1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 37 */ 38 39 #include <sys/cdefs.h> 40 #include "opt_ddb.h" 41 #include "opt_ktrace.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/sysproto.h> 46 #include <sys/capsicum.h> 47 #include <sys/eventhandler.h> 48 #include <sys/kernel.h> 49 #include <sys/ktr.h> 50 #include <sys/malloc.h> 51 #include <sys/lock.h> 52 #include <sys/mutex.h> 53 #include <sys/proc.h> 54 #include <sys/procdesc.h> 55 #include <sys/jail.h> 56 #include <sys/tty.h> 57 #include <sys/wait.h> 58 #include <sys/vmmeter.h> 59 #include <sys/vnode.h> 60 #include <sys/racct.h> 61 #include <sys/resourcevar.h> 62 #include <sys/sbuf.h> 63 #include <sys/signalvar.h> 64 #include <sys/sched.h> 65 #include <sys/sx.h> 66 #include <sys/syscallsubr.h> 67 #include <sys/sysctl.h> 68 #include <sys/syslog.h> 69 #include <sys/ptrace.h> 70 #include <sys/acct.h> /* for acct_process() function prototype */ 71 #include <sys/filedesc.h> 72 #include <sys/sdt.h> 73 #include <sys/shm.h> 74 #include <sys/sem.h> 75 #include <sys/sysent.h> 76 #include <sys/timers.h> 77 #include <sys/umtxvar.h> 78 #ifdef KTRACE 79 #include <sys/ktrace.h> 80 #endif 81 82 #include <security/audit/audit.h> 83 #include <security/mac/mac_framework.h> 84 85 #include <vm/vm.h> 86 #include <vm/vm_extern.h> 87 #include <vm/vm_param.h> 88 #include <vm/pmap.h> 89 #include <vm/vm_map.h> 90 #include <vm/vm_page.h> 91 #include <vm/uma.h> 92 93 #ifdef KDTRACE_HOOKS 94 #include <sys/dtrace_bsd.h> 95 dtrace_execexit_func_t dtrace_fasttrap_exit; 96 #endif 97 98 SDT_PROVIDER_DECLARE(proc); 99 SDT_PROBE_DEFINE1(proc, , , exit, "int"); 100 101 static int kern_kill_on_dbg_exit = 1; 102 SYSCTL_INT(_kern, OID_AUTO, kill_on_debugger_exit, CTLFLAG_RWTUN, 103 &kern_kill_on_dbg_exit, 0, 104 "Kill ptraced processes when debugger exits"); 105 106 static bool kern_wait_dequeue_sigchld = 1; 107 SYSCTL_BOOL(_kern, OID_AUTO, wait_dequeue_sigchld, CTLFLAG_RWTUN, 108 &kern_wait_dequeue_sigchld, 0, 109 "Dequeue SIGCHLD on wait(2) for live process"); 110 111 struct proc * 112 proc_realparent(struct proc *child) 113 { 114 struct proc *p, *parent; 115 116 sx_assert(&proctree_lock, SX_LOCKED); 117 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) 118 return (child->p_pptr->p_pid == child->p_oppid ? 119 child->p_pptr : child->p_reaper); 120 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 121 /* Cannot use LIST_PREV(), since the list head is not known. */ 122 p = __containerof(p->p_orphan.le_prev, struct proc, 123 p_orphan.le_next); 124 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 125 ("missing P_ORPHAN %p", p)); 126 } 127 parent = __containerof(p->p_orphan.le_prev, struct proc, 128 p_orphans.lh_first); 129 return (parent); 130 } 131 132 void 133 reaper_abandon_children(struct proc *p, bool exiting) 134 { 135 struct proc *p1, *p2, *ptmp; 136 137 sx_assert(&proctree_lock, SX_XLOCKED); 138 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 139 if ((p->p_treeflag & P_TREE_REAPER) == 0) 140 return; 141 p1 = p->p_reaper; 142 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 143 LIST_REMOVE(p2, p_reapsibling); 144 p2->p_reaper = p1; 145 p2->p_reapsubtree = p->p_reapsubtree; 146 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 147 if (exiting && p2->p_pptr == p) { 148 PROC_LOCK(p2); 149 proc_reparent(p2, p1, true); 150 PROC_UNLOCK(p2); 151 } 152 } 153 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 154 p->p_treeflag &= ~P_TREE_REAPER; 155 } 156 157 static void 158 reaper_clear(struct proc *p) 159 { 160 struct proc *p1; 161 bool clear; 162 163 sx_assert(&proctree_lock, SX_LOCKED); 164 LIST_REMOVE(p, p_reapsibling); 165 if (p->p_reapsubtree == 1) 166 return; 167 clear = true; 168 LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) { 169 if (p1->p_reapsubtree == p->p_reapsubtree) { 170 clear = false; 171 break; 172 } 173 } 174 if (clear) 175 proc_id_clear(PROC_ID_REAP, p->p_reapsubtree); 176 } 177 178 void 179 proc_clear_orphan(struct proc *p) 180 { 181 struct proc *p1; 182 183 sx_assert(&proctree_lock, SA_XLOCKED); 184 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 185 return; 186 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 187 p1 = LIST_NEXT(p, p_orphan); 188 if (p1 != NULL) 189 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 190 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 191 } 192 LIST_REMOVE(p, p_orphan); 193 p->p_treeflag &= ~P_TREE_ORPHANED; 194 } 195 196 void 197 exit_onexit(struct proc *p) 198 { 199 MPASS(p->p_numthreads == 1); 200 umtx_thread_exit(FIRST_THREAD_IN_PROC(p)); 201 } 202 203 /* 204 * exit -- death of process. 205 */ 206 int 207 sys_exit(struct thread *td, struct exit_args *uap) 208 { 209 210 exit1(td, uap->rval, 0); 211 __unreachable(); 212 } 213 214 void 215 proc_set_p2_wexit(struct proc *p) 216 { 217 PROC_LOCK_ASSERT(p, MA_OWNED); 218 p->p_flag2 |= P2_WEXIT; 219 } 220 221 /* 222 * Exit: deallocate address space and other resources, change proc state to 223 * zombie, and unlink proc from allproc and parent's lists. Save exit status 224 * and rusage for wait(). Check for child processes and orphan them. 225 */ 226 void 227 exit1(struct thread *td, int rval, int signo) 228 { 229 struct proc *p, *nq, *q, *t; 230 struct thread *tdt; 231 ksiginfo_t *ksi, *ksi1; 232 int signal_parent; 233 234 mtx_assert(&Giant, MA_NOTOWNED); 235 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 236 TSPROCEXIT(td->td_proc->p_pid); 237 238 p = td->td_proc; 239 /* 240 * In case we're rebooting we just let init die in order to 241 * work around an issues where pid 1 might get a fatal signal. 242 * For instance, if network interface serving NFS root is 243 * going down due to reboot, page-in requests for text are 244 * failing. 245 */ 246 if (p == initproc && rebooting == 0) { 247 printf("init died (signal %d, exit %d)\n", signo, rval); 248 panic("Going nowhere without my init!"); 249 } 250 251 /* 252 * Process deferred operations, designated with ASTF_KCLEAR. 253 * For instance, we need to deref SU mp, since the thread does 254 * not return to userspace, and wait for geom to stabilize. 255 */ 256 ast_kclear(td); 257 258 /* 259 * MUST abort all other threads before proceeding past here. 260 */ 261 PROC_LOCK(p); 262 proc_set_p2_wexit(p); 263 264 /* 265 * First check if some other thread or external request got 266 * here before us. If so, act appropriately: exit or suspend. 267 * We must ensure that stop requests are handled before we set 268 * P_WEXIT. 269 */ 270 thread_suspend_check(0); 271 while (p->p_flag & P_HADTHREADS) { 272 /* 273 * Kill off the other threads. This requires 274 * some co-operation from other parts of the kernel 275 * so it may not be instantaneous. With this state set 276 * any thread attempting to interruptibly 277 * sleep will return immediately with EINTR or EWOULDBLOCK 278 * which will hopefully force them to back out to userland 279 * freeing resources as they go. Any thread attempting 280 * to return to userland will thread_exit() from ast(). 281 * thread_exit() will unsuspend us when the last of the 282 * other threads exits. 283 * If there is already a thread singler after resumption, 284 * calling thread_single() will fail; in that case, we just 285 * re-check all suspension request, the thread should 286 * either be suspended there or exit. 287 */ 288 if (!thread_single(p, SINGLE_EXIT)) 289 /* 290 * All other activity in this process is now 291 * stopped. Threading support has been turned 292 * off. 293 */ 294 break; 295 /* 296 * Recheck for new stop or suspend requests which 297 * might appear while process lock was dropped in 298 * thread_single(). 299 */ 300 thread_suspend_check(0); 301 } 302 KASSERT(p->p_numthreads == 1, 303 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 304 racct_sub(p, RACCT_NTHR, 1); 305 306 /* Let event handler change exit status */ 307 p->p_xexit = rval; 308 p->p_xsig = signo; 309 310 /* 311 * Ignore any pending request to stop due to a stop signal. 312 * Once P_WEXIT is set, future requests will be ignored as 313 * well. 314 */ 315 p->p_flag &= ~P_STOPPED_SIG; 316 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 317 318 /* Note that we are exiting. */ 319 p->p_flag |= P_WEXIT; 320 321 /* 322 * Wait for any processes that have a hold on our vmspace to 323 * release their reference. 324 */ 325 while (p->p_lock > 0) 326 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 327 328 PROC_UNLOCK(p); 329 /* Drain the limit callout while we don't have the proc locked */ 330 callout_drain(&p->p_limco); 331 332 #ifdef AUDIT 333 /* 334 * The Sun BSM exit token contains two components: an exit status as 335 * passed to exit(), and a return value to indicate what sort of exit 336 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 337 * what the return value is. 338 */ 339 AUDIT_ARG_EXIT(rval, 0); 340 AUDIT_SYSCALL_EXIT(0, td); 341 #endif 342 343 /* Are we a task leader with peers? */ 344 if (p->p_peers != NULL && p == p->p_leader) { 345 mtx_lock(&ppeers_lock); 346 q = p->p_peers; 347 while (q != NULL) { 348 PROC_LOCK(q); 349 kern_psignal(q, SIGKILL); 350 PROC_UNLOCK(q); 351 q = q->p_peers; 352 } 353 while (p->p_peers != NULL) 354 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 355 mtx_unlock(&ppeers_lock); 356 } 357 358 itimers_exit(p); 359 360 /* 361 * Check if any loadable modules need anything done at process exit. 362 * E.g. SYSV IPC stuff. 363 * Event handler could change exit status. 364 * XXX what if one of these generates an error? 365 */ 366 EVENTHANDLER_DIRECT_INVOKE(process_exit, p); 367 368 /* 369 * If parent is waiting for us to exit or exec, 370 * P_PPWAIT is set; we will wakeup the parent below. 371 */ 372 PROC_LOCK(p); 373 stopprofclock(p); 374 p->p_ptevents = 0; 375 376 /* 377 * Stop the real interval timer. If the handler is currently 378 * executing, prevent it from rearming itself and let it finish. 379 */ 380 if (timevalisset(&p->p_realtimer.it_value) && 381 callout_stop(&p->p_itcallout) == 0) { 382 timevalclear(&p->p_realtimer.it_interval); 383 PROC_UNLOCK(p); 384 callout_drain(&p->p_itcallout); 385 } else { 386 PROC_UNLOCK(p); 387 } 388 389 if (p->p_sysent->sv_onexit != NULL) 390 p->p_sysent->sv_onexit(p); 391 seltdfini(td); 392 393 /* 394 * Reset any sigio structures pointing to us as a result of 395 * F_SETOWN with our pid. The P_WEXIT flag interlocks with fsetown(). 396 */ 397 funsetownlst(&p->p_sigiolst); 398 399 /* 400 * Close open files and release open-file table. 401 * This may block! 402 */ 403 pdescfree(td); 404 fdescfree(td); 405 406 /* 407 * Remove ourself from our leader's peer list and wake our leader. 408 */ 409 if (p->p_leader->p_peers != NULL) { 410 mtx_lock(&ppeers_lock); 411 if (p->p_leader->p_peers != NULL) { 412 q = p->p_leader; 413 while (q->p_peers != p) 414 q = q->p_peers; 415 q->p_peers = p->p_peers; 416 wakeup(p->p_leader); 417 } 418 mtx_unlock(&ppeers_lock); 419 } 420 421 exec_free_abi_mappings(p); 422 vmspace_exit(td); 423 (void)acct_process(td); 424 425 #ifdef KTRACE 426 ktrprocexit(td); 427 #endif 428 /* 429 * Release reference to text vnode etc 430 */ 431 if (p->p_textvp != NULL) { 432 vrele(p->p_textvp); 433 p->p_textvp = NULL; 434 } 435 if (p->p_textdvp != NULL) { 436 vrele(p->p_textdvp); 437 p->p_textdvp = NULL; 438 } 439 if (p->p_binname != NULL) { 440 free(p->p_binname, M_PARGS); 441 p->p_binname = NULL; 442 } 443 444 /* 445 * Release our limits structure. 446 */ 447 lim_free(p->p_limit); 448 p->p_limit = NULL; 449 450 tidhash_remove(td); 451 452 /* 453 * Call machine-dependent code to release any 454 * machine-dependent resources other than the address space. 455 * The address space is released by "vmspace_exitfree(p)" in 456 * vm_waitproc(). 457 */ 458 cpu_exit(td); 459 460 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 461 462 /* 463 * Remove from allproc. It still sits in the hash. 464 */ 465 sx_xlock(&allproc_lock); 466 LIST_REMOVE(p, p_list); 467 468 #ifdef DDB 469 /* 470 * Used by ddb's 'ps' command to find this process via the 471 * pidhash. 472 */ 473 p->p_list.le_prev = NULL; 474 #endif 475 prison_proc_unlink(p->p_ucred->cr_prison, p); 476 sx_xunlock(&allproc_lock); 477 478 sx_xlock(&proctree_lock); 479 if ((p->p_flag & (P_TRACED | P_PPWAIT | P_PPTRACE)) != 0) { 480 PROC_LOCK(p); 481 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 482 PROC_UNLOCK(p); 483 } 484 485 /* 486 * killjobc() might drop and re-acquire proctree_lock to 487 * revoke control tty if exiting process was a session leader. 488 */ 489 killjobc(); 490 491 /* 492 * Reparent all children processes: 493 * - traced ones to the original parent (or init if we are that parent) 494 * - the rest to init 495 */ 496 q = LIST_FIRST(&p->p_children); 497 if (q != NULL) /* only need this if any child is S_ZOMB */ 498 wakeup(q->p_reaper); 499 for (; q != NULL; q = nq) { 500 nq = LIST_NEXT(q, p_sibling); 501 ksi = ksiginfo_alloc(M_WAITOK); 502 PROC_LOCK(q); 503 q->p_sigparent = SIGCHLD; 504 505 if ((q->p_flag & P_TRACED) == 0) { 506 proc_reparent(q, q->p_reaper, true); 507 if (q->p_state == PRS_ZOMBIE) { 508 /* 509 * Inform reaper about the reparented 510 * zombie, since wait(2) has something 511 * new to report. Guarantee queueing 512 * of the SIGCHLD signal, similar to 513 * the _exit() behaviour, by providing 514 * our ksiginfo. Ksi is freed by the 515 * signal delivery. 516 */ 517 if (q->p_ksi == NULL) { 518 ksi1 = NULL; 519 } else { 520 ksiginfo_copy(q->p_ksi, ksi); 521 ksi->ksi_flags |= KSI_INS; 522 ksi1 = ksi; 523 ksi = NULL; 524 } 525 PROC_LOCK(q->p_reaper); 526 pksignal(q->p_reaper, SIGCHLD, ksi1); 527 PROC_UNLOCK(q->p_reaper); 528 } else if (q->p_pdeathsig > 0) { 529 /* 530 * The child asked to received a signal 531 * when we exit. 532 */ 533 kern_psignal(q, q->p_pdeathsig); 534 } 535 } else { 536 /* 537 * Traced processes are killed by default 538 * since their existence means someone is 539 * screwing up. 540 */ 541 t = proc_realparent(q); 542 if (t == p) { 543 proc_reparent(q, q->p_reaper, true); 544 } else { 545 PROC_LOCK(t); 546 proc_reparent(q, t, true); 547 PROC_UNLOCK(t); 548 } 549 /* 550 * Since q was found on our children list, the 551 * proc_reparent() call moved q to the orphan 552 * list due to present P_TRACED flag. Clear 553 * orphan link for q now while q is locked. 554 */ 555 proc_clear_orphan(q); 556 q->p_flag &= ~P_TRACED; 557 q->p_flag2 &= ~P2_PTRACE_FSTP; 558 q->p_ptevents = 0; 559 p->p_xthread = NULL; 560 FOREACH_THREAD_IN_PROC(q, tdt) { 561 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 562 TDB_FSTP); 563 tdt->td_xsig = 0; 564 } 565 if (kern_kill_on_dbg_exit) { 566 q->p_flag &= ~P_STOPPED_TRACE; 567 kern_psignal(q, SIGKILL); 568 } else if ((q->p_flag & (P_STOPPED_TRACE | 569 P_STOPPED_SIG)) != 0) { 570 sigqueue_delete_proc(q, SIGTRAP); 571 ptrace_unsuspend(q); 572 } 573 } 574 PROC_UNLOCK(q); 575 if (ksi != NULL) 576 ksiginfo_free(ksi); 577 } 578 579 /* 580 * Also get rid of our orphans. 581 */ 582 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 583 PROC_LOCK(q); 584 KASSERT(q->p_oppid == p->p_pid, 585 ("orphan %p of %p has unexpected oppid %d", q, p, 586 q->p_oppid)); 587 q->p_oppid = q->p_reaper->p_pid; 588 589 /* 590 * If we are the real parent of this process 591 * but it has been reparented to a debugger, then 592 * check if it asked for a signal when we exit. 593 */ 594 if (q->p_pdeathsig > 0) 595 kern_psignal(q, q->p_pdeathsig); 596 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 597 q->p_pid); 598 proc_clear_orphan(q); 599 PROC_UNLOCK(q); 600 } 601 602 #ifdef KDTRACE_HOOKS 603 if (SDT_PROBES_ENABLED()) { 604 int reason = CLD_EXITED; 605 if (WCOREDUMP(signo)) 606 reason = CLD_DUMPED; 607 else if (WIFSIGNALED(signo)) 608 reason = CLD_KILLED; 609 SDT_PROBE1(proc, , , exit, reason); 610 } 611 #endif 612 613 /* Save exit status. */ 614 PROC_LOCK(p); 615 p->p_xthread = td; 616 617 if (p->p_sysent->sv_ontdexit != NULL) 618 p->p_sysent->sv_ontdexit(td); 619 620 #ifdef KDTRACE_HOOKS 621 /* 622 * Tell the DTrace fasttrap provider about the exit if it 623 * has declared an interest. 624 */ 625 if (dtrace_fasttrap_exit) 626 dtrace_fasttrap_exit(p); 627 #endif 628 629 /* 630 * Notify interested parties of our demise. 631 */ 632 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 633 634 /* 635 * If this is a process with a descriptor, we may not need to deliver 636 * a signal to the parent. proctree_lock is held over 637 * procdesc_exit() to serialize concurrent calls to close() and 638 * exit(). 639 */ 640 signal_parent = 0; 641 if (p->p_procdesc == NULL || procdesc_exit(p)) { 642 /* 643 * Notify parent that we're gone. If parent has the 644 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 645 * notify process 1 instead (and hope it will handle this 646 * situation). 647 */ 648 PROC_LOCK(p->p_pptr); 649 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 650 if (p->p_pptr->p_sigacts->ps_flag & 651 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 652 struct proc *pp; 653 654 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 655 pp = p->p_pptr; 656 PROC_UNLOCK(pp); 657 proc_reparent(p, p->p_reaper, true); 658 p->p_sigparent = SIGCHLD; 659 PROC_LOCK(p->p_pptr); 660 661 /* 662 * Notify parent, so in case he was wait(2)ing or 663 * executing waitpid(2) with our pid, he will 664 * continue. 665 */ 666 wakeup(pp); 667 } else 668 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 669 670 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) { 671 signal_parent = 1; 672 } else if (p->p_sigparent != 0) { 673 if (p->p_sigparent == SIGCHLD) { 674 signal_parent = 1; 675 } else { /* LINUX thread */ 676 signal_parent = 2; 677 } 678 } 679 } else 680 PROC_LOCK(p->p_pptr); 681 sx_xunlock(&proctree_lock); 682 683 if (signal_parent == 1) { 684 childproc_exited(p); 685 } else if (signal_parent == 2) { 686 kern_psignal(p->p_pptr, p->p_sigparent); 687 } 688 689 /* Tell the prison that we are gone. */ 690 prison_proc_free(p->p_ucred->cr_prison); 691 692 /* 693 * The state PRS_ZOMBIE prevents other processes from sending 694 * signal to the process, to avoid memory leak, we free memory 695 * for signal queue at the time when the state is set. 696 */ 697 sigqueue_flush(&p->p_sigqueue); 698 sigqueue_flush(&td->td_sigqueue); 699 700 /* 701 * We have to wait until after acquiring all locks before 702 * changing p_state. We need to avoid all possible context 703 * switches (including ones from blocking on a mutex) while 704 * marked as a zombie. We also have to set the zombie state 705 * before we release the parent process' proc lock to avoid 706 * a lost wakeup. So, we first call wakeup, then we grab the 707 * sched lock, update the state, and release the parent process' 708 * proc lock. 709 */ 710 wakeup(p->p_pptr); 711 cv_broadcast(&p->p_pwait); 712 sched_exit(p->p_pptr, td); 713 PROC_SLOCK(p); 714 p->p_state = PRS_ZOMBIE; 715 PROC_UNLOCK(p->p_pptr); 716 717 /* 718 * Save our children's rusage information in our exit rusage. 719 */ 720 PROC_STATLOCK(p); 721 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 722 PROC_STATUNLOCK(p); 723 724 /* 725 * Make sure the scheduler takes this thread out of its tables etc. 726 * This will also release this thread's reference to the ucred. 727 * Other thread parts to release include pcb bits and such. 728 */ 729 thread_exit(); 730 } 731 732 #ifndef _SYS_SYSPROTO_H_ 733 struct abort2_args { 734 char *why; 735 int nargs; 736 void **args; 737 }; 738 #endif 739 740 int 741 sys_abort2(struct thread *td, struct abort2_args *uap) 742 { 743 void *uargs[16]; 744 void **uargsp; 745 int error, nargs; 746 747 nargs = uap->nargs; 748 if (nargs < 0 || nargs > nitems(uargs)) 749 nargs = -1; 750 uargsp = NULL; 751 if (nargs > 0) { 752 if (uap->args != NULL) { 753 error = copyin(uap->args, uargs, 754 nargs * sizeof(void *)); 755 if (error != 0) 756 nargs = -1; 757 else 758 uargsp = uargs; 759 } else 760 nargs = -1; 761 } 762 return (kern_abort2(td, uap->why, nargs, uargsp)); 763 } 764 765 /* 766 * kern_abort2() 767 * Arguments: 768 * why - user pointer to why 769 * nargs - number of arguments copied or -1 if an error occurred in copying 770 * args - pointer to an array of pointers in kernel format 771 */ 772 int 773 kern_abort2(struct thread *td, const char *why, int nargs, void **uargs) 774 { 775 struct proc *p = td->td_proc; 776 struct sbuf *sb; 777 int error, i, sig; 778 779 /* 780 * Do it right now so we can log either proper call of abort2(), or 781 * note, that invalid argument was passed. 512 is big enough to 782 * handle 16 arguments' descriptions with additional comments. 783 */ 784 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 785 sbuf_clear(sb); 786 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 787 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 788 /* 789 * Since we can't return from abort2(), send SIGKILL in cases, where 790 * abort2() was called improperly 791 */ 792 sig = SIGKILL; 793 /* Prevent from DoSes from user-space. */ 794 if (nargs == -1) 795 goto out; 796 KASSERT(nargs >= 0 && nargs <= 16, ("called with too many args (%d)", 797 nargs)); 798 /* 799 * Limit size of 'reason' string to 128. Will fit even when 800 * maximal number of arguments was chosen to be logged. 801 */ 802 if (why != NULL) { 803 error = sbuf_copyin(sb, why, 128); 804 if (error < 0) 805 goto out; 806 } else { 807 sbuf_cat(sb, "(null)"); 808 } 809 if (nargs > 0) { 810 sbuf_putc(sb, '('); 811 for (i = 0;i < nargs; i++) 812 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 813 sbuf_putc(sb, ')'); 814 } 815 /* 816 * Final stage: arguments were proper, string has been 817 * successfully copied from userspace, and copying pointers 818 * from user-space succeed. 819 */ 820 sig = SIGABRT; 821 out: 822 if (sig == SIGKILL) { 823 sbuf_trim(sb); 824 sbuf_cat(sb, " (Reason text inaccessible)"); 825 } 826 sbuf_cat(sb, "\n"); 827 sbuf_finish(sb); 828 log(LOG_INFO, "%s", sbuf_data(sb)); 829 sbuf_delete(sb); 830 PROC_LOCK(p); 831 sigexit(td, sig); 832 /* NOTREACHED */ 833 } 834 835 #ifdef COMPAT_43 836 /* 837 * The dirty work is handled by kern_wait(). 838 */ 839 int 840 owait(struct thread *td, struct owait_args *uap __unused) 841 { 842 int error, status; 843 844 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 845 if (error == 0) 846 td->td_retval[1] = status; 847 return (error); 848 } 849 #endif /* COMPAT_43 */ 850 851 /* 852 * The dirty work is handled by kern_wait(). 853 */ 854 int 855 sys_wait4(struct thread *td, struct wait4_args *uap) 856 { 857 struct rusage ru, *rup; 858 int error, status; 859 860 if (uap->rusage != NULL) 861 rup = &ru; 862 else 863 rup = NULL; 864 error = kern_wait(td, uap->pid, &status, uap->options, rup); 865 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 866 error = copyout(&status, uap->status, sizeof(status)); 867 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 868 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 869 return (error); 870 } 871 872 int 873 sys_wait6(struct thread *td, struct wait6_args *uap) 874 { 875 struct __wrusage wru, *wrup; 876 siginfo_t si, *sip; 877 idtype_t idtype; 878 id_t id; 879 int error, status; 880 881 idtype = uap->idtype; 882 id = uap->id; 883 884 if (uap->wrusage != NULL) 885 wrup = &wru; 886 else 887 wrup = NULL; 888 889 if (uap->info != NULL) { 890 sip = &si; 891 bzero(sip, sizeof(*sip)); 892 } else 893 sip = NULL; 894 895 /* 896 * We expect all callers of wait6() to know about WEXITED and 897 * WTRAPPED. 898 */ 899 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 900 901 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 902 error = copyout(&status, uap->status, sizeof(status)); 903 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 904 error = copyout(&wru, uap->wrusage, sizeof(wru)); 905 if (uap->info != NULL && error == 0) 906 error = copyout(&si, uap->info, sizeof(si)); 907 return (error); 908 } 909 910 /* 911 * Reap the remains of a zombie process and optionally return status and 912 * rusage. Asserts and will release both the proctree_lock and the process 913 * lock as part of its work. 914 */ 915 void 916 proc_reap(struct thread *td, struct proc *p, int *status, int options) 917 { 918 struct proc *q, *t; 919 920 sx_assert(&proctree_lock, SA_XLOCKED); 921 PROC_LOCK_ASSERT(p, MA_OWNED); 922 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 923 924 mtx_spin_wait_unlocked(&p->p_slock); 925 926 q = td->td_proc; 927 928 if (status) 929 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 930 if (options & WNOWAIT) { 931 /* 932 * Only poll, returning the status. Caller does not wish to 933 * release the proc struct just yet. 934 */ 935 PROC_UNLOCK(p); 936 sx_xunlock(&proctree_lock); 937 return; 938 } 939 940 PROC_LOCK(q); 941 sigqueue_take(p->p_ksi); 942 PROC_UNLOCK(q); 943 944 /* 945 * If we got the child via a ptrace 'attach', we need to give it back 946 * to the old parent. 947 */ 948 if (p->p_oppid != p->p_pptr->p_pid) { 949 PROC_UNLOCK(p); 950 t = proc_realparent(p); 951 PROC_LOCK(t); 952 PROC_LOCK(p); 953 CTR2(KTR_PTRACE, 954 "wait: traced child %d moved back to parent %d", p->p_pid, 955 t->p_pid); 956 proc_reparent(p, t, false); 957 PROC_UNLOCK(p); 958 pksignal(t, SIGCHLD, p->p_ksi); 959 wakeup(t); 960 cv_broadcast(&p->p_pwait); 961 PROC_UNLOCK(t); 962 sx_xunlock(&proctree_lock); 963 return; 964 } 965 PROC_UNLOCK(p); 966 967 /* 968 * Remove other references to this process to ensure we have an 969 * exclusive reference. 970 */ 971 sx_xlock(PIDHASHLOCK(p->p_pid)); 972 LIST_REMOVE(p, p_hash); 973 sx_xunlock(PIDHASHLOCK(p->p_pid)); 974 LIST_REMOVE(p, p_sibling); 975 reaper_abandon_children(p, true); 976 reaper_clear(p); 977 PROC_LOCK(p); 978 proc_clear_orphan(p); 979 PROC_UNLOCK(p); 980 leavepgrp(p); 981 if (p->p_procdesc != NULL) 982 procdesc_reap(p); 983 sx_xunlock(&proctree_lock); 984 985 proc_id_clear(PROC_ID_PID, p->p_pid); 986 987 PROC_LOCK(p); 988 knlist_delete(p->p_klist, td, 1); 989 knlist_detach(p->p_klist); 990 p->p_klist = NULL; 991 PROC_UNLOCK(p); 992 993 /* 994 * Removal from allproc list and process group list paired with 995 * PROC_LOCK which was executed during that time should guarantee 996 * nothing can reach this process anymore. As such further locking 997 * is unnecessary. 998 */ 999 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 1000 1001 PROC_LOCK(q); 1002 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 1003 PROC_UNLOCK(q); 1004 1005 /* 1006 * Decrement the count of procs running with this uid. 1007 */ 1008 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 1009 1010 /* 1011 * Destroy resource accounting information associated with the process. 1012 */ 1013 #ifdef RACCT 1014 if (racct_enable) { 1015 PROC_LOCK(p); 1016 racct_sub(p, RACCT_NPROC, 1); 1017 PROC_UNLOCK(p); 1018 } 1019 #endif 1020 racct_proc_exit(p); 1021 1022 /* 1023 * Free credentials, arguments, and sigacts. 1024 */ 1025 proc_unset_cred(p); 1026 pargs_drop(p->p_args); 1027 p->p_args = NULL; 1028 sigacts_free(p->p_sigacts); 1029 p->p_sigacts = NULL; 1030 1031 /* 1032 * Do any thread-system specific cleanups. 1033 */ 1034 thread_wait(p); 1035 1036 /* 1037 * Give vm and machine-dependent layer a chance to free anything that 1038 * cpu_exit couldn't release while still running in process context. 1039 */ 1040 vm_waitproc(p); 1041 #ifdef MAC 1042 mac_proc_destroy(p); 1043 #endif 1044 1045 KASSERT(FIRST_THREAD_IN_PROC(p), 1046 ("proc_reap: no residual thread!")); 1047 uma_zfree(proc_zone, p); 1048 atomic_add_int(&nprocs, -1); 1049 } 1050 1051 static int 1052 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 1053 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 1054 int check_only) 1055 { 1056 struct rusage *rup; 1057 1058 sx_assert(&proctree_lock, SA_XLOCKED); 1059 1060 PROC_LOCK(p); 1061 1062 switch (idtype) { 1063 case P_ALL: 1064 if (p->p_procdesc == NULL || 1065 (p->p_pptr == td->td_proc && 1066 (p->p_flag & P_TRACED) != 0)) { 1067 break; 1068 } 1069 1070 PROC_UNLOCK(p); 1071 return (0); 1072 case P_PID: 1073 if (p->p_pid != (pid_t)id) { 1074 PROC_UNLOCK(p); 1075 return (0); 1076 } 1077 break; 1078 case P_PGID: 1079 if (p->p_pgid != (pid_t)id) { 1080 PROC_UNLOCK(p); 1081 return (0); 1082 } 1083 break; 1084 case P_SID: 1085 if (p->p_session->s_sid != (pid_t)id) { 1086 PROC_UNLOCK(p); 1087 return (0); 1088 } 1089 break; 1090 case P_UID: 1091 if (p->p_ucred->cr_uid != (uid_t)id) { 1092 PROC_UNLOCK(p); 1093 return (0); 1094 } 1095 break; 1096 case P_GID: 1097 if (p->p_ucred->cr_gid != (gid_t)id) { 1098 PROC_UNLOCK(p); 1099 return (0); 1100 } 1101 break; 1102 case P_JAILID: 1103 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1104 PROC_UNLOCK(p); 1105 return (0); 1106 } 1107 break; 1108 /* 1109 * It seems that the thread structures get zeroed out 1110 * at process exit. This makes it impossible to 1111 * support P_SETID, P_CID or P_CPUID. 1112 */ 1113 default: 1114 PROC_UNLOCK(p); 1115 return (0); 1116 } 1117 1118 if (p_canwait(td, p)) { 1119 PROC_UNLOCK(p); 1120 return (0); 1121 } 1122 1123 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1124 PROC_UNLOCK(p); 1125 return (0); 1126 } 1127 1128 /* 1129 * This special case handles a kthread spawned by linux_clone 1130 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1131 * functions need to be able to distinguish between waiting 1132 * on a process and waiting on a thread. It is a thread if 1133 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1134 * signifies we want to wait for threads and not processes. 1135 */ 1136 if ((p->p_sigparent != SIGCHLD) ^ 1137 ((options & WLINUXCLONE) != 0)) { 1138 PROC_UNLOCK(p); 1139 return (0); 1140 } 1141 1142 if (siginfo != NULL) { 1143 bzero(siginfo, sizeof(*siginfo)); 1144 siginfo->si_errno = 0; 1145 1146 /* 1147 * SUSv4 requires that the si_signo value is always 1148 * SIGCHLD. Obey it despite the rfork(2) interface 1149 * allows to request other signal for child exit 1150 * notification. 1151 */ 1152 siginfo->si_signo = SIGCHLD; 1153 1154 /* 1155 * This is still a rough estimate. We will fix the 1156 * cases TRAPPED, STOPPED, and CONTINUED later. 1157 */ 1158 if (WCOREDUMP(p->p_xsig)) { 1159 siginfo->si_code = CLD_DUMPED; 1160 siginfo->si_status = WTERMSIG(p->p_xsig); 1161 } else if (WIFSIGNALED(p->p_xsig)) { 1162 siginfo->si_code = CLD_KILLED; 1163 siginfo->si_status = WTERMSIG(p->p_xsig); 1164 } else { 1165 siginfo->si_code = CLD_EXITED; 1166 siginfo->si_status = p->p_xexit; 1167 } 1168 1169 siginfo->si_pid = p->p_pid; 1170 siginfo->si_uid = p->p_ucred->cr_uid; 1171 1172 /* 1173 * The si_addr field would be useful additional 1174 * detail, but apparently the PC value may be lost 1175 * when we reach this point. bzero() above sets 1176 * siginfo->si_addr to NULL. 1177 */ 1178 } 1179 1180 /* 1181 * There should be no reason to limit resources usage info to 1182 * exited processes only. A snapshot about any resources used 1183 * by a stopped process may be exactly what is needed. 1184 */ 1185 if (wrusage != NULL) { 1186 rup = &wrusage->wru_self; 1187 *rup = p->p_ru; 1188 PROC_STATLOCK(p); 1189 calcru(p, &rup->ru_utime, &rup->ru_stime); 1190 PROC_STATUNLOCK(p); 1191 1192 rup = &wrusage->wru_children; 1193 *rup = p->p_stats->p_cru; 1194 calccru(p, &rup->ru_utime, &rup->ru_stime); 1195 } 1196 1197 if (p->p_state == PRS_ZOMBIE && !check_only) { 1198 proc_reap(td, p, status, options); 1199 return (-1); 1200 } 1201 return (1); 1202 } 1203 1204 int 1205 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1206 struct rusage *rusage) 1207 { 1208 struct __wrusage wru, *wrup; 1209 idtype_t idtype; 1210 id_t id; 1211 int ret; 1212 1213 /* 1214 * Translate the special pid values into the (idtype, pid) 1215 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1216 * kern_wait6() on its own. 1217 */ 1218 if (pid == WAIT_ANY) { 1219 idtype = P_ALL; 1220 id = 0; 1221 } else if (pid < 0) { 1222 idtype = P_PGID; 1223 id = (id_t)-pid; 1224 } else { 1225 idtype = P_PID; 1226 id = (id_t)pid; 1227 } 1228 1229 if (rusage != NULL) 1230 wrup = &wru; 1231 else 1232 wrup = NULL; 1233 1234 /* 1235 * For backward compatibility we implicitly add flags WEXITED 1236 * and WTRAPPED here. 1237 */ 1238 options |= WEXITED | WTRAPPED; 1239 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1240 if (rusage != NULL) 1241 *rusage = wru.wru_self; 1242 return (ret); 1243 } 1244 1245 static void 1246 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1247 int *status, int options, int si_code) 1248 { 1249 bool cont; 1250 1251 PROC_LOCK_ASSERT(p, MA_OWNED); 1252 sx_assert(&proctree_lock, SA_XLOCKED); 1253 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1254 si_code == CLD_CONTINUED); 1255 1256 cont = si_code == CLD_CONTINUED; 1257 if ((options & WNOWAIT) == 0) { 1258 if (cont) 1259 p->p_flag &= ~P_CONTINUED; 1260 else 1261 p->p_flag |= P_WAITED; 1262 if (kern_wait_dequeue_sigchld && 1263 (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) { 1264 PROC_LOCK(td->td_proc); 1265 sigqueue_take(p->p_ksi); 1266 PROC_UNLOCK(td->td_proc); 1267 } 1268 } 1269 sx_xunlock(&proctree_lock); 1270 if (siginfo != NULL) { 1271 siginfo->si_code = si_code; 1272 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1273 } 1274 if (status != NULL) 1275 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1276 td->td_retval[0] = p->p_pid; 1277 PROC_UNLOCK(p); 1278 } 1279 1280 int 1281 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1282 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1283 { 1284 struct proc *p, *q; 1285 pid_t pid; 1286 int error, nfound, ret; 1287 bool report; 1288 1289 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1290 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1291 AUDIT_ARG_VALUE(options); 1292 1293 q = td->td_proc; 1294 1295 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1296 PROC_LOCK(q); 1297 id = (id_t)q->p_pgid; 1298 PROC_UNLOCK(q); 1299 idtype = P_PGID; 1300 } 1301 1302 /* If we don't know the option, just return. */ 1303 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1304 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1305 return (EINVAL); 1306 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1307 /* 1308 * We will be unable to find any matching processes, 1309 * because there are no known events to look for. 1310 * Prefer to return error instead of blocking 1311 * indefinitely. 1312 */ 1313 return (EINVAL); 1314 } 1315 1316 loop: 1317 if (q->p_flag & P_STATCHILD) { 1318 PROC_LOCK(q); 1319 q->p_flag &= ~P_STATCHILD; 1320 PROC_UNLOCK(q); 1321 } 1322 sx_xlock(&proctree_lock); 1323 loop_locked: 1324 nfound = 0; 1325 LIST_FOREACH(p, &q->p_children, p_sibling) { 1326 pid = p->p_pid; 1327 ret = proc_to_reap(td, p, idtype, id, status, options, 1328 wrusage, siginfo, 0); 1329 if (ret == 0) 1330 continue; 1331 else if (ret != 1) { 1332 td->td_retval[0] = pid; 1333 return (0); 1334 } 1335 1336 nfound++; 1337 PROC_LOCK_ASSERT(p, MA_OWNED); 1338 1339 if ((options & WTRAPPED) != 0 && 1340 (p->p_flag & P_TRACED) != 0) { 1341 PROC_SLOCK(p); 1342 report = 1343 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) && 1344 p->p_suspcount == p->p_numthreads && 1345 (p->p_flag & P_WAITED) == 0); 1346 PROC_SUNLOCK(p); 1347 if (report) { 1348 CTR4(KTR_PTRACE, 1349 "wait: returning trapped pid %d status %#x " 1350 "(xstat %d) xthread %d", 1351 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1352 p->p_xthread != NULL ? 1353 p->p_xthread->td_tid : -1); 1354 report_alive_proc(td, p, siginfo, status, 1355 options, CLD_TRAPPED); 1356 return (0); 1357 } 1358 } 1359 if ((options & WUNTRACED) != 0 && 1360 (p->p_flag & P_STOPPED_SIG) != 0) { 1361 PROC_SLOCK(p); 1362 report = (p->p_suspcount == p->p_numthreads && 1363 ((p->p_flag & P_WAITED) == 0)); 1364 PROC_SUNLOCK(p); 1365 if (report) { 1366 report_alive_proc(td, p, siginfo, status, 1367 options, CLD_STOPPED); 1368 return (0); 1369 } 1370 } 1371 if ((options & WCONTINUED) != 0 && 1372 (p->p_flag & P_CONTINUED) != 0) { 1373 report_alive_proc(td, p, siginfo, status, options, 1374 CLD_CONTINUED); 1375 return (0); 1376 } 1377 PROC_UNLOCK(p); 1378 } 1379 1380 /* 1381 * Look in the orphans list too, to allow the parent to 1382 * collect it's child exit status even if child is being 1383 * debugged. 1384 * 1385 * Debugger detaches from the parent upon successful 1386 * switch-over from parent to child. At this point due to 1387 * re-parenting the parent loses the child to debugger and a 1388 * wait4(2) call would report that it has no children to wait 1389 * for. By maintaining a list of orphans we allow the parent 1390 * to successfully wait until the child becomes a zombie. 1391 */ 1392 if (nfound == 0) { 1393 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1394 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1395 NULL, NULL, 1); 1396 if (ret != 0) { 1397 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1398 (int)td->td_retval[0])); 1399 PROC_UNLOCK(p); 1400 nfound++; 1401 break; 1402 } 1403 } 1404 } 1405 if (nfound == 0) { 1406 sx_xunlock(&proctree_lock); 1407 return (ECHILD); 1408 } 1409 if (options & WNOHANG) { 1410 sx_xunlock(&proctree_lock); 1411 td->td_retval[0] = 0; 1412 return (0); 1413 } 1414 PROC_LOCK(q); 1415 if (q->p_flag & P_STATCHILD) { 1416 q->p_flag &= ~P_STATCHILD; 1417 PROC_UNLOCK(q); 1418 goto loop_locked; 1419 } 1420 sx_xunlock(&proctree_lock); 1421 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0); 1422 if (error) 1423 return (error); 1424 goto loop; 1425 } 1426 1427 void 1428 proc_add_orphan(struct proc *child, struct proc *parent) 1429 { 1430 1431 sx_assert(&proctree_lock, SX_XLOCKED); 1432 KASSERT((child->p_flag & P_TRACED) != 0, 1433 ("proc_add_orphan: not traced")); 1434 1435 if (LIST_EMPTY(&parent->p_orphans)) { 1436 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1437 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan); 1438 } else { 1439 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans), 1440 child, p_orphan); 1441 } 1442 child->p_treeflag |= P_TREE_ORPHANED; 1443 } 1444 1445 /* 1446 * Make process 'parent' the new parent of process 'child'. 1447 * Must be called with an exclusive hold of proctree lock. 1448 */ 1449 void 1450 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid) 1451 { 1452 1453 sx_assert(&proctree_lock, SX_XLOCKED); 1454 PROC_LOCK_ASSERT(child, MA_OWNED); 1455 if (child->p_pptr == parent) 1456 return; 1457 1458 PROC_LOCK(child->p_pptr); 1459 sigqueue_take(child->p_ksi); 1460 PROC_UNLOCK(child->p_pptr); 1461 LIST_REMOVE(child, p_sibling); 1462 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1463 1464 proc_clear_orphan(child); 1465 if ((child->p_flag & P_TRACED) != 0) { 1466 proc_add_orphan(child, child->p_pptr); 1467 } 1468 1469 child->p_pptr = parent; 1470 if (set_oppid) 1471 child->p_oppid = parent->p_pid; 1472 } 1473