1 /* $NetBSD: kern_exit.c,v 1.231 2010/12/18 01:36:19 rmind Exp $ */ 2 3 /*- 4 * Copyright (c) 1998, 1999, 2006, 2007, 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center, and by Andrew Doran. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1982, 1986, 1989, 1991, 1993 35 * The Regents of the University of California. All rights reserved. 36 * (c) UNIX System Laboratories, Inc. 37 * All or some portions of this file are derived from material licensed 38 * to the University of California by American Telephone and Telegraph 39 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 40 * the permission of UNIX System Laboratories, Inc. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)kern_exit.c 8.10 (Berkeley) 2/23/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: kern_exit.c,v 1.231 2010/12/18 01:36:19 rmind Exp $"); 71 72 #include "opt_ktrace.h" 73 #include "opt_perfctrs.h" 74 #include "opt_sa.h" 75 #include "opt_sysv.h" 76 77 #include <sys/param.h> 78 #include <sys/systm.h> 79 #include <sys/ioctl.h> 80 #include <sys/tty.h> 81 #include <sys/time.h> 82 #include <sys/resource.h> 83 #include <sys/kernel.h> 84 #include <sys/proc.h> 85 #include <sys/buf.h> 86 #include <sys/wait.h> 87 #include <sys/file.h> 88 #include <sys/vnode.h> 89 #include <sys/syslog.h> 90 #include <sys/pool.h> 91 #include <sys/uidinfo.h> 92 #if defined(PERFCTRS) 93 #include <sys/pmc.h> 94 #endif 95 #include <sys/ptrace.h> 96 #include <sys/acct.h> 97 #include <sys/filedesc.h> 98 #include <sys/ras.h> 99 #include <sys/signalvar.h> 100 #include <sys/sched.h> 101 #include <sys/sa.h> 102 #include <sys/savar.h> 103 #include <sys/mount.h> 104 #include <sys/syscallargs.h> 105 #include <sys/kauth.h> 106 #include <sys/sleepq.h> 107 #include <sys/lockdebug.h> 108 #include <sys/ktrace.h> 109 #include <sys/cpu.h> 110 #include <sys/lwpctl.h> 111 #include <sys/atomic.h> 112 #include <sys/sdt.h> 113 114 #include <uvm/uvm_extern.h> 115 116 #ifdef DEBUG_EXIT 117 int debug_exit = 0; 118 #define DPRINTF(x) if (debug_exit) printf x 119 #else 120 #define DPRINTF(x) 121 #endif 122 123 static int find_stopped_child(struct proc *, pid_t, int, struct proc **, int *); 124 static void proc_free(struct proc *, struct rusage *); 125 126 /* 127 * DTrace SDT provider definitions 128 */ 129 SDT_PROBE_DEFINE(proc,,,exit, 130 "int", NULL, /* reason */ 131 NULL, NULL, NULL, NULL, 132 NULL, NULL, NULL, NULL); 133 /* 134 * Fill in the appropriate signal information, and signal the parent. 135 */ 136 static void 137 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi) 138 { 139 140 KSI_INIT(ksi); 141 if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) { 142 if (WIFSIGNALED(p->p_xstat)) { 143 if (WCOREDUMP(p->p_xstat)) 144 ksi->ksi_code = CLD_DUMPED; 145 else 146 ksi->ksi_code = CLD_KILLED; 147 } else { 148 ksi->ksi_code = CLD_EXITED; 149 } 150 } 151 /* 152 * We fill those in, even for non-SIGCHLD. 153 * It's safe to access p->p_cred unlocked here. 154 */ 155 ksi->ksi_pid = p->p_pid; 156 ksi->ksi_uid = kauth_cred_geteuid(p->p_cred); 157 ksi->ksi_status = p->p_xstat; 158 /* XXX: is this still valid? */ 159 ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec; 160 ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec; 161 } 162 163 /* 164 * exit -- 165 * Death of process. 166 */ 167 int 168 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval) 169 { 170 /* { 171 syscallarg(int) rval; 172 } */ 173 struct proc *p = l->l_proc; 174 175 /* Don't call exit1() multiple times in the same process. */ 176 mutex_enter(p->p_lock); 177 if (p->p_sflag & PS_WEXIT) { 178 mutex_exit(p->p_lock); 179 lwp_exit(l); 180 } 181 182 /* exit1() will release the mutex. */ 183 exit1(l, W_EXITCODE(SCARG(uap, rval), 0)); 184 /* NOTREACHED */ 185 return (0); 186 } 187 188 /* 189 * Exit: deallocate address space and other resources, change proc state 190 * to zombie, and unlink proc from allproc and parent's lists. Save exit 191 * status and rusage for wait(). Check for child processes and orphan them. 192 * 193 * Must be called with p->p_lock held. Does not return. 194 */ 195 void 196 exit1(struct lwp *l, int rv) 197 { 198 struct proc *p, *q, *nq; 199 struct pgrp *pgrp; 200 ksiginfo_t ksi; 201 ksiginfoq_t kq; 202 int wakeinit, sa; 203 204 p = l->l_proc; 205 206 KASSERT(mutex_owned(p->p_lock)); 207 208 if (__predict_false(p == initproc)) 209 panic("init died (signal %d, exit %d)", 210 WTERMSIG(rv), WEXITSTATUS(rv)); 211 212 /* 213 * Disable scheduler activation upcalls. We're trying to get out of 214 * here. 215 */ 216 sa = 0; 217 #ifdef KERN_SA 218 if ((p->p_sa != NULL)) { 219 l->l_pflag |= LP_SA_NOBLOCK; 220 sa = 1; 221 } 222 #endif 223 224 p->p_sflag |= PS_WEXIT; 225 226 /* 227 * Force all other LWPs to exit before we do. Only then can we 228 * begin to tear down the rest of the process state. 229 */ 230 if (sa || p->p_nlwps > 1) 231 exit_lwps(l); 232 233 ksiginfo_queue_init(&kq); 234 235 /* 236 * If we have been asked to stop on exit, do so now. 237 */ 238 if (__predict_false(p->p_sflag & PS_STOPEXIT)) { 239 KERNEL_UNLOCK_ALL(l, &l->l_biglocks); 240 sigclearall(p, &contsigmask, &kq); 241 p->p_waited = 0; 242 membar_producer(); 243 p->p_stat = SSTOP; 244 lwp_lock(l); 245 p->p_nrlwps--; 246 l->l_stat = LSSTOP; 247 lwp_unlock(l); 248 mutex_exit(p->p_lock); 249 lwp_lock(l); 250 mi_switch(l); 251 KERNEL_LOCK(l->l_biglocks, l); 252 mutex_enter(p->p_lock); 253 } 254 255 /* 256 * Bin any remaining signals and mark the process as dying so it will 257 * not be found for, e.g. signals. 258 */ 259 sigfillset(&p->p_sigctx.ps_sigignore); 260 sigclearall(p, NULL, &kq); 261 p->p_stat = SDYING; 262 mutex_exit(p->p_lock); 263 ksiginfo_queue_drain(&kq); 264 265 /* Destroy any lwpctl info. */ 266 if (p->p_lwpctl != NULL) 267 lwp_ctl_exit(); 268 269 /* 270 * Drain all remaining references that procfs, ptrace and others may 271 * have on the process. 272 */ 273 rw_enter(&p->p_reflock, RW_WRITER); 274 275 DPRINTF(("exit1: %d.%d exiting.\n", p->p_pid, l->l_lid)); 276 277 timers_free(p, TIMERS_ALL); 278 #if defined(__HAVE_RAS) 279 ras_purgeall(); 280 #endif 281 282 /* 283 * Close open files, release open-file table and free signal 284 * actions. This may block! 285 */ 286 fd_free(); 287 cwdfree(p->p_cwdi); 288 p->p_cwdi = NULL; 289 doexithooks(p); 290 sigactsfree(p->p_sigacts); 291 292 /* 293 * Write out accounting data. 294 */ 295 (void)acct_process(l); 296 297 #ifdef KTRACE 298 /* 299 * Release trace file. 300 */ 301 if (p->p_tracep != NULL) { 302 mutex_enter(&ktrace_lock); 303 ktrderef(p); 304 mutex_exit(&ktrace_lock); 305 } 306 #endif 307 308 /* 309 * If emulation has process exit hook, call it now. 310 * Set the exit status now so that the exit hook has 311 * an opportunity to tweak it (COMPAT_LINUX requires 312 * this for thread group emulation) 313 */ 314 p->p_xstat = rv; 315 if (p->p_emul->e_proc_exit) 316 (*p->p_emul->e_proc_exit)(p); 317 318 /* 319 * Free the VM resources we're still holding on to. 320 * We must do this from a valid thread because doing 321 * so may block. This frees vmspace, which we don't 322 * need anymore. The only remaining lwp is the one 323 * we run at this moment, nothing runs in userland 324 * anymore. 325 */ 326 uvm_proc_exit(p); 327 328 /* 329 * Stop profiling. 330 */ 331 if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) { 332 mutex_spin_enter(&p->p_stmutex); 333 stopprofclock(p); 334 mutex_spin_exit(&p->p_stmutex); 335 } 336 337 /* 338 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we 339 * wake up the parent early to avoid deadlock. We can do this once 340 * the VM resources are released. 341 */ 342 mutex_enter(proc_lock); 343 if (p->p_lflag & PL_PPWAIT) { 344 p->p_lflag &= ~PL_PPWAIT; 345 cv_broadcast(&p->p_pptr->p_waitcv); 346 } 347 348 if (SESS_LEADER(p)) { 349 struct vnode *vprele = NULL, *vprevoke = NULL; 350 struct session *sp = p->p_session; 351 struct tty *tp; 352 353 if (sp->s_ttyvp) { 354 /* 355 * Controlling process. 356 * Signal foreground pgrp, 357 * drain controlling terminal 358 * and revoke access to controlling terminal. 359 */ 360 tp = sp->s_ttyp; 361 mutex_spin_enter(&tty_lock); 362 if (tp->t_session == sp) { 363 /* we can't guarantee the revoke will do this */ 364 pgrp = tp->t_pgrp; 365 tp->t_pgrp = NULL; 366 tp->t_session = NULL; 367 mutex_spin_exit(&tty_lock); 368 if (pgrp != NULL) { 369 pgsignal(pgrp, SIGHUP, 1); 370 } 371 mutex_exit(proc_lock); 372 (void) ttywait(tp); 373 mutex_enter(proc_lock); 374 375 /* The tty could have been revoked. */ 376 vprevoke = sp->s_ttyvp; 377 } else 378 mutex_spin_exit(&tty_lock); 379 vprele = sp->s_ttyvp; 380 sp->s_ttyvp = NULL; 381 /* 382 * s_ttyp is not zero'd; we use this to indicate 383 * that the session once had a controlling terminal. 384 * (for logging and informational purposes) 385 */ 386 } 387 sp->s_leader = NULL; 388 389 if (vprevoke != NULL || vprele != NULL) { 390 if (vprevoke != NULL) { 391 /* Releases proc_lock. */ 392 proc_sessrele(sp); 393 VOP_REVOKE(vprevoke, REVOKEALL); 394 } else 395 mutex_exit(proc_lock); 396 if (vprele != NULL) 397 vrele(vprele); 398 mutex_enter(proc_lock); 399 } 400 } 401 fixjobc(p, p->p_pgrp, 0); 402 403 /* 404 * Finalize the last LWP's specificdata, as well as the 405 * specificdata for the proc itself. 406 */ 407 lwp_finispecific(l); 408 proc_finispecific(p); 409 410 /* 411 * Notify interested parties of our demise. 412 */ 413 KNOTE(&p->p_klist, NOTE_EXIT); 414 415 SDT_PROBE(proc,,,exit, 416 (WCOREDUMP(rv) ? CLD_DUMPED : 417 (WIFSIGNALED(rv) ? CLD_KILLED : CLD_EXITED)), 418 0,0,0,0); 419 420 #if PERFCTRS 421 /* 422 * Save final PMC information in parent process & clean up. 423 */ 424 if (PMC_ENABLED(p)) { 425 pmc_save_context(p); 426 pmc_accumulate(p->p_pptr, p); 427 pmc_process_exit(p); 428 } 429 #endif 430 431 /* 432 * Reset p_opptr pointer of all former children which got 433 * traced by another process and were reparented. We reset 434 * it to NULL here; the trace detach code then reparents 435 * the child to initproc. We only check allproc list, since 436 * eventual former children on zombproc list won't reference 437 * p_opptr anymore. 438 */ 439 if (__predict_false(p->p_slflag & PSL_CHTRACED)) { 440 PROCLIST_FOREACH(q, &allproc) { 441 if (q->p_opptr == p) 442 q->p_opptr = NULL; 443 } 444 } 445 446 /* 447 * Give orphaned children to init(8). 448 */ 449 q = LIST_FIRST(&p->p_children); 450 wakeinit = (q != NULL); 451 for (; q != NULL; q = nq) { 452 nq = LIST_NEXT(q, p_sibling); 453 454 /* 455 * Traced processes are killed since their existence 456 * means someone is screwing up. Since we reset the 457 * trace flags, the logic in sys_wait4() would not be 458 * triggered to reparent the process to its 459 * original parent, so we must do this here. 460 */ 461 if (__predict_false(q->p_slflag & PSL_TRACED)) { 462 mutex_enter(p->p_lock); 463 q->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL); 464 mutex_exit(p->p_lock); 465 if (q->p_opptr != q->p_pptr) { 466 struct proc *t = q->p_opptr; 467 proc_reparent(q, t ? t : initproc); 468 q->p_opptr = NULL; 469 } else 470 proc_reparent(q, initproc); 471 killproc(q, "orphaned traced process"); 472 } else 473 proc_reparent(q, initproc); 474 } 475 476 /* 477 * Move proc from allproc to zombproc, it's now nearly ready to be 478 * collected by parent. 479 */ 480 LIST_REMOVE(l, l_list); 481 LIST_REMOVE(p, p_list); 482 LIST_INSERT_HEAD(&zombproc, p, p_list); 483 484 /* 485 * Mark the process as dead. We must do this before we signal 486 * the parent. 487 */ 488 p->p_stat = SDEAD; 489 490 /* Put in front of parent's sibling list for parent to collect it */ 491 q = p->p_pptr; 492 q->p_nstopchild++; 493 if (LIST_FIRST(&q->p_children) != p) { 494 /* Put child where it can be found quickly */ 495 LIST_REMOVE(p, p_sibling); 496 LIST_INSERT_HEAD(&q->p_children, p, p_sibling); 497 } 498 499 /* 500 * Notify parent that we're gone. If parent has the P_NOCLDWAIT 501 * flag set, notify init instead (and hope it will handle 502 * this situation). 503 */ 504 if (q->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) { 505 proc_reparent(p, initproc); 506 wakeinit = 1; 507 508 /* 509 * If this was the last child of our parent, notify 510 * parent, so in case he was wait(2)ing, he will 511 * continue. 512 */ 513 if (LIST_FIRST(&q->p_children) == NULL) 514 cv_broadcast(&q->p_waitcv); 515 } 516 517 /* Reload parent pointer, since p may have been reparented above */ 518 q = p->p_pptr; 519 520 if (__predict_false((p->p_slflag & PSL_FSTRACE) == 0 && 521 p->p_exitsig != 0)) { 522 exit_psignal(p, q, &ksi); 523 kpsignal(q, &ksi, NULL); 524 } 525 526 /* Calculate the final rusage info. */ 527 calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime, 528 NULL, NULL); 529 530 if (wakeinit) 531 cv_broadcast(&initproc->p_waitcv); 532 533 callout_destroy(&l->l_timeout_ch); 534 535 /* 536 * Remaining lwp resources will be freed in lwp_exit2() once we've 537 * switch to idle context; at that point, we will be marked as a 538 * full blown zombie. 539 */ 540 mutex_enter(p->p_lock); 541 lwp_drainrefs(l); 542 lwp_lock(l); 543 l->l_prflag &= ~LPR_DETACHED; 544 l->l_stat = LSZOMB; 545 lwp_unlock(l); 546 KASSERT(curlwp == l); 547 KASSERT(p->p_nrlwps == 1); 548 KASSERT(p->p_nlwps == 1); 549 p->p_stat = SZOMB; 550 p->p_nrlwps--; 551 p->p_nzlwps++; 552 p->p_ndlwps = 0; 553 mutex_exit(p->p_lock); 554 555 /* 556 * Signal the parent to collect us, and drop the proclist lock. 557 * Drop debugger/procfs lock; no new references can be gained. 558 */ 559 cv_broadcast(&p->p_pptr->p_waitcv); 560 rw_exit(&p->p_reflock); 561 mutex_exit(proc_lock); 562 563 /* Verify that we hold no locks other than the kernel lock. */ 564 LOCKDEBUG_BARRIER(&kernel_lock, 0); 565 566 /* 567 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP! 568 */ 569 570 /* 571 * Give machine-dependent code a chance to free any MD LWP 572 * resources. This must be done before uvm_lwp_exit(), in 573 * case these resources are in the PCB. 574 */ 575 cpu_lwp_free(l, 1); 576 pmap_deactivate(l); 577 578 /* This process no longer needs to hold the kernel lock. */ 579 #ifdef notyet 580 /* XXXSMP hold in lwp_userret() */ 581 KERNEL_UNLOCK_LAST(l); 582 #else 583 KERNEL_UNLOCK_ALL(l, NULL); 584 #endif 585 586 lwp_exit_switchaway(l); 587 } 588 589 void 590 exit_lwps(struct lwp *l) 591 { 592 struct proc *p; 593 struct lwp *l2; 594 int error; 595 lwpid_t waited; 596 int nlocks; 597 598 KERNEL_UNLOCK_ALL(l, &nlocks); 599 600 p = l->l_proc; 601 KASSERT(mutex_owned(p->p_lock)); 602 603 #ifdef KERN_SA 604 if (p->p_sa != NULL) { 605 struct sadata_vp *vp; 606 SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) { 607 /* 608 * Make SA-cached LWPs normal process interruptable 609 * so that the exit code can wake them. Locking 610 * savp_mutex locks all the lwps on this vp that 611 * we need to adjust. 612 */ 613 mutex_enter(&vp->savp_mutex); 614 DPRINTF(("exit_lwps: Making cached LWPs of %d on " 615 "VP %d interruptable: ", p->p_pid, vp->savp_id)); 616 TAILQ_FOREACH(l2, &vp->savp_lwpcache, l_sleepchain) { 617 l2->l_flag |= LW_SINTR; 618 DPRINTF(("%d ", l2->l_lid)); 619 } 620 DPRINTF(("\n")); 621 622 DPRINTF(("exit_lwps: Making unblocking LWPs of %d on " 623 "VP %d interruptable: ", p->p_pid, vp->savp_id)); 624 TAILQ_FOREACH(l2, &vp->savp_woken, l_sleepchain) { 625 vp->savp_woken_count--; 626 l2->l_flag |= LW_SINTR; 627 DPRINTF(("%d ", l2->l_lid)); 628 } 629 DPRINTF(("\n")); 630 mutex_exit(&vp->savp_mutex); 631 } 632 } 633 #endif 634 635 retry: 636 /* 637 * Interrupt LWPs in interruptable sleep, unsuspend suspended 638 * LWPs and then wait for everyone else to finish. 639 */ 640 LIST_FOREACH(l2, &p->p_lwps, l_sibling) { 641 if (l2 == l) 642 continue; 643 lwp_lock(l2); 644 l2->l_flag &= ~LW_SA; 645 l2->l_flag |= LW_WEXIT; 646 if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) || 647 l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) { 648 /* setrunnable() will release the lock. */ 649 setrunnable(l2); 650 DPRINTF(("exit_lwps: Made %d.%d runnable\n", 651 p->p_pid, l2->l_lid)); 652 continue; 653 } 654 lwp_unlock(l2); 655 } 656 while (p->p_nlwps > 1) { 657 DPRINTF(("exit_lwps: waiting for %d LWPs (%d zombies)\n", 658 p->p_nlwps, p->p_nzlwps)); 659 error = lwp_wait1(l, 0, &waited, LWPWAIT_EXITCONTROL); 660 if (p->p_nlwps == 1) 661 break; 662 if (error == EDEADLK) { 663 /* 664 * LWPs can get suspended/slept behind us. 665 * (eg. sa_setwoken) 666 * kick them again and retry. 667 */ 668 goto retry; 669 } 670 if (error) 671 panic("exit_lwps: lwp_wait1 failed with error %d", 672 error); 673 DPRINTF(("exit_lwps: Got LWP %d from lwp_wait1()\n", waited)); 674 } 675 676 KERNEL_LOCK(nlocks, l); 677 KASSERT(p->p_nlwps == 1); 678 } 679 680 int 681 do_sys_wait(int *pid, int *status, int options, struct rusage *ru) 682 { 683 proc_t *child; 684 int error; 685 686 if (ru != NULL) { 687 memset(ru, 0, sizeof(*ru)); 688 } 689 mutex_enter(proc_lock); 690 error = find_stopped_child(curproc, *pid, options, &child, status); 691 if (child == NULL) { 692 mutex_exit(proc_lock); 693 *pid = 0; 694 return error; 695 } 696 *pid = child->p_pid; 697 698 if (child->p_stat == SZOMB) { 699 /* proc_free() will release the proc_lock. */ 700 if (options & WNOWAIT) { 701 mutex_exit(proc_lock); 702 } else { 703 proc_free(child, ru); 704 } 705 } else { 706 /* Child state must have been SSTOP. */ 707 mutex_exit(proc_lock); 708 *status = W_STOPCODE(*status); 709 } 710 return 0; 711 } 712 713 int 714 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap, 715 register_t *retval) 716 { 717 /* { 718 syscallarg(int) pid; 719 syscallarg(int *) status; 720 syscallarg(int) options; 721 syscallarg(struct rusage *) rusage; 722 } */ 723 int error, status, pid = SCARG(uap, pid); 724 struct rusage ru; 725 726 error = do_sys_wait(&pid, &status, SCARG(uap, options), 727 SCARG(uap, rusage) != NULL ? &ru : NULL); 728 729 retval[0] = pid; 730 if (pid == 0) { 731 return error; 732 } 733 if (SCARG(uap, status)) { 734 error = copyout(&status, SCARG(uap, status), sizeof(status)); 735 } 736 if (SCARG(uap, rusage) && error == 0) { 737 error = copyout(&ru, SCARG(uap, rusage), sizeof(ru)); 738 } 739 return error; 740 } 741 742 /* 743 * Scan list of child processes for a child process that has stopped or 744 * exited. Used by sys_wait4 and 'compat' equivalents. 745 * 746 * Must be called with the proc_lock held, and may release while waiting. 747 */ 748 static int 749 find_stopped_child(struct proc *parent, pid_t pid, int options, 750 struct proc **child_p, int *status_p) 751 { 752 struct proc *child, *dead; 753 int error; 754 755 KASSERT(mutex_owned(proc_lock)); 756 757 if (options & ~(WUNTRACED|WNOHANG|WALTSIG|WALLSIG) 758 && !(options & WOPTSCHECKED)) { 759 *child_p = NULL; 760 return EINVAL; 761 } 762 763 if (pid == 0 && !(options & WOPTSCHECKED)) 764 pid = -parent->p_pgid; 765 766 for (;;) { 767 error = ECHILD; 768 dead = NULL; 769 770 LIST_FOREACH(child, &parent->p_children, p_sibling) { 771 if (pid >= 0) { 772 if (child->p_pid != pid) { 773 child = proc_find_raw(pid); 774 if (child == NULL || 775 child->p_stat == SIDL || 776 child->p_pptr != parent) { 777 child = NULL; 778 break; 779 } 780 } 781 } else if (pid != WAIT_ANY && child->p_pgid != -pid) { 782 /* Child not in correct pgrp */ 783 continue; 784 } 785 786 /* 787 * Wait for processes with p_exitsig != SIGCHLD 788 * processes only if WALTSIG is set; wait for 789 * processes with p_exitsig == SIGCHLD only 790 * if WALTSIG is clear. 791 */ 792 if (((options & WALLSIG) == 0) && 793 (options & WALTSIG ? child->p_exitsig == SIGCHLD 794 : P_EXITSIG(child) != SIGCHLD)){ 795 if (child->p_pid == pid) { 796 child = NULL; 797 break; 798 } 799 continue; 800 } 801 802 error = 0; 803 if ((options & WNOZOMBIE) == 0) { 804 if (child->p_stat == SZOMB) 805 break; 806 if (child->p_stat == SDEAD) { 807 /* 808 * We may occasionally arrive here 809 * after receiving a signal, but 810 * immediately before the child 811 * process is zombified. The wait 812 * will be short, so avoid returning 813 * to userspace. 814 */ 815 dead = child; 816 } 817 } 818 819 if (child->p_stat == SSTOP && 820 child->p_waited == 0 && 821 (child->p_slflag & PSL_TRACED || 822 options & WUNTRACED)) { 823 if ((options & WNOWAIT) == 0) { 824 child->p_waited = 1; 825 parent->p_nstopchild--; 826 } 827 break; 828 } 829 if (parent->p_nstopchild == 0 || child->p_pid == pid) { 830 child = NULL; 831 break; 832 } 833 } 834 835 if (child != NULL || error != 0 || 836 ((options & WNOHANG) != 0 && dead == NULL)) { 837 if (child != NULL) { 838 *status_p = child->p_xstat; 839 } 840 *child_p = child; 841 return error; 842 } 843 844 /* 845 * Wait for another child process to stop. 846 */ 847 error = cv_wait_sig(&parent->p_waitcv, proc_lock); 848 849 if (error != 0) { 850 *child_p = NULL; 851 return error; 852 } 853 } 854 } 855 856 /* 857 * Free a process after parent has taken all the state info. Must be called 858 * with the proclist lock held, and will release before returning. 859 * 860 * *ru is returned to the caller, and must be freed by the caller. 861 */ 862 static void 863 proc_free(struct proc *p, struct rusage *ru) 864 { 865 struct proc *parent = p->p_pptr; 866 struct lwp *l; 867 ksiginfo_t ksi; 868 kauth_cred_t cred1, cred2; 869 uid_t uid; 870 871 KASSERT(mutex_owned(proc_lock)); 872 KASSERT(p->p_nlwps == 1); 873 KASSERT(p->p_nzlwps == 1); 874 KASSERT(p->p_nrlwps == 0); 875 KASSERT(p->p_stat == SZOMB); 876 877 /* 878 * If we got the child via ptrace(2) or procfs, and 879 * the parent is different (meaning the process was 880 * attached, rather than run as a child), then we need 881 * to give it back to the old parent, and send the 882 * parent the exit signal. The rest of the cleanup 883 * will be done when the old parent waits on the child. 884 */ 885 if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) { 886 mutex_enter(p->p_lock); 887 p->p_slflag &= ~(PSL_TRACED|PSL_FSTRACE|PSL_SYSCALL); 888 mutex_exit(p->p_lock); 889 parent = (p->p_opptr == NULL) ? initproc : p->p_opptr; 890 proc_reparent(p, parent); 891 p->p_opptr = NULL; 892 if (p->p_exitsig != 0) { 893 exit_psignal(p, parent, &ksi); 894 kpsignal(parent, &ksi, NULL); 895 } 896 cv_broadcast(&parent->p_waitcv); 897 mutex_exit(proc_lock); 898 return; 899 } 900 901 sched_proc_exit(parent, p); 902 903 /* 904 * Add child times of exiting process onto its own times. 905 * This cannot be done any earlier else it might get done twice. 906 */ 907 l = LIST_FIRST(&p->p_lwps); 908 p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw); 909 p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw; 910 ruadd(&p->p_stats->p_ru, &l->l_ru); 911 ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru); 912 ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru); 913 if (ru != NULL) 914 *ru = p->p_stats->p_ru; 915 p->p_xstat = 0; 916 917 /* Release any SA state. */ 918 #ifdef KERN_SA 919 if (p->p_sa) 920 sa_release(p); 921 #endif 922 923 /* 924 * At this point we are going to start freeing the final resources. 925 * If anyone tries to access the proc structure after here they will 926 * get a shock - bits are missing. Attempt to make it hard! We 927 * don't bother with any further locking past this point. 928 */ 929 p->p_stat = SIDL; /* not even a zombie any more */ 930 LIST_REMOVE(p, p_list); /* off zombproc */ 931 parent->p_nstopchild--; 932 LIST_REMOVE(p, p_sibling); 933 934 /* 935 * Let pid be reallocated. 936 */ 937 proc_free_pid(p->p_pid); 938 939 /* 940 * Unlink process from its process group. 941 * Releases the proc_lock. 942 */ 943 proc_leavepgrp(p); 944 945 /* 946 * Delay release until after lwp_free. 947 */ 948 cred2 = l->l_cred; 949 950 /* 951 * Free the last LWP's resources. 952 * 953 * lwp_free ensures the LWP is no longer running on another CPU. 954 */ 955 lwp_free(l, false, true); 956 957 /* 958 * Now no one except us can reach the process p. 959 */ 960 961 /* 962 * Decrement the count of procs running with this uid. 963 */ 964 cred1 = p->p_cred; 965 uid = kauth_cred_getuid(cred1); 966 (void)chgproccnt(uid, -1); 967 968 /* 969 * Release substructures. 970 */ 971 972 limfree(p->p_limit); 973 pstatsfree(p->p_stats); 974 kauth_cred_free(cred1); 975 kauth_cred_free(cred2); 976 977 /* 978 * Release reference to text vnode 979 */ 980 if (p->p_textvp) 981 vrele(p->p_textvp); 982 983 mutex_destroy(&p->p_auxlock); 984 mutex_obj_free(p->p_lock); 985 mutex_destroy(&p->p_stmutex); 986 cv_destroy(&p->p_waitcv); 987 cv_destroy(&p->p_lwpcv); 988 rw_destroy(&p->p_reflock); 989 990 proc_free_mem(p); 991 } 992 993 /* 994 * make process 'parent' the new parent of process 'child'. 995 * 996 * Must be called with proc_lock held. 997 */ 998 void 999 proc_reparent(struct proc *child, struct proc *parent) 1000 { 1001 1002 KASSERT(mutex_owned(proc_lock)); 1003 1004 if (child->p_pptr == parent) 1005 return; 1006 1007 if (child->p_stat == SZOMB || 1008 (child->p_stat == SSTOP && !child->p_waited)) { 1009 child->p_pptr->p_nstopchild--; 1010 parent->p_nstopchild++; 1011 } 1012 if (parent == initproc) 1013 child->p_exitsig = SIGCHLD; 1014 1015 LIST_REMOVE(child, p_sibling); 1016 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1017 child->p_pptr = parent; 1018 child->p_ppid = parent->p_pid; 1019 } 1020