1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: stable/10/sys/kern/kern_exit.c 310585 2016-12-26 10:16:05Z kib $"); 39 40#include "opt_compat.h" 41#include "opt_kdtrace.h" 42#include "opt_ktrace.h" 43#include "opt_procdesc.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/malloc.h> 52#include <sys/lock.h> 53#include <sys/mutex.h> 54#include <sys/proc.h> 55#include <sys/procdesc.h> 56#include <sys/pioctl.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/syslog.h> 70#include <sys/ptrace.h> 71#include <sys/acct.h> /* for acct_process() function prototype */ 72#include <sys/filedesc.h> 73#include <sys/sdt.h> 74#include <sys/shm.h> 75#include <sys/sem.h> 76#include <sys/umtx.h> 77#ifdef KTRACE 78#include <sys/ktrace.h> 79#endif 80 81#include <security/audit/audit.h> 82#include <security/mac/mac_framework.h> 83 84#include <vm/vm.h> 85#include <vm/vm_extern.h> 86#include <vm/vm_param.h> 87#include <vm/pmap.h> 88#include <vm/vm_map.h> 89#include <vm/vm_page.h> 90#include <vm/uma.h> 91 92#ifdef KDTRACE_HOOKS 93#include <sys/dtrace_bsd.h> 94dtrace_execexit_func_t dtrace_fasttrap_exit; 95#endif 96 97SDT_PROVIDER_DECLARE(proc); 98SDT_PROBE_DEFINE1(proc, , , exit, "int"); 99 100/* Hook for NFS teardown procedure. */ 101void (*nlminfo_release_p)(struct proc *p); 102 103struct proc * 104proc_realparent(struct proc *child) 105{ 106 struct proc *p, *parent; 107 108 sx_assert(&proctree_lock, SX_LOCKED); 109 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 110 if (child->p_oppid == 0 || 111 child->p_pptr->p_pid == child->p_oppid) 112 parent = child->p_pptr; 113 else 114 parent = initproc; 115 return (parent); 116 } 117 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 118 /* Cannot use LIST_PREV(), since the list head is not known. */ 119 p = __containerof(p->p_orphan.le_prev, struct proc, 120 p_orphan.le_next); 121 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 122 ("missing P_ORPHAN %p", p)); 123 } 124 parent = __containerof(p->p_orphan.le_prev, struct proc, 125 p_orphans.lh_first); 126 return (parent); 127} 128 129void 130reaper_abandon_children(struct proc *p, bool exiting) 131{ 132 struct proc *p1, *p2, *ptmp; 133 134 sx_assert(&proctree_lock, SX_LOCKED); 135 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 136 if ((p->p_treeflag & P_TREE_REAPER) == 0) 137 return; 138 p1 = p->p_reaper; 139 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 140 LIST_REMOVE(p2, p_reapsibling); 141 p2->p_reaper = p1; 142 p2->p_reapsubtree = p->p_reapsubtree; 143 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 144 if (exiting && p2->p_pptr == p) { 145 PROC_LOCK(p2); 146 proc_reparent(p2, p1); 147 PROC_UNLOCK(p2); 148 } 149 } 150 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 151 p->p_treeflag &= ~P_TREE_REAPER; 152} 153 154static void 155clear_orphan(struct proc *p) 156{ 157 struct proc *p1; 158 159 sx_assert(&proctree_lock, SA_XLOCKED); 160 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 161 return; 162 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 163 p1 = LIST_NEXT(p, p_orphan); 164 if (p1 != NULL) 165 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 166 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 167 } 168 LIST_REMOVE(p, p_orphan); 169 p->p_treeflag &= ~P_TREE_ORPHANED; 170} 171 172/* 173 * exit -- death of process. 174 */ 175void 176sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 177{ 178 179 exit1(td, W_EXITCODE(uap->rval, 0)); 180 /* NOTREACHED */ 181} 182 183/* 184 * Exit: deallocate address space and other resources, change proc state to 185 * zombie, and unlink proc from allproc and parent's lists. Save exit status 186 * and rusage for wait(). Check for child processes and orphan them. 187 */ 188void 189exit1(struct thread *td, int rv) 190{ 191 struct proc *p, *nq, *q, *t; 192 struct thread *tdt; 193 struct vnode *ttyvp = NULL; 194 195 mtx_assert(&Giant, MA_NOTOWNED); 196 197 p = td->td_proc; 198 /* 199 * XXX in case we're rebooting we just let init die in order to 200 * work around an unsolved stack overflow seen very late during 201 * shutdown on sparc64 when the gmirror worker process exists. 202 */ 203 if (p == initproc && rebooting == 0) { 204 printf("init died (signal %d, exit %d)\n", 205 WTERMSIG(rv), WEXITSTATUS(rv)); 206 panic("Going nowhere without my init!"); 207 } 208 209 /* 210 * Deref SU mp, since the thread does not return to userspace. 211 */ 212 if (softdep_ast_cleanup != NULL) 213 softdep_ast_cleanup(); 214 215 /* 216 * MUST abort all other threads before proceeding past here. 217 */ 218 PROC_LOCK(p); 219 /* 220 * First check if some other thread or external request got 221 * here before us. If so, act appropriately: exit or suspend. 222 * We must ensure that stop requests are handled before we set 223 * P_WEXIT. 224 */ 225 thread_suspend_check(0); 226 while (p->p_flag & P_HADTHREADS) { 227 /* 228 * Kill off the other threads. This requires 229 * some co-operation from other parts of the kernel 230 * so it may not be instantaneous. With this state set 231 * any thread entering the kernel from userspace will 232 * thread_exit() in trap(). Any thread attempting to 233 * sleep will return immediately with EINTR or EWOULDBLOCK 234 * which will hopefully force them to back out to userland 235 * freeing resources as they go. Any thread attempting 236 * to return to userland will thread_exit() from userret(). 237 * thread_exit() will unsuspend us when the last of the 238 * other threads exits. 239 * If there is already a thread singler after resumption, 240 * calling thread_single will fail; in that case, we just 241 * re-check all suspension request, the thread should 242 * either be suspended there or exit. 243 */ 244 if (!thread_single(p, SINGLE_EXIT)) 245 /* 246 * All other activity in this process is now 247 * stopped. Threading support has been turned 248 * off. 249 */ 250 break; 251 /* 252 * Recheck for new stop or suspend requests which 253 * might appear while process lock was dropped in 254 * thread_single(). 255 */ 256 thread_suspend_check(0); 257 } 258 KASSERT(p->p_numthreads == 1, 259 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 260 racct_sub(p, RACCT_NTHR, 1); 261 /* 262 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 263 * on our vmspace, so we should block below until they have 264 * released their reference to us. Note that if they have 265 * requested S_EXIT stops we will block here until they ack 266 * via PIOCCONT. 267 */ 268 _STOPEVENT(p, S_EXIT, rv); 269 270 /* 271 * Ignore any pending request to stop due to a stop signal. 272 * Once P_WEXIT is set, future requests will be ignored as 273 * well. 274 */ 275 p->p_flag &= ~P_STOPPED_SIG; 276 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 277 278 /* 279 * Note that we are exiting and do another wakeup of anyone in 280 * PIOCWAIT in case they aren't listening for S_EXIT stops or 281 * decided to wait again after we told them we are exiting. 282 */ 283 p->p_flag |= P_WEXIT; 284 wakeup(&p->p_stype); 285 286 /* 287 * Wait for any processes that have a hold on our vmspace to 288 * release their reference. 289 */ 290 while (p->p_lock > 0) 291 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 292 293 p->p_xstat = rv; /* Let event handler change exit status */ 294 PROC_UNLOCK(p); 295 /* Drain the limit callout while we don't have the proc locked */ 296 callout_drain(&p->p_limco); 297 298#ifdef AUDIT 299 /* 300 * The Sun BSM exit token contains two components: an exit status as 301 * passed to exit(), and a return value to indicate what sort of exit 302 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 303 * what the return value is. 304 */ 305 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0); 306 AUDIT_SYSCALL_EXIT(0, td); 307#endif 308 309 /* Are we a task leader? */ 310 if (p == p->p_leader) { 311 mtx_lock(&ppeers_lock); 312 q = p->p_peers; 313 while (q != NULL) { 314 PROC_LOCK(q); 315 kern_psignal(q, SIGKILL); 316 PROC_UNLOCK(q); 317 q = q->p_peers; 318 } 319 while (p->p_peers != NULL) 320 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 321 mtx_unlock(&ppeers_lock); 322 } 323 324 /* 325 * Check if any loadable modules need anything done at process exit. 326 * E.g. SYSV IPC stuff 327 * XXX what if one of these generates an error? 328 */ 329 EVENTHANDLER_INVOKE(process_exit, p); 330 331 /* 332 * If parent is waiting for us to exit or exec, 333 * P_PPWAIT is set; we will wakeup the parent below. 334 */ 335 PROC_LOCK(p); 336 rv = p->p_xstat; /* Event handler could change exit status */ 337 stopprofclock(p); 338 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 339 p->p_ptevents = 0; 340 341 /* 342 * Stop the real interval timer. If the handler is currently 343 * executing, prevent it from rearming itself and let it finish. 344 */ 345 if (timevalisset(&p->p_realtimer.it_value) && 346 callout_stop(&p->p_itcallout) == 0) { 347 timevalclear(&p->p_realtimer.it_interval); 348 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 349 KASSERT(!timevalisset(&p->p_realtimer.it_value), 350 ("realtime timer is still armed")); 351 } 352 PROC_UNLOCK(p); 353 354 /* 355 * Reset any sigio structures pointing to us as a result of 356 * F_SETOWN with our pid. 357 */ 358 funsetownlst(&p->p_sigiolst); 359 360 /* 361 * If this process has an nlminfo data area (for lockd), release it 362 */ 363 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 364 (*nlminfo_release_p)(p); 365 366 /* 367 * Close open files and release open-file table. 368 * This may block! 369 */ 370 fdescfree(td); 371 372 /* 373 * If this thread tickled GEOM, we need to wait for the giggling to 374 * stop before we return to userland 375 */ 376 if (td->td_pflags & TDP_GEOM) 377 g_waitidle(); 378 379 /* 380 * Remove ourself from our leader's peer list and wake our leader. 381 */ 382 mtx_lock(&ppeers_lock); 383 if (p->p_leader->p_peers) { 384 q = p->p_leader; 385 while (q->p_peers != p) 386 q = q->p_peers; 387 q->p_peers = p->p_peers; 388 wakeup(p->p_leader); 389 } 390 mtx_unlock(&ppeers_lock); 391 392 vmspace_exit(td); 393 394 sx_xlock(&proctree_lock); 395 if (SESS_LEADER(p)) { 396 struct session *sp = p->p_session; 397 struct tty *tp; 398 399 /* 400 * s_ttyp is not zero'd; we use this to indicate that 401 * the session once had a controlling terminal. (for 402 * logging and informational purposes) 403 */ 404 SESS_LOCK(sp); 405 ttyvp = sp->s_ttyvp; 406 tp = sp->s_ttyp; 407 sp->s_ttyvp = NULL; 408 sp->s_ttydp = NULL; 409 sp->s_leader = NULL; 410 SESS_UNLOCK(sp); 411 412 /* 413 * Signal foreground pgrp and revoke access to 414 * controlling terminal if it has not been revoked 415 * already. 416 * 417 * Because the TTY may have been revoked in the mean 418 * time and could already have a new session associated 419 * with it, make sure we don't send a SIGHUP to a 420 * foreground process group that does not belong to this 421 * session. 422 */ 423 424 if (tp != NULL) { 425 tty_lock(tp); 426 if (tp->t_session == sp) 427 tty_signal_pgrp(tp, SIGHUP); 428 tty_unlock(tp); 429 } 430 431 if (ttyvp != NULL) { 432 sx_xunlock(&proctree_lock); 433 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 434 VOP_REVOKE(ttyvp, REVOKEALL); 435 VOP_UNLOCK(ttyvp, 0); 436 } 437 sx_xlock(&proctree_lock); 438 } 439 } 440 fixjobc(p, p->p_pgrp, 0); 441 sx_xunlock(&proctree_lock); 442 (void)acct_process(td); 443 444 /* Release the TTY now we've unlocked everything. */ 445 if (ttyvp != NULL) 446 vrele(ttyvp); 447#ifdef KTRACE 448 ktrprocexit(td); 449#endif 450 /* 451 * Release reference to text vnode 452 */ 453 if (p->p_textvp != NULL) { 454 vrele(p->p_textvp); 455 p->p_textvp = NULL; 456 } 457 458 /* 459 * Release our limits structure. 460 */ 461 lim_free(p->p_limit); 462 p->p_limit = NULL; 463 464 tidhash_remove(td); 465 466 /* 467 * Remove proc from allproc queue and pidhash chain. 468 * Place onto zombproc. Unlink from parent's child list. 469 */ 470 sx_xlock(&allproc_lock); 471 LIST_REMOVE(p, p_list); 472 LIST_INSERT_HEAD(&zombproc, p, p_list); 473 LIST_REMOVE(p, p_hash); 474 sx_xunlock(&allproc_lock); 475 476 /* 477 * Call machine-dependent code to release any 478 * machine-dependent resources other than the address space. 479 * The address space is released by "vmspace_exitfree(p)" in 480 * vm_waitproc(). 481 */ 482 cpu_exit(td); 483 484 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 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 sx_xlock(&proctree_lock); 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 PROC_LOCK(q); 498 q->p_sigparent = SIGCHLD; 499 500 if (!(q->p_flag & P_TRACED)) { 501 proc_reparent(q, q->p_reaper); 502 if (q->p_state == PRS_ZOMBIE) { 503 PROC_LOCK(q->p_reaper); 504 pksignal(q->p_reaper, SIGCHLD, q->p_ksi); 505 PROC_UNLOCK(q->p_reaper); 506 } 507 } else { 508 /* 509 * Traced processes are killed since their existence 510 * means someone is screwing up. 511 */ 512 t = proc_realparent(q); 513 if (t == p) { 514 proc_reparent(q, q->p_reaper); 515 } else { 516 PROC_LOCK(t); 517 proc_reparent(q, t); 518 PROC_UNLOCK(t); 519 } 520 /* 521 * Since q was found on our children list, the 522 * proc_reparent() call moved q to the orphan 523 * list due to present P_TRACED flag. Clear 524 * orphan link for q now while q is locked. 525 */ 526 clear_orphan(q); 527 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 528 q->p_flag2 &= ~P2_PTRACE_FSTP; 529 q->p_ptevents = 0; 530 FOREACH_THREAD_IN_PROC(q, tdt) { 531 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 532 TDB_FSTP); 533 } 534 kern_psignal(q, SIGKILL); 535 } 536 PROC_UNLOCK(q); 537 } 538 539 /* 540 * Also get rid of our orphans. 541 */ 542 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 543 PROC_LOCK(q); 544 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 545 q->p_pid); 546 clear_orphan(q); 547 PROC_UNLOCK(q); 548 } 549 550 /* Save exit status. */ 551 PROC_LOCK(p); 552 p->p_xthread = td; 553 554 /* Tell the prison that we are gone. */ 555 prison_proc_free(p->p_ucred->cr_prison); 556 557#ifdef KDTRACE_HOOKS 558 /* 559 * Tell the DTrace fasttrap provider about the exit if it 560 * has declared an interest. 561 */ 562 if (dtrace_fasttrap_exit) 563 dtrace_fasttrap_exit(p); 564#endif 565 566 /* 567 * Notify interested parties of our demise. 568 */ 569 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 570 571#ifdef KDTRACE_HOOKS 572 int reason = CLD_EXITED; 573 if (WCOREDUMP(rv)) 574 reason = CLD_DUMPED; 575 else if (WIFSIGNALED(rv)) 576 reason = CLD_KILLED; 577 SDT_PROBE1(proc, , , exit, reason); 578#endif 579 580 /* 581 * Just delete all entries in the p_klist. At this point we won't 582 * report any more events, and there are nasty race conditions that 583 * can beat us if we don't. 584 */ 585 knlist_clear(&p->p_klist, 1); 586 587 /* 588 * If this is a process with a descriptor, we may not need to deliver 589 * a signal to the parent. proctree_lock is held over 590 * procdesc_exit() to serialize concurrent calls to close() and 591 * exit(). 592 */ 593#ifdef PROCDESC 594 if (p->p_procdesc == NULL || procdesc_exit(p)) { 595#endif 596 /* 597 * Notify parent that we're gone. If parent has the 598 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 599 * notify process 1 instead (and hope it will handle this 600 * situation). 601 */ 602 PROC_LOCK(p->p_pptr); 603 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 604 if (p->p_pptr->p_sigacts->ps_flag & 605 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 606 struct proc *pp; 607 608 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 609 pp = p->p_pptr; 610 PROC_UNLOCK(pp); 611 proc_reparent(p, p->p_reaper); 612 p->p_sigparent = SIGCHLD; 613 PROC_LOCK(p->p_pptr); 614 615 /* 616 * Notify parent, so in case he was wait(2)ing or 617 * executing waitpid(2) with our pid, he will 618 * continue. 619 */ 620 wakeup(pp); 621 } else 622 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 623 624 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 625 childproc_exited(p); 626 else if (p->p_sigparent != 0) { 627 if (p->p_sigparent == SIGCHLD) 628 childproc_exited(p); 629 else /* LINUX thread */ 630 kern_psignal(p->p_pptr, p->p_sigparent); 631 } 632#ifdef PROCDESC 633 } else 634 PROC_LOCK(p->p_pptr); 635#endif 636 sx_xunlock(&proctree_lock); 637 638 /* 639 * The state PRS_ZOMBIE prevents other proesses from sending 640 * signal to the process, to avoid memory leak, we free memory 641 * for signal queue at the time when the state is set. 642 */ 643 sigqueue_flush(&p->p_sigqueue); 644 sigqueue_flush(&td->td_sigqueue); 645 646 /* 647 * We have to wait until after acquiring all locks before 648 * changing p_state. We need to avoid all possible context 649 * switches (including ones from blocking on a mutex) while 650 * marked as a zombie. We also have to set the zombie state 651 * before we release the parent process' proc lock to avoid 652 * a lost wakeup. So, we first call wakeup, then we grab the 653 * sched lock, update the state, and release the parent process' 654 * proc lock. 655 */ 656 wakeup(p->p_pptr); 657 cv_broadcast(&p->p_pwait); 658 sched_exit(p->p_pptr, td); 659 umtx_thread_exit(td); 660 PROC_SLOCK(p); 661 p->p_state = PRS_ZOMBIE; 662 PROC_UNLOCK(p->p_pptr); 663 664 /* 665 * Hopefully no one will try to deliver a signal to the process this 666 * late in the game. 667 */ 668 knlist_destroy(&p->p_klist); 669 670 /* 671 * Save our children's rusage information in our exit rusage. 672 */ 673 PROC_STATLOCK(p); 674 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 675 PROC_STATUNLOCK(p); 676 677 /* 678 * Make sure the scheduler takes this thread out of its tables etc. 679 * This will also release this thread's reference to the ucred. 680 * Other thread parts to release include pcb bits and such. 681 */ 682 thread_exit(); 683} 684 685 686#ifndef _SYS_SYSPROTO_H_ 687struct abort2_args { 688 char *why; 689 int nargs; 690 void **args; 691}; 692#endif 693 694int 695sys_abort2(struct thread *td, struct abort2_args *uap) 696{ 697 struct proc *p = td->td_proc; 698 struct sbuf *sb; 699 void *uargs[16]; 700 int error, i, sig; 701 702 /* 703 * Do it right now so we can log either proper call of abort2(), or 704 * note, that invalid argument was passed. 512 is big enough to 705 * handle 16 arguments' descriptions with additional comments. 706 */ 707 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 708 sbuf_clear(sb); 709 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 710 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 711 /* 712 * Since we can't return from abort2(), send SIGKILL in cases, where 713 * abort2() was called improperly 714 */ 715 sig = SIGKILL; 716 /* Prevent from DoSes from user-space. */ 717 if (uap->nargs < 0 || uap->nargs > 16) 718 goto out; 719 if (uap->nargs > 0) { 720 if (uap->args == NULL) 721 goto out; 722 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 723 if (error != 0) 724 goto out; 725 } 726 /* 727 * Limit size of 'reason' string to 128. Will fit even when 728 * maximal number of arguments was chosen to be logged. 729 */ 730 if (uap->why != NULL) { 731 error = sbuf_copyin(sb, uap->why, 128); 732 if (error < 0) 733 goto out; 734 } else { 735 sbuf_printf(sb, "(null)"); 736 } 737 if (uap->nargs > 0) { 738 sbuf_printf(sb, "("); 739 for (i = 0;i < uap->nargs; i++) 740 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 741 sbuf_printf(sb, ")"); 742 } 743 /* 744 * Final stage: arguments were proper, string has been 745 * successfully copied from userspace, and copying pointers 746 * from user-space succeed. 747 */ 748 sig = SIGABRT; 749out: 750 if (sig == SIGKILL) { 751 sbuf_trim(sb); 752 sbuf_printf(sb, " (Reason text inaccessible)"); 753 } 754 sbuf_cat(sb, "\n"); 755 sbuf_finish(sb); 756 log(LOG_INFO, "%s", sbuf_data(sb)); 757 sbuf_delete(sb); 758 exit1(td, W_EXITCODE(0, sig)); 759 return (0); 760} 761 762 763#ifdef COMPAT_43 764/* 765 * The dirty work is handled by kern_wait(). 766 */ 767int 768owait(struct thread *td, struct owait_args *uap __unused) 769{ 770 int error, status; 771 772 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 773 if (error == 0) 774 td->td_retval[1] = status; 775 return (error); 776} 777#endif /* COMPAT_43 */ 778 779/* 780 * The dirty work is handled by kern_wait(). 781 */ 782int 783sys_wait4(struct thread *td, struct wait4_args *uap) 784{ 785 struct rusage ru, *rup; 786 int error, status; 787 788 if (uap->rusage != NULL) 789 rup = &ru; 790 else 791 rup = NULL; 792 error = kern_wait(td, uap->pid, &status, uap->options, rup); 793 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 794 error = copyout(&status, uap->status, sizeof(status)); 795 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 796 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 797 return (error); 798} 799 800int 801sys_wait6(struct thread *td, struct wait6_args *uap) 802{ 803 struct __wrusage wru, *wrup; 804 siginfo_t si, *sip; 805 idtype_t idtype; 806 id_t id; 807 int error, status; 808 809 idtype = uap->idtype; 810 id = uap->id; 811 812 if (uap->wrusage != NULL) 813 wrup = &wru; 814 else 815 wrup = NULL; 816 817 if (uap->info != NULL) { 818 sip = &si; 819 bzero(sip, sizeof(*sip)); 820 } else 821 sip = NULL; 822 823 /* 824 * We expect all callers of wait6() to know about WEXITED and 825 * WTRAPPED. 826 */ 827 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 828 829 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 830 error = copyout(&status, uap->status, sizeof(status)); 831 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 832 error = copyout(&wru, uap->wrusage, sizeof(wru)); 833 if (uap->info != NULL && error == 0) 834 error = copyout(&si, uap->info, sizeof(si)); 835 return (error); 836} 837 838/* 839 * Reap the remains of a zombie process and optionally return status and 840 * rusage. Asserts and will release both the proctree_lock and the process 841 * lock as part of its work. 842 */ 843void 844proc_reap(struct thread *td, struct proc *p, int *status, int options) 845{ 846 struct proc *q, *t; 847 848 sx_assert(&proctree_lock, SA_XLOCKED); 849 PROC_LOCK_ASSERT(p, MA_OWNED); 850 PROC_SLOCK_ASSERT(p, MA_OWNED); 851 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 852 853 q = td->td_proc; 854 855 PROC_SUNLOCK(p); 856 td->td_retval[0] = p->p_pid; 857 if (status) 858 *status = p->p_xstat; /* convert to int */ 859 if (options & WNOWAIT) { 860 /* 861 * Only poll, returning the status. Caller does not wish to 862 * release the proc struct just yet. 863 */ 864 PROC_UNLOCK(p); 865 sx_xunlock(&proctree_lock); 866 return; 867 } 868 869 PROC_LOCK(q); 870 sigqueue_take(p->p_ksi); 871 PROC_UNLOCK(q); 872 873 /* 874 * If we got the child via a ptrace 'attach', we need to give it back 875 * to the old parent. 876 */ 877 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 878 PROC_UNLOCK(p); 879 t = proc_realparent(p); 880 PROC_LOCK(t); 881 PROC_LOCK(p); 882 CTR2(KTR_PTRACE, 883 "wait: traced child %d moved back to parent %d", p->p_pid, 884 t->p_pid); 885 proc_reparent(p, t); 886 p->p_oppid = 0; 887 PROC_UNLOCK(p); 888 pksignal(t, SIGCHLD, p->p_ksi); 889 wakeup(t); 890 cv_broadcast(&p->p_pwait); 891 PROC_UNLOCK(t); 892 sx_xunlock(&proctree_lock); 893 return; 894 } 895 p->p_oppid = 0; 896 PROC_UNLOCK(p); 897 898 /* 899 * Remove other references to this process to ensure we have an 900 * exclusive reference. 901 */ 902 sx_xlock(&allproc_lock); 903 LIST_REMOVE(p, p_list); /* off zombproc */ 904 sx_xunlock(&allproc_lock); 905 LIST_REMOVE(p, p_sibling); 906 reaper_abandon_children(p, true); 907 LIST_REMOVE(p, p_reapsibling); 908 PROC_LOCK(p); 909 clear_orphan(p); 910 PROC_UNLOCK(p); 911 leavepgrp(p); 912#ifdef PROCDESC 913 if (p->p_procdesc != NULL) 914 procdesc_reap(p); 915#endif 916 sx_xunlock(&proctree_lock); 917 918 /* 919 * As a side effect of this lock, we know that all other writes to 920 * this proc are visible now, so no more locking is needed for p. 921 */ 922 PROC_LOCK(p); 923 p->p_xstat = 0; /* XXX: why? */ 924 PROC_UNLOCK(p); 925 PROC_LOCK(q); 926 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 927 PROC_UNLOCK(q); 928 929 /* 930 * Decrement the count of procs running with this uid. 931 */ 932 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 933 934 /* 935 * Destroy resource accounting information associated with the process. 936 */ 937#ifdef RACCT 938 if (racct_enable) { 939 PROC_LOCK(p); 940 racct_sub(p, RACCT_NPROC, 1); 941 PROC_UNLOCK(p); 942 } 943#endif 944 racct_proc_exit(p); 945 946 /* 947 * Free credentials, arguments, and sigacts. 948 */ 949 crfree(p->p_ucred); 950 proc_set_cred(p, NULL); 951 pargs_drop(p->p_args); 952 p->p_args = NULL; 953 sigacts_free(p->p_sigacts); 954 p->p_sigacts = NULL; 955 956 /* 957 * Do any thread-system specific cleanups. 958 */ 959 thread_wait(p); 960 961 /* 962 * Give vm and machine-dependent layer a chance to free anything that 963 * cpu_exit couldn't release while still running in process context. 964 */ 965 vm_waitproc(p); 966#ifdef MAC 967 mac_proc_destroy(p); 968#endif 969 KASSERT(FIRST_THREAD_IN_PROC(p), 970 ("proc_reap: no residual thread!")); 971 uma_zfree(proc_zone, p); 972 atomic_add_int(&nprocs, -1); 973} 974 975static int 976proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 977 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 978 int check_only) 979{ 980 struct rusage *rup; 981 982 sx_assert(&proctree_lock, SA_XLOCKED); 983 984 PROC_LOCK(p); 985 986 switch (idtype) { 987 case P_ALL: 988 break; 989 case P_PID: 990 if (p->p_pid != (pid_t)id) { 991 PROC_UNLOCK(p); 992 return (0); 993 } 994 break; 995 case P_PGID: 996 if (p->p_pgid != (pid_t)id) { 997 PROC_UNLOCK(p); 998 return (0); 999 } 1000 break; 1001 case P_SID: 1002 if (p->p_session->s_sid != (pid_t)id) { 1003 PROC_UNLOCK(p); 1004 return (0); 1005 } 1006 break; 1007 case P_UID: 1008 if (p->p_ucred->cr_uid != (uid_t)id) { 1009 PROC_UNLOCK(p); 1010 return (0); 1011 } 1012 break; 1013 case P_GID: 1014 if (p->p_ucred->cr_gid != (gid_t)id) { 1015 PROC_UNLOCK(p); 1016 return (0); 1017 } 1018 break; 1019 case P_JAILID: 1020 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1021 PROC_UNLOCK(p); 1022 return (0); 1023 } 1024 break; 1025 /* 1026 * It seems that the thread structures get zeroed out 1027 * at process exit. This makes it impossible to 1028 * support P_SETID, P_CID or P_CPUID. 1029 */ 1030 default: 1031 PROC_UNLOCK(p); 1032 return (0); 1033 } 1034 1035 if (p_canwait(td, p)) { 1036 PROC_UNLOCK(p); 1037 return (0); 1038 } 1039 1040 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1041 PROC_UNLOCK(p); 1042 return (0); 1043 } 1044 1045 /* 1046 * This special case handles a kthread spawned by linux_clone 1047 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1048 * functions need to be able to distinguish between waiting 1049 * on a process and waiting on a thread. It is a thread if 1050 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1051 * signifies we want to wait for threads and not processes. 1052 */ 1053 if ((p->p_sigparent != SIGCHLD) ^ 1054 ((options & WLINUXCLONE) != 0)) { 1055 PROC_UNLOCK(p); 1056 return (0); 1057 } 1058 1059 if (siginfo != NULL) { 1060 bzero(siginfo, sizeof(*siginfo)); 1061 siginfo->si_errno = 0; 1062 1063 /* 1064 * SUSv4 requires that the si_signo value is always 1065 * SIGCHLD. Obey it despite the rfork(2) interface 1066 * allows to request other signal for child exit 1067 * notification. 1068 */ 1069 siginfo->si_signo = SIGCHLD; 1070 1071 /* 1072 * This is still a rough estimate. We will fix the 1073 * cases TRAPPED, STOPPED, and CONTINUED later. 1074 */ 1075 if (WCOREDUMP(p->p_xstat)) { 1076 siginfo->si_code = CLD_DUMPED; 1077 siginfo->si_status = WTERMSIG(p->p_xstat); 1078 } else if (WIFSIGNALED(p->p_xstat)) { 1079 siginfo->si_code = CLD_KILLED; 1080 siginfo->si_status = WTERMSIG(p->p_xstat); 1081 } else { 1082 siginfo->si_code = CLD_EXITED; 1083 siginfo->si_status = WEXITSTATUS(p->p_xstat); 1084 } 1085 1086 siginfo->si_pid = p->p_pid; 1087 siginfo->si_uid = p->p_ucred->cr_uid; 1088 1089 /* 1090 * The si_addr field would be useful additional 1091 * detail, but apparently the PC value may be lost 1092 * when we reach this point. bzero() above sets 1093 * siginfo->si_addr to NULL. 1094 */ 1095 } 1096 1097 /* 1098 * There should be no reason to limit resources usage info to 1099 * exited processes only. A snapshot about any resources used 1100 * by a stopped process may be exactly what is needed. 1101 */ 1102 if (wrusage != NULL) { 1103 rup = &wrusage->wru_self; 1104 *rup = p->p_ru; 1105 PROC_STATLOCK(p); 1106 calcru(p, &rup->ru_utime, &rup->ru_stime); 1107 PROC_STATUNLOCK(p); 1108 1109 rup = &wrusage->wru_children; 1110 *rup = p->p_stats->p_cru; 1111 calccru(p, &rup->ru_utime, &rup->ru_stime); 1112 } 1113 1114 if (p->p_state == PRS_ZOMBIE && !check_only) { 1115 PROC_SLOCK(p); 1116 proc_reap(td, p, status, options); 1117 return (-1); 1118 } 1119 PROC_UNLOCK(p); 1120 return (1); 1121} 1122 1123int 1124kern_wait(struct thread *td, pid_t pid, int *status, int options, 1125 struct rusage *rusage) 1126{ 1127 struct __wrusage wru, *wrup; 1128 idtype_t idtype; 1129 id_t id; 1130 int ret; 1131 1132 /* 1133 * Translate the special pid values into the (idtype, pid) 1134 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1135 * kern_wait6() on its own. 1136 */ 1137 if (pid == WAIT_ANY) { 1138 idtype = P_ALL; 1139 id = 0; 1140 } else if (pid < 0) { 1141 idtype = P_PGID; 1142 id = (id_t)-pid; 1143 } else { 1144 idtype = P_PID; 1145 id = (id_t)pid; 1146 } 1147 1148 if (rusage != NULL) 1149 wrup = &wru; 1150 else 1151 wrup = NULL; 1152 1153 /* 1154 * For backward compatibility we implicitly add flags WEXITED 1155 * and WTRAPPED here. 1156 */ 1157 options |= WEXITED | WTRAPPED; 1158 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1159 if (rusage != NULL) 1160 *rusage = wru.wru_self; 1161 return (ret); 1162} 1163 1164int 1165kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1166 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1167{ 1168 struct proc *p, *q; 1169 int error, nfound, ret; 1170 1171 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1172 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1173 AUDIT_ARG_VALUE(options); 1174 1175 q = td->td_proc; 1176 1177 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1178 PROC_LOCK(q); 1179 id = (id_t)q->p_pgid; 1180 PROC_UNLOCK(q); 1181 idtype = P_PGID; 1182 } 1183 1184 /* If we don't know the option, just return. */ 1185 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1186 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1187 return (EINVAL); 1188 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1189 /* 1190 * We will be unable to find any matching processes, 1191 * because there are no known events to look for. 1192 * Prefer to return error instead of blocking 1193 * indefinitely. 1194 */ 1195 return (EINVAL); 1196 } 1197 1198loop: 1199 if (q->p_flag & P_STATCHILD) { 1200 PROC_LOCK(q); 1201 q->p_flag &= ~P_STATCHILD; 1202 PROC_UNLOCK(q); 1203 } 1204 nfound = 0; 1205 sx_xlock(&proctree_lock); 1206 LIST_FOREACH(p, &q->p_children, p_sibling) { 1207 ret = proc_to_reap(td, p, idtype, id, status, options, 1208 wrusage, siginfo, 0); 1209 if (ret == 0) 1210 continue; 1211 else if (ret == 1) 1212 nfound++; 1213 else 1214 return (0); 1215 1216 PROC_LOCK(p); 1217 PROC_SLOCK(p); 1218 1219 if ((options & WTRAPPED) != 0 && 1220 (p->p_flag & P_TRACED) != 0 && 1221 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1222 (p->p_suspcount == p->p_numthreads) && 1223 ((p->p_flag & P_WAITED) == 0)) { 1224 PROC_SUNLOCK(p); 1225 if ((options & WNOWAIT) == 0) 1226 p->p_flag |= P_WAITED; 1227 sx_xunlock(&proctree_lock); 1228 td->td_retval[0] = p->p_pid; 1229 1230 if (status != NULL) 1231 *status = W_STOPCODE(p->p_xstat); 1232 if (siginfo != NULL) { 1233 siginfo->si_status = p->p_xstat; 1234 siginfo->si_code = CLD_TRAPPED; 1235 } 1236 if ((options & WNOWAIT) == 0) { 1237 PROC_LOCK(q); 1238 sigqueue_take(p->p_ksi); 1239 PROC_UNLOCK(q); 1240 } 1241 1242 CTR4(KTR_PTRACE, 1243 "wait: returning trapped pid %d status %#x (xstat %d) xthread %d", 1244 p->p_pid, W_STOPCODE(p->p_xstat), p->p_xstat, 1245 p->p_xthread != NULL ? p->p_xthread->td_tid : -1); 1246 PROC_UNLOCK(p); 1247 return (0); 1248 } 1249 if ((options & WUNTRACED) != 0 && 1250 (p->p_flag & P_STOPPED_SIG) != 0 && 1251 (p->p_suspcount == p->p_numthreads) && 1252 ((p->p_flag & P_WAITED) == 0)) { 1253 PROC_SUNLOCK(p); 1254 if ((options & WNOWAIT) == 0) 1255 p->p_flag |= P_WAITED; 1256 sx_xunlock(&proctree_lock); 1257 td->td_retval[0] = p->p_pid; 1258 1259 if (status != NULL) 1260 *status = W_STOPCODE(p->p_xstat); 1261 if (siginfo != NULL) { 1262 siginfo->si_status = p->p_xstat; 1263 siginfo->si_code = CLD_STOPPED; 1264 } 1265 if ((options & WNOWAIT) == 0) { 1266 PROC_LOCK(q); 1267 sigqueue_take(p->p_ksi); 1268 PROC_UNLOCK(q); 1269 } 1270 1271 PROC_UNLOCK(p); 1272 return (0); 1273 } 1274 PROC_SUNLOCK(p); 1275 if ((options & WCONTINUED) != 0 && 1276 (p->p_flag & P_CONTINUED) != 0) { 1277 sx_xunlock(&proctree_lock); 1278 td->td_retval[0] = p->p_pid; 1279 if ((options & WNOWAIT) == 0) { 1280 p->p_flag &= ~P_CONTINUED; 1281 PROC_LOCK(q); 1282 sigqueue_take(p->p_ksi); 1283 PROC_UNLOCK(q); 1284 } 1285 PROC_UNLOCK(p); 1286 1287 if (status != NULL) 1288 *status = SIGCONT; 1289 if (siginfo != NULL) { 1290 siginfo->si_status = SIGCONT; 1291 siginfo->si_code = CLD_CONTINUED; 1292 } 1293 return (0); 1294 } 1295 PROC_UNLOCK(p); 1296 } 1297 1298 /* 1299 * Look in the orphans list too, to allow the parent to 1300 * collect it's child exit status even if child is being 1301 * debugged. 1302 * 1303 * Debugger detaches from the parent upon successful 1304 * switch-over from parent to child. At this point due to 1305 * re-parenting the parent loses the child to debugger and a 1306 * wait4(2) call would report that it has no children to wait 1307 * for. By maintaining a list of orphans we allow the parent 1308 * to successfully wait until the child becomes a zombie. 1309 */ 1310 if (nfound == 0) { 1311 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1312 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1313 NULL, NULL, 1); 1314 if (ret != 0) { 1315 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1316 (int)td->td_retval[0])); 1317 nfound++; 1318 break; 1319 } 1320 } 1321 } 1322 if (nfound == 0) { 1323 sx_xunlock(&proctree_lock); 1324 return (ECHILD); 1325 } 1326 if (options & WNOHANG) { 1327 sx_xunlock(&proctree_lock); 1328 td->td_retval[0] = 0; 1329 return (0); 1330 } 1331 PROC_LOCK(q); 1332 sx_xunlock(&proctree_lock); 1333 if (q->p_flag & P_STATCHILD) { 1334 q->p_flag &= ~P_STATCHILD; 1335 error = 0; 1336 } else 1337 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1338 PROC_UNLOCK(q); 1339 if (error) 1340 return (error); 1341 goto loop; 1342} 1343 1344/* 1345 * Make process 'parent' the new parent of process 'child'. 1346 * Must be called with an exclusive hold of proctree lock. 1347 */ 1348void 1349proc_reparent(struct proc *child, struct proc *parent) 1350{ 1351 1352 sx_assert(&proctree_lock, SX_XLOCKED); 1353 PROC_LOCK_ASSERT(child, MA_OWNED); 1354 if (child->p_pptr == parent) 1355 return; 1356 1357 PROC_LOCK(child->p_pptr); 1358 sigqueue_take(child->p_ksi); 1359 PROC_UNLOCK(child->p_pptr); 1360 LIST_REMOVE(child, p_sibling); 1361 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1362 1363 clear_orphan(child); 1364 if (child->p_flag & P_TRACED) { 1365 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1366 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1367 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1368 p_orphan); 1369 } else { 1370 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1371 child, p_orphan); 1372 } 1373 child->p_treeflag |= P_TREE_ORPHANED; 1374 } 1375 1376 child->p_pptr = parent; 1377} 1378