Loading...
1/*
2 * Copyright (c) International Business Machines Corp., 2006
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 *
18 * Author: Artem Bityutskiy (Битюцкий Артём)
19 */
20
21/*
22 * The UBI Eraseblock Association (EBA) sub-system.
23 *
24 * This sub-system is responsible for I/O to/from logical eraseblock.
25 *
26 * Although in this implementation the EBA table is fully kept and managed in
27 * RAM, which assumes poor scalability, it might be (partially) maintained on
28 * flash in future implementations.
29 *
30 * The EBA sub-system implements per-logical eraseblock locking. Before
31 * accessing a logical eraseblock it is locked for reading or writing. The
32 * per-logical eraseblock locking is implemented by means of the lock tree. The
33 * lock tree is an RB-tree which refers all the currently locked logical
34 * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
35 * They are indexed by (@vol_id, @lnum) pairs.
36 *
37 * EBA also maintains the global sequence counter which is incremented each
38 * time a logical eraseblock is mapped to a physical eraseblock and it is
39 * stored in the volume identifier header. This means that each VID header has
40 * a unique sequence number. The sequence number is only increased an we assume
41 * 64 bits is enough to never overflow.
42 */
43
44#include <linux/slab.h>
45#include <linux/crc32.h>
46#include <linux/err.h>
47#include "ubi.h"
48
49/* Number of physical eraseblocks reserved for atomic LEB change operation */
50#define EBA_RESERVED_PEBS 1
51
52/**
53 * next_sqnum - get next sequence number.
54 * @ubi: UBI device description object
55 *
56 * This function returns next sequence number to use, which is just the current
57 * global sequence counter value. It also increases the global sequence
58 * counter.
59 */
60static unsigned long long next_sqnum(struct ubi_device *ubi)
61{
62 unsigned long long sqnum;
63
64 spin_lock(&ubi->ltree_lock);
65 sqnum = ubi->global_sqnum++;
66 spin_unlock(&ubi->ltree_lock);
67
68 return sqnum;
69}
70
71/**
72 * ubi_get_compat - get compatibility flags of a volume.
73 * @ubi: UBI device description object
74 * @vol_id: volume ID
75 *
76 * This function returns compatibility flags for an internal volume. User
77 * volumes have no compatibility flags, so %0 is returned.
78 */
79static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
80{
81 if (vol_id == UBI_LAYOUT_VOLUME_ID)
82 return UBI_LAYOUT_VOLUME_COMPAT;
83 return 0;
84}
85
86/**
87 * ltree_lookup - look up the lock tree.
88 * @ubi: UBI device description object
89 * @vol_id: volume ID
90 * @lnum: logical eraseblock number
91 *
92 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
93 * object if the logical eraseblock is locked and %NULL if it is not.
94 * @ubi->ltree_lock has to be locked.
95 */
96static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
97 int lnum)
98{
99 struct rb_node *p;
100
101 p = ubi->ltree.rb_node;
102 while (p) {
103 struct ubi_ltree_entry *le;
104
105 le = rb_entry(p, struct ubi_ltree_entry, rb);
106
107 if (vol_id < le->vol_id)
108 p = p->rb_left;
109 else if (vol_id > le->vol_id)
110 p = p->rb_right;
111 else {
112 if (lnum < le->lnum)
113 p = p->rb_left;
114 else if (lnum > le->lnum)
115 p = p->rb_right;
116 else
117 return le;
118 }
119 }
120
121 return NULL;
122}
123
124/**
125 * ltree_add_entry - add new entry to the lock tree.
126 * @ubi: UBI device description object
127 * @vol_id: volume ID
128 * @lnum: logical eraseblock number
129 *
130 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
131 * lock tree. If such entry is already there, its usage counter is increased.
132 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
133 * failed.
134 */
135static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
136 int vol_id, int lnum)
137{
138 struct ubi_ltree_entry *le, *le1, *le_free;
139
140 le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
141 if (!le)
142 return ERR_PTR(-ENOMEM);
143
144 le->users = 0;
145 init_rwsem(&le->mutex);
146 le->vol_id = vol_id;
147 le->lnum = lnum;
148
149 spin_lock(&ubi->ltree_lock);
150 le1 = ltree_lookup(ubi, vol_id, lnum);
151
152 if (le1) {
153 /*
154 * This logical eraseblock is already locked. The newly
155 * allocated lock entry is not needed.
156 */
157 le_free = le;
158 le = le1;
159 } else {
160 struct rb_node **p, *parent = NULL;
161
162 /*
163 * No lock entry, add the newly allocated one to the
164 * @ubi->ltree RB-tree.
165 */
166 le_free = NULL;
167
168 p = &ubi->ltree.rb_node;
169 while (*p) {
170 parent = *p;
171 le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
172
173 if (vol_id < le1->vol_id)
174 p = &(*p)->rb_left;
175 else if (vol_id > le1->vol_id)
176 p = &(*p)->rb_right;
177 else {
178 ubi_assert(lnum != le1->lnum);
179 if (lnum < le1->lnum)
180 p = &(*p)->rb_left;
181 else
182 p = &(*p)->rb_right;
183 }
184 }
185
186 rb_link_node(&le->rb, parent, p);
187 rb_insert_color(&le->rb, &ubi->ltree);
188 }
189 le->users += 1;
190 spin_unlock(&ubi->ltree_lock);
191
192 kfree(le_free);
193 return le;
194}
195
196/**
197 * leb_read_lock - lock logical eraseblock for reading.
198 * @ubi: UBI device description object
199 * @vol_id: volume ID
200 * @lnum: logical eraseblock number
201 *
202 * This function locks a logical eraseblock for reading. Returns zero in case
203 * of success and a negative error code in case of failure.
204 */
205static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
206{
207 struct ubi_ltree_entry *le;
208
209 le = ltree_add_entry(ubi, vol_id, lnum);
210 if (IS_ERR(le))
211 return PTR_ERR(le);
212 down_read(&le->mutex);
213 return 0;
214}
215
216/**
217 * leb_read_unlock - unlock logical eraseblock.
218 * @ubi: UBI device description object
219 * @vol_id: volume ID
220 * @lnum: logical eraseblock number
221 */
222static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
223{
224 struct ubi_ltree_entry *le;
225
226 spin_lock(&ubi->ltree_lock);
227 le = ltree_lookup(ubi, vol_id, lnum);
228 le->users -= 1;
229 ubi_assert(le->users >= 0);
230 up_read(&le->mutex);
231 if (le->users == 0) {
232 rb_erase(&le->rb, &ubi->ltree);
233 kfree(le);
234 }
235 spin_unlock(&ubi->ltree_lock);
236}
237
238/**
239 * leb_write_lock - lock logical eraseblock for writing.
240 * @ubi: UBI device description object
241 * @vol_id: volume ID
242 * @lnum: logical eraseblock number
243 *
244 * This function locks a logical eraseblock for writing. Returns zero in case
245 * of success and a negative error code in case of failure.
246 */
247static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
248{
249 struct ubi_ltree_entry *le;
250
251 le = ltree_add_entry(ubi, vol_id, lnum);
252 if (IS_ERR(le))
253 return PTR_ERR(le);
254 down_write(&le->mutex);
255 return 0;
256}
257
258/**
259 * leb_write_lock - lock logical eraseblock for writing.
260 * @ubi: UBI device description object
261 * @vol_id: volume ID
262 * @lnum: logical eraseblock number
263 *
264 * This function locks a logical eraseblock for writing if there is no
265 * contention and does nothing if there is contention. Returns %0 in case of
266 * success, %1 in case of contention, and and a negative error code in case of
267 * failure.
268 */
269static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
270{
271 struct ubi_ltree_entry *le;
272
273 le = ltree_add_entry(ubi, vol_id, lnum);
274 if (IS_ERR(le))
275 return PTR_ERR(le);
276 if (down_write_trylock(&le->mutex))
277 return 0;
278
279 /* Contention, cancel */
280 spin_lock(&ubi->ltree_lock);
281 le->users -= 1;
282 ubi_assert(le->users >= 0);
283 if (le->users == 0) {
284 rb_erase(&le->rb, &ubi->ltree);
285 kfree(le);
286 }
287 spin_unlock(&ubi->ltree_lock);
288
289 return 1;
290}
291
292/**
293 * leb_write_unlock - unlock logical eraseblock.
294 * @ubi: UBI device description object
295 * @vol_id: volume ID
296 * @lnum: logical eraseblock number
297 */
298static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
299{
300 struct ubi_ltree_entry *le;
301
302 spin_lock(&ubi->ltree_lock);
303 le = ltree_lookup(ubi, vol_id, lnum);
304 le->users -= 1;
305 ubi_assert(le->users >= 0);
306 up_write(&le->mutex);
307 if (le->users == 0) {
308 rb_erase(&le->rb, &ubi->ltree);
309 kfree(le);
310 }
311 spin_unlock(&ubi->ltree_lock);
312}
313
314/**
315 * ubi_eba_unmap_leb - un-map logical eraseblock.
316 * @ubi: UBI device description object
317 * @vol: volume description object
318 * @lnum: logical eraseblock number
319 *
320 * This function un-maps logical eraseblock @lnum and schedules corresponding
321 * physical eraseblock for erasure. Returns zero in case of success and a
322 * negative error code in case of failure.
323 */
324int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
325 int lnum)
326{
327 int err, pnum, vol_id = vol->vol_id;
328
329 if (ubi->ro_mode)
330 return -EROFS;
331
332 err = leb_write_lock(ubi, vol_id, lnum);
333 if (err)
334 return err;
335
336 pnum = vol->eba_tbl[lnum];
337 if (pnum < 0)
338 /* This logical eraseblock is already unmapped */
339 goto out_unlock;
340
341 dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
342
343 vol->eba_tbl[lnum] = UBI_LEB_UNMAPPED;
344 err = ubi_wl_put_peb(ubi, pnum, 0);
345
346out_unlock:
347 leb_write_unlock(ubi, vol_id, lnum);
348 return err;
349}
350
351/**
352 * ubi_eba_read_leb - read data.
353 * @ubi: UBI device description object
354 * @vol: volume description object
355 * @lnum: logical eraseblock number
356 * @buf: buffer to store the read data
357 * @offset: offset from where to read
358 * @len: how many bytes to read
359 * @check: data CRC check flag
360 *
361 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
362 * bytes. The @check flag only makes sense for static volumes and forces
363 * eraseblock data CRC checking.
364 *
365 * In case of success this function returns zero. In case of a static volume,
366 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
367 * returned for any volume type if an ECC error was detected by the MTD device
368 * driver. Other negative error cored may be returned in case of other errors.
369 */
370int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
371 void *buf, int offset, int len, int check)
372{
373 int err, pnum, scrub = 0, vol_id = vol->vol_id;
374 struct ubi_vid_hdr *vid_hdr;
375 uint32_t uninitialized_var(crc);
376
377 err = leb_read_lock(ubi, vol_id, lnum);
378 if (err)
379 return err;
380
381 pnum = vol->eba_tbl[lnum];
382 if (pnum < 0) {
383 /*
384 * The logical eraseblock is not mapped, fill the whole buffer
385 * with 0xFF bytes. The exception is static volumes for which
386 * it is an error to read unmapped logical eraseblocks.
387 */
388 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
389 len, offset, vol_id, lnum);
390 leb_read_unlock(ubi, vol_id, lnum);
391 ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
392 memset(buf, 0xFF, len);
393 return 0;
394 }
395
396 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
397 len, offset, vol_id, lnum, pnum);
398
399 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
400 check = 0;
401
402retry:
403 if (check) {
404 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
405 if (!vid_hdr) {
406 err = -ENOMEM;
407 goto out_unlock;
408 }
409
410 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
411 if (err && err != UBI_IO_BITFLIPS) {
412 if (err > 0) {
413 /*
414 * The header is either absent or corrupted.
415 * The former case means there is a bug -
416 * switch to read-only mode just in case.
417 * The latter case means a real corruption - we
418 * may try to recover data. FIXME: but this is
419 * not implemented.
420 */
421 if (err == UBI_IO_BAD_HDR_EBADMSG ||
422 err == UBI_IO_BAD_HDR) {
423 ubi_warn("corrupted VID header at PEB "
424 "%d, LEB %d:%d", pnum, vol_id,
425 lnum);
426 err = -EBADMSG;
427 } else
428 ubi_ro_mode(ubi);
429 }
430 goto out_free;
431 } else if (err == UBI_IO_BITFLIPS)
432 scrub = 1;
433
434 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
435 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
436
437 crc = be32_to_cpu(vid_hdr->data_crc);
438 ubi_free_vid_hdr(ubi, vid_hdr);
439 }
440
441 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
442 if (err) {
443 if (err == UBI_IO_BITFLIPS) {
444 scrub = 1;
445 err = 0;
446 } else if (err == -EBADMSG) {
447 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
448 goto out_unlock;
449 scrub = 1;
450 if (!check) {
451 ubi_msg("force data checking");
452 check = 1;
453 goto retry;
454 }
455 } else
456 goto out_unlock;
457 }
458
459 if (check) {
460 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
461 if (crc1 != crc) {
462 ubi_warn("CRC error: calculated %#08x, must be %#08x",
463 crc1, crc);
464 err = -EBADMSG;
465 goto out_unlock;
466 }
467 }
468
469 if (scrub)
470 err = ubi_wl_scrub_peb(ubi, pnum);
471
472 leb_read_unlock(ubi, vol_id, lnum);
473 return err;
474
475out_free:
476 ubi_free_vid_hdr(ubi, vid_hdr);
477out_unlock:
478 leb_read_unlock(ubi, vol_id, lnum);
479 return err;
480}
481
482/**
483 * recover_peb - recover from write failure.
484 * @ubi: UBI device description object
485 * @pnum: the physical eraseblock to recover
486 * @vol_id: volume ID
487 * @lnum: logical eraseblock number
488 * @buf: data which was not written because of the write failure
489 * @offset: offset of the failed write
490 * @len: how many bytes should have been written
491 *
492 * This function is called in case of a write failure and moves all good data
493 * from the potentially bad physical eraseblock to a good physical eraseblock.
494 * This function also writes the data which was not written due to the failure.
495 * Returns new physical eraseblock number in case of success, and a negative
496 * error code in case of failure.
497 */
498static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
499 const void *buf, int offset, int len)
500{
501 int err, idx = vol_id2idx(ubi, vol_id), new_pnum, data_size, tries = 0;
502 struct ubi_volume *vol = ubi->volumes[idx];
503 struct ubi_vid_hdr *vid_hdr;
504
505 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
506 if (!vid_hdr)
507 return -ENOMEM;
508
509retry:
510 new_pnum = ubi_wl_get_peb(ubi, UBI_UNKNOWN);
511 if (new_pnum < 0) {
512 ubi_free_vid_hdr(ubi, vid_hdr);
513 return new_pnum;
514 }
515
516 ubi_msg("recover PEB %d, move data to PEB %d", pnum, new_pnum);
517
518 err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
519 if (err && err != UBI_IO_BITFLIPS) {
520 if (err > 0)
521 err = -EIO;
522 goto out_put;
523 }
524
525 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
526 err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
527 if (err)
528 goto write_error;
529
530 data_size = offset + len;
531 mutex_lock(&ubi->buf_mutex);
532 memset(ubi->peb_buf1 + offset, 0xFF, len);
533
534 /* Read everything before the area where the write failure happened */
535 if (offset > 0) {
536 err = ubi_io_read_data(ubi, ubi->peb_buf1, pnum, 0, offset);
537 if (err && err != UBI_IO_BITFLIPS)
538 goto out_unlock;
539 }
540
541 memcpy(ubi->peb_buf1 + offset, buf, len);
542
543 err = ubi_io_write_data(ubi, ubi->peb_buf1, new_pnum, 0, data_size);
544 if (err) {
545 mutex_unlock(&ubi->buf_mutex);
546 goto write_error;
547 }
548
549 mutex_unlock(&ubi->buf_mutex);
550 ubi_free_vid_hdr(ubi, vid_hdr);
551
552 vol->eba_tbl[lnum] = new_pnum;
553 ubi_wl_put_peb(ubi, pnum, 1);
554
555 ubi_msg("data was successfully recovered");
556 return 0;
557
558out_unlock:
559 mutex_unlock(&ubi->buf_mutex);
560out_put:
561 ubi_wl_put_peb(ubi, new_pnum, 1);
562 ubi_free_vid_hdr(ubi, vid_hdr);
563 return err;
564
565write_error:
566 /*
567 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
568 * get another one.
569 */
570 ubi_warn("failed to write to PEB %d", new_pnum);
571 ubi_wl_put_peb(ubi, new_pnum, 1);
572 if (++tries > UBI_IO_RETRIES) {
573 ubi_free_vid_hdr(ubi, vid_hdr);
574 return err;
575 }
576 ubi_msg("try again");
577 goto retry;
578}
579
580/**
581 * ubi_eba_write_leb - write data to dynamic volume.
582 * @ubi: UBI device description object
583 * @vol: volume description object
584 * @lnum: logical eraseblock number
585 * @buf: the data to write
586 * @offset: offset within the logical eraseblock where to write
587 * @len: how many bytes to write
588 * @dtype: data type
589 *
590 * This function writes data to logical eraseblock @lnum of a dynamic volume
591 * @vol. Returns zero in case of success and a negative error code in case
592 * of failure. In case of error, it is possible that something was still
593 * written to the flash media, but may be some garbage.
594 */
595int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
596 const void *buf, int offset, int len, int dtype)
597{
598 int err, pnum, tries = 0, vol_id = vol->vol_id;
599 struct ubi_vid_hdr *vid_hdr;
600
601 if (ubi->ro_mode)
602 return -EROFS;
603
604 err = leb_write_lock(ubi, vol_id, lnum);
605 if (err)
606 return err;
607
608 pnum = vol->eba_tbl[lnum];
609 if (pnum >= 0) {
610 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
611 len, offset, vol_id, lnum, pnum);
612
613 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
614 if (err) {
615 ubi_warn("failed to write data to PEB %d", pnum);
616 if (err == -EIO && ubi->bad_allowed)
617 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
618 offset, len);
619 if (err)
620 ubi_ro_mode(ubi);
621 }
622 leb_write_unlock(ubi, vol_id, lnum);
623 return err;
624 }
625
626 /*
627 * The logical eraseblock is not mapped. We have to get a free physical
628 * eraseblock and write the volume identifier header there first.
629 */
630 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
631 if (!vid_hdr) {
632 leb_write_unlock(ubi, vol_id, lnum);
633 return -ENOMEM;
634 }
635
636 vid_hdr->vol_type = UBI_VID_DYNAMIC;
637 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
638 vid_hdr->vol_id = cpu_to_be32(vol_id);
639 vid_hdr->lnum = cpu_to_be32(lnum);
640 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
641 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
642
643retry:
644 pnum = ubi_wl_get_peb(ubi, dtype);
645 if (pnum < 0) {
646 ubi_free_vid_hdr(ubi, vid_hdr);
647 leb_write_unlock(ubi, vol_id, lnum);
648 return pnum;
649 }
650
651 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
652 len, offset, vol_id, lnum, pnum);
653
654 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
655 if (err) {
656 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
657 vol_id, lnum, pnum);
658 goto write_error;
659 }
660
661 if (len) {
662 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
663 if (err) {
664 ubi_warn("failed to write %d bytes at offset %d of "
665 "LEB %d:%d, PEB %d", len, offset, vol_id,
666 lnum, pnum);
667 goto write_error;
668 }
669 }
670
671 vol->eba_tbl[lnum] = pnum;
672
673 leb_write_unlock(ubi, vol_id, lnum);
674 ubi_free_vid_hdr(ubi, vid_hdr);
675 return 0;
676
677write_error:
678 if (err != -EIO || !ubi->bad_allowed) {
679 ubi_ro_mode(ubi);
680 leb_write_unlock(ubi, vol_id, lnum);
681 ubi_free_vid_hdr(ubi, vid_hdr);
682 return err;
683 }
684
685 /*
686 * Fortunately, this is the first write operation to this physical
687 * eraseblock, so just put it and request a new one. We assume that if
688 * this physical eraseblock went bad, the erase code will handle that.
689 */
690 err = ubi_wl_put_peb(ubi, pnum, 1);
691 if (err || ++tries > UBI_IO_RETRIES) {
692 ubi_ro_mode(ubi);
693 leb_write_unlock(ubi, vol_id, lnum);
694 ubi_free_vid_hdr(ubi, vid_hdr);
695 return err;
696 }
697
698 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
699 ubi_msg("try another PEB");
700 goto retry;
701}
702
703/**
704 * ubi_eba_write_leb_st - write data to static volume.
705 * @ubi: UBI device description object
706 * @vol: volume description object
707 * @lnum: logical eraseblock number
708 * @buf: data to write
709 * @len: how many bytes to write
710 * @dtype: data type
711 * @used_ebs: how many logical eraseblocks will this volume contain
712 *
713 * This function writes data to logical eraseblock @lnum of static volume
714 * @vol. The @used_ebs argument should contain total number of logical
715 * eraseblock in this static volume.
716 *
717 * When writing to the last logical eraseblock, the @len argument doesn't have
718 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
719 * to the real data size, although the @buf buffer has to contain the
720 * alignment. In all other cases, @len has to be aligned.
721 *
722 * It is prohibited to write more than once to logical eraseblocks of static
723 * volumes. This function returns zero in case of success and a negative error
724 * code in case of failure.
725 */
726int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
727 int lnum, const void *buf, int len, int dtype,
728 int used_ebs)
729{
730 int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
731 struct ubi_vid_hdr *vid_hdr;
732 uint32_t crc;
733
734 if (ubi->ro_mode)
735 return -EROFS;
736
737 if (lnum == used_ebs - 1)
738 /* If this is the last LEB @len may be unaligned */
739 len = ALIGN(data_size, ubi->min_io_size);
740 else
741 ubi_assert(!(len & (ubi->min_io_size - 1)));
742
743 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
744 if (!vid_hdr)
745 return -ENOMEM;
746
747 err = leb_write_lock(ubi, vol_id, lnum);
748 if (err) {
749 ubi_free_vid_hdr(ubi, vid_hdr);
750 return err;
751 }
752
753 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
754 vid_hdr->vol_id = cpu_to_be32(vol_id);
755 vid_hdr->lnum = cpu_to_be32(lnum);
756 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
757 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
758
759 crc = crc32(UBI_CRC32_INIT, buf, data_size);
760 vid_hdr->vol_type = UBI_VID_STATIC;
761 vid_hdr->data_size = cpu_to_be32(data_size);
762 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
763 vid_hdr->data_crc = cpu_to_be32(crc);
764
765retry:
766 pnum = ubi_wl_get_peb(ubi, dtype);
767 if (pnum < 0) {
768 ubi_free_vid_hdr(ubi, vid_hdr);
769 leb_write_unlock(ubi, vol_id, lnum);
770 return pnum;
771 }
772
773 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
774 len, vol_id, lnum, pnum, used_ebs);
775
776 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
777 if (err) {
778 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
779 vol_id, lnum, pnum);
780 goto write_error;
781 }
782
783 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
784 if (err) {
785 ubi_warn("failed to write %d bytes of data to PEB %d",
786 len, pnum);
787 goto write_error;
788 }
789
790 ubi_assert(vol->eba_tbl[lnum] < 0);
791 vol->eba_tbl[lnum] = pnum;
792
793 leb_write_unlock(ubi, vol_id, lnum);
794 ubi_free_vid_hdr(ubi, vid_hdr);
795 return 0;
796
797write_error:
798 if (err != -EIO || !ubi->bad_allowed) {
799 /*
800 * This flash device does not admit of bad eraseblocks or
801 * something nasty and unexpected happened. Switch to read-only
802 * mode just in case.
803 */
804 ubi_ro_mode(ubi);
805 leb_write_unlock(ubi, vol_id, lnum);
806 ubi_free_vid_hdr(ubi, vid_hdr);
807 return err;
808 }
809
810 err = ubi_wl_put_peb(ubi, pnum, 1);
811 if (err || ++tries > UBI_IO_RETRIES) {
812 ubi_ro_mode(ubi);
813 leb_write_unlock(ubi, vol_id, lnum);
814 ubi_free_vid_hdr(ubi, vid_hdr);
815 return err;
816 }
817
818 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
819 ubi_msg("try another PEB");
820 goto retry;
821}
822
823/*
824 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
825 * @ubi: UBI device description object
826 * @vol: volume description object
827 * @lnum: logical eraseblock number
828 * @buf: data to write
829 * @len: how many bytes to write
830 * @dtype: data type
831 *
832 * This function changes the contents of a logical eraseblock atomically. @buf
833 * has to contain new logical eraseblock data, and @len - the length of the
834 * data, which has to be aligned. This function guarantees that in case of an
835 * unclean reboot the old contents is preserved. Returns zero in case of
836 * success and a negative error code in case of failure.
837 *
838 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
839 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
840 */
841int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
842 int lnum, const void *buf, int len, int dtype)
843{
844 int err, pnum, tries = 0, vol_id = vol->vol_id;
845 struct ubi_vid_hdr *vid_hdr;
846 uint32_t crc;
847
848 if (ubi->ro_mode)
849 return -EROFS;
850
851 if (len == 0) {
852 /*
853 * Special case when data length is zero. In this case the LEB
854 * has to be unmapped and mapped somewhere else.
855 */
856 err = ubi_eba_unmap_leb(ubi, vol, lnum);
857 if (err)
858 return err;
859 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0, dtype);
860 }
861
862 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
863 if (!vid_hdr)
864 return -ENOMEM;
865
866 mutex_lock(&ubi->alc_mutex);
867 err = leb_write_lock(ubi, vol_id, lnum);
868 if (err)
869 goto out_mutex;
870
871 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
872 vid_hdr->vol_id = cpu_to_be32(vol_id);
873 vid_hdr->lnum = cpu_to_be32(lnum);
874 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
875 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
876
877 crc = crc32(UBI_CRC32_INIT, buf, len);
878 vid_hdr->vol_type = UBI_VID_DYNAMIC;
879 vid_hdr->data_size = cpu_to_be32(len);
880 vid_hdr->copy_flag = 1;
881 vid_hdr->data_crc = cpu_to_be32(crc);
882
883retry:
884 pnum = ubi_wl_get_peb(ubi, dtype);
885 if (pnum < 0) {
886 err = pnum;
887 goto out_leb_unlock;
888 }
889
890 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
891 vol_id, lnum, vol->eba_tbl[lnum], pnum);
892
893 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
894 if (err) {
895 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
896 vol_id, lnum, pnum);
897 goto write_error;
898 }
899
900 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
901 if (err) {
902 ubi_warn("failed to write %d bytes of data to PEB %d",
903 len, pnum);
904 goto write_error;
905 }
906
907 if (vol->eba_tbl[lnum] >= 0) {
908 err = ubi_wl_put_peb(ubi, vol->eba_tbl[lnum], 0);
909 if (err)
910 goto out_leb_unlock;
911 }
912
913 vol->eba_tbl[lnum] = pnum;
914
915out_leb_unlock:
916 leb_write_unlock(ubi, vol_id, lnum);
917out_mutex:
918 mutex_unlock(&ubi->alc_mutex);
919 ubi_free_vid_hdr(ubi, vid_hdr);
920 return err;
921
922write_error:
923 if (err != -EIO || !ubi->bad_allowed) {
924 /*
925 * This flash device does not admit of bad eraseblocks or
926 * something nasty and unexpected happened. Switch to read-only
927 * mode just in case.
928 */
929 ubi_ro_mode(ubi);
930 goto out_leb_unlock;
931 }
932
933 err = ubi_wl_put_peb(ubi, pnum, 1);
934 if (err || ++tries > UBI_IO_RETRIES) {
935 ubi_ro_mode(ubi);
936 goto out_leb_unlock;
937 }
938
939 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
940 ubi_msg("try another PEB");
941 goto retry;
942}
943
944/**
945 * is_error_sane - check whether a read error is sane.
946 * @err: code of the error happened during reading
947 *
948 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
949 * cannot read data from the target PEB (an error @err happened). If the error
950 * code is sane, then we treat this error as non-fatal. Otherwise the error is
951 * fatal and UBI will be switched to R/O mode later.
952 *
953 * The idea is that we try not to switch to R/O mode if the read error is
954 * something which suggests there was a real read problem. E.g., %-EIO. Or a
955 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
956 * mode, simply because we do not know what happened at the MTD level, and we
957 * cannot handle this. E.g., the underlying driver may have become crazy, and
958 * it is safer to switch to R/O mode to preserve the data.
959 *
960 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
961 * which we have just written.
962 */
963static int is_error_sane(int err)
964{
965 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
966 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
967 return 0;
968 return 1;
969}
970
971/**
972 * ubi_eba_copy_leb - copy logical eraseblock.
973 * @ubi: UBI device description object
974 * @from: physical eraseblock number from where to copy
975 * @to: physical eraseblock number where to copy
976 * @vid_hdr: VID header of the @from physical eraseblock
977 *
978 * This function copies logical eraseblock from physical eraseblock @from to
979 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
980 * function. Returns:
981 * o %0 in case of success;
982 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_CANCEL_BITFLIPS, etc;
983 * o a negative error code in case of failure.
984 */
985int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
986 struct ubi_vid_hdr *vid_hdr)
987{
988 int err, vol_id, lnum, data_size, aldata_size, idx;
989 struct ubi_volume *vol;
990 uint32_t crc;
991
992 vol_id = be32_to_cpu(vid_hdr->vol_id);
993 lnum = be32_to_cpu(vid_hdr->lnum);
994
995 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
996
997 if (vid_hdr->vol_type == UBI_VID_STATIC) {
998 data_size = be32_to_cpu(vid_hdr->data_size);
999 aldata_size = ALIGN(data_size, ubi->min_io_size);
1000 } else
1001 data_size = aldata_size =
1002 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1003
1004 idx = vol_id2idx(ubi, vol_id);
1005 spin_lock(&ubi->volumes_lock);
1006 /*
1007 * Note, we may race with volume deletion, which means that the volume
1008 * this logical eraseblock belongs to might be being deleted. Since the
1009 * volume deletion un-maps all the volume's logical eraseblocks, it will
1010 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1011 */
1012 vol = ubi->volumes[idx];
1013 spin_unlock(&ubi->volumes_lock);
1014 if (!vol) {
1015 /* No need to do further work, cancel */
1016 dbg_wl("volume %d is being removed, cancel", vol_id);
1017 return MOVE_CANCEL_RACE;
1018 }
1019
1020 /*
1021 * We do not want anybody to write to this logical eraseblock while we
1022 * are moving it, so lock it.
1023 *
1024 * Note, we are using non-waiting locking here, because we cannot sleep
1025 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1026 * unmapping the LEB which is mapped to the PEB we are going to move
1027 * (@from). This task locks the LEB and goes sleep in the
1028 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1029 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1030 * LEB is already locked, we just do not move it and return
1031 * %MOVE_CANCEL_RACE, which means that UBI will re-try, but later.
1032 */
1033 err = leb_write_trylock(ubi, vol_id, lnum);
1034 if (err) {
1035 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1036 return MOVE_CANCEL_RACE;
1037 }
1038
1039 /*
1040 * The LEB might have been put meanwhile, and the task which put it is
1041 * probably waiting on @ubi->move_mutex. No need to continue the work,
1042 * cancel it.
1043 */
1044 if (vol->eba_tbl[lnum] != from) {
1045 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to "
1046 "PEB %d, cancel", vol_id, lnum, from,
1047 vol->eba_tbl[lnum]);
1048 err = MOVE_CANCEL_RACE;
1049 goto out_unlock_leb;
1050 }
1051
1052 /*
1053 * OK, now the LEB is locked and we can safely start moving it. Since
1054 * this function utilizes the @ubi->peb_buf1 buffer which is shared
1055 * with some other functions - we lock the buffer by taking the
1056 * @ubi->buf_mutex.
1057 */
1058 mutex_lock(&ubi->buf_mutex);
1059 dbg_wl("read %d bytes of data", aldata_size);
1060 err = ubi_io_read_data(ubi, ubi->peb_buf1, from, 0, aldata_size);
1061 if (err && err != UBI_IO_BITFLIPS) {
1062 ubi_warn("error %d while reading data from PEB %d",
1063 err, from);
1064 err = MOVE_SOURCE_RD_ERR;
1065 goto out_unlock_buf;
1066 }
1067
1068 /*
1069 * Now we have got to calculate how much data we have to copy. In
1070 * case of a static volume it is fairly easy - the VID header contains
1071 * the data size. In case of a dynamic volume it is more difficult - we
1072 * have to read the contents, cut 0xFF bytes from the end and copy only
1073 * the first part. We must do this to avoid writing 0xFF bytes as it
1074 * may have some side-effects. And not only this. It is important not
1075 * to include those 0xFFs to CRC because later the they may be filled
1076 * by data.
1077 */
1078 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1079 aldata_size = data_size =
1080 ubi_calc_data_len(ubi, ubi->peb_buf1, data_size);
1081
1082 cond_resched();
1083 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf1, data_size);
1084 cond_resched();
1085
1086 /*
1087 * It may turn out to be that the whole @from physical eraseblock
1088 * contains only 0xFF bytes. Then we have to only write the VID header
1089 * and do not write any data. This also means we should not set
1090 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1091 */
1092 if (data_size > 0) {
1093 vid_hdr->copy_flag = 1;
1094 vid_hdr->data_size = cpu_to_be32(data_size);
1095 vid_hdr->data_crc = cpu_to_be32(crc);
1096 }
1097 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
1098
1099 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1100 if (err) {
1101 if (err == -EIO)
1102 err = MOVE_TARGET_WR_ERR;
1103 goto out_unlock_buf;
1104 }
1105
1106 cond_resched();
1107
1108 /* Read the VID header back and check if it was written correctly */
1109 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1110 if (err) {
1111 if (err != UBI_IO_BITFLIPS) {
1112 ubi_warn("error %d while reading VID header back from "
1113 "PEB %d", err, to);
1114 if (is_error_sane(err))
1115 err = MOVE_TARGET_RD_ERR;
1116 } else
1117 err = MOVE_CANCEL_BITFLIPS;
1118 goto out_unlock_buf;
1119 }
1120
1121 if (data_size > 0) {
1122 err = ubi_io_write_data(ubi, ubi->peb_buf1, to, 0, aldata_size);
1123 if (err) {
1124 if (err == -EIO)
1125 err = MOVE_TARGET_WR_ERR;
1126 goto out_unlock_buf;
1127 }
1128
1129 cond_resched();
1130
1131 /*
1132 * We've written the data and are going to read it back to make
1133 * sure it was written correctly.
1134 */
1135
1136 err = ubi_io_read_data(ubi, ubi->peb_buf2, to, 0, aldata_size);
1137 if (err) {
1138 if (err != UBI_IO_BITFLIPS) {
1139 ubi_warn("error %d while reading data back "
1140 "from PEB %d", err, to);
1141 if (is_error_sane(err))
1142 err = MOVE_TARGET_RD_ERR;
1143 } else
1144 err = MOVE_CANCEL_BITFLIPS;
1145 goto out_unlock_buf;
1146 }
1147
1148 cond_resched();
1149
1150 if (memcmp(ubi->peb_buf1, ubi->peb_buf2, aldata_size)) {
1151 ubi_warn("read data back from PEB %d and it is "
1152 "different", to);
1153 err = -EINVAL;
1154 goto out_unlock_buf;
1155 }
1156 }
1157
1158 ubi_assert(vol->eba_tbl[lnum] == from);
1159 vol->eba_tbl[lnum] = to;
1160
1161out_unlock_buf:
1162 mutex_unlock(&ubi->buf_mutex);
1163out_unlock_leb:
1164 leb_write_unlock(ubi, vol_id, lnum);
1165 return err;
1166}
1167
1168/**
1169 * print_rsvd_warning - warn about not having enough reserved PEBs.
1170 * @ubi: UBI device description object
1171 *
1172 * This is a helper function for 'ubi_eba_init_scan()' which is called when UBI
1173 * cannot reserve enough PEBs for bad block handling. This function makes a
1174 * decision whether we have to print a warning or not. The algorithm is as
1175 * follows:
1176 * o if this is a new UBI image, then just print the warning
1177 * o if this is an UBI image which has already been used for some time, print
1178 * a warning only if we can reserve less than 10% of the expected amount of
1179 * the reserved PEB.
1180 *
1181 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1182 * of PEBs becomes smaller, which is normal and we do not want to scare users
1183 * with a warning every time they attach the MTD device. This was an issue
1184 * reported by real users.
1185 */
1186static void print_rsvd_warning(struct ubi_device *ubi,
1187 struct ubi_scan_info *si)
1188{
1189 /*
1190 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1191 * large number to distinguish between newly flashed and used images.
1192 */
1193 if (si->max_sqnum > (1 << 18)) {
1194 int min = ubi->beb_rsvd_level / 10;
1195
1196 if (!min)
1197 min = 1;
1198 if (ubi->beb_rsvd_pebs > min)
1199 return;
1200 }
1201
1202 ubi_warn("cannot reserve enough PEBs for bad PEB handling, reserved %d,"
1203 " need %d", ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1204 if (ubi->corr_peb_count)
1205 ubi_warn("%d PEBs are corrupted and not used",
1206 ubi->corr_peb_count);
1207}
1208
1209/**
1210 * ubi_eba_init_scan - initialize the EBA sub-system using scanning information.
1211 * @ubi: UBI device description object
1212 * @si: scanning information
1213 *
1214 * This function returns zero in case of success and a negative error code in
1215 * case of failure.
1216 */
1217int ubi_eba_init_scan(struct ubi_device *ubi, struct ubi_scan_info *si)
1218{
1219 int i, j, err, num_volumes;
1220 struct ubi_scan_volume *sv;
1221 struct ubi_volume *vol;
1222 struct ubi_scan_leb *seb;
1223 struct rb_node *rb;
1224
1225 dbg_eba("initialize EBA sub-system");
1226
1227 spin_lock_init(&ubi->ltree_lock);
1228 mutex_init(&ubi->alc_mutex);
1229 ubi->ltree = RB_ROOT;
1230
1231 ubi->global_sqnum = si->max_sqnum + 1;
1232 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1233
1234 for (i = 0; i < num_volumes; i++) {
1235 vol = ubi->volumes[i];
1236 if (!vol)
1237 continue;
1238
1239 cond_resched();
1240
1241 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1242 GFP_KERNEL);
1243 if (!vol->eba_tbl) {
1244 err = -ENOMEM;
1245 goto out_free;
1246 }
1247
1248 for (j = 0; j < vol->reserved_pebs; j++)
1249 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1250
1251 sv = ubi_scan_find_sv(si, idx2vol_id(ubi, i));
1252 if (!sv)
1253 continue;
1254
1255 ubi_rb_for_each_entry(rb, seb, &sv->root, u.rb) {
1256 if (seb->lnum >= vol->reserved_pebs)
1257 /*
1258 * This may happen in case of an unclean reboot
1259 * during re-size.
1260 */
1261 ubi_scan_move_to_list(sv, seb, &si->erase);
1262 vol->eba_tbl[seb->lnum] = seb->pnum;
1263 }
1264 }
1265
1266 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1267 ubi_err("no enough physical eraseblocks (%d, need %d)",
1268 ubi->avail_pebs, EBA_RESERVED_PEBS);
1269 if (ubi->corr_peb_count)
1270 ubi_err("%d PEBs are corrupted and not used",
1271 ubi->corr_peb_count);
1272 err = -ENOSPC;
1273 goto out_free;
1274 }
1275 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1276 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1277
1278 if (ubi->bad_allowed) {
1279 ubi_calculate_reserved(ubi);
1280
1281 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1282 /* No enough free physical eraseblocks */
1283 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1284 print_rsvd_warning(ubi, si);
1285 } else
1286 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1287
1288 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1289 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1290 }
1291
1292 dbg_eba("EBA sub-system is initialized");
1293 return 0;
1294
1295out_free:
1296 for (i = 0; i < num_volumes; i++) {
1297 if (!ubi->volumes[i])
1298 continue;
1299 kfree(ubi->volumes[i]->eba_tbl);
1300 ubi->volumes[i]->eba_tbl = NULL;
1301 }
1302 return err;
1303}
1/*
2 * Copyright (c) International Business Machines Corp., 2006
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
12 * the GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17 *
18 * Author: Artem Bityutskiy (Битюцкий Артём)
19 */
20
21/*
22 * The UBI Eraseblock Association (EBA) sub-system.
23 *
24 * This sub-system is responsible for I/O to/from logical eraseblock.
25 *
26 * Although in this implementation the EBA table is fully kept and managed in
27 * RAM, which assumes poor scalability, it might be (partially) maintained on
28 * flash in future implementations.
29 *
30 * The EBA sub-system implements per-logical eraseblock locking. Before
31 * accessing a logical eraseblock it is locked for reading or writing. The
32 * per-logical eraseblock locking is implemented by means of the lock tree. The
33 * lock tree is an RB-tree which refers all the currently locked logical
34 * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
35 * They are indexed by (@vol_id, @lnum) pairs.
36 *
37 * EBA also maintains the global sequence counter which is incremented each
38 * time a logical eraseblock is mapped to a physical eraseblock and it is
39 * stored in the volume identifier header. This means that each VID header has
40 * a unique sequence number. The sequence number is only increased an we assume
41 * 64 bits is enough to never overflow.
42 */
43
44#include <linux/slab.h>
45#include <linux/crc32.h>
46#include <linux/err.h>
47#include "ubi.h"
48
49/* Number of physical eraseblocks reserved for atomic LEB change operation */
50#define EBA_RESERVED_PEBS 1
51
52/**
53 * 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, vol_id, lnum, 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 (mtd_is_eccerr(err)) {
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);
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_buf + 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_buf, pnum, 0, offset);
537 if (err && err != UBI_IO_BITFLIPS)
538 goto out_unlock;
539 }
540
541 memcpy(ubi->peb_buf + offset, buf, len);
542
543 err = ubi_io_write_data(ubi, ubi->peb_buf, 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, vol_id, lnum, 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, vol_id, lnum, 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, vol_id, lnum, 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 *
589 * This function writes data to logical eraseblock @lnum of a dynamic volume
590 * @vol. Returns zero in case of success and a negative error code in case
591 * of failure. In case of error, it is possible that something was still
592 * written to the flash media, but may be some garbage.
593 */
594int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
595 const void *buf, int offset, int len)
596{
597 int err, pnum, tries = 0, vol_id = vol->vol_id;
598 struct ubi_vid_hdr *vid_hdr;
599
600 if (ubi->ro_mode)
601 return -EROFS;
602
603 err = leb_write_lock(ubi, vol_id, lnum);
604 if (err)
605 return err;
606
607 pnum = vol->eba_tbl[lnum];
608 if (pnum >= 0) {
609 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
610 len, offset, vol_id, lnum, pnum);
611
612 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
613 if (err) {
614 ubi_warn("failed to write data to PEB %d", pnum);
615 if (err == -EIO && ubi->bad_allowed)
616 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
617 offset, len);
618 if (err)
619 ubi_ro_mode(ubi);
620 }
621 leb_write_unlock(ubi, vol_id, lnum);
622 return err;
623 }
624
625 /*
626 * The logical eraseblock is not mapped. We have to get a free physical
627 * eraseblock and write the volume identifier header there first.
628 */
629 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
630 if (!vid_hdr) {
631 leb_write_unlock(ubi, vol_id, lnum);
632 return -ENOMEM;
633 }
634
635 vid_hdr->vol_type = UBI_VID_DYNAMIC;
636 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
637 vid_hdr->vol_id = cpu_to_be32(vol_id);
638 vid_hdr->lnum = cpu_to_be32(lnum);
639 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
640 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
641
642retry:
643 pnum = ubi_wl_get_peb(ubi);
644 if (pnum < 0) {
645 ubi_free_vid_hdr(ubi, vid_hdr);
646 leb_write_unlock(ubi, vol_id, lnum);
647 return pnum;
648 }
649
650 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
651 len, offset, vol_id, lnum, pnum);
652
653 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
654 if (err) {
655 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
656 vol_id, lnum, pnum);
657 goto write_error;
658 }
659
660 if (len) {
661 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
662 if (err) {
663 ubi_warn("failed to write %d bytes at offset %d of "
664 "LEB %d:%d, PEB %d", len, offset, vol_id,
665 lnum, pnum);
666 goto write_error;
667 }
668 }
669
670 vol->eba_tbl[lnum] = pnum;
671
672 leb_write_unlock(ubi, vol_id, lnum);
673 ubi_free_vid_hdr(ubi, vid_hdr);
674 return 0;
675
676write_error:
677 if (err != -EIO || !ubi->bad_allowed) {
678 ubi_ro_mode(ubi);
679 leb_write_unlock(ubi, vol_id, lnum);
680 ubi_free_vid_hdr(ubi, vid_hdr);
681 return err;
682 }
683
684 /*
685 * Fortunately, this is the first write operation to this physical
686 * eraseblock, so just put it and request a new one. We assume that if
687 * this physical eraseblock went bad, the erase code will handle that.
688 */
689 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
690 if (err || ++tries > UBI_IO_RETRIES) {
691 ubi_ro_mode(ubi);
692 leb_write_unlock(ubi, vol_id, lnum);
693 ubi_free_vid_hdr(ubi, vid_hdr);
694 return err;
695 }
696
697 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
698 ubi_msg("try another PEB");
699 goto retry;
700}
701
702/**
703 * ubi_eba_write_leb_st - write data to static volume.
704 * @ubi: UBI device description object
705 * @vol: volume description object
706 * @lnum: logical eraseblock number
707 * @buf: data to write
708 * @len: how many bytes to write
709 * @used_ebs: how many logical eraseblocks will this volume contain
710 *
711 * This function writes data to logical eraseblock @lnum of static volume
712 * @vol. The @used_ebs argument should contain total number of logical
713 * eraseblock in this static volume.
714 *
715 * When writing to the last logical eraseblock, the @len argument doesn't have
716 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
717 * to the real data size, although the @buf buffer has to contain the
718 * alignment. In all other cases, @len has to be aligned.
719 *
720 * It is prohibited to write more than once to logical eraseblocks of static
721 * volumes. This function returns zero in case of success and a negative error
722 * code in case of failure.
723 */
724int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
725 int lnum, const void *buf, int len, int used_ebs)
726{
727 int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
728 struct ubi_vid_hdr *vid_hdr;
729 uint32_t crc;
730
731 if (ubi->ro_mode)
732 return -EROFS;
733
734 if (lnum == used_ebs - 1)
735 /* If this is the last LEB @len may be unaligned */
736 len = ALIGN(data_size, ubi->min_io_size);
737 else
738 ubi_assert(!(len & (ubi->min_io_size - 1)));
739
740 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
741 if (!vid_hdr)
742 return -ENOMEM;
743
744 err = leb_write_lock(ubi, vol_id, lnum);
745 if (err) {
746 ubi_free_vid_hdr(ubi, vid_hdr);
747 return err;
748 }
749
750 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
751 vid_hdr->vol_id = cpu_to_be32(vol_id);
752 vid_hdr->lnum = cpu_to_be32(lnum);
753 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
754 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
755
756 crc = crc32(UBI_CRC32_INIT, buf, data_size);
757 vid_hdr->vol_type = UBI_VID_STATIC;
758 vid_hdr->data_size = cpu_to_be32(data_size);
759 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
760 vid_hdr->data_crc = cpu_to_be32(crc);
761
762retry:
763 pnum = ubi_wl_get_peb(ubi);
764 if (pnum < 0) {
765 ubi_free_vid_hdr(ubi, vid_hdr);
766 leb_write_unlock(ubi, vol_id, lnum);
767 return pnum;
768 }
769
770 dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
771 len, vol_id, lnum, pnum, used_ebs);
772
773 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
774 if (err) {
775 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
776 vol_id, lnum, pnum);
777 goto write_error;
778 }
779
780 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
781 if (err) {
782 ubi_warn("failed to write %d bytes of data to PEB %d",
783 len, pnum);
784 goto write_error;
785 }
786
787 ubi_assert(vol->eba_tbl[lnum] < 0);
788 vol->eba_tbl[lnum] = pnum;
789
790 leb_write_unlock(ubi, vol_id, lnum);
791 ubi_free_vid_hdr(ubi, vid_hdr);
792 return 0;
793
794write_error:
795 if (err != -EIO || !ubi->bad_allowed) {
796 /*
797 * This flash device does not admit of bad eraseblocks or
798 * something nasty and unexpected happened. Switch to read-only
799 * mode just in case.
800 */
801 ubi_ro_mode(ubi);
802 leb_write_unlock(ubi, vol_id, lnum);
803 ubi_free_vid_hdr(ubi, vid_hdr);
804 return err;
805 }
806
807 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
808 if (err || ++tries > UBI_IO_RETRIES) {
809 ubi_ro_mode(ubi);
810 leb_write_unlock(ubi, vol_id, lnum);
811 ubi_free_vid_hdr(ubi, vid_hdr);
812 return err;
813 }
814
815 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
816 ubi_msg("try another PEB");
817 goto retry;
818}
819
820/*
821 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
822 * @ubi: UBI device description object
823 * @vol: volume description object
824 * @lnum: logical eraseblock number
825 * @buf: data to write
826 * @len: how many bytes to write
827 *
828 * This function changes the contents of a logical eraseblock atomically. @buf
829 * has to contain new logical eraseblock data, and @len - the length of the
830 * data, which has to be aligned. This function guarantees that in case of an
831 * unclean reboot the old contents is preserved. Returns zero in case of
832 * success and a negative error code in case of failure.
833 *
834 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
835 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
836 */
837int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
838 int lnum, const void *buf, int len)
839{
840 int err, pnum, tries = 0, vol_id = vol->vol_id;
841 struct ubi_vid_hdr *vid_hdr;
842 uint32_t crc;
843
844 if (ubi->ro_mode)
845 return -EROFS;
846
847 if (len == 0) {
848 /*
849 * Special case when data length is zero. In this case the LEB
850 * has to be unmapped and mapped somewhere else.
851 */
852 err = ubi_eba_unmap_leb(ubi, vol, lnum);
853 if (err)
854 return err;
855 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
856 }
857
858 vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
859 if (!vid_hdr)
860 return -ENOMEM;
861
862 mutex_lock(&ubi->alc_mutex);
863 err = leb_write_lock(ubi, vol_id, lnum);
864 if (err)
865 goto out_mutex;
866
867 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
868 vid_hdr->vol_id = cpu_to_be32(vol_id);
869 vid_hdr->lnum = cpu_to_be32(lnum);
870 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
871 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
872
873 crc = crc32(UBI_CRC32_INIT, buf, len);
874 vid_hdr->vol_type = UBI_VID_DYNAMIC;
875 vid_hdr->data_size = cpu_to_be32(len);
876 vid_hdr->copy_flag = 1;
877 vid_hdr->data_crc = cpu_to_be32(crc);
878
879retry:
880 pnum = ubi_wl_get_peb(ubi);
881 if (pnum < 0) {
882 err = pnum;
883 goto out_leb_unlock;
884 }
885
886 dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
887 vol_id, lnum, vol->eba_tbl[lnum], pnum);
888
889 err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
890 if (err) {
891 ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
892 vol_id, lnum, pnum);
893 goto write_error;
894 }
895
896 err = ubi_io_write_data(ubi, buf, pnum, 0, len);
897 if (err) {
898 ubi_warn("failed to write %d bytes of data to PEB %d",
899 len, pnum);
900 goto write_error;
901 }
902
903 if (vol->eba_tbl[lnum] >= 0) {
904 err = ubi_wl_put_peb(ubi, vol_id, lnum, vol->eba_tbl[lnum], 0);
905 if (err)
906 goto out_leb_unlock;
907 }
908
909 vol->eba_tbl[lnum] = pnum;
910
911out_leb_unlock:
912 leb_write_unlock(ubi, vol_id, lnum);
913out_mutex:
914 mutex_unlock(&ubi->alc_mutex);
915 ubi_free_vid_hdr(ubi, vid_hdr);
916 return err;
917
918write_error:
919 if (err != -EIO || !ubi->bad_allowed) {
920 /*
921 * This flash device does not admit of bad eraseblocks or
922 * something nasty and unexpected happened. Switch to read-only
923 * mode just in case.
924 */
925 ubi_ro_mode(ubi);
926 goto out_leb_unlock;
927 }
928
929 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
930 if (err || ++tries > UBI_IO_RETRIES) {
931 ubi_ro_mode(ubi);
932 goto out_leb_unlock;
933 }
934
935 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
936 ubi_msg("try another PEB");
937 goto retry;
938}
939
940/**
941 * is_error_sane - check whether a read error is sane.
942 * @err: code of the error happened during reading
943 *
944 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
945 * cannot read data from the target PEB (an error @err happened). If the error
946 * code is sane, then we treat this error as non-fatal. Otherwise the error is
947 * fatal and UBI will be switched to R/O mode later.
948 *
949 * The idea is that we try not to switch to R/O mode if the read error is
950 * something which suggests there was a real read problem. E.g., %-EIO. Or a
951 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
952 * mode, simply because we do not know what happened at the MTD level, and we
953 * cannot handle this. E.g., the underlying driver may have become crazy, and
954 * it is safer to switch to R/O mode to preserve the data.
955 *
956 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
957 * which we have just written.
958 */
959static int is_error_sane(int err)
960{
961 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
962 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
963 return 0;
964 return 1;
965}
966
967/**
968 * ubi_eba_copy_leb - copy logical eraseblock.
969 * @ubi: UBI device description object
970 * @from: physical eraseblock number from where to copy
971 * @to: physical eraseblock number where to copy
972 * @vid_hdr: VID header of the @from physical eraseblock
973 *
974 * This function copies logical eraseblock from physical eraseblock @from to
975 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
976 * function. Returns:
977 * o %0 in case of success;
978 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
979 * o a negative error code in case of failure.
980 */
981int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
982 struct ubi_vid_hdr *vid_hdr)
983{
984 int err, vol_id, lnum, data_size, aldata_size, idx;
985 struct ubi_volume *vol;
986 uint32_t crc;
987
988 vol_id = be32_to_cpu(vid_hdr->vol_id);
989 lnum = be32_to_cpu(vid_hdr->lnum);
990
991 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
992
993 if (vid_hdr->vol_type == UBI_VID_STATIC) {
994 data_size = be32_to_cpu(vid_hdr->data_size);
995 aldata_size = ALIGN(data_size, ubi->min_io_size);
996 } else
997 data_size = aldata_size =
998 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
999
1000 idx = vol_id2idx(ubi, vol_id);
1001 spin_lock(&ubi->volumes_lock);
1002 /*
1003 * Note, we may race with volume deletion, which means that the volume
1004 * this logical eraseblock belongs to might be being deleted. Since the
1005 * volume deletion un-maps all the volume's logical eraseblocks, it will
1006 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1007 */
1008 vol = ubi->volumes[idx];
1009 spin_unlock(&ubi->volumes_lock);
1010 if (!vol) {
1011 /* No need to do further work, cancel */
1012 dbg_wl("volume %d is being removed, cancel", vol_id);
1013 return MOVE_CANCEL_RACE;
1014 }
1015
1016 /*
1017 * We do not want anybody to write to this logical eraseblock while we
1018 * are moving it, so lock it.
1019 *
1020 * Note, we are using non-waiting locking here, because we cannot sleep
1021 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1022 * unmapping the LEB which is mapped to the PEB we are going to move
1023 * (@from). This task locks the LEB and goes sleep in the
1024 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1025 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1026 * LEB is already locked, we just do not move it and return
1027 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1028 * we do not know the reasons of the contention - it may be just a
1029 * normal I/O on this LEB, so we want to re-try.
1030 */
1031 err = leb_write_trylock(ubi, vol_id, lnum);
1032 if (err) {
1033 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1034 return MOVE_RETRY;
1035 }
1036
1037 /*
1038 * The LEB might have been put meanwhile, and the task which put it is
1039 * probably waiting on @ubi->move_mutex. No need to continue the work,
1040 * cancel it.
1041 */
1042 if (vol->eba_tbl[lnum] != from) {
1043 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to "
1044 "PEB %d, cancel", vol_id, lnum, from,
1045 vol->eba_tbl[lnum]);
1046 err = MOVE_CANCEL_RACE;
1047 goto out_unlock_leb;
1048 }
1049
1050 /*
1051 * OK, now the LEB is locked and we can safely start moving it. Since
1052 * this function utilizes the @ubi->peb_buf buffer which is shared
1053 * with some other functions - we lock the buffer by taking the
1054 * @ubi->buf_mutex.
1055 */
1056 mutex_lock(&ubi->buf_mutex);
1057 dbg_wl("read %d bytes of data", aldata_size);
1058 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1059 if (err && err != UBI_IO_BITFLIPS) {
1060 ubi_warn("error %d while reading data from PEB %d",
1061 err, from);
1062 err = MOVE_SOURCE_RD_ERR;
1063 goto out_unlock_buf;
1064 }
1065
1066 /*
1067 * Now we have got to calculate how much data we have to copy. In
1068 * case of a static volume it is fairly easy - the VID header contains
1069 * the data size. In case of a dynamic volume it is more difficult - we
1070 * have to read the contents, cut 0xFF bytes from the end and copy only
1071 * the first part. We must do this to avoid writing 0xFF bytes as it
1072 * may have some side-effects. And not only this. It is important not
1073 * to include those 0xFFs to CRC because later the they may be filled
1074 * by data.
1075 */
1076 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1077 aldata_size = data_size =
1078 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1079
1080 cond_resched();
1081 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1082 cond_resched();
1083
1084 /*
1085 * It may turn out to be that the whole @from physical eraseblock
1086 * contains only 0xFF bytes. Then we have to only write the VID header
1087 * and do not write any data. This also means we should not set
1088 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1089 */
1090 if (data_size > 0) {
1091 vid_hdr->copy_flag = 1;
1092 vid_hdr->data_size = cpu_to_be32(data_size);
1093 vid_hdr->data_crc = cpu_to_be32(crc);
1094 }
1095 vid_hdr->sqnum = cpu_to_be64(next_sqnum(ubi));
1096
1097 err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1098 if (err) {
1099 if (err == -EIO)
1100 err = MOVE_TARGET_WR_ERR;
1101 goto out_unlock_buf;
1102 }
1103
1104 cond_resched();
1105
1106 /* Read the VID header back and check if it was written correctly */
1107 err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1108 if (err) {
1109 if (err != UBI_IO_BITFLIPS) {
1110 ubi_warn("error %d while reading VID header back from "
1111 "PEB %d", err, to);
1112 if (is_error_sane(err))
1113 err = MOVE_TARGET_RD_ERR;
1114 } else
1115 err = MOVE_TARGET_BITFLIPS;
1116 goto out_unlock_buf;
1117 }
1118
1119 if (data_size > 0) {
1120 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1121 if (err) {
1122 if (err == -EIO)
1123 err = MOVE_TARGET_WR_ERR;
1124 goto out_unlock_buf;
1125 }
1126
1127 cond_resched();
1128
1129 /*
1130 * We've written the data and are going to read it back to make
1131 * sure it was written correctly.
1132 */
1133 memset(ubi->peb_buf, 0xFF, aldata_size);
1134 err = ubi_io_read_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1135 if (err) {
1136 if (err != UBI_IO_BITFLIPS) {
1137 ubi_warn("error %d while reading data back "
1138 "from PEB %d", err, to);
1139 if (is_error_sane(err))
1140 err = MOVE_TARGET_RD_ERR;
1141 } else
1142 err = MOVE_TARGET_BITFLIPS;
1143 goto out_unlock_buf;
1144 }
1145
1146 cond_resched();
1147
1148 if (crc != crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size)) {
1149 ubi_warn("read data back from PEB %d and it is "
1150 "different", to);
1151 err = -EINVAL;
1152 goto out_unlock_buf;
1153 }
1154 }
1155
1156 ubi_assert(vol->eba_tbl[lnum] == from);
1157 vol->eba_tbl[lnum] = to;
1158
1159out_unlock_buf:
1160 mutex_unlock(&ubi->buf_mutex);
1161out_unlock_leb:
1162 leb_write_unlock(ubi, vol_id, lnum);
1163 return err;
1164}
1165
1166/**
1167 * print_rsvd_warning - warn about not having enough reserved PEBs.
1168 * @ubi: UBI device description object
1169 *
1170 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1171 * cannot reserve enough PEBs for bad block handling. This function makes a
1172 * decision whether we have to print a warning or not. The algorithm is as
1173 * follows:
1174 * o if this is a new UBI image, then just print the warning
1175 * o if this is an UBI image which has already been used for some time, print
1176 * a warning only if we can reserve less than 10% of the expected amount of
1177 * the reserved PEB.
1178 *
1179 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1180 * of PEBs becomes smaller, which is normal and we do not want to scare users
1181 * with a warning every time they attach the MTD device. This was an issue
1182 * reported by real users.
1183 */
1184static void print_rsvd_warning(struct ubi_device *ubi,
1185 struct ubi_attach_info *ai)
1186{
1187 /*
1188 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1189 * large number to distinguish between newly flashed and used images.
1190 */
1191 if (ai->max_sqnum > (1 << 18)) {
1192 int min = ubi->beb_rsvd_level / 10;
1193
1194 if (!min)
1195 min = 1;
1196 if (ubi->beb_rsvd_pebs > min)
1197 return;
1198 }
1199
1200 ubi_warn("cannot reserve enough PEBs for bad PEB handling, reserved %d,"
1201 " need %d", ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1202 if (ubi->corr_peb_count)
1203 ubi_warn("%d PEBs are corrupted and not used",
1204 ubi->corr_peb_count);
1205}
1206
1207/**
1208 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1209 * @ubi: UBI device description object
1210 * @ai: attaching information
1211 *
1212 * This function returns zero in case of success and a negative error code in
1213 * case of failure.
1214 */
1215int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1216{
1217 int i, j, err, num_volumes;
1218 struct ubi_ainf_volume *av;
1219 struct ubi_volume *vol;
1220 struct ubi_ainf_peb *aeb;
1221 struct rb_node *rb;
1222
1223 dbg_eba("initialize EBA sub-system");
1224
1225 spin_lock_init(&ubi->ltree_lock);
1226 mutex_init(&ubi->alc_mutex);
1227 ubi->ltree = RB_ROOT;
1228
1229 ubi->global_sqnum = ai->max_sqnum + 1;
1230 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1231
1232 for (i = 0; i < num_volumes; i++) {
1233 vol = ubi->volumes[i];
1234 if (!vol)
1235 continue;
1236
1237 cond_resched();
1238
1239 vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1240 GFP_KERNEL);
1241 if (!vol->eba_tbl) {
1242 err = -ENOMEM;
1243 goto out_free;
1244 }
1245
1246 for (j = 0; j < vol->reserved_pebs; j++)
1247 vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1248
1249 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1250 if (!av)
1251 continue;
1252
1253 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1254 if (aeb->lnum >= vol->reserved_pebs)
1255 /*
1256 * This may happen in case of an unclean reboot
1257 * during re-size.
1258 */
1259 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1260 vol->eba_tbl[aeb->lnum] = aeb->pnum;
1261 }
1262 }
1263
1264 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1265 ubi_err("no enough physical eraseblocks (%d, need %d)",
1266 ubi->avail_pebs, EBA_RESERVED_PEBS);
1267 if (ubi->corr_peb_count)
1268 ubi_err("%d PEBs are corrupted and not used",
1269 ubi->corr_peb_count);
1270 err = -ENOSPC;
1271 goto out_free;
1272 }
1273 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1274 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1275
1276 if (ubi->bad_allowed) {
1277 ubi_calculate_reserved(ubi);
1278
1279 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1280 /* No enough free physical eraseblocks */
1281 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1282 print_rsvd_warning(ubi, ai);
1283 } else
1284 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1285
1286 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1287 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1288 }
1289
1290 dbg_eba("EBA sub-system is initialized");
1291 return 0;
1292
1293out_free:
1294 for (i = 0; i < num_volumes; i++) {
1295 if (!ubi->volumes[i])
1296 continue;
1297 kfree(ubi->volumes[i]->eba_tbl);
1298 ubi->volumes[i]->eba_tbl = NULL;
1299 }
1300 return err;
1301}