kern_exit.c revision 284665
1228990Suqs/*- 2228990Suqs * Copyright (c) 1982, 1986, 1989, 1991, 1993 3228990Suqs * The Regents of the University of California. All rights reserved. 4422Sjkh * (c) UNIX System Laboratories, Inc. 5422Sjkh * All or some portions of this file are derived from material licensed 6422Sjkh * to the University of California by American Telephone and Telegraph 74996Sjkh * Co. or Unix System Laboratories, Inc. and are reproduced herein with 84996Sjkh * the permission of UNIX System Laboratories, Inc. 97998Sjkh * 107998Sjkh * 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 284665 2015-06-21 06:28:26Z trasz $"); 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, kernel, , 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); 339 340 /* 341 * Stop the real interval timer. If the handler is currently 342 * executing, prevent it from rearming itself and let it finish. 343 */ 344 if (timevalisset(&p->p_realtimer.it_value) && 345 callout_stop(&p->p_itcallout) == 0) { 346 timevalclear(&p->p_realtimer.it_interval); 347 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 348 KASSERT(!timevalisset(&p->p_realtimer.it_value), 349 ("realtime timer is still armed")); 350 } 351 PROC_UNLOCK(p); 352 353 /* 354 * Reset any sigio structures pointing to us as a result of 355 * F_SETOWN with our pid. 356 */ 357 funsetownlst(&p->p_sigiolst); 358 359 /* 360 * If this process has an nlminfo data area (for lockd), release it 361 */ 362 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 363 (*nlminfo_release_p)(p); 364 365 /* 366 * Close open files and release open-file table. 367 * This may block! 368 */ 369 fdescfree(td); 370 371 /* 372 * If this thread tickled GEOM, we need to wait for the giggling to 373 * stop before we return to userland 374 */ 375 if (td->td_pflags & TDP_GEOM) 376 g_waitidle(); 377 378 /* 379 * Remove ourself from our leader's peer list and wake our leader. 380 */ 381 mtx_lock(&ppeers_lock); 382 if (p->p_leader->p_peers) { 383 q = p->p_leader; 384 while (q->p_peers != p) 385 q = q->p_peers; 386 q->p_peers = p->p_peers; 387 wakeup(p->p_leader); 388 } 389 mtx_unlock(&ppeers_lock); 390 391 vmspace_exit(td); 392 393 sx_xlock(&proctree_lock); 394 if (SESS_LEADER(p)) { 395 struct session *sp = p->p_session; 396 struct tty *tp; 397 398 /* 399 * s_ttyp is not zero'd; we use this to indicate that 400 * the session once had a controlling terminal. (for 401 * logging and informational purposes) 402 */ 403 SESS_LOCK(sp); 404 ttyvp = sp->s_ttyvp; 405 tp = sp->s_ttyp; 406 sp->s_ttyvp = NULL; 407 sp->s_ttydp = NULL; 408 sp->s_leader = NULL; 409 SESS_UNLOCK(sp); 410 411 /* 412 * Signal foreground pgrp and revoke access to 413 * controlling terminal if it has not been revoked 414 * already. 415 * 416 * Because the TTY may have been revoked in the mean 417 * time and could already have a new session associated 418 * with it, make sure we don't send a SIGHUP to a 419 * foreground process group that does not belong to this 420 * session. 421 */ 422 423 if (tp != NULL) { 424 tty_lock(tp); 425 if (tp->t_session == sp) 426 tty_signal_pgrp(tp, SIGHUP); 427 tty_unlock(tp); 428 } 429 430 if (ttyvp != NULL) { 431 sx_xunlock(&proctree_lock); 432 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 433 VOP_REVOKE(ttyvp, REVOKEALL); 434 VOP_UNLOCK(ttyvp, 0); 435 } 436 sx_xlock(&proctree_lock); 437 } 438 } 439 fixjobc(p, p->p_pgrp, 0); 440 sx_xunlock(&proctree_lock); 441 (void)acct_process(td); 442 443 /* Release the TTY now we've unlocked everything. */ 444 if (ttyvp != NULL) 445 vrele(ttyvp); 446#ifdef KTRACE 447 ktrprocexit(td); 448#endif 449 /* 450 * Release reference to text vnode 451 */ 452 if (p->p_textvp != NULL) { 453 vrele(p->p_textvp); 454 p->p_textvp = NULL; 455 } 456 457 /* 458 * Release our limits structure. 459 */ 460 lim_free(p->p_limit); 461 p->p_limit = NULL; 462 463 tidhash_remove(td); 464 465 /* 466 * Remove proc from allproc queue and pidhash chain. 467 * Place onto zombproc. Unlink from parent's child list. 468 */ 469 sx_xlock(&allproc_lock); 470 LIST_REMOVE(p, p_list); 471 LIST_INSERT_HEAD(&zombproc, p, p_list); 472 LIST_REMOVE(p, p_hash); 473 sx_xunlock(&allproc_lock); 474 475 /* 476 * Call machine-dependent code to release any 477 * machine-dependent resources other than the address space. 478 * The address space is released by "vmspace_exitfree(p)" in 479 * vm_waitproc(). 480 */ 481 cpu_exit(td); 482 483 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 484 485 /* 486 * Reparent all children processes: 487 * - traced ones to the original parent (or init if we are that parent) 488 * - the rest to init 489 */ 490 sx_xlock(&proctree_lock); 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 PROC_LOCK(q); 497 q->p_sigparent = SIGCHLD; 498 499 if (!(q->p_flag & P_TRACED)) { 500 proc_reparent(q, q->p_reaper); 501 } else { 502 /* 503 * Traced processes are killed since their existence 504 * means someone is screwing up. 505 */ 506 t = proc_realparent(q); 507 if (t == p) { 508 proc_reparent(q, q->p_reaper); 509 } else { 510 PROC_LOCK(t); 511 proc_reparent(q, t); 512 PROC_UNLOCK(t); 513 } 514 /* 515 * Since q was found on our children list, the 516 * proc_reparent() call moved q to the orphan 517 * list due to present P_TRACED flag. Clear 518 * orphan link for q now while q is locked. 519 */ 520 clear_orphan(q); 521 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 522 FOREACH_THREAD_IN_PROC(q, tdt) 523 tdt->td_dbgflags &= ~TDB_SUSPEND; 524 kern_psignal(q, SIGKILL); 525 } 526 PROC_UNLOCK(q); 527 } 528 529 /* 530 * Also get rid of our orphans. 531 */ 532 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 533 PROC_LOCK(q); 534 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 535 q->p_pid); 536 clear_orphan(q); 537 PROC_UNLOCK(q); 538 } 539 540 /* Save exit status. */ 541 PROC_LOCK(p); 542 p->p_xthread = td; 543 544 /* Tell the prison that we are gone. */ 545 prison_proc_free(p->p_ucred->cr_prison); 546 547#ifdef KDTRACE_HOOKS 548 /* 549 * Tell the DTrace fasttrap provider about the exit if it 550 * has declared an interest. 551 */ 552 if (dtrace_fasttrap_exit) 553 dtrace_fasttrap_exit(p); 554#endif 555 556 /* 557 * Notify interested parties of our demise. 558 */ 559 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 560 561#ifdef KDTRACE_HOOKS 562 int reason = CLD_EXITED; 563 if (WCOREDUMP(rv)) 564 reason = CLD_DUMPED; 565 else if (WIFSIGNALED(rv)) 566 reason = CLD_KILLED; 567 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 568#endif 569 570 /* 571 * Just delete all entries in the p_klist. At this point we won't 572 * report any more events, and there are nasty race conditions that 573 * can beat us if we don't. 574 */ 575 knlist_clear(&p->p_klist, 1); 576 577 /* 578 * If this is a process with a descriptor, we may not need to deliver 579 * a signal to the parent. proctree_lock is held over 580 * procdesc_exit() to serialize concurrent calls to close() and 581 * exit(). 582 */ 583#ifdef PROCDESC 584 if (p->p_procdesc == NULL || procdesc_exit(p)) { 585#endif 586 /* 587 * Notify parent that we're gone. If parent has the 588 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 589 * notify process 1 instead (and hope it will handle this 590 * situation). 591 */ 592 PROC_LOCK(p->p_pptr); 593 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 594 if (p->p_pptr->p_sigacts->ps_flag & 595 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 596 struct proc *pp; 597 598 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 599 pp = p->p_pptr; 600 PROC_UNLOCK(pp); 601 proc_reparent(p, p->p_reaper); 602 p->p_sigparent = SIGCHLD; 603 PROC_LOCK(p->p_pptr); 604 605 /* 606 * Notify parent, so in case he was wait(2)ing or 607 * executing waitpid(2) with our pid, he will 608 * continue. 609 */ 610 wakeup(pp); 611 } else 612 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 613 614 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 615 childproc_exited(p); 616 else if (p->p_sigparent != 0) { 617 if (p->p_sigparent == SIGCHLD) 618 childproc_exited(p); 619 else /* LINUX thread */ 620 kern_psignal(p->p_pptr, p->p_sigparent); 621 } 622#ifdef PROCDESC 623 } else 624 PROC_LOCK(p->p_pptr); 625#endif 626 sx_xunlock(&proctree_lock); 627 628 /* 629 * The state PRS_ZOMBIE prevents other proesses from sending 630 * signal to the process, to avoid memory leak, we free memory 631 * for signal queue at the time when the state is set. 632 */ 633 sigqueue_flush(&p->p_sigqueue); 634 sigqueue_flush(&td->td_sigqueue); 635 636 /* 637 * We have to wait until after acquiring all locks before 638 * changing p_state. We need to avoid all possible context 639 * switches (including ones from blocking on a mutex) while 640 * marked as a zombie. We also have to set the zombie state 641 * before we release the parent process' proc lock to avoid 642 * a lost wakeup. So, we first call wakeup, then we grab the 643 * sched lock, update the state, and release the parent process' 644 * proc lock. 645 */ 646 wakeup(p->p_pptr); 647 cv_broadcast(&p->p_pwait); 648 sched_exit(p->p_pptr, td); 649 umtx_thread_exit(td); 650 PROC_SLOCK(p); 651 p->p_state = PRS_ZOMBIE; 652 PROC_UNLOCK(p->p_pptr); 653 654 /* 655 * Hopefully no one will try to deliver a signal to the process this 656 * late in the game. 657 */ 658 knlist_destroy(&p->p_klist); 659 660 /* 661 * Save our children's rusage information in our exit rusage. 662 */ 663 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 664 665 /* 666 * Make sure the scheduler takes this thread out of its tables etc. 667 * This will also release this thread's reference to the ucred. 668 * Other thread parts to release include pcb bits and such. 669 */ 670 thread_exit(); 671} 672 673 674#ifndef _SYS_SYSPROTO_H_ 675struct abort2_args { 676 char *why; 677 int nargs; 678 void **args; 679}; 680#endif 681 682int 683sys_abort2(struct thread *td, struct abort2_args *uap) 684{ 685 struct proc *p = td->td_proc; 686 struct sbuf *sb; 687 void *uargs[16]; 688 int error, i, sig; 689 690 /* 691 * Do it right now so we can log either proper call of abort2(), or 692 * note, that invalid argument was passed. 512 is big enough to 693 * handle 16 arguments' descriptions with additional comments. 694 */ 695 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 696 sbuf_clear(sb); 697 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 698 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 699 /* 700 * Since we can't return from abort2(), send SIGKILL in cases, where 701 * abort2() was called improperly 702 */ 703 sig = SIGKILL; 704 /* Prevent from DoSes from user-space. */ 705 if (uap->nargs < 0 || uap->nargs > 16) 706 goto out; 707 if (uap->nargs > 0) { 708 if (uap->args == NULL) 709 goto out; 710 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 711 if (error != 0) 712 goto out; 713 } 714 /* 715 * Limit size of 'reason' string to 128. Will fit even when 716 * maximal number of arguments was chosen to be logged. 717 */ 718 if (uap->why != NULL) { 719 error = sbuf_copyin(sb, uap->why, 128); 720 if (error < 0) 721 goto out; 722 } else { 723 sbuf_printf(sb, "(null)"); 724 } 725 if (uap->nargs > 0) { 726 sbuf_printf(sb, "("); 727 for (i = 0;i < uap->nargs; i++) 728 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 729 sbuf_printf(sb, ")"); 730 } 731 /* 732 * Final stage: arguments were proper, string has been 733 * successfully copied from userspace, and copying pointers 734 * from user-space succeed. 735 */ 736 sig = SIGABRT; 737out: 738 if (sig == SIGKILL) { 739 sbuf_trim(sb); 740 sbuf_printf(sb, " (Reason text inaccessible)"); 741 } 742 sbuf_cat(sb, "\n"); 743 sbuf_finish(sb); 744 log(LOG_INFO, "%s", sbuf_data(sb)); 745 sbuf_delete(sb); 746 exit1(td, W_EXITCODE(0, sig)); 747 return (0); 748} 749 750 751#ifdef COMPAT_43 752/* 753 * The dirty work is handled by kern_wait(). 754 */ 755int 756owait(struct thread *td, struct owait_args *uap __unused) 757{ 758 int error, status; 759 760 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 761 if (error == 0) 762 td->td_retval[1] = status; 763 return (error); 764} 765#endif /* COMPAT_43 */ 766 767/* 768 * The dirty work is handled by kern_wait(). 769 */ 770int 771sys_wait4(struct thread *td, struct wait4_args *uap) 772{ 773 struct rusage ru, *rup; 774 int error, status; 775 776 if (uap->rusage != NULL) 777 rup = &ru; 778 else 779 rup = NULL; 780 error = kern_wait(td, uap->pid, &status, uap->options, rup); 781 if (uap->status != NULL && error == 0) 782 error = copyout(&status, uap->status, sizeof(status)); 783 if (uap->rusage != NULL && error == 0) 784 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 785 return (error); 786} 787 788int 789sys_wait6(struct thread *td, struct wait6_args *uap) 790{ 791 struct __wrusage wru, *wrup; 792 siginfo_t si, *sip; 793 idtype_t idtype; 794 id_t id; 795 int error, status; 796 797 idtype = uap->idtype; 798 id = uap->id; 799 800 if (uap->wrusage != NULL) 801 wrup = &wru; 802 else 803 wrup = NULL; 804 805 if (uap->info != NULL) { 806 sip = &si; 807 bzero(sip, sizeof(*sip)); 808 } else 809 sip = NULL; 810 811 /* 812 * We expect all callers of wait6() to know about WEXITED and 813 * WTRAPPED. 814 */ 815 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 816 817 if (uap->status != NULL && error == 0) 818 error = copyout(&status, uap->status, sizeof(status)); 819 if (uap->wrusage != NULL && error == 0) 820 error = copyout(&wru, uap->wrusage, sizeof(wru)); 821 if (uap->info != NULL && error == 0) 822 error = copyout(&si, uap->info, sizeof(si)); 823 return (error); 824} 825 826/* 827 * Reap the remains of a zombie process and optionally return status and 828 * rusage. Asserts and will release both the proctree_lock and the process 829 * lock as part of its work. 830 */ 831void 832proc_reap(struct thread *td, struct proc *p, int *status, int options) 833{ 834 struct proc *q, *t; 835 836 sx_assert(&proctree_lock, SA_XLOCKED); 837 PROC_LOCK_ASSERT(p, MA_OWNED); 838 PROC_SLOCK_ASSERT(p, MA_OWNED); 839 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 840 841 q = td->td_proc; 842 843 PROC_SUNLOCK(p); 844 td->td_retval[0] = p->p_pid; 845 if (status) 846 *status = p->p_xstat; /* convert to int */ 847 if (options & WNOWAIT) { 848 /* 849 * Only poll, returning the status. Caller does not wish to 850 * release the proc struct just yet. 851 */ 852 PROC_UNLOCK(p); 853 sx_xunlock(&proctree_lock); 854 return; 855 } 856 857 PROC_LOCK(q); 858 sigqueue_take(p->p_ksi); 859 PROC_UNLOCK(q); 860 PROC_UNLOCK(p); 861 862 /* 863 * If we got the child via a ptrace 'attach', we need to give it back 864 * to the old parent. 865 */ 866 if (p->p_oppid != 0) { 867 t = proc_realparent(p); 868 PROC_LOCK(t); 869 PROC_LOCK(p); 870 CTR2(KTR_PTRACE, 871 "wait: traced child %d moved back to parent %d", p->p_pid, 872 t->p_pid); 873 proc_reparent(p, t); 874 p->p_oppid = 0; 875 PROC_UNLOCK(p); 876 pksignal(t, SIGCHLD, p->p_ksi); 877 wakeup(t); 878 cv_broadcast(&p->p_pwait); 879 PROC_UNLOCK(t); 880 sx_xunlock(&proctree_lock); 881 return; 882 } 883 884 /* 885 * Remove other references to this process to ensure we have an 886 * exclusive reference. 887 */ 888 sx_xlock(&allproc_lock); 889 LIST_REMOVE(p, p_list); /* off zombproc */ 890 sx_xunlock(&allproc_lock); 891 LIST_REMOVE(p, p_sibling); 892 reaper_abandon_children(p, true); 893 LIST_REMOVE(p, p_reapsibling); 894 PROC_LOCK(p); 895 clear_orphan(p); 896 PROC_UNLOCK(p); 897 leavepgrp(p); 898#ifdef PROCDESC 899 if (p->p_procdesc != NULL) 900 procdesc_reap(p); 901#endif 902 sx_xunlock(&proctree_lock); 903 904 /* 905 * As a side effect of this lock, we know that all other writes to 906 * this proc are visible now, so no more locking is needed for p. 907 */ 908 PROC_LOCK(p); 909 p->p_xstat = 0; /* XXX: why? */ 910 PROC_UNLOCK(p); 911 PROC_LOCK(q); 912 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 913 PROC_UNLOCK(q); 914 915 /* 916 * Decrement the count of procs running with this uid. 917 */ 918 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 919 920 /* 921 * Destroy resource accounting information associated with the process. 922 */ 923#ifdef RACCT 924 if (racct_enable) { 925 PROC_LOCK(p); 926 racct_sub(p, RACCT_NPROC, 1); 927 PROC_UNLOCK(p); 928 } 929#endif 930 racct_proc_exit(p); 931 932 /* 933 * Free credentials, arguments, and sigacts. 934 */ 935 crfree(p->p_ucred); 936 p->p_ucred = NULL; 937 pargs_drop(p->p_args); 938 p->p_args = NULL; 939 sigacts_free(p->p_sigacts); 940 p->p_sigacts = NULL; 941 942 /* 943 * Do any thread-system specific cleanups. 944 */ 945 thread_wait(p); 946 947 /* 948 * Give vm and machine-dependent layer a chance to free anything that 949 * cpu_exit couldn't release while still running in process context. 950 */ 951 vm_waitproc(p); 952#ifdef MAC 953 mac_proc_destroy(p); 954#endif 955 KASSERT(FIRST_THREAD_IN_PROC(p), 956 ("proc_reap: no residual thread!")); 957 uma_zfree(proc_zone, p); 958 sx_xlock(&allproc_lock); 959 nprocs--; 960 sx_xunlock(&allproc_lock); 961} 962 963static int 964proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 965 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo) 966{ 967 struct rusage *rup; 968 969 sx_assert(&proctree_lock, SA_XLOCKED); 970 971 PROC_LOCK(p); 972 973 switch (idtype) { 974 case P_ALL: 975 break; 976 case P_PID: 977 if (p->p_pid != (pid_t)id) { 978 PROC_UNLOCK(p); 979 return (0); 980 } 981 break; 982 case P_PGID: 983 if (p->p_pgid != (pid_t)id) { 984 PROC_UNLOCK(p); 985 return (0); 986 } 987 break; 988 case P_SID: 989 if (p->p_session->s_sid != (pid_t)id) { 990 PROC_UNLOCK(p); 991 return (0); 992 } 993 break; 994 case P_UID: 995 if (p->p_ucred->cr_uid != (uid_t)id) { 996 PROC_UNLOCK(p); 997 return (0); 998 } 999 break; 1000 case P_GID: 1001 if (p->p_ucred->cr_gid != (gid_t)id) { 1002 PROC_UNLOCK(p); 1003 return (0); 1004 } 1005 break; 1006 case P_JAILID: 1007 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1008 PROC_UNLOCK(p); 1009 return (0); 1010 } 1011 break; 1012 /* 1013 * It seems that the thread structures get zeroed out 1014 * at process exit. This makes it impossible to 1015 * support P_SETID, P_CID or P_CPUID. 1016 */ 1017 default: 1018 PROC_UNLOCK(p); 1019 return (0); 1020 } 1021 1022 if (p_canwait(td, p)) { 1023 PROC_UNLOCK(p); 1024 return (0); 1025 } 1026 1027 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1028 PROC_UNLOCK(p); 1029 return (0); 1030 } 1031 1032 /* 1033 * This special case handles a kthread spawned by linux_clone 1034 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1035 * functions need to be able to distinguish between waiting 1036 * on a process and waiting on a thread. It is a thread if 1037 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1038 * signifies we want to wait for threads and not processes. 1039 */ 1040 if ((p->p_sigparent != SIGCHLD) ^ 1041 ((options & WLINUXCLONE) != 0)) { 1042 PROC_UNLOCK(p); 1043 return (0); 1044 } 1045 1046 PROC_SLOCK(p); 1047 1048 if (siginfo != NULL) { 1049 bzero(siginfo, sizeof(*siginfo)); 1050 siginfo->si_errno = 0; 1051 1052 /* 1053 * SUSv4 requires that the si_signo value is always 1054 * SIGCHLD. Obey it despite the rfork(2) interface 1055 * allows to request other signal for child exit 1056 * notification. 1057 */ 1058 siginfo->si_signo = SIGCHLD; 1059 1060 /* 1061 * This is still a rough estimate. We will fix the 1062 * cases TRAPPED, STOPPED, and CONTINUED later. 1063 */ 1064 if (WCOREDUMP(p->p_xstat)) { 1065 siginfo->si_code = CLD_DUMPED; 1066 siginfo->si_status = WTERMSIG(p->p_xstat); 1067 } else if (WIFSIGNALED(p->p_xstat)) { 1068 siginfo->si_code = CLD_KILLED; 1069 siginfo->si_status = WTERMSIG(p->p_xstat); 1070 } else { 1071 siginfo->si_code = CLD_EXITED; 1072 siginfo->si_status = WEXITSTATUS(p->p_xstat); 1073 } 1074 1075 siginfo->si_pid = p->p_pid; 1076 siginfo->si_uid = p->p_ucred->cr_uid; 1077 1078 /* 1079 * The si_addr field would be useful additional 1080 * detail, but apparently the PC value may be lost 1081 * when we reach this point. bzero() above sets 1082 * siginfo->si_addr to NULL. 1083 */ 1084 } 1085 1086 /* 1087 * There should be no reason to limit resources usage info to 1088 * exited processes only. A snapshot about any resources used 1089 * by a stopped process may be exactly what is needed. 1090 */ 1091 if (wrusage != NULL) { 1092 rup = &wrusage->wru_self; 1093 *rup = p->p_ru; 1094 calcru(p, &rup->ru_utime, &rup->ru_stime); 1095 1096 rup = &wrusage->wru_children; 1097 *rup = p->p_stats->p_cru; 1098 calccru(p, &rup->ru_utime, &rup->ru_stime); 1099 } 1100 1101 if (p->p_state == PRS_ZOMBIE) { 1102 proc_reap(td, p, status, options); 1103 return (-1); 1104 } 1105 PROC_SUNLOCK(p); 1106 PROC_UNLOCK(p); 1107 return (1); 1108} 1109 1110int 1111kern_wait(struct thread *td, pid_t pid, int *status, int options, 1112 struct rusage *rusage) 1113{ 1114 struct __wrusage wru, *wrup; 1115 idtype_t idtype; 1116 id_t id; 1117 int ret; 1118 1119 /* 1120 * Translate the special pid values into the (idtype, pid) 1121 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1122 * kern_wait6() on its own. 1123 */ 1124 if (pid == WAIT_ANY) { 1125 idtype = P_ALL; 1126 id = 0; 1127 } else if (pid < 0) { 1128 idtype = P_PGID; 1129 id = (id_t)-pid; 1130 } else { 1131 idtype = P_PID; 1132 id = (id_t)pid; 1133 } 1134 1135 if (rusage != NULL) 1136 wrup = &wru; 1137 else 1138 wrup = NULL; 1139 1140 /* 1141 * For backward compatibility we implicitly add flags WEXITED 1142 * and WTRAPPED here. 1143 */ 1144 options |= WEXITED | WTRAPPED; 1145 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1146 if (rusage != NULL) 1147 *rusage = wru.wru_self; 1148 return (ret); 1149} 1150 1151int 1152kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1153 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1154{ 1155 struct proc *p, *q; 1156 int error, nfound, ret; 1157 1158 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1159 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1160 AUDIT_ARG_VALUE(options); 1161 1162 q = td->td_proc; 1163 1164 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1165 PROC_LOCK(q); 1166 id = (id_t)q->p_pgid; 1167 PROC_UNLOCK(q); 1168 idtype = P_PGID; 1169 } 1170 1171 /* If we don't know the option, just return. */ 1172 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1173 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1174 return (EINVAL); 1175 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1176 /* 1177 * We will be unable to find any matching processes, 1178 * because there are no known events to look for. 1179 * Prefer to return error instead of blocking 1180 * indefinitely. 1181 */ 1182 return (EINVAL); 1183 } 1184 1185loop: 1186 if (q->p_flag & P_STATCHILD) { 1187 PROC_LOCK(q); 1188 q->p_flag &= ~P_STATCHILD; 1189 PROC_UNLOCK(q); 1190 } 1191 nfound = 0; 1192 sx_xlock(&proctree_lock); 1193 LIST_FOREACH(p, &q->p_children, p_sibling) { 1194 ret = proc_to_reap(td, p, idtype, id, status, options, 1195 wrusage, siginfo); 1196 if (ret == 0) 1197 continue; 1198 else if (ret == 1) 1199 nfound++; 1200 else 1201 return (0); 1202 1203 PROC_LOCK(p); 1204 PROC_SLOCK(p); 1205 1206 if ((options & WTRAPPED) != 0 && 1207 (p->p_flag & P_TRACED) != 0 && 1208 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1209 (p->p_suspcount == p->p_numthreads) && 1210 ((p->p_flag & P_WAITED) == 0)) { 1211 PROC_SUNLOCK(p); 1212 if ((options & WNOWAIT) == 0) 1213 p->p_flag |= P_WAITED; 1214 sx_xunlock(&proctree_lock); 1215 td->td_retval[0] = p->p_pid; 1216 1217 if (status != NULL) 1218 *status = W_STOPCODE(p->p_xstat); 1219 if (siginfo != NULL) { 1220 siginfo->si_status = p->p_xstat; 1221 siginfo->si_code = CLD_TRAPPED; 1222 } 1223 if ((options & WNOWAIT) == 0) { 1224 PROC_LOCK(q); 1225 sigqueue_take(p->p_ksi); 1226 PROC_UNLOCK(q); 1227 } 1228 1229 CTR4(KTR_PTRACE, 1230 "wait: returning trapped pid %d status %#x (xstat %d) xthread %d", 1231 p->p_pid, W_STOPCODE(p->p_xstat), p->p_xstat, 1232 p->p_xthread != NULL ? p->p_xthread->td_tid : -1); 1233 PROC_UNLOCK(p); 1234 return (0); 1235 } 1236 if ((options & WUNTRACED) != 0 && 1237 (p->p_flag & P_STOPPED_SIG) != 0 && 1238 (p->p_suspcount == p->p_numthreads) && 1239 ((p->p_flag & P_WAITED) == 0)) { 1240 PROC_SUNLOCK(p); 1241 if ((options & WNOWAIT) == 0) 1242 p->p_flag |= P_WAITED; 1243 sx_xunlock(&proctree_lock); 1244 td->td_retval[0] = p->p_pid; 1245 1246 if (status != NULL) 1247 *status = W_STOPCODE(p->p_xstat); 1248 if (siginfo != NULL) { 1249 siginfo->si_status = p->p_xstat; 1250 siginfo->si_code = CLD_STOPPED; 1251 } 1252 if ((options & WNOWAIT) == 0) { 1253 PROC_LOCK(q); 1254 sigqueue_take(p->p_ksi); 1255 PROC_UNLOCK(q); 1256 } 1257 1258 PROC_UNLOCK(p); 1259 return (0); 1260 } 1261 PROC_SUNLOCK(p); 1262 if ((options & WCONTINUED) != 0 && 1263 (p->p_flag & P_CONTINUED) != 0) { 1264 sx_xunlock(&proctree_lock); 1265 td->td_retval[0] = p->p_pid; 1266 if ((options & WNOWAIT) == 0) { 1267 p->p_flag &= ~P_CONTINUED; 1268 PROC_LOCK(q); 1269 sigqueue_take(p->p_ksi); 1270 PROC_UNLOCK(q); 1271 } 1272 PROC_UNLOCK(p); 1273 1274 if (status != NULL) 1275 *status = SIGCONT; 1276 if (siginfo != NULL) { 1277 siginfo->si_status = SIGCONT; 1278 siginfo->si_code = CLD_CONTINUED; 1279 } 1280 return (0); 1281 } 1282 PROC_UNLOCK(p); 1283 } 1284 1285 /* 1286 * Look in the orphans list too, to allow the parent to 1287 * collect it's child exit status even if child is being 1288 * debugged. 1289 * 1290 * Debugger detaches from the parent upon successful 1291 * switch-over from parent to child. At this point due to 1292 * re-parenting the parent loses the child to debugger and a 1293 * wait4(2) call would report that it has no children to wait 1294 * for. By maintaining a list of orphans we allow the parent 1295 * to successfully wait until the child becomes a zombie. 1296 */ 1297 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1298 ret = proc_to_reap(td, p, idtype, id, status, options, 1299 wrusage, siginfo); 1300 if (ret == 0) 1301 continue; 1302 else if (ret == 1) 1303 nfound++; 1304 else 1305 return (0); 1306 } 1307 if (nfound == 0) { 1308 sx_xunlock(&proctree_lock); 1309 return (ECHILD); 1310 } 1311 if (options & WNOHANG) { 1312 sx_xunlock(&proctree_lock); 1313 td->td_retval[0] = 0; 1314 return (0); 1315 } 1316 PROC_LOCK(q); 1317 sx_xunlock(&proctree_lock); 1318 if (q->p_flag & P_STATCHILD) { 1319 q->p_flag &= ~P_STATCHILD; 1320 error = 0; 1321 } else 1322 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1323 PROC_UNLOCK(q); 1324 if (error) 1325 return (error); 1326 goto loop; 1327} 1328 1329/* 1330 * Make process 'parent' the new parent of process 'child'. 1331 * Must be called with an exclusive hold of proctree lock. 1332 */ 1333void 1334proc_reparent(struct proc *child, struct proc *parent) 1335{ 1336 1337 sx_assert(&proctree_lock, SX_XLOCKED); 1338 PROC_LOCK_ASSERT(child, MA_OWNED); 1339 if (child->p_pptr == parent) 1340 return; 1341 1342 PROC_LOCK(child->p_pptr); 1343 sigqueue_take(child->p_ksi); 1344 PROC_UNLOCK(child->p_pptr); 1345 LIST_REMOVE(child, p_sibling); 1346 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1347 1348 clear_orphan(child); 1349 if (child->p_flag & P_TRACED) { 1350 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1351 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1352 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1353 p_orphan); 1354 } else { 1355 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1356 child, p_orphan); 1357 } 1358 child->p_treeflag |= P_TREE_ORPHANED; 1359 } 1360 1361 child->p_pptr = parent; 1362} 1363