Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright (C) 2015 Google, Inc.
4 *
5 * Author: Sami Tolvanen <samitolvanen@google.com>
6 */
7
8#include "dm-verity-fec.h"
9#include <linux/math64.h>
10
11#define DM_MSG_PREFIX "verity-fec"
12
13/*
14 * If error correction has been configured, returns true.
15 */
16bool verity_fec_is_enabled(struct dm_verity *v)
17{
18 return v->fec && v->fec->dev;
19}
20
21/*
22 * Return a pointer to dm_verity_fec_io after dm_verity_io and its variable
23 * length fields.
24 */
25static inline struct dm_verity_fec_io *fec_io(struct dm_verity_io *io)
26{
27 return (struct dm_verity_fec_io *)
28 ((char *)io + io->v->ti->per_io_data_size - sizeof(struct dm_verity_fec_io));
29}
30
31/*
32 * Return an interleaved offset for a byte in RS block.
33 */
34static inline u64 fec_interleave(struct dm_verity *v, u64 offset)
35{
36 u32 mod;
37
38 mod = do_div(offset, v->fec->rsn);
39 return offset + mod * (v->fec->rounds << v->data_dev_block_bits);
40}
41
42/*
43 * Read error-correcting codes for the requested RS block. Returns a pointer
44 * to the data block. Caller is responsible for releasing buf.
45 */
46static u8 *fec_read_parity(struct dm_verity *v, u64 rsb, int index,
47 unsigned int *offset, unsigned int par_buf_offset,
48 struct dm_buffer **buf, unsigned short ioprio)
49{
50 u64 position, block, rem;
51 u8 *res;
52
53 /* We have already part of parity bytes read, skip to the next block */
54 if (par_buf_offset)
55 index++;
56
57 position = (index + rsb) * v->fec->roots;
58 block = div64_u64_rem(position, v->fec->io_size, &rem);
59 *offset = par_buf_offset ? 0 : (unsigned int)rem;
60
61 res = dm_bufio_read_with_ioprio(v->fec->bufio, block, buf, ioprio);
62 if (IS_ERR(res)) {
63 DMERR("%s: FEC %llu: parity read failed (block %llu): %ld",
64 v->data_dev->name, (unsigned long long)rsb,
65 (unsigned long long)block, PTR_ERR(res));
66 *buf = NULL;
67 }
68
69 return res;
70}
71
72/* Loop over each preallocated buffer slot. */
73#define fec_for_each_prealloc_buffer(__i) \
74 for (__i = 0; __i < DM_VERITY_FEC_BUF_PREALLOC; __i++)
75
76/* Loop over each extra buffer slot. */
77#define fec_for_each_extra_buffer(io, __i) \
78 for (__i = DM_VERITY_FEC_BUF_PREALLOC; __i < DM_VERITY_FEC_BUF_MAX; __i++)
79
80/* Loop over each allocated buffer. */
81#define fec_for_each_buffer(io, __i) \
82 for (__i = 0; __i < (io)->nbufs; __i++)
83
84/* Loop over each RS block in each allocated buffer. */
85#define fec_for_each_buffer_rs_block(io, __i, __j) \
86 fec_for_each_buffer(io, __i) \
87 for (__j = 0; __j < 1 << DM_VERITY_FEC_BUF_RS_BITS; __j++)
88
89/*
90 * Return a pointer to the current RS block when called inside
91 * fec_for_each_buffer_rs_block.
92 */
93static inline u8 *fec_buffer_rs_block(struct dm_verity *v,
94 struct dm_verity_fec_io *fio,
95 unsigned int i, unsigned int j)
96{
97 return &fio->bufs[i][j * v->fec->rsn];
98}
99
100/*
101 * Return an index to the current RS block when called inside
102 * fec_for_each_buffer_rs_block.
103 */
104static inline unsigned int fec_buffer_rs_index(unsigned int i, unsigned int j)
105{
106 return (i << DM_VERITY_FEC_BUF_RS_BITS) + j;
107}
108
109/*
110 * Decode all RS blocks from buffers and copy corrected bytes into fio->output
111 * starting from block_offset.
112 */
113static int fec_decode_bufs(struct dm_verity *v, struct dm_verity_io *io,
114 struct dm_verity_fec_io *fio, u64 rsb, int byte_index,
115 unsigned int block_offset, int neras)
116{
117 int r, corrected = 0, res;
118 struct dm_buffer *buf;
119 unsigned int n, i, j, offset, par_buf_offset = 0;
120 uint16_t par_buf[DM_VERITY_FEC_RSM - DM_VERITY_FEC_MIN_RSN];
121 u8 *par, *block;
122 struct bio *bio = dm_bio_from_per_bio_data(io, v->ti->per_io_data_size);
123
124 par = fec_read_parity(v, rsb, block_offset, &offset,
125 par_buf_offset, &buf, bio_prio(bio));
126 if (IS_ERR(par))
127 return PTR_ERR(par);
128
129 /*
130 * Decode the RS blocks we have in bufs. Each RS block results in
131 * one corrected target byte and consumes fec->roots parity bytes.
132 */
133 fec_for_each_buffer_rs_block(fio, n, i) {
134 block = fec_buffer_rs_block(v, fio, n, i);
135 for (j = 0; j < v->fec->roots - par_buf_offset; j++)
136 par_buf[par_buf_offset + j] = par[offset + j];
137 /* Decode an RS block using Reed-Solomon */
138 res = decode_rs8(fio->rs, block, par_buf, v->fec->rsn,
139 NULL, neras, fio->erasures, 0, NULL);
140 if (res < 0) {
141 r = res;
142 goto error;
143 }
144
145 corrected += res;
146 fio->output[block_offset] = block[byte_index];
147
148 block_offset++;
149 if (block_offset >= 1 << v->data_dev_block_bits)
150 goto done;
151
152 /* Read the next block when we run out of parity bytes */
153 offset += (v->fec->roots - par_buf_offset);
154 /* Check if parity bytes are split between blocks */
155 if (offset < v->fec->io_size && (offset + v->fec->roots) > v->fec->io_size) {
156 par_buf_offset = v->fec->io_size - offset;
157 for (j = 0; j < par_buf_offset; j++)
158 par_buf[j] = par[offset + j];
159 offset += par_buf_offset;
160 } else
161 par_buf_offset = 0;
162
163 if (offset >= v->fec->io_size) {
164 dm_bufio_release(buf);
165
166 par = fec_read_parity(v, rsb, block_offset, &offset,
167 par_buf_offset, &buf, bio_prio(bio));
168 if (IS_ERR(par))
169 return PTR_ERR(par);
170 }
171 }
172done:
173 r = corrected;
174error:
175 dm_bufio_release(buf);
176
177 if (r < 0 && neras)
178 DMERR_LIMIT("%s: FEC %llu: failed to correct: %d",
179 v->data_dev->name, (unsigned long long)rsb, r);
180 else if (r > 0)
181 DMWARN_LIMIT("%s: FEC %llu: corrected %d errors",
182 v->data_dev->name, (unsigned long long)rsb, r);
183
184 return r;
185}
186
187/*
188 * Locate data block erasures using verity hashes.
189 */
190static int fec_is_erasure(struct dm_verity *v, struct dm_verity_io *io,
191 u8 *want_digest, u8 *data)
192{
193 if (unlikely(verity_hash(v, io, data, 1 << v->data_dev_block_bits,
194 verity_io_real_digest(v, io), true)))
195 return 0;
196
197 return memcmp(verity_io_real_digest(v, io), want_digest,
198 v->digest_size) != 0;
199}
200
201/*
202 * Read data blocks that are part of the RS block and deinterleave as much as
203 * fits into buffers. Check for erasure locations if @neras is non-NULL.
204 */
205static int fec_read_bufs(struct dm_verity *v, struct dm_verity_io *io,
206 u64 rsb, u64 target, unsigned int block_offset,
207 int *neras)
208{
209 bool is_zero;
210 int i, j, target_index = -1;
211 struct dm_buffer *buf;
212 struct dm_bufio_client *bufio;
213 struct dm_verity_fec_io *fio = fec_io(io);
214 u64 block, ileaved;
215 u8 *bbuf, *rs_block;
216 u8 want_digest[HASH_MAX_DIGESTSIZE];
217 unsigned int n, k;
218 struct bio *bio = dm_bio_from_per_bio_data(io, v->ti->per_io_data_size);
219
220 if (neras)
221 *neras = 0;
222
223 if (WARN_ON(v->digest_size > sizeof(want_digest)))
224 return -EINVAL;
225
226 /*
227 * read each of the rsn data blocks that are part of the RS block, and
228 * interleave contents to available bufs
229 */
230 for (i = 0; i < v->fec->rsn; i++) {
231 ileaved = fec_interleave(v, rsb * v->fec->rsn + i);
232
233 /*
234 * target is the data block we want to correct, target_index is
235 * the index of this block within the rsn RS blocks
236 */
237 if (ileaved == target)
238 target_index = i;
239
240 block = ileaved >> v->data_dev_block_bits;
241 bufio = v->fec->data_bufio;
242
243 if (block >= v->data_blocks) {
244 block -= v->data_blocks;
245
246 /*
247 * blocks outside the area were assumed to contain
248 * zeros when encoding data was generated
249 */
250 if (unlikely(block >= v->fec->hash_blocks))
251 continue;
252
253 block += v->hash_start;
254 bufio = v->bufio;
255 }
256
257 bbuf = dm_bufio_read_with_ioprio(bufio, block, &buf, bio_prio(bio));
258 if (IS_ERR(bbuf)) {
259 DMWARN_LIMIT("%s: FEC %llu: read failed (%llu): %ld",
260 v->data_dev->name,
261 (unsigned long long)rsb,
262 (unsigned long long)block, PTR_ERR(bbuf));
263
264 /* assume the block is corrupted */
265 if (neras && *neras <= v->fec->roots)
266 fio->erasures[(*neras)++] = i;
267
268 continue;
269 }
270
271 /* locate erasures if the block is on the data device */
272 if (bufio == v->fec->data_bufio &&
273 verity_hash_for_block(v, io, block, want_digest,
274 &is_zero) == 0) {
275 /* skip known zero blocks entirely */
276 if (is_zero)
277 goto done;
278
279 /*
280 * skip if we have already found the theoretical
281 * maximum number (i.e. fec->roots) of erasures
282 */
283 if (neras && *neras <= v->fec->roots &&
284 fec_is_erasure(v, io, want_digest, bbuf))
285 fio->erasures[(*neras)++] = i;
286 }
287
288 /*
289 * deinterleave and copy the bytes that fit into bufs,
290 * starting from block_offset
291 */
292 fec_for_each_buffer_rs_block(fio, n, j) {
293 k = fec_buffer_rs_index(n, j) + block_offset;
294
295 if (k >= 1 << v->data_dev_block_bits)
296 goto done;
297
298 rs_block = fec_buffer_rs_block(v, fio, n, j);
299 rs_block[i] = bbuf[k];
300 }
301done:
302 dm_bufio_release(buf);
303 }
304
305 return target_index;
306}
307
308/*
309 * Allocate RS control structure and FEC buffers from preallocated mempools,
310 * and attempt to allocate as many extra buffers as available.
311 */
312static int fec_alloc_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
313{
314 unsigned int n;
315
316 if (!fio->rs)
317 fio->rs = mempool_alloc(&v->fec->rs_pool, GFP_NOIO);
318
319 fec_for_each_prealloc_buffer(n) {
320 if (fio->bufs[n])
321 continue;
322
323 fio->bufs[n] = mempool_alloc(&v->fec->prealloc_pool, GFP_NOWAIT);
324 if (unlikely(!fio->bufs[n])) {
325 DMERR("failed to allocate FEC buffer");
326 return -ENOMEM;
327 }
328 }
329
330 /* try to allocate the maximum number of buffers */
331 fec_for_each_extra_buffer(fio, n) {
332 if (fio->bufs[n])
333 continue;
334
335 fio->bufs[n] = mempool_alloc(&v->fec->extra_pool, GFP_NOWAIT);
336 /* we can manage with even one buffer if necessary */
337 if (unlikely(!fio->bufs[n]))
338 break;
339 }
340 fio->nbufs = n;
341
342 if (!fio->output)
343 fio->output = mempool_alloc(&v->fec->output_pool, GFP_NOIO);
344
345 return 0;
346}
347
348/*
349 * Initialize buffers and clear erasures. fec_read_bufs() assumes buffers are
350 * zeroed before deinterleaving.
351 */
352static void fec_init_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
353{
354 unsigned int n;
355
356 fec_for_each_buffer(fio, n)
357 memset(fio->bufs[n], 0, v->fec->rsn << DM_VERITY_FEC_BUF_RS_BITS);
358
359 memset(fio->erasures, 0, sizeof(fio->erasures));
360}
361
362/*
363 * Decode all RS blocks in a single data block and return the target block
364 * (indicated by @offset) in fio->output. If @use_erasures is non-zero, uses
365 * hashes to locate erasures.
366 */
367static int fec_decode_rsb(struct dm_verity *v, struct dm_verity_io *io,
368 struct dm_verity_fec_io *fio, u64 rsb, u64 offset,
369 bool use_erasures)
370{
371 int r, neras = 0;
372 unsigned int pos;
373
374 r = fec_alloc_bufs(v, fio);
375 if (unlikely(r < 0))
376 return r;
377
378 for (pos = 0; pos < 1 << v->data_dev_block_bits; ) {
379 fec_init_bufs(v, fio);
380
381 r = fec_read_bufs(v, io, rsb, offset, pos,
382 use_erasures ? &neras : NULL);
383 if (unlikely(r < 0))
384 return r;
385
386 r = fec_decode_bufs(v, io, fio, rsb, r, pos, neras);
387 if (r < 0)
388 return r;
389
390 pos += fio->nbufs << DM_VERITY_FEC_BUF_RS_BITS;
391 }
392
393 /* Always re-validate the corrected block against the expected hash */
394 r = verity_hash(v, io, fio->output, 1 << v->data_dev_block_bits,
395 verity_io_real_digest(v, io), true);
396 if (unlikely(r < 0))
397 return r;
398
399 if (memcmp(verity_io_real_digest(v, io), verity_io_want_digest(v, io),
400 v->digest_size)) {
401 DMERR_LIMIT("%s: FEC %llu: failed to correct (%d erasures)",
402 v->data_dev->name, (unsigned long long)rsb, neras);
403 return -EILSEQ;
404 }
405
406 return 0;
407}
408
409/* Correct errors in a block. Copies corrected block to dest. */
410int verity_fec_decode(struct dm_verity *v, struct dm_verity_io *io,
411 enum verity_block_type type, sector_t block, u8 *dest)
412{
413 int r;
414 struct dm_verity_fec_io *fio = fec_io(io);
415 u64 offset, res, rsb;
416
417 if (!verity_fec_is_enabled(v))
418 return -EOPNOTSUPP;
419
420 if (fio->level >= DM_VERITY_FEC_MAX_RECURSION) {
421 DMWARN_LIMIT("%s: FEC: recursion too deep", v->data_dev->name);
422 return -EIO;
423 }
424
425 fio->level++;
426
427 if (type == DM_VERITY_BLOCK_TYPE_METADATA)
428 block = block - v->hash_start + v->data_blocks;
429
430 /*
431 * For RS(M, N), the continuous FEC data is divided into blocks of N
432 * bytes. Since block size may not be divisible by N, the last block
433 * is zero padded when decoding.
434 *
435 * Each byte of the block is covered by a different RS(M, N) code,
436 * and each code is interleaved over N blocks to make it less likely
437 * that bursty corruption will leave us in unrecoverable state.
438 */
439
440 offset = block << v->data_dev_block_bits;
441 res = div64_u64(offset, v->fec->rounds << v->data_dev_block_bits);
442
443 /*
444 * The base RS block we can feed to the interleaver to find out all
445 * blocks required for decoding.
446 */
447 rsb = offset - res * (v->fec->rounds << v->data_dev_block_bits);
448
449 /*
450 * Locating erasures is slow, so attempt to recover the block without
451 * them first. Do a second attempt with erasures if the corruption is
452 * bad enough.
453 */
454 r = fec_decode_rsb(v, io, fio, rsb, offset, false);
455 if (r < 0) {
456 r = fec_decode_rsb(v, io, fio, rsb, offset, true);
457 if (r < 0)
458 goto done;
459 }
460
461 memcpy(dest, fio->output, 1 << v->data_dev_block_bits);
462
463done:
464 fio->level--;
465 return r;
466}
467
468/*
469 * Clean up per-bio data.
470 */
471void verity_fec_finish_io(struct dm_verity_io *io)
472{
473 unsigned int n;
474 struct dm_verity_fec *f = io->v->fec;
475 struct dm_verity_fec_io *fio = fec_io(io);
476
477 if (!verity_fec_is_enabled(io->v))
478 return;
479
480 mempool_free(fio->rs, &f->rs_pool);
481
482 fec_for_each_prealloc_buffer(n)
483 mempool_free(fio->bufs[n], &f->prealloc_pool);
484
485 fec_for_each_extra_buffer(fio, n)
486 mempool_free(fio->bufs[n], &f->extra_pool);
487
488 mempool_free(fio->output, &f->output_pool);
489}
490
491/*
492 * Initialize per-bio data.
493 */
494void verity_fec_init_io(struct dm_verity_io *io)
495{
496 struct dm_verity_fec_io *fio = fec_io(io);
497
498 if (!verity_fec_is_enabled(io->v))
499 return;
500
501 fio->rs = NULL;
502 memset(fio->bufs, 0, sizeof(fio->bufs));
503 fio->nbufs = 0;
504 fio->output = NULL;
505 fio->level = 0;
506}
507
508/*
509 * Append feature arguments and values to the status table.
510 */
511unsigned int verity_fec_status_table(struct dm_verity *v, unsigned int sz,
512 char *result, unsigned int maxlen)
513{
514 if (!verity_fec_is_enabled(v))
515 return sz;
516
517 DMEMIT(" " DM_VERITY_OPT_FEC_DEV " %s "
518 DM_VERITY_OPT_FEC_BLOCKS " %llu "
519 DM_VERITY_OPT_FEC_START " %llu "
520 DM_VERITY_OPT_FEC_ROOTS " %d",
521 v->fec->dev->name,
522 (unsigned long long)v->fec->blocks,
523 (unsigned long long)v->fec->start,
524 v->fec->roots);
525
526 return sz;
527}
528
529void verity_fec_dtr(struct dm_verity *v)
530{
531 struct dm_verity_fec *f = v->fec;
532
533 if (!verity_fec_is_enabled(v))
534 goto out;
535
536 mempool_exit(&f->rs_pool);
537 mempool_exit(&f->prealloc_pool);
538 mempool_exit(&f->extra_pool);
539 mempool_exit(&f->output_pool);
540 kmem_cache_destroy(f->cache);
541
542 if (f->data_bufio)
543 dm_bufio_client_destroy(f->data_bufio);
544 if (f->bufio)
545 dm_bufio_client_destroy(f->bufio);
546
547 if (f->dev)
548 dm_put_device(v->ti, f->dev);
549out:
550 kfree(f);
551 v->fec = NULL;
552}
553
554static void *fec_rs_alloc(gfp_t gfp_mask, void *pool_data)
555{
556 struct dm_verity *v = pool_data;
557
558 return init_rs_gfp(8, 0x11d, 0, 1, v->fec->roots, gfp_mask);
559}
560
561static void fec_rs_free(void *element, void *pool_data)
562{
563 struct rs_control *rs = element;
564
565 if (rs)
566 free_rs(rs);
567}
568
569bool verity_is_fec_opt_arg(const char *arg_name)
570{
571 return (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV) ||
572 !strcasecmp(arg_name, DM_VERITY_OPT_FEC_BLOCKS) ||
573 !strcasecmp(arg_name, DM_VERITY_OPT_FEC_START) ||
574 !strcasecmp(arg_name, DM_VERITY_OPT_FEC_ROOTS));
575}
576
577int verity_fec_parse_opt_args(struct dm_arg_set *as, struct dm_verity *v,
578 unsigned int *argc, const char *arg_name)
579{
580 int r;
581 struct dm_target *ti = v->ti;
582 const char *arg_value;
583 unsigned long long num_ll;
584 unsigned char num_c;
585 char dummy;
586
587 if (!*argc) {
588 ti->error = "FEC feature arguments require a value";
589 return -EINVAL;
590 }
591
592 arg_value = dm_shift_arg(as);
593 (*argc)--;
594
595 if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV)) {
596 r = dm_get_device(ti, arg_value, BLK_OPEN_READ, &v->fec->dev);
597 if (r) {
598 ti->error = "FEC device lookup failed";
599 return r;
600 }
601
602 } else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_BLOCKS)) {
603 if (sscanf(arg_value, "%llu%c", &num_ll, &dummy) != 1 ||
604 ((sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT))
605 >> (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll)) {
606 ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
607 return -EINVAL;
608 }
609 v->fec->blocks = num_ll;
610
611 } else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_START)) {
612 if (sscanf(arg_value, "%llu%c", &num_ll, &dummy) != 1 ||
613 ((sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT)) >>
614 (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll)) {
615 ti->error = "Invalid " DM_VERITY_OPT_FEC_START;
616 return -EINVAL;
617 }
618 v->fec->start = num_ll;
619
620 } else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_ROOTS)) {
621 if (sscanf(arg_value, "%hhu%c", &num_c, &dummy) != 1 || !num_c ||
622 num_c < (DM_VERITY_FEC_RSM - DM_VERITY_FEC_MAX_RSN) ||
623 num_c > (DM_VERITY_FEC_RSM - DM_VERITY_FEC_MIN_RSN)) {
624 ti->error = "Invalid " DM_VERITY_OPT_FEC_ROOTS;
625 return -EINVAL;
626 }
627 v->fec->roots = num_c;
628
629 } else {
630 ti->error = "Unrecognized verity FEC feature request";
631 return -EINVAL;
632 }
633
634 return 0;
635}
636
637/*
638 * Allocate dm_verity_fec for v->fec. Must be called before verity_fec_ctr.
639 */
640int verity_fec_ctr_alloc(struct dm_verity *v)
641{
642 struct dm_verity_fec *f;
643
644 f = kzalloc(sizeof(struct dm_verity_fec), GFP_KERNEL);
645 if (!f) {
646 v->ti->error = "Cannot allocate FEC structure";
647 return -ENOMEM;
648 }
649 v->fec = f;
650
651 return 0;
652}
653
654/*
655 * Validate arguments and preallocate memory. Must be called after arguments
656 * have been parsed using verity_fec_parse_opt_args.
657 */
658int verity_fec_ctr(struct dm_verity *v)
659{
660 struct dm_verity_fec *f = v->fec;
661 struct dm_target *ti = v->ti;
662 u64 hash_blocks, fec_blocks;
663 int ret;
664
665 if (!verity_fec_is_enabled(v)) {
666 verity_fec_dtr(v);
667 return 0;
668 }
669
670 /*
671 * FEC is computed over data blocks, possible metadata, and
672 * hash blocks. In other words, FEC covers total of fec_blocks
673 * blocks consisting of the following:
674 *
675 * data blocks | hash blocks | metadata (optional)
676 *
677 * We allow metadata after hash blocks to support a use case
678 * where all data is stored on the same device and FEC covers
679 * the entire area.
680 *
681 * If metadata is included, we require it to be available on the
682 * hash device after the hash blocks.
683 */
684
685 hash_blocks = v->hash_blocks - v->hash_start;
686
687 /*
688 * Require matching block sizes for data and hash devices for
689 * simplicity.
690 */
691 if (v->data_dev_block_bits != v->hash_dev_block_bits) {
692 ti->error = "Block sizes must match to use FEC";
693 return -EINVAL;
694 }
695
696 if (!f->roots) {
697 ti->error = "Missing " DM_VERITY_OPT_FEC_ROOTS;
698 return -EINVAL;
699 }
700 f->rsn = DM_VERITY_FEC_RSM - f->roots;
701
702 if (!f->blocks) {
703 ti->error = "Missing " DM_VERITY_OPT_FEC_BLOCKS;
704 return -EINVAL;
705 }
706
707 f->rounds = f->blocks;
708 if (sector_div(f->rounds, f->rsn))
709 f->rounds++;
710
711 /*
712 * Due to optional metadata, f->blocks can be larger than
713 * data_blocks and hash_blocks combined.
714 */
715 if (f->blocks < v->data_blocks + hash_blocks || !f->rounds) {
716 ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
717 return -EINVAL;
718 }
719
720 /*
721 * Metadata is accessed through the hash device, so we require
722 * it to be large enough.
723 */
724 f->hash_blocks = f->blocks - v->data_blocks;
725 if (dm_bufio_get_device_size(v->bufio) < f->hash_blocks) {
726 ti->error = "Hash device is too small for "
727 DM_VERITY_OPT_FEC_BLOCKS;
728 return -E2BIG;
729 }
730
731 f->io_size = 1 << v->data_dev_block_bits;
732
733 f->bufio = dm_bufio_client_create(f->dev->bdev,
734 f->io_size,
735 1, 0, NULL, NULL, 0);
736 if (IS_ERR(f->bufio)) {
737 ti->error = "Cannot initialize FEC bufio client";
738 return PTR_ERR(f->bufio);
739 }
740
741 dm_bufio_set_sector_offset(f->bufio, f->start << (v->data_dev_block_bits - SECTOR_SHIFT));
742
743 fec_blocks = div64_u64(f->rounds * f->roots, v->fec->roots << SECTOR_SHIFT);
744 if (dm_bufio_get_device_size(f->bufio) < fec_blocks) {
745 ti->error = "FEC device is too small";
746 return -E2BIG;
747 }
748
749 f->data_bufio = dm_bufio_client_create(v->data_dev->bdev,
750 1 << v->data_dev_block_bits,
751 1, 0, NULL, NULL, 0);
752 if (IS_ERR(f->data_bufio)) {
753 ti->error = "Cannot initialize FEC data bufio client";
754 return PTR_ERR(f->data_bufio);
755 }
756
757 if (dm_bufio_get_device_size(f->data_bufio) < v->data_blocks) {
758 ti->error = "Data device is too small";
759 return -E2BIG;
760 }
761
762 /* Preallocate an rs_control structure for each worker thread */
763 ret = mempool_init(&f->rs_pool, num_online_cpus(), fec_rs_alloc,
764 fec_rs_free, (void *) v);
765 if (ret) {
766 ti->error = "Cannot allocate RS pool";
767 return ret;
768 }
769
770 f->cache = kmem_cache_create("dm_verity_fec_buffers",
771 f->rsn << DM_VERITY_FEC_BUF_RS_BITS,
772 0, 0, NULL);
773 if (!f->cache) {
774 ti->error = "Cannot create FEC buffer cache";
775 return -ENOMEM;
776 }
777
778 /* Preallocate DM_VERITY_FEC_BUF_PREALLOC buffers for each thread */
779 ret = mempool_init_slab_pool(&f->prealloc_pool, num_online_cpus() *
780 DM_VERITY_FEC_BUF_PREALLOC,
781 f->cache);
782 if (ret) {
783 ti->error = "Cannot allocate FEC buffer prealloc pool";
784 return ret;
785 }
786
787 ret = mempool_init_slab_pool(&f->extra_pool, 0, f->cache);
788 if (ret) {
789 ti->error = "Cannot allocate FEC buffer extra pool";
790 return ret;
791 }
792
793 /* Preallocate an output buffer for each thread */
794 ret = mempool_init_kmalloc_pool(&f->output_pool, num_online_cpus(),
795 1 << v->data_dev_block_bits);
796 if (ret) {
797 ti->error = "Cannot allocate FEC output pool";
798 return ret;
799 }
800
801 /* Reserve space for our per-bio data */
802 ti->per_io_data_size += sizeof(struct dm_verity_fec_io);
803
804 return 0;
805}
1/*
2 * Copyright (C) 2015 Google, Inc.
3 *
4 * Author: Sami Tolvanen <samitolvanen@google.com>
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License as published by the Free
8 * Software Foundation; either version 2 of the License, or (at your option)
9 * any later version.
10 */
11
12#include "dm-verity-fec.h"
13#include <linux/math64.h>
14
15#define DM_MSG_PREFIX "verity-fec"
16
17/*
18 * If error correction has been configured, returns true.
19 */
20bool verity_fec_is_enabled(struct dm_verity *v)
21{
22 return v->fec && v->fec->dev;
23}
24
25/*
26 * Return a pointer to dm_verity_fec_io after dm_verity_io and its variable
27 * length fields.
28 */
29static inline struct dm_verity_fec_io *fec_io(struct dm_verity_io *io)
30{
31 return (struct dm_verity_fec_io *) verity_io_digest_end(io->v, io);
32}
33
34/*
35 * Return an interleaved offset for a byte in RS block.
36 */
37static inline u64 fec_interleave(struct dm_verity *v, u64 offset)
38{
39 u32 mod;
40
41 mod = do_div(offset, v->fec->rsn);
42 return offset + mod * (v->fec->rounds << v->data_dev_block_bits);
43}
44
45/*
46 * Decode an RS block using Reed-Solomon.
47 */
48static int fec_decode_rs8(struct dm_verity *v, struct dm_verity_fec_io *fio,
49 u8 *data, u8 *fec, int neras)
50{
51 int i;
52 uint16_t par[DM_VERITY_FEC_RSM - DM_VERITY_FEC_MIN_RSN];
53
54 for (i = 0; i < v->fec->roots; i++)
55 par[i] = fec[i];
56
57 return decode_rs8(fio->rs, data, par, v->fec->rsn, NULL, neras,
58 fio->erasures, 0, NULL);
59}
60
61/*
62 * Read error-correcting codes for the requested RS block. Returns a pointer
63 * to the data block. Caller is responsible for releasing buf.
64 */
65static u8 *fec_read_parity(struct dm_verity *v, u64 rsb, int index,
66 unsigned *offset, struct dm_buffer **buf)
67{
68 u64 position, block;
69 u8 *res;
70
71 position = (index + rsb) * v->fec->roots;
72 block = position >> v->data_dev_block_bits;
73 *offset = (unsigned)(position - (block << v->data_dev_block_bits));
74
75 res = dm_bufio_read(v->fec->bufio, v->fec->start + block, buf);
76 if (unlikely(IS_ERR(res))) {
77 DMERR("%s: FEC %llu: parity read failed (block %llu): %ld",
78 v->data_dev->name, (unsigned long long)rsb,
79 (unsigned long long)(v->fec->start + block),
80 PTR_ERR(res));
81 *buf = NULL;
82 }
83
84 return res;
85}
86
87/* Loop over each preallocated buffer slot. */
88#define fec_for_each_prealloc_buffer(__i) \
89 for (__i = 0; __i < DM_VERITY_FEC_BUF_PREALLOC; __i++)
90
91/* Loop over each extra buffer slot. */
92#define fec_for_each_extra_buffer(io, __i) \
93 for (__i = DM_VERITY_FEC_BUF_PREALLOC; __i < DM_VERITY_FEC_BUF_MAX; __i++)
94
95/* Loop over each allocated buffer. */
96#define fec_for_each_buffer(io, __i) \
97 for (__i = 0; __i < (io)->nbufs; __i++)
98
99/* Loop over each RS block in each allocated buffer. */
100#define fec_for_each_buffer_rs_block(io, __i, __j) \
101 fec_for_each_buffer(io, __i) \
102 for (__j = 0; __j < 1 << DM_VERITY_FEC_BUF_RS_BITS; __j++)
103
104/*
105 * Return a pointer to the current RS block when called inside
106 * fec_for_each_buffer_rs_block.
107 */
108static inline u8 *fec_buffer_rs_block(struct dm_verity *v,
109 struct dm_verity_fec_io *fio,
110 unsigned i, unsigned j)
111{
112 return &fio->bufs[i][j * v->fec->rsn];
113}
114
115/*
116 * Return an index to the current RS block when called inside
117 * fec_for_each_buffer_rs_block.
118 */
119static inline unsigned fec_buffer_rs_index(unsigned i, unsigned j)
120{
121 return (i << DM_VERITY_FEC_BUF_RS_BITS) + j;
122}
123
124/*
125 * Decode all RS blocks from buffers and copy corrected bytes into fio->output
126 * starting from block_offset.
127 */
128static int fec_decode_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio,
129 u64 rsb, int byte_index, unsigned block_offset,
130 int neras)
131{
132 int r, corrected = 0, res;
133 struct dm_buffer *buf;
134 unsigned n, i, offset;
135 u8 *par, *block;
136
137 par = fec_read_parity(v, rsb, block_offset, &offset, &buf);
138 if (IS_ERR(par))
139 return PTR_ERR(par);
140
141 /*
142 * Decode the RS blocks we have in bufs. Each RS block results in
143 * one corrected target byte and consumes fec->roots parity bytes.
144 */
145 fec_for_each_buffer_rs_block(fio, n, i) {
146 block = fec_buffer_rs_block(v, fio, n, i);
147 res = fec_decode_rs8(v, fio, block, &par[offset], neras);
148 if (res < 0) {
149 r = res;
150 goto error;
151 }
152
153 corrected += res;
154 fio->output[block_offset] = block[byte_index];
155
156 block_offset++;
157 if (block_offset >= 1 << v->data_dev_block_bits)
158 goto done;
159
160 /* read the next block when we run out of parity bytes */
161 offset += v->fec->roots;
162 if (offset >= 1 << v->data_dev_block_bits) {
163 dm_bufio_release(buf);
164
165 par = fec_read_parity(v, rsb, block_offset, &offset, &buf);
166 if (unlikely(IS_ERR(par)))
167 return PTR_ERR(par);
168 }
169 }
170done:
171 r = corrected;
172error:
173 dm_bufio_release(buf);
174
175 if (r < 0 && neras)
176 DMERR_LIMIT("%s: FEC %llu: failed to correct: %d",
177 v->data_dev->name, (unsigned long long)rsb, r);
178 else if (r > 0)
179 DMWARN_LIMIT("%s: FEC %llu: corrected %d errors",
180 v->data_dev->name, (unsigned long long)rsb, r);
181
182 return r;
183}
184
185/*
186 * Locate data block erasures using verity hashes.
187 */
188static int fec_is_erasure(struct dm_verity *v, struct dm_verity_io *io,
189 u8 *want_digest, u8 *data)
190{
191 if (unlikely(verity_hash(v, verity_io_hash_desc(v, io),
192 data, 1 << v->data_dev_block_bits,
193 verity_io_real_digest(v, io))))
194 return 0;
195
196 return memcmp(verity_io_real_digest(v, io), want_digest,
197 v->digest_size) != 0;
198}
199
200/*
201 * Read data blocks that are part of the RS block and deinterleave as much as
202 * fits into buffers. Check for erasure locations if @neras is non-NULL.
203 */
204static int fec_read_bufs(struct dm_verity *v, struct dm_verity_io *io,
205 u64 rsb, u64 target, unsigned block_offset,
206 int *neras)
207{
208 bool is_zero;
209 int i, j, target_index = -1;
210 struct dm_buffer *buf;
211 struct dm_bufio_client *bufio;
212 struct dm_verity_fec_io *fio = fec_io(io);
213 u64 block, ileaved;
214 u8 *bbuf, *rs_block;
215 u8 want_digest[v->digest_size];
216 unsigned n, k;
217
218 if (neras)
219 *neras = 0;
220
221 /*
222 * read each of the rsn data blocks that are part of the RS block, and
223 * interleave contents to available bufs
224 */
225 for (i = 0; i < v->fec->rsn; i++) {
226 ileaved = fec_interleave(v, rsb * v->fec->rsn + i);
227
228 /*
229 * target is the data block we want to correct, target_index is
230 * the index of this block within the rsn RS blocks
231 */
232 if (ileaved == target)
233 target_index = i;
234
235 block = ileaved >> v->data_dev_block_bits;
236 bufio = v->fec->data_bufio;
237
238 if (block >= v->data_blocks) {
239 block -= v->data_blocks;
240
241 /*
242 * blocks outside the area were assumed to contain
243 * zeros when encoding data was generated
244 */
245 if (unlikely(block >= v->fec->hash_blocks))
246 continue;
247
248 block += v->hash_start;
249 bufio = v->bufio;
250 }
251
252 bbuf = dm_bufio_read(bufio, block, &buf);
253 if (unlikely(IS_ERR(bbuf))) {
254 DMWARN_LIMIT("%s: FEC %llu: read failed (%llu): %ld",
255 v->data_dev->name,
256 (unsigned long long)rsb,
257 (unsigned long long)block, PTR_ERR(bbuf));
258
259 /* assume the block is corrupted */
260 if (neras && *neras <= v->fec->roots)
261 fio->erasures[(*neras)++] = i;
262
263 continue;
264 }
265
266 /* locate erasures if the block is on the data device */
267 if (bufio == v->fec->data_bufio &&
268 verity_hash_for_block(v, io, block, want_digest,
269 &is_zero) == 0) {
270 /* skip known zero blocks entirely */
271 if (is_zero)
272 goto done;
273
274 /*
275 * skip if we have already found the theoretical
276 * maximum number (i.e. fec->roots) of erasures
277 */
278 if (neras && *neras <= v->fec->roots &&
279 fec_is_erasure(v, io, want_digest, bbuf))
280 fio->erasures[(*neras)++] = i;
281 }
282
283 /*
284 * deinterleave and copy the bytes that fit into bufs,
285 * starting from block_offset
286 */
287 fec_for_each_buffer_rs_block(fio, n, j) {
288 k = fec_buffer_rs_index(n, j) + block_offset;
289
290 if (k >= 1 << v->data_dev_block_bits)
291 goto done;
292
293 rs_block = fec_buffer_rs_block(v, fio, n, j);
294 rs_block[i] = bbuf[k];
295 }
296done:
297 dm_bufio_release(buf);
298 }
299
300 return target_index;
301}
302
303/*
304 * Allocate RS control structure and FEC buffers from preallocated mempools,
305 * and attempt to allocate as many extra buffers as available.
306 */
307static int fec_alloc_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
308{
309 unsigned n;
310
311 if (!fio->rs) {
312 fio->rs = mempool_alloc(v->fec->rs_pool, 0);
313 if (unlikely(!fio->rs)) {
314 DMERR("failed to allocate RS");
315 return -ENOMEM;
316 }
317 }
318
319 fec_for_each_prealloc_buffer(n) {
320 if (fio->bufs[n])
321 continue;
322
323 fio->bufs[n] = mempool_alloc(v->fec->prealloc_pool, GFP_NOIO);
324 if (unlikely(!fio->bufs[n])) {
325 DMERR("failed to allocate FEC buffer");
326 return -ENOMEM;
327 }
328 }
329
330 /* try to allocate the maximum number of buffers */
331 fec_for_each_extra_buffer(fio, n) {
332 if (fio->bufs[n])
333 continue;
334
335 fio->bufs[n] = mempool_alloc(v->fec->extra_pool, GFP_NOIO);
336 /* we can manage with even one buffer if necessary */
337 if (unlikely(!fio->bufs[n]))
338 break;
339 }
340 fio->nbufs = n;
341
342 if (!fio->output) {
343 fio->output = mempool_alloc(v->fec->output_pool, GFP_NOIO);
344
345 if (!fio->output) {
346 DMERR("failed to allocate FEC page");
347 return -ENOMEM;
348 }
349 }
350
351 return 0;
352}
353
354/*
355 * Initialize buffers and clear erasures. fec_read_bufs() assumes buffers are
356 * zeroed before deinterleaving.
357 */
358static void fec_init_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
359{
360 unsigned n;
361
362 fec_for_each_buffer(fio, n)
363 memset(fio->bufs[n], 0, v->fec->rsn << DM_VERITY_FEC_BUF_RS_BITS);
364
365 memset(fio->erasures, 0, sizeof(fio->erasures));
366}
367
368/*
369 * Decode all RS blocks in a single data block and return the target block
370 * (indicated by @offset) in fio->output. If @use_erasures is non-zero, uses
371 * hashes to locate erasures.
372 */
373static int fec_decode_rsb(struct dm_verity *v, struct dm_verity_io *io,
374 struct dm_verity_fec_io *fio, u64 rsb, u64 offset,
375 bool use_erasures)
376{
377 int r, neras = 0;
378 unsigned pos;
379
380 r = fec_alloc_bufs(v, fio);
381 if (unlikely(r < 0))
382 return r;
383
384 for (pos = 0; pos < 1 << v->data_dev_block_bits; ) {
385 fec_init_bufs(v, fio);
386
387 r = fec_read_bufs(v, io, rsb, offset, pos,
388 use_erasures ? &neras : NULL);
389 if (unlikely(r < 0))
390 return r;
391
392 r = fec_decode_bufs(v, fio, rsb, r, pos, neras);
393 if (r < 0)
394 return r;
395
396 pos += fio->nbufs << DM_VERITY_FEC_BUF_RS_BITS;
397 }
398
399 /* Always re-validate the corrected block against the expected hash */
400 r = verity_hash(v, verity_io_hash_desc(v, io), fio->output,
401 1 << v->data_dev_block_bits,
402 verity_io_real_digest(v, io));
403 if (unlikely(r < 0))
404 return r;
405
406 if (memcmp(verity_io_real_digest(v, io), verity_io_want_digest(v, io),
407 v->digest_size)) {
408 DMERR_LIMIT("%s: FEC %llu: failed to correct (%d erasures)",
409 v->data_dev->name, (unsigned long long)rsb, neras);
410 return -EILSEQ;
411 }
412
413 return 0;
414}
415
416static int fec_bv_copy(struct dm_verity *v, struct dm_verity_io *io, u8 *data,
417 size_t len)
418{
419 struct dm_verity_fec_io *fio = fec_io(io);
420
421 memcpy(data, &fio->output[fio->output_pos], len);
422 fio->output_pos += len;
423
424 return 0;
425}
426
427/*
428 * Correct errors in a block. Copies corrected block to dest if non-NULL,
429 * otherwise to a bio_vec starting from iter.
430 */
431int verity_fec_decode(struct dm_verity *v, struct dm_verity_io *io,
432 enum verity_block_type type, sector_t block, u8 *dest,
433 struct bvec_iter *iter)
434{
435 int r;
436 struct dm_verity_fec_io *fio = fec_io(io);
437 u64 offset, res, rsb;
438
439 if (!verity_fec_is_enabled(v))
440 return -EOPNOTSUPP;
441
442 if (fio->level >= DM_VERITY_FEC_MAX_RECURSION) {
443 DMWARN_LIMIT("%s: FEC: recursion too deep", v->data_dev->name);
444 return -EIO;
445 }
446
447 fio->level++;
448
449 if (type == DM_VERITY_BLOCK_TYPE_METADATA)
450 block += v->data_blocks;
451
452 /*
453 * For RS(M, N), the continuous FEC data is divided into blocks of N
454 * bytes. Since block size may not be divisible by N, the last block
455 * is zero padded when decoding.
456 *
457 * Each byte of the block is covered by a different RS(M, N) code,
458 * and each code is interleaved over N blocks to make it less likely
459 * that bursty corruption will leave us in unrecoverable state.
460 */
461
462 offset = block << v->data_dev_block_bits;
463 res = div64_u64(offset, v->fec->rounds << v->data_dev_block_bits);
464
465 /*
466 * The base RS block we can feed to the interleaver to find out all
467 * blocks required for decoding.
468 */
469 rsb = offset - res * (v->fec->rounds << v->data_dev_block_bits);
470
471 /*
472 * Locating erasures is slow, so attempt to recover the block without
473 * them first. Do a second attempt with erasures if the corruption is
474 * bad enough.
475 */
476 r = fec_decode_rsb(v, io, fio, rsb, offset, false);
477 if (r < 0) {
478 r = fec_decode_rsb(v, io, fio, rsb, offset, true);
479 if (r < 0)
480 goto done;
481 }
482
483 if (dest)
484 memcpy(dest, fio->output, 1 << v->data_dev_block_bits);
485 else if (iter) {
486 fio->output_pos = 0;
487 r = verity_for_bv_block(v, io, iter, fec_bv_copy);
488 }
489
490done:
491 fio->level--;
492 return r;
493}
494
495/*
496 * Clean up per-bio data.
497 */
498void verity_fec_finish_io(struct dm_verity_io *io)
499{
500 unsigned n;
501 struct dm_verity_fec *f = io->v->fec;
502 struct dm_verity_fec_io *fio = fec_io(io);
503
504 if (!verity_fec_is_enabled(io->v))
505 return;
506
507 mempool_free(fio->rs, f->rs_pool);
508
509 fec_for_each_prealloc_buffer(n)
510 mempool_free(fio->bufs[n], f->prealloc_pool);
511
512 fec_for_each_extra_buffer(fio, n)
513 mempool_free(fio->bufs[n], f->extra_pool);
514
515 mempool_free(fio->output, f->output_pool);
516}
517
518/*
519 * Initialize per-bio data.
520 */
521void verity_fec_init_io(struct dm_verity_io *io)
522{
523 struct dm_verity_fec_io *fio = fec_io(io);
524
525 if (!verity_fec_is_enabled(io->v))
526 return;
527
528 fio->rs = NULL;
529 memset(fio->bufs, 0, sizeof(fio->bufs));
530 fio->nbufs = 0;
531 fio->output = NULL;
532 fio->level = 0;
533}
534
535/*
536 * Append feature arguments and values to the status table.
537 */
538unsigned verity_fec_status_table(struct dm_verity *v, unsigned sz,
539 char *result, unsigned maxlen)
540{
541 if (!verity_fec_is_enabled(v))
542 return sz;
543
544 DMEMIT(" " DM_VERITY_OPT_FEC_DEV " %s "
545 DM_VERITY_OPT_FEC_BLOCKS " %llu "
546 DM_VERITY_OPT_FEC_START " %llu "
547 DM_VERITY_OPT_FEC_ROOTS " %d",
548 v->fec->dev->name,
549 (unsigned long long)v->fec->blocks,
550 (unsigned long long)v->fec->start,
551 v->fec->roots);
552
553 return sz;
554}
555
556void verity_fec_dtr(struct dm_verity *v)
557{
558 struct dm_verity_fec *f = v->fec;
559
560 if (!verity_fec_is_enabled(v))
561 goto out;
562
563 mempool_destroy(f->rs_pool);
564 mempool_destroy(f->prealloc_pool);
565 mempool_destroy(f->extra_pool);
566 kmem_cache_destroy(f->cache);
567
568 if (f->data_bufio)
569 dm_bufio_client_destroy(f->data_bufio);
570 if (f->bufio)
571 dm_bufio_client_destroy(f->bufio);
572
573 if (f->dev)
574 dm_put_device(v->ti, f->dev);
575out:
576 kfree(f);
577 v->fec = NULL;
578}
579
580static void *fec_rs_alloc(gfp_t gfp_mask, void *pool_data)
581{
582 struct dm_verity *v = (struct dm_verity *)pool_data;
583
584 return init_rs(8, 0x11d, 0, 1, v->fec->roots);
585}
586
587static void fec_rs_free(void *element, void *pool_data)
588{
589 struct rs_control *rs = (struct rs_control *)element;
590
591 if (rs)
592 free_rs(rs);
593}
594
595bool verity_is_fec_opt_arg(const char *arg_name)
596{
597 return (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV) ||
598 !strcasecmp(arg_name, DM_VERITY_OPT_FEC_BLOCKS) ||
599 !strcasecmp(arg_name, DM_VERITY_OPT_FEC_START) ||
600 !strcasecmp(arg_name, DM_VERITY_OPT_FEC_ROOTS));
601}
602
603int verity_fec_parse_opt_args(struct dm_arg_set *as, struct dm_verity *v,
604 unsigned *argc, const char *arg_name)
605{
606 int r;
607 struct dm_target *ti = v->ti;
608 const char *arg_value;
609 unsigned long long num_ll;
610 unsigned char num_c;
611 char dummy;
612
613 if (!*argc) {
614 ti->error = "FEC feature arguments require a value";
615 return -EINVAL;
616 }
617
618 arg_value = dm_shift_arg(as);
619 (*argc)--;
620
621 if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV)) {
622 r = dm_get_device(ti, arg_value, FMODE_READ, &v->fec->dev);
623 if (r) {
624 ti->error = "FEC device lookup failed";
625 return r;
626 }
627
628 } else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_BLOCKS)) {
629 if (sscanf(arg_value, "%llu%c", &num_ll, &dummy) != 1 ||
630 ((sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT))
631 >> (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll)) {
632 ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
633 return -EINVAL;
634 }
635 v->fec->blocks = num_ll;
636
637 } else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_START)) {
638 if (sscanf(arg_value, "%llu%c", &num_ll, &dummy) != 1 ||
639 ((sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT)) >>
640 (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll)) {
641 ti->error = "Invalid " DM_VERITY_OPT_FEC_START;
642 return -EINVAL;
643 }
644 v->fec->start = num_ll;
645
646 } else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_ROOTS)) {
647 if (sscanf(arg_value, "%hhu%c", &num_c, &dummy) != 1 || !num_c ||
648 num_c < (DM_VERITY_FEC_RSM - DM_VERITY_FEC_MAX_RSN) ||
649 num_c > (DM_VERITY_FEC_RSM - DM_VERITY_FEC_MIN_RSN)) {
650 ti->error = "Invalid " DM_VERITY_OPT_FEC_ROOTS;
651 return -EINVAL;
652 }
653 v->fec->roots = num_c;
654
655 } else {
656 ti->error = "Unrecognized verity FEC feature request";
657 return -EINVAL;
658 }
659
660 return 0;
661}
662
663/*
664 * Allocate dm_verity_fec for v->fec. Must be called before verity_fec_ctr.
665 */
666int verity_fec_ctr_alloc(struct dm_verity *v)
667{
668 struct dm_verity_fec *f;
669
670 f = kzalloc(sizeof(struct dm_verity_fec), GFP_KERNEL);
671 if (!f) {
672 v->ti->error = "Cannot allocate FEC structure";
673 return -ENOMEM;
674 }
675 v->fec = f;
676
677 return 0;
678}
679
680/*
681 * Validate arguments and preallocate memory. Must be called after arguments
682 * have been parsed using verity_fec_parse_opt_args.
683 */
684int verity_fec_ctr(struct dm_verity *v)
685{
686 struct dm_verity_fec *f = v->fec;
687 struct dm_target *ti = v->ti;
688 u64 hash_blocks;
689
690 if (!verity_fec_is_enabled(v)) {
691 verity_fec_dtr(v);
692 return 0;
693 }
694
695 /*
696 * FEC is computed over data blocks, possible metadata, and
697 * hash blocks. In other words, FEC covers total of fec_blocks
698 * blocks consisting of the following:
699 *
700 * data blocks | hash blocks | metadata (optional)
701 *
702 * We allow metadata after hash blocks to support a use case
703 * where all data is stored on the same device and FEC covers
704 * the entire area.
705 *
706 * If metadata is included, we require it to be available on the
707 * hash device after the hash blocks.
708 */
709
710 hash_blocks = v->hash_blocks - v->hash_start;
711
712 /*
713 * Require matching block sizes for data and hash devices for
714 * simplicity.
715 */
716 if (v->data_dev_block_bits != v->hash_dev_block_bits) {
717 ti->error = "Block sizes must match to use FEC";
718 return -EINVAL;
719 }
720
721 if (!f->roots) {
722 ti->error = "Missing " DM_VERITY_OPT_FEC_ROOTS;
723 return -EINVAL;
724 }
725 f->rsn = DM_VERITY_FEC_RSM - f->roots;
726
727 if (!f->blocks) {
728 ti->error = "Missing " DM_VERITY_OPT_FEC_BLOCKS;
729 return -EINVAL;
730 }
731
732 f->rounds = f->blocks;
733 if (sector_div(f->rounds, f->rsn))
734 f->rounds++;
735
736 /*
737 * Due to optional metadata, f->blocks can be larger than
738 * data_blocks and hash_blocks combined.
739 */
740 if (f->blocks < v->data_blocks + hash_blocks || !f->rounds) {
741 ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
742 return -EINVAL;
743 }
744
745 /*
746 * Metadata is accessed through the hash device, so we require
747 * it to be large enough.
748 */
749 f->hash_blocks = f->blocks - v->data_blocks;
750 if (dm_bufio_get_device_size(v->bufio) < f->hash_blocks) {
751 ti->error = "Hash device is too small for "
752 DM_VERITY_OPT_FEC_BLOCKS;
753 return -E2BIG;
754 }
755
756 f->bufio = dm_bufio_client_create(f->dev->bdev,
757 1 << v->data_dev_block_bits,
758 1, 0, NULL, NULL);
759 if (IS_ERR(f->bufio)) {
760 ti->error = "Cannot initialize FEC bufio client";
761 return PTR_ERR(f->bufio);
762 }
763
764 if (dm_bufio_get_device_size(f->bufio) <
765 ((f->start + f->rounds * f->roots) >> v->data_dev_block_bits)) {
766 ti->error = "FEC device is too small";
767 return -E2BIG;
768 }
769
770 f->data_bufio = dm_bufio_client_create(v->data_dev->bdev,
771 1 << v->data_dev_block_bits,
772 1, 0, NULL, NULL);
773 if (IS_ERR(f->data_bufio)) {
774 ti->error = "Cannot initialize FEC data bufio client";
775 return PTR_ERR(f->data_bufio);
776 }
777
778 if (dm_bufio_get_device_size(f->data_bufio) < v->data_blocks) {
779 ti->error = "Data device is too small";
780 return -E2BIG;
781 }
782
783 /* Preallocate an rs_control structure for each worker thread */
784 f->rs_pool = mempool_create(num_online_cpus(), fec_rs_alloc,
785 fec_rs_free, (void *) v);
786 if (!f->rs_pool) {
787 ti->error = "Cannot allocate RS pool";
788 return -ENOMEM;
789 }
790
791 f->cache = kmem_cache_create("dm_verity_fec_buffers",
792 f->rsn << DM_VERITY_FEC_BUF_RS_BITS,
793 0, 0, NULL);
794 if (!f->cache) {
795 ti->error = "Cannot create FEC buffer cache";
796 return -ENOMEM;
797 }
798
799 /* Preallocate DM_VERITY_FEC_BUF_PREALLOC buffers for each thread */
800 f->prealloc_pool = mempool_create_slab_pool(num_online_cpus() *
801 DM_VERITY_FEC_BUF_PREALLOC,
802 f->cache);
803 if (!f->prealloc_pool) {
804 ti->error = "Cannot allocate FEC buffer prealloc pool";
805 return -ENOMEM;
806 }
807
808 f->extra_pool = mempool_create_slab_pool(0, f->cache);
809 if (!f->extra_pool) {
810 ti->error = "Cannot allocate FEC buffer extra pool";
811 return -ENOMEM;
812 }
813
814 /* Preallocate an output buffer for each thread */
815 f->output_pool = mempool_create_kmalloc_pool(num_online_cpus(),
816 1 << v->data_dev_block_bits);
817 if (!f->output_pool) {
818 ti->error = "Cannot allocate FEC output pool";
819 return -ENOMEM;
820 }
821
822 /* Reserve space for our per-bio data */
823 ti->per_io_data_size += sizeof(struct dm_verity_fec_io);
824
825 return 0;
826}