Skip to contentSkip to conversation

mptcp: simplify crypto.c

RFC PATCH In review Davide Caratti 3 participants 6 messages Jun 14, 2019 – Jun 20, 2019

Subsystems (2)
lore
Share & tools
lore

Davide Caratti

author
#0 lore raw

cover mptcp: simplify crypto.c

this series wants to simplify, and hopefully improve, random generation of keys and nonces in crypto.c, While at it, I removed some unneeded #includes on top of the file. Any feedbacks are welcome!

Note:
in case you think the code is ok, I will propose to squash it into commit 34f09d53931a ("mptcp: Add key generation and token tree")

Davide Caratti (2):
net: mptcp: randomness improvements for crypto.c
net: mptcp: remove useless #includes in crypto.c

 net/mptcp/crypto.c   | 51 +++++---------------------------------------
 net/mptcp/protocol.c |  1 -
 net/mptcp/protocol.h |  6 ------
 3 files changed, 5 insertions(+), 53 deletions(-)
Signature · 1 line
2.20.1
author
#1 lore raw

RFC PATCH 1/2 net: mptcp: randomness improvements for crypto.c

- use net_get_random_once() to defer initialization of the crypto secret, instead of computing it at boot time, and remove 'crypto_init()'.

- don't use siphash for nonces, and also remove crypto_get_nonce_v{4,6}. They are not used on current MPTCP, and future users should suffice

     get_random_bytes(&nonce, sizeof(u32));

like it's done by other protocols in the Linux kernel (like SCTP).

- use tcp_cookie_time() instead of 'seed', like it's done in syncookies, so that the initial value of 'seed' is time-dependent (not always 0).

CC Florian Westphal <fwestpha(a)redhat.com>
CC Paolo Abeni <pabeni(a)redhat.com>
Signed-off-by Davide Caratti <dcaratti(a)redhat.com>
---
 net/mptcp/crypto.c   | 47 +++++---------------------------------------
 net/mptcp/protocol.c |  1 -
 net/mptcp/protocol.h |  6 ------
 3 files changed, 5 insertions(+), 49 deletions(-)
net/mptcp/crypto.c +5 -42
index 2dc453d1d5ab..48b6748a8d25 100644
@@ -29,47 +29,18 @@
 #include <linux/cryptohash.h>
 #include <linux/random.h>
 #include <linux/siphash.h>
+#include <net/tcp.h>
 #include <asm/unaligned.h>
 
 static siphash_key_t crypto_key_secret __read_mostly;
-static hsiphash_key_t crypto_nonce_secret __read_mostly;
-static u32 crypto_seed;
-
-u32 crypto_v4_get_nonce(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport)
-{
-	return hsiphash_4u32((__force u32)saddr, (__force u32)daddr,
-			    (__force u32)sport << 16 | (__force u32)dport,
-			    crypto_seed++, &crypto_nonce_secret);
-}
 
 u64 crypto_v4_get_key(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport)
 {
 	pr_debug("src=%x:%d, dst=%x:%d", saddr, sport, daddr, dport);
+	net_get_random_once(&crypto_key_secret, sizeof(crypto_key_secret));
 	return siphash_4u32((__force u32)saddr, (__force u32)daddr,
 			    (__force u32)sport << 16 | (__force u32)dport,
-			    crypto_seed++, &crypto_key_secret);
-}
-
-u32 crypto_v6_get_nonce(const struct in6_addr *saddr,
-			const struct in6_addr *daddr,
-			__be16 sport, __be16 dport)
-{
-	const struct {
-		struct in6_addr saddr;
-		struct in6_addr daddr;
-		u32 seed;
-		__be16 sport;
-		__be16 dport;
-	} __aligned(SIPHASH_ALIGNMENT) combined = {
-		.saddr = *saddr,
-		.daddr = *daddr,
-		.seed = crypto_seed++,
-		.sport = sport,
-		.dport = dport,
-	};
-
-	return hsiphash(&combined, offsetofend(typeof(combined), dport),
-			&crypto_nonce_secret);
+			    tcp_cookie_time(), &crypto_key_secret);
 }
 
 u64 crypto_v6_get_key(const struct in6_addr *saddr,
@@ -85,11 +56,12 @@ u64 crypto_v6_get_key(const struct in6_addr *saddr,
 	} __aligned(SIPHASH_ALIGNMENT) combined = {
 		.saddr = *saddr,
 		.daddr = *daddr,
-		.seed = crypto_seed++,
+		.seed = tcp_cookie_time(),
 		.sport = sport,
 		.dport = dport,
 	};
 
+	net_get_random_once(&crypto_key_secret, sizeof(crypto_key_secret));
 	return siphash(&combined, offsetofend(typeof(combined), dport),
 		       &crypto_key_secret);
 }
@@ -196,12 +168,3 @@ void crypto_hmac_sha1(u64 key1, u64 key2, u32 *hash_out,
 	for (i = 0; i < 5; i++)
 		hash_out[i] = (__force u32)cpu_to_be32(hash_out[i]);
 }
-
-void crypto_init(void)
-{
-	get_random_bytes((void *)&crypto_key_secret,
-			 sizeof(crypto_key_secret));
-	get_random_bytes((void *)&crypto_nonce_secret,
-			 sizeof(crypto_nonce_secret));
-	crypto_seed = 0;
-}
net/mptcp/protocol.c +0 -1
index 941988387ea0..1bff0d2556d5 100644
@@ -1008,7 +1008,6 @@ void __init mptcp_init(void)
 	mptcp_stream_ops.shutdown = mptcp_shutdown;
 
 	token_init();
-	crypto_init();
 	subflow_init();
 
 	if (proto_register(&mptcp_prot, 1) != 0)
net/mptcp/protocol.h +0 -6
index a4f0e7d3bd62..471b885bb2db 100644
@@ -131,17 +131,11 @@ void token_new_accept(struct sock *sk);
 void token_update_accept(struct sock *sk, struct sock *conn);
 void token_destroy(u32 token);
 
-void crypto_init(void);
-u32 crypto_v4_get_nonce(__be32 saddr, __be32 daddr,
-			__be16 sport, __be16 dport);
 u64 crypto_v4_get_key(__be32 saddr, __be32 daddr,
 			__be16 sport, __be16 dport);
 u64 crypto_v6_get_key(const struct in6_addr *saddr,
 		      const struct in6_addr *daddr,
 		      __be16 sport, __be16 dport);
-u32 crypto_v6_get_nonce(const struct in6_addr *saddr,
-			const struct in6_addr *daddr,
-			__be16 sport, __be16 dport);
 void crypto_key_sha1(u64 key, u32 *token, u64 *idsn);
 void crypto_hmac_sha1(u64 key1, u64 key2, u32 *hash_out,
 		     int arg_num, ...);
Signature · 1 line
2.20.1
#2 lore raw

Re: [MPTCP] [PATCH RFC 1/2] net: mptcp: randomness improvements for crypto.c

Quoted 38 lines · Davide Caratti

- use net_get_random_once() to defer initialization of the crypto secret, instead of computing it at boot time, and remove 'crypto_init()'.

- don't use siphash for nonces, and also remove crypto_get_nonce_v{4,6}. They are not used on current MPTCP, and future users should suffice

    get_random_bytes(&nonce, sizeof(u32));

like it's done by other protocols in the Linux kernel (like SCTP).

- use tcp_cookie_time() instead of 'seed', like it's done in syncookies, so that the initial value of 'seed' is time-dependent (not always 0).

CC Florian Westphal <fwestpha(a)redhat.com>
CC Paolo Abeni <pabeni(a)redhat.com>
Signed-off-by Davide Caratti <dcaratti(a)redhat.com>

---

net/mptcp/crypto.c   | 47 +++++---------------------------------------
net/mptcp/protocol.c |  1 -
net/mptcp/protocol.h |  6 ------
3 files changed, 5 insertions(+), 49 deletions(-)
net/mptcp/crypto.c +1 -4
index 2dc453d1d5ab..48b6748a8d25 100644
@@ -29,47 +29,18 @@
 #include <linux/cryptohash.h>
 #include <linux/random.h>
 #include <linux/siphash.h>
+#include <net/tcp.h>
 #include <asm/unaligned.h>
 
 static siphash_key_t crypto_key_secret __read_mostly;
-static hsiphash_key_t crypto_nonce_secret __read_mostly;
-static u32 crypto_seed;
-
-u32 crypto_v4_get_nonce(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport)

net/mptcp/crypto.c line 40

Quoted 10 lines
net/mptcp/crypto.c +1 -5
-{
-	return hsiphash_4u32((__force u32)saddr, (__force u32)daddr,
-			    (__force u32)sport << 16 | (__force u32)dport,
-			    crypto_seed++, &crypto_nonce_secret);
-}
 
 u64 crypto_v4_get_key(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport)
 {
 	pr_debug("src=%x:%d, dst=%x:%d", saddr, sport, daddr, dport);
+	net_get_random_once(&crypto_key_secret, sizeof(crypto_key_secret));

Thanks for this fix, initializing this later also gives more time for get_random_bytes() to get properly seeded.

Quoted 16 lines
net/mptcp/crypto.c +0 -14
 	return siphash_4u32((__force u32)saddr, (__force u32)daddr,
 			    (__force u32)sport << 16 | (__force u32)dport,
-			    crypto_seed++, &crypto_key_secret);
-}
-
-u32 crypto_v6_get_nonce(const struct in6_addr *saddr,
-			const struct in6_addr *daddr,
-			__be16 sport, __be16 dport)
-{
-	const struct {
-		struct in6_addr saddr;
-		struct in6_addr daddr;
-		u32 seed;
-		__be16 sport;
-		__be16 dport;
-	} __aligned(SIPHASH_ALIGNMENT) combined = {

net/mptcp/crypto.c line 43

Quoted 10 lines
net/mptcp/crypto.c +1 -9
-		.saddr = *saddr,
-		.daddr = *daddr,
-		.seed = crypto_seed++,
-		.sport = sport,
-		.dport = dport,
-	};
-
-	return hsiphash(&combined, offsetofend(typeof(combined), dport),
-			&crypto_nonce_secret);
+			    tcp_cookie_time(), &crypto_key_secret);

tcp_cookie_time() adds a dependency on CONFIG_SYN_COOKIES. It also looks like tcp_cookie_time() only increments once per second, so the uniqueness of the key (within a 1-second interval) becomes totally dependent on the uniqueness of the 4-tuple. I think a different seed source is needed that won't repeat so predictably.

net/mptcp/crypto.c line 45

Quoted 2 lines
net/mptcp/crypto.c +0 -0
 }
 

Thanks,

--
Mat Martineau
Intel

#3 lore raw

Re: [MPTCP] [PATCH RFC 1/2] net: mptcp: randomness improvements for crypto.c

Quoted 9 lines · Mat Martineau

net/mptcp/crypto.c line 43

Quoted 3 lines
net/mptcp/crypto.c +1 -2
-	return hsiphash(&combined, offsetofend(typeof(combined), dport),
-			&crypto_nonce_secret);
+			    tcp_cookie_time(), &crypto_key_secret);

tcp_cookie_time() adds a dependency on CONFIG_SYN_COOKIES. It also looks like tcp_cookie_time() only increments once per second, so the uniqueness of the key (within a 1-second interval) becomes totally dependent on the uniqueness of the 4-tuple. I think a different seed source is needed that won't repeat so predictably.

Thats a good point, this means we will also need to include net_hash_mix() into the input so we won't create same keys for identical tuples in different network namespaces.

But, I wonder, if the key doesn't need to be recomputeable at all, then this is just a random value to send, so two alternatives might be:

1. just use get_random_bytes
2. use prandom_u32() as additional input key instead of increment++.

#4 lore raw

Re: [MPTCP] [PATCH RFC 1/2] net: mptcp: randomness improvements for crypto.c

Quoted 15 lines · Florian Westphal
Quoted 9 lines · Mat Martineau

net/mptcp/crypto.c line 43

Quoted 3 lines
net/mptcp/crypto.c +1 -2
-	return hsiphash(&combined, offsetofend(typeof(combined), dport),
-			&crypto_nonce_secret);
+			    tcp_cookie_time(), &crypto_key_secret);

tcp_cookie_time() adds a dependency on CONFIG_SYN_COOKIES. It also looks like tcp_cookie_time() only increments once per second, so the uniqueness of the key (within a 1-second interval) becomes totally dependent on the uniqueness of the 4-tuple. I think a different seed source is needed that won't repeat so predictably.

Thats a good point, this means we will also need to include net_hash_mix() into the input so we won't create same keys for identical tuples in different network namespaces.

Quoted 5 lines

But, I wonder, if the key doesn't need to be recomputeable at all, then this is just a random value to send, so two alternatives might be:

1. just use get_random_bytes
2. use prandom_u32() as additional input key instead of increment++.

We talked about this in the meeting today. Right now there's no recomputing of the key, so a random key would be simplest. I'm not sure if that's more expensive to compute, but we are not locking ourselves in to anything with our key generation choice now.

--
Mat Martineau
Intel

author
#5 lore raw

RFC PATCH 2/2 net: mptcp: remove useless #includes in crypto.c

In reply to Davide Caratti

- crypto.c doesn't need module.h (that was included twice)
- crypto.c doesn't need netdevice.h
- remove duplicate include of kernel.h

Signed-off-by Davide Caratti <dcaratti(a)redhat.com>
---
 net/mptcp/crypto.c | 4 ----
 1 file changed, 4 deletions(-)
net/mptcp/crypto.c +0 -4
index 48b6748a8d25..117f2dd9e70a 100644
@@ -22,10 +22,6 @@
  */
 
 #include <linux/kernel.h>
-#include <linux/module.h>
-#include <linux/netdevice.h>
-#include <linux/kernel.h>
-#include <linux/module.h>
 #include <linux/cryptohash.h>
 #include <linux/random.h>
 #include <linux/siphash.h>
Signature · 1 line
2.20.1