This patch adds the SIV-CMAC algorithm to nettle (an update of the previous attempt). It is an atypical cipher which fits into the encrypt_message interface.
regards, Nikos
Nikos Mavrogiannopoulos nmav@redhat.com writes:
This patch adds the SIV-CMAC algorithm to nettle (an update of the previous attempt). It is an atypical cipher which fits into the encrypt_message interface.
Thanks. Some comments below:
--- a/nettle-types.h +++ b/nettle-types.h @@ -78,6 +78,21 @@ typedef void *nettle_realloc_func(void *ctx, void *p, size_t length); /* Ciphers */ typedef void nettle_set_key_func(void *ctx, const uint8_t *key);
+/* AEAD ciphers */ +typedef void +nettle_encrypt_message(void *ctx,
size_t nlength, const uint8_t *nonce,size_t alength, const uint8_t *adata,size_t tlength,size_t clength, uint8_t *dst, const uint8_t *src);+typedef int +nettle_decrypt_message(void *ctx,
size_t nlength, const uint8_t *nonce,size_t alength, const uint8_t *adata,size_t tlength,size_t mlength, uint8_t *dst, const uint8_t *src);
In this patch, these types used in tests only. But I imagine you'd like something in nettle-meta to represent a message-oriented aead?
+@acronym{SIV-CMAC} mode is a combination of counter mode with message +authentication based on @acronym{CMAC}. Unlike other counter @acronym{AEAD} +modes, it provides protection against accidental nonce misuse, making it +a good choice for stateless-servers that cannot ensure nonce uniqueness.
Some detail on the nonce-reuse would be helpful. As I understood RFC 5297, the nonce used in SIV is only to make the ciphertexts of otherwise identical messages look different to the attacker.
--- /dev/null +++ b/siv-aes128-cmac.c @@ -0,0 +1,79 @@
[...]
+void +siv_aes128_cmac_encrypt_message(struct siv_aes128_cmac_ctx *ctx,
size_t nlength, const uint8_t *nonce,size_t alength, const uint8_t *adata,size_t tlength,size_t slength, uint8_t *dst, const uint8_t *src)+{
- assert(tlength == SIV_DIGEST_SIZE);
The tlength argument doesn't look that useful, do we need it?. Also, I think we agreed that the message length argument should be the size of the *destination* area, i.e., ciphertext size for encrypt_message (see ccm_*_encrypt_message):
--- /dev/null +++ b/siv-cmac.c @@ -0,0 +1,194 @@
+/* This is a common structure for AES-128 and AES-256 thus
- for the cipher part we simply pad to the maximum structure
- size plus 16 bytes to account for any alignment difference in
- the original structures */
+struct cmac128_syn {
- struct cmac128_ctx ctx;
- struct {
- uint8_t pad[NETTLE_MAX_CIPHER16_CONTEXT_SIZE+16];
- } cipher;
+};
I don't think this guarantees any alignment. You would either need soemthing like
struct cipher_storage { union { uint64_t u64; char c[NETTLE_MAX_CIPHER16_CONTEXT_SIZE]; } storage; }; ... CMAC128_CTX(cipher_storage) ctx;
or use TMP_ALLOC_ALIGN (but in the latter case, you can't use the CMAC128_* macros). Or let caller pass in the cipher context (see last comment in this mail)
+static +void _siv_s2v(const struct nettle_cipher *nc,
const uint8_t *s2vk, size_t alength, const uint8_t *adata,size_t nlength, const uint8_t *nonce,size_t plength, const uint8_t *pdata,uint8_t *v)+{
- struct cmac128_syn ctx;
- union nettle_block16 D, S, T;
- const uint8_t const_one[16] = {
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01
- };
- const uint8_t const_zero[16] = {
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
- };
Use static const.
- assert(nc->context_size <= NETTLE_MAX_CIPHER16_CONTEXT_SIZE);
- /* ensure we have enough size of context plus any padding size */
- CMAC128_SET_KEY(&ctx, nc->set_encrypt_key, nc->encrypt, s2vk);
- if (nlength == 0 && alength == 0) {
- CMAC128_UPDATE(&ctx, nc->encrypt, 16, const_one);
- CMAC128_DIGEST(&ctx, nc->encrypt, 16, v);
- return;
- }
Shouldn't the plaintext, plength, pdata, still be processed in this case?
In this function, you treat empty associated data or nonce as those elements missing in the input vector to S2V. E.g., if both adata and nonce are empty, the input vector is { plaintext }, one single element. But it could also be { "", "", plaintext }, with three elements, the first two being empty strings.
To me, this sounds like a likely source of interop problems. Since RFC 5297 is general and allows the application to decide on the number of elements and meaning of the input vector, it doesn't give much guidance on this, as far as I see. The crucial case is when an application specifies that SIV is used with associated data and/or a nonce, but allows an empty string for either of those.
Also encoding empty adata, nonce "foo" in the same way as adata "foo", empty nonce, seems like a subtlety contrary to the spirit of 1.3.3.
+void +siv_cmac_set_key(struct siv_cmac_ctx *ctx, void *cipher,
const struct nettle_cipher *nc,const uint8_t *key)+{
- unsigned skey_size = nc->key_size;
- assert(skey_size <= SIV_MAX_KEY_SIZE/2);
- memcpy(ctx->s2vk, key, skey_size);
I think this function should do the underlying key setup also for the cipher instance used for s2v, not just store the key for later. So then the function would be
void siv_cmac_set_key(void *cmac_cipher, void *ctr_cipher, const struct nettle_cipher *nc, const uint8_t *key)
with no struct siv_cmac_ctx.
Regards, /Niels
On Sun, 2019-04-14 at 09:33 +0200, Niels Möller wrote:
Nikos Mavrogiannopoulos nmav@redhat.com writes:
This patch adds the SIV-CMAC algorithm to nettle (an update of the previous attempt). It is an atypical cipher which fits into the encrypt_message interface.
Thanks. Some comments below:
--- a/nettle-types.h +++ b/nettle-types.h @@ -78,6 +78,21 @@ typedef void *nettle_realloc_func(void *ctx, void *p, size_t length); /* Ciphers */ typedef void nettle_set_key_func(void *ctx, const uint8_t *key);
+/* AEAD ciphers */ +typedef void +nettle_encrypt_message(void *ctx,
size_t nlength, const uint8_t *nonce,size_t alength, const uint8_t *adata,size_t tlength,size_t clength, uint8_t *dst, const uint8_t*src);
+typedef int +nettle_decrypt_message(void *ctx,
size_t nlength, const uint8_t *nonce,size_t alength, const uint8_t *adata,size_t tlength,size_t mlength, uint8_t *dst, const uint8_t*src);
In this patch, these types used in tests only. But I imagine you'd like something in nettle-meta to represent a message-oriented aead?
I have never used the nettle-meta interface for ciphers, so I'm not sure about whether that's needed or not. CCM for example is not represented there. Would it make sense to separate the meta interface from the addition of the cipher, or should I avoid adding the type completely?
+@acronym{SIV-CMAC} mode is a combination of counter mode with message +authentication based on @acronym{CMAC}. Unlike other counter @acronym{AEAD} +modes, it provides protection against accidental nonce misuse, making it +a good choice for stateless-servers that cannot ensure nonce uniqueness.
Some detail on the nonce-reuse would be helpful. As I understood RFC 5297, the nonce used in SIV is only to make the ciphertexts of otherwise identical messages look different to the attacker.
My understanding of this RFC is that a primary goal as in 1.3.2, is to be resistant to nonce misuse (e.g., if you re-use the nonce). I can add more information but I am not sure I understand which direction of detail you mean. What are you missing or think is unclear in the text?
--- /dev/null +++ b/siv-aes128-cmac.c @@ -0,0 +1,79 @@
[...]
+void +siv_aes128_cmac_encrypt_message(struct siv_aes128_cmac_ctx *ctx,
size_t nlength, const uint8_t *nonce,size_t alength, const uint8_t *adata,size_t tlength,size_t slength, uint8_t *dst, constuint8_t *src) +{
- assert(tlength == SIV_DIGEST_SIZE);
The tlength argument doesn't look that useful, do we need it?.
If we remove the tlength then it would not implement the nettle_encrypt_message_func(). My goal in keeping tlength was to have a consistent interface across the message oriented ciphers.
Also, I think we agreed that the message length argument should be the size of the *destination* area, i.e., ciphertext size for encrypt_message (see ccm_*_encrypt_message):
I can name it clength, but there is the implicit assumption that clength = slength right? Otherwise if a larger buffer is provided for destination, I do not see how it figure the source size.
--- /dev/null +++ b/siv-cmac.c @@ -0,0 +1,194 @@ +/* This is a common structure for AES-128 and AES-256 thus
- for the cipher part we simply pad to the maximum structure
- size plus 16 bytes to account for any alignment difference in
- the original structures */
+struct cmac128_syn {
- struct cmac128_ctx ctx;
- struct {
- uint8_t pad[NETTLE_MAX_CIPHER16_CONTEXT_SIZE+16];
- } cipher;
+};
I don't think this guarantees any alignment. You would either need soemthing like
struct cipher_storage { union { uint64_t u64; char c[NETTLE_MAX_CIPHER16_CONTEXT_SIZE]; } storage; }; ... CMAC128_CTX(cipher_storage) ctx;
The goal here was not to make c aligned, but to have enough pad so that when this structure is read as cmac_aes128_ctx, to account for any alignment in the latter. Maybe actually I should do just that, and add the real types as union.
or use TMP_ALLOC_ALIGN (but in the latter case, you can't use the CMAC128_* macros). Or let caller pass in the cipher context (see last comment in this mail)
+static +void _siv_s2v(const struct nettle_cipher *nc,
const uint8_t *s2vk, size_t alength, const uint8_t*adata,
size_t nlength, const uint8_t *nonce,size_t plength, const uint8_t *pdata,uint8_t *v)+{
- struct cmac128_syn ctx;
- union nettle_block16 D, S, T;
- const uint8_t const_one[16] = {
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01
- };
- const uint8_t const_zero[16] = {
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
- };
Use static const.
Done. Why do you prefer this?
- assert(nc->context_size <= NETTLE_MAX_CIPHER16_CONTEXT_SIZE);
- /* ensure we have enough size of context plus any padding size
*/
- CMAC128_SET_KEY(&ctx, nc->set_encrypt_key, nc->encrypt, s2vk);
- if (nlength == 0 && alength == 0) {
- CMAC128_UPDATE(&ctx, nc->encrypt, 16, const_one);
- CMAC128_DIGEST(&ctx, nc->encrypt, 16, v);
- return;
- }
Shouldn't the plaintext, plength, pdata, still be processed in this case?
In this function, you treat empty associated data or nonce as those elements missing in the input vector to S2V. E.g., if both adata and nonce are empty, the input vector is { plaintext }, one single element. But it could also be { "", "", plaintext }, with three elements, the first two being empty strings.
To me, this sounds like a likely source of interop problems. Since RFC 5297 is general and allows the application to decide on the number of elements and meaning of the input vector, it doesn't give much guidance on this, as far as I see. The crucial case is when an application specifies that SIV is used with associated data and/or a nonce, but allows an empty string for either of those.
Also encoding empty adata, nonce "foo" in the same way as adata "foo", empty nonce, seems like a subtlety contrary to the spirit of 1.3.3.
I am not sure if the message interface can be used for derivation, but I'll need some time to answer your questions above.
+void +siv_cmac_set_key(struct siv_cmac_ctx *ctx, void *cipher,
const struct nettle_cipher *nc,const uint8_t *key)+{
- unsigned skey_size = nc->key_size;
- assert(skey_size <= SIV_MAX_KEY_SIZE/2);
- memcpy(ctx->s2vk, key, skey_size);
I think this function should do the underlying key setup also for the cipher instance used for s2v, not just store the key for later. So then the function would be
void siv_cmac_set_key(void *cmac_cipher, void *ctr_cipher, const struct nettle_cipher *nc, const uint8_t *key)
with no struct siv_cmac_ctx.
Why do you think that? Are there any benefits in that way?
regards, Nikos
Nikos Mavrogiannopoulos nmav@gnutls.org writes:
On Sun, 2019-04-14 at 09:33 +0200, Niels Möller wrote:
+typedef int +nettle_decrypt_message(void *ctx,
size_t nlength, const uint8_t *nonce,size_t alength, const uint8_t *adata,size_t tlength,size_t mlength, uint8_t *dst, const uint8_t*src);
In this patch, these types used in tests only. But I imagine you'd like something in nettle-meta to represent a message-oriented aead?
I have never used the nettle-meta interface for ciphers, so I'm not sure about whether that's needed or not. CCM for example is not represented there. Would it make sense to separate the meta interface from the addition of the cipher, or should I avoid adding the type completely?
Put these typedefs in testutils.h for now, if you find them useful for the tests.
Some detail on the nonce-reuse would be helpful. As I understood RFC 5297, the nonce used in SIV is only to make the ciphertexts of otherwise identical messages look different to the attacker.
My understanding of this RFC is that a primary goal as in 1.3.2, is to be resistant to nonce misuse (e.g., if you re-use the nonce). I can add more information but I am not sure I understand which direction of detail you mean. What are you missing or think is unclear in the text?
Something brief on what the consequences are if one accidentally or deliberately uses the same nonce twice. You could even quote the rtc, if it expresses it well and concisely.
--- /dev/null +++ b/siv-aes128-cmac.c @@ -0,0 +1,79 @@
[...]
+void +siv_aes128_cmac_encrypt_message(struct siv_aes128_cmac_ctx *ctx,
size_t nlength, const uint8_t *nonce,size_t alength, const uint8_t *adata,size_t tlength,size_t slength, uint8_t *dst, constuint8_t *src) +{
- assert(tlength == SIV_DIGEST_SIZE);
The tlength argument doesn't look that useful, do we need it?.
If we remove the tlength then it would not implement the nettle_encrypt_message_func(). My goal in keeping tlength was to have a consistent interface across the message oriented ciphers.
I'd prefer to not have the tlength argument. If we do a nettle-meta-style interface for messge-oriented aead, we may use a fix tlength, so that, e.g., CCM with full tag length (16 bytes) nd truncated tag (say, 4 bytes) would be considered distinct aead algorithms.
Also, I think we agreed that the message length argument should be the size of the *destination* area, i.e., ciphertext size for encrypt_message (see ccm_*_encrypt_message):
I can name it clength, but there is the implicit assumption that clength = slength right? Otherwise if a larger buffer is provided for destination, I do not see how it figure the source size.
clength == tlength + slength. This is what the docs for the corresponding ccm functions say (which do have tlength argument, since ccm supports truncated tag) say:
The CCM message fuctions provides a simple interface that will perform authentication and message encryption in a single function call. The length of the cleartext is given by MLENGTH and the length of the ciphertext is given by CLENGTH, always exactly TLENGTH bytes longer than the corresponding plaintext. The length argument passed to a function is always the size for the result, CLENGTH for the encryption functions, and MLENGTH for the decryption functions.
--- /dev/null +++ b/siv-cmac.c @@ -0,0 +1,194 @@ +/* This is a common structure for AES-128 and AES-256 thus
- for the cipher part we simply pad to the maximum structure
- size plus 16 bytes to account for any alignment difference in
- the original structures */
+struct cmac128_syn {
- struct cmac128_ctx ctx;
- struct {
- uint8_t pad[NETTLE_MAX_CIPHER16_CONTEXT_SIZE+16];
- } cipher;
+};
I don't think this guarantees any alignment. You would either need soemthing like
struct cipher_storage { union { uint64_t u64; char c[NETTLE_MAX_CIPHER16_CONTEXT_SIZE]; } storage; }; ... CMAC128_CTX(cipher_storage) ctx;
The goal here was not to make c aligned, but to have enough pad so that when this structure is read as cmac_aes128_ctx, to account for any alignment in the latter. Maybe actually I should do just that, and add the real types as union.
I think it's important that the context struct gets the alignment it needs. Now, since cmac128_ctx has aligned contents, uint8_t pad will likely get sufficient alignment anyway, but I think it's a bit brittle to depend on that; an uint8_t[...] inside a struct usually doesn't get larger alignment than a single byte.
or use TMP_ALLOC_ALIGN (but in the latter case, you can't use the CMAC128_* macros). Or let caller pass in the cipher context (see last comment in this mail)
+static +void _siv_s2v(const struct nettle_cipher *nc,
const uint8_t *s2vk, size_t alength, const uint8_t*adata,
size_t nlength, const uint8_t *nonce,size_t plength, const uint8_t *pdata,uint8_t *v)+{
- struct cmac128_syn ctx;
- union nettle_block16 D, S, T;
- const uint8_t const_one[16] = {
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01
- };
- const uint8_t const_zero[16] = {
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
- };
Use static const.
Done. Why do you prefer this?
Do ensure that the constants are allocated in the rodata segment. With just const, that's an abbreviation for const auto, and the compiler typically allocates them on the stack and generates code to initialize them on each call, like any other auto variables.
Not entirely sure why the compiler doesn't optimize it better, but using rodata for variables not explicitly declared static might violate some subtle requirement of the C standard. (I think const in C++ works differently, but I don't understand all the subtleties there either).
In addition, it would be good if these could be typed as nettle_block16 rather than char arrays, for alignment. That's easy for the const_zero block (this is also done in ax.c:eax_set_key), but I'm not sure of the top of my head how to write a union initializer for a
static const union nettle_block16 const_one = {...}
+void +siv_cmac_set_key(struct siv_cmac_ctx *ctx, void *cipher,
const struct nettle_cipher *nc,const uint8_t *key)+{
- unsigned skey_size = nc->key_size;
- assert(skey_size <= SIV_MAX_KEY_SIZE/2);
- memcpy(ctx->s2vk, key, skey_size);
I think this function should do the underlying key setup also for the cipher instance used for s2v, not just store the key for later. So then the function would be
void siv_cmac_set_key(void *cmac_cipher, void *ctr_cipher, const struct nettle_cipher *nc, const uint8_t *key)
with no struct siv_cmac_ctx.
Why do you think that? Are there any benefits in that way?
The idea of the set_key function is to do all preparations that don't depend on the actual message, so they don't have to be repeated. And I think it's a bit odd to handle the keying of the two involved cipher contexts so differently.
Slightly related: I think the cipher context struct(s) should be const void*, when passed to siv_cmac_encrypt_message and siv_cmac_decrypt_message. One usecase may be a server calling set_key at startup, and then spawning worker threads encrypting or decrypting messages. If all key-related data is const by the time the threads are spawned, it can be shared without any mutexes or the like.
Regards, /Niels
On Sun, 2019-04-14 at 09:33 +0200, Niels Möller wrote:
- assert(nc->context_size <= NETTLE_MAX_CIPHER16_CONTEXT_SIZE);
- /* ensure we have enough size of context plus any padding size
*/
- CMAC128_SET_KEY(&ctx, nc->set_encrypt_key, nc->encrypt, s2vk);
- if (nlength == 0 && alength == 0) {
- CMAC128_UPDATE(&ctx, nc->encrypt, 16, const_one);
- CMAC128_DIGEST(&ctx, nc->encrypt, 16, v);
- return;
- }
Shouldn't the plaintext, plength, pdata, still be processed in this case?
Right, there should be an and plength == 0 as well. I've added an two additional test cases to check these cases, and the case where everything is zero, doesn't seem to interoperate with two libs I tried.
Hopefully it is issue of this code.
https://github.com/miscreant/miscreant/issues/194 https://github.com/dfoxfranke/libaes_siv/issues/14
In this function, you treat empty associated data or nonce as those elements missing in the input vector to S2V. E.g., if both adata and nonce are empty, the input vector is { plaintext }, one single element. But it could also be { "", "", plaintext }, with three elements, the first two being empty strings.
While the low level function could handle it, it is not exposed to be called directly (mainly intentionally as this cipher introduces a very new paradigm which I do not quite see much of practical uses).
This patch only adds the higher level AEAD API only, so this case cannot happen as we don't have the notion of empty string in nettle. We can introduce it of course, though we may be opening a can of worms as not only empty strings are undefined in terms of AEAD API [0], but what would these mean in the other implementations?
[0]. https://tools.ietf.org/html/rfc5116#section-2
To me, this sounds like a likely source of interop problems. Since RFC 5297 is general and allows the application to decide on the number of elements and meaning of the input vector, it doesn't give much guidance on this, as far as I see. The crucial case is when an application specifies that SIV is used with associated data and/or a nonce, but allows an empty string for either of those.
I agree on that. That's one of the reasons I stuck on the higher level AEAD API (expressed by the message APIs in nettle). I added two sentences in the documentation about it.
I think this function should do the underlying key setup also for
the
cipher instance used for s2v, not just store the key for later. So then the function would be
void siv_cmac_set_key(void *cmac_cipher, void *ctr_cipher,
The idea of the set_key function is to do all preparations that don't depend on the actual message, so they don't have to be repeated. And I think it's a bit odd to handle the keying of the two involved cipher contexts so differently.
Done. It needed some reorganization, and cmac128_syn is still needed in an ugly simulation of the CMAC structure setup to use the macros. I have kept the union
The attached version should address the comments so far and also changes cmac128_set_key to use nettle_block16 as well.
regards, Nikos
On Wed, 2019-04-17 at 11:37 +0200, Nikos Mavrogiannopoulos wrote:
On Sun, 2019-04-14 at 09:33 +0200, Niels Möller wrote:
- assert(nc->context_size <= NETTLE_MAX_CIPHER16_CONTEXT_SIZE);
- /* ensure we have enough size of context plus any padding size
*/
- CMAC128_SET_KEY(&ctx, nc->set_encrypt_key, nc->encrypt, s2vk);
- if (nlength == 0 && alength == 0) {
- CMAC128_UPDATE(&ctx, nc->encrypt, 16, const_one);
- CMAC128_DIGEST(&ctx, nc->encrypt, 16, v);
- return;
- }
Shouldn't the plaintext, plength, pdata, still be processed in this case?
Right, there should be an and plength == 0 as well. I've added an two additional test cases to check these cases, and the case where everything is zero, doesn't seem to interoperate with two libs I tried.
Hopefully it is issue of this code.
https://github.com/miscreant/miscreant/issues/194 https://github.com/dfoxfranke/libaes_siv/issues/14
Based on the discussions on these bugs, I think it makes sense to strictly restrict the implementation to the common interoperable conventions (AEAD). That is, always assume (aad, nonce, plaintext), even if empty. As such I've updated the test vectors to include non empty nonce and tested them against libaes_siv.
The attached bugs update this implementation, and an additional patch adds a "make distcheck" check to CI. That ensures that any missing or incorrectly added in Makefile files are caught by the CI.
regards, Nikos
On Wed, 2019-04-17 at 20:27 +0200, Nikos Mavrogiannopoulos wrote:
- static const union nettle_block16 const_zero = { .b = {
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
- 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 }
You could save some space/eyes by using .b = 0 (assuming we can depend on modern C99 semantics) or also just via .u64 = { 0, 0 }
Simo.
Nikos Mavrogiannopoulos nmav@redhat.com writes:
In this function, you treat empty associated data or nonce as those elements missing in the input vector to S2V. E.g., if both adata and nonce are empty, the input vector is { plaintext }, one single element. But it could also be { "", "", plaintext }, with three elements, the first two being empty strings.
While the low level function could handle it, it is not exposed to be called directly (mainly intentionally as this cipher introduces a very new paradigm which I do not quite see much of practical uses).
This patch only adds the higher level AEAD API only, so this case cannot happen as we don't have the notion of empty string in nettle. We can introduce it of course, though we may be opening a can of worms as not only empty strings are undefined in terms of AEAD API [0], but what would these mean in the other implementations?
By empty string I simply meant a string of length 0. As I read RFC 5116, it requires that the key is between 1 and 255 octets (inclusive), but all of nonce, plaintext and associated data may consist of zero octets.
To me, this sounds like a likely source of interop problems. Since RFC 5297 is general and allows the application to decide on the number of elements and meaning of the input vector, it doesn't give much guidance on this, as far as I see. The crucial case is when an application specifies that SIV is used with associated data and/or a nonce, but allows an empty string for either of those.
I agree on that. That's one of the reasons I stuck on the higher level AEAD API (expressed by the message APIs in nettle). I added two sentences in the documentation about it.
The thing is, the AEAD api should allow inputs to be zero-length strings. Then the question is how to treat zero-length inputs in _siv_s2v, and I don't find RFC 5297 crystal clear on this point.
To me, it would make most sense for the AEAD construction to always use the S2V function in the spec with S1 = associated data (possibly zero length), S2 = nonce (possibly zero length), S3 = plaintext (possibly zero length). But we need to do what's needed to make it easy to interoperate with applicatinos and protocols using SIV; if everyone else does this differently, we should probably follow.
If we do it this way, then the nonce-less "key wrapping" usecase mentioned in RFC5297, with the example in A1, is *not* a special case of the AEAD construction, since this mode uses S1 = associated data, S2 = plaintext.
If we need to support several modes, maybe we should have a context struct that lets us do S2V incrementally, one element at a time,
siv_add_adata(struct siv_ctx *ctx, size_t length, const uint8_t *data);
called zero or more times, followed by
siv_encrypt(struct siv_ctx *ctx, size_t clength, uint8_t *dst, const uint8_t *src);
which does the CMAC of the plaintext (the final S2V element), extracts V, and does the encryption. (S2V is defined also for n == 0, but that's not very useful and we don't need to support it).
But I think we should start with the AEAD-style api with both adata and nonce mandatory (but possibly empty), before worrying too much about generalizations.
Done. It needed some reorganization, and cmac128_syn is still needed in an ugly simulation of the CMAC structure setup to use the macros. I have kept the union
Maybe it would be easier without using the CMAC macros. They're intended for convenience, so there's little point in using them where it doesn't bring any convenience.
The attached version should address the comments so far and also changes cmac128_set_key to use nettle_block16 as well.
Thanks! I hope to read through the new patch during the weekend.
Regards, /Niels
On Wed, 2019-04-17 at 20:41 +0200, Niels Möller wrote:
To me, this sounds like a likely source of interop problems. Since RFC 5297 is general and allows the application to decide on the number of elements and meaning of the input vector, it doesn't give much guidance on this, as far as I see. The crucial case is when an application specifies that SIV is used with associated data and/or a nonce, but allows an empty string for either of those.
I agree on that. That's one of the reasons I stuck on the higher level AEAD API (expressed by the message APIs in nettle). I added two sentences in the documentation about it.
The thing is, the AEAD api should allow inputs to be zero-length strings. Then the question is how to treat zero-length inputs in _siv_s2v, and I don't find RFC 5297 crystal clear on this point.
To me, it would make most sense for the AEAD construction to always use the S2V function in the spec with S1 = associated data (possibly zero length), S2 = nonce (possibly zero length), S3 = plaintext (possibly zero length). But we need to do what's needed to make it easy to interoperate with applicatinos and protocols using SIV; if everyone else does this differently, we should probably follow.
I agree. The patch I sent yesterday is towards that. I have verified that this approach interoperates with two implementations. The difference from what you write above is that we don't support at all the case where nonce=empty. That has interop issues (two interpretations, skip the field, or use it as empty), and I think it makes sense to leave it out. It has no use for our interface.
Today's patch adds two more vectors from another implementation and includes Simo's suggestion.
If we do it this way, then the nonce-less "key wrapping" usecase mentioned in RFC5297, with the example in A1, is *not* a special case of the AEAD construction, since this mode uses S1 = associated data, S2 = plaintext.
If we need to support several modes, maybe we should have a context struct that lets us do S2V incrementally, one element at a time,
Let's see if that is needed. For key wrapping I know no practical applications. I'd treat it as a separate algorithm, and we can add it later if needed.
Done. It needed some reorganization, and cmac128_syn is still needed in an ugly simulation of the CMAC structure setup to use the macros. I have kept the union
Maybe it would be easier without using the CMAC macros. They're intended for convenience, so there's little point in using them where it doesn't bring any convenience.
I do not think that avoiding them would change this part.
regards, Nikos
Simo informed me that I didn't update the CMAC file with the new initializer. Instead of spamming the list with numerous patches, my latest version is at:
https://gitlab.com/nmav/nettle/merge_requests/4/
Can be downloaded as patches at: https://gitlab.com/nmav/nettle/merge_requests/4.patch
On Thu, Apr 18, 2019 at 9:00 AM Nikos Mavrogiannopoulos nmav@redhat.com wrote:
On Wed, 2019-04-17 at 20:41 +0200, Niels Möller wrote:
To me, this sounds like a likely source of interop problems. Since RFC 5297 is general and allows the application to decide on the number of elements and meaning of the input vector, it doesn't give much guidance on this, as far as I see. The crucial case is when an application specifies that SIV is used with associated data and/or a nonce, but allows an empty string for either of those.
I agree on that. That's one of the reasons I stuck on the higher level AEAD API (expressed by the message APIs in nettle). I added two sentences in the documentation about it.
The thing is, the AEAD api should allow inputs to be zero-length strings. Then the question is how to treat zero-length inputs in _siv_s2v, and I don't find RFC 5297 crystal clear on this point.
To me, it would make most sense for the AEAD construction to always use the S2V function in the spec with S1 = associated data (possibly zero length), S2 = nonce (possibly zero length), S3 = plaintext (possibly zero length). But we need to do what's needed to make it easy to interoperate with applicatinos and protocols using SIV; if everyone else does this differently, we should probably follow.
I agree. The patch I sent yesterday is towards that. I have verified that this approach interoperates with two implementations. The difference from what you write above is that we don't support at all the case where nonce=empty. That has interop issues (two interpretations, skip the field, or use it as empty), and I think it makes sense to leave it out. It has no use for our interface.
Today's patch adds two more vectors from another implementation and includes Simo's suggestion.
If we do it this way, then the nonce-less "key wrapping" usecase mentioned in RFC5297, with the example in A1, is *not* a special case of the AEAD construction, since this mode uses S1 = associated data, S2 = plaintext.
If we need to support several modes, maybe we should have a context struct that lets us do S2V incrementally, one element at a time,
Let's see if that is needed. For key wrapping I know no practical applications. I'd treat it as a separate algorithm, and we can add it later if needed.
Done. It needed some reorganization, and cmac128_syn is still needed in an ugly simulation of the CMAC structure setup to use the macros. I have kept the union
Maybe it would be easier without using the CMAC macros. They're intended for convenience, so there's little point in using them where it doesn't bring any convenience.
I do not think that avoiding them would change this part.
regards, Nikos
Nikos Mavrogiannopoulos nmav@redhat.com writes:
I've integrated this onto a branch siv-mode, with some changes (renamed functions to all have the siv_cmac prefix, document that empty nonce may have interop issues and should be considered experimental (but still allow it), and dropped some macros and unrelated changes (which would belong on a different branch). Please have a look, and say if I've messed up something.
Regards, /Niels
Thanks. If you added the zero-nonce method, maybe it would be better to add test vectors for it as well. I'm copying from my last patch with it:
+ /* + * Example with no nonce, no AD and no plaintext; taken from + * https://github.com/miscreant/miscreant/blob/master/vectors/aes_siv.tjson + * however we don't interoperate. + */ + test_siv_aes128("AES_SIV_CMAC256", sizeof(struct siv_aes128_cmac_ctx), + &nettle_aes128, + SHEX("fffefdfc fbfaf9f8 f7f6f5f4 f3f2f1f0" + "f0f1f2f3 f4f5f6f7 f8f9fafb fcfdfeff"), + SHEX(""), + SHEX(""), + SHEX(""), + SHEX("949f99cb cc3eb5da6 d3c45d0 f59aa9c7")); + + /* + * Example with no nonce, no AD and plaintext; taken from + * https://github.com/miscreant/miscreant/blob/master/vectors/aes_siv.tjson + */ + test_siv_aes128("AES_SIV_CMAC256", sizeof(struct siv_aes128_cmac_ctx), + &nettle_aes128, + SHEX("fffefdfc fbfaf9f8 f7f6f5f4 f3f2f1f0" + "f0f1f2f3 f4f5f6f7 f8f9fafb fcfdfeff"), + SHEX(""), + SHEX(""), + SHEX("00112233 44556677 8899aabb ccddeeff"), + SHEX("f304f912 863e303d 5b540e50 57c7010c" + "942ffaf4 5b0e5ca5 fb9a56a5 263bb065")); + /* + * Example without nonce length < 16 (RFC5297) + */ + test_siv_aes128("AES_SIV_CMAC256", sizeof(struct siv_aes128_cmac_ctx), + &nettle_aes128, + SHEX("fffefdfc fbfaf9f8 f7f6f5f4 f3f2f1f0" + "f0f1f2f3 f4f5f6f7 f8f9fafb fcfdfeff"), + SHEX(""), + SHEX("10111213 14151617 18191a1b 1c1d1e1f" + "20212223 24252627"), + SHEX("11223344 55667788 99aabbcc ddee"), + SHEX("85632d07 c6e8f37f 950acd32 0a2ecc93" + "40c02b96 90c4dc04 daef7f6a fe5c")); + + /* + * Example without nonce length > 16 + */ + test_siv_aes128("AES_SIV_CMAC256", sizeof(struct siv_aes128_cmac_ctx), + &nettle_aes128, + SHEX("7f7e7d7c 7b7a7978 77767574 73727170" + "40414243 44454647 48494a4b 4c4d4e4f"), + SHEX(""), + SHEX("00112233 44556677 8899aabb ccddeeff" + "deaddada deaddada ffeeddcc bbaa9988" + "77665544 33221100"), + SHEX("74686973 20697320 736f6d65 20706c61" + "696e7465 78742074 6f20656e 63727970" + "74207573 696e6720 5349562d 414553"), + SHEX("c12ccaa7 54e1b3fa 4f416c18 415625ca" + "472fbee de5bc03f 34934819 a9abb20b5" + "8cd019c 470ac832 f6eb9ddf 0656c5dce" + "ffe611a 5a5ca3e1 c3c12da5 6e4bb87")); + /* AES-SIV-CMAC-512 (AES-256) + */ + test_siv_aes256("AES_SIV_CMAC512", sizeof(struct siv_aes256_cmac_ctx), + &nettle_aes256, + SHEX("c27df2fd aec35d4a 2a412a50 c3e8c47d" + "2d568e91 a38e5414 8abdc0b6 e86caf87" + "695c0a8a df4c5f8e b2c6c8b1 36529864" + "f3b84b3a e8e3676c e760c461 f3a13e83"), + SHEX(""), + SHEX("10111213 14151617 18191a1b 1c1d1e1f" + "20212223 24252627"), + SHEX("11223344 55667788 99aabbcc ddee"), + SHEX("ae2b1bd1 ba7fcd6a 4f9f7eb2 4b40f766" + "86053ffd c384cb19 76031f46 3013")); + + /* + * Example without nonce length > 16 + */ + test_siv_aes256("AES_SIV_CMAC512", sizeof(struct siv_aes256_cmac_ctx), + &nettle_aes256, + SHEX("c27df2fd aec35d4a 2a412a50 c3e8c47d" + "2d568e91 a38e5414 8abdc0b6 e86caf87" + "695c0a8a df4c5f8e b2c6c8b1 36529864" + "f3b84b3a e8e3676c e760c461 f3a13e83"), + SHEX(""), + SHEX("00112233 44556677 8899aabb ccddeeff" + "deaddada deaddada ffeeddcc bbaa9988" + "77665544 33221100"), + SHEX("74686973 20697320 736f6d65 20706c61" + "696e7465 78742074 6f20656e 63727970" + "74207573 696e6720 5349562d 414553"), + SHEX("79476aaa 388374fe 97d0db51 596cb5ee" + "a933e001 412026c7 956c82dd b753b1af" + "3d7d49ac 474a800a c14b4bab a4542067" + "83647ef9 51315dab b7a2c05b 288ba8"));
On Mon, May 6, 2019 at 7:45 PM Niels Möller nisse@lysator.liu.se wrote:
Nikos Mavrogiannopoulos nmav@redhat.com writes:
I've integrated this onto a branch siv-mode, with some changes (renamed functions to all have the siv_cmac prefix, document that empty nonce may have interop issues and should be considered experimental (but still allow it), and dropped some macros and unrelated changes (which would belong on a different branch). Please have a look, and say if I've messed up something.
Regards, /Niels
-- Niels Möller. PGP-encrypted email is preferred. Keyid 368C6677. Internet email is subject to wholesale government surveillance. _______________________________________________ nettle-bugs mailing list nettle-bugs@lists.lysator.liu.se http://lists.lysator.liu.se/mailman/listinfo/nettle-bugs
Nikos Mavrogiannopoulos n.mavrogiannopoulos@gmail.com writes:
Thanks. If you added the zero-nonce method, maybe it would be better to add test vectors for it as well. I'm copying from my last patch with it:
I was about to add the miscreant.js examples (and with nettle's output, which is different), to illustrate interop issue. Unfortunately, the RFC 5297 testvectors appear useless if one wants to test the RFC 5116 mode of operation.
And on second thought, maybe it makes more sense to change nettle to be interoperable with miscreant here? I think that's how you did it originally, and I found it confusing. RFC 5297 (SIV mode) says that for use according to RFC5116 (AEAD interface), N_MIN = 1.
Another option, which you've also tried, is to to require non-empty nonce, i.e., add back the assert (nlength > 0), and define SIV_MIN_NONCE_SIZE as one, not zero. That's perhaps the most conservative approach: support for empty nonce, however that should behave, can be added later.
Opinions?
Regards, /Niels
I prefer the second option because I think the zero nonce variant requires a disproportionate, to its usefullness and use, discussion to define the "right" semantics.
On May 11, 2019 7:49:31 AM UTC, nisse@lysator.liu.se wrote:
Nikos Mavrogiannopoulos n.mavrogiannopoulos@gmail.com writes:
Thanks. If you added the zero-nonce method, maybe it would be better to add test vectors for it as well. I'm copying from my last patch with it:
I was about to add the miscreant.js examples (and with nettle's output, which is different), to illustrate interop issue. Unfortunately, the RFC 5297 testvectors appear useless if one wants to test the RFC 5116 mode of operation.
And on second thought, maybe it makes more sense to change nettle to be interoperable with miscreant here? I think that's how you did it originally, and I found it confusing. RFC 5297 (SIV mode) says that for use according to RFC5116 (AEAD interface), N_MIN = 1.
Another option, which you've also tried, is to to require non-empty nonce, i.e., add back the assert (nlength > 0), and define SIV_MIN_NONCE_SIZE as one, not zero. That's perhaps the most conservative approach: support for empty nonce, however that should behave, can be added later.
Opinions?
Regards, /Niels
Nikos Mavrogiannopoulos n.mavrogiannopoulos@gmail.com writes:
I prefer the second option because I think the zero nonce variant requires a disproportionate, to its usefullness and use, discussion to define the "right" semantics.
Merged siv-mode to the master branch now. Does it look right to you?
Regards, /Niels
Looks good to me, but I'm adding Mirek in CC who is using SIV-AES-CMAC for NTS/NTP implementation to verify that the final code is sufficient for this implementation.
regards, Nikos
On Tue, Jul 2, 2019 at 4:25 PM Niels Möller nisse@lysator.liu.se wrote:
Nikos Mavrogiannopoulos n.mavrogiannopoulos@gmail.com writes:
I prefer the second option because I think the zero nonce variant requires a disproportionate, to its usefullness and use, discussion to define the "right" semantics.
Merged siv-mode to the master branch now. Does it look right to you?
Regards, /Niels
-- Niels Möller. PGP-encrypted email is preferred. Keyid 368C6677. Internet email is subject to wholesale government surveillance.
On Wed, Jul 03, 2019 at 06:26:07AM +0200, Nikos Mavrogiannopoulos wrote:
Looks good to me, but I'm adding Mirek in CC who is using SIV-AES-CMAC for NTS/NTP implementation to verify that the final code is sufficient for this implementation.
I've updated my code to use the Nettle's SIV-CMAC and it seems to be working fine. It interoperates with the previous version of itself (using Nikos' original SIV-CMAC implementation with slightly different API) and two other NTS implementations (one based on openssl, not sure about the other).
Thank you both!
nettle-bugs@lists.lysator.liu.se
-
Miroslav Lichvar -
Nikos Mavrogiannopoulos -
Nikos Mavrogiannopoulos -
Nikos Mavrogiannopoulos -
nisse＠lysator.liu.se -
Simo Sorce