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)
68 u64 position, block, rem;
71 position = (index + rsb) * v->fec->roots;
72 block = div64_u64_rem(position, v->fec->io_size, &rem);
73 *offset = (unsigned)rem;
75 res = dm_bufio_read(v->fec->bufio, 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)block, PTR_ERR(res));
86 /* Loop over each preallocated buffer slot. */
87 #define fec_for_each_prealloc_buffer(__i) \
88 for (__i = 0; __i < DM_VERITY_FEC_BUF_PREALLOC; __i++)
90 /* Loop over each extra buffer slot. */
91 #define fec_for_each_extra_buffer(io, __i) \
92 for (__i = DM_VERITY_FEC_BUF_PREALLOC; __i < DM_VERITY_FEC_BUF_MAX; __i++)
94 /* Loop over each allocated buffer. */
95 #define fec_for_each_buffer(io, __i) \
96 for (__i = 0; __i < (io)->nbufs; __i++)
98 /* Loop over each RS block in each allocated buffer. */
99 #define fec_for_each_buffer_rs_block(io, __i, __j) \
100 fec_for_each_buffer(io, __i) \
101 for (__j = 0; __j < 1 << DM_VERITY_FEC_BUF_RS_BITS; __j++)
104 * Return a pointer to the current RS block when called inside
105 * fec_for_each_buffer_rs_block.
107 static inline u8 *fec_buffer_rs_block(struct dm_verity *v,
108 struct dm_verity_fec_io *fio,
109 unsigned i, unsigned j)
111 return &fio->bufs[i][j * v->fec->rsn];
115 * Return an index to the current RS block when called inside
116 * fec_for_each_buffer_rs_block.
118 static inline unsigned fec_buffer_rs_index(unsigned i, unsigned j)
120 return (i << DM_VERITY_FEC_BUF_RS_BITS) + j;
124 * Decode all RS blocks from buffers and copy corrected bytes into fio->output
125 * starting from block_offset.
127 static int fec_decode_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio,
128 u64 rsb, int byte_index, unsigned block_offset,
131 int r, corrected = 0, res;
132 struct dm_buffer *buf;
133 unsigned n, i, offset;
136 par = fec_read_parity(v, rsb, block_offset, &offset, &buf);
141 * Decode the RS blocks we have in bufs. Each RS block results in
142 * one corrected target byte and consumes fec->roots parity bytes.
144 fec_for_each_buffer_rs_block(fio, n, i) {
145 block = fec_buffer_rs_block(v, fio, n, i);
146 res = fec_decode_rs8(v, fio, block, &par[offset], neras);
153 fio->output[block_offset] = block[byte_index];
156 if (block_offset >= 1 << v->data_dev_block_bits)
159 /* read the next block when we run out of parity bytes */
160 offset += v->fec->roots;
161 if (offset >= v->fec->io_size) {
162 dm_bufio_release(buf);
164 par = fec_read_parity(v, rsb, block_offset, &offset, &buf);
165 if (unlikely(IS_ERR(par)))
172 dm_bufio_release(buf);
175 DMERR_LIMIT("%s: FEC %llu: failed to correct: %d",
176 v->data_dev->name, (unsigned long long)rsb, r);
178 DMWARN_LIMIT("%s: FEC %llu: corrected %d errors",
179 v->data_dev->name, (unsigned long long)rsb, r);
185 * Locate data block erasures using verity hashes.
187 static int fec_is_erasure(struct dm_verity *v, struct dm_verity_io *io,
188 u8 *want_digest, u8 *data)
190 if (unlikely(verity_hash(v, verity_io_hash_req(v, io),
191 data, 1 << v->data_dev_block_bits,
192 verity_io_real_digest(v, io))))
195 return memcmp(verity_io_real_digest(v, io), want_digest,
196 v->digest_size) != 0;
200 * Read data blocks that are part of the RS block and deinterleave as much as
201 * fits into buffers. Check for erasure locations if @neras is non-NULL.
203 static int fec_read_bufs(struct dm_verity *v, struct dm_verity_io *io,
204 u64 rsb, u64 target, unsigned block_offset,
208 int i, j, target_index = -1;
209 struct dm_buffer *buf;
210 struct dm_bufio_client *bufio;
211 struct dm_verity_fec_io *fio = fec_io(io);
214 u8 want_digest[v->digest_size];
221 * read each of the rsn data blocks that are part of the RS block, and
222 * interleave contents to available bufs
224 for (i = 0; i < v->fec->rsn; i++) {
225 ileaved = fec_interleave(v, rsb * v->fec->rsn + i);
228 * target is the data block we want to correct, target_index is
229 * the index of this block within the rsn RS blocks
231 if (ileaved == target)
234 block = ileaved >> v->data_dev_block_bits;
235 bufio = v->fec->data_bufio;
237 if (block >= v->data_blocks) {
238 block -= v->data_blocks;
241 * blocks outside the area were assumed to contain
242 * zeros when encoding data was generated
244 if (unlikely(block >= v->fec->hash_blocks))
247 block += v->hash_start;
251 bbuf = dm_bufio_read(bufio, block, &buf);
252 if (unlikely(IS_ERR(bbuf))) {
253 DMWARN_LIMIT("%s: FEC %llu: read failed (%llu): %ld",
255 (unsigned long long)rsb,
256 (unsigned long long)block, PTR_ERR(bbuf));
258 /* assume the block is corrupted */
259 if (neras && *neras <= v->fec->roots)
260 fio->erasures[(*neras)++] = i;
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,
269 /* skip known zero blocks entirely */
274 * skip if we have already found the theoretical
275 * maximum number (i.e. fec->roots) of erasures
277 if (neras && *neras <= v->fec->roots &&
278 fec_is_erasure(v, io, want_digest, bbuf))
279 fio->erasures[(*neras)++] = i;
283 * deinterleave and copy the bytes that fit into bufs,
284 * starting from block_offset
286 fec_for_each_buffer_rs_block(fio, n, j) {
287 k = fec_buffer_rs_index(n, j) + block_offset;
289 if (k >= 1 << v->data_dev_block_bits)
292 rs_block = fec_buffer_rs_block(v, fio, n, j);
293 rs_block[i] = bbuf[k];
296 dm_bufio_release(buf);
303 * Allocate RS control structure and FEC buffers from preallocated mempools,
304 * and attempt to allocate as many extra buffers as available.
306 static int fec_alloc_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
311 fio->rs = mempool_alloc(&v->fec->rs_pool, GFP_NOIO);
313 fec_for_each_prealloc_buffer(n) {
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");
324 /* try to allocate the maximum number of buffers */
325 fec_for_each_extra_buffer(fio, n) {
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]))
337 fio->output = mempool_alloc(&v->fec->output_pool, GFP_NOIO);
343 * Initialize buffers and clear erasures. fec_read_bufs() assumes buffers are
344 * zeroed before deinterleaving.
346 static void fec_init_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)
350 fec_for_each_buffer(fio, n)
351 memset(fio->bufs[n], 0, v->fec->rsn << DM_VERITY_FEC_BUF_RS_BITS);
353 memset(fio->erasures, 0, sizeof(fio->erasures));
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.
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,
368 r = fec_alloc_bufs(v, fio);
372 for (pos = 0; pos < 1 << v->data_dev_block_bits; ) {
373 fec_init_bufs(v, fio);
375 r = fec_read_bufs(v, io, rsb, offset, pos,
376 use_erasures ? &neras : NULL);
380 r = fec_decode_bufs(v, fio, rsb, r, pos, neras);
384 pos += fio->nbufs << DM_VERITY_FEC_BUF_RS_BITS;
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));
394 if (memcmp(verity_io_real_digest(v, io), verity_io_want_digest(v, io),
396 DMERR_LIMIT("%s: FEC %llu: failed to correct (%d erasures)",
397 v->data_dev->name, (unsigned long long)rsb, neras);
404 static int fec_bv_copy(struct dm_verity *v, struct dm_verity_io *io, u8 *data,
407 struct dm_verity_fec_io *fio = fec_io(io);
409 memcpy(data, &fio->output[fio->output_pos], len);
410 fio->output_pos += len;
416 * Correct errors in a block. Copies corrected block to dest if non-NULL,
417 * otherwise to a bio_vec starting from iter.
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)
424 struct dm_verity_fec_io *fio = fec_io(io);
425 u64 offset, res, rsb;
427 if (!verity_fec_is_enabled(v))
430 if (fio->level >= DM_VERITY_FEC_MAX_RECURSION) {
431 DMWARN_LIMIT("%s: FEC: recursion too deep", v->data_dev->name);
437 if (type == DM_VERITY_BLOCK_TYPE_METADATA)
438 block = block - v->hash_start + v->data_blocks;
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.
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.
450 offset = block << v->data_dev_block_bits;
451 res = div64_u64(offset, v->fec->rounds << v->data_dev_block_bits);
454 * The base RS block we can feed to the interleaver to find out all
455 * blocks required for decoding.
457 rsb = offset - res * (v->fec->rounds << v->data_dev_block_bits);
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
464 r = fec_decode_rsb(v, io, fio, rsb, offset, false);
466 r = fec_decode_rsb(v, io, fio, rsb, offset, true);
472 memcpy(dest, fio->output, 1 << v->data_dev_block_bits);
475 r = verity_for_bv_block(v, io, iter, fec_bv_copy);
484 * Clean up per-bio data.
486 void verity_fec_finish_io(struct dm_verity_io *io)
489 struct dm_verity_fec *f = io->v->fec;
490 struct dm_verity_fec_io *fio = fec_io(io);
492 if (!verity_fec_is_enabled(io->v))
495 mempool_free(fio->rs, &f->rs_pool);
497 fec_for_each_prealloc_buffer(n)
498 mempool_free(fio->bufs[n], &f->prealloc_pool);
500 fec_for_each_extra_buffer(fio, n)
501 mempool_free(fio->bufs[n], &f->extra_pool);
503 mempool_free(fio->output, &f->output_pool);
507 * Initialize per-bio data.
509 void verity_fec_init_io(struct dm_verity_io *io)
511 struct dm_verity_fec_io *fio = fec_io(io);
513 if (!verity_fec_is_enabled(io->v))
517 memset(fio->bufs, 0, sizeof(fio->bufs));
524 * Append feature arguments and values to the status table.
526 unsigned verity_fec_status_table(struct dm_verity *v, unsigned sz,
527 char *result, unsigned maxlen)
529 if (!verity_fec_is_enabled(v))
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",
537 (unsigned long long)v->fec->blocks,
538 (unsigned long long)v->fec->start,
544 void verity_fec_dtr(struct dm_verity *v)
546 struct dm_verity_fec *f = v->fec;
548 if (!verity_fec_is_enabled(v))
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);
558 dm_bufio_client_destroy(f->data_bufio);
560 dm_bufio_client_destroy(f->bufio);
563 dm_put_device(v->ti, f->dev);
569 static void *fec_rs_alloc(gfp_t gfp_mask, void *pool_data)
571 struct dm_verity *v = (struct dm_verity *)pool_data;
573 return init_rs_gfp(8, 0x11d, 0, 1, v->fec->roots, gfp_mask);
576 static void fec_rs_free(void *element, void *pool_data)
578 struct rs_control *rs = (struct rs_control *)element;
584 bool verity_is_fec_opt_arg(const char *arg_name)
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));
592 int verity_fec_parse_opt_args(struct dm_arg_set *as, struct dm_verity *v,
593 unsigned *argc, const char *arg_name)
596 struct dm_target *ti = v->ti;
597 const char *arg_value;
598 unsigned long long num_ll;
603 ti->error = "FEC feature arguments require a value";
607 arg_value = dm_shift_arg(as);
610 if (!strcasecmp(arg_name, DM_VERITY_OPT_FEC_DEV)) {
611 r = dm_get_device(ti, arg_value, FMODE_READ, &v->fec->dev);
613 ti->error = "FEC device lookup failed";
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;
624 v->fec->blocks = num_ll;
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;
633 v->fec->start = num_ll;
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;
642 v->fec->roots = num_c;
645 ti->error = "Unrecognized verity FEC feature request";
653 * Allocate dm_verity_fec for v->fec. Must be called before verity_fec_ctr.
655 int verity_fec_ctr_alloc(struct dm_verity *v)
657 struct dm_verity_fec *f;
659 f = kzalloc(sizeof(struct dm_verity_fec), GFP_KERNEL);
661 v->ti->error = "Cannot allocate FEC structure";
670 * Validate arguments and preallocate memory. Must be called after arguments
671 * have been parsed using verity_fec_parse_opt_args.
673 int verity_fec_ctr(struct dm_verity *v)
675 struct dm_verity_fec *f = v->fec;
676 struct dm_target *ti = v->ti;
677 u64 hash_blocks, fec_blocks;
680 if (!verity_fec_is_enabled(v)) {
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:
690 * data blocks | hash blocks | metadata (optional)
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
696 * If metadata is included, we require it to be available on the
697 * hash device after the hash blocks.
700 hash_blocks = v->hash_blocks - v->hash_start;
703 * Require matching block sizes for data and hash devices for
706 if (v->data_dev_block_bits != v->hash_dev_block_bits) {
707 ti->error = "Block sizes must match to use FEC";
712 ti->error = "Missing " DM_VERITY_OPT_FEC_ROOTS;
715 f->rsn = DM_VERITY_FEC_RSM - f->roots;
718 ti->error = "Missing " DM_VERITY_OPT_FEC_BLOCKS;
722 f->rounds = f->blocks;
723 if (sector_div(f->rounds, f->rsn))
727 * Due to optional metadata, f->blocks can be larger than
728 * data_blocks and hash_blocks combined.
730 if (f->blocks < v->data_blocks + hash_blocks || !f->rounds) {
731 ti->error = "Invalid " DM_VERITY_OPT_FEC_BLOCKS;
736 * Metadata is accessed through the hash device, so we require
737 * it to be large enough.
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;
746 if ((f->roots << SECTOR_SHIFT) & ((1 << v->data_dev_block_bits) - 1))
747 f->io_size = 1 << v->data_dev_block_bits;
749 f->io_size = v->fec->roots << SECTOR_SHIFT;
751 f->bufio = dm_bufio_client_create(f->dev->bdev,
754 if (IS_ERR(f->bufio)) {
755 ti->error = "Cannot initialize FEC bufio client";
756 return PTR_ERR(f->bufio);
759 dm_bufio_set_sector_offset(f->bufio, f->start << (v->data_dev_block_bits - SECTOR_SHIFT));
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";
767 f->data_bufio = dm_bufio_client_create(v->data_dev->bdev,
768 1 << v->data_dev_block_bits,
770 if (IS_ERR(f->data_bufio)) {
771 ti->error = "Cannot initialize FEC data bufio client";
772 return PTR_ERR(f->data_bufio);
775 if (dm_bufio_get_device_size(f->data_bufio) < v->data_blocks) {
776 ti->error = "Data device is too small";
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);
784 ti->error = "Cannot allocate RS pool";
788 f->cache = kmem_cache_create("dm_verity_fec_buffers",
789 f->rsn << DM_VERITY_FEC_BUF_RS_BITS,
792 ti->error = "Cannot create FEC buffer cache";
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,
801 ti->error = "Cannot allocate FEC buffer prealloc pool";
805 ret = mempool_init_slab_pool(&f->extra_pool, 0, f->cache);
807 ti->error = "Cannot allocate FEC buffer extra pool";
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);
815 ti->error = "Cannot allocate FEC output pool";
819 /* Reserve space for our per-bio data */
820 ti->per_io_data_size += sizeof(struct dm_verity_fec_io);