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 1.67 2007/03/13 00:18:59 corecode 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>
66 #include <vm/vm_zone.h>
68 #include <sys/vmmeter.h>
69 #include <sys/thread2.h>
70 #include <sys/signal2.h>
72 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
75 * These are the stuctures used to create a callout list for things to do
76 * when forking a process
80 TAILQ_ENTRY(forklist) next;
83 TAILQ_HEAD(forklist_head, forklist);
84 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
86 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
88 int forksleep; /* Place for fork1() to sleep on. */
92 sys_fork(struct fork_args *uap)
94 struct lwp *lp = curthread->td_lwp;
98 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
100 start_forked_proc(lp, p2);
101 uap->sysmsg_fds[0] = p2->p_pid;
102 uap->sysmsg_fds[1] = 0;
109 sys_vfork(struct vfork_args *uap)
111 struct lwp *lp = curthread->td_lwp;
115 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
117 start_forked_proc(lp, p2);
118 uap->sysmsg_fds[0] = p2->p_pid;
119 uap->sysmsg_fds[1] = 0;
125 * Handle rforks. An rfork may (1) operate on the current process without
126 * creating a new, (2) create a new process that shared the current process's
127 * vmspace, signals, and/or descriptors, or (3) create a new process that does
128 * not share these things (normal fork).
130 * Note that we only call start_forked_proc() if a new process is actually
133 * rfork { int flags }
136 sys_rfork(struct rfork_args *uap)
138 struct lwp *lp = curthread->td_lwp;
142 if ((uap->flags & RFKERNELONLY) != 0)
145 error = fork1(lp, uap->flags | RFPGLOCK, &p2);
148 start_forked_proc(lp, p2);
149 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
150 uap->sysmsg_fds[1] = 0;
156 sys_lwp_create(struct lwp_create_args *uap)
158 struct proc *p = curproc;
160 struct lwp_params params;
163 error = copyin(uap->params, ¶ms, sizeof(params));
167 lp = lwp_fork(curthread->td_lwp, p, RFPROC);
168 error = cpu_prepare_lwp(lp, ¶ms);
169 if (params.tid1 != NULL &&
170 (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
172 if (params.tid2 != NULL &&
173 (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
177 * Now schedule the new lwp.
179 p->p_usched->resetpriority(lp);
181 lp->lwp_stat = LSRUN;
182 p->p_usched->setrunqueue(lp);
189 LIST_REMOVE(lp, lwp_list);
190 /* lwp_dispose expects a exited lwp */
191 lp->lwp_thread->td_flags = TDF_EXITING;
197 int nprocs = 1; /* process 0 */
200 fork1(struct lwp *lp1, int flags, struct proc **procp)
202 struct proc *p1 = lp1->lwp_proc;
203 struct proc *p2, *pptr;
207 static int curfail = 0;
208 static struct timeval lastfail;
210 struct filedesc_to_leader *fdtol;
212 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
216 * Here we don't create a new process, but we divorce
217 * certain parts of a process from itself.
219 if ((flags & RFPROC) == 0) {
222 * This kind of stunt does not work anymore if
223 * there are native threads (lwps) running
225 if (p1->p_nthreads != 1)
228 vm_fork(p1, 0, flags);
231 * Close all file descriptors.
233 if (flags & RFCFDG) {
234 struct filedesc *fdtmp;
241 * Unshare file descriptors (from parent.)
244 if (p1->p_fd->fd_refcnt > 1) {
245 struct filedesc *newfd;
256 * Interlock against process group signal delivery. If signals
257 * are pending after the interlock is obtained we have to restart
258 * the system call to process the signals. If we don't the child
259 * can miss a pgsignal (such as ^C) sent during the fork.
261 * We can't use CURSIG() here because it will process any STOPs
262 * and cause the process group lock to be held indefinitely. If
263 * a STOP occurs, the fork will be restarted after the CONT.
267 if ((flags & RFPGLOCK) && (pgrp = p1->p_pgrp) != NULL) {
268 lockmgr(&pgrp->pg_lock, LK_SHARED);
276 * Although process entries are dynamically created, we still keep
277 * a global limit on the maximum number we will create. Don't allow
278 * a nonprivileged user to use the last ten processes; don't let root
279 * exceed the limit. The variable nprocs is the current number of
280 * processes, maxproc is the limit.
282 uid = p1->p_ucred->cr_ruid;
283 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
284 if (ppsratecheck(&lastfail, &curfail, 1))
285 kprintf("maxproc limit exceeded by uid %d, please "
286 "see tuning(7) and login.conf(5).\n", uid);
287 tsleep(&forksleep, 0, "fork", hz / 2);
292 * Increment the nprocs resource before blocking can occur. There
293 * are hard-limits as to the number of processes that can run.
298 * Increment the count of procs running with this uid. Don't allow
299 * a nonprivileged user to exceed their current limit.
301 ok = chgproccnt(p1->p_ucred->cr_ruidinfo, 1,
302 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
305 * Back out the process count
308 if (ppsratecheck(&lastfail, &curfail, 1))
309 kprintf("maxproc limit exceeded by uid %d, please "
310 "see tuning(7) and login.conf(5).\n", uid);
311 tsleep(&forksleep, 0, "fork", hz / 2);
316 /* Allocate new proc. */
317 p2 = zalloc(proc_zone);
318 bzero(p2, sizeof(*p2));
321 * Setup linkage for kernel based threading XXX lwp
323 if (flags & RFTHREAD) {
324 p2->p_peers = p1->p_peers;
326 p2->p_leader = p1->p_leader;
331 LIST_INIT(&p2->p_lwps);
334 * Setting the state to SIDL protects the partially initialized
335 * process once it starts getting hooked into the rest of the system.
338 proc_add_allproc(p2);
341 * Make a proc table entry for the new process.
342 * The whole structure was zeroed above, so copy the section that is
343 * copied directly from the parent.
345 bcopy(&p1->p_startcopy, &p2->p_startcopy,
346 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
349 * Duplicate sub-structures as needed.
350 * Increase reference counts on shared objects.
352 if (p1->p_flag & P_PROFIL)
354 p2->p_ucred = crhold(p1->p_ucred);
356 if (jailed(p2->p_ucred))
357 p2->p_flag |= P_JAILED;
360 p2->p_args->ar_ref++;
362 p2->p_usched = p1->p_usched;
364 if (flags & RFSIGSHARE) {
365 p2->p_sigacts = p1->p_sigacts;
366 p2->p_sigacts->ps_refcnt++;
368 p2->p_sigacts = (struct sigacts *)kmalloc(sizeof(*p2->p_sigacts),
369 M_SUBPROC, M_WAITOK);
370 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
371 p2->p_sigacts->ps_refcnt = 1;
373 if (flags & RFLINUXTHPN)
374 p2->p_sigparent = SIGUSR1;
376 p2->p_sigparent = SIGCHLD;
378 /* bump references to the text vnode (for procfs) */
379 p2->p_textvp = p1->p_textvp;
384 * Handle file descriptors
386 if (flags & RFCFDG) {
387 p2->p_fd = fdinit(p1);
389 } else if (flags & RFFDG) {
390 p2->p_fd = fdcopy(p1);
393 p2->p_fd = fdshare(p1);
394 if (p1->p_fdtol == NULL)
396 filedesc_to_leader_alloc(NULL,
398 if ((flags & RFTHREAD) != 0) {
400 * Shared file descriptor table and
401 * shared process leaders.
404 fdtol->fdl_refcount++;
407 * Shared file descriptor table, and
408 * different process leaders
410 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
414 p2->p_limit = plimit_fork(p1->p_limit);
417 * Preserve some more flags in subprocess. P_PROFIL has already
420 p2->p_flag |= p1->p_flag & P_SUGID;
421 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
422 p2->p_flag |= P_CONTROLT;
423 if (flags & RFPPWAIT)
424 p2->p_flag |= P_PPWAIT;
427 * Inherit the virtual kernel structure (allows a virtual kernel
428 * to fork to simulate multiple cpus).
431 vkernel_inherit(p1, p2);
434 * Once we are on a pglist we may receive signals. XXX we might
435 * race a ^C being sent to the process group by not receiving it
436 * at all prior to this line.
438 LIST_INSERT_AFTER(p1, p2, p_pglist);
441 * Attach the new process to its parent.
443 * If RFNOWAIT is set, the newly created process becomes a child
444 * of init. This effectively disassociates the child from the
447 if (flags & RFNOWAIT)
452 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
453 LIST_INIT(&p2->p_children);
454 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
455 callout_init(&p2->p_ithandle);
459 * Copy traceflag and tracefile if enabled. If not inherited,
460 * these were zeroed above but we still could have a trace race
461 * so make sure p2's p_tracenode is NULL.
463 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
464 p2->p_traceflag = p1->p_traceflag;
465 p2->p_tracenode = ktrinherit(p1->p_tracenode);
470 * This begins the section where we must prevent the parent
471 * from being swapped.
473 * Gets PRELE'd in the caller in start_forked_proc().
477 vm_fork(p1, p2, flags);
480 * Create the first lwp associated with the new proc.
481 * It will return via a different execution path later, directly
482 * into userland, after it was put on the runq by
483 * start_forked_proc().
485 lwp_fork(lp1, p2, flags);
487 if (flags == (RFFDG | RFPROC)) {
488 mycpu->gd_cnt.v_forks++;
489 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
490 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM)) {
491 mycpu->gd_cnt.v_vforks++;
492 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
493 } else if (p1 == &proc0) {
494 mycpu->gd_cnt.v_kthreads++;
495 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
497 mycpu->gd_cnt.v_rforks++;
498 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
502 * Both processes are set up, now check if any loadable modules want
503 * to adjust anything.
504 * What if they have an error? XXX
506 TAILQ_FOREACH(ep, &fork_list, next) {
507 (*ep->function)(p1, p2, flags);
511 * Set the start time. Note that the process is not runnable. The
512 * caller is responsible for making it runnable.
514 microtime(&p2->p_start);
515 p2->p_acflag = AFORK;
518 * tell any interested parties about the new process
520 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
523 * Return child proc pointer to parent.
528 lockmgr(&pgrp->pg_lock, LK_RELEASE);
533 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
540 * We need to prevent wrap-around collisions.
541 * Until we have a nice tid allocator, we need to
542 * start searching for free tids once we wrap around.
544 * XXX give me a nicer allocator
546 if (destproc->p_lasttid + 1 <= 0) {
549 FOREACH_LWP_IN_PROC(lp, destproc) {
550 if (lp->lwp_tid != tid)
552 /* tids match, search next. */
555 * Wait -- the whole tid space is depleted?
559 panic("lwp_fork: All tids depleted?!");
562 /* When we come here, the tid is not occupied */
564 tid = destproc->p_lasttid++;
567 lp = zalloc(lwp_zone);
568 bzero(lp, sizeof(*lp));
569 lp->lwp_proc = destproc;
571 LIST_INSERT_HEAD(&destproc->p_lwps, lp, lwp_list);
572 destproc->p_nthreads++;
573 lp->lwp_stat = LSRUN;
574 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
575 (unsigned) ((caddr_t)&lp->lwp_endcopy -
576 (caddr_t)&lp->lwp_startcopy));
577 lp->lwp_flag |= origlp->lwp_flag & LWP_ALTSTACK;
579 * Set cpbase to the last timeout that occured (not the upcoming
582 * A critical section is required since a timer IPI can update
583 * scheduler specific data.
586 lp->lwp_cpbase = mycpu->gd_schedclock.time -
587 mycpu->gd_schedclock.periodic;
588 destproc->p_usched->heuristic_forking(origlp, lp);
591 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0);
593 td->td_proc = destproc;
595 td->td_switch = cpu_heavy_switch;
597 KKASSERT(td->td_mpcount == 1);
599 lwkt_setpri(td, TDPRI_KERN_USER);
600 lwkt_set_comm(td, "%s", destproc->p_comm);
603 * cpu_fork will copy and update the pcb, set up the kernel stack,
604 * and make the child ready to run.
606 cpu_fork(origlp, lp, flags);
607 caps_fork(origlp->lwp_thread, lp->lwp_thread);
613 * The next two functionms are general routines to handle adding/deleting
614 * items on the fork callout list.
617 * Take the arguments given and put them onto the fork callout list,
618 * However first make sure that it's not already there.
619 * Returns 0 on success or a standard error number.
622 at_fork(forklist_fn function)
627 /* let the programmer know if he's been stupid */
628 if (rm_at_fork(function)) {
629 kprintf("WARNING: fork callout entry (%p) already present\n",
633 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
634 ep->function = function;
635 TAILQ_INSERT_TAIL(&fork_list, ep, next);
640 * Scan the exit callout list for the given item and remove it..
641 * Returns the number of items removed (0 or 1)
644 rm_at_fork(forklist_fn function)
648 TAILQ_FOREACH(ep, &fork_list, next) {
649 if (ep->function == function) {
650 TAILQ_REMOVE(&fork_list, ep, next);
659 * Add a forked process to the run queue after any remaining setup, such
660 * as setting the fork handler, has been completed.
663 start_forked_proc(struct lwp *lp1, struct proc *p2)
665 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
668 * Move from SIDL to RUN queue, and activate the process's thread.
669 * Activation of the thread effectively makes the process "a"
670 * current process, so we do not setrunqueue().
672 * YYY setrunqueue works here but we should clean up the trampoline
673 * code so we just schedule the LWKT thread and let the trampoline
674 * deal with the userland scheduler on return to userland.
676 KASSERT(p2->p_stat == SIDL,
677 ("cannot start forked process, bad status: %p", p2));
678 p2->p_usched->resetpriority(lp2);
680 p2->p_stat = SACTIVE;
681 lp2->lwp_stat = LSRUN;
682 p2->p_usched->setrunqueue(lp2);
686 * Now can be swapped.
688 PRELE(lp1->lwp_proc);
691 * Preserve synchronization semantics of vfork. If waiting for
692 * child to exec or exit, set P_PPWAIT on child, and sleep on our
693 * proc (in case of exit).
695 while (p2->p_flag & P_PPWAIT)
696 tsleep(lp1->lwp_proc, 0, "ppwait", 0);