1 /* $NetBSD: kern_exit.c,v 1.293 2021/12/05 08:13:12 msaitoh 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 1.293 2021/12/05 08:13:12 msaitoh 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 * Reset p_opptr pointer of all former children which got 439 * traced by another process and were reparented. We reset 440 * it to NULL here; the trace detach code then reparents 441 * the child to initproc. We only check allproc list, since 442 * eventual former children on zombproc list won't reference 443 * p_opptr anymore. 444 */ 445 if (__predict_false(p->p_slflag & PSL_CHTRACED)) { 446 struct proc *q; 447 PROCLIST_FOREACH(q, &allproc) { 448 if (q->p_opptr == p) 449 q->p_opptr = NULL; 450 } 451 PROCLIST_FOREACH(q, &zombproc) { 452 if (q->p_opptr == p) 453 q->p_opptr = NULL; 454 } 455 } 456 457 /* 458 * Give orphaned children to init(8). 459 */ 460 child = LIST_FIRST(&p->p_children); 461 wakeinit = (child != NULL); 462 for (; child != NULL; child = next_child) { 463 next_child = LIST_NEXT(child, p_sibling); 464 465 /* 466 * Traced processes are killed since their existence 467 * means someone is screwing up. Since we reset the 468 * trace flags, the logic in sys_wait4() would not be 469 * triggered to reparent the process to its 470 * original parent, so we must do this here. 471 */ 472 if (__predict_false(child->p_slflag & PSL_TRACED)) { 473 mutex_enter(p->p_lock); 474 child->p_slflag &= 475 ~(PSL_TRACED|PSL_SYSCALL); 476 mutex_exit(p->p_lock); 477 if (child->p_opptr != child->p_pptr) { 478 struct proc *t = child->p_opptr; 479 proc_reparent(child, t ? t : initproc); 480 child->p_opptr = NULL; 481 } else 482 proc_reparent(child, initproc); 483 killproc(child, "orphaned traced process"); 484 } else 485 proc_reparent(child, initproc); 486 } 487 488 /* 489 * Move proc from allproc to zombproc, it's now nearly ready to be 490 * collected by parent. 491 */ 492 LIST_REMOVE(l, l_list); 493 LIST_REMOVE(p, p_list); 494 LIST_INSERT_HEAD(&zombproc, p, p_list); 495 496 /* 497 * Mark the process as dead. We must do this before we signal 498 * the parent. 499 */ 500 p->p_stat = SDEAD; 501 502 /* 503 * Let anyone watching this DTrace probe know what we're 504 * on our way out. 505 */ 506 SDT_PROBE(proc, kernel, , exit, 507 ((p->p_sflag & PS_COREDUMP) ? CLD_DUMPED : 508 (p->p_xsig ? CLD_KILLED : CLD_EXITED)), 509 0,0,0,0); 510 511 /* Put in front of parent's sibling list for parent to collect it */ 512 old_parent = p->p_pptr; 513 old_parent->p_nstopchild++; 514 if (LIST_FIRST(&old_parent->p_children) != p) { 515 /* Put child where it can be found quickly */ 516 LIST_REMOVE(p, p_sibling); 517 LIST_INSERT_HEAD(&old_parent->p_children, p, p_sibling); 518 } 519 520 /* 521 * Notify parent that we're gone. If parent has the P_NOCLDWAIT 522 * flag set, notify init instead (and hope it will handle 523 * this situation). 524 */ 525 if (old_parent->p_flag & (PK_NOCLDWAIT|PK_CLDSIGIGN)) { 526 proc_reparent(p, initproc); 527 wakeinit = 1; 528 529 /* 530 * If this was the last child of our parent, notify 531 * parent, so in case he was wait(2)ing, he will 532 * continue. 533 */ 534 if (LIST_FIRST(&old_parent->p_children) == NULL) 535 cv_broadcast(&old_parent->p_waitcv); 536 } 537 538 /* Reload parent pointer, since p may have been reparented above */ 539 new_parent = p->p_pptr; 540 541 if (__predict_false(p->p_exitsig != 0)) { 542 exit_psignal(p, new_parent, &ksi); 543 kpsignal(new_parent, &ksi, NULL); 544 } 545 546 /* Calculate the final rusage info. */ 547 calcru(p, &p->p_stats->p_ru.ru_utime, &p->p_stats->p_ru.ru_stime, 548 NULL, NULL); 549 550 if (wakeinit) 551 cv_broadcast(&initproc->p_waitcv); 552 553 callout_destroy(&l->l_timeout_ch); 554 555 /* 556 * Release any PCU resources before becoming a zombie. 557 */ 558 pcu_discard_all(l); 559 560 mutex_enter(p->p_lock); 561 /* 562 * Notify other processes tracking us with a knote that 563 * we're exiting. 564 * 565 * N.B. we do this here because the process is now SDEAD, 566 * and thus cannot have any more knotes attached. Also, 567 * knote_proc_exit() expects that p->p_lock is already 568 * held (and will assert so). 569 */ 570 if (!SLIST_EMPTY(&p->p_klist)) { 571 knote_proc_exit(p); 572 } 573 574 /* Free the LWP ID */ 575 proc_free_lwpid(p, l->l_lid); 576 lwp_drainrefs(l); 577 lwp_lock(l); 578 l->l_prflag &= ~LPR_DETACHED; 579 l->l_stat = LSZOMB; 580 lwp_unlock(l); 581 KASSERT(curlwp == l); 582 KASSERT(p->p_nrlwps == 1); 583 KASSERT(p->p_nlwps == 1); 584 p->p_stat = SZOMB; 585 p->p_nrlwps--; 586 p->p_nzlwps++; 587 p->p_ndlwps = 0; 588 mutex_exit(p->p_lock); 589 590 /* 591 * Signal the parent to collect us, and drop the proclist lock. 592 * Drop debugger/procfs lock; no new references can be gained. 593 */ 594 cv_broadcast(&p->p_pptr->p_waitcv); 595 rw_exit(&p->p_reflock); 596 mutex_exit(&proc_lock); 597 598 /* 599 * NOTE: WE ARE NO LONGER ALLOWED TO SLEEP! 600 */ 601 602 /* 603 * Give machine-dependent code a chance to free any MD LWP 604 * resources. This must be done before uvm_lwp_exit(), in 605 * case these resources are in the PCB. 606 */ 607 cpu_lwp_free(l, 1); 608 609 /* Switch away into oblivion. */ 610 lwp_lock(l); 611 spc_lock(l->l_cpu); 612 mi_switch(l); 613 panic("exit1"); 614 } 615 616 void 617 exit_lwps(struct lwp *l) 618 { 619 proc_t *p = l->l_proc; 620 lwp_t *l2; 621 622 retry: 623 KASSERT(mutex_owned(p->p_lock)); 624 625 /* 626 * Interrupt LWPs in interruptable sleep, unsuspend suspended 627 * LWPs and then wait for everyone else to finish. 628 */ 629 LIST_FOREACH(l2, &p->p_lwps, l_sibling) { 630 if (l2 == l) 631 continue; 632 lwp_lock(l2); 633 l2->l_flag |= LW_WEXIT; 634 if ((l2->l_stat == LSSLEEP && (l2->l_flag & LW_SINTR)) || 635 l2->l_stat == LSSUSPENDED || l2->l_stat == LSSTOP) { 636 l2->l_flag &= ~LW_DBGSUSPEND; 637 /* setrunnable() will release the lock. */ 638 setrunnable(l2); 639 continue; 640 } 641 lwp_need_userret(l2); 642 lwp_unlock(l2); 643 } 644 645 /* 646 * Wait for every LWP to exit. Note: LWPs can get suspended/slept 647 * behind us or there may even be new LWPs created. Therefore, a 648 * full retry is required on error. 649 */ 650 while (p->p_nlwps > 1) { 651 if (lwp_wait(l, 0, NULL, true)) { 652 goto retry; 653 } 654 } 655 656 KASSERT(p->p_nlwps == 1); 657 } 658 659 int 660 do_sys_waitid(idtype_t idtype, id_t id, int *pid, int *status, int options, 661 struct wrusage *wru, siginfo_t *si) 662 { 663 proc_t *child; 664 int error; 665 666 667 if (wru != NULL) 668 memset(wru, 0, sizeof(*wru)); 669 if (si != NULL) 670 memset(si, 0, sizeof(*si)); 671 672 mutex_enter(&proc_lock); 673 error = find_stopped_child(curproc, idtype, id, options, &child, 674 wru, si); 675 if (child == NULL) { 676 mutex_exit(&proc_lock); 677 *pid = 0; 678 *status = 0; 679 return error; 680 } 681 *pid = child->p_pid; 682 683 if (child->p_stat == SZOMB) { 684 /* Child is exiting */ 685 *status = P_WAITSTATUS(child); 686 /* proc_free() will release the proc_lock. */ 687 if (options & WNOWAIT) { 688 mutex_exit(&proc_lock); 689 } else { 690 proc_free(child, wru); 691 } 692 } else { 693 /* Don't mark SIGCONT if we are being stopped */ 694 *status = (child->p_xsig == SIGCONT && child->p_stat != SSTOP) ? 695 W_CONTCODE() : W_STOPCODE(child->p_xsig); 696 mutex_exit(&proc_lock); 697 } 698 return 0; 699 } 700 701 int 702 do_sys_wait(int *pid, int *status, int options, struct rusage *ru) 703 { 704 idtype_t idtype; 705 id_t id; 706 int ret; 707 struct wrusage wru; 708 709 /* 710 * Translate the special pid values into the (idtype, pid) 711 * pair for wait6. The WAIT_MYPGRP case is handled by 712 * find_stopped_child() on its own. 713 */ 714 if (*pid == WAIT_ANY) { 715 idtype = P_ALL; 716 id = 0; 717 } else if (*pid < 0) { 718 idtype = P_PGID; 719 id = (id_t)-*pid; 720 } else { 721 idtype = P_PID; 722 id = (id_t)*pid; 723 } 724 options |= WEXITED | WTRAPPED; 725 ret = do_sys_waitid(idtype, id, pid, status, options, ru ? &wru : NULL, 726 NULL); 727 if (ru) 728 *ru = wru.wru_self; 729 return ret; 730 } 731 732 int 733 sys___wait450(struct lwp *l, const struct sys___wait450_args *uap, 734 register_t *retval) 735 { 736 /* { 737 syscallarg(int) pid; 738 syscallarg(int *) status; 739 syscallarg(int) options; 740 syscallarg(struct rusage *) rusage; 741 } */ 742 int error, status, pid = SCARG(uap, pid); 743 struct rusage ru; 744 745 error = do_sys_wait(&pid, &status, SCARG(uap, options), 746 SCARG(uap, rusage) != NULL ? &ru : NULL); 747 748 retval[0] = pid; 749 if (pid == 0) { 750 return error; 751 } 752 if (SCARG(uap, status)) { 753 error = copyout(&status, SCARG(uap, status), sizeof(status)); 754 } 755 if (SCARG(uap, rusage) && error == 0) { 756 error = copyout(&ru, SCARG(uap, rusage), sizeof(ru)); 757 } 758 return error; 759 } 760 761 int 762 sys_wait6(struct lwp *l, const struct sys_wait6_args *uap, register_t *retval) 763 { 764 /* { 765 syscallarg(idtype_t) idtype; 766 syscallarg(id_t) id; 767 syscallarg(int *) status; 768 syscallarg(int) options; 769 syscallarg(struct wrusage *) wru; 770 syscallarg(siginfo_t *) si; 771 } */ 772 struct wrusage wru, *wrup; 773 siginfo_t si, *sip; 774 idtype_t idtype; 775 int pid; 776 id_t id; 777 int error, status; 778 779 idtype = SCARG(uap, idtype); 780 id = SCARG(uap, id); 781 782 if (SCARG(uap, wru) != NULL) 783 wrup = &wru; 784 else 785 wrup = NULL; 786 787 if (SCARG(uap, info) != NULL) 788 sip = &si; 789 else 790 sip = NULL; 791 792 /* 793 * We expect all callers of wait6() to know about WEXITED and 794 * WTRAPPED. 795 */ 796 error = do_sys_waitid(idtype, id, &pid, &status, SCARG(uap, options), 797 wrup, sip); 798 799 retval[0] = pid; /* tell userland who it was */ 800 801 #if 0 802 /* 803 * should we copyout if there was no process, hence no useful data? 804 * We don't for an old style wait4() (etc) but I believe 805 * FreeBSD does for wait6(), so a tossup... Go with FreeBSD for now. 806 */ 807 if (pid == 0) 808 return error; 809 #endif 810 811 if (SCARG(uap, status) != NULL && error == 0) 812 error = copyout(&status, SCARG(uap, status), sizeof(status)); 813 if (SCARG(uap, wru) != NULL && error == 0) 814 error = copyout(&wru, SCARG(uap, wru), sizeof(wru)); 815 if (SCARG(uap, info) != NULL && error == 0) 816 error = copyout(&si, SCARG(uap, info), sizeof(si)); 817 return error; 818 } 819 820 821 /* 822 * Find a process that matches the provided criteria, and fill siginfo 823 * and resources if found. 824 * Returns: 825 * -1: Not found, abort early 826 * 0: Not matched 827 * 1: Matched, there might be more matches 828 * 2: This is the only match 829 */ 830 static int 831 match_process(const struct proc *pp, struct proc **q, idtype_t idtype, id_t id, 832 int options, struct wrusage *wrusage, siginfo_t *siginfo) 833 { 834 struct rusage *rup; 835 struct proc *p = *q; 836 int rv = 1; 837 838 mutex_enter(p->p_lock); 839 switch (idtype) { 840 case P_ALL: 841 break; 842 case P_PID: 843 if (p->p_pid != (pid_t)id) { 844 mutex_exit(p->p_lock); 845 p = *q = proc_find_raw((pid_t)id); 846 if (p == NULL || p->p_stat == SIDL || p->p_pptr != pp) { 847 *q = NULL; 848 return -1; 849 } 850 mutex_enter(p->p_lock); 851 } 852 rv++; 853 break; 854 case P_PGID: 855 if (p->p_pgid != (pid_t)id) 856 goto out; 857 break; 858 case P_SID: 859 if (p->p_session->s_sid != (pid_t)id) 860 goto out; 861 break; 862 case P_UID: 863 if (kauth_cred_geteuid(p->p_cred) != (uid_t)id) 864 goto out; 865 break; 866 case P_GID: 867 if (kauth_cred_getegid(p->p_cred) != (gid_t)id) 868 goto out; 869 break; 870 case P_CID: 871 case P_PSETID: 872 case P_CPUID: 873 /* XXX: Implement me */ 874 default: 875 out: 876 mutex_exit(p->p_lock); 877 return 0; 878 } 879 880 if ((options & WEXITED) == 0 && p->p_stat == SZOMB) 881 goto out; 882 883 if (siginfo != NULL) { 884 siginfo->si_errno = 0; 885 886 /* 887 * SUSv4 requires that the si_signo value is always 888 * SIGCHLD. Obey it despite the rfork(2) interface 889 * allows to request other signal for child exit 890 * notification. 891 */ 892 siginfo->si_signo = SIGCHLD; 893 894 /* 895 * This is still a rough estimate. We will fix the 896 * cases TRAPPED, STOPPED, and CONTINUED later. 897 */ 898 if (p->p_sflag & PS_COREDUMP) { 899 siginfo->si_code = CLD_DUMPED; 900 siginfo->si_status = p->p_xsig; 901 } else if (p->p_xsig) { 902 siginfo->si_code = CLD_KILLED; 903 siginfo->si_status = p->p_xsig; 904 } else { 905 siginfo->si_code = CLD_EXITED; 906 siginfo->si_status = p->p_xexit; 907 } 908 909 siginfo->si_pid = p->p_pid; 910 siginfo->si_uid = kauth_cred_geteuid(p->p_cred); 911 siginfo->si_utime = p->p_stats->p_ru.ru_utime.tv_sec; 912 siginfo->si_stime = p->p_stats->p_ru.ru_stime.tv_sec; 913 } 914 915 /* 916 * There should be no reason to limit resources usage info to 917 * exited processes only. A snapshot about any resources used 918 * by a stopped process may be exactly what is needed. 919 */ 920 if (wrusage != NULL) { 921 rup = &wrusage->wru_self; 922 *rup = p->p_stats->p_ru; 923 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 924 925 rup = &wrusage->wru_children; 926 *rup = p->p_stats->p_cru; 927 calcru(p, &rup->ru_utime, &rup->ru_stime, NULL, NULL); 928 } 929 930 mutex_exit(p->p_lock); 931 return rv; 932 } 933 934 /* 935 * Determine if there are existing processes being debugged 936 * that used to be (and sometime later will be again) children 937 * of a specific parent (while matching wait criteria) 938 */ 939 static bool 940 debugged_child_exists(idtype_t idtype, id_t id, int options, siginfo_t *si, 941 const struct proc *parent) 942 { 943 struct proc *pp; 944 945 /* 946 * If we are searching for a specific pid, we can optimise a little 947 */ 948 if (idtype == P_PID) { 949 /* 950 * Check the specific process to see if its real parent is us 951 */ 952 pp = proc_find_raw((pid_t)id); 953 if (pp != NULL && pp->p_stat != SIDL && pp->p_opptr == parent) { 954 /* 955 * using P_ALL here avoids match_process() doing the 956 * same work that we just did, but incorrectly for 957 * this scenario. 958 */ 959 if (match_process(parent, &pp, P_ALL, id, options, 960 NULL, si)) 961 return true; 962 } 963 return false; 964 } 965 966 /* 967 * For the hard cases, just look everywhere to see if some 968 * stolen (reparented) process is really our lost child. 969 * Then check if that process could satisfy the wait conditions. 970 */ 971 972 /* 973 * XXX inefficient, but hopefully fairly rare. 974 * XXX should really use a list of reparented processes. 975 */ 976 PROCLIST_FOREACH(pp, &allproc) { 977 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 978 continue; 979 if (pp->p_opptr == parent && 980 match_process(parent, &pp, idtype, id, options, NULL, si)) 981 return true; 982 } 983 PROCLIST_FOREACH(pp, &zombproc) { 984 if (pp->p_stat == SIDL) /* XXX impossible ?? */ 985 continue; 986 if (pp->p_opptr == parent && 987 match_process(parent, &pp, idtype, id, options, NULL, si)) 988 return true; 989 } 990 991 return false; 992 } 993 994 /* 995 * Scan list of child processes for a child process that has stopped or 996 * exited. Used by sys_wait4 and 'compat' equivalents. 997 * 998 * Must be called with the proc_lock held, and may release while waiting. 999 */ 1000 static int 1001 find_stopped_child(struct proc *parent, idtype_t idtype, id_t id, int options, 1002 struct proc **child_p, struct wrusage *wru, siginfo_t *si) 1003 { 1004 struct proc *child, *dead; 1005 int error; 1006 1007 KASSERT(mutex_owned(&proc_lock)); 1008 1009 if (options & ~WALLOPTS) { 1010 *child_p = NULL; 1011 return EINVAL; 1012 } 1013 1014 if ((options & WSELECTOPTS) == 0) { 1015 /* 1016 * We will be unable to find any matching processes, 1017 * because there are no known events to look for. 1018 * Prefer to return error instead of blocking 1019 * indefinitely. 1020 */ 1021 *child_p = NULL; 1022 return EINVAL; 1023 } 1024 1025 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1026 mutex_enter(parent->p_lock); 1027 id = (id_t)parent->p_pgid; 1028 mutex_exit(parent->p_lock); 1029 idtype = P_PGID; 1030 } 1031 1032 for (;;) { 1033 error = ECHILD; 1034 dead = NULL; 1035 1036 LIST_FOREACH(child, &parent->p_children, p_sibling) { 1037 int rv = match_process(parent, &child, idtype, id, 1038 options, wru, si); 1039 if (rv == -1) 1040 break; 1041 if (rv == 0) 1042 continue; 1043 1044 /* 1045 * Wait for processes with p_exitsig != SIGCHLD 1046 * processes only if WALTSIG is set; wait for 1047 * processes with p_exitsig == SIGCHLD only 1048 * if WALTSIG is clear. 1049 */ 1050 if (((options & WALLSIG) == 0) && 1051 (options & WALTSIG ? child->p_exitsig == SIGCHLD 1052 : P_EXITSIG(child) != SIGCHLD)){ 1053 if (rv == 2) { 1054 child = NULL; 1055 break; 1056 } 1057 continue; 1058 } 1059 1060 error = 0; 1061 if ((options & WNOZOMBIE) == 0) { 1062 if (child->p_stat == SZOMB) 1063 break; 1064 if (child->p_stat == SDEAD) { 1065 /* 1066 * We may occasionally arrive here 1067 * after receiving a signal, but 1068 * immediately before the child 1069 * process is zombified. The wait 1070 * will be short, so avoid returning 1071 * to userspace. 1072 */ 1073 dead = child; 1074 } 1075 } 1076 1077 if ((options & WCONTINUED) != 0 && 1078 child->p_xsig == SIGCONT && 1079 (child->p_sflag & PS_CONTINUED)) { 1080 if ((options & WNOWAIT) == 0) { 1081 child->p_sflag &= ~PS_CONTINUED; 1082 child->p_waited = 1; 1083 parent->p_nstopchild--; 1084 } 1085 if (si) { 1086 si->si_status = child->p_xsig; 1087 si->si_code = CLD_CONTINUED; 1088 } 1089 break; 1090 } 1091 1092 if ((options & (WTRAPPED|WSTOPPED)) != 0 && 1093 child->p_stat == SSTOP && 1094 child->p_waited == 0 && 1095 ((child->p_slflag & PSL_TRACED) || 1096 options & (WUNTRACED|WSTOPPED))) { 1097 if ((options & WNOWAIT) == 0) { 1098 child->p_waited = 1; 1099 parent->p_nstopchild--; 1100 } 1101 if (si) { 1102 si->si_status = child->p_xsig; 1103 si->si_code = 1104 (child->p_slflag & PSL_TRACED) ? 1105 CLD_TRAPPED : CLD_STOPPED; 1106 } 1107 break; 1108 } 1109 if (parent->p_nstopchild == 0 || rv == 2) { 1110 child = NULL; 1111 break; 1112 } 1113 } 1114 1115 /* 1116 * If we found nothing, but we are the bereaved parent 1117 * of a stolen child, look and see if that child (or 1118 * one of them) meets our search criteria. If so, then 1119 * we cannot succeed, but we can hang (wait...), 1120 * or if WNOHANG, return 0 instead of ECHILD 1121 */ 1122 if (child == NULL && error == ECHILD && 1123 (parent->p_slflag & PSL_CHTRACED) && 1124 debugged_child_exists(idtype, id, options, si, parent)) 1125 error = 0; 1126 1127 if (child != NULL || error != 0 || 1128 ((options & WNOHANG) != 0 && dead == NULL)) { 1129 *child_p = child; 1130 return error; 1131 } 1132 1133 /* 1134 * Wait for another child process to stop. 1135 */ 1136 error = cv_wait_sig(&parent->p_waitcv, &proc_lock); 1137 1138 if (error != 0) { 1139 *child_p = NULL; 1140 return error; 1141 } 1142 } 1143 } 1144 1145 /* 1146 * Free a process after parent has taken all the state info. Must be called 1147 * with the proclist lock held, and will release before returning. 1148 * 1149 * *ru is returned to the caller, and must be freed by the caller. 1150 */ 1151 static void 1152 proc_free(struct proc *p, struct wrusage *wru) 1153 { 1154 struct proc *parent = p->p_pptr; 1155 struct lwp *l; 1156 ksiginfo_t ksi; 1157 kauth_cred_t cred1, cred2; 1158 uid_t uid; 1159 1160 KASSERT(mutex_owned(&proc_lock)); 1161 KASSERT(p->p_nlwps == 1); 1162 KASSERT(p->p_nzlwps == 1); 1163 KASSERT(p->p_nrlwps == 0); 1164 KASSERT(p->p_stat == SZOMB); 1165 1166 /* 1167 * If we got the child via ptrace(2) or procfs, and 1168 * the parent is different (meaning the process was 1169 * attached, rather than run as a child), then we need 1170 * to give it back to the old parent, and send the 1171 * parent the exit signal. The rest of the cleanup 1172 * will be done when the old parent waits on the child. 1173 */ 1174 if ((p->p_slflag & PSL_TRACED) != 0 && p->p_opptr != parent) { 1175 mutex_enter(p->p_lock); 1176 p->p_slflag &= ~(PSL_TRACED|PSL_SYSCALL); 1177 mutex_exit(p->p_lock); 1178 parent = (p->p_opptr == NULL) ? initproc : p->p_opptr; 1179 proc_reparent(p, parent); 1180 p->p_opptr = NULL; 1181 if (p->p_exitsig != 0) { 1182 exit_psignal(p, parent, &ksi); 1183 kpsignal(parent, &ksi, NULL); 1184 } 1185 cv_broadcast(&parent->p_waitcv); 1186 mutex_exit(&proc_lock); 1187 return; 1188 } 1189 1190 sched_proc_exit(parent, p); 1191 1192 /* 1193 * Add child times of exiting process onto its own times. 1194 * This cannot be done any earlier else it might get done twice. 1195 */ 1196 l = LIST_FIRST(&p->p_lwps); 1197 p->p_stats->p_ru.ru_nvcsw += (l->l_ncsw - l->l_nivcsw); 1198 p->p_stats->p_ru.ru_nivcsw += l->l_nivcsw; 1199 ruadd(&p->p_stats->p_ru, &l->l_ru); 1200 ruadd(&p->p_stats->p_ru, &p->p_stats->p_cru); 1201 ruadd(&parent->p_stats->p_cru, &p->p_stats->p_ru); 1202 if (wru != NULL) { 1203 wru->wru_self = p->p_stats->p_ru; 1204 wru->wru_children = p->p_stats->p_cru; 1205 } 1206 p->p_xsig = 0; 1207 p->p_xexit = 0; 1208 1209 /* 1210 * At this point we are going to start freeing the final resources. 1211 * If anyone tries to access the proc structure after here they will 1212 * get a shock - bits are missing. Attempt to make it hard! We 1213 * don't bother with any further locking past this point. 1214 */ 1215 p->p_stat = SIDL; /* not even a zombie any more */ 1216 LIST_REMOVE(p, p_list); /* off zombproc */ 1217 parent->p_nstopchild--; 1218 LIST_REMOVE(p, p_sibling); 1219 1220 /* 1221 * Let pid be reallocated. 1222 */ 1223 proc_free_pid(p->p_pid); 1224 atomic_dec_uint(&nprocs); 1225 1226 /* 1227 * Unlink process from its process group. 1228 * Releases the proc_lock. 1229 */ 1230 proc_leavepgrp(p); 1231 1232 /* 1233 * Delay release until after lwp_free. 1234 */ 1235 cred2 = l->l_cred; 1236 1237 /* 1238 * Free the last LWP's resources. 1239 * 1240 * lwp_free ensures the LWP is no longer running on another CPU. 1241 */ 1242 lwp_free(l, false, true); 1243 1244 /* 1245 * Now no one except us can reach the process p. 1246 */ 1247 1248 /* 1249 * Decrement the count of procs running with this uid. 1250 */ 1251 cred1 = p->p_cred; 1252 uid = kauth_cred_getuid(cred1); 1253 (void)chgproccnt(uid, -1); 1254 1255 /* 1256 * Release substructures. 1257 */ 1258 1259 lim_free(p->p_limit); 1260 pstatsfree(p->p_stats); 1261 kauth_cred_free(cred1); 1262 kauth_cred_free(cred2); 1263 1264 /* 1265 * Release reference to text vnode 1266 */ 1267 if (p->p_textvp) 1268 vrele(p->p_textvp); 1269 kmem_strfree(p->p_path); 1270 1271 mutex_destroy(&p->p_auxlock); 1272 mutex_obj_free(p->p_lock); 1273 mutex_destroy(&p->p_stmutex); 1274 cv_destroy(&p->p_waitcv); 1275 cv_destroy(&p->p_lwpcv); 1276 rw_destroy(&p->p_reflock); 1277 1278 proc_free_mem(p); 1279 } 1280 1281 /* 1282 * Change the parent of a process for tracing purposes. 1283 */ 1284 void 1285 proc_changeparent(struct proc *t, struct proc *p) 1286 { 1287 SET(t->p_slflag, PSL_TRACED); 1288 t->p_opptr = t->p_pptr; 1289 if (t->p_pptr == p) 1290 return; 1291 struct proc *parent = t->p_pptr; 1292 1293 if (parent->p_lock < t->p_lock) { 1294 if (!mutex_tryenter(parent->p_lock)) { 1295 mutex_exit(t->p_lock); 1296 mutex_enter(parent->p_lock); 1297 mutex_enter(t->p_lock); 1298 } 1299 } else if (parent->p_lock > t->p_lock) { 1300 mutex_enter(parent->p_lock); 1301 } 1302 parent->p_slflag |= PSL_CHTRACED; 1303 proc_reparent(t, p); 1304 if (parent->p_lock != t->p_lock) 1305 mutex_exit(parent->p_lock); 1306 } 1307 1308 /* 1309 * make process 'parent' the new parent of process 'child'. 1310 * 1311 * Must be called with proc_lock held. 1312 */ 1313 void 1314 proc_reparent(struct proc *child, struct proc *parent) 1315 { 1316 1317 KASSERT(mutex_owned(&proc_lock)); 1318 1319 if (child->p_pptr == parent) 1320 return; 1321 1322 if (child->p_stat == SZOMB || child->p_stat == SDEAD || 1323 (child->p_stat == SSTOP && !child->p_waited)) { 1324 child->p_pptr->p_nstopchild--; 1325 parent->p_nstopchild++; 1326 } 1327 if (parent == initproc) { 1328 child->p_exitsig = SIGCHLD; 1329 child->p_ppid = parent->p_pid; 1330 } 1331 1332 LIST_REMOVE(child, p_sibling); 1333 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1334 child->p_pptr = parent; 1335 } 1336