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 #include "opt_verbose_sysinit.h"
  50
  51 #include <sys/param.h>
  52 #include <sys/kernel.h>
  53 #include <sys/exec.h>
  54 #include <sys/file.h>
  55 #include <sys/filedesc.h>
  56 #include <sys/jail.h>
  57 #include <sys/ktr.h>
  58 #include <sys/lock.h>
  59 #include <sys/loginclass.h>
  60 #include <sys/mount.h>
  61 #include <sys/mutex.h>
  62 #include <sys/syscallsubr.h>
  63 #include <sys/sysctl.h>
  64 #include <sys/proc.h>
  65 #include <sys/racct.h>
  66 #include <sys/resourcevar.h>
  67 #include <sys/systm.h>
  68 #include <sys/signalvar.h>
  69 #include <sys/vnode.h>
  70 #include <sys/sysent.h>
  71 #include <sys/reboot.h>
  72 #include <sys/sched.h>
  73 #include <sys/sx.h>
  74 #include <sys/sysproto.h>
  75 #include <sys/vmmeter.h>
  76 #include <sys/unistd.h>
  77 #include <sys/malloc.h>
  78 #include <sys/conf.h>
  79 #include <sys/cpuset.h>
  80
  81 #include <machine/cpu.h>
  82
  83 #include <security/audit/audit.h>
  84 #include <security/mac/mac_framework.h>
  85
  86 #include <vm/vm.h>
  87 #include <vm/vm_param.h>
  88 #include <vm/pmap.h>
  89 #include <vm/vm_map.h>
  90 #include <vm/vm_domain.h>
  91 #include <sys/copyright.h>
  92
  93 #include <ddb/ddb.h>
  94 #include <ddb/db_sym.h>
  95
  96 void mi_startup(void);                          /* Should be elsewhere */
  97
  98 /* Components of the first process -- never freed. */
  99 static struct session session0;
 100 static struct pgrp pgrp0;
 101 struct  proc proc0;
 102 struct thread0_storage thread0_st __aligned(32);
 103 struct  vmspace vmspace0;
 104 struct  proc *initproc;
 105
 106 #ifndef BOOTHOWTO
 107 #define BOOTHOWTO       0
 108 #endif
 109 int     boothowto = BOOTHOWTO;  /* initialized so that it can be patched */
 110 SYSCTL_INT(_debug, OID_AUTO, boothowto, CTLFLAG_RD, &boothowto, 0,
 111         "Boot control flags, passed from loader");
 112
 113 #ifndef BOOTVERBOSE
 114 #define BOOTVERBOSE     0
 115 #endif
 116 int     bootverbose = BOOTVERBOSE;
 117 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, &bootverbose, 0,
 118         "Control the output of verbose kernel messages");
 119
 120 #ifdef VERBOSE_SYSINIT
 121 /*
 122  * We'll use the defined value of VERBOSE_SYSINIT from the kernel config to
 123  * dictate the default VERBOSE_SYSINIT behavior.  Significant values for this
 124  * option and associated tunable are:
 125  * - 0, 'compiled in but silent by default'
 126  * - 1, 'compiled in but verbose by default' (default)
 127  */
 128 int     verbose_sysinit = VERBOSE_SYSINIT;
 129 TUNABLE_INT("debug.verbose_sysinit", &verbose_sysinit);
 130 #endif
 131
 132 #ifdef INVARIANTS
 133 FEATURE(invariants, "Kernel compiled with INVARIANTS, may affect performance");
 134 #endif
 135
 136 /*
 137  * This ensures that there is at least one entry so that the sysinit_set
 138  * symbol is not undefined.  A sybsystem ID of SI_SUB_DUMMY is never
 139  * executed.
 140  */
 141 SYSINIT(placeholder, SI_SUB_DUMMY, SI_ORDER_ANY, NULL, NULL);
 142
 143 /*
 144  * The sysinit table itself.  Items are checked off as the are run.
 145  * If we want to register new sysinit types, add them to newsysinit.
 146  */
 147 SET_DECLARE(sysinit_set, struct sysinit);
 148 struct sysinit **sysinit, **sysinit_end;
 149 struct sysinit **newsysinit, **newsysinit_end;
 150
 151 EVENTHANDLER_LIST_DECLARE(process_init);
 152 EVENTHANDLER_LIST_DECLARE(thread_init);
 153 EVENTHANDLER_LIST_DECLARE(process_ctor);
 154 EVENTHANDLER_LIST_DECLARE(thread_ctor);
 155
 156 /*
 157  * Merge a new sysinit set into the current set, reallocating it if
 158  * necessary.  This can only be called after malloc is running.
 159  */
 160 void
 161 sysinit_add(struct sysinit **set, struct sysinit **set_end)
 162 {
 163         struct sysinit **newset;
 164         struct sysinit **sipp;
 165         struct sysinit **xipp;
 166         int count;
 167
 168         count = set_end - set;
 169         if (newsysinit)
 170                 count += newsysinit_end - newsysinit;
 171         else
 172                 count += sysinit_end - sysinit;
 173         newset = malloc(count * sizeof(*sipp), M_TEMP, M_NOWAIT);
 174         if (newset == NULL)
 175                 panic("cannot malloc for sysinit");
 176         xipp = newset;
 177         if (newsysinit)
 178                 for (sipp = newsysinit; sipp < newsysinit_end; sipp++)
 179                         *xipp++ = *sipp;
 180         else
 181                 for (sipp = sysinit; sipp < sysinit_end; sipp++)
 182                         *xipp++ = *sipp;
 183         for (sipp = set; sipp < set_end; sipp++)
 184                 *xipp++ = *sipp;
 185         if (newsysinit)
 186                 free(newsysinit, M_TEMP);
 187         newsysinit = newset;
 188         newsysinit_end = newset + count;
 189 }
 190
 191 #if defined (DDB) && defined(VERBOSE_SYSINIT)
 192 static const char *
 193 symbol_name(vm_offset_t va, db_strategy_t strategy)
 194 {
 195         const char *name;
 196         c_db_sym_t sym;
 197         db_expr_t  offset;
 198
 199         if (va == 0)
 200                 return (NULL);
 201         sym = db_search_symbol(va, strategy, &offset);
 202         if (offset != 0)
 203                 return (NULL);
 204         db_symbol_values(sym, &name, NULL);
 205         return (name);
 206 }
 207 #endif
 208
 209 /*
 210  * System startup; initialize the world, create process 0, mount root
 211  * filesystem, and fork to create init and pagedaemon.  Most of the
 212  * hard work is done in the lower-level initialization routines including
 213  * startup(), which does memory initialization and autoconfiguration.
 214  *
 215  * This allows simple addition of new kernel subsystems that require
 216  * boot time initialization.  It also allows substitution of subsystem
 217  * (for instance, a scheduler, kernel profiler, or VM system) by object
 218  * module.  Finally, it allows for optional "kernel threads".
 219  */
 220 void
 221 mi_startup(void)
 222 {
 223
 224         register struct sysinit **sipp;         /* system initialization*/
 225         register struct sysinit **xipp;         /* interior loop of sort*/
 226         register struct sysinit *save;          /* bubble*/
 227
 228 #if defined(VERBOSE_SYSINIT)
 229         int last;
 230         int verbose;
 231 #endif
 232
 233         if (boothowto & RB_VERBOSE)
 234                 bootverbose++;
 235
 236         if (sysinit == NULL) {
 237                 sysinit = SET_BEGIN(sysinit_set);
 238                 sysinit_end = SET_LIMIT(sysinit_set);
 239         }
 240
 241 restart:
 242         /*
 243          * Perform a bubble sort of the system initialization objects by
 244          * their subsystem (primary key) and order (secondary key).
 245          */
 246         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 247                 for (xipp = sipp + 1; xipp < sysinit_end; xipp++) {
 248                         if ((*sipp)->subsystem < (*xipp)->subsystem ||
 249                              ((*sipp)->subsystem == (*xipp)->subsystem &&
 250                               (*sipp)->order <= (*xipp)->order))
 251                                 continue;       /* skip*/
 252                         save = *sipp;
 253                         *sipp = *xipp;
 254                         *xipp = save;
 255                 }
 256         }
 257
 258 #if defined(VERBOSE_SYSINIT)
 259         last = SI_SUB_COPYRIGHT;
 260         verbose = 0;
 261 #if !defined(DDB)
 262         printf("VERBOSE_SYSINIT: DDB not enabled, symbol lookups disabled.\n");
 263 #endif
 264 #endif
 265
 266         /*
 267          * Traverse the (now) ordered list of system initialization tasks.
 268          * Perform each task, and continue on to the next task.
 269          */
 270         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
 271
 272                 if ((*sipp)->subsystem == SI_SUB_DUMMY)
 273                         continue;       /* skip dummy task(s)*/
 274
 275                 if ((*sipp)->subsystem == SI_SUB_DONE)
 276                         continue;
 277
 278 #if defined(VERBOSE_SYSINIT)
 279                 if ((*sipp)->subsystem > last && verbose_sysinit != 0) {
 280                         verbose = 1;
 281                         last = (*sipp)->subsystem;
 282                         printf("subsystem %x\n", last);
 283                 }
 284                 if (verbose) {
 285 #if defined(DDB)
 286                         const char *func, *data;
 287
 288                         func = symbol_name((vm_offset_t)(*sipp)->func,
 289                             DB_STGY_PROC);
 290                         data = symbol_name((vm_offset_t)(*sipp)->udata,
 291                             DB_STGY_ANY);
 292                         if (func != NULL && data != NULL)
 293                                 printf("   %s(&%s)... ", func, data);
 294                         else if (func != NULL)
 295                                 printf("   %s(%p)... ", func, (*sipp)->udata);
 296                         else
 297 #endif
 298                                 printf("   %p(%p)... ", (*sipp)->func,
 299                                     (*sipp)->udata);
 300                 }
 301 #endif
 302
 303                 /* Call function */
 304                 (*((*sipp)->func))((*sipp)->udata);
 305
 306 #if defined(VERBOSE_SYSINIT)
 307                 if (verbose)
 308                         printf("done.\n");
 309 #endif
 310
 311                 /* Check off the one we're just done */
 312                 (*sipp)->subsystem = SI_SUB_DONE;
 313
 314                 /* Check if we've installed more sysinit items via KLD */
 315                 if (newsysinit != NULL) {
 316                         if (sysinit != SET_BEGIN(sysinit_set))
 317                                 free(sysinit, M_TEMP);
 318                         sysinit = newsysinit;
 319                         sysinit_end = newsysinit_end;
 320                         newsysinit = NULL;
 321                         newsysinit_end = NULL;
 322                         goto restart;
 323                 }
 324         }
 325
 326         mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED);
 327         mtx_unlock(&Giant);
 328
 329         /*
 330          * Now hand over this thread to swapper.
 331          */
 332         swapper();
 333         /* NOTREACHED*/
 334 }
 335
 336
 337 /*
 338  ***************************************************************************
 339  ****
 340  **** The following SYSINIT's belong elsewhere, but have not yet
 341  **** been moved.
 342  ****
 343  ***************************************************************************
 344  */
 345 static void
 346 print_caddr_t(void *data)
 347 {
 348         printf("%s", (char *)data);
 349 }
 350
 351 static void
 352 print_version(void *data __unused)
 353 {
 354         int len;
 355
 356         /* Strip a trailing newline from version. */
 357         len = strlen(version);
 358         while (len > 0 && version[len - 1] == '\n')
 359                 len--;
 360         printf("%.*s %s\n", len, version, machine);
 361         printf("%s\n", compiler_version);
 362 }
 363
 364 SYSINIT(announce, SI_SUB_COPYRIGHT, SI_ORDER_FIRST, print_caddr_t,
 365     copyright);
 366 SYSINIT(trademark, SI_SUB_COPYRIGHT, SI_ORDER_SECOND, print_caddr_t,
 367     trademark);
 368 SYSINIT(version, SI_SUB_COPYRIGHT, SI_ORDER_THIRD, print_version, NULL);
 369
 370 #ifdef WITNESS
 371 static char wit_warn[] =
 372      "WARNING: WITNESS option enabled, expect reduced performance.\n";
 373 SYSINIT(witwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 1,
 374    print_caddr_t, wit_warn);
 375 SYSINIT(witwarn2, SI_SUB_LAST, SI_ORDER_THIRD + 1,
 376    print_caddr_t, wit_warn);
 377 #endif
 378
 379 #ifdef DIAGNOSTIC
 380 static char diag_warn[] =
 381      "WARNING: DIAGNOSTIC option enabled, expect reduced performance.\n";
 382 SYSINIT(diagwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 2,
 383     print_caddr_t, diag_warn);
 384 SYSINIT(diagwarn2, SI_SUB_LAST, SI_ORDER_THIRD + 2,
 385     print_caddr_t, diag_warn);
 386 #endif
 387
 388 static int
 389 null_fetch_syscall_args(struct thread *td __unused)
 390 {
 391
 392         panic("null_fetch_syscall_args");
 393 }
 394
 395 static void
 396 null_set_syscall_retval(struct thread *td __unused, int error __unused)
 397 {
 398
 399         panic("null_set_syscall_retval");
 400 }
 401
 402 struct sysentvec null_sysvec = {
 403         .sv_size        = 0,
 404         .sv_table       = NULL,
 405         .sv_mask        = 0,
 406         .sv_errsize     = 0,
 407         .sv_errtbl      = NULL,
 408         .sv_transtrap   = NULL,
 409         .sv_fixup       = NULL,
 410         .sv_sendsig     = NULL,
 411         .sv_sigcode     = NULL,
 412         .sv_szsigcode   = NULL,
 413         .sv_name        = "null",
 414         .sv_coredump    = NULL,
 415         .sv_imgact_try  = NULL,
 416         .sv_minsigstksz = 0,
 417         .sv_pagesize    = PAGE_SIZE,
 418         .sv_minuser     = VM_MIN_ADDRESS,
 419         .sv_maxuser     = VM_MAXUSER_ADDRESS,
 420         .sv_usrstack    = USRSTACK,
 421         .sv_psstrings   = PS_STRINGS,
 422         .sv_stackprot   = VM_PROT_ALL,
 423         .sv_copyout_strings     = NULL,
 424         .sv_setregs     = NULL,
 425         .sv_fixlimit    = NULL,
 426         .sv_maxssiz     = NULL,
 427         .sv_flags       = 0,
 428         .sv_set_syscall_retval = null_set_syscall_retval,
 429         .sv_fetch_syscall_args = null_fetch_syscall_args,
 430         .sv_syscallnames = NULL,
 431         .sv_schedtail   = NULL,
 432         .sv_thread_detach = NULL,
 433         .sv_trap        = NULL,
 434 };
 435
 436 /*
 437  ***************************************************************************
 438  ****
 439  **** The two following SYSINIT's are proc0 specific glue code.  I am not
 440  **** convinced that they can not be safely combined, but their order of
 441  **** operation has been maintained as the same as the original init_main.c
 442  **** for right now.
 443  ****
 444  **** These probably belong in init_proc.c or kern_proc.c, since they
 445  **** deal with proc0 (the fork template process).
 446  ****
 447  ***************************************************************************
 448  */
 449 /* ARGSUSED*/
 450 static void
 451 proc0_init(void *dummy __unused)
 452 {
 453         struct proc *p;
 454         struct thread *td;
 455         struct ucred *newcred;
 456         vm_paddr_t pageablemem;
 457         int i;
 458
 459         GIANT_REQUIRED;
 460         p = &proc0;
 461         td = &thread0;
 462
 463         /*
 464          * Initialize magic number and osrel.
 465          */
 466         p->p_magic = P_MAGIC;
 467         p->p_osrel = osreldate;
 468
 469         /*
 470          * Initialize thread and process structures.
 471          */
 472         procinit();     /* set up proc zone */
 473         threadinit();   /* set up UMA zones */
 474
 475         /*
 476          * Initialise scheduler resources.
 477          * Add scheduler specific parts to proc, thread as needed.
 478          */
 479         schedinit();    /* scheduler gets its house in order */
 480
 481         /*
 482          * Create process 0 (the swapper).
 483          */
 484         LIST_INSERT_HEAD(&allproc, p, p_list);
 485         LIST_INSERT_HEAD(PIDHASH(0), p, p_hash);
 486         mtx_init(&pgrp0.pg_mtx, "process group", NULL, MTX_DEF | MTX_DUPOK);
 487         p->p_pgrp = &pgrp0;
 488         LIST_INSERT_HEAD(PGRPHASH(0), &pgrp0, pg_hash);
 489         LIST_INIT(&pgrp0.pg_members);
 490         LIST_INSERT_HEAD(&pgrp0.pg_members, p, p_pglist);
 491
 492         pgrp0.pg_session = &session0;
 493         mtx_init(&session0.s_mtx, "session", NULL, MTX_DEF);
 494         refcount_init(&session0.s_count, 1);
 495         session0.s_leader = p;
 496
 497         p->p_sysent = &null_sysvec;
 498         p->p_flag = P_SYSTEM | P_INMEM | P_KPROC;
 499         p->p_flag2 = 0;
 500         p->p_state = PRS_NORMAL;
 501         p->p_klist = knlist_alloc(&p->p_mtx);
 502         STAILQ_INIT(&p->p_ktr);
 503         p->p_nice = NZERO;
 504         /* pid_max cannot be greater than PID_MAX */
 505         td->td_tid = PID_MAX + 1;
 506         LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
 507         td->td_state = TDS_RUNNING;
 508         td->td_pri_class = PRI_TIMESHARE;
 509         td->td_user_pri = PUSER;
 510         td->td_base_user_pri = PUSER;
 511         td->td_lend_user_pri = PRI_MAX;
 512         td->td_priority = PVM;
 513         td->td_base_pri = PVM;
 514         td->td_oncpu = 0;
 515         td->td_flags = TDF_INMEM;
 516         td->td_pflags = TDP_KTHREAD;
 517         td->td_cpuset = cpuset_thread0();
 518         vm_domain_policy_init(&td->td_vm_dom_policy);
 519         vm_domain_policy_set(&td->td_vm_dom_policy, VM_POLICY_NONE, -1);
 520         vm_domain_policy_init(&p->p_vm_dom_policy);
 521         vm_domain_policy_set(&p->p_vm_dom_policy, VM_POLICY_NONE, -1);
 522         prison0_init();
 523         p->p_peers = 0;
 524         p->p_leader = p;
 525         p->p_reaper = p;
 526         LIST_INIT(&p->p_reaplist);
 527
 528         strncpy(p->p_comm, "kernel", sizeof (p->p_comm));
 529         strncpy(td->td_name, "swapper", sizeof (td->td_name));
 530
 531         callout_init_mtx(&p->p_itcallout, &p->p_mtx, 0);
 532         callout_init_mtx(&p->p_limco, &p->p_mtx, 0);
 533         callout_init(&td->td_slpcallout, 1);
 534
 535         /* Create credentials. */
 536         newcred = crget();
 537         newcred->cr_ngroups = 1;        /* group 0 */
 538         newcred->cr_uidinfo = uifind(0);
 539         newcred->cr_ruidinfo = uifind(0);
 540         newcred->cr_prison = &prison0;
 541         newcred->cr_loginclass = loginclass_find("default");
 542         proc_set_cred_init(p, newcred);
 543 #ifdef AUDIT
 544         audit_cred_kproc0(newcred);
 545 #endif
 546 #ifdef MAC
 547         mac_cred_create_swapper(newcred);
 548 #endif
 549         /* Create sigacts. */
 550         p->p_sigacts = sigacts_alloc();
 551
 552         /* Initialize signal state for process 0. */
 553         siginit(&proc0);
 554
 555         /* Create the file descriptor table. */
 556         p->p_fd = fdinit(NULL, false);
 557         p->p_fdtol = NULL;
 558
 559         /* Create the limits structures. */
 560         p->p_limit = lim_alloc();
 561         for (i = 0; i < RLIM_NLIMITS; i++)
 562                 p->p_limit->pl_rlimit[i].rlim_cur =
 563                     p->p_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
 564         p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur =
 565             p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
 566         p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_cur =
 567             p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
 568         p->p_limit->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
 569         p->p_limit->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
 570         p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
 571         p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
 572         /* Cast to avoid overflow on i386/PAE. */
 573         pageablemem = ptoa((vm_paddr_t)vm_cnt.v_free_count);
 574         p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_cur =
 575             p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_max = pageablemem;
 576         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = pageablemem / 3;
 577         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_max = pageablemem;
 578         p->p_cpulimit = RLIM_INFINITY;
 579
 580         PROC_LOCK(p);
 581         thread_cow_get_proc(td, p);
 582         PROC_UNLOCK(p);
 583
 584         /* Initialize resource accounting structures. */
 585         racct_create(&p->p_racct);
 586
 587         p->p_stats = pstats_alloc();
 588
 589         /* Allocate a prototype map so we have something to fork. */
 590         p->p_vmspace = &vmspace0;
 591         vmspace0.vm_refcnt = 1;
 592         pmap_pinit0(vmspace_pmap(&vmspace0));
 593
 594         /*
 595          * proc0 is not expected to enter usermode, so there is no special
 596          * handling for sv_minuser here, like is done for exec_new_vmspace().
 597          */
 598         vm_map_init(&vmspace0.vm_map, vmspace_pmap(&vmspace0),
 599             p->p_sysent->sv_minuser, p->p_sysent->sv_maxuser);
 600
 601         /*
 602          * Call the init and ctor for the new thread and proc.  We wait
 603          * to do this until all other structures are fairly sane.
 604          */
 605         EVENTHANDLER_DIRECT_INVOKE(process_init, p);
 606         EVENTHANDLER_DIRECT_INVOKE(thread_init, td);
 607         EVENTHANDLER_DIRECT_INVOKE(process_ctor, p);
 608         EVENTHANDLER_DIRECT_INVOKE(thread_ctor, td);
 609
 610         /*
 611          * Charge root for one process.
 612          */
 613         (void)chgproccnt(p->p_ucred->cr_ruidinfo, 1, 0);
 614         PROC_LOCK(p);
 615         racct_add_force(p, RACCT_NPROC, 1);
 616         PROC_UNLOCK(p);
 617 }
 618 SYSINIT(p0init, SI_SUB_INTRINSIC, SI_ORDER_FIRST, proc0_init, NULL);
 619
 620 /* ARGSUSED*/
 621 static void
 622 proc0_post(void *dummy __unused)
 623 {
 624         struct timespec ts;
 625         struct proc *p;
 626         struct rusage ru;
 627         struct thread *td;
 628
 629         /*
 630          * Now we can look at the time, having had a chance to verify the
 631          * time from the filesystem.  Pretend that proc0 started now.
 632          */
 633         sx_slock(&allproc_lock);
 634         FOREACH_PROC_IN_SYSTEM(p) {
 635                 PROC_LOCK(p);
 636                 if (p->p_state == PRS_NEW) {
 637                         PROC_UNLOCK(p);
 638                         continue;
 639                 }
 640                 microuptime(&p->p_stats->p_start);
 641                 PROC_STATLOCK(p);
 642                 rufetch(p, &ru);        /* Clears thread stats */
 643                 p->p_rux.rux_runtime = 0;
 644                 p->p_rux.rux_uticks = 0;
 645                 p->p_rux.rux_sticks = 0;
 646                 p->p_rux.rux_iticks = 0;
 647                 PROC_STATUNLOCK(p);
 648                 FOREACH_THREAD_IN_PROC(p, td) {
 649                         td->td_runtime = 0;
 650                 }
 651                 PROC_UNLOCK(p);
 652         }
 653         sx_sunlock(&allproc_lock);
 654         PCPU_SET(switchtime, cpu_ticks());
 655         PCPU_SET(switchticks, ticks);
 656
 657         /*
 658          * Give the ``random'' number generator a thump.
 659          */
 660         nanotime(&ts);
 661         srandom(ts.tv_sec ^ ts.tv_nsec);
 662 }
 663 SYSINIT(p0post, SI_SUB_INTRINSIC_POST, SI_ORDER_FIRST, proc0_post, NULL);
 664
 665 static void
 666 random_init(void *dummy __unused)
 667 {
 668
 669         /*
 670          * After CPU has been started we have some randomness on most
 671          * platforms via get_cyclecount().  For platforms that don't
 672          * we will reseed random(9) in proc0_post() as well.
 673          */
 674         srandom(get_cyclecount());
 675 }
 676 SYSINIT(random, SI_SUB_RANDOM, SI_ORDER_FIRST, random_init, NULL);
 677
 678 /*
 679  ***************************************************************************
 680  ****
 681  **** The following SYSINIT's and glue code should be moved to the
 682  **** respective files on a per subsystem basis.
 683  ****
 684  ***************************************************************************
 685  */
 686
 687
 688 /*
 689  ***************************************************************************
 690  ****
 691  **** The following code probably belongs in another file, like
 692  **** kern/init_init.c.
 693  ****
 694  ***************************************************************************
 695  */
 696
 697 /*
 698  * List of paths to try when searching for "init".
 699  */
 700 static char init_path[MAXPATHLEN] =
 701 #ifdef  INIT_PATH
 702     __XSTRING(INIT_PATH);
 703 #else
 704     "/sbin/init:/sbin/oinit:/sbin/init.bak:/rescue/init";
 705 #endif
 706 SYSCTL_STRING(_kern, OID_AUTO, init_path, CTLFLAG_RD, init_path, 0,
 707         "Path used to search the init process");
 708
 709 /*
 710  * Shutdown timeout of init(8).
 711  * Unused within kernel, but used to control init(8), hence do not remove.
 712  */
 713 #ifndef INIT_SHUTDOWN_TIMEOUT
 714 #define INIT_SHUTDOWN_TIMEOUT 120
 715 #endif
 716 static int init_shutdown_timeout = INIT_SHUTDOWN_TIMEOUT;
 717 SYSCTL_INT(_kern, OID_AUTO, init_shutdown_timeout,
 718         CTLFLAG_RW, &init_shutdown_timeout, 0, "Shutdown timeout of init(8). "
 719         "Unused within kernel, but used to control init(8)");
 720
 721 /*
 722  * Start the initial user process; try exec'ing each pathname in init_path.
 723  * The program is invoked with one argument containing the boot flags.
 724  */
 725 static void
 726 start_init(void *dummy)
 727 {
 728         vm_offset_t addr;
 729         struct execve_args args;
 730         int options, error;
 731         char *var, *path, *next, *s;
 732         char *ucp, **uap, *arg0, *arg1;
 733         struct thread *td;
 734         struct proc *p;
 735
 736         mtx_lock(&Giant);
 737
 738         GIANT_REQUIRED;
 739
 740         td = curthread;
 741         p = td->td_proc;
 742
 743         vfs_mountroot();
 744
 745         /* Wipe GELI passphrase from the environment. */
 746         kern_unsetenv("kern.geom.eli.passphrase");
 747
 748         /*
 749          * Need just enough stack to hold the faked-up "execve()" arguments.
 750          */
 751         addr = p->p_sysent->sv_usrstack - PAGE_SIZE;
 752         if (vm_map_find(&p->p_vmspace->vm_map, NULL, 0, &addr, PAGE_SIZE, 0,
 753             VMFS_NO_SPACE, VM_PROT_ALL, VM_PROT_ALL, 0) != 0)
 754                 panic("init: couldn't allocate argument space");
 755         p->p_vmspace->vm_maxsaddr = (caddr_t)addr;
 756         p->p_vmspace->vm_ssize = 1;
 757
 758         if ((var = kern_getenv("init_path")) != NULL) {
 759                 strlcpy(init_path, var, sizeof(init_path));
 760                 freeenv(var);
 761         }
 762
 763         for (path = init_path; *path != '\0'; path = next) {
 764                 while (*path == ':')
 765                         path++;
 766                 if (*path == '\0')
 767                         break;
 768                 for (next = path; *next != '\0' && *next != ':'; next++)
 769                         /* nothing */ ;
 770                 if (bootverbose)
 771                         printf("start_init: trying %.*s\n", (int)(next - path),
 772                             path);
 773
 774                 /*
 775                  * Move out the boot flag argument.
 776                  */
 777                 options = 0;
 778                 ucp = (char *)p->p_sysent->sv_usrstack;
 779                 (void)subyte(--ucp, 0);         /* trailing zero */
 780                 if (boothowto & RB_SINGLE) {
 781                         (void)subyte(--ucp, 's');
 782                         options = 1;
 783                 }
 784 #ifdef notyet
 785                 if (boothowto & RB_FASTBOOT) {
 786                         (void)subyte(--ucp, 'f');
 787                         options = 1;
 788                 }
 789 #endif
 790
 791 #ifdef BOOTCDROM
 792                 (void)subyte(--ucp, 'C');
 793                 options = 1;
 794 #endif
 795
 796                 if (options == 0)
 797                         (void)subyte(--ucp, '-');
 798                 (void)subyte(--ucp, '-');               /* leading hyphen */
 799                 arg1 = ucp;
 800
 801                 /*
 802                  * Move out the file name (also arg 0).
 803                  */
 804                 (void)subyte(--ucp, 0);
 805                 for (s = next - 1; s >= path; s--)
 806                         (void)subyte(--ucp, *s);
 807                 arg0 = ucp;
 808
 809                 /*
 810                  * Move out the arg pointers.
 811                  */
 812                 uap = (char **)rounddown2((intptr_t)ucp, sizeof(intptr_t));
 813                 (void)suword((caddr_t)--uap, (long)0);  /* terminator */
 814                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg1);
 815                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg0);
 816
 817                 /*
 818                  * Point at the arguments.
 819                  */
 820                 args.fname = arg0;
 821                 args.argv = uap;
 822                 args.envv = NULL;
 823
 824                 /*
 825                  * Now try to exec the program.  If can't for any reason
 826                  * other than it doesn't exist, complain.
 827                  *
 828                  * Otherwise, return via fork_trampoline() all the way
 829                  * to user mode as init!
 830                  */
 831                 if ((error = sys_execve(td, &args)) == 0) {
 832                         mtx_unlock(&Giant);
 833                         return;
 834                 }
 835                 if (error != ENOENT)
 836                         printf("exec %.*s: error %d\n", (int)(next - path),
 837                             path, error);
 838         }
 839         printf("init: not found in path %s\n", init_path);
 840         panic("no init");
 841 }
 842
 843 /*
 844  * Like kproc_create(), but runs in it's own address space.
 845  * We do this early to reserve pid 1.
 846  *
 847  * Note special case - do not make it runnable yet.  Other work
 848  * in progress will change this more.
 849  */
 850 static void
 851 create_init(const void *udata __unused)
 852 {
 853         struct fork_req fr;
 854         struct ucred *newcred, *oldcred;
 855         struct thread *td;
 856         int error;
 857
 858         bzero(&fr, sizeof(fr));
 859         fr.fr_flags = RFFDG | RFPROC | RFSTOPPED;
 860         fr.fr_procp = &initproc;
 861         error = fork1(&thread0, &fr);
 862         if (error)
 863                 panic("cannot fork init: %d\n", error);
 864         KASSERT(initproc->p_pid == 1, ("create_init: initproc->p_pid != 1"));
 865         /* divorce init's credentials from the kernel's */
 866         newcred = crget();
 867         sx_xlock(&proctree_lock);
 868         PROC_LOCK(initproc);
 869         initproc->p_flag |= P_SYSTEM | P_INMEM;
 870         initproc->p_treeflag |= P_TREE_REAPER;
 871         LIST_INSERT_HEAD(&initproc->p_reaplist, &proc0, p_reapsibling);
 872         oldcred = initproc->p_ucred;
 873         crcopy(newcred, oldcred);
 874 #ifdef MAC
 875         mac_cred_create_init(newcred);
 876 #endif
 877 #ifdef AUDIT
 878         audit_cred_proc1(newcred);
 879 #endif
 880         proc_set_cred(initproc, newcred);
 881         td = FIRST_THREAD_IN_PROC(initproc);
 882         crfree(td->td_ucred);
 883         td->td_ucred = crhold(initproc->p_ucred);
 884         PROC_UNLOCK(initproc);
 885         sx_xunlock(&proctree_lock);
 886         crfree(oldcred);
 887         cpu_fork_kthread_handler(FIRST_THREAD_IN_PROC(initproc),
 888             start_init, NULL);
 889 }
 890 SYSINIT(init, SI_SUB_CREATE_INIT, SI_ORDER_FIRST, create_init, NULL);
 891
 892 /*
 893  * Make it runnable now.
 894  */
 895 static void
 896 kick_init(const void *udata __unused)
 897 {
 898         struct thread *td;
 899
 900         td = FIRST_THREAD_IN_PROC(initproc);
 901         thread_lock(td);
 902         TD_SET_CAN_RUN(td);
 903         sched_add(td, SRQ_BORING);
 904         thread_unlock(td);
 905 }
 906 SYSINIT(kickinit, SI_SUB_KTHREAD_INIT, SI_ORDER_MIDDLE, kick_init, NULL);