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. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
39 * $FreeBSD: src/sys/kern/kern_fork.c,v 1.72.2.14 2003/06/26 04:15:10 silby Exp $
42 #include "opt_ktrace.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/sysproto.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
52 #include <sys/resourcevar.h>
53 #include <sys/vnode.h>
55 #include <sys/ktrace.h>
56 #include <sys/unistd.h>
62 #include <vm/vm_map.h>
63 #include <vm/vm_extern.h>
65 #include <sys/vmmeter.h>
66 #include <sys/refcount.h>
67 #include <sys/thread2.h>
68 #include <sys/signal2.h>
69 #include <sys/spinlock2.h>
71 #include <sys/dsched.h>
73 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
76 * These are the stuctures used to create a callout list for things to do
77 * when forking a process
81 TAILQ_ENTRY(forklist) next;
84 TAILQ_HEAD(forklist_head, forklist);
85 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
87 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
89 int forksleep; /* Place for fork1() to sleep on. */
92 * Red-Black tree support for LWPs
96 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
98 if (lp1->lwp_tid < lp2->lwp_tid)
100 if (lp1->lwp_tid > lp2->lwp_tid)
105 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);
113 sys_fork(struct fork_args *uap)
115 struct lwp *lp = curthread->td_lwp;
119 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
122 start_forked_proc(lp, p2);
123 uap->sysmsg_fds[0] = p2->p_pid;
124 uap->sysmsg_fds[1] = 0;
134 sys_vfork(struct vfork_args *uap)
136 struct lwp *lp = curthread->td_lwp;
140 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
143 start_forked_proc(lp, p2);
144 uap->sysmsg_fds[0] = p2->p_pid;
145 uap->sysmsg_fds[1] = 0;
152 * Handle rforks. An rfork may (1) operate on the current process without
153 * creating a new, (2) create a new process that shared the current process's
154 * vmspace, signals, and/or descriptors, or (3) create a new process that does
155 * not share these things (normal fork).
157 * Note that we only call start_forked_proc() if a new process is actually
160 * rfork { int flags }
165 sys_rfork(struct rfork_args *uap)
167 struct lwp *lp = curthread->td_lwp;
171 if ((uap->flags & RFKERNELONLY) != 0)
174 error = fork1(lp, uap->flags | RFPGLOCK, &p2);
178 start_forked_proc(lp, p2);
179 uap->sysmsg_fds[0] = p2->p_pid;
180 uap->sysmsg_fds[1] = 0;
183 uap->sysmsg_fds[0] = 0;
184 uap->sysmsg_fds[1] = 0;
194 sys_lwp_create(struct lwp_create_args *uap)
196 struct proc *p = curproc;
198 struct lwp_params params;
201 error = copyin(uap->params, ¶ms, sizeof(params));
205 lwkt_gettoken(&p->p_token);
206 plimit_lwp_fork(p); /* force exclusive access */
207 lp = lwp_fork(curthread->td_lwp, p, RFPROC);
208 error = cpu_prepare_lwp(lp, ¶ms);
211 if (params.tid1 != NULL &&
212 (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
214 if (params.tid2 != NULL &&
215 (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
219 * Now schedule the new lwp.
221 p->p_usched->resetpriority(lp);
223 lp->lwp_stat = LSRUN;
224 p->p_usched->setrunqueue(lp);
226 lwkt_reltoken(&p->p_token);
231 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
233 /* lwp_dispose expects an exited lwp, and a held proc */
234 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
235 lp->lwp_thread->td_flags |= TDF_EXITING;
236 lwkt_remove_tdallq(lp->lwp_thread);
238 biosched_done(lp->lwp_thread);
239 dsched_exit_thread(lp->lwp_thread);
241 lwkt_reltoken(&p->p_token);
246 int nprocs = 1; /* process 0 */
249 fork1(struct lwp *lp1, int flags, struct proc **procp)
251 struct proc *p1 = lp1->lwp_proc;
258 static int curfail = 0;
259 static struct timeval lastfail;
261 struct filedesc_to_leader *fdtol;
263 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
266 lwkt_gettoken(&p1->p_token);
271 * Here we don't create a new process, but we divorce
272 * certain parts of a process from itself.
274 if ((flags & RFPROC) == 0) {
276 * This kind of stunt does not work anymore if
277 * there are native threads (lwps) running
279 if (p1->p_nthreads != 1) {
284 vm_fork(p1, 0, flags);
287 * Close all file descriptors.
289 if (flags & RFCFDG) {
290 struct filedesc *fdtmp;
296 * Unshare file descriptors (from parent.)
299 if (p1->p_fd->fd_refcnt > 1) {
300 struct filedesc *newfd;
301 error = fdcopy(p1, &newfd);
315 * Interlock against process group signal delivery. If signals
316 * are pending after the interlock is obtained we have to restart
317 * the system call to process the signals. If we don't the child
318 * can miss a pgsignal (such as ^C) sent during the fork.
320 * We can't use CURSIG() here because it will process any STOPs
321 * and cause the process group lock to be held indefinitely. If
322 * a STOP occurs, the fork will be restarted after the CONT.
325 if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) {
327 lockmgr(&plkgrp->pg_lock, LK_SHARED);
328 if (CURSIG_NOBLOCK(lp1)) {
335 * Although process entries are dynamically created, we still keep
336 * a global limit on the maximum number we will create. Don't allow
337 * a nonprivileged user to use the last ten processes; don't let root
338 * exceed the limit. The variable nprocs is the current number of
339 * processes, maxproc is the limit.
341 uid = lp1->lwp_thread->td_ucred->cr_ruid;
342 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
343 if (ppsratecheck(&lastfail, &curfail, 1))
344 kprintf("maxproc limit exceeded by uid %d, please "
345 "see tuning(7) and login.conf(5).\n", uid);
346 tsleep(&forksleep, 0, "fork", hz / 2);
352 * Increment the nprocs resource before blocking can occur. There
353 * are hard-limits as to the number of processes that can run.
355 atomic_add_int(&nprocs, 1);
358 * Increment the count of procs running with this uid. Don't allow
359 * a nonprivileged user to exceed their current limit.
361 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
362 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
365 * Back out the process count
367 atomic_add_int(&nprocs, -1);
368 if (ppsratecheck(&lastfail, &curfail, 1))
369 kprintf("maxproc limit exceeded by uid %d, please "
370 "see tuning(7) and login.conf(5).\n", uid);
371 tsleep(&forksleep, 0, "fork", hz / 2);
377 * Allocate a new process, don't get fancy: zero the structure.
379 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
382 * Core initialization. SIDL is a safety state that protects the
383 * partially initialized process once it starts getting hooked
384 * into system structures and becomes addressable.
386 * We must be sure to acquire p2->p_token as well, we must hold it
387 * once the process is on the allproc list to avoid things such
388 * as competing modifications to p_flags.
390 p2->p_lasttid = -1; /* first tid will be 0 */
393 RB_INIT(&p2->p_lwp_tree);
394 spin_init(&p2->p_spin);
395 lwkt_token_init(&p2->p_token, "proc");
396 lwkt_gettoken(&p2->p_token);
399 * Setup linkage for kernel based threading XXX lwp. Also add the
400 * process to the allproclist.
402 * The process structure is addressable after this point.
404 if (flags & RFTHREAD) {
405 p2->p_peers = p1->p_peers;
407 p2->p_leader = p1->p_leader;
411 proc_add_allproc(p2);
414 * Initialize the section which is copied verbatim from the parent.
416 bcopy(&p1->p_startcopy, &p2->p_startcopy,
417 ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
420 * Duplicate sub-structures as needed. Increase reference counts
423 * NOTE: because we are now on the allproc list it is possible for
424 * other consumers to gain temporary references to p2
425 * (p2->p_lock can change).
427 if (p1->p_flags & P_PROFIL)
429 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
431 if (jailed(p2->p_ucred))
432 p2->p_flags |= P_JAILED;
435 refcount_acquire(&p2->p_args->ar_ref);
437 p2->p_usched = p1->p_usched;
438 /* XXX: verify copy of the secondary iosched stuff */
441 if (flags & RFSIGSHARE) {
442 p2->p_sigacts = p1->p_sigacts;
443 refcount_acquire(&p2->p_sigacts->ps_refcnt);
445 p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts),
446 M_SUBPROC, M_WAITOK);
447 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
448 refcount_init(&p2->p_sigacts->ps_refcnt, 1);
450 if (flags & RFLINUXTHPN)
451 p2->p_sigparent = SIGUSR1;
453 p2->p_sigparent = SIGCHLD;
455 /* bump references to the text vnode (for procfs) */
456 p2->p_textvp = p1->p_textvp;
460 /* copy namecache handle to the text file */
461 if (p1->p_textnch.mount)
462 cache_copy(&p1->p_textnch, &p2->p_textnch);
465 * Handle file descriptors
467 if (flags & RFCFDG) {
468 p2->p_fd = fdinit(p1);
470 } else if (flags & RFFDG) {
471 error = fdcopy(p1, &p2->p_fd);
478 p2->p_fd = fdshare(p1);
479 if (p1->p_fdtol == NULL) {
480 p1->p_fdtol = filedesc_to_leader_alloc(NULL,
483 if ((flags & RFTHREAD) != 0) {
485 * Shared file descriptor table and
486 * shared process leaders.
489 fdtol->fdl_refcount++;
492 * Shared file descriptor table, and
493 * different process leaders
495 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
499 p2->p_limit = plimit_fork(p1);
502 * Preserve some more flags in subprocess. P_PROFIL has already
505 p2->p_flags |= p1->p_flags & P_SUGID;
506 if (p1->p_session->s_ttyvp != NULL && (p1->p_flags & P_CONTROLT))
507 p2->p_flags |= P_CONTROLT;
508 if (flags & RFPPWAIT)
509 p2->p_flags |= P_PPWAIT;
512 * Inherit the virtual kernel structure (allows a virtual kernel
513 * to fork to simulate multiple cpus).
516 vkernel_inherit(p1, p2);
519 * Once we are on a pglist we may receive signals. XXX we might
520 * race a ^C being sent to the process group by not receiving it
521 * at all prior to this line.
524 lwkt_gettoken(&p1grp->pg_token);
525 LIST_INSERT_AFTER(p1, p2, p_pglist);
526 lwkt_reltoken(&p1grp->pg_token);
529 * Attach the new process to its parent.
531 * If RFNOWAIT is set, the newly created process becomes a child
532 * of init. This effectively disassociates the child from the
535 if (flags & RFNOWAIT)
540 LIST_INIT(&p2->p_children);
542 lwkt_gettoken(&pptr->p_token);
543 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
544 lwkt_reltoken(&pptr->p_token);
546 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
547 callout_init_mp(&p2->p_ithandle);
551 * Copy traceflag and tracefile if enabled. If not inherited,
552 * these were zeroed above but we still could have a trace race
553 * so make sure p2's p_tracenode is NULL.
555 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
556 p2->p_traceflag = p1->p_traceflag;
557 p2->p_tracenode = ktrinherit(p1->p_tracenode);
562 * This begins the section where we must prevent the parent
563 * from being swapped.
565 * Gets PRELE'd in the caller in start_forked_proc().
569 vm_fork(p1, p2, flags);
572 * Create the first lwp associated with the new proc.
573 * It will return via a different execution path later, directly
574 * into userland, after it was put on the runq by
575 * start_forked_proc().
577 lwp_fork(lp1, p2, flags);
579 if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
580 mycpu->gd_cnt.v_forks++;
581 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize +
582 p2->p_vmspace->vm_ssize;
583 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
584 mycpu->gd_cnt.v_vforks++;
585 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
586 p2->p_vmspace->vm_ssize;
587 } else if (p1 == &proc0) {
588 mycpu->gd_cnt.v_kthreads++;
589 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
590 p2->p_vmspace->vm_ssize;
592 mycpu->gd_cnt.v_rforks++;
593 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
594 p2->p_vmspace->vm_ssize;
598 * Both processes are set up, now check if any loadable modules want
599 * to adjust anything.
600 * What if they have an error? XXX
602 TAILQ_FOREACH(ep, &fork_list, next) {
603 (*ep->function)(p1, p2, flags);
607 * Set the start time. Note that the process is not runnable. The
608 * caller is responsible for making it runnable.
610 microtime(&p2->p_start);
611 p2->p_acflag = AFORK;
614 * tell any interested parties about the new process
616 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
619 * Return child proc pointer to parent.
625 lwkt_reltoken(&p2->p_token);
626 lwkt_reltoken(&p1->p_token);
628 lockmgr(&plkgrp->pg_lock, LK_RELEASE);
635 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
637 globaldata_t gd = mycpu;
641 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
643 lp->lwp_proc = destproc;
644 lp->lwp_vmspace = destproc->p_vmspace;
645 lp->lwp_stat = LSRUN;
646 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
647 (unsigned) ((caddr_t)&lp->lwp_endcopy -
648 (caddr_t)&lp->lwp_startcopy));
649 lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK;
651 * Set cpbase to the last timeout that occured (not the upcoming
654 * A critical section is required since a timer IPI can update
655 * scheduler specific data.
658 lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic;
659 destproc->p_usched->heuristic_forking(origlp, lp);
661 lp->lwp_cpumask &= usched_mastermask;
662 lwkt_token_init(&lp->lwp_token, "lwp_token");
663 spin_init(&lp->lwp_spin);
666 * Assign the thread to the current cpu to begin with so we
669 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0);
671 td->td_proc = destproc;
673 td->td_switch = cpu_heavy_switch;
674 #ifdef NO_LWKT_SPLIT_USERPRI
675 lwkt_setpri(td, TDPRI_USER_NORM);
677 lwkt_setpri(td, TDPRI_KERN_USER);
679 lwkt_set_comm(td, "%s", destproc->p_comm);
682 * cpu_fork will copy and update the pcb, set up the kernel stack,
683 * and make the child ready to run.
685 cpu_fork(origlp, lp, flags);
686 kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
689 * Assign a TID to the lp. Loop until the insert succeeds (returns
692 lp->lwp_tid = destproc->p_lasttid;
694 if (++lp->lwp_tid < 0)
696 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
697 destproc->p_lasttid = lp->lwp_tid;
698 destproc->p_nthreads++;
704 * The next two functionms are general routines to handle adding/deleting
705 * items on the fork callout list.
708 * Take the arguments given and put them onto the fork callout list,
709 * However first make sure that it's not already there.
710 * Returns 0 on success or a standard error number.
713 at_fork(forklist_fn function)
718 /* let the programmer know if he's been stupid */
719 if (rm_at_fork(function)) {
720 kprintf("WARNING: fork callout entry (%p) already present\n",
724 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
725 ep->function = function;
726 TAILQ_INSERT_TAIL(&fork_list, ep, next);
731 * Scan the exit callout list for the given item and remove it..
732 * Returns the number of items removed (0 or 1)
735 rm_at_fork(forklist_fn function)
739 TAILQ_FOREACH(ep, &fork_list, next) {
740 if (ep->function == function) {
741 TAILQ_REMOVE(&fork_list, ep, next);
750 * Add a forked process to the run queue after any remaining setup, such
751 * as setting the fork handler, has been completed.
753 * p2 is held by the caller.
756 start_forked_proc(struct lwp *lp1, struct proc *p2)
758 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
761 * Move from SIDL to RUN queue, and activate the process's thread.
762 * Activation of the thread effectively makes the process "a"
763 * current process, so we do not setrunqueue().
765 * YYY setrunqueue works here but we should clean up the trampoline
766 * code so we just schedule the LWKT thread and let the trampoline
767 * deal with the userland scheduler on return to userland.
769 KASSERT(p2->p_stat == SIDL,
770 ("cannot start forked process, bad status: %p", p2));
771 p2->p_usched->resetpriority(lp2);
773 p2->p_stat = SACTIVE;
774 lp2->lwp_stat = LSRUN;
775 p2->p_usched->setrunqueue(lp2);
779 * Now can be swapped.
781 PRELE(lp1->lwp_proc);
784 * Preserve synchronization semantics of vfork. If waiting for
785 * child to exec or exit, set P_PPWAIT on child, and sleep on our
786 * proc (in case of exec or exit).
788 * We must hold our p_token to interlock the flag/tsleep
790 lwkt_gettoken(&p2->p_token);
791 while (p2->p_flags & P_PPWAIT)
792 tsleep(lp1->lwp_proc, 0, "ppwait", 0);
793 lwkt_reltoken(&p2->p_token);