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 __FBSDID("$FreeBSD$"); 41 42 #include "opt_ddb.h" 43 #include "opt_ktrace.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/sysproto.h> 48 #include <sys/capsicum.h> 49 #include <sys/eventhandler.h> 50 #include <sys/kernel.h> 51 #include <sys/ktr.h> 52 #include <sys/malloc.h> 53 #include <sys/lock.h> 54 #include <sys/mutex.h> 55 #include <sys/proc.h> 56 #include <sys/procdesc.h> 57 #include <sys/jail.h> 58 #include <sys/tty.h> 59 #include <sys/wait.h> 60 #include <sys/vmmeter.h> 61 #include <sys/vnode.h> 62 #include <sys/racct.h> 63 #include <sys/resourcevar.h> 64 #include <sys/sbuf.h> 65 #include <sys/signalvar.h> 66 #include <sys/sched.h> 67 #include <sys/sx.h> 68 #include <sys/syscallsubr.h> 69 #include <sys/sysctl.h> 70 #include <sys/syslog.h> 71 #include <sys/ptrace.h> 72 #include <sys/acct.h> /* for acct_process() function prototype */ 73 #include <sys/filedesc.h> 74 #include <sys/sdt.h> 75 #include <sys/shm.h> 76 #include <sys/sem.h> 77 #include <sys/sysent.h> 78 #include <sys/timers.h> 79 #include <sys/umtxvar.h> 80 #ifdef KTRACE 81 #include <sys/ktrace.h> 82 #endif 83 84 #include <security/audit/audit.h> 85 #include <security/mac/mac_framework.h> 86 87 #include <vm/vm.h> 88 #include <vm/vm_extern.h> 89 #include <vm/vm_param.h> 90 #include <vm/pmap.h> 91 #include <vm/vm_map.h> 92 #include <vm/vm_page.h> 93 #include <vm/uma.h> 94 95 #ifdef KDTRACE_HOOKS 96 #include <sys/dtrace_bsd.h> 97 dtrace_execexit_func_t dtrace_fasttrap_exit; 98 #endif 99 100 SDT_PROVIDER_DECLARE(proc); 101 SDT_PROBE_DEFINE1(proc, , , exit, "int"); 102 103 static int kern_kill_on_dbg_exit = 1; 104 SYSCTL_INT(_kern, OID_AUTO, kill_on_debugger_exit, CTLFLAG_RWTUN, 105 &kern_kill_on_dbg_exit, 0, 106 "Kill ptraced processes when debugger exits"); 107 108 static bool kern_wait_dequeue_sigchld = 1; 109 SYSCTL_BOOL(_kern, OID_AUTO, wait_dequeue_sigchld, CTLFLAG_RWTUN, 110 &kern_wait_dequeue_sigchld, 0, 111 "Dequeue SIGCHLD on wait(2) for live process"); 112 113 struct proc * 114 proc_realparent(struct proc *child) 115 { 116 struct proc *p, *parent; 117 118 sx_assert(&proctree_lock, SX_LOCKED); 119 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) 120 return (child->p_pptr->p_pid == child->p_oppid ? 121 child->p_pptr : child->p_reaper); 122 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 123 /* Cannot use LIST_PREV(), since the list head is not known. */ 124 p = __containerof(p->p_orphan.le_prev, struct proc, 125 p_orphan.le_next); 126 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 127 ("missing P_ORPHAN %p", p)); 128 } 129 parent = __containerof(p->p_orphan.le_prev, struct proc, 130 p_orphans.lh_first); 131 return (parent); 132 } 133 134 void 135 reaper_abandon_children(struct proc *p, bool exiting) 136 { 137 struct proc *p1, *p2, *ptmp; 138 139 sx_assert(&proctree_lock, SX_XLOCKED); 140 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 141 if ((p->p_treeflag & P_TREE_REAPER) == 0) 142 return; 143 p1 = p->p_reaper; 144 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 145 LIST_REMOVE(p2, p_reapsibling); 146 p2->p_reaper = p1; 147 p2->p_reapsubtree = p->p_reapsubtree; 148 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 149 if (exiting && p2->p_pptr == p) { 150 PROC_LOCK(p2); 151 proc_reparent(p2, p1, true); 152 PROC_UNLOCK(p2); 153 } 154 } 155 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 156 p->p_treeflag &= ~P_TREE_REAPER; 157 } 158 159 static void 160 reaper_clear(struct proc *p) 161 { 162 struct proc *p1; 163 bool clear; 164 165 sx_assert(&proctree_lock, SX_LOCKED); 166 LIST_REMOVE(p, p_reapsibling); 167 if (p->p_reapsubtree == 1) 168 return; 169 clear = true; 170 LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) { 171 if (p1->p_reapsubtree == p->p_reapsubtree) { 172 clear = false; 173 break; 174 } 175 } 176 if (clear) 177 proc_id_clear(PROC_ID_REAP, p->p_reapsubtree); 178 } 179 180 void 181 proc_clear_orphan(struct proc *p) 182 { 183 struct proc *p1; 184 185 sx_assert(&proctree_lock, SA_XLOCKED); 186 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 187 return; 188 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 189 p1 = LIST_NEXT(p, p_orphan); 190 if (p1 != NULL) 191 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 192 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 193 } 194 LIST_REMOVE(p, p_orphan); 195 p->p_treeflag &= ~P_TREE_ORPHANED; 196 } 197 198 void 199 exit_onexit(struct proc *p) 200 { 201 MPASS(p->p_numthreads == 1); 202 umtx_thread_exit(FIRST_THREAD_IN_PROC(p)); 203 } 204 205 /* 206 * exit -- death of process. 207 */ 208 int 209 sys_exit(struct thread *td, struct exit_args *uap) 210 { 211 212 exit1(td, uap->rval, 0); 213 __unreachable(); 214 } 215 216 void 217 proc_set_p2_wexit(struct proc *p) 218 { 219 PROC_LOCK_ASSERT(p, MA_OWNED); 220 p->p_flag2 |= P2_WEXIT; 221 } 222 223 /* 224 * Exit: deallocate address space and other resources, change proc state to 225 * zombie, and unlink proc from allproc and parent's lists. Save exit status 226 * and rusage for wait(). Check for child processes and orphan them. 227 */ 228 void 229 exit1(struct thread *td, int rval, int signo) 230 { 231 struct proc *p, *nq, *q, *t; 232 struct thread *tdt; 233 ksiginfo_t *ksi, *ksi1; 234 int signal_parent; 235 236 mtx_assert(&Giant, MA_NOTOWNED); 237 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 238 TSPROCEXIT(td->td_proc->p_pid); 239 240 p = td->td_proc; 241 /* 242 * XXX in case we're rebooting we just let init die in order to 243 * work around an unsolved stack overflow seen very late during 244 * shutdown on sparc64 when the gmirror worker process exists. 245 * XXX what to do now that sparc64 is gone... remove if? 246 */ 247 if (p == initproc && rebooting == 0) { 248 printf("init died (signal %d, exit %d)\n", signo, rval); 249 panic("Going nowhere without my init!"); 250 } 251 252 /* 253 * Process deferred operations, designated with ASTF_KCLEAR. 254 * For instance, we need to deref SU mp, since the thread does 255 * not return to userspace, and wait for geom to stabilize. 256 */ 257 ast_kclear(td); 258 259 /* 260 * MUST abort all other threads before proceeding past here. 261 */ 262 PROC_LOCK(p); 263 proc_set_p2_wexit(p); 264 265 /* 266 * First check if some other thread or external request got 267 * here before us. If so, act appropriately: exit or suspend. 268 * We must ensure that stop requests are handled before we set 269 * P_WEXIT. 270 */ 271 thread_suspend_check(0); 272 while (p->p_flag & P_HADTHREADS) { 273 /* 274 * Kill off the other threads. This requires 275 * some co-operation from other parts of the kernel 276 * so it may not be instantaneous. With this state set 277 * any thread attempting to interruptibly 278 * sleep will return immediately with EINTR or EWOULDBLOCK 279 * which will hopefully force them to back out to userland 280 * freeing resources as they go. Any thread attempting 281 * to return to userland will thread_exit() from ast(). 282 * thread_exit() will unsuspend us when the last of the 283 * other threads exits. 284 * If there is already a thread singler after resumption, 285 * calling thread_single() will fail; in that case, we just 286 * re-check all suspension request, the thread should 287 * either be suspended there or exit. 288 */ 289 if (!thread_single(p, SINGLE_EXIT)) 290 /* 291 * All other activity in this process is now 292 * stopped. Threading support has been turned 293 * off. 294 */ 295 break; 296 /* 297 * Recheck for new stop or suspend requests which 298 * might appear while process lock was dropped in 299 * thread_single(). 300 */ 301 thread_suspend_check(0); 302 } 303 KASSERT(p->p_numthreads == 1, 304 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 305 racct_sub(p, RACCT_NTHR, 1); 306 307 /* Let event handler change exit status */ 308 p->p_xexit = rval; 309 p->p_xsig = signo; 310 311 /* 312 * Ignore any pending request to stop due to a stop signal. 313 * Once P_WEXIT is set, future requests will be ignored as 314 * well. 315 */ 316 p->p_flag &= ~P_STOPPED_SIG; 317 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 318 319 /* Note that we are exiting. */ 320 p->p_flag |= P_WEXIT; 321 322 /* 323 * Wait for any processes that have a hold on our vmspace to 324 * release their reference. 325 */ 326 while (p->p_lock > 0) 327 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 328 329 PROC_UNLOCK(p); 330 /* Drain the limit callout while we don't have the proc locked */ 331 callout_drain(&p->p_limco); 332 333 #ifdef AUDIT 334 /* 335 * The Sun BSM exit token contains two components: an exit status as 336 * passed to exit(), and a return value to indicate what sort of exit 337 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 338 * what the return value is. 339 */ 340 AUDIT_ARG_EXIT(rval, 0); 341 AUDIT_SYSCALL_EXIT(0, td); 342 #endif 343 344 /* Are we a task leader with peers? */ 345 if (p->p_peers != NULL && p == p->p_leader) { 346 mtx_lock(&ppeers_lock); 347 q = p->p_peers; 348 while (q != NULL) { 349 PROC_LOCK(q); 350 kern_psignal(q, SIGKILL); 351 PROC_UNLOCK(q); 352 q = q->p_peers; 353 } 354 while (p->p_peers != NULL) 355 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 356 mtx_unlock(&ppeers_lock); 357 } 358 359 itimers_exit(p); 360 361 /* 362 * Check if any loadable modules need anything done at process exit. 363 * E.g. SYSV IPC stuff. 364 * Event handler could change exit status. 365 * XXX what if one of these generates an error? 366 */ 367 EVENTHANDLER_DIRECT_INVOKE(process_exit, p); 368 369 /* 370 * If parent is waiting for us to exit or exec, 371 * P_PPWAIT is set; we will wakeup the parent below. 372 */ 373 PROC_LOCK(p); 374 stopprofclock(p); 375 p->p_ptevents = 0; 376 377 /* 378 * Stop the real interval timer. If the handler is currently 379 * executing, prevent it from rearming itself and let it finish. 380 */ 381 if (timevalisset(&p->p_realtimer.it_value) && 382 callout_stop(&p->p_itcallout) == 0) { 383 timevalclear(&p->p_realtimer.it_interval); 384 PROC_UNLOCK(p); 385 callout_drain(&p->p_itcallout); 386 } else { 387 PROC_UNLOCK(p); 388 } 389 390 if (p->p_sysent->sv_onexit != NULL) 391 p->p_sysent->sv_onexit(p); 392 seltdfini(td); 393 394 /* 395 * Reset any sigio structures pointing to us as a result of 396 * F_SETOWN with our pid. The P_WEXIT flag interlocks with fsetown(). 397 */ 398 funsetownlst(&p->p_sigiolst); 399 400 /* 401 * Close open files and release open-file table. 402 * This may block! 403 */ 404 pdescfree(td); 405 fdescfree(td); 406 407 /* 408 * Remove ourself from our leader's peer list and wake our leader. 409 */ 410 if (p->p_leader->p_peers != NULL) { 411 mtx_lock(&ppeers_lock); 412 if (p->p_leader->p_peers != NULL) { 413 q = p->p_leader; 414 while (q->p_peers != p) 415 q = q->p_peers; 416 q->p_peers = p->p_peers; 417 wakeup(p->p_leader); 418 } 419 mtx_unlock(&ppeers_lock); 420 } 421 422 exec_free_abi_mappings(p); 423 vmspace_exit(td); 424 (void)acct_process(td); 425 426 #ifdef KTRACE 427 ktrprocexit(td); 428 #endif 429 /* 430 * Release reference to text vnode etc 431 */ 432 if (p->p_textvp != NULL) { 433 vrele(p->p_textvp); 434 p->p_textvp = NULL; 435 } 436 if (p->p_textdvp != NULL) { 437 vrele(p->p_textdvp); 438 p->p_textdvp = NULL; 439 } 440 if (p->p_binname != NULL) { 441 free(p->p_binname, M_PARGS); 442 p->p_binname = NULL; 443 } 444 445 /* 446 * Release our limits structure. 447 */ 448 lim_free(p->p_limit); 449 p->p_limit = NULL; 450 451 tidhash_remove(td); 452 453 /* 454 * Call machine-dependent code to release any 455 * machine-dependent resources other than the address space. 456 * The address space is released by "vmspace_exitfree(p)" in 457 * vm_waitproc(). 458 */ 459 cpu_exit(td); 460 461 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 462 463 /* 464 * Remove from allproc. It still sits in the hash. 465 */ 466 sx_xlock(&allproc_lock); 467 LIST_REMOVE(p, p_list); 468 469 #ifdef DDB 470 /* 471 * Used by ddb's 'ps' command to find this process via the 472 * pidhash. 473 */ 474 p->p_list.le_prev = NULL; 475 #endif 476 sx_xunlock(&allproc_lock); 477 478 sx_xlock(&proctree_lock); 479 PROC_LOCK(p); 480 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 481 PROC_UNLOCK(p); 482 483 /* 484 * killjobc() might drop and re-acquire proctree_lock to 485 * revoke control tty if exiting process was a session leader. 486 */ 487 killjobc(); 488 489 /* 490 * Reparent all children processes: 491 * - traced ones to the original parent (or init if we are that parent) 492 * - the rest to init 493 */ 494 q = LIST_FIRST(&p->p_children); 495 if (q != NULL) /* only need this if any child is S_ZOMB */ 496 wakeup(q->p_reaper); 497 for (; q != NULL; q = nq) { 498 nq = LIST_NEXT(q, p_sibling); 499 ksi = ksiginfo_alloc(M_WAITOK); 500 PROC_LOCK(q); 501 q->p_sigparent = SIGCHLD; 502 503 if ((q->p_flag & P_TRACED) == 0) { 504 proc_reparent(q, q->p_reaper, true); 505 if (q->p_state == PRS_ZOMBIE) { 506 /* 507 * Inform reaper about the reparented 508 * zombie, since wait(2) has something 509 * new to report. Guarantee queueing 510 * of the SIGCHLD signal, similar to 511 * the _exit() behaviour, by providing 512 * our ksiginfo. Ksi is freed by the 513 * signal delivery. 514 */ 515 if (q->p_ksi == NULL) { 516 ksi1 = NULL; 517 } else { 518 ksiginfo_copy(q->p_ksi, ksi); 519 ksi->ksi_flags |= KSI_INS; 520 ksi1 = ksi; 521 ksi = NULL; 522 } 523 PROC_LOCK(q->p_reaper); 524 pksignal(q->p_reaper, SIGCHLD, ksi1); 525 PROC_UNLOCK(q->p_reaper); 526 } else if (q->p_pdeathsig > 0) { 527 /* 528 * The child asked to received a signal 529 * when we exit. 530 */ 531 kern_psignal(q, q->p_pdeathsig); 532 } 533 } else { 534 /* 535 * Traced processes are killed by default 536 * since their existence means someone is 537 * screwing up. 538 */ 539 t = proc_realparent(q); 540 if (t == p) { 541 proc_reparent(q, q->p_reaper, true); 542 } else { 543 PROC_LOCK(t); 544 proc_reparent(q, t, true); 545 PROC_UNLOCK(t); 546 } 547 /* 548 * Since q was found on our children list, the 549 * proc_reparent() call moved q to the orphan 550 * list due to present P_TRACED flag. Clear 551 * orphan link for q now while q is locked. 552 */ 553 proc_clear_orphan(q); 554 q->p_flag &= ~P_TRACED; 555 q->p_flag2 &= ~P2_PTRACE_FSTP; 556 q->p_ptevents = 0; 557 p->p_xthread = NULL; 558 FOREACH_THREAD_IN_PROC(q, tdt) { 559 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 560 TDB_FSTP); 561 tdt->td_xsig = 0; 562 } 563 if (kern_kill_on_dbg_exit) { 564 q->p_flag &= ~P_STOPPED_TRACE; 565 kern_psignal(q, SIGKILL); 566 } else if ((q->p_flag & (P_STOPPED_TRACE | 567 P_STOPPED_SIG)) != 0) { 568 sigqueue_delete_proc(q, SIGTRAP); 569 ptrace_unsuspend(q); 570 } 571 } 572 PROC_UNLOCK(q); 573 if (ksi != NULL) 574 ksiginfo_free(ksi); 575 } 576 577 /* 578 * Also get rid of our orphans. 579 */ 580 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 581 PROC_LOCK(q); 582 KASSERT(q->p_oppid == p->p_pid, 583 ("orphan %p of %p has unexpected oppid %d", q, p, 584 q->p_oppid)); 585 q->p_oppid = q->p_reaper->p_pid; 586 587 /* 588 * If we are the real parent of this process 589 * but it has been reparented to a debugger, then 590 * check if it asked for a signal when we exit. 591 */ 592 if (q->p_pdeathsig > 0) 593 kern_psignal(q, q->p_pdeathsig); 594 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 595 q->p_pid); 596 proc_clear_orphan(q); 597 PROC_UNLOCK(q); 598 } 599 600 #ifdef KDTRACE_HOOKS 601 if (SDT_PROBES_ENABLED()) { 602 int reason = CLD_EXITED; 603 if (WCOREDUMP(signo)) 604 reason = CLD_DUMPED; 605 else if (WIFSIGNALED(signo)) 606 reason = CLD_KILLED; 607 SDT_PROBE1(proc, , , exit, reason); 608 } 609 #endif 610 611 /* Save exit status. */ 612 PROC_LOCK(p); 613 p->p_xthread = td; 614 615 if (p->p_sysent->sv_ontdexit != NULL) 616 p->p_sysent->sv_ontdexit(td); 617 618 #ifdef KDTRACE_HOOKS 619 /* 620 * Tell the DTrace fasttrap provider about the exit if it 621 * has declared an interest. 622 */ 623 if (dtrace_fasttrap_exit) 624 dtrace_fasttrap_exit(p); 625 #endif 626 627 /* 628 * Notify interested parties of our demise. 629 */ 630 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 631 632 /* 633 * If this is a process with a descriptor, we may not need to deliver 634 * a signal to the parent. proctree_lock is held over 635 * procdesc_exit() to serialize concurrent calls to close() and 636 * exit(). 637 */ 638 signal_parent = 0; 639 if (p->p_procdesc == NULL || procdesc_exit(p)) { 640 /* 641 * Notify parent that we're gone. If parent has the 642 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 643 * notify process 1 instead (and hope it will handle this 644 * situation). 645 */ 646 PROC_LOCK(p->p_pptr); 647 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 648 if (p->p_pptr->p_sigacts->ps_flag & 649 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 650 struct proc *pp; 651 652 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 653 pp = p->p_pptr; 654 PROC_UNLOCK(pp); 655 proc_reparent(p, p->p_reaper, true); 656 p->p_sigparent = SIGCHLD; 657 PROC_LOCK(p->p_pptr); 658 659 /* 660 * Notify parent, so in case he was wait(2)ing or 661 * executing waitpid(2) with our pid, he will 662 * continue. 663 */ 664 wakeup(pp); 665 } else 666 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 667 668 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) { 669 signal_parent = 1; 670 } else if (p->p_sigparent != 0) { 671 if (p->p_sigparent == SIGCHLD) { 672 signal_parent = 1; 673 } else { /* LINUX thread */ 674 signal_parent = 2; 675 } 676 } 677 } else 678 PROC_LOCK(p->p_pptr); 679 sx_xunlock(&proctree_lock); 680 681 if (signal_parent == 1) { 682 childproc_exited(p); 683 } else if (signal_parent == 2) { 684 kern_psignal(p->p_pptr, p->p_sigparent); 685 } 686 687 /* Tell the prison that we are gone. */ 688 prison_proc_free(p->p_ucred->cr_prison); 689 690 /* 691 * The state PRS_ZOMBIE prevents other proesses from sending 692 * signal to the process, to avoid memory leak, we free memory 693 * for signal queue at the time when the state is set. 694 */ 695 sigqueue_flush(&p->p_sigqueue); 696 sigqueue_flush(&td->td_sigqueue); 697 698 /* 699 * We have to wait until after acquiring all locks before 700 * changing p_state. We need to avoid all possible context 701 * switches (including ones from blocking on a mutex) while 702 * marked as a zombie. We also have to set the zombie state 703 * before we release the parent process' proc lock to avoid 704 * a lost wakeup. So, we first call wakeup, then we grab the 705 * sched lock, update the state, and release the parent process' 706 * proc lock. 707 */ 708 wakeup(p->p_pptr); 709 cv_broadcast(&p->p_pwait); 710 sched_exit(p->p_pptr, td); 711 PROC_SLOCK(p); 712 p->p_state = PRS_ZOMBIE; 713 PROC_UNLOCK(p->p_pptr); 714 715 /* 716 * Save our children's rusage information in our exit rusage. 717 */ 718 PROC_STATLOCK(p); 719 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 720 PROC_STATUNLOCK(p); 721 722 /* 723 * Make sure the scheduler takes this thread out of its tables etc. 724 * This will also release this thread's reference to the ucred. 725 * Other thread parts to release include pcb bits and such. 726 */ 727 thread_exit(); 728 } 729 730 #ifndef _SYS_SYSPROTO_H_ 731 struct abort2_args { 732 char *why; 733 int nargs; 734 void **args; 735 }; 736 #endif 737 738 int 739 sys_abort2(struct thread *td, struct abort2_args *uap) 740 { 741 void *uargs[16]; 742 void **uargsp; 743 int error, nargs; 744 745 nargs = uap->nargs; 746 if (nargs < 0 || nargs > nitems(uargs)) 747 nargs = -1; 748 uargsp = NULL; 749 if (nargs > 0) { 750 if (uap->args != NULL) { 751 error = copyin(uap->args, uargs, 752 nargs * sizeof(void *)); 753 if (error != 0) 754 nargs = -1; 755 else 756 uargsp = uargs; 757 } else 758 nargs = -1; 759 } 760 return (kern_abort2(td, uap->why, nargs, uargsp)); 761 } 762 763 /* 764 * kern_abort2() 765 * Arguments: 766 * why - user pointer to why 767 * nargs - number of arguments copied or -1 if an error occured in copying 768 * args - pointer to an array of pointers in kernel format 769 */ 770 int 771 kern_abort2(struct thread *td, const char *why, int nargs, void **uargs) 772 { 773 struct proc *p = td->td_proc; 774 struct sbuf *sb; 775 int error, i, sig; 776 777 /* 778 * Do it right now so we can log either proper call of abort2(), or 779 * note, that invalid argument was passed. 512 is big enough to 780 * handle 16 arguments' descriptions with additional comments. 781 */ 782 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 783 sbuf_clear(sb); 784 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 785 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 786 /* 787 * Since we can't return from abort2(), send SIGKILL in cases, where 788 * abort2() was called improperly 789 */ 790 sig = SIGKILL; 791 /* Prevent from DoSes from user-space. */ 792 if (nargs == -1) 793 goto out; 794 KASSERT(nargs >= 0 && nargs <= 16, ("called with too many args (%d)", 795 nargs)); 796 /* 797 * Limit size of 'reason' string to 128. Will fit even when 798 * maximal number of arguments was chosen to be logged. 799 */ 800 if (why != NULL) { 801 error = sbuf_copyin(sb, why, 128); 802 if (error < 0) 803 goto out; 804 } else { 805 sbuf_printf(sb, "(null)"); 806 } 807 if (nargs > 0) { 808 sbuf_printf(sb, "("); 809 for (i = 0;i < nargs; i++) 810 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 811 sbuf_printf(sb, ")"); 812 } 813 /* 814 * Final stage: arguments were proper, string has been 815 * successfully copied from userspace, and copying pointers 816 * from user-space succeed. 817 */ 818 sig = SIGABRT; 819 out: 820 if (sig == SIGKILL) { 821 sbuf_trim(sb); 822 sbuf_printf(sb, " (Reason text inaccessible)"); 823 } 824 sbuf_cat(sb, "\n"); 825 sbuf_finish(sb); 826 log(LOG_INFO, "%s", sbuf_data(sb)); 827 sbuf_delete(sb); 828 exit1(td, 0, sig); 829 return (0); 830 } 831 832 #ifdef COMPAT_43 833 /* 834 * The dirty work is handled by kern_wait(). 835 */ 836 int 837 owait(struct thread *td, struct owait_args *uap __unused) 838 { 839 int error, status; 840 841 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 842 if (error == 0) 843 td->td_retval[1] = status; 844 return (error); 845 } 846 #endif /* COMPAT_43 */ 847 848 /* 849 * The dirty work is handled by kern_wait(). 850 */ 851 int 852 sys_wait4(struct thread *td, struct wait4_args *uap) 853 { 854 struct rusage ru, *rup; 855 int error, status; 856 857 if (uap->rusage != NULL) 858 rup = &ru; 859 else 860 rup = NULL; 861 error = kern_wait(td, uap->pid, &status, uap->options, rup); 862 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 863 error = copyout(&status, uap->status, sizeof(status)); 864 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 865 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 866 return (error); 867 } 868 869 int 870 sys_wait6(struct thread *td, struct wait6_args *uap) 871 { 872 struct __wrusage wru, *wrup; 873 siginfo_t si, *sip; 874 idtype_t idtype; 875 id_t id; 876 int error, status; 877 878 idtype = uap->idtype; 879 id = uap->id; 880 881 if (uap->wrusage != NULL) 882 wrup = &wru; 883 else 884 wrup = NULL; 885 886 if (uap->info != NULL) { 887 sip = &si; 888 bzero(sip, sizeof(*sip)); 889 } else 890 sip = NULL; 891 892 /* 893 * We expect all callers of wait6() to know about WEXITED and 894 * WTRAPPED. 895 */ 896 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 897 898 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 899 error = copyout(&status, uap->status, sizeof(status)); 900 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 901 error = copyout(&wru, uap->wrusage, sizeof(wru)); 902 if (uap->info != NULL && error == 0) 903 error = copyout(&si, uap->info, sizeof(si)); 904 return (error); 905 } 906 907 /* 908 * Reap the remains of a zombie process and optionally return status and 909 * rusage. Asserts and will release both the proctree_lock and the process 910 * lock as part of its work. 911 */ 912 void 913 proc_reap(struct thread *td, struct proc *p, int *status, int options) 914 { 915 struct proc *q, *t; 916 917 sx_assert(&proctree_lock, SA_XLOCKED); 918 PROC_LOCK_ASSERT(p, MA_OWNED); 919 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 920 921 mtx_spin_wait_unlocked(&p->p_slock); 922 923 q = td->td_proc; 924 925 if (status) 926 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 927 if (options & WNOWAIT) { 928 /* 929 * Only poll, returning the status. Caller does not wish to 930 * release the proc struct just yet. 931 */ 932 PROC_UNLOCK(p); 933 sx_xunlock(&proctree_lock); 934 return; 935 } 936 937 PROC_LOCK(q); 938 sigqueue_take(p->p_ksi); 939 PROC_UNLOCK(q); 940 941 /* 942 * If we got the child via a ptrace 'attach', we need to give it back 943 * to the old parent. 944 */ 945 if (p->p_oppid != p->p_pptr->p_pid) { 946 PROC_UNLOCK(p); 947 t = proc_realparent(p); 948 PROC_LOCK(t); 949 PROC_LOCK(p); 950 CTR2(KTR_PTRACE, 951 "wait: traced child %d moved back to parent %d", p->p_pid, 952 t->p_pid); 953 proc_reparent(p, t, false); 954 PROC_UNLOCK(p); 955 pksignal(t, SIGCHLD, p->p_ksi); 956 wakeup(t); 957 cv_broadcast(&p->p_pwait); 958 PROC_UNLOCK(t); 959 sx_xunlock(&proctree_lock); 960 return; 961 } 962 PROC_UNLOCK(p); 963 964 /* 965 * Remove other references to this process to ensure we have an 966 * exclusive reference. 967 */ 968 sx_xlock(PIDHASHLOCK(p->p_pid)); 969 LIST_REMOVE(p, p_hash); 970 sx_xunlock(PIDHASHLOCK(p->p_pid)); 971 LIST_REMOVE(p, p_sibling); 972 reaper_abandon_children(p, true); 973 reaper_clear(p); 974 PROC_LOCK(p); 975 proc_clear_orphan(p); 976 PROC_UNLOCK(p); 977 leavepgrp(p); 978 if (p->p_procdesc != NULL) 979 procdesc_reap(p); 980 sx_xunlock(&proctree_lock); 981 982 proc_id_clear(PROC_ID_PID, p->p_pid); 983 984 PROC_LOCK(p); 985 knlist_detach(p->p_klist); 986 p->p_klist = NULL; 987 PROC_UNLOCK(p); 988 989 /* 990 * Removal from allproc list and process group list paired with 991 * PROC_LOCK which was executed during that time should guarantee 992 * nothing can reach this process anymore. As such further locking 993 * is unnecessary. 994 */ 995 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 996 997 PROC_LOCK(q); 998 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 999 PROC_UNLOCK(q); 1000 1001 /* 1002 * Decrement the count of procs running with this uid. 1003 */ 1004 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 1005 1006 /* 1007 * Destroy resource accounting information associated with the process. 1008 */ 1009 #ifdef RACCT 1010 if (racct_enable) { 1011 PROC_LOCK(p); 1012 racct_sub(p, RACCT_NPROC, 1); 1013 PROC_UNLOCK(p); 1014 } 1015 #endif 1016 racct_proc_exit(p); 1017 1018 /* 1019 * Free credentials, arguments, and sigacts. 1020 */ 1021 proc_unset_cred(p); 1022 pargs_drop(p->p_args); 1023 p->p_args = NULL; 1024 sigacts_free(p->p_sigacts); 1025 p->p_sigacts = NULL; 1026 1027 /* 1028 * Do any thread-system specific cleanups. 1029 */ 1030 thread_wait(p); 1031 1032 /* 1033 * Give vm and machine-dependent layer a chance to free anything that 1034 * cpu_exit couldn't release while still running in process context. 1035 */ 1036 vm_waitproc(p); 1037 #ifdef MAC 1038 mac_proc_destroy(p); 1039 #endif 1040 1041 KASSERT(FIRST_THREAD_IN_PROC(p), 1042 ("proc_reap: no residual thread!")); 1043 uma_zfree(proc_zone, p); 1044 atomic_add_int(&nprocs, -1); 1045 } 1046 1047 static int 1048 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 1049 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 1050 int check_only) 1051 { 1052 struct rusage *rup; 1053 1054 sx_assert(&proctree_lock, SA_XLOCKED); 1055 1056 PROC_LOCK(p); 1057 1058 switch (idtype) { 1059 case P_ALL: 1060 if (p->p_procdesc == NULL || 1061 (p->p_pptr == td->td_proc && 1062 (p->p_flag & P_TRACED) != 0)) { 1063 break; 1064 } 1065 1066 PROC_UNLOCK(p); 1067 return (0); 1068 case P_PID: 1069 if (p->p_pid != (pid_t)id) { 1070 PROC_UNLOCK(p); 1071 return (0); 1072 } 1073 break; 1074 case P_PGID: 1075 if (p->p_pgid != (pid_t)id) { 1076 PROC_UNLOCK(p); 1077 return (0); 1078 } 1079 break; 1080 case P_SID: 1081 if (p->p_session->s_sid != (pid_t)id) { 1082 PROC_UNLOCK(p); 1083 return (0); 1084 } 1085 break; 1086 case P_UID: 1087 if (p->p_ucred->cr_uid != (uid_t)id) { 1088 PROC_UNLOCK(p); 1089 return (0); 1090 } 1091 break; 1092 case P_GID: 1093 if (p->p_ucred->cr_gid != (gid_t)id) { 1094 PROC_UNLOCK(p); 1095 return (0); 1096 } 1097 break; 1098 case P_JAILID: 1099 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1100 PROC_UNLOCK(p); 1101 return (0); 1102 } 1103 break; 1104 /* 1105 * It seems that the thread structures get zeroed out 1106 * at process exit. This makes it impossible to 1107 * support P_SETID, P_CID or P_CPUID. 1108 */ 1109 default: 1110 PROC_UNLOCK(p); 1111 return (0); 1112 } 1113 1114 if (p_canwait(td, p)) { 1115 PROC_UNLOCK(p); 1116 return (0); 1117 } 1118 1119 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1120 PROC_UNLOCK(p); 1121 return (0); 1122 } 1123 1124 /* 1125 * This special case handles a kthread spawned by linux_clone 1126 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1127 * functions need to be able to distinguish between waiting 1128 * on a process and waiting on a thread. It is a thread if 1129 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1130 * signifies we want to wait for threads and not processes. 1131 */ 1132 if ((p->p_sigparent != SIGCHLD) ^ 1133 ((options & WLINUXCLONE) != 0)) { 1134 PROC_UNLOCK(p); 1135 return (0); 1136 } 1137 1138 if (siginfo != NULL) { 1139 bzero(siginfo, sizeof(*siginfo)); 1140 siginfo->si_errno = 0; 1141 1142 /* 1143 * SUSv4 requires that the si_signo value is always 1144 * SIGCHLD. Obey it despite the rfork(2) interface 1145 * allows to request other signal for child exit 1146 * notification. 1147 */ 1148 siginfo->si_signo = SIGCHLD; 1149 1150 /* 1151 * This is still a rough estimate. We will fix the 1152 * cases TRAPPED, STOPPED, and CONTINUED later. 1153 */ 1154 if (WCOREDUMP(p->p_xsig)) { 1155 siginfo->si_code = CLD_DUMPED; 1156 siginfo->si_status = WTERMSIG(p->p_xsig); 1157 } else if (WIFSIGNALED(p->p_xsig)) { 1158 siginfo->si_code = CLD_KILLED; 1159 siginfo->si_status = WTERMSIG(p->p_xsig); 1160 } else { 1161 siginfo->si_code = CLD_EXITED; 1162 siginfo->si_status = p->p_xexit; 1163 } 1164 1165 siginfo->si_pid = p->p_pid; 1166 siginfo->si_uid = p->p_ucred->cr_uid; 1167 1168 /* 1169 * The si_addr field would be useful additional 1170 * detail, but apparently the PC value may be lost 1171 * when we reach this point. bzero() above sets 1172 * siginfo->si_addr to NULL. 1173 */ 1174 } 1175 1176 /* 1177 * There should be no reason to limit resources usage info to 1178 * exited processes only. A snapshot about any resources used 1179 * by a stopped process may be exactly what is needed. 1180 */ 1181 if (wrusage != NULL) { 1182 rup = &wrusage->wru_self; 1183 *rup = p->p_ru; 1184 PROC_STATLOCK(p); 1185 calcru(p, &rup->ru_utime, &rup->ru_stime); 1186 PROC_STATUNLOCK(p); 1187 1188 rup = &wrusage->wru_children; 1189 *rup = p->p_stats->p_cru; 1190 calccru(p, &rup->ru_utime, &rup->ru_stime); 1191 } 1192 1193 if (p->p_state == PRS_ZOMBIE && !check_only) { 1194 proc_reap(td, p, status, options); 1195 return (-1); 1196 } 1197 return (1); 1198 } 1199 1200 int 1201 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1202 struct rusage *rusage) 1203 { 1204 struct __wrusage wru, *wrup; 1205 idtype_t idtype; 1206 id_t id; 1207 int ret; 1208 1209 /* 1210 * Translate the special pid values into the (idtype, pid) 1211 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1212 * kern_wait6() on its own. 1213 */ 1214 if (pid == WAIT_ANY) { 1215 idtype = P_ALL; 1216 id = 0; 1217 } else if (pid < 0) { 1218 idtype = P_PGID; 1219 id = (id_t)-pid; 1220 } else { 1221 idtype = P_PID; 1222 id = (id_t)pid; 1223 } 1224 1225 if (rusage != NULL) 1226 wrup = &wru; 1227 else 1228 wrup = NULL; 1229 1230 /* 1231 * For backward compatibility we implicitly add flags WEXITED 1232 * and WTRAPPED here. 1233 */ 1234 options |= WEXITED | WTRAPPED; 1235 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1236 if (rusage != NULL) 1237 *rusage = wru.wru_self; 1238 return (ret); 1239 } 1240 1241 static void 1242 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1243 int *status, int options, int si_code) 1244 { 1245 bool cont; 1246 1247 PROC_LOCK_ASSERT(p, MA_OWNED); 1248 sx_assert(&proctree_lock, SA_XLOCKED); 1249 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1250 si_code == CLD_CONTINUED); 1251 1252 cont = si_code == CLD_CONTINUED; 1253 if ((options & WNOWAIT) == 0) { 1254 if (cont) 1255 p->p_flag &= ~P_CONTINUED; 1256 else 1257 p->p_flag |= P_WAITED; 1258 if (kern_wait_dequeue_sigchld && 1259 (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) { 1260 PROC_LOCK(td->td_proc); 1261 sigqueue_take(p->p_ksi); 1262 PROC_UNLOCK(td->td_proc); 1263 } 1264 } 1265 sx_xunlock(&proctree_lock); 1266 if (siginfo != NULL) { 1267 siginfo->si_code = si_code; 1268 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1269 } 1270 if (status != NULL) 1271 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1272 td->td_retval[0] = p->p_pid; 1273 PROC_UNLOCK(p); 1274 } 1275 1276 int 1277 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1278 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1279 { 1280 struct proc *p, *q; 1281 pid_t pid; 1282 int error, nfound, ret; 1283 bool report; 1284 1285 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1286 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1287 AUDIT_ARG_VALUE(options); 1288 1289 q = td->td_proc; 1290 1291 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1292 PROC_LOCK(q); 1293 id = (id_t)q->p_pgid; 1294 PROC_UNLOCK(q); 1295 idtype = P_PGID; 1296 } 1297 1298 /* If we don't know the option, just return. */ 1299 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1300 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1301 return (EINVAL); 1302 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1303 /* 1304 * We will be unable to find any matching processes, 1305 * because there are no known events to look for. 1306 * Prefer to return error instead of blocking 1307 * indefinitely. 1308 */ 1309 return (EINVAL); 1310 } 1311 1312 loop: 1313 if (q->p_flag & P_STATCHILD) { 1314 PROC_LOCK(q); 1315 q->p_flag &= ~P_STATCHILD; 1316 PROC_UNLOCK(q); 1317 } 1318 sx_xlock(&proctree_lock); 1319 loop_locked: 1320 nfound = 0; 1321 LIST_FOREACH(p, &q->p_children, p_sibling) { 1322 pid = p->p_pid; 1323 ret = proc_to_reap(td, p, idtype, id, status, options, 1324 wrusage, siginfo, 0); 1325 if (ret == 0) 1326 continue; 1327 else if (ret != 1) { 1328 td->td_retval[0] = pid; 1329 return (0); 1330 } 1331 1332 nfound++; 1333 PROC_LOCK_ASSERT(p, MA_OWNED); 1334 1335 if ((options & WTRAPPED) != 0 && 1336 (p->p_flag & P_TRACED) != 0) { 1337 PROC_SLOCK(p); 1338 report = 1339 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) && 1340 p->p_suspcount == p->p_numthreads && 1341 (p->p_flag & P_WAITED) == 0); 1342 PROC_SUNLOCK(p); 1343 if (report) { 1344 CTR4(KTR_PTRACE, 1345 "wait: returning trapped pid %d status %#x " 1346 "(xstat %d) xthread %d", 1347 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1348 p->p_xthread != NULL ? 1349 p->p_xthread->td_tid : -1); 1350 report_alive_proc(td, p, siginfo, status, 1351 options, CLD_TRAPPED); 1352 return (0); 1353 } 1354 } 1355 if ((options & WUNTRACED) != 0 && 1356 (p->p_flag & P_STOPPED_SIG) != 0) { 1357 PROC_SLOCK(p); 1358 report = (p->p_suspcount == p->p_numthreads && 1359 ((p->p_flag & P_WAITED) == 0)); 1360 PROC_SUNLOCK(p); 1361 if (report) { 1362 report_alive_proc(td, p, siginfo, status, 1363 options, CLD_STOPPED); 1364 return (0); 1365 } 1366 } 1367 if ((options & WCONTINUED) != 0 && 1368 (p->p_flag & P_CONTINUED) != 0) { 1369 report_alive_proc(td, p, siginfo, status, options, 1370 CLD_CONTINUED); 1371 return (0); 1372 } 1373 PROC_UNLOCK(p); 1374 } 1375 1376 /* 1377 * Look in the orphans list too, to allow the parent to 1378 * collect it's child exit status even if child is being 1379 * debugged. 1380 * 1381 * Debugger detaches from the parent upon successful 1382 * switch-over from parent to child. At this point due to 1383 * re-parenting the parent loses the child to debugger and a 1384 * wait4(2) call would report that it has no children to wait 1385 * for. By maintaining a list of orphans we allow the parent 1386 * to successfully wait until the child becomes a zombie. 1387 */ 1388 if (nfound == 0) { 1389 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1390 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1391 NULL, NULL, 1); 1392 if (ret != 0) { 1393 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1394 (int)td->td_retval[0])); 1395 PROC_UNLOCK(p); 1396 nfound++; 1397 break; 1398 } 1399 } 1400 } 1401 if (nfound == 0) { 1402 sx_xunlock(&proctree_lock); 1403 return (ECHILD); 1404 } 1405 if (options & WNOHANG) { 1406 sx_xunlock(&proctree_lock); 1407 td->td_retval[0] = 0; 1408 return (0); 1409 } 1410 PROC_LOCK(q); 1411 if (q->p_flag & P_STATCHILD) { 1412 q->p_flag &= ~P_STATCHILD; 1413 PROC_UNLOCK(q); 1414 goto loop_locked; 1415 } 1416 sx_xunlock(&proctree_lock); 1417 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0); 1418 if (error) 1419 return (error); 1420 goto loop; 1421 } 1422 1423 void 1424 proc_add_orphan(struct proc *child, struct proc *parent) 1425 { 1426 1427 sx_assert(&proctree_lock, SX_XLOCKED); 1428 KASSERT((child->p_flag & P_TRACED) != 0, 1429 ("proc_add_orphan: not traced")); 1430 1431 if (LIST_EMPTY(&parent->p_orphans)) { 1432 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1433 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan); 1434 } else { 1435 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans), 1436 child, p_orphan); 1437 } 1438 child->p_treeflag |= P_TREE_ORPHANED; 1439 } 1440 1441 /* 1442 * Make process 'parent' the new parent of process 'child'. 1443 * Must be called with an exclusive hold of proctree lock. 1444 */ 1445 void 1446 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid) 1447 { 1448 1449 sx_assert(&proctree_lock, SX_XLOCKED); 1450 PROC_LOCK_ASSERT(child, MA_OWNED); 1451 if (child->p_pptr == parent) 1452 return; 1453 1454 PROC_LOCK(child->p_pptr); 1455 sigqueue_take(child->p_ksi); 1456 PROC_UNLOCK(child->p_pptr); 1457 LIST_REMOVE(child, p_sibling); 1458 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1459 1460 proc_clear_orphan(child); 1461 if ((child->p_flag & P_TRACED) != 0) { 1462 proc_add_orphan(child, child->p_pptr); 1463 } 1464 1465 child->p_pptr = parent; 1466 if (set_oppid) 1467 child->p_oppid = parent->p_pid; 1468 } 1469