]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/kern/init_main.c
Sync with upstream version (20130520) that includes same fixes made last week.
[FreeBSD/FreeBSD.git] / sys / kern / init_main.c
1 /*-
2  * Copyright (c) 1995 Terrence R. Lambert
3  * All rights reserved.
4  *
5  * Copyright (c) 1982, 1986, 1989, 1991, 1992, 1993
6  *      The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. All advertising materials mentioning features or use of this software
22  *    must display the following acknowledgement:
23  *      This product includes software developed by the University of
24  *      California, Berkeley and its contributors.
25  * 4. Neither the name of the University nor the names of its contributors
26  *    may be used to endorse or promote products derived from this software
27  *    without specific prior written permission.
28  *
29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39  * SUCH DAMAGE.
40  *
41  *      @(#)init_main.c 8.9 (Berkeley) 1/21/94
42  */
43
44 #include <sys/cdefs.h>
45 __FBSDID("$FreeBSD$");
46
47 #include "opt_ddb.h"
48 #include "opt_init_path.h"
49
50 #include <sys/param.h>
51 #include <sys/kernel.h>
52 #include <sys/exec.h>
53 #include <sys/file.h>
54 #include <sys/filedesc.h>
55 #include <sys/jail.h>
56 #include <sys/ktr.h>
57 #include <sys/lock.h>
58 #include <sys/loginclass.h>
59 #include <sys/mount.h>
60 #include <sys/mutex.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/sysctl.h>
63 #include <sys/proc.h>
64 #include <sys/racct.h>
65 #include <sys/resourcevar.h>
66 #include <sys/systm.h>
67 #include <sys/signalvar.h>
68 #include <sys/vnode.h>
69 #include <sys/sysent.h>
70 #include <sys/reboot.h>
71 #include <sys/sched.h>
72 #include <sys/sx.h>
73 #include <sys/sysproto.h>
74 #include <sys/vmmeter.h>
75 #include <sys/unistd.h>
76 #include <sys/malloc.h>
77 #include <sys/conf.h>
78 #include <sys/cpuset.h>
79
80 #include <machine/cpu.h>
81
82 #include <security/audit/audit.h>
83 #include <security/mac/mac_framework.h>
84
85 #include <vm/vm.h>
86 #include <vm/vm_param.h>
87 #include <vm/pmap.h>
88 #include <vm/vm_map.h>
89 #include <sys/copyright.h>
90
91 #include <ddb/ddb.h>
92 #include <ddb/db_sym.h>
93
94 void mi_startup(void);                          /* Should be elsewhere */
95
96 /* Components of the first process -- never freed. */
97 static struct session session0;
98 static struct pgrp pgrp0;
99 struct  proc proc0;
100 struct  thread thread0 __aligned(16);
101 struct  vmspace vmspace0;
102 struct  proc *initproc;
103
104 #ifndef BOOTHOWTO
105 #define BOOTHOWTO       0
106 #endif
107 int     boothowto = BOOTHOWTO;  /* initialized so that it can be patched */
108 SYSCTL_INT(_debug, OID_AUTO, boothowto, CTLFLAG_RD, &boothowto, 0,
109         "Boot control flags, passed from loader");
110
111 #ifndef BOOTVERBOSE
112 #define BOOTVERBOSE     0
113 #endif
114 int     bootverbose = BOOTVERBOSE;
115 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, &bootverbose, 0,
116         "Control the output of verbose kernel messages");
117
118 /*
119  * This ensures that there is at least one entry so that the sysinit_set
120  * symbol is not undefined.  A sybsystem ID of SI_SUB_DUMMY is never
121  * executed.
122  */
123 SYSINIT(placeholder, SI_SUB_DUMMY, SI_ORDER_ANY, NULL, NULL);
124
125 /*
126  * The sysinit table itself.  Items are checked off as the are run.
127  * If we want to register new sysinit types, add them to newsysinit.
128  */
129 SET_DECLARE(sysinit_set, struct sysinit);
130 struct sysinit **sysinit, **sysinit_end;
131 struct sysinit **newsysinit, **newsysinit_end;
132
133 /*
134  * Merge a new sysinit set into the current set, reallocating it if
135  * necessary.  This can only be called after malloc is running.
136  */
137 void
138 sysinit_add(struct sysinit **set, struct sysinit **set_end)
139 {
140         struct sysinit **newset;
141         struct sysinit **sipp;
142         struct sysinit **xipp;
143         int count;
144
145         count = set_end - set;
146         if (newsysinit)
147                 count += newsysinit_end - newsysinit;
148         else
149                 count += sysinit_end - sysinit;
150         newset = malloc(count * sizeof(*sipp), M_TEMP, M_NOWAIT);
151         if (newset == NULL)
152                 panic("cannot malloc for sysinit");
153         xipp = newset;
154         if (newsysinit)
155                 for (sipp = newsysinit; sipp < newsysinit_end; sipp++)
156                         *xipp++ = *sipp;
157         else
158                 for (sipp = sysinit; sipp < sysinit_end; sipp++)
159                         *xipp++ = *sipp;
160         for (sipp = set; sipp < set_end; sipp++)
161                 *xipp++ = *sipp;
162         if (newsysinit)
163                 free(newsysinit, M_TEMP);
164         newsysinit = newset;
165         newsysinit_end = newset + count;
166 }
167
168 #if defined (DDB) && defined(VERBOSE_SYSINIT)
169 static const char *
170 symbol_name(vm_offset_t va, db_strategy_t strategy)
171 {
172         const char *name;
173         c_db_sym_t sym;
174         db_expr_t  offset;
175
176         if (va == 0)
177                 return (NULL);
178         sym = db_search_symbol(va, strategy, &offset);
179         if (offset != 0)
180                 return (NULL);
181         db_symbol_values(sym, &name, NULL);
182         return (name);
183 }
184 #endif
185
186 /*
187  * System startup; initialize the world, create process 0, mount root
188  * filesystem, and fork to create init and pagedaemon.  Most of the
189  * hard work is done in the lower-level initialization routines including
190  * startup(), which does memory initialization and autoconfiguration.
191  *
192  * This allows simple addition of new kernel subsystems that require
193  * boot time initialization.  It also allows substitution of subsystem
194  * (for instance, a scheduler, kernel profiler, or VM system) by object
195  * module.  Finally, it allows for optional "kernel threads".
196  */
197 void
198 mi_startup(void)
199 {
200
201         register struct sysinit **sipp;         /* system initialization*/
202         register struct sysinit **xipp;         /* interior loop of sort*/
203         register struct sysinit *save;          /* bubble*/
204
205 #if defined(VERBOSE_SYSINIT)
206         int last;
207         int verbose;
208 #endif
209
210         if (boothowto & RB_VERBOSE)
211                 bootverbose++;
212
213         if (sysinit == NULL) {
214                 sysinit = SET_BEGIN(sysinit_set);
215                 sysinit_end = SET_LIMIT(sysinit_set);
216         }
217
218 restart:
219         /*
220          * Perform a bubble sort of the system initialization objects by
221          * their subsystem (primary key) and order (secondary key).
222          */
223         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
224                 for (xipp = sipp + 1; xipp < sysinit_end; xipp++) {
225                         if ((*sipp)->subsystem < (*xipp)->subsystem ||
226                              ((*sipp)->subsystem == (*xipp)->subsystem &&
227                               (*sipp)->order <= (*xipp)->order))
228                                 continue;       /* skip*/
229                         save = *sipp;
230                         *sipp = *xipp;
231                         *xipp = save;
232                 }
233         }
234
235 #if defined(VERBOSE_SYSINIT)
236         last = SI_SUB_COPYRIGHT;
237         verbose = 0;
238 #if !defined(DDB)
239         printf("VERBOSE_SYSINIT: DDB not enabled, symbol lookups disabled.\n");
240 #endif
241 #endif
242
243         /*
244          * Traverse the (now) ordered list of system initialization tasks.
245          * Perform each task, and continue on to the next task.
246          *
247          * The last item on the list is expected to be the scheduler,
248          * which will not return.
249          */
250         for (sipp = sysinit; sipp < sysinit_end; sipp++) {
251
252                 if ((*sipp)->subsystem == SI_SUB_DUMMY)
253                         continue;       /* skip dummy task(s)*/
254
255                 if ((*sipp)->subsystem == SI_SUB_DONE)
256                         continue;
257
258 #if defined(VERBOSE_SYSINIT)
259                 if ((*sipp)->subsystem > last) {
260                         verbose = 1;
261                         last = (*sipp)->subsystem;
262                         printf("subsystem %x\n", last);
263                 }
264                 if (verbose) {
265 #if defined(DDB)
266                         const char *func, *data;
267
268                         func = symbol_name((vm_offset_t)(*sipp)->func,
269                             DB_STGY_PROC);
270                         data = symbol_name((vm_offset_t)(*sipp)->udata,
271                             DB_STGY_ANY);
272                         if (func != NULL && data != NULL)
273                                 printf("   %s(&%s)... ", func, data);
274                         else if (func != NULL)
275                                 printf("   %s(%p)... ", func, (*sipp)->udata);
276                         else
277 #endif
278                                 printf("   %p(%p)... ", (*sipp)->func,
279                                     (*sipp)->udata);
280                 }
281 #endif
282
283                 /* Call function */
284                 (*((*sipp)->func))((*sipp)->udata);
285
286 #if defined(VERBOSE_SYSINIT)
287                 if (verbose)
288                         printf("done.\n");
289 #endif
290
291                 /* Check off the one we're just done */
292                 (*sipp)->subsystem = SI_SUB_DONE;
293
294                 /* Check if we've installed more sysinit items via KLD */
295                 if (newsysinit != NULL) {
296                         if (sysinit != SET_BEGIN(sysinit_set))
297                                 free(sysinit, M_TEMP);
298                         sysinit = newsysinit;
299                         sysinit_end = newsysinit_end;
300                         newsysinit = NULL;
301                         newsysinit_end = NULL;
302                         goto restart;
303                 }
304         }
305
306         panic("Shouldn't get here!");
307         /* NOTREACHED*/
308 }
309
310
311 /*
312  ***************************************************************************
313  ****
314  **** The following SYSINIT's belong elsewhere, but have not yet
315  **** been moved.
316  ****
317  ***************************************************************************
318  */
319 static void
320 print_caddr_t(void *data)
321 {
322         printf("%s", (char *)data);
323 }
324
325 static void
326 print_version(void *data __unused)
327 {
328         int len;
329
330         /* Strip a trailing newline from version. */
331         len = strlen(version);
332         while (len > 0 && version[len - 1] == '\n')
333                 len--;
334         printf("%.*s %s\n", len, version, machine);
335         printf("%s\n", compiler_version);
336 }
337
338 SYSINIT(announce, SI_SUB_COPYRIGHT, SI_ORDER_FIRST, print_caddr_t,
339     copyright);
340 SYSINIT(trademark, SI_SUB_COPYRIGHT, SI_ORDER_SECOND, print_caddr_t,
341     trademark);
342 SYSINIT(version, SI_SUB_COPYRIGHT, SI_ORDER_THIRD, print_version, NULL);
343
344 #ifdef WITNESS
345 static char wit_warn[] =
346      "WARNING: WITNESS option enabled, expect reduced performance.\n";
347 SYSINIT(witwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 1,
348    print_caddr_t, wit_warn);
349 SYSINIT(witwarn2, SI_SUB_RUN_SCHEDULER, SI_ORDER_THIRD + 1,
350    print_caddr_t, wit_warn);
351 #endif
352
353 #ifdef DIAGNOSTIC
354 static char diag_warn[] =
355      "WARNING: DIAGNOSTIC option enabled, expect reduced performance.\n";
356 SYSINIT(diagwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 2,
357     print_caddr_t, diag_warn);
358 SYSINIT(diagwarn2, SI_SUB_RUN_SCHEDULER, SI_ORDER_THIRD + 2,
359     print_caddr_t, diag_warn);
360 #endif
361
362 static int
363 null_fetch_syscall_args(struct thread *td __unused,
364     struct syscall_args *sa __unused)
365 {
366
367         panic("null_fetch_syscall_args");
368 }
369
370 static void
371 null_set_syscall_retval(struct thread *td __unused, int error __unused)
372 {
373
374         panic("null_set_syscall_retval");
375 }
376
377 struct sysentvec null_sysvec = {
378         .sv_size        = 0,
379         .sv_table       = NULL,
380         .sv_mask        = 0,
381         .sv_sigsize     = 0,
382         .sv_sigtbl      = NULL,
383         .sv_errsize     = 0,
384         .sv_errtbl      = NULL,
385         .sv_transtrap   = NULL,
386         .sv_fixup       = NULL,
387         .sv_sendsig     = NULL,
388         .sv_sigcode     = NULL,
389         .sv_szsigcode   = NULL,
390         .sv_prepsyscall = NULL,
391         .sv_name        = "null",
392         .sv_coredump    = NULL,
393         .sv_imgact_try  = NULL,
394         .sv_minsigstksz = 0,
395         .sv_pagesize    = PAGE_SIZE,
396         .sv_minuser     = VM_MIN_ADDRESS,
397         .sv_maxuser     = VM_MAXUSER_ADDRESS,
398         .sv_usrstack    = USRSTACK,
399         .sv_psstrings   = PS_STRINGS,
400         .sv_stackprot   = VM_PROT_ALL,
401         .sv_copyout_strings     = NULL,
402         .sv_setregs     = NULL,
403         .sv_fixlimit    = NULL,
404         .sv_maxssiz     = NULL,
405         .sv_flags       = 0,
406         .sv_set_syscall_retval = null_set_syscall_retval,
407         .sv_fetch_syscall_args = null_fetch_syscall_args,
408         .sv_syscallnames = NULL,
409         .sv_schedtail   = NULL,
410 };
411
412 /*
413  ***************************************************************************
414  ****
415  **** The two following SYSINIT's are proc0 specific glue code.  I am not
416  **** convinced that they can not be safely combined, but their order of
417  **** operation has been maintained as the same as the original init_main.c
418  **** for right now.
419  ****
420  **** These probably belong in init_proc.c or kern_proc.c, since they
421  **** deal with proc0 (the fork template process).
422  ****
423  ***************************************************************************
424  */
425 /* ARGSUSED*/
426 static void
427 proc0_init(void *dummy __unused)
428 {
429         struct proc *p;
430         struct thread *td;
431         vm_paddr_t pageablemem;
432         int i;
433
434         GIANT_REQUIRED;
435         p = &proc0;
436         td = &thread0;
437         
438         /*
439          * Initialize magic number and osrel.
440          */
441         p->p_magic = P_MAGIC;
442         p->p_osrel = osreldate;
443
444         /*
445          * Initialize thread and process structures.
446          */
447         procinit();     /* set up proc zone */
448         threadinit();   /* set up UMA zones */
449
450         /*
451          * Initialise scheduler resources.
452          * Add scheduler specific parts to proc, thread as needed.
453          */
454         schedinit();    /* scheduler gets its house in order */
455         /*
456          * Initialize sleep queue hash table
457          */
458         sleepinit();
459
460         /*
461          * additional VM structures
462          */
463         vm_init2();
464
465         /*
466          * Create process 0 (the swapper).
467          */
468         LIST_INSERT_HEAD(&allproc, p, p_list);
469         LIST_INSERT_HEAD(PIDHASH(0), p, p_hash);
470         mtx_init(&pgrp0.pg_mtx, "process group", NULL, MTX_DEF | MTX_DUPOK);
471         p->p_pgrp = &pgrp0;
472         LIST_INSERT_HEAD(PGRPHASH(0), &pgrp0, pg_hash);
473         LIST_INIT(&pgrp0.pg_members);
474         LIST_INSERT_HEAD(&pgrp0.pg_members, p, p_pglist);
475
476         pgrp0.pg_session = &session0;
477         mtx_init(&session0.s_mtx, "session", NULL, MTX_DEF);
478         refcount_init(&session0.s_count, 1);
479         session0.s_leader = p;
480
481         p->p_sysent = &null_sysvec;
482         p->p_flag = P_SYSTEM | P_INMEM;
483         p->p_state = PRS_NORMAL;
484         knlist_init_mtx(&p->p_klist, &p->p_mtx);
485         STAILQ_INIT(&p->p_ktr);
486         p->p_nice = NZERO;
487         /* pid_max cannot be greater than PID_MAX */
488         td->td_tid = PID_MAX + 1;
489         LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
490         td->td_state = TDS_RUNNING;
491         td->td_pri_class = PRI_TIMESHARE;
492         td->td_user_pri = PUSER;
493         td->td_base_user_pri = PUSER;
494         td->td_lend_user_pri = PRI_MAX;
495         td->td_priority = PVM;
496         td->td_base_pri = PVM;
497         td->td_oncpu = 0;
498         td->td_flags = TDF_INMEM;
499         td->td_pflags = TDP_KTHREAD;
500         td->td_cpuset = cpuset_thread0();
501         prison0.pr_cpuset = cpuset_ref(td->td_cpuset);
502         p->p_peers = 0;
503         p->p_leader = p;
504
505
506         strncpy(p->p_comm, "kernel", sizeof (p->p_comm));
507         strncpy(td->td_name, "swapper", sizeof (td->td_name));
508
509         callout_init_mtx(&p->p_itcallout, &p->p_mtx, 0);
510         callout_init_mtx(&p->p_limco, &p->p_mtx, 0);
511         callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
512
513         /* Create credentials. */
514         p->p_ucred = crget();
515         p->p_ucred->cr_ngroups = 1;     /* group 0 */
516         p->p_ucred->cr_uidinfo = uifind(0);
517         p->p_ucred->cr_ruidinfo = uifind(0);
518         p->p_ucred->cr_prison = &prison0;
519         p->p_ucred->cr_loginclass = loginclass_find("default");
520 #ifdef AUDIT
521         audit_cred_kproc0(p->p_ucred);
522 #endif
523 #ifdef MAC
524         mac_cred_create_swapper(p->p_ucred);
525 #endif
526         td->td_ucred = crhold(p->p_ucred);
527
528         /* Create sigacts. */
529         p->p_sigacts = sigacts_alloc();
530
531         /* Initialize signal state for process 0. */
532         siginit(&proc0);
533
534         /* Create the file descriptor table. */
535         p->p_fd = fdinit(NULL);
536         p->p_fdtol = NULL;
537
538         /* Create the limits structures. */
539         p->p_limit = lim_alloc();
540         for (i = 0; i < RLIM_NLIMITS; i++)
541                 p->p_limit->pl_rlimit[i].rlim_cur =
542                     p->p_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
543         p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur =
544             p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
545         p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_cur =
546             p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
547         p->p_limit->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
548         p->p_limit->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
549         p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
550         p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
551         /* Cast to avoid overflow on i386/PAE. */
552         pageablemem = ptoa((vm_paddr_t)cnt.v_free_count);
553         p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_cur =
554             p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_max = pageablemem;
555         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = pageablemem / 3;
556         p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_max = pageablemem;
557         p->p_cpulimit = RLIM_INFINITY;
558
559         /* Initialize resource accounting structures. */
560         racct_create(&p->p_racct);
561
562         p->p_stats = pstats_alloc();
563
564         /* Allocate a prototype map so we have something to fork. */
565         pmap_pinit0(vmspace_pmap(&vmspace0));
566         p->p_vmspace = &vmspace0;
567         vmspace0.vm_refcnt = 1;
568
569         /*
570          * proc0 is not expected to enter usermode, so there is no special
571          * handling for sv_minuser here, like is done for exec_new_vmspace().
572          */
573         vm_map_init(&vmspace0.vm_map, vmspace_pmap(&vmspace0),
574             p->p_sysent->sv_minuser, p->p_sysent->sv_maxuser);
575
576         /*
577          * Call the init and ctor for the new thread and proc.  We wait
578          * to do this until all other structures are fairly sane.
579          */
580         EVENTHANDLER_INVOKE(process_init, p);
581         EVENTHANDLER_INVOKE(thread_init, td);
582         EVENTHANDLER_INVOKE(process_ctor, p);
583         EVENTHANDLER_INVOKE(thread_ctor, td);
584
585         /*
586          * Charge root for one process.
587          */
588         (void)chgproccnt(p->p_ucred->cr_ruidinfo, 1, 0);
589         PROC_LOCK(p);
590         racct_add_force(p, RACCT_NPROC, 1);
591         PROC_UNLOCK(p);
592 }
593 SYSINIT(p0init, SI_SUB_INTRINSIC, SI_ORDER_FIRST, proc0_init, NULL);
594
595 /* ARGSUSED*/
596 static void
597 proc0_post(void *dummy __unused)
598 {
599         struct timespec ts;
600         struct proc *p;
601         struct rusage ru;
602         struct thread *td;
603
604         /*
605          * Now we can look at the time, having had a chance to verify the
606          * time from the filesystem.  Pretend that proc0 started now.
607          */
608         sx_slock(&allproc_lock);
609         FOREACH_PROC_IN_SYSTEM(p) {
610                 microuptime(&p->p_stats->p_start);
611                 PROC_SLOCK(p);
612                 rufetch(p, &ru);        /* Clears thread stats */
613                 PROC_SUNLOCK(p);
614                 p->p_rux.rux_runtime = 0;
615                 p->p_rux.rux_uticks = 0;
616                 p->p_rux.rux_sticks = 0;
617                 p->p_rux.rux_iticks = 0;
618                 FOREACH_THREAD_IN_PROC(p, td) {
619                         td->td_runtime = 0;
620                 }
621         }
622         sx_sunlock(&allproc_lock);
623         PCPU_SET(switchtime, cpu_ticks());
624         PCPU_SET(switchticks, ticks);
625
626         /*
627          * Give the ``random'' number generator a thump.
628          */
629         nanotime(&ts);
630         srandom(ts.tv_sec ^ ts.tv_nsec);
631 }
632 SYSINIT(p0post, SI_SUB_INTRINSIC_POST, SI_ORDER_FIRST, proc0_post, NULL);
633
634 static void
635 random_init(void *dummy __unused)
636 {
637
638         /*
639          * After CPU has been started we have some randomness on most
640          * platforms via get_cyclecount().  For platforms that don't
641          * we will reseed random(9) in proc0_post() as well.
642          */
643         srandom(get_cyclecount());
644 }
645 SYSINIT(random, SI_SUB_RANDOM, SI_ORDER_FIRST, random_init, NULL);
646
647 /*
648  ***************************************************************************
649  ****
650  **** The following SYSINIT's and glue code should be moved to the
651  **** respective files on a per subsystem basis.
652  ****
653  ***************************************************************************
654  */
655
656
657 /*
658  ***************************************************************************
659  ****
660  **** The following code probably belongs in another file, like
661  **** kern/init_init.c.
662  ****
663  ***************************************************************************
664  */
665
666 /*
667  * List of paths to try when searching for "init".
668  */
669 static char init_path[MAXPATHLEN] =
670 #ifdef  INIT_PATH
671     __XSTRING(INIT_PATH);
672 #else
673     "/sbin/init:/sbin/oinit:/sbin/init.bak:/rescue/init";
674 #endif
675 SYSCTL_STRING(_kern, OID_AUTO, init_path, CTLFLAG_RD, init_path, 0,
676         "Path used to search the init process");
677
678 /*
679  * Shutdown timeout of init(8).
680  * Unused within kernel, but used to control init(8), hence do not remove.
681  */
682 #ifndef INIT_SHUTDOWN_TIMEOUT
683 #define INIT_SHUTDOWN_TIMEOUT 120
684 #endif
685 static int init_shutdown_timeout = INIT_SHUTDOWN_TIMEOUT;
686 SYSCTL_INT(_kern, OID_AUTO, init_shutdown_timeout,
687         CTLFLAG_RW, &init_shutdown_timeout, 0, "Shutdown timeout of init(8). "
688         "Unused within kernel, but used to control init(8)");
689
690 /*
691  * Start the initial user process; try exec'ing each pathname in init_path.
692  * The program is invoked with one argument containing the boot flags.
693  */
694 static void
695 start_init(void *dummy)
696 {
697         vm_offset_t addr;
698         struct execve_args args;
699         int options, error;
700         char *var, *path, *next, *s;
701         char *ucp, **uap, *arg0, *arg1;
702         struct thread *td;
703         struct proc *p;
704
705         mtx_lock(&Giant);
706
707         GIANT_REQUIRED;
708
709         td = curthread;
710         p = td->td_proc;
711
712         vfs_mountroot();
713
714         /*
715          * Need just enough stack to hold the faked-up "execve()" arguments.
716          */
717         addr = p->p_sysent->sv_usrstack - PAGE_SIZE;
718         if (vm_map_find(&p->p_vmspace->vm_map, NULL, 0, &addr, PAGE_SIZE,
719                         FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != 0)
720                 panic("init: couldn't allocate argument space");
721         p->p_vmspace->vm_maxsaddr = (caddr_t)addr;
722         p->p_vmspace->vm_ssize = 1;
723
724         if ((var = getenv("init_path")) != NULL) {
725                 strlcpy(init_path, var, sizeof(init_path));
726                 freeenv(var);
727         }
728         
729         for (path = init_path; *path != '\0'; path = next) {
730                 while (*path == ':')
731                         path++;
732                 if (*path == '\0')
733                         break;
734                 for (next = path; *next != '\0' && *next != ':'; next++)
735                         /* nothing */ ;
736                 if (bootverbose)
737                         printf("start_init: trying %.*s\n", (int)(next - path),
738                             path);
739                         
740                 /*
741                  * Move out the boot flag argument.
742                  */
743                 options = 0;
744                 ucp = (char *)p->p_sysent->sv_usrstack;
745                 (void)subyte(--ucp, 0);         /* trailing zero */
746                 if (boothowto & RB_SINGLE) {
747                         (void)subyte(--ucp, 's');
748                         options = 1;
749                 }
750 #ifdef notyet
751                 if (boothowto & RB_FASTBOOT) {
752                         (void)subyte(--ucp, 'f');
753                         options = 1;
754                 }
755 #endif
756
757 #ifdef BOOTCDROM
758                 (void)subyte(--ucp, 'C');
759                 options = 1;
760 #endif
761
762                 if (options == 0)
763                         (void)subyte(--ucp, '-');
764                 (void)subyte(--ucp, '-');               /* leading hyphen */
765                 arg1 = ucp;
766
767                 /*
768                  * Move out the file name (also arg 0).
769                  */
770                 (void)subyte(--ucp, 0);
771                 for (s = next - 1; s >= path; s--)
772                         (void)subyte(--ucp, *s);
773                 arg0 = ucp;
774
775                 /*
776                  * Move out the arg pointers.
777                  */
778                 uap = (char **)((intptr_t)ucp & ~(sizeof(intptr_t)-1));
779                 (void)suword((caddr_t)--uap, (long)0);  /* terminator */
780                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg1);
781                 (void)suword((caddr_t)--uap, (long)(intptr_t)arg0);
782
783                 /*
784                  * Point at the arguments.
785                  */
786                 args.fname = arg0;
787                 args.argv = uap;
788                 args.envv = NULL;
789
790                 /*
791                  * Now try to exec the program.  If can't for any reason
792                  * other than it doesn't exist, complain.
793                  *
794                  * Otherwise, return via fork_trampoline() all the way
795                  * to user mode as init!
796                  */
797                 if ((error = sys_execve(td, &args)) == 0) {
798                         mtx_unlock(&Giant);
799                         return;
800                 }
801                 if (error != ENOENT)
802                         printf("exec %.*s: error %d\n", (int)(next - path), 
803                             path, error);
804         }
805         printf("init: not found in path %s\n", init_path);
806         panic("no init");
807 }
808
809 /*
810  * Like kproc_create(), but runs in it's own address space.
811  * We do this early to reserve pid 1.
812  *
813  * Note special case - do not make it runnable yet.  Other work
814  * in progress will change this more.
815  */
816 static void
817 create_init(const void *udata __unused)
818 {
819         struct ucred *newcred, *oldcred;
820         int error;
821
822         error = fork1(&thread0, RFFDG | RFPROC | RFSTOPPED, 0, &initproc,
823             NULL, 0);
824         if (error)
825                 panic("cannot fork init: %d\n", error);
826         KASSERT(initproc->p_pid == 1, ("create_init: initproc->p_pid != 1"));
827         /* divorce init's credentials from the kernel's */
828         newcred = crget();
829         PROC_LOCK(initproc);
830         initproc->p_flag |= P_SYSTEM | P_INMEM;
831         oldcred = initproc->p_ucred;
832         crcopy(newcred, oldcred);
833 #ifdef MAC
834         mac_cred_create_init(newcred);
835 #endif
836 #ifdef AUDIT
837         audit_cred_proc1(newcred);
838 #endif
839         initproc->p_ucred = newcred;
840         PROC_UNLOCK(initproc);
841         crfree(oldcred);
842         cred_update_thread(FIRST_THREAD_IN_PROC(initproc));
843         cpu_set_fork_handler(FIRST_THREAD_IN_PROC(initproc), start_init, NULL);
844 }
845 SYSINIT(init, SI_SUB_CREATE_INIT, SI_ORDER_FIRST, create_init, NULL);
846
847 /*
848  * Make it runnable now.
849  */
850 static void
851 kick_init(const void *udata __unused)
852 {
853         struct thread *td;
854
855         td = FIRST_THREAD_IN_PROC(initproc);
856         thread_lock(td);
857         TD_SET_CAN_RUN(td);
858         sched_add(td, SRQ_BORING);
859         thread_unlock(td);
860 }
861 SYSINIT(kickinit, SI_SUB_KTHREAD_INIT, SI_ORDER_FIRST, kick_init, NULL);