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
  50 #include <sys/param.h>
  51 #include <sys/kernel.h>
  52 #include <sys/exec.h>
  53 #include <sys/file.h>
  54 #include <sys/filedesc.h>
  55 #include <sys/jail.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 (boothowto & RB_VERBOSE)
 182                 bootverbose++;
 183
 184         if (sysinit == NULL) {
 185                 sysinit = SET_BEGIN(sysinit_set);
 186                 sysinit_end = SET_LIMIT(sysinit_set);
 187         }
 188
 189 restart:
 190         /*
 191          * Perform a bubble sort of the system initialization objects by
 192          * their subsystem (primary key) and order (secondary key).
 193          */
 194         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 195                 for (xipp = sipp + 1; xipp < sysinit_end; xipp++) {
 196                         if ((*sipp)->subsystem < (*xipp)->subsystem ||
 197                              ((*sipp)->subsystem == (*xipp)->subsystem &&
 198                               (*sipp)->order <= (*xipp)->order))
 199                                 continue;       /* skip*/
 200                         save = *sipp;
 201                         *sipp = *xipp;
 202                         *xipp = save;
 203                 }
 204         }
 205
 206 #if defined(VERBOSE_SYSINIT)
 207         last = SI_SUB_COPYRIGHT;
 208         verbose = 0;
 209 #if !defined(DDB)
 210         printf("VERBOSE_SYSINIT: DDB not enabled, symbol lookups disabled.\n");
 211 #endif
 212 #endif
 213
 214         /*
 215          * Traverse the (now) ordered list of system initialization tasks.
 216          * Perform each task, and continue on to the next task.
 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         mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED);
 274         mtx_unlock(&Giant);
 275
 276         /*
 277          * Now hand over this thread to swapper.
 278          */
 279         swapper();
 280         /* NOTREACHED*/
 281 }
 282
 283
 284 /*
 285  ***************************************************************************
 286  ****
 287  **** The following SYSINIT's belong elsewhere, but have not yet
 288  **** been moved.
 289  ****
 290  ***************************************************************************
 291  */
 292 static void
 293 print_caddr_t(void *data)
 294 {
 295         printf("%s", (char *)data);
 296 }
 297
 298 static void
 299 print_version(void *data __unused)
 300 {
 301         int len;
 302
 303         /* Strip a trailing newline from version. */
 304         len = strlen(version);
 305         while (len > 0 && version[len - 1] == '\n')
 306                 len--;
 307         printf("%.*s %s\n", len, version, machine);
 308         printf("%s\n", compiler_version);
 309 }
 310
 311 SYSINIT(announce, SI_SUB_COPYRIGHT, SI_ORDER_FIRST, print_caddr_t,
 312     copyright);
 313 SYSINIT(trademark, SI_SUB_COPYRIGHT, SI_ORDER_SECOND, print_caddr_t,
 314     trademark);
 315 SYSINIT(version, SI_SUB_COPYRIGHT, SI_ORDER_THIRD, print_version, NULL);
 316
 317 #ifdef WITNESS
 318 static char wit_warn[] =
 319      "WARNING: WITNESS option enabled, expect reduced performance.\n";
 320 SYSINIT(witwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 1,
 321    print_caddr_t, wit_warn);
 322 SYSINIT(witwarn2, SI_SUB_LAST, SI_ORDER_THIRD + 1,
 323    print_caddr_t, wit_warn);
 324 #endif
 325
 326 #ifdef DIAGNOSTIC
 327 static char diag_warn[] =
 328      "WARNING: DIAGNOSTIC option enabled, expect reduced performance.\n";
 329 SYSINIT(diagwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 2,
 330     print_caddr_t, diag_warn);
 331 SYSINIT(diagwarn2, SI_SUB_LAST, SI_ORDER_THIRD + 2,
 332     print_caddr_t, diag_warn);
 333 #endif
 334
 335 static int
 336 null_fetch_syscall_args(struct thread *td __unused,
 337     struct syscall_args *sa __unused)
 338 {
 339
 340         panic("null_fetch_syscall_args");
 341 }
 342
 343 static void
 344 null_set_syscall_retval(struct thread *td __unused, int error __unused)
 345 {
 346
 347         panic("null_set_syscall_retval");
 348 }
 349
 350 struct sysentvec null_sysvec = {
 351         .sv_size        = 0,
 352         .sv_table       = NULL,
 353         .sv_mask        = 0,
 354         .sv_sigsize     = 0,
 355         .sv_sigtbl      = NULL,
 356         .sv_errsize     = 0,
 357         .sv_errtbl      = NULL,
 358         .sv_transtrap   = NULL,
 359         .sv_fixup       = NULL,
 360         .sv_sendsig     = NULL,
 361         .sv_sigcode     = NULL,
 362         .sv_szsigcode   = NULL,
 363         .sv_prepsyscall = NULL,
 364         .sv_name        = "null",
 365         .sv_coredump    = NULL,
 366         .sv_imgact_try  = NULL,
 367         .sv_minsigstksz = 0,
 368         .sv_pagesize    = PAGE_SIZE,
 369         .sv_minuser     = VM_MIN_ADDRESS,
 370         .sv_maxuser     = VM_MAXUSER_ADDRESS,
 371         .sv_usrstack    = USRSTACK,
 372         .sv_psstrings   = PS_STRINGS,
 373         .sv_stackprot   = VM_PROT_ALL,
 374         .sv_copyout_strings     = NULL,
 375         .sv_setregs     = NULL,
 376         .sv_fixlimit    = NULL,
 377         .sv_maxssiz     = NULL,
 378         .sv_flags       = 0,
 379         .sv_set_syscall_retval = null_set_syscall_retval,
 380         .sv_fetch_syscall_args = null_fetch_syscall_args,
 381         .sv_syscallnames = NULL,
 382         .sv_schedtail   = NULL,
 383 };
 384
 385 /*
 386  ***************************************************************************
 387  ****
 388  **** The two following SYSINIT's are proc0 specific glue code.  I am not
 389  **** convinced that they can not be safely combined, but their order of
 390  **** operation has been maintained as the same as the original init_main.c
 391  **** for right now.
 392  ****
 393  **** These probably belong in init_proc.c or kern_proc.c, since they
 394  **** deal with proc0 (the fork template process).
 395  ****
 396  ***************************************************************************
 397  */
 398 /* ARGSUSED*/
 399 static void
 400 proc0_init(void *dummy __unused)
 401 {
 402         struct proc *p;
 403         unsigned i;
 404         struct thread *td;
 405
 406         GIANT_REQUIRED;
 407         p = &proc0;
 408         td = &thread0;
 409
 410         /*
 411          * Initialize magic number and osrel.
 412          */
 413         p->p_magic = P_MAGIC;
 414         p->p_osrel = osreldate;
 415
 416         /*
 417          * Initialize thread and process structures.
 418          */
 419         procinit();     /* set up proc zone */
 420         threadinit();   /* set up UMA zones */
 421
 422         /*
 423          * Initialise scheduler resources.
 424          * Add scheduler specific parts to proc, thread as needed.
 425          */
 426         schedinit();    /* scheduler gets its house in order */
 427         /*
 428          * Initialize sleep queue hash table
 429          */
 430         sleepinit();
 431
 432         /*
 433          * additional VM structures
 434          */
 435         vm_init2();
 436
 437         /*
 438          * Create process 0 (the swapper).
 439          */
 440         LIST_INSERT_HEAD(&allproc, p, p_list);
 441         LIST_INSERT_HEAD(PIDHASH(0), p, p_hash);
 442         mtx_init(&pgrp0.pg_mtx, "process group", NULL, MTX_DEF | MTX_DUPOK);
 443         p->p_pgrp = &pgrp0;
 444         LIST_INSERT_HEAD(PGRPHASH(0), &pgrp0, pg_hash);
 445         LIST_INIT(&pgrp0.pg_members);
 446         LIST_INSERT_HEAD(&pgrp0.pg_members, p, p_pglist);
 447
 448         pgrp0.pg_session = &session0;
 449         mtx_init(&session0.s_mtx, "session", NULL, MTX_DEF);
 450         refcount_init(&session0.s_count, 1);
 451         session0.s_leader = p;
 452
 453         p->p_sysent = &null_sysvec;
 454         p->p_flag = P_SYSTEM | P_INMEM | P_KTHREAD;
 455         p->p_state = PRS_NORMAL;
 456         knlist_init_mtx(&p->p_klist, &p->p_mtx);
 457         STAILQ_INIT(&p->p_ktr);
 458         p->p_nice = NZERO;
 459         td->td_tid = PID_MAX + 1;
 460         td->td_state = TDS_RUNNING;
 461         td->td_pri_class = PRI_TIMESHARE;
 462         td->td_user_pri = PUSER;
 463         td->td_base_user_pri = PUSER;
 464         td->td_priority = PVM;
 465         td->td_base_pri = PVM;
 466         td->td_oncpu = 0;
 467         td->td_flags = TDF_INMEM|TDP_KTHREAD;
 468         td->td_cpuset = cpuset_thread0();
 469         prison0.pr_cpuset = cpuset_ref(td->td_cpuset);
 470         p->p_peers = 0;
 471         p->p_leader = p;
 472
 473
 474         strncpy(p->p_comm, "kernel", sizeof (p->p_comm));
 475         strncpy(td->td_name, "swapper", sizeof (td->td_name));
 476
 477         callout_init(&p->p_itcallout, CALLOUT_MPSAFE);
 478         callout_init_mtx(&p->p_limco, &p->p_mtx, 0);
 479         callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
 480
 481         /* Create credentials. */
 482         p->p_ucred = crget();
 483         p->p_ucred->cr_ngroups = 1;     /* group 0 */
 484         p->p_ucred->cr_uidinfo = uifind(0);
 485         p->p_ucred->cr_ruidinfo = uifind(0);
 486         p->p_ucred->cr_prison = &prison0;
 487 #ifdef AUDIT
 488         audit_cred_kproc0(p->p_ucred);
 489 #endif
 490 #ifdef MAC
 491         mac_cred_create_swapper(p->p_ucred);
 492 #endif
 493         td->td_ucred = crhold(p->p_ucred);
 494
 495         /* Create sigacts. */
 496         p->p_sigacts = sigacts_alloc();
 497
 498         /* Initialize signal state for process 0. */
 499         siginit(&proc0);
 500
 501         /* Create the file descriptor table. */
 502         p->p_fd = fdinit(NULL);
 503         p->p_fdtol = NULL;
 504
 505         /* Create the limits structures. */
 506         p->p_limit = lim_alloc();
 507         for (i = 0; i < RLIM_NLIMITS; i++)
 508                 p->p_limit->pl_rlimit[i].rlim_cur =
 509                     p->p_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
 510         p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur =
 511             p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
 512         p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_cur =
 513             p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
 514         i = ptoa(cnt.v_free_count);
 515         p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_max = i;
 516         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_max = i;
 517         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = i / 3;
 518         p->p_cpulimit = RLIM_INFINITY;
 519
 520         p->p_stats = pstats_alloc();
 521
 522         /* Allocate a prototype map so we have something to fork. */
 523         pmap_pinit0(vmspace_pmap(&vmspace0));
 524         p->p_vmspace = &vmspace0;
 525         vmspace0.vm_refcnt = 1;
 526
 527         /*
 528          * proc0 is not expected to enter usermode, so there is no special
 529          * handling for sv_minuser here, like is done for exec_new_vmspace().
 530          */
 531         vm_map_init(&vmspace0.vm_map, p->p_sysent->sv_minuser,
 532             p->p_sysent->sv_maxuser);
 533         vmspace0.vm_map.pmap = vmspace_pmap(&vmspace0);
 534
 535         /*-
 536          * call the init and ctor for the new thread and proc
 537          * we wait to do this until all other structures
 538          * are fairly sane.
 539          */
 540         EVENTHANDLER_INVOKE(process_init, p);
 541         EVENTHANDLER_INVOKE(thread_init, td);
 542         EVENTHANDLER_INVOKE(process_ctor, p);
 543         EVENTHANDLER_INVOKE(thread_ctor, td);
 544
 545         /*
 546          * Charge root for one process.
 547          */
 548         (void)chgproccnt(p->p_ucred->cr_ruidinfo, 1, 0);
 549 }
 550 SYSINIT(p0init, SI_SUB_INTRINSIC, SI_ORDER_FIRST, proc0_init, NULL);
 551
 552 /* ARGSUSED*/
 553 static void
 554 proc0_post(void *dummy __unused)
 555 {
 556         struct timespec ts;
 557         struct proc *p;
 558         struct rusage ru;
 559         struct thread *td;
 560
 561         /*
 562          * Now we can look at the time, having had a chance to verify the
 563          * time from the filesystem.  Pretend that proc0 started now.
 564          */
 565         sx_slock(&allproc_lock);
 566         FOREACH_PROC_IN_SYSTEM(p) {
 567                 microuptime(&p->p_stats->p_start);
 568                 PROC_SLOCK(p);
 569                 rufetch(p, &ru);        /* Clears thread stats */
 570                 PROC_SUNLOCK(p);
 571                 p->p_rux.rux_runtime = 0;
 572                 p->p_rux.rux_uticks = 0;
 573                 p->p_rux.rux_sticks = 0;
 574                 p->p_rux.rux_iticks = 0;
 575                 FOREACH_THREAD_IN_PROC(p, td) {
 576                         td->td_runtime = 0;
 577                 }
 578         }
 579         sx_sunlock(&allproc_lock);
 580         PCPU_SET(switchtime, cpu_ticks());
 581         PCPU_SET(switchticks, ticks);
 582
 583         /*
 584          * Give the ``random'' number generator a thump.
 585          */
 586         nanotime(&ts);
 587         srandom(ts.tv_sec ^ ts.tv_nsec);
 588 }
 589 SYSINIT(p0post, SI_SUB_INTRINSIC_POST, SI_ORDER_FIRST, proc0_post, NULL);
 590
 591 static void
 592 random_init(void *dummy __unused)
 593 {
 594
 595         /*
 596          * After CPU has been started we have some randomness on most
 597          * platforms via get_cyclecount().  For platforms that don't
 598          * we will reseed random(9) in proc0_post() as well.
 599          */
 600         srandom(get_cyclecount());
 601 }
 602 SYSINIT(random, SI_SUB_RANDOM, SI_ORDER_FIRST, random_init, NULL);
 603
 604 /*
 605  ***************************************************************************
 606  ****
 607  **** The following SYSINIT's and glue code should be moved to the
 608  **** respective files on a per subsystem basis.
 609  ****
 610  ***************************************************************************
 611  */
 612
 613
 614 /*
 615  ***************************************************************************
 616  ****
 617  **** The following code probably belongs in another file, like
 618  **** kern/init_init.c.
 619  ****
 620  ***************************************************************************
 621  */
 622
 623 /*
 624  * List of paths to try when searching for "init".
 625  */
 626 static char init_path[MAXPATHLEN] =
 627 #ifdef  INIT_PATH
 628     __XSTRING(INIT_PATH);
 629 #else
 630     "/sbin/init:/sbin/oinit:/sbin/init.bak:/rescue/init:/stand/sysinstall";
 631 #endif
 632 SYSCTL_STRING(_kern, OID_AUTO, init_path, CTLFLAG_RD, init_path, 0,
 633         "Path used to search the init process");
 634
 635 /*
 636  * Shutdown timeout of init(8).
 637  * Unused within kernel, but used to control init(8), hence do not remove.
 638  */
 639 #ifndef INIT_SHUTDOWN_TIMEOUT
 640 #define INIT_SHUTDOWN_TIMEOUT 120
 641 #endif
 642 static int init_shutdown_timeout = INIT_SHUTDOWN_TIMEOUT;
 643 SYSCTL_INT(_kern, OID_AUTO, init_shutdown_timeout,
 644         CTLFLAG_RW, &init_shutdown_timeout, 0, "");
 645
 646 /*
 647  * Start the initial user process; try exec'ing each pathname in init_path.
 648  * The program is invoked with one argument containing the boot flags.
 649  */
 650 static void
 651 start_init(void *dummy)
 652 {
 653         vm_offset_t addr;
 654         struct execve_args args;
 655         int options, error;
 656         char *var, *path, *next, *s;
 657         char *ucp, **uap, *arg0, *arg1;
 658         struct thread *td;
 659         struct proc *p;
 660
 661         mtx_lock(&Giant);
 662
 663         GIANT_REQUIRED;
 664
 665         td = curthread;
 666         p = td->td_proc;
 667
 668         vfs_mountroot();
 669
 670         /*
 671          * Need just enough stack to hold the faked-up "execve()" arguments.
 672          */
 673         addr = p->p_sysent->sv_usrstack - PAGE_SIZE;
 674         if (vm_map_find(&p->p_vmspace->vm_map, NULL, 0, &addr, PAGE_SIZE,
 675                         FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != 0)
 676                 panic("init: couldn't allocate argument space");
 677         p->p_vmspace->vm_maxsaddr = (caddr_t)addr;
 678         p->p_vmspace->vm_ssize = 1;
 679
 680         if ((var = getenv("init_path")) != NULL) {
 681                 strlcpy(init_path, var, sizeof(init_path));
 682                 freeenv(var);
 683         }
 684
 685         for (path = init_path; *path != '\0'; path = next) {
 686                 while (*path == ':')
 687                         path++;
 688                 if (*path == '\0')
 689                         break;
 690                 for (next = path; *next != '\0' && *next != ':'; next++)
 691                         /* nothing */ ;
 692                 if (bootverbose)
 693                         printf("start_init: trying %.*s\n", (int)(next - path),
 694                             path);
 695
 696                 /*
 697                  * Move out the boot flag argument.
 698                  */
 699                 options = 0;
 700                 ucp = (char *)p->p_sysent->sv_usrstack;
 701                 (void)subyte(--ucp, 0);         /* trailing zero */
 702                 if (boothowto & RB_SINGLE) {
 703                         (void)subyte(--ucp, 's');
 704                         options = 1;
 705                 }
 706 #ifdef notyet
 707                 if (boothowto & RB_FASTBOOT) {
 708                         (void)subyte(--ucp, 'f');
 709                         options = 1;
 710                 }
 711 #endif
 712
 713 #ifdef BOOTCDROM
 714                 (void)subyte(--ucp, 'C');
 715                 options = 1;
 716 #endif
 717
 718                 if (options == 0)
 719                         (void)subyte(--ucp, '-');
 720                 (void)subyte(--ucp, '-');               /* leading hyphen */
 721                 arg1 = ucp;
 722
 723                 /*
 724                  * Move out the file name (also arg 0).
 725                  */
 726                 (void)subyte(--ucp, 0);
 727                 for (s = next - 1; s >= path; s--)
 728                         (void)subyte(--ucp, *s);
 729                 arg0 = ucp;
 730
 731                 /*
 732                  * Move out the arg pointers.
 733                  */
 734                 uap = (char **)((intptr_t)ucp & ~(sizeof(intptr_t)-1));
 735                 (void)suword((caddr_t)--uap, (long)0);  /* terminator */
 736                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg1);
 737                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg0);
 738
 739                 /*
 740                  * Point at the arguments.
 741                  */
 742                 args.fname = arg0;
 743                 args.argv = uap;
 744                 args.envv = NULL;
 745
 746                 /*
 747                  * Now try to exec the program.  If can't for any reason
 748                  * other than it doesn't exist, complain.
 749                  *
 750                  * Otherwise, return via fork_trampoline() all the way
 751                  * to user mode as init!
 752                  */
 753                 if ((error = execve(td, &args)) == 0) {
 754                         mtx_unlock(&Giant);
 755                         return;
 756                 }
 757                 if (error != ENOENT)
 758                         printf("exec %.*s: error %d\n", (int)(next - path),
 759                             path, error);
 760         }
 761         printf("init: not found in path %s\n", init_path);
 762         panic("no init");
 763 }
 764
 765 /*
 766  * Like kproc_create(), but runs in it's own address space.
 767  * We do this early to reserve pid 1.
 768  *
 769  * Note special case - do not make it runnable yet.  Other work
 770  * in progress will change this more.
 771  */
 772 static void
 773 create_init(const void *udata __unused)
 774 {
 775         struct ucred *newcred, *oldcred;
 776         int error;
 777
 778         error = fork1(&thread0, RFFDG | RFPROC | RFSTOPPED, 0, &initproc);
 779         if (error)
 780                 panic("cannot fork init: %d\n", error);
 781         KASSERT(initproc->p_pid == 1, ("create_init: initproc->p_pid != 1"));
 782         /* divorce init's credentials from the kernel's */
 783         newcred = crget();
 784         PROC_LOCK(initproc);
 785         initproc->p_flag |= P_SYSTEM | P_INMEM;
 786         oldcred = initproc->p_ucred;
 787         crcopy(newcred, oldcred);
 788 #ifdef MAC
 789         mac_cred_create_init(newcred);
 790 #endif
 791 #ifdef AUDIT
 792         audit_cred_proc1(newcred);
 793 #endif
 794         initproc->p_ucred = newcred;
 795         PROC_UNLOCK(initproc);
 796         crfree(oldcred);
 797         cred_update_thread(FIRST_THREAD_IN_PROC(initproc));
 798         cpu_set_fork_handler(FIRST_THREAD_IN_PROC(initproc), start_init, NULL);
 799 }
 800 SYSINIT(init, SI_SUB_CREATE_INIT, SI_ORDER_FIRST, create_init, NULL);
 801
 802 /*
 803  * Make it runnable now.
 804  */
 805 static void
 806 kick_init(const void *udata __unused)
 807 {
 808         struct thread *td;
 809
 810         td = FIRST_THREAD_IN_PROC(initproc);
 811         thread_lock(td);
 812         TD_SET_CAN_RUN(td);
 813         sched_add(td, SRQ_BORING);
 814         thread_unlock(td);
 815 }
 816 SYSINIT(kickinit, SI_SUB_KTHREAD_INIT, SI_ORDER_FIRST, kick_init, NULL);