1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
5 #include "sd-netlink.h"
7 #include "alloc-util.h"
12 #include "netlink-genl.h"
13 #include "netlink-internal.h"
14 #include "netlink-slot.h"
15 #include "netlink-util.h"
16 #include "process-util.h"
17 #include "socket-util.h"
18 #include "string-util.h"
20 /* Some really high limit, to catch programming errors */
21 #define REPLY_CALLBACKS_MAX UINT16_MAX
23 static int netlink_new(sd_netlink
**ret
) {
24 _cleanup_(sd_netlink_unrefp
) sd_netlink
*nl
= NULL
;
26 assert_return(ret
, -EINVAL
);
28 nl
= new(sd_netlink
, 1);
35 .sockaddr
.nl
.nl_family
= AF_NETLINK
,
36 .original_pid
= getpid_cached(),
39 /* Kernel change notification messages have sequence number 0. We want to avoid that with our
40 * own serials, in order not to get confused when matching up kernel replies to our earlier
43 * Moreover, when using netlink socket activation (i.e. where PID 1 binds an AF_NETLINK
44 * socket for us and passes it to us across execve()) and we get restarted multiple times
45 * while the socket sticks around we might get confused by replies from earlier runs coming
46 * in late — which is pretty likely if we'd start our sequence numbers always from 1. Hence,
47 * let's start with a value based on the system clock. This should make collisions much less
48 * likely (though still theoretically possible). We use a 32 bit µs counter starting at boot
49 * for this (and explicitly exclude the zero, see above). This counter will wrap around after
50 * a bit more than 1h, but that's hopefully OK as the kernel shouldn't take that long to
51 * reply to our requests.
53 * We only pick the initial start value this way. For each message we simply increase the
54 * sequence number by 1. This means we could enqueue 1 netlink message per µs without risking
55 * collisions, which should be OK.
57 * Note this means the serials will be in the range 1…UINT32_MAX here.
59 * (In an ideal world we'd attach the current serial counter to the netlink socket itself
60 * somehow, to avoid all this, but I couldn't come up with a nice way to do this) */
61 .serial
= (uint32_t) (now(CLOCK_MONOTONIC
) % UINT32_MAX
) + 1,
68 int sd_netlink_open_fd(sd_netlink
**ret
, int fd
) {
69 _cleanup_(sd_netlink_unrefp
) sd_netlink
*nl
= NULL
;
72 assert_return(ret
, -EINVAL
);
73 assert_return(fd
>= 0, -EBADF
);
79 r
= getsockopt_int(fd
, SOL_SOCKET
, SO_PROTOCOL
, &protocol
);
84 nl
->protocol
= protocol
;
86 r
= setsockopt_int(fd
, SOL_NETLINK
, NETLINK_EXT_ACK
, true);
88 log_debug_errno(r
, "sd-netlink: Failed to enable NETLINK_EXT_ACK option, ignoring: %m");
90 r
= setsockopt_int(fd
, SOL_NETLINK
, NETLINK_GET_STRICT_CHK
, true);
92 log_debug_errno(r
, "sd-netlink: Failed to enable NETLINK_GET_STRICT_CHK option, ignoring: %m");
96 nl
->fd
= -EBADF
; /* on failure, the caller remains owner of the fd, hence don't close it here */
106 int sd_netlink_open(sd_netlink
**ret
) {
107 return netlink_open_family(ret
, NETLINK_ROUTE
);
110 int sd_netlink_increase_rxbuf(sd_netlink
*nl
, size_t size
) {
111 assert_return(nl
, -EINVAL
);
112 assert_return(!netlink_pid_changed(nl
), -ECHILD
);
114 return fd_increase_rxbuf(nl
->fd
, size
);
117 static sd_netlink
*netlink_free(sd_netlink
*nl
) {
122 ordered_set_free(nl
->rqueue
);
123 hashmap_free(nl
->rqueue_by_serial
);
124 hashmap_free(nl
->rqueue_partial_by_serial
);
127 while ((s
= nl
->slots
)) {
129 netlink_slot_disconnect(s
, true);
131 hashmap_free(nl
->reply_callbacks
);
132 prioq_free(nl
->reply_callbacks_prioq
);
134 sd_event_source_unref(nl
->io_event_source
);
135 sd_event_source_unref(nl
->time_event_source
);
136 sd_event_unref(nl
->event
);
138 hashmap_free(nl
->broadcast_group_refs
);
140 genl_clear_family(nl
);
146 DEFINE_TRIVIAL_REF_UNREF_FUNC(sd_netlink
, sd_netlink
, netlink_free
);
150 sd_netlink_message
*message
,
155 assert_return(nl
, -EINVAL
);
156 assert_return(!netlink_pid_changed(nl
), -ECHILD
);
157 assert_return(message
, -EINVAL
);
158 assert_return(!message
->sealed
, -EPERM
);
160 netlink_seal_message(nl
, message
);
162 r
= socket_write_message(nl
, message
);
167 *serial
= message_get_serial(message
);
172 static int dispatch_rqueue(sd_netlink
*nl
, sd_netlink_message
**ret
) {
173 sd_netlink_message
*m
;
179 if (ordered_set_size(nl
->rqueue
) <= 0) {
180 /* Try to read a new message */
181 r
= socket_read_message(nl
);
182 if (r
== -ENOBUFS
) /* FIXME: ignore buffer overruns for now */
183 log_debug_errno(r
, "sd-netlink: Got ENOBUFS from netlink socket, ignoring.");
188 /* Dispatch a queued message */
189 m
= ordered_set_steal_first(nl
->rqueue
);
191 sd_netlink_message_unref(hashmap_remove_value(nl
->rqueue_by_serial
, UINT32_TO_PTR(message_get_serial(m
)), m
));
196 static int process_timeout(sd_netlink
*nl
) {
197 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
198 struct reply_callback
*c
;
199 sd_netlink_slot
*slot
;
205 c
= prioq_peek(nl
->reply_callbacks_prioq
);
209 n
= now(CLOCK_MONOTONIC
);
213 r
= message_new_synthetic_error(nl
, -ETIMEDOUT
, c
->serial
, &m
);
217 assert_se(prioq_pop(nl
->reply_callbacks_prioq
) == c
);
219 hashmap_remove(nl
->reply_callbacks
, UINT32_TO_PTR(c
->serial
));
221 slot
= container_of(c
, sd_netlink_slot
, reply_callback
);
223 r
= c
->callback(nl
, m
, slot
->userdata
);
225 log_debug_errno(r
, "sd-netlink: timedout callback %s%s%sfailed: %m",
226 slot
->description
? "'" : "",
227 strempty(slot
->description
),
228 slot
->description
? "' " : "");
231 netlink_slot_disconnect(slot
, true);
236 static int process_reply(sd_netlink
*nl
, sd_netlink_message
*m
) {
237 struct reply_callback
*c
;
238 sd_netlink_slot
*slot
;
246 serial
= message_get_serial(m
);
247 c
= hashmap_remove(nl
->reply_callbacks
, UINT32_TO_PTR(serial
));
251 if (c
->timeout
!= 0) {
252 prioq_remove(nl
->reply_callbacks_prioq
, c
, &c
->prioq_idx
);
256 r
= sd_netlink_message_get_type(m
, &type
);
260 if (type
== NLMSG_DONE
)
263 slot
= container_of(c
, sd_netlink_slot
, reply_callback
);
265 r
= c
->callback(nl
, m
, slot
->userdata
);
267 log_debug_errno(r
, "sd-netlink: reply callback %s%s%sfailed: %m",
268 slot
->description
? "'" : "",
269 strempty(slot
->description
),
270 slot
->description
? "' " : "");
273 netlink_slot_disconnect(slot
, true);
278 static int process_match(sd_netlink
*nl
, sd_netlink_message
*m
) {
286 r
= sd_netlink_message_get_type(m
, &type
);
290 if (m
->protocol
== NETLINK_GENERIC
) {
291 r
= sd_genl_message_get_command(nl
, m
, &cmd
);
297 LIST_FOREACH(match_callbacks
, c
, nl
->match_callbacks
) {
298 sd_netlink_slot
*slot
;
303 if (c
->cmd
!= 0 && c
->cmd
!= cmd
)
306 for (size_t i
= 0; i
< c
->n_groups
; i
++)
307 if (c
->groups
[i
] == m
->multicast_group
) {
315 slot
= container_of(c
, sd_netlink_slot
, match_callback
);
317 r
= c
->callback(nl
, m
, slot
->userdata
);
319 log_debug_errno(r
, "sd-netlink: match callback %s%s%sfailed: %m",
320 slot
->description
? "'" : "",
321 strempty(slot
->description
),
322 slot
->description
? "' " : "");
330 static int process_running(sd_netlink
*nl
, sd_netlink_message
**ret
) {
331 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
336 r
= process_timeout(nl
);
340 r
= dispatch_rqueue(nl
, &m
);
346 if (sd_netlink_message_is_broadcast(m
))
347 r
= process_match(nl
, m
);
349 r
= process_reply(nl
, m
);
368 int sd_netlink_process(sd_netlink
*nl
, sd_netlink_message
**ret
) {
369 NETLINK_DONT_DESTROY(nl
);
372 assert_return(nl
, -EINVAL
);
373 assert_return(!netlink_pid_changed(nl
), -ECHILD
);
374 assert_return(!nl
->processing
, -EBUSY
);
376 nl
->processing
= true;
377 r
= process_running(nl
, ret
);
378 nl
->processing
= false;
383 static usec_t
calc_elapse(uint64_t usec
) {
384 if (usec
== UINT64_MAX
)
388 usec
= NETLINK_DEFAULT_TIMEOUT_USEC
;
390 return usec_add(now(CLOCK_MONOTONIC
), usec
);
393 static int netlink_poll(sd_netlink
*nl
, bool need_more
, usec_t timeout_usec
) {
394 usec_t m
= USEC_INFINITY
;
399 e
= sd_netlink_get_events(nl
);
404 /* Caller wants more data, and doesn't care about
405 * what's been read or any other timeouts. */
410 /* Caller wants to process if there is something to
411 * process, but doesn't care otherwise */
413 r
= sd_netlink_get_timeout(nl
, &until
);
417 m
= usec_sub_unsigned(until
, now(CLOCK_MONOTONIC
));
420 r
= fd_wait_for_event(nl
->fd
, e
, MIN(m
, timeout_usec
));
427 int sd_netlink_wait(sd_netlink
*nl
, uint64_t timeout_usec
) {
430 assert_return(nl
, -EINVAL
);
431 assert_return(!netlink_pid_changed(nl
), -ECHILD
);
433 if (ordered_set_size(nl
->rqueue
) > 0)
436 r
= netlink_poll(nl
, false, timeout_usec
);
437 if (r
< 0 && ERRNO_IS_TRANSIENT(r
)) /* Convert EINTR to "something happened" and give user a chance to run some code before calling back into us */
442 static int timeout_compare(const void *a
, const void *b
) {
443 const struct reply_callback
*x
= a
, *y
= b
;
445 if (x
->timeout
!= 0 && y
->timeout
== 0)
448 if (x
->timeout
== 0 && y
->timeout
!= 0)
451 return CMP(x
->timeout
, y
->timeout
);
454 int sd_netlink_call_async(
456 sd_netlink_slot
**ret_slot
,
457 sd_netlink_message
*m
,
458 sd_netlink_message_handler_t callback
,
459 sd_netlink_destroy_t destroy_callback
,
462 const char *description
) {
464 _cleanup_free_ sd_netlink_slot
*slot
= NULL
;
467 assert_return(nl
, -EINVAL
);
468 assert_return(m
, -EINVAL
);
469 assert_return(callback
, -EINVAL
);
470 assert_return(!netlink_pid_changed(nl
), -ECHILD
);
472 if (hashmap_size(nl
->reply_callbacks
) >= REPLY_CALLBACKS_MAX
)
475 r
= hashmap_ensure_allocated(&nl
->reply_callbacks
, &trivial_hash_ops
);
479 if (usec
!= UINT64_MAX
) {
480 r
= prioq_ensure_allocated(&nl
->reply_callbacks_prioq
, timeout_compare
);
485 r
= netlink_slot_allocate(nl
, !ret_slot
, NETLINK_REPLY_CALLBACK
, sizeof(struct reply_callback
), userdata
, description
, &slot
);
489 slot
->reply_callback
.callback
= callback
;
490 slot
->reply_callback
.timeout
= calc_elapse(usec
);
492 k
= sd_netlink_send(nl
, m
, &slot
->reply_callback
.serial
);
496 r
= hashmap_put(nl
->reply_callbacks
, UINT32_TO_PTR(slot
->reply_callback
.serial
), &slot
->reply_callback
);
500 if (slot
->reply_callback
.timeout
!= 0) {
501 r
= prioq_put(nl
->reply_callbacks_prioq
, &slot
->reply_callback
, &slot
->reply_callback
.prioq_idx
);
503 (void) hashmap_remove(nl
->reply_callbacks
, UINT32_TO_PTR(slot
->reply_callback
.serial
));
508 /* Set this at last. Otherwise, some failures in above would call destroy_callback but some would not. */
509 slot
->destroy_callback
= destroy_callback
;
523 sd_netlink_message
**ret
) {
528 assert_return(nl
, -EINVAL
);
529 assert_return(!netlink_pid_changed(nl
), -ECHILD
);
531 timeout
= calc_elapse(usec
);
534 _cleanup_(sd_netlink_message_unrefp
) sd_netlink_message
*m
= NULL
;
537 m
= hashmap_remove(nl
->rqueue_by_serial
, UINT32_TO_PTR(serial
));
541 /* found a match, remove from rqueue and return it */
542 sd_netlink_message_unref(ordered_set_remove(nl
->rqueue
, m
));
544 r
= sd_netlink_message_get_errno(m
);
548 r
= sd_netlink_message_get_type(m
, &type
);
552 if (type
== NLMSG_DONE
) {
563 r
= socket_read_message(nl
);
567 /* received message, so try to process straight away */
573 n
= now(CLOCK_MONOTONIC
);
577 left
= usec_sub_unsigned(timeout
, n
);
579 left
= USEC_INFINITY
;
581 r
= netlink_poll(nl
, true, left
);
591 sd_netlink_message
*message
,
593 sd_netlink_message
**ret
) {
598 assert_return(nl
, -EINVAL
);
599 assert_return(!netlink_pid_changed(nl
), -ECHILD
);
600 assert_return(message
, -EINVAL
);
602 r
= sd_netlink_send(nl
, message
, &serial
);
606 return sd_netlink_read(nl
, serial
, usec
, ret
);
609 int sd_netlink_get_events(sd_netlink
*nl
) {
610 assert_return(nl
, -EINVAL
);
611 assert_return(!netlink_pid_changed(nl
), -ECHILD
);
613 return ordered_set_size(nl
->rqueue
) == 0 ? POLLIN
: 0;
616 int sd_netlink_get_timeout(sd_netlink
*nl
, uint64_t *timeout_usec
) {
617 struct reply_callback
*c
;
619 assert_return(nl
, -EINVAL
);
620 assert_return(timeout_usec
, -EINVAL
);
621 assert_return(!netlink_pid_changed(nl
), -ECHILD
);
623 if (ordered_set_size(nl
->rqueue
) > 0) {
628 c
= prioq_peek(nl
->reply_callbacks_prioq
);
630 *timeout_usec
= UINT64_MAX
;
634 *timeout_usec
= c
->timeout
;
638 static int io_callback(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
639 sd_netlink
*nl
= ASSERT_PTR(userdata
);
642 r
= sd_netlink_process(nl
, NULL
);
649 static int time_callback(sd_event_source
*s
, uint64_t usec
, void *userdata
) {
650 sd_netlink
*nl
= ASSERT_PTR(userdata
);
653 r
= sd_netlink_process(nl
, NULL
);
660 static int prepare_callback(sd_event_source
*s
, void *userdata
) {
661 sd_netlink
*nl
= ASSERT_PTR(userdata
);
667 r
= sd_netlink_get_events(nl
);
671 r
= sd_event_source_set_io_events(nl
->io_event_source
, r
);
675 enabled
= sd_netlink_get_timeout(nl
, &until
);
679 r
= sd_event_source_set_time(nl
->time_event_source
, until
);
684 r
= sd_event_source_set_enabled(nl
->time_event_source
,
685 enabled
> 0 ? SD_EVENT_ONESHOT
: SD_EVENT_OFF
);
692 int sd_netlink_attach_event(sd_netlink
*nl
, sd_event
*event
, int64_t priority
) {
695 assert_return(nl
, -EINVAL
);
696 assert_return(!nl
->event
, -EBUSY
);
698 assert(!nl
->io_event_source
);
699 assert(!nl
->time_event_source
);
702 nl
->event
= sd_event_ref(event
);
704 r
= sd_event_default(&nl
->event
);
709 r
= sd_event_add_io(nl
->event
, &nl
->io_event_source
, nl
->fd
, 0, io_callback
, nl
);
713 r
= sd_event_source_set_priority(nl
->io_event_source
, priority
);
717 r
= sd_event_source_set_description(nl
->io_event_source
, "netlink-receive-message");
721 r
= sd_event_source_set_prepare(nl
->io_event_source
, prepare_callback
);
725 r
= sd_event_add_time(nl
->event
, &nl
->time_event_source
, CLOCK_MONOTONIC
, 0, 0, time_callback
, nl
);
729 r
= sd_event_source_set_priority(nl
->time_event_source
, priority
);
733 r
= sd_event_source_set_description(nl
->time_event_source
, "netlink-timer");
740 sd_netlink_detach_event(nl
);
744 int sd_netlink_detach_event(sd_netlink
*nl
) {
745 assert_return(nl
, -EINVAL
);
746 assert_return(nl
->event
, -ENXIO
);
748 nl
->io_event_source
= sd_event_source_unref(nl
->io_event_source
);
750 nl
->time_event_source
= sd_event_source_unref(nl
->time_event_source
);
752 nl
->event
= sd_event_unref(nl
->event
);
757 int netlink_add_match_internal(
759 sd_netlink_slot
**ret_slot
,
760 const uint32_t *groups
,
764 sd_netlink_message_handler_t callback
,
765 sd_netlink_destroy_t destroy_callback
,
767 const char *description
) {
769 _cleanup_free_ sd_netlink_slot
*slot
= NULL
;
773 assert(n_groups
> 0);
775 for (size_t i
= 0; i
< n_groups
; i
++) {
776 r
= socket_broadcast_group_ref(nl
, groups
[i
]);
781 r
= netlink_slot_allocate(nl
, !ret_slot
, NETLINK_MATCH_CALLBACK
, sizeof(struct match_callback
),
782 userdata
, description
, &slot
);
786 slot
->match_callback
.groups
= newdup(uint32_t, groups
, n_groups
);
787 if (!slot
->match_callback
.groups
)
790 slot
->match_callback
.n_groups
= n_groups
;
791 slot
->match_callback
.callback
= callback
;
792 slot
->match_callback
.type
= type
;
793 slot
->match_callback
.cmd
= cmd
;
795 LIST_PREPEND(match_callbacks
, nl
->match_callbacks
, &slot
->match_callback
);
797 /* Set this at last. Otherwise, some failures in above call the destroy callback but some do not. */
798 slot
->destroy_callback
= destroy_callback
;
807 int sd_netlink_add_match(
809 sd_netlink_slot
**ret_slot
,
811 sd_netlink_message_handler_t callback
,
812 sd_netlink_destroy_t destroy_callback
,
814 const char *description
) {
816 static const uint32_t
817 address_groups
[] = { RTNLGRP_IPV4_IFADDR
, RTNLGRP_IPV6_IFADDR
, },
818 link_groups
[] = { RTNLGRP_LINK
, },
819 neighbor_groups
[] = { RTNLGRP_NEIGH
, },
820 nexthop_groups
[] = { RTNLGRP_NEXTHOP
, },
821 route_groups
[] = { RTNLGRP_IPV4_ROUTE
, RTNLGRP_IPV6_ROUTE
, },
822 rule_groups
[] = { RTNLGRP_IPV4_RULE
, RTNLGRP_IPV6_RULE
, },
823 tc_groups
[] = { RTNLGRP_TC
};
824 const uint32_t *groups
;
827 assert_return(rtnl
, -EINVAL
);
828 assert_return(callback
, -EINVAL
);
829 assert_return(!netlink_pid_changed(rtnl
), -ECHILD
);
834 groups
= link_groups
;
835 n_groups
= ELEMENTSOF(link_groups
);
839 groups
= address_groups
;
840 n_groups
= ELEMENTSOF(address_groups
);
844 groups
= neighbor_groups
;
845 n_groups
= ELEMENTSOF(neighbor_groups
);
849 groups
= route_groups
;
850 n_groups
= ELEMENTSOF(route_groups
);
854 groups
= rule_groups
;
855 n_groups
= ELEMENTSOF(rule_groups
);
859 groups
= nexthop_groups
;
860 n_groups
= ELEMENTSOF(nexthop_groups
);
867 n_groups
= ELEMENTSOF(tc_groups
);
873 return netlink_add_match_internal(rtnl
, ret_slot
, groups
, n_groups
, type
, 0, callback
,
874 destroy_callback
, userdata
, description
);
877 int sd_netlink_attach_filter(sd_netlink
*nl
, size_t len
, const struct sock_filter
*filter
) {
878 assert_return(nl
, -EINVAL
);
879 assert_return(len
== 0 || filter
, -EINVAL
);
881 if (setsockopt(nl
->fd
, SOL_SOCKET
,
882 len
== 0 ? SO_DETACH_FILTER
: SO_ATTACH_FILTER
,
883 &(struct sock_fprog
) {
885 .filter
= (struct sock_filter
*) filter
,
886 }, sizeof(struct sock_fprog
)) < 0)