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_ktrace.h"
41 #include "opt_kstack_pages.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/eventhandler.h>
47 #include <sys/fcntl.h>
48 #include <sys/filedesc.h>
50 #include <sys/kernel.h>
51 #include <sys/kthread.h>
52 #include <sys/sysctl.h>
54 #include <sys/malloc.h>
55 #include <sys/mutex.h>
58 #include <sys/procdesc.h>
59 #include <sys/pioctl.h>
60 #include <sys/ptrace.h>
61 #include <sys/racct.h>
62 #include <sys/resourcevar.h>
63 #include <sys/sched.h>
64 #include <sys/syscall.h>
65 #include <sys/vmmeter.h>
66 #include <sys/vnode.h>
69 #include <sys/ktrace.h>
70 #include <sys/unistd.h>
73 #include <sys/sysent.h>
74 #include <sys/signalvar.h>
76 #include <security/audit/audit.h>
77 #include <security/mac/mac_framework.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_extern.h>
84 #include <vm/vm_domain.h>
87 #include <sys/dtrace_bsd.h>
88 dtrace_fork_func_t dtrace_fasttrap_fork;
91 SDT_PROVIDER_DECLARE(proc);
92 SDT_PROBE_DEFINE3(proc, , , create, "struct proc *", "struct proc *", "int");
94 #ifndef _SYS_SYSPROTO_H_
100 EVENTHANDLER_LIST_DECLARE(process_fork);
104 sys_fork(struct thread *td, struct fork_args *uap)
109 bzero(&fr, sizeof(fr));
110 fr.fr_flags = RFFDG | RFPROC;
112 error = fork1(td, &fr);
114 td->td_retval[0] = pid;
115 td->td_retval[1] = 0;
122 sys_pdfork(struct thread *td, struct pdfork_args *uap)
127 bzero(&fr, sizeof(fr));
128 fr.fr_flags = RFFDG | RFPROC | RFPROCDESC;
131 fr.fr_pd_flags = uap->flags;
133 * It is necessary to return fd by reference because 0 is a valid file
134 * descriptor number, and the child needs to be able to distinguish
135 * itself from the parent using the return value.
137 error = fork1(td, &fr);
139 td->td_retval[0] = pid;
140 td->td_retval[1] = 0;
141 error = copyout(&fd, uap->fdp, sizeof(fd));
148 sys_vfork(struct thread *td, struct vfork_args *uap)
153 bzero(&fr, sizeof(fr));
154 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
156 error = fork1(td, &fr);
158 td->td_retval[0] = pid;
159 td->td_retval[1] = 0;
165 sys_rfork(struct thread *td, struct rfork_args *uap)
170 /* Don't allow kernel-only flags. */
171 if ((uap->flags & RFKERNELONLY) != 0)
174 AUDIT_ARG_FFLAGS(uap->flags);
175 bzero(&fr, sizeof(fr));
176 fr.fr_flags = uap->flags;
178 error = fork1(td, &fr);
180 td->td_retval[0] = pid;
181 td->td_retval[1] = 0;
186 int nprocs = 1; /* process 0 */
188 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
192 * Random component to lastpid generation. We mix in a random factor to make
193 * it a little harder to predict. We sanity check the modulus value to avoid
194 * doing it in critical paths. Don't let it be too small or we pointlessly
195 * waste randomness entropy, and don't let it be impossibly large. Using a
196 * modulus that is too big causes a LOT more process table scans and slows
197 * down fork processing as the pidchecked caching is defeated.
199 static int randompid = 0;
202 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
206 error = sysctl_wire_old_buffer(req, sizeof(int));
209 sx_xlock(&allproc_lock);
211 error = sysctl_handle_int(oidp, &pid, 0, req);
212 if (error == 0 && req->newptr != NULL) {
216 /* generate a random PID modulus between 100 and 1123 */
217 randompid = 100 + arc4random() % 1024;
218 else if (pid < 0 || pid > pid_max - 100)
220 randompid = pid_max - 100;
222 /* Make it reasonable */
227 sx_xunlock(&allproc_lock);
231 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
232 0, 0, sysctl_kern_randompid, "I", "Random PID modulus. Special values: 0: disable, 1: choose random value");
235 fork_findpid(int flags)
239 static int pidchecked = 0;
242 * Requires allproc_lock in order to iterate over the list
243 * of processes, and proctree_lock to access p_pgrp.
245 sx_assert(&allproc_lock, SX_LOCKED);
246 sx_assert(&proctree_lock, SX_LOCKED);
249 * Find an unused process ID. We remember a range of unused IDs
250 * ready to use (from lastpid+1 through pidchecked-1).
252 * If RFHIGHPID is set (used during system boot), do not allocate
255 trypid = lastpid + 1;
256 if (flags & RFHIGHPID) {
261 trypid += arc4random() % randompid;
265 * If the process ID prototype has wrapped around,
266 * restart somewhat above 0, as the low-numbered procs
267 * tend to include daemons that don't exit.
269 if (trypid >= pid_max) {
270 trypid = trypid % pid_max;
275 if (trypid >= pidchecked) {
278 pidchecked = PID_MAX;
280 * Scan the active and zombie procs to check whether this pid
281 * is in use. Remember the lowest pid that's greater
282 * than trypid, so we can avoid checking for a while.
284 * Avoid reuse of the process group id, session id or
285 * the reaper subtree id. Note that for process group
286 * and sessions, the amount of reserved pids is
287 * limited by process limit. For the subtree ids, the
288 * id is kept reserved only while there is a
289 * non-reaped process in the subtree, so amount of
290 * reserved pids is limited by process limit times
293 p = LIST_FIRST(&allproc);
295 for (; p != NULL; p = LIST_NEXT(p, p_list)) {
296 while (p->p_pid == trypid ||
297 p->p_reapsubtree == trypid ||
298 (p->p_pgrp != NULL &&
299 (p->p_pgrp->pg_id == trypid ||
300 (p->p_session != NULL &&
301 p->p_session->s_sid == trypid)))) {
303 if (trypid >= pidchecked)
306 if (p->p_pid > trypid && pidchecked > p->p_pid)
307 pidchecked = p->p_pid;
308 if (p->p_pgrp != NULL) {
309 if (p->p_pgrp->pg_id > trypid &&
310 pidchecked > p->p_pgrp->pg_id)
311 pidchecked = p->p_pgrp->pg_id;
312 if (p->p_session != NULL &&
313 p->p_session->s_sid > trypid &&
314 pidchecked > p->p_session->s_sid)
315 pidchecked = p->p_session->s_sid;
320 p = LIST_FIRST(&zombproc);
326 * RFHIGHPID does not mess with the lastpid counter during boot.
328 if (flags & RFHIGHPID)
337 fork_norfproc(struct thread *td, int flags)
342 KASSERT((flags & RFPROC) == 0,
343 ("fork_norfproc called with RFPROC set"));
346 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
347 (flags & (RFCFDG | RFFDG))) {
349 if (thread_single(p1, SINGLE_BOUNDARY)) {
356 error = vm_forkproc(td, NULL, NULL, NULL, flags);
361 * Close all file descriptors.
363 if (flags & RFCFDG) {
364 struct filedesc *fdtmp;
365 fdtmp = fdinit(td->td_proc->p_fd, false);
371 * Unshare file descriptors (from parent).
377 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
378 (flags & (RFCFDG | RFFDG))) {
380 thread_single_end(p1, SINGLE_BOUNDARY);
387 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
388 struct vmspace *vm2, struct file *fp_procdesc)
390 struct proc *p1, *pptr;
393 struct filedesc_to_leader *fdtol;
394 struct sigacts *newsigacts;
396 sx_assert(&proctree_lock, SX_SLOCKED);
397 sx_assert(&allproc_lock, SX_XLOCKED);
401 trypid = fork_findpid(fr->fr_flags);
403 sx_sunlock(&proctree_lock);
405 p2->p_state = PRS_NEW; /* protect against others */
407 AUDIT_ARG_PID(p2->p_pid);
408 LIST_INSERT_HEAD(&allproc, p2, p_list);
410 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
415 sx_xunlock(&allproc_lock);
417 bcopy(&p1->p_startcopy, &p2->p_startcopy,
418 __rangeof(struct proc, p_startcopy, p_endcopy));
419 p2->p_elf_machine = p1->p_elf_machine;
420 p2->p_elf_flags = p1->p_elf_flags;
421 pargs_hold(p2->p_args);
425 bzero(&p2->p_startzero,
426 __rangeof(struct proc, p_startzero, p_endzero));
430 /* Tell the prison that we exist. */
431 prison_proc_hold(p2->p_ucred->cr_prison);
436 * Malloc things while we don't hold any locks.
438 if (fr->fr_flags & RFSIGSHARE)
441 newsigacts = sigacts_alloc();
446 if (fr->fr_flags & RFCFDG) {
447 fd = fdinit(p1->p_fd, false);
449 } else if (fr->fr_flags & RFFDG) {
450 fd = fdcopy(p1->p_fd);
453 fd = fdshare(p1->p_fd);
454 if (p1->p_fdtol == NULL)
455 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
457 if ((fr->fr_flags & RFTHREAD) != 0) {
459 * Shared file descriptor table, and shared
463 FILEDESC_XLOCK(p1->p_fd);
464 fdtol->fdl_refcount++;
465 FILEDESC_XUNLOCK(p1->p_fd);
468 * Shared file descriptor table, and different
471 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
476 * Make a proc table entry for the new process.
477 * Start by zeroing the section of proc that is zero-initialized,
478 * then copy the section that is copied directly from the parent.
484 bzero(&td2->td_startzero,
485 __rangeof(struct thread, td_startzero, td_endzero));
486 td2->td_sleeptimo = 0;
487 td2->td_vslock_sz = 0;
488 bzero(&td2->td_si, sizeof(td2->td_si));
490 bcopy(&td->td_startcopy, &td2->td_startcopy,
491 __rangeof(struct thread, td_startcopy, td_endcopy));
492 td2->td_sa = td->td_sa;
494 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
495 td2->td_sigstk = td->td_sigstk;
496 td2->td_flags = TDF_INMEM;
497 td2->td_lend_user_pri = PRI_MAX;
501 td2->td_vnet_lpush = NULL;
505 * Allow the scheduler to initialize the child.
512 * Duplicate sub-structures as needed.
513 * Increase reference counts on shared objects.
515 p2->p_flag = P_INMEM;
516 p2->p_flag2 = p1->p_flag2 & (P2_NOTRACE | P2_NOTRACE_EXEC | P2_TRAPCAP);
517 p2->p_swtick = ticks;
518 if (p1->p_flag & P_PROFIL)
522 * Whilst the proc lock is held, copy the VM domain data out
523 * using the VM domain method.
525 vm_domain_policy_init(&p2->p_vm_dom_policy);
526 vm_domain_policy_localcopy(&p2->p_vm_dom_policy,
527 &p1->p_vm_dom_policy);
529 if (fr->fr_flags & RFSIGSHARE) {
530 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
532 sigacts_copy(newsigacts, p1->p_sigacts);
533 p2->p_sigacts = newsigacts;
536 if (fr->fr_flags & RFTSIGZMB)
537 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
538 else if (fr->fr_flags & RFLINUXTHPN)
539 p2->p_sigparent = SIGUSR1;
541 p2->p_sigparent = SIGCHLD;
543 p2->p_textvp = p1->p_textvp;
547 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
548 p2->p_flag |= P_PROTECTED;
549 p2->p_flag2 |= P2_INHERIT_PROTECTED;
553 * p_limit is copy-on-write. Bump its refcount.
557 thread_cow_get_proc(td2, p2);
559 pstats_fork(p1->p_stats, p2->p_stats);
564 /* Bump references to the text vnode (for procfs). */
566 vrefact(p2->p_textvp);
569 * Set up linkage for kernel based threading.
571 if ((fr->fr_flags & RFTHREAD) != 0) {
572 mtx_lock(&ppeers_lock);
573 p2->p_peers = p1->p_peers;
575 p2->p_leader = p1->p_leader;
576 mtx_unlock(&ppeers_lock);
577 PROC_LOCK(p1->p_leader);
578 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
579 PROC_UNLOCK(p1->p_leader);
581 * The task leader is exiting, so process p1 is
582 * going to be killed shortly. Since p1 obviously
583 * isn't dead yet, we know that the leader is either
584 * sending SIGKILL's to all the processes in this
585 * task or is sleeping waiting for all the peers to
586 * exit. We let p1 complete the fork, but we need
587 * to go ahead and kill the new process p2 since
588 * the task leader may not get a chance to send
589 * SIGKILL to it. We leave it on the list so that
590 * the task leader will wait for this new process
594 kern_psignal(p2, SIGKILL);
597 PROC_UNLOCK(p1->p_leader);
603 sx_xlock(&proctree_lock);
604 PGRP_LOCK(p1->p_pgrp);
609 * Preserve some more flags in subprocess. P_PROFIL has already
612 p2->p_flag |= p1->p_flag & P_SUGID;
613 td2->td_pflags |= (td->td_pflags & TDP_ALTSTACK) | TDP_FORKING;
614 SESS_LOCK(p1->p_session);
615 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
616 p2->p_flag |= P_CONTROLT;
617 SESS_UNLOCK(p1->p_session);
618 if (fr->fr_flags & RFPPWAIT)
619 p2->p_flag |= P_PPWAIT;
621 p2->p_pgrp = p1->p_pgrp;
622 LIST_INSERT_AFTER(p1, p2, p_pglist);
623 PGRP_UNLOCK(p1->p_pgrp);
624 LIST_INIT(&p2->p_children);
625 LIST_INIT(&p2->p_orphans);
627 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
630 * If PF_FORK is set, the child process inherits the
631 * procfs ioctl flags from its parent.
633 if (p1->p_pfsflags & PF_FORK) {
634 p2->p_stops = p1->p_stops;
635 p2->p_pfsflags = p1->p_pfsflags;
639 * This begins the section where we must prevent the parent
640 * from being swapped.
646 * Attach the new process to its parent.
648 * If RFNOWAIT is set, the newly created process becomes a child
649 * of init. This effectively disassociates the child from the
652 if ((fr->fr_flags & RFNOWAIT) != 0) {
656 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
661 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
662 LIST_INIT(&p2->p_reaplist);
663 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
664 if (p2->p_reaper == p1)
665 p2->p_reapsubtree = p2->p_pid;
666 sx_xunlock(&proctree_lock);
668 /* Inform accounting that we have forked. */
669 p2->p_acflag = AFORK;
677 * Finish creating the child process. It will return via a different
678 * execution path later. (ie: directly into user mode)
680 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
682 if (fr->fr_flags == (RFFDG | RFPROC)) {
683 PCPU_INC(cnt.v_forks);
684 PCPU_ADD(cnt.v_forkpages, p2->p_vmspace->vm_dsize +
685 p2->p_vmspace->vm_ssize);
686 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
687 PCPU_INC(cnt.v_vforks);
688 PCPU_ADD(cnt.v_vforkpages, p2->p_vmspace->vm_dsize +
689 p2->p_vmspace->vm_ssize);
690 } else if (p1 == &proc0) {
691 PCPU_INC(cnt.v_kthreads);
692 PCPU_ADD(cnt.v_kthreadpages, p2->p_vmspace->vm_dsize +
693 p2->p_vmspace->vm_ssize);
695 PCPU_INC(cnt.v_rforks);
696 PCPU_ADD(cnt.v_rforkpages, p2->p_vmspace->vm_dsize +
697 p2->p_vmspace->vm_ssize);
701 * Associate the process descriptor with the process before anything
702 * can happen that might cause that process to need the descriptor.
703 * However, don't do this until after fork(2) can no longer fail.
705 if (fr->fr_flags & RFPROCDESC)
706 procdesc_new(p2, fr->fr_pd_flags);
709 * Both processes are set up, now check if any loadable modules want
710 * to adjust anything.
712 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
715 * Set the child start time and mark the process as being complete.
719 microuptime(&p2->p_stats->p_start);
721 p2->p_state = PRS_NORMAL;
726 * Tell the DTrace fasttrap provider about the new process so that any
727 * tracepoints inherited from the parent can be removed. We have to do
728 * this only after p_state is PRS_NORMAL since the fasttrap module will
729 * use pfind() later on.
731 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
732 dtrace_fasttrap_fork(p1, p2);
735 * Hold the process so that it cannot exit after we make it runnable,
736 * but before we wait for the debugger.
739 if (fr->fr_flags & RFPPWAIT) {
740 td->td_pflags |= TDP_RFPPWAIT;
741 td->td_rfppwait_p = p2;
742 td->td_dbgflags |= TDB_VFORK;
747 * Now can be swapped.
753 * Tell any interested parties about the new process.
755 knote_fork(p1->p_klist, p2->p_pid);
756 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
758 if (fr->fr_flags & RFPROCDESC) {
759 procdesc_finit(p2->p_procdesc, fp_procdesc);
760 fdrop(fp_procdesc, td);
764 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
765 * synced with forks in progress so it is OK if we miss it
768 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
769 sx_xlock(&proctree_lock);
773 * p1->p_ptevents & p1->p_pptr are protected by both
774 * process and proctree locks for modifications,
775 * so owning proctree_lock allows the race-free read.
777 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
779 * Arrange for debugger to receive the fork event.
781 * We can report PL_FLAG_FORKED regardless of
782 * P_FOLLOWFORK settings, but it does not make a sense
785 td->td_dbgflags |= TDB_FORK;
786 td->td_dbg_forked = p2->p_pid;
787 td2->td_dbgflags |= TDB_STOPATFORK;
788 proc_set_traced(p2, true);
790 "do_fork: attaching to new child pid %d: oppid %d",
791 p2->p_pid, p2->p_oppid);
792 proc_reparent(p2, p1->p_pptr);
795 sx_xunlock(&proctree_lock);
798 if ((fr->fr_flags & RFSTOPPED) == 0) {
800 * If RFSTOPPED not requested, make child runnable and
805 sched_add(td2, SRQ_BORING);
807 if (fr->fr_pidp != NULL)
808 *fr->fr_pidp = p2->p_pid;
815 racct_proc_fork_done(p2);
820 fork1(struct thread *td, struct fork_req *fr)
822 struct proc *p1, *newproc;
825 struct file *fp_procdesc;
826 vm_ooffset_t mem_charged;
827 int error, nprocs_new, ok;
829 static struct timeval lastfail;
832 flags = fr->fr_flags;
833 pages = fr->fr_pages;
835 if ((flags & RFSTOPPED) != 0)
836 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
838 MPASS(fr->fr_procp == NULL);
840 /* Check for the undefined or unimplemented flags. */
841 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
844 /* Signal value requires RFTSIGZMB. */
845 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
848 /* Can't copy and clear. */
849 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
852 /* Check the validity of the signal number. */
853 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
856 if ((flags & RFPROCDESC) != 0) {
857 /* Can't not create a process yet get a process descriptor. */
858 if ((flags & RFPROC) == 0)
861 /* Must provide a place to put a procdesc if creating one. */
862 if (fr->fr_pd_fd == NULL)
865 /* Check if we are using supported flags. */
866 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
873 * Here we don't create a new process, but we divorce
874 * certain parts of a process from itself.
876 if ((flags & RFPROC) == 0) {
877 if (fr->fr_procp != NULL)
878 *fr->fr_procp = NULL;
879 else if (fr->fr_pidp != NULL)
881 return (fork_norfproc(td, flags));
889 * Increment the nprocs resource before allocations occur.
890 * Although process entries are dynamically created, we still
891 * keep a global limit on the maximum number we will
892 * create. There are hard-limits as to the number of processes
893 * that can run, established by the KVA and memory usage for
896 * Don't allow a nonprivileged user to use the last ten
897 * processes; don't let root exceed the limit.
899 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
900 if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred,
901 PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) {
903 sx_xlock(&allproc_lock);
904 if (ppsratecheck(&lastfail, &curfail, 1)) {
905 printf("maxproc limit exceeded by uid %u (pid %d); "
906 "see tuning(7) and login.conf(5)\n",
907 td->td_ucred->cr_ruid, p1->p_pid);
909 sx_xunlock(&allproc_lock);
914 * If required, create a process descriptor in the parent first; we
915 * will abandon it if something goes wrong. We don't finit() until
918 if (flags & RFPROCDESC) {
919 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
920 fr->fr_pd_flags, fr->fr_pd_fcaps);
927 pages = kstack_pages;
928 /* Allocate new proc. */
929 newproc = uma_zalloc(proc_zone, M_WAITOK);
930 td2 = FIRST_THREAD_IN_PROC(newproc);
932 td2 = thread_alloc(pages);
937 proc_linkup(newproc, td2);
939 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
940 if (td2->td_kstack != 0)
941 vm_thread_dispose(td2);
942 if (!thread_alloc_stack(td2, pages)) {
949 if ((flags & RFMEM) == 0) {
950 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
955 if (!swap_reserve(mem_charged)) {
957 * The swap reservation failed. The accounting
958 * from the entries of the copied vm2 will be
959 * subtracted in vmspace_free(), so force the
962 swap_reserve_force(mem_charged);
970 * XXX: This is ugly; when we copy resource usage, we need to bump
971 * per-cred resource counters.
973 proc_set_cred_init(newproc, crhold(td->td_ucred));
976 * Initialize resource accounting for the child process.
978 error = racct_proc_fork(p1, newproc);
985 mac_proc_init(newproc);
987 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
988 STAILQ_INIT(&newproc->p_ktr);
990 /* We have to lock the process tree while we look for a pid. */
991 sx_slock(&proctree_lock);
992 sx_xlock(&allproc_lock);
995 * Increment the count of procs running with this uid. Don't allow
996 * a nonprivileged user to exceed their current limit.
998 * XXXRW: Can we avoid privilege here if it's not needed?
1000 error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
1002 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
1004 ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
1005 lim_cur(td, RLIMIT_NPROC));
1008 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1013 sx_sunlock(&proctree_lock);
1014 sx_xunlock(&allproc_lock);
1016 mac_proc_destroy(newproc);
1018 racct_proc_exit(newproc);
1020 crfree(newproc->p_ucred);
1021 newproc->p_ucred = NULL;
1025 uma_zfree(proc_zone, newproc);
1026 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1027 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1028 fdrop(fp_procdesc, td);
1030 atomic_add_int(&nprocs, -1);
1031 pause("fork", hz / 2);
1036 * Handle the return of a child process from fork1(). This function
1037 * is called from the MD fork_trampoline() entry point.
1040 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1041 struct trapframe *frame)
1049 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1051 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1052 td, td_get_sched(td), p->p_pid, td->td_name);
1054 sched_fork_exit(td);
1056 * Processes normally resume in mi_switch() after being
1057 * cpu_switch()'ed to, but when children start up they arrive here
1058 * instead, so we must do much the same things as mi_switch() would.
1060 if ((dtd = PCPU_GET(deadthread))) {
1061 PCPU_SET(deadthread, NULL);
1067 * cpu_fork_kthread_handler intercepts this function call to
1068 * have this call a non-return function to stay in kernel mode.
1069 * initproc has its own fork handler, but it does return.
1071 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1072 callout(arg, frame);
1075 * Check if a kernel thread misbehaved and returned from its main
1078 if (p->p_flag & P_KPROC) {
1079 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1080 td->td_name, p->p_pid);
1083 mtx_assert(&Giant, MA_NOTOWNED);
1085 if (p->p_sysent->sv_schedtail != NULL)
1086 (p->p_sysent->sv_schedtail)(td);
1087 td->td_pflags &= ~TDP_FORKING;
1091 * Simplified back end of syscall(), used when returning from fork()
1092 * directly into user mode. This function is passed in to fork_exit()
1093 * as the first parameter and is called when returning to a new
1097 fork_return(struct thread *td, struct trapframe *frame)
1102 if (td->td_dbgflags & TDB_STOPATFORK) {
1104 if ((p->p_flag & P_TRACED) != 0) {
1106 * Inform the debugger if one is still present.
1108 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1109 ptracestop(td, SIGSTOP, NULL);
1110 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1113 * ... otherwise clear the request.
1115 td->td_dbgflags &= ~TDB_STOPATFORK;
1118 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1120 * This is the start of a new thread in a traced
1121 * process. Report a system call exit event.
1124 td->td_dbgflags |= TDB_SCX;
1125 _STOPEVENT(p, S_SCX, td->td_sa.code);
1126 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1127 (td->td_dbgflags & TDB_BORN) != 0)
1128 ptracestop(td, SIGTRAP, NULL);
1129 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1136 if (KTRPOINT(td, KTR_SYSRET))
1137 ktrsysret(SYS_fork, 0, 0);