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 $
f6c36234 40 * $DragonFly: src/sys/kern/kern_exit.c,v 1.91 2008/05/18 20:02:02 nth Exp $
984263bc
MD
41 */
42
43#include "opt_compat.h"
44#include "opt_ktrace.h"
45
46#include <sys/param.h>
47#include <sys/systm.h>
48#include <sys/sysproto.h>
49#include <sys/kernel.h>
50#include <sys/malloc.h>
51#include <sys/proc.h>
29f58392 52#include <sys/ktrace.h>
984263bc
MD
53#include <sys/pioctl.h>
54#include <sys/tty.h>
55#include <sys/wait.h>
56#include <sys/vnode.h>
57#include <sys/resourcevar.h>
58#include <sys/signalvar.h>
70d3d461 59#include <sys/taskqueue.h>
984263bc
MD
60#include <sys/ptrace.h>
61#include <sys/acct.h> /* for acct_process() function prototype */
62#include <sys/filedesc.h>
63#include <sys/shm.h>
64#include <sys/sem.h>
984263bc 65#include <sys/jail.h>
9697c509 66#include <sys/kern_syscall.h>
a722be49 67#include <sys/upcall.h>
f6bf3af1 68#include <sys/caps.h>
d5f340eb 69#include <sys/unistd.h>
8ba5f7ef 70#include <sys/eventhandler.h>
2883d2d8 71#include <sys/dsched.h>
984263bc
MD
72
73#include <vm/vm.h>
74#include <vm/vm_param.h>
75#include <sys/lock.h>
76#include <vm/pmap.h>
77#include <vm/vm_map.h>
984263bc
MD
78#include <vm/vm_extern.h>
79#include <sys/user.h>
80
e955543c 81#include <sys/refcount.h>
e43a034f 82#include <sys/thread2.h>
e3161323 83#include <sys/sysref2.h>
684a93c4 84#include <sys/mplock2.h>
e3161323
MD
85
86static void reaplwps(void *context, int dummy);
3e291793 87static void reaplwp(struct lwp *lp);
dc52e1cc 88static void killlwps(struct lwp *lp);
e43a034f 89
984263bc 90static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback");
d511d5d5 91static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
984263bc 92
b5c4d81f
MD
93static struct lwkt_token deadlwp_token = LWKT_TOKEN_INITIALIZER(deadlwp_token);
94
984263bc
MD
95/*
96 * callout list for things to do at exit time
97 */
98struct exitlist {
99 exitlist_fn function;
100 TAILQ_ENTRY(exitlist) next;
101};
102
103TAILQ_HEAD(exit_list_head, exitlist);
104static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list);
105
106/*
70d3d461
SS
107 * LWP reaper data
108 */
109struct task *deadlwp_task[MAXCPU];
110struct lwplist deadlwp_list[MAXCPU];
111
112/*
984263bc
MD
113 * exit --
114 * Death of process.
41c20dac
MD
115 *
116 * SYS_EXIT_ARGS(int rval)
984263bc 117 */
753fd850
MD
118int
119sys_exit(struct exit_args *uap)
984263bc 120{
41c20dac 121 exit1(W_EXITCODE(uap->rval, 0));
984263bc
MD
122 /* NOTREACHED */
123}
124
d5f340eb
SS
125/*
126 * Extended exit --
127 * Death of a lwp or process with optional bells and whistles.
3919ced0
MD
128 *
129 * MPALMOSTSAFE
d5f340eb
SS
130 */
131int
132sys_extexit(struct extexit_args *uap)
133{
b5c4d81f 134 struct proc *p = curproc;
d5f340eb
SS
135 int action, who;
136 int error;
137
138 action = EXTEXIT_ACTION(uap->how);
139 who = EXTEXIT_WHO(uap->how);
140
141 /* Check parameters before we might perform some action */
142 switch (who) {
143 case EXTEXIT_PROC:
144 case EXTEXIT_LWP:
145 break;
d5f340eb
SS
146 default:
147 return (EINVAL);
148 }
149
150 switch (action) {
151 case EXTEXIT_SIMPLE:
152 break;
d5f340eb
SS
153 case EXTEXIT_SETINT:
154 error = copyout(&uap->status, uap->addr, sizeof(uap->status));
155 if (error)
156 return (error);
157 break;
d5f340eb
SS
158 default:
159 return (EINVAL);
160 }
161
b5c4d81f 162 lwkt_gettoken(&p->p_token);
3919ced0 163
d5f340eb
SS
164 switch (who) {
165 case EXTEXIT_LWP:
166 /*
167 * Be sure only to perform a simple lwp exit if there is at
168 * least one more lwp in the proc, which will call exit1()
169 * later, otherwise the proc will be an UNDEAD and not even a
170 * SZOMB!
171 */
b5c4d81f
MD
172 if (p->p_nthreads > 1) {
173 lwp_exit(0); /* called w/ p_token held */
e3161323
MD
174 /* NOT REACHED */
175 }
d5f340eb
SS
176 /* else last lwp in proc: do the real thing */
177 /* FALLTHROUGH */
d5f340eb
SS
178 default: /* to help gcc */
179 case EXTEXIT_PROC:
b5c4d81f 180 lwkt_reltoken(&p->p_token);
d5f340eb
SS
181 exit1(W_EXITCODE(uap->status, 0));
182 /* NOTREACHED */
183 }
184
185 /* NOTREACHED */
b5c4d81f 186 lwkt_reltoken(&p->p_token); /* safety */
d5f340eb
SS
187}
188
e3161323 189/*
dc52e1cc
MD
190 * Kill all lwps associated with the current process except the
191 * current lwp. Return an error if we race another thread trying to
192 * do the same thing and lose the race.
193 *
194 * If forexec is non-zero the current thread and process flags are
195 * cleaned up so they can be reused.
616516c8
MD
196 *
197 * Caller must hold curproc->p_token
dc52e1cc
MD
198 */
199int
200killalllwps(int forexec)
201{
202 struct lwp *lp = curthread->td_lwp;
203 struct proc *p = lp->lwp_proc;
204
205 /*
206 * Interlock against P_WEXIT. Only one of the process's thread
207 * is allowed to do the master exit.
208 */
4643740a 209 if (p->p_flags & P_WEXIT)
dc52e1cc 210 return (EALREADY);
4643740a 211 p->p_flags |= P_WEXIT;
dc52e1cc
MD
212
213 /*
4643740a 214 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs
dc52e1cc 215 */
4643740a 216 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
dc52e1cc
MD
217 if (p->p_nthreads > 1)
218 killlwps(lp);
219
220 /*
221 * If doing this for an exec, clean up the remaining thread
222 * (us) for continuing operation after all the other threads
223 * have been killed.
224 */
225 if (forexec) {
4643740a
MD
226 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
227 p->p_flags &= ~P_WEXIT;
dc52e1cc
MD
228 }
229 return(0);
230}
231
232/*
e3161323
MD
233 * Kill all LWPs except the current one. Do not try to signal
234 * LWPs which have exited on their own or have already been
235 * signaled.
236 */
dc52e1cc 237static void
70d3d461
SS
238killlwps(struct lwp *lp)
239{
240 struct proc *p = lp->lwp_proc;
241 struct lwp *tlp;
242
e3161323 243 /*
89d9d449 244 * Kill the remaining LWPs. We must send the signal before setting
4643740a 245 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce
89d9d449
MD
246 * races. tlp must be held across the call as it might block and
247 * allow the target lwp to rip itself out from under our loop.
e3161323 248 */
70d3d461 249 FOREACH_LWP_IN_PROC(tlp, p) {
89d9d449 250 LWPHOLD(tlp);
e2b148c6 251 lwkt_gettoken(&tlp->lwp_token);
4643740a 252 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) {
73e40cb2 253 lwpsignal(p, tlp, SIGKILL);
4643740a 254 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT);
e3161323 255 }
e2b148c6 256 lwkt_reltoken(&tlp->lwp_token);
89d9d449 257 LWPRELE(tlp);
70d3d461
SS
258 }
259
e3161323
MD
260 /*
261 * Wait for everything to clear out.
262 */
70d3d461 263 while (p->p_nthreads > 1) {
ef1c4d7f 264 tsleep(&p->p_nthreads, 0, "killlwps", 0);
70d3d461
SS
265 }
266}
267
984263bc
MD
268/*
269 * Exit: deallocate address space and other resources, change proc state
270 * to zombie, and unlink proc from allproc and parent's lists. Save exit
271 * status and rusage for wait(). Check for child processes and orphan them.
272 */
273void
41c20dac 274exit1(int rv)
984263bc 275{
bb3cd951
SS
276 struct thread *td = curthread;
277 struct proc *p = td->td_proc;
278 struct lwp *lp = td->td_lwp;
41c20dac
MD
279 struct proc *q, *nq;
280 struct vmspace *vm;
984263bc
MD
281 struct vnode *vtmp;
282 struct exitlist *ep;
dc52e1cc 283 int error;
984263bc 284
b5c4d81f
MD
285 lwkt_gettoken(&p->p_token);
286
984263bc 287 if (p->p_pid == 1) {
6ea70f76 288 kprintf("init died (signal %d, exit %d)\n",
984263bc
MD
289 WTERMSIG(rv), WEXITSTATUS(rv));
290 panic("Going nowhere without my init!");
291 }
9d7a637e
AE
292 varsymset_clean(&p->p_varsymset);
293 lockuninit(&p->p_varsymset.vx_lock);
2af9d75d 294
70d3d461 295 /*
dc52e1cc
MD
296 * Kill all lwps associated with the current process, return an
297 * error if we race another thread trying to do the same thing
298 * and lose the race.
70d3d461 299 */
dc52e1cc
MD
300 error = killalllwps(0);
301 if (error) {
e3161323
MD
302 lwp_exit(0);
303 /* NOT REACHED */
222b18cd 304 }
81eea9f4 305
81eea9f4 306 caps_exit(lp->lwp_thread);
984263bc
MD
307
308 /* are we a task leader? */
e3161323 309 if (p == p->p_leader) {
984263bc
MD
310 struct kill_args killArgs;
311 killArgs.signum = SIGKILL;
312 q = p->p_peers;
313 while(q) {
314 killArgs.pid = q->p_pid;
315 /*
316 * The interface for kill is better
317 * than the internal signal
318 */
753fd850 319 sys_kill(&killArgs);
984263bc
MD
320 nq = q;
321 q = q->p_peers;
322 }
323 while (p->p_peers)
e3161323 324 tsleep((caddr_t)p, 0, "exit1", 0);
8ba5f7ef 325 }
984263bc
MD
326
327#ifdef PGINPROF
328 vmsizmon();
329#endif
330 STOPEVENT(p, S_EXIT, rv);
7c37ea07 331 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */
984263bc
MD
332
333 /*
334 * Check if any loadable modules need anything done at process exit.
335 * e.g. SYSV IPC stuff
336 * XXX what if one of these generates an error?
337 */
8ba5f7ef
AH
338 p->p_xstat = rv;
339 EVENTHANDLER_INVOKE(process_exit, p);
340
341 /*
342 * XXX: imho, the eventhandler stuff is much cleaner than this.
343 * Maybe we should move everything to use eventhandler.
344 */
984263bc 345 TAILQ_FOREACH(ep, &exit_list, next)
bb3cd951 346 (*ep->function)(td);
984263bc 347
4643740a 348 if (p->p_flags & P_PROFIL)
984263bc 349 stopprofclock(p);
82354ad8 350
984263bc 351 SIGEMPTYSET(p->p_siglist);
aa6c3de6 352 SIGEMPTYSET(lp->lwp_siglist);
984263bc 353 if (timevalisset(&p->p_realtimer.it_value))
a471eac5 354 callout_stop_sync(&p->p_ithandle);
984263bc
MD
355
356 /*
357 * Reset any sigio structures pointing to us as a result of
358 * F_SETOWN with our pid.
359 */
360 funsetownlst(&p->p_sigiolst);
361
362 /*
363 * Close open files and release open-file table.
364 * This may block!
365 */
0a4a9c77 366 fdfree(p, NULL);
984263bc
MD
367
368 if(p->p_leader->p_peers) {
369 q = p->p_leader;
370 while(q->p_peers != p)
371 q = q->p_peers;
372 q->p_peers = p->p_peers;
373 wakeup((caddr_t)p->p_leader);
374 }
375
376 /*
377 * XXX Shutdown SYSV semaphores
378 */
379 semexit(p);
380
06ecca5a
MD
381 KKASSERT(p->p_numposixlocks == 0);
382
984263bc
MD
383 /* The next two chunks should probably be moved to vmspace_exit. */
384 vm = p->p_vmspace;
a722be49
MD
385
386 /*
387 * Release upcalls associated with this process
388 */
389 if (vm->vm_upcalls)
08f2f1bb 390 upc_release(vm, lp);
a722be49 391
39005e16
MD
392 /*
393 * Clean up data related to virtual kernel operation. Clean up
394 * any vkernel context related to the current lwp now so we can
395 * destroy p_vkernel.
396 */
397 if (p->p_vkernel) {
398 vkernel_lwp_exit(lp);
4a22e893 399 vkernel_exit(p);
39005e16 400 }
0daa37a5 401
984263bc
MD
402 /*
403 * Release user portion of address space.
404 * This releases references to vnodes,
405 * which could cause I/O if the file has been unlinked.
406 * Need to do this early enough that we can still sleep.
407 * Can't free the entire vmspace as the kernel stack
408 * may be mapped within that space also.
409 *
410 * Processes sharing the same vmspace may exit in one order, and
411 * get cleaned up by vmspace_exit() in a different order. The
412 * last exiting process to reach this point releases as much of
413 * the environment as it can, and the last process cleaned up
414 * by vmspace_exit() (which decrements exitingcnt) cleans up the
415 * remainder.
416 */
46754a20 417 vmspace_exitbump(vm);
e3161323 418 sysref_put(&vm->vm_sysref);
984263bc
MD
419
420 if (SESS_LEADER(p)) {
1fd87d54 421 struct session *sp = p->p_session;
984263bc
MD
422
423 if (sp->s_ttyvp) {
424 /*
1fbb5fc0
MD
425 * We are the controlling process. Signal the
426 * foreground process group, drain the controlling
427 * terminal, and revoke access to the controlling
428 * terminal.
429 *
430 * NOTE: while waiting for the process group to exit
431 * it is possible that one of the processes in the
f1aeb0c0
MD
432 * group will revoke the tty, so the ttyclosesession()
433 * function will re-check sp->s_ttyvp.
984263bc
MD
434 */
435 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
436 if (sp->s_ttyp->t_pgrp)
437 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
f1aeb0c0
MD
438 ttywait(sp->s_ttyp);
439 ttyclosesession(sp, 1); /* also revoke */
984263bc 440 }
5fd012e0
MD
441 /*
442 * Release the tty. If someone has it open via
443 * /dev/tty then close it (since they no longer can
444 * once we've NULL'd it out).
445 */
f1aeb0c0
MD
446 ttyclosesession(sp, 0);
447
984263bc
MD
448 /*
449 * s_ttyp is not zero'd; we use this to indicate
450 * that the session once had a controlling terminal.
451 * (for logging and informational purposes)
452 */
453 }
454 sp->s_leader = NULL;
455 }
456 fixjobc(p, p->p_pgrp, 0);
457 (void)acct_process(p);
458#ifdef KTRACE
459 /*
460 * release trace file
461 */
29f58392
MD
462 if (p->p_tracenode)
463 ktrdestroy(&p->p_tracenode);
464 p->p_traceflag = 0;
984263bc
MD
465#endif
466 /*
467 * Release reference to text vnode
468 */
469 if ((vtmp = p->p_textvp) != NULL) {
470 p->p_textvp = NULL;
471 vrele(vtmp);
472 }
473
8ba5f7ef
AH
474 /* Release namecache handle to text file */
475 if (p->p_textnch.ncp)
476 cache_drop(&p->p_textnch);
477
984263bc 478 /*
82354ad8
MD
479 * We have to handle PPWAIT here or proc_move_allproc_zombie()
480 * will block on the PHOLD() the parent is doing.
481 */
482 if (p->p_flags & P_PPWAIT) {
483 p->p_flags &= ~P_PPWAIT;
484 wakeup(p->p_pptr);
485 }
486
487 /*
5bf0d9b5
MD
488 * Move the process to the zombie list. This will block
489 * until the process p_lock count reaches 0. The process will
490 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
491 * which is called from cpu_proc_exit().
ae8050a4 492 */
5bf0d9b5 493 proc_move_allproc_zombie(p);
984263bc 494
7e90d791
MD
495 /*
496 * Reparent all of this process's children to the init process.
497 * We must hold initproc->p_token in order to mess with
498 * initproc->p_children. We already hold p->p_token (to remove
499 * the children from our list).
500 */
984263bc 501 q = LIST_FIRST(&p->p_children);
7e90d791
MD
502 if (q) {
503 lwkt_gettoken(&initproc->p_token);
504 while (q) {
505 nq = LIST_NEXT(q, p_sibling);
506 LIST_REMOVE(q, p_sibling);
507 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
508 q->p_pptr = initproc;
509 q->p_sigparent = SIGCHLD;
510 /*
511 * Traced processes are killed
512 * since their existence means someone is screwing up.
513 */
4643740a
MD
514 if (q->p_flags & P_TRACED) {
515 q->p_flags &= ~P_TRACED;
7e90d791
MD
516 ksignal(q, SIGKILL);
517 }
518 q = nq;
984263bc 519 }
7e90d791
MD
520 lwkt_reltoken(&initproc->p_token);
521 wakeup(initproc);
984263bc
MD
522 }
523
524 /*
525 * Save exit status and final rusage info, adding in child rusage
526 * info and self times.
527 */
fde7ac71
SS
528 calcru_proc(p, &p->p_ru);
529 ruadd(&p->p_ru, &p->p_cru);
984263bc
MD
530
531 /*
984263bc
MD
532 * notify interested parties of our demise.
533 */
534 KNOTE(&p->p_klist, NOTE_EXIT);
535
536 /*
537 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
b8aacf5e
MR
538 * flag set, or if the handler is set to SIG_IGN, notify process 1
539 * instead (and hope it will handle this situation).
984263bc 540 */
b8aacf5e
MR
541 if (p->p_pptr->p_sigacts->ps_flag &
542 (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
984263bc 543 struct proc *pp = p->p_pptr;
58c2553a
MD
544
545 PHOLD(pp);
984263bc 546 proc_reparent(p, initproc);
58c2553a 547
984263bc
MD
548 /*
549 * If this was the last child of our parent, notify
550 * parent, so in case he was wait(2)ing, he will
58c2553a 551 * continue. This function interlocks with pptr->p_token.
984263bc
MD
552 */
553 if (LIST_EMPTY(&pp->p_children))
554 wakeup((caddr_t)pp);
58c2553a 555 PRELE(pp);
984263bc
MD
556 }
557
f2f3db5c
MD
558 /* lwkt_gettoken(&proc_token); */
559 q = p->p_pptr;
58c2553a 560 PHOLD(q);
f2f3db5c 561 if (p->p_sigparent && q != initproc) {
f2f3db5c 562 ksignal(q, p->p_sigparent);
984263bc 563 } else {
f2f3db5c 564 ksignal(q, SIGCHLD);
984263bc 565 }
82354ad8
MD
566
567 p->p_flags &= ~P_TRACED;
58c2553a 568 wakeup(p->p_pptr);
82354ad8 569
58c2553a 570 PRELE(q);
f2f3db5c
MD
571 /* lwkt_reltoken(&proc_token); */
572 /* NOTE: p->p_pptr can get ripped out */
984263bc 573 /*
8ad65e08
MD
574 * cpu_exit is responsible for clearing curproc, since
575 * it is heavily integrated with the thread/switching sequence.
576 *
984263bc
MD
577 * Other substructures are freed from wait().
578 */
8f1f6170 579 plimit_free(p);
984263bc
MD
580
581 /*
0a3f9b47
MD
582 * Release the current user process designation on the process so
583 * the userland scheduler can work in someone else.
26a0694b 584 */
553ea3c8 585 p->p_usched->release_curproc(lp);
26a0694b
MD
586
587 /*
e3161323
MD
588 * Finally, call machine-dependent code to release as many of the
589 * lwp's resources as we can and halt execution of this thread.
984263bc 590 */
e3161323 591 lwp_exit(1);
c6880072
SS
592}
593
0d355d3b
MD
594/*
595 * Eventually called by every exiting LWP
b5c4d81f
MD
596 *
597 * p->p_token must be held. mplock may be held and will be released.
0d355d3b 598 */
c6880072 599void
e3161323 600lwp_exit(int masterexit)
c6880072 601{
d86a23e0
MD
602 struct thread *td = curthread;
603 struct lwp *lp = td->td_lwp;
c6880072 604 struct proc *p = lp->lwp_proc;
8c2bce60 605 int dowake = 0;
c6880072 606
70d3d461 607 /*
4643740a 608 * lwp_exit() may be called without setting LWP_MP_WEXIT, so
e3161323
MD
609 * make sure it is set here.
610 */
b5c4d81f 611 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
4643740a 612 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
e3161323
MD
613
614 /*
39005e16
MD
615 * Clean up any virtualization
616 */
617 if (lp->lwp_vkernel)
618 vkernel_lwp_exit(lp);
619
620 /*
a591f597
MD
621 * Clean up select/poll support
622 */
623 kqueue_terminate(&lp->lwp_kqueue);
624
625 /*
0d355d3b
MD
626 * Clean up any syscall-cached ucred
627 */
d86a23e0
MD
628 if (td->td_ucred) {
629 crfree(td->td_ucred);
630 td->td_ucred = NULL;
0d355d3b
MD
631 }
632
633 /*
70d3d461
SS
634 * Nobody actually wakes us when the lock
635 * count reaches zero, so just wait one tick.
636 */
637 while (lp->lwp_lock > 0)
638 tsleep(lp, 0, "lwpexit", 1);
639
640 /* Hand down resource usage to our proc */
641 ruadd(&p->p_ru, &lp->lwp_ru);
642
e3161323
MD
643 /*
644 * If we don't hold the process until the LWP is reaped wait*()
645 * may try to dispose of its vmspace before all the LWPs have
646 * actually terminated.
647 */
648 PHOLD(p);
649
650 /*
2883d2d8
MD
651 * Do any remaining work that might block on us. We should be
652 * coded such that further blocking is ok after decrementing
653 * p_nthreads but don't take the chance.
654 */
655 dsched_exit_thread(td);
656 biosched_done(curthread);
657
658 /*
e3161323
MD
659 * We have to use the reaper for all the LWPs except the one doing
660 * the master exit. The LWP doing the master exit can just be
661 * left on p_lwps and the process reaper will deal with it
662 * synchronously, which is much faster.
ce19c8fe
MD
663 *
664 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0.
2af9d75d
MD
665 *
666 * The process is left held until the reaper calls lwp_dispose() on
667 * the lp (after calling lwp_wait()).
e3161323
MD
668 */
669 if (masterexit == 0) {
3e291793 670 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
e3161323 671 --p->p_nthreads;
ce19c8fe 672 if (p->p_nthreads <= 1)
8c2bce60 673 dowake = 1;
b5c4d81f 674 lwkt_gettoken(&deadlwp_token);
3e291793 675 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry);
2883d2d8
MD
676 taskqueue_enqueue(taskqueue_thread[mycpuid],
677 deadlwp_task[mycpuid]);
b5c4d81f 678 lwkt_reltoken(&deadlwp_token);
e3161323
MD
679 } else {
680 --p->p_nthreads;
ce19c8fe 681 if (p->p_nthreads <= 1)
8c2bce60 682 dowake = 1;
e3161323 683 }
b5c4d81f
MD
684
685 /*
8c2bce60
MD
686 * Release p_token. Issue the wakeup() on p_nthreads if necessary,
687 * as late as possible to give us a chance to actually deschedule and
688 * switch away before another cpu core hits reaplwp().
b5c4d81f
MD
689 */
690 lwkt_reltoken(&p->p_token);
8c2bce60
MD
691 if (dowake)
692 wakeup(&p->p_nthreads);
c6880072
SS
693 cpu_lwp_exit();
694}
695
696/*
2af9d75d
MD
697 * Wait until a lwp is completely dead. The final interlock in this drama
698 * is when TDF_EXITING is set in cpu_thread_exit() just before the final
699 * switchout.
c6880072 700 *
2af9d75d
MD
701 * At the point TDF_EXITING is set a complete exit is accomplished when
702 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear.
c6880072 703 *
2af9d75d
MD
704 * Returns non-zero on success, and zero if the caller needs to retry
705 * the lwp_wait().
c6880072
SS
706 */
707static int
708lwp_wait(struct lwp *lp)
709{
710 struct thread *td = lp->lwp_thread;;
711
712 KKASSERT(lwkt_preempted_proc() != lp);
713
2af9d75d
MD
714 /*
715 * Wait until the lp has entered its low level exit and wait
716 * until other cores with refs on the lp (e.g. for ps or signaling)
717 * release them.
718 */
719 if (lp->lwp_lock > 0) {
c6880072 720 tsleep(lp, 0, "lwpwait1", 1);
2af9d75d
MD
721 return(0);
722 }
c6880072 723
2af9d75d
MD
724 /*
725 * Wait until the thread is no longer references and no longer
726 * runnable or preempted (i.e. finishes its low level exit).
727 */
728 if (td->td_refs) {
729 tsleep(td, 0, "lwpwait2", 1);
730 return(0);
731 }
c6880072
SS
732
733 /*
734 * The lwp's thread may still be in the middle
735 * of switching away, we can't rip its stack out from
736 * under it until TDF_EXITING is set and both
737 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
738 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
739 * will be cleared temporarily if a thread gets
740 * preempted.
741 *
742 * YYY no wakeup occurs, so we simply return failure
743 * and let the caller deal with sleeping and calling
744 * us again.
745 */
2af9d75d
MD
746 if ((td->td_flags & (TDF_RUNNING |
747 TDF_PREEMPT_LOCK |
748 TDF_EXITING)) != TDF_EXITING) {
749 tsleep(lp, 0, "lwpwait2", 1);
c6880072 750 return (0);
cf709dd2 751 }
2af9d75d
MD
752 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0,
753 ("lwp_wait: td %p (%s) still on run or sleep queue",
754 td, td->td_comm));
c6880072
SS
755 return (1);
756}
757
758/*
759 * Release the resources associated with a lwp.
760 * The lwp must be completely dead.
761 */
762void
763lwp_dispose(struct lwp *lp)
764{
765 struct thread *td = lp->lwp_thread;;
766
767 KKASSERT(lwkt_preempted_proc() != lp);
768 KKASSERT(td->td_refs == 0);
2af9d75d
MD
769 KKASSERT((td->td_flags & (TDF_RUNNING |
770 TDF_PREEMPT_LOCK |
771 TDF_EXITING)) == TDF_EXITING);
c6880072 772
e3161323
MD
773 PRELE(lp->lwp_proc);
774 lp->lwp_proc = NULL;
c6880072
SS
775 if (td != NULL) {
776 td->td_proc = NULL;
777 td->td_lwp = NULL;
778 lp->lwp_thread = NULL;
779 lwkt_free_thread(td);
780 }
f6c36234 781 kfree(lp, M_LWP);
984263bc
MD
782}
783
3919ced0
MD
784/*
785 * MPSAFE
786 */
984263bc 787int
753fd850 788sys_wait4(struct wait_args *uap)
984263bc 789{
9697c509
DRJ
790 struct rusage rusage;
791 int error, status;
984263bc 792
3919ced0
MD
793 error = kern_wait(uap->pid, (uap->status ? &status : NULL),
794 uap->options, (uap->rusage ? &rusage : NULL),
795 &uap->sysmsg_result);
984263bc 796
9697c509
DRJ
797 if (error == 0 && uap->status)
798 error = copyout(&status, uap->status, sizeof(*uap->status));
799 if (error == 0 && uap->rusage)
800 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
801 return (error);
984263bc
MD
802}
803
41c20dac
MD
804/*
805 * wait1()
806 *
807 * wait_args(int pid, int *status, int options, struct rusage *rusage)
3919ced0
MD
808 *
809 * MPALMOSTSAFE
41c20dac 810 */
9697c509
DRJ
811int
812kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
984263bc 813{
9697c509 814 struct thread *td = curthread;
3e291793 815 struct lwp *lp;
9697c509 816 struct proc *q = td->td_proc;
41c20dac 817 struct proc *p, *t;
e955543c 818 struct pargs *pa;
6fa9e71a 819 struct sigacts *ps;
9697c509 820 int nfound, error;
984263bc 821
9697c509
DRJ
822 if (pid == 0)
823 pid = -q->p_pgid;
cc162105 824 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
984263bc 825 return (EINVAL);
b5c4d81f
MD
826
827 lwkt_gettoken(&q->p_token);
984263bc 828loop:
6d5b152e 829 /*
b5c4d81f
MD
830 * All sorts of things can change due to blocking so we have to loop
831 * all the way back up here.
6d5b152e
MD
832 *
833 * The problem is that if a process group is stopped and the parent
834 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
835 * of the child and then stop itself when it tries to return from the
836 * system call. When the process group is resumed the parent will
837 * then get the STOP status even though the child has now resumed
838 * (a followup wait*() will get the CONT status).
839 *
840 * Previously the CONT would overwrite the STOP because the tstop
841 * was handled within tsleep(), and the parent would only see
b5c4d81f 842 * the CONT when both are stopped and continued together. This little
6d5b152e
MD
843 * two-line hack restores this effect.
844 */
164b8401 845 while (q->p_stat == SSTOP)
9a379a4a 846 tstop();
6d5b152e 847
984263bc 848 nfound = 0;
b5c4d81f 849
8c986a82
MD
850 /*
851 * Loop on children.
852 *
853 * NOTE: We don't want to break q's p_token in the loop for the
854 * case where no children are found or we risk breaking the
855 * interlock between child and parent.
856 */
984263bc 857 LIST_FOREACH(p, &q->p_children, p_sibling) {
9697c509 858 if (pid != WAIT_ANY &&
b5c4d81f 859 p->p_pid != pid && p->p_pgid != -pid) {
984263bc 860 continue;
b5c4d81f 861 }
984263bc 862
5686ec5a
MD
863 /*
864 * This special case handles a kthread spawned by linux_clone
352f5709
MD
865 * (see linux_misc.c). The linux_wait4 and linux_waitpid
866 * functions need to be able to distinguish between waiting
867 * on a process and waiting on a thread. It is a thread if
868 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
869 * signifies we want to wait for threads and not processes.
984263bc 870 */
352f5709
MD
871 if ((p->p_sigparent != SIGCHLD) ^
872 ((options & WLINUXCLONE) != 0)) {
984263bc 873 continue;
352f5709 874 }
984263bc
MD
875
876 nfound++;
416d05d7 877 if (p->p_stat == SZOMB) {
ae8050a4 878 /*
8f211c4b
MD
879 * We may go into SZOMB with threads still present.
880 * We must wait for them to exit before we can reap
881 * the master thread, otherwise we may race reaping
882 * non-master threads.
883 */
b5c4d81f 884 lwkt_gettoken(&p->p_token);
8f211c4b
MD
885 while (p->p_nthreads > 0) {
886 tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
887 }
888
889 /*
e3161323
MD
890 * Reap any LWPs left in p->p_lwps. This is usually
891 * just the last LWP. This must be done before
892 * we loop on p_lock since the lwps hold a ref on
893 * it as a vmspace interlock.
894 *
895 * Once that is accomplished p_nthreads had better
896 * be zero.
897 */
3e291793
MD
898 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
899 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
900 reaplwp(lp);
901 }
e3161323
MD
902 KKASSERT(p->p_nthreads == 0);
903
904 /*
905 * Don't do anything really bad until all references
906 * to the process go away. This may include other
907 * LWPs which are still in the process of being
908 * reaped. We can't just pull the rug out from under
909 * them because they may still be using the VM space.
910 *
911 * Certain kernel facilities such as /proc will also
912 * put a hold on the process for short periods of
913 * time.
c008d3ad 914 */
82354ad8 915 PSTALL(p, "reap3", 0);
c1102e9f 916
9697c509
DRJ
917 /* Take care of our return values. */
918 *res = p->p_pid;
52cac9fb
MD
919 p->p_usched->heuristic_exiting(td->td_lwp, p);
920
9697c509
DRJ
921 if (status)
922 *status = p->p_xstat;
923 if (rusage)
fde7ac71 924 *rusage = p->p_ru;
984263bc
MD
925 /*
926 * If we got the child via a ptrace 'attach',
927 * we need to give it back to the old parent.
928 */
58c2553a 929 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
984263bc
MD
930 p->p_oppid = 0;
931 proc_reparent(p, t);
84204577 932 ksignal(t, SIGCHLD);
984263bc 933 wakeup((caddr_t)t);
3919ced0 934 error = 0;
58c2553a 935 PRELE(t);
eb2adbf5 936 lwkt_reltoken(&p->p_token);
3919ced0 937 goto done;
984263bc 938 }
3c8687d6
SS
939
940 /*
941 * Unlink the proc from its process group so that
942 * the following operations won't lead to an
943 * inconsistent state for processes running down
944 * the zombie list.
945 */
3c8687d6 946 proc_remove_zombie(p);
eb2adbf5 947 lwkt_reltoken(&p->p_token);
3c8687d6
SS
948 leavepgrp(p);
949
984263bc 950 p->p_xstat = 0;
fde7ac71 951 ruadd(&q->p_cru, &p->p_ru);
984263bc
MD
952
953 /*
954 * Decrement the count of procs running with this uid.
955 */
9697c509 956 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
984263bc
MD
957
958 /*
959 * Free up credentials.
960 */
41c20dac
MD
961 crfree(p->p_ucred);
962 p->p_ucred = NULL;
984263bc
MD
963
964 /*
965 * Remove unused arguments
966 */
e955543c
MD
967 pa = p->p_args;
968 p->p_args = NULL;
e955543c
MD
969 if (pa && refcount_release(&pa->ar_ref)) {
970 kfree(pa, M_PARGS);
971 pa = NULL;
972 }
6fa9e71a
MD
973
974 ps = p->p_sigacts;
975 p->p_sigacts = NULL;
976 if (ps && refcount_release(&ps->ps_refcnt)) {
977 kfree(ps, M_SUBPROC);
978 ps = NULL;
984263bc
MD
979 }
980
49aa3df0 981 /*
e22b17d7
MD
982 * Our exitingcount was incremented when the process
983 * became a zombie, now that the process has been
984 * removed from (almost) all lists we should be able
985 * to safely destroy its vmspace. Wait for any current
986 * holders to go away (so the vmspace remains stable),
987 * then scrap it.
49aa3df0 988 */
82354ad8 989 PSTALL(p, "reap4", 0);
e22b17d7 990 vmspace_exitfree(p);
82354ad8 991 PSTALL(p, "reap5", 0);
e22b17d7 992
fb2a331e 993 kfree(p, M_PROC);
8c2bce60 994 atomic_add_int(&nprocs, -1);
3919ced0
MD
995 error = 0;
996 goto done;
984263bc 997 }
4643740a
MD
998 if (p->p_stat == SSTOP && (p->p_flags & P_WAITED) == 0 &&
999 ((p->p_flags & P_TRACED) || (options & WUNTRACED))) {
8c986a82 1000 lwkt_gettoken(&p->p_token);
4643740a 1001 p->p_flags |= P_WAITED;
9697c509
DRJ
1002
1003 *res = p->p_pid;
52cac9fb 1004 p->p_usched->heuristic_exiting(td->td_lwp, p);
9697c509 1005 if (status)
a89c93ee 1006 *status = W_STOPCODE(p->p_xstat);
9697c509
DRJ
1007 /* Zero rusage so we get something consistent. */
1008 if (rusage)
7eeb4207 1009 bzero(rusage, sizeof(*rusage));
3919ced0 1010 error = 0;
8c986a82 1011 lwkt_reltoken(&p->p_token);
3919ced0 1012 goto done;
984263bc 1013 }
4643740a 1014 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) {
8c986a82 1015 lwkt_gettoken(&p->p_token);
cc162105 1016 *res = p->p_pid;
52cac9fb 1017 p->p_usched->heuristic_exiting(td->td_lwp, p);
4643740a 1018 p->p_flags &= ~P_CONTINUED;
cc162105
PA
1019
1020 if (status)
1021 *status = SIGCONT;
3919ced0 1022 error = 0;
8c986a82 1023 lwkt_reltoken(&p->p_token);
3919ced0 1024 goto done;
cc162105 1025 }
984263bc 1026 }
3919ced0
MD
1027 if (nfound == 0) {
1028 error = ECHILD;
1029 goto done;
1030 }
9697c509
DRJ
1031 if (options & WNOHANG) {
1032 *res = 0;
3919ced0
MD
1033 error = 0;
1034 goto done;
984263bc 1035 }
58c2553a
MD
1036
1037 /*
1038 * Wait for signal - interlocked using q->p_token.
1039 */
1040 error = tsleep(q, PCATCH, "wait", 0);
3919ced0
MD
1041 if (error) {
1042done:
b5c4d81f 1043 lwkt_reltoken(&q->p_token);
984263bc 1044 return (error);
3919ced0 1045 }
984263bc
MD
1046 goto loop;
1047}
1048
1049/*
b5c4d81f 1050 * Make process 'parent' the new parent of process 'child'.
7e90d791
MD
1051 *
1052 * p_children/p_sibling requires the parent's token, and
1053 * changing pptr requires the child's token, so we have to
1054 * get three tokens to do this operation.
984263bc
MD
1055 */
1056void
77153250 1057proc_reparent(struct proc *child, struct proc *parent)
984263bc 1058{
7e90d791
MD
1059 struct proc *opp = child->p_pptr;
1060
1061 if (opp == parent)
984263bc 1062 return;
7e90d791 1063 PHOLD(opp);
b5c4d81f 1064 PHOLD(parent);
7e90d791 1065 lwkt_gettoken(&opp->p_token);
b5c4d81f
MD
1066 lwkt_gettoken(&child->p_token);
1067 lwkt_gettoken(&parent->p_token);
7e90d791 1068 KKASSERT(child->p_pptr == opp);
984263bc
MD
1069 LIST_REMOVE(child, p_sibling);
1070 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1071 child->p_pptr = parent;
b5c4d81f
MD
1072 lwkt_reltoken(&parent->p_token);
1073 lwkt_reltoken(&child->p_token);
7e90d791 1074 lwkt_reltoken(&opp->p_token);
b5c4d81f 1075 PRELE(parent);
7e90d791 1076 PRELE(opp);
984263bc
MD
1077}
1078
1079/*
1080 * The next two functions are to handle adding/deleting items on the
1081 * exit callout list
1082 *
1083 * at_exit():
1084 * Take the arguments given and put them onto the exit callout list,
1085 * However first make sure that it's not already there.
1086 * returns 0 on success.
1087 */
1088
1089int
77153250 1090at_exit(exitlist_fn function)
984263bc
MD
1091{
1092 struct exitlist *ep;
1093
1094#ifdef INVARIANTS
1095 /* Be noisy if the programmer has lost track of things */
1096 if (rm_at_exit(function))
6ea70f76 1097 kprintf("WARNING: exit callout entry (%p) already present\n",
984263bc
MD
1098 function);
1099#endif
efda3bd0 1100 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
984263bc
MD
1101 if (ep == NULL)
1102 return (ENOMEM);
1103 ep->function = function;
1104 TAILQ_INSERT_TAIL(&exit_list, ep, next);
1105 return (0);
1106}
1107
1108/*
1109 * Scan the exit callout list for the given item and remove it.
1110 * Returns the number of items removed (0 or 1)
1111 */
1112int
77153250 1113rm_at_exit(exitlist_fn function)
984263bc
MD
1114{
1115 struct exitlist *ep;
1116
1117 TAILQ_FOREACH(ep, &exit_list, next) {
1118 if (ep->function == function) {
1119 TAILQ_REMOVE(&exit_list, ep, next);
efda3bd0 1120 kfree(ep, M_ATEXIT);
984263bc
MD
1121 return(1);
1122 }
1123 }
1124 return (0);
1125}
1126
70d3d461
SS
1127/*
1128 * LWP reaper related code.
1129 */
70d3d461
SS
1130static void
1131reaplwps(void *context, int dummy)
1132{
1133 struct lwplist *lwplist = context;
1134 struct lwp *lp;
1135
b5c4d81f 1136 lwkt_gettoken(&deadlwp_token);
70d3d461 1137 while ((lp = LIST_FIRST(lwplist))) {
3e291793
MD
1138 LIST_REMOVE(lp, u.lwp_reap_entry);
1139 reaplwp(lp);
70d3d461 1140 }
b5c4d81f 1141 lwkt_reltoken(&deadlwp_token);
70d3d461
SS
1142}
1143
1144static void
3e291793
MD
1145reaplwp(struct lwp *lp)
1146{
2af9d75d
MD
1147 while (lwp_wait(lp) == 0)
1148 ;
3e291793
MD
1149 lwp_dispose(lp);
1150}
1151
1152static void
70d3d461
SS
1153deadlwp_init(void)
1154{
1155 int cpu;
1156
1157 for (cpu = 0; cpu < ncpus; cpu++) {
1158 LIST_INIT(&deadlwp_list[cpu]);
2af9d75d
MD
1159 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]),
1160 M_DEVBUF, M_WAITOK);
70d3d461
SS
1161 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1162 }
1163}
1164
1165SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);