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. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)kern_fork.c 8.6 (Berkeley) 4/8/94
35 * $FreeBSD: src/sys/kern/kern_fork.c,v 1.72.2.14 2003/06/26 04:15:10 silby Exp $
38 #include "opt_ktrace.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/sysproto.h>
43 #include <sys/filedesc.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/malloc.h>
48 #include <sys/resourcevar.h>
49 #include <sys/vnode.h>
51 #include <sys/ktrace.h>
52 #include <sys/unistd.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_extern.h>
61 #include <sys/vmmeter.h>
62 #include <sys/refcount.h>
63 #include <sys/thread2.h>
64 #include <sys/signal2.h>
65 #include <sys/spinlock2.h>
67 #include <sys/dsched.h>
69 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
72 * These are the stuctures used to create a callout list for things to do
73 * when forking a process
77 TAILQ_ENTRY(forklist) next;
80 TAILQ_HEAD(forklist_head, forklist);
81 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
83 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
85 int forksleep; /* Place for fork1() to sleep on. */
88 * Red-Black tree support for LWPs
92 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
94 if (lp1->lwp_tid < lp2->lwp_tid)
96 if (lp1->lwp_tid > lp2->lwp_tid)
101 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);
107 sys_fork(struct fork_args *uap)
109 struct lwp *lp = curthread->td_lwp;
113 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
116 start_forked_proc(lp, p2);
117 uap->sysmsg_fds[0] = p2->p_pid;
118 uap->sysmsg_fds[1] = 0;
125 * vfork() system call
128 sys_vfork(struct vfork_args *uap)
130 struct lwp *lp = curthread->td_lwp;
134 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
137 start_forked_proc(lp, p2);
138 uap->sysmsg_fds[0] = p2->p_pid;
139 uap->sysmsg_fds[1] = 0;
146 * Handle rforks. An rfork may (1) operate on the current process without
147 * creating a new, (2) create a new process that shared the current process's
148 * vmspace, signals, and/or descriptors, or (3) create a new process that does
149 * not share these things (normal fork).
151 * Note that we only call start_forked_proc() if a new process is actually
154 * rfork { int flags }
157 sys_rfork(struct rfork_args *uap)
159 struct lwp *lp = curthread->td_lwp;
163 if ((uap->flags & RFKERNELONLY) != 0)
166 error = fork1(lp, uap->flags | RFPGLOCK, &p2);
170 start_forked_proc(lp, p2);
171 uap->sysmsg_fds[0] = p2->p_pid;
172 uap->sysmsg_fds[1] = 0;
175 uap->sysmsg_fds[0] = 0;
176 uap->sysmsg_fds[1] = 0;
183 * Low level thread create used by pthreads.
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);
203 if (params.tid1 != NULL &&
204 (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
206 if (params.tid2 != NULL &&
207 (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
211 * Now schedule the new lwp.
213 p->p_usched->resetpriority(lp);
215 lp->lwp_stat = LSRUN;
216 p->p_usched->setrunqueue(lp);
218 lwkt_reltoken(&p->p_token);
223 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
225 /* lwp_dispose expects an exited lwp, and a held proc */
226 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
227 lp->lwp_thread->td_flags |= TDF_EXITING;
228 lwkt_remove_tdallq(lp->lwp_thread);
230 biosched_done(lp->lwp_thread);
231 dsched_exit_thread(lp->lwp_thread);
233 lwkt_reltoken(&p->p_token);
238 int nprocs = 1; /* process 0 */
241 fork1(struct lwp *lp1, int flags, struct proc **procp)
243 struct proc *p1 = lp1->lwp_proc;
250 static int curfail = 0;
251 static struct timeval lastfail;
253 struct filedesc_to_leader *fdtol;
255 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
258 lwkt_gettoken(&p1->p_token);
263 * Here we don't create a new process, but we divorce
264 * certain parts of a process from itself.
266 if ((flags & RFPROC) == 0) {
268 * This kind of stunt does not work anymore if
269 * there are native threads (lwps) running
271 if (p1->p_nthreads != 1) {
276 vm_fork(p1, 0, flags);
279 * Close all file descriptors.
281 if (flags & RFCFDG) {
282 struct filedesc *fdtmp;
288 * Unshare file descriptors (from parent.)
291 if (p1->p_fd->fd_refcnt > 1) {
292 struct filedesc *newfd;
293 error = fdcopy(p1, &newfd);
307 * Interlock against process group signal delivery. If signals
308 * are pending after the interlock is obtained we have to restart
309 * the system call to process the signals. If we don't the child
310 * can miss a pgsignal (such as ^C) sent during the fork.
312 * We can't use CURSIG() here because it will process any STOPs
313 * and cause the process group lock to be held indefinitely. If
314 * a STOP occurs, the fork will be restarted after the CONT.
317 if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) {
319 lockmgr(&plkgrp->pg_lock, LK_SHARED);
320 if (CURSIG_NOBLOCK(lp1)) {
327 * Although process entries are dynamically created, we still keep
328 * a global limit on the maximum number we will create. Don't allow
329 * a nonprivileged user to use the last ten processes; don't let root
330 * exceed the limit. The variable nprocs is the current number of
331 * processes, maxproc is the limit.
333 uid = lp1->lwp_thread->td_ucred->cr_ruid;
334 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
335 if (ppsratecheck(&lastfail, &curfail, 1))
336 kprintf("maxproc limit exceeded by uid %d, please "
337 "see tuning(7) and login.conf(5).\n", uid);
338 tsleep(&forksleep, 0, "fork", hz / 2);
344 * Increment the nprocs resource before blocking can occur. There
345 * are hard-limits as to the number of processes that can run.
347 atomic_add_int(&nprocs, 1);
350 * Increment the count of procs running with this uid. Don't allow
351 * a nonprivileged user to exceed their current limit.
353 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
354 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
357 * Back out the process count
359 atomic_add_int(&nprocs, -1);
360 if (ppsratecheck(&lastfail, &curfail, 1))
361 kprintf("maxproc limit exceeded by uid %d, please "
362 "see tuning(7) and login.conf(5).\n", uid);
363 tsleep(&forksleep, 0, "fork", hz / 2);
369 * Allocate a new process, don't get fancy: zero the structure.
371 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
374 * Core initialization. SIDL is a safety state that protects the
375 * partially initialized process once it starts getting hooked
376 * into system structures and becomes addressable.
378 * We must be sure to acquire p2->p_token as well, we must hold it
379 * once the process is on the allproc list to avoid things such
380 * as competing modifications to p_flags.
382 mycpu->gd_forkid += ncpus;
383 p2->p_forkid = mycpu->gd_forkid + mycpu->gd_cpuid;
384 p2->p_lasttid = -1; /* first tid will be 0 */
387 RB_INIT(&p2->p_lwp_tree);
388 spin_init(&p2->p_spin, "procfork1");
389 lwkt_token_init(&p2->p_token, "proc");
390 lwkt_gettoken(&p2->p_token);
393 * Setup linkage for kernel based threading XXX lwp. Also add the
394 * process to the allproclist.
396 * The process structure is addressable after this point.
398 if (flags & RFTHREAD) {
399 p2->p_peers = p1->p_peers;
401 p2->p_leader = p1->p_leader;
405 proc_add_allproc(p2);
408 * Initialize the section which is copied verbatim from the parent.
410 bcopy(&p1->p_startcopy, &p2->p_startcopy,
411 ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
414 * Duplicate sub-structures as needed. Increase reference counts
417 * NOTE: because we are now on the allproc list it is possible for
418 * other consumers to gain temporary references to p2
419 * (p2->p_lock can change).
421 if (p1->p_flags & P_PROFIL)
423 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
425 if (jailed(p2->p_ucred))
426 p2->p_flags |= P_JAILED;
429 refcount_acquire(&p2->p_args->ar_ref);
431 p2->p_usched = p1->p_usched;
432 /* XXX: verify copy of the secondary iosched stuff */
435 if (flags & RFSIGSHARE) {
436 p2->p_sigacts = p1->p_sigacts;
437 refcount_acquire(&p2->p_sigacts->ps_refcnt);
439 p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts),
440 M_SUBPROC, M_WAITOK);
441 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
442 refcount_init(&p2->p_sigacts->ps_refcnt, 1);
444 if (flags & RFLINUXTHPN)
445 p2->p_sigparent = SIGUSR1;
447 p2->p_sigparent = SIGCHLD;
449 /* bump references to the text vnode (for procfs) */
450 p2->p_textvp = p1->p_textvp;
454 /* copy namecache handle to the text file */
455 if (p1->p_textnch.mount)
456 cache_copy(&p1->p_textnch, &p2->p_textnch);
459 * Handle file descriptors
461 if (flags & RFCFDG) {
462 p2->p_fd = fdinit(p1);
464 } else if (flags & RFFDG) {
465 error = fdcopy(p1, &p2->p_fd);
472 p2->p_fd = fdshare(p1);
473 if (p1->p_fdtol == NULL) {
474 p1->p_fdtol = filedesc_to_leader_alloc(NULL,
477 if ((flags & RFTHREAD) != 0) {
479 * Shared file descriptor table and
480 * shared process leaders.
483 fdtol->fdl_refcount++;
486 * Shared file descriptor table, and
487 * different process leaders
489 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
493 p2->p_limit = plimit_fork(p1);
496 * Preserve some more flags in subprocess. P_PROFIL has already
499 p2->p_flags |= p1->p_flags & P_SUGID;
500 if (p1->p_session->s_ttyvp != NULL && (p1->p_flags & P_CONTROLT))
501 p2->p_flags |= P_CONTROLT;
502 if (flags & RFPPWAIT) {
503 p2->p_flags |= P_PPWAIT;
505 p1->p_upmap->invfork = 1;
510 * Inherit the virtual kernel structure (allows a virtual kernel
511 * to fork to simulate multiple cpus).
514 vkernel_inherit(p1, p2);
517 * Once we are on a pglist we may receive signals. XXX we might
518 * race a ^C being sent to the process group by not receiving it
519 * at all prior to this line.
522 lwkt_gettoken(&p1grp->pg_token);
523 LIST_INSERT_AFTER(p1, p2, p_pglist);
524 lwkt_reltoken(&p1grp->pg_token);
527 * Attach the new process to its parent.
529 * If RFNOWAIT is set, the newly created process becomes a child
530 * of init. This effectively disassociates the child from the
533 if (flags & RFNOWAIT)
538 LIST_INIT(&p2->p_children);
540 lwkt_gettoken(&pptr->p_token);
541 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
542 lwkt_reltoken(&pptr->p_token);
544 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
545 callout_init_mp(&p2->p_ithandle);
549 * Copy traceflag and tracefile if enabled. If not inherited,
550 * these were zeroed above but we still could have a trace race
551 * so make sure p2's p_tracenode is NULL.
553 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
554 p2->p_traceflag = p1->p_traceflag;
555 p2->p_tracenode = ktrinherit(p1->p_tracenode);
560 * This begins the section where we must prevent the parent
561 * from being swapped.
563 * Gets PRELE'd in the caller in start_forked_proc().
567 vm_fork(p1, p2, flags);
570 * Create the first lwp associated with the new proc.
571 * It will return via a different execution path later, directly
572 * into userland, after it was put on the runq by
573 * start_forked_proc().
575 lwp_fork(lp1, p2, flags);
577 if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
578 mycpu->gd_cnt.v_forks++;
579 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize +
580 p2->p_vmspace->vm_ssize;
581 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
582 mycpu->gd_cnt.v_vforks++;
583 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
584 p2->p_vmspace->vm_ssize;
585 } else if (p1 == &proc0) {
586 mycpu->gd_cnt.v_kthreads++;
587 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
588 p2->p_vmspace->vm_ssize;
590 mycpu->gd_cnt.v_rforks++;
591 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
592 p2->p_vmspace->vm_ssize;
596 * Both processes are set up, now check if any loadable modules want
597 * to adjust anything.
598 * What if they have an error? XXX
600 TAILQ_FOREACH(ep, &fork_list, next) {
601 (*ep->function)(p1, p2, flags);
605 * Set the start time. Note that the process is not runnable. The
606 * caller is responsible for making it runnable.
608 microtime(&p2->p_start);
609 p2->p_acflag = AFORK;
612 * tell any interested parties about the new process
614 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
617 * Return child proc pointer to parent.
623 lwkt_reltoken(&p2->p_token);
624 lwkt_reltoken(&p1->p_token);
626 lockmgr(&plkgrp->pg_lock, LK_RELEASE);
633 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
635 globaldata_t gd = mycpu;
639 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
641 lp->lwp_proc = destproc;
642 lp->lwp_vmspace = destproc->p_vmspace;
643 lp->lwp_stat = LSRUN;
644 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
645 (unsigned) ((caddr_t)&lp->lwp_endcopy -
646 (caddr_t)&lp->lwp_startcopy));
647 lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK;
649 * Set cpbase to the last timeout that occured (not the upcoming
652 * A critical section is required since a timer IPI can update
653 * scheduler specific data.
656 lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic;
657 destproc->p_usched->heuristic_forking(origlp, lp);
659 CPUMASK_ANDMASK(lp->lwp_cpumask, usched_mastermask);
660 lwkt_token_init(&lp->lwp_token, "lwp_token");
661 spin_init(&lp->lwp_spin, "lwptoken");
664 * Assign the thread to the current cpu to begin with so we
667 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0);
669 td->td_ucred = crhold(destproc->p_ucred);
670 td->td_proc = destproc;
672 td->td_switch = cpu_heavy_switch;
673 #ifdef NO_LWKT_SPLIT_USERPRI
674 lwkt_setpri(td, TDPRI_USER_NORM);
676 lwkt_setpri(td, TDPRI_KERN_USER);
678 lwkt_set_comm(td, "%s", destproc->p_comm);
681 * cpu_fork will copy and update the pcb, set up the kernel stack,
682 * and make the child ready to run.
684 cpu_fork(origlp, lp, flags);
685 kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
688 * Assign a TID to the lp. Loop until the insert succeeds (returns
691 lp->lwp_tid = destproc->p_lasttid;
693 if (++lp->lwp_tid < 0)
695 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
696 destproc->p_lasttid = lp->lwp_tid;
697 destproc->p_nthreads++;
700 * This flag is set and never cleared. It means that the process
701 * was threaded at some point. Used to improve exit performance.
703 destproc->p_flags |= P_MAYBETHREADED;
709 * The next two functionms are general routines to handle adding/deleting
710 * items on the fork callout list.
713 * Take the arguments given and put them onto the fork callout list,
714 * However first make sure that it's not already there.
715 * Returns 0 on success or a standard error number.
718 at_fork(forklist_fn function)
723 /* let the programmer know if he's been stupid */
724 if (rm_at_fork(function)) {
725 kprintf("WARNING: fork callout entry (%p) already present\n",
729 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
730 ep->function = function;
731 TAILQ_INSERT_TAIL(&fork_list, ep, next);
736 * Scan the exit callout list for the given item and remove it..
737 * Returns the number of items removed (0 or 1)
740 rm_at_fork(forklist_fn function)
744 TAILQ_FOREACH(ep, &fork_list, next) {
745 if (ep->function == function) {
746 TAILQ_REMOVE(&fork_list, ep, next);
755 * Add a forked process to the run queue after any remaining setup, such
756 * as setting the fork handler, has been completed.
758 * p2 is held by the caller.
761 start_forked_proc(struct lwp *lp1, struct proc *p2)
763 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
767 * Move from SIDL to RUN queue, and activate the process's thread.
768 * Activation of the thread effectively makes the process "a"
769 * current process, so we do not setrunqueue().
771 * YYY setrunqueue works here but we should clean up the trampoline
772 * code so we just schedule the LWKT thread and let the trampoline
773 * deal with the userland scheduler on return to userland.
775 KASSERT(p2->p_stat == SIDL,
776 ("cannot start forked process, bad status: %p", p2));
777 p2->p_usched->resetpriority(lp2);
779 p2->p_stat = SACTIVE;
780 lp2->lwp_stat = LSRUN;
781 p2->p_usched->setrunqueue(lp2);
785 * Now can be swapped.
787 PRELE(lp1->lwp_proc);
790 * Preserve synchronization semantics of vfork. P_PPWAIT is set in
791 * the child until it has retired the parent's resources. The parent
792 * must wait for the flag to be cleared by the child.
794 * Interlock the flag/tsleep with atomic ops to avoid unnecessary
797 * XXX Is this use of an atomic op on a field that is not normally
798 * manipulated with atomic ops ok?
800 while ((pflags = p2->p_flags) & P_PPWAIT) {
802 tsleep_interlock(lp1->lwp_proc, 0);
803 if (atomic_cmpset_int(&p2->p_flags, pflags, pflags))
804 tsleep(lp1->lwp_proc, PINTERLOCKED, "ppwait", 0);