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 $
40 * $DragonFly: src/sys/kern/kern_fork.c,v 1.77 2008/05/18 20:02:02 nth Exp $
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/sysproto.h>
48 #include <sys/filedesc.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
53 #include <sys/resourcevar.h>
54 #include <sys/vnode.h>
56 #include <sys/ktrace.h>
57 #include <sys/unistd.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_extern.h>
67 #include <sys/vmmeter.h>
68 #include <sys/refcount.h>
69 #include <sys/thread2.h>
70 #include <sys/signal2.h>
71 #include <sys/spinlock2.h>
73 #include <sys/dsched.h>
75 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
78 * These are the stuctures used to create a callout list for things to do
79 * when forking a process
83 TAILQ_ENTRY(forklist) next;
86 TAILQ_HEAD(forklist_head, forklist);
87 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
89 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
91 int forksleep; /* Place for fork1() to sleep on. */
94 * Red-Black tree support for LWPs
98 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
100 if (lp1->lwp_tid < lp2->lwp_tid)
102 if (lp1->lwp_tid > lp2->lwp_tid)
107 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);
115 sys_fork(struct fork_args *uap)
117 struct lwp *lp = curthread->td_lwp;
121 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
123 start_forked_proc(lp, p2);
124 uap->sysmsg_fds[0] = p2->p_pid;
125 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);
142 start_forked_proc(lp, p2);
143 uap->sysmsg_fds[0] = p2->p_pid;
144 uap->sysmsg_fds[1] = 0;
150 * Handle rforks. An rfork may (1) operate on the current process without
151 * creating a new, (2) create a new process that shared the current process's
152 * vmspace, signals, and/or descriptors, or (3) create a new process that does
153 * not share these things (normal fork).
155 * Note that we only call start_forked_proc() if a new process is actually
158 * rfork { int flags }
163 sys_rfork(struct rfork_args *uap)
165 struct lwp *lp = curthread->td_lwp;
169 if ((uap->flags & RFKERNELONLY) != 0)
172 error = fork1(lp, uap->flags | RFPGLOCK, &p2);
175 start_forked_proc(lp, p2);
176 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
177 uap->sysmsg_fds[1] = 0;
186 sys_lwp_create(struct lwp_create_args *uap)
188 struct proc *p = curproc;
190 struct lwp_params params;
193 error = copyin(uap->params, ¶ms, sizeof(params));
197 lwkt_gettoken(&p->p_token);
198 plimit_lwp_fork(p); /* force exclusive access */
199 lp = lwp_fork(curthread->td_lwp, p, RFPROC);
200 error = cpu_prepare_lwp(lp, ¶ms);
201 if (params.tid1 != NULL &&
202 (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
204 if (params.tid2 != NULL &&
205 (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
209 * Now schedule the new lwp.
211 p->p_usched->resetpriority(lp);
213 lp->lwp_stat = LSRUN;
214 p->p_usched->setrunqueue(lp);
216 lwkt_reltoken(&p->p_token);
221 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
223 /* lwp_dispose expects an exited lwp, and a held proc */
224 lp->lwp_flag |= LWP_WEXIT;
225 lp->lwp_thread->td_flags |= TDF_EXITING;
228 lwkt_reltoken(&p->p_token);
233 int nprocs = 1; /* process 0 */
236 fork1(struct lwp *lp1, int flags, struct proc **procp)
238 struct proc *p1 = lp1->lwp_proc;
239 struct proc *p2, *pptr;
244 static int curfail = 0;
245 static struct timeval lastfail;
247 struct filedesc_to_leader *fdtol;
249 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
252 lwkt_gettoken(&p1->p_token);
256 * Here we don't create a new process, but we divorce
257 * certain parts of a process from itself.
259 if ((flags & RFPROC) == 0) {
261 * This kind of stunt does not work anymore if
262 * there are native threads (lwps) running
264 if (p1->p_nthreads != 1) {
269 vm_fork(p1, 0, flags);
272 * Close all file descriptors.
274 if (flags & RFCFDG) {
275 struct filedesc *fdtmp;
281 * Unshare file descriptors (from parent.)
284 if (p1->p_fd->fd_refcnt > 1) {
285 struct filedesc *newfd;
286 error = fdcopy(p1, &newfd);
300 * Interlock against process group signal delivery. If signals
301 * are pending after the interlock is obtained we have to restart
302 * the system call to process the signals. If we don't the child
303 * can miss a pgsignal (such as ^C) sent during the fork.
305 * We can't use CURSIG() here because it will process any STOPs
306 * and cause the process group lock to be held indefinitely. If
307 * a STOP occurs, the fork will be restarted after the CONT.
310 if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) {
312 lockmgr(&plkgrp->pg_lock, LK_SHARED);
313 if (CURSIG_NOBLOCK(lp1)) {
320 * Although process entries are dynamically created, we still keep
321 * a global limit on the maximum number we will create. Don't allow
322 * a nonprivileged user to use the last ten processes; don't let root
323 * exceed the limit. The variable nprocs is the current number of
324 * processes, maxproc is the limit.
326 uid = lp1->lwp_thread->td_ucred->cr_ruid;
327 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
328 if (ppsratecheck(&lastfail, &curfail, 1))
329 kprintf("maxproc limit exceeded by uid %d, please "
330 "see tuning(7) and login.conf(5).\n", uid);
331 tsleep(&forksleep, 0, "fork", hz / 2);
336 * Increment the nprocs resource before blocking can occur. There
337 * are hard-limits as to the number of processes that can run.
342 * Increment the count of procs running with this uid. Don't allow
343 * a nonprivileged user to exceed their current limit.
345 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
346 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
349 * Back out the process count
352 if (ppsratecheck(&lastfail, &curfail, 1))
353 kprintf("maxproc limit exceeded by uid %d, please "
354 "see tuning(7) and login.conf(5).\n", uid);
355 tsleep(&forksleep, 0, "fork", hz / 2);
360 /* Allocate new proc. */
361 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
364 * Setup linkage for kernel based threading XXX lwp
366 if (flags & RFTHREAD) {
367 p2->p_peers = p1->p_peers;
369 p2->p_leader = p1->p_leader;
374 RB_INIT(&p2->p_lwp_tree);
375 spin_init(&p2->p_spin);
376 lwkt_token_init(&p2->p_token, "iproc");
377 p2->p_lasttid = -1; /* first tid will be 0 */
380 * Setting the state to SIDL protects the partially initialized
381 * process once it starts getting hooked into the rest of the system.
384 proc_add_allproc(p2);
387 * Make a proc table entry for the new process.
388 * The whole structure was zeroed above, so copy the section that is
389 * copied directly from the parent.
391 bcopy(&p1->p_startcopy, &p2->p_startcopy,
392 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
395 * Duplicate sub-structures as needed. Increase reference counts
398 * NOTE: because we are now on the allproc list it is possible for
399 * other consumers to gain temporary references to p2
400 * (p2->p_lock can change).
402 if (p1->p_flag & P_PROFIL)
404 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
406 if (jailed(p2->p_ucred))
407 p2->p_flag |= P_JAILED;
410 refcount_acquire(&p2->p_args->ar_ref);
412 p2->p_usched = p1->p_usched;
413 /* XXX: verify copy of the secondary iosched stuff */
416 if (flags & RFSIGSHARE) {
417 p2->p_sigacts = p1->p_sigacts;
418 refcount_acquire(&p2->p_sigacts->ps_refcnt);
420 p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts),
421 M_SUBPROC, M_WAITOK);
422 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
423 refcount_init(&p2->p_sigacts->ps_refcnt, 1);
425 if (flags & RFLINUXTHPN)
426 p2->p_sigparent = SIGUSR1;
428 p2->p_sigparent = SIGCHLD;
430 /* bump references to the text vnode (for procfs) */
431 p2->p_textvp = p1->p_textvp;
435 /* copy namecache handle to the text file */
436 if (p1->p_textnch.mount)
437 cache_copy(&p1->p_textnch, &p2->p_textnch);
440 * Handle file descriptors
442 if (flags & RFCFDG) {
443 p2->p_fd = fdinit(p1);
445 } else if (flags & RFFDG) {
446 error = fdcopy(p1, &p2->p_fd);
453 p2->p_fd = fdshare(p1);
454 if (p1->p_fdtol == NULL) {
455 lwkt_gettoken(&p1->p_token);
457 filedesc_to_leader_alloc(NULL,
459 lwkt_reltoken(&p1->p_token);
461 if ((flags & RFTHREAD) != 0) {
463 * Shared file descriptor table and
464 * shared process leaders.
467 fdtol->fdl_refcount++;
470 * Shared file descriptor table, and
471 * different process leaders
473 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
477 p2->p_limit = plimit_fork(p1);
480 * Preserve some more flags in subprocess. P_PROFIL has already
483 p2->p_flag |= p1->p_flag & P_SUGID;
484 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
485 p2->p_flag |= P_CONTROLT;
486 if (flags & RFPPWAIT)
487 p2->p_flag |= P_PPWAIT;
490 * Inherit the virtual kernel structure (allows a virtual kernel
491 * to fork to simulate multiple cpus).
494 vkernel_inherit(p1, p2);
497 * Once we are on a pglist we may receive signals. XXX we might
498 * race a ^C being sent to the process group by not receiving it
499 * at all prior to this line.
502 lwkt_gettoken(&p1grp->pg_token);
503 LIST_INSERT_AFTER(p1, p2, p_pglist);
504 lwkt_reltoken(&p1grp->pg_token);
507 * Attach the new process to its parent.
509 * If RFNOWAIT is set, the newly created process becomes a child
510 * of init. This effectively disassociates the child from the
513 if (flags & RFNOWAIT)
518 LIST_INIT(&p2->p_children);
520 lwkt_gettoken(&pptr->p_token);
521 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
522 lwkt_reltoken(&pptr->p_token);
524 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
525 callout_init(&p2->p_ithandle);
529 * Copy traceflag and tracefile if enabled. If not inherited,
530 * these were zeroed above but we still could have a trace race
531 * so make sure p2's p_tracenode is NULL.
533 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
534 p2->p_traceflag = p1->p_traceflag;
535 p2->p_tracenode = ktrinherit(p1->p_tracenode);
540 * This begins the section where we must prevent the parent
541 * from being swapped.
543 * Gets PRELE'd in the caller in start_forked_proc().
547 vm_fork(p1, p2, flags);
550 * Create the first lwp associated with the new proc.
551 * It will return via a different execution path later, directly
552 * into userland, after it was put on the runq by
553 * start_forked_proc().
555 lwp_fork(lp1, p2, flags);
557 if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
558 mycpu->gd_cnt.v_forks++;
559 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
560 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
561 mycpu->gd_cnt.v_vforks++;
562 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
563 } else if (p1 == &proc0) {
564 mycpu->gd_cnt.v_kthreads++;
565 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
567 mycpu->gd_cnt.v_rforks++;
568 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
572 * Both processes are set up, now check if any loadable modules want
573 * to adjust anything.
574 * What if they have an error? XXX
576 TAILQ_FOREACH(ep, &fork_list, next) {
577 (*ep->function)(p1, p2, flags);
581 * Set the start time. Note that the process is not runnable. The
582 * caller is responsible for making it runnable.
584 microtime(&p2->p_start);
585 p2->p_acflag = AFORK;
588 * tell any interested parties about the new process
590 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
593 * Return child proc pointer to parent.
598 lwkt_reltoken(&p1->p_token);
600 lockmgr(&plkgrp->pg_lock, LK_RELEASE);
607 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
612 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
614 lp->lwp_proc = destproc;
615 lp->lwp_vmspace = destproc->p_vmspace;
616 lp->lwp_stat = LSRUN;
617 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
618 (unsigned) ((caddr_t)&lp->lwp_endcopy -
619 (caddr_t)&lp->lwp_startcopy));
620 lp->lwp_flag |= origlp->lwp_flag & LWP_ALTSTACK;
622 * Set cpbase to the last timeout that occured (not the upcoming
625 * A critical section is required since a timer IPI can update
626 * scheduler specific data.
629 lp->lwp_cpbase = mycpu->gd_schedclock.time -
630 mycpu->gd_schedclock.periodic;
631 destproc->p_usched->heuristic_forking(origlp, lp);
633 lp->lwp_cpumask &= usched_mastermask;
635 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0);
637 td->td_proc = destproc;
639 td->td_switch = cpu_heavy_switch;
640 lwkt_setpri(td, TDPRI_KERN_USER);
641 lwkt_set_comm(td, "%s", destproc->p_comm);
644 * cpu_fork will copy and update the pcb, set up the kernel stack,
645 * and make the child ready to run.
647 cpu_fork(origlp, lp, flags);
648 caps_fork(origlp->lwp_thread, lp->lwp_thread);
649 kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
652 * Assign a TID to the lp. Loop until the insert succeeds (returns
655 lp->lwp_tid = destproc->p_lasttid;
657 if (++lp->lwp_tid < 0)
659 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
660 destproc->p_lasttid = lp->lwp_tid;
661 destproc->p_nthreads++;
668 * The next two functionms are general routines to handle adding/deleting
669 * items on the fork callout list.
672 * Take the arguments given and put them onto the fork callout list,
673 * However first make sure that it's not already there.
674 * Returns 0 on success or a standard error number.
677 at_fork(forklist_fn function)
682 /* let the programmer know if he's been stupid */
683 if (rm_at_fork(function)) {
684 kprintf("WARNING: fork callout entry (%p) already present\n",
688 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
689 ep->function = function;
690 TAILQ_INSERT_TAIL(&fork_list, ep, next);
695 * Scan the exit callout list for the given item and remove it..
696 * Returns the number of items removed (0 or 1)
699 rm_at_fork(forklist_fn function)
703 TAILQ_FOREACH(ep, &fork_list, next) {
704 if (ep->function == function) {
705 TAILQ_REMOVE(&fork_list, ep, next);
714 * Add a forked process to the run queue after any remaining setup, such
715 * as setting the fork handler, has been completed.
718 start_forked_proc(struct lwp *lp1, struct proc *p2)
720 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
723 * Move from SIDL to RUN queue, and activate the process's thread.
724 * Activation of the thread effectively makes the process "a"
725 * current process, so we do not setrunqueue().
727 * YYY setrunqueue works here but we should clean up the trampoline
728 * code so we just schedule the LWKT thread and let the trampoline
729 * deal with the userland scheduler on return to userland.
731 KASSERT(p2->p_stat == SIDL,
732 ("cannot start forked process, bad status: %p", p2));
733 p2->p_usched->resetpriority(lp2);
735 p2->p_stat = SACTIVE;
736 lp2->lwp_stat = LSRUN;
737 p2->p_usched->setrunqueue(lp2);
741 * Now can be swapped.
743 PRELE(lp1->lwp_proc);
746 * Preserve synchronization semantics of vfork. If waiting for
747 * child to exec or exit, set P_PPWAIT on child, and sleep on our
748 * proc (in case of exit).
750 while (p2->p_flag & P_PPWAIT)
751 tsleep(lp1->lwp_proc, 0, "ppwait", 0);