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 * 3. 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"
41 #include "opt_kstack_pages.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/eventhandler.h>
47 #include <sys/fcntl.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/procdesc.h>
59 #include <sys/pioctl.h>
60 #include <sys/ptrace.h>
61 #include <sys/racct.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sched.h>
64 #include <sys/syscall.h>
65 #include <sys/vmmeter.h>
66 #include <sys/vnode.h>
69 #include <sys/ktrace.h>
70 #include <sys/unistd.h>
73 #include <sys/sysent.h>
74 #include <sys/signalvar.h>
76 #include <security/audit/audit.h>
77 #include <security/mac/mac_framework.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_extern.h>
84 #include <vm/vm_domain.h>
87 #include <sys/dtrace_bsd.h>
88 dtrace_fork_func_t dtrace_fasttrap_fork;
91 SDT_PROVIDER_DECLARE(proc);
92 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
94 #ifndef _SYS_SYSPROTO_H_
100 EVENTHANDLER_LIST_DECLARE(process_fork);
104 sys_fork(struct thread *td, struct fork_args *uap)
109 bzero(&fr, sizeof(fr));
110 fr.fr_flags = RFFDG | RFPROC;
112 error = fork1(td, &fr);
114 td->td_retval[0] = pid;
115 td->td_retval[1] = 0;
122 sys_pdfork(struct thread *td, struct pdfork_args *uap)
127 bzero(&fr, sizeof(fr));
128 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
131 fr.fr_pd_flags = 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)
174 AUDIT_ARG_FFLAGS(uap->flags);
175 bzero(&fr, sizeof(fr));
176 fr.fr_flags = uap->flags;
178 error = fork1(td, &fr);
180 td->td_retval[0] = pid;
181 td->td_retval[1] = 0;
186 int nprocs = 1; /* process 0 */
188 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
192 * Random component to lastpid generation. We mix in a random factor to make
193 * it a little harder to predict. We sanity check the modulus value to avoid
194 * doing it in critical paths. Don't let it be too small or we pointlessly
195 * waste randomness entropy, and don't let it be impossibly large. Using a
196 * modulus that is too big causes a LOT more process table scans and slows
197 * down fork processing as the pidchecked caching is defeated.
199 static int randompid = 0;
202 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
206 error = sysctl_wire_old_buffer(req, sizeof(int));
209 sx_xlock(&allproc_lock);
211 error = sysctl_handle_int(oidp, &pid, 0, req);
212 if (error == 0 && req->newptr != NULL) {
216 /* generate a random PID modulus between 100 and 1123 */
217 randompid = 100 + arc4random() % 1024;
218 else if (pid < 0 || pid > pid_max - 100)
220 randompid = pid_max - 100;
222 /* Make it reasonable */
227 sx_xunlock(&allproc_lock);
231 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
232 0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value");
235 fork_findpid(int flags)
239 static int pidchecked = 0;
242 * Requires allproc_lock in order to iterate over the list
243 * of processes, and proctree_lock to access p_pgrp.
245 sx_assert(&allproc_lock, SX_LOCKED);
246 sx_assert(&proctree_lock, SX_LOCKED);
249 * Find an unused process ID. We remember a range of unused IDs
250 * ready to use (from lastpid+1 through pidchecked-1).
252 * If RFHIGHPID is set (used during system boot), do not allocate
255 trypid = lastpid + 1;
256 if (flags & RFHIGHPID) {
261 trypid += arc4random() % randompid;
265 * If the process ID prototype has wrapped around,
266 * restart somewhat above 0, as the low-numbered procs
267 * tend to include daemons that don't exit.
269 if (trypid >= pid_max) {
270 trypid = trypid % pid_max;
275 if (trypid >= pidchecked) {
278 pidchecked = PID_MAX;
280 * Scan the active and zombie procs to check whether this pid
281 * is in use. Remember the lowest pid that's greater
282 * than trypid, so we can avoid checking for a while.
284 * Avoid reuse of the process group id, session id or
285 * the reaper subtree id. Note that for process group
286 * and sessions, the amount of reserved pids is
287 * limited by process limit. For the subtree ids, the
288 * id is kept reserved only while there is a
289 * non-reaped process in the subtree, so amount of
290 * reserved pids is limited by process limit times
293 p = LIST_FIRST(&allproc);
295 for (; p != NULL; p = LIST_NEXT(p, p_list)) {
296 while (p->p_pid == trypid ||
297 p->p_reapsubtree == trypid ||
298 (p->p_pgrp != NULL &&
299 (p->p_pgrp->pg_id == trypid ||
300 (p->p_session != NULL &&
301 p->p_session->s_sid == trypid)))) {
303 if (trypid >= pidchecked)
306 if (p->p_pid > trypid && pidchecked > p->p_pid)
307 pidchecked = p->p_pid;
308 if (p->p_pgrp != NULL) {
309 if (p->p_pgrp->pg_id > trypid &&
310 pidchecked > p->p_pgrp->pg_id)
311 pidchecked = p->p_pgrp->pg_id;
312 if (p->p_session != NULL &&
313 p->p_session->s_sid > trypid &&
314 pidchecked > p->p_session->s_sid)
315 pidchecked = p->p_session->s_sid;
320 p = LIST_FIRST(&zombproc);
326 * RFHIGHPID does not mess with the lastpid counter during boot.
328 if (flags & RFHIGHPID)
337 fork_norfproc(struct thread *td, int flags)
342 KASSERT((flags & RFPROC) == 0,
343 ("fork_norfproc called with RFPROC set"));
346 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
347 (flags & (RFCFDG | RFFDG))) {
349 if (thread_single(p1, SINGLE_BOUNDARY)) {
356 error = vm_forkproc(td, NULL, NULL, NULL, flags);
361 * Close all file descriptors.
363 if (flags & RFCFDG) {
364 struct filedesc *fdtmp;
365 fdtmp = fdinit(td->td_proc->p_fd, false);
371 * Unshare file descriptors (from parent).
377 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
378 (flags & (RFCFDG | RFFDG))) {
380 thread_single_end(p1, SINGLE_BOUNDARY);
387 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
388 struct vmspace *vm2, struct file *fp_procdesc)
390 struct proc *p1, *pptr;
393 struct filedesc_to_leader *fdtol;
394 struct sigacts *newsigacts;
396 sx_assert(&proctree_lock, SX_SLOCKED);
397 sx_assert(&allproc_lock, SX_XLOCKED);
401 trypid = fork_findpid(fr->fr_flags);
403 sx_sunlock(&proctree_lock);
405 p2->p_state = PRS_NEW; /* protect against others */
407 AUDIT_ARG_PID(p2->p_pid);
408 LIST_INSERT_HEAD(&allproc, p2, p_list);
410 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
415 sx_xunlock(&allproc_lock);
417 bcopy(&p1->p_startcopy, &p2->p_startcopy,
418 __rangeof(struct proc, p_startcopy, p_endcopy));
419 pargs_hold(p2->p_args);
423 bzero(&p2->p_startzero,
424 __rangeof(struct proc, p_startzero, p_endzero));
426 /* Tell the prison that we exist. */
427 prison_proc_hold(p2->p_ucred->cr_prison);
432 * Malloc things while we don't hold any locks.
434 if (fr->fr_flags & RFSIGSHARE)
437 newsigacts = sigacts_alloc();
442 if (fr->fr_flags & RFCFDG) {
443 fd = fdinit(p1->p_fd, false);
445 } else if (fr->fr_flags & RFFDG) {
446 fd = fdcopy(p1->p_fd);
449 fd = fdshare(p1->p_fd);
450 if (p1->p_fdtol == NULL)
451 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
453 if ((fr->fr_flags & RFTHREAD) != 0) {
455 * Shared file descriptor table, and shared
459 FILEDESC_XLOCK(p1->p_fd);
460 fdtol->fdl_refcount++;
461 FILEDESC_XUNLOCK(p1->p_fd);
464 * Shared file descriptor table, and different
467 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
472 * Make a proc table entry for the new process.
473 * Start by zeroing the section of proc that is zero-initialized,
474 * then copy the section that is copied directly from the parent.
480 bzero(&td2->td_startzero,
481 __rangeof(struct thread, td_startzero, td_endzero));
483 bcopy(&td->td_startcopy, &td2->td_startcopy,
484 __rangeof(struct thread, td_startcopy, td_endcopy));
486 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
487 td2->td_sigstk = td->td_sigstk;
488 td2->td_flags = TDF_INMEM;
489 td2->td_lend_user_pri = PRI_MAX;
493 td2->td_vnet_lpush = NULL;
497 * Allow the scheduler to initialize the child.
504 * Duplicate sub-structures as needed.
505 * Increase reference counts on shared objects.
507 p2->p_flag = P_INMEM;
508 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC | P2_TRAPCAP);
509 p2->p_swtick = ticks;
510 if (p1->p_flag & P_PROFIL)
514 * Whilst the proc lock is held, copy the VM domain data out
515 * using the VM domain method.
517 vm_domain_policy_init(&p2->p_vm_dom_policy);
518 vm_domain_policy_localcopy(&p2->p_vm_dom_policy,
519 &p1->p_vm_dom_policy);
521 if (fr->fr_flags & RFSIGSHARE) {
522 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
524 sigacts_copy(newsigacts, p1->p_sigacts);
525 p2->p_sigacts = newsigacts;
528 if (fr->fr_flags & RFTSIGZMB)
529 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
530 else if (fr->fr_flags & RFLINUXTHPN)
531 p2->p_sigparent = SIGUSR1;
533 p2->p_sigparent = SIGCHLD;
535 p2->p_textvp = p1->p_textvp;
539 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
540 p2->p_flag |= P_PROTECTED;
541 p2->p_flag2 |= P2_INHERIT_PROTECTED;
545 * p_limit is copy-on-write. Bump its refcount.
549 thread_cow_get_proc(td2, p2);
551 pstats_fork(p1->p_stats, p2->p_stats);
556 /* Bump references to the text vnode (for procfs). */
558 vrefact(p2->p_textvp);
561 * Set up linkage for kernel based threading.
563 if ((fr->fr_flags & RFTHREAD) != 0) {
564 mtx_lock(&ppeers_lock);
565 p2->p_peers = p1->p_peers;
567 p2->p_leader = p1->p_leader;
568 mtx_unlock(&ppeers_lock);
569 PROC_LOCK(p1->p_leader);
570 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
571 PROC_UNLOCK(p1->p_leader);
573 * The task leader is exiting, so process p1 is
574 * going to be killed shortly. Since p1 obviously
575 * isn't dead yet, we know that the leader is either
576 * sending SIGKILL's to all the processes in this
577 * task or is sleeping waiting for all the peers to
578 * exit. We let p1 complete the fork, but we need
579 * to go ahead and kill the new process p2 since
580 * the task leader may not get a chance to send
581 * SIGKILL to it. We leave it on the list so that
582 * the task leader will wait for this new process
586 kern_psignal(p2, SIGKILL);
589 PROC_UNLOCK(p1->p_leader);
595 sx_xlock(&proctree_lock);
596 PGRP_LOCK(p1->p_pgrp);
601 * Preserve some more flags in subprocess. P_PROFIL has already
604 p2->p_flag |= p1->p_flag & P_SUGID;
605 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING;
606 SESS_LOCK(p1->p_session);
607 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
608 p2->p_flag |= P_CONTROLT;
609 SESS_UNLOCK(p1->p_session);
610 if (fr->fr_flags & RFPPWAIT)
611 p2->p_flag |= P_PPWAIT;
613 p2->p_pgrp = p1->p_pgrp;
614 LIST_INSERT_AFTER(p1, p2, p_pglist);
615 PGRP_UNLOCK(p1->p_pgrp);
616 LIST_INIT(&p2->p_children);
617 LIST_INIT(&p2->p_orphans);
619 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
622 * If PF_FORK is set, the child process inherits the
623 * procfs ioctl flags from its parent.
625 if (p1->p_pfsflags & PF_FORK) {
626 p2->p_stops = p1->p_stops;
627 p2->p_pfsflags = p1->p_pfsflags;
631 * This begins the section where we must prevent the parent
632 * from being swapped.
638 * Attach the new process to its parent.
640 * If RFNOWAIT is set, the newly created process becomes a child
641 * of init. This effectively disassociates the child from the
644 if ((fr->fr_flags & RFNOWAIT) != 0) {
648 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
653 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
654 LIST_INIT(&p2->p_reaplist);
655 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
656 if (p2->p_reaper == p1)
657 p2->p_reapsubtree = p2->p_pid;
658 sx_xunlock(&proctree_lock);
660 /* Inform accounting that we have forked. */
661 p2->p_acflag = AFORK;
669 * Finish creating the child process. It will return via a different
670 * execution path later. (ie: directly into user mode)
672 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
674 if (fr->fr_flags == (RFFDG | RFPROC)) {
676 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
677 p2->p_vmspace->vm_ssize);
678 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
679 VM_CNT_INC(v_vforks);
680 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
681 p2->p_vmspace->vm_ssize);
682 } else if (p1 == &proc0) {
683 VM_CNT_INC(v_kthreads);
684 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
685 p2->p_vmspace->vm_ssize);
687 VM_CNT_INC(v_rforks);
688 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
689 p2->p_vmspace->vm_ssize);
693 * Associate the process descriptor with the process before anything
694 * can happen that might cause that process to need the descriptor.
695 * However, don't do this until after fork(2) can no longer fail.
697 if (fr->fr_flags & RFPROCDESC)
698 procdesc_new(p2, fr->fr_pd_flags);
701 * Both processes are set up, now check if any loadable modules want
702 * to adjust anything.
704 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
707 * Set the child start time and mark the process as being complete.
711 microuptime(&p2->p_stats->p_start);
713 p2->p_state = PRS_NORMAL;
718 * Tell the DTrace fasttrap provider about the new process so that any
719 * tracepoints inherited from the parent can be removed. We have to do
720 * this only after p_state is PRS_NORMAL since the fasttrap module will
721 * use pfind() later on.
723 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
724 dtrace_fasttrap_fork(p1, p2);
727 * Hold the process so that it cannot exit after we make it runnable,
728 * but before we wait for the debugger.
731 if (p1->p_ptevents & PTRACE_FORK) {
733 * Arrange for debugger to receive the fork event.
735 * We can report PL_FLAG_FORKED regardless of
736 * P_FOLLOWFORK settings, but it does not make a sense
739 td->td_dbgflags |= TDB_FORK;
740 td->td_dbg_forked = p2->p_pid;
741 td2->td_dbgflags |= TDB_STOPATFORK;
743 if (fr->fr_flags & RFPPWAIT) {
744 td->td_pflags |= TDP_RFPPWAIT;
745 td->td_rfppwait_p = p2;
746 td->td_dbgflags |= TDB_VFORK;
751 * Now can be swapped.
757 * Tell any interested parties about the new process.
759 knote_fork(p1->p_klist, p2->p_pid);
760 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
762 if (fr->fr_flags & RFPROCDESC) {
763 procdesc_finit(p2->p_procdesc, fp_procdesc);
764 fdrop(fp_procdesc, td);
767 if ((fr->fr_flags & RFSTOPPED) == 0) {
769 * If RFSTOPPED not requested, make child runnable and
774 sched_add(td2, SRQ_BORING);
776 if (fr->fr_pidp != NULL)
777 *fr->fr_pidp = p2->p_pid;
784 * Wait until debugger is attached to child.
786 while (td2->td_proc == p2 && (td2->td_dbgflags & TDB_STOPATFORK) != 0)
787 cv_wait(&p2->p_dbgwait, &p2->p_mtx);
789 racct_proc_fork_done(p2);
794 fork1(struct thread *td, struct fork_req *fr)
796 struct proc *p1, *newproc;
799 struct file *fp_procdesc;
800 vm_ooffset_t mem_charged;
801 int error, nprocs_new, ok;
803 static struct timeval lastfail;
806 flags = fr->fr_flags;
807 pages = fr->fr_pages;
809 if ((flags & RFSTOPPED) != 0)
810 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
812 MPASS(fr->fr_procp == NULL);
814 /* Check for the undefined or unimplemented flags. */
815 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
818 /* Signal value requires RFTSIGZMB. */
819 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
822 /* Can't copy and clear. */
823 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
826 /* Check the validity of the signal number. */
827 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
830 if ((flags & RFPROCDESC) != 0) {
831 /* Can't not create a process yet get a process descriptor. */
832 if ((flags & RFPROC) == 0)
835 /* Must provide a place to put a procdesc if creating one. */
836 if (fr->fr_pd_fd == NULL)
839 /* Check if we are using supported flags. */
840 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
847 * Here we don't create a new process, but we divorce
848 * certain parts of a process from itself.
850 if ((flags & RFPROC) == 0) {
851 if (fr->fr_procp != NULL)
852 *fr->fr_procp = NULL;
853 else if (fr->fr_pidp != NULL)
855 return (fork_norfproc(td, flags));
863 * Increment the nprocs resource before allocations occur.
864 * Although process entries are dynamically created, we still
865 * keep a global limit on the maximum number we will
866 * create. There are hard-limits as to the number of processes
867 * that can run, established by the KVA and memory usage for
870 * Don't allow a nonprivileged user to use the last ten
871 * processes; don't let root exceed the limit.
873 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
874 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred,
875 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) {
877 sx_xlock(&allproc_lock);
878 if (ppsratecheck(&lastfail, &curfail, 1)) {
879 printf("maxproc limit exceeded by uid %u (pid %d); "
880 "see tuning(7) and login.conf(5)\n",
881 td->td_ucred->cr_ruid, p1->p_pid);
883 sx_xunlock(&allproc_lock);
888 * If required, create a process descriptor in the parent first; we
889 * will abandon it if something goes wrong. We don't finit() until
892 if (flags & RFPROCDESC) {
893 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
894 fr->fr_pd_flags, fr->fr_pd_fcaps);
901 pages = kstack_pages;
902 /* Allocate new proc. */
903 newproc = uma_zalloc(proc_zone, M_WAITOK);
904 td2 = FIRST_THREAD_IN_PROC(newproc);
906 td2 = thread_alloc(pages);
911 proc_linkup(newproc, td2);
913 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
914 if (td2->td_kstack != 0)
915 vm_thread_dispose(td2);
916 if (!thread_alloc_stack(td2, pages)) {
923 if ((flags & RFMEM) == 0) {
924 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
929 if (!swap_reserve(mem_charged)) {
931 * The swap reservation failed. The accounting
932 * from the entries of the copied vm2 will be
933 * subtracted in vmspace_free(), so force the
936 swap_reserve_force(mem_charged);
944 * XXX: This is ugly; when we copy resource usage, we need to bump
945 * per-cred resource counters.
947 proc_set_cred_init(newproc, crhold(td->td_ucred));
950 * Initialize resource accounting for the child process.
952 error = racct_proc_fork(p1, newproc);
959 mac_proc_init(newproc);
961 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
962 STAILQ_INIT(&newproc->p_ktr);
964 /* We have to lock the process tree while we look for a pid. */
965 sx_slock(&proctree_lock);
966 sx_xlock(&allproc_lock);
969 * Increment the count of procs running with this uid. Don't allow
970 * a nonprivileged user to exceed their current limit.
972 * XXXRW: Can we avoid privilege here if it's not needed?
974 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
976 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
978 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
979 lim_cur(td, RLIMIT_NPROC));
982 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
987 sx_sunlock(&proctree_lock);
988 sx_xunlock(&allproc_lock);
990 mac_proc_destroy(newproc);
992 racct_proc_exit(newproc);
994 crfree(newproc->p_ucred);
995 newproc->p_ucred = NULL;
999 uma_zfree(proc_zone, newproc);
1000 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1001 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1002 fdrop(fp_procdesc, td);
1004 atomic_add_int(&nprocs, -1);
1005 pause("fork", hz / 2);
1010 * Handle the return of a child process from fork1(). This function
1011 * is called from the MD fork_trampoline() entry point.
1014 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1015 struct trapframe *frame)
1023 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1025 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1026 td, td_get_sched(td), p->p_pid, td->td_name);
1028 sched_fork_exit(td);
1030 * Processes normally resume in mi_switch() after being
1031 * cpu_switch()'ed to, but when children start up they arrive here
1032 * instead, so we must do much the same things as mi_switch() would.
1034 if ((dtd = PCPU_GET(deadthread))) {
1035 PCPU_SET(deadthread, NULL);
1041 * cpu_fork_kthread_handler intercepts this function call to
1042 * have this call a non-return function to stay in kernel mode.
1043 * initproc has its own fork handler, but it does return.
1045 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1046 callout(arg, frame);
1049 * Check if a kernel thread misbehaved and returned from its main
1052 if (p->p_flag & P_KPROC) {
1053 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1054 td->td_name, p->p_pid);
1057 mtx_assert(&Giant, MA_NOTOWNED);
1059 if (p->p_sysent->sv_schedtail != NULL)
1060 (p->p_sysent->sv_schedtail)(td);
1061 td->td_pflags &= ~TDP_FORKING;
1065 * Simplified back end of syscall(), used when returning from fork()
1066 * directly into user mode. This function is passed in to fork_exit()
1067 * as the first parameter and is called when returning to a new
1071 fork_return(struct thread *td, struct trapframe *frame)
1073 struct proc *p, *dbg;
1076 if (td->td_dbgflags & TDB_STOPATFORK) {
1077 sx_xlock(&proctree_lock);
1079 if (p->p_pptr->p_ptevents & PTRACE_FORK) {
1081 * If debugger still wants auto-attach for the
1082 * parent's children, do it now.
1084 dbg = p->p_pptr->p_pptr;
1085 proc_set_traced(p, true);
1087 "fork_return: attaching to new child pid %d: oppid %d",
1088 p->p_pid, p->p_oppid);
1089 proc_reparent(p, dbg);
1090 sx_xunlock(&proctree_lock);
1091 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1092 ptracestop(td, SIGSTOP, NULL);
1093 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1096 * ... otherwise clear the request.
1098 sx_xunlock(&proctree_lock);
1099 td->td_dbgflags &= ~TDB_STOPATFORK;
1100 cv_broadcast(&p->p_dbgwait);
1103 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1105 * This is the start of a new thread in a traced
1106 * process. Report a system call exit event.
1109 td->td_dbgflags |= TDB_SCX;
1110 _STOPEVENT(p, S_SCX, td->td_sa.code);
1111 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1112 (td->td_dbgflags & TDB_BORN) != 0)
1113 ptracestop(td, SIGTRAP, NULL);
1114 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1121 if (KTRPOINT(td, KTR_SYSRET))
1122 ktrsysret(SYS_fork, 0, 0);