xref: /openbmc/linux/drivers/md/dm-verity-fec.c (revision e3d786a3)
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  */
20 bool 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  */
29 static 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  */
37 static 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  */
48 static 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  */
65 static 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  */
108 static 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  */
119 static 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  */
128 static 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 	}
170 done:
171 	r = corrected;
172 error:
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  */
188 static 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_req(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  */
204 static 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[HASH_MAX_DIGESTSIZE];
216 	unsigned n, k;
217 
218 	if (neras)
219 		*neras = 0;
220 
221 	if (WARN_ON(v->digest_size > sizeof(want_digest)))
222 		return -EINVAL;
223 
224 	/*
225 	 * read each of the rsn data blocks that are part of the RS block, and
226 	 * interleave contents to available bufs
227 	 */
228 	for (i = 0; i < v->fec->rsn; i++) {
229 		ileaved = fec_interleave(v, rsb * v->fec->rsn + i);
230 
231 		/*
232 		 * target is the data block we want to correct, target_index is
233 		 * the index of this block within the rsn RS blocks
234 		 */
235 		if (ileaved == target)
236 			target_index = i;
237 
238 		block = ileaved >> v->data_dev_block_bits;
239 		bufio = v->fec->data_bufio;
240 
241 		if (block >= v->data_blocks) {
242 			block -= v->data_blocks;
243 
244 			/*
245 			 * blocks outside the area were assumed to contain
246 			 * zeros when encoding data was generated
247 			 */
248 			if (unlikely(block >= v->fec->hash_blocks))
249 				continue;
250 
251 			block += v->hash_start;
252 			bufio = v->bufio;
253 		}
254 
255 		bbuf = dm_bufio_read(bufio, block, &buf);
256 		if (unlikely(IS_ERR(bbuf))) {
257 			DMWARN_LIMIT("%s: FEC %llu: read failed (%llu): %ld",
258 				     v->data_dev->name,
259 				     (unsigned long long)rsb,
260 				     (unsigned long long)block, PTR_ERR(bbuf));
261 
262 			/* assume the block is corrupted */
263 			if (neras && *neras <= v->fec->roots)
264 				fio->erasures[(*neras)++] = i;
265 
266 			continue;
267 		}
268 
269 		/* locate erasures if the block is on the data device */
270 		if (bufio == v->fec->data_bufio &&
271 		    verity_hash_for_block(v, io, block, want_digest,
272 					  &is_zero) == 0) {
273 			/* skip known zero blocks entirely */
274 			if (is_zero)
275 				goto done;
276 
277 			/*
278 			 * skip if we have already found the theoretical
279 			 * maximum number (i.e. fec->roots) of erasures
280 			 */
281 			if (neras && *neras <= v->fec->roots &&
282 			    fec_is_erasure(v, io, want_digest, bbuf))
283 				fio->erasures[(*neras)++] = i;
284 		}
285 
286 		/*
287 		 * deinterleave and copy the bytes that fit into bufs,
288 		 * starting from block_offset
289 		 */
290 		fec_for_each_buffer_rs_block(fio, n, j) {
291 			k = fec_buffer_rs_index(n, j) + block_offset;
292 
293 			if (k >= 1 << v->data_dev_block_bits)
294 				goto done;
295 
296 			rs_block = fec_buffer_rs_block(v, fio, n, j);
297 			rs_block[i] = bbuf[k];
298 		}
299 done:
300 		dm_bufio_release(buf);
301 	}
302 
303 	return target_index;
304 }
305 
306 /*
307  * Allocate RS control structure and FEC buffers from preallocated mempools,
308  * and attempt to allocate as many extra buffers as available.
309  */
310 static int fec_alloc_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
311 {
312 	unsigned n;
313 
314 	if (!fio->rs)
315 		fio->rs = mempool_alloc(&v->fec->rs_pool, GFP_NOIO);
316 
317 	fec_for_each_prealloc_buffer(n) {
318 		if (fio->bufs[n])
319 			continue;
320 
321 		fio->bufs[n] = mempool_alloc(&v->fec->prealloc_pool, GFP_NOWAIT);
322 		if (unlikely(!fio->bufs[n])) {
323 			DMERR("failed to allocate FEC buffer");
324 			return -ENOMEM;
325 		}
326 	}
327 
328 	/* try to allocate the maximum number of buffers */
329 	fec_for_each_extra_buffer(fio, n) {
330 		if (fio->bufs[n])
331 			continue;
332 
333 		fio->bufs[n] = mempool_alloc(&v->fec->extra_pool, GFP_NOWAIT);
334 		/* we can manage with even one buffer if necessary */
335 		if (unlikely(!fio->bufs[n]))
336 			break;
337 	}
338 	fio->nbufs = n;
339 
340 	if (!fio->output)
341 		fio->output = mempool_alloc(&v->fec->output_pool, GFP_NOIO);
342 
343 	return 0;
344 }
345 
346 /*
347  * Initialize buffers and clear erasures. fec_read_bufs() assumes buffers are
348  * zeroed before deinterleaving.
349  */
350 static void fec_init_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
351 {
352 	unsigned n;
353 
354 	fec_for_each_buffer(fio, n)
355 		memset(fio->bufs[n], 0, v->fec->rsn << DM_VERITY_FEC_BUF_RS_BITS);
356 
357 	memset(fio->erasures, 0, sizeof(fio->erasures));
358 }
359 
360 /*
361  * Decode all RS blocks in a single data block and return the target block
362  * (indicated by @offset) in fio->output. If @use_erasures is non-zero, uses
363  * hashes to locate erasures.
364  */
365 static int fec_decode_rsb(struct dm_verity *v, struct dm_verity_io *io,
366 			  struct dm_verity_fec_io *fio, u64 rsb, u64 offset,
367 			  bool use_erasures)
368 {
369 	int r, neras = 0;
370 	unsigned pos;
371 
372 	r = fec_alloc_bufs(v, fio);
373 	if (unlikely(r < 0))
374 		return r;
375 
376 	for (pos = 0; pos < 1 << v->data_dev_block_bits; ) {
377 		fec_init_bufs(v, fio);
378 
379 		r = fec_read_bufs(v, io, rsb, offset, pos,
380 				  use_erasures ? &neras : NULL);
381 		if (unlikely(r < 0))
382 			return r;
383 
384 		r = fec_decode_bufs(v, fio, rsb, r, pos, neras);
385 		if (r < 0)
386 			return r;
387 
388 		pos += fio->nbufs << DM_VERITY_FEC_BUF_RS_BITS;
389 	}
390 
391 	/* Always re-validate the corrected block against the expected hash */
392 	r = verity_hash(v, verity_io_hash_req(v, io), fio->output,
393 			1 << v->data_dev_block_bits,
394 			verity_io_real_digest(v, io));
395 	if (unlikely(r < 0))
396 		return r;
397 
398 	if (memcmp(verity_io_real_digest(v, io), verity_io_want_digest(v, io),
399 		   v->digest_size)) {
400 		DMERR_LIMIT("%s: FEC %llu: failed to correct (%d erasures)",
401 			    v->data_dev->name, (unsigned long long)rsb, neras);
402 		return -EILSEQ;
403 	}
404 
405 	return 0;
406 }
407 
408 static int fec_bv_copy(struct dm_verity *v, struct dm_verity_io *io, u8 *data,
409 		       size_t len)
410 {
411 	struct dm_verity_fec_io *fio = fec_io(io);
412 
413 	memcpy(data, &fio->output[fio->output_pos], len);
414 	fio->output_pos += len;
415 
416 	return 0;
417 }
418 
419 /*
420  * Correct errors in a block. Copies corrected block to dest if non-NULL,
421  * otherwise to a bio_vec starting from iter.
422  */
423 int verity_fec_decode(struct dm_verity *v, struct dm_verity_io *io,
424 		      enum verity_block_type type, sector_t block, u8 *dest,
425 		      struct bvec_iter *iter)
426 {
427 	int r;
428 	struct dm_verity_fec_io *fio = fec_io(io);
429 	u64 offset, res, rsb;
430 
431 	if (!verity_fec_is_enabled(v))
432 		return -EOPNOTSUPP;
433 
434 	if (fio->level >= DM_VERITY_FEC_MAX_RECURSION) {
435 		DMWARN_LIMIT("%s: FEC: recursion too deep", v->data_dev->name);
436 		return -EIO;
437 	}
438 
439 	fio->level++;
440 
441 	if (type == DM_VERITY_BLOCK_TYPE_METADATA)
442 		block += v->data_blocks;
443 
444 	/*
445 	 * For RS(M, N), the continuous FEC data is divided into blocks of N
446 	 * bytes. Since block size may not be divisible by N, the last block
447 	 * is zero padded when decoding.
448 	 *
449 	 * Each byte of the block is covered by a different RS(M, N) code,
450 	 * and each code is interleaved over N blocks to make it less likely
451 	 * that bursty corruption will leave us in unrecoverable state.
452 	 */
453 
454 	offset = block << v->data_dev_block_bits;
455 	res = div64_u64(offset, v->fec->rounds << v->data_dev_block_bits);
456 
457 	/*
458 	 * The base RS block we can feed to the interleaver to find out all
459 	 * blocks required for decoding.
460 	 */
461 	rsb = offset - res * (v->fec->rounds << v->data_dev_block_bits);
462 
463 	/*
464 	 * Locating erasures is slow, so attempt to recover the block without
465 	 * them first. Do a second attempt with erasures if the corruption is
466 	 * bad enough.
467 	 */
468 	r = fec_decode_rsb(v, io, fio, rsb, offset, false);
469 	if (r < 0) {
470 		r = fec_decode_rsb(v, io, fio, rsb, offset, true);
471 		if (r < 0)
472 			goto done;
473 	}
474 
475 	if (dest)
476 		memcpy(dest, fio->output, 1 << v->data_dev_block_bits);
477 	else if (iter) {
478 		fio->output_pos = 0;
479 		r = verity_for_bv_block(v, io, iter, fec_bv_copy);
480 	}
481 
482 done:
483 	fio->level--;
484 	return r;
485 }
486 
487 /*
488  * Clean up per-bio data.
489  */
490 void verity_fec_finish_io(struct dm_verity_io *io)
491 {
492 	unsigned n;
493 	struct dm_verity_fec *f = io->v->fec;
494 	struct dm_verity_fec_io *fio = fec_io(io);
495 
496 	if (!verity_fec_is_enabled(io->v))
497 		return;
498 
499 	mempool_free(fio->rs, &f->rs_pool);
500 
501 	fec_for_each_prealloc_buffer(n)
502 		mempool_free(fio->bufs[n], &f->prealloc_pool);
503 
504 	fec_for_each_extra_buffer(fio, n)
505 		mempool_free(fio->bufs[n], &f->extra_pool);
506 
507 	mempool_free(fio->output, &f->output_pool);
508 }
509 
510 /*
511  * Initialize per-bio data.
512  */
513 void verity_fec_init_io(struct dm_verity_io *io)
514 {
515 	struct dm_verity_fec_io *fio = fec_io(io);
516 
517 	if (!verity_fec_is_enabled(io->v))
518 		return;
519 
520 	fio->rs = NULL;
521 	memset(fio->bufs, 0, sizeof(fio->bufs));
522 	fio->nbufs = 0;
523 	fio->output = NULL;
524 	fio->level = 0;
525 }
526 
527 /*
528  * Append feature arguments and values to the status table.
529  */
530 unsigned verity_fec_status_table(struct dm_verity *v, unsigned sz,
531 				 char *result, unsigned maxlen)
532 {
533 	if (!verity_fec_is_enabled(v))
534 		return sz;
535 
536 	DMEMIT(" " DM_VERITY_OPT_FEC_DEV " %s "
537 	       DM_VERITY_OPT_FEC_BLOCKS " %llu "
538 	       DM_VERITY_OPT_FEC_START " %llu "
539 	       DM_VERITY_OPT_FEC_ROOTS " %d",
540 	       v->fec->dev->name,
541 	       (unsigned long long)v->fec->blocks,
542 	       (unsigned long long)v->fec->start,
543 	       v->fec->roots);
544 
545 	return sz;
546 }
547 
548 void verity_fec_dtr(struct dm_verity *v)
549 {
550 	struct dm_verity_fec *f = v->fec;
551 
552 	if (!verity_fec_is_enabled(v))
553 		goto out;
554 
555 	mempool_exit(&f->rs_pool);
556 	mempool_exit(&f->prealloc_pool);
557 	mempool_exit(&f->extra_pool);
558 	kmem_cache_destroy(f->cache);
559 
560 	if (f->data_bufio)
561 		dm_bufio_client_destroy(f->data_bufio);
562 	if (f->bufio)
563 		dm_bufio_client_destroy(f->bufio);
564 
565 	if (f->dev)
566 		dm_put_device(v->ti, f->dev);
567 out:
568 	kfree(f);
569 	v->fec = NULL;
570 }
571 
572 static void *fec_rs_alloc(gfp_t gfp_mask, void *pool_data)
573 {
574 	struct dm_verity *v = (struct dm_verity *)pool_data;
575 
576 	return init_rs_gfp(8, 0x11d, 0, 1, v->fec->roots, gfp_mask);
577 }
578 
579 static void fec_rs_free(void *element, void *pool_data)
580 {
581 	struct rs_control *rs = (struct rs_control *)element;
582 
583 	if (rs)
584 		free_rs(rs);
585 }
586 
587 bool verity_is_fec_opt_arg(const char *arg_name)
588 {
589 	return (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV) ||
590 		!strcasecmp(arg_name, DM_VERITY_OPT_FEC_BLOCKS) ||
591 		!strcasecmp(arg_name, DM_VERITY_OPT_FEC_START) ||
592 		!strcasecmp(arg_name, DM_VERITY_OPT_FEC_ROOTS));
593 }
594 
595 int verity_fec_parse_opt_args(struct dm_arg_set *as, struct dm_verity *v,
596 			      unsigned *argc, const char *arg_name)
597 {
598 	int r;
599 	struct dm_target *ti = v->ti;
600 	const char *arg_value;
601 	unsigned long long num_ll;
602 	unsigned char num_c;
603 	char dummy;
604 
605 	if (!*argc) {
606 		ti->error = "FEC feature arguments require a value";
607 		return -EINVAL;
608 	}
609 
610 	arg_value = dm_shift_arg(as);
611 	(*argc)--;
612 
613 	if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV)) {
614 		r = dm_get_device(ti, arg_value, FMODE_READ, &v->fec->dev);
615 		if (r) {
616 			ti->error = "FEC device lookup failed";
617 			return r;
618 		}
619 
620 	} else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_BLOCKS)) {
621 		if (sscanf(arg_value, "%llu%c", &num_ll, &dummy) != 1 ||
622 		    ((sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT))
623 		     >> (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll)) {
624 			ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
625 			return -EINVAL;
626 		}
627 		v->fec->blocks = num_ll;
628 
629 	} else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_START)) {
630 		if (sscanf(arg_value, "%llu%c", &num_ll, &dummy) != 1 ||
631 		    ((sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT)) >>
632 		     (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll)) {
633 			ti->error = "Invalid " DM_VERITY_OPT_FEC_START;
634 			return -EINVAL;
635 		}
636 		v->fec->start = num_ll;
637 
638 	} else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_ROOTS)) {
639 		if (sscanf(arg_value, "%hhu%c", &num_c, &dummy) != 1 || !num_c ||
640 		    num_c < (DM_VERITY_FEC_RSM - DM_VERITY_FEC_MAX_RSN) ||
641 		    num_c > (DM_VERITY_FEC_RSM - DM_VERITY_FEC_MIN_RSN)) {
642 			ti->error = "Invalid " DM_VERITY_OPT_FEC_ROOTS;
643 			return -EINVAL;
644 		}
645 		v->fec->roots = num_c;
646 
647 	} else {
648 		ti->error = "Unrecognized verity FEC feature request";
649 		return -EINVAL;
650 	}
651 
652 	return 0;
653 }
654 
655 /*
656  * Allocate dm_verity_fec for v->fec. Must be called before verity_fec_ctr.
657  */
658 int verity_fec_ctr_alloc(struct dm_verity *v)
659 {
660 	struct dm_verity_fec *f;
661 
662 	f = kzalloc(sizeof(struct dm_verity_fec), GFP_KERNEL);
663 	if (!f) {
664 		v->ti->error = "Cannot allocate FEC structure";
665 		return -ENOMEM;
666 	}
667 	v->fec = f;
668 
669 	return 0;
670 }
671 
672 /*
673  * Validate arguments and preallocate memory. Must be called after arguments
674  * have been parsed using verity_fec_parse_opt_args.
675  */
676 int verity_fec_ctr(struct dm_verity *v)
677 {
678 	struct dm_verity_fec *f = v->fec;
679 	struct dm_target *ti = v->ti;
680 	u64 hash_blocks;
681 	int ret;
682 
683 	if (!verity_fec_is_enabled(v)) {
684 		verity_fec_dtr(v);
685 		return 0;
686 	}
687 
688 	/*
689 	 * FEC is computed over data blocks, possible metadata, and
690 	 * hash blocks. In other words, FEC covers total of fec_blocks
691 	 * blocks consisting of the following:
692 	 *
693 	 *  data blocks | hash blocks | metadata (optional)
694 	 *
695 	 * We allow metadata after hash blocks to support a use case
696 	 * where all data is stored on the same device and FEC covers
697 	 * the entire area.
698 	 *
699 	 * If metadata is included, we require it to be available on the
700 	 * hash device after the hash blocks.
701 	 */
702 
703 	hash_blocks = v->hash_blocks - v->hash_start;
704 
705 	/*
706 	 * Require matching block sizes for data and hash devices for
707 	 * simplicity.
708 	 */
709 	if (v->data_dev_block_bits != v->hash_dev_block_bits) {
710 		ti->error = "Block sizes must match to use FEC";
711 		return -EINVAL;
712 	}
713 
714 	if (!f->roots) {
715 		ti->error = "Missing " DM_VERITY_OPT_FEC_ROOTS;
716 		return -EINVAL;
717 	}
718 	f->rsn = DM_VERITY_FEC_RSM - f->roots;
719 
720 	if (!f->blocks) {
721 		ti->error = "Missing " DM_VERITY_OPT_FEC_BLOCKS;
722 		return -EINVAL;
723 	}
724 
725 	f->rounds = f->blocks;
726 	if (sector_div(f->rounds, f->rsn))
727 		f->rounds++;
728 
729 	/*
730 	 * Due to optional metadata, f->blocks can be larger than
731 	 * data_blocks and hash_blocks combined.
732 	 */
733 	if (f->blocks < v->data_blocks + hash_blocks || !f->rounds) {
734 		ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
735 		return -EINVAL;
736 	}
737 
738 	/*
739 	 * Metadata is accessed through the hash device, so we require
740 	 * it to be large enough.
741 	 */
742 	f->hash_blocks = f->blocks - v->data_blocks;
743 	if (dm_bufio_get_device_size(v->bufio) < f->hash_blocks) {
744 		ti->error = "Hash device is too small for "
745 			DM_VERITY_OPT_FEC_BLOCKS;
746 		return -E2BIG;
747 	}
748 
749 	f->bufio = dm_bufio_client_create(f->dev->bdev,
750 					  1 << v->data_dev_block_bits,
751 					  1, 0, NULL, NULL);
752 	if (IS_ERR(f->bufio)) {
753 		ti->error = "Cannot initialize FEC bufio client";
754 		return PTR_ERR(f->bufio);
755 	}
756 
757 	if (dm_bufio_get_device_size(f->bufio) <
758 	    ((f->start + f->rounds * f->roots) >> v->data_dev_block_bits)) {
759 		ti->error = "FEC device is too small";
760 		return -E2BIG;
761 	}
762 
763 	f->data_bufio = dm_bufio_client_create(v->data_dev->bdev,
764 					       1 << v->data_dev_block_bits,
765 					       1, 0, NULL, NULL);
766 	if (IS_ERR(f->data_bufio)) {
767 		ti->error = "Cannot initialize FEC data bufio client";
768 		return PTR_ERR(f->data_bufio);
769 	}
770 
771 	if (dm_bufio_get_device_size(f->data_bufio) < v->data_blocks) {
772 		ti->error = "Data device is too small";
773 		return -E2BIG;
774 	}
775 
776 	/* Preallocate an rs_control structure for each worker thread */
777 	ret = mempool_init(&f->rs_pool, num_online_cpus(), fec_rs_alloc,
778 			   fec_rs_free, (void *) v);
779 	if (ret) {
780 		ti->error = "Cannot allocate RS pool";
781 		return ret;
782 	}
783 
784 	f->cache = kmem_cache_create("dm_verity_fec_buffers",
785 				     f->rsn << DM_VERITY_FEC_BUF_RS_BITS,
786 				     0, 0, NULL);
787 	if (!f->cache) {
788 		ti->error = "Cannot create FEC buffer cache";
789 		return -ENOMEM;
790 	}
791 
792 	/* Preallocate DM_VERITY_FEC_BUF_PREALLOC buffers for each thread */
793 	ret = mempool_init_slab_pool(&f->prealloc_pool, num_online_cpus() *
794 				     DM_VERITY_FEC_BUF_PREALLOC,
795 				     f->cache);
796 	if (ret) {
797 		ti->error = "Cannot allocate FEC buffer prealloc pool";
798 		return ret;
799 	}
800 
801 	ret = mempool_init_slab_pool(&f->extra_pool, 0, f->cache);
802 	if (ret) {
803 		ti->error = "Cannot allocate FEC buffer extra pool";
804 		return ret;
805 	}
806 
807 	/* Preallocate an output buffer for each thread */
808 	ret = mempool_init_kmalloc_pool(&f->output_pool, num_online_cpus(),
809 					1 << v->data_dev_block_bits);
810 	if (ret) {
811 		ti->error = "Cannot allocate FEC output pool";
812 		return ret;
813 	}
814 
815 	/* Reserve space for our per-bio data */
816 	ti->per_io_data_size += sizeof(struct dm_verity_fec_io);
817 
818 	return 0;
819 }
820