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