xref: /freebsd/sys/netlink/netlink_io.c (revision 7fdf597e96a02165cfe22ff357b857d5fa15ed8a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2021 Ng Peng Nam Sean
5  * Copyright (c) 2022 Alexander V. Chernikov <melifaro@FreeBSD.org>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/ck.h>
31 #include <sys/lock.h>
32 #include <sys/malloc.h>
33 #include <sys/mbuf.h>
34 #include <sys/mutex.h>
35 #include <sys/socket.h>
36 #include <sys/socketvar.h>
37 #include <sys/syslog.h>
38 
39 #include <netlink/netlink.h>
40 #include <netlink/netlink_ctl.h>
41 #include <netlink/netlink_linux.h>
42 #include <netlink/netlink_var.h>
43 
44 #define	DEBUG_MOD_NAME	nl_io
45 #define	DEBUG_MAX_LEVEL	LOG_DEBUG3
46 #include <netlink/netlink_debug.h>
47 _DECLARE_DEBUG(LOG_INFO);
48 
49 /*
50  * The logic below provide a p2p interface for receiving and
51  * sending netlink data between the kernel and userland.
52  */
53 
54 static bool nl_process_nbuf(struct nl_buf *nb, struct nlpcb *nlp);
55 
56 struct nl_buf *
57 nl_buf_alloc(size_t len, int mflag)
58 {
59 	struct nl_buf *nb;
60 
61 	KASSERT(len > 0 && len <= UINT_MAX, ("%s: invalid length %zu",
62 	    __func__, len));
63 
64 	nb = malloc(sizeof(struct nl_buf) + len, M_NETLINK, mflag);
65 	if (__predict_true(nb != NULL)) {
66 		nb->buflen = len;
67 		nb->datalen = nb->offset = 0;
68 	}
69 
70 	return (nb);
71 }
72 
73 void
74 nl_buf_free(struct nl_buf *nb)
75 {
76 
77 	free(nb, M_NETLINK);
78 }
79 
80 void
81 nl_schedule_taskqueue(struct nlpcb *nlp)
82 {
83 	if (!nlp->nl_task_pending) {
84 		nlp->nl_task_pending = true;
85 		taskqueue_enqueue(nlp->nl_taskqueue, &nlp->nl_task);
86 		NL_LOG(LOG_DEBUG3, "taskqueue scheduled");
87 	} else {
88 		NL_LOG(LOG_DEBUG3, "taskqueue schedule skipped");
89 	}
90 }
91 
92 static bool
93 nl_process_received_one(struct nlpcb *nlp)
94 {
95 	struct socket *so = nlp->nl_socket;
96 	struct sockbuf *sb;
97 	struct nl_buf *nb;
98 	bool reschedule = false;
99 
100 	NLP_LOCK(nlp);
101 	nlp->nl_task_pending = false;
102 	NLP_UNLOCK(nlp);
103 
104 	/*
105 	 * Do not process queued up requests if there is no space to queue
106 	 * replies.
107 	 */
108 	sb = &so->so_rcv;
109 	SOCK_RECVBUF_LOCK(so);
110 	if (sb->sb_hiwat <= sb->sb_ccc) {
111 		SOCK_RECVBUF_UNLOCK(so);
112 		return (false);
113 	}
114 	SOCK_RECVBUF_UNLOCK(so);
115 
116 	sb = &so->so_snd;
117 	SOCK_SENDBUF_LOCK(so);
118 	while ((nb = TAILQ_FIRST(&sb->nl_queue)) != NULL) {
119 		TAILQ_REMOVE(&sb->nl_queue, nb, tailq);
120 		SOCK_SENDBUF_UNLOCK(so);
121 		reschedule = nl_process_nbuf(nb, nlp);
122 		SOCK_SENDBUF_LOCK(so);
123 		if (reschedule) {
124 			sb->sb_acc -= nb->datalen;
125 			sb->sb_ccc -= nb->datalen;
126 			/* XXXGL: potentially can reduce lock&unlock count. */
127 			sowwakeup_locked(so);
128 			nl_buf_free(nb);
129 			SOCK_SENDBUF_LOCK(so);
130 		} else {
131 			TAILQ_INSERT_HEAD(&sb->nl_queue, nb, tailq);
132 			break;
133 		}
134 	}
135 	SOCK_SENDBUF_UNLOCK(so);
136 
137 	return (reschedule);
138 }
139 
140 static void
141 nl_process_received(struct nlpcb *nlp)
142 {
143 	NL_LOG(LOG_DEBUG3, "taskqueue called");
144 
145 	if (__predict_false(nlp->nl_need_thread_setup)) {
146 		nl_set_thread_nlp(curthread, nlp);
147 		NLP_LOCK(nlp);
148 		nlp->nl_need_thread_setup = false;
149 		NLP_UNLOCK(nlp);
150 	}
151 
152 	while (nl_process_received_one(nlp))
153 		;
154 }
155 
156 /*
157  * Called after some data have been read from the socket.
158  */
159 void
160 nl_on_transmit(struct nlpcb *nlp)
161 {
162 	NLP_LOCK(nlp);
163 
164 	struct socket *so = nlp->nl_socket;
165 	if (__predict_false(nlp->nl_dropped_bytes > 0 && so != NULL)) {
166 		unsigned long dropped_bytes = nlp->nl_dropped_bytes;
167 		unsigned long dropped_messages = nlp->nl_dropped_messages;
168 		nlp->nl_dropped_bytes = 0;
169 		nlp->nl_dropped_messages = 0;
170 
171 		struct sockbuf *sb = &so->so_rcv;
172 		NLP_LOG(LOG_DEBUG, nlp,
173 		    "socket RX overflowed, %lu messages (%lu bytes) dropped. "
174 		    "bytes: [%u/%u]", dropped_messages, dropped_bytes,
175 		    sb->sb_ccc, sb->sb_hiwat);
176 		/* TODO: send netlink message */
177 	}
178 
179 	nl_schedule_taskqueue(nlp);
180 	NLP_UNLOCK(nlp);
181 }
182 
183 void
184 nl_taskqueue_handler(void *_arg, int pending)
185 {
186 	struct nlpcb *nlp = (struct nlpcb *)_arg;
187 
188 	CURVNET_SET(nlp->nl_socket->so_vnet);
189 	nl_process_received(nlp);
190 	CURVNET_RESTORE();
191 }
192 
193 /*
194  * Tries to send current data buffer from writer.
195  *
196  * Returns true on success.
197  * If no queue overrunes happened, wakes up socket owner.
198  */
199 bool
200 nl_send(struct nl_writer *nw, struct nlpcb *nlp)
201 {
202 	struct socket *so = nlp->nl_socket;
203 	struct sockbuf *sb = &so->so_rcv;
204 	struct nl_buf *nb;
205 
206 	MPASS(nw->hdr == NULL);
207 	MPASS(nw->buf != NULL);
208 	MPASS(nw->buf->datalen > 0);
209 
210 	IF_DEBUG_LEVEL(LOG_DEBUG2) {
211 		struct nlmsghdr *hdr = (struct nlmsghdr *)nw->buf->data;
212 		NLP_LOG(LOG_DEBUG2, nlp,
213 		    "TX len %u msgs %u msg type %d first hdrlen %u",
214 		    nw->buf->datalen, nw->num_messages, hdr->nlmsg_type,
215 		    hdr->nlmsg_len);
216 	}
217 
218 	if (nlp->nl_linux && linux_netlink_p != NULL &&
219 	    __predict_false(!linux_netlink_p->msgs_to_linux(nw, nlp))) {
220 		nl_buf_free(nw->buf);
221 		nw->buf = NULL;
222 		return (false);
223 	}
224 
225 	nb = nw->buf;
226 	nw->buf = NULL;
227 
228 	SOCK_RECVBUF_LOCK(so);
229 	if (!nw->ignore_limit && __predict_false(sb->sb_hiwat <= sb->sb_ccc)) {
230 		SOCK_RECVBUF_UNLOCK(so);
231 		NLP_LOCK(nlp);
232 		nlp->nl_dropped_bytes += nb->datalen;
233 		nlp->nl_dropped_messages += nw->num_messages;
234 		NLP_LOG(LOG_DEBUG2, nlp, "RX oveflow: %lu m (+%d), %lu b (+%d)",
235 		    (unsigned long)nlp->nl_dropped_messages, nw->num_messages,
236 		    (unsigned long)nlp->nl_dropped_bytes, nb->datalen);
237 		NLP_UNLOCK(nlp);
238 		nl_buf_free(nb);
239 		return (false);
240 	} else {
241 		bool full;
242 
243 		TAILQ_INSERT_TAIL(&sb->nl_queue, nb, tailq);
244 		sb->sb_acc += nb->datalen;
245 		sb->sb_ccc += nb->datalen;
246 		full = sb->sb_hiwat <= sb->sb_ccc;
247 		sorwakeup_locked(so);
248 		if (full) {
249 			NLP_LOCK(nlp);
250 			nlp->nl_tx_blocked = true;
251 			NLP_UNLOCK(nlp);
252 		}
253 		return (true);
254 	}
255 }
256 
257 static int
258 nl_receive_message(struct nlmsghdr *hdr, int remaining_length,
259     struct nlpcb *nlp, struct nl_pstate *npt)
260 {
261 	nl_handler_f handler = nl_handlers[nlp->nl_proto].cb;
262 	int error = 0;
263 
264 	NLP_LOG(LOG_DEBUG2, nlp, "msg len: %u type: %d: flags: 0x%X seq: %u pid: %u",
265 	    hdr->nlmsg_len, hdr->nlmsg_type, hdr->nlmsg_flags, hdr->nlmsg_seq,
266 	    hdr->nlmsg_pid);
267 
268 	if (__predict_false(hdr->nlmsg_len > remaining_length)) {
269 		NLP_LOG(LOG_DEBUG, nlp, "message is not entirely present: want %d got %d",
270 		    hdr->nlmsg_len, remaining_length);
271 		return (EINVAL);
272 	} else if (__predict_false(hdr->nlmsg_len < sizeof(*hdr))) {
273 		NL_LOG(LOG_DEBUG, "message too short: %d", hdr->nlmsg_len);
274 		return (EINVAL);
275 	}
276 	/* Stamp each message with sender pid */
277 	hdr->nlmsg_pid = nlp->nl_port;
278 
279 	npt->hdr = hdr;
280 
281 	if (hdr->nlmsg_flags & NLM_F_REQUEST &&
282 	    hdr->nlmsg_type >= NLMSG_MIN_TYPE) {
283 		NL_LOG(LOG_DEBUG2, "handling message with msg type: %d",
284 		   hdr->nlmsg_type);
285 		if (nlp->nl_linux) {
286 			MPASS(linux_netlink_p != NULL);
287 			error = linux_netlink_p->msg_from_linux(nlp->nl_proto,
288 			    &hdr, npt);
289 			if (error)
290 				goto ack;
291 		}
292 		error = handler(hdr, npt);
293 		NL_LOG(LOG_DEBUG2, "retcode: %d", error);
294 	}
295 ack:
296 	if ((hdr->nlmsg_flags & NLM_F_ACK) || (error != 0 && error != EINTR)) {
297 		if (!npt->nw->suppress_ack) {
298 			NL_LOG(LOG_DEBUG3, "ack");
299 			nlmsg_ack(nlp, error, hdr, npt);
300 		}
301 	}
302 
303 	return (0);
304 }
305 
306 static void
307 npt_clear(struct nl_pstate *npt)
308 {
309 	lb_clear(&npt->lb);
310 	npt->error = 0;
311 	npt->err_msg = NULL;
312 	npt->err_off = 0;
313 	npt->hdr = NULL;
314 	npt->nw->suppress_ack = false;
315 }
316 
317 /*
318  * Processes an incoming packet, which can contain multiple netlink messages
319  */
320 static bool
321 nl_process_nbuf(struct nl_buf *nb, struct nlpcb *nlp)
322 {
323 	struct nl_writer nw;
324 	struct nlmsghdr *hdr;
325 	int error;
326 
327 	NL_LOG(LOG_DEBUG3, "RX netlink buf %p on %p", nb, nlp->nl_socket);
328 
329 	if (!nl_writer_unicast(&nw, NLMSG_SMALL, nlp, false)) {
330 		NL_LOG(LOG_DEBUG, "error allocating socket writer");
331 		return (true);
332 	}
333 
334 	nlmsg_ignore_limit(&nw);
335 
336 	struct nl_pstate npt = {
337 		.nlp = nlp,
338 		.lb.base = &nb->data[roundup2(nb->datalen, 8)],
339 		.lb.size = nb->buflen - roundup2(nb->datalen, 8),
340 		.nw = &nw,
341 		.strict = nlp->nl_flags & NLF_STRICT,
342 	};
343 
344 	for (; nb->offset + sizeof(struct nlmsghdr) <= nb->datalen;) {
345 		hdr = (struct nlmsghdr *)&nb->data[nb->offset];
346 		/* Save length prior to calling handler */
347 		int msglen = NLMSG_ALIGN(hdr->nlmsg_len);
348 		NL_LOG(LOG_DEBUG3, "parsing offset %d/%d",
349 		    nb->offset, nb->datalen);
350 		npt_clear(&npt);
351 		error = nl_receive_message(hdr, nb->datalen - nb->offset, nlp,
352 		    &npt);
353 		nb->offset += msglen;
354 		if (__predict_false(error != 0 || nlp->nl_tx_blocked))
355 			break;
356 	}
357 	NL_LOG(LOG_DEBUG3, "packet parsing done");
358 	nlmsg_flush(&nw);
359 
360 	if (nlp->nl_tx_blocked) {
361 		NLP_LOCK(nlp);
362 		nlp->nl_tx_blocked = false;
363 		NLP_UNLOCK(nlp);
364 		return (false);
365 	} else
366 		return (true);
367 }
368