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/pioctl.h>
63 #include <sys/ptrace.h>
64 #include <sys/racct.h>
65 #include <sys/resourcevar.h>
66 #include <sys/sched.h>
67 #include <sys/syscall.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
72 #include <sys/ktrace.h>
73 #include <sys/unistd.h>
76 #include <sys/sysent.h>
77 #include <sys/signalvar.h>
79 #include <security/audit/audit.h>
80 #include <security/mac/mac_framework.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_extern.h>
89 #include <sys/dtrace_bsd.h>
90 dtrace_fork_func_t dtrace_fasttrap_fork;
93 SDT_PROVIDER_DECLARE(proc);
94 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
96 #ifndef _SYS_SYSPROTO_H_
102 EVENTHANDLER_LIST_DECLARE(process_fork);
106 sys_fork(struct thread *td, struct fork_args *uap)
111 bzero(&fr, sizeof(fr));
112 fr.fr_flags = RFFDG | RFPROC;
114 error = fork1(td, &fr);
116 td->td_retval[0] = pid;
117 td->td_retval[1] = 0;
124 sys_pdfork(struct thread *td, struct pdfork_args *uap)
129 bzero(&fr, sizeof(fr));
130 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
133 fr.fr_pd_flags = uap->flags;
135 * It is necessary to return fd by reference because 0 is a valid file
136 * descriptor number, and the child needs to be able to distinguish
137 * itself from the parent using the return value.
139 error = fork1(td, &fr);
141 td->td_retval[0] = pid;
142 td->td_retval[1] = 0;
143 error = copyout(&fd, uap->fdp, sizeof(fd));
150 sys_vfork(struct thread *td, struct vfork_args *uap)
155 bzero(&fr, sizeof(fr));
156 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
158 error = fork1(td, &fr);
160 td->td_retval[0] = pid;
161 td->td_retval[1] = 0;
167 sys_rfork(struct thread *td, struct rfork_args *uap)
172 /* Don't allow kernel-only flags. */
173 if ((uap->flags & RFKERNELONLY) != 0)
176 AUDIT_ARG_FFLAGS(uap->flags);
177 bzero(&fr, sizeof(fr));
178 fr.fr_flags = uap->flags;
180 error = fork1(td, &fr);
182 td->td_retval[0] = pid;
183 td->td_retval[1] = 0;
188 int __exclusive_cache_line nprocs = 1; /* process 0 */
190 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
194 * Random component to lastpid generation. We mix in a random factor to make
195 * it a little harder to predict. We sanity check the modulus value to avoid
196 * doing it in critical paths. Don't let it be too small or we pointlessly
197 * waste randomness entropy, and don't let it be impossibly large. Using a
198 * modulus that is too big causes a LOT more process table scans and slows
199 * down fork processing as the pidchecked caching is defeated.
201 static int randompid = 0;
204 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
208 error = sysctl_wire_old_buffer(req, sizeof(int));
211 sx_xlock(&allproc_lock);
213 error = sysctl_handle_int(oidp, &pid, 0, req);
214 if (error == 0 && req->newptr != NULL) {
218 /* generate a random PID modulus between 100 and 1123 */
219 randompid = 100 + arc4random() % 1024;
220 else if (pid < 0 || pid > pid_max - 100)
222 randompid = pid_max - 100;
224 /* Make it reasonable */
229 sx_xunlock(&allproc_lock);
233 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
234 0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value");
236 extern bitstr_t proc_id_pidmap;
237 extern bitstr_t proc_id_grpidmap;
238 extern bitstr_t proc_id_sessidmap;
239 extern bitstr_t proc_id_reapmap;
242 * Find an unused process ID
244 * If RFHIGHPID is set (used during system boot), do not allocate
248 fork_findpid(int flags)
253 trypid = lastpid + 1;
254 if (flags & RFHIGHPID) {
259 trypid += arc4random() % randompid;
261 mtx_lock(&procid_lock);
264 * If the process ID prototype has wrapped around,
265 * restart somewhat above 0, as the low-numbered procs
266 * tend to include daemons that don't exit.
268 if (trypid >= pid_max) {
269 trypid = trypid % pid_max;
274 bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
279 if (bit_test(&proc_id_grpidmap, result) ||
280 bit_test(&proc_id_sessidmap, result) ||
281 bit_test(&proc_id_reapmap, result)) {
287 * RFHIGHPID does not mess with the lastpid counter during boot.
289 if ((flags & RFHIGHPID) == 0)
292 bit_set(&proc_id_pidmap, result);
293 mtx_unlock(&procid_lock);
299 fork_norfproc(struct thread *td, int flags)
304 KASSERT((flags & RFPROC) == 0,
305 ("fork_norfproc called with RFPROC set"));
308 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
309 (flags & (RFCFDG | RFFDG))) {
311 if (thread_single(p1, SINGLE_BOUNDARY)) {
318 error = vm_forkproc(td, NULL, NULL, NULL, flags);
323 * Close all file descriptors.
325 if (flags & RFCFDG) {
326 struct filedesc *fdtmp;
327 fdtmp = fdinit(td->td_proc->p_fd, false);
333 * Unshare file descriptors (from parent).
339 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
340 (flags & (RFCFDG | RFFDG))) {
342 thread_single_end(p1, SINGLE_BOUNDARY);
349 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
350 struct vmspace *vm2, struct file *fp_procdesc)
352 struct proc *p1, *pptr;
355 struct filedesc_to_leader *fdtol;
356 struct sigacts *newsigacts;
358 sx_assert(&allproc_lock, SX_XLOCKED);
362 trypid = fork_findpid(fr->fr_flags);
363 p2->p_state = PRS_NEW; /* protect against others */
365 AUDIT_ARG_PID(p2->p_pid);
366 LIST_INSERT_HEAD(&allproc, p2, p_list);
368 sx_xlock(PIDHASHLOCK(p2->p_pid));
369 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
370 sx_xunlock(PIDHASHLOCK(p2->p_pid));
374 sx_xunlock(&allproc_lock);
376 bcopy(&p1->p_startcopy, &p2->p_startcopy,
377 __rangeof(struct proc, p_startcopy, p_endcopy));
378 pargs_hold(p2->p_args);
382 bzero(&p2->p_startzero,
383 __rangeof(struct proc, p_startzero, p_endzero));
385 /* Tell the prison that we exist. */
386 prison_proc_hold(p2->p_ucred->cr_prison);
393 * Malloc things while we don't hold any locks.
395 if (fr->fr_flags & RFSIGSHARE)
398 newsigacts = sigacts_alloc();
403 if (fr->fr_flags & RFCFDG) {
404 fd = fdinit(p1->p_fd, false);
406 } else if (fr->fr_flags & RFFDG) {
407 fd = fdcopy(p1->p_fd);
410 fd = fdshare(p1->p_fd);
411 if (p1->p_fdtol == NULL)
412 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
414 if ((fr->fr_flags & RFTHREAD) != 0) {
416 * Shared file descriptor table, and shared
420 FILEDESC_XLOCK(p1->p_fd);
421 fdtol->fdl_refcount++;
422 FILEDESC_XUNLOCK(p1->p_fd);
425 * Shared file descriptor table, and different
428 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
433 * Make a proc table entry for the new process.
434 * Start by zeroing the section of proc that is zero-initialized,
435 * then copy the section that is copied directly from the parent.
441 bzero(&td2->td_startzero,
442 __rangeof(struct thread, td_startzero, td_endzero));
444 bcopy(&td->td_startcopy, &td2->td_startcopy,
445 __rangeof(struct thread, td_startcopy, td_endcopy));
447 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
448 td2->td_sigstk = td->td_sigstk;
449 td2->td_flags = TDF_INMEM;
450 td2->td_lend_user_pri = PRI_MAX;
454 td2->td_vnet_lpush = NULL;
458 * Allow the scheduler to initialize the child.
465 * Duplicate sub-structures as needed.
466 * Increase reference counts on shared objects.
468 p2->p_flag = P_INMEM;
469 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
470 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC | P2_TRAPCAP);
471 p2->p_swtick = ticks;
472 if (p1->p_flag & P_PROFIL)
475 if (fr->fr_flags & RFSIGSHARE) {
476 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
478 sigacts_copy(newsigacts, p1->p_sigacts);
479 p2->p_sigacts = newsigacts;
482 if (fr->fr_flags & RFTSIGZMB)
483 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
484 else if (fr->fr_flags & RFLINUXTHPN)
485 p2->p_sigparent = SIGUSR1;
487 p2->p_sigparent = SIGCHLD;
489 p2->p_textvp = p1->p_textvp;
493 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
494 p2->p_flag |= P_PROTECTED;
495 p2->p_flag2 |= P2_INHERIT_PROTECTED;
499 * p_limit is copy-on-write. Bump its refcount.
503 thread_cow_get_proc(td2, p2);
505 pstats_fork(p1->p_stats, p2->p_stats);
510 /* Bump references to the text vnode (for procfs). */
512 vrefact(p2->p_textvp);
515 * Set up linkage for kernel based threading.
517 if ((fr->fr_flags & RFTHREAD) != 0) {
518 mtx_lock(&ppeers_lock);
519 p2->p_peers = p1->p_peers;
521 p2->p_leader = p1->p_leader;
522 mtx_unlock(&ppeers_lock);
523 PROC_LOCK(p1->p_leader);
524 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
525 PROC_UNLOCK(p1->p_leader);
527 * The task leader is exiting, so process p1 is
528 * going to be killed shortly. Since p1 obviously
529 * isn't dead yet, we know that the leader is either
530 * sending SIGKILL's to all the processes in this
531 * task or is sleeping waiting for all the peers to
532 * exit. We let p1 complete the fork, but we need
533 * to go ahead and kill the new process p2 since
534 * the task leader may not get a chance to send
535 * SIGKILL to it. We leave it on the list so that
536 * the task leader will wait for this new process
540 kern_psignal(p2, SIGKILL);
543 PROC_UNLOCK(p1->p_leader);
549 sx_xlock(&proctree_lock);
550 PGRP_LOCK(p1->p_pgrp);
555 * Preserve some more flags in subprocess. P_PROFIL has already
558 p2->p_flag |= p1->p_flag & P_SUGID;
559 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING;
560 SESS_LOCK(p1->p_session);
561 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
562 p2->p_flag |= P_CONTROLT;
563 SESS_UNLOCK(p1->p_session);
564 if (fr->fr_flags & RFPPWAIT)
565 p2->p_flag |= P_PPWAIT;
567 p2->p_pgrp = p1->p_pgrp;
568 LIST_INSERT_AFTER(p1, p2, p_pglist);
569 PGRP_UNLOCK(p1->p_pgrp);
570 LIST_INIT(&p2->p_children);
571 LIST_INIT(&p2->p_orphans);
573 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
576 * If PF_FORK is set, the child process inherits the
577 * procfs ioctl flags from its parent.
579 if (p1->p_pfsflags & PF_FORK) {
580 p2->p_stops = p1->p_stops;
581 p2->p_pfsflags = p1->p_pfsflags;
585 * This begins the section where we must prevent the parent
586 * from being swapped.
592 * Attach the new process to its parent.
594 * If RFNOWAIT is set, the newly created process becomes a child
595 * of init. This effectively disassociates the child from the
598 if ((fr->fr_flags & RFNOWAIT) != 0) {
602 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
607 p2->p_oppid = pptr->p_pid;
608 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
609 LIST_INIT(&p2->p_reaplist);
610 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
611 if (p2->p_reaper == p1 && p1 != initproc) {
612 p2->p_reapsubtree = p2->p_pid;
613 proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
615 sx_xunlock(&proctree_lock);
617 /* Inform accounting that we have forked. */
618 p2->p_acflag = AFORK;
626 * Finish creating the child process. It will return via a different
627 * execution path later. (ie: directly into user mode)
629 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
631 if (fr->fr_flags == (RFFDG | RFPROC)) {
633 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
634 p2->p_vmspace->vm_ssize);
635 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
636 VM_CNT_INC(v_vforks);
637 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
638 p2->p_vmspace->vm_ssize);
639 } else if (p1 == &proc0) {
640 VM_CNT_INC(v_kthreads);
641 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
642 p2->p_vmspace->vm_ssize);
644 VM_CNT_INC(v_rforks);
645 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
646 p2->p_vmspace->vm_ssize);
650 * Associate the process descriptor with the process before anything
651 * can happen that might cause that process to need the descriptor.
652 * However, don't do this until after fork(2) can no longer fail.
654 if (fr->fr_flags & RFPROCDESC)
655 procdesc_new(p2, fr->fr_pd_flags);
658 * Both processes are set up, now check if any loadable modules want
659 * to adjust anything.
661 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
664 * Set the child start time and mark the process as being complete.
668 microuptime(&p2->p_stats->p_start);
670 p2->p_state = PRS_NORMAL;
675 * Tell the DTrace fasttrap provider about the new process so that any
676 * tracepoints inherited from the parent can be removed. We have to do
677 * this only after p_state is PRS_NORMAL since the fasttrap module will
678 * use pfind() later on.
680 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
681 dtrace_fasttrap_fork(p1, p2);
683 if (fr->fr_flags & RFPPWAIT) {
684 td->td_pflags |= TDP_RFPPWAIT;
685 td->td_rfppwait_p = p2;
686 td->td_dbgflags |= TDB_VFORK;
691 * Tell any interested parties about the new process.
693 knote_fork(p1->p_klist, p2->p_pid);
696 * Now can be swapped.
700 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
702 if (fr->fr_flags & RFPROCDESC) {
703 procdesc_finit(p2->p_procdesc, fp_procdesc);
704 fdrop(fp_procdesc, td);
708 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
709 * synced with forks in progress so it is OK if we miss it
712 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
713 sx_xlock(&proctree_lock);
717 * p1->p_ptevents & p1->p_pptr are protected by both
718 * process and proctree locks for modifications,
719 * so owning proctree_lock allows the race-free read.
721 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
723 * Arrange for debugger to receive the fork event.
725 * We can report PL_FLAG_FORKED regardless of
726 * P_FOLLOWFORK settings, but it does not make a sense
729 td->td_dbgflags |= TDB_FORK;
730 td->td_dbg_forked = p2->p_pid;
731 td2->td_dbgflags |= TDB_STOPATFORK;
732 proc_set_traced(p2, true);
734 "do_fork: attaching to new child pid %d: oppid %d",
735 p2->p_pid, p2->p_oppid);
736 proc_reparent(p2, p1->p_pptr, false);
739 sx_xunlock(&proctree_lock);
742 racct_proc_fork_done(p2);
744 if ((fr->fr_flags & RFSTOPPED) == 0) {
745 if (fr->fr_pidp != NULL)
746 *fr->fr_pidp = p2->p_pid;
748 * If RFSTOPPED not requested, make child runnable and
753 sched_add(td2, SRQ_BORING);
761 fork_rfppwait(struct thread *td)
765 MPASS(td->td_pflags & TDP_RFPPWAIT);
769 * Preserve synchronization semantics of vfork. If
770 * waiting for child to exec or exit, fork set
771 * P_PPWAIT on child, and there we sleep on our proc
774 * Do it after the ptracestop() above is finished, to
775 * not block our debugger until child execs or exits
776 * to finish vfork wait.
778 td->td_pflags &= ~TDP_RFPPWAIT;
779 p2 = td->td_rfppwait_p;
782 while (p2->p_flag & P_PPWAIT) {
784 if (thread_suspend_check_needed()) {
786 thread_suspend_check(0);
792 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
796 if (td->td_dbgflags & TDB_VFORK) {
798 if (p->p_ptevents & PTRACE_VFORK)
799 ptracestop(td, SIGTRAP, NULL);
800 td->td_dbgflags &= ~TDB_VFORK;
806 fork1(struct thread *td, struct fork_req *fr)
808 struct proc *p1, *newproc;
811 struct file *fp_procdesc;
812 vm_ooffset_t mem_charged;
813 int error, nprocs_new, ok;
815 static struct timeval lastfail;
818 flags = fr->fr_flags;
819 pages = fr->fr_pages;
821 if ((flags & RFSTOPPED) != 0)
822 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
824 MPASS(fr->fr_procp == NULL);
826 /* Check for the undefined or unimplemented flags. */
827 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
830 /* Signal value requires RFTSIGZMB. */
831 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
834 /* Can't copy and clear. */
835 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
838 /* Check the validity of the signal number. */
839 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
842 if ((flags & RFPROCDESC) != 0) {
843 /* Can't not create a process yet get a process descriptor. */
844 if ((flags & RFPROC) == 0)
847 /* Must provide a place to put a procdesc if creating one. */
848 if (fr->fr_pd_fd == NULL)
851 /* Check if we are using supported flags. */
852 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
859 * Here we don't create a new process, but we divorce
860 * certain parts of a process from itself.
862 if ((flags & RFPROC) == 0) {
863 if (fr->fr_procp != NULL)
864 *fr->fr_procp = NULL;
865 else if (fr->fr_pidp != NULL)
867 return (fork_norfproc(td, flags));
875 * Increment the nprocs resource before allocations occur.
876 * Although process entries are dynamically created, we still
877 * keep a global limit on the maximum number we will
878 * create. There are hard-limits as to the number of processes
879 * that can run, established by the KVA and memory usage for
882 * Don't allow a nonprivileged user to use the last ten
883 * processes; don't let root exceed the limit.
885 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
886 if (nprocs_new >= maxproc - 10) {
887 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
888 nprocs_new >= maxproc) {
890 sx_xlock(&allproc_lock);
891 if (ppsratecheck(&lastfail, &curfail, 1)) {
892 printf("maxproc limit exceeded by uid %u "
893 "(pid %d); see tuning(7) and "
895 td->td_ucred->cr_ruid, p1->p_pid);
897 sx_xunlock(&allproc_lock);
903 * If required, create a process descriptor in the parent first; we
904 * will abandon it if something goes wrong. We don't finit() until
907 if (flags & RFPROCDESC) {
908 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
909 fr->fr_pd_flags, fr->fr_pd_fcaps);
916 pages = kstack_pages;
917 /* Allocate new proc. */
918 newproc = uma_zalloc(proc_zone, M_WAITOK);
919 td2 = FIRST_THREAD_IN_PROC(newproc);
921 td2 = thread_alloc(pages);
926 proc_linkup(newproc, td2);
928 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
929 if (td2->td_kstack != 0)
930 vm_thread_dispose(td2);
931 if (!thread_alloc_stack(td2, pages)) {
938 if ((flags & RFMEM) == 0) {
939 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
944 if (!swap_reserve(mem_charged)) {
946 * The swap reservation failed. The accounting
947 * from the entries of the copied vm2 will be
948 * subtracted in vmspace_free(), so force the
951 swap_reserve_force(mem_charged);
959 * XXX: This is ugly; when we copy resource usage, we need to bump
960 * per-cred resource counters.
962 proc_set_cred_init(newproc, crhold(td->td_ucred));
965 * Initialize resource accounting for the child process.
967 error = racct_proc_fork(p1, newproc);
974 mac_proc_init(newproc);
976 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
977 STAILQ_INIT(&newproc->p_ktr);
979 sx_xlock(&allproc_lock);
982 * Increment the count of procs running with this uid. Don't allow
983 * a nonprivileged user to exceed their current limit.
985 * XXXRW: Can we avoid privilege here if it's not needed?
987 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT);
989 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
991 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
992 lim_cur(td, RLIMIT_NPROC));
995 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1000 sx_xunlock(&allproc_lock);
1002 mac_proc_destroy(newproc);
1004 racct_proc_exit(newproc);
1006 crfree(newproc->p_ucred);
1007 newproc->p_ucred = NULL;
1011 uma_zfree(proc_zone, newproc);
1012 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1013 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1014 fdrop(fp_procdesc, td);
1016 atomic_add_int(&nprocs, -1);
1017 pause("fork", hz / 2);
1022 * Handle the return of a child process from fork1(). This function
1023 * is called from the MD fork_trampoline() entry point.
1026 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1027 struct trapframe *frame)
1035 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1037 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1038 td, td_get_sched(td), p->p_pid, td->td_name);
1040 sched_fork_exit(td);
1042 * Processes normally resume in mi_switch() after being
1043 * cpu_switch()'ed to, but when children start up they arrive here
1044 * instead, so we must do much the same things as mi_switch() would.
1046 if ((dtd = PCPU_GET(deadthread))) {
1047 PCPU_SET(deadthread, NULL);
1053 * cpu_fork_kthread_handler intercepts this function call to
1054 * have this call a non-return function to stay in kernel mode.
1055 * initproc has its own fork handler, but it does return.
1057 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1058 callout(arg, frame);
1061 * Check if a kernel thread misbehaved and returned from its main
1064 if (p->p_flag & P_KPROC) {
1065 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1066 td->td_name, p->p_pid);
1069 mtx_assert(&Giant, MA_NOTOWNED);
1071 if (p->p_sysent->sv_schedtail != NULL)
1072 (p->p_sysent->sv_schedtail)(td);
1073 td->td_pflags &= ~TDP_FORKING;
1077 * Simplified back end of syscall(), used when returning from fork()
1078 * directly into user mode. This function is passed in to fork_exit()
1079 * as the first parameter and is called when returning to a new
1083 fork_return(struct thread *td, struct trapframe *frame)
1088 if (td->td_dbgflags & TDB_STOPATFORK) {
1090 if ((p->p_flag & P_TRACED) != 0) {
1092 * Inform the debugger if one is still present.
1094 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1095 ptracestop(td, SIGSTOP, NULL);
1096 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1099 * ... otherwise clear the request.
1101 td->td_dbgflags &= ~TDB_STOPATFORK;
1104 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1106 * This is the start of a new thread in a traced
1107 * process. Report a system call exit event.
1110 td->td_dbgflags |= TDB_SCX;
1111 _STOPEVENT(p, S_SCX, td->td_sa.code);
1112 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1113 (td->td_dbgflags & TDB_BORN) != 0)
1114 ptracestop(td, SIGTRAP, NULL);
1115 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1122 if (KTRPOINT(td, KTR_SYSRET))
1123 ktrsysret(SYS_fork, 0, 0);