is prohibited to write more than once to logical eraseblocks of static
877 * volumes. This function returns zero in case of success and a negative error
878 * code in case of failure.
880 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
881 int lnum, const void *buf, int len, int used_ebs)
883 int err, tries, data_size = len, vol_id = vol->vol_id;
884 struct ubi_vid_hdr *vid_hdr;
890 if (lnum == used_ebs - 1)
891 /* If this is the last LEB @len may be unaligned */
892 len = ALIGN(data_size, ubi->min_io_size);
894 ubi_assert(!(len & (ubi->min_io_size - 1)));
896 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
900 err = leb_write_lock(ubi, vol_id, lnum);
904 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
905 vid_hdr->vol_id = cpu_to_be32(vol_id);
906 vid_hdr->lnum = cpu_to_be32(lnum);
907 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
908 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
910 crc = crc32(UBI_CRC32_INIT, buf, data_size);
911 vid_hdr->vol_type = UBI_VID_STATIC;
912 vid_hdr->data_size = cpu_to_be32(data_size);
913 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
914 vid_hdr->data_crc = cpu_to_be32(crc);
916 ubi_assert(vol->eba_tbl[lnum] < 0);
918 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
919 err = try_write_vid_and_data(vol, lnum, vid_hdr, buf, 0, len);
920 if (err != -EIO || !ubi->bad_allowed)
923 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
924 ubi_msg(ubi, "try another PEB");
930 leb_write_unlock(ubi, vol_id, lnum);
933 ubi_free_vid_hdr(ubi, vid_hdr);
939 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
940 * @ubi: UBI device description object
941 * @vol: volume description object
942 * @lnum: logical eraseblock number
943 * @buf: data to write
944 * @len: how many bytes to write
946 * This function changes the contents of a logical eraseblock atomically. @buf
947 * has to contain new logical eraseblock data, and @len - the length of the
948 * data, which has to be aligned. This function guarantees that in case of an
949 * unclean reboot the old contents is preserved. Returns zero in case of
950 * success and a negative error code in case of failure.
952 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
953 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
955 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
956 int lnum, const void *buf, int len)
958 int err, tries, vol_id = vol->vol_id;
959 struct ubi_vid_hdr *vid_hdr;
967 * Special case when data length is zero. In this case the LEB
968 * has to be unmapped and mapped somewhere else.
970 err = ubi_eba_unmap_leb(ubi, vol, lnum);
973 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
976 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
980 mutex_lock(&ubi->alc_mutex);
981 err = leb_write_lock(ubi, vol_id, lnum);
985 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
986 vid_hdr->vol_id = cpu_to_be32(vol_id);
987 vid_hdr->lnum = cpu_to_be32(lnum);
988 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
989 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
991 crc = crc32(UBI_CRC32_INIT, buf, len);
992 vid_hdr->vol_type = UBI_VID_DYNAMIC;
993 vid_hdr->data_size = cpu_to_be32(len);
994 vid_hdr->copy_flag = 1;
995 vid_hdr->data_crc = cpu_to_be32(crc);
997 dbg_eba("change LEB %d:%d", vol_id, lnum);
999 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1000 err = try_write_vid_and_data(vol, lnum, vid_hdr, buf, 0, len);
1001 if (err != -EIO || !ubi->bad_allowed)
1004 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1005 ubi_msg(ubi, "try another PEB");
1009 * This flash device does not admit of bad eraseblocks or
1010 * something nasty and unexpected happened. Switch to read-only
1011 * mode just in case.
1016 leb_write_unlock(ubi, vol_id, lnum);
1019 mutex_unlock(&ubi->alc_mutex);
1020 ubi_free_vid_hdr(ubi, vid_hdr);
1025 * is_error_sane - check whether a read error is sane.
1026 * @err: code of the error happened during reading
1028 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1029 * cannot read data from the target PEB (an error @err happened). If the error
1030 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1031 * fatal and UBI will be switched to R/O mode later.
1033 * The idea is that we try not to switch to R/O mode if the read error is
1034 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1035 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1036 * mode, simply because we do not know what happened at the MTD level, and we
1037 * cannot handle this. E.g., the underlying driver may have become crazy, and
1038 * it is safer to switch to R/O mode to preserve the data.
1040 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1041 * which we have just written.
1043 static int is_error_sane(int err)
1045 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1046 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1052 * ubi_eba_copy_leb - copy logical eraseblock.
1053 * @ubi: UBI device description object
1054 * @from: physical eraseblock number from where to copy
1055 * @to: physical eraseblock number where to copy
1056 * @vid_hdr: VID header of the @from physical eraseblock
1058 * This function copies logical eraseblock from physical eraseblock @from to
1059 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1060 * function. Returns:
1061 * o %0 in case of success;
1062 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1063 * o a negative error code in case of failure.
1065 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1066 struct ubi_vid_hdr *vid_hdr)
1068 int err, vol_id, lnum, data_size, aldata_size, idx;
1069 struct ubi_volume *vol;
1072 vol_id = be32_to_cpu(vid_hdr->vol_id);
1073 lnum = be32_to_cpu(vid_hdr->lnum);
1075 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1077 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1078 data_size = be32_to_cpu(vid_hdr->data_size);
1079 aldata_size = ALIGN(data_size, ubi->min_io_size);
1081 data_size = aldata_size =
1082 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1084 idx = vol_id2idx(ubi, vol_id);
1085 spin_lock(&ubi->volumes_lock);
1087 * Note, we may race with volume deletion, which means that the volume
1088 * this logical eraseblock belongs to might be being deleted. Since the
1089 * volume deletion un-maps all the volume's logical eraseblocks, it will
1090 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1092 vol = ubi->volumes[idx];
1093 spin_unlock(&ubi->volumes_lock);
1095 /* No need to do further work, cancel */
1096 dbg_wl("volume %d is being removed, cancel", vol_id);
1097 return MOVE_CANCEL_RACE;
1101 * We do not want anybody to write to this logical eraseblock while we
1102 * are moving it, so lock it.
1104 * Note, we are using non-waiting locking here, because we cannot sleep
1105 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1106 * unmapping the LEB which is mapped to the PEB we are going to move
1107 * (@from). This task locks the LEB and goes sleep in the
1108 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1109 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1110 * LEB is already locked, we just do not move it and return
1111 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1112 * we do not know the reasons of the contention - it may be just a
1113 * normal I/O on this LEB, so we want to re-try.
1115 err = leb_write_trylock(ubi, vol_id, lnum);
1117 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1122 * The LEB might have been put meanwhile, and the task which put it is
1123 * probably waiting on @ubi->move_mutex. No need to continue the work,
1126 if (vol->eba_tbl[lnum] != from) {
1127 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1128 vol_id, lnum, from, vol->eba_tbl[lnum]);
1129 err = MOVE_CANCEL_RACE;
1130 goto out_unlock_leb;
1134 * OK, now the LEB is locked and we can safely start moving it. Since
1135 * this function utilizes the @ubi->peb_buf buffer which is shared
1136 * with some other functions - we lock the buffer by taking the
1139 mutex_lock(&ubi->buf_mutex);
1140 dbg_wl("read %d bytes of data", aldata_size);
1141 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1142 if (err && err != UBI_IO_BITFLIPS) {
1143 ubi_warn(ubi, "error %d while reading data from PEB %d",
1145 err = MOVE_SOURCE_RD_ERR;
1146 goto out_unlock_buf;
1150 * Now we have got to calculate how much data we have to copy. In
1151 * case of a static volume it is fairly easy - the VID header contains
1152 * the data size. In case of a dynamic volume it is more difficult - we
1153 * have to read the contents, cut 0xFF bytes from the end and copy only
1154 * the first part. We must do this to avoid writing 0xFF bytes as it
1155 * may have some side-effects. And not only this. It is important not
1156 * to include those 0xFFs to CRC because later the they may be filled
1159 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1160 aldata_size = data_size =
1161 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1164 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1168 * It may turn out to be that the whole @from physical eraseblock
1169 * contains only 0xFF bytes. Then we have to only write the VID header
1170 * and do not write any data. This also means we should not set
1171 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1173 if (data_size > 0) {
1174 vid_hdr->copy_flag = 1;
1175 vid_hdr->data_size = cpu_to_be32(data_size);
1176 vid_hdr->data_crc = cpu_to_be32(crc);
1178 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1180 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1183 err = MOVE_TARGET_WR_ERR;
1184 goto out_unlock_buf;
1189 /* Read the VID header back and check if it was written correctly */
1190 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1192 if (err != UBI_IO_BITFLIPS) {
1193 ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1195 if (is_error_sane(err))
1196 err = MOVE_TARGET_RD_ERR;
1198 err = MOVE_TARGET_BITFLIPS;
1199 goto out_unlock_buf;
1202 if (data_size > 0) {
1203 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1206 err = MOVE_TARGET_WR_ERR;
1207 goto out_unlock_buf;
1213 ubi_assert(vol->eba_tbl[lnum] == from);
1214 down_read(&ubi->fm_eba_sem);
1215 vol->eba_tbl[lnum] = to;
1216 up_read(&ubi->fm_eba_sem);
1219 mutex_unlock(&ubi->buf_mutex);
1221 leb_write_unlock(ubi, vol_id, lnum);
1226 * print_rsvd_warning - warn about not having enough reserved PEBs.
1227 * @ubi: UBI device description object
1229 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1230 * cannot reserve enough PEBs for bad block handling. This function makes a
1231 * decision whether we have to print a warning or not. The algorithm is as
1233 * o if this is a new UBI image, then just print the warning
1234 * o if this is an UBI image which has already been used for some time, print
1235 * a warning only if we can reserve less than 10% of the expected amount of
1238 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1239 * of PEBs becomes smaller, which is normal and we do not want to scare users
1240 * with a warning every time they attach the MTD device. This was an issue
1241 * reported by real users.
1243 static void print_rsvd_warning(struct ubi_device *ubi,
1244 struct ubi_attach_info *ai)
1247 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1248 * large number to distinguish between newly flashed and used images.
1250 if (ai->max_sqnum > (1 << 18)) {
1251 int min = ubi->beb_rsvd_level / 10;
1255 if (ubi->beb_rsvd_pebs > min)
1259 ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1260 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1261 if (ubi->corr_peb_count)
1262 ubi_warn(ubi, "%d PEBs are corrupted and not used",
1263 ubi->corr_peb_count);
1267 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1268 * @ubi: UBI device description object
1269 * @ai_fastmap: UBI attach info object created by fastmap
1270 * @ai_scan: UBI attach info object created by scanning
1272 * Returns < 0 in case of an internal error, 0 otherwise.
1273 * If a bad EBA table entry was found it will be printed out and
1274 * ubi_assert() triggers.
1276 int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1277 struct ubi_attach_info *ai_scan)
1279 int i, j, num_volumes, ret = 0;
1280 int **scan_eba, **fm_eba;
1281 struct ubi_ainf_volume *av;
1282 struct ubi_volume *vol;
1283 struct ubi_ainf_peb *aeb;
1286 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1288 scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1292 fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1298 for (i = 0; i < num_volumes; i++) {
1299 vol = ubi->volumes[i];
1303 scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1310 fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1317 for (j = 0; j < vol->reserved_pebs; j++)
1318 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1320 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1324 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1325 scan_eba[i][aeb->lnum] = aeb->pnum;
1327 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1331 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1332 fm_eba[i][aeb->lnum] = aeb->pnum;
1334 for (j = 0; j < vol->reserved_pebs; j++) {
1335 if (scan_eba[i][j] != fm_eba[i][j]) {
1336 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1337 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1340 ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1341 vol->vol_id, j, fm_eba[i][j],
1349 for (i = 0; i < num_volumes; i++) {
1350 if (!ubi->volumes[i])
1363 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1364 * @ubi: UBI device description object
1365 * @ai: attaching information
1367 * This function returns zero in case of success and a negative error code in
1370 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1372 int i, j, err, num_volumes;
1373 struct ubi_ainf_volume *av;
1374 struct ubi_volume *vol;
1375 struct ubi_ainf_peb *aeb;
1378 dbg_eba("initialize EBA sub-system");
1380 spin_lock_init(&ubi->ltree_lock);
1381 mutex_init(&ubi->alc_mutex);
1382 ubi->ltree = RB_ROOT;
1384 ubi->global_sqnum = ai->max_sqnum + 1;
1385 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1387 for (i = 0; i < num_volumes; i++) {
1388 vol = ubi->volumes[i];
1394 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1396 if (!vol->eba_tbl) {
1401 for (j = 0; j < vol->reserved_pebs; j++)
1402 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1404 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1408 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1409 if (aeb->lnum >= vol->reserved_pebs)
1411 * This may happen in case of an unclean reboot
1414 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1416 vol->eba_tbl[aeb->lnum] = aeb->pnum;
1420 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1421 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1422 ubi->avail_pebs, EBA_RESERVED_PEBS);
1423 if (ubi->corr_peb_count)
1424 ubi_err(ubi, "%d PEBs are corrupted and not used",
1425 ubi->corr_peb_count);
1429 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1430 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1432 if (ubi->bad_allowed) {
1433 ubi_calculate_reserved(ubi);
1435 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1436 /* No enough free physical eraseblocks */
1437 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1438 print_rsvd_warning(ubi, ai);
1440 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1442 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1443 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1446 dbg_eba("EBA sub-system is initialized");
1450 for (i = 0; i < num_volumes; i++) {
1451 if (!ubi->volumes[i])
1453 kfree(ubi->volumes[i]->eba_tbl);
1454 ubi->volumes[i]->eba_tbl = NULL;