sys/kern/init_main.c

   1 /*-
   2  * Copyright (c) 1995 Terrence R. Lambert
   3  * All rights reserved.
   4  *
   5  * Copyright (c) 1982, 1986, 1989, 1991, 1992, 1993
   6  *      The Regents of the University of California.  All rights reserved.
   7  * (c) UNIX System Laboratories, Inc.
   8  * All or some portions of this file are derived from material licensed
   9  * to the University of California by American Telephone and Telegraph
  10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
  11  * the permission of UNIX System Laboratories, Inc.
  12  *
  13  * Redistribution and use in source and binary forms, with or without
  14  * modification, are permitted provided that the following conditions
  15  * are met:
  16  * 1. Redistributions of source code must retain the above copyright
  17  *    notice, this list of conditions and the following disclaimer.
  18  * 2. Redistributions in binary form must reproduce the above copyright
  19  *    notice, this list of conditions and the following disclaimer in the
  20  *    documentation and/or other materials provided with the distribution.
  21  * 3. All advertising materials mentioning features or use of this software
  22  *    must display the following acknowledgement:
  23  *      This product includes software developed by the University of
  24  *      California, Berkeley and its contributors.
  25  * 4. Neither the name of the University nor the names of its contributors
  26  *    may be used to endorse or promote products derived from this software
  27  *    without specific prior written permission.
  28  *
  29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  39  * SUCH DAMAGE.
  40  *
  41  *      @(#)init_main.c 8.9 (Berkeley) 1/21/94
  42  */
  43
  44 #include <sys/cdefs.h>
  45 __FBSDID("$FreeBSD$");
  46
  47 #include "opt_ddb.h"
  48 #include "opt_init_path.h"
  49
  50 #include <sys/param.h>
  51 #include <sys/kernel.h>
  52 #include <sys/exec.h>
  53 #include <sys/file.h>
  54 #include <sys/filedesc.h>
  55 #include <sys/jail.h>
  56 #include <sys/ktr.h>
  57 #include <sys/lock.h>
  58 #include <sys/mount.h>
  59 #include <sys/mutex.h>
  60 #include <sys/syscallsubr.h>
  61 #include <sys/sysctl.h>
  62 #include <sys/proc.h>
  63 #include <sys/resourcevar.h>
  64 #include <sys/systm.h>
  65 #include <sys/signalvar.h>
  66 #include <sys/vnode.h>
  67 #include <sys/sysent.h>
  68 #include <sys/reboot.h>
  69 #include <sys/sched.h>
  70 #include <sys/sx.h>
  71 #include <sys/sysproto.h>
  72 #include <sys/vmmeter.h>
  73 #include <sys/unistd.h>
  74 #include <sys/malloc.h>
  75 #include <sys/conf.h>
  76 #include <sys/cpuset.h>
  77
  78 #include <machine/cpu.h>
  79
  80 #include <security/audit/audit.h>
  81 #include <security/mac/mac_framework.h>
  82
  83 #include <vm/vm.h>
  84 #include <vm/vm_param.h>
  85 #include <vm/pmap.h>
  86 #include <vm/vm_map.h>
  87 #include <sys/copyright.h>
  88
  89 #include <ddb/ddb.h>
  90 #include <ddb/db_sym.h>
  91
  92 void mi_startup(void);                          /* Should be elsewhere */
  93
  94 /* Components of the first process -- never freed. */
  95 static struct session session0;
  96 static struct pgrp pgrp0;
  97 struct  proc proc0;
  98 struct  thread thread0 __aligned(16);
  99 struct  vmspace vmspace0;
 100 struct  proc *initproc;
 101
 102 int     boothowto = 0;          /* initialized so that it can be patched */
 103 SYSCTL_INT(_debug, OID_AUTO, boothowto, CTLFLAG_RD, &boothowto, 0, "");
 104 int     bootverbose;
 105 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, &bootverbose, 0, "");
 106
 107 /*
 108  * This ensures that there is at least one entry so that the sysinit_set
 109  * symbol is not undefined.  A sybsystem ID of SI_SUB_DUMMY is never
 110  * executed.
 111  */
 112 SYSINIT(placeholder, SI_SUB_DUMMY, SI_ORDER_ANY, NULL, NULL);
 113
 114 /*
 115  * The sysinit table itself.  Items are checked off as the are run.
 116  * If we want to register new sysinit types, add them to newsysinit.
 117  */
 118 SET_DECLARE(sysinit_set, struct sysinit);
 119 struct sysinit **sysinit, **sysinit_end;
 120 struct sysinit **newsysinit, **newsysinit_end;
 121
 122 /*
 123  * Merge a new sysinit set into the current set, reallocating it if
 124  * necessary.  This can only be called after malloc is running.
 125  */
 126 void
 127 sysinit_add(struct sysinit **set, struct sysinit **set_end)
 128 {
 129         struct sysinit **newset;
 130         struct sysinit **sipp;
 131         struct sysinit **xipp;
 132         int count;
 133
 134         count = set_end - set;
 135         if (newsysinit)
 136                 count += newsysinit_end - newsysinit;
 137         else
 138                 count += sysinit_end - sysinit;
 139         newset = malloc(count * sizeof(*sipp), M_TEMP, M_NOWAIT);
 140         if (newset == NULL)
 141                 panic("cannot malloc for sysinit");
 142         xipp = newset;
 143         if (newsysinit)
 144                 for (sipp = newsysinit; sipp < newsysinit_end; sipp++)
 145                         *xipp++ = *sipp;
 146         else
 147                 for (sipp = sysinit; sipp < sysinit_end; sipp++)
 148                         *xipp++ = *sipp;
 149         for (sipp = set; sipp < set_end; sipp++)
 150                 *xipp++ = *sipp;
 151         if (newsysinit)
 152                 free(newsysinit, M_TEMP);
 153         newsysinit = newset;
 154         newsysinit_end = newset + count;
 155 }
 156
 157 /*
 158  * System startup; initialize the world, create process 0, mount root
 159  * filesystem, and fork to create init and pagedaemon.  Most of the
 160  * hard work is done in the lower-level initialization routines including
 161  * startup(), which does memory initialization and autoconfiguration.
 162  *
 163  * This allows simple addition of new kernel subsystems that require
 164  * boot time initialization.  It also allows substitution of subsystem
 165  * (for instance, a scheduler, kernel profiler, or VM system) by object
 166  * module.  Finally, it allows for optional "kernel threads".
 167  */
 168 void
 169 mi_startup(void)
 170 {
 171
 172         register struct sysinit **sipp;         /* system initialization*/
 173         register struct sysinit **xipp;         /* interior loop of sort*/
 174         register struct sysinit *save;          /* bubble*/
 175
 176 #if defined(VERBOSE_SYSINIT)
 177         int last;
 178         int verbose;
 179 #endif
 180
 181         if (boothowto & RB_VERBOSE)
 182                 bootverbose++;
 183
 184         if (sysinit == NULL) {
 185                 sysinit = SET_BEGIN(sysinit_set);
 186                 sysinit_end = SET_LIMIT(sysinit_set);
 187         }
 188
 189 restart:
 190         /*
 191          * Perform a bubble sort of the system initialization objects by
 192          * their subsystem (primary key) and order (secondary key).
 193          */
 194         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 195                 for (xipp = sipp + 1; xipp < sysinit_end; xipp++) {
 196                         if ((*sipp)->subsystem < (*xipp)->subsystem ||
 197                              ((*sipp)->subsystem == (*xipp)->subsystem &&
 198                               (*sipp)->order <= (*xipp)->order))
 199                                 continue;       /* skip*/
 200                         save = *sipp;
 201                         *sipp = *xipp;
 202                         *xipp = save;
 203                 }
 204         }
 205
 206 #if defined(VERBOSE_SYSINIT)
 207         last = SI_SUB_COPYRIGHT;
 208         verbose = 0;
 209 #if !defined(DDB)
 210         printf("VERBOSE_SYSINIT: DDB not enabled, symbol lookups disabled.\n");
 211 #endif
 212 #endif
 213
 214         /*
 215          * Traverse the (now) ordered list of system initialization tasks.
 216          * Perform each task, and continue on to the next task.
 217          *
 218          * The last item on the list is expected to be the scheduler,
 219          * which will not return.
 220          */
 221         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 222
 223                 if ((*sipp)->subsystem == SI_SUB_DUMMY)
 224                         continue;       /* skip dummy task(s)*/
 225
 226                 if ((*sipp)->subsystem == SI_SUB_DONE)
 227                         continue;
 228
 229 #if defined(VERBOSE_SYSINIT)
 230                 if ((*sipp)->subsystem > last) {
 231                         verbose = 1;
 232                         last = (*sipp)->subsystem;
 233                         printf("subsystem %x\n", last);
 234                 }
 235                 if (verbose) {
 236 #if defined(DDB)
 237                         const char *name;
 238                         c_db_sym_t sym;
 239                         db_expr_t  offset;
 240
 241                         sym = db_search_symbol((vm_offset_t)(*sipp)->func,
 242                             DB_STGY_PROC, &offset);
 243                         db_symbol_values(sym, &name, NULL);
 244                         if (name != NULL)
 245                                 printf("   %s(%p)... ", name, (*sipp)->udata);
 246                         else
 247 #endif
 248                                 printf("   %p(%p)... ", (*sipp)->func,
 249                                     (*sipp)->udata);
 250                 }
 251 #endif
 252
 253                 /* Call function */
 254                 (*((*sipp)->func))((*sipp)->udata);
 255
 256 #if defined(VERBOSE_SYSINIT)
 257                 if (verbose)
 258                         printf("done.\n");
 259 #endif
 260
 261                 /* Check off the one we're just done */
 262                 (*sipp)->subsystem = SI_SUB_DONE;
 263
 264                 /* Check if we've installed more sysinit items via KLD */
 265                 if (newsysinit != NULL) {
 266                         if (sysinit != SET_BEGIN(sysinit_set))
 267                                 free(sysinit, M_TEMP);
 268                         sysinit = newsysinit;
 269                         sysinit_end = newsysinit_end;
 270                         newsysinit = NULL;
 271                         newsysinit_end = NULL;
 272                         goto restart;
 273                 }
 274         }
 275
 276         panic("Shouldn't get here!");
 277         /* NOTREACHED*/
 278 }
 279
 280
 281 /*
 282  ***************************************************************************
 283  ****
 284  **** The following SYSINIT's belong elsewhere, but have not yet
 285  **** been moved.
 286  ****
 287  ***************************************************************************
 288  */
 289 static void
 290 print_caddr_t(void *data)
 291 {
 292         printf("%s", (char *)data);
 293 }
 294
 295 static void
 296 print_version(void *data __unused)
 297 {
 298         int len;
 299
 300         /* Strip a trailing newline from version. */
 301         len = strlen(version);
 302         while (len > 0 && version[len - 1] == '\n')
 303                 len--;
 304         printf("%.*s %s\n", len, version, machine);
 305 }
 306
 307 SYSINIT(announce, SI_SUB_COPYRIGHT, SI_ORDER_FIRST, print_caddr_t,
 308     copyright);
 309 SYSINIT(trademark, SI_SUB_COPYRIGHT, SI_ORDER_SECOND, print_caddr_t,
 310     trademark);
 311 SYSINIT(version, SI_SUB_COPYRIGHT, SI_ORDER_THIRD, print_version, NULL);
 312
 313 #ifdef WITNESS
 314 static char wit_warn[] =
 315      "WARNING: WITNESS option enabled, expect reduced performance.\n";
 316 SYSINIT(witwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 1,
 317    print_caddr_t, wit_warn);
 318 SYSINIT(witwarn2, SI_SUB_RUN_SCHEDULER, SI_ORDER_THIRD + 1,
 319    print_caddr_t, wit_warn);
 320 #endif
 321
 322 #ifdef DIAGNOSTIC
 323 static char diag_warn[] =
 324      "WARNING: DIAGNOSTIC option enabled, expect reduced performance.\n";
 325 SYSINIT(diagwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 2,
 326     print_caddr_t, diag_warn);
 327 SYSINIT(diagwarn2, SI_SUB_RUN_SCHEDULER, SI_ORDER_THIRD + 2,
 328     print_caddr_t, diag_warn);
 329 #endif
 330
 331 static int
 332 null_fetch_syscall_args(struct thread *td __unused,
 333     struct syscall_args *sa __unused)
 334 {
 335
 336         panic("null_fetch_syscall_args");
 337 }
 338
 339 static void
 340 null_set_syscall_retval(struct thread *td __unused, int error __unused)
 341 {
 342
 343         panic("null_set_syscall_retval");
 344 }
 345
 346 struct sysentvec null_sysvec = {
 347         .sv_size        = 0,
 348         .sv_table       = NULL,
 349         .sv_mask        = 0,
 350         .sv_sigsize     = 0,
 351         .sv_sigtbl      = NULL,
 352         .sv_errsize     = 0,
 353         .sv_errtbl      = NULL,
 354         .sv_transtrap   = NULL,
 355         .sv_fixup       = NULL,
 356         .sv_sendsig     = NULL,
 357         .sv_sigcode     = NULL,
 358         .sv_szsigcode   = NULL,
 359         .sv_prepsyscall = NULL,
 360         .sv_name        = "null",
 361         .sv_coredump    = NULL,
 362         .sv_imgact_try  = NULL,
 363         .sv_minsigstksz = 0,
 364         .sv_pagesize    = PAGE_SIZE,
 365         .sv_minuser     = VM_MIN_ADDRESS,
 366         .sv_maxuser     = VM_MAXUSER_ADDRESS,
 367         .sv_usrstack    = USRSTACK,
 368         .sv_psstrings   = PS_STRINGS,
 369         .sv_stackprot   = VM_PROT_ALL,
 370         .sv_copyout_strings     = NULL,
 371         .sv_setregs     = NULL,
 372         .sv_fixlimit    = NULL,
 373         .sv_maxssiz     = NULL,
 374         .sv_flags       = 0,
 375         .sv_set_syscall_retval = null_set_syscall_retval,
 376         .sv_fetch_syscall_args = null_fetch_syscall_args,
 377         .sv_syscallnames = NULL,
 378 };
 379
 380 /*
 381  ***************************************************************************
 382  ****
 383  **** The two following SYSINIT's are proc0 specific glue code.  I am not
 384  **** convinced that they can not be safely combined, but their order of
 385  **** operation has been maintained as the same as the original init_main.c
 386  **** for right now.
 387  ****
 388  **** These probably belong in init_proc.c or kern_proc.c, since they
 389  **** deal with proc0 (the fork template process).
 390  ****
 391  ***************************************************************************
 392  */
 393 /* ARGSUSED*/
 394 static void
 395 proc0_init(void *dummy __unused)
 396 {
 397         struct proc *p;
 398         unsigned i;
 399         struct thread *td;
 400
 401         GIANT_REQUIRED;
 402         p = &proc0;
 403         td = &thread0;
 404
 405         /*
 406          * Initialize magic number and osrel.
 407          */
 408         p->p_magic = P_MAGIC;
 409         p->p_osrel = osreldate;
 410
 411         /*
 412          * Initialize thread and process structures.
 413          */
 414         procinit();     /* set up proc zone */
 415         threadinit();   /* set up UMA zones */
 416
 417         /*
 418          * Initialise scheduler resources.
 419          * Add scheduler specific parts to proc, thread as needed.
 420          */
 421         schedinit();    /* scheduler gets its house in order */
 422         /*
 423          * Initialize sleep queue hash table
 424          */
 425         sleepinit();
 426
 427         /*
 428          * additional VM structures
 429          */
 430         vm_init2();
 431
 432         /*
 433          * Create process 0 (the swapper).
 434          */
 435         LIST_INSERT_HEAD(&allproc, p, p_list);
 436         LIST_INSERT_HEAD(PIDHASH(0), p, p_hash);
 437         mtx_init(&pgrp0.pg_mtx, "process group", NULL, MTX_DEF | MTX_DUPOK);
 438         p->p_pgrp = &pgrp0;
 439         LIST_INSERT_HEAD(PGRPHASH(0), &pgrp0, pg_hash);
 440         LIST_INIT(&pgrp0.pg_members);
 441         LIST_INSERT_HEAD(&pgrp0.pg_members, p, p_pglist);
 442
 443         pgrp0.pg_session = &session0;
 444         mtx_init(&session0.s_mtx, "session", NULL, MTX_DEF);
 445         refcount_init(&session0.s_count, 1);
 446         session0.s_leader = p;
 447
 448         p->p_sysent = &null_sysvec;
 449         p->p_flag = P_SYSTEM | P_INMEM;
 450         p->p_state = PRS_NORMAL;
 451         knlist_init_mtx(&p->p_klist, &p->p_mtx);
 452         STAILQ_INIT(&p->p_ktr);
 453         p->p_nice = NZERO;
 454         td->td_tid = PID_MAX + 1;
 455         td->td_state = TDS_RUNNING;
 456         td->td_pri_class = PRI_TIMESHARE;
 457         td->td_user_pri = PUSER;
 458         td->td_base_user_pri = PUSER;
 459         td->td_priority = PVM;
 460         td->td_base_pri = PUSER;
 461         td->td_oncpu = 0;
 462         td->td_flags = TDF_INMEM|TDP_KTHREAD;
 463         td->td_cpuset = cpuset_thread0();
 464         prison0.pr_cpuset = cpuset_ref(td->td_cpuset);
 465         p->p_peers = 0;
 466         p->p_leader = p;
 467
 468
 469         strncpy(p->p_comm, "kernel", sizeof (p->p_comm));
 470         strncpy(td->td_name, "swapper", sizeof (td->td_name));
 471
 472         callout_init(&p->p_itcallout, CALLOUT_MPSAFE);
 473         callout_init_mtx(&p->p_limco, &p->p_mtx, 0);
 474         callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
 475
 476         /* Create credentials. */
 477         p->p_ucred = crget();
 478         p->p_ucred->cr_ngroups = 1;     /* group 0 */
 479         p->p_ucred->cr_uidinfo = uifind(0);
 480         p->p_ucred->cr_ruidinfo = uifind(0);
 481         p->p_ucred->cr_prison = &prison0;
 482 #ifdef AUDIT
 483         audit_cred_kproc0(p->p_ucred);
 484 #endif
 485 #ifdef MAC
 486         mac_cred_create_swapper(p->p_ucred);
 487 #endif
 488         td->td_ucred = crhold(p->p_ucred);
 489
 490         /* Create sigacts. */
 491         p->p_sigacts = sigacts_alloc();
 492
 493         /* Initialize signal state for process 0. */
 494         siginit(&proc0);
 495
 496         /* Create the file descriptor table. */
 497         p->p_fd = fdinit(NULL);
 498         p->p_fdtol = NULL;
 499
 500         /* Create the limits structures. */
 501         p->p_limit = lim_alloc();
 502         for (i = 0; i < RLIM_NLIMITS; i++)
 503                 p->p_limit->pl_rlimit[i].rlim_cur =
 504                     p->p_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
 505         p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur =
 506             p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
 507         p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_cur =
 508             p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
 509         i = ptoa(cnt.v_free_count);
 510         p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_max = i;
 511         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_max = i;
 512         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = i / 3;
 513         p->p_cpulimit = RLIM_INFINITY;
 514
 515         p->p_stats = pstats_alloc();
 516
 517         /* Allocate a prototype map so we have something to fork. */
 518         pmap_pinit0(vmspace_pmap(&vmspace0));
 519         p->p_vmspace = &vmspace0;
 520         vmspace0.vm_refcnt = 1;
 521
 522         /*
 523          * proc0 is not expected to enter usermode, so there is no special
 524          * handling for sv_minuser here, like is done for exec_new_vmspace().
 525          */
 526         vm_map_init(&vmspace0.vm_map, p->p_sysent->sv_minuser,
 527             p->p_sysent->sv_maxuser);
 528         vmspace0.vm_map.pmap = vmspace_pmap(&vmspace0);
 529
 530         /*-
 531          * call the init and ctor for the new thread and proc
 532          * we wait to do this until all other structures
 533          * are fairly sane.
 534          */
 535         EVENTHANDLER_INVOKE(process_init, p);
 536         EVENTHANDLER_INVOKE(thread_init, td);
 537         EVENTHANDLER_INVOKE(process_ctor, p);
 538         EVENTHANDLER_INVOKE(thread_ctor, td);
 539
 540         /*
 541          * Charge root for one process.
 542          */
 543         (void)chgproccnt(p->p_ucred->cr_ruidinfo, 1, 0);
 544 }
 545 SYSINIT(p0init, SI_SUB_INTRINSIC, SI_ORDER_FIRST, proc0_init, NULL);
 546
 547 /* ARGSUSED*/
 548 static void
 549 proc0_post(void *dummy __unused)
 550 {
 551         struct timespec ts;
 552         struct proc *p;
 553         struct rusage ru;
 554         struct thread *td;
 555
 556         /*
 557          * Now we can look at the time, having had a chance to verify the
 558          * time from the filesystem.  Pretend that proc0 started now.
 559          */
 560         sx_slock(&allproc_lock);
 561         FOREACH_PROC_IN_SYSTEM(p) {
 562                 microuptime(&p->p_stats->p_start);
 563                 PROC_SLOCK(p);
 564                 rufetch(p, &ru);        /* Clears thread stats */
 565                 PROC_SUNLOCK(p);
 566                 p->p_rux.rux_runtime = 0;
 567                 p->p_rux.rux_uticks = 0;
 568                 p->p_rux.rux_sticks = 0;
 569                 p->p_rux.rux_iticks = 0;
 570                 FOREACH_THREAD_IN_PROC(p, td) {
 571                         td->td_runtime = 0;
 572                 }
 573         }
 574         sx_sunlock(&allproc_lock);
 575         PCPU_SET(switchtime, cpu_ticks());
 576         PCPU_SET(switchticks, ticks);
 577
 578         /*
 579          * Give the ``random'' number generator a thump.
 580          */
 581         nanotime(&ts);
 582         srandom(ts.tv_sec ^ ts.tv_nsec);
 583 }
 584 SYSINIT(p0post, SI_SUB_INTRINSIC_POST, SI_ORDER_FIRST, proc0_post, NULL);
 585
 586 static void
 587 random_init(void *dummy __unused)
 588 {
 589
 590         /*
 591          * After CPU has been started we have some randomness on most
 592          * platforms via get_cyclecount().  For platforms that don't
 593          * we will reseed random(9) in proc0_post() as well.
 594          */
 595         srandom(get_cyclecount());
 596 }
 597 SYSINIT(random, SI_SUB_RANDOM, SI_ORDER_FIRST, random_init, NULL);
 598
 599 /*
 600  ***************************************************************************
 601  ****
 602  **** The following SYSINIT's and glue code should be moved to the
 603  **** respective files on a per subsystem basis.
 604  ****
 605  ***************************************************************************
 606  */
 607
 608
 609 /*
 610  ***************************************************************************
 611  ****
 612  **** The following code probably belongs in another file, like
 613  **** kern/init_init.c.
 614  ****
 615  ***************************************************************************
 616  */
 617
 618 /*
 619  * List of paths to try when searching for "init".
 620  */
 621 static char init_path[MAXPATHLEN] =
 622 #ifdef  INIT_PATH
 623     __XSTRING(INIT_PATH);
 624 #else
 625     "/sbin/init:/sbin/oinit:/sbin/init.bak:/rescue/init:/stand/sysinstall";
 626 #endif
 627 SYSCTL_STRING(_kern, OID_AUTO, init_path, CTLFLAG_RD, init_path, 0,
 628         "Path used to search the init process");
 629
 630 /*
 631  * Shutdown timeout of init(8).
 632  * Unused within kernel, but used to control init(8), hence do not remove.
 633  */
 634 #ifndef INIT_SHUTDOWN_TIMEOUT
 635 #define INIT_SHUTDOWN_TIMEOUT 120
 636 #endif
 637 static int init_shutdown_timeout = INIT_SHUTDOWN_TIMEOUT;
 638 SYSCTL_INT(_kern, OID_AUTO, init_shutdown_timeout,
 639         CTLFLAG_RW, &init_shutdown_timeout, 0, "");
 640
 641 /*
 642  * Start the initial user process; try exec'ing each pathname in init_path.
 643  * The program is invoked with one argument containing the boot flags.
 644  */
 645 static void
 646 start_init(void *dummy)
 647 {
 648         vm_offset_t addr;
 649         struct execve_args args;
 650         int options, error;
 651         char *var, *path, *next, *s;
 652         char *ucp, **uap, *arg0, *arg1;
 653         struct thread *td;
 654         struct proc *p;
 655
 656         mtx_lock(&Giant);
 657
 658         GIANT_REQUIRED;
 659
 660         td = curthread;
 661         p = td->td_proc;
 662
 663         vfs_mountroot();
 664
 665         /*
 666          * Need just enough stack to hold the faked-up "execve()" arguments.
 667          */
 668         addr = p->p_sysent->sv_usrstack - PAGE_SIZE;
 669         if (vm_map_find(&p->p_vmspace->vm_map, NULL, 0, &addr, PAGE_SIZE,
 670                         FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != 0)
 671                 panic("init: couldn't allocate argument space");
 672         p->p_vmspace->vm_maxsaddr = (caddr_t)addr;
 673         p->p_vmspace->vm_ssize = 1;
 674
 675         if ((var = getenv("init_path")) != NULL) {
 676                 strlcpy(init_path, var, sizeof(init_path));
 677                 freeenv(var);
 678         }
 679
 680         for (path = init_path; *path != '\0'; path = next) {
 681                 while (*path == ':')
 682                         path++;
 683                 if (*path == '\0')
 684                         break;
 685                 for (next = path; *next != '\0' && *next != ':'; next++)
 686                         /* nothing */ ;
 687                 if (bootverbose)
 688                         printf("start_init: trying %.*s\n", (int)(next - path),
 689                             path);
 690
 691                 /*
 692                  * Move out the boot flag argument.
 693                  */
 694                 options = 0;
 695                 ucp = (char *)p->p_sysent->sv_usrstack;
 696                 (void)subyte(--ucp, 0);         /* trailing zero */
 697                 if (boothowto & RB_SINGLE) {
 698                         (void)subyte(--ucp, 's');
 699                         options = 1;
 700                 }
 701 #ifdef notyet
 702                 if (boothowto & RB_FASTBOOT) {
 703                         (void)subyte(--ucp, 'f');
 704                         options = 1;
 705                 }
 706 #endif
 707
 708 #ifdef BOOTCDROM
 709                 (void)subyte(--ucp, 'C');
 710                 options = 1;
 711 #endif
 712
 713                 if (options == 0)
 714                         (void)subyte(--ucp, '-');
 715                 (void)subyte(--ucp, '-');               /* leading hyphen */
 716                 arg1 = ucp;
 717
 718                 /*
 719                  * Move out the file name (also arg 0).
 720                  */
 721                 (void)subyte(--ucp, 0);
 722                 for (s = next - 1; s >= path; s--)
 723                         (void)subyte(--ucp, *s);
 724                 arg0 = ucp;
 725
 726                 /*
 727                  * Move out the arg pointers.
 728                  */
 729                 uap = (char **)((intptr_t)ucp & ~(sizeof(intptr_t)-1));
 730                 (void)suword((caddr_t)--uap, (long)0);  /* terminator */
 731                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg1);
 732                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg0);
 733
 734                 /*
 735                  * Point at the arguments.
 736                  */
 737                 args.fname = arg0;
 738                 args.argv = uap;
 739                 args.envv = NULL;
 740
 741                 /*
 742                  * Now try to exec the program.  If can't for any reason
 743                  * other than it doesn't exist, complain.
 744                  *
 745                  * Otherwise, return via fork_trampoline() all the way
 746                  * to user mode as init!
 747                  */
 748                 if ((error = execve(td, &args)) == 0) {
 749                         mtx_unlock(&Giant);
 750                         return;
 751                 }
 752                 if (error != ENOENT)
 753                         printf("exec %.*s: error %d\n", (int)(next - path),
 754                             path, error);
 755         }
 756         printf("init: not found in path %s\n", init_path);
 757         panic("no init");
 758 }
 759
 760 /*
 761  * Like kproc_create(), but runs in it's own address space.
 762  * We do this early to reserve pid 1.
 763  *
 764  * Note special case - do not make it runnable yet.  Other work
 765  * in progress will change this more.
 766  */
 767 static void
 768 create_init(const void *udata __unused)
 769 {
 770         struct ucred *newcred, *oldcred;
 771         int error;
 772
 773         error = fork1(&thread0, RFFDG | RFPROC | RFSTOPPED, 0, &initproc);
 774         if (error)
 775                 panic("cannot fork init: %d\n", error);
 776         KASSERT(initproc->p_pid == 1, ("create_init: initproc->p_pid != 1"));
 777         /* divorce init's credentials from the kernel's */
 778         newcred = crget();
 779         PROC_LOCK(initproc);
 780         initproc->p_flag |= P_SYSTEM | P_INMEM;
 781         oldcred = initproc->p_ucred;
 782         crcopy(newcred, oldcred);
 783 #ifdef MAC
 784         mac_cred_create_init(newcred);
 785 #endif
 786 #ifdef AUDIT
 787         audit_cred_proc1(newcred);
 788 #endif
 789         initproc->p_ucred = newcred;
 790         PROC_UNLOCK(initproc);
 791         crfree(oldcred);
 792         cred_update_thread(FIRST_THREAD_IN_PROC(initproc));
 793         cpu_set_fork_handler(FIRST_THREAD_IN_PROC(initproc), start_init, NULL);
 794 }
 795 SYSINIT(init, SI_SUB_CREATE_INIT, SI_ORDER_FIRST, create_init, NULL);
 796
 797 /*
 798  * Make it runnable now.
 799  */
 800 static void
 801 kick_init(const void *udata __unused)
 802 {
 803         struct thread *td;
 804
 805         td = FIRST_THREAD_IN_PROC(initproc);
 806         thread_lock(td);
 807         TD_SET_CAN_RUN(td);
 808         sched_add(td, SRQ_BORING);
 809         thread_unlock(td);
 810 }
 811 SYSINIT(kickinit, SI_SUB_KTHREAD_INIT, SI_ORDER_FIRST, kick_init, NULL);