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>
25 #include "sd-netlink.h"
27 #include "alloc-util.h"
28 #include "ether-addr-util.h"
29 #include "lockfile-util.h"
30 #include "netlink-util.h"
31 #include "nspawn-network.h"
32 #include "siphash24.h"
33 #include "socket-util.h"
34 #include "stat-util.h"
35 #include "string-util.h"
36 #include "udev-util.h"
39 #define HOST_HASH_KEY SD_ID128_MAKE(1a,37,6f,c7,46,ec,45,0b,ad,a3,d5,31,06,60,5d,b1)
40 #define CONTAINER_HASH_KEY SD_ID128_MAKE(c3,c4,f9,19,b5,57,b2,1c,e6,cf,14,27,03,9c,ee,a2)
41 #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)
42 #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)
43 #define MACVLAN_HASH_KEY SD_ID128_MAKE(00,13,6d,bc,66,83,44,81,bb,0c,f9,51,1f,24,a6,6f)
45 static int remove_one_link(sd_netlink
*rtnl
, const char *name
) {
46 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
52 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_DELLINK
, 0);
54 return log_error_errno(r
, "Failed to allocate netlink message: %m");
56 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, name
);
58 return log_error_errno(r
, "Failed to add netlink interface name: %m");
60 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
61 if (r
== -ENODEV
) /* Already gone */
64 return log_error_errno(r
, "Failed to remove interface %s: %m", name
);
69 static int generate_mac(
70 const char *machine_name
,
71 struct ether_addr
*mac
,
80 l
= strlen(machine_name
);
81 sz
= sizeof(sd_id128_t
) + l
;
87 /* fetch some persistent data unique to the host */
88 r
= sd_id128_get_machine((sd_id128_t
*) v
);
92 /* combine with some data unique (on this host) to this
93 * container instance */
94 i
= mempcpy(v
+ sizeof(sd_id128_t
), machine_name
, l
);
97 memcpy(i
, &idx
, sizeof(idx
));
100 /* Let's hash the host machine ID plus the container name. We
101 * use a fixed, but originally randomly created hash key here. */
102 result
= htole64(siphash24(v
, sz
, hash_key
.bytes
));
104 assert_cc(ETH_ALEN
<= sizeof(result
));
105 memcpy(mac
->ether_addr_octet
, &result
, ETH_ALEN
);
107 /* see eth_random_addr in the kernel */
108 mac
->ether_addr_octet
[0] &= 0xfe; /* clear multicast bit */
109 mac
->ether_addr_octet
[0] |= 0x02; /* set local assignment bit (IEEE802) */
117 const char *ifname_host
,
118 const struct ether_addr
*mac_host
,
119 const char *ifname_container
,
120 const struct ether_addr
*mac_container
) {
122 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
128 assert(ifname_container
);
129 assert(mac_container
);
131 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_NEWLINK
, 0);
133 return log_error_errno(r
, "Failed to allocate netlink message: %m");
135 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, ifname_host
);
137 return log_error_errno(r
, "Failed to add netlink interface name: %m");
139 r
= sd_netlink_message_append_ether_addr(m
, IFLA_ADDRESS
, mac_host
);
141 return log_error_errno(r
, "Failed to add netlink MAC address: %m");
143 r
= sd_netlink_message_open_container(m
, IFLA_LINKINFO
);
145 return log_error_errno(r
, "Failed to open netlink container: %m");
147 r
= sd_netlink_message_open_container_union(m
, IFLA_INFO_DATA
, "veth");
149 return log_error_errno(r
, "Failed to open netlink container: %m");
151 r
= sd_netlink_message_open_container(m
, VETH_INFO_PEER
);
153 return log_error_errno(r
, "Failed to open netlink container: %m");
155 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, ifname_container
);
157 return log_error_errno(r
, "Failed to add netlink interface name: %m");
159 r
= sd_netlink_message_append_ether_addr(m
, IFLA_ADDRESS
, mac_container
);
161 return log_error_errno(r
, "Failed to add netlink MAC address: %m");
163 r
= sd_netlink_message_append_u32(m
, IFLA_NET_NS_PID
, pid
);
165 return log_error_errno(r
, "Failed to add netlink namespace field: %m");
167 r
= sd_netlink_message_close_container(m
);
169 return log_error_errno(r
, "Failed to close netlink container: %m");
171 r
= sd_netlink_message_close_container(m
);
173 return log_error_errno(r
, "Failed to close netlink container: %m");
175 r
= sd_netlink_message_close_container(m
);
177 return log_error_errno(r
, "Failed to close netlink container: %m");
179 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
181 return log_error_errno(r
, "Failed to add new veth interfaces (%s:%s): %m", ifname_host
, ifname_container
);
186 int setup_veth(const char *machine_name
,
188 char iface_name
[IFNAMSIZ
],
191 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
192 struct ether_addr mac_host
, mac_container
;
195 assert(machine_name
);
199 /* Use two different interface name prefixes depending whether
200 * we are in bridge mode or not. */
201 snprintf(iface_name
, IFNAMSIZ
- 1, "%s-%s",
202 bridge
? "vb" : "ve", machine_name
);
204 r
= generate_mac(machine_name
, &mac_container
, CONTAINER_HASH_KEY
, 0);
206 return log_error_errno(r
, "Failed to generate predictable MAC address for container side: %m");
208 r
= generate_mac(machine_name
, &mac_host
, HOST_HASH_KEY
, 0);
210 return log_error_errno(r
, "Failed to generate predictable MAC address for host side: %m");
212 r
= sd_netlink_open(&rtnl
);
214 return log_error_errno(r
, "Failed to connect to netlink: %m");
216 r
= add_veth(rtnl
, pid
, iface_name
, &mac_host
, "host0", &mac_container
);
220 i
= (int) if_nametoindex(iface_name
);
222 return log_error_errno(errno
, "Failed to resolve interface %s: %m", iface_name
);
227 int setup_veth_extra(
228 const char *machine_name
,
232 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
237 assert(machine_name
);
240 if (strv_isempty(pairs
))
243 r
= sd_netlink_open(&rtnl
);
245 return log_error_errno(r
, "Failed to connect to netlink: %m");
247 STRV_FOREACH_PAIR(a
, b
, pairs
) {
248 struct ether_addr mac_host
, mac_container
;
250 r
= generate_mac(machine_name
, &mac_container
, VETH_EXTRA_CONTAINER_HASH_KEY
, idx
);
252 return log_error_errno(r
, "Failed to generate predictable MAC address for container side of extra veth link: %m");
254 r
= generate_mac(machine_name
, &mac_host
, VETH_EXTRA_HOST_HASH_KEY
, idx
);
256 return log_error_errno(r
, "Failed to generate predictable MAC address for container side of extra veth link: %m");
258 r
= add_veth(rtnl
, pid
, *a
, &mac_host
, *b
, &mac_container
);
268 static int join_bridge(sd_netlink
*rtnl
, const char *veth_name
, const char *bridge_name
) {
269 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
276 bridge_ifi
= (int) if_nametoindex(bridge_name
);
280 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_SETLINK
, 0);
284 r
= sd_rtnl_message_link_set_flags(m
, IFF_UP
, IFF_UP
);
288 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, veth_name
);
292 r
= sd_netlink_message_append_u32(m
, IFLA_MASTER
, bridge_ifi
);
296 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
303 static int create_bridge(sd_netlink
*rtnl
, const char *bridge_name
) {
304 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
307 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_NEWLINK
, 0);
311 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, bridge_name
);
315 r
= sd_netlink_message_open_container(m
, IFLA_LINKINFO
);
319 r
= sd_netlink_message_open_container_union(m
, IFLA_INFO_DATA
, "bridge");
323 r
= sd_netlink_message_close_container(m
);
327 r
= sd_netlink_message_close_container(m
);
331 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
338 int setup_bridge(const char *veth_name
, const char *bridge_name
, bool create
) {
339 _cleanup_release_lock_file_ LockFile bridge_lock
= LOCK_FILE_INIT
;
340 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
347 r
= sd_netlink_open(&rtnl
);
349 return log_error_errno(r
, "Failed to connect to netlink: %m");
352 /* We take a system-wide lock here, so that we can safely check whether there's still a member in the
353 * bridge before removing it, without risking interference from other nspawn instances. */
355 r
= make_lock_file("/run/systemd/nspawn-network-zone", LOCK_EX
, &bridge_lock
);
357 return log_error_errno(r
, "Failed to take network zone lock: %m");
361 bridge_ifi
= join_bridge(rtnl
, veth_name
, bridge_name
);
364 if (bridge_ifi
!= -ENODEV
|| !create
|| n
> 10)
365 return log_error_errno(bridge_ifi
, "Failed to add interface %s to bridge %s: %m", veth_name
, bridge_name
);
367 /* Count attempts, so that we don't enter an endless loop here. */
370 /* The bridge doesn't exist yet. Let's create it */
371 r
= create_bridge(rtnl
, bridge_name
);
373 return log_error_errno(r
, "Failed to create bridge interface %s: %m", bridge_name
);
375 /* Try again, now that the bridge exists */
379 int remove_bridge(const char *bridge_name
) {
380 _cleanup_release_lock_file_ LockFile bridge_lock
= LOCK_FILE_INIT
;
381 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
385 /* Removes the specified bridge, but only if it is currently empty */
387 if (isempty(bridge_name
))
390 r
= make_lock_file("/run/systemd/nspawn-network-zone", LOCK_EX
, &bridge_lock
);
392 return log_error_errno(r
, "Failed to take network zone lock: %m");
394 path
= strjoina("/sys/class/net/", bridge_name
, "/brif");
396 r
= dir_is_empty(path
);
397 if (r
== -ENOENT
) /* Already gone? */
400 return log_error_errno(r
, "Can't detect if bridge %s is empty: %m", bridge_name
);
401 if (r
== 0) /* Still populated, leave it around */
404 r
= sd_netlink_open(&rtnl
);
406 return log_error_errno(r
, "Failed to connect to netlink: %m");
408 return remove_one_link(rtnl
, bridge_name
);
411 static int parse_interface(struct udev
*udev
, const char *name
) {
412 _cleanup_udev_device_unref_
struct udev_device
*d
= NULL
;
413 char ifi_str
[2 + DECIMAL_STR_MAX(int)];
416 ifi
= (int) if_nametoindex(name
);
418 return log_error_errno(errno
, "Failed to resolve interface %s: %m", name
);
420 sprintf(ifi_str
, "n%i", ifi
);
421 d
= udev_device_new_from_device_id(udev
, ifi_str
);
423 return log_error_errno(errno
, "Failed to get udev device for interface %s: %m", name
);
425 if (udev_device_get_is_initialized(d
) <= 0) {
426 log_error("Network interface %s is not initialized yet.", name
);
433 int move_network_interfaces(pid_t pid
, char **ifaces
) {
434 _cleanup_udev_unref_
struct udev
*udev
= NULL
;
435 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
439 if (strv_isempty(ifaces
))
442 r
= sd_netlink_open(&rtnl
);
444 return log_error_errno(r
, "Failed to connect to netlink: %m");
448 log_error("Failed to connect to udev.");
452 STRV_FOREACH(i
, ifaces
) {
453 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
456 ifi
= parse_interface(udev
, *i
);
460 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_SETLINK
, ifi
);
462 return log_error_errno(r
, "Failed to allocate netlink message: %m");
464 r
= sd_netlink_message_append_u32(m
, IFLA_NET_NS_PID
, pid
);
466 return log_error_errno(r
, "Failed to append namespace PID to netlink message: %m");
468 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
470 return log_error_errno(r
, "Failed to move interface %s to namespace: %m", *i
);
476 int setup_macvlan(const char *machine_name
, pid_t pid
, char **ifaces
) {
477 _cleanup_udev_unref_
struct udev
*udev
= NULL
;
478 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
483 if (strv_isempty(ifaces
))
486 r
= sd_netlink_open(&rtnl
);
488 return log_error_errno(r
, "Failed to connect to netlink: %m");
492 log_error("Failed to connect to udev.");
496 STRV_FOREACH(i
, ifaces
) {
497 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
498 _cleanup_free_
char *n
= NULL
;
499 struct ether_addr mac
;
502 ifi
= parse_interface(udev
, *i
);
506 r
= generate_mac(machine_name
, &mac
, MACVLAN_HASH_KEY
, idx
++);
508 return log_error_errno(r
, "Failed to create MACVLAN MAC address: %m");
510 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_NEWLINK
, 0);
512 return log_error_errno(r
, "Failed to allocate netlink message: %m");
514 r
= sd_netlink_message_append_u32(m
, IFLA_LINK
, ifi
);
516 return log_error_errno(r
, "Failed to add netlink interface index: %m");
518 n
= strappend("mv-", *i
);
522 strshorten(n
, IFNAMSIZ
-1);
524 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, n
);
526 return log_error_errno(r
, "Failed to add netlink interface name: %m");
528 r
= sd_netlink_message_append_ether_addr(m
, IFLA_ADDRESS
, &mac
);
530 return log_error_errno(r
, "Failed to add netlink MAC address: %m");
532 r
= sd_netlink_message_append_u32(m
, IFLA_NET_NS_PID
, pid
);
534 return log_error_errno(r
, "Failed to add netlink namespace field: %m");
536 r
= sd_netlink_message_open_container(m
, IFLA_LINKINFO
);
538 return log_error_errno(r
, "Failed to open netlink container: %m");
540 r
= sd_netlink_message_open_container_union(m
, IFLA_INFO_DATA
, "macvlan");
542 return log_error_errno(r
, "Failed to open netlink container: %m");
544 r
= sd_netlink_message_append_u32(m
, IFLA_MACVLAN_MODE
, MACVLAN_MODE_BRIDGE
);
546 return log_error_errno(r
, "Failed to append macvlan mode: %m");
548 r
= sd_netlink_message_close_container(m
);
550 return log_error_errno(r
, "Failed to close netlink container: %m");
552 r
= sd_netlink_message_close_container(m
);
554 return log_error_errno(r
, "Failed to close netlink container: %m");
556 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
558 return log_error_errno(r
, "Failed to add new macvlan interfaces: %m");
564 int setup_ipvlan(const char *machine_name
, pid_t pid
, char **ifaces
) {
565 _cleanup_udev_unref_
struct udev
*udev
= NULL
;
566 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
570 if (strv_isempty(ifaces
))
573 r
= sd_netlink_open(&rtnl
);
575 return log_error_errno(r
, "Failed to connect to netlink: %m");
579 log_error("Failed to connect to udev.");
583 STRV_FOREACH(i
, ifaces
) {
584 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
585 _cleanup_free_
char *n
= NULL
;
588 ifi
= parse_interface(udev
, *i
);
592 r
= sd_rtnl_message_new_link(rtnl
, &m
, RTM_NEWLINK
, 0);
594 return log_error_errno(r
, "Failed to allocate netlink message: %m");
596 r
= sd_netlink_message_append_u32(m
, IFLA_LINK
, ifi
);
598 return log_error_errno(r
, "Failed to add netlink interface index: %m");
600 n
= strappend("iv-", *i
);
604 strshorten(n
, IFNAMSIZ
-1);
606 r
= sd_netlink_message_append_string(m
, IFLA_IFNAME
, n
);
608 return log_error_errno(r
, "Failed to add netlink interface name: %m");
610 r
= sd_netlink_message_append_u32(m
, IFLA_NET_NS_PID
, pid
);
612 return log_error_errno(r
, "Failed to add netlink namespace field: %m");
614 r
= sd_netlink_message_open_container(m
, IFLA_LINKINFO
);
616 return log_error_errno(r
, "Failed to open netlink container: %m");
618 r
= sd_netlink_message_open_container_union(m
, IFLA_INFO_DATA
, "ipvlan");
620 return log_error_errno(r
, "Failed to open netlink container: %m");
622 r
= sd_netlink_message_append_u16(m
, IFLA_IPVLAN_MODE
, IPVLAN_MODE_L2
);
624 return log_error_errno(r
, "Failed to add ipvlan mode: %m");
626 r
= sd_netlink_message_close_container(m
);
628 return log_error_errno(r
, "Failed to close netlink container: %m");
630 r
= sd_netlink_message_close_container(m
);
632 return log_error_errno(r
, "Failed to close netlink container: %m");
634 r
= sd_netlink_call(rtnl
, m
, 0, NULL
);
636 return log_error_errno(r
, "Failed to add new ipvlan interfaces: %m");
642 int veth_extra_parse(char ***l
, const char *p
) {
643 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
646 r
= extract_first_word(&p
, &a
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
649 if (r
== 0 || !ifname_valid(a
))
652 r
= extract_first_word(&p
, &b
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
655 if (r
== 0 || !ifname_valid(b
)) {
665 r
= strv_push_pair(l
, a
, b
);
673 int remove_veth_links(const char *primary
, char **pairs
) {
674 _cleanup_(sd_netlink_unrefp
) sd_netlink
*rtnl
= NULL
;
678 /* In some cases the kernel might pin the veth links between host and container even after the namespace
679 * died. Hence, let's better remove them explicitly too. */
681 if (isempty(primary
) && strv_isempty(pairs
))
684 r
= sd_netlink_open(&rtnl
);
686 return log_error_errno(r
, "Failed to connect to netlink: %m");
688 remove_one_link(rtnl
, primary
);
690 STRV_FOREACH_PAIR(a
, b
, pairs
)
691 remove_one_link(rtnl
, *a
);