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 int     boothowto = 0;          /* initialized so that it can be patched */
 105 SYSCTL_INT(_debug, OID_AUTO, boothowto, CTLFLAG_RD, &boothowto, 0,
 106         "Boot control flags, passed from loader");
 107 int     bootverbose;
 108 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, &bootverbose, 0,
 109         "Control the output of verbose kernel messages");
 110
 111 /*
 112  * This ensures that there is at least one entry so that the sysinit_set
 113  * symbol is not undefined.  A sybsystem ID of SI_SUB_DUMMY is never
 114  * executed.
 115  */
 116 SYSINIT(placeholder, SI_SUB_DUMMY, SI_ORDER_ANY, NULL, NULL);
 117
 118 /*
 119  * The sysinit table itself.  Items are checked off as the are run.
 120  * If we want to register new sysinit types, add them to newsysinit.
 121  */
 122 SET_DECLARE(sysinit_set, struct sysinit);
 123 struct sysinit **sysinit, **sysinit_end;
 124 struct sysinit **newsysinit, **newsysinit_end;
 125
 126 /*
 127  * Merge a new sysinit set into the current set, reallocating it if
 128  * necessary.  This can only be called after malloc is running.
 129  */
 130 void
 131 sysinit_add(struct sysinit **set, struct sysinit **set_end)
 132 {
 133         struct sysinit **newset;
 134         struct sysinit **sipp;
 135         struct sysinit **xipp;
 136         int count;
 137
 138         count = set_end - set;
 139         if (newsysinit)
 140                 count += newsysinit_end - newsysinit;
 141         else
 142                 count += sysinit_end - sysinit;
 143         newset = malloc(count * sizeof(*sipp), M_TEMP, M_NOWAIT);
 144         if (newset == NULL)
 145                 panic("cannot malloc for sysinit");
 146         xipp = newset;
 147         if (newsysinit)
 148                 for (sipp = newsysinit; sipp < newsysinit_end; sipp++)
 149                         *xipp++ = *sipp;
 150         else
 151                 for (sipp = sysinit; sipp < sysinit_end; sipp++)
 152                         *xipp++ = *sipp;
 153         for (sipp = set; sipp < set_end; sipp++)
 154                 *xipp++ = *sipp;
 155         if (newsysinit)
 156                 free(newsysinit, M_TEMP);
 157         newsysinit = newset;
 158         newsysinit_end = newset + count;
 159 }
 160
 161 #if defined (DDB) && defined(VERBOSE_SYSINIT)
 162 static const char *
 163 symbol_name(vm_offset_t va, db_strategy_t strategy)
 164 {
 165         const char *name;
 166         c_db_sym_t sym;
 167         db_expr_t  offset;
 168
 169         if (va == 0)
 170                 return (NULL);
 171         sym = db_search_symbol(va, strategy, &offset);
 172         if (offset != 0)
 173                 return (NULL);
 174         db_symbol_values(sym, &name, NULL);
 175         return (name);
 176 }
 177 #endif
 178
 179 /*
 180  * System startup; initialize the world, create process 0, mount root
 181  * filesystem, and fork to create init and pagedaemon.  Most of the
 182  * hard work is done in the lower-level initialization routines including
 183  * startup(), which does memory initialization and autoconfiguration.
 184  *
 185  * This allows simple addition of new kernel subsystems that require
 186  * boot time initialization.  It also allows substitution of subsystem
 187  * (for instance, a scheduler, kernel profiler, or VM system) by object
 188  * module.  Finally, it allows for optional "kernel threads".
 189  */
 190 void
 191 mi_startup(void)
 192 {
 193
 194         register struct sysinit **sipp;         /* system initialization*/
 195         register struct sysinit **xipp;         /* interior loop of sort*/
 196         register struct sysinit *save;          /* bubble*/
 197
 198 #if defined(VERBOSE_SYSINIT)
 199         int last;
 200         int verbose;
 201 #endif
 202
 203         if (boothowto & RB_VERBOSE)
 204                 bootverbose++;
 205
 206         if (sysinit == NULL) {
 207                 sysinit = SET_BEGIN(sysinit_set);
 208                 sysinit_end = SET_LIMIT(sysinit_set);
 209         }
 210
 211 restart:
 212         /*
 213          * Perform a bubble sort of the system initialization objects by
 214          * their subsystem (primary key) and order (secondary key).
 215          */
 216         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 217                 for (xipp = sipp + 1; xipp < sysinit_end; xipp++) {
 218                         if ((*sipp)->subsystem < (*xipp)->subsystem ||
 219                              ((*sipp)->subsystem == (*xipp)->subsystem &&
 220                               (*sipp)->order <= (*xipp)->order))
 221                                 continue;       /* skip*/
 222                         save = *sipp;
 223                         *sipp = *xipp;
 224                         *xipp = save;
 225                 }
 226         }
 227
 228 #if defined(VERBOSE_SYSINIT)
 229         last = SI_SUB_COPYRIGHT;
 230         verbose = 0;
 231 #if !defined(DDB)
 232         printf("VERBOSE_SYSINIT: DDB not enabled, symbol lookups disabled.\n");
 233 #endif
 234 #endif
 235
 236         /*
 237          * Traverse the (now) ordered list of system initialization tasks.
 238          * Perform each task, and continue on to the next task.
 239          *
 240          * The last item on the list is expected to be the scheduler,
 241          * which will not return.
 242          */
 243         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 244
 245                 if ((*sipp)->subsystem == SI_SUB_DUMMY)
 246                         continue;       /* skip dummy task(s)*/
 247
 248                 if ((*sipp)->subsystem == SI_SUB_DONE)
 249                         continue;
 250
 251 #if defined(VERBOSE_SYSINIT)
 252                 if ((*sipp)->subsystem > last) {
 253                         verbose = 1;
 254                         last = (*sipp)->subsystem;
 255                         printf("subsystem %x\n", last);
 256                 }
 257                 if (verbose) {
 258 #if defined(DDB)
 259                         const char *func, *data;
 260
 261                         func = symbol_name((vm_offset_t)(*sipp)->func,
 262                             DB_STGY_PROC);
 263                         data = symbol_name((vm_offset_t)(*sipp)->udata,
 264                             DB_STGY_ANY);
 265                         if (func != NULL && data != NULL)
 266                                 printf("   %s(&%s)... ", func, data);
 267                         else if (func != NULL)
 268                                 printf("   %s(%p)... ", func, (*sipp)->udata);
 269                         else
 270 #endif
 271                                 printf("   %p(%p)... ", (*sipp)->func,
 272                                     (*sipp)->udata);
 273                 }
 274 #endif
 275
 276                 /* Call function */
 277                 (*((*sipp)->func))((*sipp)->udata);
 278
 279 #if defined(VERBOSE_SYSINIT)
 280                 if (verbose)
 281                         printf("done.\n");
 282 #endif
 283
 284                 /* Check off the one we're just done */
 285                 (*sipp)->subsystem = SI_SUB_DONE;
 286
 287                 /* Check if we've installed more sysinit items via KLD */
 288                 if (newsysinit != NULL) {
 289                         if (sysinit != SET_BEGIN(sysinit_set))
 290                                 free(sysinit, M_TEMP);
 291                         sysinit = newsysinit;
 292                         sysinit_end = newsysinit_end;
 293                         newsysinit = NULL;
 294                         newsysinit_end = NULL;
 295                         goto restart;
 296                 }
 297         }
 298
 299         panic("Shouldn't get here!");
 300         /* NOTREACHED*/
 301 }
 302
 303
 304 /*
 305  ***************************************************************************
 306  ****
 307  **** The following SYSINIT's belong elsewhere, but have not yet
 308  **** been moved.
 309  ****
 310  ***************************************************************************
 311  */
 312 static void
 313 print_caddr_t(void *data)
 314 {
 315         printf("%s", (char *)data);
 316 }
 317
 318 static void
 319 print_version(void *data __unused)
 320 {
 321         int len;
 322
 323         /* Strip a trailing newline from version. */
 324         len = strlen(version);
 325         while (len > 0 && version[len - 1] == '\n')
 326                 len--;
 327         printf("%.*s %s\n", len, version, machine);
 328 }
 329
 330 SYSINIT(announce, SI_SUB_COPYRIGHT, SI_ORDER_FIRST, print_caddr_t,
 331     copyright);
 332 SYSINIT(trademark, SI_SUB_COPYRIGHT, SI_ORDER_SECOND, print_caddr_t,
 333     trademark);
 334 SYSINIT(version, SI_SUB_COPYRIGHT, SI_ORDER_THIRD, print_version, NULL);
 335
 336 #ifdef WITNESS
 337 static char wit_warn[] =
 338      "WARNING: WITNESS option enabled, expect reduced performance.\n";
 339 SYSINIT(witwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 1,
 340    print_caddr_t, wit_warn);
 341 SYSINIT(witwarn2, SI_SUB_RUN_SCHEDULER, SI_ORDER_THIRD + 1,
 342    print_caddr_t, wit_warn);
 343 #endif
 344
 345 #ifdef DIAGNOSTIC
 346 static char diag_warn[] =
 347      "WARNING: DIAGNOSTIC option enabled, expect reduced performance.\n";
 348 SYSINIT(diagwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 2,
 349     print_caddr_t, diag_warn);
 350 SYSINIT(diagwarn2, SI_SUB_RUN_SCHEDULER, SI_ORDER_THIRD + 2,
 351     print_caddr_t, diag_warn);
 352 #endif
 353
 354 static int
 355 null_fetch_syscall_args(struct thread *td __unused,
 356     struct syscall_args *sa __unused)
 357 {
 358
 359         panic("null_fetch_syscall_args");
 360 }
 361
 362 static void
 363 null_set_syscall_retval(struct thread *td __unused, int error __unused)
 364 {
 365
 366         panic("null_set_syscall_retval");
 367 }
 368
 369 struct sysentvec null_sysvec = {
 370         .sv_size        = 0,
 371         .sv_table       = NULL,
 372         .sv_mask        = 0,
 373         .sv_sigsize     = 0,
 374         .sv_sigtbl      = NULL,
 375         .sv_errsize     = 0,
 376         .sv_errtbl      = NULL,
 377         .sv_transtrap   = NULL,
 378         .sv_fixup       = NULL,
 379         .sv_sendsig     = NULL,
 380         .sv_sigcode     = NULL,
 381         .sv_szsigcode   = NULL,
 382         .sv_prepsyscall = NULL,
 383         .sv_name        = "null",
 384         .sv_coredump    = NULL,
 385         .sv_imgact_try  = NULL,
 386         .sv_minsigstksz = 0,
 387         .sv_pagesize    = PAGE_SIZE,
 388         .sv_minuser     = VM_MIN_ADDRESS,
 389         .sv_maxuser     = VM_MAXUSER_ADDRESS,
 390         .sv_usrstack    = USRSTACK,
 391         .sv_psstrings   = PS_STRINGS,
 392         .sv_stackprot   = VM_PROT_ALL,
 393         .sv_copyout_strings     = NULL,
 394         .sv_setregs     = NULL,
 395         .sv_fixlimit    = NULL,
 396         .sv_maxssiz     = NULL,
 397         .sv_flags       = 0,
 398         .sv_set_syscall_retval = null_set_syscall_retval,
 399         .sv_fetch_syscall_args = null_fetch_syscall_args,
 400         .sv_syscallnames = NULL,
 401         .sv_schedtail   = NULL,
 402 };
 403
 404 /*
 405  ***************************************************************************
 406  ****
 407  **** The two following SYSINIT's are proc0 specific glue code.  I am not
 408  **** convinced that they can not be safely combined, but their order of
 409  **** operation has been maintained as the same as the original init_main.c
 410  **** for right now.
 411  ****
 412  **** These probably belong in init_proc.c or kern_proc.c, since they
 413  **** deal with proc0 (the fork template process).
 414  ****
 415  ***************************************************************************
 416  */
 417 /* ARGSUSED*/
 418 static void
 419 proc0_init(void *dummy __unused)
 420 {
 421         struct proc *p;
 422         struct thread *td;
 423         vm_paddr_t pageablemem;
 424         int i;
 425
 426         GIANT_REQUIRED;
 427         p = &proc0;
 428         td = &thread0;
 429
 430         /*
 431          * Initialize magic number and osrel.
 432          */
 433         p->p_magic = P_MAGIC;
 434         p->p_osrel = osreldate;
 435
 436         /*
 437          * Initialize thread and process structures.
 438          */
 439         procinit();     /* set up proc zone */
 440         threadinit();   /* set up UMA zones */
 441
 442         /*
 443          * Initialise scheduler resources.
 444          * Add scheduler specific parts to proc, thread as needed.
 445          */
 446         schedinit();    /* scheduler gets its house in order */
 447         /*
 448          * Initialize sleep queue hash table
 449          */
 450         sleepinit();
 451
 452         /*
 453          * additional VM structures
 454          */
 455         vm_init2();
 456
 457         /*
 458          * Create process 0 (the swapper).
 459          */
 460         LIST_INSERT_HEAD(&allproc, p, p_list);
 461         LIST_INSERT_HEAD(PIDHASH(0), p, p_hash);
 462         mtx_init(&pgrp0.pg_mtx, "process group", NULL, MTX_DEF | MTX_DUPOK);
 463         p->p_pgrp = &pgrp0;
 464         LIST_INSERT_HEAD(PGRPHASH(0), &pgrp0, pg_hash);
 465         LIST_INIT(&pgrp0.pg_members);
 466         LIST_INSERT_HEAD(&pgrp0.pg_members, p, p_pglist);
 467
 468         pgrp0.pg_session = &session0;
 469         mtx_init(&session0.s_mtx, "session", NULL, MTX_DEF);
 470         refcount_init(&session0.s_count, 1);
 471         session0.s_leader = p;
 472
 473         p->p_sysent = &null_sysvec;
 474         p->p_flag = P_SYSTEM | P_INMEM;
 475         p->p_state = PRS_NORMAL;
 476         knlist_init_mtx(&p->p_klist, &p->p_mtx);
 477         STAILQ_INIT(&p->p_ktr);
 478         p->p_nice = NZERO;
 479         /* pid_max cannot be greater than PID_MAX */
 480         td->td_tid = PID_MAX + 1;
 481         LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
 482         td->td_state = TDS_RUNNING;
 483         td->td_pri_class = PRI_TIMESHARE;
 484         td->td_user_pri = PUSER;
 485         td->td_base_user_pri = PUSER;
 486         td->td_lend_user_pri = PRI_MAX;
 487         td->td_priority = PVM;
 488         td->td_base_pri = PVM;
 489         td->td_oncpu = 0;
 490         td->td_flags = TDF_INMEM|TDP_KTHREAD;
 491         td->td_cpuset = cpuset_thread0();
 492         prison0.pr_cpuset = cpuset_ref(td->td_cpuset);
 493         p->p_peers = 0;
 494         p->p_leader = p;
 495
 496
 497         strncpy(p->p_comm, "kernel", sizeof (p->p_comm));
 498         strncpy(td->td_name, "swapper", sizeof (td->td_name));
 499
 500         callout_init(&p->p_itcallout, CALLOUT_MPSAFE);
 501         callout_init_mtx(&p->p_limco, &p->p_mtx, 0);
 502         callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
 503
 504         /* Create credentials. */
 505         p->p_ucred = crget();
 506         p->p_ucred->cr_ngroups = 1;     /* group 0 */
 507         p->p_ucred->cr_uidinfo = uifind(0);
 508         p->p_ucred->cr_ruidinfo = uifind(0);
 509         p->p_ucred->cr_prison = &prison0;
 510         p->p_ucred->cr_loginclass = loginclass_find("default");
 511 #ifdef AUDIT
 512         audit_cred_kproc0(p->p_ucred);
 513 #endif
 514 #ifdef MAC
 515         mac_cred_create_swapper(p->p_ucred);
 516 #endif
 517         td->td_ucred = crhold(p->p_ucred);
 518
 519         /* Create sigacts. */
 520         p->p_sigacts = sigacts_alloc();
 521
 522         /* Initialize signal state for process 0. */
 523         siginit(&proc0);
 524
 525         /* Create the file descriptor table. */
 526         p->p_fd = fdinit(NULL);
 527         p->p_fdtol = NULL;
 528
 529         /* Create the limits structures. */
 530         p->p_limit = lim_alloc();
 531         for (i = 0; i < RLIM_NLIMITS; i++)
 532                 p->p_limit->pl_rlimit[i].rlim_cur =
 533                     p->p_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
 534         p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur =
 535             p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
 536         p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_cur =
 537             p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
 538         p->p_limit->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
 539         p->p_limit->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
 540         p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
 541         p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
 542         /* Cast to avoid overflow on i386/PAE. */
 543         pageablemem = ptoa((vm_paddr_t)cnt.v_free_count);
 544         p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_cur =
 545             p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_max = pageablemem;
 546         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = pageablemem / 3;
 547         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_max = pageablemem;
 548         p->p_cpulimit = RLIM_INFINITY;
 549
 550         /* Initialize resource accounting structures. */
 551         racct_create(&p->p_racct);
 552
 553         p->p_stats = pstats_alloc();
 554
 555         /* Allocate a prototype map so we have something to fork. */
 556         pmap_pinit0(vmspace_pmap(&vmspace0));
 557         p->p_vmspace = &vmspace0;
 558         vmspace0.vm_refcnt = 1;
 559
 560         /*
 561          * proc0 is not expected to enter usermode, so there is no special
 562          * handling for sv_minuser here, like is done for exec_new_vmspace().
 563          */
 564         vm_map_init(&vmspace0.vm_map, vmspace_pmap(&vmspace0),
 565             p->p_sysent->sv_minuser, p->p_sysent->sv_maxuser);
 566
 567         /*
 568          * Call the init and ctor for the new thread and proc.  We wait
 569          * to do this until all other structures are fairly sane.
 570          */
 571         EVENTHANDLER_INVOKE(process_init, p);
 572         EVENTHANDLER_INVOKE(thread_init, td);
 573         EVENTHANDLER_INVOKE(process_ctor, p);
 574         EVENTHANDLER_INVOKE(thread_ctor, td);
 575
 576         /*
 577          * Charge root for one process.
 578          */
 579         (void)chgproccnt(p->p_ucred->cr_ruidinfo, 1, 0);
 580         PROC_LOCK(p);
 581         racct_add_force(p, RACCT_NPROC, 1);
 582         PROC_UNLOCK(p);
 583 }
 584 SYSINIT(p0init, SI_SUB_INTRINSIC, SI_ORDER_FIRST, proc0_init, NULL);
 585
 586 /* ARGSUSED*/
 587 static void
 588 proc0_post(void *dummy __unused)
 589 {
 590         struct timespec ts;
 591         struct proc *p;
 592         struct rusage ru;
 593         struct thread *td;
 594
 595         /*
 596          * Now we can look at the time, having had a chance to verify the
 597          * time from the filesystem.  Pretend that proc0 started now.
 598          */
 599         sx_slock(&allproc_lock);
 600         FOREACH_PROC_IN_SYSTEM(p) {
 601                 microuptime(&p->p_stats->p_start);
 602                 PROC_SLOCK(p);
 603                 rufetch(p, &ru);        /* Clears thread stats */
 604                 PROC_SUNLOCK(p);
 605                 p->p_rux.rux_runtime = 0;
 606                 p->p_rux.rux_uticks = 0;
 607                 p->p_rux.rux_sticks = 0;
 608                 p->p_rux.rux_iticks = 0;
 609                 FOREACH_THREAD_IN_PROC(p, td) {
 610                         td->td_runtime = 0;
 611                 }
 612         }
 613         sx_sunlock(&allproc_lock);
 614         PCPU_SET(switchtime, cpu_ticks());
 615         PCPU_SET(switchticks, ticks);
 616
 617         /*
 618          * Give the ``random'' number generator a thump.
 619          */
 620         nanotime(&ts);
 621         srandom(ts.tv_sec ^ ts.tv_nsec);
 622 }
 623 SYSINIT(p0post, SI_SUB_INTRINSIC_POST, SI_ORDER_FIRST, proc0_post, NULL);
 624
 625 static void
 626 random_init(void *dummy __unused)
 627 {
 628
 629         /*
 630          * After CPU has been started we have some randomness on most
 631          * platforms via get_cyclecount().  For platforms that don't
 632          * we will reseed random(9) in proc0_post() as well.
 633          */
 634         srandom(get_cyclecount());
 635 }
 636 SYSINIT(random, SI_SUB_RANDOM, SI_ORDER_FIRST, random_init, NULL);
 637
 638 /*
 639  ***************************************************************************
 640  ****
 641  **** The following SYSINIT's and glue code should be moved to the
 642  **** respective files on a per subsystem basis.
 643  ****
 644  ***************************************************************************
 645  */
 646
 647
 648 /*
 649  ***************************************************************************
 650  ****
 651  **** The following code probably belongs in another file, like
 652  **** kern/init_init.c.
 653  ****
 654  ***************************************************************************
 655  */
 656
 657 /*
 658  * List of paths to try when searching for "init".
 659  */
 660 static char init_path[MAXPATHLEN] =
 661 #ifdef  INIT_PATH
 662     __XSTRING(INIT_PATH);
 663 #else
 664     "/sbin/init:/sbin/oinit:/sbin/init.bak:/rescue/init:/stand/sysinstall";
 665 #endif
 666 SYSCTL_STRING(_kern, OID_AUTO, init_path, CTLFLAG_RD, init_path, 0,
 667         "Path used to search the init process");
 668
 669 /*
 670  * Shutdown timeout of init(8).
 671  * Unused within kernel, but used to control init(8), hence do not remove.
 672  */
 673 #ifndef INIT_SHUTDOWN_TIMEOUT
 674 #define INIT_SHUTDOWN_TIMEOUT 120
 675 #endif
 676 static int init_shutdown_timeout = INIT_SHUTDOWN_TIMEOUT;
 677 SYSCTL_INT(_kern, OID_AUTO, init_shutdown_timeout,
 678         CTLFLAG_RW, &init_shutdown_timeout, 0, "Shutdown timeout of init(8). "
 679         "Unused within kernel, but used to control init(8)");
 680
 681 /*
 682  * Start the initial user process; try exec'ing each pathname in init_path.
 683  * The program is invoked with one argument containing the boot flags.
 684  */
 685 static void
 686 start_init(void *dummy)
 687 {
 688         vm_offset_t addr;
 689         struct execve_args args;
 690         int options, error;
 691         char *var, *path, *next, *s;
 692         char *ucp, **uap, *arg0, *arg1;
 693         struct thread *td;
 694         struct proc *p;
 695
 696         mtx_lock(&Giant);
 697
 698         GIANT_REQUIRED;
 699
 700         td = curthread;
 701         p = td->td_proc;
 702
 703         vfs_mountroot();
 704
 705         /*
 706          * Need just enough stack to hold the faked-up "execve()" arguments.
 707          */
 708         addr = p->p_sysent->sv_usrstack - PAGE_SIZE;
 709         if (vm_map_find(&p->p_vmspace->vm_map, NULL, 0, &addr, PAGE_SIZE,
 710                         FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != 0)
 711                 panic("init: couldn't allocate argument space");
 712         p->p_vmspace->vm_maxsaddr = (caddr_t)addr;
 713         p->p_vmspace->vm_ssize = 1;
 714
 715         if ((var = getenv("init_path")) != NULL) {
 716                 strlcpy(init_path, var, sizeof(init_path));
 717                 freeenv(var);
 718         }
 719
 720         for (path = init_path; *path != '\0'; path = next) {
 721                 while (*path == ':')
 722                         path++;
 723                 if (*path == '\0')
 724                         break;
 725                 for (next = path; *next != '\0' && *next != ':'; next++)
 726                         /* nothing */ ;
 727                 if (bootverbose)
 728                         printf("start_init: trying %.*s\n", (int)(next - path),
 729                             path);
 730
 731                 /*
 732                  * Move out the boot flag argument.
 733                  */
 734                 options = 0;
 735                 ucp = (char *)p->p_sysent->sv_usrstack;
 736                 (void)subyte(--ucp, 0);         /* trailing zero */
 737                 if (boothowto & RB_SINGLE) {
 738                         (void)subyte(--ucp, 's');
 739                         options = 1;
 740                 }
 741 #ifdef notyet
 742                 if (boothowto & RB_FASTBOOT) {
 743                         (void)subyte(--ucp, 'f');
 744                         options = 1;
 745                 }
 746 #endif
 747
 748 #ifdef BOOTCDROM
 749                 (void)subyte(--ucp, 'C');
 750                 options = 1;
 751 #endif
 752
 753                 if (options == 0)
 754                         (void)subyte(--ucp, '-');
 755                 (void)subyte(--ucp, '-');               /* leading hyphen */
 756                 arg1 = ucp;
 757
 758                 /*
 759                  * Move out the file name (also arg 0).
 760                  */
 761                 (void)subyte(--ucp, 0);
 762                 for (s = next - 1; s >= path; s--)
 763                         (void)subyte(--ucp, *s);
 764                 arg0 = ucp;
 765
 766                 /*
 767                  * Move out the arg pointers.
 768                  */
 769                 uap = (char **)((intptr_t)ucp & ~(sizeof(intptr_t)-1));
 770                 (void)suword((caddr_t)--uap, (long)0);  /* terminator */
 771                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg1);
 772                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg0);
 773
 774                 /*
 775                  * Point at the arguments.
 776                  */
 777                 args.fname = arg0;
 778                 args.argv = uap;
 779                 args.envv = NULL;
 780
 781                 /*
 782                  * Now try to exec the program.  If can't for any reason
 783                  * other than it doesn't exist, complain.
 784                  *
 785                  * Otherwise, return via fork_trampoline() all the way
 786                  * to user mode as init!
 787                  */
 788                 if ((error = sys_execve(td, &args)) == 0) {
 789                         mtx_unlock(&Giant);
 790                         return;
 791                 }
 792                 if (error != ENOENT)
 793                         printf("exec %.*s: error %d\n", (int)(next - path),
 794                             path, error);
 795         }
 796         printf("init: not found in path %s\n", init_path);
 797         panic("no init");
 798 }
 799
 800 /*
 801  * Like kproc_create(), but runs in it's own address space.
 802  * We do this early to reserve pid 1.
 803  *
 804  * Note special case - do not make it runnable yet.  Other work
 805  * in progress will change this more.
 806  */
 807 static void
 808 create_init(const void *udata __unused)
 809 {
 810         struct ucred *newcred, *oldcred;
 811         int error;
 812
 813         error = fork1(&thread0, RFFDG | RFPROC | RFSTOPPED, 0, &initproc,
 814             NULL, 0);
 815         if (error)
 816                 panic("cannot fork init: %d\n", error);
 817         KASSERT(initproc->p_pid == 1, ("create_init: initproc->p_pid != 1"));
 818         /* divorce init's credentials from the kernel's */
 819         newcred = crget();
 820         PROC_LOCK(initproc);
 821         initproc->p_flag |= P_SYSTEM | P_INMEM;
 822         oldcred = initproc->p_ucred;
 823         crcopy(newcred, oldcred);
 824 #ifdef MAC
 825         mac_cred_create_init(newcred);
 826 #endif
 827 #ifdef AUDIT
 828         audit_cred_proc1(newcred);
 829 #endif
 830         initproc->p_ucred = newcred;
 831         PROC_UNLOCK(initproc);
 832         crfree(oldcred);
 833         cred_update_thread(FIRST_THREAD_IN_PROC(initproc));
 834         cpu_set_fork_handler(FIRST_THREAD_IN_PROC(initproc), start_init, NULL);
 835 }
 836 SYSINIT(init, SI_SUB_CREATE_INIT, SI_ORDER_FIRST, create_init, NULL);
 837
 838 /*
 839  * Make it runnable now.
 840  */
 841 static void
 842 kick_init(const void *udata __unused)
 843 {
 844         struct thread *td;
 845
 846         td = FIRST_THREAD_IN_PROC(initproc);
 847         thread_lock(td);
 848         TD_SET_CAN_RUN(td);
 849         sched_add(td, SRQ_BORING);
 850         thread_unlock(td);
 851 }
 852 SYSINIT(kickinit, SI_SUB_KTHREAD_INIT, SI_ORDER_FIRST, kick_init, NULL);