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 vm_forkproc(td, NULL, NULL, flags);
226 * Close all file descriptors.
228 if (flags & RFCFDG) {
229 struct filedesc *fdtmp;
230 fdtmp = fdinit(td->td_proc->p_fd);
236 * Unshare file descriptors (from parent).
245 * Note 1:1 allows for forking with one thread coming out on the
246 * other side with the expectation that the process is about to
249 if (p1->p_flag & P_HADTHREADS) {
251 * Idle the other threads for a second.
252 * Since the user space is copied, it must remain stable.
253 * In addition, all threads (from the user perspective)
254 * need to either be suspended or in the kernel,
255 * where they will try restart in the parent and will
256 * be aborted in the child.
259 if (thread_single(SINGLE_NO_EXIT)) {
260 /* Abort. Someone else is single threading before us. */
266 * All other activity in this process
267 * is now suspended at the user boundary,
268 * (or other safe places if we think of any).
272 /* Allocate new proc. */
273 newproc = uma_zalloc(proc_zone, M_WAITOK);
275 mac_init_proc(newproc);
278 audit_proc_alloc(newproc);
280 knlist_init(&newproc->p_klist, &newproc->p_mtx, NULL, NULL, NULL);
281 STAILQ_INIT(&newproc->p_ktr);
283 /* We have to lock the process tree while we look for a pid. */
284 sx_slock(&proctree_lock);
287 * Although process entries are dynamically created, we still keep
288 * a global limit on the maximum number we will create. Don't allow
289 * a nonprivileged user to use the last ten processes; don't let root
290 * exceed the limit. The variable nprocs is the current number of
291 * processes, maxproc is the limit.
293 sx_xlock(&allproc_lock);
294 uid = td->td_ucred->cr_ruid;
295 if ((nprocs >= maxproc - 10 &&
296 suser_cred(td->td_ucred, SUSER_RUID) != 0) ||
303 * Increment the count of procs running with this uid. Don't allow
304 * a nonprivileged user to exceed their current limit.
307 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
308 (uid != 0) ? lim_cur(p1, RLIMIT_NPROC) : 0);
316 * Increment the nprocs resource before blocking can occur. There
317 * are hard-limits as to the number of processes that can run.
322 * Find an unused process ID. We remember a range of unused IDs
323 * ready to use (from lastpid+1 through pidchecked-1).
325 * If RFHIGHPID is set (used during system boot), do not allocate
328 trypid = lastpid + 1;
329 if (flags & RFHIGHPID) {
334 trypid += arc4random() % randompid;
338 * If the process ID prototype has wrapped around,
339 * restart somewhat above 0, as the low-numbered procs
340 * tend to include daemons that don't exit.
342 if (trypid >= PID_MAX) {
343 trypid = trypid % PID_MAX;
348 if (trypid >= pidchecked) {
351 pidchecked = PID_MAX;
353 * Scan the active and zombie procs to check whether this pid
354 * is in use. Remember the lowest pid that's greater
355 * than trypid, so we can avoid checking for a while.
357 p2 = LIST_FIRST(&allproc);
359 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
361 while (p2->p_pid == trypid ||
362 (p2->p_pgrp != NULL &&
363 (p2->p_pgrp->pg_id == trypid ||
364 (p2->p_session != NULL &&
365 p2->p_session->s_sid == trypid)))) {
367 if (trypid >= pidchecked) {
372 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
373 pidchecked = p2->p_pid;
374 if (p2->p_pgrp != NULL) {
375 if (p2->p_pgrp->pg_id > trypid &&
376 pidchecked > p2->p_pgrp->pg_id)
377 pidchecked = p2->p_pgrp->pg_id;
378 if (p2->p_session != NULL &&
379 p2->p_session->s_sid > trypid &&
380 pidchecked > p2->p_session->s_sid)
381 pidchecked = p2->p_session->s_sid;
387 p2 = LIST_FIRST(&zombproc);
391 sx_sunlock(&proctree_lock);
394 * RFHIGHPID does not mess with the lastpid counter during boot.
396 if (flags & RFHIGHPID)
402 p2->p_state = PRS_NEW; /* protect against others */
404 AUDIT_ARG(pid, p2->p_pid);
405 LIST_INSERT_HEAD(&allproc, p2, p_list);
406 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
407 sx_xunlock(&allproc_lock);
410 * Malloc things while we don't hold any locks.
412 if (flags & RFSIGSHARE)
415 newsigacts = sigacts_alloc();
420 if (flags & RFCFDG) {
421 fd = fdinit(p1->p_fd);
423 } else if (flags & RFFDG) {
424 fd = fdcopy(p1->p_fd);
427 fd = fdshare(p1->p_fd);
428 if (p1->p_fdtol == NULL)
430 filedesc_to_leader_alloc(NULL,
433 if ((flags & RFTHREAD) != 0) {
435 * Shared file descriptor table and
436 * shared process leaders.
439 FILEDESC_LOCK_FAST(p1->p_fd);
440 fdtol->fdl_refcount++;
441 FILEDESC_UNLOCK_FAST(p1->p_fd);
444 * Shared file descriptor table, and
445 * different process leaders
447 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
453 * Make a proc table entry for the new process.
454 * Start by zeroing the section of proc that is zero-initialized,
455 * then copy the section that is copied directly from the parent.
457 td2 = FIRST_THREAD_IN_PROC(p2);
458 kg2 = FIRST_KSEGRP_IN_PROC(p2);
460 /* Allocate and switch to an alternate kstack if specified. */
462 vm_thread_new_altkstack(td2, pages);
467 bzero(&p2->p_startzero,
468 __rangeof(struct proc, p_startzero, p_endzero));
469 bzero(&td2->td_startzero,
470 __rangeof(struct thread, td_startzero, td_endzero));
471 bzero(&kg2->kg_startzero,
472 __rangeof(struct ksegrp, kg_startzero, kg_endzero));
474 bcopy(&p1->p_startcopy, &p2->p_startcopy,
475 __rangeof(struct proc, p_startcopy, p_endcopy));
476 bcopy(&td->td_startcopy, &td2->td_startcopy,
477 __rangeof(struct thread, td_startcopy, td_endcopy));
478 bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
479 __rangeof(struct ksegrp, kg_startcopy, kg_endcopy));
481 td2->td_sigstk = td->td_sigstk;
482 td2->td_sigmask = td->td_sigmask;
485 * Duplicate sub-structures as needed.
486 * Increase reference counts on shared objects.
489 if (p1->p_flag & P_PROFIL)
491 mtx_lock_spin(&sched_lock);
492 p2->p_sflag = PS_INMEM;
494 * Allow the scheduler to adjust the priority of the child and
495 * parent while we hold the sched_lock.
499 mtx_unlock_spin(&sched_lock);
500 p2->p_ucred = crhold(td->td_ucred);
501 td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */
503 audit_proc_fork(p1, p2);
505 pargs_hold(p2->p_args);
507 if (flags & RFSIGSHARE) {
508 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
510 sigacts_copy(newsigacts, p1->p_sigacts);
511 p2->p_sigacts = newsigacts;
513 if (flags & RFLINUXTHPN)
514 p2->p_sigparent = SIGUSR1;
516 p2->p_sigparent = SIGCHLD;
518 p2->p_textvp = p1->p_textvp;
523 * p_limit is copy-on-write. Bump its refcount.
525 p2->p_limit = lim_hold(p1->p_limit);
527 pstats_fork(p1->p_stats, p2->p_stats);
532 /* Bump references to the text vnode (for procfs) */
537 * Set up linkage for kernel based threading.
539 if ((flags & RFTHREAD) != 0) {
540 mtx_lock(&ppeers_lock);
541 p2->p_peers = p1->p_peers;
543 p2->p_leader = p1->p_leader;
544 mtx_unlock(&ppeers_lock);
545 PROC_LOCK(p1->p_leader);
546 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
547 PROC_UNLOCK(p1->p_leader);
549 * The task leader is exiting, so process p1 is
550 * going to be killed shortly. Since p1 obviously
551 * isn't dead yet, we know that the leader is either
552 * sending SIGKILL's to all the processes in this
553 * task or is sleeping waiting for all the peers to
554 * exit. We let p1 complete the fork, but we need
555 * to go ahead and kill the new process p2 since
556 * the task leader may not get a chance to send
557 * SIGKILL to it. We leave it on the list so that
558 * the task leader will wait for this new process
562 psignal(p2, SIGKILL);
565 PROC_UNLOCK(p1->p_leader);
571 sx_xlock(&proctree_lock);
572 PGRP_LOCK(p1->p_pgrp);
577 * Preserve some more flags in subprocess. P_PROFIL has already
580 p2->p_flag |= p1->p_flag & P_SUGID;
581 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
582 SESS_LOCK(p1->p_session);
583 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
584 p2->p_flag |= P_CONTROLT;
585 SESS_UNLOCK(p1->p_session);
586 if (flags & RFPPWAIT)
587 p2->p_flag |= P_PPWAIT;
589 p2->p_pgrp = p1->p_pgrp;
590 LIST_INSERT_AFTER(p1, p2, p_pglist);
591 PGRP_UNLOCK(p1->p_pgrp);
592 LIST_INIT(&p2->p_children);
594 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
598 * Copy traceflag and tracefile if enabled.
600 mtx_lock(&ktrace_mtx);
601 KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode"));
602 if (p1->p_traceflag & KTRFAC_INHERIT) {
603 p2->p_traceflag = p1->p_traceflag;
604 if ((p2->p_tracevp = p1->p_tracevp) != NULL) {
606 KASSERT(p1->p_tracecred != NULL,
607 ("ktrace vnode with no cred"));
608 p2->p_tracecred = crhold(p1->p_tracecred);
611 mtx_unlock(&ktrace_mtx);
615 * If PF_FORK is set, the child process inherits the
616 * procfs ioctl flags from its parent.
618 if (p1->p_pfsflags & PF_FORK) {
619 p2->p_stops = p1->p_stops;
620 p2->p_pfsflags = p1->p_pfsflags;
624 * This begins the section where we must prevent the parent
625 * from being swapped.
631 * Attach the new process to its parent.
633 * If RFNOWAIT is set, the newly created process becomes a child
634 * of init. This effectively disassociates the child from the
637 if (flags & RFNOWAIT)
642 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
643 sx_xunlock(&proctree_lock);
645 /* Inform accounting that we have forked. */
646 p2->p_acflag = AFORK;
650 * Finish creating the child process. It will return via a different
651 * execution path later. (ie: directly into user mode)
653 vm_forkproc(td, p2, td2, flags);
655 if (flags == (RFFDG | RFPROC)) {
656 atomic_add_int(&cnt.v_forks, 1);
657 atomic_add_int(&cnt.v_forkpages, p2->p_vmspace->vm_dsize +
658 p2->p_vmspace->vm_ssize);
659 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
660 atomic_add_int(&cnt.v_vforks, 1);
661 atomic_add_int(&cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
662 p2->p_vmspace->vm_ssize);
663 } else if (p1 == &proc0) {
664 atomic_add_int(&cnt.v_kthreads, 1);
665 atomic_add_int(&cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
666 p2->p_vmspace->vm_ssize);
668 atomic_add_int(&cnt.v_rforks, 1);
669 atomic_add_int(&cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
670 p2->p_vmspace->vm_ssize);
674 * Both processes are set up, now check if any loadable modules want
675 * to adjust anything.
676 * What if they have an error? XXX
678 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
681 * Set the child start time and mark the process as being complete.
683 microuptime(&p2->p_stats->p_start);
684 mtx_lock_spin(&sched_lock);
685 p2->p_state = PRS_NORMAL;
688 * If RFSTOPPED not requested, make child runnable and add to
691 if ((flags & RFSTOPPED) == 0) {
693 setrunqueue(td2, SRQ_BORING);
695 mtx_unlock_spin(&sched_lock);
698 * Now can be swapped.
704 * Tell any interested parties about the new process.
706 KNOTE_LOCKED(&p1->p_klist, NOTE_FORK | p2->p_pid);
711 * Preserve synchronization semantics of vfork. If waiting for
712 * child to exec or exit, set P_PPWAIT on child, and sleep on our
713 * proc (in case of exit).
716 while (p2->p_flag & P_PPWAIT)
717 msleep(p1, &p2->p_mtx, PWAIT, "ppwait", 0);
721 * If other threads are waiting, let them continue now.
723 if (p1->p_flag & P_HADTHREADS) {
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",
739 sx_xunlock(&allproc_lock);
741 mac_destroy_proc(newproc);
744 audit_proc_free(newproc);
746 uma_zfree(proc_zone, newproc);
747 if (p1->p_flag & P_HADTHREADS) {
752 tsleep(&forksleep, PUSER, "fork", hz / 2);
757 * Handle the return of a child process from fork1(). This function
758 * is called from the MD fork_trampoline() entry point.
761 fork_exit(callout, arg, frame)
762 void (*callout)(void *, struct trapframe *);
764 struct trapframe *frame;
770 * Finish setting up thread glue so that it begins execution in a
771 * non-nested critical section with sched_lock held but not recursed.
775 td->td_oncpu = PCPU_GET(cpuid);
776 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
778 sched_lock.mtx_lock = (uintptr_t)td;
779 mtx_assert(&sched_lock, MA_OWNED | MA_NOTRECURSED);
780 CTR4(KTR_PROC, "fork_exit: new thread %p (kse %p, pid %d, %s)",
781 td, td->td_sched, p->p_pid, p->p_comm);
784 * Processes normally resume in mi_switch() after being
785 * cpu_switch()'ed to, but when children start up they arrive here
786 * instead, so we must do much the same things as mi_switch() would.
789 if ((td = PCPU_GET(deadthread))) {
790 PCPU_SET(deadthread, NULL);
794 mtx_unlock_spin(&sched_lock);
797 * cpu_set_fork_handler intercepts this function call to
798 * have this call a non-return function to stay in kernel mode.
799 * initproc has its own fork handler, but it does return.
801 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
805 * Check if a kernel thread misbehaved and returned from its main
808 if (p->p_flag & P_KTHREAD) {
809 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
810 p->p_comm, p->p_pid);
813 mtx_assert(&Giant, MA_NOTOWNED);
817 * Simplified back end of syscall(), used when returning from fork()
818 * directly into user mode. Giant is not held on entry, and must not
819 * be held on return. This function is passed in to fork_exit() as the
820 * first parameter and is called when returning to a new userland process.
823 fork_return(td, frame)
825 struct trapframe *frame;
830 if (KTRPOINT(td, KTR_SYSRET))
831 ktrsysret(SYS_fork, 0, 0);
833 mtx_assert(&Giant, MA_NOTOWNED);