Add writeups/ipv6/eyeball

3 files changed, 616 insertions, 0 deletions

diff --git a/writeups/ipv6/eyeball/.gitignore b/writeups/ipv6/eyeball/.gitignore
new file mode 100644
index 0000000..08882c0
--- /dev/null
+++ b/writeups/ipv6/eyeball/.gitignore
@@ -0,0 +1 @@
+/eyeball
diff --git a/writeups/ipv6/eyeball/Makefile b/writeups/ipv6/eyeball/Makefile
new file mode 100644
index 0000000..939a53d
--- /dev/null
+++ b/writeups/ipv6/eyeball/Makefile
@@ -0,0 +1,3 @@
+
+eyeball: eyeball.c
+	cc -g -O0 -Wall -Wextra eyeball.c -o eyeball
diff --git a/writeups/ipv6/eyeball/eyeball.c b/writeups/ipv6/eyeball/eyeball.c
new file mode 100644
index 0000000..32a6033
--- /dev/null
+++ b/writeups/ipv6/eyeball/eyeball.c
@@ -0,0 +1,612 @@
+/**
+ * @file eyeball.c
+ * @author David Timber
+ * @brief Demonstrates RFC 8305
+ * @see
+ * https://datatracker.ietf.org/doc/html/rfc6555
+ * https://datatracker.ietf.org/doc/html/rfc8305
+ */
+#define _POSIX_C_SOURCE 200809L
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdbool.h>
+#include <inttypes.h>
+#include <assert.h>
+#include <string.h>
+#include <errno.h>
+#include <signal.h>
+#include <math.h>
+
+#include <unistd.h>
+#include <getopt.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <arpa/inet.h>
+#include <netdb.h>
+#include <pthread.h>
+
+#define ARGV0 "eyeball"
+#define DEFAULT_RES_DELAY 0.05
+
+
+void ts_sub (
+	struct timespec *c,
+	const struct timespec *a,
+	const struct timespec *b)
+{
+	c->tv_nsec = a->tv_nsec - b->tv_nsec;
+	c->tv_sec = a->tv_sec - b->tv_sec;
+	if (c->tv_nsec < 0) {
+		c->tv_sec -= 1;
+		c->tv_nsec += 1000000000;
+	}
+}
+
+enum happy_result {
+	// no error: connection established and handshake complete
+	HR_NULL,
+	// a function call failed and error set to errno
+	HR_ERRNO,
+	// getaddrinfo() returned error and error is set to the returned value
+	HR_GAI,
+	// error set to user-defined value
+	HR_OTHER,
+};
+
+struct happy_error {
+	struct {
+		struct timespec resolv;
+		struct timespec conn;
+		struct timespec total;
+	} delay;
+	enum happy_result result;
+	int error;
+	bool ready;
+};
+
+typedef void*(*happy_connectf_t)(
+	void *ctx,
+	const size_t aic,
+	const struct addrinfo *aiv,
+	struct happy_error *out_err);
+
+/*
+ * simple reference connectf implementation that creates a TCP socket and
+ * connects to the address
+ */
+void *happy_tcp_connectf (
+		void *ctx,
+		const size_t, // unused
+		const struct addrinfo *aiv,
+		struct happy_error *out_err)
+{
+	int ret = -1;
+	int fr;
+
+	ret = socket(aiv->ai_family, aiv->ai_socktype, aiv->ai_protocol);
+	if (ret < 0) {
+		goto ERR;
+	}
+
+	fr = connect(ret, aiv->ai_addr, aiv->ai_addrlen);
+	if (fr != 0) {
+		goto ERR;
+	}
+	goto END;
+
+ERR:
+	out_err->result = HR_ERRNO;
+	out_err->error = errno;
+	if (ret >= 0) {
+		close(ret);
+		ret = -1;
+	}
+
+END:
+	*((int*)ctx) = ret;
+	return NULL;
+}
+
+/*
+ * Converts the list returned from getaddrinfo() to a flat array
+ */
+struct addrinfo *flataddrinfo (const struct addrinfo *list, size_t *cnt) {
+	const struct addrinfo *p;
+	struct addrinfo *ret = NULL;
+	size_t i;
+
+	for (i = 0, p = list; p != NULL; p = p->ai_next, i += 1);
+	if (i == 0) {
+		goto END;
+	}
+
+	ret = malloc(sizeof(struct addrinfo) * i);
+	if (ret == NULL) {
+		goto END;
+	}
+
+	for (i = 0, p = list; p != NULL; p = p->ai_next, i += 1) {
+		ret[i] = *p;
+		ret[i].ai_next = ret + i + 1;
+	}
+	ret[i - 1].ai_next = NULL;
+
+END:
+	if (cnt != NULL) {
+		*cnt = i;
+	}
+	return ret;
+}
+
+void *happy_th_main_inner (
+		const char *node,
+		const char *service,
+		const struct addrinfo *hints,
+		void *ctx,
+		happy_connectf_t connf,
+		pthread_mutex_t *lock,
+		pthread_cond_t *cond,
+		struct happy_error *out_err)
+{
+	int fr;
+	void *ret = NULL;
+	struct happy_error err = { 0, };
+	struct addrinfo *gai_res = NULL;
+	struct addrinfo *aiv = NULL;
+	size_t aic = 0;
+	struct {
+		struct timespec start;
+		struct timespec resolv;
+		struct timespec conn;
+		struct timespec end;
+	} ts;
+
+	clock_gettime(CLOCK_MONOTONIC, &ts.start);
+
+	// do resolve
+	fr = getaddrinfo(node, service, hints, &gai_res);
+	clock_gettime(CLOCK_MONOTONIC, &ts.resolv);
+	if (fr != 0) {
+		if (fr == EAI_SYSTEM) {
+			err.result = HR_ERRNO;
+			err.error = errno;
+		}
+		else {
+			err.result = HR_GAI;
+			err.error = fr;
+		}
+
+		goto END;
+	}
+	assert(gai_res != NULL);
+
+	// bring the result to our address space by converting the list to array
+	aiv = flataddrinfo(gai_res, &aic);
+	if (aiv == NULL) {
+		goto END;
+	}
+	// free the list
+	freeaddrinfo(gai_res);
+	gai_res = NULL;
+
+	clock_gettime(CLOCK_MONOTONIC, &ts.conn);
+	if (connf != NULL) {
+		// call the inner function
+		ret = connf(ctx, aic, aiv, &err);
+	}
+
+END:
+	clock_gettime(CLOCK_MONOTONIC, &ts.end);
+	ts_sub(&err.delay.resolv, &ts.resolv, &ts.start);
+	ts_sub(&err.delay.conn, &ts.end, &ts.conn);
+	ts_sub(&err.delay.total, &ts.end, &ts.start);
+
+	freeaddrinfo(gai_res);
+	free(aiv);
+
+	err.ready = true;
+
+	fr = pthread_mutex_lock(lock);
+	assert(fr == 0);
+	if (out_err != NULL) {
+		*out_err = err;
+	}
+	pthread_cond_broadcast(cond);
+	pthread_mutex_unlock(lock);
+
+	return ret;
+}
+
+struct happy_th_args {
+	const char *node;
+	const char *service;
+	const struct addrinfo *hints;
+	void *ctx;
+	happy_connectf_t connf;
+	pthread_mutex_t *lock;
+	pthread_cond_t *cond;
+	struct happy_error *out_err;
+};
+
+/*
+ * Happy eyeballs worker thread entry
+ */
+void *happy_th_main (void *args_in) {
+	struct happy_th_args *args = args_in;
+
+	return happy_th_main_inner(
+		args->node,
+		args->service,
+		args->hints,
+		args->ctx,
+		args->connf,
+		args->lock,
+		args->cond,
+		args->out_err
+	);
+}
+
+int aihintflags (void) {
+	int ret = 0;
+#ifdef AI_IDN
+	ret |= AI_IDN;
+#endif
+	return ret;
+}
+
+struct {
+	const char *node;
+	const char *service;
+	struct timespec op_timeout;
+	double res_delay;
+	struct {
+		bool help;
+		bool both;
+	} flags;
+} opts;
+
+void init_opts (void) {
+	// connection timeout: 10 s
+	opts.op_timeout.tv_sec = 10;
+	opts.op_timeout.tv_nsec = 0;
+	// "Resolution Delay" as per RFC 8305 section 3: 50 ms
+	opts.res_delay = DEFAULT_RES_DELAY;
+}
+
+bool parse_args (const int argc, const char **argv) {
+	int fr;
+	double tmpf;
+
+	while (true) {
+		fr = getopt(argc, (char *const*)argv, "hd:b");
+		if (fr < 0) {
+			break;
+		}
+
+		switch (fr) {
+		case 'h': opts.flags.help = true; break;
+		case 'd':
+			tmpf = NAN;
+			fr = sscanf(optarg, "%lf", &tmpf);
+			if (fr != 1 || isnan(tmpf)) {
+				fprintf(stderr, ARGV0": invalid option -d: %s\n", optarg);
+				return false;
+			}
+			opts.res_delay = tmpf;
+			break;
+		case 'b':
+			opts.flags.both = true;
+			break;
+		case '?': return false;
+		}
+	}
+
+	if (argc < optind + 2) {
+		fprintf(stderr, ARGV0": too few arguments\n");
+		return false;
+	}
+
+	opts.node = argv[optind];
+	opts.service = argv[optind + 1];
+
+	return true;
+}
+
+void print_help (void) {
+	printf("Usage: "ARGV0" [-hb] [-d RES_DELAY] <NODE> <SERVICE>\n");
+}
+
+struct {
+	pthread_mutex_t lock;
+	pthread_cond_t cond;
+	struct {
+		struct addrinfo hints;
+		struct happy_th_args args;
+		struct happy_error err;
+		pthread_t th;
+		int ret;
+		bool started;
+	} th[2];
+} g;
+
+bool init_ctx (void) {
+	pthread_mutex_init(&g.lock, NULL);
+	pthread_cond_init(&g.cond, NULL);
+
+	// hints
+	g.th[0].hints.ai_socktype = g.th[1].hints.ai_socktype = SOCK_STREAM;
+	g.th[0].hints.ai_flags = g.th[1].hints.ai_flags = aihintflags();
+	g.th[0].hints.ai_family = AF_INET6;
+	g.th[1].hints.ai_family = AF_INET;
+
+	// args
+	g.th[0].args.node = g.th[1].args.node = opts.node;
+	g.th[0].args.service = g.th[1].args.service = opts.service;
+	g.th[0].args.connf = g.th[1].args.connf = happy_tcp_connectf;
+	g.th[0].args.hints = &g.th[0].hints;
+	g.th[0].args.ctx = &g.th[0].ret;
+	g.th[0].args.lock = &g.lock;
+	g.th[0].args.cond = &g.cond;
+	g.th[0].args.out_err = &g.th[0].err;
+	g.th[1].args.hints = &g.th[1].hints;
+	g.th[1].args.ctx = &g.th[1].ret;
+	g.th[1].args.lock = &g.lock;
+	g.th[1].args.cond = &g.cond;
+	g.th[1].args.out_err = &g.th[1].err;
+
+	g.th[0].ret = g.th[1].ret = -1;
+
+	return true;
+}
+
+void deinit_ctx (void) {
+	for (size_t i = 0; i < 2; i += 1) {
+		if (!g.th[i].started) {
+			continue;
+		}
+/*
+ * ain't pretty, but canceling getaddrinfo() is not possible. Portability comes
+ * first in this demo. This is why cross-platform many apps handcraft their own
+ * address or use libraries like c-res.
+ */
+		pthread_cancel(g.th[i].th);
+		pthread_join(g.th[i].th, NULL);
+	}
+
+	pthread_mutex_destroy(&g.lock);
+	pthread_cond_destroy(&g.cond);
+}
+
+bool spawn_threads (void) {
+	int fr;
+
+	for (size_t i = 0; i < 2; i += 1) {
+		fr = pthread_create(&g.th[i].th, NULL, happy_th_main, &g.th[i].args);
+		g.th[i].started = fr == 0;
+		if (!g.th[i].started) {
+			perror(ARGV0": pthread_create()");
+			return false;
+		}
+	}
+
+	return true;
+}
+
+void get_leadtime (struct timespec *ts, const struct timespec *amt) {
+	clock_gettime(CLOCK_REALTIME, ts);
+	ts->tv_sec += amt->tv_sec;
+	ts->tv_nsec += amt->tv_nsec;
+	ts->tv_sec += ts->tv_nsec / 1000000000;
+	ts->tv_nsec %= 1000000000;
+}
+
+void get_leadtimef (struct timespec *ts, const double lt) {
+	struct timespec translated;
+
+	translated.tv_sec = (time_t)lt;
+	translated.tv_nsec = (long)((lt - (time_t)lt) * 1000000000);
+	get_leadtime(ts, &translated);
+}
+
+ssize_t poll_result (void) {
+	struct timespec op_deadline;
+	int fr;
+
+	get_leadtime(&op_deadline, &opts.op_timeout);
+
+	while (true) {
+		if (g.th[0].err.ready && g.th[0].err.result == HR_NULL) {
+			// ipv6 made it already
+			return 1;
+		}
+
+		if (g.th[1].err.ready && g.th[1].err.result == HR_NULL) {
+			// ipv4 connected, but let's wait around for a bit
+			struct timespec bias_deadline;
+
+			fprintf(stderr, ARGV0": ipv4 shot first\n");
+
+			get_leadtimef(&bias_deadline, opts.res_delay);
+			pthread_cond_timedwait(&g.cond, &g.lock, &bias_deadline);
+
+			if (g.th[0].err.ready && g.th[0].err.result == HR_NULL) {
+				// ipv6 made it!
+				return 1;
+			}
+			else if (g.th[1].err.ready) {
+				// ipv6 didn't make it
+				return 2;
+			}
+			else {
+				// all timed out
+				return -1;
+			}
+		}
+
+		if (g.th[0].err.ready && g.th[1].err.ready) {
+			// both failed
+			return -1;
+		}
+
+		// both still trying
+		fr = pthread_cond_timedwait(&g.cond, &g.lock, &op_deadline);
+		if (fr != 0) {
+			// timed out or interrupted
+			break;
+		}
+	}
+
+	return -1;
+}
+
+ssize_t poll_both (void) {
+	ssize_t ret = 0;
+	struct timespec op_deadline;
+	int fr;
+
+	get_leadtime(&op_deadline, &opts.op_timeout);
+
+	while (!(g.th[0].err.ready && g.th[1].err.ready)) {
+		fr = pthread_cond_timedwait(&g.cond, &g.lock, &op_deadline);
+		if (fr != 0) {
+			break;
+		}
+	}
+
+	if (g.th[0].err.result == HR_NULL) {
+		ret |= 1;
+	}
+	if (g.th[1].err.result == HR_NULL) {
+		ret |= 2;
+	}
+	if (ret == 0) {
+		ret = -1;
+	}
+
+	return ret;
+}
+
+void print_sockname (
+		const int fd,
+		int(*namef)(int, struct sockaddr*, socklen_t*))
+{
+	union {
+		struct sockaddr_storage ss;
+		struct sockaddr sa;
+		struct sockaddr_in sin;
+		struct sockaddr_in6 sin6;
+	} addr;
+	socklen_t sl = sizeof(addr);
+	char str[INET6_ADDRSTRLEN] = { 0 };
+	int fr;
+	void *addr_data;
+	uint16_t port;
+
+	fr = namef(fd, &addr.sa, &sl);
+	assert(fr == 0);
+
+	if (addr.sa.sa_family == AF_INET) {
+		addr_data = &addr.sin.sin_addr;
+		port = ntohs(addr.sin.sin_port);
+	}
+	else {
+		addr_data = &addr.sin6.sin6_addr;
+		port = ntohs(addr.sin6.sin6_port);
+	}
+	inet_ntop(addr.sa.sa_family, addr_data, str, sizeof(str));
+
+	if (addr.sa.sa_family == AF_INET) {
+		printf("%s:%"PRIu16, str, port);
+	}
+	else {
+		printf("[%s]:%"PRIu16, str, port);
+	}
+}
+
+void do_report (const ssize_t picked) {
+	char star;
+
+	for (size_t i = 0; i < 2; i += 1) {
+		if (picked > 0 && ((1 << i) & picked)) {
+			star =  '*';
+		}
+		else {
+			star = ' ';
+		}
+		printf("%c%s: ", star, g.th[i].hints.ai_family == AF_INET ? "v4" : "v6");
+
+		if (g.th[i].err.ready) {
+			printf("%ld.%03lus %ld.%03lus %ld.%03lus ",
+				(long)g.th[i].err.delay.resolv.tv_sec,
+				(unsigned long)g.th[i].err.delay.resolv.tv_nsec / 1000000,
+				(long)g.th[i].err.delay.conn.tv_sec,
+				(unsigned long)g.th[i].err.delay.conn.tv_nsec / 1000000,
+				(long)g.th[i].err.delay.total.tv_sec,
+				(unsigned long)g.th[i].err.delay.total.tv_nsec / 1000000);
+
+
+			switch (g.th[i].err.result) {
+			case HR_GAI:
+				printf("%s", gai_strerror(g.th[i].err.error));
+				break;
+			case HR_ERRNO:
+				printf("%s", strerror(g.th[i].err.error));
+				break;
+			default:
+				print_sockname(g.th[i].ret, getsockname);
+				printf(" -> ");
+				print_sockname(g.th[i].ret, getpeername);
+			}
+		}
+		printf("\n");
+	}
+}
+
+int do_pick (void) {
+	ssize_t picked;
+
+	pthread_mutex_lock(&g.lock);
+	if (opts.flags.both) {
+		picked = poll_both();
+	}
+	else {
+		picked = poll_result();
+	}
+	do_report(picked);
+	pthread_mutex_unlock(&g.lock);
+
+	if (picked < 0) {
+		return 1;
+	}
+	return 0;
+}
+
+int main (const int argc, const char **argv) {
+	static int ec = 0;
+
+	init_opts();
+
+	if (!parse_args(argc, argv)) {
+		ec = 2;
+		goto END;
+	}
+	if (opts.flags.help) {
+		print_help();
+		goto END;
+	}
+
+	if (!init_ctx()) {
+		ec = 1;
+		goto END;
+	}
+	if (!spawn_threads()) {
+		ec = 1;
+		goto END;
+	}
+
+	ec = do_pick();
+
+END:
+	deinit_ctx();
+	return ec;
+}