Loading...
1/*
2 * Copyright (c) International Business Machines Corp., 2006
3 *
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.
8 *
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.
13 *
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
17 *
18 * Author: Artem Bityutskiy (Битюцкий Артём)
19 */
20
21/*
22 * The UBI Eraseblock Association (EBA) sub-system.
23 *
24 * This sub-system is responsible for I/O to/from logical eraseblock.
25 *
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.
29 *
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.
36 *
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.
42 */
43
44#include <linux/slab.h>
45#include <linux/crc32.h>
46#include <linux/err.h>
47#include "ubi.h"
48
49/* Number of physical eraseblocks reserved for atomic LEB change operation */
50#define EBA_RESERVED_PEBS 1
51
52/**
53 * next_sqnum - get next sequence number.
54 * @ubi: UBI device description object
55 *
56 * This function returns next sequence number to use, which is just the current
57 * global sequence counter value. It also increases the global sequence
58 * counter.
59 */
60static unsigned long long next_sqnum(struct ubi_device *ubi)
61{
62 unsigned long long sqnum;
63
64 spin_lock(&ubi->ltree_lock);
65 sqnum = ubi->global_sqnum++;
66 spin_unlock(&ubi->ltree_lock);
67
68 return sqnum;
69}
70
71/**
72 * ubi_get_compat - get compatibility flags of a volume.
73 * @ubi: UBI device description object
74 * @vol_id: volume ID
75 *
76 * This function returns compatibility flags for an internal volume. User
77 * volumes have no compatibility flags, so %0 is returned.
78 */
79static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
80{
81 if (vol_id == UBI_LAYOUT_VOLUME_ID)
82 return UBI_LAYOUT_VOLUME_COMPAT;
83 return 0;
84}
85
86/**
87 * ltree_lookup - look up the lock tree.
88 * @ubi: UBI device description object
89 * @vol_id: volume ID
90 * @lnum: logical eraseblock number
91 *
92 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
93 * object if the logical eraseblock is locked and %NULL if it is not.
94 * @ubi->ltree_lock has to be locked.
95 */
96static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
97 int lnum)
98{
99 struct rb_node *p;
100
101 p = ubi->ltree.rb_node;
102 while (p) {
103 struct ubi_ltree_entry *le;
104
105 le = rb_entry(p, struct ubi_ltree_entry, rb);
106
107 if (vol_id < le->vol_id)
108 p = p->rb_left;
109 else if (vol_id > le->vol_id)
110 p = p->rb_right;
111 else {
112 if (lnum < le->lnum)
113 p = p->rb_left;
114 else if (lnum > le->lnum)
115 p = p->rb_right;
116 else
117 return le;
118 }
119 }
120
121 return NULL;
122}
123
124/**
125 * ltree_add_entry - add new entry to the lock tree.
126 * @ubi: UBI device description object
127 * @vol_id: volume ID
128 * @lnum: logical eraseblock number
129 *
130 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
131 * lock tree. If such entry is already there, its usage counter is increased.
132 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
133 * failed.
134 */
135static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
136 int vol_id, int lnum)
137{
138 struct ubi_ltree_entry *le, *le1, *le_free;
139
140 le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
141 if (!le)
142 return ERR_PTR(-ENOMEM);
143
144 le->users = 0;
145 init_rwsem(&le->mutex);
146 le->vol_id = vol_id;
147 le->lnum = lnum;
148
149 spin_lock(&ubi->ltree_lock);
150 le1 = ltree_lookup(ubi, vol_id, lnum);
151
152 if (le1) {
153 /*
154 * This logical eraseblock is already locked. The newly
155 * allocated lock entry is not needed.
156 */
157 le_free = le;
158 le = le1;
159 } else {
160 struct rb_node **p, *parent = NULL;
161
162 /*
163 * No lock entry, add the newly allocated one to the
164 * @ubi->ltree RB-tree.
165 */
166 le_free = NULL;
167
168 p = &ubi->ltree.rb_node;
169 while (*p) {
170 parent = *p;
171 le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
172
173 if (vol_id < le1->vol_id)
174 p = &(*p)->rb_left;
175 else if (vol_id > le1->vol_id)
176 p = &(*p)->rb_right;
177 else {
178 ubi_assert(lnum != le1->lnum);
179 if (lnum < le1->lnum)
180 p = &(*p)->rb_left;
181 else
182 p = &(*p)->rb_right;
183 }
184 }
185
186 rb_link_node(&le->rb, parent, p);
187 rb_insert_color(&le->rb, &ubi->ltree);
188 }
189 le->users += 1;
190 spin_unlock(&ubi->ltree_lock);
191
192 kfree(le_free);
193 return le;
194}
195
196/**
197 * leb_read_lock - lock logical eraseblock for reading.
198 * @ubi: UBI device description object
199 * @vol_id: volume ID
200 * @lnum: logical eraseblock number
201 *
202 * This function locks a logical eraseblock for reading. Returns zero in case
203 * of success and a negative error code in case of failure.
204 */
205static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
206{
207 struct ubi_ltree_entry *le;
208
209 le = ltree_add_entry(ubi, vol_id, lnum);
210 if (IS_ERR(le))
211 return PTR_ERR(le);
212 down_read(&le->mutex);
213 return 0;
214}
215
216/**
217 * leb_read_unlock - unlock logical eraseblock.
218 * @ubi: UBI device description object
219 * @vol_id: volume ID
220 * @lnum: logical eraseblock number
221 */
222static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
223{
224 struct ubi_ltree_entry *le;
225
226 spin_lock(&ubi->ltree_lock);
227 le = ltree_lookup(ubi, vol_id, lnum);
228 le->users -= 1;
229 ubi_assert(le->users >= 0);
230 up_read(&le->mutex);
231 if (le->users == 0) {
232 rb_erase(&le->rb, &ubi->ltree);
233 kfree(le);
234 }
235 spin_unlock(&ubi->ltree_lock);
236}
237
238/**
239 * leb_write_lock - lock logical eraseblock for writing.
240 * @ubi: UBI device description object
241 * @vol_id: volume ID
242 * @lnum: logical eraseblock number
243 *
244 * This function locks a logical eraseblock for writing. Returns zero in case
245 * of success and a negative error code in case of failure.
246 */
247static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
248{
249 struct ubi_ltree_entry *le;
250
251 le = ltree_add_entry(ubi, vol_id, lnum);
252 if (IS_ERR(le))
253 return PTR_ERR(le);
254 down_write(&le->mutex);
255 return 0;
256}
257
258/**
259 * leb_write_lock - lock logical eraseblock for writing.
260 * @ubi: UBI device description object
261 * @vol_id: volume ID
262 * @lnum: logical eraseblock number
263 *
264 * This function locks a logical eraseblock for writing if there is no
265 * contention and does nothing if there is contention. Returns %0 in case of
266 * success, %1 in case of contention, and and a negative error code in case of
267 * failure.
268 */
269static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
270{
271 struct ubi_ltree_entry *le;
272
273 le = ltree_add_entry(ubi, vol_id, lnum);
274 if (IS_ERR(le))
275 return PTR_ERR(le);
276 if (down_write_trylock(&le->mutex))
277 return 0;
278
279 /* Contention, cancel */
280 spin_lock(&ubi->ltree_lock);
281 le->users -= 1;
282 ubi_assert(le->users >= 0);
283 if (le->users == 0) {
284 rb_erase(&le->rb, &ubi->ltree);
285 kfree(le);
286 }
287 spin_unlock(&ubi->ltree_lock);
288
289 return 1;
290}
291
292/**
293 * leb_write_unlock - unlock logical eraseblock.
294 * @ubi: UBI device description object
295 * @vol_id: volume ID
296 * @lnum: logical eraseblock number
297 */
298static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
299{
300 struct ubi_ltree_entry *le;
301
302 spin_lock(&ubi->ltree_lock);
303 le = ltree_lookup(ubi, vol_id, lnum);
304 le->users -= 1;
305 ubi_assert(le->users >= 0);
306 up_write(&le->mutex);
307 if (le->users == 0) {
308 rb_erase(&le->rb, &ubi->ltree);
309 kfree(le);
310 }
311 spin_unlock(&ubi->ltree_lock);
312}
313
314/**
315 * ubi_eba_unmap_leb - un-map logical eraseblock.
316 * @ubi: UBI device description object
317 * @vol: volume description object
318 * @lnum: logical eraseblock number
319 *
320 * This function un-maps logical eraseblock @lnum and schedules corresponding
321 * physical eraseblock for erasure. Returns zero in case of success and a
322 * negative error code in case of failure.
323 */
324int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
325 int lnum)
326{
327 int err, pnum, vol_id = vol->vol_id;
328
329 if (ubi->ro_mode)
330 return -EROFS;
331
332 err = leb_write_lock(ubi, vol_id, lnum);
333 if (err)
334 return err;
335
336 pnum = vol->eba_tbl[lnum];
337 if (pnum < 0)
338 /* This logical eraseblock is already unmapped */
339 goto out_unlock;
340
341 dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
342
343 vol->eba_tbl[lnum] = UBI_LEB_UNMAPPED;
344 err = ubi_wl_put_peb(ubi, pnum, 0);
345
346out_unlock:
347 leb_write_unlock(ubi, vol_id, lnum);
348 return err;
349}
350
351/**
352 * ubi_eba_read_leb - read data.
353 * @ubi: UBI device description object
354 * @vol: volume description object
355 * @lnum: logical eraseblock number
356 * @buf: buffer to store the read data
357 * @offset: offset from where to read
358 * @len: how many bytes to read
359 * @check: data CRC check flag
360 *
361 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
362 * bytes. The @check flag only makes sense for static volumes and forces
363 * eraseblock data CRC checking.
364 *
365 * In case of success this function returns zero. In case of a static volume,
366 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
367 * returned for any volume type if an ECC error was detected by the MTD device
368 * driver. Other negative error cored may be returned in case of other errors.
369 */
370int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
371 void *buf, int offset, int len, int check)
372{
373 int err, pnum, scrub = 0, vol_id = vol->vol_id;
374 struct ubi_vid_hdr *vid_hdr;
375 uint32_t uninitialized_var(crc);
376
377 err = leb_read_lock(ubi, vol_id, lnum);
378 if (err)
379 return err;
380
381 pnum = vol->eba_tbl[lnum];
382 if (pnum < 0) {
383 /*
384 * The logical eraseblock is not mapped, fill the whole buffer
385 * with 0xFF bytes. The exception is static volumes for which
386 * it is an error to read unmapped logical eraseblocks.
387 */
388 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
389 len, offset, vol_id, lnum);
390 leb_read_unlock(ubi, vol_id, lnum);
391 ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
392 memset(buf, 0xFF, len);
393 return 0;
394 }
395
396 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
397 len, offset, vol_id, lnum, pnum);
398
399 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
400 check = 0;
401
402retry:
403 if (check) {
404 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
405 if (!vid_hdr) {
406 err = -ENOMEM;
407 goto out_unlock;
408 }
409
410 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
411 if (err && err != UBI_IO_BITFLIPS) {
412 if (err > 0) {
413 /*
414 * The header is either absent or corrupted.
415 * The former case means there is a bug -
416 * switch to read-only mode just in case.
417 * The latter case means a real corruption - we
418 * may try to recover data. FIXME: but this is
419 * not implemented.
420 */
421 if (err == UBI_IO_BAD_HDR_EBADMSG ||
422 err == UBI_IO_BAD_HDR) {
423 ubi_warn("corrupted VID header at PEB "
424 "%d, LEB %d:%d", pnum, vol_id,
425 lnum);
426 err = -EBADMSG;
427 } else
428 ubi_ro_mode(ubi);
429 }
430 goto out_free;
431 } else if (err == UBI_IO_BITFLIPS)
432 scrub = 1;
433
434 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
435 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
436
437 crc = be32_to_cpu(vid_hdr->data_crc);
438 ubi_free_vid_hdr(ubi, vid_hdr);
439 }
440
441 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
442 if (err) {
443 if (err == UBI_IO_BITFLIPS) {
444 scrub = 1;
445 err = 0;
446 } else if (err == -EBADMSG) {
447 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
448 goto out_unlock;
449 scrub = 1;
450 if (!check) {
451 ubi_msg("force data checking");
452 check = 1;
453 goto retry;
454 }
455 } else
456 goto out_unlock;
457 }
458
459 if (check) {
460 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
461 if (crc1 != crc) {
462 ubi_warn("CRC error: calculated %#08x, must be %#08x",
463 crc1, crc);
464 err = -EBADMSG;
465 goto out_unlock;
466 }
467 }
468
469 if (scrub)
470 err = ubi_wl_scrub_peb(ubi, pnum);
471
472 leb_read_unlock(ubi, vol_id, lnum);
473 return err;
474
475out_free:
476 ubi_free_vid_hdr(ubi, vid_hdr);
477out_unlock:
478 leb_read_unlock(ubi, vol_id, lnum);
479 return err;
480}
481
482/**
483 * recover_peb - recover from write failure.
484 * @ubi: UBI device description object
485 * @pnum: the physical eraseblock to recover
486 * @vol_id: volume ID
487 * @lnum: logical eraseblock number
488 * @buf: data which was not written because of the write failure
489 * @offset: offset of the failed write
490 * @len: how many bytes should have been written
491 *
492 * This function is called in case of a write failure and moves all good data
493 * from the potentially bad physical eraseblock to a good physical eraseblock.
494 * This function also writes the data which was not written due to the failure.
495 * Returns new physical eraseblock number in case of success, and a negative
496 * error code in case of failure.
497 */
498static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
499 const void *buf, int offset, int len)
500{
501 int err, idx = vol_id2idx(ubi, vol_id), new_pnum, data_size, tries = 0;
502 struct ubi_volume *vol = ubi->volumes[idx];
503 struct ubi_vid_hdr *vid_hdr;
504
505 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
506 if (!vid_hdr)
507 return -ENOMEM;
508
509retry:
510 new_pnum = ubi_wl_get_peb(ubi, UBI_UNKNOWN);
511 if (new_pnum < 0) {
512 ubi_free_vid_hdr(ubi, vid_hdr);
513 return new_pnum;
514 }
515
516 ubi_msg("recover PEB %d, move data to PEB %d", pnum, new_pnum);
517
518 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
519 if (err && err != UBI_IO_BITFLIPS) {
520 if (err > 0)
521 err = -EIO;
522 goto out_put;
523 }
524
525 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
526 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
527 if (err)
528 goto write_error;
529
530 data_size = offset + len;
531 mutex_lock(&ubi->buf_mutex);
532 memset(ubi->peb_buf1 + offset, 0xFF, len);
533
534 /* Read everything before the area where the write failure happened */
535 if (offset > 0) {
536 err = ubi_io_read_data(ubi, ubi->peb_buf1, pnum, 0, offset);
537 if (err && err != UBI_IO_BITFLIPS)
538 goto out_unlock;
539 }
540
541 memcpy(ubi->peb_buf1 + offset, buf, len);
542
543 err = ubi_io_write_data(ubi, ubi->peb_buf1, new_pnum, 0, data_size);
544 if (err) {
545 mutex_unlock(&ubi->buf_mutex);
546 goto write_error;
547 }
548
549 mutex_unlock(&ubi->buf_mutex);
550 ubi_free_vid_hdr(ubi, vid_hdr);
551
552 vol->eba_tbl[lnum] = new_pnum;
553 ubi_wl_put_peb(ubi, pnum, 1);
554
555 ubi_msg("data was successfully recovered");
556 return 0;
557
558out_unlock:
559 mutex_unlock(&ubi->buf_mutex);
560out_put:
561 ubi_wl_put_peb(ubi, new_pnum, 1);
562 ubi_free_vid_hdr(ubi, vid_hdr);
563 return err;
564
565write_error:
566 /*
567 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
568 * get another one.
569 */
570 ubi_warn("failed to write to PEB %d", new_pnum);
571 ubi_wl_put_peb(ubi, new_pnum, 1);
572 if (++tries > UBI_IO_RETRIES) {
573 ubi_free_vid_hdr(ubi, vid_hdr);
574 return err;
575 }
576 ubi_msg("try again");
577 goto retry;
578}
579
580/**
581 * ubi_eba_write_leb - write data to dynamic volume.
582 * @ubi: UBI device description object
583 * @vol: volume description object
584 * @lnum: logical eraseblock number
585 * @buf: the data to write
586 * @offset: offset within the logical eraseblock where to write
587 * @len: how many bytes to write
588 * @dtype: data type
589 *
590 * This function writes data to logical eraseblock @lnum of a dynamic volume
591 * @vol. Returns zero in case of success and a negative error code in case
592 * of failure. In case of error, it is possible that something was still
593 * written to the flash media, but may be some garbage.
594 */
595int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
596 const void *buf, int offset, int len, int dtype)
597{
598 int err, pnum, tries = 0, vol_id = vol->vol_id;
599 struct ubi_vid_hdr *vid_hdr;
600
601 if (ubi->ro_mode)
602 return -EROFS;
603
604 err = leb_write_lock(ubi, vol_id, lnum);
605 if (err)
606 return err;
607
608 pnum = vol->eba_tbl[lnum];
609 if (pnum >= 0) {
610 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
611 len, offset, vol_id, lnum, pnum);
612
613 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
614 if (err) {
615 ubi_warn("failed to write data to PEB %d", pnum);
616 if (err == -EIO && ubi->bad_allowed)
617 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
618 offset, len);
619 if (err)
620 ubi_ro_mode(ubi);
621 }
622 leb_write_unlock(ubi, vol_id, lnum);
623 return err;
624 }
625
626 /*
627 * The logical eraseblock is not mapped. We have to get a free physical
628 * eraseblock and write the volume identifier header there first.
629 */
630 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
631 if (!vid_hdr) {
632 leb_write_unlock(ubi, vol_id, lnum);
633 return -ENOMEM;
634 }
635
636 vid_hdr->vol_type = UBI_VID_DYNAMIC;
637 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
638 vid_hdr->vol_id = cpu_to_be32(vol_id);
639 vid_hdr->lnum = cpu_to_be32(lnum);
640 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
641 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
642
643retry:
644 pnum = ubi_wl_get_peb(ubi, dtype);
645 if (pnum < 0) {
646 ubi_free_vid_hdr(ubi, vid_hdr);
647 leb_write_unlock(ubi, vol_id, lnum);
648 return pnum;
649 }
650
651 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
652 len, offset, vol_id, lnum, pnum);
653
654 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
655 if (err) {
656 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
657 vol_id, lnum, pnum);
658 goto write_error;
659 }
660
661 if (len) {
662 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
663 if (err) {
664 ubi_warn("failed to write %d bytes at offset %d of "
665 "LEB %d:%d, PEB %d", len, offset, vol_id,
666 lnum, pnum);
667 goto write_error;
668 }
669 }
670
671 vol->eba_tbl[lnum] = pnum;
672
673 leb_write_unlock(ubi, vol_id, lnum);
674 ubi_free_vid_hdr(ubi, vid_hdr);
675 return 0;
676
677write_error:
678 if (err != -EIO || !ubi->bad_allowed) {
679 ubi_ro_mode(ubi);
680 leb_write_unlock(ubi, vol_id, lnum);
681 ubi_free_vid_hdr(ubi, vid_hdr);
682 return err;
683 }
684
685 /*
686 * Fortunately, this is the first write operation to this physical
687 * eraseblock, so just put it and request a new one. We assume that if
688 * this physical eraseblock went bad, the erase code will handle that.
689 */
690 err = ubi_wl_put_peb(ubi, pnum, 1);
691 if (err || ++tries > UBI_IO_RETRIES) {
692 ubi_ro_mode(ubi);
693 leb_write_unlock(ubi, vol_id, lnum);
694 ubi_free_vid_hdr(ubi, vid_hdr);
695 return err;
696 }
697
698 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
699 ubi_msg("try another PEB");
700 goto retry;
701}
702
703/**
704 * ubi_eba_write_leb_st - write data to static volume.
705 * @ubi: UBI device description object
706 * @vol: volume description object
707 * @lnum: logical eraseblock number
708 * @buf: data to write
709 * @len: how many bytes to write
710 * @dtype: data type
711 * @used_ebs: how many logical eraseblocks will this volume contain
712 *
713 * This function writes data to logical eraseblock @lnum of static volume
714 * @vol. The @used_ebs argument should contain total number of logical
715 * eraseblock in this static volume.
716 *
717 * When writing to the last logical eraseblock, the @len argument doesn't have
718 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
719 * to the real data size, although the @buf buffer has to contain the
720 * alignment. In all other cases, @len has to be aligned.
721 *
722 * It is prohibited to write more than once to logical eraseblocks of static
723 * volumes. This function returns zero in case of success and a negative error
724 * code in case of failure.
725 */
726int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
727 int lnum, const void *buf, int len, int dtype,
728 int used_ebs)
729{
730 int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
731 struct ubi_vid_hdr *vid_hdr;
732 uint32_t crc;
733
734 if (ubi->ro_mode)
735 return -EROFS;
736
737 if (lnum == used_ebs - 1)
738 /* If this is the last LEB @len may be unaligned */
739 len = ALIGN(data_size, ubi->min_io_size);
740 else
741 ubi_assert(!(len & (ubi->min_io_size - 1)));
742
743 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
744 if (!vid_hdr)
745 return -ENOMEM;
746
747 err = leb_write_lock(ubi, vol_id, lnum);
748 if (err) {
749 ubi_free_vid_hdr(ubi, vid_hdr);
750 return err;
751 }
752
753 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
754 vid_hdr->vol_id = cpu_to_be32(vol_id);
755 vid_hdr->lnum = cpu_to_be32(lnum);
756 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
757 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
758
759 crc = crc32(UBI_CRC32_INIT, buf, data_size);
760 vid_hdr->vol_type = UBI_VID_STATIC;
761 vid_hdr->data_size = cpu_to_be32(data_size);
762 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
763 vid_hdr->data_crc = cpu_to_be32(crc);
764
765retry:
766 pnum = ubi_wl_get_peb(ubi, dtype);
767 if (pnum < 0) {
768 ubi_free_vid_hdr(ubi, vid_hdr);
769 leb_write_unlock(ubi, vol_id, lnum);
770 return pnum;
771 }
772
773 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
774 len, vol_id, lnum, pnum, used_ebs);
775
776 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
777 if (err) {
778 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
779 vol_id, lnum, pnum);
780 goto write_error;
781 }
782
783 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
784 if (err) {
785 ubi_warn("failed to write %d bytes of data to PEB %d",
786 len, pnum);
787 goto write_error;
788 }
789
790 ubi_assert(vol->eba_tbl[lnum] < 0);
791 vol->eba_tbl[lnum] = pnum;
792
793 leb_write_unlock(ubi, vol_id, lnum);
794 ubi_free_vid_hdr(ubi, vid_hdr);
795 return 0;
796
797write_error:
798 if (err != -EIO || !ubi->bad_allowed) {
799 /*
800 * This flash device does not admit of bad eraseblocks or
801 * something nasty and unexpected happened. Switch to read-only
802 * mode just in case.
803 */
804 ubi_ro_mode(ubi);
805 leb_write_unlock(ubi, vol_id, lnum);
806 ubi_free_vid_hdr(ubi, vid_hdr);
807 return err;
808 }
809
810 err = ubi_wl_put_peb(ubi, pnum, 1);
811 if (err || ++tries > UBI_IO_RETRIES) {
812 ubi_ro_mode(ubi);
813 leb_write_unlock(ubi, vol_id, lnum);
814 ubi_free_vid_hdr(ubi, vid_hdr);
815 return err;
816 }
817
818 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
819 ubi_msg("try another PEB");
820 goto retry;
821}
822
823/*
824 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
825 * @ubi: UBI device description object
826 * @vol: volume description object
827 * @lnum: logical eraseblock number
828 * @buf: data to write
829 * @len: how many bytes to write
830 * @dtype: data type
831 *
832 * This function changes the contents of a logical eraseblock atomically. @buf
833 * has to contain new logical eraseblock data, and @len - the length of the
834 * data, which has to be aligned. This function guarantees that in case of an
835 * unclean reboot the old contents is preserved. Returns zero in case of
836 * success and a negative error code in case of failure.
837 *
838 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
839 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
840 */
841int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
842 int lnum, const void *buf, int len, int dtype)
843{
844 int err, pnum, tries = 0, vol_id = vol->vol_id;
845 struct ubi_vid_hdr *vid_hdr;
846 uint32_t crc;
847
848 if (ubi->ro_mode)
849 return -EROFS;
850
851 if (len == 0) {
852 /*
853 * Special case when data length is zero. In this case the LEB
854 * has to be unmapped and mapped somewhere else.
855 */
856 err = ubi_eba_unmap_leb(ubi, vol, lnum);
857 if (err)
858 return err;
859 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0, dtype);
860 }
861
862 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
863 if (!vid_hdr)
864 return -ENOMEM;
865
866 mutex_lock(&ubi->alc_mutex);
867 err = leb_write_lock(ubi, vol_id, lnum);
868 if (err)
869 goto out_mutex;
870
871 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
872 vid_hdr->vol_id = cpu_to_be32(vol_id);
873 vid_hdr->lnum = cpu_to_be32(lnum);
874 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
875 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
876
877 crc = crc32(UBI_CRC32_INIT, buf, len);
878 vid_hdr->vol_type = UBI_VID_DYNAMIC;
879 vid_hdr->data_size = cpu_to_be32(len);
880 vid_hdr->copy_flag = 1;
881 vid_hdr->data_crc = cpu_to_be32(crc);
882
883retry:
884 pnum = ubi_wl_get_peb(ubi, dtype);
885 if (pnum < 0) {
886 err = pnum;
887 goto out_leb_unlock;
888 }
889
890 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
891 vol_id, lnum, vol->eba_tbl[lnum], pnum);
892
893 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
894 if (err) {
895 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
896 vol_id, lnum, pnum);
897 goto write_error;
898 }
899
900 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
901 if (err) {
902 ubi_warn("failed to write %d bytes of data to PEB %d",
903 len, pnum);
904 goto write_error;
905 }
906
907 if (vol->eba_tbl[lnum] >= 0) {
908 err = ubi_wl_put_peb(ubi, vol->eba_tbl[lnum], 0);
909 if (err)
910 goto out_leb_unlock;
911 }
912
913 vol->eba_tbl[lnum] = pnum;
914
915out_leb_unlock:
916 leb_write_unlock(ubi, vol_id, lnum);
917out_mutex:
918 mutex_unlock(&ubi->alc_mutex);
919 ubi_free_vid_hdr(ubi, vid_hdr);
920 return err;
921
922write_error:
923 if (err != -EIO || !ubi->bad_allowed) {
924 /*
925 * This flash device does not admit of bad eraseblocks or
926 * something nasty and unexpected happened. Switch to read-only
927 * mode just in case.
928 */
929 ubi_ro_mode(ubi);
930 goto out_leb_unlock;
931 }
932
933 err = ubi_wl_put_peb(ubi, pnum, 1);
934 if (err || ++tries > UBI_IO_RETRIES) {
935 ubi_ro_mode(ubi);
936 goto out_leb_unlock;
937 }
938
939 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
940 ubi_msg("try another PEB");
941 goto retry;
942}
943
944/**
945 * is_error_sane - check whether a read error is sane.
946 * @err: code of the error happened during reading
947 *
948 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
949 * cannot read data from the target PEB (an error @err happened). If the error
950 * code is sane, then we treat this error as non-fatal. Otherwise the error is
951 * fatal and UBI will be switched to R/O mode later.
952 *
953 * The idea is that we try not to switch to R/O mode if the read error is
954 * something which suggests there was a real read problem. E.g., %-EIO. Or a
955 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
956 * mode, simply because we do not know what happened at the MTD level, and we
957 * cannot handle this. E.g., the underlying driver may have become crazy, and
958 * it is safer to switch to R/O mode to preserve the data.
959 *
960 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
961 * which we have just written.
962 */
963static int is_error_sane(int err)
964{
965 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
966 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
967 return 0;
968 return 1;
969}
970
971/**
972 * ubi_eba_copy_leb - copy logical eraseblock.
973 * @ubi: UBI device description object
974 * @from: physical eraseblock number from where to copy
975 * @to: physical eraseblock number where to copy
976 * @vid_hdr: VID header of the @from physical eraseblock
977 *
978 * This function copies logical eraseblock from physical eraseblock @from to
979 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
980 * function. Returns:
981 * o %0 in case of success;
982 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_CANCEL_BITFLIPS, etc;
983 * o a negative error code in case of failure.
984 */
985int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
986 struct ubi_vid_hdr *vid_hdr)
987{
988 int err, vol_id, lnum, data_size, aldata_size, idx;
989 struct ubi_volume *vol;
990 uint32_t crc;
991
992 vol_id = be32_to_cpu(vid_hdr->vol_id);
993 lnum = be32_to_cpu(vid_hdr->lnum);
994
995 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
996
997 if (vid_hdr->vol_type == UBI_VID_STATIC) {
998 data_size = be32_to_cpu(vid_hdr->data_size);
999 aldata_size = ALIGN(data_size, ubi->min_io_size);
1000 } else
1001 data_size = aldata_size =
1002 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1003
1004 idx = vol_id2idx(ubi, vol_id);
1005 spin_lock(&ubi->volumes_lock);
1006 /*
1007 * Note, we may race with volume deletion, which means that the volume
1008 * this logical eraseblock belongs to might be being deleted. Since the
1009 * volume deletion un-maps all the volume's logical eraseblocks, it will
1010 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1011 */
1012 vol = ubi->volumes[idx];
1013 spin_unlock(&ubi->volumes_lock);
1014 if (!vol) {
1015 /* No need to do further work, cancel */
1016 dbg_wl("volume %d is being removed, cancel", vol_id);
1017 return MOVE_CANCEL_RACE;
1018 }
1019
1020 /*
1021 * We do not want anybody to write to this logical eraseblock while we
1022 * are moving it, so lock it.
1023 *
1024 * Note, we are using non-waiting locking here, because we cannot sleep
1025 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1026 * unmapping the LEB which is mapped to the PEB we are going to move
1027 * (@from). This task locks the LEB and goes sleep in the
1028 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1029 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1030 * LEB is already locked, we just do not move it and return
1031 * %MOVE_CANCEL_RACE, which means that UBI will re-try, but later.
1032 */
1033 err = leb_write_trylock(ubi, vol_id, lnum);
1034 if (err) {
1035 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1036 return MOVE_CANCEL_RACE;
1037 }
1038
1039 /*
1040 * The LEB might have been put meanwhile, and the task which put it is
1041 * probably waiting on @ubi->move_mutex. No need to continue the work,
1042 * cancel it.
1043 */
1044 if (vol->eba_tbl[lnum] != from) {
1045 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to "
1046 "PEB %d, cancel", vol_id, lnum, from,
1047 vol->eba_tbl[lnum]);
1048 err = MOVE_CANCEL_RACE;
1049 goto out_unlock_leb;
1050 }
1051
1052 /*
1053 * OK, now the LEB is locked and we can safely start moving it. Since
1054 * this function utilizes the @ubi->peb_buf1 buffer which is shared
1055 * with some other functions - we lock the buffer by taking the
1056 * @ubi->buf_mutex.
1057 */
1058 mutex_lock(&ubi->buf_mutex);
1059 dbg_wl("read %d bytes of data", aldata_size);
1060 err = ubi_io_read_data(ubi, ubi->peb_buf1, from, 0, aldata_size);
1061 if (err && err != UBI_IO_BITFLIPS) {
1062 ubi_warn("error %d while reading data from PEB %d",
1063 err, from);
1064 err = MOVE_SOURCE_RD_ERR;
1065 goto out_unlock_buf;
1066 }
1067
1068 /*
1069 * Now we have got to calculate how much data we have to copy. In
1070 * case of a static volume it is fairly easy - the VID header contains
1071 * the data size. In case of a dynamic volume it is more difficult - we
1072 * have to read the contents, cut 0xFF bytes from the end and copy only
1073 * the first part. We must do this to avoid writing 0xFF bytes as it
1074 * may have some side-effects. And not only this. It is important not
1075 * to include those 0xFFs to CRC because later the they may be filled
1076 * by data.
1077 */
1078 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1079 aldata_size = data_size =
1080 ubi_calc_data_len(ubi, ubi->peb_buf1, data_size);
1081
1082 cond_resched();
1083 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf1, data_size);
1084 cond_resched();
1085
1086 /*
1087 * It may turn out to be that the whole @from physical eraseblock
1088 * contains only 0xFF bytes. Then we have to only write the VID header
1089 * and do not write any data. This also means we should not set
1090 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1091 */
1092 if (data_size > 0) {
1093 vid_hdr->copy_flag = 1;
1094 vid_hdr->data_size = cpu_to_be32(data_size);
1095 vid_hdr->data_crc = cpu_to_be32(crc);
1096 }
1097 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
1098
1099 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1100 if (err) {
1101 if (err == -EIO)
1102 err = MOVE_TARGET_WR_ERR;
1103 goto out_unlock_buf;
1104 }
1105
1106 cond_resched();
1107
1108 /* Read the VID header back and check if it was written correctly */
1109 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1110 if (err) {
1111 if (err != UBI_IO_BITFLIPS) {
1112 ubi_warn("error %d while reading VID header back from "
1113 "PEB %d", err, to);
1114 if (is_error_sane(err))
1115 err = MOVE_TARGET_RD_ERR;
1116 } else
1117 err = MOVE_CANCEL_BITFLIPS;
1118 goto out_unlock_buf;
1119 }
1120
1121 if (data_size > 0) {
1122 err = ubi_io_write_data(ubi, ubi->peb_buf1, to, 0, aldata_size);
1123 if (err) {
1124 if (err == -EIO)
1125 err = MOVE_TARGET_WR_ERR;
1126 goto out_unlock_buf;
1127 }
1128
1129 cond_resched();
1130
1131 /*
1132 * We've written the data and are going to read it back to make
1133 * sure it was written correctly.
1134 */
1135
1136 err = ubi_io_read_data(ubi, ubi->peb_buf2, to, 0, aldata_size);
1137 if (err) {
1138 if (err != UBI_IO_BITFLIPS) {
1139 ubi_warn("error %d while reading data back "
1140 "from PEB %d", err, to);
1141 if (is_error_sane(err))
1142 err = MOVE_TARGET_RD_ERR;
1143 } else
1144 err = MOVE_CANCEL_BITFLIPS;
1145 goto out_unlock_buf;
1146 }
1147
1148 cond_resched();
1149
1150 if (memcmp(ubi->peb_buf1, ubi->peb_buf2, aldata_size)) {
1151 ubi_warn("read data back from PEB %d and it is "
1152 "different", to);
1153 err = -EINVAL;
1154 goto out_unlock_buf;
1155 }
1156 }
1157
1158 ubi_assert(vol->eba_tbl[lnum] == from);
1159 vol->eba_tbl[lnum] = to;
1160
1161out_unlock_buf:
1162 mutex_unlock(&ubi->buf_mutex);
1163out_unlock_leb:
1164 leb_write_unlock(ubi, vol_id, lnum);
1165 return err;
1166}
1167
1168/**
1169 * print_rsvd_warning - warn about not having enough reserved PEBs.
1170 * @ubi: UBI device description object
1171 *
1172 * This is a helper function for 'ubi_eba_init_scan()' which is called when UBI
1173 * cannot reserve enough PEBs for bad block handling. This function makes a
1174 * decision whether we have to print a warning or not. The algorithm is as
1175 * follows:
1176 * o if this is a new UBI image, then just print the warning
1177 * o if this is an UBI image which has already been used for some time, print
1178 * a warning only if we can reserve less than 10% of the expected amount of
1179 * the reserved PEB.
1180 *
1181 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1182 * of PEBs becomes smaller, which is normal and we do not want to scare users
1183 * with a warning every time they attach the MTD device. This was an issue
1184 * reported by real users.
1185 */
1186static void print_rsvd_warning(struct ubi_device *ubi,
1187 struct ubi_scan_info *si)
1188{
1189 /*
1190 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1191 * large number to distinguish between newly flashed and used images.
1192 */
1193 if (si->max_sqnum > (1 << 18)) {
1194 int min = ubi->beb_rsvd_level / 10;
1195
1196 if (!min)
1197 min = 1;
1198 if (ubi->beb_rsvd_pebs > min)
1199 return;
1200 }
1201
1202 ubi_warn("cannot reserve enough PEBs for bad PEB handling, reserved %d,"
1203 " need %d", ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1204 if (ubi->corr_peb_count)
1205 ubi_warn("%d PEBs are corrupted and not used",
1206 ubi->corr_peb_count);
1207}
1208
1209/**
1210 * ubi_eba_init_scan - initialize the EBA sub-system using scanning information.
1211 * @ubi: UBI device description object
1212 * @si: scanning information
1213 *
1214 * This function returns zero in case of success and a negative error code in
1215 * case of failure.
1216 */
1217int ubi_eba_init_scan(struct ubi_device *ubi, struct ubi_scan_info *si)
1218{
1219 int i, j, err, num_volumes;
1220 struct ubi_scan_volume *sv;
1221 struct ubi_volume *vol;
1222 struct ubi_scan_leb *seb;
1223 struct rb_node *rb;
1224
1225 dbg_eba("initialize EBA sub-system");
1226
1227 spin_lock_init(&ubi->ltree_lock);
1228 mutex_init(&ubi->alc_mutex);
1229 ubi->ltree = RB_ROOT;
1230
1231 ubi->global_sqnum = si->max_sqnum + 1;
1232 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1233
1234 for (i = 0; i < num_volumes; i++) {
1235 vol = ubi->volumes[i];
1236 if (!vol)
1237 continue;
1238
1239 cond_resched();
1240
1241 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1242 GFP_KERNEL);
1243 if (!vol->eba_tbl) {
1244 err = -ENOMEM;
1245 goto out_free;
1246 }
1247
1248 for (j = 0; j < vol->reserved_pebs; j++)
1249 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1250
1251 sv = ubi_scan_find_sv(si, idx2vol_id(ubi, i));
1252 if (!sv)
1253 continue;
1254
1255 ubi_rb_for_each_entry(rb, seb, &sv->root, u.rb) {
1256 if (seb->lnum >= vol->reserved_pebs)
1257 /*
1258 * This may happen in case of an unclean reboot
1259 * during re-size.
1260 */
1261 ubi_scan_move_to_list(sv, seb, &si->erase);
1262 vol->eba_tbl[seb->lnum] = seb->pnum;
1263 }
1264 }
1265
1266 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1267 ubi_err("no enough physical eraseblocks (%d, need %d)",
1268 ubi->avail_pebs, EBA_RESERVED_PEBS);
1269 if (ubi->corr_peb_count)
1270 ubi_err("%d PEBs are corrupted and not used",
1271 ubi->corr_peb_count);
1272 err = -ENOSPC;
1273 goto out_free;
1274 }
1275 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1276 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1277
1278 if (ubi->bad_allowed) {
1279 ubi_calculate_reserved(ubi);
1280
1281 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1282 /* No enough free physical eraseblocks */
1283 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1284 print_rsvd_warning(ubi, si);
1285 } else
1286 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1287
1288 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1289 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1290 }
1291
1292 dbg_eba("EBA sub-system is initialized");
1293 return 0;
1294
1295out_free:
1296 for (i = 0; i < num_volumes; i++) {
1297 if (!ubi->volumes[i])
1298 continue;
1299 kfree(ubi->volumes[i]->eba_tbl);
1300 ubi->volumes[i]->eba_tbl = NULL;
1301 }
1302 return err;
1303}
1/*
2 * Copyright (c) International Business Machines Corp., 2006
3 *
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.
8 *
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.
13 *
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
17 *
18 * Author: Artem Bityutskiy (Битюцкий Артём)
19 */
20
21/*
22 * The UBI Eraseblock Association (EBA) sub-system.
23 *
24 * This sub-system is responsible for I/O to/from logical eraseblock.
25 *
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.
29 *
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.
36 *
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.
42 */
43
44#include <linux/slab.h>
45#include <linux/crc32.h>
46#include <linux/err.h>
47#include "ubi.h"
48
49/* Number of physical eraseblocks reserved for atomic LEB change operation */
50#define EBA_RESERVED_PEBS 1
51
52/**
53 * struct ubi_eba_entry - structure encoding a single LEB -> PEB association
54 * @pnum: the physical eraseblock number attached to the LEB
55 *
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.
59 */
60struct ubi_eba_entry {
61 int pnum;
62};
63
64/**
65 * struct ubi_eba_table - LEB -> PEB association information
66 * @entries: the LEB to PEB mapping (one entry per LEB).
67 *
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.
71 */
72struct ubi_eba_table {
73 struct ubi_eba_entry *entries;
74};
75
76/**
77 * next_sqnum - get next sequence number.
78 * @ubi: UBI device description object
79 *
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.
83 */
84unsigned long long ubi_next_sqnum(struct ubi_device *ubi)
85{
86 unsigned long long sqnum;
87
88 spin_lock(&ubi->ltree_lock);
89 sqnum = ubi->global_sqnum++;
90 spin_unlock(&ubi->ltree_lock);
91
92 return sqnum;
93}
94
95/**
96 * ubi_get_compat - get compatibility flags of a volume.
97 * @ubi: UBI device description object
98 * @vol_id: volume ID
99 *
100 * This function returns compatibility flags for an internal volume. User
101 * volumes have no compatibility flags, so %0 is returned.
102 */
103static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
104{
105 if (vol_id == UBI_LAYOUT_VOLUME_ID)
106 return UBI_LAYOUT_VOLUME_COMPAT;
107 return 0;
108}
109
110/**
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
115 *
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.
119 */
120void ubi_eba_get_ldesc(struct ubi_volume *vol, int lnum,
121 struct ubi_eba_leb_desc *ldesc)
122{
123 ldesc->lnum = lnum;
124 ldesc->pnum = vol->eba_tbl->entries[lnum].pnum;
125}
126
127/**
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
132 *
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.
135 */
136struct ubi_eba_table *ubi_eba_create_table(struct ubi_volume *vol,
137 int nentries)
138{
139 struct ubi_eba_table *tbl;
140 int err = -ENOMEM;
141 int i;
142
143 tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
144 if (!tbl)
145 return ERR_PTR(-ENOMEM);
146
147 tbl->entries = kmalloc_array(nentries, sizeof(*tbl->entries),
148 GFP_KERNEL);
149 if (!tbl->entries)
150 goto err;
151
152 for (i = 0; i < nentries; i++)
153 tbl->entries[i].pnum = UBI_LEB_UNMAPPED;
154
155 return tbl;
156
157err:
158 kfree(tbl->entries);
159 kfree(tbl);
160
161 return ERR_PTR(err);
162}
163
164/**
165 * ubi_eba_destroy_table - destroy an EBA table
166 * @tbl: the table to destroy
167 *
168 * Destroy an EBA table.
169 */
170void ubi_eba_destroy_table(struct ubi_eba_table *tbl)
171{
172 if (!tbl)
173 return;
174
175 kfree(tbl->entries);
176 kfree(tbl);
177}
178
179/**
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
184 *
185 * Copy the EBA table stored in vol into the one pointed by dst.
186 */
187void ubi_eba_copy_table(struct ubi_volume *vol, struct ubi_eba_table *dst,
188 int nentries)
189{
190 struct ubi_eba_table *src;
191 int i;
192
193 ubi_assert(dst && vol && vol->eba_tbl);
194
195 src = vol->eba_tbl;
196
197 for (i = 0; i < nentries; i++)
198 dst->entries[i].pnum = src->entries[i].pnum;
199}
200
201/**
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
205 *
206 * Assign a new EBA table to the volume and release the old one.
207 */
208void ubi_eba_replace_table(struct ubi_volume *vol, struct ubi_eba_table *tbl)
209{
210 ubi_eba_destroy_table(vol->eba_tbl);
211 vol->eba_tbl = tbl;
212}
213
214/**
215 * ltree_lookup - look up the lock tree.
216 * @ubi: UBI device description object
217 * @vol_id: volume ID
218 * @lnum: logical eraseblock number
219 *
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.
223 */
224static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
225 int lnum)
226{
227 struct rb_node *p;
228
229 p = ubi->ltree.rb_node;
230 while (p) {
231 struct ubi_ltree_entry *le;
232
233 le = rb_entry(p, struct ubi_ltree_entry, rb);
234
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;
246 }
247 }
248
249 return NULL;
250}
251
252/**
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
257 *
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.
262 */
263static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
264 int vol_id, int lnum)
265{
266 struct ubi_ltree_entry *le, *le1, *le_free;
267
268 le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
269 if (!le)
270 return ERR_PTR(-ENOMEM);
271
272 le->users = 0;
273 init_rwsem(&le->mutex);
274 le->vol_id = vol_id;
275 le->lnum = lnum;
276
277 spin_lock(&ubi->ltree_lock);
278 le1 = ltree_lookup(ubi, vol_id, lnum);
279
280 if (le1) {
281 /*
282 * This logical eraseblock is already locked. The newly
283 * allocated lock entry is not needed.
284 */
285 le_free = le;
286 le = le1;
287 } else {
288 struct rb_node **p, *parent = NULL;
289
290 /*
291 * No lock entry, add the newly allocated one to the
292 * @ubi->ltree RB-tree.
293 */
294 le_free = NULL;
295
296 p = &ubi->ltree.rb_node;
297 while (*p) {
298 parent = *p;
299 le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
300
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;
311 }
312 }
313
314 rb_link_node(&le->rb, parent, p);
315 rb_insert_color(&le->rb, &ubi->ltree);
316 }
317 le->users += 1;
318 spin_unlock(&ubi->ltree_lock);
319
320 kfree(le_free);
321 return le;
322}
323
324/**
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
329 *
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.
332 */
333static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
334{
335 struct ubi_ltree_entry *le;
336
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;
342}
343
344/**
345 * leb_read_unlock - unlock logical eraseblock.
346 * @ubi: UBI device description object
347 * @vol_id: volume ID
348 * @lnum: logical eraseblock number
349 */
350static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
351{
352 struct ubi_ltree_entry *le;
353
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);
362 }
363 spin_unlock(&ubi->ltree_lock);
364}
365
366/**
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
371 *
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.
374 */
375static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
376{
377 struct ubi_ltree_entry *le;
378
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;
384}
385
386/**
387 * leb_write_trylock - try to lock logical eraseblock for writing.
388 * @ubi: UBI device description object
389 * @vol_id: volume ID
390 * @lnum: logical eraseblock number
391 *
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.
396 */
397static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
398{
399 struct ubi_ltree_entry *le;
400
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;
406
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);
414 }
415 spin_unlock(&ubi->ltree_lock);
416
417 return 1;
418}
419
420/**
421 * leb_write_unlock - unlock logical eraseblock.
422 * @ubi: UBI device description object
423 * @vol_id: volume ID
424 * @lnum: logical eraseblock number
425 */
426static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
427{
428 struct ubi_ltree_entry *le;
429
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);
438 }
439 spin_unlock(&ubi->ltree_lock);
440}
441
442/**
443 * ubi_eba_is_mapped - check if a LEB is mapped.
444 * @vol: volume description object
445 * @lnum: logical eraseblock number
446 *
447 * This function returns true if the LEB is mapped, false otherwise.
448 */
449bool ubi_eba_is_mapped(struct ubi_volume *vol, int lnum)
450{
451 return vol->eba_tbl->entries[lnum].pnum >= 0;
452}
453
454/**
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
459 *
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.
463 */
464int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
465 int lnum)
466{
467 int err, pnum, vol_id = vol->vol_id;
468
469 if (ubi->ro_mode)
470 return -EROFS;
471
472 err = leb_write_lock(ubi, vol_id, lnum);
473 if (err)
474 return err;
475
476 pnum = vol->eba_tbl->entries[lnum].pnum;
477 if (pnum < 0)
478 /* This logical eraseblock is already unmapped */
479 goto out_unlock;
480
481 dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
482
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);
487
488out_unlock:
489 leb_write_unlock(ubi, vol_id, lnum);
490 return err;
491}
492
493/**
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
502 *
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.
506 *
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.
511 */
512int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
513 void *buf, int offset, int len, int check)
514{
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);
519
520 err = leb_read_lock(ubi, vol_id, lnum);
521 if (err)
522 return err;
523
524 pnum = vol->eba_tbl->entries[lnum].pnum;
525 if (pnum < 0) {
526 /*
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.
530 */
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;
537 }
538
539 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
540 len, offset, vol_id, lnum, pnum);
541
542 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
543 check = 0;
544
545retry:
546 if (check) {
547 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
548 if (!vidb) {
549 err = -ENOMEM;
550 goto out_unlock;
551 }
552
553 vid_hdr = ubi_get_vid_hdr(vidb);
554
555 err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
556 if (err && err != UBI_IO_BITFLIPS) {
557 if (err > 0) {
558 /*
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.
565 */
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 {
572 /*
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.
584 */
585 if (ubi->fast_attach) {
586 err = -EBADMSG;
587 } else {
588 err = -EINVAL;
589 ubi_ro_mode(ubi);
590 }
591 }
592 }
593 goto out_free;
594 } else if (err == UBI_IO_BITFLIPS)
595 scrub = 1;
596
597 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
598 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
599
600 crc = be32_to_cpu(vid_hdr->data_crc);
601 ubi_free_vid_buf(vidb);
602 }
603
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;
616 }
617 } else
618 goto out_unlock;
619 }
620
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;
628 }
629 }
630
631 if (scrub)
632 err = ubi_wl_scrub_peb(ubi, pnum);
633
634 leb_read_unlock(ubi, vol_id, lnum);
635 return err;
636
637out_free:
638 ubi_free_vid_buf(vidb);
639out_unlock:
640 leb_read_unlock(ubi, vol_id, lnum);
641 return err;
642}
643
644/**
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
653 *
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.
657 */
658int 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)
661{
662 int to_read;
663 int ret;
664 struct scatterlist *sg;
665
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;
673
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;
679
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;
687 }
688
689 break;
690 }
691
692 sgl->list_pos++;
693 sgl->page_pos = 0;
694 }
695
696 return ret;
697}
698
699/**
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
709 *
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.
716 */
717static 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)
720{
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;
725
726 *retry = false;
727
728 new_pnum = ubi_wl_get_peb(ubi);
729 if (new_pnum < 0) {
730 err = new_pnum;
731 goto out_put;
732 }
733
734 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
735 pnum, new_pnum);
736
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;
742 }
743
744 vid_hdr = ubi_get_vid_hdr(vidb);
745 ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
746
747 mutex_lock(&ubi->buf_mutex);
748 memset(ubi->peb_buf + offset, 0xFF, len);
749
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;
755 }
756
757 *retry = true;
758
759 memcpy(ubi->peb_buf + offset, buf, len);
760
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;
770
771 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
772
773out_unlock:
774 mutex_unlock(&ubi->buf_mutex);
775
776 if (!err)
777 vol->eba_tbl->entries[lnum].pnum = new_pnum;
778
779out_put:
780 up_read(&ubi->fm_eba_sem);
781
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) {
786 /*
787 * Bad luck? This physical eraseblock is bad too? Crud. Let's
788 * try to get another one.
789 */
790 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
791 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
792 }
793
794 return err;
795}
796
797/**
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
806 *
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.
812 */
813static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
814 const void *buf, int offset, int len)
815{
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;
819
820 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
821 if (!vidb)
822 return -ENOMEM;
823
824 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
825 bool retry;
826
827 err = try_recover_peb(vol, pnum, lnum, buf, offset, len, vidb,
828 &retry);
829 if (!err || !retry)
830 break;
831
832 ubi_msg(ubi, "try again");
833 }
834
835 ubi_free_vid_buf(vidb);
836
837 return err;
838}
839
840/**
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
848 *
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.
854 */
855static 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)
858{
859 struct ubi_device *ubi = vol->ubi;
860 int pnum, opnum, err, vol_id = vol->vol_id;
861
862 pnum = ubi_wl_get_peb(ubi);
863 if (pnum < 0) {
864 err = pnum;
865 goto out_put;
866 }
867
868 opnum = vol->eba_tbl->entries[lnum].pnum;
869
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);
872
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;
878 }
879
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;
887 }
888 }
889
890 vol->eba_tbl->entries[lnum].pnum = pnum;
891
892out_put:
893 up_read(&ubi->fm_eba_sem);
894
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);
899
900 return err;
901}
902
903/**
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
911 *
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.
917 */
918int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
919 const void *buf, int offset, int len)
920{
921 int err, pnum, tries, vol_id = vol->vol_id;
922 struct ubi_vid_io_buf *vidb;
923 struct ubi_vid_hdr *vid_hdr;
924
925 if (ubi->ro_mode)
926 return -EROFS;
927
928 err = leb_write_lock(ubi, vol_id, lnum);
929 if (err)
930 return err;
931
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);
936
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);
943 }
944
945 goto out;
946 }
947
948 /*
949 * The logical eraseblock is not mapped. We have to get a free physical
950 * eraseblock and write the volume identifier header there first.
951 */
952 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
953 if (!vidb) {
954 leb_write_unlock(ubi, vol_id, lnum);
955 return -ENOMEM;
956 }
957
958 vid_hdr = ubi_get_vid_hdr(vidb);
959
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);
966
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;
971
972 /*
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.
977 */
978 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
979 ubi_msg(ubi, "try another PEB");
980 }
981
982 ubi_free_vid_buf(vidb);
983
984out:
985 if (err)
986 ubi_ro_mode(ubi);
987
988 leb_write_unlock(ubi, vol_id, lnum);
989
990 return err;
991}
992
993/**
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
1001 *
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.
1005 *
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.
1010 *
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.
1014 */
1015int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
1016 int lnum, const void *buf, int len, int used_ebs)
1017{
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;
1022
1023 if (ubi->ro_mode)
1024 return -EROFS;
1025
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)));
1031
1032 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1033 if (!vidb)
1034 return -ENOMEM;
1035
1036 vid_hdr = ubi_get_vid_hdr(vidb);
1037
1038 err = leb_write_lock(ubi, vol_id, lnum);
1039 if (err)
1040 goto out;
1041
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);
1047
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);
1053
1054 ubi_assert(vol->eba_tbl->entries[lnum].pnum < 0);
1055
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;
1060
1061 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1062 ubi_msg(ubi, "try another PEB");
1063 }
1064
1065 if (err)
1066 ubi_ro_mode(ubi);
1067
1068 leb_write_unlock(ubi, vol_id, lnum);
1069
1070out:
1071 ubi_free_vid_buf(vidb);
1072
1073 return err;
1074}
1075
1076/*
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
1083 *
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.
1089 *
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.
1092 */
1093int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
1094 int lnum, const void *buf, int len)
1095{
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;
1100
1101 if (ubi->ro_mode)
1102 return -EROFS;
1103
1104 if (len == 0) {
1105 /*
1106 * Special case when data length is zero. In this case the LEB
1107 * has to be unmapped and mapped somewhere else.
1108 */
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);
1113 }
1114
1115 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1116 if (!vidb)
1117 return -ENOMEM;
1118
1119 vid_hdr = ubi_get_vid_hdr(vidb);
1120
1121 mutex_lock(&ubi->alc_mutex);
1122 err = leb_write_lock(ubi, vol_id, lnum);
1123 if (err)
1124 goto out_mutex;
1125
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);
1131
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);
1137
1138 dbg_eba("change LEB %d:%d", vol_id, lnum);
1139
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;
1144
1145 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1146 ubi_msg(ubi, "try another PEB");
1147 }
1148
1149 /*
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.
1153 */
1154 if (err)
1155 ubi_ro_mode(ubi);
1156
1157 leb_write_unlock(ubi, vol_id, lnum);
1158
1159out_mutex:
1160 mutex_unlock(&ubi->alc_mutex);
1161 ubi_free_vid_buf(vidb);
1162 return err;
1163}
1164
1165/**
1166 * is_error_sane - check whether a read error is sane.
1167 * @err: code of the error happened during reading
1168 *
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.
1173 *
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.
1180 *
1181 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1182 * which we have just written.
1183 */
1184static int is_error_sane(int err)
1185{
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;
1190}
1191
1192/**
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
1198 *
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.
1205 */
1206int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1207 struct ubi_vid_io_buf *vidb)
1208{
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;
1213
1214 ubi_assert(rwsem_is_locked(&ubi->fm_eba_sem));
1215
1216 vol_id = be32_to_cpu(vid_hdr->vol_id);
1217 lnum = be32_to_cpu(vid_hdr->lnum);
1218
1219 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1220
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);
1227
1228 idx = vol_id2idx(ubi, vol_id);
1229 spin_lock(&ubi->volumes_lock);
1230 /*
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.
1235 */
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;
1242 }
1243
1244 /*
1245 * We do not want anybody to write to this logical eraseblock while we
1246 * are moving it, so lock it.
1247 *
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.
1258 */
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;
1263 }
1264
1265 /*
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.
1269 */
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;
1275 }
1276
1277 /*
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.
1282 */
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;
1291 }
1292
1293 /*
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.
1302 */
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);
1306
1307 cond_resched();
1308 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1309 cond_resched();
1310
1311 /*
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.
1316 */
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);
1321 }
1322 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1323
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;
1329 }
1330
1331 cond_resched();
1332
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;
1344 }
1345
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;
1352 }
1353
1354 cond_resched();
1355 }
1356
1357 ubi_assert(vol->eba_tbl->entries[lnum].pnum == from);
1358 vol->eba_tbl->entries[lnum].pnum = to;
1359
1360out_unlock_buf:
1361 mutex_unlock(&ubi->buf_mutex);
1362out_unlock_leb:
1363 leb_write_unlock(ubi, vol_id, lnum);
1364 return err;
1365}
1366
1367/**
1368 * print_rsvd_warning - warn about not having enough reserved PEBs.
1369 * @ubi: UBI device description object
1370 *
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.
1379 *
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.
1384 */
1385static void print_rsvd_warning(struct ubi_device *ubi,
1386 struct ubi_attach_info *ai)
1387{
1388 /*
1389 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1390 * large number to distinguish between newly flashed and used images.
1391 */
1392 if (ai->max_sqnum > (1 << 18)) {
1393 int min = ubi->beb_rsvd_level / 10;
1394
1395 if (!min)
1396 min = 1;
1397 if (ubi->beb_rsvd_pebs > min)
1398 return;
1399 }
1400
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);
1406}
1407
1408/**
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
1413 *
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.
1417 */
1418int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1419 struct ubi_attach_info *ai_scan)
1420{
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;
1427
1428 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1429
1430 scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1431 if (!scan_eba)
1432 return -ENOMEM;
1433
1434 fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1435 if (!fm_eba) {
1436 kfree(scan_eba);
1437 return -ENOMEM;
1438 }
1439
1440 for (i = 0; i < num_volumes; i++) {
1441 vol = ubi->volumes[i];
1442 if (!vol)
1443 continue;
1444
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;
1450 }
1451
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;
1457 }
1458
1459 for (j = 0; j < vol->reserved_pebs; j++)
1460 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1461
1462 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1463 if (!av)
1464 continue;
1465
1466 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1467 scan_eba[i][aeb->lnum] = aeb->pnum;
1468
1469 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1470 if (!av)
1471 continue;
1472
1473 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1474 fm_eba[i][aeb->lnum] = aeb->pnum;
1475
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;
1481
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);
1486 }
1487 }
1488 }
1489
1490out_free:
1491 for (i = 0; i < num_volumes; i++) {
1492 if (!ubi->volumes[i])
1493 continue;
1494
1495 kfree(scan_eba[i]);
1496 kfree(fm_eba[i]);
1497 }
1498
1499 kfree(scan_eba);
1500 kfree(fm_eba);
1501 return ret;
1502}
1503
1504/**
1505 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1506 * @ubi: UBI device description object
1507 * @ai: attaching information
1508 *
1509 * This function returns zero in case of success and a negative error code in
1510 * case of failure.
1511 */
1512int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1513{
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;
1519
1520 dbg_eba("initialize EBA sub-system");
1521
1522 spin_lock_init(&ubi->ltree_lock);
1523 mutex_init(&ubi->alc_mutex);
1524 ubi->ltree = RB_ROOT;
1525
1526 ubi->global_sqnum = ai->max_sqnum + 1;
1527 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1528
1529 for (i = 0; i < num_volumes; i++) {
1530 struct ubi_eba_table *tbl;
1531
1532 vol = ubi->volumes[i];
1533 if (!vol)
1534 continue;
1535
1536 cond_resched();
1537
1538 tbl = ubi_eba_create_table(vol, vol->reserved_pebs);
1539 if (IS_ERR(tbl)) {
1540 err = PTR_ERR(tbl);
1541 goto out_free;
1542 }
1543
1544 ubi_eba_replace_table(vol, tbl);
1545
1546 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1547 if (!av)
1548 continue;
1549
1550 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1551 if (aeb->lnum >= vol->reserved_pebs) {
1552 /*
1553 * This may happen in case of an unclean reboot
1554 * during re-size.
1555 */
1556 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1557 } else {
1558 struct ubi_eba_entry *entry;
1559
1560 entry = &vol->eba_tbl->entries[aeb->lnum];
1561 entry->pnum = aeb->pnum;
1562 }
1563 }
1564 }
1565
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;
1574 }
1575 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1576 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1577
1578 if (ubi->bad_allowed) {
1579 ubi_calculate_reserved(ubi);
1580
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;
1587
1588 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1589 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1590 }
1591
1592 dbg_eba("EBA sub-system is initialized");
1593 return 0;
1594
1595out_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);
1600 }
1601 return err;
1602}