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>
55 #include <sys/pioctl.h>
56 #include <sys/resourcevar.h>
57 #include <sys/sched.h>
58 #include <sys/syscall.h>
59 #include <sys/vmmeter.h>
60 #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>
73 #include <vm/vm_map.h>
74 #include <vm/vm_extern.h>
78 #ifndef _SYS_SYSPROTO_H_
84 static int forksleep; /* Place for fork1() to sleep on. */
93 struct fork_args *uap;
98 error = fork1(td, RFFDG | RFPROC, 0, &p2);
100 td->td_retval[0] = p2->p_pid;
101 td->td_retval[1] = 0;
113 struct vfork_args *uap;
118 error = fork1(td, RFFDG | RFPROC | RFPPWAIT | RFMEM, 0, &p2);
120 td->td_retval[0] = p2->p_pid;
121 td->td_retval[1] = 0;
132 struct rfork_args *uap;
137 /* Don't allow kernel-only flags. */
138 if ((uap->flags & RFKERNELONLY) != 0)
141 AUDIT_ARG(fflags, uap->flags);
142 error = fork1(td, uap->flags, 0, &p2);
144 td->td_retval[0] = p2 ? p2->p_pid : 0;
145 td->td_retval[1] = 0;
150 int nprocs = 1; /* process 0 */
152 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
156 * Random component to lastpid generation. We mix in a random factor to make
157 * it a little harder to predict. We sanity check the modulus value to avoid
158 * doing it in critical paths. Don't let it be too small or we pointlessly
159 * waste randomness entropy, and don't let it be impossibly large. Using a
160 * modulus that is too big causes a LOT more process table scans and slows
161 * down fork processing as the pidchecked caching is defeated.
163 static int randompid = 0;
166 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
170 error = sysctl_wire_old_buffer(req, sizeof(int));
173 sx_xlock(&allproc_lock);
175 error = sysctl_handle_int(oidp, &pid, 0, req);
176 if (error == 0 && req->newptr != NULL) {
177 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
179 else if (pid < 2) /* NOP */
181 else if (pid < 100) /* Make it reasonable */
185 sx_xunlock(&allproc_lock);
189 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
190 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
193 fork1(td, flags, pages, procp)
199 struct proc *p1, *p2, *pptr;
201 struct proc *newproc;
203 static int curfail, pidchecked = 0;
204 static struct timeval lastfail;
206 struct filedesc_to_leader *fdtol;
209 struct sigacts *newsigacts;
212 /* Can't copy and clear. */
213 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
219 * Here we don't create a new process, but we divorce
220 * certain parts of a process from itself.
222 if ((flags & RFPROC) == 0) {
223 if ((p1->p_flag & P_HADTHREADS) &&
224 (flags & (RFCFDG | RFFDG))) {
226 if (thread_single(SINGLE_BOUNDARY)) {
233 vm_forkproc(td, NULL, NULL, flags);
236 * Close all file descriptors.
238 if (flags & RFCFDG) {
239 struct filedesc *fdtmp;
240 fdtmp = fdinit(td->td_proc->p_fd);
246 * Unshare file descriptors (from parent).
251 if ((p1->p_flag & P_HADTHREADS) &&
252 (flags & (RFCFDG | RFFDG))) {
262 * Note 1:1 allows for forking with one thread coming out on the
263 * other side with the expectation that the process is about to
266 if (p1->p_flag & P_HADTHREADS) {
268 * Idle the other threads for a second.
269 * Since the user space is copied, it must remain stable.
270 * In addition, all threads (from the user perspective)
271 * need to either be suspended or in the kernel,
272 * where they will try restart in the parent and will
273 * be aborted in the child.
276 if (thread_single(SINGLE_NO_EXIT)) {
277 /* Abort. Someone else is single threading before us. */
283 * All other activity in this process
284 * is now suspended at the user boundary,
285 * (or other safe places if we think of any).
289 /* Allocate new proc. */
290 newproc = uma_zalloc(proc_zone, M_WAITOK);
292 mac_init_proc(newproc);
295 audit_proc_alloc(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 uid = td->td_ucred->cr_ruid;
312 if ((nprocs >= maxproc - 10 &&
313 suser_cred(td->td_ucred, SUSER_RUID) != 0) ||
320 * Increment the count of procs running with this uid. Don't allow
321 * a nonprivileged user to exceed their current limit.
324 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
325 (uid != 0) ? lim_cur(p1, RLIMIT_NPROC) : 0);
333 * Increment the nprocs resource before blocking can occur. There
334 * are hard-limits as to the number of processes that can run.
339 * Find an unused process ID. We remember a range of unused IDs
340 * ready to use (from lastpid+1 through pidchecked-1).
342 * If RFHIGHPID is set (used during system boot), do not allocate
345 trypid = lastpid + 1;
346 if (flags & RFHIGHPID) {
351 trypid += arc4random() % randompid;
355 * If the process ID prototype has wrapped around,
356 * restart somewhat above 0, as the low-numbered procs
357 * tend to include daemons that don't exit.
359 if (trypid >= PID_MAX) {
360 trypid = trypid % PID_MAX;
365 if (trypid >= pidchecked) {
368 pidchecked = PID_MAX;
370 * Scan the active and zombie procs to check whether this pid
371 * is in use. Remember the lowest pid that's greater
372 * than trypid, so we can avoid checking for a while.
374 p2 = LIST_FIRST(&allproc);
376 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
378 while (p2->p_pid == trypid ||
379 (p2->p_pgrp != NULL &&
380 (p2->p_pgrp->pg_id == trypid ||
381 (p2->p_session != NULL &&
382 p2->p_session->s_sid == trypid)))) {
384 if (trypid >= pidchecked) {
389 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
390 pidchecked = p2->p_pid;
391 if (p2->p_pgrp != NULL) {
392 if (p2->p_pgrp->pg_id > trypid &&
393 pidchecked > p2->p_pgrp->pg_id)
394 pidchecked = p2->p_pgrp->pg_id;
395 if (p2->p_session != NULL &&
396 p2->p_session->s_sid > trypid &&
397 pidchecked > p2->p_session->s_sid)
398 pidchecked = p2->p_session->s_sid;
404 p2 = LIST_FIRST(&zombproc);
408 sx_sunlock(&proctree_lock);
411 * RFHIGHPID does not mess with the lastpid counter during boot.
413 if (flags & RFHIGHPID)
419 p2->p_state = PRS_NEW; /* protect against others */
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);
424 sx_xunlock(&allproc_lock);
427 * Malloc things while we don't hold any locks.
429 if (flags & RFSIGSHARE)
432 newsigacts = sigacts_alloc();
437 if (flags & RFCFDG) {
438 fd = fdinit(p1->p_fd);
440 } else if (flags & RFFDG) {
441 fd = fdcopy(p1->p_fd);
444 fd = fdshare(p1->p_fd);
445 if (p1->p_fdtol == NULL)
447 filedesc_to_leader_alloc(NULL,
450 if ((flags & RFTHREAD) != 0) {
452 * Shared file descriptor table and
453 * shared process leaders.
456 FILEDESC_LOCK_FAST(p1->p_fd);
457 fdtol->fdl_refcount++;
458 FILEDESC_UNLOCK_FAST(p1->p_fd);
461 * Shared file descriptor table, and
462 * different process leaders
464 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
470 * Make a proc table entry for the new process.
471 * Start by zeroing the section of proc that is zero-initialized,
472 * then copy the section that is copied directly from the parent.
474 td2 = FIRST_THREAD_IN_PROC(p2);
475 kg2 = FIRST_KSEGRP_IN_PROC(p2);
477 /* Allocate and switch to an alternate kstack if specified. */
479 vm_thread_new_altkstack(td2, pages);
484 bzero(&p2->p_startzero,
485 __rangeof(struct proc, p_startzero, p_endzero));
486 bzero(&td2->td_startzero,
487 __rangeof(struct thread, td_startzero, td_endzero));
488 bzero(&kg2->kg_startzero,
489 __rangeof(struct ksegrp, kg_startzero, kg_endzero));
491 bcopy(&p1->p_startcopy, &p2->p_startcopy,
492 __rangeof(struct proc, p_startcopy, p_endcopy));
493 bcopy(&td->td_startcopy, &td2->td_startcopy,
494 __rangeof(struct thread, td_startcopy, td_endcopy));
495 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
496 __rangeof(struct ksegrp, kg_startcopy, kg_endcopy));
498 td2->td_sigstk = td->td_sigstk;
499 td2->td_sigmask = td->td_sigmask;
502 * Duplicate sub-structures as needed.
503 * Increase reference counts on shared objects.
506 if (p1->p_flag & P_PROFIL)
508 mtx_lock_spin(&sched_lock);
509 p2->p_sflag = PS_INMEM;
511 * Allow the scheduler to adjust the priority of the child and
512 * parent while we hold the sched_lock.
516 mtx_unlock_spin(&sched_lock);
517 p2->p_ucred = crhold(td->td_ucred);
518 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */
520 audit_proc_fork(p1, p2);
522 pargs_hold(p2->p_args);
524 if (flags & RFSIGSHARE) {
525 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
527 sigacts_copy(newsigacts, p1->p_sigacts);
528 p2->p_sigacts = newsigacts;
530 if (flags & RFLINUXTHPN)
531 p2->p_sigparent = SIGUSR1;
533 p2->p_sigparent = SIGCHLD;
535 p2->p_textvp = p1->p_textvp;
540 * p_limit is copy-on-write. Bump its refcount.
542 p2->p_limit = lim_hold(p1->p_limit);
544 pstats_fork(p1->p_stats, p2->p_stats);
549 /* Bump references to the text vnode (for procfs) */
554 * Set up linkage for kernel based threading.
556 if ((flags & RFTHREAD) != 0) {
557 mtx_lock(&ppeers_lock);
558 p2->p_peers = p1->p_peers;
560 p2->p_leader = p1->p_leader;
561 mtx_unlock(&ppeers_lock);
562 PROC_LOCK(p1->p_leader);
563 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
564 PROC_UNLOCK(p1->p_leader);
566 * The task leader is exiting, so process p1 is
567 * going to be killed shortly. Since p1 obviously
568 * isn't dead yet, we know that the leader is either
569 * sending SIGKILL's to all the processes in this
570 * task or is sleeping waiting for all the peers to
571 * exit. We let p1 complete the fork, but we need
572 * to go ahead and kill the new process p2 since
573 * the task leader may not get a chance to send
574 * SIGKILL to it. We leave it on the list so that
575 * the task leader will wait for this new process
579 psignal(p2, SIGKILL);
582 PROC_UNLOCK(p1->p_leader);
588 sx_xlock(&proctree_lock);
589 PGRP_LOCK(p1->p_pgrp);
594 * Preserve some more flags in subprocess. P_PROFIL has already
597 p2->p_flag |= p1->p_flag & P_SUGID;
598 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
599 SESS_LOCK(p1->p_session);
600 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
601 p2->p_flag |= P_CONTROLT;
602 SESS_UNLOCK(p1->p_session);
603 if (flags & RFPPWAIT)
604 p2->p_flag |= P_PPWAIT;
606 p2->p_pgrp = p1->p_pgrp;
607 LIST_INSERT_AFTER(p1, p2, p_pglist);
608 PGRP_UNLOCK(p1->p_pgrp);
609 LIST_INIT(&p2->p_children);
611 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
615 * Copy traceflag and tracefile if enabled.
617 mtx_lock(&ktrace_mtx);
618 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
619 if (p1->p_traceflag & KTRFAC_INHERIT) {
620 p2->p_traceflag = p1->p_traceflag;
621 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
623 KASSERT(p1->p_tracecred != NULL,
624 ("ktrace vnode with no cred"));
625 p2->p_tracecred = crhold(p1->p_tracecred);
628 mtx_unlock(&ktrace_mtx);
632 * If PF_FORK is set, the child process inherits the
633 * procfs ioctl flags from its parent.
635 if (p1->p_pfsflags & PF_FORK) {
636 p2->p_stops = p1->p_stops;
637 p2->p_pfsflags = p1->p_pfsflags;
641 * This begins the section where we must prevent the parent
642 * from being swapped.
648 * Attach the new process to its parent.
650 * If RFNOWAIT is set, the newly created process becomes a child
651 * of init. This effectively disassociates the child from the
654 if (flags & RFNOWAIT)
659 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
660 sx_xunlock(&proctree_lock);
662 /* Inform accounting that we have forked. */
663 p2->p_acflag = AFORK;
667 * Finish creating the child process. It will return via a different
668 * execution path later. (ie: directly into user mode)
670 vm_forkproc(td, p2, td2, flags);
672 if (flags == (RFFDG | RFPROC)) {
673 atomic_add_int(&cnt.v_forks, 1);
674 atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize +
675 p2->p_vmspace->vm_ssize);
676 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
677 atomic_add_int(&cnt.v_vforks, 1);
678 atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
679 p2->p_vmspace->vm_ssize);
680 } else if (p1 == &proc0) {
681 atomic_add_int(&cnt.v_kthreads, 1);
682 atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
683 p2->p_vmspace->vm_ssize);
685 atomic_add_int(&cnt.v_rforks, 1);
686 atomic_add_int(&cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
687 p2->p_vmspace->vm_ssize);
691 * Both processes are set up, now check if any loadable modules want
692 * to adjust anything.
693 * What if they have an error? XXX
695 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
698 * Set the child start time and mark the process as being complete.
700 microuptime(&p2->p_stats->p_start);
701 mtx_lock_spin(&sched_lock);
702 p2->p_state = PRS_NORMAL;
705 * If RFSTOPPED not requested, make child runnable and add to
708 if ((flags & RFSTOPPED) == 0) {
710 setrunqueue(td2, SRQ_BORING);
712 mtx_unlock_spin(&sched_lock);
715 * Now can be swapped.
721 * Tell any interested parties about the new process.
723 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
728 * Preserve synchronization semantics of vfork. If waiting for
729 * child to exec or exit, set P_PPWAIT on child, and sleep on our
730 * proc (in case of exit).
733 while (p2->p_flag & P_PPWAIT)
734 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
738 * If other threads are waiting, let them continue now.
740 if (p1->p_flag & P_HADTHREADS) {
747 * Return child proc pointer to parent.
752 sx_sunlock(&proctree_lock);
753 if (ppsratecheck(&lastfail, &curfail, 1))
754 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
756 sx_xunlock(&allproc_lock);
758 mac_destroy_proc(newproc);
761 audit_proc_free(newproc);
763 uma_zfree(proc_zone, newproc);
764 if (p1->p_flag & P_HADTHREADS) {
769 tsleep(&forksleep, PUSER, "fork", hz / 2);
774 * Handle the return of a child process from fork1(). This function
775 * is called from the MD fork_trampoline() entry point.
778 fork_exit(callout, arg, frame)
779 void (*callout)(void *, struct trapframe *);
781 struct trapframe *frame;
787 * Finish setting up thread glue so that it begins execution in a
788 * non-nested critical section with sched_lock held but not recursed.
792 td->td_oncpu = PCPU_GET(cpuid);
793 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
795 sched_lock.mtx_lock = (uintptr_t)td;
796 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
797 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
798 td, td->td_sched, p->p_pid, p->p_comm);
801 * Processes normally resume in mi_switch() after being
802 * cpu_switch()'ed to, but when children start up they arrive here
803 * instead, so we must do much the same things as mi_switch() would.
806 if ((td = PCPU_GET(deadthread))) {
807 PCPU_SET(deadthread, NULL);
811 mtx_unlock_spin(&sched_lock);
814 * cpu_set_fork_handler intercepts this function call to
815 * have this call a non-return function to stay in kernel mode.
816 * initproc has its own fork handler, but it does return.
818 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
822 * Check if a kernel thread misbehaved and returned from its main
825 if (p->p_flag & P_KTHREAD) {
826 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
827 p->p_comm, p->p_pid);
830 mtx_assert(&Giant, MA_NOTOWNED);
834 * Simplified back end of syscall(), used when returning from fork()
835 * directly into user mode. Giant is not held on entry, and must not
836 * be held on return. This function is passed in to fork_exit() as the
837 * first parameter and is called when returning to a new userland process.
840 fork_return(td, frame)
842 struct trapframe *frame;
847 if (KTRPOINT(td, KTR_SYSRET))
848 ktrsysret(SYS_fork, 0, 0);
850 mtx_assert(&Giant, MA_NOTOWNED);