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_kdtrace.h"
41 #include "opt_ktrace.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/eventhandler.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>
50 #include <sys/kthread.h>
51 #include <sys/sysctl.h>
53 #include <sys/malloc.h>
54 #include <sys/mutex.h>
58 #include <sys/pioctl.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
61 #include <sys/syscall.h>
62 #include <sys/vmmeter.h>
63 #include <sys/vnode.h>
66 #include <sys/ktrace.h>
67 #include <sys/unistd.h>
70 #include <sys/signalvar.h>
72 #include <security/audit/audit.h>
73 #include <security/mac/mac_framework.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_extern.h>
82 #include <sys/dtrace_bsd.h>
83 dtrace_fork_func_t dtrace_fasttrap_fork;
86 SDT_PROVIDER_DECLARE(proc);
87 SDT_PROBE_DEFINE(proc, kernel, , create);
88 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *");
89 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *");
90 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int");
92 #ifndef _SYS_SYSPROTO_H_
102 struct fork_args *uap;
107 error = fork1(td, RFFDG | RFPROC, 0, &p2);
109 td->td_retval[0] = p2->p_pid;
110 td->td_retval[1] = 0;
119 struct vfork_args *uap;
124 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
126 td->td_retval[0] = p2->p_pid;
127 td->td_retval[1] = 0;
135 struct rfork_args *uap;
140 /* Don't allow kernel-only flags. */
141 if ((uap->flags & RFKERNELONLY) != 0)
144 AUDIT_ARG(fflags, uap->flags);
145 error = fork1(td, uap->flags, 0, &p2);
147 td->td_retval[0] = p2 ? p2->p_pid : 0;
148 td->td_retval[1] = 0;
153 int nprocs = 1; /* process 0 */
155 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
159 * Random component to lastpid generation. We mix in a random factor to make
160 * it a little harder to predict. We sanity check the modulus value to avoid
161 * doing it in critical paths. Don't let it be too small or we pointlessly
162 * waste randomness entropy, and don't let it be impossibly large. Using a
163 * modulus that is too big causes a LOT more process table scans and slows
164 * down fork processing as the pidchecked caching is defeated.
166 static int randompid = 0;
169 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
173 error = sysctl_wire_old_buffer(req, sizeof(int));
176 sx_xlock(&allproc_lock);
178 error = sysctl_handle_int(oidp, &pid, 0, req);
179 if (error == 0 && req->newptr != NULL) {
180 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
182 else if (pid < 2) /* NOP */
184 else if (pid < 100) /* Make it reasonable */
188 sx_xunlock(&allproc_lock);
192 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
193 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
196 fork1(td, flags, pages, procp)
202 struct proc *p1, *p2, *pptr;
203 struct proc *newproc;
205 static int curfail, pidchecked = 0;
206 static struct timeval lastfail;
208 struct filedesc_to_leader *fdtol;
210 struct sigacts *newsigacts;
214 /* Can't copy and clear. */
215 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
221 * Here we don't create a new process, but we divorce
222 * certain parts of a process from itself.
224 if ((flags & RFPROC) == 0) {
225 if ((p1->p_flag & P_HADTHREADS) &&
226 (flags & (RFCFDG | RFFDG))) {
228 if (thread_single(SINGLE_BOUNDARY)) {
235 error = vm_forkproc(td, NULL, NULL, NULL, flags);
240 * Close all file descriptors.
242 if (flags & RFCFDG) {
243 struct filedesc *fdtmp;
244 fdtmp = fdinit(td->td_proc->p_fd);
250 * Unshare file descriptors (from parent).
256 if ((p1->p_flag & P_HADTHREADS) &&
257 (flags & (RFCFDG | RFFDG))) {
267 /* Allocate new proc. */
268 newproc = uma_zalloc(proc_zone, M_WAITOK);
269 if (TAILQ_EMPTY(&newproc->p_threads)) {
270 td2 = thread_alloc();
275 proc_linkup(newproc, td2);
276 sched_newproc(newproc, td2);
278 td2 = FIRST_THREAD_IN_PROC(newproc);
280 /* Allocate and switch to an alternate kstack if specified. */
282 if (!vm_thread_new_altkstack(td2, pages)) {
287 if ((flags & RFMEM) == 0) {
288 vm2 = vmspace_fork(p1->p_vmspace);
295 mac_init_proc(newproc);
297 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
298 STAILQ_INIT(&newproc->p_ktr);
300 /* We have to lock the process tree while we look for a pid. */
301 sx_slock(&proctree_lock);
304 * Although process entries are dynamically created, we still keep
305 * a global limit on the maximum number we will create. Don't allow
306 * a nonprivileged user to use the last ten processes; don't let root
307 * exceed the limit. The variable nprocs is the current number of
308 * processes, maxproc is the limit.
310 sx_xlock(&allproc_lock);
311 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
312 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
318 * Increment the count of procs running with this uid. Don't allow
319 * a nonprivileged user to exceed their current limit.
321 * XXXRW: Can we avoid privilege here if it's not needed?
323 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
325 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
328 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
329 lim_cur(p1, RLIMIT_NPROC));
338 * Increment the nprocs resource before blocking can occur. There
339 * are hard-limits as to the number of processes that can run.
344 * Find an unused process ID. We remember a range of unused IDs
345 * ready to use (from lastpid+1 through pidchecked-1).
347 * If RFHIGHPID is set (used during system boot), do not allocate
350 trypid = lastpid + 1;
351 if (flags & RFHIGHPID) {
356 trypid += arc4random() % randompid;
360 * If the process ID prototype has wrapped around,
361 * restart somewhat above 0, as the low-numbered procs
362 * tend to include daemons that don't exit.
364 if (trypid >= PID_MAX) {
365 trypid = trypid % PID_MAX;
370 if (trypid >= pidchecked) {
373 pidchecked = PID_MAX;
375 * Scan the active and zombie procs to check whether this pid
376 * is in use. Remember the lowest pid that's greater
377 * than trypid, so we can avoid checking for a while.
379 p2 = LIST_FIRST(&allproc);
381 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
382 while (p2->p_pid == trypid ||
383 (p2->p_pgrp != NULL &&
384 (p2->p_pgrp->pg_id == trypid ||
385 (p2->p_session != NULL &&
386 p2->p_session->s_sid == trypid)))) {
388 if (trypid >= pidchecked)
391 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
392 pidchecked = p2->p_pid;
393 if (p2->p_pgrp != NULL) {
394 if (p2->p_pgrp->pg_id > trypid &&
395 pidchecked > p2->p_pgrp->pg_id)
396 pidchecked = p2->p_pgrp->pg_id;
397 if (p2->p_session != NULL &&
398 p2->p_session->s_sid > trypid &&
399 pidchecked > p2->p_session->s_sid)
400 pidchecked = p2->p_session->s_sid;
405 p2 = LIST_FIRST(&zombproc);
409 sx_sunlock(&proctree_lock);
412 * RFHIGHPID does not mess with the lastpid counter during boot.
414 if (flags & RFHIGHPID)
420 p2->p_state = PRS_NEW; /* protect against others */
423 * Allow the scheduler to initialize the child.
428 AUDIT_ARG(pid, p2->p_pid);
429 LIST_INSERT_HEAD(&allproc, p2, p_list);
430 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
435 sx_xunlock(&allproc_lock);
437 bcopy(&p1->p_startcopy, &p2->p_startcopy,
438 __rangeof(struct proc, p_startcopy, p_endcopy));
439 pargs_hold(p2->p_args);
442 bzero(&p2->p_startzero,
443 __rangeof(struct proc, p_startzero, p_endzero));
445 p2->p_ucred = crhold(td->td_ucred);
447 /* In case we are jailed tell the prison that we exist. */
448 if (jailed(p2->p_ucred))
449 prison_proc_hold(p2->p_ucred->cr_prison);
454 * Malloc things while we don't hold any locks.
456 if (flags & RFSIGSHARE)
459 newsigacts = sigacts_alloc();
464 if (flags & RFCFDG) {
465 fd = fdinit(p1->p_fd);
467 } else if (flags & RFFDG) {
468 fd = fdcopy(p1->p_fd);
471 fd = fdshare(p1->p_fd);
472 if (p1->p_fdtol == NULL)
474 filedesc_to_leader_alloc(NULL,
477 if ((flags & RFTHREAD) != 0) {
479 * Shared file descriptor table and
480 * shared process leaders.
483 FILEDESC_XLOCK(p1->p_fd);
484 fdtol->fdl_refcount++;
485 FILEDESC_XUNLOCK(p1->p_fd);
488 * Shared file descriptor table, and
489 * different process leaders
491 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
497 * Make a proc table entry for the new process.
498 * Start by zeroing the section of proc that is zero-initialized,
499 * then copy the section that is copied directly from the parent.
505 bzero(&td2->td_startzero,
506 __rangeof(struct thread, td_startzero, td_endzero));
508 bcopy(&td->td_startcopy, &td2->td_startcopy,
509 __rangeof(struct thread, td_startcopy, td_endcopy));
511 td2->td_sigstk = td->td_sigstk;
512 td2->td_sigmask = td->td_sigmask;
513 td2->td_flags = TDF_INMEM;
516 * Duplicate sub-structures as needed.
517 * Increase reference counts on shared objects.
519 p2->p_flag = P_INMEM;
520 p2->p_swtick = ticks;
521 if (p1->p_flag & P_PROFIL)
523 td2->td_ucred = crhold(p2->p_ucred);
525 if (flags & RFSIGSHARE) {
526 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
528 sigacts_copy(newsigacts, p1->p_sigacts);
529 p2->p_sigacts = newsigacts;
531 if (flags & RFLINUXTHPN)
532 p2->p_sigparent = SIGUSR1;
534 p2->p_sigparent = SIGCHLD;
536 p2->p_textvp = p1->p_textvp;
541 * p_limit is copy-on-write. Bump its refcount.
545 pstats_fork(p1->p_stats, p2->p_stats);
550 /* Bump references to the text vnode (for procfs) */
555 * Set up linkage for kernel based threading.
557 if ((flags & RFTHREAD) != 0) {
558 mtx_lock(&ppeers_lock);
559 p2->p_peers = p1->p_peers;
561 p2->p_leader = p1->p_leader;
562 mtx_unlock(&ppeers_lock);
563 PROC_LOCK(p1->p_leader);
564 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
565 PROC_UNLOCK(p1->p_leader);
567 * The task leader is exiting, so process p1 is
568 * going to be killed shortly. Since p1 obviously
569 * isn't dead yet, we know that the leader is either
570 * sending SIGKILL's to all the processes in this
571 * task or is sleeping waiting for all the peers to
572 * exit. We let p1 complete the fork, but we need
573 * to go ahead and kill the new process p2 since
574 * the task leader may not get a chance to send
575 * SIGKILL to it. We leave it on the list so that
576 * the task leader will wait for this new process
580 psignal(p2, SIGKILL);
583 PROC_UNLOCK(p1->p_leader);
589 sx_xlock(&proctree_lock);
590 PGRP_LOCK(p1->p_pgrp);
595 * Preserve some more flags in subprocess. P_PROFIL has already
598 p2->p_flag |= p1->p_flag & P_SUGID;
599 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
600 SESS_LOCK(p1->p_session);
601 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
602 p2->p_flag |= P_CONTROLT;
603 SESS_UNLOCK(p1->p_session);
604 if (flags & RFPPWAIT)
605 p2->p_flag |= P_PPWAIT;
607 p2->p_pgrp = p1->p_pgrp;
608 LIST_INSERT_AFTER(p1, p2, p_pglist);
609 PGRP_UNLOCK(p1->p_pgrp);
610 LIST_INIT(&p2->p_children);
612 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
616 * Copy traceflag and tracefile if enabled.
618 mtx_lock(&ktrace_mtx);
619 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
620 if (p1->p_traceflag & KTRFAC_INHERIT) {
621 p2->p_traceflag = p1->p_traceflag;
622 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
624 KASSERT(p1->p_tracecred != NULL,
625 ("ktrace vnode with no cred"));
626 p2->p_tracecred = crhold(p1->p_tracecred);
629 mtx_unlock(&ktrace_mtx);
633 * If PF_FORK is set, the child process inherits the
634 * procfs ioctl flags from its parent.
636 if (p1->p_pfsflags & PF_FORK) {
637 p2->p_stops = p1->p_stops;
638 p2->p_pfsflags = p1->p_pfsflags;
643 * Tell the DTrace fasttrap provider about the new process
644 * if it has registered an interest.
646 if (dtrace_fasttrap_fork)
647 dtrace_fasttrap_fork(p1, p2);
651 * This begins the section where we must prevent the parent
652 * from being swapped.
658 * Attach the new process to its parent.
660 * If RFNOWAIT is set, the newly created process becomes a child
661 * of init. This effectively disassociates the child from the
664 if (flags & RFNOWAIT)
669 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
670 sx_xunlock(&proctree_lock);
672 /* Inform accounting that we have forked. */
673 p2->p_acflag = AFORK;
677 * Finish creating the child process. It will return via a different
678 * execution path later. (ie: directly into user mode)
680 vm_forkproc(td, p2, td2, vm2, flags);
682 if (flags == (RFFDG | RFPROC)) {
683 PCPU_INC(cnt.v_forks);
684 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
685 p2->p_vmspace->vm_ssize);
686 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
687 PCPU_INC(cnt.v_vforks);
688 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
689 p2->p_vmspace->vm_ssize);
690 } else if (p1 == &proc0) {
691 PCPU_INC(cnt.v_kthreads);
692 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
693 p2->p_vmspace->vm_ssize);
695 PCPU_INC(cnt.v_rforks);
696 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
697 p2->p_vmspace->vm_ssize);
701 * Both processes are set up, now check if any loadable modules want
702 * to adjust anything.
703 * What if they have an error? XXX
705 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
708 * Set the child start time and mark the process as being complete.
710 microuptime(&p2->p_stats->p_start);
712 p2->p_state = PRS_NORMAL;
716 * If RFSTOPPED not requested, make child runnable and add to
719 if ((flags & RFSTOPPED) == 0) {
722 sched_add(td2, SRQ_BORING);
727 * Now can be swapped.
734 * Tell any interested parties about the new process.
736 knote_fork(&p1->p_klist, p2->p_pid);
737 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
740 * Preserve synchronization semantics of vfork. If waiting for
741 * child to exec or exit, set P_PPWAIT on child, and sleep on our
742 * proc (in case of exit).
745 while (p2->p_flag & P_PPWAIT)
746 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
750 * Return child proc pointer to parent.
755 sx_sunlock(&proctree_lock);
756 if (ppsratecheck(&lastfail, &curfail, 1))
757 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
758 td->td_ucred->cr_ruid);
759 sx_xunlock(&allproc_lock);
761 mac_destroy_proc(newproc);
766 uma_zfree(proc_zone, newproc);
767 pause("fork", hz / 2);
772 * Handle the return of a child process from fork1(). This function
773 * is called from the MD fork_trampoline() entry point.
776 fork_exit(callout, arg, frame)
777 void (*callout)(void *, struct trapframe *);
779 struct trapframe *frame;
787 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
789 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
790 td, td->td_sched, p->p_pid, p->p_comm);
794 * Processes normally resume in mi_switch() after being
795 * cpu_switch()'ed to, but when children start up they arrive here
796 * instead, so we must do much the same things as mi_switch() would.
798 if ((dtd = PCPU_GET(deadthread))) {
799 PCPU_SET(deadthread, NULL);
805 * cpu_set_fork_handler intercepts this function call to
806 * have this call a non-return function to stay in kernel mode.
807 * initproc has its own fork handler, but it does return.
809 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
813 * Check if a kernel thread misbehaved and returned from its main
816 if (p->p_flag & P_KTHREAD) {
817 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
818 p->p_comm, p->p_pid);
821 mtx_assert(&Giant, MA_NOTOWNED);
823 EVENTHANDLER_INVOKE(schedtail, p);
827 * Simplified back end of syscall(), used when returning from fork()
828 * directly into user mode. Giant is not held on entry, and must not
829 * be held on return. This function is passed in to fork_exit() as the
830 * first parameter and is called when returning to a new userland process.
833 fork_return(td, frame)
835 struct trapframe *frame;
840 if (KTRPOINT(td, KTR_SYSRET))
841 ktrsysret(SYS_fork, 0, 0);
843 mtx_assert(&Giant, MA_NOTOWNED);