xref: /linux/drivers/md/dm-verity-fec.c (revision d358e5254674b70f34c847715ca509e46eb81e6f)
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  */
verity_fec_is_enabled(struct dm_verity * v)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  */
fec_io(struct dm_verity_io * io)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  */
fec_interleave(struct dm_verity * v,u64 offset)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  * Read error-correcting codes for the requested RS block. Returns a pointer
44  * to the data block. Caller is responsible for releasing buf.
45  */
fec_read_parity(struct dm_verity * v,u64 rsb,int index,unsigned int * offset,unsigned int par_buf_offset,struct dm_buffer ** buf,unsigned short ioprio)46 static 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  */
fec_buffer_rs_block(struct dm_verity * v,struct dm_verity_fec_io * fio,unsigned int i,unsigned int j)93 static 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  */
fec_buffer_rs_index(unsigned int i,unsigned int j)104 static 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  */
fec_decode_bufs(struct dm_verity * v,struct dm_verity_io * io,struct dm_verity_fec_io * fio,u64 rsb,int byte_index,unsigned int block_offset,int neras)113 static 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->bi_ioprio);
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->bi_ioprio);
168 			if (IS_ERR(par))
169 				return PTR_ERR(par);
170 		}
171 	}
172 done:
173 	r = corrected;
174 error:
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 		atomic64_inc(&v->fec->corrected);
184 	}
185 
186 	return r;
187 }
188 
189 /*
190  * Locate data block erasures using verity hashes.
191  */
fec_is_erasure(struct dm_verity * v,struct dm_verity_io * io,const u8 * want_digest,const u8 * data)192 static int fec_is_erasure(struct dm_verity *v, struct dm_verity_io *io,
193 			  const u8 *want_digest, const u8 *data)
194 {
195 	if (unlikely(verity_hash(v, io, data, 1 << v->data_dev_block_bits,
196 				 io->tmp_digest)))
197 		return 0;
198 
199 	return memcmp(io->tmp_digest, want_digest, v->digest_size) != 0;
200 }
201 
202 /*
203  * Read data blocks that are part of the RS block and deinterleave as much as
204  * fits into buffers. Check for erasure locations if @neras is non-NULL.
205  */
fec_read_bufs(struct dm_verity * v,struct dm_verity_io * io,u64 rsb,u64 target,unsigned int block_offset,int * neras)206 static int fec_read_bufs(struct dm_verity *v, struct dm_verity_io *io,
207 			 u64 rsb, u64 target, unsigned int block_offset,
208 			 int *neras)
209 {
210 	bool is_zero;
211 	int i, j, target_index = -1;
212 	struct dm_buffer *buf;
213 	struct dm_bufio_client *bufio;
214 	struct dm_verity_fec_io *fio = fec_io(io);
215 	u64 block, ileaved;
216 	u8 *bbuf, *rs_block;
217 	u8 want_digest[HASH_MAX_DIGESTSIZE];
218 	unsigned int n, k;
219 	struct bio *bio = dm_bio_from_per_bio_data(io, v->ti->per_io_data_size);
220 
221 	if (neras)
222 		*neras = 0;
223 
224 	if (WARN_ON(v->digest_size > sizeof(want_digest)))
225 		return -EINVAL;
226 
227 	/*
228 	 * read each of the rsn data blocks that are part of the RS block, and
229 	 * interleave contents to available bufs
230 	 */
231 	for (i = 0; i < v->fec->rsn; i++) {
232 		ileaved = fec_interleave(v, rsb * v->fec->rsn + i);
233 
234 		/*
235 		 * target is the data block we want to correct, target_index is
236 		 * the index of this block within the rsn RS blocks
237 		 */
238 		if (ileaved == target)
239 			target_index = i;
240 
241 		block = ileaved >> v->data_dev_block_bits;
242 		bufio = v->fec->data_bufio;
243 
244 		if (block >= v->data_blocks) {
245 			block -= v->data_blocks;
246 
247 			/*
248 			 * blocks outside the area were assumed to contain
249 			 * zeros when encoding data was generated
250 			 */
251 			if (unlikely(block >= v->fec->hash_blocks))
252 				continue;
253 
254 			block += v->hash_start;
255 			bufio = v->bufio;
256 		}
257 
258 		bbuf = dm_bufio_read_with_ioprio(bufio, block, &buf, bio->bi_ioprio);
259 		if (IS_ERR(bbuf)) {
260 			DMWARN_LIMIT("%s: FEC %llu: read failed (%llu): %ld",
261 				     v->data_dev->name,
262 				     (unsigned long long)rsb,
263 				     (unsigned long long)block, PTR_ERR(bbuf));
264 
265 			/* assume the block is corrupted */
266 			if (neras && *neras <= v->fec->roots)
267 				fio->erasures[(*neras)++] = i;
268 
269 			continue;
270 		}
271 
272 		/* locate erasures if the block is on the data device */
273 		if (bufio == v->fec->data_bufio &&
274 		    verity_hash_for_block(v, io, block, want_digest,
275 					  &is_zero) == 0) {
276 			/* skip known zero blocks entirely */
277 			if (is_zero)
278 				goto done;
279 
280 			/*
281 			 * skip if we have already found the theoretical
282 			 * maximum number (i.e. fec->roots) of erasures
283 			 */
284 			if (neras && *neras <= v->fec->roots &&
285 			    fec_is_erasure(v, io, want_digest, bbuf))
286 				fio->erasures[(*neras)++] = i;
287 		}
288 
289 		/*
290 		 * deinterleave and copy the bytes that fit into bufs,
291 		 * starting from block_offset
292 		 */
293 		fec_for_each_buffer_rs_block(fio, n, j) {
294 			k = fec_buffer_rs_index(n, j) + block_offset;
295 
296 			if (k >= 1 << v->data_dev_block_bits)
297 				goto done;
298 
299 			rs_block = fec_buffer_rs_block(v, fio, n, j);
300 			rs_block[i] = bbuf[k];
301 		}
302 done:
303 		dm_bufio_release(buf);
304 	}
305 
306 	return target_index;
307 }
308 
309 /*
310  * Allocate RS control structure and FEC buffers from preallocated mempools,
311  * and attempt to allocate as many extra buffers as available.
312  */
fec_alloc_bufs(struct dm_verity * v,struct dm_verity_fec_io * fio)313 static int fec_alloc_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
314 {
315 	unsigned int n;
316 
317 	if (!fio->rs)
318 		fio->rs = mempool_alloc(&v->fec->rs_pool, GFP_NOIO);
319 
320 	fec_for_each_prealloc_buffer(n) {
321 		if (fio->bufs[n])
322 			continue;
323 
324 		fio->bufs[n] = mempool_alloc(&v->fec->prealloc_pool, GFP_NOIO);
325 	}
326 
327 	/* try to allocate the maximum number of buffers */
328 	fec_for_each_extra_buffer(fio, n) {
329 		if (fio->bufs[n])
330 			continue;
331 
332 		fio->bufs[n] = kmem_cache_alloc(v->fec->cache, GFP_NOWAIT);
333 		/* we can manage with even one buffer if necessary */
334 		if (unlikely(!fio->bufs[n]))
335 			break;
336 	}
337 	fio->nbufs = n;
338 
339 	if (!fio->output)
340 		fio->output = mempool_alloc(&v->fec->output_pool, GFP_NOIO);
341 
342 	return 0;
343 }
344 
345 /*
346  * Initialize buffers and clear erasures. fec_read_bufs() assumes buffers are
347  * zeroed before deinterleaving.
348  */
fec_init_bufs(struct dm_verity * v,struct dm_verity_fec_io * fio)349 static void fec_init_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
350 {
351 	unsigned int n;
352 
353 	fec_for_each_buffer(fio, n)
354 		memset(fio->bufs[n], 0, v->fec->rsn << DM_VERITY_FEC_BUF_RS_BITS);
355 
356 	memset(fio->erasures, 0, sizeof(fio->erasures));
357 }
358 
359 /*
360  * Decode all RS blocks in a single data block and return the target block
361  * (indicated by @offset) in fio->output. If @use_erasures is non-zero, uses
362  * hashes to locate erasures.
363  */
fec_decode_rsb(struct dm_verity * v,struct dm_verity_io * io,struct dm_verity_fec_io * fio,u64 rsb,u64 offset,const u8 * want_digest,bool use_erasures)364 static int fec_decode_rsb(struct dm_verity *v, struct dm_verity_io *io,
365 			  struct dm_verity_fec_io *fio, u64 rsb, u64 offset,
366 			  const u8 *want_digest, bool use_erasures)
367 {
368 	int r, neras = 0;
369 	unsigned int pos;
370 
371 	r = fec_alloc_bufs(v, fio);
372 	if (unlikely(r < 0))
373 		return r;
374 
375 	for (pos = 0; pos < 1 << v->data_dev_block_bits; ) {
376 		fec_init_bufs(v, fio);
377 
378 		r = fec_read_bufs(v, io, rsb, offset, pos,
379 				  use_erasures ? &neras : NULL);
380 		if (unlikely(r < 0))
381 			return r;
382 
383 		r = fec_decode_bufs(v, io, fio, rsb, r, pos, neras);
384 		if (r < 0)
385 			return r;
386 
387 		pos += fio->nbufs << DM_VERITY_FEC_BUF_RS_BITS;
388 	}
389 
390 	/* Always re-validate the corrected block against the expected hash */
391 	r = verity_hash(v, io, fio->output, 1 << v->data_dev_block_bits,
392 			io->tmp_digest);
393 	if (unlikely(r < 0))
394 		return r;
395 
396 	if (memcmp(io->tmp_digest, want_digest, v->digest_size)) {
397 		DMERR_LIMIT("%s: FEC %llu: failed to correct (%d erasures)",
398 			    v->data_dev->name, (unsigned long long)rsb, neras);
399 		return -EILSEQ;
400 	}
401 
402 	return 0;
403 }
404 
405 /* Correct errors in a block. Copies corrected block to dest. */
verity_fec_decode(struct dm_verity * v,struct dm_verity_io * io,enum verity_block_type type,const u8 * want_digest,sector_t block,u8 * dest)406 int verity_fec_decode(struct dm_verity *v, struct dm_verity_io *io,
407 		      enum verity_block_type type, const u8 *want_digest,
408 		      sector_t block, u8 *dest)
409 {
410 	int r;
411 	struct dm_verity_fec_io *fio = fec_io(io);
412 	u64 offset, res, rsb;
413 
414 	if (!verity_fec_is_enabled(v))
415 		return -EOPNOTSUPP;
416 
417 	if (fio->level)
418 		return -EIO;
419 
420 	fio->level++;
421 
422 	if (type == DM_VERITY_BLOCK_TYPE_METADATA)
423 		block = block - v->hash_start + v->data_blocks;
424 
425 	/*
426 	 * For RS(M, N), the continuous FEC data is divided into blocks of N
427 	 * bytes. Since block size may not be divisible by N, the last block
428 	 * is zero padded when decoding.
429 	 *
430 	 * Each byte of the block is covered by a different RS(M, N) code,
431 	 * and each code is interleaved over N blocks to make it less likely
432 	 * that bursty corruption will leave us in unrecoverable state.
433 	 */
434 
435 	offset = block << v->data_dev_block_bits;
436 	res = div64_u64(offset, v->fec->rounds << v->data_dev_block_bits);
437 
438 	/*
439 	 * The base RS block we can feed to the interleaver to find out all
440 	 * blocks required for decoding.
441 	 */
442 	rsb = offset - res * (v->fec->rounds << v->data_dev_block_bits);
443 
444 	/*
445 	 * Locating erasures is slow, so attempt to recover the block without
446 	 * them first. Do a second attempt with erasures if the corruption is
447 	 * bad enough.
448 	 */
449 	r = fec_decode_rsb(v, io, fio, rsb, offset, want_digest, false);
450 	if (r < 0) {
451 		r = fec_decode_rsb(v, io, fio, rsb, offset, want_digest, true);
452 		if (r < 0)
453 			goto done;
454 	}
455 
456 	memcpy(dest, fio->output, 1 << v->data_dev_block_bits);
457 
458 done:
459 	fio->level--;
460 	return r;
461 }
462 
463 /*
464  * Clean up per-bio data.
465  */
verity_fec_finish_io(struct dm_verity_io * io)466 void verity_fec_finish_io(struct dm_verity_io *io)
467 {
468 	unsigned int n;
469 	struct dm_verity_fec *f = io->v->fec;
470 	struct dm_verity_fec_io *fio = fec_io(io);
471 
472 	if (!verity_fec_is_enabled(io->v))
473 		return;
474 
475 	mempool_free(fio->rs, &f->rs_pool);
476 
477 	fec_for_each_prealloc_buffer(n)
478 		mempool_free(fio->bufs[n], &f->prealloc_pool);
479 
480 	fec_for_each_extra_buffer(fio, n)
481 		if (fio->bufs[n])
482 			kmem_cache_free(f->cache, fio->bufs[n]);
483 
484 	mempool_free(fio->output, &f->output_pool);
485 }
486 
487 /*
488  * Initialize per-bio data.
489  */
verity_fec_init_io(struct dm_verity_io * io)490 void verity_fec_init_io(struct dm_verity_io *io)
491 {
492 	struct dm_verity_fec_io *fio = fec_io(io);
493 
494 	if (!verity_fec_is_enabled(io->v))
495 		return;
496 
497 	fio->rs = NULL;
498 	memset(fio->bufs, 0, sizeof(fio->bufs));
499 	fio->nbufs = 0;
500 	fio->output = NULL;
501 	fio->level = 0;
502 }
503 
504 /*
505  * Append feature arguments and values to the status table.
506  */
verity_fec_status_table(struct dm_verity * v,unsigned int sz,char * result,unsigned int maxlen)507 unsigned int verity_fec_status_table(struct dm_verity *v, unsigned int sz,
508 				 char *result, unsigned int maxlen)
509 {
510 	if (!verity_fec_is_enabled(v))
511 		return sz;
512 
513 	DMEMIT(" " DM_VERITY_OPT_FEC_DEV " %s "
514 	       DM_VERITY_OPT_FEC_BLOCKS " %llu "
515 	       DM_VERITY_OPT_FEC_START " %llu "
516 	       DM_VERITY_OPT_FEC_ROOTS " %d",
517 	       v->fec->dev->name,
518 	       (unsigned long long)v->fec->blocks,
519 	       (unsigned long long)v->fec->start,
520 	       v->fec->roots);
521 
522 	return sz;
523 }
524 
verity_fec_dtr(struct dm_verity * v)525 void verity_fec_dtr(struct dm_verity *v)
526 {
527 	struct dm_verity_fec *f = v->fec;
528 
529 	if (!verity_fec_is_enabled(v))
530 		goto out;
531 
532 	mempool_exit(&f->rs_pool);
533 	mempool_exit(&f->prealloc_pool);
534 	mempool_exit(&f->output_pool);
535 	kmem_cache_destroy(f->cache);
536 
537 	if (f->data_bufio)
538 		dm_bufio_client_destroy(f->data_bufio);
539 	if (f->bufio)
540 		dm_bufio_client_destroy(f->bufio);
541 
542 	if (f->dev)
543 		dm_put_device(v->ti, f->dev);
544 out:
545 	kfree(f);
546 	v->fec = NULL;
547 }
548 
fec_rs_alloc(gfp_t gfp_mask,void * pool_data)549 static void *fec_rs_alloc(gfp_t gfp_mask, void *pool_data)
550 {
551 	struct dm_verity *v = pool_data;
552 
553 	return init_rs_gfp(8, 0x11d, 0, 1, v->fec->roots, gfp_mask);
554 }
555 
fec_rs_free(void * element,void * pool_data)556 static void fec_rs_free(void *element, void *pool_data)
557 {
558 	struct rs_control *rs = element;
559 
560 	if (rs)
561 		free_rs(rs);
562 }
563 
verity_is_fec_opt_arg(const char * arg_name)564 bool verity_is_fec_opt_arg(const char *arg_name)
565 {
566 	return (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV) ||
567 		!strcasecmp(arg_name, DM_VERITY_OPT_FEC_BLOCKS) ||
568 		!strcasecmp(arg_name, DM_VERITY_OPT_FEC_START) ||
569 		!strcasecmp(arg_name, DM_VERITY_OPT_FEC_ROOTS));
570 }
571 
verity_fec_parse_opt_args(struct dm_arg_set * as,struct dm_verity * v,unsigned int * argc,const char * arg_name)572 int verity_fec_parse_opt_args(struct dm_arg_set *as, struct dm_verity *v,
573 			      unsigned int *argc, const char *arg_name)
574 {
575 	int r;
576 	struct dm_target *ti = v->ti;
577 	const char *arg_value;
578 	unsigned long long num_ll;
579 	unsigned char num_c;
580 	char dummy;
581 
582 	if (!*argc) {
583 		ti->error = "FEC feature arguments require a value";
584 		return -EINVAL;
585 	}
586 
587 	arg_value = dm_shift_arg(as);
588 	(*argc)--;
589 
590 	if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV)) {
591 		if (v->fec->dev) {
592 			ti->error = "FEC device already specified";
593 			return -EINVAL;
594 		}
595 		r = dm_get_device(ti, arg_value, BLK_OPEN_READ, &v->fec->dev);
596 		if (r) {
597 			ti->error = "FEC device lookup failed";
598 			return r;
599 		}
600 
601 	} else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_BLOCKS)) {
602 		if (sscanf(arg_value, "%llu%c", &num_ll, &dummy) != 1 ||
603 		    ((sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT))
604 		     >> (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll)) {
605 			ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
606 			return -EINVAL;
607 		}
608 		v->fec->blocks = num_ll;
609 
610 	} else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_START)) {
611 		if (sscanf(arg_value, "%llu%c", &num_ll, &dummy) != 1 ||
612 		    ((sector_t)(num_ll << (v->data_dev_block_bits - SECTOR_SHIFT)) >>
613 		     (v->data_dev_block_bits - SECTOR_SHIFT) != num_ll)) {
614 			ti->error = "Invalid " DM_VERITY_OPT_FEC_START;
615 			return -EINVAL;
616 		}
617 		v->fec->start = num_ll;
618 
619 	} else if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_ROOTS)) {
620 		if (sscanf(arg_value, "%hhu%c", &num_c, &dummy) != 1 || !num_c ||
621 		    num_c < (DM_VERITY_FEC_RSM - DM_VERITY_FEC_MAX_RSN) ||
622 		    num_c > (DM_VERITY_FEC_RSM - DM_VERITY_FEC_MIN_RSN)) {
623 			ti->error = "Invalid " DM_VERITY_OPT_FEC_ROOTS;
624 			return -EINVAL;
625 		}
626 		v->fec->roots = num_c;
627 
628 	} else {
629 		ti->error = "Unrecognized verity FEC feature request";
630 		return -EINVAL;
631 	}
632 
633 	return 0;
634 }
635 
636 /*
637  * Allocate dm_verity_fec for v->fec. Must be called before verity_fec_ctr.
638  */
verity_fec_ctr_alloc(struct dm_verity * v)639 int verity_fec_ctr_alloc(struct dm_verity *v)
640 {
641 	struct dm_verity_fec *f;
642 
643 	f = kzalloc(sizeof(struct dm_verity_fec), GFP_KERNEL);
644 	if (!f) {
645 		v->ti->error = "Cannot allocate FEC structure";
646 		return -ENOMEM;
647 	}
648 	v->fec = f;
649 
650 	return 0;
651 }
652 
653 /*
654  * Validate arguments and preallocate memory. Must be called after arguments
655  * have been parsed using verity_fec_parse_opt_args.
656  */
verity_fec_ctr(struct dm_verity * v)657 int verity_fec_ctr(struct dm_verity *v)
658 {
659 	struct dm_verity_fec *f = v->fec;
660 	struct dm_target *ti = v->ti;
661 	u64 hash_blocks, fec_blocks;
662 	int ret;
663 
664 	if (!verity_fec_is_enabled(v)) {
665 		verity_fec_dtr(v);
666 		return 0;
667 	}
668 
669 	/*
670 	 * FEC is computed over data blocks, possible metadata, and
671 	 * hash blocks. In other words, FEC covers total of fec_blocks
672 	 * blocks consisting of the following:
673 	 *
674 	 *  data blocks | hash blocks | metadata (optional)
675 	 *
676 	 * We allow metadata after hash blocks to support a use case
677 	 * where all data is stored on the same device and FEC covers
678 	 * the entire area.
679 	 *
680 	 * If metadata is included, we require it to be available on the
681 	 * hash device after the hash blocks.
682 	 */
683 
684 	hash_blocks = v->hash_blocks - v->hash_start;
685 
686 	/*
687 	 * Require matching block sizes for data and hash devices for
688 	 * simplicity.
689 	 */
690 	if (v->data_dev_block_bits != v->hash_dev_block_bits) {
691 		ti->error = "Block sizes must match to use FEC";
692 		return -EINVAL;
693 	}
694 
695 	if (!f->roots) {
696 		ti->error = "Missing " DM_VERITY_OPT_FEC_ROOTS;
697 		return -EINVAL;
698 	}
699 	f->rsn = DM_VERITY_FEC_RSM - f->roots;
700 
701 	if (!f->blocks) {
702 		ti->error = "Missing " DM_VERITY_OPT_FEC_BLOCKS;
703 		return -EINVAL;
704 	}
705 
706 	f->rounds = f->blocks;
707 	if (sector_div(f->rounds, f->rsn))
708 		f->rounds++;
709 
710 	/*
711 	 * Due to optional metadata, f->blocks can be larger than
712 	 * data_blocks and hash_blocks combined.
713 	 */
714 	if (f->blocks < v->data_blocks + hash_blocks || !f->rounds) {
715 		ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
716 		return -EINVAL;
717 	}
718 
719 	/*
720 	 * Metadata is accessed through the hash device, so we require
721 	 * it to be large enough.
722 	 */
723 	f->hash_blocks = f->blocks - v->data_blocks;
724 	if (dm_bufio_get_device_size(v->bufio) < f->hash_blocks) {
725 		ti->error = "Hash device is too small for "
726 			DM_VERITY_OPT_FEC_BLOCKS;
727 		return -E2BIG;
728 	}
729 
730 	f->io_size = 1 << v->data_dev_block_bits;
731 
732 	f->bufio = dm_bufio_client_create(f->dev->bdev,
733 					  f->io_size,
734 					  1, 0, NULL, NULL, 0);
735 	if (IS_ERR(f->bufio)) {
736 		ti->error = "Cannot initialize FEC bufio client";
737 		return PTR_ERR(f->bufio);
738 	}
739 
740 	dm_bufio_set_sector_offset(f->bufio, f->start << (v->data_dev_block_bits - SECTOR_SHIFT));
741 
742 	fec_blocks = div64_u64(f->rounds * f->roots, v->fec->roots << SECTOR_SHIFT);
743 	if (dm_bufio_get_device_size(f->bufio) < fec_blocks) {
744 		ti->error = "FEC device is too small";
745 		return -E2BIG;
746 	}
747 
748 	f->data_bufio = dm_bufio_client_create(v->data_dev->bdev,
749 					       1 << v->data_dev_block_bits,
750 					       1, 0, NULL, NULL, 0);
751 	if (IS_ERR(f->data_bufio)) {
752 		ti->error = "Cannot initialize FEC data bufio client";
753 		return PTR_ERR(f->data_bufio);
754 	}
755 
756 	if (dm_bufio_get_device_size(f->data_bufio) < v->data_blocks) {
757 		ti->error = "Data device is too small";
758 		return -E2BIG;
759 	}
760 
761 	/* Preallocate an rs_control structure for each worker thread */
762 	ret = mempool_init(&f->rs_pool, num_online_cpus(), fec_rs_alloc,
763 			   fec_rs_free, (void *) v);
764 	if (ret) {
765 		ti->error = "Cannot allocate RS pool";
766 		return ret;
767 	}
768 
769 	f->cache = kmem_cache_create("dm_verity_fec_buffers",
770 				     f->rsn << DM_VERITY_FEC_BUF_RS_BITS,
771 				     0, 0, NULL);
772 	if (!f->cache) {
773 		ti->error = "Cannot create FEC buffer cache";
774 		return -ENOMEM;
775 	}
776 
777 	/* Preallocate DM_VERITY_FEC_BUF_PREALLOC buffers for each thread */
778 	ret = mempool_init_slab_pool(&f->prealloc_pool, num_online_cpus() *
779 				     DM_VERITY_FEC_BUF_PREALLOC,
780 				     f->cache);
781 	if (ret) {
782 		ti->error = "Cannot allocate FEC buffer prealloc pool";
783 		return ret;
784 	}
785 
786 	/* Preallocate an output buffer for each thread */
787 	ret = mempool_init_kmalloc_pool(&f->output_pool, num_online_cpus(),
788 					1 << v->data_dev_block_bits);
789 	if (ret) {
790 		ti->error = "Cannot allocate FEC output pool";
791 		return ret;
792 	}
793 
794 	/* Reserve space for our per-bio data */
795 	ti->per_io_data_size += sizeof(struct dm_verity_fec_io);
796 
797 	return 0;
798 }
799