Kernel - Fix numerous procfs/ptrace issues
[dragonfly.git] / sys / kern / kern_exit.c
CommitLineData
984263bc
MD
1/*
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.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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.
25 *
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
36 * SUCH DAMAGE.
37 *
38 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94
39 * $FreeBSD: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $
40 */
41
42#include "opt_compat.h"
43#include "opt_ktrace.h"
44
45#include <sys/param.h>
46#include <sys/systm.h>
47#include <sys/sysproto.h>
48#include <sys/kernel.h>
49#include <sys/malloc.h>
50#include <sys/proc.h>
29f58392 51#include <sys/ktrace.h>
984263bc
MD
52#include <sys/pioctl.h>
53#include <sys/tty.h>
54#include <sys/wait.h>
55#include <sys/vnode.h>
56#include <sys/resourcevar.h>
57#include <sys/signalvar.h>
70d3d461 58#include <sys/taskqueue.h>
984263bc
MD
59#include <sys/ptrace.h>
60#include <sys/acct.h> /* for acct_process() function prototype */
61#include <sys/filedesc.h>
62#include <sys/shm.h>
63#include <sys/sem.h>
984263bc 64#include <sys/jail.h>
9697c509 65#include <sys/kern_syscall.h>
a722be49 66#include <sys/upcall.h>
f6bf3af1 67#include <sys/caps.h>
d5f340eb 68#include <sys/unistd.h>
8ba5f7ef 69#include <sys/eventhandler.h>
2883d2d8 70#include <sys/dsched.h>
984263bc
MD
71
72#include <vm/vm.h>
73#include <vm/vm_param.h>
74#include <sys/lock.h>
75#include <vm/pmap.h>
76#include <vm/vm_map.h>
984263bc
MD
77#include <vm/vm_extern.h>
78#include <sys/user.h>
79
e955543c 80#include <sys/refcount.h>
e43a034f 81#include <sys/thread2.h>
e3161323 82#include <sys/sysref2.h>
684a93c4 83#include <sys/mplock2.h>
e3161323
MD
84
85static void reaplwps(void *context, int dummy);
3e291793 86static void reaplwp(struct lwp *lp);
dc52e1cc 87static void killlwps(struct lwp *lp);
e43a034f 88
984263bc 89static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
d511d5d5 90static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
984263bc 91
b5c4d81f
MD
92static struct lwkt_token deadlwp_token = LWKT_TOKEN_INITIALIZER(deadlwp_token);
93
984263bc
MD
94/*
95 * callout list for things to do at exit time
96 */
97struct exitlist {
98 exitlist_fn function;
99 TAILQ_ENTRY(exitlist) next;
100};
101
102TAILQ_HEAD(exit_list_head, exitlist);
103static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
104
105/*
70d3d461
SS
106 * LWP reaper data
107 */
108struct task *deadlwp_task[MAXCPU];
109struct lwplist deadlwp_list[MAXCPU];
110
111/*
984263bc
MD
112 * exit --
113 * Death of process.
41c20dac
MD
114 *
115 * SYS_EXIT_ARGS(int rval)
984263bc 116 */
753fd850
MD
117int
118sys_exit(struct exit_args *uap)
984263bc 119{
41c20dac 120 exit1(W_EXITCODE(uap->rval, 0));
984263bc
MD
121 /* NOTREACHED */
122}
123
d5f340eb
SS
124/*
125 * Extended exit --
126 * Death of a lwp or process with optional bells and whistles.
3919ced0
MD
127 *
128 * MPALMOSTSAFE
d5f340eb
SS
129 */
130int
131sys_extexit(struct extexit_args *uap)
132{
b5c4d81f 133 struct proc *p = curproc;
d5f340eb
SS
134 int action, who;
135 int error;
136
137 action = EXTEXIT_ACTION(uap->how);
138 who = EXTEXIT_WHO(uap->how);
139
140 /* Check parameters before we might perform some action */
141 switch (who) {
142 case EXTEXIT_PROC:
143 case EXTEXIT_LWP:
144 break;
d5f340eb
SS
145 default:
146 return (EINVAL);
147 }
148
149 switch (action) {
150 case EXTEXIT_SIMPLE:
151 break;
d5f340eb
SS
152 case EXTEXIT_SETINT:
153 error = copyout(&uap->status, uap->addr, sizeof(uap->status));
154 if (error)
155 return (error);
156 break;
d5f340eb
SS
157 default:
158 return (EINVAL);
159 }
160
b5c4d81f 161 lwkt_gettoken(&p->p_token);
3919ced0 162
d5f340eb
SS
163 switch (who) {
164 case EXTEXIT_LWP:
165 /*
166 * Be sure only to perform a simple lwp exit if there is at
167 * least one more lwp in the proc, which will call exit1()
168 * later, otherwise the proc will be an UNDEAD and not even a
169 * SZOMB!
170 */
b5c4d81f
MD
171 if (p->p_nthreads > 1) {
172 lwp_exit(0); /* called w/ p_token held */
e3161323
MD
173 /* NOT REACHED */
174 }
d5f340eb
SS
175 /* else last lwp in proc: do the real thing */
176 /* FALLTHROUGH */
d5f340eb
SS
177 default: /* to help gcc */
178 case EXTEXIT_PROC:
b5c4d81f 179 lwkt_reltoken(&p->p_token);
d5f340eb
SS
180 exit1(W_EXITCODE(uap->status, 0));
181 /* NOTREACHED */
182 }
183
184 /* NOTREACHED */
b5c4d81f 185 lwkt_reltoken(&p->p_token); /* safety */
d5f340eb
SS
186}
187
e3161323 188/*
dc52e1cc
MD
189 * Kill all lwps associated with the current process except the
190 * current lwp. Return an error if we race another thread trying to
191 * do the same thing and lose the race.
192 *
193 * If forexec is non-zero the current thread and process flags are
194 * cleaned up so they can be reused.
616516c8
MD
195 *
196 * Caller must hold curproc->p_token
dc52e1cc
MD
197 */
198int
199killalllwps(int forexec)
200{
201 struct lwp *lp = curthread->td_lwp;
202 struct proc *p = lp->lwp_proc;
203
204 /*
205 * Interlock against P_WEXIT. Only one of the process's thread
206 * is allowed to do the master exit.
207 */
4643740a 208 if (p->p_flags & P_WEXIT)
dc52e1cc 209 return (EALREADY);
4643740a 210 p->p_flags |= P_WEXIT;
dc52e1cc
MD
211
212 /*
4643740a 213 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs
dc52e1cc 214 */
4643740a 215 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
dc52e1cc
MD
216 if (p->p_nthreads > 1)
217 killlwps(lp);
218
219 /*
220 * If doing this for an exec, clean up the remaining thread
221 * (us) for continuing operation after all the other threads
222 * have been killed.
223 */
224 if (forexec) {
4643740a
MD
225 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
226 p->p_flags &= ~P_WEXIT;
dc52e1cc
MD
227 }
228 return(0);
229}
230
231/*
e3161323
MD
232 * Kill all LWPs except the current one. Do not try to signal
233 * LWPs which have exited on their own or have already been
234 * signaled.
235 */
dc52e1cc 236static void
70d3d461
SS
237killlwps(struct lwp *lp)
238{
239 struct proc *p = lp->lwp_proc;
240 struct lwp *tlp;
241
e3161323 242 /*
89d9d449 243 * Kill the remaining LWPs. We must send the signal before setting
4643740a 244 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce
89d9d449
MD
245 * races. tlp must be held across the call as it might block and
246 * allow the target lwp to rip itself out from under our loop.
e3161323 247 */
70d3d461 248 FOREACH_LWP_IN_PROC(tlp, p) {
89d9d449 249 LWPHOLD(tlp);
e2b148c6 250 lwkt_gettoken(&tlp->lwp_token);
4643740a 251 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) {
73e40cb2 252 lwpsignal(p, tlp, SIGKILL);
4643740a 253 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT);
e3161323 254 }
e2b148c6 255 lwkt_reltoken(&tlp->lwp_token);
89d9d449 256 LWPRELE(tlp);
70d3d461
SS
257 }
258
e3161323
MD
259 /*
260 * Wait for everything to clear out.
261 */
70d3d461 262 while (p->p_nthreads > 1) {
ef1c4d7f 263 tsleep(&p->p_nthreads, 0, "killlwps", 0);
70d3d461
SS
264 }
265}
266
984263bc
MD
267/*
268 * Exit: deallocate address space and other resources, change proc state
269 * to zombie, and unlink proc from allproc and parent's lists. Save exit
270 * status and rusage for wait(). Check for child processes and orphan them.
271 */
272void
41c20dac 273exit1(int rv)
984263bc 274{
bb3cd951
SS
275 struct thread *td = curthread;
276 struct proc *p = td->td_proc;
277 struct lwp *lp = td->td_lwp;
41c20dac
MD
278 struct proc *q, *nq;
279 struct vmspace *vm;
984263bc
MD
280 struct vnode *vtmp;
281 struct exitlist *ep;
dc52e1cc 282 int error;
984263bc 283
b5c4d81f
MD
284 lwkt_gettoken(&p->p_token);
285
984263bc 286 if (p->p_pid == 1) {
6ea70f76 287 kprintf("init died (signal %d, exit %d)\n",
984263bc
MD
288 WTERMSIG(rv), WEXITSTATUS(rv));
289 panic("Going nowhere without my init!");
290 }
9d7a637e
AE
291 varsymset_clean(&p->p_varsymset);
292 lockuninit(&p->p_varsymset.vx_lock);
2af9d75d 293
70d3d461 294 /*
dc52e1cc
MD
295 * Kill all lwps associated with the current process, return an
296 * error if we race another thread trying to do the same thing
297 * and lose the race.
70d3d461 298 */
dc52e1cc
MD
299 error = killalllwps(0);
300 if (error) {
e3161323
MD
301 lwp_exit(0);
302 /* NOT REACHED */
222b18cd 303 }
81eea9f4 304
81eea9f4 305 caps_exit(lp->lwp_thread);
984263bc
MD
306
307 /* are we a task leader? */
e3161323 308 if (p == p->p_leader) {
984263bc
MD
309 struct kill_args killArgs;
310 killArgs.signum = SIGKILL;
311 q = p->p_peers;
312 while(q) {
313 killArgs.pid = q->p_pid;
314 /*
315 * The interface for kill is better
316 * than the internal signal
317 */
753fd850 318 sys_kill(&killArgs);
984263bc
MD
319 nq = q;
320 q = q->p_peers;
321 }
322 while (p->p_peers)
e3161323 323 tsleep((caddr_t)p, 0, "exit1", 0);
8ba5f7ef 324 }
984263bc
MD
325
326#ifdef PGINPROF
327 vmsizmon();
328#endif
329 STOPEVENT(p, S_EXIT, rv);
b2a6ad87 330 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */
984263bc
MD
331
332 /*
333 * Check if any loadable modules need anything done at process exit.
334 * e.g. SYSV IPC stuff
335 * XXX what if one of these generates an error?
336 */
8ba5f7ef
AH
337 p->p_xstat = rv;
338 EVENTHANDLER_INVOKE(process_exit, p);
339
340 /*
341 * XXX: imho, the eventhandler stuff is much cleaner than this.
342 * Maybe we should move everything to use eventhandler.
343 */
984263bc 344 TAILQ_FOREACH(ep, &exit_list, next)
bb3cd951 345 (*ep->function)(td);
984263bc 346
4643740a 347 if (p->p_flags & P_PROFIL)
984263bc 348 stopprofclock(p);
82354ad8 349
984263bc 350 SIGEMPTYSET(p->p_siglist);
aa6c3de6 351 SIGEMPTYSET(lp->lwp_siglist);
984263bc 352 if (timevalisset(&p->p_realtimer.it_value))
a471eac5 353 callout_stop_sync(&p->p_ithandle);
984263bc
MD
354
355 /*
356 * Reset any sigio structures pointing to us as a result of
357 * F_SETOWN with our pid.
358 */
359 funsetownlst(&p->p_sigiolst);
360
361 /*
362 * Close open files and release open-file table.
363 * This may block!
364 */
0a4a9c77 365 fdfree(p, NULL);
984263bc
MD
366
367 if(p->p_leader->p_peers) {
368 q = p->p_leader;
369 while(q->p_peers != p)
370 q = q->p_peers;
371 q->p_peers = p->p_peers;
372 wakeup((caddr_t)p->p_leader);
373 }
374
375 /*
376 * XXX Shutdown SYSV semaphores
377 */
378 semexit(p);
379
06ecca5a
MD
380 KKASSERT(p->p_numposixlocks == 0);
381
984263bc
MD
382 /* The next two chunks should probably be moved to vmspace_exit. */
383 vm = p->p_vmspace;
a722be49
MD
384
385 /*
386 * Release upcalls associated with this process
387 */
388 if (vm->vm_upcalls)
08f2f1bb 389 upc_release(vm, lp);
a722be49 390
39005e16
MD
391 /*
392 * Clean up data related to virtual kernel operation. Clean up
393 * any vkernel context related to the current lwp now so we can
394 * destroy p_vkernel.
395 */
396 if (p->p_vkernel) {
397 vkernel_lwp_exit(lp);
4a22e893 398 vkernel_exit(p);
39005e16 399 }
0daa37a5 400
984263bc
MD
401 /*
402 * Release user portion of address space.
403 * This releases references to vnodes,
404 * which could cause I/O if the file has been unlinked.
405 * Need to do this early enough that we can still sleep.
406 * Can't free the entire vmspace as the kernel stack
407 * may be mapped within that space also.
408 *
409 * Processes sharing the same vmspace may exit in one order, and
410 * get cleaned up by vmspace_exit() in a different order. The
411 * last exiting process to reach this point releases as much of
412 * the environment as it can, and the last process cleaned up
413 * by vmspace_exit() (which decrements exitingcnt) cleans up the
414 * remainder.
415 */
46754a20 416 vmspace_exitbump(vm);
e3161323 417 sysref_put(&vm->vm_sysref);
984263bc
MD
418
419 if (SESS_LEADER(p)) {
1fd87d54 420 struct session *sp = p->p_session;
984263bc
MD
421
422 if (sp->s_ttyvp) {
423 /*
1fbb5fc0
MD
424 * We are the controlling process. Signal the
425 * foreground process group, drain the controlling
426 * terminal, and revoke access to the controlling
427 * terminal.
428 *
429 * NOTE: while waiting for the process group to exit
430 * it is possible that one of the processes in the
f1aeb0c0
MD
431 * group will revoke the tty, so the ttyclosesession()
432 * function will re-check sp->s_ttyvp.
984263bc
MD
433 */
434 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
435 if (sp->s_ttyp->t_pgrp)
436 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
f1aeb0c0
MD
437 ttywait(sp->s_ttyp);
438 ttyclosesession(sp, 1); /* also revoke */
984263bc 439 }
5fd012e0
MD
440 /*
441 * Release the tty. If someone has it open via
442 * /dev/tty then close it (since they no longer can
443 * once we've NULL'd it out).
444 */
f1aeb0c0
MD
445 ttyclosesession(sp, 0);
446
984263bc
MD
447 /*
448 * s_ttyp is not zero'd; we use this to indicate
449 * that the session once had a controlling terminal.
450 * (for logging and informational purposes)
451 */
452 }
453 sp->s_leader = NULL;
454 }
455 fixjobc(p, p->p_pgrp, 0);
456 (void)acct_process(p);
457#ifdef KTRACE
458 /*
459 * release trace file
460 */
29f58392
MD
461 if (p->p_tracenode)
462 ktrdestroy(&p->p_tracenode);
463 p->p_traceflag = 0;
984263bc
MD
464#endif
465 /*
466 * Release reference to text vnode
467 */
468 if ((vtmp = p->p_textvp) != NULL) {
469 p->p_textvp = NULL;
470 vrele(vtmp);
471 }
472
8ba5f7ef
AH
473 /* Release namecache handle to text file */
474 if (p->p_textnch.ncp)
475 cache_drop(&p->p_textnch);
476
984263bc 477 /*
82354ad8
MD
478 * We have to handle PPWAIT here or proc_move_allproc_zombie()
479 * will block on the PHOLD() the parent is doing.
480 */
481 if (p->p_flags & P_PPWAIT) {
482 p->p_flags &= ~P_PPWAIT;
483 wakeup(p->p_pptr);
484 }
485
486 /*
5bf0d9b5
MD
487 * Move the process to the zombie list. This will block
488 * until the process p_lock count reaches 0. The process will
489 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
490 * which is called from cpu_proc_exit().
ae8050a4 491 */
5bf0d9b5 492 proc_move_allproc_zombie(p);
984263bc 493
7e90d791
MD
494 /*
495 * Reparent all of this process's children to the init process.
496 * We must hold initproc->p_token in order to mess with
497 * initproc->p_children. We already hold p->p_token (to remove
498 * the children from our list).
499 */
984263bc 500 q = LIST_FIRST(&p->p_children);
7e90d791
MD
501 if (q) {
502 lwkt_gettoken(&initproc->p_token);
503 while (q) {
504 nq = LIST_NEXT(q, p_sibling);
505 LIST_REMOVE(q, p_sibling);
506 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
507 q->p_pptr = initproc;
508 q->p_sigparent = SIGCHLD;
509 /*
510 * Traced processes are killed
511 * since their existence means someone is screwing up.
512 */
4643740a
MD
513 if (q->p_flags & P_TRACED) {
514 q->p_flags &= ~P_TRACED;
7e90d791
MD
515 ksignal(q, SIGKILL);
516 }
517 q = nq;
984263bc 518 }
7e90d791
MD
519 lwkt_reltoken(&initproc->p_token);
520 wakeup(initproc);
984263bc
MD
521 }
522
523 /*
524 * Save exit status and final rusage info, adding in child rusage
525 * info and self times.
526 */
fde7ac71
SS
527 calcru_proc(p, &p->p_ru);
528 ruadd(&p->p_ru, &p->p_cru);
984263bc
MD
529
530 /*
984263bc
MD
531 * notify interested parties of our demise.
532 */
533 KNOTE(&p->p_klist, NOTE_EXIT);
534
535 /*
536 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
19af1870
MR
537 * flag set, or if the handler is set to SIG_IGN, notify process 1
538 * instead (and hope it will handle this situation).
984263bc 539 */
a0ae03a7 540 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
984263bc 541 struct proc *pp = p->p_pptr;
58c2553a
MD
542
543 PHOLD(pp);
984263bc 544 proc_reparent(p, initproc);
58c2553a 545
984263bc
MD
546 /*
547 * If this was the last child of our parent, notify
548 * parent, so in case he was wait(2)ing, he will
58c2553a 549 * continue. This function interlocks with pptr->p_token.
984263bc
MD
550 */
551 if (LIST_EMPTY(&pp->p_children))
552 wakeup((caddr_t)pp);
58c2553a 553 PRELE(pp);
984263bc
MD
554 }
555
f2f3db5c
MD
556 /* lwkt_gettoken(&proc_token); */
557 q = p->p_pptr;
58c2553a 558 PHOLD(q);
f2f3db5c 559 if (p->p_sigparent && q != initproc) {
f2f3db5c 560 ksignal(q, p->p_sigparent);
984263bc 561 } else {
f2f3db5c 562 ksignal(q, SIGCHLD);
984263bc 563 }
82354ad8
MD
564
565 p->p_flags &= ~P_TRACED;
58c2553a 566 wakeup(p->p_pptr);
82354ad8 567
58c2553a 568 PRELE(q);
f2f3db5c
MD
569 /* lwkt_reltoken(&proc_token); */
570 /* NOTE: p->p_pptr can get ripped out */
984263bc 571 /*
8ad65e08
MD
572 * cpu_exit is responsible for clearing curproc, since
573 * it is heavily integrated with the thread/switching sequence.
574 *
984263bc
MD
575 * Other substructures are freed from wait().
576 */
8f1f6170 577 plimit_free(p);
984263bc
MD
578
579 /*
0a3f9b47
MD
580 * Release the current user process designation on the process so
581 * the userland scheduler can work in someone else.
26a0694b 582 */
553ea3c8 583 p->p_usched->release_curproc(lp);
26a0694b
MD
584
585 /*
e3161323
MD
586 * Finally, call machine-dependent code to release as many of the
587 * lwp's resources as we can and halt execution of this thread.
984263bc 588 */
e3161323 589 lwp_exit(1);
c6880072
SS
590}
591
0d355d3b
MD
592/*
593 * Eventually called by every exiting LWP
b5c4d81f
MD
594 *
595 * p->p_token must be held. mplock may be held and will be released.
0d355d3b 596 */
c6880072 597void
e3161323 598lwp_exit(int masterexit)
c6880072 599{
d86a23e0
MD
600 struct thread *td = curthread;
601 struct lwp *lp = td->td_lwp;
c6880072 602 struct proc *p = lp->lwp_proc;
8c2bce60 603 int dowake = 0;
c6880072 604
70d3d461 605 /*
4643740a 606 * lwp_exit() may be called without setting LWP_MP_WEXIT, so
e3161323
MD
607 * make sure it is set here.
608 */
b5c4d81f 609 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
4643740a 610 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
e3161323
MD
611
612 /*
39005e16
MD
613 * Clean up any virtualization
614 */
615 if (lp->lwp_vkernel)
616 vkernel_lwp_exit(lp);
617
618 /*
a591f597
MD
619 * Clean up select/poll support
620 */
621 kqueue_terminate(&lp->lwp_kqueue);
622
623 /*
0d355d3b
MD
624 * Clean up any syscall-cached ucred
625 */
d86a23e0
MD
626 if (td->td_ucred) {
627 crfree(td->td_ucred);
628 td->td_ucred = NULL;
0d355d3b
MD
629 }
630
631 /*
70d3d461
SS
632 * Nobody actually wakes us when the lock
633 * count reaches zero, so just wait one tick.
634 */
635 while (lp->lwp_lock > 0)
636 tsleep(lp, 0, "lwpexit", 1);
637
638 /* Hand down resource usage to our proc */
639 ruadd(&p->p_ru, &lp->lwp_ru);
640
e3161323
MD
641 /*
642 * If we don't hold the process until the LWP is reaped wait*()
643 * may try to dispose of its vmspace before all the LWPs have
644 * actually terminated.
645 */
646 PHOLD(p);
647
648 /*
2883d2d8
MD
649 * Do any remaining work that might block on us. We should be
650 * coded such that further blocking is ok after decrementing
651 * p_nthreads but don't take the chance.
652 */
653 dsched_exit_thread(td);
654 biosched_done(curthread);
655
656 /*
e3161323
MD
657 * We have to use the reaper for all the LWPs except the one doing
658 * the master exit. The LWP doing the master exit can just be
659 * left on p_lwps and the process reaper will deal with it
660 * synchronously, which is much faster.
ce19c8fe
MD
661 *
662 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0.
2af9d75d
MD
663 *
664 * The process is left held until the reaper calls lwp_dispose() on
665 * the lp (after calling lwp_wait()).
e3161323
MD
666 */
667 if (masterexit == 0) {
3e291793 668 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
e3161323 669 --p->p_nthreads;
ce19c8fe 670 if (p->p_nthreads <= 1)
8c2bce60 671 dowake = 1;
b5c4d81f 672 lwkt_gettoken(&deadlwp_token);
3e291793 673 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry);
2883d2d8
MD
674 taskqueue_enqueue(taskqueue_thread[mycpuid],
675 deadlwp_task[mycpuid]);
b5c4d81f 676 lwkt_reltoken(&deadlwp_token);
e3161323
MD
677 } else {
678 --p->p_nthreads;
ce19c8fe 679 if (p->p_nthreads <= 1)
8c2bce60 680 dowake = 1;
e3161323 681 }
b5c4d81f
MD
682
683 /*
8c2bce60
MD
684 * Release p_token. Issue the wakeup() on p_nthreads if necessary,
685 * as late as possible to give us a chance to actually deschedule and
686 * switch away before another cpu core hits reaplwp().
b5c4d81f
MD
687 */
688 lwkt_reltoken(&p->p_token);
8c2bce60
MD
689 if (dowake)
690 wakeup(&p->p_nthreads);
c6880072
SS
691 cpu_lwp_exit();
692}
693
694/*
2af9d75d
MD
695 * Wait until a lwp is completely dead. The final interlock in this drama
696 * is when TDF_EXITING is set in cpu_thread_exit() just before the final
697 * switchout.
c6880072 698 *
2af9d75d
MD
699 * At the point TDF_EXITING is set a complete exit is accomplished when
700 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear.
c6880072 701 *
2af9d75d
MD
702 * Returns non-zero on success, and zero if the caller needs to retry
703 * the lwp_wait().
c6880072
SS
704 */
705static int
706lwp_wait(struct lwp *lp)
707{
708 struct thread *td = lp->lwp_thread;;
709
710 KKASSERT(lwkt_preempted_proc() != lp);
711
2af9d75d
MD
712 /*
713 * Wait until the lp has entered its low level exit and wait
714 * until other cores with refs on the lp (e.g. for ps or signaling)
715 * release them.
716 */
717 if (lp->lwp_lock > 0) {
c6880072 718 tsleep(lp, 0, "lwpwait1", 1);
2af9d75d
MD
719 return(0);
720 }
c6880072 721
2af9d75d
MD
722 /*
723 * Wait until the thread is no longer references and no longer
724 * runnable or preempted (i.e. finishes its low level exit).
725 */
726 if (td->td_refs) {
727 tsleep(td, 0, "lwpwait2", 1);
728 return(0);
729 }
c6880072
SS
730
731 /*
732 * The lwp's thread may still be in the middle
733 * of switching away, we can't rip its stack out from
734 * under it until TDF_EXITING is set and both
735 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
736 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
737 * will be cleared temporarily if a thread gets
738 * preempted.
739 *
740 * YYY no wakeup occurs, so we simply return failure
741 * and let the caller deal with sleeping and calling
742 * us again.
743 */
2af9d75d
MD
744 if ((td->td_flags & (TDF_RUNNING |
745 TDF_PREEMPT_LOCK |
746 TDF_EXITING)) != TDF_EXITING) {
747 tsleep(lp, 0, "lwpwait2", 1);
c6880072 748 return (0);
cf709dd2 749 }
2af9d75d
MD
750 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0,
751 ("lwp_wait: td %p (%s) still on run or sleep queue",
752 td, td->td_comm));
c6880072
SS
753 return (1);
754}
755
756/*
757 * Release the resources associated with a lwp.
758 * The lwp must be completely dead.
759 */
760void
761lwp_dispose(struct lwp *lp)
762{
763 struct thread *td = lp->lwp_thread;;
764
765 KKASSERT(lwkt_preempted_proc() != lp);
766 KKASSERT(td->td_refs == 0);
2af9d75d
MD
767 KKASSERT((td->td_flags & (TDF_RUNNING |
768 TDF_PREEMPT_LOCK |
769 TDF_EXITING)) == TDF_EXITING);
c6880072 770
e3161323
MD
771 PRELE(lp->lwp_proc);
772 lp->lwp_proc = NULL;
c6880072
SS
773 if (td != NULL) {
774 td->td_proc = NULL;
775 td->td_lwp = NULL;
776 lp->lwp_thread = NULL;
777 lwkt_free_thread(td);
778 }
f6c36234 779 kfree(lp, M_LWP);
984263bc
MD
780}
781
3919ced0
MD
782/*
783 * MPSAFE
784 */
984263bc 785int
753fd850 786sys_wait4(struct wait_args *uap)
984263bc 787{
9697c509
DRJ
788 struct rusage rusage;
789 int error, status;
984263bc 790
3919ced0
MD
791 error = kern_wait(uap->pid, (uap->status ? &status : NULL),
792 uap->options, (uap->rusage ? &rusage : NULL),
793 &uap->sysmsg_result);
984263bc 794
9697c509
DRJ
795 if (error == 0 && uap->status)
796 error = copyout(&status, uap->status, sizeof(*uap->status));
797 if (error == 0 && uap->rusage)
798 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
799 return (error);
984263bc
MD
800}
801
41c20dac
MD
802/*
803 * wait1()
804 *
805 * wait_args(int pid, int *status, int options, struct rusage *rusage)
3919ced0
MD
806 *
807 * MPALMOSTSAFE
41c20dac 808 */
9697c509
DRJ
809int
810kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
984263bc 811{
9697c509 812 struct thread *td = curthread;
3e291793 813 struct lwp *lp;
9697c509 814 struct proc *q = td->td_proc;
41c20dac 815 struct proc *p, *t;
e955543c 816 struct pargs *pa;
6fa9e71a 817 struct sigacts *ps;
9697c509 818 int nfound, error;
984263bc 819
9697c509
DRJ
820 if (pid == 0)
821 pid = -q->p_pgid;
cc162105 822 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
984263bc 823 return (EINVAL);
b5c4d81f
MD
824
825 lwkt_gettoken(&q->p_token);
984263bc 826loop:
6d5b152e 827 /*
b5c4d81f
MD
828 * All sorts of things can change due to blocking so we have to loop
829 * all the way back up here.
6d5b152e
MD
830 *
831 * The problem is that if a process group is stopped and the parent
832 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
833 * of the child and then stop itself when it tries to return from the
834 * system call. When the process group is resumed the parent will
835 * then get the STOP status even though the child has now resumed
836 * (a followup wait*() will get the CONT status).
837 *
838 * Previously the CONT would overwrite the STOP because the tstop
839 * was handled within tsleep(), and the parent would only see
b5c4d81f 840 * the CONT when both are stopped and continued together. This little
6d5b152e
MD
841 * two-line hack restores this effect.
842 */
164b8401 843 while (q->p_stat == SSTOP)
9a379a4a 844 tstop();
6d5b152e 845
984263bc 846 nfound = 0;
b5c4d81f 847
8c986a82
MD
848 /*
849 * Loop on children.
850 *
851 * NOTE: We don't want to break q's p_token in the loop for the
852 * case where no children are found or we risk breaking the
853 * interlock between child and parent.
854 */
984263bc 855 LIST_FOREACH(p, &q->p_children, p_sibling) {
9697c509 856 if (pid != WAIT_ANY &&
b5c4d81f 857 p->p_pid != pid && p->p_pgid != -pid) {
984263bc 858 continue;
b5c4d81f 859 }
984263bc 860
5686ec5a
MD
861 /*
862 * This special case handles a kthread spawned by linux_clone
352f5709
MD
863 * (see linux_misc.c). The linux_wait4 and linux_waitpid
864 * functions need to be able to distinguish between waiting
865 * on a process and waiting on a thread. It is a thread if
866 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
867 * signifies we want to wait for threads and not processes.
984263bc 868 */
352f5709
MD
869 if ((p->p_sigparent != SIGCHLD) ^
870 ((options & WLINUXCLONE) != 0)) {
984263bc 871 continue;
352f5709 872 }
984263bc
MD
873
874 nfound++;
416d05d7 875 if (p->p_stat == SZOMB) {
ae8050a4 876 /*
8f211c4b
MD
877 * We may go into SZOMB with threads still present.
878 * We must wait for them to exit before we can reap
879 * the master thread, otherwise we may race reaping
880 * non-master threads.
881 */
b5c4d81f 882 lwkt_gettoken(&p->p_token);
8f211c4b
MD
883 while (p->p_nthreads > 0) {
884 tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
885 }
886
887 /*
e3161323
MD
888 * Reap any LWPs left in p->p_lwps. This is usually
889 * just the last LWP. This must be done before
890 * we loop on p_lock since the lwps hold a ref on
891 * it as a vmspace interlock.
892 *
893 * Once that is accomplished p_nthreads had better
894 * be zero.
895 */
3e291793
MD
896 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
897 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
898 reaplwp(lp);
899 }
e3161323
MD
900 KKASSERT(p->p_nthreads == 0);
901
902 /*
903 * Don't do anything really bad until all references
904 * to the process go away. This may include other
905 * LWPs which are still in the process of being
906 * reaped. We can't just pull the rug out from under
907 * them because they may still be using the VM space.
908 *
909 * Certain kernel facilities such as /proc will also
910 * put a hold on the process for short periods of
911 * time.
c008d3ad 912 */
82354ad8 913 PSTALL(p, "reap3", 0);
c1102e9f 914
9697c509
DRJ
915 /* Take care of our return values. */
916 *res = p->p_pid;
52cac9fb
MD
917 p->p_usched->heuristic_exiting(td->td_lwp, p);
918
9697c509
DRJ
919 if (status)
920 *status = p->p_xstat;
921 if (rusage)
fde7ac71 922 *rusage = p->p_ru;
984263bc
MD
923 /*
924 * If we got the child via a ptrace 'attach',
925 * we need to give it back to the old parent.
926 */
58c2553a 927 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
984263bc
MD
928 p->p_oppid = 0;
929 proc_reparent(p, t);
84204577 930 ksignal(t, SIGCHLD);
984263bc 931 wakeup((caddr_t)t);
3919ced0 932 error = 0;
58c2553a 933 PRELE(t);
eb2adbf5 934 lwkt_reltoken(&p->p_token);
3919ced0 935 goto done;
984263bc 936 }
3c8687d6
SS
937
938 /*
939 * Unlink the proc from its process group so that
940 * the following operations won't lead to an
941 * inconsistent state for processes running down
942 * the zombie list.
943 */
3c8687d6 944 proc_remove_zombie(p);
eb2adbf5 945 lwkt_reltoken(&p->p_token);
3c8687d6
SS
946 leavepgrp(p);
947
984263bc 948 p->p_xstat = 0;
fde7ac71 949 ruadd(&q->p_cru, &p->p_ru);
984263bc
MD
950
951 /*
952 * Decrement the count of procs running with this uid.
953 */
9697c509 954 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
984263bc
MD
955
956 /*
957 * Free up credentials.
958 */
41c20dac
MD
959 crfree(p->p_ucred);
960 p->p_ucred = NULL;
984263bc
MD
961
962 /*
963 * Remove unused arguments
964 */
e955543c
MD
965 pa = p->p_args;
966 p->p_args = NULL;
e955543c
MD
967 if (pa && refcount_release(&pa->ar_ref)) {
968 kfree(pa, M_PARGS);
969 pa = NULL;
970 }
6fa9e71a
MD
971
972 ps = p->p_sigacts;
973 p->p_sigacts = NULL;
974 if (ps && refcount_release(&ps->ps_refcnt)) {
975 kfree(ps, M_SUBPROC);
976 ps = NULL;
984263bc
MD
977 }
978
49aa3df0 979 /*
e22b17d7
MD
980 * Our exitingcount was incremented when the process
981 * became a zombie, now that the process has been
982 * removed from (almost) all lists we should be able
983 * to safely destroy its vmspace. Wait for any current
984 * holders to go away (so the vmspace remains stable),
985 * then scrap it.
49aa3df0 986 */
82354ad8 987 PSTALL(p, "reap4", 0);
e22b17d7 988 vmspace_exitfree(p);
82354ad8 989 PSTALL(p, "reap5", 0);
e22b17d7 990
fb2a331e 991 kfree(p, M_PROC);
8c2bce60 992 atomic_add_int(&nprocs, -1);
3919ced0
MD
993 error = 0;
994 goto done;
984263bc 995 }
4643740a
MD
996 if (p->p_stat == SSTOP && (p->p_flags & P_WAITED) == 0 &&
997 ((p->p_flags & P_TRACED) || (options & WUNTRACED))) {
8c986a82 998 lwkt_gettoken(&p->p_token);
4643740a 999 p->p_flags |= P_WAITED;
9697c509
DRJ
1000
1001 *res = p->p_pid;
52cac9fb 1002 p->p_usched->heuristic_exiting(td->td_lwp, p);
9697c509 1003 if (status)
a89c93ee 1004 *status = W_STOPCODE(p->p_xstat);
9697c509
DRJ
1005 /* Zero rusage so we get something consistent. */
1006 if (rusage)
8075c3b8 1007 bzero(rusage, sizeof(*rusage));
3919ced0 1008 error = 0;
8c986a82 1009 lwkt_reltoken(&p->p_token);
3919ced0 1010 goto done;
984263bc 1011 }
4643740a 1012 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) {
8c986a82 1013 lwkt_gettoken(&p->p_token);
cc162105 1014 *res = p->p_pid;
52cac9fb 1015 p->p_usched->heuristic_exiting(td->td_lwp, p);
4643740a 1016 p->p_flags &= ~P_CONTINUED;
cc162105
PA
1017
1018 if (status)
1019 *status = SIGCONT;
3919ced0 1020 error = 0;
8c986a82 1021 lwkt_reltoken(&p->p_token);
3919ced0 1022 goto done;
cc162105 1023 }
984263bc 1024 }
3919ced0
MD
1025 if (nfound == 0) {
1026 error = ECHILD;
1027 goto done;
1028 }
9697c509
DRJ
1029 if (options & WNOHANG) {
1030 *res = 0;
3919ced0
MD
1031 error = 0;
1032 goto done;
984263bc 1033 }
58c2553a
MD
1034
1035 /*
1036 * Wait for signal - interlocked using q->p_token.
1037 */
1038 error = tsleep(q, PCATCH, "wait", 0);
3919ced0
MD
1039 if (error) {
1040done:
b5c4d81f 1041 lwkt_reltoken(&q->p_token);
984263bc 1042 return (error);
3919ced0 1043 }
984263bc
MD
1044 goto loop;
1045}
1046
1047/*
b5c4d81f 1048 * Make process 'parent' the new parent of process 'child'.
7e90d791
MD
1049 *
1050 * p_children/p_sibling requires the parent's token, and
1051 * changing pptr requires the child's token, so we have to
1052 * get three tokens to do this operation.
984263bc
MD
1053 */
1054void
77153250 1055proc_reparent(struct proc *child, struct proc *parent)
984263bc 1056{
7e90d791
MD
1057 struct proc *opp = child->p_pptr;
1058
1059 if (opp == parent)
984263bc 1060 return;
7e90d791 1061 PHOLD(opp);
b5c4d81f 1062 PHOLD(parent);
7e90d791 1063 lwkt_gettoken(&opp->p_token);
b5c4d81f
MD
1064 lwkt_gettoken(&child->p_token);
1065 lwkt_gettoken(&parent->p_token);
7e90d791 1066 KKASSERT(child->p_pptr == opp);
984263bc
MD
1067 LIST_REMOVE(child, p_sibling);
1068 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1069 child->p_pptr = parent;
b5c4d81f
MD
1070 lwkt_reltoken(&parent->p_token);
1071 lwkt_reltoken(&child->p_token);
7e90d791 1072 lwkt_reltoken(&opp->p_token);
b5c4d81f 1073 PRELE(parent);
7e90d791 1074 PRELE(opp);
984263bc
MD
1075}
1076
1077/*
1078 * The next two functions are to handle adding/deleting items on the
1079 * exit callout list
1080 *
1081 * at_exit():
1082 * Take the arguments given and put them onto the exit callout list,
1083 * However first make sure that it's not already there.
1084 * returns 0 on success.
1085 */
1086
1087int
77153250 1088at_exit(exitlist_fn function)
984263bc
MD
1089{
1090 struct exitlist *ep;
1091
1092#ifdef INVARIANTS
1093 /* Be noisy if the programmer has lost track of things */
1094 if (rm_at_exit(function))
6ea70f76 1095 kprintf("WARNING: exit callout entry (%p) already present\n",
984263bc
MD
1096 function);
1097#endif
efda3bd0 1098 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
984263bc
MD
1099 if (ep == NULL)
1100 return (ENOMEM);
1101 ep->function = function;
1102 TAILQ_INSERT_TAIL(&exit_list, ep, next);
1103 return (0);
1104}
1105
1106/*
1107 * Scan the exit callout list for the given item and remove it.
1108 * Returns the number of items removed (0 or 1)
1109 */
1110int
77153250 1111rm_at_exit(exitlist_fn function)
984263bc
MD
1112{
1113 struct exitlist *ep;
1114
1115 TAILQ_FOREACH(ep, &exit_list, next) {
1116 if (ep->function == function) {
1117 TAILQ_REMOVE(&exit_list, ep, next);
efda3bd0 1118 kfree(ep, M_ATEXIT);
984263bc
MD
1119 return(1);
1120 }
1121 }
1122 return (0);
1123}
1124
70d3d461
SS
1125/*
1126 * LWP reaper related code.
1127 */
70d3d461
SS
1128static void
1129reaplwps(void *context, int dummy)
1130{
1131 struct lwplist *lwplist = context;
1132 struct lwp *lp;
1133
b5c4d81f 1134 lwkt_gettoken(&deadlwp_token);
70d3d461 1135 while ((lp = LIST_FIRST(lwplist))) {
3e291793
MD
1136 LIST_REMOVE(lp, u.lwp_reap_entry);
1137 reaplwp(lp);
70d3d461 1138 }
b5c4d81f 1139 lwkt_reltoken(&deadlwp_token);
70d3d461
SS
1140}
1141
1142static void
3e291793
MD
1143reaplwp(struct lwp *lp)
1144{
2af9d75d
MD
1145 while (lwp_wait(lp) == 0)
1146 ;
3e291793
MD
1147 lwp_dispose(lp);
1148}
1149
1150static void
70d3d461
SS
1151deadlwp_init(void)
1152{
1153 int cpu;
1154
1155 for (cpu = 0; cpu < ncpus; cpu++) {
1156 LIST_INIT(&deadlwp_list[cpu]);
2af9d75d
MD
1157 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]),
1158 M_DEVBUF, M_WAITOK);
70d3d461
SS
1159 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1160 }
1161}
1162
1163SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);