kernel - Major signal path adjustments to fix races, tsleep race fixes, +more
[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);
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);
330 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */
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);
984263bc
MD
349 /*
350 * If parent is waiting for us to exit or exec,
351 * P_PPWAIT is set; we will wakeup the parent below.
352 */
4643740a 353 p->p_flags &= ~(P_TRACED | P_PPWAIT);
984263bc 354 SIGEMPTYSET(p->p_siglist);
aa6c3de6 355 SIGEMPTYSET(lp->lwp_siglist);
984263bc 356 if (timevalisset(&p->p_realtimer.it_value))
8fbf9130 357 callout_stop(&p->p_ithandle);
984263bc
MD
358
359 /*
360 * Reset any sigio structures pointing to us as a result of
361 * F_SETOWN with our pid.
362 */
363 funsetownlst(&p->p_sigiolst);
364
365 /*
366 * Close open files and release open-file table.
367 * This may block!
368 */
0a4a9c77 369 fdfree(p, NULL);
984263bc
MD
370
371 if(p->p_leader->p_peers) {
372 q = p->p_leader;
373 while(q->p_peers != p)
374 q = q->p_peers;
375 q->p_peers = p->p_peers;
376 wakeup((caddr_t)p->p_leader);
377 }
378
379 /*
380 * XXX Shutdown SYSV semaphores
381 */
382 semexit(p);
383
06ecca5a
MD
384 KKASSERT(p->p_numposixlocks == 0);
385
984263bc
MD
386 /* The next two chunks should probably be moved to vmspace_exit. */
387 vm = p->p_vmspace;
a722be49
MD
388
389 /*
390 * Release upcalls associated with this process
391 */
392 if (vm->vm_upcalls)
08f2f1bb 393 upc_release(vm, lp);
a722be49 394
39005e16
MD
395 /*
396 * Clean up data related to virtual kernel operation. Clean up
397 * any vkernel context related to the current lwp now so we can
398 * destroy p_vkernel.
399 */
400 if (p->p_vkernel) {
401 vkernel_lwp_exit(lp);
4a22e893 402 vkernel_exit(p);
39005e16 403 }
0daa37a5 404
984263bc
MD
405 /*
406 * Release user portion of address space.
407 * This releases references to vnodes,
408 * which could cause I/O if the file has been unlinked.
409 * Need to do this early enough that we can still sleep.
410 * Can't free the entire vmspace as the kernel stack
411 * may be mapped within that space also.
412 *
413 * Processes sharing the same vmspace may exit in one order, and
414 * get cleaned up by vmspace_exit() in a different order. The
415 * last exiting process to reach this point releases as much of
416 * the environment as it can, and the last process cleaned up
417 * by vmspace_exit() (which decrements exitingcnt) cleans up the
418 * remainder.
419 */
46754a20 420 vmspace_exitbump(vm);
e3161323 421 sysref_put(&vm->vm_sysref);
984263bc
MD
422
423 if (SESS_LEADER(p)) {
1fd87d54 424 struct session *sp = p->p_session;
984263bc
MD
425
426 if (sp->s_ttyvp) {
427 /*
1fbb5fc0
MD
428 * We are the controlling process. Signal the
429 * foreground process group, drain the controlling
430 * terminal, and revoke access to the controlling
431 * terminal.
432 *
433 * NOTE: while waiting for the process group to exit
434 * it is possible that one of the processes in the
f1aeb0c0
MD
435 * group will revoke the tty, so the ttyclosesession()
436 * function will re-check sp->s_ttyvp.
984263bc
MD
437 */
438 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
439 if (sp->s_ttyp->t_pgrp)
440 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
f1aeb0c0
MD
441 ttywait(sp->s_ttyp);
442 ttyclosesession(sp, 1); /* also revoke */
984263bc 443 }
5fd012e0
MD
444 /*
445 * Release the tty. If someone has it open via
446 * /dev/tty then close it (since they no longer can
447 * once we've NULL'd it out).
448 */
f1aeb0c0
MD
449 ttyclosesession(sp, 0);
450
984263bc
MD
451 /*
452 * s_ttyp is not zero'd; we use this to indicate
453 * that the session once had a controlling terminal.
454 * (for logging and informational purposes)
455 */
456 }
457 sp->s_leader = NULL;
458 }
459 fixjobc(p, p->p_pgrp, 0);
460 (void)acct_process(p);
461#ifdef KTRACE
462 /*
463 * release trace file
464 */
29f58392
MD
465 if (p->p_tracenode)
466 ktrdestroy(&p->p_tracenode);
467 p->p_traceflag = 0;
984263bc
MD
468#endif
469 /*
470 * Release reference to text vnode
471 */
472 if ((vtmp = p->p_textvp) != NULL) {
473 p->p_textvp = NULL;
474 vrele(vtmp);
475 }
476
8ba5f7ef
AH
477 /* Release namecache handle to text file */
478 if (p->p_textnch.ncp)
479 cache_drop(&p->p_textnch);
480
984263bc 481 /*
5bf0d9b5
MD
482 * Move the process to the zombie list. This will block
483 * until the process p_lock count reaches 0. The process will
484 * not be reaped until TDF_EXITING is set by cpu_thread_exit(),
485 * which is called from cpu_proc_exit().
ae8050a4 486 */
5bf0d9b5 487 proc_move_allproc_zombie(p);
984263bc 488
7e90d791
MD
489 /*
490 * Reparent all of this process's children to the init process.
491 * We must hold initproc->p_token in order to mess with
492 * initproc->p_children. We already hold p->p_token (to remove
493 * the children from our list).
494 */
984263bc 495 q = LIST_FIRST(&p->p_children);
7e90d791
MD
496 if (q) {
497 lwkt_gettoken(&initproc->p_token);
498 while (q) {
499 nq = LIST_NEXT(q, p_sibling);
500 LIST_REMOVE(q, p_sibling);
501 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling);
502 q->p_pptr = initproc;
503 q->p_sigparent = SIGCHLD;
504 /*
505 * Traced processes are killed
506 * since their existence means someone is screwing up.
507 */
4643740a
MD
508 if (q->p_flags & P_TRACED) {
509 q->p_flags &= ~P_TRACED;
7e90d791
MD
510 ksignal(q, SIGKILL);
511 }
512 q = nq;
984263bc 513 }
7e90d791
MD
514 lwkt_reltoken(&initproc->p_token);
515 wakeup(initproc);
984263bc
MD
516 }
517
518 /*
519 * Save exit status and final rusage info, adding in child rusage
520 * info and self times.
521 */
fde7ac71
SS
522 calcru_proc(p, &p->p_ru);
523 ruadd(&p->p_ru, &p->p_cru);
984263bc
MD
524
525 /*
984263bc
MD
526 * notify interested parties of our demise.
527 */
528 KNOTE(&p->p_klist, NOTE_EXIT);
529
530 /*
531 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT
532 * flag set, notify process 1 instead (and hope it will handle
533 * this situation).
534 */
b1b4e5a6 535 if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) {
984263bc 536 struct proc *pp = p->p_pptr;
58c2553a
MD
537
538 PHOLD(pp);
984263bc 539 proc_reparent(p, initproc);
58c2553a 540
984263bc
MD
541 /*
542 * If this was the last child of our parent, notify
543 * parent, so in case he was wait(2)ing, he will
58c2553a 544 * continue. This function interlocks with pptr->p_token.
984263bc
MD
545 */
546 if (LIST_EMPTY(&pp->p_children))
547 wakeup((caddr_t)pp);
58c2553a 548 PRELE(pp);
984263bc
MD
549 }
550
f2f3db5c
MD
551 /* lwkt_gettoken(&proc_token); */
552 q = p->p_pptr;
58c2553a 553 PHOLD(q);
f2f3db5c 554 if (p->p_sigparent && q != initproc) {
f2f3db5c 555 ksignal(q, p->p_sigparent);
984263bc 556 } else {
f2f3db5c 557 ksignal(q, SIGCHLD);
984263bc 558 }
58c2553a
MD
559 wakeup(p->p_pptr);
560 PRELE(q);
f2f3db5c
MD
561 /* lwkt_reltoken(&proc_token); */
562 /* NOTE: p->p_pptr can get ripped out */
984263bc 563 /*
8ad65e08
MD
564 * cpu_exit is responsible for clearing curproc, since
565 * it is heavily integrated with the thread/switching sequence.
566 *
984263bc
MD
567 * Other substructures are freed from wait().
568 */
8f1f6170 569 plimit_free(p);
984263bc
MD
570
571 /*
0a3f9b47
MD
572 * Release the current user process designation on the process so
573 * the userland scheduler can work in someone else.
26a0694b 574 */
553ea3c8 575 p->p_usched->release_curproc(lp);
26a0694b
MD
576
577 /*
e3161323
MD
578 * Finally, call machine-dependent code to release as many of the
579 * lwp's resources as we can and halt execution of this thread.
984263bc 580 */
e3161323 581 lwp_exit(1);
c6880072
SS
582}
583
0d355d3b
MD
584/*
585 * Eventually called by every exiting LWP
b5c4d81f
MD
586 *
587 * p->p_token must be held. mplock may be held and will be released.
0d355d3b 588 */
c6880072 589void
e3161323 590lwp_exit(int masterexit)
c6880072 591{
d86a23e0
MD
592 struct thread *td = curthread;
593 struct lwp *lp = td->td_lwp;
c6880072 594 struct proc *p = lp->lwp_proc;
8c2bce60 595 int dowake = 0;
c6880072 596
70d3d461 597 /*
4643740a 598 * lwp_exit() may be called without setting LWP_MP_WEXIT, so
e3161323
MD
599 * make sure it is set here.
600 */
b5c4d81f 601 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
4643740a 602 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT);
e3161323
MD
603
604 /*
39005e16
MD
605 * Clean up any virtualization
606 */
607 if (lp->lwp_vkernel)
608 vkernel_lwp_exit(lp);
609
610 /*
a591f597
MD
611 * Clean up select/poll support
612 */
613 kqueue_terminate(&lp->lwp_kqueue);
614
615 /*
0d355d3b
MD
616 * Clean up any syscall-cached ucred
617 */
d86a23e0
MD
618 if (td->td_ucred) {
619 crfree(td->td_ucred);
620 td->td_ucred = NULL;
0d355d3b
MD
621 }
622
623 /*
70d3d461
SS
624 * Nobody actually wakes us when the lock
625 * count reaches zero, so just wait one tick.
626 */
627 while (lp->lwp_lock > 0)
628 tsleep(lp, 0, "lwpexit", 1);
629
630 /* Hand down resource usage to our proc */
631 ruadd(&p->p_ru, &lp->lwp_ru);
632
e3161323
MD
633 /*
634 * If we don't hold the process until the LWP is reaped wait*()
635 * may try to dispose of its vmspace before all the LWPs have
636 * actually terminated.
637 */
638 PHOLD(p);
639
640 /*
2883d2d8
MD
641 * Do any remaining work that might block on us. We should be
642 * coded such that further blocking is ok after decrementing
643 * p_nthreads but don't take the chance.
644 */
645 dsched_exit_thread(td);
646 biosched_done(curthread);
647
648 /*
e3161323
MD
649 * We have to use the reaper for all the LWPs except the one doing
650 * the master exit. The LWP doing the master exit can just be
651 * left on p_lwps and the process reaper will deal with it
652 * synchronously, which is much faster.
ce19c8fe
MD
653 *
654 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0.
e3161323
MD
655 */
656 if (masterexit == 0) {
3e291793 657 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
e3161323 658 --p->p_nthreads;
ce19c8fe 659 if (p->p_nthreads <= 1)
8c2bce60 660 dowake = 1;
b5c4d81f 661 lwkt_gettoken(&deadlwp_token);
3e291793 662 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry);
2883d2d8
MD
663 taskqueue_enqueue(taskqueue_thread[mycpuid],
664 deadlwp_task[mycpuid]);
b5c4d81f 665 lwkt_reltoken(&deadlwp_token);
e3161323
MD
666 } else {
667 --p->p_nthreads;
ce19c8fe 668 if (p->p_nthreads <= 1)
8c2bce60 669 dowake = 1;
e3161323 670 }
b5c4d81f
MD
671
672 /*
8c2bce60
MD
673 * Release p_token. Issue the wakeup() on p_nthreads if necessary,
674 * as late as possible to give us a chance to actually deschedule and
675 * switch away before another cpu core hits reaplwp().
b5c4d81f
MD
676 */
677 lwkt_reltoken(&p->p_token);
8c2bce60
MD
678 if (dowake)
679 wakeup(&p->p_nthreads);
c6880072
SS
680 cpu_lwp_exit();
681}
682
683/*
684 * Wait until a lwp is completely dead.
685 *
686 * If the thread is still executing, which can't be waited upon,
687 * return failure. The caller is responsible of waiting a little
688 * bit and checking again.
689 *
690 * Suggested use:
691 * while (!lwp_wait(lp))
692 * tsleep(lp, 0, "lwpwait", 1);
693 */
694static int
695lwp_wait(struct lwp *lp)
696{
697 struct thread *td = lp->lwp_thread;;
698
699 KKASSERT(lwkt_preempted_proc() != lp);
700
701 while (lp->lwp_lock > 0)
702 tsleep(lp, 0, "lwpwait1", 1);
703
704 lwkt_wait_free(td);
705
706 /*
707 * The lwp's thread may still be in the middle
708 * of switching away, we can't rip its stack out from
709 * under it until TDF_EXITING is set and both
710 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear.
711 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING
712 * will be cleared temporarily if a thread gets
713 * preempted.
714 *
715 * YYY no wakeup occurs, so we simply return failure
716 * and let the caller deal with sleeping and calling
717 * us again.
718 */
cf709dd2
MD
719 if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|
720 TDF_EXITING|TDF_RUNQ)) != TDF_EXITING) {
c6880072 721 return (0);
cf709dd2 722 }
74c9628e
MD
723 KASSERT((td->td_flags & TDF_TSLEEPQ) == 0,
724 ("lwp_wait: td %p (%s) still on sleep queue", td, td->td_comm));
c6880072
SS
725 return (1);
726}
727
728/*
729 * Release the resources associated with a lwp.
730 * The lwp must be completely dead.
731 */
732void
733lwp_dispose(struct lwp *lp)
734{
735 struct thread *td = lp->lwp_thread;;
736
737 KKASSERT(lwkt_preempted_proc() != lp);
738 KKASSERT(td->td_refs == 0);
739 KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) ==
740 TDF_EXITING);
741
e3161323
MD
742 PRELE(lp->lwp_proc);
743 lp->lwp_proc = NULL;
c6880072
SS
744 if (td != NULL) {
745 td->td_proc = NULL;
746 td->td_lwp = NULL;
747 lp->lwp_thread = NULL;
748 lwkt_free_thread(td);
749 }
f6c36234 750 kfree(lp, M_LWP);
984263bc
MD
751}
752
3919ced0
MD
753/*
754 * MPSAFE
755 */
984263bc 756int
753fd850 757sys_wait4(struct wait_args *uap)
984263bc 758{
9697c509
DRJ
759 struct rusage rusage;
760 int error, status;
984263bc 761
3919ced0
MD
762 error = kern_wait(uap->pid, (uap->status ? &status : NULL),
763 uap->options, (uap->rusage ? &rusage : NULL),
764 &uap->sysmsg_result);
984263bc 765
9697c509
DRJ
766 if (error == 0 && uap->status)
767 error = copyout(&status, uap->status, sizeof(*uap->status));
768 if (error == 0 && uap->rusage)
769 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage));
770 return (error);
984263bc
MD
771}
772
41c20dac
MD
773/*
774 * wait1()
775 *
776 * wait_args(int pid, int *status, int options, struct rusage *rusage)
3919ced0
MD
777 *
778 * MPALMOSTSAFE
41c20dac 779 */
9697c509
DRJ
780int
781kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res)
984263bc 782{
9697c509 783 struct thread *td = curthread;
3e291793 784 struct lwp *lp;
9697c509 785 struct proc *q = td->td_proc;
41c20dac 786 struct proc *p, *t;
e955543c 787 struct pargs *pa;
6fa9e71a 788 struct sigacts *ps;
9697c509 789 int nfound, error;
984263bc 790
9697c509
DRJ
791 if (pid == 0)
792 pid = -q->p_pgid;
cc162105 793 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
984263bc 794 return (EINVAL);
b5c4d81f
MD
795
796 lwkt_gettoken(&q->p_token);
984263bc 797loop:
6d5b152e 798 /*
b5c4d81f
MD
799 * All sorts of things can change due to blocking so we have to loop
800 * all the way back up here.
6d5b152e
MD
801 *
802 * The problem is that if a process group is stopped and the parent
803 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP
804 * of the child and then stop itself when it tries to return from the
805 * system call. When the process group is resumed the parent will
806 * then get the STOP status even though the child has now resumed
807 * (a followup wait*() will get the CONT status).
808 *
809 * Previously the CONT would overwrite the STOP because the tstop
810 * was handled within tsleep(), and the parent would only see
b5c4d81f 811 * the CONT when both are stopped and continued together. This little
6d5b152e
MD
812 * two-line hack restores this effect.
813 */
164b8401 814 while (q->p_stat == SSTOP)
9a379a4a 815 tstop();
6d5b152e 816
984263bc 817 nfound = 0;
b5c4d81f 818
8c986a82
MD
819 /*
820 * Loop on children.
821 *
822 * NOTE: We don't want to break q's p_token in the loop for the
823 * case where no children are found or we risk breaking the
824 * interlock between child and parent.
825 */
984263bc 826 LIST_FOREACH(p, &q->p_children, p_sibling) {
9697c509 827 if (pid != WAIT_ANY &&
b5c4d81f 828 p->p_pid != pid && p->p_pgid != -pid) {
984263bc 829 continue;
b5c4d81f 830 }
984263bc 831
5686ec5a
MD
832 /*
833 * This special case handles a kthread spawned by linux_clone
352f5709
MD
834 * (see linux_misc.c). The linux_wait4 and linux_waitpid
835 * functions need to be able to distinguish between waiting
836 * on a process and waiting on a thread. It is a thread if
837 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
838 * signifies we want to wait for threads and not processes.
984263bc 839 */
352f5709
MD
840 if ((p->p_sigparent != SIGCHLD) ^
841 ((options & WLINUXCLONE) != 0)) {
984263bc 842 continue;
352f5709 843 }
984263bc
MD
844
845 nfound++;
416d05d7 846 if (p->p_stat == SZOMB) {
ae8050a4 847 /*
8f211c4b
MD
848 * We may go into SZOMB with threads still present.
849 * We must wait for them to exit before we can reap
850 * the master thread, otherwise we may race reaping
851 * non-master threads.
852 */
b5c4d81f 853 lwkt_gettoken(&p->p_token);
8f211c4b
MD
854 while (p->p_nthreads > 0) {
855 tsleep(&p->p_nthreads, 0, "lwpzomb", hz);
856 }
857
858 /*
e3161323
MD
859 * Reap any LWPs left in p->p_lwps. This is usually
860 * just the last LWP. This must be done before
861 * we loop on p_lock since the lwps hold a ref on
862 * it as a vmspace interlock.
863 *
864 * Once that is accomplished p_nthreads had better
865 * be zero.
866 */
3e291793
MD
867 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) {
868 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp);
869 reaplwp(lp);
870 }
e3161323 871 KKASSERT(p->p_nthreads == 0);
b5c4d81f 872 lwkt_reltoken(&p->p_token);
e3161323
MD
873
874 /*
875 * Don't do anything really bad until all references
876 * to the process go away. This may include other
877 * LWPs which are still in the process of being
878 * reaped. We can't just pull the rug out from under
879 * them because they may still be using the VM space.
880 *
881 * Certain kernel facilities such as /proc will also
882 * put a hold on the process for short periods of
883 * time.
c008d3ad 884 */
c6880072
SS
885 while (p->p_lock)
886 tsleep(p, 0, "reap3", hz);
c1102e9f 887
9697c509
DRJ
888 /* Take care of our return values. */
889 *res = p->p_pid;
52cac9fb
MD
890 p->p_usched->heuristic_exiting(td->td_lwp, p);
891
9697c509
DRJ
892 if (status)
893 *status = p->p_xstat;
894 if (rusage)
fde7ac71 895 *rusage = p->p_ru;
984263bc
MD
896 /*
897 * If we got the child via a ptrace 'attach',
898 * we need to give it back to the old parent.
899 */
58c2553a 900 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
984263bc
MD
901 p->p_oppid = 0;
902 proc_reparent(p, t);
84204577 903 ksignal(t, SIGCHLD);
984263bc 904 wakeup((caddr_t)t);
3919ced0 905 error = 0;
58c2553a 906 PRELE(t);
3919ced0 907 goto done;
984263bc 908 }
3c8687d6
SS
909
910 /*
911 * Unlink the proc from its process group so that
912 * the following operations won't lead to an
913 * inconsistent state for processes running down
914 * the zombie list.
915 */
3c8687d6
SS
916 proc_remove_zombie(p);
917 leavepgrp(p);
918
984263bc 919 p->p_xstat = 0;
fde7ac71 920 ruadd(&q->p_cru, &p->p_ru);
984263bc
MD
921
922 /*
923 * Decrement the count of procs running with this uid.
924 */
9697c509 925 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
984263bc
MD
926
927 /*
928 * Free up credentials.
929 */
41c20dac
MD
930 crfree(p->p_ucred);
931 p->p_ucred = NULL;
984263bc
MD
932
933 /*
934 * Remove unused arguments
935 */
e955543c
MD
936 pa = p->p_args;
937 p->p_args = NULL;
e955543c
MD
938 if (pa && refcount_release(&pa->ar_ref)) {
939 kfree(pa, M_PARGS);
940 pa = NULL;
941 }
6fa9e71a
MD
942
943 ps = p->p_sigacts;
944 p->p_sigacts = NULL;
945 if (ps && refcount_release(&ps->ps_refcnt)) {
946 kfree(ps, M_SUBPROC);
947 ps = NULL;
984263bc
MD
948 }
949
49aa3df0 950 /*
e22b17d7
MD
951 * Our exitingcount was incremented when the process
952 * became a zombie, now that the process has been
953 * removed from (almost) all lists we should be able
954 * to safely destroy its vmspace. Wait for any current
955 * holders to go away (so the vmspace remains stable),
956 * then scrap it.
49aa3df0
YT
957 */
958 while (p->p_lock)
959 tsleep(p, 0, "reap4", hz);
e22b17d7
MD
960 vmspace_exitfree(p);
961 while (p->p_lock)
962 tsleep(p, 0, "reap5", hz);
963
fb2a331e 964 kfree(p, M_PROC);
8c2bce60 965 atomic_add_int(&nprocs, -1);
3919ced0
MD
966 error = 0;
967 goto done;
984263bc 968 }
4643740a
MD
969 if (p->p_stat == SSTOP && (p->p_flags & P_WAITED) == 0 &&
970 ((p->p_flags & P_TRACED) || (options & WUNTRACED))) {
8c986a82 971 lwkt_gettoken(&p->p_token);
4643740a 972 p->p_flags |= P_WAITED;
9697c509
DRJ
973
974 *res = p->p_pid;
52cac9fb 975 p->p_usched->heuristic_exiting(td->td_lwp, p);
9697c509 976 if (status)
a89c93ee 977 *status = W_STOPCODE(p->p_xstat);
9697c509
DRJ
978 /* Zero rusage so we get something consistent. */
979 if (rusage)
980 bzero(rusage, sizeof(rusage));
3919ced0 981 error = 0;
8c986a82 982 lwkt_reltoken(&p->p_token);
3919ced0 983 goto done;
984263bc 984 }
4643740a 985 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) {
8c986a82 986 lwkt_gettoken(&p->p_token);
cc162105 987 *res = p->p_pid;
52cac9fb 988 p->p_usched->heuristic_exiting(td->td_lwp, p);
4643740a 989 p->p_flags &= ~P_CONTINUED;
cc162105
PA
990
991 if (status)
992 *status = SIGCONT;
3919ced0 993 error = 0;
8c986a82 994 lwkt_reltoken(&p->p_token);
3919ced0 995 goto done;
cc162105 996 }
984263bc 997 }
3919ced0
MD
998 if (nfound == 0) {
999 error = ECHILD;
1000 goto done;
1001 }
9697c509
DRJ
1002 if (options & WNOHANG) {
1003 *res = 0;
3919ced0
MD
1004 error = 0;
1005 goto done;
984263bc 1006 }
58c2553a
MD
1007
1008 /*
1009 * Wait for signal - interlocked using q->p_token.
1010 */
1011 error = tsleep(q, PCATCH, "wait", 0);
3919ced0
MD
1012 if (error) {
1013done:
b5c4d81f 1014 lwkt_reltoken(&q->p_token);
984263bc 1015 return (error);
3919ced0 1016 }
984263bc
MD
1017 goto loop;
1018}
1019
1020/*
b5c4d81f 1021 * Make process 'parent' the new parent of process 'child'.
7e90d791
MD
1022 *
1023 * p_children/p_sibling requires the parent's token, and
1024 * changing pptr requires the child's token, so we have to
1025 * get three tokens to do this operation.
984263bc
MD
1026 */
1027void
77153250 1028proc_reparent(struct proc *child, struct proc *parent)
984263bc 1029{
7e90d791
MD
1030 struct proc *opp = child->p_pptr;
1031
1032 if (opp == parent)
984263bc 1033 return;
7e90d791 1034 PHOLD(opp);
b5c4d81f 1035 PHOLD(parent);
7e90d791 1036 lwkt_gettoken(&opp->p_token);
b5c4d81f
MD
1037 lwkt_gettoken(&child->p_token);
1038 lwkt_gettoken(&parent->p_token);
7e90d791 1039 KKASSERT(child->p_pptr == opp);
984263bc
MD
1040 LIST_REMOVE(child, p_sibling);
1041 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
1042 child->p_pptr = parent;
b5c4d81f
MD
1043 lwkt_reltoken(&parent->p_token);
1044 lwkt_reltoken(&child->p_token);
7e90d791 1045 lwkt_reltoken(&opp->p_token);
b5c4d81f 1046 PRELE(parent);
7e90d791 1047 PRELE(opp);
984263bc
MD
1048}
1049
1050/*
1051 * The next two functions are to handle adding/deleting items on the
1052 * exit callout list
1053 *
1054 * at_exit():
1055 * Take the arguments given and put them onto the exit callout list,
1056 * However first make sure that it's not already there.
1057 * returns 0 on success.
1058 */
1059
1060int
77153250 1061at_exit(exitlist_fn function)
984263bc
MD
1062{
1063 struct exitlist *ep;
1064
1065#ifdef INVARIANTS
1066 /* Be noisy if the programmer has lost track of things */
1067 if (rm_at_exit(function))
6ea70f76 1068 kprintf("WARNING: exit callout entry (%p) already present\n",
984263bc
MD
1069 function);
1070#endif
efda3bd0 1071 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT);
984263bc
MD
1072 if (ep == NULL)
1073 return (ENOMEM);
1074 ep->function = function;
1075 TAILQ_INSERT_TAIL(&exit_list, ep, next);
1076 return (0);
1077}
1078
1079/*
1080 * Scan the exit callout list for the given item and remove it.
1081 * Returns the number of items removed (0 or 1)
1082 */
1083int
77153250 1084rm_at_exit(exitlist_fn function)
984263bc
MD
1085{
1086 struct exitlist *ep;
1087
1088 TAILQ_FOREACH(ep, &exit_list, next) {
1089 if (ep->function == function) {
1090 TAILQ_REMOVE(&exit_list, ep, next);
efda3bd0 1091 kfree(ep, M_ATEXIT);
984263bc
MD
1092 return(1);
1093 }
1094 }
1095 return (0);
1096}
1097
70d3d461
SS
1098/*
1099 * LWP reaper related code.
1100 */
70d3d461
SS
1101static void
1102reaplwps(void *context, int dummy)
1103{
1104 struct lwplist *lwplist = context;
1105 struct lwp *lp;
1106
b5c4d81f 1107 lwkt_gettoken(&deadlwp_token);
70d3d461 1108 while ((lp = LIST_FIRST(lwplist))) {
3e291793
MD
1109 LIST_REMOVE(lp, u.lwp_reap_entry);
1110 reaplwp(lp);
70d3d461 1111 }
b5c4d81f 1112 lwkt_reltoken(&deadlwp_token);
70d3d461
SS
1113}
1114
1115static void
3e291793
MD
1116reaplwp(struct lwp *lp)
1117{
8c2bce60
MD
1118 if (lwp_wait(lp) == 0) {
1119 tsleep_interlock(lp, 0);
1120 while (lwp_wait(lp) == 0)
1121 tsleep(lp, PINTERLOCKED, "lwpreap", 1);
1122 }
3e291793
MD
1123 lwp_dispose(lp);
1124}
1125
1126static void
70d3d461
SS
1127deadlwp_init(void)
1128{
1129 int cpu;
1130
1131 for (cpu = 0; cpu < ncpus; cpu++) {
1132 LIST_INIT(&deadlwp_list[cpu]);
1133 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK);
1134 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]);
1135 }
1136}
1137
1138SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL);