2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_ktrace.h"
43 #include "opt_kstack_pages.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/bitstring.h>
48 #include <sys/sysproto.h>
49 #include <sys/eventhandler.h>
50 #include <sys/fcntl.h>
51 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/sysctl.h>
57 #include <sys/malloc.h>
58 #include <sys/mutex.h>
61 #include <sys/procdesc.h>
62 #include <sys/ptrace.h>
63 #include <sys/racct.h>
64 #include <sys/resourcevar.h>
65 #include <sys/sched.h>
66 #include <sys/syscall.h>
67 #include <sys/vmmeter.h>
68 #include <sys/vnode.h>
71 #include <sys/ktrace.h>
72 #include <sys/unistd.h>
75 #include <sys/sysent.h>
76 #include <sys/signalvar.h>
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_extern.h>
88 #include <sys/dtrace_bsd.h>
89 dtrace_fork_func_t dtrace_fasttrap_fork;
92 SDT_PROVIDER_DECLARE(proc);
93 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
95 #ifndef _SYS_SYSPROTO_H_
103 sys_fork(struct thread *td, struct fork_args *uap)
108 bzero(&fr, sizeof(fr));
109 fr.fr_flags = RFFDG | RFPROC;
111 error = fork1(td, &fr);
113 td->td_retval[0] = pid;
114 td->td_retval[1] = 0;
121 sys_pdfork(struct thread *td, struct pdfork_args *uap)
126 bzero(&fr, sizeof(fr));
127 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
130 fr.fr_pd_flags = uap->flags;
131 AUDIT_ARG_FFLAGS(uap->flags);
133 * It is necessary to return fd by reference because 0 is a valid file
134 * descriptor number, and the child needs to be able to distinguish
135 * itself from the parent using the return value.
137 error = fork1(td, &fr);
139 td->td_retval[0] = pid;
140 td->td_retval[1] = 0;
141 error = copyout(&fd, uap->fdp, sizeof(fd));
148 sys_vfork(struct thread *td, struct vfork_args *uap)
153 bzero(&fr, sizeof(fr));
154 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
156 error = fork1(td, &fr);
158 td->td_retval[0] = pid;
159 td->td_retval[1] = 0;
165 sys_rfork(struct thread *td, struct rfork_args *uap)
170 /* Don't allow kernel-only flags. */
171 if ((uap->flags & RFKERNELONLY) != 0)
173 /* RFSPAWN must not appear with others */
174 if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN)
177 AUDIT_ARG_FFLAGS(uap->flags);
178 bzero(&fr, sizeof(fr));
179 if ((uap->flags & RFSPAWN) != 0) {
180 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
181 fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
183 fr.fr_flags = uap->flags;
186 error = fork1(td, &fr);
188 td->td_retval[0] = pid;
189 td->td_retval[1] = 0;
194 int __exclusive_cache_line nprocs = 1; /* process 0 */
196 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
200 * Random component to lastpid generation. We mix in a random factor to make
201 * it a little harder to predict. We sanity check the modulus value to avoid
202 * doing it in critical paths. Don't let it be too small or we pointlessly
203 * waste randomness entropy, and don't let it be impossibly large. Using a
204 * modulus that is too big causes a LOT more process table scans and slows
205 * down fork processing as the pidchecked caching is defeated.
207 static int randompid = 0;
210 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
214 error = sysctl_wire_old_buffer(req, sizeof(int));
217 sx_xlock(&allproc_lock);
219 error = sysctl_handle_int(oidp, &pid, 0, req);
220 if (error == 0 && req->newptr != NULL) {
224 /* generate a random PID modulus between 100 and 1123 */
225 randompid = 100 + arc4random() % 1024;
226 else if (pid < 0 || pid > pid_max - 100)
228 randompid = pid_max - 100;
230 /* Make it reasonable */
235 sx_xunlock(&allproc_lock);
239 SYSCTL_PROC(_kern, OID_AUTO, randompid,
240 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
241 sysctl_kern_randompid, "I",
242 "Random PID modulus. Special values: 0: disable, 1: choose random value");
244 extern bitstr_t proc_id_pidmap;
245 extern bitstr_t proc_id_grpidmap;
246 extern bitstr_t proc_id_sessidmap;
247 extern bitstr_t proc_id_reapmap;
250 * Find an unused process ID
252 * If RFHIGHPID is set (used during system boot), do not allocate
256 fork_findpid(int flags)
262 * Avoid calling arc4random with procid_lock held.
265 if (__predict_false(randompid))
266 random = arc4random() % randompid;
268 mtx_lock(&procid_lock);
270 trypid = lastpid + 1;
271 if (flags & RFHIGHPID) {
278 if (trypid >= pid_max)
281 bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
283 KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
287 if (bit_test(&proc_id_grpidmap, result) ||
288 bit_test(&proc_id_sessidmap, result) ||
289 bit_test(&proc_id_reapmap, result)) {
295 * RFHIGHPID does not mess with the lastpid counter during boot.
297 if ((flags & RFHIGHPID) == 0)
300 bit_set(&proc_id_pidmap, result);
301 mtx_unlock(&procid_lock);
307 fork_norfproc(struct thread *td, int flags)
312 KASSERT((flags & RFPROC) == 0,
313 ("fork_norfproc called with RFPROC set"));
317 * Quiesce other threads if necessary. If RFMEM is not specified we
318 * must ensure that other threads do not concurrently create a second
319 * process sharing the vmspace, see vmspace_unshare().
321 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
322 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
324 if (thread_single(p1, SINGLE_BOUNDARY)) {
331 error = vm_forkproc(td, NULL, NULL, NULL, flags);
336 * Close all file descriptors.
338 if ((flags & RFCFDG) != 0) {
339 struct filedesc *fdtmp;
340 struct pwddesc *pdtmp;
342 pdtmp = pdinit(td->td_proc->p_pd, false);
343 fdtmp = fdinit(td->td_proc->p_fd, false, NULL);
351 * Unshare file descriptors (from parent).
353 if ((flags & RFFDG) != 0) {
359 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
360 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
362 thread_single_end(p1, SINGLE_BOUNDARY);
369 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
370 struct vmspace *vm2, struct file *fp_procdesc)
372 struct proc *p1, *pptr;
374 struct filedesc_to_leader *fdtol;
376 struct sigacts *newsigacts;
381 bcopy(&p1->p_startcopy, &p2->p_startcopy,
382 __rangeof(struct proc, p_startcopy, p_endcopy));
383 pargs_hold(p2->p_args);
384 p2->p_umtx_min_timeout = p1->p_umtx_min_timeout;
387 bzero(&p2->p_startzero,
388 __rangeof(struct proc, p_startzero, p_endzero));
390 /* Tell the prison that we exist. */
391 prison_proc_hold(p2->p_ucred->cr_prison);
393 p2->p_state = PRS_NEW; /* protect against others */
394 p2->p_pid = fork_findpid(fr->fr_flags);
395 AUDIT_ARG_PID(p2->p_pid);
396 TSFORK(p2->p_pid, p1->p_pid);
398 sx_xlock(&allproc_lock);
399 LIST_INSERT_HEAD(&allproc, p2, p_list);
401 sx_xunlock(&allproc_lock);
403 sx_xlock(PIDHASHLOCK(p2->p_pid));
404 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
405 sx_xunlock(PIDHASHLOCK(p2->p_pid));
410 * Malloc things while we don't hold any locks.
412 if (fr->fr_flags & RFSIGSHARE)
415 newsigacts = sigacts_alloc();
420 if (fr->fr_flags & RFCFDG) {
421 pd = pdinit(p1->p_pd, false);
422 fd = fdinit(p1->p_fd, false, NULL);
424 } else if (fr->fr_flags & RFFDG) {
425 if (fr->fr_flags2 & FR2_SHARE_PATHS)
426 pd = pdshare(p1->p_pd);
428 pd = pdcopy(p1->p_pd);
429 fd = fdcopy(p1->p_fd);
432 if (fr->fr_flags2 & FR2_SHARE_PATHS)
433 pd = pdcopy(p1->p_pd);
435 pd = pdshare(p1->p_pd);
436 fd = fdshare(p1->p_fd);
437 if (p1->p_fdtol == NULL)
438 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
440 if ((fr->fr_flags & RFTHREAD) != 0) {
442 * Shared file descriptor table, and shared
445 fdtol = filedesc_to_leader_share(p1->p_fdtol, p1->p_fd);
448 * Shared file descriptor table, and different
451 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
456 * Make a proc table entry for the new process.
457 * Start by zeroing the section of proc that is zero-initialized,
458 * then copy the section that is copied directly from the parent.
464 bzero(&td2->td_startzero,
465 __rangeof(struct thread, td_startzero, td_endzero));
467 bcopy(&td->td_startcopy, &td2->td_startcopy,
468 __rangeof(struct thread, td_startcopy, td_endcopy));
470 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
471 td2->td_sigstk = td->td_sigstk;
472 td2->td_flags = TDF_INMEM;
473 td2->td_lend_user_pri = PRI_MAX;
477 td2->td_vnet_lpush = NULL;
481 * Allow the scheduler to initialize the child.
488 * Duplicate sub-structures as needed.
489 * Increase reference counts on shared objects.
491 p2->p_flag = P_INMEM;
492 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
493 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
494 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
495 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS |
496 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
497 p2->p_swtick = ticks;
498 if (p1->p_flag & P_PROFIL)
501 if (fr->fr_flags & RFSIGSHARE) {
502 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
504 sigacts_copy(newsigacts, p1->p_sigacts);
505 p2->p_sigacts = newsigacts;
506 if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) {
507 mtx_lock(&p2->p_sigacts->ps_mtx);
508 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0)
510 if ((fr->fr_flags2 & FR2_KPROC) != 0)
511 p2->p_sigacts->ps_flag |= PS_NOCLDWAIT;
512 mtx_unlock(&p2->p_sigacts->ps_mtx);
516 if (fr->fr_flags & RFTSIGZMB)
517 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
518 else if (fr->fr_flags & RFLINUXTHPN)
519 p2->p_sigparent = SIGUSR1;
521 p2->p_sigparent = SIGCHLD;
523 if ((fr->fr_flags2 & FR2_KPROC) != 0) {
524 p2->p_flag |= P_SYSTEM | P_KPROC;
525 td2->td_pflags |= TDP_KTHREAD;
528 p2->p_textvp = p1->p_textvp;
529 p2->p_textdvp = p1->p_textdvp;
533 p2->p_elf_brandinfo = p1->p_elf_brandinfo;
535 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
536 p2->p_flag |= P_PROTECTED;
537 p2->p_flag2 |= P2_INHERIT_PROTECTED;
541 * p_limit is copy-on-write. Bump its refcount.
545 thread_cow_get_proc(td2, p2);
547 pstats_fork(p1->p_stats, p2->p_stats);
553 * Bump references to the text vnode and directory, and copy
556 if (p2->p_textvp != NULL)
557 vrefact(p2->p_textvp);
558 if (p2->p_textdvp != NULL)
559 vrefact(p2->p_textdvp);
560 p2->p_binname = p1->p_binname == NULL ? NULL :
561 strdup(p1->p_binname, M_PARGS);
564 * Set up linkage for kernel based threading.
566 if ((fr->fr_flags & RFTHREAD) != 0) {
567 mtx_lock(&ppeers_lock);
568 p2->p_peers = p1->p_peers;
570 p2->p_leader = p1->p_leader;
571 mtx_unlock(&ppeers_lock);
572 PROC_LOCK(p1->p_leader);
573 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
574 PROC_UNLOCK(p1->p_leader);
576 * The task leader is exiting, so process p1 is
577 * going to be killed shortly. Since p1 obviously
578 * isn't dead yet, we know that the leader is either
579 * sending SIGKILL's to all the processes in this
580 * task or is sleeping waiting for all the peers to
581 * exit. We let p1 complete the fork, but we need
582 * to go ahead and kill the new process p2 since
583 * the task leader may not get a chance to send
584 * SIGKILL to it. We leave it on the list so that
585 * the task leader will wait for this new process
589 kern_psignal(p2, SIGKILL);
592 PROC_UNLOCK(p1->p_leader);
598 sx_xlock(&proctree_lock);
599 PGRP_LOCK(p1->p_pgrp);
604 * Preserve some more flags in subprocess. P_PROFIL has already
607 p2->p_flag |= p1->p_flag & P_SUGID;
608 td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK |
609 TDP_SIGFASTBLOCK)) | TDP_FORKING;
610 SESS_LOCK(p1->p_session);
611 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
612 p2->p_flag |= P_CONTROLT;
613 SESS_UNLOCK(p1->p_session);
614 if (fr->fr_flags & RFPPWAIT)
615 p2->p_flag |= P_PPWAIT;
617 p2->p_pgrp = p1->p_pgrp;
618 LIST_INSERT_AFTER(p1, p2, p_pglist);
619 PGRP_UNLOCK(p1->p_pgrp);
620 LIST_INIT(&p2->p_children);
621 LIST_INIT(&p2->p_orphans);
623 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
624 TAILQ_INIT(&p2->p_kqtim_stop);
627 * This begins the section where we must prevent the parent
628 * from being swapped.
634 * Attach the new process to its parent.
636 * If RFNOWAIT is set, the newly created process becomes a child
637 * of init. This effectively disassociates the child from the
640 if ((fr->fr_flags & RFNOWAIT) != 0) {
644 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
649 p2->p_oppid = pptr->p_pid;
650 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
651 LIST_INIT(&p2->p_reaplist);
652 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
653 if (p2->p_reaper == p1 && p1 != initproc) {
654 p2->p_reapsubtree = p2->p_pid;
655 proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
657 sx_xunlock(&proctree_lock);
659 /* Inform accounting that we have forked. */
660 p2->p_acflag = AFORK;
668 * Finish creating the child process. It will return via a different
669 * execution path later. (ie: directly into user mode)
671 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
673 if (fr->fr_flags == (RFFDG | RFPROC)) {
675 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
676 p2->p_vmspace->vm_ssize);
677 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
678 VM_CNT_INC(v_vforks);
679 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
680 p2->p_vmspace->vm_ssize);
681 } else if (p1 == &proc0) {
682 VM_CNT_INC(v_kthreads);
683 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
684 p2->p_vmspace->vm_ssize);
686 VM_CNT_INC(v_rforks);
687 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
688 p2->p_vmspace->vm_ssize);
692 * Associate the process descriptor with the process before anything
693 * can happen that might cause that process to need the descriptor.
694 * However, don't do this until after fork(2) can no longer fail.
696 if (fr->fr_flags & RFPROCDESC)
697 procdesc_new(p2, fr->fr_pd_flags);
700 * Both processes are set up, now check if any loadable modules want
701 * to adjust anything.
703 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
706 * Set the child start time and mark the process as being complete.
710 microuptime(&p2->p_stats->p_start);
712 p2->p_state = PRS_NORMAL;
717 * Tell the DTrace fasttrap provider about the new process so that any
718 * tracepoints inherited from the parent can be removed. We have to do
719 * this only after p_state is PRS_NORMAL since the fasttrap module will
720 * use pfind() later on.
722 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
723 dtrace_fasttrap_fork(p1, p2);
725 if (fr->fr_flags & RFPPWAIT) {
726 td->td_pflags |= TDP_RFPPWAIT;
727 td->td_rfppwait_p = p2;
728 td->td_dbgflags |= TDB_VFORK;
733 * Tell any interested parties about the new process.
735 knote_fork(p1->p_klist, p2->p_pid);
738 * Now can be swapped.
742 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
744 if (fr->fr_flags & RFPROCDESC) {
745 procdesc_finit(p2->p_procdesc, fp_procdesc);
746 fdrop(fp_procdesc, td);
750 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
751 * synced with forks in progress so it is OK if we miss it
754 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
755 sx_xlock(&proctree_lock);
759 * p1->p_ptevents & p1->p_pptr are protected by both
760 * process and proctree locks for modifications,
761 * so owning proctree_lock allows the race-free read.
763 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
765 * Arrange for debugger to receive the fork event.
767 * We can report PL_FLAG_FORKED regardless of
768 * P_FOLLOWFORK settings, but it does not make a sense
771 td->td_dbgflags |= TDB_FORK;
772 td->td_dbg_forked = p2->p_pid;
773 td2->td_dbgflags |= TDB_STOPATFORK;
774 proc_set_traced(p2, true);
776 "do_fork: attaching to new child pid %d: oppid %d",
777 p2->p_pid, p2->p_oppid);
778 proc_reparent(p2, p1->p_pptr, false);
781 sx_xunlock(&proctree_lock);
784 racct_proc_fork_done(p2);
786 if ((fr->fr_flags & RFSTOPPED) == 0) {
787 if (fr->fr_pidp != NULL)
788 *fr->fr_pidp = p2->p_pid;
790 * If RFSTOPPED not requested, make child runnable and
795 sched_add(td2, SRQ_BORING);
802 fork_rfppwait(struct thread *td)
806 MPASS(td->td_pflags & TDP_RFPPWAIT);
810 * Preserve synchronization semantics of vfork. If
811 * waiting for child to exec or exit, fork set
812 * P_PPWAIT on child, and there we sleep on our proc
815 * Do it after the ptracestop() above is finished, to
816 * not block our debugger until child execs or exits
817 * to finish vfork wait.
819 td->td_pflags &= ~TDP_RFPPWAIT;
820 p2 = td->td_rfppwait_p;
823 while (p2->p_flag & P_PPWAIT) {
825 if (thread_suspend_check_needed()) {
827 thread_suspend_check(0);
833 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
837 if (td->td_dbgflags & TDB_VFORK) {
839 if (p->p_ptevents & PTRACE_VFORK)
840 ptracestop(td, SIGTRAP, NULL);
841 td->td_dbgflags &= ~TDB_VFORK;
847 fork1(struct thread *td, struct fork_req *fr)
849 struct proc *p1, *newproc;
853 struct file *fp_procdesc;
854 vm_ooffset_t mem_charged;
855 int error, nprocs_new;
857 static struct timeval lastfail;
860 flags = fr->fr_flags;
861 pages = fr->fr_pages;
863 if ((flags & RFSTOPPED) != 0)
864 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
866 MPASS(fr->fr_procp == NULL);
868 /* Check for the undefined or unimplemented flags. */
869 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
872 /* Signal value requires RFTSIGZMB. */
873 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
876 /* Can't copy and clear. */
877 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
880 /* Check the validity of the signal number. */
881 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
884 if ((flags & RFPROCDESC) != 0) {
885 /* Can't not create a process yet get a process descriptor. */
886 if ((flags & RFPROC) == 0)
889 /* Must provide a place to put a procdesc if creating one. */
890 if (fr->fr_pd_fd == NULL)
893 /* Check if we are using supported flags. */
894 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
901 * Here we don't create a new process, but we divorce
902 * certain parts of a process from itself.
904 if ((flags & RFPROC) == 0) {
905 if (fr->fr_procp != NULL)
906 *fr->fr_procp = NULL;
907 else if (fr->fr_pidp != NULL)
909 return (fork_norfproc(td, flags));
917 * Increment the nprocs resource before allocations occur.
918 * Although process entries are dynamically created, we still
919 * keep a global limit on the maximum number we will
920 * create. There are hard-limits as to the number of processes
921 * that can run, established by the KVA and memory usage for
924 * Don't allow a nonprivileged user to use the last ten
925 * processes; don't let root exceed the limit.
927 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
928 if (nprocs_new >= maxproc - 10) {
929 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
930 nprocs_new >= maxproc) {
932 sx_xlock(&allproc_lock);
933 if (ppsratecheck(&lastfail, &curfail, 1)) {
934 printf("maxproc limit exceeded by uid %u "
935 "(pid %d); see tuning(7) and "
937 td->td_ucred->cr_ruid, p1->p_pid);
939 sx_xunlock(&allproc_lock);
945 * If required, create a process descriptor in the parent first; we
946 * will abandon it if something goes wrong. We don't finit() until
949 if (flags & RFPROCDESC) {
950 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
951 fr->fr_pd_flags, fr->fr_pd_fcaps);
954 AUDIT_ARG_FD(*fr->fr_pd_fd);
959 pages = kstack_pages;
960 /* Allocate new proc. */
961 newproc = uma_zalloc(proc_zone, M_WAITOK);
962 td2 = FIRST_THREAD_IN_PROC(newproc);
964 td2 = thread_alloc(pages);
969 proc_linkup(newproc, td2);
971 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
972 if (td2->td_kstack != 0)
973 vm_thread_dispose(td2);
974 if (!thread_alloc_stack(td2, pages)) {
981 if ((flags & RFMEM) == 0) {
982 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
987 if (!swap_reserve(mem_charged)) {
989 * The swap reservation failed. The accounting
990 * from the entries of the copied vm2 will be
991 * subtracted in vmspace_free(), so force the
994 swap_reserve_force(mem_charged);
1002 * XXX: This is ugly; when we copy resource usage, we need to bump
1003 * per-cred resource counters.
1005 proc_set_cred_init(newproc, td->td_ucred);
1008 * Initialize resource accounting for the child process.
1010 error = racct_proc_fork(p1, newproc);
1017 mac_proc_init(newproc);
1019 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
1020 STAILQ_INIT(&newproc->p_ktr);
1023 * Increment the count of procs running with this uid. Don't allow
1024 * a nonprivileged user to exceed their current limit.
1026 cred = td->td_ucred;
1027 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1028 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1030 chgproccnt(cred->cr_ruidinfo, 1, 0);
1033 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1038 mac_proc_destroy(newproc);
1040 racct_proc_exit(newproc);
1042 proc_unset_cred(newproc);
1046 uma_zfree(proc_zone, newproc);
1047 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1048 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1049 fdrop(fp_procdesc, td);
1051 atomic_add_int(&nprocs, -1);
1052 pause("fork", hz / 2);
1057 * Handle the return of a child process from fork1(). This function
1058 * is called from the MD fork_trampoline() entry point.
1061 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1062 struct trapframe *frame)
1070 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1072 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1073 td, td_get_sched(td), p->p_pid, td->td_name);
1075 sched_fork_exit(td);
1078 * Processes normally resume in mi_switch() after being
1079 * cpu_switch()'ed to, but when children start up they arrive here
1080 * instead, so we must do much the same things as mi_switch() would.
1082 if ((dtd = PCPU_GET(deadthread))) {
1083 PCPU_SET(deadthread, NULL);
1089 * cpu_fork_kthread_handler intercepts this function call to
1090 * have this call a non-return function to stay in kernel mode.
1091 * initproc has its own fork handler, but it does return.
1093 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1094 callout(arg, frame);
1097 * Check if a kernel thread misbehaved and returned from its main
1100 if (p->p_flag & P_KPROC) {
1101 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1102 td->td_name, p->p_pid);
1105 mtx_assert(&Giant, MA_NOTOWNED);
1107 if (p->p_sysent->sv_schedtail != NULL)
1108 (p->p_sysent->sv_schedtail)(td);
1109 td->td_pflags &= ~TDP_FORKING;
1113 * Simplified back end of syscall(), used when returning from fork()
1114 * directly into user mode. This function is passed in to fork_exit()
1115 * as the first parameter and is called when returning to a new
1119 fork_return(struct thread *td, struct trapframe *frame)
1124 if (td->td_dbgflags & TDB_STOPATFORK) {
1126 if ((p->p_flag & P_TRACED) != 0) {
1128 * Inform the debugger if one is still present.
1130 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1131 ptracestop(td, SIGSTOP, NULL);
1132 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1135 * ... otherwise clear the request.
1137 td->td_dbgflags &= ~TDB_STOPATFORK;
1140 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1142 * This is the start of a new thread in a traced
1143 * process. Report a system call exit event.
1146 td->td_dbgflags |= TDB_SCX;
1147 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1148 (td->td_dbgflags & TDB_BORN) != 0)
1149 ptracestop(td, SIGTRAP, NULL);
1150 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1155 * If the prison was killed mid-fork, die along with it.
1157 if (!prison_isalive(td->td_ucred->cr_prison))
1158 exit1(td, 0, SIGKILL);
1163 if (KTRPOINT(td, KTR_SYSRET))
1164 ktrsysret(td->td_sa.code, 0, 0);