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
;
139 int netlink_talk_filter(struct nlmsghdr
*h
, ns_id_t ns_id
, int startup
)
142 * This is an error condition that must be handled during
145 * The netlink_talk_filter function is used for communication
146 * down the netlink_cmd pipe and we are expecting
147 * an ack being received. So if we get here
148 * then we did not receive the ack and instead
149 * received some other message in an unexpected
152 zlog_debug("%s: ignoring message type 0x%04x(%s) NS %u", __func__
,
153 h
->nlmsg_type
, nl_msg_type_to_str(h
->nlmsg_type
), ns_id
);
157 static int netlink_recvbuf(struct nlsock
*nl
, uint32_t newsize
)
160 socklen_t newlen
= sizeof(newsize
);
161 socklen_t oldlen
= sizeof(oldsize
);
164 ret
= getsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUF
, &oldsize
, &oldlen
);
166 flog_err_sys(LIB_ERR_SOCKET
,
167 "Can't get %s receive buffer size: %s", nl
->name
,
168 safe_strerror(errno
));
172 /* Try force option (linux >= 2.6.14) and fall back to normal set */
173 if (zserv_privs
.change(ZPRIVS_RAISE
))
174 flog_err(LIB_ERR_PRIVILEGES
,
175 "routing_socket: Can't raise privileges");
176 ret
= setsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUFFORCE
, &nl_rcvbufsize
,
177 sizeof(nl_rcvbufsize
));
178 if (zserv_privs
.change(ZPRIVS_LOWER
))
179 flog_err(LIB_ERR_PRIVILEGES
,
180 "routing_socket: Can't lower privileges");
182 ret
= setsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUF
,
183 &nl_rcvbufsize
, sizeof(nl_rcvbufsize
));
185 flog_err_sys(LIB_ERR_SOCKET
,
186 "Can't set %s receive buffer size: %s", nl
->name
,
187 safe_strerror(errno
));
191 ret
= getsockopt(nl
->sock
, SOL_SOCKET
, SO_RCVBUF
, &newsize
, &newlen
);
193 flog_err_sys(LIB_ERR_SOCKET
,
194 "Can't get %s receive buffer size: %s", nl
->name
,
195 safe_strerror(errno
));
199 zlog_info("Setting netlink socket receive buffer size: %u -> %u",
204 /* Make socket for Linux netlink interface. */
205 static int netlink_socket(struct nlsock
*nl
, unsigned long groups
,
209 struct sockaddr_nl snl
;
214 if (zserv_privs
.change(ZPRIVS_RAISE
)) {
215 flog_err(LIB_ERR_PRIVILEGES
, "Can't raise privileges");
219 sock
= ns_socket(AF_NETLINK
, SOCK_RAW
, NETLINK_ROUTE
, ns_id
);
221 flog_err_sys(LIB_ERR_SOCKET
, "Can't open %s socket: %s",
222 nl
->name
, safe_strerror(errno
));
226 memset(&snl
, 0, sizeof snl
);
227 snl
.nl_family
= AF_NETLINK
;
228 snl
.nl_groups
= groups
;
230 /* Bind the socket to the netlink structure for anything. */
231 ret
= bind(sock
, (struct sockaddr
*)&snl
, sizeof snl
);
233 if (zserv_privs
.change(ZPRIVS_LOWER
))
234 flog_err(LIB_ERR_PRIVILEGES
, "Can't lower privileges");
237 flog_err_sys(LIB_ERR_SOCKET
,
238 "Can't bind %s socket to group 0x%x: %s", nl
->name
,
239 snl
.nl_groups
, safe_strerror(save_errno
));
244 /* multiple netlink sockets will have different nl_pid */
245 namelen
= sizeof snl
;
246 ret
= getsockname(sock
, (struct sockaddr
*)&snl
, (socklen_t
*)&namelen
);
247 if (ret
< 0 || namelen
!= sizeof snl
) {
248 flog_err_sys(LIB_ERR_SOCKET
, "Can't get %s socket name: %s",
249 nl
->name
, safe_strerror(errno
));
259 static int netlink_information_fetch(struct nlmsghdr
*h
, ns_id_t ns_id
,
263 * When we handle new message types here
264 * because we are starting to install them
265 * then lets check the netlink_install_filter
266 * and see if we should add the corresponding
267 * allow through entry there.
268 * Probably not needed to do but please
271 switch (h
->nlmsg_type
) {
273 return netlink_route_change(h
, ns_id
, startup
);
275 return netlink_route_change(h
, ns_id
, startup
);
277 return netlink_link_change(h
, ns_id
, startup
);
279 return netlink_link_change(h
, ns_id
, startup
);
281 return netlink_interface_addr(h
, ns_id
, startup
);
283 return netlink_interface_addr(h
, ns_id
, startup
);
285 return netlink_neigh_change(h
, ns_id
);
287 return netlink_neigh_change(h
, ns_id
);
289 return netlink_rule_change(h
, ns_id
, startup
);
291 return netlink_rule_change(h
, ns_id
, startup
);
294 * If we have received this message then
295 * we have made a mistake during development
296 * and we need to write some code to handle
297 * this message type or not ask for
298 * it to be sent up to us
300 flog_err(ZEBRA_ERR_UNKNOWN_NLMSG
,
301 "Unknown netlink nlmsg_type %s(%d) vrf %u\n",
302 nl_msg_type_to_str(h
->nlmsg_type
), h
->nlmsg_type
,
309 #if defined(HANDLE_NETLINK_FUZZING)
310 /* Using globals here to avoid adding function parameters */
312 /* Keep distinct filenames for netlink fuzzy collection */
313 static unsigned int netlink_file_counter
= 1;
315 /* File name to read fuzzed netlink from */
316 static char netlink_fuzz_file
[MAXPATHLEN
] = "";
318 /* Flag for whether to read from file or not */
322 * netlink_read_init() - Starts the message parser
323 * @fname: Filename to read.
325 void netlink_read_init(const char *fname
)
327 snprintf(netlink_fuzz_file
, MAXPATHLEN
, "%s", fname
);
328 /* Creating this fake socket for testing purposes */
329 struct zebra_ns
*zns
= zebra_ns_lookup(NS_DEFAULT
);
331 netlink_parse_info(netlink_information_fetch
, &zns
->netlink
, zns
, 1, 0);
335 * netlink_write_incoming() - Writes all data received from netlink to a file
336 * @buf: Data from netlink.
337 * @size: Size of data.
338 * @counter: Counter for keeping filenames distinct.
340 static void netlink_write_incoming(const char *buf
, const unsigned int size
,
341 unsigned int counter
)
343 char fname
[MAXPATHLEN
];
346 zserv_privs
.change(ZPRIVS_RAISE
);
347 snprintf(fname
, MAXPATHLEN
, "%s/%s_%u", DAEMON_VTY_DIR
, "netlink",
349 f
= fopen(fname
, "w");
351 fwrite(buf
, 1, size
, f
);
354 zserv_privs
.change(ZPRIVS_LOWER
);
358 * netlink_read_file() - Reads netlink data from file
359 * @buf: Netlink buffer being overwritten.
360 * @fname: File name to read from.
362 * Return: Size of file.
364 static long netlink_read_file(char *buf
, const char *fname
)
367 long file_bytes
= -1;
369 zserv_privs
.change(ZPRIVS_RAISE
);
370 f
= fopen(fname
, "r");
372 fseek(f
, 0, SEEK_END
);
373 file_bytes
= ftell(f
);
375 fread(buf
, NL_RCV_PKT_BUF_SIZE
, 1, f
);
378 zserv_privs
.change(ZPRIVS_LOWER
);
382 #endif /* HANDLE_NETLINK_FUZZING */
384 static int kernel_read(struct thread
*thread
)
386 struct zebra_ns
*zns
= (struct zebra_ns
*)THREAD_ARG(thread
);
387 netlink_parse_info(netlink_information_fetch
, &zns
->netlink
, zns
, 5, 0);
388 zns
->t_netlink
= NULL
;
389 thread_add_read(zebrad
.master
, kernel_read
, zns
, zns
->netlink
.sock
,
396 * Filter out messages from self that occur on listener socket,
397 * caused by our actions on the command socket
399 * When we add new Netlink message types we probably
400 * do not need to add them here as that we are filtering
401 * on the routes we actually care to receive( which is rarer
402 * then the normal course of operations). We are intentionally
403 * allowing some messages from ourselves through
404 * ( I'm looking at you Interface based netlink messages )
405 * so that we only had to write one way to handle incoming
406 * address add/delete changes.
408 static void netlink_install_filter(int sock
, __u32 pid
)
411 * BPF_JUMP instructions and where you jump to are based upon
412 * 0 as being the next statement. So count from 0. Writing
413 * this down because every time I look at this I have to
416 struct sock_filter filter
[] = {
419 * if (nlmsg_pid == pid) {
420 * if (the incoming nlmsg_type ==
421 * RTM_NEWADDR | RTM_DELADDR)
426 * keep this netlink message
429 * 0: Load the nlmsg_pid into the BPF register
431 BPF_STMT(BPF_LD
| BPF_ABS
| BPF_W
,
432 offsetof(struct nlmsghdr
, nlmsg_pid
)),
436 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htonl(pid
), 0, 4),
438 * 2: Load the nlmsg_type into BPF register
440 BPF_STMT(BPF_LD
| BPF_ABS
| BPF_H
,
441 offsetof(struct nlmsghdr
, nlmsg_type
)),
443 * 3: Compare to RTM_NEWADDR
445 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htons(RTM_NEWADDR
), 2, 0),
447 * 4: Compare to RTM_DELADDR
449 BPF_JUMP(BPF_JMP
| BPF_JEQ
| BPF_K
, htons(RTM_DELADDR
), 1, 0),
451 * 5: This is the end state of we want to skip the
454 BPF_STMT(BPF_RET
| BPF_K
, 0),
455 /* 6: This is the end state of we want to keep
458 BPF_STMT(BPF_RET
| BPF_K
, 0xffff),
461 struct sock_fprog prog
= {
462 .len
= array_size(filter
), .filter
= filter
,
465 if (setsockopt(sock
, SOL_SOCKET
, SO_ATTACH_FILTER
, &prog
, sizeof(prog
))
467 zlog_warn("Can't install socket filter: %s\n",
468 safe_strerror(errno
));
471 void netlink_parse_rtattr(struct rtattr
**tb
, int max
, struct rtattr
*rta
,
474 while (RTA_OK(rta
, len
)) {
475 if (rta
->rta_type
<= max
)
476 tb
[rta
->rta_type
] = rta
;
477 rta
= RTA_NEXT(rta
, len
);
481 int addattr_l(struct nlmsghdr
*n
, unsigned int maxlen
, int type
,
482 const void *data
, unsigned int alen
)
487 len
= RTA_LENGTH(alen
);
489 if (NLMSG_ALIGN(n
->nlmsg_len
) + RTA_ALIGN(len
) > maxlen
)
492 rta
= (struct rtattr
*)(((char *)n
) + NLMSG_ALIGN(n
->nlmsg_len
));
493 rta
->rta_type
= type
;
497 memcpy(RTA_DATA(rta
), data
, alen
);
501 n
->nlmsg_len
= NLMSG_ALIGN(n
->nlmsg_len
) + RTA_ALIGN(len
);
506 int rta_addattr_l(struct rtattr
*rta
, unsigned int maxlen
, int type
,
507 const void *data
, unsigned int alen
)
510 struct rtattr
*subrta
;
512 len
= RTA_LENGTH(alen
);
514 if (RTA_ALIGN(rta
->rta_len
) + RTA_ALIGN(len
) > maxlen
)
517 subrta
= (struct rtattr
*)(((char *)rta
) + RTA_ALIGN(rta
->rta_len
));
518 subrta
->rta_type
= type
;
519 subrta
->rta_len
= len
;
522 memcpy(RTA_DATA(subrta
), data
, alen
);
526 rta
->rta_len
= NLMSG_ALIGN(rta
->rta_len
) + RTA_ALIGN(len
);
531 int addattr16(struct nlmsghdr
*n
, unsigned int maxlen
, int type
, uint16_t data
)
533 return addattr_l(n
, maxlen
, type
, &data
, sizeof(uint16_t));
536 int addattr32(struct nlmsghdr
*n
, unsigned int maxlen
, int type
, int data
)
538 return addattr_l(n
, maxlen
, type
, &data
, sizeof(uint32_t));
541 struct rtattr
*addattr_nest(struct nlmsghdr
*n
, int maxlen
, int type
)
543 struct rtattr
*nest
= NLMSG_TAIL(n
);
545 addattr_l(n
, maxlen
, type
, NULL
, 0);
549 int addattr_nest_end(struct nlmsghdr
*n
, struct rtattr
*nest
)
551 nest
->rta_len
= (uint8_t *)NLMSG_TAIL(n
) - (uint8_t *)nest
;
555 struct rtattr
*rta_nest(struct rtattr
*rta
, int maxlen
, int type
)
557 struct rtattr
*nest
= RTA_TAIL(rta
);
559 rta_addattr_l(rta
, maxlen
, type
, NULL
, 0);
563 int rta_nest_end(struct rtattr
*rta
, struct rtattr
*nest
)
565 nest
->rta_len
= (uint8_t *)RTA_TAIL(rta
) - (uint8_t *)nest
;
569 const char *nl_msg_type_to_str(uint16_t msg_type
)
571 return lookup_msg(nlmsg_str
, msg_type
, "");
574 const char *nl_rtproto_to_str(uint8_t rtproto
)
576 return lookup_msg(rtproto_str
, rtproto
, "");
579 const char *nl_family_to_str(uint8_t family
)
581 return lookup_msg(family_str
, family
, "");
584 const char *nl_rttype_to_str(uint8_t rttype
)
586 return lookup_msg(rttype_str
, rttype
, "");
589 #define NL_OK(nla, len) \
590 ((len) >= (int)sizeof(struct nlattr) \
591 && (nla)->nla_len >= sizeof(struct nlattr) \
592 && (nla)->nla_len <= (len))
593 #define NL_NEXT(nla, attrlen) \
594 ((attrlen) -= RTA_ALIGN((nla)->nla_len), \
595 (struct nlattr *)(((char *)(nla)) + RTA_ALIGN((nla)->nla_len)))
597 ((struct nlattr *)(((char *)(r)) \
598 + NLMSG_ALIGN(sizeof(struct nlmsgerr))))
600 static void netlink_parse_nlattr(struct nlattr
**tb
, int max
,
601 struct nlattr
*nla
, int len
)
603 while (NL_OK(nla
, len
)) {
604 if (nla
->nla_type
<= max
)
605 tb
[nla
->nla_type
] = nla
;
606 nla
= NL_NEXT(nla
, len
);
610 static void netlink_parse_extended_ack(struct nlmsghdr
*h
)
612 struct nlattr
*tb
[NLMSGERR_ATTR_MAX
+ 1];
613 const struct nlmsgerr
*err
=
614 (const struct nlmsgerr
*)((uint8_t *)h
616 sizeof(struct nlmsghdr
)));
617 const struct nlmsghdr
*err_nlh
= NULL
;
618 uint32_t hlen
= sizeof(*err
);
619 const char *msg
= NULL
;
622 if (!(h
->nlmsg_flags
& NLM_F_CAPPED
))
623 hlen
+= h
->nlmsg_len
- NLMSG_ALIGN(sizeof(struct nlmsghdr
));
625 memset(tb
, 0, sizeof(tb
));
626 netlink_parse_nlattr(tb
, NLMSGERR_ATTR_MAX
, NL_RTA(h
), hlen
);
628 if (tb
[NLMSGERR_ATTR_MSG
])
629 msg
= (const char *)RTA_DATA(tb
[NLMSGERR_ATTR_MSG
]);
631 if (tb
[NLMSGERR_ATTR_OFFS
]) {
632 off
= *(uint32_t *)RTA_DATA(tb
[NLMSGERR_ATTR_OFFS
]);
634 if (off
> h
->nlmsg_len
) {
635 zlog_err("Invalid offset for NLMSGERR_ATTR_OFFS\n");
636 } else if (!(h
->nlmsg_flags
& NLM_F_CAPPED
)) {
638 * Header of failed message
639 * we are not doing anything currently with it
640 * but noticing it for later.
643 zlog_warn("%s: Received %d extended Ack",
644 __PRETTY_FUNCTION__
, err_nlh
->nlmsg_type
);
648 if (msg
&& *msg
!= '\0') {
649 bool is_err
= !!err
->error
;
652 zlog_err("Extended Error: %s", msg
);
654 zlog_warn("Extended Warning: %s", msg
);
661 * Receive message from netlink interface and pass those information
662 * to the given function.
664 * filter -> Function to call to read the results
665 * nl -> netlink socket information
666 * zns -> The zebra namespace data
667 * count -> How many we should read in, 0 means as much as possible
668 * startup -> Are we reading in under startup conditions? passed to
671 int netlink_parse_info(int (*filter
)(struct nlmsghdr
*, ns_id_t
, int),
672 struct nlsock
*nl
, struct zebra_ns
*zns
, int count
,
681 char buf
[NL_RCV_PKT_BUF_SIZE
];
682 struct iovec iov
= {.iov_base
= buf
, .iov_len
= sizeof buf
};
683 struct sockaddr_nl snl
;
684 struct msghdr msg
= {.msg_name
= (void *)&snl
,
685 .msg_namelen
= sizeof snl
,
690 if (count
&& read_in
>= count
)
693 #if defined(HANDLE_NETLINK_FUZZING)
694 /* Check if reading and filename is set */
695 if (netlink_read
&& '\0' != netlink_fuzz_file
[0]) {
696 zlog_debug("Reading netlink fuzz file");
697 status
= netlink_read_file(buf
, netlink_fuzz_file
);
700 status
= recvmsg(nl
->sock
, &msg
, 0);
703 status
= recvmsg(nl
->sock
, &msg
, 0);
704 #endif /* HANDLE_NETLINK_FUZZING */
708 if (errno
== EWOULDBLOCK
|| errno
== EAGAIN
)
710 flog_err(ZEBRA_ERR_RECVMSG_OVERRUN
,
711 "%s recvmsg overrun: %s", nl
->name
,
712 safe_strerror(errno
));
714 * In this case we are screwed.
715 * There is no good way to
716 * recover zebra at this point.
723 flog_err_sys(LIB_ERR_SOCKET
, "%s EOF", nl
->name
);
727 if (msg
.msg_namelen
!= sizeof snl
) {
728 flog_err(ZEBRA_ERR_NETLINK_LENGTH_ERROR
,
729 "%s sender address length error: length %d",
730 nl
->name
, msg
.msg_namelen
);
734 if (IS_ZEBRA_DEBUG_KERNEL_MSGDUMP_RECV
) {
735 zlog_debug("%s: << netlink message dump [recv]",
737 zlog_hexdump(buf
, status
);
740 #if defined(HANDLE_NETLINK_FUZZING)
742 zlog_debug("Writing incoming netlink message");
743 netlink_write_incoming(buf
, status
,
744 netlink_file_counter
++);
746 #endif /* HANDLE_NETLINK_FUZZING */
749 for (h
= (struct nlmsghdr
*)buf
;
750 (status
>= 0 && NLMSG_OK(h
, (unsigned int)status
));
751 h
= NLMSG_NEXT(h
, status
)) {
752 /* Finish of reading. */
753 if (h
->nlmsg_type
== NLMSG_DONE
)
756 /* Error handling. */
757 if (h
->nlmsg_type
== NLMSG_ERROR
) {
758 struct nlmsgerr
*err
=
759 (struct nlmsgerr
*)NLMSG_DATA(h
);
760 int errnum
= err
->error
;
761 int msg_type
= err
->msg
.nlmsg_type
;
764 < NLMSG_LENGTH(sizeof(struct nlmsgerr
))) {
765 zlog_err("%s error: message truncated",
771 * Parse the extended information before
772 * we actually handle it.
773 * At this point in time we do not
774 * do anything other than report the
777 if (h
->nlmsg_flags
& NLM_F_ACK_TLVS
)
778 netlink_parse_extended_ack(h
);
780 /* If the error field is zero, then this is an
782 if (err
->error
== 0) {
783 if (IS_ZEBRA_DEBUG_KERNEL
) {
785 "%s: %s ACK: type=%s(%u), seq=%u, pid=%u",
786 __FUNCTION__
, nl
->name
,
788 err
->msg
.nlmsg_type
),
794 /* return if not a multipart message,
795 * otherwise continue */
796 if (!(h
->nlmsg_flags
& NLM_F_MULTI
))
802 < NLMSG_LENGTH(sizeof(struct nlmsgerr
))) {
804 ZEBRA_ERR_NETLINK_LENGTH_ERROR
,
805 "%s error: message truncated",
810 /* Deal with errors that occur because of races
811 * in link handling */
812 if (nl
== &zns
->netlink_cmd
813 && ((msg_type
== RTM_DELROUTE
814 && (-errnum
== ENODEV
815 || -errnum
== ESRCH
))
816 || (msg_type
== RTM_NEWROUTE
817 && (-errnum
== ENETDOWN
818 || -errnum
== EEXIST
)))) {
819 if (IS_ZEBRA_DEBUG_KERNEL
)
821 "%s: error: %s type=%s(%u), seq=%u, pid=%u",
823 safe_strerror(-errnum
),
832 /* We see RTM_DELNEIGH when shutting down an
833 * interface with an IPv4
834 * link-local. The kernel should have already
835 * deleted the neighbor
836 * so do not log these as an error.
838 if (msg_type
== RTM_DELNEIGH
839 || (nl
== &zns
->netlink_cmd
840 && msg_type
== RTM_NEWROUTE
842 || -errnum
== ENETUNREACH
))) {
843 /* This is known to happen in some
844 * situations, don't log
847 if (IS_ZEBRA_DEBUG_KERNEL
)
849 "%s error: %s, type=%s(%u), seq=%u, pid=%u",
851 safe_strerror(-errnum
),
859 ZEBRA_ERR_UNEXPECTED_MESSAGE
,
860 "%s error: %s, type=%s(%u), seq=%u, pid=%u",
862 safe_strerror(-errnum
),
863 nl_msg_type_to_str(msg_type
),
864 msg_type
, err
->msg
.nlmsg_seq
,
870 /* OK we got netlink message. */
871 if (IS_ZEBRA_DEBUG_KERNEL
)
873 "netlink_parse_info: %s type %s(%u), len=%d, seq=%u, pid=%u",
875 nl_msg_type_to_str(h
->nlmsg_type
),
876 h
->nlmsg_type
, h
->nlmsg_len
,
877 h
->nlmsg_seq
, h
->nlmsg_pid
);
881 * Ignore messages that maybe sent from
882 * other actors besides the kernel
884 if (snl
.nl_pid
!= 0) {
885 zlog_debug("Ignoring message from pid %u",
890 error
= (*filter
)(h
, zns
->ns_id
, startup
);
892 zlog_warn("%s filter function error", nl
->name
);
897 /* After error care. */
898 if (msg
.msg_flags
& MSG_TRUNC
) {
899 flog_err(ZEBRA_ERR_NETLINK_LENGTH_ERROR
,
900 "%s error: message truncated", nl
->name
);
904 flog_err(ZEBRA_ERR_NETLINK_LENGTH_ERROR
,
905 "%s error: data remnant size %d", nl
->name
,
916 * sendmsg() to netlink socket then recvmsg().
917 * Calls netlink_parse_info to parse returned data
919 * filter -> The filter to read final results from kernel
920 * nlmsghdr -> The data to send to the kernel
921 * nl -> The netlink socket information
922 * zns -> The zebra namespace information
923 * startup -> Are we reading in under startup conditions
924 * This is passed through eventually to filter.
926 int netlink_talk(int (*filter
)(struct nlmsghdr
*, ns_id_t
, int startup
),
927 struct nlmsghdr
*n
, struct nlsock
*nl
, struct zebra_ns
*zns
,
931 struct sockaddr_nl snl
;
936 memset(&snl
, 0, sizeof snl
);
937 memset(&iov
, 0, sizeof iov
);
938 memset(&msg
, 0, sizeof msg
);
941 iov
.iov_len
= n
->nlmsg_len
;
942 msg
.msg_name
= (void *)&snl
;
943 msg
.msg_namelen
= sizeof snl
;
947 snl
.nl_family
= AF_NETLINK
;
949 n
->nlmsg_seq
= ++nl
->seq
;
950 n
->nlmsg_pid
= nl
->snl
.nl_pid
;
952 if (IS_ZEBRA_DEBUG_KERNEL
)
954 "netlink_talk: %s type %s(%u), len=%d seq=%u flags 0x%x",
955 nl
->name
, nl_msg_type_to_str(n
->nlmsg_type
),
956 n
->nlmsg_type
, n
->nlmsg_len
, n
->nlmsg_seq
,
959 /* Send message to netlink interface. */
960 if (zserv_privs
.change(ZPRIVS_RAISE
))
961 flog_err(LIB_ERR_PRIVILEGES
, "Can't raise privileges");
962 status
= sendmsg(nl
->sock
, &msg
, 0);
964 if (zserv_privs
.change(ZPRIVS_LOWER
))
965 flog_err(LIB_ERR_PRIVILEGES
, "Can't lower privileges");
967 if (IS_ZEBRA_DEBUG_KERNEL_MSGDUMP_SEND
) {
968 zlog_debug("%s: >> netlink message dump [sent]", __func__
);
969 zlog_hexdump(n
, n
->nlmsg_len
);
973 flog_err_sys(LIB_ERR_SOCKET
, "netlink_talk sendmsg() error: %s",
974 safe_strerror(save_errno
));
980 * Get reply from netlink socket.
981 * The reply should either be an acknowlegement or an error.
983 return netlink_parse_info(filter
, nl
, zns
, 0, startup
);
986 /* Issue request message to kernel via netlink socket. GET messages
987 * are issued through this interface.
989 int netlink_request(struct nlsock
*nl
, struct nlmsghdr
*n
)
992 struct sockaddr_nl snl
;
995 /* Check netlink socket. */
997 flog_err_sys(LIB_ERR_SOCKET
, "%s socket isn't active.",
1002 /* Fill common fields for all requests. */
1003 n
->nlmsg_flags
= NLM_F_ROOT
| NLM_F_MATCH
| NLM_F_REQUEST
;
1004 n
->nlmsg_pid
= nl
->snl
.nl_pid
;
1005 n
->nlmsg_seq
= ++nl
->seq
;
1007 memset(&snl
, 0, sizeof snl
);
1008 snl
.nl_family
= AF_NETLINK
;
1010 /* Raise capabilities and send message, then lower capabilities. */
1011 if (zserv_privs
.change(ZPRIVS_RAISE
)) {
1012 flog_err(LIB_ERR_PRIVILEGES
, "Can't raise privileges");
1016 ret
= sendto(nl
->sock
, (void *)n
, n
->nlmsg_len
, 0,
1017 (struct sockaddr
*)&snl
, sizeof snl
);
1020 if (zserv_privs
.change(ZPRIVS_LOWER
))
1021 flog_err(LIB_ERR_PRIVILEGES
, "Can't lower privileges");
1024 flog_err_sys(LIB_ERR_SOCKET
, "%s sendto failed: %s", nl
->name
,
1025 safe_strerror(save_errno
));
1032 /* Exported interface function. This function simply calls
1033 netlink_socket (). */
1034 void kernel_init(struct zebra_ns
*zns
)
1036 unsigned long groups
;
1037 #if defined SOL_NETLINK
1042 * Initialize netlink sockets
1044 * If RTMGRP_XXX exists use that, but at some point
1045 * I think the kernel developers realized that
1046 * keeping track of all the different values would
1047 * lead to confusion, so we need to convert the
1048 * RTNLGRP_XXX to a bit position for ourself
1050 groups
= RTMGRP_LINK
|
1052 RTMGRP_IPV4_IFADDR
|
1054 RTMGRP_IPV6_IFADDR
|
1055 RTMGRP_IPV4_MROUTE
|
1057 (1 << (RTNLGRP_IPV4_RULE
- 1)) |
1058 (1 << (RTNLGRP_IPV6_RULE
- 1));
1060 snprintf(zns
->netlink
.name
, sizeof(zns
->netlink
.name
),
1061 "netlink-listen (NS %u)", zns
->ns_id
);
1062 zns
->netlink
.sock
= -1;
1063 if (netlink_socket(&zns
->netlink
, groups
, zns
->ns_id
) < 0) {
1064 zlog_err("Failure to create %s socket",
1069 snprintf(zns
->netlink_cmd
.name
, sizeof(zns
->netlink_cmd
.name
),
1070 "netlink-cmd (NS %u)", zns
->ns_id
);
1071 zns
->netlink_cmd
.sock
= -1;
1072 if (netlink_socket(&zns
->netlink_cmd
, 0, zns
->ns_id
) < 0) {
1073 zlog_err("Failure to create %s socket",
1074 zns
->netlink_cmd
.name
);
1079 * SOL_NETLINK is not available on all platforms yet
1080 * apparently. It's in bits/socket.h which I am not
1081 * sure that we want to pull into our build system.
1083 #if defined SOL_NETLINK
1085 * Let's tell the kernel that we want to receive extended
1086 * ACKS over our command socket
1089 ret
= setsockopt(zns
->netlink_cmd
.sock
, SOL_NETLINK
, NETLINK_EXT_ACK
,
1093 zlog_notice("Registration for extended ACK failed : %d %s",
1094 errno
, safe_strerror(errno
));
1097 /* Register kernel socket. */
1098 if (fcntl(zns
->netlink
.sock
, F_SETFL
, O_NONBLOCK
) < 0)
1099 flog_err_sys(LIB_ERR_SOCKET
, "Can't set %s socket flags: %s",
1100 zns
->netlink
.name
, safe_strerror(errno
));
1102 if (fcntl(zns
->netlink_cmd
.sock
, F_SETFL
, O_NONBLOCK
) < 0)
1103 zlog_err("Can't set %s socket error: %s(%d)",
1104 zns
->netlink_cmd
.name
, safe_strerror(errno
), errno
);
1106 /* Set receive buffer size if it's set from command line */
1108 netlink_recvbuf(&zns
->netlink
, nl_rcvbufsize
);
1110 assert(zns
->netlink
.sock
>= 0);
1111 netlink_install_filter(zns
->netlink
.sock
,
1112 zns
->netlink_cmd
.snl
.nl_pid
);
1113 zns
->t_netlink
= NULL
;
1115 thread_add_read(zebrad
.master
, kernel_read
, zns
,
1116 zns
->netlink
.sock
, &zns
->t_netlink
);
1121 void kernel_terminate(struct zebra_ns
*zns
)
1123 THREAD_READ_OFF(zns
->t_netlink
);
1125 if (zns
->netlink
.sock
>= 0) {
1126 close(zns
->netlink
.sock
);
1127 zns
->netlink
.sock
= -1;
1130 if (zns
->netlink_cmd
.sock
>= 0) {
1131 close(zns
->netlink_cmd
.sock
);
1132 zns
->netlink_cmd
.sock
= -1;
1136 #endif /* HAVE_NETLINK */