1 /* $NetBSD: kern_exit.c,v 1.273 2018/11/29 12:37:22 maxv 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 1.273 2018/11/29 12:37:22 maxv 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/vnode.h> 88 #include <sys/syslog.h> 89 #include <sys/pool.h> 90 #include <sys/uidinfo.h> 91 #include <sys/ptrace.h> 92 #include <sys/acct.h> 93 #include <sys/filedesc.h> 94 #include <sys/ras.h> 95 #include <sys/signalvar.h> 96 #include <sys/sched.h> 97 #include <sys/mount.h> 98 #include <sys/syscallargs.h> 99 #include <sys/kauth.h> 100 #include <sys/sleepq.h> 101 #include <sys/lockdebug.h> 102 #include <sys/ktrace.h> 103 #include <sys/cpu.h> 104 #include <sys/lwpctl.h> 105 #include <sys/atomic.h> 106 #include <sys/sdt.h> 107 108 #include <uvm/uvm_extern.h> 109 110 #ifdef DEBUG_EXIT 111 int debug_exit = 0; 112 #define DPRINTF(x) if (debug_exit) printf x 113 #else 114 #define DPRINTF(x) 115 #endif 116 117 static int find_stopped_child(struct proc *, idtype_t, id_t, int, 118 struct proc **, struct wrusage *, siginfo_t *); 119 static void proc_free(struct proc *, struct wrusage *); 120 121 /* 122 * DTrace SDT provider definitions 123 */ 124 SDT_PROVIDER_DECLARE(proc); 125 SDT_PROBE_DEFINE1(proc, kernel, , exit, "int"); 126 127 /* 128 * Fill in the appropriate signal information, and signal the parent. 129 */ 130 /* XXX noclone works around a gcc 4.5 bug on arm */ 131 static void __noclone 132 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi) 133 { 134 135 KSI_INIT(ksi); 136 if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) { 137 if (p->p_xsig) { 138 if (p->p_sflag & PS_COREDUMP) 139 ksi->ksi_code = CLD_DUMPED; 140 else 141 ksi->ksi_code = CLD_KILLED; 142 ksi->ksi_status = p->p_xsig; 143 } else { 144 ksi->ksi_code = CLD_EXITED; 145 ksi->ksi_status = p->p_xexit; 146 } 147 } else { 148 ksi->ksi_code = SI_USER; 149 ksi->ksi_status = p->p_xsig; 150 } 151 /* 152 * We fill those in, even for non-SIGCHLD. 153 * It's safe to access p->p_cred unlocked here. 154 */ 155 ksi->ksi_pid = p->p_pid; 156 ksi->ksi_uid = kauth_cred_geteuid(p->p_cred); 157 /* XXX: is this still valid? */ 158 ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec; 159 ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec; 160 } 161 162 /* 163 * exit -- 164 * Death of process. 165 */ 166 int 167 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval) 168 { 169 /* { 170 syscallarg(int) rval; 171 } */ 172 struct proc *p = l->l_proc; 173 174 /* Don't call exit1() multiple times in the same process. */ 175 mutex_enter(p->p_lock); 176 if (p->p_sflag & PS_WEXIT) { 177 mutex_exit(p->p_lock); 178 lwp_exit(l); 179 } 180 181 /* exit1() will release the mutex. */ 182 exit1(l, SCARG(uap, rval), 0); 183 /* NOTREACHED */ 184 return (0); 185 } 186 187 /* 188 * Exit: deallocate address space and other resources, change proc state 189 * to zombie, and unlink proc from allproc and parent's lists. Save exit 190 * status and rusage for wait(). Check for child processes and orphan them. 191 * 192 * Must be called with p->p_lock held. Does not return. 193 */ 194 void 195 exit1(struct lwp *l, int exitcode, int signo) 196 { 197 struct proc *p, *child, *next_child, *old_parent, *new_parent; 198 struct pgrp *pgrp; 199 ksiginfo_t ksi; 200 ksiginfoq_t kq; 201 int wakeinit; 202 203 p = l->l_proc; 204 205 KASSERT(mutex_owned(p->p_lock)); 206 KASSERT(p->p_vmspace != NULL); 207 208 if (__predict_false(p == initproc)) { 209 panic("init died (signal %d, exit %d)", signo, exitcode); 210 } 211 212 p->p_sflag |= PS_WEXIT; 213 214 /* 215 * Force all other LWPs to exit before we do. Only then can we 216 * begin to tear down the rest of the process state. 217 */ 218 if (p->p_nlwps > 1) { 219 exit_lwps(l); 220 } 221 222 ksiginfo_queue_init(&kq); 223 224 /* 225 * If we have been asked to stop on exit, do so now. 226 */ 227 if (__predict_false(p->p_sflag & PS_STOPEXIT)) { 228 KERNEL_UNLOCK_ALL(l, &l->l_biglocks); 229 sigclearall(p, &contsigmask, &kq); 230 231 if (!mutex_tryenter(proc_lock)) { 232 mutex_exit(p->p_lock); 233 mutex_enter(proc_lock); 234 mutex_enter(p->p_lock); 235 } 236 p->p_waited = 0; 237 p->p_pptr->p_nstopchild++; 238 p->p_stat = SSTOP; 239 mutex_exit(proc_lock); 240 lwp_lock(l); 241 p->p_nrlwps--; 242 l->l_stat = LSSTOP; 243 lwp_unlock(l); 244 mutex_exit(p->p_lock); 245 lwp_lock(l); 246 mi_switch(l); 247 KERNEL_LOCK(l->l_biglocks, l); 248 mutex_enter(p->p_lock); 249 } 250 251 /* 252 * Bin any remaining signals and mark the process as dying so it will 253 * not be found for, e.g. signals. 254 */ 255 sigfillset(&p->p_sigctx.ps_sigignore); 256 sigclearall(p, NULL, &kq); 257 p->p_stat = SDYING; 258 mutex_exit(p->p_lock); 259 ksiginfo_queue_drain(&kq); 260 261 /* Destroy any lwpctl info. */ 262 if (p->p_lwpctl != NULL) 263 lwp_ctl_exit(); 264 265 /* 266 * Drain all remaining references that procfs, ptrace and others may 267 * have on the process. 268 */ 269 rw_enter(&p->p_reflock, RW_WRITER); 270 271 DPRINTF(("%s: %d.%d exiting.\n", __func__, p->p_pid, l->l_lid)); 272 273 timers_free(p, TIMERS_ALL); 274 #if defined(__HAVE_RAS) 275 ras_purgeall(); 276 #endif 277 278 /* 279 * Close open files, release open-file table and free signal 280 * actions. This may block! 281 */ 282 fd_free(); 283 cwdfree(p->p_cwdi); 284 p->p_cwdi = NULL; 285 doexithooks(p); 286 sigactsfree(p->p_sigacts); 287 288 /* 289 * Write out accounting data. 290 */ 291 (void)acct_process(l); 292 293 #ifdef KTRACE 294 /* 295 * Release trace file. 296 */ 297 if (p->p_tracep != NULL) { 298 mutex_enter(&ktrace_lock); 299 ktrderef(p); 300 mutex_exit(&ktrace_lock); 301 } 302 #endif 303 304 p->p_xexit = exitcode; 305 p->p_xsig = signo; 306 307 /* 308 * If emulation has process exit hook, call it now. 309 * Set the exit status now so that the exit hook has 310 * an opportunity to tweak it (COMPAT_LINUX requires 311 * this for thread group emulation) 312 */ 313 if (p->p_emul->e_proc_exit) 314 (*p->p_emul->e_proc_exit)(p); 315 316 /* 317 * Free the VM resources we're still holding on to. 318 * We must do this from a valid thread because doing 319 * so may block. This frees vmspace, which we don't 320 * need anymore. The only remaining lwp is the one 321 * we run at this moment, nothing runs in userland 322 * anymore. 323 */ 324 ruspace(p); /* Update our vm resource use */ 325 uvm_proc_exit(p); 326 327 /* 328 * Stop profiling. 329 */ 330 if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) { 331 mutex_spin_enter(&p->p_stmutex); 332 stopprofclock(p); 333 mutex_spin_exit(&p->p_stmutex); 334 } 335 336 /* 337 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we 338 * wake up the parent early to avoid deadlock. We can do this once 339 * the VM resources are released. 340 */ 341 mutex_enter(proc_lock); 342 if (p->p_lflag & PL_PPWAIT) { 343 l->l_lwpctl = NULL; /* was on loan from blocked parent */ 344 p->p_lflag &= ~PL_PPWAIT; 345 cv_broadcast(&p->p_pptr->p_waitcv); 346 } 347 348 if (SESS_LEADER(p)) { 349 struct vnode *vprele = NULL, *vprevoke = NULL; 350 struct session *sp = p->p_session; 351 struct tty *tp; 352 353 if (sp->s_ttyvp) { 354 /* 355 * Controlling process. 356 * Signal foreground pgrp, 357 * drain controlling terminal 358 * and revoke access to controlling terminal. 359 */ 360 tp = sp->s_ttyp; 361 mutex_spin_enter(&tty_lock); 362 if (tp->t_session == sp) { 363 /* we can't guarantee the revoke will do this */ 364 pgrp = tp->t_pgrp; 365 tp->t_pgrp = NULL; 366 tp->t_session = NULL; 367 mutex_spin_exit(&tty_lock); 368 if (pgrp != NULL) { 369 pgsignal(pgrp, SIGHUP, 1); 370 } 371 mutex_exit(proc_lock); 372 (void) ttywait(tp); 373 mutex_enter(proc_lock); 374 375 /* The tty could have been revoked. */ 376 vprevoke = sp->s_ttyvp; 377 } else 378 mutex_spin_exit(&tty_lock); 379 vprele = sp->s_ttyvp; 380 sp->s_ttyvp = NULL; 381 /* 382 * s_ttyp is not zero'd; we use this to indicate 383 * that the session once had a controlling terminal. 384 * (for logging and informational purposes) 385 */ 386 } 387 sp->s_leader = NULL; 388 389 if (vprevoke != NULL || vprele != NULL) { 390 if (vprevoke != NULL) { 391 /* Releases proc_lock. */ 392 proc_sessrele(sp); 393 VOP_REVOKE(vprevoke, REVOKEALL); 394 } else 395 mutex_exit(proc_lock); 396 if (vprele != NULL) 397 vrele(vprele); 398 mutex_enter(proc_lock); 399 } 400 } 401 fixjobc(p, p->p_pgrp, 0); 402 403 /* 404 * Finalize the last LWP's specificdata, as well as the 405 * specificdata for the proc itself. 406 */ 407 lwp_finispecific(l); 408 proc_finispecific(p); 409 410 /* 411 * Notify interested parties of our demise. 412 */ 413 KNOTE(&p->p_klist, NOTE_EXIT); 414 415 SDT_PROBE(proc, kernel, , exit, 416 ((p->p_sflag & PS_COREDUMP) ? CLD_DUMPED : 417 (p->p_xsig ? CLD_KILLED : CLD_EXITED)), 418 0,0,0,0); 419 420 /* 421 * Reset p_opptr pointer of all former children which got 422 * traced by another process and were reparented. We reset 423 * it to NULL here; the trace detach code then reparents 424 * the child to initproc. We only check allproc list, since 425 * eventual former children on zombproc list won't reference 426 * p_opptr anymore. 427 */ 428 if (__predict_false(p->p_slflag & PSL_CHTRACED)) { 429 struct proc *q; 430 PROCLIST_FOREACH(q, &allproc) { 431 if (q->p_opptr == p) 432 q->p_opptr = NULL; 433 } 434 PROCLIST_FOREACH(q, &zombproc) { 435 if (q->p_opptr == p) 436 q->p_opptr = NULL; 437 } 438 } 439 440 /* 441 * Give orphaned children to init(8). 442 */ 443 child = LIST_FIRST(&p->p_children); 444 wakeinit = (child != NULL); 445 for (; child != NULL; child = next_child) { 446 next_child = LIST_NEXT(child, p_sibling); 447 448 /* 449 * Traced processes are killed since their existence 450 * means someone is screwing up. Since we reset the 451 * trace flags, the logic in sys_wait4() would not be 452 * triggered to reparent the process to its 453 * original parent, so we must do this here. 454 */ 455 if (__predict_false(child->p_slflag & PSL_TRACED)) { 456 mutex_enter(p->p_lock); 457 child->p_slflag &= 458 ~(PSL_TRACED|PSL_SYSCALL); 459 mutex_exit(p->p_lock); 460 if (child->p_opptr != child->p_pptr) { 461 struct proc *t = child->p_opptr; 462 proc_reparent(child, t ? t : initproc); 463 child->p_opptr = NULL; 464 } else 465 proc_reparent(child, initproc); 466 killproc(child, "orphaned traced process"); 467 } else 468 proc_reparent(child, initproc); 469 } 470 471 /* 472 * Move proc from allproc to zombproc, it's now nearly ready to be 473 * collected by parent. 474 */ 475 LIST_REMOVE(l, l_list); 476 LIST_REMOVE(p, p_list); 477 LIST_INSERT_HEAD(&zombproc, p, p_list); 478 479 /* 480 * Mark the process as dead. We must do this before we signal 481 * the parent. 482 */ 483 p->p_stat = SDEAD; 484 485 /* Put in front of parent's sibling list for parent to collect it */ 486 old_parent = p->p_pptr; 487 old_parent->p_nstopchild++; 488 if (LIST_FIRST(&old_parent->p_children) != p) { 489 /* Put child where it can be found quickly */ 490 LIST_REMOVE(p, p_sibling); 491 LIST_INSERT_HEAD(&old_parent->p_children, p, p_sibling); 492 } 493 494 /* 495 * Notify parent that we're gone. If parent has the P_NOCLDWAIT 496 * flag set, notify init instead (and hope it will handle 497 * this situation). 498 */ 499 if (old_parent->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) { 500 proc_reparent(p, initproc); 501 wakeinit = 1; 502 503 /* 504 * If this was the last child of our parent, notify 505 * parent, so in case he was wait(2)ing, he will 506 * continue. 507 */ 508 if (LIST_FIRST(&old_parent->p_children) == NULL) 509 cv_broadcast(&old_parent->p_waitcv); 510 } 511 512 /* Reload parent pointer, since p may have been reparented above */ 513 new_parent = p->p_pptr; 514 515 if (__predict_false(p->p_exitsig != 0)) { 516 exit_psignal(p, new_parent, &ksi); 517 kpsignal(new_parent, &ksi, NULL); 518 } 519 520 /* Calculate the final rusage info. */ 521 calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime, 522 NULL, NULL); 523 524 if (wakeinit) 525 cv_broadcast(&initproc->p_waitcv); 526 527 callout_destroy(&l->l_timeout_ch); 528 529 /* 530 * Release any PCU resources before becoming a zombie. 531 */ 532 pcu_discard_all(l); 533 534 mutex_enter(p->p_lock); 535 /* Free the linux lwp id */ 536 if ((l->l_pflag & LP_PIDLID) != 0 && l->l_lid != p->p_pid) 537 proc_free_pid(l->l_lid); 538 lwp_drainrefs(l); 539 lwp_lock(l); 540 l->l_prflag &= ~LPR_DETACHED; 541 l->l_stat = LSZOMB; 542 lwp_unlock(l); 543 KASSERT(curlwp == l); 544 KASSERT(p->p_nrlwps == 1); 545 KASSERT(p->p_nlwps == 1); 546 p->p_stat = SZOMB; 547 p->p_nrlwps--; 548 p->p_nzlwps++; 549 p->p_ndlwps = 0; 550 mutex_exit(p->p_lock); 551 552 /* 553 * Signal the parent to collect us, and drop the proclist lock. 554 * Drop debugger/procfs lock; no new references can be gained. 555 */ 556 cv_broadcast(&p->p_pptr->p_waitcv); 557 rw_exit(&p->p_reflock); 558 mutex_exit(proc_lock); 559 560 /* Verify that we hold no locks other than the kernel lock. */ 561 LOCKDEBUG_BARRIER(&kernel_lock, 0); 562 563 /* 564 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP! 565 */ 566 567 /* 568 * Give machine-dependent code a chance to free any MD LWP 569 * resources. This must be done before uvm_lwp_exit(), in 570 * case these resources are in the PCB. 571 */ 572 cpu_lwp_free(l, 1); 573 574 pmap_deactivate(l); 575 576 /* This process no longer needs to hold the kernel lock. */ 577 #ifdef notyet 578 /* XXXSMP hold in lwp_userret() */ 579 KERNEL_UNLOCK_LAST(l); 580 #else 581 KERNEL_UNLOCK_ALL(l, NULL); 582 #endif 583 584 lwp_exit_switchaway(l); 585 } 586 587 void 588 exit_lwps(struct lwp *l) 589 { 590 proc_t *p = l->l_proc; 591 lwp_t *l2; 592 int nlocks; 593 594 KERNEL_UNLOCK_ALL(l, &nlocks); 595 retry: 596 KASSERT(mutex_owned(p->p_lock)); 597 598 /* 599 * Interrupt LWPs in interruptable sleep, unsuspend suspended 600 * LWPs and then wait for everyone else to finish. 601 */ 602 LIST_FOREACH(l2, &p->p_lwps, l_sibling) { 603 if (l2 == l) 604 continue; 605 lwp_lock(l2); 606 l2->l_flag |= LW_WEXIT; 607 if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) || 608 l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) { 609 /* setrunnable() will release the lock. */ 610 setrunnable(l2); 611 continue; 612 } 613 lwp_unlock(l2); 614 } 615 616 /* 617 * Wait for every LWP to exit. Note: LWPs can get suspended/slept 618 * behind us or there may even be new LWPs created. Therefore, a 619 * full retry is required on error. 620 */ 621 while (p->p_nlwps > 1) { 622 if (lwp_wait(l, 0, NULL, true)) { 623 goto retry; 624 } 625 } 626 627 KERNEL_LOCK(nlocks, l); 628 KASSERT(p->p_nlwps == 1); 629 } 630 631 int 632 do_sys_waitid(idtype_t idtype, id_t id, int *pid, int *status, int options, 633 struct wrusage *wru, siginfo_t *si) 634 { 635 proc_t *child; 636 int error; 637 638 639 if (wru != NULL) 640 memset(wru, 0, sizeof(*wru)); 641 if (si != NULL) 642 memset(si, 0, sizeof(*si)); 643 644 mutex_enter(proc_lock); 645 error = find_stopped_child(curproc, idtype, id, options, &child, 646 wru, si); 647 if (child == NULL) { 648 mutex_exit(proc_lock); 649 *pid = 0; 650 *status = 0; 651 return error; 652 } 653 *pid = child->p_pid; 654 655 if (child->p_stat == SZOMB) { 656 /* Child is exiting */ 657 *status = P_WAITSTATUS(child); 658 /* proc_free() will release the proc_lock. */ 659 if (options & WNOWAIT) { 660 mutex_exit(proc_lock); 661 } else { 662 proc_free(child, wru); 663 } 664 } else { 665 /* Don't mark SIGCONT if we are being stopped */ 666 *status = (child->p_xsig == SIGCONT && child->p_stat != SSTOP) ? 667 W_CONTCODE() : W_STOPCODE(child->p_xsig); 668 mutex_exit(proc_lock); 669 } 670 return 0; 671 } 672 673 int 674 do_sys_wait(int *pid, int *status, int options, struct rusage *ru) 675 { 676 idtype_t idtype; 677 id_t id; 678 int ret; 679 struct wrusage wru; 680 681 /* 682 * Translate the special pid values into the (idtype, pid) 683 * pair for wait6. The WAIT_MYPGRP case is handled by 684 * find_stopped_child() on its own. 685 */ 686 if (*pid == WAIT_ANY) { 687 idtype = P_ALL; 688 id = 0; 689 } else if (*pid < 0) { 690 idtype = P_PGID; 691 id = (id_t)-*pid; 692 } else { 693 idtype = P_PID; 694 id = (id_t)*pid; 695 } 696 options |= WEXITED | WTRAPPED; 697 ret = do_sys_waitid(idtype, id, pid, status, options, ru ? &wru : NULL, 698 NULL); 699 if (ru) 700 *ru = wru.wru_self; 701 return ret; 702 } 703 704 int 705 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap, 706 register_t *retval) 707 { 708 /* { 709 syscallarg(int) pid; 710 syscallarg(int *) status; 711 syscallarg(int) options; 712 syscallarg(struct rusage *) rusage; 713 } */ 714 int error, status, pid = SCARG(uap, pid); 715 struct rusage ru; 716 717 error = do_sys_wait(&pid, &status, SCARG(uap, options), 718 SCARG(uap, rusage) != NULL ? &ru : NULL); 719 720 retval[0] = pid; 721 if (pid == 0) { 722 return error; 723 } 724 if (SCARG(uap, status)) { 725 error = copyout(&status, SCARG(uap, status), sizeof(status)); 726 } 727 if (SCARG(uap, rusage) && error == 0) { 728 error = copyout(&ru, SCARG(uap, rusage), sizeof(ru)); 729 } 730 return error; 731 } 732 733 int 734 sys_wait6(struct lwp *l, const struct sys_wait6_args *uap, register_t *retval) 735 { 736 /* { 737 syscallarg(idtype_t) idtype; 738 syscallarg(id_t) id; 739 syscallarg(int *) status; 740 syscallarg(int) options; 741 syscallarg(struct wrusage *) wru; 742 syscallarg(siginfo_t *) si; 743 } */ 744 struct wrusage wru, *wrup; 745 siginfo_t si, *sip; 746 idtype_t idtype; 747 int pid; 748 id_t id; 749 int error, status; 750 751 idtype = SCARG(uap, idtype); 752 id = SCARG(uap, id); 753 754 if (SCARG(uap, wru) != NULL) 755 wrup = &wru; 756 else 757 wrup = NULL; 758 759 if (SCARG(uap, info) != NULL) 760 sip = &si; 761 else 762 sip = NULL; 763 764 /* 765 * We expect all callers of wait6() to know about WEXITED and 766 * WTRAPPED. 767 */ 768 error = do_sys_waitid(idtype, id, &pid, &status, SCARG(uap, options), 769 wrup, sip); 770 771 retval[0] = pid; /* tell userland who it was */ 772 773 #if 0 774 /* 775 * should we copyout if there was no process, hence no useful data? 776 * We don't for an old sytle wait4() (etc) but I believe 777 * FreeBSD does for wait6(), so a tossup... Go with FreeBSD for now. 778 */ 779 if (pid == 0) 780 return error; 781 #endif 782 783 if (SCARG(uap, status) != NULL && error == 0) 784 error = copyout(&status, SCARG(uap, status), sizeof(status)); 785 if (SCARG(uap, wru) != NULL && error == 0) 786 error = copyout(&wru, SCARG(uap, wru), sizeof(wru)); 787 if (SCARG(uap, info) != NULL && error == 0) 788 error = copyout(&si, SCARG(uap, info), sizeof(si)); 789 return error; 790 } 791 792 793 /* 794 * Find a process that matches the provided criteria, and fill siginfo 795 * and resources if found. 796 * Returns: 797 * -1: Not found, abort early 798 * 0: Not matched 799 * 1: Matched, there might be more matches 800 * 2: This is the only match 801 */ 802 static int 803 match_process(const struct proc *pp, struct proc **q, idtype_t idtype, id_t id, 804 int options, struct wrusage *wrusage, siginfo_t *siginfo) 805 { 806 struct rusage *rup; 807 struct proc *p = *q; 808 int rv = 1; 809 810 mutex_enter(p->p_lock); 811 switch (idtype) { 812 case P_ALL: 813 break; 814 case P_PID: 815 if (p->p_pid != (pid_t)id) { 816 mutex_exit(p->p_lock); 817 p = *q = proc_find_raw((pid_t)id); 818 if (p == NULL || p->p_stat == SIDL || p->p_pptr != pp) { 819 *q = NULL; 820 return -1; 821 } 822 mutex_enter(p->p_lock); 823 } 824 rv++; 825 break; 826 case P_PGID: 827 if (p->p_pgid != (pid_t)id) 828 goto out; 829 break; 830 case P_SID: 831 if (p->p_session->s_sid != (pid_t)id) 832 goto out; 833 break; 834 case P_UID: 835 if (kauth_cred_geteuid(p->p_cred) != (uid_t)id) 836 goto out; 837 break; 838 case P_GID: 839 if (kauth_cred_getegid(p->p_cred) != (gid_t)id) 840 goto out; 841 break; 842 case P_CID: 843 case P_PSETID: 844 case P_CPUID: 845 /* XXX: Implement me */ 846 default: 847 out: 848 mutex_exit(p->p_lock); 849 return 0; 850 } 851 852 if ((options & WEXITED) == 0 && p->p_stat == SZOMB) 853 goto out; 854 855 if (siginfo != NULL) { 856 siginfo->si_errno = 0; 857 858 /* 859 * SUSv4 requires that the si_signo value is always 860 * SIGCHLD. Obey it despite the rfork(2) interface 861 * allows to request other signal for child exit 862 * notification. 863 */ 864 siginfo->si_signo = SIGCHLD; 865 866 /* 867 * This is still a rough estimate. We will fix the 868 * cases TRAPPED, STOPPED, and CONTINUED later. 869 */ 870 if (p->p_sflag & PS_COREDUMP) { 871 siginfo->si_code = CLD_DUMPED; 872 siginfo->si_status = p->p_xsig; 873 } else if (p->p_xsig) { 874 siginfo->si_code = CLD_KILLED; 875 siginfo->si_status = p->p_xsig; 876 } else { 877 siginfo->si_code = CLD_EXITED; 878 siginfo->si_status = p->p_xexit; 879 } 880 881 siginfo->si_pid = p->p_pid; 882 siginfo->si_uid = kauth_cred_geteuid(p->p_cred); 883 siginfo->si_utime = p->p_stats->p_ru.ru_utime.tv_sec; 884 siginfo->si_stime = p->p_stats->p_ru.ru_stime.tv_sec; 885 } 886 887 /* 888 * There should be no reason to limit resources usage info to 889 * exited processes only. A snapshot about any resources used 890 * by a stopped process may be exactly what is needed. 891 */ 892 if (wrusage != NULL) { 893 rup = &wrusage->wru_self; 894 *rup = p->p_stats->p_ru; 895 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 896 897 rup = &wrusage->wru_children; 898 *rup = p->p_stats->p_cru; 899 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 900 } 901 902 mutex_exit(p->p_lock); 903 return rv; 904 } 905 906 /* 907 * Determine if there are existing processes being debugged 908 * that used to be (and sometime later will be again) children 909 * of a specific parent (while matching wait criteria) 910 */ 911 static bool 912 debugged_child_exists(idtype_t idtype, id_t id, int options, siginfo_t *si, 913 const struct proc *parent) 914 { 915 struct proc *pp; 916 917 /* 918 * If we are searching for a specific pid, we can optimise a little 919 */ 920 if (idtype == P_PID) { 921 /* 922 * Check the specific process to see if its real parent is us 923 */ 924 pp = proc_find_raw((pid_t)id); 925 if (pp != NULL && pp->p_stat != SIDL && pp->p_opptr == parent) { 926 /* 927 * using P_ALL here avoids match_process() doing the 928 * same work that we just did, but incorrectly for 929 * this scenario. 930 */ 931 if (match_process(parent, &pp, P_ALL, id, options, 932 NULL, si)) 933 return true; 934 } 935 return false; 936 } 937 938 /* 939 * For the hard cases, just look everywhere to see if some 940 * stolen (reparented) process is really our lost child. 941 * Then check if that process could satisfy the wait conditions. 942 */ 943 944 /* 945 * XXX inefficient, but hopefully fairly rare. 946 * XXX should really use a list of reparented processes. 947 */ 948 PROCLIST_FOREACH(pp, &allproc) { 949 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 950 continue; 951 if (pp->p_opptr == parent && 952 match_process(parent, &pp, idtype, id, options, NULL, si)) 953 return true; 954 } 955 PROCLIST_FOREACH(pp, &zombproc) { 956 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 957 continue; 958 if (pp->p_opptr == parent && 959 match_process(parent, &pp, idtype, id, options, NULL, si)) 960 return true; 961 } 962 963 return false; 964 } 965 966 /* 967 * Scan list of child processes for a child process that has stopped or 968 * exited. Used by sys_wait4 and 'compat' equivalents. 969 * 970 * Must be called with the proc_lock held, and may release while waiting. 971 */ 972 static int 973 find_stopped_child(struct proc *parent, idtype_t idtype, id_t id, int options, 974 struct proc **child_p, struct wrusage *wru, siginfo_t *si) 975 { 976 struct proc *child, *dead; 977 int error; 978 979 KASSERT(mutex_owned(proc_lock)); 980 981 if (options & ~WALLOPTS) { 982 *child_p = NULL; 983 return EINVAL; 984 } 985 986 if ((options & WSELECTOPTS) == 0) { 987 /* 988 * We will be unable to find any matching processes, 989 * because there are no known events to look for. 990 * Prefer to return error instead of blocking 991 * indefinitely. 992 */ 993 *child_p = NULL; 994 return EINVAL; 995 } 996 997 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 998 mutex_enter(parent->p_lock); 999 id = (id_t)parent->p_pgid; 1000 mutex_exit(parent->p_lock); 1001 idtype = P_PGID; 1002 } 1003 1004 for (;;) { 1005 error = ECHILD; 1006 dead = NULL; 1007 1008 LIST_FOREACH(child, &parent->p_children, p_sibling) { 1009 int rv = match_process(parent, &child, idtype, id, 1010 options, wru, si); 1011 if (rv == -1) 1012 break; 1013 if (rv == 0) 1014 continue; 1015 1016 /* 1017 * Wait for processes with p_exitsig != SIGCHLD 1018 * processes only if WALTSIG is set; wait for 1019 * processes with p_exitsig == SIGCHLD only 1020 * if WALTSIG is clear. 1021 */ 1022 if (((options & WALLSIG) == 0) && 1023 (options & WALTSIG ? child->p_exitsig == SIGCHLD 1024 : P_EXITSIG(child) != SIGCHLD)){ 1025 if (rv == 2) { 1026 child = NULL; 1027 break; 1028 } 1029 continue; 1030 } 1031 1032 error = 0; 1033 if ((options & WNOZOMBIE) == 0) { 1034 if (child->p_stat == SZOMB) 1035 break; 1036 if (child->p_stat == SDEAD) { 1037 /* 1038 * We may occasionally arrive here 1039 * after receiving a signal, but 1040 * immediately before the child 1041 * process is zombified. The wait 1042 * will be short, so avoid returning 1043 * to userspace. 1044 */ 1045 dead = child; 1046 } 1047 } 1048 1049 if ((options & WCONTINUED) != 0 && 1050 child->p_xsig == SIGCONT && 1051 (child->p_sflag & PS_CONTINUED)) { 1052 if ((options & WNOWAIT) == 0) { 1053 child->p_sflag &= ~PS_CONTINUED; 1054 child->p_waited = 1; 1055 parent->p_nstopchild--; 1056 } 1057 if (si) { 1058 si->si_status = child->p_xsig; 1059 si->si_code = CLD_CONTINUED; 1060 } 1061 break; 1062 } 1063 1064 if ((options & (WTRAPPED|WSTOPPED)) != 0 && 1065 child->p_stat == SSTOP && 1066 child->p_waited == 0 && 1067 ((child->p_slflag & PSL_TRACED) || 1068 options & (WUNTRACED|WSTOPPED))) { 1069 if ((options & WNOWAIT) == 0) { 1070 child->p_waited = 1; 1071 parent->p_nstopchild--; 1072 } 1073 if (si) { 1074 si->si_status = child->p_xsig; 1075 si->si_code = 1076 (child->p_slflag & PSL_TRACED) ? 1077 CLD_TRAPPED : CLD_STOPPED; 1078 } 1079 break; 1080 } 1081 if (parent->p_nstopchild == 0 || rv == 2) { 1082 child = NULL; 1083 break; 1084 } 1085 } 1086 1087 /* 1088 * If we found nothing, but we are the bereaved parent 1089 * of a stolen child, look and see if that child (or 1090 * one of them) meets our search criteria. If so, then 1091 * we cannot succeed, but we can hang (wait...), 1092 * or if WNOHANG, return 0 instead of ECHILD 1093 */ 1094 if (child == NULL && error == ECHILD && 1095 (parent->p_slflag & PSL_CHTRACED) && 1096 debugged_child_exists(idtype, id, options, si, parent)) 1097 error = 0; 1098 1099 if (child != NULL || error != 0 || 1100 ((options & WNOHANG) != 0 && dead == NULL)) { 1101 *child_p = child; 1102 return error; 1103 } 1104 1105 /* 1106 * Wait for another child process to stop. 1107 */ 1108 error = cv_wait_sig(&parent->p_waitcv, proc_lock); 1109 1110 if (error != 0) { 1111 *child_p = NULL; 1112 return error; 1113 } 1114 } 1115 } 1116 1117 /* 1118 * Free a process after parent has taken all the state info. Must be called 1119 * with the proclist lock held, and will release before returning. 1120 * 1121 * *ru is returned to the caller, and must be freed by the caller. 1122 */ 1123 static void 1124 proc_free(struct proc *p, struct wrusage *wru) 1125 { 1126 struct proc *parent = p->p_pptr; 1127 struct lwp *l; 1128 ksiginfo_t ksi; 1129 kauth_cred_t cred1, cred2; 1130 uid_t uid; 1131 1132 KASSERT(mutex_owned(proc_lock)); 1133 KASSERT(p->p_nlwps == 1); 1134 KASSERT(p->p_nzlwps == 1); 1135 KASSERT(p->p_nrlwps == 0); 1136 KASSERT(p->p_stat == SZOMB); 1137 1138 /* 1139 * If we got the child via ptrace(2) or procfs, and 1140 * the parent is different (meaning the process was 1141 * attached, rather than run as a child), then we need 1142 * to give it back to the old parent, and send the 1143 * parent the exit signal. The rest of the cleanup 1144 * will be done when the old parent waits on the child. 1145 */ 1146 if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) { 1147 mutex_enter(p->p_lock); 1148 p->p_slflag &= ~(PSL_TRACED|PSL_SYSCALL); 1149 mutex_exit(p->p_lock); 1150 parent = (p->p_opptr == NULL) ? initproc : p->p_opptr; 1151 proc_reparent(p, parent); 1152 p->p_opptr = NULL; 1153 if (p->p_exitsig != 0) { 1154 exit_psignal(p, parent, &ksi); 1155 kpsignal(parent, &ksi, NULL); 1156 } 1157 cv_broadcast(&parent->p_waitcv); 1158 mutex_exit(proc_lock); 1159 return; 1160 } 1161 1162 sched_proc_exit(parent, p); 1163 1164 /* 1165 * Add child times of exiting process onto its own times. 1166 * This cannot be done any earlier else it might get done twice. 1167 */ 1168 l = LIST_FIRST(&p->p_lwps); 1169 p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw); 1170 p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw; 1171 ruadd(&p->p_stats->p_ru, &l->l_ru); 1172 ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru); 1173 ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru); 1174 if (wru != NULL) { 1175 wru->wru_self = p->p_stats->p_ru; 1176 wru->wru_children = p->p_stats->p_cru; 1177 } 1178 p->p_xsig = 0; 1179 p->p_xexit = 0; 1180 1181 /* 1182 * At this point we are going to start freeing the final resources. 1183 * If anyone tries to access the proc structure after here they will 1184 * get a shock - bits are missing. Attempt to make it hard! We 1185 * don't bother with any further locking past this point. 1186 */ 1187 p->p_stat = SIDL; /* not even a zombie any more */ 1188 LIST_REMOVE(p, p_list); /* off zombproc */ 1189 parent->p_nstopchild--; 1190 LIST_REMOVE(p, p_sibling); 1191 1192 /* 1193 * Let pid be reallocated. 1194 */ 1195 proc_free_pid(p->p_pid); 1196 1197 /* 1198 * Unlink process from its process group. 1199 * Releases the proc_lock. 1200 */ 1201 proc_leavepgrp(p); 1202 1203 /* 1204 * Delay release until after lwp_free. 1205 */ 1206 cred2 = l->l_cred; 1207 1208 /* 1209 * Free the last LWP's resources. 1210 * 1211 * lwp_free ensures the LWP is no longer running on another CPU. 1212 */ 1213 lwp_free(l, false, true); 1214 1215 /* 1216 * Now no one except us can reach the process p. 1217 */ 1218 1219 /* 1220 * Decrement the count of procs running with this uid. 1221 */ 1222 cred1 = p->p_cred; 1223 uid = kauth_cred_getuid(cred1); 1224 (void)chgproccnt(uid, -1); 1225 1226 /* 1227 * Release substructures. 1228 */ 1229 1230 lim_free(p->p_limit); 1231 pstatsfree(p->p_stats); 1232 kauth_cred_free(cred1); 1233 kauth_cred_free(cred2); 1234 1235 /* 1236 * Release reference to text vnode 1237 */ 1238 if (p->p_textvp) 1239 vrele(p->p_textvp); 1240 kmem_strfree(p->p_path); 1241 1242 mutex_destroy(&p->p_auxlock); 1243 mutex_obj_free(p->p_lock); 1244 mutex_destroy(&p->p_stmutex); 1245 cv_destroy(&p->p_waitcv); 1246 cv_destroy(&p->p_lwpcv); 1247 rw_destroy(&p->p_reflock); 1248 1249 proc_free_mem(p); 1250 } 1251 1252 /* 1253 * Change the parent of a process for tracing purposes. 1254 */ 1255 void 1256 proc_changeparent(struct proc *t, struct proc *p) 1257 { 1258 SET(t->p_slflag, PSL_TRACED); 1259 t->p_opptr = t->p_pptr; 1260 if (t->p_pptr == p) 1261 return; 1262 struct proc *parent = t->p_pptr; 1263 1264 if (parent->p_lock < t->p_lock) { 1265 if (!mutex_tryenter(parent->p_lock)) { 1266 mutex_exit(t->p_lock); 1267 mutex_enter(parent->p_lock); 1268 mutex_enter(t->p_lock); 1269 } 1270 } else if (parent->p_lock > t->p_lock) { 1271 mutex_enter(parent->p_lock); 1272 } 1273 parent->p_slflag |= PSL_CHTRACED; 1274 proc_reparent(t, p); 1275 if (parent->p_lock != t->p_lock) 1276 mutex_exit(parent->p_lock); 1277 } 1278 1279 /* 1280 * make process 'parent' the new parent of process 'child'. 1281 * 1282 * Must be called with proc_lock held. 1283 */ 1284 void 1285 proc_reparent(struct proc *child, struct proc *parent) 1286 { 1287 1288 KASSERT(mutex_owned(proc_lock)); 1289 1290 if (child->p_pptr == parent) 1291 return; 1292 1293 if (child->p_stat == SZOMB || child->p_stat == SDEAD || 1294 (child->p_stat == SSTOP && !child->p_waited)) { 1295 child->p_pptr->p_nstopchild--; 1296 parent->p_nstopchild++; 1297 } 1298 if (parent == initproc) { 1299 child->p_exitsig = SIGCHLD; 1300 child->p_ppid = parent->p_pid; 1301 } 1302 1303 LIST_REMOVE(child, p_sibling); 1304 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1305 child->p_pptr = parent; 1306 } 1307