xref: /freebsd/lib/libnv/msgio.c (revision a2f733abcff64628b7771a47089628b7327a88bd)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2013 The FreeBSD Foundation
5  * Copyright (c) 2013 Mariusz Zaborski <oshogbo@FreeBSD.org>
6  * All rights reserved.
7  *
8  * This software was developed by Pawel Jakub Dawidek under sponsorship from
9  * the FreeBSD Foundation.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/param.h>
34 #include <sys/socket.h>
35 #include <sys/select.h>
36 
37 #include <errno.h>
38 #include <fcntl.h>
39 #include <stdbool.h>
40 #include <stdint.h>
41 #include <stdlib.h>
42 #include <string.h>
43 #include <unistd.h>
44 
45 #ifdef HAVE_PJDLOG
46 #include <pjdlog.h>
47 #endif
48 
49 #include "common_impl.h"
50 #include "msgio.h"
51 
52 #ifndef	HAVE_PJDLOG
53 #include <assert.h>
54 #define	PJDLOG_ASSERT(...)		assert(__VA_ARGS__)
55 #define	PJDLOG_RASSERT(expr, ...)	assert(expr)
56 #define	PJDLOG_ABORT(...)		abort()
57 #endif
58 
59 #ifdef __linux__
60 /* Linux: arbitrary size, but must be lower than SCM_MAX_FD. */
61 #define	PKG_MAX_SIZE	((64U - 1) * CMSG_SPACE(sizeof(int)))
62 #else
63 /*
64  * To work around limitations in 32-bit emulation on 64-bit kernels, use a
65  * machine-independent limit on the number of FDs per message.  Each control
66  * message contains 1 FD and requires 12 bytes for the header, 4 pad bytes,
67  * 4 bytes for the descriptor, and another 4 pad bytes.
68  */
69 #define	PKG_MAX_SIZE	(MCLBYTES / 24)
70 #endif
71 
72 static int
msghdr_add_fd(struct cmsghdr * cmsg,int fd)73 msghdr_add_fd(struct cmsghdr *cmsg, int fd)
74 {
75 
76 	PJDLOG_ASSERT(fd >= 0);
77 
78 	cmsg->cmsg_level = SOL_SOCKET;
79 	cmsg->cmsg_type = SCM_RIGHTS;
80 	cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
81 	bcopy(&fd, CMSG_DATA(cmsg), sizeof(fd));
82 
83 	return (0);
84 }
85 
86 static void
fd_wait(int fd,bool doread)87 fd_wait(int fd, bool doread)
88 {
89 	fd_set fds;
90 
91 	PJDLOG_ASSERT(fd >= 0);
92 
93 	FD_ZERO(&fds);
94 	FD_SET(fd, &fds);
95 	(void)select(fd + 1, doread ? &fds : NULL, doread ? NULL : &fds,
96 	    NULL, NULL);
97 }
98 
99 static int
msg_recv(int sock,struct msghdr * msg)100 msg_recv(int sock, struct msghdr *msg)
101 {
102 	int flags;
103 
104 	PJDLOG_ASSERT(sock >= 0);
105 
106 #ifdef MSG_CMSG_CLOEXEC
107 	flags = MSG_CMSG_CLOEXEC;
108 #else
109 	flags = 0;
110 #endif
111 
112 	for (;;) {
113 		fd_wait(sock, true);
114 		if (recvmsg(sock, msg, flags) == -1) {
115 			if (errno == EINTR)
116 				continue;
117 			return (-1);
118 		}
119 		break;
120 	}
121 
122 	return (0);
123 }
124 
125 static int
msg_send(int sock,const struct msghdr * msg)126 msg_send(int sock, const struct msghdr *msg)
127 {
128 
129 	PJDLOG_ASSERT(sock >= 0);
130 
131 	for (;;) {
132 		fd_wait(sock, false);
133 		if (sendmsg(sock, msg, 0) == -1) {
134 			if (errno == EINTR)
135 				continue;
136 			return (-1);
137 		}
138 		break;
139 	}
140 
141 	return (0);
142 }
143 
144 #ifdef __FreeBSD__
145 int
cred_send(int sock)146 cred_send(int sock)
147 {
148 	unsigned char credbuf[CMSG_SPACE(sizeof(struct cmsgcred))];
149 	struct msghdr msg;
150 	struct cmsghdr *cmsg;
151 	struct iovec iov;
152 	uint8_t dummy;
153 
154 	bzero(credbuf, sizeof(credbuf));
155 	bzero(&msg, sizeof(msg));
156 	bzero(&iov, sizeof(iov));
157 
158 	/*
159 	 * XXX: We send one byte along with the control message, because
160 	 *      setting msg_iov to NULL only works if this is the first
161 	 *      packet send over the socket. Once we send some data we
162 	 *      won't be able to send credentials anymore. This is most
163 	 *      likely a kernel bug.
164 	 */
165 	dummy = 0;
166 	iov.iov_base = &dummy;
167 	iov.iov_len = sizeof(dummy);
168 
169 	msg.msg_iov = &iov;
170 	msg.msg_iovlen = 1;
171 	msg.msg_control = credbuf;
172 	msg.msg_controllen = sizeof(credbuf);
173 
174 	cmsg = CMSG_FIRSTHDR(&msg);
175 	cmsg->cmsg_len = CMSG_LEN(sizeof(struct cmsgcred));
176 	cmsg->cmsg_level = SOL_SOCKET;
177 	cmsg->cmsg_type = SCM_CREDS;
178 
179 	if (msg_send(sock, &msg) == -1)
180 		return (-1);
181 
182 	return (0);
183 }
184 
185 int
cred_recv(int sock,struct cmsgcred * cred)186 cred_recv(int sock, struct cmsgcred *cred)
187 {
188 	unsigned char credbuf[CMSG_SPACE(sizeof(struct cmsgcred))];
189 	struct msghdr msg;
190 	struct cmsghdr *cmsg;
191 	struct iovec iov;
192 	uint8_t dummy;
193 
194 	bzero(credbuf, sizeof(credbuf));
195 	bzero(&msg, sizeof(msg));
196 	bzero(&iov, sizeof(iov));
197 
198 	iov.iov_base = &dummy;
199 	iov.iov_len = sizeof(dummy);
200 
201 	msg.msg_iov = &iov;
202 	msg.msg_iovlen = 1;
203 	msg.msg_control = credbuf;
204 	msg.msg_controllen = sizeof(credbuf);
205 
206 	if (msg_recv(sock, &msg) == -1)
207 		return (-1);
208 
209 	cmsg = CMSG_FIRSTHDR(&msg);
210 	if (cmsg == NULL ||
211 	    cmsg->cmsg_len != CMSG_LEN(sizeof(struct cmsgcred)) ||
212 	    cmsg->cmsg_level != SOL_SOCKET || cmsg->cmsg_type != SCM_CREDS) {
213 		errno = EINVAL;
214 		return (-1);
215 	}
216 	bcopy(CMSG_DATA(cmsg), cred, sizeof(*cred));
217 
218 	return (0);
219 }
220 #endif
221 
222 static int
fd_package_send(int sock,const int * fds,size_t nfds)223 fd_package_send(int sock, const int *fds, size_t nfds)
224 {
225 	struct msghdr msg;
226 	struct cmsghdr *cmsg;
227 	struct iovec iov;
228 	unsigned int i;
229 	int serrno, ret;
230 	uint8_t dummy;
231 
232 	PJDLOG_ASSERT(sock >= 0);
233 	PJDLOG_ASSERT(fds != NULL);
234 	PJDLOG_ASSERT(nfds > 0);
235 
236 	bzero(&msg, sizeof(msg));
237 
238 	/*
239 	 * XXX: Look into cred_send function for more details.
240 	 */
241 	dummy = 0;
242 	iov.iov_base = &dummy;
243 	iov.iov_len = sizeof(dummy);
244 
245 	msg.msg_iov = &iov;
246 	msg.msg_iovlen = 1;
247 	msg.msg_controllen = nfds * CMSG_SPACE(sizeof(int));
248 	msg.msg_control = calloc(1, msg.msg_controllen);
249 	if (msg.msg_control == NULL)
250 		return (-1);
251 
252 	ret = -1;
253 
254 	for (i = 0, cmsg = CMSG_FIRSTHDR(&msg); i < nfds && cmsg != NULL;
255 	    i++, cmsg = CMSG_NXTHDR(&msg, cmsg)) {
256 		if (msghdr_add_fd(cmsg, fds[i]) == -1)
257 			goto end;
258 	}
259 
260 	if (msg_send(sock, &msg) == -1)
261 		goto end;
262 
263 	ret = 0;
264 end:
265 	serrno = errno;
266 	free(msg.msg_control);
267 	errno = serrno;
268 	return (ret);
269 }
270 
271 static int
fd_package_recv(int sock,int * fds,size_t nfds)272 fd_package_recv(int sock, int *fds, size_t nfds)
273 {
274 	struct msghdr msg;
275 	struct cmsghdr *cmsg;
276 	unsigned int i;
277 	int serrno, ret;
278 	struct iovec iov;
279 	uint8_t dummy;
280 
281 	PJDLOG_ASSERT(sock >= 0);
282 	PJDLOG_ASSERT(nfds > 0);
283 	PJDLOG_ASSERT(fds != NULL);
284 
285 	bzero(&msg, sizeof(msg));
286 	bzero(&iov, sizeof(iov));
287 
288 	/*
289 	 * XXX: Look into cred_send function for more details.
290 	 */
291 	iov.iov_base = &dummy;
292 	iov.iov_len = sizeof(dummy);
293 
294 	msg.msg_iov = &iov;
295 	msg.msg_iovlen = 1;
296 	msg.msg_controllen = nfds * CMSG_SPACE(sizeof(int));
297 	msg.msg_control = calloc(1, msg.msg_controllen);
298 	if (msg.msg_control == NULL)
299 		return (-1);
300 
301 	ret = -1;
302 
303 	if (msg_recv(sock, &msg) == -1)
304 		goto end;
305 
306 	i = 0;
307 	cmsg = CMSG_FIRSTHDR(&msg);
308 	while (cmsg && i < nfds) {
309 		unsigned int n;
310 
311 		if (cmsg->cmsg_level != SOL_SOCKET ||
312 		    cmsg->cmsg_type != SCM_RIGHTS) {
313 			errno = EINVAL;
314 			break;
315 		}
316 		n = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
317 		if (i + n > nfds) {
318 			errno = EINVAL;
319 			break;
320 		}
321 		bcopy(CMSG_DATA(cmsg), fds + i, sizeof(int) * n);
322 		cmsg = CMSG_NXTHDR(&msg, cmsg);
323 		i += n;
324 	}
325 
326 	if (cmsg != NULL || i < nfds) {
327 		unsigned int last;
328 
329 		/*
330 		 * We need to close all received descriptors, even if we have
331 		 * different control message (eg. SCM_CREDS) in between.
332 		 */
333 		last = i;
334 		for (i = 0; i < last; i++) {
335 			if (fds[i] >= 0) {
336 				close(fds[i]);
337 			}
338 		}
339 		errno = EINVAL;
340 		goto end;
341 	}
342 
343 #ifndef MSG_CMSG_CLOEXEC
344 	/*
345 	 * If the MSG_CMSG_CLOEXEC flag is not available we cannot set the
346 	 * close-on-exec flag atomically, but we still want to set it for
347 	 * consistency.
348 	 */
349 	for (i = 0; i < nfds; i++) {
350 		(void) fcntl(fds[i], F_SETFD, FD_CLOEXEC);
351 	}
352 #endif
353 
354 	ret = 0;
355 end:
356 	serrno = errno;
357 	free(msg.msg_control);
358 	errno = serrno;
359 	return (ret);
360 }
361 
362 int
fd_recv(int sock,int * fds,size_t nfds)363 fd_recv(int sock, int *fds, size_t nfds)
364 {
365 	unsigned int i, step, j;
366 	int ret, serrno;
367 
368 	if (nfds == 0 || fds == NULL) {
369 		errno = EINVAL;
370 		return (-1);
371 	}
372 
373 	ret = i = step = 0;
374 	while (i < nfds) {
375 		if (PKG_MAX_SIZE < nfds - i)
376 			step = PKG_MAX_SIZE;
377 		else
378 			step = nfds - i;
379 		ret = fd_package_recv(sock, fds + i, step);
380 		if (ret != 0) {
381 			/* Close all received descriptors. */
382 			serrno = errno;
383 			for (j = 0; j < i; j++)
384 				close(fds[j]);
385 			errno = serrno;
386 			break;
387 		}
388 		i += step;
389 	}
390 
391 	return (ret);
392 }
393 
394 int
fd_send(int sock,const int * fds,size_t nfds)395 fd_send(int sock, const int *fds, size_t nfds)
396 {
397 	unsigned int i, step;
398 	int ret;
399 
400 	if (nfds == 0 || fds == NULL) {
401 		errno = EINVAL;
402 		return (-1);
403 	}
404 
405 	ret = i = step = 0;
406 	while (i < nfds) {
407 		if (PKG_MAX_SIZE < nfds - i)
408 			step = PKG_MAX_SIZE;
409 		else
410 			step = nfds - i;
411 		ret = fd_package_send(sock, fds + i, step);
412 		if (ret != 0)
413 			break;
414 		i += step;
415 	}
416 
417 	return (ret);
418 }
419 
420 int
buf_send(int sock,void * buf,size_t size)421 buf_send(int sock, void *buf, size_t size)
422 {
423 	ssize_t done;
424 	unsigned char *ptr;
425 
426 	PJDLOG_ASSERT(sock >= 0);
427 	PJDLOG_ASSERT(size > 0);
428 	PJDLOG_ASSERT(buf != NULL);
429 
430 	ptr = buf;
431 	do {
432 		fd_wait(sock, false);
433 		done = send(sock, ptr, size, 0);
434 		if (done == -1) {
435 			if (errno == EINTR)
436 				continue;
437 			return (-1);
438 		} else if (done == 0) {
439 			errno = ENOTCONN;
440 			return (-1);
441 		}
442 		size -= done;
443 		ptr += done;
444 	} while (size > 0);
445 
446 	return (0);
447 }
448 
449 int
buf_recv(int sock,void * buf,size_t size,int flags)450 buf_recv(int sock, void *buf, size_t size, int flags)
451 {
452 	ssize_t done;
453 	unsigned char *ptr;
454 
455 	PJDLOG_ASSERT(sock >= 0);
456 	PJDLOG_ASSERT(buf != NULL);
457 
458 	ptr = buf;
459 	while (size > 0) {
460 		fd_wait(sock, true);
461 		done = recv(sock, ptr, size, flags);
462 		if (done == -1) {
463 			if (errno == EINTR)
464 				continue;
465 			return (-1);
466 		} else if (done == 0) {
467 			errno = ENOTCONN;
468 			return (-1);
469 		}
470 		size -= done;
471 		ptr += done;
472 	}
473 
474 	return (0);
475 }
476