]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/nspawn/nspawn-network.c
pkgconfig: define variables relative to ${prefix}/${rootprefix}/${sysconfdir}
[thirdparty/systemd.git] / src / nspawn / nspawn-network.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <linux/veth.h>
4 #include <net/if.h>
5 #include <sys/file.h>
6
7 #include "sd-device.h"
8 #include "sd-id128.h"
9 #include "sd-netlink.h"
10
11 #include "alloc-util.h"
12 #include "ether-addr-util.h"
13 #include "lockfile-util.h"
14 #include "netlink-util.h"
15 #include "nspawn-network.h"
16 #include "siphash24.h"
17 #include "socket-util.h"
18 #include "stat-util.h"
19 #include "string-util.h"
20 #include "strv.h"
21 #include "util.h"
22
23 #define HOST_HASH_KEY SD_ID128_MAKE(1a,37,6f,c7,46,ec,45,0b,ad,a3,d5,31,06,60,5d,b1)
24 #define CONTAINER_HASH_KEY SD_ID128_MAKE(c3,c4,f9,19,b5,57,b2,1c,e6,cf,14,27,03,9c,ee,a2)
25 #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)
26 #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)
27 #define MACVLAN_HASH_KEY SD_ID128_MAKE(00,13,6d,bc,66,83,44,81,bb,0c,f9,51,1f,24,a6,6f)
28
29 static int remove_one_link(sd_netlink *rtnl, const char *name) {
30 _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
31 int r;
32
33 if (isempty(name))
34 return 0;
35
36 r = sd_rtnl_message_new_link(rtnl, &m, RTM_DELLINK, 0);
37 if (r < 0)
38 return log_error_errno(r, "Failed to allocate netlink message: %m");
39
40 r = sd_netlink_message_append_string(m, IFLA_IFNAME, name);
41 if (r < 0)
42 return log_error_errno(r, "Failed to add netlink interface name: %m");
43
44 r = sd_netlink_call(rtnl, m, 0, NULL);
45 if (r == -ENODEV) /* Already gone */
46 return 0;
47 if (r < 0)
48 return log_error_errno(r, "Failed to remove interface %s: %m", name);
49
50 return 1;
51 }
52
53 static int generate_mac(
54 const char *machine_name,
55 struct ether_addr *mac,
56 sd_id128_t hash_key,
57 uint64_t idx) {
58
59 uint64_t result;
60 size_t l, sz;
61 uint8_t *v, *i;
62 int r;
63
64 l = strlen(machine_name);
65 sz = sizeof(sd_id128_t) + l;
66 if (idx > 0)
67 sz += sizeof(idx);
68
69 v = alloca(sz);
70
71 /* fetch some persistent data unique to the host */
72 r = sd_id128_get_machine((sd_id128_t*) v);
73 if (r < 0)
74 return r;
75
76 /* combine with some data unique (on this host) to this
77 * container instance */
78 i = mempcpy(v + sizeof(sd_id128_t), machine_name, l);
79 if (idx > 0) {
80 idx = htole64(idx);
81 memcpy(i, &idx, sizeof(idx));
82 }
83
84 /* Let's hash the host machine ID plus the container name. We
85 * use a fixed, but originally randomly created hash key here. */
86 result = htole64(siphash24(v, sz, hash_key.bytes));
87
88 assert_cc(ETH_ALEN <= sizeof(result));
89 memcpy(mac->ether_addr_octet, &result, ETH_ALEN);
90
91 /* see eth_random_addr in the kernel */
92 mac->ether_addr_octet[0] &= 0xfe; /* clear multicast bit */
93 mac->ether_addr_octet[0] |= 0x02; /* set local assignment bit (IEEE802) */
94
95 return 0;
96 }
97
98 static int add_veth(
99 sd_netlink *rtnl,
100 pid_t pid,
101 const char *ifname_host,
102 const struct ether_addr *mac_host,
103 const char *ifname_container,
104 const struct ether_addr *mac_container) {
105
106 _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
107 int r;
108
109 assert(rtnl);
110 assert(ifname_host);
111 assert(mac_host);
112 assert(ifname_container);
113 assert(mac_container);
114
115 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
116 if (r < 0)
117 return log_error_errno(r, "Failed to allocate netlink message: %m");
118
119 r = sd_netlink_message_append_string(m, IFLA_IFNAME, ifname_host);
120 if (r < 0)
121 return log_error_errno(r, "Failed to add netlink interface name: %m");
122
123 r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, mac_host);
124 if (r < 0)
125 return log_error_errno(r, "Failed to add netlink MAC address: %m");
126
127 r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
128 if (r < 0)
129 return log_error_errno(r, "Failed to open netlink container: %m");
130
131 r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "veth");
132 if (r < 0)
133 return log_error_errno(r, "Failed to open netlink container: %m");
134
135 r = sd_netlink_message_open_container(m, VETH_INFO_PEER);
136 if (r < 0)
137 return log_error_errno(r, "Failed to open netlink container: %m");
138
139 r = sd_netlink_message_append_string(m, IFLA_IFNAME, ifname_container);
140 if (r < 0)
141 return log_error_errno(r, "Failed to add netlink interface name: %m");
142
143 r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, mac_container);
144 if (r < 0)
145 return log_error_errno(r, "Failed to add netlink MAC address: %m");
146
147 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
148 if (r < 0)
149 return log_error_errno(r, "Failed to add netlink namespace field: %m");
150
151 r = sd_netlink_message_close_container(m);
152 if (r < 0)
153 return log_error_errno(r, "Failed to close netlink container: %m");
154
155 r = sd_netlink_message_close_container(m);
156 if (r < 0)
157 return log_error_errno(r, "Failed to close netlink container: %m");
158
159 r = sd_netlink_message_close_container(m);
160 if (r < 0)
161 return log_error_errno(r, "Failed to close netlink container: %m");
162
163 r = sd_netlink_call(rtnl, m, 0, NULL);
164 if (r < 0)
165 return log_error_errno(r, "Failed to add new veth interfaces (%s:%s): %m", ifname_host, ifname_container);
166
167 return 0;
168 }
169
170 int setup_veth(const char *machine_name,
171 pid_t pid,
172 char iface_name[IFNAMSIZ],
173 bool bridge) {
174
175 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
176 struct ether_addr mac_host, mac_container;
177 int r, i;
178
179 assert(machine_name);
180 assert(pid > 0);
181 assert(iface_name);
182
183 /* Use two different interface name prefixes depending whether
184 * we are in bridge mode or not. */
185 snprintf(iface_name, IFNAMSIZ - 1, "%s-%s",
186 bridge ? "vb" : "ve", machine_name);
187
188 r = generate_mac(machine_name, &mac_container, CONTAINER_HASH_KEY, 0);
189 if (r < 0)
190 return log_error_errno(r, "Failed to generate predictable MAC address for container side: %m");
191
192 r = generate_mac(machine_name, &mac_host, HOST_HASH_KEY, 0);
193 if (r < 0)
194 return log_error_errno(r, "Failed to generate predictable MAC address for host side: %m");
195
196 r = sd_netlink_open(&rtnl);
197 if (r < 0)
198 return log_error_errno(r, "Failed to connect to netlink: %m");
199
200 r = add_veth(rtnl, pid, iface_name, &mac_host, "host0", &mac_container);
201 if (r < 0)
202 return r;
203
204 i = (int) if_nametoindex(iface_name);
205 if (i <= 0)
206 return log_error_errno(errno, "Failed to resolve interface %s: %m", iface_name);
207
208 return i;
209 }
210
211 int setup_veth_extra(
212 const char *machine_name,
213 pid_t pid,
214 char **pairs) {
215
216 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
217 uint64_t idx = 0;
218 char **a, **b;
219 int r;
220
221 assert(machine_name);
222 assert(pid > 0);
223
224 if (strv_isempty(pairs))
225 return 0;
226
227 r = sd_netlink_open(&rtnl);
228 if (r < 0)
229 return log_error_errno(r, "Failed to connect to netlink: %m");
230
231 STRV_FOREACH_PAIR(a, b, pairs) {
232 struct ether_addr mac_host, mac_container;
233
234 r = generate_mac(machine_name, &mac_container, VETH_EXTRA_CONTAINER_HASH_KEY, idx);
235 if (r < 0)
236 return log_error_errno(r, "Failed to generate predictable MAC address for container side of extra veth link: %m");
237
238 r = generate_mac(machine_name, &mac_host, VETH_EXTRA_HOST_HASH_KEY, idx);
239 if (r < 0)
240 return log_error_errno(r, "Failed to generate predictable MAC address for container side of extra veth link: %m");
241
242 r = add_veth(rtnl, pid, *a, &mac_host, *b, &mac_container);
243 if (r < 0)
244 return r;
245
246 idx++;
247 }
248
249 return 0;
250 }
251
252 static int join_bridge(sd_netlink *rtnl, const char *veth_name, const char *bridge_name) {
253 _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
254 int r, bridge_ifi;
255
256 assert(rtnl);
257 assert(veth_name);
258 assert(bridge_name);
259
260 bridge_ifi = (int) if_nametoindex(bridge_name);
261 if (bridge_ifi <= 0)
262 return -errno;
263
264 r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, 0);
265 if (r < 0)
266 return r;
267
268 r = sd_rtnl_message_link_set_flags(m, IFF_UP, IFF_UP);
269 if (r < 0)
270 return r;
271
272 r = sd_netlink_message_append_string(m, IFLA_IFNAME, veth_name);
273 if (r < 0)
274 return r;
275
276 r = sd_netlink_message_append_u32(m, IFLA_MASTER, bridge_ifi);
277 if (r < 0)
278 return r;
279
280 r = sd_netlink_call(rtnl, m, 0, NULL);
281 if (r < 0)
282 return r;
283
284 return bridge_ifi;
285 }
286
287 static int create_bridge(sd_netlink *rtnl, const char *bridge_name) {
288 _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
289 int r;
290
291 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
292 if (r < 0)
293 return r;
294
295 r = sd_netlink_message_append_string(m, IFLA_IFNAME, bridge_name);
296 if (r < 0)
297 return r;
298
299 r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
300 if (r < 0)
301 return r;
302
303 r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "bridge");
304 if (r < 0)
305 return r;
306
307 r = sd_netlink_message_close_container(m);
308 if (r < 0)
309 return r;
310
311 r = sd_netlink_message_close_container(m);
312 if (r < 0)
313 return r;
314
315 r = sd_netlink_call(rtnl, m, 0, NULL);
316 if (r < 0)
317 return r;
318
319 return 0;
320 }
321
322 int setup_bridge(const char *veth_name, const char *bridge_name, bool create) {
323 _cleanup_(release_lock_file) LockFile bridge_lock = LOCK_FILE_INIT;
324 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
325 int r, bridge_ifi;
326 unsigned n = 0;
327
328 assert(veth_name);
329 assert(bridge_name);
330
331 r = sd_netlink_open(&rtnl);
332 if (r < 0)
333 return log_error_errno(r, "Failed to connect to netlink: %m");
334
335 if (create) {
336 /* We take a system-wide lock here, so that we can safely check whether there's still a member in the
337 * bridge before removing it, without risking interference from other nspawn instances. */
338
339 r = make_lock_file("/run/systemd/nspawn-network-zone", LOCK_EX, &bridge_lock);
340 if (r < 0)
341 return log_error_errno(r, "Failed to take network zone lock: %m");
342 }
343
344 for (;;) {
345 bridge_ifi = join_bridge(rtnl, veth_name, bridge_name);
346 if (bridge_ifi >= 0)
347 return bridge_ifi;
348 if (bridge_ifi != -ENODEV || !create || n > 10)
349 return log_error_errno(bridge_ifi, "Failed to add interface %s to bridge %s: %m", veth_name, bridge_name);
350
351 /* Count attempts, so that we don't enter an endless loop here. */
352 n++;
353
354 /* The bridge doesn't exist yet. Let's create it */
355 r = create_bridge(rtnl, bridge_name);
356 if (r < 0)
357 return log_error_errno(r, "Failed to create bridge interface %s: %m", bridge_name);
358
359 /* Try again, now that the bridge exists */
360 }
361 }
362
363 int remove_bridge(const char *bridge_name) {
364 _cleanup_(release_lock_file) LockFile bridge_lock = LOCK_FILE_INIT;
365 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
366 const char *path;
367 int r;
368
369 /* Removes the specified bridge, but only if it is currently empty */
370
371 if (isempty(bridge_name))
372 return 0;
373
374 r = make_lock_file("/run/systemd/nspawn-network-zone", LOCK_EX, &bridge_lock);
375 if (r < 0)
376 return log_error_errno(r, "Failed to take network zone lock: %m");
377
378 path = strjoina("/sys/class/net/", bridge_name, "/brif");
379
380 r = dir_is_empty(path);
381 if (r == -ENOENT) /* Already gone? */
382 return 0;
383 if (r < 0)
384 return log_error_errno(r, "Can't detect if bridge %s is empty: %m", bridge_name);
385 if (r == 0) /* Still populated, leave it around */
386 return 0;
387
388 r = sd_netlink_open(&rtnl);
389 if (r < 0)
390 return log_error_errno(r, "Failed to connect to netlink: %m");
391
392 return remove_one_link(rtnl, bridge_name);
393 }
394
395 static int parse_interface(const char *name) {
396 _cleanup_(sd_device_unrefp) sd_device *d = NULL;
397 char ifi_str[2 + DECIMAL_STR_MAX(int)];
398 int ifi, r;
399
400 ifi = (int) if_nametoindex(name);
401 if (ifi <= 0)
402 return log_error_errno(errno, "Failed to resolve interface %s: %m", name);
403
404 sprintf(ifi_str, "n%i", ifi);
405 r = sd_device_new_from_device_id(&d, ifi_str);
406 if (r < 0)
407 return log_error_errno(r, "Failed to get device for interface %s: %m", name);
408
409 r = sd_device_get_is_initialized(d);
410 if (r < 0)
411 return log_error_errno(r, "Failed to determine whether interface %s is initialized or not: %m", name);
412 if (r == 0) {
413 log_error("Network interface %s is not initialized yet.", name);
414 return -EBUSY;
415 }
416
417 return ifi;
418 }
419
420 int move_network_interfaces(pid_t pid, char **ifaces) {
421 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
422 char **i;
423 int r;
424
425 if (strv_isempty(ifaces))
426 return 0;
427
428 r = sd_netlink_open(&rtnl);
429 if (r < 0)
430 return log_error_errno(r, "Failed to connect to netlink: %m");
431
432 STRV_FOREACH(i, ifaces) {
433 _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
434 int ifi;
435
436 ifi = parse_interface(*i);
437 if (ifi < 0)
438 return ifi;
439
440 r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, ifi);
441 if (r < 0)
442 return log_error_errno(r, "Failed to allocate netlink message: %m");
443
444 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
445 if (r < 0)
446 return log_error_errno(r, "Failed to append namespace PID to netlink message: %m");
447
448 r = sd_netlink_call(rtnl, m, 0, NULL);
449 if (r < 0)
450 return log_error_errno(r, "Failed to move interface %s to namespace: %m", *i);
451 }
452
453 return 0;
454 }
455
456 int setup_macvlan(const char *machine_name, pid_t pid, char **ifaces) {
457 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
458 unsigned idx = 0;
459 char **i;
460 int r;
461
462 if (strv_isempty(ifaces))
463 return 0;
464
465 r = sd_netlink_open(&rtnl);
466 if (r < 0)
467 return log_error_errno(r, "Failed to connect to netlink: %m");
468
469 STRV_FOREACH(i, ifaces) {
470 _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
471 _cleanup_free_ char *n = NULL;
472 struct ether_addr mac;
473 int ifi;
474
475 ifi = parse_interface(*i);
476 if (ifi < 0)
477 return ifi;
478
479 r = generate_mac(machine_name, &mac, MACVLAN_HASH_KEY, idx++);
480 if (r < 0)
481 return log_error_errno(r, "Failed to create MACVLAN MAC address: %m");
482
483 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
484 if (r < 0)
485 return log_error_errno(r, "Failed to allocate netlink message: %m");
486
487 r = sd_netlink_message_append_u32(m, IFLA_LINK, ifi);
488 if (r < 0)
489 return log_error_errno(r, "Failed to add netlink interface index: %m");
490
491 n = strappend("mv-", *i);
492 if (!n)
493 return log_oom();
494
495 strshorten(n, IFNAMSIZ-1);
496
497 r = sd_netlink_message_append_string(m, IFLA_IFNAME, n);
498 if (r < 0)
499 return log_error_errno(r, "Failed to add netlink interface name: %m");
500
501 r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, &mac);
502 if (r < 0)
503 return log_error_errno(r, "Failed to add netlink MAC address: %m");
504
505 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
506 if (r < 0)
507 return log_error_errno(r, "Failed to add netlink namespace field: %m");
508
509 r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
510 if (r < 0)
511 return log_error_errno(r, "Failed to open netlink container: %m");
512
513 r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "macvlan");
514 if (r < 0)
515 return log_error_errno(r, "Failed to open netlink container: %m");
516
517 r = sd_netlink_message_append_u32(m, IFLA_MACVLAN_MODE, MACVLAN_MODE_BRIDGE);
518 if (r < 0)
519 return log_error_errno(r, "Failed to append macvlan mode: %m");
520
521 r = sd_netlink_message_close_container(m);
522 if (r < 0)
523 return log_error_errno(r, "Failed to close netlink container: %m");
524
525 r = sd_netlink_message_close_container(m);
526 if (r < 0)
527 return log_error_errno(r, "Failed to close netlink container: %m");
528
529 r = sd_netlink_call(rtnl, m, 0, NULL);
530 if (r < 0)
531 return log_error_errno(r, "Failed to add new macvlan interfaces: %m");
532 }
533
534 return 0;
535 }
536
537 int setup_ipvlan(const char *machine_name, pid_t pid, char **ifaces) {
538 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
539 char **i;
540 int r;
541
542 if (strv_isempty(ifaces))
543 return 0;
544
545 r = sd_netlink_open(&rtnl);
546 if (r < 0)
547 return log_error_errno(r, "Failed to connect to netlink: %m");
548
549 STRV_FOREACH(i, ifaces) {
550 _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
551 _cleanup_free_ char *n = NULL;
552 int ifi;
553
554 ifi = parse_interface(*i);
555 if (ifi < 0)
556 return ifi;
557
558 r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
559 if (r < 0)
560 return log_error_errno(r, "Failed to allocate netlink message: %m");
561
562 r = sd_netlink_message_append_u32(m, IFLA_LINK, ifi);
563 if (r < 0)
564 return log_error_errno(r, "Failed to add netlink interface index: %m");
565
566 n = strappend("iv-", *i);
567 if (!n)
568 return log_oom();
569
570 strshorten(n, IFNAMSIZ-1);
571
572 r = sd_netlink_message_append_string(m, IFLA_IFNAME, n);
573 if (r < 0)
574 return log_error_errno(r, "Failed to add netlink interface name: %m");
575
576 r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
577 if (r < 0)
578 return log_error_errno(r, "Failed to add netlink namespace field: %m");
579
580 r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
581 if (r < 0)
582 return log_error_errno(r, "Failed to open netlink container: %m");
583
584 r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "ipvlan");
585 if (r < 0)
586 return log_error_errno(r, "Failed to open netlink container: %m");
587
588 r = sd_netlink_message_append_u16(m, IFLA_IPVLAN_MODE, IPVLAN_MODE_L2);
589 if (r < 0)
590 return log_error_errno(r, "Failed to add ipvlan mode: %m");
591
592 r = sd_netlink_message_close_container(m);
593 if (r < 0)
594 return log_error_errno(r, "Failed to close netlink container: %m");
595
596 r = sd_netlink_message_close_container(m);
597 if (r < 0)
598 return log_error_errno(r, "Failed to close netlink container: %m");
599
600 r = sd_netlink_call(rtnl, m, 0, NULL);
601 if (r < 0)
602 return log_error_errno(r, "Failed to add new ipvlan interfaces: %m");
603 }
604
605 return 0;
606 }
607
608 int veth_extra_parse(char ***l, const char *p) {
609 _cleanup_free_ char *a = NULL, *b = NULL;
610 int r;
611
612 r = extract_first_word(&p, &a, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
613 if (r < 0)
614 return r;
615 if (r == 0 || !ifname_valid(a))
616 return -EINVAL;
617
618 r = extract_first_word(&p, &b, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
619 if (r < 0)
620 return r;
621 if (r == 0 || !ifname_valid(b)) {
622 free(b);
623 b = strdup(a);
624 if (!b)
625 return -ENOMEM;
626 }
627
628 if (p)
629 return -EINVAL;
630
631 r = strv_push_pair(l, a, b);
632 if (r < 0)
633 return -ENOMEM;
634
635 a = b = NULL;
636 return 0;
637 }
638
639 int remove_veth_links(const char *primary, char **pairs) {
640 _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
641 char **a, **b;
642 int r;
643
644 /* In some cases the kernel might pin the veth links between host and container even after the namespace
645 * died. Hence, let's better remove them explicitly too. */
646
647 if (isempty(primary) && strv_isempty(pairs))
648 return 0;
649
650 r = sd_netlink_open(&rtnl);
651 if (r < 0)
652 return log_error_errno(r, "Failed to connect to netlink: %m");
653
654 remove_one_link(rtnl, primary);
655
656 STRV_FOREACH_PAIR(a, b, pairs)
657 remove_one_link(rtnl, *a);
658
659 return 0;
660 }