ARM: pxa/raumfeld: mark rotary encoder properties as __initconst
[linux-2.6/btrfs-unstable.git] / drivers / mtd / ubi / eba.c
blob388e46be6ad92805f2a6633da6960d8c56b1b837
1 /*
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>
47 #include "ubi.h"
49 /* Number of physical eraseblocks reserved for atomic LEB change operation */
50 #define EBA_RESERVED_PEBS 1
52 /**
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
58 * entries table.
60 struct ubi_eba_entry {
61 int pnum;
64 /**
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
70 * changes.
72 struct ubi_eba_table {
73 struct ubi_eba_entry *entries;
76 /**
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
82 * counter.
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);
92 return sqnum;
95 /**
96 * ubi_get_compat - get compatibility flags of a volume.
97 * @ubi: UBI device description object
98 * @vol_id: volume ID
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;
107 return 0;
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)
123 ldesc->lnum = lnum;
124 ldesc->pnum = vol->eba_tbl->entries[lnum].pnum;
128 * ubi_eba_create_table - allocate a new EBA table and initialize it with all
129 * LEBs unmapped
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,
137 int nentries)
139 struct ubi_eba_table *tbl;
140 int err = -ENOMEM;
141 int i;
143 tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
144 if (!tbl)
145 return ERR_PTR(-ENOMEM);
147 tbl->entries = kmalloc_array(nentries, sizeof(*tbl->entries),
148 GFP_KERNEL);
149 if (!tbl->entries)
150 goto err;
152 for (i = 0; i < nentries; i++)
153 tbl->entries[i].pnum = UBI_LEB_UNMAPPED;
155 return tbl;
157 err:
158 kfree(tbl->entries);
159 kfree(tbl);
161 return ERR_PTR(err);
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)
172 if (!tbl)
173 return;
175 kfree(tbl->entries);
176 kfree(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
182 * @dst: destination
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,
188 int nentries)
190 struct ubi_eba_table *src;
191 int i;
193 ubi_assert(dst && vol && vol->eba_tbl);
195 src = 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);
211 vol->eba_tbl = tbl;
215 * ltree_lookup - look up the lock tree.
216 * @ubi: UBI device description object
217 * @vol_id: volume ID
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,
225 int lnum)
227 struct rb_node *p;
229 p = ubi->ltree.rb_node;
230 while (p) {
231 struct ubi_ltree_entry *le;
233 le = rb_entry(p, struct ubi_ltree_entry, rb);
235 if (vol_id < le->vol_id)
236 p = p->rb_left;
237 else if (vol_id > le->vol_id)
238 p = p->rb_right;
239 else {
240 if (lnum < le->lnum)
241 p = p->rb_left;
242 else if (lnum > le->lnum)
243 p = p->rb_right;
244 else
245 return le;
249 return NULL;
253 * ltree_add_entry - add new entry to the lock tree.
254 * @ubi: UBI device description object
255 * @vol_id: volume ID
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
261 * failed.
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);
269 if (!le)
270 return ERR_PTR(-ENOMEM);
272 le->users = 0;
273 init_rwsem(&le->mutex);
274 le->vol_id = vol_id;
275 le->lnum = lnum;
277 spin_lock(&ubi->ltree_lock);
278 le1 = ltree_lookup(ubi, vol_id, lnum);
280 if (le1) {
282 * This logical eraseblock is already locked. The newly
283 * allocated lock entry is not needed.
285 le_free = le;
286 le = le1;
287 } else {
288 struct rb_node **p, *parent = NULL;
291 * No lock entry, add the newly allocated one to the
292 * @ubi->ltree RB-tree.
294 le_free = NULL;
296 p = &ubi->ltree.rb_node;
297 while (*p) {
298 parent = *p;
299 le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
301 if (vol_id < le1->vol_id)
302 p = &(*p)->rb_left;
303 else if (vol_id > le1->vol_id)
304 p = &(*p)->rb_right;
305 else {
306 ubi_assert(lnum != le1->lnum);
307 if (lnum < le1->lnum)
308 p = &(*p)->rb_left;
309 else
310 p = &(*p)->rb_right;
314 rb_link_node(&le->rb, parent, p);
315 rb_insert_color(&le->rb, &ubi->ltree);
317 le->users += 1;
318 spin_unlock(&ubi->ltree_lock);
320 kfree(le_free);
321 return le;
325 * leb_read_lock - lock logical eraseblock for reading.
326 * @ubi: UBI device description object
327 * @vol_id: volume ID
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);
338 if (IS_ERR(le))
339 return PTR_ERR(le);
340 down_read(&le->mutex);
341 return 0;
345 * leb_read_unlock - unlock logical eraseblock.
346 * @ubi: UBI device description object
347 * @vol_id: volume ID
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);
356 le->users -= 1;
357 ubi_assert(le->users >= 0);
358 up_read(&le->mutex);
359 if (le->users == 0) {
360 rb_erase(&le->rb, &ubi->ltree);
361 kfree(le);
363 spin_unlock(&ubi->ltree_lock);
367 * leb_write_lock - lock logical eraseblock for writing.
368 * @ubi: UBI device description object
369 * @vol_id: volume ID
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);
380 if (IS_ERR(le))
381 return PTR_ERR(le);
382 down_write(&le->mutex);
383 return 0;
387 * leb_write_lock - lock logical eraseblock for writing.
388 * @ubi: UBI device description object
389 * @vol_id: volume ID
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
395 * failure.
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);
402 if (IS_ERR(le))
403 return PTR_ERR(le);
404 if (down_write_trylock(&le->mutex))
405 return 0;
407 /* Contention, cancel */
408 spin_lock(&ubi->ltree_lock);
409 le->users -= 1;
410 ubi_assert(le->users >= 0);
411 if (le->users == 0) {
412 rb_erase(&le->rb, &ubi->ltree);
413 kfree(le);
415 spin_unlock(&ubi->ltree_lock);
417 return 1;
421 * leb_write_unlock - unlock logical eraseblock.
422 * @ubi: UBI device description object
423 * @vol_id: volume ID
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);
432 le->users -= 1;
433 ubi_assert(le->users >= 0);
434 up_write(&le->mutex);
435 if (le->users == 0) {
436 rb_erase(&le->rb, &ubi->ltree);
437 kfree(le);
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,
465 int lnum)
467 int err, pnum, vol_id = vol->vol_id;
469 if (ubi->ro_mode)
470 return -EROFS;
472 err = leb_write_lock(ubi, vol_id, lnum);
473 if (err)
474 return err;
476 pnum = vol->eba_tbl->entries[lnum].pnum;
477 if (pnum < 0)
478 /* This logical eraseblock is already unmapped */
479 goto out_unlock;
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);
488 out_unlock:
489 leb_write_unlock(ubi, vol_id, lnum);
490 return err;
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);
521 if (err)
522 return err;
524 pnum = vol->eba_tbl->entries[lnum].pnum;
525 if (pnum < 0) {
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);
536 return 0;
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)
543 check = 0;
545 retry:
546 if (check) {
547 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
548 if (!vidb) {
549 err = -ENOMEM;
550 goto out_unlock;
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) {
557 if (err > 0) {
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
564 * not implemented.
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",
569 pnum, vol_id, lnum);
570 err = -EBADMSG;
571 } else {
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
583 * LEBs in any case.
585 if (ubi->fast_attach) {
586 err = -EBADMSG;
587 } else {
588 err = -EINVAL;
589 ubi_ro_mode(ubi);
593 goto out_free;
594 } else if (err == UBI_IO_BITFLIPS)
595 scrub = 1;
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);
605 if (err) {
606 if (err == UBI_IO_BITFLIPS)
607 scrub = 1;
608 else if (mtd_is_eccerr(err)) {
609 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
610 goto out_unlock;
611 scrub = 1;
612 if (!check) {
613 ubi_msg(ubi, "force data checking");
614 check = 1;
615 goto retry;
617 } else
618 goto out_unlock;
621 if (check) {
622 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
623 if (crc1 != crc) {
624 ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
625 crc1, crc);
626 err = -EBADMSG;
627 goto out_unlock;
631 if (scrub)
632 err = ubi_wl_scrub_peb(ubi, pnum);
634 leb_read_unlock(ubi, vol_id, lnum);
635 return err;
637 out_free:
638 ubi_free_vid_buf(vidb);
639 out_unlock:
640 leb_read_unlock(ubi, vol_id, lnum);
641 return err;
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
656 * list.
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,
660 int check)
662 int to_read;
663 int ret;
664 struct scatterlist *sg;
666 for (;;) {
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)
670 to_read = len;
671 else
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,
676 to_read, check);
677 if (ret < 0)
678 return ret;
680 offset += to_read;
681 len -= to_read;
682 if (!len) {
683 sgl->page_pos += to_read;
684 if (sgl->page_pos == sg->length) {
685 sgl->list_pos++;
686 sgl->page_pos = 0;
689 break;
692 sgl->list_pos++;
693 sgl->page_pos = 0;
696 return ret;
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
707 * @vidb: VID buffer
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;
724 uint32_t crc;
726 *retry = false;
728 new_pnum = ubi_wl_get_peb(ubi);
729 if (new_pnum < 0) {
730 err = new_pnum;
731 goto out_put;
734 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
735 pnum, new_pnum);
737 err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
738 if (err && err != UBI_IO_BITFLIPS) {
739 if (err > 0)
740 err = -EIO;
741 goto out_put;
744 vid_hdr = ubi_get_vid_hdr(vidb);
745 ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
747 mutex_lock(&ubi->buf_mutex);
748 memset(ubi->peb_buf + offset, 0xFF, len);
750 /* Read everything before the area where the write failure happened */
751 if (offset > 0) {
752 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
753 if (err && err != UBI_IO_BITFLIPS)
754 goto out_unlock;
757 *retry = true;
759 memcpy(ubi->peb_buf + offset, buf, len);
761 data_size = offset + len;
762 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
763 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
764 vid_hdr->copy_flag = 1;
765 vid_hdr->data_size = cpu_to_be32(data_size);
766 vid_hdr->data_crc = cpu_to_be32(crc);
767 err = ubi_io_write_vid_hdr(ubi, new_pnum, vidb);
768 if (err)
769 goto out_unlock;
771 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
773 out_unlock:
774 mutex_unlock(&ubi->buf_mutex);
776 if (!err)
777 vol->eba_tbl->entries[lnum].pnum = new_pnum;
779 out_put:
780 up_read(&ubi->fm_eba_sem);
782 if (!err) {
783 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
784 ubi_msg(ubi, "data was successfully recovered");
785 } else if (new_pnum >= 0) {
787 * Bad luck? This physical eraseblock is bad too? Crud. Let's
788 * try to get another one.
790 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
791 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
794 return err;
798 * recover_peb - recover from write failure.
799 * @ubi: UBI device description object
800 * @pnum: the physical eraseblock to recover
801 * @vol_id: volume ID
802 * @lnum: logical eraseblock number
803 * @buf: data which was not written because of the write failure
804 * @offset: offset of the failed write
805 * @len: how many bytes should have been written
807 * This function is called in case of a write failure and moves all good data
808 * from the potentially bad physical eraseblock to a good physical eraseblock.
809 * This function also writes the data which was not written due to the failure.
810 * Returns 0 in case of success, and a negative error code in case of failure.
811 * This function tries %UBI_IO_RETRIES before giving up.
813 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
814 const void *buf, int offset, int len)
816 int err, idx = vol_id2idx(ubi, vol_id), tries;
817 struct ubi_volume *vol = ubi->volumes[idx];
818 struct ubi_vid_io_buf *vidb;
820 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
821 if (!vidb)
822 return -ENOMEM;
824 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
825 bool retry;
827 err = try_recover_peb(vol, pnum, lnum, buf, offset, len, vidb,
828 &retry);
829 if (!err || !retry)
830 break;
832 ubi_msg(ubi, "try again");
835 ubi_free_vid_buf(vidb);
837 return err;
841 * try_write_vid_and_data - try to write VID header and data to a new PEB.
842 * @vol: volume description object
843 * @lnum: logical eraseblock number
844 * @vidb: the VID buffer to write
845 * @buf: buffer containing the data
846 * @offset: where to start writing data
847 * @len: how many bytes should be written
849 * This function tries to write VID header and data belonging to logical
850 * eraseblock @lnum of volume @vol to a new physical eraseblock. Returns zero
851 * in case of success and a negative error code in case of failure.
852 * In case of error, it is possible that something was still written to the
853 * flash media, but may be some garbage.
855 static int try_write_vid_and_data(struct ubi_volume *vol, int lnum,
856 struct ubi_vid_io_buf *vidb, const void *buf,
857 int offset, int len)
859 struct ubi_device *ubi = vol->ubi;
860 int pnum, opnum, err, vol_id = vol->vol_id;
862 pnum = ubi_wl_get_peb(ubi);
863 if (pnum < 0) {
864 err = pnum;
865 goto out_put;
868 opnum = vol->eba_tbl->entries[lnum].pnum;
870 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
871 len, offset, vol_id, lnum, pnum);
873 err = ubi_io_write_vid_hdr(ubi, pnum, vidb);
874 if (err) {
875 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
876 vol_id, lnum, pnum);
877 goto out_put;
880 if (len) {
881 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
882 if (err) {
883 ubi_warn(ubi,
884 "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
885 len, offset, vol_id, lnum, pnum);
886 goto out_put;
890 vol->eba_tbl->entries[lnum].pnum = pnum;
892 out_put:
893 up_read(&ubi->fm_eba_sem);
895 if (err && pnum >= 0)
896 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
897 else if (!err && opnum >= 0)
898 err = ubi_wl_put_peb(ubi, vol_id, lnum, opnum, 0);
900 return err;
904 * ubi_eba_write_leb - write data to dynamic volume.
905 * @ubi: UBI device description object
906 * @vol: volume description object
907 * @lnum: logical eraseblock number
908 * @buf: the data to write
909 * @offset: offset within the logical eraseblock where to write
910 * @len: how many bytes to write
912 * This function writes data to logical eraseblock @lnum of a dynamic volume
913 * @vol. Returns zero in case of success and a negative error code in case
914 * of failure. In case of error, it is possible that something was still
915 * written to the flash media, but may be some garbage.
916 * This function retries %UBI_IO_RETRIES times before giving up.
918 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
919 const void *buf, int offset, int len)
921 int err, pnum, tries, vol_id = vol->vol_id;
922 struct ubi_vid_io_buf *vidb;
923 struct ubi_vid_hdr *vid_hdr;
925 if (ubi->ro_mode)
926 return -EROFS;
928 err = leb_write_lock(ubi, vol_id, lnum);
929 if (err)
930 return err;
932 pnum = vol->eba_tbl->entries[lnum].pnum;
933 if (pnum >= 0) {
934 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
935 len, offset, vol_id, lnum, pnum);
937 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
938 if (err) {
939 ubi_warn(ubi, "failed to write data to PEB %d", pnum);
940 if (err == -EIO && ubi->bad_allowed)
941 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
942 offset, len);
945 goto out;
949 * The logical eraseblock is not mapped. We have to get a free physical
950 * eraseblock and write the volume identifier header there first.
952 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
953 if (!vidb) {
954 leb_write_unlock(ubi, vol_id, lnum);
955 return -ENOMEM;
958 vid_hdr = ubi_get_vid_hdr(vidb);
960 vid_hdr->vol_type = UBI_VID_DYNAMIC;
961 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
962 vid_hdr->vol_id = cpu_to_be32(vol_id);
963 vid_hdr->lnum = cpu_to_be32(lnum);
964 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
965 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
967 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
968 err = try_write_vid_and_data(vol, lnum, vidb, buf, offset, len);
969 if (err != -EIO || !ubi->bad_allowed)
970 break;
973 * Fortunately, this is the first write operation to this
974 * physical eraseblock, so just put it and request a new one.
975 * We assume that if this physical eraseblock went bad, the
976 * erase code will handle that.
978 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
979 ubi_msg(ubi, "try another PEB");
982 ubi_free_vid_buf(vidb);
984 out:
985 if (err)
986 ubi_ro_mode(ubi);
988 leb_write_unlock(ubi, vol_id, lnum);
990 return err;
994 * ubi_eba_write_leb_st - write data to static volume.
995 * @ubi: UBI device description object
996 * @vol: volume description object
997 * @lnum: logical eraseblock number
998 * @buf: data to write
999 * @len: how many bytes to write
1000 * @used_ebs: how many logical eraseblocks will this volume contain
1002 * This function writes data to logical eraseblock @lnum of static volume
1003 * @vol. The @used_ebs argument should contain total number of logical
1004 * eraseblock in this static volume.
1006 * When writing to the last logical eraseblock, the @len argument doesn't have
1007 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
1008 * to the real data size, although the @buf buffer has to contain the
1009 * alignment. In all other cases, @len has to be aligned.
1011 * It is prohibited to write more than once to logical eraseblocks of static
1012 * volumes. This function returns zero in case of success and a negative error
1013 * code in case of failure.
1015 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
1016 int lnum, const void *buf, int len, int used_ebs)
1018 int err, tries, data_size = len, vol_id = vol->vol_id;
1019 struct ubi_vid_io_buf *vidb;
1020 struct ubi_vid_hdr *vid_hdr;
1021 uint32_t crc;
1023 if (ubi->ro_mode)
1024 return -EROFS;
1026 if (lnum == used_ebs - 1)
1027 /* If this is the last LEB @len may be unaligned */
1028 len = ALIGN(data_size, ubi->min_io_size);
1029 else
1030 ubi_assert(!(len & (ubi->min_io_size - 1)));
1032 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1033 if (!vidb)
1034 return -ENOMEM;
1036 vid_hdr = ubi_get_vid_hdr(vidb);
1038 err = leb_write_lock(ubi, vol_id, lnum);
1039 if (err)
1040 goto out;
1042 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1043 vid_hdr->vol_id = cpu_to_be32(vol_id);
1044 vid_hdr->lnum = cpu_to_be32(lnum);
1045 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1046 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1048 crc = crc32(UBI_CRC32_INIT, buf, data_size);
1049 vid_hdr->vol_type = UBI_VID_STATIC;
1050 vid_hdr->data_size = cpu_to_be32(data_size);
1051 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
1052 vid_hdr->data_crc = cpu_to_be32(crc);
1054 ubi_assert(vol->eba_tbl->entries[lnum].pnum < 0);
1056 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1057 err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1058 if (err != -EIO || !ubi->bad_allowed)
1059 break;
1061 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1062 ubi_msg(ubi, "try another PEB");
1065 if (err)
1066 ubi_ro_mode(ubi);
1068 leb_write_unlock(ubi, vol_id, lnum);
1070 out:
1071 ubi_free_vid_buf(vidb);
1073 return err;
1077 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
1078 * @ubi: UBI device description object
1079 * @vol: volume description object
1080 * @lnum: logical eraseblock number
1081 * @buf: data to write
1082 * @len: how many bytes to write
1084 * This function changes the contents of a logical eraseblock atomically. @buf
1085 * has to contain new logical eraseblock data, and @len - the length of the
1086 * data, which has to be aligned. This function guarantees that in case of an
1087 * unclean reboot the old contents is preserved. Returns zero in case of
1088 * success and a negative error code in case of failure.
1090 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
1091 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
1093 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
1094 int lnum, const void *buf, int len)
1096 int err, tries, vol_id = vol->vol_id;
1097 struct ubi_vid_io_buf *vidb;
1098 struct ubi_vid_hdr *vid_hdr;
1099 uint32_t crc;
1101 if (ubi->ro_mode)
1102 return -EROFS;
1104 if (len == 0) {
1106 * Special case when data length is zero. In this case the LEB
1107 * has to be unmapped and mapped somewhere else.
1109 err = ubi_eba_unmap_leb(ubi, vol, lnum);
1110 if (err)
1111 return err;
1112 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
1115 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1116 if (!vidb)
1117 return -ENOMEM;
1119 vid_hdr = ubi_get_vid_hdr(vidb);
1121 mutex_lock(&ubi->alc_mutex);
1122 err = leb_write_lock(ubi, vol_id, lnum);
1123 if (err)
1124 goto out_mutex;
1126 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1127 vid_hdr->vol_id = cpu_to_be32(vol_id);
1128 vid_hdr->lnum = cpu_to_be32(lnum);
1129 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1130 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1132 crc = crc32(UBI_CRC32_INIT, buf, len);
1133 vid_hdr->vol_type = UBI_VID_DYNAMIC;
1134 vid_hdr->data_size = cpu_to_be32(len);
1135 vid_hdr->copy_flag = 1;
1136 vid_hdr->data_crc = cpu_to_be32(crc);
1138 dbg_eba("change LEB %d:%d", vol_id, lnum);
1140 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1141 err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1142 if (err != -EIO || !ubi->bad_allowed)
1143 break;
1145 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1146 ubi_msg(ubi, "try another PEB");
1150 * This flash device does not admit of bad eraseblocks or
1151 * something nasty and unexpected happened. Switch to read-only
1152 * mode just in case.
1154 if (err)
1155 ubi_ro_mode(ubi);
1157 leb_write_unlock(ubi, vol_id, lnum);
1159 out_mutex:
1160 mutex_unlock(&ubi->alc_mutex);
1161 ubi_free_vid_buf(vidb);
1162 return err;
1166 * is_error_sane - check whether a read error is sane.
1167 * @err: code of the error happened during reading
1169 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1170 * cannot read data from the target PEB (an error @err happened). If the error
1171 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1172 * fatal and UBI will be switched to R/O mode later.
1174 * The idea is that we try not to switch to R/O mode if the read error is
1175 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1176 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1177 * mode, simply because we do not know what happened at the MTD level, and we
1178 * cannot handle this. E.g., the underlying driver may have become crazy, and
1179 * it is safer to switch to R/O mode to preserve the data.
1181 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1182 * which we have just written.
1184 static int is_error_sane(int err)
1186 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1187 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1188 return 0;
1189 return 1;
1193 * ubi_eba_copy_leb - copy logical eraseblock.
1194 * @ubi: UBI device description object
1195 * @from: physical eraseblock number from where to copy
1196 * @to: physical eraseblock number where to copy
1197 * @vid_hdr: VID header of the @from physical eraseblock
1199 * This function copies logical eraseblock from physical eraseblock @from to
1200 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1201 * function. Returns:
1202 * o %0 in case of success;
1203 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1204 * o a negative error code in case of failure.
1206 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1207 struct ubi_vid_io_buf *vidb)
1209 int err, vol_id, lnum, data_size, aldata_size, idx;
1210 struct ubi_vid_hdr *vid_hdr = ubi_get_vid_hdr(vidb);
1211 struct ubi_volume *vol;
1212 uint32_t crc;
1214 ubi_assert(rwsem_is_locked(&ubi->fm_eba_sem));
1216 vol_id = be32_to_cpu(vid_hdr->vol_id);
1217 lnum = be32_to_cpu(vid_hdr->lnum);
1219 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1221 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1222 data_size = be32_to_cpu(vid_hdr->data_size);
1223 aldata_size = ALIGN(data_size, ubi->min_io_size);
1224 } else
1225 data_size = aldata_size =
1226 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1228 idx = vol_id2idx(ubi, vol_id);
1229 spin_lock(&ubi->volumes_lock);
1231 * Note, we may race with volume deletion, which means that the volume
1232 * this logical eraseblock belongs to might be being deleted. Since the
1233 * volume deletion un-maps all the volume's logical eraseblocks, it will
1234 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1236 vol = ubi->volumes[idx];
1237 spin_unlock(&ubi->volumes_lock);
1238 if (!vol) {
1239 /* No need to do further work, cancel */
1240 dbg_wl("volume %d is being removed, cancel", vol_id);
1241 return MOVE_CANCEL_RACE;
1245 * We do not want anybody to write to this logical eraseblock while we
1246 * are moving it, so lock it.
1248 * Note, we are using non-waiting locking here, because we cannot sleep
1249 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1250 * unmapping the LEB which is mapped to the PEB we are going to move
1251 * (@from). This task locks the LEB and goes sleep in the
1252 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1253 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1254 * LEB is already locked, we just do not move it and return
1255 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1256 * we do not know the reasons of the contention - it may be just a
1257 * normal I/O on this LEB, so we want to re-try.
1259 err = leb_write_trylock(ubi, vol_id, lnum);
1260 if (err) {
1261 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1262 return MOVE_RETRY;
1266 * The LEB might have been put meanwhile, and the task which put it is
1267 * probably waiting on @ubi->move_mutex. No need to continue the work,
1268 * cancel it.
1270 if (vol->eba_tbl->entries[lnum].pnum != from) {
1271 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1272 vol_id, lnum, from, vol->eba_tbl->entries[lnum].pnum);
1273 err = MOVE_CANCEL_RACE;
1274 goto out_unlock_leb;
1278 * OK, now the LEB is locked and we can safely start moving it. Since
1279 * this function utilizes the @ubi->peb_buf buffer which is shared
1280 * with some other functions - we lock the buffer by taking the
1281 * @ubi->buf_mutex.
1283 mutex_lock(&ubi->buf_mutex);
1284 dbg_wl("read %d bytes of data", aldata_size);
1285 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1286 if (err && err != UBI_IO_BITFLIPS) {
1287 ubi_warn(ubi, "error %d while reading data from PEB %d",
1288 err, from);
1289 err = MOVE_SOURCE_RD_ERR;
1290 goto out_unlock_buf;
1294 * Now we have got to calculate how much data we have to copy. In
1295 * case of a static volume it is fairly easy - the VID header contains
1296 * the data size. In case of a dynamic volume it is more difficult - we
1297 * have to read the contents, cut 0xFF bytes from the end and copy only
1298 * the first part. We must do this to avoid writing 0xFF bytes as it
1299 * may have some side-effects. And not only this. It is important not
1300 * to include those 0xFFs to CRC because later the they may be filled
1301 * by data.
1303 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1304 aldata_size = data_size =
1305 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1307 cond_resched();
1308 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1309 cond_resched();
1312 * It may turn out to be that the whole @from physical eraseblock
1313 * contains only 0xFF bytes. Then we have to only write the VID header
1314 * and do not write any data. This also means we should not set
1315 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1317 if (data_size > 0) {
1318 vid_hdr->copy_flag = 1;
1319 vid_hdr->data_size = cpu_to_be32(data_size);
1320 vid_hdr->data_crc = cpu_to_be32(crc);
1322 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1324 err = ubi_io_write_vid_hdr(ubi, to, vidb);
1325 if (err) {
1326 if (err == -EIO)
1327 err = MOVE_TARGET_WR_ERR;
1328 goto out_unlock_buf;
1331 cond_resched();
1333 /* Read the VID header back and check if it was written correctly */
1334 err = ubi_io_read_vid_hdr(ubi, to, vidb, 1);
1335 if (err) {
1336 if (err != UBI_IO_BITFLIPS) {
1337 ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1338 err, to);
1339 if (is_error_sane(err))
1340 err = MOVE_TARGET_RD_ERR;
1341 } else
1342 err = MOVE_TARGET_BITFLIPS;
1343 goto out_unlock_buf;
1346 if (data_size > 0) {
1347 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1348 if (err) {
1349 if (err == -EIO)
1350 err = MOVE_TARGET_WR_ERR;
1351 goto out_unlock_buf;
1354 cond_resched();
1357 ubi_assert(vol->eba_tbl->entries[lnum].pnum == from);
1358 vol->eba_tbl->entries[lnum].pnum = to;
1360 out_unlock_buf:
1361 mutex_unlock(&ubi->buf_mutex);
1362 out_unlock_leb:
1363 leb_write_unlock(ubi, vol_id, lnum);
1364 return err;
1368 * print_rsvd_warning - warn about not having enough reserved PEBs.
1369 * @ubi: UBI device description object
1371 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1372 * cannot reserve enough PEBs for bad block handling. This function makes a
1373 * decision whether we have to print a warning or not. The algorithm is as
1374 * follows:
1375 * o if this is a new UBI image, then just print the warning
1376 * o if this is an UBI image which has already been used for some time, print
1377 * a warning only if we can reserve less than 10% of the expected amount of
1378 * the reserved PEB.
1380 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1381 * of PEBs becomes smaller, which is normal and we do not want to scare users
1382 * with a warning every time they attach the MTD device. This was an issue
1383 * reported by real users.
1385 static void print_rsvd_warning(struct ubi_device *ubi,
1386 struct ubi_attach_info *ai)
1389 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1390 * large number to distinguish between newly flashed and used images.
1392 if (ai->max_sqnum > (1 << 18)) {
1393 int min = ubi->beb_rsvd_level / 10;
1395 if (!min)
1396 min = 1;
1397 if (ubi->beb_rsvd_pebs > min)
1398 return;
1401 ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1402 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1403 if (ubi->corr_peb_count)
1404 ubi_warn(ubi, "%d PEBs are corrupted and not used",
1405 ubi->corr_peb_count);
1409 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1410 * @ubi: UBI device description object
1411 * @ai_fastmap: UBI attach info object created by fastmap
1412 * @ai_scan: UBI attach info object created by scanning
1414 * Returns < 0 in case of an internal error, 0 otherwise.
1415 * If a bad EBA table entry was found it will be printed out and
1416 * ubi_assert() triggers.
1418 int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1419 struct ubi_attach_info *ai_scan)
1421 int i, j, num_volumes, ret = 0;
1422 int **scan_eba, **fm_eba;
1423 struct ubi_ainf_volume *av;
1424 struct ubi_volume *vol;
1425 struct ubi_ainf_peb *aeb;
1426 struct rb_node *rb;
1428 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1430 scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1431 if (!scan_eba)
1432 return -ENOMEM;
1434 fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1435 if (!fm_eba) {
1436 kfree(scan_eba);
1437 return -ENOMEM;
1440 for (i = 0; i < num_volumes; i++) {
1441 vol = ubi->volumes[i];
1442 if (!vol)
1443 continue;
1445 scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1446 GFP_KERNEL);
1447 if (!scan_eba[i]) {
1448 ret = -ENOMEM;
1449 goto out_free;
1452 fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1453 GFP_KERNEL);
1454 if (!fm_eba[i]) {
1455 ret = -ENOMEM;
1456 goto out_free;
1459 for (j = 0; j < vol->reserved_pebs; j++)
1460 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1462 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1463 if (!av)
1464 continue;
1466 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1467 scan_eba[i][aeb->lnum] = aeb->pnum;
1469 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1470 if (!av)
1471 continue;
1473 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1474 fm_eba[i][aeb->lnum] = aeb->pnum;
1476 for (j = 0; j < vol->reserved_pebs; j++) {
1477 if (scan_eba[i][j] != fm_eba[i][j]) {
1478 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1479 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1480 continue;
1482 ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1483 vol->vol_id, j, fm_eba[i][j],
1484 scan_eba[i][j]);
1485 ubi_assert(0);
1490 out_free:
1491 for (i = 0; i < num_volumes; i++) {
1492 if (!ubi->volumes[i])
1493 continue;
1495 kfree(scan_eba[i]);
1496 kfree(fm_eba[i]);
1499 kfree(scan_eba);
1500 kfree(fm_eba);
1501 return ret;
1505 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1506 * @ubi: UBI device description object
1507 * @ai: attaching information
1509 * This function returns zero in case of success and a negative error code in
1510 * case of failure.
1512 int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1514 int i, err, num_volumes;
1515 struct ubi_ainf_volume *av;
1516 struct ubi_volume *vol;
1517 struct ubi_ainf_peb *aeb;
1518 struct rb_node *rb;
1520 dbg_eba("initialize EBA sub-system");
1522 spin_lock_init(&ubi->ltree_lock);
1523 mutex_init(&ubi->alc_mutex);
1524 ubi->ltree = RB_ROOT;
1526 ubi->global_sqnum = ai->max_sqnum + 1;
1527 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1529 for (i = 0; i < num_volumes; i++) {
1530 struct ubi_eba_table *tbl;
1532 vol = ubi->volumes[i];
1533 if (!vol)
1534 continue;
1536 cond_resched();
1538 tbl = ubi_eba_create_table(vol, vol->reserved_pebs);
1539 if (IS_ERR(tbl)) {
1540 err = PTR_ERR(tbl);
1541 goto out_free;
1544 ubi_eba_replace_table(vol, tbl);
1546 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1547 if (!av)
1548 continue;
1550 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1551 if (aeb->lnum >= vol->reserved_pebs) {
1553 * This may happen in case of an unclean reboot
1554 * during re-size.
1556 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1557 } else {
1558 struct ubi_eba_entry *entry;
1560 entry = &vol->eba_tbl->entries[aeb->lnum];
1561 entry->pnum = aeb->pnum;
1566 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1567 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1568 ubi->avail_pebs, EBA_RESERVED_PEBS);
1569 if (ubi->corr_peb_count)
1570 ubi_err(ubi, "%d PEBs are corrupted and not used",
1571 ubi->corr_peb_count);
1572 err = -ENOSPC;
1573 goto out_free;
1575 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1576 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1578 if (ubi->bad_allowed) {
1579 ubi_calculate_reserved(ubi);
1581 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1582 /* No enough free physical eraseblocks */
1583 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1584 print_rsvd_warning(ubi, ai);
1585 } else
1586 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1588 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1589 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1592 dbg_eba("EBA sub-system is initialized");
1593 return 0;
1595 out_free:
1596 for (i = 0; i < num_volumes; i++) {
1597 if (!ubi->volumes[i])
1598 continue;
1599 ubi_eba_replace_table(ubi->volumes[i], NULL);
1601 return err;