2 * Copyright (C) 2015 Google, Inc.
4 * Author: Sami Tolvanen <samitolvanen@google.com>
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)
12 #include "dm-verity-fec.h"
13 #include <linux/math64.h>
15 #define DM_MSG_PREFIX "verity-fec"
18 * If error correction has been configured, returns true.
20 bool verity_fec_is_enabled(struct dm_verity *v)
22 return v->fec && v->fec->dev;
26 * Return a pointer to dm_verity_fec_io after dm_verity_io and its variable
29 static inline struct dm_verity_fec_io *fec_io(struct dm_verity_io *io)
31 return (struct dm_verity_fec_io *) verity_io_digest_end(io->v, io);
35 * Return an interleaved offset for a byte in RS block.
37 static inline u64 fec_interleave(struct dm_verity *v, u64 offset)
41 mod = do_div(offset, v->fec->rsn);
42 return offset + mod * (v->fec->rounds << v->data_dev_block_bits);
46 * Decode an RS block using Reed-Solomon.
48 static int fec_decode_rs8(struct dm_verity *v, struct dm_verity_fec_io *fio,
49 u8 *data, u8 *fec, int neras)
52 uint16_t par[DM_VERITY_FEC_RSM - DM_VERITY_FEC_MIN_RSN];
54 for (i = 0; i < v->fec->roots; i++)
57 return decode_rs8(fio->rs, data, par, v->fec->rsn, NULL, neras,
58 fio->erasures, 0, NULL);
62 * Read error-correcting codes for the requested RS block. Returns a pointer
63 * to the data block. Caller is responsible for releasing buf.
65 static u8 *fec_read_parity(struct dm_verity *v, u64 rsb, int index,
66 unsigned *offset, struct dm_buffer **buf)
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));
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),
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++)
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++)
95 /* Loop over each allocated buffer. */
96 #define fec_for_each_buffer(io, __i) \
97 for (__i = 0; __i < (io)->nbufs; __i++)
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++)
105 * Return a pointer to the current RS block when called inside
106 * fec_for_each_buffer_rs_block.
108 static inline u8 *fec_buffer_rs_block(struct dm_verity *v,
109 struct dm_verity_fec_io *fio,
110 unsigned i, unsigned j)
112 return &fio->bufs[i][j * v->fec->rsn];
116 * Return an index to the current RS block when called inside
117 * fec_for_each_buffer_rs_block.
119 static inline unsigned fec_buffer_rs_index(unsigned i, unsigned j)
121 return (i << DM_VERITY_FEC_BUF_RS_BITS) + j;
125 * Decode all RS blocks from buffers and copy corrected bytes into fio->output
126 * starting from block_offset.
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,
132 int r, corrected = 0, res;
133 struct dm_buffer *buf;
134 unsigned n, i, offset;
137 par = fec_read_parity(v, rsb, block_offset, &offset, &buf);
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.
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);
154 fio->output[block_offset] = block[byte_index];
157 if (block_offset >= 1 << v->data_dev_block_bits)
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);
165 par = fec_read_parity(v, rsb, block_offset, &offset, &buf);
166 if (unlikely(IS_ERR(par)))
173 dm_bufio_release(buf);
176 DMERR_LIMIT("%s: FEC %llu: failed to correct: %d",
177 v->data_dev->name, (unsigned long long)rsb, r);
179 DMWARN_LIMIT("%s: FEC %llu: corrected %d errors",
180 v->data_dev->name, (unsigned long long)rsb, r);
186 * Locate data block erasures using verity hashes.
188 static int fec_is_erasure(struct dm_verity *v, struct dm_verity_io *io,
189 u8 *want_digest, u8 *data)
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))))
196 return memcmp(verity_io_real_digest(v, io), want_digest,
197 v->digest_size) != 0;
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.
204 static int fec_read_bufs(struct dm_verity *v, struct dm_verity_io *io,
205 u64 rsb, u64 target, unsigned block_offset,
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);
215 u8 want_digest[HASH_MAX_DIGESTSIZE];
221 if (WARN_ON(v->digest_size > sizeof(want_digest)))
225 * read each of the rsn data blocks that are part of the RS block, and
226 * interleave contents to available bufs
228 for (i = 0; i < v->fec->rsn; i++) {
229 ileaved = fec_interleave(v, rsb * v->fec->rsn + i);
232 * target is the data block we want to correct, target_index is
233 * the index of this block within the rsn RS blocks
235 if (ileaved == target)
238 block = ileaved >> v->data_dev_block_bits;
239 bufio = v->fec->data_bufio;
241 if (block >= v->data_blocks) {
242 block -= v->data_blocks;
245 * blocks outside the area were assumed to contain
246 * zeros when encoding data was generated
248 if (unlikely(block >= v->fec->hash_blocks))
251 block += v->hash_start;
255 bbuf = dm_bufio_read(bufio, block, &buf);
256 if (unlikely(IS_ERR(bbuf))) {
257 DMWARN_LIMIT("%s: FEC %llu: read failed (%llu): %ld",
259 (unsigned long long)rsb,
260 (unsigned long long)block, PTR_ERR(bbuf));
262 /* assume the block is corrupted */
263 if (neras && *neras <= v->fec->roots)
264 fio->erasures[(*neras)++] = i;
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,
273 /* skip known zero blocks entirely */
278 * skip if we have already found the theoretical
279 * maximum number (i.e. fec->roots) of erasures
281 if (neras && *neras <= v->fec->roots &&
282 fec_is_erasure(v, io, want_digest, bbuf))
283 fio->erasures[(*neras)++] = i;
287 * deinterleave and copy the bytes that fit into bufs,
288 * starting from block_offset
290 fec_for_each_buffer_rs_block(fio, n, j) {
291 k = fec_buffer_rs_index(n, j) + block_offset;
293 if (k >= 1 << v->data_dev_block_bits)
296 rs_block = fec_buffer_rs_block(v, fio, n, j);
297 rs_block[i] = bbuf[k];
300 dm_bufio_release(buf);
307 * Allocate RS control structure and FEC buffers from preallocated mempools,
308 * and attempt to allocate as many extra buffers as available.
310 static int fec_alloc_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
315 fio->rs = mempool_alloc(&v->fec->rs_pool, GFP_NOIO);
317 fec_for_each_prealloc_buffer(n) {
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");
328 /* try to allocate the maximum number of buffers */
329 fec_for_each_extra_buffer(fio, n) {
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]))
341 fio->output = mempool_alloc(&v->fec->output_pool, GFP_NOIO);
347 * Initialize buffers and clear erasures. fec_read_bufs() assumes buffers are
348 * zeroed before deinterleaving.
350 static void fec_init_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
354 fec_for_each_buffer(fio, n)
355 memset(fio->bufs[n], 0, v->fec->rsn << DM_VERITY_FEC_BUF_RS_BITS);
357 memset(fio->erasures, 0, sizeof(fio->erasures));
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.
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,
372 r = fec_alloc_bufs(v, fio);
376 for (pos = 0; pos < 1 << v->data_dev_block_bits; ) {
377 fec_init_bufs(v, fio);
379 r = fec_read_bufs(v, io, rsb, offset, pos,
380 use_erasures ? &neras : NULL);
384 r = fec_decode_bufs(v, fio, rsb, r, pos, neras);
388 pos += fio->nbufs << DM_VERITY_FEC_BUF_RS_BITS;
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));
398 if (memcmp(verity_io_real_digest(v, io), verity_io_want_digest(v, io),
400 DMERR_LIMIT("%s: FEC %llu: failed to correct (%d erasures)",
401 v->data_dev->name, (unsigned long long)rsb, neras);
408 static int fec_bv_copy(struct dm_verity *v, struct dm_verity_io *io, u8 *data,
411 struct dm_verity_fec_io *fio = fec_io(io);
413 memcpy(data, &fio->output[fio->output_pos], len);
414 fio->output_pos += len;
420 * Correct errors in a block. Copies corrected block to dest if non-NULL,
421 * otherwise to a bio_vec starting from iter.
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)
428 struct dm_verity_fec_io *fio = fec_io(io);
429 u64 offset, res, rsb;
431 if (!verity_fec_is_enabled(v))
434 if (fio->level >= DM_VERITY_FEC_MAX_RECURSION) {
435 DMWARN_LIMIT("%s: FEC: recursion too deep", v->data_dev->name);
441 if (type == DM_VERITY_BLOCK_TYPE_METADATA)
442 block += v->data_blocks;
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.
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.
454 offset = block << v->data_dev_block_bits;
455 res = div64_u64(offset, v->fec->rounds << v->data_dev_block_bits);
458 * The base RS block we can feed to the interleaver to find out all
459 * blocks required for decoding.
461 rsb = offset - res * (v->fec->rounds << v->data_dev_block_bits);
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
468 r = fec_decode_rsb(v, io, fio, rsb, offset, false);
470 r = fec_decode_rsb(v, io, fio, rsb, offset, true);
476 memcpy(dest, fio->output, 1 << v->data_dev_block_bits);
479 r = verity_for_bv_block(v, io, iter, fec_bv_copy);
488 * Clean up per-bio data.
490 void verity_fec_finish_io(struct dm_verity_io *io)
493 struct dm_verity_fec *f = io->v->fec;
494 struct dm_verity_fec_io *fio = fec_io(io);
496 if (!verity_fec_is_enabled(io->v))
499 mempool_free(fio->rs, &f->rs_pool);
501 fec_for_each_prealloc_buffer(n)
502 mempool_free(fio->bufs[n], &f->prealloc_pool);
504 fec_for_each_extra_buffer(fio, n)
505 mempool_free(fio->bufs[n], &f->extra_pool);
507 mempool_free(fio->output, &f->output_pool);
511 * Initialize per-bio data.
513 void verity_fec_init_io(struct dm_verity_io *io)
515 struct dm_verity_fec_io *fio = fec_io(io);
517 if (!verity_fec_is_enabled(io->v))
521 memset(fio->bufs, 0, sizeof(fio->bufs));
528 * Append feature arguments and values to the status table.
530 unsigned verity_fec_status_table(struct dm_verity *v, unsigned sz,
531 char *result, unsigned maxlen)
533 if (!verity_fec_is_enabled(v))
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",
541 (unsigned long long)v->fec->blocks,
542 (unsigned long long)v->fec->start,
548 void verity_fec_dtr(struct dm_verity *v)
550 struct dm_verity_fec *f = v->fec;
552 if (!verity_fec_is_enabled(v))
555 mempool_exit(&f->rs_pool);
556 mempool_exit(&f->prealloc_pool);
557 mempool_exit(&f->extra_pool);
558 kmem_cache_destroy(f->cache);
561 dm_bufio_client_destroy(f->data_bufio);
563 dm_bufio_client_destroy(f->bufio);
566 dm_put_device(v->ti, f->dev);
572 static void *fec_rs_alloc(gfp_t gfp_mask, void *pool_data)
574 struct dm_verity *v = (struct dm_verity *)pool_data;
576 return init_rs_gfp(8, 0x11d, 0, 1, v->fec->roots, gfp_mask);
579 static void fec_rs_free(void *element, void *pool_data)
581 struct rs_control *rs = (struct rs_control *)element;
587 bool verity_is_fec_opt_arg(const char *arg_name)
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));
595 int verity_fec_parse_opt_args(struct dm_arg_set *as, struct dm_verity *v,
596 unsigned *argc, const char *arg_name)
599 struct dm_target *ti = v->ti;
600 const char *arg_value;
601 unsigned long long num_ll;
606 ti->error = "FEC feature arguments require a value";
610 arg_value = dm_shift_arg(as);
613 if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV)) {
614 r = dm_get_device(ti, arg_value, FMODE_READ, &v->fec->dev);
616 ti->error = "FEC device lookup failed";
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;
627 v->fec->blocks = num_ll;
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;
636 v->fec->start = num_ll;
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;
645 v->fec->roots = num_c;
648 ti->error = "Unrecognized verity FEC feature request";
656 * Allocate dm_verity_fec for v->fec. Must be called before verity_fec_ctr.
658 int verity_fec_ctr_alloc(struct dm_verity *v)
660 struct dm_verity_fec *f;
662 f = kzalloc(sizeof(struct dm_verity_fec), GFP_KERNEL);
664 v->ti->error = "Cannot allocate FEC structure";
673 * Validate arguments and preallocate memory. Must be called after arguments
674 * have been parsed using verity_fec_parse_opt_args.
676 int verity_fec_ctr(struct dm_verity *v)
678 struct dm_verity_fec *f = v->fec;
679 struct dm_target *ti = v->ti;
683 if (!verity_fec_is_enabled(v)) {
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:
693 * data blocks | hash blocks | metadata (optional)
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
699 * If metadata is included, we require it to be available on the
700 * hash device after the hash blocks.
703 hash_blocks = v->hash_blocks - v->hash_start;
706 * Require matching block sizes for data and hash devices for
709 if (v->data_dev_block_bits != v->hash_dev_block_bits) {
710 ti->error = "Block sizes must match to use FEC";
715 ti->error = "Missing " DM_VERITY_OPT_FEC_ROOTS;
718 f->rsn = DM_VERITY_FEC_RSM - f->roots;
721 ti->error = "Missing " DM_VERITY_OPT_FEC_BLOCKS;
725 f->rounds = f->blocks;
726 if (sector_div(f->rounds, f->rsn))
730 * Due to optional metadata, f->blocks can be larger than
731 * data_blocks and hash_blocks combined.
733 if (f->blocks < v->data_blocks + hash_blocks || !f->rounds) {
734 ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
739 * Metadata is accessed through the hash device, so we require
740 * it to be large enough.
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;
749 f->bufio = dm_bufio_client_create(f->dev->bdev,
750 1 << v->data_dev_block_bits,
752 if (IS_ERR(f->bufio)) {
753 ti->error = "Cannot initialize FEC bufio client";
754 return PTR_ERR(f->bufio);
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";
763 f->data_bufio = dm_bufio_client_create(v->data_dev->bdev,
764 1 << v->data_dev_block_bits,
766 if (IS_ERR(f->data_bufio)) {
767 ti->error = "Cannot initialize FEC data bufio client";
768 return PTR_ERR(f->data_bufio);
771 if (dm_bufio_get_device_size(f->data_bufio) < v->data_blocks) {
772 ti->error = "Data device is too small";
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);
780 ti->error = "Cannot allocate RS pool";
784 f->cache = kmem_cache_create("dm_verity_fec_buffers",
785 f->rsn << DM_VERITY_FEC_BUF_RS_BITS,
788 ti->error = "Cannot create FEC buffer cache";
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,
797 ti->error = "Cannot allocate FEC buffer prealloc pool";
801 ret = mempool_init_slab_pool(&f->extra_pool, 0, f->cache);
803 ti->error = "Cannot allocate FEC buffer extra pool";
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);
811 ti->error = "Cannot allocate FEC output pool";
815 /* Reserve space for our per-bio data */
816 ti->per_io_data_size += sizeof(struct dm_verity_fec_io);