Optimize dm-verity and fsverity using multibuffer hashing
Subsystems (6)
Eric Biggers
maintainerv5 cover Optimize dm-verity and fsverity using multibuffer hashing
On many modern CPUs, it is possible to compute the SHA-256 hash of two equal-length messages in about the same time as a single message, if all the instructions are interleaved. This is because each SHA-256 (and also most other cryptographic hash functions) is inherently serialized and therefore can't always take advantage of the CPU's full throughput.
An earlier attempt to support multibuffer hashing in Linux was based around the ahash API. That approach had some major issues, as does the alternative ahash-based approach proposed by Herbert (see my response at
https://lore.kernel.org/linux-crypto/[email protected]/).
This patchset instead takes a much simpler approach of just adding a synchronous API for hashing equal-length messages.
This works well for dm-verity and fsverity, which use Merkle trees and therefore hash large numbers of equal-length messages.
This patchset is organized as follows:
- Patch 1-3 add crypto_shash_finup_mb() and tests for it.
- Patch 4-5 implement finup_mb on x86_64 and arm64, using an interleaving factor of 2.
- Patch 6 adds multibuffer hashing support to fsverity.
- Patches 7-14 are cleanups and optimizations to dm-verity that prepare the way for adding multibuffer hashing support. These don't depend on any of the previous patches.
- Patch 15 adds multibuffer hashing support to dm-verity.
On CPUs that support multiple concurrent SHA-256's (all arm64 CPUs I tested, and AMD Zen CPUs), raw SHA-256 hashing throughput increases by 70-98%, and the throughput of cold-cache reads from dm-verity and fsverity increases by very roughly 35%.
Changed in v5:
- Reworked the dm-verity patches again. Split the preparation work into separate patches, fixed two bugs, and added some new cleanups.
- Other small cleanups
Changed in v4:
- Reorganized the fsverity and dm-verity code to have a unified code path for single-block vs. multi-block processing. For data blocks they now use only crypto_shash_finup_mb().
Changed in v3:
- Change API from finup2x to finup_mb. It now takes arrays of data buffer and output buffers, avoiding hardcoding 2x in the API.
Changed in v2:
- Rebase onto cryptodev/master
- Add more comments to assembly
- Reorganize some of the assembly slightly
- Fix the claimed throughput improvement on arm64
- Fix incorrect kunmap order in fs/verity/verify.c
- Adjust testmgr generation logic slightly
- Explicitly check for INT_MAX before casting unsigned int to int
- Mention SHA3 based parallel hashes
- Mention AVX512-based approach
Eric Biggers (15):
crypto: shash - add support for finup_mb
crypto: testmgr - generate power-of-2 lengths more often
crypto: testmgr - add tests for finup_mb
crypto: x86/sha256-ni - add support for finup_mb
crypto: arm64/sha256-ce - add support for finup_mb
fsverity: improve performance by using multibuffer hashing
dm-verity: move hash algorithm setup into its own function
dm-verity: move data hash mismatch handling into its own function
dm-verity: make real_digest and want_digest fixed-length
dm-verity: provide dma_alignment limit in io_hints
dm-verity: always "map" the data blocks
dm-verity: make verity_hash() take dm_verity_io instead of ahash_request
dm-verity: hash blocks with shash import+finup when possible
dm-verity: reduce scope of real and wanted digests
dm-verity: improve performance by using multibuffer hashing
arch/arm64/crypto/sha2-ce-core.S | 281 +++++++++++++- arch/arm64/crypto/sha2-ce-glue.c | 40 ++ arch/x86/crypto/sha256_ni_asm.S | 368 ++++++++++++++++++ arch/x86/crypto/sha256_ssse3_glue.c | 39 ++ crypto/shash.c | 58 +++ crypto/testmgr.c | 90 ++++- drivers/md/dm-verity-fec.c | 49 +-- drivers/md/dm-verity-fec.h | 9 +- drivers/md/dm-verity-target.c | 582 ++++++++++++++++------------ drivers/md/dm-verity.h | 66 ++-- fs/verity/fsverity_private.h | 7 + fs/verity/hash_algs.c | 8 +- fs/verity/verify.c | 170 ++++++-- include/crypto/hash.h | 52 ++- 14 files changed, 1440 insertions(+), 379 deletions(-)
base-commit: 6d4e1993a30539f556da2ebd36f1936c583eb812
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 1/15 crypto: shash - add support for finup_mb
From: Eric Biggers <[email protected]>
Most cryptographic hash functions are serialized, in the sense that they have an internal block size and the blocks must be processed serially. (BLAKE3 is a notable exception that has tree-based hashing built-in, but all the more common choices such as the SHAs and BLAKE2 are serialized. ParallelHash and Sakura are parallel hashes based on SHA3, but SHA3 is much slower than SHA256 in software even with the ARMv8 SHA3 extension.)
This limits the performance of computing a single hash. Yet, computing multiple hashes simultaneously does not have this limitation. Modern CPUs are superscalar and often can execute independent instructions in parallel. As a result, on many modern CPUs, it is possible to hash two equal-length messages in about the same time as a single message, if all the instructions are interleaved.
Meanwhile, a very common use case for hashing in the Linux kernel is dm-verity and fs-verity. Both use a Merkle tree that has a fixed block size, usually 4096 bytes with an empty or 32-byte salt prepended. The hash algorithm is usually SHA-256. Usually, many blocks need to be hashed at a time. This is an ideal scenario for multibuffer hashing.
Linux actually used to support SHA-256 multibuffer hashing on x86_64, before it was removed by commit ab8085c130ed ("crypto: x86 - remove SHA multibuffer routines and mcryptd"). However, it was integrated with the crypto API in a weird way, where it behaved as an asynchronous hash that queued up and executed all requests on a global queue. This made it very complex, buggy, and virtually unusable.
This patch takes a new approach of just adding an API crypto_shash_finup_mb() that synchronously computes the hash of multiple equal-length messages, starting from a common state that represents the (possibly empty) common prefix shared by the messages.
The new API is part of the "shash" algorithm type, as it does not make sense in "ahash". It does a "finup" operation rather than a "digest" operation in order to support the salt that is used by dm-verity and fs-verity. The data and output buffers are provided in arrays of length @num_msgs in order to make the API itself extensible to interleaving factors other than 2. (Though, initially only 2x will actually be used. There are some platforms in which a higher factor could help, but there are significant trade-offs.)
--- crypto/shash.c | 58 +++++++++++++++++++++++++++++++++++++++++++ include/crypto/hash.h | 52 +++++++++++++++++++++++++++++++++++++- 2 files changed, 109 insertions(+), 1 deletion(-)
crypto/shash.c +58 -0
@@ -73,10 +73,57 @@ int crypto_shash_finup(struct shash_desc *desc, const u8 *data, {return crypto_shash_alg (desc->tfm)->finup (desc, data, len, out); }EXPORT_SYMBOL_GPL (crypto_shash_finup); +static noinline_for_stack int +shash_finup_mb_fallback (struct shash_desc *desc,const u8 *const data[], +unsigned int len,u8 *const outs[], +unsigned int num_msgs) +{ +struct crypto_shash *tfm = desc->tfm; +SHASH_DESC_ON_STACK (desc2, tfm); +unsigned int i; +int err; + +for (i =0 ; i < num_msgs -1 ; i++) { + desc2->tfm = tfm; +memcpy (shash_desc_ctx (desc2),shash_desc_ctx (desc), +crypto_shash_descsize (tfm)); + err =crypto_shash_finup (desc2, data[i], len, outs[i]); +if (err) +return err; + } +return crypto_shash_finup (desc, data[i], len, outs[i]); +} + +int crypto_shash_finup_mb (struct shash_desc *desc,const u8 *const data[], +unsigned int len,u8 *const outs[], +unsigned int num_msgs) +{ +struct shash_alg *alg =crypto_shash_alg (desc->tfm); +int err; + +if (num_msgs ==1 ) +return crypto_shash_finup (desc, data[0 ], len, outs[0 ]); + +if (num_msgs ==0 ) +return 0 ; + +if (WARN_ON_ONCE (num_msgs > alg->mb_max_msgs)) + goto fallback; + + err = alg->finup_mb (desc, data, len, outs, num_msgs); +if (unlikely (err == -EOPNOTSUPP )) + goto fallback; +return err; + +fallback: +return shash_finup_mb_fallback (desc, data, len, outs, num_msgs); +} +EXPORT_SYMBOL_GPL (crypto_shash_finup_mb); +static int shash_default_digest (struct shash_desc *desc,const u8 *data,unsigned int len,u8 *out) {struct shash_alg *shash =crypto_shash_alg (desc->tfm);
@@ -312,10 +359,21 @@ static int shash_prepare_alg(struct shash_alg *alg)return -EINVAL ;if ((alg->export && !alg->import) || (alg->import && !alg->export))return -EINVAL ; +if (alg->mb_max_msgs >1 ) { +if (alg->mb_max_msgs >HASH_MAX_MB_MSGS ) +return -EINVAL ; +if (!alg->finup_mb) +return -EINVAL ; + }else { +if (alg->finup_mb) +return -EINVAL ; + alg->mb_max_msgs =1 ; + } + err =hash_prepare_alg (&alg->halg);if (err)return err; base->cra_type = &crypto_shash_type;
include/crypto/hash.h +51 -1
117 lines not on this page · raw message
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 2/15 crypto: testmgr - generate power-of-2 lengths more often
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
Implementations of hash functions often have special cases when lengths are a multiple of the hash function's internal block size (e.g. 64 for SHA-256, 128 for SHA-512). Currently, when the fuzz testing code generates lengths, it doesn't prefer any length mod 64 over any other. This limits the coverage of these special cases.
Therefore, this patch updates the fuzz testing code to generate power-of-2 lengths and divide messages exactly in half a bit more often.
--- crypto/testmgr.c | 16 ++++++++++++---- 1 file changed, 12 insertions(+), 4 deletions(-)
crypto/testmgr.c +12 -4
@@ -914,18 +914,24 @@ static unsigned int generate_random_length(struct rnd_state *rng, {unsigned int len =prandom_u32_below (rng, max_len +1 );switch (prandom_u32_below (rng,4 )) {case 0 : -return len %64 ; + len %=64 ; +break ;case 1 : -return len %256 ; + len %=256 ; +break ;case 2 : -return len %1024 ; + len %=1024 ; +break ;default : -return len; +break ; } +if (prandom_u32_below (rng,4 ) ==0 ) + len =rounddown_pow_of_two (len); +return len; }/* Flip a random bit in the given nonempty data buffer */ static void flip_random_bit (struct rnd_state *rng,u8 *buf,size_t size) {
@@ -1017,10 +1023,12 @@ static char *generate_random_sgl_divisions(struct rnd_state *rng,unsigned int this_len;const char *flushtype_str;if (div == &divs[max_divs -1 ] ||prandom_bool (rng)) this_len = remaining; +else if (prandom_u32_below (rng,4 ) ==0 ) + this_len = (remaining +1 ) /2 ;else this_len =prandom_u32_inclusive (rng,1 , remaining); div->proportion_of_total = this_len;if (prandom_u32_below (rng,4 ) ==0 )
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 3/15 crypto: testmgr - add tests for finup_mb
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
Update the shash self-tests to test the new finup_mb method when CONFIG_CRYPTO_MANAGER_EXTRA_TESTS=y.
--- crypto/testmgr.c | 74 +++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 67 insertions(+), 7 deletions(-)
crypto/testmgr.c +67 -7
@@ -227,10 +227,11 @@ enum flush_type {/* finalization function for hash algorithms */ enum finalization_type {FINALIZATION_TYPE_FINAL ,/* use final() */ FINALIZATION_TYPE_FINUP ,/* use finup() */ +FINALIZATION_TYPE_FINUP_MB ,/* use finup_mb() */ FINALIZATION_TYPE_DIGEST ,/* use digest() */ };/* * Whether the crypto operation will occur in-place, and if so whether the
@@ -290,10 +291,15 @@ struct test_sg_division {* the @iv_offset * @key_offset: misalignment of the key, where 0 is default alignment * @key_offset_relative_to_alignmask: if true, add the algorithm's alignmask to * the @key_offset * @finalization_type: what finalization function to use for hashes +* @multibuffer_index: random number used to generate the message index to use +* for finup_mb (when finup_mb is used). +* @multibuffer_count: random number used to generate the num_msgs parameter to +* finup_mb (when finup_mb is used). +* * @nosimd: execute with SIMD disabled? Requires !CRYPTO_TFM_REQ_MAY_SLEEP. * This applies to the parts of the operation that aren't controlled * individually by @nosimd_setkey or @src_divs[].nosimd. * @nosimd_setkey: set the key (if applicable) with SIMD disabled? Requires * !CRYPTO_TFM_REQ_MAY_SLEEP.
@@ -307,10 +313,12 @@ struct testvec_config {unsigned int iv_offset;unsigned int key_offset;bool iv_offset_relative_to_alignmask;bool key_offset_relative_to_alignmask;enum finalization_type finalization_type; +unsigned int multibuffer_index; +unsigned int multibuffer_count;bool nosimd;bool nosimd_setkey; };#define TESTVEC_CONFIG_NAMELEN 192
@@ -1122,19 +1130,27 @@ static void generate_random_testvec_config(struct rnd_state *rng,if (prandom_bool (rng)) { cfg->req_flags |=CRYPTO_TFM_REQ_MAY_SLEEP ; p +=scnprintf (p, end - p," may_sleep" ); } -switch (prandom_u32_below (rng,4 )) { +switch (prandom_u32_below (rng,8 )) {case 0 : +case 1 : cfg->finalization_type =FINALIZATION_TYPE_FINAL ; p +=scnprintf (p, end - p," use_final" );break ; -case 1 : +case 2 : cfg->finalization_type =FINALIZATION_TYPE_FINUP ; p +=scnprintf (p, end - p," use_finup" );break ; +case 3 : +case 4 : + cfg->finalization_type =FINALIZATION_TYPE_FINUP_MB ; + cfg->multibuffer_index =prandom_u32_state (rng); + cfg->multibuffer_count =prandom_u32_state (rng); + p +=scnprintf (p, end - p," use_finup_mb" ); +break ;default : cfg->finalization_type =FINALIZATION_TYPE_DIGEST ; p +=scnprintf (p, end - p," use_digest" );break ; }
@@ -1289,10 +1305,37 @@ static inline int check_shash_op(const char *op, int err,pr_err ("alg: shash: %s %s() failed with err %d on test vector %s, cfg=\"%s\"\n" , driver, op, err, vec_name, cfg->name);return err; } +static int do_finup_mb (struct shash_desc *desc, +const u8 *data,unsigned int len,u8 *result, +const struct testvec_config *cfg, +const struct test_sglist *tsgl) +{ +struct crypto_shash *tfm = desc->tfm; +const u8 *unused_data = tsgl->bufs[XBUFSIZE -1 ]; +u8 unused_result[HASH_MAX_DIGESTSIZE ]; +const u8 *datas[HASH_MAX_MB_MSGS ]; +u8 *outs[HASH_MAX_MB_MSGS ]; +unsigned int num_msgs; +unsigned int msg_idx; +unsigned int i; + + num_msgs =1 + (cfg->multibuffer_count %crypto_shash_mb_max_msgs (tfm)); +if (WARN_ON_ONCE (num_msgs >HASH_MAX_MB_MSGS )) +return -EINVAL ; + msg_idx = cfg->multibuffer_index % num_msgs; +for (i =0 ; i < num_msgs; i++) { + datas[i] = unused_data; + outs[i] = unused_result; + } + datas[msg_idx] = data; + outs[msg_idx] = result; +return crypto_shash_finup_mb (desc, datas, len, outs, num_msgs); +} +/* Test one hash test vector in one configuration, using the shash API */ static int test_shash_vec_cfg(const struct hash_testvec *vec, const char *vec_name, const struct testvec_config *cfg, struct shash_desc *desc,
@@ -1365,11 +1408,14 @@ static int test_shash_vec_cfg(const struct hash_testvec *vec,return -EINVAL ; } goto result_ready; } -/* Using init(), zero or more update(), then final() or finup() */ +/* +* Using init(), zero or more update(), then either final(), finup(), or +* finup_mb(). +*/ if (cfg->nosimd)crypto_disable_simd_for_test (); err =crypto_shash_init (desc);if (cfg->nosimd)
@@ -1377,28 +1423,42 @@ static int test_shash_vec_cfg(const struct hash_testvec *vec, err =check_shash_op ("init" , err, driver, vec_name, cfg);if (err)return err;for (i =0 ; i < tsgl->nents; i++) { +const u8 *data =sg_virt (&tsgl->sgl[i]); +unsigned int len = tsgl->sgl[i].length; +if (i +1 == tsgl->nents && cfg->finalization_type ==FINALIZATION_TYPE_FINUP ) {if (divs[i]->nosimd)crypto_disable_simd_for_test (); - err =crypto_shash_finup (desc,sg_virt (&tsgl->sgl[i]), - tsgl->sgl[i].length, result); + err =crypto_shash_finup (desc, data, len, result);if (divs[i]->nosimd)crypto_reenable_simd_for_test (); err =check_shash_op ("finup" , err, driver, vec_name, cfg);if (err)return err; goto result_ready; } +if (i +1 == tsgl->nents && + cfg->finalization_type ==FINALIZATION_TYPE_FINUP_MB ) { +if (divs[i]->nosimd) +crypto_disable_simd_for_test (); + err =do_finup_mb (desc, data, len, result, cfg, tsgl); +if (divs[i]->nosimd) +crypto_reenable_simd_for_test (); + err =check_shash_op ("finup_mb" , err, driver, vec_name, + cfg); +if (err) +return err; + goto result_ready; + }if (divs[i]->nosimd)crypto_disable_simd_for_test (); - err =crypto_shash_update (desc,sg_virt (&tsgl->sgl[i]), - tsgl->sgl[i].length); + err =crypto_shash_update (desc, data, len);if (divs[i]->nosimd)crypto_reenable_simd_for_test (); err =check_shash_op ("update" , err, driver, vec_name, cfg);if (err)return err;
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 4/15 crypto: x86/sha256-ni - add support for finup_mb
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
Add an implementation of finup_mb to sha256-ni, using an interleaving factor of 2. It interleaves a finup operation for two equal-length messages that share a common prefix. dm-verity and fs-verity will take advantage of this for greatly improved performance on capable CPUs.
This increases the throughput of SHA-256 hashing 4096-byte messages by the following amounts on the following CPUs:
AMD Zen 1: 84%
AMD Zen 4: 98%
Intel Ice Lake: 4%
Intel Sapphire Rapids: 20%
For now, this seems to benefit AMD much more than Intel. This seems to be because current AMD CPUs support concurrent execution of the SHA-NI instructions, but unfortunately current Intel CPUs don't, except for the sha256msg2 instruction. Hopefully future Intel CPUs will support SHA-NI on more execution ports. Zen 1 supports 2 concurrent sha256rnds2, and Zen 4 supports 4 concurrent sha256rnds2, which suggests that even better performance may be achievable on Zen 4 by interleaving more than two hashes; however, doing so poses a number of trade-offs.
It's been reported that the method that achieves the highest SHA-256 throughput on Intel CPUs is actually computing 16 hashes simultaneously using AVX512. That method would be quite different to the SHA-NI method used in this patch. However, such a high interleaving factor isn't practical for the use cases being targeted in the kernel.
--- arch/x86/crypto/sha256_ni_asm.S | 368 ++++++++++++++++++++++++++++ arch/x86/crypto/sha256_ssse3_glue.c | 39 +++ 2 files changed, 407 insertions(+)
arch/x86/crypto/sha256_ni_asm.S +368 -0
@@ -172,10 +172,378 @@ SYM_TYPED_FUNC_START(sha256_ni_transform) .Ldone_hash: RET SYM_FUNC_END(sha256_ni_transform) +#undef DIGEST_PTR +#undef DATA_PTR +#undef NUM_BLKS +#undef SHA256CONSTANTS +#undef MSG +#undef STATE0 +#undef STATE1 +#undef MSG0 +#undef MSG1 +#undef MSG2 +#undef MSG3 +#undef TMP +#undef SHUF_MASK +#undef ABEF_SAVE +#undef CDGH_SAVE + +// parameters for __sha256_ni_finup2x() +#define SCTX %rdi +#define DATA1 %rsi +#define DATA2 %rdx +#define LEN %ecx +#define LEN8 %cl +#define LEN64 %rcx +#define OUT1 %r8 +#define OUT2 %r9 + +// other scalar variables +#define SHA256CONSTANTS %rax +#define COUNT %r10 +#define COUNT32 %r10d +#define FINAL_STEP %r11d + +// rbx is used as a temporary. + +#define MSG %xmm0 // sha256rnds2 implicit operand +#define STATE0_A %xmm1 +#define STATE1_A %xmm2 +#define STATE0_B %xmm3 +#define STATE1_B %xmm4 +#define TMP_A %xmm5 +#define TMP_B %xmm6 +#define MSG0_A %xmm7 +#define MSG1_A %xmm8 +#define MSG2_A %xmm9 +#define MSG3_A %xmm10 +#define MSG0_B %xmm11 +#define MSG1_B %xmm12 +#define MSG2_B %xmm13 +#define MSG3_B %xmm14 +#define SHUF_MASK %xmm15 + +#define OFFSETOF_STATE 0 // offsetof(struct sha256_state, state) +#define OFFSETOF_COUNT 32 // offsetof(struct sha256_state, count) +#define OFFSETOF_BUF 40 // offsetof(struct sha256_state, buf) + +// Do 4 rounds of SHA-256 for each of two messages (interleaved). m0_a and m0_b +// contain the current 4 message schedule words for the first and second message +// respectively. +// +// If not all the message schedule words have been computed yet, then this also +// computes 4 more message schedule words for each message. m1_a-m3_a contain +// the next 3 groups of 4 message schedule words for the first message, and +// likewise m1_b-m3_b for the second. After consuming the current value of +// m0_a, this macro computes the group after m3_a and writes it to m0_a, and +// likewise for *_b. This means that the next (m0_a, m1_a, m2_a, m3_a) is the +// current (m1_a, m2_a, m3_a, m0_a), and likewise for *_b, so the caller must +// cycle through the registers accordingly. +.macro do_4rounds_2x i, m0_a, m1_a, m2_a, m3_a, m0_b, m1_b, m2_b, m3_b + movdqa (\i-32)*4(SHA256CONSTANTS), TMP_A + movdqa TMP_A, TMP_B + paddd \m0_a, TMP_A + paddd \m0_b, TMP_B +.if \i < 48 + sha256msg1 \m1_a, \m0_a + sha256msg1 \m1_b, \m0_b +.endif + movdqa TMP_A, MSG + sha256rnds2 STATE0_A, STATE1_A + movdqa TMP_B, MSG + sha256rnds2 STATE0_B, STATE1_B + pshufd $0x0E, TMP_A, MSG + sha256rnds2 STATE1_A, STATE0_A + pshufd $0x0E, TMP_B, MSG + sha256rnds2 STATE1_B, STATE0_B +.if \i < 48 + movdqa \m3_a, TMP_A + movdqa \m3_b, TMP_B + palignr $4, \m2_a, TMP_A + palignr $4, \m2_b, TMP_B + paddd TMP_A, \m0_a + paddd TMP_B, \m0_b + sha256msg2 \m3_a, \m0_a + sha256msg2 \m3_b, \m0_b +.endif +.endm + +// +// void __sha256_ni_finup2x(const struct sha256_state *sctx, +// const u8 *data1, const u8 *data2, int len, +// u8 out1[SHA256_DIGEST_SIZE], +// u8 out2[SHA256_DIGEST_SIZE]); +// +// This function computes the SHA-256 digests of two messages |data1| and +// |data2| that are both |len| bytes long, starting from the initial state +// |sctx|. |len| must be at least SHA256_BLOCK_SIZE. +// +// The instructions for the two SHA-256 operations are interleaved. On many +// CPUs, this is almost twice as fast as hashing each message individually due +// to taking better advantage of the CPU's SHA-256 and SIMD throughput. +// +SYM_FUNC_START(__sha256_ni_finup2x) + // Allocate 128 bytes of stack space, 16-byte aligned. + push %rbx + push %rbp + mov %rsp, %rbp + sub $128, %rsp + and $~15, %rsp + + // Load the shuffle mask for swapping the endianness of 32-bit words. + movdqa PSHUFFLE_BYTE_FLIP_MASK(%rip), SHUF_MASK + + // Set up pointer to the round constants. + lea K256+32*4(%rip), SHA256CONSTANTS + + // Initially we're not processing the final blocks. + xor FINAL_STEP, FINAL_STEP + + // Load the initial state from sctx->state. + movdqu OFFSETOF_STATE+0*16(SCTX), STATE0_A // DCBA + movdqu OFFSETOF_STATE+1*16(SCTX), STATE1_A // HGFE + movdqa STATE0_A, TMP_A + punpcklqdq STATE1_A, STATE0_A // FEBA + punpckhqdq TMP_A, STATE1_A // DCHG + pshufd $0x1B, STATE0_A, STATE0_A // ABEF + pshufd $0xB1, STATE1_A, STATE1_A // CDGH + + // Load sctx->count. Take the mod 64 of it to get the number of bytes + // that are buffered in sctx->buf. Also save it in a register with LEN + // added to it. + mov LEN, LEN + mov OFFSETOF_COUNT(SCTX), %rbx + lea (%rbx, LEN64, 1), COUNT + and $63, %ebx + jz .Lfinup2x_enter_loop // No bytes buffered? + + // %ebx bytes (1 to 63) are currently buffered in sctx->buf. Load them + // followed by the first 64 - %ebx bytes of data. Since LEN >= 64, we + // just load 64 bytes from each of sctx->buf, DATA1, and DATA2 + // unconditionally and rearrange the data as needed. + + movdqu OFFSETOF_BUF+0*16(SCTX), MSG0_A + movdqu OFFSETOF_BUF+1*16(SCTX), MSG1_A + movdqu OFFSETOF_BUF+2*16(SCTX), MSG2_A + movdqu OFFSETOF_BUF+3*16(SCTX), MSG3_A + movdqa MSG0_A, 0*16(%rsp) + movdqa MSG1_A, 1*16(%rsp) + movdqa MSG2_A, 2*16(%rsp) + movdqa MSG3_A, 3*16(%rsp) + + movdqu 0*16(DATA1), MSG0_A + movdqu 1*16(DATA1), MSG1_A + movdqu 2*16(DATA1), MSG2_A + movdqu 3*16(DATA1), MSG3_A + movdqu MSG0_A, 0*16(%rsp,%rbx) + movdqu MSG1_A, 1*16(%rsp,%rbx) + movdqu MSG2_A, 2*16(%rsp,%rbx) + movdqu MSG3_A, 3*16(%rsp,%rbx) + movdqa 0*16(%rsp), MSG0_A + movdqa 1*16(%rsp), MSG1_A + movdqa 2*16(%rsp), MSG2_A + movdqa 3*16(%rsp), MSG3_A + + movdqu 0*16(DATA2), MSG0_B + movdqu 1*16(DATA2), MSG1_B + movdqu 2*16(DATA2), MSG2_B + movdqu 3*16(DATA2), MSG3_B + movdqu MSG0_B, 0*16(%rsp,%rbx) + movdqu MSG1_B, 1*16(%rsp,%rbx) + movdqu MSG2_B, 2*16(%rsp,%rbx) + movdqu MSG3_B, 3*16(%rsp,%rbx) + movdqa 0*16(%rsp), MSG0_B + movdqa 1*16(%rsp), MSG1_B + movdqa 2*16(%rsp), MSG2_B + movdqa 3*16(%rsp), MSG3_B + + sub $64, %rbx // rbx = buffered - 64 + sub %rbx, DATA1 // DATA1 += 64 - buffered + sub %rbx, DATA2 // DATA2 += 64 - buffered + add %ebx, LEN // LEN += buffered - 64 + movdqa STATE0_A, STATE0_B + movdqa STATE1_A, STATE1_B + jmp .Lfinup2x_loop_have_data + +.Lfinup2x_enter_loop: + sub $64, LEN + movdqa STATE0_A, STATE0_B + movdqa STATE1_A, STATE1_B +.Lfinup2x_loop: + // Load the next two data blocks. + movdqu 0*16(DATA1), MSG0_A + movdqu 0*16(DATA2), MSG0_B + movdqu 1*16(DATA1), MSG1_A + movdqu 1*16(DATA2), MSG1_B + movdqu 2*16(DATA1), MSG2_A + movdqu 2*16(DATA2), MSG2_B + movdqu 3*16(DATA1), MSG3_A + movdqu 3*16(DATA2), MSG3_B + add $64, DATA1 + add $64, DATA2 +.Lfinup2x_loop_have_data: + // Convert the words of the data blocks from big endian. + pshufb SHUF_MASK, MSG0_A + pshufb SHUF_MASK, MSG0_B + pshufb SHUF_MASK, MSG1_A + pshufb SHUF_MASK, MSG1_B + pshufb SHUF_MASK, MSG2_A + pshufb SHUF_MASK, MSG2_B + pshufb SHUF_MASK, MSG3_A + pshufb SHUF_MASK, MSG3_B +.Lfinup2x_loop_have_bswapped_data: + + // Save the original state for each block. + movdqa STATE0_A, 0*16(%rsp) + movdqa STATE0_B, 1*16(%rsp) + movdqa STATE1_A, 2*16(%rsp) + movdqa STATE1_B, 3*16(%rsp) + + // Do the SHA-256 rounds on each block. +.irp i, 0, 16, 32, 48 + do_4rounds_2x (\i + 0), MSG0_A, MSG1_A, MSG2_A, MSG3_A, \ + MSG0_B, MSG1_B, MSG2_B, MSG3_B + do_4rounds_2x (\i + 4), MSG1_A, MSG2_A, MSG3_A, MSG0_A, \ + MSG1_B, MSG2_B, MSG3_B, MSG0_B + do_4rounds_2x (\i + 8), MSG2_A, MSG3_A, MSG0_A, MSG1_A, \ + MSG2_B, MSG3_B, MSG0_B, MSG1_B + do_4rounds_2x (\i + 12), MSG3_A, MSG0_A, MSG1_A, MSG2_A, \ + MSG3_B, MSG0_B, MSG1_B, MSG2_B +.endr + + // Add the original state for each block. + paddd 0*16(%rsp), STATE0_A + paddd 1*16(%rsp), STATE0_B + paddd 2*16(%rsp), STATE1_A + paddd 3*16(%rsp), STATE1_B + + // Update LEN and loop back if more blocks remain. + sub $64, LEN + jge .Lfinup2x_loop + + // Check if any final blocks need to be handled. + // FINAL_STEP = 2: all done + // FINAL_STEP = 1: need to do count-only padding block + // FINAL_STEP = 0: need to do the block with 0x80 padding byte + cmp $1, FINAL_STEP + jg .Lfinup2x_done + je .Lfinup2x_finalize_countonly + add $64, LEN + jz .Lfinup2x_finalize_blockaligned + + // Not block-aligned; 1 <= LEN <= 63 data bytes remain. Pad the block. + // To do this, write the padding starting with the 0x80 byte to + // &sp[64]. Then for each message, copy the last 64 data bytes to sp + // and load from &sp[64 - LEN] to get the needed padding block. This + // code relies on the data buffers being >= 64 bytes in length. + mov $64, %ebx + sub LEN, %ebx // ebx = 64 - LEN + sub %rbx, DATA1 // DATA1 -= 64 - LEN + sub %rbx, DATA2 // DATA2 -= 64 - LEN + mov $0x80, FINAL_STEP // using FINAL_STEP as a temporary + movd FINAL_STEP, MSG0_A + pxor MSG1_A, MSG1_A + movdqa MSG0_A, 4*16(%rsp) + movdqa MSG1_A, 5*16(%rsp) + movdqa MSG1_A, 6*16(%rsp) + movdqa MSG1_A, 7*16(%rsp) + cmp $56, LEN + jge 1f // will COUNT spill into its own block? + shl $3, COUNT + bswap COUNT + mov COUNT, 56(%rsp,%rbx) + mov $2, FINAL_STEP // won't need count-only block + jmp 2f +1: + mov $1, FINAL_STEP // will need count-only block +2: + movdqu 0*16(DATA1), MSG0_A + movdqu 1*16(DATA1), MSG1_A + movdqu 2*16(DATA1), MSG2_A + movdqu 3*16(DATA1), MSG3_A + movdqa MSG0_A, 0*16(%rsp) + movdqa MSG1_A, 1*16(%rsp) + movdqa MSG2_A, 2*16(%rsp) + movdqa MSG3_A, 3*16(%rsp) + movdqu 0*16(%rsp,%rbx), MSG0_A + movdqu 1*16(%rsp,%rbx), MSG1_A + movdqu 2*16(%rsp,%rbx), MSG2_A + movdqu 3*16(%rsp,%rbx), MSG3_A + + movdqu 0*16(DATA2), MSG0_B + movdqu 1*16(DATA2), MSG1_B + movdqu 2*16(DATA2), MSG2_B + movdqu 3*16(DATA2), MSG3_B + movdqa MSG0_B, 0*16(%rsp) + movdqa MSG1_B, 1*16(%rsp) + movdqa MSG2_B, 2*16(%rsp) + movdqa MSG3_B, 3*16(%rsp) + movdqu 0*16(%rsp,%rbx), MSG0_B + movdqu 1*16(%rsp,%rbx), MSG1_B + movdqu 2*16(%rsp,%rbx), MSG2_B + movdqu 3*16(%rsp,%rbx), MSG3_B + jmp .Lfinup2x_loop_have_data + + // Prepare a padding block, either: + // + // {0x80, 0, 0, 0, ..., count (as __be64)} + // This is for a block aligned message. + // + // { 0, 0, 0, 0, ..., count (as __be64)} + // This is for a message whose length mod 64 is >= 56. + // + // Pre-swap the endianness of the words. +.Lfinup2x_finalize_countonly: + pxor MSG0_A, MSG0_A + jmp 1f + +.Lfinup2x_finalize_blockaligned: + mov $0x80000000, %ebx + movd %ebx, MSG0_A +1: + pxor MSG1_A, MSG1_A + pxor MSG2_A, MSG2_A + ror $29, COUNT + movq COUNT, MSG3_A + pslldq $8, MSG3_A + movdqa MSG0_A, MSG0_B + pxor MSG1_B, MSG1_B + pxor MSG2_B, MSG2_B + movdqa MSG3_A, MSG3_B + mov $2, FINAL_STEP + jmp .Lfinup2x_loop_have_bswapped_data + +.Lfinup2x_done: + // Write the two digests with all bytes in the correct order. + movdqa STATE0_A, TMP_A + movdqa STATE0_B, TMP_B + punpcklqdq STATE1_A, STATE0_A // GHEF + punpcklqdq STATE1_B, STATE0_B + punpckhqdq TMP_A, STATE1_A // ABCD + punpckhqdq TMP_B, STATE1_B + pshufd $0xB1, STATE0_A, STATE0_A // HGFE + pshufd $0xB1, STATE0_B, STATE0_B + pshufd $0x1B, STATE1_A, STATE1_A // DCBA + pshufd $0x1B, STATE1_B, STATE1_B + pshufb SHUF_MASK, STATE0_A + pshufb SHUF_MASK, STATE0_B + pshufb SHUF_MASK, STATE1_A + pshufb SHUF_MASK, STATE1_B + movdqu STATE0_A, 1*16(OUT1) + movdqu STATE0_B, 1*16(OUT2) + movdqu STATE1_A, 0*16(OUT1) + movdqu STATE1_B, 0*16(OUT2) + + mov %rbp, %rsp + pop %rbp + pop %rbx + RET +SYM_FUNC_END(__sha256_ni_finup2x) + .section .rodata.cst256.K256, "aM", @progbits, 256 .align 64 K256: .long 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5 .long 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
arch/x86/crypto/sha256_ssse3_glue.c +39 -0
@@ -331,10 +331,15 @@ static void unregister_sha256_avx2(void)#ifdef CONFIG_AS_SHA256_NI asmlinkage void sha256_ni_transform(struct sha256_state *digest,const u8 *data,int rounds); +asmlinkage void __sha256_ni_finup2x(const struct sha256_state *sctx, +const u8 *data1,const u8 *data2,int len, +u8 out1[SHA256_DIGEST_SIZE ], +u8 out2[SHA256_DIGEST_SIZE ]); +static int sha256_ni_update (struct shash_desc *desc,const u8 *data,unsigned int len) {return _sha256_update (desc, data, len, sha256_ni_transform); }
@@ -355,18 +360,52 @@ static int sha256_ni_digest(struct shash_desc *desc, const u8 *data, {return sha256_base_init (desc) ?:sha256_ni_finup (desc, data, len, out); } +static int sha256_ni_finup_mb (struct shash_desc *desc, +const u8 *const data[],unsigned int len, +u8 *const outs[],unsigned int num_msgs) +{ +struct sha256_state *sctx =shash_desc_ctx (desc); + +/* +* num_msgs != 2 should not happen here, since this algorithm sets +* mb_max_msgs=2, and the crypto API handles num_msgs <= 1 before +* calling into the algorithm's finup_mb method. +*/ +if (WARN_ON_ONCE (num_msgs !=2 )) +return -EOPNOTSUPP ; + +if (unlikely (!crypto_simd_usable ())) +return -EOPNOTSUPP ; + +/* __sha256_ni_finup2x() assumes SHA256_BLOCK_SIZE <= len <= INT_MAX. */ +if (unlikely (len <SHA256_BLOCK_SIZE || len >INT_MAX )) +return -EOPNOTSUPP ; + +/* __sha256_ni_finup2x() assumes the following offsets. */ +BUILD_BUG_ON (offsetof(struct sha256_state , state) !=0 ); +BUILD_BUG_ON (offsetof(struct sha256_state , count) !=32 ); +BUILD_BUG_ON (offsetof(struct sha256_state , buf) !=40 ); + +kernel_fpu_begin (); +__sha256_ni_finup2x (sctx, data[0 ], data[1 ], len, outs[0 ], outs[1 ]); +kernel_fpu_end (); +return 0 ; +} +static struct shash_alg sha256_ni_algs[] = { { .digestsize =SHA256_DIGEST_SIZE , .init = sha256_base_init, .update = sha256_ni_update, .final = sha256_ni_final, .finup = sha256_ni_finup, .digest = sha256_ni_digest, + .finup_mb = sha256_ni_finup_mb, .descsize =sizeof (struct sha256_state ), + .mb_max_msgs =2 , .base = { .cra_name ="sha256" , .cra_driver_name ="sha256-ni" , .cra_priority =250 , .cra_blocksize =SHA256_BLOCK_SIZE ,
Signature · 1 line
2.45.1
Herbert Xu
maintainerRe: [PATCH v5 04/15] crypto: x86/sha256-ni - add support for finup_mb
Quoted 5 lines · Eric Biggers
From: Eric Biggers <[email protected]>
Add an implementation of
finup_mbto sha256-ni, using an interleaving factor of 2. It interleaves a finup operation for two equal-length messages that share a common prefix. dm-verity and fs-verity will take
I think the limitation on equal length is artificial. There is no reason why the code couldn't handle two messages with different lengths. Simply execute in dual mode up until the shorter message runs out. Then carry on as if you have a single message.
In fact, there is no reason why the two hashes have to start from the same initial state either. It has no bearing on the performance of the actual hashing as far as I can see.
Cheers,
Signature · 3 lines
Email: Herbert Xu <[email protected]> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
Eric Biggers
maintainerRe: [PATCH v5 04/15] crypto: x86/sha256-ni - add support for finup_mb
Quoted 11 lines · Herbert Xu
Quoted 5 lines · Eric Biggers
From: Eric Biggers <[email protected]>
Add an implementation of
finup_mbto sha256-ni, using an interleaving factor of 2. It interleaves a finup operation for two equal-length messages that share a common prefix. dm-verity and fs-verity will takeI think the limitation on equal length is artificial. There is no reason why the code couldn't handle two messages with different lengths. Simply execute in dual mode up until the shorter message runs out. Then carry on as if you have a single message.
Sure, as I mentioned the algorithm could fall back to single-buffer hashing once the messages get out of sync. This would actually have to be implemented and tested, of course, which gets especially tricky with your proposal to support arbitrary scatterlists. And there's no actual use case for adding that complexity yet.
Quoted 3 lines
In fact, there is no reason why the two hashes have to start from the same initial state either. It has no bearing on the performance of the actual hashing as far as I can see.
The SHA-256 inner loop would indeed be the same, but the single state has several advantages:
- The caller only needs to allocate and prepare a single state. This saves per-IO memory and reduces overhead.
- The glue code doesn't need to check that the number of internally buffered bytes are synced up.
- The assembly code only needs to load from the one state.
All of this simplifies the code slightly and boosts performance slightly.
These advantages aren't *too* large, of course, and if a use case for supporting update arose, then support for multiple states would be added. But it doesn't make sense to add this functionality prematurely before it actually has a user.
- Eric
Eric Biggers
maintainerPATCH v5 5/15 crypto: arm64/sha256-ce - add support for finup_mb
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
Add an implementation of finup_mb to sha256-ce, using an interleaving factor of 2. It interleaves a finup operation for two equal-length messages that share a common prefix. dm-verity and fs-verity will take advantage of this for greatly improved performance on capable CPUs.
On an ARM Cortex-X1, this increases the throughput of SHA-256 hashing 4096-byte messages by 70%.
--- arch/arm64/crypto/sha2-ce-core.S | 281 ++++++++++++++++++++++++++++++- arch/arm64/crypto/sha2-ce-glue.c | 40 +++++ 2 files changed, 315 insertions(+), 6 deletions(-)
arch/arm64/crypto/sha2-ce-core.S +275 -6
@@ -68,22 +68,26 @@ .word 0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5 .word 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3 .word 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208 .word 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2 + .macro load_round_constants tmp + adr_l \tmp, .Lsha2_rcon + ld1 { v0.4s- v3.4s}, [\tmp], #64 + ld1 { v4.4s- v7.4s}, [\tmp], #64 + ld1 { v8.4s-v11.4s}, [\tmp], #64 + ld1 {v12.4s-v15.4s}, [\tmp] + .endm + /* * int __sha256_ce_transform(struct sha256_ce_state *sst, u8 const *src, * int blocks) */ .text SYM_FUNC_START(__sha256_ce_transform) - /* load round constants */ - adr_l x8, .Lsha2_rcon - ld1 { v0.4s- v3.4s}, [x8], #64 - ld1 { v4.4s- v7.4s}, [x8], #64 - ld1 { v8.4s-v11.4s}, [x8], #64 - ld1 {v12.4s-v15.4s}, [x8] + + load_round_constants x8 /* load state */ ld1 {dgav.4s, dgbv.4s}, [x0] /* load sha256_ce_state::finalize */
@@ -153,5 +157,270 @@ CPU_LE( rev32 v19.16b, v19.16b ) /* store new state */ 3: st1 {dgav.4s, dgbv.4s}, [x0] mov w0, w2 ret SYM_FUNC_END(__sha256_ce_transform) + + .unreq dga + .unreq dgav + .unreq dgb + .unreq dgbv + .unreq t0 + .unreq t1 + .unreq dg0q + .unreq dg0v + .unreq dg1q + .unreq dg1v + .unreq dg2q + .unreq dg2v + + // parameters for __sha256_ce_finup2x() + sctx .req x0 + data1 .req x1 + data2 .req x2 + len .req w3 + out1 .req x4 + out2 .req x5 + + // other scalar variables + count .req x6 + final_step .req w7 + + // x8-x9 are used as temporaries. + + // v0-v15 are used to cache the SHA-256 round constants. + // v16-v19 are used for the message schedule for the first message. + // v20-v23 are used for the message schedule for the second message. + // v24-v31 are used for the state and temporaries as given below. + // *_a are for the first message and *_b for the second. + state0_a_q .req q24 + state0_a .req v24 + state1_a_q .req q25 + state1_a .req v25 + state0_b_q .req q26 + state0_b .req v26 + state1_b_q .req q27 + state1_b .req v27 + t0_a .req v28 + t0_b .req v29 + t1_a_q .req q30 + t1_a .req v30 + t1_b_q .req q31 + t1_b .req v31 + +#define OFFSETOF_COUNT 32 // offsetof(struct sha256_state, count) +#define OFFSETOF_BUF 40 // offsetof(struct sha256_state, buf) +// offsetof(struct sha256_state, state) is assumed to be 0. + + // Do 4 rounds of SHA-256 for each of two messages (interleaved). m0_a + // and m0_b contain the current 4 message schedule words for the first + // and second message respectively. + // + // If not all the message schedule words have been computed yet, then + // this also computes 4 more message schedule words for each message. + // m1_a-m3_a contain the next 3 groups of 4 message schedule words for + // the first message, and likewise m1_b-m3_b for the second. After + // consuming the current value of m0_a, this macro computes the group + // after m3_a and writes it to m0_a, and likewise for *_b. This means + // that the next (m0_a, m1_a, m2_a, m3_a) is the current (m1_a, m2_a, + // m3_a, m0_a), and likewise for *_b, so the caller must cycle through + // the registers accordingly. + .macro do_4rounds_2x i, k, m0_a, m1_a, m2_a, m3_a, \ + m0_b, m1_b, m2_b, m3_b + add t0_a\().4s, \m0_a\().4s, \k\().4s + add t0_b\().4s, \m0_b\().4s, \k\().4s + .if \i < 48 + sha256su0 \m0_a\().4s, \m1_a\().4s + sha256su0 \m0_b\().4s, \m1_b\().4s + sha256su1 \m0_a\().4s, \m2_a\().4s, \m3_a\().4s + sha256su1 \m0_b\().4s, \m2_b\().4s, \m3_b\().4s + .endif + mov t1_a.16b, state0_a.16b + mov t1_b.16b, state0_b.16b + sha256h state0_a_q, state1_a_q, t0_a\().4s + sha256h state0_b_q, state1_b_q, t0_b\().4s + sha256h2 state1_a_q, t1_a_q, t0_a\().4s + sha256h2 state1_b_q, t1_b_q, t0_b\().4s + .endm + + .macro do_16rounds_2x i, k0, k1, k2, k3 + do_4rounds_2x \i + 0, \k0, v16, v17, v18, v19, v20, v21, v22, v23 + do_4rounds_2x \i + 4, \k1, v17, v18, v19, v16, v21, v22, v23, v20 + do_4rounds_2x \i + 8, \k2, v18, v19, v16, v17, v22, v23, v20, v21 + do_4rounds_2x \i + 12, \k3, v19, v16, v17, v18, v23, v20, v21, v22 + .endm + +// +// void __sha256_ce_finup2x(const struct sha256_state *sctx, +// const u8 *data1, const u8 *data2, int len, +// u8 out1[SHA256_DIGEST_SIZE], +// u8 out2[SHA256_DIGEST_SIZE]); +// +// This function computes the SHA-256 digests of two messages |data1| and +// |data2| that are both |len| bytes long, starting from the initial state +// |sctx|. |len| must be at least SHA256_BLOCK_SIZE. +// +// The instructions for the two SHA-256 operations are interleaved. On many +// CPUs, this is almost twice as fast as hashing each message individually due +// to taking better advantage of the CPU's SHA-256 and SIMD throughput. +// +SYM_FUNC_START(__sha256_ce_finup2x) + sub sp, sp, #128 + mov final_step, #0 + load_round_constants x8 + + // Load the initial state from sctx->state. + ld1 {state0_a.4s-state1_a.4s}, [sctx] + + // Load sctx->count. Take the mod 64 of it to get the number of bytes + // that are buffered in sctx->buf. Also save it in a register with len + // added to it. + ldr x8, [sctx, #OFFSETOF_COUNT] + add count, x8, len, sxtw + and x8, x8, #63 + cbz x8, .Lfinup2x_enter_loop // No bytes buffered? + + // x8 bytes (1 to 63) are currently buffered in sctx->buf. Load them + // followed by the first 64 - x8 bytes of data. Since len >= 64, we + // just load 64 bytes from each of sctx->buf, data1, and data2 + // unconditionally and rearrange the data as needed. + add x9, sctx, #OFFSETOF_BUF + ld1 {v16.16b-v19.16b}, [x9] + st1 {v16.16b-v19.16b}, [sp] + + ld1 {v16.16b-v19.16b}, [data1], #64 + add x9, sp, x8 + st1 {v16.16b-v19.16b}, [x9] + ld1 {v16.4s-v19.4s}, [sp] + + ld1 {v20.16b-v23.16b}, [data2], #64 + st1 {v20.16b-v23.16b}, [x9] + ld1 {v20.4s-v23.4s}, [sp] + + sub len, len, #64 + sub data1, data1, x8 + sub data2, data2, x8 + add len, len, w8 + mov state0_b.16b, state0_a.16b + mov state1_b.16b, state1_a.16b + b .Lfinup2x_loop_have_data + +.Lfinup2x_enter_loop: + sub len, len, #64 + mov state0_b.16b, state0_a.16b + mov state1_b.16b, state1_a.16b +.Lfinup2x_loop: + // Load the next two data blocks. + ld1 {v16.4s-v19.4s}, [data1], #64 + ld1 {v20.4s-v23.4s}, [data2], #64 +.Lfinup2x_loop_have_data: + // Convert the words of the data blocks from big endian. +CPU_LE( rev32 v16.16b, v16.16b ) +CPU_LE( rev32 v17.16b, v17.16b ) +CPU_LE( rev32 v18.16b, v18.16b ) +CPU_LE( rev32 v19.16b, v19.16b ) +CPU_LE( rev32 v20.16b, v20.16b ) +CPU_LE( rev32 v21.16b, v21.16b ) +CPU_LE( rev32 v22.16b, v22.16b ) +CPU_LE( rev32 v23.16b, v23.16b ) +.Lfinup2x_loop_have_bswapped_data: + + // Save the original state for each block. + st1 {state0_a.4s-state1_b.4s}, [sp] + + // Do the SHA-256 rounds on each block. + do_16rounds_2x 0, v0, v1, v2, v3 + do_16rounds_2x 16, v4, v5, v6, v7 + do_16rounds_2x 32, v8, v9, v10, v11 + do_16rounds_2x 48, v12, v13, v14, v15 + + // Add the original state for each block. + ld1 {v16.4s-v19.4s}, [sp] + add state0_a.4s, state0_a.4s, v16.4s + add state1_a.4s, state1_a.4s, v17.4s + add state0_b.4s, state0_b.4s, v18.4s + add state1_b.4s, state1_b.4s, v19.4s + + // Update len and loop back if more blocks remain. + sub len, len, #64 + tbz len, #31, .Lfinup2x_loop // len >= 0? + + // Check if any final blocks need to be handled. + // final_step = 2: all done + // final_step = 1: need to do count-only padding block + // final_step = 0: need to do the block with 0x80 padding byte + tbnz final_step, #1, .Lfinup2x_done + tbnz final_step, #0, .Lfinup2x_finalize_countonly + add len, len, #64 + cbz len, .Lfinup2x_finalize_blockaligned + + // Not block-aligned; 1 <= len <= 63 data bytes remain. Pad the block. + // To do this, write the padding starting with the 0x80 byte to + // &sp[64]. Then for each message, copy the last 64 data bytes to sp + // and load from &sp[64 - len] to get the needed padding block. This + // code relies on the data buffers being >= 64 bytes in length. + sub w8, len, #64 // w8 = len - 64 + add data1, data1, w8, sxtw // data1 += len - 64 + add data2, data2, w8, sxtw // data2 += len - 64 + mov x9, 0x80 + fmov d16, x9 + movi v17.16b, #0 + stp q16, q17, [sp, #64] + stp q17, q17, [sp, #96] + sub x9, sp, w8, sxtw // x9 = &sp[64 - len] + cmp len, #56 + b.ge 1f // will count spill into its own block? + lsl count, count, #3 + rev count, count + str count, [x9, #56] + mov final_step, #2 // won't need count-only block + b 2f +1: + mov final_step, #1 // will need count-only block +2: + ld1 {v16.16b-v19.16b}, [data1] + st1 {v16.16b-v19.16b}, [sp] + ld1 {v16.4s-v19.4s}, [x9] + ld1 {v20.16b-v23.16b}, [data2] + st1 {v20.16b-v23.16b}, [sp] + ld1 {v20.4s-v23.4s}, [x9] + b .Lfinup2x_loop_have_data + + // Prepare a padding block, either: + // + // {0x80, 0, 0, 0, ..., count (as __be64)} + // This is for a block aligned message. + // + // { 0, 0, 0, 0, ..., count (as __be64)} + // This is for a message whose length mod 64 is >= 56. + // + // Pre-swap the endianness of the words. +.Lfinup2x_finalize_countonly: + movi v16.2d, #0 + b 1f +.Lfinup2x_finalize_blockaligned: + mov x8, #0x80000000 + fmov d16, x8 +1: + movi v17.2d, #0 + movi v18.2d, #0 + ror count, count, #29 // ror(lsl(count, 3), 32) + mov v19.d[0], xzr + mov v19.d[1], count + mov v20.16b, v16.16b + movi v21.2d, #0 + movi v22.2d, #0 + mov v23.16b, v19.16b + mov final_step, #2 + b .Lfinup2x_loop_have_bswapped_data + +.Lfinup2x_done: + // Write the two digests with all bytes in the correct order. +CPU_LE( rev32 state0_a.16b, state0_a.16b ) +CPU_LE( rev32 state1_a.16b, state1_a.16b ) +CPU_LE( rev32 state0_b.16b, state0_b.16b ) +CPU_LE( rev32 state1_b.16b, state1_b.16b ) + st1 {state0_a.4s-state1_a.4s}, [out1] + st1 {state0_b.4s-state1_b.4s}, [out2] + add sp, sp, #128 + ret +SYM_FUNC_END(__sha256_ce_finup2x)
arch/arm64/crypto/sha2-ce-glue.c +40 -0
@@ -31,10 +31,15 @@ extern const u32 sha256_ce_offsetof_count;extern const u32 sha256_ce_offsetof_finalize; asmlinkage int __sha256_ce_transform(struct sha256_ce_state *sst, u8 const *src,int blocks); +asmlinkage void __sha256_ce_finup2x(const struct sha256_state *sctx, +const u8 *data1,const u8 *data2,int len, +u8 out1[SHA256_DIGEST_SIZE ], +u8 out2[SHA256_DIGEST_SIZE ]); +static void sha256_ce_transform (struct sha256_state *sst,u8 const *src,int blocks) {while (blocks) {int rem;
@@ -122,10 +127,43 @@ static int sha256_ce_digest(struct shash_desc *desc, const u8 *data, {sha256_base_init (desc);return sha256_ce_finup (desc, data, len, out); } +static int sha256_ce_finup_mb (struct shash_desc *desc, +const u8 *const data[],unsigned int len, +u8 *const outs[],unsigned int num_msgs) +{ +struct sha256_ce_state *sctx =shash_desc_ctx (desc); + +/* +* num_msgs != 2 should not happen here, since this algorithm sets +* mb_max_msgs=2, and the crypto API handles num_msgs <= 1 before +* calling into the algorithm's finup_mb method. +*/ +if (WARN_ON_ONCE (num_msgs !=2 )) +return -EOPNOTSUPP ; + +if (unlikely (!crypto_simd_usable ())) +return -EOPNOTSUPP ; + +/* __sha256_ce_finup2x() assumes SHA256_BLOCK_SIZE <= len <= INT_MAX. */ +if (unlikely (len <SHA256_BLOCK_SIZE || len >INT_MAX )) +return -EOPNOTSUPP ; + +/* __sha256_ce_finup2x() assumes the following offsets. */ +BUILD_BUG_ON (offsetof(struct sha256_state , state) !=0 ); +BUILD_BUG_ON (offsetof(struct sha256_state , count) !=32 ); +BUILD_BUG_ON (offsetof(struct sha256_state , buf) !=40 ); + +kernel_neon_begin (); +__sha256_ce_finup2x (&sctx->sst, data[0 ], data[1 ], len, outs[0 ], + outs[1 ]); +kernel_neon_end (); +return 0 ; +} +static int sha256_ce_export (struct shash_desc *desc,void *out) {struct sha256_ce_state *sctx =shash_desc_ctx (desc);memcpy (out, &sctx->sst,sizeof (struct sha256_state ));
@@ -162,13 +200,15 @@ static struct shash_alg algs[] = { { .init = sha256_base_init, .update = sha256_ce_update, .final = sha256_ce_final, .finup = sha256_ce_finup, .digest = sha256_ce_digest, + .finup_mb = sha256_ce_finup_mb, .export = sha256_ce_export, .import = sha256_ce_import, .descsize = sizeof(struct sha256_ce_state), + .mb_max_msgs = 2, .statesize = sizeof(struct sha256_state), .digestsize = SHA256_DIGEST_SIZE, .base = { .cra_name = "sha256", .cra_driver_name = "sha256-ce",
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 6/15 fsverity: improve performance by using multibuffer hashing
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
When supported by the hash algorithm, use crypto_shash_finup_mb() to interleave the hashing of pairs of data blocks. On some CPUs this nearly doubles hashing performance. The increase in overall throughput of cold-cache fsverity reads that I'm seeing on arm64 and x86_64 is roughly 35% (though this metric is hard to measure as it jumps around a lot).
For now this is only done on the verification path, and only for data blocks, not Merkle tree blocks. We could use finup_mb on Merkle tree blocks too, but that is less important as there aren't as many Merkle tree blocks as data blocks, and that would require some additional code restructuring. We could also use finup_mb to accelerate building the Merkle tree, but verification performance is more important.
--- fs/verity/fsverity_private.h | 7 ++ fs/verity/hash_algs.c | 8 +- fs/verity/verify.c | 170 +++++++++++++++++++++++++++++------ 3 files changed, 154 insertions(+), 31 deletions(-)
fs/verity/fsverity_private.h +7 -0
@@ -27,10 +27,15 @@ struct fsverity_hash_alg {/* * The HASH_ALGO_* constant for this algorithm. This is different from * FS_VERITY_HASH_ALG_*, which uses a different numbering scheme. */ enum hash_algo algo_id; +/* +* The maximum supported interleaving factor for multibuffer hashing, or +* 1 if the algorithm doesn't support multibuffer hashing +*/ +int mb_max_msgs; };/* Merkle tree parameters: hash algorithm, initial hash state, and topology */ struct merkle_tree_params {const struct fsverity_hash_alg *hash_alg;/* the hash algorithm */
@@ -150,8 +155,10 @@ static inline void fsverity_init_signature(void) }#endif /* !CONFIG_FS_VERITY_BUILTIN_SIGNATURES */ /* verify.c */ +#define FS_VERITY_MAX_PENDING_DATA_BLOCKS 2 +void __init fsverity_init_workqueue (void );#endif /* _FSVERITY_PRIVATE_H */
fs/verity/hash_algs.c +6 -2
@@ -82,12 +82,16 @@ const struct fsverity_hash_alg *fsverity_get_hash_alg(const struct inode *inode,if (WARN_ON_ONCE (alg->digest_size !=crypto_shash_digestsize (tfm))) goto err_free_tfm;if (WARN_ON_ONCE (alg->block_size !=crypto_shash_blocksize (tfm))) goto err_free_tfm; -pr_info ("%s using implementation \"%s\"\n" , - alg->name,crypto_shash_driver_name (tfm)); + alg->mb_max_msgs =min (crypto_shash_mb_max_msgs (tfm), +FS_VERITY_MAX_PENDING_DATA_BLOCKS ); + +pr_info ("%s using implementation \"%s\"%s\n" , + alg->name,crypto_shash_driver_name (tfm), + alg->mb_max_msgs >1 ?" (multibuffer)" :"" );/* pairs with smp_load_acquire() above */ smp_store_release (&alg->tfm, tfm); goto out_unlock;
fs/verity/verify.c +141 -29
289 lines not on this page · raw message
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 7/15 dm-verity: move hash algorithm setup into its own function
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
Move the code that sets up the hash transformation into its own function. No change in behavior.
--- drivers/md/dm-verity-target.c | 70 +++++++++++++++++++---------------- 1 file changed, 39 insertions(+), 31 deletions(-)
drivers/md/dm-verity-target.c +39 -31
@@ -1224,10 +1224,47 @@ static int verity_parse_opt_args(struct dm_arg_set *as, struct dm_verity *v, }while (argc && !r);return r; } +static int verity_setup_hash_alg (struct dm_verity *v,const char *alg_name) +{ +struct dm_target *ti = v->ti; +struct crypto_ahash *ahash; + + v->alg_name =kstrdup (alg_name,GFP_KERNEL ); +if (!v->alg_name) { + ti->error ="Cannot allocate algorithm name" ; +return -ENOMEM ; + } + + ahash =crypto_alloc_ahash (alg_name,0 , + v->use_bh_wq ?CRYPTO_ALG_ASYNC :0 ); +if (IS_ERR (ahash)) { + ti->error ="Cannot initialize hash function" ; +return PTR_ERR (ahash); + } + v->tfm = ahash; + +/* +* dm-verity performance can vary greatly depending on which hash +* algorithm implementation is used. Help people debug performance +* problems by logging the ->cra_driver_name. +*/ +DMINFO ("%s using implementation \"%s\"" , alg_name, +crypto_hash_alg_common (ahash)->base.cra_driver_name); + + v->digest_size =crypto_ahash_digestsize (ahash); +if ((1 << v->hash_dev_block_bits) < v->digest_size *2 ) { + ti->error ="Digest size too big" ; +return -EINVAL ; + } + v->ahash_reqsize =sizeof (struct ahash_request ) + +crypto_ahash_reqsize (ahash); +return 0 ; +} +/* * Target parameters: * <version> The current format is version 1. * Vsn 0 is compatible with original Chromium OS releases. * <data device>
@@ -1348,42 +1385,13 @@ static int verity_ctr(struct dm_target *ti, unsigned int argc, char **argv) r = -EINVAL ; goto bad; } v->hash_start = num_ll; - v->alg_name =kstrdup (argv[7 ],GFP_KERNEL ); -if (!v->alg_name) { - ti->error ="Cannot allocate algorithm name" ; - r = -ENOMEM ; - goto bad; - } - - v->tfm =crypto_alloc_ahash (v->alg_name,0 , - v->use_bh_wq ?CRYPTO_ALG_ASYNC :0 ); -if (IS_ERR (v->tfm)) { - ti->error ="Cannot initialize hash function" ; - r =PTR_ERR (v->tfm); - v->tfm =NULL ; - goto bad; - } - -/* -* dm-verity performance can vary greatly depending on which hash -* algorithm implementation is used. Help people debug performance -* problems by logging the ->cra_driver_name. -*/ -DMINFO ("%s using implementation \"%s\"" , v->alg_name, -crypto_hash_alg_common (v->tfm)->base.cra_driver_name); - - v->digest_size =crypto_ahash_digestsize (v->tfm); -if ((1 << v->hash_dev_block_bits) < v->digest_size *2 ) { - ti->error ="Digest size too big" ; - r = -EINVAL ; + r =verity_setup_hash_alg (v, argv[7 ]); +if (r) goto bad; - } - v->ahash_reqsize =sizeof (struct ahash_request ) + -crypto_ahash_reqsize (v->tfm); v->root_digest =kmalloc (v->digest_size,GFP_KERNEL );if (!v->root_digest) { ti->error ="Cannot allocate root digest" ; r = -ENOMEM ;
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 8/15 dm-verity: move data hash mismatch handling into its own function
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
Move the code that handles mismatches of data block hashes into its own function so that it doesn't clutter up verity_verify_io().
--- drivers/md/dm-verity-target.c | 64 ++++++++++++++++++++--------------- 1 file changed, 36 insertions(+), 28 deletions(-)
drivers/md/dm-verity-target.c +36 -28
@@ -540,10 +540,42 @@ static noinline int verity_recheck(struct dm_verity *v, struct dm_verity_io *io,mempool_free (page, &v->recheck_pool);return r; } +static int verity_handle_data_hash_mismatch (struct dm_verity *v, +struct dm_verity_io *io, +struct bio *bio,sector_t blkno, +struct bvec_iter *start) +{ +if (static_branch_unlikely (&use_bh_wq_enabled) && io->in_bh) { +/* +* Error handling code (FEC included) cannot be run in the +* BH workqueue, so fallback to a standard workqueue. +*/ +return -EAGAIN ; + } +if (verity_recheck (v, io, *start, blkno) ==0 ) { +if (v->validated_blocks) +set_bit (blkno, v->validated_blocks); +return 0 ; + } +#if defined(CONFIG_DM_VERITY_FEC ) +if (verity_fec_decode (v, io,DM_VERITY_BLOCK_TYPE_DATA , blkno, +NULL , start) ==0 ) +return 0 ; +#endif +if (bio->bi_status) +return -EIO ;/* Error correction failed; Just return error */ + +if (verity_handle_err (v,DM_VERITY_BLOCK_TYPE_DATA , blkno)) { +dm_audit_log_bio (DM_MSG_PREFIX ,"verify-data" , bio, blkno,0 ); +return -EIO ; + } +return 0 ; +} +static int verity_bv_zero (struct dm_verity *v,struct dm_verity_io *io,u8 *data,size_t len) {memset (data,0 , len);return 0 ;
@@ -632,39 +664,15 @@ static int verity_verify_io(struct dm_verity_io *io)if (likely (memcmp (verity_io_real_digest (v, io),verity_io_want_digest (v, io), v->digest_size) ==0 )) {if (v->validated_blocks)set_bit (cur_block, v->validated_blocks);continue ; - }else if (static_branch_unlikely (&use_bh_wq_enabled) && io->in_bh) { -/* -* Error handling code (FEC included) cannot be run in a -* tasklet since it may sleep, so fallback to work-queue. -*/ -return -EAGAIN ; - }else if (verity_recheck (v, io, start, cur_block) ==0 ) { -if (v->validated_blocks) -set_bit (cur_block, v->validated_blocks); -continue ; -#if defined(CONFIG_DM_VERITY_FEC ) - } else if (verity_fec_decode(v, io, DM_VERITY_BLOCK_TYPE_DATA, - cur_block,NULL , &start) ==0 ) { -continue ; -#endif - }else { -if (bio->bi_status) { -/* -* Error correction failed; Just return error -*/ -return -EIO ; - } -if (verity_handle_err (v,DM_VERITY_BLOCK_TYPE_DATA , - cur_block)) { -dm_audit_log_bio (DM_MSG_PREFIX ,"verify-data" , - bio, cur_block,0 ); -return -EIO ; - } } + r =verity_handle_data_hash_mismatch (v, io, bio, cur_block, + &start); +if (unlikely (r)) +return r; }return 0 ; }
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 9/15 dm-verity: make real_digest and want_digest fixed-length
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
Change the digest fields in struct dm_verity_io from variable-length to fixed-length, since their maximum length is fixed at HASH_MAX_DIGESTSIZE, i.e. 64 bytes, which is not too big. This is simpler and makes the fields a bit faster to access.
(HASH_MAX_DIGESTSIZE did not exist when this code was written, which may explain why it wasn't used.)
This makes the verity_io_real_digest() and verity_io_want_digest() functions trivial, but this patch leaves them in place temporarily since most of their callers will go away in a later patch anyway.
--- drivers/md/dm-verity-target.c | 3 +-- drivers/md/dm-verity.h | 17 +++++++---------- 2 files changed, 8 insertions(+), 12 deletions(-)
drivers/md/dm-verity-target.c +1 -2
@@ -1527,12 +1527,11 @@ static int verity_ctr(struct dm_target *ti, unsigned int argc, char **argv) ti->error ="Cannot allocate workqueue" ; r = -ENOMEM ; goto bad; } - ti->per_io_data_size =sizeof (struct dm_verity_io ) + - v->ahash_reqsize + v->digest_size *2 ; + ti->per_io_data_size =sizeof (struct dm_verity_io ) + v->ahash_reqsize; r =verity_fec_ctr (v);if (r) goto bad;
drivers/md/dm-verity.h +7 -10
@@ -89,19 +89,16 @@ struct dm_verity_io {struct work_struct work;struct work_struct bh_work;char *recheck_buffer; +u8 real_digest[HASH_MAX_DIGESTSIZE ]; +u8 want_digest[HASH_MAX_DIGESTSIZE ]; +/* -* Three variably-size fields follow this struct: -* -* u8 hash_req[v->ahash_reqsize]; -* u8 real_digest[v->digest_size]; -* u8 want_digest[v->digest_size]; -* -* To access them use: verity_io_hash_req(), verity_io_real_digest() -* and verity_io_want_digest(). +* This struct is followed by a variable-sized struct ahash_request of +* size v->ahash_reqsize. To access it, use verity_io_hash_req(). */ };static inline struct ahash_request *verity_io_hash_req (struct dm_verity *v,struct dm_verity_io *io)
@@ -110,17 +107,17 @@ static inline struct ahash_request *verity_io_hash_req(struct dm_verity *v, }static inline u8 *verity_io_real_digest (struct dm_verity *v,struct dm_verity_io *io) { -return (u8 *)(io +1 ) + v->ahash_reqsize; +return io->real_digest; }static inline u8 *verity_io_want_digest (struct dm_verity *v,struct dm_verity_io *io) { -return (u8 *)(io +1 ) + v->ahash_reqsize + v->digest_size; +return io->want_digest; } extern int verity_for_bv_block(struct dm_verity *v, struct dm_verity_io *io, struct bvec_iter *iter, int (*process)(struct dm_verity *v,
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 10/15 dm-verity: provide dma_alignment limit in io_hints
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
Since Linux v6.1, some filesystems support submitting direct I/O that is aligned to only dma_alignment instead of the logical_block_size alignment that was required before. I/O that is not aligned to the logical_block_size is difficult to handle in device-mapper targets that do cryptographic processing of data, as it makes the units of data that are hashed or encrypted possibly be split across pages, creating rarely used and rarely tested edge cases.
As such, dm-crypt and dm-integrity have already opted out of this by setting dma_alignment to 'logical_block_size - 1'.
Although dm-verity does have code that handles these cases (or at least is intended to do so), supporting direct I/O with such a low amount of alignment is not really useful on dm-verity devices. So, opt dm-verity out of it too so that it's not necessary to handle these edge cases.
--- drivers/md/dm-verity-target.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/md/dm-verity-target.c +2 -0
@@ -1021,10 +1021,12 @@ static void verity_io_hints(struct dm_target *ti, struct queue_limits *limits)if (limits->physical_block_size <1 << v->data_dev_block_bits) limits->physical_block_size =1 << v->data_dev_block_bits;blk_limits_io_min (limits, limits->logical_block_size); + + limits->dma_alignment = limits->logical_block_size -1 ; }static void verity_dtr (struct dm_target *ti) {struct dm_verity *v = ti->private;
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 11/15 dm-verity: always "map" the data blocks
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
dm-verity needs to access data blocks by virtual address in three different cases (zeroization, recheck, and forward error correction), and one more case (shash support) is coming. Since it's guaranteed that dm-verity data blocks never cross pages, and kmap_local_page and kunmap_local are no-ops on modern platforms anyway, just unconditionally "map" every data block's page and work with the virtual buffer directly. This simplifies the code and eliminates unnecessary overhead.
--- drivers/md/dm-verity-fec.c | 26 +---- drivers/md/dm-verity-fec.h | 6 +- drivers/md/dm-verity-target.c | 182 +++++++--------------------------- drivers/md/dm-verity.h | 8 -- 4 files changed, 42 insertions(+), 180 deletions(-)
drivers/md/dm-verity-fec.c +3 -23
@@ -402,28 +402,13 @@ static int fec_decode_rsb(struct dm_verity *v, struct dm_verity_io *io, }return 0 ; } -static int fec_bv_copy (struct dm_verity *v,struct dm_verity_io *io,u8 *data, -size_t len) -{ -struct dm_verity_fec_io *fio =fec_io (io); - -memcpy (data, &fio->output[fio->output_pos], len); - fio->output_pos += len; - -return 0 ; -} - -/* -* Correct errors in a block. Copies corrected block to dest if non-NULL, -* otherwise to a bio_vec starting from iter. -*/ +/* Correct errors in a block. Copies corrected block to dest. */ int verity_fec_decode (struct dm_verity *v,struct dm_verity_io *io, -enum verity_block_type type,sector_t block,u8 *dest, -struct bvec_iter *iter) +enum verity_block_type type,sector_t block,u8 *dest) {int r;struct dm_verity_fec_io *fio =fec_io (io);u64 offset, res, rsb;
@@ -469,16 +454,11 @@ int verity_fec_decode(struct dm_verity *v, struct dm_verity_io *io, r =fec_decode_rsb (v, io, fio, rsb, offset, true);if (r <0 ) goto done; } -if (dest) -memcpy (dest, fio->output,1 << v->data_dev_block_bits); -else if (iter) { - fio->output_pos =0 ; - r =verity_for_bv_block (v, io, iter, fec_bv_copy); - } +memcpy (dest, fio->output,1 << v->data_dev_block_bits); done: fio->level--;return r; }
drivers/md/dm-verity-fec.h +2 -4
@@ -55,11 +55,10 @@ struct dm_verity_fec_io {struct rs_control *rs;/* Reed-Solomon state */ int erasures[DM_VERITY_FEC_MAX_RSN ];/* erasures for decode_rs8 */ u8 *bufs[DM_VERITY_FEC_BUF_MAX ];/* bufs for deinterleaving */ unsigned int nbufs;/* number of buffers allocated */ u8 *output;/* buffer for corrected output */ -size_t output_pos;unsigned int level;/* recursion level */ };#ifdef CONFIG_DM_VERITY_FEC
@@ -68,11 +67,11 @@ struct dm_verity_fec_io {extern bool verity_fec_is_enabled (struct dm_verity *v);extern int verity_fec_decode (struct dm_verity *v,struct dm_verity_io *io,enum verity_block_type type,sector_t block, -u8 *dest,struct bvec_iter *iter); +u8 *dest);extern unsigned int verity_fec_status_table (struct dm_verity *v,unsigned int sz,char *result,unsigned int maxlen);extern void verity_fec_finish_io (struct dm_verity_io *io);
@@ -98,12 +97,11 @@ static inline bool verity_fec_is_enabled(struct dm_verity *v) }static inline int verity_fec_decode (struct dm_verity *v,struct dm_verity_io *io,enum verity_block_type type, -sector_t block,u8 *dest, -struct bvec_iter *iter) +sector_t block,u8 *dest) {return -EOPNOTSUPP ; } static inline unsigned int verity_fec_status_table(struct dm_verity *v,
drivers/md/dm-verity-target.c +37 -145
294 lines not on this page · raw message
drivers/md/dm-verity.h +0 -8
@@ -87,12 +87,10 @@ struct dm_verity_io {bool in_bh;struct work_struct work;struct work_struct bh_work; -char *recheck_buffer; -u8 real_digest[HASH_MAX_DIGESTSIZE ];u8 want_digest[HASH_MAX_DIGESTSIZE ];/* * This struct is followed by a variable-sized struct ahash_request of
@@ -116,16 +114,10 @@ static inline u8 *verity_io_want_digest(struct dm_verity *v, struct dm_verity_io *io) {return io->want_digest; } -extern int verity_for_bv_block (struct dm_verity *v,struct dm_verity_io *io, -struct bvec_iter *iter, -int (*process)(struct dm_verity *v, -struct dm_verity_io *io, -u8 *data,size_t len)); -extern int verity_hash (struct dm_verity *v,struct ahash_request *req,const u8 *data,size_t len,u8 *digest,bool may_sleep);extern int verity_hash_for_block (struct dm_verity *v,struct dm_verity_io *io,sector_t block,u8 *digest,bool *is_zero);
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 12/15 dm-verity: make verity_hash() take dm_verity_io instead of ahash_request
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
In preparation for adding shash support to dm-verity, change verity_hash() to take a pointer to a struct dm_verity_io instead of a pointer to the ahash_request embedded inside it.
--- drivers/md/dm-verity-fec.c | 6 ++---- drivers/md/dm-verity-target.c | 21 ++++++++++----------- drivers/md/dm-verity.h | 2 +- 3 files changed, 13 insertions(+), 16 deletions(-)
drivers/md/dm-verity-fec.c +2 -4
@@ -184,12 +184,11 @@ static int fec_decode_bufs(struct dm_verity *v, struct dm_verity_io *io,* Locate data block erasures using verity hashes. */ static int fec_is_erasure (struct dm_verity *v,struct dm_verity_io *io,u8 *want_digest,u8 *data) { -if (unlikely (verity_hash (v,verity_io_hash_req (v, io), - data,1 << v->data_dev_block_bits, +if (unlikely (verity_hash (v, io, data,1 << v->data_dev_block_bits,verity_io_real_digest (v, io), true)))return 0 ;return memcmp (verity_io_real_digest (v, io), want_digest, v->digest_size) !=0 ;
@@ -386,12 +385,11 @@ static int fec_decode_rsb(struct dm_verity *v, struct dm_verity_io *io, pos += fio->nbufs <<DM_VERITY_FEC_BUF_RS_BITS ; }/* Always re-validate the corrected block against the expected hash */ - r =verity_hash (v,verity_io_hash_req (v, io), fio->output, -1 << v->data_dev_block_bits, + r =verity_hash (v, io, fio->output,1 << v->data_dev_block_bits,verity_io_real_digest (v, io), true);if (unlikely (r <0 ))return r; if (memcmp(verity_io_real_digest(v, io), verity_io_want_digest(v, io),
drivers/md/dm-verity-target.c +10 -11
@@ -178,13 +178,14 @@ static int verity_hash_final(struct dm_verity *v, struct ahash_request *req, r =crypto_wait_req (crypto_ahash_final (req), wait); out:return r; } -int verity_hash (struct dm_verity *v,struct ahash_request *req, +int verity_hash (struct dm_verity *v,struct dm_verity_io *io,const u8 *data,size_t len,u8 *digest,bool may_sleep) { +struct ahash_request *req =verity_io_hash_req (v, io);int r;struct crypto_wait wait; r =verity_hash_init (v, req, &wait, may_sleep);if (unlikely (r <0 ))
@@ -323,12 +324,11 @@ static int verity_verify_level(struct dm_verity *v, struct dm_verity_io *io,if (skip_unverified) { r =1 ; goto release_ret_r; } - r =verity_hash (v,verity_io_hash_req (v, io), - data,1 << v->hash_dev_block_bits, + r =verity_hash (v, io, data,1 << v->hash_dev_block_bits,verity_io_real_digest (v, io), !io->in_bh);if (unlikely (r <0 )) goto release_ret_r; if (likely(memcmp(verity_io_real_digest(v, io), want_digest,
@@ -426,12 +426,11 @@ static noinline int verity_recheck(struct dm_verity *v, struct dm_verity_io *io, io_loc.count =1 << (v->data_dev_block_bits -SECTOR_SHIFT ); r =dm_io (&io_req,1 , &io_loc,NULL ,IOPRIO_DEFAULT );if (unlikely (r)) goto free_ret; - r =verity_hash (v,verity_io_hash_req (v, io), buffer, -1 << v->data_dev_block_bits, + r =verity_hash (v, io, buffer,1 << v->data_dev_block_bits,verity_io_real_digest (v, io), true);if (unlikely (r)) goto free_ret; if (memcmp(verity_io_real_digest(v, io),
@@ -542,11 +541,11 @@ static int verity_verify_io(struct dm_verity_io *io)memset (data,0 , block_size);kunmap_local (data);continue ; } - r =verity_hash (v,verity_io_hash_req (v, io), data, block_size, + r =verity_hash (v, io, data, block_size,verity_io_real_digest (v, io), !io->in_bh);if (unlikely (r <0 )) {kunmap_local (data);return r; }
@@ -983,33 +982,33 @@ static int verity_alloc_most_once(struct dm_verity *v) }static int verity_alloc_zero_digest (struct dm_verity *v) {int r = -ENOMEM ; -struct ahash_request *req; +struct dm_verity_io *io;u8 *zero_data; v->zero_digest =kmalloc (v->digest_size,GFP_KERNEL );if (!v->zero_digest)return r; - req =kmalloc (v->ahash_reqsize,GFP_KERNEL ); + io =kmalloc (sizeof (*io) + v->ahash_reqsize,GFP_KERNEL ); -if (!req) +if (!io)return r;/* verity_dtr will free zero_digest */ zero_data =kzalloc (1 << v->data_dev_block_bits,GFP_KERNEL );if (!zero_data) goto out; - r =verity_hash (v, req, zero_data,1 << v->data_dev_block_bits, + r =verity_hash (v, io, zero_data,1 << v->data_dev_block_bits, v->zero_digest, true); out: -kfree (req); +kfree (io);kfree (zero_data);return r; }
drivers/md/dm-verity.h +1 -1
@@ -114,11 +114,11 @@ static inline u8 *verity_io_want_digest(struct dm_verity *v, struct dm_verity_io *io) {return io->want_digest; } -extern int verity_hash (struct dm_verity *v,struct ahash_request *req, +extern int verity_hash (struct dm_verity *v,struct dm_verity_io *io,const u8 *data,size_t len,u8 *digest,bool may_sleep);extern int verity_hash_for_block (struct dm_verity *v,struct dm_verity_io *io,sector_t block,u8 *digest,bool *is_zero);
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 13/15 dm-verity: hash blocks with shash import+finup when possible
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
Currently dm-verity computes the hash of each block by using multiple calls to the "ahash" crypto API. While the exact sequence depends on the chosen dm-verity settings, in the vast majority of cases it is:
1. crypto_ahash_init()
2. crypto_ahash_update() [salt]
3. crypto_ahash_update() [data]
4. crypto_ahash_final()
This is inefficient for two main reasons:
- It makes multiple indirect calls, which is expensive on modern CPUs especially when mitigations for CPU vulnerabilities are enabled.
Since the salt is the same across all blocks on a given dm-verity device, a much more efficient sequence would be to do an import of the pre-salted state, then a finup.
- It uses the ahash (asynchronous hash) API, despite the fact that CPU-based hashing is almost always used in practice, and therefore it experiences the overhead of the ahash-based wrapper for shash.
Because dm-verity was intentionally converted to ahash to support off-CPU crypto accelerators, a full reversion to shash might not be acceptable. Yet, we should still provide a fast path for shash with the most common dm-verity settings.
Another reason for shash over ahash is that the upcoming multibuffer hashing support, which is specific to CPU-based hashing, is much better suited for shash than for ahash. Supporting it via ahash would add significant complexity and overhead. And it's not possible for the "same" code to properly support both multibuffer hashing and HW accelerators at the same time anyway, given the different computation models. Unfortunately there will always be code specific to each model needed (for users who want to support both).
Therefore, this patch adds a new shash import+finup based fast path to dm-verity. It is used automatically when appropriate. This makes dm-verity optimized for what the vast majority of users want: CPU-based hashing with the most common settings, while still retaining support for rarer settings and off-CPU crypto accelerators.
In benchmarks with veritysetup's default parameters (SHA-256, 4K data and hash block sizes, 32-byte salt), which also match the parameters that Android currently uses, this patch improves block hashing performance by about 15% on x86_64 using the SHA-NI instructions, or by about 5% on arm64 using the ARMv8 SHA2 instructions. On x86_64 roughly two-thirds of the improvement comes from the use of import and finup, while the remaining third comes from the switch from ahash to shash.
Note that another benefit of using "import" to handle the salt is that if the salt size is equal to the input size of the hash algorithm's compression function, e.g. 64 bytes for SHA-256, then the performance is exactly the same as no salt. This doesn't seem to be much better than veritysetup's current default of 32-byte salts, due to the way SHA-256's finalization padding works, but it should be marginally better.
--- drivers/md/dm-verity-target.c | 169 ++++++++++++++++++++++++---------- drivers/md/dm-verity.h | 18 ++-- 2 files changed, 130 insertions(+), 57 deletions(-)
drivers/md/dm-verity-target.c +119 -50
@@ -46,10 +46,13 @@ static unsigned int dm_verity_prefetch_cluster = DM_VERITY_DEFAULT_PREFETCH_SIZEmodule_param_named (prefetch_cluster, dm_verity_prefetch_cluster, uint,0644 );static DEFINE_STATIC_KEY_FALSE (use_bh_wq_enabled); +/* Is at least one dm-verity instance using ahash_tfm instead of shash_tfm? */ +static DEFINE_STATIC_KEY_FALSE (ahash_enabled); +struct dm_verity_prefetch_work {struct work_struct work;struct dm_verity *v;unsigned short ioprio;sector_t block;
@@ -100,11 +103,11 @@ static sector_t verity_position_at_level(struct dm_verity *v, sector_t block, int level) {return block >> (level * v->hash_per_block_bits); } -static int verity_hash_update (struct dm_verity *v,struct ahash_request *req, +static int verity_ahash_update (struct dm_verity *v,struct ahash_request *req,const u8 *data,size_t len,struct crypto_wait *wait) {struct scatterlist sg;
@@ -133,16 +136,16 @@ static int verity_hash_update(struct dm_verity *v, struct ahash_request *req, }/* * Wrapper for crypto_ahash_init, which handles verity salting. */ -static int verity_hash_init (struct dm_verity *v,struct ahash_request *req, +static int verity_ahash_init (struct dm_verity *v,struct ahash_request *req,struct crypto_wait *wait,bool may_sleep) {int r; -ahash_request_set_tfm (req, v->tfm); +ahash_request_set_tfm (req, v->ahash_tfm);ahash_request_set_callback (req, may_sleep ?CRYPTO_TFM_REQ_MAY_SLEEP |CRYPTO_TFM_REQ_MAY_BACKLOG :0 , crypto_req_done, (void *)wait);crypto_init_wait (wait);
@@ -153,22 +156,22 @@ static int verity_hash_init(struct dm_verity *v, struct ahash_request *req,DMERR ("crypto_ahash_init failed: %d" , r);return r; }if (likely (v->salt_size && (v->version >=1 ))) - r =verity_hash_update (v, req, v->salt, v->salt_size, wait); + r =verity_ahash_update (v, req, v->salt, v->salt_size, wait);return r; } -static int verity_hash_final (struct dm_verity *v,struct ahash_request *req, -u8 *digest,struct crypto_wait *wait) +static int verity_ahash_final (struct dm_verity *v,struct ahash_request *req, +u8 *digest,struct crypto_wait *wait) {int r;if (unlikely (v->salt_size && (!v->version))) { - r =verity_hash_update (v, req, v->salt, v->salt_size, wait); + r =verity_ahash_update (v, req, v->salt, v->salt_size, wait);if (r <0 ) {DMERR ("%s failed updating salt: %d" , __func__, r); goto out; }
@@ -181,25 +184,28 @@ static int verity_hash_final(struct dm_verity *v, struct ahash_request *req, }int verity_hash (struct dm_verity *v,struct dm_verity_io *io,const u8 *data,size_t len,u8 *digest,bool may_sleep) { -struct ahash_request *req =verity_io_hash_req (v, io);int r; -struct crypto_wait wait; - - r =verity_hash_init (v, req, &wait, may_sleep); -if (unlikely (r <0 )) - goto out; - r =verity_hash_update (v, req, data, len, &wait); -if (unlikely (r <0 )) - goto out; +if (static_branch_unlikely (&ahash_enabled) && !v->shash_tfm) { +struct ahash_request *req =verity_io_hash_req (v, io); +struct crypto_wait wait; - r =verity_hash_final (v, req, digest, &wait); + r =verity_ahash_init (v, req, &wait, may_sleep) ?: +verity_ahash_update (v, req, data, len, &wait) ?: +verity_ahash_final (v, req, digest, &wait); + }else { +struct shash_desc *desc =verity_io_hash_req (v, io); -out: + desc->tfm = v->shash_tfm; + r =crypto_shash_import (desc, v->initial_hashstate) ?: +crypto_shash_finup (desc, data, len, digest); + } +if (unlikely (r)) +DMERR ("Error hashing block: %d" , r);return r; }static void verity_hash_at_level (struct dm_verity *v,sector_t block,int level,sector_t *hash_block,unsigned int *offset)
@@ -932,15 +938,20 @@ static void verity_dtr(struct dm_target *ti)if (v->bufio)dm_bufio_client_destroy (v->bufio);kvfree (v->validated_blocks);kfree (v->salt); +kfree (v->initial_hashstate);kfree (v->root_digest);kfree (v->zero_digest); -if (v->tfm) -crypto_free_ahash (v->tfm); +if (v->ahash_tfm) { +static_branch_dec (&ahash_enabled); +crypto_free_ahash (v->ahash_tfm); + }else { +crypto_free_shash (v->shash_tfm); + }kfree (v->alg_name);if (v->hash_dev)dm_put_device (ti, v->hash_dev);
@@ -990,11 +1001,11 @@ static int verity_alloc_zero_digest(struct dm_verity *v) v->zero_digest =kmalloc (v->digest_size,GFP_KERNEL );if (!v->zero_digest)return r; - io =kmalloc (sizeof (*io) + v->ahash_reqsize,GFP_KERNEL ); + io =kmalloc (sizeof (*io) + v->hash_reqsize,GFP_KERNEL );if (!io)return r;/* verity_dtr will free zero_digest */ zero_data =kzalloc (1 << v->data_dev_block_bits,GFP_KERNEL );
@@ -1129,40 +1140,110 @@ static int verity_parse_opt_args(struct dm_arg_set *as, struct dm_verity *v,static int verity_setup_hash_alg (struct dm_verity *v,const char *alg_name) {struct dm_target *ti = v->ti;struct crypto_ahash *ahash; +struct crypto_shash *shash =NULL ; +const char *driver_name; v->alg_name =kstrdup (alg_name,GFP_KERNEL );if (!v->alg_name) { ti->error ="Cannot allocate algorithm name" ;return -ENOMEM ; } +/* +* Allocate the hash transformation object that this dm-verity instance +* will use. The vast majority of dm-verity users use CPU-based +* hashing, so when possible use the shash API to minimize the crypto +* API overhead. If the ahash API resolves to a different driver +* (likely an off-CPU hardware offload), use ahash instead. Also use +* ahash if the obsolete dm-verity format with the appended salt is +* being used, so that quirk only needs to be handled in one place. +*/ ahash =crypto_alloc_ahash (alg_name,0 , v->use_bh_wq ?CRYPTO_ALG_ASYNC :0 );if (IS_ERR (ahash)) { ti->error ="Cannot initialize hash function" ;return PTR_ERR (ahash); } - v->tfm = ahash; - -/* -* dm-verity performance can vary greatly depending on which hash -* algorithm implementation is used. Help people debug performance -* problems by logging the ->cra_driver_name. -*/ -DMINFO ("%s using implementation \"%s\"" , alg_name, -crypto_hash_alg_common (ahash)->base.cra_driver_name); - - v->digest_size =crypto_ahash_digestsize (ahash); + driver_name =crypto_ahash_driver_name (ahash); +if (v->version >=1 /* salt prepended, not appended? */ ) { + shash =crypto_alloc_shash (alg_name,0 ,0 ); +if (!IS_ERR (shash) && +strcmp (crypto_shash_driver_name (shash), driver_name) !=0 ) { +/* +* ahash gave a different driver than shash, so probably +* this is a case of real hardware offload. Use ahash. +*/ +crypto_free_shash (shash); + shash =NULL ; + } + } +if (!IS_ERR_OR_NULL (shash)) { +crypto_free_ahash (ahash); + ahash =NULL ; + v->shash_tfm = shash; + v->digest_size =crypto_shash_digestsize (shash); + v->hash_reqsize =sizeof (struct shash_desc ) + +crypto_shash_descsize (shash); +DMINFO ("%s using shash \"%s\"" , alg_name, driver_name); + }else { + v->ahash_tfm = ahash; +static_branch_inc (&ahash_enabled); + v->digest_size =crypto_ahash_digestsize (ahash); + v->hash_reqsize =sizeof (struct ahash_request ) + +crypto_ahash_reqsize (ahash); +DMINFO ("%s using ahash \"%s\"" , alg_name, driver_name); + }if ((1 << v->hash_dev_block_bits) < v->digest_size *2 ) { ti->error ="Digest size too big" ;return -EINVAL ; } - v->ahash_reqsize =sizeof (struct ahash_request ) + -crypto_ahash_reqsize (ahash); +return 0 ; +} + +static int verity_setup_salt_and_hashstate (struct dm_verity *v,const char *arg) +{ +struct dm_target *ti = v->ti; + +if (strcmp (arg,"-" ) !=0 ) { + v->salt_size =strlen (arg) /2 ; + v->salt =kmalloc (v->salt_size,GFP_KERNEL ); +if (!v->salt) { + ti->error ="Cannot allocate salt" ; +return -ENOMEM ; + } +if (strlen (arg) != v->salt_size *2 || +hex2bin (v->salt, arg, v->salt_size)) { + ti->error ="Invalid salt" ; +return -EINVAL ; + } + } +if (v->shash_tfm) { +SHASH_DESC_ON_STACK (desc, v->shash_tfm); +int r; + +/* +* Compute the pre-salted hash state that can be passed to +* crypto_shash_import() for each block later. +*/ + v->initial_hashstate =kmalloc ( +crypto_shash_statesize (v->shash_tfm),GFP_KERNEL ); +if (!v->initial_hashstate) { + ti->error ="Cannot allocate initial hash state" ; +return -ENOMEM ; + } + desc->tfm = v->shash_tfm; + r =crypto_shash_init (desc) ?: +crypto_shash_update (desc, v->salt, v->salt_size) ?: +crypto_shash_export (desc, v->initial_hashstate); +if (r) { + ti->error ="Cannot set up initial hash state" ; +return r; + } + }return 0 ; }/* * Target parameters:
@@ -1304,25 +1385,13 @@ static int verity_ctr(struct dm_target *ti, unsigned int argc, char **argv) r = -EINVAL ; goto bad; } root_hash_digest_to_validate = argv[8 ]; -if (strcmp (argv[9 ],"-" )) { - v->salt_size =strlen (argv[9 ]) /2 ; - v->salt =kmalloc (v->salt_size,GFP_KERNEL ); -if (!v->salt) { - ti->error ="Cannot allocate salt" ; - r = -ENOMEM ; - goto bad; - } -if (strlen (argv[9 ]) != v->salt_size *2 || -hex2bin (v->salt, argv[9 ], v->salt_size)) { - ti->error ="Invalid salt" ; - r = -EINVAL ; - goto bad; - } - } + r =verity_setup_salt_and_hashstate (v, argv[9 ]); +if (r) + goto bad; argv +=10 ; argc -=10 ;/* Optional parameters */
@@ -1420,11 +1489,11 @@ static int verity_ctr(struct dm_target *ti, unsigned int argc, char **argv) ti->error ="Cannot allocate workqueue" ; r = -ENOMEM ; goto bad; } - ti->per_io_data_size =sizeof (struct dm_verity_io ) + v->ahash_reqsize; + ti->per_io_data_size =sizeof (struct dm_verity_io ) + v->hash_reqsize; r =verity_fec_ctr (v);if (r) goto bad;
drivers/md/dm-verity.h +11 -7
@@ -37,13 +37,15 @@ struct dm_verity {struct dm_dev *data_dev;struct dm_dev *hash_dev;struct dm_target *ti;struct dm_bufio_client *bufio;char *alg_name; -struct crypto_ahash *tfm; +struct crypto_ahash *ahash_tfm;/* either this or shash_tfm is set */ +struct crypto_shash *shash_tfm;/* either this or ahash_tfm is set */ u8 *root_digest;/* digest of the root block */ u8 *salt;/* salt: its size is salt_size */ +u8 *initial_hashstate;/* salted initial state, if shash_tfm is set */ u8 *zero_digest;/* digest for a zero block */ unsigned int salt_size;sector_t data_start;/* data offset in 512-byte sectors */ sector_t hash_start;/* hash start in blocks */ sector_t data_blocks;/* the number of data blocks */
@@ -54,11 +56,11 @@ struct dm_verity {unsigned char levels;/* the number of tree levels */ unsigned char version; bool hash_failed:1; /* set if hash of any block failed */ bool use_bh_wq:1; /* try to verify in BH wq before normal work-queue */unsigned int digest_size;/* digest size for the current hash algorithm */ -unsigned int ahash_reqsize;/* the size of temporary space for crypto */ +unsigned int hash_reqsize;/* the size of temporary space for crypto */ enum verity_mode mode;/* mode for handling verification errors */ unsigned int corrupted_errs;/* Number of errors for corrupted blocks */ struct workqueue_struct *verify_wq;
@@ -91,19 +93,21 @@ struct dm_verity_io {u8 real_digest[HASH_MAX_DIGESTSIZE ];u8 want_digest[HASH_MAX_DIGESTSIZE ];/* -* This struct is followed by a variable-sized struct ahash_request of -* size v->ahash_reqsize. To access it, use verity_io_hash_req(). +* This struct is followed by a variable-sized hash request of size +* v->hash_reqsize, either a struct ahash_request or a struct shash_desc +* (depending on whether ahash_tfm or shash_tfm is being used). To +* access it, use verity_io_hash_req(). */ }; -static inline struct ahash_request *verity_io_hash_req (struct dm_verity *v, -struct dm_verity_io *io) +static inline void *verity_io_hash_req (struct dm_verity *v, +struct dm_verity_io *io) { -return (struct ahash_request *)(io +1 ); +return io +1 ; }static inline u8 *verity_io_real_digest (struct dm_verity *v,struct dm_verity_io *io) {
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 14/15 dm-verity: reduce scope of real and wanted digests
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
In preparation for supporting multibuffer hashing where dm-verity will need to keep track of the real and wanted digests for multiple data blocks simultaneously, stop using the want_digest and real_digest fields of struct dm_verity_io from so many different places. Specifically:
- Make various functions take want_digest as a parameter rather than having it be implicitly passed via the struct dm_verity_io.
- Add a new tmp_digest field, and use this instead of real_digest when computing a hash solely for the purpose of immediately checking it.
The result is that real_digest and want_digest are only used by verity_verify_io() and verity_check_data_block_hash().
--- drivers/md/dm-verity-fec.c | 19 +++++++++--------- drivers/md/dm-verity-fec.h | 5 +++-- drivers/md/dm-verity-target.c | 36 ++++++++++++++++++----------------- drivers/md/dm-verity.h | 1 + 4 files changed, 32 insertions(+), 29 deletions(-)
drivers/md/dm-verity-fec.c +9 -10
@@ -185,15 +185,14 @@ static int fec_decode_bufs(struct dm_verity *v, struct dm_verity_io *io,*/ static int fec_is_erasure (struct dm_verity *v,struct dm_verity_io *io,u8 *want_digest,u8 *data) {if (unlikely (verity_hash (v, io, data,1 << v->data_dev_block_bits, -verity_io_real_digest (v, io), true))) + io->tmp_digest, true)))return 0 ; -return memcmp (verity_io_real_digest (v, io), want_digest, - v->digest_size) !=0 ; +return memcmp (io->tmp_digest, want_digest, v->digest_size) !=0 ; }/* * Read data blocks that are part of the RS block and deinterleave as much as * fits into buffers. Check for erasure locations if @neras is non-NULL.
@@ -360,11 +359,11 @@ static void fec_init_bufs(struct dm_verity *v, struct dm_verity_fec_io *fio)* (indicated by @offset) in fio->output. If @use_erasures is non-zero, uses * hashes to locate erasures. */ static int fec_decode_rsb (struct dm_verity *v,struct dm_verity_io *io,struct dm_verity_fec_io *fio,u64 rsb,u64 offset, -bool use_erasures) +const u8 *want_digest,bool use_erasures) {int r, neras =0 ;unsigned int pos; r =fec_alloc_bufs (v, fio);
@@ -386,27 +385,27 @@ static int fec_decode_rsb(struct dm_verity *v, struct dm_verity_io *io, pos += fio->nbufs <<DM_VERITY_FEC_BUF_RS_BITS ; }/* Always re-validate the corrected block against the expected hash */ r =verity_hash (v, io, fio->output,1 << v->data_dev_block_bits, -verity_io_real_digest (v, io), true); + io->tmp_digest, true);if (unlikely (r <0 ))return r; -if (memcmp (verity_io_real_digest (v, io),verity_io_want_digest (v, io), - v->digest_size)) { +if (memcmp (io->tmp_digest, want_digest, v->digest_size)) {DMERR_LIMIT ("%s: FEC %llu: failed to correct (%d erasures)" , v->data_dev->name, (unsigned long long )rsb, neras);return -EILSEQ ; }return 0 ; }/* Correct errors in a block. Copies corrected block to dest. */ int verity_fec_decode (struct dm_verity *v,struct dm_verity_io *io, -enum verity_block_type type,sector_t block,u8 *dest) +enum verity_block_type type,const u8 *want_digest, +sector_t block,u8 *dest) {int r;struct dm_verity_fec_io *fio =fec_io (io);u64 offset, res, rsb;
@@ -445,13 +444,13 @@ int verity_fec_decode(struct dm_verity *v, struct dm_verity_io *io,/* * Locating erasures is slow, so attempt to recover the block without * them first. Do a second attempt with erasures if the corruption is * bad enough. */ - r =fec_decode_rsb (v, io, fio, rsb, offset, false); + r =fec_decode_rsb (v, io, fio, rsb, offset, want_digest, false);if (r <0 ) { - r =fec_decode_rsb (v, io, fio, rsb, offset, true); + r =fec_decode_rsb (v, io, fio, rsb, offset, want_digest, true);if (r <0 ) goto done; }memcpy (dest, fio->output,1 << v->data_dev_block_bits);
drivers/md/dm-verity-fec.h +3 -2
@@ -66,12 +66,12 @@ struct dm_verity_fec_io {#define DM_VERITY_OPTS_FEC 8extern bool verity_fec_is_enabled (struct dm_verity *v);extern int verity_fec_decode (struct dm_verity *v,struct dm_verity_io *io, -enum verity_block_type type,sector_t block, -u8 *dest); +enum verity_block_type type,const u8 *want_digest, +sector_t block,u8 *dest);extern unsigned int verity_fec_status_table (struct dm_verity *v,unsigned int sz,char *result,unsigned int maxlen);extern void verity_fec_finish_io (struct dm_verity_io *io);
@@ -97,10 +97,11 @@ static inline bool verity_fec_is_enabled(struct dm_verity *v) }static inline int 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) {return -EOPNOTSUPP ; }
drivers/md/dm-verity-target.c +19 -17
@@ -283,16 +283,16 @@ static int verity_handle_err(struct dm_verity *v, enum verity_block_type type,/* * Verify hash of a metadata block pertaining to the specified data block * ("block" argument) at a specified level ("level" argument). * -* On successful return, verity_io_want_digest(v, io) contains the hash value -* for a lower tree level or for the data block (if we're at the lowest level). +* On successful return, want_digest contains the hash value for a lower tree +* level or for the data block (if we're at the lowest level). * * If "skip_unverified" is true, unverified buffer is skipped and 1 is returned. * If "skip_unverified" is false, unverified buffer is hashed and verified -* against current value of verity_io_want_digest(v, io). +* against current value of want_digest. */ static int verity_verify_level (struct dm_verity *v,struct dm_verity_io *io,sector_t block,int level,bool skip_unverified,u8 *want_digest) {
@@ -331,26 +331,26 @@ static int verity_verify_level(struct dm_verity *v, struct dm_verity_io *io, r =1 ; goto release_ret_r; } r =verity_hash (v, io, data,1 << v->hash_dev_block_bits, -verity_io_real_digest (v, io), !io->in_bh); + io->tmp_digest, !io->in_bh);if (unlikely (r <0 )) goto release_ret_r; -if (likely (memcmp (verity_io_real_digest (v, io), want_digest, +if (likely (memcmp (io->tmp_digest, want_digest, v->digest_size) ==0 )) aux->hash_verified =1 ;else if (static_branch_unlikely (&use_bh_wq_enabled) && io->in_bh) {/* * Error handling code (FEC included) cannot be run in a * tasklet since it may sleep, so fallback to work-queue. */ r = -EAGAIN ; goto release_ret_r; }else if (verity_fec_decode (v, io,DM_VERITY_BLOCK_TYPE_METADATA , - hash_block, data) ==0 ) + want_digest, hash_block, data) ==0 ) aux->hash_verified =1 ;else if (verity_handle_err (v,DM_VERITY_BLOCK_TYPE_METADATA , hash_block)) { struct bio *bio =
@@ -409,11 +409,12 @@ int verity_hash_for_block(struct dm_verity *v, struct dm_verity_io *io,return r; } static noinline int verity_recheck(struct dm_verity *v, struct dm_verity_io *io, -sector_t cur_block,u8 *dest) +const u8 *want_digest,sector_t cur_block, +u8 *dest) {struct page *page;void *buffer;int r;struct dm_io_request io_req;
@@ -433,16 +434,15 @@ static noinline int verity_recheck(struct dm_verity *v, struct dm_verity_io *io, r =dm_io (&io_req,1 , &io_loc,NULL ,IOPRIO_DEFAULT );if (unlikely (r)) goto free_ret; r =verity_hash (v, io, buffer,1 << v->data_dev_block_bits, -verity_io_real_digest (v, io), true); + io->tmp_digest, true);if (unlikely (r)) goto free_ret; -if (memcmp (verity_io_real_digest (v, io), -verity_io_want_digest (v, io), v->digest_size)) { +if (memcmp (io->tmp_digest, want_digest, v->digest_size)) { r = -EIO ; goto free_ret; }memcpy (dest, buffer,1 << v->data_dev_block_bits);
@@ -453,28 +453,29 @@ static noinline int verity_recheck(struct dm_verity *v, struct dm_verity_io *io,return r; }static int verity_handle_data_hash_mismatch (struct dm_verity *v,struct dm_verity_io *io, -struct bio *bio,sector_t blkno, -u8 *data) +struct bio *bio, +const u8 *want_digest, +sector_t blkno,u8 *data) {if (static_branch_unlikely (&use_bh_wq_enabled) && io->in_bh) {/* * Error handling code (FEC included) cannot be run in the * BH workqueue, so fallback to a standard workqueue. */ return -EAGAIN ; } -if (verity_recheck (v, io, blkno, data) ==0 ) { +if (verity_recheck (v, io, want_digest, blkno, data) ==0 ) {if (v->validated_blocks)set_bit (blkno, v->validated_blocks);return 0 ; }#if defined(CONFIG_DM_VERITY_FEC ) -if (verity_fec_decode (v, io,DM_VERITY_BLOCK_TYPE_DATA , blkno, - data) ==0 ) +if (verity_fec_decode (v, io,DM_VERITY_BLOCK_TYPE_DATA , want_digest, + blkno, data) ==0 )return 0 ;#endif if (bio->bi_status)return -EIO ;/* Error correction failed; Just return error */
@@ -561,12 +562,13 @@ static int verity_verify_io(struct dm_verity_io *io)if (v->validated_blocks)set_bit (cur_block, v->validated_blocks);kunmap_local (data);continue ; } - r =verity_handle_data_hash_mismatch (v, io, bio, cur_block, - data); + r =verity_handle_data_hash_mismatch (v, io, bio, +verity_io_want_digest (v, io), + cur_block, data);kunmap_local (data);if (unlikely (r))return r; }
drivers/md/dm-verity.h +1 -0
@@ -89,10 +89,11 @@ struct dm_verity_io {bool in_bh;struct work_struct work;struct work_struct bh_work; +u8 tmp_digest[HASH_MAX_DIGESTSIZE ];u8 real_digest[HASH_MAX_DIGESTSIZE ];u8 want_digest[HASH_MAX_DIGESTSIZE ];/* * This struct is followed by a variable-sized hash request of size
Signature · 1 line
2.45.1
Eric Biggers
maintainerPATCH v5 15/15 dm-verity: improve performance by using multibuffer hashing
In reply to Eric BiggersFrom: Eric Biggers <[email protected]>
When supported by the hash algorithm, use crypto_shash_finup_mb() to interleave the hashing of pairs of data blocks. On some CPUs this nearly doubles hashing performance. The increase in overall throughput of cold-cache dm-verity reads that I'm seeing on arm64 and x86_64 is roughly 35% (though this metric is hard to measure as it jumps around a lot).
For now this is only done on data blocks, not Merkle tree blocks. We could use finup_mb on Merkle tree blocks too, but that is less important as there aren't as many Merkle tree blocks as data blocks, and that would require some additional code restructuring.
--- drivers/md/dm-verity-target.c | 171 ++++++++++++++++++++++++++-------- drivers/md/dm-verity.h | 34 ++++--- 2 files changed, 151 insertions(+), 54 deletions(-)
drivers/md/dm-verity-target.c +131 -40
@@ -181,22 +181,28 @@ static int verity_ahash_final(struct dm_verity *v, struct ahash_request *req, r =crypto_wait_req (crypto_ahash_final (req), wait); out:return r; } +static int verity_ahash (struct dm_verity *v,struct dm_verity_io *io, +const u8 *data,size_t len,u8 *digest,bool may_sleep) +{ +struct ahash_request *req =verity_io_hash_req (v, io); +struct crypto_wait wait; + +return verity_ahash_init (v, req, &wait, may_sleep) ?: +verity_ahash_update (v, req, data, len, &wait) ?: +verity_ahash_final (v, req, digest, &wait); +} +int verity_hash (struct dm_verity *v,struct dm_verity_io *io,const u8 *data,size_t len,u8 *digest,bool may_sleep) {int r;if (static_branch_unlikely (&ahash_enabled) && !v->shash_tfm) { -struct ahash_request *req =verity_io_hash_req (v, io); -struct crypto_wait wait; - - r =verity_ahash_init (v, req, &wait, may_sleep) ?: -verity_ahash_update (v, req, data, len, &wait) ?: -verity_ahash_final (v, req, digest, &wait); + r =verity_ahash (v, io, data, len, digest, may_sleep); }else {struct shash_desc *desc =verity_io_hash_req (v, io); desc->tfm = v->shash_tfm; r = crypto_shash_import(desc, v->initial_hashstate) ?:
@@ -205,10 +211,38 @@ int verity_hash(struct dm_verity *v, struct dm_verity_io *io,if (unlikely (r))DMERR ("Error hashing block: %d" , r);return r; } +static int verity_hash_mb (struct dm_verity *v,struct dm_verity_io *io, +const u8 *data[],size_t len,u8 *digests[], +int num_blocks) +{ +int r =0 ; + +if (static_branch_unlikely (&ahash_enabled) && !v->shash_tfm) { +int i; + +/* Note: in practice num_blocks is always 1 in this case. */ +for (i =0 ; i < num_blocks; i++) { + r =verity_ahash (v, io, data[i], len, digests[i], + !io->in_bh); +if (r) +break ; + } + }else { +struct shash_desc *desc =verity_io_hash_req (v, io); + + desc->tfm = v->shash_tfm; + r =crypto_shash_import (desc, v->initial_hashstate) ?: +crypto_shash_finup_mb (desc, data, len, digests, num_blocks); + } +if (unlikely (r)) +DMERR ("Error hashing blocks: %d" , r); +return r; +} +static void verity_hash_at_level (struct dm_verity *v,sector_t block,int level,sector_t *hash_block,unsigned int *offset) {sector_t position =verity_position_at_level (v, block, level);unsigned int idx;
@@ -454,13 +488,16 @@ static noinline int verity_recheck(struct dm_verity *v, struct dm_verity_io *io, }static int verity_handle_data_hash_mismatch (struct dm_verity *v,struct dm_verity_io *io,struct bio *bio, -const u8 *want_digest, -sector_t blkno,u8 *data) +struct pending_block *block) { +const u8 *want_digest = block->want_digest; +sector_t blkno = block->blkno; +u8 *data = block->data; +if (static_branch_unlikely (&use_bh_wq_enabled) && io->in_bh) {/* * Error handling code (FEC included) cannot be run in the * BH workqueue, so fallback to a standard workqueue. */
@@ -484,10 +521,60 @@ static int verity_handle_data_hash_mismatch(struct dm_verity *v,return -EIO ; }return 0 ; } +static void verity_clear_pending_blocks (struct dm_verity_io *io) +{ +int i; + +for (i = io->num_pending -1 ; i >=0 ; i--) { +kunmap_local (io->pending_blocks[i].data); + io->pending_blocks[i].data =NULL ; + } + io->num_pending =0 ; +} + +static int verity_verify_pending_blocks (struct dm_verity *v, +struct dm_verity_io *io, +struct bio *bio) +{ +const u8 *data[DM_VERITY_MAX_PENDING_DATA_BLOCKS ]; +u8 *real_digests[DM_VERITY_MAX_PENDING_DATA_BLOCKS ]; +int i; +int r; + +if (io->num_pending ==0 ) +return 0 ; + +for (i =0 ; i < io->num_pending; i++) { + data[i] = io->pending_blocks[i].data; + real_digests[i] = io->pending_blocks[i].real_digest; + } + + r =verity_hash_mb (v, io, data,1 << v->data_dev_block_bits, + real_digests, io->num_pending); +if (unlikely (r)) +return r; + +for (i =0 ; i < io->num_pending; i++) { +struct pending_block *block = &io->pending_blocks[i]; + +if (likely (memcmp (real_digests[i], block->want_digest, + v->digest_size) ==0 )) { +if (v->validated_blocks) +set_bit (block->blkno, v->validated_blocks); + }else { + r =verity_handle_data_hash_mismatch (v, io, bio, block); +if (unlikely (r)) +return r; + } + } +verity_clear_pending_blocks (io); +return 0 ; +} +/* * Verify one "dm_verity_io" structure. */ static int verity_verify_io (struct dm_verity_io *io) {
@@ -495,10 +582,13 @@ static int verity_verify_io(struct dm_verity_io *io)const unsigned int block_size =1 << v->data_dev_block_bits;struct bvec_iter iter_copy;struct bvec_iter *iter;struct bio *bio =dm_bio_from_per_bio_data (io, v->ti->per_io_data_size);unsigned int b; +int r; + + io->num_pending =0 ;if (static_branch_unlikely (&use_bh_wq_enabled) && io->in_bh) {/* * Copy the iterator in case we need to restart * verification in a work-queue.
@@ -508,36 +598,38 @@ static int verity_verify_io(struct dm_verity_io *io) }else iter = &io->iter;for (b =0 ; b < io->n_blocks; b++,bio_advance_iter (bio, iter, block_size)) { -int r; -sector_t cur_block = io->block + b; +sector_t blkno = io->block + b; +struct pending_block *block;bool is_zero;struct bio_vec bv;void *data;if (v->validated_blocks && bio->bi_status ==BLK_STS_OK && -likely (test_bit (cur_block, v->validated_blocks))) +likely (test_bit (blkno, v->validated_blocks)))continue ; - r =verity_hash_for_block (v, io, cur_block, -verity_io_want_digest (v, io), + block = &io->pending_blocks[io->num_pending]; + + r =verity_hash_for_block (v, io, blkno, block->want_digest, &is_zero);if (unlikely (r <0 )) -return r; + goto error; bv =bio_iter_iovec (bio, *iter);if (unlikely (bv.bv_len < block_size)) {/* * Data block spans pages. This should not happen, * since dm-verity sets dma_alignment to the data block * size minus 1, and dm-verity also doesn't allow the * data block size to be greater than PAGE_SIZE. */ DMERR_LIMIT ("unaligned io (data block spans pages)" ); -return -EIO ; + r = -EIO ; + goto error; } data =bvec_kmap_local (&bv);if (is_zero) {
@@ -547,34 +639,28 @@ static int verity_verify_io(struct dm_verity_io *io)*/ memset (data,0 , block_size);kunmap_local (data);continue ; } - - r =verity_hash (v, io, data, block_size, -verity_io_real_digest (v, io), !io->in_bh); -if (unlikely (r <0 )) { -kunmap_local (data); -return r; + block->data = data; + block->blkno = blkno; +if (++io->num_pending == v->mb_max_msgs) { + r =verity_verify_pending_blocks (v, io, bio); +if (unlikely (r)) + goto error; } - -if (likely (memcmp (verity_io_real_digest (v, io), -verity_io_want_digest (v, io), v->digest_size) ==0 )) { -if (v->validated_blocks) -set_bit (cur_block, v->validated_blocks); -kunmap_local (data); -continue ; - } - r =verity_handle_data_hash_mismatch (v, io, bio, -verity_io_want_digest (v, io), - cur_block, data); -kunmap_local (data); -if (unlikely (r)) -return r; } + r =verity_verify_pending_blocks (v, io, bio); +if (unlikely (r)) + goto error; +return 0 ; + +error: +verity_clear_pending_blocks (io); +return r; }/* * Skip verity work in response to I/O error when system is shutting down. */
@@ -1155,14 +1241,15 @@ static int verity_setup_hash_alg(struct dm_verity *v, const char *alg_name)/* * Allocate the hash transformation object that this dm-verity instance * will use. The vast majority of dm-verity users use CPU-based * hashing, so when possible use the shash API to minimize the crypto -* API overhead. If the ahash API resolves to a different driver -* (likely an off-CPU hardware offload), use ahash instead. Also use -* ahash if the obsolete dm-verity format with the appended salt is -* being used, so that quirk only needs to be handled in one place. +* API overhead, especially when multibuffer hashing is used. If the +* ahash API resolves to a different driver (likely an off-CPU hardware +* offload), use ahash instead. Also use ahash if the obsolete +* dm-verity format with the appended salt is being used, so that quirk +* only needs to be handled in one place. */ ahash =crypto_alloc_ahash (alg_name,0 , v->use_bh_wq ?CRYPTO_ALG_ASYNC :0 );if (IS_ERR (ahash)) { ti->error ="Cannot initialize hash function" ;
@@ -1186,17 +1273,21 @@ static int verity_setup_hash_alg(struct dm_verity *v, const char *alg_name) ahash =NULL ; v->shash_tfm = shash; v->digest_size =crypto_shash_digestsize (shash); v->hash_reqsize =sizeof (struct shash_desc ) +crypto_shash_descsize (shash); -DMINFO ("%s using shash \"%s\"" , alg_name, driver_name); + v->mb_max_msgs =min (crypto_shash_mb_max_msgs (shash), +DM_VERITY_MAX_PENDING_DATA_BLOCKS ); +DMINFO ("%s using shash \"%s\"%s" , alg_name, driver_name, + v->mb_max_msgs >1 ?" (multibuffer)" :"" ); }else { v->ahash_tfm = ahash;static_branch_inc (&ahash_enabled); v->digest_size =crypto_ahash_digestsize (ahash); v->hash_reqsize =sizeof (struct ahash_request ) +crypto_ahash_reqsize (ahash); + v->mb_max_msgs =1 ;DMINFO ("%s using ahash \"%s\"" , alg_name, driver_name); }if ((1 << v->hash_dev_block_bits) < v->digest_size *2 ) { ti->error ="Digest size too big" ;return -EINVAL ;
drivers/md/dm-verity.h +20 -14
@@ -55,10 +55,11 @@ struct dm_verity {unsigned char hash_per_block_bits;/* log2(hashes in hash block) */ unsigned char levels;/* the number of tree levels */ unsigned char version; bool hash_failed:1; /* set if hash of any block failed */ bool use_bh_wq:1; /* try to verify in BH wq before normal work-queue */ +unsigned char mb_max_msgs;/* max multibuffer hashing interleaving factor */ unsigned int digest_size;/* digest size for the current hash algorithm */ unsigned int hash_reqsize;/* the size of temporary space for crypto */ enum verity_mode mode;/* mode for handling verification errors */ unsigned int corrupted_errs;/* Number of errors for corrupted blocks */
@@ -74,10 +75,19 @@ struct dm_verity {struct dm_io_client *io;mempool_t recheck_pool; }; +#define DM_VERITY_MAX_PENDING_DATA_BLOCKS HASH_MAX_MB_MSGS + +struct pending_block { +void *data; +sector_t blkno; +u8 want_digest[HASH_MAX_DIGESTSIZE ]; +u8 real_digest[HASH_MAX_DIGESTSIZE ]; +}; +struct dm_verity_io {struct dm_verity *v;/* original value of bio->bi_end_io */ bio_end_io_t *orig_bi_end_io;
@@ -90,12 +100,20 @@ struct dm_verity_io {struct work_struct work;struct work_struct bh_work;u8 tmp_digest[HASH_MAX_DIGESTSIZE ]; -u8 real_digest[HASH_MAX_DIGESTSIZE ]; -u8 want_digest[HASH_MAX_DIGESTSIZE ]; + +/* +* This is the queue of data blocks that are pending verification. We +* allow multiple blocks to be queued up in order to support hash +* algorithm implementations that provide support for multibuffer +* hashing, i.e. interleaving the hashing of multiple messages. On many +* CPUs this improves performance significantly. +*/ +int num_pending; +struct pending_block pending_blocks[DM_VERITY_MAX_PENDING_DATA_BLOCKS ];/* * This struct is followed by a variable-sized hash request of size * v->hash_reqsize, either a struct ahash_request or a struct shash_desc * (depending on whether ahash_tfm or shash_tfm is being used). To
@@ -107,22 +125,10 @@ static inline void *verity_io_hash_req(struct dm_verity *v, struct dm_verity_io *io) {return io +1 ; } -static inline u8 *verity_io_real_digest (struct dm_verity *v, -struct dm_verity_io *io) -{ -return io->real_digest; -} - -static inline u8 *verity_io_want_digest (struct dm_verity *v, -struct dm_verity_io *io) -{ -return io->want_digest; -} -extern int verity_hash (struct dm_verity *v,struct dm_verity_io *io,const u8 *data,size_t len,u8 *digest,bool may_sleep);extern int verity_hash_for_block (struct dm_verity *v,struct dm_verity_io *io,sector_t block,u8 *digest,bool *is_zero);
Signature · 1 line
2.45.1
Herbert Xu
maintainerRe: [PATCH v5 15/15] dm-verity: improve performance by using multibuffer hashing
drivers/md/dm-verity-target.c static int verity_verify_io(struct dm_verity_io *io) · line 650
Quoted 6 lines · Eric Biggers
drivers/md/dm-verity-target.c +4
-0+if (++io->num_pending == v->mb_max_msgs) { + r =verity_verify_pending_blocks (v, io, bio); +if (unlikely (r)) + goto error; }
What is the overhead if you just let it accumulate as large a request as possible? We should let the underlying algorithm decide how to divide this up in the most optimal fashion.
Cheers,
Signature · 3 lines
Email: Herbert Xu <[email protected]> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
Eric Biggers
maintainerRe: [PATCH v5 15/15] dm-verity: improve performance by using multibuffer hashing
Quoted 12 lines · Herbert Xu
drivers/md/dm-verity-target.c static int verity_verify_io(struct dm_verity_io *io) · line 650
Quoted 6 lines · Eric Biggers
drivers/md/dm-verity-target.c +4
-0+if (++io->num_pending == v->mb_max_msgs) { + r =verity_verify_pending_blocks (v, io, bio); +if (unlikely (r)) + goto error; }What is the overhead if you just let it accumulate as large a request as possible? We should let the underlying algorithm decide how to divide this up in the most optimal fashion.
The queue adds 144*num_messages bytes to each bio. It's desirable to keep this memory overhead down. So it makes sense to limit the queue length to the multibuffer hashing interleaving factor. Yes we could build something where you could get a marginal performance benefit from amounts higher than that by saving indirect calls, but I think it wouldn't be worth bloating the per-IO memory.
Another thing to keep in mind is that with how the dm-verity code is currently structured, for each data block it gets the wanted hash from the Merkle tree (which it prefetched earlier) before hashing the data block. So I also worry that if we wait too long before starting to hash the data blocks, dm-verity will spend more time unnecessarily blocked on waiting for Merkle tree I/O.
- Eric
Eric Biggers
maintainerRe: [PATCH v5 15/15] dm-verity: improve performance by using multibuffer hashing
Quoted 20 lines · Eric Biggers
Quoted 12 lines · Herbert Xu
drivers/md/dm-verity-target.c static int verity_verify_io(struct dm_verity_io *io) · line 650
Quoted 6 lines · Eric Biggers
drivers/md/dm-verity-target.c +4
-0+if (++io->num_pending == v->mb_max_msgs) { + r =verity_verify_pending_blocks (v, io, bio); +if (unlikely (r)) + goto error; }What is the overhead if you just let it accumulate as large a request as possible? We should let the underlying algorithm decide how to divide this up in the most optimal fashion.
The queue adds 144*num_messages bytes to each bio. It's desirable to keep this memory overhead down. So it makes sense to limit the queue length to the multibuffer hashing interleaving factor. Yes we could build something where you could get a marginal performance benefit from amounts higher than that by saving indirect calls, but I think it wouldn't be worth bloating the per-IO memory.
Quoted 5 lines
Another thing to keep in mind is that with how the dm-verity code is currently structured, for each data block it gets the wanted hash from the Merkle tree (which it prefetched earlier) before hashing the data block. So I also worry that if we wait too long before starting to hash the data blocks, dm-verity will spend more time unnecessarily blocked on waiting for Merkle tree I/O.
To clarify, 144*num_messages bytes is for the current version that doesn't have a redundant shash_desc (or ahash_request which is even bigger) for each block. If you force those to be included, then that would of course go up.
Also, we can't really swap the order of "hashing a data block" and "getting the wanted hash of the block" in order to avoid the second issue I mentioned, because that would require removing the optimization where dm-verity zero-fills blocks whose hashes are the hash of the zero block instead of verifying them.
So it seems like the downsides of queueing up too many blocks outweigh the potential benefits.
- Eric
Quoted 7 lines · Eric Biggers
On many modern CPUs, it is possible to compute the SHA-256 hash of two equal-length messages in about the same time as a single message, if all the instructions are interleaved. This is because each SHA-256 (and also most other cryptographic hash functions) is inherently serialized and therefore can't always take advantage of the CPU's full throughput.
Quoted 7 lines
An earlier attempt to support multibuffer hashing in Linux was based around the ahash API. That approach had some major issues, as does the alternative ahash-based approach proposed by Herbert (see my response at
https://lore.kernel.org/linux-crypto/[email protected]/).
This patchset instead takes a much simpler approach of just adding a synchronous API for hashing equal-length messages.
I share Eric's skepticism that shoehorning this into ahash for theoretical reasons is going to lead anywhere. So I would strongly prefer this approach. We can always revisit this if/when this generic multibuffer ahash materializes.
So for this series
Hi Eric,
Quoted 14 lines · Eric Biggers
On many modern CPUs, it is possible to compute the SHA-256 hash of two equal-length messages in about the same time as a single message, if all the instructions are interleaved. This is because each SHA-256 (and also most other cryptographic hash functions) is inherently serialized and therefore can't always take advantage of the CPU's full throughput.
An earlier attempt to support multibuffer hashing in Linux was based around the ahash API. That approach had some major issues, as does the alternative ahash-based approach proposed by Herbert (see my response at
https://lore.kernel.org/linux-crypto/[email protected]/).
This patchset instead takes a much simpler approach of just adding a synchronous API for hashing equal-length messages.
Quoted 2 lines
This works well for dm-verity and fsverity, which use Merkle trees and therefore hash large numbers of equal-length messages.
Thank you for continuing to work on this! Improving dm-verity performance is a high priority for Android, and this patch series shows very promising results. FWIW, I would like to see this merged upstream, and any ahash improvements handled in follow-up patches. For the series:
Sami
Mikulas Patocka
maintainerHi
I'd like to ask what's the status of this patchset.
Will Herbert accept it? What's the planned kernel version where it will appear?
Mikulas
Quoted 96 lines · Eric Biggers
On many modern CPUs, it is possible to compute the SHA-256 hash of two equal-length messages in about the same time as a single message, if all the instructions are interleaved. This is because each SHA-256 (and also most other cryptographic hash functions) is inherently serialized and therefore can't always take advantage of the CPU's full throughput.
An earlier attempt to support multibuffer hashing in Linux was based around the ahash API. That approach had some major issues, as does the alternative ahash-based approach proposed by Herbert (see my response at
https://lore.kernel.org/linux-crypto/[email protected]/).
This patchset instead takes a much simpler approach of just adding a synchronous API for hashing equal-length messages.This works well for dm-verity and fsverity, which use Merkle trees and therefore hash large numbers of equal-length messages.
This patchset is organized as follows:
- Patch 1-3 add
crypto_shash_finup_mb()and tests for it.
- Patch 4-5 implementfinup_mbonx86_64and arm64, using an interleaving factor of 2.
- Patch 6 adds multibuffer hashing support to fsverity.
- Patches 7-14 are cleanups and optimizations to dm-verity that prepare the way for adding multibuffer hashing support. These don't depend on any of the previous patches.
- Patch 15 adds multibuffer hashing support to dm-verity.On CPUs that support multiple concurrent SHA-256's (all arm64 CPUs I tested, and AMD Zen CPUs), raw SHA-256 hashing throughput increases by 70-98%, and the throughput of cold-cache reads from dm-verity and fsverity increases by very roughly 35%.
Changed in v5:
- Reworked the dm-verity patches again. Split the preparation work into separate patches, fixed two bugs, and added some new cleanups.
- Other small cleanupsChanged in v4:
- Reorganized the fsverity and dm-verity code to have a unified code path for single-block vs. multi-block processing. For data blocks they now use onlycrypto_shash_finup_mb().Changed in v3:
- Change API from finup2x tofinup_mb. It now takes arrays of data buffer and output buffers, avoiding hardcoding 2x in the API.Changed in v2:
- Rebase onto cryptodev/master
- Add more comments to assembly
- Reorganize some of the assembly slightly
- Fix the claimed throughput improvement on arm64
- Fix incorrect kunmap order infs/verity/verify.c
- Adjust testmgr generation logic slightly
- Explicitly check forINT_MAXbefore casting unsigned int to int
- Mention SHA3 based parallel hashes
- Mention AVX512-based approachEric Biggers (15):
crypto: shash - add support forfinup_mb
crypto: testmgr - generate power-of-2 lengths more often
crypto: testmgr - add tests forfinup_mb
crypto: x86/sha256-ni - add support forfinup_mb
crypto: arm64/sha256-ce - add support forfinup_mb
fsverity: improve performance by using multibuffer hashing
dm-verity: move hash algorithm setup into its own function
dm-verity: move data hash mismatch handling into its own function
dm-verity: makereal_digestandwant_digestfixed-length
dm-verity: providedma_alignmentlimit inio_hints
dm-verity: always "map" the data blocks
dm-verity: makeverity_hash()takedm_verity_ioinstead ofahash_request
dm-verity: hash blocks with shash import+finup when possible
dm-verity: reduce scope of real and wanted digests
dm-verity: improve performance by using multibuffer hashingarch/arm64/crypto/sha2-ce-core.S | 281 +++++++++++++- arch/arm64/crypto/sha2-ce-glue.c | 40 ++ arch/x86/crypto/sha256_ni_asm.S | 368 ++++++++++++++++++ arch/x86/crypto/sha256_ssse3_glue.c | 39 ++ crypto/shash.c | 58 +++ crypto/testmgr.c | 90 ++++- drivers/md/dm-verity-fec.c | 49 +-- drivers/md/dm-verity-fec.h | 9 +- drivers/md/dm-verity-target.c | 582 ++++++++++++++++------------ drivers/md/dm-verity.h | 66 ++-- fs/verity/fsverity_private.h | 7 + fs/verity/hash_algs.c | 8 +- fs/verity/verify.c | 170 ++++++-- include/crypto/hash.h | 52 ++- 14 files changed, 1440 insertions(+), 379 deletions(-)base-commit:
6d4e1993a30539f556da2ebd36f1936c583eb812Signature · 1 line
2.45.1
Eric Biggers
maintainerQuoted 8 lines · Mikulas Patocka
Hi
I'd like to ask what's the status of this patchset.
Will Herbert accept it? What's the planned kernel version where it will appear?
Mikulas
It's blocked by Herbert wanting to design the multibuffer hashing API in a more complex way that doesn't make sense. See the previous discussions. I don't know when Herbert will change his mind, so for now I've shifted my focus to the Android kernels. (In the mean time, the improvements that don't depend on the actual multibuffer hashing API still make sense to apply upstream, though.)
- Eric
Quoted 9 lines · Eric Biggers
Quoted 7 lines · Mikulas Patocka
Hi
I'd like to ask what's the status of this patchset.
Will Herbert accept it? What's the planned kernel version where it will appear?
Quoted 6 lines
Quoted 1 line
Mikulas
It's blocked by Herbert wanting to design the multibuffer hashing API in a more complex way that doesn't make sense. See the previous discussions. I don't know when Herbert will change his mind, so for now I've shifted my focus to the Android kernels.
Yeah, this is really quite unfortunate, especially since the alternative approach doesn't appear to be forthcoming.
I'd prefer Eric's approach over what Herbert is proposing, as the former is available today and only addressess problems that are actually known to exist, rather than handwavy claims about what future developments in IPsec or h/w crypto accelerators might make meaningful use of.