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/sysproto.h>
48 #include <sys/eventhandler.h>
49 #include <sys/fcntl.h>
50 #include <sys/filedesc.h>
52 #include <sys/kernel.h>
53 #include <sys/kthread.h>
54 #include <sys/sysctl.h>
56 #include <sys/malloc.h>
57 #include <sys/mutex.h>
60 #include <sys/procdesc.h>
61 #include <sys/pioctl.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>
86 #include <vm/vm_domain.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 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");
237 fork_findpid(int flags)
241 static int pidchecked = 0;
244 * Requires allproc_lock in order to iterate over the list
245 * of processes, and proctree_lock to access p_pgrp.
247 sx_assert(&allproc_lock, SX_LOCKED);
248 sx_assert(&proctree_lock, SX_LOCKED);
251 * Find an unused process ID. We remember a range of unused IDs
252 * ready to use (from lastpid+1 through pidchecked-1).
254 * If RFHIGHPID is set (used during system boot), do not allocate
257 trypid = lastpid + 1;
258 if (flags & RFHIGHPID) {
263 trypid += arc4random() % randompid;
267 * If the process ID prototype has wrapped around,
268 * restart somewhat above 0, as the low-numbered procs
269 * tend to include daemons that don't exit.
271 if (trypid >= pid_max) {
272 trypid = trypid % pid_max;
277 if (trypid >= pidchecked) {
280 pidchecked = PID_MAX;
282 * Scan the active and zombie procs to check whether this pid
283 * is in use. Remember the lowest pid that's greater
284 * than trypid, so we can avoid checking for a while.
286 * Avoid reuse of the process group id, session id or
287 * the reaper subtree id. Note that for process group
288 * and sessions, the amount of reserved pids is
289 * limited by process limit. For the subtree ids, the
290 * id is kept reserved only while there is a
291 * non-reaped process in the subtree, so amount of
292 * reserved pids is limited by process limit times
295 p = LIST_FIRST(&allproc);
297 for (; p != NULL; p = LIST_NEXT(p, p_list)) {
298 while (p->p_pid == trypid ||
299 p->p_reapsubtree == trypid ||
300 (p->p_pgrp != NULL &&
301 (p->p_pgrp->pg_id == trypid ||
302 (p->p_session != NULL &&
303 p->p_session->s_sid == trypid)))) {
305 if (trypid >= pidchecked)
308 if (p->p_pid > trypid && pidchecked > p->p_pid)
309 pidchecked = p->p_pid;
310 if (p->p_pgrp != NULL) {
311 if (p->p_pgrp->pg_id > trypid &&
312 pidchecked > p->p_pgrp->pg_id)
313 pidchecked = p->p_pgrp->pg_id;
314 if (p->p_session != NULL &&
315 p->p_session->s_sid > trypid &&
316 pidchecked > p->p_session->s_sid)
317 pidchecked = p->p_session->s_sid;
322 p = LIST_FIRST(&zombproc);
328 * RFHIGHPID does not mess with the lastpid counter during boot.
330 if (flags & RFHIGHPID)
339 fork_norfproc(struct thread *td, int flags)
344 KASSERT((flags & RFPROC) == 0,
345 ("fork_norfproc called with RFPROC set"));
348 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
349 (flags & (RFCFDG | RFFDG))) {
351 if (thread_single(p1, SINGLE_BOUNDARY)) {
358 error = vm_forkproc(td, NULL, NULL, NULL, flags);
363 * Close all file descriptors.
365 if (flags & RFCFDG) {
366 struct filedesc *fdtmp;
367 fdtmp = fdinit(td->td_proc->p_fd, false);
373 * Unshare file descriptors (from parent).
379 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
380 (flags & (RFCFDG | RFFDG))) {
382 thread_single_end(p1, SINGLE_BOUNDARY);
389 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
390 struct vmspace *vm2, struct file *fp_procdesc)
392 struct proc *p1, *pptr;
395 struct filedesc_to_leader *fdtol;
396 struct sigacts *newsigacts;
398 sx_assert(&proctree_lock, SX_SLOCKED);
399 sx_assert(&allproc_lock, SX_XLOCKED);
403 trypid = fork_findpid(fr->fr_flags);
405 sx_sunlock(&proctree_lock);
407 p2->p_state = PRS_NEW; /* protect against others */
409 AUDIT_ARG_PID(p2->p_pid);
410 LIST_INSERT_HEAD(&allproc, p2, p_list);
412 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
417 sx_xunlock(&allproc_lock);
419 bcopy(&p1->p_startcopy, &p2->p_startcopy,
420 __rangeof(struct proc, p_startcopy, p_endcopy));
421 pargs_hold(p2->p_args);
425 bzero(&p2->p_startzero,
426 __rangeof(struct proc, p_startzero, p_endzero));
428 /* Tell the prison that we exist. */
429 prison_proc_hold(p2->p_ucred->cr_prison);
434 * Malloc things while we don't hold any locks.
436 if (fr->fr_flags & RFSIGSHARE)
439 newsigacts = sigacts_alloc();
444 if (fr->fr_flags & RFCFDG) {
445 fd = fdinit(p1->p_fd, false);
447 } else if (fr->fr_flags & RFFDG) {
448 fd = fdcopy(p1->p_fd);
451 fd = fdshare(p1->p_fd);
452 if (p1->p_fdtol == NULL)
453 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
455 if ((fr->fr_flags & RFTHREAD) != 0) {
457 * Shared file descriptor table, and shared
461 FILEDESC_XLOCK(p1->p_fd);
462 fdtol->fdl_refcount++;
463 FILEDESC_XUNLOCK(p1->p_fd);
466 * Shared file descriptor table, and different
469 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
474 * Make a proc table entry for the new process.
475 * Start by zeroing the section of proc that is zero-initialized,
476 * then copy the section that is copied directly from the parent.
482 bzero(&td2->td_startzero,
483 __rangeof(struct thread, td_startzero, td_endzero));
485 bcopy(&td->td_startcopy, &td2->td_startcopy,
486 __rangeof(struct thread, td_startcopy, td_endcopy));
488 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
489 td2->td_sigstk = td->td_sigstk;
490 td2->td_flags = TDF_INMEM;
491 td2->td_lend_user_pri = PRI_MAX;
495 td2->td_vnet_lpush = NULL;
499 * Allow the scheduler to initialize the child.
506 * Duplicate sub-structures as needed.
507 * Increase reference counts on shared objects.
509 p2->p_flag = P_INMEM;
510 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC | P2_TRAPCAP);
511 p2->p_swtick = ticks;
512 if (p1->p_flag & P_PROFIL)
516 * Whilst the proc lock is held, copy the VM domain data out
517 * using the VM domain method.
519 vm_domain_policy_init(&p2->p_vm_dom_policy);
520 vm_domain_policy_localcopy(&p2->p_vm_dom_policy,
521 &p1->p_vm_dom_policy);
523 if (fr->fr_flags & RFSIGSHARE) {
524 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
526 sigacts_copy(newsigacts, p1->p_sigacts);
527 p2->p_sigacts = newsigacts;
530 if (fr->fr_flags & RFTSIGZMB)
531 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
532 else if (fr->fr_flags & RFLINUXTHPN)
533 p2->p_sigparent = SIGUSR1;
535 p2->p_sigparent = SIGCHLD;
537 p2->p_textvp = p1->p_textvp;
541 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
542 p2->p_flag |= P_PROTECTED;
543 p2->p_flag2 |= P2_INHERIT_PROTECTED;
547 * p_limit is copy-on-write. Bump its refcount.
551 thread_cow_get_proc(td2, p2);
553 pstats_fork(p1->p_stats, p2->p_stats);
558 /* Bump references to the text vnode (for procfs). */
560 vrefact(p2->p_textvp);
563 * Set up linkage for kernel based threading.
565 if ((fr->fr_flags & RFTHREAD) != 0) {
566 mtx_lock(&ppeers_lock);
567 p2->p_peers = p1->p_peers;
569 p2->p_leader = p1->p_leader;
570 mtx_unlock(&ppeers_lock);
571 PROC_LOCK(p1->p_leader);
572 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
573 PROC_UNLOCK(p1->p_leader);
575 * The task leader is exiting, so process p1 is
576 * going to be killed shortly. Since p1 obviously
577 * isn't dead yet, we know that the leader is either
578 * sending SIGKILL's to all the processes in this
579 * task or is sleeping waiting for all the peers to
580 * exit. We let p1 complete the fork, but we need
581 * to go ahead and kill the new process p2 since
582 * the task leader may not get a chance to send
583 * SIGKILL to it. We leave it on the list so that
584 * the task leader will wait for this new process
588 kern_psignal(p2, SIGKILL);
591 PROC_UNLOCK(p1->p_leader);
597 sx_xlock(&proctree_lock);
598 PGRP_LOCK(p1->p_pgrp);
603 * Preserve some more flags in subprocess. P_PROFIL has already
606 p2->p_flag |= p1->p_flag & P_SUGID;
607 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING;
608 SESS_LOCK(p1->p_session);
609 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
610 p2->p_flag |= P_CONTROLT;
611 SESS_UNLOCK(p1->p_session);
612 if (fr->fr_flags & RFPPWAIT)
613 p2->p_flag |= P_PPWAIT;
615 p2->p_pgrp = p1->p_pgrp;
616 LIST_INSERT_AFTER(p1, p2, p_pglist);
617 PGRP_UNLOCK(p1->p_pgrp);
618 LIST_INIT(&p2->p_children);
619 LIST_INIT(&p2->p_orphans);
621 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
624 * If PF_FORK is set, the child process inherits the
625 * procfs ioctl flags from its parent.
627 if (p1->p_pfsflags & PF_FORK) {
628 p2->p_stops = p1->p_stops;
629 p2->p_pfsflags = p1->p_pfsflags;
633 * This begins the section where we must prevent the parent
634 * from being swapped.
640 * Attach the new process to its parent.
642 * If RFNOWAIT is set, the newly created process becomes a child
643 * of init. This effectively disassociates the child from the
646 if ((fr->fr_flags & RFNOWAIT) != 0) {
650 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
655 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
656 LIST_INIT(&p2->p_reaplist);
657 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
658 if (p2->p_reaper == p1)
659 p2->p_reapsubtree = p2->p_pid;
660 sx_xunlock(&proctree_lock);
662 /* Inform accounting that we have forked. */
663 p2->p_acflag = AFORK;
671 * Finish creating the child process. It will return via a different
672 * execution path later. (ie: directly into user mode)
674 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
676 if (fr->fr_flags == (RFFDG | RFPROC)) {
678 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
679 p2->p_vmspace->vm_ssize);
680 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
681 VM_CNT_INC(v_vforks);
682 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
683 p2->p_vmspace->vm_ssize);
684 } else if (p1 == &proc0) {
685 VM_CNT_INC(v_kthreads);
686 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
687 p2->p_vmspace->vm_ssize);
689 VM_CNT_INC(v_rforks);
690 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
691 p2->p_vmspace->vm_ssize);
695 * Associate the process descriptor with the process before anything
696 * can happen that might cause that process to need the descriptor.
697 * However, don't do this until after fork(2) can no longer fail.
699 if (fr->fr_flags & RFPROCDESC)
700 procdesc_new(p2, fr->fr_pd_flags);
703 * Both processes are set up, now check if any loadable modules want
704 * to adjust anything.
706 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
709 * Set the child start time and mark the process as being complete.
713 microuptime(&p2->p_stats->p_start);
715 p2->p_state = PRS_NORMAL;
720 * Tell the DTrace fasttrap provider about the new process so that any
721 * tracepoints inherited from the parent can be removed. We have to do
722 * this only after p_state is PRS_NORMAL since the fasttrap module will
723 * use pfind() later on.
725 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
726 dtrace_fasttrap_fork(p1, p2);
729 * Hold the process so that it cannot exit after we make it runnable,
730 * but before we wait for the debugger.
733 if (p1->p_ptevents & PTRACE_FORK) {
735 * Arrange for debugger to receive the fork event.
737 * We can report PL_FLAG_FORKED regardless of
738 * P_FOLLOWFORK settings, but it does not make a sense
741 td->td_dbgflags |= TDB_FORK;
742 td->td_dbg_forked = p2->p_pid;
743 td2->td_dbgflags |= TDB_STOPATFORK;
745 if (fr->fr_flags & RFPPWAIT) {
746 td->td_pflags |= TDP_RFPPWAIT;
747 td->td_rfppwait_p = p2;
748 td->td_dbgflags |= TDB_VFORK;
753 * Now can be swapped.
759 * Tell any interested parties about the new process.
761 knote_fork(p1->p_klist, p2->p_pid);
762 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
764 if (fr->fr_flags & RFPROCDESC) {
765 procdesc_finit(p2->p_procdesc, fp_procdesc);
766 fdrop(fp_procdesc, td);
769 if ((fr->fr_flags & RFSTOPPED) == 0) {
771 * If RFSTOPPED not requested, make child runnable and
776 sched_add(td2, SRQ_BORING);
778 if (fr->fr_pidp != NULL)
779 *fr->fr_pidp = p2->p_pid;
786 * Wait until debugger is attached to child.
788 while (td2->td_proc == p2 && (td2->td_dbgflags & TDB_STOPATFORK) != 0)
789 cv_wait(&p2->p_dbgwait, &p2->p_mtx);
791 racct_proc_fork_done(p2);
796 fork1(struct thread *td, struct fork_req *fr)
798 struct proc *p1, *newproc;
801 struct file *fp_procdesc;
802 vm_ooffset_t mem_charged;
803 int error, nprocs_new, ok;
805 static struct timeval lastfail;
808 flags = fr->fr_flags;
809 pages = fr->fr_pages;
811 if ((flags & RFSTOPPED) != 0)
812 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
814 MPASS(fr->fr_procp == NULL);
816 /* Check for the undefined or unimplemented flags. */
817 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
820 /* Signal value requires RFTSIGZMB. */
821 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
824 /* Can't copy and clear. */
825 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
828 /* Check the validity of the signal number. */
829 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
832 if ((flags & RFPROCDESC) != 0) {
833 /* Can't not create a process yet get a process descriptor. */
834 if ((flags & RFPROC) == 0)
837 /* Must provide a place to put a procdesc if creating one. */
838 if (fr->fr_pd_fd == NULL)
841 /* Check if we are using supported flags. */
842 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
849 * Here we don't create a new process, but we divorce
850 * certain parts of a process from itself.
852 if ((flags & RFPROC) == 0) {
853 if (fr->fr_procp != NULL)
854 *fr->fr_procp = NULL;
855 else if (fr->fr_pidp != NULL)
857 return (fork_norfproc(td, flags));
865 * Increment the nprocs resource before allocations occur.
866 * Although process entries are dynamically created, we still
867 * keep a global limit on the maximum number we will
868 * create. There are hard-limits as to the number of processes
869 * that can run, established by the KVA and memory usage for
872 * Don't allow a nonprivileged user to use the last ten
873 * processes; don't let root exceed the limit.
875 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
876 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred,
877 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) {
879 sx_xlock(&allproc_lock);
880 if (ppsratecheck(&lastfail, &curfail, 1)) {
881 printf("maxproc limit exceeded by uid %u (pid %d); "
882 "see tuning(7) and login.conf(5)\n",
883 td->td_ucred->cr_ruid, p1->p_pid);
885 sx_xunlock(&allproc_lock);
890 * If required, create a process descriptor in the parent first; we
891 * will abandon it if something goes wrong. We don't finit() until
894 if (flags & RFPROCDESC) {
895 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
896 fr->fr_pd_flags, fr->fr_pd_fcaps);
903 pages = kstack_pages;
904 /* Allocate new proc. */
905 newproc = uma_zalloc(proc_zone, M_WAITOK);
906 td2 = FIRST_THREAD_IN_PROC(newproc);
908 td2 = thread_alloc(pages);
913 proc_linkup(newproc, td2);
915 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
916 if (td2->td_kstack != 0)
917 vm_thread_dispose(td2);
918 if (!thread_alloc_stack(td2, pages)) {
925 if ((flags & RFMEM) == 0) {
926 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
931 if (!swap_reserve(mem_charged)) {
933 * The swap reservation failed. The accounting
934 * from the entries of the copied vm2 will be
935 * subtracted in vmspace_free(), so force the
938 swap_reserve_force(mem_charged);
946 * XXX: This is ugly; when we copy resource usage, we need to bump
947 * per-cred resource counters.
949 proc_set_cred_init(newproc, crhold(td->td_ucred));
952 * Initialize resource accounting for the child process.
954 error = racct_proc_fork(p1, newproc);
961 mac_proc_init(newproc);
963 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
964 STAILQ_INIT(&newproc->p_ktr);
966 /* We have to lock the process tree while we look for a pid. */
967 sx_slock(&proctree_lock);
968 sx_xlock(&allproc_lock);
971 * Increment the count of procs running with this uid. Don't allow
972 * a nonprivileged user to exceed their current limit.
974 * XXXRW: Can we avoid privilege here if it's not needed?
976 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
978 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
980 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
981 lim_cur(td, RLIMIT_NPROC));
984 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
989 sx_sunlock(&proctree_lock);
990 sx_xunlock(&allproc_lock);
992 mac_proc_destroy(newproc);
994 racct_proc_exit(newproc);
996 crfree(newproc->p_ucred);
997 newproc->p_ucred = NULL;
1001 uma_zfree(proc_zone, newproc);
1002 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1003 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1004 fdrop(fp_procdesc, td);
1006 atomic_add_int(&nprocs, -1);
1007 pause("fork", hz / 2);
1012 * Handle the return of a child process from fork1(). This function
1013 * is called from the MD fork_trampoline() entry point.
1016 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1017 struct trapframe *frame)
1025 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1027 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1028 td, td_get_sched(td), p->p_pid, td->td_name);
1030 sched_fork_exit(td);
1032 * Processes normally resume in mi_switch() after being
1033 * cpu_switch()'ed to, but when children start up they arrive here
1034 * instead, so we must do much the same things as mi_switch() would.
1036 if ((dtd = PCPU_GET(deadthread))) {
1037 PCPU_SET(deadthread, NULL);
1043 * cpu_fork_kthread_handler intercepts this function call to
1044 * have this call a non-return function to stay in kernel mode.
1045 * initproc has its own fork handler, but it does return.
1047 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1048 callout(arg, frame);
1051 * Check if a kernel thread misbehaved and returned from its main
1054 if (p->p_flag & P_KPROC) {
1055 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1056 td->td_name, p->p_pid);
1059 mtx_assert(&Giant, MA_NOTOWNED);
1061 if (p->p_sysent->sv_schedtail != NULL)
1062 (p->p_sysent->sv_schedtail)(td);
1063 td->td_pflags &= ~TDP_FORKING;
1067 * Simplified back end of syscall(), used when returning from fork()
1068 * directly into user mode. This function is passed in to fork_exit()
1069 * as the first parameter and is called when returning to a new
1073 fork_return(struct thread *td, struct trapframe *frame)
1075 struct proc *p, *dbg;
1078 if (td->td_dbgflags & TDB_STOPATFORK) {
1079 sx_xlock(&proctree_lock);
1081 if (p->p_pptr->p_ptevents & PTRACE_FORK) {
1083 * If debugger still wants auto-attach for the
1084 * parent's children, do it now.
1086 dbg = p->p_pptr->p_pptr;
1087 proc_set_traced(p, true);
1089 "fork_return: attaching to new child pid %d: oppid %d",
1090 p->p_pid, p->p_oppid);
1091 proc_reparent(p, dbg);
1092 sx_xunlock(&proctree_lock);
1093 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1094 ptracestop(td, SIGSTOP, NULL);
1095 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1098 * ... otherwise clear the request.
1100 sx_xunlock(&proctree_lock);
1101 td->td_dbgflags &= ~TDB_STOPATFORK;
1102 cv_broadcast(&p->p_dbgwait);
1105 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1107 * This is the start of a new thread in a traced
1108 * process. Report a system call exit event.
1111 td->td_dbgflags |= TDB_SCX;
1112 _STOPEVENT(p, S_SCX, td->td_sa.code);
1113 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1114 (td->td_dbgflags & TDB_BORN) != 0)
1115 ptracestop(td, SIGTRAP, NULL);
1116 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1123 if (KTRPOINT(td, KTR_SYSRET))
1124 ktrsysret(SYS_fork, 0, 0);