1 /* $NetBSD: kern_exit.c,v 1.278 2019/12/06 21:36:10 ad 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 1.278 2019/12/06 21:36:10 ad Exp $"); 71 72 #include "opt_ktrace.h" 73 #include "opt_dtrace.h" 74 #include "opt_sysv.h" 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/ioctl.h> 79 #include <sys/tty.h> 80 #include <sys/time.h> 81 #include <sys/resource.h> 82 #include <sys/kernel.h> 83 #include <sys/proc.h> 84 #include <sys/buf.h> 85 #include <sys/wait.h> 86 #include <sys/file.h> 87 #include <sys/fstrans.h> 88 #include <sys/vnode.h> 89 #include <sys/syslog.h> 90 #include <sys/pool.h> 91 #include <sys/uidinfo.h> 92 #include <sys/ptrace.h> 93 #include <sys/acct.h> 94 #include <sys/filedesc.h> 95 #include <sys/ras.h> 96 #include <sys/signalvar.h> 97 #include <sys/sched.h> 98 #include <sys/mount.h> 99 #include <sys/syscallargs.h> 100 #include <sys/kauth.h> 101 #include <sys/sleepq.h> 102 #include <sys/lockdebug.h> 103 #include <sys/ktrace.h> 104 #include <sys/cpu.h> 105 #include <sys/lwpctl.h> 106 #include <sys/atomic.h> 107 #include <sys/sdt.h> 108 #include <sys/psref.h> 109 110 #include <uvm/uvm_extern.h> 111 112 #ifdef DEBUG_EXIT 113 int debug_exit = 0; 114 #define DPRINTF(x) if (debug_exit) printf x 115 #else 116 #define DPRINTF(x) 117 #endif 118 119 static int find_stopped_child(struct proc *, idtype_t, id_t, int, 120 struct proc **, struct wrusage *, siginfo_t *); 121 static void proc_free(struct proc *, struct wrusage *); 122 123 /* 124 * DTrace SDT provider definitions 125 */ 126 SDT_PROVIDER_DECLARE(proc); 127 SDT_PROBE_DEFINE1(proc, kernel, , exit, "int"); 128 129 /* 130 * Fill in the appropriate signal information, and signal the parent. 131 */ 132 /* XXX noclone works around a gcc 4.5 bug on arm */ 133 static void __noclone 134 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi) 135 { 136 137 KSI_INIT(ksi); 138 if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) { 139 if (p->p_xsig) { 140 if (p->p_sflag & PS_COREDUMP) 141 ksi->ksi_code = CLD_DUMPED; 142 else 143 ksi->ksi_code = CLD_KILLED; 144 ksi->ksi_status = p->p_xsig; 145 } else { 146 ksi->ksi_code = CLD_EXITED; 147 ksi->ksi_status = p->p_xexit; 148 } 149 } else { 150 ksi->ksi_code = SI_USER; 151 ksi->ksi_status = p->p_xsig; 152 } 153 /* 154 * We fill those in, even for non-SIGCHLD. 155 * It's safe to access p->p_cred unlocked here. 156 */ 157 ksi->ksi_pid = p->p_pid; 158 ksi->ksi_uid = kauth_cred_geteuid(p->p_cred); 159 /* XXX: is this still valid? */ 160 ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec; 161 ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec; 162 } 163 164 /* 165 * exit -- 166 * Death of process. 167 */ 168 int 169 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval) 170 { 171 /* { 172 syscallarg(int) rval; 173 } */ 174 struct proc *p = l->l_proc; 175 176 /* Don't call exit1() multiple times in the same process. */ 177 mutex_enter(p->p_lock); 178 if (p->p_sflag & PS_WEXIT) { 179 mutex_exit(p->p_lock); 180 lwp_exit(l); 181 } 182 183 /* exit1() will release the mutex. */ 184 exit1(l, SCARG(uap, rval), 0); 185 /* NOTREACHED */ 186 return (0); 187 } 188 189 /* 190 * Exit: deallocate address space and other resources, change proc state 191 * to zombie, and unlink proc from allproc and parent's lists. Save exit 192 * status and rusage for wait(). Check for child processes and orphan them. 193 * 194 * Must be called with p->p_lock held. Does not return. 195 */ 196 void 197 exit1(struct lwp *l, int exitcode, int signo) 198 { 199 struct proc *p, *child, *next_child, *old_parent, *new_parent; 200 struct pgrp *pgrp; 201 ksiginfo_t ksi; 202 ksiginfoq_t kq; 203 int wakeinit; 204 205 p = l->l_proc; 206 207 KASSERT(mutex_owned(p->p_lock)); 208 KASSERT(p->p_vmspace != NULL); 209 210 if (__predict_false(p == initproc)) { 211 panic("init died (signal %d, exit %d)", signo, exitcode); 212 } 213 214 p->p_sflag |= PS_WEXIT; 215 216 /* 217 * Force all other LWPs to exit before we do. Only then can we 218 * begin to tear down the rest of the process state. 219 */ 220 if (p->p_nlwps > 1) { 221 exit_lwps(l); 222 } 223 224 ksiginfo_queue_init(&kq); 225 226 /* 227 * If we have been asked to stop on exit, do so now. 228 */ 229 if (__predict_false(p->p_sflag & PS_STOPEXIT)) { 230 KERNEL_UNLOCK_ALL(l, &l->l_biglocks); 231 sigclearall(p, &contsigmask, &kq); 232 233 if (!mutex_tryenter(proc_lock)) { 234 mutex_exit(p->p_lock); 235 mutex_enter(proc_lock); 236 mutex_enter(p->p_lock); 237 } 238 p->p_waited = 0; 239 p->p_pptr->p_nstopchild++; 240 p->p_stat = SSTOP; 241 mutex_exit(proc_lock); 242 lwp_lock(l); 243 p->p_nrlwps--; 244 l->l_stat = LSSTOP; 245 lwp_unlock(l); 246 mutex_exit(p->p_lock); 247 lwp_lock(l); 248 spc_lock(l->l_cpu); 249 mi_switch(l); 250 KERNEL_LOCK(l->l_biglocks, l); 251 mutex_enter(p->p_lock); 252 } 253 254 /* 255 * Bin any remaining signals and mark the process as dying so it will 256 * not be found for, e.g. signals. 257 */ 258 sigfillset(&p->p_sigctx.ps_sigignore); 259 sigclearall(p, NULL, &kq); 260 p->p_stat = SDYING; 261 mutex_exit(p->p_lock); 262 ksiginfo_queue_drain(&kq); 263 264 /* Destroy any lwpctl info. */ 265 if (p->p_lwpctl != NULL) 266 lwp_ctl_exit(); 267 268 /* 269 * Drain all remaining references that procfs, ptrace and others may 270 * have on the process. 271 */ 272 rw_enter(&p->p_reflock, RW_WRITER); 273 274 DPRINTF(("%s: %d.%d exiting.\n", __func__, p->p_pid, l->l_lid)); 275 276 timers_free(p, TIMERS_ALL); 277 #if defined(__HAVE_RAS) 278 ras_purgeall(); 279 #endif 280 281 /* 282 * Close open files, release open-file table and free signal 283 * actions. This may block! 284 */ 285 fd_free(); 286 cwdfree(p->p_cwdi); 287 p->p_cwdi = NULL; 288 doexithooks(p); 289 sigactsfree(p->p_sigacts); 290 291 /* 292 * Write out accounting data. 293 */ 294 (void)acct_process(l); 295 296 #ifdef KTRACE 297 /* 298 * Release trace file. 299 */ 300 if (p->p_tracep != NULL) { 301 mutex_enter(&ktrace_lock); 302 ktrderef(p); 303 mutex_exit(&ktrace_lock); 304 } 305 #endif 306 307 p->p_xexit = exitcode; 308 p->p_xsig = signo; 309 310 /* 311 * If emulation has process exit hook, call it now. 312 * Set the exit status now so that the exit hook has 313 * an opportunity to tweak it (COMPAT_LINUX requires 314 * this for thread group emulation) 315 */ 316 if (p->p_emul->e_proc_exit) 317 (*p->p_emul->e_proc_exit)(p); 318 319 /* 320 * Free the VM resources we're still holding on to. 321 * We must do this from a valid thread because doing 322 * so may block. This frees vmspace, which we don't 323 * need anymore. The only remaining lwp is the one 324 * we run at this moment, nothing runs in userland 325 * anymore. 326 */ 327 ruspace(p); /* Update our vm resource use */ 328 uvm_proc_exit(p); 329 330 /* 331 * Stop profiling. 332 */ 333 if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) { 334 mutex_spin_enter(&p->p_stmutex); 335 stopprofclock(p); 336 mutex_spin_exit(&p->p_stmutex); 337 } 338 339 /* 340 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we 341 * wake up the parent early to avoid deadlock. We can do this once 342 * the VM resources are released. 343 */ 344 mutex_enter(proc_lock); 345 if (p->p_lflag & PL_PPWAIT) { 346 lwp_t *lp; 347 348 l->l_lwpctl = NULL; /* was on loan from blocked parent */ 349 p->p_lflag &= ~PL_PPWAIT; 350 351 lp = p->p_vforklwp; 352 p->p_vforklwp = NULL; 353 lp->l_vforkwaiting = false; 354 cv_broadcast(&lp->l_waitcv); 355 } 356 357 if (SESS_LEADER(p)) { 358 struct vnode *vprele = NULL, *vprevoke = NULL; 359 struct session *sp = p->p_session; 360 struct tty *tp; 361 362 if (sp->s_ttyvp) { 363 /* 364 * Controlling process. 365 * Signal foreground pgrp, 366 * drain controlling terminal 367 * and revoke access to controlling terminal. 368 */ 369 tp = sp->s_ttyp; 370 mutex_spin_enter(&tty_lock); 371 if (tp->t_session == sp) { 372 /* we can't guarantee the revoke will do this */ 373 pgrp = tp->t_pgrp; 374 tp->t_pgrp = NULL; 375 tp->t_session = NULL; 376 mutex_spin_exit(&tty_lock); 377 if (pgrp != NULL) { 378 pgsignal(pgrp, SIGHUP, 1); 379 } 380 mutex_exit(proc_lock); 381 (void) ttywait(tp); 382 mutex_enter(proc_lock); 383 384 /* The tty could have been revoked. */ 385 vprevoke = sp->s_ttyvp; 386 } else 387 mutex_spin_exit(&tty_lock); 388 vprele = sp->s_ttyvp; 389 sp->s_ttyvp = NULL; 390 /* 391 * s_ttyp is not zero'd; we use this to indicate 392 * that the session once had a controlling terminal. 393 * (for logging and informational purposes) 394 */ 395 } 396 sp->s_leader = NULL; 397 398 if (vprevoke != NULL || vprele != NULL) { 399 if (vprevoke != NULL) { 400 /* Releases proc_lock. */ 401 proc_sessrele(sp); 402 VOP_REVOKE(vprevoke, REVOKEALL); 403 } else 404 mutex_exit(proc_lock); 405 if (vprele != NULL) 406 vrele(vprele); 407 mutex_enter(proc_lock); 408 } 409 } 410 fixjobc(p, p->p_pgrp, 0); 411 412 /* Release fstrans private data. */ 413 fstrans_lwp_dtor(l); 414 415 /* 416 * Finalize the last LWP's specificdata, as well as the 417 * specificdata for the proc itself. 418 */ 419 lwp_finispecific(l); 420 proc_finispecific(p); 421 422 /* 423 * Notify interested parties of our demise. 424 */ 425 KNOTE(&p->p_klist, NOTE_EXIT); 426 427 SDT_PROBE(proc, kernel, , exit, 428 ((p->p_sflag & PS_COREDUMP) ? CLD_DUMPED : 429 (p->p_xsig ? CLD_KILLED : CLD_EXITED)), 430 0,0,0,0); 431 432 /* 433 * Reset p_opptr pointer of all former children which got 434 * traced by another process and were reparented. We reset 435 * it to NULL here; the trace detach code then reparents 436 * the child to initproc. We only check allproc list, since 437 * eventual former children on zombproc list won't reference 438 * p_opptr anymore. 439 */ 440 if (__predict_false(p->p_slflag & PSL_CHTRACED)) { 441 struct proc *q; 442 PROCLIST_FOREACH(q, &allproc) { 443 if (q->p_opptr == p) 444 q->p_opptr = NULL; 445 } 446 PROCLIST_FOREACH(q, &zombproc) { 447 if (q->p_opptr == p) 448 q->p_opptr = NULL; 449 } 450 } 451 452 /* 453 * Give orphaned children to init(8). 454 */ 455 child = LIST_FIRST(&p->p_children); 456 wakeinit = (child != NULL); 457 for (; child != NULL; child = next_child) { 458 next_child = LIST_NEXT(child, p_sibling); 459 460 /* 461 * Traced processes are killed since their existence 462 * means someone is screwing up. Since we reset the 463 * trace flags, the logic in sys_wait4() would not be 464 * triggered to reparent the process to its 465 * original parent, so we must do this here. 466 */ 467 if (__predict_false(child->p_slflag & PSL_TRACED)) { 468 mutex_enter(p->p_lock); 469 child->p_slflag &= 470 ~(PSL_TRACED|PSL_SYSCALL); 471 mutex_exit(p->p_lock); 472 if (child->p_opptr != child->p_pptr) { 473 struct proc *t = child->p_opptr; 474 proc_reparent(child, t ? t : initproc); 475 child->p_opptr = NULL; 476 } else 477 proc_reparent(child, initproc); 478 killproc(child, "orphaned traced process"); 479 } else 480 proc_reparent(child, initproc); 481 } 482 483 /* 484 * Move proc from allproc to zombproc, it's now nearly ready to be 485 * collected by parent. 486 */ 487 LIST_REMOVE(l, l_list); 488 LIST_REMOVE(p, p_list); 489 LIST_INSERT_HEAD(&zombproc, p, p_list); 490 491 /* 492 * Mark the process as dead. We must do this before we signal 493 * the parent. 494 */ 495 p->p_stat = SDEAD; 496 497 /* Put in front of parent's sibling list for parent to collect it */ 498 old_parent = p->p_pptr; 499 old_parent->p_nstopchild++; 500 if (LIST_FIRST(&old_parent->p_children) != p) { 501 /* Put child where it can be found quickly */ 502 LIST_REMOVE(p, p_sibling); 503 LIST_INSERT_HEAD(&old_parent->p_children, p, p_sibling); 504 } 505 506 /* 507 * Notify parent that we're gone. If parent has the P_NOCLDWAIT 508 * flag set, notify init instead (and hope it will handle 509 * this situation). 510 */ 511 if (old_parent->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) { 512 proc_reparent(p, initproc); 513 wakeinit = 1; 514 515 /* 516 * If this was the last child of our parent, notify 517 * parent, so in case he was wait(2)ing, he will 518 * continue. 519 */ 520 if (LIST_FIRST(&old_parent->p_children) == NULL) 521 cv_broadcast(&old_parent->p_waitcv); 522 } 523 524 /* Reload parent pointer, since p may have been reparented above */ 525 new_parent = p->p_pptr; 526 527 if (__predict_false(p->p_exitsig != 0)) { 528 exit_psignal(p, new_parent, &ksi); 529 kpsignal(new_parent, &ksi, NULL); 530 } 531 532 /* Calculate the final rusage info. */ 533 calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime, 534 NULL, NULL); 535 536 if (wakeinit) 537 cv_broadcast(&initproc->p_waitcv); 538 539 callout_destroy(&l->l_timeout_ch); 540 541 /* 542 * Release any PCU resources before becoming a zombie. 543 */ 544 pcu_discard_all(l); 545 546 mutex_enter(p->p_lock); 547 /* Free the linux lwp id */ 548 if ((l->l_pflag & LP_PIDLID) != 0 && l->l_lid != p->p_pid) 549 proc_free_pid(l->l_lid); 550 lwp_drainrefs(l); 551 lwp_lock(l); 552 l->l_prflag &= ~LPR_DETACHED; 553 l->l_stat = LSZOMB; 554 lwp_unlock(l); 555 KASSERT(curlwp == l); 556 KASSERT(p->p_nrlwps == 1); 557 KASSERT(p->p_nlwps == 1); 558 p->p_stat = SZOMB; 559 p->p_nrlwps--; 560 p->p_nzlwps++; 561 p->p_ndlwps = 0; 562 mutex_exit(p->p_lock); 563 564 /* 565 * Signal the parent to collect us, and drop the proclist lock. 566 * Drop debugger/procfs lock; no new references can be gained. 567 */ 568 cv_broadcast(&p->p_pptr->p_waitcv); 569 rw_exit(&p->p_reflock); 570 mutex_exit(proc_lock); 571 572 /* Verify that we hold no locks other than the kernel lock. */ 573 LOCKDEBUG_BARRIER(&kernel_lock, 0); 574 575 /* 576 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP! 577 */ 578 579 /* 580 * Give machine-dependent code a chance to free any MD LWP 581 * resources. This must be done before uvm_lwp_exit(), in 582 * case these resources are in the PCB. 583 */ 584 cpu_lwp_free(l, 1); 585 586 pmap_deactivate(l); 587 588 /* This process no longer needs to hold the kernel lock. */ 589 #ifdef notyet 590 /* XXXSMP hold in lwp_userret() */ 591 KERNEL_UNLOCK_LAST(l); 592 #else 593 KERNEL_UNLOCK_ALL(l, NULL); 594 #endif 595 596 lwp_exit_switchaway(l); 597 } 598 599 void 600 exit_lwps(struct lwp *l) 601 { 602 proc_t *p = l->l_proc; 603 lwp_t *l2; 604 int nlocks; 605 606 KERNEL_UNLOCK_ALL(l, &nlocks); 607 retry: 608 KASSERT(mutex_owned(p->p_lock)); 609 610 /* 611 * Interrupt LWPs in interruptable sleep, unsuspend suspended 612 * LWPs and then wait for everyone else to finish. 613 */ 614 LIST_FOREACH(l2, &p->p_lwps, l_sibling) { 615 if (l2 == l) 616 continue; 617 lwp_lock(l2); 618 l2->l_flag |= LW_WEXIT; 619 if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) || 620 l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) { 621 l2->l_flag &= ~LW_DBGSUSPEND; 622 /* setrunnable() will release the lock. */ 623 setrunnable(l2); 624 continue; 625 } 626 lwp_unlock(l2); 627 } 628 629 /* 630 * Wait for every LWP to exit. Note: LWPs can get suspended/slept 631 * behind us or there may even be new LWPs created. Therefore, a 632 * full retry is required on error. 633 */ 634 while (p->p_nlwps > 1) { 635 if (lwp_wait(l, 0, NULL, true)) { 636 goto retry; 637 } 638 } 639 640 KERNEL_LOCK(nlocks, l); 641 KASSERT(p->p_nlwps == 1); 642 } 643 644 int 645 do_sys_waitid(idtype_t idtype, id_t id, int *pid, int *status, int options, 646 struct wrusage *wru, siginfo_t *si) 647 { 648 proc_t *child; 649 int error; 650 651 652 if (wru != NULL) 653 memset(wru, 0, sizeof(*wru)); 654 if (si != NULL) 655 memset(si, 0, sizeof(*si)); 656 657 mutex_enter(proc_lock); 658 error = find_stopped_child(curproc, idtype, id, options, &child, 659 wru, si); 660 if (child == NULL) { 661 mutex_exit(proc_lock); 662 *pid = 0; 663 *status = 0; 664 return error; 665 } 666 *pid = child->p_pid; 667 668 if (child->p_stat == SZOMB) { 669 /* Child is exiting */ 670 *status = P_WAITSTATUS(child); 671 /* proc_free() will release the proc_lock. */ 672 if (options & WNOWAIT) { 673 mutex_exit(proc_lock); 674 } else { 675 proc_free(child, wru); 676 } 677 } else { 678 /* Don't mark SIGCONT if we are being stopped */ 679 *status = (child->p_xsig == SIGCONT && child->p_stat != SSTOP) ? 680 W_CONTCODE() : W_STOPCODE(child->p_xsig); 681 mutex_exit(proc_lock); 682 } 683 return 0; 684 } 685 686 int 687 do_sys_wait(int *pid, int *status, int options, struct rusage *ru) 688 { 689 idtype_t idtype; 690 id_t id; 691 int ret; 692 struct wrusage wru; 693 694 /* 695 * Translate the special pid values into the (idtype, pid) 696 * pair for wait6. The WAIT_MYPGRP case is handled by 697 * find_stopped_child() on its own. 698 */ 699 if (*pid == WAIT_ANY) { 700 idtype = P_ALL; 701 id = 0; 702 } else if (*pid < 0) { 703 idtype = P_PGID; 704 id = (id_t)-*pid; 705 } else { 706 idtype = P_PID; 707 id = (id_t)*pid; 708 } 709 options |= WEXITED | WTRAPPED; 710 ret = do_sys_waitid(idtype, id, pid, status, options, ru ? &wru : NULL, 711 NULL); 712 if (ru) 713 *ru = wru.wru_self; 714 return ret; 715 } 716 717 int 718 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap, 719 register_t *retval) 720 { 721 /* { 722 syscallarg(int) pid; 723 syscallarg(int *) status; 724 syscallarg(int) options; 725 syscallarg(struct rusage *) rusage; 726 } */ 727 int error, status, pid = SCARG(uap, pid); 728 struct rusage ru; 729 730 error = do_sys_wait(&pid, &status, SCARG(uap, options), 731 SCARG(uap, rusage) != NULL ? &ru : NULL); 732 733 retval[0] = pid; 734 if (pid == 0) { 735 return error; 736 } 737 if (SCARG(uap, status)) { 738 error = copyout(&status, SCARG(uap, status), sizeof(status)); 739 } 740 if (SCARG(uap, rusage) && error == 0) { 741 error = copyout(&ru, SCARG(uap, rusage), sizeof(ru)); 742 } 743 return error; 744 } 745 746 int 747 sys_wait6(struct lwp *l, const struct sys_wait6_args *uap, register_t *retval) 748 { 749 /* { 750 syscallarg(idtype_t) idtype; 751 syscallarg(id_t) id; 752 syscallarg(int *) status; 753 syscallarg(int) options; 754 syscallarg(struct wrusage *) wru; 755 syscallarg(siginfo_t *) si; 756 } */ 757 struct wrusage wru, *wrup; 758 siginfo_t si, *sip; 759 idtype_t idtype; 760 int pid; 761 id_t id; 762 int error, status; 763 764 idtype = SCARG(uap, idtype); 765 id = SCARG(uap, id); 766 767 if (SCARG(uap, wru) != NULL) 768 wrup = &wru; 769 else 770 wrup = NULL; 771 772 if (SCARG(uap, info) != NULL) 773 sip = &si; 774 else 775 sip = NULL; 776 777 /* 778 * We expect all callers of wait6() to know about WEXITED and 779 * WTRAPPED. 780 */ 781 error = do_sys_waitid(idtype, id, &pid, &status, SCARG(uap, options), 782 wrup, sip); 783 784 retval[0] = pid; /* tell userland who it was */ 785 786 #if 0 787 /* 788 * should we copyout if there was no process, hence no useful data? 789 * We don't for an old sytle wait4() (etc) but I believe 790 * FreeBSD does for wait6(), so a tossup... Go with FreeBSD for now. 791 */ 792 if (pid == 0) 793 return error; 794 #endif 795 796 if (SCARG(uap, status) != NULL && error == 0) 797 error = copyout(&status, SCARG(uap, status), sizeof(status)); 798 if (SCARG(uap, wru) != NULL && error == 0) 799 error = copyout(&wru, SCARG(uap, wru), sizeof(wru)); 800 if (SCARG(uap, info) != NULL && error == 0) 801 error = copyout(&si, SCARG(uap, info), sizeof(si)); 802 return error; 803 } 804 805 806 /* 807 * Find a process that matches the provided criteria, and fill siginfo 808 * and resources if found. 809 * Returns: 810 * -1: Not found, abort early 811 * 0: Not matched 812 * 1: Matched, there might be more matches 813 * 2: This is the only match 814 */ 815 static int 816 match_process(const struct proc *pp, struct proc **q, idtype_t idtype, id_t id, 817 int options, struct wrusage *wrusage, siginfo_t *siginfo) 818 { 819 struct rusage *rup; 820 struct proc *p = *q; 821 int rv = 1; 822 823 mutex_enter(p->p_lock); 824 switch (idtype) { 825 case P_ALL: 826 break; 827 case P_PID: 828 if (p->p_pid != (pid_t)id) { 829 mutex_exit(p->p_lock); 830 p = *q = proc_find_raw((pid_t)id); 831 if (p == NULL || p->p_stat == SIDL || p->p_pptr != pp) { 832 *q = NULL; 833 return -1; 834 } 835 mutex_enter(p->p_lock); 836 } 837 rv++; 838 break; 839 case P_PGID: 840 if (p->p_pgid != (pid_t)id) 841 goto out; 842 break; 843 case P_SID: 844 if (p->p_session->s_sid != (pid_t)id) 845 goto out; 846 break; 847 case P_UID: 848 if (kauth_cred_geteuid(p->p_cred) != (uid_t)id) 849 goto out; 850 break; 851 case P_GID: 852 if (kauth_cred_getegid(p->p_cred) != (gid_t)id) 853 goto out; 854 break; 855 case P_CID: 856 case P_PSETID: 857 case P_CPUID: 858 /* XXX: Implement me */ 859 default: 860 out: 861 mutex_exit(p->p_lock); 862 return 0; 863 } 864 865 if ((options & WEXITED) == 0 && p->p_stat == SZOMB) 866 goto out; 867 868 if (siginfo != NULL) { 869 siginfo->si_errno = 0; 870 871 /* 872 * SUSv4 requires that the si_signo value is always 873 * SIGCHLD. Obey it despite the rfork(2) interface 874 * allows to request other signal for child exit 875 * notification. 876 */ 877 siginfo->si_signo = SIGCHLD; 878 879 /* 880 * This is still a rough estimate. We will fix the 881 * cases TRAPPED, STOPPED, and CONTINUED later. 882 */ 883 if (p->p_sflag & PS_COREDUMP) { 884 siginfo->si_code = CLD_DUMPED; 885 siginfo->si_status = p->p_xsig; 886 } else if (p->p_xsig) { 887 siginfo->si_code = CLD_KILLED; 888 siginfo->si_status = p->p_xsig; 889 } else { 890 siginfo->si_code = CLD_EXITED; 891 siginfo->si_status = p->p_xexit; 892 } 893 894 siginfo->si_pid = p->p_pid; 895 siginfo->si_uid = kauth_cred_geteuid(p->p_cred); 896 siginfo->si_utime = p->p_stats->p_ru.ru_utime.tv_sec; 897 siginfo->si_stime = p->p_stats->p_ru.ru_stime.tv_sec; 898 } 899 900 /* 901 * There should be no reason to limit resources usage info to 902 * exited processes only. A snapshot about any resources used 903 * by a stopped process may be exactly what is needed. 904 */ 905 if (wrusage != NULL) { 906 rup = &wrusage->wru_self; 907 *rup = p->p_stats->p_ru; 908 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 909 910 rup = &wrusage->wru_children; 911 *rup = p->p_stats->p_cru; 912 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 913 } 914 915 mutex_exit(p->p_lock); 916 return rv; 917 } 918 919 /* 920 * Determine if there are existing processes being debugged 921 * that used to be (and sometime later will be again) children 922 * of a specific parent (while matching wait criteria) 923 */ 924 static bool 925 debugged_child_exists(idtype_t idtype, id_t id, int options, siginfo_t *si, 926 const struct proc *parent) 927 { 928 struct proc *pp; 929 930 /* 931 * If we are searching for a specific pid, we can optimise a little 932 */ 933 if (idtype == P_PID) { 934 /* 935 * Check the specific process to see if its real parent is us 936 */ 937 pp = proc_find_raw((pid_t)id); 938 if (pp != NULL && pp->p_stat != SIDL && pp->p_opptr == parent) { 939 /* 940 * using P_ALL here avoids match_process() doing the 941 * same work that we just did, but incorrectly for 942 * this scenario. 943 */ 944 if (match_process(parent, &pp, P_ALL, id, options, 945 NULL, si)) 946 return true; 947 } 948 return false; 949 } 950 951 /* 952 * For the hard cases, just look everywhere to see if some 953 * stolen (reparented) process is really our lost child. 954 * Then check if that process could satisfy the wait conditions. 955 */ 956 957 /* 958 * XXX inefficient, but hopefully fairly rare. 959 * XXX should really use a list of reparented processes. 960 */ 961 PROCLIST_FOREACH(pp, &allproc) { 962 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 963 continue; 964 if (pp->p_opptr == parent && 965 match_process(parent, &pp, idtype, id, options, NULL, si)) 966 return true; 967 } 968 PROCLIST_FOREACH(pp, &zombproc) { 969 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 970 continue; 971 if (pp->p_opptr == parent && 972 match_process(parent, &pp, idtype, id, options, NULL, si)) 973 return true; 974 } 975 976 return false; 977 } 978 979 /* 980 * Scan list of child processes for a child process that has stopped or 981 * exited. Used by sys_wait4 and 'compat' equivalents. 982 * 983 * Must be called with the proc_lock held, and may release while waiting. 984 */ 985 static int 986 find_stopped_child(struct proc *parent, idtype_t idtype, id_t id, int options, 987 struct proc **child_p, struct wrusage *wru, siginfo_t *si) 988 { 989 struct proc *child, *dead; 990 int error; 991 992 KASSERT(mutex_owned(proc_lock)); 993 994 if (options & ~WALLOPTS) { 995 *child_p = NULL; 996 return EINVAL; 997 } 998 999 if ((options & WSELECTOPTS) == 0) { 1000 /* 1001 * We will be unable to find any matching processes, 1002 * because there are no known events to look for. 1003 * Prefer to return error instead of blocking 1004 * indefinitely. 1005 */ 1006 *child_p = NULL; 1007 return EINVAL; 1008 } 1009 1010 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1011 mutex_enter(parent->p_lock); 1012 id = (id_t)parent->p_pgid; 1013 mutex_exit(parent->p_lock); 1014 idtype = P_PGID; 1015 } 1016 1017 for (;;) { 1018 error = ECHILD; 1019 dead = NULL; 1020 1021 LIST_FOREACH(child, &parent->p_children, p_sibling) { 1022 int rv = match_process(parent, &child, idtype, id, 1023 options, wru, si); 1024 if (rv == -1) 1025 break; 1026 if (rv == 0) 1027 continue; 1028 1029 /* 1030 * Wait for processes with p_exitsig != SIGCHLD 1031 * processes only if WALTSIG is set; wait for 1032 * processes with p_exitsig == SIGCHLD only 1033 * if WALTSIG is clear. 1034 */ 1035 if (((options & WALLSIG) == 0) && 1036 (options & WALTSIG ? child->p_exitsig == SIGCHLD 1037 : P_EXITSIG(child) != SIGCHLD)){ 1038 if (rv == 2) { 1039 child = NULL; 1040 break; 1041 } 1042 continue; 1043 } 1044 1045 error = 0; 1046 if ((options & WNOZOMBIE) == 0) { 1047 if (child->p_stat == SZOMB) 1048 break; 1049 if (child->p_stat == SDEAD) { 1050 /* 1051 * We may occasionally arrive here 1052 * after receiving a signal, but 1053 * immediately before the child 1054 * process is zombified. The wait 1055 * will be short, so avoid returning 1056 * to userspace. 1057 */ 1058 dead = child; 1059 } 1060 } 1061 1062 if ((options & WCONTINUED) != 0 && 1063 child->p_xsig == SIGCONT && 1064 (child->p_sflag & PS_CONTINUED)) { 1065 if ((options & WNOWAIT) == 0) { 1066 child->p_sflag &= ~PS_CONTINUED; 1067 child->p_waited = 1; 1068 parent->p_nstopchild--; 1069 } 1070 if (si) { 1071 si->si_status = child->p_xsig; 1072 si->si_code = CLD_CONTINUED; 1073 } 1074 break; 1075 } 1076 1077 if ((options & (WTRAPPED|WSTOPPED)) != 0 && 1078 child->p_stat == SSTOP && 1079 child->p_waited == 0 && 1080 ((child->p_slflag & PSL_TRACED) || 1081 options & (WUNTRACED|WSTOPPED))) { 1082 if ((options & WNOWAIT) == 0) { 1083 child->p_waited = 1; 1084 parent->p_nstopchild--; 1085 } 1086 if (si) { 1087 si->si_status = child->p_xsig; 1088 si->si_code = 1089 (child->p_slflag & PSL_TRACED) ? 1090 CLD_TRAPPED : CLD_STOPPED; 1091 } 1092 break; 1093 } 1094 if (parent->p_nstopchild == 0 || rv == 2) { 1095 child = NULL; 1096 break; 1097 } 1098 } 1099 1100 /* 1101 * If we found nothing, but we are the bereaved parent 1102 * of a stolen child, look and see if that child (or 1103 * one of them) meets our search criteria. If so, then 1104 * we cannot succeed, but we can hang (wait...), 1105 * or if WNOHANG, return 0 instead of ECHILD 1106 */ 1107 if (child == NULL && error == ECHILD && 1108 (parent->p_slflag & PSL_CHTRACED) && 1109 debugged_child_exists(idtype, id, options, si, parent)) 1110 error = 0; 1111 1112 if (child != NULL || error != 0 || 1113 ((options & WNOHANG) != 0 && dead == NULL)) { 1114 *child_p = child; 1115 return error; 1116 } 1117 1118 /* 1119 * Wait for another child process to stop. 1120 */ 1121 error = cv_wait_sig(&parent->p_waitcv, proc_lock); 1122 1123 if (error != 0) { 1124 *child_p = NULL; 1125 return error; 1126 } 1127 } 1128 } 1129 1130 /* 1131 * Free a process after parent has taken all the state info. Must be called 1132 * with the proclist lock held, and will release before returning. 1133 * 1134 * *ru is returned to the caller, and must be freed by the caller. 1135 */ 1136 static void 1137 proc_free(struct proc *p, struct wrusage *wru) 1138 { 1139 struct proc *parent = p->p_pptr; 1140 struct lwp *l; 1141 ksiginfo_t ksi; 1142 kauth_cred_t cred1, cred2; 1143 uid_t uid; 1144 1145 KASSERT(mutex_owned(proc_lock)); 1146 KASSERT(p->p_nlwps == 1); 1147 KASSERT(p->p_nzlwps == 1); 1148 KASSERT(p->p_nrlwps == 0); 1149 KASSERT(p->p_stat == SZOMB); 1150 1151 /* 1152 * If we got the child via ptrace(2) or procfs, and 1153 * the parent is different (meaning the process was 1154 * attached, rather than run as a child), then we need 1155 * to give it back to the old parent, and send the 1156 * parent the exit signal. The rest of the cleanup 1157 * will be done when the old parent waits on the child. 1158 */ 1159 if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) { 1160 mutex_enter(p->p_lock); 1161 p->p_slflag &= ~(PSL_TRACED|PSL_SYSCALL); 1162 mutex_exit(p->p_lock); 1163 parent = (p->p_opptr == NULL) ? initproc : p->p_opptr; 1164 proc_reparent(p, parent); 1165 p->p_opptr = NULL; 1166 if (p->p_exitsig != 0) { 1167 exit_psignal(p, parent, &ksi); 1168 kpsignal(parent, &ksi, NULL); 1169 } 1170 cv_broadcast(&parent->p_waitcv); 1171 mutex_exit(proc_lock); 1172 return; 1173 } 1174 1175 sched_proc_exit(parent, p); 1176 1177 /* 1178 * Add child times of exiting process onto its own times. 1179 * This cannot be done any earlier else it might get done twice. 1180 */ 1181 l = LIST_FIRST(&p->p_lwps); 1182 p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw); 1183 p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw; 1184 ruadd(&p->p_stats->p_ru, &l->l_ru); 1185 ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru); 1186 ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru); 1187 if (wru != NULL) { 1188 wru->wru_self = p->p_stats->p_ru; 1189 wru->wru_children = p->p_stats->p_cru; 1190 } 1191 p->p_xsig = 0; 1192 p->p_xexit = 0; 1193 1194 /* 1195 * At this point we are going to start freeing the final resources. 1196 * If anyone tries to access the proc structure after here they will 1197 * get a shock - bits are missing. Attempt to make it hard! We 1198 * don't bother with any further locking past this point. 1199 */ 1200 p->p_stat = SIDL; /* not even a zombie any more */ 1201 LIST_REMOVE(p, p_list); /* off zombproc */ 1202 parent->p_nstopchild--; 1203 LIST_REMOVE(p, p_sibling); 1204 1205 /* 1206 * Let pid be reallocated. 1207 */ 1208 proc_free_pid(p->p_pid); 1209 1210 /* 1211 * Unlink process from its process group. 1212 * Releases the proc_lock. 1213 */ 1214 proc_leavepgrp(p); 1215 1216 /* 1217 * Delay release until after lwp_free. 1218 */ 1219 cred2 = l->l_cred; 1220 1221 /* 1222 * Free the last LWP's resources. 1223 * 1224 * lwp_free ensures the LWP is no longer running on another CPU. 1225 */ 1226 lwp_free(l, false, true); 1227 1228 /* 1229 * Now no one except us can reach the process p. 1230 */ 1231 1232 /* 1233 * Decrement the count of procs running with this uid. 1234 */ 1235 cred1 = p->p_cred; 1236 uid = kauth_cred_getuid(cred1); 1237 (void)chgproccnt(uid, -1); 1238 1239 /* 1240 * Release substructures. 1241 */ 1242 1243 lim_free(p->p_limit); 1244 pstatsfree(p->p_stats); 1245 kauth_cred_free(cred1); 1246 kauth_cred_free(cred2); 1247 1248 /* 1249 * Release reference to text vnode 1250 */ 1251 if (p->p_textvp) 1252 vrele(p->p_textvp); 1253 kmem_strfree(p->p_path); 1254 1255 mutex_destroy(&p->p_auxlock); 1256 mutex_obj_free(p->p_lock); 1257 mutex_destroy(&p->p_stmutex); 1258 cv_destroy(&p->p_waitcv); 1259 cv_destroy(&p->p_lwpcv); 1260 rw_destroy(&p->p_reflock); 1261 1262 proc_free_mem(p); 1263 } 1264 1265 /* 1266 * Change the parent of a process for tracing purposes. 1267 */ 1268 void 1269 proc_changeparent(struct proc *t, struct proc *p) 1270 { 1271 SET(t->p_slflag, PSL_TRACED); 1272 t->p_opptr = t->p_pptr; 1273 if (t->p_pptr == p) 1274 return; 1275 struct proc *parent = t->p_pptr; 1276 1277 if (parent->p_lock < t->p_lock) { 1278 if (!mutex_tryenter(parent->p_lock)) { 1279 mutex_exit(t->p_lock); 1280 mutex_enter(parent->p_lock); 1281 mutex_enter(t->p_lock); 1282 } 1283 } else if (parent->p_lock > t->p_lock) { 1284 mutex_enter(parent->p_lock); 1285 } 1286 parent->p_slflag |= PSL_CHTRACED; 1287 proc_reparent(t, p); 1288 if (parent->p_lock != t->p_lock) 1289 mutex_exit(parent->p_lock); 1290 } 1291 1292 /* 1293 * make process 'parent' the new parent of process 'child'. 1294 * 1295 * Must be called with proc_lock held. 1296 */ 1297 void 1298 proc_reparent(struct proc *child, struct proc *parent) 1299 { 1300 1301 KASSERT(mutex_owned(proc_lock)); 1302 1303 if (child->p_pptr == parent) 1304 return; 1305 1306 if (child->p_stat == SZOMB || child->p_stat == SDEAD || 1307 (child->p_stat == SSTOP && !child->p_waited)) { 1308 child->p_pptr->p_nstopchild--; 1309 parent->p_nstopchild++; 1310 } 1311 if (parent == initproc) { 1312 child->p_exitsig = SIGCHLD; 1313 child->p_ppid = parent->p_pid; 1314 } 1315 1316 LIST_REMOVE(child, p_sibling); 1317 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1318 child->p_pptr = parent; 1319 } 1320