devd.cc revision 263758
1/*-
2 * Copyright (c) 2002-2010 M. Warner Losh.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * my_system is a variation on lib/libc/stdlib/system.c:
27 *
28 * Copyright (c) 1988, 1993
29 *	The Regents of the University of California.  All rights reserved.
30 *
31 * Redistribution and use in source and binary forms, with or without
32 * modification, are permitted provided that the following conditions
33 * are met:
34 * 1. Redistributions of source code must retain the above copyright
35 *    notice, this list of conditions and the following disclaimer.
36 * 2. Redistributions in binary form must reproduce the above copyright
37 *    notice, this list of conditions and the following disclaimer in the
38 *    documentation and/or other materials provided with the distribution.
39 * 4. Neither the name of the University nor the names of its contributors
40 *    may be used to endorse or promote products derived from this software
41 *    without specific prior written permission.
42 *
43 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
44 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
45 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
46 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
47 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
48 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
49 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
50 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
51 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
52 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
53 * SUCH DAMAGE.
54 */
55
56/*
57 * DEVD control daemon.
58 */
59
60// TODO list:
61//	o devd.conf and devd man pages need a lot of help:
62//	  - devd needs to document the unix domain socket
63//	  - devd.conf needs more details on the supported statements.
64
65#include <sys/cdefs.h>
66__FBSDID("$FreeBSD: head/sbin/devd/devd.cc 263758 2014-03-26 02:25:40Z mjg $");
67
68#include <sys/param.h>
69#include <sys/socket.h>
70#include <sys/stat.h>
71#include <sys/sysctl.h>
72#include <sys/types.h>
73#include <sys/wait.h>
74#include <sys/un.h>
75
76#include <cctype>
77#include <cerrno>
78#include <cstdlib>
79#include <cstdio>
80#include <csignal>
81#include <cstring>
82#include <cstdarg>
83
84#include <dirent.h>
85#include <err.h>
86#include <fcntl.h>
87#include <libutil.h>
88#include <paths.h>
89#include <poll.h>
90#include <regex.h>
91#include <syslog.h>
92#include <unistd.h>
93
94#include <algorithm>
95#include <map>
96#include <string>
97#include <list>
98#include <vector>
99
100#include "devd.h"		/* C compatible definitions */
101#include "devd.hh"		/* C++ class definitions */
102
103#define PIPE "/var/run/devd.pipe"
104#define CF "/etc/devd.conf"
105#define SYSCTL "hw.bus.devctl_queue"
106
107/*
108 * Since the client socket is nonblocking, we must increase its send buffer to
109 * handle brief event storms.  On FreeBSD, AF_UNIX sockets don't have a receive
110 * buffer, so the client can't increate the buffersize by itself.
111 *
112 * For example, when creating a ZFS pool, devd emits one 165 character
113 * resource.fs.zfs.statechange message for each vdev in the pool.  A 64k
114 * buffer has enough space for almost 400 drives, which would be very large but
115 * not impossibly large pool.  A 128k buffer has enough space for 794 drives,
116 * which is more than can fit in a rack with modern technology.
117 */
118#define CLIENT_BUFSIZE 131072
119
120using namespace std;
121
122extern FILE *yyin;
123extern int lineno;
124
125static const char notify = '!';
126static const char nomatch = '?';
127static const char attach = '+';
128static const char detach = '-';
129
130static struct pidfh *pfh;
131
132static int no_daemon = 0;
133static int daemonize_quick = 0;
134static int quiet_mode = 0;
135static unsigned total_events = 0;
136static volatile sig_atomic_t got_siginfo = 0;
137static volatile sig_atomic_t romeo_must_die = 0;
138
139static const char *configfile = CF;
140
141static void devdlog(int priority, const char* message, ...)
142	__printflike(2, 3);
143static void event_loop(void);
144static void usage(void);
145
146template <class T> void
147delete_and_clear(vector<T *> &v)
148{
149	typename vector<T *>::const_iterator i;
150
151	for (i = v.begin(); i != v.end(); ++i)
152		delete *i;
153	v.clear();
154}
155
156config cfg;
157
158event_proc::event_proc() : _prio(-1)
159{
160	_epsvec.reserve(4);
161}
162
163event_proc::~event_proc()
164{
165	delete_and_clear(_epsvec);
166}
167
168void
169event_proc::add(eps *eps)
170{
171	_epsvec.push_back(eps);
172}
173
174bool
175event_proc::matches(config &c) const
176{
177	vector<eps *>::const_iterator i;
178
179	for (i = _epsvec.begin(); i != _epsvec.end(); ++i)
180		if (!(*i)->do_match(c))
181			return (false);
182	return (true);
183}
184
185bool
186event_proc::run(config &c) const
187{
188	vector<eps *>::const_iterator i;
189
190	for (i = _epsvec.begin(); i != _epsvec.end(); ++i)
191		if (!(*i)->do_action(c))
192			return (false);
193	return (true);
194}
195
196action::action(const char *cmd)
197	: _cmd(cmd)
198{
199	// nothing
200}
201
202action::~action()
203{
204	// nothing
205}
206
207static int
208my_system(const char *command)
209{
210	pid_t pid, savedpid;
211	int pstat;
212	struct sigaction ign, intact, quitact;
213	sigset_t newsigblock, oldsigblock;
214
215	if (!command)		/* just checking... */
216		return (1);
217
218	/*
219	 * Ignore SIGINT and SIGQUIT, block SIGCHLD. Remember to save
220	 * existing signal dispositions.
221	 */
222	ign.sa_handler = SIG_IGN;
223	::sigemptyset(&ign.sa_mask);
224	ign.sa_flags = 0;
225	::sigaction(SIGINT, &ign, &intact);
226	::sigaction(SIGQUIT, &ign, &quitact);
227	::sigemptyset(&newsigblock);
228	::sigaddset(&newsigblock, SIGCHLD);
229	::sigprocmask(SIG_BLOCK, &newsigblock, &oldsigblock);
230	switch (pid = ::fork()) {
231	case -1:			/* error */
232		break;
233	case 0:				/* child */
234		/*
235		 * Restore original signal dispositions and exec the command.
236		 */
237		::sigaction(SIGINT, &intact, NULL);
238		::sigaction(SIGQUIT,  &quitact, NULL);
239		::sigprocmask(SIG_SETMASK, &oldsigblock, NULL);
240		/*
241		 * Close the PID file, and all other open descriptors.
242		 * Inherit std{in,out,err} only.
243		 */
244		cfg.close_pidfile();
245		::closefrom(3);
246		::execl(_PATH_BSHELL, "sh", "-c", command, (char *)NULL);
247		::_exit(127);
248	default:			/* parent */
249		savedpid = pid;
250		do {
251			pid = ::wait4(savedpid, &pstat, 0, (struct rusage *)0);
252		} while (pid == -1 && errno == EINTR);
253		break;
254	}
255	::sigaction(SIGINT, &intact, NULL);
256	::sigaction(SIGQUIT,  &quitact, NULL);
257	::sigprocmask(SIG_SETMASK, &oldsigblock, NULL);
258	return (pid == -1 ? -1 : pstat);
259}
260
261bool
262action::do_action(config &c)
263{
264	string s = c.expand_string(_cmd.c_str());
265	devdlog(LOG_INFO, "Executing '%s'\n", s.c_str());
266	my_system(s.c_str());
267	return (true);
268}
269
270match::match(config &c, const char *var, const char *re) :
271	_inv(re[0] == '!'),
272	_var(var),
273	_re(c.expand_string(_inv ? re + 1 : re, "^", "$"))
274{
275	regcomp(&_regex, _re.c_str(), REG_EXTENDED | REG_NOSUB | REG_ICASE);
276}
277
278match::~match()
279{
280	regfree(&_regex);
281}
282
283bool
284match::do_match(config &c)
285{
286	const string &value = c.get_variable(_var);
287	bool retval;
288
289	/*
290	 * This function gets called WAY too often to justify calling syslog()
291	 * each time, even at LOG_DEBUG.  Because if syslogd isn't running, it
292	 * can consume excessive amounts of systime inside of connect().  Only
293	 * log when we're in -d mode.
294	 */
295	if (no_daemon) {
296		devdlog(LOG_DEBUG, "Testing %s=%s against %s, invert=%d\n",
297		    _var.c_str(), value.c_str(), _re.c_str(), _inv);
298	}
299
300	retval = (regexec(&_regex, value.c_str(), 0, NULL, 0) == 0);
301	if (_inv == 1)
302		retval = (retval == 0) ? 1 : 0;
303
304	return (retval);
305}
306
307#include <sys/sockio.h>
308#include <net/if.h>
309#include <net/if_media.h>
310
311media::media(config &, const char *var, const char *type)
312	: _var(var), _type(-1)
313{
314	static struct ifmedia_description media_types[] = {
315		{ IFM_ETHER,		"Ethernet" },
316		{ IFM_TOKEN,		"Tokenring" },
317		{ IFM_FDDI,		"FDDI" },
318		{ IFM_IEEE80211,	"802.11" },
319		{ IFM_ATM,		"ATM" },
320		{ -1,			"unknown" },
321		{ 0, NULL },
322	};
323	for (int i = 0; media_types[i].ifmt_string != NULL; ++i)
324		if (strcasecmp(type, media_types[i].ifmt_string) == 0) {
325			_type = media_types[i].ifmt_word;
326			break;
327		}
328}
329
330media::~media()
331{
332}
333
334bool
335media::do_match(config &c)
336{
337	string value;
338	struct ifmediareq ifmr;
339	bool retval;
340	int s;
341
342	// Since we can be called from both a device attach/detach
343	// context where device-name is defined and what we want,
344	// as well as from a link status context, where subsystem is
345	// the name of interest, first try device-name and fall back
346	// to subsystem if none exists.
347	value = c.get_variable("device-name");
348	if (value.empty())
349		value = c.get_variable("subsystem");
350	devdlog(LOG_DEBUG, "Testing media type of %s against 0x%x\n",
351		    value.c_str(), _type);
352
353	retval = false;
354
355	s = socket(PF_INET, SOCK_DGRAM, 0);
356	if (s >= 0) {
357		memset(&ifmr, 0, sizeof(ifmr));
358		strncpy(ifmr.ifm_name, value.c_str(), sizeof(ifmr.ifm_name));
359
360		if (ioctl(s, SIOCGIFMEDIA, (caddr_t)&ifmr) >= 0 &&
361		    ifmr.ifm_status & IFM_AVALID) {
362			devdlog(LOG_DEBUG, "%s has media type 0x%x\n",
363				    value.c_str(), IFM_TYPE(ifmr.ifm_active));
364			retval = (IFM_TYPE(ifmr.ifm_active) == _type);
365		} else if (_type == -1) {
366			devdlog(LOG_DEBUG, "%s has unknown media type\n",
367				    value.c_str());
368			retval = true;
369		}
370		close(s);
371	}
372
373	return (retval);
374}
375
376const string var_list::bogus = "_$_$_$_$_B_O_G_U_S_$_$_$_$_";
377const string var_list::nothing = "";
378
379const string &
380var_list::get_variable(const string &var) const
381{
382	map<string, string>::const_iterator i;
383
384	i = _vars.find(var);
385	if (i == _vars.end())
386		return (var_list::bogus);
387	return (i->second);
388}
389
390bool
391var_list::is_set(const string &var) const
392{
393	return (_vars.find(var) != _vars.end());
394}
395
396void
397var_list::set_variable(const string &var, const string &val)
398{
399	/*
400	 * This function gets called WAY too often to justify calling syslog()
401	 * each time, even at LOG_DEBUG.  Because if syslogd isn't running, it
402	 * can consume excessive amounts of systime inside of connect().  Only
403	 * log when we're in -d mode.
404	 */
405	if (no_daemon)
406		devdlog(LOG_DEBUG, "setting %s=%s\n", var.c_str(), val.c_str());
407	_vars[var] = val;
408}
409
410void
411config::reset(void)
412{
413	_dir_list.clear();
414	delete_and_clear(_var_list_table);
415	delete_and_clear(_attach_list);
416	delete_and_clear(_detach_list);
417	delete_and_clear(_nomatch_list);
418	delete_and_clear(_notify_list);
419}
420
421void
422config::parse_one_file(const char *fn)
423{
424	devdlog(LOG_DEBUG, "Parsing %s\n", fn);
425	yyin = fopen(fn, "r");
426	if (yyin == NULL)
427		err(1, "Cannot open config file %s", fn);
428	lineno = 1;
429	if (yyparse() != 0)
430		errx(1, "Cannot parse %s at line %d", fn, lineno);
431	fclose(yyin);
432}
433
434void
435config::parse_files_in_dir(const char *dirname)
436{
437	DIR *dirp;
438	struct dirent *dp;
439	char path[PATH_MAX];
440
441	devdlog(LOG_DEBUG, "Parsing files in %s\n", dirname);
442	dirp = opendir(dirname);
443	if (dirp == NULL)
444		return;
445	readdir(dirp);		/* Skip . */
446	readdir(dirp);		/* Skip .. */
447	while ((dp = readdir(dirp)) != NULL) {
448		if (strcmp(dp->d_name + dp->d_namlen - 5, ".conf") == 0) {
449			snprintf(path, sizeof(path), "%s/%s",
450			    dirname, dp->d_name);
451			parse_one_file(path);
452		}
453	}
454	closedir(dirp);
455}
456
457class epv_greater {
458public:
459	int operator()(event_proc *const&l1, event_proc *const&l2) const
460	{
461		return (l1->get_priority() > l2->get_priority());
462	}
463};
464
465void
466config::sort_vector(vector<event_proc *> &v)
467{
468	stable_sort(v.begin(), v.end(), epv_greater());
469}
470
471void
472config::parse(void)
473{
474	vector<string>::const_iterator i;
475
476	parse_one_file(configfile);
477	for (i = _dir_list.begin(); i != _dir_list.end(); ++i)
478		parse_files_in_dir((*i).c_str());
479	sort_vector(_attach_list);
480	sort_vector(_detach_list);
481	sort_vector(_nomatch_list);
482	sort_vector(_notify_list);
483}
484
485void
486config::open_pidfile()
487{
488	pid_t otherpid;
489
490	if (_pidfile.empty())
491		return;
492	pfh = pidfile_open(_pidfile.c_str(), 0600, &otherpid);
493	if (pfh == NULL) {
494		if (errno == EEXIST)
495			errx(1, "devd already running, pid: %d", (int)otherpid);
496		warn("cannot open pid file");
497	}
498}
499
500void
501config::write_pidfile()
502{
503
504	pidfile_write(pfh);
505}
506
507void
508config::close_pidfile()
509{
510
511	pidfile_close(pfh);
512}
513
514void
515config::remove_pidfile()
516{
517
518	pidfile_remove(pfh);
519}
520
521void
522config::add_attach(int prio, event_proc *p)
523{
524	p->set_priority(prio);
525	_attach_list.push_back(p);
526}
527
528void
529config::add_detach(int prio, event_proc *p)
530{
531	p->set_priority(prio);
532	_detach_list.push_back(p);
533}
534
535void
536config::add_directory(const char *dir)
537{
538	_dir_list.push_back(string(dir));
539}
540
541void
542config::add_nomatch(int prio, event_proc *p)
543{
544	p->set_priority(prio);
545	_nomatch_list.push_back(p);
546}
547
548void
549config::add_notify(int prio, event_proc *p)
550{
551	p->set_priority(prio);
552	_notify_list.push_back(p);
553}
554
555void
556config::set_pidfile(const char *fn)
557{
558	_pidfile = fn;
559}
560
561void
562config::push_var_table()
563{
564	var_list *vl;
565
566	vl = new var_list();
567	_var_list_table.push_back(vl);
568	devdlog(LOG_DEBUG, "Pushing table\n");
569}
570
571void
572config::pop_var_table()
573{
574	delete _var_list_table.back();
575	_var_list_table.pop_back();
576	devdlog(LOG_DEBUG, "Popping table\n");
577}
578
579void
580config::set_variable(const char *var, const char *val)
581{
582	_var_list_table.back()->set_variable(var, val);
583}
584
585const string &
586config::get_variable(const string &var)
587{
588	vector<var_list *>::reverse_iterator i;
589
590	for (i = _var_list_table.rbegin(); i != _var_list_table.rend(); ++i) {
591		if ((*i)->is_set(var))
592			return ((*i)->get_variable(var));
593	}
594	return (var_list::nothing);
595}
596
597bool
598config::is_id_char(char ch) const
599{
600	return (ch != '\0' && (isalpha(ch) || isdigit(ch) || ch == '_' ||
601	    ch == '-'));
602}
603
604void
605config::expand_one(const char *&src, string &dst)
606{
607	int count;
608	string buffer;
609
610	src++;
611	// $$ -> $
612	if (*src == '$') {
613		dst += *src++;
614		return;
615	}
616
617	// $(foo) -> $(foo)
618	// Not sure if I want to support this or not, so for now we just pass
619	// it through.
620	if (*src == '(') {
621		dst += '$';
622		count = 1;
623		/* If the string ends before ) is matched , return. */
624		while (count > 0 && *src) {
625			if (*src == ')')
626				count--;
627			else if (*src == '(')
628				count++;
629			dst += *src++;
630		}
631		return;
632	}
633
634	// $[^A-Za-z] -> $\1
635	if (!isalpha(*src)) {
636		dst += '$';
637		dst += *src++;
638		return;
639	}
640
641	// $var -> replace with value
642	do {
643		buffer += *src++;
644	} while (is_id_char(*src));
645	dst.append(get_variable(buffer));
646}
647
648const string
649config::expand_string(const char *src, const char *prepend, const char *append)
650{
651	const char *var_at;
652	string dst;
653
654	/*
655	 * 128 bytes is enough for 2427 of 2438 expansions that happen
656	 * while parsing config files, as tested on 2013-01-30.
657	 */
658	dst.reserve(128);
659
660	if (prepend != NULL)
661		dst = prepend;
662
663	for (;;) {
664		var_at = strchr(src, '$');
665		if (var_at == NULL) {
666			dst.append(src);
667			break;
668		}
669		dst.append(src, var_at - src);
670		src = var_at;
671		expand_one(src, dst);
672	}
673
674	if (append != NULL)
675		dst.append(append);
676
677	return (dst);
678}
679
680bool
681config::chop_var(char *&buffer, char *&lhs, char *&rhs) const
682{
683	char *walker;
684
685	if (*buffer == '\0')
686		return (false);
687	walker = lhs = buffer;
688	while (is_id_char(*walker))
689		walker++;
690	if (*walker != '=')
691		return (false);
692	walker++;		// skip =
693	if (*walker == '"') {
694		walker++;	// skip "
695		rhs = walker;
696		while (*walker && *walker != '"')
697			walker++;
698		if (*walker != '"')
699			return (false);
700		rhs[-2] = '\0';
701		*walker++ = '\0';
702	} else {
703		rhs = walker;
704		while (*walker && !isspace(*walker))
705			walker++;
706		if (*walker != '\0')
707			*walker++ = '\0';
708		rhs[-1] = '\0';
709	}
710	while (isspace(*walker))
711		walker++;
712	buffer = walker;
713	return (true);
714}
715
716
717char *
718config::set_vars(char *buffer)
719{
720	char *lhs;
721	char *rhs;
722
723	while (1) {
724		if (!chop_var(buffer, lhs, rhs))
725			break;
726		set_variable(lhs, rhs);
727	}
728	return (buffer);
729}
730
731void
732config::find_and_execute(char type)
733{
734	vector<event_proc *> *l;
735	vector<event_proc *>::const_iterator i;
736	const char *s;
737
738	switch (type) {
739	default:
740		return;
741	case notify:
742		l = &_notify_list;
743		s = "notify";
744		break;
745	case nomatch:
746		l = &_nomatch_list;
747		s = "nomatch";
748		break;
749	case attach:
750		l = &_attach_list;
751		s = "attach";
752		break;
753	case detach:
754		l = &_detach_list;
755		s = "detach";
756		break;
757	}
758	devdlog(LOG_DEBUG, "Processing %s event\n", s);
759	for (i = l->begin(); i != l->end(); ++i) {
760		if ((*i)->matches(*this)) {
761			(*i)->run(*this);
762			break;
763		}
764	}
765
766}
767
768
769static void
770process_event(char *buffer)
771{
772	char type;
773	char *sp;
774
775	sp = buffer + 1;
776	devdlog(LOG_INFO, "Processing event '%s'\n", buffer);
777	type = *buffer++;
778	cfg.push_var_table();
779	// No match doesn't have a device, and the format is a little
780	// different, so handle it separately.
781	switch (type) {
782	case notify:
783		sp = cfg.set_vars(sp);
784		break;
785	case nomatch:
786		//? at location pnp-info on bus
787		sp = strchr(sp, ' ');
788		if (sp == NULL)
789			return;	/* Can't happen? */
790		*sp++ = '\0';
791		while (isspace(*sp))
792			sp++;
793		if (strncmp(sp, "at ", 3) == 0)
794			sp += 3;
795		sp = cfg.set_vars(sp);
796		while (isspace(*sp))
797			sp++;
798		if (strncmp(sp, "on ", 3) == 0)
799			cfg.set_variable("bus", sp + 3);
800		break;
801	case attach:	/*FALLTHROUGH*/
802	case detach:
803		sp = strchr(sp, ' ');
804		if (sp == NULL)
805			return;	/* Can't happen? */
806		*sp++ = '\0';
807		cfg.set_variable("device-name", buffer);
808		while (isspace(*sp))
809			sp++;
810		if (strncmp(sp, "at ", 3) == 0)
811			sp += 3;
812		sp = cfg.set_vars(sp);
813		while (isspace(*sp))
814			sp++;
815		if (strncmp(sp, "on ", 3) == 0)
816			cfg.set_variable("bus", sp + 3);
817		break;
818	}
819
820	cfg.find_and_execute(type);
821	cfg.pop_var_table();
822}
823
824int
825create_socket(const char *name)
826{
827	int fd, slen;
828	struct sockaddr_un sun;
829
830	if ((fd = socket(PF_LOCAL, SOCK_STREAM, 0)) < 0)
831		err(1, "socket");
832	bzero(&sun, sizeof(sun));
833	sun.sun_family = AF_UNIX;
834	strlcpy(sun.sun_path, name, sizeof(sun.sun_path));
835	slen = SUN_LEN(&sun);
836	unlink(name);
837	if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0)
838	    	err(1, "fcntl");
839	if (::bind(fd, (struct sockaddr *) & sun, slen) < 0)
840		err(1, "bind");
841	listen(fd, 4);
842	chown(name, 0, 0);	/* XXX - root.wheel */
843	chmod(name, 0666);
844	return (fd);
845}
846
847unsigned int max_clients = 10;	/* Default, can be overriden on cmdline. */
848unsigned int num_clients;
849list<int> clients;
850
851void
852notify_clients(const char *data, int len)
853{
854	list<int>::iterator i;
855
856	/*
857	 * Deliver the data to all clients.  Throw clients overboard at the
858	 * first sign of trouble.  This reaps clients who've died or closed
859	 * their sockets, and also clients who are alive but failing to keep up
860	 * (or who are maliciously not reading, to consume buffer space in
861	 * kernel memory or tie up the limited number of available connections).
862	 */
863	for (i = clients.begin(); i != clients.end(); ) {
864		if (write(*i, data, len) != len) {
865			--num_clients;
866			close(*i);
867			i = clients.erase(i);
868			devdlog(LOG_WARNING, "notify_clients: write() failed; "
869			    "dropping unresponsive client\n");
870		} else
871			++i;
872	}
873}
874
875void
876check_clients(void)
877{
878	int s;
879	struct pollfd pfd;
880	list<int>::iterator i;
881
882	/*
883	 * Check all existing clients to see if any of them have disappeared.
884	 * Normally we reap clients when we get an error trying to send them an
885	 * event.  This check eliminates the problem of an ever-growing list of
886	 * zombie clients because we're never writing to them on a system
887	 * without frequent device-change activity.
888	 */
889	pfd.events = 0;
890	for (i = clients.begin(); i != clients.end(); ) {
891		pfd.fd = *i;
892		s = poll(&pfd, 1, 0);
893		if ((s < 0 && s != EINTR ) ||
894		    (s > 0 && (pfd.revents & POLLHUP))) {
895			--num_clients;
896			close(*i);
897			i = clients.erase(i);
898			devdlog(LOG_NOTICE, "check_clients:  "
899			    "dropping disconnected client\n");
900		} else
901			++i;
902	}
903}
904
905void
906new_client(int fd)
907{
908	int s;
909	int sndbuf_size;
910
911	/*
912	 * First go reap any zombie clients, then accept the connection, and
913	 * shut down the read side to stop clients from consuming kernel memory
914	 * by sending large buffers full of data we'll never read.
915	 */
916	check_clients();
917	s = accept(fd, NULL, NULL);
918	if (s != -1) {
919		sndbuf_size = CLIENT_BUFSIZE;
920		if (setsockopt(s, SOL_SOCKET, SO_SNDBUF, &sndbuf_size,
921		    sizeof(sndbuf_size)))
922			err(1, "setsockopt");
923		shutdown(s, SHUT_RD);
924		clients.push_back(s);
925		++num_clients;
926	} else
927		err(1, "accept");
928}
929
930static void
931event_loop(void)
932{
933	int rv;
934	int fd;
935	char buffer[DEVCTL_MAXBUF];
936	int once = 0;
937	int server_fd, max_fd;
938	int accepting;
939	timeval tv;
940	fd_set fds;
941
942	fd = open(PATH_DEVCTL, O_RDONLY | O_CLOEXEC);
943	if (fd == -1)
944		err(1, "Can't open devctl device %s", PATH_DEVCTL);
945	server_fd = create_socket(PIPE);
946	accepting = 1;
947	max_fd = max(fd, server_fd) + 1;
948	while (!romeo_must_die) {
949		if (!once && !no_daemon && !daemonize_quick) {
950			// Check to see if we have any events pending.
951			tv.tv_sec = 0;
952			tv.tv_usec = 0;
953			FD_ZERO(&fds);
954			FD_SET(fd, &fds);
955			rv = select(fd + 1, &fds, &fds, &fds, &tv);
956			// No events -> we've processed all pending events
957			if (rv == 0) {
958				devdlog(LOG_DEBUG, "Calling daemon\n");
959				cfg.remove_pidfile();
960				cfg.open_pidfile();
961				daemon(0, 0);
962				cfg.write_pidfile();
963				once++;
964			}
965		}
966		/*
967		 * When we've already got the max number of clients, stop
968		 * accepting new connections (don't put server_fd in the set),
969		 * shrink the accept() queue to reject connections quickly, and
970		 * poll the existing clients more often, so that we notice more
971		 * quickly when any of them disappear to free up client slots.
972		 */
973		FD_ZERO(&fds);
974		FD_SET(fd, &fds);
975		if (num_clients < max_clients) {
976			if (!accepting) {
977				listen(server_fd, max_clients);
978				accepting = 1;
979			}
980			FD_SET(server_fd, &fds);
981			tv.tv_sec = 60;
982			tv.tv_usec = 0;
983		} else {
984			if (accepting) {
985				listen(server_fd, 0);
986				accepting = 0;
987			}
988			tv.tv_sec = 2;
989			tv.tv_usec = 0;
990		}
991		rv = select(max_fd, &fds, NULL, NULL, &tv);
992		if (got_siginfo) {
993			devdlog(LOG_NOTICE, "Events received so far=%u\n",
994			    total_events);
995			got_siginfo = 0;
996		}
997		if (rv == -1) {
998			if (errno == EINTR)
999				continue;
1000			err(1, "select");
1001		} else if (rv == 0)
1002			check_clients();
1003		if (FD_ISSET(fd, &fds)) {
1004			rv = read(fd, buffer, sizeof(buffer) - 1);
1005			if (rv > 0) {
1006				total_events++;
1007				if (rv == sizeof(buffer) - 1) {
1008					devdlog(LOG_WARNING, "Warning: "
1009					    "available event data exceeded "
1010					    "buffer space\n");
1011				}
1012				notify_clients(buffer, rv);
1013				buffer[rv] = '\0';
1014				while (buffer[--rv] == '\n')
1015					buffer[rv] = '\0';
1016				process_event(buffer);
1017			} else if (rv < 0) {
1018				if (errno != EINTR)
1019					break;
1020			} else {
1021				/* EOF */
1022				break;
1023			}
1024		}
1025		if (FD_ISSET(server_fd, &fds))
1026			new_client(server_fd);
1027	}
1028	close(fd);
1029}
1030
1031/*
1032 * functions that the parser uses.
1033 */
1034void
1035add_attach(int prio, event_proc *p)
1036{
1037	cfg.add_attach(prio, p);
1038}
1039
1040void
1041add_detach(int prio, event_proc *p)
1042{
1043	cfg.add_detach(prio, p);
1044}
1045
1046void
1047add_directory(const char *dir)
1048{
1049	cfg.add_directory(dir);
1050	free(const_cast<char *>(dir));
1051}
1052
1053void
1054add_nomatch(int prio, event_proc *p)
1055{
1056	cfg.add_nomatch(prio, p);
1057}
1058
1059void
1060add_notify(int prio, event_proc *p)
1061{
1062	cfg.add_notify(prio, p);
1063}
1064
1065event_proc *
1066add_to_event_proc(event_proc *ep, eps *eps)
1067{
1068	if (ep == NULL)
1069		ep = new event_proc();
1070	ep->add(eps);
1071	return (ep);
1072}
1073
1074eps *
1075new_action(const char *cmd)
1076{
1077	eps *e = new action(cmd);
1078	free(const_cast<char *>(cmd));
1079	return (e);
1080}
1081
1082eps *
1083new_match(const char *var, const char *re)
1084{
1085	eps *e = new match(cfg, var, re);
1086	free(const_cast<char *>(var));
1087	free(const_cast<char *>(re));
1088	return (e);
1089}
1090
1091eps *
1092new_media(const char *var, const char *re)
1093{
1094	eps *e = new media(cfg, var, re);
1095	free(const_cast<char *>(var));
1096	free(const_cast<char *>(re));
1097	return (e);
1098}
1099
1100void
1101set_pidfile(const char *name)
1102{
1103	cfg.set_pidfile(name);
1104	free(const_cast<char *>(name));
1105}
1106
1107void
1108set_variable(const char *var, const char *val)
1109{
1110	cfg.set_variable(var, val);
1111	free(const_cast<char *>(var));
1112	free(const_cast<char *>(val));
1113}
1114
1115
1116
1117static void
1118gensighand(int)
1119{
1120	romeo_must_die = 1;
1121}
1122
1123/*
1124 * SIGINFO handler.  Will print useful statistics to the syslog or stderr
1125 * as appropriate
1126 */
1127static void
1128siginfohand(int)
1129{
1130	got_siginfo = 1;
1131}
1132
1133/*
1134 * Local logging function.  Prints to syslog if we're daemonized; stderr
1135 * otherwise.
1136 */
1137static void
1138devdlog(int priority, const char* fmt, ...)
1139{
1140	va_list argp;
1141
1142	va_start(argp, fmt);
1143	if (no_daemon)
1144		vfprintf(stderr, fmt, argp);
1145	else if ((! quiet_mode) || (priority <= LOG_WARNING))
1146		vsyslog(priority, fmt, argp);
1147	va_end(argp);
1148}
1149
1150static void
1151usage()
1152{
1153	fprintf(stderr, "usage: %s [-dnq] [-l connlimit] [-f file]\n",
1154	    getprogname());
1155	exit(1);
1156}
1157
1158static void
1159check_devd_enabled()
1160{
1161	int val = 0;
1162	size_t len;
1163
1164	len = sizeof(val);
1165	if (sysctlbyname(SYSCTL, &val, &len, NULL, 0) != 0)
1166		errx(1, "devctl sysctl missing from kernel!");
1167	if (val == 0) {
1168		warnx("Setting " SYSCTL " to 1000");
1169		val = 1000;
1170		sysctlbyname(SYSCTL, NULL, NULL, &val, sizeof(val));
1171	}
1172}
1173
1174/*
1175 * main
1176 */
1177int
1178main(int argc, char **argv)
1179{
1180	int ch;
1181
1182	check_devd_enabled();
1183	while ((ch = getopt(argc, argv, "df:l:nq")) != -1) {
1184		switch (ch) {
1185		case 'd':
1186			no_daemon = 1;
1187			break;
1188		case 'f':
1189			configfile = optarg;
1190			break;
1191		case 'l':
1192			max_clients = MAX(1, strtoul(optarg, NULL, 0));
1193			break;
1194		case 'n':
1195			daemonize_quick = 1;
1196			break;
1197		case 'q':
1198			quiet_mode = 1;
1199			break;
1200		default:
1201			usage();
1202		}
1203	}
1204
1205	cfg.parse();
1206	if (!no_daemon && daemonize_quick) {
1207		cfg.open_pidfile();
1208		daemon(0, 0);
1209		cfg.write_pidfile();
1210	}
1211	signal(SIGPIPE, SIG_IGN);
1212	signal(SIGHUP, gensighand);
1213	signal(SIGINT, gensighand);
1214	signal(SIGTERM, gensighand);
1215	signal(SIGINFO, siginfohand);
1216	event_loop();
1217	return (0);
1218}
1219