chiark / gitweb /
[PATCH] udevd race conditions and performance, assorted cleanups
[elogind.git] / udevd.c
diff --git a/udevd.c b/udevd.c
index 476cb64309a2ab9f71caf3ed4f29fb8e23f49516..3b91a458da7f5bebcfca0547f3ec92a7df46bc84 100644 (file)
--- a/udevd.c
+++ b/udevd.c
 #include <sys/socket.h>
 #include <sys/un.h>
 #include <sys/time.h>
+#include <fcntl.h>
 
 #include "list.h"
 #include "udev.h"
+#include "udev_lib.h"
 #include "udev_version.h"
 #include "udevd.h"
 #include "logging.h"
 
+static int pipefds[2];
 static int expected_seqnum = 0;
 volatile static int children_waiting;
-volatile static int msg_q_timeout;
+volatile static int run_msg_q;
+volatile static int sig_flag;
+static int run_exec_q;
 
-LIST_HEAD(msg_list);
-LIST_HEAD(exec_list);
-LIST_HEAD(running_list);
+static LIST_HEAD(msg_list);
+static LIST_HEAD(exec_list);
+static LIST_HEAD(running_list);
 
 static void exec_queue_manager(void);
 static void msg_queue_manager(void);
+static void user_sighandler(void);
+static void reap_kids(void);
 
-unsigned char logname[42];
-
-int log_ok(void)
+#ifdef LOG
+unsigned char logname[LOGNAME_SIZE];
+void log_message (int level, const char *format, ...)
 {
-       return 1;
+       va_list args;
+
+       va_start(args, format);
+       vsyslog(level, format, args);
+       va_end(args);
 }
+#endif
 
 static void msg_dump_queue(void)
 {
+#ifdef DEBUG
        struct hotplug_msg *msg;
 
        list_for_each_entry(msg, &msg_list, list)
                dbg("sequence %d in queue", msg->seqnum);
+#endif
 }
 
 static void msg_dump(struct hotplug_msg *msg)
@@ -86,7 +100,6 @@ static void run_queue_delete(struct hotplug_msg *msg)
 {
        list_del(&msg->list);
        free(msg);
-       exec_queue_manager();
 }
 
 /* orders the message in the queue by sequence number */
@@ -94,18 +107,20 @@ static void msg_queue_insert(struct hotplug_msg *msg)
 {
        struct hotplug_msg *loop_msg;
 
-       /* sort message by sequence number into list*/
-       list_for_each_entry(loop_msg, &msg_list, list)
-               if (loop_msg->seqnum > msg->seqnum)
+       /* sort message by sequence number into list. events
+        * will tend to come in order, so scan the list backwards
+        */
+       list_for_each_entry_reverse(loop_msg, &msg_list, list)
+               if (loop_msg->seqnum < msg->seqnum)
                        break;
-       list_add_tail(&msg->list, &loop_msg->list);
+       list_add(&msg->list, &loop_msg->list);
        dbg("queued message seq %d", msg->seqnum);
 
        /* store timestamp of queuing */
        msg->queue_time = time(NULL);
 
        /* run msg queue manager */
-       msg_queue_manager();
+       run_msg_q = 1;
 
        return ;
 }
@@ -114,20 +129,27 @@ static void msg_queue_insert(struct hotplug_msg *msg)
 static void udev_run(struct hotplug_msg *msg)
 {
        pid_t pid;
-       setenv("ACTION", msg->action, 1);
-       setenv("DEVPATH", msg->devpath, 1);
+       char action[ACTION_SIZE];
+       char devpath[DEVPATH_SIZE];
+       char *env[] = { action, devpath, NULL };
+
+       snprintf(action, sizeof(action), "ACTION=%s", msg->action);
+       snprintf(devpath, sizeof(devpath), "DEVPATH=%s", msg->devpath);
 
        pid = fork();
        switch (pid) {
        case 0:
                /* child */
-               execl(UDEV_BIN, "udev", msg->subsystem, NULL);
+               execle(UDEV_BIN, "udev", msg->subsystem, NULL, env);
                dbg("exec of child failed");
                exit(1);
                break;
        case -1:
                dbg("fork of child failed");
                run_queue_delete(msg);
+               /* note: we never managed to run, so we had no impact on 
+                * running_with_devpath(), so don't bother setting run_exec_q
+                */
                break;
        default:
                /* get SIGCHLD in main loop */
@@ -170,7 +192,7 @@ static void exec_queue_manager()
 static void msg_move_exec(struct hotplug_msg *msg)
 {
        list_move_tail(&msg->list, &exec_list);
-       exec_queue_manager();
+       run_exec_q = 1;
        expected_seqnum = msg->seqnum+1;
        dbg("moved seq %d to exec, next expected is %d",
                msg->seqnum, expected_seqnum);
@@ -218,6 +240,11 @@ static void handle_msg(int sock)
 {
        struct hotplug_msg *msg;
        int retval;
+       struct msghdr smsg;
+       struct cmsghdr *cmsg;
+       struct iovec iov;
+       struct ucred *cred;
+       char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
 
        msg = msg_create();
        if (msg == NULL) {
@@ -225,43 +252,86 @@ static void handle_msg(int sock)
                return;
        }
 
-       retval = recv(sock, msg, sizeof(struct hotplug_msg), 0);
+       iov.iov_base = msg;
+       iov.iov_len = sizeof(struct hotplug_msg);
+
+       memset(&smsg, 0x00, sizeof(struct msghdr));
+       smsg.msg_iov = &iov;
+       smsg.msg_iovlen = 1;
+       smsg.msg_control = cred_msg;
+       smsg.msg_controllen = sizeof(cred_msg);
+
+       retval = recvmsg(sock, &smsg, 0);
        if (retval <  0) {
                if (errno != EINTR)
                        dbg("unable to receive message");
                return;
        }
-       
+       cmsg = CMSG_FIRSTHDR(&smsg);
+       cred = (struct ucred *) CMSG_DATA(cmsg);
+
+       if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
+               dbg("no sender credentials received, message ignored");
+               goto skip;
+       }
+
+       if (cred->uid != 0) {
+               dbg("sender uid=%i, message ignored", cred->uid);
+               goto skip;
+       }
+
        if (strncmp(msg->magic, UDEV_MAGIC, sizeof(UDEV_MAGIC)) != 0 ) {
                dbg("message magic '%s' doesn't match, ignore it", msg->magic);
-               free(msg);
-               return;
+               goto skip;
        }
 
        /* if no seqnum is given, we move straight to exec queue */
        if (msg->seqnum == -1) {
                list_add(&msg->list, &exec_list);
-               exec_queue_manager();
+               run_exec_q = 1;
        } else {
                msg_queue_insert(msg);
        }
+       return;
+
+skip:
+       free(msg);
+       return;
 }
 
 static void sig_handler(int signum)
 {
+       int rc;
        switch (signum) {
                case SIGINT:
                case SIGTERM:
                        exit(20 + signum);
                        break;
                case SIGALRM:
-                       msg_q_timeout = 1;
+                       /* set flag, then write to pipe if needed */
+                       run_msg_q = 1;
+                       goto do_write;
                        break;
                case SIGCHLD:
+                       /* set flag, then write to pipe if needed */
                        children_waiting = 1;
+                       goto do_write;
                        break;
                default:
                        dbg("unhandled signal");
+                       return;
+       }
+       
+do_write:
+       /* if pipe is empty, write to pipe to force select to return
+        * immediately when it gets called
+        */
+       if (!sig_flag) {
+               rc = write(pipefds[1],&signum,sizeof(signum));
+               if (rc < 0)
+                       dbg("unable to write to pipe");
+               else
+                       sig_flag = 1;
        }
 }
 
@@ -274,28 +344,89 @@ static void udev_done(int pid)
                if (msg->pid == pid) {
                        dbg("<== exec seq %d came back", msg->seqnum);
                        run_queue_delete(msg);
+                       
+                       /* we want to run the exec queue manager since there may
+                        * be events waiting with the devpath of the one that
+                        * just finished
+                        */
+                       run_exec_q = 1;
                        return;
                }
        }
 }
 
+static void reap_kids()
+{
+       /* reap all dead children */
+       while(1) {
+               int pid = waitpid(-1, 0, WNOHANG);
+               if ((pid == -1) || (pid == 0))
+                       break;
+               udev_done(pid);
+       }
+}
+
+/* just read everything from the pipe and clear the flag,
+ * the useful flags were set in the signal handler
+ */
+static void user_sighandler()
+{
+       int sig;
+       while(1) {
+               int rc = read(pipefds[0],&sig,sizeof(sig));
+               if (rc < 0)
+                       break;
+
+               sig_flag = 0;
+       }
+}
+
+
 int main(int argc, char *argv[])
 {
-       int ssock;
+       int ssock, maxsockplus;
        struct sockaddr_un saddr;
        socklen_t addrlen;
        int retval;
+       const int on = 1;
+       struct sigaction act;
+       fd_set readfds;
 
        init_logging("udevd");
+       dbg("version %s", UDEV_VERSION);
 
-       signal(SIGINT, sig_handler);
-       signal(SIGTERM, sig_handler);
-       signal(SIGALRM, sig_handler);
-       signal(SIGCHLD, sig_handler);
-
-       /* we want these two to interrupt system calls */
-       siginterrupt(SIGALRM, 1);
-       siginterrupt(SIGCHLD, 1);
+       if (getuid() != 0) {
+               dbg("need to be root, exit");
+               exit(1);
+       }
+       
+       /* setup signal handler pipe */
+   retval = pipe(pipefds);
+   if (retval < 0) {
+      dbg("error getting pipes: %s", strerror(errno));
+      exit(1);
+   }
+       
+   retval = fcntl(pipefds[0], F_SETFL, O_NONBLOCK);
+   if (retval < 0) {
+      dbg("fcntl on read pipe: %s", strerror(errno));
+      exit(1);
+   }
+   
+   retval = fcntl(pipefds[1], F_SETFL, O_NONBLOCK);
+   if (retval < 0) {
+      dbg("fcntl on write pipe: %s", strerror(errno));
+      exit(1);
+   }
+
+       /* set signal handlers */
+       act.sa_handler = sig_handler;
+       sigemptyset(&act.sa_mask);
+       act.sa_flags = SA_RESTART;
+       sigaction(SIGINT, &act, NULL);
+       sigaction(SIGTERM, &act, NULL);
+       sigaction(SIGALRM, &act, NULL);
+       sigaction(SIGCHLD, &act, NULL);
 
        memset(&saddr, 0x00, sizeof(saddr));
        saddr.sun_family = AF_LOCAL;
@@ -305,34 +436,64 @@ int main(int argc, char *argv[])
 
        ssock = socket(AF_LOCAL, SOCK_DGRAM, 0);
        if (ssock == -1) {
-               dbg("error getting socket");
+               dbg("error getting socket, exit");
                exit(1);
        }
 
        /* the bind takes care of ensuring only one copy running */
-       retval = bind(ssock, &saddr, addrlen);
+       retval = bind(ssock, (struct sockaddr *) &saddr, addrlen);
        if (retval < 0) {
-               dbg("bind failed\n");
+               dbg("bind failed, exit");
                goto exit;
        }
 
-       while (1) {
-               handle_msg(ssock);
+       /* enable receiving of the sender credentials */
+       setsockopt(ssock, SOL_SOCKET, SO_PASSCRED, &on, sizeof(on));
 
-               while(msg_q_timeout) {
-                       msg_q_timeout = 0;
-                       msg_queue_manager();
+   FD_ZERO(&readfds);
+   FD_SET(ssock, &readfds);
+   FD_SET(pipefds[0], &readfds);
+       maxsockplus = ssock+1;
+       while (1) {
+               fd_set workreadfds = readfds;
+               retval = select(maxsockplus, &workreadfds, NULL, NULL, NULL);
+               
+               if (retval < 0) {
+                       dbg("error in select: %s", strerror(errno));
+                       continue;
                }
-
-               while(children_waiting) {
+               
+               if (FD_ISSET(ssock, &workreadfds))
+                       handle_msg(ssock);
+               
+               if (FD_ISSET(pipefds[0], &workreadfds))
+                       user_sighandler();
+               
+               if (children_waiting) {
                        children_waiting = 0;
-                       /* reap all dead children */
-                       while(1) {
-                               int pid = waitpid(-1, 0, WNOHANG);
-                               if ((pid == -1) || (pid == 0))
-                                       break;
-                               udev_done(pid);
+                       reap_kids();
+               }
+               
+               if (run_msg_q) {
+                       run_msg_q = 0;
+                       msg_queue_manager();
+               }
+               
+               if (run_exec_q) {
+                       
+                       /* this is tricky.  exec_queue_manager() loops over exec_list, and
+                        * calls running_with_devpath(), which loops over running_list. This gives
+                        * O(N*M), which can get *nasty*.  Clean up running_list before
+                        * calling exec_queue_manager().
+                        */
+                       
+                       if (children_waiting) {
+                               children_waiting = 0;
+                               reap_kids();
                        }
+
+                       run_exec_q = 0;
+                       exec_queue_manager();
                }
        }
 exit: