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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_ktrace.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/eventhandler.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/kthread.h>
50 #include <sys/sysctl.h>
52 #include <sys/malloc.h>
53 #include <sys/mutex.h>
56 #include <sys/pioctl.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sched.h>
59 #include <sys/syscall.h>
60 #include <sys/vmmeter.h>
61 #include <sys/vnode.h>
64 #include <sys/ktrace.h>
65 #include <sys/unistd.h>
67 #include <sys/signalvar.h>
69 #include <security/audit/audit.h>
70 #include <security/mac/mac_framework.h>
74 #include <vm/vm_map.h>
75 #include <vm/vm_extern.h>
79 #ifndef _SYS_SYSPROTO_H_
89 struct fork_args *uap;
94 error = fork1(td, RFFDG | RFPROC, 0, &p2);
96 td->td_retval[0] = p2->p_pid;
106 struct vfork_args *uap;
111 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
113 td->td_retval[0] = p2->p_pid;
114 td->td_retval[1] = 0;
122 struct rfork_args *uap;
127 /* Don't allow kernel-only flags. */
128 if ((uap->flags & RFKERNELONLY) != 0)
131 AUDIT_ARG(fflags, uap->flags);
132 error = fork1(td, uap->flags, 0, &p2);
134 td->td_retval[0] = p2 ? p2->p_pid : 0;
135 td->td_retval[1] = 0;
140 int nprocs = 1; /* process 0 */
142 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
146 * Random component to lastpid 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)
160 error = sysctl_wire_old_buffer(req, sizeof(int));
163 sx_xlock(&allproc_lock);
165 error = sysctl_handle_int(oidp, &pid, 0, req);
166 if (error == 0 && req->newptr != NULL) {
167 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
169 else if (pid < 2) /* NOP */
171 else if (pid < 100) /* Make it reasonable */
175 sx_xunlock(&allproc_lock);
179 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
180 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
183 fork1(td, flags, pages, procp)
189 struct proc *p1, *p2, *pptr;
190 struct proc *newproc;
192 static int curfail, pidchecked = 0;
193 static struct timeval lastfail;
195 struct filedesc_to_leader *fdtol;
197 struct sigacts *newsigacts;
201 /* Can't copy and clear. */
202 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
208 * Here we don't create a new process, but we divorce
209 * certain parts of a process from itself.
211 if ((flags & RFPROC) == 0) {
212 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
213 (flags & (RFCFDG | RFFDG))) {
215 if (thread_single(SINGLE_BOUNDARY)) {
222 error = vm_forkproc(td, NULL, NULL, NULL, flags);
227 * Close all file descriptors.
229 if (flags & RFCFDG) {
230 struct filedesc *fdtmp;
231 fdtmp = fdinit(td->td_proc->p_fd);
237 * Unshare file descriptors (from parent).
243 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
244 (flags & (RFCFDG | RFFDG))) {
255 * We did have single-threading code here
256 * however it proved un-needed and caused problems
259 /* Allocate new proc. */
260 newproc = uma_zalloc(proc_zone, M_WAITOK);
261 if (TAILQ_EMPTY(&newproc->p_threads)) {
262 td2 = thread_alloc();
267 proc_linkup(newproc, td2);
268 sched_newproc(newproc, td2);
270 td2 = FIRST_THREAD_IN_PROC(newproc);
272 /* Allocate and switch to an alternate kstack if specified. */
274 if (!vm_thread_new_altkstack(td2, pages)) {
279 if ((flags & RFMEM) == 0) {
280 vm2 = vmspace_fork(p1->p_vmspace);
288 mac_proc_init(newproc);
290 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
291 STAILQ_INIT(&newproc->p_ktr);
293 /* We have to lock the process tree while we look for a pid. */
294 sx_slock(&proctree_lock);
297 * Although process entries are dynamically created, we still keep
298 * a global limit on the maximum number we will create. Don't allow
299 * a nonprivileged user to use the last ten processes; don't let root
300 * exceed the limit. The variable nprocs is the current number of
301 * processes, maxproc is the limit.
303 sx_xlock(&allproc_lock);
304 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
305 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
311 * Increment the count of procs running with this uid. Don't allow
312 * a nonprivileged user to exceed their current limit.
314 * XXXRW: Can we avoid privilege here if it's not needed?
316 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
318 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
321 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
322 lim_cur(p1, RLIMIT_NPROC));
331 * Increment the nprocs resource before blocking can occur. There
332 * are hard-limits as to the number of processes that can run.
337 * Find an unused process ID. We remember a range of unused IDs
338 * ready to use (from lastpid+1 through pidchecked-1).
340 * If RFHIGHPID is set (used during system boot), do not allocate
343 trypid = lastpid + 1;
344 if (flags & RFHIGHPID) {
349 trypid += arc4random() % randompid;
353 * If the process ID prototype has wrapped around,
354 * restart somewhat above 0, as the low-numbered procs
355 * tend to include daemons that don't exit.
357 if (trypid >= PID_MAX) {
358 trypid = trypid % PID_MAX;
363 if (trypid >= pidchecked) {
366 pidchecked = PID_MAX;
368 * Scan the active and zombie procs to check whether this pid
369 * is in use. Remember the lowest pid that's greater
370 * than trypid, so we can avoid checking for a while.
372 p2 = LIST_FIRST(&allproc);
374 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
375 while (p2->p_pid == trypid ||
376 (p2->p_pgrp != NULL &&
377 (p2->p_pgrp->pg_id == trypid ||
378 (p2->p_session != NULL &&
379 p2->p_session->s_sid == trypid)))) {
381 if (trypid >= pidchecked)
384 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
385 pidchecked = p2->p_pid;
386 if (p2->p_pgrp != NULL) {
387 if (p2->p_pgrp->pg_id > trypid &&
388 pidchecked > p2->p_pgrp->pg_id)
389 pidchecked = p2->p_pgrp->pg_id;
390 if (p2->p_session != NULL &&
391 p2->p_session->s_sid > trypid &&
392 pidchecked > p2->p_session->s_sid)
393 pidchecked = p2->p_session->s_sid;
398 p2 = LIST_FIRST(&zombproc);
402 sx_sunlock(&proctree_lock);
405 * RFHIGHPID does not mess with the lastpid counter during boot.
407 if (flags & RFHIGHPID)
413 p2->p_state = PRS_NEW; /* protect against others */
416 * Allow the scheduler to initialize the child.
421 AUDIT_ARG(pid, p2->p_pid);
422 LIST_INSERT_HEAD(&allproc, p2, p_list);
423 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
428 sx_xunlock(&allproc_lock);
430 bcopy(&p1->p_startcopy, &p2->p_startcopy,
431 __rangeof(struct proc, p_startcopy, p_endcopy));
434 bzero(&p2->p_startzero,
435 __rangeof(struct proc, p_startzero, p_endzero));
437 p2->p_ucred = crhold(td->td_ucred);
441 * Malloc things while we don't hold any locks.
443 if (flags & RFSIGSHARE)
446 newsigacts = sigacts_alloc();
451 if (flags & RFCFDG) {
452 fd = fdinit(p1->p_fd);
454 } else if (flags & RFFDG) {
455 fd = fdcopy(p1->p_fd);
458 fd = fdshare(p1->p_fd);
459 if (p1->p_fdtol == NULL)
461 filedesc_to_leader_alloc(NULL,
464 if ((flags & RFTHREAD) != 0) {
466 * Shared file descriptor table and
467 * shared process leaders.
470 FILEDESC_XLOCK(p1->p_fd);
471 fdtol->fdl_refcount++;
472 FILEDESC_XUNLOCK(p1->p_fd);
475 * Shared file descriptor table, and
476 * different process leaders
478 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
484 * Make a proc table entry for the new process.
485 * Start by zeroing the section of proc that is zero-initialized,
486 * then copy the section that is copied directly from the parent.
492 bzero(&td2->td_startzero,
493 __rangeof(struct thread, td_startzero, td_endzero));
495 bcopy(&td->td_startcopy, &td2->td_startcopy,
496 __rangeof(struct thread, td_startcopy, td_endcopy));
498 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
499 td2->td_sigstk = td->td_sigstk;
500 td2->td_sigmask = td->td_sigmask;
501 td2->td_flags = TDF_INMEM;
504 * Duplicate sub-structures as needed.
505 * Increase reference counts on shared objects.
507 p2->p_flag = P_INMEM;
508 p2->p_swtick = ticks;
509 if (p1->p_flag & P_PROFIL)
511 td2->td_ucred = crhold(p2->p_ucred);
512 pargs_hold(p2->p_args);
514 if (flags & RFSIGSHARE) {
515 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
517 sigacts_copy(newsigacts, p1->p_sigacts);
518 p2->p_sigacts = newsigacts;
520 if (flags & RFLINUXTHPN)
521 p2->p_sigparent = SIGUSR1;
523 p2->p_sigparent = SIGCHLD;
525 p2->p_textvp = p1->p_textvp;
530 * p_limit is copy-on-write. Bump its refcount.
534 pstats_fork(p1->p_stats, p2->p_stats);
539 /* Bump references to the text vnode (for procfs) */
544 * Set up linkage for kernel based threading.
546 if ((flags & RFTHREAD) != 0) {
547 mtx_lock(&ppeers_lock);
548 p2->p_peers = p1->p_peers;
550 p2->p_leader = p1->p_leader;
551 mtx_unlock(&ppeers_lock);
552 PROC_LOCK(p1->p_leader);
553 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
554 PROC_UNLOCK(p1->p_leader);
556 * The task leader is exiting, so process p1 is
557 * going to be killed shortly. Since p1 obviously
558 * isn't dead yet, we know that the leader is either
559 * sending SIGKILL's to all the processes in this
560 * task or is sleeping waiting for all the peers to
561 * exit. We let p1 complete the fork, but we need
562 * to go ahead and kill the new process p2 since
563 * the task leader may not get a chance to send
564 * SIGKILL to it. We leave it on the list so that
565 * the task leader will wait for this new process
569 psignal(p2, SIGKILL);
572 PROC_UNLOCK(p1->p_leader);
578 sx_xlock(&proctree_lock);
579 PGRP_LOCK(p1->p_pgrp);
584 * Preserve some more flags in subprocess. P_PROFIL has already
587 p2->p_flag |= p1->p_flag & P_SUGID;
588 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
589 SESS_LOCK(p1->p_session);
590 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
591 p2->p_flag |= P_CONTROLT;
592 SESS_UNLOCK(p1->p_session);
593 if (flags & RFPPWAIT)
594 p2->p_flag |= P_PPWAIT;
596 p2->p_pgrp = p1->p_pgrp;
597 LIST_INSERT_AFTER(p1, p2, p_pglist);
598 PGRP_UNLOCK(p1->p_pgrp);
599 LIST_INIT(&p2->p_children);
601 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
605 * Copy traceflag and tracefile if enabled.
607 mtx_lock(&ktrace_mtx);
608 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
609 if (p1->p_traceflag & KTRFAC_INHERIT) {
610 p2->p_traceflag = p1->p_traceflag;
611 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
613 KASSERT(p1->p_tracecred != NULL,
614 ("ktrace vnode with no cred"));
615 p2->p_tracecred = crhold(p1->p_tracecred);
618 mtx_unlock(&ktrace_mtx);
622 * If PF_FORK is set, the child process inherits the
623 * procfs ioctl flags from its parent.
625 if (p1->p_pfsflags & PF_FORK) {
626 p2->p_stops = p1->p_stops;
627 p2->p_pfsflags = p1->p_pfsflags;
631 * This begins the section where we must prevent the parent
632 * from being swapped.
638 * Attach the new process to its parent.
640 * If RFNOWAIT is set, the newly created process becomes a child
641 * of init. This effectively disassociates the child from the
644 if (flags & RFNOWAIT)
649 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
650 sx_xunlock(&proctree_lock);
652 /* Inform accounting that we have forked. */
653 p2->p_acflag = AFORK;
657 * Finish creating the child process. It will return via a different
658 * execution path later. (ie: directly into user mode)
660 vm_forkproc(td, p2, td2, vm2, flags);
662 if (flags == (RFFDG | RFPROC)) {
663 PCPU_INC(cnt.v_forks);
664 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
665 p2->p_vmspace->vm_ssize);
666 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
667 PCPU_INC(cnt.v_vforks);
668 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
669 p2->p_vmspace->vm_ssize);
670 } else if (p1 == &proc0) {
671 PCPU_INC(cnt.v_kthreads);
672 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
673 p2->p_vmspace->vm_ssize);
675 PCPU_INC(cnt.v_rforks);
676 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
677 p2->p_vmspace->vm_ssize);
681 * Both processes are set up, now check if any loadable modules want
682 * to adjust anything.
683 * What if they have an error? XXX
685 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
688 * Set the child start time and mark the process as being complete.
690 microuptime(&p2->p_stats->p_start);
692 p2->p_state = PRS_NORMAL;
696 * If RFSTOPPED not requested, make child runnable and add to
699 if ((flags & RFSTOPPED) == 0) {
702 sched_add(td2, SRQ_BORING);
707 * Now can be swapped.
713 * Tell any interested parties about the new process.
715 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
720 * Preserve synchronization semantics of vfork. If waiting for
721 * child to exec or exit, set P_PPWAIT on child, and sleep on our
722 * proc (in case of exit).
725 while (p2->p_flag & P_PPWAIT)
726 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
730 * Return child proc pointer to parent.
735 sx_sunlock(&proctree_lock);
736 if (ppsratecheck(&lastfail, &curfail, 1))
737 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
738 td->td_ucred->cr_ruid);
739 sx_xunlock(&allproc_lock);
741 mac_proc_destroy(newproc);
744 uma_zfree(proc_zone, newproc);
745 pause("fork", hz / 2);
750 * Handle the return of a child process from fork1(). This function
751 * is called from the MD fork_trampoline() entry point.
754 fork_exit(callout, arg, frame)
755 void (*callout)(void *, struct trapframe *);
757 struct trapframe *frame;
765 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
767 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
768 td, td->td_sched, p->p_pid, td->td_name);
772 * Processes normally resume in mi_switch() after being
773 * cpu_switch()'ed to, but when children start up they arrive here
774 * instead, so we must do much the same things as mi_switch() would.
776 if ((dtd = PCPU_GET(deadthread))) {
777 PCPU_SET(deadthread, NULL);
783 * cpu_set_fork_handler intercepts this function call to
784 * have this call a non-return function to stay in kernel mode.
785 * initproc has its own fork handler, but it does return.
787 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
791 * Check if a kernel thread misbehaved and returned from its main
794 if (p->p_flag & P_KTHREAD) {
795 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
796 td->td_name, p->p_pid);
799 mtx_assert(&Giant, MA_NOTOWNED);
801 EVENTHANDLER_INVOKE(schedtail, p);
805 * Simplified back end of syscall(), used when returning from fork()
806 * directly into user mode. Giant is not held on entry, and must not
807 * be held on return. This function is passed in to fork_exit() as the
808 * first parameter and is called when returning to a new userland process.
811 fork_return(td, frame)
813 struct trapframe *frame;
818 if (KTRPOINT(td, KTR_SYSRET))
819 ktrsysret(SYS_fork, 0, 0);
821 mtx_assert(&Giant, MA_NOTOWNED);