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"
52 #ifndef NETLINK_KOBJECT_UEVENT
53 #define NETLINK_KOBJECT_UEVENT 15
57 static int udevd_sock;
58 static int uevent_netlink_sock;
61 static int pipefds[2];
62 static volatile int sigchilds_waiting;
63 static volatile int run_msg_q;
64 static volatile int sig_flag;
65 static int init_phase = 1;
66 static int run_exec_q;
67 static int stop_exec_q;
69 static LIST_HEAD(msg_list);
70 static LIST_HEAD(exec_list);
71 static LIST_HEAD(running_list);
73 static void exec_queue_manager(void);
74 static void msg_queue_manager(void);
75 static void user_sighandler(void);
76 static void reap_sigchilds(void);
78 static char *udev_bin;
79 static unsigned long long expected_seqnum;
80 static int event_timeout;
81 static int max_childs;
82 static int max_childs_running;
86 void log_message (int priority, const char *format, ...)
90 if (priority > udev_log_priority)
93 va_start(args, format);
94 vsyslog(priority, format, args);
99 static void msg_dump_queue(void)
102 struct uevent_msg *msg;
104 list_for_each_entry(msg, &msg_list, node)
105 dbg("sequence %llu in queue", msg->seqnum);
109 static void msg_queue_delete(struct uevent_msg *msg)
111 list_del(&msg->node);
115 /* orders the message in the queue by sequence number */
116 static void msg_queue_insert(struct uevent_msg *msg)
118 struct uevent_msg *loop_msg;
121 if (msg->seqnum == 0) {
122 dbg("no SEQNUM, move straight to the exec queue");
123 list_add(&msg->node, &exec_list);
128 /* store timestamp of queuing */
130 msg->queue_time = info.uptime;
132 /* with the first event we provide a phase of shorter timeout */
134 static long init_time;
137 init_time = info.uptime;
138 if (info.uptime - init_time >= UDEVD_INIT_TIME)
142 /* don't delay messages with timeout set */
144 dbg("move seq %llu with timeout %u to exec queue", msg->seqnum, msg->timeout);
145 list_add(&msg->node, &exec_list);
150 /* sort message by sequence number into list */
151 list_for_each_entry_reverse(loop_msg, &msg_list, node) {
152 if (loop_msg->seqnum < msg->seqnum)
155 if (loop_msg->seqnum == msg->seqnum) {
156 dbg("ignoring duplicate message seq %llu", msg->seqnum);
161 list_add(&msg->node, &loop_msg->node);
162 info("seq %llu queued, devpath '%s'", msg->seqnum, msg->devpath);
164 /* run msg queue manager */
170 /* forks event and removes event from run queue when finished */
171 static void udev_event_fork(struct uevent_msg *msg)
173 char *const argv[] = { "udev", msg->subsystem, NULL };
181 if (uevent_netlink_sock != -1)
182 close(uevent_netlink_sock);
185 setpriority(PRIO_PROCESS, 0, UDEV_PRIORITY);
186 execve(udev_bin, argv, msg->envp);
187 err("exec of child failed");
190 err("fork of child failed");
191 msg_queue_delete(msg);
194 /* get SIGCHLD in main loop */
196 info("seq %llu forked, pid %d, %ld seconds old",
197 msg->seqnum, pid, info.uptime - msg->queue_time);
202 static int running_processes(void)
205 static char buf[4096];
210 f = open("/proc/stat", O_RDONLY);
214 len = read(f, buf, sizeof(buf));
222 pos = strstr(buf, "procs_running ");
226 if (sscanf(pos, "procs_running %u", &running) != 1)
232 /* return the number of process es in our session, count only until limit */
233 static int running_processes_in_session(pid_t session, int limit)
239 dir = opendir("/proc");
243 /* read process info from /proc */
244 for (dent = readdir(dir); dent != NULL; dent = readdir(dir)) {
250 pid_t ppid, pgrp, sess;
253 if (!isdigit(dent->d_name[0]))
256 snprintf(procdir, sizeof(procdir), "/proc/%s/stat", dent->d_name);
257 procdir[sizeof(procdir)-1] = '\0';
259 f = open(procdir, O_RDONLY);
263 len = read(f, line, sizeof(line));
271 /* skip ugly program name */
272 pos = strrchr(line, ')') + 2;
276 if (sscanf(pos, "%c %d %d %d ", &state, &ppid, &pgrp, &sess) != 4)
279 /* count only processes in our session */
283 /* count only running, no sleeping processes */
288 if (limit > 0 && running >= limit)
296 static int compare_devpath(const char *running, const char *waiting)
300 for (i = 0; i < PATH_SIZE; i++) {
301 /* identical device event found */
302 if (running[i] == '\0' && waiting[i] == '\0')
305 /* parent device event found */
306 if (running[i] == '\0' && waiting[i] == '/')
309 /* child device event found */
310 if (running[i] == '/' && waiting[i] == '\0')
313 /* no matching event */
314 if (running[i] != waiting[i])
321 /* returns still running task for the same device, its parent or its physical device */
322 static int running_with_devpath(struct uevent_msg *msg, int limit)
324 struct uevent_msg *loop_msg;
325 int childs_count = 0;
327 if (msg->devpath == NULL)
330 /* skip any events with a timeout set */
331 if (msg->timeout != 0)
334 list_for_each_entry(loop_msg, &running_list, node) {
335 if (limit && childs_count++ > limit) {
336 dbg("%llu, maximum number (%i) of child reached", msg->seqnum, childs_count);
339 if (loop_msg->devpath == NULL)
342 /* return running parent/child device event */
343 if (compare_devpath(loop_msg->devpath, msg->devpath) != 0) {
344 dbg("%llu, child device event still running %llu (%s)",
345 msg->seqnum, loop_msg->seqnum, loop_msg->devpath);
349 /* return running physical device event */
350 if (msg->physdevpath && msg->action && strcmp(msg->action, "add") == 0)
351 if (compare_devpath(loop_msg->devpath, msg->physdevpath) != 0) {
352 dbg("%llu, physical device event still running %llu (%s)",
353 msg->seqnum, loop_msg->seqnum, loop_msg->devpath);
361 /* exec queue management routine executes the events and serializes events in the same sequence */
362 static void exec_queue_manager(void)
364 struct uevent_msg *loop_msg;
365 struct uevent_msg *tmp_msg;
368 if (list_empty(&exec_list))
371 running = running_processes();
372 dbg("%d processes runnning on system", running);
374 running = max_childs_running;
376 list_for_each_entry_safe(loop_msg, tmp_msg, &exec_list, node) {
377 /* check running processes in our session and possibly throttle */
378 if (running >= max_childs_running) {
379 running = running_processes_in_session(sid, max_childs_running+10);
380 dbg("at least %d processes running in session", running);
381 if (running >= max_childs_running) {
382 dbg("delay seq %llu, cause too many processes already running",
388 if (running_with_devpath(loop_msg, max_childs) == 0) {
389 /* move event to run list */
390 list_move_tail(&loop_msg->node, &running_list);
391 udev_event_fork(loop_msg);
393 dbg("moved seq %llu to running list", loop_msg->seqnum);
395 dbg("delay seq %llu (%s)", loop_msg->seqnum, loop_msg->devpath);
399 static void msg_move_exec(struct uevent_msg *msg)
401 list_move_tail(&msg->node, &exec_list);
403 expected_seqnum = msg->seqnum+1;
404 dbg("moved seq %llu to exec, next expected is %llu",
405 msg->seqnum, expected_seqnum);
408 /* msg queue management routine handles the timeouts and dispatches the events */
409 static void msg_queue_manager(void)
411 struct uevent_msg *loop_msg;
412 struct uevent_msg *tmp_msg;
415 int timeout = event_timeout;
417 dbg("msg queue manager, next expected is %llu", expected_seqnum);
420 list_for_each_entry_safe(loop_msg, tmp_msg, &msg_list, node) {
421 /* move event with expected sequence to the exec list */
422 if (loop_msg->seqnum == expected_seqnum) {
423 msg_move_exec(loop_msg);
427 /* limit timeout during initialization phase */
429 timeout = UDEVD_INIT_EVENT_TIMEOUT;
430 dbg("initialization phase, limit timeout to %i seconds", UDEVD_INIT_EVENT_TIMEOUT);
433 /* move event with expired timeout to the exec list */
434 msg_age = info.uptime - loop_msg->queue_time;
435 dbg("seq %llu is %li seconds old", loop_msg->seqnum, msg_age);
436 if (msg_age >= timeout) {
437 msg_move_exec(loop_msg);
446 /* set timeout for remaining queued events */
447 if (list_empty(&msg_list) == 0) {
448 struct itimerval itv = {{0, 0}, {timeout - msg_age, 0}};
449 dbg("next event expires in %li seconds", timeout - msg_age);
450 setitimer(ITIMER_REAL, &itv, NULL);
454 static struct uevent_msg *get_msg_from_envbuf(const char *buf, int buf_size)
458 struct uevent_msg *msg;
462 msg = malloc(sizeof(struct uevent_msg) + buf_size);
465 memset(msg, 0x00, sizeof(struct uevent_msg) + buf_size);
467 /* copy environment buffer and reconstruct envp */
468 memcpy(msg->envbuf, buf, buf_size);
470 for (i = 0; (bufpos < buf_size) && (i < UEVENT_NUM_ENVP-2); i++) {
474 key = &msg->envbuf[bufpos];
475 keylen = strlen(key);
477 bufpos += keylen + 1;
478 dbg("add '%s' to msg.envp[%i]", msg->envp[i], i);
480 /* remember some keys for further processing */
481 if (strncmp(key, "ACTION=", 7) == 0)
482 msg->action = &key[7];
483 else if (strncmp(key, "DEVPATH=", 8) == 0)
484 msg->devpath = &key[8];
485 else if (strncmp(key, "SUBSYSTEM=", 10) == 0)
486 msg->subsystem = &key[10];
487 else if (strncmp(key, "SEQNUM=", 7) == 0)
488 msg->seqnum = strtoull(&key[7], NULL, 10);
489 else if (strncmp(key, "PHYSDEVPATH=", 12) == 0)
490 msg->physdevpath = &key[12];
491 else if (strncmp(key, "MAJOR=", 6) == 0)
492 major = strtoull(&key[6], NULL, 10);
493 else if (strncmp(key, "MINOR=", 6) == 0)
494 minor = strtoull(&key[6], NULL, 10);
495 else if (strncmp(key, "TIMEOUT=", 8) == 0)
496 msg->timeout = strtoull(&key[8], NULL, 10);
498 msg->devt = makedev(major, minor);
499 msg->envp[i++] = "UDEVD_EVENT=1";
505 /* receive the udevd message from userspace */
506 static struct uevent_msg *get_udevd_msg(void)
508 static struct udevd_msg usend_msg;
509 struct uevent_msg *msg;
512 struct cmsghdr *cmsg;
515 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
519 memset(&usend_msg, 0x00, sizeof(struct udevd_msg));
520 iov.iov_base = &usend_msg;
521 iov.iov_len = sizeof(struct udevd_msg);
523 memset(&smsg, 0x00, sizeof(struct msghdr));
526 smsg.msg_control = cred_msg;
527 smsg.msg_controllen = sizeof(cred_msg);
529 size = recvmsg(udevd_sock, &smsg, 0);
532 dbg("unable to receive udevd message");
535 cmsg = CMSG_FIRSTHDR(&smsg);
536 cred = (struct ucred *) CMSG_DATA(cmsg);
538 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
539 info("no sender credentials received, message ignored");
543 if (cred->uid != 0) {
544 info("sender uid=%i, message ignored", cred->uid);
548 if (strncmp(usend_msg.magic, UDEV_MAGIC, sizeof(UDEV_MAGIC)) != 0 ) {
549 info("message magic '%s' doesn't match, ignore it", usend_msg.magic);
553 switch (usend_msg.type) {
554 case UDEVD_UEVENT_UDEVSEND:
555 case UDEVD_UEVENT_INITSEND:
556 info("udevd event message received");
557 envbuf_size = size - offsetof(struct udevd_msg, envbuf);
558 dbg("envbuf_size=%i", envbuf_size);
559 msg = get_msg_from_envbuf(usend_msg.envbuf, envbuf_size);
562 msg->type = usend_msg.type;
564 case UDEVD_STOP_EXEC_QUEUE:
565 info("udevd message (STOP_EXEC_QUEUE) received");
568 case UDEVD_START_EXEC_QUEUE:
569 info("udevd message (START_EXEC_QUEUE) received");
571 exec_queue_manager();
573 case UDEVD_SET_LOG_LEVEL:
574 intval = (int *) usend_msg.envbuf;
575 info("udevd message (SET_LOG_PRIORITY) received, udev_log_priority=%i", *intval);
576 udev_log_priority = *intval;
578 case UDEVD_SET_MAX_CHILDS:
579 intval = (int *) usend_msg.envbuf;
580 info("udevd message (UDEVD_SET_MAX_CHILDS) received, max_childs=%i", *intval);
581 max_childs = *intval;
584 dbg("unknown message type");
589 /* receive the kernel user event message and do some sanity checks */
590 static struct uevent_msg *get_netlink_msg(void)
592 struct uevent_msg *msg;
595 static char buffer[UEVENT_BUFFER_SIZE + 512];
598 size = recv(uevent_netlink_sock, &buffer, sizeof(buffer), 0);
601 dbg("unable to receive udevd message");
605 if ((size_t)size > sizeof(buffer)-1)
606 size = sizeof(buffer)-1;
608 dbg("uevent_size=%zi", size);
610 /* start of event payload */
611 bufpos = strlen(buffer)+1;
612 msg = get_msg_from_envbuf(&buffer[bufpos], size-bufpos);
615 msg->type = UDEVD_UEVENT_NETLINK;
617 /* validate message */
618 pos = strchr(buffer, '@');
620 dbg("invalid uevent '%s'", buffer);
626 if (msg->action == NULL) {
627 dbg("no ACTION in payload found, skip event '%s'", buffer);
632 if (strcmp(msg->action, buffer) != 0) {
633 dbg("ACTION in payload does not match uevent, skip event '%s'", buffer);
641 static void asmlinkage sig_handler(int signum)
651 /* set flag, then write to pipe if needed */
656 /* set flag, then write to pipe if needed */
657 sigchilds_waiting = 1;
663 /* if pipe is empty, write to pipe to force select to return
664 * immediately when it gets called
667 rc = write(pipefds[1],&signum,sizeof(signum));
673 static void udev_done(int pid)
675 /* find msg associated with pid and delete it */
676 struct uevent_msg *msg;
679 list_for_each_entry(msg, &running_list, node) {
680 if (msg->pid == pid) {
682 info("seq %llu exit, %ld seconds old", msg->seqnum, info.uptime - msg->queue_time);
683 msg_queue_delete(msg);
685 /* we want to run the exec queue manager since there may
686 * be events waiting with the devpath of the one that
695 static void reap_sigchilds(void)
698 int pid = waitpid(-1, NULL, WNOHANG);
699 if ((pid == -1) || (pid == 0))
705 /* just read everything from the pipe and clear the flag,
706 * the flags was set in the signal handler
708 static void user_sighandler(void)
713 int rc = read(pipefds[0], &sig, sizeof(sig));
721 static int init_udevd_socket(void)
723 struct sockaddr_un saddr;
724 const int buffersize = 1024 * 1024;
726 const int feature_on = 1;
729 memset(&saddr, 0x00, sizeof(saddr));
730 saddr.sun_family = AF_LOCAL;
731 /* use abstract namespace for socket path */
732 strcpy(&saddr.sun_path[1], UDEVD_SOCK_PATH);
733 addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(saddr.sun_path+1) + 1;
735 udevd_sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
736 if (udevd_sock == -1) {
737 err("error getting socket, %s", strerror(errno));
741 /* set receive buffersize */
742 setsockopt(udevd_sock, SOL_SOCKET, SO_RCVBUF, &buffersize, sizeof(buffersize));
744 /* the bind takes care of ensuring only one copy running */
745 retval = bind(udevd_sock, (struct sockaddr *) &saddr, addrlen);
747 err("bind failed, %s", strerror(errno));
752 /* enable receiving of the sender credentials */
753 setsockopt(udevd_sock, SOL_SOCKET, SO_PASSCRED, &feature_on, sizeof(feature_on));
758 static int init_uevent_netlink_sock(void)
760 struct sockaddr_nl snl;
761 const int buffersize = 1024 * 1024;
764 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
765 snl.nl_family = AF_NETLINK;
766 snl.nl_pid = getpid();
767 snl.nl_groups = 0xffffffff;
769 uevent_netlink_sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
770 if (uevent_netlink_sock == -1) {
771 dbg("error getting socket, %s", strerror(errno));
775 /* set receive buffersize */
776 setsockopt(uevent_netlink_sock, SOL_SOCKET, SO_RCVBUF, &buffersize, sizeof(buffersize));
778 retval = bind(uevent_netlink_sock, (struct sockaddr *) &snl,
779 sizeof(struct sockaddr_nl));
781 dbg("bind failed, %s", strerror(errno));
782 close(uevent_netlink_sock);
783 uevent_netlink_sock = -1;
790 int main(int argc, char *argv[], char *envp[])
795 struct sigaction act;
798 int uevent_netlink_active = 0;
802 logging_init("udevd");
804 dbg("version %s", UDEV_VERSION);
807 err("need to be root, exit");
811 for (i = 1 ; i < argc; i++) {
813 if (strcmp(arg, "--daemon") == 0 || strcmp(arg, "-d") == 0) {
814 info("will daemonize");
817 if (strcmp(arg, "--stop-exec-queue") == 0) {
818 info("will not execute events until START_EXEC_QUEUE is received");
828 dbg("damonized fork running");
831 err("fork of daemon failed");
839 /* become session leader */
841 dbg("our session is %d", sid);
844 umask(umask(077) | 022);
846 /*set a reasonable scheduling priority for the daemon */
847 setpriority(PRIO_PROCESS, 0, UDEVD_PRIORITY);
849 /* Set fds to dev/null */
850 fd = open( "/dev/null", O_RDWR );
858 err("error opening /dev/null %s", strerror(errno));
860 /* setup signal handler pipe */
861 retval = pipe(pipefds);
863 err("error getting pipes: %s", strerror(errno));
867 retval = fcntl(pipefds[0], F_SETFL, O_NONBLOCK);
869 err("error fcntl on read pipe: %s", strerror(errno));
872 retval = fcntl(pipefds[0], F_SETFD, FD_CLOEXEC);
874 err("error fcntl on read pipe: %s", strerror(errno));
876 retval = fcntl(pipefds[1], F_SETFL, O_NONBLOCK);
878 err("error fcntl on write pipe: %s", strerror(errno));
881 retval = fcntl(pipefds[1], F_SETFD, FD_CLOEXEC);
883 err("error fcntl on write pipe: %s", strerror(errno));
885 /* set signal handlers */
886 memset(&act, 0x00, sizeof(struct sigaction));
887 act.sa_handler = (void (*)(int)) sig_handler;
888 sigemptyset(&act.sa_mask);
889 act.sa_flags = SA_RESTART;
890 sigaction(SIGINT, &act, NULL);
891 sigaction(SIGTERM, &act, NULL);
892 sigaction(SIGALRM, &act, NULL);
893 sigaction(SIGCHLD, &act, NULL);
895 if (init_uevent_netlink_sock() < 0) {
896 dbg("uevent socket not available");
899 if (init_udevd_socket() < 0) {
900 if (errno == EADDRINUSE)
901 dbg("another udevd running, exit");
903 dbg("error initialising udevd socket: %s", strerror(errno));
908 /* override of forked udev binary, used for testing */
909 udev_bin = getenv("UDEV_BIN");
910 if (udev_bin != NULL)
911 info("udev binary is set to '%s'", udev_bin);
915 /* init of expected_seqnum value */
916 value = getenv("UDEVD_EXPECTED_SEQNUM");
918 expected_seqnum = strtoull(value, NULL, 10);
919 info("initialize expected_seqnum to %llu", expected_seqnum);
922 /* timeout to wait for missing events */
923 value = getenv("UDEVD_EVENT_TIMEOUT");
925 event_timeout = strtoul(value, NULL, 10);
927 event_timeout = UDEVD_EVENT_TIMEOUT;
928 info("initialize event_timeout to %u", event_timeout);
930 /* maximum limit of forked childs */
931 value = getenv("UDEVD_MAX_CHILDS");
933 max_childs = strtoul(value, NULL, 10);
935 max_childs = UDEVD_MAX_CHILDS;
936 info("initialize max_childs to %u", max_childs);
938 /* start to throttle forking if maximum number of _running_ childs is reached */
939 value = getenv("UDEVD_MAX_CHILDS_RUNNING");
941 max_childs_running = strtoull(value, NULL, 10);
943 max_childs_running = UDEVD_MAX_CHILDS_RUNNING;
944 info("initialize max_childs_running to %u", max_childs_running);
947 FD_SET(udevd_sock, &readfds);
948 if (uevent_netlink_sock != -1)
949 FD_SET(uevent_netlink_sock, &readfds);
950 FD_SET(pipefds[0], &readfds);
951 maxsockplus = udevd_sock+1;
953 struct uevent_msg *msg;
955 fd_set workreadfds = readfds;
956 retval = select(maxsockplus, &workreadfds, NULL, NULL, NULL);
960 dbg("error in select: %s", strerror(errno));
964 if (FD_ISSET(udevd_sock, &workreadfds)) {
965 msg = get_udevd_msg();
967 /* discard kernel messages if netlink is active */
968 if (uevent_netlink_active && msg->type == UDEVD_UEVENT_UDEVSEND && msg->seqnum != 0) {
969 dbg("skip uevent_helper message, netlink is active");
973 msg_queue_insert(msg);
977 if (FD_ISSET(uevent_netlink_sock, &workreadfds)) {
978 msg = get_netlink_msg();
980 msg_queue_insert(msg);
981 /* disable udevsend with first netlink message */
982 if (!uevent_netlink_active) {
983 info("uevent_nl message received, disable udevsend messages");
984 uevent_netlink_active = 1;
989 if (FD_ISSET(pipefds[0], &workreadfds))
992 if (sigchilds_waiting) {
993 sigchilds_waiting = 0;
1003 /* clean up running_list before calling exec_queue_manager() */
1004 if (sigchilds_waiting) {
1005 sigchilds_waiting = 0;
1011 exec_queue_manager();