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);
113 sys_fork(struct fork_args *uap)
115 struct lwp *lp = curthread->td_lwp;
119 error = fork1(lp, RFFDG | RFPROC | RFPGLOCK, &p2);
122 start_forked_proc(lp, p2);
123 uap->sysmsg_fds[0] = p2->p_pid;
124 uap->sysmsg_fds[1] = 0;
131 * vfork() system call
134 sys_vfork(struct vfork_args *uap)
136 struct lwp *lp = curthread->td_lwp;
140 error = fork1(lp, RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK, &p2);
143 start_forked_proc(lp, p2);
144 uap->sysmsg_fds[0] = p2->p_pid;
145 uap->sysmsg_fds[1] = 0;
152 * Handle rforks. An rfork may (1) operate on the current process without
153 * creating a new, (2) create a new process that shared the current process's
154 * vmspace, signals, and/or descriptors, or (3) create a new process that does
155 * not share these things (normal fork).
157 * Note that we only call start_forked_proc() if a new process is actually
160 * rfork { int flags }
163 sys_rfork(struct rfork_args *uap)
165 struct lwp *lp = curthread->td_lwp;
169 if ((uap->flags & RFKERNELONLY) != 0)
172 error = fork1(lp, uap->flags | RFPGLOCK, &p2);
176 start_forked_proc(lp, p2);
177 uap->sysmsg_fds[0] = p2->p_pid;
178 uap->sysmsg_fds[1] = 0;
181 uap->sysmsg_fds[0] = 0;
182 uap->sysmsg_fds[1] = 0;
189 lwp_create1(struct lwp_params *uprm, const cpumask_t *umask)
191 struct proc *p = curproc;
193 struct lwp_params params;
194 cpumask_t *mask = NULL, mask0;
197 error = copyin(uprm, ¶ms, sizeof(params));
202 error = copyin(umask, &mask0, sizeof(mask0));
205 CPUMASK_ANDMASK(mask0, smp_active_mask);
206 if (CPUMASK_TESTNZERO(mask0))
210 lwkt_gettoken(&p->p_token);
211 plimit_lwp_fork(p); /* force exclusive access */
212 lp = lwp_fork(curthread->td_lwp, p, RFPROC | RFMEM, mask);
213 error = cpu_prepare_lwp(lp, ¶ms);
216 if (params.lwp_tid1 != NULL &&
217 (error = copyout(&lp->lwp_tid, params.lwp_tid1, sizeof(lp->lwp_tid))))
219 if (params.lwp_tid2 != NULL &&
220 (error = copyout(&lp->lwp_tid, params.lwp_tid2, sizeof(lp->lwp_tid))))
224 * Now schedule the new lwp.
226 p->p_usched->resetpriority(lp);
228 lp->lwp_stat = LSRUN;
229 p->p_usched->setrunqueue(lp);
231 lwkt_reltoken(&p->p_token);
237 * Make sure no one is using this lwp, before it is removed from
238 * the tree. If we didn't wait it here, lwp tree iteration with
239 * blocking operation would be broken.
241 while (lp->lwp_lock > 0)
242 tsleep(lp, 0, "lwpfail", 1);
243 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
245 /* lwp_dispose expects an exited lwp, and a held proc */
246 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
247 lp->lwp_thread->td_flags |= TDF_EXITING;
248 lwkt_remove_tdallq(lp->lwp_thread);
250 biosched_done(lp->lwp_thread);
251 dsched_exit_thread(lp->lwp_thread);
253 lwkt_reltoken(&p->p_token);
259 * Low level thread create used by pthreads.
262 sys_lwp_create(struct lwp_create_args *uap)
265 return (lwp_create1(uap->params, NULL));
269 sys_lwp_create2(struct lwp_create2_args *uap)
272 return (lwp_create1(uap->params, uap->mask));
275 int nprocs = 1; /* process 0 */
278 fork1(struct lwp *lp1, int flags, struct proc **procp)
280 struct proc *p1 = lp1->lwp_proc;
285 struct sysreaper *reap;
288 static int curfail = 0;
289 static struct timeval lastfail;
291 struct filedesc_to_leader *fdtol;
293 if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
296 lwkt_gettoken(&p1->p_token);
301 * Here we don't create a new process, but we divorce
302 * certain parts of a process from itself.
304 if ((flags & RFPROC) == 0) {
306 * This kind of stunt does not work anymore if
307 * there are native threads (lwps) running
309 if (p1->p_nthreads != 1) {
314 vm_fork(p1, 0, flags);
317 * Close all file descriptors.
319 if (flags & RFCFDG) {
320 struct filedesc *fdtmp;
326 * Unshare file descriptors (from parent.)
329 if (p1->p_fd->fd_refcnt > 1) {
330 struct filedesc *newfd;
331 error = fdcopy(p1, &newfd);
345 * Interlock against process group signal delivery. If signals
346 * are pending after the interlock is obtained we have to restart
347 * the system call to process the signals. If we don't the child
348 * can miss a pgsignal (such as ^C) sent during the fork.
350 * We can't use CURSIG() here because it will process any STOPs
351 * and cause the process group lock to be held indefinitely. If
352 * a STOP occurs, the fork will be restarted after the CONT.
355 if ((flags & RFPGLOCK) && (plkgrp = p1->p_pgrp) != NULL) {
357 lockmgr(&plkgrp->pg_lock, LK_SHARED);
358 if (CURSIG_NOBLOCK(lp1)) {
365 * Although process entries are dynamically created, we still keep
366 * a global limit on the maximum number we will create. Don't allow
367 * a nonprivileged user to use the last ten processes; don't let root
368 * exceed the limit. The variable nprocs is the current number of
369 * processes, maxproc is the limit.
371 uid = lp1->lwp_thread->td_ucred->cr_ruid;
372 if ((nprocs >= maxproc - 10 && uid != 0) || nprocs >= maxproc) {
373 if (ppsratecheck(&lastfail, &curfail, 1))
374 kprintf("maxproc limit exceeded by uid %d, please "
375 "see tuning(7) and login.conf(5).\n", uid);
376 tsleep(&forksleep, 0, "fork", hz / 2);
382 * Increment the nprocs resource before blocking can occur. There
383 * are hard-limits as to the number of processes that can run.
385 atomic_add_int(&nprocs, 1);
388 * Increment the count of procs running with this uid. This also
391 ok = chgproccnt(lp1->lwp_thread->td_ucred->cr_ruidinfo, 1,
392 plimit_getadjvalue(RLIMIT_NPROC));
395 * Back out the process count
397 atomic_add_int(&nprocs, -1);
398 if (ppsratecheck(&lastfail, &curfail, 1)) {
399 kprintf("maxproc limit of %jd "
400 "exceeded by \"%s\" uid %d, "
401 "please see tuning(7) and login.conf(5).\n",
402 plimit_getadjvalue(RLIMIT_NPROC),
406 tsleep(&forksleep, 0, "fork", hz / 2);
412 * Allocate a new process, don't get fancy: zero the structure.
414 p2 = kmalloc(sizeof(struct proc), M_PROC, M_WAITOK|M_ZERO);
417 * Core initialization. SIDL is a safety state that protects the
418 * partially initialized process once it starts getting hooked
419 * into system structures and becomes addressable.
421 * We must be sure to acquire p2->p_token as well, we must hold it
422 * once the process is on the allproc list to avoid things such
423 * as competing modifications to p_flags.
425 mycpu->gd_forkid += ncpus;
426 p2->p_forkid = mycpu->gd_forkid + mycpu->gd_cpuid;
427 p2->p_lasttid = 0; /* first tid will be 1 */
431 * NOTE: Process 0 will not have a reaper, but process 1 (init) and
432 * all other processes always will.
434 if ((reap = p1->p_reaper) != NULL) {
441 RB_INIT(&p2->p_lwp_tree);
442 spin_init(&p2->p_spin, "procfork1");
443 lwkt_token_init(&p2->p_token, "proc");
444 lwkt_gettoken(&p2->p_token);
447 * Setup linkage for kernel based threading XXX lwp. Also add the
448 * process to the allproclist.
450 * The process structure is addressable after this point.
452 if (flags & RFTHREAD) {
453 p2->p_peers = p1->p_peers;
455 p2->p_leader = p1->p_leader;
459 proc_add_allproc(p2);
462 * Initialize the section which is copied verbatim from the parent.
464 bcopy(&p1->p_startcopy, &p2->p_startcopy,
465 ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy));
468 * Duplicate sub-structures as needed. Increase reference counts
471 * NOTE: because we are now on the allproc list it is possible for
472 * other consumers to gain temporary references to p2
473 * (p2->p_lock can change).
475 if (p1->p_flags & P_PROFIL)
477 p2->p_ucred = crhold(lp1->lwp_thread->td_ucred);
479 if (jailed(p2->p_ucred))
480 p2->p_flags |= P_JAILED;
483 refcount_acquire(&p2->p_args->ar_ref);
485 p2->p_usched = p1->p_usched;
486 /* XXX: verify copy of the secondary iosched stuff */
487 dsched_enter_proc(p2);
489 if (flags & RFSIGSHARE) {
490 p2->p_sigacts = p1->p_sigacts;
491 refcount_acquire(&p2->p_sigacts->ps_refcnt);
493 p2->p_sigacts = kmalloc(sizeof(*p2->p_sigacts),
494 M_SUBPROC, M_WAITOK);
495 bcopy(p1->p_sigacts, p2->p_sigacts, sizeof(*p2->p_sigacts));
496 refcount_init(&p2->p_sigacts->ps_refcnt, 1);
498 if (flags & RFLINUXTHPN)
499 p2->p_sigparent = SIGUSR1;
501 p2->p_sigparent = SIGCHLD;
503 /* bump references to the text vnode (for procfs) */
504 p2->p_textvp = p1->p_textvp;
508 /* copy namecache handle to the text file */
509 if (p1->p_textnch.mount)
510 cache_copy(&p1->p_textnch, &p2->p_textnch);
513 * Handle file descriptors
515 if (flags & RFCFDG) {
516 p2->p_fd = fdinit(p1);
518 } else if (flags & RFFDG) {
519 error = fdcopy(p1, &p2->p_fd);
526 p2->p_fd = fdshare(p1);
527 if (p1->p_fdtol == NULL) {
528 p1->p_fdtol = filedesc_to_leader_alloc(NULL,
531 if ((flags & RFTHREAD) != 0) {
533 * Shared file descriptor table and
534 * shared process leaders.
537 fdtol->fdl_refcount++;
540 * Shared file descriptor table, and
541 * different process leaders
543 fdtol = filedesc_to_leader_alloc(p1->p_fdtol, p2);
547 p2->p_limit = plimit_fork(p1);
550 * Adjust depth for resource downscaling
552 if ((p2->p_depth & 31) != 31)
556 * Preserve some more flags in subprocess. P_PROFIL has already
559 p2->p_flags |= p1->p_flags & P_SUGID;
560 if (p1->p_session->s_ttyvp != NULL && (p1->p_flags & P_CONTROLT))
561 p2->p_flags |= P_CONTROLT;
562 if (flags & RFPPWAIT) {
563 p2->p_flags |= P_PPWAIT;
565 atomic_add_int(&p1->p_upmap->invfork, 1);
569 * Inherit the virtual kernel structure (allows a virtual kernel
570 * to fork to simulate multiple cpus).
573 vkernel_inherit(p1, p2);
576 * Once we are on a pglist we may receive signals. XXX we might
577 * race a ^C being sent to the process group by not receiving it
578 * at all prior to this line.
581 lwkt_gettoken(&p1grp->pg_token);
582 LIST_INSERT_AFTER(p1, p2, p_pglist);
583 lwkt_reltoken(&p1grp->pg_token);
586 * Attach the new process to its parent.
588 * If RFNOWAIT is set, the newly created process becomes a child
589 * of the reaper (typically init). This effectively disassociates
590 * the child from the parent.
592 * Temporarily hold pptr for the RFNOWAIT case to avoid ripouts.
594 if (flags & RFNOWAIT) {
595 pptr = reaper_get(reap);
604 LIST_INIT(&p2->p_children);
606 lwkt_gettoken(&pptr->p_token);
607 LIST_INSERT_HEAD(&pptr->p_children, p2, p_sibling);
608 lwkt_reltoken(&pptr->p_token);
610 if (flags & RFNOWAIT)
613 varsymset_init(&p2->p_varsymset, &p1->p_varsymset);
614 callout_init_mp(&p2->p_ithandle);
618 * Copy traceflag and tracefile if enabled. If not inherited,
619 * these were zeroed above but we still could have a trace race
620 * so make sure p2's p_tracenode is NULL.
622 if ((p1->p_traceflag & KTRFAC_INHERIT) && p2->p_tracenode == NULL) {
623 p2->p_traceflag = p1->p_traceflag;
624 p2->p_tracenode = ktrinherit(p1->p_tracenode);
629 * This begins the section where we must prevent the parent
630 * from being swapped.
632 * Gets PRELE'd in the caller in start_forked_proc().
636 vm_fork(p1, p2, flags);
639 * Create the first lwp associated with the new proc.
640 * It will return via a different execution path later, directly
641 * into userland, after it was put on the runq by
642 * start_forked_proc().
644 lwp_fork(lp1, p2, flags, NULL);
646 if (flags == (RFFDG | RFPROC | RFPGLOCK)) {
647 mycpu->gd_cnt.v_forks++;
648 mycpu->gd_cnt.v_forkpages += p2->p_vmspace->vm_dsize +
649 p2->p_vmspace->vm_ssize;
650 } else if (flags == (RFFDG | RFPROC | RFPPWAIT | RFMEM | RFPGLOCK)) {
651 mycpu->gd_cnt.v_vforks++;
652 mycpu->gd_cnt.v_vforkpages += p2->p_vmspace->vm_dsize +
653 p2->p_vmspace->vm_ssize;
654 } else if (p1 == &proc0) {
655 mycpu->gd_cnt.v_kthreads++;
656 mycpu->gd_cnt.v_kthreadpages += p2->p_vmspace->vm_dsize +
657 p2->p_vmspace->vm_ssize;
659 mycpu->gd_cnt.v_rforks++;
660 mycpu->gd_cnt.v_rforkpages += p2->p_vmspace->vm_dsize +
661 p2->p_vmspace->vm_ssize;
665 * Both processes are set up, now check if any loadable modules want
666 * to adjust anything.
667 * What if they have an error? XXX
669 TAILQ_FOREACH(ep, &fork_list, next) {
670 (*ep->function)(p1, p2, flags);
674 * Set the start time. Note that the process is not runnable. The
675 * caller is responsible for making it runnable.
677 microtime(&p2->p_start);
678 p2->p_acflag = AFORK;
681 * tell any interested parties about the new process
683 KNOTE(&p1->p_klist, NOTE_FORK | p2->p_pid);
686 * Return child proc pointer to parent.
692 lwkt_reltoken(&p2->p_token);
693 lwkt_reltoken(&p1->p_token);
695 lockmgr(&plkgrp->pg_lock, LK_RELEASE);
702 lwp_fork(struct lwp *origlp, struct proc *destproc, int flags,
703 const cpumask_t *mask)
705 globaldata_t gd = mycpu;
709 lp = kmalloc(sizeof(struct lwp), M_LWP, M_WAITOK|M_ZERO);
711 lp->lwp_proc = destproc;
712 lp->lwp_vmspace = destproc->p_vmspace;
713 lp->lwp_stat = LSRUN;
714 bcopy(&origlp->lwp_startcopy, &lp->lwp_startcopy,
715 (unsigned) ((caddr_t)&lp->lwp_endcopy -
716 (caddr_t)&lp->lwp_startcopy));
718 lp->lwp_cpumask = *mask;
721 * Reset the sigaltstack if memory is shared, otherwise inherit
725 lp->lwp_sigstk.ss_flags = SS_DISABLE;
726 lp->lwp_sigstk.ss_size = 0;
727 lp->lwp_sigstk.ss_sp = NULL;
728 lp->lwp_flags &= ~LWP_ALTSTACK;
730 lp->lwp_flags |= origlp->lwp_flags & LWP_ALTSTACK;
734 * Set cpbase to the last timeout that occured (not the upcoming
737 * A critical section is required since a timer IPI can update
738 * scheduler specific data.
741 lp->lwp_cpbase = gd->gd_schedclock.time - gd->gd_schedclock.periodic;
742 destproc->p_usched->heuristic_forking(origlp, lp);
744 CPUMASK_ANDMASK(lp->lwp_cpumask, usched_mastermask);
745 lwkt_token_init(&lp->lwp_token, "lwp_token");
746 spin_init(&lp->lwp_spin, "lwptoken");
749 * Assign the thread to the current cpu to begin with so we
752 td = lwkt_alloc_thread(NULL, LWKT_THREAD_STACK, gd->gd_cpuid, 0);
754 td->td_ucred = crhold(destproc->p_ucred);
755 td->td_proc = destproc;
757 td->td_switch = cpu_heavy_switch;
758 #ifdef NO_LWKT_SPLIT_USERPRI
759 lwkt_setpri(td, TDPRI_USER_NORM);
761 lwkt_setpri(td, TDPRI_KERN_USER);
763 lwkt_set_comm(td, "%s", destproc->p_comm);
766 * cpu_fork will copy and update the pcb, set up the kernel stack,
767 * and make the child ready to run.
769 cpu_fork(origlp, lp, flags);
770 kqueue_init(&lp->lwp_kqueue, destproc->p_fd);
773 * Assign a TID to the lp. Loop until the insert succeeds (returns
776 * If we are in a vfork assign the same TID as the lwp that did the
777 * vfork(). This way if the user program messes around with
778 * pthread calls inside the vfork(), it will operate like an
779 * extension of the (blocked) parent. Also note that since the
780 * address space is being shared, insofar as pthreads is concerned,
781 * the code running in the vfork() is part of the original process.
783 if (flags & RFPPWAIT) {
784 lp->lwp_tid = origlp->lwp_tid - 1;
786 lp->lwp_tid = destproc->p_lasttid;
790 if (++lp->lwp_tid <= 0)
792 } while (lwp_rb_tree_RB_INSERT(&destproc->p_lwp_tree, lp) != NULL);
794 destproc->p_lasttid = lp->lwp_tid;
795 destproc->p_nthreads++;
798 * This flag is set and never cleared. It means that the process
799 * was threaded at some point. Used to improve exit performance.
801 destproc->p_flags |= P_MAYBETHREADED;
807 * The next two functionms are general routines to handle adding/deleting
808 * items on the fork callout list.
811 * Take the arguments given and put them onto the fork callout list,
812 * However first make sure that it's not already there.
813 * Returns 0 on success or a standard error number.
816 at_fork(forklist_fn function)
821 /* let the programmer know if he's been stupid */
822 if (rm_at_fork(function)) {
823 kprintf("WARNING: fork callout entry (%p) already present\n",
827 ep = kmalloc(sizeof(*ep), M_ATFORK, M_WAITOK|M_ZERO);
828 ep->function = function;
829 TAILQ_INSERT_TAIL(&fork_list, ep, next);
834 * Scan the exit callout list for the given item and remove it..
835 * Returns the number of items removed (0 or 1)
838 rm_at_fork(forklist_fn function)
842 TAILQ_FOREACH(ep, &fork_list, next) {
843 if (ep->function == function) {
844 TAILQ_REMOVE(&fork_list, ep, next);
853 * Add a forked process to the run queue after any remaining setup, such
854 * as setting the fork handler, has been completed.
856 * p2 is held by the caller.
859 start_forked_proc(struct lwp *lp1, struct proc *p2)
861 struct lwp *lp2 = ONLY_LWP_IN_PROC(p2);
865 * Move from SIDL to RUN queue, and activate the process's thread.
866 * Activation of the thread effectively makes the process "a"
867 * current process, so we do not setrunqueue().
869 * YYY setrunqueue works here but we should clean up the trampoline
870 * code so we just schedule the LWKT thread and let the trampoline
871 * deal with the userland scheduler on return to userland.
873 KASSERT(p2->p_stat == SIDL,
874 ("cannot start forked process, bad status: %p", p2));
875 p2->p_usched->resetpriority(lp2);
877 p2->p_stat = SACTIVE;
878 lp2->lwp_stat = LSRUN;
879 p2->p_usched->setrunqueue(lp2);
883 * Now can be swapped.
885 PRELE(lp1->lwp_proc);
888 * Preserve synchronization semantics of vfork. P_PPWAIT is set in
889 * the child until it has retired the parent's resources. The parent
890 * must wait for the flag to be cleared by the child.
892 * Interlock the flag/tsleep with atomic ops to avoid unnecessary
895 * XXX Is this use of an atomic op on a field that is not normally
896 * manipulated with atomic ops ok?
898 while ((pflags = p2->p_flags) & P_PPWAIT) {
900 tsleep_interlock(lp1->lwp_proc, 0);
901 if (atomic_cmpset_int(&p2->p_flags, pflags, pflags))
902 tsleep(lp1->lwp_proc, PINTERLOCKED, "ppwait", 0);
907 * procctl (idtype_t idtype, id_t id, int cmd, void *arg)
910 sys_procctl(struct procctl_args *uap)
912 struct proc *p = curproc;
914 struct sysreaper *reap;
915 union reaper_info udata;
918 if (uap->idtype != P_PID || uap->id != (id_t)p->p_pid)
922 case PROC_REAP_ACQUIRE:
923 lwkt_gettoken(&p->p_token);
924 reap = kmalloc(sizeof(*reap), M_REAPER, M_WAITOK|M_ZERO);
925 if (p->p_reaper == NULL || p->p_reaper->p != p) {
926 reaper_init(p, reap);
929 kfree(reap, M_REAPER);
932 lwkt_reltoken(&p->p_token);
934 case PROC_REAP_RELEASE:
935 lwkt_gettoken(&p->p_token);
938 KKASSERT(reap != NULL);
940 reaper_hold(reap); /* in case of thread race */
941 lockmgr(&reap->lock, LK_EXCLUSIVE);
943 lockmgr(&reap->lock, LK_RELEASE);
948 p->p_reaper = reap->parent;
950 reaper_hold(p->p_reaper);
951 lockmgr(&reap->lock, LK_RELEASE);
952 reaper_drop(reap); /* our ref */
953 reaper_drop(reap); /* old p_reaper ref */
958 lwkt_reltoken(&p->p_token);
960 case PROC_REAP_STATUS:
961 bzero(&udata, sizeof(udata));
962 lwkt_gettoken_shared(&p->p_token);
963 if ((reap = p->p_reaper) != NULL && reap->p == p) {
964 udata.status.flags = reap->flags;
965 udata.status.refs = reap->refs - 1; /* minus ours */
967 p2 = LIST_FIRST(&p->p_children);
968 udata.status.pid_head = p2 ? p2->p_pid : -1;
969 lwkt_reltoken(&p->p_token);
972 error = copyout(&udata, uap->data,
973 sizeof(udata.status));
986 * Bump ref on reaper, preventing destruction
989 reaper_hold(struct sysreaper *reap)
991 KKASSERT(reap->refs > 0);
992 refcount_acquire(&reap->refs);
996 * Drop ref on reaper, destroy the structure on the 1->0
997 * transition and loop on the parent.
1000 reaper_drop(struct sysreaper *next)
1002 struct sysreaper *reap;
1004 while ((reap = next) != NULL) {
1005 if (refcount_release(&reap->refs)) {
1006 next = reap->parent;
1007 KKASSERT(reap->p == NULL);
1008 lockmgr(&reaper_lock, LK_EXCLUSIVE);
1009 reap->parent = NULL;
1010 kfree(reap, M_REAPER);
1011 lockmgr(&reaper_lock, LK_RELEASE);
1019 * Initialize a static or newly allocated reaper structure
1022 reaper_init(struct proc *p, struct sysreaper *reap)
1024 reap->parent = p->p_reaper;
1026 if (p == initproc) {
1027 reap->flags = REAPER_STAT_OWNED | REAPER_STAT_REALINIT;
1030 reap->flags = REAPER_STAT_OWNED;
1033 lockinit(&reap->lock, "subrp", 0, 0);
1039 * Called with p->p_token held during exit.
1041 * This is a bit simpler than RELEASE because there are no threads remaining
1042 * to race. We only release if we own the reaper, the exit code will handle
1043 * the final p_reaper release.
1046 reaper_exit(struct proc *p)
1048 struct sysreaper *reap;
1051 * Release acquired reaper
1053 if ((reap = p->p_reaper) != NULL && reap->p == p) {
1054 lockmgr(&reap->lock, LK_EXCLUSIVE);
1055 p->p_reaper = reap->parent;
1057 reaper_hold(p->p_reaper);
1059 lockmgr(&reap->lock, LK_RELEASE);
1064 * Return and clear reaper (caller is holding p_token for us)
1065 * (reap->p does not equal p). Caller must drop it.
1067 if ((reap = p->p_reaper) != NULL) {
1074 * Return a held (PHOLD) process representing the reaper for process (p).
1075 * NULL should not normally be returned. Caller should PRELE() the returned
1076 * reaper process when finished.
1078 * Remove dead internal nodes while we are at it.
1080 * Process (p)'s token must be held on call.
1081 * The returned process's token is NOT acquired by this routine.
1084 reaper_get(struct sysreaper *reap)
1086 struct sysreaper *next;
1087 struct proc *reproc;
1093 * Extra hold for loop
1098 lockmgr(&reap->lock, LK_SHARED);
1106 lockmgr(&reap->lock, LK_RELEASE);
1114 lockmgr(&reap->lock, LK_RELEASE);
1119 * Traverse upwards in the reaper topology, destroy
1120 * dead internal nodes when possible.
1122 * NOTE: Our ref on next means that a dead node should
1123 * have 2 (ours and reap->parent's).
1125 next = reap->parent;
1128 if (next->refs == 2 && next->p == NULL) {
1129 lockmgr(&reap->lock, LK_RELEASE);
1130 lockmgr(&reap->lock, LK_EXCLUSIVE);
1131 if (next->refs == 2 &&
1132 reap->parent == next &&
1135 * reap->parent inherits ref from next.
1137 reap->parent = next->parent;
1138 next->parent = NULL;
1139 reaper_drop(next); /* ours */
1140 reaper_drop(next); /* old parent */
1141 next = reap->parent;
1142 continue; /* possible chain */
1147 lockmgr(&reap->lock, LK_RELEASE);
1155 * Test that the sender is allowed to send a signal to the target.
1156 * The sender process is assumed to have a stable reaper. The
1157 * target can be e.g. from a scan callback.
1159 * Target cannot be the reaper process itself unless reaper_ok is specified,
1160 * or sender == target.
1163 reaper_sigtest(struct proc *sender, struct proc *target, int reaper_ok)
1165 struct sysreaper *sreap;
1166 struct sysreaper *reap;
1169 sreap = sender->p_reaper;
1173 if (sreap == target->p_reaper) {
1174 if (sreap->p == target && sreap->p != sender && reaper_ok == 0)
1178 lockmgr(&reaper_lock, LK_SHARED);
1180 for (reap = target->p_reaper; reap; reap = reap->parent) {
1181 if (sreap == reap) {
1182 if (sreap->p != target || reaper_ok)
1187 lockmgr(&reaper_lock, LK_RELEASE);