xref: /linux/drivers/md/dm-verity-fec.c (revision 257ca10c7317d4a424e48bb95d14ca53a1f1dd6f)
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  */
16 bool 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  */
25 static 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  */
34 static 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  * Decode an RS block using Reed-Solomon.
44  */
45 static int fec_decode_rs8(struct dm_verity *v, struct dm_verity_fec_io *fio,
46 			  u8 *data, u8 *fec, int neras)
47 {
48 	int i;
49 	uint16_t par[DM_VERITY_FEC_RSM - DM_VERITY_FEC_MIN_RSN];
50 
51 	for (i = 0; i < v->fec->roots; i++)
52 		par[i] = fec[i];
53 
54 	return decode_rs8(fio->rs, data, par, v->fec->rsn, NULL, neras,
55 			  fio->erasures, 0, NULL);
56 }
57 
58 /*
59  * Read error-correcting codes for the requested RS block. Returns a pointer
60  * to the data block. Caller is responsible for releasing buf.
61  */
62 static u8 *fec_read_parity(struct dm_verity *v, u64 rsb, int index,
63 			   unsigned int *offset, struct dm_buffer **buf)
64 {
65 	u64 position, block, rem;
66 	u8 *res;
67 
68 	position = (index + rsb) * v->fec->roots;
69 	block = div64_u64_rem(position, v->fec->io_size, &rem);
70 	*offset = (unsigned int)rem;
71 
72 	res = dm_bufio_read(v->fec->bufio, block, buf);
73 	if (IS_ERR(res)) {
74 		DMERR("%s: FEC %llu: parity read failed (block %llu): %ld",
75 		      v->data_dev->name, (unsigned long long)rsb,
76 		      (unsigned long long)block, PTR_ERR(res));
77 		*buf = NULL;
78 	}
79 
80 	return res;
81 }
82 
83 /* Loop over each preallocated buffer slot. */
84 #define fec_for_each_prealloc_buffer(__i) \
85 	for (__i = 0; __i < DM_VERITY_FEC_BUF_PREALLOC; __i++)
86 
87 /* Loop over each extra buffer slot. */
88 #define fec_for_each_extra_buffer(io, __i) \
89 	for (__i = DM_VERITY_FEC_BUF_PREALLOC; __i < DM_VERITY_FEC_BUF_MAX; __i++)
90 
91 /* Loop over each allocated buffer. */
92 #define fec_for_each_buffer(io, __i) \
93 	for (__i = 0; __i < (io)->nbufs; __i++)
94 
95 /* Loop over each RS block in each allocated buffer. */
96 #define fec_for_each_buffer_rs_block(io, __i, __j) \
97 	fec_for_each_buffer(io, __i) \
98 		for (__j = 0; __j < 1 << DM_VERITY_FEC_BUF_RS_BITS; __j++)
99 
100 /*
101  * Return a pointer to the current RS block when called inside
102  * fec_for_each_buffer_rs_block.
103  */
104 static inline u8 *fec_buffer_rs_block(struct dm_verity *v,
105 				      struct dm_verity_fec_io *fio,
106 				      unsigned int i, unsigned int j)
107 {
108 	return &fio->bufs[i][j * v->fec->rsn];
109 }
110 
111 /*
112  * Return an index to the current RS block when called inside
113  * fec_for_each_buffer_rs_block.
114  */
115 static inline unsigned int fec_buffer_rs_index(unsigned int i, unsigned int j)
116 {
117 	return (i << DM_VERITY_FEC_BUF_RS_BITS) + j;
118 }
119 
120 /*
121  * Decode all RS blocks from buffers and copy corrected bytes into fio->output
122  * starting from block_offset.
123  */
124 static int fec_decode_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio,
125 			   u64 rsb, int byte_index, unsigned int block_offset,
126 			   int neras)
127 {
128 	int r, corrected = 0, res;
129 	struct dm_buffer *buf;
130 	unsigned int n, i, offset;
131 	u8 *par, *block;
132 
133 	par = fec_read_parity(v, rsb, block_offset, &offset, &buf);
134 	if (IS_ERR(par))
135 		return PTR_ERR(par);
136 
137 	/*
138 	 * Decode the RS blocks we have in bufs. Each RS block results in
139 	 * one corrected target byte and consumes fec->roots parity bytes.
140 	 */
141 	fec_for_each_buffer_rs_block(fio, n, i) {
142 		block = fec_buffer_rs_block(v, fio, n, i);
143 		res = fec_decode_rs8(v, fio, block, &par[offset], neras);
144 		if (res < 0) {
145 			r = res;
146 			goto error;
147 		}
148 
149 		corrected += res;
150 		fio->output[block_offset] = block[byte_index];
151 
152 		block_offset++;
153 		if (block_offset >= 1 << v->data_dev_block_bits)
154 			goto done;
155 
156 		/* read the next block when we run out of parity bytes */
157 		offset += v->fec->roots;
158 		if (offset >= v->fec->io_size) {
159 			dm_bufio_release(buf);
160 
161 			par = fec_read_parity(v, rsb, block_offset, &offset, &buf);
162 			if (IS_ERR(par))
163 				return PTR_ERR(par);
164 		}
165 	}
166 done:
167 	r = corrected;
168 error:
169 	dm_bufio_release(buf);
170 
171 	if (r < 0 && neras)
172 		DMERR_LIMIT("%s: FEC %llu: failed to correct: %d",
173 			    v->data_dev->name, (unsigned long long)rsb, r);
174 	else if (r > 0)
175 		DMWARN_LIMIT("%s: FEC %llu: corrected %d errors",
176 			     v->data_dev->name, (unsigned long long)rsb, r);
177 
178 	return r;
179 }
180 
181 /*
182  * Locate data block erasures using verity hashes.
183  */
184 static int fec_is_erasure(struct dm_verity *v, struct dm_verity_io *io,
185 			  u8 *want_digest, u8 *data)
186 {
187 	if (unlikely(verity_hash(v, verity_io_hash_req(v, io),
188 				 data, 1 << v->data_dev_block_bits,
189 				 verity_io_real_digest(v, io), true)))
190 		return 0;
191 
192 	return memcmp(verity_io_real_digest(v, io), want_digest,
193 		      v->digest_size) != 0;
194 }
195 
196 /*
197  * Read data blocks that are part of the RS block and deinterleave as much as
198  * fits into buffers. Check for erasure locations if @neras is non-NULL.
199  */
200 static int fec_read_bufs(struct dm_verity *v, struct dm_verity_io *io,
201 			 u64 rsb, u64 target, unsigned int block_offset,
202 			 int *neras)
203 {
204 	bool is_zero;
205 	int i, j, target_index = -1;
206 	struct dm_buffer *buf;
207 	struct dm_bufio_client *bufio;
208 	struct dm_verity_fec_io *fio = fec_io(io);
209 	u64 block, ileaved;
210 	u8 *bbuf, *rs_block;
211 	u8 want_digest[HASH_MAX_DIGESTSIZE];
212 	unsigned int n, k;
213 
214 	if (neras)
215 		*neras = 0;
216 
217 	if (WARN_ON(v->digest_size > sizeof(want_digest)))
218 		return -EINVAL;
219 
220 	/*
221 	 * read each of the rsn data blocks that are part of the RS block, and
222 	 * interleave contents to available bufs
223 	 */
224 	for (i = 0; i < v->fec->rsn; i++) {
225 		ileaved = fec_interleave(v, rsb * v->fec->rsn + i);
226 
227 		/*
228 		 * target is the data block we want to correct, target_index is
229 		 * the index of this block within the rsn RS blocks
230 		 */
231 		if (ileaved == target)
232 			target_index = i;
233 
234 		block = ileaved >> v->data_dev_block_bits;
235 		bufio = v->fec->data_bufio;
236 
237 		if (block >= v->data_blocks) {
238 			block -= v->data_blocks;
239 
240 			/*
241 			 * blocks outside the area were assumed to contain
242 			 * zeros when encoding data was generated
243 			 */
244 			if (unlikely(block >= v->fec->hash_blocks))
245 				continue;
246 
247 			block += v->hash_start;
248 			bufio = v->bufio;
249 		}
250 
251 		bbuf = dm_bufio_read(bufio, block, &buf);
252 		if (IS_ERR(bbuf)) {
253 			DMWARN_LIMIT("%s: FEC %llu: read failed (%llu): %ld",
254 				     v->data_dev->name,
255 				     (unsigned long long)rsb,
256 				     (unsigned long long)block, PTR_ERR(bbuf));
257 
258 			/* assume the block is corrupted */
259 			if (neras && *neras <= v->fec->roots)
260 				fio->erasures[(*neras)++] = i;
261 
262 			continue;
263 		}
264 
265 		/* locate erasures if the block is on the data device */
266 		if (bufio == v->fec->data_bufio &&
267 		    verity_hash_for_block(v, io, block, want_digest,
268 					  &is_zero) == 0) {
269 			/* skip known zero blocks entirely */
270 			if (is_zero)
271 				goto done;
272 
273 			/*
274 			 * skip if we have already found the theoretical
275 			 * maximum number (i.e. fec->roots) of erasures
276 			 */
277 			if (neras && *neras <= v->fec->roots &&
278 			    fec_is_erasure(v, io, want_digest, bbuf))
279 				fio->erasures[(*neras)++] = i;
280 		}
281 
282 		/*
283 		 * deinterleave and copy the bytes that fit into bufs,
284 		 * starting from block_offset
285 		 */
286 		fec_for_each_buffer_rs_block(fio, n, j) {
287 			k = fec_buffer_rs_index(n, j) + block_offset;
288 
289 			if (k >= 1 << v->data_dev_block_bits)
290 				goto done;
291 
292 			rs_block = fec_buffer_rs_block(v, fio, n, j);
293 			rs_block[i] = bbuf[k];
294 		}
295 done:
296 		dm_bufio_release(buf);
297 	}
298 
299 	return target_index;
300 }
301 
302 /*
303  * Allocate RS control structure and FEC buffers from preallocated mempools,
304  * and attempt to allocate as many extra buffers as available.
305  */
306 static int fec_alloc_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
307 {
308 	unsigned int n;
309 
310 	if (!fio->rs)
311 		fio->rs = mempool_alloc(&v->fec->rs_pool, GFP_NOIO);
312 
313 	fec_for_each_prealloc_buffer(n) {
314 		if (fio->bufs[n])
315 			continue;
316 
317 		fio->bufs[n] = mempool_alloc(&v->fec->prealloc_pool, GFP_NOWAIT);
318 		if (unlikely(!fio->bufs[n])) {
319 			DMERR("failed to allocate FEC buffer");
320 			return -ENOMEM;
321 		}
322 	}
323 
324 	/* try to allocate the maximum number of buffers */
325 	fec_for_each_extra_buffer(fio, n) {
326 		if (fio->bufs[n])
327 			continue;
328 
329 		fio->bufs[n] = mempool_alloc(&v->fec->extra_pool, GFP_NOWAIT);
330 		/* we can manage with even one buffer if necessary */
331 		if (unlikely(!fio->bufs[n]))
332 			break;
333 	}
334 	fio->nbufs = n;
335 
336 	if (!fio->output)
337 		fio->output = mempool_alloc(&v->fec->output_pool, GFP_NOIO);
338 
339 	return 0;
340 }
341 
342 /*
343  * Initialize buffers and clear erasures. fec_read_bufs() assumes buffers are
344  * zeroed before deinterleaving.
345  */
346 static void fec_init_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
347 {
348 	unsigned int n;
349 
350 	fec_for_each_buffer(fio, n)
351 		memset(fio->bufs[n], 0, v->fec->rsn << DM_VERITY_FEC_BUF_RS_BITS);
352 
353 	memset(fio->erasures, 0, sizeof(fio->erasures));
354 }
355 
356 /*
357  * Decode all RS blocks in a single data block and return the target block
358  * (indicated by @offset) in fio->output. If @use_erasures is non-zero, uses
359  * hashes to locate erasures.
360  */
361 static int fec_decode_rsb(struct dm_verity *v, struct dm_verity_io *io,
362 			  struct dm_verity_fec_io *fio, u64 rsb, u64 offset,
363 			  bool use_erasures)
364 {
365 	int r, neras = 0;
366 	unsigned int pos;
367 
368 	r = fec_alloc_bufs(v, fio);
369 	if (unlikely(r < 0))
370 		return r;
371 
372 	for (pos = 0; pos < 1 << v->data_dev_block_bits; ) {
373 		fec_init_bufs(v, fio);
374 
375 		r = fec_read_bufs(v, io, rsb, offset, pos,
376 				  use_erasures ? &neras : NULL);
377 		if (unlikely(r < 0))
378 			return r;
379 
380 		r = fec_decode_bufs(v, fio, rsb, r, pos, neras);
381 		if (r < 0)
382 			return r;
383 
384 		pos += fio->nbufs << DM_VERITY_FEC_BUF_RS_BITS;
385 	}
386 
387 	/* Always re-validate the corrected block against the expected hash */
388 	r = verity_hash(v, verity_io_hash_req(v, io), fio->output,
389 			1 << v->data_dev_block_bits,
390 			verity_io_real_digest(v, io), true);
391 	if (unlikely(r < 0))
392 		return r;
393 
394 	if (memcmp(verity_io_real_digest(v, io), verity_io_want_digest(v, io),
395 		   v->digest_size)) {
396 		DMERR_LIMIT("%s: FEC %llu: failed to correct (%d erasures)",
397 			    v->data_dev->name, (unsigned long long)rsb, neras);
398 		return -EILSEQ;
399 	}
400 
401 	return 0;
402 }
403 
404 static int fec_bv_copy(struct dm_verity *v, struct dm_verity_io *io, u8 *data,
405 		       size_t len)
406 {
407 	struct dm_verity_fec_io *fio = fec_io(io);
408 
409 	memcpy(data, &fio->output[fio->output_pos], len);
410 	fio->output_pos += len;
411 
412 	return 0;
413 }
414 
415 /*
416  * Correct errors in a block. Copies corrected block to dest if non-NULL,
417  * otherwise to a bio_vec starting from iter.
418  */
419 int verity_fec_decode(struct dm_verity *v, struct dm_verity_io *io,
420 		      enum verity_block_type type, sector_t block, u8 *dest,
421 		      struct bvec_iter *iter)
422 {
423 	int r;
424 	struct dm_verity_fec_io *fio = fec_io(io);
425 	u64 offset, res, rsb;
426 
427 	if (!verity_fec_is_enabled(v))
428 		return -EOPNOTSUPP;
429 
430 	if (fio->level >= DM_VERITY_FEC_MAX_RECURSION) {
431 		DMWARN_LIMIT("%s: FEC: recursion too deep", v->data_dev->name);
432 		return -EIO;
433 	}
434 
435 	fio->level++;
436 
437 	if (type == DM_VERITY_BLOCK_TYPE_METADATA)
438 		block = block - v->hash_start + v->data_blocks;
439 
440 	/*
441 	 * For RS(M, N), the continuous FEC data is divided into blocks of N
442 	 * bytes. Since block size may not be divisible by N, the last block
443 	 * is zero padded when decoding.
444 	 *
445 	 * Each byte of the block is covered by a different RS(M, N) code,
446 	 * and each code is interleaved over N blocks to make it less likely
447 	 * that bursty corruption will leave us in unrecoverable state.
448 	 */
449 
450 	offset = block << v->data_dev_block_bits;
451 	res = div64_u64(offset, v->fec->rounds << v->data_dev_block_bits);
452 
453 	/*
454 	 * The base RS block we can feed to the interleaver to find out all
455 	 * blocks required for decoding.
456 	 */
457 	rsb = offset - res * (v->fec->rounds << v->data_dev_block_bits);
458 
459 	/*
460 	 * Locating erasures is slow, so attempt to recover the block without
461 	 * them first. Do a second attempt with erasures if the corruption is
462 	 * bad enough.
463 	 */
464 	r = fec_decode_rsb(v, io, fio, rsb, offset, false);
465 	if (r < 0) {
466 		r = fec_decode_rsb(v, io, fio, rsb, offset, true);
467 		if (r < 0)
468 			goto done;
469 	}
470 
471 	if (dest)
472 		memcpy(dest, fio->output, 1 << v->data_dev_block_bits);
473 	else if (iter) {
474 		fio->output_pos = 0;
475 		r = verity_for_bv_block(v, io, iter, fec_bv_copy);
476 	}
477 
478 done:
479 	fio->level--;
480 	return r;
481 }
482 
483 /*
484  * Clean up per-bio data.
485  */
486 void verity_fec_finish_io(struct dm_verity_io *io)
487 {
488 	unsigned int n;
489 	struct dm_verity_fec *f = io->v->fec;
490 	struct dm_verity_fec_io *fio = fec_io(io);
491 
492 	if (!verity_fec_is_enabled(io->v))
493 		return;
494 
495 	mempool_free(fio->rs, &f->rs_pool);
496 
497 	fec_for_each_prealloc_buffer(n)
498 		mempool_free(fio->bufs[n], &f->prealloc_pool);
499 
500 	fec_for_each_extra_buffer(fio, n)
501 		mempool_free(fio->bufs[n], &f->extra_pool);
502 
503 	mempool_free(fio->output, &f->output_pool);
504 }
505 
506 /*
507  * Initialize per-bio data.
508  */
509 void verity_fec_init_io(struct dm_verity_io *io)
510 {
511 	struct dm_verity_fec_io *fio = fec_io(io);
512 
513 	if (!verity_fec_is_enabled(io->v))
514 		return;
515 
516 	fio->rs = NULL;
517 	memset(fio->bufs, 0, sizeof(fio->bufs));
518 	fio->nbufs = 0;
519 	fio->output = NULL;
520 	fio->level = 0;
521 }
522 
523 /*
524  * Append feature arguments and values to the status table.
525  */
526 unsigned int verity_fec_status_table(struct dm_verity *v, unsigned int sz,
527 				 char *result, unsigned int maxlen)
528 {
529 	if (!verity_fec_is_enabled(v))
530 		return sz;
531 
532 	DMEMIT(" " DM_VERITY_OPT_FEC_DEV " %s "
533 	       DM_VERITY_OPT_FEC_BLOCKS " %llu "
534 	       DM_VERITY_OPT_FEC_START " %llu "
535 	       DM_VERITY_OPT_FEC_ROOTS " %d",
536 	       v->fec->dev->name,
537 	       (unsigned long long)v->fec->blocks,
538 	       (unsigned long long)v->fec->start,
539 	       v->fec->roots);
540 
541 	return sz;
542 }
543 
544 void verity_fec_dtr(struct dm_verity *v)
545 {
546 	struct dm_verity_fec *f = v->fec;
547 
548 	if (!verity_fec_is_enabled(v))
549 		goto out;
550 
551 	mempool_exit(&f->rs_pool);
552 	mempool_exit(&f->prealloc_pool);
553 	mempool_exit(&f->extra_pool);
554 	mempool_exit(&f->output_pool);
555 	kmem_cache_destroy(f->cache);
556 
557 	if (f->data_bufio)
558 		dm_bufio_client_destroy(f->data_bufio);
559 	if (f->bufio)
560 		dm_bufio_client_destroy(f->bufio);
561 
562 	if (f->dev)
563 		dm_put_device(v->ti, f->dev);
564 out:
565 	kfree(f);
566 	v->fec = NULL;
567 }
568 
569 static void *fec_rs_alloc(gfp_t gfp_mask, void *pool_data)
570 {
571 	struct dm_verity *v = pool_data;
572 
573 	return init_rs_gfp(8, 0x11d, 0, 1, v->fec->roots, gfp_mask);
574 }
575 
576 static void fec_rs_free(void *element, void *pool_data)
577 {
578 	struct rs_control *rs = element;
579 
580 	if (rs)
581 		free_rs(rs);
582 }
583 
584 bool verity_is_fec_opt_arg(const char *arg_name)
585 {
586 	return (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV) ||
587 		!strcasecmp(arg_name, DM_VERITY_OPT_FEC_BLOCKS) ||
588 		!strcasecmp(arg_name, DM_VERITY_OPT_FEC_START) ||
589 		!strcasecmp(arg_name, DM_VERITY_OPT_FEC_ROOTS));
590 }
591 
592 int verity_fec_parse_opt_args(struct dm_arg_set *as, struct dm_verity *v,
593 			      unsigned int *argc, const char *arg_name)
594 {
595 	int r;
596 	struct dm_target *ti = v->ti;
597 	const char *arg_value;
598 	unsigned long long num_ll;
599 	unsigned char num_c;
600 	char dummy;
601 
602 	if (!*argc) {
603 		ti->error = "FEC feature arguments require a value";
604 		return -EINVAL;
605 	}
606 
607 	arg_value = dm_shift_arg(as);
608 	(*argc)--;
609 
610 	if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV)) {
611 		r = dm_get_device(ti, arg_value, BLK_OPEN_READ, &v->fec->dev);
612 		if (r) {
613 			ti->error = "FEC device lookup failed";
614 			return r;
615 		}
616 
617 	} else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_BLOCKS)) {
618 		if (sscanf(arg_value, "%llu%c", &num_ll, &dummy) != 1 ||
619 		    ((sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT))
620 		     >> (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll)) {
621 			ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
622 			return -EINVAL;
623 		}
624 		v->fec->blocks = num_ll;
625 
626 	} else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_START)) {
627 		if (sscanf(arg_value, "%llu%c", &num_ll, &dummy) != 1 ||
628 		    ((sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT)) >>
629 		     (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll)) {
630 			ti->error = "Invalid " DM_VERITY_OPT_FEC_START;
631 			return -EINVAL;
632 		}
633 		v->fec->start = num_ll;
634 
635 	} else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_ROOTS)) {
636 		if (sscanf(arg_value, "%hhu%c", &num_c, &dummy) != 1 || !num_c ||
637 		    num_c < (DM_VERITY_FEC_RSM - DM_VERITY_FEC_MAX_RSN) ||
638 		    num_c > (DM_VERITY_FEC_RSM - DM_VERITY_FEC_MIN_RSN)) {
639 			ti->error = "Invalid " DM_VERITY_OPT_FEC_ROOTS;
640 			return -EINVAL;
641 		}
642 		v->fec->roots = num_c;
643 
644 	} else {
645 		ti->error = "Unrecognized verity FEC feature request";
646 		return -EINVAL;
647 	}
648 
649 	return 0;
650 }
651 
652 /*
653  * Allocate dm_verity_fec for v->fec. Must be called before verity_fec_ctr.
654  */
655 int verity_fec_ctr_alloc(struct dm_verity *v)
656 {
657 	struct dm_verity_fec *f;
658 
659 	f = kzalloc(sizeof(struct dm_verity_fec), GFP_KERNEL);
660 	if (!f) {
661 		v->ti->error = "Cannot allocate FEC structure";
662 		return -ENOMEM;
663 	}
664 	v->fec = f;
665 
666 	return 0;
667 }
668 
669 /*
670  * Validate arguments and preallocate memory. Must be called after arguments
671  * have been parsed using verity_fec_parse_opt_args.
672  */
673 int verity_fec_ctr(struct dm_verity *v)
674 {
675 	struct dm_verity_fec *f = v->fec;
676 	struct dm_target *ti = v->ti;
677 	u64 hash_blocks, fec_blocks;
678 	int ret;
679 
680 	if (!verity_fec_is_enabled(v)) {
681 		verity_fec_dtr(v);
682 		return 0;
683 	}
684 
685 	/*
686 	 * FEC is computed over data blocks, possible metadata, and
687 	 * hash blocks. In other words, FEC covers total of fec_blocks
688 	 * blocks consisting of the following:
689 	 *
690 	 *  data blocks | hash blocks | metadata (optional)
691 	 *
692 	 * We allow metadata after hash blocks to support a use case
693 	 * where all data is stored on the same device and FEC covers
694 	 * the entire area.
695 	 *
696 	 * If metadata is included, we require it to be available on the
697 	 * hash device after the hash blocks.
698 	 */
699 
700 	hash_blocks = v->hash_blocks - v->hash_start;
701 
702 	/*
703 	 * Require matching block sizes for data and hash devices for
704 	 * simplicity.
705 	 */
706 	if (v->data_dev_block_bits != v->hash_dev_block_bits) {
707 		ti->error = "Block sizes must match to use FEC";
708 		return -EINVAL;
709 	}
710 
711 	if (!f->roots) {
712 		ti->error = "Missing " DM_VERITY_OPT_FEC_ROOTS;
713 		return -EINVAL;
714 	}
715 	f->rsn = DM_VERITY_FEC_RSM - f->roots;
716 
717 	if (!f->blocks) {
718 		ti->error = "Missing " DM_VERITY_OPT_FEC_BLOCKS;
719 		return -EINVAL;
720 	}
721 
722 	f->rounds = f->blocks;
723 	if (sector_div(f->rounds, f->rsn))
724 		f->rounds++;
725 
726 	/*
727 	 * Due to optional metadata, f->blocks can be larger than
728 	 * data_blocks and hash_blocks combined.
729 	 */
730 	if (f->blocks < v->data_blocks + hash_blocks || !f->rounds) {
731 		ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
732 		return -EINVAL;
733 	}
734 
735 	/*
736 	 * Metadata is accessed through the hash device, so we require
737 	 * it to be large enough.
738 	 */
739 	f->hash_blocks = f->blocks - v->data_blocks;
740 	if (dm_bufio_get_device_size(v->bufio) < f->hash_blocks) {
741 		ti->error = "Hash device is too small for "
742 			DM_VERITY_OPT_FEC_BLOCKS;
743 		return -E2BIG;
744 	}
745 
746 	if ((f->roots << SECTOR_SHIFT) & ((1 << v->data_dev_block_bits) - 1))
747 		f->io_size = 1 << v->data_dev_block_bits;
748 	else
749 		f->io_size = v->fec->roots << SECTOR_SHIFT;
750 
751 	f->bufio = dm_bufio_client_create(f->dev->bdev,
752 					  f->io_size,
753 					  1, 0, NULL, NULL, 0);
754 	if (IS_ERR(f->bufio)) {
755 		ti->error = "Cannot initialize FEC bufio client";
756 		return PTR_ERR(f->bufio);
757 	}
758 
759 	dm_bufio_set_sector_offset(f->bufio, f->start << (v->data_dev_block_bits - SECTOR_SHIFT));
760 
761 	fec_blocks = div64_u64(f->rounds * f->roots, v->fec->roots << SECTOR_SHIFT);
762 	if (dm_bufio_get_device_size(f->bufio) < fec_blocks) {
763 		ti->error = "FEC device is too small";
764 		return -E2BIG;
765 	}
766 
767 	f->data_bufio = dm_bufio_client_create(v->data_dev->bdev,
768 					       1 << v->data_dev_block_bits,
769 					       1, 0, NULL, NULL, 0);
770 	if (IS_ERR(f->data_bufio)) {
771 		ti->error = "Cannot initialize FEC data bufio client";
772 		return PTR_ERR(f->data_bufio);
773 	}
774 
775 	if (dm_bufio_get_device_size(f->data_bufio) < v->data_blocks) {
776 		ti->error = "Data device is too small";
777 		return -E2BIG;
778 	}
779 
780 	/* Preallocate an rs_control structure for each worker thread */
781 	ret = mempool_init(&f->rs_pool, num_online_cpus(), fec_rs_alloc,
782 			   fec_rs_free, (void *) v);
783 	if (ret) {
784 		ti->error = "Cannot allocate RS pool";
785 		return ret;
786 	}
787 
788 	f->cache = kmem_cache_create("dm_verity_fec_buffers",
789 				     f->rsn << DM_VERITY_FEC_BUF_RS_BITS,
790 				     0, 0, NULL);
791 	if (!f->cache) {
792 		ti->error = "Cannot create FEC buffer cache";
793 		return -ENOMEM;
794 	}
795 
796 	/* Preallocate DM_VERITY_FEC_BUF_PREALLOC buffers for each thread */
797 	ret = mempool_init_slab_pool(&f->prealloc_pool, num_online_cpus() *
798 				     DM_VERITY_FEC_BUF_PREALLOC,
799 				     f->cache);
800 	if (ret) {
801 		ti->error = "Cannot allocate FEC buffer prealloc pool";
802 		return ret;
803 	}
804 
805 	ret = mempool_init_slab_pool(&f->extra_pool, 0, f->cache);
806 	if (ret) {
807 		ti->error = "Cannot allocate FEC buffer extra pool";
808 		return ret;
809 	}
810 
811 	/* Preallocate an output buffer for each thread */
812 	ret = mempool_init_kmalloc_pool(&f->output_pool, num_online_cpus(),
813 					1 << v->data_dev_block_bits);
814 	if (ret) {
815 		ti->error = "Cannot allocate FEC output pool";
816 		return ret;
817 	}
818 
819 	/* Reserve space for our per-bio data */
820 	ti->per_io_data_size += sizeof(struct dm_verity_fec_io);
821 
822 	return 0;
823 }
824