Home | History | Annotate | Line # | Download | only in kern
init_main.c revision 1.94
      1 /*	$NetBSD: init_main.c,v 1.94 1997/01/31 00:50:38 mouse Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1995 Christopher G. Demetriou.  All rights reserved.
      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/param.h>
     45 #include <sys/filedesc.h>
     46 #include <sys/errno.h>
     47 #include <sys/exec.h>
     48 #include <sys/kernel.h>
     49 #include <sys/mount.h>
     50 #include <sys/map.h>
     51 #include <sys/proc.h>
     52 #include <sys/resourcevar.h>
     53 #include <sys/signalvar.h>
     54 #include <sys/systm.h>
     55 #include <sys/vnode.h>
     56 #include <sys/tty.h>
     57 #include <sys/conf.h>
     58 #include <sys/buf.h>
     59 #ifdef REAL_CLISTS
     60 #include <sys/clist.h>
     61 #endif
     62 #include <sys/device.h>
     63 #include <sys/protosw.h>
     64 #include <sys/reboot.h>
     65 #include <sys/user.h>
     66 #ifdef SYSVSHM
     67 #include <sys/shm.h>
     68 #endif
     69 #ifdef SYSVSEM
     70 #include <sys/sem.h>
     71 #endif
     72 #ifdef SYSVMSG
     73 #include <sys/msg.h>
     74 #endif
     75 #include <sys/domain.h>
     76 #include <sys/mbuf.h>
     77 #include <sys/namei.h>
     78 
     79 #include <sys/syscall.h>
     80 #include <sys/syscallargs.h>
     81 
     82 #include <ufs/ufs/quota.h>
     83 
     84 #include <machine/cpu.h>
     85 
     86 #include <vm/vm.h>
     87 #include <vm/vm_pageout.h>
     88 
     89 #include <net/if.h>
     90 #include <net/raw_cb.h>
     91 
     92 char	copyright[] =
     93 "Copyright (c) 1982, 1986, 1989, 1991, 1993\n    The Regents of the University of California.  All rights reserved.\n\n";
     94 
     95 /* Components of the first process -- never freed. */
     96 struct	session session0;
     97 struct	pgrp pgrp0;
     98 struct	proc proc0;
     99 struct	pcred cred0;
    100 struct	filedesc0 filedesc0;
    101 struct	plimit limit0;
    102 struct	vmspace vmspace0;
    103 struct	proc *curproc = &proc0;
    104 struct	proc *initproc, *pageproc;
    105 
    106 int	cmask = CMASK;
    107 extern	struct user *proc0paddr;
    108 
    109 struct	vnode *rootvp, *swapdev_vp;
    110 int	boothowto;
    111 struct	timeval boottime;
    112 struct	timeval runtime;
    113 
    114 static void start_init __P((struct proc *));
    115 static void start_pagedaemon __P((struct proc *));
    116 void main __P((void *));
    117 
    118 #ifdef cpu_set_init_frame
    119 void *initframep;				/* XXX should go away */
    120 #endif
    121 
    122 extern char sigcode[], esigcode[];
    123 #ifdef SYSCALL_DEBUG
    124 extern char *syscallnames[];
    125 #endif
    126 
    127 struct emul emul_netbsd = {
    128 	"netbsd",
    129 	NULL,
    130 	sendsig,
    131 	SYS_syscall,
    132 	SYS_MAXSYSCALL,
    133 	sysent,
    134 #ifdef SYSCALL_DEBUG
    135 	syscallnames,
    136 #else
    137 	NULL,
    138 #endif
    139 	0,
    140 	copyargs,
    141 	setregs,
    142 	sigcode,
    143 	esigcode,
    144 };
    145 
    146 /*
    147  * System startup; initialize the world, create process 0, mount root
    148  * filesystem, and fork to create init and pagedaemon.  Most of the
    149  * hard work is done in the lower-level initialization routines including
    150  * startup(), which does memory initialization and autoconfiguration.
    151  */
    152 void
    153 main(framep)
    154 	void *framep;				/* XXX should go away */
    155 {
    156 	register struct proc *p;
    157 	register struct pdevinit *pdev;
    158 	register int i;
    159 	int s;
    160 	register_t rval[2];
    161 	extern int (*mountroot) __P((void));
    162 	extern struct pdevinit pdevinit[];
    163 	extern void roundrobin __P((void *));
    164 	extern void schedcpu __P((void *));
    165 	extern void disk_init __P((void));
    166 #if defined(NFSSERVER) || defined(NFSCLIENT)
    167 	extern void nfs_init __P((void));
    168 #endif
    169 
    170 	/*
    171 	 * Initialize the current process pointer (curproc) before
    172 	 * any possible traps/probes to simplify trap processing.
    173 	 */
    174 	p = &proc0;
    175 	curproc = p;
    176 	/*
    177 	 * Attempt to find console and initialize
    178 	 * in case of early panic or other messages.
    179 	 */
    180 	consinit();
    181 	printf(copyright);
    182 
    183 	vm_mem_init();
    184 	kmeminit();
    185 	disk_init();		/* must come before autoconfiguration */
    186 	tty_init();		/* initialise tty list */
    187 	config_init();		/* init autoconfiguration data structures */
    188 	cpu_startup();
    189 
    190 	/*
    191 	 * Initialize process and pgrp structures.
    192 	 */
    193 	procinit();
    194 
    195 	/*
    196 	 * Create process 0 (the swapper).
    197 	 */
    198 	LIST_INSERT_HEAD(&allproc, p, p_list);
    199 	p->p_pgrp = &pgrp0;
    200 	LIST_INSERT_HEAD(PGRPHASH(0), &pgrp0, pg_hash);
    201 	LIST_INIT(&pgrp0.pg_members);
    202 	LIST_INSERT_HEAD(&pgrp0.pg_members, p, p_pglist);
    203 
    204 	pgrp0.pg_session = &session0;
    205 	session0.s_count = 1;
    206 	session0.s_leader = p;
    207 
    208 	p->p_flag = P_INMEM | P_SYSTEM;
    209 	p->p_stat = SRUN;
    210 	p->p_nice = NZERO;
    211 	p->p_emul = &emul_netbsd;
    212 	bcopy("swapper", p->p_comm, sizeof ("swapper"));
    213 
    214 	/* Create credentials. */
    215 	cred0.p_refcnt = 1;
    216 	p->p_cred = &cred0;
    217 	p->p_ucred = crget();
    218 	p->p_ucred->cr_ngroups = 1;	/* group 0 */
    219 
    220 	/* Create the file descriptor table. */
    221 	p->p_fd = &filedesc0.fd_fd;
    222 	filedesc0.fd_fd.fd_refcnt = 1;
    223 	filedesc0.fd_fd.fd_cmask = cmask;
    224 	filedesc0.fd_fd.fd_ofiles = filedesc0.fd_dfiles;
    225 	filedesc0.fd_fd.fd_ofileflags = filedesc0.fd_dfileflags;
    226 	filedesc0.fd_fd.fd_nfiles = NDFILE;
    227 
    228 	/* Create the limits structures. */
    229 	p->p_limit = &limit0;
    230 	for (i = 0; i < sizeof(p->p_rlimit)/sizeof(p->p_rlimit[0]); i++)
    231 		limit0.pl_rlimit[i].rlim_cur =
    232 		    limit0.pl_rlimit[i].rlim_max = RLIM_INFINITY;
    233 	limit0.pl_rlimit[RLIMIT_NOFILE].rlim_cur = NOFILE;
    234 	limit0.pl_rlimit[RLIMIT_NPROC].rlim_cur = MAXUPRC;
    235 	i = ptoa(cnt.v_free_count);
    236 	limit0.pl_rlimit[RLIMIT_RSS].rlim_max = i;
    237 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_max = i;
    238 	limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = i / 3;
    239 	limit0.p_refcnt = 1;
    240 
    241 	/* Allocate a prototype map so we have something to fork. */
    242 	p->p_vmspace = &vmspace0;
    243 	vmspace0.vm_refcnt = 1;
    244 	pmap_pinit(&vmspace0.vm_pmap);
    245 	vm_map_init(&p->p_vmspace->vm_map, round_page(VM_MIN_ADDRESS),
    246 	    trunc_page(VM_MAX_ADDRESS), TRUE);
    247 	vmspace0.vm_map.pmap = &vmspace0.vm_pmap;
    248 	p->p_addr = proc0paddr;				/* XXX */
    249 
    250 	/*
    251 	 * We continue to place resource usage info and signal
    252 	 * actions in the user struct so they're pageable.
    253 	 */
    254 	p->p_stats = &p->p_addr->u_stats;
    255 	p->p_sigacts = &p->p_addr->u_sigacts;
    256 
    257 	/*
    258 	 * Charge root for one process.
    259 	 */
    260 	(void)chgproccnt(0, 1);
    261 
    262 	rqinit();
    263 
    264 	/* Configure virtual memory system, set vm rlimits. */
    265 	vm_init_limits(p);
    266 
    267 	/* Initialize the file systems. */
    268 #if defined(NFSSERVER) || defined(NFSCLIENT)
    269 	nfs_init();			/* initialize server/shared data */
    270 #endif
    271 	vfsinit();
    272 
    273 	/* Start real time and statistics clocks. */
    274 	initclocks();
    275 
    276 	/* Initialize mbuf's. */
    277 	mbinit();
    278 
    279 #ifdef REAL_CLISTS
    280 	/* Initialize clists. */
    281 	clist_init();
    282 #endif
    283 
    284 #ifdef SYSVSHM
    285 	/* Initialize System V style shared memory. */
    286 	shminit();
    287 #endif
    288 
    289 #ifdef SYSVSEM
    290 	/* Initialize System V style semaphores. */
    291 	seminit();
    292 #endif
    293 
    294 #ifdef SYSVMSG
    295 	/* Initialize System V style message queues. */
    296 	msginit();
    297 #endif
    298 
    299 	/* Attach pseudo-devices. */
    300 	for (pdev = pdevinit; pdev->pdev_attach != NULL; pdev++)
    301 		(*pdev->pdev_attach)(pdev->pdev_count);
    302 
    303 	/*
    304 	 * Initialize protocols.  Block reception of incoming packets
    305 	 * until everything is ready.
    306 	 */
    307 	s = splimp();
    308 	ifinit();
    309 	domaininit();
    310 	splx(s);
    311 
    312 #ifdef GPROF
    313 	/* Initialize kernel profiling. */
    314 	kmstartup();
    315 #endif
    316 
    317 	/* Kick off timeout driven events by calling first time. */
    318 	roundrobin(NULL);
    319 	schedcpu(NULL);
    320 
    321 	/* Mount the root file system. */
    322 	if ((*mountroot)())
    323 		panic("cannot mount root");
    324 	mountlist.cqh_first->mnt_flag |= MNT_ROOTFS;
    325 	mountlist.cqh_first->mnt_op->vfs_refcount++;
    326 
    327 	/* Get the vnode for '/'.  Set filedesc0.fd_fd.fd_cdir to reference it. */
    328 	if (VFS_ROOT(mountlist.cqh_first, &rootvnode))
    329 		panic("cannot find root vnode");
    330 	filedesc0.fd_fd.fd_cdir = rootvnode;
    331 	VREF(filedesc0.fd_fd.fd_cdir);
    332 	VOP_UNLOCK(rootvnode);
    333 	filedesc0.fd_fd.fd_rdir = NULL;
    334 	swapinit();
    335 
    336 	/*
    337 	 * Now can look at time, having had a chance to verify the time
    338 	 * from the file system.  Reset p->p_rtime as it may have been
    339 	 * munched in mi_switch() after the time got set.
    340 	 */
    341 	p->p_stats->p_start = runtime = mono_time = boottime = time;
    342 	p->p_rtime.tv_sec = p->p_rtime.tv_usec = 0;
    343 
    344 	/* Initialize signal state for process 0. */
    345 	siginit(p);
    346 
    347 	/* Create process 1 (init(8)). */
    348 	if (sys_fork(p, NULL, rval))
    349 		panic("fork init");
    350 #ifdef cpu_set_init_frame			/* XXX should go away */
    351 	if (rval[1]) {
    352 		/*
    353 		 * Now in process 1.
    354 		 */
    355 		initframep = framep;
    356 		start_init(curproc);
    357 		return;
    358 	}
    359 #else
    360 	cpu_set_kpc(pfind(1), start_init);
    361 #endif
    362 
    363 	/* Create process 2 (the pageout daemon). */
    364 	if (sys_fork(p, NULL, rval))
    365 		panic("fork pager");
    366 #ifdef cpu_set_init_frame			/* XXX should go away */
    367 	if (rval[1]) {
    368 		/*
    369 		 * Now in process 2.
    370 		 */
    371 		start_pagedaemon(curproc);
    372 	}
    373 #else
    374 	cpu_set_kpc(pfind(2), start_pagedaemon);
    375 #endif
    376 
    377 	/* The scheduler is an infinite loop. */
    378 	scheduler();
    379 	/* NOTREACHED */
    380 }
    381 
    382 static void
    383 check_console(p)
    384 	struct proc    *p
    385 {
    386 	struct nameidata nd;
    387 	int error;
    388 
    389 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/dev/console", p);
    390 	error = namei(&nd);
    391 	if (error)
    392 		if (error == ENOENT)
    393 			printf("warning: no /dev/console\n");
    394 		else
    395 			printf("warning: lookup /dev/console: error %d\n", error);
    396 	else
    397 		vrele(nd.ni_vp);
    398 }
    399 
    400 /*
    401  * List of paths to try when searching for "init".
    402  */
    403 static char *initpaths[] = {
    404 	"/sbin/init",
    405 	"/sbin/oinit",
    406 	"/sbin/init.bak",
    407 	NULL,
    408 };
    409 
    410 /*
    411  * Start the initial user process; try exec'ing each pathname in "initpaths".
    412  * The program is invoked with one argument containing the boot flags.
    413  */
    414 static void
    415 start_init(p)
    416 	struct proc *p;
    417 {
    418 	vm_offset_t addr;
    419 	struct sys_execve_args /* {
    420 		syscallarg(char *) path;
    421 		syscallarg(char * const *) argp;
    422 		syscallarg(char * const *) envp;
    423 	} */ args;
    424 	int options, i, error;
    425 	register_t retval[2];
    426 	char flags[4], *flagsp;
    427 	char **pathp, *path, *ucp, **uap, *arg0, *arg1 = NULL;
    428 
    429 	/*
    430 	 * Now in process 1.
    431 	 */
    432 	initproc = p;
    433 
    434 #ifdef cpu_set_init_frame			/* XXX should go away */
    435 	/*
    436 	 * We need to set the system call frame as if we were entered through
    437 	 * a syscall() so that when we call sys_execve() below, it will be able
    438 	 * to set the entry point (see setregs) when it tries to exec.  The
    439 	 * startup code in "locore.s" has allocated space for the frame and
    440 	 * passed a pointer to that space as main's argument.
    441 	 */
    442 	cpu_set_init_frame(p, initframep);
    443 #endif
    444 
    445 	/*
    446 	 * This is not the right way to do this.  We really should
    447 	 * hand-craft a descriptor onto /dev/console to hand to init,
    448 	 * but that's a _lot_ more work, and the benefit from this easy
    449 	 * hack makes up for the "good is the enemy of the best" effect.
    450 	 */
    451 	check_console(p);
    452 
    453 	/*
    454 	 * Need just enough stack to hold the faked-up "execve()" arguments.
    455 	 */
    456 	addr = USRSTACK - PAGE_SIZE;
    457 	if (vm_allocate(&p->p_vmspace->vm_map, &addr, (vm_size_t)PAGE_SIZE,
    458 	    FALSE) != 0)
    459 		panic("init: couldn't allocate argument space");
    460 	p->p_vmspace->vm_maxsaddr = (caddr_t)addr;
    461 
    462 	for (pathp = &initpaths[0]; (path = *pathp) != NULL; pathp++) {
    463 		ucp = (char *)(addr + PAGE_SIZE);
    464 
    465 		/*
    466 		 * Construct the boot flag argument.
    467 		 */
    468 		flagsp = flags;
    469 		*flagsp++ = '-';
    470 		options = 0;
    471 
    472 		if (boothowto & RB_SINGLE) {
    473 			*flagsp++ = 's';
    474 			options = 1;
    475 		}
    476 #ifdef notyet
    477 		if (boothowto & RB_FASTBOOT) {
    478 			*flagsp++ = 'f';
    479 			options = 1;
    480 		}
    481 #endif
    482 
    483 		/*
    484 		 * Move out the flags (arg 1), if necessary.
    485 		 */
    486 		if (options != 0) {
    487 			*flagsp++ = '\0';
    488 			i = flagsp - flags;
    489 #ifdef DEBUG
    490 			printf("init: copying out flags `%s' %d\n", flags, i);
    491 #endif
    492 			(void)copyout((caddr_t)flags, (caddr_t)(ucp -= i), i);
    493 			arg1 = ucp;
    494 		}
    495 
    496 		/*
    497 		 * Move out the file name (also arg 0).
    498 		 */
    499 		i = strlen(path) + 1;
    500 #ifdef DEBUG
    501 		printf("init: copying out path `%s' %d\n", path, i);
    502 #endif
    503 		(void)copyout((caddr_t)path, (caddr_t)(ucp -= i), i);
    504 		arg0 = ucp;
    505 
    506 		/*
    507 		 * Move out the arg pointers.
    508 		 */
    509 		uap = (char **)((long)ucp & ~ALIGNBYTES);
    510 		(void)suword((caddr_t)--uap, 0);	/* terminator */
    511 		if (options != 0)
    512 			(void)suword((caddr_t)--uap, (long)arg1);
    513 		(void)suword((caddr_t)--uap, (long)arg0);
    514 
    515 		/*
    516 		 * Point at the arguments.
    517 		 */
    518 		SCARG(&args, path) = arg0;
    519 		SCARG(&args, argp) = uap;
    520 		SCARG(&args, envp) = NULL;
    521 
    522 		/*
    523 		 * Now try to exec the program.  If can't for any reason
    524 		 * other than it doesn't exist, complain.
    525 		 */
    526 		if ((error = sys_execve(p, &args, retval)) == 0)
    527 			return;
    528 		if (error != ENOENT)
    529 			printf("exec %s: error %d\n", path, error);
    530 	}
    531 	printf("init: not found\n");
    532 	panic("no init");
    533 }
    534 
    535 static void
    536 start_pagedaemon(p)
    537 	struct proc *p;
    538 {
    539 
    540 	/*
    541 	 * Now in process 2.
    542 	 */
    543 	pageproc = p;
    544 	p->p_flag |= P_INMEM | P_SYSTEM;	/* XXX */
    545 	bcopy("pagedaemon", curproc->p_comm, sizeof ("pagedaemon"));
    546 	vm_pageout();
    547 	/* NOTREACHED */
    548 }
    549