1 /* $NetBSD: kern_exit.c,v 1.276 2019/06/13 20:20:18 kamil 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 1.276 2019/06/13 20:20:18 kamil 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 lwp_t *lp; 346 347 l->l_lwpctl = NULL; /* was on loan from blocked parent */ 348 p->p_lflag &= ~PL_PPWAIT; 349 350 lp = p->p_vforklwp; 351 p->p_vforklwp = NULL; 352 lp->l_vforkwaiting = false; 353 cv_broadcast(&lp->l_waitcv); 354 } 355 356 if (SESS_LEADER(p)) { 357 struct vnode *vprele = NULL, *vprevoke = NULL; 358 struct session *sp = p->p_session; 359 struct tty *tp; 360 361 if (sp->s_ttyvp) { 362 /* 363 * Controlling process. 364 * Signal foreground pgrp, 365 * drain controlling terminal 366 * and revoke access to controlling terminal. 367 */ 368 tp = sp->s_ttyp; 369 mutex_spin_enter(&tty_lock); 370 if (tp->t_session == sp) { 371 /* we can't guarantee the revoke will do this */ 372 pgrp = tp->t_pgrp; 373 tp->t_pgrp = NULL; 374 tp->t_session = NULL; 375 mutex_spin_exit(&tty_lock); 376 if (pgrp != NULL) { 377 pgsignal(pgrp, SIGHUP, 1); 378 } 379 mutex_exit(proc_lock); 380 (void) ttywait(tp); 381 mutex_enter(proc_lock); 382 383 /* The tty could have been revoked. */ 384 vprevoke = sp->s_ttyvp; 385 } else 386 mutex_spin_exit(&tty_lock); 387 vprele = sp->s_ttyvp; 388 sp->s_ttyvp = NULL; 389 /* 390 * s_ttyp is not zero'd; we use this to indicate 391 * that the session once had a controlling terminal. 392 * (for logging and informational purposes) 393 */ 394 } 395 sp->s_leader = NULL; 396 397 if (vprevoke != NULL || vprele != NULL) { 398 if (vprevoke != NULL) { 399 /* Releases proc_lock. */ 400 proc_sessrele(sp); 401 VOP_REVOKE(vprevoke, REVOKEALL); 402 } else 403 mutex_exit(proc_lock); 404 if (vprele != NULL) 405 vrele(vprele); 406 mutex_enter(proc_lock); 407 } 408 } 409 fixjobc(p, p->p_pgrp, 0); 410 411 /* Release fstrans private data. */ 412 fstrans_lwp_dtor(l); 413 414 /* 415 * Finalize the last LWP's specificdata, as well as the 416 * specificdata for the proc itself. 417 */ 418 lwp_finispecific(l); 419 proc_finispecific(p); 420 421 /* 422 * Notify interested parties of our demise. 423 */ 424 KNOTE(&p->p_klist, NOTE_EXIT); 425 426 SDT_PROBE(proc, kernel, , exit, 427 ((p->p_sflag & PS_COREDUMP) ? CLD_DUMPED : 428 (p->p_xsig ? CLD_KILLED : CLD_EXITED)), 429 0,0,0,0); 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 struct proc *q; 441 PROCLIST_FOREACH(q, &allproc) { 442 if (q->p_opptr == p) 443 q->p_opptr = NULL; 444 } 445 PROCLIST_FOREACH(q, &zombproc) { 446 if (q->p_opptr == p) 447 q->p_opptr = NULL; 448 } 449 } 450 451 /* 452 * Give orphaned children to init(8). 453 */ 454 child = LIST_FIRST(&p->p_children); 455 wakeinit = (child != NULL); 456 for (; child != NULL; child = next_child) { 457 next_child = LIST_NEXT(child, p_sibling); 458 459 /* 460 * Traced processes are killed since their existence 461 * means someone is screwing up. Since we reset the 462 * trace flags, the logic in sys_wait4() would not be 463 * triggered to reparent the process to its 464 * original parent, so we must do this here. 465 */ 466 if (__predict_false(child->p_slflag & PSL_TRACED)) { 467 mutex_enter(p->p_lock); 468 child->p_slflag &= 469 ~(PSL_TRACED|PSL_SYSCALL); 470 mutex_exit(p->p_lock); 471 if (child->p_opptr != child->p_pptr) { 472 struct proc *t = child->p_opptr; 473 proc_reparent(child, t ? t : initproc); 474 child->p_opptr = NULL; 475 } else 476 proc_reparent(child, initproc); 477 killproc(child, "orphaned traced process"); 478 } else 479 proc_reparent(child, initproc); 480 } 481 482 /* 483 * Move proc from allproc to zombproc, it's now nearly ready to be 484 * collected by parent. 485 */ 486 LIST_REMOVE(l, l_list); 487 LIST_REMOVE(p, p_list); 488 LIST_INSERT_HEAD(&zombproc, p, p_list); 489 490 /* 491 * Mark the process as dead. We must do this before we signal 492 * the parent. 493 */ 494 p->p_stat = SDEAD; 495 496 /* Put in front of parent's sibling list for parent to collect it */ 497 old_parent = p->p_pptr; 498 old_parent->p_nstopchild++; 499 if (LIST_FIRST(&old_parent->p_children) != p) { 500 /* Put child where it can be found quickly */ 501 LIST_REMOVE(p, p_sibling); 502 LIST_INSERT_HEAD(&old_parent->p_children, p, p_sibling); 503 } 504 505 /* 506 * Notify parent that we're gone. If parent has the P_NOCLDWAIT 507 * flag set, notify init instead (and hope it will handle 508 * this situation). 509 */ 510 if (old_parent->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) { 511 proc_reparent(p, initproc); 512 wakeinit = 1; 513 514 /* 515 * If this was the last child of our parent, notify 516 * parent, so in case he was wait(2)ing, he will 517 * continue. 518 */ 519 if (LIST_FIRST(&old_parent->p_children) == NULL) 520 cv_broadcast(&old_parent->p_waitcv); 521 } 522 523 /* Reload parent pointer, since p may have been reparented above */ 524 new_parent = p->p_pptr; 525 526 if (__predict_false(p->p_exitsig != 0)) { 527 exit_psignal(p, new_parent, &ksi); 528 kpsignal(new_parent, &ksi, NULL); 529 } 530 531 /* Calculate the final rusage info. */ 532 calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime, 533 NULL, NULL); 534 535 if (wakeinit) 536 cv_broadcast(&initproc->p_waitcv); 537 538 callout_destroy(&l->l_timeout_ch); 539 540 /* 541 * Release any PCU resources before becoming a zombie. 542 */ 543 pcu_discard_all(l); 544 545 mutex_enter(p->p_lock); 546 /* Free the linux lwp id */ 547 if ((l->l_pflag & LP_PIDLID) != 0 && l->l_lid != p->p_pid) 548 proc_free_pid(l->l_lid); 549 lwp_drainrefs(l); 550 lwp_lock(l); 551 l->l_prflag &= ~LPR_DETACHED; 552 l->l_stat = LSZOMB; 553 lwp_unlock(l); 554 KASSERT(curlwp == l); 555 KASSERT(p->p_nrlwps == 1); 556 KASSERT(p->p_nlwps == 1); 557 p->p_stat = SZOMB; 558 p->p_nrlwps--; 559 p->p_nzlwps++; 560 p->p_ndlwps = 0; 561 mutex_exit(p->p_lock); 562 563 /* 564 * Signal the parent to collect us, and drop the proclist lock. 565 * Drop debugger/procfs lock; no new references can be gained. 566 */ 567 cv_broadcast(&p->p_pptr->p_waitcv); 568 rw_exit(&p->p_reflock); 569 mutex_exit(proc_lock); 570 571 /* Verify that we hold no locks other than the kernel lock. */ 572 LOCKDEBUG_BARRIER(&kernel_lock, 0); 573 574 /* 575 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP! 576 */ 577 578 /* 579 * Give machine-dependent code a chance to free any MD LWP 580 * resources. This must be done before uvm_lwp_exit(), in 581 * case these resources are in the PCB. 582 */ 583 cpu_lwp_free(l, 1); 584 585 pmap_deactivate(l); 586 587 /* This process no longer needs to hold the kernel lock. */ 588 #ifdef notyet 589 /* XXXSMP hold in lwp_userret() */ 590 KERNEL_UNLOCK_LAST(l); 591 #else 592 KERNEL_UNLOCK_ALL(l, NULL); 593 #endif 594 595 lwp_exit_switchaway(l); 596 } 597 598 void 599 exit_lwps(struct lwp *l) 600 { 601 proc_t *p = l->l_proc; 602 lwp_t *l2; 603 int nlocks; 604 605 KERNEL_UNLOCK_ALL(l, &nlocks); 606 retry: 607 KASSERT(mutex_owned(p->p_lock)); 608 609 /* 610 * Interrupt LWPs in interruptable sleep, unsuspend suspended 611 * LWPs and then wait for everyone else to finish. 612 */ 613 LIST_FOREACH(l2, &p->p_lwps, l_sibling) { 614 if (l2 == l) 615 continue; 616 lwp_lock(l2); 617 l2->l_flag |= LW_WEXIT; 618 if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) || 619 l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) { 620 /* setrunnable() will release the lock. */ 621 setrunnable(l2); 622 continue; 623 } 624 lwp_unlock(l2); 625 } 626 627 /* 628 * Wait for every LWP to exit. Note: LWPs can get suspended/slept 629 * behind us or there may even be new LWPs created. Therefore, a 630 * full retry is required on error. 631 */ 632 while (p->p_nlwps > 1) { 633 if (lwp_wait(l, 0, NULL, true)) { 634 goto retry; 635 } 636 } 637 638 KERNEL_LOCK(nlocks, l); 639 KASSERT(p->p_nlwps == 1); 640 } 641 642 int 643 do_sys_waitid(idtype_t idtype, id_t id, int *pid, int *status, int options, 644 struct wrusage *wru, siginfo_t *si) 645 { 646 proc_t *child; 647 int error; 648 649 650 if (wru != NULL) 651 memset(wru, 0, sizeof(*wru)); 652 if (si != NULL) 653 memset(si, 0, sizeof(*si)); 654 655 mutex_enter(proc_lock); 656 error = find_stopped_child(curproc, idtype, id, options, &child, 657 wru, si); 658 if (child == NULL) { 659 mutex_exit(proc_lock); 660 *pid = 0; 661 *status = 0; 662 return error; 663 } 664 *pid = child->p_pid; 665 666 if (child->p_stat == SZOMB) { 667 /* Child is exiting */ 668 *status = P_WAITSTATUS(child); 669 /* proc_free() will release the proc_lock. */ 670 if (options & WNOWAIT) { 671 mutex_exit(proc_lock); 672 } else { 673 proc_free(child, wru); 674 } 675 } else { 676 /* Don't mark SIGCONT if we are being stopped */ 677 *status = (child->p_xsig == SIGCONT && child->p_stat != SSTOP) ? 678 W_CONTCODE() : W_STOPCODE(child->p_xsig); 679 mutex_exit(proc_lock); 680 } 681 return 0; 682 } 683 684 int 685 do_sys_wait(int *pid, int *status, int options, struct rusage *ru) 686 { 687 idtype_t idtype; 688 id_t id; 689 int ret; 690 struct wrusage wru; 691 692 /* 693 * Translate the special pid values into the (idtype, pid) 694 * pair for wait6. The WAIT_MYPGRP case is handled by 695 * find_stopped_child() on its own. 696 */ 697 if (*pid == WAIT_ANY) { 698 idtype = P_ALL; 699 id = 0; 700 } else if (*pid < 0) { 701 idtype = P_PGID; 702 id = (id_t)-*pid; 703 } else { 704 idtype = P_PID; 705 id = (id_t)*pid; 706 } 707 options |= WEXITED | WTRAPPED; 708 ret = do_sys_waitid(idtype, id, pid, status, options, ru ? &wru : NULL, 709 NULL); 710 if (ru) 711 *ru = wru.wru_self; 712 return ret; 713 } 714 715 int 716 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap, 717 register_t *retval) 718 { 719 /* { 720 syscallarg(int) pid; 721 syscallarg(int *) status; 722 syscallarg(int) options; 723 syscallarg(struct rusage *) rusage; 724 } */ 725 int error, status, pid = SCARG(uap, pid); 726 struct rusage ru; 727 728 error = do_sys_wait(&pid, &status, SCARG(uap, options), 729 SCARG(uap, rusage) != NULL ? &ru : NULL); 730 731 retval[0] = pid; 732 if (pid == 0) { 733 return error; 734 } 735 if (SCARG(uap, status)) { 736 error = copyout(&status, SCARG(uap, status), sizeof(status)); 737 } 738 if (SCARG(uap, rusage) && error == 0) { 739 error = copyout(&ru, SCARG(uap, rusage), sizeof(ru)); 740 } 741 return error; 742 } 743 744 int 745 sys_wait6(struct lwp *l, const struct sys_wait6_args *uap, register_t *retval) 746 { 747 /* { 748 syscallarg(idtype_t) idtype; 749 syscallarg(id_t) id; 750 syscallarg(int *) status; 751 syscallarg(int) options; 752 syscallarg(struct wrusage *) wru; 753 syscallarg(siginfo_t *) si; 754 } */ 755 struct wrusage wru, *wrup; 756 siginfo_t si, *sip; 757 idtype_t idtype; 758 int pid; 759 id_t id; 760 int error, status; 761 762 idtype = SCARG(uap, idtype); 763 id = SCARG(uap, id); 764 765 if (SCARG(uap, wru) != NULL) 766 wrup = &wru; 767 else 768 wrup = NULL; 769 770 if (SCARG(uap, info) != NULL) 771 sip = &si; 772 else 773 sip = NULL; 774 775 /* 776 * We expect all callers of wait6() to know about WEXITED and 777 * WTRAPPED. 778 */ 779 error = do_sys_waitid(idtype, id, &pid, &status, SCARG(uap, options), 780 wrup, sip); 781 782 retval[0] = pid; /* tell userland who it was */ 783 784 #if 0 785 /* 786 * should we copyout if there was no process, hence no useful data? 787 * We don't for an old sytle wait4() (etc) but I believe 788 * FreeBSD does for wait6(), so a tossup... Go with FreeBSD for now. 789 */ 790 if (pid == 0) 791 return error; 792 #endif 793 794 if (SCARG(uap, status) != NULL && error == 0) 795 error = copyout(&status, SCARG(uap, status), sizeof(status)); 796 if (SCARG(uap, wru) != NULL && error == 0) 797 error = copyout(&wru, SCARG(uap, wru), sizeof(wru)); 798 if (SCARG(uap, info) != NULL && error == 0) 799 error = copyout(&si, SCARG(uap, info), sizeof(si)); 800 return error; 801 } 802 803 804 /* 805 * Find a process that matches the provided criteria, and fill siginfo 806 * and resources if found. 807 * Returns: 808 * -1: Not found, abort early 809 * 0: Not matched 810 * 1: Matched, there might be more matches 811 * 2: This is the only match 812 */ 813 static int 814 match_process(const struct proc *pp, struct proc **q, idtype_t idtype, id_t id, 815 int options, struct wrusage *wrusage, siginfo_t *siginfo) 816 { 817 struct rusage *rup; 818 struct proc *p = *q; 819 int rv = 1; 820 821 mutex_enter(p->p_lock); 822 switch (idtype) { 823 case P_ALL: 824 break; 825 case P_PID: 826 if (p->p_pid != (pid_t)id) { 827 mutex_exit(p->p_lock); 828 p = *q = proc_find_raw((pid_t)id); 829 if (p == NULL || p->p_stat == SIDL || p->p_pptr != pp) { 830 *q = NULL; 831 return -1; 832 } 833 mutex_enter(p->p_lock); 834 } 835 rv++; 836 break; 837 case P_PGID: 838 if (p->p_pgid != (pid_t)id) 839 goto out; 840 break; 841 case P_SID: 842 if (p->p_session->s_sid != (pid_t)id) 843 goto out; 844 break; 845 case P_UID: 846 if (kauth_cred_geteuid(p->p_cred) != (uid_t)id) 847 goto out; 848 break; 849 case P_GID: 850 if (kauth_cred_getegid(p->p_cred) != (gid_t)id) 851 goto out; 852 break; 853 case P_CID: 854 case P_PSETID: 855 case P_CPUID: 856 /* XXX: Implement me */ 857 default: 858 out: 859 mutex_exit(p->p_lock); 860 return 0; 861 } 862 863 if ((options & WEXITED) == 0 && p->p_stat == SZOMB) 864 goto out; 865 866 if (siginfo != NULL) { 867 siginfo->si_errno = 0; 868 869 /* 870 * SUSv4 requires that the si_signo value is always 871 * SIGCHLD. Obey it despite the rfork(2) interface 872 * allows to request other signal for child exit 873 * notification. 874 */ 875 siginfo->si_signo = SIGCHLD; 876 877 /* 878 * This is still a rough estimate. We will fix the 879 * cases TRAPPED, STOPPED, and CONTINUED later. 880 */ 881 if (p->p_sflag & PS_COREDUMP) { 882 siginfo->si_code = CLD_DUMPED; 883 siginfo->si_status = p->p_xsig; 884 } else if (p->p_xsig) { 885 siginfo->si_code = CLD_KILLED; 886 siginfo->si_status = p->p_xsig; 887 } else { 888 siginfo->si_code = CLD_EXITED; 889 siginfo->si_status = p->p_xexit; 890 } 891 892 siginfo->si_pid = p->p_pid; 893 siginfo->si_uid = kauth_cred_geteuid(p->p_cred); 894 siginfo->si_utime = p->p_stats->p_ru.ru_utime.tv_sec; 895 siginfo->si_stime = p->p_stats->p_ru.ru_stime.tv_sec; 896 } 897 898 /* 899 * There should be no reason to limit resources usage info to 900 * exited processes only. A snapshot about any resources used 901 * by a stopped process may be exactly what is needed. 902 */ 903 if (wrusage != NULL) { 904 rup = &wrusage->wru_self; 905 *rup = p->p_stats->p_ru; 906 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 907 908 rup = &wrusage->wru_children; 909 *rup = p->p_stats->p_cru; 910 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 911 } 912 913 mutex_exit(p->p_lock); 914 return rv; 915 } 916 917 /* 918 * Determine if there are existing processes being debugged 919 * that used to be (and sometime later will be again) children 920 * of a specific parent (while matching wait criteria) 921 */ 922 static bool 923 debugged_child_exists(idtype_t idtype, id_t id, int options, siginfo_t *si, 924 const struct proc *parent) 925 { 926 struct proc *pp; 927 928 /* 929 * If we are searching for a specific pid, we can optimise a little 930 */ 931 if (idtype == P_PID) { 932 /* 933 * Check the specific process to see if its real parent is us 934 */ 935 pp = proc_find_raw((pid_t)id); 936 if (pp != NULL && pp->p_stat != SIDL && pp->p_opptr == parent) { 937 /* 938 * using P_ALL here avoids match_process() doing the 939 * same work that we just did, but incorrectly for 940 * this scenario. 941 */ 942 if (match_process(parent, &pp, P_ALL, id, options, 943 NULL, si)) 944 return true; 945 } 946 return false; 947 } 948 949 /* 950 * For the hard cases, just look everywhere to see if some 951 * stolen (reparented) process is really our lost child. 952 * Then check if that process could satisfy the wait conditions. 953 */ 954 955 /* 956 * XXX inefficient, but hopefully fairly rare. 957 * XXX should really use a list of reparented processes. 958 */ 959 PROCLIST_FOREACH(pp, &allproc) { 960 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 961 continue; 962 if (pp->p_opptr == parent && 963 match_process(parent, &pp, idtype, id, options, NULL, si)) 964 return true; 965 } 966 PROCLIST_FOREACH(pp, &zombproc) { 967 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 968 continue; 969 if (pp->p_opptr == parent && 970 match_process(parent, &pp, idtype, id, options, NULL, si)) 971 return true; 972 } 973 974 return false; 975 } 976 977 /* 978 * Scan list of child processes for a child process that has stopped or 979 * exited. Used by sys_wait4 and 'compat' equivalents. 980 * 981 * Must be called with the proc_lock held, and may release while waiting. 982 */ 983 static int 984 find_stopped_child(struct proc *parent, idtype_t idtype, id_t id, int options, 985 struct proc **child_p, struct wrusage *wru, siginfo_t *si) 986 { 987 struct proc *child, *dead; 988 int error; 989 990 KASSERT(mutex_owned(proc_lock)); 991 992 if (options & ~WALLOPTS) { 993 *child_p = NULL; 994 return EINVAL; 995 } 996 997 if ((options & WSELECTOPTS) == 0) { 998 /* 999 * We will be unable to find any matching processes, 1000 * because there are no known events to look for. 1001 * Prefer to return error instead of blocking 1002 * indefinitely. 1003 */ 1004 *child_p = NULL; 1005 return EINVAL; 1006 } 1007 1008 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1009 mutex_enter(parent->p_lock); 1010 id = (id_t)parent->p_pgid; 1011 mutex_exit(parent->p_lock); 1012 idtype = P_PGID; 1013 } 1014 1015 for (;;) { 1016 error = ECHILD; 1017 dead = NULL; 1018 1019 LIST_FOREACH(child, &parent->p_children, p_sibling) { 1020 int rv = match_process(parent, &child, idtype, id, 1021 options, wru, si); 1022 if (rv == -1) 1023 break; 1024 if (rv == 0) 1025 continue; 1026 1027 /* 1028 * Wait for processes with p_exitsig != SIGCHLD 1029 * processes only if WALTSIG is set; wait for 1030 * processes with p_exitsig == SIGCHLD only 1031 * if WALTSIG is clear. 1032 */ 1033 if (((options & WALLSIG) == 0) && 1034 (options & WALTSIG ? child->p_exitsig == SIGCHLD 1035 : P_EXITSIG(child) != SIGCHLD)){ 1036 if (rv == 2) { 1037 child = NULL; 1038 break; 1039 } 1040 continue; 1041 } 1042 1043 error = 0; 1044 if ((options & WNOZOMBIE) == 0) { 1045 if (child->p_stat == SZOMB) 1046 break; 1047 if (child->p_stat == SDEAD) { 1048 /* 1049 * We may occasionally arrive here 1050 * after receiving a signal, but 1051 * immediately before the child 1052 * process is zombified. The wait 1053 * will be short, so avoid returning 1054 * to userspace. 1055 */ 1056 dead = child; 1057 } 1058 } 1059 1060 if ((options & WCONTINUED) != 0 && 1061 child->p_xsig == SIGCONT && 1062 (child->p_sflag & PS_CONTINUED)) { 1063 if ((options & WNOWAIT) == 0) { 1064 child->p_sflag &= ~PS_CONTINUED; 1065 child->p_waited = 1; 1066 parent->p_nstopchild--; 1067 } 1068 if (si) { 1069 si->si_status = child->p_xsig; 1070 si->si_code = CLD_CONTINUED; 1071 } 1072 break; 1073 } 1074 1075 if ((options & (WTRAPPED|WSTOPPED)) != 0 && 1076 child->p_stat == SSTOP && 1077 child->p_waited == 0 && 1078 ((child->p_slflag & PSL_TRACED) || 1079 options & (WUNTRACED|WSTOPPED))) { 1080 if ((options & WNOWAIT) == 0) { 1081 child->p_waited = 1; 1082 parent->p_nstopchild--; 1083 } 1084 if (si) { 1085 si->si_status = child->p_xsig; 1086 si->si_code = 1087 (child->p_slflag & PSL_TRACED) ? 1088 CLD_TRAPPED : CLD_STOPPED; 1089 } 1090 break; 1091 } 1092 if (parent->p_nstopchild == 0 || rv == 2) { 1093 child = NULL; 1094 break; 1095 } 1096 } 1097 1098 /* 1099 * If we found nothing, but we are the bereaved parent 1100 * of a stolen child, look and see if that child (or 1101 * one of them) meets our search criteria. If so, then 1102 * we cannot succeed, but we can hang (wait...), 1103 * or if WNOHANG, return 0 instead of ECHILD 1104 */ 1105 if (child == NULL && error == ECHILD && 1106 (parent->p_slflag & PSL_CHTRACED) && 1107 debugged_child_exists(idtype, id, options, si, parent)) 1108 error = 0; 1109 1110 if (child != NULL || error != 0 || 1111 ((options & WNOHANG) != 0 && dead == NULL)) { 1112 *child_p = child; 1113 return error; 1114 } 1115 1116 /* 1117 * Wait for another child process to stop. 1118 */ 1119 error = cv_wait_sig(&parent->p_waitcv, proc_lock); 1120 1121 if (error != 0) { 1122 *child_p = NULL; 1123 return error; 1124 } 1125 } 1126 } 1127 1128 /* 1129 * Free a process after parent has taken all the state info. Must be called 1130 * with the proclist lock held, and will release before returning. 1131 * 1132 * *ru is returned to the caller, and must be freed by the caller. 1133 */ 1134 static void 1135 proc_free(struct proc *p, struct wrusage *wru) 1136 { 1137 struct proc *parent = p->p_pptr; 1138 struct lwp *l; 1139 ksiginfo_t ksi; 1140 kauth_cred_t cred1, cred2; 1141 uid_t uid; 1142 1143 KASSERT(mutex_owned(proc_lock)); 1144 KASSERT(p->p_nlwps == 1); 1145 KASSERT(p->p_nzlwps == 1); 1146 KASSERT(p->p_nrlwps == 0); 1147 KASSERT(p->p_stat == SZOMB); 1148 1149 /* 1150 * If we got the child via ptrace(2) or procfs, and 1151 * the parent is different (meaning the process was 1152 * attached, rather than run as a child), then we need 1153 * to give it back to the old parent, and send the 1154 * parent the exit signal. The rest of the cleanup 1155 * will be done when the old parent waits on the child. 1156 */ 1157 if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) { 1158 mutex_enter(p->p_lock); 1159 p->p_slflag &= ~(PSL_TRACED|PSL_SYSCALL); 1160 mutex_exit(p->p_lock); 1161 parent = (p->p_opptr == NULL) ? initproc : p->p_opptr; 1162 proc_reparent(p, parent); 1163 p->p_opptr = NULL; 1164 if (p->p_exitsig != 0) { 1165 exit_psignal(p, parent, &ksi); 1166 kpsignal(parent, &ksi, NULL); 1167 } 1168 cv_broadcast(&parent->p_waitcv); 1169 mutex_exit(proc_lock); 1170 return; 1171 } 1172 1173 sched_proc_exit(parent, p); 1174 1175 /* 1176 * Add child times of exiting process onto its own times. 1177 * This cannot be done any earlier else it might get done twice. 1178 */ 1179 l = LIST_FIRST(&p->p_lwps); 1180 p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw); 1181 p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw; 1182 ruadd(&p->p_stats->p_ru, &l->l_ru); 1183 ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru); 1184 ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru); 1185 if (wru != NULL) { 1186 wru->wru_self = p->p_stats->p_ru; 1187 wru->wru_children = p->p_stats->p_cru; 1188 } 1189 p->p_xsig = 0; 1190 p->p_xexit = 0; 1191 1192 /* 1193 * At this point we are going to start freeing the final resources. 1194 * If anyone tries to access the proc structure after here they will 1195 * get a shock - bits are missing. Attempt to make it hard! We 1196 * don't bother with any further locking past this point. 1197 */ 1198 p->p_stat = SIDL; /* not even a zombie any more */ 1199 LIST_REMOVE(p, p_list); /* off zombproc */ 1200 parent->p_nstopchild--; 1201 LIST_REMOVE(p, p_sibling); 1202 1203 /* 1204 * Let pid be reallocated. 1205 */ 1206 proc_free_pid(p->p_pid); 1207 1208 /* 1209 * Unlink process from its process group. 1210 * Releases the proc_lock. 1211 */ 1212 proc_leavepgrp(p); 1213 1214 /* 1215 * Delay release until after lwp_free. 1216 */ 1217 cred2 = l->l_cred; 1218 1219 /* 1220 * Free the last LWP's resources. 1221 * 1222 * lwp_free ensures the LWP is no longer running on another CPU. 1223 */ 1224 lwp_free(l, false, true); 1225 1226 /* 1227 * Now no one except us can reach the process p. 1228 */ 1229 1230 /* 1231 * Decrement the count of procs running with this uid. 1232 */ 1233 cred1 = p->p_cred; 1234 uid = kauth_cred_getuid(cred1); 1235 (void)chgproccnt(uid, -1); 1236 1237 /* 1238 * Release substructures. 1239 */ 1240 1241 lim_free(p->p_limit); 1242 pstatsfree(p->p_stats); 1243 kauth_cred_free(cred1); 1244 kauth_cred_free(cred2); 1245 1246 /* 1247 * Release reference to text vnode 1248 */ 1249 if (p->p_textvp) 1250 vrele(p->p_textvp); 1251 kmem_strfree(p->p_path); 1252 1253 mutex_destroy(&p->p_auxlock); 1254 mutex_obj_free(p->p_lock); 1255 mutex_destroy(&p->p_stmutex); 1256 cv_destroy(&p->p_waitcv); 1257 cv_destroy(&p->p_lwpcv); 1258 rw_destroy(&p->p_reflock); 1259 1260 proc_free_mem(p); 1261 } 1262 1263 /* 1264 * Change the parent of a process for tracing purposes. 1265 */ 1266 void 1267 proc_changeparent(struct proc *t, struct proc *p) 1268 { 1269 SET(t->p_slflag, PSL_TRACED); 1270 t->p_opptr = t->p_pptr; 1271 if (t->p_pptr == p) 1272 return; 1273 struct proc *parent = t->p_pptr; 1274 1275 if (parent->p_lock < t->p_lock) { 1276 if (!mutex_tryenter(parent->p_lock)) { 1277 mutex_exit(t->p_lock); 1278 mutex_enter(parent->p_lock); 1279 mutex_enter(t->p_lock); 1280 } 1281 } else if (parent->p_lock > t->p_lock) { 1282 mutex_enter(parent->p_lock); 1283 } 1284 parent->p_slflag |= PSL_CHTRACED; 1285 proc_reparent(t, p); 1286 if (parent->p_lock != t->p_lock) 1287 mutex_exit(parent->p_lock); 1288 } 1289 1290 /* 1291 * make process 'parent' the new parent of process 'child'. 1292 * 1293 * Must be called with proc_lock held. 1294 */ 1295 void 1296 proc_reparent(struct proc *child, struct proc *parent) 1297 { 1298 1299 KASSERT(mutex_owned(proc_lock)); 1300 1301 if (child->p_pptr == parent) 1302 return; 1303 1304 if (child->p_stat == SZOMB || child->p_stat == SDEAD || 1305 (child->p_stat == SSTOP && !child->p_waited)) { 1306 child->p_pptr->p_nstopchild--; 1307 parent->p_nstopchild++; 1308 } 1309 if (parent == initproc) { 1310 child->p_exitsig = SIGCHLD; 1311 child->p_ppid = parent->p_pid; 1312 } 1313 1314 LIST_REMOVE(child, p_sibling); 1315 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1316 child->p_pptr = parent; 1317 } 1318