linux-next/crypto/xts.c

368 lines
8.9 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/* XTS: as defined in IEEE1619/D16
* http://grouper.ieee.org/groups/1619/email/pdf00086.pdf
* (sector sizes which are not a multiple of 16 bytes are,
* however currently unsupported)
*
* Copyright (c) 2007 Rik Snel <rsnel@cube.dyndns.org>
*
* Based on ecb.c
* Copyright (c) 2006 Herbert Xu <herbert@gondor.apana.org.au>
*/
#include <crypto/internal/skcipher.h>
#include <crypto/scatterwalk.h>
#include <linux/err.h>
#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/scatterlist.h>
#include <linux/slab.h>
#include <crypto/xts.h>
#include <crypto/b128ops.h>
#include <crypto/gf128mul.h>
struct priv {
struct crypto_skcipher *child;
struct crypto_cipher *tweak;
};
struct xts_instance_ctx {
struct crypto_skcipher_spawn spawn;
char name[CRYPTO_MAX_ALG_NAME];
};
struct rctx {
le128 t;
struct skcipher_request subreq;
};
static int setkey(struct crypto_skcipher *parent, const u8 *key,
unsigned int keylen)
{
struct priv *ctx = crypto_skcipher_ctx(parent);
struct crypto_skcipher *child;
struct crypto_cipher *tweak;
int err;
err = xts_verify_key(parent, key, keylen);
if (err)
return err;
keylen /= 2;
/* we need two cipher instances: one to compute the initial 'tweak'
* by encrypting the IV (usually the 'plain' iv) and the other
* one to encrypt and decrypt the data */
/* tweak cipher, uses Key2 i.e. the second half of *key */
tweak = ctx->tweak;
crypto_cipher_clear_flags(tweak, CRYPTO_TFM_REQ_MASK);
crypto_cipher_set_flags(tweak, crypto_skcipher_get_flags(parent) &
CRYPTO_TFM_REQ_MASK);
err = crypto_cipher_setkey(tweak, key + keylen, keylen);
crypto_skcipher_set_flags(parent, crypto_cipher_get_flags(tweak) &
CRYPTO_TFM_RES_MASK);
if (err)
return err;
/* data cipher, uses Key1 i.e. the first half of *key */
child = ctx->child;
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;
}
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
/*
* We compute the tweak masks twice (both before and after the ECB encryption or
* decryption) to avoid having to allocate a temporary buffer and/or make
* mutliple calls to the 'ecb(..)' instance, which usually would be slower than
* just doing the gf128mul_x_ble() calls again.
*/
static int xor_tweak(struct skcipher_request *req, bool second_pass)
{
struct rctx *rctx = skcipher_request_ctx(req);
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
const int bs = XTS_BLOCK_SIZE;
struct skcipher_walk w;
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
le128 t = rctx->t;
int err;
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
if (second_pass) {
req = &rctx->subreq;
/* set to our TFM to enforce correct alignment: */
skcipher_request_set_tfm(req, tfm);
}
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
err = skcipher_walk_virt(&w, req, false);
while (w.nbytes) {
unsigned int avail = w.nbytes;
le128 *wsrc;
le128 *wdst;
wsrc = w.src.virt.addr;
wdst = w.dst.virt.addr;
do {
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
le128_xor(wdst++, &t, wsrc++);
gf128mul_x_ble(&t, &t);
} while ((avail -= bs) >= bs);
err = skcipher_walk_done(&w, avail);
}
return err;
}
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
static int xor_tweak_pre(struct skcipher_request *req)
{
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
return xor_tweak(req, false);
}
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
static int xor_tweak_post(struct skcipher_request *req)
{
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
return xor_tweak(req, true);
}
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
static void crypt_done(struct crypto_async_request *areq, int err)
{
struct skcipher_request *req = areq->data;
if (!err) {
struct rctx *rctx = skcipher_request_ctx(req);
rctx->subreq.base.flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
err = xor_tweak_post(req);
}
skcipher_request_complete(req, err);
}
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
static void init_crypt(struct skcipher_request *req)
{
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
struct priv *ctx = crypto_skcipher_ctx(crypto_skcipher_reqtfm(req));
struct rctx *rctx = skcipher_request_ctx(req);
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
struct skcipher_request *subreq = &rctx->subreq;
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
skcipher_request_set_tfm(subreq, ctx->child);
skcipher_request_set_callback(subreq, req->base.flags, crypt_done, req);
skcipher_request_set_crypt(subreq, req->dst, req->dst,
req->cryptlen, NULL);
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
/* calculate first value of T */
crypto_cipher_encrypt_one(ctx->tweak, (u8 *)&rctx->t, req->iv);
}
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
static int encrypt(struct skcipher_request *req)
{
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
struct rctx *rctx = skcipher_request_ctx(req);
struct skcipher_request *subreq = &rctx->subreq;
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
init_crypt(req);
return xor_tweak_pre(req) ?:
crypto_skcipher_encrypt(subreq) ?:
xor_tweak_post(req);
}
static int decrypt(struct skcipher_request *req)
{
crypto: xts - Drop use of auxiliary buffer Since commit acb9b159c784 ("crypto: gf128mul - define gf128mul_x_* in gf128mul.h"), the gf128mul_x_*() functions are very fast and therefore caching the computed XTS tweaks has only negligible advantage over computing them twice. In fact, since the current caching implementation limits the size of the calls to the child ecb(...) algorithm to PAGE_SIZE (usually 4096 B), it is often actually slower than the simple recomputing implementation. This patch simplifies the XTS template to recompute the XTS tweaks from scratch in the second pass and thus also removes the need to allocate a dynamic buffer using kmalloc(). As discussed at [1], the use of kmalloc causes deadlocks with dm-crypt. PERFORMANCE RESULTS I measured time to encrypt/decrypt a memory buffer of varying sizes with xts(ecb-aes-aesni) using a tool I wrote ([2]) and the results suggest that after this patch the performance is either better or comparable for both small and large buffers. Note that there is a lot of noise in the measurements, but the overall difference is easy to see. Old code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 331 328 xts(aes) 384 64 332 333 xts(aes) 512 64 338 348 xts(aes) 256 512 889 920 xts(aes) 384 512 1019 993 xts(aes) 512 512 1032 990 xts(aes) 256 4096 2152 2292 xts(aes) 384 4096 2453 2597 xts(aes) 512 4096 3041 2641 xts(aes) 256 16384 9443 8027 xts(aes) 384 16384 8536 8925 xts(aes) 512 16384 9232 9417 xts(aes) 256 32768 16383 14897 xts(aes) 384 32768 17527 16102 xts(aes) 512 32768 18483 17322 New code: ALGORITHM KEY (b) DATA (B) TIME ENC (ns) TIME DEC (ns) xts(aes) 256 64 328 324 xts(aes) 384 64 324 319 xts(aes) 512 64 320 322 xts(aes) 256 512 476 473 xts(aes) 384 512 509 492 xts(aes) 512 512 531 514 xts(aes) 256 4096 2132 1829 xts(aes) 384 4096 2357 2055 xts(aes) 512 4096 2178 2027 xts(aes) 256 16384 6920 6983 xts(aes) 384 16384 8597 7505 xts(aes) 512 16384 7841 8164 xts(aes) 256 32768 13468 12307 xts(aes) 384 32768 14808 13402 xts(aes) 512 32768 15753 14636 [1] https://lkml.org/lkml/2018/8/23/1315 [2] https://gitlab.com/omos/linux-crypto-bench Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2018-09-11 07:40:08 +00:00
struct rctx *rctx = skcipher_request_ctx(req);
struct skcipher_request *subreq = &rctx->subreq;
init_crypt(req);
return xor_tweak_pre(req) ?:
crypto_skcipher_decrypt(subreq) ?:
xor_tweak_post(req);
}
static int init_tfm(struct crypto_skcipher *tfm)
{
struct skcipher_instance *inst = skcipher_alg_instance(tfm);
struct xts_instance_ctx *ictx = skcipher_instance_ctx(inst);
struct priv *ctx = crypto_skcipher_ctx(tfm);
struct crypto_skcipher *child;
struct crypto_cipher *tweak;
child = crypto_spawn_skcipher(&ictx->spawn);
if (IS_ERR(child))
return PTR_ERR(child);
ctx->child = child;
tweak = crypto_alloc_cipher(ictx->name, 0, 0);
if (IS_ERR(tweak)) {
crypto_free_skcipher(ctx->child);
return PTR_ERR(tweak);
}
ctx->tweak = tweak;
crypto_skcipher_set_reqsize(tfm, crypto_skcipher_reqsize(child) +
sizeof(struct rctx));
return 0;
}
static void exit_tfm(struct crypto_skcipher *tfm)
{
struct priv *ctx = crypto_skcipher_ctx(tfm);
crypto_free_skcipher(ctx->child);
crypto_free_cipher(ctx->tweak);
}
static void free(struct skcipher_instance *inst)
{
crypto_drop_skcipher(skcipher_instance_ctx(inst));
kfree(inst);
}
static int create(struct crypto_template *tmpl, struct rtattr **tb)
{
struct skcipher_instance *inst;
struct crypto_attr_type *algt;
struct xts_instance_ctx *ctx;
struct skcipher_alg *alg;
const char *cipher_name;
u32 mask;
int err;
algt = crypto_get_attr_type(tb);
if (IS_ERR(algt))
return PTR_ERR(algt);
if ((algt->type ^ CRYPTO_ALG_TYPE_SKCIPHER) & algt->mask)
return -EINVAL;
cipher_name = crypto_attr_alg_name(tb[1]);
if (IS_ERR(cipher_name))
return PTR_ERR(cipher_name);
inst = kzalloc(sizeof(*inst) + sizeof(*ctx), GFP_KERNEL);
if (!inst)
return -ENOMEM;
ctx = skcipher_instance_ctx(inst);
crypto_set_skcipher_spawn(&ctx->spawn, skcipher_crypto_instance(inst));
mask = crypto_requires_off(algt->type, algt->mask,
CRYPTO_ALG_NEED_FALLBACK |
CRYPTO_ALG_ASYNC);
err = crypto_grab_skcipher(&ctx->spawn, cipher_name, 0, mask);
if (err == -ENOENT) {
err = -ENAMETOOLONG;
if (snprintf(ctx->name, CRYPTO_MAX_ALG_NAME, "ecb(%s)",
cipher_name) >= CRYPTO_MAX_ALG_NAME)
goto err_free_inst;
err = crypto_grab_skcipher(&ctx->spawn, ctx->name, 0, mask);
}
if (err)
goto err_free_inst;
alg = crypto_skcipher_spawn_alg(&ctx->spawn);
err = -EINVAL;
if (alg->base.cra_blocksize != XTS_BLOCK_SIZE)
goto err_drop_spawn;
if (crypto_skcipher_alg_ivsize(alg))
goto err_drop_spawn;
err = crypto_inst_setname(skcipher_crypto_instance(inst), "xts",
&alg->base);
if (err)
goto err_drop_spawn;
err = -EINVAL;
cipher_name = alg->base.cra_name;
/* Alas we screwed up the naming so we have to mangle the
* cipher name.
*/
if (!strncmp(cipher_name, "ecb(", 4)) {
unsigned len;
len = strlcpy(ctx->name, cipher_name + 4, sizeof(ctx->name));
if (len < 2 || len >= sizeof(ctx->name))
goto err_drop_spawn;
if (ctx->name[len - 1] != ')')
goto err_drop_spawn;
ctx->name[len - 1] = 0;
if (snprintf(inst->alg.base.cra_name, CRYPTO_MAX_ALG_NAME,
"xts(%s)", ctx->name) >= CRYPTO_MAX_ALG_NAME) {
err = -ENAMETOOLONG;
goto err_drop_spawn;
}
} else
goto err_drop_spawn;
inst->alg.base.cra_flags = alg->base.cra_flags & CRYPTO_ALG_ASYNC;
inst->alg.base.cra_priority = alg->base.cra_priority;
inst->alg.base.cra_blocksize = XTS_BLOCK_SIZE;
inst->alg.base.cra_alignmask = alg->base.cra_alignmask |
(__alignof__(u64) - 1);
inst->alg.ivsize = XTS_BLOCK_SIZE;
inst->alg.min_keysize = crypto_skcipher_alg_min_keysize(alg) * 2;
inst->alg.max_keysize = crypto_skcipher_alg_max_keysize(alg) * 2;
inst->alg.base.cra_ctxsize = sizeof(struct priv);
inst->alg.init = init_tfm;
inst->alg.exit = exit_tfm;
inst->alg.setkey = setkey;
inst->alg.encrypt = encrypt;
inst->alg.decrypt = decrypt;
inst->free = free;
err = skcipher_register_instance(tmpl, inst);
if (err)
goto err_drop_spawn;
out:
return err;
err_drop_spawn:
crypto_drop_skcipher(&ctx->spawn);
err_free_inst:
kfree(inst);
goto out;
}
static struct crypto_template crypto_tmpl = {
.name = "xts",
.create = create,
.module = THIS_MODULE,
};
static int __init crypto_module_init(void)
{
return crypto_register_template(&crypto_tmpl);
}
static void __exit crypto_module_exit(void)
{
crypto_unregister_template(&crypto_tmpl);
}
subsys_initcall(crypto_module_init);
module_exit(crypto_module_exit);
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("XTS block cipher mode");
MODULE_ALIAS_CRYPTO("xts");