2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_kdtrace.h"
41 #include "opt_ktrace.h"
42 #include "opt_kstack_pages.h"
43 #include "opt_procdesc.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/racct.h>
63 #include <sys/resourcevar.h>
64 #include <sys/sched.h>
65 #include <sys/syscall.h>
66 #include <sys/vmmeter.h>
67 #include <sys/vnode.h>
70 #include <sys/ktrace.h>
71 #include <sys/unistd.h>
74 #include <sys/sysent.h>
75 #include <sys/signalvar.h>
77 #include <security/audit/audit.h>
78 #include <security/mac/mac_framework.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_extern.h>
87 #include <sys/dtrace_bsd.h>
88 dtrace_fork_func_t dtrace_fasttrap_fork;
91 SDT_PROVIDER_DECLARE(proc);
92 SDT_PROBE_DEFINE(proc, kernel, , create, create);
93 SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *");
94 SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *");
95 SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int");
97 #ifndef _SYS_SYSPROTO_H_
105 sys_fork(struct thread *td, struct fork_args *uap)
110 error = fork1(td, RFFDG | RFPROC, 0, &p2, NULL, 0);
112 td->td_retval[0] = p2->p_pid;
113 td->td_retval[1] = 0;
122 struct pdfork_args *uap;
129 * It is necessary to return fd by reference because 0 is a valid file
130 * descriptor number, and the child needs to be able to distinguish
131 * itself from the parent using the return value.
133 error = fork1(td, RFFDG | RFPROC | RFPROCDESC, 0, &p2,
136 td->td_retval[0] = p2->p_pid;
137 td->td_retval[1] = 0;
138 error = copyout(&fd, uap->fdp, sizeof(fd));
148 sys_vfork(struct thread *td, struct vfork_args *uap)
153 flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
154 error = fork1(td, flags, 0, &p2, NULL, 0);
156 td->td_retval[0] = p2->p_pid;
157 td->td_retval[1] = 0;
163 sys_rfork(struct thread *td, struct rfork_args *uap)
168 /* Don't allow kernel-only flags. */
169 if ((uap->flags & RFKERNELONLY) != 0)
172 AUDIT_ARG_FFLAGS(uap->flags);
173 error = fork1(td, uap->flags, 0, &p2, NULL, 0);
175 td->td_retval[0] = p2 ? p2->p_pid : 0;
176 td->td_retval[1] = 0;
181 int nprocs = 1; /* process 0 */
183 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
187 * Random component to lastpid generation. We mix in a random factor to make
188 * it a little harder to predict. We sanity check the modulus value to avoid
189 * doing it in critical paths. Don't let it be too small or we pointlessly
190 * waste randomness entropy, and don't let it be impossibly large. Using a
191 * modulus that is too big causes a LOT more process table scans and slows
192 * down fork processing as the pidchecked caching is defeated.
194 static int randompid = 0;
197 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
201 error = sysctl_wire_old_buffer(req, sizeof(int));
204 sx_xlock(&allproc_lock);
206 error = sysctl_handle_int(oidp, &pid, 0, req);
207 if (error == 0 && req->newptr != NULL) {
208 if (pid < 0 || pid > pid_max - 100) /* out of range */
210 else if (pid < 2) /* NOP */
212 else if (pid < 100) /* Make it reasonable */
216 sx_xunlock(&allproc_lock);
220 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
221 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
224 fork_findpid(int flags)
228 static int pidchecked = 0;
231 * Requires allproc_lock in order to iterate over the list
232 * of processes, and proctree_lock to access p_pgrp.
234 sx_assert(&allproc_lock, SX_LOCKED);
235 sx_assert(&proctree_lock, SX_LOCKED);
238 * Find an unused process ID. We remember a range of unused IDs
239 * ready to use (from lastpid+1 through pidchecked-1).
241 * If RFHIGHPID is set (used during system boot), do not allocate
244 trypid = lastpid + 1;
245 if (flags & RFHIGHPID) {
250 trypid += arc4random() % randompid;
254 * If the process ID prototype has wrapped around,
255 * restart somewhat above 0, as the low-numbered procs
256 * tend to include daemons that don't exit.
258 if (trypid >= pid_max) {
259 trypid = trypid % pid_max;
264 if (trypid >= pidchecked) {
267 pidchecked = PID_MAX;
269 * Scan the active and zombie procs to check whether this pid
270 * is in use. Remember the lowest pid that's greater
271 * than trypid, so we can avoid checking for a while.
273 p = LIST_FIRST(&allproc);
275 for (; p != NULL; p = LIST_NEXT(p, p_list)) {
276 while (p->p_pid == trypid ||
277 (p->p_pgrp != NULL &&
278 (p->p_pgrp->pg_id == trypid ||
279 (p->p_session != NULL &&
280 p->p_session->s_sid == trypid)))) {
282 if (trypid >= pidchecked)
285 if (p->p_pid > trypid && pidchecked > p->p_pid)
286 pidchecked = p->p_pid;
287 if (p->p_pgrp != NULL) {
288 if (p->p_pgrp->pg_id > trypid &&
289 pidchecked > p->p_pgrp->pg_id)
290 pidchecked = p->p_pgrp->pg_id;
291 if (p->p_session != NULL &&
292 p->p_session->s_sid > trypid &&
293 pidchecked > p->p_session->s_sid)
294 pidchecked = p->p_session->s_sid;
299 p = LIST_FIRST(&zombproc);
305 * RFHIGHPID does not mess with the lastpid counter during boot.
307 if (flags & RFHIGHPID)
316 fork_norfproc(struct thread *td, int flags)
321 KASSERT((flags & RFPROC) == 0,
322 ("fork_norfproc called with RFPROC set"));
325 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
326 (flags & (RFCFDG | RFFDG))) {
328 if (thread_single(SINGLE_BOUNDARY)) {
335 error = vm_forkproc(td, NULL, NULL, NULL, flags);
340 * Close all file descriptors.
342 if (flags & RFCFDG) {
343 struct filedesc *fdtmp;
344 fdtmp = fdinit(td->td_proc->p_fd);
350 * Unshare file descriptors (from parent).
356 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
357 (flags & (RFCFDG | RFFDG))) {
366 do_fork(struct thread *td, int flags, struct proc *p2, struct thread *td2,
367 struct vmspace *vm2, int pdflags)
369 struct proc *p1, *pptr;
372 struct filedesc_to_leader *fdtol;
373 struct sigacts *newsigacts;
375 sx_assert(&proctree_lock, SX_SLOCKED);
376 sx_assert(&allproc_lock, SX_XLOCKED);
382 * Increment the nprocs resource before blocking can occur. There
383 * are hard-limits as to the number of processes that can run.
387 trypid = fork_findpid(flags);
389 sx_sunlock(&proctree_lock);
391 p2->p_state = PRS_NEW; /* protect against others */
393 AUDIT_ARG_PID(p2->p_pid);
394 LIST_INSERT_HEAD(&allproc, p2, p_list);
395 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
400 sx_xunlock(&allproc_lock);
402 bcopy(&p1->p_startcopy, &p2->p_startcopy,
403 __rangeof(struct proc, p_startcopy, p_endcopy));
404 pargs_hold(p2->p_args);
407 bzero(&p2->p_startzero,
408 __rangeof(struct proc, p_startzero, p_endzero));
410 p2->p_ucred = crhold(td->td_ucred);
412 /* Tell the prison that we exist. */
413 prison_proc_hold(p2->p_ucred->cr_prison);
418 * Malloc things while we don't hold any locks.
420 if (flags & RFSIGSHARE)
423 newsigacts = sigacts_alloc();
428 if (flags & RFCFDG) {
429 fd = fdinit(p1->p_fd);
431 } else if (flags & RFFDG) {
432 fd = fdcopy(p1->p_fd);
435 fd = fdshare(p1->p_fd);
436 if (p1->p_fdtol == NULL)
437 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
439 if ((flags & RFTHREAD) != 0) {
441 * Shared file descriptor table, and shared
445 FILEDESC_XLOCK(p1->p_fd);
446 fdtol->fdl_refcount++;
447 FILEDESC_XUNLOCK(p1->p_fd);
450 * Shared file descriptor table, and different
453 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
458 * Make a proc table entry for the new process.
459 * Start by zeroing the section of proc that is zero-initialized,
460 * then copy the section that is copied directly from the parent.
466 bzero(&td2->td_startzero,
467 __rangeof(struct thread, td_startzero, td_endzero));
469 bcopy(&td->td_startcopy, &td2->td_startcopy,
470 __rangeof(struct thread, td_startcopy, td_endcopy));
472 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
473 td2->td_sigstk = td->td_sigstk;
474 td2->td_flags = TDF_INMEM;
475 td2->td_lend_user_pri = PRI_MAX;
479 td2->td_vnet_lpush = NULL;
483 * Allow the scheduler to initialize the child.
490 * Duplicate sub-structures as needed.
491 * Increase reference counts on shared objects.
493 p2->p_flag = P_INMEM;
494 p2->p_swtick = ticks;
495 if (p1->p_flag & P_PROFIL)
497 td2->td_ucred = crhold(p2->p_ucred);
499 if (flags & RFSIGSHARE) {
500 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
502 sigacts_copy(newsigacts, p1->p_sigacts);
503 p2->p_sigacts = newsigacts;
506 if (flags & RFTSIGZMB)
507 p2->p_sigparent = RFTSIGNUM(flags);
508 else if (flags & RFLINUXTHPN)
509 p2->p_sigparent = SIGUSR1;
511 p2->p_sigparent = SIGCHLD;
513 p2->p_textvp = p1->p_textvp;
518 * p_limit is copy-on-write. Bump its refcount.
522 pstats_fork(p1->p_stats, p2->p_stats);
527 /* Bump references to the text vnode (for procfs). */
532 * Set up linkage for kernel based threading.
534 if ((flags & RFTHREAD) != 0) {
535 mtx_lock(&ppeers_lock);
536 p2->p_peers = p1->p_peers;
538 p2->p_leader = p1->p_leader;
539 mtx_unlock(&ppeers_lock);
540 PROC_LOCK(p1->p_leader);
541 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
542 PROC_UNLOCK(p1->p_leader);
544 * The task leader is exiting, so process p1 is
545 * going to be killed shortly. Since p1 obviously
546 * isn't dead yet, we know that the leader is either
547 * sending SIGKILL's to all the processes in this
548 * task or is sleeping waiting for all the peers to
549 * exit. We let p1 complete the fork, but we need
550 * to go ahead and kill the new process p2 since
551 * the task leader may not get a chance to send
552 * SIGKILL to it. We leave it on the list so that
553 * the task leader will wait for this new process
557 kern_psignal(p2, SIGKILL);
560 PROC_UNLOCK(p1->p_leader);
566 sx_xlock(&proctree_lock);
567 PGRP_LOCK(p1->p_pgrp);
572 * Preserve some more flags in subprocess. P_PROFIL has already
575 p2->p_flag |= p1->p_flag & P_SUGID;
576 td2->td_pflags |= td->td_pflags & TDP_ALTSTACK;
577 SESS_LOCK(p1->p_session);
578 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
579 p2->p_flag |= P_CONTROLT;
580 SESS_UNLOCK(p1->p_session);
581 if (flags & RFPPWAIT)
582 p2->p_flag |= P_PPWAIT;
584 p2->p_pgrp = p1->p_pgrp;
585 LIST_INSERT_AFTER(p1, p2, p_pglist);
586 PGRP_UNLOCK(p1->p_pgrp);
587 LIST_INIT(&p2->p_children);
588 LIST_INIT(&p2->p_orphans);
590 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
593 * If PF_FORK is set, the child process inherits the
594 * procfs ioctl flags from its parent.
596 if (p1->p_pfsflags & PF_FORK) {
597 p2->p_stops = p1->p_stops;
598 p2->p_pfsflags = p1->p_pfsflags;
602 * This begins the section where we must prevent the parent
603 * from being swapped.
609 * Attach the new process to its parent.
611 * If RFNOWAIT is set, the newly created process becomes a child
612 * of init. This effectively disassociates the child from the
615 if (flags & RFNOWAIT)
620 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
621 sx_xunlock(&proctree_lock);
623 /* Inform accounting that we have forked. */
624 p2->p_acflag = AFORK;
632 * Finish creating the child process. It will return via a different
633 * execution path later. (ie: directly into user mode)
635 vm_forkproc(td, p2, td2, vm2, flags);
637 if (flags == (RFFDG | RFPROC)) {
638 PCPU_INC(cnt.v_forks);
639 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
640 p2->p_vmspace->vm_ssize);
641 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
642 PCPU_INC(cnt.v_vforks);
643 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
644 p2->p_vmspace->vm_ssize);
645 } else if (p1 == &proc0) {
646 PCPU_INC(cnt.v_kthreads);
647 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
648 p2->p_vmspace->vm_ssize);
650 PCPU_INC(cnt.v_rforks);
651 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
652 p2->p_vmspace->vm_ssize);
657 * Associate the process descriptor with the process before anything
658 * can happen that might cause that process to need the descriptor.
659 * However, don't do this until after fork(2) can no longer fail.
661 if (flags & RFPROCDESC)
662 procdesc_new(p2, pdflags);
666 * Both processes are set up, now check if any loadable modules want
667 * to adjust anything.
669 EVENTHANDLER_INVOKE(process_fork, p1, p2, flags);
672 * Set the child start time and mark the process as being complete.
676 microuptime(&p2->p_stats->p_start);
678 p2->p_state = PRS_NORMAL;
683 * Tell the DTrace fasttrap provider about the new process
684 * if it has registered an interest. We have to do this only after
685 * p_state is PRS_NORMAL since the fasttrap module will use pfind()
688 if (dtrace_fasttrap_fork)
689 dtrace_fasttrap_fork(p1, p2);
691 if ((p1->p_flag & (P_TRACED | P_FOLLOWFORK)) == (P_TRACED |
694 * Arrange for debugger to receive the fork event.
696 * We can report PL_FLAG_FORKED regardless of
697 * P_FOLLOWFORK settings, but it does not make a sense
700 td->td_dbgflags |= TDB_FORK;
701 td->td_dbg_forked = p2->p_pid;
702 td2->td_dbgflags |= TDB_STOPATFORK;
706 if (flags & RFPPWAIT) {
707 td->td_pflags |= TDP_RFPPWAIT;
708 td->td_rfppwait_p = p2;
711 if ((flags & RFSTOPPED) == 0) {
713 * If RFSTOPPED not requested, make child runnable and
718 sched_add(td2, SRQ_BORING);
723 * Now can be swapped.
729 * Tell any interested parties about the new process.
731 knote_fork(&p1->p_klist, p2->p_pid);
732 SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0);
735 * Wait until debugger is attached to child.
738 while ((td2->td_dbgflags & TDB_STOPATFORK) != 0)
739 cv_wait(&p2->p_dbgwait, &p2->p_mtx);
746 fork1(struct thread *td, int flags, int pages, struct proc **procp,
747 int *procdescp, int pdflags)
750 struct proc *newproc;
754 vm_ooffset_t mem_charged;
757 static struct timeval lastfail;
759 struct file *fp_procdesc = NULL;
762 /* Check for the undefined or unimplemented flags. */
763 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
766 /* Signal value requires RFTSIGZMB. */
767 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
770 /* Can't copy and clear. */
771 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
774 /* Check the validity of the signal number. */
775 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
779 if ((flags & RFPROCDESC) != 0) {
780 /* Can't not create a process yet get a process descriptor. */
781 if ((flags & RFPROC) == 0)
784 /* Must provide a place to put a procdesc if creating one. */
785 if (procdescp == NULL)
793 * Here we don't create a new process, but we divorce
794 * certain parts of a process from itself.
796 if ((flags & RFPROC) == 0) {
798 return (fork_norfproc(td, flags));
803 * If required, create a process descriptor in the parent first; we
804 * will abandon it if something goes wrong. We don't finit() until
807 if (flags & RFPROCDESC) {
808 error = falloc(td, &fp_procdesc, procdescp, 0);
817 pages = KSTACK_PAGES;
818 /* Allocate new proc. */
819 newproc = uma_zalloc(proc_zone, M_WAITOK);
820 td2 = FIRST_THREAD_IN_PROC(newproc);
822 td2 = thread_alloc(pages);
827 proc_linkup(newproc, td2);
829 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
830 if (td2->td_kstack != 0)
831 vm_thread_dispose(td2);
832 if (!thread_alloc_stack(td2, pages)) {
839 if ((flags & RFMEM) == 0) {
840 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
845 if (!swap_reserve(mem_charged)) {
847 * The swap reservation failed. The accounting
848 * from the entries of the copied vm2 will be
849 * substracted in vmspace_free(), so force the
852 swap_reserve_force(mem_charged);
860 * XXX: This is ugly; when we copy resource usage, we need to bump
861 * per-cred resource counters.
863 newproc->p_ucred = p1->p_ucred;
866 * Initialize resource accounting for the child process.
868 error = racct_proc_fork(p1, newproc);
875 mac_proc_init(newproc);
877 knlist_init_mtx(&newproc->p_klist, &newproc->p_mtx);
878 STAILQ_INIT(&newproc->p_ktr);
880 /* We have to lock the process tree while we look for a pid. */
881 sx_slock(&proctree_lock);
884 * Although process entries are dynamically created, we still keep
885 * a global limit on the maximum number we will create. Don't allow
886 * a nonprivileged user to use the last ten processes; don't let root
887 * exceed the limit. The variable nprocs is the current number of
888 * processes, maxproc is the limit.
890 sx_xlock(&allproc_lock);
891 if ((nprocs >= maxproc - 10 && priv_check_cred(td->td_ucred,
892 PRIV_MAXPROC, 0) != 0) || nprocs >= maxproc) {
898 * Increment the count of procs running with this uid. Don't allow
899 * a nonprivileged user to exceed their current limit.
901 * XXXRW: Can we avoid privilege here if it's not needed?
903 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
905 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
908 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
909 lim_cur(p1, RLIMIT_NPROC));
913 do_fork(td, flags, newproc, td2, vm2, pdflags);
916 * Return child proc pointer to parent.
920 if (flags & RFPROCDESC) {
921 procdesc_finit(newproc->p_procdesc, fp_procdesc);
922 fdrop(fp_procdesc, td);
925 racct_proc_fork_done(newproc);
931 sx_sunlock(&proctree_lock);
932 if (ppsratecheck(&lastfail, &curfail, 1))
933 printf("maxproc limit exceeded by uid %i, please see tuning(7) and login.conf(5).\n",
934 td->td_ucred->cr_ruid);
935 sx_xunlock(&allproc_lock);
937 mac_proc_destroy(newproc);
939 racct_proc_exit(newproc);
943 uma_zfree(proc_zone, newproc);
945 if (((flags & RFPROCDESC) != 0) && (fp_procdesc != NULL)) {
946 fdclose(td->td_proc->p_fd, fp_procdesc, *procdescp, td);
947 fdrop(fp_procdesc, td);
950 pause("fork", hz / 2);
955 * Handle the return of a child process from fork1(). This function
956 * is called from the MD fork_trampoline() entry point.
959 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
960 struct trapframe *frame)
968 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
970 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
971 td, td->td_sched, p->p_pid, td->td_name);
975 * Processes normally resume in mi_switch() after being
976 * cpu_switch()'ed to, but when children start up they arrive here
977 * instead, so we must do much the same things as mi_switch() would.
979 if ((dtd = PCPU_GET(deadthread))) {
980 PCPU_SET(deadthread, NULL);
986 * cpu_set_fork_handler intercepts this function call to
987 * have this call a non-return function to stay in kernel mode.
988 * initproc has its own fork handler, but it does return.
990 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
994 * Check if a kernel thread misbehaved and returned from its main
997 if (p->p_flag & P_KTHREAD) {
998 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
999 td->td_name, p->p_pid);
1002 mtx_assert(&Giant, MA_NOTOWNED);
1004 if (p->p_sysent->sv_schedtail != NULL)
1005 (p->p_sysent->sv_schedtail)(td);
1009 * Simplified back end of syscall(), used when returning from fork()
1010 * directly into user mode. Giant is not held on entry, and must not
1011 * be held on return. This function is passed in to fork_exit() as the
1012 * first parameter and is called when returning to a new userland process.
1015 fork_return(struct thread *td, struct trapframe *frame)
1017 struct proc *p, *dbg;
1019 if (td->td_dbgflags & TDB_STOPATFORK) {
1021 sx_xlock(&proctree_lock);
1023 if ((p->p_pptr->p_flag & (P_TRACED | P_FOLLOWFORK)) ==
1024 (P_TRACED | P_FOLLOWFORK)) {
1026 * If debugger still wants auto-attach for the
1027 * parent's children, do it now.
1029 dbg = p->p_pptr->p_pptr;
1030 p->p_flag |= P_TRACED;
1031 p->p_oppid = p->p_pptr->p_pid;
1032 proc_reparent(p, dbg);
1033 sx_xunlock(&proctree_lock);
1034 td->td_dbgflags |= TDB_CHILD;
1035 ptracestop(td, SIGSTOP);
1036 td->td_dbgflags &= ~TDB_CHILD;
1039 * ... otherwise clear the request.
1041 sx_xunlock(&proctree_lock);
1042 td->td_dbgflags &= ~TDB_STOPATFORK;
1043 cv_broadcast(&p->p_dbgwait);
1051 if (KTRPOINT(td, KTR_SYSRET))
1052 ktrsysret(SYS_fork, 0, 0);