xref: /freebsd/sys/kern/init_main.c (revision 0957b409a90fd597c1e9124cbaf3edd2b488f4ac)
1 /*-
2  * SPDX-License-Identifier: BSD-4-Clause
3  *
4  * Copyright (c) 1995 Terrence R. Lambert
5  * All rights reserved.
6  *
7  * Copyright (c) 1982, 1986, 1989, 1991, 1992, 1993
8  *	The Regents of the University of California.  All rights reserved.
9  * (c) UNIX System Laboratories, Inc.
10  * All or some portions of this file are derived from material licensed
11  * to the University of California by American Telephone and Telegraph
12  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
13  * the permission of UNIX System Laboratories, Inc.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. All advertising materials mentioning features or use of this software
24  *    must display the following acknowledgement:
25  *	This product includes software developed by the University of
26  *	California, Berkeley and its contributors.
27  * 4. Neither the name of the University nor the names of its contributors
28  *    may be used to endorse or promote products derived from this software
29  *    without specific prior written permission.
30  *
31  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
32  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
33  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
34  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
35  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
36  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
37  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
38  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
39  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
40  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41  * SUCH DAMAGE.
42  *
43  *	@(#)init_main.c	8.9 (Berkeley) 1/21/94
44  */
45 
46 #include <sys/cdefs.h>
47 __FBSDID("$FreeBSD$");
48 
49 #include "opt_ddb.h"
50 #include "opt_init_path.h"
51 #include "opt_verbose_sysinit.h"
52 
53 #include <sys/param.h>
54 #include <sys/kernel.h>
55 #include <sys/epoch.h>
56 #include <sys/exec.h>
57 #include <sys/file.h>
58 #include <sys/filedesc.h>
59 #include <sys/imgact.h>
60 #include <sys/jail.h>
61 #include <sys/ktr.h>
62 #include <sys/lock.h>
63 #include <sys/loginclass.h>
64 #include <sys/mount.h>
65 #include <sys/mutex.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/proc.h>
69 #include <sys/racct.h>
70 #include <sys/resourcevar.h>
71 #include <sys/systm.h>
72 #include <sys/signalvar.h>
73 #include <sys/vnode.h>
74 #include <sys/sysent.h>
75 #include <sys/reboot.h>
76 #include <sys/sched.h>
77 #include <sys/sx.h>
78 #include <sys/sysproto.h>
79 #include <sys/vmmeter.h>
80 #include <sys/unistd.h>
81 #include <sys/malloc.h>
82 #include <sys/conf.h>
83 #include <sys/cpuset.h>
84 
85 #include <machine/cpu.h>
86 
87 #include <security/audit/audit.h>
88 #include <security/mac/mac_framework.h>
89 
90 #include <vm/vm.h>
91 #include <vm/vm_param.h>
92 #include <vm/vm_extern.h>
93 #include <vm/pmap.h>
94 #include <vm/vm_map.h>
95 #include <sys/copyright.h>
96 
97 #include <ddb/ddb.h>
98 #include <ddb/db_sym.h>
99 
100 void mi_startup(void);				/* Should be elsewhere */
101 
102 /* Components of the first process -- never freed. */
103 static struct session session0;
104 static struct pgrp pgrp0;
105 struct	proc proc0;
106 struct thread0_storage thread0_st __aligned(32);
107 struct	vmspace vmspace0;
108 struct	proc *initproc;
109 
110 int
111 linux_alloc_current_noop(struct thread *td __unused, int flags __unused)
112 {
113 	return (0);
114 }
115 int (*lkpi_alloc_current)(struct thread *, int) = linux_alloc_current_noop;
116 
117 
118 #ifndef BOOTHOWTO
119 #define	BOOTHOWTO	0
120 #endif
121 int	boothowto = BOOTHOWTO;	/* initialized so that it can be patched */
122 SYSCTL_INT(_debug, OID_AUTO, boothowto, CTLFLAG_RD, &boothowto, 0,
123 	"Boot control flags, passed from loader");
124 
125 #ifndef BOOTVERBOSE
126 #define	BOOTVERBOSE	0
127 #endif
128 int	bootverbose = BOOTVERBOSE;
129 SYSCTL_INT(_debug, OID_AUTO, bootverbose, CTLFLAG_RW, &bootverbose, 0,
130 	"Control the output of verbose kernel messages");
131 
132 #ifdef VERBOSE_SYSINIT
133 /*
134  * We'll use the defined value of VERBOSE_SYSINIT from the kernel config to
135  * dictate the default VERBOSE_SYSINIT behavior.  Significant values for this
136  * option and associated tunable are:
137  * - 0, 'compiled in but silent by default'
138  * - 1, 'compiled in but verbose by default' (default)
139  */
140 int	verbose_sysinit = VERBOSE_SYSINIT;
141 TUNABLE_INT("debug.verbose_sysinit", &verbose_sysinit);
142 #endif
143 
144 #ifdef INVARIANTS
145 FEATURE(invariants, "Kernel compiled with INVARIANTS, may affect performance");
146 #endif
147 
148 /*
149  * This ensures that there is at least one entry so that the sysinit_set
150  * symbol is not undefined.  A sybsystem ID of SI_SUB_DUMMY is never
151  * executed.
152  */
153 SYSINIT(placeholder, SI_SUB_DUMMY, SI_ORDER_ANY, NULL, NULL);
154 
155 /*
156  * The sysinit table itself.  Items are checked off as the are run.
157  * If we want to register new sysinit types, add them to newsysinit.
158  */
159 SET_DECLARE(sysinit_set, struct sysinit);
160 struct sysinit **sysinit, **sysinit_end;
161 struct sysinit **newsysinit, **newsysinit_end;
162 
163 EVENTHANDLER_LIST_DECLARE(process_init);
164 EVENTHANDLER_LIST_DECLARE(thread_init);
165 EVENTHANDLER_LIST_DECLARE(process_ctor);
166 EVENTHANDLER_LIST_DECLARE(thread_ctor);
167 
168 /*
169  * Merge a new sysinit set into the current set, reallocating it if
170  * necessary.  This can only be called after malloc is running.
171  */
172 void
173 sysinit_add(struct sysinit **set, struct sysinit **set_end)
174 {
175 	struct sysinit **newset;
176 	struct sysinit **sipp;
177 	struct sysinit **xipp;
178 	int count;
179 
180 	count = set_end - set;
181 	if (newsysinit)
182 		count += newsysinit_end - newsysinit;
183 	else
184 		count += sysinit_end - sysinit;
185 	newset = malloc(count * sizeof(*sipp), M_TEMP, M_NOWAIT);
186 	if (newset == NULL)
187 		panic("cannot malloc for sysinit");
188 	xipp = newset;
189 	if (newsysinit)
190 		for (sipp = newsysinit; sipp < newsysinit_end; sipp++)
191 			*xipp++ = *sipp;
192 	else
193 		for (sipp = sysinit; sipp < sysinit_end; sipp++)
194 			*xipp++ = *sipp;
195 	for (sipp = set; sipp < set_end; sipp++)
196 		*xipp++ = *sipp;
197 	if (newsysinit)
198 		free(newsysinit, M_TEMP);
199 	newsysinit = newset;
200 	newsysinit_end = newset + count;
201 }
202 
203 #if defined (DDB) && defined(VERBOSE_SYSINIT)
204 static const char *
205 symbol_name(vm_offset_t va, db_strategy_t strategy)
206 {
207 	const char *name;
208 	c_db_sym_t sym;
209 	db_expr_t  offset;
210 
211 	if (va == 0)
212 		return (NULL);
213 	sym = db_search_symbol(va, strategy, &offset);
214 	if (offset != 0)
215 		return (NULL);
216 	db_symbol_values(sym, &name, NULL);
217 	return (name);
218 }
219 #endif
220 
221 /*
222  * System startup; initialize the world, create process 0, mount root
223  * filesystem, and fork to create init and pagedaemon.  Most of the
224  * hard work is done in the lower-level initialization routines including
225  * startup(), which does memory initialization and autoconfiguration.
226  *
227  * This allows simple addition of new kernel subsystems that require
228  * boot time initialization.  It also allows substitution of subsystem
229  * (for instance, a scheduler, kernel profiler, or VM system) by object
230  * module.  Finally, it allows for optional "kernel threads".
231  */
232 void
233 mi_startup(void)
234 {
235 
236 	struct sysinit **sipp;	/* system initialization*/
237 	struct sysinit **xipp;	/* interior loop of sort*/
238 	struct sysinit *save;	/* bubble*/
239 
240 #if defined(VERBOSE_SYSINIT)
241 	int last;
242 	int verbose;
243 #endif
244 
245 	TSENTER();
246 
247 	if (boothowto & RB_VERBOSE)
248 		bootverbose++;
249 
250 	if (sysinit == NULL) {
251 		sysinit = SET_BEGIN(sysinit_set);
252 		sysinit_end = SET_LIMIT(sysinit_set);
253 	}
254 
255 restart:
256 	/*
257 	 * Perform a bubble sort of the system initialization objects by
258 	 * their subsystem (primary key) and order (secondary key).
259 	 */
260 	for (sipp = sysinit; sipp < sysinit_end; sipp++) {
261 		for (xipp = sipp + 1; xipp < sysinit_end; xipp++) {
262 			if ((*sipp)->subsystem < (*xipp)->subsystem ||
263 			     ((*sipp)->subsystem == (*xipp)->subsystem &&
264 			      (*sipp)->order <= (*xipp)->order))
265 				continue;	/* skip*/
266 			save = *sipp;
267 			*sipp = *xipp;
268 			*xipp = save;
269 		}
270 	}
271 
272 #if defined(VERBOSE_SYSINIT)
273 	last = SI_SUB_COPYRIGHT;
274 	verbose = 0;
275 #if !defined(DDB)
276 	printf("VERBOSE_SYSINIT: DDB not enabled, symbol lookups disabled.\n");
277 #endif
278 #endif
279 
280 	/*
281 	 * Traverse the (now) ordered list of system initialization tasks.
282 	 * Perform each task, and continue on to the next task.
283 	 */
284 	for (sipp = sysinit; sipp < sysinit_end; sipp++) {
285 
286 		if ((*sipp)->subsystem == SI_SUB_DUMMY)
287 			continue;	/* skip dummy task(s)*/
288 
289 		if ((*sipp)->subsystem == SI_SUB_DONE)
290 			continue;
291 
292 #if defined(VERBOSE_SYSINIT)
293 		if ((*sipp)->subsystem > last && verbose_sysinit != 0) {
294 			verbose = 1;
295 			last = (*sipp)->subsystem;
296 			printf("subsystem %x\n", last);
297 		}
298 		if (verbose) {
299 #if defined(DDB)
300 			const char *func, *data;
301 
302 			func = symbol_name((vm_offset_t)(*sipp)->func,
303 			    DB_STGY_PROC);
304 			data = symbol_name((vm_offset_t)(*sipp)->udata,
305 			    DB_STGY_ANY);
306 			if (func != NULL && data != NULL)
307 				printf("   %s(&%s)... ", func, data);
308 			else if (func != NULL)
309 				printf("   %s(%p)... ", func, (*sipp)->udata);
310 			else
311 #endif
312 				printf("   %p(%p)... ", (*sipp)->func,
313 				    (*sipp)->udata);
314 		}
315 #endif
316 
317 		/* Call function */
318 		(*((*sipp)->func))((*sipp)->udata);
319 
320 #if defined(VERBOSE_SYSINIT)
321 		if (verbose)
322 			printf("done.\n");
323 #endif
324 
325 		/* Check off the one we're just done */
326 		(*sipp)->subsystem = SI_SUB_DONE;
327 
328 		/* Check if we've installed more sysinit items via KLD */
329 		if (newsysinit != NULL) {
330 			if (sysinit != SET_BEGIN(sysinit_set))
331 				free(sysinit, M_TEMP);
332 			sysinit = newsysinit;
333 			sysinit_end = newsysinit_end;
334 			newsysinit = NULL;
335 			newsysinit_end = NULL;
336 			goto restart;
337 		}
338 	}
339 
340 	TSEXIT();	/* Here so we don't overlap with start_init. */
341 
342 	mtx_assert(&Giant, MA_OWNED | MA_NOTRECURSED);
343 	mtx_unlock(&Giant);
344 
345 	/*
346 	 * Now hand over this thread to swapper.
347 	 */
348 	swapper();
349 	/* NOTREACHED*/
350 }
351 
352 static void
353 print_caddr_t(void *data)
354 {
355 	printf("%s", (char *)data);
356 }
357 
358 static void
359 print_version(void *data __unused)
360 {
361 	int len;
362 
363 	/* Strip a trailing newline from version. */
364 	len = strlen(version);
365 	while (len > 0 && version[len - 1] == '\n')
366 		len--;
367 	printf("%.*s %s\n", len, version, machine);
368 	printf("%s\n", compiler_version);
369 }
370 
371 SYSINIT(announce, SI_SUB_COPYRIGHT, SI_ORDER_FIRST, print_caddr_t,
372     copyright);
373 SYSINIT(trademark, SI_SUB_COPYRIGHT, SI_ORDER_SECOND, print_caddr_t,
374     trademark);
375 SYSINIT(version, SI_SUB_COPYRIGHT, SI_ORDER_THIRD, print_version, NULL);
376 
377 #ifdef WITNESS
378 static char wit_warn[] =
379      "WARNING: WITNESS option enabled, expect reduced performance.\n";
380 SYSINIT(witwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 1,
381    print_caddr_t, wit_warn);
382 SYSINIT(witwarn2, SI_SUB_LAST, SI_ORDER_THIRD + 1,
383    print_caddr_t, wit_warn);
384 #endif
385 
386 #ifdef DIAGNOSTIC
387 static char diag_warn[] =
388      "WARNING: DIAGNOSTIC option enabled, expect reduced performance.\n";
389 SYSINIT(diagwarn, SI_SUB_COPYRIGHT, SI_ORDER_THIRD + 2,
390     print_caddr_t, diag_warn);
391 SYSINIT(diagwarn2, SI_SUB_LAST, SI_ORDER_THIRD + 2,
392     print_caddr_t, diag_warn);
393 #endif
394 
395 static int
396 null_fetch_syscall_args(struct thread *td __unused)
397 {
398 
399 	panic("null_fetch_syscall_args");
400 }
401 
402 static void
403 null_set_syscall_retval(struct thread *td __unused, int error __unused)
404 {
405 
406 	panic("null_set_syscall_retval");
407 }
408 
409 struct sysentvec null_sysvec = {
410 	.sv_size	= 0,
411 	.sv_table	= NULL,
412 	.sv_errsize	= 0,
413 	.sv_errtbl	= NULL,
414 	.sv_transtrap	= NULL,
415 	.sv_fixup	= NULL,
416 	.sv_sendsig	= NULL,
417 	.sv_sigcode	= NULL,
418 	.sv_szsigcode	= NULL,
419 	.sv_name	= "null",
420 	.sv_coredump	= NULL,
421 	.sv_imgact_try	= NULL,
422 	.sv_minsigstksz	= 0,
423 	.sv_pagesize	= PAGE_SIZE,
424 	.sv_minuser	= VM_MIN_ADDRESS,
425 	.sv_maxuser	= VM_MAXUSER_ADDRESS,
426 	.sv_usrstack	= USRSTACK,
427 	.sv_psstrings	= PS_STRINGS,
428 	.sv_stackprot	= VM_PROT_ALL,
429 	.sv_copyout_strings	= NULL,
430 	.sv_setregs	= NULL,
431 	.sv_fixlimit	= NULL,
432 	.sv_maxssiz	= NULL,
433 	.sv_flags	= 0,
434 	.sv_set_syscall_retval = null_set_syscall_retval,
435 	.sv_fetch_syscall_args = null_fetch_syscall_args,
436 	.sv_syscallnames = NULL,
437 	.sv_schedtail	= NULL,
438 	.sv_thread_detach = NULL,
439 	.sv_trap	= NULL,
440 };
441 
442 /*
443  * The two following SYSINIT's are proc0 specific glue code.  I am not
444  * convinced that they can not be safely combined, but their order of
445  * operation has been maintained as the same as the original init_main.c
446  * for right now.
447  */
448 /* ARGSUSED*/
449 static void
450 proc0_init(void *dummy __unused)
451 {
452 	struct proc *p;
453 	struct thread *td;
454 	struct ucred *newcred;
455 	struct uidinfo tmpuinfo;
456 	struct loginclass tmplc = {
457 		.lc_name = "",
458 	};
459 	vm_paddr_t pageablemem;
460 	int i;
461 
462 	GIANT_REQUIRED;
463 	p = &proc0;
464 	td = &thread0;
465 
466 	/*
467 	 * Initialize magic number and osrel.
468 	 */
469 	p->p_magic = P_MAGIC;
470 	p->p_osrel = osreldate;
471 
472 	/*
473 	 * Initialize thread and process structures.
474 	 */
475 	procinit();	/* set up proc zone */
476 	threadinit();	/* set up UMA zones */
477 
478 	/*
479 	 * Initialise scheduler resources.
480 	 * Add scheduler specific parts to proc, thread as needed.
481 	 */
482 	schedinit();	/* scheduler gets its house in order */
483 
484 	/*
485 	 * Create process 0 (the swapper).
486 	 */
487 	LIST_INSERT_HEAD(&allproc, p, p_list);
488 	LIST_INSERT_HEAD(PIDHASH(0), p, p_hash);
489 	mtx_init(&pgrp0.pg_mtx, "process group", NULL, MTX_DEF | MTX_DUPOK);
490 	p->p_pgrp = &pgrp0;
491 	LIST_INSERT_HEAD(PGRPHASH(0), &pgrp0, pg_hash);
492 	LIST_INIT(&pgrp0.pg_members);
493 	LIST_INSERT_HEAD(&pgrp0.pg_members, p, p_pglist);
494 
495 	pgrp0.pg_session = &session0;
496 	mtx_init(&session0.s_mtx, "session", NULL, MTX_DEF);
497 	refcount_init(&session0.s_count, 1);
498 	session0.s_leader = p;
499 
500 	p->p_sysent = &null_sysvec;
501 	p->p_flag = P_SYSTEM | P_INMEM | P_KPROC;
502 	p->p_flag2 = 0;
503 	p->p_state = PRS_NORMAL;
504 	p->p_klist = knlist_alloc(&p->p_mtx);
505 	STAILQ_INIT(&p->p_ktr);
506 	p->p_nice = NZERO;
507 	/* pid_max cannot be greater than PID_MAX */
508 	td->td_tid = PID_MAX + 1;
509 	LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
510 	td->td_state = TDS_RUNNING;
511 	td->td_pri_class = PRI_TIMESHARE;
512 	td->td_user_pri = PUSER;
513 	td->td_base_user_pri = PUSER;
514 	td->td_lend_user_pri = PRI_MAX;
515 	td->td_priority = PVM;
516 	td->td_base_pri = PVM;
517 	td->td_oncpu = curcpu;
518 	td->td_flags = TDF_INMEM;
519 	td->td_pflags = TDP_KTHREAD;
520 	td->td_cpuset = cpuset_thread0();
521 	td->td_domain.dr_policy = td->td_cpuset->cs_domain;
522 	epoch_thread_init(td);
523 	prison0_init();
524 	p->p_peers = 0;
525 	p->p_leader = p;
526 	p->p_reaper = p;
527 	p->p_treeflag |= P_TREE_REAPER;
528 	LIST_INIT(&p->p_reaplist);
529 
530 	strncpy(p->p_comm, "kernel", sizeof (p->p_comm));
531 	strncpy(td->td_name, "swapper", sizeof (td->td_name));
532 
533 	callout_init_mtx(&p->p_itcallout, &p->p_mtx, 0);
534 	callout_init_mtx(&p->p_limco, &p->p_mtx, 0);
535 	callout_init(&td->td_slpcallout, 1);
536 
537 	/* Create credentials. */
538 	newcred = crget();
539 	newcred->cr_ngroups = 1;	/* group 0 */
540 	/* A hack to prevent uifind from tripping over NULL pointers. */
541 	curthread->td_ucred = newcred;
542 	tmpuinfo.ui_uid = 1;
543 	newcred->cr_uidinfo = newcred->cr_ruidinfo = &tmpuinfo;
544 	newcred->cr_uidinfo = uifind(0);
545 	newcred->cr_ruidinfo = uifind(0);
546 	newcred->cr_loginclass = &tmplc;
547 	newcred->cr_loginclass = loginclass_find("default");
548 	/* End hack. creds get properly set later with thread_cow_get_proc */
549 	curthread->td_ucred = NULL;
550 	newcred->cr_prison = &prison0;
551 	proc_set_cred_init(p, newcred);
552 #ifdef AUDIT
553 	audit_cred_kproc0(newcred);
554 #endif
555 #ifdef MAC
556 	mac_cred_create_swapper(newcred);
557 #endif
558 	/* Create sigacts. */
559 	p->p_sigacts = sigacts_alloc();
560 
561 	/* Initialize signal state for process 0. */
562 	siginit(&proc0);
563 
564 	/* Create the file descriptor table. */
565 	p->p_fd = fdinit(NULL, false);
566 	p->p_fdtol = NULL;
567 
568 	/* Create the limits structures. */
569 	p->p_limit = lim_alloc();
570 	for (i = 0; i < RLIM_NLIMITS; i++)
571 		p->p_limit->pl_rlimit[i].rlim_cur =
572 		    p->p_limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
573 	p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur =
574 	    p->p_limit->pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
575 	p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_cur =
576 	    p->p_limit->pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
577 	p->p_limit->pl_rlimit[RLIMIT_DATA].rlim_cur = dfldsiz;
578 	p->p_limit->pl_rlimit[RLIMIT_DATA].rlim_max = maxdsiz;
579 	p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur = dflssiz;
580 	p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_max = maxssiz;
581 	/* Cast to avoid overflow on i386/PAE. */
582 	pageablemem = ptoa((vm_paddr_t)vm_free_count());
583 	p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_cur =
584 	    p->p_limit->pl_rlimit[RLIMIT_RSS].rlim_max = pageablemem;
585 	p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = pageablemem / 3;
586 	p->p_limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_max = pageablemem;
587 	p->p_cpulimit = RLIM_INFINITY;
588 
589 	PROC_LOCK(p);
590 	thread_cow_get_proc(td, p);
591 	PROC_UNLOCK(p);
592 
593 	/* Initialize resource accounting structures. */
594 	racct_create(&p->p_racct);
595 
596 	p->p_stats = pstats_alloc();
597 
598 	/* Allocate a prototype map so we have something to fork. */
599 	p->p_vmspace = &vmspace0;
600 	vmspace0.vm_refcnt = 1;
601 	pmap_pinit0(vmspace_pmap(&vmspace0));
602 
603 	/*
604 	 * proc0 is not expected to enter usermode, so there is no special
605 	 * handling for sv_minuser here, like is done for exec_new_vmspace().
606 	 */
607 	vm_map_init(&vmspace0.vm_map, vmspace_pmap(&vmspace0),
608 	    p->p_sysent->sv_minuser, p->p_sysent->sv_maxuser);
609 
610 	/*
611 	 * Call the init and ctor for the new thread and proc.  We wait
612 	 * to do this until all other structures are fairly sane.
613 	 */
614 	EVENTHANDLER_DIRECT_INVOKE(process_init, p);
615 	EVENTHANDLER_DIRECT_INVOKE(thread_init, td);
616 	EVENTHANDLER_DIRECT_INVOKE(process_ctor, p);
617 	EVENTHANDLER_DIRECT_INVOKE(thread_ctor, td);
618 
619 	/*
620 	 * Charge root for one process.
621 	 */
622 	(void)chgproccnt(p->p_ucred->cr_ruidinfo, 1, 0);
623 	PROC_LOCK(p);
624 	racct_add_force(p, RACCT_NPROC, 1);
625 	PROC_UNLOCK(p);
626 }
627 SYSINIT(p0init, SI_SUB_INTRINSIC, SI_ORDER_FIRST, proc0_init, NULL);
628 
629 /* ARGSUSED*/
630 static void
631 proc0_post(void *dummy __unused)
632 {
633 	struct timespec ts;
634 	struct proc *p;
635 	struct rusage ru;
636 	struct thread *td;
637 
638 	/*
639 	 * Now we can look at the time, having had a chance to verify the
640 	 * time from the filesystem.  Pretend that proc0 started now.
641 	 */
642 	sx_slock(&allproc_lock);
643 	FOREACH_PROC_IN_SYSTEM(p) {
644 		PROC_LOCK(p);
645 		if (p->p_state == PRS_NEW) {
646 			PROC_UNLOCK(p);
647 			continue;
648 		}
649 		microuptime(&p->p_stats->p_start);
650 		PROC_STATLOCK(p);
651 		rufetch(p, &ru);	/* Clears thread stats */
652 		p->p_rux.rux_runtime = 0;
653 		p->p_rux.rux_uticks = 0;
654 		p->p_rux.rux_sticks = 0;
655 		p->p_rux.rux_iticks = 0;
656 		PROC_STATUNLOCK(p);
657 		FOREACH_THREAD_IN_PROC(p, td) {
658 			td->td_runtime = 0;
659 		}
660 		PROC_UNLOCK(p);
661 	}
662 	sx_sunlock(&allproc_lock);
663 	PCPU_SET(switchtime, cpu_ticks());
664 	PCPU_SET(switchticks, ticks);
665 
666 	/*
667 	 * Give the ``random'' number generator a thump.
668 	 */
669 	nanotime(&ts);
670 	srandom(ts.tv_sec ^ ts.tv_nsec);
671 }
672 SYSINIT(p0post, SI_SUB_INTRINSIC_POST, SI_ORDER_FIRST, proc0_post, NULL);
673 
674 static void
675 random_init(void *dummy __unused)
676 {
677 
678 	/*
679 	 * After CPU has been started we have some randomness on most
680 	 * platforms via get_cyclecount().  For platforms that don't
681 	 * we will reseed random(9) in proc0_post() as well.
682 	 */
683 	srandom(get_cyclecount());
684 }
685 SYSINIT(random, SI_SUB_RANDOM, SI_ORDER_FIRST, random_init, NULL);
686 
687 /*
688  ***************************************************************************
689  ****
690  **** The following SYSINIT's and glue code should be moved to the
691  **** respective files on a per subsystem basis.
692  ****
693  ***************************************************************************
694  */
695 
696 /*
697  * List of paths to try when searching for "init".
698  */
699 static char init_path[MAXPATHLEN] =
700 #ifdef	INIT_PATH
701     __XSTRING(INIT_PATH);
702 #else
703     "/sbin/init:/sbin/oinit:/sbin/init.bak:/rescue/init";
704 #endif
705 SYSCTL_STRING(_kern, OID_AUTO, init_path, CTLFLAG_RD, init_path, 0,
706 	"Path used to search the init process");
707 
708 /*
709  * Shutdown timeout of init(8).
710  * Unused within kernel, but used to control init(8), hence do not remove.
711  */
712 #ifndef INIT_SHUTDOWN_TIMEOUT
713 #define INIT_SHUTDOWN_TIMEOUT 120
714 #endif
715 static int init_shutdown_timeout = INIT_SHUTDOWN_TIMEOUT;
716 SYSCTL_INT(_kern, OID_AUTO, init_shutdown_timeout,
717 	CTLFLAG_RW, &init_shutdown_timeout, 0, "Shutdown timeout of init(8). "
718 	"Unused within kernel, but used to control init(8)");
719 
720 /*
721  * Start the initial user process; try exec'ing each pathname in init_path.
722  * The program is invoked with one argument containing the boot flags.
723  */
724 static void
725 start_init(void *dummy)
726 {
727 	struct image_args args;
728 	int error;
729 	char *var, *path;
730 	char *free_init_path, *tmp_init_path;
731 	struct thread *td;
732 	struct proc *p;
733 	struct vmspace *oldvmspace;
734 
735 	TSENTER();	/* Here so we don't overlap with mi_startup. */
736 
737 	td = curthread;
738 	p = td->td_proc;
739 
740 	vfs_mountroot();
741 
742 	/* Wipe GELI passphrase from the environment. */
743 	kern_unsetenv("kern.geom.eli.passphrase");
744 
745 	if ((var = kern_getenv("init_path")) != NULL) {
746 		strlcpy(init_path, var, sizeof(init_path));
747 		freeenv(var);
748 	}
749 	free_init_path = tmp_init_path = strdup(init_path, M_TEMP);
750 
751 	while ((path = strsep(&tmp_init_path, ":")) != NULL) {
752 		if (bootverbose)
753 			printf("start_init: trying %s\n", path);
754 
755 		memset(&args, 0, sizeof(args));
756 		error = exec_alloc_args(&args);
757 		if (error != 0)
758 			panic("%s: Can't allocate space for init arguments %d",
759 			    __func__, error);
760 
761 		error = exec_args_add_fname(&args, path, UIO_SYSSPACE);
762 		if (error != 0)
763 			panic("%s: Can't add fname %d", __func__, error);
764 		error = exec_args_add_arg(&args, path, UIO_SYSSPACE);
765 		if (error != 0)
766 			panic("%s: Can't add argv[0] %d", __func__, error);
767 		if (boothowto & RB_SINGLE)
768 			error = exec_args_add_arg(&args, "-s", UIO_SYSSPACE);
769 		if (error != 0)
770 			panic("%s: Can't add argv[0] %d", __func__, error);
771 
772 		/*
773 		 * Now try to exec the program.  If can't for any reason
774 		 * other than it doesn't exist, complain.
775 		 *
776 		 * Otherwise, return via fork_trampoline() all the way
777 		 * to user mode as init!
778 		 */
779 		KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0,
780 		    ("nested execve"));
781 		oldvmspace = td->td_proc->p_vmspace;
782 		error = kern_execve(td, &args, NULL);
783 		KASSERT(error != 0,
784 		    ("kern_execve returned success, not EJUSTRETURN"));
785 		if (error == EJUSTRETURN) {
786 			if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
787 				KASSERT(p->p_vmspace != oldvmspace,
788 				    ("oldvmspace still used"));
789 				vmspace_free(oldvmspace);
790 				td->td_pflags &= ~TDP_EXECVMSPC;
791 			}
792 			free(free_init_path, M_TEMP);
793 			TSEXIT();
794 			return;
795 		}
796 		if (error != ENOENT)
797 			printf("exec %s: error %d\n", path, error);
798 	}
799 	free(free_init_path, M_TEMP);
800 	printf("init: not found in path %s\n", init_path);
801 	panic("no init");
802 }
803 
804 /*
805  * Like kproc_create(), but runs in its own address space.
806  * We do this early to reserve pid 1.
807  *
808  * Note special case - do not make it runnable yet.  Other work
809  * in progress will change this more.
810  */
811 static void
812 create_init(const void *udata __unused)
813 {
814 	struct fork_req fr;
815 	struct ucred *newcred, *oldcred;
816 	struct thread *td;
817 	int error;
818 
819 	bzero(&fr, sizeof(fr));
820 	fr.fr_flags = RFFDG | RFPROC | RFSTOPPED;
821 	fr.fr_procp = &initproc;
822 	error = fork1(&thread0, &fr);
823 	if (error)
824 		panic("cannot fork init: %d\n", error);
825 	KASSERT(initproc->p_pid == 1, ("create_init: initproc->p_pid != 1"));
826 	/* divorce init's credentials from the kernel's */
827 	newcred = crget();
828 	sx_xlock(&proctree_lock);
829 	PROC_LOCK(initproc);
830 	initproc->p_flag |= P_SYSTEM | P_INMEM;
831 	initproc->p_treeflag |= P_TREE_REAPER;
832 	oldcred = initproc->p_ucred;
833 	crcopy(newcred, oldcred);
834 #ifdef MAC
835 	mac_cred_create_init(newcred);
836 #endif
837 #ifdef AUDIT
838 	audit_cred_proc1(newcred);
839 #endif
840 	proc_set_cred(initproc, newcred);
841 	td = FIRST_THREAD_IN_PROC(initproc);
842 	crfree(td->td_ucred);
843 	td->td_ucred = crhold(initproc->p_ucred);
844 	PROC_UNLOCK(initproc);
845 	sx_xunlock(&proctree_lock);
846 	crfree(oldcred);
847 	cpu_fork_kthread_handler(FIRST_THREAD_IN_PROC(initproc),
848 	    start_init, NULL);
849 }
850 SYSINIT(init, SI_SUB_CREATE_INIT, SI_ORDER_FIRST, create_init, NULL);
851 
852 /*
853  * Make it runnable now.
854  */
855 static void
856 kick_init(const void *udata __unused)
857 {
858 	struct thread *td;
859 
860 	td = FIRST_THREAD_IN_PROC(initproc);
861 	thread_lock(td);
862 	TD_SET_CAN_RUN(td);
863 	sched_add(td, SRQ_BORING);
864 	thread_unlock(td);
865 }
866 SYSINIT(kickinit, SI_SUB_KTHREAD_INIT, SI_ORDER_MIDDLE, kick_init, NULL);
867