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