1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 // Copyright (C) 2018 Facebook
14 #include <bpf/libbpf.h>
16 #include <linux/rtnetlink.h>
17 #include <linux/socket.h>
18 #include <linux/tc_act/tc_bpf.h>
19 #include <sys/socket.h>
21 #include <sys/types.h>
23 #include "bpf/nlattr.h"
25 #include "netlink_dumper.h"
28 #define SOL_NETLINK 270
31 struct ip_devname_ifindex {
37 struct ip_devname_ifindex *devices;
43 struct tc_kind_handle {
49 struct tc_kind_handle *handle_array;
61 struct bpf_attach_info {
62 __u32 flow_dissector_id;
65 enum net_attach_type {
67 NET_ATTACH_TYPE_XDP_GENERIC,
68 NET_ATTACH_TYPE_XDP_DRIVER,
69 NET_ATTACH_TYPE_XDP_OFFLOAD,
72 static const char * const attach_type_strings[] = {
73 [NET_ATTACH_TYPE_XDP] = "xdp",
74 [NET_ATTACH_TYPE_XDP_GENERIC] = "xdpgeneric",
75 [NET_ATTACH_TYPE_XDP_DRIVER] = "xdpdrv",
76 [NET_ATTACH_TYPE_XDP_OFFLOAD] = "xdpoffload",
79 const size_t net_attach_type_size = ARRAY_SIZE(attach_type_strings);
81 static enum net_attach_type parse_attach_type(const char *str)
83 enum net_attach_type type;
85 for (type = 0; type < net_attach_type_size; type++) {
86 if (attach_type_strings[type] &&
87 is_prefix(str, attach_type_strings[type]))
91 return net_attach_type_size;
94 typedef int (*dump_nlmsg_t)(void *cookie, void *msg, struct nlattr **tb);
96 typedef int (*__dump_nlmsg_t)(struct nlmsghdr *nlmsg, dump_nlmsg_t, void *cookie);
98 static int netlink_open(__u32 *nl_pid)
100 struct sockaddr_nl sa;
105 memset(&sa, 0, sizeof(sa));
106 sa.nl_family = AF_NETLINK;
108 sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
112 if (setsockopt(sock, SOL_NETLINK, NETLINK_EXT_ACK,
113 &one, sizeof(one)) < 0) {
114 p_err("Netlink error reporting not supported");
117 if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
122 addrlen = sizeof(sa);
123 if (getsockname(sock, (struct sockaddr *)&sa, &addrlen) < 0) {
128 if (addrlen != sizeof(sa)) {
129 ret = -LIBBPF_ERRNO__INTERNAL;
141 static int netlink_recv(int sock, __u32 nl_pid, __u32 seq,
142 __dump_nlmsg_t _fn, dump_nlmsg_t fn,
145 bool multipart = true;
146 struct nlmsgerr *err;
153 len = recv(sock, buf, sizeof(buf), 0);
162 for (nh = (struct nlmsghdr *)buf; NLMSG_OK(nh, (unsigned int)len);
163 nh = NLMSG_NEXT(nh, len)) {
164 if (nh->nlmsg_pid != nl_pid) {
165 ret = -LIBBPF_ERRNO__WRNGPID;
168 if (nh->nlmsg_seq != seq) {
169 ret = -LIBBPF_ERRNO__INVSEQ;
172 if (nh->nlmsg_flags & NLM_F_MULTI)
174 switch (nh->nlmsg_type) {
176 err = (struct nlmsgerr *)NLMSG_DATA(nh);
180 libbpf_nla_dump_errormsg(nh);
188 ret = _fn(nh, fn, cookie);
199 static int __dump_class_nlmsg(struct nlmsghdr *nlh,
200 dump_nlmsg_t dump_class_nlmsg,
203 struct nlattr *tb[TCA_MAX + 1], *attr;
204 struct tcmsg *t = NLMSG_DATA(nlh);
207 len = nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*t));
208 attr = (struct nlattr *) ((void *) t + NLMSG_ALIGN(sizeof(*t)));
209 if (libbpf_nla_parse(tb, TCA_MAX, attr, len, NULL) != 0)
210 return -LIBBPF_ERRNO__NLPARSE;
212 return dump_class_nlmsg(cookie, t, tb);
215 static int netlink_get_class(int sock, unsigned int nl_pid, int ifindex,
216 dump_nlmsg_t dump_class_nlmsg, void *cookie)
222 .nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct tcmsg)),
223 .nlh.nlmsg_type = RTM_GETTCLASS,
224 .nlh.nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST,
225 .t.tcm_family = AF_UNSPEC,
226 .t.tcm_ifindex = ifindex,
228 int seq = time(NULL);
230 req.nlh.nlmsg_seq = seq;
231 if (send(sock, &req, req.nlh.nlmsg_len, 0) < 0)
234 return netlink_recv(sock, nl_pid, seq, __dump_class_nlmsg,
235 dump_class_nlmsg, cookie);
238 static int __dump_qdisc_nlmsg(struct nlmsghdr *nlh,
239 dump_nlmsg_t dump_qdisc_nlmsg,
242 struct nlattr *tb[TCA_MAX + 1], *attr;
243 struct tcmsg *t = NLMSG_DATA(nlh);
246 len = nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*t));
247 attr = (struct nlattr *) ((void *) t + NLMSG_ALIGN(sizeof(*t)));
248 if (libbpf_nla_parse(tb, TCA_MAX, attr, len, NULL) != 0)
249 return -LIBBPF_ERRNO__NLPARSE;
251 return dump_qdisc_nlmsg(cookie, t, tb);
254 static int netlink_get_qdisc(int sock, unsigned int nl_pid, int ifindex,
255 dump_nlmsg_t dump_qdisc_nlmsg, void *cookie)
261 .nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct tcmsg)),
262 .nlh.nlmsg_type = RTM_GETQDISC,
263 .nlh.nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST,
264 .t.tcm_family = AF_UNSPEC,
265 .t.tcm_ifindex = ifindex,
267 int seq = time(NULL);
269 req.nlh.nlmsg_seq = seq;
270 if (send(sock, &req, req.nlh.nlmsg_len, 0) < 0)
273 return netlink_recv(sock, nl_pid, seq, __dump_qdisc_nlmsg,
274 dump_qdisc_nlmsg, cookie);
277 static int __dump_filter_nlmsg(struct nlmsghdr *nlh,
278 dump_nlmsg_t dump_filter_nlmsg,
281 struct nlattr *tb[TCA_MAX + 1], *attr;
282 struct tcmsg *t = NLMSG_DATA(nlh);
285 len = nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*t));
286 attr = (struct nlattr *) ((void *) t + NLMSG_ALIGN(sizeof(*t)));
287 if (libbpf_nla_parse(tb, TCA_MAX, attr, len, NULL) != 0)
288 return -LIBBPF_ERRNO__NLPARSE;
290 return dump_filter_nlmsg(cookie, t, tb);
293 static int netlink_get_filter(int sock, unsigned int nl_pid, int ifindex, int handle,
294 dump_nlmsg_t dump_filter_nlmsg, void *cookie)
300 .nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct tcmsg)),
301 .nlh.nlmsg_type = RTM_GETTFILTER,
302 .nlh.nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST,
303 .t.tcm_family = AF_UNSPEC,
304 .t.tcm_ifindex = ifindex,
305 .t.tcm_parent = handle,
307 int seq = time(NULL);
309 req.nlh.nlmsg_seq = seq;
310 if (send(sock, &req, req.nlh.nlmsg_len, 0) < 0)
313 return netlink_recv(sock, nl_pid, seq, __dump_filter_nlmsg,
314 dump_filter_nlmsg, cookie);
317 static int __dump_link_nlmsg(struct nlmsghdr *nlh,
318 dump_nlmsg_t dump_link_nlmsg, void *cookie)
320 struct nlattr *tb[IFLA_MAX + 1], *attr;
321 struct ifinfomsg *ifi = NLMSG_DATA(nlh);
324 len = nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*ifi));
325 attr = (struct nlattr *) ((void *) ifi + NLMSG_ALIGN(sizeof(*ifi)));
326 if (libbpf_nla_parse(tb, IFLA_MAX, attr, len, NULL) != 0)
327 return -LIBBPF_ERRNO__NLPARSE;
329 return dump_link_nlmsg(cookie, ifi, tb);
332 static int netlink_get_link(int sock, unsigned int nl_pid,
333 dump_nlmsg_t dump_link_nlmsg, void *cookie)
337 struct ifinfomsg ifm;
339 .nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg)),
340 .nlh.nlmsg_type = RTM_GETLINK,
341 .nlh.nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST,
342 .ifm.ifi_family = AF_PACKET,
344 int seq = time(NULL);
346 req.nlh.nlmsg_seq = seq;
347 if (send(sock, &req, req.nlh.nlmsg_len, 0) < 0)
350 return netlink_recv(sock, nl_pid, seq, __dump_link_nlmsg,
351 dump_link_nlmsg, cookie);
354 static int dump_link_nlmsg(void *cookie, void *msg, struct nlattr **tb)
356 struct bpf_netdev_t *netinfo = cookie;
357 struct ifinfomsg *ifinfo = msg;
359 if (netinfo->filter_idx > 0 && netinfo->filter_idx != ifinfo->ifi_index)
362 if (netinfo->used_len == netinfo->array_len) {
363 netinfo->devices = realloc(netinfo->devices,
364 (netinfo->array_len + 16) *
365 sizeof(struct ip_devname_ifindex));
366 if (!netinfo->devices)
369 netinfo->array_len += 16;
371 netinfo->devices[netinfo->used_len].ifindex = ifinfo->ifi_index;
372 snprintf(netinfo->devices[netinfo->used_len].devname,
373 sizeof(netinfo->devices[netinfo->used_len].devname),
376 ? libbpf_nla_getattr_str(tb[IFLA_IFNAME])
380 return do_xdp_dump(ifinfo, tb);
383 static int dump_class_qdisc_nlmsg(void *cookie, void *msg, struct nlattr **tb)
385 struct bpf_tcinfo_t *tcinfo = cookie;
386 struct tcmsg *info = msg;
388 if (tcinfo->is_qdisc) {
389 /* skip clsact qdisc */
391 strcmp(libbpf_nla_data(tb[TCA_KIND]), "clsact") == 0)
393 if (info->tcm_handle == 0)
397 if (tcinfo->used_len == tcinfo->array_len) {
398 tcinfo->handle_array = realloc(tcinfo->handle_array,
399 (tcinfo->array_len + 16) * sizeof(struct tc_kind_handle));
400 if (!tcinfo->handle_array)
403 tcinfo->array_len += 16;
405 tcinfo->handle_array[tcinfo->used_len].handle = info->tcm_handle;
406 snprintf(tcinfo->handle_array[tcinfo->used_len].kind,
407 sizeof(tcinfo->handle_array[tcinfo->used_len].kind),
410 ? libbpf_nla_getattr_str(tb[TCA_KIND])
417 static int dump_filter_nlmsg(void *cookie, void *msg, struct nlattr **tb)
419 const struct bpf_filter_t *filter_info = cookie;
421 return do_filter_dump((struct tcmsg *)msg, tb, filter_info->kind,
422 filter_info->devname, filter_info->ifindex);
425 static int show_dev_tc_bpf(int sock, unsigned int nl_pid,
426 struct ip_devname_ifindex *dev)
428 struct bpf_filter_t filter_info;
429 struct bpf_tcinfo_t tcinfo;
430 int i, handle, ret = 0;
432 tcinfo.handle_array = NULL;
434 tcinfo.array_len = 0;
436 tcinfo.is_qdisc = false;
437 ret = netlink_get_class(sock, nl_pid, dev->ifindex,
438 dump_class_qdisc_nlmsg, &tcinfo);
442 tcinfo.is_qdisc = true;
443 ret = netlink_get_qdisc(sock, nl_pid, dev->ifindex,
444 dump_class_qdisc_nlmsg, &tcinfo);
448 filter_info.devname = dev->devname;
449 filter_info.ifindex = dev->ifindex;
450 for (i = 0; i < tcinfo.used_len; i++) {
451 filter_info.kind = tcinfo.handle_array[i].kind;
452 ret = netlink_get_filter(sock, nl_pid, dev->ifindex,
453 tcinfo.handle_array[i].handle,
454 dump_filter_nlmsg, &filter_info);
459 /* root, ingress and egress handle */
461 filter_info.kind = "root";
462 ret = netlink_get_filter(sock, nl_pid, dev->ifindex, handle,
463 dump_filter_nlmsg, &filter_info);
467 handle = TC_H_MAKE(TC_H_CLSACT, TC_H_MIN_INGRESS);
468 filter_info.kind = "clsact/ingress";
469 ret = netlink_get_filter(sock, nl_pid, dev->ifindex, handle,
470 dump_filter_nlmsg, &filter_info);
474 handle = TC_H_MAKE(TC_H_CLSACT, TC_H_MIN_EGRESS);
475 filter_info.kind = "clsact/egress";
476 ret = netlink_get_filter(sock, nl_pid, dev->ifindex, handle,
477 dump_filter_nlmsg, &filter_info);
482 free(tcinfo.handle_array);
486 static int query_flow_dissector(struct bpf_attach_info *attach_info)
494 fd = open("/proc/self/ns/net", O_RDONLY);
496 p_err("can't open /proc/self/ns/net: %s",
500 prog_cnt = ARRAY_SIZE(prog_ids);
501 err = bpf_prog_query(fd, BPF_FLOW_DISSECTOR, 0,
502 &attach_flags, prog_ids, &prog_cnt);
505 if (errno == EINVAL) {
506 /* Older kernel's don't support querying
507 * flow dissector programs.
512 p_err("can't query prog: %s", strerror(errno));
517 attach_info->flow_dissector_id = prog_ids[0];
522 static int net_parse_dev(int *argc, char ***argv)
526 if (is_prefix(**argv, "dev")) {
529 ifindex = if_nametoindex(**argv);
531 p_err("invalid devname %s", **argv);
535 p_err("expected 'dev', got: '%s'?", **argv);
542 static int do_attach_detach_xdp(int progfd, enum net_attach_type attach_type,
543 int ifindex, bool overwrite)
548 flags = XDP_FLAGS_UPDATE_IF_NOEXIST;
549 if (attach_type == NET_ATTACH_TYPE_XDP_GENERIC)
550 flags |= XDP_FLAGS_SKB_MODE;
551 if (attach_type == NET_ATTACH_TYPE_XDP_DRIVER)
552 flags |= XDP_FLAGS_DRV_MODE;
553 if (attach_type == NET_ATTACH_TYPE_XDP_OFFLOAD)
554 flags |= XDP_FLAGS_HW_MODE;
556 return bpf_xdp_attach(ifindex, progfd, flags, NULL);
559 static int do_attach(int argc, char **argv)
561 enum net_attach_type attach_type;
562 int progfd, ifindex, err = 0;
563 bool overwrite = false;
565 /* parse attach args */
569 attach_type = parse_attach_type(*argv);
570 if (attach_type == net_attach_type_size) {
571 p_err("invalid net attach/detach type: %s", *argv);
576 progfd = prog_parse_fd(&argc, &argv);
580 ifindex = net_parse_dev(&argc, &argv);
587 if (is_prefix(*argv, "overwrite")) {
590 p_err("expected 'overwrite', got: '%s'?", *argv);
596 /* attach xdp prog */
597 if (is_prefix("xdp", attach_type_strings[attach_type]))
598 err = do_attach_detach_xdp(progfd, attach_type, ifindex,
601 p_err("interface %s attach failed: %s",
602 attach_type_strings[attach_type], strerror(-err));
607 jsonw_null(json_wtr);
613 static int do_detach(int argc, char **argv)
615 enum net_attach_type attach_type;
616 int progfd, ifindex, err = 0;
618 /* parse detach args */
622 attach_type = parse_attach_type(*argv);
623 if (attach_type == net_attach_type_size) {
624 p_err("invalid net attach/detach type: %s", *argv);
629 ifindex = net_parse_dev(&argc, &argv);
633 /* detach xdp prog */
635 if (is_prefix("xdp", attach_type_strings[attach_type]))
636 err = do_attach_detach_xdp(progfd, attach_type, ifindex, NULL);
639 p_err("interface %s detach failed: %s",
640 attach_type_strings[attach_type], strerror(-err));
645 jsonw_null(json_wtr);
650 static int netfilter_link_compar(const void *a, const void *b)
652 const struct bpf_link_info *nfa = a;
653 const struct bpf_link_info *nfb = b;
656 delta = nfa->netfilter.pf - nfb->netfilter.pf;
660 delta = nfa->netfilter.hooknum - nfb->netfilter.hooknum;
664 if (nfa->netfilter.priority < nfb->netfilter.priority)
666 if (nfa->netfilter.priority > nfb->netfilter.priority)
669 return nfa->netfilter.flags - nfb->netfilter.flags;
672 static void show_link_netfilter(void)
674 unsigned int nf_link_len = 0, nf_link_count = 0;
675 struct bpf_link_info *nf_link_info = NULL;
679 struct bpf_link_info info;
683 err = bpf_link_get_next_id(id, &id);
687 p_err("can't get next link: %s (id %d)", strerror(errno), id);
691 fd = bpf_link_get_fd_by_id(id);
693 p_err("can't get link by id (%u): %s", id, strerror(errno));
697 memset(&info, 0, sizeof(info));
700 err = bpf_link_get_info_by_fd(fd, &info, &len);
705 p_err("can't get link info for fd %d: %s", fd, strerror(errno));
709 if (info.type != BPF_LINK_TYPE_NETFILTER)
712 if (nf_link_count >= nf_link_len) {
713 static const unsigned int max_link_count = INT_MAX / sizeof(info);
714 struct bpf_link_info *expand;
716 if (nf_link_count > max_link_count) {
717 p_err("cannot handle more than %u links\n", max_link_count);
723 expand = realloc(nf_link_info, nf_link_len * sizeof(info));
725 p_err("realloc: %s", strerror(errno));
729 nf_link_info = expand;
732 nf_link_info[nf_link_count] = info;
736 qsort(nf_link_info, nf_link_count, sizeof(*nf_link_info), netfilter_link_compar);
738 for (id = 0; id < nf_link_count; id++) {
741 netfilter_dump_json(&nf_link_info[id], json_wtr);
743 netfilter_dump_plain(&nf_link_info[id]);
745 NET_DUMP_UINT("id", " prog_id %u", nf_link_info[id].prog_id);
752 static int do_show(int argc, char **argv)
754 struct bpf_attach_info attach_info = {};
755 int i, sock, ret, filter_idx = -1;
756 struct bpf_netdev_t dev_array;
757 unsigned int nl_pid = 0;
761 filter_idx = net_parse_dev(&argc, &argv);
764 } else if (argc != 0) {
768 ret = query_flow_dissector(&attach_info);
772 sock = netlink_open(&nl_pid);
774 fprintf(stderr, "failed to open netlink sock\n");
778 dev_array.devices = NULL;
779 dev_array.used_len = 0;
780 dev_array.array_len = 0;
781 dev_array.filter_idx = filter_idx;
784 jsonw_start_array(json_wtr);
786 NET_START_ARRAY("xdp", "%s:\n");
787 ret = netlink_get_link(sock, nl_pid, dump_link_nlmsg, &dev_array);
791 NET_START_ARRAY("tc", "%s:\n");
792 for (i = 0; i < dev_array.used_len; i++) {
793 ret = show_dev_tc_bpf(sock, nl_pid,
794 &dev_array.devices[i]);
801 NET_START_ARRAY("flow_dissector", "%s:\n");
802 if (attach_info.flow_dissector_id > 0)
803 NET_DUMP_UINT("id", "id %u", attach_info.flow_dissector_id);
806 NET_START_ARRAY("netfilter", "%s:\n");
807 show_link_netfilter();
812 jsonw_end_array(json_wtr);
816 jsonw_null(json_wtr);
817 libbpf_strerror(ret, err_buf, sizeof(err_buf));
818 fprintf(stderr, "Error: %s\n", err_buf);
820 free(dev_array.devices);
825 static int do_help(int argc, char **argv)
828 jsonw_null(json_wtr);
833 "Usage: %1$s %2$s { show | list } [dev <devname>]\n"
834 " %1$s %2$s attach ATTACH_TYPE PROG dev <devname> [ overwrite ]\n"
835 " %1$s %2$s detach ATTACH_TYPE dev <devname>\n"
838 " " HELP_SPEC_PROGRAM "\n"
839 " ATTACH_TYPE := { xdp | xdpgeneric | xdpdrv | xdpoffload }\n"
840 " " HELP_SPEC_OPTIONS " }\n"
842 "Note: Only xdp and tc attachments are supported now.\n"
843 " For progs attached to cgroups, use \"bpftool cgroup\"\n"
844 " to dump program attachments. For program types\n"
845 " sk_{filter,skb,msg,reuseport} and lwt/seg6, please\n"
846 " consult iproute2.\n"
853 static const struct cmd cmds[] = {
856 { "attach", do_attach },
857 { "detach", do_detach },
862 int do_net(int argc, char **argv)
864 return cmd_select(cmds, argc, argv, do_help);