kern_exit.c revision 304188
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 304188 2016-08-15 21:10:41Z jhb $"); 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 } else { 503 /* 504 * Traced processes are killed since their existence 505 * means someone is screwing up. 506 */ 507 t = proc_realparent(q); 508 if (t == p) { 509 proc_reparent(q, q->p_reaper); 510 } else { 511 PROC_LOCK(t); 512 proc_reparent(q, t); 513 PROC_UNLOCK(t); 514 } 515 /* 516 * Since q was found on our children list, the 517 * proc_reparent() call moved q to the orphan 518 * list due to present P_TRACED flag. Clear 519 * orphan link for q now while q is locked. 520 */ 521 clear_orphan(q); 522 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 523 q->p_ptevents = 0; 524 FOREACH_THREAD_IN_PROC(q, tdt) 525 tdt->td_dbgflags &= ~TDB_SUSPEND; 526 kern_psignal(q, SIGKILL); 527 } 528 PROC_UNLOCK(q); 529 } 530 531 /* 532 * Also get rid of our orphans. 533 */ 534 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 535 PROC_LOCK(q); 536 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 537 q->p_pid); 538 clear_orphan(q); 539 PROC_UNLOCK(q); 540 } 541 542 /* Save exit status. */ 543 PROC_LOCK(p); 544 p->p_xthread = td; 545 546 /* Tell the prison that we are gone. */ 547 prison_proc_free(p->p_ucred->cr_prison); 548 549#ifdef KDTRACE_HOOKS 550 /* 551 * Tell the DTrace fasttrap provider about the exit if it 552 * has declared an interest. 553 */ 554 if (dtrace_fasttrap_exit) 555 dtrace_fasttrap_exit(p); 556#endif 557 558 /* 559 * Notify interested parties of our demise. 560 */ 561 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 562 563#ifdef KDTRACE_HOOKS 564 int reason = CLD_EXITED; 565 if (WCOREDUMP(rv)) 566 reason = CLD_DUMPED; 567 else if (WIFSIGNALED(rv)) 568 reason = CLD_KILLED; 569 SDT_PROBE1(proc, , , exit, reason); 570#endif 571 572 /* 573 * Just delete all entries in the p_klist. At this point we won't 574 * report any more events, and there are nasty race conditions that 575 * can beat us if we don't. 576 */ 577 knlist_clear(&p->p_klist, 1); 578 579 /* 580 * If this is a process with a descriptor, we may not need to deliver 581 * a signal to the parent. proctree_lock is held over 582 * procdesc_exit() to serialize concurrent calls to close() and 583 * exit(). 584 */ 585#ifdef PROCDESC 586 if (p->p_procdesc == NULL || procdesc_exit(p)) { 587#endif 588 /* 589 * Notify parent that we're gone. If parent has the 590 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 591 * notify process 1 instead (and hope it will handle this 592 * situation). 593 */ 594 PROC_LOCK(p->p_pptr); 595 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 596 if (p->p_pptr->p_sigacts->ps_flag & 597 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 598 struct proc *pp; 599 600 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 601 pp = p->p_pptr; 602 PROC_UNLOCK(pp); 603 proc_reparent(p, p->p_reaper); 604 p->p_sigparent = SIGCHLD; 605 PROC_LOCK(p->p_pptr); 606 607 /* 608 * Notify parent, so in case he was wait(2)ing or 609 * executing waitpid(2) with our pid, he will 610 * continue. 611 */ 612 wakeup(pp); 613 } else 614 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 615 616 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 617 childproc_exited(p); 618 else if (p->p_sigparent != 0) { 619 if (p->p_sigparent == SIGCHLD) 620 childproc_exited(p); 621 else /* LINUX thread */ 622 kern_psignal(p->p_pptr, p->p_sigparent); 623 } 624#ifdef PROCDESC 625 } else 626 PROC_LOCK(p->p_pptr); 627#endif 628 sx_xunlock(&proctree_lock); 629 630 /* 631 * The state PRS_ZOMBIE prevents other proesses from sending 632 * signal to the process, to avoid memory leak, we free memory 633 * for signal queue at the time when the state is set. 634 */ 635 sigqueue_flush(&p->p_sigqueue); 636 sigqueue_flush(&td->td_sigqueue); 637 638 /* 639 * We have to wait until after acquiring all locks before 640 * changing p_state. We need to avoid all possible context 641 * switches (including ones from blocking on a mutex) while 642 * marked as a zombie. We also have to set the zombie state 643 * before we release the parent process' proc lock to avoid 644 * a lost wakeup. So, we first call wakeup, then we grab the 645 * sched lock, update the state, and release the parent process' 646 * proc lock. 647 */ 648 wakeup(p->p_pptr); 649 cv_broadcast(&p->p_pwait); 650 sched_exit(p->p_pptr, td); 651 umtx_thread_exit(td); 652 PROC_SLOCK(p); 653 p->p_state = PRS_ZOMBIE; 654 PROC_UNLOCK(p->p_pptr); 655 656 /* 657 * Hopefully no one will try to deliver a signal to the process this 658 * late in the game. 659 */ 660 knlist_destroy(&p->p_klist); 661 662 /* 663 * Save our children's rusage information in our exit rusage. 664 */ 665 PROC_STATLOCK(p); 666 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 667 PROC_STATUNLOCK(p); 668 669 /* 670 * Make sure the scheduler takes this thread out of its tables etc. 671 * This will also release this thread's reference to the ucred. 672 * Other thread parts to release include pcb bits and such. 673 */ 674 thread_exit(); 675} 676 677 678#ifndef _SYS_SYSPROTO_H_ 679struct abort2_args { 680 char *why; 681 int nargs; 682 void **args; 683}; 684#endif 685 686int 687sys_abort2(struct thread *td, struct abort2_args *uap) 688{ 689 struct proc *p = td->td_proc; 690 struct sbuf *sb; 691 void *uargs[16]; 692 int error, i, sig; 693 694 /* 695 * Do it right now so we can log either proper call of abort2(), or 696 * note, that invalid argument was passed. 512 is big enough to 697 * handle 16 arguments' descriptions with additional comments. 698 */ 699 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 700 sbuf_clear(sb); 701 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 702 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 703 /* 704 * Since we can't return from abort2(), send SIGKILL in cases, where 705 * abort2() was called improperly 706 */ 707 sig = SIGKILL; 708 /* Prevent from DoSes from user-space. */ 709 if (uap->nargs < 0 || uap->nargs > 16) 710 goto out; 711 if (uap->nargs > 0) { 712 if (uap->args == NULL) 713 goto out; 714 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 715 if (error != 0) 716 goto out; 717 } 718 /* 719 * Limit size of 'reason' string to 128. Will fit even when 720 * maximal number of arguments was chosen to be logged. 721 */ 722 if (uap->why != NULL) { 723 error = sbuf_copyin(sb, uap->why, 128); 724 if (error < 0) 725 goto out; 726 } else { 727 sbuf_printf(sb, "(null)"); 728 } 729 if (uap->nargs > 0) { 730 sbuf_printf(sb, "("); 731 for (i = 0;i < uap->nargs; i++) 732 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 733 sbuf_printf(sb, ")"); 734 } 735 /* 736 * Final stage: arguments were proper, string has been 737 * successfully copied from userspace, and copying pointers 738 * from user-space succeed. 739 */ 740 sig = SIGABRT; 741out: 742 if (sig == SIGKILL) { 743 sbuf_trim(sb); 744 sbuf_printf(sb, " (Reason text inaccessible)"); 745 } 746 sbuf_cat(sb, "\n"); 747 sbuf_finish(sb); 748 log(LOG_INFO, "%s", sbuf_data(sb)); 749 sbuf_delete(sb); 750 exit1(td, W_EXITCODE(0, sig)); 751 return (0); 752} 753 754 755#ifdef COMPAT_43 756/* 757 * The dirty work is handled by kern_wait(). 758 */ 759int 760owait(struct thread *td, struct owait_args *uap __unused) 761{ 762 int error, status; 763 764 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 765 if (error == 0) 766 td->td_retval[1] = status; 767 return (error); 768} 769#endif /* COMPAT_43 */ 770 771/* 772 * The dirty work is handled by kern_wait(). 773 */ 774int 775sys_wait4(struct thread *td, struct wait4_args *uap) 776{ 777 struct rusage ru, *rup; 778 int error, status; 779 780 if (uap->rusage != NULL) 781 rup = &ru; 782 else 783 rup = NULL; 784 error = kern_wait(td, uap->pid, &status, uap->options, rup); 785 if (uap->status != NULL && error == 0) 786 error = copyout(&status, uap->status, sizeof(status)); 787 if (uap->rusage != NULL && error == 0) 788 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 789 return (error); 790} 791 792int 793sys_wait6(struct thread *td, struct wait6_args *uap) 794{ 795 struct __wrusage wru, *wrup; 796 siginfo_t si, *sip; 797 idtype_t idtype; 798 id_t id; 799 int error, status; 800 801 idtype = uap->idtype; 802 id = uap->id; 803 804 if (uap->wrusage != NULL) 805 wrup = &wru; 806 else 807 wrup = NULL; 808 809 if (uap->info != NULL) { 810 sip = &si; 811 bzero(sip, sizeof(*sip)); 812 } else 813 sip = NULL; 814 815 /* 816 * We expect all callers of wait6() to know about WEXITED and 817 * WTRAPPED. 818 */ 819 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 820 821 if (uap->status != NULL && error == 0) 822 error = copyout(&status, uap->status, sizeof(status)); 823 if (uap->wrusage != NULL && error == 0) 824 error = copyout(&wru, uap->wrusage, sizeof(wru)); 825 if (uap->info != NULL && error == 0) 826 error = copyout(&si, uap->info, sizeof(si)); 827 return (error); 828} 829 830/* 831 * Reap the remains of a zombie process and optionally return status and 832 * rusage. Asserts and will release both the proctree_lock and the process 833 * lock as part of its work. 834 */ 835void 836proc_reap(struct thread *td, struct proc *p, int *status, int options) 837{ 838 struct proc *q, *t; 839 840 sx_assert(&proctree_lock, SA_XLOCKED); 841 PROC_LOCK_ASSERT(p, MA_OWNED); 842 PROC_SLOCK_ASSERT(p, MA_OWNED); 843 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 844 845 q = td->td_proc; 846 847 PROC_SUNLOCK(p); 848 td->td_retval[0] = p->p_pid; 849 if (status) 850 *status = p->p_xstat; /* convert to int */ 851 if (options & WNOWAIT) { 852 /* 853 * Only poll, returning the status. Caller does not wish to 854 * release the proc struct just yet. 855 */ 856 PROC_UNLOCK(p); 857 sx_xunlock(&proctree_lock); 858 return; 859 } 860 861 PROC_LOCK(q); 862 sigqueue_take(p->p_ksi); 863 PROC_UNLOCK(q); 864 865 /* 866 * If we got the child via a ptrace 'attach', we need to give it back 867 * to the old parent. 868 */ 869 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 870 PROC_UNLOCK(p); 871 t = proc_realparent(p); 872 PROC_LOCK(t); 873 PROC_LOCK(p); 874 CTR2(KTR_PTRACE, 875 "wait: traced child %d moved back to parent %d", p->p_pid, 876 t->p_pid); 877 proc_reparent(p, t); 878 p->p_oppid = 0; 879 PROC_UNLOCK(p); 880 pksignal(t, SIGCHLD, p->p_ksi); 881 wakeup(t); 882 cv_broadcast(&p->p_pwait); 883 PROC_UNLOCK(t); 884 sx_xunlock(&proctree_lock); 885 return; 886 } 887 p->p_oppid = 0; 888 PROC_UNLOCK(p); 889 890 /* 891 * Remove other references to this process to ensure we have an 892 * exclusive reference. 893 */ 894 sx_xlock(&allproc_lock); 895 LIST_REMOVE(p, p_list); /* off zombproc */ 896 sx_xunlock(&allproc_lock); 897 LIST_REMOVE(p, p_sibling); 898 reaper_abandon_children(p, true); 899 LIST_REMOVE(p, p_reapsibling); 900 PROC_LOCK(p); 901 clear_orphan(p); 902 PROC_UNLOCK(p); 903 leavepgrp(p); 904#ifdef PROCDESC 905 if (p->p_procdesc != NULL) 906 procdesc_reap(p); 907#endif 908 sx_xunlock(&proctree_lock); 909 910 /* 911 * As a side effect of this lock, we know that all other writes to 912 * this proc are visible now, so no more locking is needed for p. 913 */ 914 PROC_LOCK(p); 915 p->p_xstat = 0; /* XXX: why? */ 916 PROC_UNLOCK(p); 917 PROC_LOCK(q); 918 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 919 PROC_UNLOCK(q); 920 921 /* 922 * Decrement the count of procs running with this uid. 923 */ 924 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 925 926 /* 927 * Destroy resource accounting information associated with the process. 928 */ 929#ifdef RACCT 930 if (racct_enable) { 931 PROC_LOCK(p); 932 racct_sub(p, RACCT_NPROC, 1); 933 PROC_UNLOCK(p); 934 } 935#endif 936 racct_proc_exit(p); 937 938 /* 939 * Free credentials, arguments, and sigacts. 940 */ 941 crfree(p->p_ucred); 942 proc_set_cred(p, NULL); 943 pargs_drop(p->p_args); 944 p->p_args = NULL; 945 sigacts_free(p->p_sigacts); 946 p->p_sigacts = NULL; 947 948 /* 949 * Do any thread-system specific cleanups. 950 */ 951 thread_wait(p); 952 953 /* 954 * Give vm and machine-dependent layer a chance to free anything that 955 * cpu_exit couldn't release while still running in process context. 956 */ 957 vm_waitproc(p); 958#ifdef MAC 959 mac_proc_destroy(p); 960#endif 961 KASSERT(FIRST_THREAD_IN_PROC(p), 962 ("proc_reap: no residual thread!")); 963 uma_zfree(proc_zone, p); 964 atomic_add_int(&nprocs, -1); 965} 966 967static int 968proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 969 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 970 int check_only) 971{ 972 struct rusage *rup; 973 974 sx_assert(&proctree_lock, SA_XLOCKED); 975 976 PROC_LOCK(p); 977 978 switch (idtype) { 979 case P_ALL: 980 break; 981 case P_PID: 982 if (p->p_pid != (pid_t)id) { 983 PROC_UNLOCK(p); 984 return (0); 985 } 986 break; 987 case P_PGID: 988 if (p->p_pgid != (pid_t)id) { 989 PROC_UNLOCK(p); 990 return (0); 991 } 992 break; 993 case P_SID: 994 if (p->p_session->s_sid != (pid_t)id) { 995 PROC_UNLOCK(p); 996 return (0); 997 } 998 break; 999 case P_UID: 1000 if (p->p_ucred->cr_uid != (uid_t)id) { 1001 PROC_UNLOCK(p); 1002 return (0); 1003 } 1004 break; 1005 case P_GID: 1006 if (p->p_ucred->cr_gid != (gid_t)id) { 1007 PROC_UNLOCK(p); 1008 return (0); 1009 } 1010 break; 1011 case P_JAILID: 1012 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1013 PROC_UNLOCK(p); 1014 return (0); 1015 } 1016 break; 1017 /* 1018 * It seems that the thread structures get zeroed out 1019 * at process exit. This makes it impossible to 1020 * support P_SETID, P_CID or P_CPUID. 1021 */ 1022 default: 1023 PROC_UNLOCK(p); 1024 return (0); 1025 } 1026 1027 if (p_canwait(td, p)) { 1028 PROC_UNLOCK(p); 1029 return (0); 1030 } 1031 1032 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1033 PROC_UNLOCK(p); 1034 return (0); 1035 } 1036 1037 /* 1038 * This special case handles a kthread spawned by linux_clone 1039 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1040 * functions need to be able to distinguish between waiting 1041 * on a process and waiting on a thread. It is a thread if 1042 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1043 * signifies we want to wait for threads and not processes. 1044 */ 1045 if ((p->p_sigparent != SIGCHLD) ^ 1046 ((options & WLINUXCLONE) != 0)) { 1047 PROC_UNLOCK(p); 1048 return (0); 1049 } 1050 1051 if (siginfo != NULL) { 1052 bzero(siginfo, sizeof(*siginfo)); 1053 siginfo->si_errno = 0; 1054 1055 /* 1056 * SUSv4 requires that the si_signo value is always 1057 * SIGCHLD. Obey it despite the rfork(2) interface 1058 * allows to request other signal for child exit 1059 * notification. 1060 */ 1061 siginfo->si_signo = SIGCHLD; 1062 1063 /* 1064 * This is still a rough estimate. We will fix the 1065 * cases TRAPPED, STOPPED, and CONTINUED later. 1066 */ 1067 if (WCOREDUMP(p->p_xstat)) { 1068 siginfo->si_code = CLD_DUMPED; 1069 siginfo->si_status = WTERMSIG(p->p_xstat); 1070 } else if (WIFSIGNALED(p->p_xstat)) { 1071 siginfo->si_code = CLD_KILLED; 1072 siginfo->si_status = WTERMSIG(p->p_xstat); 1073 } else { 1074 siginfo->si_code = CLD_EXITED; 1075 siginfo->si_status = WEXITSTATUS(p->p_xstat); 1076 } 1077 1078 siginfo->si_pid = p->p_pid; 1079 siginfo->si_uid = p->p_ucred->cr_uid; 1080 1081 /* 1082 * The si_addr field would be useful additional 1083 * detail, but apparently the PC value may be lost 1084 * when we reach this point. bzero() above sets 1085 * siginfo->si_addr to NULL. 1086 */ 1087 } 1088 1089 /* 1090 * There should be no reason to limit resources usage info to 1091 * exited processes only. A snapshot about any resources used 1092 * by a stopped process may be exactly what is needed. 1093 */ 1094 if (wrusage != NULL) { 1095 rup = &wrusage->wru_self; 1096 *rup = p->p_ru; 1097 PROC_STATLOCK(p); 1098 calcru(p, &rup->ru_utime, &rup->ru_stime); 1099 PROC_STATUNLOCK(p); 1100 1101 rup = &wrusage->wru_children; 1102 *rup = p->p_stats->p_cru; 1103 calccru(p, &rup->ru_utime, &rup->ru_stime); 1104 } 1105 1106 if (p->p_state == PRS_ZOMBIE && !check_only) { 1107 PROC_SLOCK(p); 1108 proc_reap(td, p, status, options); 1109 return (-1); 1110 } 1111 PROC_UNLOCK(p); 1112 return (1); 1113} 1114 1115int 1116kern_wait(struct thread *td, pid_t pid, int *status, int options, 1117 struct rusage *rusage) 1118{ 1119 struct __wrusage wru, *wrup; 1120 idtype_t idtype; 1121 id_t id; 1122 int ret; 1123 1124 /* 1125 * Translate the special pid values into the (idtype, pid) 1126 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1127 * kern_wait6() on its own. 1128 */ 1129 if (pid == WAIT_ANY) { 1130 idtype = P_ALL; 1131 id = 0; 1132 } else if (pid < 0) { 1133 idtype = P_PGID; 1134 id = (id_t)-pid; 1135 } else { 1136 idtype = P_PID; 1137 id = (id_t)pid; 1138 } 1139 1140 if (rusage != NULL) 1141 wrup = &wru; 1142 else 1143 wrup = NULL; 1144 1145 /* 1146 * For backward compatibility we implicitly add flags WEXITED 1147 * and WTRAPPED here. 1148 */ 1149 options |= WEXITED | WTRAPPED; 1150 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1151 if (rusage != NULL) 1152 *rusage = wru.wru_self; 1153 return (ret); 1154} 1155 1156int 1157kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1158 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1159{ 1160 struct proc *p, *q; 1161 int error, nfound, ret; 1162 1163 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1164 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1165 AUDIT_ARG_VALUE(options); 1166 1167 q = td->td_proc; 1168 1169 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1170 PROC_LOCK(q); 1171 id = (id_t)q->p_pgid; 1172 PROC_UNLOCK(q); 1173 idtype = P_PGID; 1174 } 1175 1176 /* If we don't know the option, just return. */ 1177 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1178 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1179 return (EINVAL); 1180 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1181 /* 1182 * We will be unable to find any matching processes, 1183 * because there are no known events to look for. 1184 * Prefer to return error instead of blocking 1185 * indefinitely. 1186 */ 1187 return (EINVAL); 1188 } 1189 1190loop: 1191 if (q->p_flag & P_STATCHILD) { 1192 PROC_LOCK(q); 1193 q->p_flag &= ~P_STATCHILD; 1194 PROC_UNLOCK(q); 1195 } 1196 nfound = 0; 1197 sx_xlock(&proctree_lock); 1198 LIST_FOREACH(p, &q->p_children, p_sibling) { 1199 ret = proc_to_reap(td, p, idtype, id, status, options, 1200 wrusage, siginfo, 0); 1201 if (ret == 0) 1202 continue; 1203 else if (ret == 1) 1204 nfound++; 1205 else 1206 return (0); 1207 1208 PROC_LOCK(p); 1209 PROC_SLOCK(p); 1210 1211 if ((options & WTRAPPED) != 0 && 1212 (p->p_flag & P_TRACED) != 0 && 1213 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1214 (p->p_suspcount == p->p_numthreads) && 1215 ((p->p_flag & P_WAITED) == 0)) { 1216 PROC_SUNLOCK(p); 1217 if ((options & WNOWAIT) == 0) 1218 p->p_flag |= P_WAITED; 1219 sx_xunlock(&proctree_lock); 1220 td->td_retval[0] = p->p_pid; 1221 1222 if (status != NULL) 1223 *status = W_STOPCODE(p->p_xstat); 1224 if (siginfo != NULL) { 1225 siginfo->si_status = p->p_xstat; 1226 siginfo->si_code = CLD_TRAPPED; 1227 } 1228 if ((options & WNOWAIT) == 0) { 1229 PROC_LOCK(q); 1230 sigqueue_take(p->p_ksi); 1231 PROC_UNLOCK(q); 1232 } 1233 1234 CTR4(KTR_PTRACE, 1235 "wait: returning trapped pid %d status %#x (xstat %d) xthread %d", 1236 p->p_pid, W_STOPCODE(p->p_xstat), p->p_xstat, 1237 p->p_xthread != NULL ? p->p_xthread->td_tid : -1); 1238 PROC_UNLOCK(p); 1239 return (0); 1240 } 1241 if ((options & WUNTRACED) != 0 && 1242 (p->p_flag & P_STOPPED_SIG) != 0 && 1243 (p->p_suspcount == p->p_numthreads) && 1244 ((p->p_flag & P_WAITED) == 0)) { 1245 PROC_SUNLOCK(p); 1246 if ((options & WNOWAIT) == 0) 1247 p->p_flag |= P_WAITED; 1248 sx_xunlock(&proctree_lock); 1249 td->td_retval[0] = p->p_pid; 1250 1251 if (status != NULL) 1252 *status = W_STOPCODE(p->p_xstat); 1253 if (siginfo != NULL) { 1254 siginfo->si_status = p->p_xstat; 1255 siginfo->si_code = CLD_STOPPED; 1256 } 1257 if ((options & WNOWAIT) == 0) { 1258 PROC_LOCK(q); 1259 sigqueue_take(p->p_ksi); 1260 PROC_UNLOCK(q); 1261 } 1262 1263 PROC_UNLOCK(p); 1264 return (0); 1265 } 1266 PROC_SUNLOCK(p); 1267 if ((options & WCONTINUED) != 0 && 1268 (p->p_flag & P_CONTINUED) != 0) { 1269 sx_xunlock(&proctree_lock); 1270 td->td_retval[0] = p->p_pid; 1271 if ((options & WNOWAIT) == 0) { 1272 p->p_flag &= ~P_CONTINUED; 1273 PROC_LOCK(q); 1274 sigqueue_take(p->p_ksi); 1275 PROC_UNLOCK(q); 1276 } 1277 PROC_UNLOCK(p); 1278 1279 if (status != NULL) 1280 *status = SIGCONT; 1281 if (siginfo != NULL) { 1282 siginfo->si_status = SIGCONT; 1283 siginfo->si_code = CLD_CONTINUED; 1284 } 1285 return (0); 1286 } 1287 PROC_UNLOCK(p); 1288 } 1289 1290 /* 1291 * Look in the orphans list too, to allow the parent to 1292 * collect it's child exit status even if child is being 1293 * debugged. 1294 * 1295 * Debugger detaches from the parent upon successful 1296 * switch-over from parent to child. At this point due to 1297 * re-parenting the parent loses the child to debugger and a 1298 * wait4(2) call would report that it has no children to wait 1299 * for. By maintaining a list of orphans we allow the parent 1300 * to successfully wait until the child becomes a zombie. 1301 */ 1302 if (nfound == 0) { 1303 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1304 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1305 NULL, NULL, 1); 1306 if (ret != 0) { 1307 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1308 (int)td->td_retval[0])); 1309 nfound++; 1310 break; 1311 } 1312 } 1313 } 1314 if (nfound == 0) { 1315 sx_xunlock(&proctree_lock); 1316 return (ECHILD); 1317 } 1318 if (options & WNOHANG) { 1319 sx_xunlock(&proctree_lock); 1320 td->td_retval[0] = 0; 1321 return (0); 1322 } 1323 PROC_LOCK(q); 1324 sx_xunlock(&proctree_lock); 1325 if (q->p_flag & P_STATCHILD) { 1326 q->p_flag &= ~P_STATCHILD; 1327 error = 0; 1328 } else 1329 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1330 PROC_UNLOCK(q); 1331 if (error) 1332 return (error); 1333 goto loop; 1334} 1335 1336/* 1337 * Make process 'parent' the new parent of process 'child'. 1338 * Must be called with an exclusive hold of proctree lock. 1339 */ 1340void 1341proc_reparent(struct proc *child, struct proc *parent) 1342{ 1343 1344 sx_assert(&proctree_lock, SX_XLOCKED); 1345 PROC_LOCK_ASSERT(child, MA_OWNED); 1346 if (child->p_pptr == parent) 1347 return; 1348 1349 PROC_LOCK(child->p_pptr); 1350 sigqueue_take(child->p_ksi); 1351 PROC_UNLOCK(child->p_pptr); 1352 LIST_REMOVE(child, p_sibling); 1353 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1354 1355 clear_orphan(child); 1356 if (child->p_flag & P_TRACED) { 1357 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1358 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1359 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1360 p_orphan); 1361 } else { 1362 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1363 child, p_orphan); 1364 } 1365 child->p_treeflag |= P_TREE_ORPHANED; 1366 } 1367 1368 child->p_pptr = parent; 1369} 1370