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