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"
42 #include "opt_kstack_pages.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>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
52 #include <sys/sysctl.h>
54 #include <sys/malloc.h>
55 #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/sysent.h>
71 #include <sys/signalvar.h>
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_extern.h>
83 #include <sys/dtrace_bsd.h>
84 dtrace_fork_func_t dtrace_fasttrap_fork;
87 SDT_PROVIDER_DECLARE(proc);
88 SDT_PROBE_DEFINE(proc, kernel, , create, create);
89 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *");
90 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *");
91 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int");
93 #ifndef _SYS_SYSPROTO_H_
103 struct fork_args *uap;
108 error = fork1(td, RFFDG | RFPROC, 0, &p2);
110 td->td_retval[0] = p2->p_pid;
111 td->td_retval[1] = 0;
120 struct vfork_args *uap;
125 flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
126 error = fork1(td, flags, 0, &p2);
128 td->td_retval[0] = p2->p_pid;
129 td->td_retval[1] = 0;
137 struct rfork_args *uap;
142 /* Don't allow kernel-only flags. */
143 if ((uap->flags & RFKERNELONLY) != 0)
146 AUDIT_ARG_FFLAGS(uap->flags);
147 error = fork1(td, uap->flags, 0, &p2);
149 td->td_retval[0] = p2 ? p2->p_pid : 0;
150 td->td_retval[1] = 0;
155 int nprocs = 1; /* process 0 */
157 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
161 * Random component to lastpid generation. We mix in a random factor to make
162 * it a little harder to predict. We sanity check the modulus value to avoid
163 * doing it in critical paths. Don't let it be too small or we pointlessly
164 * waste randomness entropy, and don't let it be impossibly large. Using a
165 * modulus that is too big causes a LOT more process table scans and slows
166 * down fork processing as the pidchecked caching is defeated.
168 static int randompid = 0;
171 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
175 error = sysctl_wire_old_buffer(req, sizeof(int));
178 sx_xlock(&allproc_lock);
180 error = sysctl_handle_int(oidp, &pid, 0, req);
181 if (error == 0 && req->newptr != NULL) {
182 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
184 else if (pid < 2) /* NOP */
186 else if (pid < 100) /* Make it reasonable */
190 sx_xunlock(&allproc_lock);
194 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
195 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
198 fork1(td, flags, pages, procp)
204 struct proc *p1, *p2, *pptr;
205 struct proc *newproc;
206 int ok, p2_held, trypid;
207 static int curfail, pidchecked = 0;
208 static struct timeval lastfail;
210 struct filedesc_to_leader *fdtol;
212 struct sigacts *newsigacts;
214 vm_ooffset_t mem_charged;
217 /* Check for the undefined or unimplemented flags. */
218 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
221 /* Signal value requires RFTSIGZMB. */
222 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
225 /* Can't copy and clear. */
226 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
229 /* Check the validity of the signal number. */
230 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
237 * Here we don't create a new process, but we divorce
238 * certain parts of a process from itself.
240 if ((flags & RFPROC) == 0) {
241 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
242 (flags & (RFCFDG | RFFDG))) {
244 if (thread_single(SINGLE_BOUNDARY)) {
251 error = vm_forkproc(td, NULL, NULL, NULL, flags);
256 * Close all file descriptors.
258 if (flags & RFCFDG) {
259 struct filedesc *fdtmp;
260 fdtmp = fdinit(td->td_proc->p_fd);
266 * Unshare file descriptors (from parent).
272 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
273 (flags & (RFCFDG | RFFDG))) {
284 * We did have single-threading code here
285 * however it proved un-needed and caused problems
291 pages = KSTACK_PAGES;
292 /* Allocate new proc. */
293 newproc = uma_zalloc(proc_zone, M_WAITOK);
294 td2 = FIRST_THREAD_IN_PROC(newproc);
296 td2 = thread_alloc(pages);
301 proc_linkup(newproc, td2);
303 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
304 if (td2->td_kstack != 0)
305 vm_thread_dispose(td2);
306 if (!thread_alloc_stack(td2, pages)) {
313 if ((flags & RFMEM) == 0) {
314 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
319 if (!swap_reserve(mem_charged)) {
321 * The swap reservation failed. The accounting
322 * from the entries of the copied vm2 will be
323 * substracted in vmspace_free(), so force the
326 swap_reserve_force(mem_charged);
333 mac_proc_init(newproc);
335 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
336 STAILQ_INIT(&newproc->p_ktr);
338 /* We have to lock the process tree while we look for a pid. */
339 sx_slock(&proctree_lock);
342 * Although process entries are dynamically created, we still keep
343 * a global limit on the maximum number we will create. Don't allow
344 * a nonprivileged user to use the last ten processes; don't let root
345 * exceed the limit. The variable nprocs is the current number of
346 * processes, maxproc is the limit.
348 sx_xlock(&allproc_lock);
349 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
350 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
356 * Increment the count of procs running with this uid. Don't allow
357 * a nonprivileged user to exceed their current limit.
359 * XXXRW: Can we avoid privilege here if it's not needed?
361 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
363 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
366 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
367 lim_cur(p1, RLIMIT_NPROC));
376 * Increment the nprocs resource before blocking can occur. There
377 * are hard-limits as to the number of processes that can run.
382 * Find an unused process ID. We remember a range of unused IDs
383 * ready to use (from lastpid+1 through pidchecked-1).
385 * If RFHIGHPID is set (used during system boot), do not allocate
388 trypid = lastpid + 1;
389 if (flags & RFHIGHPID) {
394 trypid += arc4random() % randompid;
398 * If the process ID prototype has wrapped around,
399 * restart somewhat above 0, as the low-numbered procs
400 * tend to include daemons that don't exit.
402 if (trypid >= PID_MAX) {
403 trypid = trypid % PID_MAX;
408 if (trypid >= pidchecked) {
411 pidchecked = PID_MAX;
413 * Scan the active and zombie procs to check whether this pid
414 * is in use. Remember the lowest pid that's greater
415 * than trypid, so we can avoid checking for a while.
417 p2 = LIST_FIRST(&allproc);
419 for (; p2 != NULL; p2 = LIST_NEXT(p2, p_list)) {
420 while (p2->p_pid == trypid ||
421 (p2->p_pgrp != NULL &&
422 (p2->p_pgrp->pg_id == trypid ||
423 (p2->p_session != NULL &&
424 p2->p_session->s_sid == trypid)))) {
426 if (trypid >= pidchecked)
429 if (p2->p_pid > trypid && pidchecked > p2->p_pid)
430 pidchecked = p2->p_pid;
431 if (p2->p_pgrp != NULL) {
432 if (p2->p_pgrp->pg_id > trypid &&
433 pidchecked > p2->p_pgrp->pg_id)
434 pidchecked = p2->p_pgrp->pg_id;
435 if (p2->p_session != NULL &&
436 p2->p_session->s_sid > trypid &&
437 pidchecked > p2->p_session->s_sid)
438 pidchecked = p2->p_session->s_sid;
443 p2 = LIST_FIRST(&zombproc);
447 sx_sunlock(&proctree_lock);
450 * RFHIGHPID does not mess with the lastpid counter during boot.
452 if (flags & RFHIGHPID)
458 p2->p_state = PRS_NEW; /* protect against others */
460 AUDIT_ARG_PID(p2->p_pid);
461 LIST_INSERT_HEAD(&allproc, p2, p_list);
462 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
467 sx_xunlock(&allproc_lock);
469 bcopy(&p1->p_startcopy, &p2->p_startcopy,
470 __rangeof(struct proc, p_startcopy, p_endcopy));
471 pargs_hold(p2->p_args);
474 bzero(&p2->p_startzero,
475 __rangeof(struct proc, p_startzero, p_endzero));
477 p2->p_ucred = crhold(td->td_ucred);
479 /* Tell the prison that we exist. */
480 prison_proc_hold(p2->p_ucred->cr_prison);
485 * Malloc things while we don't hold any locks.
487 if (flags & RFSIGSHARE)
490 newsigacts = sigacts_alloc();
495 if (flags & RFCFDG) {
496 fd = fdinit(p1->p_fd);
498 } else if (flags & RFFDG) {
499 fd = fdcopy(p1->p_fd);
502 fd = fdshare(p1->p_fd);
503 if (p1->p_fdtol == NULL)
505 filedesc_to_leader_alloc(NULL,
508 if ((flags & RFTHREAD) != 0) {
510 * Shared file descriptor table and
511 * shared process leaders.
514 FILEDESC_XLOCK(p1->p_fd);
515 fdtol->fdl_refcount++;
516 FILEDESC_XUNLOCK(p1->p_fd);
519 * Shared file descriptor table, and
520 * different process leaders
522 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
528 * Make a proc table entry for the new process.
529 * Start by zeroing the section of proc that is zero-initialized,
530 * then copy the section that is copied directly from the parent.
536 bzero(&td2->td_startzero,
537 __rangeof(struct thread, td_startzero, td_endzero));
538 bzero(&td2->td_rux, sizeof(td2->td_rux));
539 td2->td_map_def_user = NULL;
540 td2->td_dbg_forked = 0;
542 bcopy(&td->td_startcopy, &td2->td_startcopy,
543 __rangeof(struct thread, td_startcopy, td_endcopy));
545 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
546 td2->td_sigstk = td->td_sigstk;
547 td2->td_sigmask = td->td_sigmask;
548 td2->td_flags = TDF_INMEM;
552 td2->td_vnet_lpush = NULL;
556 * Allow the scheduler to initialize the child.
563 * Duplicate sub-structures as needed.
564 * Increase reference counts on shared objects.
566 p2->p_flag = P_INMEM;
567 p2->p_swtick = ticks;
568 if (p1->p_flag & P_PROFIL)
570 td2->td_ucred = crhold(p2->p_ucred);
572 if (flags & RFSIGSHARE) {
573 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
575 sigacts_copy(newsigacts, p1->p_sigacts);
576 p2->p_sigacts = newsigacts;
579 if (flags & RFTSIGZMB)
580 p2->p_sigparent = RFTSIGNUM(flags);
581 else if (flags & RFLINUXTHPN)
582 p2->p_sigparent = SIGUSR1;
584 p2->p_sigparent = SIGCHLD;
586 p2->p_textvp = p1->p_textvp;
591 * p_limit is copy-on-write. Bump its refcount.
595 pstats_fork(p1->p_stats, p2->p_stats);
600 /* Bump references to the text vnode (for procfs) */
605 * Set up linkage for kernel based threading.
607 if ((flags & RFTHREAD) != 0) {
608 mtx_lock(&ppeers_lock);
609 p2->p_peers = p1->p_peers;
611 p2->p_leader = p1->p_leader;
612 mtx_unlock(&ppeers_lock);
613 PROC_LOCK(p1->p_leader);
614 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
615 PROC_UNLOCK(p1->p_leader);
617 * The task leader is exiting, so process p1 is
618 * going to be killed shortly. Since p1 obviously
619 * isn't dead yet, we know that the leader is either
620 * sending SIGKILL's to all the processes in this
621 * task or is sleeping waiting for all the peers to
622 * exit. We let p1 complete the fork, but we need
623 * to go ahead and kill the new process p2 since
624 * the task leader may not get a chance to send
625 * SIGKILL to it. We leave it on the list so that
626 * the task leader will wait for this new process
630 psignal(p2, SIGKILL);
633 PROC_UNLOCK(p1->p_leader);
639 sx_xlock(&proctree_lock);
640 PGRP_LOCK(p1->p_pgrp);
645 * Preserve some more flags in subprocess. P_PROFIL has already
648 p2->p_flag |= p1->p_flag & P_SUGID;
649 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
650 SESS_LOCK(p1->p_session);
651 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
652 p2->p_flag |= P_CONTROLT;
653 SESS_UNLOCK(p1->p_session);
654 if (flags & RFPPWAIT)
655 p2->p_flag |= P_PPWAIT;
657 p2->p_pgrp = p1->p_pgrp;
658 LIST_INSERT_AFTER(p1, p2, p_pglist);
659 PGRP_UNLOCK(p1->p_pgrp);
660 LIST_INIT(&p2->p_children);
662 callout_init(&p2->p_itcallout, CALLOUT_MPSAFE);
665 * If PF_FORK is set, the child process inherits the
666 * procfs ioctl flags from its parent.
668 if (p1->p_pfsflags & PF_FORK) {
669 p2->p_stops = p1->p_stops;
670 p2->p_pfsflags = p1->p_pfsflags;
674 * This begins the section where we must prevent the parent
675 * from being swapped.
681 * Attach the new process to its parent.
683 * If RFNOWAIT is set, the newly created process becomes a child
684 * of init. This effectively disassociates the child from the
687 if (flags & RFNOWAIT)
692 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
693 sx_xunlock(&proctree_lock);
695 /* Inform accounting that we have forked. */
696 p2->p_acflag = AFORK;
704 * Finish creating the child process. It will return via a different
705 * execution path later. (ie: directly into user mode)
707 vm_forkproc(td, p2, td2, vm2, flags);
709 if (flags == (RFFDG | RFPROC)) {
710 PCPU_INC(cnt.v_forks);
711 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
712 p2->p_vmspace->vm_ssize);
713 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
714 PCPU_INC(cnt.v_vforks);
715 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
716 p2->p_vmspace->vm_ssize);
717 } else if (p1 == &proc0) {
718 PCPU_INC(cnt.v_kthreads);
719 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
720 p2->p_vmspace->vm_ssize);
722 PCPU_INC(cnt.v_rforks);
723 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
724 p2->p_vmspace->vm_ssize);
728 * Both processes are set up, now check if any loadable modules want
729 * to adjust anything.
730 * What if they have an error? XXX
732 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
735 * Set the child start time and mark the process as being complete.
739 microuptime(&p2->p_stats->p_start);
741 p2->p_state = PRS_NORMAL;
746 * Tell the DTrace fasttrap provider about the new process
747 * if it has registered an interest. We have to do this only after
748 * p_state is PRS_NORMAL since the fasttrap module will use pfind()
751 if (dtrace_fasttrap_fork)
752 dtrace_fasttrap_fork(p1, p2);
754 if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED |
757 * Arrange for debugger to receive the fork event.
759 * We can report PL_FLAG_FORKED regardless of
760 * P_FOLLOWFORK settings, but it does not make a sense
763 td->td_dbgflags |= TDB_FORK;
764 td->td_dbg_forked = p2->p_pid;
765 td2->td_dbgflags |= TDB_STOPATFORK;
770 if ((flags & RFSTOPPED) == 0) {
772 * If RFSTOPPED not requested, make child runnable and
777 sched_add(td2, SRQ_BORING);
782 * Now can be swapped.
788 * Tell any interested parties about the new process.
790 knote_fork(&p1->p_klist, p2->p_pid);
791 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
794 * Wait until debugger is attached to child.
797 while ((td2->td_dbgflags & TDB_STOPATFORK) != 0)
798 cv_wait(&p2->p_dbgwait, &p2->p_mtx);
803 * Preserve synchronization semantics of vfork. If waiting for
804 * child to exec or exit, set P_PPWAIT on child, and sleep on our
805 * proc (in case of exit).
807 while (p2->p_flag & P_PPWAIT)
808 cv_wait(&p2->p_pwait, &p2->p_mtx);
812 * Return child proc pointer to parent.
817 sx_sunlock(&proctree_lock);
818 if (ppsratecheck(&lastfail, &curfail, 1))
819 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
820 td->td_ucred->cr_ruid);
821 sx_xunlock(&allproc_lock);
823 mac_proc_destroy(newproc);
828 uma_zfree(proc_zone, newproc);
829 pause("fork", hz / 2);
834 * Handle the return of a child process from fork1(). This function
835 * is called from the MD fork_trampoline() entry point.
838 fork_exit(callout, arg, frame)
839 void (*callout)(void *, struct trapframe *);
841 struct trapframe *frame;
849 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
851 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
852 td, td->td_sched, p->p_pid, td->td_name);
856 * Processes normally resume in mi_switch() after being
857 * cpu_switch()'ed to, but when children start up they arrive here
858 * instead, so we must do much the same things as mi_switch() would.
860 if ((dtd = PCPU_GET(deadthread))) {
861 PCPU_SET(deadthread, NULL);
867 * cpu_set_fork_handler intercepts this function call to
868 * have this call a non-return function to stay in kernel mode.
869 * initproc has its own fork handler, but it does return.
871 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
875 * Check if a kernel thread misbehaved and returned from its main
878 if (p->p_flag & P_KTHREAD) {
879 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
880 td->td_name, p->p_pid);
883 mtx_assert(&Giant, MA_NOTOWNED);
885 if (p->p_sysent->sv_schedtail != NULL)
886 (p->p_sysent->sv_schedtail)(td);
890 * Simplified back end of syscall(), used when returning from fork()
891 * directly into user mode. Giant is not held on entry, and must not
892 * be held on return. This function is passed in to fork_exit() as the
893 * first parameter and is called when returning to a new userland process.
896 fork_return(td, frame)
898 struct trapframe *frame;
900 struct proc *p, *dbg;
902 if (td->td_dbgflags & TDB_STOPATFORK) {
904 sx_xlock(&proctree_lock);
906 if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) ==
907 (P_TRACED | P_FOLLOWFORK)) {
909 * If debugger still wants auto-attach for the
910 * parent's children, do it now.
912 dbg = p->p_pptr->p_pptr;
913 p->p_flag |= P_TRACED;
914 p->p_oppid = p->p_pptr->p_pid;
915 proc_reparent(p, dbg);
916 sx_xunlock(&proctree_lock);
917 ptracestop(td, SIGSTOP);
920 * ... otherwise clear the request.
922 sx_xunlock(&proctree_lock);
923 td->td_dbgflags &= ~TDB_STOPATFORK;
924 cv_broadcast(&p->p_dbgwait);
932 if (KTRPOINT(td, KTR_SYSRET))
933 ktrsysret(SYS_fork, 0, 0);
935 mtx_assert(&Giant, MA_NOTOWNED);