crypto: Fix next IV issue for CTS template
PATCH 1/1 crypto: Fix next IV issue for CTS template
From: Libo Wang <Libo [email protected]>
CTS template assumes underlying CBC algorithm will carry out next IV for further process.But some implementations of CBC algorithm in kernel break this assumption, for example, some hardware crypto drivers ignore next IV for performance consider, inthis case, tcry cts(cbc(aes)) test case will fail. This patch is trying to fix it by getting next IV information ready before last two blocks processed.
--- crypto/cts.c | 29 +++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)
crypto/cts.c +25 -4
@@ -154,6 +154,7 @@ static int crypto_cts_encrypt(struct skcipher_request *req)unsigned int nbytes = req->cryptlen;int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset; +int ret =0 ;skcipher_request_set_tfm (subreq, ctx->child);
@@ -174,8 +175,17 @@ static int crypto_cts_encrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv); -return crypto_skcipher_encrypt (subreq) ?: -cts_cbc_encrypt (req); +/* process CBC blocks */ + ret =crypto_skcipher_encrypt (subreq); +/* process last two blocks */ +if (!ret) { +/* Get IVn-1 back */ +scatterwalk_map_and_copy (req->iv, req->dst, (offset - bsize), bsize,0 ); +/* Continue last two blocks */ +return cts_cbc_encrypt (req); + } + +return ret; }static int cts_cbc_decrypt (struct skcipher_request *req)
@@ -248,6 +258,8 @@ static int crypto_cts_decrypt(struct skcipher_request *req)int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset;u8 *space; +int ret =0 ; +u8 iv_next[bsize];skcipher_request_set_tfm (subreq, ctx->child);
@@ -277,8 +289,17 @@ static int crypto_cts_decrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv); -return crypto_skcipher_decrypt (subreq) ?: -cts_cbc_decrypt (req); +/* process last two blocks */ +scatterwalk_map_and_copy (iv_next, req->src, (offset - bsize), bsize,0 ); + ret =crypto_skcipher_decrypt (subreq); +if (!ret) { +/* Set Next IV */ + subreq->iv = iv_next; +/* process last two blocks */ +return cts_cbc_decrypt (req); + } + +return ret; }static int crypto_cts_init_tfm (struct crypto_skcipher *tfm)
Hello Libo,
Quoted 11 lines · On 17 February 2017 at 03:47, <[email protected]> wrote:
From: Libo Wang <Libo [email protected]>
CTS template assumes underlying CBC algorithm will carry out next IV for further process.But some implementations of CBC algorithm in kernel break this assumption, for example, some hardware crypto drivers ignore next IV for performance consider, inthis case, tcry cts(cbc(aes)) test case will fail. This patch is trying to fix it by getting next IV information ready before last two blocks processed.
Signed-off-by Libo Wang <[email protected]>Signed-off-by Dennis Chen <[email protected]>
Which algorithms in particular break this assumption? I recently fixed some ARM accelerated software drivers for this reason. If there are others, we should fix those rather than try to fix it in the CTS driver.
Quoted 17 lines
--- crypto/cts.c | 29 +++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)crypto/cts.c +1
-0@@ -154,6 +154,7 @@ static int crypto_cts_encrypt(struct skcipher_request *req)unsigned int nbytes = req->cryptlen;int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset; +int ret =0 ;skcipher_request_set_tfm (subreq, ctx->child);
crypto/cts.c static int crypto_cts_encrypt(struct skcipher_request *req) · line 181
Quoted 9 lines
crypto/cts.c +4
-2@@ -174,8 +175,17 @@ static int crypto_cts_encrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv); -return crypto_skcipher_encrypt (subreq) ?: -cts_cbc_encrypt (req); +/* process CBC blocks */ + ret =crypto_skcipher_encrypt (subreq); +/* process last two blocks */ +if (!ret) {
What happens if an async driver returns -EINPROGRESS here?
Quoted 42 lines
crypto/cts.c +20
-3+/* Get IVn-1 back */ +scatterwalk_map_and_copy (req->iv, req->dst, (offset - bsize), bsize,0 ); +/* Continue last two blocks */ +return cts_cbc_encrypt (req); + } + +return ret; }static int cts_cbc_decrypt (struct skcipher_request *req)@@ -248,6 +258,8 @@ static int crypto_cts_decrypt(struct skcipher_request *req)int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset;u8 *space; +int ret =0 ; +u8 iv_next[bsize];skcipher_request_set_tfm (subreq, ctx->child);@@ -277,8 +289,17 @@ static int crypto_cts_decrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv); -return crypto_skcipher_decrypt (subreq) ?: -cts_cbc_decrypt (req); +/* process last two blocks */ +scatterwalk_map_and_copy (iv_next, req->src, (offset - bsize), bsize,0 ); + ret =crypto_skcipher_decrypt (subreq); +if (!ret) { +/* Set Next IV */ + subreq->iv = iv_next; +/* process last two blocks */ +return cts_cbc_decrypt (req); + } + +return ret; }static int crypto_cts_init_tfm (struct crypto_skcipher *tfm) --1.8.3.1
Quoted 1 line
IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
Please configure your email client so it doesn't spit out these.
Hello Ard,
Morning!
Quoted 12 lines · Ard Biesheuvel
Hello Libo,
Quoted 9 lines · On 17 February 2017 at 03:47, <[email protected]> wrote:
From: Libo Wang <Libo [email protected]>
CTS template assumes underlying CBC algorithm will carry out next IV for further process.But some implementations of CBC algorithm in kernel break this assumption, for example, some hardware crypto drivers ignore next IV for performance consider, inthis case, tcry cts(cbc(aes)) test case will fail. This patch is trying to fix it by getting next IV information ready before last two blocks processed.
Quoted 8 lines
Quoted 2 lines
Signed-off-by Libo Wang <[email protected]>Signed-off-by Dennis Chen <[email protected]>Which algorithms in particular break this assumption? I recently fixed some ARM accelerated software drivers for this reason. If there are others, we should fix those rather than try to fix it in the CTS driver.
There're some ARM HW accelerated drivers which are not upstream yet(IP license) breaks this assumption. The current CTS template requires the underlying CBC algorithm always fill the next IV, but in some cases like CBC standalone mode, it doesn't need to fill the next IV, so apparently it will induce performance degrade from the point of drivers, that's why we fix it in the CTS template,except that it will also mitigate the potential defect of other HW-based CBC drivers.
Quoted 21 lines
Quoted 21 lines
--- crypto/cts.c | 29 +++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-)crypto/cts.c +1
-0@@ -154,6 +154,7 @@ static int crypto_cts_encrypt(struct skcipher_request *req)unsigned int nbytes = req->cryptlen;int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset; +int ret =0 ;skcipher_request_set_tfm (subreq, ctx->child);@@ -174,8 +175,17 @@ static int crypto_cts_encrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv);
Quoted 8 lines
crypto/cts.c static int crypto_cts_encrypt(struct skcipher_request *req) · line 181
Quoted 6 lines
crypto/cts.c +4
-2-return crypto_skcipher_encrypt (subreq) ?: -cts_cbc_encrypt (req); +/* process CBC blocks */ + ret =crypto_skcipher_encrypt (subreq); +/* process last two blocks */ +if (!ret) {What happens if an async driver returns
-EINPROGRESShere?
For this case, the CTS will return -EINPROGRESS directly.
Thanks,
Dennis
Quoted 43 lines
Quoted 43 lines
crypto/cts.c +20
-3+/* Get IVn-1 back */ +scatterwalk_map_and_copy (req->iv, req->dst, (offset - bsize), bsize,0 ); +/* Continue last two blocks */ +return cts_cbc_encrypt (req); + } + +return ret; }static int cts_cbc_decrypt (struct skcipher_request *req)@@ -248,6 +258,8 @@ static int crypto_cts_decrypt(struct skcipher_request *req)int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset;u8 *space; +int ret =0 ; +u8 iv_next[bsize];skcipher_request_set_tfm (subreq, ctx->child);@@ -277,8 +289,17 @@ static int crypto_cts_decrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv); -return crypto_skcipher_decrypt (subreq) ?: -cts_cbc_decrypt (req); +/* process last two blocks */ +scatterwalk_map_and_copy (iv_next, req->src, (offset - bsize), bsize,0 ); + ret =crypto_skcipher_decrypt (subreq); +if (!ret) { +/* Set Next IV */ + subreq->iv = iv_next; +/* process last two blocks */ +return cts_cbc_decrypt (req); + } + +return ret; }static int crypto_cts_init_tfm (struct crypto_skcipher *tfm) --1.8.3.1
Quoted 26 lines · Dennis Chen
Hello Ard,
Morning!Quoted 13 lines · Ard Biesheuvel
Hello Libo,
Quoted 10 lines · On 17 February 2017 at 03:47, <[email protected]> wrote:
From: Libo Wang <Libo [email protected]>
CTS template assumes underlying CBC algorithm will carry out next IV for further process.But some implementations of CBC algorithm in kernel break this assumption, for example, some hardware crypto drivers ignore next IV for performance consider, inthis case, tcry cts(cbc(aes)) test case will fail. This patch is trying to fix it by getting next IV information ready before last two blocks processed.
Quoted 8 lines
Quoted 2 lines
Signed-off-by Libo Wang <[email protected]>Signed-off-by Dennis Chen <[email protected]>Which algorithms in particular break this assumption? I recently fixed some ARM accelerated software drivers for this reason. If there are others, we should fix those rather than try to fix it in the CTS driver.
Quoted 6 lines
There're some ARM HW accelerated drivers which are not upstream yet(IP license) breaks this assumption. The current CTS template requires the underlying CBC algorithm always fill the next IV, but in some cases like CBC standalone mode, it doesn't need to fill the next IV, so apparently it will induce performance degrade from the point of drivers, that's why we fix it in the CTS template,except that it will also mitigate the potential defect of other HW-based CBC drivers.
You should fix this in the driver, not in the cts code.
Quoted 29 lines
Quoted 21 lines
Quoted 21 lines
---
crypto/cts.c| 29 +++++++++++++++++++++++++----1 file changed, 25 insertions(+), 4 deletions(-)crypto/cts.c +1
-0@@ -154,6 +154,7 @@ static int crypto_cts_encrypt(struct skcipher_request *req)unsigned int nbytes = req->cryptlen;int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset; +int ret =0 ;skcipher_request_set_tfm (subreq, ctx->child);@@ -174,8 +175,17 @@ static int crypto_cts_encrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv);Quoted 8 lines
crypto/cts.c static int crypto_cts_encrypt(struct skcipher_request *req) · line 181
Quoted 6 lines
crypto/cts.c +4
-2-return crypto_skcipher_encrypt (subreq) ?: -cts_cbc_encrypt (req); +/* process CBC blocks */ + ret =crypto_skcipher_encrypt (subreq); +/* process last two blocks */ +if (!ret) {What happens if an async driver returns
-EINPROGRESShere?
Quoted 2 lines
For this case, the CTS will return
-EINPROGRESSdirectly.
I see. But how do you handle the missing out IV when the async call completes?
Quoted 43 lines
Quoted 43 lines
Quoted 43 lines
crypto/cts.c +20
-3+/* Get IVn-1 back */ +scatterwalk_map_and_copy (req->iv, req->dst, (offset - bsize), bsize,0 ); +/* Continue last two blocks */ +return cts_cbc_encrypt (req); + } + +return ret; }static int cts_cbc_decrypt (struct skcipher_request *req)@@ -248,6 +258,8 @@ static int crypto_cts_decrypt(struct skcipher_request *req)int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset;u8 *space; +int ret =0 ; +u8 iv_next[bsize];skcipher_request_set_tfm (subreq, ctx->child);@@ -277,8 +289,17 @@ static int crypto_cts_decrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv); -return crypto_skcipher_decrypt (subreq) ?: -cts_cbc_decrypt (req); +/* process last two blocks */ +scatterwalk_map_and_copy (iv_next, req->src, (offset - bsize), bsize,0 ); + ret =crypto_skcipher_decrypt (subreq); +if (!ret) { +/* Set Next IV */ + subreq->iv = iv_next; +/* process last two blocks */ +return cts_cbc_decrypt (req); + } + +return ret; }static int crypto_cts_init_tfm (struct crypto_skcipher *tfm) --1.8.3.1
Quoted 26 lines · Ard Biesheuvel
Quoted 26 lines · Dennis Chen
Hello Ard,
Morning!Quoted 21 lines · Ard Biesheuvel
Hello Libo,
Quoted 10 lines · On 17 February 2017 at 03:47, <[email protected]> wrote:
From: Libo Wang <Libo [email protected]>
CTS template assumes underlying CBC algorithm will carry out next IV for further process.But some implementations of CBC algorithm in kernel break this assumption, for example, some hardware crypto drivers ignore next IV for performance consider, inthis case, tcry cts(cbc(aes)) test case will fail. This patch is trying to fix it by getting next IV information ready before last two blocks processed.
Quoted 2 lines
Signed-off-by Libo Wang <[email protected]>Signed-off-by Dennis Chen <[email protected]>Which algorithms in particular break this assumption? I recently fixed some ARM accelerated software drivers for this reason. If there are others, we should fix those rather than try to fix it in the CTS driver.
Quoted 9 lines
Quoted 6 lines
There're some ARM HW accelerated drivers which are not upstream yet(IP license) breaks this assumption. The current CTS template requires the underlying CBC algorithm always fill the next IV, but in some cases like CBC standalone mode, it doesn't need to fill the next IV, so apparently it will induce performance degrade from the point of drivers, that's why we fix it in the CTS template,except that it will also mitigate the potential defect of other HW-based CBC drivers.
You should fix this in the driver, not in the cts code.
Ah, we're open for the change, but is it better to give the reason to fix it in the driver instead of in cts code?
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
Quoted 18 lines
---
crypto/cts.c| 29 +++++++++++++++++++++++++----1 file changed, 25 insertions(+), 4 deletions(-)crypto/cts.c +1
-0@@ -154,6 +154,7 @@ static int crypto_cts_encrypt(struct skcipher_request *req)unsigned int nbytes = req->cryptlen;int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset; +int ret =0 ;skcipher_request_set_tfm (subreq, ctx->child);crypto/cts.c static int crypto_cts_encrypt(struct skcipher_request *req) · line 181
Quoted 9 lines
crypto/cts.c +4
-2@@ -174,8 +175,17 @@ static int crypto_cts_encrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv); -return crypto_skcipher_encrypt (subreq) ?: -cts_cbc_encrypt (req); +/* process CBC blocks */ + ret =crypto_skcipher_encrypt (subreq); +/* process last two blocks */ +if (!ret) {What happens if an async driver returns
-EINPROGRESShere?
Quoted 5 lines
Quoted 2 lines
For this case, the CTS will return
-EINPROGRESSdirectly.I see. But how do you handle the missing out IV when the async call completes?
Not quite understand here, is there some logic inconsistent of the change here comparing the original one to handle this case? We think it will have the same result if the fix is in driver layer.
Thanks,
Dennis
Quoted 44 lines
Quoted 44 lines
Quoted 44 lines
Quoted 44 lines
crypto/cts.c +20
-3+/* Get IVn-1 back */ +scatterwalk_map_and_copy (req->iv, req->dst, (offset - bsize), bsize,0 ); +/* Continue last two blocks */ +return cts_cbc_encrypt (req); + } + +return ret; }static int cts_cbc_decrypt (struct skcipher_request *req)@@ -248,6 +258,8 @@ static int crypto_cts_decrypt(struct skcipher_request *req)int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset;u8 *space; +int ret =0 ; +u8 iv_next[bsize];skcipher_request_set_tfm (subreq, ctx->child);@@ -277,8 +289,17 @@ static int crypto_cts_decrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv); -return crypto_skcipher_decrypt (subreq) ?: -cts_cbc_decrypt (req); +/* process last two blocks */ +scatterwalk_map_and_copy (iv_next, req->src, (offset - bsize), bsize,0 ); + ret =crypto_skcipher_decrypt (subreq); +if (!ret) { +/* Set Next IV */ + subreq->iv = iv_next; +/* process last two blocks */ +return cts_cbc_decrypt (req); + } + +return ret; }static int crypto_cts_init_tfm (struct crypto_skcipher *tfm) --1.8.3.1
Quoted 37 lines · Dennis Chen
Quoted 35 lines · Ard Biesheuvel
Quoted 32 lines · Dennis Chen
Hello Ard,
Morning!Quoted 13 lines · Ard Biesheuvel
Hello Libo,
Quoted 10 lines · On 17 February 2017 at 03:47, <[email protected]> wrote:
From: Libo Wang <Libo [email protected]>
CTS template assumes underlying CBC algorithm will carry out next IV for further process.But some implementations of CBC algorithm in kernel break this assumption, for example, some hardware crypto drivers ignore next IV for performance consider, inthis case, tcry cts(cbc(aes)) test case will fail. This patch is trying to fix it by getting next IV information ready before last two blocks processed.
Quoted 8 lines
Quoted 2 lines
Signed-off-by Libo Wang <[email protected]>Signed-off-by Dennis Chen <[email protected]>Which algorithms in particular break this assumption? I recently fixed some ARM accelerated software drivers for this reason. If there are others, we should fix those rather than try to fix it in the CTS driver.
There're some ARM HW accelerated drivers which are not upstream yet(IP license) breaks this assumption. The current CTS template requires the underlying CBC algorithm always fill the next IV, but in some cases like CBC standalone mode, it doesn't need to fill the next IV, so apparently it will induce performance degrade from the point of drivers, that's why we fix it in the CTS template,except that it will also mitigate the potential defect of other HW-based CBC drivers.
Quoted 4 lines
Quoted 2 lines
You should fix this in the driver, not in the cts code.
Ah, we're open for the change, but is it better to give the reason to fix it in the driver instead of in cts code?
Because it is the h/w driver that violates the api.
Please search the list: I discussed this with Herbert very recently
Quoted 83 lines
Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
Quoted 18 lines
---
crypto/cts.c| 29 +++++++++++++++++++++++++----1 file changed, 25 insertions(+), 4 deletions(-)crypto/cts.c +1
-0@@ -154,6 +154,7 @@ static int crypto_cts_encrypt(struct skcipher_request *req)unsigned int nbytes = req->cryptlen;int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset; +int ret =0 ;skcipher_request_set_tfm (subreq, ctx->child);crypto/cts.c static int crypto_cts_encrypt(struct skcipher_request *req) · line 181
Quoted 9 lines
crypto/cts.c +4
-2@@ -174,8 +175,17 @@ static int crypto_cts_encrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv); -return crypto_skcipher_encrypt (subreq) ?: -cts_cbc_encrypt (req); +/* process CBC blocks */ + ret =crypto_skcipher_encrypt (subreq); +/* process last two blocks */ +if (!ret) {What happens if an async driver returns
-EINPROGRESShere?Quoted 5 lines
Quoted 2 lines
For this case, the CTS will return
-EINPROGRESSdirectly.I see. But how do you handle the missing out IV when the async call completes?
Not quite understand here, is there some logic inconsistent of the change here comparing the original one to handle this case? We think it will have the same result if the fix is in driver layer.
Thanks,
DennisQuoted 44 lines
Quoted 44 lines
Quoted 44 lines
Quoted 44 lines
crypto/cts.c +20
-3+/* Get IVn-1 back */ +scatterwalk_map_and_copy (req->iv, req->dst, (offset - bsize), bsize,0 ); +/* Continue last two blocks */ +return cts_cbc_encrypt (req); + } + +return ret; }static int cts_cbc_decrypt (struct skcipher_request *req)@@ -248,6 +258,8 @@ static int crypto_cts_decrypt(struct skcipher_request *req)int cbc_blocks = (nbytes + bsize -1 ) / bsize -1 ;unsigned int offset;u8 *space; +int ret =0 ; +u8 iv_next[bsize];skcipher_request_set_tfm (subreq, ctx->child);@@ -277,8 +289,17 @@ static int crypto_cts_decrypt(struct skcipher_request *req)skcipher_request_set_crypt (subreq, req->src, req->dst, offset, req->iv); -return crypto_skcipher_decrypt (subreq) ?: -cts_cbc_decrypt (req); +/* process last two blocks */ +scatterwalk_map_and_copy (iv_next, req->src, (offset - bsize), bsize,0 ); + ret =crypto_skcipher_decrypt (subreq); +if (!ret) { +/* Set Next IV */ + subreq->iv = iv_next; +/* process last two blocks */ +return cts_cbc_decrypt (req); + } + +return ret; }static int crypto_cts_init_tfm (struct crypto_skcipher *tfm) --1.8.3.1