Change H264 encoder from baseline to main

Add a `struct nvnc* nvnc_open_from_fd(int fd)` function which takes an
existing connection-based socket file descriptor bound by the library
user or a parent process and just calls listen() on it, as an
alternative to letting neatvnc handle socket configuration.

.gitignore

@@ -8,3 +8,4 @@ build
 experiments
 subprojects
 sandbox
+.vscode

FUNDING.yml

@@ -1 +1,2 @@
+github: any1
 patreon: andriyngvason

README.md

@@ -18,6 +18,9 @@ neat.
  * gnutls (optional)
  * libdrm (optional)
  * libturbojpeg (optional)
+ * nettle (optional)
+ * hogweed (optional)
+ * gmp (optional)
  * pixman
  * zlib
 

bench/zrle-bench.c

@@ -71,10 +71,10 @@ static int run_benchmark(const char *image)
 	z_stream zs = { 0 };
 
 	deflateInit2(&zs, /* compression level: */ 1,
-			  /*            method: */ Z_DEFLATED,
-			  /*       window bits: */ 15,
-			  /*         mem level: */ 9,
-			  /*          strategy: */ Z_DEFAULT_STRATEGY);
+			/*            method: */ Z_DEFLATED,
+			/*       window bits: */ 15,
+			/*         mem level: */ 9,
+			/*          strategy: */ Z_DEFAULT_STRATEGY);
 
 	void *dummy = malloc(stride * height * 4);
 	if (!dummy)
@@ -86,7 +86,7 @@ static int run_benchmark(const char *image)
 
 	uint64_t end_time = gettime_us(CLOCK_PROCESS_CPUTIME_ID);
 	printf("memcpy baseline for %s took %"PRIu64" micro seconds\n", image,
-	       end_time - start_time);
+			end_time - start_time);
 
 	free(dummy);
 
@@ -95,7 +95,7 @@ static int run_benchmark(const char *image)
 
 	end_time = gettime_us(CLOCK_PROCESS_CPUTIME_ID);
 	printf("Encoding %s took %"PRIu64" micro seconds\n", image,
-	       end_time - start_time);
+			end_time - start_time);
 
 	double orig_size = stride * height * 4;
 	double compressed_size = frame.len;

examples/draw.c

@@ -198,7 +198,7 @@ static void composite_dot(struct draw *draw, uint32_t* image,
 		}
 
 	pixman_region_init_rect(damage, start.x, start.y,
-	                        stop.x - start.x, stop.y - start.y);
+			stop.x - start.x, stop.y - start.y);
 }
 
 static void draw_dot(struct draw *draw, struct coord coord, int radius,
@@ -219,7 +219,7 @@ static void draw_dot(struct draw *draw, struct coord coord, int radius,
 }
 
 static void on_pointer_event(struct nvnc_client* client, uint16_t x, uint16_t y,
-                             enum nvnc_button_mask buttons)
+		enum nvnc_button_mask buttons)
 {
 	if (!(buttons & NVNC_BUTTON_LEFT))
 		return;

include/base64.h

@@ -0,0 +1,25 @@
+/* Copyright (c) 2023 Andri Yngvason
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
+ * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+ * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+ * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+ * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#pragma once
+
+#include <stdint.h>
+#include <unistd.h>
+
+#define BASE64_ENCODED_SIZE(x) ((((x) + 2) / 3) * 4 + 1)
+#define BASE64_DECODED_MAX_SIZE(x) ((((x) + 3) / 4) * 3)
+
+void base64_encode(char* dst, const uint8_t* src, size_t src_len);
+ssize_t base64_decode(uint8_t* dst, const char* src);

include/common.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 - 2020 Andri Yngvason
+ * Copyright (c) 2019 - 2024 Andri Yngvason
  *
  * Permission to use, copy, modify, and/or distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -26,6 +26,10 @@
 #include "neatvnc.h"
 #include "config.h"
 
+#ifdef HAVE_CRYPTO
+#include "crypto.h"
+#endif
+
 #ifdef ENABLE_TLS
 #include <gnutls/gnutls.h>
 #endif
@@ -43,6 +47,13 @@ enum nvnc_client_state {
 	VNC_CLIENT_STATE_WAITING_FOR_VENCRYPT_VERSION,
 	VNC_CLIENT_STATE_WAITING_FOR_VENCRYPT_SUBTYPE,
 	VNC_CLIENT_STATE_WAITING_FOR_VENCRYPT_PLAIN_AUTH,
+#endif
+#ifdef HAVE_CRYPTO
+	VNC_CLIENT_STATE_WAITING_FOR_APPLE_DH_RESPONSE,
+	VNC_CLIENT_STATE_WAITING_FOR_RSA_AES_PUBLIC_KEY,
+	VNC_CLIENT_STATE_WAITING_FOR_RSA_AES_CHALLENGE,
+	VNC_CLIENT_STATE_WAITING_FOR_RSA_AES_CLIENT_HASH,
+	VNC_CLIENT_STATE_WAITING_FOR_RSA_AES_CREDENTIALS,
 #endif
 	VNC_CLIENT_STATE_WAITING_FOR_INIT,
 	VNC_CLIENT_STATE_READY,
@@ -53,6 +64,9 @@ struct stream;
 struct aml_handler;
 struct aml_idle;
 struct nvnc_display;
+struct crypto_key;
+struct crypto_rsa_pub_key;
+struct crypto_rsa_priv_key;
 
 struct nvnc_common {
 	void* userdata;
@@ -69,7 +83,6 @@ struct nvnc_client {
 	struct nvnc_common common;
 	int ref;
 	struct stream* net_stream;
-	char hostname[256];
 	char username[256];
 	struct nvnc* server;
 	enum nvnc_client_state state;
@@ -89,10 +102,27 @@ struct nvnc_client {
 	uint32_t known_width;
 	uint32_t known_height;
 	struct cut_text cut_text;
-	bool is_qemu_key_ext_notified;
+	bool is_ext_notified;
 	struct encoder* encoder;
+	struct encoder* zrle_encoder;
+	struct encoder* tight_encoder;
 	uint32_t cursor_seq;
 	int quality;
+	bool formats_changed;
+	enum nvnc_keyboard_led_state led_state;
+	enum nvnc_keyboard_led_state pending_led_state;
+
+#ifdef HAVE_CRYPTO
+	struct crypto_key* apple_dh_secret;
+
+	struct {
+		enum crypto_hash_type hash_type;
+		enum crypto_cipher_type cipher_type;
+		size_t challenge_len;
+		uint8_t challenge[32];
+		struct crypto_rsa_pub_key* pub;
+	} rsa;
+#endif
 };
 
 LIST_HEAD(nvnc_client_list, nvnc_client);
@@ -101,6 +131,7 @@ enum nvnc__socket_type {
 	NVNC__SOCKET_TCP,
 	NVNC__SOCKET_UNIX,
 	NVNC__SOCKET_WEBSOCKET,
+	NVNC__SOCKET_FROM_FD,
 };
 
 struct nvnc {
@@ -126,10 +157,17 @@ struct nvnc {
 	} cursor;
 	uint32_t cursor_seq;
 
-#ifdef ENABLE_TLS
-	gnutls_certificate_credentials_t tls_creds;
+	enum nvnc_auth_flags auth_flags;
 	nvnc_auth_fn auth_fn;
 	void* auth_ud;
+
+#ifdef ENABLE_TLS
+	gnutls_certificate_credentials_t tls_creds;
+#endif
+
+#ifdef HAVE_CRYPTO
+	struct crypto_rsa_pub_key* rsa_pub;
+	struct crypto_rsa_priv_key* rsa_priv;
 #endif
 
 	uint32_t n_damage_clients;

include/crypto.h

@@ -0,0 +1,113 @@
+#pragma once
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <unistd.h>
+
+struct crypto_key;
+struct crypto_cipher;
+struct crypto_hash;
+struct crypto_rsa_pub_key;
+struct crypto_rsa_priv_key;
+struct vec;
+
+enum crypto_cipher_type {
+	CRYPTO_CIPHER_INVALID = 0,
+	CRYPTO_CIPHER_AES128_ECB,
+	CRYPTO_CIPHER_AES_EAX,
+	CRYPTO_CIPHER_AES256_EAX,
+};
+
+enum crypto_hash_type {
+	CRYPTO_HASH_INVALID = 0,
+	CRYPTO_HASH_MD5,
+	CRYPTO_HASH_SHA1,
+	CRYPTO_HASH_SHA256,
+};
+
+struct crypto_data_entry {
+	uint8_t* data;
+	size_t len;
+};
+
+void crypto_dump_base16(const char* msg, const uint8_t* bytes, size_t len);
+void crypto_dump_base64(const char* msg, const uint8_t* bytes, size_t len);
+
+void crypto_random(uint8_t* dst, size_t len);
+
+// Key generation
+struct crypto_key* crypto_key_new(int g, const uint8_t *p, uint32_t p_len,
+		const uint8_t* q, uint32_t q_len);
+void crypto_key_del(struct crypto_key* key);
+
+int crypto_key_g(const struct crypto_key* key);
+uint32_t crypto_key_p(const struct crypto_key* key, uint8_t* dst,
+		uint32_t dst_size);
+uint32_t crypto_key_q(const struct crypto_key* key, uint8_t* dst,
+		uint32_t dst_size);
+
+struct crypto_key* crypto_keygen(void);
+
+// Diffie-Hellman
+struct crypto_key* crypto_derive_public_key(const struct crypto_key* priv);
+struct crypto_key* crypto_derive_shared_secret(
+		const struct crypto_key* own_secret,
+		const struct crypto_key* remote_public_key);
+
+// Ciphers
+struct crypto_cipher* crypto_cipher_new(const uint8_t* enc_key,
+		const uint8_t* dec_key, enum crypto_cipher_type type);
+void crypto_cipher_del(struct crypto_cipher* self);
+
+bool crypto_cipher_encrypt(struct crypto_cipher* self, struct vec* dst,
+		uint8_t* mac, const uint8_t* src, size_t len,
+		const uint8_t* ad, size_t ad_len);
+ssize_t crypto_cipher_decrypt(struct crypto_cipher* self, uint8_t* dst,
+		uint8_t* mac, const uint8_t* src, size_t len,
+		const uint8_t* ad, size_t ad_len);
+
+// Hashing
+struct crypto_hash* crypto_hash_new(enum crypto_hash_type type);
+void crypto_hash_del(struct crypto_hash* self);
+
+void crypto_hash_append(struct crypto_hash* self, const uint8_t* src,
+		size_t len);
+void crypto_hash_digest(struct crypto_hash* self, uint8_t* dst,
+		size_t len);
+
+void crypto_hash_one(uint8_t* dst, size_t dst_len, enum crypto_hash_type type,
+		const uint8_t* src, size_t src_len);
+void crypto_hash_many(uint8_t* dst, size_t dst_len, enum crypto_hash_type type,
+		const struct crypto_data_entry *src);
+
+// RSA
+struct crypto_rsa_pub_key* crypto_rsa_pub_key_new(void);
+void crypto_rsa_pub_key_del(struct crypto_rsa_pub_key*);
+
+// Returns length in bytes
+size_t crypto_rsa_pub_key_length(const struct crypto_rsa_pub_key* key);
+
+struct crypto_rsa_pub_key* crypto_rsa_pub_key_import(const uint8_t* modulus,
+		const uint8_t* exponent, size_t size);
+
+void crypto_rsa_pub_key_modulus(const struct crypto_rsa_pub_key* key,
+		uint8_t* dst, size_t dst_size);
+void crypto_rsa_pub_key_exponent(const struct crypto_rsa_pub_key* key,
+		uint8_t* dst, size_t dst_size);
+
+bool crypto_rsa_priv_key_import_pkcs1_der(struct crypto_rsa_priv_key* priv,
+		struct crypto_rsa_pub_key* pub, const uint8_t* key,
+		size_t size);
+
+bool crypto_rsa_priv_key_load(struct crypto_rsa_priv_key* priv,
+		struct crypto_rsa_pub_key* pub, const char* path);
+
+struct crypto_rsa_priv_key *crypto_rsa_priv_key_new(void);
+void crypto_rsa_priv_key_del(struct crypto_rsa_priv_key*);
+
+bool crypto_rsa_keygen(struct crypto_rsa_pub_key*, struct crypto_rsa_priv_key*);
+
+ssize_t crypto_rsa_encrypt(struct crypto_rsa_pub_key* pub, uint8_t* dst,
+		size_t dst_size, const uint8_t* src, size_t src_size);
+ssize_t crypto_rsa_decrypt(struct crypto_rsa_priv_key* priv, uint8_t* dst,
+		size_t dst_size, const uint8_t* src, size_t src_size);

include/damage-refinery.h

@@ -20,8 +20,10 @@
 
 struct pixman_region16;
 struct nvnc_fb;
+struct XXH3_state_s;
 
 struct damage_refinery {
+	struct XXH3_state_s* state;
 	uint32_t* hashes;
 	uint32_t width;
 	uint32_t height;

include/h264-encoder.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021 - 2022 Andri Yngvason
+ * Copyright (c) 2021 - 2024 Andri Yngvason
  *
  * Permission to use, copy, modify, and/or distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -17,13 +17,28 @@
 
 #include <stdint.h>
 #include <unistd.h>
+#include <stdbool.h>
 
-struct h264_encoder;
 struct nvnc_fb;
+struct h264_encoder;
 
 typedef void (*h264_encoder_packet_handler_fn)(const void* payload, size_t size,
 		uint64_t pts, void* userdata);
 
+struct h264_encoder_impl {
+	struct h264_encoder* (*create)(uint32_t width, uint32_t height,
+			uint32_t format, int quality);
+	void (*destroy)(struct h264_encoder*);
+	void (*feed)(struct h264_encoder*, struct nvnc_fb*);
+};
+
+struct h264_encoder {
+	struct h264_encoder_impl *impl;
+	h264_encoder_packet_handler_fn on_packet_ready;
+	void* userdata;
+	bool next_frame_should_be_keyframe;
+};
+
 struct h264_encoder* h264_encoder_create(uint32_t width, uint32_t height,
 		uint32_t format, int quality);
 

include/murmurhash.h

@@ -1,48 +0,0 @@
-/*
- * The MIT License (MIT)
- * 
- * Copyright (c) 2014 Joseph Werle
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- * 
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-#ifndef MURMURHASH_H
-#define MURMURHASH_H 1
-
-#include <stdint.h>
-
-#define MURMURHASH_VERSION "0.0.3"
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-/**
- * Returns a murmur hash of `key' based on `seed'
- * using the MurmurHash3 algorithm
- */
-
-uint32_t
-murmurhash (const char *, uint32_t, uint32_t);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif

include/neatvnc.h

@@ -22,6 +22,7 @@
 #include <stdarg.h>
 #include <stdlib.h>
 #include <assert.h>
+#include <sys/socket.h>
 
 #define NVNC_NO_PTS UINT64_MAX
 
@@ -85,6 +86,12 @@ enum nvnc_transform {
 	NVNC_TRANSFORM_FLIPPED_270 = 7,
 };
 
+enum nvnc_keyboard_led_state {
+	NVNC_KEYBOARD_LED_SCROLL_LOCK = 1 << 0,
+	NVNC_KEYBOARD_LED_NUM_LOCK = 1 << 1,
+	NVNC_KEYBOARD_LED_CAPS_LOCK = 1 << 2,
+};
+
 enum nvnc_log_level {
 	NVNC_LOG_PANIC = 0,
 	NVNC_LOG_ERROR = 1,
@@ -94,6 +101,11 @@ enum nvnc_log_level {
 	NVNC_LOG_TRACE = 5,
 };
 
+enum nvnc_auth_flags {
+	NVNC_AUTH_REQUIRE_AUTH = 1 << 0,
+	NVNC_AUTH_REQUIRE_ENCRYPTION = 1 << 1,
+};
+
 struct nvnc_log_data {
 	enum nvnc_log_level level;
 	const char* file;
@@ -126,6 +138,7 @@ extern const char nvnc_version[];
 struct nvnc* nvnc_open(const char* addr, uint16_t port);
 struct nvnc* nvnc_open_unix(const char *addr);
 struct nvnc* nvnc_open_websocket(const char* addr, uint16_t port);
+struct nvnc* nvnc_open_from_fd(int fd);
 void nvnc_close(struct nvnc* self);
 
 void nvnc_add_display(struct nvnc*, struct nvnc_display*);
@@ -136,13 +149,17 @@ void* nvnc_get_userdata(const void* self);
 
 struct nvnc* nvnc_client_get_server(const struct nvnc_client* client);
 bool nvnc_client_supports_cursor(const struct nvnc_client* client);
-const char* nvnc_client_get_hostname(const struct nvnc_client* client);
+int nvnc_client_get_address(const struct nvnc_client* client,
+		struct sockaddr* restrict addr, socklen_t* restrict addrlen);
 const char* nvnc_client_get_auth_username(const struct nvnc_client* client);
 
 struct nvnc_client* nvnc_client_first(struct nvnc* self);
 struct nvnc_client* nvnc_client_next(struct nvnc_client* client);
 void nvnc_client_close(struct nvnc_client* client);
 
+void nvnc_client_set_led_state(struct nvnc_client*,
+		enum nvnc_keyboard_led_state);
+
 void nvnc_set_name(struct nvnc* self, const char* name);
 
 void nvnc_set_key_fn(struct nvnc* self, nvnc_key_fn);
@@ -155,8 +172,11 @@ void nvnc_set_cut_text_fn(struct nvnc*, nvnc_cut_text_fn fn);
 void nvnc_set_desktop_layout_fn(struct nvnc* self, nvnc_desktop_layout_fn);
 
 bool nvnc_has_auth(void);
-int nvnc_enable_auth(struct nvnc* self, const char* privkey_path,
-                     const char* cert_path, nvnc_auth_fn, void* userdata);
+int nvnc_enable_auth(struct nvnc* self, enum nvnc_auth_flags flags,
+		nvnc_auth_fn, void* userdata);
+int nvnc_set_tls_creds(struct nvnc* self, const char* privkey_path,
+                     const char* cert_path);
+int nvnc_set_rsa_creds(struct nvnc* self, const char* private_key_path);
 
 struct nvnc_fb* nvnc_fb_new(uint16_t width, uint16_t height,
                             uint32_t fourcc_format, uint16_t stride);
@@ -227,6 +247,9 @@ void nvnc_set_cursor(struct nvnc*, struct nvnc_fb*, uint16_t width,
 		     uint16_t height, uint16_t hotspot_x, uint16_t hotspot_y,
 		     bool is_damaged);
 
+void nvnc_default_logger(const struct nvnc_log_data* meta, const char* message);
+
 void nvnc_set_log_fn(nvnc_log_fn);
+void nvnc_set_log_fn_thread_local(nvnc_log_fn fn);
 void nvnc_set_log_level(enum nvnc_log_level);
 void nvnc__log(const struct nvnc_log_data*, const char* fmt, ...);

include/pixels.h

@@ -22,10 +22,11 @@
 #include <stdbool.h>
 
 struct rfb_pixel_format;
+struct rfb_set_colour_map_entries_msg;
 
-void pixel32_to_cpixel(uint8_t* restrict dst,
+void pixel_to_cpixel(uint8_t* restrict dst,
                        const struct rfb_pixel_format* dst_fmt,
-                       const uint32_t* restrict src,
+                       const uint8_t* restrict src,
                        const struct rfb_pixel_format* src_fmt,
                        size_t bytes_per_cpixel, size_t len);
 
@@ -41,3 +42,4 @@ bool extract_alpha_mask(uint8_t* dst, const void* src, uint32_t format,
 
 const char* drm_format_to_string(uint32_t fmt);
 const char* rfb_pixfmt_to_string(const struct rfb_pixel_format* fmt);
+void make_rgb332_pal8_map(struct rfb_set_colour_map_entries_msg* msg);

include/rfb-proto.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 - 2022 Andri Yngvason
+ * Copyright (c) 2019 - 2024 Andri Yngvason
  *
  * Permission to use, copy, modify, and/or distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -29,8 +29,11 @@ enum rfb_security_type {
 	RFB_SECURITY_TYPE_INVALID = 0,
 	RFB_SECURITY_TYPE_NONE = 1,
 	RFB_SECURITY_TYPE_VNC_AUTH = 2,
+	RFB_SECURITY_TYPE_RSA_AES = 5,
 	RFB_SECURITY_TYPE_TIGHT = 16,
 	RFB_SECURITY_TYPE_VENCRYPT = 19,
+	RFB_SECURITY_TYPE_APPLE_DH = 30,
+	RFB_SECURITY_TYPE_RSA_AES256 = 129,
 };
 
 enum rfb_security_handshake_result {
@@ -66,9 +69,11 @@ enum rfb_encodings {
 	RFB_ENCODING_CURSOR = -239,
 	RFB_ENCODING_DESKTOPSIZE = -223,
 	RFB_ENCODING_QEMU_EXT_KEY_EVENT = -258,
+	RFB_ENCODING_QEMU_LED_STATE = -261,
 	RFB_ENCODING_EXTENDEDDESKTOPSIZE = -308,
 	RFB_ENCODING_PTS = -1000,
 	RFB_ENCODING_NTP = -1001,
+	RFB_ENCODING_VMWARE_LED_STATE = 0x574d5668,
 };
 
 #define RFB_ENCODING_JPEG_HIGHQ -23
@@ -106,9 +111,21 @@ enum rfb_resize_status {
 	RFB_RESIZE_STATUS_REQUEST_FORWARDED = 4,
 };
 
+enum rfb_rsa_aes_cred_subtype {
+	RFB_RSA_AES_CRED_SUBTYPE_USER_AND_PASS = 1,
+	RFB_RSA_AES_CRED_SUBTYPE_ONLY_PASS = 2,
+};
+
+// This is the same for both qemu and vmware extensions
+enum rfb_led_state {
+	RFB_LED_STATE_SCROLL_LOCK = 1 << 0,
+	RFB_LED_STATE_NUM_LOCK = 1 << 1,
+	RFB_LED_STATE_CAPS_LOCK = 1 << 2,
+};
+
 struct rfb_security_types_msg {
 	uint8_t n;
-	uint8_t types[1];
+	uint8_t types[0];
 } RFB_PACKED;
 
 struct rfb_error_reason {
@@ -237,3 +254,36 @@ struct rfb_ntp_msg {
 	uint8_t padding[3];
 	uint32_t t0, t1, t2, t3;
 } RFB_PACKED;
+
+struct rfb_apple_dh_server_msg {
+	uint16_t generator;
+	uint16_t key_size;
+	uint8_t modulus_and_key[0];
+} RFB_PACKED;
+
+struct rfb_apple_dh_client_msg {
+	uint8_t encrypted_credentials[128];
+	uint8_t public_key[0];
+} RFB_PACKED;
+
+struct rfb_rsa_aes_pub_key_msg {
+	uint32_t length;
+	uint8_t modulus_and_exponent[0];
+} RFB_PACKED;
+
+struct rfb_rsa_aes_challenge_msg {
+	uint16_t length;
+	uint8_t challenge[0];
+} RFB_PACKED;
+
+struct rfb_colour_map_entry {
+	uint16_t r, g, b;
+} RFB_PACKED;
+
+struct rfb_set_colour_map_entries_msg {
+	uint8_t type;
+	uint8_t padding;
+	uint16_t first_colour;
+	uint16_t n_colours;
+	struct rfb_colour_map_entry colours[0];
+} RFB_PACKED;

include/stream-tcp.h

@@ -0,0 +1,36 @@
+/*
+ * Copyright (c) 2023 Andri Yngvason
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
+ * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+ * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+ * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+ * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#pragma once
+
+#include "stream.h"
+
+#include <unistd.h>
+
+struct stream;
+
+int stream_tcp_init(struct stream* self, int fd, stream_event_fn on_event,
+		void* userdata);
+int stream_tcp_close(struct stream* self);
+void stream_tcp_destroy(struct stream* self);
+ssize_t stream_tcp_read(struct stream* self, void* dst, size_t size);
+int stream_tcp_send(struct stream* self, struct rcbuf* payload,
+                stream_req_fn on_done, void* userdata);
+int stream_tcp_send_first(struct stream* self, struct rcbuf* payload);
+void stream_tcp_exec_and_send(struct stream* self,
+		stream_exec_fn exec_fn, void* userdata);
+int stream_tcp_install_cipher(struct stream* self,
+		struct crypto_cipher* cipher);

include/stream.h

@@ -19,6 +19,11 @@
 #include "config.h"
 #include "sys/queue.h"
 #include "rcbuf.h"
+#include "vec.h"
+
+#ifdef HAVE_CRYPTO
+#include "crypto.h"
+#endif
 
 #include <stdint.h>
 #include <stdbool.h>
@@ -45,6 +50,7 @@ enum stream_event {
 };
 
 struct stream;
+struct crypto_cipher;
 
 typedef void (*stream_event_fn)(struct stream*, enum stream_event);
 typedef void (*stream_req_fn)(void*, enum stream_req_status);
@@ -86,6 +92,9 @@ struct stream {
 	uint32_t bytes_received;
 
 	bool cork;
+
+	struct crypto_cipher* cipher;
+	struct vec tmp_buf;
 };
 
 #ifdef ENABLE_WEBSOCKET
@@ -108,3 +117,9 @@ void stream_exec_and_send(struct stream* self, stream_exec_fn, void* userdata);
 #ifdef ENABLE_TLS
 int stream_upgrade_to_tls(struct stream* self, void* context);
 #endif
+
+#ifdef HAVE_CRYPTO
+int stream_upgrade_to_rsa_eas(struct stream* base,
+		enum crypto_cipher_type cipher_type,
+		const uint8_t* enc_key, const uint8_t* dec_key);
+#endif

include/websocket.h

@@ -1,3 +1,19 @@
+/*
+ * Copyright (c) 2023 Andri Yngvason
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
+ * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+ * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+ * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+ * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
 #pragma once
 
 #include <stdint.h>

meson.build

@@ -1,7 +1,7 @@
 project(
 	'neatvnc',
 	'c',
-	version: '0.6.0',
+	version: '0.9-dev',
 	license: 'ISC',
 	default_options: [
 		'c_std=gnu11',
@@ -13,7 +13,6 @@ buildtype = get_option('buildtype')
 host_system = host_machine.system()
 
 c_args = [
-	'-DPROJECT_VERSION="@0@"'.format(meson.project_version()),
 	'-D_GNU_SOURCE',
 	'-fvisibility=hidden',
 	'-DAML_UNSTABLE_API=1',
@@ -27,17 +26,20 @@ if buildtype != 'debug' and buildtype != 'debugoptimized'
 	c_args += '-DNDEBUG'
 endif
 
+version = '"@0@"'.format(meson.project_version())
 git = find_program('git', native: true, required: false)
 if git.found()
-	git_describe = run_command([git, 'describe', '--tags', '--long'])
+	git_commit = run_command([git, 'rev-parse', '--short', 'HEAD'])
 	git_branch = run_command([git, 'rev-parse', '--abbrev-ref', 'HEAD'])
-	if git_describe.returncode() == 0 and git_branch.returncode() == 0
-		c_args += '-DGIT_VERSION="@0@ (@1@)"'.format(
-			git_describe.stdout().strip(),
+	if git_commit.returncode() == 0 and git_branch.returncode() == 0
+		version = '"v@0@-@1@ (@2@)"'.format(
+			meson.project_version(),
+			git_commit.stdout().strip(),
 			git_branch.stdout().strip(),
 		)
 	endif
 endif
+add_project_arguments('-DPROJECT_VERSION=@0@'.format(version), language: 'c')
 
 libdrm_inc = dependency('libdrm').partial_dependency(compile_args: true)
 
@@ -51,6 +53,8 @@ pixman = dependency('pixman-1')
 libturbojpeg = dependency('libturbojpeg', required: get_option('jpeg'))
 gnutls = dependency('gnutls', required: get_option('tls'))
 nettle = dependency('nettle', required: get_option('nettle'))
+hogweed = dependency('hogweed', required: get_option('nettle'))
+gmp = dependency('gmp', required: get_option('nettle'))
 zlib = dependency('zlib')
 gbm = dependency('gbm', required: get_option('gbm'))
 libdrm = dependency('libdrm', required: get_option('h264'))
@@ -89,10 +93,10 @@ sources = [
 	'src/resampler.c',
 	'src/transform-util.c',
 	'src/damage-refinery.c',
-	'src/murmurhash.c',
 	'src/encoder.c',
 	'src/cursor.c',
 	'src/logging.c',
+	'src/base64.c',
 ]
 
 dependencies = [
@@ -118,9 +122,11 @@ if gnutls.found()
 	config.set('ENABLE_TLS', true)
 endif
 
-if nettle.found()
-	dependencies += nettle
+if nettle.found() and hogweed.found() and gmp.found()
+	dependencies += [ nettle, hogweed, gmp ]
 	enable_websocket = true
+	config.set('HAVE_CRYPTO', true)
+	sources += ['src/crypto-nettle.c', 'src/stream-rsa-aes.c']
 endif
 
 if host_system == 'linux' and get_option('systemtap') and cc.has_header('sys/sdt.h')
@@ -132,13 +138,26 @@ if gbm.found()
 	config.set('HAVE_GBM', true)
 endif
 
-if gbm.found() and libdrm.found() and libavcodec.found() and libavfilter.found() and libavutil.found()
-	sources += [ 'src/h264-encoder.c', 'src/open-h264.c' ]
+have_ffmpeg = gbm.found() and libdrm.found() and libavcodec.found() and libavfilter.found() and libavutil.found()
+have_v4l2 = gbm.found() and libdrm.found() and cc.check_header('linux/videodev2.h')
+
+if have_ffmpeg
+	sources += [ 'src/h264-encoder-ffmpeg-impl.c' ]
 	dependencies += [libdrm, libavcodec, libavfilter, libavutil]
-	config.set('ENABLE_OPEN_H264', true)
+	config.set('HAVE_FFMPEG', true)
 	config.set('HAVE_LIBAVUTIL', true)
 endif
 
+if have_v4l2
+	sources += [ 'src/h264-encoder-v4l2m2m-impl.c' ]
+	config.set('HAVE_V4L2', true)
+endif
+
+if have_ffmpeg or have_v4l2
+	sources += [ 'src/h264-encoder.c', 'src/open-h264.c' ]
+	config.set('ENABLE_OPEN_H264', true)
+endif
+
 if enable_websocket
 	sources += [
 		'src/ws-handshake.c',
@@ -149,6 +168,13 @@ if enable_websocket
 	config.set('ENABLE_WEBSOCKET', true)
 endif
 
+if get_option('experimental')
+	if buildtype == 'release'
+		warning('Experimental features enabled in release build')
+	endif
+	config.set('ENABLE_EXPERIMENTAL', true)
+endif
+
 configure_file(
 	output: 'config.h',
 	configuration: config,

meson_options.txt

@@ -7,3 +7,4 @@ option('nettle', type: 'feature', value: 'auto', description: 'Enable nettle low
 option('systemtap', type: 'boolean', value: false, description: 'Enable tracing using sdt')
 option('gbm', type: 'feature', value: 'auto', description: 'Enable GBM integration')
 option('h264', type: 'feature', value: 'auto', description: 'Enable open h264 encoding')
+option('experimental', type: 'boolean', value: false, description: 'Enable experimental features')

src/base64.c

@@ -0,0 +1,155 @@
+/* Copyright (c) 2023 Andri Yngvason
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
+ * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+ * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+ * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+ * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "base64.h"
+
+#include <unistd.h>
+#include <stdint.h>
+#include <stdbool.h>
+#include <stdlib.h>
+#include <string.h>
+
+static const char base64_enc_lut[] =
+	"ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
+
+static const uint8_t base64_validation_lut[256] = {
+	['A'] = 1, ['B'] = 1, ['C'] = 1, ['D'] = 1,
+	['E'] = 1, ['F'] = 1, ['G'] = 1, ['H'] = 1,
+	['I'] = 1, ['J'] = 1, ['K'] = 1, ['L'] = 1,
+	['M'] = 1, ['N'] = 1, ['O'] = 1, ['P'] = 1,
+	['Q'] = 1, ['R'] = 1, ['S'] = 1, ['T'] = 1,
+	['U'] = 1, ['V'] = 1, ['W'] = 1, ['X'] = 1,
+	['Y'] = 1, ['Z'] = 1, ['a'] = 1, ['b'] = 1,
+	['c'] = 1, ['d'] = 1, ['e'] = 1, ['f'] = 1,
+	['g'] = 1, ['h'] = 1, ['i'] = 1, ['j'] = 1,
+	['k'] = 1, ['l'] = 1, ['m'] = 1, ['n'] = 1,
+	['o'] = 1, ['p'] = 1, ['q'] = 1, ['r'] = 1,
+	['s'] = 1, ['t'] = 1, ['u'] = 1, ['v'] = 1,
+	['w'] = 1, ['x'] = 1, ['y'] = 1, ['z'] = 1,
+	['0'] = 1, ['1'] = 1, ['2'] = 1, ['3'] = 1,
+	['4'] = 1, ['5'] = 1, ['6'] = 1, ['7'] = 1,
+	['8'] = 1, ['9'] = 1, ['+'] = 1, ['/'] = 1,
+	['-'] = 1, ['_'] = 1, ['='] = 1,
+};
+
+static const uint8_t base64_dec_lut[256] = {
+	['A'] = 0x00, ['B'] = 0x01, ['C'] = 0x02, ['D'] = 0x03,
+	['E'] = 0x04, ['F'] = 0x05, ['G'] = 0x06, ['H'] = 0x07,
+	['I'] = 0x08, ['J'] = 0x09, ['K'] = 0x0a, ['L'] = 0x0b,
+	['M'] = 0x0c, ['N'] = 0x0d, ['O'] = 0x0e, ['P'] = 0x0f,
+	['Q'] = 0x10, ['R'] = 0x11, ['S'] = 0x12, ['T'] = 0x13,
+	['U'] = 0x14, ['V'] = 0x15, ['W'] = 0x16, ['X'] = 0x17,
+	['Y'] = 0x18, ['Z'] = 0x19, ['a'] = 0x1a, ['b'] = 0x1b,
+	['c'] = 0x1c, ['d'] = 0x1d, ['e'] = 0x1e, ['f'] = 0x1f,
+	['g'] = 0x20, ['h'] = 0x21, ['i'] = 0x22, ['j'] = 0x23,
+	['k'] = 0x24, ['l'] = 0x25, ['m'] = 0x26, ['n'] = 0x27,
+	['o'] = 0x28, ['p'] = 0x29, ['q'] = 0x2a, ['r'] = 0x2b,
+	['s'] = 0x2c, ['t'] = 0x2d, ['u'] = 0x2e, ['v'] = 0x2f,
+	['w'] = 0x30, ['x'] = 0x31, ['y'] = 0x32, ['z'] = 0x33,
+	['0'] = 0x34, ['1'] = 0x35, ['2'] = 0x36, ['3'] = 0x37,
+	['4'] = 0x38, ['5'] = 0x39, ['6'] = 0x3a, ['7'] = 0x3b,
+	['8'] = 0x3c, ['9'] = 0x3d, ['+'] = 0x3e, ['/'] = 0x3f,
+	['-'] = 0x3e, ['_'] = 0x3f,
+};
+
+void base64_encode(char* dst, const uint8_t* src, size_t src_len)
+{
+	size_t i = 0;
+
+	for (; i < src_len / 3; ++i) {
+		uint32_t tmp = 0;
+		tmp |= (uint32_t)src[i * 3 + 0] << 16;
+		tmp |= (uint32_t)src[i * 3 + 1] << 8;
+		tmp |= (uint32_t)src[i * 3 + 2];
+
+		dst[i * 4 + 0] = base64_enc_lut[tmp >> 18];
+		dst[i * 4 + 1] = base64_enc_lut[(tmp >> 12) & 0x3f];
+		dst[i * 4 + 2] = base64_enc_lut[(tmp >> 6) & 0x3f];
+		dst[i * 4 + 3] = base64_enc_lut[tmp & 0x3f];
+	}
+
+	size_t rem = src_len % 3;
+	if (rem == 0) {
+		dst[i * 4] = '\0';
+		return;
+	}
+
+	uint32_t tmp = 0;
+	for (size_t r = 0; r < rem; ++r) {
+		size_t s = (2 - r) * 8;
+		tmp |= (uint32_t)src[i * 3 + r] << s;
+	}
+
+	size_t di = 0;
+	for (; di < rem + 1; ++di) {
+		size_t s = (3 - di) * 6;
+		dst[i * 4 + di] = base64_enc_lut[(tmp >> s) & 0x3f];
+	}
+
+	for (; di < 4; ++di) {
+		dst[i * 4 + di] = '=';
+	}
+
+	dst[i * 4 + di] = '\0';
+
+}
+
+static bool base64_is_valid(const char* src)
+{
+	for (int i = 0; src[i]; i++)
+		if (!base64_validation_lut[(uint8_t)src[i]])
+			return false;
+	return true;
+}
+
+ssize_t base64_decode(uint8_t* dst, const char* src)
+{
+	if (!base64_is_valid(src))
+		return -1;
+
+	size_t src_len = strcspn(src, "=");
+	size_t i = 0;
+
+	for (; i < src_len / 4; ++i) {
+		uint32_t tmp = 0;
+		tmp |= (uint32_t)base64_dec_lut[(uint8_t)src[i * 4 + 0]] << 18;
+		tmp |= (uint32_t)base64_dec_lut[(uint8_t)src[i * 4 + 1]] << 12;
+		tmp |= (uint32_t)base64_dec_lut[(uint8_t)src[i * 4 + 2]] << 6;
+		tmp |= (uint32_t)base64_dec_lut[(uint8_t)src[i * 4 + 3]];
+
+		dst[i * 3 + 0] = tmp >> 16;
+		dst[i * 3 + 1] = (tmp >> 8) & 0xff;
+		dst[i * 3 + 2] = tmp & 0xff;
+	}
+
+	size_t rem = src_len % 4;
+	if (rem == 0)
+		return i * 3;
+
+	size_t di = 0;
+
+	uint32_t tmp = 0;
+	for (size_t r = 0; r < rem; ++r) {
+		size_t s = (3 - r) * 6;
+		tmp |= (uint32_t)base64_dec_lut[(uint8_t)src[i * 4 + r]] << s;
+	}
+
+	for (; di < (rem * 3) / 4; ++di) {
+		size_t s = (2 - di) * 8;
+		dst[i * 3 + di] = (tmp >> s) & 0xff;
+	}
+
+	return i * 3 + di;
+}

src/crypto-nettle.c

@@ -0,0 +1,739 @@
+#include "crypto.h"
+#include "neatvnc.h"
+#include "vec.h"
+#include "base64.h"
+
+#include <gmp.h>
+#include <nettle/base64.h>
+#include <nettle/base16.h>
+#include <nettle/aes.h>
+#include <nettle/eax.h>
+#include <nettle/md5.h>
+#include <nettle/sha1.h>
+#include <nettle/sha.h>
+#include <nettle/rsa.h>
+
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdbool.h>
+#include <string.h>
+#include <sys/param.h>
+#include <arpa/inet.h>
+
+// TODO: This is linux specific
+#include <sys/random.h>
+
+#define UDIV_UP(a, b) (((a) + (b) - 1) / (b))
+
+struct vec;
+
+struct crypto_key {
+	int g;
+	mpz_t p;
+	mpz_t q;
+};
+
+struct crypto_aes_eax {
+	struct eax_aes128_ctx ctx;
+	uint64_t count[2];
+};
+
+struct crypto_aes256_eax {
+	struct EAX_CTX(struct aes256_ctx) ctx;
+	uint64_t count[2];
+};
+
+struct crypto_cipher {
+	union {
+		struct aes128_ctx aes128_ecb;
+		struct crypto_aes_eax aes_eax;
+		struct crypto_aes256_eax aes256_eax;
+	} enc_ctx;
+
+	union {
+		struct aes128_ctx aes128_ecb;
+		struct crypto_aes_eax aes_eax;
+		struct crypto_aes256_eax aes256_eax;
+	} dec_ctx;
+
+	bool (*encrypt)(struct crypto_cipher*, struct vec* dst, uint8_t* mac,
+			const uint8_t* src, size_t src_len, const uint8_t* ad,
+			size_t ad_len);
+	ssize_t (*decrypt)(struct crypto_cipher*, uint8_t* dst, uint8_t* mac,
+			const uint8_t* src, size_t src_len, const uint8_t* ad,
+			size_t ad_len);
+};
+
+struct crypto_hash {
+	union {
+		struct md5_ctx md5;
+		struct sha1_ctx sha1;
+		struct sha256_ctx sha256;
+	} ctx;
+
+	void (*update)(void* ctx, size_t len, const uint8_t* src);
+	void (*digest)(void* ctx, size_t len, uint8_t* dst);
+};
+
+struct crypto_rsa_pub_key {
+	struct rsa_public_key key;
+};
+
+struct crypto_rsa_priv_key {
+	struct rsa_private_key key;
+};
+
+void crypto_dump_base64(const char* msg, const uint8_t* bytes, size_t len)
+{
+	struct base64_encode_ctx ctx = {};
+	size_t buflen = BASE64_ENCODE_LENGTH(len);
+	char* buffer = malloc(buflen + BASE64_ENCODE_FINAL_LENGTH + 1);
+	assert(buffer);
+	nettle_base64_encode_init(&ctx);
+	size_t count = nettle_base64_encode_update(&ctx, buffer, len, bytes);
+	count += nettle_base64_encode_final(&ctx, buffer + count);
+	buffer[count] = '\0';
+
+	nvnc_log(NVNC_LOG_DEBUG, "%s: %s", msg, buffer);
+	free(buffer);
+}
+
+void crypto_dump_base16(const char* msg, const uint8_t* bytes, size_t len)
+{
+	size_t buflen = BASE16_ENCODE_LENGTH(len);
+	char* buffer = calloc(1, buflen + 1);
+	assert(buffer);
+	nettle_base16_encode_update(buffer, len, bytes);
+
+	nvnc_log(NVNC_LOG_DEBUG, "%s: %s", msg, buffer);
+	free(buffer);
+}
+
+void crypto_random(uint8_t* dst, size_t len)
+{
+	getrandom(dst, len, 0);
+}
+
+static void crypto_import(mpz_t n, const uint8_t* src, size_t len)
+{
+	int order = 1;
+	int unit_size = 1;
+	int endian = 1;
+	int skip_bits = 0;
+
+	mpz_import(n, len, order, unit_size, endian, skip_bits, src);
+}
+
+struct crypto_key *crypto_key_new(int g, const uint8_t* p, uint32_t p_len,
+		const uint8_t* q, uint32_t q_len)
+{
+	struct crypto_key* self = calloc(1, sizeof(*self));
+	if (!self)
+		return NULL;
+
+	self->g = g;
+
+	mpz_init(self->p);
+	crypto_import(self->p, p, p_len);
+
+	mpz_init(self->q);
+	crypto_import(self->q, q, q_len);
+
+	return self;
+}
+
+void crypto_key_del(struct crypto_key* key)
+{
+	if (!key)
+		return;
+	mpz_clear(key->q);
+	mpz_clear(key->p);
+	free(key);
+}
+
+int crypto_key_g(const struct crypto_key* key)
+{
+	return key->g;
+}
+
+static size_t crypto_export(uint8_t* dst, size_t dst_size, const mpz_t n)
+{
+	int order = 1; // msb first
+	int unit_size = 1; // byte
+	int endian = 1; // msb first
+	int skip_bits = 0;
+
+	size_t bitsize = mpz_sizeinbase(n, 2);
+	size_t bytesize = (bitsize + 7) / 8;
+
+	assert(bytesize <= dst_size);
+
+	memset(dst, 0, dst_size);
+	mpz_export(dst + dst_size - bytesize, &bytesize, order, unit_size,
+			endian, skip_bits, n);
+
+	return bytesize;
+}
+
+uint32_t crypto_key_p(const struct crypto_key* key, uint8_t* dst,
+		uint32_t dst_size)
+{
+	return crypto_export(dst, dst_size, key->p);
+}
+
+uint32_t crypto_key_q(const struct crypto_key* key, uint8_t* dst,
+		uint32_t dst_size)
+{
+	return crypto_export(dst, dst_size, key->q);
+}
+
+static void initialise_p(mpz_t p)
+{
+	// RFC 3526, section 3
+	static const char s[] =
+		"FFFFFFFFFFFFFFFFC90FDAA22168C234C4C6628B80DC1CD1"
+		"29024E088A67CC74020BBEA63B139B22514A08798E3404DD"
+		"EF9519B3CD3A431B302B0A6DF25F14374FE1356D6D51C245"
+		"E485B576625E7EC6F44C42E9A637ED6B0BFF5CB6F406B7ED"
+		"EE386BFB5A899FA5AE9F24117C4B1FE649286651ECE45B3D"
+		"C2007CB8A163BF0598DA48361C55D39A69163FA8FD24CF5F"
+		"83655D23DCA3AD961C62F356208552BB9ED529077096966D"
+		"670C354E4ABC9804F1746C08CA18217C32905E462E36CE3B"
+		"E39E772C180E86039B2783A2EC07A28FB5C55DF06F4C52C9"
+		"DE2BCBF6955817183995497CEA956AE515D2261898FA0510"
+		"15728E5A8AACAA68FFFFFFFFFFFFFFFF";
+
+	char buf[256];
+	size_t len = 0;
+	struct base16_decode_ctx ctx;
+	nettle_base16_decode_init(&ctx);
+	nettle_base16_decode_update(&ctx, &len, (uint8_t*)buf, sizeof(s) - 1, s);
+	nettle_base16_decode_final(&ctx);
+	assert(len == sizeof(buf));
+
+	crypto_import(p, (const uint8_t*)buf, sizeof(buf));
+}
+
+static void generate_random(mpz_t n)
+{
+	uint8_t buf[256];
+	getrandom(buf, sizeof(buf), 0);
+	crypto_import(n, buf, sizeof(buf));
+}
+
+struct crypto_key* crypto_keygen(void)
+{
+	struct crypto_key* self = calloc(1, sizeof(*self));
+	if (!self)
+		return NULL;
+
+	self->g = 2;
+
+	mpz_init(self->p);
+	initialise_p(self->p);
+
+	mpz_init(self->q);
+	generate_random(self->q);
+
+	return self;
+}
+
+struct crypto_key* crypto_derive_public_key(const struct crypto_key* priv)
+{
+	struct crypto_key* pub = calloc(1, sizeof(*pub));
+	if (!pub)
+		return NULL;
+
+	pub->g = priv->g;
+	mpz_set(pub->p, priv->p);
+	mpz_init(pub->q);
+
+	mpz_t g;
+	mpz_init(g);
+	mpz_set_ui(g, priv->g);
+
+	mpz_powm_sec(pub->q, g, priv->q, priv->p);
+	mpz_clear(g);
+
+	return pub;
+}
+
+struct crypto_key* crypto_derive_shared_secret(
+		const struct crypto_key* own_secret,
+		const struct crypto_key* remote_public_key)
+{
+	if (own_secret->g != remote_public_key->g) {
+		return NULL;
+	}
+
+	if (mpz_cmp(own_secret->p, remote_public_key->p) != 0) {
+		return NULL;
+	}
+
+	struct crypto_key* shared = calloc(1, sizeof(*shared));
+	if (!shared)
+		return NULL;
+
+	shared->g = own_secret->g;
+	mpz_set(shared->p, own_secret->p);
+
+	mpz_t g;
+	mpz_init(g);
+	mpz_set_ui(g, own_secret->g);
+
+	mpz_powm_sec(shared->q, remote_public_key->q, own_secret->q,
+			own_secret->p);
+	mpz_clear(g);
+
+	return shared;
+}
+
+static bool crypto_cipher_aes128_ecb_encrypt(struct crypto_cipher* self,
+		struct vec* dst, uint8_t* mac, const uint8_t* src,
+		size_t len, const uint8_t* ad, size_t ad_len)
+{
+	vec_reserve(dst, dst->len + len);
+	aes128_encrypt(&self->enc_ctx.aes128_ecb, len, dst->data, src);
+	dst->len = len;
+	return true;
+}
+
+static ssize_t crypto_cipher_aes128_ecb_decrypt(struct crypto_cipher* self,
+		uint8_t* dst, uint8_t* mac, const uint8_t* src, size_t len,
+		const uint8_t* ad, size_t ad_len)
+{
+	aes128_decrypt(&self->dec_ctx.aes128_ecb, len, dst, src);
+	return len;
+}
+
+static struct crypto_cipher* crypto_cipher_new_aes128_ecb(
+		const uint8_t* enc_key, const uint8_t* dec_key)
+{
+	struct crypto_cipher* self = calloc(1, sizeof(*self));
+	if (!self)
+		return NULL;
+
+	if (enc_key)
+		aes128_set_encrypt_key(&self->enc_ctx.aes128_ecb, enc_key);
+
+	if (dec_key)
+		aes128_set_decrypt_key(&self->dec_ctx.aes128_ecb, dec_key);
+
+	self->encrypt = crypto_cipher_aes128_ecb_encrypt;
+	self->decrypt = crypto_cipher_aes128_ecb_decrypt;
+
+	return self;
+}
+
+static void crypto_aes_eax_update_nonce(struct crypto_aes_eax* self)
+{
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+	nettle_eax_aes128_set_nonce(&self->ctx, 16, (const uint8_t*)self->count);
+#else
+	uint64_t c[2];
+	c[0] = __builtin_bswap64(self->count[0]);
+	c[1] = __builtin_bswap64(self->count[1]);
+	nettle_eax_aes128_set_nonce(&self->ctx, 16, (const uint8_t*)c);
+#endif
+
+	if (++self->count[0] == 0)
+		++self->count[1];
+}
+
+static bool crypto_cipher_aes_eax_encrypt(struct crypto_cipher* self,
+		struct vec* dst, uint8_t* mac, const uint8_t* src,
+		size_t src_len, const uint8_t* ad, size_t ad_len)
+{
+	vec_reserve(dst, dst->len + src_len);
+
+	crypto_aes_eax_update_nonce(&self->enc_ctx.aes_eax);
+	nettle_eax_aes128_update(&self->enc_ctx.aes_eax.ctx, ad_len,
+			(uint8_t*)ad);
+	nettle_eax_aes128_encrypt(&self->enc_ctx.aes_eax.ctx, src_len,
+			(uint8_t*)dst->data + dst->len, src);
+	dst->len += src_len;
+
+	nettle_eax_aes128_digest(&self->enc_ctx.aes_eax.ctx, 16, mac);
+
+	return true;
+}
+
+static ssize_t crypto_cipher_aes_eax_decrypt(struct crypto_cipher* self,
+		uint8_t* dst, uint8_t* mac, const uint8_t* src, size_t len,
+		const uint8_t* ad, size_t ad_len)
+{
+	crypto_aes_eax_update_nonce(&self->dec_ctx.aes_eax);
+	nettle_eax_aes128_update(&self->dec_ctx.aes_eax.ctx, ad_len, ad);
+	nettle_eax_aes128_decrypt(&self->dec_ctx.aes_eax.ctx, len, dst, src);
+	nettle_eax_aes128_digest(&self->dec_ctx.aes_eax.ctx, 16, mac);
+	return len;
+}
+
+static struct crypto_cipher* crypto_cipher_new_aes_eax(const uint8_t* enc_key,
+		const uint8_t* dec_key)
+{
+	struct crypto_cipher* self = calloc(1, sizeof(*self));
+	if (!self)
+		return NULL;
+
+	eax_aes128_set_key(&self->enc_ctx.aes_eax.ctx, enc_key);
+	eax_aes128_set_key(&self->dec_ctx.aes_eax.ctx, dec_key);
+
+	self->encrypt = crypto_cipher_aes_eax_encrypt;
+	self->decrypt = crypto_cipher_aes_eax_decrypt;
+
+	return self;
+}
+
+static void crypto_aes256_eax_update_nonce(struct crypto_aes256_eax* self)
+{
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+	EAX_SET_NONCE(&self->ctx, aes256_encrypt, 16, (const uint8_t*)self->count);
+#else
+	uint64_t c[2];
+	c[0] = __builtin_bswap64(self->count[0]);
+	c[1] = __builtin_bswap64(self->count[1]);
+	EAX_SET_NONCE(&self->ctx, aes256_encrypt, 16, (const uint8_t*)c);
+#endif
+
+	if (++self->count[0] == 0)
+		++self->count[1];
+}
+
+static bool crypto_cipher_aes256_eax_encrypt(struct crypto_cipher* self,
+		struct vec* dst, uint8_t* mac, const uint8_t* src,
+		size_t src_len, const uint8_t* ad, size_t ad_len)
+{
+	vec_reserve(dst, dst->len + src_len);
+
+	crypto_aes256_eax_update_nonce(&self->enc_ctx.aes256_eax);
+
+	EAX_UPDATE(&self->enc_ctx.aes256_eax.ctx, aes256_encrypt, ad_len, ad);
+	EAX_ENCRYPT(&self->enc_ctx.aes256_eax.ctx, aes256_encrypt, src_len,
+			(uint8_t*)dst->data + dst->len, src);
+	dst->len += src_len;
+
+	EAX_DIGEST(&self->enc_ctx.aes256_eax.ctx, aes256_encrypt, 16, mac);
+
+	return true;
+}
+
+static ssize_t crypto_cipher_aes256_eax_decrypt(struct crypto_cipher* self,
+		uint8_t* dst, uint8_t* mac, const uint8_t* src, size_t len,
+		const uint8_t* ad, size_t ad_len)
+{
+	crypto_aes256_eax_update_nonce(&self->dec_ctx.aes256_eax);
+	EAX_UPDATE(&self->dec_ctx.aes256_eax.ctx, aes256_encrypt, ad_len, ad);
+	EAX_DECRYPT(&self->dec_ctx.aes256_eax.ctx, aes256_encrypt, len, dst, src);
+	EAX_DIGEST(&self->dec_ctx.aes256_eax.ctx, aes256_encrypt, 16, mac);
+	return len;
+}
+
+static struct crypto_cipher* crypto_cipher_new_aes256_eax(const uint8_t* enc_key,
+		const uint8_t* dec_key)
+{
+	struct crypto_cipher* self = calloc(1, sizeof(*self));
+	if (!self)
+		return NULL;
+
+	EAX_SET_KEY(&self->enc_ctx.aes256_eax.ctx, aes256_set_encrypt_key,
+			aes256_encrypt, enc_key);
+	EAX_SET_KEY(&self->dec_ctx.aes256_eax.ctx, aes256_set_encrypt_key,
+			aes256_encrypt, dec_key);
+
+	self->encrypt = crypto_cipher_aes256_eax_encrypt;
+	self->decrypt = crypto_cipher_aes256_eax_decrypt;
+
+	return self;
+}
+
+struct crypto_cipher* crypto_cipher_new(const uint8_t* enc_key,
+		const uint8_t* dec_key, enum crypto_cipher_type type)
+{
+	switch (type) {
+	case CRYPTO_CIPHER_AES128_ECB:
+		return crypto_cipher_new_aes128_ecb(enc_key, dec_key);
+	case CRYPTO_CIPHER_AES_EAX:
+		return crypto_cipher_new_aes_eax(enc_key, dec_key);
+	case CRYPTO_CIPHER_AES256_EAX:
+		return crypto_cipher_new_aes256_eax(enc_key, dec_key);
+	case CRYPTO_CIPHER_INVALID:
+		break;
+	}
+
+	nvnc_log(NVNC_LOG_PANIC, "Invalid type: %d", type);
+	return NULL;
+}
+
+void crypto_cipher_del(struct crypto_cipher* self)
+{
+	free(self);
+}
+
+bool crypto_cipher_encrypt(struct crypto_cipher* self, struct vec* dst,
+		uint8_t* mac, const uint8_t* src, size_t src_len,
+		const uint8_t* ad, size_t ad_len)
+{
+	return self->encrypt(self, dst, mac, src, src_len, ad, ad_len);
+}
+
+ssize_t crypto_cipher_decrypt(struct crypto_cipher* self, uint8_t* dst,
+		uint8_t* mac, const uint8_t* src, size_t src_len,
+		const uint8_t* ad, size_t ad_len)
+{
+	return self->decrypt(self, dst, mac, src, src_len, ad, ad_len);
+}
+
+struct crypto_hash* crypto_hash_new(enum crypto_hash_type type)
+{
+	struct crypto_hash* self = calloc(1, sizeof(*self));
+	if (!self)
+		return NULL;
+
+	switch (type) {
+	case CRYPTO_HASH_INVALID:
+		nvnc_log(NVNC_LOG_PANIC, "Invalid hash type");
+		break;
+	case CRYPTO_HASH_MD5:
+		md5_init(&self->ctx.md5);
+		self->update = (void*)nettle_md5_update;
+		self->digest = (void*)nettle_md5_digest;
+		break;
+	case CRYPTO_HASH_SHA1:
+		sha1_init(&self->ctx.sha1);
+		self->update = (void*)nettle_sha1_update;
+		self->digest = (void*)nettle_sha1_digest;
+		break;
+	case CRYPTO_HASH_SHA256:
+		sha256_init(&self->ctx.sha256);
+		self->update = (void*)nettle_sha256_update;
+		self->digest = (void*)nettle_sha256_digest;
+		break;
+	}
+
+	return self;
+}
+
+void crypto_hash_del(struct crypto_hash* self)
+{
+	free(self);
+}
+
+void crypto_hash_append(struct crypto_hash* self, const uint8_t* src,
+		size_t len)
+{
+	self->update(&self->ctx, len, src);
+}
+
+void crypto_hash_digest(struct crypto_hash* self, uint8_t* dst, size_t len)
+{
+	self->digest(&self->ctx, len, dst);
+}
+
+void crypto_hash_one(uint8_t* dst, size_t dst_len, enum crypto_hash_type type,
+		const uint8_t* src, size_t src_len)
+{
+	struct crypto_hash *hash = crypto_hash_new(type);
+	crypto_hash_append(hash, src, src_len);
+	crypto_hash_digest(hash, dst, dst_len);
+	crypto_hash_del(hash);
+}
+
+void crypto_hash_many(uint8_t* dst, size_t dst_len, enum crypto_hash_type type,
+		const struct crypto_data_entry *src)
+{
+	struct crypto_hash *hash = crypto_hash_new(type);
+	for (int i = 0; src[i].data && src[i].len; ++i)
+		crypto_hash_append(hash, src[i].data, src[i].len);
+	crypto_hash_digest(hash, dst, dst_len);
+	crypto_hash_del(hash);
+}
+
+struct crypto_rsa_pub_key *crypto_rsa_pub_key_new(void)
+{
+	struct crypto_rsa_pub_key* self = calloc(1, sizeof(*self));
+	if (!self)
+		return NULL;
+
+	rsa_public_key_init(&self->key);
+	return self;
+}
+
+void crypto_rsa_pub_key_del(struct crypto_rsa_pub_key* self)
+{
+	if (!self)
+		return;
+	rsa_public_key_clear(&self->key);
+	free(self);
+}
+
+struct crypto_rsa_pub_key* crypto_rsa_pub_key_import(const uint8_t* modulus,
+		const uint8_t* exponent, size_t size)
+{
+	struct crypto_rsa_pub_key* self = crypto_rsa_pub_key_new();
+	if (!self)
+		return NULL;
+
+	rsa_public_key_init(&self->key);
+	mpz_init(self->key.n);
+	crypto_import(self->key.n, modulus, size);
+	mpz_init(self->key.e);
+	crypto_import(self->key.e, exponent, size);
+	rsa_public_key_prepare(&self->key);
+
+	return self;
+}
+
+bool crypto_rsa_priv_key_import_pkcs1_der(struct crypto_rsa_priv_key* priv,
+		struct crypto_rsa_pub_key* pub, const uint8_t* key,
+		size_t size)
+{
+	return rsa_keypair_from_der(&pub->key, &priv->key, 0, size, key);
+}
+
+bool crypto_rsa_priv_key_load(struct crypto_rsa_priv_key* priv,
+		struct crypto_rsa_pub_key* pub, const char* path)
+{
+	FILE* stream = fopen(path, "r");
+	if (!stream) {
+		nvnc_log(NVNC_LOG_ERROR, "Could not open file: %m");
+		return false;
+	}
+
+	char* line = NULL;
+	size_t n = 0;
+	if (getline(&line, &n, stream) < 0) {
+		nvnc_log(NVNC_LOG_ERROR, "RSA private key file is not PEM");
+		return false;
+	}
+
+	char head[128];
+	strncpy(head, line, sizeof(head));
+	head[sizeof(head) - 1] = '\0';
+	char* end = strchr(head, '\n');
+	if (end)
+		*end = '\0';
+
+	nvnc_trace("Read PEM head: \"%s\"\n", head);
+
+	struct vec base64_der;
+	vec_init(&base64_der, 4096);
+
+	while (getline(&line, &n, stream) >= 0) {
+		if (strncmp(line, "-----END", 8) == 0)
+			break;
+
+
+		vec_append(&base64_der, line, strcspn(line, "\n"));
+	}
+	free(line);
+	fclose(stream);
+
+	vec_append_zero(&base64_der, 1);
+
+	uint8_t* der = malloc(BASE64_DECODED_MAX_SIZE(base64_der.len));
+	assert(der);
+
+	ssize_t der_len = base64_decode(der, base64_der.data);
+	vec_destroy(&base64_der);
+	if (der_len < 0) {
+		free(der);
+		return false;
+	}
+
+	bool ok = false;
+	if (strcmp(head, "-----BEGIN RSA PRIVATE KEY-----") == 0) {
+		ok = crypto_rsa_priv_key_import_pkcs1_der(priv, pub, der, der_len);
+	} else {
+		nvnc_log(NVNC_LOG_ERROR, "Unsupported RSA private key format");
+	}
+
+	nvnc_trace("Private key is %d bits long", priv->key.size * 8);
+
+	free(der);
+	return ok;
+}
+
+void crypto_rsa_pub_key_modulus(const struct crypto_rsa_pub_key* key,
+		uint8_t* dst, size_t dst_size)
+{
+	crypto_export(dst, dst_size, key->key.n);
+}
+
+void crypto_rsa_pub_key_exponent(const struct crypto_rsa_pub_key* key,
+		uint8_t* dst, size_t dst_size)
+{
+	char* str = mpz_get_str(NULL, 16, key->key.e);
+	free(str);
+
+	crypto_export(dst, dst_size, key->key.e);
+}
+
+struct crypto_rsa_priv_key *crypto_rsa_priv_key_new(void)
+{
+	struct crypto_rsa_priv_key* self = calloc(1, sizeof(*self));
+	if (!self)
+		return NULL;
+
+	rsa_private_key_init(&self->key);
+	return self;
+}
+
+void crypto_rsa_priv_key_del(struct crypto_rsa_priv_key* self)
+{
+	if (!self)
+		return;
+	rsa_private_key_clear(&self->key);
+	free(self);
+}
+
+size_t crypto_rsa_pub_key_length(const struct crypto_rsa_pub_key* key)
+{
+	return key->key.size;
+}
+
+static void generate_random_for_rsa(void* random_ctx, size_t len, uint8_t* dst)
+{
+	getrandom(dst, len, 0);
+}
+
+bool crypto_rsa_keygen(struct crypto_rsa_pub_key* pub,
+		struct crypto_rsa_priv_key* priv)
+{
+	void* random_ctx = NULL;
+	nettle_random_func* random_func = generate_random_for_rsa;
+	void* progress_ctx = NULL;
+	nettle_progress_func* progress = NULL;
+
+	int rc = rsa_generate_keypair(&pub->key, &priv->key, random_ctx,
+			random_func, progress_ctx, progress, 2048, 30);
+	return rc != 0;
+}
+
+ssize_t crypto_rsa_encrypt(struct crypto_rsa_pub_key* pub, uint8_t* dst,
+		size_t dst_size, const uint8_t* src, size_t src_size)
+{
+	mpz_t ciphertext;
+	mpz_init(ciphertext);
+	int r = rsa_encrypt(&pub->key, NULL, generate_random_for_rsa,
+			src_size, src, ciphertext);
+	if (r == 0) {
+		mpz_clear(ciphertext);
+		return -1;
+	}
+	size_t len = crypto_export(dst, dst_size, ciphertext);
+	mpz_clear(ciphertext);
+	return len;
+}
+
+ssize_t crypto_rsa_decrypt(struct crypto_rsa_priv_key* priv, uint8_t* dst,
+		size_t dst_size, const uint8_t* src, size_t src_size)
+{
+	mpz_t ciphertext;
+	mpz_init(ciphertext);
+	crypto_import(ciphertext, src, src_size);
+	int r = rsa_decrypt(&priv->key, &dst_size, dst, ciphertext);
+	mpz_clear(ciphertext);
+	return r != 0 ? (ssize_t)dst_size : -1;
+}

src/cursor.c

@@ -94,12 +94,14 @@ int cursor_encode(struct vec* dst, struct rfb_pixel_format* pixfmt,
 	uint8_t* dstdata = dst->data;
 	dstdata += dst->len;
 
+	int32_t src_byte_stride = image->stride * (srcfmt.bits_per_pixel / 8);
+
 	if((int32_t)width == image->stride) {
-		pixel32_to_cpixel(dstdata, pixfmt, image->addr, &srcfmt, bpp, size);
+		pixel_to_cpixel(dstdata, pixfmt, image->addr, &srcfmt, bpp, size);
 	} else {
 		for (uint32_t y = 0; y < height; ++y) {
-			pixel32_to_cpixel(dstdata + y * bpp * width, pixfmt,
-					(uint32_t*)image->addr + y * image->stride,
+			pixel_to_cpixel(dstdata + y * bpp * width, pixfmt,
+					(uint8_t*)image->addr + y * src_byte_stride,
 					&srcfmt, bpp, width);
 		}
 	}
@@ -110,8 +112,8 @@ int cursor_encode(struct vec* dst, struct rfb_pixel_format* pixfmt,
 
 	for (uint32_t y = 0; y < height; ++y) {
 		if (!extract_alpha_mask(dstdata + y * UDIV_UP(width, 8),
-					(uint32_t*)image->addr + y * image->stride,
-					image->fourcc_format, width))
+				(uint32_t*)image->addr + y * image->stride,
+				image->fourcc_format, width))
 			goto failure;
 
 		dst->len += UDIV_UP(width, 8);

src/damage-refinery.c

@@ -22,8 +22,13 @@
 #include <sys/param.h>
 
 #include "fb.h"
+#include "pixels.h"
 #include "damage-refinery.h"
-#include "murmurhash.h"
+
+#define XXH_STATIC_LINKING_ONLY
+#define XXH_IMPLEMENTATION
+#define XXH_VECTOR XXH_SCALAR
+#include "xxhash.h"
 
 #define UDIV_UP(a, b) (((a) + (b) - 1) / (b))
 
@@ -38,10 +43,16 @@ int damage_refinery_init(struct damage_refinery* self, uint32_t width,
 	uint32_t twidth = UDIV_UP(width, 32);
 	uint32_t theight = UDIV_UP(height, 32);
 
-	self->hashes = calloc(twidth * theight, sizeof(*self->hashes));
-	if (!self->hashes)
+	self->state = XXH3_createState();
+	if (!self->state)
 		return -1;
 
+	self->hashes = calloc(twidth * theight, sizeof(*self->hashes));
+	if (!self->hashes) {
+		XXH3_freeState(self->state);
+		return -1;
+	}
+
 	return 0;
 }
 
@@ -57,28 +68,31 @@ int damage_refinery_resize(struct damage_refinery* self, uint32_t width,
 
 void damage_refinery_destroy(struct damage_refinery* self)
 {
+	XXH3_freeState(self->state);
 	free(self->hashes);
 }
 
 static uint32_t damage_hash_tile(struct damage_refinery* self, uint32_t tx,
 		uint32_t ty, const struct nvnc_fb* buffer)
 {
-	uint32_t* pixels = buffer->addr;
-	int pixel_stride = buffer->stride;
+	uint8_t* pixels = buffer->addr;
+	int bpp = pixel_size_from_fourcc(buffer->fourcc_format);
+	int byte_stride = buffer->stride * bpp;
 
 	int x_start = tx * 32;
 	int x_stop = MIN((tx + 1) * 32, self->width);
 	int y_start = ty * 32;
 	int y_stop = MIN((ty + 1) * 32, self->height);
 
-	uint32_t hash = 0;
+	int32_t xoff = x_start * bpp;
 
-	// TODO: Support different pixel sizes
-	for (int y = y_start; y < y_stop; ++y)
-		hash = murmurhash((void*)&(pixels[x_start + y * pixel_stride]),
-				4 * (x_stop - x_start), hash);
+	XXH3_64bits_reset(self->state);
+	for (int y = y_start; y < y_stop; ++y) {
+		XXH3_64bits_update(self->state, pixels + xoff + y * byte_stride,
+				bpp * (x_stop - x_start));
+	}
 
-	return hash;
+	return XXH3_64bits_digest(self->state);
 }
 
 static uint32_t* damage_tile_hash_ptr(struct damage_refinery* self,
@@ -119,12 +133,12 @@ static void tile_region_from_region(struct pixman_region16* dst,
 }
 
 void damage_refine(struct damage_refinery* self,
-		struct pixman_region16* refined, 
+		struct pixman_region16* refined,
 		struct pixman_region16* hint,
 		struct nvnc_fb* buffer)
 {
 	assert(self->width == (uint32_t)buffer->width &&
-	       self->height == (uint32_t)buffer->height);
+			self->height == (uint32_t)buffer->height);
 
 	nvnc_fb_map(buffer);
 

src/enc-util.c

@@ -50,7 +50,7 @@ uint32_t calculate_region_area(struct pixman_region16* region)
 
 	int n_rects = 0;
 	struct pixman_box16* rects = pixman_region_rectangles(region,
-		&n_rects);
+			&n_rects);
 
 	for (int i = 0; i < n_rects; ++i) {
 		int width = rects[i].x2 - rects[i].x1;

src/fb.c

@@ -35,12 +35,14 @@
 
 EXPORT
 struct nvnc_fb* nvnc_fb_new(uint16_t width, uint16_t height,
-                            uint32_t fourcc_format, uint16_t stride)
+		uint32_t fourcc_format, uint16_t stride)
 {
 	struct nvnc_fb* fb = calloc(1, sizeof(*fb));
 	if (!fb)
 		return NULL;
 
+	uint32_t bpp = pixel_size_from_fourcc(fourcc_format);
+
 	fb->type = NVNC_FB_SIMPLE;
 	fb->ref = 1;
 	fb->width = width;
@@ -49,7 +51,7 @@ struct nvnc_fb* nvnc_fb_new(uint16_t width, uint16_t height,
 	fb->stride = stride;
 	fb->pts = NVNC_NO_PTS;
 
-	size_t size = height * stride * 4; /* Assume 4 byte format for now */
+	size_t size = height * stride * bpp;
 	size_t alignment = MAX(4, sizeof(void*));
 	size_t aligned_size = ALIGN_UP(size, alignment);
 
@@ -64,7 +66,7 @@ struct nvnc_fb* nvnc_fb_new(uint16_t width, uint16_t height,
 
 EXPORT
 struct nvnc_fb* nvnc_fb_from_buffer(void* buffer, uint16_t width, uint16_t height,
-                            uint32_t fourcc_format, int32_t stride)
+		uint32_t fourcc_format, int32_t stride)
 {
 	struct nvnc_fb* fb = calloc(1, sizeof(*fb));
 	if (!fb)

src/fb_pool.c

@@ -165,7 +165,7 @@ void nvnc_fb_pool_release(struct nvnc_fb_pool* self, struct nvnc_fb* fb)
 	}
 
 	nvnc_fb_ref(fb);
-	
+
 	struct fbq_item* item = calloc(1, sizeof(*item));
 	assert(item);
 	item->fb = fb;

src/h264-encoder-ffmpeg-impl.c

@@ -0,0 +1,627 @@
+/*
+ * Copyright (c) 2021 - 2024 Andri Yngvason
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
+ * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+ * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+ * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+ * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "h264-encoder.h"
+#include "neatvnc.h"
+#include "fb.h"
+#include "sys/queue.h"
+#include "vec.h"
+#include "usdt.h"
+
+#include <stdlib.h>
+#include <stdint.h>
+#include <stdbool.h>
+#include <unistd.h>
+#include <assert.h>
+#include <gbm.h>
+#include <xf86drm.h>
+#include <aml.h>
+
+#include <libavcodec/avcodec.h>
+#include <libavutil/hwcontext.h>
+#include <libavutil/hwcontext_drm.h>
+#include <libavutil/pixdesc.h>
+#include <libavutil/dict.h>
+#include <libavfilter/avfilter.h>
+#include <libavfilter/buffersink.h>
+#include <libavfilter/buffersrc.h>
+
+#include <libdrm/drm_fourcc.h>
+
+struct h264_encoder;
+
+struct fb_queue_entry {
+	struct nvnc_fb* fb;
+	TAILQ_ENTRY(fb_queue_entry) link;
+};
+
+TAILQ_HEAD(fb_queue, fb_queue_entry);
+
+struct h264_encoder_ffmpeg {
+	struct h264_encoder base;
+
+	uint32_t width;
+	uint32_t height;
+	uint32_t format;
+
+	AVRational timebase;
+	AVRational sample_aspect_ratio;
+	enum AVPixelFormat av_pixel_format;
+
+	/* type: AVHWDeviceContext */
+	AVBufferRef* hw_device_ctx;
+
+	/* type: AVHWFramesContext */
+	AVBufferRef* hw_frames_ctx;
+
+	AVCodecContext* codec_ctx;
+
+	AVFilterGraph* filter_graph;
+	AVFilterContext* filter_in;
+	AVFilterContext* filter_out;
+
+	struct fb_queue fb_queue;
+
+	struct aml_work* work;
+	struct nvnc_fb* current_fb;
+	struct vec current_packet;
+	bool current_frame_is_keyframe;
+
+	bool please_destroy;
+};
+
+struct h264_encoder_impl h264_encoder_ffmpeg_impl;
+
+static enum AVPixelFormat drm_to_av_pixel_format(uint32_t format)
+{
+	switch (format) {
+	case DRM_FORMAT_XRGB8888:
+	case DRM_FORMAT_ARGB8888:
+		return AV_PIX_FMT_BGR0;
+	case DRM_FORMAT_XBGR8888:
+	case DRM_FORMAT_ABGR8888:
+		return AV_PIX_FMT_RGB0;
+	case DRM_FORMAT_RGBX8888:
+	case DRM_FORMAT_RGBA8888:
+		return AV_PIX_FMT_0BGR;
+	case DRM_FORMAT_BGRX8888:
+	case DRM_FORMAT_BGRA8888:
+		return AV_PIX_FMT_0RGB;
+	}
+
+	return AV_PIX_FMT_NONE;
+}
+
+static void hw_frame_desc_free(void* opaque, uint8_t* data)
+{
+	struct AVDRMFrameDescriptor* desc = (void*)data;
+	assert(desc);
+
+	for (int i = 0; i < desc->nb_objects; ++i)
+		close(desc->objects[i].fd);
+
+	free(desc);
+}
+
+// TODO: Maybe do this once per frame inside nvnc_fb?
+static AVFrame* fb_to_avframe(struct nvnc_fb* fb)
+{
+	struct gbm_bo* bo = fb->bo;
+
+	int n_planes = gbm_bo_get_plane_count(bo);
+
+	AVDRMFrameDescriptor* desc = calloc(1, sizeof(*desc));
+	desc->nb_objects = n_planes;
+
+	desc->nb_layers = 1;
+	desc->layers[0].format = gbm_bo_get_format(bo);
+	desc->layers[0].nb_planes = n_planes;
+
+	for (int i = 0; i < n_planes; ++i) {
+		uint32_t stride = gbm_bo_get_stride_for_plane(bo, i);
+
+		desc->objects[i].fd = gbm_bo_get_fd_for_plane(bo, i);
+		desc->objects[i].size = stride * fb->height;
+		desc->objects[i].format_modifier = gbm_bo_get_modifier(bo);
+
+		desc->layers[0].format = gbm_bo_get_format(bo);
+		desc->layers[0].planes[i].object_index = i;
+		desc->layers[0].planes[i].offset = gbm_bo_get_offset(bo, i);
+		desc->layers[0].planes[i].pitch = stride;
+	}
+
+	AVFrame* frame = av_frame_alloc();
+	if (!frame) {
+		hw_frame_desc_free(NULL, (void*)desc);
+		return NULL;
+	}
+
+	frame->opaque = fb;
+	frame->width = fb->width;
+	frame->height = fb->height;
+	frame->format = AV_PIX_FMT_DRM_PRIME;
+	frame->sample_aspect_ratio = (AVRational){1, 1};
+
+	AVBufferRef* desc_ref = av_buffer_create((void*)desc, sizeof(*desc),
+			hw_frame_desc_free, NULL, 0);
+	if (!desc_ref) {
+		hw_frame_desc_free(NULL, (void*)desc);
+		av_frame_free(&frame);
+		return NULL;
+	}
+
+	frame->buf[0] = desc_ref;
+	frame->data[0] = (void*)desc_ref->data;
+
+	// TODO: Set colorspace?
+
+	return frame;
+}
+
+static struct nvnc_fb* fb_queue_dequeue(struct fb_queue* queue)
+{
+	if (TAILQ_EMPTY(queue))
+		return NULL;
+
+	struct fb_queue_entry* entry = TAILQ_FIRST(queue);
+	TAILQ_REMOVE(queue, entry, link);
+	struct nvnc_fb* fb = entry->fb;
+	free(entry);
+
+	return fb;
+}
+
+static int fb_queue_enqueue(struct fb_queue* queue, struct nvnc_fb* fb)
+{
+	struct fb_queue_entry* entry = calloc(1, sizeof(*entry));
+	if (!entry)
+		return -1;
+
+	entry->fb = fb;
+	nvnc_fb_ref(fb);
+	TAILQ_INSERT_TAIL(queue, entry, link);
+
+	return 0;
+}
+
+static int h264_encoder__init_buffersrc(struct h264_encoder_ffmpeg* self)
+{
+	int rc;
+
+	/* Placeholder values are used to pacify input checking and the real
+	 * values are set below.
+	 */
+	rc = avfilter_graph_create_filter(&self->filter_in,
+			avfilter_get_by_name("buffer"), "in",
+			"width=1:height=1:pix_fmt=drm_prime:time_base=1/1", NULL,
+			self->filter_graph);
+	if (rc != 0)
+		return -1;
+
+	AVBufferSrcParameters *params = av_buffersrc_parameters_alloc();
+	if (!params)
+		return -1;
+
+	params->format = AV_PIX_FMT_DRM_PRIME;
+	params->width = self->width;
+	params->height = self->height;
+	params->sample_aspect_ratio = self->sample_aspect_ratio;
+	params->time_base = self->timebase;
+	params->hw_frames_ctx = self->hw_frames_ctx;
+
+	rc = av_buffersrc_parameters_set(self->filter_in, params);
+	assert(rc == 0);
+
+	av_free(params);
+	return 0;
+}
+
+static int h264_encoder__init_filters(struct h264_encoder_ffmpeg* self)
+{
+	int rc;
+
+	self->filter_graph = avfilter_graph_alloc();
+	if (!self->filter_graph)
+		return -1;
+
+	rc = h264_encoder__init_buffersrc(self);
+	if (rc != 0)
+		goto failure;
+
+	rc = avfilter_graph_create_filter(&self->filter_out,
+			avfilter_get_by_name("buffersink"), "out", NULL,
+			NULL, self->filter_graph);
+	if (rc != 0)
+		goto failure;
+
+	AVFilterInOut* inputs = avfilter_inout_alloc();
+	if (!inputs)
+		goto failure;
+
+	inputs->name = av_strdup("in");
+	inputs->filter_ctx = self->filter_in;
+	inputs->pad_idx = 0;
+	inputs->next = NULL;
+
+	AVFilterInOut* outputs = avfilter_inout_alloc();
+	if (!outputs) {
+		avfilter_inout_free(&inputs);
+		goto failure;
+	}
+
+	outputs->name = av_strdup("out");
+	outputs->filter_ctx = self->filter_out;
+	outputs->pad_idx = 0;
+	outputs->next = NULL;
+
+	rc = avfilter_graph_parse(self->filter_graph,
+			"hwmap=mode=direct:derive_device=vaapi"
+			",scale_vaapi=format=nv12:mode=fast",
+			outputs, inputs, NULL);
+	if (rc != 0)
+		goto failure;
+
+	assert(self->hw_device_ctx);
+
+	for (unsigned int i = 0; i < self->filter_graph->nb_filters; ++i) {
+		self->filter_graph->filters[i]->hw_device_ctx =
+			av_buffer_ref(self->hw_device_ctx);
+	}
+
+	rc = avfilter_graph_config(self->filter_graph, NULL);
+	if (rc != 0)
+		goto failure;
+
+	return 0;
+
+failure:
+	avfilter_graph_free(&self->filter_graph);
+	return -1;
+}
+
+static int h264_encoder__init_codec_context(struct h264_encoder_ffmpeg* self,
+		const AVCodec* codec, int quality)
+{
+	self->codec_ctx = avcodec_alloc_context3(codec);
+	if (!self->codec_ctx)
+		return -1;
+
+	struct AVCodecContext* c = self->codec_ctx;
+	c->width = self->width;
+	c->height = self->height;
+	c->time_base = self->timebase;
+	c->sample_aspect_ratio = self->sample_aspect_ratio;
+	c->pix_fmt = AV_PIX_FMT_VAAPI;
+	c->gop_size = INT32_MAX; /* We'll select key frames manually */
+	c->max_b_frames = 0; /* B-frames are bad for latency */
+	c->global_quality = quality;
+
+	/* open-h264 requires baseline profile, so we use constrained
+	 * baseline: AV_PROFILE_H264_BASELINE.
+	 * But that is not supported by many clients. So we use a "DEFAULT" profile.
+	 * 
+	 */
+	c->profile = AV_PROFILE_H264_MAIN;
+
+	return 0;
+}
+
+static int h264_encoder__init_hw_frames_context(struct h264_encoder_ffmpeg* self)
+{
+	self->hw_frames_ctx = av_hwframe_ctx_alloc(self->hw_device_ctx);
+	if (!self->hw_frames_ctx)
+		return -1;
+
+	AVHWFramesContext* c = (AVHWFramesContext*)self->hw_frames_ctx->data;
+	c->format = AV_PIX_FMT_DRM_PRIME;
+	c->sw_format = drm_to_av_pixel_format(self->format);
+	c->width = self->width;
+	c->height = self->height;
+
+	if (av_hwframe_ctx_init(self->hw_frames_ctx) < 0)
+		av_buffer_unref(&self->hw_frames_ctx);
+
+	return 0;
+}
+
+static int h264_encoder__schedule_work(struct h264_encoder_ffmpeg* self)
+{
+	if (self->current_fb)
+		return 0;
+
+	self->current_fb = fb_queue_dequeue(&self->fb_queue);
+	if (!self->current_fb)
+		return 0;
+
+	DTRACE_PROBE1(neatvnc, h264_encode_frame_begin, self->current_fb->pts);
+
+	self->current_frame_is_keyframe = self->base.next_frame_should_be_keyframe;
+	self->base.next_frame_should_be_keyframe = false;
+
+	return aml_start(aml_get_default(), self->work);
+}
+
+static int h264_encoder__encode(struct h264_encoder_ffmpeg* self,
+		AVFrame* frame_in)
+{
+	int rc;
+
+	rc = av_buffersrc_add_frame_flags(self->filter_in, frame_in,
+			AV_BUFFERSRC_FLAG_KEEP_REF);
+	if (rc != 0)
+		return -1;
+
+	AVFrame* filtered_frame = av_frame_alloc();
+	if (!filtered_frame)
+		return -1;
+
+	rc = av_buffersink_get_frame(self->filter_out, filtered_frame);
+	if (rc != 0)
+		goto get_frame_failure;
+
+	rc = avcodec_send_frame(self->codec_ctx, filtered_frame);
+	if (rc != 0)
+		goto send_frame_failure;
+
+	AVPacket* packet = av_packet_alloc();
+	assert(packet); // TODO
+
+	while (1) {
+		rc = avcodec_receive_packet(self->codec_ctx, packet);
+		if (rc != 0)
+			break;
+
+		vec_append(&self->current_packet, packet->data, packet->size);
+
+		packet->stream_index = 0;
+		av_packet_unref(packet);
+	}
+
+	// Frame should always start with a zero:
+	assert(self->current_packet.len == 0 ||
+			((char*)self->current_packet.data)[0] == 0);
+
+	av_packet_free(&packet);
+send_frame_failure:
+	av_frame_unref(filtered_frame);
+get_frame_failure:
+	av_frame_free(&filtered_frame);
+	return rc == AVERROR(EAGAIN) ? 0 : rc;
+}
+
+static void h264_encoder__do_work(void* handle)
+{
+	struct h264_encoder_ffmpeg* self = aml_get_userdata(handle);
+
+	AVFrame* frame = fb_to_avframe(self->current_fb);
+	assert(frame); // TODO
+
+	frame->hw_frames_ctx = av_buffer_ref(self->hw_frames_ctx);
+
+	if (self->current_frame_is_keyframe) {
+#if LIBAVUTIL_VERSION_INT >= AV_VERSION_INT(58, 7, 100)
+		frame->flags |= AV_FRAME_FLAG_KEY;
+#else
+		frame->key_frame = 1;
+#endif
+		frame->pict_type = AV_PICTURE_TYPE_I;
+	} else {
+#if LIBAVUTIL_VERSION_INT >= AV_VERSION_INT(58, 7, 100)
+		frame->flags &= ~AV_FRAME_FLAG_KEY;
+#else
+		frame->key_frame = 0;
+#endif
+		frame->pict_type = AV_PICTURE_TYPE_P;
+	}
+
+	int rc = h264_encoder__encode(self, frame);
+	if (rc != 0) {
+		char err[256];
+		av_strerror(rc, err, sizeof(err));
+		nvnc_log(NVNC_LOG_ERROR, "Failed to encode packet: %s", err);
+		goto failure;
+	}
+
+failure:
+	av_frame_unref(frame);
+	av_frame_free(&frame);
+}
+
+static void h264_encoder__on_work_done(void* handle)
+{
+	struct h264_encoder_ffmpeg* self = aml_get_userdata(handle);
+
+	uint64_t pts = nvnc_fb_get_pts(self->current_fb);
+	nvnc_fb_release(self->current_fb);
+	nvnc_fb_unref(self->current_fb);
+	self->current_fb = NULL;
+
+	DTRACE_PROBE1(neatvnc, h264_encode_frame_end, pts);
+
+	if (self->please_destroy) {
+		vec_destroy(&self->current_packet);
+		h264_encoder_destroy(&self->base);
+		return;
+	}
+
+	if (self->current_packet.len == 0) {
+		nvnc_log(NVNC_LOG_WARNING, "Whoops, encoded packet length is 0");
+		return;
+	}
+
+	void* userdata = self->base.userdata;
+
+	// Must make a copy of packet because the callback might destroy the
+	// encoder object.
+	struct vec packet;
+	vec_init(&packet, self->current_packet.len);
+	vec_append(&packet, self->current_packet.data,
+			self->current_packet.len);
+
+	vec_clear(&self->current_packet);
+	h264_encoder__schedule_work(self);
+
+	self->base.on_packet_ready(packet.data, packet.len, pts, userdata);
+	vec_destroy(&packet);
+}
+
+static int find_render_node(char *node, size_t maxlen) {
+	bool r = -1;
+	drmDevice *devices[64];
+
+	int n = drmGetDevices2(0, devices, sizeof(devices) / sizeof(devices[0]));
+	for (int i = 0; i < n; ++i) {
+		drmDevice *dev = devices[i];
+		if (!(dev->available_nodes & (1 << DRM_NODE_RENDER)))
+			continue;
+
+		strncpy(node, dev->nodes[DRM_NODE_RENDER], maxlen);
+		node[maxlen - 1] = '\0';
+		r = 0;
+		break;
+	}
+
+	drmFreeDevices(devices, n);
+	return r;
+}
+
+static struct h264_encoder* h264_encoder_ffmpeg_create(uint32_t width,
+		uint32_t height, uint32_t format, int quality)
+{
+	int rc;
+
+	struct h264_encoder_ffmpeg* self = calloc(1, sizeof(*self));
+	if (!self)
+		return NULL;
+
+	self->base.impl = &h264_encoder_ffmpeg_impl;
+
+	if (vec_init(&self->current_packet, 65536) < 0)
+		goto packet_failure;
+
+	self->work = aml_work_new(h264_encoder__do_work,
+			h264_encoder__on_work_done, self, NULL);
+	if (!self->work)
+		goto worker_failure;
+
+	char render_node[64];
+	if (find_render_node(render_node, sizeof(render_node)) < 0)
+		goto render_node_failure;
+
+	rc = av_hwdevice_ctx_create(&self->hw_device_ctx,
+			AV_HWDEVICE_TYPE_DRM, render_node, NULL, 0);
+	if (rc != 0)
+		goto hwdevice_ctx_failure;
+
+	self->base.next_frame_should_be_keyframe = true;
+	TAILQ_INIT(&self->fb_queue);
+
+	self->width = width;
+	self->height = height;
+	self->format = format;
+	self->timebase = (AVRational){1, 1000000};
+	self->sample_aspect_ratio = (AVRational){1, 1};
+	self->av_pixel_format = drm_to_av_pixel_format(format);
+	if (self->av_pixel_format == AV_PIX_FMT_NONE)
+		goto pix_fmt_failure;
+
+	const AVCodec* codec = avcodec_find_encoder_by_name("h264_vaapi");
+	if (!codec)
+		goto codec_failure;
+
+	if (h264_encoder__init_hw_frames_context(self) < 0)
+		goto hw_frames_context_failure;
+
+	if (h264_encoder__init_filters(self) < 0)
+		goto filter_failure;
+
+	if (h264_encoder__init_codec_context(self, codec, quality) < 0)
+		goto codec_context_failure;
+
+	self->codec_ctx->hw_frames_ctx =
+		av_buffer_ref(self->filter_out->inputs[0]->hw_frames_ctx);
+
+	AVDictionary *opts = NULL;
+	av_dict_set_int(&opts, "async_depth", 1, 0);
+
+	rc = avcodec_open2(self->codec_ctx, codec, &opts);
+	av_dict_free(&opts);
+
+	if (rc != 0)
+		goto avcodec_open_failure;
+
+	return &self->base;
+
+avcodec_open_failure:
+	avcodec_free_context(&self->codec_ctx);
+codec_context_failure:
+filter_failure:
+	av_buffer_unref(&self->hw_frames_ctx);
+hw_frames_context_failure:
+codec_failure:
+pix_fmt_failure:
+	av_buffer_unref(&self->hw_device_ctx);
+hwdevice_ctx_failure:
+render_node_failure:
+	aml_unref(self->work);
+worker_failure:
+	vec_destroy(&self->current_packet);
+packet_failure:
+	free(self);
+	return NULL;
+}
+
+static void h264_encoder_ffmpeg_destroy(struct h264_encoder* base)
+{
+	struct h264_encoder_ffmpeg* self = (struct h264_encoder_ffmpeg*)base;
+
+	if (self->current_fb) {
+		self->please_destroy = true;
+		return;
+	}
+
+	vec_destroy(&self->current_packet);
+	av_buffer_unref(&self->hw_frames_ctx);
+	avcodec_free_context(&self->codec_ctx);
+	av_buffer_unref(&self->hw_device_ctx);
+	avfilter_graph_free(&self->filter_graph);
+	aml_unref(self->work);
+	free(self);
+}
+
+static void h264_encoder_ffmpeg_feed(struct h264_encoder* base,
+		struct nvnc_fb* fb)
+{
+	struct h264_encoder_ffmpeg* self = (struct h264_encoder_ffmpeg*)base;
+	assert(fb->type == NVNC_FB_GBM_BO);
+
+	// TODO: Add transform filter
+	assert(fb->transform == NVNC_TRANSFORM_NORMAL);
+
+	int rc = fb_queue_enqueue(&self->fb_queue, fb);
+	assert(rc == 0); // TODO
+
+	nvnc_fb_hold(fb);
+
+	rc = h264_encoder__schedule_work(self);
+	assert(rc == 0); // TODO
+}
+
+struct h264_encoder_impl h264_encoder_ffmpeg_impl = {
+	.create = h264_encoder_ffmpeg_create,
+	.destroy = h264_encoder_ffmpeg_destroy,
+	.feed = h264_encoder_ffmpeg_feed,
+};

src/h264-encoder-v4l2m2m-impl.c

@@ -0,0 +1,741 @@
+/*
+ * Copyright (c) 2024 Andri Yngvason
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
+ * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+ * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+ * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+ * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "h264-encoder.h"
+#include "neatvnc.h"
+#include "fb.h"
+#include "pixels.h"
+
+#include <assert.h>
+#include <string.h>
+#include <stdio.h>
+#include <inttypes.h>
+#include <fcntl.h>
+#include <sys/mman.h>
+#include <sys/ioctl.h>
+#include <linux/videodev2.h>
+#include <drm_fourcc.h>
+#include <gbm.h>
+#include <aml.h>
+#include <dirent.h>
+
+#define UDIV_UP(a, b) (((a) + (b) - 1) / (b))
+#define ALIGN_UP(a, b) ((b) * UDIV_UP((a), (b)))
+#define ARRAY_LENGTH(a) (sizeof(a) / sizeof((a)[0]))
+
+#define N_SRC_BUFS 3
+#define N_DST_BUFS 3
+
+struct h264_encoder_v4l2m2m_dst_buf {
+	struct v4l2_buffer buffer;
+	struct v4l2_plane plane;
+	void* payload;
+};
+
+struct h264_encoder_v4l2m2m_src_buf {
+	struct v4l2_buffer buffer;
+	struct v4l2_plane planes[4];
+	int fd;
+	bool is_taken;
+	struct nvnc_fb* fb;
+};
+
+struct h264_encoder_v4l2m2m {
+	struct h264_encoder base;
+
+	uint32_t width;
+	uint32_t height;
+	uint32_t format;
+	int quality; // TODO: Can we affect the quality?
+
+	char driver[16];
+
+	int fd;
+	struct aml_handler* handler;
+
+	struct h264_encoder_v4l2m2m_src_buf src_bufs[N_SRC_BUFS];
+	int src_buf_index;
+
+	struct h264_encoder_v4l2m2m_dst_buf dst_bufs[N_DST_BUFS];
+};
+
+struct h264_encoder_impl h264_encoder_v4l2m2m_impl;
+
+static int v4l2_qbuf(int fd, const struct v4l2_buffer* inbuf)
+{
+	assert(inbuf->length <= 4);
+	struct v4l2_plane planes[4];
+	struct v4l2_buffer outbuf;
+	outbuf = *inbuf;
+	memcpy(&planes, inbuf->m.planes, inbuf->length * sizeof(planes[0]));
+	outbuf.m.planes = planes;
+	return ioctl(fd, VIDIOC_QBUF, &outbuf);
+}
+
+static inline int v4l2_dqbuf(int fd, struct v4l2_buffer* buf)
+{
+	return ioctl(fd, VIDIOC_DQBUF, buf);
+}
+
+static struct h264_encoder_v4l2m2m_src_buf* take_src_buffer(
+		struct h264_encoder_v4l2m2m* self)
+{
+	unsigned int count = 0;
+	int i = self->src_buf_index;
+
+	struct h264_encoder_v4l2m2m_src_buf* buffer;
+	do {
+		buffer = &self->src_bufs[i++];
+		i %= ARRAY_LENGTH(self->src_bufs);
+	} while (++count < ARRAY_LENGTH(self->src_bufs) && buffer->is_taken);
+
+	if (buffer->is_taken)
+		return NULL;
+
+	self->src_buf_index = i;
+	buffer->is_taken = true;
+
+	return buffer;
+}
+
+static bool any_src_buf_is_taken(struct h264_encoder_v4l2m2m* self)
+{
+	bool result = false;
+	for (unsigned int i = 0; i < ARRAY_LENGTH(self->src_bufs); ++i)
+		if (self->src_bufs[i].is_taken)
+			result = true;
+	return result;
+}
+
+static int u32_cmp(const void* pa, const void* pb)
+{
+	const uint32_t *a = pa;
+	const uint32_t *b = pb;
+	return *a < *b ? -1 : *a > *b;
+}
+
+static size_t get_supported_formats(struct h264_encoder_v4l2m2m* self,
+		uint32_t* formats, size_t max_len)
+{
+	size_t i = 0;
+	for (;; ++i) {
+		struct v4l2_fmtdesc desc = {
+			.index = i,
+			.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE,
+		};
+		int rc = ioctl(self->fd, VIDIOC_ENUM_FMT, &desc);
+		if (rc < 0)
+			break;
+
+		nvnc_trace("Got pixel format: %s", desc.description);
+
+		formats[i] = desc.pixelformat;
+	}
+
+	qsort(formats, i, sizeof(*formats), u32_cmp);
+
+	return i;
+}
+
+static bool have_v4l2_format(const uint32_t* formats, size_t n_formats,
+		uint32_t format)
+{
+	return bsearch(&format, formats, n_formats, sizeof(format), u32_cmp);
+}
+
+static uint32_t v4l2_format_from_drm(const uint32_t* formats,
+		size_t n_formats, uint32_t drm_format)
+{
+#define TRY_FORMAT(f) \
+	if (have_v4l2_format(formats, n_formats, f)) \
+		return f
+
+	switch (drm_format) {
+	case DRM_FORMAT_RGBX8888:
+	case DRM_FORMAT_RGBA8888:
+		TRY_FORMAT(V4L2_PIX_FMT_RGBX32);
+		TRY_FORMAT(V4L2_PIX_FMT_RGBA32);
+		break;
+	case DRM_FORMAT_XRGB8888:
+	case DRM_FORMAT_ARGB8888:
+		TRY_FORMAT(V4L2_PIX_FMT_XRGB32);
+		TRY_FORMAT(V4L2_PIX_FMT_ARGB32);
+		TRY_FORMAT(V4L2_PIX_FMT_RGB32);
+		break;
+	case DRM_FORMAT_BGRX8888:
+	case DRM_FORMAT_BGRA8888:
+		TRY_FORMAT(V4L2_PIX_FMT_XBGR32);
+		TRY_FORMAT(V4L2_PIX_FMT_ABGR32);
+		TRY_FORMAT(V4L2_PIX_FMT_BGR32);
+		break;
+	case DRM_FORMAT_XBGR8888:
+	case DRM_FORMAT_ABGR8888:
+		TRY_FORMAT(V4L2_PIX_FMT_BGRX32);
+		TRY_FORMAT(V4L2_PIX_FMT_BGRA32);
+		break;
+	// TODO: More formats
+	}
+
+	return 0;
+#undef TRY_FORMAT
+}
+
+// This driver mixes up pixel formats...
+static uint32_t v4l2_format_from_drm_bcm2835(const uint32_t* formats,
+		size_t n_formats, uint32_t drm_format)
+{
+	switch (drm_format) {
+	case DRM_FORMAT_XRGB8888:
+	case DRM_FORMAT_ARGB8888:
+		return V4L2_PIX_FMT_RGBA32;
+	case DRM_FORMAT_BGRX8888:
+	case DRM_FORMAT_BGRA8888:
+		// TODO: This could also be ABGR, based on how this driver
+		// behaves
+		return V4L2_PIX_FMT_BGR32;
+	}
+	return 0;
+}
+
+static int set_src_fmt(struct h264_encoder_v4l2m2m* self)
+{
+	int rc;
+
+	uint32_t supported_formats[256];
+	size_t n_formats = get_supported_formats(self, supported_formats,
+			ARRAY_LENGTH(supported_formats));
+
+	uint32_t format;
+	if (strcmp(self->driver, "bcm2835-codec") == 0)
+		format = v4l2_format_from_drm_bcm2835(supported_formats,
+				n_formats, self->format);
+	else
+		format = v4l2_format_from_drm(supported_formats, n_formats,
+				self->format);
+	if (!format) {
+		nvnc_log(NVNC_LOG_DEBUG, "Failed to find a proper pixel format");
+		return -1;
+	}
+
+	struct v4l2_format fmt = {
+		.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE,
+	};
+	rc = ioctl(self->fd, VIDIOC_G_FMT, &fmt);
+	if (rc < 0) {
+		return -1;
+	}
+
+	struct v4l2_pix_format_mplane* pix_fmt = &fmt.fmt.pix_mp;
+	pix_fmt->pixelformat = format;
+	pix_fmt->width = ALIGN_UP(self->width, 16);
+	pix_fmt->height = ALIGN_UP(self->height, 16);
+
+	rc = ioctl(self->fd, VIDIOC_S_FMT, &fmt);
+	if (rc < 0) {
+		return -1;
+	}
+
+	return 0;
+}
+
+static int set_dst_fmt(struct h264_encoder_v4l2m2m* self)
+{
+	int rc;
+
+	struct v4l2_format fmt = {
+		.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE,
+	};
+	rc = ioctl(self->fd, VIDIOC_G_FMT, &fmt);
+	if (rc < 0) {
+		return -1;
+	}
+
+	struct v4l2_pix_format_mplane* pix_fmt = &fmt.fmt.pix_mp;
+	pix_fmt->pixelformat = V4L2_PIX_FMT_H264;
+	pix_fmt->width = self->width;
+	pix_fmt->height = self->height;
+
+	rc = ioctl(self->fd, VIDIOC_S_FMT, &fmt);
+	if (rc < 0) {
+		return -1;
+	}
+
+	return 0;
+}
+
+static int alloc_dst_buffers(struct h264_encoder_v4l2m2m* self)
+{
+	int n_bufs = ARRAY_LENGTH(self->dst_bufs);
+	int rc;
+
+	struct v4l2_requestbuffers req = {
+		.memory = V4L2_MEMORY_MMAP,
+		.count = n_bufs,
+		.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE,
+	};
+	rc = ioctl(self->fd, VIDIOC_REQBUFS, &req);
+	if (rc < 0)
+		return -1;
+
+	for (unsigned int i = 0; i < req.count; ++i) {
+		struct h264_encoder_v4l2m2m_dst_buf* buffer = &self->dst_bufs[i];
+		struct v4l2_buffer* buf = &buffer->buffer;
+
+		buf->index = i;
+		buf->type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
+		buf->memory = V4L2_MEMORY_MMAP;
+		buf->length = 1;
+		buf->m.planes = &buffer->plane;
+
+		rc = ioctl(self->fd, VIDIOC_QUERYBUF, buf);
+		if (rc < 0)
+			return -1;
+
+		buffer->payload = mmap(0, buffer->plane.length,
+				PROT_READ | PROT_WRITE, MAP_SHARED, self->fd,
+				buffer->plane.m.mem_offset);
+		if (buffer->payload == MAP_FAILED) {
+			nvnc_log(NVNC_LOG_ERROR, "Whoops, mapping failed: %m");
+			return -1;
+		}
+	}
+
+	return 0;
+}
+
+static void enqueue_dst_buffers(struct h264_encoder_v4l2m2m* self)
+{
+	for (unsigned int i = 0; i < ARRAY_LENGTH(self->dst_bufs); ++i) {
+		int rc = v4l2_qbuf(self->fd, &self->dst_bufs[i].buffer);
+		assert(rc >= 0);
+	}
+}
+
+static void process_dst_bufs(struct h264_encoder_v4l2m2m* self)
+{
+	int rc;
+	struct v4l2_plane plane = { 0 };
+	struct v4l2_buffer buf = {
+		.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE,
+		.memory = V4L2_MEMORY_MMAP,
+		.length = 1,
+		.m.planes = &plane,
+	};
+
+	while (true) {
+		rc = v4l2_dqbuf(self->fd, &buf);
+		if (rc < 0)
+			break;
+
+		uint64_t pts = buf.timestamp.tv_sec * UINT64_C(1000000) +
+			buf.timestamp.tv_usec;
+		struct h264_encoder_v4l2m2m_dst_buf* dstbuf =
+			&self->dst_bufs[buf.index];
+		size_t size = buf.m.planes[0].bytesused;
+
+		static uint64_t last_pts;
+		if (last_pts && last_pts > pts) {
+			nvnc_log(NVNC_LOG_ERROR, "pts - last_pts = %"PRIi64,
+					(int64_t)pts - (int64_t)last_pts);
+		}
+		last_pts = pts;
+
+		nvnc_trace("Encoded frame (index %d) at %"PRIu64" µs with size: %zu",
+				buf.index, pts, size);
+
+		self->base.on_packet_ready(dstbuf->payload, size, pts,
+				self->base.userdata);
+
+		v4l2_qbuf(self->fd, &buf);
+	}
+}
+
+static void process_src_bufs(struct h264_encoder_v4l2m2m* self)
+{
+	int rc;
+	struct v4l2_plane planes[4] = { 0 };
+	struct v4l2_buffer buf = {
+		.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE,
+		.memory = V4L2_MEMORY_DMABUF,
+		.length = 1,
+		.m.planes = planes,
+	};
+
+	while (true) {
+		rc = v4l2_dqbuf(self->fd, &buf);
+		if (rc < 0)
+			break;
+
+		struct h264_encoder_v4l2m2m_src_buf* srcbuf =
+			&self->src_bufs[buf.index];
+		srcbuf->is_taken = false;
+
+		// TODO: This assumes that there's only one fd
+		close(srcbuf->planes[0].m.fd);
+
+		nvnc_fb_unmap(srcbuf->fb);
+		nvnc_fb_release(srcbuf->fb);
+		nvnc_fb_unref(srcbuf->fb);
+		srcbuf->fb = NULL;
+	}
+}
+
+static void stream_off(struct h264_encoder_v4l2m2m* self)
+{
+	int type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
+	ioctl(self->fd, VIDIOC_STREAMOFF, &type);
+
+	type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
+	ioctl(self->fd, VIDIOC_STREAMOFF, &type);
+}
+
+static void free_dst_buffers(struct h264_encoder_v4l2m2m* self)
+{
+	for (unsigned int i = 0; i < ARRAY_LENGTH(self->dst_bufs); ++i) {
+		struct h264_encoder_v4l2m2m_dst_buf* buf = &self->dst_bufs[i];
+		munmap(buf->payload, buf->plane.length);
+	}
+}
+
+static int stream_on(struct h264_encoder_v4l2m2m* self)
+{
+	int type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
+	ioctl(self->fd, VIDIOC_STREAMON, &type);
+
+	type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;
+	return ioctl(self->fd, VIDIOC_STREAMON, &type);
+}
+
+static int alloc_src_buffers(struct h264_encoder_v4l2m2m* self)
+{
+	int rc;
+
+	struct v4l2_requestbuffers req = {
+		.memory = V4L2_MEMORY_DMABUF,
+		.count = N_SRC_BUFS,
+		.type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE,
+	};
+	rc = ioctl(self->fd, VIDIOC_REQBUFS, &req);
+	if (rc < 0)
+		return -1;
+
+	for (int i = 0; i < N_SRC_BUFS; ++i) {
+		struct h264_encoder_v4l2m2m_src_buf* buffer = &self->src_bufs[i];
+		struct v4l2_buffer* buf = &buffer->buffer;
+
+		buf->index = i;
+		buf->type = V4L2_BUF_TYPE_VIDEO_OUTPUT_MPLANE;
+		buf->memory = V4L2_MEMORY_DMABUF;
+		buf->length = 1;
+		buf->m.planes = buffer->planes;
+
+		rc = ioctl(self->fd, VIDIOC_QUERYBUF, buf);
+		if (rc < 0)
+			return -1;
+	}
+
+	return 0;
+}
+
+static void force_key_frame(struct h264_encoder_v4l2m2m* self)
+{
+	struct v4l2_control ctrl = { 0 };
+	ctrl.id = V4L2_CID_MPEG_VIDEO_FORCE_KEY_FRAME;
+	ctrl.value = 0;
+	ioctl(self->fd, VIDIOC_S_CTRL, &ctrl);
+}
+
+static void encode_buffer(struct h264_encoder_v4l2m2m* self,
+		struct nvnc_fb* fb)
+{
+	struct h264_encoder_v4l2m2m_src_buf* srcbuf = take_src_buffer(self);
+	if (!srcbuf) {
+		nvnc_log(NVNC_LOG_ERROR, "Out of source buffers. Dropping frame...");
+		return;
+	}
+
+	assert(!srcbuf->fb);
+
+	nvnc_fb_ref(fb);
+	nvnc_fb_hold(fb);
+
+	/* For some reason the v4l2m2m h264 encoder in the Rapberry Pi 4 gets
+	 * really glitchy unless the buffer is mapped first.
+	 * This should probably be handled by the driver, but it's not.
+	 */
+	nvnc_fb_map(fb);
+
+	srcbuf->fb = fb;
+
+	struct gbm_bo* bo = nvnc_fb_get_gbm_bo(fb);
+
+	int n_planes = gbm_bo_get_plane_count(bo);
+	int fd = gbm_bo_get_fd(bo);
+	uint32_t height = ALIGN_UP(gbm_bo_get_height(bo), 16);
+
+	for (int i = 0; i < n_planes; ++i) {
+		uint32_t stride = gbm_bo_get_stride_for_plane(bo, i);
+		uint32_t offset = gbm_bo_get_offset(bo, i);
+		uint32_t size = stride * height;
+
+		srcbuf->buffer.m.planes[i].m.fd = fd;
+		srcbuf->buffer.m.planes[i].bytesused = size;
+		srcbuf->buffer.m.planes[i].length = size;
+		srcbuf->buffer.m.planes[i].data_offset = offset;
+	}
+
+	srcbuf->buffer.timestamp.tv_sec = fb->pts / UINT64_C(1000000);
+	srcbuf->buffer.timestamp.tv_usec = fb->pts % UINT64_C(1000000);
+
+	if (self->base.next_frame_should_be_keyframe)
+		force_key_frame(self);
+	self->base.next_frame_should_be_keyframe = false;
+
+	int rc = v4l2_qbuf(self->fd, &srcbuf->buffer);
+	if (rc < 0) {
+		nvnc_log(NVNC_LOG_PANIC, "Failed to enqueue buffer: %m");
+	}
+}
+
+static void process_fd_events(void* handle)
+{
+	struct h264_encoder_v4l2m2m* self = aml_get_userdata(handle);
+	process_dst_bufs(self);
+}
+
+static void h264_encoder_v4l2m2m_configure(struct h264_encoder_v4l2m2m* self)
+{
+	struct v4l2_control ctrl = { 0 };
+
+	ctrl.id = V4L2_CID_MPEG_VIDEO_H264_PROFILE;
+	ctrl.value = V4L2_MPEG_VIDEO_H264_PROFILE_CONSTRAINED_BASELINE;
+	ioctl(self->fd, VIDIOC_S_CTRL, &ctrl);
+
+	ctrl.id = V4L2_CID_MPEG_VIDEO_H264_I_PERIOD;
+	ctrl.value = INT_MAX;
+	ioctl(self->fd, VIDIOC_S_CTRL, &ctrl);
+
+	ctrl.id = V4L2_CID_MPEG_VIDEO_BITRATE_MODE;
+	ctrl.value = V4L2_MPEG_VIDEO_BITRATE_MODE_CQ;
+	ioctl(self->fd, VIDIOC_S_CTRL, &ctrl);
+
+	ctrl.id = V4L2_CID_MPEG_VIDEO_CONSTANT_QUALITY;
+	ctrl.value = self->quality;
+	ioctl(self->fd, VIDIOC_S_CTRL, &ctrl);
+}
+
+static bool can_encode_to_h264(int fd)
+{
+	size_t i = 0;
+	for (;; ++i) {
+		struct v4l2_fmtdesc desc = {
+			.index = i,
+			.type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE,
+		};
+		int rc = ioctl(fd, VIDIOC_ENUM_FMT, &desc);
+		if (rc < 0)
+			break;
+
+		if (desc.pixelformat == V4L2_PIX_FMT_H264)
+			return true;
+	}
+	return false;
+}
+
+static bool can_handle_frame_size(int fd, uint32_t width, uint32_t height)
+{
+	size_t i = 0;
+	for (;; ++i) {
+		struct v4l2_frmsizeenum size = {
+			.index = i,
+			.pixel_format = V4L2_PIX_FMT_H264,
+		};
+		int rc = ioctl(fd, VIDIOC_ENUM_FRAMESIZES, &size);
+		if (rc < 0)
+			break;
+
+		switch (size.type) {
+		case V4L2_FRMSIZE_TYPE_DISCRETE:
+			if (size.discrete.width == width &&
+					size.discrete.height == height)
+				return true;
+			break;
+		case V4L2_FRMSIZE_TYPE_CONTINUOUS:
+		case V4L2_FRMSIZE_TYPE_STEPWISE:
+			if (size.stepwise.min_width <= width &&
+					width <= size.stepwise.max_width &&
+					size.stepwise.min_height <= height &&
+					height <= size.stepwise.max_height &&
+					(16 % size.stepwise.step_width) == 0 &&
+					(16 % size.stepwise.step_height) == 0)
+				return true;
+			break;
+		}
+	}
+	return false;
+}
+
+static bool is_device_capable(int fd, uint32_t width, uint32_t height)
+{
+	struct v4l2_capability cap = { 0 };
+	int rc = ioctl(fd, VIDIOC_QUERYCAP, &cap);
+	if (rc < 0)
+		return false;
+
+	uint32_t required_caps = V4L2_CAP_VIDEO_M2M_MPLANE | V4L2_CAP_STREAMING;
+	if ((cap.capabilities & required_caps) != required_caps)
+		return false;
+
+	if (!can_encode_to_h264(fd))
+		return false;
+
+	if (!can_handle_frame_size(fd, width, height))
+		return false;
+
+	return true;
+}
+
+static int find_capable_device(uint32_t width, uint32_t height)
+{
+	int fd = -1;
+	DIR *dir = opendir("/dev");
+	assert(dir);
+
+	for (;;) {
+		struct dirent* entry = readdir(dir);
+		if (!entry)
+			break;
+
+		if (strncmp(entry->d_name, "video", 5) != 0)
+			continue;
+
+		char path[256];
+		snprintf(path, sizeof(path), "/dev/%s", entry->d_name);
+		fd = open(path, O_RDWR | O_CLOEXEC);
+		if (fd < 0) {
+			continue;
+		}
+
+		if (is_device_capable(fd, width, height)) {
+			nvnc_log(NVNC_LOG_DEBUG, "Using v4l2m2m device: %s",
+					path);
+			break;
+		}
+		close(fd);
+		fd = -1;
+	}
+
+	closedir(dir);
+	return fd;
+}
+
+static struct h264_encoder* h264_encoder_v4l2m2m_create(uint32_t width,
+		uint32_t height, uint32_t format, int quality)
+{
+	struct h264_encoder_v4l2m2m* self = calloc(1, sizeof(*self));
+	if (!self)
+		return NULL;
+
+	self->base.impl = &h264_encoder_v4l2m2m_impl;
+	self->fd = -1;
+	self->width = width;
+	self->height = height;
+	self->format = format;
+	self->quality = quality;
+
+	self->fd = find_capable_device(width, height);
+	if (self->fd < 0)
+		goto failure;
+
+	struct v4l2_capability cap = { 0 };
+	ioctl(self->fd, VIDIOC_QUERYCAP, &cap);
+	strncpy(self->driver, (const char*)cap.driver, sizeof(self->driver));
+
+	if (set_src_fmt(self) < 0)
+		goto failure;
+
+	if (set_dst_fmt(self) < 0)
+		goto failure;
+
+	h264_encoder_v4l2m2m_configure(self);
+
+	if (alloc_dst_buffers(self) < 0)
+		goto failure;
+
+	if (alloc_src_buffers(self) < 0)
+		goto failure;
+
+	enqueue_dst_buffers(self);
+
+	if (stream_on(self) < 0)
+		goto failure;
+
+	int flags = fcntl(self->fd, F_GETFL);
+	fcntl(self->fd, F_SETFL, flags | O_NONBLOCK);
+
+	self->handler = aml_handler_new(self->fd, process_fd_events, self, NULL);
+	aml_set_event_mask(self->handler, AML_EVENT_READ);
+
+	if (aml_start(aml_get_default(), self->handler) < 0) {
+		aml_unref(self->handler);
+		goto failure;
+	}
+
+	return &self->base;
+
+failure:
+	if (self->fd >= 0)
+		close(self->fd);
+	return NULL;
+}
+
+static void claim_all_src_bufs(
+		struct h264_encoder_v4l2m2m* self)
+{
+	for (;;) {
+		 process_src_bufs(self);
+		 if (!any_src_buf_is_taken(self))
+			 break;
+		 usleep(10000);
+	}
+}
+
+static void h264_encoder_v4l2m2m_destroy(struct h264_encoder* base)
+{
+	struct h264_encoder_v4l2m2m* self = (struct h264_encoder_v4l2m2m*)base;
+	claim_all_src_bufs(self);
+	aml_stop(aml_get_default(), self->handler);
+	aml_unref(self->handler);
+	stream_off(self);
+	free_dst_buffers(self);
+	if (self->fd >= 0)
+		close(self->fd);
+	free(self);
+}
+
+static void h264_encoder_v4l2m2m_feed(struct h264_encoder* base,
+		struct nvnc_fb* fb)
+{
+	struct h264_encoder_v4l2m2m* self = (struct h264_encoder_v4l2m2m*)base;
+	process_src_bufs(self);
+	encode_buffer(self, fb);
+}
+
+struct h264_encoder_impl h264_encoder_v4l2m2m_impl = {
+	.create = h264_encoder_v4l2m2m_create,
+	.destroy = h264_encoder_v4l2m2m_destroy,
+	.feed = h264_encoder_v4l2m2m_feed,
+};

src/h264-encoder.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021 - 2022 Andri Yngvason
+ * Copyright (c) 2024 Andri Yngvason
  *
  * Permission to use, copy, modify, and/or distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -15,606 +15,60 @@
  */
 
 #include "h264-encoder.h"
-#include "neatvnc.h"
-#include "fb.h"
-#include "sys/queue.h"
-#include "vec.h"
-#include "usdt.h"
+#include "config.h"
 
-#include <stdlib.h>
-#include <stdint.h>
-#include <stdbool.h>
-#include <unistd.h>
-#include <assert.h>
-#include <gbm.h>
-#include <xf86drm.h>
-#include <aml.h>
+#ifdef HAVE_FFMPEG
+extern struct h264_encoder_impl h264_encoder_ffmpeg_impl;
+#endif
 
-#include <libavcodec/avcodec.h>
-#include <libavutil/hwcontext.h>
-#include <libavutil/hwcontext_drm.h>
-#include <libavutil/pixdesc.h>
-#include <libavutil/dict.h>
-#include <libavfilter/avfilter.h>
-#include <libavfilter/buffersink.h>
-#include <libavfilter/buffersrc.h>
-
-#include <libdrm/drm_fourcc.h>
-
-struct h264_encoder;
-
-struct fb_queue_entry {
-	struct nvnc_fb* fb;
-	TAILQ_ENTRY(fb_queue_entry) link;
-};
-
-TAILQ_HEAD(fb_queue, fb_queue_entry);
-
-struct h264_encoder {
-	h264_encoder_packet_handler_fn on_packet_ready;
-	void* userdata;
-
-	uint32_t width;
-	uint32_t height;
-	uint32_t format;
-
-	AVRational timebase;
-	AVRational sample_aspect_ratio;
-	enum AVPixelFormat av_pixel_format;
-
-	/* type: AVHWDeviceContext */
-	AVBufferRef* hw_device_ctx;
-
-	/* type: AVHWFramesContext */
-	AVBufferRef* hw_frames_ctx;
-
-	AVCodecContext* codec_ctx;
-
-	AVFilterGraph* filter_graph;
-	AVFilterContext* filter_in;
-	AVFilterContext* filter_out;
-
-	bool next_frame_should_be_keyframe;
-	struct fb_queue fb_queue;
-
-	struct aml_work* work;
-	struct nvnc_fb* current_fb;
-	struct vec current_packet;
-	bool current_frame_is_keyframe;
-
-	bool please_destroy;
-};
-
-static enum AVPixelFormat drm_to_av_pixel_format(uint32_t format)
-{
-	switch (format) {
-	case DRM_FORMAT_XRGB8888:
-	case DRM_FORMAT_ARGB8888:
-		return AV_PIX_FMT_BGR0;
-	case DRM_FORMAT_XBGR8888:
-	case DRM_FORMAT_ABGR8888:
-		return AV_PIX_FMT_RGB0;
-	case DRM_FORMAT_RGBX8888:
-	case DRM_FORMAT_RGBA8888:
-		return AV_PIX_FMT_0BGR;
-	case DRM_FORMAT_BGRX8888:
-	case DRM_FORMAT_BGRA8888:
-		return AV_PIX_FMT_0RGB;
-	}
-
-	return AV_PIX_FMT_NONE;
-}
-
-static void hw_frame_desc_free(void* opaque, uint8_t* data)
-{
-	struct AVDRMFrameDescriptor* desc = (void*)data;
-	assert(desc);
-
-	for (int i = 0; i < desc->nb_objects; ++i)
-		close(desc->objects[i].fd);
-
-	free(desc);
-}
-
-// TODO: Maybe do this once per frame inside nvnc_fb?
-static AVFrame* fb_to_avframe(struct nvnc_fb* fb)
-{
-	struct gbm_bo* bo = fb->bo;
-
-	int n_planes = gbm_bo_get_plane_count(bo);
-
-	AVDRMFrameDescriptor* desc = calloc(1, sizeof(*desc));
-	desc->nb_objects = n_planes;
-
-	desc->nb_layers = 1;
-	desc->layers[0].format = gbm_bo_get_format(bo);
-	desc->layers[0].nb_planes = n_planes;
-
-	for (int i = 0; i < n_planes; ++i) {
-		uint32_t stride = gbm_bo_get_stride_for_plane(bo, i);
-
-		desc->objects[i].fd = gbm_bo_get_fd_for_plane(bo, i);
-		desc->objects[i].size = stride * fb->height;
-		desc->objects[i].format_modifier = gbm_bo_get_modifier(bo);
-
-		desc->layers[0].format = gbm_bo_get_format(bo);
-		desc->layers[0].planes[i].object_index = i;
-		desc->layers[0].planes[i].offset = gbm_bo_get_offset(bo, i);
-		desc->layers[0].planes[i].pitch = stride;
-	}
-
-	AVFrame* frame = av_frame_alloc();
-	if (!frame) {
-		hw_frame_desc_free(NULL, (void*)desc);
-		return NULL;
-	}
-
-	frame->opaque = fb;
-	frame->width = fb->width;
-	frame->height = fb->height;
-	frame->format = AV_PIX_FMT_DRM_PRIME;
-	frame->sample_aspect_ratio = (AVRational){1, 1};
-
-	AVBufferRef* desc_ref = av_buffer_create((void*)desc, sizeof(*desc),
-			hw_frame_desc_free, NULL, 0);
-	if (!desc_ref) {
-		hw_frame_desc_free(NULL, (void*)desc);
-		av_frame_free(&frame);
-		return NULL;
-	}
-
-	frame->buf[0] = desc_ref;
-	frame->data[0] = (void*)desc_ref->data;
-
-	// TODO: Set colorspace?
-
-	return frame;
-}
-
-static struct nvnc_fb* fb_queue_dequeue(struct fb_queue* queue)
-{
-	if (TAILQ_EMPTY(queue))
-		return NULL;
-
-	struct fb_queue_entry* entry = TAILQ_FIRST(queue);
-	TAILQ_REMOVE(queue, entry, link);
-	struct nvnc_fb* fb = entry->fb;
-	free(entry);
-
-	return fb;
-}
-
-static int fb_queue_enqueue(struct fb_queue* queue, struct nvnc_fb* fb)
-{
-	struct fb_queue_entry* entry = calloc(1, sizeof(*entry));
-	if (!entry)
-		return -1;
-
-	entry->fb = fb;
-	nvnc_fb_ref(fb);
-	TAILQ_INSERT_TAIL(queue, entry, link);
-
-	return 0;
-}
-
-static int h264_encoder__init_buffersrc(struct h264_encoder* self)
-{
-	int rc;
-
-	/* Placeholder values are used to pacify input checking and the real
-	 * values are set below.
-	 */
-	rc = avfilter_graph_create_filter(&self->filter_in,
-			avfilter_get_by_name("buffer"), "in",
-			"width=1:height=1:pix_fmt=drm_prime:time_base=1/1", NULL,
-			self->filter_graph);
-	if (rc != 0)
-		return -1;
-
-	AVBufferSrcParameters *params = av_buffersrc_parameters_alloc();
-	if (!params)
-		return -1;
-
-	params->format = AV_PIX_FMT_DRM_PRIME;
-	params->width = self->width;
-	params->height = self->height;
-	params->sample_aspect_ratio = self->sample_aspect_ratio;
-	params->time_base = self->timebase;
-	params->hw_frames_ctx = self->hw_frames_ctx;
-
-	rc = av_buffersrc_parameters_set(self->filter_in, params);
-	assert(rc == 0);
-
-	av_free(params);
-	return 0;
-}
-
-static int h264_encoder__init_filters(struct h264_encoder* self)
-{
-	int rc;
-
-	self->filter_graph = avfilter_graph_alloc();
-	if (!self->filter_graph)
-		return -1;
-
-	rc = h264_encoder__init_buffersrc(self);
-	if (rc != 0)
-		goto failure;
-
-	rc = avfilter_graph_create_filter(&self->filter_out,
-			avfilter_get_by_name("buffersink"), "out", NULL,
-			NULL, self->filter_graph);
-	if (rc != 0)
-		goto failure;
-
-	AVFilterInOut* inputs = avfilter_inout_alloc();
-	if (!inputs)
-		goto failure;
-
-	inputs->name = av_strdup("in");
-	inputs->filter_ctx = self->filter_in;
-	inputs->pad_idx = 0;
-	inputs->next = NULL;
-
-	AVFilterInOut* outputs = avfilter_inout_alloc();
-	if (!outputs) {
-		avfilter_inout_free(&inputs);
-		goto failure;
-	}
-
-	outputs->name = av_strdup("out");
-	outputs->filter_ctx = self->filter_out;
-	outputs->pad_idx = 0;
-	outputs->next = NULL;
-
-	rc = avfilter_graph_parse(self->filter_graph,
-			"hwmap=mode=direct:derive_device=vaapi"
-			",scale_vaapi=format=nv12:mode=fast",
-			outputs, inputs, NULL);
-	if (rc != 0)
-		goto failure;
-
-	assert(self->hw_device_ctx);
-
-	for (unsigned int i = 0; i < self->filter_graph->nb_filters; ++i) {
-		self->filter_graph->filters[i]->hw_device_ctx =
-			av_buffer_ref(self->hw_device_ctx);
-	}
-
-	rc = avfilter_graph_config(self->filter_graph, NULL);
-	if (rc != 0)
-		goto failure;
-
-	return 0;
-
-failure:
-	avfilter_graph_free(&self->filter_graph);
-	return -1;
-}
-
-static int h264_encoder__init_codec_context(struct h264_encoder* self,
-		const AVCodec* codec, int quality)
-{
-	self->codec_ctx = avcodec_alloc_context3(codec);
-	if (!self->codec_ctx)
-		return -1;
-
-	struct AVCodecContext* c = self->codec_ctx;
-	c->width = self->width;
-	c->height = self->height;
-	c->time_base = self->timebase;
-	c->sample_aspect_ratio = self->sample_aspect_ratio;
-	c->pix_fmt = AV_PIX_FMT_VAAPI;
-	c->gop_size = INT32_MAX; /* We'll select key frames manually */
-	c->max_b_frames = 0; /* B-frames are bad for latency */
-	c->global_quality = quality;
-
-	/* open-h264 requires baseline profile, so we use constrained
-	 * baseline.
-	 */
-	c->profile = 578;
-
-	return 0;
-}
-
-static int h264_encoder__init_hw_frames_context(struct h264_encoder* self)
-{
-	self->hw_frames_ctx = av_hwframe_ctx_alloc(self->hw_device_ctx);
-	if (!self->hw_frames_ctx)
-		return -1;
-
-	AVHWFramesContext* c = (AVHWFramesContext*)self->hw_frames_ctx->data;
-	c->format = AV_PIX_FMT_DRM_PRIME;
-	c->sw_format = drm_to_av_pixel_format(self->format);
-	c->width = self->width;
-	c->height = self->height;
-	
-	if (av_hwframe_ctx_init(self->hw_frames_ctx) < 0)
-		av_buffer_unref(&self->hw_frames_ctx);
-
-	return 0;
-}
-
-static int h264_encoder__schedule_work(struct h264_encoder* self)
-{
-	if (self->current_fb)
-		return 0;
-
-	self->current_fb = fb_queue_dequeue(&self->fb_queue);
-	if (!self->current_fb)
-		return 0;
-
-	DTRACE_PROBE1(neatvnc, h264_encode_frame_begin, self->current_fb->pts);
-
-	self->current_frame_is_keyframe = self->next_frame_should_be_keyframe;
-	self->next_frame_should_be_keyframe = false;
-
-	return aml_start(aml_get_default(), self->work);
-}
-
-static int h264_encoder__encode(struct h264_encoder* self, AVFrame* frame_in)
-{
-	int rc;
-
-	rc = av_buffersrc_add_frame_flags(self->filter_in, frame_in,
-			AV_BUFFERSRC_FLAG_KEEP_REF);
-	if (rc != 0)
-		return -1;
-
-	AVFrame* filtered_frame = av_frame_alloc();
-	if (!filtered_frame)
-		return -1;
-
-	rc = av_buffersink_get_frame(self->filter_out, filtered_frame);
-	if (rc != 0)
-		goto get_frame_failure;
-
-	rc = avcodec_send_frame(self->codec_ctx, filtered_frame);
-	if (rc != 0)
-		goto send_frame_failure;
-
-	AVPacket* packet = av_packet_alloc();
-	assert(packet); // TODO
-
-	while (1) {
-		rc = avcodec_receive_packet(self->codec_ctx, packet);
-		if (rc != 0)
-			break;
-
-		vec_append(&self->current_packet, packet->data, packet->size);
-
-		packet->stream_index = 0;
-		av_packet_unref(packet);
-	}
-
-	// Frame should always start with a zero:
-	assert(self->current_packet.len == 0 ||
-			((char*)self->current_packet.data)[0] == 0);
-
-	av_packet_free(&packet);
-send_frame_failure:
-	av_frame_unref(filtered_frame);
-get_frame_failure:
-	av_frame_free(&filtered_frame);
-	return rc == AVERROR(EAGAIN) ? 0 : rc;
-}
-
-static void h264_encoder__do_work(void* handle)
-{
-	struct h264_encoder* self = aml_get_userdata(handle);
-
-	AVFrame* frame = fb_to_avframe(self->current_fb);
-	assert(frame); // TODO
-
-	frame->hw_frames_ctx = av_buffer_ref(self->hw_frames_ctx);
-
-	if (self->current_frame_is_keyframe) {
-		frame->key_frame = 1;
-		frame->pict_type = AV_PICTURE_TYPE_I;
-	} else {
-		frame->key_frame = 0;
-		frame->pict_type = AV_PICTURE_TYPE_P;
-	}
-
-	int rc = h264_encoder__encode(self, frame);
-	if (rc != 0) {
-		char err[256];
-		av_strerror(rc, err, sizeof(err));
-		nvnc_log(NVNC_LOG_ERROR, "Failed to encode packet: %s", err);
-		goto failure;
-	}
-
-failure:
-	av_frame_unref(frame);
-	av_frame_free(&frame);
-}
-
-static void h264_encoder__on_work_done(void* handle)
-{
-	struct h264_encoder* self = aml_get_userdata(handle);
-
-	uint64_t pts = nvnc_fb_get_pts(self->current_fb);
-	nvnc_fb_release(self->current_fb);
-	nvnc_fb_unref(self->current_fb);
-	self->current_fb = NULL;
-
-	DTRACE_PROBE1(neatvnc, h264_encode_frame_end, pts);
-
-	if (self->please_destroy) {
-		vec_destroy(&self->current_packet);
-		h264_encoder_destroy(self);
-		return;
-	}
-
-	if (self->current_packet.len == 0) {
-		nvnc_log(NVNC_LOG_WARNING, "Whoops, encoded packet length is 0");
-		return;
-	}
-
-	void* userdata = self->userdata;
-
-	// Must make a copy of packet because the callback might destroy the
-	// encoder object.
-	struct vec packet;
-	vec_init(&packet, self->current_packet.len);
-	vec_append(&packet, self->current_packet.data,
-			self->current_packet.len);
-
-	vec_clear(&self->current_packet);
-	h264_encoder__schedule_work(self);
-
-	self->on_packet_ready(packet.data, packet.len, pts, userdata);
-	vec_destroy(&packet);
-}
-
-static int find_render_node(char *node, size_t maxlen) {
-	bool r = -1;
-	drmDevice *devices[64];
-
-	int n = drmGetDevices2(0, devices, sizeof(devices) / sizeof(devices[0]));
-	for (int i = 0; i < n; ++i) {
-		drmDevice *dev = devices[i];
-		if (!(dev->available_nodes & (1 << DRM_NODE_RENDER)))
-			continue;
-
-		strncpy(node, dev->nodes[DRM_NODE_RENDER], maxlen);
-		node[maxlen - 1] = '\0';
-		r = 0;
-		break;
-	}
-
-	drmFreeDevices(devices, n);
-	return r;
-}
+#ifdef HAVE_V4L2
+extern struct h264_encoder_impl h264_encoder_v4l2m2m_impl;
+#endif
 
 struct h264_encoder* h264_encoder_create(uint32_t width, uint32_t height,
 		uint32_t format, int quality)
 {
-	int rc;
+	struct h264_encoder* encoder = NULL;
 
-	struct h264_encoder* self = calloc(1, sizeof(*self));
-	if (!self)
-		return NULL;
+#ifdef HAVE_V4L2
+	encoder = h264_encoder_v4l2m2m_impl.create(width, height, format, quality);
+	if (encoder) {
+		return encoder;
+	}
+#endif
 
-	if (vec_init(&self->current_packet, 65536) < 0)
-		goto packet_failure;
+#ifdef HAVE_FFMPEG
+	encoder = h264_encoder_ffmpeg_impl.create(width, height, format, quality);
+	if (encoder) {
+		return encoder;
+	}
+#endif
 
-	self->work = aml_work_new(h264_encoder__do_work,
-			h264_encoder__on_work_done, self, NULL);
-	if (!self->work)
-		goto worker_failure;
-
-	char render_node[64];
-	if (find_render_node(render_node, sizeof(render_node)) < 0)
-		goto render_node_failure;
-
-	rc = av_hwdevice_ctx_create(&self->hw_device_ctx,
-			AV_HWDEVICE_TYPE_DRM, render_node, NULL, 0);
-	if (rc != 0)
-		goto hwdevice_ctx_failure;
-
-	self->next_frame_should_be_keyframe = true;
-	TAILQ_INIT(&self->fb_queue);
-
-	self->width = width;
-	self->height = height;
-	self->format = format;
-	self->timebase = (AVRational){1, 1000000};
-	self->sample_aspect_ratio = (AVRational){1, 1};
-	self->av_pixel_format = drm_to_av_pixel_format(format);
-	if (self->av_pixel_format == AV_PIX_FMT_NONE)
-		goto pix_fmt_failure;
-
-	const AVCodec* codec = avcodec_find_encoder_by_name("h264_vaapi");
-	if (!codec)
-		goto codec_failure;
-
-	if (h264_encoder__init_hw_frames_context(self) < 0)
-		goto hw_frames_context_failure;
-
-	if (h264_encoder__init_filters(self) < 0)
-		goto filter_failure;
-
-	if (h264_encoder__init_codec_context(self, codec, quality) < 0)
-		goto codec_context_failure;
-
-	self->codec_ctx->hw_frames_ctx =
-		av_buffer_ref(self->filter_out->inputs[0]->hw_frames_ctx);
-
-	AVDictionary *opts = NULL;
-	av_dict_set_int(&opts, "async_depth", 1, 0);
-
-	rc = avcodec_open2(self->codec_ctx, codec, &opts);
-	av_dict_free(&opts);
-
-	if (rc != 0)
-		goto avcodec_open_failure;
-
-	return self;
-
-avcodec_open_failure:
-	avcodec_free_context(&self->codec_ctx);
-codec_context_failure:
-filter_failure:
-	av_buffer_unref(&self->hw_frames_ctx);
-hw_frames_context_failure:
-codec_failure:
-pix_fmt_failure:
-	av_buffer_unref(&self->hw_device_ctx);
-hwdevice_ctx_failure:
-render_node_failure:
-	aml_unref(self->work);
-worker_failure:
-	vec_destroy(&self->current_packet);
-packet_failure:
-	free(self);
-	return NULL;
+	return encoder;
 }
 
 void h264_encoder_destroy(struct h264_encoder* self)
 {
-	if (self->current_fb) {
-		self->please_destroy = true;
-		return;
-	}
-
-	vec_destroy(&self->current_packet);
-	av_buffer_unref(&self->hw_frames_ctx);
-	avcodec_free_context(&self->codec_ctx);
-	av_buffer_unref(&self->hw_device_ctx);
-	avfilter_graph_free(&self->filter_graph);
-	aml_unref(self->work);
-	free(self);
+	self->impl->destroy(self);
 }
 
 void h264_encoder_set_packet_handler_fn(struct h264_encoder* self,
-		h264_encoder_packet_handler_fn value)
+		h264_encoder_packet_handler_fn fn)
 {
-	self->on_packet_ready = value;
+	self->on_packet_ready = fn;
 }
 
-void h264_encoder_set_userdata(struct h264_encoder* self, void* value)
+void h264_encoder_set_userdata(struct h264_encoder* self, void* userdata)
 {
-	self->userdata = value;
+	self->userdata = userdata;
+}
+
+void h264_encoder_feed(struct h264_encoder* self, struct nvnc_fb* fb)
+{
+	self->impl->feed(self, fb);
 }
 
 void h264_encoder_request_keyframe(struct h264_encoder* self)
 {
 	self->next_frame_should_be_keyframe = true;
 }
-
-void h264_encoder_feed(struct h264_encoder* self, struct nvnc_fb* fb)
-{
-	assert(fb->type == NVNC_FB_GBM_BO);
-
-	// TODO: Add transform filter
-	assert(fb->transform == NVNC_TRANSFORM_NORMAL);
-
-	int rc = fb_queue_enqueue(&self->fb_queue, fb);
-	assert(rc == 0); // TODO
-
-	nvnc_fb_hold(fb);
-
-	rc = h264_encoder__schedule_work(self);
-	assert(rc == 0); // TODO
-}

src/logging.c

@@ -25,6 +25,7 @@
 #include <stdarg.h>
 #include <string.h>
 #include <ctype.h>
+#include <threads.h>
 
 #ifdef HAVE_LIBAVUTIL
 #include <libavutil/avutil.h>
@@ -32,10 +33,8 @@
 
 #define EXPORT __attribute__((visibility("default")))
 
-static void default_logger(const struct nvnc_log_data* meta,
-		const char* message);
-
-static nvnc_log_fn log_fn = default_logger;
+static nvnc_log_fn log_fn = nvnc_default_logger;
+static thread_local nvnc_log_fn thread_local_log_fn = NULL;
 
 #ifndef NDEBUG
 static enum nvnc_log_level log_level = NVNC_LOG_DEBUG;
@@ -45,6 +44,11 @@ static enum nvnc_log_level log_level = NVNC_LOG_WARNING;
 
 static bool is_initialised = false;
 
+static nvnc_log_fn get_log_fn(void)
+{
+	return thread_local_log_fn ? thread_local_log_fn : log_fn;
+}
+
 static char* trim_left(char* str)
 {
 	while (isspace(*str))
@@ -100,14 +104,15 @@ static void nvnc__vlog(const struct nvnc_log_data* meta, const char* fmt,
 
 	if (meta->level <= log_level) {
 		vsnprintf(message, sizeof(message), fmt, args);
-		log_fn(meta, trim(message));
+		get_log_fn()(meta, trim(message));
 	}
 
 	if (meta->level == NVNC_LOG_PANIC)
 		abort();
 }
 
-static void default_logger(const struct nvnc_log_data* meta,
+EXPORT
+void nvnc_default_logger(const struct nvnc_log_data* meta,
 		const char* message)
 {
 	const char* level = log_level_to_string(meta->level);
@@ -178,7 +183,13 @@ void nvnc_set_log_level(enum nvnc_log_level level)
 EXPORT
 void nvnc_set_log_fn(nvnc_log_fn fn)
 {
-	log_fn = fn;
+	log_fn = fn ? fn : nvnc_default_logger;
+}
+
+EXPORT
+void nvnc_set_log_fn_thread_local(nvnc_log_fn fn)
+{
+	thread_local_log_fn = fn;
 }
 
 EXPORT

src/murmurhash.c

@@ -1,92 +0,0 @@
-/*
- * The MIT License (MIT)
- * 
- * Copyright (c) 2014 Joseph Werle
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to deal
- * in the Software without restriction, including without limitation the rights
- * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
- * copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- * 
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-#include <stdlib.h>
-#include <stdio.h>
-#include <stdint.h>
-#include "murmurhash.h"
-
-uint32_t
-murmurhash (const char *key, uint32_t len, uint32_t seed) {
-  uint32_t c1 = 0xcc9e2d51;
-  uint32_t c2 = 0x1b873593;
-  uint32_t r1 = 15;
-  uint32_t r2 = 13;
-  uint32_t m = 5;
-  uint32_t n = 0xe6546b64;
-  uint32_t h = 0;
-  uint32_t k = 0;
-  uint8_t *d = (uint8_t *) key; // 32 bit extract from `key'
-  const uint32_t *chunks = NULL;
-  const uint8_t *tail = NULL; // tail - last 8 bytes
-  int i = 0;
-  int l = len / 4; // chunk length
-
-  h = seed;
-
-  chunks = (const uint32_t *) (d + l * 4); // body
-  tail = (const uint8_t *) (d + l * 4); // last 8 byte chunk of `key'
-
-  // for each 4 byte chunk of `key'
-  for (i = -l; i != 0; ++i) {
-    // next 4 byte chunk of `key'
-    k = chunks[i];
-
-    // encode next 4 byte chunk of `key'
-    k *= c1;
-    k = (k << r1) | (k >> (32 - r1));
-    k *= c2;
-
-    // append to hash
-    h ^= k;
-    h = (h << r2) | (h >> (32 - r2));
-    h = h * m + n;
-  }
-
-  k = 0;
-
-  // remainder
-  switch (len & 3) { // `len % 4'
-    case 3: k ^= (tail[2] << 16);
-      // fallthrough
-    case 2: k ^= (tail[1] << 8);
-      // fallthrough
-    case 1:
-      k ^= tail[0];
-      k *= c1;
-      k = (k << r1) | (k >> (32 - r1));
-      k *= c2;
-      h ^= k;
-  }
-
-  h ^= len;
-
-  h ^= (h >> 16);
-  h *= 0x85ebca6b;
-  h ^= (h >> 13);
-  h *= 0xc2b2ae35;
-  h ^= (h >> 16);
-
-  return h;
-}

src/pixels.c

@@ -19,20 +19,20 @@
 #include <stdlib.h>
 #include <assert.h>
 #include <libdrm/drm_fourcc.h>
+#include <math.h>
 
 #define POPCOUNT(x) __builtin_popcount(x)
 #define UDIV_UP(a, b) (((a) + (b) - 1) / (b))
 #define XSTR(s) STR(s)
 #define STR(s) #s
 
-void pixel32_to_cpixel(uint8_t* restrict dst,
-                       const struct rfb_pixel_format* dst_fmt,
-                       const uint32_t* restrict src,
-                       const struct rfb_pixel_format* src_fmt,
-                       size_t bytes_per_cpixel, size_t len)
+static void pixel32_to_cpixel(uint8_t* restrict dst,
+		const struct rfb_pixel_format* dst_fmt,
+		const uint32_t* restrict src,
+		const struct rfb_pixel_format* src_fmt,
+		size_t bytes_per_cpixel, size_t len)
 {
 	assert(src_fmt->true_colour_flag);
-	assert(src_fmt->bits_per_pixel == 32);
 	assert(src_fmt->depth <= 32);
 	assert(dst_fmt->true_colour_flag);
 	assert(dst_fmt->bits_per_pixel <= 32);
@@ -152,6 +152,148 @@ void pixel32_to_cpixel(uint8_t* restrict dst,
 #undef CONVERT_PIXELS
 }
 
+void pixel_to_cpixel(uint8_t* restrict dst,
+		const struct rfb_pixel_format* dst_fmt,
+		const uint8_t* restrict src,
+		const struct rfb_pixel_format* src_fmt,
+		size_t bytes_per_cpixel, size_t len)
+{
+	if (src_fmt->bits_per_pixel == 32) {
+		pixel32_to_cpixel(dst, dst_fmt, (uint32_t*)src, src_fmt, bytes_per_cpixel, len);
+		return;
+	}
+
+	assert(src_fmt->true_colour_flag);
+	assert(src_fmt->depth <= 32);
+	assert(dst_fmt->true_colour_flag);
+	assert(dst_fmt->bits_per_pixel <= 32);
+	assert(dst_fmt->depth <= 32);
+	assert(bytes_per_cpixel <= 4 && bytes_per_cpixel >= 1);
+
+	uint32_t src_bpp = src_fmt->bits_per_pixel / 8;
+	uint32_t src_red_shift = src_fmt->red_shift;
+	uint32_t src_green_shift = src_fmt->green_shift;
+	uint32_t src_blue_shift = src_fmt->blue_shift;
+
+	uint32_t dst_red_shift = dst_fmt->red_shift;
+	uint32_t dst_green_shift = dst_fmt->green_shift;
+	uint32_t dst_blue_shift = dst_fmt->blue_shift;
+
+	uint32_t src_red_max = src_fmt->red_max;
+	uint32_t src_green_max = src_fmt->green_max;
+	uint32_t src_blue_max = src_fmt->blue_max;
+
+	uint32_t src_red_bits = POPCOUNT(src_fmt->red_max);
+	uint32_t src_green_bits = POPCOUNT(src_fmt->green_max);
+	uint32_t src_blue_bits = POPCOUNT(src_fmt->blue_max);
+
+	uint32_t dst_red_bits = POPCOUNT(dst_fmt->red_max);
+	uint32_t dst_green_bits = POPCOUNT(dst_fmt->green_max);
+	uint32_t dst_blue_bits = POPCOUNT(dst_fmt->blue_max);
+
+	uint32_t dst_endian_correction;
+
+#define CONVERT_PIXELS(cpx, px)                                                \
+	{                                                                      \
+		uint32_t r, g, b;                                              \
+		r = ((px >> src_red_shift) & src_red_max) << dst_red_bits      \
+		        >> src_red_bits << dst_red_shift;                      \
+		g = ((px >> src_green_shift) & src_green_max) << dst_green_bits\
+		        >> src_green_bits << dst_green_shift;                  \
+		b = ((px >> src_blue_shift) & src_blue_max) << dst_blue_bits   \
+		        >> src_blue_bits << dst_blue_shift;                    \
+		cpx = r | g | b;                                               \
+	}
+
+	switch (bytes_per_cpixel) {
+	case 4:
+		if (dst_fmt->big_endian_flag) {
+			while (len--) {
+				uint32_t cpx, px = 0;
+				memcpy(&px, src, src_bpp);
+				src += src_bpp;
+
+				CONVERT_PIXELS(cpx, px)
+
+				*dst++ = (cpx >> 24) & 0xff;
+				*dst++ = (cpx >> 16) & 0xff;
+				*dst++ = (cpx >> 8) & 0xff;
+				*dst++ = (cpx >> 0) & 0xff;
+			}
+		} else {
+			while (len--) {
+				uint32_t cpx, px = 0;
+				memcpy(&px, src, src_bpp);
+				src += src_bpp;
+
+				CONVERT_PIXELS(cpx, px)
+
+				*dst++ = (cpx >> 0) & 0xff;
+				*dst++ = (cpx >> 8) & 0xff;
+				*dst++ = (cpx >> 16) & 0xff;
+				*dst++ = (cpx >> 24) & 0xff;
+			}
+		}
+		break;
+	case 3:
+		if (dst_fmt->bits_per_pixel == 32 && dst_fmt->depth <= 24) {
+			uint32_t min_dst_shift = dst_red_shift;
+			if (min_dst_shift > dst_green_shift)
+				min_dst_shift = dst_green_shift;
+			if (min_dst_shift > dst_blue_shift)
+				min_dst_shift = dst_blue_shift;
+
+			dst_red_shift -= min_dst_shift;
+			dst_green_shift -= min_dst_shift;
+			dst_blue_shift -= min_dst_shift;
+		}
+
+		dst_endian_correction = dst_fmt->big_endian_flag ? 16 : 0;
+
+		while (len--) {
+			uint32_t cpx, px = 0;
+			memcpy(&px, src, src_bpp);
+			src += src_bpp;
+
+			CONVERT_PIXELS(cpx, px)
+
+			*dst++ = (cpx >> (0 ^ dst_endian_correction)) & 0xff;
+			*dst++ = (cpx >> 8) & 0xff;
+			*dst++ = (cpx >> (16 ^ dst_endian_correction)) & 0xff;
+		}
+		break;
+	case 2:
+		dst_endian_correction = dst_fmt->big_endian_flag ? 8 : 0;
+
+		while (len--) {
+			uint32_t cpx, px = 0;
+			memcpy(&px, src, src_bpp);
+			src += src_bpp;
+
+			CONVERT_PIXELS(cpx, px)
+
+			*dst++ = (cpx >> (0 ^ dst_endian_correction)) & 0xff;
+			*dst++ = (cpx >> (8 ^ dst_endian_correction)) & 0xff;
+		}
+		break;
+	case 1:
+		while (len--) {
+			uint32_t cpx, px = 0;
+			memcpy(&px, src, src_bpp);
+			src += src_bpp;
+
+			CONVERT_PIXELS(cpx, px)
+
+			*dst++ = cpx & 0xff;
+		}
+		break;
+	default:
+		abort();
+	}
+
+#undef CONVERT_PIXELS
+}
+
 /* clang-format off */
 int rfb_pixfmt_from_fourcc(struct rfb_pixel_format *dst, uint32_t src) {
 	switch (src & ~DRM_FORMAT_BIG_ENDIAN) {
@@ -215,6 +357,22 @@ bpp_32:
 		dst->green_max = 0xff;
 		dst->blue_max = 0xff;
 		break;
+	case DRM_FORMAT_BGR888:
+		dst->red_shift = 0;
+		dst->green_shift = 8;
+		dst->blue_shift = 16;
+		goto bpp_24;
+	case DRM_FORMAT_RGB888:
+		dst->red_shift = 16;
+		dst->green_shift = 8;
+		dst->blue_shift = 0;
+bpp_24:
+		dst->bits_per_pixel = 24;
+		dst->depth = 24;
+		dst->red_max = 0xff;
+		dst->green_max = 0xff;
+		dst->blue_max = 0xff;
+		break;
 	case DRM_FORMAT_RGBA4444:
 	case DRM_FORMAT_RGBX4444:
 		dst->red_shift = 12;
@@ -275,6 +433,9 @@ int pixel_size_from_fourcc(uint32_t fourcc)
 	case DRM_FORMAT_ABGR8888:
 	case DRM_FORMAT_XBGR8888:
 		return 4;
+	case DRM_FORMAT_BGR888:
+	case DRM_FORMAT_RGB888:
+		return 3;
 	case DRM_FORMAT_RGBA4444:
 	case DRM_FORMAT_RGBX4444:
 	case DRM_FORMAT_BGRA4444:
@@ -457,6 +618,8 @@ const char* drm_format_to_string(uint32_t fmt)
 	X(XRGB8888) \
 	X(ABGR8888) \
 	X(XBGR8888) \
+	X(RGB888) \
+	X(BGR888) \
 	X(RGBA4444) \
 	X(RGBX4444) \
 	X(BGRA4444) \
@@ -474,9 +637,6 @@ const char* drm_format_to_string(uint32_t fmt)
 // Not exact, but close enough for debugging
 const char* rfb_pixfmt_to_string(const struct rfb_pixel_format* fmt)
 {
-	if (!(fmt->red_max == fmt->green_max && fmt->red_max == fmt->blue_max))
-		goto failure;
-
 	uint32_t profile = (fmt->red_shift << 16) | (fmt->green_shift << 8)
 		| (fmt->blue_shift);
 
@@ -495,9 +655,25 @@ const char* rfb_pixfmt_to_string(const struct rfb_pixel_format* fmt)
 	CASE(8, 4, 0):   return "XRGB4444";
 	CASE(0, 4, 8):   return "XBGR4444";
 	CASE(11, 5, 0):  return "RGB565";
+	CASE(5, 2, 0):   return "RGB332";
+	CASE(0, 2, 5):   return "RGB332";
+	CASE(4, 2, 0):   return "RGB222";
+	CASE(0, 2, 4):   return "BGR222";
 #undef CASE
 	}
-
-failure:
 	return "UNKNOWN";
 }
+
+void make_rgb332_pal8_map(struct rfb_set_colour_map_entries_msg* msg)
+{
+	msg->type = RFB_SERVER_TO_CLIENT_SET_COLOUR_MAP_ENTRIES;
+	msg->padding = 0;
+	msg->first_colour = htons(0);
+	msg->n_colours = htons(256);
+
+	for (unsigned int i = 0; i < 256; ++i) {
+		msg->colours[i].r = htons(round(65535.0 / 7.0 * ((i >> 5) & 7)));
+		msg->colours[i].g = htons(round(65535.0 / 7.0 * ((i >> 2) & 7)));
+		msg->colours[i].b = htons(round(65535.0 / 3.0 * (i & 3)));
+	}
+}

src/raw-encoding.c

@@ -54,10 +54,10 @@ static inline struct raw_encoder* raw_encoder(struct encoder* encoder)
 }
 
 static int raw_encode_box(struct raw_encoder_work* ctx, struct vec* dst,
-                          const struct rfb_pixel_format* dst_fmt,
-                          const struct nvnc_fb* fb,
-                          const struct rfb_pixel_format* src_fmt, int x_start,
-                          int y_start, int stride, int width, int height)
+		const struct rfb_pixel_format* dst_fmt,
+		const struct nvnc_fb* fb,
+		const struct rfb_pixel_format* src_fmt, int x_start,
+		int y_start, int stride, int width, int height)
 {
 	uint16_t x_pos = ctx->x_pos;
 	uint16_t y_pos = ctx->y_pos;
@@ -69,7 +69,10 @@ static int raw_encode_box(struct raw_encoder_work* ctx, struct vec* dst,
 	if (rc < 0)
 		return -1;
 
-	uint32_t* b = fb->addr;
+	uint8_t* b = fb->addr;
+	int32_t src_bpp = src_fmt->bits_per_pixel / 8;
+	int32_t xoff = x_start * src_bpp;
+	int32_t src_stride = fb->stride * src_bpp;
 
 	int bpp = dst_fmt->bits_per_pixel / 8;
 
@@ -80,9 +83,9 @@ static int raw_encode_box(struct raw_encoder_work* ctx, struct vec* dst,
 	uint8_t* d = dst->data;
 
 	for (int y = y_start; y < y_start + height; ++y) {
-		pixel32_to_cpixel(d + dst->len, dst_fmt,
-		                  b + x_start + y * stride, src_fmt,
-		                  bpp, width);
+		pixel_to_cpixel(d + dst->len, dst_fmt,
+				b + xoff + y * src_stride, src_fmt,
+				bpp, width);
 		dst->len += width * bpp;
 	}
 
@@ -114,7 +117,7 @@ static int raw_encode_frame(struct raw_encoder_work* ctx, struct vec* dst,
 		int box_height = box[i].y2 - y;
 
 		rc = raw_encode_box(ctx, dst, dst_fmt, src, src_fmt, x, y,
-				    src->stride, box_width, box_height);
+				src->stride, box_width, box_height);
 		if (rc < 0)
 			return -1;
 	}
@@ -131,7 +134,7 @@ static void raw_encoder_do_work(void* obj)
 	struct nvnc_fb* fb = ctx->fb;
 	assert(fb);
 
-	size_t bpp = nvnc_fb_get_pixel_size(fb);
+	size_t bpp = ctx->output_format.bits_per_pixel / 8;
 	size_t n_rects = pixman_region_n_rects(&ctx->damage);
 	if (n_rects > UINT16_MAX)
 		n_rects = 1;

src/stream-gnutls.c

@@ -82,6 +82,8 @@ static int stream_gnutls__flush(struct stream* base)
 {
 	struct stream_gnutls* self = (struct stream_gnutls*)base;
 	while (!TAILQ_EMPTY(&self->base.send_queue)) {
+		assert(self->base.state != STREAM_STATE_CLOSED);
+
 		struct stream_req* req = TAILQ_FIRST(&self->base.send_queue);
 
 		ssize_t rc = gnutls_record_send(self->session,
@@ -167,7 +169,7 @@ static void stream_gnutls__on_event(void* obj)
 }
 
 static int stream_gnutls_send(struct stream* self, struct rcbuf* payload,
-                stream_req_fn on_done, void* userdata)
+		stream_req_fn on_done, void* userdata)
 {
 	if (self->state == STREAM_STATE_CLOSED)
 		return -1;
@@ -264,7 +266,7 @@ int stream_upgrade_to_tls(struct stream* base, void* context)
 		goto failure;
 
 	rc = gnutls_credentials_set(self->session, GNUTLS_CRD_CERTIFICATE,
-	                            context);
+			context);
 	if (rc != GNUTLS_E_SUCCESS)
 		goto failure;
 

src/stream-rsa-aes.c

@@ -0,0 +1,211 @@
+/*
+ * Copyright (c) 2023 Andri Yngvason
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
+ * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+ * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+ * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+ * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+#include <assert.h>
+#include <errno.h>
+#include <sys/param.h>
+#include <arpa/inet.h>
+
+#include "rcbuf.h"
+#include "stream.h"
+#include "stream-tcp.h"
+#include "stream-common.h"
+#include "crypto.h"
+#include "neatvnc.h"
+
+#define RSA_AES_BUFFER_SIZE 8192
+#define UDIV_UP(a, b) (((a) + (b) - 1) / (b))
+
+struct stream_rsa_aes {
+	struct stream base;
+
+	size_t read_index;
+	uint8_t* read_buffer;
+
+	struct crypto_cipher* cipher;
+};
+
+static_assert(sizeof(struct stream_rsa_aes) <= STREAM_ALLOC_SIZE,
+		"struct stream_rsa_aes has grown too large, increase STREAM_ALLOC_SIZE");
+
+static void stream_rsa_aes_destroy(struct stream* base)
+{
+	struct stream_rsa_aes* self = (struct stream_rsa_aes*)base;
+	crypto_cipher_del(self->cipher);
+	free(self->read_buffer);
+	stream_tcp_destroy(base);
+}
+
+static void stream_rsa_aes_read_into_buffer(struct stream_rsa_aes* self)
+{
+	ssize_t n_read = stream_tcp_read(&self->base,
+			self->read_buffer + self->read_index,
+			RSA_AES_BUFFER_SIZE - self->read_index);
+	if (n_read > 0)
+		self->read_index += n_read;
+}
+
+static ssize_t stream_rsa_aes_parse_header(struct stream_rsa_aes* self)
+{
+	if (self->read_index <= 2) {
+		return -1;
+	}
+
+	uint16_t len_be;
+	memcpy(&len_be, self->read_buffer, sizeof(len_be));
+	size_t len = ntohs(len_be);
+
+	if (self->read_index < 2 + 16 + len) {
+		return -1;
+	}
+
+	return len;
+}
+
+static ssize_t stream_rsa_aes_read_message(struct stream_rsa_aes* self,
+		uint8_t* dst, size_t size)
+{
+	ssize_t msg_len = stream_rsa_aes_parse_header(self);
+	if (msg_len < 0) {
+		return 0;
+	}
+
+	// The entire message must fit in dst
+	/* TODO: With this, stream_tcp__on_event won't run until network input
+	 * is received. We need to somehow schedule on_event or also buffer the
+	 * decrypted data here.
+	 * Another option would be to keep back the message counter in the
+	 * cipher until the message has been fully read.
+	 */
+	if ((size_t)msg_len > size)
+		return 0;
+
+	uint16_t msg_len_be = htons(msg_len);
+
+	uint8_t expected_mac[16];
+	ssize_t n = crypto_cipher_decrypt(self->cipher, dst, expected_mac,
+			self->read_buffer + 2, msg_len,
+			(uint8_t*)&msg_len_be, sizeof(msg_len_be));
+
+	uint8_t* actual_mac = self->read_buffer + 2 + msg_len;
+	if (memcmp(expected_mac, actual_mac, 16) != 0) {
+		nvnc_log(NVNC_LOG_DEBUG, "Message authentication failed");
+		errno = EBADMSG;
+		return -1;
+	}
+
+	self->read_index -= 2 + 16 + msg_len;
+	memmove(self->read_buffer, self->read_buffer + 2 + 16 + msg_len,
+			self->read_index);
+
+	return n;
+}
+
+static ssize_t stream_rsa_aes_read(struct stream* base, void* dst, size_t size)
+{
+	struct stream_rsa_aes* self = (struct stream_rsa_aes*)base;
+
+	stream_rsa_aes_read_into_buffer(self);
+	if (self->base.state == STREAM_STATE_CLOSED)
+		return 0;
+
+	size_t total_read = 0;
+
+	while (true) {
+		ssize_t n_read = stream_rsa_aes_read_message(self, dst, size);
+		if (n_read == 0)
+			break;
+
+		if (n_read < 0) {
+			if (errno == EAGAIN) {
+				break;
+			}
+			return -1;
+		}
+
+		total_read += n_read;
+		dst += n_read;
+		size -= n_read;
+	}
+
+	return total_read;
+}
+
+static int stream_rsa_aes_send(struct stream* base, struct rcbuf* payload,
+		stream_req_fn on_done, void* userdata)
+{
+	struct stream_rsa_aes* self = (struct stream_rsa_aes*)base;
+	size_t n_msg = UDIV_UP(payload->size, RSA_AES_BUFFER_SIZE);
+
+	struct vec buf;
+	vec_init(&buf, payload->size + n_msg * (2 + 16));
+
+	for (size_t i = 0; i < n_msg; ++i) {
+		size_t msglen = MIN(payload->size - i * RSA_AES_BUFFER_SIZE,
+				RSA_AES_BUFFER_SIZE);
+		uint16_t msglen_be = htons(msglen);
+
+		vec_append(&buf, &msglen_be, sizeof(msglen_be));
+
+		uint8_t mac[16];
+		crypto_cipher_encrypt(self->cipher, &buf, mac,
+				payload->payload + i * RSA_AES_BUFFER_SIZE,
+				msglen, (uint8_t*)&msglen_be, sizeof(msglen_be));
+		vec_append(&buf, mac, sizeof(mac));
+	}
+
+	size_t payload_size = payload->size;
+	rcbuf_unref(payload);
+
+	int r = stream_tcp_send(base, rcbuf_new(buf.data, buf.len), on_done,
+			userdata);
+	if (r < 0) {
+		return r;
+	}
+
+	return payload_size;
+}
+
+static struct stream_impl impl = {
+	.close = stream_tcp_close,
+	.destroy = stream_rsa_aes_destroy,
+	.read = stream_rsa_aes_read,
+	.send = stream_rsa_aes_send,
+};
+
+int stream_upgrade_to_rsa_eas(struct stream* base,
+		enum crypto_cipher_type cipher_type,
+		const uint8_t* enc_key, const uint8_t* dec_key)
+{
+	struct stream_rsa_aes* self = (struct stream_rsa_aes*)base;
+
+	self->read_index = 0;
+	self->read_buffer = malloc(RSA_AES_BUFFER_SIZE);
+	if (!self->read_buffer)
+		return -1;
+
+	self->cipher = crypto_cipher_new(enc_key, dec_key, cipher_type);
+	if (!self->cipher) {
+		free(self->read_buffer);
+		return -1;
+	}
+
+	self->base.impl = &impl;
+	return 0;
+}

src/stream-tcp.c

@@ -30,12 +30,14 @@
 #include "rcbuf.h"
 #include "stream.h"
 #include "stream-common.h"
+#include "stream-tcp.h"
 #include "sys/queue.h"
+#include "neatvnc.h"
 
 static_assert(sizeof(struct stream) <= STREAM_ALLOC_SIZE,
 		"struct stream has grown too large, increase STREAM_ALLOC_SIZE");
 
-static int stream_tcp_close(struct stream* self)
+int stream_tcp_close(struct stream* self)
 {
 	if (self->state == STREAM_STATE_CLOSED)
 		return -1;
@@ -55,8 +57,9 @@ static int stream_tcp_close(struct stream* self)
 	return 0;
 }
 
-static void stream_tcp_destroy(struct stream* self)
+void stream_tcp_destroy(struct stream* self)
 {
+	vec_destroy(&self->tmp_buf);
 	stream_close(self);
 	aml_unref(self->handler);
 	free(self);
@@ -76,7 +79,8 @@ static int stream_tcp__flush(struct stream* self)
 		if (req->exec) {
 			if (req->payload)
 				rcbuf_unref(req->payload);
-			req->payload = req->exec(self, req->userdata);
+			struct rcbuf* payload  = req->exec(self, req->userdata);
+			req->payload = payload;
 		}
 
 		iov[n_msgs].iov_base = req->payload->payload;
@@ -182,21 +186,29 @@ static void stream_tcp__on_event(void* obj)
 		stream_tcp__on_writable(self);
 }
 
-static ssize_t stream_tcp_read(struct stream* self, void* dst, size_t size)
+ssize_t stream_tcp_read(struct stream* self, void* dst, size_t size)
 {
 	if (self->state != STREAM_STATE_NORMAL)
 		return -1;
 
-	ssize_t rc = read(self->fd, dst, size);
+	uint8_t* read_buffer = dst;
+
+	if (self->cipher) {
+		vec_reserve(&self->tmp_buf, size);
+		read_buffer = self->tmp_buf.data;
+	}
+
+	ssize_t rc = read(self->fd, read_buffer, size);
 	if (rc == 0)
 		stream__remote_closed(self);
 	if (rc > 0)
 		self->bytes_received += rc;
+
 	return rc;
 }
 
-static int stream_tcp_send(struct stream* self, struct rcbuf* payload,
-                stream_req_fn on_done, void* userdata)
+int stream_tcp_send(struct stream* self, struct rcbuf* payload,
+		stream_req_fn on_done, void* userdata)
 {
 	if (self->state == STREAM_STATE_CLOSED)
 		return -1;
@@ -214,7 +226,7 @@ static int stream_tcp_send(struct stream* self, struct rcbuf* payload,
 	return stream_tcp__flush(self);
 }
 
-static int stream_tcp_send_first(struct stream* self, struct rcbuf* payload)
+int stream_tcp_send_first(struct stream* self, struct rcbuf* payload)
 {
 	if (self->state == STREAM_STATE_CLOSED)
 		return -1;
@@ -229,7 +241,7 @@ static int stream_tcp_send_first(struct stream* self, struct rcbuf* payload)
 	return stream_tcp__flush(self);
 }
 
-static void stream_tcp_exec_and_send(struct stream* self,
+void stream_tcp_exec_and_send(struct stream* self,
 		stream_exec_fn exec_fn, void* userdata)
 {
 	if (self->state == STREAM_STATE_CLOSED)
@@ -256,12 +268,9 @@ static struct stream_impl impl = {
 	.exec_and_send = stream_tcp_exec_and_send,
 };
 
-struct stream* stream_new(int fd, stream_event_fn on_event, void* userdata)
+int stream_tcp_init(struct stream* self, int fd, stream_event_fn on_event,
+		void* userdata)
 {
-	struct stream* self = calloc(1, STREAM_ALLOC_SIZE);
-	if (!self)
-		return NULL;
-
 	self->impl = &impl,
 	self->fd = fd;
 	self->on_event = on_event;
@@ -273,19 +282,31 @@ struct stream* stream_new(int fd, stream_event_fn on_event, void* userdata)
 
 	self->handler = aml_handler_new(fd, stream_tcp__on_event, self, NULL);
 	if (!self->handler)
-		goto failure;
+		return -1;
 
 	if (aml_start(aml_get_default(), self->handler) < 0)
 		goto start_failure;
 
 	stream__poll_r(self);
 
-	return self;
+	return 0;
 
 start_failure:
 	aml_unref(self->handler);
-	self = NULL; /* Handled in unref */
-failure:
-	free(self);
-	return NULL;
+	return -1;
+}
+
+struct stream* stream_new(int fd, stream_event_fn on_event, void* userdata)
+{
+	struct stream* self = calloc(1, STREAM_ALLOC_SIZE);
+	if (!self)
+		return NULL;
+
+	if (stream_tcp_init(self, fd, on_event, userdata) < 0) {
+		free(self);
+		return NULL;
+	}
+
+	return self;
+
 }

src/stream-ws.c

@@ -16,6 +16,7 @@
 
 #include "stream.h"
 #include "stream-common.h"
+#include "stream-tcp.h"
 #include "websocket.h"
 #include "vec.h"
 #include "neatvnc.h"
@@ -38,31 +39,17 @@ struct stream_ws_exec_ctx {
 
 struct stream_ws {
 	struct stream base;
+	stream_event_fn on_event;
 	enum stream_ws_state ws_state;
 	struct ws_frame_header header;
 	enum ws_opcode current_opcode;
-	uint8_t read_buffer[4096]; // TODO: Is this a reasonable size?
 	size_t read_index;
-	struct stream* tcp_stream;
+	uint8_t read_buffer[4096]; // TODO: Is this a reasonable size?
 };
 
-static int stream_ws_close(struct stream* self)
-{
-	struct stream_ws* ws = (struct stream_ws*)self;
-	self->state = STREAM_STATE_CLOSED;
-	return stream_close(ws->tcp_stream);
-}
-
-static void stream_ws_destroy(struct stream* self)
-{
-	struct stream_ws* ws = (struct stream_ws*)self;
-	stream_destroy(ws->tcp_stream);
-	free(self);
-}
-
 static void stream_ws_read_into_buffer(struct stream_ws* ws)
 {
-	ssize_t n_read = stream_read(ws->tcp_stream,
+	ssize_t n_read = stream_tcp_read(&ws->base,
 			ws->read_buffer + ws->read_index,
 			sizeof(ws->read_buffer) - ws->read_index);
 	if (n_read > 0)
@@ -104,14 +91,15 @@ static ssize_t stream_ws_process_ping(struct stream_ws* ws, size_t offset)
 
 		uint8_t buf[WS_HEADER_MIN_SIZE];
 		int reply_len = ws_write_frame_header(buf, &reply);
-		stream_write(ws->tcp_stream, buf, reply_len, NULL, NULL);
+		stream_tcp_send(&ws->base, rcbuf_from_mem(buf, reply_len),
+				NULL, NULL);
 	}
 
 	int payload_len = MIN(ws->read_index, ws->header.payload_length);
 
 	// Feed back the payload:
-	stream_write(ws->tcp_stream, ws->read_buffer + offset,
-			payload_len, NULL, NULL);
+	stream_tcp_send(&ws->base, rcbuf_from_mem(ws->read_buffer + offset,
+			payload_len), NULL, NULL);
 
 	stream_ws_advance_read_buffer(ws, payload_len, offset);
 	return 0;
@@ -124,7 +112,7 @@ static ssize_t stream_ws_process_payload(struct stream_ws* ws, void* dst,
 	case WS_OPCODE_CONT:
 		// Remote end started with a continuation frame. This is
 		// unexpected, so we'll just close.
-		stream__remote_closed(ws->tcp_stream);
+		stream__remote_closed(&ws->base);
 		return 0;
 	case WS_OPCODE_TEXT:
 		// This is unexpected, but let's just ignore it...
@@ -133,7 +121,7 @@ static ssize_t stream_ws_process_payload(struct stream_ws* ws, void* dst,
 	case WS_OPCODE_BIN:
 		return stream_ws_copy_payload(ws, dst, size, offset);
 	case WS_OPCODE_CLOSE:
-		stream__remote_closed(ws->tcp_stream);
+		stream__remote_closed(&ws->base);
 		return 0;
 	case WS_OPCODE_PING:
 		return stream_ws_process_ping(ws, offset);
@@ -157,7 +145,7 @@ static ssize_t stream_ws_read_frame(struct stream_ws* ws, void* dst,
 	}
 
 	if (!ws_parse_frame_header(&ws->header, ws->read_buffer,
-				ws->read_index)) {
+			ws->read_index)) {
 		return 0;
 	}
 
@@ -213,8 +201,8 @@ static ssize_t stream_ws_read_handshake(struct stream_ws* ws, void* dst,
 	if (header_len < 0)
 		return 0;
 
-	ws->tcp_stream->cork = false;
-	stream_send_first(ws->tcp_stream, rcbuf_from_mem(reply, strlen(reply)));
+	ws->base.cork = false;
+	stream_tcp_send_first(&ws->base, rcbuf_from_mem(reply, strlen(reply)));
 
 	ws->read_index -= header_len;
 	memmove(ws->read_buffer, ws->read_buffer + header_len, ws->read_index);
@@ -242,7 +230,7 @@ static ssize_t stream_ws_read(struct stream* self, void* dst, size_t size)
 }
 
 static int stream_ws_send(struct stream* self, struct rcbuf* payload,
-                stream_req_fn on_done, void* userdata)
+		stream_req_fn on_done, void* userdata)
 {
 	struct stream_ws* ws = (struct stream_ws*)self;
 
@@ -255,18 +243,23 @@ static int stream_ws_send(struct stream* self, struct rcbuf* payload,
 	uint8_t raw_head[WS_HEADER_MIN_SIZE];
 	int head_len = ws_write_frame_header(raw_head, &head);
 
-	stream_write(ws->tcp_stream, &raw_head, head_len, NULL, NULL);
-	return stream_send(ws->tcp_stream, payload, on_done, userdata);
+	stream_tcp_send(&ws->base, rcbuf_from_mem(&raw_head, head_len),
+			NULL, NULL);
+	return stream_tcp_send(&ws->base, payload, on_done, userdata);
 }
 
 static struct rcbuf* stream_ws_chained_exec(struct stream* tcp_stream,
 		void* userdata)
 {
+	struct stream_ws* ws = (struct stream_ws*)tcp_stream;
 	struct stream_ws_exec_ctx* ctx = userdata;
-	struct stream_ws* ws = tcp_stream->userdata;
 
 	struct rcbuf* buf = ctx->exec(&ws->base, ctx->userdata);
 
+	// TODO: This also needs to be cleaned it it's left on the send queue
+	// when the stream is destroyed.
+	free(ctx->userdata);
+
 	struct vec out;
 	vec_init(&out, WS_HEADER_MIN_SIZE + buf->size + 1);
 
@@ -279,6 +272,7 @@ static struct rcbuf* stream_ws_chained_exec(struct stream* tcp_stream,
 	out.len += head_len;
 
 	vec_append(&out, buf->payload, buf->size);
+	rcbuf_unref(buf);
 	return rcbuf_new(out.data, out.len);
 }
 
@@ -293,23 +287,12 @@ static void stream_ws_exec_and_send(struct stream* self, stream_exec_fn exec,
 	ctx->exec = exec;
 	ctx->userdata = userdata;
 
-	stream_exec_and_send(ws->tcp_stream, stream_ws_chained_exec, ctx);
-}
-
-static void stream_ws_event(struct stream* self, enum stream_event event)
-{
-	struct stream_ws* ws = self->userdata;
-
-	if (event == STREAM_EVENT_REMOTE_CLOSED) {
-		ws->base.state = STREAM_STATE_CLOSED;
-	}
-
-	ws->base.on_event(&ws->base, event);
+	stream_tcp_exec_and_send(&ws->base, stream_ws_chained_exec, ctx);
 }
 
 static struct stream_impl impl = {
-	.close = stream_ws_close,
-	.destroy = stream_ws_destroy,
+	.close = stream_tcp_close,
+	.destroy = stream_tcp_destroy,
 	.read = stream_ws_read,
 	.send = stream_ws_send,
 	.exec_and_send = stream_ws_exec_and_send,
@@ -321,19 +304,11 @@ struct stream* stream_ws_new(int fd, stream_event_fn on_event, void* userdata)
 	if (!self)
 		return NULL;
 
-	self->base.state = STREAM_STATE_NORMAL;
-	self->base.impl = &impl;
-	self->base.on_event = on_event;
-	self->base.userdata = userdata;
-
-	self->tcp_stream = stream_new(fd, stream_ws_event, self);
-	if (!self->tcp_stream) {
-		free(self);
-		return NULL;
-	}
+	stream_tcp_init(&self->base, fd, on_event, userdata);
+	self->base.impl =  &impl;
 
 	// Don't send anything until handshake is done:
-	self->tcp_stream->cork = true;
+	self->base.cork = true;
 
 	return &self->base;
 }

src/stream.c

@@ -31,7 +31,7 @@ void stream_destroy(struct stream* self)
 }
 
 int stream_send(struct stream* self, struct rcbuf* payload,
-                stream_req_fn on_done, void* userdata)
+		stream_req_fn on_done, void* userdata)
 {
 	assert(self->impl && self->impl->send);
 	return self->impl->send(self, payload, on_done, userdata);
@@ -44,7 +44,7 @@ int stream_send_first(struct stream* self, struct rcbuf* payload)
 }
 
 int stream_write(struct stream* self, const void* payload, size_t len,
-                 stream_req_fn on_done, void* userdata)
+		stream_req_fn on_done, void* userdata)
 {
 	struct rcbuf* buf = rcbuf_from_mem(payload, len);
 	return buf ? stream_send(self, buf, on_done, userdata) : -1;

src/tight.c

@@ -119,11 +119,11 @@ static inline struct tight_encoder* tight_encoder(struct encoder* encoder)
 static int tight_encoder_init_stream(z_stream* zs)
 {
 	int rc = deflateInit2(zs,
-	                      /* compression level: */ 1,
-	                      /*            method: */ Z_DEFLATED,
-	                      /*       window bits: */ 15,
-	                      /*         mem level: */ 9,
-	                      /*          strategy: */ Z_DEFAULT_STRATEGY);
+			/* compression level: */ 1,
+			/*            method: */ Z_DEFLATED,
+			/*       window bits: */ 15,
+			/*         mem level: */ 9,
+			/*          strategy: */ Z_DEFAULT_STRATEGY);
 	return rc == Z_OK ? 0 : -1;
 }
 
@@ -261,7 +261,7 @@ static void tight_encode_size(struct vec* dst, size_t size)
 }
 
 static int tight_deflate(struct tight_tile* tile, void* src,
-			 size_t len, z_stream* zs, bool flush)
+		size_t len, z_stream* zs, bool flush)
 {
 	zs->next_in = src;
 	zs->avail_in = len;
@@ -302,13 +302,14 @@ static void tight_encode_tile_basic(struct tight_encoder* self,
 	else
 		memcpy(&cfmt, &self->dfmt, sizeof(cfmt));
 
-	uint32_t* addr = nvnc_fb_get_addr(self->fb);
-	int32_t stride = nvnc_fb_get_stride(self->fb);
-
+	uint8_t* addr = nvnc_fb_get_addr(self->fb);
+	int32_t bpp = self->sfmt.bits_per_pixel / 8;
+	int32_t byte_stride = nvnc_fb_get_stride(self->fb) * bpp;
+	int32_t xoff = x * bpp;
 	// TODO: Limit width and hight to the sides
 	for (uint32_t y = y_start; y < y_start + height; ++y) {
-		void* img = addr + x + y * stride;
-		pixel32_to_cpixel(row, &cfmt, img, &self->sfmt,
+		uint8_t* img = addr + xoff + y * byte_stride;
+		pixel_to_cpixel(row, &cfmt, img, &self->sfmt,
 				bytes_per_cpixel, width);
 
 		// TODO What to do if the buffer fills up?
@@ -335,6 +336,10 @@ static enum TJPF tight_get_jpeg_pixfmt(uint32_t fourcc)
 	case DRM_FORMAT_ABGR8888:
 	case DRM_FORMAT_XBGR8888:
 		return TJPF_RGBX;
+	case DRM_FORMAT_BGR888:
+		return TJPF_RGB;
+	case DRM_FORMAT_RGB888:
+		return TJPF_BGR;
 	}
 
 	return TJPF_UNKNOWN;
@@ -360,14 +365,16 @@ static int tight_encode_tile_jpeg(struct tight_encoder* self,
 	if (!handle)
 		return -1;
 
-	uint32_t* addr = nvnc_fb_get_addr(self->fb);
-	int32_t stride = nvnc_fb_get_stride(self->fb);
-	void* img = (uint32_t*)addr + x + y * stride;
+	uint8_t* addr = nvnc_fb_get_addr(self->fb);
+	int32_t bpp = self->sfmt.bits_per_pixel / 8;
+	int32_t byte_stride = nvnc_fb_get_stride(self->fb) * bpp;
+	int32_t xoff = x * bpp;
+	uint8_t* img = addr + xoff + y * byte_stride;
 
 	enum TJSAMP subsampling = (quality == 9) ? TJSAMP_444 : TJSAMP_420;
 
 	int rc = -1;
-	rc = tjCompress2(handle, img, width, stride * 4, height, tjfmt, &buffer,
+	rc = tjCompress2(handle, img, width, byte_stride, height, tjfmt, &buffer,
 			&size, subsampling, quality, TJFLAG_FASTDCT);
 	if (rc < 0) {
 		nvnc_log(NVNC_LOG_ERROR, "Failed to encode tight JPEG box: %s",

src/ws-framing.c

@@ -1,3 +1,19 @@
+/*
+ * Copyright (c) 2023 Andri Yngvason
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
+ * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+ * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+ * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+ * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
 #include "websocket.h"
 
 #include <stdint.h>

src/ws-handshake.c

@@ -1,8 +1,23 @@
+/*
+ * Copyright (c) 2023 Andri Yngvason
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
+ * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+ * AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
+ * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+ * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+ * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
 #include "websocket.h"
 #include "http.h"
-
-#include <nettle/sha1.h>
-#include <nettle/base64.h>
+#include "crypto.h"
+#include "base64.h"
 
 #include <stdint.h>
 #include <stdio.h>
@@ -61,27 +76,27 @@ ssize_t ws_handshake(char* output, size_t output_maxlen, const char* input)
 	if (have_versions && !strstr(versions, ",13,"))
 		goto failure;
 
-	struct sha1_ctx ctx;
-	sha1_init(&ctx);
-	sha1_update(&ctx, strlen(challenge), (const uint8_t*)challenge);
-	sha1_update(&ctx, strlen(magic_uuid), (const uint8_t*)magic_uuid);
+	uint8_t hash[20];
+	crypto_hash_many(hash, sizeof(hash), CRYPTO_HASH_SHA1,
+			(struct crypto_data_entry[]){
+		{ (uint8_t*)challenge, strlen(challenge) },
+		{ (uint8_t*)magic_uuid, strlen(magic_uuid) },
+		{}
+	});
 
-	uint8_t hash[SHA1_DIGEST_SIZE];
-	sha1_digest(&ctx, sizeof(hash), hash);
-
-	char response[BASE64_ENCODE_RAW_LENGTH(SHA1_DIGEST_SIZE) + 1] = {};
-	base64_encode_raw(response, SHA1_DIGEST_SIZE, hash);
+	char response[BASE64_ENCODED_SIZE(sizeof(hash))] = {};
+	base64_encode(response, hash, sizeof(hash));
 
 	size_t len = snprintf(output, output_maxlen,
-		"HTTP/1.1 101 Switching Protocols\r\n"
-		"Upgrade: websocket\r\n"
-		"Connection: Upgrade\r\n"
-		"Sec-WebSocket-Accept: %s\r\n"
-		"%s%s"
-		"\r\n",
-		response,
-		have_protocols ? "Sec-WebSocket-Protocol: char\r\n" : "",
-		have_versions ? "Sec-WebSocket-Version: 13\r\n" : "");
+			"HTTP/1.1 101 Switching Protocols\r\n"
+			"Upgrade: websocket\r\n"
+			"Connection: Upgrade\r\n"
+			"Sec-WebSocket-Accept: %s\r\n"
+			"%s%s"
+			"\r\n",
+			response,
+			have_protocols ? "Sec-WebSocket-Protocol: char\r\n" : "",
+			have_versions ? "Sec-WebSocket-Version: 13\r\n" : "");
 
 	ssize_t header_len = req.header_length;
 	ok = len < output_maxlen;

src/zrle.c

@@ -61,32 +61,32 @@ static inline struct zrle_encoder* zrle_encoder(struct encoder* encoder)
 	return (struct zrle_encoder*)encoder;
 }
 
-static inline int find_colour_in_palette(uint32_t* palette, int len,
-                                         uint32_t colour)
+static inline int find_colour_in_palette(uint8_t* palette, int len,
+		const uint8_t* colour, int bpp)
 {
 	for (int i = 0; i < len; ++i)
-		if (palette[i] == colour)
+		if (memcmp(palette + i * bpp, colour, bpp) == 0)
 			return i;
 
 	return -1;
 }
 
-static int zrle_get_tile_palette(uint32_t* palette, const uint32_t* src,
-                                 size_t length)
+static int zrle_get_tile_palette(uint8_t* palette, const uint8_t* src,
+		const int src_bpp, size_t length)
 {
 	int n = 0;
 
 	/* TODO: Maybe ignore the alpha channel */
-	palette[n++] = src[0];
+	memcpy(palette + (n++ * src_bpp), src, src_bpp);
 
 	for (size_t i = 0; i < length; ++i) {
-		uint32_t colour = src[i];
+		const uint8_t* colour_addr = src + i * src_bpp;
 
-		if (find_colour_in_palette(palette, n, colour) < 0) {
+		if (find_colour_in_palette(palette, n, colour_addr, src_bpp) < 0) {
 			if (n >= 16)
 				return -1;
 
-			palette[n++] = colour;
+			memcpy(palette + (n++ * src_bpp), colour_addr, src_bpp);
 		}
 	}
 
@@ -94,16 +94,16 @@ static int zrle_get_tile_palette(uint32_t* palette, const uint32_t* src,
 }
 
 static void zrle_encode_unichrome_tile(struct vec* dst,
-                                       const struct rfb_pixel_format* dst_fmt,
-                                       uint32_t colour,
-                                       const struct rfb_pixel_format* src_fmt)
+		const struct rfb_pixel_format* dst_fmt,
+		uint8_t* colour,
+		const struct rfb_pixel_format* src_fmt)
 {
 	int bytes_per_cpixel = calc_bytes_per_cpixel(dst_fmt);
 
 	vec_fast_append_8(dst, 1);
 
-	pixel32_to_cpixel(((uint8_t*)dst->data) + 1, dst_fmt, &colour, src_fmt,
-	                  bytes_per_cpixel, 1);
+	pixel_to_cpixel(((uint8_t*)dst->data) + 1, dst_fmt, colour, src_fmt,
+			bytes_per_cpixel, 1);
 
 	dst->len += bytes_per_cpixel;
 }
@@ -126,17 +126,18 @@ static void encode_run_length(struct vec* dst, uint8_t index, int run_length)
 }
 
 static void zrle_encode_packed_tile(struct vec* dst,
-                                    const struct rfb_pixel_format* dst_fmt,
-                                    const uint32_t* src,
-                                    const struct rfb_pixel_format* src_fmt,
-                                    size_t length, uint32_t* palette,
-                                    int palette_size)
+		const struct rfb_pixel_format* dst_fmt,
+		const uint8_t* src,
+		const struct rfb_pixel_format* src_fmt,
+		size_t length, uint8_t* palette,
+		int palette_size)
 {
 	int bytes_per_cpixel = calc_bytes_per_cpixel(dst_fmt);
+	int src_bpp = src_fmt->bits_per_pixel / 8;
 
 	uint8_t cpalette[16 * 3];
-	pixel32_to_cpixel((uint8_t*)cpalette, dst_fmt, palette, src_fmt,
-	                  bytes_per_cpixel, palette_size);
+	pixel_to_cpixel(cpalette, dst_fmt, palette, src_fmt,
+			bytes_per_cpixel, palette_size);
 
 	vec_fast_append_8(dst, 128 | palette_size);
 
@@ -146,64 +147,65 @@ static void zrle_encode_packed_tile(struct vec* dst,
 	int run_length = 1;
 
 	for (size_t i = 1; i < length; ++i) {
-		if (src[i] == src[i - 1]) {
+		if (memcmp(src + i * src_bpp, src + (i - 1) * src_bpp, src_bpp) == 0) {
 			run_length++;
 			continue;
 		}
 
-		index = find_colour_in_palette(palette, palette_size, src[i - 1]);
+		index = find_colour_in_palette(palette, palette_size, src + (i - 1) * src_bpp, src_bpp);
 		encode_run_length(dst, index, run_length);
 		run_length = 1;
 	}
 
 	if (run_length > 0) {
 		index = find_colour_in_palette(palette, palette_size,
-		                               src[length - 1]);
+				src + (length - 1) * src_bpp, src_bpp);
 		encode_run_length(dst, index, run_length);
 	}
 }
 
-static void zrle_copy_tile(uint32_t* dst, const uint32_t* src, int stride,
-                           int width, int height)
+static void zrle_copy_tile(uint8_t* tile, const uint8_t* src, int src_bpp,
+		int stride, int width, int height)
 {
+	int byte_stride = stride * src_bpp;
 	for (int y = 0; y < height; ++y)
-		memcpy(dst + y * width, src + y * stride, width * 4);
+		memcpy(tile + y * width * src_bpp, src + y * byte_stride, width * src_bpp);
 }
 
 static void zrle_encode_tile(struct vec* dst,
-                             const struct rfb_pixel_format* dst_fmt,
-                             const uint32_t* src,
-                             const struct rfb_pixel_format* src_fmt,
-                             size_t length)
+		const struct rfb_pixel_format* dst_fmt,
+		const uint8_t* src,
+		const struct rfb_pixel_format* src_fmt,
+		size_t length)
 {
 	int bytes_per_cpixel = calc_bytes_per_cpixel(dst_fmt);
-
+	int src_bpp = src_fmt->bits_per_pixel / 8;
 	vec_clear(dst);
 
-	uint32_t palette[16];
-	int palette_size = zrle_get_tile_palette(palette, src, length);
+	uint8_t palette[16 * 4];
+	int palette_size = zrle_get_tile_palette(palette, src, src_bpp, length);
 
 	if (palette_size == 1) {
-		zrle_encode_unichrome_tile(dst, dst_fmt, palette[0], src_fmt);
+		zrle_encode_unichrome_tile(dst, dst_fmt, &palette[0], src_fmt);
 		return;
 	}
 
 	if (palette_size > 1) {
 		zrle_encode_packed_tile(dst, dst_fmt, src, src_fmt, length,
-		                        palette, palette_size);
+				palette, palette_size);
 		return;
 	}
 
 	vec_fast_append_8(dst, 0);
 
-	pixel32_to_cpixel(((uint8_t*)dst->data) + 1, dst_fmt, src, src_fmt,
-	                  bytes_per_cpixel, length);
+	pixel_to_cpixel(((uint8_t*)dst->data) + 1, dst_fmt, (uint8_t*)src, src_fmt,
+			bytes_per_cpixel, length);
 
 	dst->len += bytes_per_cpixel * length;
 }
 
 static int zrle_deflate(struct vec* dst, const struct vec* src, z_stream* zs,
-                        bool flush)
+		bool flush)
 {
 	zs->next_in = src->data;
 	zs->avail_in = src->len;
@@ -228,19 +230,20 @@ static int zrle_deflate(struct vec* dst, const struct vec* src, z_stream* zs,
 }
 
 static int zrle_encode_box(struct zrle_encoder* self, struct vec* out,
-                           const struct rfb_pixel_format* dst_fmt,
-                           const struct nvnc_fb* fb,
-                           const struct rfb_pixel_format* src_fmt, int x, int y,
-                           int stride, int width, int height, z_stream* zs)
+		const struct rfb_pixel_format* dst_fmt,
+		const struct nvnc_fb* fb,
+		const struct rfb_pixel_format* src_fmt, int x, int y,
+		int stride, int width, int height, z_stream* zs)
 {
 	int r = -1;
 	int bytes_per_cpixel = calc_bytes_per_cpixel(dst_fmt);
+	int src_bpp = src_fmt->bits_per_pixel / 8;
 	struct vec in;
 
 	uint16_t x_pos = self->encoder.x_pos;
 	uint16_t y_pos = self->encoder.y_pos;
 
-	uint32_t* tile = malloc(TILE_LENGTH * TILE_LENGTH * 4);
+	uint8_t* tile = malloc(TILE_LENGTH * TILE_LENGTH * 4);
 	if (!tile)
 		goto failure;
 
@@ -268,14 +271,15 @@ static int zrle_encode_box(struct zrle_encoder* self, struct vec* out,
 		                          ? TILE_LENGTH
 		                          : height - tile_y;
 
-		int y_off = y + tile_y;
+		int y_off = (y + tile_y) * stride * src_bpp;
+		int x_off = (x + tile_x) * src_bpp;
 
 		zrle_copy_tile(tile,
-		               ((uint32_t*)fb->addr) + x + tile_x + y_off * stride,
-		               stride, tile_width, tile_height);
+				((uint8_t*)fb->addr) + x_off + y_off, src_bpp,
+				stride, tile_width, tile_height);
 
 		zrle_encode_tile(&in, dst_fmt, tile, src_fmt,
-		                 tile_width * tile_height);
+				tile_width * tile_height);
 
 		r = zrle_deflate(out, &in, zs, i == n_tiles - 1);
 		if (r < 0)
@@ -319,7 +323,7 @@ static int zrle_encode_frame(struct zrle_encoder* self, z_stream* zs,
 		int box_height = box[i].y2 - y;
 
 		rc = zrle_encode_box(self, dst, dst_fmt, src, src_fmt, x, y,
-		                     src->stride, box_width, box_height, zs);
+				src->stride, box_width, box_height, zs);
 		if (rc < 0)
 			return -1;
 	}

test/meson.build

@@ -7,6 +7,16 @@ pixels = executable('pixels',
 	dependencies: [
 		pixman,
 		libdrm_inc,
+		libm,
 	],
 )
 test('pixels', pixels)
+
+base64 = executable('base64',
+	[
+		'test-base64.c',
+		'../src/base64.c',
+	],
+	include_directories: inc,
+)
+test('base64', base64)

test/test-base64.c

@@ -0,0 +1,129 @@
+#include "base64.h"
+
+#include <stdio.h>
+#include <stdbool.h>
+#include <string.h>
+
+static bool test_encode_0(void)
+{
+	char buf[1] = {};
+	base64_encode(buf, (const uint8_t*)"", 0);
+	return strlen(buf) == 0;
+}
+
+static bool test_encode_1(void)
+{
+	static const char input[] = "a";
+	char buf[BASE64_ENCODED_SIZE(sizeof(input) - 1)] = {};
+	base64_encode(buf, (const uint8_t*)input, sizeof(input) - 1);
+	return strcmp(buf, "YQ==") == 0;
+}
+
+static bool test_encode_2(void)
+{
+	static const char input[] = "ab";
+	char buf[BASE64_ENCODED_SIZE(sizeof(input) - 1)] = {};
+	base64_encode(buf, (const uint8_t*)input, sizeof(input) - 1);
+	return strcmp(buf, "YWI=") == 0;
+}
+
+static bool test_encode_3(void)
+{
+	static const char input[] = "abc";
+	char buf[BASE64_ENCODED_SIZE(sizeof(input) - 1)] = {};
+	base64_encode(buf, (const uint8_t*)input, sizeof(input) - 1);
+	return strcmp(buf, "YWJj") == 0;
+}
+
+static bool test_encode_4(void)
+{
+	static const char input[] = "abcd";
+	char buf[BASE64_ENCODED_SIZE(sizeof(input) - 1)] = {};
+	base64_encode(buf, (const uint8_t*)input, sizeof(input) - 1);
+	return strcmp(buf, "YWJjZA==") == 0;
+}
+
+static bool test_encode_5(void)
+{
+	static const char input[] = "abcde";
+	char buf[BASE64_ENCODED_SIZE(sizeof(input) - 1)] = {};
+	base64_encode(buf, (const uint8_t*)input, sizeof(input) - 1);
+	return strcmp(buf, "YWJjZGU=") == 0;
+}
+
+static bool test_decode_0(void)
+{
+	uint8_t buf[1] = {};
+	ssize_t r = base64_decode(buf, "");
+	return r == 0 && buf[0] == 0;
+}
+
+static bool test_decode_1(void)
+{
+	static const char input[] = "YQ==";
+	uint8_t buf[BASE64_DECODED_MAX_SIZE(sizeof(input) - 1)] = {};
+	ssize_t r = base64_decode(buf, input);
+	return r == 1 && memcmp(buf, "a", r) == 0;
+}
+
+static bool test_decode_2(void)
+{
+	static const char input[] = "YWI=";
+	uint8_t buf[BASE64_DECODED_MAX_SIZE(sizeof(input) - 1)] = {};
+	ssize_t r = base64_decode(buf, input);
+	return r == 2 && memcmp(buf, "ab", r) == 0;
+}
+
+static bool test_decode_3(void)
+{
+	static const char input[] = "YWJj";
+	uint8_t buf[BASE64_DECODED_MAX_SIZE(sizeof(input) - 1)] = {};
+	ssize_t r = base64_decode(buf, input);
+	return r == 3 && memcmp(buf, "abc", r) == 0;
+}
+
+static bool test_decode_4(void)
+{
+	static const char input[] = "YWJjZA==";
+	uint8_t buf[BASE64_DECODED_MAX_SIZE(sizeof(input) - 1)] = {};
+	ssize_t r = base64_decode(buf, input);
+	return r == 4 && memcmp(buf, "abcd", r) == 0;
+}
+
+static bool test_decode_5(void)
+{
+	static const char input[] = "YWJjZGU=";
+	uint8_t buf[BASE64_DECODED_MAX_SIZE(sizeof(input) - 1)] = {};
+	ssize_t r = base64_decode(buf, input);
+	return r == 5 && memcmp(buf, "abcde", r) == 0;
+}
+
+#define XSTR(s) STR(s)
+#define STR(s) #s
+
+#define RUN_TEST(name) ({ \
+	bool ok = test_ ## name(); \
+	printf("[%s] %s\n", ok ? " OK " : "FAIL", XSTR(name)); \
+	ok; \
+})
+
+int main()
+{
+	bool ok = true;
+
+	ok &= RUN_TEST(encode_0);
+	ok &= RUN_TEST(encode_1);
+	ok &= RUN_TEST(encode_2);
+	ok &= RUN_TEST(encode_3);
+	ok &= RUN_TEST(encode_4);
+	ok &= RUN_TEST(encode_5);
+
+	ok &= RUN_TEST(decode_0);
+	ok &= RUN_TEST(decode_1);
+	ok &= RUN_TEST(decode_2);
+	ok &= RUN_TEST(decode_3);
+	ok &= RUN_TEST(decode_4);
+	ok &= RUN_TEST(decode_5);
+
+	return ok ? 0 : 1;
+}

test/test-pixels.c

@@ -22,10 +22,11 @@
 #define UDIV_UP(a, b) (((a) + (b) - 1) / (b))
 #define ARRAY_LEN(a) (sizeof(a) / (sizeof(a[0])))
 
-static bool test_pixel32_to_cpixel_4bpp(void)
+static bool test_pixel_to_cpixel_4bpp(void)
 {
 	uint32_t src = u32_le(0x11223344u);
 	uint32_t dst;
+	uint8_t* src_addr = (uint8_t*)&src;
 
 	struct rfb_pixel_format dstfmt = { 0 }, srcfmt = { 0 };
 
@@ -33,25 +34,63 @@ static bool test_pixel32_to_cpixel_4bpp(void)
 
 	dst = 0;
 	rfb_pixfmt_from_fourcc(&srcfmt, DRM_FORMAT_RGBA8888);
-	pixel32_to_cpixel((uint8_t*)&dst, &dstfmt, &src, &srcfmt, 4, 1);
+	pixel_to_cpixel((uint8_t*)&dst, &dstfmt, src_addr, &srcfmt, 4, 1);
 	if ((src & 0xffffff00u) != (dst & 0xffffff00u))
 		return false;
 
 	dst = 0;
 	rfb_pixfmt_from_fourcc(&dstfmt, DRM_FORMAT_ABGR8888);
-	pixel32_to_cpixel((uint8_t*)&dst, &dstfmt, &src, &srcfmt, 4, 1);
+	pixel_to_cpixel((uint8_t*)&dst, &dstfmt, src_addr, &srcfmt, 4, 1);
 	if (dst != u32_le(0x00332211u))
 		return false;
 
 	dst = 0;
 	rfb_pixfmt_from_fourcc(&dstfmt, DRM_FORMAT_ARGB8888);
-	pixel32_to_cpixel((uint8_t*)&dst, &dstfmt, &src, &srcfmt, 4, 1);
+	pixel_to_cpixel((uint8_t*)&dst, &dstfmt, src_addr, &srcfmt, 4, 1);
 	if (dst != u32_le(0x00112233u))
 		return false;
 
 	dst = 0;
 	rfb_pixfmt_from_fourcc(&dstfmt, DRM_FORMAT_BGRA8888);
-	pixel32_to_cpixel((uint8_t*)&dst, &dstfmt, &src, &srcfmt, 4, 1);
+	pixel_to_cpixel((uint8_t*)&dst, &dstfmt, src_addr, &srcfmt, 4, 1);
+	if (dst != u32_le(0x33221100u))
+		return false;
+
+	return true;
+}
+
+static bool test_pixel_to_cpixel_3bpp(void)
+{
+	//44 is extra data that should not be copied anywhere below.
+	uint32_t src = u32_le(0x44112233u);
+	uint32_t dst;
+	uint8_t* src_addr = (uint8_t*)&src;
+
+	struct rfb_pixel_format dstfmt = { 0 }, srcfmt = { 0 };
+
+	rfb_pixfmt_from_fourcc(&srcfmt, DRM_FORMAT_RGB888);
+
+	dst = 0;
+	rfb_pixfmt_from_fourcc(&dstfmt, DRM_FORMAT_RGBA8888);
+	pixel_to_cpixel((uint8_t*)&dst, &dstfmt, src_addr, &srcfmt, 4, 1);
+	if (dst != u32_le(0x11223300u))
+		return false;
+
+	dst = 0;
+	rfb_pixfmt_from_fourcc(&dstfmt, DRM_FORMAT_ABGR8888);
+	pixel_to_cpixel((uint8_t*)&dst, &dstfmt, src_addr, &srcfmt, 4, 1);
+	if (dst != u32_le(0x00332211u))
+		return false;
+
+	dst = 0;
+	rfb_pixfmt_from_fourcc(&dstfmt, DRM_FORMAT_ARGB8888);
+	pixel_to_cpixel((uint8_t*)&dst, &dstfmt, src_addr, &srcfmt, 4, 1);
+	if (dst != u32_le(0x00112233u))
+		return false;
+
+	dst = 0;
+	rfb_pixfmt_from_fourcc(&dstfmt, DRM_FORMAT_BGRA8888);
+	pixel_to_cpixel((uint8_t*)&dst, &dstfmt, src_addr, &srcfmt, 4, 1);
 	if (dst != u32_le(0x33221100u))
 		return false;
 
@@ -173,7 +212,8 @@ static bool test_rfb_pixfmt_to_string(void)
 
 int main()
 {
-	bool ok = test_pixel32_to_cpixel_4bpp() &&
+	bool ok = test_pixel_to_cpixel_4bpp() &&
+		test_pixel_to_cpixel_3bpp() &&
 		test_fourcc_to_pixman_fmt() &&
 		test_extract_alpha_mask_rgba8888() &&
 		test_drm_format_to_string() &&