| Message ID | 20240603183731.108986-2-ebiggers@kernel.org (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Herbert Xu |
| Headers | show |
| Series | Optimize dm-verity and fsverity using multibuffer hashing | expand |
On Mon, 3 Jun 2024 at 20:39, Eric Biggers <ebiggers@kernel.org> wrote: > > From: Eric Biggers <ebiggers@google.com> > > 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. > It's not only about out-of-order/superscalar execution. In some cases (at least on ARM), it takes more than a cycle for the result of an instruction to become available to the next one, even if the computation itself completes in a single cycle, and this affects in-order cores as well. The crux here is that the associated benefit only exists if the independent inputs can be interleaved at the instruction level. OoO cores will have some tolerance for deviations from this, but in the general case, this kind of multi-stream processing requires meticulous parallellization. That also means that it is substantially different from the asynchronous accelerator use case where a single IP block may have different queues that can be used in parallel. For these, it might make sense to provide some infrastructure to mix inputs from disparate sources, but the same logic is unlikely to be useful for the CPU based parallel hashing case. ... > > 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. > This is an independent optimization, right? This could be useful even more sequential hashing, and is not a fundamental aspect of parallel hashing? > 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.) > I could imagine cases where 3-way would have an advantage over 2-way - it is highly uarch dependent, though, so I wouldn't spend too much time accommodating this before a use case actually materializes. > Signed-off-by: Eric Biggers <ebiggers@google.com> > --- > crypto/shash.c | 60 +++++++++++++++++++++++++++++++++++++++++++ > include/crypto/hash.h | 45 +++++++++++++++++++++++++++++++- > 2 files changed, 104 insertions(+), 1 deletion(-) > > diff --git a/crypto/shash.c b/crypto/shash.c > index 301ab42bf849..5a2352933fbf 100644 > --- a/crypto/shash.c > +++ b/crypto/shash.c > @@ -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 +359,20 @@ 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) { > + if (alg->mb_max_msgs > HASH_MAX_MB_MSGS) > + return -EINVAL; > + if (!alg->finup_mb) > + return -EINVAL; > + } else { > + if (alg->finup_mb) > + return -EINVAL; > + } > + > err = hash_prepare_alg(&alg->halg); > if (err) > return err; > > base->cra_type = &crypto_shash_type; > @@ -339,10 +396,13 @@ static int shash_prepare_alg(struct shash_alg *alg) > if (!alg->export) > alg->halg.statesize = alg->descsize; > if (!alg->setkey) > alg->setkey = shash_no_setkey; > > + if (!alg->mb_max_msgs) > + alg->mb_max_msgs = 1; > + > return 0; > } > > int crypto_register_shash(struct shash_alg *alg) > { > diff --git a/include/crypto/hash.h b/include/crypto/hash.h > index 2d5ea9f9ff43..002099610755 100644 > --- a/include/crypto/hash.h > +++ b/include/crypto/hash.h > @@ -154,11 +154,13 @@ struct ahash_alg { > struct shash_desc { > struct crypto_shash *tfm; > void *__ctx[] __aligned(ARCH_SLAB_MINALIGN); > }; > > -#define HASH_MAX_DIGESTSIZE 64 > +#define HASH_MAX_DIGESTSIZE 64 > + > +#define HASH_MAX_MB_MSGS 2 /* max value of crypto_shash_mb_max_msgs() */ > > /* > * Worst case is hmac(sha3-224-generic). Its context is a nested 'shash_desc' > * containing a 'struct sha3_state'. > */ > @@ -177,10 +179,19 @@ struct shash_desc { > * @finup: see struct ahash_alg > * @digest: see struct ahash_alg > * @export: see struct ahash_alg > * @import: see struct ahash_alg > * @setkey: see struct ahash_alg > + * @finup_mb: **[optional]** Multibuffer hashing support. Finish calculating > + * the digests of multiple messages, interleaving the instructions to > + * potentially achieve better performance than hashing each message > + * individually. The num_msgs argument will be between 2 and > + * @mb_max_msgs inclusively. If there are particular values of len > + * or num_msgs, or a particular calling context (e.g. no-SIMD) that > + * the implementation does not support with this method, the > + * implementation may return -EOPNOTSUPP from this method in those > + * cases to cause the crypto API to fall back to repeated finups. > * @init_tfm: Initialize the cryptographic transformation object. > * This function is called only once at the instantiation > * time, right after the transformation context was > * allocated. In case the cryptographic hardware has > * some special requirements which need to be handled > @@ -192,10 +203,11 @@ struct shash_desc { > * various changes set in @init_tfm. > * @clone_tfm: Copy transform into new object, may allocate memory. > * @descsize: Size of the operational state for the message digest. This state > * size is the memory size that needs to be allocated for > * shash_desc.__ctx > + * @mb_max_msgs: Maximum supported value of num_msgs argument to @finup_mb > * @halg: see struct hash_alg_common > * @HASH_ALG_COMMON: see struct hash_alg_common > */ > struct shash_alg { > int (*init)(struct shash_desc *desc); > @@ -208,15 +220,19 @@ struct shash_alg { > unsigned int len, u8 *out); > int (*export)(struct shash_desc *desc, void *out); > int (*import)(struct shash_desc *desc, const void *in); > int (*setkey)(struct crypto_shash *tfm, const u8 *key, > unsigned int keylen); > + int (*finup_mb)(struct shash_desc *desc, const u8 * const data[], > + unsigned int len, u8 * const outs[], > + unsigned int num_msgs); > int (*init_tfm)(struct crypto_shash *tfm); > void (*exit_tfm)(struct crypto_shash *tfm); > int (*clone_tfm)(struct crypto_shash *dst, struct crypto_shash *src); > > unsigned int descsize; > + unsigned int mb_max_msgs; > > union { > struct HASH_ALG_COMMON; > struct hash_alg_common halg; > }; > @@ -750,10 +766,20 @@ static inline unsigned int crypto_shash_digestsize(struct crypto_shash *tfm) > static inline unsigned int crypto_shash_statesize(struct crypto_shash *tfm) > { > return crypto_shash_alg(tfm)->statesize; > } > > +/* > + * Return the maximum supported multibuffer hashing interleaving factor, i.e. > + * the maximum num_msgs that can be passed to crypto_shash_finup_mb(). The > + * return value will be between 1 and HASH_MAX_MB_MSGS inclusively. > + */ > +static inline unsigned int crypto_shash_mb_max_msgs(struct crypto_shash *tfm) > +{ > + return crypto_shash_alg(tfm)->mb_max_msgs; > +} > + > static inline u32 crypto_shash_get_flags(struct crypto_shash *tfm) > { > return crypto_tfm_get_flags(crypto_shash_tfm(tfm)); > } > > @@ -843,10 +869,27 @@ int crypto_shash_digest(struct shash_desc *desc, const u8 *data, > * Return: 0 on success; < 0 if an error occurred. > */ > int crypto_shash_tfm_digest(struct crypto_shash *tfm, const u8 *data, > unsigned int len, u8 *out); > > +/** > + * crypto_shash_finup_mb() - multibuffer message hashing > + * @desc: the starting state that is forked for each message. It contains the > + * state after hashing a (possibly-empty) common prefix of the messages. > + * @data: the data of each message (not including any common prefix from @desc) > + * @len: length of each data buffer in bytes > + * @outs: output buffer for each message digest > + * @num_msgs: number of messages, i.e. the number of entries in @data and @outs. > + * This can't be more than crypto_shash_mb_max_msgs(). > + * > + * Context: Any context. > + * Return: 0 on success; a negative errno value on failure. > + */ > +int crypto_shash_finup_mb(struct shash_desc *desc, const u8 * const data[], > + unsigned int len, u8 * const outs[], > + unsigned int num_msgs); > + > /** > * crypto_shash_export() - extract operational state for message digest > * @desc: reference to the operational state handle whose state is exported > * @out: output buffer of sufficient size that can hold the hash state > * > -- > 2.45.1 > >
On Tue, Jun 04, 2024 at 08:55:48PM +0200, Ard Biesheuvel wrote: > > > > 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. > > > > This is an independent optimization, right? This could be useful even > more sequential hashing, and is not a fundamental aspect of parallel > hashing? If you're referring to the part about using a common starting state, that's not an independent optimization. Only multibuffer hashing processes multiple messages in one call and therefore has an opportunity to share a starting shash_desc for finup. This isn't just an optimization but it also makes the multibuffer hashing API and its implementation much simpler. With single-buffer there has to be one shash_desc per message as usual. If you're asking if crypto_shash_finup_mb() can be used even without multibuffer hashing support, the answer is yes. This patchset makes crypto_shash_finup_mb() fall back to crypto_shash_finup() as needed, and this is used by fsverity and dm-verity to have one code path that uses crypto_shash_finup_mb() instead of separate code paths that use crypto_shash_finup_mb() and crypto_shash_finup(). This just makes things a bit simpler and isn't an optimization; note that the fallback has to copy the shash_desc for each message beyond the first. - Eric
diff --git a/crypto/shash.c b/crypto/shash.c index 301ab42bf849..5a2352933fbf 100644 --- a/crypto/shash.c +++ b/crypto/shash.c @@ -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 +359,20 @@ 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) { + if (alg->mb_max_msgs > HASH_MAX_MB_MSGS) + return -EINVAL; + if (!alg->finup_mb) + return -EINVAL; + } else { + if (alg->finup_mb) + return -EINVAL; + } + err = hash_prepare_alg(&alg->halg); if (err) return err; base->cra_type = &crypto_shash_type; @@ -339,10 +396,13 @@ static int shash_prepare_alg(struct shash_alg *alg) if (!alg->export) alg->halg.statesize = alg->descsize; if (!alg->setkey) alg->setkey = shash_no_setkey; + if (!alg->mb_max_msgs) + alg->mb_max_msgs = 1; + return 0; } int crypto_register_shash(struct shash_alg *alg) { diff --git a/include/crypto/hash.h b/include/crypto/hash.h index 2d5ea9f9ff43..002099610755 100644 --- a/include/crypto/hash.h +++ b/include/crypto/hash.h @@ -154,11 +154,13 @@ struct ahash_alg { struct shash_desc { struct crypto_shash *tfm; void *__ctx[] __aligned(ARCH_SLAB_MINALIGN); }; -#define HASH_MAX_DIGESTSIZE 64 +#define HASH_MAX_DIGESTSIZE 64 + +#define HASH_MAX_MB_MSGS 2 /* max value of crypto_shash_mb_max_msgs() */ /* * Worst case is hmac(sha3-224-generic). Its context is a nested 'shash_desc' * containing a 'struct sha3_state'. */ @@ -177,10 +179,19 @@ struct shash_desc { * @finup: see struct ahash_alg * @digest: see struct ahash_alg * @export: see struct ahash_alg * @import: see struct ahash_alg * @setkey: see struct ahash_alg + * @finup_mb: **[optional]** Multibuffer hashing support. Finish calculating + * the digests of multiple messages, interleaving the instructions to + * potentially achieve better performance than hashing each message + * individually. The num_msgs argument will be between 2 and + * @mb_max_msgs inclusively. If there are particular values of len + * or num_msgs, or a particular calling context (e.g. no-SIMD) that + * the implementation does not support with this method, the + * implementation may return -EOPNOTSUPP from this method in those + * cases to cause the crypto API to fall back to repeated finups. * @init_tfm: Initialize the cryptographic transformation object. * This function is called only once at the instantiation * time, right after the transformation context was * allocated. In case the cryptographic hardware has * some special requirements which need to be handled @@ -192,10 +203,11 @@ struct shash_desc { * various changes set in @init_tfm. * @clone_tfm: Copy transform into new object, may allocate memory. * @descsize: Size of the operational state for the message digest. This state * size is the memory size that needs to be allocated for * shash_desc.__ctx + * @mb_max_msgs: Maximum supported value of num_msgs argument to @finup_mb * @halg: see struct hash_alg_common * @HASH_ALG_COMMON: see struct hash_alg_common */ struct shash_alg { int (*init)(struct shash_desc *desc); @@ -208,15 +220,19 @@ struct shash_alg { unsigned int len, u8 *out); int (*export)(struct shash_desc *desc, void *out); int (*import)(struct shash_desc *desc, const void *in); int (*setkey)(struct crypto_shash *tfm, const u8 *key, unsigned int keylen); + int (*finup_mb)(struct shash_desc *desc, const u8 * const data[], + unsigned int len, u8 * const outs[], + unsigned int num_msgs); int (*init_tfm)(struct crypto_shash *tfm); void (*exit_tfm)(struct crypto_shash *tfm); int (*clone_tfm)(struct crypto_shash *dst, struct crypto_shash *src); unsigned int descsize; + unsigned int mb_max_msgs; union { struct HASH_ALG_COMMON; struct hash_alg_common halg; }; @@ -750,10 +766,20 @@ static inline unsigned int crypto_shash_digestsize(struct crypto_shash *tfm) static inline unsigned int crypto_shash_statesize(struct crypto_shash *tfm) { return crypto_shash_alg(tfm)->statesize; } +/* + * Return the maximum supported multibuffer hashing interleaving factor, i.e. + * the maximum num_msgs that can be passed to crypto_shash_finup_mb(). The + * return value will be between 1 and HASH_MAX_MB_MSGS inclusively. + */ +static inline unsigned int crypto_shash_mb_max_msgs(struct crypto_shash *tfm) +{ + return crypto_shash_alg(tfm)->mb_max_msgs; +} + static inline u32 crypto_shash_get_flags(struct crypto_shash *tfm) { return crypto_tfm_get_flags(crypto_shash_tfm(tfm)); } @@ -843,10 +869,27 @@ int crypto_shash_digest(struct shash_desc *desc, const u8 *data, * Return: 0 on success; < 0 if an error occurred. */ int crypto_shash_tfm_digest(struct crypto_shash *tfm, const u8 *data, unsigned int len, u8 *out); +/** + * crypto_shash_finup_mb() - multibuffer message hashing + * @desc: the starting state that is forked for each message. It contains the + * state after hashing a (possibly-empty) common prefix of the messages. + * @data: the data of each message (not including any common prefix from @desc) + * @len: length of each data buffer in bytes + * @outs: output buffer for each message digest + * @num_msgs: number of messages, i.e. the number of entries in @data and @outs. + * This can't be more than crypto_shash_mb_max_msgs(). + * + * Context: Any context. + * Return: 0 on success; a negative errno value on failure. + */ +int crypto_shash_finup_mb(struct shash_desc *desc, const u8 * const data[], + unsigned int len, u8 * const outs[], + unsigned int num_msgs); + /** * crypto_shash_export() - extract operational state for message digest * @desc: reference to the operational state handle whose state is exported * @out: output buffer of sufficient size that can hold the hash state *