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_
102 sys_fork(struct thread *td, struct fork_args *uap)
107 bzero(&fr, sizeof(fr));
108 fr.fr_flags = RFFDG | RFPROC;
110 error = fork1(td, &fr);
112 td->td_retval[0] = pid;
113 td->td_retval[1] = 0;
120 sys_pdfork(struct thread *td, struct pdfork_args *uap)
125 bzero(&fr, sizeof(fr));
126 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
129 fr.fr_pd_flags = uap->flags;
131 * It is necessary to return fd by reference because 0 is a valid file
132 * descriptor number, and the child needs to be able to distinguish
133 * itself from the parent using the return value.
135 error = fork1(td, &fr);
137 td->td_retval[0] = pid;
138 td->td_retval[1] = 0;
139 error = copyout(&fd, uap->fdp, sizeof(fd));
146 sys_vfork(struct thread *td, struct vfork_args *uap)
151 bzero(&fr, sizeof(fr));
152 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
154 error = fork1(td, &fr);
156 td->td_retval[0] = 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 bzero(&fr, sizeof(fr));
174 fr.fr_flags = uap->flags;
176 error = fork1(td, &fr);
178 td->td_retval[0] = pid;
179 td->td_retval[1] = 0;
184 int nprocs = 1; /* process 0 */
186 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
190 * Random component to lastpid generation. We mix in a random factor to make
191 * it a little harder to predict. We sanity check the modulus value to avoid
192 * doing it in critical paths. Don't let it be too small or we pointlessly
193 * waste randomness entropy, and don't let it be impossibly large. Using a
194 * modulus that is too big causes a LOT more process table scans and slows
195 * down fork processing as the pidchecked caching is defeated.
197 static int randompid = 0;
200 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
204 error = sysctl_wire_old_buffer(req, sizeof(int));
207 sx_xlock(&allproc_lock);
209 error = sysctl_handle_int(oidp, &pid, 0, req);
210 if (error == 0 && req->newptr != NULL) {
214 /* generate a random PID modulus between 100 and 1123 */
215 randompid = 100 + arc4random() % 1024;
216 else if (pid < 0 || pid > pid_max - 100)
218 randompid = pid_max - 100;
220 /* Make it reasonable */
225 sx_xunlock(&allproc_lock);
229 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
230 0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value");
233 fork_findpid(int flags)
237 static int pidchecked = 0;
240 * Requires allproc_lock in order to iterate over the list
241 * of processes, and proctree_lock to access p_pgrp.
243 sx_assert(&allproc_lock, SX_LOCKED);
244 sx_assert(&proctree_lock, SX_LOCKED);
247 * Find an unused process ID. We remember a range of unused IDs
248 * ready to use (from lastpid+1 through pidchecked-1).
250 * If RFHIGHPID is set (used during system boot), do not allocate
253 trypid = lastpid + 1;
254 if (flags & RFHIGHPID) {
259 trypid += arc4random() % randompid;
263 * If the process ID prototype has wrapped around,
264 * restart somewhat above 0, as the low-numbered procs
265 * tend to include daemons that don't exit.
267 if (trypid >= pid_max) {
268 trypid = trypid % pid_max;
273 if (trypid >= pidchecked) {
276 pidchecked = PID_MAX;
278 * Scan the active and zombie procs to check whether this pid
279 * is in use. Remember the lowest pid that's greater
280 * than trypid, so we can avoid checking for a while.
282 * Avoid reuse of the process group id, session id or
283 * the reaper subtree id. Note that for process group
284 * and sessions, the amount of reserved pids is
285 * limited by process limit. For the subtree ids, the
286 * id is kept reserved only while there is a
287 * non-reaped process in the subtree, so amount of
288 * reserved pids is limited by process limit times
291 p = LIST_FIRST(&allproc);
293 for (; p != NULL; p = LIST_NEXT(p, p_list)) {
294 while (p->p_pid == trypid ||
295 p->p_reapsubtree == trypid ||
296 (p->p_pgrp != NULL &&
297 (p->p_pgrp->pg_id == trypid ||
298 (p->p_session != NULL &&
299 p->p_session->s_sid == trypid)))) {
301 if (trypid >= pidchecked)
304 if (p->p_pid > trypid && pidchecked > p->p_pid)
305 pidchecked = p->p_pid;
306 if (p->p_pgrp != NULL) {
307 if (p->p_pgrp->pg_id > trypid &&
308 pidchecked > p->p_pgrp->pg_id)
309 pidchecked = p->p_pgrp->pg_id;
310 if (p->p_session != NULL &&
311 p->p_session->s_sid > trypid &&
312 pidchecked > p->p_session->s_sid)
313 pidchecked = p->p_session->s_sid;
318 p = LIST_FIRST(&zombproc);
324 * RFHIGHPID does not mess with the lastpid counter during boot.
326 if (flags & RFHIGHPID)
335 fork_norfproc(struct thread *td, int flags)
340 KASSERT((flags & RFPROC) == 0,
341 ("fork_norfproc called with RFPROC set"));
344 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
345 (flags & (RFCFDG | RFFDG))) {
347 if (thread_single(p1, SINGLE_BOUNDARY)) {
354 error = vm_forkproc(td, NULL, NULL, NULL, flags);
359 * Close all file descriptors.
361 if (flags & RFCFDG) {
362 struct filedesc *fdtmp;
363 fdtmp = fdinit(td->td_proc->p_fd, false);
369 * Unshare file descriptors (from parent).
375 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
376 (flags & (RFCFDG | RFFDG))) {
378 thread_single_end(p1, SINGLE_BOUNDARY);
385 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
386 struct vmspace *vm2, struct file *fp_procdesc)
388 struct proc *p1, *pptr;
391 struct filedesc_to_leader *fdtol;
392 struct sigacts *newsigacts;
394 sx_assert(&proctree_lock, SX_SLOCKED);
395 sx_assert(&allproc_lock, SX_XLOCKED);
399 trypid = fork_findpid(fr->fr_flags);
401 sx_sunlock(&proctree_lock);
403 p2->p_state = PRS_NEW; /* protect against others */
405 AUDIT_ARG_PID(p2->p_pid);
406 LIST_INSERT_HEAD(&allproc, p2, p_list);
408 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
413 sx_xunlock(&allproc_lock);
415 bcopy(&p1->p_startcopy, &p2->p_startcopy,
416 __rangeof(struct proc, p_startcopy, p_endcopy));
417 pargs_hold(p2->p_args);
421 bzero(&p2->p_startzero,
422 __rangeof(struct proc, p_startzero, p_endzero));
424 /* Tell the prison that we exist. */
425 prison_proc_hold(p2->p_ucred->cr_prison);
430 * Malloc things while we don't hold any locks.
432 if (fr->fr_flags & RFSIGSHARE)
435 newsigacts = sigacts_alloc();
440 if (fr->fr_flags & RFCFDG) {
441 fd = fdinit(p1->p_fd, false);
443 } else if (fr->fr_flags & RFFDG) {
444 fd = fdcopy(p1->p_fd);
447 fd = fdshare(p1->p_fd);
448 if (p1->p_fdtol == NULL)
449 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
451 if ((fr->fr_flags & RFTHREAD) != 0) {
453 * Shared file descriptor table, and shared
457 FILEDESC_XLOCK(p1->p_fd);
458 fdtol->fdl_refcount++;
459 FILEDESC_XUNLOCK(p1->p_fd);
462 * Shared file descriptor table, and different
465 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
470 * Make a proc table entry for the new process.
471 * Start by zeroing the section of proc that is zero-initialized,
472 * then copy the section that is copied directly from the parent.
478 bzero(&td2->td_startzero,
479 __rangeof(struct thread, td_startzero, td_endzero));
481 bcopy(&td->td_startcopy, &td2->td_startcopy,
482 __rangeof(struct thread, td_startcopy, td_endcopy));
484 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
485 td2->td_sigstk = td->td_sigstk;
486 td2->td_flags = TDF_INMEM;
487 td2->td_lend_user_pri = PRI_MAX;
491 td2->td_vnet_lpush = NULL;
495 * Allow the scheduler to initialize the child.
502 * Duplicate sub-structures as needed.
503 * Increase reference counts on shared objects.
505 p2->p_flag = P_INMEM;
506 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC | P2_TRAPCAP);
507 p2->p_swtick = ticks;
508 if (p1->p_flag & P_PROFIL)
512 * Whilst the proc lock is held, copy the VM domain data out
513 * using the VM domain method.
515 vm_domain_policy_init(&p2->p_vm_dom_policy);
516 vm_domain_policy_localcopy(&p2->p_vm_dom_policy,
517 &p1->p_vm_dom_policy);
519 if (fr->fr_flags & RFSIGSHARE) {
520 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
522 sigacts_copy(newsigacts, p1->p_sigacts);
523 p2->p_sigacts = newsigacts;
526 if (fr->fr_flags & RFTSIGZMB)
527 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
528 else if (fr->fr_flags & RFLINUXTHPN)
529 p2->p_sigparent = SIGUSR1;
531 p2->p_sigparent = SIGCHLD;
533 p2->p_textvp = p1->p_textvp;
537 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
538 p2->p_flag |= P_PROTECTED;
539 p2->p_flag2 |= P2_INHERIT_PROTECTED;
543 * p_limit is copy-on-write. Bump its refcount.
547 thread_cow_get_proc(td2, p2);
549 pstats_fork(p1->p_stats, p2->p_stats);
554 /* Bump references to the text vnode (for procfs). */
556 vrefact(p2->p_textvp);
559 * Set up linkage for kernel based threading.
561 if ((fr->fr_flags & RFTHREAD) != 0) {
562 mtx_lock(&ppeers_lock);
563 p2->p_peers = p1->p_peers;
565 p2->p_leader = p1->p_leader;
566 mtx_unlock(&ppeers_lock);
567 PROC_LOCK(p1->p_leader);
568 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
569 PROC_UNLOCK(p1->p_leader);
571 * The task leader is exiting, so process p1 is
572 * going to be killed shortly. Since p1 obviously
573 * isn't dead yet, we know that the leader is either
574 * sending SIGKILL's to all the processes in this
575 * task or is sleeping waiting for all the peers to
576 * exit. We let p1 complete the fork, but we need
577 * to go ahead and kill the new process p2 since
578 * the task leader may not get a chance to send
579 * SIGKILL to it. We leave it on the list so that
580 * the task leader will wait for this new process
584 kern_psignal(p2, SIGKILL);
587 PROC_UNLOCK(p1->p_leader);
593 sx_xlock(&proctree_lock);
594 PGRP_LOCK(p1->p_pgrp);
599 * Preserve some more flags in subprocess. P_PROFIL has already
602 p2->p_flag |= p1->p_flag & P_SUGID;
603 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING;
604 SESS_LOCK(p1->p_session);
605 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
606 p2->p_flag |= P_CONTROLT;
607 SESS_UNLOCK(p1->p_session);
608 if (fr->fr_flags & RFPPWAIT)
609 p2->p_flag |= P_PPWAIT;
611 p2->p_pgrp = p1->p_pgrp;
612 LIST_INSERT_AFTER(p1, p2, p_pglist);
613 PGRP_UNLOCK(p1->p_pgrp);
614 LIST_INIT(&p2->p_children);
615 LIST_INIT(&p2->p_orphans);
617 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
620 * If PF_FORK is set, the child process inherits the
621 * procfs ioctl flags from its parent.
623 if (p1->p_pfsflags & PF_FORK) {
624 p2->p_stops = p1->p_stops;
625 p2->p_pfsflags = p1->p_pfsflags;
629 * This begins the section where we must prevent the parent
630 * from being swapped.
636 * Attach the new process to its parent.
638 * If RFNOWAIT is set, the newly created process becomes a child
639 * of init. This effectively disassociates the child from the
642 if ((fr->fr_flags & RFNOWAIT) != 0) {
646 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
651 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
652 LIST_INIT(&p2->p_reaplist);
653 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
654 if (p2->p_reaper == p1)
655 p2->p_reapsubtree = p2->p_pid;
656 sx_xunlock(&proctree_lock);
658 /* Inform accounting that we have forked. */
659 p2->p_acflag = AFORK;
667 * Finish creating the child process. It will return via a different
668 * execution path later. (ie: directly into user mode)
670 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
672 if (fr->fr_flags == (RFFDG | RFPROC)) {
674 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
675 p2->p_vmspace->vm_ssize);
676 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
677 VM_CNT_INC(v_vforks);
678 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
679 p2->p_vmspace->vm_ssize);
680 } else if (p1 == &proc0) {
681 VM_CNT_INC(v_kthreads);
682 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
683 p2->p_vmspace->vm_ssize);
685 VM_CNT_INC(v_rforks);
686 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
687 p2->p_vmspace->vm_ssize);
691 * Associate the process descriptor with the process before anything
692 * can happen that might cause that process to need the descriptor.
693 * However, don't do this until after fork(2) can no longer fail.
695 if (fr->fr_flags & RFPROCDESC)
696 procdesc_new(p2, fr->fr_pd_flags);
699 * Both processes are set up, now check if any loadable modules want
700 * to adjust anything.
702 EVENTHANDLER_INVOKE(process_fork, p1, p2, fr->fr_flags);
705 * Set the child start time and mark the process as being complete.
709 microuptime(&p2->p_stats->p_start);
711 p2->p_state = PRS_NORMAL;
716 * Tell the DTrace fasttrap provider about the new process so that any
717 * tracepoints inherited from the parent can be removed. We have to do
718 * this only after p_state is PRS_NORMAL since the fasttrap module will
719 * use pfind() later on.
721 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
722 dtrace_fasttrap_fork(p1, p2);
725 * Hold the process so that it cannot exit after we make it runnable,
726 * but before we wait for the debugger.
729 if (p1->p_ptevents & PTRACE_FORK) {
731 * Arrange for debugger to receive the fork event.
733 * We can report PL_FLAG_FORKED regardless of
734 * P_FOLLOWFORK settings, but it does not make a sense
737 td->td_dbgflags |= TDB_FORK;
738 td->td_dbg_forked = p2->p_pid;
739 td2->td_dbgflags |= TDB_STOPATFORK;
741 if (fr->fr_flags & RFPPWAIT) {
742 td->td_pflags |= TDP_RFPPWAIT;
743 td->td_rfppwait_p = p2;
744 td->td_dbgflags |= TDB_VFORK;
749 * Now can be swapped.
755 * Tell any interested parties about the new process.
757 knote_fork(p1->p_klist, p2->p_pid);
758 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
760 if (fr->fr_flags & RFPROCDESC) {
761 procdesc_finit(p2->p_procdesc, fp_procdesc);
762 fdrop(fp_procdesc, td);
765 if ((fr->fr_flags & RFSTOPPED) == 0) {
767 * If RFSTOPPED not requested, make child runnable and
772 sched_add(td2, SRQ_BORING);
774 if (fr->fr_pidp != NULL)
775 *fr->fr_pidp = p2->p_pid;
782 * Wait until debugger is attached to child.
784 while (td2->td_proc == p2 && (td2->td_dbgflags & TDB_STOPATFORK) != 0)
785 cv_wait(&p2->p_dbgwait, &p2->p_mtx);
787 racct_proc_fork_done(p2);
792 fork1(struct thread *td, struct fork_req *fr)
794 struct proc *p1, *newproc;
797 struct file *fp_procdesc;
798 vm_ooffset_t mem_charged;
799 int error, nprocs_new, ok;
801 static struct timeval lastfail;
804 flags = fr->fr_flags;
805 pages = fr->fr_pages;
807 if ((flags & RFSTOPPED) != 0)
808 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
810 MPASS(fr->fr_procp == NULL);
812 /* Check for the undefined or unimplemented flags. */
813 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
816 /* Signal value requires RFTSIGZMB. */
817 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
820 /* Can't copy and clear. */
821 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
824 /* Check the validity of the signal number. */
825 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
828 if ((flags & RFPROCDESC) != 0) {
829 /* Can't not create a process yet get a process descriptor. */
830 if ((flags & RFPROC) == 0)
833 /* Must provide a place to put a procdesc if creating one. */
834 if (fr->fr_pd_fd == NULL)
837 /* Check if we are using supported flags. */
838 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
845 * Here we don't create a new process, but we divorce
846 * certain parts of a process from itself.
848 if ((flags & RFPROC) == 0) {
849 if (fr->fr_procp != NULL)
850 *fr->fr_procp = NULL;
851 else if (fr->fr_pidp != NULL)
853 return (fork_norfproc(td, flags));
861 * Increment the nprocs resource before allocations occur.
862 * Although process entries are dynamically created, we still
863 * keep a global limit on the maximum number we will
864 * create. There are hard-limits as to the number of processes
865 * that can run, established by the KVA and memory usage for
868 * Don't allow a nonprivileged user to use the last ten
869 * processes; don't let root exceed the limit.
871 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
872 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred,
873 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) {
875 sx_xlock(&allproc_lock);
876 if (ppsratecheck(&lastfail, &curfail, 1)) {
877 printf("maxproc limit exceeded by uid %u (pid %d); "
878 "see tuning(7) and login.conf(5)\n",
879 td->td_ucred->cr_ruid, p1->p_pid);
881 sx_xunlock(&allproc_lock);
886 * If required, create a process descriptor in the parent first; we
887 * will abandon it if something goes wrong. We don't finit() until
890 if (flags & RFPROCDESC) {
891 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
892 fr->fr_pd_flags, fr->fr_pd_fcaps);
899 pages = kstack_pages;
900 /* Allocate new proc. */
901 newproc = uma_zalloc(proc_zone, M_WAITOK);
902 td2 = FIRST_THREAD_IN_PROC(newproc);
904 td2 = thread_alloc(pages);
909 proc_linkup(newproc, td2);
911 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
912 if (td2->td_kstack != 0)
913 vm_thread_dispose(td2);
914 if (!thread_alloc_stack(td2, pages)) {
921 if ((flags & RFMEM) == 0) {
922 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
927 if (!swap_reserve(mem_charged)) {
929 * The swap reservation failed. The accounting
930 * from the entries of the copied vm2 will be
931 * subtracted in vmspace_free(), so force the
934 swap_reserve_force(mem_charged);
942 * XXX: This is ugly; when we copy resource usage, we need to bump
943 * per-cred resource counters.
945 proc_set_cred_init(newproc, crhold(td->td_ucred));
948 * Initialize resource accounting for the child process.
950 error = racct_proc_fork(p1, newproc);
957 mac_proc_init(newproc);
959 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
960 STAILQ_INIT(&newproc->p_ktr);
962 /* We have to lock the process tree while we look for a pid. */
963 sx_slock(&proctree_lock);
964 sx_xlock(&allproc_lock);
967 * Increment the count of procs running with this uid. Don't allow
968 * a nonprivileged user to exceed their current limit.
970 * XXXRW: Can we avoid privilege here if it's not needed?
972 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
974 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
976 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
977 lim_cur(td, RLIMIT_NPROC));
980 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
985 sx_sunlock(&proctree_lock);
986 sx_xunlock(&allproc_lock);
988 mac_proc_destroy(newproc);
990 racct_proc_exit(newproc);
992 crfree(newproc->p_ucred);
993 newproc->p_ucred = NULL;
997 uma_zfree(proc_zone, newproc);
998 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
999 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1000 fdrop(fp_procdesc, td);
1002 atomic_add_int(&nprocs, -1);
1003 pause("fork", hz / 2);
1008 * Handle the return of a child process from fork1(). This function
1009 * is called from the MD fork_trampoline() entry point.
1012 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1013 struct trapframe *frame)
1021 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1023 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1024 td, td_get_sched(td), p->p_pid, td->td_name);
1026 sched_fork_exit(td);
1028 * Processes normally resume in mi_switch() after being
1029 * cpu_switch()'ed to, but when children start up they arrive here
1030 * instead, so we must do much the same things as mi_switch() would.
1032 if ((dtd = PCPU_GET(deadthread))) {
1033 PCPU_SET(deadthread, NULL);
1039 * cpu_fork_kthread_handler intercepts this function call to
1040 * have this call a non-return function to stay in kernel mode.
1041 * initproc has its own fork handler, but it does return.
1043 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1044 callout(arg, frame);
1047 * Check if a kernel thread misbehaved and returned from its main
1050 if (p->p_flag & P_KPROC) {
1051 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1052 td->td_name, p->p_pid);
1055 mtx_assert(&Giant, MA_NOTOWNED);
1057 if (p->p_sysent->sv_schedtail != NULL)
1058 (p->p_sysent->sv_schedtail)(td);
1059 td->td_pflags &= ~TDP_FORKING;
1063 * Simplified back end of syscall(), used when returning from fork()
1064 * directly into user mode. This function is passed in to fork_exit()
1065 * as the first parameter and is called when returning to a new
1069 fork_return(struct thread *td, struct trapframe *frame)
1071 struct proc *p, *dbg;
1074 if (td->td_dbgflags & TDB_STOPATFORK) {
1075 sx_xlock(&proctree_lock);
1077 if (p->p_pptr->p_ptevents & PTRACE_FORK) {
1079 * If debugger still wants auto-attach for the
1080 * parent's children, do it now.
1082 dbg = p->p_pptr->p_pptr;
1083 proc_set_traced(p, true);
1085 "fork_return: attaching to new child pid %d: oppid %d",
1086 p->p_pid, p->p_oppid);
1087 proc_reparent(p, dbg);
1088 sx_xunlock(&proctree_lock);
1089 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1090 ptracestop(td, SIGSTOP, NULL);
1091 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1094 * ... otherwise clear the request.
1096 sx_xunlock(&proctree_lock);
1097 td->td_dbgflags &= ~TDB_STOPATFORK;
1098 cv_broadcast(&p->p_dbgwait);
1101 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1103 * This is the start of a new thread in a traced
1104 * process. Report a system call exit event.
1107 td->td_dbgflags |= TDB_SCX;
1108 _STOPEVENT(p, S_SCX, td->td_sa.code);
1109 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1110 (td->td_dbgflags & TDB_BORN) != 0)
1111 ptracestop(td, SIGTRAP, NULL);
1112 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1119 if (KTRPOINT(td, KTR_SYSRET))
1120 ktrsysret(SYS_fork, 0, 0);