xref: /freebsd/usr.sbin/iscsid/iscsid.c (revision 1c4ee7dfb8affed302171232b0f612e6bcba3c10)
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/types.h>
33 #include <sys/time.h>
34 #include <sys/ioctl.h>
35 #include <sys/param.h>
36 #include <sys/linker.h>
37 #include <sys/socket.h>
38 #include <sys/sysctl.h>
39 #include <sys/capsicum.h>
40 #include <sys/wait.h>
41 #include <netinet/in.h>
42 #include <netinet/tcp.h>
43 #include <assert.h>
44 #include <capsicum_helpers.h>
45 #include <errno.h>
46 #include <fcntl.h>
47 #include <libutil.h>
48 #include <netdb.h>
49 #include <signal.h>
50 #include <stdbool.h>
51 #include <stdint.h>
52 #include <stdio.h>
53 #include <stdlib.h>
54 #include <string.h>
55 #include <unistd.h>
56 
57 #include "iscsid.h"
58 
59 static bool	timed_out(void);
60 #ifdef ICL_KERNEL_PROXY
61 static void	pdu_receive_proxy(struct pdu *pdu);
62 static void	pdu_send_proxy(struct pdu *pdu);
63 #endif /* ICL_KERNEL_PROXY */
64 
65 static volatile bool sigalrm_received = false;
66 
67 static int nchildren = 0;
68 
69 static struct connection_ops conn_ops = {
70 	.timed_out = timed_out,
71 #ifdef ICL_KERNEL_PROXY
72 	.pdu_receive_proxy = pdu_receive_proxy,
73 	.pdu_send_proxy = pdu_send_proxy,
74 #endif
75 	.fail = fail,
76 };
77 
78 static void
79 usage(void)
80 {
81 
82 	fprintf(stderr, "usage: iscsid [-P pidfile][-d][-m maxproc][-t timeout]\n");
83 	exit(1);
84 }
85 
86 #ifdef ICL_KERNEL_PROXY
87 
88 static void
89 pdu_receive_proxy(struct pdu *pdu)
90 {
91 	struct iscsid_connection *conn;
92 	struct iscsi_daemon_receive idr;
93 	size_t len;
94 	int error;
95 
96 	conn = (struct iscsid_connection *)pdu->pdu_connection;
97 	assert(conn->conn_conf.isc_iser != 0);
98 
99 	pdu->pdu_data = malloc(conn->conn.conn_max_recv_data_segment_length);
100 	if (pdu->pdu_data == NULL)
101 		log_err(1, "malloc");
102 
103 	memset(&idr, 0, sizeof(idr));
104 	idr.idr_session_id = conn->conn_session_id;
105 	idr.idr_bhs = pdu->pdu_bhs;
106 	idr.idr_data_segment_len = conn->conn.conn_max_recv_data_segment_length;
107 	idr.idr_data_segment = pdu->pdu_data;
108 
109 	error = ioctl(conn->conn_iscsi_fd, ISCSIDRECEIVE, &idr);
110 	if (error != 0)
111 		log_err(1, "ISCSIDRECEIVE");
112 
113 	len = pdu_ahs_length(pdu);
114 	if (len > 0)
115 		log_errx(1, "protocol error: non-empty AHS");
116 
117 	len = pdu_data_segment_length(pdu);
118 	assert(len <= (size_t)conn->conn.conn_max_recv_data_segment_length);
119 	pdu->pdu_data_len = len;
120 }
121 
122 static void
123 pdu_send_proxy(struct pdu *pdu)
124 {
125 	struct iscsid_connection *conn;
126 	struct iscsi_daemon_send ids;
127 	int error;
128 
129 	conn = (struct iscsid_connection *)pdu->pdu_connection;
130 	assert(conn->conn_conf.isc_iser != 0);
131 
132 	pdu_set_data_segment_length(pdu, pdu->pdu_data_len);
133 
134 	memset(&ids, 0, sizeof(ids));
135 	ids.ids_session_id = conn->conn_session_id;
136 	ids.ids_bhs = pdu->pdu_bhs;
137 	ids.ids_data_segment_len = pdu->pdu_data_len;
138 	ids.ids_data_segment = pdu->pdu_data;
139 
140 	error = ioctl(conn->conn_iscsi_fd, ISCSIDSEND, &ids);
141 	if (error != 0)
142 		log_err(1, "ISCSIDSEND");
143 }
144 
145 #endif /* ICL_KERNEL_PROXY */
146 
147 static void
148 resolve_addr(const struct connection *conn, const char *address,
149     struct addrinfo **ai, bool initiator_side)
150 {
151 	struct addrinfo hints;
152 	char *arg, *addr, *ch, *tofree;
153 	const char *port;
154 	int error, colons = 0;
155 
156 	tofree = arg = checked_strdup(address);
157 
158 	if (arg[0] == '\0') {
159 		fail(conn, "empty address");
160 		log_errx(1, "empty address");
161 	}
162 	if (arg[0] == '[') {
163 		/*
164 		 * IPv6 address in square brackets, perhaps with port.
165 		 */
166 		arg++;
167 		addr = strsep(&arg, "]");
168 		if (arg == NULL) {
169 			fail(conn, "malformed address");
170 			log_errx(1, "malformed address %s", address);
171 		}
172 		if (arg[0] == '\0') {
173 			port = NULL;
174 		} else if (arg[0] == ':') {
175 			port = arg + 1;
176 		} else {
177 			fail(conn, "malformed address");
178 			log_errx(1, "malformed address %s", address);
179 		}
180 	} else {
181 		/*
182 		 * Either IPv6 address without brackets - and without
183 		 * a port - or IPv4 address.  Just count the colons.
184 		 */
185 		for (ch = arg; *ch != '\0'; ch++) {
186 			if (*ch == ':')
187 				colons++;
188 		}
189 		if (colons > 1) {
190 			addr = arg;
191 			port = NULL;
192 		} else {
193 			addr = strsep(&arg, ":");
194 			if (arg == NULL)
195 				port = NULL;
196 			else
197 				port = arg;
198 		}
199 	}
200 
201 	if (port == NULL && !initiator_side)
202 		port = "3260";
203 
204 	memset(&hints, 0, sizeof(hints));
205 	hints.ai_family = PF_UNSPEC;
206 	hints.ai_socktype = SOCK_STREAM;
207 	hints.ai_flags = AI_ADDRCONFIG | AI_NUMERICSERV;
208 	if (initiator_side)
209 		hints.ai_flags |= AI_PASSIVE;
210 
211 	error = getaddrinfo(addr, port, &hints, ai);
212 	if (error != 0) {
213 		fail(conn, gai_strerror(error));
214 		log_errx(1, "getaddrinfo for %s failed: %s",
215 		    address, gai_strerror(error));
216 	}
217 
218 	free(tofree);
219 }
220 
221 static struct iscsid_connection *
222 connection_new(int iscsi_fd, const struct iscsi_daemon_request *request)
223 {
224 	struct iscsid_connection *conn;
225 	struct addrinfo *from_ai, *to_ai;
226 	const char *from_addr, *to_addr;
227 #ifdef ICL_KERNEL_PROXY
228 	struct iscsi_daemon_connect idc;
229 #endif
230 	int error, optval;
231 
232 	conn = calloc(1, sizeof(*conn));
233 	if (conn == NULL)
234 		log_err(1, "calloc");
235 
236 	connection_init(&conn->conn, &conn_ops,
237 	    request->idr_conf.isc_iser != 0);
238 	conn->conn_protocol_level = 0;
239 	conn->conn_initial_r2t = true;
240 	conn->conn_iscsi_fd = iscsi_fd;
241 
242 	conn->conn_session_id = request->idr_session_id;
243 	memcpy(&conn->conn_conf, &request->idr_conf, sizeof(conn->conn_conf));
244 	memcpy(&conn->conn.conn_isid, &request->idr_isid,
245 	    sizeof(conn->conn.conn_isid));
246 	conn->conn.conn_tsih = request->idr_tsih;
247 
248 	from_addr = conn->conn_conf.isc_initiator_addr;
249 	to_addr = conn->conn_conf.isc_target_addr;
250 
251 	if (from_addr[0] != '\0')
252 		resolve_addr(&conn->conn, from_addr, &from_ai, true);
253 	else
254 		from_ai = NULL;
255 
256 	resolve_addr(&conn->conn, to_addr, &to_ai, false);
257 
258 #ifdef ICL_KERNEL_PROXY
259 	if (conn->conn_conf.isc_iser) {
260 		memset(&idc, 0, sizeof(idc));
261 		idc.idc_session_id = conn->conn_session_id;
262 		if (conn->conn_conf.isc_iser)
263 			idc.idc_iser = 1;
264 		idc.idc_domain = to_ai->ai_family;
265 		idc.idc_socktype = to_ai->ai_socktype;
266 		idc.idc_protocol = to_ai->ai_protocol;
267 		if (from_ai != NULL) {
268 			idc.idc_from_addr = from_ai->ai_addr;
269 			idc.idc_from_addrlen = from_ai->ai_addrlen;
270 		}
271 		idc.idc_to_addr = to_ai->ai_addr;
272 		idc.idc_to_addrlen = to_ai->ai_addrlen;
273 
274 		log_debugx("connecting to %s using ICL kernel proxy", to_addr);
275 		error = ioctl(iscsi_fd, ISCSIDCONNECT, &idc);
276 		if (error != 0) {
277 			fail(&conn->conn, strerror(errno));
278 			log_err(1, "failed to connect to %s "
279 			    "using ICL kernel proxy: ISCSIDCONNECT", to_addr);
280 		}
281 
282 		if (from_ai != NULL)
283 			freeaddrinfo(from_ai);
284 		freeaddrinfo(to_ai);
285 
286 		return (conn);
287 	}
288 #endif /* ICL_KERNEL_PROXY */
289 
290 	if (conn->conn_conf.isc_iser) {
291 		fail(&conn->conn, "iSER not supported");
292 		log_errx(1, "iscsid(8) compiled without ICL_KERNEL_PROXY "
293 		    "does not support iSER");
294 	}
295 
296 	conn->conn.conn_socket = socket(to_ai->ai_family, to_ai->ai_socktype,
297 	    to_ai->ai_protocol);
298 	if (conn->conn.conn_socket < 0) {
299 		fail(&conn->conn, strerror(errno));
300 		log_err(1, "failed to create socket for %s", from_addr);
301 	}
302 	optval = SOCKBUF_SIZE;
303 	if (setsockopt(conn->conn.conn_socket, SOL_SOCKET, SO_RCVBUF,
304 	    &optval, sizeof(optval)) == -1)
305 		log_warn("setsockopt(SO_RCVBUF) failed");
306 	optval = SOCKBUF_SIZE;
307 	if (setsockopt(conn->conn.conn_socket, SOL_SOCKET, SO_SNDBUF,
308 	    &optval, sizeof(optval)) == -1)
309 		log_warn("setsockopt(SO_SNDBUF) failed");
310 	optval = 1;
311 	if (setsockopt(conn->conn.conn_socket, SOL_SOCKET, SO_NO_DDP,
312 	    &optval, sizeof(optval)) == -1)
313 		log_warn("setsockopt(SO_NO_DDP) failed");
314 	if (conn->conn_conf.isc_dscp != -1) {
315 		int tos = conn->conn_conf.isc_dscp << 2;
316 		if (to_ai->ai_family == AF_INET) {
317 			if (setsockopt(conn->conn.conn_socket,
318 			    IPPROTO_IP, IP_TOS,
319 			    &tos, sizeof(tos)) == -1)
320 				log_warn("setsockopt(IP_TOS) "
321 				    "failed for %s",
322 				    from_addr);
323 		} else
324 		if (to_ai->ai_family == AF_INET6) {
325 			if (setsockopt(conn->conn.conn_socket,
326 			    IPPROTO_IPV6, IPV6_TCLASS,
327 			    &tos, sizeof(tos)) == -1)
328 				log_warn("setsockopt(IPV6_TCLASS) "
329 				    "failed for %s",
330 				    from_addr);
331 		}
332 	}
333 	if (conn->conn_conf.isc_pcp != -1) {
334 		int pcp = conn->conn_conf.isc_pcp;
335 		if (to_ai->ai_family == AF_INET) {
336 			if (setsockopt(conn->conn.conn_socket,
337 			    IPPROTO_IP, IP_VLAN_PCP,
338 			    &pcp, sizeof(pcp)) == -1)
339 				log_warn("setsockopt(IP_VLAN_PCP) "
340 				    "failed for %s",
341 				    from_addr);
342 		} else
343 		if (to_ai->ai_family == AF_INET6) {
344 			if (setsockopt(conn->conn.conn_socket,
345 			    IPPROTO_IPV6, IPV6_VLAN_PCP,
346 			    &pcp, sizeof(pcp)) == -1)
347 				log_warn("setsockopt(IPV6_VLAN_PCP) "
348 				    "failed for %s",
349 				    from_addr);
350 		}
351 	}
352 	/*
353 	 * Reduce TCP SYN_SENT timeout while
354 	 * no connectivity exists, to allow
355 	 * rapid reuse of the available slots.
356 	 */
357 	int keepinit = 0;
358 	if (conn->conn_conf.isc_login_timeout > 0) {
359 		keepinit = conn->conn_conf.isc_login_timeout;
360 		log_debugx("session specific LoginTimeout at %d sec",
361 			keepinit);
362 	}
363 	if (conn->conn_conf.isc_login_timeout == -1) {
364 		int value;
365 		size_t size = sizeof(value);
366 		if (sysctlbyname("kern.iscsi.login_timeout",
367 		    &value, &size, NULL, 0) == 0) {
368 			keepinit = value;
369 			log_debugx("global login_timeout at %d sec",
370 				keepinit);
371 		}
372 	}
373 	if (keepinit > 0) {
374 		if (setsockopt(conn->conn.conn_socket,
375 		    IPPROTO_TCP, TCP_KEEPINIT,
376 		    &keepinit, sizeof(keepinit)) == -1)
377 			log_warnx("setsockopt(TCP_KEEPINIT) "
378 			    "failed for %s", to_addr);
379 	}
380 	if (from_ai != NULL) {
381 		error = bind(conn->conn.conn_socket, from_ai->ai_addr,
382 		    from_ai->ai_addrlen);
383 		if (error != 0) {
384 			fail(&conn->conn, strerror(errno));
385 			log_err(1, "failed to bind to %s", from_addr);
386 		}
387 	}
388 	log_debugx("connecting to %s", to_addr);
389 	error = connect(conn->conn.conn_socket, to_ai->ai_addr,
390 	    to_ai->ai_addrlen);
391 	if (error != 0) {
392 		fail(&conn->conn, strerror(errno));
393 		log_err(1, "failed to connect to %s", to_addr);
394 	}
395 
396 	if (from_ai != NULL)
397 		freeaddrinfo(from_ai);
398 	freeaddrinfo(to_ai);
399 
400 	return (conn);
401 }
402 
403 static void
404 limits(struct iscsid_connection *conn)
405 {
406 	struct iscsi_daemon_limits idl;
407 	struct iscsi_session_limits *isl;
408 	int error;
409 
410 	log_debugx("fetching limits from the kernel");
411 
412 	memset(&idl, 0, sizeof(idl));
413 	idl.idl_session_id = conn->conn_session_id;
414 	idl.idl_socket = conn->conn.conn_socket;
415 
416 	error = ioctl(conn->conn_iscsi_fd, ISCSIDLIMITS, &idl);
417 	if (error != 0)
418 		log_err(1, "ISCSIDLIMITS");
419 
420 	/*
421 	 * Read the driver limits and provide reasonable defaults for the ones
422 	 * the driver doesn't care about.  If a max_snd_dsl is not explicitly
423 	 * provided by the driver then we'll make sure both conn->max_snd_dsl
424 	 * and isl->max_snd_dsl are set to the rcv_dsl.  This preserves historic
425 	 * behavior.
426 	 */
427 	isl = &conn->conn_limits;
428 	memcpy(isl, &idl.idl_limits, sizeof(*isl));
429 	if (isl->isl_max_recv_data_segment_length == 0)
430 		isl->isl_max_recv_data_segment_length = (1 << 24) - 1;
431 	if (isl->isl_max_send_data_segment_length == 0)
432 		isl->isl_max_send_data_segment_length =
433 		    isl->isl_max_recv_data_segment_length;
434 	if (isl->isl_max_burst_length == 0)
435 		isl->isl_max_burst_length = (1 << 24) - 1;
436 	if (isl->isl_first_burst_length == 0)
437 		isl->isl_first_burst_length = (1 << 24) - 1;
438 	if (isl->isl_first_burst_length > isl->isl_max_burst_length)
439 		isl->isl_first_burst_length = isl->isl_max_burst_length;
440 
441 	/*
442 	 * Limit default send length in case it won't be negotiated.
443 	 * We can't do it for other limits, since they may affect both
444 	 * sender and receiver operation, and we must obey defaults.
445 	 */
446 	if (conn->conn.conn_max_send_data_segment_length >
447 	    isl->isl_max_send_data_segment_length) {
448 		conn->conn.conn_max_send_data_segment_length =
449 		    isl->isl_max_send_data_segment_length;
450 	}
451 }
452 
453 static void
454 handoff(struct iscsid_connection *conn)
455 {
456 	struct iscsi_daemon_handoff idh;
457 	int error;
458 
459 	log_debugx("handing off connection to the kernel");
460 
461 	memset(&idh, 0, sizeof(idh));
462 	idh.idh_session_id = conn->conn_session_id;
463 	idh.idh_socket = conn->conn.conn_socket;
464 	strlcpy(idh.idh_target_alias, conn->conn_target_alias,
465 	    sizeof(idh.idh_target_alias));
466 	idh.idh_tsih = conn->conn.conn_tsih;
467 	idh.idh_statsn = conn->conn.conn_statsn;
468 	idh.idh_protocol_level = conn->conn_protocol_level;
469 	idh.idh_header_digest = conn->conn.conn_header_digest;
470 	idh.idh_data_digest = conn->conn.conn_data_digest;
471 	idh.idh_initial_r2t = conn->conn_initial_r2t;
472 	idh.idh_immediate_data = conn->conn.conn_immediate_data;
473 	idh.idh_max_recv_data_segment_length =
474 	    conn->conn.conn_max_recv_data_segment_length;
475 	idh.idh_max_send_data_segment_length =
476 	    conn->conn.conn_max_send_data_segment_length;
477 	idh.idh_max_burst_length = conn->conn.conn_max_burst_length;
478 	idh.idh_first_burst_length = conn->conn.conn_first_burst_length;
479 
480 	error = ioctl(conn->conn_iscsi_fd, ISCSIDHANDOFF, &idh);
481 	if (error != 0)
482 		log_err(1, "ISCSIDHANDOFF");
483 }
484 
485 void
486 fail(const struct connection *base_conn, const char *reason)
487 {
488 	const struct iscsid_connection *conn;
489 	struct iscsi_daemon_fail idf;
490 	int error, saved_errno;
491 
492 	conn = (const struct iscsid_connection *)base_conn;
493 	saved_errno = errno;
494 
495 	memset(&idf, 0, sizeof(idf));
496 	idf.idf_session_id = conn->conn_session_id;
497 	strlcpy(idf.idf_reason, reason, sizeof(idf.idf_reason));
498 
499 	error = ioctl(conn->conn_iscsi_fd, ISCSIDFAIL, &idf);
500 	if (error != 0)
501 		log_err(1, "ISCSIDFAIL");
502 
503 	errno = saved_errno;
504 }
505 
506 /*
507  * XXX: I CANT INTO LATIN
508  */
509 static void
510 capsicate(struct iscsid_connection *conn)
511 {
512 	cap_rights_t rights;
513 	const unsigned long cmds[] = {
514 #ifdef ICL_KERNEL_PROXY
515 		ISCSIDCONNECT,
516 		ISCSIDSEND,
517 		ISCSIDRECEIVE,
518 #endif
519 		ISCSIDLIMITS,
520 		ISCSIDHANDOFF,
521 		ISCSIDFAIL,
522 		ISCSISADD,
523 		ISCSISREMOVE,
524 		ISCSISMODIFY
525 	};
526 
527 	cap_rights_init(&rights, CAP_IOCTL);
528 	if (caph_rights_limit(conn->conn_iscsi_fd, &rights) < 0)
529 		log_err(1, "cap_rights_limit");
530 
531 	if (caph_ioctls_limit(conn->conn_iscsi_fd, cmds, nitems(cmds)) < 0)
532 		log_err(1, "cap_ioctls_limit");
533 
534 	if (caph_enter() != 0)
535 		log_err(1, "cap_enter");
536 
537 	if (cap_sandboxed())
538 		log_debugx("Capsicum capability mode enabled");
539 	else
540 		log_warnx("Capsicum capability mode not supported");
541 }
542 
543 static bool
544 timed_out(void)
545 {
546 
547 	return (sigalrm_received);
548 }
549 
550 static void
551 sigalrm_handler(int dummy __unused)
552 {
553 	/*
554 	 * It would be easiest to just log an error and exit.  We can't
555 	 * do this, though, because log_errx() is not signal safe, since
556 	 * it calls syslog(3).  Instead, set a flag checked by pdu_send()
557 	 * and pdu_receive(), to call log_errx() there.  Should they fail
558 	 * to notice, we'll exit here one second later.
559 	 */
560 	if (sigalrm_received) {
561 		/*
562 		 * Oh well.  Just give up and quit.
563 		 */
564 		_exit(2);
565 	}
566 
567 	sigalrm_received = true;
568 }
569 
570 static void
571 set_timeout(int timeout)
572 {
573 	struct sigaction sa;
574 	struct itimerval itv;
575 	int error;
576 
577 	if (timeout <= 0) {
578 		log_debugx("session timeout disabled");
579 		return;
580 	}
581 
582 	bzero(&sa, sizeof(sa));
583 	sa.sa_handler = sigalrm_handler;
584 	sigfillset(&sa.sa_mask);
585 	error = sigaction(SIGALRM, &sa, NULL);
586 	if (error != 0)
587 		log_err(1, "sigaction");
588 
589 	/*
590 	 * First SIGALRM will arive after conf_timeout seconds.
591 	 * If we do nothing, another one will arrive a second later.
592 	 */
593 	bzero(&itv, sizeof(itv));
594 	itv.it_interval.tv_sec = 1;
595 	itv.it_value.tv_sec = timeout;
596 
597 	log_debugx("setting session timeout to %d seconds",
598 	    timeout);
599 	error = setitimer(ITIMER_REAL, &itv, NULL);
600 	if (error != 0)
601 		log_err(1, "setitimer");
602 }
603 
604 static void
605 sigchld_handler(int dummy __unused)
606 {
607 
608 	/*
609 	 * The only purpose of this handler is to make SIGCHLD
610 	 * interrupt the ISCSIDWAIT ioctl(2), so we can call
611 	 * wait_for_children().
612 	 */
613 }
614 
615 static void
616 register_sigchld(void)
617 {
618 	struct sigaction sa;
619 	int error;
620 
621 	bzero(&sa, sizeof(sa));
622 	sa.sa_handler = sigchld_handler;
623 	sigfillset(&sa.sa_mask);
624 	error = sigaction(SIGCHLD, &sa, NULL);
625 	if (error != 0)
626 		log_err(1, "sigaction");
627 
628 }
629 
630 static void
631 handle_request(int iscsi_fd, const struct iscsi_daemon_request *request, int timeout)
632 {
633 	struct iscsid_connection *conn;
634 
635 	log_set_peer_addr(request->idr_conf.isc_target_addr);
636 	if (request->idr_conf.isc_target[0] != '\0') {
637 		log_set_peer_name(request->idr_conf.isc_target);
638 		setproctitle("%s (%s)", request->idr_conf.isc_target_addr, request->idr_conf.isc_target);
639 	} else {
640 		setproctitle("%s", request->idr_conf.isc_target_addr);
641 	}
642 
643 	conn = connection_new(iscsi_fd, request);
644 	capsicate(conn);
645 	limits(conn);
646 	set_timeout(timeout);
647 	login(conn);
648 	if (conn->conn_conf.isc_discovery != 0)
649 		discovery(conn);
650 	else
651 		handoff(conn);
652 
653 	log_debugx("nothing more to do; exiting");
654 	exit (0);
655 }
656 
657 static int
658 wait_for_children(bool block)
659 {
660 	pid_t pid;
661 	int status;
662 	int num = 0;
663 
664 	for (;;) {
665 		/*
666 		 * If "block" is true, wait for at least one process.
667 		 */
668 		if (block && num == 0)
669 			pid = wait4(-1, &status, 0, NULL);
670 		else
671 			pid = wait4(-1, &status, WNOHANG, NULL);
672 		if (pid <= 0)
673 			break;
674 		if (WIFSIGNALED(status)) {
675 			log_warnx("child process %d terminated with signal %d",
676 			    pid, WTERMSIG(status));
677 		} else if (WEXITSTATUS(status) != 0) {
678 			log_warnx("child process %d terminated with exit status %d",
679 			    pid, WEXITSTATUS(status));
680 		} else {
681 			log_debugx("child process %d terminated gracefully", pid);
682 		}
683 		num++;
684 	}
685 
686 	return (num);
687 }
688 
689 int
690 main(int argc, char **argv)
691 {
692 	int ch, debug = 0, error, iscsi_fd, maxproc = 30, retval, saved_errno,
693 	    timeout = 60;
694 	bool dont_daemonize = false;
695 	struct pidfh *pidfh;
696 	pid_t pid, otherpid;
697 	const char *pidfile_path = DEFAULT_PIDFILE;
698 	struct iscsi_daemon_request request;
699 
700 	while ((ch = getopt(argc, argv, "P:dl:m:t:")) != -1) {
701 		switch (ch) {
702 		case 'P':
703 			pidfile_path = optarg;
704 			break;
705 		case 'd':
706 			dont_daemonize = true;
707 			debug++;
708 			break;
709 		case 'l':
710 			debug = atoi(optarg);
711 			break;
712 		case 'm':
713 			maxproc = atoi(optarg);
714 			break;
715 		case 't':
716 			timeout = atoi(optarg);
717 			break;
718 		case '?':
719 		default:
720 			usage();
721 		}
722 	}
723 	argc -= optind;
724 	if (argc != 0)
725 		usage();
726 
727 	log_init(debug);
728 
729 	pidfh = pidfile_open(pidfile_path, 0600, &otherpid);
730 	if (pidfh == NULL) {
731 		if (errno == EEXIST)
732 			log_errx(1, "daemon already running, pid: %jd.",
733 			    (intmax_t)otherpid);
734 		log_err(1, "cannot open or create pidfile \"%s\"",
735 		    pidfile_path);
736 	}
737 
738 	iscsi_fd = open(ISCSI_PATH, O_RDWR);
739 	if (iscsi_fd < 0 && errno == ENOENT) {
740 		saved_errno = errno;
741 		retval = kldload("iscsi");
742 		if (retval != -1)
743 			iscsi_fd = open(ISCSI_PATH, O_RDWR);
744 		else
745 			errno = saved_errno;
746 	}
747 	if (iscsi_fd < 0)
748 		log_err(1, "failed to open %s", ISCSI_PATH);
749 
750 	if (dont_daemonize == false) {
751 		if (daemon(0, 0) == -1) {
752 			log_warn("cannot daemonize");
753 			pidfile_remove(pidfh);
754 			exit(1);
755 		}
756 	}
757 
758 	pidfile_write(pidfh);
759 
760 	register_sigchld();
761 
762 	for (;;) {
763 		log_debugx("waiting for request from the kernel");
764 
765 		memset(&request, 0, sizeof(request));
766 		error = ioctl(iscsi_fd, ISCSIDWAIT, &request);
767 		if (error != 0) {
768 			if (errno == EINTR) {
769 				nchildren -= wait_for_children(false);
770 				assert(nchildren >= 0);
771 				continue;
772 			}
773 
774 			log_err(1, "ISCSIDWAIT");
775 		}
776 
777 		if (dont_daemonize) {
778 			log_debugx("not forking due to -d flag; "
779 			    "will exit after servicing a single request");
780 		} else {
781 			nchildren -= wait_for_children(false);
782 			assert(nchildren >= 0);
783 
784 			while (maxproc > 0 && nchildren >= maxproc) {
785 				log_debugx("maxproc limit of %d child processes hit; "
786 				    "waiting for child process to exit", maxproc);
787 				nchildren -= wait_for_children(true);
788 				assert(nchildren >= 0);
789 			}
790 			log_debugx("incoming connection; forking child process #%d",
791 			    nchildren);
792 			nchildren++;
793 
794 			pid = fork();
795 			if (pid < 0)
796 				log_err(1, "fork");
797 			if (pid > 0)
798 				continue;
799 		}
800 
801 		pidfile_close(pidfh);
802 		handle_request(iscsi_fd, &request, timeout);
803 	}
804 
805 	return (0);
806 }
807