1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 // Copyright (C) 2018 Facebook
12 #include <bpf/libbpf.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/socket.h>
16 #include <linux/tc_act/tc_bpf.h>
17 #include <sys/socket.h>
19 #include <sys/types.h>
21 #include "bpf/nlattr.h"
23 #include "netlink_dumper.h"
26 #define SOL_NETLINK 270
29 struct ip_devname_ifindex {
35 struct ip_devname_ifindex *devices;
41 struct tc_kind_handle {
47 struct tc_kind_handle *handle_array;
59 struct bpf_attach_info {
60 __u32 flow_dissector_id;
63 enum net_attach_type {
65 NET_ATTACH_TYPE_XDP_GENERIC,
66 NET_ATTACH_TYPE_XDP_DRIVER,
67 NET_ATTACH_TYPE_XDP_OFFLOAD,
70 static const char * const attach_type_strings[] = {
71 [NET_ATTACH_TYPE_XDP] = "xdp",
72 [NET_ATTACH_TYPE_XDP_GENERIC] = "xdpgeneric",
73 [NET_ATTACH_TYPE_XDP_DRIVER] = "xdpdrv",
74 [NET_ATTACH_TYPE_XDP_OFFLOAD] = "xdpoffload",
77 const size_t net_attach_type_size = ARRAY_SIZE(attach_type_strings);
79 static enum net_attach_type parse_attach_type(const char *str)
81 enum net_attach_type type;
83 for (type = 0; type < net_attach_type_size; type++) {
84 if (attach_type_strings[type] &&
85 is_prefix(str, attach_type_strings[type]))
89 return net_attach_type_size;
92 typedef int (*dump_nlmsg_t)(void *cookie, void *msg, struct nlattr **tb);
94 typedef int (*__dump_nlmsg_t)(struct nlmsghdr *nlmsg, dump_nlmsg_t, void *cookie);
96 static int netlink_open(__u32 *nl_pid)
98 struct sockaddr_nl sa;
103 memset(&sa, 0, sizeof(sa));
104 sa.nl_family = AF_NETLINK;
106 sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
110 if (setsockopt(sock, SOL_NETLINK, NETLINK_EXT_ACK,
111 &one, sizeof(one)) < 0) {
112 p_err("Netlink error reporting not supported");
115 if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
120 addrlen = sizeof(sa);
121 if (getsockname(sock, (struct sockaddr *)&sa, &addrlen) < 0) {
126 if (addrlen != sizeof(sa)) {
127 ret = -LIBBPF_ERRNO__INTERNAL;
139 static int netlink_recv(int sock, __u32 nl_pid, __u32 seq,
140 __dump_nlmsg_t _fn, dump_nlmsg_t fn,
143 bool multipart = true;
144 struct nlmsgerr *err;
151 len = recv(sock, buf, sizeof(buf), 0);
160 for (nh = (struct nlmsghdr *)buf; NLMSG_OK(nh, (unsigned int)len);
161 nh = NLMSG_NEXT(nh, len)) {
162 if (nh->nlmsg_pid != nl_pid) {
163 ret = -LIBBPF_ERRNO__WRNGPID;
166 if (nh->nlmsg_seq != seq) {
167 ret = -LIBBPF_ERRNO__INVSEQ;
170 if (nh->nlmsg_flags & NLM_F_MULTI)
172 switch (nh->nlmsg_type) {
174 err = (struct nlmsgerr *)NLMSG_DATA(nh);
178 libbpf_nla_dump_errormsg(nh);
186 ret = _fn(nh, fn, cookie);
197 static int __dump_class_nlmsg(struct nlmsghdr *nlh,
198 dump_nlmsg_t dump_class_nlmsg,
201 struct nlattr *tb[TCA_MAX + 1], *attr;
202 struct tcmsg *t = NLMSG_DATA(nlh);
205 len = nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*t));
206 attr = (struct nlattr *) ((void *) t + NLMSG_ALIGN(sizeof(*t)));
207 if (libbpf_nla_parse(tb, TCA_MAX, attr, len, NULL) != 0)
208 return -LIBBPF_ERRNO__NLPARSE;
210 return dump_class_nlmsg(cookie, t, tb);
213 static int netlink_get_class(int sock, unsigned int nl_pid, int ifindex,
214 dump_nlmsg_t dump_class_nlmsg, void *cookie)
220 .nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct tcmsg)),
221 .nlh.nlmsg_type = RTM_GETTCLASS,
222 .nlh.nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST,
223 .t.tcm_family = AF_UNSPEC,
224 .t.tcm_ifindex = ifindex,
226 int seq = time(NULL);
228 req.nlh.nlmsg_seq = seq;
229 if (send(sock, &req, req.nlh.nlmsg_len, 0) < 0)
232 return netlink_recv(sock, nl_pid, seq, __dump_class_nlmsg,
233 dump_class_nlmsg, cookie);
236 static int __dump_qdisc_nlmsg(struct nlmsghdr *nlh,
237 dump_nlmsg_t dump_qdisc_nlmsg,
240 struct nlattr *tb[TCA_MAX + 1], *attr;
241 struct tcmsg *t = NLMSG_DATA(nlh);
244 len = nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*t));
245 attr = (struct nlattr *) ((void *) t + NLMSG_ALIGN(sizeof(*t)));
246 if (libbpf_nla_parse(tb, TCA_MAX, attr, len, NULL) != 0)
247 return -LIBBPF_ERRNO__NLPARSE;
249 return dump_qdisc_nlmsg(cookie, t, tb);
252 static int netlink_get_qdisc(int sock, unsigned int nl_pid, int ifindex,
253 dump_nlmsg_t dump_qdisc_nlmsg, void *cookie)
259 .nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct tcmsg)),
260 .nlh.nlmsg_type = RTM_GETQDISC,
261 .nlh.nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST,
262 .t.tcm_family = AF_UNSPEC,
263 .t.tcm_ifindex = ifindex,
265 int seq = time(NULL);
267 req.nlh.nlmsg_seq = seq;
268 if (send(sock, &req, req.nlh.nlmsg_len, 0) < 0)
271 return netlink_recv(sock, nl_pid, seq, __dump_qdisc_nlmsg,
272 dump_qdisc_nlmsg, cookie);
275 static int __dump_filter_nlmsg(struct nlmsghdr *nlh,
276 dump_nlmsg_t dump_filter_nlmsg,
279 struct nlattr *tb[TCA_MAX + 1], *attr;
280 struct tcmsg *t = NLMSG_DATA(nlh);
283 len = nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*t));
284 attr = (struct nlattr *) ((void *) t + NLMSG_ALIGN(sizeof(*t)));
285 if (libbpf_nla_parse(tb, TCA_MAX, attr, len, NULL) != 0)
286 return -LIBBPF_ERRNO__NLPARSE;
288 return dump_filter_nlmsg(cookie, t, tb);
291 static int netlink_get_filter(int sock, unsigned int nl_pid, int ifindex, int handle,
292 dump_nlmsg_t dump_filter_nlmsg, void *cookie)
298 .nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct tcmsg)),
299 .nlh.nlmsg_type = RTM_GETTFILTER,
300 .nlh.nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST,
301 .t.tcm_family = AF_UNSPEC,
302 .t.tcm_ifindex = ifindex,
303 .t.tcm_parent = handle,
305 int seq = time(NULL);
307 req.nlh.nlmsg_seq = seq;
308 if (send(sock, &req, req.nlh.nlmsg_len, 0) < 0)
311 return netlink_recv(sock, nl_pid, seq, __dump_filter_nlmsg,
312 dump_filter_nlmsg, cookie);
315 static int __dump_link_nlmsg(struct nlmsghdr *nlh,
316 dump_nlmsg_t dump_link_nlmsg, void *cookie)
318 struct nlattr *tb[IFLA_MAX + 1], *attr;
319 struct ifinfomsg *ifi = NLMSG_DATA(nlh);
322 len = nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*ifi));
323 attr = (struct nlattr *) ((void *) ifi + NLMSG_ALIGN(sizeof(*ifi)));
324 if (libbpf_nla_parse(tb, IFLA_MAX, attr, len, NULL) != 0)
325 return -LIBBPF_ERRNO__NLPARSE;
327 return dump_link_nlmsg(cookie, ifi, tb);
330 static int netlink_get_link(int sock, unsigned int nl_pid,
331 dump_nlmsg_t dump_link_nlmsg, void *cookie)
335 struct ifinfomsg ifm;
337 .nlh.nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg)),
338 .nlh.nlmsg_type = RTM_GETLINK,
339 .nlh.nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST,
340 .ifm.ifi_family = AF_PACKET,
342 int seq = time(NULL);
344 req.nlh.nlmsg_seq = seq;
345 if (send(sock, &req, req.nlh.nlmsg_len, 0) < 0)
348 return netlink_recv(sock, nl_pid, seq, __dump_link_nlmsg,
349 dump_link_nlmsg, cookie);
352 static int dump_link_nlmsg(void *cookie, void *msg, struct nlattr **tb)
354 struct bpf_netdev_t *netinfo = cookie;
355 struct ifinfomsg *ifinfo = msg;
357 if (netinfo->filter_idx > 0 && netinfo->filter_idx != ifinfo->ifi_index)
360 if (netinfo->used_len == netinfo->array_len) {
361 netinfo->devices = realloc(netinfo->devices,
362 (netinfo->array_len + 16) *
363 sizeof(struct ip_devname_ifindex));
364 if (!netinfo->devices)
367 netinfo->array_len += 16;
369 netinfo->devices[netinfo->used_len].ifindex = ifinfo->ifi_index;
370 snprintf(netinfo->devices[netinfo->used_len].devname,
371 sizeof(netinfo->devices[netinfo->used_len].devname),
374 ? libbpf_nla_getattr_str(tb[IFLA_IFNAME])
378 return do_xdp_dump(ifinfo, tb);
381 static int dump_class_qdisc_nlmsg(void *cookie, void *msg, struct nlattr **tb)
383 struct bpf_tcinfo_t *tcinfo = cookie;
384 struct tcmsg *info = msg;
386 if (tcinfo->is_qdisc) {
387 /* skip clsact qdisc */
389 strcmp(libbpf_nla_data(tb[TCA_KIND]), "clsact") == 0)
391 if (info->tcm_handle == 0)
395 if (tcinfo->used_len == tcinfo->array_len) {
396 tcinfo->handle_array = realloc(tcinfo->handle_array,
397 (tcinfo->array_len + 16) * sizeof(struct tc_kind_handle));
398 if (!tcinfo->handle_array)
401 tcinfo->array_len += 16;
403 tcinfo->handle_array[tcinfo->used_len].handle = info->tcm_handle;
404 snprintf(tcinfo->handle_array[tcinfo->used_len].kind,
405 sizeof(tcinfo->handle_array[tcinfo->used_len].kind),
408 ? libbpf_nla_getattr_str(tb[TCA_KIND])
415 static int dump_filter_nlmsg(void *cookie, void *msg, struct nlattr **tb)
417 const struct bpf_filter_t *filter_info = cookie;
419 return do_filter_dump((struct tcmsg *)msg, tb, filter_info->kind,
420 filter_info->devname, filter_info->ifindex);
423 static int show_dev_tc_bpf(int sock, unsigned int nl_pid,
424 struct ip_devname_ifindex *dev)
426 struct bpf_filter_t filter_info;
427 struct bpf_tcinfo_t tcinfo;
428 int i, handle, ret = 0;
430 tcinfo.handle_array = NULL;
432 tcinfo.array_len = 0;
434 tcinfo.is_qdisc = false;
435 ret = netlink_get_class(sock, nl_pid, dev->ifindex,
436 dump_class_qdisc_nlmsg, &tcinfo);
440 tcinfo.is_qdisc = true;
441 ret = netlink_get_qdisc(sock, nl_pid, dev->ifindex,
442 dump_class_qdisc_nlmsg, &tcinfo);
446 filter_info.devname = dev->devname;
447 filter_info.ifindex = dev->ifindex;
448 for (i = 0; i < tcinfo.used_len; i++) {
449 filter_info.kind = tcinfo.handle_array[i].kind;
450 ret = netlink_get_filter(sock, nl_pid, dev->ifindex,
451 tcinfo.handle_array[i].handle,
452 dump_filter_nlmsg, &filter_info);
457 /* root, ingress and egress handle */
459 filter_info.kind = "root";
460 ret = netlink_get_filter(sock, nl_pid, dev->ifindex, handle,
461 dump_filter_nlmsg, &filter_info);
465 handle = TC_H_MAKE(TC_H_CLSACT, TC_H_MIN_INGRESS);
466 filter_info.kind = "clsact/ingress";
467 ret = netlink_get_filter(sock, nl_pid, dev->ifindex, handle,
468 dump_filter_nlmsg, &filter_info);
472 handle = TC_H_MAKE(TC_H_CLSACT, TC_H_MIN_EGRESS);
473 filter_info.kind = "clsact/egress";
474 ret = netlink_get_filter(sock, nl_pid, dev->ifindex, handle,
475 dump_filter_nlmsg, &filter_info);
480 free(tcinfo.handle_array);
484 static int query_flow_dissector(struct bpf_attach_info *attach_info)
492 fd = open("/proc/self/ns/net", O_RDONLY);
494 p_err("can't open /proc/self/ns/net: %s",
498 prog_cnt = ARRAY_SIZE(prog_ids);
499 err = bpf_prog_query(fd, BPF_FLOW_DISSECTOR, 0,
500 &attach_flags, prog_ids, &prog_cnt);
503 if (errno == EINVAL) {
504 /* Older kernel's don't support querying
505 * flow dissector programs.
510 p_err("can't query prog: %s", strerror(errno));
515 attach_info->flow_dissector_id = prog_ids[0];
520 static int net_parse_dev(int *argc, char ***argv)
524 if (is_prefix(**argv, "dev")) {
527 ifindex = if_nametoindex(**argv);
529 p_err("invalid devname %s", **argv);
533 p_err("expected 'dev', got: '%s'?", **argv);
540 static int do_attach_detach_xdp(int progfd, enum net_attach_type attach_type,
541 int ifindex, bool overwrite)
546 flags = XDP_FLAGS_UPDATE_IF_NOEXIST;
547 if (attach_type == NET_ATTACH_TYPE_XDP_GENERIC)
548 flags |= XDP_FLAGS_SKB_MODE;
549 if (attach_type == NET_ATTACH_TYPE_XDP_DRIVER)
550 flags |= XDP_FLAGS_DRV_MODE;
551 if (attach_type == NET_ATTACH_TYPE_XDP_OFFLOAD)
552 flags |= XDP_FLAGS_HW_MODE;
554 return bpf_set_link_xdp_fd(ifindex, progfd, flags);
557 static int do_attach(int argc, char **argv)
559 enum net_attach_type attach_type;
560 int progfd, ifindex, err = 0;
561 bool overwrite = false;
563 /* parse attach args */
567 attach_type = parse_attach_type(*argv);
568 if (attach_type == net_attach_type_size) {
569 p_err("invalid net attach/detach type: %s", *argv);
574 progfd = prog_parse_fd(&argc, &argv);
578 ifindex = net_parse_dev(&argc, &argv);
585 if (is_prefix(*argv, "overwrite")) {
588 p_err("expected 'overwrite', got: '%s'?", *argv);
594 /* attach xdp prog */
595 if (is_prefix("xdp", attach_type_strings[attach_type]))
596 err = do_attach_detach_xdp(progfd, attach_type, ifindex,
599 p_err("interface %s attach failed: %s",
600 attach_type_strings[attach_type], strerror(-err));
605 jsonw_null(json_wtr);
611 static int do_detach(int argc, char **argv)
613 enum net_attach_type attach_type;
614 int progfd, ifindex, err = 0;
616 /* parse detach args */
620 attach_type = parse_attach_type(*argv);
621 if (attach_type == net_attach_type_size) {
622 p_err("invalid net attach/detach type: %s", *argv);
627 ifindex = net_parse_dev(&argc, &argv);
631 /* detach xdp prog */
633 if (is_prefix("xdp", attach_type_strings[attach_type]))
634 err = do_attach_detach_xdp(progfd, attach_type, ifindex, NULL);
637 p_err("interface %s detach failed: %s",
638 attach_type_strings[attach_type], strerror(-err));
643 jsonw_null(json_wtr);
648 static int do_show(int argc, char **argv)
650 struct bpf_attach_info attach_info = {};
651 int i, sock, ret, filter_idx = -1;
652 struct bpf_netdev_t dev_array;
653 unsigned int nl_pid = 0;
657 filter_idx = net_parse_dev(&argc, &argv);
660 } else if (argc != 0) {
664 ret = query_flow_dissector(&attach_info);
668 sock = netlink_open(&nl_pid);
670 fprintf(stderr, "failed to open netlink sock\n");
674 dev_array.devices = NULL;
675 dev_array.used_len = 0;
676 dev_array.array_len = 0;
677 dev_array.filter_idx = filter_idx;
680 jsonw_start_array(json_wtr);
682 NET_START_ARRAY("xdp", "%s:\n");
683 ret = netlink_get_link(sock, nl_pid, dump_link_nlmsg, &dev_array);
687 NET_START_ARRAY("tc", "%s:\n");
688 for (i = 0; i < dev_array.used_len; i++) {
689 ret = show_dev_tc_bpf(sock, nl_pid,
690 &dev_array.devices[i]);
697 NET_START_ARRAY("flow_dissector", "%s:\n");
698 if (attach_info.flow_dissector_id > 0)
699 NET_DUMP_UINT("id", "id %u", attach_info.flow_dissector_id);
704 jsonw_end_array(json_wtr);
708 jsonw_null(json_wtr);
709 libbpf_strerror(ret, err_buf, sizeof(err_buf));
710 fprintf(stderr, "Error: %s\n", err_buf);
712 free(dev_array.devices);
717 static int do_help(int argc, char **argv)
720 jsonw_null(json_wtr);
725 "Usage: %1$s %2$s { show | list } [dev <devname>]\n"
726 " %1$s %2$s attach ATTACH_TYPE PROG dev <devname> [ overwrite ]\n"
727 " %1$s %2$s detach ATTACH_TYPE dev <devname>\n"
730 " " HELP_SPEC_PROGRAM "\n"
731 " ATTACH_TYPE := { xdp | xdpgeneric | xdpdrv | xdpoffload }\n"
732 " " HELP_SPEC_OPTIONS " }\n"
734 "Note: Only xdp and tc attachments are supported now.\n"
735 " For progs attached to cgroups, use \"bpftool cgroup\"\n"
736 " to dump program attachments. For program types\n"
737 " sk_{filter,skb,msg,reuseport} and lwt/seg6, please\n"
738 " consult iproute2.\n"
745 static const struct cmd cmds[] = {
748 { "attach", do_attach },
749 { "detach", do_detach },
754 int do_net(int argc, char **argv)
756 return cmd_select(cmds, argc, argv, do_help);