udevd.c

   1 /*
   2  * Copyright (C) 2004-2006 Kay Sievers <kay.sievers@vrfy.org>
   3  * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
   4  *
   5  *      This program is free software; you can redistribute it and/or modify it
   6  *      under the terms of the GNU General Public License as published by the
   7  *      Free Software Foundation version 2 of the License.
   8  *
   9  *      This program is distributed in the hope that it will be useful, but
  10  *      WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *      MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12  *      General Public License for more details.
  13  *
  14  *      You should have received a copy of the GNU General Public License along
  15  *      with this program; if not, write to the Free Software Foundation, Inc.,
  16  *      51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  *
  18  */
  19
  20 #include <stddef.h>
  21 #include <signal.h>
  22 #include <unistd.h>
  23 #include <errno.h>
  24 #include <stdio.h>
  25 #include <stdlib.h>
  26 #include <string.h>
  27 #include <ctype.h>
  28 #include <dirent.h>
  29 #include <fcntl.h>
  30 #include <syslog.h>
  31 #include <time.h>
  32 #include <getopt.h>
  33 #include <sys/select.h>
  34 #include <sys/wait.h>
  35 #include <sys/types.h>
  36 #include <sys/socket.h>
  37 #include <sys/un.h>
  38 #include <sys/stat.h>
  39 #include <sys/ioctl.h>
  40 #include <linux/types.h>
  41 #include <linux/netlink.h>
  42
  43 #include "udev.h"
  44 #include "udev_rules.h"
  45 #include "udevd.h"
  46 #include "udev_selinux.h"
  47
  48 static int debug_trace;
  49 static int verbose;
  50
  51 static struct udev_rules rules;
  52 static int udevd_sock = -1;
  53 static int uevent_netlink_sock = -1;
  54 static int inotify_fd = -1;
  55 static pid_t sid;
  56
  57 static int signal_pipe[2] = {-1, -1};
  58 static volatile int sigchilds_waiting;
  59 static volatile int udev_exit;
  60 static volatile int reload_config;
  61 static int run_exec_q;
  62 static int stop_exec_q;
  63 static int max_childs;
  64 static int max_childs_running;
  65 static char udev_log[32];
  66
  67 static LIST_HEAD(exec_list);
  68 static LIST_HEAD(running_list);
  69
  70
  71 #ifdef USE_LOG
  72 void log_message(int priority, const char *format, ...)
  73 {
  74         va_list args;
  75
  76         if (priority > udev_log_priority)
  77                 return;
  78
  79         if (verbose) {
  80                 printf("[%d] ", (int) getpid());
  81                 va_start(args, format);
  82                 vprintf(format, args);
  83                 va_end(args);
  84                 printf("\n");
  85         } else {
  86                 va_start(args, format);
  87                 vsyslog(priority, format, args);
  88                 va_end(args);
  89         }
  90 }
  91
  92 #endif
  93
  94 static void asmlinkage udev_event_sig_handler(int signum)
  95 {
  96         if (signum == SIGALRM)
  97                 exit(1);
  98 }
  99
 100 static int udev_event_process(struct udevd_uevent_msg *msg)
 101 {
 102         struct sigaction act;
 103         struct udevice *udev;
 104         int i;
 105         int retval;
 106
 107         /* set signal handlers */
 108         memset(&act, 0x00, sizeof(act));
 109         act.sa_handler = (void (*)(int)) udev_event_sig_handler;
 110         sigemptyset (&act.sa_mask);
 111         act.sa_flags = 0;
 112         sigaction(SIGALRM, &act, NULL);
 113
 114         /* reset to default */
 115         act.sa_handler = SIG_DFL;
 116         sigaction(SIGINT, &act, NULL);
 117         sigaction(SIGTERM, &act, NULL);
 118         sigaction(SIGCHLD, &act, NULL);
 119         sigaction(SIGHUP, &act, NULL);
 120
 121         /* trigger timeout to prevent hanging processes */
 122         alarm(UDEV_ALARM_TIMEOUT);
 123
 124         /* reconstruct event environment from message */
 125         for (i = 0; msg->envp[i]; i++)
 126                 putenv(msg->envp[i]);
 127
 128         udev = udev_device_init(NULL);
 129         if (udev == NULL)
 130                 return -1;
 131         strlcpy(udev->action, msg->action, sizeof(udev->action));
 132         sysfs_device_set_values(udev->dev, msg->devpath, msg->subsystem, msg->driver);
 133         udev->devpath_old = msg->devpath_old;
 134         udev->devt = msg->devt;
 135
 136         retval = udev_device_event(&rules, udev);
 137
 138         /* run programs collected by RUN-key*/
 139         if (retval == 0 && !udev->ignore_device && udev_run)
 140                 retval = udev_rules_run(udev);
 141
 142         udev_device_cleanup(udev);
 143         return retval;
 144 }
 145
 146 enum event_state {
 147         EVENT_QUEUED,
 148         EVENT_FINISHED,
 149         EVENT_FAILED,
 150 };
 151
 152 static void export_event_state(struct udevd_uevent_msg *msg, enum event_state state)
 153 {
 154         char filename[PATH_SIZE];
 155         char filename_failed[PATH_SIZE];
 156         size_t start;
 157         struct udevd_uevent_msg *loop_msg;
 158         int fd;
 159
 160         /* location of queue file */
 161         strlcpy(filename, udev_root, sizeof(filename));
 162         strlcat(filename, "/", sizeof(filename));
 163         start = strlcat(filename, EVENT_QUEUE_DIR"/", sizeof(filename));
 164         strlcat(filename, msg->devpath, sizeof(filename));
 165         path_encode(&filename[start], sizeof(filename) - start);
 166
 167         /* location of failed file */
 168         strlcpy(filename_failed, udev_root, sizeof(filename_failed));
 169         strlcat(filename_failed, "/", sizeof(filename_failed));
 170         start = strlcat(filename_failed, EVENT_FAILED_DIR"/", sizeof(filename_failed));
 171         strlcat(filename_failed, msg->devpath, sizeof(filename_failed));
 172         path_encode(&filename_failed[start], sizeof(filename) - start);
 173
 174         switch (state) {
 175         case EVENT_QUEUED:
 176                 unlink(filename_failed);
 177                 delete_path(filename_failed);
 178
 179                 create_path(filename);
 180                 fd = open(filename, O_WRONLY|O_TRUNC|O_CREAT, 0644);
 181                 if (fd > 0)
 182                         close(fd);
 183                 return;
 184         case EVENT_FINISHED:
 185         case EVENT_FAILED:
 186                 if (msg->devpath_old != NULL) {
 187                         /* "move" event - rename failed file to current name, do not delete failed */
 188                         char filename_failed_old[PATH_SIZE];
 189
 190                         strlcpy(filename_failed_old, udev_root, sizeof(filename_failed_old));
 191                         strlcat(filename_failed_old, "/", sizeof(filename_failed_old));
 192                         start = strlcat(filename_failed_old, EVENT_FAILED_DIR"/", sizeof(filename_failed_old));
 193                         strlcat(filename_failed_old, msg->devpath_old, sizeof(filename_failed_old));
 194                         path_encode(&filename_failed_old[start], sizeof(filename) - start);
 195
 196                         if (rename(filename_failed_old, filename_failed) == 0)
 197                                 info("renamed devpath, moved failed state of '%s' to %s'",
 198                                      msg->devpath_old, msg->devpath);
 199                 } else {
 200                         unlink(filename_failed);
 201                         delete_path(filename_failed);
 202                 }
 203
 204                 /* skip if events for the same path are still pending */
 205                 list_for_each_entry(loop_msg, &running_list, node)
 206                         if (loop_msg->devpath && strcmp(loop_msg->devpath, msg->devpath) == 0)
 207                                 return;
 208
 209                 list_for_each_entry(loop_msg, &exec_list, node)
 210                         if (loop_msg->devpath && strcmp(loop_msg->devpath, msg->devpath) == 0)
 211                                 return;
 212
 213                 /* move failed event to the failed directory */
 214                 if (state == EVENT_FAILED) {
 215                         create_path(filename_failed);
 216                         rename(filename, filename_failed);
 217                 } else {
 218                         unlink(filename);
 219                 }
 220
 221                 /* clean up possibly empty queue directory */
 222                 delete_path(filename);
 223
 224                 return;
 225         }
 226 }
 227
 228 static void msg_queue_delete(struct udevd_uevent_msg *msg)
 229 {
 230         list_del(&msg->node);
 231
 232         /* mark as failed, if "add" event returns non-zero */
 233         if (msg->exitstatus && strcmp(msg->action, "add") == 0)
 234                 export_event_state(msg, EVENT_FAILED);
 235         else
 236                 export_event_state(msg, EVENT_FINISHED);
 237
 238         free(msg);
 239 }
 240
 241 static void udev_event_run(struct udevd_uevent_msg *msg)
 242 {
 243         pid_t pid;
 244         int retval;
 245
 246         pid = fork();
 247         switch (pid) {
 248         case 0:
 249                 /* child */
 250                 close(uevent_netlink_sock);
 251                 close(udevd_sock);
 252                 if (inotify_fd >= 0)
 253                         close(inotify_fd);
 254                 close(signal_pipe[READ_END]);
 255                 close(signal_pipe[WRITE_END]);
 256                 logging_close();
 257
 258                 logging_init("udevd-event");
 259                 setpriority(PRIO_PROCESS, 0, UDEV_PRIORITY);
 260
 261                 retval = udev_event_process(msg);
 262                 info("seq %llu finished with %i", msg->seqnum, retval);
 263
 264                 logging_close();
 265                 if (retval)
 266                         exit(1);
 267                 exit(0);
 268         case -1:
 269                 err("fork of child failed: %s", strerror(errno));
 270                 msg_queue_delete(msg);
 271                 break;
 272         default:
 273                 /* get SIGCHLD in main loop */
 274                 info("seq %llu forked, pid [%d], '%s' '%s', %ld seconds old",
 275                      msg->seqnum, pid,  msg->action, msg->subsystem, time(NULL) - msg->queue_time);
 276                 msg->pid = pid;
 277         }
 278 }
 279
 280 static void msg_queue_insert(struct udevd_uevent_msg *msg)
 281 {
 282         char filename[PATH_SIZE];
 283         int fd;
 284
 285         msg->queue_time = time(NULL);
 286
 287         strlcpy(filename, udev_root, sizeof(filename));
 288         strlcat(filename, "/" EVENT_SEQNUM, sizeof(filename));
 289         fd = open(filename, O_WRONLY|O_TRUNC|O_CREAT, 0644);
 290         if (fd >= 0) {
 291                 char str[32];
 292                 int len;
 293
 294                 len = sprintf(str, "%llu\n", msg->seqnum);
 295                 write(fd, str, len);
 296                 close(fd);
 297         }
 298
 299         export_event_state(msg, EVENT_QUEUED);
 300         info("seq %llu forked, '%s' '%s'", msg->seqnum, msg->action, msg->subsystem);
 301
 302         /* run one event after the other in debug mode */
 303         if (debug_trace) {
 304                 list_add_tail(&msg->node, &running_list);
 305                 udev_event_run(msg);
 306                 waitpid(msg->pid, NULL, 0);
 307                 msg_queue_delete(msg);
 308                 return;
 309         }
 310
 311         /* run all events with a timeout set immediately */
 312         if (msg->timeout != 0) {
 313                 list_add_tail(&msg->node, &running_list);
 314                 udev_event_run(msg);
 315                 return;
 316         }
 317
 318         list_add_tail(&msg->node, &exec_list);
 319         run_exec_q = 1;
 320 }
 321
 322 static int mem_size_mb(void)
 323 {
 324         FILE* f;
 325         char buf[4096];
 326         long int memsize = -1;
 327
 328         f = fopen("/proc/meminfo", "r");
 329         if (f == NULL)
 330                 return -1;
 331
 332         while (fgets(buf, sizeof(buf), f) != NULL) {
 333                 long int value;
 334
 335                 if (sscanf(buf, "MemTotal: %ld kB", &value) == 1) {
 336                         memsize = value / 1024;
 337                         break;
 338                 }
 339         }
 340
 341         fclose(f);
 342         return memsize;
 343 }
 344
 345 static int cpu_count(void)
 346 {
 347         FILE* f;
 348         char buf[4096];
 349         int count = 0;
 350
 351         f = fopen("/proc/stat", "r");
 352         if (f == NULL)
 353                 return -1;
 354
 355         while (fgets(buf, sizeof(buf), f) != NULL) {
 356                 if (strncmp(buf, "cpu", 3) == 0 && isdigit(buf[3]))
 357                         count++;
 358         }
 359
 360         fclose(f);
 361         if (count == 0)
 362                 return -1;
 363         return count;
 364 }
 365
 366 static int running_processes(void)
 367 {
 368         FILE* f;
 369         char buf[4096];
 370         int running = -1;
 371
 372         f = fopen("/proc/stat", "r");
 373         if (f == NULL)
 374                 return -1;
 375
 376         while (fgets(buf, sizeof(buf), f) != NULL) {
 377                 int value;
 378
 379                 if (sscanf(buf, "procs_running %u", &value) == 1) {
 380                         running = value;
 381                         break;
 382                 }
 383         }
 384
 385         fclose(f);
 386         return running;
 387 }
 388
 389 /* return the number of process es in our session, count only until limit */
 390 static int running_processes_in_session(pid_t session, int limit)
 391 {
 392         DIR *dir;
 393         struct dirent *dent;
 394         int running = 0;
 395
 396         dir = opendir("/proc");
 397         if (!dir)
 398                 return -1;
 399
 400         /* read process info from /proc */
 401         for (dent = readdir(dir); dent != NULL; dent = readdir(dir)) {
 402                 int f;
 403                 char procdir[64];
 404                 char line[256];
 405                 const char *pos;
 406                 char state;
 407                 pid_t ppid, pgrp, sess;
 408                 int len;
 409
 410                 if (!isdigit(dent->d_name[0]))
 411                         continue;
 412
 413                 snprintf(procdir, sizeof(procdir), "/proc/%s/stat", dent->d_name);
 414                 procdir[sizeof(procdir)-1] = '\0';
 415
 416                 f = open(procdir, O_RDONLY);
 417                 if (f == -1)
 418                         continue;
 419
 420                 len = read(f, line, sizeof(line)-1);
 421                 close(f);
 422
 423                 if (len <= 0)
 424                         continue;
 425                 else
 426                         line[len] = '\0';
 427
 428                 /* skip ugly program name */
 429                 pos = strrchr(line, ')') + 2;
 430                 if (pos == NULL)
 431                         continue;
 432
 433                 if (sscanf(pos, "%c %d %d %d ", &state, &ppid, &pgrp, &sess) != 4)
 434                         continue;
 435
 436                 /* count only processes in our session */
 437                 if (sess != session)
 438                         continue;
 439
 440                 /* count only running, no sleeping processes */
 441                 if (state != 'R')
 442                         continue;
 443
 444                 running++;
 445                 if (limit > 0 && running >= limit)
 446                         break;
 447         }
 448         closedir(dir);
 449
 450         return running;
 451 }
 452
 453 static int compare_devpath(const char *running, const char *waiting)
 454 {
 455         int i;
 456
 457         for (i = 0; i < PATH_SIZE; i++) {
 458                 /* identical device event found */
 459                 if (running[i] == '\0' && waiting[i] == '\0')
 460                         return 1;
 461
 462                 /* parent device event found */
 463                 if (running[i] == '\0' && waiting[i] == '/')
 464                         return 2;
 465
 466                 /* child device event found */
 467                 if (running[i] == '/' && waiting[i] == '\0')
 468                         return 3;
 469
 470                 /* no matching event */
 471                 if (running[i] != waiting[i])
 472                         break;
 473         }
 474
 475         return 0;
 476 }
 477
 478 /* lookup event for identical, parent, child, or physical device */
 479 static int devpath_busy(struct udevd_uevent_msg *msg, int limit)
 480 {
 481         struct udevd_uevent_msg *loop_msg;
 482         int childs_count = 0;
 483
 484         /* check exec-queue which may still contain delayed events we depend on */
 485         list_for_each_entry(loop_msg, &exec_list, node) {
 486                 /* skip ourself and all later events */
 487                 if (loop_msg->seqnum >= msg->seqnum)
 488                         break;
 489
 490                 /* check our old name */
 491                 if (msg->devpath_old != NULL)
 492                         if (strcmp(loop_msg->devpath , msg->devpath_old) == 0)
 493                                 return 2;
 494
 495                 /* check identical, parent, or child device event */
 496                 if (compare_devpath(loop_msg->devpath, msg->devpath) != 0) {
 497                         dbg("%llu, device event still pending %llu (%s)",
 498                             msg->seqnum, loop_msg->seqnum, loop_msg->devpath);
 499                         return 3;
 500                 }
 501
 502                 /* check physical device event (special case of parent) */
 503                 if (msg->physdevpath && msg->action && strcmp(msg->action, "add") == 0)
 504                         if (compare_devpath(loop_msg->devpath, msg->physdevpath) != 0) {
 505                                 dbg("%llu, physical device event still pending %llu (%s)",
 506                                     msg->seqnum, loop_msg->seqnum, loop_msg->devpath);
 507                                 return 4;
 508                         }
 509         }
 510
 511         /* check run queue for still running events */
 512         list_for_each_entry(loop_msg, &running_list, node) {
 513                 if (limit && childs_count++ > limit) {
 514                         dbg("%llu, maximum number (%i) of childs reached", msg->seqnum, childs_count);
 515                         return 1;
 516                 }
 517
 518                 /* check our old name */
 519                 if (msg->devpath_old != NULL)
 520                         if (strcmp(loop_msg->devpath , msg->devpath_old) == 0)
 521                                 return 2;
 522
 523                 /* check identical, parent, or child device event */
 524                 if (compare_devpath(loop_msg->devpath, msg->devpath) != 0) {
 525                         dbg("%llu, device event still running %llu (%s)",
 526                             msg->seqnum, loop_msg->seqnum, loop_msg->devpath);
 527                         return 3;
 528                 }
 529
 530                 /* check physical device event (special case of parent) */
 531                 if (msg->physdevpath && msg->action && strcmp(msg->action, "add") == 0)
 532                         if (compare_devpath(loop_msg->devpath, msg->physdevpath) != 0) {
 533                                 dbg("%llu, physical device event still running %llu (%s)",
 534                                     msg->seqnum, loop_msg->seqnum, loop_msg->devpath);
 535                                 return 4;
 536                         }
 537         }
 538         return 0;
 539 }
 540
 541 /* serializes events for the identical and parent and child devices */
 542 static void msg_queue_manager(void)
 543 {
 544         struct udevd_uevent_msg *loop_msg;
 545         struct udevd_uevent_msg *tmp_msg;
 546         int running;
 547
 548         if (list_empty(&exec_list))
 549                 return;
 550
 551         running = running_processes();
 552         dbg("%d processes runnning on system", running);
 553         if (running < 0)
 554                 running = max_childs_running;
 555
 556         list_for_each_entry_safe(loop_msg, tmp_msg, &exec_list, node) {
 557                 /* check running processes in our session and possibly throttle */
 558                 if (running >= max_childs_running) {
 559                         running = running_processes_in_session(sid, max_childs_running+10);
 560                         dbg("at least %d processes running in session", running);
 561                         if (running >= max_childs_running) {
 562                                 dbg("delay seq %llu, too many processes already running", loop_msg->seqnum);
 563                                 return;
 564                         }
 565                 }
 566
 567                 /* serialize and wait for parent or child events */
 568                 if (devpath_busy(loop_msg, max_childs) != 0) {
 569                         dbg("delay seq %llu (%s)", loop_msg->seqnum, loop_msg->devpath);
 570                         continue;
 571                 }
 572
 573                 /* move event to run list */
 574                 list_move_tail(&loop_msg->node, &running_list);
 575                 udev_event_run(loop_msg);
 576                 running++;
 577                 dbg("moved seq %llu to running list", loop_msg->seqnum);
 578         }
 579 }
 580
 581 static struct udevd_uevent_msg *get_msg_from_envbuf(const char *buf, int buf_size)
 582 {
 583         int bufpos;
 584         int i;
 585         struct udevd_uevent_msg *msg;
 586         char *physdevdriver_key = NULL;
 587         int maj = 0;
 588         int min = 0;
 589
 590         msg = malloc(sizeof(struct udevd_uevent_msg) + buf_size);
 591         if (msg == NULL)
 592                 return NULL;
 593         memset(msg, 0x00, sizeof(struct udevd_uevent_msg) + buf_size);
 594
 595         /* copy environment buffer and reconstruct envp */
 596         memcpy(msg->envbuf, buf, buf_size);
 597         bufpos = 0;
 598         for (i = 0; (bufpos < buf_size) && (i < UEVENT_NUM_ENVP-2); i++) {
 599                 int keylen;
 600                 char *key;
 601
 602                 key = &msg->envbuf[bufpos];
 603                 keylen = strlen(key);
 604                 msg->envp[i] = key;
 605                 bufpos += keylen + 1;
 606                 dbg("add '%s' to msg.envp[%i]", msg->envp[i], i);
 607
 608                 /* remember some keys for further processing */
 609                 if (strncmp(key, "ACTION=", 7) == 0)
 610                         msg->action = &key[7];
 611                 else if (strncmp(key, "DEVPATH=", 8) == 0)
 612                         msg->devpath = &key[8];
 613                 else if (strncmp(key, "SUBSYSTEM=", 10) == 0)
 614                         msg->subsystem = &key[10];
 615                 else if (strncmp(key, "DRIVER=", 7) == 0)
 616                         msg->driver = &key[7];
 617                 else if (strncmp(key, "SEQNUM=", 7) == 0)
 618                         msg->seqnum = strtoull(&key[7], NULL, 10);
 619                 else if (strncmp(key, "DEVPATH_OLD=", 12) == 0)
 620                         msg->devpath_old = &key[12];
 621                 else if (strncmp(key, "PHYSDEVPATH=", 12) == 0)
 622                         msg->physdevpath = &key[12];
 623                 else if (strncmp(key, "PHYSDEVDRIVER=", 14) == 0)
 624                         physdevdriver_key = key;
 625                 else if (strncmp(key, "MAJOR=", 6) == 0)
 626                         maj = strtoull(&key[6], NULL, 10);
 627                 else if (strncmp(key, "MINOR=", 6) == 0)
 628                         min = strtoull(&key[6], NULL, 10);
 629                 else if (strncmp(key, "TIMEOUT=", 8) == 0)
 630                         msg->timeout = strtoull(&key[8], NULL, 10);
 631         }
 632         msg->devt = makedev(maj, min);
 633         msg->envp[i++] = "UDEVD_EVENT=1";
 634
 635         if (msg->driver == NULL && msg->physdevpath == NULL && physdevdriver_key != NULL) {
 636                 /* for older kernels DRIVER is empty for a bus device, export PHYSDEVDRIVER as DRIVER */
 637                 msg->envp[i++] = &physdevdriver_key[7];
 638                 msg->driver = &physdevdriver_key[14];
 639         }
 640
 641         msg->envp[i] = NULL;
 642
 643         if (msg->devpath == NULL || msg->action == NULL) {
 644                 info("DEVPATH or ACTION missing, ignore message");
 645                 free(msg);
 646                 return NULL;
 647         }
 648         return msg;
 649 }
 650
 651 /* receive the udevd message from userspace */
 652 static void get_ctrl_msg(void)
 653 {
 654         struct udevd_ctrl_msg ctrl_msg;
 655         ssize_t size;
 656         struct msghdr smsg;
 657         struct cmsghdr *cmsg;
 658         struct iovec iov;
 659         struct ucred *cred;
 660         char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
 661         int *intval;
 662         char *pos;
 663
 664         memset(&ctrl_msg, 0x00, sizeof(struct udevd_ctrl_msg));
 665         iov.iov_base = &ctrl_msg;
 666         iov.iov_len = sizeof(struct udevd_ctrl_msg);
 667
 668         memset(&smsg, 0x00, sizeof(struct msghdr));
 669         smsg.msg_iov = &iov;
 670         smsg.msg_iovlen = 1;
 671         smsg.msg_control = cred_msg;
 672         smsg.msg_controllen = sizeof(cred_msg);
 673
 674         size = recvmsg(udevd_sock, &smsg, 0);
 675         if (size <  0) {
 676                 if (errno != EINTR)
 677                         err("unable to receive user udevd message: %s", strerror(errno));
 678                 return;
 679         }
 680         cmsg = CMSG_FIRSTHDR(&smsg);
 681         cred = (struct ucred *) CMSG_DATA(cmsg);
 682
 683         if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
 684                 err("no sender credentials received, message ignored");
 685                 return;
 686         }
 687
 688         if (cred->uid != 0) {
 689                 err("sender uid=%i, message ignored", cred->uid);
 690                 return;
 691         }
 692
 693         if (strncmp(ctrl_msg.magic, UDEVD_CTRL_MAGIC, sizeof(UDEVD_CTRL_MAGIC)) != 0 ) {
 694                 err("message magic '%s' doesn't match, ignore it", ctrl_msg.magic);
 695                 return;
 696         }
 697
 698         switch (ctrl_msg.type) {
 699         case UDEVD_CTRL_ENV:
 700                 pos = strchr(ctrl_msg.buf, '=');
 701                 if (pos == NULL) {
 702                         err("wrong key format '%s'", ctrl_msg.buf);
 703                         break;
 704                 }
 705                 pos[0] = '\0';
 706                 if (pos[1] == '\0') {
 707                         info("udevd message (ENV) received, unset '%s'", ctrl_msg.buf);
 708                         unsetenv(ctrl_msg.buf);
 709                 } else {
 710                         info("udevd message (ENV) received, set '%s=%s'", ctrl_msg.buf, &pos[1]);
 711                         setenv(ctrl_msg.buf, &pos[1], 1);
 712                 }
 713                 break;
 714         case UDEVD_CTRL_STOP_EXEC_QUEUE:
 715                 info("udevd message (STOP_EXEC_QUEUE) received");
 716                 stop_exec_q = 1;
 717                 break;
 718         case UDEVD_CTRL_START_EXEC_QUEUE:
 719                 info("udevd message (START_EXEC_QUEUE) received");
 720                 stop_exec_q = 0;
 721                 msg_queue_manager();
 722                 break;
 723         case UDEVD_CTRL_SET_LOG_LEVEL:
 724                 intval = (int *) ctrl_msg.buf;
 725                 info("udevd message (SET_LOG_PRIORITY) received, udev_log_priority=%i", *intval);
 726                 udev_log_priority = *intval;
 727                 sprintf(udev_log, "UDEV_LOG=%i", udev_log_priority);
 728                 putenv(udev_log);
 729                 break;
 730         case UDEVD_CTRL_SET_MAX_CHILDS:
 731                 intval = (int *) ctrl_msg.buf;
 732                 info("udevd message (UDEVD_SET_MAX_CHILDS) received, max_childs=%i", *intval);
 733                 max_childs = *intval;
 734                 break;
 735         case UDEVD_CTRL_SET_MAX_CHILDS_RUNNING:
 736                 intval = (int *) ctrl_msg.buf;
 737                 info("udevd message (UDEVD_SET_MAX_CHILDS_RUNNING) received, max_childs=%i", *intval);
 738                 max_childs_running = *intval;
 739                 break;
 740         case UDEVD_CTRL_RELOAD_RULES:
 741                 info("udevd message (RELOAD_RULES) received");
 742                 reload_config = 1;
 743                 break;
 744         default:
 745                 err("unknown control message type");
 746         }
 747 }
 748
 749 /* receive the kernel user event message and do some sanity checks */
 750 static struct udevd_uevent_msg *get_netlink_msg(void)
 751 {
 752         struct udevd_uevent_msg *msg;
 753         int bufpos;
 754         ssize_t size;
 755         static char buffer[UEVENT_BUFFER_SIZE+512];
 756         char *pos;
 757
 758         size = recv(uevent_netlink_sock, &buffer, sizeof(buffer), 0);
 759         if (size <  0) {
 760                 if (errno != EINTR)
 761                         err("unable to receive kernel netlink message: %s", strerror(errno));
 762                 return NULL;
 763         }
 764
 765         if ((size_t)size > sizeof(buffer)-1)
 766                 size = sizeof(buffer)-1;
 767         buffer[size] = '\0';
 768         dbg("uevent_size=%zi", size);
 769
 770         /* start of event payload */
 771         bufpos = strlen(buffer)+1;
 772         msg = get_msg_from_envbuf(&buffer[bufpos], size-bufpos);
 773         if (msg == NULL)
 774                 return NULL;
 775
 776         /* validate message */
 777         pos = strchr(buffer, '@');
 778         if (pos == NULL) {
 779                 err("invalid uevent '%s'", buffer);
 780                 free(msg);
 781                 return NULL;
 782         }
 783         pos[0] = '\0';
 784
 785         if (msg->action == NULL) {
 786                 info("no ACTION in payload found, skip event '%s'", buffer);
 787                 free(msg);
 788                 return NULL;
 789         }
 790
 791         if (strcmp(msg->action, buffer) != 0) {
 792                 err("ACTION in payload does not match uevent, skip event '%s'", buffer);
 793                 free(msg);
 794                 return NULL;
 795         }
 796
 797         return msg;
 798 }
 799
 800 static void asmlinkage sig_handler(int signum)
 801 {
 802         switch (signum) {
 803                 case SIGINT:
 804                 case SIGTERM:
 805                         udev_exit = 1;
 806                         break;
 807                 case SIGCHLD:
 808                         /* set flag, then write to pipe if needed */
 809                         sigchilds_waiting = 1;
 810                         break;
 811                 case SIGHUP:
 812                         reload_config = 1;
 813                         break;
 814         }
 815
 816         /* write to pipe, which will wakeup select() in our mainloop */
 817         write(signal_pipe[WRITE_END], "", 1);
 818 }
 819
 820 static void udev_done(int pid, int exitstatus)
 821 {
 822         /* find msg associated with pid and delete it */
 823         struct udevd_uevent_msg *msg;
 824
 825         list_for_each_entry(msg, &running_list, node) {
 826                 if (msg->pid == pid) {
 827                         info("seq %llu, pid [%d] exit with %i, %ld seconds old", msg->seqnum, msg->pid,
 828                              exitstatus, time(NULL) - msg->queue_time);
 829                         msg->exitstatus = exitstatus;
 830                         msg_queue_delete(msg);
 831
 832                         /* there may be events waiting with the same devpath */
 833                         run_exec_q = 1;
 834                         return;
 835                 }
 836         }
 837 }
 838
 839 static void reap_sigchilds(void)
 840 {
 841         pid_t pid;
 842         int status;
 843
 844         while (1) {
 845                 pid = waitpid(-1, &status, WNOHANG);
 846                 if (pid <= 0)
 847                         break;
 848                 if (WIFEXITED(status))
 849                         status = WEXITSTATUS(status);
 850                 else if (WIFSIGNALED(status))
 851                         status = WTERMSIG(status) + 128;
 852                 else
 853                         status = 0;
 854                 udev_done(pid, status);
 855         }
 856 }
 857
 858 static int init_udevd_socket(void)
 859 {
 860         struct sockaddr_un saddr;
 861         socklen_t addrlen;
 862         const int feature_on = 1;
 863         int retval;
 864
 865         memset(&saddr, 0x00, sizeof(saddr));
 866         saddr.sun_family = AF_LOCAL;
 867         /* use abstract namespace for socket path */
 868         strcpy(&saddr.sun_path[1], UDEVD_CTRL_SOCK_PATH);
 869         addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(saddr.sun_path+1) + 1;
 870
 871         udevd_sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
 872         if (udevd_sock == -1) {
 873                 err("error getting socket: %s", strerror(errno));
 874                 return -1;
 875         }
 876
 877         /* the bind takes care of ensuring only one copy running */
 878         retval = bind(udevd_sock, (struct sockaddr *) &saddr, addrlen);
 879         if (retval < 0) {
 880                 err("bind failed: %s", strerror(errno));
 881                 close(udevd_sock);
 882                 udevd_sock = -1;
 883                 return -1;
 884         }
 885
 886         /* enable receiving of the sender credentials */
 887         setsockopt(udevd_sock, SOL_SOCKET, SO_PASSCRED, &feature_on, sizeof(feature_on));
 888
 889         return 0;
 890 }
 891
 892 static int init_uevent_netlink_sock(void)
 893 {
 894         struct sockaddr_nl snl;
 895         const int buffersize = 16 * 1024 * 1024;
 896         int retval;
 897
 898         memset(&snl, 0x00, sizeof(struct sockaddr_nl));
 899         snl.nl_family = AF_NETLINK;
 900         snl.nl_pid = getpid();
 901         snl.nl_groups = 1;
 902
 903         uevent_netlink_sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
 904         if (uevent_netlink_sock == -1) {
 905                 err("error getting socket: %s", strerror(errno));
 906                 return -1;
 907         }
 908
 909         /* set receive buffersize */
 910         setsockopt(uevent_netlink_sock, SOL_SOCKET, SO_RCVBUFFORCE, &buffersize, sizeof(buffersize));
 911
 912         retval = bind(uevent_netlink_sock, (struct sockaddr *) &snl, sizeof(struct sockaddr_nl));
 913         if (retval < 0) {
 914                 err("bind failed: %s", strerror(errno));
 915                 close(uevent_netlink_sock);
 916                 uevent_netlink_sock = -1;
 917                 return -1;
 918         }
 919         return 0;
 920 }
 921
 922 static void export_initial_seqnum(void)
 923 {
 924         char filename[PATH_SIZE];
 925         int fd;
 926         char seqnum[32];
 927         ssize_t len = 0;
 928
 929         strlcpy(filename, sysfs_path, sizeof(filename));
 930         strlcat(filename, "/kernel/uevent_seqnum", sizeof(filename));
 931         fd = open(filename, O_RDONLY);
 932         if (fd >= 0) {
 933                 len = read(fd, seqnum, sizeof(seqnum)-1);
 934                 close(fd);
 935         }
 936         if (len <= 0) {
 937                 strcpy(seqnum, "0\n");
 938                 len = 3;
 939         }
 940         strlcpy(filename, udev_root, sizeof(filename));
 941         strlcat(filename, "/" EVENT_SEQNUM, sizeof(filename));
 942         create_path(filename);
 943         fd = open(filename, O_WRONLY|O_TRUNC|O_CREAT, 0644);
 944         if (fd >= 0) {
 945                 write(fd, seqnum, len);
 946                 close(fd);
 947         }
 948 }
 949
 950 int main(int argc, char *argv[], char *envp[])
 951 {
 952         int retval;
 953         int fd;
 954         struct sigaction act;
 955         fd_set readfds;
 956         const char *value;
 957         int daemonize = 0;
 958         int option;
 959         static const struct option options[] = {
 960                 { "daemon", 0, NULL, 'd' },
 961                 { "debug-trace", 0, NULL, 't' },
 962                 { "verbose", 0, NULL, 'v' },
 963                 { "help", 0, NULL, 'h' },
 964                 { "version", 0, NULL, 'V' },
 965                 {}
 966         };
 967         int rc = 1;
 968         int maxfd;
 969
 970         logging_init("udevd");
 971         udev_config_init();
 972         selinux_init();
 973         dbg("version %s", UDEV_VERSION);
 974
 975         while (1) {
 976                 option = getopt_long(argc, argv, "dtvhV", options, NULL);
 977                 if (option == -1)
 978                         break;
 979
 980                 switch (option) {
 981                 case 'd':
 982                         daemonize = 1;
 983                         break;
 984                 case 't':
 985                         debug_trace = 1;
 986                         break;
 987                 case 'v':
 988                         verbose = 1;
 989                         if (udev_log_priority < LOG_INFO)
 990                                 udev_log_priority = LOG_INFO;
 991                         break;
 992                 case 'h':
 993                         printf("Usage: udevd [--help] [--daemon] [--debug-trace] [--verbose] [--version]\n");
 994                         goto exit;
 995                 case 'V':
 996                         printf("%s\n", UDEV_VERSION);
 997                         goto exit;
 998                 default:
 999                         goto exit;
1000                 }
1001         }
1002
1003         if (getuid() != 0) {
1004                 fprintf(stderr, "root privileges required\n");
1005                 err("root privileges required");
1006                 goto exit;
1007         }
1008
1009         /* make sure std{in,out,err} fd's are in a sane state */
1010         fd = open("/dev/null", O_RDWR);
1011         if (fd < 0) {
1012                 fprintf(stderr, "cannot open /dev/null\n");
1013                 err("cannot open /dev/null");
1014         }
1015         if (fd > STDIN_FILENO)
1016                 dup2(fd, STDIN_FILENO);
1017         if (write(STDOUT_FILENO, 0, 0) < 0)
1018                 dup2(fd, STDOUT_FILENO);
1019         if (write(STDERR_FILENO, 0, 0) < 0)
1020                 dup2(fd, STDERR_FILENO);
1021
1022         /* init sockets to receive events */
1023         if (init_udevd_socket() < 0) {
1024                 if (errno == EADDRINUSE) {
1025                         fprintf(stderr, "another udev daemon already running\n");
1026                         err("another udev daemon already running");
1027                         rc = 1;
1028                 } else {
1029                         fprintf(stderr, "error initializing udevd socket\n");
1030                         err("error initializing udevd socket");
1031                         rc = 2;
1032                 }
1033                 goto exit;
1034         }
1035
1036         if (init_uevent_netlink_sock() < 0) {
1037                 fprintf(stderr, "error initializing netlink socket\n");
1038                 err("error initializing netlink socket");
1039                 rc = 3;
1040                 goto exit;
1041         }
1042
1043         /* setup signal handler pipe */
1044         retval = pipe(signal_pipe);
1045         if (retval < 0) {
1046                 err("error getting pipes: %s", strerror(errno));
1047                 goto exit;
1048         }
1049
1050         retval = fcntl(signal_pipe[READ_END], F_GETFL, 0);
1051         if (retval < 0) {
1052                 err("error fcntl on read pipe: %s", strerror(errno));
1053                 goto exit;
1054         }
1055         retval = fcntl(signal_pipe[READ_END], F_SETFL, retval | O_NONBLOCK);
1056         if (retval < 0) {
1057                 err("error fcntl on read pipe: %s", strerror(errno));
1058                 goto exit;
1059         }
1060
1061         retval = fcntl(signal_pipe[WRITE_END], F_GETFL, 0);
1062         if (retval < 0) {
1063                 err("error fcntl on write pipe: %s", strerror(errno));
1064                 goto exit;
1065         }
1066         retval = fcntl(signal_pipe[WRITE_END], F_SETFL, retval | O_NONBLOCK);
1067         if (retval < 0) {
1068                 err("error fcntl on write pipe: %s", strerror(errno));
1069                 goto exit;
1070         }
1071
1072         /* parse the rules and keep them in memory */
1073         sysfs_init();
1074         udev_rules_init(&rules, 1);
1075
1076         export_initial_seqnum();
1077
1078         if (daemonize) {
1079                 pid_t pid;
1080
1081                 pid = fork();
1082                 switch (pid) {
1083                 case 0:
1084                         dbg("daemonized fork running");
1085                         break;
1086                 case -1:
1087                         err("fork of daemon failed: %s", strerror(errno));
1088                         rc = 4;
1089                         goto exit;
1090                 default:
1091                         dbg("child [%u] running, parent exits", pid);
1092                         rc = 0;
1093                         goto exit;
1094                 }
1095         }
1096
1097         /* redirect std{out,err} fd's */
1098         if (!verbose)
1099                 dup2(fd, STDOUT_FILENO);
1100         dup2(fd, STDERR_FILENO);
1101         if (fd > STDERR_FILENO)
1102                 close(fd);
1103
1104         /* set scheduling priority for the daemon */
1105         setpriority(PRIO_PROCESS, 0, UDEVD_PRIORITY);
1106
1107         chdir("/");
1108         umask(022);
1109
1110         /* become session leader */
1111         sid = setsid();
1112         dbg("our session is %d", sid);
1113
1114         /* OOM_DISABLE == -17 */
1115         fd = open("/proc/self/oom_adj", O_RDWR);
1116         if (fd < 0)
1117                 err("error disabling OOM: %s", strerror(errno));
1118         else {
1119                 write(fd, "-17", 3);
1120                 close(fd);
1121         }
1122
1123         /* set signal handlers */
1124         memset(&act, 0x00, sizeof(struct sigaction));
1125         act.sa_handler = (void (*)(int)) sig_handler;
1126         sigemptyset(&act.sa_mask);
1127         act.sa_flags = SA_RESTART;
1128         sigaction(SIGINT, &act, NULL);
1129         sigaction(SIGTERM, &act, NULL);
1130         sigaction(SIGCHLD, &act, NULL);
1131         sigaction(SIGHUP, &act, NULL);
1132
1133         /* watch rules directory */
1134         inotify_fd = inotify_init();
1135         if (inotify_fd >= 0) {
1136                 char filename[PATH_MAX];
1137
1138                 inotify_add_watch(inotify_fd, udev_rules_dir, IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1139
1140                 /* watch dynamic rules directory */
1141                 strlcpy(filename, udev_root, sizeof(filename));
1142                 strlcat(filename, "/"RULES_DYN_DIR, sizeof(filename));
1143                 inotify_add_watch(inotify_fd, filename, IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1144         } else if (errno == ENOSYS)
1145                 err("the kernel does not support inotify, udevd can't monitor rules file changes");
1146         else
1147                 err("inotify_init failed: %s", strerror(errno));
1148
1149         /* maximum limit of forked childs */
1150         value = getenv("UDEVD_MAX_CHILDS");
1151         if (value)
1152                 max_childs = strtoul(value, NULL, 10);
1153         else {
1154                 int memsize = mem_size_mb();
1155                 if (memsize > 0)
1156                         max_childs = 128 + (memsize / 4);
1157                 else
1158                         max_childs = UDEVD_MAX_CHILDS;
1159         }
1160         info("initialize max_childs to %u", max_childs);
1161
1162         /* start to throttle forking if maximum number of _running_ childs is reached */
1163         value = getenv("UDEVD_MAX_CHILDS_RUNNING");
1164         if (value)
1165                 max_childs_running = strtoull(value, NULL, 10);
1166         else {
1167                 int cpus = cpu_count();
1168                 if (cpus > 0)
1169                         max_childs_running = 8 + (8 * cpus);
1170                 else
1171                         max_childs_running = UDEVD_MAX_CHILDS_RUNNING;
1172         }
1173         info("initialize max_childs_running to %u", max_childs_running);
1174
1175         /* clear environment for forked event processes */
1176         clearenv();
1177
1178         /* export log_priority , as called programs may want to follow that setting */
1179         sprintf(udev_log, "UDEV_LOG=%i", udev_log_priority);
1180         putenv(udev_log);
1181         if (debug_trace)
1182                 putenv("DEBUG=1");
1183
1184         maxfd = udevd_sock;
1185         maxfd = UDEV_MAX(maxfd, uevent_netlink_sock);
1186         maxfd = UDEV_MAX(maxfd, signal_pipe[READ_END]);
1187         maxfd = UDEV_MAX(maxfd, inotify_fd);
1188
1189         while (!udev_exit) {
1190                 struct udevd_uevent_msg *msg;
1191                 int fdcount;
1192
1193                 FD_ZERO(&readfds);
1194                 FD_SET(signal_pipe[READ_END], &readfds);
1195                 FD_SET(udevd_sock, &readfds);
1196                 FD_SET(uevent_netlink_sock, &readfds);
1197                 if (inotify_fd >= 0)
1198                         FD_SET(inotify_fd, &readfds);
1199
1200                 fdcount = select(maxfd+1, &readfds, NULL, NULL, NULL);
1201                 if (fdcount < 0) {
1202                         if (errno != EINTR)
1203                                 err("error in select: %s", strerror(errno));
1204                         continue;
1205                 }
1206
1207                 /* get control message */
1208                 if (FD_ISSET(udevd_sock, &readfds))
1209                         get_ctrl_msg();
1210
1211                 /* get netlink message */
1212                 if (FD_ISSET(uevent_netlink_sock, &readfds)) {
1213                         msg = get_netlink_msg();
1214                         if (msg)
1215                                 msg_queue_insert(msg);
1216                 }
1217
1218                 /* received a signal, clear our notification pipe */
1219                 if (FD_ISSET(signal_pipe[READ_END], &readfds)) {
1220                         char buf[256];
1221
1222                         read(signal_pipe[READ_END], &buf, sizeof(buf));
1223                 }
1224
1225                 /* rules directory inotify watch */
1226                 if ((inotify_fd >= 0) && FD_ISSET(inotify_fd, &readfds)) {
1227                         int nbytes;
1228
1229                         /* discard all possible events, we can just reload the config */
1230                         if ((ioctl(inotify_fd, FIONREAD, &nbytes) == 0) && nbytes > 0) {
1231                                 char *buf;
1232
1233                                 reload_config = 1;
1234                                 buf = malloc(nbytes);
1235                                 if (buf == NULL) {
1236                                         err("error getting buffer for inotify, disable watching");
1237                                         close(inotify_fd);
1238                                         inotify_fd = -1;
1239                                 }
1240                                 read(inotify_fd, buf, nbytes);
1241                                 free(buf);
1242                         }
1243                 }
1244
1245                 /* rules changed, set by inotify or a HUP signal */
1246                 if (reload_config) {
1247                         reload_config = 0;
1248                         udev_rules_cleanup(&rules);
1249                         udev_rules_init(&rules, 1);
1250                 }
1251
1252                 /* forked child has returned */
1253                 if (sigchilds_waiting) {
1254                         sigchilds_waiting = 0;
1255                         reap_sigchilds();
1256                 }
1257
1258                 if (run_exec_q) {
1259                         run_exec_q = 0;
1260                         if (!stop_exec_q)
1261                                 msg_queue_manager();
1262                 }
1263         }
1264         rc = 0;
1265
1266 exit:
1267         udev_rules_cleanup(&rules);
1268         sysfs_cleanup();
1269         selinux_exit();
1270
1271         if (signal_pipe[READ_END] >= 0)
1272                 close(signal_pipe[READ_END]);
1273         if (signal_pipe[WRITE_END] >= 0)
1274                 close(signal_pipe[WRITE_END]);
1275
1276         if (udevd_sock >= 0)
1277                 close(udevd_sock);
1278         if (inotify_fd >= 0)
1279                 close(inotify_fd);
1280         if (uevent_netlink_sock >= 0)
1281                 close(uevent_netlink_sock);
1282
1283         logging_close();
1284
1285         return rc;
1286 }