Home | History | Annotate | Line # | Download | only in kern
kern_fork.c revision 1.46
      1 /*	$NetBSD: kern_fork.c,v 1.46 1998/08/13 02:10:57 eeh 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/pool.h>
     53 #include <sys/mount.h>
     54 #include <sys/proc.h>
     55 #include <sys/resourcevar.h>
     56 #include <sys/vnode.h>
     57 #include <sys/file.h>
     58 #include <sys/acct.h>
     59 #include <sys/ktrace.h>
     60 #include <sys/vmmeter.h>
     61 
     62 #include <sys/syscallargs.h>
     63 
     64 #include <vm/vm.h>
     65 #include <vm/vm_kern.h>
     66 
     67 #if defined(UVM)
     68 #include <uvm/uvm_extern.h>
     69 #endif
     70 
     71 int	nprocs = 1;		/* process 0 */
     72 
     73 /*ARGSUSED*/
     74 int
     75 sys_fork(p, v, retval)
     76 	struct proc *p;
     77 	void *v;
     78 	register_t *retval;
     79 {
     80 
     81 	return (fork1(p, 0, retval, NULL));
     82 }
     83 
     84 /*
     85  * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM).
     86  * Address space is not shared, but parent is blocked until child exit.
     87  */
     88 /*ARGSUSED*/
     89 int
     90 sys_vfork(p, v, retval)
     91 	struct proc *p;
     92 	void *v;
     93 	register_t *retval;
     94 {
     95 
     96 	return (fork1(p, FORK_PPWAIT, retval, NULL));
     97 }
     98 
     99 /*
    100  * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2)
    101  * semantics.  Address space is shared, and parent is blocked until child exit.
    102  */
    103 /*ARGSUSED*/
    104 int
    105 sys___vfork14(p, v, retval)
    106 	struct proc *p;
    107 	void *v;
    108 	register_t *retval;
    109 {
    110 
    111 	return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, retval, NULL));
    112 }
    113 
    114 int
    115 fork1(p1, flags, retval, rnewprocp)
    116 	register struct proc *p1;
    117 	int flags;
    118 	register_t *retval;
    119 	struct proc **rnewprocp;
    120 {
    121 	register struct proc *p2;
    122 	register uid_t uid;
    123 	struct proc *newproc;
    124 	int count;
    125 	vaddr_t uaddr;
    126 	static int nextpid, pidchecked = 0;
    127 
    128 	/*
    129 	 * Although process entries are dynamically created, we still keep
    130 	 * a global limit on the maximum number we will create.  Don't allow
    131 	 * a nonprivileged user to use the last process; don't let root
    132 	 * exceed the limit. The variable nprocs is the current number of
    133 	 * processes, maxproc is the limit.
    134 	 */
    135 	uid = p1->p_cred->p_ruid;
    136 	if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) {
    137 		tablefull("proc");
    138 		return (EAGAIN);
    139 	}
    140 
    141 	/*
    142 	 * Increment the count of procs running with this uid. Don't allow
    143 	 * a nonprivileged user to exceed their current limit.
    144 	 */
    145 	count = chgproccnt(uid, 1);
    146 	if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) {
    147 		(void)chgproccnt(uid, -1);
    148 		return (EAGAIN);
    149 	}
    150 
    151 	/*
    152 	 * Allocate virtual address space for the U-area now, while it
    153 	 * is still easy to abort the fork operation if we're out of
    154 	 * kernel virtual address space.  The actual U-area pages will
    155 	 * be allocated and wired in vm_fork().
    156 	 */
    157 #if defined(UVM)
    158 	uaddr = uvm_km_valloc(kernel_map, USPACE);
    159 #else
    160 	uaddr = kmem_alloc_pageable(kernel_map, USPACE);
    161 #endif
    162 	if (uaddr == 0) {
    163 		(void)chgproccnt(uid, -1);
    164 		return (ENOMEM);
    165 	}
    166 
    167 	/*
    168 	 * We are now committed to the fork.  From here on, we may
    169 	 * block on resources, but resource allocation may NOT fail.
    170 	 */
    171 
    172 	/* Allocate new proc. */
    173 	newproc = pool_get(&proc_pool, PR_WAITOK);
    174 
    175 	/*
    176 	 * Find an unused process ID.  We remember a range of unused IDs
    177 	 * ready to use (from nextpid+1 through pidchecked-1).
    178 	 */
    179 	nextpid++;
    180 retry:
    181 	/*
    182 	 * If the process ID prototype has wrapped around,
    183 	 * restart somewhat above 0, as the low-numbered procs
    184 	 * tend to include daemons that don't exit.
    185 	 */
    186 	if (nextpid >= PID_MAX) {
    187 		nextpid = 100;
    188 		pidchecked = 0;
    189 	}
    190 	if (nextpid >= pidchecked) {
    191 		int doingzomb = 0;
    192 
    193 		pidchecked = PID_MAX;
    194 		/*
    195 		 * Scan the active and zombie procs to check whether this pid
    196 		 * is in use.  Remember the lowest pid that's greater
    197 		 * than nextpid, so we can avoid checking for a while.
    198 		 */
    199 		p2 = allproc.lh_first;
    200 again:
    201 		for (; p2 != 0; p2 = p2->p_list.le_next) {
    202 			while (p2->p_pid == nextpid ||
    203 			    p2->p_pgrp->pg_id == nextpid ||
    204 			    p2->p_session->s_sid == nextpid) {
    205 				nextpid++;
    206 				if (nextpid >= pidchecked)
    207 					goto retry;
    208 			}
    209 			if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
    210 				pidchecked = p2->p_pid;
    211 
    212 			if (p2->p_pgrp->pg_id > nextpid &&
    213 			    pidchecked > p2->p_pgrp->pg_id)
    214 				pidchecked = p2->p_pgrp->pg_id;
    215 
    216 			if (p2->p_session->s_sid > nextpid &&
    217 			    pidchecked > p2->p_session->s_sid)
    218 				pidchecked = p2->p_session->s_sid;
    219 		}
    220 		if (!doingzomb) {
    221 			doingzomb = 1;
    222 			p2 = zombproc.lh_first;
    223 			goto again;
    224 		}
    225 	}
    226 
    227 	nprocs++;
    228 	p2 = newproc;
    229 	p2->p_stat = SIDL;			/* protect against others */
    230 	p2->p_pid = nextpid;
    231 	LIST_INSERT_HEAD(&allproc, p2, p_list);
    232 	p2->p_forw = p2->p_back = NULL;		/* shouldn't be necessary */
    233 	LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
    234 
    235 	/*
    236 	 * Make a proc table entry for the new process.
    237 	 * Start by zeroing the section of proc that is zero-initialized,
    238 	 * then copy the section that is copied directly from the parent.
    239 	 */
    240 	memset(&p2->p_startzero, 0,
    241 	    (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
    242 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
    243 	    (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
    244 
    245 	/*
    246 	 * Duplicate sub-structures as needed.
    247 	 * Increase reference counts on shared objects.
    248 	 * The p_stats and p_sigacts substructs are set in vm_fork.
    249 	 */
    250 	p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID);
    251 	p2->p_emul = p1->p_emul;
    252 	if (p1->p_flag & P_PROFIL)
    253 		startprofclock(p2);
    254 	MALLOC(p2->p_cred, struct pcred *, sizeof(struct pcred),
    255 	    M_SUBPROC, M_WAITOK);
    256 	memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred));
    257 	p2->p_cred->p_refcnt = 1;
    258 	crhold(p1->p_ucred);
    259 
    260 	/* bump references to the text vnode (for procfs) */
    261 	p2->p_textvp = p1->p_textvp;
    262 	if (p2->p_textvp)
    263 		VREF(p2->p_textvp);
    264 
    265 	p2->p_fd = fdcopy(p1);
    266 	/*
    267 	 * If p_limit is still copy-on-write, bump refcnt,
    268 	 * otherwise get a copy that won't be modified.
    269 	 * (If PL_SHAREMOD is clear, the structure is shared
    270 	 * copy-on-write.)
    271 	 */
    272 	if (p1->p_limit->p_lflags & PL_SHAREMOD)
    273 		p2->p_limit = limcopy(p1->p_limit);
    274 	else {
    275 		p2->p_limit = p1->p_limit;
    276 		p2->p_limit->p_refcnt++;
    277 	}
    278 
    279 	if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
    280 		p2->p_flag |= P_CONTROLT;
    281 	if (flags & FORK_PPWAIT)
    282 		p2->p_flag |= P_PPWAIT;
    283 	LIST_INSERT_AFTER(p1, p2, p_pglist);
    284 	p2->p_pptr = p1;
    285 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
    286 	LIST_INIT(&p2->p_children);
    287 
    288 #ifdef KTRACE
    289 	/*
    290 	 * Copy traceflag and tracefile if enabled.
    291 	 * If not inherited, these were zeroed above.
    292 	 */
    293 	if (p1->p_traceflag&KTRFAC_INHERIT) {
    294 		p2->p_traceflag = p1->p_traceflag;
    295 		if ((p2->p_tracep = p1->p_tracep) != NULL)
    296 			ktradref(p2);
    297 	}
    298 #endif
    299 
    300 	/*
    301 	 * This begins the section where we must prevent the parent
    302 	 * from being swapped.
    303 	 */
    304 	PHOLD(p1);
    305 
    306 	/*
    307 	 * Finish creating the child process.  It will return through a
    308 	 * different path later.
    309 	 */
    310 	p2->p_addr = (struct user *)uaddr;
    311 #if defined(UVM)
    312 	uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
    313 #else
    314 	vm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE);
    315 #endif
    316 
    317 	/*
    318 	 * Make child runnable, set start time, and add to run queue.
    319 	 */
    320 	(void) splstatclock();
    321 	p2->p_stats->p_start = time;
    322 	p2->p_acflag = AFORK;
    323 	p2->p_stat = SRUN;
    324 	setrunqueue(p2);
    325 	(void) spl0();
    326 
    327 	/*
    328 	 * Now can be swapped.
    329 	 */
    330 	PRELE(p1);
    331 
    332 	/*
    333 	 * Update stats now that we know the fork was successful.
    334 	 */
    335 #if defined(UVM)
    336 	uvmexp.forks++;
    337 	if (flags & FORK_PPWAIT)
    338 		uvmexp.forks_ppwait++;
    339 	if (flags & FORK_SHAREVM)
    340 		uvmexp.forks_sharevm++;
    341 #else
    342 	cnt.v_forks++;
    343 	if (flags & FORK_PPWAIT)
    344 		cnt.v_forks_ppwait++;
    345 	if (flags & FORK_SHAREVM)
    346 		cnt.v_forks_sharevm++;
    347 #endif
    348 
    349 	/*
    350 	 * Pass a pointer to the new process to the caller.
    351 	 */
    352 	if (rnewprocp != NULL)
    353 		*rnewprocp = p2;
    354 
    355 	/*
    356 	 * Preserve synchronization semantics of vfork.  If waiting for
    357 	 * child to exec or exit, set P_PPWAIT on child, and sleep on our
    358 	 * proc (in case of exit).
    359 	 */
    360 	if (flags & FORK_PPWAIT)
    361 		while (p2->p_flag & P_PPWAIT)
    362 			tsleep(p1, PWAIT, "ppwait", 0);
    363 
    364 	/*
    365 	 * Return child pid to parent process,
    366 	 * marking us as parent via retval[1].
    367 	 */
    368 	if (retval != NULL) {
    369 		retval[0] = p2->p_pid;
    370 		retval[1] = 0;
    371 	}
    372 	return (0);
    373 }
    374