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 prison_proc_unlink(p->p_ucred->cr_prison, p); 477 sx_xunlock(&allproc_lock); 478 479 sx_xlock(&proctree_lock); 480 PROC_LOCK(p); 481 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 482 PROC_UNLOCK(p); 483 484 /* 485 * killjobc() might drop and re-acquire proctree_lock to 486 * revoke control tty if exiting process was a session leader. 487 */ 488 killjobc(); 489 490 /* 491 * Reparent all children processes: 492 * - traced ones to the original parent (or init if we are that parent) 493 * - the rest to init 494 */ 495 q = LIST_FIRST(&p->p_children); 496 if (q != NULL) /* only need this if any child is S_ZOMB */ 497 wakeup(q->p_reaper); 498 for (; q != NULL; q = nq) { 499 nq = LIST_NEXT(q, p_sibling); 500 ksi = ksiginfo_alloc(M_WAITOK); 501 PROC_LOCK(q); 502 q->p_sigparent = SIGCHLD; 503 504 if ((q->p_flag & P_TRACED) == 0) { 505 proc_reparent(q, q->p_reaper, true); 506 if (q->p_state == PRS_ZOMBIE) { 507 /* 508 * Inform reaper about the reparented 509 * zombie, since wait(2) has something 510 * new to report. Guarantee queueing 511 * of the SIGCHLD signal, similar to 512 * the _exit() behaviour, by providing 513 * our ksiginfo. Ksi is freed by the 514 * signal delivery. 515 */ 516 if (q->p_ksi == NULL) { 517 ksi1 = NULL; 518 } else { 519 ksiginfo_copy(q->p_ksi, ksi); 520 ksi->ksi_flags |= KSI_INS; 521 ksi1 = ksi; 522 ksi = NULL; 523 } 524 PROC_LOCK(q->p_reaper); 525 pksignal(q->p_reaper, SIGCHLD, ksi1); 526 PROC_UNLOCK(q->p_reaper); 527 } else if (q->p_pdeathsig > 0) { 528 /* 529 * The child asked to received a signal 530 * when we exit. 531 */ 532 kern_psignal(q, q->p_pdeathsig); 533 } 534 } else { 535 /* 536 * Traced processes are killed by default 537 * since their existence means someone is 538 * screwing up. 539 */ 540 t = proc_realparent(q); 541 if (t == p) { 542 proc_reparent(q, q->p_reaper, true); 543 } else { 544 PROC_LOCK(t); 545 proc_reparent(q, t, true); 546 PROC_UNLOCK(t); 547 } 548 /* 549 * Since q was found on our children list, the 550 * proc_reparent() call moved q to the orphan 551 * list due to present P_TRACED flag. Clear 552 * orphan link for q now while q is locked. 553 */ 554 proc_clear_orphan(q); 555 q->p_flag &= ~P_TRACED; 556 q->p_flag2 &= ~P2_PTRACE_FSTP; 557 q->p_ptevents = 0; 558 p->p_xthread = NULL; 559 FOREACH_THREAD_IN_PROC(q, tdt) { 560 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 561 TDB_FSTP); 562 tdt->td_xsig = 0; 563 } 564 if (kern_kill_on_dbg_exit) { 565 q->p_flag &= ~P_STOPPED_TRACE; 566 kern_psignal(q, SIGKILL); 567 } else if ((q->p_flag & (P_STOPPED_TRACE | 568 P_STOPPED_SIG)) != 0) { 569 sigqueue_delete_proc(q, SIGTRAP); 570 ptrace_unsuspend(q); 571 } 572 } 573 PROC_UNLOCK(q); 574 if (ksi != NULL) 575 ksiginfo_free(ksi); 576 } 577 578 /* 579 * Also get rid of our orphans. 580 */ 581 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 582 PROC_LOCK(q); 583 KASSERT(q->p_oppid == p->p_pid, 584 ("orphan %p of %p has unexpected oppid %d", q, p, 585 q->p_oppid)); 586 q->p_oppid = q->p_reaper->p_pid; 587 588 /* 589 * If we are the real parent of this process 590 * but it has been reparented to a debugger, then 591 * check if it asked for a signal when we exit. 592 */ 593 if (q->p_pdeathsig > 0) 594 kern_psignal(q, q->p_pdeathsig); 595 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 596 q->p_pid); 597 proc_clear_orphan(q); 598 PROC_UNLOCK(q); 599 } 600 601 #ifdef KDTRACE_HOOKS 602 if (SDT_PROBES_ENABLED()) { 603 int reason = CLD_EXITED; 604 if (WCOREDUMP(signo)) 605 reason = CLD_DUMPED; 606 else if (WIFSIGNALED(signo)) 607 reason = CLD_KILLED; 608 SDT_PROBE1(proc, , , exit, reason); 609 } 610 #endif 611 612 /* Save exit status. */ 613 PROC_LOCK(p); 614 p->p_xthread = td; 615 616 if (p->p_sysent->sv_ontdexit != NULL) 617 p->p_sysent->sv_ontdexit(td); 618 619 #ifdef KDTRACE_HOOKS 620 /* 621 * Tell the DTrace fasttrap provider about the exit if it 622 * has declared an interest. 623 */ 624 if (dtrace_fasttrap_exit) 625 dtrace_fasttrap_exit(p); 626 #endif 627 628 /* 629 * Notify interested parties of our demise. 630 */ 631 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 632 633 /* 634 * If this is a process with a descriptor, we may not need to deliver 635 * a signal to the parent. proctree_lock is held over 636 * procdesc_exit() to serialize concurrent calls to close() and 637 * exit(). 638 */ 639 signal_parent = 0; 640 if (p->p_procdesc == NULL || procdesc_exit(p)) { 641 /* 642 * Notify parent that we're gone. If parent has the 643 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 644 * notify process 1 instead (and hope it will handle this 645 * situation). 646 */ 647 PROC_LOCK(p->p_pptr); 648 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 649 if (p->p_pptr->p_sigacts->ps_flag & 650 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 651 struct proc *pp; 652 653 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 654 pp = p->p_pptr; 655 PROC_UNLOCK(pp); 656 proc_reparent(p, p->p_reaper, true); 657 p->p_sigparent = SIGCHLD; 658 PROC_LOCK(p->p_pptr); 659 660 /* 661 * Notify parent, so in case he was wait(2)ing or 662 * executing waitpid(2) with our pid, he will 663 * continue. 664 */ 665 wakeup(pp); 666 } else 667 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 668 669 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) { 670 signal_parent = 1; 671 } else if (p->p_sigparent != 0) { 672 if (p->p_sigparent == SIGCHLD) { 673 signal_parent = 1; 674 } else { /* LINUX thread */ 675 signal_parent = 2; 676 } 677 } 678 } else 679 PROC_LOCK(p->p_pptr); 680 sx_xunlock(&proctree_lock); 681 682 if (signal_parent == 1) { 683 childproc_exited(p); 684 } else if (signal_parent == 2) { 685 kern_psignal(p->p_pptr, p->p_sigparent); 686 } 687 688 /* Tell the prison that we are gone. */ 689 prison_proc_free(p->p_ucred->cr_prison); 690 691 /* 692 * The state PRS_ZOMBIE prevents other proesses from sending 693 * signal to the process, to avoid memory leak, we free memory 694 * for signal queue at the time when the state is set. 695 */ 696 sigqueue_flush(&p->p_sigqueue); 697 sigqueue_flush(&td->td_sigqueue); 698 699 /* 700 * We have to wait until after acquiring all locks before 701 * changing p_state. We need to avoid all possible context 702 * switches (including ones from blocking on a mutex) while 703 * marked as a zombie. We also have to set the zombie state 704 * before we release the parent process' proc lock to avoid 705 * a lost wakeup. So, we first call wakeup, then we grab the 706 * sched lock, update the state, and release the parent process' 707 * proc lock. 708 */ 709 wakeup(p->p_pptr); 710 cv_broadcast(&p->p_pwait); 711 sched_exit(p->p_pptr, td); 712 PROC_SLOCK(p); 713 p->p_state = PRS_ZOMBIE; 714 PROC_UNLOCK(p->p_pptr); 715 716 /* 717 * Save our children's rusage information in our exit rusage. 718 */ 719 PROC_STATLOCK(p); 720 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 721 PROC_STATUNLOCK(p); 722 723 /* 724 * Make sure the scheduler takes this thread out of its tables etc. 725 * This will also release this thread's reference to the ucred. 726 * Other thread parts to release include pcb bits and such. 727 */ 728 thread_exit(); 729 } 730 731 #ifndef _SYS_SYSPROTO_H_ 732 struct abort2_args { 733 char *why; 734 int nargs; 735 void **args; 736 }; 737 #endif 738 739 int 740 sys_abort2(struct thread *td, struct abort2_args *uap) 741 { 742 void *uargs[16]; 743 void **uargsp; 744 int error, nargs; 745 746 nargs = uap->nargs; 747 if (nargs < 0 || nargs > nitems(uargs)) 748 nargs = -1; 749 uargsp = NULL; 750 if (nargs > 0) { 751 if (uap->args != NULL) { 752 error = copyin(uap->args, uargs, 753 nargs * sizeof(void *)); 754 if (error != 0) 755 nargs = -1; 756 else 757 uargsp = uargs; 758 } else 759 nargs = -1; 760 } 761 return (kern_abort2(td, uap->why, nargs, uargsp)); 762 } 763 764 /* 765 * kern_abort2() 766 * Arguments: 767 * why - user pointer to why 768 * nargs - number of arguments copied or -1 if an error occurred in copying 769 * args - pointer to an array of pointers in kernel format 770 */ 771 int 772 kern_abort2(struct thread *td, const char *why, int nargs, void **uargs) 773 { 774 struct proc *p = td->td_proc; 775 struct sbuf *sb; 776 int error, i, sig; 777 778 /* 779 * Do it right now so we can log either proper call of abort2(), or 780 * note, that invalid argument was passed. 512 is big enough to 781 * handle 16 arguments' descriptions with additional comments. 782 */ 783 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 784 sbuf_clear(sb); 785 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 786 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 787 /* 788 * Since we can't return from abort2(), send SIGKILL in cases, where 789 * abort2() was called improperly 790 */ 791 sig = SIGKILL; 792 /* Prevent from DoSes from user-space. */ 793 if (nargs == -1) 794 goto out; 795 KASSERT(nargs >= 0 && nargs <= 16, ("called with too many args (%d)", 796 nargs)); 797 /* 798 * Limit size of 'reason' string to 128. Will fit even when 799 * maximal number of arguments was chosen to be logged. 800 */ 801 if (why != NULL) { 802 error = sbuf_copyin(sb, why, 128); 803 if (error < 0) 804 goto out; 805 } else { 806 sbuf_printf(sb, "(null)"); 807 } 808 if (nargs > 0) { 809 sbuf_printf(sb, "("); 810 for (i = 0;i < nargs; i++) 811 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 812 sbuf_printf(sb, ")"); 813 } 814 /* 815 * Final stage: arguments were proper, string has been 816 * successfully copied from userspace, and copying pointers 817 * from user-space succeed. 818 */ 819 sig = SIGABRT; 820 out: 821 if (sig == SIGKILL) { 822 sbuf_trim(sb); 823 sbuf_printf(sb, " (Reason text inaccessible)"); 824 } 825 sbuf_cat(sb, "\n"); 826 sbuf_finish(sb); 827 log(LOG_INFO, "%s", sbuf_data(sb)); 828 sbuf_delete(sb); 829 exit1(td, 0, sig); 830 return (0); 831 } 832 833 #ifdef COMPAT_43 834 /* 835 * The dirty work is handled by kern_wait(). 836 */ 837 int 838 owait(struct thread *td, struct owait_args *uap __unused) 839 { 840 int error, status; 841 842 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 843 if (error == 0) 844 td->td_retval[1] = status; 845 return (error); 846 } 847 #endif /* COMPAT_43 */ 848 849 /* 850 * The dirty work is handled by kern_wait(). 851 */ 852 int 853 sys_wait4(struct thread *td, struct wait4_args *uap) 854 { 855 struct rusage ru, *rup; 856 int error, status; 857 858 if (uap->rusage != NULL) 859 rup = &ru; 860 else 861 rup = NULL; 862 error = kern_wait(td, uap->pid, &status, uap->options, rup); 863 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 864 error = copyout(&status, uap->status, sizeof(status)); 865 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 866 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 867 return (error); 868 } 869 870 int 871 sys_wait6(struct thread *td, struct wait6_args *uap) 872 { 873 struct __wrusage wru, *wrup; 874 siginfo_t si, *sip; 875 idtype_t idtype; 876 id_t id; 877 int error, status; 878 879 idtype = uap->idtype; 880 id = uap->id; 881 882 if (uap->wrusage != NULL) 883 wrup = &wru; 884 else 885 wrup = NULL; 886 887 if (uap->info != NULL) { 888 sip = &si; 889 bzero(sip, sizeof(*sip)); 890 } else 891 sip = NULL; 892 893 /* 894 * We expect all callers of wait6() to know about WEXITED and 895 * WTRAPPED. 896 */ 897 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 898 899 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 900 error = copyout(&status, uap->status, sizeof(status)); 901 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 902 error = copyout(&wru, uap->wrusage, sizeof(wru)); 903 if (uap->info != NULL && error == 0) 904 error = copyout(&si, uap->info, sizeof(si)); 905 return (error); 906 } 907 908 /* 909 * Reap the remains of a zombie process and optionally return status and 910 * rusage. Asserts and will release both the proctree_lock and the process 911 * lock as part of its work. 912 */ 913 void 914 proc_reap(struct thread *td, struct proc *p, int *status, int options) 915 { 916 struct proc *q, *t; 917 918 sx_assert(&proctree_lock, SA_XLOCKED); 919 PROC_LOCK_ASSERT(p, MA_OWNED); 920 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 921 922 mtx_spin_wait_unlocked(&p->p_slock); 923 924 q = td->td_proc; 925 926 if (status) 927 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 928 if (options & WNOWAIT) { 929 /* 930 * Only poll, returning the status. Caller does not wish to 931 * release the proc struct just yet. 932 */ 933 PROC_UNLOCK(p); 934 sx_xunlock(&proctree_lock); 935 return; 936 } 937 938 PROC_LOCK(q); 939 sigqueue_take(p->p_ksi); 940 PROC_UNLOCK(q); 941 942 /* 943 * If we got the child via a ptrace 'attach', we need to give it back 944 * to the old parent. 945 */ 946 if (p->p_oppid != p->p_pptr->p_pid) { 947 PROC_UNLOCK(p); 948 t = proc_realparent(p); 949 PROC_LOCK(t); 950 PROC_LOCK(p); 951 CTR2(KTR_PTRACE, 952 "wait: traced child %d moved back to parent %d", p->p_pid, 953 t->p_pid); 954 proc_reparent(p, t, false); 955 PROC_UNLOCK(p); 956 pksignal(t, SIGCHLD, p->p_ksi); 957 wakeup(t); 958 cv_broadcast(&p->p_pwait); 959 PROC_UNLOCK(t); 960 sx_xunlock(&proctree_lock); 961 return; 962 } 963 PROC_UNLOCK(p); 964 965 /* 966 * Remove other references to this process to ensure we have an 967 * exclusive reference. 968 */ 969 sx_xlock(PIDHASHLOCK(p->p_pid)); 970 LIST_REMOVE(p, p_hash); 971 sx_xunlock(PIDHASHLOCK(p->p_pid)); 972 LIST_REMOVE(p, p_sibling); 973 reaper_abandon_children(p, true); 974 reaper_clear(p); 975 PROC_LOCK(p); 976 proc_clear_orphan(p); 977 PROC_UNLOCK(p); 978 leavepgrp(p); 979 if (p->p_procdesc != NULL) 980 procdesc_reap(p); 981 sx_xunlock(&proctree_lock); 982 983 proc_id_clear(PROC_ID_PID, p->p_pid); 984 985 PROC_LOCK(p); 986 knlist_detach(p->p_klist); 987 p->p_klist = NULL; 988 PROC_UNLOCK(p); 989 990 /* 991 * Removal from allproc list and process group list paired with 992 * PROC_LOCK which was executed during that time should guarantee 993 * nothing can reach this process anymore. As such further locking 994 * is unnecessary. 995 */ 996 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 997 998 PROC_LOCK(q); 999 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 1000 PROC_UNLOCK(q); 1001 1002 /* 1003 * Decrement the count of procs running with this uid. 1004 */ 1005 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 1006 1007 /* 1008 * Destroy resource accounting information associated with the process. 1009 */ 1010 #ifdef RACCT 1011 if (racct_enable) { 1012 PROC_LOCK(p); 1013 racct_sub(p, RACCT_NPROC, 1); 1014 PROC_UNLOCK(p); 1015 } 1016 #endif 1017 racct_proc_exit(p); 1018 1019 /* 1020 * Free credentials, arguments, and sigacts. 1021 */ 1022 proc_unset_cred(p); 1023 pargs_drop(p->p_args); 1024 p->p_args = NULL; 1025 sigacts_free(p->p_sigacts); 1026 p->p_sigacts = NULL; 1027 1028 /* 1029 * Do any thread-system specific cleanups. 1030 */ 1031 thread_wait(p); 1032 1033 /* 1034 * Give vm and machine-dependent layer a chance to free anything that 1035 * cpu_exit couldn't release while still running in process context. 1036 */ 1037 vm_waitproc(p); 1038 #ifdef MAC 1039 mac_proc_destroy(p); 1040 #endif 1041 1042 KASSERT(FIRST_THREAD_IN_PROC(p), 1043 ("proc_reap: no residual thread!")); 1044 uma_zfree(proc_zone, p); 1045 atomic_add_int(&nprocs, -1); 1046 } 1047 1048 static int 1049 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 1050 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 1051 int check_only) 1052 { 1053 struct rusage *rup; 1054 1055 sx_assert(&proctree_lock, SA_XLOCKED); 1056 1057 PROC_LOCK(p); 1058 1059 switch (idtype) { 1060 case P_ALL: 1061 if (p->p_procdesc == NULL || 1062 (p->p_pptr == td->td_proc && 1063 (p->p_flag & P_TRACED) != 0)) { 1064 break; 1065 } 1066 1067 PROC_UNLOCK(p); 1068 return (0); 1069 case P_PID: 1070 if (p->p_pid != (pid_t)id) { 1071 PROC_UNLOCK(p); 1072 return (0); 1073 } 1074 break; 1075 case P_PGID: 1076 if (p->p_pgid != (pid_t)id) { 1077 PROC_UNLOCK(p); 1078 return (0); 1079 } 1080 break; 1081 case P_SID: 1082 if (p->p_session->s_sid != (pid_t)id) { 1083 PROC_UNLOCK(p); 1084 return (0); 1085 } 1086 break; 1087 case P_UID: 1088 if (p->p_ucred->cr_uid != (uid_t)id) { 1089 PROC_UNLOCK(p); 1090 return (0); 1091 } 1092 break; 1093 case P_GID: 1094 if (p->p_ucred->cr_gid != (gid_t)id) { 1095 PROC_UNLOCK(p); 1096 return (0); 1097 } 1098 break; 1099 case P_JAILID: 1100 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1101 PROC_UNLOCK(p); 1102 return (0); 1103 } 1104 break; 1105 /* 1106 * It seems that the thread structures get zeroed out 1107 * at process exit. This makes it impossible to 1108 * support P_SETID, P_CID or P_CPUID. 1109 */ 1110 default: 1111 PROC_UNLOCK(p); 1112 return (0); 1113 } 1114 1115 if (p_canwait(td, p)) { 1116 PROC_UNLOCK(p); 1117 return (0); 1118 } 1119 1120 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1121 PROC_UNLOCK(p); 1122 return (0); 1123 } 1124 1125 /* 1126 * This special case handles a kthread spawned by linux_clone 1127 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1128 * functions need to be able to distinguish between waiting 1129 * on a process and waiting on a thread. It is a thread if 1130 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1131 * signifies we want to wait for threads and not processes. 1132 */ 1133 if ((p->p_sigparent != SIGCHLD) ^ 1134 ((options & WLINUXCLONE) != 0)) { 1135 PROC_UNLOCK(p); 1136 return (0); 1137 } 1138 1139 if (siginfo != NULL) { 1140 bzero(siginfo, sizeof(*siginfo)); 1141 siginfo->si_errno = 0; 1142 1143 /* 1144 * SUSv4 requires that the si_signo value is always 1145 * SIGCHLD. Obey it despite the rfork(2) interface 1146 * allows to request other signal for child exit 1147 * notification. 1148 */ 1149 siginfo->si_signo = SIGCHLD; 1150 1151 /* 1152 * This is still a rough estimate. We will fix the 1153 * cases TRAPPED, STOPPED, and CONTINUED later. 1154 */ 1155 if (WCOREDUMP(p->p_xsig)) { 1156 siginfo->si_code = CLD_DUMPED; 1157 siginfo->si_status = WTERMSIG(p->p_xsig); 1158 } else if (WIFSIGNALED(p->p_xsig)) { 1159 siginfo->si_code = CLD_KILLED; 1160 siginfo->si_status = WTERMSIG(p->p_xsig); 1161 } else { 1162 siginfo->si_code = CLD_EXITED; 1163 siginfo->si_status = p->p_xexit; 1164 } 1165 1166 siginfo->si_pid = p->p_pid; 1167 siginfo->si_uid = p->p_ucred->cr_uid; 1168 1169 /* 1170 * The si_addr field would be useful additional 1171 * detail, but apparently the PC value may be lost 1172 * when we reach this point. bzero() above sets 1173 * siginfo->si_addr to NULL. 1174 */ 1175 } 1176 1177 /* 1178 * There should be no reason to limit resources usage info to 1179 * exited processes only. A snapshot about any resources used 1180 * by a stopped process may be exactly what is needed. 1181 */ 1182 if (wrusage != NULL) { 1183 rup = &wrusage->wru_self; 1184 *rup = p->p_ru; 1185 PROC_STATLOCK(p); 1186 calcru(p, &rup->ru_utime, &rup->ru_stime); 1187 PROC_STATUNLOCK(p); 1188 1189 rup = &wrusage->wru_children; 1190 *rup = p->p_stats->p_cru; 1191 calccru(p, &rup->ru_utime, &rup->ru_stime); 1192 } 1193 1194 if (p->p_state == PRS_ZOMBIE && !check_only) { 1195 proc_reap(td, p, status, options); 1196 return (-1); 1197 } 1198 return (1); 1199 } 1200 1201 int 1202 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1203 struct rusage *rusage) 1204 { 1205 struct __wrusage wru, *wrup; 1206 idtype_t idtype; 1207 id_t id; 1208 int ret; 1209 1210 /* 1211 * Translate the special pid values into the (idtype, pid) 1212 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1213 * kern_wait6() on its own. 1214 */ 1215 if (pid == WAIT_ANY) { 1216 idtype = P_ALL; 1217 id = 0; 1218 } else if (pid < 0) { 1219 idtype = P_PGID; 1220 id = (id_t)-pid; 1221 } else { 1222 idtype = P_PID; 1223 id = (id_t)pid; 1224 } 1225 1226 if (rusage != NULL) 1227 wrup = &wru; 1228 else 1229 wrup = NULL; 1230 1231 /* 1232 * For backward compatibility we implicitly add flags WEXITED 1233 * and WTRAPPED here. 1234 */ 1235 options |= WEXITED | WTRAPPED; 1236 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1237 if (rusage != NULL) 1238 *rusage = wru.wru_self; 1239 return (ret); 1240 } 1241 1242 static void 1243 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1244 int *status, int options, int si_code) 1245 { 1246 bool cont; 1247 1248 PROC_LOCK_ASSERT(p, MA_OWNED); 1249 sx_assert(&proctree_lock, SA_XLOCKED); 1250 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1251 si_code == CLD_CONTINUED); 1252 1253 cont = si_code == CLD_CONTINUED; 1254 if ((options & WNOWAIT) == 0) { 1255 if (cont) 1256 p->p_flag &= ~P_CONTINUED; 1257 else 1258 p->p_flag |= P_WAITED; 1259 if (kern_wait_dequeue_sigchld && 1260 (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) { 1261 PROC_LOCK(td->td_proc); 1262 sigqueue_take(p->p_ksi); 1263 PROC_UNLOCK(td->td_proc); 1264 } 1265 } 1266 sx_xunlock(&proctree_lock); 1267 if (siginfo != NULL) { 1268 siginfo->si_code = si_code; 1269 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1270 } 1271 if (status != NULL) 1272 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1273 td->td_retval[0] = p->p_pid; 1274 PROC_UNLOCK(p); 1275 } 1276 1277 int 1278 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1279 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1280 { 1281 struct proc *p, *q; 1282 pid_t pid; 1283 int error, nfound, ret; 1284 bool report; 1285 1286 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1287 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1288 AUDIT_ARG_VALUE(options); 1289 1290 q = td->td_proc; 1291 1292 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1293 PROC_LOCK(q); 1294 id = (id_t)q->p_pgid; 1295 PROC_UNLOCK(q); 1296 idtype = P_PGID; 1297 } 1298 1299 /* If we don't know the option, just return. */ 1300 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1301 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1302 return (EINVAL); 1303 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1304 /* 1305 * We will be unable to find any matching processes, 1306 * because there are no known events to look for. 1307 * Prefer to return error instead of blocking 1308 * indefinitely. 1309 */ 1310 return (EINVAL); 1311 } 1312 1313 loop: 1314 if (q->p_flag & P_STATCHILD) { 1315 PROC_LOCK(q); 1316 q->p_flag &= ~P_STATCHILD; 1317 PROC_UNLOCK(q); 1318 } 1319 sx_xlock(&proctree_lock); 1320 loop_locked: 1321 nfound = 0; 1322 LIST_FOREACH(p, &q->p_children, p_sibling) { 1323 pid = p->p_pid; 1324 ret = proc_to_reap(td, p, idtype, id, status, options, 1325 wrusage, siginfo, 0); 1326 if (ret == 0) 1327 continue; 1328 else if (ret != 1) { 1329 td->td_retval[0] = pid; 1330 return (0); 1331 } 1332 1333 nfound++; 1334 PROC_LOCK_ASSERT(p, MA_OWNED); 1335 1336 if ((options & WTRAPPED) != 0 && 1337 (p->p_flag & P_TRACED) != 0) { 1338 PROC_SLOCK(p); 1339 report = 1340 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) && 1341 p->p_suspcount == p->p_numthreads && 1342 (p->p_flag & P_WAITED) == 0); 1343 PROC_SUNLOCK(p); 1344 if (report) { 1345 CTR4(KTR_PTRACE, 1346 "wait: returning trapped pid %d status %#x " 1347 "(xstat %d) xthread %d", 1348 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1349 p->p_xthread != NULL ? 1350 p->p_xthread->td_tid : -1); 1351 report_alive_proc(td, p, siginfo, status, 1352 options, CLD_TRAPPED); 1353 return (0); 1354 } 1355 } 1356 if ((options & WUNTRACED) != 0 && 1357 (p->p_flag & P_STOPPED_SIG) != 0) { 1358 PROC_SLOCK(p); 1359 report = (p->p_suspcount == p->p_numthreads && 1360 ((p->p_flag & P_WAITED) == 0)); 1361 PROC_SUNLOCK(p); 1362 if (report) { 1363 report_alive_proc(td, p, siginfo, status, 1364 options, CLD_STOPPED); 1365 return (0); 1366 } 1367 } 1368 if ((options & WCONTINUED) != 0 && 1369 (p->p_flag & P_CONTINUED) != 0) { 1370 report_alive_proc(td, p, siginfo, status, options, 1371 CLD_CONTINUED); 1372 return (0); 1373 } 1374 PROC_UNLOCK(p); 1375 } 1376 1377 /* 1378 * Look in the orphans list too, to allow the parent to 1379 * collect it's child exit status even if child is being 1380 * debugged. 1381 * 1382 * Debugger detaches from the parent upon successful 1383 * switch-over from parent to child. At this point due to 1384 * re-parenting the parent loses the child to debugger and a 1385 * wait4(2) call would report that it has no children to wait 1386 * for. By maintaining a list of orphans we allow the parent 1387 * to successfully wait until the child becomes a zombie. 1388 */ 1389 if (nfound == 0) { 1390 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1391 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1392 NULL, NULL, 1); 1393 if (ret != 0) { 1394 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1395 (int)td->td_retval[0])); 1396 PROC_UNLOCK(p); 1397 nfound++; 1398 break; 1399 } 1400 } 1401 } 1402 if (nfound == 0) { 1403 sx_xunlock(&proctree_lock); 1404 return (ECHILD); 1405 } 1406 if (options & WNOHANG) { 1407 sx_xunlock(&proctree_lock); 1408 td->td_retval[0] = 0; 1409 return (0); 1410 } 1411 PROC_LOCK(q); 1412 if (q->p_flag & P_STATCHILD) { 1413 q->p_flag &= ~P_STATCHILD; 1414 PROC_UNLOCK(q); 1415 goto loop_locked; 1416 } 1417 sx_xunlock(&proctree_lock); 1418 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0); 1419 if (error) 1420 return (error); 1421 goto loop; 1422 } 1423 1424 void 1425 proc_add_orphan(struct proc *child, struct proc *parent) 1426 { 1427 1428 sx_assert(&proctree_lock, SX_XLOCKED); 1429 KASSERT((child->p_flag & P_TRACED) != 0, 1430 ("proc_add_orphan: not traced")); 1431 1432 if (LIST_EMPTY(&parent->p_orphans)) { 1433 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1434 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan); 1435 } else { 1436 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans), 1437 child, p_orphan); 1438 } 1439 child->p_treeflag |= P_TREE_ORPHANED; 1440 } 1441 1442 /* 1443 * Make process 'parent' the new parent of process 'child'. 1444 * Must be called with an exclusive hold of proctree lock. 1445 */ 1446 void 1447 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid) 1448 { 1449 1450 sx_assert(&proctree_lock, SX_XLOCKED); 1451 PROC_LOCK_ASSERT(child, MA_OWNED); 1452 if (child->p_pptr == parent) 1453 return; 1454 1455 PROC_LOCK(child->p_pptr); 1456 sigqueue_take(child->p_ksi); 1457 PROC_UNLOCK(child->p_pptr); 1458 LIST_REMOVE(child, p_sibling); 1459 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1460 1461 proc_clear_orphan(child); 1462 if ((child->p_flag & P_TRACED) != 0) { 1463 proc_add_orphan(child, child->p_pptr); 1464 } 1465 1466 child->p_pptr = parent; 1467 if (set_oppid) 1468 child->p_oppid = parent->p_pid; 1469 } 1470