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_io_buf *vidb;
517 struct ubi_vid_hdr *vid_hdr;
518 uint32_t uninitialized_var(crc);
520 err = leb_read_lock(ubi, vol_id, lnum);
524 pnum = vol->eba_tbl->entries[lnum].pnum;
527 * The logical eraseblock is not mapped, fill the whole buffer
528 * with 0xFF bytes. The exception is static volumes for which
529 * it is an error to read unmapped logical eraseblocks.
531 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
532 len, offset, vol_id, lnum);
533 leb_read_unlock(ubi, vol_id, lnum);
534 ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
535 memset(buf, 0xFF, len);
539 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
540 len, offset, vol_id, lnum, pnum);
542 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
547 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
553 vid_hdr = ubi_get_vid_hdr(vidb);
555 err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
556 if (err && err != UBI_IO_BITFLIPS) {
559 * The header is either absent or corrupted.
560 * The former case means there is a bug -
561 * switch to read-only mode just in case.
562 * The latter case means a real corruption - we
563 * may try to recover data. FIXME: but this is
566 if (err == UBI_IO_BAD_HDR_EBADMSG ||
567 err == UBI_IO_BAD_HDR) {
568 ubi_warn(ubi, "corrupted VID header at PEB %d, LEB %d:%d",
573 * Ending up here in the non-Fastmap case
574 * is a clear bug as the VID header had to
575 * be present at scan time to have it referenced.
576 * With fastmap the story is more complicated.
577 * Fastmap has the mapping info without the need
578 * of a full scan. So the LEB could have been
579 * unmapped, Fastmap cannot know this and keeps
580 * the LEB referenced.
581 * This is valid and works as the layer above UBI
582 * has to do bookkeeping about used/referenced
585 if (ubi->fast_attach) {
594 } else if (err == UBI_IO_BITFLIPS)
597 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
598 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
600 crc = be32_to_cpu(vid_hdr->data_crc);
601 ubi_free_vid_buf(vidb);
604 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
606 if (err == UBI_IO_BITFLIPS)
608 else if (mtd_is_eccerr(err)) {
609 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
613 ubi_msg(ubi, "force data checking");
622 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
624 ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
632 err = ubi_wl_scrub_peb(ubi, pnum);
634 leb_read_unlock(ubi, vol_id, lnum);
638 ubi_free_vid_buf(vidb);
640 leb_read_unlock(ubi, vol_id, lnum);
645 * ubi_eba_read_leb_sg - read data into a scatter gather list.
646 * @ubi: UBI device description object
647 * @vol: volume description object
648 * @lnum: logical eraseblock number
649 * @sgl: UBI scatter gather list to store the read data
650 * @offset: offset from where to read
651 * @len: how many bytes to read
652 * @check: data CRC check flag
654 * This function works exactly like ubi_eba_read_leb(). But instead of
655 * storing the read data into a buffer it writes to an UBI scatter gather
658 int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
659 struct ubi_sgl *sgl, int lnum, int offset, int len,
664 struct scatterlist *sg;
667 ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
668 sg = &sgl->sg[sgl->list_pos];
669 if (len < sg->length - sgl->page_pos)
672 to_read = sg->length - sgl->page_pos;
674 ret = ubi_eba_read_leb(ubi, vol, lnum,
675 sg_virt(sg) + sgl->page_pos, offset,
683 sgl->page_pos += to_read;
684 if (sgl->page_pos == sg->length) {
700 * try_recover_peb - try to recover from write failure.
701 * @vol: volume description object
702 * @pnum: the physical eraseblock to recover
703 * @lnum: logical eraseblock number
704 * @buf: data which was not written because of the write failure
705 * @offset: offset of the failed write
706 * @len: how many bytes should have been written
708 * @retry: whether the caller should retry in case of failure
710 * This function is called in case of a write failure and moves all good data
711 * from the potentially bad physical eraseblock to a good physical eraseblock.
712 * This function also writes the data which was not written due to the failure.
713 * Returns 0 in case of success, and a negative error code in case of failure.
714 * In case of failure, the %retry parameter is set to false if this is a fatal
715 * error (retrying won't help), and true otherwise.
717 static int try_recover_peb(struct ubi_volume *vol, int pnum, int lnum,
718 const void *buf, int offset, int len,
719 struct ubi_vid_io_buf *vidb, bool *retry)
721 struct ubi_device *ubi = vol->ubi;
722 struct ubi_vid_hdr *vid_hdr;
723 int new_pnum, err, vol_id = vol->vol_id, data_size;
728 new_pnum = ubi_wl_get_peb(ubi);
734 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
737 err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
738 if (err && err != UBI_IO_BITFLIPS) {
744 ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
746 mutex_lock(&ubi->buf_mutex);
747 memset(ubi->peb_buf + offset, 0xFF, len);
749 /* Read everything before the area where the write failure happened */
751 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
752 if (err && err != UBI_IO_BITFLIPS)
758 memcpy(ubi->peb_buf + offset, buf, len);
760 data_size = offset + len;
761 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
762 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
763 vid_hdr->copy_flag = 1;
764 vid_hdr->data_size = cpu_to_be32(data_size);
765 vid_hdr->data_crc = cpu_to_be32(crc);
766 err = ubi_io_write_vid_hdr(ubi, new_pnum, vidb);
770 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
773 mutex_unlock(&ubi->buf_mutex);
776 vol->eba_tbl->entries[lnum].pnum = new_pnum;
779 up_read(&ubi->fm_eba_sem);
782 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
783 ubi_msg(ubi, "data was successfully recovered");
784 } else if (new_pnum >= 0) {
786 * Bad luck? This physical eraseblock is bad too? Crud. Let's
787 * try to get another one.
789 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
790 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
797 * recover_peb - recover from write failure.
798 * @ubi: UBI device description object
799 * @pnum: the physical eraseblock to recover
801 * @lnum: logical eraseblock number
802 * @buf: data which was not written because of the write failure
803 * @offset: offset of the failed write
804 * @len: how many bytes should have been written
806 * This function is called in case of a write failure and moves all good data
807 * from the potentially bad physical eraseblock to a good physical eraseblock.
808 * This function also writes the data which was not written due to the failure.
809 * Returns 0 in case of success, and a negative error code in case of failure.
810 * This function tries %UBI_IO_RETRIES before giving up.
812 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
813 const void *buf, int offset, int len)
815 int err, idx = vol_id2idx(ubi, vol_id), tries;
816 struct ubi_volume *vol = ubi->volumes[idx];
817 struct ubi_vid_io_buf *vidb;
819 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
823 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
826 err = try_recover_peb(vol, pnum, lnum, buf, offset, len, vidb,
831 ubi_msg(ubi, "try again");
834 ubi_free_vid_buf(vidb);
840 * try_write_vid_and_data - try to write VID header and data to a new PEB.
841 * @vol: volume description object
842 * @lnum: logical eraseblock number
843 * @vidb: the VID buffer to write
844 * @buf: buffer containing the data
845 * @offset: where to start writing data
846 * @len: how many bytes should be written
848 * This function tries to write VID header and data belonging to logical
849 * eraseblock @lnum of volume @vol to a new physical eraseblock. Returns zero
850 * in case of success and a negative error code in case of failure.
851 * In case of error, it is possible that something was still written to the
852 * flash media, but may be some garbage.
854 static int try_write_vid_and_data(struct ubi_volume *vol, int lnum,
855 struct ubi_vid_io_buf *vidb, const void *buf,
858 struct ubi_device *ubi = vol->ubi;
859 int pnum, opnum, err, vol_id = vol->vol_id;
861 pnum = ubi_wl_get_peb(ubi);
867 opnum = vol->eba_tbl->entries[lnum].pnum;
869 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
870 len, offset, vol_id, lnum, pnum);
872 err = ubi_io_write_vid_hdr(ubi, pnum, vidb);
874 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
880 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
883 "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
884 len, offset, vol_id, lnum, pnum);
889 vol->eba_tbl->entries[lnum].pnum = pnum;
892 up_read(&ubi->fm_eba_sem);
894 if (err && pnum >= 0)
895 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
896 else if (!err && opnum >= 0)
897 err = ubi_wl_put_peb(ubi, vol_id, lnum, opnum, 0);
903 * ubi_eba_write_leb - write data to dynamic volume.
904 * @ubi: UBI device description object
905 * @vol: volume description object
906 * @lnum: logical eraseblock number
907 * @buf: the data to write
908 * @offset: offset within the logical eraseblock where to write
909 * @len: how many bytes to write
911 * This function writes data to logical eraseblock @lnum of a dynamic volume
912 * @vol. Returns zero in case of success and a negative error code in case
913 * of failure. In case of error, it is possible that something was still
914 * written to the flash media, but may be some garbage.
915 * This function retries %UBI_IO_RETRIES times before giving up.
917 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
918 const void *buf, int offset, int len)
920 int err, pnum, tries, vol_id = vol->vol_id;
921 struct ubi_vid_io_buf *vidb;
922 struct ubi_vid_hdr *vid_hdr;
927 err = leb_write_lock(ubi, vol_id, lnum);
931 pnum = vol->eba_tbl->entries[lnum].pnum;
933 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
934 len, offset, vol_id, lnum, pnum);
936 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
938 ubi_warn(ubi, "failed to write data to PEB %d", pnum);
939 if (err == -EIO && ubi->bad_allowed)
940 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
948 * The logical eraseblock is not mapped. We have to get a free physical
949 * eraseblock and write the volume identifier header there first.
951 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
953 leb_write_unlock(ubi, vol_id, lnum);
957 vid_hdr = ubi_get_vid_hdr(vidb);
959 vid_hdr->vol_type = UBI_VID_DYNAMIC;
960 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
961 vid_hdr->vol_id = cpu_to_be32(vol_id);
962 vid_hdr->lnum = cpu_to_be32(lnum);
963 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
964 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
966 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
967 err = try_write_vid_and_data(vol, lnum, vidb, buf, offset, len);
968 if (err != -EIO || !ubi->bad_allowed)
972 * Fortunately, this is the first write operation to this
973 * physical eraseblock, so just put it and request a new one.
974 * We assume that if this physical eraseblock went bad, the
975 * erase code will handle that.
977 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
978 ubi_msg(ubi, "try another PEB");
981 ubi_free_vid_buf(vidb);
987 leb_write_unlock(ubi, vol_id, lnum);
993 * ubi_eba_write_leb_st - write data to static volume.
994 * @ubi: UBI device description object
995 * @vol: volume description object
996 * @lnum: logical eraseblock number
997 * @buf: data to write
998 * @len: how many bytes to write
999 * @used_ebs: how many logical eraseblocks will this volume contain
1001 * This function writes data to logical eraseblock @lnum of static volume
1002 * @vol. The @used_ebs argument should contain total number of logical
1003 * eraseblock in this static volume.
1005 * When writing to the last logical eraseblock, the @len argument doesn't have
1006 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
1007 * to the real data size, although the @buf buffer has to contain the
1008 * alignment. In all other cases, @len has to be aligned.
1010 * It is prohibited to write more than once to logical eraseblocks of static
1011 * volumes. This function returns zero in case of success and a negative error
1012 * code in case of failure.
1014 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
1015 int lnum, const void *buf, int len, int used_ebs)
1017 int err, tries, data_size = len, vol_id = vol->vol_id;
1018 struct ubi_vid_io_buf *vidb;
1019 struct ubi_vid_hdr *vid_hdr;
1025 if (lnum == used_ebs - 1)
1026 /* If this is the last LEB @len may be unaligned */
1027 len = ALIGN(data_size, ubi->min_io_size);
1029 ubi_assert(!(len & (ubi->min_io_size - 1)));
1031 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1035 vid_hdr = ubi_get_vid_hdr(vidb);
1037 err = leb_write_lock(ubi, vol_id, lnum);
1041 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1042 vid_hdr->vol_id = cpu_to_be32(vol_id);
1043 vid_hdr->lnum = cpu_to_be32(lnum);
1044 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1045 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1047 crc = crc32(UBI_CRC32_INIT, buf, data_size);
1048 vid_hdr->vol_type = UBI_VID_STATIC;
1049 vid_hdr->data_size = cpu_to_be32(data_size);
1050 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
1051 vid_hdr->data_crc = cpu_to_be32(crc);
1053 ubi_assert(vol->eba_tbl->entries[lnum].pnum < 0);
1055 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1056 err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1057 if (err != -EIO || !ubi->bad_allowed)
1060 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1061 ubi_msg(ubi, "try another PEB");
1067 leb_write_unlock(ubi, vol_id, lnum);
1070 ubi_free_vid_buf(vidb);
1076 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
1077 * @ubi: UBI device description object
1078 * @vol: volume description object
1079 * @lnum: logical eraseblock number
1080 * @buf: data to write
1081 * @len: how many bytes to write
1083 * This function changes the contents of a logical eraseblock atomically. @buf
1084 * has to contain new logical eraseblock data, and @len - the length of the
1085 * data, which has to be aligned. This function guarantees that in case of an
1086 * unclean reboot the old contents is preserved. Returns zero in case of
1087 * success and a negative error code in case of failure.
1089 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
1090 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
1092 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
1093 int lnum, const void *buf, int len)
1095 int err, tries, vol_id = vol->vol_id;
1096 struct ubi_vid_io_buf *vidb;
1097 struct ubi_vid_hdr *vid_hdr;
1105 * Special case when data length is zero. In this case the LEB
1106 * has to be unmapped and mapped somewhere else.
1108 err = ubi_eba_unmap_leb(ubi, vol, lnum);
1111 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
1114 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1118 vid_hdr = ubi_get_vid_hdr(vidb);
1120 mutex_lock(&ubi->alc_mutex);
1121 err = leb_write_lock(ubi, vol_id, lnum);
1125 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1126 vid_hdr->vol_id = cpu_to_be32(vol_id);
1127 vid_hdr->lnum = cpu_to_be32(lnum);
1128 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1129 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1131 crc = crc32(UBI_CRC32_INIT, buf, len);
1132 vid_hdr->vol_type = UBI_VID_DYNAMIC;
1133 vid_hdr->data_size = cpu_to_be32(len);
1134 vid_hdr->copy_flag = 1;
1135 vid_hdr->data_crc = cpu_to_be32(crc);
1137 dbg_eba("change LEB %d:%d", vol_id, lnum);
1139 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1140 err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1141 if (err != -EIO || !ubi->bad_allowed)
1144 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1145 ubi_msg(ubi, "try another PEB");
1149 * This flash device does not admit of bad eraseblocks or
1150 * something nasty and unexpected happened. Switch to read-only
1151 * mode just in case.
1156 leb_write_unlock(ubi, vol_id, lnum);
1159 mutex_unlock(&ubi->alc_mutex);
1160 ubi_free_vid_buf(vidb);
1165 * is_error_sane - check whether a read error is sane.
1166 * @err: code of the error happened during reading
1168 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1169 * cannot read data from the target PEB (an error @err happened). If the error
1170 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1171 * fatal and UBI will be switched to R/O mode later.
1173 * The idea is that we try not to switch to R/O mode if the read error is
1174 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1175 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1176 * mode, simply because we do not know what happened at the MTD level, and we
1177 * cannot handle this. E.g., the underlying driver may have become crazy, and
1178 * it is safer to switch to R/O mode to preserve the data.
1180 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1181 * which we have just written.
1183 static int is_error_sane(int err)
1185 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1186 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1192 * ubi_eba_copy_leb - copy logical eraseblock.
1193 * @ubi: UBI device description object
1194 * @from: physical eraseblock number from where to copy
1195 * @to: physical eraseblock number where to copy
1196 * @vid_hdr: VID header of the @from physical eraseblock
1198 * This function copies logical eraseblock from physical eraseblock @from to
1199 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1200 * function. Returns:
1201 * o %0 in case of success;
1202 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1203 * o a negative error code in case of failure.
1205 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1206 struct ubi_vid_io_buf *vidb)
1208 int err, vol_id, lnum, data_size, aldata_size, idx;
1209 struct ubi_vid_hdr *vid_hdr = ubi_get_vid_hdr(vidb);
1210 struct ubi_volume *vol;
1213 ubi_assert(rwsem_is_locked(&ubi->fm_eba_sem));
1215 vol_id = be32_to_cpu(vid_hdr->vol_id);
1216 lnum = be32_to_cpu(vid_hdr->lnum);
1218 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1220 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1221 data_size = be32_to_cpu(vid_hdr->data_size);
1222 aldata_size = ALIGN(data_size, ubi->min_io_size);
1224 data_size = aldata_size =
1225 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1227 idx = vol_id2idx(ubi, vol_id);
1228 spin_lock(&ubi->volumes_lock);
1230 * Note, we may race with volume deletion, which means that the volume
1231 * this logical eraseblock belongs to might be being deleted. Since the
1232 * volume deletion un-maps all the volume's logical eraseblocks, it will
1233 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1235 vol = ubi->volumes[idx];
1236 spin_unlock(&ubi->volumes_lock);
1238 /* No need to do further work, cancel */
1239 dbg_wl("volume %d is being removed, cancel", vol_id);
1240 return MOVE_CANCEL_RACE;
1244 * We do not want anybody to write to this logical eraseblock while we
1245 * are moving it, so lock it.
1247 * Note, we are using non-waiting locking here, because we cannot sleep
1248 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1249 * unmapping the LEB which is mapped to the PEB we are going to move
1250 * (@from). This task locks the LEB and goes sleep in the
1251 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1252 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1253 * LEB is already locked, we just do not move it and return
1254 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1255 * we do not know the reasons of the contention - it may be just a
1256 * normal I/O on this LEB, so we want to re-try.
1258 err = leb_write_trylock(ubi, vol_id, lnum);
1260 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1265 * The LEB might have been put meanwhile, and the task which put it is
1266 * probably waiting on @ubi->move_mutex. No need to continue the work,
1269 if (vol->eba_tbl->entries[lnum].pnum != from) {
1270 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1271 vol_id, lnum, from, vol->eba_tbl->entries[lnum].pnum);
1272 err = MOVE_CANCEL_RACE;
1273 goto out_unlock_leb;
1277 * OK, now the LEB is locked and we can safely start moving it. Since
1278 * this function utilizes the @ubi->peb_buf buffer which is shared
1279 * with some other functions - we lock the buffer by taking the
1282 mutex_lock(&ubi->buf_mutex);
1283 dbg_wl("read %d bytes of data", aldata_size);
1284 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1285 if (err && err != UBI_IO_BITFLIPS) {
1286 ubi_warn(ubi, "error %d while reading data from PEB %d",
1288 err = MOVE_SOURCE_RD_ERR;
1289 goto out_unlock_buf;
1293 * Now we have got to calculate how much data we have to copy. In
1294 * case of a static volume it is fairly easy - the VID header contains
1295 * the data size. In case of a dynamic volume it is more difficult - we
1296 * have to read the contents, cut 0xFF bytes from the end and copy only
1297 * the first part. We must do this to avoid writing 0xFF bytes as it
1298 * may have some side-effects. And not only this. It is important not
1299 * to include those 0xFFs to CRC because later the they may be filled
1302 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1303 aldata_size = data_size =
1304 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1307 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1311 * It may turn out to be that the whole @from physical eraseblock
1312 * contains only 0xFF bytes. Then we have to only write the VID header
1313 * and do not write any data. This also means we should not set
1314 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1316 if (data_size > 0) {
1317 vid_hdr->copy_flag = 1;
1318 vid_hdr->data_size = cpu_to_be32(data_size);
1319 vid_hdr->data_crc = cpu_to_be32(crc);
1321 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1323 err = ubi_io_write_vid_hdr(ubi, to, vidb);
1326 err = MOVE_TARGET_WR_ERR;
1327 goto out_unlock_buf;
1332 /* Read the VID header back and check if it was written correctly */
1333 err = ubi_io_read_vid_hdr(ubi, to, vidb, 1);
1335 if (err != UBI_IO_BITFLIPS) {
1336 ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1338 if (is_error_sane(err))
1339 err = MOVE_TARGET_RD_ERR;
1341 err = MOVE_TARGET_BITFLIPS;
1342 goto out_unlock_buf;
1345 if (data_size > 0) {
1346 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1349 err = MOVE_TARGET_WR_ERR;
1350 goto out_unlock_buf;
1356 ubi_assert(vol->eba_tbl->entries[lnum].pnum == from);
1357 vol->eba_tbl->entries[lnum].pnum = to;
1360 mutex_unlock(&ubi->buf_mutex);
1362 leb_write_unlock(ubi, vol_id, lnum);
1367 * print_rsvd_warning - warn about not having enough reserved PEBs.
1368 * @ubi: UBI device description object
1370 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1371 * cannot reserve enough PEBs for bad block handling. This function makes a
1372 * decision whether we have to print a warning or not. The algorithm is as
1374 * o if this is a new UBI image, then just print the warning
1375 * o if this is an UBI image which has already been used for some time, print
1376 * a warning only if we can reserve less than 10% of the expected amount of
1379 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1380 * of PEBs becomes smaller, which is normal and we do not want to scare users
1381 * with a warning every time they attach the MTD device. This was an issue
1382 * reported by real users.
1384 static void print_rsvd_warning(struct ubi_device *ubi,
1385 struct ubi_attach_info *ai)
1388 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1389 * large number to distinguish between newly flashed and used images.
1391 if (ai->max_sqnum > (1 << 18)) {
1392 int min = ubi->beb_rsvd_level / 10;
1396 if (ubi->beb_rsvd_pebs > min)
1400 ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1401 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1402 if (ubi->corr_peb_count)
1403 ubi_warn(ubi, "%d PEBs are corrupted and not used",
1404 ubi->corr_peb_count);
1408 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1409 * @ubi: UBI device description object
1410 * @ai_fastmap: UBI attach info object created by fastmap
1411 * @ai_scan: UBI attach info object created by scanning
1413 * Returns < 0 in case of an internal error, 0 otherwise.
1414 * If a bad EBA table entry was found it will be printed out and
1415 * ubi_assert() triggers.
1417 int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1418 struct ubi_attach_info *ai_scan)
1420 int i, j, num_volumes, ret = 0;
1421 int **scan_eba, **fm_eba;
1422 struct ubi_ainf_volume *av;
1423 struct ubi_volume *vol;
1424 struct ubi_ainf_peb *aeb;
1427 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1429 scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1433 fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1439 for (i = 0; i < num_volumes; i++) {
1440 vol = ubi->volumes[i];
1444 scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1451 fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1458 for (j = 0; j < vol->reserved_pebs; j++)
1459 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1461 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1465 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1466 scan_eba[i][aeb->lnum] = aeb->pnum;
1468 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1472 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1473 fm_eba[i][aeb->lnum] = aeb->pnum;
1475 for (j = 0; j < vol->reserved_pebs; j++) {
1476 if (scan_eba[i][j] != fm_eba[i][j]) {
1477 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1478 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1481 ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1482 vol->vol_id, j, fm_eba[i][j],
1490 for (i = 0; i < num_volumes; i++) {
1491 if (!ubi->volumes[i])
1504 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1505 * @ubi: UBI device description object
1506 * @ai: attaching information
1508 * This function returns zero in case of success and a negative error code in
1511 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1513 int i, err, num_volumes;
1514 struct ubi_ainf_volume *av;
1515 struct ubi_volume *vol;
1516 struct ubi_ainf_peb *aeb;
1519 dbg_eba("initialize EBA sub-system");
1521 spin_lock_init(&ubi->ltree_lock);
1522 mutex_init(&ubi->alc_mutex);
1523 ubi->ltree = RB_ROOT;
1525 ubi->global_sqnum = ai->max_sqnum + 1;
1526 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1528 for (i = 0; i < num_volumes; i++) {
1529 struct ubi_eba_table *tbl;
1531 vol = ubi->volumes[i];
1537 tbl = ubi_eba_create_table(vol, vol->reserved_pebs);
1543 ubi_eba_replace_table(vol, tbl);
1545 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1549 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1550 if (aeb->lnum >= vol->reserved_pebs) {
1552 * This may happen in case of an unclean reboot
1555 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1557 struct ubi_eba_entry *entry;
1559 entry = &vol->eba_tbl->entries[aeb->lnum];
1560 entry->pnum = aeb->pnum;
1565 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1566 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1567 ubi->avail_pebs, EBA_RESERVED_PEBS);
1568 if (ubi->corr_peb_count)
1569 ubi_err(ubi, "%d PEBs are corrupted and not used",
1570 ubi->corr_peb_count);
1574 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1575 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1577 if (ubi->bad_allowed) {
1578 ubi_calculate_reserved(ubi);
1580 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1581 /* No enough free physical eraseblocks */
1582 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1583 print_rsvd_warning(ubi, ai);
1585 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1587 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1588 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1591 dbg_eba("EBA sub-system is initialized");
1595 for (i = 0; i < num_volumes; i++) {
1596 if (!ubi->volumes[i])
1598 ubi_eba_replace_table(ubi->volumes[i], NULL);