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 1.72.2.14 2003/06/26 04:15:10 silby Exp $
40 * $DragonFly: src/sys/kern/kern_fork.c,v 1.19 2004/03/06 22:14:09 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>
64 #include <vm/vm_map.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_zone.h>
68 #include <sys/vmmeter.h>
71 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
74 * These are the stuctures used to create a callout list for things to do
75 * when forking a process
79 TAILQ_ENTRY(forklist) next;
82 TAILQ_HEAD(forklist_head, forklist);
83 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
85 int forksleep; /* Place for fork1() to sleep on. */
89 fork(struct fork_args *uap)
91 struct proc *p = curproc;
95 error = fork1(p, RFFDG | RFPROC, &p2);
97 start_forked_proc(p, p2);
98 uap->sysmsg_fds[0] = p2->p_pid;
99 uap->sysmsg_fds[1] = 0;
106 vfork(struct vfork_args *uap)
108 struct proc *p = curproc;
112 error = fork1(p, RFFDG | RFPROC | RFPPWAIT | RFMEM, &p2);
114 start_forked_proc(p, p2);
115 uap->sysmsg_fds[0] = p2->p_pid;
116 uap->sysmsg_fds[1] = 0;
122 rfork(struct rfork_args *uap)
124 struct proc *p = curproc;
128 /* Don't allow kernel only flags */
129 if ((uap->flags & RFKERNELONLY) != 0)
132 error = fork1(p, uap->flags, &p2);
134 start_forked_proc(p, p2);
135 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
136 uap->sysmsg_fds[1] = 0;
142 int nprocs = 1; /* process 0 */
143 static int nextpid = 0;
146 * Random component to nextpid generation. We mix in a random factor to make
147 * it a little harder to predict. We sanity check the modulus value to avoid
148 * doing it in critical paths. Don't let it be too small or we pointlessly
149 * waste randomness entropy, and don't let it be impossibly large. Using a
150 * modulus that is too big causes a LOT more process table scans and slows
151 * down fork processing as the pidchecked caching is defeated.
153 static int randompid = 0;
156 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
161 error = sysctl_handle_int(oidp, &pid, 0, req);
162 if (error || !req->newptr)
164 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
166 else if (pid < 2) /* NOP */
168 else if (pid < 100) /* Make it reasonable */
174 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
175 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
178 fork1(p1, flags, procp)
183 struct proc *p2, *pptr;
185 struct proc *newproc;
187 static int curfail = 0, pidchecked = 0;
188 static struct timeval lastfail;
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 if (ppsratecheck(&lastfail, &curfail, 1))
238 printf("maxproc limit exceeded by uid %d, please "
239 "see tuning(7) and login.conf(5).\n", uid);
240 tsleep(&forksleep, 0, "fork", hz / 2);
244 * Increment the nprocs resource before blocking can occur. There
245 * are hard-limits as to the number of processes that can run.
250 * Increment the count of procs running with this uid. Don't allow
251 * a nonprivileged user to exceed their current limit.
253 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
254 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
257 * Back out the process count
260 if (ppsratecheck(&lastfail, &curfail, 1))
261 printf("maxproc limit exceeded by uid %d, please "
262 "see tuning(7) and login.conf(5).\n", uid);
263 tsleep(&forksleep, 0, "fork", hz / 2);
267 /* Allocate new proc. */
268 newproc = zalloc(proc_zone);
271 * Setup linkage for kernel based threading
273 if((flags & RFTHREAD) != 0) {
274 newproc->p_peers = p1->p_peers;
275 p1->p_peers = newproc;
276 newproc->p_leader = p1->p_leader;
278 newproc->p_peers = 0;
279 newproc->p_leader = newproc;
282 newproc->p_wakeup = 0;
283 newproc->p_vmspace = NULL;
286 * Find an unused process ID. We remember a range of unused IDs
287 * ready to use (from nextpid+1 through pidchecked-1).
291 nextpid += arc4random() % randompid;
294 * If the process ID prototype has wrapped around,
295 * restart somewhat above 0, as the low-numbered procs
296 * tend to include daemons that don't exit.
298 if (nextpid >= PID_MAX) {
299 nextpid = nextpid % PID_MAX;
304 if (nextpid >= pidchecked) {
307 pidchecked = PID_MAX;
309 * Scan the active and zombie procs to check whether this pid
310 * is in use. Remember the lowest pid that's greater
311 * than nextpid, so we can avoid checking for a while.
313 p2 = LIST_FIRST(&allproc);
315 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) {
316 while (p2->p_pid == nextpid ||
317 p2->p_pgrp->pg_id == nextpid ||
318 p2->p_session->s_sid == nextpid) {
320 if (nextpid >= pidchecked)
323 if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
324 pidchecked = p2->p_pid;
325 if (p2->p_pgrp->pg_id > nextpid &&
326 pidchecked > p2->p_pgrp->pg_id)
327 pidchecked = p2->p_pgrp->pg_id;
328 if (p2->p_session->s_sid > nextpid &&
329 pidchecked > p2->p_session->s_sid)
330 pidchecked = p2->p_session->s_sid;
334 p2 = LIST_FIRST(&zombproc);
340 p2->p_stat = SIDL; /* protect against others */
342 LIST_INSERT_HEAD(&allproc, p2, p_list);
343 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
346 * Make a proc table entry for the new process.
347 * Start by zeroing the section of proc that is zero-initialized,
348 * then copy the section that is copied directly from the parent.
350 bzero(&p2->p_startzero,
351 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
352 bcopy(&p1->p_startcopy, &p2->p_startcopy,
353 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
355 p2->p_aioinfo = NULL;
358 * Duplicate sub-structures as needed.
359 * Increase reference counts on shared objects.
360 * The p_stats and p_sigacts substructs are set in vm_fork.
362 * P_CP_RELEASED indicates that the process is starting out in
363 * the kernel (in the fork trampoline). The flag will be converted
364 * to P_CURPROC when the new process calls userret() and attempts
365 * to return to userland
367 p2->p_flag = P_INMEM | P_CP_RELEASED;
368 if (p1->p_flag & P_PROFIL)
370 p2->p_ucred = crhold(p1->p_ucred);
372 if (p2->p_ucred->cr_prison) {
373 p2->p_ucred->cr_prison->pr_ref++;
374 p2->p_flag |= P_JAILED;
378 p2->p_args->ar_ref++;
380 if (flags & RFSIGSHARE) {
381 p2->p_procsig = p1->p_procsig;
382 p2->p_procsig->ps_refcnt++;
383 if (p1->p_sigacts == &p1->p_addr->u_sigacts) {
384 struct sigacts *newsigacts;
387 /* Create the shared sigacts structure */
388 MALLOC(newsigacts, struct sigacts *,
389 sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
392 * Set p_sigacts to the new shared structure.
393 * Note that this is updating p1->p_sigacts at the
394 * same time, since p_sigacts is just a pointer to
395 * the shared p_procsig->ps_sigacts.
397 p2->p_sigacts = newsigacts;
398 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts,
399 sizeof(*p2->p_sigacts));
400 *p2->p_sigacts = p1->p_addr->u_sigacts;
404 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
405 M_SUBPROC, M_WAITOK);
406 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
407 p2->p_procsig->ps_refcnt = 1;
408 p2->p_sigacts = NULL; /* finished in vm_fork() */
410 if (flags & RFLINUXTHPN)
411 p2->p_sigparent = SIGUSR1;
413 p2->p_sigparent = SIGCHLD;
415 /* bump references to the text vnode (for procfs) */
416 p2->p_textvp = p1->p_textvp;
420 if (flags & RFCFDG) {
421 p2->p_fd = fdinit(p1);
423 } else if (flags & RFFDG) {
424 p2->p_fd = fdcopy(p1);
427 p2->p_fd = fdshare(p1);
428 if (p1->p_fdtol == NULL)
430 filedesc_to_leader_alloc(NULL,
432 if ((flags & RFTHREAD) != 0) {
434 * Shared file descriptor table and
435 * shared process leaders.
438 fdtol->fdl_refcount++;
441 * Shared file descriptor table, and
442 * different process leaders
444 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
450 * If p_limit is still copy-on-write, bump refcnt,
451 * otherwise get a copy that won't be modified.
452 * (If PL_SHAREMOD is clear, the structure is shared
455 if (p1->p_limit->p_lflags & PL_SHAREMOD)
456 p2->p_limit = limcopy(p1->p_limit);
458 p2->p_limit = p1->p_limit;
459 p2->p_limit->p_refcnt++;
463 * Preserve some more flags in subprocess. P_PROFIL has already
466 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
467 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
468 p2->p_flag |= P_CONTROLT;
469 if (flags & RFPPWAIT)
470 p2->p_flag |= P_PPWAIT;
472 LIST_INSERT_AFTER(p1, p2, p_pglist);
475 * Attach the new process to its parent.
477 * If RFNOWAIT is set, the newly created process becomes a child
478 * of init. This effectively disassociates the child from the
481 if (flags & RFNOWAIT)
486 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
487 LIST_INIT(&p2->p_children);
488 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
492 * Copy traceflag and tracefile if enabled. If not inherited,
493 * these were zeroed above but we still could have a trace race
494 * so make sure p2's p_tracep is NULL.
496 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) {
497 p2->p_traceflag = p1->p_traceflag;
498 if ((p2->p_tracep = p1->p_tracep) != NULL)
504 * set priority of child to be that of parent
506 p2->p_estcpu = p1->p_estcpu;
509 * This begins the section where we must prevent the parent
510 * from being swapped.
515 * Finish creating the child process. It will return via a different
516 * execution path later. (ie: directly into user mode)
518 vm_fork(p1, p2, flags);
519 caps_fork(p1, p2, flags);
521 if (flags == (RFFDG | RFPROC)) {
522 mycpu->gd_cnt.v_forks++;
523 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
524 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
525 mycpu->gd_cnt.v_vforks++;
526 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
527 } else if (p1 == &proc0) {
528 mycpu->gd_cnt.v_kthreads++;
529 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
531 mycpu->gd_cnt.v_rforks++;
532 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
536 * Both processes are set up, now check if any loadable modules want
537 * to adjust anything.
538 * What if they have an error? XXX
540 TAILQ_FOREACH(ep, &fork_list, next) {
541 (*ep->function)(p1, p2, flags);
545 * Make child runnable and add to run queue.
547 microtime(&(p2->p_stats->p_start));
548 p2->p_acflag = AFORK;
551 * tell any interested parties about the new process
553 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
556 * Return child proc pointer to parent.
563 * The next two functionms are general routines to handle adding/deleting
564 * items on the fork callout list.
567 * Take the arguments given and put them onto the fork callout list,
568 * However first make sure that it's not already there.
569 * Returns 0 on success or a standard error number.
574 forklist_fn function;
579 /* let the programmer know if he's been stupid */
580 if (rm_at_fork(function))
581 printf("WARNING: fork callout entry (%p) already present\n",
584 ep = malloc(sizeof(*ep), M_ATFORK, M_NOWAIT);
587 ep->function = function;
588 TAILQ_INSERT_TAIL(&fork_list, ep, next);
593 * Scan the exit callout list for the given item and remove it..
594 * Returns the number of items removed (0 or 1)
599 forklist_fn function;
603 TAILQ_FOREACH(ep, &fork_list, next) {
604 if (ep->function == function) {
605 TAILQ_REMOVE(&fork_list, ep, next);
614 * Add a forked process to the run queue after any remaining setup, such
615 * as setting the fork handler, has been completed.
619 start_forked_proc(struct proc *p1, struct proc *p2)
622 * Move from SIDL to RUN queue, and activate the process's thread.
623 * Activation of the thread effectively makes the process "a"
624 * current process, so we do not setrunqueue().
626 KASSERT(p2->p_stat == SIDL,
627 ("cannot start forked process, bad status: %p", p2));
634 * Now can be swapped.
639 * Preserve synchronization semantics of vfork. If waiting for
640 * child to exec or exit, set P_PPWAIT on child, and sleep on our
641 * proc (in case of exit).
643 while (p2->p_flag & P_PPWAIT)
644 tsleep(p1, 0, "ppwait", 0);