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 1.23 2004/04/12 17:49:27 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(struct proc *p1, int flags, struct proc **procp)
180 struct proc *p2, *pptr;
182 struct proc *newproc;
184 static int curfail = 0, pidchecked = 0;
185 static struct timeval lastfail;
187 struct filedesc_to_leader *fdtol;
189 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
193 * Here we don't create a new process, but we divorce
194 * certain parts of a process from itself.
196 if ((flags & RFPROC) == 0) {
198 vm_fork(p1, 0, flags);
201 * Close all file descriptors.
203 if (flags & RFCFDG) {
204 struct filedesc *fdtmp;
211 * Unshare file descriptors (from parent.)
214 if (p1->p_fd->fd_refcnt > 1) {
215 struct filedesc *newfd;
226 * Although process entries are dynamically created, we still keep
227 * a global limit on the maximum number we will create. Don't allow
228 * a nonprivileged user to use the last ten processes; don't let root
229 * exceed the limit. The variable nprocs is the current number of
230 * processes, maxproc is the limit.
232 uid = p1->p_ucred->cr_ruid;
233 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
234 if (ppsratecheck(&lastfail, &curfail, 1))
235 printf("maxproc limit exceeded by uid %d, please "
236 "see tuning(7) and login.conf(5).\n", uid);
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 if (ppsratecheck(&lastfail, &curfail, 1))
258 printf("maxproc limit exceeded by uid %d, please "
259 "see tuning(7) and login.conf(5).\n", uid);
260 tsleep(&forksleep, 0, "fork", hz / 2);
264 /* Allocate new proc. */
265 newproc = zalloc(proc_zone);
268 * Setup linkage for kernel based threading
270 if ((flags & RFTHREAD) != 0) {
271 newproc->p_peers = p1->p_peers;
272 p1->p_peers = newproc;
273 newproc->p_leader = p1->p_leader;
275 newproc->p_peers = 0;
276 newproc->p_leader = newproc;
279 newproc->p_wakeup = 0;
280 newproc->p_vmspace = NULL;
283 * Find an unused process ID. We remember a range of unused IDs
284 * ready to use (from nextpid+1 through pidchecked-1).
288 nextpid += arc4random() % randompid;
291 * If the process ID prototype has wrapped around,
292 * restart somewhat above 0, as the low-numbered procs
293 * tend to include daemons that don't exit.
295 if (nextpid >= PID_MAX) {
296 nextpid = nextpid % PID_MAX;
301 if (nextpid >= pidchecked) {
304 pidchecked = PID_MAX;
306 * Scan the active and zombie procs to check whether this pid
307 * is in use. Remember the lowest pid that's greater
308 * than nextpid, so we can avoid checking for a while.
310 p2 = LIST_FIRST(&allproc);
312 for (; p2 != 0; p2 = LIST_NEXT(p2, p_list)) {
313 while (p2->p_pid == nextpid ||
314 p2->p_pgrp->pg_id == nextpid ||
315 p2->p_session->s_sid == nextpid) {
317 if (nextpid >= pidchecked)
320 if (p2->p_pid > nextpid && pidchecked > p2->p_pid)
321 pidchecked = p2->p_pid;
322 if (p2->p_pgrp->pg_id > nextpid &&
323 pidchecked > p2->p_pgrp->pg_id)
324 pidchecked = p2->p_pgrp->pg_id;
325 if (p2->p_session->s_sid > nextpid &&
326 pidchecked > p2->p_session->s_sid)
327 pidchecked = p2->p_session->s_sid;
331 p2 = LIST_FIRST(&zombproc);
337 p2->p_stat = SIDL; /* protect against others */
339 LIST_INSERT_HEAD(&allproc, p2, p_list);
340 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
343 * Make a proc table entry for the new process.
344 * Start by zeroing the section of proc that is zero-initialized,
345 * then copy the section that is copied directly from the parent.
347 bzero(&p2->p_startzero,
348 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero));
349 bcopy(&p1->p_startcopy, &p2->p_startcopy,
350 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
352 p2->p_aioinfo = NULL;
355 * Duplicate sub-structures as needed.
356 * Increase reference counts on shared objects.
357 * The p_stats and p_sigacts substructs are set in vm_fork.
359 * P_CP_RELEASED indicates that the process is starting out in
360 * the kernel (in the fork trampoline). The flag will be cleared
361 * when the new process calls userret() and acquires its current
362 * process designation for the return to userland.
364 p2->p_flag = P_INMEM | P_CP_RELEASED;
365 if (p1->p_flag & P_PROFIL)
367 p2->p_ucred = crhold(p1->p_ucred);
369 if (p2->p_ucred->cr_prison) {
370 p2->p_ucred->cr_prison->pr_ref++;
371 p2->p_flag |= P_JAILED;
375 p2->p_args->ar_ref++;
377 if (flags & RFSIGSHARE) {
378 p2->p_procsig = p1->p_procsig;
379 p2->p_procsig->ps_refcnt++;
380 if (p1->p_sigacts == &p1->p_addr->u_sigacts) {
381 struct sigacts *newsigacts;
384 /* Create the shared sigacts structure */
385 MALLOC(newsigacts, struct sigacts *,
386 sizeof(struct sigacts), M_SUBPROC, M_WAITOK);
389 * Set p_sigacts to the new shared structure.
390 * Note that this is updating p1->p_sigacts at the
391 * same time, since p_sigacts is just a pointer to
392 * the shared p_procsig->ps_sigacts.
394 p2->p_sigacts = newsigacts;
395 bcopy(&p1->p_addr->u_sigacts, p2->p_sigacts,
396 sizeof(*p2->p_sigacts));
397 *p2->p_sigacts = p1->p_addr->u_sigacts;
401 MALLOC(p2->p_procsig, struct procsig *, sizeof(struct procsig),
402 M_SUBPROC, M_WAITOK);
403 bcopy(p1->p_procsig, p2->p_procsig, sizeof(*p2->p_procsig));
404 p2->p_procsig->ps_refcnt = 1;
405 p2->p_sigacts = NULL; /* finished in vm_fork() */
407 if (flags & RFLINUXTHPN)
408 p2->p_sigparent = SIGUSR1;
410 p2->p_sigparent = SIGCHLD;
412 /* bump references to the text vnode (for procfs) */
413 p2->p_textvp = p1->p_textvp;
417 if (flags & RFCFDG) {
418 p2->p_fd = fdinit(p1);
420 } else if (flags & RFFDG) {
421 p2->p_fd = fdcopy(p1);
424 p2->p_fd = fdshare(p1);
425 if (p1->p_fdtol == NULL)
427 filedesc_to_leader_alloc(NULL,
429 if ((flags & RFTHREAD) != 0) {
431 * Shared file descriptor table and
432 * shared process leaders.
435 fdtol->fdl_refcount++;
438 * Shared file descriptor table, and
439 * different process leaders
441 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
447 * If p_limit is still copy-on-write, bump refcnt,
448 * otherwise get a copy that won't be modified.
449 * (If PL_SHAREMOD is clear, the structure is shared
452 if (p1->p_limit->p_lflags & PL_SHAREMOD) {
453 p2->p_limit = limcopy(p1->p_limit);
455 p2->p_limit = p1->p_limit;
456 p2->p_limit->p_refcnt++;
460 * Preserve some more flags in subprocess. P_PROFIL has already
463 p2->p_flag |= p1->p_flag & (P_SUGID | P_ALTSTACK);
464 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
465 p2->p_flag |= P_CONTROLT;
466 if (flags & RFPPWAIT)
467 p2->p_flag |= P_PPWAIT;
469 LIST_INSERT_AFTER(p1, p2, p_pglist);
472 * Attach the new process to its parent.
474 * If RFNOWAIT is set, the newly created process becomes a child
475 * of init. This effectively disassociates the child from the
478 if (flags & RFNOWAIT)
483 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
484 LIST_INIT(&p2->p_children);
485 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
489 * Copy traceflag and tracefile if enabled. If not inherited,
490 * these were zeroed above but we still could have a trace race
491 * so make sure p2's p_tracep is NULL.
493 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracep == NULL) {
494 p2->p_traceflag = p1->p_traceflag;
495 if ((p2->p_tracep = p1->p_tracep) != NULL)
501 * Give the child process an estcpu skewed towards the batch side
502 * of the parent. This prevents batch programs from glitching
503 * interactive programs when they are first started. If the child
504 * is not a batch program it's priority will be corrected by the
507 p2->p_estcpu_fork = p2->p_estcpu =
508 ESTCPULIM(p1->p_estcpu + ESTCPURAMP);
511 * This begins the section where we must prevent the parent
512 * from being swapped.
517 * Finish creating the child process. It will return via a different
518 * execution path later. (ie: directly into user mode)
520 vm_fork(p1, p2, flags);
521 caps_fork(p1, p2, flags);
523 if (flags == (RFFDG | RFPROC)) {
524 mycpu->gd_cnt.v_forks++;
525 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
526 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
527 mycpu->gd_cnt.v_vforks++;
528 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
529 } else if (p1 == &proc0) {
530 mycpu->gd_cnt.v_kthreads++;
531 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
533 mycpu->gd_cnt.v_rforks++;
534 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
538 * Both processes are set up, now check if any loadable modules want
539 * to adjust anything.
540 * What if they have an error? XXX
542 TAILQ_FOREACH(ep, &fork_list, next) {
543 (*ep->function)(p1, p2, flags);
547 * Make child runnable and add to run queue.
549 microtime(&(p2->p_stats->p_start));
550 p2->p_acflag = AFORK;
553 * tell any interested parties about the new process
555 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
558 * Return child proc pointer to parent.
565 * The next two functionms are general routines to handle adding/deleting
566 * items on the fork callout list.
569 * Take the arguments given and put them onto the fork callout list,
570 * However first make sure that it's not already there.
571 * Returns 0 on success or a standard error number.
574 at_fork(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",
585 ep = malloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
586 ep->function = function;
587 TAILQ_INSERT_TAIL(&fork_list, ep, next);
592 * Scan the exit callout list for the given item and remove it..
593 * Returns the number of items removed (0 or 1)
596 rm_at_fork(forklist_fn function)
600 TAILQ_FOREACH(ep, &fork_list, next) {
601 if (ep->function == function) {
602 TAILQ_REMOVE(&fork_list, ep, next);
611 * Add a forked process to the run queue after any remaining setup, such
612 * as setting the fork handler, has been completed.
615 start_forked_proc(struct proc *p1, struct proc *p2)
618 * Move from SIDL to RUN queue, and activate the process's thread.
619 * Activation of the thread effectively makes the process "a"
620 * current process, so we do not setrunqueue().
622 KASSERT(p2->p_stat == SIDL,
623 ("cannot start forked process, bad status: %p", p2));
631 * Now can be swapped.
636 * Preserve synchronization semantics of vfork. If waiting for
637 * child to exec or exit, set P_PPWAIT on child, and sleep on our
638 * proc (in case of exit).
640 while (p2->p_flag & P_PPWAIT)
641 tsleep(p1, 0, "ppwait", 0);