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>
59 #include <vm/vm_map.h>
60 #include <vm/vm_extern.h>
62 #include <sys/vmmeter.h>
63 #include <sys/refcount.h>
64 #include <sys/thread2.h>
65 #include <sys/signal2.h>
66 #include <sys/spinlock2.h>
68 #include <sys/dsched.h>
70 static MALLOC_DEFINE(M_ATFORK, "atfork", "atfork callback");
71 static MALLOC_DEFINE(M_REAPER, "reaper", "process reapers");
74 * These are the stuctures used to create a callout list for things to do
75 * when forking a process
79 TAILQ_ENTRY(forklist) next;
82 TAILQ_HEAD(forklist_head, forklist);
83 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
85 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags,
86 const cpumask_t *mask);
87 static int lwp_create1(struct lwp_params *params,
88 const cpumask_t *mask);
89 static struct lock reaper_lock = LOCK_INITIALIZER("reapgl", 0, 0);
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);
110 * When forking, memory underpinning umtx-supported mutexes may be set
111 * COW causing the physical address to change. We must wakeup any threads
112 * blocked on the physical address to allow them to re-resolve their VM.
114 * (caller is holding p->p_token)
117 wake_umtx_threads(struct proc *p1)
122 RB_FOREACH(lp, lwp_rb_tree, &p1->p_lwp_tree) {
124 if (td && (td->td_flags & TDF_TSLEEPQ) &&
125 (td->td_wdomain & PDOMAIN_MASK) == PDOMAIN_UMTX) {
126 wakeup_domain(td->td_wchan, PDOMAIN_UMTX);
135 sys_fork(struct fork_args *uap)
137 struct lwp *lp = curthread->td_lwp;
141 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
144 start_forked_proc(lp, p2);
145 uap->sysmsg_fds[0] = p2->p_pid;
146 uap->sysmsg_fds[1] = 0;
153 * vfork() system call
156 sys_vfork(struct vfork_args *uap)
158 struct lwp *lp = curthread->td_lwp;
162 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
165 start_forked_proc(lp, p2);
166 uap->sysmsg_fds[0] = p2->p_pid;
167 uap->sysmsg_fds[1] = 0;
174 * Handle rforks. An rfork may (1) operate on the current process without
175 * creating a new, (2) create a new process that shared the current process's
176 * vmspace, signals, and/or descriptors, or (3) create a new process that does
177 * not share these things (normal fork).
179 * Note that we only call start_forked_proc() if a new process is actually
182 * rfork { int flags }
185 sys_rfork(struct rfork_args *uap)
187 struct lwp *lp = curthread->td_lwp;
191 if ((uap->flags & RFKERNELONLY) != 0)
194 error = fork1(lp, uap->flags | RFPGLOCK, &p2);
198 start_forked_proc(lp, p2);
199 uap->sysmsg_fds[0] = p2->p_pid;
200 uap->sysmsg_fds[1] = 0;
203 uap->sysmsg_fds[0] = 0;
204 uap->sysmsg_fds[1] = 0;
211 lwp_create1(struct lwp_params *uprm, const cpumask_t *umask)
213 struct proc *p = curproc;
215 struct lwp_params params;
216 cpumask_t *mask = NULL, mask0;
219 error = copyin(uprm, ¶ms, sizeof(params));
224 error = copyin(umask, &mask0, sizeof(mask0));
227 CPUMASK_ANDMASK(mask0, smp_active_mask);
228 if (CPUMASK_TESTNZERO(mask0))
232 lwkt_gettoken(&p->p_token);
233 plimit_lwp_fork(p); /* force exclusive access */
234 lp = lwp_fork(curthread->td_lwp, p, RFPROC | RFMEM, mask);
235 error = cpu_prepare_lwp(lp, ¶ms);
238 if (params.lwp_tid1 != NULL &&
239 (error = copyout(&lp->lwp_tid, params.lwp_tid1, sizeof(lp->lwp_tid))))
241 if (params.lwp_tid2 != NULL &&
242 (error = copyout(&lp->lwp_tid, params.lwp_tid2, sizeof(lp->lwp_tid))))
246 * Now schedule the new lwp.
248 p->p_usched->resetpriority(lp);
250 lp->lwp_stat = LSRUN;
251 p->p_usched->setrunqueue(lp);
253 lwkt_reltoken(&p->p_token);
259 * Make sure no one is using this lwp, before it is removed from
260 * the tree. If we didn't wait it here, lwp tree iteration with
261 * blocking operation would be broken.
263 while (lp->lwp_lock > 0)
264 tsleep(lp, 0, "lwpfail", 1);
265 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
267 /* lwp_dispose expects an exited lwp, and a held proc */
268 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
269 lp->lwp_thread->td_flags |= TDF_EXITING;
270 lwkt_remove_tdallq(lp->lwp_thread);
272 biosched_done(lp->lwp_thread);
273 dsched_exit_thread(lp->lwp_thread);
275 lwkt_reltoken(&p->p_token);
281 * Low level thread create used by pthreads.
284 sys_lwp_create(struct lwp_create_args *uap)
287 return (lwp_create1(uap->params, NULL));
291 sys_lwp_create2(struct lwp_create2_args *uap)
294 return (lwp_create1(uap->params, uap->mask));
297 int nprocs = 1; /* process 0 */
300 fork1(struct lwp *lp1, int flags, struct proc **procp)
302 struct proc *p1 = lp1->lwp_proc;
307 struct sysreaper *reap;
310 static int curfail = 0;
311 static struct timeval lastfail;
313 struct filedesc_to_leader *fdtol;
315 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
318 lwkt_gettoken(&p1->p_token);
323 * Here we don't create a new process, but we divorce
324 * certain parts of a process from itself.
326 if ((flags & RFPROC) == 0) {
328 * This kind of stunt does not work anymore if
329 * there are native threads (lwps) running
331 if (p1->p_nthreads != 1) {
336 vm_fork(p1, 0, flags);
337 if ((flags & RFMEM) == 0)
338 wake_umtx_threads(p1);
341 * Close all file descriptors.
343 if (flags & RFCFDG) {
344 struct filedesc *fdtmp;
350 * Unshare file descriptors (from parent.)
353 if (p1->p_fd->fd_refcnt > 1) {
354 struct filedesc *newfd;
355 error = fdcopy(p1, &newfd);
369 * Interlock against process group signal delivery. If signals
370 * are pending after the interlock is obtained we have to restart
371 * the system call to process the signals. If we don't the child
372 * can miss a pgsignal (such as ^C) sent during the fork.
374 * We can't use CURSIG() here because it will process any STOPs
375 * and cause the process group lock to be held indefinitely. If
376 * a STOP occurs, the fork will be restarted after the CONT.
379 if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) {
381 lockmgr(&plkgrp->pg_lock, LK_SHARED);
382 if (CURSIG_NOBLOCK(lp1)) {
389 * Although process entries are dynamically created, we still keep
390 * a global limit on the maximum number we will create. Don't allow
391 * a nonprivileged user to use the last ten processes; don't let root
392 * exceed the limit. The variable nprocs is the current number of
393 * processes, maxproc is the limit.
395 uid = lp1->lwp_thread->td_ucred->cr_ruid;
396 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
397 if (ppsratecheck(&lastfail, &curfail, 1))
398 kprintf("maxproc limit exceeded by uid %d, please "
399 "see tuning(7) and login.conf(5).\n", uid);
400 tsleep(&forksleep, 0, "fork", hz / 2);
406 * Increment the nprocs resource before blocking can occur. There
407 * are hard-limits as to the number of processes that can run.
409 atomic_add_int(&nprocs, 1);
412 * Increment the count of procs running with this uid. This also
415 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
416 plimit_getadjvalue(RLIMIT_NPROC));
419 * Back out the process count
421 atomic_add_int(&nprocs, -1);
422 if (ppsratecheck(&lastfail, &curfail, 1)) {
423 kprintf("maxproc limit of %jd "
424 "exceeded by \"%s\" uid %d, "
425 "please see tuning(7) and login.conf(5).\n",
426 plimit_getadjvalue(RLIMIT_NPROC),
430 tsleep(&forksleep, 0, "fork", hz / 2);
436 * Allocate a new process, don't get fancy: zero the structure.
438 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
441 * Core initialization. SIDL is a safety state that protects the
442 * partially initialized process once it starts getting hooked
443 * into system structures and becomes addressable.
445 * We must be sure to acquire p2->p_token as well, we must hold it
446 * once the process is on the allproc list to avoid things such
447 * as competing modifications to p_flags.
449 mycpu->gd_forkid += ncpus;
450 p2->p_forkid = mycpu->gd_forkid + mycpu->gd_cpuid;
451 p2->p_lasttid = 0; /* first tid will be 1 */
455 * NOTE: Process 0 will not have a reaper, but process 1 (init) and
456 * all other processes always will.
458 if ((reap = p1->p_reaper) != NULL) {
465 RB_INIT(&p2->p_lwp_tree);
466 spin_init(&p2->p_spin, "procfork1");
467 lwkt_token_init(&p2->p_token, "proc");
468 lwkt_gettoken(&p2->p_token);
469 p2->p_uidpcpu = kmalloc(sizeof(*p2->p_uidpcpu) * ncpus,
470 M_SUBPROC, M_WAITOK | M_ZERO);
473 * Setup linkage for kernel based threading XXX lwp. Also add the
474 * process to the allproclist.
476 * The process structure is addressable after this point.
478 if (flags & RFTHREAD) {
479 p2->p_peers = p1->p_peers;
481 p2->p_leader = p1->p_leader;
485 proc_add_allproc(p2);
488 * Initialize the section which is copied verbatim from the parent.
490 bcopy(&p1->p_startcopy, &p2->p_startcopy,
491 ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
494 * Duplicate sub-structures as needed. Increase reference counts
497 * NOTE: because we are now on the allproc list it is possible for
498 * other consumers to gain temporary references to p2
499 * (p2->p_lock can change).
501 if (p1->p_flags & P_PROFIL)
503 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
505 if (jailed(p2->p_ucred))
506 p2->p_flags |= P_JAILED;
509 refcount_acquire(&p2->p_args->ar_ref);
511 p2->p_usched = p1->p_usched;
512 /* XXX: verify copy of the secondary iosched stuff */
513 dsched_enter_proc(p2);
515 if (flags & RFSIGSHARE) {
516 p2->p_sigacts = p1->p_sigacts;
517 refcount_acquire(&p2->p_sigacts->ps_refcnt);
519 p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts),
520 M_SUBPROC, M_WAITOK);
521 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
522 refcount_init(&p2->p_sigacts->ps_refcnt, 1);
524 if (flags & RFLINUXTHPN)
525 p2->p_sigparent = SIGUSR1;
527 p2->p_sigparent = SIGCHLD;
529 /* bump references to the text vnode (for procfs) */
530 p2->p_textvp = p1->p_textvp;
534 /* copy namecache handle to the text file */
535 if (p1->p_textnch.mount)
536 cache_copy(&p1->p_textnch, &p2->p_textnch);
539 * Handle file descriptors
541 if (flags & RFCFDG) {
542 p2->p_fd = fdinit(p1);
544 } else if (flags & RFFDG) {
545 error = fdcopy(p1, &p2->p_fd);
552 p2->p_fd = fdshare(p1);
553 if (p1->p_fdtol == NULL) {
554 p1->p_fdtol = filedesc_to_leader_alloc(NULL,
557 if ((flags & RFTHREAD) != 0) {
559 * Shared file descriptor table and
560 * shared process leaders.
563 fdtol->fdl_refcount++;
566 * Shared file descriptor table, and
567 * different process leaders
569 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
573 p2->p_limit = plimit_fork(p1);
576 * Adjust depth for resource downscaling
578 if ((p2->p_depth & 31) != 31)
582 * Preserve some more flags in subprocess. P_PROFIL has already
585 p2->p_flags |= p1->p_flags & P_SUGID;
586 if (p1->p_session->s_ttyvp != NULL && (p1->p_flags & P_CONTROLT))
587 p2->p_flags |= P_CONTROLT;
588 if (flags & RFPPWAIT) {
589 p2->p_flags |= P_PPWAIT;
591 atomic_add_int(&p1->p_upmap->invfork, 1);
595 * Inherit the virtual kernel structure (allows a virtual kernel
596 * to fork to simulate multiple cpus).
599 vkernel_inherit(p1, p2);
602 * Once we are on a pglist we may receive signals. XXX we might
603 * race a ^C being sent to the process group by not receiving it
604 * at all prior to this line.
607 lwkt_gettoken(&p1grp->pg_token);
608 LIST_INSERT_AFTER(p1, p2, p_pglist);
609 lwkt_reltoken(&p1grp->pg_token);
612 * Attach the new process to its parent.
614 * If RFNOWAIT is set, the newly created process becomes a child
615 * of the reaper (typically init). This effectively disassociates
616 * the child from the parent.
618 * Temporarily hold pptr for the RFNOWAIT case to avoid ripouts.
620 if (flags & RFNOWAIT) {
621 pptr = reaper_get(reap);
630 p2->p_ppid = pptr->p_pid;
631 LIST_INIT(&p2->p_children);
633 lwkt_gettoken(&pptr->p_token);
634 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
635 lwkt_reltoken(&pptr->p_token);
637 if (flags & RFNOWAIT)
640 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
641 callout_init_mp(&p2->p_ithandle);
645 * Copy traceflag and tracefile if enabled. If not inherited,
646 * these were zeroed above but we still could have a trace race
647 * so make sure p2's p_tracenode is NULL.
649 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
650 p2->p_traceflag = p1->p_traceflag;
651 p2->p_tracenode = ktrinherit(p1->p_tracenode);
656 * This begins the section where we must prevent the parent
657 * from being swapped.
659 * Gets PRELE'd in the caller in start_forked_proc().
663 vm_fork(p1, p2, flags);
664 if ((flags & RFMEM) == 0)
665 wake_umtx_threads(p1);
668 * Create the first lwp associated with the new proc.
669 * It will return via a different execution path later, directly
670 * into userland, after it was put on the runq by
671 * start_forked_proc().
673 lwp_fork(lp1, p2, flags, NULL);
675 if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
676 mycpu->gd_cnt.v_forks++;
677 mycpu->gd_cnt.v_forkpages += btoc(p2->p_vmspace->vm_dsize) +
678 btoc(p2->p_vmspace->vm_ssize);
679 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
680 mycpu->gd_cnt.v_vforks++;
681 mycpu->gd_cnt.v_vforkpages += btoc(p2->p_vmspace->vm_dsize) +
682 btoc(p2->p_vmspace->vm_ssize);
683 } else if (p1 == &proc0) {
684 mycpu->gd_cnt.v_kthreads++;
685 mycpu->gd_cnt.v_kthreadpages += btoc(p2->p_vmspace->vm_dsize) +
686 btoc(p2->p_vmspace->vm_ssize);
688 mycpu->gd_cnt.v_rforks++;
689 mycpu->gd_cnt.v_rforkpages += btoc(p2->p_vmspace->vm_dsize) +
690 btoc(p2->p_vmspace->vm_ssize);
694 * Both processes are set up, now check if any loadable modules want
695 * to adjust anything.
696 * What if they have an error? XXX
698 TAILQ_FOREACH(ep, &fork_list, next) {
699 (*ep->function)(p1, p2, flags);
703 * Set the start time. Note that the process is not runnable. The
704 * caller is responsible for making it runnable.
706 microtime(&p2->p_start);
707 p2->p_acflag = AFORK;
710 * tell any interested parties about the new process
712 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
715 * Return child proc pointer to parent.
721 lwkt_reltoken(&p2->p_token);
722 lwkt_reltoken(&p1->p_token);
724 lockmgr(&plkgrp->pg_lock, LK_RELEASE);
731 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags,
732 const cpumask_t *mask)
734 globaldata_t gd = mycpu;
738 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
740 lp->lwp_proc = destproc;
741 lp->lwp_vmspace = destproc->p_vmspace;
742 lp->lwp_stat = LSRUN;
743 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
744 (unsigned) ((caddr_t)&lp->lwp_endcopy -
745 (caddr_t)&lp->lwp_startcopy));
747 lp->lwp_cpumask = *mask;
750 * Reset the sigaltstack if memory is shared, otherwise inherit
754 lp->lwp_sigstk.ss_flags = SS_DISABLE;
755 lp->lwp_sigstk.ss_size = 0;
756 lp->lwp_sigstk.ss_sp = NULL;
757 lp->lwp_flags &= ~LWP_ALTSTACK;
759 lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK;
763 * Set cpbase to the last timeout that occured (not the upcoming
766 * A critical section is required since a timer IPI can update
767 * scheduler specific data.
770 lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic;
771 destproc->p_usched->heuristic_forking(origlp, lp);
773 CPUMASK_ANDMASK(lp->lwp_cpumask, usched_mastermask);
774 lwkt_token_init(&lp->lwp_token, "lwp_token");
775 spin_init(&lp->lwp_spin, "lwptoken");
778 * Assign the thread to the current cpu to begin with so we
781 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0);
783 td->td_wakefromcpu = gd->gd_cpuid;
784 td->td_ucred = crhold(destproc->p_ucred);
785 td->td_proc = destproc;
787 td->td_switch = cpu_heavy_switch;
788 #ifdef NO_LWKT_SPLIT_USERPRI
789 lwkt_setpri(td, TDPRI_USER_NORM);
791 lwkt_setpri(td, TDPRI_KERN_USER);
793 lwkt_set_comm(td, "%s", destproc->p_comm);
796 * cpu_fork will copy and update the pcb, set up the kernel stack,
797 * and make the child ready to run.
799 cpu_fork(origlp, lp, flags);
800 kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
803 * Assign a TID to the lp. Loop until the insert succeeds (returns
806 * If we are in a vfork assign the same TID as the lwp that did the
807 * vfork(). This way if the user program messes around with
808 * pthread calls inside the vfork(), it will operate like an
809 * extension of the (blocked) parent. Also note that since the
810 * address space is being shared, insofar as pthreads is concerned,
811 * the code running in the vfork() is part of the original process.
813 if (flags & RFPPWAIT) {
814 lp->lwp_tid = origlp->lwp_tid - 1;
816 lp->lwp_tid = destproc->p_lasttid;
820 if (++lp->lwp_tid <= 0)
822 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
824 destproc->p_lasttid = lp->lwp_tid;
825 destproc->p_nthreads++;
828 * This flag is set and never cleared. It means that the process
829 * was threaded at some point. Used to improve exit performance.
831 pmap_maybethreaded(&destproc->p_vmspace->vm_pmap);
832 destproc->p_flags |= P_MAYBETHREADED;
838 * The next two functionms are general routines to handle adding/deleting
839 * items on the fork callout list.
842 * Take the arguments given and put them onto the fork callout list,
843 * However first make sure that it's not already there.
844 * Returns 0 on success or a standard error number.
847 at_fork(forklist_fn function)
852 /* let the programmer know if he's been stupid */
853 if (rm_at_fork(function)) {
854 kprintf("WARNING: fork callout entry (%p) already present\n",
858 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
859 ep->function = function;
860 TAILQ_INSERT_TAIL(&fork_list, ep, next);
865 * Scan the exit callout list for the given item and remove it..
866 * Returns the number of items removed (0 or 1)
869 rm_at_fork(forklist_fn function)
873 TAILQ_FOREACH(ep, &fork_list, next) {
874 if (ep->function == function) {
875 TAILQ_REMOVE(&fork_list, ep, next);
884 * Add a forked process to the run queue after any remaining setup, such
885 * as setting the fork handler, has been completed.
887 * p2 is held by the caller.
890 start_forked_proc(struct lwp *lp1, struct proc *p2)
892 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
896 * Move from SIDL to RUN queue, and activate the process's thread.
897 * Activation of the thread effectively makes the process "a"
898 * current process, so we do not setrunqueue().
900 * YYY setrunqueue works here but we should clean up the trampoline
901 * code so we just schedule the LWKT thread and let the trampoline
902 * deal with the userland scheduler on return to userland.
904 KASSERT(p2->p_stat == SIDL,
905 ("cannot start forked process, bad status: %p", p2));
906 p2->p_usched->resetpriority(lp2);
908 p2->p_stat = SACTIVE;
909 lp2->lwp_stat = LSRUN;
910 p2->p_usched->setrunqueue(lp2);
914 * Now can be swapped.
916 PRELE(lp1->lwp_proc);
919 * Preserve synchronization semantics of vfork. P_PPWAIT is set in
920 * the child until it has retired the parent's resources. The parent
921 * must wait for the flag to be cleared by the child.
923 * Interlock the flag/tsleep with atomic ops to avoid unnecessary
926 * XXX Is this use of an atomic op on a field that is not normally
927 * manipulated with atomic ops ok?
929 while ((pflags = p2->p_flags) & P_PPWAIT) {
931 tsleep_interlock(lp1->lwp_proc, 0);
932 if (atomic_cmpset_int(&p2->p_flags, pflags, pflags))
933 tsleep(lp1->lwp_proc, PINTERLOCKED, "ppwait", 0);
938 * procctl (idtype_t idtype, id_t id, int cmd, void *arg)
941 sys_procctl(struct procctl_args *uap)
943 struct proc *p = curproc;
945 struct sysreaper *reap;
946 union reaper_info udata;
949 if (uap->idtype != P_PID || uap->id != (id_t)p->p_pid)
953 case PROC_REAP_ACQUIRE:
954 lwkt_gettoken(&p->p_token);
955 reap = kmalloc(sizeof(*reap), M_REAPER, M_WAITOK|M_ZERO);
956 if (p->p_reaper == NULL || p->p_reaper->p != p) {
957 reaper_init(p, reap);
960 kfree(reap, M_REAPER);
963 lwkt_reltoken(&p->p_token);
965 case PROC_REAP_RELEASE:
966 lwkt_gettoken(&p->p_token);
969 KKASSERT(reap != NULL);
971 reaper_hold(reap); /* in case of thread race */
972 lockmgr(&reap->lock, LK_EXCLUSIVE);
974 lockmgr(&reap->lock, LK_RELEASE);
979 p->p_reaper = reap->parent;
981 reaper_hold(p->p_reaper);
982 lockmgr(&reap->lock, LK_RELEASE);
983 reaper_drop(reap); /* our ref */
984 reaper_drop(reap); /* old p_reaper ref */
989 lwkt_reltoken(&p->p_token);
991 case PROC_REAP_STATUS:
992 bzero(&udata, sizeof(udata));
993 lwkt_gettoken_shared(&p->p_token);
994 if ((reap = p->p_reaper) != NULL && reap->p == p) {
995 udata.status.flags = reap->flags;
996 udata.status.refs = reap->refs - 1; /* minus ours */
998 p2 = LIST_FIRST(&p->p_children);
999 udata.status.pid_head = p2 ? p2->p_pid : -1;
1000 lwkt_reltoken(&p->p_token);
1003 error = copyout(&udata, uap->data,
1004 sizeof(udata.status));
1017 * Bump ref on reaper, preventing destruction
1020 reaper_hold(struct sysreaper *reap)
1022 KKASSERT(reap->refs > 0);
1023 refcount_acquire(&reap->refs);
1027 * Drop ref on reaper, destroy the structure on the 1->0
1028 * transition and loop on the parent.
1031 reaper_drop(struct sysreaper *next)
1033 struct sysreaper *reap;
1035 while ((reap = next) != NULL) {
1036 if (refcount_release(&reap->refs)) {
1037 next = reap->parent;
1038 KKASSERT(reap->p == NULL);
1039 lockmgr(&reaper_lock, LK_EXCLUSIVE);
1040 reap->parent = NULL;
1041 kfree(reap, M_REAPER);
1042 lockmgr(&reaper_lock, LK_RELEASE);
1050 * Initialize a static or newly allocated reaper structure
1053 reaper_init(struct proc *p, struct sysreaper *reap)
1055 reap->parent = p->p_reaper;
1057 if (p == initproc) {
1058 reap->flags = REAPER_STAT_OWNED | REAPER_STAT_REALINIT;
1061 reap->flags = REAPER_STAT_OWNED;
1064 lockinit(&reap->lock, "subrp", 0, 0);
1070 * Called with p->p_token held during exit.
1072 * This is a bit simpler than RELEASE because there are no threads remaining
1073 * to race. We only release if we own the reaper, the exit code will handle
1074 * the final p_reaper release.
1077 reaper_exit(struct proc *p)
1079 struct sysreaper *reap;
1082 * Release acquired reaper
1084 if ((reap = p->p_reaper) != NULL && reap->p == p) {
1085 lockmgr(&reap->lock, LK_EXCLUSIVE);
1086 p->p_reaper = reap->parent;
1088 reaper_hold(p->p_reaper);
1090 lockmgr(&reap->lock, LK_RELEASE);
1095 * Return and clear reaper (caller is holding p_token for us)
1096 * (reap->p does not equal p). Caller must drop it.
1098 if ((reap = p->p_reaper) != NULL) {
1105 * Return a held (PHOLD) process representing the reaper for process (p).
1106 * NULL should not normally be returned. Caller should PRELE() the returned
1107 * reaper process when finished.
1109 * Remove dead internal nodes while we are at it.
1111 * Process (p)'s token must be held on call.
1112 * The returned process's token is NOT acquired by this routine.
1115 reaper_get(struct sysreaper *reap)
1117 struct sysreaper *next;
1118 struct proc *reproc;
1124 * Extra hold for loop
1129 lockmgr(&reap->lock, LK_SHARED);
1137 lockmgr(&reap->lock, LK_RELEASE);
1145 lockmgr(&reap->lock, LK_RELEASE);
1150 * Traverse upwards in the reaper topology, destroy
1151 * dead internal nodes when possible.
1153 * NOTE: Our ref on next means that a dead node should
1154 * have 2 (ours and reap->parent's).
1156 next = reap->parent;
1159 if (next->refs == 2 && next->p == NULL) {
1160 lockmgr(&reap->lock, LK_RELEASE);
1161 lockmgr(&reap->lock, LK_EXCLUSIVE);
1162 if (next->refs == 2 &&
1163 reap->parent == next &&
1166 * reap->parent inherits ref from next.
1168 reap->parent = next->parent;
1169 next->parent = NULL;
1170 reaper_drop(next); /* ours */
1171 reaper_drop(next); /* old parent */
1172 next = reap->parent;
1173 continue; /* possible chain */
1178 lockmgr(&reap->lock, LK_RELEASE);
1186 * Test that the sender is allowed to send a signal to the target.
1187 * The sender process is assumed to have a stable reaper. The
1188 * target can be e.g. from a scan callback.
1190 * Target cannot be the reaper process itself unless reaper_ok is specified,
1191 * or sender == target.
1194 reaper_sigtest(struct proc *sender, struct proc *target, int reaper_ok)
1196 struct sysreaper *sreap;
1197 struct sysreaper *reap;
1200 sreap = sender->p_reaper;
1204 if (sreap == target->p_reaper) {
1205 if (sreap->p == target && sreap->p != sender && reaper_ok == 0)
1209 lockmgr(&reaper_lock, LK_SHARED);
1211 for (reap = target->p_reaper; reap; reap = reap->parent) {
1212 if (sreap == reap) {
1213 if (sreap->p != target || reaper_ok)
1218 lockmgr(&reaper_lock, LK_RELEASE);