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/loginclass.h>
  59 #include <sys/mount.h>
  60 #include <sys/mutex.h>
  61 #include <sys/syscallsubr.h>
  62 #include <sys/sysctl.h>
  63 #include <sys/proc.h>
  64 #include <sys/racct.h>
  65 #include <sys/resourcevar.h>
  66 #include <sys/systm.h>
  67 #include <sys/signalvar.h>
  68 #include <sys/vnode.h>
  69 #include <sys/sysent.h>
  70 #include <sys/reboot.h>
  71 #include <sys/sched.h>
  72 #include <sys/sx.h>
  73 #include <sys/sysproto.h>
  74 #include <sys/vmmeter.h>
  75 #include <sys/unistd.h>
  76 #include <sys/malloc.h>
  77 #include <sys/conf.h>
  78 #include <sys/cpuset.h>
  79
  80 #include <machine/cpu.h>
  81
  82 #include <security/audit/audit.h>
  83 #include <security/mac/mac_framework.h>
  84
  85 #include <vm/vm.h>
  86 #include <vm/vm_param.h>
  87 #include <vm/pmap.h>
  88 #include <vm/vm_map.h>
  89 #include <sys/copyright.h>
  90
  91 #include <ddb/ddb.h>
  92 #include <ddb/db_sym.h>
  93
  94 void mi_startup(void);                          /* Should be elsewhere */
  95
  96 /* Components of the first process -- never freed. */
  97 static struct session session0;
  98 static struct pgrp pgrp0;
  99 struct  proc proc0;
 100 struct  thread thread0 __aligned(16);
 101 struct  vmspace vmspace0;
 102 struct  proc *initproc;
 103
 104 #ifndef BOOTHOWTO
 105 #define BOOTHOWTO       0
 106 #endif
 107 int     boothowto = BOOTHOWTO;  /* initialized so that it can be patched */
 108 SYSCTL_INT(_debug, OID_AUTO, boothowto, CTLFLAG_RD, &boothowto, 0,
 109         "Boot control flags, passed from loader");
 110
 111 #ifndef BOOTVERBOSE
 112 #define BOOTVERBOSE     0
 113 #endif
 114 int     bootverbose = BOOTVERBOSE;
 115 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, &bootverbose, 0,
 116         "Control the output of verbose kernel messages");
 117
 118 #ifdef INVARIANTS
 119 FEATURE(invariants, "Kernel compiled with INVARIANTS, may affect performance");
 120 #endif
 121
 122 /*
 123  * This ensures that there is at least one entry so that the sysinit_set
 124  * symbol is not undefined.  A sybsystem ID of SI_SUB_DUMMY is never
 125  * executed.
 126  */
 127 SYSINIT(placeholder, SI_SUB_DUMMY, SI_ORDER_ANY, NULL, NULL);
 128
 129 /*
 130  * The sysinit table itself.  Items are checked off as the are run.
 131  * If we want to register new sysinit types, add them to newsysinit.
 132  */
 133 SET_DECLARE(sysinit_set, struct sysinit);
 134 struct sysinit **sysinit, **sysinit_end;
 135 struct sysinit **newsysinit, **newsysinit_end;
 136
 137 /*
 138  * Merge a new sysinit set into the current set, reallocating it if
 139  * necessary.  This can only be called after malloc is running.
 140  */
 141 void
 142 sysinit_add(struct sysinit **set, struct sysinit **set_end)
 143 {
 144         struct sysinit **newset;
 145         struct sysinit **sipp;
 146         struct sysinit **xipp;
 147         int count;
 148
 149         count = set_end - set;
 150         if (newsysinit)
 151                 count += newsysinit_end - newsysinit;
 152         else
 153                 count += sysinit_end - sysinit;
 154         newset = malloc(count * sizeof(*sipp), M_TEMP, M_NOWAIT);
 155         if (newset == NULL)
 156                 panic("cannot malloc for sysinit");
 157         xipp = newset;
 158         if (newsysinit)
 159                 for (sipp = newsysinit; sipp < newsysinit_end; sipp++)
 160                         *xipp++ = *sipp;
 161         else
 162                 for (sipp = sysinit; sipp < sysinit_end; sipp++)
 163                         *xipp++ = *sipp;
 164         for (sipp = set; sipp < set_end; sipp++)
 165                 *xipp++ = *sipp;
 166         if (newsysinit)
 167                 free(newsysinit, M_TEMP);
 168         newsysinit = newset;
 169         newsysinit_end = newset + count;
 170 }
 171
 172 #if defined (DDB) && defined(VERBOSE_SYSINIT)
 173 static const char *
 174 symbol_name(vm_offset_t va, db_strategy_t strategy)
 175 {
 176         const char *name;
 177         c_db_sym_t sym;
 178         db_expr_t  offset;
 179
 180         if (va == 0)
 181                 return (NULL);
 182         sym = db_search_symbol(va, strategy, &offset);
 183         if (offset != 0)
 184                 return (NULL);
 185         db_symbol_values(sym, &name, NULL);
 186         return (name);
 187 }
 188 #endif
 189
 190 /*
 191  * System startup; initialize the world, create process 0, mount root
 192  * filesystem, and fork to create init and pagedaemon.  Most of the
 193  * hard work is done in the lower-level initialization routines including
 194  * startup(), which does memory initialization and autoconfiguration.
 195  *
 196  * This allows simple addition of new kernel subsystems that require
 197  * boot time initialization.  It also allows substitution of subsystem
 198  * (for instance, a scheduler, kernel profiler, or VM system) by object
 199  * module.  Finally, it allows for optional "kernel threads".
 200  */
 201 void
 202 mi_startup(void)
 203 {
 204
 205         register struct sysinit **sipp;         /* system initialization*/
 206         register struct sysinit **xipp;         /* interior loop of sort*/
 207         register struct sysinit *save;          /* bubble*/
 208
 209 #if defined(VERBOSE_SYSINIT)
 210         int last;
 211         int verbose;
 212 #endif
 213
 214         if (boothowto & RB_VERBOSE)
 215                 bootverbose++;
 216
 217         if (sysinit == NULL) {
 218                 sysinit = SET_BEGIN(sysinit_set);
 219                 sysinit_end = SET_LIMIT(sysinit_set);
 220         }
 221
 222 restart:
 223         /*
 224          * Perform a bubble sort of the system initialization objects by
 225          * their subsystem (primary key) and order (secondary key).
 226          */
 227         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 228                 for (xipp = sipp + 1; xipp < sysinit_end; xipp++) {
 229                         if ((*sipp)->subsystem < (*xipp)->subsystem ||
 230                              ((*sipp)->subsystem == (*xipp)->subsystem &&
 231                               (*sipp)->order <= (*xipp)->order))
 232                                 continue;       /* skip*/
 233                         save = *sipp;
 234                         *sipp = *xipp;
 235                         *xipp = save;
 236                 }
 237         }
 238
 239 #if defined(VERBOSE_SYSINIT)
 240         last = SI_SUB_COPYRIGHT;
 241         verbose = 0;
 242 #if !defined(DDB)
 243         printf("VERBOSE_SYSINIT: DDB not enabled, symbol lookups disabled.\n");
 244 #endif
 245 #endif
 246
 247         /*
 248          * Traverse the (now) ordered list of system initialization tasks.
 249          * Perform each task, and continue on to the next task.
 250          */
 251         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 252
 253                 if ((*sipp)->subsystem == SI_SUB_DUMMY)
 254                         continue;       /* skip dummy task(s)*/
 255
 256                 if ((*sipp)->subsystem == SI_SUB_DONE)
 257                         continue;
 258
 259 #if defined(VERBOSE_SYSINIT)
 260                 if ((*sipp)->subsystem > last) {
 261                         verbose = 1;
 262                         last = (*sipp)->subsystem;
 263                         printf("subsystem %x\n", last);
 264                 }
 265                 if (verbose) {
 266 #if defined(DDB)
 267                         const char *func, *data;
 268
 269                         func = symbol_name((vm_offset_t)(*sipp)->func,
 270                             DB_STGY_PROC);
 271                         data = symbol_name((vm_offset_t)(*sipp)->udata,
 272                             DB_STGY_ANY);
 273                         if (func != NULL && data != NULL)
 274                                 printf("   %s(&%s)... ", func, data);
 275                         else if (func != NULL)
 276                                 printf("   %s(%p)... ", func, (*sipp)->udata);
 277                         else
 278 #endif
 279                                 printf("   %p(%p)... ", (*sipp)->func,
 280                                     (*sipp)->udata);
 281                 }
 282 #endif
 283
 284                 /* Call function */
 285                 (*((*sipp)->func))((*sipp)->udata);
 286
 287 #if defined(VERBOSE_SYSINIT)
 288                 if (verbose)
 289                         printf("done.\n");
 290 #endif
 291
 292                 /* Check off the one we're just done */
 293                 (*sipp)->subsystem = SI_SUB_DONE;
 294
 295                 /* Check if we've installed more sysinit items via KLD */
 296                 if (newsysinit != NULL) {
 297                         if (sysinit != SET_BEGIN(sysinit_set))
 298                                 free(sysinit, M_TEMP);
 299                         sysinit = newsysinit;
 300                         sysinit_end = newsysinit_end;
 301                         newsysinit = NULL;
 302                         newsysinit_end = NULL;
 303                         goto restart;
 304                 }
 305         }
 306
 307         mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED);
 308         mtx_unlock(&Giant);
 309
 310         /*
 311          * Now hand over this thread to swapper.
 312          */
 313         swapper();
 314         /* NOTREACHED*/
 315 }
 316
 317
 318 /*
 319  ***************************************************************************
 320  ****
 321  **** The following SYSINIT's belong elsewhere, but have not yet
 322  **** been moved.
 323  ****
 324  ***************************************************************************
 325  */
 326 static void
 327 print_caddr_t(void *data)
 328 {
 329         printf("%s", (char *)data);
 330 }
 331
 332 static void
 333 print_version(void *data __unused)
 334 {
 335         int len;
 336
 337         /* Strip a trailing newline from version. */
 338         len = strlen(version);
 339         while (len > 0 && version[len - 1] == '\n')
 340                 len--;
 341         printf("%.*s %s\n", len, version, machine);
 342         printf("%s\n", compiler_version);
 343 }
 344
 345 SYSINIT(announce, SI_SUB_COPYRIGHT, SI_ORDER_FIRST, print_caddr_t,
 346     copyright);
 347 SYSINIT(trademark, SI_SUB_COPYRIGHT, SI_ORDER_SECOND, print_caddr_t,
 348     trademark);
 349 SYSINIT(version, SI_SUB_COPYRIGHT, SI_ORDER_THIRD, print_version, NULL);
 350
 351 #ifdef WITNESS
 352 static char wit_warn[] =
 353      "WARNING: WITNESS option enabled, expect reduced performance.\n";
 354 SYSINIT(witwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 1,
 355    print_caddr_t, wit_warn);
 356 SYSINIT(witwarn2, SI_SUB_LAST, SI_ORDER_THIRD + 1,
 357    print_caddr_t, wit_warn);
 358 #endif
 359
 360 #ifdef DIAGNOSTIC
 361 static char diag_warn[] =
 362      "WARNING: DIAGNOSTIC option enabled, expect reduced performance.\n";
 363 SYSINIT(diagwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 2,
 364     print_caddr_t, diag_warn);
 365 SYSINIT(diagwarn2, SI_SUB_LAST, SI_ORDER_THIRD + 2,
 366     print_caddr_t, diag_warn);
 367 #endif
 368
 369 static int
 370 null_fetch_syscall_args(struct thread *td __unused,
 371     struct syscall_args *sa __unused)
 372 {
 373
 374         panic("null_fetch_syscall_args");
 375 }
 376
 377 static void
 378 null_set_syscall_retval(struct thread *td __unused, int error __unused)
 379 {
 380
 381         panic("null_set_syscall_retval");
 382 }
 383
 384 struct sysentvec null_sysvec = {
 385         .sv_size        = 0,
 386         .sv_table       = NULL,
 387         .sv_mask        = 0,
 388         .sv_sigsize     = 0,
 389         .sv_sigtbl      = NULL,
 390         .sv_errsize     = 0,
 391         .sv_errtbl      = NULL,
 392         .sv_transtrap   = NULL,
 393         .sv_fixup       = NULL,
 394         .sv_sendsig     = NULL,
 395         .sv_sigcode     = NULL,
 396         .sv_szsigcode   = NULL,
 397         .sv_prepsyscall = NULL,
 398         .sv_name        = "null",
 399         .sv_coredump    = NULL,
 400         .sv_imgact_try  = NULL,
 401         .sv_minsigstksz = 0,
 402         .sv_pagesize    = PAGE_SIZE,
 403         .sv_minuser     = VM_MIN_ADDRESS,
 404         .sv_maxuser     = VM_MAXUSER_ADDRESS,
 405         .sv_usrstack    = USRSTACK,
 406         .sv_psstrings   = PS_STRINGS,
 407         .sv_stackprot   = VM_PROT_ALL,
 408         .sv_copyout_strings     = NULL,
 409         .sv_setregs     = NULL,
 410         .sv_fixlimit    = NULL,
 411         .sv_maxssiz     = NULL,
 412         .sv_flags       = 0,
 413         .sv_set_syscall_retval = null_set_syscall_retval,
 414         .sv_fetch_syscall_args = null_fetch_syscall_args,
 415         .sv_syscallnames = NULL,
 416         .sv_schedtail   = NULL,
 417         .sv_thread_detach = NULL,
 418 };
 419
 420 /*
 421  ***************************************************************************
 422  ****
 423  **** The two following SYSINIT's are proc0 specific glue code.  I am not
 424  **** convinced that they can not be safely combined, but their order of
 425  **** operation has been maintained as the same as the original init_main.c
 426  **** for right now.
 427  ****
 428  **** These probably belong in init_proc.c or kern_proc.c, since they
 429  **** deal with proc0 (the fork template process).
 430  ****
 431  ***************************************************************************
 432  */
 433 /* ARGSUSED*/
 434 static void
 435 proc0_init(void *dummy __unused)
 436 {
 437         struct proc *p;
 438         struct thread *td;
 439         vm_paddr_t pageablemem;
 440         int i;
 441
 442         GIANT_REQUIRED;
 443         p = &proc0;
 444         td = &thread0;
 445
 446         /*
 447          * Initialize magic number and osrel.
 448          */
 449         p->p_magic = P_MAGIC;
 450         p->p_osrel = osreldate;
 451
 452         /*
 453          * Initialize thread and process structures.
 454          */
 455         procinit();     /* set up proc zone */
 456         threadinit();   /* set up UMA zones */
 457
 458         /*
 459          * Initialise scheduler resources.
 460          * Add scheduler specific parts to proc, thread as needed.
 461          */
 462         schedinit();    /* scheduler gets its house in order */
 463
 464         /*
 465          * Create process 0 (the swapper).
 466          */
 467         LIST_INSERT_HEAD(&allproc, p, p_list);
 468         LIST_INSERT_HEAD(PIDHASH(0), p, p_hash);
 469         mtx_init(&pgrp0.pg_mtx, "process group", NULL, MTX_DEF | MTX_DUPOK);
 470         p->p_pgrp = &pgrp0;
 471         LIST_INSERT_HEAD(PGRPHASH(0), &pgrp0, pg_hash);
 472         LIST_INIT(&pgrp0.pg_members);
 473         LIST_INSERT_HEAD(&pgrp0.pg_members, p, p_pglist);
 474
 475         pgrp0.pg_session = &session0;
 476         mtx_init(&session0.s_mtx, "session", NULL, MTX_DEF);
 477         refcount_init(&session0.s_count, 1);
 478         session0.s_leader = p;
 479
 480         p->p_sysent = &null_sysvec;
 481         p->p_flag = P_SYSTEM | P_INMEM;
 482         p->p_flag2 = 0;
 483         p->p_state = PRS_NORMAL;
 484         knlist_init_mtx(&p->p_klist, &p->p_mtx);
 485         STAILQ_INIT(&p->p_ktr);
 486         p->p_nice = NZERO;
 487         /* pid_max cannot be greater than PID_MAX */
 488         td->td_tid = PID_MAX + 1;
 489         LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
 490         td->td_state = TDS_RUNNING;
 491         td->td_pri_class = PRI_TIMESHARE;
 492         td->td_user_pri = PUSER;
 493         td->td_base_user_pri = PUSER;
 494         td->td_lend_user_pri = PRI_MAX;
 495         td->td_priority = PVM;
 496         td->td_base_pri = PVM;
 497         td->td_oncpu = 0;
 498         td->td_flags = TDF_INMEM;
 499         td->td_pflags = TDP_KTHREAD;
 500         td->td_cpuset = cpuset_thread0();
 501         prison0_init();
 502         p->p_peers = 0;
 503         p->p_leader = p;
 504         p->p_reaper = p;
 505         LIST_INIT(&p->p_reaplist);
 506
 507         strncpy(p->p_comm, "kernel", sizeof (p->p_comm));
 508         strncpy(td->td_name, "swapper", sizeof (td->td_name));
 509
 510         callout_init_mtx(&p->p_itcallout, &p->p_mtx, 0);
 511         callout_init_mtx(&p->p_limco, &p->p_mtx, 0);
 512         callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
 513
 514         /* Create credentials. */
 515         p->p_ucred = crget();
 516         p->p_ucred->cr_ngroups = 1;     /* group 0 */
 517         p->p_ucred->cr_uidinfo = uifind(0);
 518         p->p_ucred->cr_ruidinfo = uifind(0);
 519         p->p_ucred->cr_prison = &prison0;
 520         p->p_ucred->cr_loginclass = loginclass_find("default");
 521 #ifdef AUDIT
 522         audit_cred_kproc0(p->p_ucred);
 523 #endif
 524 #ifdef MAC
 525         mac_cred_create_swapper(p->p_ucred);
 526 #endif
 527         td->td_ucred = crhold(p->p_ucred);
 528
 529         /* Create sigacts. */
 530         p->p_sigacts = sigacts_alloc();
 531
 532         /* Initialize signal state for process 0. */
 533         siginit(&proc0);
 534
 535         /* Create the file descriptor table. */
 536         p->p_fd = fdinit(NULL);
 537         p->p_fdtol = NULL;
 538
 539         /* Create the limits structures. */
 540         p->p_limit = lim_alloc();
 541         for (i = 0; i < RLIM_NLIMITS; i++)
 542                 p->p_limit->pl_rlimit[i].rlim_cur =
 543                     p->p_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
 544         p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur =
 545             p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
 546         p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_cur =
 547             p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
 548         p->p_limit->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
 549         p->p_limit->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
 550         p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
 551         p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
 552         /* Cast to avoid overflow on i386/PAE. */
 553         pageablemem = ptoa((vm_paddr_t)cnt.v_free_count);
 554         p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_cur =
 555             p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_max = pageablemem;
 556         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = pageablemem / 3;
 557         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_max = pageablemem;
 558         p->p_cpulimit = RLIM_INFINITY;
 559
 560         /* Initialize resource accounting structures. */
 561         racct_create(&p->p_racct);
 562
 563         p->p_stats = pstats_alloc();
 564
 565         /* Allocate a prototype map so we have something to fork. */
 566         pmap_pinit0(vmspace_pmap(&vmspace0));
 567         p->p_vmspace = &vmspace0;
 568         vmspace0.vm_refcnt = 1;
 569
 570         /*
 571          * proc0 is not expected to enter usermode, so there is no special
 572          * handling for sv_minuser here, like is done for exec_new_vmspace().
 573          */
 574         vm_map_init(&vmspace0.vm_map, vmspace_pmap(&vmspace0),
 575             p->p_sysent->sv_minuser, p->p_sysent->sv_maxuser);
 576
 577         /*
 578          * Call the init and ctor for the new thread and proc.  We wait
 579          * to do this until all other structures are fairly sane.
 580          */
 581         EVENTHANDLER_INVOKE(process_init, p);
 582         EVENTHANDLER_INVOKE(thread_init, td);
 583         EVENTHANDLER_INVOKE(process_ctor, p);
 584         EVENTHANDLER_INVOKE(thread_ctor, td);
 585
 586         /*
 587          * Charge root for one process.
 588          */
 589         (void)chgproccnt(p->p_ucred->cr_ruidinfo, 1, 0);
 590         PROC_LOCK(p);
 591         racct_add_force(p, RACCT_NPROC, 1);
 592         PROC_UNLOCK(p);
 593 }
 594 SYSINIT(p0init, SI_SUB_INTRINSIC, SI_ORDER_FIRST, proc0_init, NULL);
 595
 596 /* ARGSUSED*/
 597 static void
 598 proc0_post(void *dummy __unused)
 599 {
 600         struct timespec ts;
 601         struct proc *p;
 602         struct rusage ru;
 603         struct thread *td;
 604
 605         /*
 606          * Now we can look at the time, having had a chance to verify the
 607          * time from the filesystem.  Pretend that proc0 started now.
 608          */
 609         sx_slock(&allproc_lock);
 610         FOREACH_PROC_IN_SYSTEM(p) {
 611                 microuptime(&p->p_stats->p_start);
 612                 PROC_STATLOCK(p);
 613                 rufetch(p, &ru);        /* Clears thread stats */
 614                 PROC_STATUNLOCK(p);
 615                 p->p_rux.rux_runtime = 0;
 616                 p->p_rux.rux_uticks = 0;
 617                 p->p_rux.rux_sticks = 0;
 618                 p->p_rux.rux_iticks = 0;
 619                 FOREACH_THREAD_IN_PROC(p, td) {
 620                         td->td_runtime = 0;
 621                 }
 622         }
 623         sx_sunlock(&allproc_lock);
 624         PCPU_SET(switchtime, cpu_ticks());
 625         PCPU_SET(switchticks, ticks);
 626
 627         /*
 628          * Give the ``random'' number generator a thump.
 629          */
 630         nanotime(&ts);
 631         srandom(ts.tv_sec ^ ts.tv_nsec);
 632 }
 633 SYSINIT(p0post, SI_SUB_INTRINSIC_POST, SI_ORDER_FIRST, proc0_post, NULL);
 634
 635 static void
 636 random_init(void *dummy __unused)
 637 {
 638
 639         /*
 640          * After CPU has been started we have some randomness on most
 641          * platforms via get_cyclecount().  For platforms that don't
 642          * we will reseed random(9) in proc0_post() as well.
 643          */
 644         srandom(get_cyclecount());
 645 }
 646 SYSINIT(random, SI_SUB_RANDOM, SI_ORDER_FIRST, random_init, NULL);
 647
 648 /*
 649  ***************************************************************************
 650  ****
 651  **** The following SYSINIT's and glue code should be moved to the
 652  **** respective files on a per subsystem basis.
 653  ****
 654  ***************************************************************************
 655  */
 656
 657
 658 /*
 659  ***************************************************************************
 660  ****
 661  **** The following code probably belongs in another file, like
 662  **** kern/init_init.c.
 663  ****
 664  ***************************************************************************
 665  */
 666
 667 /*
 668  * List of paths to try when searching for "init".
 669  */
 670 static char init_path[MAXPATHLEN] =
 671 #ifdef  INIT_PATH
 672     __XSTRING(INIT_PATH);
 673 #else
 674     "/sbin/init:/sbin/oinit:/sbin/init.bak:/rescue/init";
 675 #endif
 676 SYSCTL_STRING(_kern, OID_AUTO, init_path, CTLFLAG_RD, init_path, 0,
 677         "Path used to search the init process");
 678
 679 /*
 680  * Shutdown timeout of init(8).
 681  * Unused within kernel, but used to control init(8), hence do not remove.
 682  */
 683 #ifndef INIT_SHUTDOWN_TIMEOUT
 684 #define INIT_SHUTDOWN_TIMEOUT 120
 685 #endif
 686 static int init_shutdown_timeout = INIT_SHUTDOWN_TIMEOUT;
 687 SYSCTL_INT(_kern, OID_AUTO, init_shutdown_timeout,
 688         CTLFLAG_RW, &init_shutdown_timeout, 0, "Shutdown timeout of init(8). "
 689         "Unused within kernel, but used to control init(8)");
 690
 691 /*
 692  * Start the initial user process; try exec'ing each pathname in init_path.
 693  * The program is invoked with one argument containing the boot flags.
 694  */
 695 static void
 696 start_init(void *dummy)
 697 {
 698         vm_offset_t addr;
 699         struct execve_args args;
 700         int options, error;
 701         char *var, *path, *next, *s;
 702         char *ucp, **uap, *arg0, *arg1;
 703         struct thread *td;
 704         struct proc *p;
 705
 706         mtx_lock(&Giant);
 707
 708         GIANT_REQUIRED;
 709
 710         td = curthread;
 711         p = td->td_proc;
 712
 713         vfs_mountroot();
 714
 715         /* Wipe GELI passphrase from the environment. */
 716         unsetenv("kern.geom.eli.passphrase");
 717
 718         /*
 719          * Need just enough stack to hold the faked-up "execve()" arguments.
 720          */
 721         addr = p->p_sysent->sv_usrstack - PAGE_SIZE;
 722         if (vm_map_find(&p->p_vmspace->vm_map, NULL, 0, &addr, PAGE_SIZE, 0,
 723             VMFS_NO_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0) != 0)
 724                 panic("init: couldn't allocate argument space");
 725         p->p_vmspace->vm_maxsaddr = (caddr_t)addr;
 726         p->p_vmspace->vm_ssize = 1;
 727
 728         if ((var = getenv("init_path")) != NULL) {
 729                 strlcpy(init_path, var, sizeof(init_path));
 730                 freeenv(var);
 731         }
 732
 733         for (path = init_path; *path != '\0'; path = next) {
 734                 while (*path == ':')
 735                         path++;
 736                 if (*path == '\0')
 737                         break;
 738                 for (next = path; *next != '\0' && *next != ':'; next++)
 739                         /* nothing */ ;
 740                 if (bootverbose)
 741                         printf("start_init: trying %.*s\n", (int)(next - path),
 742                             path);
 743
 744                 /*
 745                  * Move out the boot flag argument.
 746                  */
 747                 options = 0;
 748                 ucp = (char *)p->p_sysent->sv_usrstack;
 749                 (void)subyte(--ucp, 0);         /* trailing zero */
 750                 if (boothowto & RB_SINGLE) {
 751                         (void)subyte(--ucp, 's');
 752                         options = 1;
 753                 }
 754 #ifdef notyet
 755                 if (boothowto & RB_FASTBOOT) {
 756                         (void)subyte(--ucp, 'f');
 757                         options = 1;
 758                 }
 759 #endif
 760
 761 #ifdef BOOTCDROM
 762                 (void)subyte(--ucp, 'C');
 763                 options = 1;
 764 #endif
 765
 766                 if (options == 0)
 767                         (void)subyte(--ucp, '-');
 768                 (void)subyte(--ucp, '-');               /* leading hyphen */
 769                 arg1 = ucp;
 770
 771                 /*
 772                  * Move out the file name (also arg 0).
 773                  */
 774                 (void)subyte(--ucp, 0);
 775                 for (s = next - 1; s >= path; s--)
 776                         (void)subyte(--ucp, *s);
 777                 arg0 = ucp;
 778
 779                 /*
 780                  * Move out the arg pointers.
 781                  */
 782                 uap = (char **)((intptr_t)ucp & ~(sizeof(intptr_t)-1));
 783                 (void)suword((caddr_t)--uap, (long)0);  /* terminator */
 784                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg1);
 785                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg0);
 786
 787                 /*
 788                  * Point at the arguments.
 789                  */
 790                 args.fname = arg0;
 791                 args.argv = uap;
 792                 args.envv = NULL;
 793
 794                 /*
 795                  * Now try to exec the program.  If can't for any reason
 796                  * other than it doesn't exist, complain.
 797                  *
 798                  * Otherwise, return via fork_trampoline() all the way
 799                  * to user mode as init!
 800                  */
 801                 if ((error = sys_execve(td, &args)) == 0) {
 802                         mtx_unlock(&Giant);
 803                         return;
 804                 }
 805                 if (error != ENOENT)
 806                         printf("exec %.*s: error %d\n", (int)(next - path),
 807                             path, error);
 808         }
 809         printf("init: not found in path %s\n", init_path);
 810         panic("no init");
 811 }
 812
 813 /*
 814  * Like kproc_create(), but runs in it's own address space.
 815  * We do this early to reserve pid 1.
 816  *
 817  * Note special case - do not make it runnable yet.  Other work
 818  * in progress will change this more.
 819  */
 820 static void
 821 create_init(const void *udata __unused)
 822 {
 823         struct ucred *newcred, *oldcred;
 824         int error;
 825
 826         error = fork1(&thread0, RFFDG | RFPROC | RFSTOPPED, 0, &initproc,
 827             NULL, 0);
 828         if (error)
 829                 panic("cannot fork init: %d\n", error);
 830         KASSERT(initproc->p_pid == 1, ("create_init: initproc->p_pid != 1"));
 831         /* divorce init's credentials from the kernel's */
 832         newcred = crget();
 833         sx_xlock(&proctree_lock);
 834         PROC_LOCK(initproc);
 835         initproc->p_flag |= P_SYSTEM | P_INMEM;
 836         initproc->p_treeflag |= P_TREE_REAPER;
 837         LIST_INSERT_HEAD(&initproc->p_reaplist, &proc0, p_reapsibling);
 838         oldcred = initproc->p_ucred;
 839         crcopy(newcred, oldcred);
 840 #ifdef MAC
 841         mac_cred_create_init(newcred);
 842 #endif
 843 #ifdef AUDIT
 844         audit_cred_proc1(newcred);
 845 #endif
 846         initproc->p_ucred = newcred;
 847         PROC_UNLOCK(initproc);
 848         sx_xunlock(&proctree_lock);
 849         crfree(oldcred);
 850         cred_update_thread(FIRST_THREAD_IN_PROC(initproc));
 851         cpu_set_fork_handler(FIRST_THREAD_IN_PROC(initproc), start_init, NULL);
 852 }
 853 SYSINIT(init, SI_SUB_CREATE_INIT, SI_ORDER_FIRST, create_init, NULL);
 854
 855 /*
 856  * Make it runnable now.
 857  */
 858 static void
 859 kick_init(const void *udata __unused)
 860 {
 861         struct thread *td;
 862
 863         td = FIRST_THREAD_IN_PROC(initproc);
 864         thread_lock(td);
 865         TD_SET_CAN_RUN(td);
 866         sched_add(td, SRQ_BORING);
 867         thread_unlock(td);
 868 }
 869 SYSINIT(kickinit, SI_SUB_KTHREAD_INIT, SI_ORDER_FIRST, kick_init, NULL);