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_LOCKED); 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 void 209 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 210 { 211 212 exit1(td, uap->rval, 0); 213 /* NOTREACHED */ 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_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) { 373 timevalclear(&p->p_realtimer.it_interval); 374 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 375 KASSERT(!timevalisset(&p->p_realtimer.it_value), 376 ("realtime timer is still armed")); 377 } 378 379 PROC_UNLOCK(p); 380 381 if (p->p_sysent->sv_onexit != NULL) 382 p->p_sysent->sv_onexit(p); 383 seltdfini(td); 384 385 /* 386 * Reset any sigio structures pointing to us as a result of 387 * F_SETOWN with our pid. The P_WEXIT flag interlocks with fsetown(). 388 */ 389 funsetownlst(&p->p_sigiolst); 390 391 /* 392 * Close open files and release open-file table. 393 * This may block! 394 */ 395 pdescfree(td); 396 fdescfree(td); 397 398 /* 399 * If this thread tickled GEOM, we need to wait for the giggling to 400 * stop before we return to userland 401 */ 402 if (td->td_pflags & TDP_GEOM) 403 g_waitidle(); 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 sx_xunlock(&allproc_lock); 475 476 sx_xlock(&proctree_lock); 477 PROC_LOCK(p); 478 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 479 PROC_UNLOCK(p); 480 481 /* 482 * killjobc() might drop and re-acquire proctree_lock to 483 * revoke control tty if exiting process was a session leader. 484 */ 485 killjobc(); 486 487 /* 488 * Reparent all children processes: 489 * - traced ones to the original parent (or init if we are that parent) 490 * - the rest to init 491 */ 492 q = LIST_FIRST(&p->p_children); 493 if (q != NULL) /* only need this if any child is S_ZOMB */ 494 wakeup(q->p_reaper); 495 for (; q != NULL; q = nq) { 496 nq = LIST_NEXT(q, p_sibling); 497 ksi = ksiginfo_alloc(TRUE); 498 PROC_LOCK(q); 499 q->p_sigparent = SIGCHLD; 500 501 if ((q->p_flag & P_TRACED) == 0) { 502 proc_reparent(q, q->p_reaper, true); 503 if (q->p_state == PRS_ZOMBIE) { 504 /* 505 * Inform reaper about the reparented 506 * zombie, since wait(2) has something 507 * new to report. Guarantee queueing 508 * of the SIGCHLD signal, similar to 509 * the _exit() behaviour, by providing 510 * our ksiginfo. Ksi is freed by the 511 * signal delivery. 512 */ 513 if (q->p_ksi == NULL) { 514 ksi1 = NULL; 515 } else { 516 ksiginfo_copy(q->p_ksi, ksi); 517 ksi->ksi_flags |= KSI_INS; 518 ksi1 = ksi; 519 ksi = NULL; 520 } 521 PROC_LOCK(q->p_reaper); 522 pksignal(q->p_reaper, SIGCHLD, ksi1); 523 PROC_UNLOCK(q->p_reaper); 524 } else if (q->p_pdeathsig > 0) { 525 /* 526 * The child asked to received a signal 527 * when we exit. 528 */ 529 kern_psignal(q, q->p_pdeathsig); 530 } 531 } else { 532 /* 533 * Traced processes are killed by default 534 * since their existence means someone is 535 * screwing up. 536 */ 537 t = proc_realparent(q); 538 if (t == p) { 539 proc_reparent(q, q->p_reaper, true); 540 } else { 541 PROC_LOCK(t); 542 proc_reparent(q, t, true); 543 PROC_UNLOCK(t); 544 } 545 /* 546 * Since q was found on our children list, the 547 * proc_reparent() call moved q to the orphan 548 * list due to present P_TRACED flag. Clear 549 * orphan link for q now while q is locked. 550 */ 551 proc_clear_orphan(q); 552 q->p_flag &= ~P_TRACED; 553 q->p_flag2 &= ~P2_PTRACE_FSTP; 554 q->p_ptevents = 0; 555 p->p_xthread = NULL; 556 FOREACH_THREAD_IN_PROC(q, tdt) { 557 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 558 TDB_FSTP); 559 tdt->td_xsig = 0; 560 } 561 if (kern_kill_on_dbg_exit) { 562 q->p_flag &= ~P_STOPPED_TRACE; 563 kern_psignal(q, SIGKILL); 564 } else if ((q->p_flag & (P_STOPPED_TRACE | 565 P_STOPPED_SIG)) != 0) { 566 sigqueue_delete_proc(q, SIGTRAP); 567 ptrace_unsuspend(q); 568 } 569 } 570 PROC_UNLOCK(q); 571 if (ksi != NULL) 572 ksiginfo_free(ksi); 573 } 574 575 /* 576 * Also get rid of our orphans. 577 */ 578 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 579 PROC_LOCK(q); 580 KASSERT(q->p_oppid == p->p_pid, 581 ("orphan %p of %p has unexpected oppid %d", q, p, 582 q->p_oppid)); 583 q->p_oppid = q->p_reaper->p_pid; 584 585 /* 586 * If we are the real parent of this process 587 * but it has been reparented to a debugger, then 588 * check if it asked for a signal when we exit. 589 */ 590 if (q->p_pdeathsig > 0) 591 kern_psignal(q, q->p_pdeathsig); 592 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 593 q->p_pid); 594 proc_clear_orphan(q); 595 PROC_UNLOCK(q); 596 } 597 598 #ifdef KDTRACE_HOOKS 599 if (SDT_PROBES_ENABLED()) { 600 int reason = CLD_EXITED; 601 if (WCOREDUMP(signo)) 602 reason = CLD_DUMPED; 603 else if (WIFSIGNALED(signo)) 604 reason = CLD_KILLED; 605 SDT_PROBE1(proc, , , exit, reason); 606 } 607 #endif 608 609 /* Save exit status. */ 610 PROC_LOCK(p); 611 p->p_xthread = td; 612 613 if (p->p_sysent->sv_ontdexit != NULL) 614 p->p_sysent->sv_ontdexit(td); 615 616 #ifdef KDTRACE_HOOKS 617 /* 618 * Tell the DTrace fasttrap provider about the exit if it 619 * has declared an interest. 620 */ 621 if (dtrace_fasttrap_exit) 622 dtrace_fasttrap_exit(p); 623 #endif 624 625 /* 626 * Notify interested parties of our demise. 627 */ 628 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 629 630 /* 631 * If this is a process with a descriptor, we may not need to deliver 632 * a signal to the parent. proctree_lock is held over 633 * procdesc_exit() to serialize concurrent calls to close() and 634 * exit(). 635 */ 636 signal_parent = 0; 637 if (p->p_procdesc == NULL || procdesc_exit(p)) { 638 /* 639 * Notify parent that we're gone. If parent has the 640 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 641 * notify process 1 instead (and hope it will handle this 642 * situation). 643 */ 644 PROC_LOCK(p->p_pptr); 645 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 646 if (p->p_pptr->p_sigacts->ps_flag & 647 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 648 struct proc *pp; 649 650 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 651 pp = p->p_pptr; 652 PROC_UNLOCK(pp); 653 proc_reparent(p, p->p_reaper, true); 654 p->p_sigparent = SIGCHLD; 655 PROC_LOCK(p->p_pptr); 656 657 /* 658 * Notify parent, so in case he was wait(2)ing or 659 * executing waitpid(2) with our pid, he will 660 * continue. 661 */ 662 wakeup(pp); 663 } else 664 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 665 666 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) { 667 signal_parent = 1; 668 } else if (p->p_sigparent != 0) { 669 if (p->p_sigparent == SIGCHLD) { 670 signal_parent = 1; 671 } else { /* LINUX thread */ 672 signal_parent = 2; 673 } 674 } 675 } else 676 PROC_LOCK(p->p_pptr); 677 sx_xunlock(&proctree_lock); 678 679 if (signal_parent == 1) { 680 childproc_exited(p); 681 } else if (signal_parent == 2) { 682 kern_psignal(p->p_pptr, p->p_sigparent); 683 } 684 685 /* Tell the prison that we are gone. */ 686 prison_proc_free(p->p_ucred->cr_prison); 687 688 /* 689 * The state PRS_ZOMBIE prevents other proesses from sending 690 * signal to the process, to avoid memory leak, we free memory 691 * for signal queue at the time when the state is set. 692 */ 693 sigqueue_flush(&p->p_sigqueue); 694 sigqueue_flush(&td->td_sigqueue); 695 696 /* 697 * We have to wait until after acquiring all locks before 698 * changing p_state. We need to avoid all possible context 699 * switches (including ones from blocking on a mutex) while 700 * marked as a zombie. We also have to set the zombie state 701 * before we release the parent process' proc lock to avoid 702 * a lost wakeup. So, we first call wakeup, then we grab the 703 * sched lock, update the state, and release the parent process' 704 * proc lock. 705 */ 706 wakeup(p->p_pptr); 707 cv_broadcast(&p->p_pwait); 708 sched_exit(p->p_pptr, td); 709 PROC_SLOCK(p); 710 p->p_state = PRS_ZOMBIE; 711 PROC_UNLOCK(p->p_pptr); 712 713 /* 714 * Save our children's rusage information in our exit rusage. 715 */ 716 PROC_STATLOCK(p); 717 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 718 PROC_STATUNLOCK(p); 719 720 /* 721 * Make sure the scheduler takes this thread out of its tables etc. 722 * This will also release this thread's reference to the ucred. 723 * Other thread parts to release include pcb bits and such. 724 */ 725 thread_exit(); 726 } 727 728 #ifndef _SYS_SYSPROTO_H_ 729 struct abort2_args { 730 char *why; 731 int nargs; 732 void **args; 733 }; 734 #endif 735 736 int 737 sys_abort2(struct thread *td, struct abort2_args *uap) 738 { 739 struct proc *p = td->td_proc; 740 struct sbuf *sb; 741 void *uargs[16]; 742 int error, i, sig; 743 744 /* 745 * Do it right now so we can log either proper call of abort2(), or 746 * note, that invalid argument was passed. 512 is big enough to 747 * handle 16 arguments' descriptions with additional comments. 748 */ 749 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 750 sbuf_clear(sb); 751 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 752 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 753 /* 754 * Since we can't return from abort2(), send SIGKILL in cases, where 755 * abort2() was called improperly 756 */ 757 sig = SIGKILL; 758 /* Prevent from DoSes from user-space. */ 759 if (uap->nargs < 0 || uap->nargs > 16) 760 goto out; 761 if (uap->nargs > 0) { 762 if (uap->args == NULL) 763 goto out; 764 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 765 if (error != 0) 766 goto out; 767 } 768 /* 769 * Limit size of 'reason' string to 128. Will fit even when 770 * maximal number of arguments was chosen to be logged. 771 */ 772 if (uap->why != NULL) { 773 error = sbuf_copyin(sb, uap->why, 128); 774 if (error < 0) 775 goto out; 776 } else { 777 sbuf_printf(sb, "(null)"); 778 } 779 if (uap->nargs > 0) { 780 sbuf_printf(sb, "("); 781 for (i = 0;i < uap->nargs; i++) 782 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 783 sbuf_printf(sb, ")"); 784 } 785 /* 786 * Final stage: arguments were proper, string has been 787 * successfully copied from userspace, and copying pointers 788 * from user-space succeed. 789 */ 790 sig = SIGABRT; 791 out: 792 if (sig == SIGKILL) { 793 sbuf_trim(sb); 794 sbuf_printf(sb, " (Reason text inaccessible)"); 795 } 796 sbuf_cat(sb, "\n"); 797 sbuf_finish(sb); 798 log(LOG_INFO, "%s", sbuf_data(sb)); 799 sbuf_delete(sb); 800 exit1(td, 0, sig); 801 return (0); 802 } 803 804 #ifdef COMPAT_43 805 /* 806 * The dirty work is handled by kern_wait(). 807 */ 808 int 809 owait(struct thread *td, struct owait_args *uap __unused) 810 { 811 int error, status; 812 813 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 814 if (error == 0) 815 td->td_retval[1] = status; 816 return (error); 817 } 818 #endif /* COMPAT_43 */ 819 820 /* 821 * The dirty work is handled by kern_wait(). 822 */ 823 int 824 sys_wait4(struct thread *td, struct wait4_args *uap) 825 { 826 struct rusage ru, *rup; 827 int error, status; 828 829 if (uap->rusage != NULL) 830 rup = &ru; 831 else 832 rup = NULL; 833 error = kern_wait(td, uap->pid, &status, uap->options, rup); 834 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 835 error = copyout(&status, uap->status, sizeof(status)); 836 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 837 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 838 return (error); 839 } 840 841 int 842 sys_wait6(struct thread *td, struct wait6_args *uap) 843 { 844 struct __wrusage wru, *wrup; 845 siginfo_t si, *sip; 846 idtype_t idtype; 847 id_t id; 848 int error, status; 849 850 idtype = uap->idtype; 851 id = uap->id; 852 853 if (uap->wrusage != NULL) 854 wrup = &wru; 855 else 856 wrup = NULL; 857 858 if (uap->info != NULL) { 859 sip = &si; 860 bzero(sip, sizeof(*sip)); 861 } else 862 sip = NULL; 863 864 /* 865 * We expect all callers of wait6() to know about WEXITED and 866 * WTRAPPED. 867 */ 868 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 869 870 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 871 error = copyout(&status, uap->status, sizeof(status)); 872 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 873 error = copyout(&wru, uap->wrusage, sizeof(wru)); 874 if (uap->info != NULL && error == 0) 875 error = copyout(&si, uap->info, sizeof(si)); 876 return (error); 877 } 878 879 /* 880 * Reap the remains of a zombie process and optionally return status and 881 * rusage. Asserts and will release both the proctree_lock and the process 882 * lock as part of its work. 883 */ 884 void 885 proc_reap(struct thread *td, struct proc *p, int *status, int options) 886 { 887 struct proc *q, *t; 888 889 sx_assert(&proctree_lock, SA_XLOCKED); 890 PROC_LOCK_ASSERT(p, MA_OWNED); 891 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 892 893 mtx_spin_wait_unlocked(&p->p_slock); 894 895 q = td->td_proc; 896 897 if (status) 898 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 899 if (options & WNOWAIT) { 900 /* 901 * Only poll, returning the status. Caller does not wish to 902 * release the proc struct just yet. 903 */ 904 PROC_UNLOCK(p); 905 sx_xunlock(&proctree_lock); 906 return; 907 } 908 909 PROC_LOCK(q); 910 sigqueue_take(p->p_ksi); 911 PROC_UNLOCK(q); 912 913 /* 914 * If we got the child via a ptrace 'attach', we need to give it back 915 * to the old parent. 916 */ 917 if (p->p_oppid != p->p_pptr->p_pid) { 918 PROC_UNLOCK(p); 919 t = proc_realparent(p); 920 PROC_LOCK(t); 921 PROC_LOCK(p); 922 CTR2(KTR_PTRACE, 923 "wait: traced child %d moved back to parent %d", p->p_pid, 924 t->p_pid); 925 proc_reparent(p, t, false); 926 PROC_UNLOCK(p); 927 pksignal(t, SIGCHLD, p->p_ksi); 928 wakeup(t); 929 cv_broadcast(&p->p_pwait); 930 PROC_UNLOCK(t); 931 sx_xunlock(&proctree_lock); 932 return; 933 } 934 PROC_UNLOCK(p); 935 936 /* 937 * Remove other references to this process to ensure we have an 938 * exclusive reference. 939 */ 940 sx_xlock(PIDHASHLOCK(p->p_pid)); 941 LIST_REMOVE(p, p_hash); 942 sx_xunlock(PIDHASHLOCK(p->p_pid)); 943 LIST_REMOVE(p, p_sibling); 944 reaper_abandon_children(p, true); 945 reaper_clear(p); 946 PROC_LOCK(p); 947 proc_clear_orphan(p); 948 PROC_UNLOCK(p); 949 leavepgrp(p); 950 if (p->p_procdesc != NULL) 951 procdesc_reap(p); 952 sx_xunlock(&proctree_lock); 953 954 proc_id_clear(PROC_ID_PID, p->p_pid); 955 956 PROC_LOCK(p); 957 knlist_detach(p->p_klist); 958 p->p_klist = NULL; 959 PROC_UNLOCK(p); 960 961 /* 962 * Removal from allproc list and process group list paired with 963 * PROC_LOCK which was executed during that time should guarantee 964 * nothing can reach this process anymore. As such further locking 965 * is unnecessary. 966 */ 967 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 968 969 PROC_LOCK(q); 970 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 971 PROC_UNLOCK(q); 972 973 /* 974 * Decrement the count of procs running with this uid. 975 */ 976 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 977 978 /* 979 * Destroy resource accounting information associated with the process. 980 */ 981 #ifdef RACCT 982 if (racct_enable) { 983 PROC_LOCK(p); 984 racct_sub(p, RACCT_NPROC, 1); 985 PROC_UNLOCK(p); 986 } 987 #endif 988 racct_proc_exit(p); 989 990 /* 991 * Free credentials, arguments, and sigacts. 992 */ 993 proc_unset_cred(p); 994 pargs_drop(p->p_args); 995 p->p_args = NULL; 996 sigacts_free(p->p_sigacts); 997 p->p_sigacts = NULL; 998 999 /* 1000 * Do any thread-system specific cleanups. 1001 */ 1002 thread_wait(p); 1003 1004 /* 1005 * Give vm and machine-dependent layer a chance to free anything that 1006 * cpu_exit couldn't release while still running in process context. 1007 */ 1008 vm_waitproc(p); 1009 #ifdef MAC 1010 mac_proc_destroy(p); 1011 #endif 1012 1013 KASSERT(FIRST_THREAD_IN_PROC(p), 1014 ("proc_reap: no residual thread!")); 1015 uma_zfree(proc_zone, p); 1016 atomic_add_int(&nprocs, -1); 1017 } 1018 1019 static int 1020 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 1021 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 1022 int check_only) 1023 { 1024 struct rusage *rup; 1025 1026 sx_assert(&proctree_lock, SA_XLOCKED); 1027 1028 PROC_LOCK(p); 1029 1030 switch (idtype) { 1031 case P_ALL: 1032 if (p->p_procdesc == NULL || 1033 (p->p_pptr == td->td_proc && 1034 (p->p_flag & P_TRACED) != 0)) { 1035 break; 1036 } 1037 1038 PROC_UNLOCK(p); 1039 return (0); 1040 case P_PID: 1041 if (p->p_pid != (pid_t)id) { 1042 PROC_UNLOCK(p); 1043 return (0); 1044 } 1045 break; 1046 case P_PGID: 1047 if (p->p_pgid != (pid_t)id) { 1048 PROC_UNLOCK(p); 1049 return (0); 1050 } 1051 break; 1052 case P_SID: 1053 if (p->p_session->s_sid != (pid_t)id) { 1054 PROC_UNLOCK(p); 1055 return (0); 1056 } 1057 break; 1058 case P_UID: 1059 if (p->p_ucred->cr_uid != (uid_t)id) { 1060 PROC_UNLOCK(p); 1061 return (0); 1062 } 1063 break; 1064 case P_GID: 1065 if (p->p_ucred->cr_gid != (gid_t)id) { 1066 PROC_UNLOCK(p); 1067 return (0); 1068 } 1069 break; 1070 case P_JAILID: 1071 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1072 PROC_UNLOCK(p); 1073 return (0); 1074 } 1075 break; 1076 /* 1077 * It seems that the thread structures get zeroed out 1078 * at process exit. This makes it impossible to 1079 * support P_SETID, P_CID or P_CPUID. 1080 */ 1081 default: 1082 PROC_UNLOCK(p); 1083 return (0); 1084 } 1085 1086 if (p_canwait(td, p)) { 1087 PROC_UNLOCK(p); 1088 return (0); 1089 } 1090 1091 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1092 PROC_UNLOCK(p); 1093 return (0); 1094 } 1095 1096 /* 1097 * This special case handles a kthread spawned by linux_clone 1098 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1099 * functions need to be able to distinguish between waiting 1100 * on a process and waiting on a thread. It is a thread if 1101 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1102 * signifies we want to wait for threads and not processes. 1103 */ 1104 if ((p->p_sigparent != SIGCHLD) ^ 1105 ((options & WLINUXCLONE) != 0)) { 1106 PROC_UNLOCK(p); 1107 return (0); 1108 } 1109 1110 if (siginfo != NULL) { 1111 bzero(siginfo, sizeof(*siginfo)); 1112 siginfo->si_errno = 0; 1113 1114 /* 1115 * SUSv4 requires that the si_signo value is always 1116 * SIGCHLD. Obey it despite the rfork(2) interface 1117 * allows to request other signal for child exit 1118 * notification. 1119 */ 1120 siginfo->si_signo = SIGCHLD; 1121 1122 /* 1123 * This is still a rough estimate. We will fix the 1124 * cases TRAPPED, STOPPED, and CONTINUED later. 1125 */ 1126 if (WCOREDUMP(p->p_xsig)) { 1127 siginfo->si_code = CLD_DUMPED; 1128 siginfo->si_status = WTERMSIG(p->p_xsig); 1129 } else if (WIFSIGNALED(p->p_xsig)) { 1130 siginfo->si_code = CLD_KILLED; 1131 siginfo->si_status = WTERMSIG(p->p_xsig); 1132 } else { 1133 siginfo->si_code = CLD_EXITED; 1134 siginfo->si_status = p->p_xexit; 1135 } 1136 1137 siginfo->si_pid = p->p_pid; 1138 siginfo->si_uid = p->p_ucred->cr_uid; 1139 1140 /* 1141 * The si_addr field would be useful additional 1142 * detail, but apparently the PC value may be lost 1143 * when we reach this point. bzero() above sets 1144 * siginfo->si_addr to NULL. 1145 */ 1146 } 1147 1148 /* 1149 * There should be no reason to limit resources usage info to 1150 * exited processes only. A snapshot about any resources used 1151 * by a stopped process may be exactly what is needed. 1152 */ 1153 if (wrusage != NULL) { 1154 rup = &wrusage->wru_self; 1155 *rup = p->p_ru; 1156 PROC_STATLOCK(p); 1157 calcru(p, &rup->ru_utime, &rup->ru_stime); 1158 PROC_STATUNLOCK(p); 1159 1160 rup = &wrusage->wru_children; 1161 *rup = p->p_stats->p_cru; 1162 calccru(p, &rup->ru_utime, &rup->ru_stime); 1163 } 1164 1165 if (p->p_state == PRS_ZOMBIE && !check_only) { 1166 proc_reap(td, p, status, options); 1167 return (-1); 1168 } 1169 return (1); 1170 } 1171 1172 int 1173 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1174 struct rusage *rusage) 1175 { 1176 struct __wrusage wru, *wrup; 1177 idtype_t idtype; 1178 id_t id; 1179 int ret; 1180 1181 /* 1182 * Translate the special pid values into the (idtype, pid) 1183 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1184 * kern_wait6() on its own. 1185 */ 1186 if (pid == WAIT_ANY) { 1187 idtype = P_ALL; 1188 id = 0; 1189 } else if (pid < 0) { 1190 idtype = P_PGID; 1191 id = (id_t)-pid; 1192 } else { 1193 idtype = P_PID; 1194 id = (id_t)pid; 1195 } 1196 1197 if (rusage != NULL) 1198 wrup = &wru; 1199 else 1200 wrup = NULL; 1201 1202 /* 1203 * For backward compatibility we implicitly add flags WEXITED 1204 * and WTRAPPED here. 1205 */ 1206 options |= WEXITED | WTRAPPED; 1207 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1208 if (rusage != NULL) 1209 *rusage = wru.wru_self; 1210 return (ret); 1211 } 1212 1213 static void 1214 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1215 int *status, int options, int si_code) 1216 { 1217 bool cont; 1218 1219 PROC_LOCK_ASSERT(p, MA_OWNED); 1220 sx_assert(&proctree_lock, SA_XLOCKED); 1221 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1222 si_code == CLD_CONTINUED); 1223 1224 cont = si_code == CLD_CONTINUED; 1225 if ((options & WNOWAIT) == 0) { 1226 if (cont) 1227 p->p_flag &= ~P_CONTINUED; 1228 else 1229 p->p_flag |= P_WAITED; 1230 if (kern_wait_dequeue_sigchld && 1231 (td->td_proc->p_sysent->sv_flags & SV_SIG_WAITNDQ) == 0) { 1232 PROC_LOCK(td->td_proc); 1233 sigqueue_take(p->p_ksi); 1234 PROC_UNLOCK(td->td_proc); 1235 } 1236 } 1237 sx_xunlock(&proctree_lock); 1238 if (siginfo != NULL) { 1239 siginfo->si_code = si_code; 1240 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1241 } 1242 if (status != NULL) 1243 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1244 PROC_UNLOCK(p); 1245 td->td_retval[0] = p->p_pid; 1246 } 1247 1248 int 1249 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1250 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1251 { 1252 struct proc *p, *q; 1253 pid_t pid; 1254 int error, nfound, ret; 1255 bool report; 1256 1257 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1258 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1259 AUDIT_ARG_VALUE(options); 1260 1261 q = td->td_proc; 1262 1263 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1264 PROC_LOCK(q); 1265 id = (id_t)q->p_pgid; 1266 PROC_UNLOCK(q); 1267 idtype = P_PGID; 1268 } 1269 1270 /* If we don't know the option, just return. */ 1271 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1272 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1273 return (EINVAL); 1274 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1275 /* 1276 * We will be unable to find any matching processes, 1277 * because there are no known events to look for. 1278 * Prefer to return error instead of blocking 1279 * indefinitely. 1280 */ 1281 return (EINVAL); 1282 } 1283 1284 loop: 1285 if (q->p_flag & P_STATCHILD) { 1286 PROC_LOCK(q); 1287 q->p_flag &= ~P_STATCHILD; 1288 PROC_UNLOCK(q); 1289 } 1290 sx_xlock(&proctree_lock); 1291 loop_locked: 1292 nfound = 0; 1293 LIST_FOREACH(p, &q->p_children, p_sibling) { 1294 pid = p->p_pid; 1295 ret = proc_to_reap(td, p, idtype, id, status, options, 1296 wrusage, siginfo, 0); 1297 if (ret == 0) 1298 continue; 1299 else if (ret != 1) { 1300 td->td_retval[0] = pid; 1301 return (0); 1302 } 1303 1304 nfound++; 1305 PROC_LOCK_ASSERT(p, MA_OWNED); 1306 1307 if ((options & WTRAPPED) != 0 && 1308 (p->p_flag & P_TRACED) != 0) { 1309 PROC_SLOCK(p); 1310 report = 1311 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) && 1312 p->p_suspcount == p->p_numthreads && 1313 (p->p_flag & P_WAITED) == 0); 1314 PROC_SUNLOCK(p); 1315 if (report) { 1316 CTR4(KTR_PTRACE, 1317 "wait: returning trapped pid %d status %#x " 1318 "(xstat %d) xthread %d", 1319 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1320 p->p_xthread != NULL ? 1321 p->p_xthread->td_tid : -1); 1322 report_alive_proc(td, p, siginfo, status, 1323 options, CLD_TRAPPED); 1324 return (0); 1325 } 1326 } 1327 if ((options & WUNTRACED) != 0 && 1328 (p->p_flag & P_STOPPED_SIG) != 0) { 1329 PROC_SLOCK(p); 1330 report = (p->p_suspcount == p->p_numthreads && 1331 ((p->p_flag & P_WAITED) == 0)); 1332 PROC_SUNLOCK(p); 1333 if (report) { 1334 report_alive_proc(td, p, siginfo, status, 1335 options, CLD_STOPPED); 1336 return (0); 1337 } 1338 } 1339 if ((options & WCONTINUED) != 0 && 1340 (p->p_flag & P_CONTINUED) != 0) { 1341 report_alive_proc(td, p, siginfo, status, options, 1342 CLD_CONTINUED); 1343 return (0); 1344 } 1345 PROC_UNLOCK(p); 1346 } 1347 1348 /* 1349 * Look in the orphans list too, to allow the parent to 1350 * collect it's child exit status even if child is being 1351 * debugged. 1352 * 1353 * Debugger detaches from the parent upon successful 1354 * switch-over from parent to child. At this point due to 1355 * re-parenting the parent loses the child to debugger and a 1356 * wait4(2) call would report that it has no children to wait 1357 * for. By maintaining a list of orphans we allow the parent 1358 * to successfully wait until the child becomes a zombie. 1359 */ 1360 if (nfound == 0) { 1361 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1362 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1363 NULL, NULL, 1); 1364 if (ret != 0) { 1365 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1366 (int)td->td_retval[0])); 1367 PROC_UNLOCK(p); 1368 nfound++; 1369 break; 1370 } 1371 } 1372 } 1373 if (nfound == 0) { 1374 sx_xunlock(&proctree_lock); 1375 return (ECHILD); 1376 } 1377 if (options & WNOHANG) { 1378 sx_xunlock(&proctree_lock); 1379 td->td_retval[0] = 0; 1380 return (0); 1381 } 1382 PROC_LOCK(q); 1383 if (q->p_flag & P_STATCHILD) { 1384 q->p_flag &= ~P_STATCHILD; 1385 PROC_UNLOCK(q); 1386 goto loop_locked; 1387 } 1388 sx_xunlock(&proctree_lock); 1389 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0); 1390 if (error) 1391 return (error); 1392 goto loop; 1393 } 1394 1395 void 1396 proc_add_orphan(struct proc *child, struct proc *parent) 1397 { 1398 1399 sx_assert(&proctree_lock, SX_XLOCKED); 1400 KASSERT((child->p_flag & P_TRACED) != 0, 1401 ("proc_add_orphan: not traced")); 1402 1403 if (LIST_EMPTY(&parent->p_orphans)) { 1404 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1405 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan); 1406 } else { 1407 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans), 1408 child, p_orphan); 1409 } 1410 child->p_treeflag |= P_TREE_ORPHANED; 1411 } 1412 1413 /* 1414 * Make process 'parent' the new parent of process 'child'. 1415 * Must be called with an exclusive hold of proctree lock. 1416 */ 1417 void 1418 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid) 1419 { 1420 1421 sx_assert(&proctree_lock, SX_XLOCKED); 1422 PROC_LOCK_ASSERT(child, MA_OWNED); 1423 if (child->p_pptr == parent) 1424 return; 1425 1426 PROC_LOCK(child->p_pptr); 1427 sigqueue_take(child->p_ksi); 1428 PROC_UNLOCK(child->p_pptr); 1429 LIST_REMOVE(child, p_sibling); 1430 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1431 1432 proc_clear_orphan(child); 1433 if ((child->p_flag & P_TRACED) != 0) { 1434 proc_add_orphan(child, child->p_pptr); 1435 } 1436 1437 child->p_pptr = parent; 1438 if (set_oppid) 1439 child->p_oppid = parent->p_pid; 1440 } 1441