]>
Commit | Line | Data |
---|---|---|
718e3744 | 1 | /* Kernel communication using netlink interface. |
2 | * Copyright (C) 1999 Kunihiro Ishiguro | |
3 | * | |
4 | * This file is part of GNU Zebra. | |
5 | * | |
6 | * GNU Zebra is free software; you can redistribute it and/or modify it | |
7 | * under the terms of the GNU General Public License as published by the | |
8 | * Free Software Foundation; either version 2, or (at your option) any | |
9 | * later version. | |
10 | * | |
11 | * GNU Zebra 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 | * General Public License for more details. | |
15 | * | |
896014f4 DL |
16 | * You should have received a copy of the GNU General Public License along |
17 | * with this program; see the file COPYING; if not, write to the Free Software | |
18 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
718e3744 | 19 | */ |
1fdc9eae | 20 | |
21 | #include <zebra.h> | |
22 | ||
ddfeb486 DL |
23 | #ifdef HAVE_NETLINK |
24 | ||
1fdc9eae | 25 | #include "linklist.h" |
26 | #include "if.h" | |
27 | #include "log.h" | |
28 | #include "prefix.h" | |
29 | #include "connected.h" | |
30 | #include "table.h" | |
31 | #include "memory.h" | |
1fdc9eae | 32 | #include "rib.h" |
33 | #include "thread.h" | |
34 | #include "privs.h" | |
35 | #include "nexthop.h" | |
36 | #include "vrf.h" | |
37 | #include "mpls.h" | |
174482ef | 38 | #include "lib_errors.h" |
d4000d7b | 39 | #include "hash.h" |
1fdc9eae | 40 | |
3801e764 | 41 | #include "zebra/zebra_router.h" |
1fdc9eae | 42 | #include "zebra/zebra_ns.h" |
43 | #include "zebra/zebra_vrf.h" | |
05f7f5db | 44 | #include "zebra/rt.h" |
1fdc9eae | 45 | #include "zebra/debug.h" |
46 | #include "zebra/kernel_netlink.h" | |
47 | #include "zebra/rt_netlink.h" | |
48 | #include "zebra/if_netlink.h" | |
942bf97b | 49 | #include "zebra/rule_netlink.h" |
ebb61fca | 50 | #include "zebra/netconf_netlink.h" |
43e52561 | 51 | #include "zebra/zebra_errors.h" |
1fdc9eae | 52 | |
53 | #ifndef SO_RCVBUFFORCE | |
54 | #define SO_RCVBUFFORCE (33) | |
55 | #endif | |
56 | ||
57 | /* Hack for GNU libc version 2. */ | |
58 | #ifndef MSG_TRUNC | |
59 | #define MSG_TRUNC 0x20 | |
60 | #endif /* MSG_TRUNC */ | |
61 | ||
62 | #ifndef NLMSG_TAIL | |
d62a17ae | 63 | #define NLMSG_TAIL(nmsg) \ |
d7c0a89a QY |
64 | ((struct rtattr *)(((uint8_t *)(nmsg)) \ |
65 | + NLMSG_ALIGN((nmsg)->nlmsg_len))) | |
1fdc9eae | 66 | #endif |
67 | ||
68 | #ifndef RTA_TAIL | |
d62a17ae | 69 | #define RTA_TAIL(rta) \ |
d7c0a89a | 70 | ((struct rtattr *)(((uint8_t *)(rta)) + RTA_ALIGN((rta)->rta_len))) |
1fdc9eae | 71 | #endif |
72 | ||
f909c673 DS |
73 | #ifndef RTNL_FAMILY_IP6MR |
74 | #define RTNL_FAMILY_IP6MR 129 | |
75 | #endif | |
76 | ||
77 | #ifndef RTPROT_MROUTED | |
78 | #define RTPROT_MROUTED 17 | |
79 | #endif | |
80 | ||
531c92b8 | 81 | #define NL_DEFAULT_BATCH_BUFSIZE (16 * NL_PKT_BUF_SIZE) |
e63c7622 JU |
82 | |
83 | /* | |
84 | * We limit the batch's size to a number smaller than the length of the | |
85 | * underlying buffer since the last message that wouldn't fit the batch would go | |
86 | * over the upper boundary and then it would have to be encoded again into a new | |
87 | * buffer. If the difference between the limit and the length of the buffer is | |
88 | * big enough (bigger than the biggest Netlink message) then this situation | |
89 | * won't occur. | |
90 | */ | |
531c92b8 JU |
91 | #define NL_DEFAULT_BATCH_SEND_THRESHOLD (15 * NL_PKT_BUF_SIZE) |
92 | ||
d62a17ae | 93 | static const struct message nlmsg_str[] = {{RTM_NEWROUTE, "RTM_NEWROUTE"}, |
94 | {RTM_DELROUTE, "RTM_DELROUTE"}, | |
95 | {RTM_GETROUTE, "RTM_GETROUTE"}, | |
96 | {RTM_NEWLINK, "RTM_NEWLINK"}, | |
5d414138 | 97 | {RTM_SETLINK, "RTM_SETLINK"}, |
d62a17ae | 98 | {RTM_DELLINK, "RTM_DELLINK"}, |
99 | {RTM_GETLINK, "RTM_GETLINK"}, | |
100 | {RTM_NEWADDR, "RTM_NEWADDR"}, | |
101 | {RTM_DELADDR, "RTM_DELADDR"}, | |
102 | {RTM_GETADDR, "RTM_GETADDR"}, | |
103 | {RTM_NEWNEIGH, "RTM_NEWNEIGH"}, | |
104 | {RTM_DELNEIGH, "RTM_DELNEIGH"}, | |
105 | {RTM_GETNEIGH, "RTM_GETNEIGH"}, | |
942bf97b | 106 | {RTM_NEWRULE, "RTM_NEWRULE"}, |
107 | {RTM_DELRULE, "RTM_DELRULE"}, | |
108 | {RTM_GETRULE, "RTM_GETRULE"}, | |
79580b5a SW |
109 | {RTM_NEWNEXTHOP, "RTM_NEWNEXTHOP"}, |
110 | {RTM_DELNEXTHOP, "RTM_DELNEXTHOP"}, | |
111 | {RTM_GETNEXTHOP, "RTM_GETNEXTHOP"}, | |
ebb61fca DS |
112 | {RTM_NEWNETCONF, "RTM_NEWNETCONF"}, |
113 | {RTM_DELNETCONF, "RTM_DELNETCONF"}, | |
d62a17ae | 114 | {0}}; |
1fdc9eae | 115 | |
116 | static const struct message rtproto_str[] = { | |
d62a17ae | 117 | {RTPROT_REDIRECT, "redirect"}, |
118 | {RTPROT_KERNEL, "kernel"}, | |
119 | {RTPROT_BOOT, "boot"}, | |
120 | {RTPROT_STATIC, "static"}, | |
121 | {RTPROT_GATED, "GateD"}, | |
122 | {RTPROT_RA, "router advertisement"}, | |
123 | {RTPROT_MRT, "MRT"}, | |
124 | {RTPROT_ZEBRA, "Zebra"}, | |
1fdc9eae | 125 | #ifdef RTPROT_BIRD |
d62a17ae | 126 | {RTPROT_BIRD, "BIRD"}, |
1fdc9eae | 127 | #endif /* RTPROT_BIRD */ |
d62a17ae | 128 | {RTPROT_MROUTED, "mroute"}, |
129 | {RTPROT_BGP, "BGP"}, | |
130 | {RTPROT_OSPF, "OSPF"}, | |
131 | {RTPROT_ISIS, "IS-IS"}, | |
132 | {RTPROT_RIP, "RIP"}, | |
133 | {RTPROT_RIPNG, "RIPNG"}, | |
d4d71f11 | 134 | {RTPROT_ZSTATIC, "static"}, |
d62a17ae | 135 | {0}}; |
136 | ||
137 | static const struct message family_str[] = {{AF_INET, "ipv4"}, | |
138 | {AF_INET6, "ipv6"}, | |
139 | {AF_BRIDGE, "bridge"}, | |
140 | {RTNL_FAMILY_IPMR, "ipv4MR"}, | |
141 | {RTNL_FAMILY_IP6MR, "ipv6MR"}, | |
142 | {0}}; | |
143 | ||
8c8f250b DS |
144 | static const struct message rttype_str[] = {{RTN_UNSPEC, "none"}, |
145 | {RTN_UNICAST, "unicast"}, | |
146 | {RTN_LOCAL, "local"}, | |
147 | {RTN_BROADCAST, "broadcast"}, | |
148 | {RTN_ANYCAST, "anycast"}, | |
d62a17ae | 149 | {RTN_MULTICAST, "multicast"}, |
8c8f250b DS |
150 | {RTN_BLACKHOLE, "blackhole"}, |
151 | {RTN_UNREACHABLE, "unreachable"}, | |
152 | {RTN_PROHIBIT, "prohibited"}, | |
153 | {RTN_THROW, "throw"}, | |
154 | {RTN_NAT, "nat"}, | |
155 | {RTN_XRESOLVE, "resolver"}, | |
d62a17ae | 156 | {0}}; |
b339bde7 | 157 | |
1fdc9eae | 158 | extern struct thread_master *master; |
1fdc9eae | 159 | |
160 | extern struct zebra_privs_t zserv_privs; | |
161 | ||
bf8d3d6a | 162 | DEFINE_MTYPE_STATIC(ZEBRA, NL_BUF, "Zebra Netlink buffers"); |
531c92b8 | 163 | |
34869809 MS |
164 | /* Hashtable and mutex to allow lookup of nlsock structs by socket/fd value. |
165 | * We have both the main and dplane pthreads using these structs, so we have | |
166 | * to protect the hash with a lock. | |
167 | */ | |
168 | static struct hash *nlsock_hash; | |
169 | pthread_mutex_t nlsock_mutex; | |
170 | ||
171 | /* Lock and unlock wrappers for nlsock hash */ | |
172 | #define NLSOCK_LOCK() pthread_mutex_lock(&nlsock_mutex) | |
173 | #define NLSOCK_UNLOCK() pthread_mutex_unlock(&nlsock_mutex) | |
174 | ||
531c92b8 JU |
175 | size_t nl_batch_tx_bufsize; |
176 | char *nl_batch_tx_buf; | |
177 | ||
531c92b8 JU |
178 | _Atomic uint32_t nl_batch_bufsize = NL_DEFAULT_BATCH_BUFSIZE; |
179 | _Atomic uint32_t nl_batch_send_threshold = NL_DEFAULT_BATCH_SEND_THRESHOLD; | |
180 | ||
e63c7622 JU |
181 | struct nl_batch { |
182 | void *buf; | |
183 | size_t bufsiz; | |
184 | size_t limit; | |
185 | ||
186 | void *buf_head; | |
187 | size_t curlen; | |
188 | size_t msgcnt; | |
189 | ||
190 | const struct zebra_dplane_info *zns; | |
e63c7622 | 191 | |
f6feb48b | 192 | struct dplane_ctx_q ctx_list; |
e63c7622 | 193 | |
f6feb48b JU |
194 | /* |
195 | * Pointer to the queue of completed contexts outbound back | |
196 | * towards the dataplane module. | |
197 | */ | |
198 | struct dplane_ctx_q *ctx_out_q; | |
e63c7622 JU |
199 | }; |
200 | ||
531c92b8 JU |
201 | int netlink_config_write_helper(struct vty *vty) |
202 | { | |
203 | uint32_t size = | |
204 | atomic_load_explicit(&nl_batch_bufsize, memory_order_relaxed); | |
205 | uint32_t threshold = atomic_load_explicit(&nl_batch_send_threshold, | |
206 | memory_order_relaxed); | |
207 | ||
208 | if (size != NL_DEFAULT_BATCH_BUFSIZE | |
209 | || threshold != NL_DEFAULT_BATCH_SEND_THRESHOLD) | |
210 | vty_out(vty, "zebra kernel netlink batch-tx-buf %u %u\n", size, | |
211 | threshold); | |
212 | ||
c40e1b1c SW |
213 | if (if_netlink_frr_protodown_r_bit_is_set()) |
214 | vty_out(vty, "zebra protodown reason-bit %u\n", | |
215 | if_netlink_get_frr_protodown_r_bit()); | |
216 | ||
531c92b8 JU |
217 | return 0; |
218 | } | |
219 | ||
220 | void netlink_set_batch_buffer_size(uint32_t size, uint32_t threshold, bool set) | |
221 | { | |
222 | if (!set) { | |
223 | size = NL_DEFAULT_BATCH_BUFSIZE; | |
224 | threshold = NL_DEFAULT_BATCH_SEND_THRESHOLD; | |
225 | } | |
226 | ||
227 | atomic_store_explicit(&nl_batch_bufsize, size, memory_order_relaxed); | |
228 | atomic_store_explicit(&nl_batch_send_threshold, threshold, | |
229 | memory_order_relaxed); | |
230 | } | |
231 | ||
2414abd3 | 232 | int netlink_talk_filter(struct nlmsghdr *h, ns_id_t ns_id, int startup) |
1fdc9eae | 233 | { |
3575d9e8 DS |
234 | /* |
235 | * This is an error condition that must be handled during | |
236 | * development. | |
237 | * | |
238 | * The netlink_talk_filter function is used for communication | |
239 | * down the netlink_cmd pipe and we are expecting | |
240 | * an ack being received. So if we get here | |
241 | * then we did not receive the ack and instead | |
242 | * received some other message in an unexpected | |
243 | * way. | |
244 | */ | |
43e52561 QY |
245 | zlog_debug("%s: ignoring message type 0x%04x(%s) NS %u", __func__, |
246 | h->nlmsg_type, nl_msg_type_to_str(h->nlmsg_type), ns_id); | |
d62a17ae | 247 | return 0; |
1fdc9eae | 248 | } |
249 | ||
d62a17ae | 250 | static int netlink_recvbuf(struct nlsock *nl, uint32_t newsize) |
1fdc9eae | 251 | { |
d7c0a89a | 252 | uint32_t oldsize; |
d62a17ae | 253 | socklen_t newlen = sizeof(newsize); |
254 | socklen_t oldlen = sizeof(oldsize); | |
255 | int ret; | |
256 | ||
257 | ret = getsockopt(nl->sock, SOL_SOCKET, SO_RCVBUF, &oldsize, &oldlen); | |
258 | if (ret < 0) { | |
450971aa | 259 | flog_err_sys(EC_LIB_SOCKET, |
09c866e3 QY |
260 | "Can't get %s receive buffer size: %s", nl->name, |
261 | safe_strerror(errno)); | |
d62a17ae | 262 | return -1; |
263 | } | |
264 | ||
265 | /* Try force option (linux >= 2.6.14) and fall back to normal set */ | |
0cf6db21 | 266 | frr_with_privs(&zserv_privs) { |
01b9e3fd | 267 | ret = setsockopt(nl->sock, SOL_SOCKET, SO_RCVBUFFORCE, |
9fb83b55 | 268 | &rcvbufsize, sizeof(rcvbufsize)); |
01b9e3fd | 269 | } |
d62a17ae | 270 | if (ret < 0) |
9fb83b55 DS |
271 | ret = setsockopt(nl->sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsize, |
272 | sizeof(rcvbufsize)); | |
d62a17ae | 273 | if (ret < 0) { |
450971aa | 274 | flog_err_sys(EC_LIB_SOCKET, |
09c866e3 QY |
275 | "Can't set %s receive buffer size: %s", nl->name, |
276 | safe_strerror(errno)); | |
d62a17ae | 277 | return -1; |
278 | } | |
279 | ||
280 | ret = getsockopt(nl->sock, SOL_SOCKET, SO_RCVBUF, &newsize, &newlen); | |
281 | if (ret < 0) { | |
450971aa | 282 | flog_err_sys(EC_LIB_SOCKET, |
09c866e3 QY |
283 | "Can't get %s receive buffer size: %s", nl->name, |
284 | safe_strerror(errno)); | |
d62a17ae | 285 | return -1; |
286 | } | |
d62a17ae | 287 | return 0; |
1fdc9eae | 288 | } |
289 | ||
290 | /* Make socket for Linux netlink interface. */ | |
d62a17ae | 291 | static int netlink_socket(struct nlsock *nl, unsigned long groups, |
292 | ns_id_t ns_id) | |
1fdc9eae | 293 | { |
d62a17ae | 294 | int ret; |
295 | struct sockaddr_nl snl; | |
296 | int sock; | |
297 | int namelen; | |
d62a17ae | 298 | |
0cf6db21 | 299 | frr_with_privs(&zserv_privs) { |
6bb30c2c DL |
300 | sock = ns_socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE, ns_id); |
301 | if (sock < 0) { | |
302 | zlog_err("Can't open %s socket: %s", nl->name, | |
303 | safe_strerror(errno)); | |
304 | return -1; | |
305 | } | |
d62a17ae | 306 | |
0d6f7fd6 | 307 | memset(&snl, 0, sizeof(snl)); |
6bb30c2c DL |
308 | snl.nl_family = AF_NETLINK; |
309 | snl.nl_groups = groups; | |
d62a17ae | 310 | |
6bb30c2c | 311 | /* Bind the socket to the netlink structure for anything. */ |
0d6f7fd6 | 312 | ret = bind(sock, (struct sockaddr *)&snl, sizeof(snl)); |
6bb30c2c | 313 | } |
d62a17ae | 314 | |
315 | if (ret < 0) { | |
6bb30c2c DL |
316 | zlog_err("Can't bind %s socket to group 0x%x: %s", nl->name, |
317 | snl.nl_groups, safe_strerror(errno)); | |
d62a17ae | 318 | close(sock); |
319 | return -1; | |
320 | } | |
321 | ||
322 | /* multiple netlink sockets will have different nl_pid */ | |
0d6f7fd6 | 323 | namelen = sizeof(snl); |
d62a17ae | 324 | ret = getsockname(sock, (struct sockaddr *)&snl, (socklen_t *)&namelen); |
0d6f7fd6 | 325 | if (ret < 0 || namelen != sizeof(snl)) { |
450971aa | 326 | flog_err_sys(EC_LIB_SOCKET, "Can't get %s socket name: %s", |
09c866e3 | 327 | nl->name, safe_strerror(errno)); |
d62a17ae | 328 | close(sock); |
329 | return -1; | |
330 | } | |
331 | ||
332 | nl->snl = snl; | |
333 | nl->sock = sock; | |
2cf7651f DS |
334 | nl->buflen = NL_RCV_PKT_BUF_SIZE; |
335 | nl->buf = XMALLOC(MTYPE_NL_BUF, nl->buflen); | |
336 | ||
d62a17ae | 337 | return ret; |
1fdc9eae | 338 | } |
339 | ||
d166308b MS |
340 | /* |
341 | * Dispatch an incoming netlink message; used by the zebra main pthread's | |
342 | * netlink event reader. | |
343 | */ | |
2414abd3 | 344 | static int netlink_information_fetch(struct nlmsghdr *h, ns_id_t ns_id, |
d62a17ae | 345 | int startup) |
1fdc9eae | 346 | { |
3575d9e8 DS |
347 | /* |
348 | * When we handle new message types here | |
349 | * because we are starting to install them | |
350 | * then lets check the netlink_install_filter | |
351 | * and see if we should add the corresponding | |
352 | * allow through entry there. | |
353 | * Probably not needed to do but please | |
354 | * think about it. | |
355 | */ | |
d62a17ae | 356 | switch (h->nlmsg_type) { |
357 | case RTM_NEWROUTE: | |
2414abd3 | 358 | return netlink_route_change(h, ns_id, startup); |
d62a17ae | 359 | case RTM_DELROUTE: |
2414abd3 | 360 | return netlink_route_change(h, ns_id, startup); |
d62a17ae | 361 | case RTM_NEWLINK: |
2414abd3 | 362 | return netlink_link_change(h, ns_id, startup); |
d62a17ae | 363 | case RTM_DELLINK: |
2414abd3 | 364 | return netlink_link_change(h, ns_id, startup); |
d62a17ae | 365 | case RTM_NEWNEIGH: |
d62a17ae | 366 | case RTM_DELNEIGH: |
951f8bcb | 367 | case RTM_GETNEIGH: |
7a52f27e | 368 | return netlink_neigh_change(h, ns_id); |
942bf97b | 369 | case RTM_NEWRULE: |
2414abd3 | 370 | return netlink_rule_change(h, ns_id, startup); |
942bf97b | 371 | case RTM_DELRULE: |
2414abd3 | 372 | return netlink_rule_change(h, ns_id, startup); |
79580b5a | 373 | case RTM_NEWNEXTHOP: |
d9f5b2f5 | 374 | return netlink_nexthop_change(h, ns_id, startup); |
79580b5a | 375 | case RTM_DELNEXTHOP: |
d9f5b2f5 | 376 | return netlink_nexthop_change(h, ns_id, startup); |
d166308b MS |
377 | |
378 | /* Messages handled in the dplane thread */ | |
379 | case RTM_NEWADDR: | |
380 | case RTM_DELADDR: | |
cd787a8a MS |
381 | case RTM_NEWNETCONF: |
382 | case RTM_DELNETCONF: | |
d166308b MS |
383 | return 0; |
384 | ||
d62a17ae | 385 | default: |
3575d9e8 DS |
386 | /* |
387 | * If we have received this message then | |
388 | * we have made a mistake during development | |
389 | * and we need to write some code to handle | |
390 | * this message type or not ask for | |
391 | * it to be sent up to us | |
392 | */ | |
e914ccbe | 393 | flog_err(EC_ZEBRA_UNKNOWN_NLMSG, |
1d5453d6 | 394 | "Unknown netlink nlmsg_type %s(%d) vrf %u", |
1c50c1c0 QY |
395 | nl_msg_type_to_str(h->nlmsg_type), h->nlmsg_type, |
396 | ns_id); | |
d62a17ae | 397 | break; |
398 | } | |
399 | return 0; | |
1fdc9eae | 400 | } |
401 | ||
d166308b MS |
402 | /* |
403 | * Dispatch an incoming netlink message; used by the dataplane pthread's | |
404 | * netlink event reader code. | |
405 | */ | |
406 | static int dplane_netlink_information_fetch(struct nlmsghdr *h, ns_id_t ns_id, | |
407 | int startup) | |
408 | { | |
409 | /* | |
410 | * Dispatch the incoming messages that the dplane pthread handles | |
411 | */ | |
412 | switch (h->nlmsg_type) { | |
413 | case RTM_NEWADDR: | |
414 | case RTM_DELADDR: | |
415 | return netlink_interface_addr_dplane(h, ns_id, startup); | |
416 | ||
9f3f1486 MS |
417 | case RTM_NEWNETCONF: |
418 | case RTM_DELNETCONF: | |
419 | return netlink_netconf_change(h, ns_id, startup); | |
420 | ||
421 | /* TODO -- other messages for the dplane socket and pthread */ | |
422 | ||
d166308b MS |
423 | case RTM_NEWLINK: |
424 | case RTM_DELLINK: | |
425 | ||
426 | default: | |
427 | break; | |
428 | } | |
429 | ||
430 | return 0; | |
431 | } | |
432 | ||
cc9f21da | 433 | static void kernel_read(struct thread *thread) |
1fdc9eae | 434 | { |
d62a17ae | 435 | struct zebra_ns *zns = (struct zebra_ns *)THREAD_ARG(thread); |
85a75f1e MS |
436 | struct zebra_dplane_info dp_info; |
437 | ||
438 | /* Capture key info from ns struct */ | |
439 | zebra_dplane_info_from_zns(&dp_info, zns, false); | |
440 | ||
441 | netlink_parse_info(netlink_information_fetch, &zns->netlink, &dp_info, | |
9bfadae8 | 442 | 5, false); |
d166308b | 443 | |
3801e764 | 444 | thread_add_read(zrouter.master, kernel_read, zns, zns->netlink.sock, |
d62a17ae | 445 | &zns->t_netlink); |
1fdc9eae | 446 | } |
447 | ||
d166308b MS |
448 | /* |
449 | * Called by the dplane pthread to read incoming OS messages and dispatch them. | |
450 | */ | |
451 | int kernel_dplane_read(struct zebra_dplane_info *info) | |
452 | { | |
d4000d7b DS |
453 | struct nlsock *nl = kernel_netlink_nlsock_lookup(info->sock); |
454 | ||
455 | netlink_parse_info(dplane_netlink_information_fetch, nl, info, 5, | |
456 | false); | |
d166308b MS |
457 | |
458 | return 0; | |
459 | } | |
460 | ||
3575d9e8 DS |
461 | /* |
462 | * Filter out messages from self that occur on listener socket, | |
62b8bb7a | 463 | * caused by our actions on the command socket(s) |
3575d9e8 DS |
464 | * |
465 | * When we add new Netlink message types we probably | |
466 | * do not need to add them here as that we are filtering | |
467 | * on the routes we actually care to receive( which is rarer | |
468 | * then the normal course of operations). We are intentionally | |
469 | * allowing some messages from ourselves through | |
470 | * ( I'm looking at you Interface based netlink messages ) | |
9f3f1486 MS |
471 | * so that we only have to write one way to handle incoming |
472 | * address add/delete and xxxNETCONF changes. | |
1fdc9eae | 473 | */ |
ff45112c | 474 | static void netlink_install_filter(int sock, uint32_t pid, uint32_t dplane_pid) |
1fdc9eae | 475 | { |
3575d9e8 DS |
476 | /* |
477 | * BPF_JUMP instructions and where you jump to are based upon | |
478 | * 0 as being the next statement. So count from 0. Writing | |
479 | * this down because every time I look at this I have to | |
480 | * re-remember it. | |
481 | */ | |
d62a17ae | 482 | struct sock_filter filter[] = { |
3575d9e8 DS |
483 | /* |
484 | * Logic: | |
62b8bb7a MS |
485 | * if (nlmsg_pid == pid || |
486 | * nlmsg_pid == dplane_pid) { | |
3575d9e8 | 487 | * if (the incoming nlmsg_type == |
9f3f1486 MS |
488 | * RTM_NEWADDR || RTM_DELADDR || RTM_NEWNETCONF || |
489 | * RTM_DELNETCONF) | |
3575d9e8 DS |
490 | * keep this message |
491 | * else | |
492 | * skip this message | |
493 | * } else | |
494 | * keep this netlink message | |
495 | */ | |
496 | /* | |
497 | * 0: Load the nlmsg_pid into the BPF register | |
498 | */ | |
d62a17ae | 499 | BPF_STMT(BPF_LD | BPF_ABS | BPF_W, |
500 | offsetof(struct nlmsghdr, nlmsg_pid)), | |
3575d9e8 DS |
501 | /* |
502 | * 1: Compare to pid | |
503 | */ | |
62b8bb7a | 504 | BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htonl(pid), 1, 0), |
3575d9e8 | 505 | /* |
62b8bb7a MS |
506 | * 2: Compare to dplane pid |
507 | */ | |
9f3f1486 | 508 | BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htonl(dplane_pid), 0, 6), |
62b8bb7a MS |
509 | /* |
510 | * 3: Load the nlmsg_type into BPF register | |
3575d9e8 DS |
511 | */ |
512 | BPF_STMT(BPF_LD | BPF_ABS | BPF_H, | |
513 | offsetof(struct nlmsghdr, nlmsg_type)), | |
514 | /* | |
62b8bb7a | 515 | * 4: Compare to RTM_NEWADDR |
3575d9e8 | 516 | */ |
9f3f1486 | 517 | BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htons(RTM_NEWADDR), 4, 0), |
3575d9e8 | 518 | /* |
62b8bb7a | 519 | * 5: Compare to RTM_DELADDR |
3575d9e8 | 520 | */ |
9f3f1486 MS |
521 | BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htons(RTM_DELADDR), 3, 0), |
522 | /* | |
523 | * 6: Compare to RTM_NEWNETCONF | |
524 | */ | |
525 | BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htons(RTM_NEWNETCONF), 2, | |
526 | 0), | |
527 | /* | |
528 | * 7: Compare to RTM_DELNETCONF | |
529 | */ | |
530 | BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, htons(RTM_DELNETCONF), 1, | |
531 | 0), | |
3575d9e8 | 532 | /* |
9f3f1486 | 533 | * 8: This is the end state of we want to skip the |
3575d9e8 DS |
534 | * message |
535 | */ | |
d62a17ae | 536 | BPF_STMT(BPF_RET | BPF_K, 0), |
9f3f1486 | 537 | /* 9: This is the end state of we want to keep |
3575d9e8 DS |
538 | * the message |
539 | */ | |
d62a17ae | 540 | BPF_STMT(BPF_RET | BPF_K, 0xffff), |
541 | }; | |
542 | ||
543 | struct sock_fprog prog = { | |
9d303b37 | 544 | .len = array_size(filter), .filter = filter, |
d62a17ae | 545 | }; |
546 | ||
547 | if (setsockopt(sock, SOL_SOCKET, SO_ATTACH_FILTER, &prog, sizeof(prog)) | |
548 | < 0) | |
1d5453d6 | 549 | flog_err_sys(EC_LIB_SOCKET, "Can't install socket filter: %s", |
9df414fe | 550 | safe_strerror(errno)); |
1fdc9eae | 551 | } |
552 | ||
d166308b MS |
553 | void netlink_parse_rtattr_flags(struct rtattr **tb, int max, struct rtattr *rta, |
554 | int len, unsigned short flags) | |
4bcdb608 NA |
555 | { |
556 | unsigned short type; | |
557 | ||
269b69d7 | 558 | memset(tb, 0, sizeof(struct rtattr *) * (max + 1)); |
4bcdb608 NA |
559 | while (RTA_OK(rta, len)) { |
560 | type = rta->rta_type & ~flags; | |
561 | if ((type <= max) && (!tb[type])) | |
562 | tb[type] = rta; | |
563 | rta = RTA_NEXT(rta, len); | |
564 | } | |
565 | } | |
566 | ||
d62a17ae | 567 | void netlink_parse_rtattr(struct rtattr **tb, int max, struct rtattr *rta, |
568 | int len) | |
1fdc9eae | 569 | { |
269b69d7 | 570 | memset(tb, 0, sizeof(struct rtattr *) * (max + 1)); |
d62a17ae | 571 | while (RTA_OK(rta, len)) { |
572 | if (rta->rta_type <= max) | |
573 | tb[rta->rta_type] = rta; | |
574 | rta = RTA_NEXT(rta, len); | |
575 | } | |
1fdc9eae | 576 | } |
577 | ||
87da6a60 SW |
578 | /** |
579 | * netlink_parse_rtattr_nested() - Parses a nested route attribute | |
580 | * @tb: Pointer to array for storing rtattr in. | |
581 | * @max: Max number to store. | |
582 | * @rta: Pointer to rtattr to look for nested items in. | |
583 | */ | |
584 | void netlink_parse_rtattr_nested(struct rtattr **tb, int max, | |
585 | struct rtattr *rta) | |
586 | { | |
587 | netlink_parse_rtattr(tb, max, RTA_DATA(rta), RTA_PAYLOAD(rta)); | |
588 | } | |
589 | ||
312a6bee JU |
590 | bool nl_attr_put(struct nlmsghdr *n, unsigned int maxlen, int type, |
591 | const void *data, unsigned int alen) | |
1fdc9eae | 592 | { |
d62a17ae | 593 | int len; |
594 | struct rtattr *rta; | |
1fdc9eae | 595 | |
d62a17ae | 596 | len = RTA_LENGTH(alen); |
1fdc9eae | 597 | |
d62a17ae | 598 | if (NLMSG_ALIGN(n->nlmsg_len) + RTA_ALIGN(len) > maxlen) |
312a6bee | 599 | return false; |
1fdc9eae | 600 | |
d62a17ae | 601 | rta = (struct rtattr *)(((char *)n) + NLMSG_ALIGN(n->nlmsg_len)); |
602 | rta->rta_type = type; | |
603 | rta->rta_len = len; | |
4b2792b5 | 604 | |
d62a17ae | 605 | if (data) |
606 | memcpy(RTA_DATA(rta), data, alen); | |
607 | else | |
608 | assert(alen == 0); | |
4b2792b5 | 609 | |
d62a17ae | 610 | n->nlmsg_len = NLMSG_ALIGN(n->nlmsg_len) + RTA_ALIGN(len); |
1fdc9eae | 611 | |
312a6bee | 612 | return true; |
1fdc9eae | 613 | } |
614 | ||
94d70a65 DS |
615 | bool nl_attr_put8(struct nlmsghdr *n, unsigned int maxlen, int type, |
616 | uint8_t data) | |
617 | { | |
618 | return nl_attr_put(n, maxlen, type, &data, sizeof(uint8_t)); | |
619 | } | |
620 | ||
312a6bee JU |
621 | bool nl_attr_put16(struct nlmsghdr *n, unsigned int maxlen, int type, |
622 | uint16_t data) | |
bbc16902 | 623 | { |
312a6bee | 624 | return nl_attr_put(n, maxlen, type, &data, sizeof(uint16_t)); |
bbc16902 | 625 | } |
626 | ||
312a6bee JU |
627 | bool nl_attr_put32(struct nlmsghdr *n, unsigned int maxlen, int type, |
628 | uint32_t data) | |
1fdc9eae | 629 | { |
312a6bee | 630 | return nl_attr_put(n, maxlen, type, &data, sizeof(uint32_t)); |
1fdc9eae | 631 | } |
632 | ||
312a6bee | 633 | struct rtattr *nl_attr_nest(struct nlmsghdr *n, unsigned int maxlen, int type) |
1fdc9eae | 634 | { |
d62a17ae | 635 | struct rtattr *nest = NLMSG_TAIL(n); |
1fdc9eae | 636 | |
312a6bee JU |
637 | if (!nl_attr_put(n, maxlen, type, NULL, 0)) |
638 | return NULL; | |
639 | ||
40d86eba | 640 | nest->rta_type |= NLA_F_NESTED; |
d62a17ae | 641 | return nest; |
1fdc9eae | 642 | } |
643 | ||
312a6bee | 644 | int nl_attr_nest_end(struct nlmsghdr *n, struct rtattr *nest) |
1fdc9eae | 645 | { |
d7c0a89a | 646 | nest->rta_len = (uint8_t *)NLMSG_TAIL(n) - (uint8_t *)nest; |
d62a17ae | 647 | return n->nlmsg_len; |
1fdc9eae | 648 | } |
649 | ||
312a6bee | 650 | struct rtnexthop *nl_attr_rtnh(struct nlmsghdr *n, unsigned int maxlen) |
1fdc9eae | 651 | { |
312a6bee | 652 | struct rtnexthop *rtnh = (struct rtnexthop *)NLMSG_TAIL(n); |
1fdc9eae | 653 | |
312a6bee JU |
654 | if (NLMSG_ALIGN(n->nlmsg_len) + RTNH_ALIGN(sizeof(struct rtnexthop)) |
655 | > maxlen) | |
656 | return NULL; | |
657 | ||
658 | memset(rtnh, 0, sizeof(struct rtnexthop)); | |
659 | n->nlmsg_len = | |
660 | NLMSG_ALIGN(n->nlmsg_len) + RTA_ALIGN(sizeof(struct rtnexthop)); | |
661 | ||
662 | return rtnh; | |
1fdc9eae | 663 | } |
664 | ||
312a6bee | 665 | void nl_attr_rtnh_end(struct nlmsghdr *n, struct rtnexthop *rtnh) |
1fdc9eae | 666 | { |
312a6bee | 667 | rtnh->rtnh_len = (uint8_t *)NLMSG_TAIL(n) - (uint8_t *)rtnh; |
1fdc9eae | 668 | } |
669 | ||
d62a17ae | 670 | const char *nl_msg_type_to_str(uint16_t msg_type) |
1fdc9eae | 671 | { |
d62a17ae | 672 | return lookup_msg(nlmsg_str, msg_type, ""); |
1fdc9eae | 673 | } |
674 | ||
d7c0a89a | 675 | const char *nl_rtproto_to_str(uint8_t rtproto) |
1fdc9eae | 676 | { |
d62a17ae | 677 | return lookup_msg(rtproto_str, rtproto, ""); |
1fdc9eae | 678 | } |
b339bde7 | 679 | |
d7c0a89a | 680 | const char *nl_family_to_str(uint8_t family) |
b339bde7 | 681 | { |
d62a17ae | 682 | return lookup_msg(family_str, family, ""); |
b339bde7 DS |
683 | } |
684 | ||
d7c0a89a | 685 | const char *nl_rttype_to_str(uint8_t rttype) |
b339bde7 | 686 | { |
d62a17ae | 687 | return lookup_msg(rttype_str, rttype, ""); |
b339bde7 DS |
688 | } |
689 | ||
4cebb2b6 | 690 | #define NLA_OK(nla, len) \ |
5d307d5d DS |
691 | ((len) >= (int)sizeof(struct nlattr) \ |
692 | && (nla)->nla_len >= sizeof(struct nlattr) \ | |
693 | && (nla)->nla_len <= (len)) | |
4cebb2b6 SW |
694 | #define NLA_NEXT(nla, attrlen) \ |
695 | ((attrlen) -= NLA_ALIGN((nla)->nla_len), \ | |
696 | (struct nlattr *)(((char *)(nla)) + NLA_ALIGN((nla)->nla_len))) | |
697 | #define NLA_LENGTH(len) (NLA_ALIGN(sizeof(struct nlattr)) + (len)) | |
698 | #define NLA_DATA(nla) ((struct nlattr *)(((char *)(nla)) + NLA_LENGTH(0))) | |
699 | ||
700 | #define ERR_NLA(err, inner_len) \ | |
701 | ((struct nlattr *)(((char *)(err)) \ | |
702 | + NLMSG_ALIGN(sizeof(struct nlmsgerr)) \ | |
703 | + NLMSG_ALIGN((inner_len)))) | |
5d307d5d DS |
704 | |
705 | static void netlink_parse_nlattr(struct nlattr **tb, int max, | |
706 | struct nlattr *nla, int len) | |
707 | { | |
4cebb2b6 | 708 | while (NLA_OK(nla, len)) { |
5d307d5d DS |
709 | if (nla->nla_type <= max) |
710 | tb[nla->nla_type] = nla; | |
4cebb2b6 | 711 | nla = NLA_NEXT(nla, len); |
5d307d5d DS |
712 | } |
713 | } | |
714 | ||
715 | static void netlink_parse_extended_ack(struct nlmsghdr *h) | |
716 | { | |
4cebb2b6 SW |
717 | struct nlattr *tb[NLMSGERR_ATTR_MAX + 1] = {}; |
718 | const struct nlmsgerr *err = (const struct nlmsgerr *)NLMSG_DATA(h); | |
5d307d5d | 719 | const struct nlmsghdr *err_nlh = NULL; |
4cebb2b6 SW |
720 | /* Length not including nlmsghdr */ |
721 | uint32_t len = 0; | |
722 | /* Inner error netlink message length */ | |
723 | uint32_t inner_len = 0; | |
5d307d5d DS |
724 | const char *msg = NULL; |
725 | uint32_t off = 0; | |
726 | ||
727 | if (!(h->nlmsg_flags & NLM_F_CAPPED)) | |
4cebb2b6 SW |
728 | inner_len = (uint32_t)NLMSG_PAYLOAD(&err->msg, 0); |
729 | ||
730 | len = (uint32_t)(NLMSG_PAYLOAD(h, sizeof(struct nlmsgerr)) - inner_len); | |
5d307d5d | 731 | |
4cebb2b6 SW |
732 | netlink_parse_nlattr(tb, NLMSGERR_ATTR_MAX, ERR_NLA(err, inner_len), |
733 | len); | |
5d307d5d DS |
734 | |
735 | if (tb[NLMSGERR_ATTR_MSG]) | |
4cebb2b6 | 736 | msg = (const char *)NLA_DATA(tb[NLMSGERR_ATTR_MSG]); |
5d307d5d DS |
737 | |
738 | if (tb[NLMSGERR_ATTR_OFFS]) { | |
4cebb2b6 | 739 | off = *(uint32_t *)NLA_DATA(tb[NLMSGERR_ATTR_OFFS]); |
5d307d5d DS |
740 | |
741 | if (off > h->nlmsg_len) { | |
9165c5f5 | 742 | zlog_err("Invalid offset for NLMSGERR_ATTR_OFFS"); |
5d307d5d DS |
743 | } else if (!(h->nlmsg_flags & NLM_F_CAPPED)) { |
744 | /* | |
745 | * Header of failed message | |
746 | * we are not doing anything currently with it | |
747 | * but noticing it for later. | |
748 | */ | |
749 | err_nlh = &err->msg; | |
15569c58 | 750 | zlog_debug("%s: Received %s extended Ack", __func__, |
87b5d1b0 | 751 | nl_msg_type_to_str(err_nlh->nlmsg_type)); |
5d307d5d DS |
752 | } |
753 | } | |
754 | ||
755 | if (msg && *msg != '\0') { | |
756 | bool is_err = !!err->error; | |
757 | ||
758 | if (is_err) | |
759 | zlog_err("Extended Error: %s", msg); | |
760 | else | |
e914ccbe | 761 | flog_warn(EC_ZEBRA_NETLINK_EXTENDED_WARNING, |
9df414fe | 762 | "Extended Warning: %s", msg); |
5d307d5d DS |
763 | } |
764 | } | |
765 | ||
ae6138bf JU |
766 | /* |
767 | * netlink_send_msg - send a netlink message of a certain size. | |
768 | * | |
769 | * Returns -1 on error. Otherwise, it returns the number of bytes sent. | |
770 | */ | |
f8653393 JU |
771 | static ssize_t netlink_send_msg(const struct nlsock *nl, void *buf, |
772 | size_t buflen) | |
ae6138bf | 773 | { |
f8653393 JU |
774 | struct sockaddr_nl snl = {}; |
775 | struct iovec iov = {}; | |
776 | struct msghdr msg = {}; | |
777 | ssize_t status; | |
778 | int save_errno = 0; | |
ae6138bf JU |
779 | |
780 | iov.iov_base = buf; | |
781 | iov.iov_len = buflen; | |
f8653393 | 782 | msg.msg_name = &snl; |
ae6138bf JU |
783 | msg.msg_namelen = sizeof(snl); |
784 | msg.msg_iov = &iov; | |
785 | msg.msg_iovlen = 1; | |
786 | ||
787 | snl.nl_family = AF_NETLINK; | |
788 | ||
789 | /* Send message to netlink interface. */ | |
790 | frr_with_privs(&zserv_privs) { | |
791 | status = sendmsg(nl->sock, &msg, 0); | |
792 | save_errno = errno; | |
793 | } | |
794 | ||
795 | if (IS_ZEBRA_DEBUG_KERNEL_MSGDUMP_SEND) { | |
796 | zlog_debug("%s: >> netlink message dump [sent]", __func__); | |
eead0bc4 RZ |
797 | #ifdef NETLINK_DEBUG |
798 | nl_dump(buf, buflen); | |
799 | #else | |
ae6138bf | 800 | zlog_hexdump(buf, buflen); |
eead0bc4 | 801 | #endif /* NETLINK_DEBUG */ |
ae6138bf JU |
802 | } |
803 | ||
f8653393 | 804 | if (status == -1) { |
ae6138bf JU |
805 | flog_err_sys(EC_LIB_SOCKET, "%s error: %s", __func__, |
806 | safe_strerror(save_errno)); | |
807 | return -1; | |
808 | } | |
809 | ||
810 | return status; | |
811 | } | |
812 | ||
813 | /* | |
814 | * netlink_recv_msg - receive a netlink message. | |
815 | * | |
816 | * Returns -1 on error, 0 if read would block or the number of bytes received. | |
817 | */ | |
2cf7651f | 818 | static int netlink_recv_msg(struct nlsock *nl, struct msghdr *msg) |
ae6138bf JU |
819 | { |
820 | struct iovec iov; | |
821 | int status; | |
822 | ||
2cf7651f DS |
823 | iov.iov_base = nl->buf; |
824 | iov.iov_len = nl->buflen; | |
825 | msg->msg_iov = &iov; | |
826 | msg->msg_iovlen = 1; | |
ae6138bf JU |
827 | |
828 | do { | |
2cf7651f DS |
829 | int bytes; |
830 | ||
831 | bytes = recv(nl->sock, NULL, 0, MSG_PEEK | MSG_TRUNC); | |
832 | ||
833 | if (bytes >= 0 && (size_t)bytes > nl->buflen) { | |
834 | nl->buf = XREALLOC(MTYPE_NL_BUF, nl->buf, bytes); | |
835 | nl->buflen = bytes; | |
836 | iov.iov_base = nl->buf; | |
837 | iov.iov_len = nl->buflen; | |
838 | } | |
839 | ||
840 | status = recvmsg(nl->sock, msg, 0); | |
f8653393 | 841 | } while (status == -1 && errno == EINTR); |
ae6138bf | 842 | |
f8653393 | 843 | if (status == -1) { |
ae6138bf JU |
844 | if (errno == EWOULDBLOCK || errno == EAGAIN) |
845 | return 0; | |
846 | flog_err(EC_ZEBRA_RECVMSG_OVERRUN, "%s recvmsg overrun: %s", | |
847 | nl->name, safe_strerror(errno)); | |
848 | /* | |
849 | * In this case we are screwed. There is no good way to recover | |
850 | * zebra at this point. | |
851 | */ | |
852 | exit(-1); | |
853 | } | |
854 | ||
855 | if (status == 0) { | |
856 | flog_err_sys(EC_LIB_SOCKET, "%s EOF", nl->name); | |
857 | return -1; | |
858 | } | |
859 | ||
2cf7651f | 860 | if (msg->msg_namelen != sizeof(struct sockaddr_nl)) { |
ae6138bf JU |
861 | flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR, |
862 | "%s sender address length error: length %d", nl->name, | |
2cf7651f | 863 | msg->msg_namelen); |
ae6138bf JU |
864 | return -1; |
865 | } | |
866 | ||
867 | if (IS_ZEBRA_DEBUG_KERNEL_MSGDUMP_RECV) { | |
868 | zlog_debug("%s: << netlink message dump [recv]", __func__); | |
eead0bc4 | 869 | #ifdef NETLINK_DEBUG |
2cf7651f | 870 | nl_dump(nl->buf, status); |
eead0bc4 | 871 | #else |
2cf7651f | 872 | zlog_hexdump(nl->buf, status); |
eead0bc4 | 873 | #endif /* NETLINK_DEBUG */ |
ae6138bf JU |
874 | } |
875 | ||
ae6138bf JU |
876 | return status; |
877 | } | |
878 | ||
879 | /* | |
880 | * netlink_parse_error - parse a netlink error message | |
881 | * | |
882 | * Returns 1 if this message is acknowledgement, 0 if this error should be | |
883 | * ignored, -1 otherwise. | |
884 | */ | |
885 | static int netlink_parse_error(const struct nlsock *nl, struct nlmsghdr *h, | |
d166308b | 886 | bool is_cmd, bool startup) |
ae6138bf JU |
887 | { |
888 | struct nlmsgerr *err = (struct nlmsgerr *)NLMSG_DATA(h); | |
889 | int errnum = err->error; | |
890 | int msg_type = err->msg.nlmsg_type; | |
891 | ||
892 | if (h->nlmsg_len < NLMSG_LENGTH(sizeof(struct nlmsgerr))) { | |
893 | flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR, | |
894 | "%s error: message truncated", nl->name); | |
895 | return -1; | |
896 | } | |
897 | ||
898 | /* | |
899 | * Parse the extended information before we actually handle it. At this | |
900 | * point in time we do not do anything other than report the issue. | |
901 | */ | |
902 | if (h->nlmsg_flags & NLM_F_ACK_TLVS) | |
903 | netlink_parse_extended_ack(h); | |
904 | ||
905 | /* If the error field is zero, then this is an ACK. */ | |
906 | if (err->error == 0) { | |
907 | if (IS_ZEBRA_DEBUG_KERNEL) { | |
908 | zlog_debug("%s: %s ACK: type=%s(%u), seq=%u, pid=%u", | |
909 | __func__, nl->name, | |
910 | nl_msg_type_to_str(err->msg.nlmsg_type), | |
911 | err->msg.nlmsg_type, err->msg.nlmsg_seq, | |
912 | err->msg.nlmsg_pid); | |
913 | } | |
914 | ||
915 | return 1; | |
916 | } | |
917 | ||
918 | /* Deal with errors that occur because of races in link handling. */ | |
d166308b | 919 | if (is_cmd |
ae6138bf JU |
920 | && ((msg_type == RTM_DELROUTE |
921 | && (-errnum == ENODEV || -errnum == ESRCH)) | |
922 | || (msg_type == RTM_NEWROUTE | |
923 | && (-errnum == ENETDOWN || -errnum == EEXIST)))) { | |
924 | if (IS_ZEBRA_DEBUG_KERNEL) | |
925 | zlog_debug("%s: error: %s type=%s(%u), seq=%u, pid=%u", | |
926 | nl->name, safe_strerror(-errnum), | |
927 | nl_msg_type_to_str(msg_type), msg_type, | |
928 | err->msg.nlmsg_seq, err->msg.nlmsg_pid); | |
929 | return 0; | |
930 | } | |
931 | ||
932 | /* | |
933 | * We see RTM_DELNEIGH when shutting down an interface with an IPv4 | |
934 | * link-local. The kernel should have already deleted the neighbor so | |
935 | * do not log these as an error. | |
936 | */ | |
937 | if (msg_type == RTM_DELNEIGH | |
d166308b | 938 | || (is_cmd && msg_type == RTM_NEWROUTE |
ae6138bf JU |
939 | && (-errnum == ESRCH || -errnum == ENETUNREACH))) { |
940 | /* | |
941 | * This is known to happen in some situations, don't log as | |
942 | * error. | |
943 | */ | |
944 | if (IS_ZEBRA_DEBUG_KERNEL) | |
945 | zlog_debug("%s error: %s, type=%s(%u), seq=%u, pid=%u", | |
946 | nl->name, safe_strerror(-errnum), | |
947 | nl_msg_type_to_str(msg_type), msg_type, | |
948 | err->msg.nlmsg_seq, err->msg.nlmsg_pid); | |
949 | } else { | |
950 | if ((msg_type != RTM_GETNEXTHOP) || !startup) | |
951 | flog_err(EC_ZEBRA_UNEXPECTED_MESSAGE, | |
952 | "%s error: %s, type=%s(%u), seq=%u, pid=%u", | |
953 | nl->name, safe_strerror(-errnum), | |
954 | nl_msg_type_to_str(msg_type), msg_type, | |
955 | err->msg.nlmsg_seq, err->msg.nlmsg_pid); | |
956 | } | |
957 | ||
958 | return -1; | |
959 | } | |
960 | ||
936ebf0a DS |
961 | /* |
962 | * netlink_parse_info | |
963 | * | |
964 | * Receive message from netlink interface and pass those information | |
965 | * to the given function. | |
966 | * | |
967 | * filter -> Function to call to read the results | |
968 | * nl -> netlink socket information | |
969 | * zns -> The zebra namespace data | |
970 | * count -> How many we should read in, 0 means as much as possible | |
971 | * startup -> Are we reading in under startup conditions? passed to | |
972 | * the filter. | |
973 | */ | |
2414abd3 | 974 | int netlink_parse_info(int (*filter)(struct nlmsghdr *, ns_id_t, int), |
2cf7651f | 975 | struct nlsock *nl, const struct zebra_dplane_info *zns, |
9bfadae8 | 976 | int count, bool startup) |
1fdc9eae | 977 | { |
d62a17ae | 978 | int status; |
979 | int ret = 0; | |
980 | int error; | |
981 | int read_in = 0; | |
982 | ||
983 | while (1) { | |
d62a17ae | 984 | struct sockaddr_nl snl; |
985 | struct msghdr msg = {.msg_name = (void *)&snl, | |
ae6138bf | 986 | .msg_namelen = sizeof(snl)}; |
d62a17ae | 987 | struct nlmsghdr *h; |
988 | ||
989 | if (count && read_in >= count) | |
990 | return 0; | |
991 | ||
2cf7651f | 992 | status = netlink_recv_msg(nl, &msg); |
ae6138bf | 993 | if (status == -1) |
d62a17ae | 994 | return -1; |
ae6138bf JU |
995 | else if (status == 0) |
996 | break; | |
81a2f870 | 997 | |
d62a17ae | 998 | read_in++; |
2cf7651f | 999 | for (h = (struct nlmsghdr *)nl->buf; |
e6a0e0d1 | 1000 | (status >= 0 && NLMSG_OK(h, (unsigned int)status)); |
d62a17ae | 1001 | h = NLMSG_NEXT(h, status)) { |
1002 | /* Finish of reading. */ | |
1003 | if (h->nlmsg_type == NLMSG_DONE) | |
1004 | return ret; | |
1005 | ||
1006 | /* Error handling. */ | |
1007 | if (h->nlmsg_type == NLMSG_ERROR) { | |
d166308b MS |
1008 | int err = netlink_parse_error( |
1009 | nl, h, zns->is_cmd, startup); | |
1010 | ||
ae6138bf | 1011 | if (err == 1) { |
d62a17ae | 1012 | if (!(h->nlmsg_flags & NLM_F_MULTI)) |
1013 | return 0; | |
1014 | continue; | |
ae6138bf JU |
1015 | } else |
1016 | return err; | |
d62a17ae | 1017 | } |
1018 | ||
1019 | /* OK we got netlink message. */ | |
1020 | if (IS_ZEBRA_DEBUG_KERNEL) | |
1021 | zlog_debug( | |
d166308b MS |
1022 | "%s: %s type %s(%u), len=%d, seq=%u, pid=%u", |
1023 | __func__, nl->name, | |
d62a17ae | 1024 | nl_msg_type_to_str(h->nlmsg_type), |
1025 | h->nlmsg_type, h->nlmsg_len, | |
1026 | h->nlmsg_seq, h->nlmsg_pid); | |
1027 | ||
783827ae DS |
1028 | |
1029 | /* | |
1030 | * Ignore messages that maybe sent from | |
1031 | * other actors besides the kernel | |
1032 | */ | |
1033 | if (snl.nl_pid != 0) { | |
43e52561 QY |
1034 | zlog_debug("Ignoring message from pid %u", |
1035 | snl.nl_pid); | |
d62a17ae | 1036 | continue; |
1037 | } | |
1038 | ||
2414abd3 | 1039 | error = (*filter)(h, zns->ns_id, startup); |
d62a17ae | 1040 | if (error < 0) { |
9df414fe QY |
1041 | zlog_debug("%s filter function error", |
1042 | nl->name); | |
d62a17ae | 1043 | ret = error; |
1044 | } | |
1045 | } | |
1046 | ||
1047 | /* After error care. */ | |
1048 | if (msg.msg_flags & MSG_TRUNC) { | |
e914ccbe | 1049 | flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR, |
1c50c1c0 | 1050 | "%s error: message truncated", nl->name); |
d62a17ae | 1051 | continue; |
1052 | } | |
1053 | if (status) { | |
e914ccbe | 1054 | flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR, |
1c50c1c0 QY |
1055 | "%s error: data remnant size %d", nl->name, |
1056 | status); | |
d62a17ae | 1057 | return -1; |
1058 | } | |
1059 | } | |
1060 | return ret; | |
1fdc9eae | 1061 | } |
1062 | ||
936ebf0a | 1063 | /* |
7cdb1a84 | 1064 | * netlink_talk_info |
936ebf0a DS |
1065 | * |
1066 | * sendmsg() to netlink socket then recvmsg(). | |
1067 | * Calls netlink_parse_info to parse returned data | |
1068 | * | |
1069 | * filter -> The filter to read final results from kernel | |
1070 | * nlmsghdr -> The data to send to the kernel | |
8b962e77 | 1071 | * dp_info -> The dataplane and netlink socket information |
936ebf0a DS |
1072 | * startup -> Are we reading in under startup conditions |
1073 | * This is passed through eventually to filter. | |
1074 | */ | |
2cf7651f DS |
1075 | static int netlink_talk_info(int (*filter)(struct nlmsghdr *, ns_id_t, |
1076 | int startup), | |
1077 | struct nlmsghdr *n, | |
1078 | struct zebra_dplane_info *dp_info, bool startup) | |
1fdc9eae | 1079 | { |
d4000d7b | 1080 | struct nlsock *nl; |
d62a17ae | 1081 | |
d4000d7b | 1082 | nl = kernel_netlink_nlsock_lookup(dp_info->sock); |
3670f504 | 1083 | n->nlmsg_seq = dp_info->seq; |
d62a17ae | 1084 | n->nlmsg_pid = nl->snl.nl_pid; |
1085 | ||
d62a17ae | 1086 | if (IS_ZEBRA_DEBUG_KERNEL) |
1087 | zlog_debug( | |
1088 | "netlink_talk: %s type %s(%u), len=%d seq=%u flags 0x%x", | |
1089 | nl->name, nl_msg_type_to_str(n->nlmsg_type), | |
1090 | n->nlmsg_type, n->nlmsg_len, n->nlmsg_seq, | |
1091 | n->nlmsg_flags); | |
1092 | ||
f8653393 | 1093 | if (netlink_send_msg(nl, n, n->nlmsg_len) == -1) |
d62a17ae | 1094 | return -1; |
d62a17ae | 1095 | |
d62a17ae | 1096 | /* |
1097 | * Get reply from netlink socket. | |
1098 | * The reply should either be an acknowlegement or an error. | |
1099 | */ | |
7cdb1a84 MS |
1100 | return netlink_parse_info(filter, nl, dp_info, 0, startup); |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * Synchronous version of netlink_talk_info. Converts args to suit the | |
1105 | * common version, which is suitable for both sync and async use. | |
7cdb1a84 MS |
1106 | */ |
1107 | int netlink_talk(int (*filter)(struct nlmsghdr *, ns_id_t, int startup), | |
1108 | struct nlmsghdr *n, struct nlsock *nl, struct zebra_ns *zns, | |
9bfadae8 | 1109 | bool startup) |
7cdb1a84 MS |
1110 | { |
1111 | struct zebra_dplane_info dp_info; | |
1112 | ||
1113 | /* Increment sequence number before capturing snapshot of ns socket | |
1114 | * info. | |
1115 | */ | |
1116 | nl->seq++; | |
1117 | ||
1118 | /* Capture info in intermediate info struct */ | |
85a75f1e | 1119 | zebra_dplane_info_from_zns(&dp_info, zns, (nl == &(zns->netlink_cmd))); |
7cdb1a84 | 1120 | |
5709131c | 1121 | return netlink_talk_info(filter, n, &dp_info, startup); |
1fdc9eae | 1122 | } |
1123 | ||
289602d7 | 1124 | /* Issue request message to kernel via netlink socket. GET messages |
1125 | * are issued through this interface. | |
1126 | */ | |
fd3f8e52 | 1127 | int netlink_request(struct nlsock *nl, void *req) |
1fdc9eae | 1128 | { |
fd3f8e52 | 1129 | struct nlmsghdr *n = (struct nlmsghdr *)req; |
d62a17ae | 1130 | |
1131 | /* Check netlink socket. */ | |
1132 | if (nl->sock < 0) { | |
450971aa | 1133 | flog_err_sys(EC_LIB_SOCKET, "%s socket isn't active.", |
09c866e3 | 1134 | nl->name); |
d62a17ae | 1135 | return -1; |
1136 | } | |
1137 | ||
1138 | /* Fill common fields for all requests. */ | |
d62a17ae | 1139 | n->nlmsg_pid = nl->snl.nl_pid; |
1140 | n->nlmsg_seq = ++nl->seq; | |
1141 | ||
f8653393 | 1142 | if (netlink_send_msg(nl, req, n->nlmsg_len) == -1) |
d62a17ae | 1143 | return -1; |
d62a17ae | 1144 | |
1145 | return 0; | |
1fdc9eae | 1146 | } |
1147 | ||
e63c7622 JU |
1148 | static int nl_batch_read_resp(struct nl_batch *bth) |
1149 | { | |
1150 | struct nlmsghdr *h; | |
1151 | struct sockaddr_nl snl; | |
9d06e121 | 1152 | struct msghdr msg = {}; |
f6feb48b | 1153 | int status, seq; |
d4000d7b | 1154 | struct nlsock *nl; |
f6feb48b JU |
1155 | struct zebra_dplane_ctx *ctx; |
1156 | bool ignore_msg; | |
e63c7622 | 1157 | |
d4000d7b | 1158 | nl = kernel_netlink_nlsock_lookup(bth->zns->sock); |
e63c7622 JU |
1159 | |
1160 | msg.msg_name = (void *)&snl; | |
1161 | msg.msg_namelen = sizeof(snl); | |
1162 | ||
2f9dbd3a JU |
1163 | /* |
1164 | * The responses are not batched, so we need to read and process one | |
1165 | * message at a time. | |
1166 | */ | |
1167 | while (true) { | |
2cf7651f | 1168 | status = netlink_recv_msg(nl, &msg); |
00249e25 DS |
1169 | /* |
1170 | * status == -1 is a full on failure somewhere | |
1171 | * since we don't know where the problem happened | |
1172 | * we must mark all as failed | |
1173 | * | |
1174 | * Else we mark everything as worked | |
1175 | * | |
1176 | */ | |
1177 | if (status == -1 || status == 0) { | |
1178 | while ((ctx = dplane_ctx_dequeue(&(bth->ctx_list))) != | |
1179 | NULL) { | |
1180 | if (status == -1) | |
1181 | dplane_ctx_set_status( | |
1182 | ctx, | |
1183 | ZEBRA_DPLANE_REQUEST_FAILURE); | |
1184 | dplane_ctx_enqueue_tail(bth->ctx_out_q, ctx); | |
1185 | } | |
2f9dbd3a | 1186 | return status; |
00249e25 | 1187 | } |
e63c7622 | 1188 | |
2cf7651f | 1189 | h = (struct nlmsghdr *)nl->buf; |
f6feb48b JU |
1190 | ignore_msg = false; |
1191 | seq = h->nlmsg_seq; | |
e63c7622 | 1192 | /* |
f6feb48b JU |
1193 | * Find the corresponding context object. Received responses are |
1194 | * in the same order as requests we sent, so we can simply | |
1195 | * iterate over the context list and match responses with | |
1196 | * requests at same time. | |
e63c7622 | 1197 | */ |
f6feb48b | 1198 | while (true) { |
c8453cd7 DS |
1199 | ctx = dplane_ctx_get_head(&(bth->ctx_list)); |
1200 | if (ctx == NULL) { | |
1201 | /* | |
1202 | * This is a situation where we have gotten | |
1203 | * into a bad spot. We need to know that | |
1204 | * this happens( does it? ) | |
1205 | */ | |
1206 | zlog_err( | |
1207 | "%s:WARNING Received netlink Response for an error and no Contexts to associate with it", | |
1208 | __func__); | |
e63c7622 | 1209 | break; |
c8453cd7 | 1210 | } |
f6feb48b JU |
1211 | |
1212 | /* | |
1213 | * 'update' context objects take two consecutive | |
1214 | * sequence numbers. | |
1215 | */ | |
3670f504 DS |
1216 | if (dplane_ctx_is_update(ctx) && |
1217 | dplane_ctx_get_ns(ctx)->seq + 1 == seq) { | |
f6feb48b JU |
1218 | /* |
1219 | * This is the situation where we get a response | |
1220 | * to a message that should be ignored. | |
1221 | */ | |
1222 | ignore_msg = true; | |
1223 | break; | |
1224 | } | |
c8453cd7 DS |
1225 | |
1226 | ctx = dplane_ctx_dequeue(&(bth->ctx_list)); | |
1227 | dplane_ctx_enqueue_tail(bth->ctx_out_q, ctx); | |
1228 | ||
1229 | /* We have found corresponding context object. */ | |
3670f504 | 1230 | if (dplane_ctx_get_ns(ctx)->seq == seq) |
c8453cd7 DS |
1231 | break; |
1232 | ||
3670f504 | 1233 | if (dplane_ctx_get_ns(ctx)->seq > seq) |
c8453cd7 | 1234 | zlog_warn( |
b9d95135 | 1235 | "%s:WARNING Received %u is less than any context on the queue ctx->seq %u", |
c8453cd7 | 1236 | __func__, seq, |
3670f504 | 1237 | dplane_ctx_get_ns(ctx)->seq); |
e63c7622 JU |
1238 | } |
1239 | ||
c8453cd7 DS |
1240 | if (ignore_msg) { |
1241 | /* | |
1242 | * If we ignore the message due to an update | |
1243 | * above we should still fricking decode the | |
1244 | * message for our operator to understand | |
1245 | * what is going on | |
1246 | */ | |
1247 | int err = netlink_parse_error(nl, h, bth->zns->is_cmd, | |
1248 | false); | |
1249 | ||
1250 | zlog_debug("%s: netlink error message seq=%d %d", | |
1251 | __func__, h->nlmsg_seq, err); | |
f6feb48b | 1252 | continue; |
c8453cd7 | 1253 | } |
f6feb48b | 1254 | |
e63c7622 JU |
1255 | /* |
1256 | * We received a message with the sequence number that isn't | |
1257 | * associated with any dplane context object. | |
1258 | */ | |
f6feb48b | 1259 | if (ctx == NULL) { |
4c99d413 MS |
1260 | if (IS_ZEBRA_DEBUG_KERNEL) |
1261 | zlog_debug( | |
1262 | "%s: skipping unassociated response, seq number %d NS %u", | |
1263 | __func__, h->nlmsg_seq, | |
1264 | bth->zns->ns_id); | |
e63c7622 JU |
1265 | continue; |
1266 | } | |
1267 | ||
1268 | if (h->nlmsg_type == NLMSG_ERROR) { | |
d166308b MS |
1269 | int err = netlink_parse_error(nl, h, bth->zns->is_cmd, |
1270 | false); | |
e63c7622 JU |
1271 | |
1272 | if (err == -1) | |
f6feb48b JU |
1273 | dplane_ctx_set_status( |
1274 | ctx, ZEBRA_DPLANE_REQUEST_FAILURE); | |
e63c7622 | 1275 | |
4c99d413 MS |
1276 | if (IS_ZEBRA_DEBUG_KERNEL) |
1277 | zlog_debug("%s: netlink error message seq=%d ", | |
1278 | __func__, h->nlmsg_seq); | |
e63c7622 JU |
1279 | continue; |
1280 | } | |
1281 | ||
1282 | /* | |
1283 | * If we get here then we did not receive neither the ack nor | |
1284 | * the error and instead received some other message in an | |
1285 | * unexpected way. | |
1286 | */ | |
4c99d413 MS |
1287 | if (IS_ZEBRA_DEBUG_KERNEL) |
1288 | zlog_debug("%s: ignoring message type 0x%04x(%s) NS %u", | |
1289 | __func__, h->nlmsg_type, | |
1290 | nl_msg_type_to_str(h->nlmsg_type), | |
1291 | bth->zns->ns_id); | |
e63c7622 JU |
1292 | } |
1293 | ||
1294 | return 0; | |
1295 | } | |
1296 | ||
1297 | static void nl_batch_reset(struct nl_batch *bth) | |
1298 | { | |
e63c7622 JU |
1299 | bth->buf_head = bth->buf; |
1300 | bth->curlen = 0; | |
1301 | bth->msgcnt = 0; | |
1302 | bth->zns = NULL; | |
1303 | ||
f6feb48b | 1304 | TAILQ_INIT(&(bth->ctx_list)); |
e63c7622 JU |
1305 | } |
1306 | ||
f6feb48b | 1307 | static void nl_batch_init(struct nl_batch *bth, struct dplane_ctx_q *ctx_out_q) |
e63c7622 | 1308 | { |
531c92b8 JU |
1309 | /* |
1310 | * If the size of the buffer has changed, free and then allocate a new | |
1311 | * one. | |
1312 | */ | |
1313 | size_t bufsize = | |
1314 | atomic_load_explicit(&nl_batch_bufsize, memory_order_relaxed); | |
1315 | if (bufsize != nl_batch_tx_bufsize) { | |
1316 | if (nl_batch_tx_buf) | |
1317 | XFREE(MTYPE_NL_BUF, nl_batch_tx_buf); | |
1318 | ||
1319 | nl_batch_tx_buf = XCALLOC(MTYPE_NL_BUF, bufsize); | |
1320 | nl_batch_tx_bufsize = bufsize; | |
1321 | } | |
1322 | ||
f6feb48b | 1323 | bth->buf = nl_batch_tx_buf; |
531c92b8 JU |
1324 | bth->bufsiz = bufsize; |
1325 | bth->limit = atomic_load_explicit(&nl_batch_send_threshold, | |
1326 | memory_order_relaxed); | |
e63c7622 | 1327 | |
f6feb48b | 1328 | bth->ctx_out_q = ctx_out_q; |
e63c7622 | 1329 | |
f6feb48b JU |
1330 | nl_batch_reset(bth); |
1331 | } | |
1332 | ||
1333 | static void nl_batch_send(struct nl_batch *bth) | |
1334 | { | |
1335 | struct zebra_dplane_ctx *ctx; | |
1336 | bool err = false; | |
e63c7622 | 1337 | |
f6feb48b | 1338 | if (bth->curlen != 0 && bth->zns != NULL) { |
d4000d7b DS |
1339 | struct nlsock *nl = |
1340 | kernel_netlink_nlsock_lookup(bth->zns->sock); | |
1341 | ||
f6feb48b JU |
1342 | if (IS_ZEBRA_DEBUG_KERNEL) |
1343 | zlog_debug("%s: %s, batch size=%zu, msg cnt=%zu", | |
d4000d7b | 1344 | __func__, nl->name, bth->curlen, |
f6feb48b | 1345 | bth->msgcnt); |
e63c7622 | 1346 | |
d4000d7b | 1347 | if (netlink_send_msg(nl, bth->buf, bth->curlen) == -1) |
e63c7622 | 1348 | err = true; |
e63c7622 | 1349 | |
f6feb48b JU |
1350 | if (!err) { |
1351 | if (nl_batch_read_resp(bth) == -1) | |
1352 | err = true; | |
1353 | } | |
1354 | } | |
e63c7622 | 1355 | |
f6feb48b JU |
1356 | /* Move remaining contexts to the outbound queue. */ |
1357 | while (true) { | |
1358 | ctx = dplane_ctx_dequeue(&(bth->ctx_list)); | |
1359 | if (ctx == NULL) | |
1360 | break; | |
e63c7622 | 1361 | |
f6feb48b JU |
1362 | if (err) |
1363 | dplane_ctx_set_status(ctx, | |
1364 | ZEBRA_DPLANE_REQUEST_FAILURE); | |
e63c7622 | 1365 | |
f6feb48b | 1366 | dplane_ctx_enqueue_tail(bth->ctx_out_q, ctx); |
e63c7622 JU |
1367 | } |
1368 | ||
1369 | nl_batch_reset(bth); | |
1370 | } | |
1371 | ||
e63c7622 JU |
1372 | enum netlink_msg_status netlink_batch_add_msg( |
1373 | struct nl_batch *bth, struct zebra_dplane_ctx *ctx, | |
1374 | ssize_t (*msg_encoder)(struct zebra_dplane_ctx *, void *, size_t), | |
f6feb48b | 1375 | bool ignore_res) |
e63c7622 JU |
1376 | { |
1377 | int seq; | |
1378 | ssize_t size; | |
1379 | struct nlmsghdr *msgh; | |
d4000d7b | 1380 | struct nlsock *nl; |
e63c7622 | 1381 | |
e63c7622 JU |
1382 | size = (*msg_encoder)(ctx, bth->buf_head, bth->bufsiz - bth->curlen); |
1383 | ||
1384 | /* | |
1385 | * If there was an error while encoding the message (other than buffer | |
1386 | * overflow) then return an error. | |
1387 | */ | |
1388 | if (size < 0) | |
1389 | return FRR_NETLINK_ERROR; | |
1390 | ||
1391 | /* | |
1392 | * If the message doesn't fit entirely in the buffer then send the batch | |
1393 | * and retry. | |
1394 | */ | |
1395 | if (size == 0) { | |
1396 | nl_batch_send(bth); | |
1397 | size = (*msg_encoder)(ctx, bth->buf_head, | |
1398 | bth->bufsiz - bth->curlen); | |
1399 | /* | |
1400 | * If the message doesn't fit in the empty buffer then just | |
1401 | * return an error. | |
1402 | */ | |
1403 | if (size <= 0) | |
1404 | return FRR_NETLINK_ERROR; | |
1405 | } | |
1406 | ||
3670f504 | 1407 | seq = dplane_ctx_get_ns(ctx)->seq; |
d4000d7b DS |
1408 | nl = kernel_netlink_nlsock_lookup(dplane_ctx_get_ns_sock(ctx)); |
1409 | ||
f6feb48b | 1410 | if (ignore_res) |
e63c7622 JU |
1411 | seq++; |
1412 | ||
1413 | msgh = (struct nlmsghdr *)bth->buf_head; | |
1414 | msgh->nlmsg_seq = seq; | |
d4000d7b | 1415 | msgh->nlmsg_pid = nl->snl.nl_pid; |
e63c7622 | 1416 | |
e63c7622 JU |
1417 | bth->zns = dplane_ctx_get_ns(ctx); |
1418 | bth->buf_head = ((char *)bth->buf_head) + size; | |
1419 | bth->curlen += size; | |
1420 | bth->msgcnt++; | |
1421 | ||
e63c7622 JU |
1422 | return FRR_NETLINK_QUEUED; |
1423 | } | |
1424 | ||
67e3369e JU |
1425 | static enum netlink_msg_status nl_put_msg(struct nl_batch *bth, |
1426 | struct zebra_dplane_ctx *ctx) | |
1427 | { | |
1428 | if (dplane_ctx_is_skip_kernel(ctx)) | |
1429 | return FRR_NETLINK_SUCCESS; | |
1430 | ||
1431 | switch (dplane_ctx_get_op(ctx)) { | |
1432 | ||
1433 | case DPLANE_OP_ROUTE_INSTALL: | |
1434 | case DPLANE_OP_ROUTE_UPDATE: | |
1435 | case DPLANE_OP_ROUTE_DELETE: | |
1436 | return netlink_put_route_update_msg(bth, ctx); | |
1437 | ||
1438 | case DPLANE_OP_NH_INSTALL: | |
1439 | case DPLANE_OP_NH_UPDATE: | |
1440 | case DPLANE_OP_NH_DELETE: | |
1441 | return netlink_put_nexthop_update_msg(bth, ctx); | |
1442 | ||
1443 | case DPLANE_OP_LSP_INSTALL: | |
1444 | case DPLANE_OP_LSP_UPDATE: | |
1445 | case DPLANE_OP_LSP_DELETE: | |
1446 | return netlink_put_lsp_update_msg(bth, ctx); | |
1447 | ||
1448 | case DPLANE_OP_PW_INSTALL: | |
1449 | case DPLANE_OP_PW_UNINSTALL: | |
1450 | return netlink_put_pw_update_msg(bth, ctx); | |
1451 | ||
1452 | case DPLANE_OP_ADDR_INSTALL: | |
1453 | case DPLANE_OP_ADDR_UNINSTALL: | |
1454 | return netlink_put_address_update_msg(bth, ctx); | |
1455 | ||
1456 | case DPLANE_OP_MAC_INSTALL: | |
1457 | case DPLANE_OP_MAC_DELETE: | |
1458 | return netlink_put_mac_update_msg(bth, ctx); | |
1459 | ||
1460 | case DPLANE_OP_NEIGH_INSTALL: | |
1461 | case DPLANE_OP_NEIGH_UPDATE: | |
1462 | case DPLANE_OP_NEIGH_DELETE: | |
1463 | case DPLANE_OP_VTEP_ADD: | |
1464 | case DPLANE_OP_VTEP_DELETE: | |
d68e74b4 | 1465 | case DPLANE_OP_NEIGH_DISCOVER: |
0a27a2fe PG |
1466 | case DPLANE_OP_NEIGH_IP_INSTALL: |
1467 | case DPLANE_OP_NEIGH_IP_DELETE: | |
e18747a9 | 1468 | case DPLANE_OP_NEIGH_TABLE_UPDATE: |
67e3369e JU |
1469 | return netlink_put_neigh_update_msg(bth, ctx); |
1470 | ||
1471 | case DPLANE_OP_RULE_ADD: | |
1472 | case DPLANE_OP_RULE_DELETE: | |
1473 | case DPLANE_OP_RULE_UPDATE: | |
1474 | return netlink_put_rule_update_msg(bth, ctx); | |
1475 | ||
1476 | case DPLANE_OP_SYS_ROUTE_ADD: | |
1477 | case DPLANE_OP_SYS_ROUTE_DELETE: | |
1478 | case DPLANE_OP_ROUTE_NOTIFY: | |
1479 | case DPLANE_OP_LSP_NOTIFY: | |
c60522f7 | 1480 | case DPLANE_OP_BR_PORT_UPDATE: |
67e3369e JU |
1481 | return FRR_NETLINK_SUCCESS; |
1482 | ||
5162e000 PG |
1483 | case DPLANE_OP_IPTABLE_ADD: |
1484 | case DPLANE_OP_IPTABLE_DELETE: | |
ef524230 PG |
1485 | case DPLANE_OP_IPSET_ADD: |
1486 | case DPLANE_OP_IPSET_DELETE: | |
1487 | case DPLANE_OP_IPSET_ENTRY_ADD: | |
1488 | case DPLANE_OP_IPSET_ENTRY_DELETE: | |
5162e000 PG |
1489 | return FRR_NETLINK_ERROR; |
1490 | ||
62b4b7e4 PG |
1491 | case DPLANE_OP_GRE_SET: |
1492 | return netlink_put_gre_set_msg(bth, ctx); | |
1493 | ||
9d59df63 MS |
1494 | case DPLANE_OP_INTF_ADDR_ADD: |
1495 | case DPLANE_OP_INTF_ADDR_DEL: | |
728f2017 | 1496 | case DPLANE_OP_INTF_NETCONFIG: |
67e3369e JU |
1497 | case DPLANE_OP_NONE: |
1498 | return FRR_NETLINK_ERROR; | |
5d414138 SW |
1499 | |
1500 | case DPLANE_OP_INTF_INSTALL: | |
1501 | case DPLANE_OP_INTF_UPDATE: | |
1502 | case DPLANE_OP_INTF_DELETE: | |
1503 | return netlink_put_intf_update_msg(bth, ctx); | |
67e3369e JU |
1504 | } |
1505 | ||
1506 | return FRR_NETLINK_ERROR; | |
1507 | } | |
1508 | ||
fef24b03 JU |
1509 | void kernel_update_multi(struct dplane_ctx_q *ctx_list) |
1510 | { | |
67e3369e JU |
1511 | struct nl_batch batch; |
1512 | struct zebra_dplane_ctx *ctx; | |
1513 | struct dplane_ctx_q handled_list; | |
1514 | enum netlink_msg_status res; | |
1515 | ||
67e3369e | 1516 | TAILQ_INIT(&handled_list); |
f6feb48b | 1517 | nl_batch_init(&batch, &handled_list); |
67e3369e JU |
1518 | |
1519 | while (true) { | |
1520 | ctx = dplane_ctx_dequeue(ctx_list); | |
1521 | if (ctx == NULL) | |
1522 | break; | |
1523 | ||
f6feb48b JU |
1524 | if (batch.zns != NULL |
1525 | && batch.zns->ns_id != dplane_ctx_get_ns(ctx)->ns_id) | |
1526 | nl_batch_send(&batch); | |
67e3369e JU |
1527 | |
1528 | /* | |
f6feb48b JU |
1529 | * Assume all messages will succeed and then mark only the ones |
1530 | * that failed. | |
67e3369e | 1531 | */ |
f6feb48b JU |
1532 | dplane_ctx_set_status(ctx, ZEBRA_DPLANE_REQUEST_SUCCESS); |
1533 | ||
1534 | res = nl_put_msg(&batch, ctx); | |
1535 | ||
1536 | dplane_ctx_enqueue_tail(&(batch.ctx_list), ctx); | |
1537 | if (res == FRR_NETLINK_ERROR) | |
67e3369e JU |
1538 | dplane_ctx_set_status(ctx, |
1539 | ZEBRA_DPLANE_REQUEST_FAILURE); | |
1540 | ||
f6feb48b JU |
1541 | if (batch.curlen > batch.limit) |
1542 | nl_batch_send(&batch); | |
67e3369e JU |
1543 | } |
1544 | ||
1545 | nl_batch_send(&batch); | |
1546 | ||
1547 | TAILQ_INIT(ctx_list); | |
1548 | dplane_ctx_list_append(ctx_list, &handled_list); | |
fef24b03 JU |
1549 | } |
1550 | ||
d4000d7b DS |
1551 | struct nlsock *kernel_netlink_nlsock_lookup(int sock) |
1552 | { | |
34869809 | 1553 | struct nlsock lookup, *retval; |
d4000d7b DS |
1554 | |
1555 | lookup.sock = sock; | |
1556 | ||
34869809 MS |
1557 | NLSOCK_LOCK(); |
1558 | retval = hash_lookup(nlsock_hash, &lookup); | |
1559 | NLSOCK_UNLOCK(); | |
1560 | ||
1561 | return retval; | |
1562 | } | |
1563 | ||
1564 | /* Insert nlsock entry into hash */ | |
1565 | static void kernel_netlink_nlsock_insert(struct nlsock *nls) | |
1566 | { | |
1567 | NLSOCK_LOCK(); | |
1568 | (void)hash_get(nlsock_hash, nls, hash_alloc_intern); | |
1569 | NLSOCK_UNLOCK(); | |
1570 | } | |
1571 | ||
1572 | /* Remove nlsock entry from hash */ | |
1573 | static void kernel_netlink_nlsock_remove(struct nlsock *nls) | |
1574 | { | |
1575 | NLSOCK_LOCK(); | |
1576 | (void)hash_release(nlsock_hash, nls); | |
1577 | NLSOCK_UNLOCK(); | |
d4000d7b DS |
1578 | } |
1579 | ||
1580 | static uint32_t kernel_netlink_nlsock_key(const void *arg) | |
1581 | { | |
1582 | const struct nlsock *nl = arg; | |
1583 | ||
1584 | return nl->sock; | |
1585 | } | |
1586 | ||
1587 | static bool kernel_netlink_nlsock_hash_equal(const void *arg1, const void *arg2) | |
1588 | { | |
1589 | const struct nlsock *nl1 = arg1; | |
1590 | const struct nlsock *nl2 = arg2; | |
1591 | ||
1592 | if (nl1->sock == nl2->sock) | |
1593 | return true; | |
1594 | ||
1595 | return false; | |
1596 | } | |
1597 | ||
1fdc9eae | 1598 | /* Exported interface function. This function simply calls |
1599 | netlink_socket (). */ | |
d62a17ae | 1600 | void kernel_init(struct zebra_ns *zns) |
1fdc9eae | 1601 | { |
ceab66b7 | 1602 | uint32_t groups, dplane_groups; |
5d307d5d DS |
1603 | #if defined SOL_NETLINK |
1604 | int one, ret; | |
1605 | #endif | |
d62a17ae | 1606 | |
026a316f DS |
1607 | /* |
1608 | * Initialize netlink sockets | |
1609 | * | |
1610 | * If RTMGRP_XXX exists use that, but at some point | |
1611 | * I think the kernel developers realized that | |
1612 | * keeping track of all the different values would | |
1613 | * lead to confusion, so we need to convert the | |
1614 | * RTNLGRP_XXX to a bit position for ourself | |
1615 | */ | |
1616 | groups = RTMGRP_LINK | | |
1617 | RTMGRP_IPV4_ROUTE | | |
1618 | RTMGRP_IPV4_IFADDR | | |
1619 | RTMGRP_IPV6_ROUTE | | |
1620 | RTMGRP_IPV6_IFADDR | | |
1621 | RTMGRP_IPV4_MROUTE | | |
1622 | RTMGRP_NEIGH | | |
67188ca2 QY |
1623 | ((uint32_t) 1 << (RTNLGRP_IPV4_RULE - 1)) | |
1624 | ((uint32_t) 1 << (RTNLGRP_IPV6_RULE - 1)) | | |
cd787a8a | 1625 | ((uint32_t) 1 << (RTNLGRP_NEXTHOP - 1)); |
d62a17ae | 1626 | |
ceab66b7 MS |
1627 | dplane_groups = (RTMGRP_LINK | |
1628 | RTMGRP_IPV4_IFADDR | | |
cd787a8a MS |
1629 | RTMGRP_IPV6_IFADDR | |
1630 | ((uint32_t) 1 << (RTNLGRP_IPV4_NETCONF - 1)) | | |
1631 | ((uint32_t) 1 << (RTNLGRP_IPV6_NETCONF - 1)) | | |
1632 | ((uint32_t) 1 << (RTNLGRP_MPLS_NETCONF - 1))); | |
1633 | ||
ceab66b7 | 1634 | |
d62a17ae | 1635 | snprintf(zns->netlink.name, sizeof(zns->netlink.name), |
1636 | "netlink-listen (NS %u)", zns->ns_id); | |
1637 | zns->netlink.sock = -1; | |
19d5a4fe DS |
1638 | if (netlink_socket(&zns->netlink, groups, zns->ns_id) < 0) { |
1639 | zlog_err("Failure to create %s socket", | |
1640 | zns->netlink.name); | |
1641 | exit(-1); | |
1642 | } | |
34869809 MS |
1643 | |
1644 | kernel_netlink_nlsock_insert(&zns->netlink); | |
d62a17ae | 1645 | |
1646 | snprintf(zns->netlink_cmd.name, sizeof(zns->netlink_cmd.name), | |
1647 | "netlink-cmd (NS %u)", zns->ns_id); | |
1648 | zns->netlink_cmd.sock = -1; | |
19d5a4fe DS |
1649 | if (netlink_socket(&zns->netlink_cmd, 0, zns->ns_id) < 0) { |
1650 | zlog_err("Failure to create %s socket", | |
1651 | zns->netlink_cmd.name); | |
1652 | exit(-1); | |
1653 | } | |
34869809 MS |
1654 | |
1655 | kernel_netlink_nlsock_insert(&zns->netlink_cmd); | |
d62a17ae | 1656 | |
80dcc388 MS |
1657 | /* Outbound socket for dplane programming of the host OS. */ |
1658 | snprintf(zns->netlink_dplane_out.name, | |
1659 | sizeof(zns->netlink_dplane_out.name), "netlink-dp (NS %u)", | |
1660 | zns->ns_id); | |
1661 | zns->netlink_dplane_out.sock = -1; | |
1662 | if (netlink_socket(&zns->netlink_dplane_out, 0, zns->ns_id) < 0) { | |
62b8bb7a | 1663 | zlog_err("Failure to create %s socket", |
80dcc388 MS |
1664 | zns->netlink_dplane_out.name); |
1665 | exit(-1); | |
1666 | } | |
34869809 MS |
1667 | |
1668 | kernel_netlink_nlsock_insert(&zns->netlink_dplane_out); | |
80dcc388 MS |
1669 | |
1670 | /* Inbound socket for OS events coming to the dplane. */ | |
1671 | snprintf(zns->netlink_dplane_in.name, | |
1672 | sizeof(zns->netlink_dplane_in.name), "netlink-dp-in (NS %u)", | |
1673 | zns->ns_id); | |
1674 | zns->netlink_dplane_in.sock = -1; | |
ceab66b7 MS |
1675 | if (netlink_socket(&zns->netlink_dplane_in, dplane_groups, |
1676 | zns->ns_id) < 0) { | |
80dcc388 MS |
1677 | zlog_err("Failure to create %s socket", |
1678 | zns->netlink_dplane_in.name); | |
62b8bb7a MS |
1679 | exit(-1); |
1680 | } | |
34869809 MS |
1681 | |
1682 | kernel_netlink_nlsock_insert(&zns->netlink_dplane_in); | |
62b8bb7a | 1683 | |
5d307d5d DS |
1684 | /* |
1685 | * SOL_NETLINK is not available on all platforms yet | |
1686 | * apparently. It's in bits/socket.h which I am not | |
1687 | * sure that we want to pull into our build system. | |
1688 | */ | |
1689 | #if defined SOL_NETLINK | |
1690 | /* | |
1691 | * Let's tell the kernel that we want to receive extended | |
62b8bb7a | 1692 | * ACKS over our command socket(s) |
5d307d5d DS |
1693 | */ |
1694 | one = 1; | |
1695 | ret = setsockopt(zns->netlink_cmd.sock, SOL_NETLINK, NETLINK_EXT_ACK, | |
1696 | &one, sizeof(one)); | |
1697 | ||
1698 | if (ret < 0) | |
62b8bb7a MS |
1699 | zlog_notice("Registration for extended cmd ACK failed : %d %s", |
1700 | errno, safe_strerror(errno)); | |
1701 | ||
1702 | one = 1; | |
80dcc388 MS |
1703 | ret = setsockopt(zns->netlink_dplane_out.sock, SOL_NETLINK, |
1704 | NETLINK_EXT_ACK, &one, sizeof(one)); | |
62b8bb7a MS |
1705 | |
1706 | if (ret < 0) | |
1707 | zlog_notice("Registration for extended dp ACK failed : %d %s", | |
5d307d5d | 1708 | errno, safe_strerror(errno)); |
97f85144 JU |
1709 | |
1710 | /* | |
1711 | * Trim off the payload of the original netlink message in the | |
1712 | * acknowledgment. This option is available since Linux 4.2, so if | |
1713 | * setsockopt fails, ignore the error. | |
1714 | */ | |
1715 | one = 1; | |
80dcc388 MS |
1716 | ret = setsockopt(zns->netlink_dplane_out.sock, SOL_NETLINK, |
1717 | NETLINK_CAP_ACK, &one, sizeof(one)); | |
9781e6a0 DS |
1718 | if (ret < 0) |
1719 | zlog_notice( | |
1720 | "Registration for reduced ACK packet size failed, probably running an early kernel"); | |
5d307d5d DS |
1721 | #endif |
1722 | ||
d62a17ae | 1723 | /* Register kernel socket. */ |
19d5a4fe | 1724 | if (fcntl(zns->netlink.sock, F_SETFL, O_NONBLOCK) < 0) |
450971aa | 1725 | flog_err_sys(EC_LIB_SOCKET, "Can't set %s socket flags: %s", |
09c866e3 | 1726 | zns->netlink.name, safe_strerror(errno)); |
8c85e8ea DS |
1727 | |
1728 | if (fcntl(zns->netlink_cmd.sock, F_SETFL, O_NONBLOCK) < 0) | |
1729 | zlog_err("Can't set %s socket error: %s(%d)", | |
1730 | zns->netlink_cmd.name, safe_strerror(errno), errno); | |
19d5a4fe | 1731 | |
80dcc388 | 1732 | if (fcntl(zns->netlink_dplane_out.sock, F_SETFL, O_NONBLOCK) < 0) |
62b8bb7a | 1733 | zlog_err("Can't set %s socket error: %s(%d)", |
80dcc388 MS |
1734 | zns->netlink_dplane_out.name, safe_strerror(errno), |
1735 | errno); | |
1736 | ||
1737 | if (fcntl(zns->netlink_dplane_in.sock, F_SETFL, O_NONBLOCK) < 0) | |
1738 | zlog_err("Can't set %s socket error: %s(%d)", | |
1739 | zns->netlink_dplane_in.name, safe_strerror(errno), | |
1740 | errno); | |
62b8bb7a | 1741 | |
19d5a4fe | 1742 | /* Set receive buffer size if it's set from command line */ |
9fb83b55 DS |
1743 | if (rcvbufsize) { |
1744 | netlink_recvbuf(&zns->netlink, rcvbufsize); | |
1745 | netlink_recvbuf(&zns->netlink_cmd, rcvbufsize); | |
1746 | netlink_recvbuf(&zns->netlink_dplane_out, rcvbufsize); | |
1747 | netlink_recvbuf(&zns->netlink_dplane_in, rcvbufsize); | |
97f85144 | 1748 | } |
19d5a4fe | 1749 | |
80dcc388 MS |
1750 | /* Set filter for inbound sockets, to exclude events we've generated |
1751 | * ourselves. | |
1752 | */ | |
1753 | netlink_install_filter(zns->netlink.sock, zns->netlink_cmd.snl.nl_pid, | |
1754 | zns->netlink_dplane_out.snl.nl_pid); | |
1755 | ||
1756 | netlink_install_filter(zns->netlink_dplane_in.sock, | |
62b8bb7a | 1757 | zns->netlink_cmd.snl.nl_pid, |
80dcc388 | 1758 | zns->netlink_dplane_out.snl.nl_pid); |
62b8bb7a | 1759 | |
19d5a4fe DS |
1760 | zns->t_netlink = NULL; |
1761 | ||
3801e764 | 1762 | thread_add_read(zrouter.master, kernel_read, zns, |
19d5a4fe | 1763 | zns->netlink.sock, &zns->t_netlink); |
d62a17ae | 1764 | |
1765 | rt_netlink_init(); | |
1fdc9eae | 1766 | } |
1767 | ||
34869809 MS |
1768 | /* Helper to clean up an nlsock */ |
1769 | static void kernel_nlsock_fini(struct nlsock *nls) | |
1770 | { | |
1771 | if (nls && nls->sock >= 0) { | |
1772 | kernel_netlink_nlsock_remove(nls); | |
1773 | close(nls->sock); | |
1774 | nls->sock = -1; | |
1775 | XFREE(MTYPE_NL_BUF, nls->buf); | |
1776 | nls->buflen = 0; | |
1777 | } | |
1778 | } | |
1779 | ||
62b8bb7a | 1780 | void kernel_terminate(struct zebra_ns *zns, bool complete) |
1fdc9eae | 1781 | { |
50478845 | 1782 | thread_cancel(&zns->t_netlink); |
d62a17ae | 1783 | |
34869809 | 1784 | kernel_nlsock_fini(&zns->netlink); |
d62a17ae | 1785 | |
34869809 | 1786 | kernel_nlsock_fini(&zns->netlink_cmd); |
ddfeb486 | 1787 | |
34869809 | 1788 | kernel_nlsock_fini(&zns->netlink_dplane_in); |
80dcc388 | 1789 | |
62b8bb7a MS |
1790 | /* During zebra shutdown, we need to leave the dataplane socket |
1791 | * around until all work is done. | |
1792 | */ | |
34869809 MS |
1793 | if (complete) |
1794 | kernel_nlsock_fini(&zns->netlink_dplane_out); | |
1795 | } | |
d4000d7b | 1796 | |
34869809 MS |
1797 | /* |
1798 | * Global init for platform-/OS-specific things | |
1799 | */ | |
1800 | void kernel_router_init(void) | |
1801 | { | |
1802 | /* Init nlsock hash and lock */ | |
1803 | pthread_mutex_init(&nlsock_mutex, NULL); | |
1804 | nlsock_hash = hash_create_size(8, kernel_netlink_nlsock_key, | |
1805 | kernel_netlink_nlsock_hash_equal, | |
1806 | "Netlink Socket Hash"); | |
1807 | } | |
1808 | ||
1809 | /* | |
1810 | * Global deinit for platform-/OS-specific things | |
1811 | */ | |
1812 | void kernel_router_terminate(void) | |
1813 | { | |
1814 | pthread_mutex_destroy(&nlsock_mutex); | |
1815 | ||
1816 | hash_free(nlsock_hash); | |
1817 | nlsock_hash = NULL; | |
62b8bb7a | 1818 | } |
34869809 | 1819 | |
ddfeb486 | 1820 | #endif /* HAVE_NETLINK */ |