add chacha20 implementation

Signed-off-by: Felix Fietkau <nbd@nbd.name>

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 651ac804ad6714a9ed22b2232fc4691c2e37544b..4c3f67025bafb66298da7f4f477bd02f1ee39ba8 100644 (file)
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -30,7 +30,7 @@ ELSE()
   SET(ubus "")
 ENDIF()
 
-ADD_LIBRARY(unet SHARED curve25519.c siphash.c sha512.c fprime.c f25519.c ed25519.c edsign.c auth-data.c)
+ADD_LIBRARY(unet SHARED curve25519.c siphash.c sha512.c fprime.c f25519.c ed25519.c edsign.c auth-data.c chacha20.c)
 TARGET_LINK_LIBRARIES(unet ubox)
 
 ADD_EXECUTABLE(unetd ${SOURCES})

diff --git a/chacha20.c b/chacha20.c
new file mode 100644 (file)
index 0000000..71f7eac
--- /dev/null
+++ b/chacha20.c
@@ -0,0 +1,236 @@
+
+/*
+ chacha-merged.c version 20080118
+ D. J. Bernstein
+ Public domain.
+ */
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include "utils.h"
+#include "chacha20.h"
+
+struct chacha_ctx {
+    uint32_t input[16];
+};
+
+#define LOAD32_LE(SRC) get_unaligned_le32(SRC)
+
+#define STORE32_LE(DST, W) store32_le((DST), (W))
+
+static inline void
+store32_le(uint8_t dst[4], uint32_t w)
+{
+       dst[0] = (uint8_t) w; w >>= 8;
+       dst[1] = (uint8_t) w; w >>= 8;
+       dst[2] = (uint8_t) w; w >>= 8;
+       dst[3] = (uint8_t) w;
+}
+
+
+#define ROTL32(X, B) rotl32((X), (B))
+static inline uint32_t
+rotl32(const uint32_t x, const int b)
+{
+       return (x << b) | (x >> (32 - b));
+}
+
+typedef struct chacha_ctx chacha_ctx;
+
+#define U32C(v) (v##U)
+
+#define U32V(v) ((uint32_t)(v) &U32C(0xFFFFFFFF))
+
+#define ROTATE(v, c) (ROTL32(v, c))
+#define XOR(v, w) ((v) ^ (w))
+#define PLUS(v, w) (U32V((v) + (w)))
+#define PLUSONE(v) (PLUS((v), 1))
+
+#define QUARTERROUND(a, b, c, d) \
+    a = PLUS(a, b);              \
+    d = ROTATE(XOR(d, a), 16);   \
+    c = PLUS(c, d);              \
+    b = ROTATE(XOR(b, c), 12);   \
+    a = PLUS(a, b);              \
+    d = ROTATE(XOR(d, a), 8);    \
+    c = PLUS(c, d);              \
+    b = ROTATE(XOR(b, c), 7);
+
+static void
+chacha_keysetup(chacha_ctx *ctx, const uint8_t *k)
+{
+    ctx->input[0]  = U32C(0x61707865);
+    ctx->input[1]  = U32C(0x3320646e);
+    ctx->input[2]  = U32C(0x79622d32);
+    ctx->input[3]  = U32C(0x6b206574);
+    ctx->input[4]  = LOAD32_LE(k + 0);
+    ctx->input[5]  = LOAD32_LE(k + 4);
+    ctx->input[6]  = LOAD32_LE(k + 8);
+    ctx->input[7]  = LOAD32_LE(k + 12);
+    ctx->input[8]  = LOAD32_LE(k + 16);
+    ctx->input[9]  = LOAD32_LE(k + 20);
+    ctx->input[10] = LOAD32_LE(k + 24);
+    ctx->input[11] = LOAD32_LE(k + 28);
+}
+
+static void
+chacha_ivsetup(chacha_ctx *ctx, const uint8_t *iv, const uint8_t *counter)
+{
+    ctx->input[12] = counter == NULL ? 0 : LOAD32_LE(counter + 0);
+    ctx->input[13] = counter == NULL ? 0 : LOAD32_LE(counter + 4);
+    ctx->input[14] = LOAD32_LE(iv + 0);
+    ctx->input[15] = LOAD32_LE(iv + 4);
+}
+
+static void
+chacha20_encrypt_bytes(chacha_ctx *ctx, const uint8_t *m, uint8_t *c,
+                       unsigned long long bytes)
+{
+    uint32_t x0, x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14,
+        x15;
+    uint32_t j0, j1, j2, j3, j4, j5, j6, j7, j8, j9, j10, j11, j12, j13, j14,
+        j15;
+    uint8_t     *ctarget = NULL;
+    uint8_t      tmp[64];
+    unsigned int i;
+
+    if (!bytes) {
+        return; /* LCOV_EXCL_LINE */
+    }
+    j0  = ctx->input[0];
+    j1  = ctx->input[1];
+    j2  = ctx->input[2];
+    j3  = ctx->input[3];
+    j4  = ctx->input[4];
+    j5  = ctx->input[5];
+    j6  = ctx->input[6];
+    j7  = ctx->input[7];
+    j8  = ctx->input[8];
+    j9  = ctx->input[9];
+    j10 = ctx->input[10];
+    j11 = ctx->input[11];
+    j12 = ctx->input[12];
+    j13 = ctx->input[13];
+    j14 = ctx->input[14];
+    j15 = ctx->input[15];
+
+    for (;;) {
+        if (bytes < 64) {
+            memset(tmp, 0, 64);
+            for (i = 0; i < bytes; ++i) {
+                tmp[i] = m[i];
+            }
+            m       = tmp;
+            ctarget = c;
+            c       = tmp;
+        }
+        x0  = j0;
+        x1  = j1;
+        x2  = j2;
+        x3  = j3;
+        x4  = j4;
+        x5  = j5;
+        x6  = j6;
+        x7  = j7;
+        x8  = j8;
+        x9  = j9;
+        x10 = j10;
+        x11 = j11;
+        x12 = j12;
+        x13 = j13;
+        x14 = j14;
+        x15 = j15;
+        for (i = 20; i > 0; i -= 2) {
+            QUARTERROUND(x0, x4, x8, x12)
+            QUARTERROUND(x1, x5, x9, x13)
+            QUARTERROUND(x2, x6, x10, x14)
+            QUARTERROUND(x3, x7, x11, x15)
+            QUARTERROUND(x0, x5, x10, x15)
+            QUARTERROUND(x1, x6, x11, x12)
+            QUARTERROUND(x2, x7, x8, x13)
+            QUARTERROUND(x3, x4, x9, x14)
+        }
+        x0  = PLUS(x0, j0);
+        x1  = PLUS(x1, j1);
+        x2  = PLUS(x2, j2);
+        x3  = PLUS(x3, j3);
+        x4  = PLUS(x4, j4);
+        x5  = PLUS(x5, j5);
+        x6  = PLUS(x6, j6);
+        x7  = PLUS(x7, j7);
+        x8  = PLUS(x8, j8);
+        x9  = PLUS(x9, j9);
+        x10 = PLUS(x10, j10);
+        x11 = PLUS(x11, j11);
+        x12 = PLUS(x12, j12);
+        x13 = PLUS(x13, j13);
+        x14 = PLUS(x14, j14);
+        x15 = PLUS(x15, j15);
+
+        x0  = XOR(x0, LOAD32_LE(m + 0));
+        x1  = XOR(x1, LOAD32_LE(m + 4));
+        x2  = XOR(x2, LOAD32_LE(m + 8));
+        x3  = XOR(x3, LOAD32_LE(m + 12));
+        x4  = XOR(x4, LOAD32_LE(m + 16));
+        x5  = XOR(x5, LOAD32_LE(m + 20));
+        x6  = XOR(x6, LOAD32_LE(m + 24));
+        x7  = XOR(x7, LOAD32_LE(m + 28));
+        x8  = XOR(x8, LOAD32_LE(m + 32));
+        x9  = XOR(x9, LOAD32_LE(m + 36));
+        x10 = XOR(x10, LOAD32_LE(m + 40));
+        x11 = XOR(x11, LOAD32_LE(m + 44));
+        x12 = XOR(x12, LOAD32_LE(m + 48));
+        x13 = XOR(x13, LOAD32_LE(m + 52));
+        x14 = XOR(x14, LOAD32_LE(m + 56));
+        x15 = XOR(x15, LOAD32_LE(m + 60));
+
+        j12 = PLUSONE(j12);
+        /* LCOV_EXCL_START */
+        if (!j12) {
+            j13 = PLUSONE(j13);
+        }
+        /* LCOV_EXCL_STOP */
+
+        STORE32_LE(c + 0, x0);
+        STORE32_LE(c + 4, x1);
+        STORE32_LE(c + 8, x2);
+        STORE32_LE(c + 12, x3);
+        STORE32_LE(c + 16, x4);
+        STORE32_LE(c + 20, x5);
+        STORE32_LE(c + 24, x6);
+        STORE32_LE(c + 28, x7);
+        STORE32_LE(c + 32, x8);
+        STORE32_LE(c + 36, x9);
+        STORE32_LE(c + 40, x10);
+        STORE32_LE(c + 44, x11);
+        STORE32_LE(c + 48, x12);
+        STORE32_LE(c + 52, x13);
+        STORE32_LE(c + 56, x14);
+        STORE32_LE(c + 60, x15);
+
+        if (bytes <= 64) {
+            if (bytes < 64) {
+                for (i = 0; i < (unsigned int) bytes; ++i) {
+                    ctarget[i] = c[i]; /* ctarget cannot be NULL */
+                }
+            }
+            ctx->input[12] = j12;
+            ctx->input[13] = j13;
+
+            return;
+        }
+        bytes -= 64;
+        c += 64;
+        m += 64;
+    }
+}
+
+void chacha20_encrypt_msg(void *msg, size_t len, const void *nonce, const void *key)
+{
+    struct chacha_ctx ctx;
+
+    chacha_keysetup(&ctx, key);
+    chacha_ivsetup(&ctx, nonce, NULL);
+       chacha20_encrypt_bytes(&ctx, msg, msg, len);
+}

diff --git a/chacha20.h b/chacha20.h
new file mode 100644 (file)
index 0000000..5519ef8
--- /dev/null
+++ b/chacha20.h
@@ -0,0 +1,13 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * Copyright (C) 2022 Felix Fietkau <nbd@nbd.name>
+ */
+#ifndef __UNETD_CHACHA20_H
+#define __UNETD_CHACHA20_H
+
+#define CHACHA20_NONCE_SIZE    8
+#define CHACHA20_KEY_SIZE      32
+
+void chacha20_encrypt_msg(void *msg, size_t len, const void *nonce, const void *key);
+
+#endif

diff --git a/siphash.h b/siphash.h
index ff76d9c57891e7758a364ff7590412b593327494..233d33aed21340729618fa4b9a83a6a73000bd52 100644 (file)
--- a/siphash.h
+++ b/siphash.h
@@ -16,29 +16,13 @@
 #include <stdint.h>
 #include <stdlib.h>
 #include <stdbool.h>
-#include <libubox/utils.h>
+#include "utils.h"
 
 #define SIPHASH_ALIGNMENT __alignof__(uint64_t)
 typedef struct {
        uint64_t key[2];
 } siphash_key_t;
 
-static inline uint16_t get_unaligned_le16(const uint8_t *p)
-{
-       return p[0] | p[1] << 8;
-}
-
-static inline uint32_t get_unaligned_le32(const uint8_t *p)
-{
-       return p[0] | p[1] << 8 | p[2] << 16 | p[3] << 24;
-}
-
-static inline uint64_t get_unaligned_le64(const uint8_t *p)
-{
-       return (uint64_t)get_unaligned_le32(p + 4) << 32 |
-              get_unaligned_le32(p);
-}
-
 static inline bool siphash_key_is_zero(const siphash_key_t *key)
 {
        return !(key->key[0] | key->key[1]);
@@ -54,4 +38,12 @@ static inline void siphash_to_le64(void *dest, const void *data, size_t len,
        *(uint64_t *)dest = cpu_to_le64(hash);
 }
 
+static inline void siphash_to_be64(void *dest, const void *data, size_t len,
+                                  const siphash_key_t *key)
+{
+       uint64_t hash = siphash(data, len, key);
+
+       *(uint64_t *)dest = cpu_to_be64(hash);
+}
+
 #endif /* _LINUX_SIPHASH_H */

diff --git a/utils.h b/utils.h
index 077080783c314af51b09fee04fac8295b12e2293..ce8943c49088b4d96d04e514eab23345f3d3bb65 100644 (file)
--- a/utils.h
+++ b/utils.h
@@ -5,7 +5,9 @@
 #ifndef __UNETD_UTILS_H
 #define __UNETD_UTILS_H
 
+#include <string.h>
 #include <netinet/in.h>
+#include <libubox/utils.h>
 
 struct nl_msg;
 
@@ -84,6 +86,38 @@ static inline void bitmask_set_val(uint32_t *mask, unsigned int i, bool val)
                bitmask_clear(mask, i);
 }
 
+static inline uint16_t get_unaligned_be16(const uint8_t *p)
+{
+       return p[1] | p[0] << 8;
+}
+
+static inline uint32_t get_unaligned_be32(const uint8_t *p)
+{
+       return p[3] | p[2] << 8 | p[1] << 16 | p[0] << 24;
+}
+
+static inline uint64_t get_unaligned_be64(const uint8_t *p)
+{
+       return (uint64_t)get_unaligned_be32(p) << 32 |
+              get_unaligned_be32(p + 4);
+}
+
+static inline uint16_t get_unaligned_le16(const uint8_t *p)
+{
+       return p[0] | p[1] << 8;
+}
+
+static inline uint32_t get_unaligned_le32(const uint8_t *p)
+{
+       return p[0] | p[1] << 8 | p[2] << 16 | p[3] << 24;
+}
+
+static inline uint64_t get_unaligned_le64(const uint8_t *p)
+{
+       return (uint64_t)get_unaligned_le32(p + 4) << 32 |
+              get_unaligned_le32(p);
+}
+
 int rtnl_init(void);
 int rtnl_call(struct nl_msg *msg);