kern_exit.c revision 1.132 1 1.132 jdolecek /* $NetBSD: kern_exit.c,v 1.132 2004/01/03 19:43:55 jdolecek Exp $ */
2 1.56 thorpej
3 1.56 thorpej /*-
4 1.72 thorpej * Copyright (c) 1998, 1999 The NetBSD Foundation, Inc.
5 1.56 thorpej * All rights reserved.
6 1.58 christos *
7 1.56 thorpej * This code is derived from software contributed to The NetBSD Foundation
8 1.56 thorpej * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9 1.58 christos * NASA Ames Research Center.
10 1.56 thorpej *
11 1.56 thorpej * Redistribution and use in source and binary forms, with or without
12 1.56 thorpej * modification, are permitted provided that the following conditions
13 1.56 thorpej * are met:
14 1.56 thorpej * 1. Redistributions of source code must retain the above copyright
15 1.56 thorpej * notice, this list of conditions and the following disclaimer.
16 1.58 christos * 2. Redistributions in binary form must reproduce the above copyright
17 1.56 thorpej * notice, this list of conditions and the following disclaimer in the
18 1.56 thorpej * documentation and/or other materials provided with the distribution.
19 1.56 thorpej * 3. All advertising materials mentioning features or use of this software
20 1.56 thorpej * must display the following acknowledgement:
21 1.58 christos * This product includes software developed by the NetBSD
22 1.58 christos * Foundation, Inc. and its contributors.
23 1.56 thorpej * 4. Neither the name of The NetBSD Foundation nor the names of its
24 1.56 thorpej * contributors may be used to endorse or promote products derived
25 1.56 thorpej * from this software without specific prior written permission.
26 1.58 christos *
27 1.56 thorpej * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28 1.56 thorpej * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29 1.56 thorpej * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30 1.56 thorpej * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31 1.56 thorpej * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 1.56 thorpej * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 1.56 thorpej * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 1.56 thorpej * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 1.56 thorpej * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 1.56 thorpej * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 1.56 thorpej * POSSIBILITY OF SUCH DAMAGE.
38 1.56 thorpej */
39 1.24 cgd
40 1.24 cgd /*
41 1.24 cgd * Copyright (c) 1982, 1986, 1989, 1991, 1993
42 1.24 cgd * The Regents of the University of California. All rights reserved.
43 1.24 cgd * (c) UNIX System Laboratories, Inc.
44 1.24 cgd * All or some portions of this file are derived from material licensed
45 1.24 cgd * to the University of California by American Telephone and Telegraph
46 1.24 cgd * Co. or Unix System Laboratories, Inc. and are reproduced herein with
47 1.24 cgd * the permission of UNIX System Laboratories, Inc.
48 1.24 cgd *
49 1.24 cgd * Redistribution and use in source and binary forms, with or without
50 1.24 cgd * modification, are permitted provided that the following conditions
51 1.24 cgd * are met:
52 1.24 cgd * 1. Redistributions of source code must retain the above copyright
53 1.24 cgd * notice, this list of conditions and the following disclaimer.
54 1.24 cgd * 2. Redistributions in binary form must reproduce the above copyright
55 1.24 cgd * notice, this list of conditions and the following disclaimer in the
56 1.24 cgd * documentation and/or other materials provided with the distribution.
57 1.119 agc * 3. Neither the name of the University nor the names of its contributors
58 1.24 cgd * may be used to endorse or promote products derived from this software
59 1.24 cgd * without specific prior written permission.
60 1.24 cgd *
61 1.24 cgd * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
62 1.24 cgd * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
63 1.24 cgd * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
64 1.24 cgd * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
65 1.24 cgd * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
66 1.24 cgd * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
67 1.24 cgd * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
68 1.24 cgd * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
69 1.24 cgd * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
70 1.24 cgd * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71 1.24 cgd * SUCH DAMAGE.
72 1.24 cgd *
73 1.49 fvdl * @(#)kern_exit.c 8.10 (Berkeley) 2/23/95
74 1.24 cgd */
75 1.92 lukem
76 1.92 lukem #include <sys/cdefs.h>
77 1.132 jdolecek __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.132 2004/01/03 19:43:55 jdolecek Exp $");
78 1.48 mrg
79 1.51 thorpej #include "opt_ktrace.h"
80 1.97 briggs #include "opt_perfctrs.h"
81 1.94 christos #include "opt_systrace.h"
82 1.60 tron #include "opt_sysv.h"
83 1.24 cgd
84 1.24 cgd #include <sys/param.h>
85 1.24 cgd #include <sys/systm.h>
86 1.24 cgd #include <sys/ioctl.h>
87 1.24 cgd #include <sys/proc.h>
88 1.24 cgd #include <sys/tty.h>
89 1.24 cgd #include <sys/time.h>
90 1.24 cgd #include <sys/resource.h>
91 1.24 cgd #include <sys/kernel.h>
92 1.50 christos #include <sys/ktrace.h>
93 1.24 cgd #include <sys/proc.h>
94 1.24 cgd #include <sys/buf.h>
95 1.24 cgd #include <sys/wait.h>
96 1.24 cgd #include <sys/file.h>
97 1.24 cgd #include <sys/vnode.h>
98 1.24 cgd #include <sys/syslog.h>
99 1.24 cgd #include <sys/malloc.h>
100 1.53 thorpej #include <sys/pool.h>
101 1.24 cgd #include <sys/resourcevar.h>
102 1.98 briggs #if defined(PERFCTRS)
103 1.97 briggs #include <sys/pmc.h>
104 1.98 briggs #endif
105 1.24 cgd #include <sys/ptrace.h>
106 1.29 cgd #include <sys/acct.h>
107 1.36 christos #include <sys/filedesc.h>
108 1.100 gmcgarry #include <sys/ras.h>
109 1.36 christos #include <sys/signalvar.h>
110 1.64 ross #include <sys/sched.h>
111 1.107 thorpej #include <sys/sa.h>
112 1.107 thorpej #include <sys/savar.h>
113 1.26 cgd #include <sys/mount.h>
114 1.26 cgd #include <sys/syscallargs.h>
115 1.94 christos #include <sys/systrace.h>
116 1.26 cgd
117 1.24 cgd #include <machine/cpu.h>
118 1.24 cgd
119 1.47 mrg #include <uvm/uvm_extern.h>
120 1.47 mrg
121 1.107 thorpej #define DEBUG_EXIT
122 1.107 thorpej
123 1.107 thorpej #ifdef DEBUG_EXIT
124 1.107 thorpej int debug_exit = 0;
125 1.107 thorpej #define DPRINTF(x) if (debug_exit) printf x
126 1.107 thorpej #else
127 1.107 thorpej #define DPRINTF(x)
128 1.107 thorpej #endif
129 1.107 thorpej
130 1.107 thorpej static void lwp_exit_hook(struct lwp *, void *);
131 1.123 christos static void exit_psignal(struct proc *, struct proc *);
132 1.123 christos
133 1.123 christos /*
134 1.132 jdolecek * Fill in the appropriate signal information, and signal the parent.
135 1.123 christos */
136 1.123 christos static void
137 1.123 christos exit_psignal(struct proc *p, struct proc *pp)
138 1.123 christos {
139 1.123 christos ksiginfo_t ksi;
140 1.123 christos
141 1.123 christos (void)memset(&ksi, 0, sizeof(ksi));
142 1.123 christos if ((ksi.ksi_signo = P_EXITSIG(p)) == SIGCHLD) {
143 1.123 christos if (WIFSIGNALED(p->p_xstat)) {
144 1.123 christos if (WCOREDUMP(p->p_xstat))
145 1.123 christos ksi.ksi_code = CLD_DUMPED;
146 1.123 christos else
147 1.123 christos ksi.ksi_code = CLD_KILLED;
148 1.123 christos } else {
149 1.123 christos ksi.ksi_code = CLD_EXITED;
150 1.123 christos }
151 1.123 christos }
152 1.123 christos /*
153 1.123 christos * we fill those in, even for non-SIGCHLD.
154 1.123 christos */
155 1.123 christos ksi.ksi_pid = p->p_pid;
156 1.123 christos ksi.ksi_uid = p->p_ucred->cr_uid;
157 1.123 christos ksi.ksi_status = p->p_xstat;
158 1.123 christos /* XXX: is this still valid? */
159 1.123 christos ksi.ksi_utime = p->p_ru->ru_utime.tv_sec;
160 1.123 christos ksi.ksi_stime = p->p_ru->ru_stime.tv_sec;
161 1.123 christos kpsignal(pp, &ksi, NULL);
162 1.123 christos }
163 1.94 christos
164 1.24 cgd /*
165 1.24 cgd * exit --
166 1.24 cgd * Death of process.
167 1.24 cgd */
168 1.31 thorpej int
169 1.107 thorpej sys_exit(struct lwp *l, void *v, register_t *retval)
170 1.31 thorpej {
171 1.33 mycroft struct sys_exit_args /* {
172 1.89 lukem syscallarg(int) rval;
173 1.31 thorpej } */ *uap = v;
174 1.24 cgd
175 1.107 thorpej /* Don't call exit1() multiple times in the same process.*/
176 1.107 thorpej if (l->l_proc->p_flag & P_WEXIT)
177 1.107 thorpej lwp_exit(l);
178 1.107 thorpej
179 1.107 thorpej exit1(l, W_EXITCODE(SCARG(uap, rval), 0));
180 1.24 cgd /* NOTREACHED */
181 1.31 thorpej return (0);
182 1.24 cgd }
183 1.24 cgd
184 1.24 cgd /*
185 1.24 cgd * Exit: deallocate address space and other resources, change proc state
186 1.24 cgd * to zombie, and unlink proc from allproc and parent's lists. Save exit
187 1.24 cgd * status and rusage for wait(). Check for child processes and orphan them.
188 1.24 cgd */
189 1.31 thorpej void
190 1.107 thorpej exit1(struct lwp *l, int rv)
191 1.24 cgd {
192 1.107 thorpej struct proc *p, *q, *nq;
193 1.108 nathanw int s, sa;
194 1.24 cgd
195 1.107 thorpej p = l->l_proc;
196 1.107 thorpej
197 1.78 thorpej if (__predict_false(p == initproc))
198 1.24 cgd panic("init died (signal %d, exit %d)",
199 1.24 cgd WTERMSIG(rv), WEXITSTATUS(rv));
200 1.73 thorpej
201 1.112 nathanw p->p_flag |= P_WEXIT;
202 1.130 atatat if (p->p_flag & P_STOPEXIT) {
203 1.130 atatat int s;
204 1.130 atatat
205 1.130 atatat sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
206 1.130 atatat SCHED_LOCK(s);
207 1.130 atatat p->p_stat = SSTOP;
208 1.130 atatat l->l_stat = LSSTOP;
209 1.130 atatat p->p_nrlwps--;
210 1.130 atatat mi_switch(l, NULL);
211 1.130 atatat SCHED_ASSERT_UNLOCKED();
212 1.130 atatat splx(s);
213 1.130 atatat }
214 1.112 nathanw
215 1.107 thorpej DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid));
216 1.107 thorpej /*
217 1.107 thorpej * Disable scheduler activation upcalls.
218 1.107 thorpej * We're trying to get out of here.
219 1.107 thorpej */
220 1.108 nathanw sa = 0;
221 1.110 nathanw if (p->p_sa != NULL) {
222 1.107 thorpej l->l_flag &= ~L_SA;
223 1.118 fvdl #if 0
224 1.107 thorpej p->p_flag &= ~P_SA;
225 1.118 fvdl #endif
226 1.108 nathanw sa = 1;
227 1.107 thorpej }
228 1.107 thorpej
229 1.24 cgd #ifdef PGINPROF
230 1.24 cgd vmsizmon();
231 1.24 cgd #endif
232 1.24 cgd if (p->p_flag & P_PROFIL)
233 1.24 cgd stopprofclock(p);
234 1.55 thorpej p->p_ru = pool_get(&rusage_pool, PR_WAITOK);
235 1.24 cgd /*
236 1.37 mycroft * If parent is waiting for us to exit or exec, P_PPWAIT is set; we
237 1.37 mycroft * wake up the parent early to avoid deadlock.
238 1.24 cgd */
239 1.37 mycroft if (p->p_flag & P_PPWAIT) {
240 1.37 mycroft p->p_flag &= ~P_PPWAIT;
241 1.126 dsl wakeup(p->p_pptr);
242 1.37 mycroft }
243 1.87 jdolecek sigfillset(&p->p_sigctx.ps_sigignore);
244 1.87 jdolecek sigemptyset(&p->p_sigctx.ps_siglist);
245 1.87 jdolecek p->p_sigctx.ps_sigcheck = 0;
246 1.107 thorpej timers_free(p, TIMERS_ALL);
247 1.107 thorpej
248 1.108 nathanw if (sa || (p->p_nlwps > 1))
249 1.107 thorpej exit_lwps(l);
250 1.100 gmcgarry
251 1.100 gmcgarry #if defined(__HAVE_RAS)
252 1.100 gmcgarry ras_purgeall(p);
253 1.100 gmcgarry #endif
254 1.24 cgd
255 1.24 cgd /*
256 1.24 cgd * Close open files and release open-file table.
257 1.24 cgd * This may block!
258 1.24 cgd */
259 1.117 fvdl fdfree(p);
260 1.66 thorpej cwdfree(p);
261 1.24 cgd
262 1.93 christos doexithooks(p);
263 1.93 christos
264 1.24 cgd if (SESS_LEADER(p)) {
265 1.76 augustss struct session *sp = p->p_session;
266 1.127 dsl struct tty *tp;
267 1.24 cgd
268 1.24 cgd if (sp->s_ttyvp) {
269 1.24 cgd /*
270 1.24 cgd * Controlling process.
271 1.24 cgd * Signal foreground pgrp,
272 1.24 cgd * drain controlling terminal
273 1.24 cgd * and revoke access to controlling terminal.
274 1.24 cgd */
275 1.127 dsl tp = sp->s_ttyp;
276 1.127 dsl TTY_LOCK(tp);
277 1.127 dsl if (tp->t_session == sp) {
278 1.127 dsl if (tp->t_pgrp)
279 1.127 dsl pgsignal(tp->t_pgrp, SIGHUP, 1);
280 1.127 dsl /* we can't guarantee the revoke will do this */
281 1.127 dsl tp->t_pgrp = NULL;
282 1.127 dsl tp->t_session = NULL;
283 1.127 dsl TTY_UNLOCK(tp);
284 1.127 dsl SESSRELE(sp);
285 1.127 dsl (void) ttywait(tp);
286 1.24 cgd /*
287 1.24 cgd * The tty could have been revoked
288 1.24 cgd * if we blocked.
289 1.24 cgd */
290 1.24 cgd if (sp->s_ttyvp)
291 1.49 fvdl VOP_REVOKE(sp->s_ttyvp, REVOKEALL);
292 1.127 dsl } else
293 1.127 dsl TTY_UNLOCK(tp);
294 1.24 cgd if (sp->s_ttyvp)
295 1.24 cgd vrele(sp->s_ttyvp);
296 1.24 cgd sp->s_ttyvp = NULL;
297 1.24 cgd /*
298 1.24 cgd * s_ttyp is not zero'd; we use this to indicate
299 1.24 cgd * that the session once had a controlling terminal.
300 1.24 cgd * (for logging and informational purposes)
301 1.24 cgd */
302 1.24 cgd }
303 1.24 cgd sp->s_leader = NULL;
304 1.24 cgd }
305 1.24 cgd fixjobc(p, p->p_pgrp, 0);
306 1.117 fvdl (void)acct_process(p);
307 1.24 cgd #ifdef KTRACE
308 1.24 cgd /*
309 1.24 cgd * release trace file
310 1.24 cgd */
311 1.50 christos ktrderef(p);
312 1.94 christos #endif
313 1.94 christos #ifdef SYSTRACE
314 1.94 christos systrace_sys_exit(p);
315 1.24 cgd #endif
316 1.24 cgd /*
317 1.99 manu * If emulation has process exit hook, call it now.
318 1.99 manu */
319 1.99 manu if (p->p_emul->e_proc_exit)
320 1.99 manu (*p->p_emul->e_proc_exit)(p);
321 1.99 manu
322 1.99 manu /*
323 1.128 dsl * Reset p_opptr pointer of all former children which got
324 1.128 dsl * traced by another process and were reparented. We reset
325 1.128 dsl * it to NULL here; the trace detach code then reparents
326 1.128 dsl * the child to initproc. We only check allproc list, since
327 1.128 dsl * eventual former children on zombproc list won't reference
328 1.128 dsl * p_opptr anymore.
329 1.128 dsl */
330 1.128 dsl s = proclist_lock_write();
331 1.128 dsl if (p->p_flag & P_CHTRACED) {
332 1.128 dsl LIST_FOREACH(q, &allproc, p_list) {
333 1.128 dsl if (q->p_opptr == p)
334 1.128 dsl q->p_opptr = NULL;
335 1.128 dsl }
336 1.128 dsl }
337 1.128 dsl
338 1.128 dsl /*
339 1.72 thorpej * Give orphaned children to init(8).
340 1.72 thorpej */
341 1.101 matt q = LIST_FIRST(&p->p_children);
342 1.132 jdolecek if (q) /* only need this if any child is SZOMB */
343 1.126 dsl wakeup(initproc);
344 1.132 jdolecek for (; q != NULL; q = nq) {
345 1.101 matt nq = LIST_NEXT(q, p_sibling);
346 1.104 jdolecek
347 1.24 cgd /*
348 1.104 jdolecek * Traced processes are killed since their existence
349 1.104 jdolecek * means someone is screwing up. Since we reset the
350 1.104 jdolecek * trace flags, the logic in sys_wait4() would not be
351 1.104 jdolecek * triggered to reparent the process to its
352 1.106 jdolecek * original parent, so we must do this here.
353 1.24 cgd */
354 1.24 cgd if (q->p_flag & P_TRACED) {
355 1.104 jdolecek if (q->p_opptr != q->p_pptr) {
356 1.104 jdolecek struct proc *t = q->p_opptr;
357 1.104 jdolecek proc_reparent(q, t ? t : initproc);
358 1.104 jdolecek q->p_opptr = NULL;
359 1.105 jdolecek } else
360 1.105 jdolecek proc_reparent(q, initproc);
361 1.45 mycroft q->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
362 1.24 cgd psignal(q, SIGKILL);
363 1.104 jdolecek } else {
364 1.104 jdolecek proc_reparent(q, initproc);
365 1.24 cgd }
366 1.24 cgd }
367 1.128 dsl proclist_unlock_write(s);
368 1.103 jdolecek
369 1.103 jdolecek /*
370 1.115 dsl * Save exit status and final rusage info, adding in child rusage
371 1.115 dsl * info and self times.
372 1.115 dsl * In order to pick up the time for the current execution, we must
373 1.115 dsl * do this before unlinking the lwp from l_list.
374 1.115 dsl */
375 1.115 dsl p->p_xstat = rv;
376 1.115 dsl *p->p_ru = p->p_stats->p_ru;
377 1.115 dsl calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL);
378 1.115 dsl ruadd(p->p_ru, &p->p_stats->p_cru);
379 1.115 dsl
380 1.115 dsl /*
381 1.115 dsl * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP!
382 1.115 dsl */
383 1.115 dsl
384 1.115 dsl /*
385 1.115 dsl * Move proc from allproc to zombproc, but do not yet
386 1.115 dsl * wake up the reaper. We will put the proc on the
387 1.115 dsl * deadproc list later (using the p_dead member), and
388 1.115 dsl * wake up the reaper when we do.
389 1.115 dsl * Changing the state to SDEAD stops it being found by pfind().
390 1.115 dsl */
391 1.115 dsl s = proclist_lock_write();
392 1.115 dsl p->p_stat = SDEAD;
393 1.115 dsl p->p_nrlwps--;
394 1.121 yamt l->l_stat = LSDEAD;
395 1.115 dsl LIST_REMOVE(p, p_list);
396 1.115 dsl LIST_INSERT_HEAD(&zombproc, p, p_list);
397 1.115 dsl LIST_REMOVE(l, l_list);
398 1.115 dsl l->l_flag |= L_DETACHED;
399 1.115 dsl proclist_unlock_write(s);
400 1.115 dsl
401 1.115 dsl /*
402 1.103 jdolecek * Notify interested parties of our demise.
403 1.103 jdolecek */
404 1.103 jdolecek KNOTE(&p->p_klist, NOTE_EXIT);
405 1.97 briggs
406 1.97 briggs #if PERFCTRS
407 1.97 briggs /*
408 1.97 briggs * Save final PMC information in parent process & clean up.
409 1.97 briggs */
410 1.97 briggs if (PMC_ENABLED(p)) {
411 1.97 briggs pmc_save_context(p);
412 1.97 briggs pmc_accumulate(p->p_pptr, p);
413 1.97 briggs pmc_process_exit(p);
414 1.97 briggs }
415 1.97 briggs #endif
416 1.59 christos
417 1.59 christos /*
418 1.59 christos * Notify parent that we're gone. If parent has the P_NOCLDWAIT
419 1.59 christos * flag set, notify init instead (and hope it will handle
420 1.59 christos * this situation).
421 1.59 christos */
422 1.59 christos if (p->p_pptr->p_flag & P_NOCLDWAIT) {
423 1.59 christos struct proc *pp = p->p_pptr;
424 1.59 christos proc_reparent(p, initproc);
425 1.59 christos /*
426 1.59 christos * If this was the last child of our parent, notify
427 1.59 christos * parent, so in case he was wait(2)ing, he will
428 1.59 christos * continue.
429 1.59 christos */
430 1.101 matt if (LIST_FIRST(&pp->p_children) == NULL)
431 1.126 dsl wakeup(pp);
432 1.59 christos }
433 1.67 thorpej
434 1.67 thorpej /*
435 1.67 thorpej * Release the process's signal state.
436 1.67 thorpej */
437 1.67 thorpej sigactsfree(p);
438 1.96 manu
439 1.96 manu /*
440 1.107 thorpej * Clear curlwp after we've done all operations
441 1.24 cgd * that could block, and before tearing down the rest
442 1.24 cgd * of the process state that might be used from clock, etc.
443 1.107 thorpej * Also, can't clear curlwp while we're still runnable,
444 1.24 cgd * as we're not on a run queue (we are current, just not
445 1.24 cgd * a proper proc any longer!).
446 1.24 cgd *
447 1.24 cgd * Other substructures are freed from wait().
448 1.24 cgd */
449 1.107 thorpej curlwp = NULL;
450 1.74 bouyer limfree(p->p_limit);
451 1.107 thorpej pstatsfree(p->p_stats);
452 1.77 thorpej p->p_limit = NULL;
453 1.24 cgd
454 1.85 thorpej /* This process no longer needs to hold the kernel lock. */
455 1.107 thorpej KERNEL_PROC_UNLOCK(l);
456 1.85 thorpej
457 1.24 cgd /*
458 1.56 thorpej * Finally, call machine-dependent code to switch to a new
459 1.56 thorpej * context (possibly the idle context). Once we are no longer
460 1.56 thorpej * using the dead process's vmspace and stack, exit2() will be
461 1.56 thorpej * called to schedule those resources to be released by the
462 1.56 thorpej * reaper thread.
463 1.56 thorpej *
464 1.56 thorpej * Note that cpu_exit() will end with a call equivalent to
465 1.56 thorpej * cpu_switch(), finishing our execution (pun intended).
466 1.24 cgd */
467 1.107 thorpej
468 1.107 thorpej cpu_exit(l, 1);
469 1.107 thorpej }
470 1.107 thorpej
471 1.107 thorpej void
472 1.107 thorpej exit_lwps(struct lwp *l)
473 1.107 thorpej {
474 1.107 thorpej struct proc *p;
475 1.107 thorpej struct lwp *l2;
476 1.107 thorpej int s, error;
477 1.107 thorpej lwpid_t waited;
478 1.107 thorpej
479 1.107 thorpej p = l->l_proc;
480 1.107 thorpej
481 1.107 thorpej /* XXX SMP
482 1.107 thorpej * This would be the right place to IPI any LWPs running on
483 1.107 thorpej * other processors so that they can notice the userret exit hook.
484 1.107 thorpej */
485 1.107 thorpej p->p_userret = lwp_exit_hook;
486 1.107 thorpej p->p_userret_arg = NULL;
487 1.107 thorpej
488 1.131 cl if (p->p_sa) {
489 1.131 cl /*
490 1.131 cl * Make SA-cached LWPs normal process runnable LWPs so
491 1.131 cl * that they'll also self-destruct.
492 1.131 cl */
493 1.107 thorpej DPRINTF(("exit_lwps: Making cached LWPs of %d runnable: ",
494 1.107 thorpej p->p_pid));
495 1.107 thorpej SCHED_LOCK(s);
496 1.107 thorpej while ((l2 = sa_getcachelwp(p)) != 0) {
497 1.107 thorpej l2->l_priority = l2->l_usrpri;
498 1.107 thorpej setrunnable(l2);
499 1.107 thorpej DPRINTF(("%d ", l2->l_lid));
500 1.107 thorpej }
501 1.107 thorpej DPRINTF(("\n"));
502 1.107 thorpej SCHED_UNLOCK(s);
503 1.131 cl
504 1.131 cl /* Clear wokenq, the LWPs on the queue will run below */
505 1.131 cl p->p_sa->sa_wokenq_head = NULL;
506 1.107 thorpej }
507 1.107 thorpej
508 1.107 thorpej /*
509 1.107 thorpej * Interrupt LWPs in interruptable sleep, unsuspend suspended
510 1.111 skrll * LWPs, make detached LWPs undetached (so we can wait for
511 1.107 thorpej * them) and then wait for everyone else to finish.
512 1.107 thorpej */
513 1.107 thorpej LIST_FOREACH(l2, &p->p_lwps, l_sibling) {
514 1.107 thorpej l2->l_flag &= ~(L_DETACHED|L_SA);
515 1.128 dsl
516 1.131 cl if (p->p_sa && l2->l_wchan == p->p_sa)
517 1.131 cl l2->l_flag |= L_SINTR;
518 1.118 fvdl
519 1.107 thorpej if ((l2->l_stat == LSSLEEP && (l2->l_flag & L_SINTR)) ||
520 1.118 fvdl l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) {
521 1.107 thorpej SCHED_LOCK(s);
522 1.107 thorpej setrunnable(l2);
523 1.107 thorpej SCHED_UNLOCK(s);
524 1.107 thorpej DPRINTF(("exit_lwps: Made %d.%d runnable\n",
525 1.107 thorpej p->p_pid, l2->l_lid));
526 1.107 thorpej }
527 1.107 thorpej }
528 1.107 thorpej
529 1.107 thorpej
530 1.107 thorpej while (p->p_nlwps > 1) {
531 1.107 thorpej DPRINTF(("exit_lwps: waiting for %d LWPs (%d runnable, %d zombies)\n",
532 1.107 thorpej p->p_nlwps, p->p_nrlwps, p->p_nzlwps));
533 1.107 thorpej error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL);
534 1.107 thorpej if (error)
535 1.107 thorpej panic("exit_lwps: lwp_wait1 failed with error %d\n",
536 1.107 thorpej error);
537 1.107 thorpej DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited));
538 1.107 thorpej }
539 1.107 thorpej
540 1.107 thorpej p->p_userret = NULL;
541 1.107 thorpej }
542 1.107 thorpej
543 1.107 thorpej /* Wrapper function for use in p_userret */
544 1.107 thorpej static void
545 1.107 thorpej lwp_exit_hook(struct lwp *l, void *arg)
546 1.107 thorpej {
547 1.107 thorpej KERNEL_PROC_LOCK(l);
548 1.107 thorpej lwp_exit(l);
549 1.24 cgd }
550 1.24 cgd
551 1.56 thorpej /*
552 1.56 thorpej * We are called from cpu_exit() once it is safe to schedule the
553 1.85 thorpej * dead process's resources to be freed (i.e., once we've switched to
554 1.85 thorpej * the idle PCB for the current CPU).
555 1.72 thorpej *
556 1.72 thorpej * NOTE: One must be careful with locking in this routine. It's
557 1.72 thorpej * called from a critical section in machine-dependent code, so
558 1.72 thorpej * we should refrain from changing any interrupt state.
559 1.72 thorpej *
560 1.72 thorpej * We lock the deadproc list (a spin lock), place the proc on that
561 1.115 dsl * list (using the p_dead member), and wake up the reaper.
562 1.56 thorpej */
563 1.56 thorpej void
564 1.107 thorpej exit2(struct lwp *l)
565 1.56 thorpej {
566 1.107 thorpej struct proc *p = l->l_proc;
567 1.56 thorpej
568 1.72 thorpej simple_lock(&deadproc_slock);
569 1.115 dsl SLIST_INSERT_HEAD(&deadprocs, p, p_dead);
570 1.72 thorpej simple_unlock(&deadproc_slock);
571 1.72 thorpej
572 1.107 thorpej /* lwp_exit2() will wake up deadproc for us. */
573 1.107 thorpej lwp_exit2(l);
574 1.56 thorpej }
575 1.56 thorpej
576 1.56 thorpej /*
577 1.56 thorpej * Process reaper. This is run by a kernel thread to free the resources
578 1.56 thorpej * of a dead process. Once the resources are free, the process becomes
579 1.56 thorpej * a zombie, and the parent is allowed to read the undead's status.
580 1.56 thorpej */
581 1.56 thorpej void
582 1.84 thorpej reaper(void *arg)
583 1.56 thorpej {
584 1.128 dsl struct proc *p, *parent;
585 1.107 thorpej struct lwp *l;
586 1.56 thorpej
587 1.107 thorpej KERNEL_PROC_UNLOCK(curlwp);
588 1.90 fvdl
589 1.56 thorpej for (;;) {
590 1.72 thorpej simple_lock(&deadproc_slock);
591 1.115 dsl p = SLIST_FIRST(&deadprocs);
592 1.107 thorpej l = LIST_FIRST(&deadlwp);
593 1.107 thorpej if (p == NULL && l == NULL) {
594 1.56 thorpej /* No work for us; go to sleep until someone exits. */
595 1.115 dsl (void) ltsleep(&deadprocs, PVM|PNORELOCK,
596 1.83 thorpej "reaper", 0, &deadproc_slock);
597 1.56 thorpej continue;
598 1.56 thorpej }
599 1.56 thorpej
600 1.107 thorpej if (l != NULL ) {
601 1.107 thorpej p = l->l_proc;
602 1.56 thorpej
603 1.115 dsl /* Remove lwp from the deadlwp list. */
604 1.107 thorpej LIST_REMOVE(l, l_list);
605 1.107 thorpej simple_unlock(&deadproc_slock);
606 1.107 thorpej KERNEL_PROC_LOCK(curlwp);
607 1.107 thorpej
608 1.107 thorpej /*
609 1.107 thorpej * Give machine-dependent code a chance to free any
610 1.107 thorpej * resources it couldn't free while still running on
611 1.107 thorpej * that process's context. This must be done before
612 1.107 thorpej * uvm_lwp_exit(), in case these resources are in the
613 1.107 thorpej * PCB.
614 1.107 thorpej */
615 1.107 thorpej cpu_wait(l);
616 1.71 thorpej
617 1.107 thorpej /*
618 1.107 thorpej * Free the VM resources we're still holding on to.
619 1.107 thorpej */
620 1.107 thorpej uvm_lwp_exit(l);
621 1.56 thorpej
622 1.107 thorpej l->l_stat = LSZOMB;
623 1.107 thorpej if (l->l_flag & L_DETACHED) {
624 1.107 thorpej /* Nobody waits for detached LWPs. */
625 1.107 thorpej LIST_REMOVE(l, l_sibling);
626 1.107 thorpej p->p_nlwps--;
627 1.107 thorpej pool_put(&lwp_pool, l);
628 1.107 thorpej } else {
629 1.107 thorpej p->p_nzlwps++;
630 1.126 dsl wakeup(&p->p_nlwps);
631 1.107 thorpej }
632 1.107 thorpej /* XXXNJW where should this be with respect to
633 1.107 thorpej * the wakeup() above? */
634 1.107 thorpej KERNEL_PROC_UNLOCK(curlwp);
635 1.107 thorpej } else {
636 1.115 dsl /* Remove proc from the deadproc list. */
637 1.115 dsl SLIST_REMOVE_HEAD(&deadprocs, p_dead);
638 1.107 thorpej simple_unlock(&deadproc_slock);
639 1.107 thorpej KERNEL_PROC_LOCK(curlwp);
640 1.56 thorpej
641 1.107 thorpej /*
642 1.107 thorpej * Free the VM resources we're still holding on to.
643 1.107 thorpej * We must do this from a valid thread because doing
644 1.107 thorpej * so may block.
645 1.107 thorpej */
646 1.107 thorpej uvm_proc_exit(p);
647 1.107 thorpej
648 1.107 thorpej /* Process is now a true zombie. */
649 1.107 thorpej p->p_stat = SZOMB;
650 1.128 dsl parent = p->p_pptr;
651 1.128 dsl parent->p_nstopchild++;
652 1.128 dsl if (LIST_FIRST(&parent->p_children) != p) {
653 1.128 dsl /* Put child where it can be found quickly */
654 1.128 dsl LIST_REMOVE(p, p_sibling);
655 1.128 dsl LIST_INSERT_HEAD(&parent->p_children,
656 1.128 dsl p, p_sibling);
657 1.128 dsl }
658 1.107 thorpej
659 1.107 thorpej /* Wake up the parent so it can get exit status. */
660 1.107 thorpej if ((p->p_flag & P_FSTRACE) == 0 && p->p_exitsig != 0)
661 1.123 christos exit_psignal(p, p->p_pptr);
662 1.107 thorpej KERNEL_PROC_UNLOCK(curlwp);
663 1.126 dsl wakeup(p->p_pptr);
664 1.107 thorpej }
665 1.56 thorpej }
666 1.56 thorpej }
667 1.56 thorpej
668 1.24 cgd int
669 1.107 thorpej sys_wait4(struct lwp *l, void *v, register_t *retval)
670 1.31 thorpej {
671 1.76 augustss struct sys_wait4_args /* {
672 1.89 lukem syscallarg(int) pid;
673 1.89 lukem syscallarg(int *) status;
674 1.89 lukem syscallarg(int) options;
675 1.89 lukem syscallarg(struct rusage *) rusage;
676 1.31 thorpej } */ *uap = v;
677 1.109 dsl struct proc *child, *parent;
678 1.109 dsl int status, error;
679 1.24 cgd
680 1.109 dsl parent = l->l_proc;
681 1.107 thorpej
682 1.26 cgd if (SCARG(uap, pid) == 0)
683 1.109 dsl SCARG(uap, pid) = -parent->p_pgid;
684 1.109 dsl if (SCARG(uap, options) & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG))
685 1.24 cgd return (EINVAL);
686 1.41 mikel
687 1.109 dsl error = find_stopped_child(parent, SCARG(uap,pid), SCARG(uap,options),
688 1.109 dsl &child);
689 1.109 dsl if (error != 0)
690 1.109 dsl return error;
691 1.109 dsl if (child == NULL) {
692 1.109 dsl *retval = 0;
693 1.109 dsl return 0;
694 1.109 dsl }
695 1.109 dsl
696 1.109 dsl retval[0] = child->p_pid;
697 1.109 dsl
698 1.109 dsl if (child->p_stat == SZOMB) {
699 1.109 dsl if (SCARG(uap, status)) {
700 1.109 dsl status = child->p_xstat; /* convert to int */
701 1.126 dsl error = copyout(&status, SCARG(uap, status),
702 1.109 dsl sizeof(status));
703 1.109 dsl if (error)
704 1.109 dsl return (error);
705 1.109 dsl }
706 1.109 dsl if (SCARG(uap, rusage)) {
707 1.126 dsl error = copyout(child->p_ru, SCARG(uap, rusage),
708 1.109 dsl sizeof(struct rusage));
709 1.109 dsl if (error)
710 1.109 dsl return (error);
711 1.109 dsl }
712 1.109 dsl
713 1.109 dsl proc_free(child);
714 1.109 dsl return 0;
715 1.109 dsl }
716 1.109 dsl
717 1.109 dsl /* child state must be SSTOP */
718 1.109 dsl if (SCARG(uap, status)) {
719 1.109 dsl status = W_STOPCODE(child->p_xstat);
720 1.128 dsl return copyout(&status, SCARG(uap, status), sizeof(status));
721 1.109 dsl }
722 1.109 dsl return 0;
723 1.109 dsl }
724 1.109 dsl
725 1.109 dsl /*
726 1.109 dsl * Scan list of child processes for a child process that has stopped or
727 1.109 dsl * exited. Used by sys_wait4 and 'compat' equivalents.
728 1.109 dsl */
729 1.109 dsl int
730 1.109 dsl find_stopped_child(struct proc *parent, pid_t pid, int options,
731 1.109 dsl struct proc **child_p)
732 1.109 dsl {
733 1.109 dsl struct proc *child;
734 1.128 dsl int error;
735 1.68 thorpej
736 1.120 yamt for (;;) {
737 1.128 dsl proclist_lock_read();
738 1.128 dsl error = ECHILD;
739 1.109 dsl LIST_FOREACH(child, &parent->p_children, p_sibling) {
740 1.128 dsl if (pid >= 0) {
741 1.128 dsl if (child->p_pid != pid) {
742 1.128 dsl child = p_find(pid, PFIND_ZOMBIE |
743 1.128 dsl PFIND_LOCKED);
744 1.128 dsl if (child == NULL
745 1.128 dsl || child->p_pptr != parent) {
746 1.128 dsl child = NULL;
747 1.128 dsl break;
748 1.128 dsl }
749 1.128 dsl }
750 1.128 dsl } else
751 1.128 dsl if (pid != WAIT_ANY && child->p_pgid != -pid)
752 1.128 dsl /* child not in correct pgrp */
753 1.128 dsl continue;
754 1.109 dsl /*
755 1.109 dsl * Wait for processes with p_exitsig != SIGCHLD
756 1.109 dsl * processes only if WALTSIG is set; wait for
757 1.109 dsl * processes with p_exitsig == SIGCHLD only
758 1.109 dsl * if WALTSIG is clear.
759 1.109 dsl */
760 1.109 dsl if (((options & WALLSIG) == 0) &&
761 1.109 dsl (options & WALTSIG ? child->p_exitsig == SIGCHLD
762 1.128 dsl : P_EXITSIG(child) != SIGCHLD)){
763 1.128 dsl if (child->p_pid == pid) {
764 1.128 dsl child = NULL;
765 1.128 dsl break;
766 1.128 dsl }
767 1.109 dsl continue;
768 1.128 dsl }
769 1.109 dsl
770 1.128 dsl error = 0;
771 1.128 dsl if (child->p_stat == SZOMB && !(options & WNOZOMBIE))
772 1.128 dsl break;
773 1.109 dsl
774 1.109 dsl if (child->p_stat == SSTOP &&
775 1.109 dsl (child->p_flag & P_WAITED) == 0 &&
776 1.109 dsl (child->p_flag & P_TRACED || options & WUNTRACED)) {
777 1.128 dsl if ((options & WNOWAIT) == 0) {
778 1.109 dsl child->p_flag |= P_WAITED;
779 1.128 dsl parent->p_nstopchild--;
780 1.128 dsl }
781 1.128 dsl break;
782 1.128 dsl }
783 1.128 dsl if (parent->p_nstopchild == 0 || child->p_pid == pid) {
784 1.128 dsl child = NULL;
785 1.128 dsl break;
786 1.24 cgd }
787 1.109 dsl }
788 1.128 dsl proclist_unlock_read();
789 1.128 dsl if (child != NULL || error != 0 || options & WNOHANG) {
790 1.128 dsl *child_p = child;
791 1.128 dsl return error;
792 1.109 dsl }
793 1.126 dsl error = tsleep(parent, PWAIT | PCATCH, "wait", 0);
794 1.109 dsl if (error != 0)
795 1.109 dsl return error;
796 1.109 dsl }
797 1.109 dsl }
798 1.109 dsl
799 1.109 dsl /*
800 1.109 dsl * Free a process after parent has taken all the state info.
801 1.109 dsl */
802 1.109 dsl void
803 1.109 dsl proc_free(struct proc *p)
804 1.109 dsl {
805 1.109 dsl struct proc *parent = p->p_pptr;
806 1.109 dsl int s;
807 1.24 cgd
808 1.109 dsl /*
809 1.109 dsl * If we got the child via ptrace(2) or procfs, and
810 1.109 dsl * the parent is different (meaning the process was
811 1.109 dsl * attached, rather than run as a child), then we need
812 1.109 dsl * to give it back to the old parent, and send the
813 1.109 dsl * parent the exit signal. The rest of the cleanup
814 1.109 dsl * will be done when the old parent waits on the child.
815 1.109 dsl */
816 1.109 dsl if ((p->p_flag & P_TRACED) && p->p_opptr != parent){
817 1.109 dsl parent = p->p_opptr;
818 1.109 dsl if (parent == NULL)
819 1.109 dsl parent = initproc;
820 1.109 dsl proc_reparent(p, parent);
821 1.109 dsl p->p_opptr = NULL;
822 1.109 dsl p->p_flag &= ~(P_TRACED|P_WAITED|P_FSTRACE);
823 1.109 dsl if (p->p_exitsig != 0)
824 1.123 christos exit_psignal(p, parent);
825 1.126 dsl wakeup(parent);
826 1.109 dsl return;
827 1.109 dsl }
828 1.109 dsl
829 1.109 dsl scheduler_wait_hook(parent, p);
830 1.109 dsl p->p_xstat = 0;
831 1.115 dsl
832 1.109 dsl ruadd(&parent->p_stats->p_cru, p->p_ru);
833 1.115 dsl
834 1.115 dsl /*
835 1.115 dsl * At this point we are going to start freeing the final resources.
836 1.115 dsl * If anyone tries to access the proc structure after here they
837 1.115 dsl * will get a shock - bits are missing.
838 1.115 dsl * Attempt to make it hard!
839 1.115 dsl */
840 1.115 dsl
841 1.115 dsl p->p_stat = SIDL; /* not even a zombie any more */
842 1.115 dsl
843 1.109 dsl pool_put(&rusage_pool, p->p_ru);
844 1.56 thorpej
845 1.109 dsl /*
846 1.109 dsl * Finally finished with old proc entry.
847 1.109 dsl * Unlink it from its process group and free it.
848 1.109 dsl */
849 1.109 dsl leavepgrp(p);
850 1.56 thorpej
851 1.109 dsl s = proclist_lock_write();
852 1.109 dsl LIST_REMOVE(p, p_list); /* off zombproc */
853 1.128 dsl p->p_pptr->p_nstopchild--;
854 1.115 dsl LIST_REMOVE(p, p_sibling);
855 1.109 dsl proclist_unlock_write(s);
856 1.35 mycroft
857 1.109 dsl /*
858 1.109 dsl * Decrement the count of procs running with this uid.
859 1.109 dsl */
860 1.109 dsl (void)chgproccnt(p->p_cred->p_ruid, -1);
861 1.24 cgd
862 1.109 dsl /*
863 1.109 dsl * Free up credentials.
864 1.109 dsl */
865 1.109 dsl if (--p->p_cred->p_refcnt == 0) {
866 1.109 dsl crfree(p->p_cred->pc_ucred);
867 1.109 dsl pool_put(&pcred_pool, p->p_cred);
868 1.109 dsl }
869 1.107 thorpej
870 1.109 dsl /*
871 1.109 dsl * Release reference to text vnode
872 1.109 dsl */
873 1.109 dsl if (p->p_textvp)
874 1.109 dsl vrele(p->p_textvp);
875 1.24 cgd
876 1.129 cl /* Release any SA state. */
877 1.129 cl if (p->p_sa)
878 1.129 cl sa_release(p);
879 1.109 dsl
880 1.115 dsl /* Free proc structure and let pid be reallocated */
881 1.115 dsl proc_free_mem(p);
882 1.24 cgd }
883 1.24 cgd
884 1.24 cgd /*
885 1.24 cgd * make process 'parent' the new parent of process 'child'.
886 1.128 dsl *
887 1.128 dsl * Must be called with proclist_lock_write() held.
888 1.24 cgd */
889 1.24 cgd void
890 1.82 thorpej proc_reparent(struct proc *child, struct proc *parent)
891 1.24 cgd {
892 1.24 cgd
893 1.24 cgd if (child->p_pptr == parent)
894 1.24 cgd return;
895 1.70 thorpej
896 1.128 dsl if (child->p_stat == SZOMB
897 1.128 dsl || (child->p_stat == SSTOP && !(child->p_flag & P_WAITED))) {
898 1.128 dsl child->p_pptr->p_nstopchild--;
899 1.128 dsl parent->p_nstopchild++;
900 1.128 dsl }
901 1.70 thorpej if (parent == initproc)
902 1.70 thorpej child->p_exitsig = SIGCHLD;
903 1.24 cgd
904 1.25 mycroft LIST_REMOVE(child, p_sibling);
905 1.25 mycroft LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
906 1.24 cgd child->p_pptr = parent;
907 1.24 cgd }
908