projects/linux·blob master · refs

1709 lines · 46.1 KiB

1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (c) International Business Machines Corp., 2006
4 *
5 * Author: Artem Bityutskiy (Битюцкий Артём)
6 */
7
8/*
9 * The UBI Eraseblock Association (EBA) sub-system.
10 *
11 * This sub-system is responsible for I/O to/from logical eraseblock.
12 *
13 * Although in this implementation the EBA table is fully kept and managed in
14 * RAM, which assumes poor scalability, it might be (partially) maintained on
15 * flash in future implementations.
16 *
17 * The EBA sub-system implements per-logical eraseblock locking. Before
18 * accessing a logical eraseblock it is locked for reading or writing. The
19 * per-logical eraseblock locking is implemented by means of the lock tree. The
20 * lock tree is an RB-tree which refers all the currently locked logical
21 * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
22 * They are indexed by (@vol_id, @lnum) pairs.
23 *
24 * EBA also maintains the global sequence counter which is incremented each
25 * time a logical eraseblock is mapped to a physical eraseblock and it is
26 * stored in the volume identifier header. This means that each VID header has
27 * a unique sequence number. The sequence number is only increased an we assume
28 * 64 bits is enough to never overflow.
29 */
30
31#include <linux/slab.h>
32#include <linux/crc32.h>
33#include <linux/err.h>
34#include "ubi.h"
35
36/**
37 * struct ubi_eba_entry - structure encoding a single LEB -> PEB association
38 * @pnum: the physical eraseblock number attached to the LEB
39 *
40 * This structure is encoding a LEB -> PEB association. Note that the LEB
41 * number is not stored here, because it is the index used to access the
42 * entries table.
43 */
44struct ubi_eba_entry {
45 int pnum;
46};
47
48/**
49 * struct ubi_eba_table - LEB -> PEB association information
50 * @entries: the LEB to PEB mapping (one entry per LEB).
51 *
52 * This structure is private to the EBA logic and should be kept here.
53 * It is encoding the LEB to PEB association table, and is subject to
54 * changes.
55 */
56struct ubi_eba_table {
57 struct ubi_eba_entry *entries;
58};
59
60/**
61 * ubi_next_sqnum - get next sequence number.
62 * @ubi: UBI device description object
63 *
64 * This function returns next sequence number to use, which is just the current
65 * global sequence counter value. It also increases the global sequence
66 * counter.
67 */
68unsigned long long ubi_next_sqnum(struct ubi_device *ubi)
69{
70 unsigned long long sqnum;
71
72 spin_lock(&ubi->ltree_lock);
73 sqnum = ubi->global_sqnum++;
74 spin_unlock(&ubi->ltree_lock);
75
76 return sqnum;
77}
78
79/**
80 * ubi_get_compat - get compatibility flags of a volume.
81 * @ubi: UBI device description object
82 * @vol_id: volume ID
83 *
84 * This function returns compatibility flags for an internal volume. User
85 * volumes have no compatibility flags, so %0 is returned.
86 */
87static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
88{
89 if (vol_id == UBI_LAYOUT_VOLUME_ID)
90 return UBI_LAYOUT_VOLUME_COMPAT;
91 return 0;
92}
93
94/**
95 * ubi_eba_get_ldesc - get information about a LEB
96 * @vol: volume description object
97 * @lnum: logical eraseblock number
98 * @ldesc: the LEB descriptor to fill
99 *
100 * Used to query information about a specific LEB.
101 * It is currently only returning the physical position of the LEB, but will be
102 * extended to provide more information.
103 */
104void ubi_eba_get_ldesc(struct ubi_volume *vol, int lnum,
105 struct ubi_eba_leb_desc *ldesc)
106{
107 ldesc->lnum = lnum;
108 ldesc->pnum = vol->eba_tbl->entries[lnum].pnum;
109}
110
111/**
112 * ubi_eba_create_table - allocate a new EBA table and initialize it with all
113 * LEBs unmapped
114 * @vol: volume containing the EBA table to copy
115 * @nentries: number of entries in the table
116 *
117 * Allocate a new EBA table and initialize it with all LEBs unmapped.
118 * Returns a valid pointer if it succeed, an ERR_PTR() otherwise.
119 */
120struct ubi_eba_table *ubi_eba_create_table(struct ubi_volume *vol,
121 int nentries)
122{
123 struct ubi_eba_table *tbl;
124 int err = -ENOMEM;
125 int i;
126
127 tbl = kzalloc_obj(*tbl);
128 if (!tbl)
129 return ERR_PTR(-ENOMEM);
130
131 tbl->entries = kmalloc_objs(*tbl->entries, nentries);
132 if (!tbl->entries)
133 goto err;
134
135 for (i = 0; i < nentries; i++)
136 tbl->entries[i].pnum = UBI_LEB_UNMAPPED;
137
138 return tbl;
139
140err:
141 kfree(tbl);
142
143 return ERR_PTR(err);
144}
145
146/**
147 * ubi_eba_destroy_table - destroy an EBA table
148 * @tbl: the table to destroy
149 *
150 * Destroy an EBA table.
151 */
152void ubi_eba_destroy_table(struct ubi_eba_table *tbl)
153{
154 if (!tbl)
155 return;
156
157 kfree(tbl->entries);
158 kfree(tbl);
159}
160
161/**
162 * ubi_eba_copy_table - copy the EBA table attached to vol into another table
163 * @vol: volume containing the EBA table to copy
164 * @dst: destination
165 * @nentries: number of entries to copy
166 *
167 * Copy the EBA table stored in vol into the one pointed by dst.
168 */
169void ubi_eba_copy_table(struct ubi_volume *vol, struct ubi_eba_table *dst,
170 int nentries)
171{
172 struct ubi_eba_table *src;
173 int i;
174
175 ubi_assert(dst && vol && vol->eba_tbl);
176
177 src = vol->eba_tbl;
178
179 for (i = 0; i < nentries; i++)
180 dst->entries[i].pnum = src->entries[i].pnum;
181}
182
183/**
184 * ubi_eba_replace_table - assign a new EBA table to a volume
185 * @vol: volume containing the EBA table to copy
186 * @tbl: new EBA table
187 *
188 * Assign a new EBA table to the volume and release the old one.
189 */
190void ubi_eba_replace_table(struct ubi_volume *vol, struct ubi_eba_table *tbl)
191{
192 ubi_eba_destroy_table(vol->eba_tbl);
193 vol->eba_tbl = tbl;
194}
195
196/**
197 * ltree_lookup - look up the lock tree.
198 * @ubi: UBI device description object
199 * @vol_id: volume ID
200 * @lnum: logical eraseblock number
201 *
202 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
203 * object if the logical eraseblock is locked and %NULL if it is not.
204 * @ubi->ltree_lock has to be locked.
205 */
206static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
207 int lnum)
208{
209 struct rb_node *p;
210
211 p = ubi->ltree.rb_node;
212 while (p) {
213 struct ubi_ltree_entry *le;
214
215 le = rb_entry(p, struct ubi_ltree_entry, rb);
216
217 if (vol_id < le->vol_id)
218 p = p->rb_left;
219 else if (vol_id > le->vol_id)
220 p = p->rb_right;
221 else {
222 if (lnum < le->lnum)
223 p = p->rb_left;
224 else if (lnum > le->lnum)
225 p = p->rb_right;
226 else
227 return le;
228 }
229 }
230
231 return NULL;
232}
233
234/**
235 * ltree_add_entry - add new entry to the lock tree.
236 * @ubi: UBI device description object
237 * @vol_id: volume ID
238 * @lnum: logical eraseblock number
239 *
240 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
241 * lock tree. If such entry is already there, its usage counter is increased.
242 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
243 * failed.
244 */
245static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
246 int vol_id, int lnum)
247{
248 struct ubi_ltree_entry *le, *le1, *le_free;
249
250 le = kmalloc_obj(struct ubi_ltree_entry, GFP_NOFS);
251 if (!le)
252 return ERR_PTR(-ENOMEM);
253
254 le->users = 0;
255 init_rwsem(&le->mutex);
256 le->vol_id = vol_id;
257 le->lnum = lnum;
258
259 spin_lock(&ubi->ltree_lock);
260 le1 = ltree_lookup(ubi, vol_id, lnum);
261
262 if (le1) {
263 /*
264 * This logical eraseblock is already locked. The newly
265 * allocated lock entry is not needed.
266 */
267 le_free = le;
268 le = le1;
269 } else {
270 struct rb_node **p, *parent = NULL;
271
272 /*
273 * No lock entry, add the newly allocated one to the
274 * @ubi->ltree RB-tree.
275 */
276 le_free = NULL;
277
278 p = &ubi->ltree.rb_node;
279 while (*p) {
280 parent = *p;
281 le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
282
283 if (vol_id < le1->vol_id)
284 p = &(*p)->rb_left;
285 else if (vol_id > le1->vol_id)
286 p = &(*p)->rb_right;
287 else {
288 ubi_assert(lnum != le1->lnum);
289 if (lnum < le1->lnum)
290 p = &(*p)->rb_left;
291 else
292 p = &(*p)->rb_right;
293 }
294 }
295
296 rb_link_node(&le->rb, parent, p);
297 rb_insert_color(&le->rb, &ubi->ltree);
298 }
299 le->users += 1;
300 spin_unlock(&ubi->ltree_lock);
301
302 kfree(le_free);
303 return le;
304}
305
306/**
307 * leb_read_lock - lock logical eraseblock for reading.
308 * @ubi: UBI device description object
309 * @vol_id: volume ID
310 * @lnum: logical eraseblock number
311 *
312 * This function locks a logical eraseblock for reading. Returns zero in case
313 * of success and a negative error code in case of failure.
314 */
315static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
316{
317 struct ubi_ltree_entry *le;
318
319 le = ltree_add_entry(ubi, vol_id, lnum);
320 if (IS_ERR(le))
321 return PTR_ERR(le);
322 down_read(&le->mutex);
323 return 0;
324}
325
326/**
327 * leb_read_unlock - unlock logical eraseblock.
328 * @ubi: UBI device description object
329 * @vol_id: volume ID
330 * @lnum: logical eraseblock number
331 */
332static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
333{
334 struct ubi_ltree_entry *le;
335
336 spin_lock(&ubi->ltree_lock);
337 le = ltree_lookup(ubi, vol_id, lnum);
338 le->users -= 1;
339 ubi_assert(le->users >= 0);
340 up_read(&le->mutex);
341 if (le->users == 0) {
342 rb_erase(&le->rb, &ubi->ltree);
343 kfree(le);
344 }
345 spin_unlock(&ubi->ltree_lock);
346}
347
348/**
349 * leb_write_lock - lock logical eraseblock for writing.
350 * @ubi: UBI device description object
351 * @vol_id: volume ID
352 * @lnum: logical eraseblock number
353 *
354 * This function locks a logical eraseblock for writing. Returns zero in case
355 * of success and a negative error code in case of failure.
356 */
357static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
358{
359 struct ubi_ltree_entry *le;
360
361 le = ltree_add_entry(ubi, vol_id, lnum);
362 if (IS_ERR(le))
363 return PTR_ERR(le);
364 down_write(&le->mutex);
365 return 0;
366}
367
368/**
369 * leb_write_trylock - try to lock logical eraseblock for writing.
370 * @ubi: UBI device description object
371 * @vol_id: volume ID
372 * @lnum: logical eraseblock number
373 *
374 * This function locks a logical eraseblock for writing if there is no
375 * contention and does nothing if there is contention. Returns %0 in case of
376 * success, %1 in case of contention, and a negative error code in case of
377 * failure.
378 */
379static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
380{
381 struct ubi_ltree_entry *le;
382
383 le = ltree_add_entry(ubi, vol_id, lnum);
384 if (IS_ERR(le))
385 return PTR_ERR(le);
386 if (down_write_trylock(&le->mutex))
387 return 0;
388
389 /* Contention, cancel */
390 spin_lock(&ubi->ltree_lock);
391 le->users -= 1;
392 ubi_assert(le->users >= 0);
393 if (le->users == 0) {
394 rb_erase(&le->rb, &ubi->ltree);
395 kfree(le);
396 }
397 spin_unlock(&ubi->ltree_lock);
398
399 return 1;
400}
401
402/**
403 * leb_write_unlock - unlock logical eraseblock.
404 * @ubi: UBI device description object
405 * @vol_id: volume ID
406 * @lnum: logical eraseblock number
407 */
408static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
409{
410 struct ubi_ltree_entry *le;
411
412 spin_lock(&ubi->ltree_lock);
413 le = ltree_lookup(ubi, vol_id, lnum);
414 le->users -= 1;
415 ubi_assert(le->users >= 0);
416 up_write(&le->mutex);
417 if (le->users == 0) {
418 rb_erase(&le->rb, &ubi->ltree);
419 kfree(le);
420 }
421 spin_unlock(&ubi->ltree_lock);
422}
423
424/**
425 * ubi_eba_is_mapped - check if a LEB is mapped.
426 * @vol: volume description object
427 * @lnum: logical eraseblock number
428 *
429 * This function returns true if the LEB is mapped, false otherwise.
430 */
431bool ubi_eba_is_mapped(struct ubi_volume *vol, int lnum)
432{
433 return vol->eba_tbl->entries[lnum].pnum >= 0;
434}
435
436/**
437 * ubi_eba_unmap_leb - un-map logical eraseblock.
438 * @ubi: UBI device description object
439 * @vol: volume description object
440 * @lnum: logical eraseblock number
441 *
442 * This function un-maps logical eraseblock @lnum and schedules corresponding
443 * physical eraseblock for erasure. Returns zero in case of success and a
444 * negative error code in case of failure.
445 */
446int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
447 int lnum)
448{
449 int err, pnum, vol_id = vol->vol_id;
450
451 if (ubi->ro_mode)
452 return -EROFS;
453
454 err = leb_write_lock(ubi, vol_id, lnum);
455 if (err)
456 return err;
457
458 pnum = vol->eba_tbl->entries[lnum].pnum;
459 if (pnum < 0)
460 /* This logical eraseblock is already unmapped */
461 goto out_unlock;
462
463 dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
464
465 down_read(&ubi->fm_eba_sem);
466 vol->eba_tbl->entries[lnum].pnum = UBI_LEB_UNMAPPED;
467 up_read(&ubi->fm_eba_sem);
468 err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
469
470out_unlock:
471 leb_write_unlock(ubi, vol_id, lnum);
472 return err;
473}
474
475#ifdef CONFIG_MTD_UBI_FASTMAP
476/**
477 * check_mapping - check and fixup a mapping
478 * @ubi: UBI device description object
479 * @vol: volume description object
480 * @lnum: logical eraseblock number
481 * @pnum: physical eraseblock number
482 *
483 * Checks whether a given mapping is valid. Fastmap cannot track LEB unmap
484 * operations, if such an operation is interrupted the mapping still looks
485 * good, but upon first read an ECC is reported to the upper layer.
486 * Normaly during the full-scan at attach time this is fixed, for Fastmap
487 * we have to deal with it while reading.
488 * If the PEB behind a LEB shows this symthom we change the mapping to
489 * %UBI_LEB_UNMAPPED and schedule the PEB for erasure.
490 *
491 * Returns 0 on success, negative error code in case of failure.
492 */
493static int check_mapping(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
494 int *pnum)
495{
496 int err;
497 struct ubi_vid_io_buf *vidb;
498 struct ubi_vid_hdr *vid_hdr;
499
500 if (!ubi->fast_attach)
501 return 0;
502
503 if (!vol->checkmap || test_bit(lnum, vol->checkmap))
504 return 0;
505
506 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
507 if (!vidb)
508 return -ENOMEM;
509
510 err = ubi_io_read_vid_hdr(ubi, *pnum, vidb, 0);
511 if (err > 0 && err != UBI_IO_BITFLIPS) {
512 int torture = 0;
513
514 switch (err) {
515 case UBI_IO_FF:
516 case UBI_IO_FF_BITFLIPS:
517 case UBI_IO_BAD_HDR:
518 case UBI_IO_BAD_HDR_EBADMSG:
519 break;
520 default:
521 ubi_assert(0);
522 }
523
524 if (err == UBI_IO_BAD_HDR_EBADMSG || err == UBI_IO_FF_BITFLIPS)
525 torture = 1;
526
527 down_read(&ubi->fm_eba_sem);
528 vol->eba_tbl->entries[lnum].pnum = UBI_LEB_UNMAPPED;
529 up_read(&ubi->fm_eba_sem);
530 ubi_wl_put_peb(ubi, vol->vol_id, lnum, *pnum, torture);
531
532 *pnum = UBI_LEB_UNMAPPED;
533 } else if (err < 0) {
534 ubi_err(ubi, "unable to read VID header back from PEB %i: %i",
535 *pnum, err);
536
537 goto out_free;
538 } else {
539 int found_vol_id, found_lnum;
540
541 ubi_assert(err == 0 || err == UBI_IO_BITFLIPS);
542
543 vid_hdr = ubi_get_vid_hdr(vidb);
544 found_vol_id = be32_to_cpu(vid_hdr->vol_id);
545 found_lnum = be32_to_cpu(vid_hdr->lnum);
546
547 if (found_lnum != lnum || found_vol_id != vol->vol_id) {
548 ubi_err(ubi, "EBA mismatch! PEB %i is LEB %i:%i instead of LEB %i:%i",
549 *pnum, found_vol_id, found_lnum, vol->vol_id, lnum);
550 ubi_ro_mode(ubi);
551 err = -EINVAL;
552 goto out_free;
553 }
554 }
555
556 set_bit(lnum, vol->checkmap);
557 err = 0;
558
559out_free:
560 ubi_free_vid_buf(vidb);
561
562 return err;
563}
564#else
565static int check_mapping(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
566 int *pnum)
567{
568 return 0;
569}
570#endif
571
572/**
573 * ubi_eba_read_leb - read data.
574 * @ubi: UBI device description object
575 * @vol: volume description object
576 * @lnum: logical eraseblock number
577 * @buf: buffer to store the read data
578 * @offset: offset from where to read
579 * @len: how many bytes to read
580 * @check: data CRC check flag
581 *
582 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
583 * bytes. The @check flag only makes sense for static volumes and forces
584 * eraseblock data CRC checking.
585 *
586 * In case of success this function returns zero. In case of a static volume,
587 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
588 * returned for any volume type if an ECC error was detected by the MTD device
589 * driver. Other negative error cored may be returned in case of other errors.
590 */
591int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
592 void *buf, int offset, int len, int check)
593{
594 int err, pnum, scrub = 0, vol_id = vol->vol_id;
595 struct ubi_vid_io_buf *vidb;
596 struct ubi_vid_hdr *vid_hdr;
597 uint32_t crc;
598
599 err = leb_read_lock(ubi, vol_id, lnum);
600 if (err)
601 return err;
602
603 pnum = vol->eba_tbl->entries[lnum].pnum;
604 if (pnum >= 0) {
605 err = check_mapping(ubi, vol, lnum, &pnum);
606 if (err < 0)
607 goto out_unlock;
608 }
609
610 if (pnum == UBI_LEB_UNMAPPED) {
611 /*
612 * The logical eraseblock is not mapped, fill the whole buffer
613 * with 0xFF bytes. The exception is static volumes for which
614 * it is an error to read unmapped logical eraseblocks.
615 */
616 dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
617 len, offset, vol_id, lnum);
618 leb_read_unlock(ubi, vol_id, lnum);
619 ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
620 memset(buf, 0xFF, len);
621 return 0;
622 }
623
624 dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
625 len, offset, vol_id, lnum, pnum);
626
627 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
628 check = 0;
629
630retry:
631 if (check) {
632 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
633 if (!vidb) {
634 err = -ENOMEM;
635 goto out_unlock;
636 }
637
638 vid_hdr = ubi_get_vid_hdr(vidb);
639
640 err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
641 if (err && err != UBI_IO_BITFLIPS) {
642 if (err > 0) {
643 /*
644 * The header is either absent or corrupted.
645 * The former case means there is a bug -
646 * switch to read-only mode just in case.
647 * The latter case means a real corruption - we
648 * may try to recover data. FIXME: but this is
649 * not implemented.
650 */
651 if (err == UBI_IO_BAD_HDR_EBADMSG ||
652 err == UBI_IO_BAD_HDR) {
653 ubi_warn(ubi, "corrupted VID header at PEB %d, LEB %d:%d",
654 pnum, vol_id, lnum);
655 err = -EBADMSG;
656 } else {
657 /*
658 * Ending up here in the non-Fastmap case
659 * is a clear bug as the VID header had to
660 * be present at scan time to have it referenced.
661 * With fastmap the story is more complicated.
662 * Fastmap has the mapping info without the need
663 * of a full scan. So the LEB could have been
664 * unmapped, Fastmap cannot know this and keeps
665 * the LEB referenced.
666 * This is valid and works as the layer above UBI
667 * has to do bookkeeping about used/referenced
668 * LEBs in any case.
669 */
670 if (ubi->fast_attach) {
671 err = -EBADMSG;
672 } else {
673 err = -EINVAL;
674 ubi_ro_mode(ubi);
675 }
676 }
677 }
678 goto out_free;
679 } else if (err == UBI_IO_BITFLIPS)
680 scrub = 1;
681
682 ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
683 ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
684
685 crc = be32_to_cpu(vid_hdr->data_crc);
686 ubi_free_vid_buf(vidb);
687 }
688
689 err = ubi_io_read_data(ubi, buf, pnum, offset, len);
690 if (err) {
691 if (err == UBI_IO_BITFLIPS)
692 scrub = 1;
693 else if (mtd_is_eccerr(err)) {
694 if (vol->vol_type == UBI_DYNAMIC_VOLUME)
695 goto out_unlock;
696 scrub = 1;
697 if (!check) {
698 ubi_msg(ubi, "force data checking");
699 check = 1;
700 goto retry;
701 }
702 } else
703 goto out_unlock;
704 }
705
706 if (check) {
707 uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
708 if (crc1 != crc) {
709 ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
710 crc1, crc);
711 err = -EBADMSG;
712 goto out_unlock;
713 }
714 }
715
716 if (scrub)
717 err = ubi_wl_scrub_peb(ubi, pnum);
718
719 leb_read_unlock(ubi, vol_id, lnum);
720 return err;
721
722out_free:
723 ubi_free_vid_buf(vidb);
724out_unlock:
725 leb_read_unlock(ubi, vol_id, lnum);
726 return err;
727}
728
729/**
730 * ubi_eba_read_leb_sg - read data into a scatter gather list.
731 * @ubi: UBI device description object
732 * @vol: volume description object
733 * @lnum: logical eraseblock number
734 * @sgl: UBI scatter gather list to store the read data
735 * @offset: offset from where to read
736 * @len: how many bytes to read
737 * @check: data CRC check flag
738 *
739 * This function works exactly like ubi_eba_read_leb(). But instead of
740 * storing the read data into a buffer it writes to an UBI scatter gather
741 * list.
742 */
743int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
744 struct ubi_sgl *sgl, int lnum, int offset, int len,
745 int check)
746{
747 int to_read;
748 int ret;
749 struct scatterlist *sg;
750
751 for (;;) {
752 ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
753 sg = &sgl->sg[sgl->list_pos];
754 if (len < sg->length - sgl->page_pos)
755 to_read = len;
756 else
757 to_read = sg->length - sgl->page_pos;
758
759 ret = ubi_eba_read_leb(ubi, vol, lnum,
760 sg_virt(sg) + sgl->page_pos, offset,
761 to_read, check);
762 if (ret < 0)
763 return ret;
764
765 offset += to_read;
766 len -= to_read;
767 if (!len) {
768 sgl->page_pos += to_read;
769 if (sgl->page_pos == sg->length) {
770 sgl->list_pos++;
771 sgl->page_pos = 0;
772 }
773
774 break;
775 }
776
777 sgl->list_pos++;
778 sgl->page_pos = 0;
779 }
780
781 return ret;
782}
783
784/**
785 * try_recover_peb - try to recover from write failure.
786 * @vol: volume description object
787 * @pnum: the physical eraseblock to recover
788 * @lnum: logical eraseblock number
789 * @buf: data which was not written because of the write failure
790 * @offset: offset of the failed write
791 * @len: how many bytes should have been written
792 * @vidb: VID buffer
793 * @retry: whether the caller should retry in case of failure
794 *
795 * This function is called in case of a write failure and moves all good data
796 * from the potentially bad physical eraseblock to a good physical eraseblock.
797 * This function also writes the data which was not written due to the failure.
798 * Returns 0 in case of success, and a negative error code in case of failure.
799 * In case of failure, the %retry parameter is set to false if this is a fatal
800 * error (retrying won't help), and true otherwise.
801 */
802static int try_recover_peb(struct ubi_volume *vol, int pnum, int lnum,
803 const void *buf, int offset, int len,
804 struct ubi_vid_io_buf *vidb, bool *retry)
805{
806 struct ubi_device *ubi = vol->ubi;
807 struct ubi_vid_hdr *vid_hdr;
808 int new_pnum, err, vol_id = vol->vol_id, data_size;
809 uint32_t crc;
810
811 *retry = false;
812
813 new_pnum = ubi_wl_get_peb(ubi);
814 if (new_pnum < 0) {
815 err = new_pnum;
816 goto out_put;
817 }
818
819 ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
820 pnum, new_pnum);
821
822 err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
823 if (err && err != UBI_IO_BITFLIPS) {
824 if (err > 0)
825 err = -EIO;
826 goto out_put;
827 }
828
829 vid_hdr = ubi_get_vid_hdr(vidb);
830 ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
831
832 mutex_lock(&ubi->buf_mutex);
833 memset(ubi->peb_buf + offset, 0xFF, len);
834
835 /* Read everything before the area where the write failure happened */
836 if (offset > 0) {
837 err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
838 if (err && err != UBI_IO_BITFLIPS)
839 goto out_unlock;
840 }
841
842 *retry = true;
843
844 memcpy(ubi->peb_buf + offset, buf, len);
845
846 data_size = offset + len;
847 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
848 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
849 vid_hdr->copy_flag = 1;
850 vid_hdr->data_size = cpu_to_be32(data_size);
851 vid_hdr->data_crc = cpu_to_be32(crc);
852 err = ubi_io_write_vid_hdr(ubi, new_pnum, vidb);
853 if (err)
854 goto out_unlock;
855
856 err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
857
858out_unlock:
859 mutex_unlock(&ubi->buf_mutex);
860
861 if (!err)
862 vol->eba_tbl->entries[lnum].pnum = new_pnum;
863
864out_put:
865 up_read(&ubi->fm_eba_sem);
866
867 if (!err) {
868 ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
869 ubi_msg(ubi, "data was successfully recovered");
870 } else if (new_pnum >= 0) {
871 /*
872 * Bad luck? This physical eraseblock is bad too? Crud. Let's
873 * try to get another one.
874 */
875 ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
876 ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
877 }
878
879 return err;
880}
881
882/**
883 * recover_peb - recover from write failure.
884 * @ubi: UBI device description object
885 * @pnum: the physical eraseblock to recover
886 * @vol_id: volume ID
887 * @lnum: logical eraseblock number
888 * @buf: data which was not written because of the write failure
889 * @offset: offset of the failed write
890 * @len: how many bytes should have been written
891 *
892 * This function is called in case of a write failure and moves all good data
893 * from the potentially bad physical eraseblock to a good physical eraseblock.
894 * This function also writes the data which was not written due to the failure.
895 * Returns 0 in case of success, and a negative error code in case of failure.
896 * This function tries %UBI_IO_RETRIES before giving up.
897 */
898static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
899 const void *buf, int offset, int len)
900{
901 int err, idx = vol_id2idx(ubi, vol_id), tries;
902 struct ubi_volume *vol = ubi->volumes[idx];
903 struct ubi_vid_io_buf *vidb;
904
905 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
906 if (!vidb)
907 return -ENOMEM;
908
909 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
910 bool retry;
911
912 err = try_recover_peb(vol, pnum, lnum, buf, offset, len, vidb,
913 &retry);
914 if (!err || !retry)
915 break;
916
917 ubi_msg(ubi, "try again");
918 }
919
920 ubi_free_vid_buf(vidb);
921
922 return err;
923}
924
925/**
926 * try_write_vid_and_data - try to write VID header and data to a new PEB.
927 * @vol: volume description object
928 * @lnum: logical eraseblock number
929 * @vidb: the VID buffer to write
930 * @buf: buffer containing the data
931 * @offset: where to start writing data
932 * @len: how many bytes should be written
933 *
934 * This function tries to write VID header and data belonging to logical
935 * eraseblock @lnum of volume @vol to a new physical eraseblock. Returns zero
936 * in case of success and a negative error code in case of failure.
937 * In case of error, it is possible that something was still written to the
938 * flash media, but may be some garbage.
939 */
940static int try_write_vid_and_data(struct ubi_volume *vol, int lnum,
941 struct ubi_vid_io_buf *vidb, const void *buf,
942 int offset, int len)
943{
944 struct ubi_device *ubi = vol->ubi;
945 int pnum, opnum, err, err2, vol_id = vol->vol_id;
946
947 pnum = ubi_wl_get_peb(ubi);
948 if (pnum < 0) {
949 err = pnum;
950 goto out_put;
951 }
952
953 opnum = vol->eba_tbl->entries[lnum].pnum;
954
955 dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
956 len, offset, vol_id, lnum, pnum);
957
958 err = ubi_io_write_vid_hdr(ubi, pnum, vidb);
959 if (err) {
960 ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
961 vol_id, lnum, pnum);
962 goto out_put;
963 }
964
965 if (len) {
966 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
967 if (err) {
968 ubi_warn(ubi,
969 "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
970 len, offset, vol_id, lnum, pnum);
971 goto out_put;
972 }
973 }
974
975 vol->eba_tbl->entries[lnum].pnum = pnum;
976
977out_put:
978 up_read(&ubi->fm_eba_sem);
979
980 if (err && pnum >= 0) {
981 err2 = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
982 if (err2) {
983 ubi_warn(ubi, "failed to return physical eraseblock %d, error %d",
984 pnum, err2);
985 }
986 } else if (!err && opnum >= 0) {
987 err2 = ubi_wl_put_peb(ubi, vol_id, lnum, opnum, 0);
988 if (err2) {
989 ubi_warn(ubi, "failed to return physical eraseblock %d, error %d",
990 opnum, err2);
991 }
992 }
993
994 return err;
995}
996
997/**
998 * ubi_eba_write_leb - write data to dynamic volume.
999 * @ubi: UBI device description object
1000 * @vol: volume description object
1001 * @lnum: logical eraseblock number
1002 * @buf: the data to write
1003 * @offset: offset within the logical eraseblock where to write
1004 * @len: how many bytes to write
1005 *
1006 * This function writes data to logical eraseblock @lnum of a dynamic volume
1007 * @vol. Returns zero in case of success and a negative error code in case
1008 * of failure. In case of error, it is possible that something was still
1009 * written to the flash media, but may be some garbage.
1010 * This function retries %UBI_IO_RETRIES times before giving up.
1011 */
1012int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
1013 const void *buf, int offset, int len)
1014{
1015 int err, pnum, tries, vol_id = vol->vol_id;
1016 struct ubi_vid_io_buf *vidb;
1017 struct ubi_vid_hdr *vid_hdr;
1018
1019 if (ubi->ro_mode)
1020 return -EROFS;
1021
1022 err = leb_write_lock(ubi, vol_id, lnum);
1023 if (err)
1024 return err;
1025
1026 pnum = vol->eba_tbl->entries[lnum].pnum;
1027 if (pnum >= 0) {
1028 err = check_mapping(ubi, vol, lnum, &pnum);
1029 if (err < 0)
1030 goto out;
1031 }
1032
1033 if (pnum >= 0) {
1034 dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
1035 len, offset, vol_id, lnum, pnum);
1036
1037 err = ubi_io_write_data(ubi, buf, pnum, offset, len);
1038 if (err) {
1039 ubi_warn(ubi, "failed to write data to PEB %d", pnum);
1040 if (err == -EIO && ubi->bad_allowed)
1041 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
1042 offset, len);
1043 }
1044
1045 goto out;
1046 }
1047
1048 /*
1049 * The logical eraseblock is not mapped. We have to get a free physical
1050 * eraseblock and write the volume identifier header there first.
1051 */
1052 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1053 if (!vidb) {
1054 leb_write_unlock(ubi, vol_id, lnum);
1055 return -ENOMEM;
1056 }
1057
1058 vid_hdr = ubi_get_vid_hdr(vidb);
1059
1060 vid_hdr->vol_type = UBI_VID_DYNAMIC;
1061 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1062 vid_hdr->vol_id = cpu_to_be32(vol_id);
1063 vid_hdr->lnum = cpu_to_be32(lnum);
1064 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1065 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1066
1067 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1068 err = try_write_vid_and_data(vol, lnum, vidb, buf, offset, len);
1069 if (err != -EIO || !ubi->bad_allowed)
1070 break;
1071
1072 /*
1073 * Fortunately, this is the first write operation to this
1074 * physical eraseblock, so just put it and request a new one.
1075 * We assume that if this physical eraseblock went bad, the
1076 * erase code will handle that.
1077 */
1078 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1079 ubi_msg(ubi, "try another PEB");
1080 }
1081
1082 ubi_free_vid_buf(vidb);
1083
1084out:
1085 if (err)
1086 ubi_ro_mode(ubi);
1087
1088 leb_write_unlock(ubi, vol_id, lnum);
1089
1090 return err;
1091}
1092
1093/**
1094 * ubi_eba_write_leb_st - write data to static volume.
1095 * @ubi: UBI device description object
1096 * @vol: volume description object
1097 * @lnum: logical eraseblock number
1098 * @buf: data to write
1099 * @len: how many bytes to write
1100 * @used_ebs: how many logical eraseblocks will this volume contain
1101 *
1102 * This function writes data to logical eraseblock @lnum of static volume
1103 * @vol. The @used_ebs argument should contain total number of logical
1104 * eraseblock in this static volume.
1105 *
1106 * When writing to the last logical eraseblock, the @len argument doesn't have
1107 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
1108 * to the real data size, although the @buf buffer has to contain the
1109 * alignment. In all other cases, @len has to be aligned.
1110 *
1111 * It is prohibited to write more than once to logical eraseblocks of static
1112 * volumes. This function returns zero in case of success and a negative error
1113 * code in case of failure.
1114 */
1115int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
1116 int lnum, const void *buf, int len, int used_ebs)
1117{
1118 int err, tries, data_size = len, vol_id = vol->vol_id;
1119 struct ubi_vid_io_buf *vidb;
1120 struct ubi_vid_hdr *vid_hdr;
1121 uint32_t crc;
1122
1123 if (ubi->ro_mode)
1124 return -EROFS;
1125
1126 if (lnum == used_ebs - 1)
1127 /* If this is the last LEB @len may be unaligned */
1128 len = ALIGN(data_size, ubi->min_io_size);
1129 else
1130 ubi_assert(!(len & (ubi->min_io_size - 1)));
1131
1132 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1133 if (!vidb)
1134 return -ENOMEM;
1135
1136 vid_hdr = ubi_get_vid_hdr(vidb);
1137
1138 err = leb_write_lock(ubi, vol_id, lnum);
1139 if (err)
1140 goto out;
1141
1142 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1143 vid_hdr->vol_id = cpu_to_be32(vol_id);
1144 vid_hdr->lnum = cpu_to_be32(lnum);
1145 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1146 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1147
1148 crc = crc32(UBI_CRC32_INIT, buf, data_size);
1149 vid_hdr->vol_type = UBI_VID_STATIC;
1150 vid_hdr->data_size = cpu_to_be32(data_size);
1151 vid_hdr->used_ebs = cpu_to_be32(used_ebs);
1152 vid_hdr->data_crc = cpu_to_be32(crc);
1153
1154 ubi_assert(vol->eba_tbl->entries[lnum].pnum < 0);
1155
1156 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1157 err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1158 if (err != -EIO || !ubi->bad_allowed)
1159 break;
1160
1161 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1162 ubi_msg(ubi, "try another PEB");
1163 }
1164
1165 if (err)
1166 ubi_ro_mode(ubi);
1167
1168 leb_write_unlock(ubi, vol_id, lnum);
1169
1170out:
1171 ubi_free_vid_buf(vidb);
1172
1173 return err;
1174}
1175
1176/*
1177 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
1178 * @ubi: UBI device description object
1179 * @vol: volume description object
1180 * @lnum: logical eraseblock number
1181 * @buf: data to write
1182 * @len: how many bytes to write
1183 *
1184 * This function changes the contents of a logical eraseblock atomically. @buf
1185 * has to contain new logical eraseblock data, and @len - the length of the
1186 * data, which has to be aligned. This function guarantees that in case of an
1187 * unclean reboot the old contents is preserved. Returns zero in case of
1188 * success and a negative error code in case of failure.
1189 *
1190 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
1191 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
1192 */
1193int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
1194 int lnum, const void *buf, int len)
1195{
1196 int err, tries, vol_id = vol->vol_id;
1197 struct ubi_vid_io_buf *vidb;
1198 struct ubi_vid_hdr *vid_hdr;
1199 uint32_t crc;
1200
1201 if (ubi->ro_mode)
1202 return -EROFS;
1203
1204 if (len == 0) {
1205 /*
1206 * Special case when data length is zero. In this case the LEB
1207 * has to be unmapped and mapped somewhere else.
1208 */
1209 err = ubi_eba_unmap_leb(ubi, vol, lnum);
1210 if (err)
1211 return err;
1212 return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
1213 }
1214
1215 vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1216 if (!vidb)
1217 return -ENOMEM;
1218
1219 vid_hdr = ubi_get_vid_hdr(vidb);
1220
1221 mutex_lock(&ubi->alc_mutex);
1222 err = leb_write_lock(ubi, vol_id, lnum);
1223 if (err)
1224 goto out_mutex;
1225
1226 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1227 vid_hdr->vol_id = cpu_to_be32(vol_id);
1228 vid_hdr->lnum = cpu_to_be32(lnum);
1229 vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1230 vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1231
1232 crc = crc32(UBI_CRC32_INIT, buf, len);
1233 vid_hdr->vol_type = UBI_VID_DYNAMIC;
1234 vid_hdr->data_size = cpu_to_be32(len);
1235 vid_hdr->copy_flag = 1;
1236 vid_hdr->data_crc = cpu_to_be32(crc);
1237
1238 dbg_eba("change LEB %d:%d", vol_id, lnum);
1239
1240 for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1241 err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1242 if (err != -EIO || !ubi->bad_allowed)
1243 break;
1244
1245 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1246 ubi_msg(ubi, "try another PEB");
1247 }
1248
1249 /*
1250 * This flash device does not admit of bad eraseblocks or
1251 * something nasty and unexpected happened. Switch to read-only
1252 * mode just in case.
1253 */
1254 if (err)
1255 ubi_ro_mode(ubi);
1256
1257 leb_write_unlock(ubi, vol_id, lnum);
1258
1259out_mutex:
1260 mutex_unlock(&ubi->alc_mutex);
1261 ubi_free_vid_buf(vidb);
1262 return err;
1263}
1264
1265/**
1266 * is_error_sane - check whether a read error is sane.
1267 * @err: code of the error happened during reading
1268 *
1269 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1270 * cannot read data from the target PEB (an error @err happened). If the error
1271 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1272 * fatal and UBI will be switched to R/O mode later.
1273 *
1274 * The idea is that we try not to switch to R/O mode if the read error is
1275 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1276 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1277 * mode, simply because we do not know what happened at the MTD level, and we
1278 * cannot handle this. E.g., the underlying driver may have become crazy, and
1279 * it is safer to switch to R/O mode to preserve the data.
1280 *
1281 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1282 * which we have just written.
1283 */
1284static int is_error_sane(int err)
1285{
1286 if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1287 err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1288 return 0;
1289 return 1;
1290}
1291
1292/**
1293 * ubi_eba_copy_leb - copy logical eraseblock.
1294 * @ubi: UBI device description object
1295 * @from: physical eraseblock number from where to copy
1296 * @to: physical eraseblock number where to copy
1297 * @vidb: data structure from where the VID header is derived
1298 *
1299 * This function copies logical eraseblock from physical eraseblock @from to
1300 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1301 * function. Returns:
1302 * o %0 in case of success;
1303 * o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1304 * o a negative error code in case of failure.
1305 */
1306int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1307 struct ubi_vid_io_buf *vidb)
1308{
1309 int err, vol_id, lnum, data_size, aldata_size, idx;
1310 struct ubi_vid_hdr *vid_hdr = ubi_get_vid_hdr(vidb);
1311 struct ubi_volume *vol;
1312 uint32_t crc;
1313
1314 ubi_assert(rwsem_is_locked(&ubi->fm_eba_sem));
1315
1316 vol_id = be32_to_cpu(vid_hdr->vol_id);
1317 lnum = be32_to_cpu(vid_hdr->lnum);
1318
1319 dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1320
1321 if (vid_hdr->vol_type == UBI_VID_STATIC) {
1322 data_size = be32_to_cpu(vid_hdr->data_size);
1323 aldata_size = ALIGN(data_size, ubi->min_io_size);
1324 } else
1325 data_size = aldata_size =
1326 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1327
1328 idx = vol_id2idx(ubi, vol_id);
1329 spin_lock(&ubi->volumes_lock);
1330 /*
1331 * Note, we may race with volume deletion, which means that the volume
1332 * this logical eraseblock belongs to might be being deleted. Since the
1333 * volume deletion un-maps all the volume's logical eraseblocks, it will
1334 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1335 */
1336 vol = ubi->volumes[idx];
1337 spin_unlock(&ubi->volumes_lock);
1338 if (!vol) {
1339 /* No need to do further work, cancel */
1340 dbg_wl("volume %d is being removed, cancel", vol_id);
1341 return MOVE_CANCEL_RACE;
1342 }
1343
1344 /*
1345 * We do not want anybody to write to this logical eraseblock while we
1346 * are moving it, so lock it.
1347 *
1348 * Note, we are using non-waiting locking here, because we cannot sleep
1349 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1350 * unmapping the LEB which is mapped to the PEB we are going to move
1351 * (@from). This task locks the LEB and goes sleep in the
1352 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1353 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1354 * LEB is already locked, we just do not move it and return
1355 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1356 * we do not know the reasons of the contention - it may be just a
1357 * normal I/O on this LEB, so we want to re-try.
1358 */
1359 err = leb_write_trylock(ubi, vol_id, lnum);
1360 if (err) {
1361 dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1362 return MOVE_RETRY;
1363 }
1364
1365 /*
1366 * The LEB might have been put meanwhile, and the task which put it is
1367 * probably waiting on @ubi->move_mutex. No need to continue the work,
1368 * cancel it.
1369 */
1370 if (vol->eba_tbl->entries[lnum].pnum != from) {
1371 dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1372 vol_id, lnum, from, vol->eba_tbl->entries[lnum].pnum);
1373 err = MOVE_CANCEL_RACE;
1374 goto out_unlock_leb;
1375 }
1376
1377 /*
1378 * OK, now the LEB is locked and we can safely start moving it. Since
1379 * this function utilizes the @ubi->peb_buf buffer which is shared
1380 * with some other functions - we lock the buffer by taking the
1381 * @ubi->buf_mutex.
1382 */
1383 mutex_lock(&ubi->buf_mutex);
1384 dbg_wl("read %d bytes of data", aldata_size);
1385 err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1386 if (err && err != UBI_IO_BITFLIPS) {
1387 ubi_warn(ubi, "error %d while reading data from PEB %d",
1388 err, from);
1389 err = MOVE_SOURCE_RD_ERR;
1390 goto out_unlock_buf;
1391 }
1392
1393 /*
1394 * Now we have got to calculate how much data we have to copy. In
1395 * case of a static volume it is fairly easy - the VID header contains
1396 * the data size. In case of a dynamic volume it is more difficult - we
1397 * have to read the contents, cut 0xFF bytes from the end and copy only
1398 * the first part. We must do this to avoid writing 0xFF bytes as it
1399 * may have some side-effects. And not only this. It is important not
1400 * to include those 0xFFs to CRC because later the they may be filled
1401 * by data.
1402 */
1403 if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1404 aldata_size = data_size =
1405 ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1406
1407 cond_resched();
1408 crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1409 cond_resched();
1410
1411 /*
1412 * It may turn out to be that the whole @from physical eraseblock
1413 * contains only 0xFF bytes. Then we have to only write the VID header
1414 * and do not write any data. This also means we should not set
1415 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1416 */
1417 if (data_size > 0) {
1418 vid_hdr->copy_flag = 1;
1419 vid_hdr->data_size = cpu_to_be32(data_size);
1420 vid_hdr->data_crc = cpu_to_be32(crc);
1421 }
1422 vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1423
1424 err = ubi_io_write_vid_hdr(ubi, to, vidb);
1425 if (err) {
1426 if (err == -EIO)
1427 err = MOVE_TARGET_WR_ERR;
1428 goto out_unlock_buf;
1429 }
1430
1431 cond_resched();
1432
1433 /* Read the VID header back and check if it was written correctly */
1434 err = ubi_io_read_vid_hdr(ubi, to, vidb, 1);
1435 if (err) {
1436 if (err != UBI_IO_BITFLIPS) {
1437 ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1438 err, to);
1439 if (is_error_sane(err))
1440 err = MOVE_TARGET_RD_ERR;
1441 } else
1442 err = MOVE_TARGET_BITFLIPS;
1443 goto out_unlock_buf;
1444 }
1445
1446 if (data_size > 0) {
1447 err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1448 if (err) {
1449 if (err == -EIO)
1450 err = MOVE_TARGET_WR_ERR;
1451 goto out_unlock_buf;
1452 }
1453
1454 cond_resched();
1455 }
1456
1457 ubi_assert(vol->eba_tbl->entries[lnum].pnum == from);
1458
1459 /**
1460 * The volumes_lock lock is needed here to prevent the expired old eba_tbl
1461 * being updated when the eba_tbl is copied in the ubi_resize_volume() process.
1462 */
1463 spin_lock(&ubi->volumes_lock);
1464 vol->eba_tbl->entries[lnum].pnum = to;
1465 spin_unlock(&ubi->volumes_lock);
1466
1467out_unlock_buf:
1468 mutex_unlock(&ubi->buf_mutex);
1469out_unlock_leb:
1470 leb_write_unlock(ubi, vol_id, lnum);
1471 return err;
1472}
1473
1474/**
1475 * print_rsvd_warning - warn about not having enough reserved PEBs.
1476 * @ubi: UBI device description object
1477 * @ai: UBI attach info object
1478 *
1479 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1480 * cannot reserve enough PEBs for bad block handling. This function makes a
1481 * decision whether we have to print a warning or not. The algorithm is as
1482 * follows:
1483 * o if this is a new UBI image, then just print the warning
1484 * o if this is an UBI image which has already been used for some time, print
1485 * a warning only if we can reserve less than 10% of the expected amount of
1486 * the reserved PEB.
1487 *
1488 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1489 * of PEBs becomes smaller, which is normal and we do not want to scare users
1490 * with a warning every time they attach the MTD device. This was an issue
1491 * reported by real users.
1492 */
1493static void print_rsvd_warning(struct ubi_device *ubi,
1494 struct ubi_attach_info *ai)
1495{
1496 /*
1497 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1498 * large number to distinguish between newly flashed and used images.
1499 */
1500 if (ai->max_sqnum > (1 << 18)) {
1501 int min = ubi->beb_rsvd_level / 10;
1502
1503 if (!min)
1504 min = 1;
1505 if (ubi->beb_rsvd_pebs > min)
1506 return;
1507 }
1508
1509 ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1510 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1511 if (ubi->corr_peb_count)
1512 ubi_warn(ubi, "%d PEBs are corrupted and not used",
1513 ubi->corr_peb_count);
1514}
1515
1516/**
1517 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1518 * @ubi: UBI device description object
1519 * @ai_fastmap: UBI attach info object created by fastmap
1520 * @ai_scan: UBI attach info object created by scanning
1521 *
1522 * Returns < 0 in case of an internal error, 0 otherwise.
1523 * If a bad EBA table entry was found it will be printed out and
1524 * ubi_assert() triggers.
1525 */
1526int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1527 struct ubi_attach_info *ai_scan)
1528{
1529 int i, j, num_volumes, ret = 0;
1530 int **scan_eba, **fm_eba;
1531 struct ubi_ainf_volume *av;
1532 struct ubi_volume *vol;
1533 struct ubi_ainf_peb *aeb;
1534 struct rb_node *rb;
1535
1536 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1537
1538 scan_eba = kmalloc_objs(*scan_eba, num_volumes);
1539 if (!scan_eba)
1540 return -ENOMEM;
1541
1542 fm_eba = kmalloc_objs(*fm_eba, num_volumes);
1543 if (!fm_eba) {
1544 kfree(scan_eba);
1545 return -ENOMEM;
1546 }
1547
1548 for (i = 0; i < num_volumes; i++) {
1549 vol = ubi->volumes[i];
1550 if (!vol)
1551 continue;
1552
1553 scan_eba[i] = kmalloc_objs(**scan_eba, vol->reserved_pebs);
1554 if (!scan_eba[i]) {
1555 ret = -ENOMEM;
1556 goto out_free;
1557 }
1558
1559 fm_eba[i] = kmalloc_objs(**fm_eba, vol->reserved_pebs);
1560 if (!fm_eba[i]) {
1561 ret = -ENOMEM;
1562 kfree(scan_eba[i]);
1563 goto out_free;
1564 }
1565
1566 for (j = 0; j < vol->reserved_pebs; j++)
1567 scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1568
1569 av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1570 if (!av)
1571 continue;
1572
1573 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1574 scan_eba[i][aeb->lnum] = aeb->pnum;
1575
1576 av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1577 if (!av)
1578 continue;
1579
1580 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1581 fm_eba[i][aeb->lnum] = aeb->pnum;
1582
1583 for (j = 0; j < vol->reserved_pebs; j++) {
1584 if (scan_eba[i][j] != fm_eba[i][j]) {
1585 if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1586 fm_eba[i][j] == UBI_LEB_UNMAPPED)
1587 continue;
1588
1589 ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1590 vol->vol_id, j, fm_eba[i][j],
1591 scan_eba[i][j]);
1592 ubi_assert(0);
1593 }
1594 }
1595 }
1596
1597out_free:
1598 while (--i >= 0) {
1599 if (!ubi->volumes[i])
1600 continue;
1601
1602 kfree(scan_eba[i]);
1603 kfree(fm_eba[i]);
1604 }
1605
1606 kfree(scan_eba);
1607 kfree(fm_eba);
1608 return ret;
1609}
1610
1611/**
1612 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1613 * @ubi: UBI device description object
1614 * @ai: attaching information
1615 *
1616 * This function returns zero in case of success and a negative error code in
1617 * case of failure.
1618 */
1619int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1620{
1621 int i, err, num_volumes;
1622 struct ubi_ainf_volume *av;
1623 struct ubi_volume *vol;
1624 struct ubi_ainf_peb *aeb;
1625 struct rb_node *rb;
1626
1627 dbg_eba("initialize EBA sub-system");
1628
1629 spin_lock_init(&ubi->ltree_lock);
1630 mutex_init(&ubi->alc_mutex);
1631 ubi->ltree = RB_ROOT;
1632
1633 ubi->global_sqnum = ai->max_sqnum + 1;
1634 num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1635
1636 for (i = 0; i < num_volumes; i++) {
1637 struct ubi_eba_table *tbl;
1638
1639 vol = ubi->volumes[i];
1640 if (!vol)
1641 continue;
1642
1643 cond_resched();
1644
1645 tbl = ubi_eba_create_table(vol, vol->reserved_pebs);
1646 if (IS_ERR(tbl)) {
1647 err = PTR_ERR(tbl);
1648 goto out_free;
1649 }
1650
1651 ubi_eba_replace_table(vol, tbl);
1652
1653 av = ubi_find_av(ai, idx2vol_id(ubi, i));
1654 if (!av)
1655 continue;
1656
1657 ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1658 if (aeb->lnum >= vol->reserved_pebs) {
1659 /*
1660 * This may happen in case of an unclean reboot
1661 * during re-size.
1662 */
1663 ubi_move_aeb_to_list(av, aeb, &ai->erase);
1664 } else {
1665 struct ubi_eba_entry *entry;
1666
1667 entry = &vol->eba_tbl->entries[aeb->lnum];
1668 entry->pnum = aeb->pnum;
1669 }
1670 }
1671 }
1672
1673 if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1674 ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1675 ubi->avail_pebs, EBA_RESERVED_PEBS);
1676 if (ubi->corr_peb_count)
1677 ubi_err(ubi, "%d PEBs are corrupted and not used",
1678 ubi->corr_peb_count);
1679 err = -ENOSPC;
1680 goto out_free;
1681 }
1682 ubi->avail_pebs -= EBA_RESERVED_PEBS;
1683 ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1684
1685 if (ubi->bad_allowed) {
1686 ubi_calculate_reserved(ubi);
1687
1688 if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1689 /* No enough free physical eraseblocks */
1690 ubi->beb_rsvd_pebs = ubi->avail_pebs;
1691 print_rsvd_warning(ubi, ai);
1692 } else
1693 ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1694
1695 ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1696 ubi->rsvd_pebs += ubi->beb_rsvd_pebs;
1697 }
1698
1699 dbg_eba("EBA sub-system is initialized");
1700 return 0;
1701
1702out_free:
1703 for (i = 0; i < num_volumes; i++) {
1704 if (!ubi->volumes[i])
1705 continue;
1706 ubi_eba_replace_table(ubi->volumes[i], NULL);
1707 }
1708 return err;
1709}