2 * Copyright (c) 2002-2010 M. Warner Losh.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
26 * my_system is a variation on lib/libc/stdlib/system.c:
28 * Copyright (c) 1988, 1993
29 * The Regents of the University of California. All rights reserved.
31 * Redistribution and use in source and binary forms, with or without
32 * modification, are permitted provided that the following conditions
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.
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
55 * $FreeBSD: head/sbin/devd/devd.cc 262914 2014-03-07 23:30:48Z asomers $
59 * DEVD control daemon.
63 // o devd.conf and devd man pages need a lot of help:
64 // - devd needs to document the unix domain socket
65 // - devd.conf needs more details on the supported statements.
67 #include <sys/param.h>
68 #include <sys/socket.h>
70 #include <sys/sysctl.h>
71 #include <sys/types.h>
99 #include "devd.h" /* C compatible definitions */
100 #include "devd.hh" /* C++ class definitions */
102 #define PIPE "/var/run/devd.pipe"
103 #define CF "/etc/devd.conf"
104 #define SYSCTL "hw.bus.devctl_disable"
107 * Since the client socket is nonblocking, we must increase its send buffer to
108 * handle brief event storms. On FreeBSD, AF_UNIX sockets don't have a receive
109 * buffer, so the client can't increate the buffersize by itself.
111 * For example, when creating a ZFS pool, devd emits one 165 character
112 * resource.fs.zfs.statechange message for each vdev in the pool. A 64k
113 * buffer has enough space for almost 400 drives, which would be very large but
114 * not impossibly large pool. A 128k buffer has enough space for 794 drives,
115 * which is more than can fit in a rack with modern technology.
117 #define CLIENT_BUFSIZE 131072
124 static const char notify = '!';
125 static const char nomatch = '?';
126 static const char attach = '+';
127 static const char detach = '-';
129 static struct pidfh *pfh;
131 static int no_daemon = 0;
132 static int daemonize_quick = 0;
133 static int quiet_mode = 0;
134 static unsigned total_events = 0;
135 static volatile sig_atomic_t got_siginfo = 0;
136 static volatile sig_atomic_t romeo_must_die = 0;
138 static const char *configfile = CF;
140 static void devdlog(int priority, const char* message, ...)
142 static void event_loop(void);
143 static void usage(void);
145 template <class T> void
146 delete_and_clear(vector<T *> &v)
148 typename vector<T *>::const_iterator i;
150 for (i = v.begin(); i != v.end(); ++i)
157 event_proc::event_proc() : _prio(-1)
162 event_proc::~event_proc()
164 delete_and_clear(_epsvec);
168 event_proc::add(eps *eps)
170 _epsvec.push_back(eps);
174 event_proc::matches(config &c) const
176 vector<eps *>::const_iterator i;
178 for (i = _epsvec.begin(); i != _epsvec.end(); ++i)
179 if (!(*i)->do_match(c))
185 event_proc::run(config &c) const
187 vector<eps *>::const_iterator i;
189 for (i = _epsvec.begin(); i != _epsvec.end(); ++i)
190 if (!(*i)->do_action(c))
195 action::action(const char *cmd)
207 my_system(const char *command)
211 struct sigaction ign, intact, quitact;
212 sigset_t newsigblock, oldsigblock;
214 if (!command) /* just checking... */
218 * Ignore SIGINT and SIGQUIT, block SIGCHLD. Remember to save
219 * existing signal dispositions.
221 ign.sa_handler = SIG_IGN;
222 ::sigemptyset(&ign.sa_mask);
224 ::sigaction(SIGINT, &ign, &intact);
225 ::sigaction(SIGQUIT, &ign, &quitact);
226 ::sigemptyset(&newsigblock);
227 ::sigaddset(&newsigblock, SIGCHLD);
228 ::sigprocmask(SIG_BLOCK, &newsigblock, &oldsigblock);
229 switch (pid = ::fork()) {
234 * Restore original signal dispositions and exec the command.
236 ::sigaction(SIGINT, &intact, NULL);
237 ::sigaction(SIGQUIT, &quitact, NULL);
238 ::sigprocmask(SIG_SETMASK, &oldsigblock, NULL);
240 * Close the PID file, and all other open descriptors.
241 * Inherit std{in,out,err} only.
245 ::execl(_PATH_BSHELL, "sh", "-c", command, (char *)NULL);
247 default: /* parent */
250 pid = ::wait4(savedpid, &pstat, 0, (struct rusage *)0);
251 } while (pid == -1 && errno == EINTR);
254 ::sigaction(SIGINT, &intact, NULL);
255 ::sigaction(SIGQUIT, &quitact, NULL);
256 ::sigprocmask(SIG_SETMASK, &oldsigblock, NULL);
257 return (pid == -1 ? -1 : pstat);
261 action::do_action(config &c)
263 string s = c.expand_string(_cmd.c_str());
264 devdlog(LOG_INFO, "Executing '%s'\n", s.c_str());
265 my_system(s.c_str());
269 match::match(config &c, const char *var, const char *re) :
272 _re(c.expand_string(_inv ? re + 1 : re, "^", "$"))
274 regcomp(&_regex, _re.c_str(), REG_EXTENDED | REG_NOSUB | REG_ICASE);
283 match::do_match(config &c)
285 const string &value = c.get_variable(_var);
289 * This function gets called WAY too often to justify calling syslog()
290 * each time, even at LOG_DEBUG. Because if syslogd isn't running, it
291 * can consume excessive amounts of systime inside of connect(). Only
292 * log when we're in -d mode.
295 devdlog(LOG_DEBUG, "Testing %s=%s against %s, invert=%d\n",
296 _var.c_str(), value.c_str(), _re.c_str(), _inv);
299 retval = (regexec(&_regex, value.c_str(), 0, NULL, 0) == 0);
301 retval = (retval == 0) ? 1 : 0;
306 #include <sys/sockio.h>
308 #include <net/if_media.h>
310 media::media(config &, const char *var, const char *type)
311 : _var(var), _type(-1)
313 static struct ifmedia_description media_types[] = {
314 { IFM_ETHER, "Ethernet" },
315 { IFM_IEEE80211, "802.11" },
317 { IFM_CARP, "CARP" },
321 for (int i = 0; media_types[i].ifmt_string != NULL; ++i)
322 if (strcasecmp(type, media_types[i].ifmt_string) == 0) {
323 _type = media_types[i].ifmt_word;
333 media::do_match(config &c)
336 struct ifmediareq ifmr;
340 // Since we can be called from both a device attach/detach
341 // context where device-name is defined and what we want,
342 // as well as from a link status context, where subsystem is
343 // the name of interest, first try device-name and fall back
344 // to subsystem if none exists.
345 value = c.get_variable("device-name");
347 value = c.get_variable("subsystem");
348 devdlog(LOG_DEBUG, "Testing media type of %s against 0x%x\n",
349 value.c_str(), _type);
353 s = socket(PF_INET, SOCK_DGRAM, 0);
355 memset(&ifmr, 0, sizeof(ifmr));
356 strncpy(ifmr.ifm_name, value.c_str(), sizeof(ifmr.ifm_name));
358 if (ioctl(s, SIOCGIFMEDIA, (caddr_t)&ifmr) >= 0 &&
359 ifmr.ifm_status & IFM_AVALID) {
360 devdlog(LOG_DEBUG, "%s has media type 0x%x\n",
361 value.c_str(), IFM_TYPE(ifmr.ifm_active));
362 retval = (IFM_TYPE(ifmr.ifm_active) == _type);
363 } else if (_type == -1) {
364 devdlog(LOG_DEBUG, "%s has unknown media type\n",
374 const string var_list::bogus = "_$_$_$_$_B_O_G_U_S_$_$_$_$_";
375 const string var_list::nothing = "";
378 var_list::get_variable(const string &var) const
380 map<string, string>::const_iterator i;
383 if (i == _vars.end())
384 return (var_list::bogus);
389 var_list::is_set(const string &var) const
391 return (_vars.find(var) != _vars.end());
395 var_list::set_variable(const string &var, const string &val)
398 * This function gets called WAY too often to justify calling syslog()
399 * each time, even at LOG_DEBUG. Because if syslogd isn't running, it
400 * can consume excessive amounts of systime inside of connect(). Only
401 * log when we're in -d mode.
404 devdlog(LOG_DEBUG, "setting %s=%s\n", var.c_str(), val.c_str());
412 delete_and_clear(_var_list_table);
413 delete_and_clear(_attach_list);
414 delete_and_clear(_detach_list);
415 delete_and_clear(_nomatch_list);
416 delete_and_clear(_notify_list);
420 config::parse_one_file(const char *fn)
422 devdlog(LOG_DEBUG, "Parsing %s\n", fn);
423 yyin = fopen(fn, "r");
425 err(1, "Cannot open config file %s", fn);
428 errx(1, "Cannot parse %s at line %d", fn, lineno);
433 config::parse_files_in_dir(const char *dirname)
439 devdlog(LOG_DEBUG, "Parsing files in %s\n", dirname);
440 dirp = opendir(dirname);
443 readdir(dirp); /* Skip . */
444 readdir(dirp); /* Skip .. */
445 while ((dp = readdir(dirp)) != NULL) {
446 if (strcmp(dp->d_name + dp->d_namlen - 5, ".conf") == 0) {
447 snprintf(path, sizeof(path), "%s/%s",
448 dirname, dp->d_name);
449 parse_one_file(path);
457 int operator()(event_proc *const&l1, event_proc *const&l2) const
459 return (l1->get_priority() > l2->get_priority());
464 config::sort_vector(vector<event_proc *> &v)
466 stable_sort(v.begin(), v.end(), epv_greater());
472 vector<string>::const_iterator i;
474 parse_one_file(configfile);
475 for (i = _dir_list.begin(); i != _dir_list.end(); ++i)
476 parse_files_in_dir((*i).c_str());
477 sort_vector(_attach_list);
478 sort_vector(_detach_list);
479 sort_vector(_nomatch_list);
480 sort_vector(_notify_list);
484 config::open_pidfile()
488 if (_pidfile.empty())
490 pfh = pidfile_open(_pidfile.c_str(), 0600, &otherpid);
493 errx(1, "devd already running, pid: %d", (int)otherpid);
494 warn("cannot open pid file");
499 config::write_pidfile()
506 config::close_pidfile()
513 config::remove_pidfile()
520 config::add_attach(int prio, event_proc *p)
522 p->set_priority(prio);
523 _attach_list.push_back(p);
527 config::add_detach(int prio, event_proc *p)
529 p->set_priority(prio);
530 _detach_list.push_back(p);
534 config::add_directory(const char *dir)
536 _dir_list.push_back(string(dir));
540 config::add_nomatch(int prio, event_proc *p)
542 p->set_priority(prio);
543 _nomatch_list.push_back(p);
547 config::add_notify(int prio, event_proc *p)
549 p->set_priority(prio);
550 _notify_list.push_back(p);
554 config::set_pidfile(const char *fn)
560 config::push_var_table()
565 _var_list_table.push_back(vl);
566 devdlog(LOG_DEBUG, "Pushing table\n");
570 config::pop_var_table()
572 delete _var_list_table.back();
573 _var_list_table.pop_back();
574 devdlog(LOG_DEBUG, "Popping table\n");
578 config::set_variable(const char *var, const char *val)
580 _var_list_table.back()->set_variable(var, val);
584 config::get_variable(const string &var)
586 vector<var_list *>::reverse_iterator i;
588 for (i = _var_list_table.rbegin(); i != _var_list_table.rend(); ++i) {
589 if ((*i)->is_set(var))
590 return ((*i)->get_variable(var));
592 return (var_list::nothing);
596 config::is_id_char(char ch) const
598 return (ch != '\0' && (isalpha(ch) || isdigit(ch) || ch == '_' ||
603 config::expand_one(const char *&src, string &dst)
616 // Not sure if I want to support this or not, so for now we just pass
621 /* If the string ends before ) is matched , return. */
622 while (count > 0 && *src) {
625 else if (*src == '(')
633 if (!isalpha(*src)) {
639 // $var -> replace with value
642 } while (is_id_char(*src));
643 dst.append(get_variable(buffer));
647 config::expand_string(const char *src, const char *prepend, const char *append)
653 * 128 bytes is enough for 2427 of 2438 expansions that happen
654 * while parsing config files, as tested on 2013-01-30.
662 var_at = strchr(src, '$');
663 if (var_at == NULL) {
667 dst.append(src, var_at - src);
669 expand_one(src, dst);
679 config::chop_var(char *&buffer, char *&lhs, char *&rhs) const
685 walker = lhs = buffer;
686 while (is_id_char(*walker))
691 if (*walker == '"') {
694 while (*walker && *walker != '"')
702 while (*walker && !isspace(*walker))
708 while (isspace(*walker))
716 config::set_vars(char *buffer)
722 if (!chop_var(buffer, lhs, rhs))
724 set_variable(lhs, rhs);
730 config::find_and_execute(char type)
732 vector<event_proc *> *l;
733 vector<event_proc *>::const_iterator i;
756 devdlog(LOG_DEBUG, "Processing %s event\n", s);
757 for (i = l->begin(); i != l->end(); ++i) {
758 if ((*i)->matches(*this)) {
768 process_event(char *buffer)
774 devdlog(LOG_INFO, "Processing event '%s'\n", buffer);
776 cfg.push_var_table();
777 // No match doesn't have a device, and the format is a little
778 // different, so handle it separately.
781 sp = cfg.set_vars(sp);
784 //? at location pnp-info on bus
785 sp = strchr(sp, ' ');
787 return; /* Can't happen? */
791 if (strncmp(sp, "at ", 3) == 0)
793 sp = cfg.set_vars(sp);
796 if (strncmp(sp, "on ", 3) == 0)
797 cfg.set_variable("bus", sp + 3);
799 case attach: /*FALLTHROUGH*/
801 sp = strchr(sp, ' ');
803 return; /* Can't happen? */
805 cfg.set_variable("device-name", buffer);
808 if (strncmp(sp, "at ", 3) == 0)
810 sp = cfg.set_vars(sp);
813 if (strncmp(sp, "on ", 3) == 0)
814 cfg.set_variable("bus", sp + 3);
818 cfg.find_and_execute(type);
823 create_socket(const char *name)
826 struct sockaddr_un sun;
828 if ((fd = socket(PF_LOCAL, SOCK_STREAM, 0)) < 0)
830 bzero(&sun, sizeof(sun));
831 sun.sun_family = AF_UNIX;
832 strlcpy(sun.sun_path, name, sizeof(sun.sun_path));
833 slen = SUN_LEN(&sun);
835 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0)
837 if (::bind(fd, (struct sockaddr *) & sun, slen) < 0)
840 chown(name, 0, 0); /* XXX - root.wheel */
845 unsigned int max_clients = 10; /* Default, can be overriden on cmdline. */
846 unsigned int num_clients;
850 notify_clients(const char *data, int len)
852 list<int>::iterator i;
855 * Deliver the data to all clients. Throw clients overboard at the
856 * first sign of trouble. This reaps clients who've died or closed
857 * their sockets, and also clients who are alive but failing to keep up
858 * (or who are maliciously not reading, to consume buffer space in
859 * kernel memory or tie up the limited number of available connections).
861 for (i = clients.begin(); i != clients.end(); ) {
862 if (write(*i, data, len) != len) {
865 i = clients.erase(i);
866 devdlog(LOG_WARNING, "notify_clients: write() failed; "
867 "dropping unresponsive client\n");
878 list<int>::iterator i;
881 * Check all existing clients to see if any of them have disappeared.
882 * Normally we reap clients when we get an error trying to send them an
883 * event. This check eliminates the problem of an ever-growing list of
884 * zombie clients because we're never writing to them on a system
885 * without frequent device-change activity.
888 for (i = clients.begin(); i != clients.end(); ) {
890 s = poll(&pfd, 1, 0);
891 if ((s < 0 && s != EINTR ) ||
892 (s > 0 && (pfd.revents & POLLHUP))) {
895 i = clients.erase(i);
896 devdlog(LOG_NOTICE, "check_clients: "
897 "dropping disconnected client\n");
910 * First go reap any zombie clients, then accept the connection, and
911 * shut down the read side to stop clients from consuming kernel memory
912 * by sending large buffers full of data we'll never read.
915 s = accept(fd, NULL, NULL);
917 sndbuf_size = CLIENT_BUFSIZE;
918 if (setsockopt(s, SOL_SOCKET, SO_SNDBUF, &sndbuf_size,
919 sizeof(sndbuf_size)))
920 err(1, "setsockopt");
921 shutdown(s, SHUT_RD);
922 clients.push_back(s);
933 char buffer[DEVCTL_MAXBUF];
935 int server_fd, max_fd;
940 fd = open(PATH_DEVCTL, O_RDONLY | O_CLOEXEC);
942 err(1, "Can't open devctl device %s", PATH_DEVCTL);
943 server_fd = create_socket(PIPE);
945 max_fd = max(fd, server_fd) + 1;
946 while (!romeo_must_die) {
947 if (!once && !no_daemon && !daemonize_quick) {
948 // Check to see if we have any events pending.
953 rv = select(fd + 1, &fds, &fds, &fds, &tv);
954 // No events -> we've processed all pending events
956 devdlog(LOG_DEBUG, "Calling daemon\n");
957 cfg.remove_pidfile();
965 * When we've already got the max number of clients, stop
966 * accepting new connections (don't put server_fd in the set),
967 * shrink the accept() queue to reject connections quickly, and
968 * poll the existing clients more often, so that we notice more
969 * quickly when any of them disappear to free up client slots.
973 if (num_clients < max_clients) {
975 listen(server_fd, max_clients);
978 FD_SET(server_fd, &fds);
983 listen(server_fd, 0);
989 rv = select(max_fd, &fds, NULL, NULL, &tv);
991 devdlog(LOG_NOTICE, "Events received so far=%u\n",
1001 if (FD_ISSET(fd, &fds)) {
1002 rv = read(fd, buffer, sizeof(buffer) - 1);
1005 if (rv == sizeof(buffer) - 1) {
1006 devdlog(LOG_WARNING, "Warning: "
1007 "available event data exceeded "
1010 notify_clients(buffer, rv);
1012 while (buffer[--rv] == '\n')
1014 process_event(buffer);
1015 } else if (rv < 0) {
1023 if (FD_ISSET(server_fd, &fds))
1024 new_client(server_fd);
1030 * functions that the parser uses.
1033 add_attach(int prio, event_proc *p)
1035 cfg.add_attach(prio, p);
1039 add_detach(int prio, event_proc *p)
1041 cfg.add_detach(prio, p);
1045 add_directory(const char *dir)
1047 cfg.add_directory(dir);
1048 free(const_cast<char *>(dir));
1052 add_nomatch(int prio, event_proc *p)
1054 cfg.add_nomatch(prio, p);
1058 add_notify(int prio, event_proc *p)
1060 cfg.add_notify(prio, p);
1064 add_to_event_proc(event_proc *ep, eps *eps)
1067 ep = new event_proc();
1073 new_action(const char *cmd)
1075 eps *e = new action(cmd);
1076 free(const_cast<char *>(cmd));
1081 new_match(const char *var, const char *re)
1083 eps *e = new match(cfg, var, re);
1084 free(const_cast<char *>(var));
1085 free(const_cast<char *>(re));
1090 new_media(const char *var, const char *re)
1092 eps *e = new media(cfg, var, re);
1093 free(const_cast<char *>(var));
1094 free(const_cast<char *>(re));
1099 set_pidfile(const char *name)
1101 cfg.set_pidfile(name);
1102 free(const_cast<char *>(name));
1106 set_variable(const char *var, const char *val)
1108 cfg.set_variable(var, val);
1109 free(const_cast<char *>(var));
1110 free(const_cast<char *>(val));
1122 * SIGINFO handler. Will print useful statistics to the syslog or stderr
1132 * Local logging function. Prints to syslog if we're daemonized; stderr
1136 devdlog(int priority, const char* fmt, ...)
1140 va_start(argp, fmt);
1142 vfprintf(stderr, fmt, argp);
1143 else if ((! quiet_mode) || (priority <= LOG_WARNING))
1144 vsyslog(priority, fmt, argp);
1151 fprintf(stderr, "usage: %s [-dnq] [-l connlimit] [-f file]\n",
1157 check_devd_enabled()
1163 if (sysctlbyname(SYSCTL, &val, &len, NULL, 0) != 0)
1164 errx(1, "devctl sysctl missing from kernel!");
1166 warnx("Setting " SYSCTL " to 0");
1168 sysctlbyname(SYSCTL, NULL, NULL, &val, sizeof(val));
1176 main(int argc, char **argv)
1180 check_devd_enabled();
1181 while ((ch = getopt(argc, argv, "df:l:nq")) != -1) {
1187 configfile = optarg;
1190 max_clients = MAX(1, strtoul(optarg, NULL, 0));
1193 daemonize_quick = 1;
1204 if (!no_daemon && daemonize_quick) {
1207 cfg.write_pidfile();
1209 signal(SIGPIPE, SIG_IGN);
1210 signal(SIGHUP, gensighand);
1211 signal(SIGINT, gensighand);
1212 signal(SIGTERM, gensighand);
1213 signal(SIGINFO, siginfohand);