1 /* Kernel communication using netlink interface.
2 * Copyright (C) 1999 Kunihiro Ishiguro
4 * This file is part of GNU Zebra.
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
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.
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
23 #if defined(HANDLE_NETLINK_FUZZING)
26 #endif /* HANDLE_NETLINK_FUZZING */
34 #include "connected.h"
37 #include "zebra_memory.h"
44 #include "lib_errors.h"
46 #include "zebra/zserv.h"
47 #include "zebra/zebra_ns.h"
48 #include "zebra/zebra_vrf.h"
50 #include "zebra/debug.h"
51 #include "zebra/kernel_netlink.h"
52 #include "zebra/rt_netlink.h"
53 #include "zebra/if_netlink.h"
54 #include "zebra/rule_netlink.h"
55 #include "zebra/zebra_errors.h"
57 #ifndef SO_RCVBUFFORCE
58 #define SO_RCVBUFFORCE (33)
61 /* Hack for GNU libc version 2. */
63 #define MSG_TRUNC 0x20
64 #endif /* MSG_TRUNC */
67 #define NLMSG_TAIL(nmsg) \
68 ((struct rtattr *)(((uint8_t *)(nmsg)) \
69 + NLMSG_ALIGN((nmsg)->nlmsg_len)))
73 #define RTA_TAIL(rta) \
74 ((struct rtattr *)(((uint8_t *)(rta)) + RTA_ALIGN((rta)->rta_len)))
77 #ifndef RTNL_FAMILY_IP6MR
78 #define RTNL_FAMILY_IP6MR 129
81 #ifndef RTPROT_MROUTED
82 #define RTPROT_MROUTED 17
85 static const struct message nlmsg_str
[] = {{RTM_NEWROUTE
, "RTM_NEWROUTE"},
86 {RTM_DELROUTE
, "RTM_DELROUTE"},
87 {RTM_GETROUTE
, "RTM_GETROUTE"},
88 {RTM_NEWLINK
, "RTM_NEWLINK"},
89 {RTM_DELLINK
, "RTM_DELLINK"},
90 {RTM_GETLINK
, "RTM_GETLINK"},
91 {RTM_NEWADDR
, "RTM_NEWADDR"},
92 {RTM_DELADDR
, "RTM_DELADDR"},
93 {RTM_GETADDR
, "RTM_GETADDR"},
94 {RTM_NEWNEIGH
, "RTM_NEWNEIGH"},
95 {RTM_DELNEIGH
, "RTM_DELNEIGH"},
96 {RTM_GETNEIGH
, "RTM_GETNEIGH"},
97 {RTM_NEWRULE
, "RTM_NEWRULE"},
98 {RTM_DELRULE
, "RTM_DELRULE"},
99 {RTM_GETRULE
, "RTM_GETRULE"},
102 static const struct message rtproto_str
[] = {
103 {RTPROT_REDIRECT
, "redirect"},
104 {RTPROT_KERNEL
, "kernel"},
105 {RTPROT_BOOT
, "boot"},
106 {RTPROT_STATIC
, "static"},
107 {RTPROT_GATED
, "GateD"},
108 {RTPROT_RA
, "router advertisement"},
110 {RTPROT_ZEBRA
, "Zebra"},
112 {RTPROT_BIRD
, "BIRD"},
113 #endif /* RTPROT_BIRD */
114 {RTPROT_MROUTED
, "mroute"},
116 {RTPROT_OSPF
, "OSPF"},
117 {RTPROT_ISIS
, "IS-IS"},
119 {RTPROT_RIPNG
, "RIPNG"},
120 {RTPROT_ZSTATIC
, "static"},
123 static const struct message family_str
[] = {{AF_INET
, "ipv4"},
125 {AF_BRIDGE
, "bridge"},
126 {RTNL_FAMILY_IPMR
, "ipv4MR"},
127 {RTNL_FAMILY_IP6MR
, "ipv6MR"},
130 static const struct message rttype_str
[] = {{RTN_UNICAST
, "unicast"},
131 {RTN_MULTICAST
, "multicast"},
134 extern struct thread_master
*master
;
135 extern uint32_t nl_rcvbufsize
;
137 extern struct zebra_privs_t zserv_privs
;
140 int netlink_talk_filter(struct nlmsghdr
*h
, ns_id_t ns_id
, int startup
)
143 * This is an error condition that must be handled during
146 * The netlink_talk_filter function is used for communication
147 * down the netlink_cmd pipe and we are expecting
148 * an ack being received. So if we get here
149 * then we did not receive the ack and instead
150 * received some other message in an unexpected
153 zlog_debug("%s: ignoring message type 0x%04x(%s) NS %u", __func__
,
154 h
->nlmsg_type
, nl_msg_type_to_str(h
->nlmsg_type
), ns_id
);
158 static int netlink_recvbuf(struct nlsock
*nl
, uint32_t newsize
)
161 socklen_t newlen
= sizeof(newsize
);
162 socklen_t oldlen
= sizeof(oldsize
);
165 ret
= getsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUF
, &oldsize
, &oldlen
);
167 flog_err_sys(EC_LIB_SOCKET
,
168 "Can't get %s receive buffer size: %s", nl
->name
,
169 safe_strerror(errno
));
173 /* Try force option (linux >= 2.6.14) and fall back to normal set */
174 frr_elevate_privs(&zserv_privs
) {
175 ret
= setsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUFFORCE
,
177 sizeof(nl_rcvbufsize
));
180 ret
= setsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUF
,
181 &nl_rcvbufsize
, sizeof(nl_rcvbufsize
));
183 flog_err_sys(EC_LIB_SOCKET
,
184 "Can't set %s receive buffer size: %s", nl
->name
,
185 safe_strerror(errno
));
189 ret
= getsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUF
, &newsize
, &newlen
);
191 flog_err_sys(EC_LIB_SOCKET
,
192 "Can't get %s receive buffer size: %s", nl
->name
,
193 safe_strerror(errno
));
197 zlog_info("Setting netlink socket receive buffer size: %u -> %u",
202 /* Make socket for Linux netlink interface. */
203 static int netlink_socket(struct nlsock
*nl
, unsigned long groups
,
207 struct sockaddr_nl snl
;
211 frr_elevate_privs(&zserv_privs
) {
212 sock
= ns_socket(AF_NETLINK
, SOCK_RAW
, NETLINK_ROUTE
, ns_id
);
214 zlog_err("Can't open %s socket: %s", nl
->name
,
215 safe_strerror(errno
));
219 memset(&snl
, 0, sizeof snl
);
220 snl
.nl_family
= AF_NETLINK
;
221 snl
.nl_groups
= groups
;
223 /* Bind the socket to the netlink structure for anything. */
224 ret
= bind(sock
, (struct sockaddr
*)&snl
, sizeof snl
);
228 zlog_err("Can't bind %s socket to group 0x%x: %s", nl
->name
,
229 snl
.nl_groups
, safe_strerror(errno
));
234 /* multiple netlink sockets will have different nl_pid */
235 namelen
= sizeof snl
;
236 ret
= getsockname(sock
, (struct sockaddr
*)&snl
, (socklen_t
*)&namelen
);
237 if (ret
< 0 || namelen
!= sizeof snl
) {
238 flog_err_sys(EC_LIB_SOCKET
, "Can't get %s socket name: %s",
239 nl
->name
, safe_strerror(errno
));
249 static int netlink_information_fetch(struct nlmsghdr
*h
, ns_id_t ns_id
,
253 * When we handle new message types here
254 * because we are starting to install them
255 * then lets check the netlink_install_filter
256 * and see if we should add the corresponding
257 * allow through entry there.
258 * Probably not needed to do but please
261 switch (h
->nlmsg_type
) {
263 return netlink_route_change(h
, ns_id
, startup
);
265 return netlink_route_change(h
, ns_id
, startup
);
267 return netlink_link_change(h
, ns_id
, startup
);
269 return netlink_link_change(h
, ns_id
, startup
);
271 return netlink_interface_addr(h
, ns_id
, startup
);
273 return netlink_interface_addr(h
, ns_id
, startup
);
275 return netlink_neigh_change(h
, ns_id
);
277 return netlink_neigh_change(h
, ns_id
);
279 return netlink_rule_change(h
, ns_id
, startup
);
281 return netlink_rule_change(h
, ns_id
, startup
);
284 * If we have received this message then
285 * we have made a mistake during development
286 * and we need to write some code to handle
287 * this message type or not ask for
288 * it to be sent up to us
290 flog_err(EC_ZEBRA_UNKNOWN_NLMSG
,
291 "Unknown netlink nlmsg_type %s(%d) vrf %u\n",
292 nl_msg_type_to_str(h
->nlmsg_type
), h
->nlmsg_type
,
299 #if defined(HANDLE_NETLINK_FUZZING)
300 /* Using globals here to avoid adding function parameters */
302 /* Keep distinct filenames for netlink fuzzy collection */
303 static unsigned int netlink_file_counter
= 1;
305 /* File name to read fuzzed netlink from */
306 static char netlink_fuzz_file
[MAXPATHLEN
] = "";
308 /* Flag for whether to read from file or not */
312 * netlink_read_init() - Starts the message parser
313 * @fname: Filename to read.
315 void netlink_read_init(const char *fname
)
317 struct zebra_dplane_info dp_info
;
319 snprintf(netlink_fuzz_file
, MAXPATHLEN
, "%s", fname
);
320 /* Creating this fake socket for testing purposes */
321 struct zebra_ns
*zns
= zebra_ns_lookup(NS_DEFAULT
);
323 /* Capture key info from zns struct */
324 zebra_dplane_info_from_zns(&dp_info
, zns
, false);
326 netlink_parse_info(netlink_information_fetch
, &zns
->netlink
,
331 * netlink_write_incoming() - Writes all data received from netlink to a file
332 * @buf: Data from netlink.
333 * @size: Size of data.
334 * @counter: Counter for keeping filenames distinct.
336 static void netlink_write_incoming(const char *buf
, const unsigned int size
,
337 unsigned int counter
)
339 char fname
[MAXPATHLEN
];
342 snprintf(fname
, MAXPATHLEN
, "%s/%s_%u", DAEMON_VTY_DIR
, "netlink",
344 frr_elevate_privs(&zserv_privs
) {
345 f
= fopen(fname
, "w");
348 fwrite(buf
, 1, size
, f
);
354 * netlink_read_file() - Reads netlink data from file
355 * @buf: Netlink buffer being overwritten.
356 * @fname: File name to read from.
358 * Return: Size of file.
360 static long netlink_read_file(char *buf
, const char *fname
)
363 long file_bytes
= -1;
365 frr_elevate_privs(&zserv_privs
) {
366 f
= fopen(fname
, "r");
369 fseek(f
, 0, SEEK_END
);
370 file_bytes
= ftell(f
);
372 fread(buf
, NL_RCV_PKT_BUF_SIZE
, 1, f
);
378 #endif /* HANDLE_NETLINK_FUZZING */
380 static int kernel_read(struct thread
*thread
)
382 struct zebra_ns
*zns
= (struct zebra_ns
*)THREAD_ARG(thread
);
383 struct zebra_dplane_info dp_info
;
385 /* Capture key info from ns struct */
386 zebra_dplane_info_from_zns(&dp_info
, zns
, false);
388 netlink_parse_info(netlink_information_fetch
, &zns
->netlink
, &dp_info
,
390 zns
->t_netlink
= NULL
;
391 thread_add_read(zebrad
.master
, kernel_read
, zns
, zns
->netlink
.sock
,
398 * Filter out messages from self that occur on listener socket,
399 * caused by our actions on the command socket
401 * When we add new Netlink message types we probably
402 * do not need to add them here as that we are filtering
403 * on the routes we actually care to receive( which is rarer
404 * then the normal course of operations). We are intentionally
405 * allowing some messages from ourselves through
406 * ( I'm looking at you Interface based netlink messages )
407 * so that we only had to write one way to handle incoming
408 * address add/delete changes.
410 static void netlink_install_filter(int sock
, __u32 pid
)
413 * BPF_JUMP instructions and where you jump to are based upon
414 * 0 as being the next statement. So count from 0. Writing
415 * this down because every time I look at this I have to
418 struct sock_filter filter
[] = {
421 * if (nlmsg_pid == pid) {
422 * if (the incoming nlmsg_type ==
423 * RTM_NEWADDR | RTM_DELADDR)
428 * keep this netlink message
431 * 0: Load the nlmsg_pid into the BPF register
433 BPF_STMT(BPF_LD
| BPF_ABS
| BPF_W
,
434 offsetof(struct nlmsghdr
, nlmsg_pid
)),
438 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htonl(pid
), 0, 4),
440 * 2: Load the nlmsg_type into BPF register
442 BPF_STMT(BPF_LD
| BPF_ABS
| BPF_H
,
443 offsetof(struct nlmsghdr
, nlmsg_type
)),
445 * 3: Compare to RTM_NEWADDR
447 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htons(RTM_NEWADDR
), 2, 0),
449 * 4: Compare to RTM_DELADDR
451 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htons(RTM_DELADDR
), 1, 0),
453 * 5: This is the end state of we want to skip the
456 BPF_STMT(BPF_RET
| BPF_K
, 0),
457 /* 6: This is the end state of we want to keep
460 BPF_STMT(BPF_RET
| BPF_K
, 0xffff),
463 struct sock_fprog prog
= {
464 .len
= array_size(filter
), .filter
= filter
,
467 if (setsockopt(sock
, SOL_SOCKET
, SO_ATTACH_FILTER
, &prog
, sizeof(prog
))
469 flog_err_sys(EC_LIB_SOCKET
, "Can't install socket filter: %s\n",
470 safe_strerror(errno
));
473 void netlink_parse_rtattr(struct rtattr
**tb
, int max
, struct rtattr
*rta
,
476 while (RTA_OK(rta
, len
)) {
477 if (rta
->rta_type
<= max
)
478 tb
[rta
->rta_type
] = rta
;
479 rta
= RTA_NEXT(rta
, len
);
484 * netlink_parse_rtattr_nested() - Parses a nested route attribute
485 * @tb: Pointer to array for storing rtattr in.
486 * @max: Max number to store.
487 * @rta: Pointer to rtattr to look for nested items in.
489 void netlink_parse_rtattr_nested(struct rtattr
**tb
, int max
,
492 netlink_parse_rtattr(tb
, max
, RTA_DATA(rta
), RTA_PAYLOAD(rta
));
495 int addattr_l(struct nlmsghdr
*n
, unsigned int maxlen
, int type
,
496 const void *data
, unsigned int alen
)
501 len
= RTA_LENGTH(alen
);
503 if (NLMSG_ALIGN(n
->nlmsg_len
) + RTA_ALIGN(len
) > maxlen
)
506 rta
= (struct rtattr
*)(((char *)n
) + NLMSG_ALIGN(n
->nlmsg_len
));
507 rta
->rta_type
= type
;
511 memcpy(RTA_DATA(rta
), data
, alen
);
515 n
->nlmsg_len
= NLMSG_ALIGN(n
->nlmsg_len
) + RTA_ALIGN(len
);
520 int rta_addattr_l(struct rtattr
*rta
, unsigned int maxlen
, int type
,
521 const void *data
, unsigned int alen
)
524 struct rtattr
*subrta
;
526 len
= RTA_LENGTH(alen
);
528 if (RTA_ALIGN(rta
->rta_len
) + RTA_ALIGN(len
) > maxlen
)
531 subrta
= (struct rtattr
*)(((char *)rta
) + RTA_ALIGN(rta
->rta_len
));
532 subrta
->rta_type
= type
;
533 subrta
->rta_len
= len
;
536 memcpy(RTA_DATA(subrta
), data
, alen
);
540 rta
->rta_len
= NLMSG_ALIGN(rta
->rta_len
) + RTA_ALIGN(len
);
545 int addattr16(struct nlmsghdr
*n
, unsigned int maxlen
, int type
, uint16_t data
)
547 return addattr_l(n
, maxlen
, type
, &data
, sizeof(uint16_t));
550 int addattr32(struct nlmsghdr
*n
, unsigned int maxlen
, int type
, int data
)
552 return addattr_l(n
, maxlen
, type
, &data
, sizeof(uint32_t));
555 struct rtattr
*addattr_nest(struct nlmsghdr
*n
, int maxlen
, int type
)
557 struct rtattr
*nest
= NLMSG_TAIL(n
);
559 addattr_l(n
, maxlen
, type
, NULL
, 0);
563 int addattr_nest_end(struct nlmsghdr
*n
, struct rtattr
*nest
)
565 nest
->rta_len
= (uint8_t *)NLMSG_TAIL(n
) - (uint8_t *)nest
;
569 struct rtattr
*rta_nest(struct rtattr
*rta
, int maxlen
, int type
)
571 struct rtattr
*nest
= RTA_TAIL(rta
);
573 rta_addattr_l(rta
, maxlen
, type
, NULL
, 0);
577 int rta_nest_end(struct rtattr
*rta
, struct rtattr
*nest
)
579 nest
->rta_len
= (uint8_t *)RTA_TAIL(rta
) - (uint8_t *)nest
;
583 const char *nl_msg_type_to_str(uint16_t msg_type
)
585 return lookup_msg(nlmsg_str
, msg_type
, "");
588 const char *nl_rtproto_to_str(uint8_t rtproto
)
590 return lookup_msg(rtproto_str
, rtproto
, "");
593 const char *nl_family_to_str(uint8_t family
)
595 return lookup_msg(family_str
, family
, "");
598 const char *nl_rttype_to_str(uint8_t rttype
)
600 return lookup_msg(rttype_str
, rttype
, "");
603 #define NL_OK(nla, len) \
604 ((len) >= (int)sizeof(struct nlattr) \
605 && (nla)->nla_len >= sizeof(struct nlattr) \
606 && (nla)->nla_len <= (len))
607 #define NL_NEXT(nla, attrlen) \
608 ((attrlen) -= RTA_ALIGN((nla)->nla_len), \
609 (struct nlattr *)(((char *)(nla)) + RTA_ALIGN((nla)->nla_len)))
611 ((struct nlattr *)(((char *)(r)) \
612 + NLMSG_ALIGN(sizeof(struct nlmsgerr))))
614 static void netlink_parse_nlattr(struct nlattr
**tb
, int max
,
615 struct nlattr
*nla
, int len
)
617 while (NL_OK(nla
, len
)) {
618 if (nla
->nla_type
<= max
)
619 tb
[nla
->nla_type
] = nla
;
620 nla
= NL_NEXT(nla
, len
);
624 static void netlink_parse_extended_ack(struct nlmsghdr
*h
)
626 struct nlattr
*tb
[NLMSGERR_ATTR_MAX
+ 1];
627 const struct nlmsgerr
*err
=
628 (const struct nlmsgerr
*)((uint8_t *)h
630 sizeof(struct nlmsghdr
)));
631 const struct nlmsghdr
*err_nlh
= NULL
;
632 uint32_t hlen
= sizeof(*err
);
633 const char *msg
= NULL
;
636 if (!(h
->nlmsg_flags
& NLM_F_CAPPED
))
637 hlen
+= h
->nlmsg_len
- NLMSG_ALIGN(sizeof(struct nlmsghdr
));
639 memset(tb
, 0, sizeof(tb
));
640 netlink_parse_nlattr(tb
, NLMSGERR_ATTR_MAX
, NL_RTA(h
), hlen
);
642 if (tb
[NLMSGERR_ATTR_MSG
])
643 msg
= (const char *)RTA_DATA(tb
[NLMSGERR_ATTR_MSG
]);
645 if (tb
[NLMSGERR_ATTR_OFFS
]) {
646 off
= *(uint32_t *)RTA_DATA(tb
[NLMSGERR_ATTR_OFFS
]);
648 if (off
> h
->nlmsg_len
) {
649 zlog_err("Invalid offset for NLMSGERR_ATTR_OFFS\n");
650 } else if (!(h
->nlmsg_flags
& NLM_F_CAPPED
)) {
652 * Header of failed message
653 * we are not doing anything currently with it
654 * but noticing it for later.
657 zlog_debug("%s: Received %s extended Ack",
659 nl_msg_type_to_str(err_nlh
->nlmsg_type
));
663 if (msg
&& *msg
!= '\0') {
664 bool is_err
= !!err
->error
;
667 zlog_err("Extended Error: %s", msg
);
669 flog_warn(EC_ZEBRA_NETLINK_EXTENDED_WARNING
,
670 "Extended Warning: %s", msg
);
677 * Receive message from netlink interface and pass those information
678 * to the given function.
680 * filter -> Function to call to read the results
681 * nl -> netlink socket information
682 * zns -> The zebra namespace data
683 * count -> How many we should read in, 0 means as much as possible
684 * startup -> Are we reading in under startup conditions? passed to
687 int netlink_parse_info(int (*filter
)(struct nlmsghdr
*, ns_id_t
, int),
688 const struct nlsock
*nl
,
689 const struct zebra_dplane_info
*zns
,
690 int count
, int startup
)
698 char buf
[NL_RCV_PKT_BUF_SIZE
];
699 struct iovec iov
= {.iov_base
= buf
, .iov_len
= sizeof buf
};
700 struct sockaddr_nl snl
;
701 struct msghdr msg
= {.msg_name
= (void *)&snl
,
702 .msg_namelen
= sizeof snl
,
707 if (count
&& read_in
>= count
)
710 #if defined(HANDLE_NETLINK_FUZZING)
711 /* Check if reading and filename is set */
712 if (netlink_read
&& '\0' != netlink_fuzz_file
[0]) {
713 zlog_debug("Reading netlink fuzz file");
714 status
= netlink_read_file(buf
, netlink_fuzz_file
);
717 status
= recvmsg(nl
->sock
, &msg
, 0);
720 status
= recvmsg(nl
->sock
, &msg
, 0);
721 #endif /* HANDLE_NETLINK_FUZZING */
725 if (errno
== EWOULDBLOCK
|| errno
== EAGAIN
)
727 flog_err(EC_ZEBRA_RECVMSG_OVERRUN
,
728 "%s recvmsg overrun: %s", nl
->name
,
729 safe_strerror(errno
));
731 * In this case we are screwed.
732 * There is no good way to
733 * recover zebra at this point.
740 flog_err_sys(EC_LIB_SOCKET
, "%s EOF", nl
->name
);
744 if (msg
.msg_namelen
!= sizeof snl
) {
745 flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR
,
746 "%s sender address length error: length %d",
747 nl
->name
, msg
.msg_namelen
);
751 if (IS_ZEBRA_DEBUG_KERNEL_MSGDUMP_RECV
) {
752 zlog_debug("%s: << netlink message dump [recv]",
754 zlog_hexdump(buf
, status
);
757 #if defined(HANDLE_NETLINK_FUZZING)
759 zlog_debug("Writing incoming netlink message");
760 netlink_write_incoming(buf
, status
,
761 netlink_file_counter
++);
763 #endif /* HANDLE_NETLINK_FUZZING */
766 for (h
= (struct nlmsghdr
*)buf
;
767 (status
>= 0 && NLMSG_OK(h
, (unsigned int)status
));
768 h
= NLMSG_NEXT(h
, status
)) {
769 /* Finish of reading. */
770 if (h
->nlmsg_type
== NLMSG_DONE
)
773 /* Error handling. */
774 if (h
->nlmsg_type
== NLMSG_ERROR
) {
775 struct nlmsgerr
*err
=
776 (struct nlmsgerr
*)NLMSG_DATA(h
);
777 int errnum
= err
->error
;
778 int msg_type
= err
->msg
.nlmsg_type
;
781 < NLMSG_LENGTH(sizeof(struct nlmsgerr
))) {
782 zlog_err("%s error: message truncated",
788 * Parse the extended information before
789 * we actually handle it.
790 * At this point in time we do not
791 * do anything other than report the
794 if (h
->nlmsg_flags
& NLM_F_ACK_TLVS
)
795 netlink_parse_extended_ack(h
);
797 /* If the error field is zero, then this is an
799 if (err
->error
== 0) {
800 if (IS_ZEBRA_DEBUG_KERNEL
) {
802 "%s: %s ACK: type=%s(%u), seq=%u, pid=%u",
803 __FUNCTION__
, nl
->name
,
805 err
->msg
.nlmsg_type
),
811 /* return if not a multipart message,
812 * otherwise continue */
813 if (!(h
->nlmsg_flags
& NLM_F_MULTI
))
819 < NLMSG_LENGTH(sizeof(struct nlmsgerr
))) {
820 flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR
,
821 "%s error: message truncated",
826 /* Deal with errors that occur because of races
827 * in link handling */
829 && ((msg_type
== RTM_DELROUTE
830 && (-errnum
== ENODEV
831 || -errnum
== ESRCH
))
832 || (msg_type
== RTM_NEWROUTE
833 && (-errnum
== ENETDOWN
834 || -errnum
== EEXIST
)))) {
835 if (IS_ZEBRA_DEBUG_KERNEL
)
837 "%s: error: %s type=%s(%u), seq=%u, pid=%u",
839 safe_strerror(-errnum
),
848 /* We see RTM_DELNEIGH when shutting down an
849 * interface with an IPv4
850 * link-local. The kernel should have already
851 * deleted the neighbor
852 * so do not log these as an error.
854 if (msg_type
== RTM_DELNEIGH
855 || (zns
->is_cmd
&& msg_type
== RTM_NEWROUTE
857 || -errnum
== ENETUNREACH
))) {
858 /* This is known to happen in some
859 * situations, don't log
862 if (IS_ZEBRA_DEBUG_KERNEL
)
864 "%s error: %s, type=%s(%u), seq=%u, pid=%u",
866 safe_strerror(-errnum
),
874 EC_ZEBRA_UNEXPECTED_MESSAGE
,
875 "%s error: %s, type=%s(%u), seq=%u, pid=%u",
877 safe_strerror(-errnum
),
878 nl_msg_type_to_str(msg_type
),
879 msg_type
, err
->msg
.nlmsg_seq
,
885 /* OK we got netlink message. */
886 if (IS_ZEBRA_DEBUG_KERNEL
)
888 "netlink_parse_info: %s type %s(%u), len=%d, seq=%u, pid=%u",
890 nl_msg_type_to_str(h
->nlmsg_type
),
891 h
->nlmsg_type
, h
->nlmsg_len
,
892 h
->nlmsg_seq
, h
->nlmsg_pid
);
896 * Ignore messages that maybe sent from
897 * other actors besides the kernel
899 if (snl
.nl_pid
!= 0) {
900 zlog_debug("Ignoring message from pid %u",
905 error
= (*filter
)(h
, zns
->ns_id
, startup
);
907 zlog_debug("%s filter function error",
913 /* After error care. */
914 if (msg
.msg_flags
& MSG_TRUNC
) {
915 flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR
,
916 "%s error: message truncated", nl
->name
);
920 flog_err(EC_ZEBRA_NETLINK_LENGTH_ERROR
,
921 "%s error: data remnant size %d", nl
->name
,
932 * sendmsg() to netlink socket then recvmsg().
933 * Calls netlink_parse_info to parse returned data
935 * filter -> The filter to read final results from kernel
936 * nlmsghdr -> The data to send to the kernel
937 * dp_info -> The dataplane and netlink socket information
938 * startup -> Are we reading in under startup conditions
939 * This is passed through eventually to filter.
941 int netlink_talk_info(int (*filter
)(struct nlmsghdr
*, ns_id_t
, int startup
),
943 const struct zebra_dplane_info
*dp_info
, int startup
)
946 struct sockaddr_nl snl
;
950 const struct nlsock
*nl
;
952 memset(&snl
, 0, sizeof snl
);
953 memset(&iov
, 0, sizeof iov
);
954 memset(&msg
, 0, sizeof msg
);
957 iov
.iov_len
= n
->nlmsg_len
;
958 msg
.msg_name
= (void *)&snl
;
959 msg
.msg_namelen
= sizeof snl
;
963 snl
.nl_family
= AF_NETLINK
;
965 nl
= &(dp_info
->nls
);
966 n
->nlmsg_seq
= nl
->seq
;
967 n
->nlmsg_pid
= nl
->snl
.nl_pid
;
969 if (IS_ZEBRA_DEBUG_KERNEL
)
971 "netlink_talk: %s type %s(%u), len=%d seq=%u flags 0x%x",
972 nl
->name
, nl_msg_type_to_str(n
->nlmsg_type
),
973 n
->nlmsg_type
, n
->nlmsg_len
, n
->nlmsg_seq
,
976 /* Send message to netlink interface. */
977 frr_elevate_privs(&zserv_privs
) {
978 status
= sendmsg(nl
->sock
, &msg
, 0);
982 if (IS_ZEBRA_DEBUG_KERNEL_MSGDUMP_SEND
) {
983 zlog_debug("%s: >> netlink message dump [sent]", __func__
);
984 zlog_hexdump(n
, n
->nlmsg_len
);
988 flog_err_sys(EC_LIB_SOCKET
, "netlink_talk sendmsg() error: %s",
989 safe_strerror(save_errno
));
994 * Get reply from netlink socket.
995 * The reply should either be an acknowlegement or an error.
997 return netlink_parse_info(filter
, nl
, dp_info
, 0, startup
);
1001 * Synchronous version of netlink_talk_info. Converts args to suit the
1002 * common version, which is suitable for both sync and async use.
1004 int netlink_talk(int (*filter
)(struct nlmsghdr
*, ns_id_t
, int startup
),
1005 struct nlmsghdr
*n
, struct nlsock
*nl
, struct zebra_ns
*zns
,
1008 struct zebra_dplane_info dp_info
;
1010 /* Increment sequence number before capturing snapshot of ns socket
1015 /* Capture info in intermediate info struct */
1016 zebra_dplane_info_from_zns(&dp_info
, zns
, (nl
== &(zns
->netlink_cmd
)));
1018 return netlink_talk_info(filter
, n
, &dp_info
, startup
);
1021 /* Issue request message to kernel via netlink socket. GET messages
1022 * are issued through this interface.
1024 int netlink_request(struct nlsock
*nl
, struct nlmsghdr
*n
)
1027 struct sockaddr_nl snl
;
1029 /* Check netlink socket. */
1031 flog_err_sys(EC_LIB_SOCKET
, "%s socket isn't active.",
1036 /* Fill common fields for all requests. */
1037 n
->nlmsg_flags
= NLM_F_ROOT
| NLM_F_MATCH
| NLM_F_REQUEST
;
1038 n
->nlmsg_pid
= nl
->snl
.nl_pid
;
1039 n
->nlmsg_seq
= ++nl
->seq
;
1041 memset(&snl
, 0, sizeof snl
);
1042 snl
.nl_family
= AF_NETLINK
;
1044 /* Raise capabilities and send message, then lower capabilities. */
1045 frr_elevate_privs(&zserv_privs
) {
1046 ret
= sendto(nl
->sock
, (void *)n
, n
->nlmsg_len
, 0,
1047 (struct sockaddr
*)&snl
, sizeof snl
);
1051 zlog_err("%s sendto failed: %s", nl
->name
,
1052 safe_strerror(errno
));
1059 /* Exported interface function. This function simply calls
1060 netlink_socket (). */
1061 void kernel_init(struct zebra_ns
*zns
)
1063 unsigned long groups
;
1064 #if defined SOL_NETLINK
1069 * Initialize netlink sockets
1071 * If RTMGRP_XXX exists use that, but at some point
1072 * I think the kernel developers realized that
1073 * keeping track of all the different values would
1074 * lead to confusion, so we need to convert the
1075 * RTNLGRP_XXX to a bit position for ourself
1077 groups
= RTMGRP_LINK
|
1079 RTMGRP_IPV4_IFADDR
|
1081 RTMGRP_IPV6_IFADDR
|
1082 RTMGRP_IPV4_MROUTE
|
1084 (1 << (RTNLGRP_IPV4_RULE
- 1)) |
1085 (1 << (RTNLGRP_IPV6_RULE
- 1));
1087 snprintf(zns
->netlink
.name
, sizeof(zns
->netlink
.name
),
1088 "netlink-listen (NS %u)", zns
->ns_id
);
1089 zns
->netlink
.sock
= -1;
1090 if (netlink_socket(&zns
->netlink
, groups
, zns
->ns_id
) < 0) {
1091 zlog_err("Failure to create %s socket",
1096 snprintf(zns
->netlink_cmd
.name
, sizeof(zns
->netlink_cmd
.name
),
1097 "netlink-cmd (NS %u)", zns
->ns_id
);
1098 zns
->netlink_cmd
.sock
= -1;
1099 if (netlink_socket(&zns
->netlink_cmd
, 0, zns
->ns_id
) < 0) {
1100 zlog_err("Failure to create %s socket",
1101 zns
->netlink_cmd
.name
);
1106 * SOL_NETLINK is not available on all platforms yet
1107 * apparently. It's in bits/socket.h which I am not
1108 * sure that we want to pull into our build system.
1110 #if defined SOL_NETLINK
1112 * Let's tell the kernel that we want to receive extended
1113 * ACKS over our command socket
1116 ret
= setsockopt(zns
->netlink_cmd
.sock
, SOL_NETLINK
, NETLINK_EXT_ACK
,
1120 zlog_notice("Registration for extended ACK failed : %d %s",
1121 errno
, safe_strerror(errno
));
1124 /* Register kernel socket. */
1125 if (fcntl(zns
->netlink
.sock
, F_SETFL
, O_NONBLOCK
) < 0)
1126 flog_err_sys(EC_LIB_SOCKET
, "Can't set %s socket flags: %s",
1127 zns
->netlink
.name
, safe_strerror(errno
));
1129 if (fcntl(zns
->netlink_cmd
.sock
, F_SETFL
, O_NONBLOCK
) < 0)
1130 zlog_err("Can't set %s socket error: %s(%d)",
1131 zns
->netlink_cmd
.name
, safe_strerror(errno
), errno
);
1133 /* Set receive buffer size if it's set from command line */
1135 netlink_recvbuf(&zns
->netlink
, nl_rcvbufsize
);
1137 netlink_install_filter(zns
->netlink
.sock
,
1138 zns
->netlink_cmd
.snl
.nl_pid
);
1139 zns
->t_netlink
= NULL
;
1141 thread_add_read(zebrad
.master
, kernel_read
, zns
,
1142 zns
->netlink
.sock
, &zns
->t_netlink
);
1147 void kernel_terminate(struct zebra_ns
*zns
)
1149 THREAD_READ_OFF(zns
->t_netlink
);
1151 if (zns
->netlink
.sock
>= 0) {
1152 close(zns
->netlink
.sock
);
1153 zns
->netlink
.sock
= -1;
1156 if (zns
->netlink_cmd
.sock
>= 0) {
1157 close(zns
->netlink_cmd
.sock
);
1158 zns
->netlink_cmd
.sock
= -1;
1162 #endif /* HAVE_NETLINK */