2 * udevd.c - hotplug event serializer
4 * Copyright (C) 2004-2005 Kay Sievers <kay.sievers@vrfy.org>
5 * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
8 * This program is free software; you can redistribute it and/or modify it
9 * under the terms of the GNU General Public License as published by the
10 * Free Software Foundation version 2 of the License.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
17 * You should have received a copy of the GNU General Public License along
18 * with this program; if not, write to the Free Software Foundation, Inc.,
19 * 675 Mass Ave, Cambridge, MA 02139, USA.
33 #include <sys/select.h>
36 #include <sys/types.h>
37 #include <sys/socket.h>
39 #include <sys/sysinfo.h>
41 #include <linux/types.h>
42 #include <linux/netlink.h>
45 #include "udev_libc_wrapper.h"
47 #include "udev_version.h"
48 #include "udev_utils.h"
53 static int udevd_sock;
54 static int uevent_netlink_sock;
57 static int pipefds[2];
58 static volatile int sigchilds_waiting;
59 static volatile int run_msg_q;
60 static volatile int sig_flag;
61 static volatile int udev_exit;
62 static int init_phase = 1;
63 static int run_exec_q;
64 static int stop_exec_q;
66 static LIST_HEAD(msg_list);
67 static LIST_HEAD(exec_list);
68 static LIST_HEAD(running_list);
70 static void exec_queue_manager(void);
71 static void msg_queue_manager(void);
72 static void user_sighandler(void);
73 static void reap_sigchilds(void);
75 static char *udev_bin;
76 static unsigned long long expected_seqnum;
77 static int event_timeout;
78 static int max_childs;
79 static int max_childs_running;
83 void log_message (int priority, const char *format, ...)
87 if (priority > udev_log_priority)
90 va_start(args, format);
91 vsyslog(priority, format, args);
96 static void msg_dump_queue(void)
99 struct uevent_msg *msg;
101 list_for_each_entry(msg, &msg_list, node)
102 dbg("sequence %llu in queue", msg->seqnum);
106 static void msg_queue_delete(struct uevent_msg *msg)
108 list_del(&msg->node);
112 /* orders the message in the queue by sequence number */
113 static void msg_queue_insert(struct uevent_msg *msg)
115 struct uevent_msg *loop_msg;
118 if (msg->seqnum == 0) {
119 dbg("no SEQNUM, move straight to the exec queue");
120 list_add(&msg->node, &exec_list);
125 /* store timestamp of queuing */
127 msg->queue_time = info.uptime;
129 /* with the first event we provide a phase of shorter timeout */
131 static long init_time;
134 init_time = info.uptime;
135 if (info.uptime - init_time >= UDEVD_INIT_TIME)
139 /* don't delay messages with timeout set */
141 dbg("move seq %llu with timeout %u to exec queue", msg->seqnum, msg->timeout);
142 list_add(&msg->node, &exec_list);
147 /* sort message by sequence number into list */
148 list_for_each_entry_reverse(loop_msg, &msg_list, node) {
149 if (loop_msg->seqnum < msg->seqnum)
152 if (loop_msg->seqnum == msg->seqnum) {
153 dbg("ignoring duplicate message seq %llu", msg->seqnum);
158 list_add(&msg->node, &loop_msg->node);
159 info("seq %llu queued, devpath '%s'", msg->seqnum, msg->devpath);
161 /* run msg queue manager */
167 /* forks event and removes event from run queue when finished */
168 static void udev_event_fork(struct uevent_msg *msg)
170 char *const argv[] = { "udev", msg->subsystem, NULL };
178 if (uevent_netlink_sock != -1)
179 close(uevent_netlink_sock);
182 setpriority(PRIO_PROCESS, 0, UDEV_PRIORITY);
183 execve(udev_bin, argv, msg->envp);
184 err("exec of child failed");
187 err("fork of child failed");
188 msg_queue_delete(msg);
191 /* get SIGCHLD in main loop */
193 info("seq %llu forked, pid %d, %ld seconds old",
194 msg->seqnum, pid, info.uptime - msg->queue_time);
199 static int running_processes(void)
202 static char buf[4096];
207 f = open("/proc/stat", O_RDONLY);
211 len = read(f, buf, sizeof(buf));
219 pos = strstr(buf, "procs_running ");
223 if (sscanf(pos, "procs_running %u", &running) != 1)
229 /* return the number of process es in our session, count only until limit */
230 static int running_processes_in_session(pid_t session, int limit)
236 dir = opendir("/proc");
240 /* read process info from /proc */
241 for (dent = readdir(dir); dent != NULL; dent = readdir(dir)) {
247 pid_t ppid, pgrp, sess;
250 if (!isdigit(dent->d_name[0]))
253 snprintf(procdir, sizeof(procdir), "/proc/%s/stat", dent->d_name);
254 procdir[sizeof(procdir)-1] = '\0';
256 f = open(procdir, O_RDONLY);
260 len = read(f, line, sizeof(line));
268 /* skip ugly program name */
269 pos = strrchr(line, ')') + 2;
273 if (sscanf(pos, "%c %d %d %d ", &state, &ppid, &pgrp, &sess) != 4)
276 /* count only processes in our session */
280 /* count only running, no sleeping processes */
285 if (limit > 0 && running >= limit)
293 static int compare_devpath(const char *running, const char *waiting)
297 for (i = 0; i < PATH_SIZE; i++) {
298 /* identical device event found */
299 if (running[i] == '\0' && waiting[i] == '\0')
302 /* parent device event found */
303 if (running[i] == '\0' && waiting[i] == '/')
306 /* child device event found */
307 if (running[i] == '/' && waiting[i] == '\0')
310 /* no matching event */
311 if (running[i] != waiting[i])
318 /* returns still running task for the same device, its parent or its physical device */
319 static int running_with_devpath(struct uevent_msg *msg, int limit)
321 struct uevent_msg *loop_msg;
322 int childs_count = 0;
324 if (msg->devpath == NULL)
327 /* skip any events with a timeout set */
328 if (msg->timeout != 0)
331 list_for_each_entry(loop_msg, &running_list, node) {
332 if (limit && childs_count++ > limit) {
333 dbg("%llu, maximum number (%i) of child reached", msg->seqnum, childs_count);
336 if (loop_msg->devpath == NULL)
339 /* return running parent/child device event */
340 if (compare_devpath(loop_msg->devpath, msg->devpath) != 0) {
341 dbg("%llu, child device event still running %llu (%s)",
342 msg->seqnum, loop_msg->seqnum, loop_msg->devpath);
346 /* return running physical device event */
347 if (msg->physdevpath && msg->action && strcmp(msg->action, "add") == 0)
348 if (compare_devpath(loop_msg->devpath, msg->physdevpath) != 0) {
349 dbg("%llu, physical device event still running %llu (%s)",
350 msg->seqnum, loop_msg->seqnum, loop_msg->devpath);
358 /* exec queue management routine executes the events and serializes events in the same sequence */
359 static void exec_queue_manager(void)
361 struct uevent_msg *loop_msg;
362 struct uevent_msg *tmp_msg;
365 if (list_empty(&exec_list))
368 running = running_processes();
369 dbg("%d processes runnning on system", running);
371 running = max_childs_running;
373 list_for_each_entry_safe(loop_msg, tmp_msg, &exec_list, node) {
374 /* check running processes in our session and possibly throttle */
375 if (running >= max_childs_running) {
376 running = running_processes_in_session(sid, max_childs_running+10);
377 dbg("at least %d processes running in session", running);
378 if (running >= max_childs_running) {
379 dbg("delay seq %llu, cause too many processes already running",
385 if (running_with_devpath(loop_msg, max_childs) == 0) {
386 /* move event to run list */
387 list_move_tail(&loop_msg->node, &running_list);
388 udev_event_fork(loop_msg);
390 dbg("moved seq %llu to running list", loop_msg->seqnum);
392 dbg("delay seq %llu (%s)", loop_msg->seqnum, loop_msg->devpath);
396 static void msg_move_exec(struct uevent_msg *msg)
398 list_move_tail(&msg->node, &exec_list);
400 expected_seqnum = msg->seqnum+1;
401 dbg("moved seq %llu to exec, next expected is %llu",
402 msg->seqnum, expected_seqnum);
405 /* msg queue management routine handles the timeouts and dispatches the events */
406 static void msg_queue_manager(void)
408 struct uevent_msg *loop_msg;
409 struct uevent_msg *tmp_msg;
412 int timeout = event_timeout;
414 dbg("msg queue manager, next expected is %llu", expected_seqnum);
417 list_for_each_entry_safe(loop_msg, tmp_msg, &msg_list, node) {
418 /* move event with expected sequence to the exec list */
419 if (loop_msg->seqnum == expected_seqnum) {
420 msg_move_exec(loop_msg);
424 /* limit timeout during initialization phase */
426 timeout = UDEVD_INIT_EVENT_TIMEOUT;
427 dbg("initialization phase, limit timeout to %i seconds", UDEVD_INIT_EVENT_TIMEOUT);
430 /* move event with expired timeout to the exec list */
431 msg_age = info.uptime - loop_msg->queue_time;
432 dbg("seq %llu is %li seconds old", loop_msg->seqnum, msg_age);
433 if (msg_age >= timeout) {
434 msg_move_exec(loop_msg);
443 /* set timeout for remaining queued events */
444 if (list_empty(&msg_list) == 0) {
445 struct itimerval itv = {{0, 0}, {timeout - msg_age, 0}};
446 dbg("next event expires in %li seconds", timeout - msg_age);
447 setitimer(ITIMER_REAL, &itv, NULL);
451 static struct uevent_msg *get_msg_from_envbuf(const char *buf, int buf_size)
455 struct uevent_msg *msg;
459 msg = malloc(sizeof(struct uevent_msg) + buf_size);
462 memset(msg, 0x00, sizeof(struct uevent_msg) + buf_size);
464 /* copy environment buffer and reconstruct envp */
465 memcpy(msg->envbuf, buf, buf_size);
467 for (i = 0; (bufpos < buf_size) && (i < UEVENT_NUM_ENVP-2); i++) {
471 key = &msg->envbuf[bufpos];
472 keylen = strlen(key);
474 bufpos += keylen + 1;
475 dbg("add '%s' to msg.envp[%i]", msg->envp[i], i);
477 /* remember some keys for further processing */
478 if (strncmp(key, "ACTION=", 7) == 0)
479 msg->action = &key[7];
480 else if (strncmp(key, "DEVPATH=", 8) == 0)
481 msg->devpath = &key[8];
482 else if (strncmp(key, "SUBSYSTEM=", 10) == 0)
483 msg->subsystem = &key[10];
484 else if (strncmp(key, "SEQNUM=", 7) == 0)
485 msg->seqnum = strtoull(&key[7], NULL, 10);
486 else if (strncmp(key, "PHYSDEVPATH=", 12) == 0)
487 msg->physdevpath = &key[12];
488 else if (strncmp(key, "MAJOR=", 6) == 0)
489 major = strtoull(&key[6], NULL, 10);
490 else if (strncmp(key, "MINOR=", 6) == 0)
491 minor = strtoull(&key[6], NULL, 10);
492 else if (strncmp(key, "TIMEOUT=", 8) == 0)
493 msg->timeout = strtoull(&key[8], NULL, 10);
495 msg->devt = makedev(major, minor);
496 msg->envp[i++] = "UDEVD_EVENT=1";
502 /* receive the udevd message from userspace */
503 static struct uevent_msg *get_udevd_msg(void)
505 static struct udevd_msg usend_msg;
506 struct uevent_msg *msg;
509 struct cmsghdr *cmsg;
512 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
516 memset(&usend_msg, 0x00, sizeof(struct udevd_msg));
517 iov.iov_base = &usend_msg;
518 iov.iov_len = sizeof(struct udevd_msg);
520 memset(&smsg, 0x00, sizeof(struct msghdr));
523 smsg.msg_control = cred_msg;
524 smsg.msg_controllen = sizeof(cred_msg);
526 size = recvmsg(udevd_sock, &smsg, 0);
529 dbg("unable to receive udevd message");
532 cmsg = CMSG_FIRSTHDR(&smsg);
533 cred = (struct ucred *) CMSG_DATA(cmsg);
535 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
536 info("no sender credentials received, message ignored");
540 if (cred->uid != 0) {
541 info("sender uid=%i, message ignored", cred->uid);
545 if (strncmp(usend_msg.magic, UDEV_MAGIC, sizeof(UDEV_MAGIC)) != 0 ) {
546 info("message magic '%s' doesn't match, ignore it", usend_msg.magic);
550 switch (usend_msg.type) {
551 case UDEVD_UEVENT_UDEVSEND:
552 case UDEVD_UEVENT_INITSEND:
553 info("udevd event message received");
554 envbuf_size = size - offsetof(struct udevd_msg, envbuf);
555 dbg("envbuf_size=%i", envbuf_size);
556 msg = get_msg_from_envbuf(usend_msg.envbuf, envbuf_size);
559 msg->type = usend_msg.type;
561 case UDEVD_STOP_EXEC_QUEUE:
562 info("udevd message (STOP_EXEC_QUEUE) received");
565 case UDEVD_START_EXEC_QUEUE:
566 info("udevd message (START_EXEC_QUEUE) received");
568 exec_queue_manager();
570 case UDEVD_SET_LOG_LEVEL:
571 intval = (int *) usend_msg.envbuf;
572 info("udevd message (SET_LOG_PRIORITY) received, udev_log_priority=%i", *intval);
573 udev_log_priority = *intval;
575 case UDEVD_SET_MAX_CHILDS:
576 intval = (int *) usend_msg.envbuf;
577 info("udevd message (UDEVD_SET_MAX_CHILDS) received, max_childs=%i", *intval);
578 max_childs = *intval;
581 dbg("unknown message type");
586 /* receive the kernel user event message and do some sanity checks */
587 static struct uevent_msg *get_netlink_msg(void)
589 struct uevent_msg *msg;
592 static char buffer[UEVENT_BUFFER_SIZE + 512];
595 size = recv(uevent_netlink_sock, &buffer, sizeof(buffer), 0);
598 dbg("unable to receive udevd message");
602 if ((size_t)size > sizeof(buffer)-1)
603 size = sizeof(buffer)-1;
605 dbg("uevent_size=%zi", size);
607 /* start of event payload */
608 bufpos = strlen(buffer)+1;
609 msg = get_msg_from_envbuf(&buffer[bufpos], size-bufpos);
612 msg->type = UDEVD_UEVENT_NETLINK;
614 /* validate message */
615 pos = strchr(buffer, '@');
617 dbg("invalid uevent '%s'", buffer);
623 if (msg->action == NULL) {
624 dbg("no ACTION in payload found, skip event '%s'", buffer);
629 if (strcmp(msg->action, buffer) != 0) {
630 dbg("ACTION in payload does not match uevent, skip event '%s'", buffer);
638 static void asmlinkage sig_handler(int signum)
648 /* set flag, then write to pipe if needed */
652 /* set flag, then write to pipe if needed */
653 sigchilds_waiting = 1;
657 /* if pipe is empty, write to pipe to force select to return,
658 * which will wakeup our mainloop
661 rc = write(pipefds[1],&signum,sizeof(signum));
667 static void udev_done(int pid)
669 /* find msg associated with pid and delete it */
670 struct uevent_msg *msg;
673 list_for_each_entry(msg, &running_list, node) {
674 if (msg->pid == pid) {
676 info("seq %llu exit, %ld seconds old", msg->seqnum, info.uptime - msg->queue_time);
677 msg_queue_delete(msg);
679 /* we want to run the exec queue manager since there may
680 * be events waiting with the devpath of the one that
689 static void reap_sigchilds(void)
692 int pid = waitpid(-1, NULL, WNOHANG);
693 if ((pid == -1) || (pid == 0))
699 /* just read everything from the pipe and clear the flag,
700 * the flags was set in the signal handler
702 static void user_sighandler(void)
707 int rc = read(pipefds[0], &sig, sizeof(sig));
715 static int init_udevd_socket(void)
717 struct sockaddr_un saddr;
718 const int buffersize = 1024 * 1024;
720 const int feature_on = 1;
723 memset(&saddr, 0x00, sizeof(saddr));
724 saddr.sun_family = AF_LOCAL;
725 /* use abstract namespace for socket path */
726 strcpy(&saddr.sun_path[1], UDEVD_SOCK_PATH);
727 addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(saddr.sun_path+1) + 1;
729 udevd_sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
730 if (udevd_sock == -1) {
731 err("error getting socket, %s", strerror(errno));
735 /* set receive buffersize */
736 setsockopt(udevd_sock, SOL_SOCKET, SO_RCVBUF, &buffersize, sizeof(buffersize));
738 /* the bind takes care of ensuring only one copy running */
739 retval = bind(udevd_sock, (struct sockaddr *) &saddr, addrlen);
741 err("bind failed, %s", strerror(errno));
746 /* enable receiving of the sender credentials */
747 setsockopt(udevd_sock, SOL_SOCKET, SO_PASSCRED, &feature_on, sizeof(feature_on));
752 static int init_uevent_netlink_sock(void)
754 struct sockaddr_nl snl;
755 const int buffersize = 1024 * 1024;
758 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
759 snl.nl_family = AF_NETLINK;
760 snl.nl_pid = getpid();
761 snl.nl_groups = 0xffffffff;
763 uevent_netlink_sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
764 if (uevent_netlink_sock == -1) {
765 dbg("error getting socket, %s", strerror(errno));
769 /* set receive buffersize */
770 setsockopt(uevent_netlink_sock, SOL_SOCKET, SO_RCVBUF, &buffersize, sizeof(buffersize));
772 retval = bind(uevent_netlink_sock, (struct sockaddr *) &snl,
773 sizeof(struct sockaddr_nl));
775 dbg("bind failed, %s", strerror(errno));
776 close(uevent_netlink_sock);
777 uevent_netlink_sock = -1;
784 int main(int argc, char *argv[], char *envp[])
789 struct sigaction act;
792 int uevent_netlink_active = 0;
796 logging_init("udevd");
798 dbg("version %s", UDEV_VERSION);
801 err("need to be root, exit");
805 for (i = 1 ; i < argc; i++) {
807 if (strcmp(arg, "--daemon") == 0 || strcmp(arg, "-d") == 0) {
808 info("will daemonize");
811 if (strcmp(arg, "--stop-exec-queue") == 0) {
812 info("will not execute events until START_EXEC_QUEUE is received");
822 dbg("damonized fork running");
825 err("fork of daemon failed");
833 /* become session leader */
835 dbg("our session is %d", sid);
838 umask(umask(077) | 022);
840 /* set a reasonable scheduling priority for the daemon */
841 setpriority(PRIO_PROCESS, 0, UDEVD_PRIORITY);
843 /* Set fds to dev/null */
844 fd = open( "/dev/null", O_RDWR );
852 err("error opening /dev/null %s", strerror(errno));
854 /* setup signal handler pipe */
855 retval = pipe(pipefds);
857 err("error getting pipes: %s", strerror(errno));
861 retval = fcntl(pipefds[0], F_SETFL, O_NONBLOCK);
863 err("error fcntl on read pipe: %s", strerror(errno));
866 retval = fcntl(pipefds[0], F_SETFD, FD_CLOEXEC);
868 err("error fcntl on read pipe: %s", strerror(errno));
870 retval = fcntl(pipefds[1], F_SETFL, O_NONBLOCK);
872 err("error fcntl on write pipe: %s", strerror(errno));
875 retval = fcntl(pipefds[1], F_SETFD, FD_CLOEXEC);
877 err("error fcntl on write pipe: %s", strerror(errno));
879 /* set signal handlers */
880 memset(&act, 0x00, sizeof(struct sigaction));
881 act.sa_handler = (void (*)(int)) sig_handler;
882 sigemptyset(&act.sa_mask);
883 act.sa_flags = SA_RESTART;
884 sigaction(SIGINT, &act, NULL);
885 sigaction(SIGTERM, &act, NULL);
886 sigaction(SIGALRM, &act, NULL);
887 sigaction(SIGCHLD, &act, NULL);
888 sigaction(SIGHUP, &act, NULL);
890 if (init_uevent_netlink_sock() < 0) {
891 dbg("uevent socket not available");
894 if (init_udevd_socket() < 0) {
895 if (errno == EADDRINUSE)
896 dbg("another udevd running, exit");
898 dbg("error initialising udevd socket: %s", strerror(errno));
903 /* override of forked udev binary, used for testing */
904 udev_bin = getenv("UDEV_BIN");
905 if (udev_bin != NULL)
906 info("udev binary is set to '%s'", udev_bin);
910 /* init of expected_seqnum value */
911 value = getenv("UDEVD_EXPECTED_SEQNUM");
913 expected_seqnum = strtoull(value, NULL, 10);
914 info("initialize expected_seqnum to %llu", expected_seqnum);
917 /* timeout to wait for missing events */
918 value = getenv("UDEVD_EVENT_TIMEOUT");
920 event_timeout = strtoul(value, NULL, 10);
922 event_timeout = UDEVD_EVENT_TIMEOUT;
923 info("initialize event_timeout to %u", event_timeout);
925 /* maximum limit of forked childs */
926 value = getenv("UDEVD_MAX_CHILDS");
928 max_childs = strtoul(value, NULL, 10);
930 max_childs = UDEVD_MAX_CHILDS;
931 info("initialize max_childs to %u", max_childs);
933 /* start to throttle forking if maximum number of _running_ childs is reached */
934 value = getenv("UDEVD_MAX_CHILDS_RUNNING");
936 max_childs_running = strtoull(value, NULL, 10);
938 max_childs_running = UDEVD_MAX_CHILDS_RUNNING;
939 info("initialize max_childs_running to %u", max_childs_running);
942 FD_SET(udevd_sock, &readfds);
943 if (uevent_netlink_sock > 0)
944 FD_SET(uevent_netlink_sock, &readfds);
945 FD_SET(pipefds[0], &readfds);
946 maxsockplus = udevd_sock+1;
948 struct uevent_msg *msg;
950 fd_set workreadfds = readfds;
951 retval = select(maxsockplus, &workreadfds, NULL, NULL, NULL);
955 dbg("error in select: %s", strerror(errno));
959 /* get user socket message */
960 if (FD_ISSET(udevd_sock, &workreadfds)) {
961 msg = get_udevd_msg();
963 /* discard kernel messages if netlink is active */
964 if (uevent_netlink_active && msg->type == UDEVD_UEVENT_UDEVSEND && msg->seqnum != 0) {
965 dbg("skip uevent_helper message, netlink is active");
969 msg_queue_insert(msg);
973 /* get kernel netlink message */
974 if (FD_ISSET(uevent_netlink_sock, &workreadfds)) {
975 msg = get_netlink_msg();
977 msg_queue_insert(msg);
978 /* disable udevsend with first netlink message */
979 if (!uevent_netlink_active) {
980 info("uevent_nl message received, disable udevsend messages");
981 uevent_netlink_active = 1;
986 /* received a signal, clear our notification pipe */
987 if (FD_ISSET(pipefds[0], &workreadfds))
990 /* forked child have returned */
991 if (sigchilds_waiting) {
992 sigchilds_waiting = 0;
1002 /* clean up running_list before calling exec_queue_manager() */
1003 if (sigchilds_waiting) {
1004 sigchilds_waiting = 0;
1010 exec_queue_manager();
1018 if (uevent_netlink_sock > 0)
1019 close(uevent_netlink_sock);