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"
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>
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>
57 #include <sys/pioctl.h>
58 #include <sys/resourcevar.h>
59 #include <sys/sched.h>
60 #include <sys/syscall.h>
61 #include <sys/vmmeter.h>
62 #include <sys/vnode.h>
65 #include <sys/ktrace.h>
66 #include <sys/unistd.h>
69 #include <sys/signalvar.h>
71 #include <security/audit/audit.h>
72 #include <security/mac/mac_framework.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_extern.h>
81 #include <sys/dtrace_bsd.h>
82 dtrace_fork_func_t dtrace_fasttrap_fork;
85 SDT_PROVIDER_DECLARE(proc);
86 SDT_PROBE_DEFINE(proc, kernel, , create);
87 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *");
88 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *");
89 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int");
91 #ifndef _SYS_SYSPROTO_H_
101 struct fork_args *uap;
106 error = fork1(td, RFFDG | RFPROC, 0, &p2);
108 td->td_retval[0] = p2->p_pid;
109 td->td_retval[1] = 0;
118 struct vfork_args *uap;
124 flags = RFFDG | RFPROC; /* validate that this is still an issue */
126 flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
128 error = fork1(td, flags, 0, &p2);
130 td->td_retval[0] = p2->p_pid;
131 td->td_retval[1] = 0;
139 struct rfork_args *uap;
144 /* Don't allow kernel-only flags. */
145 if ((uap->flags & RFKERNELONLY) != 0)
148 AUDIT_ARG_FFLAGS(uap->flags);
149 error = fork1(td, uap->flags, 0, &p2);
151 td->td_retval[0] = p2 ? p2->p_pid : 0;
152 td->td_retval[1] = 0;
157 int nprocs = 1; /* process 0 */
159 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
163 * Random component to lastpid generation. We mix in a random factor to make
164 * it a little harder to predict. We sanity check the modulus value to avoid
165 * doing it in critical paths. Don't let it be too small or we pointlessly
166 * waste randomness entropy, and don't let it be impossibly large. Using a
167 * modulus that is too big causes a LOT more process table scans and slows
168 * down fork processing as the pidchecked caching is defeated.
170 static int randompid = 0;
173 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
177 error = sysctl_wire_old_buffer(req, sizeof(int));
180 sx_xlock(&allproc_lock);
182 error = sysctl_handle_int(oidp, &pid, 0, req);
183 if (error == 0 && req->newptr != NULL) {
184 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
186 else if (pid < 2) /* NOP */
188 else if (pid < 100) /* Make it reasonable */
192 sx_xunlock(&allproc_lock);
196 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
197 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
200 fork1(td, flags, pages, procp)
206 struct proc *p1, *p2, *pptr;
207 struct proc *newproc;
209 static int curfail, pidchecked = 0;
210 static struct timeval lastfail;
212 struct filedesc_to_leader *fdtol;
214 struct sigacts *newsigacts;
216 vm_ooffset_t mem_charged;
219 /* Can't copy and clear. */
220 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
226 * Here we don't create a new process, but we divorce
227 * certain parts of a process from itself.
229 if ((flags & RFPROC) == 0) {
230 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
231 (flags & (RFCFDG | RFFDG))) {
233 if (thread_single(SINGLE_BOUNDARY)) {
240 error = vm_forkproc(td, NULL, NULL, NULL, flags);
245 * Close all file descriptors.
247 if (flags & RFCFDG) {
248 struct filedesc *fdtmp;
249 fdtmp = fdinit(td->td_proc->p_fd);
255 * Unshare file descriptors (from parent).
261 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
262 (flags & (RFCFDG | RFFDG))) {
273 * We did have single-threading code here
274 * however it proved un-needed and caused problems
279 /* Allocate new proc. */
280 newproc = uma_zalloc(proc_zone, M_WAITOK);
281 if (TAILQ_EMPTY(&newproc->p_threads)) {
282 td2 = thread_alloc();
287 proc_linkup(newproc, td2);
289 td2 = FIRST_THREAD_IN_PROC(newproc);
291 /* Allocate and switch to an alternate kstack if specified. */
293 if (!vm_thread_new_altkstack(td2, pages)) {
298 if ((flags & RFMEM) == 0) {
299 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
304 if (!swap_reserve(mem_charged)) {
306 * The swap reservation failed. The accounting
307 * from the entries of the copied vm2 will be
308 * substracted in vmspace_free(), so force the
311 swap_reserve_force(mem_charged);
318 mac_proc_init(newproc);
320 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
321 STAILQ_INIT(&newproc->p_ktr);
323 /* We have to lock the process tree while we look for a pid. */
324 sx_slock(&proctree_lock);
327 * Although process entries are dynamically created, we still keep
328 * a global limit on the maximum number we will create. Don't allow
329 * a nonprivileged user to use the last ten processes; don't let root
330 * exceed the limit. The variable nprocs is the current number of
331 * processes, maxproc is the limit.
333 sx_xlock(&allproc_lock);
334 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
335 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
341 * Increment the count of procs running with this uid. Don't allow
342 * a nonprivileged user to exceed their current limit.
344 * XXXRW: Can we avoid privilege here if it's not needed?
346 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
348 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
351 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
352 lim_cur(p1, RLIMIT_NPROC));
361 * Increment the nprocs resource before blocking can occur. There
362 * are hard-limits as to the number of processes that can run.
367 * Find an unused process ID. We remember a range of unused IDs
368 * ready to use (from lastpid+1 through pidchecked-1).
370 * If RFHIGHPID is set (used during system boot), do not allocate
373 trypid = lastpid + 1;
374 if (flags & RFHIGHPID) {
379 trypid += arc4random() % randompid;
383 * If the process ID prototype has wrapped around,
384 * restart somewhat above 0, as the low-numbered procs
385 * tend to include daemons that don't exit.
387 if (trypid >= PID_MAX) {
388 trypid = trypid % PID_MAX;
393 if (trypid >= pidchecked) {
396 pidchecked = PID_MAX;
398 * Scan the active and zombie procs to check whether this pid
399 * is in use. Remember the lowest pid that's greater
400 * than trypid, so we can avoid checking for a while.
402 p2 = LIST_FIRST(&allproc);
404 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
405 while (p2->p_pid == trypid ||
406 (p2->p_pgrp != NULL &&
407 (p2->p_pgrp->pg_id == trypid ||
408 (p2->p_session != NULL &&
409 p2->p_session->s_sid == trypid)))) {
411 if (trypid >= pidchecked)
414 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
415 pidchecked = p2->p_pid;
416 if (p2->p_pgrp != NULL) {
417 if (p2->p_pgrp->pg_id > trypid &&
418 pidchecked > p2->p_pgrp->pg_id)
419 pidchecked = p2->p_pgrp->pg_id;
420 if (p2->p_session != NULL &&
421 p2->p_session->s_sid > trypid &&
422 pidchecked > p2->p_session->s_sid)
423 pidchecked = p2->p_session->s_sid;
428 p2 = LIST_FIRST(&zombproc);
432 sx_sunlock(&proctree_lock);
435 * RFHIGHPID does not mess with the lastpid counter during boot.
437 if (flags & RFHIGHPID)
443 p2->p_state = PRS_NEW; /* protect against others */
446 * Allow the scheduler to initialize the child.
451 AUDIT_ARG_PID(p2->p_pid);
452 LIST_INSERT_HEAD(&allproc, p2, p_list);
453 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
458 sx_xunlock(&allproc_lock);
460 bcopy(&p1->p_startcopy, &p2->p_startcopy,
461 __rangeof(struct proc, p_startcopy, p_endcopy));
462 pargs_hold(p2->p_args);
465 bzero(&p2->p_startzero,
466 __rangeof(struct proc, p_startzero, p_endzero));
468 p2->p_ucred = crhold(td->td_ucred);
470 /* Tell the prison that we exist. */
471 prison_proc_hold(p2->p_ucred->cr_prison);
476 * Malloc things while we don't hold any locks.
478 if (flags & RFSIGSHARE)
481 newsigacts = sigacts_alloc();
486 if (flags & RFCFDG) {
487 fd = fdinit(p1->p_fd);
489 } else if (flags & RFFDG) {
490 fd = fdcopy(p1->p_fd);
493 fd = fdshare(p1->p_fd);
494 if (p1->p_fdtol == NULL)
496 filedesc_to_leader_alloc(NULL,
499 if ((flags & RFTHREAD) != 0) {
501 * Shared file descriptor table and
502 * shared process leaders.
505 FILEDESC_XLOCK(p1->p_fd);
506 fdtol->fdl_refcount++;
507 FILEDESC_XUNLOCK(p1->p_fd);
510 * Shared file descriptor table, and
511 * different process leaders
513 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
519 * Make a proc table entry for the new process.
520 * Start by zeroing the section of proc that is zero-initialized,
521 * then copy the section that is copied directly from the parent.
527 bzero(&td2->td_startzero,
528 __rangeof(struct thread, td_startzero, td_endzero));
530 bcopy(&td->td_startcopy, &td2->td_startcopy,
531 __rangeof(struct thread, td_startcopy, td_endcopy));
533 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
534 td2->td_sigstk = td->td_sigstk;
535 td2->td_sigmask = td->td_sigmask;
536 td2->td_flags = TDF_INMEM;
540 td2->td_vnet_lpush = NULL;
544 * Duplicate sub-structures as needed.
545 * Increase reference counts on shared objects.
547 p2->p_flag = P_INMEM;
548 p2->p_swtick = ticks;
549 if (p1->p_flag & P_PROFIL)
551 td2->td_ucred = crhold(p2->p_ucred);
553 if (flags & RFSIGSHARE) {
554 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
556 sigacts_copy(newsigacts, p1->p_sigacts);
557 p2->p_sigacts = newsigacts;
559 if (flags & RFLINUXTHPN)
560 p2->p_sigparent = SIGUSR1;
562 p2->p_sigparent = SIGCHLD;
564 p2->p_textvp = p1->p_textvp;
569 * p_limit is copy-on-write. Bump its refcount.
573 pstats_fork(p1->p_stats, p2->p_stats);
578 /* Bump references to the text vnode (for procfs) */
583 * Set up linkage for kernel based threading.
585 if ((flags & RFTHREAD) != 0) {
586 mtx_lock(&ppeers_lock);
587 p2->p_peers = p1->p_peers;
589 p2->p_leader = p1->p_leader;
590 mtx_unlock(&ppeers_lock);
591 PROC_LOCK(p1->p_leader);
592 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
593 PROC_UNLOCK(p1->p_leader);
595 * The task leader is exiting, so process p1 is
596 * going to be killed shortly. Since p1 obviously
597 * isn't dead yet, we know that the leader is either
598 * sending SIGKILL's to all the processes in this
599 * task or is sleeping waiting for all the peers to
600 * exit. We let p1 complete the fork, but we need
601 * to go ahead and kill the new process p2 since
602 * the task leader may not get a chance to send
603 * SIGKILL to it. We leave it on the list so that
604 * the task leader will wait for this new process
608 psignal(p2, SIGKILL);
611 PROC_UNLOCK(p1->p_leader);
617 sx_xlock(&proctree_lock);
618 PGRP_LOCK(p1->p_pgrp);
623 * Preserve some more flags in subprocess. P_PROFIL has already
626 p2->p_flag |= p1->p_flag & P_SUGID;
627 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
628 SESS_LOCK(p1->p_session);
629 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
630 p2->p_flag |= P_CONTROLT;
631 SESS_UNLOCK(p1->p_session);
632 if (flags & RFPPWAIT)
633 p2->p_flag |= P_PPWAIT;
635 p2->p_pgrp = p1->p_pgrp;
636 LIST_INSERT_AFTER(p1, p2, p_pglist);
637 PGRP_UNLOCK(p1->p_pgrp);
638 LIST_INIT(&p2->p_children);
640 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
644 * Copy traceflag and tracefile if enabled.
646 mtx_lock(&ktrace_mtx);
647 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
648 if (p1->p_traceflag & KTRFAC_INHERIT) {
649 p2->p_traceflag = p1->p_traceflag;
650 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
652 KASSERT(p1->p_tracecred != NULL,
653 ("ktrace vnode with no cred"));
654 p2->p_tracecred = crhold(p1->p_tracecred);
657 mtx_unlock(&ktrace_mtx);
661 * If PF_FORK is set, the child process inherits the
662 * procfs ioctl flags from its parent.
664 if (p1->p_pfsflags & PF_FORK) {
665 p2->p_stops = p1->p_stops;
666 p2->p_pfsflags = p1->p_pfsflags;
671 * Tell the DTrace fasttrap provider about the new process
672 * if it has registered an interest.
674 if (dtrace_fasttrap_fork)
675 dtrace_fasttrap_fork(p1, p2);
679 * This begins the section where we must prevent the parent
680 * from being swapped.
686 * Attach the new process to its parent.
688 * If RFNOWAIT is set, the newly created process becomes a child
689 * of init. This effectively disassociates the child from the
692 if (flags & RFNOWAIT)
697 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
698 sx_xunlock(&proctree_lock);
700 /* Inform accounting that we have forked. */
701 p2->p_acflag = AFORK;
705 * Finish creating the child process. It will return via a different
706 * execution path later. (ie: directly into user mode)
708 vm_forkproc(td, p2, td2, vm2, flags);
710 if (flags == (RFFDG | RFPROC)) {
711 PCPU_INC(cnt.v_forks);
712 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
713 p2->p_vmspace->vm_ssize);
714 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
715 PCPU_INC(cnt.v_vforks);
716 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
717 p2->p_vmspace->vm_ssize);
718 } else if (p1 == &proc0) {
719 PCPU_INC(cnt.v_kthreads);
720 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
721 p2->p_vmspace->vm_ssize);
723 PCPU_INC(cnt.v_rforks);
724 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
725 p2->p_vmspace->vm_ssize);
729 * Both processes are set up, now check if any loadable modules want
730 * to adjust anything.
731 * What if they have an error? XXX
733 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
736 * Set the child start time and mark the process as being complete.
738 microuptime(&p2->p_stats->p_start);
740 p2->p_state = PRS_NORMAL;
744 * If RFSTOPPED not requested, make child runnable and add to
747 if ((flags & RFSTOPPED) == 0) {
750 sched_add(td2, SRQ_BORING);
755 * Now can be swapped.
762 * Tell any interested parties about the new process.
764 knote_fork(&p1->p_klist, p2->p_pid);
765 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
768 * Preserve synchronization semantics of vfork. If waiting for
769 * child to exec or exit, set P_PPWAIT on child, and sleep on our
770 * proc (in case of exit).
773 while (p2->p_flag & P_PPWAIT)
774 cv_wait(&p2->p_pwait, &p2->p_mtx);
778 * Return child proc pointer to parent.
783 sx_sunlock(&proctree_lock);
784 if (ppsratecheck(&lastfail, &curfail, 1))
785 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
786 td->td_ucred->cr_ruid);
787 sx_xunlock(&allproc_lock);
789 mac_proc_destroy(newproc);
794 uma_zfree(proc_zone, newproc);
795 pause("fork", hz / 2);
800 * Handle the return of a child process from fork1(). This function
801 * is called from the MD fork_trampoline() entry point.
804 fork_exit(callout, arg, frame)
805 void (*callout)(void *, struct trapframe *);
807 struct trapframe *frame;
815 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
817 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
818 td, td->td_sched, p->p_pid, td->td_name);
822 * Processes normally resume in mi_switch() after being
823 * cpu_switch()'ed to, but when children start up they arrive here
824 * instead, so we must do much the same things as mi_switch() would.
826 if ((dtd = PCPU_GET(deadthread))) {
827 PCPU_SET(deadthread, NULL);
833 * cpu_set_fork_handler intercepts this function call to
834 * have this call a non-return function to stay in kernel mode.
835 * initproc has its own fork handler, but it does return.
837 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
841 * Check if a kernel thread misbehaved and returned from its main
844 if (p->p_flag & P_KTHREAD) {
845 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
846 td->td_name, p->p_pid);
849 mtx_assert(&Giant, MA_NOTOWNED);
851 EVENTHANDLER_INVOKE(schedtail, p);
855 * Simplified back end of syscall(), used when returning from fork()
856 * directly into user mode. Giant is not held on entry, and must not
857 * be held on return. This function is passed in to fork_exit() as the
858 * first parameter and is called when returning to a new userland process.
861 fork_return(td, frame)
863 struct trapframe *frame;
868 if (KTRPOINT(td, KTR_SYSRET))
869 ktrsysret(SYS_fork, 0, 0);
871 mtx_assert(&Giant, MA_NOTOWNED);