sys/kern/init_main.c

   1 /*-
   2  * Copyright (c) 1995 Terrence R. Lambert
   3  * All rights reserved.
   4  *
   5  * Copyright (c) 1982, 1986, 1989, 1991, 1992, 1993
   6  *      The Regents of the University of California.  All rights reserved.
   7  * (c) UNIX System Laboratories, Inc.
   8  * All or some portions of this file are derived from material licensed
   9  * to the University of California by American Telephone and Telegraph
  10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
  11  * the permission of UNIX System Laboratories, Inc.
  12  *
  13  * Redistribution and use in source and binary forms, with or without
  14  * modification, are permitted provided that the following conditions
  15  * are met:
  16  * 1. Redistributions of source code must retain the above copyright
  17  *    notice, this list of conditions and the following disclaimer.
  18  * 2. Redistributions in binary form must reproduce the above copyright
  19  *    notice, this list of conditions and the following disclaimer in the
  20  *    documentation and/or other materials provided with the distribution.
  21  * 3. All advertising materials mentioning features or use of this software
  22  *    must display the following acknowledgement:
  23  *      This product includes software developed by the University of
  24  *      California, Berkeley and its contributors.
  25  * 4. Neither the name of the University nor the names of its contributors
  26  *    may be used to endorse or promote products derived from this software
  27  *    without specific prior written permission.
  28  *
  29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  39  * SUCH DAMAGE.
  40  *
  41  *      @(#)init_main.c 8.9 (Berkeley) 1/21/94
  42  */
  43
  44 #include <sys/cdefs.h>
  45 __FBSDID("$FreeBSD$");
  46
  47 #include "opt_ddb.h"
  48 #include "opt_init_path.h"
  49 #include "opt_mac.h"
  50
  51 #include <sys/param.h>
  52 #include <sys/kernel.h>
  53 #include <sys/exec.h>
  54 #include <sys/file.h>
  55 #include <sys/filedesc.h>
  56 #include <sys/ktr.h>
  57 #include <sys/lock.h>
  58 #include <sys/mount.h>
  59 #include <sys/mutex.h>
  60 #include <sys/syscallsubr.h>
  61 #include <sys/sysctl.h>
  62 #include <sys/proc.h>
  63 #include <sys/resourcevar.h>
  64 #include <sys/systm.h>
  65 #include <sys/signalvar.h>
  66 #include <sys/vnode.h>
  67 #include <sys/sysent.h>
  68 #include <sys/reboot.h>
  69 #include <sys/sched.h>
  70 #include <sys/sx.h>
  71 #include <sys/sysproto.h>
  72 #include <sys/vmmeter.h>
  73 #include <sys/unistd.h>
  74 #include <sys/malloc.h>
  75 #include <sys/conf.h>
  76 #include <sys/cpuset.h>
  77
  78 #include <machine/cpu.h>
  79
  80 #include <security/audit/audit.h>
  81 #include <security/mac/mac_framework.h>
  82
  83 #include <vm/vm.h>
  84 #include <vm/vm_param.h>
  85 #include <vm/pmap.h>
  86 #include <vm/vm_map.h>
  87 #include <sys/copyright.h>
  88
  89 #include <ddb/ddb.h>
  90 #include <ddb/db_sym.h>
  91
  92 void mi_startup(void);                          /* Should be elsewhere */
  93
  94 /* Components of the first process -- never freed. */
  95 static struct session session0;
  96 static struct pgrp pgrp0;
  97 struct  proc proc0;
  98 struct  thread thread0 __aligned(16);
  99 struct  vmspace vmspace0;
 100 struct  proc *initproc;
 101
 102 int     boothowto = 0;          /* initialized so that it can be patched */
 103 SYSCTL_INT(_debug, OID_AUTO, boothowto, CTLFLAG_RD, &boothowto, 0, "");
 104 int     bootverbose;
 105 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, &bootverbose, 0, "");
 106
 107 /*
 108  * This ensures that there is at least one entry so that the sysinit_set
 109  * symbol is not undefined.  A sybsystem ID of SI_SUB_DUMMY is never
 110  * executed.
 111  */
 112 SYSINIT(placeholder, SI_SUB_DUMMY, SI_ORDER_ANY, NULL, NULL);
 113
 114 /*
 115  * The sysinit table itself.  Items are checked off as the are run.
 116  * If we want to register new sysinit types, add them to newsysinit.
 117  */
 118 SET_DECLARE(sysinit_set, struct sysinit);
 119 struct sysinit **sysinit, **sysinit_end;
 120 struct sysinit **newsysinit, **newsysinit_end;
 121
 122 /*
 123  * Merge a new sysinit set into the current set, reallocating it if
 124  * necessary.  This can only be called after malloc is running.
 125  */
 126 void
 127 sysinit_add(struct sysinit **set, struct sysinit **set_end)
 128 {
 129         struct sysinit **newset;
 130         struct sysinit **sipp;
 131         struct sysinit **xipp;
 132         int count;
 133
 134         count = set_end - set;
 135         if (newsysinit)
 136                 count += newsysinit_end - newsysinit;
 137         else
 138                 count += sysinit_end - sysinit;
 139         newset = malloc(count * sizeof(*sipp), M_TEMP, M_NOWAIT);
 140         if (newset == NULL)
 141                 panic("cannot malloc for sysinit");
 142         xipp = newset;
 143         if (newsysinit)
 144                 for (sipp = newsysinit; sipp < newsysinit_end; sipp++)
 145                         *xipp++ = *sipp;
 146         else
 147                 for (sipp = sysinit; sipp < sysinit_end; sipp++)
 148                         *xipp++ = *sipp;
 149         for (sipp = set; sipp < set_end; sipp++)
 150                 *xipp++ = *sipp;
 151         if (newsysinit)
 152                 free(newsysinit, M_TEMP);
 153         newsysinit = newset;
 154         newsysinit_end = newset + count;
 155 }
 156
 157 /*
 158  * System startup; initialize the world, create process 0, mount root
 159  * filesystem, and fork to create init and pagedaemon.  Most of the
 160  * hard work is done in the lower-level initialization routines including
 161  * startup(), which does memory initialization and autoconfiguration.
 162  *
 163  * This allows simple addition of new kernel subsystems that require
 164  * boot time initialization.  It also allows substitution of subsystem
 165  * (for instance, a scheduler, kernel profiler, or VM system) by object
 166  * module.  Finally, it allows for optional "kernel threads".
 167  */
 168 void
 169 mi_startup(void)
 170 {
 171
 172         register struct sysinit **sipp;         /* system initialization*/
 173         register struct sysinit **xipp;         /* interior loop of sort*/
 174         register struct sysinit *save;          /* bubble*/
 175
 176 #if defined(VERBOSE_SYSINIT)
 177         int last;
 178         int verbose;
 179 #endif
 180
 181         if (sysinit == NULL) {
 182                 sysinit = SET_BEGIN(sysinit_set);
 183                 sysinit_end = SET_LIMIT(sysinit_set);
 184         }
 185
 186 restart:
 187         /*
 188          * Perform a bubble sort of the system initialization objects by
 189          * their subsystem (primary key) and order (secondary key).
 190          */
 191         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 192                 for (xipp = sipp + 1; xipp < sysinit_end; xipp++) {
 193                         if ((*sipp)->subsystem < (*xipp)->subsystem ||
 194                              ((*sipp)->subsystem == (*xipp)->subsystem &&
 195                               (*sipp)->order <= (*xipp)->order))
 196                                 continue;       /* skip*/
 197                         save = *sipp;
 198                         *sipp = *xipp;
 199                         *xipp = save;
 200                 }
 201         }
 202
 203 #if defined(VERBOSE_SYSINIT)
 204         last = SI_SUB_COPYRIGHT;
 205         verbose = 0;
 206 #if !defined(DDB)
 207         printf("VERBOSE_SYSINIT: DDB not enabled, symbol lookups disabled.\n");
 208 #endif
 209 #endif
 210
 211         /*
 212          * Traverse the (now) ordered list of system initialization tasks.
 213          * Perform each task, and continue on to the next task.
 214          *
 215          * The last item on the list is expected to be the scheduler,
 216          * which will not return.
 217          */
 218         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 219
 220                 if ((*sipp)->subsystem == SI_SUB_DUMMY)
 221                         continue;       /* skip dummy task(s)*/
 222
 223                 if ((*sipp)->subsystem == SI_SUB_DONE)
 224                         continue;
 225
 226 #if defined(VERBOSE_SYSINIT)
 227                 if ((*sipp)->subsystem > last) {
 228                         verbose = 1;
 229                         last = (*sipp)->subsystem;
 230                         printf("subsystem %x\n", last);
 231                 }
 232                 if (verbose) {
 233 #if defined(DDB)
 234                         const char *name;
 235                         c_db_sym_t sym;
 236                         db_expr_t  offset;
 237
 238                         sym = db_search_symbol((vm_offset_t)(*sipp)->func,
 239                             DB_STGY_PROC, &offset);
 240                         db_symbol_values(sym, &name, NULL);
 241                         if (name != NULL)
 242                                 printf("   %s(%p)... ", name, (*sipp)->udata);
 243                         else
 244 #endif
 245                                 printf("   %p(%p)... ", (*sipp)->func,
 246                                     (*sipp)->udata);
 247                 }
 248 #endif
 249
 250                 /* Call function */
 251                 (*((*sipp)->func))((*sipp)->udata);
 252
 253 #if defined(VERBOSE_SYSINIT)
 254                 if (verbose)
 255                         printf("done.\n");
 256 #endif
 257
 258                 /* Check off the one we're just done */
 259                 (*sipp)->subsystem = SI_SUB_DONE;
 260
 261                 /* Check if we've installed more sysinit items via KLD */
 262                 if (newsysinit != NULL) {
 263                         if (sysinit != SET_BEGIN(sysinit_set))
 264                                 free(sysinit, M_TEMP);
 265                         sysinit = newsysinit;
 266                         sysinit_end = newsysinit_end;
 267                         newsysinit = NULL;
 268                         newsysinit_end = NULL;
 269                         goto restart;
 270                 }
 271         }
 272
 273         panic("Shouldn't get here!");
 274         /* NOTREACHED*/
 275 }
 276
 277
 278 /*
 279  ***************************************************************************
 280  ****
 281  **** The following SYSINIT's belong elsewhere, but have not yet
 282  **** been moved.
 283  ****
 284  ***************************************************************************
 285  */
 286 static void
 287 print_caddr_t(void *data __unused)
 288 {
 289         printf("%s", (char *)data);
 290 }
 291 SYSINIT(announce, SI_SUB_COPYRIGHT, SI_ORDER_FIRST, print_caddr_t,
 292     copyright);
 293 SYSINIT(trademark, SI_SUB_COPYRIGHT, SI_ORDER_SECOND, print_caddr_t,
 294     trademark);
 295 SYSINIT(version, SI_SUB_COPYRIGHT, SI_ORDER_THIRD, print_caddr_t, version);
 296
 297 #ifdef WITNESS
 298 static char wit_warn[] =
 299      "WARNING: WITNESS option enabled, expect reduced performance.\n";
 300 SYSINIT(witwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 1,
 301    print_caddr_t, wit_warn);
 302 SYSINIT(witwarn2, SI_SUB_RUN_SCHEDULER, SI_ORDER_THIRD + 1,
 303    print_caddr_t, wit_warn);
 304 #endif
 305
 306 #ifdef DIAGNOSTIC
 307 static char diag_warn[] =
 308      "WARNING: DIAGNOSTIC option enabled, expect reduced performance.\n";
 309 SYSINIT(diagwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 2,
 310     print_caddr_t, diag_warn);
 311 SYSINIT(diagwarn2, SI_SUB_RUN_SCHEDULER, SI_ORDER_THIRD + 2,
 312     print_caddr_t, diag_warn);
 313 #endif
 314
 315 static void
 316 set_boot_verbose(void *data __unused)
 317 {
 318
 319         if (boothowto & RB_VERBOSE)
 320                 bootverbose++;
 321 }
 322 SYSINIT(boot_verbose, SI_SUB_TUNABLES, SI_ORDER_ANY, set_boot_verbose, NULL);
 323
 324 struct sysentvec null_sysvec = {
 325         .sv_size        = 0,
 326         .sv_table       = NULL,
 327         .sv_mask        = 0,
 328         .sv_sigsize     = 0,
 329         .sv_sigtbl      = NULL,
 330         .sv_errsize     = 0,
 331         .sv_errtbl      = NULL,
 332         .sv_transtrap   = NULL,
 333         .sv_fixup       = NULL,
 334         .sv_sendsig     = NULL,
 335         .sv_sigcode     = NULL,
 336         .sv_szsigcode   = NULL,
 337         .sv_prepsyscall = NULL,
 338         .sv_name        = "null",
 339         .sv_coredump    = NULL,
 340         .sv_imgact_try  = NULL,
 341         .sv_minsigstksz = 0,
 342         .sv_pagesize    = PAGE_SIZE,
 343         .sv_minuser     = VM_MIN_ADDRESS,
 344         .sv_maxuser     = VM_MAXUSER_ADDRESS,
 345         .sv_usrstack    = USRSTACK,
 346         .sv_psstrings   = PS_STRINGS,
 347         .sv_stackprot   = VM_PROT_ALL,
 348         .sv_copyout_strings     = NULL,
 349         .sv_setregs     = NULL,
 350         .sv_fixlimit    = NULL,
 351         .sv_maxssiz     = NULL
 352 };
 353
 354 /*
 355  ***************************************************************************
 356  ****
 357  **** The two following SYSINIT's are proc0 specific glue code.  I am not
 358  **** convinced that they can not be safely combined, but their order of
 359  **** operation has been maintained as the same as the original init_main.c
 360  **** for right now.
 361  ****
 362  **** These probably belong in init_proc.c or kern_proc.c, since they
 363  **** deal with proc0 (the fork template process).
 364  ****
 365  ***************************************************************************
 366  */
 367 /* ARGSUSED*/
 368 static void
 369 proc0_init(void *dummy __unused)
 370 {
 371         struct proc *p;
 372         unsigned i;
 373         struct thread *td;
 374
 375         GIANT_REQUIRED;
 376         p = &proc0;
 377         td = &thread0;
 378
 379         /*
 380          * Initialize magic number and osrel.
 381          */
 382         p->p_magic = P_MAGIC;
 383         p->p_osrel = osreldate;
 384
 385         /*
 386          * Initialize thread and process structures.
 387          */
 388         procinit();     /* set up proc zone */
 389         threadinit();   /* set up UMA zones */
 390
 391         /*
 392          * Initialise scheduler resources.
 393          * Add scheduler specific parts to proc, thread as needed.
 394          */
 395         schedinit();    /* scheduler gets its house in order */
 396         /*
 397          * Initialize sleep queue hash table
 398          */
 399         sleepinit();
 400
 401         /*
 402          * additional VM structures
 403          */
 404         vm_init2();
 405
 406         /*
 407          * Create process 0 (the swapper).
 408          */
 409         LIST_INSERT_HEAD(&allproc, p, p_list);
 410         LIST_INSERT_HEAD(PIDHASH(0), p, p_hash);
 411         mtx_init(&pgrp0.pg_mtx, "process group", NULL, MTX_DEF | MTX_DUPOK);
 412         p->p_pgrp = &pgrp0;
 413         LIST_INSERT_HEAD(PGRPHASH(0), &pgrp0, pg_hash);
 414         LIST_INIT(&pgrp0.pg_members);
 415         LIST_INSERT_HEAD(&pgrp0.pg_members, p, p_pglist);
 416
 417         pgrp0.pg_session = &session0;
 418         mtx_init(&session0.s_mtx, "session", NULL, MTX_DEF);
 419         session0.s_count = 1;
 420         session0.s_leader = p;
 421
 422         p->p_sysent = &null_sysvec;
 423         p->p_flag = P_SYSTEM | P_INMEM;
 424         p->p_state = PRS_NORMAL;
 425         knlist_init(&p->p_klist, &p->p_mtx, NULL, NULL, NULL);
 426         STAILQ_INIT(&p->p_ktr);
 427         p->p_nice = NZERO;
 428         td->td_state = TDS_RUNNING;
 429         td->td_pri_class = PRI_TIMESHARE;
 430         td->td_user_pri = PUSER;
 431         td->td_base_user_pri = PUSER;
 432         td->td_priority = PVM;
 433         td->td_base_pri = PUSER;
 434         td->td_oncpu = 0;
 435         td->td_flags = TDF_INMEM;
 436         td->td_cpuset = cpuset_thread0();
 437         p->p_peers = 0;
 438         p->p_leader = p;
 439
 440
 441         bcopy("swapper", p->p_comm, sizeof ("swapper"));
 442
 443         callout_init(&p->p_itcallout, CALLOUT_MPSAFE);
 444         callout_init_mtx(&p->p_limco, &p->p_mtx, 0);
 445         callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
 446
 447         /* Create credentials. */
 448         p->p_ucred = crget();
 449         p->p_ucred->cr_ngroups = 1;     /* group 0 */
 450         p->p_ucred->cr_uidinfo = uifind(0);
 451         p->p_ucred->cr_ruidinfo = uifind(0);
 452         p->p_ucred->cr_prison = NULL;   /* Don't jail it. */
 453 #ifdef AUDIT
 454         audit_cred_kproc0(p->p_ucred);
 455 #endif
 456 #ifdef MAC
 457         mac_create_proc0(p->p_ucred);
 458 #endif
 459         td->td_ucred = crhold(p->p_ucred);
 460
 461         /* Create sigacts. */
 462         p->p_sigacts = sigacts_alloc();
 463
 464         /* Initialize signal state for process 0. */
 465         siginit(&proc0);
 466
 467         /* Create the file descriptor table. */
 468         p->p_fd = fdinit(NULL);
 469         p->p_fdtol = NULL;
 470
 471         /* Create the limits structures. */
 472         p->p_limit = lim_alloc();
 473         for (i = 0; i < RLIM_NLIMITS; i++)
 474                 p->p_limit->pl_rlimit[i].rlim_cur =
 475                     p->p_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
 476         p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur =
 477             p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
 478         p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_cur =
 479             p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
 480         i = ptoa(cnt.v_free_count);
 481         p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_max = i;
 482         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_max = i;
 483         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = i / 3;
 484         p->p_cpulimit = RLIM_INFINITY;
 485
 486         p->p_stats = pstats_alloc();
 487
 488         /* Allocate a prototype map so we have something to fork. */
 489         pmap_pinit0(vmspace_pmap(&vmspace0));
 490         p->p_vmspace = &vmspace0;
 491         vmspace0.vm_refcnt = 1;
 492         vm_map_init(&vmspace0.vm_map, p->p_sysent->sv_minuser,
 493             p->p_sysent->sv_maxuser);
 494         vmspace0.vm_map.pmap = vmspace_pmap(&vmspace0);
 495         /*-
 496          * call the init and ctor for the new thread and proc
 497          * we wait to do this until all other structures
 498          * are fairly sane.
 499          */
 500         EVENTHANDLER_INVOKE(process_init, p);
 501         EVENTHANDLER_INVOKE(thread_init, td);
 502         EVENTHANDLER_INVOKE(process_ctor, p);
 503         EVENTHANDLER_INVOKE(thread_ctor, td);
 504
 505         /*
 506          * Charge root for one process.
 507          */
 508         (void)chgproccnt(p->p_ucred->cr_ruidinfo, 1, 0);
 509 }
 510 SYSINIT(p0init, SI_SUB_INTRINSIC, SI_ORDER_FIRST, proc0_init, NULL);
 511
 512 /* ARGSUSED*/
 513 static void
 514 proc0_post(void *dummy __unused)
 515 {
 516         struct timespec ts;
 517         struct proc *p;
 518         struct rusage ru;
 519         struct thread *td;
 520
 521         /*
 522          * Now we can look at the time, having had a chance to verify the
 523          * time from the filesystem.  Pretend that proc0 started now.
 524          */
 525         sx_slock(&allproc_lock);
 526         FOREACH_PROC_IN_SYSTEM(p) {
 527                 microuptime(&p->p_stats->p_start);
 528                 PROC_SLOCK(p);
 529                 rufetch(p, &ru);        /* Clears thread stats */
 530                 PROC_SUNLOCK(p);
 531                 p->p_rux.rux_runtime = 0;
 532                 p->p_rux.rux_uticks = 0;
 533                 p->p_rux.rux_sticks = 0;
 534                 p->p_rux.rux_iticks = 0;
 535                 FOREACH_THREAD_IN_PROC(p, td) {
 536                         td->td_runtime = 0;
 537                 }
 538         }
 539         sx_sunlock(&allproc_lock);
 540         PCPU_SET(switchtime, cpu_ticks());
 541         PCPU_SET(switchticks, ticks);
 542
 543         /*
 544          * Give the ``random'' number generator a thump.
 545          */
 546         nanotime(&ts);
 547         srandom(ts.tv_sec ^ ts.tv_nsec);
 548 }
 549 SYSINIT(p0post, SI_SUB_INTRINSIC_POST, SI_ORDER_FIRST, proc0_post, NULL);
 550
 551 /*
 552  ***************************************************************************
 553  ****
 554  **** The following SYSINIT's and glue code should be moved to the
 555  **** respective files on a per subsystem basis.
 556  ****
 557  ***************************************************************************
 558  */
 559
 560
 561 /*
 562  ***************************************************************************
 563  ****
 564  **** The following code probably belongs in another file, like
 565  **** kern/init_init.c.
 566  ****
 567  ***************************************************************************
 568  */
 569
 570 /*
 571  * List of paths to try when searching for "init".
 572  */
 573 static char init_path[MAXPATHLEN] =
 574 #ifdef  INIT_PATH
 575     __XSTRING(INIT_PATH);
 576 #else
 577     "/sbin/init:/sbin/oinit:/sbin/init.bak:/rescue/init:/stand/sysinstall";
 578 #endif
 579 SYSCTL_STRING(_kern, OID_AUTO, init_path, CTLFLAG_RD, init_path, 0,
 580         "Path used to search the init process");
 581
 582 /*
 583  * Shutdown timeout of init(8).
 584  * Unused within kernel, but used to control init(8), hence do not remove.
 585  */
 586 #ifndef INIT_SHUTDOWN_TIMEOUT
 587 #define INIT_SHUTDOWN_TIMEOUT 120
 588 #endif
 589 static int init_shutdown_timeout = INIT_SHUTDOWN_TIMEOUT;
 590 SYSCTL_INT(_kern, OID_AUTO, init_shutdown_timeout,
 591         CTLFLAG_RW, &init_shutdown_timeout, 0, "");
 592
 593 /*
 594  * Start the initial user process; try exec'ing each pathname in init_path.
 595  * The program is invoked with one argument containing the boot flags.
 596  */
 597 static void
 598 start_init(void *dummy)
 599 {
 600         vm_offset_t addr;
 601         struct execve_args args;
 602         int options, error;
 603         char *var, *path, *next, *s;
 604         char *ucp, **uap, *arg0, *arg1;
 605         struct thread *td;
 606         struct proc *p;
 607
 608         mtx_lock(&Giant);
 609
 610         GIANT_REQUIRED;
 611
 612         td = curthread;
 613         p = td->td_proc;
 614
 615         vfs_mountroot();
 616
 617         /*
 618          * Need just enough stack to hold the faked-up "execve()" arguments.
 619          */
 620         addr = p->p_sysent->sv_usrstack - PAGE_SIZE;
 621         if (vm_map_find(&p->p_vmspace->vm_map, NULL, 0, &addr, PAGE_SIZE,
 622                         FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != 0)
 623                 panic("init: couldn't allocate argument space");
 624         p->p_vmspace->vm_maxsaddr = (caddr_t)addr;
 625         p->p_vmspace->vm_ssize = 1;
 626
 627         if ((var = getenv("init_path")) != NULL) {
 628                 strlcpy(init_path, var, sizeof(init_path));
 629                 freeenv(var);
 630         }
 631
 632         for (path = init_path; *path != '\0'; path = next) {
 633                 while (*path == ':')
 634                         path++;
 635                 if (*path == '\0')
 636                         break;
 637                 for (next = path; *next != '\0' && *next != ':'; next++)
 638                         /* nothing */ ;
 639                 if (bootverbose)
 640                         printf("start_init: trying %.*s\n", (int)(next - path),
 641                             path);
 642
 643                 /*
 644                  * Move out the boot flag argument.
 645                  */
 646                 options = 0;
 647                 ucp = (char *)p->p_sysent->sv_usrstack;
 648                 (void)subyte(--ucp, 0);         /* trailing zero */
 649                 if (boothowto & RB_SINGLE) {
 650                         (void)subyte(--ucp, 's');
 651                         options = 1;
 652                 }
 653 #ifdef notyet
 654                 if (boothowto & RB_FASTBOOT) {
 655                         (void)subyte(--ucp, 'f');
 656                         options = 1;
 657                 }
 658 #endif
 659
 660 #ifdef BOOTCDROM
 661                 (void)subyte(--ucp, 'C');
 662                 options = 1;
 663 #endif
 664
 665                 if (options == 0)
 666                         (void)subyte(--ucp, '-');
 667                 (void)subyte(--ucp, '-');               /* leading hyphen */
 668                 arg1 = ucp;
 669
 670                 /*
 671                  * Move out the file name (also arg 0).
 672                  */
 673                 (void)subyte(--ucp, 0);
 674                 for (s = next - 1; s >= path; s--)
 675                         (void)subyte(--ucp, *s);
 676                 arg0 = ucp;
 677
 678                 /*
 679                  * Move out the arg pointers.
 680                  */
 681                 uap = (char **)((intptr_t)ucp & ~(sizeof(intptr_t)-1));
 682                 (void)suword((caddr_t)--uap, (long)0);  /* terminator */
 683                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg1);
 684                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg0);
 685
 686                 /*
 687                  * Point at the arguments.
 688                  */
 689                 args.fname = arg0;
 690                 args.argv = uap;
 691                 args.envv = NULL;
 692
 693                 /*
 694                  * Now try to exec the program.  If can't for any reason
 695                  * other than it doesn't exist, complain.
 696                  *
 697                  * Otherwise, return via fork_trampoline() all the way
 698                  * to user mode as init!
 699                  */
 700                 if ((error = execve(td, &args)) == 0) {
 701                         mtx_unlock(&Giant);
 702                         return;
 703                 }
 704                 if (error != ENOENT)
 705                         printf("exec %.*s: error %d\n", (int)(next - path),
 706                             path, error);
 707         }
 708         printf("init: not found in path %s\n", init_path);
 709         panic("no init");
 710 }
 711
 712 /*
 713  * Like kthread_create(), but runs in it's own address space.
 714  * We do this early to reserve pid 1.
 715  *
 716  * Note special case - do not make it runnable yet.  Other work
 717  * in progress will change this more.
 718  */
 719 static void
 720 create_init(const void *udata __unused)
 721 {
 722         struct ucred *newcred, *oldcred;
 723         int error;
 724
 725         error = fork1(&thread0, RFFDG | RFPROC | RFSTOPPED, 0, &initproc);
 726         if (error)
 727                 panic("cannot fork init: %d\n", error);
 728         KASSERT(initproc->p_pid == 1, ("create_init: initproc->p_pid != 1"));
 729         /* divorce init's credentials from the kernel's */
 730         newcred = crget();
 731         PROC_LOCK(initproc);
 732         initproc->p_flag |= P_SYSTEM | P_INMEM;
 733         oldcred = initproc->p_ucred;
 734         crcopy(newcred, oldcred);
 735 #ifdef MAC
 736         mac_create_proc1(newcred);
 737 #endif
 738 #ifdef AUDIT
 739         audit_cred_proc1(newcred);
 740 #endif
 741         initproc->p_ucred = newcred;
 742         PROC_UNLOCK(initproc);
 743         crfree(oldcred);
 744         cred_update_thread(FIRST_THREAD_IN_PROC(initproc));
 745         cpu_set_fork_handler(FIRST_THREAD_IN_PROC(initproc), start_init, NULL);
 746 }
 747 SYSINIT(init, SI_SUB_CREATE_INIT, SI_ORDER_FIRST, create_init, NULL);
 748
 749 /*
 750  * Make it runnable now.
 751  */
 752 static void
 753 kick_init(const void *udata __unused)
 754 {
 755         struct thread *td;
 756
 757         td = FIRST_THREAD_IN_PROC(initproc);
 758         thread_lock(td);
 759         TD_SET_CAN_RUN(td);
 760         sched_add(td, SRQ_BORING);
 761         thread_unlock(td);
 762 }
 763 SYSINIT(kickinit, SI_SUB_KTHREAD_INIT, SI_ORDER_FIRST, kick_init, NULL);