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// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (c) International Business Machines Corp., 2006
4 *
5 * Author: Artem Bityutskiy (Битюцкий Артём)
6 */
7
8/*
9 * The UBI Eraseblock Association (EBA) sub-system.
10 *
11 * This sub-system is responsible for I/O to/from logical eraseblock.
12 *
13 * Although in this implementation the EBA table is fully kept and managed in
14 * RAM, which assumes poor scalability, it might be (partially) maintained on
15 * flash in future implementations.
16 *
17 * The EBA sub-system implements per-logical eraseblock locking. Before
18 * accessing a logical eraseblock it is locked for reading or writing. The
19 * per-logical eraseblock locking is implemented by means of the lock tree. The
20 * lock tree is an RB-tree which refers all the currently locked logical
21 * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
22 * They are indexed by (@vol_id, @lnum) pairs.
23 *
24 * EBA also maintains the global sequence counter which is incremented each
25 * time a logical eraseblock is mapped to a physical eraseblock and it is
26 * stored in the volume identifier header. This means that each VID header has
27 * a unique sequence number. The sequence number is only increased an we assume
28 * 64 bits is enough to never overflow.
29 */
30
31#include <linux/slab.h>
32#include <linux/crc32.h>
33#include <linux/err.h>
34#include "ubi.h"
35
36/**
37 * struct ubi_eba_entry - structure encoding a single LEB -> PEB association
38 * @pnum: the physical eraseblock number attached to the LEB
39 *
40 * This structure is encoding a LEB -> PEB association. Note that the LEB
41 * number is not stored here, because it is the index used to access the
42 * entries table.
43 */
44struct ubi_eba_entry {
45 int pnum;
46};
47
48/**
49 * struct ubi_eba_table - LEB -> PEB association information
50 * @entries: the LEB to PEB mapping (one entry per LEB).
51 *
52 * This structure is private to the EBA logic and should be kept here.
53 * It is encoding the LEB to PEB association table, and is subject to
54 * changes.
55 */
56struct ubi_eba_table {
57 struct ubi_eba_entry *entries;
58};
59
60/**
61 * ubi_next_sqnum - get next sequence number.
62 * @ubi: UBI device description object
63 *
64 * This function returns next sequence number to use, which is just the current
65 * global sequence counter value. It also increases the global sequence
66 * counter.
67 */
68unsigned long long ubi_next_sqnum(struct ubi_device *ubi)
69{
70 unsigned long long sqnum;
71
72 spin_lock(&ubi->ltree_lock);
73 sqnum = ubi->global_sqnum++;
74 spin_unlock(&ubi->ltree_lock);
75
76 return sqnum;
77}
78
79/**
80 * ubi_get_compat - get compatibility flags of a volume.
81 * @ubi: UBI device description object
82 * @vol_id: volume ID
83 *
84 * This function returns compatibility flags for an internal volume. User
85 * volumes have no compatibility flags, so %0 is returned.
86 */
87static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
88{
89 if (vol_id == UBI_LAYOUT_VOLUME_ID)
90 return UBI_LAYOUT_VOLUME_COMPAT;
91 return 0;
92}
93
94/**
95 * ubi_eba_get_ldesc - get information about a LEB
96 * @vol: volume description object
97 * @lnum: logical eraseblock number
98 * @ldesc: the LEB descriptor to fill
99 *
100 * Used to query information about a specific LEB.
101 * It is currently only returning the physical position of the LEB, but will be
102 * extended to provide more information.
103 */
104void ubi_eba_get_ldesc(struct ubi_volume *vol, int lnum,
105 struct ubi_eba_leb_desc *ldesc)
106{
107 ldesc->lnum = lnum;
108 ldesc->pnum = vol->eba_tbl->entries[lnum].pnum;
109}
110
111/**
112 * ubi_eba_create_table - allocate a new EBA table and initialize it with all
113 * LEBs unmapped
114 * @vol: volume containing the EBA table to copy
115 * @nentries: number of entries in the table
116 *
117 * Allocate a new EBA table and initialize it with all LEBs unmapped.
118 * Returns a valid pointer if it succeed, an ERR_PTR() otherwise.
119 */
120struct ubi_eba_table *ubi_eba_create_table(struct ubi_volume *vol,
121 int nentries)
122{
123 struct ubi_eba_table *tbl;
124 int err = -ENOMEM;
125 int i;
126
127 tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
128 if (!tbl)
129 return ERR_PTR(-ENOMEM);
130
131 tbl->entries = kmalloc_array(nentries, sizeof(*tbl->entries),
132 GFP_KERNEL);
133 if (!tbl->entries)
134 goto err;
135
136 for (i = 0; i < nentries; i++)
137 tbl->entries[i].pnum = UBI_LEB_UNMAPPED;
138
139 return tbl;
140
141err:
142 kfree(tbl);
143
144 return ERR_PTR(err);
145}
146
147/**
148 * ubi_eba_destroy_table - destroy an EBA table
149 * @tbl: the table to destroy
150 *
151 * Destroy an EBA table.
152 */
153void ubi_eba_destroy_table(struct ubi_eba_table *tbl)
154{
155 if (!tbl)
156 return;
157
158 kfree(tbl->entries);
159 kfree(tbl);
160}
161
162/**
163 * ubi_eba_copy_table - copy the EBA table attached to vol into another table
164 * @vol: volume containing the EBA table to copy
165 * @dst: destination
166 * @nentries: number of entries to copy
167 *
168 * Copy the EBA table stored in vol into the one pointed by dst.
169 */
170void ubi_eba_copy_table(struct ubi_volume *vol, struct ubi_eba_table *dst,
171 int nentries)
172{
173 struct ubi_eba_table *src;
174 int i;
175
176 ubi_assert(dst && vol && vol->eba_tbl);
177
178 src = vol->eba_tbl;
179
180 for (i = 0; i < nentries; i++)
181 dst->entries[i].pnum = src->entries[i].pnum;
182}
183
184/**
185 * ubi_eba_replace_table - assign a new EBA table to a volume
186 * @vol: volume containing the EBA table to copy
187 * @tbl: new EBA table
188 *
189 * Assign a new EBA table to the volume and release the old one.
190 */
191void ubi_eba_replace_table(struct ubi_volume *vol, struct ubi_eba_table *tbl)
192{
193 ubi_eba_destroy_table(vol->eba_tbl);
194 vol->eba_tbl = tbl;
195}
196
197/**
198 * ltree_lookup - look up the lock tree.
199 * @ubi: UBI device description object
200 * @vol_id: volume ID
201 * @lnum: logical eraseblock number
202 *
203 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
204 * object if the logical eraseblock is locked and %NULL if it is not.
205 * @ubi->ltree_lock has to be locked.
206 */
207static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
208 int lnum)
209{
210 struct rb_node *p;
211
212 p = ubi->ltree.rb_node;
213 while (p) {
214 struct ubi_ltree_entry *le;
215
216 le = rb_entry(p, struct ubi_ltree_entry, rb);
217
218 if (vol_id < le->vol_id)
219 p = p->rb_left;
220 else if (vol_id > le->vol_id)
221 p = p->rb_right;
222 else {
223 if (lnum < le->lnum)
224 p = p->rb_left;
225 else if (lnum > le->lnum)
226 p = p->rb_right;
227 else
228 return le;
229 }
230 }
231
232 return NULL;
233}
234
235/**
236 * ltree_add_entry - add new entry to the lock tree.
237 * @ubi: UBI device description object
238 * @vol_id: volume ID
239 * @lnum: logical eraseblock number
240 *
241 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
242 * lock tree. If such entry is already there, its usage counter is increased.
243 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
244 * failed.
245 */
246static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
247 int vol_id, int lnum)
248{
249 struct ubi_ltree_entry *le, *le1, *le_free;
250
251 le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
252 if (!le)
253 return ERR_PTR(-ENOMEM);
254
255 le->users = 0;
256 init_rwsem(&le->mutex);
257 le->vol_id = vol_id;
258 le->lnum = lnum;
259
260 spin_lock(&ubi->ltree_lock);
261 le1 = ltree_lookup(ubi, vol_id, lnum);
262
263 if (le1) {
264 /*
265 * This logical eraseblock is already locked. The newly
266 * allocated lock entry is not needed.
267 */
268 le_free = le;
269 le = le1;
270 } else {
271 struct rb_node **p, *parent = NULL;
272
273 /*
274 * No lock entry, add the newly allocated one to the
275 * @ubi->ltree RB-tree.
276 */
277 le_free = NULL;
278
279 p = &ubi->ltree.rb_node;
280 while (*p) {
281 parent = *p;
282 le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
283
284 if (vol_id < le1->vol_id)
285 p = &(*p)->rb_left;
286 else if (vol_id > le1->vol_id)
287 p = &(*p)->rb_right;
288 else {
289 ubi_assert(lnum != le1->lnum);
290 if (lnum < le1->lnum)
291 p = &(*p)->rb_left;
292 else
293 p = &(*p)->rb_right;
294 }
295 }
296
297 rb_link_node(&le->rb, parent, p);
298 rb_insert_color(&le->rb, &ubi->ltree);
299 }
300 le->users += 1;
301 spin_unlock(&ubi->ltree_lock);
302
303 kfree(le_free);
304 return le;
305}
306
307/**
308 * leb_read_lock - lock logical eraseblock for reading.
309 * @ubi: UBI device description object
310 * @vol_id: volume ID
311 * @lnum: logical eraseblock number
312 *
313 * This function locks a logical eraseblock for reading. Returns zero in case
314 * of success and a negative error code in case of failure.
315 */
316static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
317{
318 struct ubi_ltree_entry *le;
319
320 le = ltree_add_entry(ubi, vol_id, lnum);
321 if (IS_ERR(le))
322 return PTR_ERR(le);
323 down_read(&le->mutex);
324 return 0;
325}
326
327/**
328 * leb_read_unlock - unlock logical eraseblock.
329 * @ubi: UBI device description object
330 * @vol_id: volume ID
331 * @lnum: logical eraseblock number
332 */
333static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
334{
335 struct ubi_ltree_entry *le;
336
337 spin_lock(&ubi->ltree_lock);
338 le = ltree_lookup(ubi, vol_id, lnum);
339 le->users -= 1;
340 ubi_assert(le->users >= 0);
341 up_read(&le->mutex);
342 if (le->users == 0) {
343 rb_erase(&le->rb, &ubi->ltree);
344 kfree(le);
345 }
346 spin_unlock(&ubi->ltree_lock);
347}
348
349/**
350 * leb_write_lock - lock logical eraseblock for writing.
351 * @ubi: UBI device description object
352 * @vol_id: volume ID
353 * @lnum: logical eraseblock number
354 *
355 * This function locks a logical eraseblock for writing. Returns zero in case
356 * of success and a negative error code in case of failure.
357 */
358static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
359{
360 struct ubi_ltree_entry *le;
361
362 le = ltree_add_entry(ubi, vol_id, lnum);
363 if (IS_ERR(le))
364 return PTR_ERR(le);
365 down_write(&le->mutex);
366 return 0;
367}
368
369/**
370 * leb_write_trylock - try to lock logical eraseblock for writing.
371 * @ubi: UBI device description object
372 * @vol_id: volume ID
373 * @lnum: logical eraseblock number
374 *
375 * This function locks a logical eraseblock for writing if there is no
376 * contention and does nothing if there is contention. Returns %0 in case of
377 * success, %1 in case of contention, and a negative error code in case of
378 * failure.
379 */
380static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
381{
382 struct ubi_ltree_entry *le;
383
384 le = ltree_add_entry(ubi, vol_id, lnum);
385 if (IS_ERR(le))
386 return PTR_ERR(le);
387 if (down_write_trylock(&le->mutex))
388 return 0;
389
390 /* Contention, cancel */
391 spin_lock(&ubi->ltree_lock);
392 le->users -= 1;
393 ubi_assert(le->users >= 0);
394 if (le->users == 0) {
395 rb_erase(&le->rb, &ubi->ltree);
396 kfree(le);
397 }
398 spin_unlock(&ubi->ltree_lock);
399
400 return 1;
401}
402
403/**
404 * leb_write_unlock - unlock logical eraseblock.
405 * @ubi: UBI device description object
406 * @vol_id: volume ID
407 * @lnum: logical eraseblock number
408 */
409static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
410{
411 struct ubi_ltree_entry *le;
412
413 spin_lock(&ubi->ltree_lock);
414 le = ltree_lookup(ubi, vol_id, lnum);
415 le->users -= 1;
416 ubi_assert(le->users >= 0);
417 up_write(&le->mutex);
418 if (le->users == 0) {
419 rb_erase(&le->rb, &ubi->ltree);
420 kfree(le);
421 }
422 spin_unlock(&ubi->ltree_lock);
423}
424
425/**
426 * ubi_eba_is_mapped - check if a LEB is mapped.
427 * @vol: volume description object
428 * @lnum: logical eraseblock number
429 *
430 * This function returns true if the LEB is mapped, false otherwise.
431 */
432bool ubi_eba_is_mapped(struct ubi_volume *vol, int lnum)
433{
434 return vol->eba_tbl->entries[lnum].pnum >= 0;
435}
436
437/**
438 * ubi_eba_unmap_leb - un-map logical eraseblock.
439 * @ubi: UBI device description object
440 * @vol: volume description object
441 * @lnum: logical eraseblock number
442 *
443 * This function un-maps logical eraseblock @lnum and schedules corresponding
444 * physical eraseblock for erasure. Returns zero in case of success and a
445 * negative error code in case of failure.
446 */
447int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
448 int lnum)
449{
450 int err, pnum, vol_id = vol->vol_id;
451
452 if (ubi->ro_mode)
453 return -EROFS;
454
455 err = leb_write_lock(ubi, vol_id, lnum);
456 if (err)
457 return err;
458
459 pnum = vol->eba_tbl->entries[lnum].pnum;
460 if (pnum < 0)
461 /* This logical eraseblock is already unmapped */
462 goto out_unlock;
463
464 dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
465
466 down_read(&ubi->fm_eba_sem);
467 vol->eba_tbl->entries[lnum].pnum = UBI_LEB_UNMAPPED;
468 up_read(&ubi->fm_eba_sem);
469 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
470
471out_unlock:
472 leb_write_unlock(ubi, vol_id, lnum);
473 return err;
474}
475
476#ifdef CONFIG_MTD_UBI_FASTMAP
477/**
478 * check_mapping - check and fixup a mapping
479 * @ubi: UBI device description object
480 * @vol: volume description object
481 * @lnum: logical eraseblock number
482 * @pnum: physical eraseblock number
483 *
484 * Checks whether a given mapping is valid. Fastmap cannot track LEB unmap
485 * operations, if such an operation is interrupted the mapping still looks
486 * good, but upon first read an ECC is reported to the upper layer.
487 * Normaly during the full-scan at attach time this is fixed, for Fastmap
488 * we have to deal with it while reading.
489 * If the PEB behind a LEB shows this symthom we change the mapping to
490 * %UBI_LEB_UNMAPPED and schedule the PEB for erasure.
491 *
492 * Returns 0 on success, negative error code in case of failure.
493 */
494static int check_mapping(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
495 int *pnum)
496{
497 int err;
498 struct ubi_vid_io_buf *vidb;
499 struct ubi_vid_hdr *vid_hdr;
500
501 if (!ubi->fast_attach)
502 return 0;
503
504 if (!vol->checkmap || test_bit(lnum, vol->checkmap))
505 return 0;
506
507 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
508 if (!vidb)
509 return -ENOMEM;
510
511 err = ubi_io_read_vid_hdr(ubi, *pnum, vidb, 0);
512 if (err > 0 && err != UBI_IO_BITFLIPS) {
513 int torture = 0;
514
515 switch (err) {
516 case UBI_IO_FF:
517 case UBI_IO_FF_BITFLIPS:
518 case UBI_IO_BAD_HDR:
519 case UBI_IO_BAD_HDR_EBADMSG:
520 break;
521 default:
522 ubi_assert(0);
523 }
524
525 if (err == UBI_IO_BAD_HDR_EBADMSG || err == UBI_IO_FF_BITFLIPS)
526 torture = 1;
527
528 down_read(&ubi->fm_eba_sem);
529 vol->eba_tbl->entries[lnum].pnum = UBI_LEB_UNMAPPED;
530 up_read(&ubi->fm_eba_sem);
531 ubi_wl_put_peb(ubi, vol->vol_id, lnum, *pnum, torture);
532
533 *pnum = UBI_LEB_UNMAPPED;
534 } else if (err < 0) {
535 ubi_err(ubi, "unable to read VID header back from PEB %i: %i",
536 *pnum, err);
537
538 goto out_free;
539 } else {
540 int found_vol_id, found_lnum;
541
542 ubi_assert(err == 0 || err == UBI_IO_BITFLIPS);
543
544 vid_hdr = ubi_get_vid_hdr(vidb);
545 found_vol_id = be32_to_cpu(vid_hdr->vol_id);
546 found_lnum = be32_to_cpu(vid_hdr->lnum);
547
548 if (found_lnum != lnum || found_vol_id != vol->vol_id) {
549 ubi_err(ubi, "EBA mismatch! PEB %i is LEB %i:%i instead of LEB %i:%i",
550 *pnum, found_vol_id, found_lnum, vol->vol_id, lnum);
551 ubi_ro_mode(ubi);
552 err = -EINVAL;
553 goto out_free;
554 }
555 }
556
557 set_bit(lnum, vol->checkmap);
558 err = 0;
559
560out_free:
561 ubi_free_vid_buf(vidb);
562
563 return err;
564}
565#else
566static int check_mapping(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
567 int *pnum)
568{
569 return 0;
570}
571#endif
572
573/**
574 * ubi_eba_read_leb - read data.
575 * @ubi: UBI device description object
576 * @vol: volume description object
577 * @lnum: logical eraseblock number
578 * @buf: buffer to store the read data
579 * @offset: offset from where to read
580 * @len: how many bytes to read
581 * @check: data CRC check flag
582 *
583 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
584 * bytes. The @check flag only makes sense for static volumes and forces
585 * eraseblock data CRC checking.
586 *
587 * In case of success this function returns zero. In case of a static volume,
588 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
589 * returned for any volume type if an ECC error was detected by the MTD device
590 * driver. Other negative error cored may be returned in case of other errors.
591 */
592int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
593 void *buf, int offset, int len, int check)
594{
595 int err, pnum, scrub = 0, vol_id = vol->vol_id;
596 struct ubi_vid_io_buf *vidb;
597 struct ubi_vid_hdr *vid_hdr;
598 uint32_t crc;
599
600 err = leb_read_lock(ubi, vol_id, lnum);
601 if (err)
602 return err;
603
604 pnum = vol->eba_tbl->entries[lnum].pnum;
605 if (pnum >= 0) {
606 err = check_mapping(ubi, vol, lnum, &pnum);
607 if (err < 0)
608 goto out_unlock;
609 }
610
611 if (pnum == UBI_LEB_UNMAPPED) {
612 /*
613 * The logical eraseblock is not mapped, fill the whole buffer
614 * with 0xFF bytes. The exception is static volumes for which
615 * it is an error to read unmapped logical eraseblocks.
616 */
617 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
618 len, offset, vol_id, lnum);
619 leb_read_unlock(ubi, vol_id, lnum);
620 ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
621 memset(buf, 0xFF, len);
622 return 0;
623 }
624
625 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
626 len, offset, vol_id, lnum, pnum);
627
628 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
629 check = 0;
630
631retry:
632 if (check) {
633 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
634 if (!vidb) {
635 err = -ENOMEM;
636 goto out_unlock;
637 }
638
639 vid_hdr = ubi_get_vid_hdr(vidb);
640
641 err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
642 if (err && err != UBI_IO_BITFLIPS) {
643 if (err > 0) {
644 /*
645 * The header is either absent or corrupted.
646 * The former case means there is a bug -
647 * switch to read-only mode just in case.
648 * The latter case means a real corruption - we
649 * may try to recover data. FIXME: but this is
650 * not implemented.
651 */
652 if (err == UBI_IO_BAD_HDR_EBADMSG ||
653 err == UBI_IO_BAD_HDR) {
654 ubi_warn(ubi, "corrupted VID header at PEB %d, LEB %d:%d",
655 pnum, vol_id, lnum);
656 err = -EBADMSG;
657 } else {
658 /*
659 * Ending up here in the non-Fastmap case
660 * is a clear bug as the VID header had to
661 * be present at scan time to have it referenced.
662 * With fastmap the story is more complicated.
663 * Fastmap has the mapping info without the need
664 * of a full scan. So the LEB could have been
665 * unmapped, Fastmap cannot know this and keeps
666 * the LEB referenced.
667 * This is valid and works as the layer above UBI
668 * has to do bookkeeping about used/referenced
669 * LEBs in any case.
670 */
671 if (ubi->fast_attach) {
672 err = -EBADMSG;
673 } else {
674 err = -EINVAL;
675 ubi_ro_mode(ubi);
676 }
677 }
678 }
679 goto out_free;
680 } else if (err == UBI_IO_BITFLIPS)
681 scrub = 1;
682
683 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
684 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
685
686 crc = be32_to_cpu(vid_hdr->data_crc);
687 ubi_free_vid_buf(vidb);
688 }
689
690 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
691 if (err) {
692 if (err == UBI_IO_BITFLIPS)
693 scrub = 1;
694 else if (mtd_is_eccerr(err)) {
695 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
696 goto out_unlock;
697 scrub = 1;
698 if (!check) {
699 ubi_msg(ubi, "force data checking");
700 check = 1;
701 goto retry;
702 }
703 } else
704 goto out_unlock;
705 }
706
707 if (check) {
708 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
709 if (crc1 != crc) {
710 ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
711 crc1, crc);
712 err = -EBADMSG;
713 goto out_unlock;
714 }
715 }
716
717 if (scrub)
718 err = ubi_wl_scrub_peb(ubi, pnum);
719
720 leb_read_unlock(ubi, vol_id, lnum);
721 return err;
722
723out_free:
724 ubi_free_vid_buf(vidb);
725out_unlock:
726 leb_read_unlock(ubi, vol_id, lnum);
727 return err;
728}
729
730/**
731 * ubi_eba_read_leb_sg - read data into a scatter gather list.
732 * @ubi: UBI device description object
733 * @vol: volume description object
734 * @lnum: logical eraseblock number
735 * @sgl: UBI scatter gather list to store the read data
736 * @offset: offset from where to read
737 * @len: how many bytes to read
738 * @check: data CRC check flag
739 *
740 * This function works exactly like ubi_eba_read_leb(). But instead of
741 * storing the read data into a buffer it writes to an UBI scatter gather
742 * list.
743 */
744int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
745 struct ubi_sgl *sgl, int lnum, int offset, int len,
746 int check)
747{
748 int to_read;
749 int ret;
750 struct scatterlist *sg;
751
752 for (;;) {
753 ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
754 sg = &sgl->sg[sgl->list_pos];
755 if (len < sg->length - sgl->page_pos)
756 to_read = len;
757 else
758 to_read = sg->length - sgl->page_pos;
759
760 ret = ubi_eba_read_leb(ubi, vol, lnum,
761 sg_virt(sg) + sgl->page_pos, offset,
762 to_read, check);
763 if (ret < 0)
764 return ret;
765
766 offset += to_read;
767 len -= to_read;
768 if (!len) {
769 sgl->page_pos += to_read;
770 if (sgl->page_pos == sg->length) {
771 sgl->list_pos++;
772 sgl->page_pos = 0;
773 }
774
775 break;
776 }
777
778 sgl->list_pos++;
779 sgl->page_pos = 0;
780 }
781
782 return ret;
783}
784
785/**
786 * try_recover_peb - try to recover from write failure.
787 * @vol: volume description object
788 * @pnum: the physical eraseblock to recover
789 * @lnum: logical eraseblock number
790 * @buf: data which was not written because of the write failure
791 * @offset: offset of the failed write
792 * @len: how many bytes should have been written
793 * @vidb: VID buffer
794 * @retry: whether the caller should retry in case of failure
795 *
796 * This function is called in case of a write failure and moves all good data
797 * from the potentially bad physical eraseblock to a good physical eraseblock.
798 * This function also writes the data which was not written due to the failure.
799 * Returns 0 in case of success, and a negative error code in case of failure.
800 * In case of failure, the %retry parameter is set to false if this is a fatal
801 * error (retrying won't help), and true otherwise.
802 */
803static int try_recover_peb(struct ubi_volume *vol, int pnum, int lnum,
804 const void *buf, int offset, int len,
805 struct ubi_vid_io_buf *vidb, bool *retry)
806{
807 struct ubi_device *ubi = vol->ubi;
808 struct ubi_vid_hdr *vid_hdr;
809 int new_pnum, err, vol_id = vol->vol_id, data_size;
810 uint32_t crc;
811
812 *retry = false;
813
814 new_pnum = ubi_wl_get_peb(ubi);
815 if (new_pnum < 0) {
816 err = new_pnum;
817 goto out_put;
818 }
819
820 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
821 pnum, new_pnum);
822
823 err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
824 if (err && err != UBI_IO_BITFLIPS) {
825 if (err > 0)
826 err = -EIO;
827 goto out_put;
828 }
829
830 vid_hdr = ubi_get_vid_hdr(vidb);
831 ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
832
833 mutex_lock(&ubi->buf_mutex);
834 memset(ubi->peb_buf + offset, 0xFF, len);
835
836 /* Read everything before the area where the write failure happened */
837 if (offset > 0) {
838 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
839 if (err && err != UBI_IO_BITFLIPS)
840 goto out_unlock;
841 }
842
843 *retry = true;
844
845 memcpy(ubi->peb_buf + offset, buf, len);
846
847 data_size = offset + len;
848 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
849 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
850 vid_hdr->copy_flag = 1;
851 vid_hdr->data_size = cpu_to_be32(data_size);
852 vid_hdr->data_crc = cpu_to_be32(crc);
853 err = ubi_io_write_vid_hdr(ubi, new_pnum, vidb);
854 if (err)
855 goto out_unlock;
856
857 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
858
859out_unlock:
860 mutex_unlock(&ubi->buf_mutex);
861
862 if (!err)
863 vol->eba_tbl->entries[lnum].pnum = new_pnum;
864
865out_put:
866 up_read(&ubi->fm_eba_sem);
867
868 if (!err) {
869 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
870 ubi_msg(ubi, "data was successfully recovered");
871 } else if (new_pnum >= 0) {
872 /*
873 * Bad luck? This physical eraseblock is bad too? Crud. Let's
874 * try to get another one.
875 */
876 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
877 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
878 }
879
880 return err;
881}
882
883/**
884 * recover_peb - recover from write failure.
885 * @ubi: UBI device description object
886 * @pnum: the physical eraseblock to recover
887 * @vol_id: volume ID
888 * @lnum: logical eraseblock number
889 * @buf: data which was not written because of the write failure
890 * @offset: offset of the failed write
891 * @len: how many bytes should have been written
892 *
893 * This function is called in case of a write failure and moves all good data
894 * from the potentially bad physical eraseblock to a good physical eraseblock.
895 * This function also writes the data which was not written due to the failure.
896 * Returns 0 in case of success, and a negative error code in case of failure.
897 * This function tries %UBI_IO_RETRIES before giving up.
898 */
899static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
900 const void *buf, int offset, int len)
901{
902 int err, idx = vol_id2idx(ubi, vol_id), tries;
903 struct ubi_volume *vol = ubi->volumes[idx];
904 struct ubi_vid_io_buf *vidb;
905
906 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
907 if (!vidb)
908 return -ENOMEM;
909
910 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
911 bool retry;
912
913 err = try_recover_peb(vol, pnum, lnum, buf, offset, len, vidb,
914 &retry);
915 if (!err || !retry)
916 break;
917
918 ubi_msg(ubi, "try again");
919 }
920
921 ubi_free_vid_buf(vidb);
922
923 return err;
924}
925
926/**
927 * try_write_vid_and_data - try to write VID header and data to a new PEB.
928 * @vol: volume description object
929 * @lnum: logical eraseblock number
930 * @vidb: the VID buffer to write
931 * @buf: buffer containing the data
932 * @offset: where to start writing data
933 * @len: how many bytes should be written
934 *
935 * This function tries to write VID header and data belonging to logical
936 * eraseblock @lnum of volume @vol to a new physical eraseblock. Returns zero
937 * in case of success and a negative error code in case of failure.
938 * In case of error, it is possible that something was still written to the
939 * flash media, but may be some garbage.
940 */
941static int try_write_vid_and_data(struct ubi_volume *vol, int lnum,
942 struct ubi_vid_io_buf *vidb, const void *buf,
943 int offset, int len)
944{
945 struct ubi_device *ubi = vol->ubi;
946 int pnum, opnum, err, err2, vol_id = vol->vol_id;
947
948 pnum = ubi_wl_get_peb(ubi);
949 if (pnum < 0) {
950 err = pnum;
951 goto out_put;
952 }
953
954 opnum = vol->eba_tbl->entries[lnum].pnum;
955
956 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
957 len, offset, vol_id, lnum, pnum);
958
959 err = ubi_io_write_vid_hdr(ubi, pnum, vidb);
960 if (err) {
961 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
962 vol_id, lnum, pnum);
963 goto out_put;
964 }
965
966 if (len) {
967 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
968 if (err) {
969 ubi_warn(ubi,
970 "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
971 len, offset, vol_id, lnum, pnum);
972 goto out_put;
973 }
974 }
975
976 vol->eba_tbl->entries[lnum].pnum = pnum;
977
978out_put:
979 up_read(&ubi->fm_eba_sem);
980
981 if (err && pnum >= 0) {
982 err2 = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
983 if (err2) {
984 ubi_warn(ubi, "failed to return physical eraseblock %d, error %d",
985 pnum, err2);
986 }
987 } else if (!err && opnum >= 0) {
988 err2 = ubi_wl_put_peb(ubi, vol_id, lnum, opnum, 0);
989 if (err2) {
990 ubi_warn(ubi, "failed to return physical eraseblock %d, error %d",
991 opnum, err2);
992 }
993 }
994
995 return err;
996}
997
998/**
999 * ubi_eba_write_leb - write data to dynamic volume.
1000 * @ubi: UBI device description object
1001 * @vol: volume description object
1002 * @lnum: logical eraseblock number
1003 * @buf: the data to write
1004 * @offset: offset within the logical eraseblock where to write
1005 * @len: how many bytes to write
1006 *
1007 * This function writes data to logical eraseblock @lnum of a dynamic volume
1008 * @vol. Returns zero in case of success and a negative error code in case
1009 * of failure. In case of error, it is possible that something was still
1010 * written to the flash media, but may be some garbage.
1011 * This function retries %UBI_IO_RETRIES times before giving up.
1012 */
1013int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
1014 const void *buf, int offset, int len)
1015{
1016 int err, pnum, tries, vol_id = vol->vol_id;
1017 struct ubi_vid_io_buf *vidb;
1018 struct ubi_vid_hdr *vid_hdr;
1019
1020 if (ubi->ro_mode)
1021 return -EROFS;
1022
1023 err = leb_write_lock(ubi, vol_id, lnum);
1024 if (err)
1025 return err;
1026
1027 pnum = vol->eba_tbl->entries[lnum].pnum;
1028 if (pnum >= 0) {
1029 err = check_mapping(ubi, vol, lnum, &pnum);
1030 if (err < 0)
1031 goto out;
1032 }
1033
1034 if (pnum >= 0) {
1035 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
1036 len, offset, vol_id, lnum, pnum);
1037
1038 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
1039 if (err) {
1040 ubi_warn(ubi, "failed to write data to PEB %d", pnum);
1041 if (err == -EIO && ubi->bad_allowed)
1042 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
1043 offset, len);
1044 }
1045
1046 goto out;
1047 }
1048
1049 /*
1050 * The logical eraseblock is not mapped. We have to get a free physical
1051 * eraseblock and write the volume identifier header there first.
1052 */
1053 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1054 if (!vidb) {
1055 leb_write_unlock(ubi, vol_id, lnum);
1056 return -ENOMEM;
1057 }
1058
1059 vid_hdr = ubi_get_vid_hdr(vidb);
1060
1061 vid_hdr->vol_type = UBI_VID_DYNAMIC;
1062 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1063 vid_hdr->vol_id = cpu_to_be32(vol_id);
1064 vid_hdr->lnum = cpu_to_be32(lnum);
1065 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1066 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1067
1068 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1069 err = try_write_vid_and_data(vol, lnum, vidb, buf, offset, len);
1070 if (err != -EIO || !ubi->bad_allowed)
1071 break;
1072
1073 /*
1074 * Fortunately, this is the first write operation to this
1075 * physical eraseblock, so just put it and request a new one.
1076 * We assume that if this physical eraseblock went bad, the
1077 * erase code will handle that.
1078 */
1079 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1080 ubi_msg(ubi, "try another PEB");
1081 }
1082
1083 ubi_free_vid_buf(vidb);
1084
1085out:
1086 if (err)
1087 ubi_ro_mode(ubi);
1088
1089 leb_write_unlock(ubi, vol_id, lnum);
1090
1091 return err;
1092}
1093
1094/**
1095 * ubi_eba_write_leb_st - write data to static volume.
1096 * @ubi: UBI device description object
1097 * @vol: volume description object
1098 * @lnum: logical eraseblock number
1099 * @buf: data to write
1100 * @len: how many bytes to write
1101 * @used_ebs: how many logical eraseblocks will this volume contain
1102 *
1103 * This function writes data to logical eraseblock @lnum of static volume
1104 * @vol. The @used_ebs argument should contain total number of logical
1105 * eraseblock in this static volume.
1106 *
1107 * When writing to the last logical eraseblock, the @len argument doesn't have
1108 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
1109 * to the real data size, although the @buf buffer has to contain the
1110 * alignment. In all other cases, @len has to be aligned.
1111 *
1112 * It is prohibited to write more than once to logical eraseblocks of static
1113 * volumes. This function returns zero in case of success and a negative error
1114 * code in case of failure.
1115 */
1116int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
1117 int lnum, const void *buf, int len, int used_ebs)
1118{
1119 int err, tries, data_size = len, vol_id = vol->vol_id;
1120 struct ubi_vid_io_buf *vidb;
1121 struct ubi_vid_hdr *vid_hdr;
1122 uint32_t crc;
1123
1124 if (ubi->ro_mode)
1125 return -EROFS;
1126
1127 if (lnum == used_ebs - 1)
1128 /* If this is the last LEB @len may be unaligned */
1129 len = ALIGN(data_size, ubi->min_io_size);
1130 else
1131 ubi_assert(!(len & (ubi->min_io_size - 1)));
1132
1133 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1134 if (!vidb)
1135 return -ENOMEM;
1136
1137 vid_hdr = ubi_get_vid_hdr(vidb);
1138
1139 err = leb_write_lock(ubi, vol_id, lnum);
1140 if (err)
1141 goto out;
1142
1143 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1144 vid_hdr->vol_id = cpu_to_be32(vol_id);
1145 vid_hdr->lnum = cpu_to_be32(lnum);
1146 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1147 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1148
1149 crc = crc32(UBI_CRC32_INIT, buf, data_size);
1150 vid_hdr->vol_type = UBI_VID_STATIC;
1151 vid_hdr->data_size = cpu_to_be32(data_size);
1152 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
1153 vid_hdr->data_crc = cpu_to_be32(crc);
1154
1155 ubi_assert(vol->eba_tbl->entries[lnum].pnum < 0);
1156
1157 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1158 err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1159 if (err != -EIO || !ubi->bad_allowed)
1160 break;
1161
1162 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1163 ubi_msg(ubi, "try another PEB");
1164 }
1165
1166 if (err)
1167 ubi_ro_mode(ubi);
1168
1169 leb_write_unlock(ubi, vol_id, lnum);
1170
1171out:
1172 ubi_free_vid_buf(vidb);
1173
1174 return err;
1175}
1176
1177/*
1178 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
1179 * @ubi: UBI device description object
1180 * @vol: volume description object
1181 * @lnum: logical eraseblock number
1182 * @buf: data to write
1183 * @len: how many bytes to write
1184 *
1185 * This function changes the contents of a logical eraseblock atomically. @buf
1186 * has to contain new logical eraseblock data, and @len - the length of the
1187 * data, which has to be aligned. This function guarantees that in case of an
1188 * unclean reboot the old contents is preserved. Returns zero in case of
1189 * success and a negative error code in case of failure.
1190 *
1191 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
1192 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
1193 */
1194int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
1195 int lnum, const void *buf, int len)
1196{
1197 int err, tries, vol_id = vol->vol_id;
1198 struct ubi_vid_io_buf *vidb;
1199 struct ubi_vid_hdr *vid_hdr;
1200 uint32_t crc;
1201
1202 if (ubi->ro_mode)
1203 return -EROFS;
1204
1205 if (len == 0) {
1206 /*
1207 * Special case when data length is zero. In this case the LEB
1208 * has to be unmapped and mapped somewhere else.
1209 */
1210 err = ubi_eba_unmap_leb(ubi, vol, lnum);
1211 if (err)
1212 return err;
1213 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
1214 }
1215
1216 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1217 if (!vidb)
1218 return -ENOMEM;
1219
1220 vid_hdr = ubi_get_vid_hdr(vidb);
1221
1222 mutex_lock(&ubi->alc_mutex);
1223 err = leb_write_lock(ubi, vol_id, lnum);
1224 if (err)
1225 goto out_mutex;
1226
1227 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1228 vid_hdr->vol_id = cpu_to_be32(vol_id);
1229 vid_hdr->lnum = cpu_to_be32(lnum);
1230 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1231 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1232
1233 crc = crc32(UBI_CRC32_INIT, buf, len);
1234 vid_hdr->vol_type = UBI_VID_DYNAMIC;
1235 vid_hdr->data_size = cpu_to_be32(len);
1236 vid_hdr->copy_flag = 1;
1237 vid_hdr->data_crc = cpu_to_be32(crc);
1238
1239 dbg_eba("change LEB %d:%d", vol_id, lnum);
1240
1241 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1242 err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1243 if (err != -EIO || !ubi->bad_allowed)
1244 break;
1245
1246 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1247 ubi_msg(ubi, "try another PEB");
1248 }
1249
1250 /*
1251 * This flash device does not admit of bad eraseblocks or
1252 * something nasty and unexpected happened. Switch to read-only
1253 * mode just in case.
1254 */
1255 if (err)
1256 ubi_ro_mode(ubi);
1257
1258 leb_write_unlock(ubi, vol_id, lnum);
1259
1260out_mutex:
1261 mutex_unlock(&ubi->alc_mutex);
1262 ubi_free_vid_buf(vidb);
1263 return err;
1264}
1265
1266/**
1267 * is_error_sane - check whether a read error is sane.
1268 * @err: code of the error happened during reading
1269 *
1270 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1271 * cannot read data from the target PEB (an error @err happened). If the error
1272 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1273 * fatal and UBI will be switched to R/O mode later.
1274 *
1275 * The idea is that we try not to switch to R/O mode if the read error is
1276 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1277 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1278 * mode, simply because we do not know what happened at the MTD level, and we
1279 * cannot handle this. E.g., the underlying driver may have become crazy, and
1280 * it is safer to switch to R/O mode to preserve the data.
1281 *
1282 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1283 * which we have just written.
1284 */
1285static int is_error_sane(int err)
1286{
1287 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1288 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1289 return 0;
1290 return 1;
1291}
1292
1293/**
1294 * ubi_eba_copy_leb - copy logical eraseblock.
1295 * @ubi: UBI device description object
1296 * @from: physical eraseblock number from where to copy
1297 * @to: physical eraseblock number where to copy
1298 * @vidb: data structure from where the VID header is derived
1299 *
1300 * This function copies logical eraseblock from physical eraseblock @from to
1301 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1302 * function. Returns:
1303 * o %0 in case of success;
1304 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1305 * o a negative error code in case of failure.
1306 */
1307int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1308 struct ubi_vid_io_buf *vidb)
1309{
1310 int err, vol_id, lnum, data_size, aldata_size, idx;
1311 struct ubi_vid_hdr *vid_hdr = ubi_get_vid_hdr(vidb);
1312 struct ubi_volume *vol;
1313 uint32_t crc;
1314
1315 ubi_assert(rwsem_is_locked(&ubi->fm_eba_sem));
1316
1317 vol_id = be32_to_cpu(vid_hdr->vol_id);
1318 lnum = be32_to_cpu(vid_hdr->lnum);
1319
1320 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1321
1322 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1323 data_size = be32_to_cpu(vid_hdr->data_size);
1324 aldata_size = ALIGN(data_size, ubi->min_io_size);
1325 } else
1326 data_size = aldata_size =
1327 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1328
1329 idx = vol_id2idx(ubi, vol_id);
1330 spin_lock(&ubi->volumes_lock);
1331 /*
1332 * Note, we may race with volume deletion, which means that the volume
1333 * this logical eraseblock belongs to might be being deleted. Since the
1334 * volume deletion un-maps all the volume's logical eraseblocks, it will
1335 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1336 */
1337 vol = ubi->volumes[idx];
1338 spin_unlock(&ubi->volumes_lock);
1339 if (!vol) {
1340 /* No need to do further work, cancel */
1341 dbg_wl("volume %d is being removed, cancel", vol_id);
1342 return MOVE_CANCEL_RACE;
1343 }
1344
1345 /*
1346 * We do not want anybody to write to this logical eraseblock while we
1347 * are moving it, so lock it.
1348 *
1349 * Note, we are using non-waiting locking here, because we cannot sleep
1350 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1351 * unmapping the LEB which is mapped to the PEB we are going to move
1352 * (@from). This task locks the LEB and goes sleep in the
1353 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1354 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1355 * LEB is already locked, we just do not move it and return
1356 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1357 * we do not know the reasons of the contention - it may be just a
1358 * normal I/O on this LEB, so we want to re-try.
1359 */
1360 err = leb_write_trylock(ubi, vol_id, lnum);
1361 if (err) {
1362 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1363 return MOVE_RETRY;
1364 }
1365
1366 /*
1367 * The LEB might have been put meanwhile, and the task which put it is
1368 * probably waiting on @ubi->move_mutex. No need to continue the work,
1369 * cancel it.
1370 */
1371 if (vol->eba_tbl->entries[lnum].pnum != from) {
1372 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1373 vol_id, lnum, from, vol->eba_tbl->entries[lnum].pnum);
1374 err = MOVE_CANCEL_RACE;
1375 goto out_unlock_leb;
1376 }
1377
1378 /*
1379 * OK, now the LEB is locked and we can safely start moving it. Since
1380 * this function utilizes the @ubi->peb_buf buffer which is shared
1381 * with some other functions - we lock the buffer by taking the
1382 * @ubi->buf_mutex.
1383 */
1384 mutex_lock(&ubi->buf_mutex);
1385 dbg_wl("read %d bytes of data", aldata_size);
1386 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1387 if (err && err != UBI_IO_BITFLIPS) {
1388 ubi_warn(ubi, "error %d while reading data from PEB %d",
1389 err, from);
1390 err = MOVE_SOURCE_RD_ERR;
1391 goto out_unlock_buf;
1392 }
1393
1394 /*
1395 * Now we have got to calculate how much data we have to copy. In
1396 * case of a static volume it is fairly easy - the VID header contains
1397 * the data size. In case of a dynamic volume it is more difficult - we
1398 * have to read the contents, cut 0xFF bytes from the end and copy only
1399 * the first part. We must do this to avoid writing 0xFF bytes as it
1400 * may have some side-effects. And not only this. It is important not
1401 * to include those 0xFFs to CRC because later the they may be filled
1402 * by data.
1403 */
1404 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1405 aldata_size = data_size =
1406 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1407
1408 cond_resched();
1409 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1410 cond_resched();
1411
1412 /*
1413 * It may turn out to be that the whole @from physical eraseblock
1414 * contains only 0xFF bytes. Then we have to only write the VID header
1415 * and do not write any data. This also means we should not set
1416 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1417 */
1418 if (data_size > 0) {
1419 vid_hdr->copy_flag = 1;
1420 vid_hdr->data_size = cpu_to_be32(data_size);
1421 vid_hdr->data_crc = cpu_to_be32(crc);
1422 }
1423 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1424
1425 err = ubi_io_write_vid_hdr(ubi, to, vidb);
1426 if (err) {
1427 if (err == -EIO)
1428 err = MOVE_TARGET_WR_ERR;
1429 goto out_unlock_buf;
1430 }
1431
1432 cond_resched();
1433
1434 /* Read the VID header back and check if it was written correctly */
1435 err = ubi_io_read_vid_hdr(ubi, to, vidb, 1);
1436 if (err) {
1437 if (err != UBI_IO_BITFLIPS) {
1438 ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1439 err, to);
1440 if (is_error_sane(err))
1441 err = MOVE_TARGET_RD_ERR;
1442 } else
1443 err = MOVE_TARGET_BITFLIPS;
1444 goto out_unlock_buf;
1445 }
1446
1447 if (data_size > 0) {
1448 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1449 if (err) {
1450 if (err == -EIO)
1451 err = MOVE_TARGET_WR_ERR;
1452 goto out_unlock_buf;
1453 }
1454
1455 cond_resched();
1456 }
1457
1458 ubi_assert(vol->eba_tbl->entries[lnum].pnum == from);
1459 vol->eba_tbl->entries[lnum].pnum = to;
1460
1461out_unlock_buf:
1462 mutex_unlock(&ubi->buf_mutex);
1463out_unlock_leb:
1464 leb_write_unlock(ubi, vol_id, lnum);
1465 return err;
1466}
1467
1468/**
1469 * print_rsvd_warning - warn about not having enough reserved PEBs.
1470 * @ubi: UBI device description object
1471 * @ai: UBI attach info object
1472 *
1473 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1474 * cannot reserve enough PEBs for bad block handling. This function makes a
1475 * decision whether we have to print a warning or not. The algorithm is as
1476 * follows:
1477 * o if this is a new UBI image, then just print the warning
1478 * o if this is an UBI image which has already been used for some time, print
1479 * a warning only if we can reserve less than 10% of the expected amount of
1480 * the reserved PEB.
1481 *
1482 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1483 * of PEBs becomes smaller, which is normal and we do not want to scare users
1484 * with a warning every time they attach the MTD device. This was an issue
1485 * reported by real users.
1486 */
1487static void print_rsvd_warning(struct ubi_device *ubi,
1488 struct ubi_attach_info *ai)
1489{
1490 /*
1491 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1492 * large number to distinguish between newly flashed and used images.
1493 */
1494 if (ai->max_sqnum > (1 << 18)) {
1495 int min = ubi->beb_rsvd_level / 10;
1496
1497 if (!min)
1498 min = 1;
1499 if (ubi->beb_rsvd_pebs > min)
1500 return;
1501 }
1502
1503 ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1504 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1505 if (ubi->corr_peb_count)
1506 ubi_warn(ubi, "%d PEBs are corrupted and not used",
1507 ubi->corr_peb_count);
1508}
1509
1510/**
1511 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1512 * @ubi: UBI device description object
1513 * @ai_fastmap: UBI attach info object created by fastmap
1514 * @ai_scan: UBI attach info object created by scanning
1515 *
1516 * Returns < 0 in case of an internal error, 0 otherwise.
1517 * If a bad EBA table entry was found it will be printed out and
1518 * ubi_assert() triggers.
1519 */
1520int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1521 struct ubi_attach_info *ai_scan)
1522{
1523 int i, j, num_volumes, ret = 0;
1524 int **scan_eba, **fm_eba;
1525 struct ubi_ainf_volume *av;
1526 struct ubi_volume *vol;
1527 struct ubi_ainf_peb *aeb;
1528 struct rb_node *rb;
1529
1530 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1531
1532 scan_eba = kmalloc_array(num_volumes, sizeof(*scan_eba), GFP_KERNEL);
1533 if (!scan_eba)
1534 return -ENOMEM;
1535
1536 fm_eba = kmalloc_array(num_volumes, sizeof(*fm_eba), GFP_KERNEL);
1537 if (!fm_eba) {
1538 kfree(scan_eba);
1539 return -ENOMEM;
1540 }
1541
1542 for (i = 0; i < num_volumes; i++) {
1543 vol = ubi->volumes[i];
1544 if (!vol)
1545 continue;
1546
1547 scan_eba[i] = kmalloc_array(vol->reserved_pebs,
1548 sizeof(**scan_eba),
1549 GFP_KERNEL);
1550 if (!scan_eba[i]) {
1551 ret = -ENOMEM;
1552 goto out_free;
1553 }
1554
1555 fm_eba[i] = kmalloc_array(vol->reserved_pebs,
1556 sizeof(**fm_eba),
1557 GFP_KERNEL);
1558 if (!fm_eba[i]) {
1559 ret = -ENOMEM;
1560 goto out_free;
1561 }
1562
1563 for (j = 0; j < vol->reserved_pebs; j++)
1564 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1565
1566 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1567 if (!av)
1568 continue;
1569
1570 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1571 scan_eba[i][aeb->lnum] = aeb->pnum;
1572
1573 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1574 if (!av)
1575 continue;
1576
1577 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1578 fm_eba[i][aeb->lnum] = aeb->pnum;
1579
1580 for (j = 0; j < vol->reserved_pebs; j++) {
1581 if (scan_eba[i][j] != fm_eba[i][j]) {
1582 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1583 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1584 continue;
1585
1586 ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1587 vol->vol_id, j, fm_eba[i][j],
1588 scan_eba[i][j]);
1589 ubi_assert(0);
1590 }
1591 }
1592 }
1593
1594out_free:
1595 for (i = 0; i < num_volumes; i++) {
1596 if (!ubi->volumes[i])
1597 continue;
1598
1599 kfree(scan_eba[i]);
1600 kfree(fm_eba[i]);
1601 }
1602
1603 kfree(scan_eba);
1604 kfree(fm_eba);
1605 return ret;
1606}
1607
1608/**
1609 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1610 * @ubi: UBI device description object
1611 * @ai: attaching information
1612 *
1613 * This function returns zero in case of success and a negative error code in
1614 * case of failure.
1615 */
1616int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1617{
1618 int i, err, num_volumes;
1619 struct ubi_ainf_volume *av;
1620 struct ubi_volume *vol;
1621 struct ubi_ainf_peb *aeb;
1622 struct rb_node *rb;
1623
1624 dbg_eba("initialize EBA sub-system");
1625
1626 spin_lock_init(&ubi->ltree_lock);
1627 mutex_init(&ubi->alc_mutex);
1628 ubi->ltree = RB_ROOT;
1629
1630 ubi->global_sqnum = ai->max_sqnum + 1;
1631 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1632
1633 for (i = 0; i < num_volumes; i++) {
1634 struct ubi_eba_table *tbl;
1635
1636 vol = ubi->volumes[i];
1637 if (!vol)
1638 continue;
1639
1640 cond_resched();
1641
1642 tbl = ubi_eba_create_table(vol, vol->reserved_pebs);
1643 if (IS_ERR(tbl)) {
1644 err = PTR_ERR(tbl);
1645 goto out_free;
1646 }
1647
1648 ubi_eba_replace_table(vol, tbl);
1649
1650 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1651 if (!av)
1652 continue;
1653
1654 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1655 if (aeb->lnum >= vol->reserved_pebs) {
1656 /*
1657 * This may happen in case of an unclean reboot
1658 * during re-size.
1659 */
1660 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1661 } else {
1662 struct ubi_eba_entry *entry;
1663
1664 entry = &vol->eba_tbl->entries[aeb->lnum];
1665 entry->pnum = aeb->pnum;
1666 }
1667 }
1668 }
1669
1670 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1671 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1672 ubi->avail_pebs, EBA_RESERVED_PEBS);
1673 if (ubi->corr_peb_count)
1674 ubi_err(ubi, "%d PEBs are corrupted and not used",
1675 ubi->corr_peb_count);
1676 err = -ENOSPC;
1677 goto out_free;
1678 }
1679 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1680 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1681
1682 if (ubi->bad_allowed) {
1683 ubi_calculate_reserved(ubi);
1684
1685 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1686 /* No enough free physical eraseblocks */
1687 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1688 print_rsvd_warning(ubi, ai);
1689 } else
1690 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1691
1692 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1693 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1694 }
1695
1696 dbg_eba("EBA sub-system is initialized");
1697 return 0;
1698
1699out_free:
1700 for (i = 0; i < num_volumes; i++) {
1701 if (!ubi->volumes[i])
1702 continue;
1703 ubi_eba_replace_table(ubi->volumes[i], NULL);
1704 }
1705 return err;
1706}