diff options
Diffstat (limited to 'doc')
-rw-r--r-- | doc/migration/.gitignore | 2 | ||||
-rw-r--r-- | doc/migration/Makefile | 20 | ||||
-rw-r--r-- | doc/migration/README | 51 | ||||
-rw-r--r-- | doc/migration/source.c | 92 | ||||
-rw-r--r-- | doc/migration/target.c | 102 | ||||
-rw-r--r-- | doc/platform-requirements/.gitignore | 1 | ||||
-rw-r--r-- | doc/platform-requirements/Makefile | 4 | ||||
-rw-r--r-- | doc/platform-requirements/common.h | 1 | ||||
-rw-r--r-- | doc/platform-requirements/listen-vs-repair.c | 128 | ||||
-rw-r--r-- | doc/platform-requirements/reuseaddr-priority.c | 6 |
10 files changed, 402 insertions, 5 deletions
diff --git a/doc/migration/.gitignore b/doc/migration/.gitignore new file mode 100644 index 0000000..59cb765 --- /dev/null +++ b/doc/migration/.gitignore @@ -0,0 +1,2 @@ +/source +/target diff --git a/doc/migration/Makefile b/doc/migration/Makefile new file mode 100644 index 0000000..04f6891 --- /dev/null +++ b/doc/migration/Makefile @@ -0,0 +1,20 @@ +# SPDX-License-Identifier: GPL-2.0-or-later +# +# PASST - Plug A Simple Socket Transport +# for qemu/UNIX domain socket mode +# +# PASTA - Pack A Subtle Tap Abstraction +# for network namespace/tap device mode +# +# Copyright (c) 2025 Red Hat GmbH +# Author: Stefano Brivio <sbrivio@redhat.com> + +TARGETS = source target +CFLAGS = -Wall -Wextra -pedantic + +all: $(TARGETS) + +$(TARGETS): %: %.c + +clean: + rm -f $(TARGETS) diff --git a/doc/migration/README b/doc/migration/README new file mode 100644 index 0000000..375603b --- /dev/null +++ b/doc/migration/README @@ -0,0 +1,51 @@ +<!--- +SPDX-License-Identifier: GPL-2.0-or-later +Copyright (c) 2025 Red Hat GmbH +Author: Stefano Brivio <sbrivio@redhat.com> +--> + +Migration +========= + +These test programs show a migration of a TCP connection from one process to +another using the TCP_REPAIR socket option. + +The two processes are a mock of the matching implementation in passt(1), and run +unprivileged, so they rely on the passt-repair helper to connect to them and set +or clear TCP_REPAIR on the connection socket, transferred to the helper using +SCM_RIGHTS. + +The passt-repair helper needs to have the CAP_NET_ADMIN capability, or run as +root. + +Example of usage +---------------- + +* Start the test server + + $ nc -l 9999 + +* Start the source side of the TCP client (mock of the source instance of passt) + + $ ./source 127.0.0.1 9999 9998 /tmp/repair.sock + +* The client sends a test string, and waits for a connection from passt-repair + + # passt-repair /tmp/repair.sock + +* The socket is now in repair mode, and `source` dumps sequences, then exits + + sending sequence: 3244673313 + receiving sequence: 2250449386 + +* Continue the connection on the target side, restarting from those sequences + + $ ./target 127.0.0.1 9999 9998 /tmp/repair.sock 3244673313 2250449386 + +* The target side now waits for a connection from passt-repair + + # passt-repair /tmp/repair.sock + +* The target side asks passt-repair to switch the socket to repair mode, sets up + the TCP sequences, then asks passt-repair to clear repair mode, and sends a + test string to the server diff --git a/doc/migration/source.c b/doc/migration/source.c new file mode 100644 index 0000000..d44ebf1 --- /dev/null +++ b/doc/migration/source.c @@ -0,0 +1,92 @@ +// SPDX-License-Identifier: GPL-2.0-or-later + +/* PASST - Plug A Simple Socket Transport + * for qemu/UNIX domain socket mode + * + * PASTA - Pack A Subtle Tap Abstraction + * for network namespace/tap device mode + * + * doc/migration/source.c - Mock of TCP migration source, use with passt-repair + * + * Copyright (c) 2025 Red Hat GmbH + * Author: Stefano Brivio <sbrivio@redhat.com> + */ + +#include <arpa/inet.h> +#include <sys/types.h> +#include <sys/socket.h> +#include <sys/un.h> +#include <stdint.h> +#include <stdio.h> +#include <stdlib.h> +#include <limits.h> +#include <unistd.h> +#include <netdb.h> +#include <netinet/tcp.h> + +int main(int argc, char **argv) +{ + struct sockaddr_in a = { AF_INET, htons(atoi(argv[3])), { 0 }, { 0 } }; + struct addrinfo hints = { 0, AF_UNSPEC, SOCK_STREAM, 0, 0, + NULL, NULL, NULL }; + struct sockaddr_un a_helper = { AF_UNIX, { 0 } }; + int seq, s, s_helper; + int8_t cmd; + struct iovec iov = { &cmd, sizeof(cmd) }; + char buf[CMSG_SPACE(sizeof(int))]; + struct msghdr msg = { NULL, 0, &iov, 1, buf, sizeof(buf), 0 }; + struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg); + socklen_t seqlen = sizeof(int); + struct addrinfo *r; + + (void)argc; + + if (argc != 5) { + fprintf(stderr, "%s DST_ADDR DST_PORT SRC_PORT HELPER_PATH\n", + argv[0]); + return -1; + } + + strcpy(a_helper.sun_path, argv[4]); + getaddrinfo(argv[1], argv[2], &hints, &r); + + /* Connect socket to server and send some data */ + s = socket(r->ai_family, SOCK_STREAM, IPPROTO_TCP); + setsockopt(s, SOL_SOCKET, SO_REUSEADDR, &((int){ 1 }), sizeof(int)); + bind(s, (struct sockaddr *)&a, sizeof(a)); + connect(s, r->ai_addr, r->ai_addrlen); + send(s, "before migration\n", sizeof("before migration\n"), 0); + + /* Wait for helper */ + s_helper = socket(AF_UNIX, SOCK_STREAM, 0); + unlink(a_helper.sun_path); + bind(s_helper, (struct sockaddr *)&a_helper, sizeof(a_helper)); + listen(s_helper, 1); + s_helper = accept(s_helper, NULL, NULL); + + /* Set up message for helper, with socket */ + cmsg->cmsg_level = SOL_SOCKET; + cmsg->cmsg_type = SCM_RIGHTS; + cmsg->cmsg_len = CMSG_LEN(sizeof(int)); + memcpy(CMSG_DATA(cmsg), &s, sizeof(s)); + + /* Send command to helper: turn repair mode on, wait for reply */ + cmd = TCP_REPAIR_ON; + sendmsg(s_helper, &msg, 0); + recv(s_helper, &((int8_t){ 0 }), 1, 0); + + /* Terminate helper */ + close(s_helper); + + /* Get sending sequence */ + seq = TCP_SEND_QUEUE; + setsockopt(s, SOL_TCP, TCP_REPAIR_QUEUE, &seq, sizeof(seq)); + getsockopt(s, SOL_TCP, TCP_QUEUE_SEQ, &seq, &seqlen); + fprintf(stdout, "%u ", seq); + + /* Get receiving sequence */ + seq = TCP_RECV_QUEUE; + setsockopt(s, SOL_TCP, TCP_REPAIR_QUEUE, &seq, sizeof(seq)); + getsockopt(s, SOL_TCP, TCP_QUEUE_SEQ, &seq, &seqlen); + fprintf(stdout, "%u\n", seq); +} diff --git a/doc/migration/target.c b/doc/migration/target.c new file mode 100644 index 0000000..f7d3108 --- /dev/null +++ b/doc/migration/target.c @@ -0,0 +1,102 @@ +// SPDX-License-Identifier: GPL-2.0-or-later + +/* PASST - Plug A Simple Socket Transport + * for qemu/UNIX domain socket mode + * + * PASTA - Pack A Subtle Tap Abstraction + * for network namespace/tap device mode + * + * doc/migration/target.c - Mock of TCP migration target, use with passt-repair + * + * Copyright (c) 2025 Red Hat GmbH + * Author: Stefano Brivio <sbrivio@redhat.com> + */ + +#include <arpa/inet.h> +#include <sys/types.h> +#include <sys/socket.h> +#include <sys/un.h> +#include <stdint.h> +#include <stdio.h> +#include <stdlib.h> +#include <limits.h> +#include <unistd.h> +#include <netdb.h> +#include <netinet/tcp.h> + +int main(int argc, char **argv) +{ + struct sockaddr_in a = { AF_INET, htons(atoi(argv[3])), { 0 }, { 0 } }; + struct addrinfo hints = { 0, AF_UNSPEC, SOCK_STREAM, 0, 0, + NULL, NULL, NULL }; + struct sockaddr_un a_helper = { AF_UNIX, { 0 } }; + int s, s_helper, seq; + int8_t cmd; + struct iovec iov = { &cmd, sizeof(cmd) }; + char buf[CMSG_SPACE(sizeof(int))]; + struct msghdr msg = { NULL, 0, &iov, 1, buf, sizeof(buf), 0 }; + struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg); + struct addrinfo *r; + + (void)argc; + + strcpy(a_helper.sun_path, argv[4]); + getaddrinfo(argv[1], argv[2], &hints, &r); + + if (argc != 7) { + fprintf(stderr, + "%s DST_ADDR DST_PORT SRC_PORT HELPER_PATH SSEQ RSEQ\n", + argv[0]); + return -1; + } + + /* Prepare socket, bind to source port */ + s = socket(r->ai_family, SOCK_STREAM, IPPROTO_TCP); + setsockopt(s, SOL_SOCKET, SO_REUSEADDR, &((int){ 1 }), sizeof(int)); + bind(s, (struct sockaddr *)&a, sizeof(a)); + + /* Wait for helper */ + s_helper = socket(AF_UNIX, SOCK_STREAM, 0); + unlink(a_helper.sun_path); + bind(s_helper, (struct sockaddr *)&a_helper, sizeof(a_helper)); + listen(s_helper, 1); + s_helper = accept(s_helper, NULL, NULL); + + /* Set up message for helper, with socket */ + cmsg->cmsg_level = SOL_SOCKET; + cmsg->cmsg_type = SCM_RIGHTS; + cmsg->cmsg_len = CMSG_LEN(sizeof(int)); + memcpy(CMSG_DATA(cmsg), &s, sizeof(s)); + + /* Send command to helper: turn repair mode on, wait for reply */ + cmd = TCP_REPAIR_ON; + sendmsg(s_helper, &msg, 0); + recv(s_helper, &((int){ 0 }), 1, 0); + + /* Set sending sequence */ + seq = TCP_SEND_QUEUE; + setsockopt(s, SOL_TCP, TCP_REPAIR_QUEUE, &seq, sizeof(seq)); + seq = atoi(argv[5]); + setsockopt(s, SOL_TCP, TCP_QUEUE_SEQ, &seq, sizeof(seq)); + + /* Set receiving sequence */ + seq = TCP_RECV_QUEUE; + setsockopt(s, SOL_TCP, TCP_REPAIR_QUEUE, &seq, sizeof(seq)); + seq = atoi(argv[6]); + setsockopt(s, SOL_TCP, TCP_QUEUE_SEQ, &seq, sizeof(seq)); + + /* Connect setting kernel state only, without actual SYN / handshake */ + connect(s, r->ai_addr, r->ai_addrlen); + + /* Send command to helper: turn repair mode off, wait for reply */ + cmd = TCP_REPAIR_OFF; + sendmsg(s_helper, &msg, 0); + + recv(s_helper, &((int8_t){ 0 }), 1, 0); + + /* Terminate helper */ + close(s_helper); + + /* Send some more data */ + send(s, "after migration\n", sizeof("after migration\n"), 0); +} diff --git a/doc/platform-requirements/.gitignore b/doc/platform-requirements/.gitignore index 3b5a10a..f6272cf 100644 --- a/doc/platform-requirements/.gitignore +++ b/doc/platform-requirements/.gitignore @@ -1,3 +1,4 @@ +/listen-vs-repair /reuseaddr-priority /recv-zero /udp-close-dup diff --git a/doc/platform-requirements/Makefile b/doc/platform-requirements/Makefile index 6a7d374..83930ef 100644 --- a/doc/platform-requirements/Makefile +++ b/doc/platform-requirements/Makefile @@ -3,8 +3,8 @@ # Copyright Red Hat # Author: David Gibson <david@gibson.dropbear.id.au> -TARGETS = reuseaddr-priority recv-zero udp-close-dup -SRCS = reuseaddr-priority.c recv-zero.c udp-close-dup.c +TARGETS = reuseaddr-priority recv-zero udp-close-dup listen-vs-repair +SRCS = reuseaddr-priority.c recv-zero.c udp-close-dup.c listen-vs-repair.c CFLAGS = -Wall all: cppcheck clang-tidy $(TARGETS:%=check-%) diff --git a/doc/platform-requirements/common.h b/doc/platform-requirements/common.h index 8844b1e..e85fc2b 100644 --- a/doc/platform-requirements/common.h +++ b/doc/platform-requirements/common.h @@ -15,6 +15,7 @@ #include <stdio.h> #include <stdlib.h> +__attribute__((format(printf, 1, 2), noreturn)) static inline void die(const char *fmt, ...) { va_list ap; diff --git a/doc/platform-requirements/listen-vs-repair.c b/doc/platform-requirements/listen-vs-repair.c new file mode 100644 index 0000000..d31fe3f --- /dev/null +++ b/doc/platform-requirements/listen-vs-repair.c @@ -0,0 +1,128 @@ +// SPDX-License-Identifier: GPL-2.0-or-later + +/* liste-vs-repair.c + * + * Do listening sockets have address conflicts with sockets under repair + * ==================================================================== + * + * When we accept() an incoming connection the accept()ed socket will have the + * same local address as the listening socket. This can be a complication on + * migration. On the migration target we've already set up listening sockets + * according to the command line. However to restore connections that we're + * migrating in we need to bind the new sockets to the same address, which would + * be an address conflict on the face of it. This test program verifies that + * enabling repair mode before bind() correctly suppresses that conflict. + * + * Copyright Red Hat + * Author: David Gibson <david@gibson.dropbear.id.au> + */ + +/* NOLINTNEXTLINE(bugprone-reserved-identifier,cert-dcl37-c,cert-dcl51-cpp) */ +#define _GNU_SOURCE + +#include <arpa/inet.h> +#include <errno.h> +#include <linux/netlink.h> +#include <linux/rtnetlink.h> +#include <net/if.h> +#include <netinet/in.h> +#include <netinet/tcp.h> +#include <sched.h> +#include <stdbool.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> + +#include "common.h" + +#define PORT 13256U +#define CPORT 13257U + +/* 127.0.0.1:PORT */ +static const struct sockaddr_in addr = SOCKADDR_INIT(INADDR_LOOPBACK, PORT); + +/* 127.0.0.1:CPORT */ +static const struct sockaddr_in caddr = SOCKADDR_INIT(INADDR_LOOPBACK, CPORT); + +/* Put ourselves into a network sandbox */ +static void net_sandbox(void) +{ + /* NOLINTNEXTLINE(altera-struct-pack-align) */ + const struct req_t { + struct nlmsghdr nlh; + struct ifinfomsg ifm; + } __attribute__((packed)) req = { + .nlh.nlmsg_type = RTM_NEWLINK, + .nlh.nlmsg_flags = NLM_F_REQUEST, + .nlh.nlmsg_len = sizeof(req), + .nlh.nlmsg_seq = 1, + .ifm.ifi_family = AF_UNSPEC, + .ifm.ifi_index = 1, + .ifm.ifi_flags = IFF_UP, + .ifm.ifi_change = IFF_UP, + }; + int nl; + + if (unshare(CLONE_NEWUSER | CLONE_NEWNET)) + die("unshare(): %s\n", strerror(errno)); + + /* Bring up lo in the new netns */ + nl = socket(AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC, NETLINK_ROUTE); + if (nl < 0) + die("Can't create netlink socket: %s\n", strerror(errno)); + + if (send(nl, &req, sizeof(req), 0) < 0) + die("Netlink send(): %s\n", strerror(errno)); + close(nl); +} + +static void check(void) +{ + int s1, s2, op; + + s1 = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP); + if (s1 < 0) + die("socket() 1: %s\n", strerror(errno)); + + if (bind(s1, (struct sockaddr *)&addr, sizeof(addr))) + die("bind() 1: %s\n", strerror(errno)); + + if (listen(s1, 0)) + die("listen(): %s\n", strerror(errno)); + + s2 = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP); + if (s2 < 0) + die("socket() 2: %s\n", strerror(errno)); + + op = TCP_REPAIR_ON; + if (setsockopt(s2, SOL_TCP, TCP_REPAIR, &op, sizeof(op))) + die("TCP_REPAIR: %s\n", strerror(errno)); + + if (bind(s2, (struct sockaddr *)&addr, sizeof(addr))) + die("bind() 2: %s\n", strerror(errno)); + + if (connect(s2, (struct sockaddr *)&caddr, sizeof(caddr))) + die("connect(): %s\n", strerror(errno)); + + op = TCP_REPAIR_OFF_NO_WP; + if (setsockopt(s2, SOL_TCP, TCP_REPAIR, &op, sizeof(op))) + die("TCP_REPAIR: %s\n", strerror(errno)); + + close(s1); + close(s2); +} + +int main(int argc, char *argv[]) +{ + (void)argc; + (void)argv; + + net_sandbox(); + + check(); + + printf("Repair mode appears to properly suppress conflicts with listening sockets\n"); + + exit(0); +} diff --git a/doc/platform-requirements/reuseaddr-priority.c b/doc/platform-requirements/reuseaddr-priority.c index 701b6ff..af39a39 100644 --- a/doc/platform-requirements/reuseaddr-priority.c +++ b/doc/platform-requirements/reuseaddr-priority.c @@ -46,13 +46,13 @@ /* Different cases for receiving socket configuration */ enum sock_type { /* Socket is bound to 0.0.0.0:DSTPORT and not connected */ - SOCK_BOUND_ANY = 0, + SOCK_BOUND_ANY, /* Socket is bound to 127.0.0.1:DSTPORT and not connected */ - SOCK_BOUND_LO = 1, + SOCK_BOUND_LO, /* Socket is bound to 0.0.0.0:DSTPORT and connected to 127.0.0.1:SRCPORT */ - SOCK_CONNECTED = 2, + SOCK_CONNECTED, NUM_SOCK_TYPES, }; |