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/thread2.h>
69 #include <sys/signal2.h>
70 #include <sys/spinlock2.h>
71 #include <sys/mplock2.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;
122 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
124 start_forked_proc(lp, p2);
125 uap->sysmsg_fds[0] = p2->p_pid;
126 uap->sysmsg_fds[1] = 0;
136 sys_vfork(struct vfork_args *uap)
138 struct lwp *lp = curthread->td_lwp;
143 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
145 start_forked_proc(lp, p2);
146 uap->sysmsg_fds[0] = p2->p_pid;
147 uap->sysmsg_fds[1] = 0;
154 * Handle rforks. An rfork may (1) operate on the current process without
155 * creating a new, (2) create a new process that shared the current process's
156 * vmspace, signals, and/or descriptors, or (3) create a new process that does
157 * not share these things (normal fork).
159 * Note that we only call start_forked_proc() if a new process is actually
162 * rfork { int flags }
167 sys_rfork(struct rfork_args *uap)
169 struct lwp *lp = curthread->td_lwp;
173 if ((uap->flags & RFKERNELONLY) != 0)
177 error = fork1(lp, uap->flags | RFPGLOCK, &p2);
180 start_forked_proc(lp, p2);
181 uap->sysmsg_fds[0] = p2 ? p2->p_pid : 0;
182 uap->sysmsg_fds[1] = 0;
192 sys_lwp_create(struct lwp_create_args *uap)
194 struct proc *p = curproc;
196 struct lwp_params params;
199 error = copyin(uap->params, ¶ms, sizeof(params));
204 plimit_lwp_fork(p); /* force exclusive access */
205 lp = lwp_fork(curthread->td_lwp, p, RFPROC);
206 error = cpu_prepare_lwp(lp, ¶ms);
207 if (params.tid1 != NULL &&
208 (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
210 if (params.tid2 != NULL &&
211 (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
215 * Now schedule the new lwp.
217 p->p_usched->resetpriority(lp);
219 lp->lwp_stat = LSRUN;
220 p->p_usched->setrunqueue(lp);
227 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
229 /* lwp_dispose expects an exited lwp, and a held proc */
230 lp->lwp_flag |= LWP_WEXIT;
231 lp->lwp_thread->td_flags |= TDF_EXITING;
239 int nprocs = 1; /* process 0 */
242 fork1(struct lwp *lp1, int flags, struct proc **procp)
244 struct proc *p1 = lp1->lwp_proc;
245 struct proc *p2, *pptr;
249 static int curfail = 0;
250 static struct timeval lastfail;
252 struct filedesc_to_leader *fdtol;
254 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
258 * Here we don't create a new process, but we divorce
259 * certain parts of a process from itself.
261 if ((flags & RFPROC) == 0) {
263 * This kind of stunt does not work anymore if
264 * there are native threads (lwps) running
266 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;
295 * Interlock against process group signal delivery. If signals
296 * are pending after the interlock is obtained we have to restart
297 * the system call to process the signals. If we don't the child
298 * can miss a pgsignal (such as ^C) sent during the fork.
300 * We can't use CURSIG() here because it will process any STOPs
301 * and cause the process group lock to be held indefinitely. If
302 * a STOP occurs, the fork will be restarted after the CONT.
306 if ((flags & RFPGLOCK) && (pgrp = p1->p_pgrp) != NULL) {
307 lockmgr(&pgrp->pg_lock, LK_SHARED);
308 if (CURSIG_NOBLOCK(lp1)) {
315 * Although process entries are dynamically created, we still keep
316 * a global limit on the maximum number we will create. Don't allow
317 * a nonprivileged user to use the last ten processes; don't let root
318 * exceed the limit. The variable nprocs is the current number of
319 * processes, maxproc is the limit.
321 uid = lp1->lwp_thread->td_ucred->cr_ruid;
322 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
323 if (ppsratecheck(&lastfail, &curfail, 1))
324 kprintf("maxproc limit exceeded by uid %d, please "
325 "see tuning(7) and login.conf(5).\n", uid);
326 tsleep(&forksleep, 0, "fork", hz / 2);
331 * Increment the nprocs resource before blocking can occur. There
332 * are hard-limits as to the number of processes that can run.
337 * Increment the count of procs running with this uid. Don't allow
338 * a nonprivileged user to exceed their current limit.
340 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
341 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
344 * Back out the process count
347 if (ppsratecheck(&lastfail, &curfail, 1))
348 kprintf("maxproc limit exceeded by uid %d, please "
349 "see tuning(7) and login.conf(5).\n", uid);
350 tsleep(&forksleep, 0, "fork", hz / 2);
355 /* Allocate new proc. */
356 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
359 * Setup linkage for kernel based threading XXX lwp
361 if (flags & RFTHREAD) {
362 p2->p_peers = p1->p_peers;
364 p2->p_leader = p1->p_leader;
369 RB_INIT(&p2->p_lwp_tree);
370 spin_init(&p2->p_spin);
371 p2->p_lasttid = -1; /* first tid will be 0 */
374 * Setting the state to SIDL protects the partially initialized
375 * process once it starts getting hooked into the rest of the system.
378 proc_add_allproc(p2);
381 * Make a proc table entry for the new process.
382 * The whole structure was zeroed above, so copy the section that is
383 * copied directly from the parent.
385 bcopy(&p1->p_startcopy, &p2->p_startcopy,
386 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
389 * Duplicate sub-structures as needed.
390 * Increase reference counts on shared objects.
392 if (p1->p_flag & P_PROFIL)
394 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
395 KKASSERT(p2->p_lock == 0);
397 if (jailed(p2->p_ucred))
398 p2->p_flag |= P_JAILED;
401 p2->p_args->ar_ref++;
403 p2->p_usched = p1->p_usched;
404 /* XXX: verify copy of the secondary iosched stuff */
407 if (flags & RFSIGSHARE) {
408 p2->p_sigacts = p1->p_sigacts;
409 p2->p_sigacts->ps_refcnt++;
411 p2->p_sigacts = (struct sigacts *)kmalloc(sizeof(*p2->p_sigacts),
412 M_SUBPROC, M_WAITOK);
413 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
414 p2->p_sigacts->ps_refcnt = 1;
416 if (flags & RFLINUXTHPN)
417 p2->p_sigparent = SIGUSR1;
419 p2->p_sigparent = SIGCHLD;
421 /* bump references to the text vnode (for procfs) */
422 p2->p_textvp = p1->p_textvp;
426 /* copy namecache handle to the text file */
427 if (p1->p_textnch.mount)
428 cache_copy(&p1->p_textnch, &p2->p_textnch);
431 * Handle file descriptors
433 if (flags & RFCFDG) {
434 p2->p_fd = fdinit(p1);
436 } else if (flags & RFFDG) {
437 p2->p_fd = fdcopy(p1);
440 p2->p_fd = fdshare(p1);
441 if (p1->p_fdtol == NULL)
443 filedesc_to_leader_alloc(NULL,
445 if ((flags & RFTHREAD) != 0) {
447 * Shared file descriptor table and
448 * shared process leaders.
451 fdtol->fdl_refcount++;
454 * Shared file descriptor table, and
455 * different process leaders
457 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
461 p2->p_limit = plimit_fork(p1);
464 * Preserve some more flags in subprocess. P_PROFIL has already
467 p2->p_flag |= p1->p_flag & P_SUGID;
468 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT)
469 p2->p_flag |= P_CONTROLT;
470 if (flags & RFPPWAIT)
471 p2->p_flag |= P_PPWAIT;
474 * Inherit the virtual kernel structure (allows a virtual kernel
475 * to fork to simulate multiple cpus).
478 vkernel_inherit(p1, p2);
481 * Once we are on a pglist we may receive signals. XXX we might
482 * race a ^C being sent to the process group by not receiving it
483 * at all prior to this line.
485 LIST_INSERT_AFTER(p1, p2, p_pglist);
488 * Attach the new process to its parent.
490 * If RFNOWAIT is set, the newly created process becomes a child
491 * of init. This effectively disassociates the child from the
494 if (flags & RFNOWAIT)
499 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
500 LIST_INIT(&p2->p_children);
501 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
502 callout_init(&p2->p_ithandle);
506 * Copy traceflag and tracefile if enabled. If not inherited,
507 * these were zeroed above but we still could have a trace race
508 * so make sure p2's p_tracenode is NULL.
510 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
511 p2->p_traceflag = p1->p_traceflag;
512 p2->p_tracenode = ktrinherit(p1->p_tracenode);
517 * This begins the section where we must prevent the parent
518 * from being swapped.
520 * Gets PRELE'd in the caller in start_forked_proc().
524 vm_fork(p1, p2, flags);
527 * Create the first lwp associated with the new proc.
528 * It will return via a different execution path later, directly
529 * into userland, after it was put on the runq by
530 * start_forked_proc().
532 lwp_fork(lp1, p2, flags);
534 if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
535 mycpu->gd_cnt.v_forks++;
536 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
537 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
538 mycpu->gd_cnt.v_vforks++;
539 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
540 } else if (p1 == &proc0) {
541 mycpu->gd_cnt.v_kthreads++;
542 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
544 mycpu->gd_cnt.v_rforks++;
545 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize + p2->p_vmspace->vm_ssize;
549 * Both processes are set up, now check if any loadable modules want
550 * to adjust anything.
551 * What if they have an error? XXX
553 TAILQ_FOREACH(ep, &fork_list, next) {
554 (*ep->function)(p1, p2, flags);
558 * Set the start time. Note that the process is not runnable. The
559 * caller is responsible for making it runnable.
561 microtime(&p2->p_start);
562 p2->p_acflag = AFORK;
565 * tell any interested parties about the new process
567 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
570 * Return child proc pointer to parent.
575 lockmgr(&pgrp->pg_lock, LK_RELEASE);
580 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
585 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
587 lp->lwp_proc = destproc;
588 lp->lwp_vmspace = destproc->p_vmspace;
589 lp->lwp_stat = LSRUN;
590 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
591 (unsigned) ((caddr_t)&lp->lwp_endcopy -
592 (caddr_t)&lp->lwp_startcopy));
593 lp->lwp_flag |= origlp->lwp_flag & LWP_ALTSTACK;
595 * Set cpbase to the last timeout that occured (not the upcoming
598 * A critical section is required since a timer IPI can update
599 * scheduler specific data.
602 lp->lwp_cpbase = mycpu->gd_schedclock.time -
603 mycpu->gd_schedclock.periodic;
604 destproc->p_usched->heuristic_forking(origlp, lp);
606 lp->lwp_cpumask &= usched_mastermask;
609 * Assign a TID to the lp. Loop until the insert succeeds (returns
612 lp->lwp_tid = destproc->p_lasttid;
614 if (++lp->lwp_tid < 0)
616 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
617 destproc->p_lasttid = lp->lwp_tid;
618 destproc->p_nthreads++;
620 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, -1, 0);
622 td->td_proc = destproc;
624 td->td_switch = cpu_heavy_switch;
625 lwkt_setpri(td, TDPRI_KERN_USER);
626 lwkt_set_comm(td, "%s", destproc->p_comm);
629 * cpu_fork will copy and update the pcb, set up the kernel stack,
630 * and make the child ready to run.
632 cpu_fork(origlp, lp, flags);
633 caps_fork(origlp->lwp_thread, lp->lwp_thread);
634 kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
640 * The next two functionms are general routines to handle adding/deleting
641 * items on the fork callout list.
644 * Take the arguments given and put them onto the fork callout list,
645 * However first make sure that it's not already there.
646 * Returns 0 on success or a standard error number.
649 at_fork(forklist_fn function)
654 /* let the programmer know if he's been stupid */
655 if (rm_at_fork(function)) {
656 kprintf("WARNING: fork callout entry (%p) already present\n",
660 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
661 ep->function = function;
662 TAILQ_INSERT_TAIL(&fork_list, ep, next);
667 * Scan the exit callout list for the given item and remove it..
668 * Returns the number of items removed (0 or 1)
671 rm_at_fork(forklist_fn function)
675 TAILQ_FOREACH(ep, &fork_list, next) {
676 if (ep->function == function) {
677 TAILQ_REMOVE(&fork_list, ep, next);
686 * Add a forked process to the run queue after any remaining setup, such
687 * as setting the fork handler, has been completed.
690 start_forked_proc(struct lwp *lp1, struct proc *p2)
692 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
695 * Move from SIDL to RUN queue, and activate the process's thread.
696 * Activation of the thread effectively makes the process "a"
697 * current process, so we do not setrunqueue().
699 * YYY setrunqueue works here but we should clean up the trampoline
700 * code so we just schedule the LWKT thread and let the trampoline
701 * deal with the userland scheduler on return to userland.
703 KASSERT(p2->p_stat == SIDL,
704 ("cannot start forked process, bad status: %p", p2));
705 p2->p_usched->resetpriority(lp2);
707 p2->p_stat = SACTIVE;
708 lp2->lwp_stat = LSRUN;
709 p2->p_usched->setrunqueue(lp2);
713 * Now can be swapped.
715 PRELE(lp1->lwp_proc);
718 * Preserve synchronization semantics of vfork. If waiting for
719 * child to exec or exit, set P_PPWAIT on child, and sleep on our
720 * proc (in case of exit).
722 while (p2->p_flag & P_PPWAIT)
723 tsleep(lp1->lwp_proc, 0, "ppwait", 0);