summaryrefslogtreecommitdiffstats
path: root/bin/named/fuzz.c
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-05 18:37:14 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-05-05 18:37:14 +0000
commitea648e70a989cca190cd7403fe892fd2dcc290b4 (patch)
treee2b6b1c647da68b0d4d66082835e256eb30970e8 /bin/named/fuzz.c
parentInitial commit. (diff)
downloadbind9-upstream/1%9.11.5.P4+dfsg.tar.xz
bind9-upstream/1%9.11.5.P4+dfsg.zip
Adding upstream version 1:9.11.5.P4+dfsg.upstream/1%9.11.5.P4+dfsgupstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to '')
-rw-r--r--bin/named/fuzz.c483
1 files changed, 483 insertions, 0 deletions
diff --git a/bin/named/fuzz.c b/bin/named/fuzz.c
new file mode 100644
index 0000000..92f3f8b
--- /dev/null
+++ b/bin/named/fuzz.c
@@ -0,0 +1,483 @@
+/*
+ * Copyright (C) Internet Systems Consortium, Inc. ("ISC")
+ *
+ * This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/.
+ *
+ * See the COPYRIGHT file distributed with this work for additional
+ * information regarding copyright ownership.
+ */
+
+#include "config.h"
+
+#include <inttypes.h>
+#include <stdbool.h>
+
+#include <named/fuzz.h>
+
+#ifdef ENABLE_AFL
+#include <named/globals.h>
+#include <named/server.h>
+#include <sys/errno.h>
+
+#include <isc/app.h>
+#include <isc/condition.h>
+#include <isc/mutex.h>
+#include <isc/thread.h>
+#include <isc/util.h>
+#include <named/log.h>
+#include <dns/log.h>
+
+#include <stdlib.h>
+#include <string.h>
+#include <signal.h>
+#include <arpa/inet.h>
+#include <unistd.h>
+#include <pthread.h>
+
+#ifndef __AFL_LOOP
+#error To use American Fuzzy Lop you have to set CC to afl-clang-fast!!!
+#endif
+
+/*
+ * We are using pthreads directly because we might be using it with unthreaded
+ * version of BIND, where all thread functions are mocks. Since AFL for now only
+ * works on Linux it's not a problem.
+ */
+static pthread_cond_t cond;
+static pthread_mutex_t mutex;
+static bool ready;
+
+
+static void *
+fuzz_main_client(void *arg) {
+ char *host;
+ char *port;
+ struct sockaddr_in servaddr;
+ int sockfd;
+ int loop;
+ void *buf;
+
+ UNUSED(arg);
+
+ /*
+ * Parse named -A argument in the "address:port" syntax. Due to
+ * the syntax used, this only supports IPv4 addresses.
+ */
+
+ host = strdup(ns_g_fuzz_named_addr);
+ RUNTIME_CHECK(host != NULL);
+ port = strchr(host, ':');
+ RUNTIME_CHECK(port != NULL);
+ *port = 0;
+ ++port;
+
+ memset(&servaddr, 0, sizeof (servaddr));
+ servaddr.sin_family = AF_INET;
+ RUNTIME_CHECK(inet_pton(AF_INET, host, &servaddr.sin_addr) == 1);
+ servaddr.sin_port = htons(atoi(port));
+
+ free(host);
+
+ /* Wait for named to start. */
+ while (!ns_g_run_done) {
+ usleep(10000);
+ }
+
+ sockfd = socket(AF_INET, SOCK_DGRAM, 0);
+ RUNTIME_CHECK(sockfd != -1);
+
+ buf = malloc(65536);
+ RUNTIME_CHECK(buf != NULL);
+
+ loop = 100000;
+ while (loop--) {
+ ssize_t length;
+
+ length = read(0, buf, 65536);
+ if (length <= 0) {
+ usleep(1000000);
+ continue;
+ }
+
+ if (length > 4096) {
+ if (getenv("AFL_CMIN")) {
+ ns_server_flushonshutdown(ns_g_server,
+ false);
+ isc_app_shutdown();
+ return (NULL);
+ }
+ raise(SIGSTOP);
+ continue;
+ }
+
+ RUNTIME_CHECK(pthread_mutex_lock(&mutex) == ISC_R_SUCCESS);
+
+ ready = false;
+
+ ssize_t sent;
+
+ sent = sendto(sockfd, buf, length, 0,
+ (struct sockaddr *) &servaddr, sizeof(servaddr));
+ RUNTIME_CHECK(sent == length);
+
+ /* unclog */
+ recvfrom(sockfd, buf, 65536, MSG_DONTWAIT, NULL, NULL);
+
+ while (!ready)
+ pthread_cond_wait(&cond, &mutex);
+
+ RUNTIME_CHECK(pthread_mutex_unlock(&mutex) == ISC_R_SUCCESS);
+ }
+
+ free(buf);
+ close(sockfd);
+
+ ns_server_flushonshutdown(ns_g_server, false);
+ isc_app_shutdown();
+
+ return (NULL);
+}
+
+static void *
+fuzz_main_resolver(void *arg) {
+ char *shost, *sport, *rhost, *rport;
+ /* Query for A? aaaaaaaaaa.example. */
+ char respacket[] =
+ "\0\0\1 \0\1\0\0\0\0\0\0\naaaaaaaaaa\7example\0\0\1\0\1";
+ struct sockaddr_in servaddr, recaddr, recvaddr;
+ int sockfd;
+ int listenfd;
+ int loop;
+ char *buf, *rbuf;
+
+ UNUSED(arg);
+
+ /*
+ * Parse named -A argument in the "laddress:sport:raddress:rport"
+ * syntax. Due to the syntax used, this only supports IPv4 addresses.
+ */
+
+ shost = strdup(ns_g_fuzz_named_addr);
+ RUNTIME_CHECK(shost != NULL);
+ sport = strchr(shost, ':');
+ RUNTIME_CHECK(sport != NULL);
+ *sport = 0;
+ sport++;
+ rhost = strchr(sport, ':');
+ RUNTIME_CHECK(rhost != NULL);
+ *rhost = 0;
+ rhost++;
+ rport = strchr(rhost, ':');
+ RUNTIME_CHECK(rport != NULL);
+ *rport = 0;
+ rport++;
+
+ memset(&servaddr, 0, sizeof (servaddr));
+ servaddr.sin_family = AF_INET;
+ RUNTIME_CHECK(inet_pton(AF_INET, shost, &servaddr.sin_addr) == 1);
+ servaddr.sin_port = htons(atoi(sport));
+
+ memset(&recaddr, 0, sizeof (recaddr));
+ recaddr.sin_family = AF_INET;
+ RUNTIME_CHECK(inet_pton(AF_INET, rhost, &recaddr.sin_addr) == 1);
+ recaddr.sin_port = htons(atoi(rport));
+
+ free(shost);
+
+ /* Wait for named to start */
+ while (!ns_g_run_done) {
+ usleep(10000);
+ }
+
+ sockfd = socket(AF_INET, SOCK_DGRAM, 0);
+ RUNTIME_CHECK(sockfd != -1);
+
+ listenfd = socket(AF_INET, SOCK_DGRAM, 0);
+ RUNTIME_CHECK(listenfd != -1);
+ RUNTIME_CHECK(bind(listenfd, (struct sockaddr *)&recaddr,
+ sizeof(struct sockaddr_in)) == 0);
+
+ buf = malloc(65536);
+ rbuf = malloc(65536);
+ RUNTIME_CHECK(buf != NULL);
+ RUNTIME_CHECK(rbuf != NULL);
+
+ loop = 100000;
+ while (loop--) {
+ ssize_t length;
+ memset(buf, 0, 16);
+ length = read(0, buf, 65536);
+ if (length <= 0) {
+ usleep(1000000);
+ continue;
+ }
+
+ if (length > 4096) {
+ if (getenv("AFL_CMIN")) {
+ ns_server_flushonshutdown(ns_g_server,
+ false);
+ isc_app_shutdown();
+ return (NULL);
+ }
+ raise(SIGSTOP);
+ continue;
+ }
+
+ if (length < 16) {
+ length = 16;
+ }
+
+ RUNTIME_CHECK(pthread_mutex_lock(&mutex) == ISC_R_SUCCESS);
+
+ ready = false;
+
+ ssize_t sent;
+ /* Randomize query ID. */
+ int id = random();
+ respacket[0] = id >> 8;
+ respacket[1] = id & 0xff;
+
+ /* flush */
+ socklen_t socklen = sizeof(recvaddr);
+ sent = recvfrom(listenfd, rbuf, 65536, MSG_DONTWAIT,
+ (struct sockaddr *) &recvaddr, &socklen);
+
+ sent = sendto(sockfd, respacket, sizeof(respacket), 0,
+ (struct sockaddr *) &servaddr, sizeof(servaddr));
+ RUNTIME_CHECK(sent == sizeof(respacket));
+
+ socklen = sizeof(recvaddr);
+ sent = recvfrom(listenfd, rbuf, 65536, 0,
+ (struct sockaddr *) &recvaddr, &socklen);
+ RUNTIME_CHECK(sent > 0);
+
+ /* Copy QID and set QR so that response is always processed. */
+ buf[0] = rbuf[0];
+ buf[1] = rbuf[1];
+ buf[2] |= 0x80;
+
+ sent = sendto(listenfd, buf, length, 0,
+ (struct sockaddr *) &recvaddr, sizeof(recvaddr));
+ RUNTIME_CHECK(sent == length);
+
+ /* We might get additional questions here (e.g. for CNAME). */
+ for (;;) {
+ fd_set fds;
+ struct timeval tv;
+ int rv;
+ int max;
+
+ FD_ZERO(&fds);
+ FD_SET(listenfd, &fds);
+ FD_SET(sockfd, &fds);
+ tv.tv_sec = 10;
+ tv.tv_usec = 0;
+ max = (listenfd > sockfd ? listenfd : sockfd)+1;
+
+ rv = select(max, &fds, NULL, NULL, &tv);
+ RUNTIME_CHECK(rv > 0);
+
+ if (FD_ISSET(sockfd, &fds)) {
+ /* It's the reply, we're done. */
+ recvfrom(sockfd, buf, 65536, 0, NULL, NULL);
+ break;
+ }
+
+ /*
+ * We've got additional question (eg. cname chain)
+ * We are bouncing it - setting QR flag and NOERROR
+ * rcode and sending it back.
+ */
+
+ length = recvfrom(listenfd, buf, 65536, 0,
+ (struct sockaddr *) &recvaddr, &socklen);
+ buf[2] |= 0x80;
+ buf[3] &= 0xF0;
+ sent = sendto(listenfd, buf, length, 0,
+ (struct sockaddr *) &recvaddr,
+ sizeof(recvaddr));
+ RUNTIME_CHECK(sent == length);
+ }
+
+ while (!ready)
+ pthread_cond_wait(&cond, &mutex);
+
+ RUNTIME_CHECK(pthread_mutex_unlock(&mutex) == 0);
+ }
+
+ free(buf);
+ free(rbuf);
+ close(sockfd);
+ ns_server_flushonshutdown(ns_g_server, false);
+ isc_app_shutdown();
+
+ /*
+ * It's here just for the signature, that's how AFL detects if it's
+ * a 'persistent mode' binary.
+ */
+ __AFL_LOOP(0);
+
+ return (NULL);
+}
+
+static void *
+fuzz_main_tcp(void *arg) {
+ char *host;
+ char *port;
+ struct sockaddr_in servaddr;
+ int sockfd;
+ char *buf;
+ int loop;
+
+ UNUSED(arg);
+
+ /*
+ * Parse named -A argument in the "address:port" syntax. Due to
+ * the syntax used, this only supports IPv4 addresses.
+ */
+
+ host = strdup(ns_g_fuzz_named_addr);
+ RUNTIME_CHECK(host != NULL);
+ port = strchr(host, ':');
+ RUNTIME_CHECK(port != NULL);
+ *port = 0;
+ ++port;
+
+ memset(&servaddr, 0, sizeof (servaddr));
+ servaddr.sin_family = AF_INET;
+ RUNTIME_CHECK(inet_pton(AF_INET, host, &servaddr.sin_addr) == 1);
+ servaddr.sin_port = htons(atoi(port));
+
+ free(host);
+
+ /* Wait for named to start */
+ while (!ns_g_run_done) {
+ usleep(10000);
+ }
+
+ buf = malloc(65539);
+ RUNTIME_CHECK(buf != NULL);
+
+ loop = 100000;
+ while (loop--) {
+ ssize_t length;
+
+ if (ns_g_fuzz_type == ns_fuzz_tcpclient) {
+ /*
+ * To fuzz TCP client we have to put length at
+ * the start of packet.
+ */
+ length = read(0, buf+2, 65535);
+ buf[0] = length >> 8;
+ buf[1] = length & 0xff;
+ length += 2;
+ } else {
+ length = read(0, buf, 65535);
+ }
+ if (length <= 0) {
+ usleep(1000000);
+ continue;
+ }
+ if (ns_g_fuzz_type == ns_fuzz_http) {
+ /*
+ * This guarantees that the request will be processed.
+ */
+ buf[length++]='\r';
+ buf[length++]='\n';
+ buf[length++]='\r';
+ buf[length++]='\n';
+ }
+
+ RUNTIME_CHECK(pthread_mutex_lock(&mutex) == ISC_R_SUCCESS);
+
+ ready = false;
+
+ ssize_t sent;
+ int yes = 1;
+ int r;
+ sockfd = socket(AF_INET, SOCK_STREAM, 0);
+
+ RUNTIME_CHECK(sockfd != -1);
+ RUNTIME_CHECK(setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR,
+ &yes, sizeof(int)) == 0);
+
+ do {
+ r = connect(sockfd, (struct sockaddr*)&servaddr,
+ sizeof(servaddr));
+ } while (r != 0);
+
+ sent = write(sockfd, buf, length);
+ RUNTIME_CHECK(sent == length);
+ close(sockfd);
+
+ /* unclog */
+ recvfrom(sockfd, buf, 65537, MSG_DONTWAIT, NULL, NULL);
+
+ while (!ready)
+ pthread_cond_wait(&cond, &mutex);
+
+ RUNTIME_CHECK(pthread_mutex_unlock(&mutex) == ISC_R_SUCCESS);
+ }
+
+ free(buf);
+ close(sockfd);
+ ns_server_flushonshutdown(ns_g_server, false);
+ isc_app_shutdown();
+
+ return (NULL);
+}
+
+#endif /* ENABLE_AFL */
+
+void
+named_fuzz_notify(void) {
+#ifdef ENABLE_AFL
+ if (getenv("AFL_CMIN")) {
+ ns_server_flushonshutdown(ns_g_server, false);
+ isc_app_shutdown();
+ return;
+ }
+
+ raise(SIGSTOP);
+
+ RUNTIME_CHECK(pthread_mutex_lock(&mutex) == 0);
+
+ ready = true;
+
+ RUNTIME_CHECK(pthread_cond_signal(&cond) == 0);
+ RUNTIME_CHECK(pthread_mutex_unlock(&mutex) == 0);
+#endif /* ENABLE_AFL */
+}
+
+void
+named_fuzz_setup(void) {
+#ifdef ENABLE_AFL
+ if (getenv("__AFL_PERSISTENT") || getenv("AFL_CMIN")) {
+ pthread_t thread;
+ void *(fn) = NULL;
+
+ switch (ns_g_fuzz_type) {
+ case ns_fuzz_client:
+ fn = fuzz_main_client;
+ break;
+
+ case ns_fuzz_http:
+ case ns_fuzz_tcpclient:
+ case ns_fuzz_rndc:
+ fn = fuzz_main_tcp;
+ break;
+ case ns_fuzz_resolver:
+ fn = fuzz_main_resolver;
+ break;
+ default:
+ RUNTIME_CHECK(fn != NULL);
+ }
+
+ RUNTIME_CHECK(pthread_mutex_init(&mutex, NULL) == 0);
+ RUNTIME_CHECK(pthread_cond_init(&cond, NULL) == 0);
+ RUNTIME_CHECK(pthread_create(&thread, NULL, fn, NULL) == 0);
+ }
+#endif /* ENABLE_AFL */
+}