Merge pull request #6449 from libgit2/1.4_sshv1.4.5

ssh: perform host key checking by default

7 files changed, 348 insertions, 97 deletions

diff --git a/CMakeLists.txt b/CMakeLists.txt
index c497ab1fe..92bd2db2b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -3,7 +3,7 @@
 
 cmake_minimum_required(VERSION 3.5.1)
 
-project(libgit2 VERSION "1.4.4" LANGUAGES C)
+project(libgit2 VERSION "1.4.5" LANGUAGES C)
 
 # Add find modules to the path
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${PROJECT_SOURCE_DIR}/cmake")
diff --git a/ci/test.sh b/ci/test.sh
index 0815522a9..60d94caf8 100755
--- a/ci/test.sh
+++ b/ci/test.sh
@@ -144,6 +144,11 @@ if [ -z "$SKIP_SSH_TESTS" ]; then
 		echo "[localhost]:2222 $algorithm $key" >>"${HOME}/.ssh/known_hosts"
 	done <"${SSHD_DIR}/id_rsa.pub"
 
+	# Append the github.com keys for the tests that don't override checks.
+	# We ask for ssh-rsa to test that the selection based off of known_hosts
+	# is working.
+	ssh-keyscan -t ssh-rsa github.com >>"${HOME}/.ssh/known_hosts"
+
 	# Get the fingerprint for localhost and remove the colons so we can
 	# parse it as a hex number. Older versions have a different output
 	# format.
diff --git a/docs/changelog.md b/docs/changelog.md
index a6794ab9d..7190777b7 100644
--- a/docs/changelog.md
+++ b/docs/changelog.md
@@ -1,3 +1,14 @@
+v1.4.5
+------
+
+🔒 This is a security release to address CVE-2023-22742: when compiled using the optional, included libssh2 backend, libgit2 fails to verify SSH keys by default.
+
+When using an SSH remote with the optional, included libssh2 backend, libgit2 does not perform certificate checking by default. Prior versions of libgit2 require the caller to set the `certificate_check` field of libgit2's `git_remote_callbacks` structure - if a certificate check callback is not set, libgit2 does not perform any certificate checking. This means that by default - without configuring a certificate check callback, clients will not perform validation on the server SSH keys and may be subject to a man-in-the-middle attack.
+
+The libgit2 security team would like to thank the Julia and Rust security teams for responsibly disclosing this vulnerability and assisting with fixing the vulnerability.
+
+All users of the v1.4 release line are recommended to upgrade.
+
 v1.4.4
 ------
 
diff --git a/include/git2/version.h b/include/git2/version.h
index f6f0dee24..9d5433009 100644
--- a/include/git2/version.h
+++ b/include/git2/version.h
@@ -7,10 +7,10 @@
 #ifndef INCLUDE_git_version_h__
 #define INCLUDE_git_version_h__
 
-#define LIBGIT2_VERSION "1.4.4"
+#define LIBGIT2_VERSION "1.4.5"
 #define LIBGIT2_VER_MAJOR 1
 #define LIBGIT2_VER_MINOR 4
-#define LIBGIT2_VER_REVISION 4
+#define LIBGIT2_VER_REVISION 5
 #define LIBGIT2_VER_PATCH 0
 
 #define LIBGIT2_SOVERSION "1.4"
diff --git a/package.json b/package.json
index a52437b00..f8715a88d 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "libgit2",
-  "version": "1.4.4",
+  "version": "1.4.5",
   "repo": "https://github.com/libgit2/libgit2",
   "description": " A cross-platform, linkable library implementation of Git that you can use in your application.",
   "install": "mkdir build && cd build && cmake .. && cmake --build ."
diff --git a/src/transports/ssh.c b/src/transports/ssh.c
index 89f085230..85e779744 100644
--- a/src/transports/ssh.c
+++ b/src/transports/ssh.c
@@ -421,15 +421,119 @@ static int request_creds(git_credential **out, ssh_subtransport *t, const char *
 	return 0;
 }
 
+#define KNOWN_HOSTS_FILE ".ssh/known_hosts"
+
+/*
+ * Load the known_hosts file.
+ *
+ * Returns success but leaves the output NULL if we couldn't find the file.
+ */
+static int load_known_hosts(LIBSSH2_KNOWNHOSTS **hosts, LIBSSH2_SESSION *session)
+{
+	git_str path = GIT_STR_INIT, home = GIT_STR_INIT;
+	LIBSSH2_KNOWNHOSTS *known_hosts = NULL;
+	int error;
+
+	GIT_ASSERT_ARG(hosts);
+
+	if ((error = git__getenv(&home, "HOME")) < 0)
+		return error;
+
+	if ((error = git_str_joinpath(&path, git_str_cstr(&home), KNOWN_HOSTS_FILE)) < 0)
+		goto out;
+
+	if ((known_hosts = libssh2_knownhost_init(session)) == NULL) {
+		ssh_error(session, "error initializing known hosts");
+		error = -1;
+		goto out;
+	}
+
+	/*
+	 * Try to read the file and consider not finding it as not trusting the
+	 * host rather than an error.
+	 */
+	error = libssh2_knownhost_readfile(known_hosts, git_str_cstr(&path), LIBSSH2_KNOWNHOST_FILE_OPENSSH);
+	if (error == LIBSSH2_ERROR_FILE)
+		error = 0;
+	if (error < 0)
+		ssh_error(session, "error reading known_hosts");
+
+out:
+	*hosts = known_hosts;
+
+	git_str_clear(&home);
+	git_str_clear(&path);
+
+	return error;
+}
+
+static const char *hostkey_type_to_string(int type)
+{
+	switch (type) {
+	case LIBSSH2_KNOWNHOST_KEY_SSHRSA:
+		return "ssh-rsa";
+	case LIBSSH2_KNOWNHOST_KEY_SSHDSS:
+		return "ssh-dss";
+#ifdef LIBSSH2_KNOWNHOST_KEY_ECDSA_256
+	case LIBSSH2_KNOWNHOST_KEY_ECDSA_256:
+		return "ecdsa-sha2-nistp256";
+	case LIBSSH2_KNOWNHOST_KEY_ECDSA_384:
+		return "ecdsa-sha2-nistp384";
+	case LIBSSH2_KNOWNHOST_KEY_ECDSA_521:
+		return "ecdsa-sha2-nistp521";
+#endif
+#ifdef LIBSSH2_KNOWNHOST_KEY_ED25519
+	case LIBSSH2_KNOWNHOST_KEY_ED25519:
+		return "ssh-ed25519";
+#endif
+	}
+
+	return NULL;
+}
+
+/*
+ * We figure out what kind of key we want to ask the remote for by trying to
+ * look it up with a nonsense key and using that mismatch to figure out what key
+ * we do have stored for the host.
+ *
+ * Returns the string to pass to libssh2_session_method_pref or NULL if we were
+ * unable to find anything or an error happened.
+ */
+static const char *find_hostkey_preference(LIBSSH2_KNOWNHOSTS *known_hosts, const char *hostname, int port)
+{
+	struct libssh2_knownhost *host = NULL;
+	/* Specify no key type so we don't filter on that */
+	int type = LIBSSH2_KNOWNHOST_TYPE_PLAIN | LIBSSH2_KNOWNHOST_KEYENC_RAW;
+	const char key = '\0';
+	int error;
+
+	/*
+	 * In case of mismatch, we can find the type of key from known_hosts in
+	 * the returned host's information as it means that an entry was found
+	 * but our nonsense key obviously didn't match.
+	 */
+	error = libssh2_knownhost_checkp(known_hosts, hostname, port, &key, 1, type, &host);
+	if (error == LIBSSH2_KNOWNHOST_CHECK_MISMATCH)
+		return hostkey_type_to_string(host->typemask & LIBSSH2_KNOWNHOST_KEY_MASK);
+
+	return NULL;
+}
+
 static int _git_ssh_session_create(
 	LIBSSH2_SESSION **session,
+	LIBSSH2_KNOWNHOSTS **hosts,
+	const char *hostname,
+	int port,
 	git_stream *io)
 {
 	int rc = 0;
 	LIBSSH2_SESSION *s;
+	LIBSSH2_KNOWNHOSTS *known_hosts;
 	git_socket_stream *socket = GIT_CONTAINER_OF(io, git_socket_stream, parent);
+	const char *keytype = NULL;
 
 	GIT_ASSERT_ARG(session);
+	GIT_ASSERT_ARG(hosts);
 
 	s = libssh2_session_init();
 	if (!s) {
@@ -437,21 +541,225 @@ static int _git_ssh_session_create(
 		return -1;
 	}
 
+	if ((rc = load_known_hosts(&known_hosts, s)) < 0) {
+		ssh_error(s, "error loading known_hosts");
+		libssh2_session_free(s);
+		return -1;
+	}
+
+	if ((keytype = find_hostkey_preference(known_hosts, hostname, port)) != NULL) {
+		do {
+			rc = libssh2_session_method_pref(s, LIBSSH2_METHOD_HOSTKEY, keytype);
+		} while (LIBSSH2_ERROR_EAGAIN == rc || LIBSSH2_ERROR_TIMEOUT == rc);
+		if (rc != LIBSSH2_ERROR_NONE) {
+			ssh_error(s, "failed to set hostkey preference");
+			goto on_error;
+		}
+	}
+
+
 	do {
 		rc = libssh2_session_handshake(s, socket->s);
 	} while (LIBSSH2_ERROR_EAGAIN == rc || LIBSSH2_ERROR_TIMEOUT == rc);
 
 	if (rc != LIBSSH2_ERROR_NONE) {
 		ssh_error(s, "failed to start SSH session");
-		libssh2_session_free(s);
-		return -1;
+		goto on_error;
 	}
 
 	libssh2_session_set_blocking(s, 1);
 
 	*session = s;
+	*hosts = known_hosts;
 
 	return 0;
+
+on_error:
+	libssh2_knownhost_free(known_hosts);
+	libssh2_session_free(s);
+	return -1;
+}
+
+
+/*
+ * Returns the typemask argument to pass to libssh2_knownhost_check{,p} based on
+ * the type of key that libssh2_session_hostkey returns.
+ */
+static int fingerprint_type_mask(int keytype)
+{
+	int mask = LIBSSH2_KNOWNHOST_TYPE_PLAIN | LIBSSH2_KNOWNHOST_KEYENC_RAW;
+	return mask;
+
+	switch (keytype) {
+	case LIBSSH2_HOSTKEY_TYPE_RSA:
+		mask |= LIBSSH2_KNOWNHOST_KEY_SSHRSA;
+		break;
+	case LIBSSH2_HOSTKEY_TYPE_DSS:
+		mask |= LIBSSH2_KNOWNHOST_KEY_SSHDSS;
+		break;
+#ifdef LIBSSH2_HOSTKEY_TYPE_ECDSA_256
+	case LIBSSH2_HOSTKEY_TYPE_ECDSA_256:
+		mask |= LIBSSH2_KNOWNHOST_KEY_ECDSA_256;
+		break;
+	case LIBSSH2_HOSTKEY_TYPE_ECDSA_384:
+		mask |= LIBSSH2_KNOWNHOST_KEY_ECDSA_384;
+		break;
+	case LIBSSH2_HOSTKEY_TYPE_ECDSA_521:
+		mask |= LIBSSH2_KNOWNHOST_KEY_ECDSA_521;
+		break;
+#endif
+#ifdef LIBSSH2_HOSTKEY_TYPE_ED25519
+	case LIBSSH2_HOSTKEY_TYPE_ED25519:
+		mask |= LIBSSH2_KNOWNHOST_KEY_ED25519;
+		break;
+#endif
+	}
+
+	return mask;
+}
+
+/*
+ * Check the host against the user's known_hosts file.
+ *
+ * Returns 1/0 for valid/''not-valid or <0 for an error
+ */
+static int check_against_known_hosts(
+	LIBSSH2_SESSION *session,
+	LIBSSH2_KNOWNHOSTS *known_hosts,
+	const char *hostname,
+	int port,
+	const char *key,
+	size_t key_len,
+	int key_type)
+{
+	int check, typemask, ret = 0;
+	struct libssh2_knownhost *host = NULL;
+
+	if (known_hosts == NULL)
+		return 0;
+
+	typemask = fingerprint_type_mask(key_type);
+	check = libssh2_knownhost_checkp(known_hosts, hostname, port, key, key_len, typemask, &host);
+	if (check == LIBSSH2_KNOWNHOST_CHECK_FAILURE) {
+		ssh_error(session, "error checking for known host");
+		return -1;
+	}
+
+	ret = check == LIBSSH2_KNOWNHOST_CHECK_MATCH ? 1 : 0;
+
+	return ret;
+}
+
+/*
+ * Perform the check for the session's certificate against known hosts if
+ * possible and then ask the user if they have a callback.
+ *
+ * Returns 1/0 for valid/not-valid or <0 for an error
+ */
+static int check_certificate(
+	LIBSSH2_SESSION *session,
+	LIBSSH2_KNOWNHOSTS *known_hosts,
+	git_transport_certificate_check_cb check_cb,
+	void *check_cb_payload,
+	const char *host,
+	int port)
+{
+	git_cert_hostkey cert = {{ 0 }};
+	const char *key;
+	size_t cert_len;
+	int cert_type, cert_valid = 0, error = 0;
+
+	if ((key = libssh2_session_hostkey(session, &cert_len, &cert_type)) == NULL) {
+		ssh_error(session, "failed to retrieve hostkey");
+		return -1;
+	}
+
+	if ((cert_valid = check_against_known_hosts(session, known_hosts, host, port, key, cert_len, cert_type)) < 0)
+		return -1;
+
+	cert.parent.cert_type = GIT_CERT_HOSTKEY_LIBSSH2;
+	if (key != NULL) {
+		cert.type |= GIT_CERT_SSH_RAW;
+		cert.hostkey = key;
+		cert.hostkey_len = cert_len;
+		switch (cert_type) {
+		case LIBSSH2_HOSTKEY_TYPE_RSA:
+			cert.raw_type = GIT_CERT_SSH_RAW_TYPE_RSA;
+			break;
+		case LIBSSH2_HOSTKEY_TYPE_DSS:
+			cert.raw_type = GIT_CERT_SSH_RAW_TYPE_DSS;
+			break;
+
+#ifdef LIBSSH2_HOSTKEY_TYPE_ECDSA_256
+		case LIBSSH2_HOSTKEY_TYPE_ECDSA_256:
+			cert.raw_type = GIT_CERT_SSH_RAW_TYPE_KEY_ECDSA_256;
+			break;
+		case LIBSSH2_HOSTKEY_TYPE_ECDSA_384:
+			cert.raw_type = GIT_CERT_SSH_RAW_TYPE_KEY_ECDSA_384;
+			break;
+		case LIBSSH2_KNOWNHOST_KEY_ECDSA_521:
+			cert.raw_type = GIT_CERT_SSH_RAW_TYPE_KEY_ECDSA_521;
+			break;
+#endif
+
+#ifdef LIBSSH2_HOSTKEY_TYPE_ED25519
+		case LIBSSH2_HOSTKEY_TYPE_ED25519:
+			cert.raw_type = GIT_CERT_SSH_RAW_TYPE_KEY_ED25519;
+			break;
+#endif
+		default:
+			cert.raw_type = GIT_CERT_SSH_RAW_TYPE_UNKNOWN;
+		}
+	}
+
+#ifdef LIBSSH2_HOSTKEY_HASH_SHA256
+	key = libssh2_hostkey_hash(session, LIBSSH2_HOSTKEY_HASH_SHA256);
+	if (key != NULL) {
+		cert.type |= GIT_CERT_SSH_SHA256;
+		memcpy(&cert.hash_sha256, key, 32);
+	}
+#endif
+
+	key = libssh2_hostkey_hash(session, LIBSSH2_HOSTKEY_HASH_SHA1);
+	if (key != NULL) {
+		cert.type |= GIT_CERT_SSH_SHA1;
+		memcpy(&cert.hash_sha1, key, 20);
+	}
+
+	key = libssh2_hostkey_hash(session, LIBSSH2_HOSTKEY_HASH_MD5);
+	if (key != NULL) {
+		cert.type |= GIT_CERT_SSH_MD5;
+		memcpy(&cert.hash_md5, key, 16);
+	}
+
+	if (cert.type == 0) {
+		git_error_set(GIT_ERROR_SSH, "unable to get the host key");
+		return -1;
+	}
+
+	git_error_clear();
+	error = 0;
+	if (!cert_valid) {
+		git_error_set(GIT_ERROR_SSH, "invalid or unknown remote ssh hostkey");
+		error = GIT_ECERTIFICATE;
+	}
+
+	if (check_cb != NULL) {
+		git_cert_hostkey *cert_ptr = &cert;
+		git_error_state previous_error = {0};
+
+		git_error_state_capture(&previous_error, error);
+		error = check_cb((git_cert *) cert_ptr, cert_valid, host, check_cb_payload);
+		if (error == GIT_PASSTHROUGH) {
+			error = git_error_state_restore(&previous_error);
+		} else if (error < 0 && !git_error_last()) {
+			git_error_set(GIT_ERROR_NET, "user canceled hostkey check");
+		}
+
+		git_error_state_free(&previous_error);
+	}
+
+	return error;
 }
 
 #define SSH_DEFAULT_PORT "22"
@@ -462,11 +770,12 @@ static int _git_ssh_setup_conn(
 	const char *cmd,
 	git_smart_subtransport_stream **stream)
 {
-	int auth_methods, error = 0;
+	int auth_methods, error = 0, port;
 	ssh_stream *s;
 	git_credential *cred = NULL;
 	LIBSSH2_SESSION *session=NULL;
 	LIBSSH2_CHANNEL *channel=NULL;
+	LIBSSH2_KNOWNHOSTS *known_hosts = NULL;
 
 	t->current_stream = NULL;
 
@@ -490,96 +799,20 @@ static int _git_ssh_setup_conn(
 	    (error = git_stream_connect(s->io)) < 0)
 		goto done;
 
-	if ((error = _git_ssh_session_create(&session, s->io)) < 0)
-		goto done;
-
-	if (t->owner->connect_opts.callbacks.certificate_check != NULL) {
-		git_cert_hostkey cert = {{ 0 }}, *cert_ptr;
-		const char *key;
-		size_t cert_len;
-		int cert_type;
-
-		cert.parent.cert_type = GIT_CERT_HOSTKEY_LIBSSH2;
-
-		key = libssh2_session_hostkey(session, &cert_len, &cert_type);
-		if (key != NULL) {
-			cert.type |= GIT_CERT_SSH_RAW;
-			cert.hostkey = key;
-			cert.hostkey_len = cert_len;
-			switch (cert_type) {
-				case LIBSSH2_HOSTKEY_TYPE_RSA:
-					cert.raw_type = GIT_CERT_SSH_RAW_TYPE_RSA;
-					break;
-				case LIBSSH2_HOSTKEY_TYPE_DSS:
-					cert.raw_type = GIT_CERT_SSH_RAW_TYPE_DSS;
-					break;
-
-#ifdef LIBSSH2_HOSTKEY_TYPE_ECDSA_256
-				case LIBSSH2_HOSTKEY_TYPE_ECDSA_256:
-					cert.raw_type = GIT_CERT_SSH_RAW_TYPE_KEY_ECDSA_256;
-					break;
-				case LIBSSH2_HOSTKEY_TYPE_ECDSA_384:
-					cert.raw_type = GIT_CERT_SSH_RAW_TYPE_KEY_ECDSA_384;
-					break;
-				case LIBSSH2_KNOWNHOST_KEY_ECDSA_521:
-					cert.raw_type = GIT_CERT_SSH_RAW_TYPE_KEY_ECDSA_521;
-					break;
-#endif
-
-#ifdef LIBSSH2_HOSTKEY_TYPE_ED25519
-				case LIBSSH2_HOSTKEY_TYPE_ED25519:
-					cert.raw_type = GIT_CERT_SSH_RAW_TYPE_KEY_ED25519;
-					break;
-#endif
-				default:
-					cert.raw_type = GIT_CERT_SSH_RAW_TYPE_UNKNOWN;
-			}
-		}
-
-#ifdef LIBSSH2_HOSTKEY_HASH_SHA256
-		key = libssh2_hostkey_hash(session, LIBSSH2_HOSTKEY_HASH_SHA256);
-		if (key != NULL) {
-			cert.type |= GIT_CERT_SSH_SHA256;
-			memcpy(&cert.hash_sha256, key, 32);
-		}
-#endif
-
-		key = libssh2_hostkey_hash(session, LIBSSH2_HOSTKEY_HASH_SHA1);
-		if (key != NULL) {
-			cert.type |= GIT_CERT_SSH_SHA1;
-			memcpy(&cert.hash_sha1, key, 20);
-		}
-
-		key = libssh2_hostkey_hash(session, LIBSSH2_HOSTKEY_HASH_MD5);
-		if (key != NULL) {
-			cert.type |= GIT_CERT_SSH_MD5;
-			memcpy(&cert.hash_md5, key, 16);
-		}
-
-		if (cert.type == 0) {
-			git_error_set(GIT_ERROR_SSH, "unable to get the host key");
-			error = -1;
-			goto done;
-		}
-
-		/* We don't currently trust any hostkeys */
-		git_error_clear();
-
-		cert_ptr = &cert;
-
-		error = t->owner->connect_opts.callbacks.certificate_check(
-			(git_cert *)cert_ptr,
-			0,
-			s->url.host,
-			t->owner->connect_opts.callbacks.payload);
+	/*
+	 * Try to parse the port as a number, if we can't then fall back to
+	 * default. It would be nice if we could get the port that was resolved
+	 * as part of the stream connection, but that's not something that's
+	 * exposed.
+	 */
+	if (git__strntol32(&port, s->url.port, strlen(s->url.port), NULL, 10) < 0)
+		port = -1;
 
-		if (error < 0 && error != GIT_PASSTHROUGH) {
-			if (!git_error_last())
-				git_error_set(GIT_ERROR_NET, "user cancelled hostkey check");
+	if ((error = _git_ssh_session_create(&session, &known_hosts, s->url.host, port, s->io)) < 0)
+		goto done;
 
-			goto done;
-		}
-	}
+	if ((error = check_certificate(session, known_hosts, t->owner->connect_opts.callbacks.certificate_check, t->owner->connect_opts.callbacks.payload, s->url.host, port)) < 0)
+		goto done;
 
 	/* we need the username to ask for auth methods */
 	if (!s->url.username) {
@@ -651,6 +884,8 @@ done:
 	if (error < 0) {
 		ssh_stream_free(*stream);
 
+		if (known_hosts)
+			libssh2_knownhost_free(known_hosts);
 		if (session)
 			libssh2_session_free(session);
 	}
diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
index f293c158d..336fc3b3d 100644
--- a/tests/CMakeLists.txt
+++ b/tests/CMakeLists.txt
@@ -70,10 +70,10 @@ endfunction(ADD_CLAR_TEST)
 
 add_clar_test(offline             -v -xonline)
 add_clar_test(invasive            -v -score::ftruncate -sfilter::stream::bigfile -sodb::largefiles -siterator::workdir::filesystem_gunk -srepo::init -srepo::init::at_filesystem_root)
-add_clar_test(online              -v -sonline -xonline::customcert)
+add_clar_test(online              -v -sonline -xonline::customcert -xonline::clone::ssh_auth_methods)
 add_clar_test(online_customcert   -v -sonline::customcert)
 add_clar_test(gitdaemon           -v -sonline::push)
-add_clar_test(ssh                 -v -sonline::push -sonline::clone::ssh_cert -sonline::clone::ssh_with_paths -sonline::clone::path_whitespace_ssh)
+add_clar_test(ssh                 -v -sonline::push -sonline::clone::ssh_cert -sonline::clone::ssh_with_paths -sonline::clone::path_whitespace_ssh -sonline::clone::ssh_auth_methods)
 add_clar_test(proxy               -v -sonline::clone::proxy)
 add_clar_test(auth_clone          -v -sonline::clone::cred)
 add_clar_test(auth_clone_and_push -v -sonline::clone::push -sonline::push)