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>
58 #include <sys/mutex.h>
61 #include <sys/procdesc.h>
62 #include <sys/pioctl.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;
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)
173 /* RFSPAWN must not appear with others */
174 if ((uap->flags & RFSPAWN) != 0 && uap->flags != RFSPAWN)
177 AUDIT_ARG_FFLAGS(uap->flags);
178 bzero(&fr, sizeof(fr));
179 if ((uap->flags & RFSPAWN) != 0) {
180 fr.fr_flags = RFFDG | RFPROC | RFPPWAIT | RFMEM;
181 fr.fr_flags2 = FR2_DROPSIG_CAUGHT;
183 fr.fr_flags = uap->flags;
186 error = fork1(td, &fr);
188 td->td_retval[0] = pid;
189 td->td_retval[1] = 0;
194 int __exclusive_cache_line nprocs = 1; /* process 0 */
196 SYSCTL_INT(_kern, OID_AUTO, lastpid, CTLFLAG_RD, &lastpid, 0,
200 * Random component to lastpid generation. We mix in a random factor to make
201 * it a little harder to predict. We sanity check the modulus value to avoid
202 * doing it in critical paths. Don't let it be too small or we pointlessly
203 * waste randomness entropy, and don't let it be impossibly large. Using a
204 * modulus that is too big causes a LOT more process table scans and slows
205 * down fork processing as the pidchecked caching is defeated.
207 static int randompid = 0;
210 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
214 error = sysctl_wire_old_buffer(req, sizeof(int));
217 sx_xlock(&allproc_lock);
219 error = sysctl_handle_int(oidp, &pid, 0, req);
220 if (error == 0 && req->newptr != NULL) {
224 /* generate a random PID modulus between 100 and 1123 */
225 randompid = 100 + arc4random() % 1024;
226 else if (pid < 0 || pid > pid_max - 100)
228 randompid = pid_max - 100;
230 /* Make it reasonable */
235 sx_xunlock(&allproc_lock);
239 SYSCTL_PROC(_kern, OID_AUTO, randompid,
240 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
241 sysctl_kern_randompid, "I",
242 "Random PID modulus. Special values: 0: disable, 1: choose random value");
244 extern bitstr_t proc_id_pidmap;
245 extern bitstr_t proc_id_grpidmap;
246 extern bitstr_t proc_id_sessidmap;
247 extern bitstr_t proc_id_reapmap;
250 * Find an unused process ID
252 * If RFHIGHPID is set (used during system boot), do not allocate
256 fork_findpid(int flags)
262 * Avoid calling arc4random with procid_lock held.
265 if (__predict_false(randompid))
266 random = arc4random() % randompid;
268 mtx_lock(&procid_lock);
270 trypid = lastpid + 1;
271 if (flags & RFHIGHPID) {
278 if (trypid >= pid_max)
281 bit_ffc_at(&proc_id_pidmap, trypid, pid_max, &result);
283 KASSERT(trypid != 2, ("unexpectedly ran out of IDs"));
287 if (bit_test(&proc_id_grpidmap, result) ||
288 bit_test(&proc_id_sessidmap, result) ||
289 bit_test(&proc_id_reapmap, result)) {
295 * RFHIGHPID does not mess with the lastpid counter during boot.
297 if ((flags & RFHIGHPID) == 0)
300 bit_set(&proc_id_pidmap, result);
301 mtx_unlock(&procid_lock);
307 fork_norfproc(struct thread *td, int flags)
312 KASSERT((flags & RFPROC) == 0,
313 ("fork_norfproc called with RFPROC set"));
316 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
317 (flags & (RFCFDG | RFFDG))) {
319 if (thread_single(p1, SINGLE_BOUNDARY)) {
326 error = vm_forkproc(td, NULL, NULL, NULL, flags);
331 * Close all file descriptors.
333 if (flags & RFCFDG) {
334 struct filedesc *fdtmp;
335 fdtmp = fdinit(td->td_proc->p_fd, false);
341 * Unshare file descriptors (from parent).
347 if (((p1->p_flag & (P_HADTHREADS|P_SYSTEM)) == P_HADTHREADS) &&
348 (flags & (RFCFDG | RFFDG))) {
350 thread_single_end(p1, SINGLE_BOUNDARY);
357 do_fork(struct thread *td, struct fork_req *fr, struct proc *p2, struct thread *td2,
358 struct vmspace *vm2, struct file *fp_procdesc)
360 struct proc *p1, *pptr;
362 struct filedesc_to_leader *fdtol;
363 struct sigacts *newsigacts;
368 bcopy(&p1->p_startcopy, &p2->p_startcopy,
369 __rangeof(struct proc, p_startcopy, p_endcopy));
370 pargs_hold(p2->p_args);
373 bzero(&p2->p_startzero,
374 __rangeof(struct proc, p_startzero, p_endzero));
376 /* Tell the prison that we exist. */
377 prison_proc_hold(p2->p_ucred->cr_prison);
379 p2->p_state = PRS_NEW; /* protect against others */
380 p2->p_pid = fork_findpid(fr->fr_flags);
381 AUDIT_ARG_PID(p2->p_pid);
383 sx_xlock(&allproc_lock);
384 LIST_INSERT_HEAD(&allproc, p2, p_list);
386 sx_xunlock(&allproc_lock);
388 sx_xlock(PIDHASHLOCK(p2->p_pid));
389 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash);
390 sx_xunlock(PIDHASHLOCK(p2->p_pid));
395 * Malloc things while we don't hold any locks.
397 if (fr->fr_flags & RFSIGSHARE)
400 newsigacts = sigacts_alloc();
405 if (fr->fr_flags & RFCFDG) {
406 fd = fdinit(p1->p_fd, false);
408 } else if (fr->fr_flags & RFFDG) {
409 fd = fdcopy(p1->p_fd);
412 fd = fdshare(p1->p_fd);
413 if (p1->p_fdtol == NULL)
414 p1->p_fdtol = filedesc_to_leader_alloc(NULL, NULL,
416 if ((fr->fr_flags & RFTHREAD) != 0) {
418 * Shared file descriptor table, and shared
422 FILEDESC_XLOCK(p1->p_fd);
423 fdtol->fdl_refcount++;
424 FILEDESC_XUNLOCK(p1->p_fd);
427 * Shared file descriptor table, and different
430 fdtol = filedesc_to_leader_alloc(p1->p_fdtol,
435 * Make a proc table entry for the new process.
436 * Start by zeroing the section of proc that is zero-initialized,
437 * then copy the section that is copied directly from the parent.
443 bzero(&td2->td_startzero,
444 __rangeof(struct thread, td_startzero, td_endzero));
446 bcopy(&td->td_startcopy, &td2->td_startcopy,
447 __rangeof(struct thread, td_startcopy, td_endcopy));
449 bcopy(&p2->p_comm, &td2->td_name, sizeof(td2->td_name));
450 td2->td_sigstk = td->td_sigstk;
451 td2->td_flags = TDF_INMEM;
452 td2->td_lend_user_pri = PRI_MAX;
456 td2->td_vnet_lpush = NULL;
460 * Allow the scheduler to initialize the child.
467 * Duplicate sub-structures as needed.
468 * Increase reference counts on shared objects.
470 p2->p_flag = P_INMEM;
471 p2->p_flag2 = p1->p_flag2 & (P2_ASLR_DISABLE | P2_ASLR_ENABLE |
472 P2_ASLR_IGNSTART | P2_NOTRACE | P2_NOTRACE_EXEC |
473 P2_PROTMAX_ENABLE | P2_PROTMAX_DISABLE | P2_TRAPCAP |
474 P2_STKGAP_DISABLE | P2_STKGAP_DISABLE_EXEC);
475 p2->p_swtick = ticks;
476 if (p1->p_flag & P_PROFIL)
479 if (fr->fr_flags & RFSIGSHARE) {
480 p2->p_sigacts = sigacts_hold(p1->p_sigacts);
482 sigacts_copy(newsigacts, p1->p_sigacts);
483 p2->p_sigacts = newsigacts;
484 if ((fr->fr_flags2 & FR2_DROPSIG_CAUGHT) != 0) {
485 mtx_lock(&p2->p_sigacts->ps_mtx);
487 mtx_unlock(&p2->p_sigacts->ps_mtx);
491 if (fr->fr_flags & RFTSIGZMB)
492 p2->p_sigparent = RFTSIGNUM(fr->fr_flags);
493 else if (fr->fr_flags & RFLINUXTHPN)
494 p2->p_sigparent = SIGUSR1;
496 p2->p_sigparent = SIGCHLD;
498 p2->p_textvp = p1->p_textvp;
502 if (p1->p_flag2 & P2_INHERIT_PROTECTED) {
503 p2->p_flag |= P_PROTECTED;
504 p2->p_flag2 |= P2_INHERIT_PROTECTED;
508 * p_limit is copy-on-write. Bump its refcount.
512 thread_cow_get_proc(td2, p2);
514 pstats_fork(p1->p_stats, p2->p_stats);
519 /* Bump references to the text vnode (for procfs). */
521 vrefact(p2->p_textvp);
524 * Set up linkage for kernel based threading.
526 if ((fr->fr_flags & RFTHREAD) != 0) {
527 mtx_lock(&ppeers_lock);
528 p2->p_peers = p1->p_peers;
530 p2->p_leader = p1->p_leader;
531 mtx_unlock(&ppeers_lock);
532 PROC_LOCK(p1->p_leader);
533 if ((p1->p_leader->p_flag & P_WEXIT) != 0) {
534 PROC_UNLOCK(p1->p_leader);
536 * The task leader is exiting, so process p1 is
537 * going to be killed shortly. Since p1 obviously
538 * isn't dead yet, we know that the leader is either
539 * sending SIGKILL's to all the processes in this
540 * task or is sleeping waiting for all the peers to
541 * exit. We let p1 complete the fork, but we need
542 * to go ahead and kill the new process p2 since
543 * the task leader may not get a chance to send
544 * SIGKILL to it. We leave it on the list so that
545 * the task leader will wait for this new process
549 kern_psignal(p2, SIGKILL);
552 PROC_UNLOCK(p1->p_leader);
558 sx_xlock(&proctree_lock);
559 PGRP_LOCK(p1->p_pgrp);
564 * Preserve some more flags in subprocess. P_PROFIL has already
567 p2->p_flag |= p1->p_flag & P_SUGID;
568 td2->td_pflags |= (td->td_pflags & (TDP_ALTSTACK |
569 TDP_SIGFASTBLOCK)) | TDP_FORKING;
570 SESS_LOCK(p1->p_session);
571 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
572 p2->p_flag |= P_CONTROLT;
573 SESS_UNLOCK(p1->p_session);
574 if (fr->fr_flags & RFPPWAIT)
575 p2->p_flag |= P_PPWAIT;
577 p2->p_pgrp = p1->p_pgrp;
578 LIST_INSERT_AFTER(p1, p2, p_pglist);
579 PGRP_UNLOCK(p1->p_pgrp);
580 LIST_INIT(&p2->p_children);
581 LIST_INIT(&p2->p_orphans);
583 callout_init_mtx(&p2->p_itcallout, &p2->p_mtx, 0);
586 * If PF_FORK is set, the child process inherits the
587 * procfs ioctl flags from its parent.
589 if (p1->p_pfsflags & PF_FORK) {
590 p2->p_stops = p1->p_stops;
591 p2->p_pfsflags = p1->p_pfsflags;
595 * This begins the section where we must prevent the parent
596 * from being swapped.
602 * Attach the new process to its parent.
604 * If RFNOWAIT is set, the newly created process becomes a child
605 * of init. This effectively disassociates the child from the
608 if ((fr->fr_flags & RFNOWAIT) != 0) {
612 p2->p_reaper = (p1->p_treeflag & P_TREE_REAPER) != 0 ?
617 p2->p_oppid = pptr->p_pid;
618 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
619 LIST_INIT(&p2->p_reaplist);
620 LIST_INSERT_HEAD(&p2->p_reaper->p_reaplist, p2, p_reapsibling);
621 if (p2->p_reaper == p1 && p1 != initproc) {
622 p2->p_reapsubtree = p2->p_pid;
623 proc_id_set_cond(PROC_ID_REAP, p2->p_pid);
625 sx_xunlock(&proctree_lock);
627 /* Inform accounting that we have forked. */
628 p2->p_acflag = AFORK;
636 * Finish creating the child process. It will return via a different
637 * execution path later. (ie: directly into user mode)
639 vm_forkproc(td, p2, td2, vm2, fr->fr_flags);
641 if (fr->fr_flags == (RFFDG | RFPROC)) {
643 VM_CNT_ADD(v_forkpages, p2->p_vmspace->vm_dsize +
644 p2->p_vmspace->vm_ssize);
645 } else if (fr->fr_flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
646 VM_CNT_INC(v_vforks);
647 VM_CNT_ADD(v_vforkpages, p2->p_vmspace->vm_dsize +
648 p2->p_vmspace->vm_ssize);
649 } else if (p1 == &proc0) {
650 VM_CNT_INC(v_kthreads);
651 VM_CNT_ADD(v_kthreadpages, p2->p_vmspace->vm_dsize +
652 p2->p_vmspace->vm_ssize);
654 VM_CNT_INC(v_rforks);
655 VM_CNT_ADD(v_rforkpages, p2->p_vmspace->vm_dsize +
656 p2->p_vmspace->vm_ssize);
660 * Associate the process descriptor with the process before anything
661 * can happen that might cause that process to need the descriptor.
662 * However, don't do this until after fork(2) can no longer fail.
664 if (fr->fr_flags & RFPROCDESC)
665 procdesc_new(p2, fr->fr_pd_flags);
668 * Both processes are set up, now check if any loadable modules want
669 * to adjust anything.
671 EVENTHANDLER_DIRECT_INVOKE(process_fork, p1, p2, fr->fr_flags);
674 * Set the child start time and mark the process as being complete.
678 microuptime(&p2->p_stats->p_start);
680 p2->p_state = PRS_NORMAL;
685 * Tell the DTrace fasttrap provider about the new process so that any
686 * tracepoints inherited from the parent can be removed. We have to do
687 * this only after p_state is PRS_NORMAL since the fasttrap module will
688 * use pfind() later on.
690 if ((fr->fr_flags & RFMEM) == 0 && dtrace_fasttrap_fork)
691 dtrace_fasttrap_fork(p1, p2);
693 if (fr->fr_flags & RFPPWAIT) {
694 td->td_pflags |= TDP_RFPPWAIT;
695 td->td_rfppwait_p = p2;
696 td->td_dbgflags |= TDB_VFORK;
701 * Tell any interested parties about the new process.
703 knote_fork(p1->p_klist, p2->p_pid);
706 * Now can be swapped.
710 SDT_PROBE3(proc, , , create, p2, p1, fr->fr_flags);
712 if (fr->fr_flags & RFPROCDESC) {
713 procdesc_finit(p2->p_procdesc, fp_procdesc);
714 fdrop(fp_procdesc, td);
718 * Speculative check for PTRACE_FORK. PTRACE_FORK is not
719 * synced with forks in progress so it is OK if we miss it
722 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
723 sx_xlock(&proctree_lock);
727 * p1->p_ptevents & p1->p_pptr are protected by both
728 * process and proctree locks for modifications,
729 * so owning proctree_lock allows the race-free read.
731 if ((p1->p_ptevents & PTRACE_FORK) != 0) {
733 * Arrange for debugger to receive the fork event.
735 * We can report PL_FLAG_FORKED regardless of
736 * P_FOLLOWFORK settings, but it does not make a sense
739 td->td_dbgflags |= TDB_FORK;
740 td->td_dbg_forked = p2->p_pid;
741 td2->td_dbgflags |= TDB_STOPATFORK;
742 proc_set_traced(p2, true);
744 "do_fork: attaching to new child pid %d: oppid %d",
745 p2->p_pid, p2->p_oppid);
746 proc_reparent(p2, p1->p_pptr, false);
749 sx_xunlock(&proctree_lock);
752 racct_proc_fork_done(p2);
754 if ((fr->fr_flags & RFSTOPPED) == 0) {
755 if (fr->fr_pidp != NULL)
756 *fr->fr_pidp = p2->p_pid;
758 * If RFSTOPPED not requested, make child runnable and
763 sched_add(td2, SRQ_BORING);
770 fork_rfppwait(struct thread *td)
774 MPASS(td->td_pflags & TDP_RFPPWAIT);
778 * Preserve synchronization semantics of vfork. If
779 * waiting for child to exec or exit, fork set
780 * P_PPWAIT on child, and there we sleep on our proc
783 * Do it after the ptracestop() above is finished, to
784 * not block our debugger until child execs or exits
785 * to finish vfork wait.
787 td->td_pflags &= ~TDP_RFPPWAIT;
788 p2 = td->td_rfppwait_p;
791 while (p2->p_flag & P_PPWAIT) {
793 if (thread_suspend_check_needed()) {
795 thread_suspend_check(0);
801 cv_timedwait(&p2->p_pwait, &p2->p_mtx, hz);
805 if (td->td_dbgflags & TDB_VFORK) {
807 if (p->p_ptevents & PTRACE_VFORK)
808 ptracestop(td, SIGTRAP, NULL);
809 td->td_dbgflags &= ~TDB_VFORK;
815 fork1(struct thread *td, struct fork_req *fr)
817 struct proc *p1, *newproc;
821 struct file *fp_procdesc;
822 vm_ooffset_t mem_charged;
823 int error, nprocs_new;
825 static struct timeval lastfail;
828 flags = fr->fr_flags;
829 pages = fr->fr_pages;
831 if ((flags & RFSTOPPED) != 0)
832 MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
834 MPASS(fr->fr_procp == NULL);
836 /* Check for the undefined or unimplemented flags. */
837 if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
840 /* Signal value requires RFTSIGZMB. */
841 if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
844 /* Can't copy and clear. */
845 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
848 /* Check the validity of the signal number. */
849 if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
852 if ((flags & RFPROCDESC) != 0) {
853 /* Can't not create a process yet get a process descriptor. */
854 if ((flags & RFPROC) == 0)
857 /* Must provide a place to put a procdesc if creating one. */
858 if (fr->fr_pd_fd == NULL)
861 /* Check if we are using supported flags. */
862 if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
869 * Here we don't create a new process, but we divorce
870 * certain parts of a process from itself.
872 if ((flags & RFPROC) == 0) {
873 if (fr->fr_procp != NULL)
874 *fr->fr_procp = NULL;
875 else if (fr->fr_pidp != NULL)
877 return (fork_norfproc(td, flags));
885 * Increment the nprocs resource before allocations occur.
886 * Although process entries are dynamically created, we still
887 * keep a global limit on the maximum number we will
888 * create. There are hard-limits as to the number of processes
889 * that can run, established by the KVA and memory usage for
892 * Don't allow a nonprivileged user to use the last ten
893 * processes; don't let root exceed the limit.
895 nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
896 if (nprocs_new >= maxproc - 10) {
897 if (priv_check_cred(td->td_ucred, PRIV_MAXPROC) != 0 ||
898 nprocs_new >= maxproc) {
900 sx_xlock(&allproc_lock);
901 if (ppsratecheck(&lastfail, &curfail, 1)) {
902 printf("maxproc limit exceeded by uid %u "
903 "(pid %d); see tuning(7) and "
905 td->td_ucred->cr_ruid, p1->p_pid);
907 sx_xunlock(&allproc_lock);
913 * If required, create a process descriptor in the parent first; we
914 * will abandon it if something goes wrong. We don't finit() until
917 if (flags & RFPROCDESC) {
918 error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
919 fr->fr_pd_flags, fr->fr_pd_fcaps);
926 pages = kstack_pages;
927 /* Allocate new proc. */
928 newproc = uma_zalloc(proc_zone, M_WAITOK);
929 td2 = FIRST_THREAD_IN_PROC(newproc);
931 td2 = thread_alloc(pages);
936 proc_linkup(newproc, td2);
938 if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
939 if (td2->td_kstack != 0)
940 vm_thread_dispose(td2);
941 if (!thread_alloc_stack(td2, pages)) {
948 if ((flags & RFMEM) == 0) {
949 vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
954 if (!swap_reserve(mem_charged)) {
956 * The swap reservation failed. The accounting
957 * from the entries of the copied vm2 will be
958 * subtracted in vmspace_free(), so force the
961 swap_reserve_force(mem_charged);
969 * XXX: This is ugly; when we copy resource usage, we need to bump
970 * per-cred resource counters.
972 proc_set_cred_init(newproc, crhold(td->td_ucred));
975 * Initialize resource accounting for the child process.
977 error = racct_proc_fork(p1, newproc);
984 mac_proc_init(newproc);
986 newproc->p_klist = knlist_alloc(&newproc->p_mtx);
987 STAILQ_INIT(&newproc->p_ktr);
990 * Increment the count of procs running with this uid. Don't allow
991 * a nonprivileged user to exceed their current limit.
994 if (!chgproccnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_NPROC))) {
995 if (priv_check_cred(cred, PRIV_PROC_LIMIT) != 0)
997 chgproccnt(cred->cr_ruidinfo, 1, 0);
1000 do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
1005 mac_proc_destroy(newproc);
1007 racct_proc_exit(newproc);
1009 crfree(newproc->p_ucred);
1010 newproc->p_ucred = NULL;
1014 uma_zfree(proc_zone, newproc);
1015 if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
1016 fdclose(td, fp_procdesc, *fr->fr_pd_fd);
1017 fdrop(fp_procdesc, td);
1019 atomic_add_int(&nprocs, -1);
1020 pause("fork", hz / 2);
1025 * Handle the return of a child process from fork1(). This function
1026 * is called from the MD fork_trampoline() entry point.
1029 fork_exit(void (*callout)(void *, struct trapframe *), void *arg,
1030 struct trapframe *frame)
1038 KASSERT(p->p_state == PRS_NORMAL, ("executing process is still new"));
1040 CTR4(KTR_PROC, "fork_exit: new thread %p (td_sched %p, pid %d, %s)",
1041 td, td_get_sched(td), p->p_pid, td->td_name);
1043 sched_fork_exit(td);
1045 * Processes normally resume in mi_switch() after being
1046 * cpu_switch()'ed to, but when children start up they arrive here
1047 * instead, so we must do much the same things as mi_switch() would.
1049 if ((dtd = PCPU_GET(deadthread))) {
1050 PCPU_SET(deadthread, NULL);
1056 * cpu_fork_kthread_handler intercepts this function call to
1057 * have this call a non-return function to stay in kernel mode.
1058 * initproc has its own fork handler, but it does return.
1060 KASSERT(callout != NULL, ("NULL callout in fork_exit"));
1061 callout(arg, frame);
1064 * Check if a kernel thread misbehaved and returned from its main
1067 if (p->p_flag & P_KPROC) {
1068 printf("Kernel thread \"%s\" (pid %d) exited prematurely.\n",
1069 td->td_name, p->p_pid);
1072 mtx_assert(&Giant, MA_NOTOWNED);
1074 if (p->p_sysent->sv_schedtail != NULL)
1075 (p->p_sysent->sv_schedtail)(td);
1076 td->td_pflags &= ~TDP_FORKING;
1080 * Simplified back end of syscall(), used when returning from fork()
1081 * directly into user mode. This function is passed in to fork_exit()
1082 * as the first parameter and is called when returning to a new
1086 fork_return(struct thread *td, struct trapframe *frame)
1091 if (td->td_dbgflags & TDB_STOPATFORK) {
1093 if ((p->p_flag & P_TRACED) != 0) {
1095 * Inform the debugger if one is still present.
1097 td->td_dbgflags |= TDB_CHILD | TDB_SCX | TDB_FSTP;
1098 ptracestop(td, SIGSTOP, NULL);
1099 td->td_dbgflags &= ~(TDB_CHILD | TDB_SCX);
1102 * ... otherwise clear the request.
1104 td->td_dbgflags &= ~TDB_STOPATFORK;
1107 } else if (p->p_flag & P_TRACED || td->td_dbgflags & TDB_BORN) {
1109 * This is the start of a new thread in a traced
1110 * process. Report a system call exit event.
1113 td->td_dbgflags |= TDB_SCX;
1114 _STOPEVENT(p, S_SCX, td->td_sa.code);
1115 if ((p->p_ptevents & PTRACE_SCX) != 0 ||
1116 (td->td_dbgflags & TDB_BORN) != 0)
1117 ptracestop(td, SIGTRAP, NULL);
1118 td->td_dbgflags &= ~(TDB_SCX | TDB_BORN);
1125 if (KTRPOINT(td, KTR_SYSRET))
1126 ktrsysret(SYS_fork, 0, 0);