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");
70 static MALLOC_DEFINE(M_REAPER, "reaper", "process reapers");
73 * These are the stuctures used to create a callout list for things to do
74 * when forking a process
78 TAILQ_ENTRY(forklist) next;
81 TAILQ_HEAD(forklist_head, forklist);
82 static struct forklist_head fork_list = TAILQ_HEAD_INITIALIZER(fork_list);
84 static struct lwp *lwp_fork(struct lwp *, struct proc *, int flags);
86 int forksleep; /* Place for fork1() to sleep on. */
89 * Red-Black tree support for LWPs
93 rb_lwp_compare(struct lwp *lp1, struct lwp *lp2)
95 if (lp1->lwp_tid < lp2->lwp_tid)
97 if (lp1->lwp_tid > lp2->lwp_tid)
102 RB_GENERATE2(lwp_rb_tree, lwp, u.lwp_rbnode, rb_lwp_compare, lwpid_t, lwp_tid);
108 sys_fork(struct fork_args *uap)
110 struct lwp *lp = curthread->td_lwp;
114 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
117 start_forked_proc(lp, p2);
118 uap->sysmsg_fds[0] = p2->p_pid;
119 uap->sysmsg_fds[1] = 0;
126 * vfork() system call
129 sys_vfork(struct vfork_args *uap)
131 struct lwp *lp = curthread->td_lwp;
135 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
138 start_forked_proc(lp, p2);
139 uap->sysmsg_fds[0] = p2->p_pid;
140 uap->sysmsg_fds[1] = 0;
147 * Handle rforks. An rfork may (1) operate on the current process without
148 * creating a new, (2) create a new process that shared the current process's
149 * vmspace, signals, and/or descriptors, or (3) create a new process that does
150 * not share these things (normal fork).
152 * Note that we only call start_forked_proc() if a new process is actually
155 * rfork { int flags }
158 sys_rfork(struct rfork_args *uap)
160 struct lwp *lp = curthread->td_lwp;
164 if ((uap->flags & RFKERNELONLY) != 0)
167 error = fork1(lp, uap->flags | RFPGLOCK, &p2);
171 start_forked_proc(lp, p2);
172 uap->sysmsg_fds[0] = p2->p_pid;
173 uap->sysmsg_fds[1] = 0;
176 uap->sysmsg_fds[0] = 0;
177 uap->sysmsg_fds[1] = 0;
184 * Low level thread create used by pthreads.
187 sys_lwp_create(struct lwp_create_args *uap)
189 struct proc *p = curproc;
191 struct lwp_params params;
194 error = copyin(uap->params, ¶ms, sizeof(params));
198 lwkt_gettoken(&p->p_token);
199 plimit_lwp_fork(p); /* force exclusive access */
200 lp = lwp_fork(curthread->td_lwp, p, RFPROC);
201 error = cpu_prepare_lwp(lp, ¶ms);
204 if (params.tid1 != NULL &&
205 (error = copyout(&lp->lwp_tid, params.tid1, sizeof(lp->lwp_tid))))
207 if (params.tid2 != NULL &&
208 (error = copyout(&lp->lwp_tid, params.tid2, sizeof(lp->lwp_tid))))
212 * Now schedule the new lwp.
214 p->p_usched->resetpriority(lp);
216 lp->lwp_stat = LSRUN;
217 p->p_usched->setrunqueue(lp);
219 lwkt_reltoken(&p->p_token);
224 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
226 /* lwp_dispose expects an exited lwp, and a held proc */
227 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
228 lp->lwp_thread->td_flags |= TDF_EXITING;
229 lwkt_remove_tdallq(lp->lwp_thread);
231 biosched_done(lp->lwp_thread);
232 dsched_exit_thread(lp->lwp_thread);
234 lwkt_reltoken(&p->p_token);
239 int nprocs = 1; /* process 0 */
242 fork1(struct lwp *lp1, int flags, struct proc **procp)
244 struct proc *p1 = lp1->lwp_proc;
251 static int curfail = 0;
252 static struct timeval lastfail;
254 struct filedesc_to_leader *fdtol;
256 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
259 lwkt_gettoken(&p1->p_token);
264 * Here we don't create a new process, but we divorce
265 * certain parts of a process from itself.
267 if ((flags & RFPROC) == 0) {
269 * This kind of stunt does not work anymore if
270 * there are native threads (lwps) running
272 if (p1->p_nthreads != 1) {
277 vm_fork(p1, 0, flags);
280 * Close all file descriptors.
282 if (flags & RFCFDG) {
283 struct filedesc *fdtmp;
289 * Unshare file descriptors (from parent.)
292 if (p1->p_fd->fd_refcnt > 1) {
293 struct filedesc *newfd;
294 error = fdcopy(p1, &newfd);
308 * Interlock against process group signal delivery. If signals
309 * are pending after the interlock is obtained we have to restart
310 * the system call to process the signals. If we don't the child
311 * can miss a pgsignal (such as ^C) sent during the fork.
313 * We can't use CURSIG() here because it will process any STOPs
314 * and cause the process group lock to be held indefinitely. If
315 * a STOP occurs, the fork will be restarted after the CONT.
318 if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) {
320 lockmgr(&plkgrp->pg_lock, LK_SHARED);
321 if (CURSIG_NOBLOCK(lp1)) {
328 * Although process entries are dynamically created, we still keep
329 * a global limit on the maximum number we will create. Don't allow
330 * a nonprivileged user to use the last ten processes; don't let root
331 * exceed the limit. The variable nprocs is the current number of
332 * processes, maxproc is the limit.
334 uid = lp1->lwp_thread->td_ucred->cr_ruid;
335 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
336 if (ppsratecheck(&lastfail, &curfail, 1))
337 kprintf("maxproc limit exceeded by uid %d, please "
338 "see tuning(7) and login.conf(5).\n", uid);
339 tsleep(&forksleep, 0, "fork", hz / 2);
345 * Increment the nprocs resource before blocking can occur. There
346 * are hard-limits as to the number of processes that can run.
348 atomic_add_int(&nprocs, 1);
351 * Increment the count of procs running with this uid. Don't allow
352 * a nonprivileged user to exceed their current limit.
354 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
355 (uid != 0) ? p1->p_rlimit[RLIMIT_NPROC].rlim_cur : 0);
358 * Back out the process count
360 atomic_add_int(&nprocs, -1);
361 if (ppsratecheck(&lastfail, &curfail, 1))
362 kprintf("maxproc limit exceeded by uid %d, please "
363 "see tuning(7) and login.conf(5).\n", uid);
364 tsleep(&forksleep, 0, "fork", hz / 2);
370 * Allocate a new process, don't get fancy: zero the structure.
372 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
375 * Core initialization. SIDL is a safety state that protects the
376 * partially initialized process once it starts getting hooked
377 * into system structures and becomes addressable.
379 * We must be sure to acquire p2->p_token as well, we must hold it
380 * once the process is on the allproc list to avoid things such
381 * as competing modifications to p_flags.
383 mycpu->gd_forkid += ncpus;
384 p2->p_forkid = mycpu->gd_forkid + mycpu->gd_cpuid;
385 p2->p_lasttid = -1; /* first tid will be 0 */
389 * NOTE: Process 0 will not have a reaper, but process 1 (init) and
390 * all other processes always will.
392 if ((p2->p_reaper = p1->p_reaper) != NULL)
393 reaper_hold(p2->p_reaper);
395 RB_INIT(&p2->p_lwp_tree);
396 spin_init(&p2->p_spin, "procfork1");
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;
513 p1->p_upmap->invfork = 1;
518 * Inherit the virtual kernel structure (allows a virtual kernel
519 * to fork to simulate multiple cpus).
522 vkernel_inherit(p1, p2);
525 * Once we are on a pglist we may receive signals. XXX we might
526 * race a ^C being sent to the process group by not receiving it
527 * at all prior to this line.
530 lwkt_gettoken(&p1grp->pg_token);
531 LIST_INSERT_AFTER(p1, p2, p_pglist);
532 lwkt_reltoken(&p1grp->pg_token);
535 * Attach the new process to its parent.
537 * If RFNOWAIT is set, the newly created process becomes a child
538 * of init. This effectively disassociates the child from the
541 if (flags & RFNOWAIT)
546 LIST_INIT(&p2->p_children);
548 lwkt_gettoken(&pptr->p_token);
549 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
550 lwkt_reltoken(&pptr->p_token);
552 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
553 callout_init_mp(&p2->p_ithandle);
557 * Copy traceflag and tracefile if enabled. If not inherited,
558 * these were zeroed above but we still could have a trace race
559 * so make sure p2's p_tracenode is NULL.
561 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
562 p2->p_traceflag = p1->p_traceflag;
563 p2->p_tracenode = ktrinherit(p1->p_tracenode);
568 * This begins the section where we must prevent the parent
569 * from being swapped.
571 * Gets PRELE'd in the caller in start_forked_proc().
575 vm_fork(p1, p2, flags);
578 * Create the first lwp associated with the new proc.
579 * It will return via a different execution path later, directly
580 * into userland, after it was put on the runq by
581 * start_forked_proc().
583 lwp_fork(lp1, p2, flags);
585 if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
586 mycpu->gd_cnt.v_forks++;
587 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize +
588 p2->p_vmspace->vm_ssize;
589 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
590 mycpu->gd_cnt.v_vforks++;
591 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
592 p2->p_vmspace->vm_ssize;
593 } else if (p1 == &proc0) {
594 mycpu->gd_cnt.v_kthreads++;
595 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
596 p2->p_vmspace->vm_ssize;
598 mycpu->gd_cnt.v_rforks++;
599 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
600 p2->p_vmspace->vm_ssize;
604 * Both processes are set up, now check if any loadable modules want
605 * to adjust anything.
606 * What if they have an error? XXX
608 TAILQ_FOREACH(ep, &fork_list, next) {
609 (*ep->function)(p1, p2, flags);
613 * Set the start time. Note that the process is not runnable. The
614 * caller is responsible for making it runnable.
616 microtime(&p2->p_start);
617 p2->p_acflag = AFORK;
620 * tell any interested parties about the new process
622 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
625 * Return child proc pointer to parent.
631 lwkt_reltoken(&p2->p_token);
632 lwkt_reltoken(&p1->p_token);
634 lockmgr(&plkgrp->pg_lock, LK_RELEASE);
641 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags)
643 globaldata_t gd = mycpu;
647 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
649 lp->lwp_proc = destproc;
650 lp->lwp_vmspace = destproc->p_vmspace;
651 lp->lwp_stat = LSRUN;
652 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
653 (unsigned) ((caddr_t)&lp->lwp_endcopy -
654 (caddr_t)&lp->lwp_startcopy));
655 lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK;
657 * Set cpbase to the last timeout that occured (not the upcoming
660 * A critical section is required since a timer IPI can update
661 * scheduler specific data.
664 lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic;
665 destproc->p_usched->heuristic_forking(origlp, lp);
667 CPUMASK_ANDMASK(lp->lwp_cpumask, usched_mastermask);
668 lwkt_token_init(&lp->lwp_token, "lwp_token");
669 spin_init(&lp->lwp_spin, "lwptoken");
672 * Assign the thread to the current cpu to begin with so we
675 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0);
677 td->td_ucred = crhold(destproc->p_ucred);
678 td->td_proc = destproc;
680 td->td_switch = cpu_heavy_switch;
681 #ifdef NO_LWKT_SPLIT_USERPRI
682 lwkt_setpri(td, TDPRI_USER_NORM);
684 lwkt_setpri(td, TDPRI_KERN_USER);
686 lwkt_set_comm(td, "%s", destproc->p_comm);
689 * cpu_fork will copy and update the pcb, set up the kernel stack,
690 * and make the child ready to run.
692 cpu_fork(origlp, lp, flags);
693 kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
696 * Assign a TID to the lp. Loop until the insert succeeds (returns
699 lp->lwp_tid = destproc->p_lasttid;
701 if (++lp->lwp_tid < 0)
703 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
704 destproc->p_lasttid = lp->lwp_tid;
705 destproc->p_nthreads++;
708 * This flag is set and never cleared. It means that the process
709 * was threaded at some point. Used to improve exit performance.
711 destproc->p_flags |= P_MAYBETHREADED;
717 * The next two functionms are general routines to handle adding/deleting
718 * items on the fork callout list.
721 * Take the arguments given and put them onto the fork callout list,
722 * However first make sure that it's not already there.
723 * Returns 0 on success or a standard error number.
726 at_fork(forklist_fn function)
731 /* let the programmer know if he's been stupid */
732 if (rm_at_fork(function)) {
733 kprintf("WARNING: fork callout entry (%p) already present\n",
737 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
738 ep->function = function;
739 TAILQ_INSERT_TAIL(&fork_list, ep, next);
744 * Scan the exit callout list for the given item and remove it..
745 * Returns the number of items removed (0 or 1)
748 rm_at_fork(forklist_fn function)
752 TAILQ_FOREACH(ep, &fork_list, next) {
753 if (ep->function == function) {
754 TAILQ_REMOVE(&fork_list, ep, next);
763 * Add a forked process to the run queue after any remaining setup, such
764 * as setting the fork handler, has been completed.
766 * p2 is held by the caller.
769 start_forked_proc(struct lwp *lp1, struct proc *p2)
771 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
775 * Move from SIDL to RUN queue, and activate the process's thread.
776 * Activation of the thread effectively makes the process "a"
777 * current process, so we do not setrunqueue().
779 * YYY setrunqueue works here but we should clean up the trampoline
780 * code so we just schedule the LWKT thread and let the trampoline
781 * deal with the userland scheduler on return to userland.
783 KASSERT(p2->p_stat == SIDL,
784 ("cannot start forked process, bad status: %p", p2));
785 p2->p_usched->resetpriority(lp2);
787 p2->p_stat = SACTIVE;
788 lp2->lwp_stat = LSRUN;
789 p2->p_usched->setrunqueue(lp2);
793 * Now can be swapped.
795 PRELE(lp1->lwp_proc);
798 * Preserve synchronization semantics of vfork. P_PPWAIT is set in
799 * the child until it has retired the parent's resources. The parent
800 * must wait for the flag to be cleared by the child.
802 * Interlock the flag/tsleep with atomic ops to avoid unnecessary
805 * XXX Is this use of an atomic op on a field that is not normally
806 * manipulated with atomic ops ok?
808 while ((pflags = p2->p_flags) & P_PPWAIT) {
810 tsleep_interlock(lp1->lwp_proc, 0);
811 if (atomic_cmpset_int(&p2->p_flags, pflags, pflags))
812 tsleep(lp1->lwp_proc, PINTERLOCKED, "ppwait", 0);
817 * procctl (idtype_t idtype, id_t id, int cmd, void *arg)
820 sys_procctl(struct procctl_args *uap)
822 struct proc *p = curproc;
824 struct sysreaper *reap;
825 union reaper_info udata;
828 if (uap->idtype != P_PID || uap->id != (id_t)p->p_pid)
832 case PROC_REAP_ACQUIRE:
833 lwkt_gettoken(&p->p_token);
834 reap = kmalloc(sizeof(*reap), M_REAPER, M_WAITOK|M_ZERO);
835 if (p->p_reaper == NULL || p->p_reaper->p != p) {
836 reaper_init(p, reap);
839 kfree(reap, M_REAPER);
842 lwkt_reltoken(&p->p_token);
844 case PROC_REAP_RELEASE:
845 lwkt_gettoken(&p->p_token);
848 KKASSERT(reap != NULL);
850 reaper_hold(reap); /* in case of thread race */
851 lockmgr(&reap->lock, LK_EXCLUSIVE);
853 lockmgr(&reap->lock, LK_RELEASE);
858 p->p_reaper = reap->parent;
860 reaper_hold(p->p_reaper);
861 lockmgr(&reap->lock, LK_RELEASE);
862 reaper_drop(reap); /* our ref */
863 reaper_drop(reap); /* old p_reaper ref */
868 lwkt_reltoken(&p->p_token);
870 case PROC_REAP_STATUS:
871 bzero(&udata, sizeof(udata));
872 lwkt_gettoken_shared(&p->p_token);
873 if ((reap = p->p_reaper) != NULL && reap->p == p) {
874 udata.status.flags = reap->flags;
875 udata.status.refs = reap->refs - 1; /* minus ours */
877 p2 = LIST_FIRST(&p->p_children);
878 udata.status.pid_head = p2 ? p2->p_pid : -1;
879 lwkt_reltoken(&p->p_token);
882 error = copyout(&udata, uap->data,
883 sizeof(udata.status));
896 * Bump ref on reaper, preventing destruction
899 reaper_hold(struct sysreaper *reap)
901 KKASSERT(reap->refs > 0);
902 refcount_acquire(&reap->refs);
906 * Drop ref on reaper, destroy the structure on the 1->0
907 * transition and loop on the parent.
910 reaper_drop(struct sysreaper *next)
912 struct sysreaper *reap;
914 while ((reap = next) != NULL) {
915 if (refcount_release(&reap->refs)) {
917 KKASSERT(reap->p == NULL);
919 kfree(reap, M_REAPER);
927 * Initialize a static or newly allocated reaper structure
930 reaper_init(struct proc *p, struct sysreaper *reap)
932 reap->parent = p->p_reaper;
935 reap->flags = REAPER_STAT_OWNED | REAPER_STAT_REALINIT;
938 reap->flags = REAPER_STAT_OWNED;
941 lockinit(&reap->lock, "subrp", 0, 0);
947 * Called with p->p_token held during exit.
949 * This is a bit simpler than RELEASE because there are no threads remaining
950 * to race. We only release if we own the reaper, the exit code will handle
951 * the final p_reaper release.
954 reaper_exit(struct proc *p)
956 struct sysreaper *reap;
959 * Release acquired reaper
961 if ((reap = p->p_reaper) != NULL && reap->p == p) {
962 lockmgr(&reap->lock, LK_EXCLUSIVE);
963 p->p_reaper = reap->parent;
965 reaper_hold(p->p_reaper);
967 lockmgr(&reap->lock, LK_RELEASE);
972 * Return and clear reaper (caller is holding p_token for us)
973 * (reap->p does not equal p). Caller must drop it.
975 if ((reap = p->p_reaper) != NULL) {
982 * Return a held (PHOLD) process representing the reaper for process (p).
983 * NULL should not normally be returned. Caller should PRELE() the returned
984 * reaper process when finished.
986 * Remove dead internal nodes while we are at it.
988 * Process (p)'s token must be held on call.
989 * The returned process's token is NOT acquired by this routine.
992 reaper_get(struct sysreaper *reap)
994 struct sysreaper *next;
1001 * Extra hold for loop
1006 lockmgr(&reap->lock, LK_SHARED);
1014 lockmgr(&reap->lock, LK_RELEASE);
1022 lockmgr(&reap->lock, LK_RELEASE);
1027 * Traverse upwards in the reaper topology, destroy
1028 * dead internal nodes when possible.
1030 * NOTE: Our ref on next means that a dead node should
1031 * have 2 (ours and reap->parent's).
1033 next = reap->parent;
1036 if (next->refs == 2 && next->p == NULL) {
1037 lockmgr(&reap->lock, LK_RELEASE);
1038 lockmgr(&reap->lock, LK_EXCLUSIVE);
1039 if (next->refs == 2 &&
1040 reap->parent == next &&
1043 * reap->parent inherits ref from next.
1045 reap->parent = next->parent;
1046 next->parent = NULL;
1047 reaper_drop(next); /* ours */
1048 reaper_drop(next); /* old parent */
1049 next = reap->parent;
1050 continue; /* possible chain */
1055 lockmgr(&reap->lock, LK_RELEASE);