1 /* $NetBSD: kern_exit.c,v 1.291 2020/12/05 18:17:01 thorpej Exp $ */ 2 3 /*- 4 * Copyright (c) 1998, 1999, 2006, 2007, 2008, 2020 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 1.291 2020/12/05 18:17:01 thorpej 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/lock.h> 103 #include <sys/lockdebug.h> 104 #include <sys/ktrace.h> 105 #include <sys/cpu.h> 106 #include <sys/lwpctl.h> 107 #include <sys/atomic.h> 108 #include <sys/sdt.h> 109 #include <sys/psref.h> 110 111 #include <uvm/uvm_extern.h> 112 113 #ifdef DEBUG_EXIT 114 int debug_exit = 0; 115 #define DPRINTF(x) if (debug_exit) printf x 116 #else 117 #define DPRINTF(x) 118 #endif 119 120 static int find_stopped_child(struct proc *, idtype_t, id_t, int, 121 struct proc **, struct wrusage *, siginfo_t *); 122 static void proc_free(struct proc *, struct wrusage *); 123 124 /* 125 * DTrace SDT provider definitions 126 */ 127 SDT_PROVIDER_DECLARE(proc); 128 SDT_PROBE_DEFINE1(proc, kernel, , exit, "int"); 129 130 /* 131 * Fill in the appropriate signal information, and signal the parent. 132 */ 133 /* XXX noclone works around a gcc 4.5 bug on arm */ 134 static void __noclone 135 exit_psignal(struct proc *p, struct proc *pp, ksiginfo_t *ksi) 136 { 137 138 KSI_INIT(ksi); 139 if ((ksi->ksi_signo = P_EXITSIG(p)) == SIGCHLD) { 140 if (p->p_xsig) { 141 if (p->p_sflag & PS_COREDUMP) 142 ksi->ksi_code = CLD_DUMPED; 143 else 144 ksi->ksi_code = CLD_KILLED; 145 ksi->ksi_status = p->p_xsig; 146 } else { 147 ksi->ksi_code = CLD_EXITED; 148 ksi->ksi_status = p->p_xexit; 149 } 150 } else { 151 ksi->ksi_code = SI_USER; 152 ksi->ksi_status = p->p_xsig; 153 } 154 /* 155 * We fill those in, even for non-SIGCHLD. 156 * It's safe to access p->p_cred unlocked here. 157 */ 158 ksi->ksi_pid = p->p_pid; 159 ksi->ksi_uid = kauth_cred_geteuid(p->p_cred); 160 /* XXX: is this still valid? */ 161 ksi->ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec; 162 ksi->ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec; 163 } 164 165 /* 166 * exit -- 167 * Death of process. 168 */ 169 int 170 sys_exit(struct lwp *l, const struct sys_exit_args *uap, register_t *retval) 171 { 172 /* { 173 syscallarg(int) rval; 174 } */ 175 struct proc *p = l->l_proc; 176 177 /* Don't call exit1() multiple times in the same process. */ 178 mutex_enter(p->p_lock); 179 if (p->p_sflag & PS_WEXIT) { 180 mutex_exit(p->p_lock); 181 lwp_exit(l); 182 } 183 184 /* exit1() will release the mutex. */ 185 exit1(l, SCARG(uap, rval), 0); 186 /* NOTREACHED */ 187 return (0); 188 } 189 190 /* 191 * Exit: deallocate address space and other resources, change proc state 192 * to zombie, and unlink proc from allproc and parent's lists. Save exit 193 * status and rusage for wait(). Check for child processes and orphan them. 194 * 195 * Must be called with p->p_lock held. Does not return. 196 */ 197 void 198 exit1(struct lwp *l, int exitcode, int signo) 199 { 200 struct proc *p, *child, *next_child, *old_parent, *new_parent; 201 struct pgrp *pgrp; 202 ksiginfo_t ksi; 203 ksiginfoq_t kq; 204 int wakeinit; 205 206 p = l->l_proc; 207 208 /* Verify that we hold no locks other than p->p_lock. */ 209 LOCKDEBUG_BARRIER(p->p_lock, 0); 210 211 /* XXX Temporary: something is leaking kernel_lock. */ 212 KERNEL_UNLOCK_ALL(l, NULL); 213 214 KASSERT(mutex_owned(p->p_lock)); 215 KASSERT(p->p_vmspace != NULL); 216 217 if (__predict_false(p == initproc)) { 218 panic("init died (signal %d, exit %d)", signo, exitcode); 219 } 220 221 p->p_sflag |= PS_WEXIT; 222 223 /* 224 * Force all other LWPs to exit before we do. Only then can we 225 * begin to tear down the rest of the process state. 226 */ 227 if (p->p_nlwps > 1) { 228 exit_lwps(l); 229 } 230 231 ksiginfo_queue_init(&kq); 232 233 /* 234 * If we have been asked to stop on exit, do so now. 235 */ 236 if (__predict_false(p->p_sflag & PS_STOPEXIT)) { 237 KERNEL_UNLOCK_ALL(l, &l->l_biglocks); 238 sigclearall(p, &contsigmask, &kq); 239 240 if (!mutex_tryenter(&proc_lock)) { 241 mutex_exit(p->p_lock); 242 mutex_enter(&proc_lock); 243 mutex_enter(p->p_lock); 244 } 245 p->p_waited = 0; 246 p->p_pptr->p_nstopchild++; 247 p->p_stat = SSTOP; 248 mutex_exit(&proc_lock); 249 lwp_lock(l); 250 p->p_nrlwps--; 251 l->l_stat = LSSTOP; 252 lwp_unlock(l); 253 mutex_exit(p->p_lock); 254 lwp_lock(l); 255 spc_lock(l->l_cpu); 256 mi_switch(l); 257 mutex_enter(p->p_lock); 258 } 259 260 /* 261 * Bin any remaining signals and mark the process as dying so it will 262 * not be found for, e.g. signals. 263 */ 264 sigfillset(&p->p_sigctx.ps_sigignore); 265 sigclearall(p, NULL, &kq); 266 p->p_stat = SDYING; 267 268 /* 269 * Perform any required thread cleanup. Do this early so 270 * anyone wanting to look us up by our global thread ID 271 * will fail to find us. 272 * 273 * N.B. this will unlock p->p_lock on our behalf. 274 */ 275 lwp_thread_cleanup(l); 276 277 ksiginfo_queue_drain(&kq); 278 279 /* Destroy any lwpctl info. */ 280 if (p->p_lwpctl != NULL) 281 lwp_ctl_exit(); 282 283 /* 284 * Drain all remaining references that procfs, ptrace and others may 285 * have on the process. 286 */ 287 rw_enter(&p->p_reflock, RW_WRITER); 288 289 DPRINTF(("%s: %d.%d exiting.\n", __func__, p->p_pid, l->l_lid)); 290 291 ptimers_free(p, TIMERS_ALL); 292 #if defined(__HAVE_RAS) 293 ras_purgeall(); 294 #endif 295 296 /* 297 * Close open files, release open-file table and free signal 298 * actions. This may block! 299 */ 300 fd_free(); 301 cwdfree(p->p_cwdi); 302 p->p_cwdi = NULL; 303 doexithooks(p); 304 sigactsfree(p->p_sigacts); 305 306 /* 307 * Write out accounting data. 308 */ 309 (void)acct_process(l); 310 311 #ifdef KTRACE 312 /* 313 * Release trace file. 314 */ 315 if (p->p_tracep != NULL) { 316 mutex_enter(&ktrace_lock); 317 ktrderef(p); 318 mutex_exit(&ktrace_lock); 319 } 320 #endif 321 322 p->p_xexit = exitcode; 323 p->p_xsig = signo; 324 325 /* 326 * If emulation has process exit hook, call it now. 327 * Set the exit status now so that the exit hook has 328 * an opportunity to tweak it (COMPAT_LINUX requires 329 * this for thread group emulation) 330 */ 331 if (p->p_emul->e_proc_exit) 332 (*p->p_emul->e_proc_exit)(p); 333 334 /* 335 * Free the VM resources we're still holding on to. 336 * We must do this from a valid thread because doing 337 * so may block. This frees vmspace, which we don't 338 * need anymore. The only remaining lwp is the one 339 * we run at this moment, nothing runs in userland 340 * anymore. 341 */ 342 ruspace(p); /* Update our vm resource use */ 343 uvm_proc_exit(p); 344 345 /* 346 * Stop profiling. 347 */ 348 if (__predict_false((p->p_stflag & PST_PROFIL) != 0)) { 349 mutex_spin_enter(&p->p_stmutex); 350 stopprofclock(p); 351 mutex_spin_exit(&p->p_stmutex); 352 } 353 354 /* 355 * If parent is waiting for us to exit or exec, PL_PPWAIT is set; we 356 * wake up the parent early to avoid deadlock. We can do this once 357 * the VM resources are released. 358 */ 359 mutex_enter(&proc_lock); 360 if (p->p_lflag & PL_PPWAIT) { 361 lwp_t *lp; 362 363 l->l_lwpctl = NULL; /* was on loan from blocked parent */ 364 p->p_lflag &= ~PL_PPWAIT; 365 366 lp = p->p_vforklwp; 367 p->p_vforklwp = NULL; 368 lp->l_vforkwaiting = false; 369 cv_broadcast(&lp->l_waitcv); 370 } 371 372 if (SESS_LEADER(p)) { 373 struct vnode *vprele = NULL, *vprevoke = NULL; 374 struct session *sp = p->p_session; 375 struct tty *tp; 376 377 if (sp->s_ttyvp) { 378 /* 379 * Controlling process. 380 * Signal foreground pgrp, 381 * drain controlling terminal 382 * and revoke access to controlling terminal. 383 */ 384 tp = sp->s_ttyp; 385 mutex_spin_enter(&tty_lock); 386 if (tp->t_session == sp) { 387 /* we can't guarantee the revoke will do this */ 388 pgrp = tp->t_pgrp; 389 tp->t_pgrp = NULL; 390 tp->t_session = NULL; 391 mutex_spin_exit(&tty_lock); 392 if (pgrp != NULL) { 393 pgsignal(pgrp, SIGHUP, 1); 394 } 395 mutex_exit(&proc_lock); 396 (void) ttywait(tp); 397 mutex_enter(&proc_lock); 398 399 /* The tty could have been revoked. */ 400 vprevoke = sp->s_ttyvp; 401 } else 402 mutex_spin_exit(&tty_lock); 403 vprele = sp->s_ttyvp; 404 sp->s_ttyvp = NULL; 405 /* 406 * s_ttyp is not zero'd; we use this to indicate 407 * that the session once had a controlling terminal. 408 * (for logging and informational purposes) 409 */ 410 } 411 sp->s_leader = NULL; 412 413 if (vprevoke != NULL || vprele != NULL) { 414 if (vprevoke != NULL) { 415 /* Releases proc_lock. */ 416 proc_sessrele(sp); 417 VOP_REVOKE(vprevoke, REVOKEALL); 418 } else 419 mutex_exit(&proc_lock); 420 if (vprele != NULL) 421 vrele(vprele); 422 mutex_enter(&proc_lock); 423 } 424 } 425 fixjobc(p, p->p_pgrp, 0); 426 427 /* Release fstrans private data. */ 428 fstrans_lwp_dtor(l); 429 430 /* 431 * Finalize the last LWP's specificdata, as well as the 432 * specificdata for the proc itself. 433 */ 434 lwp_finispecific(l); 435 proc_finispecific(p); 436 437 /* 438 * Notify interested parties of our demise. 439 */ 440 KNOTE(&p->p_klist, NOTE_EXIT); 441 442 SDT_PROBE(proc, kernel, , exit, 443 ((p->p_sflag & PS_COREDUMP) ? CLD_DUMPED : 444 (p->p_xsig ? CLD_KILLED : CLD_EXITED)), 445 0,0,0,0); 446 447 /* 448 * Reset p_opptr pointer of all former children which got 449 * traced by another process and were reparented. We reset 450 * it to NULL here; the trace detach code then reparents 451 * the child to initproc. We only check allproc list, since 452 * eventual former children on zombproc list won't reference 453 * p_opptr anymore. 454 */ 455 if (__predict_false(p->p_slflag & PSL_CHTRACED)) { 456 struct proc *q; 457 PROCLIST_FOREACH(q, &allproc) { 458 if (q->p_opptr == p) 459 q->p_opptr = NULL; 460 } 461 PROCLIST_FOREACH(q, &zombproc) { 462 if (q->p_opptr == p) 463 q->p_opptr = NULL; 464 } 465 } 466 467 /* 468 * Give orphaned children to init(8). 469 */ 470 child = LIST_FIRST(&p->p_children); 471 wakeinit = (child != NULL); 472 for (; child != NULL; child = next_child) { 473 next_child = LIST_NEXT(child, p_sibling); 474 475 /* 476 * Traced processes are killed since their existence 477 * means someone is screwing up. Since we reset the 478 * trace flags, the logic in sys_wait4() would not be 479 * triggered to reparent the process to its 480 * original parent, so we must do this here. 481 */ 482 if (__predict_false(child->p_slflag & PSL_TRACED)) { 483 mutex_enter(p->p_lock); 484 child->p_slflag &= 485 ~(PSL_TRACED|PSL_SYSCALL); 486 mutex_exit(p->p_lock); 487 if (child->p_opptr != child->p_pptr) { 488 struct proc *t = child->p_opptr; 489 proc_reparent(child, t ? t : initproc); 490 child->p_opptr = NULL; 491 } else 492 proc_reparent(child, initproc); 493 killproc(child, "orphaned traced process"); 494 } else 495 proc_reparent(child, initproc); 496 } 497 498 /* 499 * Move proc from allproc to zombproc, it's now nearly ready to be 500 * collected by parent. 501 */ 502 LIST_REMOVE(l, l_list); 503 LIST_REMOVE(p, p_list); 504 LIST_INSERT_HEAD(&zombproc, p, p_list); 505 506 /* 507 * Mark the process as dead. We must do this before we signal 508 * the parent. 509 */ 510 p->p_stat = SDEAD; 511 512 /* Put in front of parent's sibling list for parent to collect it */ 513 old_parent = p->p_pptr; 514 old_parent->p_nstopchild++; 515 if (LIST_FIRST(&old_parent->p_children) != p) { 516 /* Put child where it can be found quickly */ 517 LIST_REMOVE(p, p_sibling); 518 LIST_INSERT_HEAD(&old_parent->p_children, p, p_sibling); 519 } 520 521 /* 522 * Notify parent that we're gone. If parent has the P_NOCLDWAIT 523 * flag set, notify init instead (and hope it will handle 524 * this situation). 525 */ 526 if (old_parent->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) { 527 proc_reparent(p, initproc); 528 wakeinit = 1; 529 530 /* 531 * If this was the last child of our parent, notify 532 * parent, so in case he was wait(2)ing, he will 533 * continue. 534 */ 535 if (LIST_FIRST(&old_parent->p_children) == NULL) 536 cv_broadcast(&old_parent->p_waitcv); 537 } 538 539 /* Reload parent pointer, since p may have been reparented above */ 540 new_parent = p->p_pptr; 541 542 if (__predict_false(p->p_exitsig != 0)) { 543 exit_psignal(p, new_parent, &ksi); 544 kpsignal(new_parent, &ksi, NULL); 545 } 546 547 /* Calculate the final rusage info. */ 548 calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime, 549 NULL, NULL); 550 551 if (wakeinit) 552 cv_broadcast(&initproc->p_waitcv); 553 554 callout_destroy(&l->l_timeout_ch); 555 556 /* 557 * Release any PCU resources before becoming a zombie. 558 */ 559 pcu_discard_all(l); 560 561 mutex_enter(p->p_lock); 562 /* Free the LWP ID */ 563 proc_free_lwpid(p, l->l_lid); 564 lwp_drainrefs(l); 565 lwp_lock(l); 566 l->l_prflag &= ~LPR_DETACHED; 567 l->l_stat = LSZOMB; 568 lwp_unlock(l); 569 KASSERT(curlwp == l); 570 KASSERT(p->p_nrlwps == 1); 571 KASSERT(p->p_nlwps == 1); 572 p->p_stat = SZOMB; 573 p->p_nrlwps--; 574 p->p_nzlwps++; 575 p->p_ndlwps = 0; 576 mutex_exit(p->p_lock); 577 578 /* 579 * Signal the parent to collect us, and drop the proclist lock. 580 * Drop debugger/procfs lock; no new references can be gained. 581 */ 582 cv_broadcast(&p->p_pptr->p_waitcv); 583 rw_exit(&p->p_reflock); 584 mutex_exit(&proc_lock); 585 586 /* 587 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP! 588 */ 589 590 /* 591 * Give machine-dependent code a chance to free any MD LWP 592 * resources. This must be done before uvm_lwp_exit(), in 593 * case these resources are in the PCB. 594 */ 595 cpu_lwp_free(l, 1); 596 597 /* Switch away into oblivion. */ 598 lwp_lock(l); 599 spc_lock(l->l_cpu); 600 mi_switch(l); 601 panic("exit1"); 602 } 603 604 void 605 exit_lwps(struct lwp *l) 606 { 607 proc_t *p = l->l_proc; 608 lwp_t *l2; 609 610 retry: 611 KASSERT(mutex_owned(p->p_lock)); 612 613 /* 614 * Interrupt LWPs in interruptable sleep, unsuspend suspended 615 * LWPs and then wait for everyone else to finish. 616 */ 617 LIST_FOREACH(l2, &p->p_lwps, l_sibling) { 618 if (l2 == l) 619 continue; 620 lwp_lock(l2); 621 l2->l_flag |= LW_WEXIT; 622 if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) || 623 l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) { 624 l2->l_flag &= ~LW_DBGSUSPEND; 625 /* setrunnable() will release the lock. */ 626 setrunnable(l2); 627 continue; 628 } 629 lwp_need_userret(l2); 630 lwp_unlock(l2); 631 } 632 633 /* 634 * Wait for every LWP to exit. Note: LWPs can get suspended/slept 635 * behind us or there may even be new LWPs created. Therefore, a 636 * full retry is required on error. 637 */ 638 while (p->p_nlwps > 1) { 639 if (lwp_wait(l, 0, NULL, true)) { 640 goto retry; 641 } 642 } 643 644 KASSERT(p->p_nlwps == 1); 645 } 646 647 int 648 do_sys_waitid(idtype_t idtype, id_t id, int *pid, int *status, int options, 649 struct wrusage *wru, siginfo_t *si) 650 { 651 proc_t *child; 652 int error; 653 654 655 if (wru != NULL) 656 memset(wru, 0, sizeof(*wru)); 657 if (si != NULL) 658 memset(si, 0, sizeof(*si)); 659 660 mutex_enter(&proc_lock); 661 error = find_stopped_child(curproc, idtype, id, options, &child, 662 wru, si); 663 if (child == NULL) { 664 mutex_exit(&proc_lock); 665 *pid = 0; 666 *status = 0; 667 return error; 668 } 669 *pid = child->p_pid; 670 671 if (child->p_stat == SZOMB) { 672 /* Child is exiting */ 673 *status = P_WAITSTATUS(child); 674 /* proc_free() will release the proc_lock. */ 675 if (options & WNOWAIT) { 676 mutex_exit(&proc_lock); 677 } else { 678 proc_free(child, wru); 679 } 680 } else { 681 /* Don't mark SIGCONT if we are being stopped */ 682 *status = (child->p_xsig == SIGCONT && child->p_stat != SSTOP) ? 683 W_CONTCODE() : W_STOPCODE(child->p_xsig); 684 mutex_exit(&proc_lock); 685 } 686 return 0; 687 } 688 689 int 690 do_sys_wait(int *pid, int *status, int options, struct rusage *ru) 691 { 692 idtype_t idtype; 693 id_t id; 694 int ret; 695 struct wrusage wru; 696 697 /* 698 * Translate the special pid values into the (idtype, pid) 699 * pair for wait6. The WAIT_MYPGRP case is handled by 700 * find_stopped_child() on its own. 701 */ 702 if (*pid == WAIT_ANY) { 703 idtype = P_ALL; 704 id = 0; 705 } else if (*pid < 0) { 706 idtype = P_PGID; 707 id = (id_t)-*pid; 708 } else { 709 idtype = P_PID; 710 id = (id_t)*pid; 711 } 712 options |= WEXITED | WTRAPPED; 713 ret = do_sys_waitid(idtype, id, pid, status, options, ru ? &wru : NULL, 714 NULL); 715 if (ru) 716 *ru = wru.wru_self; 717 return ret; 718 } 719 720 int 721 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap, 722 register_t *retval) 723 { 724 /* { 725 syscallarg(int) pid; 726 syscallarg(int *) status; 727 syscallarg(int) options; 728 syscallarg(struct rusage *) rusage; 729 } */ 730 int error, status, pid = SCARG(uap, pid); 731 struct rusage ru; 732 733 error = do_sys_wait(&pid, &status, SCARG(uap, options), 734 SCARG(uap, rusage) != NULL ? &ru : NULL); 735 736 retval[0] = pid; 737 if (pid == 0) { 738 return error; 739 } 740 if (SCARG(uap, status)) { 741 error = copyout(&status, SCARG(uap, status), sizeof(status)); 742 } 743 if (SCARG(uap, rusage) && error == 0) { 744 error = copyout(&ru, SCARG(uap, rusage), sizeof(ru)); 745 } 746 return error; 747 } 748 749 int 750 sys_wait6(struct lwp *l, const struct sys_wait6_args *uap, register_t *retval) 751 { 752 /* { 753 syscallarg(idtype_t) idtype; 754 syscallarg(id_t) id; 755 syscallarg(int *) status; 756 syscallarg(int) options; 757 syscallarg(struct wrusage *) wru; 758 syscallarg(siginfo_t *) si; 759 } */ 760 struct wrusage wru, *wrup; 761 siginfo_t si, *sip; 762 idtype_t idtype; 763 int pid; 764 id_t id; 765 int error, status; 766 767 idtype = SCARG(uap, idtype); 768 id = SCARG(uap, id); 769 770 if (SCARG(uap, wru) != NULL) 771 wrup = &wru; 772 else 773 wrup = NULL; 774 775 if (SCARG(uap, info) != NULL) 776 sip = &si; 777 else 778 sip = NULL; 779 780 /* 781 * We expect all callers of wait6() to know about WEXITED and 782 * WTRAPPED. 783 */ 784 error = do_sys_waitid(idtype, id, &pid, &status, SCARG(uap, options), 785 wrup, sip); 786 787 retval[0] = pid; /* tell userland who it was */ 788 789 #if 0 790 /* 791 * should we copyout if there was no process, hence no useful data? 792 * We don't for an old sytle wait4() (etc) but I believe 793 * FreeBSD does for wait6(), so a tossup... Go with FreeBSD for now. 794 */ 795 if (pid == 0) 796 return error; 797 #endif 798 799 if (SCARG(uap, status) != NULL && error == 0) 800 error = copyout(&status, SCARG(uap, status), sizeof(status)); 801 if (SCARG(uap, wru) != NULL && error == 0) 802 error = copyout(&wru, SCARG(uap, wru), sizeof(wru)); 803 if (SCARG(uap, info) != NULL && error == 0) 804 error = copyout(&si, SCARG(uap, info), sizeof(si)); 805 return error; 806 } 807 808 809 /* 810 * Find a process that matches the provided criteria, and fill siginfo 811 * and resources if found. 812 * Returns: 813 * -1: Not found, abort early 814 * 0: Not matched 815 * 1: Matched, there might be more matches 816 * 2: This is the only match 817 */ 818 static int 819 match_process(const struct proc *pp, struct proc **q, idtype_t idtype, id_t id, 820 int options, struct wrusage *wrusage, siginfo_t *siginfo) 821 { 822 struct rusage *rup; 823 struct proc *p = *q; 824 int rv = 1; 825 826 mutex_enter(p->p_lock); 827 switch (idtype) { 828 case P_ALL: 829 break; 830 case P_PID: 831 if (p->p_pid != (pid_t)id) { 832 mutex_exit(p->p_lock); 833 p = *q = proc_find_raw((pid_t)id); 834 if (p == NULL || p->p_stat == SIDL || p->p_pptr != pp) { 835 *q = NULL; 836 return -1; 837 } 838 mutex_enter(p->p_lock); 839 } 840 rv++; 841 break; 842 case P_PGID: 843 if (p->p_pgid != (pid_t)id) 844 goto out; 845 break; 846 case P_SID: 847 if (p->p_session->s_sid != (pid_t)id) 848 goto out; 849 break; 850 case P_UID: 851 if (kauth_cred_geteuid(p->p_cred) != (uid_t)id) 852 goto out; 853 break; 854 case P_GID: 855 if (kauth_cred_getegid(p->p_cred) != (gid_t)id) 856 goto out; 857 break; 858 case P_CID: 859 case P_PSETID: 860 case P_CPUID: 861 /* XXX: Implement me */ 862 default: 863 out: 864 mutex_exit(p->p_lock); 865 return 0; 866 } 867 868 if ((options & WEXITED) == 0 && p->p_stat == SZOMB) 869 goto out; 870 871 if (siginfo != NULL) { 872 siginfo->si_errno = 0; 873 874 /* 875 * SUSv4 requires that the si_signo value is always 876 * SIGCHLD. Obey it despite the rfork(2) interface 877 * allows to request other signal for child exit 878 * notification. 879 */ 880 siginfo->si_signo = SIGCHLD; 881 882 /* 883 * This is still a rough estimate. We will fix the 884 * cases TRAPPED, STOPPED, and CONTINUED later. 885 */ 886 if (p->p_sflag & PS_COREDUMP) { 887 siginfo->si_code = CLD_DUMPED; 888 siginfo->si_status = p->p_xsig; 889 } else if (p->p_xsig) { 890 siginfo->si_code = CLD_KILLED; 891 siginfo->si_status = p->p_xsig; 892 } else { 893 siginfo->si_code = CLD_EXITED; 894 siginfo->si_status = p->p_xexit; 895 } 896 897 siginfo->si_pid = p->p_pid; 898 siginfo->si_uid = kauth_cred_geteuid(p->p_cred); 899 siginfo->si_utime = p->p_stats->p_ru.ru_utime.tv_sec; 900 siginfo->si_stime = p->p_stats->p_ru.ru_stime.tv_sec; 901 } 902 903 /* 904 * There should be no reason to limit resources usage info to 905 * exited processes only. A snapshot about any resources used 906 * by a stopped process may be exactly what is needed. 907 */ 908 if (wrusage != NULL) { 909 rup = &wrusage->wru_self; 910 *rup = p->p_stats->p_ru; 911 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 912 913 rup = &wrusage->wru_children; 914 *rup = p->p_stats->p_cru; 915 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 916 } 917 918 mutex_exit(p->p_lock); 919 return rv; 920 } 921 922 /* 923 * Determine if there are existing processes being debugged 924 * that used to be (and sometime later will be again) children 925 * of a specific parent (while matching wait criteria) 926 */ 927 static bool 928 debugged_child_exists(idtype_t idtype, id_t id, int options, siginfo_t *si, 929 const struct proc *parent) 930 { 931 struct proc *pp; 932 933 /* 934 * If we are searching for a specific pid, we can optimise a little 935 */ 936 if (idtype == P_PID) { 937 /* 938 * Check the specific process to see if its real parent is us 939 */ 940 pp = proc_find_raw((pid_t)id); 941 if (pp != NULL && pp->p_stat != SIDL && pp->p_opptr == parent) { 942 /* 943 * using P_ALL here avoids match_process() doing the 944 * same work that we just did, but incorrectly for 945 * this scenario. 946 */ 947 if (match_process(parent, &pp, P_ALL, id, options, 948 NULL, si)) 949 return true; 950 } 951 return false; 952 } 953 954 /* 955 * For the hard cases, just look everywhere to see if some 956 * stolen (reparented) process is really our lost child. 957 * Then check if that process could satisfy the wait conditions. 958 */ 959 960 /* 961 * XXX inefficient, but hopefully fairly rare. 962 * XXX should really use a list of reparented processes. 963 */ 964 PROCLIST_FOREACH(pp, &allproc) { 965 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 966 continue; 967 if (pp->p_opptr == parent && 968 match_process(parent, &pp, idtype, id, options, NULL, si)) 969 return true; 970 } 971 PROCLIST_FOREACH(pp, &zombproc) { 972 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 973 continue; 974 if (pp->p_opptr == parent && 975 match_process(parent, &pp, idtype, id, options, NULL, si)) 976 return true; 977 } 978 979 return false; 980 } 981 982 /* 983 * Scan list of child processes for a child process that has stopped or 984 * exited. Used by sys_wait4 and 'compat' equivalents. 985 * 986 * Must be called with the proc_lock held, and may release while waiting. 987 */ 988 static int 989 find_stopped_child(struct proc *parent, idtype_t idtype, id_t id, int options, 990 struct proc **child_p, struct wrusage *wru, siginfo_t *si) 991 { 992 struct proc *child, *dead; 993 int error; 994 995 KASSERT(mutex_owned(&proc_lock)); 996 997 if (options & ~WALLOPTS) { 998 *child_p = NULL; 999 return EINVAL; 1000 } 1001 1002 if ((options & WSELECTOPTS) == 0) { 1003 /* 1004 * We will be unable to find any matching processes, 1005 * because there are no known events to look for. 1006 * Prefer to return error instead of blocking 1007 * indefinitely. 1008 */ 1009 *child_p = NULL; 1010 return EINVAL; 1011 } 1012 1013 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1014 mutex_enter(parent->p_lock); 1015 id = (id_t)parent->p_pgid; 1016 mutex_exit(parent->p_lock); 1017 idtype = P_PGID; 1018 } 1019 1020 for (;;) { 1021 error = ECHILD; 1022 dead = NULL; 1023 1024 LIST_FOREACH(child, &parent->p_children, p_sibling) { 1025 int rv = match_process(parent, &child, idtype, id, 1026 options, wru, si); 1027 if (rv == -1) 1028 break; 1029 if (rv == 0) 1030 continue; 1031 1032 /* 1033 * Wait for processes with p_exitsig != SIGCHLD 1034 * processes only if WALTSIG is set; wait for 1035 * processes with p_exitsig == SIGCHLD only 1036 * if WALTSIG is clear. 1037 */ 1038 if (((options & WALLSIG) == 0) && 1039 (options & WALTSIG ? child->p_exitsig == SIGCHLD 1040 : P_EXITSIG(child) != SIGCHLD)){ 1041 if (rv == 2) { 1042 child = NULL; 1043 break; 1044 } 1045 continue; 1046 } 1047 1048 error = 0; 1049 if ((options & WNOZOMBIE) == 0) { 1050 if (child->p_stat == SZOMB) 1051 break; 1052 if (child->p_stat == SDEAD) { 1053 /* 1054 * We may occasionally arrive here 1055 * after receiving a signal, but 1056 * immediately before the child 1057 * process is zombified. The wait 1058 * will be short, so avoid returning 1059 * to userspace. 1060 */ 1061 dead = child; 1062 } 1063 } 1064 1065 if ((options & WCONTINUED) != 0 && 1066 child->p_xsig == SIGCONT && 1067 (child->p_sflag & PS_CONTINUED)) { 1068 if ((options & WNOWAIT) == 0) { 1069 child->p_sflag &= ~PS_CONTINUED; 1070 child->p_waited = 1; 1071 parent->p_nstopchild--; 1072 } 1073 if (si) { 1074 si->si_status = child->p_xsig; 1075 si->si_code = CLD_CONTINUED; 1076 } 1077 break; 1078 } 1079 1080 if ((options & (WTRAPPED|WSTOPPED)) != 0 && 1081 child->p_stat == SSTOP && 1082 child->p_waited == 0 && 1083 ((child->p_slflag & PSL_TRACED) || 1084 options & (WUNTRACED|WSTOPPED))) { 1085 if ((options & WNOWAIT) == 0) { 1086 child->p_waited = 1; 1087 parent->p_nstopchild--; 1088 } 1089 if (si) { 1090 si->si_status = child->p_xsig; 1091 si->si_code = 1092 (child->p_slflag & PSL_TRACED) ? 1093 CLD_TRAPPED : CLD_STOPPED; 1094 } 1095 break; 1096 } 1097 if (parent->p_nstopchild == 0 || rv == 2) { 1098 child = NULL; 1099 break; 1100 } 1101 } 1102 1103 /* 1104 * If we found nothing, but we are the bereaved parent 1105 * of a stolen child, look and see if that child (or 1106 * one of them) meets our search criteria. If so, then 1107 * we cannot succeed, but we can hang (wait...), 1108 * or if WNOHANG, return 0 instead of ECHILD 1109 */ 1110 if (child == NULL && error == ECHILD && 1111 (parent->p_slflag & PSL_CHTRACED) && 1112 debugged_child_exists(idtype, id, options, si, parent)) 1113 error = 0; 1114 1115 if (child != NULL || error != 0 || 1116 ((options & WNOHANG) != 0 && dead == NULL)) { 1117 *child_p = child; 1118 return error; 1119 } 1120 1121 /* 1122 * Wait for another child process to stop. 1123 */ 1124 error = cv_wait_sig(&parent->p_waitcv, &proc_lock); 1125 1126 if (error != 0) { 1127 *child_p = NULL; 1128 return error; 1129 } 1130 } 1131 } 1132 1133 /* 1134 * Free a process after parent has taken all the state info. Must be called 1135 * with the proclist lock held, and will release before returning. 1136 * 1137 * *ru is returned to the caller, and must be freed by the caller. 1138 */ 1139 static void 1140 proc_free(struct proc *p, struct wrusage *wru) 1141 { 1142 struct proc *parent = p->p_pptr; 1143 struct lwp *l; 1144 ksiginfo_t ksi; 1145 kauth_cred_t cred1, cred2; 1146 uid_t uid; 1147 1148 KASSERT(mutex_owned(&proc_lock)); 1149 KASSERT(p->p_nlwps == 1); 1150 KASSERT(p->p_nzlwps == 1); 1151 KASSERT(p->p_nrlwps == 0); 1152 KASSERT(p->p_stat == SZOMB); 1153 1154 /* 1155 * If we got the child via ptrace(2) or procfs, and 1156 * the parent is different (meaning the process was 1157 * attached, rather than run as a child), then we need 1158 * to give it back to the old parent, and send the 1159 * parent the exit signal. The rest of the cleanup 1160 * will be done when the old parent waits on the child. 1161 */ 1162 if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) { 1163 mutex_enter(p->p_lock); 1164 p->p_slflag &= ~(PSL_TRACED|PSL_SYSCALL); 1165 mutex_exit(p->p_lock); 1166 parent = (p->p_opptr == NULL) ? initproc : p->p_opptr; 1167 proc_reparent(p, parent); 1168 p->p_opptr = NULL; 1169 if (p->p_exitsig != 0) { 1170 exit_psignal(p, parent, &ksi); 1171 kpsignal(parent, &ksi, NULL); 1172 } 1173 cv_broadcast(&parent->p_waitcv); 1174 mutex_exit(&proc_lock); 1175 return; 1176 } 1177 1178 sched_proc_exit(parent, p); 1179 1180 /* 1181 * Add child times of exiting process onto its own times. 1182 * This cannot be done any earlier else it might get done twice. 1183 */ 1184 l = LIST_FIRST(&p->p_lwps); 1185 p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw); 1186 p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw; 1187 ruadd(&p->p_stats->p_ru, &l->l_ru); 1188 ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru); 1189 ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru); 1190 if (wru != NULL) { 1191 wru->wru_self = p->p_stats->p_ru; 1192 wru->wru_children = p->p_stats->p_cru; 1193 } 1194 p->p_xsig = 0; 1195 p->p_xexit = 0; 1196 1197 /* 1198 * At this point we are going to start freeing the final resources. 1199 * If anyone tries to access the proc structure after here they will 1200 * get a shock - bits are missing. Attempt to make it hard! We 1201 * don't bother with any further locking past this point. 1202 */ 1203 p->p_stat = SIDL; /* not even a zombie any more */ 1204 LIST_REMOVE(p, p_list); /* off zombproc */ 1205 parent->p_nstopchild--; 1206 LIST_REMOVE(p, p_sibling); 1207 1208 /* 1209 * Let pid be reallocated. 1210 */ 1211 proc_free_pid(p->p_pid); 1212 atomic_dec_uint(&nprocs); 1213 1214 /* 1215 * Unlink process from its process group. 1216 * Releases the proc_lock. 1217 */ 1218 proc_leavepgrp(p); 1219 1220 /* 1221 * Delay release until after lwp_free. 1222 */ 1223 cred2 = l->l_cred; 1224 1225 /* 1226 * Free the last LWP's resources. 1227 * 1228 * lwp_free ensures the LWP is no longer running on another CPU. 1229 */ 1230 lwp_free(l, false, true); 1231 1232 /* 1233 * Now no one except us can reach the process p. 1234 */ 1235 1236 /* 1237 * Decrement the count of procs running with this uid. 1238 */ 1239 cred1 = p->p_cred; 1240 uid = kauth_cred_getuid(cred1); 1241 (void)chgproccnt(uid, -1); 1242 1243 /* 1244 * Release substructures. 1245 */ 1246 1247 lim_free(p->p_limit); 1248 pstatsfree(p->p_stats); 1249 kauth_cred_free(cred1); 1250 kauth_cred_free(cred2); 1251 1252 /* 1253 * Release reference to text vnode 1254 */ 1255 if (p->p_textvp) 1256 vrele(p->p_textvp); 1257 kmem_strfree(p->p_path); 1258 1259 mutex_destroy(&p->p_auxlock); 1260 mutex_obj_free(p->p_lock); 1261 mutex_destroy(&p->p_stmutex); 1262 cv_destroy(&p->p_waitcv); 1263 cv_destroy(&p->p_lwpcv); 1264 rw_destroy(&p->p_reflock); 1265 1266 proc_free_mem(p); 1267 } 1268 1269 /* 1270 * Change the parent of a process for tracing purposes. 1271 */ 1272 void 1273 proc_changeparent(struct proc *t, struct proc *p) 1274 { 1275 SET(t->p_slflag, PSL_TRACED); 1276 t->p_opptr = t->p_pptr; 1277 if (t->p_pptr == p) 1278 return; 1279 struct proc *parent = t->p_pptr; 1280 1281 if (parent->p_lock < t->p_lock) { 1282 if (!mutex_tryenter(parent->p_lock)) { 1283 mutex_exit(t->p_lock); 1284 mutex_enter(parent->p_lock); 1285 mutex_enter(t->p_lock); 1286 } 1287 } else if (parent->p_lock > t->p_lock) { 1288 mutex_enter(parent->p_lock); 1289 } 1290 parent->p_slflag |= PSL_CHTRACED; 1291 proc_reparent(t, p); 1292 if (parent->p_lock != t->p_lock) 1293 mutex_exit(parent->p_lock); 1294 } 1295 1296 /* 1297 * make process 'parent' the new parent of process 'child'. 1298 * 1299 * Must be called with proc_lock held. 1300 */ 1301 void 1302 proc_reparent(struct proc *child, struct proc *parent) 1303 { 1304 1305 KASSERT(mutex_owned(&proc_lock)); 1306 1307 if (child->p_pptr == parent) 1308 return; 1309 1310 if (child->p_stat == SZOMB || child->p_stat == SDEAD || 1311 (child->p_stat == SSTOP && !child->p_waited)) { 1312 child->p_pptr->p_nstopchild--; 1313 parent->p_nstopchild++; 1314 } 1315 if (parent == initproc) { 1316 child->p_exitsig = SIGCHLD; 1317 child->p_ppid = parent->p_pid; 1318 } 1319 1320 LIST_REMOVE(child, p_sibling); 1321 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1322 child->p_pptr = parent; 1323 } 1324