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 * struct ubi_eba_entry - structure encoding a single LEB -> PEB association
54 * @pnum: the physical eraseblock number attached to the LEB
56 * This structure is encoding a LEB -> PEB association. Note that the LEB
57 * number is not stored here, because it is the index used to access the
60 struct ubi_eba_entry {
65 * struct ubi_eba_table - LEB -> PEB association information
66 * @entries: the LEB to PEB mapping (one entry per LEB).
68 * This structure is private to the EBA logic and should be kept here.
69 * It is encoding the LEB to PEB association table, and is subject to
72 struct ubi_eba_table {
73 struct ubi_eba_entry *entries;
77 * next_sqnum - get next sequence number.
78 * @ubi: UBI device description object
80 * This function returns next sequence number to use, which is just the current
81 * global sequence counter value. It also increases the global sequence
84 unsigned long long ubi_next_sqnum(struct ubi_device *ubi)
86 unsigned long long sqnum;
88 spin_lock(&ubi->ltree_lock);
89 sqnum = ubi->global_sqnum++;
90 spin_unlock(&ubi->ltree_lock);
96 * ubi_get_compat - get compatibility flags of a volume.
97 * @ubi: UBI device description object
100 * This function returns compatibility flags for an internal volume. User
101 * volumes have no compatibility flags, so %0 is returned.
103 static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
105 if (vol_id == UBI_LAYOUT_VOLUME_ID)
106 return UBI_LAYOUT_VOLUME_COMPAT;
111 * ubi_eba_get_ldesc - get information about a LEB
112 * @vol: volume description object
113 * @lnum: logical eraseblock number
114 * @ldesc: the LEB descriptor to fill
116 * Used to query information about a specific LEB.
117 * It is currently only returning the physical position of the LEB, but will be
118 * extended to provide more information.
120 void ubi_eba_get_ldesc(struct ubi_volume *vol, int lnum,
121 struct ubi_eba_leb_desc *ldesc)
124 ldesc->pnum = vol->eba_tbl->entries[lnum].pnum;
128 * ubi_eba_create_table - allocate a new EBA table and initialize it with all
130 * @vol: volume containing the EBA table to copy
131 * @nentries: number of entries in the table
133 * Allocate a new EBA table and initialize it with all LEBs unmapped.
134 * Returns a valid pointer if it succeed, an ERR_PTR() otherwise.
136 struct ubi_eba_table *ubi_eba_create_table(struct ubi_volume *vol,
139 struct ubi_eba_table *tbl;
143 tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
145 return ERR_PTR(-ENOMEM);
147 tbl->entries = kmalloc_array(nentries, sizeof(*tbl->entries),
152 for (i = 0; i < nentries; i++)
153 tbl->entries[i].pnum = UBI_LEB_UNMAPPED;
165 * ubi_eba_destroy_table - destroy an EBA table
166 * @tbl: the table to destroy
168 * Destroy an EBA table.
170 void ubi_eba_destroy_table(struct ubi_eba_table *tbl)
180 * ubi_eba_copy_table - copy the EBA table attached to vol into another table
181 * @vol: volume containing the EBA table to copy
183 * @nentries: number of entries to copy
185 * Copy the EBA table stored in vol into the one pointed by dst.
187 void ubi_eba_copy_table(struct ubi_volume *vol, struct ubi_eba_table *dst,
190 struct ubi_eba_table *src;
193 ubi_assert(dst && vol && vol->eba_tbl);
197 for (i = 0; i < nentries; i++)
198 dst->entries[i].pnum = src->entries[i].pnum;
202 * ubi_eba_replace_table - assign a new EBA table to a volume
203 * @vol: volume containing the EBA table to copy
204 * @tbl: new EBA table
206 * Assign a new EBA table to the volume and release the old one.
208 void ubi_eba_replace_table(struct ubi_volume *vol, struct ubi_eba_table *tbl)
210 ubi_eba_destroy_table(vol->eba_tbl);
215 * ltree_lookup - look up the lock tree.
216 * @ubi: UBI device description object
218 * @lnum: logical eraseblock number
220 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
221 * object if the logical eraseblock is locked and %NULL if it is not.
222 * @ubi->ltree_lock has to be locked.
224 static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
229 p = ubi->ltree.rb_node;
231 struct ubi_ltree_entry *le;
233 le = rb_entry(p, struct ubi_ltree_entry, rb);
235 if (vol_id < le->vol_id)
237 else if (vol_id > le->vol_id)
242 else if (lnum > le->lnum)
253 * ltree_add_entry - add new entry to the lock tree.
254 * @ubi: UBI device description object
256 * @lnum: logical eraseblock number
258 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
259 * lock tree. If such entry is already there, its usage counter is increased.
260 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
263 static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
264 int vol_id, int lnum)
266 struct ubi_ltree_entry *le, *le1, *le_free;
268 le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
270 return ERR_PTR(-ENOMEM);
273 init_rwsem(&le->mutex);
277 spin_lock(&ubi->ltree_lock);
278 le1 = ltree_lookup(ubi, vol_id, lnum);
282 * This logical eraseblock is already locked. The newly
283 * allocated lock entry is not needed.
288 struct rb_node **p, *parent = NULL;
291 * No lock entry, add the newly allocated one to the
292 * @ubi->ltree RB-tree.
296 p = &ubi->ltree.rb_node;
299 le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
301 if (vol_id < le1->vol_id)
303 else if (vol_id > le1->vol_id)
306 ubi_assert(lnum != le1->lnum);
307 if (lnum < le1->lnum)
314 rb_link_node(&le->rb, parent, p);
315 rb_insert_color(&le->rb, &ubi->ltree);
318 spin_unlock(&ubi->ltree_lock);
325 * leb_read_lock - lock logical eraseblock for reading.
326 * @ubi: UBI device description object
328 * @lnum: logical eraseblock number
330 * This function locks a logical eraseblock for reading. Returns zero in case
331 * of success and a negative error code in case of failure.
333 static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
335 struct ubi_ltree_entry *le;
337 le = ltree_add_entry(ubi, vol_id, lnum);
340 down_read(&le->mutex);
345 * leb_read_unlock - unlock logical eraseblock.
346 * @ubi: UBI device description object
348 * @lnum: logical eraseblock number
350 static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
352 struct ubi_ltree_entry *le;
354 spin_lock(&ubi->ltree_lock);
355 le = ltree_lookup(ubi, vol_id, lnum);
357 ubi_assert(le->users >= 0);
359 if (le->users == 0) {
360 rb_erase(&le->rb, &ubi->ltree);
363 spin_unlock(&ubi->ltree_lock);
367 * leb_write_lock - lock logical eraseblock for writing.
368 * @ubi: UBI device description object
370 * @lnum: logical eraseblock number
372 * This function locks a logical eraseblock for writing. Returns zero in case
373 * of success and a negative error code in case of failure.
375 static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
377 struct ubi_ltree_entry *le;
379 le = ltree_add_entry(ubi, vol_id, lnum);
382 down_write(&le->mutex);
387 * leb_write_lock - lock logical eraseblock for writing.
388 * @ubi: UBI device description object
390 * @lnum: logical eraseblock number
392 * This function locks a logical eraseblock for writing if there is no
393 * contention and does nothing if there is contention. Returns %0 in case of
394 * success, %1 in case of contention, and and a negative error code in case of
397 static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
399 struct ubi_ltree_entry *le;
401 le = ltree_add_entry(ubi, vol_id, lnum);
404 if (down_write_trylock(&le->mutex))
407 /* Contention, cancel */
408 spin_lock(&ubi->ltree_lock);
410 ubi_assert(le->users >= 0);
411 if (le->users == 0) {
412 rb_erase(&le->rb, &ubi->ltree);
415 spin_unlock(&ubi->ltree_lock);
421 * leb_write_unlock - unlock logical eraseblock.
422 * @ubi: UBI device description object
424 * @lnum: logical eraseblock number
426 static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
428 struct ubi_ltree_entry *le;
430 spin_lock(&ubi->ltree_lock);
431 le = ltree_lookup(ubi, vol_id, lnum);
433 ubi_assert(le->users >= 0);
434 up_write(&le->mutex);
435 if (le->users == 0) {
436 rb_erase(&le->rb, &ubi->ltree);
439 spin_unlock(&ubi->ltree_lock);
443 * ubi_eba_is_mapped - check if a LEB is mapped.
444 * @vol: volume description object
445 * @lnum: logical eraseblock number
447 * This function returns true if the LEB is mapped, false otherwise.
449 bool ubi_eba_is_mapped(struct ubi_volume *vol, int lnum)
451 return vol->eba_tbl->entries[lnum].pnum >= 0;
455 * ubi_eba_unmap_leb - un-map logical eraseblock.
456 * @ubi: UBI device description object
457 * @vol: volume description object
458 * @lnum: logical eraseblock number
460 * This function un-maps logical eraseblock @lnum and schedules corresponding
461 * physical eraseblock for erasure. Returns zero in case of success and a
462 * negative error code in case of failure.
464 int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
467 int err, pnum, vol_id = vol->vol_id;
472 err = leb_write_lock(ubi, vol_id, lnum);
476 pnum = vol->eba_tbl->entries[lnum].pnum;
478 /* This logical eraseblock is already unmapped */
481 dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
483 down_read(&ubi->fm_eba_sem);
484 vol->eba_tbl->entries[lnum].pnum = UBI_LEB_UNMAPPED;
485 up_read(&ubi->fm_eba_sem);
486 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
489 leb_write_unlock(ubi, vol_id, lnum);
494 * ubi_eba_read_leb - read data.
495 * @ubi: UBI device description object
496 * @vol: volume description object
497 * @lnum: logical eraseblock number
498 * @buf: buffer to store the read data
499 * @offset: offset from where to read
500 * @len: how many bytes to read
501 * @check: data CRC check flag
503 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
504 * bytes. The @check flag only makes sense for static volumes and forces
505 * eraseblock data CRC checking.
507 * In case of success this function returns zero. In case of a static volume,
508 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
509 * returned for any volume type if an ECC error was detected by the MTD device
510 * driver. Other negative error cored may be returned in case of other errors.
512 int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
513 void *buf, int offset, int len, int check)
515 int err, pnum, scrub = 0, vol_id = vol->vol_id;
516 struct ubi_vid_hdr *vid_hdr;
517 uint32_t uninitialized_var(crc);
519 err = leb_read_lock(ubi, vol_id, lnum);
523 pnum = vol->eba_tbl->entries[lnum].pnum;
526 * The logical eraseblock is not mapped, fill the whole buffer
527 * with 0xFF bytes. The exception is static volumes for which
528 * it is an error to read unmapped logical eraseblocks.
530 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
531 len, offset, vol_id, lnum);
532 leb_read_unlock(ubi, vol_id, lnum);
533 ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
534 memset(buf, 0xFF, len);
538 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
539 len, offset, vol_id, lnum, pnum);
541 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
546 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
552 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
553 if (err && err != UBI_IO_BITFLIPS) {
556 * The header is either absent or corrupted.
557 * The former case means there is a bug -
558 * switch to read-only mode just in case.
559 * The latter case means a real corruption - we
560 * may try to recover data. FIXME: but this is
563 if (err == UBI_IO_BAD_HDR_EBADMSG ||
564 err == UBI_IO_BAD_HDR) {
565 ubi_warn(ubi, "corrupted VID header at PEB %d, LEB %d:%d",
570 * Ending up here in the non-Fastmap case
571 * is a clear bug as the VID header had to
572 * be present at scan time to have it referenced.
573 * With fastmap the story is more complicated.
574 * Fastmap has the mapping info without the need
575 * of a full scan. So the LEB could have been
576 * unmapped, Fastmap cannot know this and keeps
577 * the LEB referenced.
578 * This is valid and works as the layer above UBI
579 * has to do bookkeeping about used/referenced
582 if (ubi->fast_attach) {
591 } else if (err == UBI_IO_BITFLIPS)
594 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
595 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
597 crc = be32_to_cpu(vid_hdr->data_crc);
598 ubi_free_vid_hdr(ubi, vid_hdr);
601 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
603 if (err == UBI_IO_BITFLIPS)
605 else if (mtd_is_eccerr(err)) {
606 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
610 ubi_msg(ubi, "force data checking");
619 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
621 ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
629 err = ubi_wl_scrub_peb(ubi, pnum);
631 leb_read_unlock(ubi, vol_id, lnum);
635 ubi_free_vid_hdr(ubi, vid_hdr);
637 leb_read_unlock(ubi, vol_id, lnum);
642 * ubi_eba_read_leb_sg - read data into a scatter gather list.
643 * @ubi: UBI device description object
644 * @vol: volume description object
645 * @lnum: logical eraseblock number
646 * @sgl: UBI scatter gather list to store the read data
647 * @offset: offset from where to read
648 * @len: how many bytes to read
649 * @check: data CRC check flag
651 * This function works exactly like ubi_eba_read_leb(). But instead of
652 * storing the read data into a buffer it writes to an UBI scatter gather
655 int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
656 struct ubi_sgl *sgl, int lnum, int offset, int len,
661 struct scatterlist *sg;
664 ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
665 sg = &sgl->sg[sgl->list_pos];
666 if (len < sg->length - sgl->page_pos)
669 to_read = sg->length - sgl->page_pos;
671 ret = ubi_eba_read_leb(ubi, vol, lnum,
672 sg_virt(sg) + sgl->page_pos, offset,
680 sgl->page_pos += to_read;
681 if (sgl->page_pos == sg->length) {
697 * try_recover_peb - try to recover from write failure.
698 * @vol: volume description object
699 * @pnum: the physical eraseblock to recover
700 * @lnum: logical eraseblock number
701 * @buf: data which was not written because of the write failure
702 * @offset: offset of the failed write
703 * @len: how many bytes should have been written
705 * @retry: whether the caller should retry in case of failure
707 * This function is called in case of a write failure and moves all good data
708 * from the potentially bad physical eraseblock to a good physical eraseblock.
709 * This function also writes the data which was not written due to the failure.
710 * Returns 0 in case of success, and a negative error code in case of failure.
711 * In case of failure, the %retry parameter is set to false if this is a fatal
712 * error (retrying won't help), and true otherwise.
714 static int try_recover_peb(struct ubi_volume *vol, int pnum, int lnum,
715 const void *buf, int offset, int len,
716 struct ubi_vid_hdr *vid_hdr, bool *retry)
718 struct ubi_device *ubi = vol->ubi;
719 int new_pnum, err, vol_id = vol->vol_id, data_size;
724 new_pnum = ubi_wl_get_peb(ubi);
730 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
733 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
734 if (err && err != UBI_IO_BITFLIPS) {
740 ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
742 mutex_lock(&ubi->buf_mutex);
743 memset(ubi->peb_buf + offset, 0xFF, len);
745 /* Read everything before the area where the write failure happened */
747 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
748 if (err && err != UBI_IO_BITFLIPS)
754 memcpy(ubi->peb_buf + offset, buf, len);
756 data_size = offset + len;
757 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
758 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
759 vid_hdr->copy_flag = 1;
760 vid_hdr->data_size = cpu_to_be32(data_size);
761 vid_hdr->data_crc = cpu_to_be32(crc);
762 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
766 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
769 mutex_unlock(&ubi->buf_mutex);
772 vol->eba_tbl->entries[lnum].pnum = new_pnum;
775 up_read(&ubi->fm_eba_sem);
778 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
779 ubi_msg(ubi, "data was successfully recovered");
780 } else if (new_pnum >= 0) {
782 * Bad luck? This physical eraseblock is bad too? Crud. Let's
783 * try to get another one.
785 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
786 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
793 * recover_peb - recover from write failure.
794 * @ubi: UBI device description object
795 * @pnum: the physical eraseblock to recover
797 * @lnum: logical eraseblock number
798 * @buf: data which was not written because of the write failure
799 * @offset: offset of the failed write
800 * @len: how many bytes should have been written
802 * This function is called in case of a write failure and moves all good data
803 * from the potentially bad physical eraseblock to a good physical eraseblock.
804 * This function also writes the data which was not written due to the failure.
805 * Returns 0 in case of success, and a negative error code in case of failure.
806 * This function tries %UBI_IO_RETRIES before giving up.
808 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
809 const void *buf, int offset, int len)
811 int err, idx = vol_id2idx(ubi, vol_id), tries;
812 struct ubi_volume *vol = ubi->volumes[idx];
813 struct ubi_vid_hdr *vid_hdr;
815 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
819 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
822 err = try_recover_peb(vol, pnum, lnum, buf, offset, len,
827 ubi_msg(ubi, "try again");
830 ubi_free_vid_hdr(ubi, vid_hdr);
836 * try_write_vid_and_data - try to write VID header and data to a new PEB.
837 * @vol: volume description object
838 * @lnum: logical eraseblock number
839 * @vid_hdr: VID header to write
840 * @buf: buffer containing the data
841 * @offset: where to start writing data
842 * @len: how many bytes should be written
844 * This function tries to write VID header and data belonging to logical
845 * eraseblock @lnum of volume @vol to a new physical eraseblock. Returns zero
846 * in case of success and a negative error code in case of failure.
847 * In case of error, it is possible that something was still written to the
848 * flash media, but may be some garbage.
850 static int try_write_vid_and_data(struct ubi_volume *vol, int lnum,
851 struct ubi_vid_hdr *vid_hdr, const void *buf,
854 struct ubi_device *ubi = vol->ubi;
855 int pnum, opnum, err, vol_id = vol->vol_id;
857 pnum = ubi_wl_get_peb(ubi);
863 opnum = vol->eba_tbl->entries[lnum].pnum;
865 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
866 len, offset, vol_id, lnum, pnum);
868 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
870 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
876 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
879 "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
880 len, offset, vol_id, lnum, pnum);
885 vol->eba_tbl->entries[lnum].pnum = pnum;
888 up_read(&ubi->fm_eba_sem);
890 if (err && pnum >= 0)
891 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
892 else if (!err && opnum >= 0)
893 err = ubi_wl_put_peb(ubi, vol_id, lnum, opnum, 0);
899 * ubi_eba_write_leb - write data to dynamic volume.
900 * @ubi: UBI device description object
901 * @vol: volume description object
902 * @lnum: logical eraseblock number
903 * @buf: the data to write
904 * @offset: offset within the logical eraseblock where to write
905 * @len: how many bytes to write
907 * This function writes data to logical eraseblock @lnum of a dynamic volume
908 * @vol. Returns zero in case of success and a negative error code in case
909 * of failure. In case of error, it is possible that something was still
910 * written to the flash media, but may be some garbage.
911 * This function retries %UBI_IO_RETRIES times before giving up.
913 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
914 const void *buf, int offset, int len)
916 int err, pnum, tries, vol_id = vol->vol_id;
917 struct ubi_vid_hdr *vid_hdr;
922 err = leb_write_lock(ubi, vol_id, lnum);
926 pnum = vol->eba_tbl->entries[lnum].pnum;
928 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
929 len, offset, vol_id, lnum, pnum);
931 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
933 ubi_warn(ubi, "failed to write data to PEB %d", pnum);
934 if (err == -EIO && ubi->bad_allowed)
935 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
943 * The logical eraseblock is not mapped. We have to get a free physical
944 * eraseblock and write the volume identifier header there first.
946 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
948 leb_write_unlock(ubi, vol_id, lnum);
952 vid_hdr->vol_type = UBI_VID_DYNAMIC;
953 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
954 vid_hdr->vol_id = cpu_to_be32(vol_id);
955 vid_hdr->lnum = cpu_to_be32(lnum);
956 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
957 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
959 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
960 err = try_write_vid_and_data(vol, lnum, vid_hdr, buf, offset,
962 if (err != -EIO || !ubi->bad_allowed)
966 * Fortunately, this is the first write operation to this
967 * physical eraseblock, so just put it and request a new one.
968 * We assume that if this physical eraseblock went bad, the
969 * erase code will handle that.
971 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
972 ubi_msg(ubi, "try another PEB");
975 ubi_free_vid_hdr(ubi, vid_hdr);
981 leb_write_unlock(ubi, vol_id, lnum);
987 * ubi_eba_write_leb_st - write data to static volume.
988 * @ubi: UBI device description object
989 * @vol: volume description object
990 * @lnum: logical eraseblock number
991 * @buf: data to write
992 * @len: how many bytes to write
993 * @used_ebs: how many logical eraseblocks will this volume contain
995 * This function writes data to logical eraseblock @lnum of static volume
996 * @vol. The @used_ebs argument should contain total number of logical
997 * eraseblock in this static volume.
999 * When writing to the last logical eraseblock, the @len argument doesn't have
1000 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
1001 * to the real data size, although the @buf buffer has to contain the
1002 * alignment. In all other cases, @len has to be aligned.
1004 * It is prohibited to write more than once to logical eraseblocks of static
1005 * volumes. This function returns zero in case of success and a negative error
1006 * code in case of failure.
1008 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
1009 int lnum, const void *buf, int len, int used_ebs)
1011 int err, tries, data_size = len, vol_id = vol->vol_id;
1012 struct ubi_vid_hdr *vid_hdr;
1018 if (lnum == used_ebs - 1)
1019 /* If this is the last LEB @len may be unaligned */
1020 len = ALIGN(data_size, ubi->min_io_size);
1022 ubi_assert(!(len & (ubi->min_io_size - 1)));
1024 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1028 err = leb_write_lock(ubi, vol_id, lnum);
1032 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1033 vid_hdr->vol_id = cpu_to_be32(vol_id);
1034 vid_hdr->lnum = cpu_to_be32(lnum);
1035 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1036 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1038 crc = crc32(UBI_CRC32_INIT, buf, data_size);
1039 vid_hdr->vol_type = UBI_VID_STATIC;
1040 vid_hdr->data_size = cpu_to_be32(data_size);
1041 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
1042 vid_hdr->data_crc = cpu_to_be32(crc);
1044 ubi_assert(vol->eba_tbl->entries[lnum].pnum < 0);
1046 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1047 err = try_write_vid_and_data(vol, lnum, vid_hdr, buf, 0, len);
1048 if (err != -EIO || !ubi->bad_allowed)
1051 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1052 ubi_msg(ubi, "try another PEB");
1058 leb_write_unlock(ubi, vol_id, lnum);
1061 ubi_free_vid_hdr(ubi, vid_hdr);
1067 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
1068 * @ubi: UBI device description object
1069 * @vol: volume description object
1070 * @lnum: logical eraseblock number
1071 * @buf: data to write
1072 * @len: how many bytes to write
1074 * This function changes the contents of a logical eraseblock atomically. @buf
1075 * has to contain new logical eraseblock data, and @len - the length of the
1076 * data, which has to be aligned. This function guarantees that in case of an
1077 * unclean reboot the old contents is preserved. Returns zero in case of
1078 * success and a negative error code in case of failure.
1080 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
1081 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
1083 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
1084 int lnum, const void *buf, int len)
1086 int err, tries, vol_id = vol->vol_id;
1087 struct ubi_vid_hdr *vid_hdr;
1095 * Special case when data length is zero. In this case the LEB
1096 * has to be unmapped and mapped somewhere else.
1098 err = ubi_eba_unmap_leb(ubi, vol, lnum);
1101 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
1104 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
1108 mutex_lock(&ubi->alc_mutex);
1109 err = leb_write_lock(ubi, vol_id, lnum);
1113 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1114 vid_hdr->vol_id = cpu_to_be32(vol_id);
1115 vid_hdr->lnum = cpu_to_be32(lnum);
1116 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1117 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1119 crc = crc32(UBI_CRC32_INIT, buf, len);
1120 vid_hdr->vol_type = UBI_VID_DYNAMIC;
1121 vid_hdr->data_size = cpu_to_be32(len);
1122 vid_hdr->copy_flag = 1;
1123 vid_hdr->data_crc = cpu_to_be32(crc);
1125 dbg_eba("change LEB %d:%d", vol_id, lnum);
1127 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1128 err = try_write_vid_and_data(vol, lnum, vid_hdr, buf, 0, len);
1129 if (err != -EIO || !ubi->bad_allowed)
1132 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1133 ubi_msg(ubi, "try another PEB");
1137 * This flash device does not admit of bad eraseblocks or
1138 * something nasty and unexpected happened. Switch to read-only
1139 * mode just in case.
1144 leb_write_unlock(ubi, vol_id, lnum);
1147 mutex_unlock(&ubi->alc_mutex);
1148 ubi_free_vid_hdr(ubi, vid_hdr);
1153 * is_error_sane - check whether a read error is sane.
1154 * @err: code of the error happened during reading
1156 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1157 * cannot read data from the target PEB (an error @err happened). If the error
1158 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1159 * fatal and UBI will be switched to R/O mode later.
1161 * The idea is that we try not to switch to R/O mode if the read error is
1162 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1163 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1164 * mode, simply because we do not know what happened at the MTD level, and we
1165 * cannot handle this. E.g., the underlying driver may have become crazy, and
1166 * it is safer to switch to R/O mode to preserve the data.
1168 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1169 * which we have just written.
1171 static int is_error_sane(int err)
1173 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1174 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1180 * ubi_eba_copy_leb - copy logical eraseblock.
1181 * @ubi: UBI device description object
1182 * @from: physical eraseblock number from where to copy
1183 * @to: physical eraseblock number where to copy
1184 * @vid_hdr: VID header of the @from physical eraseblock
1186 * This function copies logical eraseblock from physical eraseblock @from to
1187 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1188 * function. Returns:
1189 * o %0 in case of success;
1190 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1191 * o a negative error code in case of failure.
1193 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1194 struct ubi_vid_hdr *vid_hdr)
1196 int err, vol_id, lnum, data_size, aldata_size, idx;
1197 struct ubi_volume *vol;
1200 vol_id = be32_to_cpu(vid_hdr->vol_id);
1201 lnum = be32_to_cpu(vid_hdr->lnum);
1203 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1205 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1206 data_size = be32_to_cpu(vid_hdr->data_size);
1207 aldata_size = ALIGN(data_size, ubi->min_io_size);
1209 data_size = aldata_size =
1210 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1212 idx = vol_id2idx(ubi, vol_id);
1213 spin_lock(&ubi->volumes_lock);
1215 * Note, we may race with volume deletion, which means that the volume
1216 * this logical eraseblock belongs to might be being deleted. Since the
1217 * volume deletion un-maps all the volume's logical eraseblocks, it will
1218 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1220 vol = ubi->volumes[idx];
1221 spin_unlock(&ubi->volumes_lock);
1223 /* No need to do further work, cancel */
1224 dbg_wl("volume %d is being removed, cancel", vol_id);
1225 return MOVE_CANCEL_RACE;
1229 * We do not want anybody to write to this logical eraseblock while we
1230 * are moving it, so lock it.
1232 * Note, we are using non-waiting locking here, because we cannot sleep
1233 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1234 * unmapping the LEB which is mapped to the PEB we are going to move
1235 * (@from). This task locks the LEB and goes sleep in the
1236 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1237 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1238 * LEB is already locked, we just do not move it and return
1239 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1240 * we do not know the reasons of the contention - it may be just a
1241 * normal I/O on this LEB, so we want to re-try.
1243 err = leb_write_trylock(ubi, vol_id, lnum);
1245 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1250 * The LEB might have been put meanwhile, and the task which put it is
1251 * probably waiting on @ubi->move_mutex. No need to continue the work,
1254 if (vol->eba_tbl->entries[lnum].pnum != from) {
1255 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1256 vol_id, lnum, from, vol->eba_tbl->entries[lnum].pnum);
1257 err = MOVE_CANCEL_RACE;
1258 goto out_unlock_leb;
1262 * OK, now the LEB is locked and we can safely start moving it. Since
1263 * this function utilizes the @ubi->peb_buf buffer which is shared
1264 * with some other functions - we lock the buffer by taking the
1267 mutex_lock(&ubi->buf_mutex);
1268 dbg_wl("read %d bytes of data", aldata_size);
1269 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1270 if (err && err != UBI_IO_BITFLIPS) {
1271 ubi_warn(ubi, "error %d while reading data from PEB %d",
1273 err = MOVE_SOURCE_RD_ERR;
1274 goto out_unlock_buf;
1278 * Now we have got to calculate how much data we have to copy. In
1279 * case of a static volume it is fairly easy - the VID header contains
1280 * the data size. In case of a dynamic volume it is more difficult - we
1281 * have to read the contents, cut 0xFF bytes from the end and copy only
1282 * the first part. We must do this to avoid writing 0xFF bytes as it
1283 * may have some side-effects. And not only this. It is important not
1284 * to include those 0xFFs to CRC because later the they may be filled
1287 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1288 aldata_size = data_size =
1289 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1292 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1296 * It may turn out to be that the whole @from physical eraseblock
1297 * contains only 0xFF bytes. Then we have to only write the VID header
1298 * and do not write any data. This also means we should not set
1299 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1301 if (data_size > 0) {
1302 vid_hdr->copy_flag = 1;
1303 vid_hdr->data_size = cpu_to_be32(data_size);
1304 vid_hdr->data_crc = cpu_to_be32(crc);
1306 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1308 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1311 err = MOVE_TARGET_WR_ERR;
1312 goto out_unlock_buf;
1317 /* Read the VID header back and check if it was written correctly */
1318 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1320 if (err != UBI_IO_BITFLIPS) {
1321 ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1323 if (is_error_sane(err))
1324 err = MOVE_TARGET_RD_ERR;
1326 err = MOVE_TARGET_BITFLIPS;
1327 goto out_unlock_buf;
1330 if (data_size > 0) {
1331 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1334 err = MOVE_TARGET_WR_ERR;
1335 goto out_unlock_buf;
1341 ubi_assert(vol->eba_tbl->entries[lnum].pnum == from);
1342 down_read(&ubi->fm_eba_sem);
1343 vol->eba_tbl->entries[lnum].pnum = to;
1344 up_read(&ubi->fm_eba_sem);
1347 mutex_unlock(&ubi->buf_mutex);
1349 leb_write_unlock(ubi, vol_id, lnum);
1354 * print_rsvd_warning - warn about not having enough reserved PEBs.
1355 * @ubi: UBI device description object
1357 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1358 * cannot reserve enough PEBs for bad block handling. This function makes a
1359 * decision whether we have to print a warning or not. The algorithm is as
1361 * o if this is a new UBI image, then just print the warning
1362 * o if this is an UBI image which has already been used for some time, print
1363 * a warning only if we can reserve less than 10% of the expected amount of
1366 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1367 * of PEBs becomes smaller, which is normal and we do not want to scare users
1368 * with a warning every time they attach the MTD device. This was an issue
1369 * reported by real users.
1371 static void print_rsvd_warning(struct ubi_device *ubi,
1372 struct ubi_attach_info *ai)
1375 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1376 * large number to distinguish between newly flashed and used images.
1378 if (ai->max_sqnum > (1 << 18)) {
1379 int min = ubi->beb_rsvd_level / 10;
1383 if (ubi->beb_rsvd_pebs > min)
1387 ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1388 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1389 if (ubi->corr_peb_count)
1390 ubi_warn(ubi, "%d PEBs are corrupted and not used",
1391 ubi->corr_peb_count);
1395 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1396 * @ubi: UBI device description object
1397 * @ai_fastmap: UBI attach info object created by fastmap
1398 * @ai_scan: UBI attach info object created by scanning
1400 * Returns < 0 in case of an internal error, 0 otherwise.
1401 * If a bad EBA table entry was found it will be printed out and
1402 * ubi_assert() triggers.
1404 int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1405 struct ubi_attach_info *ai_scan)
1407 int i, j, num_volumes, ret = 0;
1408 int **scan_eba, **fm_eba;
1409 struct ubi_ainf_volume *av;
1410 struct ubi_volume *vol;
1411 struct ubi_ainf_peb *aeb;
1414 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1416 scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1420 fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1426 for (i = 0; i < num_volumes; i++) {
1427 vol = ubi->volumes[i];
1431 scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1438 fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1445 for (j = 0; j < vol->reserved_pebs; j++)
1446 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1448 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1452 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1453 scan_eba[i][aeb->lnum] = aeb->pnum;
1455 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1459 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1460 fm_eba[i][aeb->lnum] = aeb->pnum;
1462 for (j = 0; j < vol->reserved_pebs; j++) {
1463 if (scan_eba[i][j] != fm_eba[i][j]) {
1464 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1465 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1468 ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1469 vol->vol_id, j, fm_eba[i][j],
1477 for (i = 0; i < num_volumes; i++) {
1478 if (!ubi->volumes[i])
1491 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1492 * @ubi: UBI device description object
1493 * @ai: attaching information
1495 * This function returns zero in case of success and a negative error code in
1498 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1500 int i, err, num_volumes;
1501 struct ubi_ainf_volume *av;
1502 struct ubi_volume *vol;
1503 struct ubi_ainf_peb *aeb;
1506 dbg_eba("initialize EBA sub-system");
1508 spin_lock_init(&ubi->ltree_lock);
1509 mutex_init(&ubi->alc_mutex);
1510 ubi->ltree = RB_ROOT;
1512 ubi->global_sqnum = ai->max_sqnum + 1;
1513 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1515 for (i = 0; i < num_volumes; i++) {
1516 struct ubi_eba_table *tbl;
1518 vol = ubi->volumes[i];
1524 tbl = ubi_eba_create_table(vol, vol->reserved_pebs);
1530 ubi_eba_replace_table(vol, tbl);
1532 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1536 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1537 if (aeb->lnum >= vol->reserved_pebs) {
1539 * This may happen in case of an unclean reboot
1542 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1544 struct ubi_eba_entry *entry;
1546 entry = &vol->eba_tbl->entries[aeb->lnum];
1547 entry->pnum = aeb->pnum;
1552 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1553 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1554 ubi->avail_pebs, EBA_RESERVED_PEBS);
1555 if (ubi->corr_peb_count)
1556 ubi_err(ubi, "%d PEBs are corrupted and not used",
1557 ubi->corr_peb_count);
1561 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1562 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1564 if (ubi->bad_allowed) {
1565 ubi_calculate_reserved(ubi);
1567 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1568 /* No enough free physical eraseblocks */
1569 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1570 print_rsvd_warning(ubi, ai);
1572 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1574 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1575 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1578 dbg_eba("EBA sub-system is initialized");
1582 for (i = 0; i < num_volumes; i++) {
1583 if (!ubi->volumes[i])
1585 ubi_eba_replace_table(ubi->volumes[i], NULL);