1 /* $NetBSD: kern_exit.c,v 1.275 2019/05/17 03:34:26 ozaki-r 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 1.275 2019/05/17 03:34:26 ozaki-r 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 mi_switch(l); 249 KERNEL_LOCK(l->l_biglocks, l); 250 mutex_enter(p->p_lock); 251 } 252 253 /* 254 * Bin any remaining signals and mark the process as dying so it will 255 * not be found for, e.g. signals. 256 */ 257 sigfillset(&p->p_sigctx.ps_sigignore); 258 sigclearall(p, NULL, &kq); 259 p->p_stat = SDYING; 260 mutex_exit(p->p_lock); 261 ksiginfo_queue_drain(&kq); 262 263 /* Destroy any lwpctl info. */ 264 if (p->p_lwpctl != NULL) 265 lwp_ctl_exit(); 266 267 /* 268 * Drain all remaining references that procfs, ptrace and others may 269 * have on the process. 270 */ 271 rw_enter(&p->p_reflock, RW_WRITER); 272 273 DPRINTF(("%s: %d.%d exiting.\n", __func__, p->p_pid, l->l_lid)); 274 275 timers_free(p, TIMERS_ALL); 276 #if defined(__HAVE_RAS) 277 ras_purgeall(); 278 #endif 279 280 /* 281 * Close open files, release open-file table and free signal 282 * actions. This may block! 283 */ 284 fd_free(); 285 cwdfree(p->p_cwdi); 286 p->p_cwdi = NULL; 287 doexithooks(p); 288 sigactsfree(p->p_sigacts); 289 290 /* 291 * Write out accounting data. 292 */ 293 (void)acct_process(l); 294 295 #ifdef KTRACE 296 /* 297 * Release trace file. 298 */ 299 if (p->p_tracep != NULL) { 300 mutex_enter(&ktrace_lock); 301 ktrderef(p); 302 mutex_exit(&ktrace_lock); 303 } 304 #endif 305 306 p->p_xexit = exitcode; 307 p->p_xsig = signo; 308 309 /* 310 * If emulation has process exit hook, call it now. 311 * Set the exit status now so that the exit hook has 312 * an opportunity to tweak it (COMPAT_LINUX requires 313 * this for thread group emulation) 314 */ 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 ruspace(p); /* Update our vm resource use */ 327 uvm_proc_exit(p); 328 329 /* 330 * Stop profiling. 331 */ 332 if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) { 333 mutex_spin_enter(&p->p_stmutex); 334 stopprofclock(p); 335 mutex_spin_exit(&p->p_stmutex); 336 } 337 338 /* 339 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we 340 * wake up the parent early to avoid deadlock. We can do this once 341 * the VM resources are released. 342 */ 343 mutex_enter(proc_lock); 344 if (p->p_lflag & PL_PPWAIT) { 345 l->l_lwpctl = NULL; /* was on loan from blocked parent */ 346 p->p_lflag &= ~PL_PPWAIT; 347 cv_broadcast(&p->p_pptr->p_waitcv); 348 } 349 350 if (SESS_LEADER(p)) { 351 struct vnode *vprele = NULL, *vprevoke = NULL; 352 struct session *sp = p->p_session; 353 struct tty *tp; 354 355 if (sp->s_ttyvp) { 356 /* 357 * Controlling process. 358 * Signal foreground pgrp, 359 * drain controlling terminal 360 * and revoke access to controlling terminal. 361 */ 362 tp = sp->s_ttyp; 363 mutex_spin_enter(&tty_lock); 364 if (tp->t_session == sp) { 365 /* we can't guarantee the revoke will do this */ 366 pgrp = tp->t_pgrp; 367 tp->t_pgrp = NULL; 368 tp->t_session = NULL; 369 mutex_spin_exit(&tty_lock); 370 if (pgrp != NULL) { 371 pgsignal(pgrp, SIGHUP, 1); 372 } 373 mutex_exit(proc_lock); 374 (void) ttywait(tp); 375 mutex_enter(proc_lock); 376 377 /* The tty could have been revoked. */ 378 vprevoke = sp->s_ttyvp; 379 } else 380 mutex_spin_exit(&tty_lock); 381 vprele = sp->s_ttyvp; 382 sp->s_ttyvp = NULL; 383 /* 384 * s_ttyp is not zero'd; we use this to indicate 385 * that the session once had a controlling terminal. 386 * (for logging and informational purposes) 387 */ 388 } 389 sp->s_leader = NULL; 390 391 if (vprevoke != NULL || vprele != NULL) { 392 if (vprevoke != NULL) { 393 /* Releases proc_lock. */ 394 proc_sessrele(sp); 395 VOP_REVOKE(vprevoke, REVOKEALL); 396 } else 397 mutex_exit(proc_lock); 398 if (vprele != NULL) 399 vrele(vprele); 400 mutex_enter(proc_lock); 401 } 402 } 403 fixjobc(p, p->p_pgrp, 0); 404 405 /* Release fstrans private data. */ 406 fstrans_lwp_dtor(l); 407 408 /* 409 * Finalize the last LWP's specificdata, as well as the 410 * specificdata for the proc itself. 411 */ 412 lwp_finispecific(l); 413 proc_finispecific(p); 414 415 /* 416 * Notify interested parties of our demise. 417 */ 418 KNOTE(&p->p_klist, NOTE_EXIT); 419 420 SDT_PROBE(proc, kernel, , exit, 421 ((p->p_sflag & PS_COREDUMP) ? CLD_DUMPED : 422 (p->p_xsig ? CLD_KILLED : CLD_EXITED)), 423 0,0,0,0); 424 425 /* 426 * Reset p_opptr pointer of all former children which got 427 * traced by another process and were reparented. We reset 428 * it to NULL here; the trace detach code then reparents 429 * the child to initproc. We only check allproc list, since 430 * eventual former children on zombproc list won't reference 431 * p_opptr anymore. 432 */ 433 if (__predict_false(p->p_slflag & PSL_CHTRACED)) { 434 struct proc *q; 435 PROCLIST_FOREACH(q, &allproc) { 436 if (q->p_opptr == p) 437 q->p_opptr = NULL; 438 } 439 PROCLIST_FOREACH(q, &zombproc) { 440 if (q->p_opptr == p) 441 q->p_opptr = NULL; 442 } 443 } 444 445 /* 446 * Give orphaned children to init(8). 447 */ 448 child = LIST_FIRST(&p->p_children); 449 wakeinit = (child != NULL); 450 for (; child != NULL; child = next_child) { 451 next_child = LIST_NEXT(child, p_sibling); 452 453 /* 454 * Traced processes are killed since their existence 455 * means someone is screwing up. Since we reset the 456 * trace flags, the logic in sys_wait4() would not be 457 * triggered to reparent the process to its 458 * original parent, so we must do this here. 459 */ 460 if (__predict_false(child->p_slflag & PSL_TRACED)) { 461 mutex_enter(p->p_lock); 462 child->p_slflag &= 463 ~(PSL_TRACED|PSL_SYSCALL); 464 mutex_exit(p->p_lock); 465 if (child->p_opptr != child->p_pptr) { 466 struct proc *t = child->p_opptr; 467 proc_reparent(child, t ? t : initproc); 468 child->p_opptr = NULL; 469 } else 470 proc_reparent(child, initproc); 471 killproc(child, "orphaned traced process"); 472 } else 473 proc_reparent(child, 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 old_parent = p->p_pptr; 492 old_parent->p_nstopchild++; 493 if (LIST_FIRST(&old_parent->p_children) != p) { 494 /* Put child where it can be found quickly */ 495 LIST_REMOVE(p, p_sibling); 496 LIST_INSERT_HEAD(&old_parent->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 (old_parent->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(&old_parent->p_children) == NULL) 514 cv_broadcast(&old_parent->p_waitcv); 515 } 516 517 /* Reload parent pointer, since p may have been reparented above */ 518 new_parent = p->p_pptr; 519 520 if (__predict_false(p->p_exitsig != 0)) { 521 exit_psignal(p, new_parent, &ksi); 522 kpsignal(new_parent, &ksi, NULL); 523 } 524 525 /* Calculate the final rusage info. */ 526 calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime, 527 NULL, NULL); 528 529 if (wakeinit) 530 cv_broadcast(&initproc->p_waitcv); 531 532 callout_destroy(&l->l_timeout_ch); 533 534 /* 535 * Release any PCU resources before becoming a zombie. 536 */ 537 pcu_discard_all(l); 538 539 mutex_enter(p->p_lock); 540 /* Free the linux lwp id */ 541 if ((l->l_pflag & LP_PIDLID) != 0 && l->l_lid != p->p_pid) 542 proc_free_pid(l->l_lid); 543 lwp_drainrefs(l); 544 lwp_lock(l); 545 l->l_prflag &= ~LPR_DETACHED; 546 l->l_stat = LSZOMB; 547 lwp_unlock(l); 548 KASSERT(curlwp == l); 549 KASSERT(p->p_nrlwps == 1); 550 KASSERT(p->p_nlwps == 1); 551 p->p_stat = SZOMB; 552 p->p_nrlwps--; 553 p->p_nzlwps++; 554 p->p_ndlwps = 0; 555 mutex_exit(p->p_lock); 556 557 /* 558 * Signal the parent to collect us, and drop the proclist lock. 559 * Drop debugger/procfs lock; no new references can be gained. 560 */ 561 cv_broadcast(&p->p_pptr->p_waitcv); 562 rw_exit(&p->p_reflock); 563 mutex_exit(proc_lock); 564 565 /* Verify that we hold no locks other than the kernel lock. */ 566 LOCKDEBUG_BARRIER(&kernel_lock, 0); 567 568 /* 569 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP! 570 */ 571 572 /* 573 * Give machine-dependent code a chance to free any MD LWP 574 * resources. This must be done before uvm_lwp_exit(), in 575 * case these resources are in the PCB. 576 */ 577 cpu_lwp_free(l, 1); 578 579 pmap_deactivate(l); 580 581 /* This process no longer needs to hold the kernel lock. */ 582 #ifdef notyet 583 /* XXXSMP hold in lwp_userret() */ 584 KERNEL_UNLOCK_LAST(l); 585 #else 586 KERNEL_UNLOCK_ALL(l, NULL); 587 #endif 588 589 lwp_exit_switchaway(l); 590 } 591 592 void 593 exit_lwps(struct lwp *l) 594 { 595 proc_t *p = l->l_proc; 596 lwp_t *l2; 597 int nlocks; 598 599 KERNEL_UNLOCK_ALL(l, &nlocks); 600 retry: 601 KASSERT(mutex_owned(p->p_lock)); 602 603 /* 604 * Interrupt LWPs in interruptable sleep, unsuspend suspended 605 * LWPs and then wait for everyone else to finish. 606 */ 607 LIST_FOREACH(l2, &p->p_lwps, l_sibling) { 608 if (l2 == l) 609 continue; 610 lwp_lock(l2); 611 l2->l_flag |= LW_WEXIT; 612 if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) || 613 l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) { 614 /* setrunnable() will release the lock. */ 615 setrunnable(l2); 616 continue; 617 } 618 lwp_unlock(l2); 619 } 620 621 /* 622 * Wait for every LWP to exit. Note: LWPs can get suspended/slept 623 * behind us or there may even be new LWPs created. Therefore, a 624 * full retry is required on error. 625 */ 626 while (p->p_nlwps > 1) { 627 if (lwp_wait(l, 0, NULL, true)) { 628 goto retry; 629 } 630 } 631 632 KERNEL_LOCK(nlocks, l); 633 KASSERT(p->p_nlwps == 1); 634 } 635 636 int 637 do_sys_waitid(idtype_t idtype, id_t id, int *pid, int *status, int options, 638 struct wrusage *wru, siginfo_t *si) 639 { 640 proc_t *child; 641 int error; 642 643 644 if (wru != NULL) 645 memset(wru, 0, sizeof(*wru)); 646 if (si != NULL) 647 memset(si, 0, sizeof(*si)); 648 649 mutex_enter(proc_lock); 650 error = find_stopped_child(curproc, idtype, id, options, &child, 651 wru, si); 652 if (child == NULL) { 653 mutex_exit(proc_lock); 654 *pid = 0; 655 *status = 0; 656 return error; 657 } 658 *pid = child->p_pid; 659 660 if (child->p_stat == SZOMB) { 661 /* Child is exiting */ 662 *status = P_WAITSTATUS(child); 663 /* proc_free() will release the proc_lock. */ 664 if (options & WNOWAIT) { 665 mutex_exit(proc_lock); 666 } else { 667 proc_free(child, wru); 668 } 669 } else { 670 /* Don't mark SIGCONT if we are being stopped */ 671 *status = (child->p_xsig == SIGCONT && child->p_stat != SSTOP) ? 672 W_CONTCODE() : W_STOPCODE(child->p_xsig); 673 mutex_exit(proc_lock); 674 } 675 return 0; 676 } 677 678 int 679 do_sys_wait(int *pid, int *status, int options, struct rusage *ru) 680 { 681 idtype_t idtype; 682 id_t id; 683 int ret; 684 struct wrusage wru; 685 686 /* 687 * Translate the special pid values into the (idtype, pid) 688 * pair for wait6. The WAIT_MYPGRP case is handled by 689 * find_stopped_child() on its own. 690 */ 691 if (*pid == WAIT_ANY) { 692 idtype = P_ALL; 693 id = 0; 694 } else if (*pid < 0) { 695 idtype = P_PGID; 696 id = (id_t)-*pid; 697 } else { 698 idtype = P_PID; 699 id = (id_t)*pid; 700 } 701 options |= WEXITED | WTRAPPED; 702 ret = do_sys_waitid(idtype, id, pid, status, options, ru ? &wru : NULL, 703 NULL); 704 if (ru) 705 *ru = wru.wru_self; 706 return ret; 707 } 708 709 int 710 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap, 711 register_t *retval) 712 { 713 /* { 714 syscallarg(int) pid; 715 syscallarg(int *) status; 716 syscallarg(int) options; 717 syscallarg(struct rusage *) rusage; 718 } */ 719 int error, status, pid = SCARG(uap, pid); 720 struct rusage ru; 721 722 error = do_sys_wait(&pid, &status, SCARG(uap, options), 723 SCARG(uap, rusage) != NULL ? &ru : NULL); 724 725 retval[0] = pid; 726 if (pid == 0) { 727 return error; 728 } 729 if (SCARG(uap, status)) { 730 error = copyout(&status, SCARG(uap, status), sizeof(status)); 731 } 732 if (SCARG(uap, rusage) && error == 0) { 733 error = copyout(&ru, SCARG(uap, rusage), sizeof(ru)); 734 } 735 return error; 736 } 737 738 int 739 sys_wait6(struct lwp *l, const struct sys_wait6_args *uap, register_t *retval) 740 { 741 /* { 742 syscallarg(idtype_t) idtype; 743 syscallarg(id_t) id; 744 syscallarg(int *) status; 745 syscallarg(int) options; 746 syscallarg(struct wrusage *) wru; 747 syscallarg(siginfo_t *) si; 748 } */ 749 struct wrusage wru, *wrup; 750 siginfo_t si, *sip; 751 idtype_t idtype; 752 int pid; 753 id_t id; 754 int error, status; 755 756 idtype = SCARG(uap, idtype); 757 id = SCARG(uap, id); 758 759 if (SCARG(uap, wru) != NULL) 760 wrup = &wru; 761 else 762 wrup = NULL; 763 764 if (SCARG(uap, info) != NULL) 765 sip = &si; 766 else 767 sip = NULL; 768 769 /* 770 * We expect all callers of wait6() to know about WEXITED and 771 * WTRAPPED. 772 */ 773 error = do_sys_waitid(idtype, id, &pid, &status, SCARG(uap, options), 774 wrup, sip); 775 776 retval[0] = pid; /* tell userland who it was */ 777 778 #if 0 779 /* 780 * should we copyout if there was no process, hence no useful data? 781 * We don't for an old sytle wait4() (etc) but I believe 782 * FreeBSD does for wait6(), so a tossup... Go with FreeBSD for now. 783 */ 784 if (pid == 0) 785 return error; 786 #endif 787 788 if (SCARG(uap, status) != NULL && error == 0) 789 error = copyout(&status, SCARG(uap, status), sizeof(status)); 790 if (SCARG(uap, wru) != NULL && error == 0) 791 error = copyout(&wru, SCARG(uap, wru), sizeof(wru)); 792 if (SCARG(uap, info) != NULL && error == 0) 793 error = copyout(&si, SCARG(uap, info), sizeof(si)); 794 return error; 795 } 796 797 798 /* 799 * Find a process that matches the provided criteria, and fill siginfo 800 * and resources if found. 801 * Returns: 802 * -1: Not found, abort early 803 * 0: Not matched 804 * 1: Matched, there might be more matches 805 * 2: This is the only match 806 */ 807 static int 808 match_process(const struct proc *pp, struct proc **q, idtype_t idtype, id_t id, 809 int options, struct wrusage *wrusage, siginfo_t *siginfo) 810 { 811 struct rusage *rup; 812 struct proc *p = *q; 813 int rv = 1; 814 815 mutex_enter(p->p_lock); 816 switch (idtype) { 817 case P_ALL: 818 break; 819 case P_PID: 820 if (p->p_pid != (pid_t)id) { 821 mutex_exit(p->p_lock); 822 p = *q = proc_find_raw((pid_t)id); 823 if (p == NULL || p->p_stat == SIDL || p->p_pptr != pp) { 824 *q = NULL; 825 return -1; 826 } 827 mutex_enter(p->p_lock); 828 } 829 rv++; 830 break; 831 case P_PGID: 832 if (p->p_pgid != (pid_t)id) 833 goto out; 834 break; 835 case P_SID: 836 if (p->p_session->s_sid != (pid_t)id) 837 goto out; 838 break; 839 case P_UID: 840 if (kauth_cred_geteuid(p->p_cred) != (uid_t)id) 841 goto out; 842 break; 843 case P_GID: 844 if (kauth_cred_getegid(p->p_cred) != (gid_t)id) 845 goto out; 846 break; 847 case P_CID: 848 case P_PSETID: 849 case P_CPUID: 850 /* XXX: Implement me */ 851 default: 852 out: 853 mutex_exit(p->p_lock); 854 return 0; 855 } 856 857 if ((options & WEXITED) == 0 && p->p_stat == SZOMB) 858 goto out; 859 860 if (siginfo != NULL) { 861 siginfo->si_errno = 0; 862 863 /* 864 * SUSv4 requires that the si_signo value is always 865 * SIGCHLD. Obey it despite the rfork(2) interface 866 * allows to request other signal for child exit 867 * notification. 868 */ 869 siginfo->si_signo = SIGCHLD; 870 871 /* 872 * This is still a rough estimate. We will fix the 873 * cases TRAPPED, STOPPED, and CONTINUED later. 874 */ 875 if (p->p_sflag & PS_COREDUMP) { 876 siginfo->si_code = CLD_DUMPED; 877 siginfo->si_status = p->p_xsig; 878 } else if (p->p_xsig) { 879 siginfo->si_code = CLD_KILLED; 880 siginfo->si_status = p->p_xsig; 881 } else { 882 siginfo->si_code = CLD_EXITED; 883 siginfo->si_status = p->p_xexit; 884 } 885 886 siginfo->si_pid = p->p_pid; 887 siginfo->si_uid = kauth_cred_geteuid(p->p_cred); 888 siginfo->si_utime = p->p_stats->p_ru.ru_utime.tv_sec; 889 siginfo->si_stime = p->p_stats->p_ru.ru_stime.tv_sec; 890 } 891 892 /* 893 * There should be no reason to limit resources usage info to 894 * exited processes only. A snapshot about any resources used 895 * by a stopped process may be exactly what is needed. 896 */ 897 if (wrusage != NULL) { 898 rup = &wrusage->wru_self; 899 *rup = p->p_stats->p_ru; 900 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 901 902 rup = &wrusage->wru_children; 903 *rup = p->p_stats->p_cru; 904 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 905 } 906 907 mutex_exit(p->p_lock); 908 return rv; 909 } 910 911 /* 912 * Determine if there are existing processes being debugged 913 * that used to be (and sometime later will be again) children 914 * of a specific parent (while matching wait criteria) 915 */ 916 static bool 917 debugged_child_exists(idtype_t idtype, id_t id, int options, siginfo_t *si, 918 const struct proc *parent) 919 { 920 struct proc *pp; 921 922 /* 923 * If we are searching for a specific pid, we can optimise a little 924 */ 925 if (idtype == P_PID) { 926 /* 927 * Check the specific process to see if its real parent is us 928 */ 929 pp = proc_find_raw((pid_t)id); 930 if (pp != NULL && pp->p_stat != SIDL && pp->p_opptr == parent) { 931 /* 932 * using P_ALL here avoids match_process() doing the 933 * same work that we just did, but incorrectly for 934 * this scenario. 935 */ 936 if (match_process(parent, &pp, P_ALL, id, options, 937 NULL, si)) 938 return true; 939 } 940 return false; 941 } 942 943 /* 944 * For the hard cases, just look everywhere to see if some 945 * stolen (reparented) process is really our lost child. 946 * Then check if that process could satisfy the wait conditions. 947 */ 948 949 /* 950 * XXX inefficient, but hopefully fairly rare. 951 * XXX should really use a list of reparented processes. 952 */ 953 PROCLIST_FOREACH(pp, &allproc) { 954 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 955 continue; 956 if (pp->p_opptr == parent && 957 match_process(parent, &pp, idtype, id, options, NULL, si)) 958 return true; 959 } 960 PROCLIST_FOREACH(pp, &zombproc) { 961 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 962 continue; 963 if (pp->p_opptr == parent && 964 match_process(parent, &pp, idtype, id, options, NULL, si)) 965 return true; 966 } 967 968 return false; 969 } 970 971 /* 972 * Scan list of child processes for a child process that has stopped or 973 * exited. Used by sys_wait4 and 'compat' equivalents. 974 * 975 * Must be called with the proc_lock held, and may release while waiting. 976 */ 977 static int 978 find_stopped_child(struct proc *parent, idtype_t idtype, id_t id, int options, 979 struct proc **child_p, struct wrusage *wru, siginfo_t *si) 980 { 981 struct proc *child, *dead; 982 int error; 983 984 KASSERT(mutex_owned(proc_lock)); 985 986 if (options & ~WALLOPTS) { 987 *child_p = NULL; 988 return EINVAL; 989 } 990 991 if ((options & WSELECTOPTS) == 0) { 992 /* 993 * We will be unable to find any matching processes, 994 * because there are no known events to look for. 995 * Prefer to return error instead of blocking 996 * indefinitely. 997 */ 998 *child_p = NULL; 999 return EINVAL; 1000 } 1001 1002 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1003 mutex_enter(parent->p_lock); 1004 id = (id_t)parent->p_pgid; 1005 mutex_exit(parent->p_lock); 1006 idtype = P_PGID; 1007 } 1008 1009 for (;;) { 1010 error = ECHILD; 1011 dead = NULL; 1012 1013 LIST_FOREACH(child, &parent->p_children, p_sibling) { 1014 int rv = match_process(parent, &child, idtype, id, 1015 options, wru, si); 1016 if (rv == -1) 1017 break; 1018 if (rv == 0) 1019 continue; 1020 1021 /* 1022 * Wait for processes with p_exitsig != SIGCHLD 1023 * processes only if WALTSIG is set; wait for 1024 * processes with p_exitsig == SIGCHLD only 1025 * if WALTSIG is clear. 1026 */ 1027 if (((options & WALLSIG) == 0) && 1028 (options & WALTSIG ? child->p_exitsig == SIGCHLD 1029 : P_EXITSIG(child) != SIGCHLD)){ 1030 if (rv == 2) { 1031 child = NULL; 1032 break; 1033 } 1034 continue; 1035 } 1036 1037 error = 0; 1038 if ((options & WNOZOMBIE) == 0) { 1039 if (child->p_stat == SZOMB) 1040 break; 1041 if (child->p_stat == SDEAD) { 1042 /* 1043 * We may occasionally arrive here 1044 * after receiving a signal, but 1045 * immediately before the child 1046 * process is zombified. The wait 1047 * will be short, so avoid returning 1048 * to userspace. 1049 */ 1050 dead = child; 1051 } 1052 } 1053 1054 if ((options & WCONTINUED) != 0 && 1055 child->p_xsig == SIGCONT && 1056 (child->p_sflag & PS_CONTINUED)) { 1057 if ((options & WNOWAIT) == 0) { 1058 child->p_sflag &= ~PS_CONTINUED; 1059 child->p_waited = 1; 1060 parent->p_nstopchild--; 1061 } 1062 if (si) { 1063 si->si_status = child->p_xsig; 1064 si->si_code = CLD_CONTINUED; 1065 } 1066 break; 1067 } 1068 1069 if ((options & (WTRAPPED|WSTOPPED)) != 0 && 1070 child->p_stat == SSTOP && 1071 child->p_waited == 0 && 1072 ((child->p_slflag & PSL_TRACED) || 1073 options & (WUNTRACED|WSTOPPED))) { 1074 if ((options & WNOWAIT) == 0) { 1075 child->p_waited = 1; 1076 parent->p_nstopchild--; 1077 } 1078 if (si) { 1079 si->si_status = child->p_xsig; 1080 si->si_code = 1081 (child->p_slflag & PSL_TRACED) ? 1082 CLD_TRAPPED : CLD_STOPPED; 1083 } 1084 break; 1085 } 1086 if (parent->p_nstopchild == 0 || rv == 2) { 1087 child = NULL; 1088 break; 1089 } 1090 } 1091 1092 /* 1093 * If we found nothing, but we are the bereaved parent 1094 * of a stolen child, look and see if that child (or 1095 * one of them) meets our search criteria. If so, then 1096 * we cannot succeed, but we can hang (wait...), 1097 * or if WNOHANG, return 0 instead of ECHILD 1098 */ 1099 if (child == NULL && error == ECHILD && 1100 (parent->p_slflag & PSL_CHTRACED) && 1101 debugged_child_exists(idtype, id, options, si, parent)) 1102 error = 0; 1103 1104 if (child != NULL || error != 0 || 1105 ((options & WNOHANG) != 0 && dead == NULL)) { 1106 *child_p = child; 1107 return error; 1108 } 1109 1110 /* 1111 * Wait for another child process to stop. 1112 */ 1113 error = cv_wait_sig(&parent->p_waitcv, proc_lock); 1114 1115 if (error != 0) { 1116 *child_p = NULL; 1117 return error; 1118 } 1119 } 1120 } 1121 1122 /* 1123 * Free a process after parent has taken all the state info. Must be called 1124 * with the proclist lock held, and will release before returning. 1125 * 1126 * *ru is returned to the caller, and must be freed by the caller. 1127 */ 1128 static void 1129 proc_free(struct proc *p, struct wrusage *wru) 1130 { 1131 struct proc *parent = p->p_pptr; 1132 struct lwp *l; 1133 ksiginfo_t ksi; 1134 kauth_cred_t cred1, cred2; 1135 uid_t uid; 1136 1137 KASSERT(mutex_owned(proc_lock)); 1138 KASSERT(p->p_nlwps == 1); 1139 KASSERT(p->p_nzlwps == 1); 1140 KASSERT(p->p_nrlwps == 0); 1141 KASSERT(p->p_stat == SZOMB); 1142 1143 /* 1144 * If we got the child via ptrace(2) or procfs, and 1145 * the parent is different (meaning the process was 1146 * attached, rather than run as a child), then we need 1147 * to give it back to the old parent, and send the 1148 * parent the exit signal. The rest of the cleanup 1149 * will be done when the old parent waits on the child. 1150 */ 1151 if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) { 1152 mutex_enter(p->p_lock); 1153 p->p_slflag &= ~(PSL_TRACED|PSL_SYSCALL); 1154 mutex_exit(p->p_lock); 1155 parent = (p->p_opptr == NULL) ? initproc : p->p_opptr; 1156 proc_reparent(p, parent); 1157 p->p_opptr = NULL; 1158 if (p->p_exitsig != 0) { 1159 exit_psignal(p, parent, &ksi); 1160 kpsignal(parent, &ksi, NULL); 1161 } 1162 cv_broadcast(&parent->p_waitcv); 1163 mutex_exit(proc_lock); 1164 return; 1165 } 1166 1167 sched_proc_exit(parent, p); 1168 1169 /* 1170 * Add child times of exiting process onto its own times. 1171 * This cannot be done any earlier else it might get done twice. 1172 */ 1173 l = LIST_FIRST(&p->p_lwps); 1174 p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw); 1175 p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw; 1176 ruadd(&p->p_stats->p_ru, &l->l_ru); 1177 ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru); 1178 ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru); 1179 if (wru != NULL) { 1180 wru->wru_self = p->p_stats->p_ru; 1181 wru->wru_children = p->p_stats->p_cru; 1182 } 1183 p->p_xsig = 0; 1184 p->p_xexit = 0; 1185 1186 /* 1187 * At this point we are going to start freeing the final resources. 1188 * If anyone tries to access the proc structure after here they will 1189 * get a shock - bits are missing. Attempt to make it hard! We 1190 * don't bother with any further locking past this point. 1191 */ 1192 p->p_stat = SIDL; /* not even a zombie any more */ 1193 LIST_REMOVE(p, p_list); /* off zombproc */ 1194 parent->p_nstopchild--; 1195 LIST_REMOVE(p, p_sibling); 1196 1197 /* 1198 * Let pid be reallocated. 1199 */ 1200 proc_free_pid(p->p_pid); 1201 1202 /* 1203 * Unlink process from its process group. 1204 * Releases the proc_lock. 1205 */ 1206 proc_leavepgrp(p); 1207 1208 /* 1209 * Delay release until after lwp_free. 1210 */ 1211 cred2 = l->l_cred; 1212 1213 /* 1214 * Free the last LWP's resources. 1215 * 1216 * lwp_free ensures the LWP is no longer running on another CPU. 1217 */ 1218 lwp_free(l, false, true); 1219 1220 /* 1221 * Now no one except us can reach the process p. 1222 */ 1223 1224 /* 1225 * Decrement the count of procs running with this uid. 1226 */ 1227 cred1 = p->p_cred; 1228 uid = kauth_cred_getuid(cred1); 1229 (void)chgproccnt(uid, -1); 1230 1231 /* 1232 * Release substructures. 1233 */ 1234 1235 lim_free(p->p_limit); 1236 pstatsfree(p->p_stats); 1237 kauth_cred_free(cred1); 1238 kauth_cred_free(cred2); 1239 1240 /* 1241 * Release reference to text vnode 1242 */ 1243 if (p->p_textvp) 1244 vrele(p->p_textvp); 1245 kmem_strfree(p->p_path); 1246 1247 mutex_destroy(&p->p_auxlock); 1248 mutex_obj_free(p->p_lock); 1249 mutex_destroy(&p->p_stmutex); 1250 cv_destroy(&p->p_waitcv); 1251 cv_destroy(&p->p_lwpcv); 1252 rw_destroy(&p->p_reflock); 1253 1254 proc_free_mem(p); 1255 } 1256 1257 /* 1258 * Change the parent of a process for tracing purposes. 1259 */ 1260 void 1261 proc_changeparent(struct proc *t, struct proc *p) 1262 { 1263 SET(t->p_slflag, PSL_TRACED); 1264 t->p_opptr = t->p_pptr; 1265 if (t->p_pptr == p) 1266 return; 1267 struct proc *parent = t->p_pptr; 1268 1269 if (parent->p_lock < t->p_lock) { 1270 if (!mutex_tryenter(parent->p_lock)) { 1271 mutex_exit(t->p_lock); 1272 mutex_enter(parent->p_lock); 1273 mutex_enter(t->p_lock); 1274 } 1275 } else if (parent->p_lock > t->p_lock) { 1276 mutex_enter(parent->p_lock); 1277 } 1278 parent->p_slflag |= PSL_CHTRACED; 1279 proc_reparent(t, p); 1280 if (parent->p_lock != t->p_lock) 1281 mutex_exit(parent->p_lock); 1282 } 1283 1284 /* 1285 * make process 'parent' the new parent of process 'child'. 1286 * 1287 * Must be called with proc_lock held. 1288 */ 1289 void 1290 proc_reparent(struct proc *child, struct proc *parent) 1291 { 1292 1293 KASSERT(mutex_owned(proc_lock)); 1294 1295 if (child->p_pptr == parent) 1296 return; 1297 1298 if (child->p_stat == SZOMB || child->p_stat == SDEAD || 1299 (child->p_stat == SSTOP && !child->p_waited)) { 1300 child->p_pptr->p_nstopchild--; 1301 parent->p_nstopchild++; 1302 } 1303 if (parent == initproc) { 1304 child->p_exitsig = SIGCHLD; 1305 child->p_ppid = parent->p_pid; 1306 } 1307 1308 LIST_REMOVE(child, p_sibling); 1309 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1310 child->p_pptr = parent; 1311 } 1312