1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25 
  26 /*
  27  * fork.c - safe forking for svc.startd
  28  *
  29  * fork_configd() and fork_sulogin() are related, special cases that handle the
  30  * spawning of specific client processes for svc.startd.
  31  */
  32 
  33 #include <sys/contract/process.h>
  34 #include <sys/corectl.h>
  35 #include <sys/ctfs.h>
  36 #include <sys/stat.h>
  37 #include <sys/types.h>
  38 #include <sys/uio.h>
  39 #include <sys/wait.h>
  40 #include <assert.h>
  41 #include <errno.h>
  42 #include <fcntl.h>
  43 #include <libcontract.h>
  44 #include <libcontract_priv.h>
  45 #include <libscf_priv.h>
  46 #include <limits.h>
  47 #include <port.h>
  48 #include <signal.h>
  49 #include <stdarg.h>
  50 #include <stdio.h>
  51 #include <stdlib.h>
  52 #include <string.h>
  53 #include <unistd.h>
  54 #include <utmpx.h>
  55 
  56 #include "configd_exit.h"
  57 #include "protocol.h"
  58 #include "startd.h"
  59 
  60 static  struct  utmpx   *utmpp; /* pointer for getutxent() */
  61 
  62 pid_t
  63 startd_fork1(int *forkerr)
  64 {
  65         pid_t p;
  66 
  67         /*
  68          * prefork stack
  69          */
  70         wait_prefork();
  71 
  72         p = fork1();
  73 
  74         if (p == -1 && forkerr != NULL)
  75                 *forkerr = errno;
  76 
  77         /*
  78          * postfork stack
  79          */
  80         wait_postfork(p);
  81 
  82         return (p);
  83 }
  84 
  85 /*
  86  * void fork_mount(char *, char *)
  87  *   Run mount(1M) with the given options and mount point.  (mount(1M) has much
  88  *   hidden knowledge; it's much less correct to reimplement that logic here to
  89  *   save a fork(2)/exec(2) invocation.)
  90  */
  91 int
  92 fork_mount(char *path, char *opts)
  93 {
  94         pid_t pid;
  95         uint_t tries = 0;
  96         int status;
  97 
  98         for (pid = fork1(); pid == -1; pid = fork1()) {
  99                 if (++tries > MAX_MOUNT_RETRIES)
 100                         return (-1);
 101 
 102                 (void) sleep(tries);
 103         }
 104 
 105         if (pid != 0) {
 106                 (void) waitpid(pid, &status, 0);
 107 
 108                 /*
 109                  * If our mount(1M) invocation exited by peculiar means, or with
 110                  * a non-zero status, our mount likelihood is low.
 111                  */
 112                 if (!WIFEXITED(status) ||
 113                     WEXITSTATUS(status) != 0)
 114                         return (-1);
 115 
 116                 return (0);
 117         }
 118 
 119         (void) execl("/sbin/mount", "mount", "-o", opts, path, NULL);
 120 
 121         return (-1);
 122 }
 123 
 124 /*
 125  * pid_t fork_common(...)
 126  *   Common routine used by fork_sulogin and fork_configd to fork a
 127  *   process in a contract with the provided terms.  Invokes
 128  *   fork_sulogin (with its no-fork argument set) on errors.
 129  */
 130 static pid_t
 131 fork_common(const char *name, const char *svc_fmri, int retries, ctid_t *ctidp,
 132     uint_t inf, uint_t crit, uint_t fatal, uint_t param, uint64_t cookie)
 133 {
 134         uint_t tries = 0;
 135         int ctfd, err;
 136         pid_t pid;
 137 
 138         /*
 139          * Establish process contract terms.
 140          */
 141         if ((ctfd = open64(CTFS_ROOT "/process/template", O_RDWR)) == -1) {
 142                 fork_sulogin(B_TRUE, "Could not open process contract template "
 143                     "for %s: %s\n", name, strerror(errno));
 144                 /* NOTREACHED */
 145         }
 146 
 147         err = ct_tmpl_set_critical(ctfd, crit);
 148         err |= ct_pr_tmpl_set_fatal(ctfd, fatal);
 149         err |= ct_tmpl_set_informative(ctfd, inf);
 150         err |= ct_pr_tmpl_set_param(ctfd, param);
 151         err |= ct_tmpl_set_cookie(ctfd, cookie);
 152         err |= ct_pr_tmpl_set_svc_fmri(ctfd, svc_fmri);
 153         err |= ct_pr_tmpl_set_svc_aux(ctfd, name);
 154         if (err) {
 155                 (void) close(ctfd);
 156                 fork_sulogin(B_TRUE, "Could not set %s process contract "
 157                     "terms\n", name);
 158                 /* NOTREACHED */
 159         }
 160 
 161         if (err = ct_tmpl_activate(ctfd)) {
 162                 (void) close(ctfd);
 163                 fork_sulogin(B_TRUE, "Could not activate %s process contract "
 164                     "template: %s\n", name, strerror(err));
 165                 /* NOTREACHED */
 166         }
 167 
 168         /*
 169          * Attempt to fork "retries" times.
 170          */
 171         for (pid = fork1(); pid == -1; pid = fork1()) {
 172                 if (++tries > retries) {
 173                         /*
 174                          * When we exit the sulogin session, init(1M)
 175                          * will restart svc.startd(1M).
 176                          */
 177                         err = errno;
 178                         (void) ct_tmpl_clear(ctfd);
 179                         (void) close(ctfd);
 180                         fork_sulogin(B_TRUE, "Could not fork to start %s: %s\n",
 181                             name, strerror(err));
 182                         /* NOTREACHED */
 183                 }
 184                 (void) sleep(tries);
 185         }
 186 
 187         /*
 188          * Clean up, return pid and ctid.
 189          */
 190         if (pid != 0 && (errno = contract_latest(ctidp)) != 0)
 191                 uu_die("Could not get new contract id for %s\n", name);
 192         (void) ct_tmpl_clear(ctfd);
 193         (void) close(ctfd);
 194 
 195         return (pid);
 196 }
 197 
 198 /*
 199  * void fork_sulogin(boolean_t, const char *, ...)
 200  *   When we are invoked with the -s flag from boot (or run into an unfixable
 201  *   situation), we run a private copy of sulogin.  When the sulogin session
 202  *   is ended, we continue.  This is the last fallback action for system
 203  *   maintenance.
 204  *
 205  *   If immediate is true, fork_sulogin() executes sulogin(1M) directly, without
 206  *   forking.
 207  *
 208  *   Because fork_sulogin() is needed potentially before we daemonize, we leave
 209  *   it outside the wait_register() framework.
 210  */
 211 /*PRINTFLIKE2*/
 212 void
 213 fork_sulogin(boolean_t immediate, const char *format, ...)
 214 {
 215         va_list args;
 216         int fd_console;
 217 
 218         (void) printf("Requesting System Maintenance Mode\n");
 219 
 220         if (!booting_to_single_user)
 221                 (void) printf("(See /lib/svc/share/README for more "
 222                     "information.)\n");
 223 
 224         va_start(args, format);
 225         (void) vprintf(format, args);
 226         va_end(args);
 227 
 228         if (!immediate) {
 229                 ctid_t  ctid;
 230                 pid_t   pid;
 231 
 232                 pid = fork_common("sulogin", SVC_SULOGIN_FMRI,
 233                     MAX_SULOGIN_RETRIES, &ctid, CT_PR_EV_HWERR, 0,
 234                     CT_PR_EV_HWERR, CT_PR_PGRPONLY, SULOGIN_COOKIE);
 235 
 236                 if (pid != 0) {
 237                         (void) waitpid(pid, NULL, 0);
 238                         contract_abandon(ctid);
 239                         return;
 240                 }
 241                 /* close all inherited fds */
 242                 closefrom(0);
 243         } else {
 244                 (void) printf("Directly executing sulogin.\n");
 245                 /*
 246                  * Can't call closefrom() in this MT section
 247                  * so safely close a minimum set of fds.
 248                  */
 249                 (void) close(STDIN_FILENO);
 250                 (void) close(STDOUT_FILENO);
 251                 (void) close(STDERR_FILENO);
 252         }
 253 
 254         (void) setpgrp();
 255 
 256         /* open the console for sulogin */
 257         if ((fd_console = open("/dev/console", O_RDWR)) >= 0) {
 258                 if (fd_console != STDIN_FILENO)
 259                         while (dup2(fd_console, STDIN_FILENO) < 0 &&
 260                             errno == EINTR)
 261                                 ;
 262                 if (fd_console != STDOUT_FILENO)
 263                         while (dup2(fd_console, STDOUT_FILENO) < 0 &&
 264                             errno == EINTR)
 265                                 ;
 266                 if (fd_console != STDERR_FILENO)
 267                         while (dup2(fd_console, STDERR_FILENO) < 0 &&
 268                             errno == EINTR)
 269                                 ;
 270                 if (fd_console > STDERR_FILENO)
 271                         (void) close(fd_console);
 272         }
 273 
 274         setutxent();
 275         while ((utmpp = getutxent()) != NULL) {
 276                 if (strcmp(utmpp->ut_user, "LOGIN") != 0) {
 277                         if (strcmp(utmpp->ut_line, "console") == 0) {
 278                                 (void) kill(utmpp->ut_pid, 9);
 279                                 break;
 280                         }
 281                 }
 282         }
 283 
 284         (void) execl("/sbin/sulogin", "sulogin", NULL);
 285 
 286         uu_warn("Could not exec() sulogin");
 287 
 288         exit(1);
 289 }
 290 
 291 #define CONFIGD_PATH    "/lib/svc/bin/svc.configd"
 292 
 293 /*
 294  * void fork_configd(int status)
 295  *   We are interested in exit events (since the parent's exiting means configd
 296  *   is ready to run and since the child's exiting indicates an error case) and
 297  *   in empty events.  This means we have a unique template for initiating
 298  *   configd.
 299  */
 300 void
 301 fork_configd(int exitstatus)
 302 {
 303         pid_t pid;
 304         ctid_t ctid = -1;
 305         int err;
 306         char path[PATH_MAX];
 307 
 308         /*
 309          * Checking the existatus for the potential failure of the
 310          * daemonized svc.configd.  If this is not the first time
 311          * through, but a call from the svc.configd monitoring thread
 312          * after a failure this is the status that is expected.  Other
 313          * failures are exposed during initialization or are fixed
 314          * by a restart (e.g door closings).
 315          *
 316          * If this is on-disk database corruption it will also be
 317          * caught by a restart but could be cleared before the restart.
 318          *
 319          * Or this could be internal database corruption due to a
 320          * rogue service that needs to be cleared before restart.
 321          */
 322         if (WEXITSTATUS(exitstatus) == CONFIGD_EXIT_DATABASE_BAD) {
 323                 fork_sulogin(B_FALSE, "svc.configd exited with database "
 324                     "corrupt error after initialization of the repository\n");
 325         }
 326 
 327 retry:
 328         log_framework(LOG_DEBUG, "fork_configd trying to start svc.configd\n");
 329 
 330         /*
 331          * If we're retrying, we will have an old contract lying around
 332          * from the failure.  Since we're going to be creating a new
 333          * contract shortly, we abandon the old one now.
 334          */
 335         if (ctid != -1)
 336                 contract_abandon(ctid);
 337         ctid = -1;
 338 
 339         pid = fork_common("svc.configd", SCF_SERVICE_CONFIGD,
 340             MAX_CONFIGD_RETRIES, &ctid, 0, CT_PR_EV_EXIT, 0,
 341             CT_PR_INHERIT | CT_PR_REGENT, CONFIGD_COOKIE);
 342 
 343         if (pid != 0) {
 344                 int exitstatus;
 345 
 346                 st->st_configd_pid = pid;
 347 
 348                 if (waitpid(pid, &exitstatus, 0) == -1) {
 349                         fork_sulogin(B_FALSE, "waitpid on svc.configd "
 350                             "failed: %s\n", strerror(errno));
 351                 } else if (WIFEXITED(exitstatus)) {
 352                         char *errstr;
 353 
 354                         /*
 355                          * Examine exitstatus.  This will eventually get more
 356                          * complicated, as we will want to teach startd how to
 357                          * invoke configd with alternate repositories, etc.
 358                          *
 359                          * Note that exec(2) failure results in an exit status
 360                          * of 1, resulting in the default clause below.
 361                          */
 362 
 363                         /*
 364                          * Assign readable strings to cases we don't handle, or
 365                          * have error outcomes that cannot be eliminated.
 366                          */
 367                         switch (WEXITSTATUS(exitstatus)) {
 368                         case CONFIGD_EXIT_BAD_ARGS:
 369                                 errstr = "bad arguments";
 370                                 break;
 371 
 372                         case CONFIGD_EXIT_DATABASE_BAD:
 373                                 errstr = "database corrupt";
 374                                 break;
 375 
 376                         case CONFIGD_EXIT_DATABASE_LOCKED:
 377                                 errstr = "database locked";
 378                                 break;
 379                         case CONFIGD_EXIT_INIT_FAILED:
 380                                 errstr = "initialization failure";
 381                                 break;
 382                         case CONFIGD_EXIT_DOOR_INIT_FAILED:
 383                                 errstr = "door initialization failure";
 384                                 break;
 385                         case CONFIGD_EXIT_DATABASE_INIT_FAILED:
 386                                 errstr = "database initialization failure";
 387                                 break;
 388                         case CONFIGD_EXIT_NO_THREADS:
 389                                 errstr = "no threads available";
 390                                 break;
 391                         case CONFIGD_EXIT_LOST_MAIN_DOOR:
 392                                 errstr = "lost door server attachment";
 393                                 break;
 394                         case 1:
 395                                 errstr = "execution failure";
 396                                 break;
 397                         default:
 398                                 errstr = "unknown error";
 399                                 break;
 400                         }
 401 
 402                         /*
 403                          * Remedial actions for various configd failures.
 404                          */
 405                         switch (WEXITSTATUS(exitstatus)) {
 406                         case CONFIGD_EXIT_OKAY:
 407                                 break;
 408 
 409                         case CONFIGD_EXIT_DATABASE_LOCKED:
 410                                 /* attempt remount of / read-write */
 411                                 if (fs_is_read_only("/", NULL) == 1) {
 412                                         if (fs_remount("/") == -1)
 413                                                 fork_sulogin(B_FALSE,
 414                                                     "remount of root "
 415                                                     "filesystem failed\n");
 416 
 417                                         goto retry;
 418                                 }
 419                                 break;
 420 
 421                         default:
 422                                 fork_sulogin(B_FALSE, "svc.configd exited "
 423                                     "with status %d (%s)\n",
 424                                     WEXITSTATUS(exitstatus), errstr);
 425                                 goto retry;
 426                         }
 427                 } else if (WIFSIGNALED(exitstatus)) {
 428                         char signame[SIG2STR_MAX];
 429 
 430                         if (sig2str(WTERMSIG(exitstatus), signame))
 431                                 (void) snprintf(signame, SIG2STR_MAX,
 432                                     "signum %d", WTERMSIG(exitstatus));
 433 
 434                         fork_sulogin(B_FALSE, "svc.configd signalled:"
 435                             " %s\n", signame);
 436 
 437                         goto retry;
 438                 } else {
 439                         fork_sulogin(B_FALSE, "svc.configd non-exit "
 440                             "condition: 0x%x\n", exitstatus);
 441 
 442                         goto retry;
 443                 }
 444 
 445                 /*
 446                  * Announce that we have a valid svc.configd status.
 447                  */
 448                 MUTEX_LOCK(&st->st_configd_live_lock);
 449                 st->st_configd_lives = 1;
 450                 err = pthread_cond_broadcast(&st->st_configd_live_cv);
 451                 assert(err == 0);
 452                 MUTEX_UNLOCK(&st->st_configd_live_lock);
 453 
 454                 log_framework(LOG_DEBUG, "fork_configd broadcasts configd is "
 455                     "live\n");
 456                 return;
 457         }
 458 
 459         /*
 460          * Set our per-process core file path to leave core files in
 461          * /etc/svc/volatile directory, named after the PID to aid in debugging.
 462          */
 463         (void) snprintf(path, sizeof (path),
 464             "/etc/svc/volatile/core.configd.%%p");
 465 
 466         (void) core_set_process_path(path, strlen(path) + 1, getpid());
 467 
 468         log_framework(LOG_DEBUG, "executing svc.configd\n");
 469 
 470         (void) execl(CONFIGD_PATH, CONFIGD_PATH, NULL);
 471 
 472         /*
 473          * Status code is used above to identify configd exec failure.
 474          */
 475         exit(1);
 476 }
 477 
 478 void *
 479 fork_configd_thread(void *vctid)
 480 {
 481         int fd, err;
 482         ctid_t configd_ctid = (ctid_t)vctid;
 483 
 484         if (configd_ctid == -1) {
 485                 log_framework(LOG_DEBUG,
 486                     "fork_configd_thread starting svc.configd\n");
 487                 fork_configd(0);
 488         } else {
 489                 /*
 490                  * configd_ctid is known:  we broadcast and continue.
 491                  * test contract for appropriate state by verifying that
 492                  * there is one or more processes within it?
 493                  */
 494                 log_framework(LOG_DEBUG,
 495                     "fork_configd_thread accepting svc.configd with CTID %ld\n",
 496                     configd_ctid);
 497                 MUTEX_LOCK(&st->st_configd_live_lock);
 498                 st->st_configd_lives = 1;
 499                 (void) pthread_cond_broadcast(&st->st_configd_live_cv);
 500                 MUTEX_UNLOCK(&st->st_configd_live_lock);
 501         }
 502 
 503         fd = open64(CTFS_ROOT "/process/pbundle", O_RDONLY);
 504         if (fd == -1)
 505                 uu_die("process bundle open failed");
 506 
 507         /*
 508          * Make sure we get all events (including those generated by configd
 509          * before this thread was started).
 510          */
 511         err = ct_event_reset(fd);
 512         assert(err == 0);
 513 
 514         for (;;) {
 515                 int efd, sfd;
 516                 ct_evthdl_t ev;
 517                 uint32_t type;
 518                 ctevid_t evid;
 519                 ct_stathdl_t status;
 520                 ctid_t ctid;
 521                 uint64_t cookie;
 522                 pid_t pid;
 523 
 524                 if (err = ct_event_read_critical(fd, &ev)) {
 525                         assert(err != EINVAL && err != EAGAIN);
 526                         log_error(LOG_WARNING,
 527                             "Error reading next contract event: %s",
 528                             strerror(err));
 529                         continue;
 530                 }
 531 
 532                 evid = ct_event_get_evid(ev);
 533                 ctid = ct_event_get_ctid(ev);
 534                 type = ct_event_get_type(ev);
 535 
 536                 /* Fetch cookie. */
 537                 sfd = contract_open(ctid, "process", "status", O_RDONLY);
 538                 if (sfd < 0) {
 539                         ct_event_free(ev);
 540                         continue;
 541                 }
 542 
 543                 if (err = ct_status_read(sfd, CTD_COMMON, &status)) {
 544                         log_framework(LOG_WARNING, "Could not get status for "
 545                             "contract %ld: %s\n", ctid, strerror(err));
 546 
 547                         ct_event_free(ev);
 548                         startd_close(sfd);
 549                         continue;
 550                 }
 551 
 552                 cookie = ct_status_get_cookie(status);
 553 
 554                 ct_status_free(status);
 555 
 556                 startd_close(sfd);
 557 
 558                 /*
 559                  * Don't process events from contracts we aren't interested in.
 560                  */
 561                 if (cookie != CONFIGD_COOKIE) {
 562                         ct_event_free(ev);
 563                         continue;
 564                 }
 565 
 566                 if (type == CT_PR_EV_EXIT) {
 567                         int exitstatus;
 568 
 569                         (void) ct_pr_event_get_pid(ev, &pid);
 570                         (void) ct_pr_event_get_exitstatus(ev,
 571                             &exitstatus);
 572 
 573                         if (st->st_configd_pid != pid) {
 574                                 /*
 575                                  * This is the child exiting, so we
 576                                  * abandon the contract and restart
 577                                  * configd.
 578                                  */
 579                                 contract_abandon(ctid);
 580                                 fork_configd(exitstatus);
 581                         }
 582                 }
 583 
 584                 efd = contract_open(ctid, "process", "ctl", O_WRONLY);
 585                 if (efd != -1) {
 586                         (void) ct_ctl_ack(efd, evid);
 587                         startd_close(efd);
 588                 }
 589 
 590                 ct_event_free(ev);
 591 
 592         }
 593 
 594         /*NOTREACHED*/
 595         return (NULL);
 596 }
 597 
 598 void
 599 fork_rc_script(char rl, const char *arg, boolean_t wait)
 600 {
 601         pid_t pid;
 602         int tmpl, err, stat;
 603         char path[20] = "/sbin/rc.", log[20] = "rc..log", timebuf[20];
 604         time_t now;
 605         struct tm ltime;
 606         size_t sz;
 607         char *pathenv;
 608         char **nenv;
 609 
 610         path[8] = rl;
 611 
 612         tmpl = open64(CTFS_ROOT "/process/template", O_RDWR);
 613         if (tmpl >= 0) {
 614                 err = ct_tmpl_set_critical(tmpl, 0);
 615                 assert(err == 0);
 616 
 617                 err = ct_tmpl_set_informative(tmpl, 0);
 618                 assert(err == 0);
 619 
 620                 err = ct_pr_tmpl_set_fatal(tmpl, 0);
 621                 assert(err == 0);
 622 
 623                 err = ct_tmpl_activate(tmpl);
 624                 assert(err == 0);
 625 
 626                 err = close(tmpl);
 627                 assert(err == 0);
 628         } else {
 629                 uu_warn("Could not create contract template for %s.\n", path);
 630         }
 631 
 632         pid = startd_fork1(NULL);
 633         if (pid < 0) {
 634                 return;
 635         } else if (pid != 0) {
 636                 /* parent */
 637                 if (wait) {
 638                         do
 639                                 err = waitpid(pid, &stat, 0);
 640                         while (err != 0 && errno == EINTR)
 641                                 ;
 642 
 643                         if (!WIFEXITED(stat)) {
 644                                 log_framework(LOG_INFO,
 645                                     "%s terminated with waitpid() status %d.\n",
 646                                     path, stat);
 647                         } else if (WEXITSTATUS(stat) != 0) {
 648                                 log_framework(LOG_INFO,
 649                                     "%s failed with status %d.\n", path,
 650                                     WEXITSTATUS(stat));
 651                         }
 652                 }
 653 
 654                 return;
 655         }
 656 
 657         /* child */
 658 
 659         log[2] = rl;
 660 
 661         setlog(log);
 662 
 663         now = time(NULL);
 664         sz = strftime(timebuf, sizeof (timebuf), "%b %e %T",
 665             localtime_r(&now, &ltime));
 666         assert(sz != 0);
 667 
 668         (void) fprintf(stderr, "%s Executing %s %s\n", timebuf, path, arg);
 669 
 670         if (rl == 'S')
 671                 pathenv = "PATH=/sbin:/usr/sbin:/usr/bin";
 672         else
 673                 pathenv = "PATH=/usr/sbin:/usr/bin";
 674 
 675         nenv = set_smf_env(NULL, 0, pathenv, NULL, NULL);
 676 
 677         (void) execle(path, path, arg, 0, nenv);
 678 
 679         perror("exec");
 680         exit(0);
 681 }