2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_ktrace.h"
43 #include "opt_kstack_pages.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/bitstring.h>
48 #include <sys/sysproto.h>
49 #include <sys/eventhandler.h>
50 #include <sys/fcntl.h>
51 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/sysctl.h>
57 #include <sys/malloc.h>
59 #include <sys/mutex.h>
62 #include <sys/procdesc.h>
63 #include <sys/ptrace.h>
64 #include <sys/racct.h>
65 #include <sys/resourcevar.h>
66 #include <sys/sched.h>
67 #include <sys/syscall.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
72 #include <sys/ktrace.h>
73 #include <sys/unistd.h>
76 #include <sys/sysent.h>
77 #include <sys/signalvar.h>
79 #include <security/audit/audit.h>
80 #include <security/mac/mac_framework.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_extern.h>
89 #include <sys/dtrace_bsd.h>
90 dtrace_fork_func_t dtrace_fasttrap_fork;
93 SDT_PROVIDER_DECLARE(proc);
94 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
96 #ifndef _SYS_SYSPROTO_H_
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;
132 AUDIT_ARG_FFLAGS(uap->flags);
134 * It is necessary to return fd by reference because 0 is a valid file
135 * descriptor number, and the child needs to be able to distinguish
136 * itself from the parent using the return value.
138 error = fork1(td, &fr);
140 td->td_retval[0] = pid;
141 td->td_retval[1] = 0;
142 error = copyout(&fd, uap->fdp, sizeof(fd));
149 sys_vfork(struct thread *td, struct vfork_args *uap)
154 bzero(&fr, sizeof(fr));
155 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
157 error = fork1(td, &fr);
159 td->td_retval[0] = pid;
160 td->td_retval[1] = 0;
166 sys_rfork(struct thread *td, struct rfork_args *uap)
171 /* Don't allow kernel-only flags. */
172 if ((uap->flags & RFKERNELONLY) != 0)
174 /* RFSPAWN must not appear with others */
175 if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN)
178 AUDIT_ARG_FFLAGS(uap->flags);
179 bzero(&fr, sizeof(fr));
180 if ((uap->flags & RFSPAWN) != 0) {
181 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
182 fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
184 fr.fr_flags = uap->flags;
187 error = fork1(td, &fr);
189 td->td_retval[0] = pid;
190 td->td_retval[1] = 0;
195 int __exclusive_cache_line nprocs = 1; /* process 0 */
197 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
201 * Random component to lastpid generation. We mix in a random factor to make
202 * it a little harder to predict. We sanity check the modulus value to avoid
203 * doing it in critical paths. Don't let it be too small or we pointlessly
204 * waste randomness entropy, and don't let it be impossibly large. Using a
205 * modulus that is too big causes a LOT more process table scans and slows
206 * down fork processing as the pidchecked caching is defeated.
208 static int randompid = 0;
211 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
215 error = sysctl_wire_old_buffer(req, sizeof(int));
218 sx_xlock(&allproc_lock);
220 error = sysctl_handle_int(oidp, &pid, 0, req);
221 if (error == 0 && req->newptr != NULL) {
225 /* generate a random PID modulus between 100 and 1123 */
226 randompid = 100 + arc4random() % 1024;
227 else if (pid < 0 || pid > pid_max - 100)
229 randompid = pid_max - 100;
231 /* Make it reasonable */
236 sx_xunlock(&allproc_lock);
240 SYSCTL_PROC(_kern, OID_AUTO, randompid,
241 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
242 sysctl_kern_randompid, "I",
243 "Random PID modulus. Special values: 0: disable, 1: choose random value");
245 extern bitstr_t proc_id_pidmap;
246 extern bitstr_t proc_id_grpidmap;
247 extern bitstr_t proc_id_sessidmap;
248 extern bitstr_t proc_id_reapmap;
251 * Find an unused process ID
253 * If RFHIGHPID is set (used during system boot), do not allocate
257 fork_findpid(int flags)
263 * Avoid calling arc4random with procid_lock held.
266 if (__predict_false(randompid))
267 random = arc4random() % randompid;
269 mtx_lock(&procid_lock);
271 trypid = lastpid + 1;
272 if (flags & RFHIGHPID) {
279 if (trypid >= pid_max)
282 bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
284 KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
288 if (bit_test(&proc_id_grpidmap, result) ||
289 bit_test(&proc_id_sessidmap, result) ||
290 bit_test(&proc_id_reapmap, result)) {
296 * RFHIGHPID does not mess with the lastpid counter during boot.
298 if ((flags & RFHIGHPID) == 0)
301 bit_set(&proc_id_pidmap, result);
302 mtx_unlock(&procid_lock);
308 fork_norfproc(struct thread *td, int flags)
313 KASSERT((flags & RFPROC) == 0,
314 ("fork_norfproc called with RFPROC set"));
318 * Quiesce other threads if necessary. If RFMEM is not specified we
319 * must ensure that other threads do not concurrently create a second
320 * process sharing the vmspace, see vmspace_unshare().
322 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
323 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
325 if (thread_single(p1, SINGLE_BOUNDARY)) {
332 error = vm_forkproc(td, NULL, NULL, NULL, flags);
337 * Close all file descriptors.
339 if ((flags & RFCFDG) != 0) {
340 struct filedesc *fdtmp;
341 struct pwddesc *pdtmp;
343 pdtmp = pdinit(td->td_proc->p_pd, false);
352 * Unshare file descriptors (from parent).
354 if ((flags & RFFDG) != 0) {
360 if ((p1->p_flag & (P_HADTHREADS | P_SYSTEM)) == P_HADTHREADS &&
361 ((flags & (RFCFDG | RFFDG)) != 0 || (flags & RFMEM) == 0)) {
363 thread_single_end(p1, SINGLE_BOUNDARY);
370 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
371 struct vmspace *vm2, struct file *fp_procdesc)
373 struct proc *p1, *pptr;
375 struct filedesc_to_leader *fdtol;
377 struct sigacts *newsigacts;
382 bcopy(&p1->p_startcopy, &p2->p_startcopy,
383 __rangeof(struct proc, p_startcopy, p_endcopy));
384 pargs_hold(p2->p_args);
387 bzero(&p2->p_startzero,
388 __rangeof(struct proc, p_startzero, p_endzero));
390 /* Tell the prison that we exist. */
391 prison_proc_hold(p2->p_ucred->cr_prison);
393 p2->p_state = PRS_NEW; /* protect against others */
394 p2->p_pid = fork_findpid(fr->fr_flags);
395 AUDIT_ARG_PID(p2->p_pid);
396 TSFORK(p2->p_pid, p1->p_pid);
398 sx_xlock(&allproc_lock);
399 LIST_INSERT_HEAD(&allproc, p2, p_list);
401 prison_proc_link(p2->p_ucred->cr_prison, p2);
402 sx_xunlock(&allproc_lock);
404 sx_xlock(PIDHASHLOCK(p2->p_pid));
405 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
406 sx_xunlock(PIDHASHLOCK(p2->p_pid));
411 * Malloc things while we don't hold any locks.
413 if (fr->fr_flags & RFSIGSHARE)
416 newsigacts = sigacts_alloc();
421 if (fr->fr_flags & RFCFDG) {
422 pd = pdinit(p1->p_pd, false);
425 } else if (fr->fr_flags & RFFDG) {
426 if (fr->fr_flags2 & FR2_SHARE_PATHS)
427 pd = pdshare(p1->p_pd);
429 pd = pdcopy(p1->p_pd);
430 fd = fdcopy(p1->p_fd);
433 if (fr->fr_flags2 & FR2_SHARE_PATHS)
434 pd = pdcopy(p1->p_pd);
436 pd = pdshare(p1->p_pd);
437 fd = fdshare(p1->p_fd);
438 if (p1->p_fdtol == NULL)
439 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
441 if ((fr->fr_flags & RFTHREAD) != 0) {
443 * Shared file descriptor table, and shared
446 fdtol = filedesc_to_leader_share(p1->p_fdtol, p1->p_fd);
449 * Shared file descriptor table, and different
452 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
457 * Make a proc table entry for the new process.
458 * Start by zeroing the section of proc that is zero-initialized,
459 * then copy the section that is copied directly from the parent.
465 bzero(&td2->td_startzero,
466 __rangeof(struct thread, td_startzero, td_endzero));
468 bcopy(&td->td_startcopy, &td2->td_startcopy,
469 __rangeof(struct thread, td_startcopy, td_endcopy));
471 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
472 td2->td_sigstk = td->td_sigstk;
473 td2->td_flags = TDF_INMEM;
474 td2->td_lend_user_pri = PRI_MAX;
478 td2->td_vnet_lpush = NULL;
482 * Allow the scheduler to initialize the child.
487 * Request AST to check for TDP_RFPPWAIT. Do it here
488 * to avoid calling thread_lock() again.
490 if ((fr->fr_flags & RFPPWAIT) != 0)
491 ast_sched_locked(td, TDA_VFORK);
495 * Duplicate sub-structures as needed.
496 * Increase reference counts on shared objects.
498 p2->p_flag = P_INMEM;
499 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
500 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
501 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
502 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC | P2_NO_NEW_PRIVS |
503 P2_WXORX_DISABLE | P2_WXORX_ENABLE_EXEC);
504 p2->p_swtick = ticks;
505 if (p1->p_flag & P_PROFIL)
508 if (fr->fr_flags & RFSIGSHARE) {
509 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
511 sigacts_copy(newsigacts, p1->p_sigacts);
512 p2->p_sigacts = newsigacts;
513 if ((fr->fr_flags2 & (FR2_DROPSIG_CAUGHT | FR2_KPROC)) != 0) {
514 mtx_lock(&p2->p_sigacts->ps_mtx);
515 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0)
517 if ((fr->fr_flags2 & FR2_KPROC) != 0)
518 p2->p_sigacts->ps_flag |= PS_NOCLDWAIT;
519 mtx_unlock(&p2->p_sigacts->ps_mtx);
523 if (fr->fr_flags & RFTSIGZMB)
524 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
525 else if (fr->fr_flags & RFLINUXTHPN)
526 p2->p_sigparent = SIGUSR1;
528 p2->p_sigparent = SIGCHLD;
530 if ((fr->fr_flags2 & FR2_KPROC) != 0) {
531 p2->p_flag |= P_SYSTEM | P_KPROC;
532 td2->td_pflags |= TDP_KTHREAD;
535 p2->p_textvp = p1->p_textvp;
536 p2->p_textdvp = p1->p_textdvp;
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);
559 * Bump references to the text vnode and directory, and copy
562 if (p2->p_textvp != NULL)
563 vrefact(p2->p_textvp);
564 if (p2->p_textdvp != NULL)
565 vrefact(p2->p_textdvp);
566 p2->p_binname = p1->p_binname == NULL ? NULL :
567 strdup(p1->p_binname, M_PARGS);
570 * Set up linkage for kernel based threading.
572 if ((fr->fr_flags & RFTHREAD) != 0) {
573 mtx_lock(&ppeers_lock);
574 p2->p_peers = p1->p_peers;
576 p2->p_leader = p1->p_leader;
577 mtx_unlock(&ppeers_lock);
578 PROC_LOCK(p1->p_leader);
579 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
580 PROC_UNLOCK(p1->p_leader);
582 * The task leader is exiting, so process p1 is
583 * going to be killed shortly. Since p1 obviously
584 * isn't dead yet, we know that the leader is either
585 * sending SIGKILL's to all the processes in this
586 * task or is sleeping waiting for all the peers to
587 * exit. We let p1 complete the fork, but we need
588 * to go ahead and kill the new process p2 since
589 * the task leader may not get a chance to send
590 * SIGKILL to it. We leave it on the list so that
591 * the task leader will wait for this new process
595 kern_psignal(p2, SIGKILL);
598 PROC_UNLOCK(p1->p_leader);
604 sx_xlock(&proctree_lock);
605 PGRP_LOCK(p1->p_pgrp);
610 * Preserve some more flags in subprocess. P_PROFIL has already
613 p2->p_flag |= p1->p_flag & P_SUGID;
614 td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK | TDP_SIGFASTBLOCK));
615 SESS_LOCK(p1->p_session);
616 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
617 p2->p_flag |= P_CONTROLT;
618 SESS_UNLOCK(p1->p_session);
619 if (fr->fr_flags & RFPPWAIT)
620 p2->p_flag |= P_PPWAIT;
622 p2->p_pgrp = p1->p_pgrp;
623 LIST_INSERT_AFTER(p1, p2, p_pglist);
624 PGRP_UNLOCK(p1->p_pgrp);
625 LIST_INIT(&p2->p_children);
626 LIST_INIT(&p2->p_orphans);
628 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
629 TAILQ_INIT(&p2->p_kqtim_stop);
632 * This begins the section where we must prevent the parent
633 * from being swapped.
639 * Attach the new process to its parent.
641 * If RFNOWAIT is set, the newly created process becomes a child
642 * of init. This effectively disassociates the child from the
645 if ((fr->fr_flags & RFNOWAIT) != 0) {
649 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
654 p2->p_oppid = pptr->p_pid;
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 && p1 != initproc) {
659 p2->p_reapsubtree = p2->p_pid;
660 proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
662 sx_xunlock(&proctree_lock);
664 /* Inform accounting that we have forked. */
665 p2->p_acflag = AFORK;
673 * Finish creating the child process. It will return via a different
674 * execution path later. (ie: directly into user mode)
676 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
678 if (fr->fr_flags == (RFFDG | RFPROC)) {
680 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
681 p2->p_vmspace->vm_ssize);
682 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
683 VM_CNT_INC(v_vforks);
684 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
685 p2->p_vmspace->vm_ssize);
686 } else if (p1 == &proc0) {
687 VM_CNT_INC(v_kthreads);
688 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
689 p2->p_vmspace->vm_ssize);
691 VM_CNT_INC(v_rforks);
692 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
693 p2->p_vmspace->vm_ssize);
697 * Associate the process descriptor with the process before anything
698 * can happen that might cause that process to need the descriptor.
699 * However, don't do this until after fork(2) can no longer fail.
701 if (fr->fr_flags & RFPROCDESC)
702 procdesc_new(p2, fr->fr_pd_flags);
705 * Both processes are set up, now check if any loadable modules want
706 * to adjust anything.
708 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
711 * Set the child start time and mark the process as being complete.
715 microuptime(&p2->p_stats->p_start);
717 p2->p_state = PRS_NORMAL;
722 * Tell the DTrace fasttrap provider about the new process so that any
723 * tracepoints inherited from the parent can be removed. We have to do
724 * this only after p_state is PRS_NORMAL since the fasttrap module will
725 * use pfind() later on.
727 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
728 dtrace_fasttrap_fork(p1, p2);
730 if (fr->fr_flags & RFPPWAIT) {
731 td->td_pflags |= TDP_RFPPWAIT;
732 td->td_rfppwait_p = p2;
733 td->td_dbgflags |= TDB_VFORK;
738 * Tell any interested parties about the new process.
740 knote_fork(p1->p_klist, p2->p_pid);
743 * Now can be swapped.
747 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
749 if (fr->fr_flags & RFPROCDESC) {
750 procdesc_finit(p2->p_procdesc, fp_procdesc);
751 fdrop(fp_procdesc, td);
755 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
756 * synced with forks in progress so it is OK if we miss it
759 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
760 sx_xlock(&proctree_lock);
764 * p1->p_ptevents & p1->p_pptr are protected by both
765 * process and proctree locks for modifications,
766 * so owning proctree_lock allows the race-free read.
768 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
770 * Arrange for debugger to receive the fork event.
772 * We can report PL_FLAG_FORKED regardless of
773 * P_FOLLOWFORK settings, but it does not make a sense
776 td->td_dbgflags |= TDB_FORK;
777 td->td_dbg_forked = p2->p_pid;
778 td2->td_dbgflags |= TDB_STOPATFORK;
779 proc_set_traced(p2, true);
781 "do_fork: attaching to new child pid %d: oppid %d",
782 p2->p_pid, p2->p_oppid);
783 proc_reparent(p2, p1->p_pptr, false);
786 sx_xunlock(&proctree_lock);
789 racct_proc_fork_done(p2);
791 if ((fr->fr_flags & RFSTOPPED) == 0) {
792 if (fr->fr_pidp != NULL)
793 *fr->fr_pidp = p2->p_pid;
795 * If RFSTOPPED not requested, make child runnable and
800 sched_add(td2, SRQ_BORING);
807 ast_vfork(struct thread *td, int tda __unused)
811 MPASS(td->td_pflags & TDP_RFPPWAIT);
815 * Preserve synchronization semantics of vfork. If
816 * waiting for child to exec or exit, fork set
817 * P_PPWAIT on child, and there we sleep on our proc
820 * Do it after the ptracestop() above is finished, to
821 * not block our debugger until child execs or exits
822 * to finish vfork wait.
824 td->td_pflags &= ~TDP_RFPPWAIT;
825 p2 = td->td_rfppwait_p;
828 while (p2->p_flag & P_PPWAIT) {
830 if (thread_suspend_check_needed()) {
832 thread_suspend_check(0);
838 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
842 if (td->td_dbgflags & TDB_VFORK) {
844 if (p->p_ptevents & PTRACE_VFORK)
845 ptracestop(td, SIGTRAP, NULL);
846 td->td_dbgflags &= ~TDB_VFORK;
852 fork1(struct thread *td, struct fork_req *fr)
854 struct proc *p1, *newproc;
858 struct file *fp_procdesc;
859 vm_ooffset_t mem_charged;
860 int error, nprocs_new;
862 static struct timeval lastfail;
865 flags = fr->fr_flags;
866 pages = fr->fr_pages;
868 if ((flags & RFSTOPPED) != 0)
869 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
871 MPASS(fr->fr_procp == NULL);
873 /* Check for the undefined or unimplemented flags. */
874 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
877 /* Signal value requires RFTSIGZMB. */
878 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
881 /* Can't copy and clear. */
882 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
885 /* Check the validity of the signal number. */
886 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
889 if ((flags & RFPROCDESC) != 0) {
890 /* Can't not create a process yet get a process descriptor. */
891 if ((flags & RFPROC) == 0)
894 /* Must provide a place to put a procdesc if creating one. */
895 if (fr->fr_pd_fd == NULL)
898 /* Check if we are using supported flags. */
899 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
906 * Here we don't create a new process, but we divorce
907 * certain parts of a process from itself.
909 if ((flags & RFPROC) == 0) {
910 if (fr->fr_procp != NULL)
911 *fr->fr_procp = NULL;
912 else if (fr->fr_pidp != NULL)
914 return (fork_norfproc(td, flags));
922 * Increment the nprocs resource before allocations occur.
923 * Although process entries are dynamically created, we still
924 * keep a global limit on the maximum number we will
925 * create. There are hard-limits as to the number of processes
926 * that can run, established by the KVA and memory usage for
929 * Don't allow a nonprivileged user to use the last ten
930 * processes; don't let root exceed the limit.
932 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
933 if (nprocs_new >= maxproc - 10) {
934 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
935 nprocs_new >= maxproc) {
937 sx_xlock(&allproc_lock);
938 if (ppsratecheck(&lastfail, &curfail, 1)) {
939 printf("maxproc limit exceeded by uid %u "
940 "(pid %d); see tuning(7) and "
942 td->td_ucred->cr_ruid, p1->p_pid);
944 sx_xunlock(&allproc_lock);
950 * If required, create a process descriptor in the parent first; we
951 * will abandon it if something goes wrong. We don't finit() until
954 if (flags & RFPROCDESC) {
955 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
956 fr->fr_pd_flags, fr->fr_pd_fcaps);
959 AUDIT_ARG_FD(*fr->fr_pd_fd);
964 pages = kstack_pages;
965 /* Allocate new proc. */
966 newproc = uma_zalloc(proc_zone, M_WAITOK);
967 td2 = FIRST_THREAD_IN_PROC(newproc);
969 td2 = thread_alloc(pages);
974 proc_linkup(newproc, td2);
976 kmsan_thread_alloc(td2);
977 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
978 if (td2->td_kstack != 0)
979 vm_thread_dispose(td2);
980 if (!thread_alloc_stack(td2, pages)) {
987 if ((flags & RFMEM) == 0) {
988 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
993 if (!swap_reserve(mem_charged)) {
995 * The swap reservation failed. The accounting
996 * from the entries of the copied vm2 will be
997 * subtracted in vmspace_free(), so force the
1000 swap_reserve_force(mem_charged);
1008 * XXX: This is ugly; when we copy resource usage, we need to bump
1009 * per-cred resource counters.
1011 proc_set_cred_init(newproc, td->td_ucred);
1014 * Initialize resource accounting for the child process.
1016 error = racct_proc_fork(p1, newproc);
1023 mac_proc_init(newproc);
1025 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
1026 STAILQ_INIT(&newproc->p_ktr);
1029 * Increment the count of procs running with this uid. Don't allow
1030 * a nonprivileged user to exceed their current limit.
1032 cred = td->td_ucred;
1033 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
1034 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
1036 chgproccnt(cred->cr_ruidinfo, 1, 0);
1039 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1044 mac_proc_destroy(newproc);
1046 racct_proc_exit(newproc);
1048 proc_unset_cred(newproc);
1052 uma_zfree(proc_zone, newproc);
1053 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1054 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1055 fdrop(fp_procdesc, td);
1057 atomic_add_int(&nprocs, -1);
1058 pause("fork", hz / 2);
1063 * Handle the return of a child process from fork1(). This function
1064 * is called from the MD fork_trampoline() entry point.
1067 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1068 struct trapframe *frame)
1074 kmsan_mark(frame, sizeof(*frame), KMSAN_STATE_INITED);
1078 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1080 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1081 td, td_get_sched(td), p->p_pid, td->td_name);
1083 sched_fork_exit(td);
1086 * Processes normally resume in mi_switch() after being
1087 * cpu_switch()'ed to, but when children start up they arrive here
1088 * instead, so we must do much the same things as mi_switch() would.
1090 if ((dtd = PCPU_GET(deadthread))) {
1091 PCPU_SET(deadthread, NULL);
1097 * cpu_fork_kthread_handler intercepts this function call to
1098 * have this call a non-return function to stay in kernel mode.
1099 * initproc has its own fork handler, but it does return.
1101 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1102 callout(arg, frame);
1105 * Check if a kernel thread misbehaved and returned from its main
1108 if (p->p_flag & P_KPROC) {
1109 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1110 td->td_name, p->p_pid);
1113 mtx_assert(&Giant, MA_NOTOWNED);
1115 if (p->p_sysent->sv_schedtail != NULL)
1116 (p->p_sysent->sv_schedtail)(td);
1120 * Simplified back end of syscall(), used when returning from fork()
1121 * directly into user mode. This function is passed in to fork_exit()
1122 * as the first parameter and is called when returning to a new
1126 fork_return(struct thread *td, struct trapframe *frame)
1131 if (td->td_dbgflags & TDB_STOPATFORK) {
1133 if ((p->p_flag & P_TRACED) != 0) {
1135 * Inform the debugger if one is still present.
1137 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1138 ptracestop(td, SIGSTOP, NULL);
1139 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1142 * ... otherwise clear the request.
1144 td->td_dbgflags &= ~TDB_STOPATFORK;
1147 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1149 * This is the start of a new thread in a traced
1150 * process. Report a system call exit event.
1153 td->td_dbgflags |= TDB_SCX;
1154 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1155 (td->td_dbgflags & TDB_BORN) != 0)
1156 ptracestop(td, SIGTRAP, NULL);
1157 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1162 * If the prison was killed mid-fork, die along with it.
1164 if (!prison_isalive(td->td_ucred->cr_prison))
1165 exit1(td, 0, SIGKILL);
1170 if (KTRPOINT(td, KTR_SYSRET))
1171 ktrsysret(SYS_fork, 0, 0);
1176 fork_init(void *arg __unused)
1178 ast_register(TDA_VFORK, ASTR_ASTF_REQUIRED | ASTR_TDP, TDP_RFPPWAIT,
1181 SYSINIT(fork, SI_SUB_INTRINSIC, SI_ORDER_ANY, fork_init, NULL);