xref: /freebsd/usr.sbin/iscsid/iscsid.c (revision 3416500aef140042c64bc149cb1ec6620483bc44)
1 /*-
2  * Copyright (c) 2012 The FreeBSD Foundation
3  * All rights reserved.
4  *
5  * This software was developed by Edward Tomasz Napierala under sponsorship
6  * from the FreeBSD Foundation.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/types.h>
35 #include <sys/time.h>
36 #include <sys/ioctl.h>
37 #include <sys/param.h>
38 #include <sys/linker.h>
39 #include <sys/socket.h>
40 #include <sys/capsicum.h>
41 #include <sys/wait.h>
42 #include <assert.h>
43 #include <errno.h>
44 #include <fcntl.h>
45 #include <libutil.h>
46 #include <netdb.h>
47 #include <signal.h>
48 #include <stdbool.h>
49 #include <stdint.h>
50 #include <stdio.h>
51 #include <stdlib.h>
52 #include <string.h>
53 #include <unistd.h>
54 
55 #include "iscsid.h"
56 
57 static volatile bool sigalrm_received = false;
58 
59 static int nchildren = 0;
60 
61 static void
62 usage(void)
63 {
64 
65 	fprintf(stderr, "usage: iscsid [-P pidfile][-d][-m maxproc][-t timeout]\n");
66 	exit(1);
67 }
68 
69 char *
70 checked_strdup(const char *s)
71 {
72 	char *c;
73 
74 	c = strdup(s);
75 	if (c == NULL)
76 		log_err(1, "strdup");
77 	return (c);
78 }
79 
80 static void
81 resolve_addr(const struct connection *conn, const char *address,
82     struct addrinfo **ai, bool initiator_side)
83 {
84 	struct addrinfo hints;
85 	char *arg, *addr, *ch;
86 	const char *port;
87 	int error, colons = 0;
88 
89 	arg = checked_strdup(address);
90 
91 	if (arg[0] == '\0') {
92 		fail(conn, "empty address");
93 		log_errx(1, "empty address");
94 	}
95 	if (arg[0] == '[') {
96 		/*
97 		 * IPv6 address in square brackets, perhaps with port.
98 		 */
99 		arg++;
100 		addr = strsep(&arg, "]");
101 		if (arg == NULL) {
102 			fail(conn, "malformed address");
103 			log_errx(1, "malformed address %s", address);
104 		}
105 		if (arg[0] == '\0') {
106 			port = NULL;
107 		} else if (arg[0] == ':') {
108 			port = arg + 1;
109 		} else {
110 			fail(conn, "malformed address");
111 			log_errx(1, "malformed address %s", address);
112 		}
113 	} else {
114 		/*
115 		 * Either IPv6 address without brackets - and without
116 		 * a port - or IPv4 address.  Just count the colons.
117 		 */
118 		for (ch = arg; *ch != '\0'; ch++) {
119 			if (*ch == ':')
120 				colons++;
121 		}
122 		if (colons > 1) {
123 			addr = arg;
124 			port = NULL;
125 		} else {
126 			addr = strsep(&arg, ":");
127 			if (arg == NULL)
128 				port = NULL;
129 			else
130 				port = arg;
131 		}
132 	}
133 
134 	if (port == NULL && !initiator_side)
135 		port = "3260";
136 
137 	memset(&hints, 0, sizeof(hints));
138 	hints.ai_family = PF_UNSPEC;
139 	hints.ai_socktype = SOCK_STREAM;
140 	hints.ai_flags = AI_ADDRCONFIG | AI_NUMERICSERV;
141 	if (initiator_side)
142 		hints.ai_flags |= AI_PASSIVE;
143 
144 	error = getaddrinfo(addr, port, &hints, ai);
145 	if (error != 0) {
146 		fail(conn, gai_strerror(error));
147 		log_errx(1, "getaddrinfo for %s failed: %s",
148 		    address, gai_strerror(error));
149 	}
150 }
151 
152 static struct connection *
153 connection_new(int iscsi_fd, const struct iscsi_daemon_request *request)
154 {
155 	struct connection *conn;
156 	struct iscsi_session_limits *isl;
157 	struct addrinfo *from_ai, *to_ai;
158 	const char *from_addr, *to_addr;
159 #ifdef ICL_KERNEL_PROXY
160 	struct iscsi_daemon_connect idc;
161 #endif
162 	int error, sockbuf;
163 
164 	conn = calloc(1, sizeof(*conn));
165 	if (conn == NULL)
166 		log_err(1, "calloc");
167 
168 	/*
169 	 * Default values, from RFC 3720, section 12.
170 	 */
171 	conn->conn_header_digest = CONN_DIGEST_NONE;
172 	conn->conn_data_digest = CONN_DIGEST_NONE;
173 	conn->conn_initial_r2t = true;
174 	conn->conn_immediate_data = true;
175 	conn->conn_max_recv_data_segment_length = 8192;
176 	conn->conn_max_send_data_segment_length = 8192;
177 	conn->conn_max_burst_length = 262144;
178 	conn->conn_first_burst_length = 65536;
179 	conn->conn_iscsi_fd = iscsi_fd;
180 
181 	conn->conn_session_id = request->idr_session_id;
182 	memcpy(&conn->conn_conf, &request->idr_conf, sizeof(conn->conn_conf));
183 	memcpy(&conn->conn_isid, &request->idr_isid, sizeof(conn->conn_isid));
184 	conn->conn_tsih = request->idr_tsih;
185 
186 	/*
187 	 * Read the driver limits and provide reasonable defaults for the ones
188 	 * the driver doesn't care about.  If a max_snd_dsl is not explicitly
189 	 * provided by the driver then we'll make sure both conn->max_snd_dsl
190 	 * and isl->max_snd_dsl are set to the rcv_dsl.  This preserves historic
191 	 * behavior.
192 	 */
193 	isl = &conn->conn_limits;
194 	memcpy(isl, &request->idr_limits, sizeof(*isl));
195 	if (isl->isl_max_recv_data_segment_length == 0)
196 		isl->isl_max_recv_data_segment_length = (1 << 24) - 1;
197 	if (isl->isl_max_send_data_segment_length == 0)
198 		isl->isl_max_send_data_segment_length =
199 		    isl->isl_max_recv_data_segment_length;
200 	if (isl->isl_max_burst_length == 0)
201 		isl->isl_max_burst_length = (1 << 24) - 1;
202 	if (isl->isl_first_burst_length == 0)
203 		isl->isl_first_burst_length = (1 << 24) - 1;
204 	if (isl->isl_first_burst_length > isl->isl_max_burst_length)
205 		isl->isl_first_burst_length = isl->isl_max_burst_length;
206 
207 	/*
208 	 * Limit default send length in case it won't be negotiated.
209 	 * We can't do it for other limits, since they may affect both
210 	 * sender and receiver operation, and we must obey defaults.
211 	 */
212 	if (conn->conn_max_send_data_segment_length >
213 	    isl->isl_max_send_data_segment_length) {
214 		conn->conn_max_send_data_segment_length =
215 		    isl->isl_max_send_data_segment_length;
216 	}
217 
218 	from_addr = conn->conn_conf.isc_initiator_addr;
219 	to_addr = conn->conn_conf.isc_target_addr;
220 
221 	if (from_addr[0] != '\0')
222 		resolve_addr(conn, from_addr, &from_ai, true);
223 	else
224 		from_ai = NULL;
225 
226 	resolve_addr(conn, to_addr, &to_ai, false);
227 
228 #ifdef ICL_KERNEL_PROXY
229 	if (conn->conn_conf.isc_iser) {
230 		memset(&idc, 0, sizeof(idc));
231 		idc.idc_session_id = conn->conn_session_id;
232 		if (conn->conn_conf.isc_iser)
233 			idc.idc_iser = 1;
234 		idc.idc_domain = to_ai->ai_family;
235 		idc.idc_socktype = to_ai->ai_socktype;
236 		idc.idc_protocol = to_ai->ai_protocol;
237 		if (from_ai != NULL) {
238 			idc.idc_from_addr = from_ai->ai_addr;
239 			idc.idc_from_addrlen = from_ai->ai_addrlen;
240 		}
241 		idc.idc_to_addr = to_ai->ai_addr;
242 		idc.idc_to_addrlen = to_ai->ai_addrlen;
243 
244 		log_debugx("connecting to %s using ICL kernel proxy", to_addr);
245 		error = ioctl(iscsi_fd, ISCSIDCONNECT, &idc);
246 		if (error != 0) {
247 			fail(conn, strerror(errno));
248 			log_err(1, "failed to connect to %s "
249 			    "using ICL kernel proxy: ISCSIDCONNECT", to_addr);
250 		}
251 
252 		return (conn);
253 	}
254 #endif /* ICL_KERNEL_PROXY */
255 
256 	if (conn->conn_conf.isc_iser) {
257 		fail(conn, "iSER not supported");
258 		log_errx(1, "iscsid(8) compiled without ICL_KERNEL_PROXY "
259 		    "does not support iSER");
260 	}
261 
262 	conn->conn_socket = socket(to_ai->ai_family, to_ai->ai_socktype,
263 	    to_ai->ai_protocol);
264 	if (conn->conn_socket < 0) {
265 		fail(conn, strerror(errno));
266 		log_err(1, "failed to create socket for %s", from_addr);
267 	}
268 	sockbuf = SOCKBUF_SIZE;
269 	if (setsockopt(conn->conn_socket, SOL_SOCKET, SO_RCVBUF,
270 	    &sockbuf, sizeof(sockbuf)) == -1)
271 		log_warn("setsockopt(SO_RCVBUF) failed");
272 	sockbuf = SOCKBUF_SIZE;
273 	if (setsockopt(conn->conn_socket, SOL_SOCKET, SO_SNDBUF,
274 	    &sockbuf, sizeof(sockbuf)) == -1)
275 		log_warn("setsockopt(SO_SNDBUF) failed");
276 	if (from_ai != NULL) {
277 		error = bind(conn->conn_socket, from_ai->ai_addr,
278 		    from_ai->ai_addrlen);
279 		if (error != 0) {
280 			fail(conn, strerror(errno));
281 			log_err(1, "failed to bind to %s", from_addr);
282 		}
283 	}
284 	log_debugx("connecting to %s", to_addr);
285 	error = connect(conn->conn_socket, to_ai->ai_addr, to_ai->ai_addrlen);
286 	if (error != 0) {
287 		fail(conn, strerror(errno));
288 		log_err(1, "failed to connect to %s", to_addr);
289 	}
290 
291 	return (conn);
292 }
293 
294 static void
295 handoff(struct connection *conn)
296 {
297 	struct iscsi_daemon_handoff idh;
298 	int error;
299 
300 	log_debugx("handing off connection to the kernel");
301 
302 	memset(&idh, 0, sizeof(idh));
303 	idh.idh_session_id = conn->conn_session_id;
304 	idh.idh_socket = conn->conn_socket;
305 	strlcpy(idh.idh_target_alias, conn->conn_target_alias,
306 	    sizeof(idh.idh_target_alias));
307 	idh.idh_tsih = conn->conn_tsih;
308 	idh.idh_statsn = conn->conn_statsn;
309 	idh.idh_header_digest = conn->conn_header_digest;
310 	idh.idh_data_digest = conn->conn_data_digest;
311 	idh.idh_initial_r2t = conn->conn_initial_r2t;
312 	idh.idh_immediate_data = conn->conn_immediate_data;
313 	idh.idh_max_recv_data_segment_length =
314 	    conn->conn_max_recv_data_segment_length;
315 	idh.idh_max_send_data_segment_length =
316 	    conn->conn_max_send_data_segment_length;
317 	idh.idh_max_burst_length = conn->conn_max_burst_length;
318 	idh.idh_first_burst_length = conn->conn_first_burst_length;
319 
320 	error = ioctl(conn->conn_iscsi_fd, ISCSIDHANDOFF, &idh);
321 	if (error != 0)
322 		log_err(1, "ISCSIDHANDOFF");
323 }
324 
325 void
326 fail(const struct connection *conn, const char *reason)
327 {
328 	struct iscsi_daemon_fail idf;
329 	int error, saved_errno;
330 
331 	saved_errno = errno;
332 
333 	memset(&idf, 0, sizeof(idf));
334 	idf.idf_session_id = conn->conn_session_id;
335 	strlcpy(idf.idf_reason, reason, sizeof(idf.idf_reason));
336 
337 	error = ioctl(conn->conn_iscsi_fd, ISCSIDFAIL, &idf);
338 	if (error != 0)
339 		log_err(1, "ISCSIDFAIL");
340 
341 	errno = saved_errno;
342 }
343 
344 /*
345  * XXX: I CANT INTO LATIN
346  */
347 static void
348 capsicate(struct connection *conn)
349 {
350 	int error;
351 	cap_rights_t rights;
352 #ifdef ICL_KERNEL_PROXY
353 	const unsigned long cmds[] = { ISCSIDCONNECT, ISCSIDSEND, ISCSIDRECEIVE,
354 	    ISCSIDHANDOFF, ISCSIDFAIL, ISCSISADD, ISCSISREMOVE, ISCSISMODIFY };
355 #else
356 	const unsigned long cmds[] = { ISCSIDHANDOFF, ISCSIDFAIL, ISCSISADD,
357 	    ISCSISREMOVE, ISCSISMODIFY };
358 #endif
359 
360 	cap_rights_init(&rights, CAP_IOCTL);
361 	error = cap_rights_limit(conn->conn_iscsi_fd, &rights);
362 	if (error != 0 && errno != ENOSYS)
363 		log_err(1, "cap_rights_limit");
364 
365 	error = cap_ioctls_limit(conn->conn_iscsi_fd, cmds, nitems(cmds));
366 
367 	if (error != 0 && errno != ENOSYS)
368 		log_err(1, "cap_ioctls_limit");
369 
370 	error = cap_enter();
371 	if (error != 0 && errno != ENOSYS)
372 		log_err(1, "cap_enter");
373 
374 	if (cap_sandboxed())
375 		log_debugx("Capsicum capability mode enabled");
376 	else
377 		log_warnx("Capsicum capability mode not supported");
378 }
379 
380 bool
381 timed_out(void)
382 {
383 
384 	return (sigalrm_received);
385 }
386 
387 static void
388 sigalrm_handler(int dummy __unused)
389 {
390 	/*
391 	 * It would be easiest to just log an error and exit.  We can't
392 	 * do this, though, because log_errx() is not signal safe, since
393 	 * it calls syslog(3).  Instead, set a flag checked by pdu_send()
394 	 * and pdu_receive(), to call log_errx() there.  Should they fail
395 	 * to notice, we'll exit here one second later.
396 	 */
397 	if (sigalrm_received) {
398 		/*
399 		 * Oh well.  Just give up and quit.
400 		 */
401 		_exit(2);
402 	}
403 
404 	sigalrm_received = true;
405 }
406 
407 static void
408 set_timeout(int timeout)
409 {
410 	struct sigaction sa;
411 	struct itimerval itv;
412 	int error;
413 
414 	if (timeout <= 0) {
415 		log_debugx("session timeout disabled");
416 		return;
417 	}
418 
419 	bzero(&sa, sizeof(sa));
420 	sa.sa_handler = sigalrm_handler;
421 	sigfillset(&sa.sa_mask);
422 	error = sigaction(SIGALRM, &sa, NULL);
423 	if (error != 0)
424 		log_err(1, "sigaction");
425 
426 	/*
427 	 * First SIGALRM will arive after conf_timeout seconds.
428 	 * If we do nothing, another one will arrive a second later.
429 	 */
430 	bzero(&itv, sizeof(itv));
431 	itv.it_interval.tv_sec = 1;
432 	itv.it_value.tv_sec = timeout;
433 
434 	log_debugx("setting session timeout to %d seconds",
435 	    timeout);
436 	error = setitimer(ITIMER_REAL, &itv, NULL);
437 	if (error != 0)
438 		log_err(1, "setitimer");
439 }
440 
441 static void
442 sigchld_handler(int dummy __unused)
443 {
444 
445 	/*
446 	 * The only purpose of this handler is to make SIGCHLD
447 	 * interrupt the ISCSIDWAIT ioctl(2), so we can call
448 	 * wait_for_children().
449 	 */
450 }
451 
452 static void
453 register_sigchld(void)
454 {
455 	struct sigaction sa;
456 	int error;
457 
458 	bzero(&sa, sizeof(sa));
459 	sa.sa_handler = sigchld_handler;
460 	sigfillset(&sa.sa_mask);
461 	error = sigaction(SIGCHLD, &sa, NULL);
462 	if (error != 0)
463 		log_err(1, "sigaction");
464 
465 }
466 
467 static void
468 handle_request(int iscsi_fd, const struct iscsi_daemon_request *request, int timeout)
469 {
470 	struct connection *conn;
471 
472 	log_set_peer_addr(request->idr_conf.isc_target_addr);
473 	if (request->idr_conf.isc_target[0] != '\0') {
474 		log_set_peer_name(request->idr_conf.isc_target);
475 		setproctitle("%s (%s)", request->idr_conf.isc_target_addr, request->idr_conf.isc_target);
476 	} else {
477 		setproctitle("%s", request->idr_conf.isc_target_addr);
478 	}
479 
480 	conn = connection_new(iscsi_fd, request);
481 	set_timeout(timeout);
482 	capsicate(conn);
483 	login(conn);
484 	if (conn->conn_conf.isc_discovery != 0)
485 		discovery(conn);
486 	else
487 		handoff(conn);
488 
489 	log_debugx("nothing more to do; exiting");
490 	exit (0);
491 }
492 
493 static int
494 wait_for_children(bool block)
495 {
496 	pid_t pid;
497 	int status;
498 	int num = 0;
499 
500 	for (;;) {
501 		/*
502 		 * If "block" is true, wait for at least one process.
503 		 */
504 		if (block && num == 0)
505 			pid = wait4(-1, &status, 0, NULL);
506 		else
507 			pid = wait4(-1, &status, WNOHANG, NULL);
508 		if (pid <= 0)
509 			break;
510 		if (WIFSIGNALED(status)) {
511 			log_warnx("child process %d terminated with signal %d",
512 			    pid, WTERMSIG(status));
513 		} else if (WEXITSTATUS(status) != 0) {
514 			log_warnx("child process %d terminated with exit status %d",
515 			    pid, WEXITSTATUS(status));
516 		} else {
517 			log_debugx("child process %d terminated gracefully", pid);
518 		}
519 		num++;
520 	}
521 
522 	return (num);
523 }
524 
525 int
526 main(int argc, char **argv)
527 {
528 	int ch, debug = 0, error, iscsi_fd, maxproc = 30, retval, saved_errno,
529 	    timeout = 60;
530 	bool dont_daemonize = false;
531 	struct pidfh *pidfh;
532 	pid_t pid, otherpid;
533 	const char *pidfile_path = DEFAULT_PIDFILE;
534 	struct iscsi_daemon_request request;
535 
536 	while ((ch = getopt(argc, argv, "P:dl:m:t:")) != -1) {
537 		switch (ch) {
538 		case 'P':
539 			pidfile_path = optarg;
540 			break;
541 		case 'd':
542 			dont_daemonize = true;
543 			debug++;
544 			break;
545 		case 'l':
546 			debug = atoi(optarg);
547 			break;
548 		case 'm':
549 			maxproc = atoi(optarg);
550 			break;
551 		case 't':
552 			timeout = atoi(optarg);
553 			break;
554 		case '?':
555 		default:
556 			usage();
557 		}
558 	}
559 	argc -= optind;
560 	if (argc != 0)
561 		usage();
562 
563 	log_init(debug);
564 
565 	pidfh = pidfile_open(pidfile_path, 0600, &otherpid);
566 	if (pidfh == NULL) {
567 		if (errno == EEXIST)
568 			log_errx(1, "daemon already running, pid: %jd.",
569 			    (intmax_t)otherpid);
570 		log_err(1, "cannot open or create pidfile \"%s\"",
571 		    pidfile_path);
572 	}
573 
574 	iscsi_fd = open(ISCSI_PATH, O_RDWR);
575 	if (iscsi_fd < 0 && errno == ENOENT) {
576 		saved_errno = errno;
577 		retval = kldload("iscsi");
578 		if (retval != -1)
579 			iscsi_fd = open(ISCSI_PATH, O_RDWR);
580 		else
581 			errno = saved_errno;
582 	}
583 	if (iscsi_fd < 0)
584 		log_err(1, "failed to open %s", ISCSI_PATH);
585 
586 	if (dont_daemonize == false) {
587 		if (daemon(0, 0) == -1) {
588 			log_warn("cannot daemonize");
589 			pidfile_remove(pidfh);
590 			exit(1);
591 		}
592 	}
593 
594 	pidfile_write(pidfh);
595 
596 	register_sigchld();
597 
598 	for (;;) {
599 		log_debugx("waiting for request from the kernel");
600 
601 		memset(&request, 0, sizeof(request));
602 		error = ioctl(iscsi_fd, ISCSIDWAIT, &request);
603 		if (error != 0) {
604 			if (errno == EINTR) {
605 				nchildren -= wait_for_children(false);
606 				assert(nchildren >= 0);
607 				continue;
608 			}
609 
610 			log_err(1, "ISCSIDWAIT");
611 		}
612 
613 		if (dont_daemonize) {
614 			log_debugx("not forking due to -d flag; "
615 			    "will exit after servicing a single request");
616 		} else {
617 			nchildren -= wait_for_children(false);
618 			assert(nchildren >= 0);
619 
620 			while (maxproc > 0 && nchildren >= maxproc) {
621 				log_debugx("maxproc limit of %d child processes hit; "
622 				    "waiting for child process to exit", maxproc);
623 				nchildren -= wait_for_children(true);
624 				assert(nchildren >= 0);
625 			}
626 			log_debugx("incoming connection; forking child process #%d",
627 			    nchildren);
628 			nchildren++;
629 
630 			pid = fork();
631 			if (pid < 0)
632 				log_err(1, "fork");
633 			if (pid > 0)
634 				continue;
635 		}
636 
637 		pidfile_close(pidfh);
638 		handle_request(iscsi_fd, &request, timeout);
639 	}
640 
641 	return (0);
642 }
643