2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
39 * $FreeBSD: src/sys/kern/kern_fork.c,v 1.72.2.13 2003/06/06 20:21:32 tegge Exp $
40 * $DragonFly: src/sys/kern/kern_fork.c,v 1.12 2003/07/19 21:14:38 dillon Exp $
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
53 #include <sys/resourcevar.h>
54 #include <sys/vnode.h>
56 #include <sys/ktrace.h>
57 #include <sys/unistd.h>
63 #include <vm/vm_map.h>
64 #include <vm/vm_extern.h>
65 #include <vm/vm_zone.h>
67 #include <sys/vmmeter.h>
70 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
73 * These are the stuctures used to create a callout list for things to do
74 * when forking a process
78 TAILQ_ENTRY(forklist) next;
81 TAILQ_HEAD(forklist_head, forklist);
82 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
84 #ifndef _SYS_SYSPROTO_H_
90 int forksleep; /* Place for fork1() to sleep on. */
94 fork(struct fork_args *uap)
96 struct proc *p = curproc;
100 error = fork1(p, RFFDG | RFPROC, &p2);
102 start_forked_proc(p, p2);
103 p->p_retval[0] = p2->p_pid;
111 vfork(struct vfork_args *uap)
113 struct proc *p = curproc;
117 error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
119 start_forked_proc(p, p2);
120 p->p_retval[0] = p2->p_pid;
127 rfork(struct rfork_args *uap)
129 struct proc *p = curproc;
133 error = fork1(p, uap->flags, &p2);
135 start_forked_proc(p, p2);
136 p->p_retval[0] = p2 ? p2->p_pid : 0;
143 int nprocs = 1; /* process 0 */
144 static int nextpid = 0;
147 * Random component to nextpid generation. We mix in a random factor to make
148 * it a little harder to predict. We sanity check the modulus value to avoid
149 * doing it in critical paths. Don't let it be too small or we pointlessly
150 * waste randomness entropy, and don't let it be impossibly large. Using a
151 * modulus that is too big causes a LOT more process table scans and slows
152 * down fork processing as the pidchecked caching is defeated.
154 static int randompid = 0;
157 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
162 error = sysctl_handle_int(oidp, &pid, 0, req);
163 if (error || !req->newptr)
165 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
167 else if (pid < 2) /* NOP */
169 else if (pid < 100) /* Make it reasonable */
175 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
176 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
179 fork1(p1, flags, procp)
184 struct proc *p2, *pptr;
186 struct proc *newproc;
188 static int pidchecked = 0;
190 struct filedesc_to_leader *fdtol;
192 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
196 * Here we don't create a new process, but we divorce
197 * certain parts of a process from itself.
199 if ((flags & RFPROC) == 0) {
201 vm_fork(p1, 0, flags);
204 * Close all file descriptors.
206 if (flags & RFCFDG) {
207 struct filedesc *fdtmp;
214 * Unshare file descriptors (from parent.)
217 if (p1->p_fd->fd_refcnt > 1) {
218 struct filedesc *newfd;
229 * Although process entries are dynamically created, we still keep
230 * a global limit on the maximum number we will create. Don't allow
231 * a nonprivileged user to use the last ten processes; don't let root
232 * exceed the limit. The variable nprocs is the current number of
233 * processes, maxproc is the limit.
235 uid = p1->p_ucred->cr_ruid;
236 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
237 tsleep(&forksleep, 0, "fork", hz / 2);
241 * Increment the nprocs resource before blocking can occur. There
242 * are hard-limits as to the number of processes that can run.
247 * Increment the count of procs running with this uid. Don't allow
248 * a nonprivileged user to exceed their current limit.
250 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
251 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
254 * Back out the process count
257 tsleep(&forksleep, 0, "fork", hz / 2);
261 /* Allocate new proc. */
262 newproc = zalloc(proc_zone);
265 * Setup linkage for kernel based threading
267 if((flags & RFTHREAD) != 0) {
268 newproc->p_peers = p1->p_peers;
269 p1->p_peers = newproc;
270 newproc->p_leader = p1->p_leader;
272 newproc->p_peers = 0;
273 newproc->p_leader = newproc;
276 newproc->p_wakeup = 0;
277 newproc->p_vmspace = NULL;
280 * Find an unused process ID. We remember a range of unused IDs
281 * ready to use (from nextpid+1 through pidchecked-1).
285 nextpid += arc4random() % randompid;
288 * If the process ID prototype has wrapped around,
289 * restart somewhat above 0, as the low-numbered procs
290 * tend to include daemons that don't exit.
292 if (nextpid >= PID_MAX) {
293 nextpid = nextpid % PID_MAX;
298 if (nextpid >= pidchecked) {
301 pidchecked = PID_MAX;
303 * Scan the active and zombie procs to check whether this pid
304 * is in use. Remember the lowest pid that's greater
305 * than nextpid, so we can avoid checking for a while.
307 p2 = LIST_FIRST(&allproc);
309 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) {
310 while (p2->p_pid == nextpid ||
311 p2->p_pgrp->pg_id == nextpid ||
312 p2->p_session->s_sid == nextpid) {
314 if (nextpid >= pidchecked)
317 if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
318 pidchecked = p2->p_pid;
319 if (p2->p_pgrp->pg_id > nextpid &&
320 pidchecked > p2->p_pgrp->pg_id)
321 pidchecked = p2->p_pgrp->pg_id;
322 if (p2->p_session->s_sid > nextpid &&
323 pidchecked > p2->p_session->s_sid)
324 pidchecked = p2->p_session->s_sid;
328 p2 = LIST_FIRST(&zombproc);
334 p2->p_stat = SIDL; /* protect against others */
336 LIST_INSERT_HEAD(&allproc, p2, p_list);
337 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
340 * Make a proc table entry for the new process.
341 * Start by zeroing the section of proc that is zero-initialized,
342 * then copy the section that is copied directly from the parent.
344 bzero(&p2->p_startzero,
345 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
346 bcopy(&p1->p_startcopy, &p2->p_startcopy,
347 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
349 p2->p_aioinfo = NULL;
352 * Duplicate sub-structures as needed.
353 * Increase reference counts on shared objects.
354 * The p_stats and p_sigacts substructs are set in vm_fork.
356 * P_CP_RELEASED indicates that the process is starting out in
357 * the kernel (in the fork trampoline). The flag will be converted
358 * to P_CURPROC when the new process calls userret() and attempts
359 * to return to userland
361 p2->p_flag = P_INMEM | P_CP_RELEASED;
362 if (p1->p_flag & P_PROFIL)
364 p2->p_ucred = crhold(p1->p_ucred);
366 if (p2->p_ucred->cr_prison) {
367 p2->p_ucred->cr_prison->pr_ref++;
368 p2->p_flag |= P_JAILED;
372 p2->p_args->ar_ref++;
374 if (flags & RFSIGSHARE) {
375 p2->p_procsig = p1->p_procsig;
376 p2->p_procsig->ps_refcnt++;
377 if (p1->p_sigacts == &p1->p_addr->u_sigacts) {
378 struct sigacts *newsigacts;
381 /* Create the shared sigacts structure */
382 MALLOC(newsigacts, struct sigacts *,
383 sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
386 * Set p_sigacts to the new shared structure.
387 * Note that this is updating p1->p_sigacts at the
388 * same time, since p_sigacts is just a pointer to
389 * the shared p_procsig->ps_sigacts.
391 p2->p_sigacts = newsigacts;
392 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts,
393 sizeof(*p2->p_sigacts));
394 *p2->p_sigacts = p1->p_addr->u_sigacts;
398 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
399 M_SUBPROC, M_WAITOK);
400 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
401 p2->p_procsig->ps_refcnt = 1;
402 p2->p_sigacts = NULL; /* finished in vm_fork() */
404 if (flags & RFLINUXTHPN)
405 p2->p_sigparent = SIGUSR1;
407 p2->p_sigparent = SIGCHLD;
409 /* bump references to the text vnode (for procfs) */
410 p2->p_textvp = p1->p_textvp;
414 if (flags & RFCFDG) {
415 p2->p_fd = fdinit(p1);
417 } else if (flags & RFFDG) {
418 p2->p_fd = fdcopy(p1);
421 p2->p_fd = fdshare(p1);
422 if (p1->p_fdtol == NULL)
424 filedesc_to_leader_alloc(NULL,
426 if ((flags & RFTHREAD) != 0) {
428 * Shared file descriptor table and
429 * shared process leaders.
432 fdtol->fdl_refcount++;
435 * Shared file descriptor table, and
436 * different process leaders
438 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
445 * If p_limit is still copy-on-write, bump refcnt,
446 * otherwise get a copy that won't be modified.
447 * (If PL_SHAREMOD is clear, the structure is shared
450 if (p1->p_limit->p_lflags & PL_SHAREMOD)
451 p2->p_limit = limcopy(p1->p_limit);
453 p2->p_limit = p1->p_limit;
454 p2->p_limit->p_refcnt++;
458 * Preserve some more flags in subprocess. P_PROFIL has already
461 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
462 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
463 p2->p_flag |= P_CONTROLT;
464 if (flags & RFPPWAIT)
465 p2->p_flag |= P_PPWAIT;
467 LIST_INSERT_AFTER(p1, p2, p_pglist);
470 * Attach the new process to its parent.
472 * If RFNOWAIT is set, the newly created process becomes a child
473 * of init. This effectively disassociates the child from the
476 if (flags & RFNOWAIT)
481 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
482 LIST_INIT(&p2->p_children);
486 * Copy traceflag and tracefile if enabled. If not inherited,
487 * these were zeroed above but we still could have a trace race
488 * so make sure p2's p_tracep is NULL.
490 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) {
491 p2->p_traceflag = p1->p_traceflag;
492 if ((p2->p_tracep = p1->p_tracep) != NULL)
498 * set priority of child to be that of parent
500 p2->p_estcpu = p1->p_estcpu;
503 * This begins the section where we must prevent the parent
504 * from being swapped.
509 * Finish creating the child process. It will return via a different
510 * execution path later. (ie: directly into user mode)
512 vm_fork(p1, p2, flags);
514 if (flags == (RFFDG | RFPROC)) {
515 mycpu->gd_cnt.v_forks++;
516 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
517 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
518 mycpu->gd_cnt.v_vforks++;
519 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
520 } else if (p1 == &proc0) {
521 mycpu->gd_cnt.v_kthreads++;
522 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
524 mycpu->gd_cnt.v_rforks++;
525 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
529 * Both processes are set up, now check if any loadable modules want
530 * to adjust anything.
531 * What if they have an error? XXX
533 TAILQ_FOREACH(ep, &fork_list, next) {
534 (*ep->function)(p1, p2, flags);
538 * Make child runnable and add to run queue.
540 microtime(&(p2->p_stats->p_start));
541 p2->p_acflag = AFORK;
544 * tell any interested parties about the new process
546 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
549 * Return child proc pointer to parent.
556 * The next two functionms are general routines to handle adding/deleting
557 * items on the fork callout list.
560 * Take the arguments given and put them onto the fork callout list,
561 * However first make sure that it's not already there.
562 * Returns 0 on success or a standard error number.
567 forklist_fn function;
572 /* let the programmer know if he's been stupid */
573 if (rm_at_fork(function))
574 printf("WARNING: fork callout entry (%p) already present\n",
577 ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT);
580 ep->function = function;
581 TAILQ_INSERT_TAIL(&fork_list, ep, next);
586 * Scan the exit callout list for the given item and remove it..
587 * Returns the number of items removed (0 or 1)
592 forklist_fn function;
596 TAILQ_FOREACH(ep, &fork_list, next) {
597 if (ep->function == function) {
598 TAILQ_REMOVE(&fork_list, ep, next);
607 * Add a forked process to the run queue after any remaining setup, such
608 * as setting the fork handler, has been completed.
612 start_forked_proc(struct proc *p1, struct proc *p2)
615 * Move from SIDL to RUN queue, and activate the process's thread.
616 * Activation of the thread effectively makes the process "a"
617 * current process, so we do not setrunqueue().
619 KASSERT(p2->p_stat == SIDL,
620 ("cannot start forked process, bad status: %p", p2));
627 * Now can be swapped.
632 * Preserve synchronization semantics of vfork. If waiting for
633 * child to exec or exit, set P_PPWAIT on child, and sleep on our
634 * proc (in case of exit).
636 while (p2->p_flag & P_PPWAIT)
637 tsleep(p1, 0, "ppwait", 0);