2 This file is part of systemd.
4 Copyright 2015 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
20 #include <linux/veth.h>
26 #include "sd-netlink.h"
28 #include "alloc-util.h"
29 #include "ether-addr-util.h"
30 #include "lockfile-util.h"
31 #include "netlink-util.h"
32 #include "nspawn-network.h"
33 #include "siphash24.h"
34 #include "socket-util.h"
35 #include "stat-util.h"
36 #include "string-util.h"
37 #include "udev-util.h"
40 #define HOST_HASH_KEY SD_ID128_MAKE(1a,37,6f,c7,46,ec,45,0b,ad,a3,d5,31,06,60,5d,b1)
41 #define CONTAINER_HASH_KEY SD_ID128_MAKE(c3,c4,f9,19,b5,57,b2,1c,e6,cf,14,27,03,9c,ee,a2)
42 #define VETH_EXTRA_HOST_HASH_KEY SD_ID128_MAKE(48,c7,f6,b7,ea,9d,4c,9e,b7,28,d4,de,91,d5,bf,66)
43 #define VETH_EXTRA_CONTAINER_HASH_KEY SD_ID128_MAKE(af,50,17,61,ce,f9,4d,35,84,0d,2b,20,54,be,ce,59)
44 #define MACVLAN_HASH_KEY SD_ID128_MAKE(00,13,6d,bc,66,83,44,81,bb,0c,f9,51,1f,24,a6,6f)
46 static int remove_one_link(sd_netlink
*rtnl
, const char *name
) {
47 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
53 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_DELLINK
, 0);
55 return log_error_errno(r
, "Failed to allocate netlink message: %m");
57 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, name
);
59 return log_error_errno(r
, "Failed to add netlink interface name: %m");
61 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
62 if (r
== -ENODEV
) /* Already gone */
65 return log_error_errno(r
, "Failed to remove interface %s: %m", name
);
70 static int generate_mac(
71 const char *machine_name
,
72 struct ether_addr
*mac
,
81 l
= strlen(machine_name
);
82 sz
= sizeof(sd_id128_t
) + l
;
88 /* fetch some persistent data unique to the host */
89 r
= sd_id128_get_machine((sd_id128_t
*) v
);
93 /* combine with some data unique (on this host) to this
94 * container instance */
95 i
= mempcpy(v
+ sizeof(sd_id128_t
), machine_name
, l
);
98 memcpy(i
, &idx
, sizeof(idx
));
101 /* Let's hash the host machine ID plus the container name. We
102 * use a fixed, but originally randomly created hash key here. */
103 result
= htole64(siphash24(v
, sz
, hash_key
.bytes
));
105 assert_cc(ETH_ALEN
<= sizeof(result
));
106 memcpy(mac
->ether_addr_octet
, &result
, ETH_ALEN
);
108 /* see eth_random_addr in the kernel */
109 mac
->ether_addr_octet
[0] &= 0xfe; /* clear multicast bit */
110 mac
->ether_addr_octet
[0] |= 0x02; /* set local assignment bit (IEEE802) */
118 const char *ifname_host
,
119 const struct ether_addr
*mac_host
,
120 const char *ifname_container
,
121 const struct ether_addr
*mac_container
) {
123 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
129 assert(ifname_container
);
130 assert(mac_container
);
132 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_NEWLINK
, 0);
134 return log_error_errno(r
, "Failed to allocate netlink message: %m");
136 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, ifname_host
);
138 return log_error_errno(r
, "Failed to add netlink interface name: %m");
140 r
= sd_netlink_message_append_ether_addr(m
, IFLA_ADDRESS
, mac_host
);
142 return log_error_errno(r
, "Failed to add netlink MAC address: %m");
144 r
= sd_netlink_message_open_container(m
, IFLA_LINKINFO
);
146 return log_error_errno(r
, "Failed to open netlink container: %m");
148 r
= sd_netlink_message_open_container_union(m
, IFLA_INFO_DATA
, "veth");
150 return log_error_errno(r
, "Failed to open netlink container: %m");
152 r
= sd_netlink_message_open_container(m
, VETH_INFO_PEER
);
154 return log_error_errno(r
, "Failed to open netlink container: %m");
156 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, ifname_container
);
158 return log_error_errno(r
, "Failed to add netlink interface name: %m");
160 r
= sd_netlink_message_append_ether_addr(m
, IFLA_ADDRESS
, mac_container
);
162 return log_error_errno(r
, "Failed to add netlink MAC address: %m");
164 r
= sd_netlink_message_append_u32(m
, IFLA_NET_NS_PID
, pid
);
166 return log_error_errno(r
, "Failed to add netlink namespace field: %m");
168 r
= sd_netlink_message_close_container(m
);
170 return log_error_errno(r
, "Failed to close netlink container: %m");
172 r
= sd_netlink_message_close_container(m
);
174 return log_error_errno(r
, "Failed to close netlink container: %m");
176 r
= sd_netlink_message_close_container(m
);
178 return log_error_errno(r
, "Failed to close netlink container: %m");
180 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
182 return log_error_errno(r
, "Failed to add new veth interfaces (%s:%s): %m", ifname_host
, ifname_container
);
187 int setup_veth(const char *machine_name
,
189 char iface_name
[IFNAMSIZ
],
192 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
193 struct ether_addr mac_host
, mac_container
;
196 assert(machine_name
);
200 /* Use two different interface name prefixes depending whether
201 * we are in bridge mode or not. */
202 snprintf(iface_name
, IFNAMSIZ
- 1, "%s-%s",
203 bridge
? "vb" : "ve", machine_name
);
205 r
= generate_mac(machine_name
, &mac_container
, CONTAINER_HASH_KEY
, 0);
207 return log_error_errno(r
, "Failed to generate predictable MAC address for container side: %m");
209 r
= generate_mac(machine_name
, &mac_host
, HOST_HASH_KEY
, 0);
211 return log_error_errno(r
, "Failed to generate predictable MAC address for host side: %m");
213 r
= sd_netlink_open(&rtnl
);
215 return log_error_errno(r
, "Failed to connect to netlink: %m");
217 r
= add_veth(rtnl
, pid
, iface_name
, &mac_host
, "host0", &mac_container
);
221 i
= (int) if_nametoindex(iface_name
);
223 return log_error_errno(errno
, "Failed to resolve interface %s: %m", iface_name
);
228 int setup_veth_extra(
229 const char *machine_name
,
233 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
238 assert(machine_name
);
241 if (strv_isempty(pairs
))
244 r
= sd_netlink_open(&rtnl
);
246 return log_error_errno(r
, "Failed to connect to netlink: %m");
248 STRV_FOREACH_PAIR(a
, b
, pairs
) {
249 struct ether_addr mac_host
, mac_container
;
251 r
= generate_mac(machine_name
, &mac_container
, VETH_EXTRA_CONTAINER_HASH_KEY
, idx
);
253 return log_error_errno(r
, "Failed to generate predictable MAC address for container side of extra veth link: %m");
255 r
= generate_mac(machine_name
, &mac_host
, VETH_EXTRA_HOST_HASH_KEY
, idx
);
257 return log_error_errno(r
, "Failed to generate predictable MAC address for container side of extra veth link: %m");
259 r
= add_veth(rtnl
, pid
, *a
, &mac_host
, *b
, &mac_container
);
269 static int join_bridge(sd_netlink
*rtnl
, const char *veth_name
, const char *bridge_name
) {
270 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
277 bridge_ifi
= (int) if_nametoindex(bridge_name
);
281 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_SETLINK
, 0);
285 r
= sd_rtnl_message_link_set_flags(m
, IFF_UP
, IFF_UP
);
289 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, veth_name
);
293 r
= sd_netlink_message_append_u32(m
, IFLA_MASTER
, bridge_ifi
);
297 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
304 static int create_bridge(sd_netlink
*rtnl
, const char *bridge_name
) {
305 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
308 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_NEWLINK
, 0);
312 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, bridge_name
);
316 r
= sd_netlink_message_open_container(m
, IFLA_LINKINFO
);
320 r
= sd_netlink_message_open_container_union(m
, IFLA_INFO_DATA
, "bridge");
324 r
= sd_netlink_message_close_container(m
);
328 r
= sd_netlink_message_close_container(m
);
332 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
339 int setup_bridge(const char *veth_name
, const char *bridge_name
, bool create
) {
340 _cleanup_release_lock_file_ LockFile bridge_lock
= LOCK_FILE_INIT
;
341 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
348 r
= sd_netlink_open(&rtnl
);
350 return log_error_errno(r
, "Failed to connect to netlink: %m");
353 /* We take a system-wide lock here, so that we can safely check whether there's still a member in the
354 * bridge before removing it, without risking interference from other nspawn instances. */
356 r
= make_lock_file("/run/systemd/nspawn-network-zone", LOCK_EX
, &bridge_lock
);
358 return log_error_errno(r
, "Failed to take network zone lock: %m");
362 bridge_ifi
= join_bridge(rtnl
, veth_name
, bridge_name
);
365 if (bridge_ifi
!= -ENODEV
|| !create
|| n
> 10)
366 return log_error_errno(bridge_ifi
, "Failed to add interface %s to bridge %s: %m", veth_name
, bridge_name
);
368 /* Count attempts, so that we don't enter an endless loop here. */
371 /* The bridge doesn't exist yet. Let's create it */
372 r
= create_bridge(rtnl
, bridge_name
);
374 return log_error_errno(r
, "Failed to create bridge interface %s: %m", bridge_name
);
376 /* Try again, now that the bridge exists */
380 int remove_bridge(const char *bridge_name
) {
381 _cleanup_release_lock_file_ LockFile bridge_lock
= LOCK_FILE_INIT
;
382 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
386 /* Removes the specified bridge, but only if it is currently empty */
388 if (isempty(bridge_name
))
391 r
= make_lock_file("/run/systemd/nspawn-network-zone", LOCK_EX
, &bridge_lock
);
393 return log_error_errno(r
, "Failed to take network zone lock: %m");
395 path
= strjoina("/sys/class/net/", bridge_name
, "/brif");
397 r
= dir_is_empty(path
);
398 if (r
== -ENOENT
) /* Already gone? */
401 return log_error_errno(r
, "Can't detect if bridge %s is empty: %m", bridge_name
);
402 if (r
== 0) /* Still populated, leave it around */
405 r
= sd_netlink_open(&rtnl
);
407 return log_error_errno(r
, "Failed to connect to netlink: %m");
409 return remove_one_link(rtnl
, bridge_name
);
412 static int parse_interface(struct udev
*udev
, const char *name
) {
413 _cleanup_udev_device_unref_
struct udev_device
*d
= NULL
;
414 char ifi_str
[2 + DECIMAL_STR_MAX(int)];
417 ifi
= (int) if_nametoindex(name
);
419 return log_error_errno(errno
, "Failed to resolve interface %s: %m", name
);
421 sprintf(ifi_str
, "n%i", ifi
);
422 d
= udev_device_new_from_device_id(udev
, ifi_str
);
424 return log_error_errno(errno
, "Failed to get udev device for interface %s: %m", name
);
426 if (udev_device_get_is_initialized(d
) <= 0) {
427 log_error("Network interface %s is not initialized yet.", name
);
434 int move_network_interfaces(pid_t pid
, char **ifaces
) {
435 _cleanup_udev_unref_
struct udev
*udev
= NULL
;
436 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
440 if (strv_isempty(ifaces
))
443 r
= sd_netlink_open(&rtnl
);
445 return log_error_errno(r
, "Failed to connect to netlink: %m");
449 log_error("Failed to connect to udev.");
453 STRV_FOREACH(i
, ifaces
) {
454 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
457 ifi
= parse_interface(udev
, *i
);
461 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_SETLINK
, ifi
);
463 return log_error_errno(r
, "Failed to allocate netlink message: %m");
465 r
= sd_netlink_message_append_u32(m
, IFLA_NET_NS_PID
, pid
);
467 return log_error_errno(r
, "Failed to append namespace PID to netlink message: %m");
469 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
471 return log_error_errno(r
, "Failed to move interface %s to namespace: %m", *i
);
477 int setup_macvlan(const char *machine_name
, pid_t pid
, char **ifaces
) {
478 _cleanup_udev_unref_
struct udev
*udev
= NULL
;
479 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
484 if (strv_isempty(ifaces
))
487 r
= sd_netlink_open(&rtnl
);
489 return log_error_errno(r
, "Failed to connect to netlink: %m");
493 log_error("Failed to connect to udev.");
497 STRV_FOREACH(i
, ifaces
) {
498 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
499 _cleanup_free_
char *n
= NULL
;
500 struct ether_addr mac
;
503 ifi
= parse_interface(udev
, *i
);
507 r
= generate_mac(machine_name
, &mac
, MACVLAN_HASH_KEY
, idx
++);
509 return log_error_errno(r
, "Failed to create MACVLAN MAC address: %m");
511 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_NEWLINK
, 0);
513 return log_error_errno(r
, "Failed to allocate netlink message: %m");
515 r
= sd_netlink_message_append_u32(m
, IFLA_LINK
, ifi
);
517 return log_error_errno(r
, "Failed to add netlink interface index: %m");
519 n
= strappend("mv-", *i
);
523 strshorten(n
, IFNAMSIZ
-1);
525 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, n
);
527 return log_error_errno(r
, "Failed to add netlink interface name: %m");
529 r
= sd_netlink_message_append_ether_addr(m
, IFLA_ADDRESS
, &mac
);
531 return log_error_errno(r
, "Failed to add netlink MAC address: %m");
533 r
= sd_netlink_message_append_u32(m
, IFLA_NET_NS_PID
, pid
);
535 return log_error_errno(r
, "Failed to add netlink namespace field: %m");
537 r
= sd_netlink_message_open_container(m
, IFLA_LINKINFO
);
539 return log_error_errno(r
, "Failed to open netlink container: %m");
541 r
= sd_netlink_message_open_container_union(m
, IFLA_INFO_DATA
, "macvlan");
543 return log_error_errno(r
, "Failed to open netlink container: %m");
545 r
= sd_netlink_message_append_u32(m
, IFLA_MACVLAN_MODE
, MACVLAN_MODE_BRIDGE
);
547 return log_error_errno(r
, "Failed to append macvlan mode: %m");
549 r
= sd_netlink_message_close_container(m
);
551 return log_error_errno(r
, "Failed to close netlink container: %m");
553 r
= sd_netlink_message_close_container(m
);
555 return log_error_errno(r
, "Failed to close netlink container: %m");
557 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
559 return log_error_errno(r
, "Failed to add new macvlan interfaces: %m");
565 int setup_ipvlan(const char *machine_name
, pid_t pid
, char **ifaces
) {
566 _cleanup_udev_unref_
struct udev
*udev
= NULL
;
567 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
571 if (strv_isempty(ifaces
))
574 r
= sd_netlink_open(&rtnl
);
576 return log_error_errno(r
, "Failed to connect to netlink: %m");
580 log_error("Failed to connect to udev.");
584 STRV_FOREACH(i
, ifaces
) {
585 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
586 _cleanup_free_
char *n
= NULL
;
589 ifi
= parse_interface(udev
, *i
);
593 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_NEWLINK
, 0);
595 return log_error_errno(r
, "Failed to allocate netlink message: %m");
597 r
= sd_netlink_message_append_u32(m
, IFLA_LINK
, ifi
);
599 return log_error_errno(r
, "Failed to add netlink interface index: %m");
601 n
= strappend("iv-", *i
);
605 strshorten(n
, IFNAMSIZ
-1);
607 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, n
);
609 return log_error_errno(r
, "Failed to add netlink interface name: %m");
611 r
= sd_netlink_message_append_u32(m
, IFLA_NET_NS_PID
, pid
);
613 return log_error_errno(r
, "Failed to add netlink namespace field: %m");
615 r
= sd_netlink_message_open_container(m
, IFLA_LINKINFO
);
617 return log_error_errno(r
, "Failed to open netlink container: %m");
619 r
= sd_netlink_message_open_container_union(m
, IFLA_INFO_DATA
, "ipvlan");
621 return log_error_errno(r
, "Failed to open netlink container: %m");
623 r
= sd_netlink_message_append_u16(m
, IFLA_IPVLAN_MODE
, IPVLAN_MODE_L2
);
625 return log_error_errno(r
, "Failed to add ipvlan mode: %m");
627 r
= sd_netlink_message_close_container(m
);
629 return log_error_errno(r
, "Failed to close netlink container: %m");
631 r
= sd_netlink_message_close_container(m
);
633 return log_error_errno(r
, "Failed to close netlink container: %m");
635 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
637 return log_error_errno(r
, "Failed to add new ipvlan interfaces: %m");
643 int veth_extra_parse(char ***l
, const char *p
) {
644 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
647 r
= extract_first_word(&p
, &a
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
650 if (r
== 0 || !ifname_valid(a
))
653 r
= extract_first_word(&p
, &b
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
656 if (r
== 0 || !ifname_valid(b
)) {
666 r
= strv_push_pair(l
, a
, b
);
674 int remove_veth_links(const char *primary
, char **pairs
) {
675 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
679 /* In some cases the kernel might pin the veth links between host and container even after the namespace
680 * died. Hence, let's better remove them explicitly too. */
682 if (isempty(primary
) && strv_isempty(pairs
))
685 r
= sd_netlink_open(&rtnl
);
687 return log_error_errno(r
, "Failed to connect to netlink: %m");
689 remove_one_link(rtnl
, primary
);
691 STRV_FOREACH_PAIR(a
, b
, pairs
)
692 remove_one_link(rtnl
, *a
);