xref: /openbmc/linux/tools/lib/bpf/netlink.c (revision dd3cb467)
1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 /* Copyright (c) 2018 Facebook */
3 
4 #include <stdlib.h>
5 #include <memory.h>
6 #include <unistd.h>
7 #include <arpa/inet.h>
8 #include <linux/bpf.h>
9 #include <linux/if_ether.h>
10 #include <linux/pkt_cls.h>
11 #include <linux/rtnetlink.h>
12 #include <sys/socket.h>
13 #include <errno.h>
14 #include <time.h>
15 
16 #include "bpf.h"
17 #include "libbpf.h"
18 #include "libbpf_internal.h"
19 #include "nlattr.h"
20 
21 #ifndef SOL_NETLINK
22 #define SOL_NETLINK 270
23 #endif
24 
25 typedef int (*libbpf_dump_nlmsg_t)(void *cookie, void *msg, struct nlattr **tb);
26 
27 typedef int (*__dump_nlmsg_t)(struct nlmsghdr *nlmsg, libbpf_dump_nlmsg_t,
28 			      void *cookie);
29 
30 struct xdp_link_info {
31 	__u32 prog_id;
32 	__u32 drv_prog_id;
33 	__u32 hw_prog_id;
34 	__u32 skb_prog_id;
35 	__u8 attach_mode;
36 };
37 
38 struct xdp_id_md {
39 	int ifindex;
40 	__u32 flags;
41 	struct xdp_link_info info;
42 };
43 
44 static int libbpf_netlink_open(__u32 *nl_pid)
45 {
46 	struct sockaddr_nl sa;
47 	socklen_t addrlen;
48 	int one = 1, ret;
49 	int sock;
50 
51 	memset(&sa, 0, sizeof(sa));
52 	sa.nl_family = AF_NETLINK;
53 
54 	sock = socket(AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC, NETLINK_ROUTE);
55 	if (sock < 0)
56 		return -errno;
57 
58 	if (setsockopt(sock, SOL_NETLINK, NETLINK_EXT_ACK,
59 		       &one, sizeof(one)) < 0) {
60 		pr_warn("Netlink error reporting not supported\n");
61 	}
62 
63 	if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
64 		ret = -errno;
65 		goto cleanup;
66 	}
67 
68 	addrlen = sizeof(sa);
69 	if (getsockname(sock, (struct sockaddr *)&sa, &addrlen) < 0) {
70 		ret = -errno;
71 		goto cleanup;
72 	}
73 
74 	if (addrlen != sizeof(sa)) {
75 		ret = -LIBBPF_ERRNO__INTERNAL;
76 		goto cleanup;
77 	}
78 
79 	*nl_pid = sa.nl_pid;
80 	return sock;
81 
82 cleanup:
83 	close(sock);
84 	return ret;
85 }
86 
87 static void libbpf_netlink_close(int sock)
88 {
89 	close(sock);
90 }
91 
92 enum {
93 	NL_CONT,
94 	NL_NEXT,
95 	NL_DONE,
96 };
97 
98 static int netlink_recvmsg(int sock, struct msghdr *mhdr, int flags)
99 {
100 	int len;
101 
102 	do {
103 		len = recvmsg(sock, mhdr, flags);
104 	} while (len < 0 && (errno == EINTR || errno == EAGAIN));
105 
106 	if (len < 0)
107 		return -errno;
108 	return len;
109 }
110 
111 static int alloc_iov(struct iovec *iov, int len)
112 {
113 	void *nbuf;
114 
115 	nbuf = realloc(iov->iov_base, len);
116 	if (!nbuf)
117 		return -ENOMEM;
118 
119 	iov->iov_base = nbuf;
120 	iov->iov_len = len;
121 	return 0;
122 }
123 
124 static int libbpf_netlink_recv(int sock, __u32 nl_pid, int seq,
125 			       __dump_nlmsg_t _fn, libbpf_dump_nlmsg_t fn,
126 			       void *cookie)
127 {
128 	struct iovec iov = {};
129 	struct msghdr mhdr = {
130 		.msg_iov = &iov,
131 		.msg_iovlen = 1,
132 	};
133 	bool multipart = true;
134 	struct nlmsgerr *err;
135 	struct nlmsghdr *nh;
136 	int len, ret;
137 
138 	ret = alloc_iov(&iov, 4096);
139 	if (ret)
140 		goto done;
141 
142 	while (multipart) {
143 start:
144 		multipart = false;
145 		len = netlink_recvmsg(sock, &mhdr, MSG_PEEK | MSG_TRUNC);
146 		if (len < 0) {
147 			ret = len;
148 			goto done;
149 		}
150 
151 		if (len > iov.iov_len) {
152 			ret = alloc_iov(&iov, len);
153 			if (ret)
154 				goto done;
155 		}
156 
157 		len = netlink_recvmsg(sock, &mhdr, 0);
158 		if (len < 0) {
159 			ret = len;
160 			goto done;
161 		}
162 
163 		if (len == 0)
164 			break;
165 
166 		for (nh = (struct nlmsghdr *)iov.iov_base; NLMSG_OK(nh, len);
167 		     nh = NLMSG_NEXT(nh, len)) {
168 			if (nh->nlmsg_pid != nl_pid) {
169 				ret = -LIBBPF_ERRNO__WRNGPID;
170 				goto done;
171 			}
172 			if (nh->nlmsg_seq != seq) {
173 				ret = -LIBBPF_ERRNO__INVSEQ;
174 				goto done;
175 			}
176 			if (nh->nlmsg_flags & NLM_F_MULTI)
177 				multipart = true;
178 			switch (nh->nlmsg_type) {
179 			case NLMSG_ERROR:
180 				err = (struct nlmsgerr *)NLMSG_DATA(nh);
181 				if (!err->error)
182 					continue;
183 				ret = err->error;
184 				libbpf_nla_dump_errormsg(nh);
185 				goto done;
186 			case NLMSG_DONE:
187 				ret = 0;
188 				goto done;
189 			default:
190 				break;
191 			}
192 			if (_fn) {
193 				ret = _fn(nh, fn, cookie);
194 				switch (ret) {
195 				case NL_CONT:
196 					break;
197 				case NL_NEXT:
198 					goto start;
199 				case NL_DONE:
200 					ret = 0;
201 					goto done;
202 				default:
203 					goto done;
204 				}
205 			}
206 		}
207 	}
208 	ret = 0;
209 done:
210 	free(iov.iov_base);
211 	return ret;
212 }
213 
214 static int libbpf_netlink_send_recv(struct libbpf_nla_req *req,
215 				    __dump_nlmsg_t parse_msg,
216 				    libbpf_dump_nlmsg_t parse_attr,
217 				    void *cookie)
218 {
219 	__u32 nl_pid = 0;
220 	int sock, ret;
221 
222 	sock = libbpf_netlink_open(&nl_pid);
223 	if (sock < 0)
224 		return sock;
225 
226 	req->nh.nlmsg_pid = 0;
227 	req->nh.nlmsg_seq = time(NULL);
228 
229 	if (send(sock, req, req->nh.nlmsg_len, 0) < 0) {
230 		ret = -errno;
231 		goto out;
232 	}
233 
234 	ret = libbpf_netlink_recv(sock, nl_pid, req->nh.nlmsg_seq,
235 				  parse_msg, parse_attr, cookie);
236 out:
237 	libbpf_netlink_close(sock);
238 	return ret;
239 }
240 
241 static int __bpf_set_link_xdp_fd_replace(int ifindex, int fd, int old_fd,
242 					 __u32 flags)
243 {
244 	struct nlattr *nla;
245 	int ret;
246 	struct libbpf_nla_req req;
247 
248 	memset(&req, 0, sizeof(req));
249 	req.nh.nlmsg_len      = NLMSG_LENGTH(sizeof(struct ifinfomsg));
250 	req.nh.nlmsg_flags    = NLM_F_REQUEST | NLM_F_ACK;
251 	req.nh.nlmsg_type     = RTM_SETLINK;
252 	req.ifinfo.ifi_family = AF_UNSPEC;
253 	req.ifinfo.ifi_index  = ifindex;
254 
255 	nla = nlattr_begin_nested(&req, IFLA_XDP);
256 	if (!nla)
257 		return -EMSGSIZE;
258 	ret = nlattr_add(&req, IFLA_XDP_FD, &fd, sizeof(fd));
259 	if (ret < 0)
260 		return ret;
261 	if (flags) {
262 		ret = nlattr_add(&req, IFLA_XDP_FLAGS, &flags, sizeof(flags));
263 		if (ret < 0)
264 			return ret;
265 	}
266 	if (flags & XDP_FLAGS_REPLACE) {
267 		ret = nlattr_add(&req, IFLA_XDP_EXPECTED_FD, &old_fd,
268 				 sizeof(old_fd));
269 		if (ret < 0)
270 			return ret;
271 	}
272 	nlattr_end_nested(&req, nla);
273 
274 	return libbpf_netlink_send_recv(&req, NULL, NULL, NULL);
275 }
276 
277 int bpf_xdp_attach(int ifindex, int prog_fd, __u32 flags, const struct bpf_xdp_attach_opts *opts)
278 {
279 	int old_prog_fd, err;
280 
281 	if (!OPTS_VALID(opts, bpf_xdp_attach_opts))
282 		return libbpf_err(-EINVAL);
283 
284 	old_prog_fd = OPTS_GET(opts, old_prog_fd, 0);
285 	if (old_prog_fd)
286 		flags |= XDP_FLAGS_REPLACE;
287 	else
288 		old_prog_fd = -1;
289 
290 	err = __bpf_set_link_xdp_fd_replace(ifindex, prog_fd, old_prog_fd, flags);
291 	return libbpf_err(err);
292 }
293 
294 int bpf_xdp_detach(int ifindex, __u32 flags, const struct bpf_xdp_attach_opts *opts)
295 {
296 	return bpf_xdp_attach(ifindex, -1, flags, opts);
297 }
298 
299 static int __dump_link_nlmsg(struct nlmsghdr *nlh,
300 			     libbpf_dump_nlmsg_t dump_link_nlmsg, void *cookie)
301 {
302 	struct nlattr *tb[IFLA_MAX + 1], *attr;
303 	struct ifinfomsg *ifi = NLMSG_DATA(nlh);
304 	int len;
305 
306 	len = nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*ifi));
307 	attr = (struct nlattr *) ((void *) ifi + NLMSG_ALIGN(sizeof(*ifi)));
308 
309 	if (libbpf_nla_parse(tb, IFLA_MAX, attr, len, NULL) != 0)
310 		return -LIBBPF_ERRNO__NLPARSE;
311 
312 	return dump_link_nlmsg(cookie, ifi, tb);
313 }
314 
315 static int get_xdp_info(void *cookie, void *msg, struct nlattr **tb)
316 {
317 	struct nlattr *xdp_tb[IFLA_XDP_MAX + 1];
318 	struct xdp_id_md *xdp_id = cookie;
319 	struct ifinfomsg *ifinfo = msg;
320 	int ret;
321 
322 	if (xdp_id->ifindex && xdp_id->ifindex != ifinfo->ifi_index)
323 		return 0;
324 
325 	if (!tb[IFLA_XDP])
326 		return 0;
327 
328 	ret = libbpf_nla_parse_nested(xdp_tb, IFLA_XDP_MAX, tb[IFLA_XDP], NULL);
329 	if (ret)
330 		return ret;
331 
332 	if (!xdp_tb[IFLA_XDP_ATTACHED])
333 		return 0;
334 
335 	xdp_id->info.attach_mode = libbpf_nla_getattr_u8(
336 		xdp_tb[IFLA_XDP_ATTACHED]);
337 
338 	if (xdp_id->info.attach_mode == XDP_ATTACHED_NONE)
339 		return 0;
340 
341 	if (xdp_tb[IFLA_XDP_PROG_ID])
342 		xdp_id->info.prog_id = libbpf_nla_getattr_u32(
343 			xdp_tb[IFLA_XDP_PROG_ID]);
344 
345 	if (xdp_tb[IFLA_XDP_SKB_PROG_ID])
346 		xdp_id->info.skb_prog_id = libbpf_nla_getattr_u32(
347 			xdp_tb[IFLA_XDP_SKB_PROG_ID]);
348 
349 	if (xdp_tb[IFLA_XDP_DRV_PROG_ID])
350 		xdp_id->info.drv_prog_id = libbpf_nla_getattr_u32(
351 			xdp_tb[IFLA_XDP_DRV_PROG_ID]);
352 
353 	if (xdp_tb[IFLA_XDP_HW_PROG_ID])
354 		xdp_id->info.hw_prog_id = libbpf_nla_getattr_u32(
355 			xdp_tb[IFLA_XDP_HW_PROG_ID]);
356 
357 	return 0;
358 }
359 
360 int bpf_xdp_query(int ifindex, int xdp_flags, struct bpf_xdp_query_opts *opts)
361 {
362 	struct libbpf_nla_req req = {
363 		.nh.nlmsg_len      = NLMSG_LENGTH(sizeof(struct ifinfomsg)),
364 		.nh.nlmsg_type     = RTM_GETLINK,
365 		.nh.nlmsg_flags    = NLM_F_DUMP | NLM_F_REQUEST,
366 		.ifinfo.ifi_family = AF_PACKET,
367 	};
368 	struct xdp_id_md xdp_id = {};
369 	int err;
370 
371 	if (!OPTS_VALID(opts, bpf_xdp_query_opts))
372 		return libbpf_err(-EINVAL);
373 
374 	if (xdp_flags & ~XDP_FLAGS_MASK)
375 		return libbpf_err(-EINVAL);
376 
377 	/* Check whether the single {HW,DRV,SKB} mode is set */
378 	xdp_flags &= XDP_FLAGS_SKB_MODE | XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE;
379 	if (xdp_flags & (xdp_flags - 1))
380 		return libbpf_err(-EINVAL);
381 
382 	xdp_id.ifindex = ifindex;
383 	xdp_id.flags = xdp_flags;
384 
385 	err = libbpf_netlink_send_recv(&req, __dump_link_nlmsg,
386 				       get_xdp_info, &xdp_id);
387 	if (err)
388 		return libbpf_err(err);
389 
390 	OPTS_SET(opts, prog_id, xdp_id.info.prog_id);
391 	OPTS_SET(opts, drv_prog_id, xdp_id.info.drv_prog_id);
392 	OPTS_SET(opts, hw_prog_id, xdp_id.info.hw_prog_id);
393 	OPTS_SET(opts, skb_prog_id, xdp_id.info.skb_prog_id);
394 	OPTS_SET(opts, attach_mode, xdp_id.info.attach_mode);
395 
396 	return 0;
397 }
398 
399 int bpf_xdp_query_id(int ifindex, int flags, __u32 *prog_id)
400 {
401 	LIBBPF_OPTS(bpf_xdp_query_opts, opts);
402 	int ret;
403 
404 	ret = bpf_xdp_query(ifindex, flags, &opts);
405 	if (ret)
406 		return libbpf_err(ret);
407 
408 	flags &= XDP_FLAGS_MODES;
409 
410 	if (opts.attach_mode != XDP_ATTACHED_MULTI && !flags)
411 		*prog_id = opts.prog_id;
412 	else if (flags & XDP_FLAGS_DRV_MODE)
413 		*prog_id = opts.drv_prog_id;
414 	else if (flags & XDP_FLAGS_HW_MODE)
415 		*prog_id = opts.hw_prog_id;
416 	else if (flags & XDP_FLAGS_SKB_MODE)
417 		*prog_id = opts.skb_prog_id;
418 	else
419 		*prog_id = 0;
420 
421 	return 0;
422 }
423 
424 
425 typedef int (*qdisc_config_t)(struct libbpf_nla_req *req);
426 
427 static int clsact_config(struct libbpf_nla_req *req)
428 {
429 	req->tc.tcm_parent = TC_H_CLSACT;
430 	req->tc.tcm_handle = TC_H_MAKE(TC_H_CLSACT, 0);
431 
432 	return nlattr_add(req, TCA_KIND, "clsact", sizeof("clsact"));
433 }
434 
435 static int attach_point_to_config(struct bpf_tc_hook *hook,
436 				  qdisc_config_t *config)
437 {
438 	switch (OPTS_GET(hook, attach_point, 0)) {
439 	case BPF_TC_INGRESS:
440 	case BPF_TC_EGRESS:
441 	case BPF_TC_INGRESS | BPF_TC_EGRESS:
442 		if (OPTS_GET(hook, parent, 0))
443 			return -EINVAL;
444 		*config = &clsact_config;
445 		return 0;
446 	case BPF_TC_CUSTOM:
447 		return -EOPNOTSUPP;
448 	default:
449 		return -EINVAL;
450 	}
451 }
452 
453 static int tc_get_tcm_parent(enum bpf_tc_attach_point attach_point,
454 			     __u32 *parent)
455 {
456 	switch (attach_point) {
457 	case BPF_TC_INGRESS:
458 	case BPF_TC_EGRESS:
459 		if (*parent)
460 			return -EINVAL;
461 		*parent = TC_H_MAKE(TC_H_CLSACT,
462 				    attach_point == BPF_TC_INGRESS ?
463 				    TC_H_MIN_INGRESS : TC_H_MIN_EGRESS);
464 		break;
465 	case BPF_TC_CUSTOM:
466 		if (!*parent)
467 			return -EINVAL;
468 		break;
469 	default:
470 		return -EINVAL;
471 	}
472 	return 0;
473 }
474 
475 static int tc_qdisc_modify(struct bpf_tc_hook *hook, int cmd, int flags)
476 {
477 	qdisc_config_t config;
478 	int ret;
479 	struct libbpf_nla_req req;
480 
481 	ret = attach_point_to_config(hook, &config);
482 	if (ret < 0)
483 		return ret;
484 
485 	memset(&req, 0, sizeof(req));
486 	req.nh.nlmsg_len   = NLMSG_LENGTH(sizeof(struct tcmsg));
487 	req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | flags;
488 	req.nh.nlmsg_type  = cmd;
489 	req.tc.tcm_family  = AF_UNSPEC;
490 	req.tc.tcm_ifindex = OPTS_GET(hook, ifindex, 0);
491 
492 	ret = config(&req);
493 	if (ret < 0)
494 		return ret;
495 
496 	return libbpf_netlink_send_recv(&req, NULL, NULL, NULL);
497 }
498 
499 static int tc_qdisc_create_excl(struct bpf_tc_hook *hook)
500 {
501 	return tc_qdisc_modify(hook, RTM_NEWQDISC, NLM_F_CREATE | NLM_F_EXCL);
502 }
503 
504 static int tc_qdisc_delete(struct bpf_tc_hook *hook)
505 {
506 	return tc_qdisc_modify(hook, RTM_DELQDISC, 0);
507 }
508 
509 int bpf_tc_hook_create(struct bpf_tc_hook *hook)
510 {
511 	int ret;
512 
513 	if (!hook || !OPTS_VALID(hook, bpf_tc_hook) ||
514 	    OPTS_GET(hook, ifindex, 0) <= 0)
515 		return libbpf_err(-EINVAL);
516 
517 	ret = tc_qdisc_create_excl(hook);
518 	return libbpf_err(ret);
519 }
520 
521 static int __bpf_tc_detach(const struct bpf_tc_hook *hook,
522 			   const struct bpf_tc_opts *opts,
523 			   const bool flush);
524 
525 int bpf_tc_hook_destroy(struct bpf_tc_hook *hook)
526 {
527 	if (!hook || !OPTS_VALID(hook, bpf_tc_hook) ||
528 	    OPTS_GET(hook, ifindex, 0) <= 0)
529 		return libbpf_err(-EINVAL);
530 
531 	switch (OPTS_GET(hook, attach_point, 0)) {
532 	case BPF_TC_INGRESS:
533 	case BPF_TC_EGRESS:
534 		return libbpf_err(__bpf_tc_detach(hook, NULL, true));
535 	case BPF_TC_INGRESS | BPF_TC_EGRESS:
536 		return libbpf_err(tc_qdisc_delete(hook));
537 	case BPF_TC_CUSTOM:
538 		return libbpf_err(-EOPNOTSUPP);
539 	default:
540 		return libbpf_err(-EINVAL);
541 	}
542 }
543 
544 struct bpf_cb_ctx {
545 	struct bpf_tc_opts *opts;
546 	bool processed;
547 };
548 
549 static int __get_tc_info(void *cookie, struct tcmsg *tc, struct nlattr **tb,
550 			 bool unicast)
551 {
552 	struct nlattr *tbb[TCA_BPF_MAX + 1];
553 	struct bpf_cb_ctx *info = cookie;
554 
555 	if (!info || !info->opts)
556 		return -EINVAL;
557 	if (unicast && info->processed)
558 		return -EINVAL;
559 	if (!tb[TCA_OPTIONS])
560 		return NL_CONT;
561 
562 	libbpf_nla_parse_nested(tbb, TCA_BPF_MAX, tb[TCA_OPTIONS], NULL);
563 	if (!tbb[TCA_BPF_ID])
564 		return -EINVAL;
565 
566 	OPTS_SET(info->opts, prog_id, libbpf_nla_getattr_u32(tbb[TCA_BPF_ID]));
567 	OPTS_SET(info->opts, handle, tc->tcm_handle);
568 	OPTS_SET(info->opts, priority, TC_H_MAJ(tc->tcm_info) >> 16);
569 
570 	info->processed = true;
571 	return unicast ? NL_NEXT : NL_DONE;
572 }
573 
574 static int get_tc_info(struct nlmsghdr *nh, libbpf_dump_nlmsg_t fn,
575 		       void *cookie)
576 {
577 	struct tcmsg *tc = NLMSG_DATA(nh);
578 	struct nlattr *tb[TCA_MAX + 1];
579 
580 	libbpf_nla_parse(tb, TCA_MAX,
581 			 (struct nlattr *)((void *)tc + NLMSG_ALIGN(sizeof(*tc))),
582 			 NLMSG_PAYLOAD(nh, sizeof(*tc)), NULL);
583 	if (!tb[TCA_KIND])
584 		return NL_CONT;
585 	return __get_tc_info(cookie, tc, tb, nh->nlmsg_flags & NLM_F_ECHO);
586 }
587 
588 static int tc_add_fd_and_name(struct libbpf_nla_req *req, int fd)
589 {
590 	struct bpf_prog_info info = {};
591 	__u32 info_len = sizeof(info);
592 	char name[256];
593 	int len, ret;
594 
595 	ret = bpf_obj_get_info_by_fd(fd, &info, &info_len);
596 	if (ret < 0)
597 		return ret;
598 
599 	ret = nlattr_add(req, TCA_BPF_FD, &fd, sizeof(fd));
600 	if (ret < 0)
601 		return ret;
602 	len = snprintf(name, sizeof(name), "%s:[%u]", info.name, info.id);
603 	if (len < 0)
604 		return -errno;
605 	if (len >= sizeof(name))
606 		return -ENAMETOOLONG;
607 	return nlattr_add(req, TCA_BPF_NAME, name, len + 1);
608 }
609 
610 int bpf_tc_attach(const struct bpf_tc_hook *hook, struct bpf_tc_opts *opts)
611 {
612 	__u32 protocol, bpf_flags, handle, priority, parent, prog_id, flags;
613 	int ret, ifindex, attach_point, prog_fd;
614 	struct bpf_cb_ctx info = {};
615 	struct libbpf_nla_req req;
616 	struct nlattr *nla;
617 
618 	if (!hook || !opts ||
619 	    !OPTS_VALID(hook, bpf_tc_hook) ||
620 	    !OPTS_VALID(opts, bpf_tc_opts))
621 		return libbpf_err(-EINVAL);
622 
623 	ifindex      = OPTS_GET(hook, ifindex, 0);
624 	parent       = OPTS_GET(hook, parent, 0);
625 	attach_point = OPTS_GET(hook, attach_point, 0);
626 
627 	handle       = OPTS_GET(opts, handle, 0);
628 	priority     = OPTS_GET(opts, priority, 0);
629 	prog_fd      = OPTS_GET(opts, prog_fd, 0);
630 	prog_id      = OPTS_GET(opts, prog_id, 0);
631 	flags        = OPTS_GET(opts, flags, 0);
632 
633 	if (ifindex <= 0 || !prog_fd || prog_id)
634 		return libbpf_err(-EINVAL);
635 	if (priority > UINT16_MAX)
636 		return libbpf_err(-EINVAL);
637 	if (flags & ~BPF_TC_F_REPLACE)
638 		return libbpf_err(-EINVAL);
639 
640 	flags = (flags & BPF_TC_F_REPLACE) ? NLM_F_REPLACE : NLM_F_EXCL;
641 	protocol = ETH_P_ALL;
642 
643 	memset(&req, 0, sizeof(req));
644 	req.nh.nlmsg_len   = NLMSG_LENGTH(sizeof(struct tcmsg));
645 	req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE |
646 			     NLM_F_ECHO | flags;
647 	req.nh.nlmsg_type  = RTM_NEWTFILTER;
648 	req.tc.tcm_family  = AF_UNSPEC;
649 	req.tc.tcm_ifindex = ifindex;
650 	req.tc.tcm_handle  = handle;
651 	req.tc.tcm_info    = TC_H_MAKE(priority << 16, htons(protocol));
652 
653 	ret = tc_get_tcm_parent(attach_point, &parent);
654 	if (ret < 0)
655 		return libbpf_err(ret);
656 	req.tc.tcm_parent = parent;
657 
658 	ret = nlattr_add(&req, TCA_KIND, "bpf", sizeof("bpf"));
659 	if (ret < 0)
660 		return libbpf_err(ret);
661 	nla = nlattr_begin_nested(&req, TCA_OPTIONS);
662 	if (!nla)
663 		return libbpf_err(-EMSGSIZE);
664 	ret = tc_add_fd_and_name(&req, prog_fd);
665 	if (ret < 0)
666 		return libbpf_err(ret);
667 	bpf_flags = TCA_BPF_FLAG_ACT_DIRECT;
668 	ret = nlattr_add(&req, TCA_BPF_FLAGS, &bpf_flags, sizeof(bpf_flags));
669 	if (ret < 0)
670 		return libbpf_err(ret);
671 	nlattr_end_nested(&req, nla);
672 
673 	info.opts = opts;
674 
675 	ret = libbpf_netlink_send_recv(&req, get_tc_info, NULL, &info);
676 	if (ret < 0)
677 		return libbpf_err(ret);
678 	if (!info.processed)
679 		return libbpf_err(-ENOENT);
680 	return ret;
681 }
682 
683 static int __bpf_tc_detach(const struct bpf_tc_hook *hook,
684 			   const struct bpf_tc_opts *opts,
685 			   const bool flush)
686 {
687 	__u32 protocol = 0, handle, priority, parent, prog_id, flags;
688 	int ret, ifindex, attach_point, prog_fd;
689 	struct libbpf_nla_req req;
690 
691 	if (!hook ||
692 	    !OPTS_VALID(hook, bpf_tc_hook) ||
693 	    !OPTS_VALID(opts, bpf_tc_opts))
694 		return -EINVAL;
695 
696 	ifindex      = OPTS_GET(hook, ifindex, 0);
697 	parent       = OPTS_GET(hook, parent, 0);
698 	attach_point = OPTS_GET(hook, attach_point, 0);
699 
700 	handle       = OPTS_GET(opts, handle, 0);
701 	priority     = OPTS_GET(opts, priority, 0);
702 	prog_fd      = OPTS_GET(opts, prog_fd, 0);
703 	prog_id      = OPTS_GET(opts, prog_id, 0);
704 	flags        = OPTS_GET(opts, flags, 0);
705 
706 	if (ifindex <= 0 || flags || prog_fd || prog_id)
707 		return -EINVAL;
708 	if (priority > UINT16_MAX)
709 		return -EINVAL;
710 	if (!flush) {
711 		if (!handle || !priority)
712 			return -EINVAL;
713 		protocol = ETH_P_ALL;
714 	} else {
715 		if (handle || priority)
716 			return -EINVAL;
717 	}
718 
719 	memset(&req, 0, sizeof(req));
720 	req.nh.nlmsg_len   = NLMSG_LENGTH(sizeof(struct tcmsg));
721 	req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
722 	req.nh.nlmsg_type  = RTM_DELTFILTER;
723 	req.tc.tcm_family  = AF_UNSPEC;
724 	req.tc.tcm_ifindex = ifindex;
725 	if (!flush) {
726 		req.tc.tcm_handle = handle;
727 		req.tc.tcm_info   = TC_H_MAKE(priority << 16, htons(protocol));
728 	}
729 
730 	ret = tc_get_tcm_parent(attach_point, &parent);
731 	if (ret < 0)
732 		return ret;
733 	req.tc.tcm_parent = parent;
734 
735 	if (!flush) {
736 		ret = nlattr_add(&req, TCA_KIND, "bpf", sizeof("bpf"));
737 		if (ret < 0)
738 			return ret;
739 	}
740 
741 	return libbpf_netlink_send_recv(&req, NULL, NULL, NULL);
742 }
743 
744 int bpf_tc_detach(const struct bpf_tc_hook *hook,
745 		  const struct bpf_tc_opts *opts)
746 {
747 	int ret;
748 
749 	if (!opts)
750 		return libbpf_err(-EINVAL);
751 
752 	ret = __bpf_tc_detach(hook, opts, false);
753 	return libbpf_err(ret);
754 }
755 
756 int bpf_tc_query(const struct bpf_tc_hook *hook, struct bpf_tc_opts *opts)
757 {
758 	__u32 protocol, handle, priority, parent, prog_id, flags;
759 	int ret, ifindex, attach_point, prog_fd;
760 	struct bpf_cb_ctx info = {};
761 	struct libbpf_nla_req req;
762 
763 	if (!hook || !opts ||
764 	    !OPTS_VALID(hook, bpf_tc_hook) ||
765 	    !OPTS_VALID(opts, bpf_tc_opts))
766 		return libbpf_err(-EINVAL);
767 
768 	ifindex      = OPTS_GET(hook, ifindex, 0);
769 	parent       = OPTS_GET(hook, parent, 0);
770 	attach_point = OPTS_GET(hook, attach_point, 0);
771 
772 	handle       = OPTS_GET(opts, handle, 0);
773 	priority     = OPTS_GET(opts, priority, 0);
774 	prog_fd      = OPTS_GET(opts, prog_fd, 0);
775 	prog_id      = OPTS_GET(opts, prog_id, 0);
776 	flags        = OPTS_GET(opts, flags, 0);
777 
778 	if (ifindex <= 0 || flags || prog_fd || prog_id ||
779 	    !handle || !priority)
780 		return libbpf_err(-EINVAL);
781 	if (priority > UINT16_MAX)
782 		return libbpf_err(-EINVAL);
783 
784 	protocol = ETH_P_ALL;
785 
786 	memset(&req, 0, sizeof(req));
787 	req.nh.nlmsg_len   = NLMSG_LENGTH(sizeof(struct tcmsg));
788 	req.nh.nlmsg_flags = NLM_F_REQUEST;
789 	req.nh.nlmsg_type  = RTM_GETTFILTER;
790 	req.tc.tcm_family  = AF_UNSPEC;
791 	req.tc.tcm_ifindex = ifindex;
792 	req.tc.tcm_handle  = handle;
793 	req.tc.tcm_info    = TC_H_MAKE(priority << 16, htons(protocol));
794 
795 	ret = tc_get_tcm_parent(attach_point, &parent);
796 	if (ret < 0)
797 		return libbpf_err(ret);
798 	req.tc.tcm_parent = parent;
799 
800 	ret = nlattr_add(&req, TCA_KIND, "bpf", sizeof("bpf"));
801 	if (ret < 0)
802 		return libbpf_err(ret);
803 
804 	info.opts = opts;
805 
806 	ret = libbpf_netlink_send_recv(&req, get_tc_info, NULL, &info);
807 	if (ret < 0)
808 		return libbpf_err(ret);
809 	if (!info.processed)
810 		return libbpf_err(-ENOENT);
811 	return ret;
812 }
813