1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Kernel communication using netlink interface.
3 * Copyright (C) 1999 Kunihiro Ishiguro
14 #include "connected.h"
23 #include "lib_errors.h"
26 #include "zebra/zebra_router.h"
27 #include "zebra/zebra_ns.h"
28 #include "zebra/zebra_vrf.h"
30 #include "zebra/debug.h"
31 #include "zebra/kernel_netlink.h"
32 #include "zebra/rt_netlink.h"
33 #include "zebra/if_netlink.h"
34 #include "zebra/rule_netlink.h"
35 #include "zebra/tc_netlink.h"
36 #include "zebra/netconf_netlink.h"
37 #include "zebra/zebra_errors.h"
39 #ifndef SO_RCVBUFFORCE
40 #define SO_RCVBUFFORCE (33)
43 /* Hack for GNU libc version 2. */
45 #define MSG_TRUNC 0x20
46 #endif /* MSG_TRUNC */
49 #define NLMSG_TAIL(nmsg) \
50 ((struct rtattr *)(((uint8_t *)(nmsg)) \
51 + NLMSG_ALIGN((nmsg)->nlmsg_len)))
55 #define RTA_TAIL(rta) \
56 ((struct rtattr *)(((uint8_t *)(rta)) + RTA_ALIGN((rta)->rta_len)))
59 #ifndef RTNL_FAMILY_IP6MR
60 #define RTNL_FAMILY_IP6MR 129
63 #ifndef RTPROT_MROUTED
64 #define RTPROT_MROUTED 17
67 #define NL_DEFAULT_BATCH_BUFSIZE (16 * NL_PKT_BUF_SIZE)
70 * We limit the batch's size to a number smaller than the length of the
71 * underlying buffer since the last message that wouldn't fit the batch would go
72 * over the upper boundary and then it would have to be encoded again into a new
73 * buffer. If the difference between the limit and the length of the buffer is
74 * big enough (bigger than the biggest Netlink message) then this situation
77 #define NL_DEFAULT_BATCH_SEND_THRESHOLD (15 * NL_PKT_BUF_SIZE)
79 static const struct message nlmsg_str
[] = {{RTM_NEWROUTE
, "RTM_NEWROUTE"},
80 {RTM_DELROUTE
, "RTM_DELROUTE"},
81 {RTM_GETROUTE
, "RTM_GETROUTE"},
82 {RTM_NEWLINK
, "RTM_NEWLINK"},
83 {RTM_SETLINK
, "RTM_SETLINK"},
84 {RTM_DELLINK
, "RTM_DELLINK"},
85 {RTM_GETLINK
, "RTM_GETLINK"},
86 {RTM_NEWADDR
, "RTM_NEWADDR"},
87 {RTM_DELADDR
, "RTM_DELADDR"},
88 {RTM_GETADDR
, "RTM_GETADDR"},
89 {RTM_NEWNEIGH
, "RTM_NEWNEIGH"},
90 {RTM_DELNEIGH
, "RTM_DELNEIGH"},
91 {RTM_GETNEIGH
, "RTM_GETNEIGH"},
92 {RTM_NEWRULE
, "RTM_NEWRULE"},
93 {RTM_DELRULE
, "RTM_DELRULE"},
94 {RTM_GETRULE
, "RTM_GETRULE"},
95 {RTM_NEWNEXTHOP
, "RTM_NEWNEXTHOP"},
96 {RTM_DELNEXTHOP
, "RTM_DELNEXTHOP"},
97 {RTM_GETNEXTHOP
, "RTM_GETNEXTHOP"},
98 {RTM_NEWNETCONF
, "RTM_NEWNETCONF"},
99 {RTM_DELNETCONF
, "RTM_DELNETCONF"},
100 {RTM_NEWTUNNEL
, "RTM_NEWTUNNEL"},
101 {RTM_DELTUNNEL
, "RTM_DELTUNNEL"},
102 {RTM_GETTUNNEL
, "RTM_GETTUNNEL"},
103 {RTM_NEWQDISC
, "RTM_NEWQDISC"},
104 {RTM_DELQDISC
, "RTM_DELQDISC"},
105 {RTM_GETQDISC
, "RTM_GETQDISC"},
106 {RTM_NEWTCLASS
, "RTM_NEWTCLASS"},
107 {RTM_DELTCLASS
, "RTM_DELTCLASS"},
108 {RTM_GETTCLASS
, "RTM_GETTCLASS"},
109 {RTM_NEWTFILTER
, "RTM_NEWTFILTER"},
110 {RTM_DELTFILTER
, "RTM_DELTFILTER"},
111 {RTM_GETTFILTER
, "RTM_GETTFILTER"},
112 {RTM_NEWVLAN
, "RTM_NEWVLAN"},
113 {RTM_DELVLAN
, "RTM_DELVLAN"},
114 {RTM_GETVLAN
, "RTM_GETVLAN"},
117 static const struct message rtproto_str
[] = {
118 {RTPROT_REDIRECT
, "redirect"},
119 {RTPROT_KERNEL
, "kernel"},
120 {RTPROT_BOOT
, "boot"},
121 {RTPROT_STATIC
, "static"},
122 {RTPROT_GATED
, "GateD"},
123 {RTPROT_RA
, "router advertisement"},
125 {RTPROT_ZEBRA
, "Zebra"},
127 {RTPROT_BIRD
, "BIRD"},
128 #endif /* RTPROT_BIRD */
129 {RTPROT_MROUTED
, "mroute"},
131 {RTPROT_OSPF
, "OSPF"},
132 {RTPROT_ISIS
, "IS-IS"},
134 {RTPROT_RIPNG
, "RIPNG"},
135 {RTPROT_ZSTATIC
, "static"},
138 static const struct message family_str
[] = {{AF_INET
, "ipv4"},
140 {AF_BRIDGE
, "bridge"},
141 {RTNL_FAMILY_IPMR
, "ipv4MR"},
142 {RTNL_FAMILY_IP6MR
, "ipv6MR"},
145 static const struct message rttype_str
[] = {{RTN_UNSPEC
, "none"},
146 {RTN_UNICAST
, "unicast"},
147 {RTN_LOCAL
, "local"},
148 {RTN_BROADCAST
, "broadcast"},
149 {RTN_ANYCAST
, "anycast"},
150 {RTN_MULTICAST
, "multicast"},
151 {RTN_BLACKHOLE
, "blackhole"},
152 {RTN_UNREACHABLE
, "unreachable"},
153 {RTN_PROHIBIT
, "prohibited"},
154 {RTN_THROW
, "throw"},
156 {RTN_XRESOLVE
, "resolver"},
159 extern struct event_loop
*master
;
161 extern struct zebra_privs_t zserv_privs
;
163 DEFINE_MTYPE_STATIC(ZEBRA
, NL_BUF
, "Zebra Netlink buffers");
165 /* Hashtable and mutex to allow lookup of nlsock structs by socket/fd value.
166 * We have both the main and dplane pthreads using these structs, so we have
167 * to protect the hash with a lock.
169 static struct hash
*nlsock_hash
;
170 pthread_mutex_t nlsock_mutex
;
172 /* Lock and unlock wrappers for nlsock hash */
173 #define NLSOCK_LOCK() pthread_mutex_lock(&nlsock_mutex)
174 #define NLSOCK_UNLOCK() pthread_mutex_unlock(&nlsock_mutex)
176 size_t nl_batch_tx_bufsize
;
177 char *nl_batch_tx_buf
;
179 _Atomic
uint32_t nl_batch_bufsize
= NL_DEFAULT_BATCH_BUFSIZE
;
180 _Atomic
uint32_t nl_batch_send_threshold
= NL_DEFAULT_BATCH_SEND_THRESHOLD
;
191 const struct zebra_dplane_info
*zns
;
193 struct dplane_ctx_list_head ctx_list
;
196 * Pointer to the queue of completed contexts outbound back
197 * towards the dataplane module.
199 struct dplane_ctx_list_head
*ctx_out_q
;
202 int netlink_config_write_helper(struct vty
*vty
)
205 atomic_load_explicit(&nl_batch_bufsize
, memory_order_relaxed
);
206 uint32_t threshold
= atomic_load_explicit(&nl_batch_send_threshold
,
207 memory_order_relaxed
);
209 if (size
!= NL_DEFAULT_BATCH_BUFSIZE
210 || threshold
!= NL_DEFAULT_BATCH_SEND_THRESHOLD
)
211 vty_out(vty
, "zebra kernel netlink batch-tx-buf %u %u\n", size
,
214 if (if_netlink_frr_protodown_r_bit_is_set())
215 vty_out(vty
, "zebra protodown reason-bit %u\n",
216 if_netlink_get_frr_protodown_r_bit());
221 void netlink_set_batch_buffer_size(uint32_t size
, uint32_t threshold
, bool set
)
224 size
= NL_DEFAULT_BATCH_BUFSIZE
;
225 threshold
= NL_DEFAULT_BATCH_SEND_THRESHOLD
;
228 atomic_store_explicit(&nl_batch_bufsize
, size
, memory_order_relaxed
);
229 atomic_store_explicit(&nl_batch_send_threshold
, threshold
,
230 memory_order_relaxed
);
233 int netlink_talk_filter(struct nlmsghdr
*h
, ns_id_t ns_id
, int startup
)
236 * This is an error condition that must be handled during
239 * The netlink_talk_filter function is used for communication
240 * down the netlink_cmd pipe and we are expecting
241 * an ack being received. So if we get here
242 * then we did not receive the ack and instead
243 * received some other message in an unexpected
246 zlog_debug("%s: ignoring message type 0x%04x(%s) NS %u", __func__
,
247 h
->nlmsg_type
, nl_msg_type_to_str(h
->nlmsg_type
), ns_id
);
251 static int netlink_recvbuf(struct nlsock
*nl
, uint32_t newsize
)
254 socklen_t newlen
= sizeof(newsize
);
255 socklen_t oldlen
= sizeof(oldsize
);
258 ret
= getsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUF
, &oldsize
, &oldlen
);
260 flog_err_sys(EC_LIB_SOCKET
,
261 "Can't get %s receive buffer size: %s", nl
->name
,
262 safe_strerror(errno
));
266 /* Try force option (linux >= 2.6.14) and fall back to normal set */
267 frr_with_privs(&zserv_privs
) {
268 ret
= setsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUFFORCE
,
269 &rcvbufsize
, sizeof(rcvbufsize
));
272 ret
= setsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUF
, &rcvbufsize
,
275 flog_err_sys(EC_LIB_SOCKET
,
276 "Can't set %s receive buffer size: %s", nl
->name
,
277 safe_strerror(errno
));
281 ret
= getsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUF
, &newsize
, &newlen
);
283 flog_err_sys(EC_LIB_SOCKET
,
284 "Can't get %s receive buffer size: %s", nl
->name
,
285 safe_strerror(errno
));
291 static const char *group2str(uint32_t group
)
295 return "RTNLGRP_TUNNEL";
301 /* Make socket for Linux netlink interface. */
302 static int netlink_socket(struct nlsock
*nl
, unsigned long groups
,
303 uint32_t ext_groups
[], uint8_t ext_group_size
,
307 struct sockaddr_nl snl
;
311 frr_with_privs(&zserv_privs
) {
312 sock
= ns_socket(AF_NETLINK
, SOCK_RAW
, NETLINK_ROUTE
, ns_id
);
314 zlog_err("Can't open %s socket: %s", nl
->name
,
315 safe_strerror(errno
));
319 memset(&snl
, 0, sizeof(snl
));
320 snl
.nl_family
= AF_NETLINK
;
321 snl
.nl_groups
= groups
;
323 if (ext_group_size
) {
326 for (i
= 0; i
< ext_group_size
; i
++) {
327 #if defined SOL_NETLINK
328 ret
= setsockopt(sock
, SOL_NETLINK
,
329 NETLINK_ADD_MEMBERSHIP
,
331 sizeof(ext_groups
[i
]));
334 "can't setsockopt NETLINK_ADD_MEMBERSHIP for group %s(%u), this linux kernel does not support it: %s(%d)",
335 group2str(ext_groups
[i
]),
337 safe_strerror(errno
), errno
);
341 "Unable to use NETLINK_ADD_MEMBERSHIP via SOL_NETLINK for %s(%u) since the linux kernel does not support the socket option",
342 group2str(ext_groups
[i
]),
348 /* Bind the socket to the netlink structure for anything. */
349 ret
= bind(sock
, (struct sockaddr
*)&snl
, sizeof(snl
));
353 zlog_err("Can't bind %s socket to group 0x%x: %s", nl
->name
,
354 snl
.nl_groups
, safe_strerror(errno
));
359 /* multiple netlink sockets will have different nl_pid */
360 namelen
= sizeof(snl
);
361 ret
= getsockname(sock
, (struct sockaddr
*)&snl
, (socklen_t
*)&namelen
);
362 if (ret
< 0 || namelen
!= sizeof(snl
)) {
363 flog_err_sys(EC_LIB_SOCKET
, "Can't get %s socket name: %s",
364 nl
->name
, safe_strerror(errno
));
371 nl
->buflen
= NL_RCV_PKT_BUF_SIZE
;
372 nl
->buf
= XMALLOC(MTYPE_NL_BUF
, nl
->buflen
);
378 * Dispatch an incoming netlink message; used by the zebra main pthread's
379 * netlink event reader.
381 static int netlink_information_fetch(struct nlmsghdr
*h
, ns_id_t ns_id
,
385 * When we handle new message types here
386 * because we are starting to install them
387 * then lets check the netlink_install_filter
388 * and see if we should add the corresponding
389 * allow through entry there.
390 * Probably not needed to do but please
393 switch (h
->nlmsg_type
) {
395 return netlink_route_change(h
, ns_id
, startup
);
397 return netlink_route_change(h
, ns_id
, startup
);
399 return netlink_link_change(h
, ns_id
, startup
);
401 return netlink_link_change(h
, ns_id
, startup
);
405 return netlink_neigh_change(h
, ns_id
);
407 return netlink_rule_change(h
, ns_id
, startup
);
409 return netlink_rule_change(h
, ns_id
, startup
);
411 return netlink_nexthop_change(h
, ns_id
, startup
);
413 return netlink_nexthop_change(h
, ns_id
, startup
);
416 return netlink_qdisc_change(h
, ns_id
, startup
);
419 return netlink_tclass_change(h
, ns_id
, startup
);
422 return netlink_tfilter_change(h
, ns_id
, startup
);
424 return netlink_vlan_change(h
, ns_id
, startup
);
426 return netlink_vlan_change(h
, ns_id
, startup
);
428 /* Messages handled in the dplane thread */
439 * If we have received this message then
440 * we have made a mistake during development
441 * and we need to write some code to handle
442 * this message type or not ask for
443 * it to be sent up to us
445 flog_err(EC_ZEBRA_UNKNOWN_NLMSG
,
446 "Unknown netlink nlmsg_type %s(%d) vrf %u",
447 nl_msg_type_to_str(h
->nlmsg_type
), h
->nlmsg_type
,
455 * Dispatch an incoming netlink message; used by the dataplane pthread's
456 * netlink event reader code.
458 static int dplane_netlink_information_fetch(struct nlmsghdr
*h
, ns_id_t ns_id
,
462 * Dispatch the incoming messages that the dplane pthread handles
464 switch (h
->nlmsg_type
) {
467 return netlink_interface_addr_dplane(h
, ns_id
, startup
);
471 return netlink_netconf_change(h
, ns_id
, startup
);
473 /* TODO -- other messages for the dplane socket and pthread */
485 static void kernel_read(struct event
*thread
)
487 struct zebra_ns
*zns
= (struct zebra_ns
*)EVENT_ARG(thread
);
488 struct zebra_dplane_info dp_info
;
490 /* Capture key info from ns struct */
491 zebra_dplane_info_from_zns(&dp_info
, zns
, false);
493 netlink_parse_info(netlink_information_fetch
, &zns
->netlink
, &dp_info
,
496 event_add_read(zrouter
.master
, kernel_read
, zns
, zns
->netlink
.sock
,
501 * Called by the dplane pthread to read incoming OS messages and dispatch them.
503 int kernel_dplane_read(struct zebra_dplane_info
*info
)
505 struct nlsock
*nl
= kernel_netlink_nlsock_lookup(info
->sock
);
507 netlink_parse_info(dplane_netlink_information_fetch
, nl
, info
, 5,
514 * Filter out messages from self that occur on listener socket,
515 * caused by our actions on the command socket(s)
517 * When we add new Netlink message types we probably
518 * do not need to add them here as that we are filtering
519 * on the routes we actually care to receive( which is rarer
520 * then the normal course of operations). We are intentionally
521 * allowing some messages from ourselves through
522 * ( I'm looking at you Interface based netlink messages )
523 * so that we only have to write one way to handle incoming
524 * address add/delete and xxxNETCONF changes.
526 static void netlink_install_filter(int sock
, uint32_t pid
, uint32_t dplane_pid
)
529 * BPF_JUMP instructions and where you jump to are based upon
530 * 0 as being the next statement. So count from 0. Writing
531 * this down because every time I look at this I have to
534 struct sock_filter filter
[] = {
537 * if (nlmsg_pid == pid ||
538 * nlmsg_pid == dplane_pid) {
539 * if (the incoming nlmsg_type ==
540 * RTM_NEWADDR || RTM_DELADDR || RTM_NEWNETCONF ||
546 * keep this netlink message
549 * 0: Load the nlmsg_pid into the BPF register
551 BPF_STMT(BPF_LD
| BPF_ABS
| BPF_W
,
552 offsetof(struct nlmsghdr
, nlmsg_pid
)),
556 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htonl(pid
), 1, 0),
558 * 2: Compare to dplane pid
560 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htonl(dplane_pid
), 0, 6),
562 * 3: Load the nlmsg_type into BPF register
564 BPF_STMT(BPF_LD
| BPF_ABS
| BPF_H
,
565 offsetof(struct nlmsghdr
, nlmsg_type
)),
567 * 4: Compare to RTM_NEWADDR
569 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htons(RTM_NEWADDR
), 4, 0),
571 * 5: Compare to RTM_DELADDR
573 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htons(RTM_DELADDR
), 3, 0),
575 * 6: Compare to RTM_NEWNETCONF
577 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htons(RTM_NEWNETCONF
), 2,
580 * 7: Compare to RTM_DELNETCONF
582 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htons(RTM_DELNETCONF
), 1,
585 * 8: This is the end state of we want to skip the
588 BPF_STMT(BPF_RET
| BPF_K
, 0),
589 /* 9: This is the end state of we want to keep
592 BPF_STMT(BPF_RET
| BPF_K
, 0xffff),
595 struct sock_fprog prog
= {
596 .len
= array_size(filter
), .filter
= filter
,
599 if (setsockopt(sock
, SOL_SOCKET
, SO_ATTACH_FILTER
, &prog
, sizeof(prog
))
601 flog_err_sys(EC_LIB_SOCKET
, "Can't install socket filter: %s",
602 safe_strerror(errno
));
605 void netlink_parse_rtattr_flags(struct rtattr
**tb
, int max
, struct rtattr
*rta
,
606 int len
, unsigned short flags
)
610 memset(tb
, 0, sizeof(struct rtattr
*) * (max
+ 1));
611 while (RTA_OK(rta
, len
)) {
612 type
= rta
->rta_type
& ~flags
;
613 if ((type
<= max
) && (!tb
[type
]))
615 rta
= RTA_NEXT(rta
, len
);
619 void netlink_parse_rtattr(struct rtattr
**tb
, int max
, struct rtattr
*rta
,
622 memset(tb
, 0, sizeof(struct rtattr
*) * (max
+ 1));
623 while (RTA_OK(rta
, len
)) {
624 if (rta
->rta_type
<= max
)
625 tb
[rta
->rta_type
] = rta
;
626 rta
= RTA_NEXT(rta
, len
);
631 * netlink_parse_rtattr_nested() - Parses a nested route attribute
632 * @tb: Pointer to array for storing rtattr in.
633 * @max: Max number to store.
634 * @rta: Pointer to rtattr to look for nested items in.
636 void netlink_parse_rtattr_nested(struct rtattr
**tb
, int max
,
639 netlink_parse_rtattr(tb
, max
, RTA_DATA(rta
), RTA_PAYLOAD(rta
));
642 bool nl_addraw_l(struct nlmsghdr
*n
, unsigned int maxlen
, const void *data
,
645 if (NLMSG_ALIGN(n
->nlmsg_len
) + NLMSG_ALIGN(len
) > maxlen
) {
646 zlog_err("ERROR message exceeded bound of %d", maxlen
);
650 memcpy(NLMSG_TAIL(n
), data
, len
);
651 memset((uint8_t *)NLMSG_TAIL(n
) + len
, 0, NLMSG_ALIGN(len
) - len
);
652 n
->nlmsg_len
= NLMSG_ALIGN(n
->nlmsg_len
) + NLMSG_ALIGN(len
);
657 bool nl_attr_put(struct nlmsghdr
*n
, unsigned int maxlen
, int type
,
658 const void *data
, unsigned int alen
)
663 len
= RTA_LENGTH(alen
);
665 if (NLMSG_ALIGN(n
->nlmsg_len
) + RTA_ALIGN(len
) > maxlen
)
668 rta
= (struct rtattr
*)(((char *)n
) + NLMSG_ALIGN(n
->nlmsg_len
));
669 rta
->rta_type
= type
;
673 memcpy(RTA_DATA(rta
), data
, alen
);
677 n
->nlmsg_len
= NLMSG_ALIGN(n
->nlmsg_len
) + RTA_ALIGN(len
);
682 bool nl_attr_put8(struct nlmsghdr
*n
, unsigned int maxlen
, int type
,
685 return nl_attr_put(n
, maxlen
, type
, &data
, sizeof(uint8_t));
688 bool nl_attr_put16(struct nlmsghdr
*n
, unsigned int maxlen
, int type
,
691 return nl_attr_put(n
, maxlen
, type
, &data
, sizeof(uint16_t));
694 bool nl_attr_put32(struct nlmsghdr
*n
, unsigned int maxlen
, int type
,
697 return nl_attr_put(n
, maxlen
, type
, &data
, sizeof(uint32_t));
700 bool nl_attr_put64(struct nlmsghdr
*n
, unsigned int maxlen
, int type
,
703 return nl_attr_put(n
, maxlen
, type
, &data
, sizeof(uint64_t));
706 struct rtattr
*nl_attr_nest(struct nlmsghdr
*n
, unsigned int maxlen
, int type
)
708 struct rtattr
*nest
= NLMSG_TAIL(n
);
710 if (!nl_attr_put(n
, maxlen
, type
, NULL
, 0))
713 nest
->rta_type
|= NLA_F_NESTED
;
717 int nl_attr_nest_end(struct nlmsghdr
*n
, struct rtattr
*nest
)
719 nest
->rta_len
= (uint8_t *)NLMSG_TAIL(n
) - (uint8_t *)nest
;
723 struct rtnexthop
*nl_attr_rtnh(struct nlmsghdr
*n
, unsigned int maxlen
)
725 struct rtnexthop
*rtnh
= (struct rtnexthop
*)NLMSG_TAIL(n
);
727 if (NLMSG_ALIGN(n
->nlmsg_len
) + RTNH_ALIGN(sizeof(struct rtnexthop
))
731 memset(rtnh
, 0, sizeof(struct rtnexthop
));
733 NLMSG_ALIGN(n
->nlmsg_len
) + RTA_ALIGN(sizeof(struct rtnexthop
));
738 void nl_attr_rtnh_end(struct nlmsghdr
*n
, struct rtnexthop
*rtnh
)
740 rtnh
->rtnh_len
= (uint8_t *)NLMSG_TAIL(n
) - (uint8_t *)rtnh
;
743 bool nl_rta_put(struct rtattr
*rta
, unsigned int maxlen
, int type
,
744 const void *data
, int alen
)
746 struct rtattr
*subrta
;
747 int len
= RTA_LENGTH(alen
);
749 if (RTA_ALIGN(rta
->rta_len
) + RTA_ALIGN(len
) > maxlen
) {
750 zlog_err("ERROR max allowed bound %d exceeded for rtattr",
754 subrta
= (struct rtattr
*)(((char *)rta
) + RTA_ALIGN(rta
->rta_len
));
755 subrta
->rta_type
= type
;
756 subrta
->rta_len
= len
;
758 memcpy(RTA_DATA(subrta
), data
, alen
);
759 rta
->rta_len
= NLMSG_ALIGN(rta
->rta_len
) + RTA_ALIGN(len
);
764 bool nl_rta_put16(struct rtattr
*rta
, unsigned int maxlen
, int type
,
767 return nl_rta_put(rta
, maxlen
, type
, &data
, sizeof(uint16_t));
770 bool nl_rta_put64(struct rtattr
*rta
, unsigned int maxlen
, int type
,
773 return nl_rta_put(rta
, maxlen
, type
, &data
, sizeof(uint64_t));
776 struct rtattr
*nl_rta_nest(struct rtattr
*rta
, unsigned int maxlen
, int type
)
778 struct rtattr
*nest
= RTA_TAIL(rta
);
780 if (nl_rta_put(rta
, maxlen
, type
, NULL
, 0))
783 nest
->rta_type
|= NLA_F_NESTED
;
788 int nl_rta_nest_end(struct rtattr
*rta
, struct rtattr
*nest
)
790 nest
->rta_len
= (uint8_t *)RTA_TAIL(rta
) - (uint8_t *)nest
;
795 const char *nl_msg_type_to_str(uint16_t msg_type
)
797 return lookup_msg(nlmsg_str
, msg_type
, "");
800 const char *nl_rtproto_to_str(uint8_t rtproto
)
802 return lookup_msg(rtproto_str
, rtproto
, "");
805 const char *nl_family_to_str(uint8_t family
)
807 return lookup_msg(family_str
, family
, "");
810 const char *nl_rttype_to_str(uint8_t rttype
)
812 return lookup_msg(rttype_str
, rttype
, "");
815 #define NLA_OK(nla, len) \
816 ((len) >= (int)sizeof(struct nlattr) \
817 && (nla)->nla_len >= sizeof(struct nlattr) \
818 && (nla)->nla_len <= (len))
819 #define NLA_NEXT(nla, attrlen) \
820 ((attrlen) -= NLA_ALIGN((nla)->nla_len), \
821 (struct nlattr *)(((char *)(nla)) + NLA_ALIGN((nla)->nla_len)))
822 #define NLA_LENGTH(len) (NLA_ALIGN(sizeof(struct nlattr)) + (len))
823 #define NLA_DATA(nla) ((struct nlattr *)(((char *)(nla)) + NLA_LENGTH(0)))
825 #define ERR_NLA(err, inner_len) \
826 ((struct nlattr *)(((char *)(err)) \
827 + NLMSG_ALIGN(sizeof(struct nlmsgerr)) \
828 + NLMSG_ALIGN((inner_len))))
830 static void netlink_parse_nlattr(struct nlattr
**tb
, int max
,
831 struct nlattr
*nla
, int len
)
833 while (NLA_OK(nla
, len
)) {
834 if (nla
->nla_type
<= max
)
835 tb
[nla
->nla_type
] = nla
;
836 nla
= NLA_NEXT(nla
, len
);
840 static void netlink_parse_extended_ack(struct nlmsghdr
*h
)
842 struct nlattr
*tb
[NLMSGERR_ATTR_MAX
+ 1] = {};
843 const struct nlmsgerr
*err
= (const struct nlmsgerr
*)NLMSG_DATA(h
);
844 const struct nlmsghdr
*err_nlh
= NULL
;
845 /* Length not including nlmsghdr */
847 /* Inner error netlink message length */
848 uint32_t inner_len
= 0;
849 const char *msg
= NULL
;
852 if (!(h
->nlmsg_flags
& NLM_F_CAPPED
))
853 inner_len
= (uint32_t)NLMSG_PAYLOAD(&err
->msg
, 0);
855 len
= (uint32_t)(NLMSG_PAYLOAD(h
, sizeof(struct nlmsgerr
)) - inner_len
);
857 netlink_parse_nlattr(tb
, NLMSGERR_ATTR_MAX
, ERR_NLA(err
, inner_len
),
860 if (tb
[NLMSGERR_ATTR_MSG
])
861 msg
= (const char *)NLA_DATA(tb
[NLMSGERR_ATTR_MSG
]);
863 if (tb
[NLMSGERR_ATTR_OFFS
]) {
864 off
= *(uint32_t *)NLA_DATA(tb
[NLMSGERR_ATTR_OFFS
]);
866 if (off
> h
->nlmsg_len
) {
867 zlog_err("Invalid offset for NLMSGERR_ATTR_OFFS");
868 } else if (!(h
->nlmsg_flags
& NLM_F_CAPPED
)) {
870 * Header of failed message
871 * we are not doing anything currently with it
872 * but noticing it for later.
875 zlog_debug("%s: Received %s extended Ack", __func__
,
876 nl_msg_type_to_str(err_nlh
->nlmsg_type
));
880 if (msg
&& *msg
!= '\0') {
881 bool is_err
= !!err
->error
;
884 zlog_err("Extended Error: %s", msg
);
886 flog_warn(EC_ZEBRA_NETLINK_EXTENDED_WARNING
,
887 "Extended Warning: %s", msg
);
892 * netlink_send_msg - send a netlink message of a certain size.
894 * Returns -1 on error. Otherwise, it returns the number of bytes sent.
896 static ssize_t
netlink_send_msg(const struct nlsock
*nl
, void *buf
,
899 struct sockaddr_nl snl
= {};
900 struct iovec iov
= {};
901 struct msghdr msg
= {};
906 iov
.iov_len
= buflen
;
908 msg
.msg_namelen
= sizeof(snl
);
912 snl
.nl_family
= AF_NETLINK
;
914 /* Send message to netlink interface. */
915 frr_with_privs(&zserv_privs
) {
916 status
= sendmsg(nl
->sock
, &msg
, 0);
920 if (IS_ZEBRA_DEBUG_KERNEL_MSGDUMP_SEND
) {
921 zlog_debug("%s: >> netlink message dump [sent]", __func__
);
923 nl_dump(buf
, buflen
);
925 zlog_hexdump(buf
, buflen
);
926 #endif /* NETLINK_DEBUG */
930 flog_err_sys(EC_LIB_SOCKET
, "%s error: %s", __func__
,
931 safe_strerror(save_errno
));
939 * netlink_recv_msg - receive a netlink message.
941 * Returns -1 on error, 0 if read would block or the number of bytes received.
943 static int netlink_recv_msg(struct nlsock
*nl
, struct msghdr
*msg
)
948 iov
.iov_base
= nl
->buf
;
949 iov
.iov_len
= nl
->buflen
;
956 bytes
= recv(nl
->sock
, NULL
, 0, MSG_PEEK
| MSG_TRUNC
);
958 if (bytes
>= 0 && (size_t)bytes
> nl
->buflen
) {
959 nl
->buf
= XREALLOC(MTYPE_NL_BUF
, nl
->buf
, bytes
);
961 iov
.iov_base
= nl
->buf
;
962 iov
.iov_len
= nl
->buflen
;
965 status
= recvmsg(nl
->sock
, msg
, 0);
966 } while (status
== -1 && errno
== EINTR
);
969 if (errno
== EWOULDBLOCK
|| errno
== EAGAIN
)
971 flog_err(EC_ZEBRA_RECVMSG_OVERRUN
, "%s recvmsg overrun: %s",
972 nl
->name
, safe_strerror(errno
));
974 * In this case we are screwed. There is no good way to recover
975 * zebra at this point.
981 flog_err_sys(EC_LIB_SOCKET
, "%s EOF", nl
->name
);
985 if (msg
->msg_namelen
!= sizeof(struct sockaddr_nl
)) {
986 flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR
,
987 "%s sender address length error: length %d", nl
->name
,
992 if (IS_ZEBRA_DEBUG_KERNEL_MSGDUMP_RECV
) {
993 zlog_debug("%s: << netlink message dump [recv]", __func__
);
995 nl_dump(nl
->buf
, status
);
997 zlog_hexdump(nl
->buf
, status
);
998 #endif /* NETLINK_DEBUG */
1005 * netlink_parse_error - parse a netlink error message
1007 * Returns 1 if this message is acknowledgement, 0 if this error should be
1008 * ignored, -1 otherwise.
1010 static int netlink_parse_error(const struct nlsock
*nl
, struct nlmsghdr
*h
,
1011 bool is_cmd
, bool startup
)
1013 struct nlmsgerr
*err
= (struct nlmsgerr
*)NLMSG_DATA(h
);
1014 int errnum
= err
->error
;
1015 int msg_type
= err
->msg
.nlmsg_type
;
1017 if (h
->nlmsg_len
< NLMSG_LENGTH(sizeof(struct nlmsgerr
))) {
1018 flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR
,
1019 "%s error: message truncated", nl
->name
);
1024 * Parse the extended information before we actually handle it. At this
1025 * point in time we do not do anything other than report the issue.
1027 if (h
->nlmsg_flags
& NLM_F_ACK_TLVS
)
1028 netlink_parse_extended_ack(h
);
1030 /* If the error field is zero, then this is an ACK. */
1031 if (err
->error
== 0) {
1032 if (IS_ZEBRA_DEBUG_KERNEL
) {
1033 zlog_debug("%s: %s ACK: type=%s(%u), seq=%u, pid=%u",
1035 nl_msg_type_to_str(err
->msg
.nlmsg_type
),
1036 err
->msg
.nlmsg_type
, err
->msg
.nlmsg_seq
,
1037 err
->msg
.nlmsg_pid
);
1044 * Deal with errors that occur because of races in link handling
1045 * or types are not supported in kernel.
1048 ((msg_type
== RTM_DELROUTE
&&
1049 (-errnum
== ENODEV
|| -errnum
== ESRCH
)) ||
1050 (msg_type
== RTM_NEWROUTE
&&
1051 (-errnum
== ENETDOWN
|| -errnum
== EEXIST
)) ||
1052 ((msg_type
== RTM_NEWTUNNEL
|| msg_type
== RTM_DELTUNNEL
||
1053 msg_type
== RTM_GETTUNNEL
) &&
1054 (-errnum
== EOPNOTSUPP
)))) {
1055 if (IS_ZEBRA_DEBUG_KERNEL
)
1056 zlog_debug("%s: error: %s type=%s(%u), seq=%u, pid=%u",
1057 nl
->name
, safe_strerror(-errnum
),
1058 nl_msg_type_to_str(msg_type
), msg_type
,
1059 err
->msg
.nlmsg_seq
, err
->msg
.nlmsg_pid
);
1064 * We see RTM_DELNEIGH when shutting down an interface with an IPv4
1065 * link-local. The kernel should have already deleted the neighbor so
1066 * do not log these as an error.
1068 if (msg_type
== RTM_DELNEIGH
1069 || (is_cmd
&& msg_type
== RTM_NEWROUTE
1070 && (-errnum
== ESRCH
|| -errnum
== ENETUNREACH
))) {
1072 * This is known to happen in some situations, don't log as
1075 if (IS_ZEBRA_DEBUG_KERNEL
)
1076 zlog_debug("%s error: %s, type=%s(%u), seq=%u, pid=%u",
1077 nl
->name
, safe_strerror(-errnum
),
1078 nl_msg_type_to_str(msg_type
), msg_type
,
1079 err
->msg
.nlmsg_seq
, err
->msg
.nlmsg_pid
);
1081 if ((msg_type
!= RTM_GETNEXTHOP
&& msg_type
!= RTM_GETVLAN
) ||
1083 flog_err(EC_ZEBRA_UNEXPECTED_MESSAGE
,
1084 "%s error: %s, type=%s(%u), seq=%u, pid=%u",
1085 nl
->name
, safe_strerror(-errnum
),
1086 nl_msg_type_to_str(msg_type
), msg_type
,
1087 err
->msg
.nlmsg_seq
, err
->msg
.nlmsg_pid
);
1094 * netlink_parse_info
1096 * Receive message from netlink interface and pass those information
1097 * to the given function.
1099 * filter -> Function to call to read the results
1100 * nl -> netlink socket information
1101 * zns -> The zebra namespace data
1102 * count -> How many we should read in, 0 means as much as possible
1103 * startup -> Are we reading in under startup conditions? passed to
1106 int netlink_parse_info(int (*filter
)(struct nlmsghdr
*, ns_id_t
, int),
1107 struct nlsock
*nl
, const struct zebra_dplane_info
*zns
,
1108 int count
, bool startup
)
1116 struct sockaddr_nl snl
;
1117 struct msghdr msg
= {.msg_name
= (void *)&snl
,
1118 .msg_namelen
= sizeof(snl
)};
1121 if (count
&& read_in
>= count
)
1124 status
= netlink_recv_msg(nl
, &msg
);
1127 else if (status
== 0)
1131 for (h
= (struct nlmsghdr
*)nl
->buf
;
1132 (status
>= 0 && NLMSG_OK(h
, (unsigned int)status
));
1133 h
= NLMSG_NEXT(h
, status
)) {
1134 /* Finish of reading. */
1135 if (h
->nlmsg_type
== NLMSG_DONE
)
1138 /* Error handling. */
1139 if (h
->nlmsg_type
== NLMSG_ERROR
) {
1140 int err
= netlink_parse_error(
1141 nl
, h
, zns
->is_cmd
, startup
);
1144 if (!(h
->nlmsg_flags
& NLM_F_MULTI
))
1152 * What is the right thing to do? The kernel
1153 * is telling us that the dump request was interrupted
1154 * and we more than likely are out of luck and have
1155 * missed data from the kernel. At this point in time
1156 * lets just note that this is happening.
1158 if (h
->nlmsg_flags
& NLM_F_DUMP_INTR
)
1160 EC_ZEBRA_NETLINK_BAD_SEQUENCE
,
1161 "netlink recvmsg: The Dump request was interrupted");
1163 /* OK we got netlink message. */
1164 if (IS_ZEBRA_DEBUG_KERNEL
)
1166 "%s: %s type %s(%u), len=%d, seq=%u, pid=%u",
1168 nl_msg_type_to_str(h
->nlmsg_type
),
1169 h
->nlmsg_type
, h
->nlmsg_len
,
1170 h
->nlmsg_seq
, h
->nlmsg_pid
);
1174 * Ignore messages that maybe sent from
1175 * other actors besides the kernel
1177 if (snl
.nl_pid
!= 0) {
1178 zlog_debug("Ignoring message from pid %u",
1183 error
= (*filter
)(h
, zns
->ns_id
, startup
);
1185 zlog_debug("%s filter function error",
1191 /* After error care. */
1192 if (msg
.msg_flags
& MSG_TRUNC
) {
1193 flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR
,
1194 "%s error: message truncated", nl
->name
);
1198 flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR
,
1199 "%s error: data remnant size %d", nl
->name
,
1210 * sendmsg() to netlink socket then recvmsg().
1211 * Calls netlink_parse_info to parse returned data
1213 * filter -> The filter to read final results from kernel
1214 * nlmsghdr -> The data to send to the kernel
1215 * dp_info -> The dataplane and netlink socket information
1216 * startup -> Are we reading in under startup conditions
1217 * This is passed through eventually to filter.
1219 static int netlink_talk_info(int (*filter
)(struct nlmsghdr
*, ns_id_t
,
1222 struct zebra_dplane_info
*dp_info
, bool startup
)
1226 nl
= kernel_netlink_nlsock_lookup(dp_info
->sock
);
1227 n
->nlmsg_seq
= dp_info
->seq
;
1228 n
->nlmsg_pid
= nl
->snl
.nl_pid
;
1230 if (IS_ZEBRA_DEBUG_KERNEL
)
1232 "netlink_talk: %s type %s(%u), len=%d seq=%u flags 0x%x",
1233 nl
->name
, nl_msg_type_to_str(n
->nlmsg_type
),
1234 n
->nlmsg_type
, n
->nlmsg_len
, n
->nlmsg_seq
,
1237 if (netlink_send_msg(nl
, n
, n
->nlmsg_len
) == -1)
1241 * Get reply from netlink socket.
1242 * The reply should either be an acknowlegement or an error.
1244 return netlink_parse_info(filter
, nl
, dp_info
, 0, startup
);
1248 * Synchronous version of netlink_talk_info. Converts args to suit the
1249 * common version, which is suitable for both sync and async use.
1251 int netlink_talk(int (*filter
)(struct nlmsghdr
*, ns_id_t
, int startup
),
1252 struct nlmsghdr
*n
, struct nlsock
*nl
, struct zebra_ns
*zns
,
1255 struct zebra_dplane_info dp_info
;
1257 /* Increment sequence number before capturing snapshot of ns socket
1262 /* Capture info in intermediate info struct */
1263 zebra_dplane_info_from_zns(&dp_info
, zns
, (nl
== &(zns
->netlink_cmd
)));
1265 return netlink_talk_info(filter
, n
, &dp_info
, startup
);
1268 /* Issue request message to kernel via netlink socket. GET messages
1269 * are issued through this interface.
1271 int netlink_request(struct nlsock
*nl
, void *req
)
1273 struct nlmsghdr
*n
= (struct nlmsghdr
*)req
;
1275 /* Check netlink socket. */
1277 flog_err_sys(EC_LIB_SOCKET
, "%s socket isn't active.",
1282 /* Fill common fields for all requests. */
1283 n
->nlmsg_pid
= nl
->snl
.nl_pid
;
1284 n
->nlmsg_seq
= ++nl
->seq
;
1286 if (netlink_send_msg(nl
, req
, n
->nlmsg_len
) == -1)
1292 static int nl_batch_read_resp(struct nl_batch
*bth
)
1295 struct sockaddr_nl snl
;
1296 struct msghdr msg
= {};
1299 struct zebra_dplane_ctx
*ctx
;
1302 nl
= kernel_netlink_nlsock_lookup(bth
->zns
->sock
);
1304 msg
.msg_name
= (void *)&snl
;
1305 msg
.msg_namelen
= sizeof(snl
);
1308 * The responses are not batched, so we need to read and process one
1309 * message at a time.
1312 status
= netlink_recv_msg(nl
, &msg
);
1314 * status == -1 is a full on failure somewhere
1315 * since we don't know where the problem happened
1316 * we must mark all as failed
1318 * Else we mark everything as worked
1321 if (status
== -1 || status
== 0) {
1322 while ((ctx
= dplane_ctx_dequeue(&(bth
->ctx_list
))) !=
1325 dplane_ctx_set_status(
1327 ZEBRA_DPLANE_REQUEST_FAILURE
);
1328 dplane_ctx_enqueue_tail(bth
->ctx_out_q
, ctx
);
1333 h
= (struct nlmsghdr
*)nl
->buf
;
1337 * Find the corresponding context object. Received responses are
1338 * in the same order as requests we sent, so we can simply
1339 * iterate over the context list and match responses with
1340 * requests at same time.
1343 ctx
= dplane_ctx_get_head(&(bth
->ctx_list
));
1346 * This is a situation where we have gotten
1347 * into a bad spot. We need to know that
1348 * this happens( does it? )
1351 "%s:WARNING Received netlink Response for an error and no Contexts to associate with it",
1357 * 'update' context objects take two consecutive
1360 if (dplane_ctx_is_update(ctx
) &&
1361 dplane_ctx_get_ns(ctx
)->seq
+ 1 == seq
) {
1363 * This is the situation where we get a response
1364 * to a message that should be ignored.
1370 ctx
= dplane_ctx_dequeue(&(bth
->ctx_list
));
1371 dplane_ctx_enqueue_tail(bth
->ctx_out_q
, ctx
);
1373 /* We have found corresponding context object. */
1374 if (dplane_ctx_get_ns(ctx
)->seq
== seq
)
1377 if (dplane_ctx_get_ns(ctx
)->seq
> seq
)
1379 "%s:WARNING Received %u is less than any context on the queue ctx->seq %u",
1381 dplane_ctx_get_ns(ctx
)->seq
);
1386 * If we ignore the message due to an update
1387 * above we should still fricking decode the
1388 * message for our operator to understand
1391 int err
= netlink_parse_error(nl
, h
, bth
->zns
->is_cmd
,
1394 zlog_debug("%s: netlink error message seq=%d %d",
1395 __func__
, h
->nlmsg_seq
, err
);
1400 * We received a message with the sequence number that isn't
1401 * associated with any dplane context object.
1404 if (IS_ZEBRA_DEBUG_KERNEL
)
1406 "%s: skipping unassociated response, seq number %d NS %u",
1407 __func__
, h
->nlmsg_seq
,
1412 if (h
->nlmsg_type
== NLMSG_ERROR
) {
1413 int err
= netlink_parse_error(nl
, h
, bth
->zns
->is_cmd
,
1417 dplane_ctx_set_status(
1418 ctx
, ZEBRA_DPLANE_REQUEST_FAILURE
);
1420 if (IS_ZEBRA_DEBUG_KERNEL
)
1421 zlog_debug("%s: netlink error message seq=%d ",
1422 __func__
, h
->nlmsg_seq
);
1427 * If we get here then we did not receive neither the ack nor
1428 * the error and instead received some other message in an
1431 if (IS_ZEBRA_DEBUG_KERNEL
)
1432 zlog_debug("%s: ignoring message type 0x%04x(%s) NS %u",
1433 __func__
, h
->nlmsg_type
,
1434 nl_msg_type_to_str(h
->nlmsg_type
),
1441 static void nl_batch_reset(struct nl_batch
*bth
)
1443 bth
->buf_head
= bth
->buf
;
1448 dplane_ctx_q_init(&(bth
->ctx_list
));
1451 static void nl_batch_init(struct nl_batch
*bth
,
1452 struct dplane_ctx_list_head
*ctx_out_q
)
1455 * If the size of the buffer has changed, free and then allocate a new
1459 atomic_load_explicit(&nl_batch_bufsize
, memory_order_relaxed
);
1460 if (bufsize
!= nl_batch_tx_bufsize
) {
1461 if (nl_batch_tx_buf
)
1462 XFREE(MTYPE_NL_BUF
, nl_batch_tx_buf
);
1464 nl_batch_tx_buf
= XCALLOC(MTYPE_NL_BUF
, bufsize
);
1465 nl_batch_tx_bufsize
= bufsize
;
1468 bth
->buf
= nl_batch_tx_buf
;
1469 bth
->bufsiz
= bufsize
;
1470 bth
->limit
= atomic_load_explicit(&nl_batch_send_threshold
,
1471 memory_order_relaxed
);
1473 bth
->ctx_out_q
= ctx_out_q
;
1475 nl_batch_reset(bth
);
1478 static void nl_batch_send(struct nl_batch
*bth
)
1480 struct zebra_dplane_ctx
*ctx
;
1483 if (bth
->curlen
!= 0 && bth
->zns
!= NULL
) {
1485 kernel_netlink_nlsock_lookup(bth
->zns
->sock
);
1487 if (IS_ZEBRA_DEBUG_KERNEL
)
1488 zlog_debug("%s: %s, batch size=%zu, msg cnt=%zu",
1489 __func__
, nl
->name
, bth
->curlen
,
1492 if (netlink_send_msg(nl
, bth
->buf
, bth
->curlen
) == -1)
1496 if (nl_batch_read_resp(bth
) == -1)
1501 /* Move remaining contexts to the outbound queue. */
1503 ctx
= dplane_ctx_dequeue(&(bth
->ctx_list
));
1508 dplane_ctx_set_status(ctx
,
1509 ZEBRA_DPLANE_REQUEST_FAILURE
);
1511 dplane_ctx_enqueue_tail(bth
->ctx_out_q
, ctx
);
1514 nl_batch_reset(bth
);
1517 enum netlink_msg_status
netlink_batch_add_msg(
1518 struct nl_batch
*bth
, struct zebra_dplane_ctx
*ctx
,
1519 ssize_t (*msg_encoder
)(struct zebra_dplane_ctx
*, void *, size_t),
1524 struct nlmsghdr
*msgh
;
1527 size
= (*msg_encoder
)(ctx
, bth
->buf_head
, bth
->bufsiz
- bth
->curlen
);
1530 * If there was an error while encoding the message (other than buffer
1531 * overflow) then return an error.
1534 return FRR_NETLINK_ERROR
;
1537 * If the message doesn't fit entirely in the buffer then send the batch
1542 size
= (*msg_encoder
)(ctx
, bth
->buf_head
,
1543 bth
->bufsiz
- bth
->curlen
);
1545 * If the message doesn't fit in the empty buffer then just
1549 return FRR_NETLINK_ERROR
;
1552 seq
= dplane_ctx_get_ns(ctx
)->seq
;
1553 nl
= kernel_netlink_nlsock_lookup(dplane_ctx_get_ns_sock(ctx
));
1558 msgh
= (struct nlmsghdr
*)bth
->buf_head
;
1559 msgh
->nlmsg_seq
= seq
;
1560 msgh
->nlmsg_pid
= nl
->snl
.nl_pid
;
1562 bth
->zns
= dplane_ctx_get_ns(ctx
);
1563 bth
->buf_head
= ((char *)bth
->buf_head
) + size
;
1564 bth
->curlen
+= size
;
1567 return FRR_NETLINK_QUEUED
;
1570 static enum netlink_msg_status
nl_put_msg(struct nl_batch
*bth
,
1571 struct zebra_dplane_ctx
*ctx
)
1573 if (dplane_ctx_is_skip_kernel(ctx
))
1574 return FRR_NETLINK_SUCCESS
;
1576 switch (dplane_ctx_get_op(ctx
)) {
1578 case DPLANE_OP_ROUTE_INSTALL
:
1579 case DPLANE_OP_ROUTE_UPDATE
:
1580 case DPLANE_OP_ROUTE_DELETE
:
1581 return netlink_put_route_update_msg(bth
, ctx
);
1583 case DPLANE_OP_NH_INSTALL
:
1584 case DPLANE_OP_NH_UPDATE
:
1585 case DPLANE_OP_NH_DELETE
:
1586 return netlink_put_nexthop_update_msg(bth
, ctx
);
1588 case DPLANE_OP_LSP_INSTALL
:
1589 case DPLANE_OP_LSP_UPDATE
:
1590 case DPLANE_OP_LSP_DELETE
:
1591 return netlink_put_lsp_update_msg(bth
, ctx
);
1593 case DPLANE_OP_PW_INSTALL
:
1594 case DPLANE_OP_PW_UNINSTALL
:
1595 return netlink_put_pw_update_msg(bth
, ctx
);
1597 case DPLANE_OP_ADDR_INSTALL
:
1598 case DPLANE_OP_ADDR_UNINSTALL
:
1599 return netlink_put_address_update_msg(bth
, ctx
);
1601 case DPLANE_OP_MAC_INSTALL
:
1602 case DPLANE_OP_MAC_DELETE
:
1603 return netlink_put_mac_update_msg(bth
, ctx
);
1605 case DPLANE_OP_NEIGH_INSTALL
:
1606 case DPLANE_OP_NEIGH_UPDATE
:
1607 case DPLANE_OP_NEIGH_DELETE
:
1608 case DPLANE_OP_VTEP_ADD
:
1609 case DPLANE_OP_VTEP_DELETE
:
1610 case DPLANE_OP_NEIGH_DISCOVER
:
1611 case DPLANE_OP_NEIGH_IP_INSTALL
:
1612 case DPLANE_OP_NEIGH_IP_DELETE
:
1613 case DPLANE_OP_NEIGH_TABLE_UPDATE
:
1614 return netlink_put_neigh_update_msg(bth
, ctx
);
1616 case DPLANE_OP_RULE_ADD
:
1617 case DPLANE_OP_RULE_DELETE
:
1618 case DPLANE_OP_RULE_UPDATE
:
1619 return netlink_put_rule_update_msg(bth
, ctx
);
1621 case DPLANE_OP_SYS_ROUTE_ADD
:
1622 case DPLANE_OP_SYS_ROUTE_DELETE
:
1623 case DPLANE_OP_ROUTE_NOTIFY
:
1624 case DPLANE_OP_LSP_NOTIFY
:
1625 case DPLANE_OP_BR_PORT_UPDATE
:
1626 return FRR_NETLINK_SUCCESS
;
1628 case DPLANE_OP_IPTABLE_ADD
:
1629 case DPLANE_OP_IPTABLE_DELETE
:
1630 case DPLANE_OP_IPSET_ADD
:
1631 case DPLANE_OP_IPSET_DELETE
:
1632 case DPLANE_OP_IPSET_ENTRY_ADD
:
1633 case DPLANE_OP_IPSET_ENTRY_DELETE
:
1634 return FRR_NETLINK_ERROR
;
1636 case DPLANE_OP_GRE_SET
:
1637 return netlink_put_gre_set_msg(bth
, ctx
);
1639 case DPLANE_OP_INTF_ADDR_ADD
:
1640 case DPLANE_OP_INTF_ADDR_DEL
:
1641 case DPLANE_OP_NONE
:
1642 return FRR_NETLINK_ERROR
;
1644 case DPLANE_OP_INTF_NETCONFIG
:
1645 return netlink_put_intf_netconfig(bth
, ctx
);
1647 case DPLANE_OP_INTF_INSTALL
:
1648 case DPLANE_OP_INTF_UPDATE
:
1649 case DPLANE_OP_INTF_DELETE
:
1650 return netlink_put_intf_update_msg(bth
, ctx
);
1652 case DPLANE_OP_TC_QDISC_INSTALL
:
1653 case DPLANE_OP_TC_QDISC_UNINSTALL
:
1654 return netlink_put_tc_qdisc_update_msg(bth
, ctx
);
1655 case DPLANE_OP_TC_CLASS_ADD
:
1656 case DPLANE_OP_TC_CLASS_DELETE
:
1657 case DPLANE_OP_TC_CLASS_UPDATE
:
1658 return netlink_put_tc_class_update_msg(bth
, ctx
);
1659 case DPLANE_OP_TC_FILTER_ADD
:
1660 case DPLANE_OP_TC_FILTER_DELETE
:
1661 case DPLANE_OP_TC_FILTER_UPDATE
:
1662 return netlink_put_tc_filter_update_msg(bth
, ctx
);
1665 return FRR_NETLINK_ERROR
;
1668 void kernel_update_multi(struct dplane_ctx_list_head
*ctx_list
)
1670 struct nl_batch batch
;
1671 struct zebra_dplane_ctx
*ctx
;
1672 struct dplane_ctx_list_head handled_list
;
1673 enum netlink_msg_status res
;
1675 dplane_ctx_q_init(&handled_list
);
1676 nl_batch_init(&batch
, &handled_list
);
1679 ctx
= dplane_ctx_dequeue(ctx_list
);
1683 if (batch
.zns
!= NULL
1684 && batch
.zns
->ns_id
!= dplane_ctx_get_ns(ctx
)->ns_id
)
1685 nl_batch_send(&batch
);
1688 * Assume all messages will succeed and then mark only the ones
1691 dplane_ctx_set_status(ctx
, ZEBRA_DPLANE_REQUEST_SUCCESS
);
1693 res
= nl_put_msg(&batch
, ctx
);
1695 dplane_ctx_enqueue_tail(&(batch
.ctx_list
), ctx
);
1696 if (res
== FRR_NETLINK_ERROR
)
1697 dplane_ctx_set_status(ctx
,
1698 ZEBRA_DPLANE_REQUEST_FAILURE
);
1700 if (batch
.curlen
> batch
.limit
)
1701 nl_batch_send(&batch
);
1704 nl_batch_send(&batch
);
1706 dplane_ctx_q_init(ctx_list
);
1707 dplane_ctx_list_append(ctx_list
, &handled_list
);
1710 struct nlsock
*kernel_netlink_nlsock_lookup(int sock
)
1712 struct nlsock lookup
, *retval
;
1717 retval
= hash_lookup(nlsock_hash
, &lookup
);
1723 /* Insert nlsock entry into hash */
1724 static void kernel_netlink_nlsock_insert(struct nlsock
*nls
)
1727 (void)hash_get(nlsock_hash
, nls
, hash_alloc_intern
);
1731 /* Remove nlsock entry from hash */
1732 static void kernel_netlink_nlsock_remove(struct nlsock
*nls
)
1735 (void)hash_release(nlsock_hash
, nls
);
1739 static uint32_t kernel_netlink_nlsock_key(const void *arg
)
1741 const struct nlsock
*nl
= arg
;
1746 static bool kernel_netlink_nlsock_hash_equal(const void *arg1
, const void *arg2
)
1748 const struct nlsock
*nl1
= arg1
;
1749 const struct nlsock
*nl2
= arg2
;
1751 if (nl1
->sock
== nl2
->sock
)
1757 /* Exported interface function. This function simply calls
1758 netlink_socket (). */
1759 void kernel_init(struct zebra_ns
*zns
)
1761 uint32_t groups
, dplane_groups
, ext_groups
;
1762 #if defined SOL_NETLINK
1767 * Initialize netlink sockets
1769 * If RTMGRP_XXX exists use that, but at some point
1770 * I think the kernel developers realized that
1771 * keeping track of all the different values would
1772 * lead to confusion, so we need to convert the
1773 * RTNLGRP_XXX to a bit position for ourself
1776 * NOTE: If the bit is >= 32, you must use setsockopt(). Those
1777 * groups are added further below after SOL_NETLINK is verified to
1780 groups
= RTMGRP_LINK
|
1782 RTMGRP_IPV4_IFADDR
|
1784 RTMGRP_IPV6_IFADDR
|
1785 RTMGRP_IPV4_MROUTE
|
1787 ((uint32_t) 1 << (RTNLGRP_IPV4_RULE
- 1)) |
1788 ((uint32_t) 1 << (RTNLGRP_IPV6_RULE
- 1)) |
1789 ((uint32_t) 1 << (RTNLGRP_NEXTHOP
- 1)) |
1790 ((uint32_t) 1 << (RTNLGRP_TC
- 1));
1792 dplane_groups
= (RTMGRP_LINK
|
1793 RTMGRP_IPV4_IFADDR
|
1794 RTMGRP_IPV6_IFADDR
|
1795 ((uint32_t) 1 << (RTNLGRP_IPV4_NETCONF
- 1)) |
1796 ((uint32_t) 1 << (RTNLGRP_IPV6_NETCONF
- 1)) |
1797 ((uint32_t) 1 << (RTNLGRP_MPLS_NETCONF
- 1)));
1799 /* Use setsockopt for > 31 group */
1800 ext_groups
= RTNLGRP_TUNNEL
;
1802 snprintf(zns
->netlink
.name
, sizeof(zns
->netlink
.name
),
1803 "netlink-listen (NS %u)", zns
->ns_id
);
1804 zns
->netlink
.sock
= -1;
1805 if (netlink_socket(&zns
->netlink
, groups
, &ext_groups
, 1, zns
->ns_id
) <
1807 zlog_err("Failure to create %s socket",
1812 kernel_netlink_nlsock_insert(&zns
->netlink
);
1814 snprintf(zns
->netlink_cmd
.name
, sizeof(zns
->netlink_cmd
.name
),
1815 "netlink-cmd (NS %u)", zns
->ns_id
);
1816 zns
->netlink_cmd
.sock
= -1;
1817 if (netlink_socket(&zns
->netlink_cmd
, 0, 0, 0, zns
->ns_id
) < 0) {
1818 zlog_err("Failure to create %s socket",
1819 zns
->netlink_cmd
.name
);
1823 kernel_netlink_nlsock_insert(&zns
->netlink_cmd
);
1825 /* Outbound socket for dplane programming of the host OS. */
1826 snprintf(zns
->netlink_dplane_out
.name
,
1827 sizeof(zns
->netlink_dplane_out
.name
), "netlink-dp (NS %u)",
1829 zns
->netlink_dplane_out
.sock
= -1;
1830 if (netlink_socket(&zns
->netlink_dplane_out
, 0, 0, 0, zns
->ns_id
) < 0) {
1831 zlog_err("Failure to create %s socket",
1832 zns
->netlink_dplane_out
.name
);
1836 kernel_netlink_nlsock_insert(&zns
->netlink_dplane_out
);
1838 /* Inbound socket for OS events coming to the dplane. */
1839 snprintf(zns
->netlink_dplane_in
.name
,
1840 sizeof(zns
->netlink_dplane_in
.name
), "netlink-dp-in (NS %u)",
1842 zns
->netlink_dplane_in
.sock
= -1;
1843 if (netlink_socket(&zns
->netlink_dplane_in
, dplane_groups
, 0, 0,
1845 zlog_err("Failure to create %s socket",
1846 zns
->netlink_dplane_in
.name
);
1850 kernel_netlink_nlsock_insert(&zns
->netlink_dplane_in
);
1853 * SOL_NETLINK is not available on all platforms yet
1854 * apparently. It's in bits/socket.h which I am not
1855 * sure that we want to pull into our build system.
1857 #if defined SOL_NETLINK
1860 * setsockopt multicast group subscriptions that don't fit in nl_groups
1862 grp
= RTNLGRP_BRVLAN
;
1863 ret
= setsockopt(zns
->netlink
.sock
, SOL_NETLINK
, NETLINK_ADD_MEMBERSHIP
,
1868 "Registration for RTNLGRP_BRVLAN Membership failed : %d %s",
1869 errno
, safe_strerror(errno
));
1871 * Let's tell the kernel that we want to receive extended
1872 * ACKS over our command socket(s)
1875 ret
= setsockopt(zns
->netlink_cmd
.sock
, SOL_NETLINK
, NETLINK_EXT_ACK
,
1879 zlog_notice("Registration for extended cmd ACK failed : %d %s",
1880 errno
, safe_strerror(errno
));
1883 ret
= setsockopt(zns
->netlink_dplane_out
.sock
, SOL_NETLINK
,
1884 NETLINK_EXT_ACK
, &one
, sizeof(one
));
1887 zlog_notice("Registration for extended dp ACK failed : %d %s",
1888 errno
, safe_strerror(errno
));
1891 * Trim off the payload of the original netlink message in the
1892 * acknowledgment. This option is available since Linux 4.2, so if
1893 * setsockopt fails, ignore the error.
1896 ret
= setsockopt(zns
->netlink_dplane_out
.sock
, SOL_NETLINK
,
1897 NETLINK_CAP_ACK
, &one
, sizeof(one
));
1900 "Registration for reduced ACK packet size failed, probably running an early kernel");
1903 /* Register kernel socket. */
1904 if (fcntl(zns
->netlink
.sock
, F_SETFL
, O_NONBLOCK
) < 0)
1905 flog_err_sys(EC_LIB_SOCKET
, "Can't set %s socket flags: %s",
1906 zns
->netlink
.name
, safe_strerror(errno
));
1908 if (fcntl(zns
->netlink_cmd
.sock
, F_SETFL
, O_NONBLOCK
) < 0)
1909 zlog_err("Can't set %s socket error: %s(%d)",
1910 zns
->netlink_cmd
.name
, safe_strerror(errno
), errno
);
1912 if (fcntl(zns
->netlink_dplane_out
.sock
, F_SETFL
, O_NONBLOCK
) < 0)
1913 zlog_err("Can't set %s socket error: %s(%d)",
1914 zns
->netlink_dplane_out
.name
, safe_strerror(errno
),
1917 if (fcntl(zns
->netlink_dplane_in
.sock
, F_SETFL
, O_NONBLOCK
) < 0)
1918 zlog_err("Can't set %s socket error: %s(%d)",
1919 zns
->netlink_dplane_in
.name
, safe_strerror(errno
),
1922 /* Set receive buffer size if it's set from command line */
1924 netlink_recvbuf(&zns
->netlink
, rcvbufsize
);
1925 netlink_recvbuf(&zns
->netlink_cmd
, rcvbufsize
);
1926 netlink_recvbuf(&zns
->netlink_dplane_out
, rcvbufsize
);
1927 netlink_recvbuf(&zns
->netlink_dplane_in
, rcvbufsize
);
1930 /* Set filter for inbound sockets, to exclude events we've generated
1933 netlink_install_filter(zns
->netlink
.sock
, zns
->netlink_cmd
.snl
.nl_pid
,
1934 zns
->netlink_dplane_out
.snl
.nl_pid
);
1936 netlink_install_filter(zns
->netlink_dplane_in
.sock
,
1937 zns
->netlink_cmd
.snl
.nl_pid
,
1938 zns
->netlink_dplane_out
.snl
.nl_pid
);
1940 zns
->t_netlink
= NULL
;
1942 event_add_read(zrouter
.master
, kernel_read
, zns
, zns
->netlink
.sock
,
1948 /* Helper to clean up an nlsock */
1949 static void kernel_nlsock_fini(struct nlsock
*nls
)
1951 if (nls
&& nls
->sock
>= 0) {
1952 kernel_netlink_nlsock_remove(nls
);
1955 XFREE(MTYPE_NL_BUF
, nls
->buf
);
1960 void kernel_terminate(struct zebra_ns
*zns
, bool complete
)
1962 EVENT_OFF(zns
->t_netlink
);
1964 kernel_nlsock_fini(&zns
->netlink
);
1966 kernel_nlsock_fini(&zns
->netlink_cmd
);
1968 kernel_nlsock_fini(&zns
->netlink_dplane_in
);
1970 /* During zebra shutdown, we need to leave the dataplane socket
1971 * around until all work is done.
1974 kernel_nlsock_fini(&zns
->netlink_dplane_out
);
1978 * Global init for platform-/OS-specific things
1980 void kernel_router_init(void)
1982 /* Init nlsock hash and lock */
1983 pthread_mutex_init(&nlsock_mutex
, NULL
);
1984 nlsock_hash
= hash_create_size(8, kernel_netlink_nlsock_key
,
1985 kernel_netlink_nlsock_hash_equal
,
1986 "Netlink Socket Hash");
1990 * Global deinit for platform-/OS-specific things
1992 void kernel_router_terminate(void)
1994 pthread_mutex_destroy(&nlsock_mutex
);
1996 hash_free(nlsock_hash
);
2000 #endif /* HAVE_NETLINK */