sys/kern/kern_exit.c

   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  * 4. Neither the name of the University nor the names of its contributors
  19  *    may be used to endorse or promote products derived from this software
  20  *    without specific prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  *
  34  *      @(#)kern_exit.c 8.7 (Berkeley) 2/12/94
  35  */
  36
  37 #include <sys/cdefs.h>
  38 __FBSDID("$FreeBSD$");
  39
  40 #include "opt_compat.h"
  41 #include "opt_ktrace.h"
  42 #include "opt_mac.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/sysproto.h>
  47 #include <sys/eventhandler.h>
  48 #include <sys/kernel.h>
  49 #include <sys/malloc.h>
  50 #include <sys/lock.h>
  51 #include <sys/mutex.h>
  52 #include <sys/proc.h>
  53 #include <sys/pioctl.h>
  54 #include <sys/tty.h>
  55 #include <sys/wait.h>
  56 #include <sys/vmmeter.h>
  57 #include <sys/vnode.h>
  58 #include <sys/resourcevar.h>
  59 #include <sys/signalvar.h>
  60 #include <sys/sched.h>
  61 #include <sys/sx.h>
  62 #include <sys/syscallsubr.h>
  63 #include <sys/ptrace.h>
  64 #include <sys/acct.h>           /* for acct_process() function prototype */
  65 #include <sys/filedesc.h>
  66 #include <sys/mac.h>
  67 #include <sys/shm.h>
  68 #include <sys/sem.h>
  69 #ifdef KTRACE
  70 #include <sys/ktrace.h>
  71 #endif
  72
  73 #include <vm/vm.h>
  74 #include <vm/vm_extern.h>
  75 #include <vm/vm_param.h>
  76 #include <vm/pmap.h>
  77 #include <vm/vm_map.h>
  78 #include <vm/vm_page.h>
  79 #include <vm/uma.h>
  80
  81 /* Required to be non-static for SysVR4 emulator */
  82 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status");
  83
  84 /*
  85  * exit --
  86  *      Death of process.
  87  *
  88  * MPSAFE
  89  */
  90 void
  91 sys_exit(struct thread *td, struct sys_exit_args *uap)
  92 {
  93
  94         exit1(td, W_EXITCODE(uap->rval, 0));
  95         /* NOTREACHED */
  96 }
  97
  98 /*
  99  * Exit: deallocate address space and other resources, change proc state
 100  * to zombie, and unlink proc from allproc and parent's lists.  Save exit
 101  * status and rusage for wait().  Check for child processes and orphan them.
 102  */
 103 void
 104 exit1(struct thread *td, int rv)
 105 {
 106         struct bintime new_switchtime;
 107         struct proc *p, *nq, *q;
 108         struct tty *tp;
 109         struct vnode *ttyvp;
 110         struct vmspace *vm;
 111         struct vnode *vtmp;
 112 #ifdef KTRACE
 113         struct vnode *tracevp;
 114         struct ucred *tracecred;
 115 #endif
 116         struct plimit *plim;
 117         int refcnt;
 118
 119         /*
 120          * Drop Giant if caller has it.  Eventually we should warn about
 121          * being called with Giant held.
 122          */
 123         while (mtx_owned(&Giant))
 124                 mtx_unlock(&Giant);
 125
 126         p = td->td_proc;
 127         if (p == initproc) {
 128                 printf("init died (signal %d, exit %d)\n",
 129                     WTERMSIG(rv), WEXITSTATUS(rv));
 130                 panic("Going nowhere without my init!");
 131         }
 132
 133         /*
 134          * MUST abort all other threads before proceeding past here.
 135          */
 136         PROC_LOCK(p);
 137         if (p->p_flag & P_HADTHREADS) {
 138 retry:
 139                 /*
 140                  * First check if some other thread got here before us..
 141                  * if so, act apropriatly, (exit or suspend);
 142                  */
 143                 thread_suspend_check(0);
 144
 145                 /*
 146                  * Kill off the other threads. This requires
 147                  * some co-operation from other parts of the kernel
 148                  * so it may not be instantaneous.  With this state set
 149                  * any thread entering the kernel from userspace will
 150                  * thread_exit() in trap().  Any thread attempting to
 151                  * sleep will return immediately with EINTR or EWOULDBLOCK
 152                  * which will hopefully force them to back out to userland
 153                  * freeing resources as they go.  Any thread attempting
 154                  * to return to userland will thread_exit() from userret().
 155                  * thread_exit() will unsuspend us when the last of the
 156                  * other threads exits.
 157                  * If there is already a thread singler after resumption,
 158                  * calling thread_single will fail; in that case, we just
 159                  * re-check all suspension request, the thread should
 160                  * either be suspended there or exit.
 161                  */
 162                 if (thread_single(SINGLE_EXIT))
 163                         goto retry;
 164
 165                 /*
 166                  * All other activity in this process is now stopped.
 167                  * Threading support has been turned off.
 168                  */
 169         }
 170
 171         p->p_flag |= P_WEXIT;
 172         PROC_UNLOCK(p);
 173
 174         /* Are we a task leader? */
 175         if (p == p->p_leader) {
 176                 mtx_lock(&ppeers_lock);
 177                 q = p->p_peers;
 178                 while (q != NULL) {
 179                         PROC_LOCK(q);
 180                         psignal(q, SIGKILL);
 181                         PROC_UNLOCK(q);
 182                         q = q->p_peers;
 183                 }
 184                 while (p->p_peers != NULL)
 185                         msleep(p, &ppeers_lock, PWAIT, "exit1", 0);
 186                 mtx_unlock(&ppeers_lock);
 187         }
 188
 189         PROC_LOCK(p);
 190         _STOPEVENT(p, S_EXIT, rv);
 191         wakeup(&p->p_stype);    /* Wakeup anyone in procfs' PIOCWAIT */
 192         PROC_UNLOCK(p);
 193
 194         /*
 195          * Check if any loadable modules need anything done at process exit.
 196          * E.g. SYSV IPC stuff
 197          * XXX what if one of these generates an error?
 198          */
 199         EVENTHANDLER_INVOKE(process_exit, p);
 200
 201         MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage),
 202                 M_ZOMBIE, M_WAITOK);
 203         /*
 204          * If parent is waiting for us to exit or exec,
 205          * P_PPWAIT is set; we will wakeup the parent below.
 206          */
 207         PROC_LOCK(p);
 208         stopprofclock(p);
 209         p->p_flag &= ~(P_TRACED | P_PPWAIT);
 210         SIGEMPTYSET(p->p_siglist);
 211         SIGEMPTYSET(td->td_siglist);
 212
 213         /*
 214          * Stop the real interval timer.  If the handler is currently
 215          * executing, prevent it from rearming itself and let it finish.
 216          */
 217         if (timevalisset(&p->p_realtimer.it_value) &&
 218             callout_stop(&p->p_itcallout) == 0) {
 219                 timevalclear(&p->p_realtimer.it_interval);
 220                 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0);
 221                 KASSERT(!timevalisset(&p->p_realtimer.it_value),
 222                     ("realtime timer is still armed"));
 223         }
 224         PROC_UNLOCK(p);
 225
 226         /*
 227          * Reset any sigio structures pointing to us as a result of
 228          * F_SETOWN with our pid.
 229          */
 230         mtx_lock(&Giant);       /* XXX: not sure if needed */
 231         funsetownlst(&p->p_sigiolst);
 232
 233         /*
 234          * Close open files and release open-file table.
 235          * This may block!
 236          */
 237         fdfree(td);
 238         mtx_unlock(&Giant);
 239
 240         /*
 241          * If this thread tickled GEOM, we need to wait for the giggling to
 242          * stop before we return to userland
 243          */
 244         if (td->td_pflags & TDP_GEOM)
 245                 g_waitidle();
 246
 247         /*
 248          * Remove ourself from our leader's peer list and wake our leader.
 249          */
 250         mtx_lock(&ppeers_lock);
 251         if (p->p_leader->p_peers) {
 252                 q = p->p_leader;
 253                 while (q->p_peers != p)
 254                         q = q->p_peers;
 255                 q->p_peers = p->p_peers;
 256                 wakeup(p->p_leader);
 257         }
 258         mtx_unlock(&ppeers_lock);
 259
 260         /* The next two chunks should probably be moved to vmspace_exit. */
 261         vm = p->p_vmspace;
 262         /*
 263          * Release user portion of address space.
 264          * This releases references to vnodes,
 265          * which could cause I/O if the file has been unlinked.
 266          * Need to do this early enough that we can still sleep.
 267          * Can't free the entire vmspace as the kernel stack
 268          * may be mapped within that space also.
 269          *
 270          * Processes sharing the same vmspace may exit in one order, and
 271          * get cleaned up by vmspace_exit() in a different order.  The
 272          * last exiting process to reach this point releases as much of
 273          * the environment as it can, and the last process cleaned up
 274          * by vmspace_exit() (which decrements exitingcnt) cleans up the
 275          * remainder.
 276          */
 277         atomic_add_int(&vm->vm_exitingcnt, 1);
 278         do
 279                 refcnt = vm->vm_refcnt;
 280         while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
 281         if (refcnt == 1) {
 282                 shmexit(vm);
 283                 pmap_remove_pages(vmspace_pmap(vm), vm_map_min(&vm->vm_map),
 284                     vm_map_max(&vm->vm_map));
 285                 (void) vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
 286                     vm_map_max(&vm->vm_map));
 287         }
 288
 289         mtx_lock(&Giant);
 290         sx_xlock(&proctree_lock);
 291         if (SESS_LEADER(p)) {
 292                 struct session *sp;
 293
 294                 sp = p->p_session;
 295                 if (sp->s_ttyvp) {
 296                         /*
 297                          * Controlling process.
 298                          * Signal foreground pgrp,
 299                          * drain controlling terminal
 300                          * and revoke access to controlling terminal.
 301                          */
 302                         if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) {
 303                                 tp = sp->s_ttyp;
 304                                 if (sp->s_ttyp->t_pgrp) {
 305                                         PGRP_LOCK(sp->s_ttyp->t_pgrp);
 306                                         pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1);
 307                                         PGRP_UNLOCK(sp->s_ttyp->t_pgrp);
 308                                 }
 309                                 /* XXX tp should be locked. */
 310                                 sx_xunlock(&proctree_lock);
 311                                 (void) ttywait(tp);
 312                                 sx_xlock(&proctree_lock);
 313                                 /*
 314                                  * The tty could have been revoked
 315                                  * if we blocked.
 316                                  */
 317                                 if (sp->s_ttyvp) {
 318                                         ttyvp = sp->s_ttyvp;
 319                                         SESS_LOCK(p->p_session);
 320                                         sp->s_ttyvp = NULL;
 321                                         SESS_UNLOCK(p->p_session);
 322                                         sx_xunlock(&proctree_lock);
 323                                         VOP_LOCK(ttyvp, LK_EXCLUSIVE, td);
 324                                         VOP_REVOKE(ttyvp, REVOKEALL);
 325                                         vput(ttyvp);
 326                                         sx_xlock(&proctree_lock);
 327                                 }
 328                         }
 329                         if (sp->s_ttyvp) {
 330                                 ttyvp = sp->s_ttyvp;
 331                                 SESS_LOCK(p->p_session);
 332                                 sp->s_ttyvp = NULL;
 333                                 SESS_UNLOCK(p->p_session);
 334                                 vrele(ttyvp);
 335                         }
 336                         /*
 337                          * s_ttyp is not zero'd; we use this to indicate
 338                          * that the session once had a controlling terminal.
 339                          * (for logging and informational purposes)
 340                          */
 341                 }
 342                 SESS_LOCK(p->p_session);
 343                 sp->s_leader = NULL;
 344                 SESS_UNLOCK(p->p_session);
 345         }
 346         fixjobc(p, p->p_pgrp, 0);
 347         sx_xunlock(&proctree_lock);
 348         (void)acct_process(td);
 349         mtx_unlock(&Giant);
 350 #ifdef KTRACE
 351         /*
 352          * release trace file
 353          */
 354         PROC_LOCK(p);
 355         mtx_lock(&ktrace_mtx);
 356         p->p_traceflag = 0;     /* don't trace the vrele() */
 357         tracevp = p->p_tracevp;
 358         p->p_tracevp = NULL;
 359         tracecred = p->p_tracecred;
 360         p->p_tracecred = NULL;
 361         mtx_unlock(&ktrace_mtx);
 362         PROC_UNLOCK(p);
 363         if (tracevp != NULL) {
 364                 mtx_lock(&Giant);
 365                 vrele(tracevp);
 366                 mtx_unlock(&Giant);
 367         }
 368         if (tracecred != NULL)
 369                 crfree(tracecred);
 370 #endif
 371         /*
 372          * Release reference to text vnode
 373          */
 374         if ((vtmp = p->p_textvp) != NULL) {
 375                 p->p_textvp = NULL;
 376                 mtx_lock(&Giant);
 377                 vrele(vtmp);
 378                 mtx_unlock(&Giant);
 379         }
 380
 381         /*
 382          * Release our limits structure.
 383          */
 384         PROC_LOCK(p);
 385         plim = p->p_limit;
 386         p->p_limit = NULL;
 387         PROC_UNLOCK(p);
 388         lim_free(plim);
 389
 390         /*
 391          * Remove proc from allproc queue and pidhash chain.
 392          * Place onto zombproc.  Unlink from parent's child list.
 393          */
 394         sx_xlock(&allproc_lock);
 395         LIST_REMOVE(p, p_list);
 396         LIST_INSERT_HEAD(&zombproc, p, p_list);
 397         LIST_REMOVE(p, p_hash);
 398         sx_xunlock(&allproc_lock);
 399
 400         sx_xlock(&proctree_lock);
 401         q = LIST_FIRST(&p->p_children);
 402         if (q != NULL)          /* only need this if any child is S_ZOMB */
 403                 wakeup(initproc);
 404         for (; q != NULL; q = nq) {
 405                 nq = LIST_NEXT(q, p_sibling);
 406                 PROC_LOCK(q);
 407                 proc_reparent(q, initproc);
 408                 q->p_sigparent = SIGCHLD;
 409                 /*
 410                  * Traced processes are killed
 411                  * since their existence means someone is screwing up.
 412                  */
 413                 if (q->p_flag & P_TRACED) {
 414                         q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE);
 415                         psignal(q, SIGKILL);
 416                 }
 417                 PROC_UNLOCK(q);
 418         }
 419
 420         /*
 421          * Save exit status and finalize rusage info except for times,
 422          * adding in child rusage info.
 423          */
 424         PROC_LOCK(p);
 425         p->p_xstat = rv;
 426         p->p_xthread = td;
 427         p->p_stats->p_ru.ru_nvcsw++;
 428         *p->p_ru = p->p_stats->p_ru;
 429         ruadd(p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux);
 430
 431         /*
 432          * Notify interested parties of our demise.
 433          */
 434         KNOTE_LOCKED(&p->p_klist, NOTE_EXIT);
 435
 436         /*
 437          * Just delete all entries in the p_klist. At this point we won't
 438          * report any more events, and there are nasty race conditions that
 439          * can beat us if we don't.
 440          */
 441         knlist_clear(&p->p_klist, 1);
 442
 443         /*
 444          * Notify parent that we're gone.  If parent has the PS_NOCLDWAIT
 445          * flag set, or if the handler is set to SIG_IGN, notify process
 446          * 1 instead (and hope it will handle this situation).
 447          */
 448         PROC_LOCK(p->p_pptr);
 449         mtx_lock(&p->p_pptr->p_sigacts->ps_mtx);
 450         if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) {
 451                 struct proc *pp;
 452
 453                 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
 454                 pp = p->p_pptr;
 455                 PROC_UNLOCK(pp);
 456                 proc_reparent(p, initproc);
 457                 p->p_sigparent = SIGCHLD;
 458                 PROC_LOCK(p->p_pptr);
 459                 /*
 460                  * If this was the last child of our parent, notify
 461                  * parent, so in case he was wait(2)ing, he will
 462                  * continue.
 463                  */
 464                 if (LIST_EMPTY(&pp->p_children))
 465                         wakeup(pp);
 466         } else
 467                 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx);
 468
 469         if (p->p_pptr == initproc)
 470                 psignal(p->p_pptr, SIGCHLD);
 471         else if (p->p_sigparent != 0)
 472                 psignal(p->p_pptr, p->p_sigparent);
 473         PROC_UNLOCK(p->p_pptr);
 474
 475         /*
 476          * If this is a kthread, then wakeup anyone waiting for it to exit.
 477          */
 478         if (p->p_flag & P_KTHREAD)
 479                 wakeup(p);
 480         PROC_UNLOCK(p);
 481
 482         /*
 483          * Finally, call machine-dependent code to release the remaining
 484          * resources including address space.
 485          * The address space is released by "vmspace_exitfree(p)" in
 486          * vm_waitproc().
 487          */
 488         cpu_exit(td);
 489
 490         PROC_LOCK(p);
 491         PROC_LOCK(p->p_pptr);
 492         sx_xunlock(&proctree_lock);
 493
 494         /*
 495          * We have to wait until after acquiring all locks before
 496          * changing p_state.  We need to avoid all possible context
 497          * switches (including ones from blocking on a mutex) while
 498          * marked as a zombie.
 499          */
 500         mtx_lock_spin(&sched_lock);
 501         p->p_state = PRS_ZOMBIE;
 502
 503         critical_enter();
 504         mtx_unlock_spin(&sched_lock);
 505         wakeup(p->p_pptr);
 506
 507         PROC_UNLOCK(p->p_pptr);
 508         mtx_lock_spin(&sched_lock);
 509         critical_exit();
 510
 511         /* Do the same timestamp bookkeeping that mi_switch() would do. */
 512         binuptime(&new_switchtime);
 513         bintime_add(&p->p_rux.rux_runtime, &new_switchtime);
 514         bintime_sub(&p->p_rux.rux_runtime, PCPU_PTR(switchtime));
 515         PCPU_SET(switchtime, new_switchtime);
 516         PCPU_SET(switchticks, ticks);
 517         cnt.v_swtch++;
 518
 519         sched_exit(p->p_pptr, td);
 520
 521         /*
 522          * Hopefully no one will try to deliver a signal to the process this
 523          * late in the game.
 524          */
 525         knlist_destroy(&p->p_klist);
 526
 527         /*
 528          * Make sure the scheduler takes this thread out of its tables etc.
 529          * This will also release this thread's reference to the ucred.
 530          * Other thread parts to release include pcb bits and such.
 531          */
 532         thread_exit();
 533 }
 534
 535 #ifdef COMPAT_43
 536 /*
 537  * The dirty work is handled by kern_wait().
 538  *
 539  * MPSAFE.
 540  */
 541 int
 542 owait(struct thread *td, struct owait_args *uap __unused)
 543 {
 544         int error, status;
 545
 546         error = kern_wait(td, WAIT_ANY, &status, 0, NULL);
 547         if (error == 0)
 548                 td->td_retval[1] = status;
 549         return (error);
 550 }
 551 #endif /* COMPAT_43 */
 552
 553 /*
 554  * The dirty work is handled by kern_wait().
 555  *
 556  * MPSAFE.
 557  */
 558 int
 559 wait4(struct thread *td, struct wait_args *uap)
 560 {
 561         struct rusage ru, *rup;
 562         int error, status;
 563
 564         if (uap->rusage != NULL)
 565                 rup = &ru;
 566         else
 567                 rup = NULL;
 568         error = kern_wait(td, uap->pid, &status, uap->options, rup);
 569         if (uap->status != NULL && error == 0)
 570                 error = copyout(&status, uap->status, sizeof(status));
 571         if (uap->rusage != NULL && error == 0)
 572                 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
 573         return (error);
 574 }
 575
 576 int
 577 kern_wait(struct thread *td, pid_t pid, int *status, int options,
 578     struct rusage *rusage)
 579 {
 580         struct proc *p, *q, *t;
 581         int error, nfound;
 582
 583         q = td->td_proc;
 584         if (pid == 0) {
 585                 PROC_LOCK(q);
 586                 pid = -q->p_pgid;
 587                 PROC_UNLOCK(q);
 588         }
 589         if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE))
 590                 return (EINVAL);
 591 loop:
 592         if (q->p_flag & P_STATCHILD) {
 593                 PROC_LOCK(q);
 594                 q->p_flag &= ~P_STATCHILD;
 595                 PROC_UNLOCK(q);
 596         }
 597         nfound = 0;
 598         sx_xlock(&proctree_lock);
 599         LIST_FOREACH(p, &q->p_children, p_sibling) {
 600                 PROC_LOCK(p);
 601                 if (pid != WAIT_ANY &&
 602                     p->p_pid != pid && p->p_pgid != -pid) {
 603                         PROC_UNLOCK(p);
 604                         continue;
 605                 }
 606                 if (p_canwait(td, p)) {
 607                         PROC_UNLOCK(p);
 608                         continue;
 609                 }
 610
 611                 /*
 612                  * This special case handles a kthread spawned by linux_clone
 613                  * (see linux_misc.c).  The linux_wait4 and linux_waitpid
 614                  * functions need to be able to distinguish between waiting
 615                  * on a process and waiting on a thread.  It is a thread if
 616                  * p_sigparent is not SIGCHLD, and the WLINUXCLONE option
 617                  * signifies we want to wait for threads and not processes.
 618                  */
 619                 if ((p->p_sigparent != SIGCHLD) ^
 620                     ((options & WLINUXCLONE) != 0)) {
 621                         PROC_UNLOCK(p);
 622                         continue;
 623                 }
 624
 625                 nfound++;
 626                 if (p->p_state == PRS_ZOMBIE) {
 627                         td->td_retval[0] = p->p_pid;
 628                         if (status)
 629                                 *status = p->p_xstat;   /* convert to int */
 630                         if (rusage) {
 631                                 *rusage = *p->p_ru;
 632                                 calcru(p, &rusage->ru_utime, &rusage->ru_stime);
 633                         }
 634
 635                         /*
 636                          * If we got the child via a ptrace 'attach',
 637                          * we need to give it back to the old parent.
 638                          */
 639                         PROC_UNLOCK(p);
 640                         if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) {
 641                                 PROC_LOCK(p);
 642                                 p->p_oppid = 0;
 643                                 proc_reparent(p, t);
 644                                 PROC_UNLOCK(p);
 645                                 psignal(t, SIGCHLD);
 646                                 wakeup(t);
 647                                 PROC_UNLOCK(t);
 648                                 sx_xunlock(&proctree_lock);
 649                                 return (0);
 650                         }
 651
 652                         /*
 653                          * Remove other references to this process to ensure
 654                          * we have an exclusive reference.
 655                          */
 656                         sx_xlock(&allproc_lock);
 657                         LIST_REMOVE(p, p_list); /* off zombproc */
 658                         sx_xunlock(&allproc_lock);
 659                         LIST_REMOVE(p, p_sibling);
 660                         leavepgrp(p);
 661                         sx_xunlock(&proctree_lock);
 662
 663                         /*
 664                          * As a side effect of this lock, we know that
 665                          * all other writes to this proc are visible now, so
 666                          * no more locking is needed for p.
 667                          */
 668                         PROC_LOCK(p);
 669                         p->p_xstat = 0;         /* XXX: why? */
 670                         PROC_UNLOCK(p);
 671                         PROC_LOCK(q);
 672                         ruadd(&q->p_stats->p_cru, &q->p_crux, p->p_ru,
 673                             &p->p_rux);
 674                         PROC_UNLOCK(q);
 675                         FREE(p->p_ru, M_ZOMBIE);
 676                         p->p_ru = NULL;
 677
 678                         /*
 679                          * Decrement the count of procs running with this uid.
 680                          */
 681                         (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0);
 682
 683                         /*
 684                          * Free credentials, arguments, and sigacts.
 685                          */
 686                         crfree(p->p_ucred);
 687                         p->p_ucred = NULL;
 688                         pargs_drop(p->p_args);
 689                         p->p_args = NULL;
 690                         sigacts_free(p->p_sigacts);
 691                         p->p_sigacts = NULL;
 692
 693                         /*
 694                          * Do any thread-system specific cleanups.
 695                          */
 696                         thread_wait(p);
 697
 698                         /*
 699                          * Give vm and machine-dependent layer a chance
 700                          * to free anything that cpu_exit couldn't
 701                          * release while still running in process context.
 702                          */
 703                         vm_waitproc(p);
 704 #ifdef MAC
 705                         mac_destroy_proc(p);
 706 #endif
 707                         KASSERT(FIRST_THREAD_IN_PROC(p),
 708                             ("kern_wait: no residual thread!"));
 709                         uma_zfree(proc_zone, p);
 710                         sx_xlock(&allproc_lock);
 711                         nprocs--;
 712                         sx_xunlock(&allproc_lock);
 713                         return (0);
 714                 }
 715                 mtx_lock_spin(&sched_lock);
 716                 if ((p->p_flag & P_STOPPED_SIG) &&
 717                     (p->p_suspcount == p->p_numthreads) &&
 718                     (p->p_flag & P_WAITED) == 0 &&
 719                     (p->p_flag & P_TRACED || options & WUNTRACED)) {
 720                         mtx_unlock_spin(&sched_lock);
 721                         p->p_flag |= P_WAITED;
 722                         sx_xunlock(&proctree_lock);
 723                         td->td_retval[0] = p->p_pid;
 724                         if (status)
 725                                 *status = W_STOPCODE(p->p_xstat);
 726                         PROC_UNLOCK(p);
 727                         return (0);
 728                 }
 729                 mtx_unlock_spin(&sched_lock);
 730                 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) {
 731                         sx_xunlock(&proctree_lock);
 732                         td->td_retval[0] = p->p_pid;
 733                         p->p_flag &= ~P_CONTINUED;
 734                         PROC_UNLOCK(p);
 735
 736                         if (status)
 737                                 *status = SIGCONT;
 738                         return (0);
 739                 }
 740                 PROC_UNLOCK(p);
 741         }
 742         if (nfound == 0) {
 743                 sx_xunlock(&proctree_lock);
 744                 return (ECHILD);
 745         }
 746         if (options & WNOHANG) {
 747                 sx_xunlock(&proctree_lock);
 748                 td->td_retval[0] = 0;
 749                 return (0);
 750         }
 751         PROC_LOCK(q);
 752         sx_xunlock(&proctree_lock);
 753         if (q->p_flag & P_STATCHILD) {
 754                 q->p_flag &= ~P_STATCHILD;
 755                 error = 0;
 756         } else
 757                 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0);
 758         PROC_UNLOCK(q);
 759         if (error)
 760                 return (error);
 761         goto loop;
 762 }
 763
 764 /*
 765  * Make process 'parent' the new parent of process 'child'.
 766  * Must be called with an exclusive hold of proctree lock.
 767  */
 768 void
 769 proc_reparent(struct proc *child, struct proc *parent)
 770 {
 771
 772         sx_assert(&proctree_lock, SX_XLOCKED);
 773         PROC_LOCK_ASSERT(child, MA_OWNED);
 774         if (child->p_pptr == parent)
 775                 return;
 776
 777         LIST_REMOVE(child, p_sibling);
 778         LIST_INSERT_HEAD(&parent->p_children, child, p_sibling);
 779         child->p_pptr = parent;
 780 }