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);
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;
142 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)
176 error = fork1(lp, uap->flags | RFPGLOCK, &p2);
180 start_forked_proc(lp, p2);
181 uap->sysmsg_fds[0] = p2->p_pid;
182 uap->sysmsg_fds[1] = 0;
185 uap->sysmsg_fds[0] = 0;
186 uap->sysmsg_fds[1] = 0;
196 sys_lwp_create(struct lwp_create_args *uap)
198 struct proc *p = curproc;
200 struct lwp_params params;
203 error = copyin(uap->params, ¶ms, sizeof(params));
207 lwkt_gettoken(&p->p_token);
208 plimit_lwp_fork(p); /* force exclusive access */
209 lp = lwp_fork(curthread->td_lwp, p, RFPROC);
210 error = cpu_prepare_lwp(lp, ¶ms);
213 if (params.tid1 != NULL &&
214 (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
216 if (params.tid2 != NULL &&
217 (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
221 * Now schedule the new lwp.
223 p->p_usched->resetpriority(lp);
225 lp->lwp_stat = LSRUN;
226 p->p_usched->setrunqueue(lp);
228 lwkt_reltoken(&p->p_token);
233 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
235 /* lwp_dispose expects an exited lwp, and a held proc */
236 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
237 lp->lwp_thread->td_flags |= TDF_EXITING;
238 lwkt_remove_tdallq(lp->lwp_thread);
240 biosched_done(lp->lwp_thread);
241 dsched_exit_thread(lp->lwp_thread);
243 lwkt_reltoken(&p->p_token);
248 int nprocs = 1; /* process 0 */
251 fork1(struct lwp *lp1, int flags, struct proc **procp)
253 struct proc *p1 = lp1->lwp_proc;
260 static int curfail = 0;
261 static struct timeval lastfail;
263 struct filedesc_to_leader *fdtol;
265 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
268 lwkt_gettoken(&p1->p_token);
273 * Here we don't create a new process, but we divorce
274 * certain parts of a process from itself.
276 if ((flags & RFPROC) == 0) {
278 * This kind of stunt does not work anymore if
279 * there are native threads (lwps) running
281 if (p1->p_nthreads != 1) {
286 vm_fork(p1, 0, flags);
289 * Close all file descriptors.
291 if (flags & RFCFDG) {
292 struct filedesc *fdtmp;
298 * Unshare file descriptors (from parent.)
301 if (p1->p_fd->fd_refcnt > 1) {
302 struct filedesc *newfd;
303 error = fdcopy(p1, &newfd);
317 * Interlock against process group signal delivery. If signals
318 * are pending after the interlock is obtained we have to restart
319 * the system call to process the signals. If we don't the child
320 * can miss a pgsignal (such as ^C) sent during the fork.
322 * We can't use CURSIG() here because it will process any STOPs
323 * and cause the process group lock to be held indefinitely. If
324 * a STOP occurs, the fork will be restarted after the CONT.
327 if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) {
329 lockmgr(&plkgrp->pg_lock, LK_SHARED);
330 if (CURSIG_NOBLOCK(lp1)) {
337 * Although process entries are dynamically created, we still keep
338 * a global limit on the maximum number we will create. Don't allow
339 * a nonprivileged user to use the last ten processes; don't let root
340 * exceed the limit. The variable nprocs is the current number of
341 * processes, maxproc is the limit.
343 uid = lp1->lwp_thread->td_ucred->cr_ruid;
344 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
345 if (ppsratecheck(&lastfail, &curfail, 1))
346 kprintf("maxproc limit exceeded by uid %d, please "
347 "see tuning(7) and login.conf(5).\n", uid);
348 tsleep(&forksleep, 0, "fork", hz / 2);
354 * Increment the nprocs resource before blocking can occur. There
355 * are hard-limits as to the number of processes that can run.
357 atomic_add_int(&nprocs, 1);
360 * Increment the count of procs running with this uid. Don't allow
361 * a nonprivileged user to exceed their current limit.
363 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
364 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
367 * Back out the process count
369 atomic_add_int(&nprocs, -1);
370 if (ppsratecheck(&lastfail, &curfail, 1))
371 kprintf("maxproc limit exceeded by uid %d, please "
372 "see tuning(7) and login.conf(5).\n", uid);
373 tsleep(&forksleep, 0, "fork", hz / 2);
379 * Allocate a new process, don't get fancy: zero the structure.
381 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
384 * Core initialization. SIDL is a safety state that protects the
385 * partially initialized process once it starts getting hooked
386 * into system structures and becomes addressable.
388 * We must be sure to acquire p2->p_token as well, we must hold it
389 * once the process is on the allproc list to avoid things such
390 * as competing modifications to p_flags.
392 p2->p_lasttid = -1; /* first tid will be 0 */
395 RB_INIT(&p2->p_lwp_tree);
396 spin_init(&p2->p_spin);
397 lwkt_token_init(&p2->p_token, "proc");
398 lwkt_gettoken(&p2->p_token);
401 * Setup linkage for kernel based threading XXX lwp. Also add the
402 * process to the allproclist.
404 * The process structure is addressable after this point.
406 if (flags & RFTHREAD) {
407 p2->p_peers = p1->p_peers;
409 p2->p_leader = p1->p_leader;
413 proc_add_allproc(p2);
416 * Initialize the section which is copied verbatim from the parent.
418 bcopy(&p1->p_startcopy, &p2->p_startcopy,
419 ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
422 * Duplicate sub-structures as needed. Increase reference counts
425 * NOTE: because we are now on the allproc list it is possible for
426 * other consumers to gain temporary references to p2
427 * (p2->p_lock can change).
429 if (p1->p_flags & P_PROFIL)
431 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
433 if (jailed(p2->p_ucred))
434 p2->p_flags |= P_JAILED;
437 refcount_acquire(&p2->p_args->ar_ref);
439 p2->p_usched = p1->p_usched;
440 /* XXX: verify copy of the secondary iosched stuff */
443 if (flags & RFSIGSHARE) {
444 p2->p_sigacts = p1->p_sigacts;
445 refcount_acquire(&p2->p_sigacts->ps_refcnt);
447 p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts),
448 M_SUBPROC, M_WAITOK);
449 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
450 refcount_init(&p2->p_sigacts->ps_refcnt, 1);
452 if (flags & RFLINUXTHPN)
453 p2->p_sigparent = SIGUSR1;
455 p2->p_sigparent = SIGCHLD;
457 /* bump references to the text vnode (for procfs) */
458 p2->p_textvp = p1->p_textvp;
462 /* copy namecache handle to the text file */
463 if (p1->p_textnch.mount)
464 cache_copy(&p1->p_textnch, &p2->p_textnch);
467 * Handle file descriptors
469 if (flags & RFCFDG) {
470 p2->p_fd = fdinit(p1);
472 } else if (flags & RFFDG) {
473 error = fdcopy(p1, &p2->p_fd);
480 p2->p_fd = fdshare(p1);
481 if (p1->p_fdtol == NULL) {
482 p1->p_fdtol = filedesc_to_leader_alloc(NULL,
485 if ((flags & RFTHREAD) != 0) {
487 * Shared file descriptor table and
488 * shared process leaders.
491 fdtol->fdl_refcount++;
494 * Shared file descriptor table, and
495 * different process leaders
497 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
501 p2->p_limit = plimit_fork(p1);
504 * Preserve some more flags in subprocess. P_PROFIL has already
507 p2->p_flags |= p1->p_flags & P_SUGID;
508 if (p1->p_session->s_ttyvp != NULL && (p1->p_flags & P_CONTROLT))
509 p2->p_flags |= P_CONTROLT;
510 if (flags & RFPPWAIT)
511 p2->p_flags |= P_PPWAIT;
514 * Inherit the virtual kernel structure (allows a virtual kernel
515 * to fork to simulate multiple cpus).
518 vkernel_inherit(p1, p2);
521 * Once we are on a pglist we may receive signals. XXX we might
522 * race a ^C being sent to the process group by not receiving it
523 * at all prior to this line.
526 lwkt_gettoken(&p1grp->pg_token);
527 LIST_INSERT_AFTER(p1, p2, p_pglist);
528 lwkt_reltoken(&p1grp->pg_token);
531 * Attach the new process to its parent.
533 * If RFNOWAIT is set, the newly created process becomes a child
534 * of init. This effectively disassociates the child from the
537 if (flags & RFNOWAIT)
542 LIST_INIT(&p2->p_children);
544 lwkt_gettoken(&pptr->p_token);
545 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
546 lwkt_reltoken(&pptr->p_token);
548 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
549 callout_init_mp(&p2->p_ithandle);
553 * Copy traceflag and tracefile if enabled. If not inherited,
554 * these were zeroed above but we still could have a trace race
555 * so make sure p2's p_tracenode is NULL.
557 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
558 p2->p_traceflag = p1->p_traceflag;
559 p2->p_tracenode = ktrinherit(p1->p_tracenode);
564 * This begins the section where we must prevent the parent
565 * from being swapped.
567 * Gets PRELE'd in the caller in start_forked_proc().
571 vm_fork(p1, p2, flags);
574 * Create the first lwp associated with the new proc.
575 * It will return via a different execution path later, directly
576 * into userland, after it was put on the runq by
577 * start_forked_proc().
579 lwp_fork(lp1, p2, flags);
581 if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
582 mycpu->gd_cnt.v_forks++;
583 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize +
584 p2->p_vmspace->vm_ssize;
585 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
586 mycpu->gd_cnt.v_vforks++;
587 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
588 p2->p_vmspace->vm_ssize;
589 } else if (p1 == &proc0) {
590 mycpu->gd_cnt.v_kthreads++;
591 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
592 p2->p_vmspace->vm_ssize;
594 mycpu->gd_cnt.v_rforks++;
595 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
596 p2->p_vmspace->vm_ssize;
600 * Both processes are set up, now check if any loadable modules want
601 * to adjust anything.
602 * What if they have an error? XXX
604 TAILQ_FOREACH(ep, &fork_list, next) {
605 (*ep->function)(p1, p2, flags);
609 * Set the start time. Note that the process is not runnable. The
610 * caller is responsible for making it runnable.
612 microtime(&p2->p_start);
613 p2->p_acflag = AFORK;
616 * tell any interested parties about the new process
618 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
621 * Return child proc pointer to parent.
627 lwkt_reltoken(&p2->p_token);
628 lwkt_reltoken(&p1->p_token);
630 lockmgr(&plkgrp->pg_lock, LK_RELEASE);
637 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
639 globaldata_t gd = mycpu;
643 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
645 lp->lwp_proc = destproc;
646 lp->lwp_vmspace = destproc->p_vmspace;
647 lp->lwp_stat = LSRUN;
648 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
649 (unsigned) ((caddr_t)&lp->lwp_endcopy -
650 (caddr_t)&lp->lwp_startcopy));
651 lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK;
653 * Set cpbase to the last timeout that occured (not the upcoming
656 * A critical section is required since a timer IPI can update
657 * scheduler specific data.
660 lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic;
661 destproc->p_usched->heuristic_forking(origlp, lp);
663 lp->lwp_cpumask &= usched_mastermask;
664 lwkt_token_init(&lp->lwp_token, "lwp_token");
665 spin_init(&lp->lwp_spin);
668 * Assign the thread to the current cpu to begin with so we
671 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0);
673 td->td_proc = destproc;
675 td->td_switch = cpu_heavy_switch;
676 lwkt_setpri(td, TDPRI_KERN_USER);
677 lwkt_set_comm(td, "%s", destproc->p_comm);
680 * cpu_fork will copy and update the pcb, set up the kernel stack,
681 * and make the child ready to run.
683 cpu_fork(origlp, lp, flags);
684 caps_fork(origlp->lwp_thread, lp->lwp_thread);
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++;
703 * The next two functionms are general routines to handle adding/deleting
704 * items on the fork callout list.
707 * Take the arguments given and put them onto the fork callout list,
708 * However first make sure that it's not already there.
709 * Returns 0 on success or a standard error number.
712 at_fork(forklist_fn function)
717 /* let the programmer know if he's been stupid */
718 if (rm_at_fork(function)) {
719 kprintf("WARNING: fork callout entry (%p) already present\n",
723 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
724 ep->function = function;
725 TAILQ_INSERT_TAIL(&fork_list, ep, next);
730 * Scan the exit callout list for the given item and remove it..
731 * Returns the number of items removed (0 or 1)
734 rm_at_fork(forklist_fn function)
738 TAILQ_FOREACH(ep, &fork_list, next) {
739 if (ep->function == function) {
740 TAILQ_REMOVE(&fork_list, ep, next);
749 * Add a forked process to the run queue after any remaining setup, such
750 * as setting the fork handler, has been completed.
752 * p2 is held by the caller.
755 start_forked_proc(struct lwp *lp1, struct proc *p2)
757 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
760 * Move from SIDL to RUN queue, and activate the process's thread.
761 * Activation of the thread effectively makes the process "a"
762 * current process, so we do not setrunqueue().
764 * YYY setrunqueue works here but we should clean up the trampoline
765 * code so we just schedule the LWKT thread and let the trampoline
766 * deal with the userland scheduler on return to userland.
768 KASSERT(p2->p_stat == SIDL,
769 ("cannot start forked process, bad status: %p", p2));
770 p2->p_usched->resetpriority(lp2);
772 p2->p_stat = SACTIVE;
773 lp2->lwp_stat = LSRUN;
774 p2->p_usched->setrunqueue(lp2);
778 * Now can be swapped.
780 PRELE(lp1->lwp_proc);
783 * Preserve synchronization semantics of vfork. If waiting for
784 * child to exec or exit, set P_PPWAIT on child, and sleep on our
785 * proc (in case of exec or exit).
787 * We must hold our p_token to interlock the flag/tsleep
789 lwkt_gettoken(&p2->p_token);
790 while (p2->p_flags & P_PPWAIT)
791 tsleep(lp1->lwp_proc, 0, "ppwait", 0);
792 lwkt_reltoken(&p2->p_token);