In this patch, the infrastructure needed to support multibuffer encryption implementation is added: a) Enhance mcryptd daemon to support skcipher requests. b) Add multi-buffer simd skcipher helper which presents the top-level algorithm as an skcipher. b) Update configuration to include multi-buffer encryption build support. For an introduction to the multi-buffer implementation, please see http://www.intel.com/content/www/us/en/communications/communications-ia-multi-buffer-paper.html Originally-by: Chandramouli Narayanan <mouli_7982@xxxxxxxxx> Signed-off-by: Megha Dey <megha.dey@xxxxxxxxxxxxxxx> Acked-by: Tim Chen <tim.c.chen@xxxxxxxxxxxxxxx> --- crypto/Kconfig | 15 +++ crypto/mcryptd.c | 298 +++++++++++++++++++++++++++++++++++++++++ crypto/simd.c | 164 +++++++++++++++++++++++ include/crypto/internal/simd.h | 3 + include/crypto/mcryptd.h | 35 +++++ 5 files changed, 515 insertions(+) diff --git a/crypto/Kconfig b/crypto/Kconfig index aac4bc9..d172459 100644 --- a/crypto/Kconfig +++ b/crypto/Kconfig @@ -1008,6 +1008,21 @@ config CRYPTO_AES_NI_INTEL ECB, CBC, LRW, PCBC, XTS. The 64 bit version has additional acceleration for CTR. +config CRYPTO_AES_CBC_MB + tristate "AES CBC algorithm (x86_64 Multi-Buffer, Experimental)" + depends on X86 && 64BIT + select CRYPTO_SIMD + select CRYPTO_MCRYPTD + help + AES CBC encryption implemented using multi-buffer technique. + This algorithm computes on multiple data lanes concurrently with + SIMD instructions for better throughput. It should only be used + when we expect many concurrent crypto requests to keep all the + data lanes filled to realize the performance benefit. If the data + lanes are unfilled, a flush operation will be initiated after some + delay to process the exisiting crypto jobs, adding some extra + latency to low load case. + config CRYPTO_AES_SPARC64 tristate "AES cipher algorithms (SPARC64)" depends on SPARC64 diff --git a/crypto/mcryptd.c b/crypto/mcryptd.c index 4e64726..4e4830e 100644 --- a/crypto/mcryptd.c +++ b/crypto/mcryptd.c @@ -273,6 +273,266 @@ static inline bool mcryptd_check_internal(struct rtattr **tb, u32 *type, return false; } +static int mcryptd_enqueue_skcipher_request(struct mcryptd_queue *queue, + struct crypto_async_request *request, + struct mcryptd_skcipher_request_ctx *rctx) +{ + int cpu, err; + struct mcryptd_cpu_queue *cpu_queue; + + cpu = get_cpu(); + cpu_queue = this_cpu_ptr(queue->cpu_queue); + rctx->tag.cpu = cpu; + + err = crypto_enqueue_request(&cpu_queue->queue, request); + pr_debug("enqueue request: cpu %d cpu_queue %p request %p\n", + cpu, cpu_queue, request); + queue_work_on(cpu, kcrypto_wq, &cpu_queue->work); + put_cpu(); + + return err; +} + +static int mcryptd_skcipher_setkey(struct crypto_skcipher *parent, + const u8 *key, unsigned int keylen) +{ + struct mcryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(parent); + struct crypto_skcipher *child = ctx->child; + int err; + + crypto_skcipher_clear_flags(child, CRYPTO_TFM_REQ_MASK); + crypto_skcipher_set_flags(child, crypto_skcipher_get_flags(parent) & + CRYPTO_TFM_REQ_MASK); + err = crypto_skcipher_setkey(child, key, keylen); + crypto_skcipher_set_flags(parent, crypto_skcipher_get_flags(child) & + CRYPTO_TFM_RES_MASK); + return err; +} + +static void mcryptd_skcipher_complete(struct skcipher_request *req, int err) +{ + struct mcryptd_skcipher_request_ctx *rctx = skcipher_request_ctx(req); + + local_bh_disable(); + rctx->complete(&req->base, err); + local_bh_enable(); +} + +static void mcryptd_skcipher_encrypt(struct crypto_async_request *base, + int err) +{ + struct skcipher_request *req = skcipher_request_cast(base); + struct mcryptd_skcipher_request_ctx *rctx = skcipher_request_ctx(req); + struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req); + struct mcryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(tfm); + struct crypto_skcipher *child = ctx->child; + struct skcipher_request subreq; + + if (unlikely(err == -EINPROGRESS)) + goto out; + + /* set up the skcipher request to work on */ + skcipher_request_set_tfm(&subreq, child); + skcipher_request_set_callback(&subreq, + CRYPTO_TFM_REQ_MAY_SLEEP, 0, 0); + skcipher_request_set_crypt(&subreq, req->src, req->dst, + req->cryptlen, req->iv); + + /* + * pass addr of descriptor stored in the request context + * so that the callee can get to the request context + */ + rctx->desc = subreq; + err = crypto_skcipher_encrypt(&rctx->desc); + + if (err) { + req->base.complete = rctx->complete; + goto out; + } + return; + +out: + mcryptd_skcipher_complete(req, err); +} + +static void mcryptd_skcipher_decrypt(struct crypto_async_request *base, + int err) +{ + struct skcipher_request *req = skcipher_request_cast(base); + struct mcryptd_skcipher_request_ctx *rctx = skcipher_request_ctx(req); + struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req); + struct mcryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(tfm); + struct crypto_skcipher *child = ctx->child; + struct skcipher_request subreq; + + if (unlikely(err == -EINPROGRESS)) + goto out; + + /* set up the skcipher request to work on */ + skcipher_request_set_tfm(&subreq, child); + skcipher_request_set_callback(&subreq, + CRYPTO_TFM_REQ_MAY_SLEEP, NULL, NULL); + skcipher_request_set_crypt(&subreq, req->src, req->dst, + req->cryptlen, req->iv); + + /* + * pass addr of descriptor stored in the request context + * so that the callee can get to the request context + */ + rctx->desc = subreq; + err = crypto_skcipher_decrypt(&rctx->desc); + + if (err) { + req->base.complete = rctx->complete; + goto out; + } + return; + +out: + mcryptd_skcipher_complete(req, err); +} + +static int mcryptd_skcipher_enqueue(struct skcipher_request *req, + crypto_completion_t complete) +{ + struct mcryptd_skcipher_request_ctx *rctx = + skcipher_request_ctx(req); + struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req); + struct mcryptd_queue *queue; + + queue = mcryptd_get_queue(crypto_skcipher_tfm(tfm)); + rctx->complete = req->base.complete; + req->base.complete = complete; + + return mcryptd_enqueue_skcipher_request(queue, &req->base, rctx); +} + +static int mcryptd_skcipher_encrypt_enqueue(struct skcipher_request *req) +{ + return mcryptd_skcipher_enqueue(req, mcryptd_skcipher_encrypt); +} + +static int mcryptd_skcipher_decrypt_enqueue(struct skcipher_request *req) +{ + return mcryptd_skcipher_enqueue(req, mcryptd_skcipher_decrypt); +} + +static int mcryptd_skcipher_init_tfm(struct crypto_skcipher *tfm) +{ + struct skcipher_instance *inst = skcipher_alg_instance(tfm); + struct mskcipherd_instance_ctx *ictx = skcipher_instance_ctx(inst); + struct crypto_skcipher_spawn *spawn = &ictx->spawn; + struct mcryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(tfm); + struct crypto_skcipher *cipher; + + cipher = crypto_spawn_skcipher(spawn); + if (IS_ERR(cipher)) + return PTR_ERR(cipher); + + ctx->child = cipher; + crypto_skcipher_set_reqsize(tfm, + sizeof(struct mcryptd_skcipher_request_ctx)); + return 0; +} + +static void mcryptd_skcipher_exit_tfm(struct crypto_skcipher *tfm) +{ + struct mcryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(tfm); + + crypto_free_skcipher(ctx->child); +} + +static void mcryptd_skcipher_free(struct skcipher_instance *inst) +{ + struct mskcipherd_instance_ctx *ctx = skcipher_instance_ctx(inst); + + crypto_drop_skcipher(&ctx->spawn); +} + +static int mcryptd_init_instance(struct crypto_instance *inst, + struct crypto_alg *alg) +{ + if (snprintf(inst->alg.cra_driver_name, CRYPTO_MAX_ALG_NAME, + "mcryptd(%s)", + alg->cra_driver_name) >= CRYPTO_MAX_ALG_NAME) + return -ENAMETOOLONG; + + memcpy(inst->alg.cra_name, alg->cra_name, CRYPTO_MAX_ALG_NAME); + inst->alg.cra_priority = alg->cra_priority + 50; + inst->alg.cra_blocksize = alg->cra_blocksize; + inst->alg.cra_alignmask = alg->cra_alignmask; + + return 0; +} + +static int mcryptd_create_skcipher(struct crypto_template *tmpl, + struct rtattr **tb, + struct mcryptd_queue *queue) +{ + struct mskcipherd_instance_ctx *ctx; + struct skcipher_instance *inst; + struct skcipher_alg *alg; + const char *name; + u32 type; + u32 mask; + int err; + + type = 0; + mask = CRYPTO_ALG_ASYNC; + + mcryptd_check_internal(tb, &type, &mask); + + name = crypto_attr_alg_name(tb[1]); + if (IS_ERR(name)) + return PTR_ERR(name); + + inst = kzalloc(sizeof(*inst) + sizeof(*ctx), GFP_KERNEL); + if (!inst) + return -ENOMEM; + + ctx = skcipher_instance_ctx(inst); + ctx->queue = queue; + + crypto_set_skcipher_spawn(&ctx->spawn, skcipher_crypto_instance(inst)); + err = crypto_grab_skcipher(&ctx->spawn, name, type, mask); + + if (err) + goto out_free_inst; + + alg = crypto_spawn_skcipher_alg(&ctx->spawn); + err = mcryptd_init_instance(skcipher_crypto_instance(inst), &alg->base); + if (err) + goto out_drop_skcipher; + + inst->alg.base.cra_flags = CRYPTO_ALG_ASYNC | + (alg->base.cra_flags & CRYPTO_ALG_INTERNAL); + + inst->alg.ivsize = crypto_skcipher_alg_ivsize(alg); + inst->alg.chunksize = crypto_skcipher_alg_chunksize(alg); + inst->alg.min_keysize = crypto_skcipher_alg_min_keysize(alg); + inst->alg.max_keysize = crypto_skcipher_alg_max_keysize(alg); + + inst->alg.base.cra_ctxsize = sizeof(struct mcryptd_skcipher_ctx); + + inst->alg.init = mcryptd_skcipher_init_tfm; + inst->alg.exit = mcryptd_skcipher_exit_tfm; + + inst->alg.setkey = mcryptd_skcipher_setkey; + inst->alg.encrypt = mcryptd_skcipher_encrypt_enqueue; + inst->alg.decrypt = mcryptd_skcipher_decrypt_enqueue; + + inst->free = mcryptd_skcipher_free; + + err = skcipher_register_instance(tmpl, inst); + if (err) { +out_drop_skcipher: + crypto_drop_skcipher(&ctx->spawn); +out_free_inst: + kfree(inst); + } + return err; +} + static int mcryptd_hash_init_tfm(struct crypto_tfm *tfm) { struct crypto_instance *inst = crypto_tfm_alg_instance(tfm); @@ -564,6 +824,8 @@ static int mcryptd_create(struct crypto_template *tmpl, struct rtattr **tb) return PTR_ERR(algt); switch (algt->type & algt->mask & CRYPTO_ALG_TYPE_MASK) { + case CRYPTO_ALG_TYPE_BLKCIPHER: + return mcryptd_create_skcipher(tmpl, tb, &mqueue); case CRYPTO_ALG_TYPE_DIGEST: return mcryptd_create_hash(tmpl, tb, &mqueue); break; @@ -595,6 +857,42 @@ static void mcryptd_free(struct crypto_instance *inst) .module = THIS_MODULE, }; +struct mcryptd_skcipher *mcryptd_alloc_skcipher(const char *alg_name, + u32 type, u32 mask) +{ + char cryptd_alg_name[CRYPTO_MAX_ALG_NAME]; + struct crypto_skcipher *tfm; + + if (snprintf(cryptd_alg_name, CRYPTO_MAX_ALG_NAME, + "mcryptd(%s)", alg_name) >= CRYPTO_MAX_ALG_NAME) + return ERR_PTR(-EINVAL); + tfm = crypto_alloc_skcipher(cryptd_alg_name, type, mask); + if (IS_ERR(tfm)) + return ERR_CAST(tfm); + if (tfm->base.__crt_alg->cra_module != THIS_MODULE) { + crypto_free_skcipher(tfm); + return ERR_PTR(-EINVAL); + } + + return container_of(tfm, struct mcryptd_skcipher, base); +} +EXPORT_SYMBOL_GPL(mcryptd_alloc_skcipher); + +struct crypto_skcipher *mcryptd_skcipher_child( + struct mcryptd_skcipher *tfm) +{ + struct mcryptd_skcipher_ctx *ctx = crypto_skcipher_ctx(&tfm->base); + + return ctx->child; +} +EXPORT_SYMBOL_GPL(mcryptd_skcipher_child); + +void mcryptd_free_skcipher(struct mcryptd_skcipher *tfm) +{ + crypto_free_skcipher(&tfm->base); +} +EXPORT_SYMBOL_GPL(mcryptd_free_skcipher); + struct mcryptd_ahash *mcryptd_alloc_ahash(const char *alg_name, u32 type, u32 mask) { diff --git a/crypto/simd.c b/crypto/simd.c index 8820337..6757d44 100644 --- a/crypto/simd.c +++ b/crypto/simd.c @@ -28,11 +28,14 @@ #include <crypto/cryptd.h> #include <crypto/internal/simd.h> #include <crypto/internal/skcipher.h> +#include <crypto/mcryptd.h> #include <linux/kernel.h> #include <linux/module.h> #include <linux/preempt.h> #include <asm/simd.h> +static struct mcryptd_alg_state cbc_mb_alg_state; + struct simd_skcipher_alg { const char *ialg_name; struct skcipher_alg alg; @@ -42,6 +45,10 @@ struct simd_skcipher_ctx { struct cryptd_skcipher *cryptd_tfm; }; +struct simd_skcipher_ctx_mb { + struct mcryptd_skcipher *mcryptd_tfm; +}; + static int simd_skcipher_setkey(struct crypto_skcipher *tfm, const u8 *key, unsigned int key_len) { @@ -203,6 +210,163 @@ struct simd_skcipher_alg *simd_skcipher_create_compat(const char *algname, } EXPORT_SYMBOL_GPL(simd_skcipher_create_compat); +static int simd_skcipher_setkey_mb(struct crypto_skcipher *tfm, const u8 *key, + unsigned int key_len) +{ + struct simd_skcipher_ctx_mb *ctx = crypto_skcipher_ctx(tfm); + struct crypto_skcipher *child = &ctx->mcryptd_tfm->base; + int err; + + crypto_skcipher_clear_flags(child, CRYPTO_TFM_REQ_MASK); + crypto_skcipher_set_flags(child, crypto_skcipher_get_flags(tfm) & + CRYPTO_TFM_REQ_MASK); + err = crypto_skcipher_setkey(child, key, key_len); + crypto_skcipher_set_flags(tfm, crypto_skcipher_get_flags(child) & + CRYPTO_TFM_RES_MASK); + return err; +} + +static int simd_skcipher_decrypt_mb(struct skcipher_request *req) +{ + struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req); + struct simd_skcipher_ctx_mb *ctx = crypto_skcipher_ctx(tfm); + struct skcipher_request *subreq; + struct crypto_skcipher *child; + + subreq = skcipher_request_ctx(req); + *subreq = *req; + + child = &ctx->mcryptd_tfm->base; + + skcipher_request_set_tfm(subreq, child); + + return crypto_skcipher_decrypt(subreq); +} + +static int simd_skcipher_encrypt_mb(struct skcipher_request *req) +{ + struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req); + struct simd_skcipher_ctx_mb *ctx = crypto_skcipher_ctx(tfm); + struct skcipher_request *subreq; + struct crypto_skcipher *child; + + subreq = skcipher_request_ctx(req); + *subreq = *req; + + child = &ctx->mcryptd_tfm->base; + + skcipher_request_set_tfm(subreq, child); + + return crypto_skcipher_encrypt(subreq); +} + +static void simd_skcipher_exit_mb(struct crypto_skcipher *tfm) +{ + struct simd_skcipher_ctx_mb *ctx = crypto_skcipher_ctx(tfm); + + mcryptd_free_skcipher(ctx->mcryptd_tfm); +} + +static int simd_skcipher_init_mb(struct crypto_skcipher *tfm) +{ + struct simd_skcipher_ctx_mb *ctx = crypto_skcipher_ctx(tfm); + struct mcryptd_skcipher *mcryptd_tfm; + struct simd_skcipher_alg *salg; + struct skcipher_alg *alg; + unsigned reqsize; + struct mcryptd_skcipher_ctx *mctx; + + alg = crypto_skcipher_alg(tfm); + salg = container_of(alg, struct simd_skcipher_alg, alg); + + mcryptd_tfm = mcryptd_alloc_skcipher(salg->ialg_name, + CRYPTO_ALG_INTERNAL, + CRYPTO_ALG_INTERNAL); + if (IS_ERR(mcryptd_tfm)) + return PTR_ERR(mcryptd_tfm); + + mctx = crypto_skcipher_ctx(&mcryptd_tfm->base); + + mctx->alg_state = &cbc_mb_alg_state; + ctx->mcryptd_tfm = mcryptd_tfm; + + reqsize = sizeof(struct skcipher_request); + reqsize += crypto_skcipher_reqsize(&mcryptd_tfm->base); + + crypto_skcipher_set_reqsize(tfm, reqsize); + + return 0; +} + +struct simd_skcipher_alg *simd_skcipher_create_compat_mb(const char *algname, + const char *drvname, + const char *basename) +{ + struct simd_skcipher_alg *salg; + struct crypto_skcipher *tfm; + struct skcipher_alg *ialg; + struct skcipher_alg *alg; + int err; + + tfm = crypto_alloc_skcipher(basename, CRYPTO_ALG_INTERNAL, + CRYPTO_ALG_INTERNAL | CRYPTO_ALG_ASYNC); + if (IS_ERR(tfm)) + return ERR_CAST(tfm); + + ialg = crypto_skcipher_alg(tfm); + + salg = kzalloc(sizeof(*salg), GFP_KERNEL); + if (!salg) { + salg = ERR_PTR(-ENOMEM); + goto out_put_tfm; + } + + salg->ialg_name = basename; + alg = &salg->alg; + + err = -ENAMETOOLONG; + if (snprintf(alg->base.cra_name, CRYPTO_MAX_ALG_NAME, "%s", algname) >= + CRYPTO_MAX_ALG_NAME) + goto out_free_salg; + + if (snprintf(alg->base.cra_driver_name, CRYPTO_MAX_ALG_NAME, "%s", + drvname) >= CRYPTO_MAX_ALG_NAME) + goto out_free_salg; + + alg->base.cra_flags = CRYPTO_ALG_ASYNC; + alg->base.cra_priority = ialg->base.cra_priority; + alg->base.cra_blocksize = ialg->base.cra_blocksize; + alg->base.cra_alignmask = ialg->base.cra_alignmask; + alg->base.cra_module = ialg->base.cra_module; + alg->base.cra_ctxsize = sizeof(struct simd_skcipher_ctx_mb); + + alg->ivsize = ialg->ivsize; + alg->chunksize = ialg->chunksize; + alg->min_keysize = ialg->min_keysize; + alg->max_keysize = ialg->max_keysize; + + alg->init = simd_skcipher_init_mb; + alg->exit = simd_skcipher_exit_mb; + + alg->setkey = simd_skcipher_setkey_mb; + alg->encrypt = simd_skcipher_encrypt_mb; + alg->decrypt = simd_skcipher_decrypt_mb; + + err = crypto_register_skcipher(alg); + if (err) + goto out_free_salg; + +out_put_tfm: + crypto_free_skcipher(tfm); + return salg; + +out_free_salg: + kfree(salg); + salg = ERR_PTR(err); + goto out_put_tfm; +} +EXPORT_SYMBOL_GPL(simd_skcipher_create_compat_mb); + struct simd_skcipher_alg *simd_skcipher_create(const char *algname, const char *basename) { diff --git a/include/crypto/internal/simd.h b/include/crypto/internal/simd.h index 4295099..2c518a6 100644 --- a/include/crypto/internal/simd.h +++ b/include/crypto/internal/simd.h @@ -10,6 +10,9 @@ struct simd_skcipher_alg *simd_skcipher_create_compat(const char *algname, const char *drvname, const char *basename); +struct simd_skcipher_alg *simd_skcipher_create_compat_mb(const char *algname, + const char *drvname, + const char *basename); struct simd_skcipher_alg *simd_skcipher_create(const char *algname, const char *basename); void simd_skcipher_free(struct simd_skcipher_alg *alg); diff --git a/include/crypto/mcryptd.h b/include/crypto/mcryptd.h index 4a53c0d..5b17ce6 100644 --- a/include/crypto/mcryptd.h +++ b/include/crypto/mcryptd.h @@ -13,6 +13,28 @@ #include <linux/crypto.h> #include <linux/kernel.h> #include <crypto/hash.h> +#include <crypto/b128ops.h> +#include <crypto/internal/skcipher.h> +#include <crypto/internal/hash.h> + +struct mskcipherd_instance_ctx { + struct crypto_skcipher_spawn spawn; + struct mcryptd_queue *queue; +}; + +struct mcryptd_skcipher_ctx { + struct crypto_skcipher *child; + struct mcryptd_alg_state *alg_state; +}; + +struct mcryptd_skcipher { + struct crypto_skcipher base; +}; + +struct mcryptd_skcipher *mcryptd_alloc_skcipher(const char *alg_name, + u32 type, u32 mask); +struct crypto_skcipher *mcryptd_skcipher_child(struct mcryptd_skcipher *tfm); +void mcryptd_free_skcipher(struct mcryptd_skcipher *tfm); struct mcryptd_ahash { struct crypto_ahash base; @@ -62,6 +84,19 @@ struct mcryptd_hash_request_ctx { struct ahash_request areq; }; +struct mcryptd_skcipher_request_ctx { + struct list_head waiter; + crypto_completion_t complete; + struct mcryptd_tag tag; + struct skcipher_walk walk; + u8 flag; + int nbytes; + int error; + struct skcipher_request desc; + void *job; + u128 seq_iv; +}; + struct mcryptd_ahash *mcryptd_alloc_ahash(const char *alg_name, u32 type, u32 mask); struct crypto_ahash *mcryptd_ahash_child(struct mcryptd_ahash *tfm); -- 1.9.1
