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_id_md {
31 	int ifindex;
32 	__u32 flags;
33 	struct xdp_link_info info;
34 };
35 
libbpf_netlink_open(__u32 * nl_pid)36 static int libbpf_netlink_open(__u32 *nl_pid)
37 {
38 	struct sockaddr_nl sa;
39 	socklen_t addrlen;
40 	int one = 1, ret;
41 	int sock;
42 
43 	memset(&sa, 0, sizeof(sa));
44 	sa.nl_family = AF_NETLINK;
45 
46 	sock = socket(AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC, NETLINK_ROUTE);
47 	if (sock < 0)
48 		return -errno;
49 
50 	if (setsockopt(sock, SOL_NETLINK, NETLINK_EXT_ACK,
51 		       &one, sizeof(one)) < 0) {
52 		pr_warn("Netlink error reporting not supported\n");
53 	}
54 
55 	if (bind(sock, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
56 		ret = -errno;
57 		goto cleanup;
58 	}
59 
60 	addrlen = sizeof(sa);
61 	if (getsockname(sock, (struct sockaddr *)&sa, &addrlen) < 0) {
62 		ret = -errno;
63 		goto cleanup;
64 	}
65 
66 	if (addrlen != sizeof(sa)) {
67 		ret = -LIBBPF_ERRNO__INTERNAL;
68 		goto cleanup;
69 	}
70 
71 	*nl_pid = sa.nl_pid;
72 	return sock;
73 
74 cleanup:
75 	close(sock);
76 	return ret;
77 }
78 
libbpf_netlink_close(int sock)79 static void libbpf_netlink_close(int sock)
80 {
81 	close(sock);
82 }
83 
84 enum {
85 	NL_CONT,
86 	NL_NEXT,
87 	NL_DONE,
88 };
89 
netlink_recvmsg(int sock,struct msghdr * mhdr,int flags)90 static int netlink_recvmsg(int sock, struct msghdr *mhdr, int flags)
91 {
92 	int len;
93 
94 	do {
95 		len = recvmsg(sock, mhdr, flags);
96 	} while (len < 0 && (errno == EINTR || errno == EAGAIN));
97 
98 	if (len < 0)
99 		return -errno;
100 	return len;
101 }
102 
alloc_iov(struct iovec * iov,int len)103 static int alloc_iov(struct iovec *iov, int len)
104 {
105 	void *nbuf;
106 
107 	nbuf = realloc(iov->iov_base, len);
108 	if (!nbuf)
109 		return -ENOMEM;
110 
111 	iov->iov_base = nbuf;
112 	iov->iov_len = len;
113 	return 0;
114 }
115 
libbpf_netlink_recv(int sock,__u32 nl_pid,int seq,__dump_nlmsg_t _fn,libbpf_dump_nlmsg_t fn,void * cookie)116 static int libbpf_netlink_recv(int sock, __u32 nl_pid, int seq,
117 			       __dump_nlmsg_t _fn, libbpf_dump_nlmsg_t fn,
118 			       void *cookie)
119 {
120 	struct iovec iov = {};
121 	struct msghdr mhdr = {
122 		.msg_iov = &iov,
123 		.msg_iovlen = 1,
124 	};
125 	bool multipart = true;
126 	struct nlmsgerr *err;
127 	struct nlmsghdr *nh;
128 	int len, ret;
129 
130 	ret = alloc_iov(&iov, 4096);
131 	if (ret)
132 		goto done;
133 
134 	while (multipart) {
135 start:
136 		multipart = false;
137 		len = netlink_recvmsg(sock, &mhdr, MSG_PEEK | MSG_TRUNC);
138 		if (len < 0) {
139 			ret = len;
140 			goto done;
141 		}
142 
143 		if (len > iov.iov_len) {
144 			ret = alloc_iov(&iov, len);
145 			if (ret)
146 				goto done;
147 		}
148 
149 		len = netlink_recvmsg(sock, &mhdr, 0);
150 		if (len < 0) {
151 			ret = len;
152 			goto done;
153 		}
154 
155 		if (len == 0)
156 			break;
157 
158 		for (nh = (struct nlmsghdr *)iov.iov_base; NLMSG_OK(nh, len);
159 		     nh = NLMSG_NEXT(nh, len)) {
160 			if (nh->nlmsg_pid != nl_pid) {
161 				ret = -LIBBPF_ERRNO__WRNGPID;
162 				goto done;
163 			}
164 			if (nh->nlmsg_seq != seq) {
165 				ret = -LIBBPF_ERRNO__INVSEQ;
166 				goto done;
167 			}
168 			if (nh->nlmsg_flags & NLM_F_MULTI)
169 				multipart = true;
170 			switch (nh->nlmsg_type) {
171 			case NLMSG_ERROR:
172 				err = (struct nlmsgerr *)NLMSG_DATA(nh);
173 				if (!err->error)
174 					continue;
175 				ret = err->error;
176 				libbpf_nla_dump_errormsg(nh);
177 				goto done;
178 			case NLMSG_DONE:
179 				ret = 0;
180 				goto done;
181 			default:
182 				break;
183 			}
184 			if (_fn) {
185 				ret = _fn(nh, fn, cookie);
186 				switch (ret) {
187 				case NL_CONT:
188 					break;
189 				case NL_NEXT:
190 					goto start;
191 				case NL_DONE:
192 					ret = 0;
193 					goto done;
194 				default:
195 					goto done;
196 				}
197 			}
198 		}
199 	}
200 	ret = 0;
201 done:
202 	free(iov.iov_base);
203 	return ret;
204 }
205 
libbpf_netlink_send_recv(struct libbpf_nla_req * req,__dump_nlmsg_t parse_msg,libbpf_dump_nlmsg_t parse_attr,void * cookie)206 static int libbpf_netlink_send_recv(struct libbpf_nla_req *req,
207 				    __dump_nlmsg_t parse_msg,
208 				    libbpf_dump_nlmsg_t parse_attr,
209 				    void *cookie)
210 {
211 	__u32 nl_pid = 0;
212 	int sock, ret;
213 
214 	sock = libbpf_netlink_open(&nl_pid);
215 	if (sock < 0)
216 		return sock;
217 
218 	req->nh.nlmsg_pid = 0;
219 	req->nh.nlmsg_seq = time(NULL);
220 
221 	if (send(sock, req, req->nh.nlmsg_len, 0) < 0) {
222 		ret = -errno;
223 		goto out;
224 	}
225 
226 	ret = libbpf_netlink_recv(sock, nl_pid, req->nh.nlmsg_seq,
227 				  parse_msg, parse_attr, cookie);
228 out:
229 	libbpf_netlink_close(sock);
230 	return ret;
231 }
232 
__bpf_set_link_xdp_fd_replace(int ifindex,int fd,int old_fd,__u32 flags)233 static int __bpf_set_link_xdp_fd_replace(int ifindex, int fd, int old_fd,
234 					 __u32 flags)
235 {
236 	struct nlattr *nla;
237 	int ret;
238 	struct libbpf_nla_req req;
239 
240 	memset(&req, 0, sizeof(req));
241 	req.nh.nlmsg_len      = NLMSG_LENGTH(sizeof(struct ifinfomsg));
242 	req.nh.nlmsg_flags    = NLM_F_REQUEST | NLM_F_ACK;
243 	req.nh.nlmsg_type     = RTM_SETLINK;
244 	req.ifinfo.ifi_family = AF_UNSPEC;
245 	req.ifinfo.ifi_index  = ifindex;
246 
247 	nla = nlattr_begin_nested(&req, IFLA_XDP);
248 	if (!nla)
249 		return -EMSGSIZE;
250 	ret = nlattr_add(&req, IFLA_XDP_FD, &fd, sizeof(fd));
251 	if (ret < 0)
252 		return ret;
253 	if (flags) {
254 		ret = nlattr_add(&req, IFLA_XDP_FLAGS, &flags, sizeof(flags));
255 		if (ret < 0)
256 			return ret;
257 	}
258 	if (flags & XDP_FLAGS_REPLACE) {
259 		ret = nlattr_add(&req, IFLA_XDP_EXPECTED_FD, &old_fd,
260 				 sizeof(old_fd));
261 		if (ret < 0)
262 			return ret;
263 	}
264 	nlattr_end_nested(&req, nla);
265 
266 	return libbpf_netlink_send_recv(&req, NULL, NULL, NULL);
267 }
268 
bpf_xdp_attach(int ifindex,int prog_fd,__u32 flags,const struct bpf_xdp_attach_opts * opts)269 int bpf_xdp_attach(int ifindex, int prog_fd, __u32 flags, const struct bpf_xdp_attach_opts *opts)
270 {
271 	int old_prog_fd, err;
272 
273 	if (!OPTS_VALID(opts, bpf_xdp_attach_opts))
274 		return libbpf_err(-EINVAL);
275 
276 	old_prog_fd = OPTS_GET(opts, old_prog_fd, 0);
277 	if (old_prog_fd)
278 		flags |= XDP_FLAGS_REPLACE;
279 	else
280 		old_prog_fd = -1;
281 
282 	err = __bpf_set_link_xdp_fd_replace(ifindex, prog_fd, old_prog_fd, flags);
283 	return libbpf_err(err);
284 }
285 
bpf_xdp_detach(int ifindex,__u32 flags,const struct bpf_xdp_attach_opts * opts)286 int bpf_xdp_detach(int ifindex, __u32 flags, const struct bpf_xdp_attach_opts *opts)
287 {
288 	return bpf_xdp_attach(ifindex, -1, flags, opts);
289 }
290 
bpf_set_link_xdp_fd_opts(int ifindex,int fd,__u32 flags,const struct bpf_xdp_set_link_opts * opts)291 int bpf_set_link_xdp_fd_opts(int ifindex, int fd, __u32 flags,
292 			     const struct bpf_xdp_set_link_opts *opts)
293 {
294 	int old_fd = -1, ret;
295 
296 	if (!OPTS_VALID(opts, bpf_xdp_set_link_opts))
297 		return libbpf_err(-EINVAL);
298 
299 	if (OPTS_HAS(opts, old_fd)) {
300 		old_fd = OPTS_GET(opts, old_fd, -1);
301 		flags |= XDP_FLAGS_REPLACE;
302 	}
303 
304 	ret = __bpf_set_link_xdp_fd_replace(ifindex, fd, old_fd, flags);
305 	return libbpf_err(ret);
306 }
307 
bpf_set_link_xdp_fd(int ifindex,int fd,__u32 flags)308 int bpf_set_link_xdp_fd(int ifindex, int fd, __u32 flags)
309 {
310 	int ret;
311 
312 	ret = __bpf_set_link_xdp_fd_replace(ifindex, fd, 0, flags);
313 	return libbpf_err(ret);
314 }
315 
__dump_link_nlmsg(struct nlmsghdr * nlh,libbpf_dump_nlmsg_t dump_link_nlmsg,void * cookie)316 static int __dump_link_nlmsg(struct nlmsghdr *nlh,
317 			     libbpf_dump_nlmsg_t dump_link_nlmsg, void *cookie)
318 {
319 	struct nlattr *tb[IFLA_MAX + 1], *attr;
320 	struct ifinfomsg *ifi = NLMSG_DATA(nlh);
321 	int len;
322 
323 	len = nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*ifi));
324 	attr = (struct nlattr *) ((void *) ifi + NLMSG_ALIGN(sizeof(*ifi)));
325 
326 	if (libbpf_nla_parse(tb, IFLA_MAX, attr, len, NULL) != 0)
327 		return -LIBBPF_ERRNO__NLPARSE;
328 
329 	return dump_link_nlmsg(cookie, ifi, tb);
330 }
331 
get_xdp_info(void * cookie,void * msg,struct nlattr ** tb)332 static int get_xdp_info(void *cookie, void *msg, struct nlattr **tb)
333 {
334 	struct nlattr *xdp_tb[IFLA_XDP_MAX + 1];
335 	struct xdp_id_md *xdp_id = cookie;
336 	struct ifinfomsg *ifinfo = msg;
337 	int ret;
338 
339 	if (xdp_id->ifindex && xdp_id->ifindex != ifinfo->ifi_index)
340 		return 0;
341 
342 	if (!tb[IFLA_XDP])
343 		return 0;
344 
345 	ret = libbpf_nla_parse_nested(xdp_tb, IFLA_XDP_MAX, tb[IFLA_XDP], NULL);
346 	if (ret)
347 		return ret;
348 
349 	if (!xdp_tb[IFLA_XDP_ATTACHED])
350 		return 0;
351 
352 	xdp_id->info.attach_mode = libbpf_nla_getattr_u8(
353 		xdp_tb[IFLA_XDP_ATTACHED]);
354 
355 	if (xdp_id->info.attach_mode == XDP_ATTACHED_NONE)
356 		return 0;
357 
358 	if (xdp_tb[IFLA_XDP_PROG_ID])
359 		xdp_id->info.prog_id = libbpf_nla_getattr_u32(
360 			xdp_tb[IFLA_XDP_PROG_ID]);
361 
362 	if (xdp_tb[IFLA_XDP_SKB_PROG_ID])
363 		xdp_id->info.skb_prog_id = libbpf_nla_getattr_u32(
364 			xdp_tb[IFLA_XDP_SKB_PROG_ID]);
365 
366 	if (xdp_tb[IFLA_XDP_DRV_PROG_ID])
367 		xdp_id->info.drv_prog_id = libbpf_nla_getattr_u32(
368 			xdp_tb[IFLA_XDP_DRV_PROG_ID]);
369 
370 	if (xdp_tb[IFLA_XDP_HW_PROG_ID])
371 		xdp_id->info.hw_prog_id = libbpf_nla_getattr_u32(
372 			xdp_tb[IFLA_XDP_HW_PROG_ID]);
373 
374 	return 0;
375 }
376 
bpf_xdp_query(int ifindex,int xdp_flags,struct bpf_xdp_query_opts * opts)377 int bpf_xdp_query(int ifindex, int xdp_flags, struct bpf_xdp_query_opts *opts)
378 {
379 	struct libbpf_nla_req req = {
380 		.nh.nlmsg_len      = NLMSG_LENGTH(sizeof(struct ifinfomsg)),
381 		.nh.nlmsg_type     = RTM_GETLINK,
382 		.nh.nlmsg_flags    = NLM_F_DUMP | NLM_F_REQUEST,
383 		.ifinfo.ifi_family = AF_PACKET,
384 	};
385 	struct xdp_id_md xdp_id = {};
386 	int err;
387 
388 	if (!OPTS_VALID(opts, bpf_xdp_query_opts))
389 		return libbpf_err(-EINVAL);
390 
391 	if (xdp_flags & ~XDP_FLAGS_MASK)
392 		return libbpf_err(-EINVAL);
393 
394 	/* Check whether the single {HW,DRV,SKB} mode is set */
395 	xdp_flags &= XDP_FLAGS_SKB_MODE | XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE;
396 	if (xdp_flags & (xdp_flags - 1))
397 		return libbpf_err(-EINVAL);
398 
399 	xdp_id.ifindex = ifindex;
400 	xdp_id.flags = xdp_flags;
401 
402 	err = libbpf_netlink_send_recv(&req, __dump_link_nlmsg,
403 				       get_xdp_info, &xdp_id);
404 	if (err)
405 		return libbpf_err(err);
406 
407 	OPTS_SET(opts, prog_id, xdp_id.info.prog_id);
408 	OPTS_SET(opts, drv_prog_id, xdp_id.info.drv_prog_id);
409 	OPTS_SET(opts, hw_prog_id, xdp_id.info.hw_prog_id);
410 	OPTS_SET(opts, skb_prog_id, xdp_id.info.skb_prog_id);
411 	OPTS_SET(opts, attach_mode, xdp_id.info.attach_mode);
412 
413 	return 0;
414 }
415 
bpf_get_link_xdp_info(int ifindex,struct xdp_link_info * info,size_t info_size,__u32 flags)416 int bpf_get_link_xdp_info(int ifindex, struct xdp_link_info *info,
417 			  size_t info_size, __u32 flags)
418 {
419 	LIBBPF_OPTS(bpf_xdp_query_opts, opts);
420 	size_t sz;
421 	int err;
422 
423 	if (!info_size)
424 		return libbpf_err(-EINVAL);
425 
426 	err = bpf_xdp_query(ifindex, flags, &opts);
427 	if (err)
428 		return libbpf_err(err);
429 
430 	/* struct xdp_link_info field layout matches struct bpf_xdp_query_opts
431 	 * layout after sz field
432 	 */
433 	sz = min(info_size, offsetofend(struct xdp_link_info, attach_mode));
434 	memcpy(info, &opts.prog_id, sz);
435 	memset((void *)info + sz, 0, info_size - sz);
436 
437 	return 0;
438 }
439 
bpf_xdp_query_id(int ifindex,int flags,__u32 * prog_id)440 int bpf_xdp_query_id(int ifindex, int flags, __u32 *prog_id)
441 {
442 	LIBBPF_OPTS(bpf_xdp_query_opts, opts);
443 	int ret;
444 
445 	ret = bpf_xdp_query(ifindex, flags, &opts);
446 	if (ret)
447 		return libbpf_err(ret);
448 
449 	flags &= XDP_FLAGS_MODES;
450 
451 	if (opts.attach_mode != XDP_ATTACHED_MULTI && !flags)
452 		*prog_id = opts.prog_id;
453 	else if (flags & XDP_FLAGS_DRV_MODE)
454 		*prog_id = opts.drv_prog_id;
455 	else if (flags & XDP_FLAGS_HW_MODE)
456 		*prog_id = opts.hw_prog_id;
457 	else if (flags & XDP_FLAGS_SKB_MODE)
458 		*prog_id = opts.skb_prog_id;
459 	else
460 		*prog_id = 0;
461 
462 	return 0;
463 }
464 
465 
bpf_get_link_xdp_id(int ifindex,__u32 * prog_id,__u32 flags)466 int bpf_get_link_xdp_id(int ifindex, __u32 *prog_id, __u32 flags)
467 {
468 	return bpf_xdp_query_id(ifindex, flags, prog_id);
469 }
470 
471 typedef int (*qdisc_config_t)(struct libbpf_nla_req *req);
472 
clsact_config(struct libbpf_nla_req * req)473 static int clsact_config(struct libbpf_nla_req *req)
474 {
475 	req->tc.tcm_parent = TC_H_CLSACT;
476 	req->tc.tcm_handle = TC_H_MAKE(TC_H_CLSACT, 0);
477 
478 	return nlattr_add(req, TCA_KIND, "clsact", sizeof("clsact"));
479 }
480 
attach_point_to_config(struct bpf_tc_hook * hook,qdisc_config_t * config)481 static int attach_point_to_config(struct bpf_tc_hook *hook,
482 				  qdisc_config_t *config)
483 {
484 	switch (OPTS_GET(hook, attach_point, 0)) {
485 	case BPF_TC_INGRESS:
486 	case BPF_TC_EGRESS:
487 	case BPF_TC_INGRESS | BPF_TC_EGRESS:
488 		if (OPTS_GET(hook, parent, 0))
489 			return -EINVAL;
490 		*config = &clsact_config;
491 		return 0;
492 	case BPF_TC_CUSTOM:
493 		return -EOPNOTSUPP;
494 	default:
495 		return -EINVAL;
496 	}
497 }
498 
tc_get_tcm_parent(enum bpf_tc_attach_point attach_point,__u32 * parent)499 static int tc_get_tcm_parent(enum bpf_tc_attach_point attach_point,
500 			     __u32 *parent)
501 {
502 	switch (attach_point) {
503 	case BPF_TC_INGRESS:
504 	case BPF_TC_EGRESS:
505 		if (*parent)
506 			return -EINVAL;
507 		*parent = TC_H_MAKE(TC_H_CLSACT,
508 				    attach_point == BPF_TC_INGRESS ?
509 				    TC_H_MIN_INGRESS : TC_H_MIN_EGRESS);
510 		break;
511 	case BPF_TC_CUSTOM:
512 		if (!*parent)
513 			return -EINVAL;
514 		break;
515 	default:
516 		return -EINVAL;
517 	}
518 	return 0;
519 }
520 
tc_qdisc_modify(struct bpf_tc_hook * hook,int cmd,int flags)521 static int tc_qdisc_modify(struct bpf_tc_hook *hook, int cmd, int flags)
522 {
523 	qdisc_config_t config;
524 	int ret;
525 	struct libbpf_nla_req req;
526 
527 	ret = attach_point_to_config(hook, &config);
528 	if (ret < 0)
529 		return ret;
530 
531 	memset(&req, 0, sizeof(req));
532 	req.nh.nlmsg_len   = NLMSG_LENGTH(sizeof(struct tcmsg));
533 	req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | flags;
534 	req.nh.nlmsg_type  = cmd;
535 	req.tc.tcm_family  = AF_UNSPEC;
536 	req.tc.tcm_ifindex = OPTS_GET(hook, ifindex, 0);
537 
538 	ret = config(&req);
539 	if (ret < 0)
540 		return ret;
541 
542 	return libbpf_netlink_send_recv(&req, NULL, NULL, NULL);
543 }
544 
tc_qdisc_create_excl(struct bpf_tc_hook * hook)545 static int tc_qdisc_create_excl(struct bpf_tc_hook *hook)
546 {
547 	return tc_qdisc_modify(hook, RTM_NEWQDISC, NLM_F_CREATE | NLM_F_EXCL);
548 }
549 
tc_qdisc_delete(struct bpf_tc_hook * hook)550 static int tc_qdisc_delete(struct bpf_tc_hook *hook)
551 {
552 	return tc_qdisc_modify(hook, RTM_DELQDISC, 0);
553 }
554 
bpf_tc_hook_create(struct bpf_tc_hook * hook)555 int bpf_tc_hook_create(struct bpf_tc_hook *hook)
556 {
557 	int ret;
558 
559 	if (!hook || !OPTS_VALID(hook, bpf_tc_hook) ||
560 	    OPTS_GET(hook, ifindex, 0) <= 0)
561 		return libbpf_err(-EINVAL);
562 
563 	ret = tc_qdisc_create_excl(hook);
564 	return libbpf_err(ret);
565 }
566 
567 static int __bpf_tc_detach(const struct bpf_tc_hook *hook,
568 			   const struct bpf_tc_opts *opts,
569 			   const bool flush);
570 
bpf_tc_hook_destroy(struct bpf_tc_hook * hook)571 int bpf_tc_hook_destroy(struct bpf_tc_hook *hook)
572 {
573 	if (!hook || !OPTS_VALID(hook, bpf_tc_hook) ||
574 	    OPTS_GET(hook, ifindex, 0) <= 0)
575 		return libbpf_err(-EINVAL);
576 
577 	switch (OPTS_GET(hook, attach_point, 0)) {
578 	case BPF_TC_INGRESS:
579 	case BPF_TC_EGRESS:
580 		return libbpf_err(__bpf_tc_detach(hook, NULL, true));
581 	case BPF_TC_INGRESS | BPF_TC_EGRESS:
582 		return libbpf_err(tc_qdisc_delete(hook));
583 	case BPF_TC_CUSTOM:
584 		return libbpf_err(-EOPNOTSUPP);
585 	default:
586 		return libbpf_err(-EINVAL);
587 	}
588 }
589 
590 struct bpf_cb_ctx {
591 	struct bpf_tc_opts *opts;
592 	bool processed;
593 };
594 
__get_tc_info(void * cookie,struct tcmsg * tc,struct nlattr ** tb,bool unicast)595 static int __get_tc_info(void *cookie, struct tcmsg *tc, struct nlattr **tb,
596 			 bool unicast)
597 {
598 	struct nlattr *tbb[TCA_BPF_MAX + 1];
599 	struct bpf_cb_ctx *info = cookie;
600 
601 	if (!info || !info->opts)
602 		return -EINVAL;
603 	if (unicast && info->processed)
604 		return -EINVAL;
605 	if (!tb[TCA_OPTIONS])
606 		return NL_CONT;
607 
608 	libbpf_nla_parse_nested(tbb, TCA_BPF_MAX, tb[TCA_OPTIONS], NULL);
609 	if (!tbb[TCA_BPF_ID])
610 		return -EINVAL;
611 
612 	OPTS_SET(info->opts, prog_id, libbpf_nla_getattr_u32(tbb[TCA_BPF_ID]));
613 	OPTS_SET(info->opts, handle, tc->tcm_handle);
614 	OPTS_SET(info->opts, priority, TC_H_MAJ(tc->tcm_info) >> 16);
615 
616 	info->processed = true;
617 	return unicast ? NL_NEXT : NL_DONE;
618 }
619 
get_tc_info(struct nlmsghdr * nh,libbpf_dump_nlmsg_t fn,void * cookie)620 static int get_tc_info(struct nlmsghdr *nh, libbpf_dump_nlmsg_t fn,
621 		       void *cookie)
622 {
623 	struct tcmsg *tc = NLMSG_DATA(nh);
624 	struct nlattr *tb[TCA_MAX + 1];
625 
626 	libbpf_nla_parse(tb, TCA_MAX,
627 			 (struct nlattr *)((void *)tc + NLMSG_ALIGN(sizeof(*tc))),
628 			 NLMSG_PAYLOAD(nh, sizeof(*tc)), NULL);
629 	if (!tb[TCA_KIND])
630 		return NL_CONT;
631 	return __get_tc_info(cookie, tc, tb, nh->nlmsg_flags & NLM_F_ECHO);
632 }
633 
tc_add_fd_and_name(struct libbpf_nla_req * req,int fd)634 static int tc_add_fd_and_name(struct libbpf_nla_req *req, int fd)
635 {
636 	struct bpf_prog_info info = {};
637 	__u32 info_len = sizeof(info);
638 	char name[256];
639 	int len, ret;
640 
641 	ret = bpf_obj_get_info_by_fd(fd, &info, &info_len);
642 	if (ret < 0)
643 		return ret;
644 
645 	ret = nlattr_add(req, TCA_BPF_FD, &fd, sizeof(fd));
646 	if (ret < 0)
647 		return ret;
648 	len = snprintf(name, sizeof(name), "%s:[%u]", info.name, info.id);
649 	if (len < 0)
650 		return -errno;
651 	if (len >= sizeof(name))
652 		return -ENAMETOOLONG;
653 	return nlattr_add(req, TCA_BPF_NAME, name, len + 1);
654 }
655 
bpf_tc_attach(const struct bpf_tc_hook * hook,struct bpf_tc_opts * opts)656 int bpf_tc_attach(const struct bpf_tc_hook *hook, struct bpf_tc_opts *opts)
657 {
658 	__u32 protocol, bpf_flags, handle, priority, parent, prog_id, flags;
659 	int ret, ifindex, attach_point, prog_fd;
660 	struct bpf_cb_ctx info = {};
661 	struct libbpf_nla_req req;
662 	struct nlattr *nla;
663 
664 	if (!hook || !opts ||
665 	    !OPTS_VALID(hook, bpf_tc_hook) ||
666 	    !OPTS_VALID(opts, bpf_tc_opts))
667 		return libbpf_err(-EINVAL);
668 
669 	ifindex      = OPTS_GET(hook, ifindex, 0);
670 	parent       = OPTS_GET(hook, parent, 0);
671 	attach_point = OPTS_GET(hook, attach_point, 0);
672 
673 	handle       = OPTS_GET(opts, handle, 0);
674 	priority     = OPTS_GET(opts, priority, 0);
675 	prog_fd      = OPTS_GET(opts, prog_fd, 0);
676 	prog_id      = OPTS_GET(opts, prog_id, 0);
677 	flags        = OPTS_GET(opts, flags, 0);
678 
679 	if (ifindex <= 0 || !prog_fd || prog_id)
680 		return libbpf_err(-EINVAL);
681 	if (priority > UINT16_MAX)
682 		return libbpf_err(-EINVAL);
683 	if (flags & ~BPF_TC_F_REPLACE)
684 		return libbpf_err(-EINVAL);
685 
686 	flags = (flags & BPF_TC_F_REPLACE) ? NLM_F_REPLACE : NLM_F_EXCL;
687 	protocol = ETH_P_ALL;
688 
689 	memset(&req, 0, sizeof(req));
690 	req.nh.nlmsg_len   = NLMSG_LENGTH(sizeof(struct tcmsg));
691 	req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_CREATE |
692 			     NLM_F_ECHO | flags;
693 	req.nh.nlmsg_type  = RTM_NEWTFILTER;
694 	req.tc.tcm_family  = AF_UNSPEC;
695 	req.tc.tcm_ifindex = ifindex;
696 	req.tc.tcm_handle  = handle;
697 	req.tc.tcm_info    = TC_H_MAKE(priority << 16, htons(protocol));
698 
699 	ret = tc_get_tcm_parent(attach_point, &parent);
700 	if (ret < 0)
701 		return libbpf_err(ret);
702 	req.tc.tcm_parent = parent;
703 
704 	ret = nlattr_add(&req, TCA_KIND, "bpf", sizeof("bpf"));
705 	if (ret < 0)
706 		return libbpf_err(ret);
707 	nla = nlattr_begin_nested(&req, TCA_OPTIONS);
708 	if (!nla)
709 		return libbpf_err(-EMSGSIZE);
710 	ret = tc_add_fd_and_name(&req, prog_fd);
711 	if (ret < 0)
712 		return libbpf_err(ret);
713 	bpf_flags = TCA_BPF_FLAG_ACT_DIRECT;
714 	ret = nlattr_add(&req, TCA_BPF_FLAGS, &bpf_flags, sizeof(bpf_flags));
715 	if (ret < 0)
716 		return libbpf_err(ret);
717 	nlattr_end_nested(&req, nla);
718 
719 	info.opts = opts;
720 
721 	ret = libbpf_netlink_send_recv(&req, get_tc_info, NULL, &info);
722 	if (ret < 0)
723 		return libbpf_err(ret);
724 	if (!info.processed)
725 		return libbpf_err(-ENOENT);
726 	return ret;
727 }
728 
__bpf_tc_detach(const struct bpf_tc_hook * hook,const struct bpf_tc_opts * opts,const bool flush)729 static int __bpf_tc_detach(const struct bpf_tc_hook *hook,
730 			   const struct bpf_tc_opts *opts,
731 			   const bool flush)
732 {
733 	__u32 protocol = 0, handle, priority, parent, prog_id, flags;
734 	int ret, ifindex, attach_point, prog_fd;
735 	struct libbpf_nla_req req;
736 
737 	if (!hook ||
738 	    !OPTS_VALID(hook, bpf_tc_hook) ||
739 	    !OPTS_VALID(opts, bpf_tc_opts))
740 		return -EINVAL;
741 
742 	ifindex      = OPTS_GET(hook, ifindex, 0);
743 	parent       = OPTS_GET(hook, parent, 0);
744 	attach_point = OPTS_GET(hook, attach_point, 0);
745 
746 	handle       = OPTS_GET(opts, handle, 0);
747 	priority     = OPTS_GET(opts, priority, 0);
748 	prog_fd      = OPTS_GET(opts, prog_fd, 0);
749 	prog_id      = OPTS_GET(opts, prog_id, 0);
750 	flags        = OPTS_GET(opts, flags, 0);
751 
752 	if (ifindex <= 0 || flags || prog_fd || prog_id)
753 		return -EINVAL;
754 	if (priority > UINT16_MAX)
755 		return -EINVAL;
756 	if (!flush) {
757 		if (!handle || !priority)
758 			return -EINVAL;
759 		protocol = ETH_P_ALL;
760 	} else {
761 		if (handle || priority)
762 			return -EINVAL;
763 	}
764 
765 	memset(&req, 0, sizeof(req));
766 	req.nh.nlmsg_len   = NLMSG_LENGTH(sizeof(struct tcmsg));
767 	req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
768 	req.nh.nlmsg_type  = RTM_DELTFILTER;
769 	req.tc.tcm_family  = AF_UNSPEC;
770 	req.tc.tcm_ifindex = ifindex;
771 	if (!flush) {
772 		req.tc.tcm_handle = handle;
773 		req.tc.tcm_info   = TC_H_MAKE(priority << 16, htons(protocol));
774 	}
775 
776 	ret = tc_get_tcm_parent(attach_point, &parent);
777 	if (ret < 0)
778 		return ret;
779 	req.tc.tcm_parent = parent;
780 
781 	if (!flush) {
782 		ret = nlattr_add(&req, TCA_KIND, "bpf", sizeof("bpf"));
783 		if (ret < 0)
784 			return ret;
785 	}
786 
787 	return libbpf_netlink_send_recv(&req, NULL, NULL, NULL);
788 }
789 
bpf_tc_detach(const struct bpf_tc_hook * hook,const struct bpf_tc_opts * opts)790 int bpf_tc_detach(const struct bpf_tc_hook *hook,
791 		  const struct bpf_tc_opts *opts)
792 {
793 	int ret;
794 
795 	if (!opts)
796 		return libbpf_err(-EINVAL);
797 
798 	ret = __bpf_tc_detach(hook, opts, false);
799 	return libbpf_err(ret);
800 }
801 
bpf_tc_query(const struct bpf_tc_hook * hook,struct bpf_tc_opts * opts)802 int bpf_tc_query(const struct bpf_tc_hook *hook, struct bpf_tc_opts *opts)
803 {
804 	__u32 protocol, handle, priority, parent, prog_id, flags;
805 	int ret, ifindex, attach_point, prog_fd;
806 	struct bpf_cb_ctx info = {};
807 	struct libbpf_nla_req req;
808 
809 	if (!hook || !opts ||
810 	    !OPTS_VALID(hook, bpf_tc_hook) ||
811 	    !OPTS_VALID(opts, bpf_tc_opts))
812 		return libbpf_err(-EINVAL);
813 
814 	ifindex      = OPTS_GET(hook, ifindex, 0);
815 	parent       = OPTS_GET(hook, parent, 0);
816 	attach_point = OPTS_GET(hook, attach_point, 0);
817 
818 	handle       = OPTS_GET(opts, handle, 0);
819 	priority     = OPTS_GET(opts, priority, 0);
820 	prog_fd      = OPTS_GET(opts, prog_fd, 0);
821 	prog_id      = OPTS_GET(opts, prog_id, 0);
822 	flags        = OPTS_GET(opts, flags, 0);
823 
824 	if (ifindex <= 0 || flags || prog_fd || prog_id ||
825 	    !handle || !priority)
826 		return libbpf_err(-EINVAL);
827 	if (priority > UINT16_MAX)
828 		return libbpf_err(-EINVAL);
829 
830 	protocol = ETH_P_ALL;
831 
832 	memset(&req, 0, sizeof(req));
833 	req.nh.nlmsg_len   = NLMSG_LENGTH(sizeof(struct tcmsg));
834 	req.nh.nlmsg_flags = NLM_F_REQUEST;
835 	req.nh.nlmsg_type  = RTM_GETTFILTER;
836 	req.tc.tcm_family  = AF_UNSPEC;
837 	req.tc.tcm_ifindex = ifindex;
838 	req.tc.tcm_handle  = handle;
839 	req.tc.tcm_info    = TC_H_MAKE(priority << 16, htons(protocol));
840 
841 	ret = tc_get_tcm_parent(attach_point, &parent);
842 	if (ret < 0)
843 		return libbpf_err(ret);
844 	req.tc.tcm_parent = parent;
845 
846 	ret = nlattr_add(&req, TCA_KIND, "bpf", sizeof("bpf"));
847 	if (ret < 0)
848 		return libbpf_err(ret);
849 
850 	info.opts = opts;
851 
852 	ret = libbpf_netlink_send_recv(&req, get_tc_info, NULL, &info);
853 	if (ret < 0)
854 		return libbpf_err(ret);
855 	if (!info.processed)
856 		return libbpf_err(-ENOENT);
857 	return ret;
858 }
859