Home | History | Annotate | Line # | Download | only in kern
kern_fork.c revision 1.43
      1 /*	$NetBSD: kern_fork.c,v 1.43 1998/06/25 21:17:16 thorpej Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
      5  *	The Regents of the University of California.  All rights reserved.
      6  * (c) UNIX System Laboratories, Inc.
      7  * All or some portions of this file are derived from material licensed
      8  * to the University of California by American Telephone and Telegraph
      9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     10  * the permission of UNIX System Laboratories, Inc.
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. All advertising materials mentioning features or use of this software
     21  *    must display the following acknowledgement:
     22  *	This product includes software developed by the University of
     23  *	California, Berkeley and its contributors.
     24  * 4. Neither the name of the University nor the names of its contributors
     25  *    may be used to endorse or promote products derived from this software
     26  *    without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38  * SUCH DAMAGE.
     39  *
     40  *	@(#)kern_fork.c	8.8 (Berkeley) 2/14/95
     41  */
     42 
     43 #include "opt_ktrace.h"
     44 #include "opt_uvm.h"
     45 
     46 #include <sys/param.h>
     47 #include <sys/systm.h>
     48 #include <sys/map.h>
     49 #include <sys/filedesc.h>
     50 #include <sys/kernel.h>
     51 #include <sys/malloc.h>
     52 #include <sys/mount.h>
     53 #include <sys/proc.h>
     54 #include <sys/resourcevar.h>
     55 #include <sys/vnode.h>
     56 #include <sys/file.h>
     57 #include <sys/acct.h>
     58 #include <sys/ktrace.h>
     59 #include <sys/vmmeter.h>
     60 
     61 #include <sys/syscallargs.h>
     62 
     63 #include <vm/vm.h>
     64 #include <vm/vm_kern.h>
     65 
     66 #if defined(UVM)
     67 #include <uvm/uvm_extern.h>
     68 #endif
     69 
     70 int	nprocs = 1;		/* process 0 */
     71 
     72 /*ARGSUSED*/
     73 int
     74 sys_fork(p, v, retval)
     75 	struct proc *p;
     76 	void *v;
     77 	register_t *retval;
     78 {
     79 
     80 	return (fork1(p, 0, retval, NULL));
     81 }
     82 
     83 /*
     84  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
     85  * Address space is not shared, but parent is blocked until child exit.
     86  */
     87 /*ARGSUSED*/
     88 int
     89 sys_vfork(p, v, retval)
     90 	struct proc *p;
     91 	void *v;
     92 	register_t *retval;
     93 {
     94 
     95 	return (fork1(p, FORK_PPWAIT, retval, NULL));
     96 }
     97 
     98 /*
     99  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
    100  * semantics.  Address space is shared, and parent is blocked until child exit.
    101  */
    102 /*ARGSUSED*/
    103 int
    104 sys___vfork14(p, v, retval)
    105 	struct proc *p;
    106 	void *v;
    107 	register_t *retval;
    108 {
    109 
    110 	return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, retval, NULL));
    111 }
    112 
    113 int
    114 fork1(p1, flags, retval, rnewprocp)
    115 	register struct proc *p1;
    116 	int flags;
    117 	register_t *retval;
    118 	struct proc **rnewprocp;
    119 {
    120 	register struct proc *p2;
    121 	register uid_t uid;
    122 	struct proc *newproc;
    123 	int count;
    124 	vm_offset_t uaddr;
    125 	static int nextpid, pidchecked = 0;
    126 
    127 	/*
    128 	 * Although process entries are dynamically created, we still keep
    129 	 * a global limit on the maximum number we will create.  Don't allow
    130 	 * a nonprivileged user to use the last process; don't let root
    131 	 * exceed the limit. The variable nprocs is the current number of
    132 	 * processes, maxproc is the limit.
    133 	 */
    134 	uid = p1->p_cred->p_ruid;
    135 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
    136 		tablefull("proc");
    137 		return (EAGAIN);
    138 	}
    139 
    140 	/*
    141 	 * Increment the count of procs running with this uid. Don't allow
    142 	 * a nonprivileged user to exceed their current limit.
    143 	 */
    144 	count = chgproccnt(uid, 1);
    145 	if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) {
    146 		(void)chgproccnt(uid, -1);
    147 		return (EAGAIN);
    148 	}
    149 
    150 	/*
    151 	 * Allocate virtual address space for the U-area now, while it
    152 	 * is still easy to abort the fork operation if we're out of
    153 	 * kernel virtual address space.  The actual U-area pages will
    154 	 * be allocated and wired in vm_fork().
    155 	 */
    156 #if defined(UVM)
    157 	uaddr = uvm_km_valloc(kernel_map, USPACE);
    158 #else
    159 	uaddr = kmem_alloc_pageable(kernel_map, USPACE);
    160 #endif
    161 	if (uaddr == 0) {
    162 		(void)chgproccnt(uid, -1);
    163 		return (ENOMEM);
    164 	}
    165 
    166 	/*
    167 	 * We are now committed to the fork.  From here on, we may
    168 	 * block on resources, but resource allocation may NOT fail.
    169 	 */
    170 
    171 	/* Allocate new proc. */
    172 	MALLOC(newproc, struct proc *, sizeof(struct proc), M_PROC, M_WAITOK);
    173 
    174 	/*
    175 	 * Find an unused process ID.  We remember a range of unused IDs
    176 	 * ready to use (from nextpid+1 through pidchecked-1).
    177 	 */
    178 	nextpid++;
    179 retry:
    180 	/*
    181 	 * If the process ID prototype has wrapped around,
    182 	 * restart somewhat above 0, as the low-numbered procs
    183 	 * tend to include daemons that don't exit.
    184 	 */
    185 	if (nextpid >= PID_MAX) {
    186 		nextpid = 100;
    187 		pidchecked = 0;
    188 	}
    189 	if (nextpid >= pidchecked) {
    190 		int doingzomb = 0;
    191 
    192 		pidchecked = PID_MAX;
    193 		/*
    194 		 * Scan the active and zombie procs to check whether this pid
    195 		 * is in use.  Remember the lowest pid that's greater
    196 		 * than nextpid, so we can avoid checking for a while.
    197 		 */
    198 		p2 = allproc.lh_first;
    199 again:
    200 		for (; p2 != 0; p2 = p2->p_list.le_next) {
    201 			while (p2->p_pid == nextpid ||
    202 			    p2->p_pgrp->pg_id == nextpid ||
    203 			    p2->p_session->s_sid == nextpid) {
    204 				nextpid++;
    205 				if (nextpid >= pidchecked)
    206 					goto retry;
    207 			}
    208 			if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
    209 				pidchecked = p2->p_pid;
    210 
    211 			if (p2->p_pgrp->pg_id > nextpid &&
    212 			    pidchecked > p2->p_pgrp->pg_id)
    213 				pidchecked = p2->p_pgrp->pg_id;
    214 
    215 			if (p2->p_session->s_sid > nextpid &&
    216 			    pidchecked > p2->p_session->s_sid)
    217 				pidchecked = p2->p_session->s_sid;
    218 		}
    219 		if (!doingzomb) {
    220 			doingzomb = 1;
    221 			p2 = zombproc.lh_first;
    222 			goto again;
    223 		}
    224 	}
    225 
    226 	nprocs++;
    227 	p2 = newproc;
    228 	p2->p_stat = SIDL;			/* protect against others */
    229 	p2->p_pid = nextpid;
    230 	LIST_INSERT_HEAD(&allproc, p2, p_list);
    231 	p2->p_forw = p2->p_back = NULL;		/* shouldn't be necessary */
    232 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
    233 
    234 	/*
    235 	 * Make a proc table entry for the new process.
    236 	 * Start by zeroing the section of proc that is zero-initialized,
    237 	 * then copy the section that is copied directly from the parent.
    238 	 */
    239 	bzero(&p2->p_startzero,
    240 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
    241 	bcopy(&p1->p_startcopy, &p2->p_startcopy,
    242 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
    243 
    244 	/*
    245 	 * Duplicate sub-structures as needed.
    246 	 * Increase reference counts on shared objects.
    247 	 * The p_stats and p_sigacts substructs are set in vm_fork.
    248 	 */
    249 	p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID);
    250 	p2->p_emul = p1->p_emul;
    251 	if (p1->p_flag & P_PROFIL)
    252 		startprofclock(p2);
    253 	MALLOC(p2->p_cred, struct pcred *, sizeof(struct pcred),
    254 	    M_SUBPROC, M_WAITOK);
    255 	bcopy(p1->p_cred, p2->p_cred, sizeof(*p2->p_cred));
    256 	p2->p_cred->p_refcnt = 1;
    257 	crhold(p1->p_ucred);
    258 
    259 	/* bump references to the text vnode (for procfs) */
    260 	p2->p_textvp = p1->p_textvp;
    261 	if (p2->p_textvp)
    262 		VREF(p2->p_textvp);
    263 
    264 	p2->p_fd = fdcopy(p1);
    265 	/*
    266 	 * If p_limit is still copy-on-write, bump refcnt,
    267 	 * otherwise get a copy that won't be modified.
    268 	 * (If PL_SHAREMOD is clear, the structure is shared
    269 	 * copy-on-write.)
    270 	 */
    271 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
    272 		p2->p_limit = limcopy(p1->p_limit);
    273 	else {
    274 		p2->p_limit = p1->p_limit;
    275 		p2->p_limit->p_refcnt++;
    276 	}
    277 
    278 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
    279 		p2->p_flag |= P_CONTROLT;
    280 	if (flags & FORK_PPWAIT)
    281 		p2->p_flag |= P_PPWAIT;
    282 	LIST_INSERT_AFTER(p1, p2, p_pglist);
    283 	p2->p_pptr = p1;
    284 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
    285 	LIST_INIT(&p2->p_children);
    286 
    287 #ifdef KTRACE
    288 	/*
    289 	 * Copy traceflag and tracefile if enabled.
    290 	 * If not inherited, these were zeroed above.
    291 	 */
    292 	if (p1->p_traceflag&KTRFAC_INHERIT) {
    293 		p2->p_traceflag = p1->p_traceflag;
    294 		if ((p2->p_tracep = p1->p_tracep) != NULL)
    295 			ktradref(p2);
    296 	}
    297 #endif
    298 
    299 	/*
    300 	 * This begins the section where we must prevent the parent
    301 	 * from being swapped.
    302 	 */
    303 	PHOLD(p1);
    304 
    305 	/*
    306 	 * Finish creating the child process.  It will return through a
    307 	 * different path later.
    308 	 */
    309 	p2->p_addr = (struct user *)uaddr;
    310 #if defined(UVM)
    311 	uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
    312 #else
    313 	vm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
    314 #endif
    315 
    316 	/*
    317 	 * Make child runnable, set start time, and add to run queue.
    318 	 */
    319 	(void) splstatclock();
    320 	p2->p_stats->p_start = time;
    321 	p2->p_acflag = AFORK;
    322 	p2->p_stat = SRUN;
    323 	setrunqueue(p2);
    324 	(void) spl0();
    325 
    326 	/*
    327 	 * Now can be swapped.
    328 	 */
    329 	PRELE(p1);
    330 
    331 	/*
    332 	 * Update stats now that we know the fork was successful.
    333 	 */
    334 #if defined(UVM)
    335 	uvmexp.forks++;
    336 	if (flags & FORK_PPWAIT)
    337 		uvmexp.forks_ppwait++;
    338 	if (flags & FORK_SHAREVM)
    339 		uvmexp.forks_sharevm++;
    340 #else
    341 	cnt.v_forks++;
    342 	if (flags & FORK_PPWAIT)
    343 		cnt.v_forks_ppwait++;
    344 	if (flags & FORK_SHAREVM)
    345 		cnt.v_forks_sharevm++;
    346 #endif
    347 
    348 	/*
    349 	 * Pass a pointer to the new process to the caller.
    350 	 */
    351 	if (rnewprocp != NULL)
    352 		*rnewprocp = p2;
    353 
    354 	/*
    355 	 * Preserve synchronization semantics of vfork.  If waiting for
    356 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
    357 	 * proc (in case of exit).
    358 	 */
    359 	if (flags & FORK_PPWAIT)
    360 		while (p2->p_flag & P_PPWAIT)
    361 			tsleep(p1, PWAIT, "ppwait", 0);
    362 
    363 	/*
    364 	 * Return child pid to parent process,
    365 	 * marking us as parent via retval[1].
    366 	 */
    367 	if (retval != NULL) {
    368 		retval[0] = p2->p_pid;
    369 		retval[1] = 0;
    370 	}
    371 	return (0);
    372 }
    373