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;
200 /* Can't copy and clear. */
201 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
207 * Here we don't create a new process, but we divorce
208 * certain parts of a process from itself.
210 if ((flags & RFPROC) == 0) {
211 if ((p1->p_flag & P_HADTHREADS) &&
212 (flags & (RFCFDG | RFFDG))) {
214 if (thread_single(SINGLE_BOUNDARY)) {
221 vm_forkproc(td, NULL, NULL, flags);
224 * Close all file descriptors.
226 if (flags & RFCFDG) {
227 struct filedesc *fdtmp;
228 fdtmp = fdinit(td->td_proc->p_fd);
234 * Unshare file descriptors (from parent).
239 if ((p1->p_flag & P_HADTHREADS) &&
240 (flags & (RFCFDG | RFFDG))) {
250 * Note 1:1 allows for forking with one thread coming out on the
251 * other side with the expectation that the process is about to
254 if (p1->p_flag & P_HADTHREADS) {
256 * Idle the other threads for a second.
257 * Since the user space is copied, it must remain stable.
258 * In addition, all threads (from the user perspective)
259 * need to either be suspended or in the kernel,
260 * where they will try restart in the parent and will
261 * be aborted in the child.
264 if (thread_single(SINGLE_NO_EXIT)) {
265 /* Abort. Someone else is single threading before us. */
271 * All other activity in this process
272 * is now suspended at the user boundary,
273 * (or other safe places if we think of any).
277 /* Allocate new proc. */
278 newproc = uma_zalloc(proc_zone, M_WAITOK);
280 mac_init_proc(newproc);
283 audit_proc_alloc(newproc);
285 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
286 STAILQ_INIT(&newproc->p_ktr);
288 /* We have to lock the process tree while we look for a pid. */
289 sx_slock(&proctree_lock);
292 * Although process entries are dynamically created, we still keep
293 * a global limit on the maximum number we will create. Don't allow
294 * a nonprivileged user to use the last ten processes; don't let root
295 * exceed the limit. The variable nprocs is the current number of
296 * processes, maxproc is the limit.
298 sx_xlock(&allproc_lock);
299 if ((nprocs >= maxproc - 10 &&
300 priv_check_cred(td->td_ucred, PRIV_MAXPROC, SUSER_RUID) != 0) ||
307 * Increment the count of procs running with this uid. Don't allow
308 * a nonprivileged user to exceed their current limit.
310 * XXXRW: Can we avoid privilege here if it's not needed?
312 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, SUSER_RUID |
315 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
318 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
319 lim_cur(p1, RLIMIT_NPROC));
328 * Increment the nprocs resource before blocking can occur. There
329 * are hard-limits as to the number of processes that can run.
334 * Find an unused process ID. We remember a range of unused IDs
335 * ready to use (from lastpid+1 through pidchecked-1).
337 * If RFHIGHPID is set (used during system boot), do not allocate
340 trypid = lastpid + 1;
341 if (flags & RFHIGHPID) {
346 trypid += arc4random() % randompid;
350 * If the process ID prototype has wrapped around,
351 * restart somewhat above 0, as the low-numbered procs
352 * tend to include daemons that don't exit.
354 if (trypid >= PID_MAX) {
355 trypid = trypid % PID_MAX;
360 if (trypid >= pidchecked) {
363 pidchecked = PID_MAX;
365 * Scan the active and zombie procs to check whether this pid
366 * is in use. Remember the lowest pid that's greater
367 * than trypid, so we can avoid checking for a while.
369 p2 = LIST_FIRST(&allproc);
371 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
372 while (p2->p_pid == trypid ||
373 (p2->p_pgrp != NULL &&
374 (p2->p_pgrp->pg_id == trypid ||
375 (p2->p_session != NULL &&
376 p2->p_session->s_sid == trypid)))) {
378 if (trypid >= pidchecked)
381 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
382 pidchecked = p2->p_pid;
383 if (p2->p_pgrp != NULL) {
384 if (p2->p_pgrp->pg_id > trypid &&
385 pidchecked > p2->p_pgrp->pg_id)
386 pidchecked = p2->p_pgrp->pg_id;
387 if (p2->p_session != NULL &&
388 p2->p_session->s_sid > trypid &&
389 pidchecked > p2->p_session->s_sid)
390 pidchecked = p2->p_session->s_sid;
395 p2 = LIST_FIRST(&zombproc);
399 sx_sunlock(&proctree_lock);
402 * RFHIGHPID does not mess with the lastpid counter during boot.
404 if (flags & RFHIGHPID)
410 p2->p_state = PRS_NEW; /* protect against others */
412 AUDIT_ARG(pid, p2->p_pid);
413 LIST_INSERT_HEAD(&allproc, p2, p_list);
414 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
419 sx_xunlock(&allproc_lock);
421 bcopy(&p1->p_startcopy, &p2->p_startcopy,
422 __rangeof(struct proc, p_startcopy, p_endcopy));
425 bzero(&p2->p_startzero,
426 __rangeof(struct proc, p_startzero, p_endzero));
428 p2->p_ucred = crhold(td->td_ucred);
432 * Malloc things while we don't hold any locks.
434 if (flags & RFSIGSHARE)
437 newsigacts = sigacts_alloc();
442 if (flags & RFCFDG) {
443 fd = fdinit(p1->p_fd);
445 } else if (flags & RFFDG) {
446 fd = fdcopy(p1->p_fd);
449 fd = fdshare(p1->p_fd);
450 if (p1->p_fdtol == NULL)
452 filedesc_to_leader_alloc(NULL,
455 if ((flags & RFTHREAD) != 0) {
457 * Shared file descriptor table and
458 * shared process leaders.
461 FILEDESC_XLOCK(p1->p_fd);
462 fdtol->fdl_refcount++;
463 FILEDESC_XUNLOCK(p1->p_fd);
466 * Shared file descriptor table, and
467 * different process leaders
469 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
475 * Make a proc table entry for the new process.
476 * Start by zeroing the section of proc that is zero-initialized,
477 * then copy the section that is copied directly from the parent.
479 td2 = FIRST_THREAD_IN_PROC(p2);
481 /* Allocate and switch to an alternate kstack if specified. */
483 vm_thread_new_altkstack(td2, pages);
488 bzero(&td2->td_startzero,
489 __rangeof(struct thread, td_startzero, td_endzero));
491 bcopy(&td->td_startcopy, &td2->td_startcopy,
492 __rangeof(struct thread, td_startcopy, td_endcopy));
494 td2->td_sigstk = td->td_sigstk;
495 td2->td_sigmask = td->td_sigmask;
498 * Duplicate sub-structures as needed.
499 * Increase reference counts on shared objects.
502 if (p1->p_flag & P_PROFIL)
504 mtx_lock_spin(&sched_lock);
505 p2->p_sflag = PS_INMEM;
507 * Allow the scheduler to adjust the priority of the child and
508 * parent while we hold the sched_lock.
512 mtx_unlock_spin(&sched_lock);
513 td2->td_ucred = crhold(p2->p_ucred);
515 audit_proc_fork(p1, p2);
517 pargs_hold(p2->p_args);
519 if (flags & RFSIGSHARE) {
520 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
522 sigacts_copy(newsigacts, p1->p_sigacts);
523 p2->p_sigacts = newsigacts;
525 if (flags & RFLINUXTHPN)
526 p2->p_sigparent = SIGUSR1;
528 p2->p_sigparent = SIGCHLD;
530 p2->p_textvp = p1->p_textvp;
535 * p_limit is copy-on-write. Bump its refcount.
537 p2->p_limit = lim_hold(p1->p_limit);
539 pstats_fork(p1->p_stats, p2->p_stats);
544 /* Bump references to the text vnode (for procfs) */
549 * Set up linkage for kernel based threading.
551 if ((flags & RFTHREAD) != 0) {
552 mtx_lock(&ppeers_lock);
553 p2->p_peers = p1->p_peers;
555 p2->p_leader = p1->p_leader;
556 mtx_unlock(&ppeers_lock);
557 PROC_LOCK(p1->p_leader);
558 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
559 PROC_UNLOCK(p1->p_leader);
561 * The task leader is exiting, so process p1 is
562 * going to be killed shortly. Since p1 obviously
563 * isn't dead yet, we know that the leader is either
564 * sending SIGKILL's to all the processes in this
565 * task or is sleeping waiting for all the peers to
566 * exit. We let p1 complete the fork, but we need
567 * to go ahead and kill the new process p2 since
568 * the task leader may not get a chance to send
569 * SIGKILL to it. We leave it on the list so that
570 * the task leader will wait for this new process
574 psignal(p2, SIGKILL);
577 PROC_UNLOCK(p1->p_leader);
583 sx_xlock(&proctree_lock);
584 PGRP_LOCK(p1->p_pgrp);
589 * Preserve some more flags in subprocess. P_PROFIL has already
592 p2->p_flag |= p1->p_flag & P_SUGID;
593 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
594 SESS_LOCK(p1->p_session);
595 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
596 p2->p_flag |= P_CONTROLT;
597 SESS_UNLOCK(p1->p_session);
598 if (flags & RFPPWAIT)
599 p2->p_flag |= P_PPWAIT;
601 p2->p_pgrp = p1->p_pgrp;
602 LIST_INSERT_AFTER(p1, p2, p_pglist);
603 PGRP_UNLOCK(p1->p_pgrp);
604 LIST_INIT(&p2->p_children);
606 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
610 * Copy traceflag and tracefile if enabled.
612 mtx_lock(&ktrace_mtx);
613 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
614 if (p1->p_traceflag & KTRFAC_INHERIT) {
615 p2->p_traceflag = p1->p_traceflag;
616 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
618 KASSERT(p1->p_tracecred != NULL,
619 ("ktrace vnode with no cred"));
620 p2->p_tracecred = crhold(p1->p_tracecred);
623 mtx_unlock(&ktrace_mtx);
627 * If PF_FORK is set, the child process inherits the
628 * procfs ioctl flags from its parent.
630 if (p1->p_pfsflags & PF_FORK) {
631 p2->p_stops = p1->p_stops;
632 p2->p_pfsflags = p1->p_pfsflags;
636 * This begins the section where we must prevent the parent
637 * from being swapped.
643 * Attach the new process to its parent.
645 * If RFNOWAIT is set, the newly created process becomes a child
646 * of init. This effectively disassociates the child from the
649 if (flags & RFNOWAIT)
654 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
655 sx_xunlock(&proctree_lock);
657 /* Inform accounting that we have forked. */
658 p2->p_acflag = AFORK;
662 * Finish creating the child process. It will return via a different
663 * execution path later. (ie: directly into user mode)
665 vm_forkproc(td, p2, td2, flags);
667 if (flags == (RFFDG | RFPROC)) {
669 VMCNT_ADD(forkpages, p2->p_vmspace->vm_dsize +
670 p2->p_vmspace->vm_ssize);
671 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
673 VMCNT_ADD(forkpages, p2->p_vmspace->vm_dsize +
674 p2->p_vmspace->vm_ssize);
675 } else if (p1 == &proc0) {
676 VMCNT_ADD(kthreads, 1);
677 VMCNT_ADD(kthreadpages, p2->p_vmspace->vm_dsize +
678 p2->p_vmspace->vm_ssize);
680 VMCNT_ADD(rforks, 1);
681 VMCNT_ADD(rforkpages, p2->p_vmspace->vm_dsize +
682 p2->p_vmspace->vm_ssize);
686 * Both processes are set up, now check if any loadable modules want
687 * to adjust anything.
688 * What if they have an error? XXX
690 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
693 * Set the child start time and mark the process as being complete.
695 microuptime(&p2->p_stats->p_start);
696 mtx_lock_spin(&sched_lock);
697 p2->p_state = PRS_NORMAL;
700 * If RFSTOPPED not requested, make child runnable and add to
703 if ((flags & RFSTOPPED) == 0) {
705 sched_add(td2, SRQ_BORING);
707 mtx_unlock_spin(&sched_lock);
710 * Now can be swapped.
716 * Tell any interested parties about the new process.
718 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
723 * Preserve synchronization semantics of vfork. If waiting for
724 * child to exec or exit, set P_PPWAIT on child, and sleep on our
725 * proc (in case of exit).
728 while (p2->p_flag & P_PPWAIT)
729 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
733 * If other threads are waiting, let them continue now.
735 if (p1->p_flag & P_HADTHREADS) {
742 * Return child proc pointer to parent.
747 sx_sunlock(&proctree_lock);
748 if (ppsratecheck(&lastfail, &curfail, 1))
749 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
750 td->td_ucred->cr_ruid);
751 sx_xunlock(&allproc_lock);
753 mac_destroy_proc(newproc);
756 audit_proc_free(newproc);
758 uma_zfree(proc_zone, newproc);
759 if (p1->p_flag & P_HADTHREADS) {
764 pause("fork", hz / 2);
769 * Handle the return of a child process from fork1(). This function
770 * is called from the MD fork_trampoline() entry point.
773 fork_exit(callout, arg, frame)
774 void (*callout)(void *, struct trapframe *);
776 struct trapframe *frame;
782 * Finish setting up thread glue so that it begins execution in a
783 * non-nested critical section with sched_lock held but not recursed.
787 td->td_oncpu = PCPU_GET(cpuid);
788 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
790 sched_lock.mtx_lock = (uintptr_t)td;
791 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
792 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
793 td, td->td_sched, p->p_pid, p->p_comm);
796 * Processes normally resume in mi_switch() after being
797 * cpu_switch()'ed to, but when children start up they arrive here
798 * instead, so we must do much the same things as mi_switch() would.
800 if ((td = PCPU_GET(deadthread))) {
801 PCPU_SET(deadthread, NULL);
804 mtx_unlock_spin(&sched_lock);
807 * cpu_set_fork_handler intercepts this function call to
808 * have this call a non-return function to stay in kernel mode.
809 * initproc has its own fork handler, but it does return.
811 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
815 * Check if a kernel thread misbehaved and returned from its main
818 if (p->p_flag & P_KTHREAD) {
819 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
820 p->p_comm, p->p_pid);
823 mtx_assert(&Giant, MA_NOTOWNED);
825 EVENTHANDLER_INVOKE(schedtail, p);
829 * Simplified back end of syscall(), used when returning from fork()
830 * directly into user mode. Giant is not held on entry, and must not
831 * be held on return. This function is passed in to fork_exit() as the
832 * first parameter and is called when returning to a new userland process.
835 fork_return(td, frame)
837 struct trapframe *frame;
842 if (KTRPOINT(td, KTR_SYSRET))
843 ktrsysret(SYS_fork, 0, 0);
845 mtx_assert(&Giant, MA_NOTOWNED);