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 37 #include <sys/cdefs.h> 38 #include "opt_ddb.h" 39 #include "opt_ktrace.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/sysproto.h> 44 #include <sys/capsicum.h> 45 #include <sys/eventhandler.h> 46 #include <sys/kernel.h> 47 #include <sys/ktr.h> 48 #include <sys/malloc.h> 49 #include <sys/lock.h> 50 #include <sys/mutex.h> 51 #include <sys/proc.h> 52 #include <sys/procdesc.h> 53 #include <sys/jail.h> 54 #include <sys/tty.h> 55 #include <sys/wait.h> 56 #include <sys/vmmeter.h> 57 #include <sys/vnode.h> 58 #include <sys/racct.h> 59 #include <sys/resourcevar.h> 60 #include <sys/sbuf.h> 61 #include <sys/signalvar.h> 62 #include <sys/sched.h> 63 #include <sys/sx.h> 64 #include <sys/syscallsubr.h> 65 #include <sys/sysctl.h> 66 #include <sys/syslog.h> 67 #include <sys/ptrace.h> 68 #include <sys/acct.h> /* for acct_process() function prototype */ 69 #include <sys/filedesc.h> 70 #include <sys/sdt.h> 71 #include <sys/shm.h> 72 #include <sys/sem.h> 73 #include <sys/sysent.h> 74 #include <sys/timers.h> 75 #include <sys/umtxvar.h> 76 #ifdef KTRACE 77 #include <sys/ktrace.h> 78 #endif 79 80 #include <security/audit/audit.h> 81 #include <security/mac/mac_framework.h> 82 83 #include <vm/vm.h> 84 #include <vm/vm_extern.h> 85 #include <vm/vm_param.h> 86 #include <vm/pmap.h> 87 #include <vm/vm_map.h> 88 #include <vm/vm_page.h> 89 #include <vm/uma.h> 90 91 #ifdef KDTRACE_HOOKS 92 #include <sys/dtrace_bsd.h> 93 dtrace_execexit_func_t dtrace_fasttrap_exit; 94 #endif 95 96 SDT_PROVIDER_DECLARE(proc); 97 SDT_PROBE_DEFINE1(proc, , , exit, "int"); 98 99 static int kern_kill_on_dbg_exit = 1; 100 SYSCTL_INT(_kern, OID_AUTO, kill_on_debugger_exit, CTLFLAG_RWTUN, 101 &kern_kill_on_dbg_exit, 0, 102 "Kill ptraced processes when debugger exits"); 103 104 static bool kern_wait_dequeue_sigchld = 1; 105 SYSCTL_BOOL(_kern, OID_AUTO, wait_dequeue_sigchld, CTLFLAG_RWTUN, 106 &kern_wait_dequeue_sigchld, 0, 107 "Dequeue SIGCHLD on wait(2) for live process"); 108 109 struct proc * 110 proc_realparent(struct proc *child) 111 { 112 struct proc *p, *parent; 113 114 sx_assert(&proctree_lock, SX_LOCKED); 115 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) 116 return (child->p_pptr->p_pid == child->p_oppid ? 117 child->p_pptr : child->p_reaper); 118 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 119 /* Cannot use LIST_PREV(), since the list head is not known. */ 120 p = __containerof(p->p_orphan.le_prev, struct proc, 121 p_orphan.le_next); 122 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 123 ("missing P_ORPHAN %p", p)); 124 } 125 parent = __containerof(p->p_orphan.le_prev, struct proc, 126 p_orphans.lh_first); 127 return (parent); 128 } 129 130 void 131 reaper_abandon_children(struct proc *p, bool exiting) 132 { 133 struct proc *p1, *p2, *ptmp; 134 135 sx_assert(&proctree_lock, SX_XLOCKED); 136 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 137 if ((p->p_treeflag & P_TREE_REAPER) == 0) 138 return; 139 p1 = p->p_reaper; 140 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 141 LIST_REMOVE(p2, p_reapsibling); 142 p2->p_reaper = p1; 143 p2->p_reapsubtree = p->p_reapsubtree; 144 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 145 if (exiting && p2->p_pptr == p) { 146 PROC_LOCK(p2); 147 proc_reparent(p2, p1, true); 148 PROC_UNLOCK(p2); 149 } 150 } 151 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 152 p->p_treeflag &= ~P_TREE_REAPER; 153 } 154 155 static void 156 reaper_clear(struct proc *p) 157 { 158 struct proc *p1; 159 bool clear; 160 161 sx_assert(&proctree_lock, SX_LOCKED); 162 LIST_REMOVE(p, p_reapsibling); 163 if (p->p_reapsubtree == 1) 164 return; 165 clear = true; 166 LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) { 167 if (p1->p_reapsubtree == p->p_reapsubtree) { 168 clear = false; 169 break; 170 } 171 } 172 if (clear) 173 proc_id_clear(PROC_ID_REAP, p->p_reapsubtree); 174 } 175 176 void 177 proc_clear_orphan(struct proc *p) 178 { 179 struct proc *p1; 180 181 sx_assert(&proctree_lock, SA_XLOCKED); 182 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 183 return; 184 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 185 p1 = LIST_NEXT(p, p_orphan); 186 if (p1 != NULL) 187 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 188 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 189 } 190 LIST_REMOVE(p, p_orphan); 191 p->p_treeflag &= ~P_TREE_ORPHANED; 192 } 193 194 void 195 exit_onexit(struct proc *p) 196 { 197 MPASS(p->p_numthreads == 1); 198 umtx_thread_exit(FIRST_THREAD_IN_PROC(p)); 199 } 200 201 /* 202 * exit -- death of process. 203 */ 204 int 205 sys_exit(struct thread *td, struct exit_args *uap) 206 { 207 208 exit1(td, uap->rval, 0); 209 __unreachable(); 210 } 211 212 void 213 proc_set_p2_wexit(struct proc *p) 214 { 215 PROC_LOCK_ASSERT(p, MA_OWNED); 216 p->p_flag2 |= P2_WEXIT; 217 } 218 219 /* 220 * Exit: deallocate address space and other resources, change proc state to 221 * zombie, and unlink proc from allproc and parent's lists. Save exit status 222 * and rusage for wait(). Check for child processes and orphan them. 223 */ 224 void 225 exit1(struct thread *td, int rval, int signo) 226 { 227 struct proc *p, *nq, *q, *t; 228 struct thread *tdt; 229 ksiginfo_t *ksi, *ksi1; 230 int signal_parent; 231 232 mtx_assert(&Giant, MA_NOTOWNED); 233 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 234 TSPROCEXIT(td->td_proc->p_pid); 235 236 p = td->td_proc; 237 /* 238 * In case we're rebooting we just let init die in order to 239 * work around an issues where pid 1 might get a fatal signal. 240 * For instance, if network interface serving NFS root is 241 * going down due to reboot, page-in requests for text are 242 * failing. 243 */ 244 if (p == initproc && rebooting == 0) { 245 printf("init died (signal %d, exit %d)\n", signo, rval); 246 panic("Going nowhere without my init!"); 247 } 248 249 /* 250 * Process deferred operations, designated with ASTF_KCLEAR. 251 * For instance, we need to deref SU mp, since the thread does 252 * not return to userspace, and wait for geom to stabilize. 253 */ 254 ast_kclear(td); 255 256 /* 257 * MUST abort all other threads before proceeding past here. 258 */ 259 PROC_LOCK(p); 260 proc_set_p2_wexit(p); 261 262 /* 263 * First check if some other thread or external request got 264 * here before us. If so, act appropriately: exit or suspend. 265 * We must ensure that stop requests are handled before we set 266 * P_WEXIT. 267 */ 268 thread_suspend_check(0); 269 while (p->p_flag & P_HADTHREADS) { 270 /* 271 * Kill off the other threads. This requires 272 * some co-operation from other parts of the kernel 273 * so it may not be instantaneous. With this state set 274 * any thread attempting to interruptibly 275 * sleep will return immediately with EINTR or EWOULDBLOCK 276 * which will hopefully force them to back out to userland 277 * freeing resources as they go. Any thread attempting 278 * to return to userland will thread_exit() from ast(). 279 * thread_exit() will unsuspend us when the last of the 280 * other threads exits. 281 * If there is already a thread singler after resumption, 282 * calling thread_single() will fail; in that case, we just 283 * re-check all suspension request, the thread should 284 * either be suspended there or exit. 285 */ 286 if (!thread_single(p, SINGLE_EXIT)) 287 /* 288 * All other activity in this process is now 289 * stopped. Threading support has been turned 290 * off. 291 */ 292 break; 293 /* 294 * Recheck for new stop or suspend requests which 295 * might appear while process lock was dropped in 296 * thread_single(). 297 */ 298 thread_suspend_check(0); 299 } 300 KASSERT(p->p_numthreads == 1, 301 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 302 racct_sub(p, RACCT_NTHR, 1); 303 304 /* Let event handler change exit status */ 305 p->p_xexit = rval; 306 p->p_xsig = signo; 307 308 /* 309 * Ignore any pending request to stop due to a stop signal. 310 * Once P_WEXIT is set, future requests will be ignored as 311 * well. 312 */ 313 p->p_flag &= ~P_STOPPED_SIG; 314 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 315 316 /* Note that we are exiting. */ 317 p->p_flag |= P_WEXIT; 318 319 /* 320 * Wait for any processes that have a hold on our vmspace to 321 * release their reference. 322 */ 323 while (p->p_lock > 0) 324 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 325 326 PROC_UNLOCK(p); 327 /* Drain the limit callout while we don't have the proc locked */ 328 callout_drain(&p->p_limco); 329 330 #ifdef AUDIT 331 /* 332 * The Sun BSM exit token contains two components: an exit status as 333 * passed to exit(), and a return value to indicate what sort of exit 334 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 335 * what the return value is. 336 */ 337 AUDIT_ARG_EXIT(rval, 0); 338 AUDIT_SYSCALL_EXIT(0, td); 339 #endif 340 341 /* Are we a task leader with peers? */ 342 if (p->p_peers != NULL && p == p->p_leader) { 343 mtx_lock(&ppeers_lock); 344 q = p->p_peers; 345 while (q != NULL) { 346 PROC_LOCK(q); 347 kern_psignal(q, SIGKILL); 348 PROC_UNLOCK(q); 349 q = q->p_peers; 350 } 351 while (p->p_peers != NULL) 352 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 353 mtx_unlock(&ppeers_lock); 354 } 355 356 itimers_exit(p); 357 358 /* 359 * Check if any loadable modules need anything done at process exit. 360 * E.g. SYSV IPC stuff. 361 * Event handler could change exit status. 362 * XXX what if one of these generates an error? 363 */ 364 EVENTHANDLER_DIRECT_INVOKE(process_exit, p); 365 366 /* 367 * If parent is waiting for us to exit or exec, 368 * P_PPWAIT is set; we will wakeup the parent below. 369 */ 370 PROC_LOCK(p); 371 stopprofclock(p); 372 p->p_ptevents = 0; 373 374 /* 375 * Stop the real interval timer. If the handler is currently 376 * executing, prevent it from rearming itself and let it finish. 377 */ 378 p->p_flag2 &= ~P2_ITSTOPPED; 379 if (timevalisset(&p->p_realtimer.it_value) && 380 callout_stop(&p->p_itcallout) == 0) { 381 timevalclear(&p->p_realtimer.it_interval); 382 PROC_UNLOCK(p); 383 callout_drain(&p->p_itcallout); 384 } else { 385 PROC_UNLOCK(p); 386 } 387 388 if (p->p_sysent->sv_onexit != NULL) 389 p->p_sysent->sv_onexit(p); 390 seltdfini(td); 391 392 /* 393 * Reset any sigio structures pointing to us as a result of 394 * F_SETOWN with our pid. The P_WEXIT flag interlocks with fsetown(). 395 */ 396 funsetownlst(&p->p_sigiolst); 397 398 /* 399 * Close open files and release open-file table. 400 * This may block! 401 */ 402 pdescfree(td); 403 fdescfree(td); 404 405 /* 406 * Remove ourself from our leader's peer list and wake our leader. 407 */ 408 if (p->p_leader->p_peers != NULL) { 409 mtx_lock(&ppeers_lock); 410 if (p->p_leader->p_peers != NULL) { 411 q = p->p_leader; 412 while (q->p_peers != p) 413 q = q->p_peers; 414 q->p_peers = p->p_peers; 415 wakeup(p->p_leader); 416 } 417 mtx_unlock(&ppeers_lock); 418 } 419 420 exec_free_abi_mappings(p); 421 vmspace_exit(td); 422 (void)acct_process(td); 423 424 #ifdef KTRACE 425 ktrprocexit(td); 426 #endif 427 /* 428 * Release reference to text vnode etc 429 */ 430 if (p->p_textvp != NULL) { 431 vrele(p->p_textvp); 432 p->p_textvp = NULL; 433 } 434 if (p->p_textdvp != NULL) { 435 vrele(p->p_textdvp); 436 p->p_textdvp = NULL; 437 } 438 if (p->p_binname != NULL) { 439 free(p->p_binname, M_PARGS); 440 p->p_binname = NULL; 441 } 442 443 /* 444 * Release our limits structure. 445 */ 446 lim_free(p->p_limit); 447 p->p_limit = NULL; 448 449 tidhash_remove(td); 450 451 /* 452 * Call machine-dependent code to release any 453 * machine-dependent resources other than the address space. 454 * The address space is released by "vmspace_exitfree(p)" in 455 * vm_waitproc(). 456 */ 457 cpu_exit(td); 458 459 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 460 461 /* 462 * Remove from allproc. It still sits in the hash. 463 */ 464 sx_xlock(&allproc_lock); 465 LIST_REMOVE(p, p_list); 466 467 #ifdef DDB 468 /* 469 * Used by ddb's 'ps' command to find this process via the 470 * pidhash. 471 */ 472 p->p_list.le_prev = NULL; 473 #endif 474 prison_proc_unlink(p->p_ucred->cr_prison, p); 475 sx_xunlock(&allproc_lock); 476 477 sx_xlock(&proctree_lock); 478 if ((p->p_flag & (P_TRACED | P_PPWAIT | P_PPTRACE)) != 0) { 479 PROC_LOCK(p); 480 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 481 PROC_UNLOCK(p); 482 } 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 processes 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_cat(sb, "(null)"); 807 } 808 if (nargs > 0) { 809 sbuf_putc(sb, '('); 810 for (i = 0;i < nargs; i++) 811 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 812 sbuf_putc(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_cat(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 PROC_LOCK(p); 830 sigexit(td, sig); 831 /* NOTREACHED */ 832 } 833 834 #ifdef COMPAT_43 835 /* 836 * The dirty work is handled by kern_wait(). 837 */ 838 int 839 owait(struct thread *td, struct owait_args *uap __unused) 840 { 841 int error, status; 842 843 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 844 if (error == 0) 845 td->td_retval[1] = status; 846 return (error); 847 } 848 #endif /* COMPAT_43 */ 849 850 /* 851 * The dirty work is handled by kern_wait(). 852 */ 853 int 854 sys_wait4(struct thread *td, struct wait4_args *uap) 855 { 856 struct rusage ru, *rup; 857 int error, status; 858 859 if (uap->rusage != NULL) 860 rup = &ru; 861 else 862 rup = NULL; 863 error = kern_wait(td, uap->pid, &status, uap->options, rup); 864 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 865 error = copyout(&status, uap->status, sizeof(status)); 866 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 867 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 868 return (error); 869 } 870 871 int 872 sys_wait6(struct thread *td, struct wait6_args *uap) 873 { 874 struct __wrusage wru, *wrup; 875 siginfo_t si, *sip; 876 idtype_t idtype; 877 id_t id; 878 int error, status; 879 880 idtype = uap->idtype; 881 id = uap->id; 882 883 if (uap->wrusage != NULL) 884 wrup = &wru; 885 else 886 wrup = NULL; 887 888 if (uap->info != NULL) { 889 sip = &si; 890 bzero(sip, sizeof(*sip)); 891 } else 892 sip = NULL; 893 894 /* 895 * We expect all callers of wait6() to know about WEXITED and 896 * WTRAPPED. 897 */ 898 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 899 900 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 901 error = copyout(&status, uap->status, sizeof(status)); 902 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 903 error = copyout(&wru, uap->wrusage, sizeof(wru)); 904 if (uap->info != NULL && error == 0) 905 error = copyout(&si, uap->info, sizeof(si)); 906 return (error); 907 } 908 909 /* 910 * Reap the remains of a zombie process and optionally return status and 911 * rusage. Asserts and will release both the proctree_lock and the process 912 * lock as part of its work. 913 */ 914 void 915 proc_reap(struct thread *td, struct proc *p, int *status, int options) 916 { 917 struct proc *q, *t; 918 919 sx_assert(&proctree_lock, SA_XLOCKED); 920 PROC_LOCK_ASSERT(p, MA_OWNED); 921 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 922 923 mtx_spin_wait_unlocked(&p->p_slock); 924 925 q = td->td_proc; 926 927 if (status) 928 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 929 if (options & WNOWAIT) { 930 /* 931 * Only poll, returning the status. Caller does not wish to 932 * release the proc struct just yet. 933 */ 934 PROC_UNLOCK(p); 935 sx_xunlock(&proctree_lock); 936 return; 937 } 938 939 PROC_LOCK(q); 940 sigqueue_take(p->p_ksi); 941 PROC_UNLOCK(q); 942 943 /* 944 * If we got the child via a ptrace 'attach', we need to give it back 945 * to the old parent. 946 */ 947 if (p->p_oppid != p->p_pptr->p_pid) { 948 PROC_UNLOCK(p); 949 t = proc_realparent(p); 950 PROC_LOCK(t); 951 PROC_LOCK(p); 952 CTR2(KTR_PTRACE, 953 "wait: traced child %d moved back to parent %d", p->p_pid, 954 t->p_pid); 955 proc_reparent(p, t, false); 956 PROC_UNLOCK(p); 957 pksignal(t, SIGCHLD, p->p_ksi); 958 wakeup(t); 959 cv_broadcast(&p->p_pwait); 960 PROC_UNLOCK(t); 961 sx_xunlock(&proctree_lock); 962 return; 963 } 964 PROC_UNLOCK(p); 965 966 /* 967 * Remove other references to this process to ensure we have an 968 * exclusive reference. 969 */ 970 sx_xlock(PIDHASHLOCK(p->p_pid)); 971 LIST_REMOVE(p, p_hash); 972 sx_xunlock(PIDHASHLOCK(p->p_pid)); 973 LIST_REMOVE(p, p_sibling); 974 reaper_abandon_children(p, true); 975 reaper_clear(p); 976 PROC_LOCK(p); 977 proc_clear_orphan(p); 978 PROC_UNLOCK(p); 979 leavepgrp(p); 980 if (p->p_procdesc != NULL) 981 procdesc_reap(p); 982 sx_xunlock(&proctree_lock); 983 984 proc_id_clear(PROC_ID_PID, p->p_pid); 985 986 PROC_LOCK(p); 987 knlist_detach(p->p_klist); 988 p->p_klist = NULL; 989 PROC_UNLOCK(p); 990 991 /* 992 * Removal from allproc list and process group list paired with 993 * PROC_LOCK which was executed during that time should guarantee 994 * nothing can reach this process anymore. As such further locking 995 * is unnecessary. 996 */ 997 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 998 999 PROC_LOCK(q); 1000 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 1001 PROC_UNLOCK(q); 1002 1003 /* 1004 * Destroy resource accounting information associated with the process. 1005 */ 1006 #ifdef RACCT 1007 if (racct_enable) { 1008 PROC_LOCK(p); 1009 racct_sub(p, RACCT_NPROC, 1); 1010 PROC_UNLOCK(p); 1011 } 1012 #endif 1013 racct_proc_exit(p); 1014 1015 /* 1016 * Free credentials, arguments, and sigacts, and decrement the count of 1017 * processes running with this uid. 1018 */ 1019 proc_unset_cred(p, true); 1020 pargs_drop(p->p_args); 1021 p->p_args = NULL; 1022 sigacts_free(p->p_sigacts); 1023 p->p_sigacts = NULL; 1024 1025 /* 1026 * Do any thread-system specific cleanups. 1027 */ 1028 thread_wait(p); 1029 1030 /* 1031 * Give vm and machine-dependent layer a chance to free anything that 1032 * cpu_exit couldn't release while still running in process context. 1033 */ 1034 vm_waitproc(p); 1035 #ifdef MAC 1036 mac_proc_destroy(p); 1037 #endif 1038 1039 KASSERT(FIRST_THREAD_IN_PROC(p), 1040 ("proc_reap: no residual thread!")); 1041 uma_zfree(proc_zone, p); 1042 atomic_add_int(&nprocs, -1); 1043 } 1044 1045 static int 1046 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 1047 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 1048 int check_only) 1049 { 1050 struct rusage *rup; 1051 1052 sx_assert(&proctree_lock, SA_XLOCKED); 1053 1054 PROC_LOCK(p); 1055 1056 switch (idtype) { 1057 case P_ALL: 1058 if (p->p_procdesc == NULL || 1059 (p->p_pptr == td->td_proc && 1060 (p->p_flag & P_TRACED) != 0)) { 1061 break; 1062 } 1063 1064 PROC_UNLOCK(p); 1065 return (0); 1066 case P_PID: 1067 if (p->p_pid != (pid_t)id) { 1068 PROC_UNLOCK(p); 1069 return (0); 1070 } 1071 break; 1072 case P_PGID: 1073 if (p->p_pgid != (pid_t)id) { 1074 PROC_UNLOCK(p); 1075 return (0); 1076 } 1077 break; 1078 case P_SID: 1079 if (p->p_session->s_sid != (pid_t)id) { 1080 PROC_UNLOCK(p); 1081 return (0); 1082 } 1083 break; 1084 case P_UID: 1085 if (p->p_ucred->cr_uid != (uid_t)id) { 1086 PROC_UNLOCK(p); 1087 return (0); 1088 } 1089 break; 1090 case P_GID: 1091 if (p->p_ucred->cr_gid != (gid_t)id) { 1092 PROC_UNLOCK(p); 1093 return (0); 1094 } 1095 break; 1096 case P_JAILID: 1097 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1098 PROC_UNLOCK(p); 1099 return (0); 1100 } 1101 break; 1102 /* 1103 * It seems that the thread structures get zeroed out 1104 * at process exit. This makes it impossible to 1105 * support P_SETID, P_CID or P_CPUID. 1106 */ 1107 default: 1108 PROC_UNLOCK(p); 1109 return (0); 1110 } 1111 1112 if (p_canwait(td, p)) { 1113 PROC_UNLOCK(p); 1114 return (0); 1115 } 1116 1117 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1118 PROC_UNLOCK(p); 1119 return (0); 1120 } 1121 1122 /* 1123 * This special case handles a kthread spawned by linux_clone 1124 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1125 * functions need to be able to distinguish between waiting 1126 * on a process and waiting on a thread. It is a thread if 1127 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1128 * signifies we want to wait for threads and not processes. 1129 */ 1130 if ((p->p_sigparent != SIGCHLD) ^ 1131 ((options & WLINUXCLONE) != 0)) { 1132 PROC_UNLOCK(p); 1133 return (0); 1134 } 1135 1136 if (siginfo != NULL) { 1137 bzero(siginfo, sizeof(*siginfo)); 1138 siginfo->si_errno = 0; 1139 1140 /* 1141 * SUSv4 requires that the si_signo value is always 1142 * SIGCHLD. Obey it despite the rfork(2) interface 1143 * allows to request other signal for child exit 1144 * notification. 1145 */ 1146 siginfo->si_signo = SIGCHLD; 1147 1148 /* 1149 * This is still a rough estimate. We will fix the 1150 * cases TRAPPED, STOPPED, and CONTINUED later. 1151 */ 1152 if (WCOREDUMP(p->p_xsig)) { 1153 siginfo->si_code = CLD_DUMPED; 1154 siginfo->si_status = WTERMSIG(p->p_xsig); 1155 } else if (WIFSIGNALED(p->p_xsig)) { 1156 siginfo->si_code = CLD_KILLED; 1157 siginfo->si_status = WTERMSIG(p->p_xsig); 1158 } else { 1159 siginfo->si_code = CLD_EXITED; 1160 siginfo->si_status = p->p_xexit; 1161 } 1162 1163 siginfo->si_pid = p->p_pid; 1164 siginfo->si_uid = p->p_ucred->cr_uid; 1165 1166 /* 1167 * The si_addr field would be useful additional 1168 * detail, but apparently the PC value may be lost 1169 * when we reach this point. bzero() above sets 1170 * siginfo->si_addr to NULL. 1171 */ 1172 } 1173 1174 /* 1175 * There should be no reason to limit resources usage info to 1176 * exited processes only. A snapshot about any resources used 1177 * by a stopped process may be exactly what is needed. 1178 */ 1179 if (wrusage != NULL) { 1180 rup = &wrusage->wru_self; 1181 *rup = p->p_ru; 1182 PROC_STATLOCK(p); 1183 calcru(p, &rup->ru_utime, &rup->ru_stime); 1184 PROC_STATUNLOCK(p); 1185 1186 rup = &wrusage->wru_children; 1187 *rup = p->p_stats->p_cru; 1188 calccru(p, &rup->ru_utime, &rup->ru_stime); 1189 } 1190 1191 if (p->p_state == PRS_ZOMBIE && !check_only) { 1192 proc_reap(td, p, status, options); 1193 return (-1); 1194 } 1195 return (1); 1196 } 1197 1198 int 1199 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1200 struct rusage *rusage) 1201 { 1202 struct __wrusage wru, *wrup; 1203 idtype_t idtype; 1204 id_t id; 1205 int ret; 1206 1207 /* 1208 * Translate the special pid values into the (idtype, pid) 1209 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1210 * kern_wait6() on its own. 1211 */ 1212 if (pid == WAIT_ANY) { 1213 idtype = P_ALL; 1214 id = 0; 1215 } else if (pid < 0) { 1216 idtype = P_PGID; 1217 id = (id_t)-pid; 1218 } else { 1219 idtype = P_PID; 1220 id = (id_t)pid; 1221 } 1222 1223 if (rusage != NULL) 1224 wrup = &wru; 1225 else 1226 wrup = NULL; 1227 1228 /* 1229 * For backward compatibility we implicitly add flags WEXITED 1230 * and WTRAPPED here. 1231 */ 1232 options |= WEXITED | WTRAPPED; 1233 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1234 if (rusage != NULL) 1235 *rusage = wru.wru_self; 1236 return (ret); 1237 } 1238 1239 static void 1240 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1241 int *status, int options, int si_code) 1242 { 1243 bool cont; 1244 1245 PROC_LOCK_ASSERT(p, MA_OWNED); 1246 sx_assert(&proctree_lock, SA_XLOCKED); 1247 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1248 si_code == CLD_CONTINUED); 1249 1250 cont = si_code == CLD_CONTINUED; 1251 if ((options & WNOWAIT) == 0) { 1252 if (cont) 1253 p->p_flag &= ~P_CONTINUED; 1254 else 1255 p->p_flag |= P_WAITED; 1256 if (kern_wait_dequeue_sigchld && 1257 (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) { 1258 PROC_LOCK(td->td_proc); 1259 sigqueue_take(p->p_ksi); 1260 PROC_UNLOCK(td->td_proc); 1261 } 1262 } 1263 sx_xunlock(&proctree_lock); 1264 if (siginfo != NULL) { 1265 siginfo->si_code = si_code; 1266 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1267 } 1268 if (status != NULL) 1269 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1270 td->td_retval[0] = p->p_pid; 1271 PROC_UNLOCK(p); 1272 } 1273 1274 int 1275 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1276 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1277 { 1278 struct proc *p, *q; 1279 pid_t pid; 1280 int error, nfound, ret; 1281 bool report; 1282 1283 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1284 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1285 AUDIT_ARG_VALUE(options); 1286 1287 q = td->td_proc; 1288 1289 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1290 PROC_LOCK(q); 1291 id = (id_t)q->p_pgid; 1292 PROC_UNLOCK(q); 1293 idtype = P_PGID; 1294 } 1295 1296 /* If we don't know the option, just return. */ 1297 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1298 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1299 return (EINVAL); 1300 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1301 /* 1302 * We will be unable to find any matching processes, 1303 * because there are no known events to look for. 1304 * Prefer to return error instead of blocking 1305 * indefinitely. 1306 */ 1307 return (EINVAL); 1308 } 1309 1310 loop: 1311 if (q->p_flag & P_STATCHILD) { 1312 PROC_LOCK(q); 1313 q->p_flag &= ~P_STATCHILD; 1314 PROC_UNLOCK(q); 1315 } 1316 sx_xlock(&proctree_lock); 1317 loop_locked: 1318 nfound = 0; 1319 LIST_FOREACH(p, &q->p_children, p_sibling) { 1320 pid = p->p_pid; 1321 ret = proc_to_reap(td, p, idtype, id, status, options, 1322 wrusage, siginfo, 0); 1323 if (ret == 0) 1324 continue; 1325 else if (ret != 1) { 1326 td->td_retval[0] = pid; 1327 return (0); 1328 } 1329 1330 /* 1331 * When running in capsicum(4) mode, make wait(2) ignore 1332 * processes created with pdfork(2). This is because one can 1333 * disown them - by passing their process descriptor to another 1334 * process - which means it needs to be prevented from touching 1335 * them afterwards. 1336 */ 1337 if (IN_CAPABILITY_MODE(td) && p->p_procdesc != NULL) { 1338 PROC_UNLOCK(p); 1339 continue; 1340 } 1341 1342 nfound++; 1343 PROC_LOCK_ASSERT(p, MA_OWNED); 1344 1345 if ((options & WTRAPPED) != 0 && 1346 (p->p_flag & P_TRACED) != 0) { 1347 PROC_SLOCK(p); 1348 report = 1349 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) && 1350 p->p_suspcount == p->p_numthreads && 1351 (p->p_flag & P_WAITED) == 0); 1352 PROC_SUNLOCK(p); 1353 if (report) { 1354 CTR4(KTR_PTRACE, 1355 "wait: returning trapped pid %d status %#x " 1356 "(xstat %d) xthread %d", 1357 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1358 p->p_xthread != NULL ? 1359 p->p_xthread->td_tid : -1); 1360 report_alive_proc(td, p, siginfo, status, 1361 options, CLD_TRAPPED); 1362 return (0); 1363 } 1364 } 1365 if ((options & WUNTRACED) != 0 && 1366 (p->p_flag & P_STOPPED_SIG) != 0) { 1367 PROC_SLOCK(p); 1368 report = (p->p_suspcount == p->p_numthreads && 1369 ((p->p_flag & P_WAITED) == 0)); 1370 PROC_SUNLOCK(p); 1371 if (report) { 1372 report_alive_proc(td, p, siginfo, status, 1373 options, CLD_STOPPED); 1374 return (0); 1375 } 1376 } 1377 if ((options & WCONTINUED) != 0 && 1378 (p->p_flag & P_CONTINUED) != 0) { 1379 report_alive_proc(td, p, siginfo, status, options, 1380 CLD_CONTINUED); 1381 return (0); 1382 } 1383 PROC_UNLOCK(p); 1384 } 1385 1386 /* 1387 * Look in the orphans list too, to allow the parent to 1388 * collect it's child exit status even if child is being 1389 * debugged. 1390 * 1391 * Debugger detaches from the parent upon successful 1392 * switch-over from parent to child. At this point due to 1393 * re-parenting the parent loses the child to debugger and a 1394 * wait4(2) call would report that it has no children to wait 1395 * for. By maintaining a list of orphans we allow the parent 1396 * to successfully wait until the child becomes a zombie. 1397 */ 1398 if (nfound == 0) { 1399 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1400 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1401 NULL, NULL, 1); 1402 if (ret != 0) { 1403 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1404 (int)td->td_retval[0])); 1405 PROC_UNLOCK(p); 1406 nfound++; 1407 break; 1408 } 1409 } 1410 } 1411 if (nfound == 0) { 1412 sx_xunlock(&proctree_lock); 1413 return (ECHILD); 1414 } 1415 if (options & WNOHANG) { 1416 sx_xunlock(&proctree_lock); 1417 td->td_retval[0] = 0; 1418 return (0); 1419 } 1420 PROC_LOCK(q); 1421 if (q->p_flag & P_STATCHILD) { 1422 q->p_flag &= ~P_STATCHILD; 1423 PROC_UNLOCK(q); 1424 goto loop_locked; 1425 } 1426 sx_xunlock(&proctree_lock); 1427 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0); 1428 if (error) 1429 return (error); 1430 goto loop; 1431 } 1432 1433 void 1434 proc_add_orphan(struct proc *child, struct proc *parent) 1435 { 1436 1437 sx_assert(&proctree_lock, SX_XLOCKED); 1438 KASSERT((child->p_flag & P_TRACED) != 0, 1439 ("proc_add_orphan: not traced")); 1440 1441 if (LIST_EMPTY(&parent->p_orphans)) { 1442 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1443 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan); 1444 } else { 1445 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans), 1446 child, p_orphan); 1447 } 1448 child->p_treeflag |= P_TREE_ORPHANED; 1449 } 1450 1451 /* 1452 * Make process 'parent' the new parent of process 'child'. 1453 * Must be called with an exclusive hold of proctree lock. 1454 */ 1455 void 1456 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid) 1457 { 1458 1459 sx_assert(&proctree_lock, SX_XLOCKED); 1460 PROC_LOCK_ASSERT(child, MA_OWNED); 1461 if (child->p_pptr == parent) 1462 return; 1463 1464 PROC_LOCK(child->p_pptr); 1465 sigqueue_take(child->p_ksi); 1466 PROC_UNLOCK(child->p_pptr); 1467 LIST_REMOVE(child, p_sibling); 1468 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1469 1470 proc_clear_orphan(child); 1471 if ((child->p_flag & P_TRACED) != 0) { 1472 proc_add_orphan(child, child->p_pptr); 1473 } 1474 1475 child->p_pptr = parent; 1476 if (set_oppid) 1477 child->p_oppid = parent->p_pid; 1478 } 1479