1/*-
2 * SPDX-License-Identifier: BSD-3-Clause AND BSD-2-Clause
3 *
4 * Copyright (c) 2002-2010 M. Warner Losh <imp@FreeBSD.org>
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 *
27 * my_system is a variation on lib/libc/stdlib/system.c:
28 *
29 * Copyright (c) 1988, 1993
30 *	The Regents of the University of California.  All rights reserved.
31 *
32 * Redistribution and use in source and binary forms, with or without
33 * modification, are permitted provided that the following conditions
34 * are met:
35 * 1. Redistributions of source code must retain the above copyright
36 *    notice, this list of conditions and the following disclaimer.
37 * 2. Redistributions in binary form must reproduce the above copyright
38 *    notice, this list of conditions and the following disclaimer in the
39 *    documentation and/or other materials provided with the distribution.
40 * 3. Neither the name of the University nor the names of its contributors
41 *    may be used to endorse or promote products derived from this software
42 *    without specific prior written permission.
43 *
44 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
45 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
46 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
47 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
48 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
49 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
50 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
51 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
52 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
53 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
54 * SUCH DAMAGE.
55 */
56
57/*
58 * DEVD control daemon.
59 */
60
61// TODO list:
62//	o devd.conf and devd man pages need a lot of help:
63//	  - devd needs to document the unix domain socket
64//	  - devd.conf needs more details on the supported statements.
65
66#include <sys/param.h>
67#include <sys/socket.h>
68#include <sys/stat.h>
69#include <sys/sysctl.h>
70#include <sys/types.h>
71#include <sys/wait.h>
72#include <sys/un.h>
73
74#include <cctype>
75#include <cerrno>
76#include <cstdlib>
77#include <cstdio>
78#include <csignal>
79#include <cstring>
80#include <cstdarg>
81
82#include <dirent.h>
83#include <err.h>
84#include <fcntl.h>
85#include <libutil.h>
86#include <paths.h>
87#include <poll.h>
88#include <regex.h>
89#include <syslog.h>
90#include <unistd.h>
91
92#include <algorithm>
93#include <map>
94#include <string>
95#include <list>
96#include <stdexcept>
97#include <vector>
98
99#include "devd.h"		/* C compatible definitions */
100#include "devd.hh"		/* C++ class definitions */
101
102#define STREAMPIPE "/var/run/devd.pipe"
103#define SEQPACKETPIPE "/var/run/devd.seqpacket.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 increase 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.  The kernel
114 * allocates a 4608B mbuf for each message.  Modern technology places a limit of
115 * roughly 450 drives/rack, and it's unlikely that a zpool will ever be larger
116 * than that.
117 *
118 * 450 drives * 165 bytes / drive = 74250B of data in the sockbuf
119 * 450 drives * 4608B / drive = 2073600B of mbufs in the sockbuf
120 *
121 * We can't directly set the sockbuf's mbuf limit, but we can do it indirectly.
122 * The kernel sets it to the minimum of a hard-coded maximum value and sbcc *
123 * kern.ipc.sockbuf_waste_factor, where sbcc is the socket buffer size set by
124 * the user.  The default value of kern.ipc.sockbuf_waste_factor is 8.  If we
125 * set the bufsize to 256k and use the kern.ipc.sockbuf_waste_factor, then the
126 * kernel will set the mbuf limit to 2MB, which is just large enough for 450
127 * drives.  It also happens to be the same as the hardcoded maximum value.
128 */
129#define CLIENT_BUFSIZE 262144
130
131using namespace std;
132
133typedef struct client {
134	int fd;
135	int socktype;
136} client_t;
137
138extern FILE *yyin;
139
140static const char notify = '!';
141static const char nomatch = '?';
142static const char attach = '+';
143static const char detach = '-';
144
145static struct pidfh *pfh;
146
147static int no_daemon = 0;
148static int daemonize_quick = 0;
149static int quiet_mode = 0;
150static unsigned total_events = 0;
151static volatile sig_atomic_t got_siginfo = 0;
152static volatile sig_atomic_t romeo_must_die = 0;
153
154static const char *configfile = CF;
155
156static void devdlog(int priority, const char* message, ...)
157	__printflike(2, 3);
158static void event_loop(void);
159static void usage(void) __dead2;
160
161template <class T> void
162delete_and_clear(vector<T *> &v)
163{
164	typename vector<T *>::const_iterator i;
165
166	for (i = v.begin(); i != v.end(); ++i)
167		delete *i;
168	v.clear();
169}
170
171static config cfg;
172
173static const char *curr_cf = NULL;
174
175event_proc::event_proc() : _prio(-1)
176{
177	_epsvec.reserve(4);
178}
179
180event_proc::~event_proc()
181{
182	delete_and_clear(_epsvec);
183}
184
185void
186event_proc::add(eps *eps)
187{
188	_epsvec.push_back(eps);
189}
190
191bool
192event_proc::matches(config &c) const
193{
194	vector<eps *>::const_iterator i;
195
196	for (i = _epsvec.begin(); i != _epsvec.end(); ++i)
197		if (!(*i)->do_match(c))
198			return (false);
199	return (true);
200}
201
202bool
203event_proc::run(config &c) const
204{
205	vector<eps *>::const_iterator i;
206
207	for (i = _epsvec.begin(); i != _epsvec.end(); ++i)
208		if (!(*i)->do_action(c))
209			return (false);
210	return (true);
211}
212
213action::action(const char *cmd)
214	: _cmd(cmd)
215{
216	// nothing
217}
218
219action::~action()
220{
221	// nothing
222}
223
224static int
225my_system(const char *command)
226{
227	pid_t pid, savedpid;
228	int pstat;
229	struct sigaction ign, intact, quitact;
230	sigset_t newsigblock, oldsigblock;
231
232	if (!command)		/* just checking... */
233		return (1);
234
235	/*
236	 * Ignore SIGINT and SIGQUIT, block SIGCHLD. Remember to save
237	 * existing signal dispositions.
238	 */
239	ign.sa_handler = SIG_IGN;
240	::sigemptyset(&ign.sa_mask);
241	ign.sa_flags = 0;
242	::sigaction(SIGINT, &ign, &intact);
243	::sigaction(SIGQUIT, &ign, &quitact);
244	::sigemptyset(&newsigblock);
245	::sigaddset(&newsigblock, SIGCHLD);
246	::sigprocmask(SIG_BLOCK, &newsigblock, &oldsigblock);
247	switch (pid = ::fork()) {
248	case -1:			/* error */
249		break;
250	case 0:				/* child */
251		/*
252		 * Restore original signal dispositions and exec the command.
253		 */
254		::sigaction(SIGINT, &intact, NULL);
255		::sigaction(SIGQUIT,  &quitact, NULL);
256		::sigprocmask(SIG_SETMASK, &oldsigblock, NULL);
257		/*
258		 * Close the PID file, and all other open descriptors.
259		 * Inherit std{in,out,err} only.
260		 */
261		cfg.close_pidfile();
262		::closefrom(3);
263		::execl(_PATH_BSHELL, "sh", "-c", command, (char *)NULL);
264		::_exit(127);
265	default:			/* parent */
266		savedpid = pid;
267		do {
268			pid = ::wait4(savedpid, &pstat, 0, (struct rusage *)0);
269		} while (pid == -1 && errno == EINTR);
270		break;
271	}
272	::sigaction(SIGINT, &intact, NULL);
273	::sigaction(SIGQUIT,  &quitact, NULL);
274	::sigprocmask(SIG_SETMASK, &oldsigblock, NULL);
275	return (pid == -1 ? -1 : pstat);
276}
277
278bool
279action::do_action(config &c)
280{
281	string s = c.expand_string(_cmd.c_str());
282	devdlog(LOG_INFO, "Executing '%s'\n", s.c_str());
283	my_system(s.c_str());
284	return (true);
285}
286
287match::match(config &c, const char *var, const char *re) :
288	_inv(re[0] == '!'),
289	_var(var),
290	_re(c.expand_string(_inv ? re + 1 : re, "^", "$"))
291{
292	regcomp(&_regex, _re.c_str(), REG_EXTENDED | REG_NOSUB | REG_ICASE);
293}
294
295match::~match()
296{
297	regfree(&_regex);
298}
299
300bool
301match::do_match(config &c)
302{
303	const string &value = c.get_variable(_var);
304	bool retval;
305
306	/*
307	 * This function gets called WAY too often to justify calling syslog()
308	 * each time, even at LOG_DEBUG.  Because if syslogd isn't running, it
309	 * can consume excessive amounts of systime inside of connect().  Only
310	 * log when we're in -d mode.
311	 */
312	if (no_daemon) {
313		devdlog(LOG_DEBUG, "Testing %s=%s against %s, invert=%d\n",
314		    _var.c_str(), value.c_str(), _re.c_str(), _inv);
315	}
316
317	retval = (regexec(&_regex, value.c_str(), 0, NULL, 0) == 0);
318	if (_inv == 1)
319		retval = (retval == 0) ? 1 : 0;
320
321	return (retval);
322}
323
324#include <sys/sockio.h>
325#include <net/if.h>
326#include <net/if_media.h>
327
328media::media(config &, const char *var, const char *type)
329	: _var(var), _type(-1)
330{
331	static struct ifmedia_description media_types[] = {
332		{ IFM_ETHER,		"Ethernet" },
333		{ IFM_IEEE80211,	"802.11" },
334		{ IFM_ATM,		"ATM" },
335		{ -1,			"unknown" },
336		{ 0, NULL },
337	};
338	for (int i = 0; media_types[i].ifmt_string != NULL; ++i)
339		if (strcasecmp(type, media_types[i].ifmt_string) == 0) {
340			_type = media_types[i].ifmt_word;
341			break;
342		}
343}
344
345media::~media()
346{
347}
348
349bool
350media::do_match(config &c)
351{
352	string value;
353	struct ifmediareq ifmr;
354	bool retval;
355	int s;
356
357	// Since we can be called from both a device attach/detach
358	// context where device-name is defined and what we want,
359	// as well as from a link status context, where subsystem is
360	// the name of interest, first try device-name and fall back
361	// to subsystem if none exists.
362	value = c.get_variable("device-name");
363	if (value.empty())
364		value = c.get_variable("subsystem");
365	devdlog(LOG_DEBUG, "Testing media type of %s against 0x%x\n",
366		    value.c_str(), _type);
367
368	retval = false;
369
370	s = socket(PF_INET, SOCK_DGRAM, 0);
371	if (s >= 0) {
372		memset(&ifmr, 0, sizeof(ifmr));
373		strlcpy(ifmr.ifm_name, value.c_str(), sizeof(ifmr.ifm_name));
374
375		if (ioctl(s, SIOCGIFMEDIA, (caddr_t)&ifmr) >= 0 &&
376		    ifmr.ifm_status & IFM_AVALID) {
377			devdlog(LOG_DEBUG, "%s has media type 0x%x\n",
378				    value.c_str(), IFM_TYPE(ifmr.ifm_active));
379			retval = (IFM_TYPE(ifmr.ifm_active) == _type);
380		} else if (_type == -1) {
381			devdlog(LOG_DEBUG, "%s has unknown media type\n",
382				    value.c_str());
383			retval = true;
384		}
385		close(s);
386	}
387
388	return (retval);
389}
390
391const string var_list::bogus = "_$_$_$_$_B_O_G_U_S_$_$_$_$_";
392const string var_list::nothing = "";
393
394const string &
395var_list::get_variable(const string &var) const
396{
397	map<string, string>::const_iterator i;
398
399	i = _vars.find(var);
400	if (i == _vars.end())
401		return (var_list::bogus);
402	return (i->second);
403}
404
405bool
406var_list::is_set(const string &var) const
407{
408	return (_vars.find(var) != _vars.end());
409}
410
411/** fix_value
412 *
413 * Removes quoted characters that have made it this far. \" are
414 * converted to ". For all other characters, both \ and following
415 * character. So the string 'fre\:\"' is translated to 'fred\:"'.
416 */
417std::string
418var_list::fix_value(const std::string &val) const
419{
420        std::string rv(val);
421        std::string::size_type pos(0);
422
423        while ((pos = rv.find("\\\"", pos)) != rv.npos) {
424                rv.erase(pos, 1);
425        }
426        return (rv);
427}
428
429void
430var_list::set_variable(const string &var, const string &val)
431{
432	/*
433	 * This function gets called WAY too often to justify calling syslog()
434	 * each time, even at LOG_DEBUG.  Because if syslogd isn't running, it
435	 * can consume excessive amounts of systime inside of connect().  Only
436	 * log when we're in -d mode.
437	 */
438	_vars[var] = fix_value(val);
439	if (no_daemon)
440		devdlog(LOG_DEBUG, "setting %s=%s\n", var.c_str(), val.c_str());
441}
442
443void
444config::reset(void)
445{
446	_dir_list.clear();
447	delete_and_clear(_var_list_table);
448	delete_and_clear(_attach_list);
449	delete_and_clear(_detach_list);
450	delete_and_clear(_nomatch_list);
451	delete_and_clear(_notify_list);
452}
453
454/*
455 * Called recursively as new files are included, so current stack of old names
456 * saved in each instance of 'old' on the call stack. Called single threaded
457 * so global varaibles curr_cf and lineno (and all of yacc's parser state)
458 * are safe to access w/o a lock.
459 */
460void
461config::parse_one_file(const char *fn)
462{
463	const char *old;
464
465	devdlog(LOG_DEBUG, "Parsing %s\n", fn);
466	yyin = fopen(fn, "r");
467	old = curr_cf;
468	curr_cf = fn;
469	if (yyin == NULL)
470		err(1, "Cannot open config file %s", fn);
471	lineno = 1;
472	if (yyparse() != 0)
473		errx(1, "Cannot parse %s at line %d", fn, lineno);
474	fclose(yyin);
475	curr_cf = old;
476}
477
478void
479config::parse_files_in_dir(const char *dirname)
480{
481	DIR *dirp;
482	struct dirent *dp;
483	char path[PATH_MAX];
484
485	devdlog(LOG_DEBUG, "Parsing files in %s\n", dirname);
486	dirp = opendir(dirname);
487	if (dirp == NULL)
488		return;
489	readdir(dirp);		/* Skip . */
490	readdir(dirp);		/* Skip .. */
491	while ((dp = readdir(dirp)) != NULL) {
492		if (strcmp(dp->d_name + dp->d_namlen - 5, ".conf") == 0) {
493			snprintf(path, sizeof(path), "%s/%s",
494			    dirname, dp->d_name);
495			parse_one_file(path);
496		}
497	}
498	closedir(dirp);
499}
500
501class epv_greater {
502public:
503	int operator()(event_proc *const&l1, event_proc *const&l2) const
504	{
505		return (l1->get_priority() > l2->get_priority());
506	}
507};
508
509void
510config::sort_vector(vector<event_proc *> &v)
511{
512	stable_sort(v.begin(), v.end(), epv_greater());
513}
514
515void
516config::parse(void)
517{
518	vector<string>::const_iterator i;
519
520	parse_one_file(configfile);
521	for (i = _dir_list.begin(); i != _dir_list.end(); ++i)
522		parse_files_in_dir((*i).c_str());
523	sort_vector(_attach_list);
524	sort_vector(_detach_list);
525	sort_vector(_nomatch_list);
526	sort_vector(_notify_list);
527}
528
529void
530config::open_pidfile()
531{
532	pid_t otherpid;
533
534	if (_pidfile.empty())
535		return;
536	pfh = pidfile_open(_pidfile.c_str(), 0600, &otherpid);
537	if (pfh == NULL) {
538		if (errno == EEXIST)
539			errx(1, "devd already running, pid: %d", (int)otherpid);
540		warn("cannot open pid file");
541	}
542}
543
544void
545config::write_pidfile()
546{
547
548	pidfile_write(pfh);
549}
550
551void
552config::close_pidfile()
553{
554
555	pidfile_close(pfh);
556}
557
558void
559config::remove_pidfile()
560{
561
562	pidfile_remove(pfh);
563}
564
565void
566config::add_attach(int prio, event_proc *p)
567{
568	p->set_priority(prio);
569	_attach_list.push_back(p);
570}
571
572void
573config::add_detach(int prio, event_proc *p)
574{
575	p->set_priority(prio);
576	_detach_list.push_back(p);
577}
578
579void
580config::add_directory(const char *dir)
581{
582	_dir_list.push_back(string(dir));
583}
584
585void
586config::add_nomatch(int prio, event_proc *p)
587{
588	p->set_priority(prio);
589	_nomatch_list.push_back(p);
590}
591
592void
593config::add_notify(int prio, event_proc *p)
594{
595	p->set_priority(prio);
596	_notify_list.push_back(p);
597}
598
599void
600config::set_pidfile(const char *fn)
601{
602	_pidfile = fn;
603}
604
605void
606config::push_var_table()
607{
608	var_list *vl;
609
610	vl = new var_list();
611	_var_list_table.push_back(vl);
612	devdlog(LOG_DEBUG, "Pushing table\n");
613}
614
615void
616config::pop_var_table()
617{
618	delete _var_list_table.back();
619	_var_list_table.pop_back();
620	devdlog(LOG_DEBUG, "Popping table\n");
621}
622
623void
624config::set_variable(const char *var, const char *val)
625{
626	_var_list_table.back()->set_variable(var, val);
627}
628
629const string &
630config::get_variable(const string &var)
631{
632	vector<var_list *>::reverse_iterator i;
633
634	for (i = _var_list_table.rbegin(); i != _var_list_table.rend(); ++i) {
635		if ((*i)->is_set(var))
636			return ((*i)->get_variable(var));
637	}
638	return (var_list::nothing);
639}
640
641bool
642config::is_id_char(char ch) const
643{
644	return (ch != '\0' && (isalpha(ch) || isdigit(ch) || ch == '_' ||
645	    ch == '-'));
646}
647
648string
649config::shell_quote(const string &s)
650{
651	string buffer;
652	const char *cs, *ce;
653	char c;
654
655	/*
656	 * Enclose the string in $' ' with escapes for ' and / characters making
657	 * it one argument and ensuring the shell won't be affected by its
658	 * usual list of candidates.
659	 */
660	buffer.reserve(s.length() * 3 / 2);
661	buffer += '$';
662	buffer += '\'';
663	cs = s.c_str();
664	ce = cs + strlen(cs);
665	for (; cs < ce; cs++) {
666		c = *cs;
667		if (c == '\'' || c == '\\') {
668			buffer += '\\';
669		}
670		buffer += c;
671	}
672	buffer += '\'';
673
674	return buffer;
675}
676
677void
678config::expand_one(const char *&src, string &dst, bool is_shell)
679{
680	int count;
681	string buffer;
682
683	src++;
684	// $$ -> $
685	if (*src == '$') {
686		dst += *src++;
687		return;
688	}
689
690	// $(foo) -> $(foo)
691	// This is the escape hatch for passing down shell subcommands
692	if (*src == '(') {
693		dst += '$';
694		count = 0;
695		/* If the string ends before ) is matched , return. */
696		do {
697			if (*src == ')')
698				count--;
699			else if (*src == '(')
700				count++;
701			dst += *src++;
702		} while (count > 0 && *src);
703		return;
704	}
705
706	// $[^-A-Za-z_*] -> $\1
707	if (!isalpha(*src) && *src != '_' && *src != '-' && *src != '*') {
708		dst += '$';
709		dst += *src++;
710		return;
711	}
712
713	// $var -> replace with value
714	do {
715		buffer += *src++;
716	} while (is_id_char(*src));
717	dst.append(is_shell ? shell_quote(get_variable(buffer)) : get_variable(buffer));
718}
719
720const string
721config::expand_string(const char *src, const char *prepend, const char *append)
722{
723	const char *var_at;
724	string dst;
725
726	/*
727	 * 128 bytes is enough for 2427 of 2438 expansions that happen
728	 * while parsing config files, as tested on 2013-01-30.
729	 */
730	dst.reserve(128);
731
732	if (prepend != NULL)
733		dst = prepend;
734
735	for (;;) {
736		var_at = strchr(src, '$');
737		if (var_at == NULL) {
738			dst.append(src);
739			break;
740		}
741		dst.append(src, var_at - src);
742		src = var_at;
743		expand_one(src, dst, prepend == NULL);
744	}
745
746	if (append != NULL)
747		dst.append(append);
748
749	return (dst);
750}
751
752bool
753config::chop_var(char *&buffer, char *&lhs, char *&rhs) const
754{
755	char *walker;
756
757	if (*buffer == '\0')
758		return (false);
759	walker = lhs = buffer;
760	while (is_id_char(*walker))
761		walker++;
762	if (*walker != '=')
763		return (false);
764	walker++;		// skip =
765	if (*walker == '"') {
766		walker++;	// skip "
767		rhs = walker;
768		while (*walker && *walker != '"') {
769			// Skip \" ... We leave it in the string and strip the \ later.
770			// due to the super simplistic parser that we have here.
771			if (*walker == '\\' && walker[1] == '"')
772				walker++;
773			walker++;
774		}
775		if (*walker != '"')
776			return (false);
777		rhs[-2] = '\0';
778		*walker++ = '\0';
779	} else {
780		rhs = walker;
781		while (*walker && !isspace(*walker))
782			walker++;
783		if (*walker != '\0')
784			*walker++ = '\0';
785		rhs[-1] = '\0';
786	}
787	while (isspace(*walker))
788		walker++;
789	buffer = walker;
790	return (true);
791}
792
793
794char *
795config::set_vars(char *buffer)
796{
797	char *lhs;
798	char *rhs;
799
800	while (1) {
801		if (!chop_var(buffer, lhs, rhs))
802			break;
803		set_variable(lhs, rhs);
804	}
805	return (buffer);
806}
807
808void
809config::find_and_execute(char type)
810{
811	vector<event_proc *> *l;
812	vector<event_proc *>::const_iterator i;
813	const char *s;
814
815	switch (type) {
816	default:
817		return;
818	case notify:
819		l = &_notify_list;
820		s = "notify";
821		break;
822	case nomatch:
823		l = &_nomatch_list;
824		s = "nomatch";
825		break;
826	case attach:
827		l = &_attach_list;
828		s = "attach";
829		break;
830	case detach:
831		l = &_detach_list;
832		s = "detach";
833		break;
834	}
835	devdlog(LOG_DEBUG, "Processing %s event\n", s);
836	for (i = l->begin(); i != l->end(); ++i) {
837		if ((*i)->matches(*this)) {
838			(*i)->run(*this);
839			break;
840		}
841	}
842
843}
844
845
846static void
847process_event(char *buffer)
848{
849	char type;
850	char *sp;
851	struct timeval tv;
852	char *timestr;
853
854	sp = buffer + 1;
855	devdlog(LOG_INFO, "Processing event '%s'\n", buffer);
856	type = *buffer++;
857	cfg.push_var_table();
858	// $* is the entire line
859	cfg.set_variable("*", buffer - 1);
860	// $_ is the entire line without the initial character
861	cfg.set_variable("_", buffer);
862
863	// Save the time this happened (as approximated by when we got
864	// around to processing it).
865	gettimeofday(&tv, NULL);
866	asprintf(&timestr, "%jd.%06ld", (uintmax_t)tv.tv_sec, tv.tv_usec);
867	cfg.set_variable("timestamp", timestr);
868	free(timestr);
869
870	// Match doesn't have a device, and the format is a little
871	// different, so handle it separately.
872	switch (type) {
873	case notify:
874		//! (k=v)*
875		sp = cfg.set_vars(sp);
876		break;
877	case nomatch:
878		//? at location pnp-info on bus
879		sp = strchr(sp, ' ');
880		if (sp == NULL)
881			return;	/* Can't happen? */
882		*sp++ = '\0';
883		while (isspace(*sp))
884			sp++;
885		if (strncmp(sp, "at ", 3) == 0)
886			sp += 3;
887		sp = cfg.set_vars(sp);
888		while (isspace(*sp))
889			sp++;
890		if (strncmp(sp, "on ", 3) == 0)
891			cfg.set_variable("bus", sp + 3);
892		break;
893	case attach:	/*FALLTHROUGH*/
894	case detach:
895		sp = strchr(sp, ' ');
896		if (sp == NULL)
897			return;	/* Can't happen? */
898		*sp++ = '\0';
899		cfg.set_variable("device-name", buffer);
900		while (isspace(*sp))
901			sp++;
902		if (strncmp(sp, "at ", 3) == 0)
903			sp += 3;
904		sp = cfg.set_vars(sp);
905		while (isspace(*sp))
906			sp++;
907		if (strncmp(sp, "on ", 3) == 0)
908			cfg.set_variable("bus", sp + 3);
909		break;
910	}
911
912	cfg.find_and_execute(type);
913	cfg.pop_var_table();
914}
915
916static int
917create_socket(const char *name, int socktype)
918{
919	int fd, slen;
920	struct sockaddr_un sun;
921
922	if ((fd = socket(PF_LOCAL, socktype, 0)) < 0)
923		err(1, "socket");
924	bzero(&sun, sizeof(sun));
925	sun.sun_family = AF_UNIX;
926	strlcpy(sun.sun_path, name, sizeof(sun.sun_path));
927	slen = SUN_LEN(&sun);
928	unlink(name);
929	if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0)
930	    	err(1, "fcntl");
931	if (::bind(fd, (struct sockaddr *) & sun, slen) < 0)
932		err(1, "bind");
933	listen(fd, 4);
934	if (chown(name, 0, 0))	/* XXX - root.wheel */
935		err(1, "chown");
936	if (chmod(name, 0666))
937		err(1, "chmod");
938	return (fd);
939}
940
941static unsigned int max_clients = 10;	/* Default, can be overridden on cmdline. */
942static unsigned int num_clients;
943
944static list<client_t> clients;
945
946static void
947notify_clients(const char *data, int len)
948{
949	list<client_t>::iterator i;
950
951	/*
952	 * Deliver the data to all clients.  Throw clients overboard at the
953	 * first sign of trouble.  This reaps clients who've died or closed
954	 * their sockets, and also clients who are alive but failing to keep up
955	 * (or who are maliciously not reading, to consume buffer space in
956	 * kernel memory or tie up the limited number of available connections).
957	 */
958	for (i = clients.begin(); i != clients.end(); ) {
959		int flags;
960		if (i->socktype == SOCK_SEQPACKET)
961			flags = MSG_EOR;
962		else
963			flags = 0;
964
965		if (send(i->fd, data, len, flags) != len) {
966			--num_clients;
967			close(i->fd);
968			i = clients.erase(i);
969			devdlog(LOG_WARNING, "notify_clients: send() failed; "
970			    "dropping unresponsive client\n");
971		} else
972			++i;
973	}
974}
975
976static void
977check_clients(void)
978{
979	int s;
980	struct pollfd pfd;
981	list<client_t>::iterator i;
982
983	/*
984	 * Check all existing clients to see if any of them have disappeared.
985	 * Normally we reap clients when we get an error trying to send them an
986	 * event.  This check eliminates the problem of an ever-growing list of
987	 * zombie clients because we're never writing to them on a system
988	 * without frequent device-change activity.
989	 */
990	pfd.events = 0;
991	for (i = clients.begin(); i != clients.end(); ) {
992		pfd.fd = i->fd;
993		s = poll(&pfd, 1, 0);
994		if ((s < 0 && s != EINTR ) ||
995		    (s > 0 && (pfd.revents & POLLHUP))) {
996			--num_clients;
997			close(i->fd);
998			i = clients.erase(i);
999			devdlog(LOG_NOTICE, "check_clients:  "
1000			    "dropping disconnected client\n");
1001		} else
1002			++i;
1003	}
1004}
1005
1006static void
1007new_client(int fd, int socktype)
1008{
1009	client_t s;
1010	int sndbuf_size;
1011
1012	/*
1013	 * First go reap any zombie clients, then accept the connection, and
1014	 * shut down the read side to stop clients from consuming kernel memory
1015	 * by sending large buffers full of data we'll never read.
1016	 */
1017	check_clients();
1018	s.socktype = socktype;
1019	s.fd = accept(fd, NULL, NULL);
1020	if (s.fd != -1) {
1021		sndbuf_size = CLIENT_BUFSIZE;
1022		if (setsockopt(s.fd, SOL_SOCKET, SO_SNDBUF, &sndbuf_size,
1023		    sizeof(sndbuf_size)))
1024			err(1, "setsockopt");
1025		shutdown(s.fd, SHUT_RD);
1026		clients.push_back(s);
1027		++num_clients;
1028	} else
1029		err(1, "accept");
1030}
1031
1032static void
1033event_loop(void)
1034{
1035	int rv;
1036	int fd;
1037	char buffer[DEVCTL_MAXBUF];
1038	int once = 0;
1039	int stream_fd, seqpacket_fd, max_fd;
1040	int accepting;
1041	timeval tv;
1042	fd_set fds;
1043
1044	fd = open(PATH_DEVCTL, O_RDONLY | O_CLOEXEC);
1045	if (fd == -1)
1046		err(1, "Can't open devctl device %s", PATH_DEVCTL);
1047	stream_fd = create_socket(STREAMPIPE, SOCK_STREAM);
1048	seqpacket_fd = create_socket(SEQPACKETPIPE, SOCK_SEQPACKET);
1049	accepting = 1;
1050	max_fd = max(fd, max(stream_fd, seqpacket_fd)) + 1;
1051	while (!romeo_must_die) {
1052		if (!once && !no_daemon && !daemonize_quick) {
1053			// Check to see if we have any events pending.
1054			tv.tv_sec = 0;
1055			tv.tv_usec = 0;
1056			FD_ZERO(&fds);
1057			FD_SET(fd, &fds);
1058			rv = select(fd + 1, &fds, NULL, NULL, &tv);
1059			// No events -> we've processed all pending events
1060			if (rv == 0) {
1061				devdlog(LOG_DEBUG, "Calling daemon\n");
1062				cfg.remove_pidfile();
1063				cfg.open_pidfile();
1064				daemon(0, 0);
1065				cfg.write_pidfile();
1066				once++;
1067			}
1068		}
1069		/*
1070		 * When we've already got the max number of clients, stop
1071		 * accepting new connections (don't put the listening sockets in
1072		 * the set), shrink the accept() queue to reject connections
1073		 * quickly, and poll the existing clients more often, so that we
1074		 * notice more quickly when any of them disappear to free up
1075		 * client slots.
1076		 */
1077		FD_ZERO(&fds);
1078		FD_SET(fd, &fds);
1079		if (num_clients < max_clients) {
1080			if (!accepting) {
1081				listen(stream_fd, max_clients);
1082				listen(seqpacket_fd, max_clients);
1083				accepting = 1;
1084			}
1085			FD_SET(stream_fd, &fds);
1086			FD_SET(seqpacket_fd, &fds);
1087			tv.tv_sec = 60;
1088			tv.tv_usec = 0;
1089		} else {
1090			if (accepting) {
1091				listen(stream_fd, 0);
1092				listen(seqpacket_fd, 0);
1093				accepting = 0;
1094			}
1095			tv.tv_sec = 2;
1096			tv.tv_usec = 0;
1097		}
1098		rv = select(max_fd, &fds, NULL, NULL, &tv);
1099		if (got_siginfo) {
1100			devdlog(LOG_NOTICE, "Events received so far=%u\n",
1101			    total_events);
1102			got_siginfo = 0;
1103		}
1104		if (rv == -1) {
1105			if (errno == EINTR)
1106				continue;
1107			err(1, "select");
1108		} else if (rv == 0)
1109			check_clients();
1110		if (FD_ISSET(fd, &fds)) {
1111			rv = read(fd, buffer, sizeof(buffer) - 1);
1112			if (rv > 0) {
1113				total_events++;
1114				if (rv == sizeof(buffer) - 1) {
1115					devdlog(LOG_WARNING, "Warning: "
1116					    "available event data exceeded "
1117					    "buffer space\n");
1118				}
1119				notify_clients(buffer, rv);
1120				buffer[rv] = '\0';
1121				while (buffer[--rv] == '\n')
1122					buffer[rv] = '\0';
1123				try {
1124					process_event(buffer);
1125				}
1126				catch (const std::length_error& e) {
1127					devdlog(LOG_ERR, "Dropping event %s "
1128					    "due to low memory", buffer);
1129				}
1130			} else if (rv < 0) {
1131				if (errno != EINTR)
1132					break;
1133			} else {
1134				/* EOF */
1135				break;
1136			}
1137		}
1138		if (FD_ISSET(stream_fd, &fds))
1139			new_client(stream_fd, SOCK_STREAM);
1140		/*
1141		 * Aside from the socket type, both sockets use the same
1142		 * protocol, so we can process clients the same way.
1143		 */
1144		if (FD_ISSET(seqpacket_fd, &fds))
1145			new_client(seqpacket_fd, SOCK_SEQPACKET);
1146	}
1147	cfg.remove_pidfile();
1148	close(seqpacket_fd);
1149	close(stream_fd);
1150	close(fd);
1151}
1152
1153/*
1154 * functions that the parser uses.
1155 */
1156void
1157add_attach(int prio, event_proc *p)
1158{
1159	cfg.add_attach(prio, p);
1160}
1161
1162void
1163add_detach(int prio, event_proc *p)
1164{
1165	cfg.add_detach(prio, p);
1166}
1167
1168void
1169add_directory(const char *dir)
1170{
1171	cfg.add_directory(dir);
1172	free(const_cast<char *>(dir));
1173}
1174
1175void
1176add_nomatch(int prio, event_proc *p)
1177{
1178	cfg.add_nomatch(prio, p);
1179}
1180
1181void
1182add_notify(int prio, event_proc *p)
1183{
1184	cfg.add_notify(prio, p);
1185}
1186
1187event_proc *
1188add_to_event_proc(event_proc *ep, eps *eps)
1189{
1190	if (ep == NULL)
1191		ep = new event_proc();
1192	ep->add(eps);
1193	return (ep);
1194}
1195
1196eps *
1197new_action(const char *cmd)
1198{
1199	eps *e = new action(cmd);
1200	free(const_cast<char *>(cmd));
1201	return (e);
1202}
1203
1204eps *
1205new_match(const char *var, const char *re)
1206{
1207	/*
1208	 * In FreeBSD 14, we changed the system=kern to system=kernel for the
1209	 * resume message to match all the other 'kernel' messages. Generate a
1210	 * warning for the life of 14.x that we've 'fixed' the file on the fly,
1211	 * but make it a fatal error in 15.x and newer.
1212	 */
1213	if (strcmp(var, "kern") == 0) {
1214#if __FreeBSD_version < 1500000
1215		devdlog(LOG_WARNING,
1216		    "Changing deprecated system='kern' to new name 'kernel' in %s line %d.",
1217		    curr_cf, lineno);
1218		free(const_cast<char *>(var));
1219		var = strdup("kernel");
1220#elif  __FreeBSD_version < 1600000
1221		errx(1, "Encountered deprecated system=\"kern\" rule in %s line %d",
1222		    curr_cf, lineno);
1223#else
1224#error "Remove this gross hack"
1225#endif
1226	}
1227
1228	eps *e = new match(cfg, var, re);
1229	free(const_cast<char *>(var));
1230	free(const_cast<char *>(re));
1231	return (e);
1232}
1233
1234eps *
1235new_media(const char *var, const char *re)
1236{
1237	eps *e = new media(cfg, var, re);
1238	free(const_cast<char *>(var));
1239	free(const_cast<char *>(re));
1240	return (e);
1241}
1242
1243void
1244set_pidfile(const char *name)
1245{
1246	cfg.set_pidfile(name);
1247	free(const_cast<char *>(name));
1248}
1249
1250void
1251set_variable(const char *var, const char *val)
1252{
1253	cfg.set_variable(var, val);
1254	free(const_cast<char *>(var));
1255	free(const_cast<char *>(val));
1256}
1257
1258
1259
1260static void
1261gensighand(int)
1262{
1263	romeo_must_die = 1;
1264}
1265
1266/*
1267 * SIGINFO handler.  Will print useful statistics to the syslog or stderr
1268 * as appropriate
1269 */
1270static void
1271siginfohand(int)
1272{
1273	got_siginfo = 1;
1274}
1275
1276/*
1277 * Local logging function.  Prints to syslog if we're daemonized; stderr
1278 * otherwise.
1279 */
1280static void
1281devdlog(int priority, const char* fmt, ...)
1282{
1283	va_list argp;
1284
1285	va_start(argp, fmt);
1286	if (no_daemon)
1287		vfprintf(stderr, fmt, argp);
1288	else if (quiet_mode == 0 || priority <= LOG_WARNING)
1289		vsyslog(priority, fmt, argp);
1290	va_end(argp);
1291}
1292
1293static void
1294usage()
1295{
1296	fprintf(stderr, "usage: %s [-dnq] [-l connlimit] [-f file]\n",
1297	    getprogname());
1298	exit(1);
1299}
1300
1301static void
1302check_devd_enabled()
1303{
1304	int val = 0;
1305	size_t len;
1306
1307	len = sizeof(val);
1308	if (sysctlbyname(SYSCTL, &val, &len, NULL, 0) != 0)
1309		errx(1, "devctl sysctl missing from kernel!");
1310	if (val == 0) {
1311		warnx("Setting " SYSCTL " to 1000");
1312		val = 1000;
1313		if (sysctlbyname(SYSCTL, NULL, NULL, &val, sizeof(val)))
1314			err(1, "sysctlbyname");
1315	}
1316}
1317
1318/*
1319 * main
1320 */
1321int
1322main(int argc, char **argv)
1323{
1324	int ch;
1325
1326	check_devd_enabled();
1327	while ((ch = getopt(argc, argv, "df:l:nq")) != -1) {
1328		switch (ch) {
1329		case 'd':
1330			no_daemon = 1;
1331			break;
1332		case 'f':
1333			configfile = optarg;
1334			break;
1335		case 'l':
1336			max_clients = MAX(1, strtoul(optarg, NULL, 0));
1337			break;
1338		case 'n':
1339			daemonize_quick = 1;
1340			break;
1341		case 'q':
1342			quiet_mode = 1;
1343			break;
1344		default:
1345			usage();
1346		}
1347	}
1348
1349	cfg.parse();
1350	if (!no_daemon && daemonize_quick) {
1351		cfg.open_pidfile();
1352		daemon(0, 0);
1353		cfg.write_pidfile();
1354	}
1355	signal(SIGPIPE, SIG_IGN);
1356	signal(SIGHUP, gensighand);
1357	signal(SIGINT, gensighand);
1358	signal(SIGTERM, gensighand);
1359	signal(SIGINFO, siginfohand);
1360	event_loop();
1361	return (0);
1362}
1363