xref: /freebsd/usr.sbin/iscsid/iscsid.c (revision d0b2dbfa0ecf2bbc9709efc5e20baf8e4b44bbbf)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2012 The FreeBSD Foundation
5  *
6  * This software was developed by Edward Tomasz Napierala under sponsorship
7  * from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  */
31 
32 #include <sys/cdefs.h>
33 #include <sys/types.h>
34 #include <sys/time.h>
35 #include <sys/ioctl.h>
36 #include <sys/param.h>
37 #include <sys/linker.h>
38 #include <sys/socket.h>
39 #include <sys/sysctl.h>
40 #include <sys/capsicum.h>
41 #include <sys/wait.h>
42 #include <netinet/in.h>
43 #include <netinet/tcp.h>
44 #include <assert.h>
45 #include <capsicum_helpers.h>
46 #include <errno.h>
47 #include <fcntl.h>
48 #include <libutil.h>
49 #include <netdb.h>
50 #include <signal.h>
51 #include <stdbool.h>
52 #include <stdint.h>
53 #include <stdio.h>
54 #include <stdlib.h>
55 #include <string.h>
56 #include <unistd.h>
57 
58 #include "iscsid.h"
59 
60 static bool	timed_out(void);
61 #ifdef ICL_KERNEL_PROXY
62 static void	pdu_receive_proxy(struct pdu *pdu);
63 static void	pdu_send_proxy(struct pdu *pdu);
64 #endif /* ICL_KERNEL_PROXY */
65 
66 static volatile bool sigalrm_received = false;
67 
68 static int nchildren = 0;
69 
70 static struct connection_ops conn_ops = {
71 	.timed_out = timed_out,
72 #ifdef ICL_KERNEL_PROXY
73 	.pdu_receive_proxy = pdu_receive_proxy,
74 	.pdu_send_proxy = pdu_send_proxy,
75 #endif
76 	.fail = fail,
77 };
78 
79 static void
80 usage(void)
81 {
82 
83 	fprintf(stderr, "usage: iscsid [-P pidfile][-d][-m maxproc][-t timeout]\n");
84 	exit(1);
85 }
86 
87 #ifdef ICL_KERNEL_PROXY
88 
89 static void
90 pdu_receive_proxy(struct pdu *pdu)
91 {
92 	struct iscsid_connection *conn;
93 	struct iscsi_daemon_receive idr;
94 	size_t len;
95 	int error;
96 
97 	conn = (struct iscsid_connection *)pdu->pdu_connection;
98 	assert(conn->conn_conf.isc_iser != 0);
99 
100 	pdu->pdu_data = malloc(conn->conn.conn_max_recv_data_segment_length);
101 	if (pdu->pdu_data == NULL)
102 		log_err(1, "malloc");
103 
104 	memset(&idr, 0, sizeof(idr));
105 	idr.idr_session_id = conn->conn_session_id;
106 	idr.idr_bhs = pdu->pdu_bhs;
107 	idr.idr_data_segment_len = conn->conn.conn_max_recv_data_segment_length;
108 	idr.idr_data_segment = pdu->pdu_data;
109 
110 	error = ioctl(conn->conn_iscsi_fd, ISCSIDRECEIVE, &idr);
111 	if (error != 0)
112 		log_err(1, "ISCSIDRECEIVE");
113 
114 	len = pdu_ahs_length(pdu);
115 	if (len > 0)
116 		log_errx(1, "protocol error: non-empty AHS");
117 
118 	len = pdu_data_segment_length(pdu);
119 	assert(len <= (size_t)conn->conn.conn_max_recv_data_segment_length);
120 	pdu->pdu_data_len = len;
121 }
122 
123 static void
124 pdu_send_proxy(struct pdu *pdu)
125 {
126 	struct iscsid_connection *conn;
127 	struct iscsi_daemon_send ids;
128 	int error;
129 
130 	conn = (struct iscsid_connection *)pdu->pdu_connection;
131 	assert(conn->conn_conf.isc_iser != 0);
132 
133 	pdu_set_data_segment_length(pdu, pdu->pdu_data_len);
134 
135 	memset(&ids, 0, sizeof(ids));
136 	ids.ids_session_id = conn->conn_session_id;
137 	ids.ids_bhs = pdu->pdu_bhs;
138 	ids.ids_data_segment_len = pdu->pdu_data_len;
139 	ids.ids_data_segment = pdu->pdu_data;
140 
141 	error = ioctl(conn->conn_iscsi_fd, ISCSIDSEND, &ids);
142 	if (error != 0)
143 		log_err(1, "ISCSIDSEND");
144 }
145 
146 #endif /* ICL_KERNEL_PROXY */
147 
148 static void
149 resolve_addr(const struct connection *conn, const char *address,
150     struct addrinfo **ai, bool initiator_side)
151 {
152 	struct addrinfo hints;
153 	char *arg, *addr, *ch, *tofree;
154 	const char *port;
155 	int error, colons = 0;
156 
157 	tofree = arg = checked_strdup(address);
158 
159 	if (arg[0] == '\0') {
160 		fail(conn, "empty address");
161 		log_errx(1, "empty address");
162 	}
163 	if (arg[0] == '[') {
164 		/*
165 		 * IPv6 address in square brackets, perhaps with port.
166 		 */
167 		arg++;
168 		addr = strsep(&arg, "]");
169 		if (arg == NULL) {
170 			fail(conn, "malformed address");
171 			log_errx(1, "malformed address %s", address);
172 		}
173 		if (arg[0] == '\0') {
174 			port = NULL;
175 		} else if (arg[0] == ':') {
176 			port = arg + 1;
177 		} else {
178 			fail(conn, "malformed address");
179 			log_errx(1, "malformed address %s", address);
180 		}
181 	} else {
182 		/*
183 		 * Either IPv6 address without brackets - and without
184 		 * a port - or IPv4 address.  Just count the colons.
185 		 */
186 		for (ch = arg; *ch != '\0'; ch++) {
187 			if (*ch == ':')
188 				colons++;
189 		}
190 		if (colons > 1) {
191 			addr = arg;
192 			port = NULL;
193 		} else {
194 			addr = strsep(&arg, ":");
195 			if (arg == NULL)
196 				port = NULL;
197 			else
198 				port = arg;
199 		}
200 	}
201 
202 	if (port == NULL && !initiator_side)
203 		port = "3260";
204 
205 	memset(&hints, 0, sizeof(hints));
206 	hints.ai_family = PF_UNSPEC;
207 	hints.ai_socktype = SOCK_STREAM;
208 	hints.ai_flags = AI_ADDRCONFIG | AI_NUMERICSERV;
209 	if (initiator_side)
210 		hints.ai_flags |= AI_PASSIVE;
211 
212 	error = getaddrinfo(addr, port, &hints, ai);
213 	if (error != 0) {
214 		fail(conn, gai_strerror(error));
215 		log_errx(1, "getaddrinfo for %s failed: %s",
216 		    address, gai_strerror(error));
217 	}
218 
219 	free(tofree);
220 }
221 
222 static struct iscsid_connection *
223 connection_new(int iscsi_fd, const struct iscsi_daemon_request *request)
224 {
225 	struct iscsid_connection *conn;
226 	struct addrinfo *from_ai, *to_ai;
227 	const char *from_addr, *to_addr;
228 #ifdef ICL_KERNEL_PROXY
229 	struct iscsi_daemon_connect idc;
230 #endif
231 	int error, optval;
232 
233 	conn = calloc(1, sizeof(*conn));
234 	if (conn == NULL)
235 		log_err(1, "calloc");
236 
237 	connection_init(&conn->conn, &conn_ops,
238 	    request->idr_conf.isc_iser != 0);
239 	conn->conn_protocol_level = 0;
240 	conn->conn_initial_r2t = true;
241 	conn->conn_iscsi_fd = iscsi_fd;
242 
243 	conn->conn_session_id = request->idr_session_id;
244 	memcpy(&conn->conn_conf, &request->idr_conf, sizeof(conn->conn_conf));
245 	memcpy(&conn->conn.conn_isid, &request->idr_isid,
246 	    sizeof(conn->conn.conn_isid));
247 	conn->conn.conn_tsih = request->idr_tsih;
248 
249 	from_addr = conn->conn_conf.isc_initiator_addr;
250 	to_addr = conn->conn_conf.isc_target_addr;
251 
252 	if (from_addr[0] != '\0')
253 		resolve_addr(&conn->conn, from_addr, &from_ai, true);
254 	else
255 		from_ai = NULL;
256 
257 	resolve_addr(&conn->conn, to_addr, &to_ai, false);
258 
259 #ifdef ICL_KERNEL_PROXY
260 	if (conn->conn_conf.isc_iser) {
261 		memset(&idc, 0, sizeof(idc));
262 		idc.idc_session_id = conn->conn_session_id;
263 		if (conn->conn_conf.isc_iser)
264 			idc.idc_iser = 1;
265 		idc.idc_domain = to_ai->ai_family;
266 		idc.idc_socktype = to_ai->ai_socktype;
267 		idc.idc_protocol = to_ai->ai_protocol;
268 		if (from_ai != NULL) {
269 			idc.idc_from_addr = from_ai->ai_addr;
270 			idc.idc_from_addrlen = from_ai->ai_addrlen;
271 		}
272 		idc.idc_to_addr = to_ai->ai_addr;
273 		idc.idc_to_addrlen = to_ai->ai_addrlen;
274 
275 		log_debugx("connecting to %s using ICL kernel proxy", to_addr);
276 		error = ioctl(iscsi_fd, ISCSIDCONNECT, &idc);
277 		if (error != 0) {
278 			fail(&conn->conn, strerror(errno));
279 			log_err(1, "failed to connect to %s "
280 			    "using ICL kernel proxy: ISCSIDCONNECT", to_addr);
281 		}
282 
283 		if (from_ai != NULL)
284 			freeaddrinfo(from_ai);
285 		freeaddrinfo(to_ai);
286 
287 		return (conn);
288 	}
289 #endif /* ICL_KERNEL_PROXY */
290 
291 	if (conn->conn_conf.isc_iser) {
292 		fail(&conn->conn, "iSER not supported");
293 		log_errx(1, "iscsid(8) compiled without ICL_KERNEL_PROXY "
294 		    "does not support iSER");
295 	}
296 
297 	conn->conn.conn_socket = socket(to_ai->ai_family, to_ai->ai_socktype,
298 	    to_ai->ai_protocol);
299 	if (conn->conn.conn_socket < 0) {
300 		fail(&conn->conn, strerror(errno));
301 		log_err(1, "failed to create socket for %s", from_addr);
302 	}
303 	optval = SOCKBUF_SIZE;
304 	if (setsockopt(conn->conn.conn_socket, SOL_SOCKET, SO_RCVBUF,
305 	    &optval, sizeof(optval)) == -1)
306 		log_warn("setsockopt(SO_RCVBUF) failed");
307 	optval = SOCKBUF_SIZE;
308 	if (setsockopt(conn->conn.conn_socket, SOL_SOCKET, SO_SNDBUF,
309 	    &optval, sizeof(optval)) == -1)
310 		log_warn("setsockopt(SO_SNDBUF) failed");
311 	optval = 1;
312 	if (setsockopt(conn->conn.conn_socket, SOL_SOCKET, SO_NO_DDP,
313 	    &optval, sizeof(optval)) == -1)
314 		log_warn("setsockopt(SO_NO_DDP) failed");
315 	if (conn->conn_conf.isc_dscp != -1) {
316 		int tos = conn->conn_conf.isc_dscp << 2;
317 		if (to_ai->ai_family == AF_INET) {
318 			if (setsockopt(conn->conn.conn_socket,
319 			    IPPROTO_IP, IP_TOS,
320 			    &tos, sizeof(tos)) == -1)
321 				log_warn("setsockopt(IP_TOS) "
322 				    "failed for %s",
323 				    from_addr);
324 		} else
325 		if (to_ai->ai_family == AF_INET6) {
326 			if (setsockopt(conn->conn.conn_socket,
327 			    IPPROTO_IPV6, IPV6_TCLASS,
328 			    &tos, sizeof(tos)) == -1)
329 				log_warn("setsockopt(IPV6_TCLASS) "
330 				    "failed for %s",
331 				    from_addr);
332 		}
333 	}
334 	if (conn->conn_conf.isc_pcp != -1) {
335 		int pcp = conn->conn_conf.isc_pcp;
336 		if (to_ai->ai_family == AF_INET) {
337 			if (setsockopt(conn->conn.conn_socket,
338 			    IPPROTO_IP, IP_VLAN_PCP,
339 			    &pcp, sizeof(pcp)) == -1)
340 				log_warn("setsockopt(IP_VLAN_PCP) "
341 				    "failed for %s",
342 				    from_addr);
343 		} else
344 		if (to_ai->ai_family == AF_INET6) {
345 			if (setsockopt(conn->conn.conn_socket,
346 			    IPPROTO_IPV6, IPV6_VLAN_PCP,
347 			    &pcp, sizeof(pcp)) == -1)
348 				log_warn("setsockopt(IPV6_VLAN_PCP) "
349 				    "failed for %s",
350 				    from_addr);
351 		}
352 	}
353 	/*
354 	 * Reduce TCP SYN_SENT timeout while
355 	 * no connectivity exists, to allow
356 	 * rapid reuse of the available slots.
357 	 */
358 	int keepinit = 0;
359 	if (conn->conn_conf.isc_login_timeout > 0) {
360 		keepinit = conn->conn_conf.isc_login_timeout;
361 		log_debugx("session specific LoginTimeout at %d sec",
362 			keepinit);
363 	}
364 	if (conn->conn_conf.isc_login_timeout == -1) {
365 		int value;
366 		size_t size = sizeof(value);
367 		if (sysctlbyname("kern.iscsi.login_timeout",
368 		    &value, &size, NULL, 0) == 0) {
369 			keepinit = value;
370 			log_debugx("global login_timeout at %d sec",
371 				keepinit);
372 		}
373 	}
374 	if (keepinit > 0) {
375 		if (setsockopt(conn->conn.conn_socket,
376 		    IPPROTO_TCP, TCP_KEEPINIT,
377 		    &keepinit, sizeof(keepinit)) == -1)
378 			log_warnx("setsockopt(TCP_KEEPINIT) "
379 			    "failed for %s", to_addr);
380 	}
381 	if (from_ai != NULL) {
382 		error = bind(conn->conn.conn_socket, from_ai->ai_addr,
383 		    from_ai->ai_addrlen);
384 		if (error != 0) {
385 			fail(&conn->conn, strerror(errno));
386 			log_err(1, "failed to bind to %s", from_addr);
387 		}
388 	}
389 	log_debugx("connecting to %s", to_addr);
390 	error = connect(conn->conn.conn_socket, to_ai->ai_addr,
391 	    to_ai->ai_addrlen);
392 	if (error != 0) {
393 		fail(&conn->conn, strerror(errno));
394 		log_err(1, "failed to connect to %s", to_addr);
395 	}
396 
397 	if (from_ai != NULL)
398 		freeaddrinfo(from_ai);
399 	freeaddrinfo(to_ai);
400 
401 	return (conn);
402 }
403 
404 static void
405 limits(struct iscsid_connection *conn)
406 {
407 	struct iscsi_daemon_limits idl;
408 	struct iscsi_session_limits *isl;
409 	int error;
410 
411 	log_debugx("fetching limits from the kernel");
412 
413 	memset(&idl, 0, sizeof(idl));
414 	idl.idl_session_id = conn->conn_session_id;
415 	idl.idl_socket = conn->conn.conn_socket;
416 
417 	error = ioctl(conn->conn_iscsi_fd, ISCSIDLIMITS, &idl);
418 	if (error != 0)
419 		log_err(1, "ISCSIDLIMITS");
420 
421 	/*
422 	 * Read the driver limits and provide reasonable defaults for the ones
423 	 * the driver doesn't care about.  If a max_snd_dsl is not explicitly
424 	 * provided by the driver then we'll make sure both conn->max_snd_dsl
425 	 * and isl->max_snd_dsl are set to the rcv_dsl.  This preserves historic
426 	 * behavior.
427 	 */
428 	isl = &conn->conn_limits;
429 	memcpy(isl, &idl.idl_limits, sizeof(*isl));
430 	if (isl->isl_max_recv_data_segment_length == 0)
431 		isl->isl_max_recv_data_segment_length = (1 << 24) - 1;
432 	if (isl->isl_max_send_data_segment_length == 0)
433 		isl->isl_max_send_data_segment_length =
434 		    isl->isl_max_recv_data_segment_length;
435 	if (isl->isl_max_burst_length == 0)
436 		isl->isl_max_burst_length = (1 << 24) - 1;
437 	if (isl->isl_first_burst_length == 0)
438 		isl->isl_first_burst_length = (1 << 24) - 1;
439 	if (isl->isl_first_burst_length > isl->isl_max_burst_length)
440 		isl->isl_first_burst_length = isl->isl_max_burst_length;
441 
442 	/*
443 	 * Limit default send length in case it won't be negotiated.
444 	 * We can't do it for other limits, since they may affect both
445 	 * sender and receiver operation, and we must obey defaults.
446 	 */
447 	if (conn->conn.conn_max_send_data_segment_length >
448 	    isl->isl_max_send_data_segment_length) {
449 		conn->conn.conn_max_send_data_segment_length =
450 		    isl->isl_max_send_data_segment_length;
451 	}
452 }
453 
454 static void
455 handoff(struct iscsid_connection *conn)
456 {
457 	struct iscsi_daemon_handoff idh;
458 	int error;
459 
460 	log_debugx("handing off connection to the kernel");
461 
462 	memset(&idh, 0, sizeof(idh));
463 	idh.idh_session_id = conn->conn_session_id;
464 	idh.idh_socket = conn->conn.conn_socket;
465 	strlcpy(idh.idh_target_alias, conn->conn_target_alias,
466 	    sizeof(idh.idh_target_alias));
467 	idh.idh_tsih = conn->conn.conn_tsih;
468 	idh.idh_statsn = conn->conn.conn_statsn;
469 	idh.idh_protocol_level = conn->conn_protocol_level;
470 	idh.idh_header_digest = conn->conn.conn_header_digest;
471 	idh.idh_data_digest = conn->conn.conn_data_digest;
472 	idh.idh_initial_r2t = conn->conn_initial_r2t;
473 	idh.idh_immediate_data = conn->conn.conn_immediate_data;
474 	idh.idh_max_recv_data_segment_length =
475 	    conn->conn.conn_max_recv_data_segment_length;
476 	idh.idh_max_send_data_segment_length =
477 	    conn->conn.conn_max_send_data_segment_length;
478 	idh.idh_max_burst_length = conn->conn.conn_max_burst_length;
479 	idh.idh_first_burst_length = conn->conn.conn_first_burst_length;
480 
481 	error = ioctl(conn->conn_iscsi_fd, ISCSIDHANDOFF, &idh);
482 	if (error != 0)
483 		log_err(1, "ISCSIDHANDOFF");
484 }
485 
486 void
487 fail(const struct connection *base_conn, const char *reason)
488 {
489 	const struct iscsid_connection *conn;
490 	struct iscsi_daemon_fail idf;
491 	int error, saved_errno;
492 
493 	conn = (const struct iscsid_connection *)base_conn;
494 	saved_errno = errno;
495 
496 	memset(&idf, 0, sizeof(idf));
497 	idf.idf_session_id = conn->conn_session_id;
498 	strlcpy(idf.idf_reason, reason, sizeof(idf.idf_reason));
499 
500 	error = ioctl(conn->conn_iscsi_fd, ISCSIDFAIL, &idf);
501 	if (error != 0)
502 		log_err(1, "ISCSIDFAIL");
503 
504 	errno = saved_errno;
505 }
506 
507 /*
508  * XXX: I CANT INTO LATIN
509  */
510 static void
511 capsicate(struct iscsid_connection *conn)
512 {
513 	cap_rights_t rights;
514 	const unsigned long cmds[] = {
515 #ifdef ICL_KERNEL_PROXY
516 		ISCSIDCONNECT,
517 		ISCSIDSEND,
518 		ISCSIDRECEIVE,
519 #endif
520 		ISCSIDLIMITS,
521 		ISCSIDHANDOFF,
522 		ISCSIDFAIL,
523 		ISCSISADD,
524 		ISCSISREMOVE,
525 		ISCSISMODIFY
526 	};
527 
528 	cap_rights_init(&rights, CAP_IOCTL);
529 	if (caph_rights_limit(conn->conn_iscsi_fd, &rights) < 0)
530 		log_err(1, "cap_rights_limit");
531 
532 	if (caph_ioctls_limit(conn->conn_iscsi_fd, cmds, nitems(cmds)) < 0)
533 		log_err(1, "cap_ioctls_limit");
534 
535 	if (caph_enter() != 0)
536 		log_err(1, "cap_enter");
537 
538 	if (cap_sandboxed())
539 		log_debugx("Capsicum capability mode enabled");
540 	else
541 		log_warnx("Capsicum capability mode not supported");
542 }
543 
544 static bool
545 timed_out(void)
546 {
547 
548 	return (sigalrm_received);
549 }
550 
551 static void
552 sigalrm_handler(int dummy __unused)
553 {
554 	/*
555 	 * It would be easiest to just log an error and exit.  We can't
556 	 * do this, though, because log_errx() is not signal safe, since
557 	 * it calls syslog(3).  Instead, set a flag checked by pdu_send()
558 	 * and pdu_receive(), to call log_errx() there.  Should they fail
559 	 * to notice, we'll exit here one second later.
560 	 */
561 	if (sigalrm_received) {
562 		/*
563 		 * Oh well.  Just give up and quit.
564 		 */
565 		_exit(2);
566 	}
567 
568 	sigalrm_received = true;
569 }
570 
571 static void
572 set_timeout(int timeout)
573 {
574 	struct sigaction sa;
575 	struct itimerval itv;
576 	int error;
577 
578 	if (timeout <= 0) {
579 		log_debugx("session timeout disabled");
580 		return;
581 	}
582 
583 	bzero(&sa, sizeof(sa));
584 	sa.sa_handler = sigalrm_handler;
585 	sigfillset(&sa.sa_mask);
586 	error = sigaction(SIGALRM, &sa, NULL);
587 	if (error != 0)
588 		log_err(1, "sigaction");
589 
590 	/*
591 	 * First SIGALRM will arive after conf_timeout seconds.
592 	 * If we do nothing, another one will arrive a second later.
593 	 */
594 	bzero(&itv, sizeof(itv));
595 	itv.it_interval.tv_sec = 1;
596 	itv.it_value.tv_sec = timeout;
597 
598 	log_debugx("setting session timeout to %d seconds",
599 	    timeout);
600 	error = setitimer(ITIMER_REAL, &itv, NULL);
601 	if (error != 0)
602 		log_err(1, "setitimer");
603 }
604 
605 static void
606 sigchld_handler(int dummy __unused)
607 {
608 
609 	/*
610 	 * The only purpose of this handler is to make SIGCHLD
611 	 * interrupt the ISCSIDWAIT ioctl(2), so we can call
612 	 * wait_for_children().
613 	 */
614 }
615 
616 static void
617 register_sigchld(void)
618 {
619 	struct sigaction sa;
620 	int error;
621 
622 	bzero(&sa, sizeof(sa));
623 	sa.sa_handler = sigchld_handler;
624 	sigfillset(&sa.sa_mask);
625 	error = sigaction(SIGCHLD, &sa, NULL);
626 	if (error != 0)
627 		log_err(1, "sigaction");
628 
629 }
630 
631 static void
632 handle_request(int iscsi_fd, const struct iscsi_daemon_request *request, int timeout)
633 {
634 	struct iscsid_connection *conn;
635 
636 	log_set_peer_addr(request->idr_conf.isc_target_addr);
637 	if (request->idr_conf.isc_target[0] != '\0') {
638 		log_set_peer_name(request->idr_conf.isc_target);
639 		setproctitle("%s (%s)", request->idr_conf.isc_target_addr, request->idr_conf.isc_target);
640 	} else {
641 		setproctitle("%s", request->idr_conf.isc_target_addr);
642 	}
643 
644 	conn = connection_new(iscsi_fd, request);
645 	capsicate(conn);
646 	limits(conn);
647 	set_timeout(timeout);
648 	login(conn);
649 	if (conn->conn_conf.isc_discovery != 0)
650 		discovery(conn);
651 	else
652 		handoff(conn);
653 
654 	log_debugx("nothing more to do; exiting");
655 	exit (0);
656 }
657 
658 static int
659 wait_for_children(bool block)
660 {
661 	pid_t pid;
662 	int status;
663 	int num = 0;
664 
665 	for (;;) {
666 		/*
667 		 * If "block" is true, wait for at least one process.
668 		 */
669 		if (block && num == 0)
670 			pid = wait4(-1, &status, 0, NULL);
671 		else
672 			pid = wait4(-1, &status, WNOHANG, NULL);
673 		if (pid <= 0)
674 			break;
675 		if (WIFSIGNALED(status)) {
676 			log_warnx("child process %d terminated with signal %d",
677 			    pid, WTERMSIG(status));
678 		} else if (WEXITSTATUS(status) != 0) {
679 			log_warnx("child process %d terminated with exit status %d",
680 			    pid, WEXITSTATUS(status));
681 		} else {
682 			log_debugx("child process %d terminated gracefully", pid);
683 		}
684 		num++;
685 	}
686 
687 	return (num);
688 }
689 
690 int
691 main(int argc, char **argv)
692 {
693 	int ch, debug = 0, error, iscsi_fd, maxproc = 30, retval, saved_errno,
694 	    timeout = 60;
695 	bool dont_daemonize = false;
696 	struct pidfh *pidfh;
697 	pid_t pid, otherpid;
698 	const char *pidfile_path = DEFAULT_PIDFILE;
699 	struct iscsi_daemon_request request;
700 
701 	while ((ch = getopt(argc, argv, "P:dl:m:t:")) != -1) {
702 		switch (ch) {
703 		case 'P':
704 			pidfile_path = optarg;
705 			break;
706 		case 'd':
707 			dont_daemonize = true;
708 			debug++;
709 			break;
710 		case 'l':
711 			debug = atoi(optarg);
712 			break;
713 		case 'm':
714 			maxproc = atoi(optarg);
715 			break;
716 		case 't':
717 			timeout = atoi(optarg);
718 			break;
719 		case '?':
720 		default:
721 			usage();
722 		}
723 	}
724 	argc -= optind;
725 	if (argc != 0)
726 		usage();
727 
728 	log_init(debug);
729 
730 	pidfh = pidfile_open(pidfile_path, 0600, &otherpid);
731 	if (pidfh == NULL) {
732 		if (errno == EEXIST)
733 			log_errx(1, "daemon already running, pid: %jd.",
734 			    (intmax_t)otherpid);
735 		log_err(1, "cannot open or create pidfile \"%s\"",
736 		    pidfile_path);
737 	}
738 
739 	iscsi_fd = open(ISCSI_PATH, O_RDWR);
740 	if (iscsi_fd < 0 && errno == ENOENT) {
741 		saved_errno = errno;
742 		retval = kldload("iscsi");
743 		if (retval != -1)
744 			iscsi_fd = open(ISCSI_PATH, O_RDWR);
745 		else
746 			errno = saved_errno;
747 	}
748 	if (iscsi_fd < 0)
749 		log_err(1, "failed to open %s", ISCSI_PATH);
750 
751 	if (dont_daemonize == false) {
752 		if (daemon(0, 0) == -1) {
753 			log_warn("cannot daemonize");
754 			pidfile_remove(pidfh);
755 			exit(1);
756 		}
757 	}
758 
759 	pidfile_write(pidfh);
760 
761 	register_sigchld();
762 
763 	for (;;) {
764 		log_debugx("waiting for request from the kernel");
765 
766 		memset(&request, 0, sizeof(request));
767 		error = ioctl(iscsi_fd, ISCSIDWAIT, &request);
768 		if (error != 0) {
769 			if (errno == EINTR) {
770 				nchildren -= wait_for_children(false);
771 				assert(nchildren >= 0);
772 				continue;
773 			}
774 
775 			log_err(1, "ISCSIDWAIT");
776 		}
777 
778 		if (dont_daemonize) {
779 			log_debugx("not forking due to -d flag; "
780 			    "will exit after servicing a single request");
781 		} else {
782 			nchildren -= wait_for_children(false);
783 			assert(nchildren >= 0);
784 
785 			while (maxproc > 0 && nchildren >= maxproc) {
786 				log_debugx("maxproc limit of %d child processes hit; "
787 				    "waiting for child process to exit", maxproc);
788 				nchildren -= wait_for_children(true);
789 				assert(nchildren >= 0);
790 			}
791 			log_debugx("incoming connection; forking child process #%d",
792 			    nchildren);
793 			nchildren++;
794 
795 			pid = fork();
796 			if (pid < 0)
797 				log_err(1, "fork");
798 			if (pid > 0)
799 				continue;
800 		}
801 
802 		pidfile_close(pidfh);
803 		handle_request(iscsi_fd, &request, timeout);
804 	}
805 
806 	return (0);
807 }
808