| Message ID | 15957213.Wu7Np8vQse@tachyon.chronox.de (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Herbert Xu |
| Headers | show |
On Tue, Dec 30, 2014 at 10:16:03PM +0100, Stephan Mueller wrote: > The AEAD decryption operation requires the authentication tag to be > present as part of the cipher text buffer. The added check verifies that > the caller provides a cipher text buffer with at least the > authentication tag. > > As the cipher text is provided as a scatterlist, loop through the > scatterlist until we know we have sufficient cipher text bytes to invoke > the AEAD decryption operation. There is no need to vet the scatterlist. For kernel users it's up to them to ensure that the length is valid with respect to the sg list. For the user-space interface obviously you should ensure that a valid length is supplied. Cheers,
Am Montag, 5. Januar 2015, 21:19:01 schrieb Herbert Xu: Hi Herbert, > On Tue, Dec 30, 2014 at 10:16:03PM +0100, Stephan Mueller wrote: > > The AEAD decryption operation requires the authentication tag to be > > present as part of the cipher text buffer. The added check verifies that > > the caller provides a cipher text buffer with at least the > > authentication tag. > > > > As the cipher text is provided as a scatterlist, loop through the > > scatterlist until we know we have sufficient cipher text bytes to invoke > > the AEAD decryption operation. > > There is no need to vet the scatterlist. For kernel users it's > up to them to ensure that the length is valid with respect to the > sg list. > > For the user-space interface obviously you should ensure that a > valid length is supplied. Then it seems I misunderstood you in your last comments. I thought the check that the authentication tag is present should be moved to the crypto API so that all users benefit from it. As the authentication tag is "inline" in the ciphertext data stream, the only way I see to check for its presence is to verify that sufficient ciphertext is provided. If the AEAD user space interface should perform the check, then the code drop provided with patch v6 would already cover that. Could you please help me understand where that check should be and what should be done at the crypto API level? Thanks
On Mon, Jan 05, 2015 at 11:26:48AM +0100, Stephan Mueller wrote: > > Could you please help me understand where that check should be and what should > be done at the crypto API level? All I was asking for is a length check like the one at the top of crypto_authenc_decypt. Cheers,
Am Montag, 5. Januar 2015, 21:39:57 schrieb Herbert Xu: Hi Herbert, > On Mon, Jan 05, 2015 at 11:26:48AM +0100, Stephan Mueller wrote: > > Could you please help me understand where that check should be and what > > should be done at the crypto API level? > > All I was asking for is a length check like the one at the top > of crypto_authenc_decypt. Could you please check: I am unable to find the function. Note, there is no size parameter provided in crypto_aead_decrypt. To obtain the size, I would need to walk the scatterlists, no? > > Cheers,
On Mon, Jan 05, 2015 at 11:53:38AM +0100, Stephan Mueller wrote: > > Could you please check: I am unable to find the function. It's in crypto/authenc.c. > Note, there is no size parameter provided in crypto_aead_decrypt. To obtain > the size, I would need to walk the scatterlists, no? The cryptlen is supposed to contain the entire input apart from the AD, including the authentication tag if one is present. Cheers,
Am Montag, 5. Januar 2015, 21:55:40 schrieb Herbert Xu: Hi Herbert, > On Mon, Jan 05, 2015 at 11:53:38AM +0100, Stephan Mueller wrote: > > Could you please check: I am unable to find the function. > > It's in crypto/authenc.c. > > > Note, there is no size parameter provided in crypto_aead_decrypt. To > > obtain > > the size, I would need to walk the scatterlists, no? > > The cryptlen is supposed to contain the entire input apart from the > AD, including the authentication tag if one is present. Thanks a lot for the pointer, I overlooked the cryptlen member variable. I will update the patch to replace the scatterlist walk with such small check.
diff --git a/include/linux/crypto.h b/include/linux/crypto.h index 9c8776d..badc53b 100644 --- a/include/linux/crypto.h +++ b/include/linux/crypto.h @@ -24,6 +24,7 @@ #include <linux/slab.h> #include <linux/string.h> #include <linux/uaccess.h> +#include <linux/scatterlist.h> /* * Autoloaded crypto modules should only use a prefixed name to avoid allowing @@ -1412,7 +1413,18 @@ static inline int crypto_aead_encrypt(struct aead_request *req) */ static inline int crypto_aead_decrypt(struct aead_request *req) { - return crypto_aead_crt(crypto_aead_reqtfm(req))->decrypt(req); + struct scatterlist *sg = req->src; + unsigned int size = 0; + + /* ensure that the ciphertext at least contains the auth tag */ + while (sg) { + size += sg->length; + if (crypto_aead_authsize(crypto_aead_reqtfm(req)) <= size) + return crypto_aead_crt( + crypto_aead_reqtfm(req))->decrypt(req); + sg = sg_next(sg); + } + return -EINVAL; } /**
The AEAD decryption operation requires the authentication tag to be present as part of the cipher text buffer. The added check verifies that the caller provides a cipher text buffer with at least the authentication tag. As the cipher text is provided as a scatterlist, loop through the scatterlist until we know we have sufficient cipher text bytes to invoke the AEAD decryption operation. As the authentication tag is typically smaller or equal to one block of the cipher (except for authenc operations), it is expected that the while loop iterating through the scatterlist is traversed not more than once. Thus, the speed penalty should be marginal. Signed-off-by: Stephan Mueller <smueller@chronox.de> --- include/linux/crypto.h | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)