kern_exit.c revision 304190
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 304190 2016-08-15 21:44:06Z 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 } 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_flag2 &= ~P2_PTRACE_FSTP; 524 q->p_ptevents = 0; 525 FOREACH_THREAD_IN_PROC(q, tdt) { 526 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 527 TDB_FSTP); 528 } 529 kern_psignal(q, SIGKILL); 530 } 531 PROC_UNLOCK(q); 532 } 533 534 /* 535 * Also get rid of our orphans. 536 */ 537 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 538 PROC_LOCK(q); 539 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 540 q->p_pid); 541 clear_orphan(q); 542 PROC_UNLOCK(q); 543 } 544 545 /* Save exit status. */ 546 PROC_LOCK(p); 547 p->p_xthread = td; 548 549 /* Tell the prison that we are gone. */ 550 prison_proc_free(p->p_ucred->cr_prison); 551 552#ifdef KDTRACE_HOOKS 553 /* 554 * Tell the DTrace fasttrap provider about the exit if it 555 * has declared an interest. 556 */ 557 if (dtrace_fasttrap_exit) 558 dtrace_fasttrap_exit(p); 559#endif 560 561 /* 562 * Notify interested parties of our demise. 563 */ 564 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 565 566#ifdef KDTRACE_HOOKS 567 int reason = CLD_EXITED; 568 if (WCOREDUMP(rv)) 569 reason = CLD_DUMPED; 570 else if (WIFSIGNALED(rv)) 571 reason = CLD_KILLED; 572 SDT_PROBE1(proc, , , exit, reason); 573#endif 574 575 /* 576 * Just delete all entries in the p_klist. At this point we won't 577 * report any more events, and there are nasty race conditions that 578 * can beat us if we don't. 579 */ 580 knlist_clear(&p->p_klist, 1); 581 582 /* 583 * If this is a process with a descriptor, we may not need to deliver 584 * a signal to the parent. proctree_lock is held over 585 * procdesc_exit() to serialize concurrent calls to close() and 586 * exit(). 587 */ 588#ifdef PROCDESC 589 if (p->p_procdesc == NULL || procdesc_exit(p)) { 590#endif 591 /* 592 * Notify parent that we're gone. If parent has the 593 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 594 * notify process 1 instead (and hope it will handle this 595 * situation). 596 */ 597 PROC_LOCK(p->p_pptr); 598 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 599 if (p->p_pptr->p_sigacts->ps_flag & 600 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 601 struct proc *pp; 602 603 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 604 pp = p->p_pptr; 605 PROC_UNLOCK(pp); 606 proc_reparent(p, p->p_reaper); 607 p->p_sigparent = SIGCHLD; 608 PROC_LOCK(p->p_pptr); 609 610 /* 611 * Notify parent, so in case he was wait(2)ing or 612 * executing waitpid(2) with our pid, he will 613 * continue. 614 */ 615 wakeup(pp); 616 } else 617 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 618 619 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 620 childproc_exited(p); 621 else if (p->p_sigparent != 0) { 622 if (p->p_sigparent == SIGCHLD) 623 childproc_exited(p); 624 else /* LINUX thread */ 625 kern_psignal(p->p_pptr, p->p_sigparent); 626 } 627#ifdef PROCDESC 628 } else 629 PROC_LOCK(p->p_pptr); 630#endif 631 sx_xunlock(&proctree_lock); 632 633 /* 634 * The state PRS_ZOMBIE prevents other proesses from sending 635 * signal to the process, to avoid memory leak, we free memory 636 * for signal queue at the time when the state is set. 637 */ 638 sigqueue_flush(&p->p_sigqueue); 639 sigqueue_flush(&td->td_sigqueue); 640 641 /* 642 * We have to wait until after acquiring all locks before 643 * changing p_state. We need to avoid all possible context 644 * switches (including ones from blocking on a mutex) while 645 * marked as a zombie. We also have to set the zombie state 646 * before we release the parent process' proc lock to avoid 647 * a lost wakeup. So, we first call wakeup, then we grab the 648 * sched lock, update the state, and release the parent process' 649 * proc lock. 650 */ 651 wakeup(p->p_pptr); 652 cv_broadcast(&p->p_pwait); 653 sched_exit(p->p_pptr, td); 654 umtx_thread_exit(td); 655 PROC_SLOCK(p); 656 p->p_state = PRS_ZOMBIE; 657 PROC_UNLOCK(p->p_pptr); 658 659 /* 660 * Hopefully no one will try to deliver a signal to the process this 661 * late in the game. 662 */ 663 knlist_destroy(&p->p_klist); 664 665 /* 666 * Save our children's rusage information in our exit rusage. 667 */ 668 PROC_STATLOCK(p); 669 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 670 PROC_STATUNLOCK(p); 671 672 /* 673 * Make sure the scheduler takes this thread out of its tables etc. 674 * This will also release this thread's reference to the ucred. 675 * Other thread parts to release include pcb bits and such. 676 */ 677 thread_exit(); 678} 679 680 681#ifndef _SYS_SYSPROTO_H_ 682struct abort2_args { 683 char *why; 684 int nargs; 685 void **args; 686}; 687#endif 688 689int 690sys_abort2(struct thread *td, struct abort2_args *uap) 691{ 692 struct proc *p = td->td_proc; 693 struct sbuf *sb; 694 void *uargs[16]; 695 int error, i, sig; 696 697 /* 698 * Do it right now so we can log either proper call of abort2(), or 699 * note, that invalid argument was passed. 512 is big enough to 700 * handle 16 arguments' descriptions with additional comments. 701 */ 702 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 703 sbuf_clear(sb); 704 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 705 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 706 /* 707 * Since we can't return from abort2(), send SIGKILL in cases, where 708 * abort2() was called improperly 709 */ 710 sig = SIGKILL; 711 /* Prevent from DoSes from user-space. */ 712 if (uap->nargs < 0 || uap->nargs > 16) 713 goto out; 714 if (uap->nargs > 0) { 715 if (uap->args == NULL) 716 goto out; 717 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 718 if (error != 0) 719 goto out; 720 } 721 /* 722 * Limit size of 'reason' string to 128. Will fit even when 723 * maximal number of arguments was chosen to be logged. 724 */ 725 if (uap->why != NULL) { 726 error = sbuf_copyin(sb, uap->why, 128); 727 if (error < 0) 728 goto out; 729 } else { 730 sbuf_printf(sb, "(null)"); 731 } 732 if (uap->nargs > 0) { 733 sbuf_printf(sb, "("); 734 for (i = 0;i < uap->nargs; i++) 735 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 736 sbuf_printf(sb, ")"); 737 } 738 /* 739 * Final stage: arguments were proper, string has been 740 * successfully copied from userspace, and copying pointers 741 * from user-space succeed. 742 */ 743 sig = SIGABRT; 744out: 745 if (sig == SIGKILL) { 746 sbuf_trim(sb); 747 sbuf_printf(sb, " (Reason text inaccessible)"); 748 } 749 sbuf_cat(sb, "\n"); 750 sbuf_finish(sb); 751 log(LOG_INFO, "%s", sbuf_data(sb)); 752 sbuf_delete(sb); 753 exit1(td, W_EXITCODE(0, sig)); 754 return (0); 755} 756 757 758#ifdef COMPAT_43 759/* 760 * The dirty work is handled by kern_wait(). 761 */ 762int 763owait(struct thread *td, struct owait_args *uap __unused) 764{ 765 int error, status; 766 767 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 768 if (error == 0) 769 td->td_retval[1] = status; 770 return (error); 771} 772#endif /* COMPAT_43 */ 773 774/* 775 * The dirty work is handled by kern_wait(). 776 */ 777int 778sys_wait4(struct thread *td, struct wait4_args *uap) 779{ 780 struct rusage ru, *rup; 781 int error, status; 782 783 if (uap->rusage != NULL) 784 rup = &ru; 785 else 786 rup = NULL; 787 error = kern_wait(td, uap->pid, &status, uap->options, rup); 788 if (uap->status != NULL && error == 0) 789 error = copyout(&status, uap->status, sizeof(status)); 790 if (uap->rusage != NULL && error == 0) 791 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 792 return (error); 793} 794 795int 796sys_wait6(struct thread *td, struct wait6_args *uap) 797{ 798 struct __wrusage wru, *wrup; 799 siginfo_t si, *sip; 800 idtype_t idtype; 801 id_t id; 802 int error, status; 803 804 idtype = uap->idtype; 805 id = uap->id; 806 807 if (uap->wrusage != NULL) 808 wrup = &wru; 809 else 810 wrup = NULL; 811 812 if (uap->info != NULL) { 813 sip = &si; 814 bzero(sip, sizeof(*sip)); 815 } else 816 sip = NULL; 817 818 /* 819 * We expect all callers of wait6() to know about WEXITED and 820 * WTRAPPED. 821 */ 822 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 823 824 if (uap->status != NULL && error == 0) 825 error = copyout(&status, uap->status, sizeof(status)); 826 if (uap->wrusage != NULL && error == 0) 827 error = copyout(&wru, uap->wrusage, sizeof(wru)); 828 if (uap->info != NULL && error == 0) 829 error = copyout(&si, uap->info, sizeof(si)); 830 return (error); 831} 832 833/* 834 * Reap the remains of a zombie process and optionally return status and 835 * rusage. Asserts and will release both the proctree_lock and the process 836 * lock as part of its work. 837 */ 838void 839proc_reap(struct thread *td, struct proc *p, int *status, int options) 840{ 841 struct proc *q, *t; 842 843 sx_assert(&proctree_lock, SA_XLOCKED); 844 PROC_LOCK_ASSERT(p, MA_OWNED); 845 PROC_SLOCK_ASSERT(p, MA_OWNED); 846 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 847 848 q = td->td_proc; 849 850 PROC_SUNLOCK(p); 851 td->td_retval[0] = p->p_pid; 852 if (status) 853 *status = p->p_xstat; /* convert to int */ 854 if (options & WNOWAIT) { 855 /* 856 * Only poll, returning the status. Caller does not wish to 857 * release the proc struct just yet. 858 */ 859 PROC_UNLOCK(p); 860 sx_xunlock(&proctree_lock); 861 return; 862 } 863 864 PROC_LOCK(q); 865 sigqueue_take(p->p_ksi); 866 PROC_UNLOCK(q); 867 868 /* 869 * If we got the child via a ptrace 'attach', we need to give it back 870 * to the old parent. 871 */ 872 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 873 PROC_UNLOCK(p); 874 t = proc_realparent(p); 875 PROC_LOCK(t); 876 PROC_LOCK(p); 877 CTR2(KTR_PTRACE, 878 "wait: traced child %d moved back to parent %d", p->p_pid, 879 t->p_pid); 880 proc_reparent(p, t); 881 p->p_oppid = 0; 882 PROC_UNLOCK(p); 883 pksignal(t, SIGCHLD, p->p_ksi); 884 wakeup(t); 885 cv_broadcast(&p->p_pwait); 886 PROC_UNLOCK(t); 887 sx_xunlock(&proctree_lock); 888 return; 889 } 890 p->p_oppid = 0; 891 PROC_UNLOCK(p); 892 893 /* 894 * Remove other references to this process to ensure we have an 895 * exclusive reference. 896 */ 897 sx_xlock(&allproc_lock); 898 LIST_REMOVE(p, p_list); /* off zombproc */ 899 sx_xunlock(&allproc_lock); 900 LIST_REMOVE(p, p_sibling); 901 reaper_abandon_children(p, true); 902 LIST_REMOVE(p, p_reapsibling); 903 PROC_LOCK(p); 904 clear_orphan(p); 905 PROC_UNLOCK(p); 906 leavepgrp(p); 907#ifdef PROCDESC 908 if (p->p_procdesc != NULL) 909 procdesc_reap(p); 910#endif 911 sx_xunlock(&proctree_lock); 912 913 /* 914 * As a side effect of this lock, we know that all other writes to 915 * this proc are visible now, so no more locking is needed for p. 916 */ 917 PROC_LOCK(p); 918 p->p_xstat = 0; /* XXX: why? */ 919 PROC_UNLOCK(p); 920 PROC_LOCK(q); 921 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 922 PROC_UNLOCK(q); 923 924 /* 925 * Decrement the count of procs running with this uid. 926 */ 927 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 928 929 /* 930 * Destroy resource accounting information associated with the process. 931 */ 932#ifdef RACCT 933 if (racct_enable) { 934 PROC_LOCK(p); 935 racct_sub(p, RACCT_NPROC, 1); 936 PROC_UNLOCK(p); 937 } 938#endif 939 racct_proc_exit(p); 940 941 /* 942 * Free credentials, arguments, and sigacts. 943 */ 944 crfree(p->p_ucred); 945 proc_set_cred(p, NULL); 946 pargs_drop(p->p_args); 947 p->p_args = NULL; 948 sigacts_free(p->p_sigacts); 949 p->p_sigacts = NULL; 950 951 /* 952 * Do any thread-system specific cleanups. 953 */ 954 thread_wait(p); 955 956 /* 957 * Give vm and machine-dependent layer a chance to free anything that 958 * cpu_exit couldn't release while still running in process context. 959 */ 960 vm_waitproc(p); 961#ifdef MAC 962 mac_proc_destroy(p); 963#endif 964 KASSERT(FIRST_THREAD_IN_PROC(p), 965 ("proc_reap: no residual thread!")); 966 uma_zfree(proc_zone, p); 967 atomic_add_int(&nprocs, -1); 968} 969 970static int 971proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 972 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 973 int check_only) 974{ 975 struct rusage *rup; 976 977 sx_assert(&proctree_lock, SA_XLOCKED); 978 979 PROC_LOCK(p); 980 981 switch (idtype) { 982 case P_ALL: 983 break; 984 case P_PID: 985 if (p->p_pid != (pid_t)id) { 986 PROC_UNLOCK(p); 987 return (0); 988 } 989 break; 990 case P_PGID: 991 if (p->p_pgid != (pid_t)id) { 992 PROC_UNLOCK(p); 993 return (0); 994 } 995 break; 996 case P_SID: 997 if (p->p_session->s_sid != (pid_t)id) { 998 PROC_UNLOCK(p); 999 return (0); 1000 } 1001 break; 1002 case P_UID: 1003 if (p->p_ucred->cr_uid != (uid_t)id) { 1004 PROC_UNLOCK(p); 1005 return (0); 1006 } 1007 break; 1008 case P_GID: 1009 if (p->p_ucred->cr_gid != (gid_t)id) { 1010 PROC_UNLOCK(p); 1011 return (0); 1012 } 1013 break; 1014 case P_JAILID: 1015 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1016 PROC_UNLOCK(p); 1017 return (0); 1018 } 1019 break; 1020 /* 1021 * It seems that the thread structures get zeroed out 1022 * at process exit. This makes it impossible to 1023 * support P_SETID, P_CID or P_CPUID. 1024 */ 1025 default: 1026 PROC_UNLOCK(p); 1027 return (0); 1028 } 1029 1030 if (p_canwait(td, p)) { 1031 PROC_UNLOCK(p); 1032 return (0); 1033 } 1034 1035 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1036 PROC_UNLOCK(p); 1037 return (0); 1038 } 1039 1040 /* 1041 * This special case handles a kthread spawned by linux_clone 1042 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1043 * functions need to be able to distinguish between waiting 1044 * on a process and waiting on a thread. It is a thread if 1045 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1046 * signifies we want to wait for threads and not processes. 1047 */ 1048 if ((p->p_sigparent != SIGCHLD) ^ 1049 ((options & WLINUXCLONE) != 0)) { 1050 PROC_UNLOCK(p); 1051 return (0); 1052 } 1053 1054 if (siginfo != NULL) { 1055 bzero(siginfo, sizeof(*siginfo)); 1056 siginfo->si_errno = 0; 1057 1058 /* 1059 * SUSv4 requires that the si_signo value is always 1060 * SIGCHLD. Obey it despite the rfork(2) interface 1061 * allows to request other signal for child exit 1062 * notification. 1063 */ 1064 siginfo->si_signo = SIGCHLD; 1065 1066 /* 1067 * This is still a rough estimate. We will fix the 1068 * cases TRAPPED, STOPPED, and CONTINUED later. 1069 */ 1070 if (WCOREDUMP(p->p_xstat)) { 1071 siginfo->si_code = CLD_DUMPED; 1072 siginfo->si_status = WTERMSIG(p->p_xstat); 1073 } else if (WIFSIGNALED(p->p_xstat)) { 1074 siginfo->si_code = CLD_KILLED; 1075 siginfo->si_status = WTERMSIG(p->p_xstat); 1076 } else { 1077 siginfo->si_code = CLD_EXITED; 1078 siginfo->si_status = WEXITSTATUS(p->p_xstat); 1079 } 1080 1081 siginfo->si_pid = p->p_pid; 1082 siginfo->si_uid = p->p_ucred->cr_uid; 1083 1084 /* 1085 * The si_addr field would be useful additional 1086 * detail, but apparently the PC value may be lost 1087 * when we reach this point. bzero() above sets 1088 * siginfo->si_addr to NULL. 1089 */ 1090 } 1091 1092 /* 1093 * There should be no reason to limit resources usage info to 1094 * exited processes only. A snapshot about any resources used 1095 * by a stopped process may be exactly what is needed. 1096 */ 1097 if (wrusage != NULL) { 1098 rup = &wrusage->wru_self; 1099 *rup = p->p_ru; 1100 PROC_STATLOCK(p); 1101 calcru(p, &rup->ru_utime, &rup->ru_stime); 1102 PROC_STATUNLOCK(p); 1103 1104 rup = &wrusage->wru_children; 1105 *rup = p->p_stats->p_cru; 1106 calccru(p, &rup->ru_utime, &rup->ru_stime); 1107 } 1108 1109 if (p->p_state == PRS_ZOMBIE && !check_only) { 1110 PROC_SLOCK(p); 1111 proc_reap(td, p, status, options); 1112 return (-1); 1113 } 1114 PROC_UNLOCK(p); 1115 return (1); 1116} 1117 1118int 1119kern_wait(struct thread *td, pid_t pid, int *status, int options, 1120 struct rusage *rusage) 1121{ 1122 struct __wrusage wru, *wrup; 1123 idtype_t idtype; 1124 id_t id; 1125 int ret; 1126 1127 /* 1128 * Translate the special pid values into the (idtype, pid) 1129 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1130 * kern_wait6() on its own. 1131 */ 1132 if (pid == WAIT_ANY) { 1133 idtype = P_ALL; 1134 id = 0; 1135 } else if (pid < 0) { 1136 idtype = P_PGID; 1137 id = (id_t)-pid; 1138 } else { 1139 idtype = P_PID; 1140 id = (id_t)pid; 1141 } 1142 1143 if (rusage != NULL) 1144 wrup = &wru; 1145 else 1146 wrup = NULL; 1147 1148 /* 1149 * For backward compatibility we implicitly add flags WEXITED 1150 * and WTRAPPED here. 1151 */ 1152 options |= WEXITED | WTRAPPED; 1153 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1154 if (rusage != NULL) 1155 *rusage = wru.wru_self; 1156 return (ret); 1157} 1158 1159int 1160kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1161 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1162{ 1163 struct proc *p, *q; 1164 int error, nfound, ret; 1165 1166 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1167 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1168 AUDIT_ARG_VALUE(options); 1169 1170 q = td->td_proc; 1171 1172 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1173 PROC_LOCK(q); 1174 id = (id_t)q->p_pgid; 1175 PROC_UNLOCK(q); 1176 idtype = P_PGID; 1177 } 1178 1179 /* If we don't know the option, just return. */ 1180 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1181 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1182 return (EINVAL); 1183 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1184 /* 1185 * We will be unable to find any matching processes, 1186 * because there are no known events to look for. 1187 * Prefer to return error instead of blocking 1188 * indefinitely. 1189 */ 1190 return (EINVAL); 1191 } 1192 1193loop: 1194 if (q->p_flag & P_STATCHILD) { 1195 PROC_LOCK(q); 1196 q->p_flag &= ~P_STATCHILD; 1197 PROC_UNLOCK(q); 1198 } 1199 nfound = 0; 1200 sx_xlock(&proctree_lock); 1201 LIST_FOREACH(p, &q->p_children, p_sibling) { 1202 ret = proc_to_reap(td, p, idtype, id, status, options, 1203 wrusage, siginfo, 0); 1204 if (ret == 0) 1205 continue; 1206 else if (ret == 1) 1207 nfound++; 1208 else 1209 return (0); 1210 1211 PROC_LOCK(p); 1212 PROC_SLOCK(p); 1213 1214 if ((options & WTRAPPED) != 0 && 1215 (p->p_flag & P_TRACED) != 0 && 1216 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1217 (p->p_suspcount == p->p_numthreads) && 1218 ((p->p_flag & P_WAITED) == 0)) { 1219 PROC_SUNLOCK(p); 1220 if ((options & WNOWAIT) == 0) 1221 p->p_flag |= P_WAITED; 1222 sx_xunlock(&proctree_lock); 1223 td->td_retval[0] = p->p_pid; 1224 1225 if (status != NULL) 1226 *status = W_STOPCODE(p->p_xstat); 1227 if (siginfo != NULL) { 1228 siginfo->si_status = p->p_xstat; 1229 siginfo->si_code = CLD_TRAPPED; 1230 } 1231 if ((options & WNOWAIT) == 0) { 1232 PROC_LOCK(q); 1233 sigqueue_take(p->p_ksi); 1234 PROC_UNLOCK(q); 1235 } 1236 1237 CTR4(KTR_PTRACE, 1238 "wait: returning trapped pid %d status %#x (xstat %d) xthread %d", 1239 p->p_pid, W_STOPCODE(p->p_xstat), p->p_xstat, 1240 p->p_xthread != NULL ? p->p_xthread->td_tid : -1); 1241 PROC_UNLOCK(p); 1242 return (0); 1243 } 1244 if ((options & WUNTRACED) != 0 && 1245 (p->p_flag & P_STOPPED_SIG) != 0 && 1246 (p->p_suspcount == p->p_numthreads) && 1247 ((p->p_flag & P_WAITED) == 0)) { 1248 PROC_SUNLOCK(p); 1249 if ((options & WNOWAIT) == 0) 1250 p->p_flag |= P_WAITED; 1251 sx_xunlock(&proctree_lock); 1252 td->td_retval[0] = p->p_pid; 1253 1254 if (status != NULL) 1255 *status = W_STOPCODE(p->p_xstat); 1256 if (siginfo != NULL) { 1257 siginfo->si_status = p->p_xstat; 1258 siginfo->si_code = CLD_STOPPED; 1259 } 1260 if ((options & WNOWAIT) == 0) { 1261 PROC_LOCK(q); 1262 sigqueue_take(p->p_ksi); 1263 PROC_UNLOCK(q); 1264 } 1265 1266 PROC_UNLOCK(p); 1267 return (0); 1268 } 1269 PROC_SUNLOCK(p); 1270 if ((options & WCONTINUED) != 0 && 1271 (p->p_flag & P_CONTINUED) != 0) { 1272 sx_xunlock(&proctree_lock); 1273 td->td_retval[0] = p->p_pid; 1274 if ((options & WNOWAIT) == 0) { 1275 p->p_flag &= ~P_CONTINUED; 1276 PROC_LOCK(q); 1277 sigqueue_take(p->p_ksi); 1278 PROC_UNLOCK(q); 1279 } 1280 PROC_UNLOCK(p); 1281 1282 if (status != NULL) 1283 *status = SIGCONT; 1284 if (siginfo != NULL) { 1285 siginfo->si_status = SIGCONT; 1286 siginfo->si_code = CLD_CONTINUED; 1287 } 1288 return (0); 1289 } 1290 PROC_UNLOCK(p); 1291 } 1292 1293 /* 1294 * Look in the orphans list too, to allow the parent to 1295 * collect it's child exit status even if child is being 1296 * debugged. 1297 * 1298 * Debugger detaches from the parent upon successful 1299 * switch-over from parent to child. At this point due to 1300 * re-parenting the parent loses the child to debugger and a 1301 * wait4(2) call would report that it has no children to wait 1302 * for. By maintaining a list of orphans we allow the parent 1303 * to successfully wait until the child becomes a zombie. 1304 */ 1305 if (nfound == 0) { 1306 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1307 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1308 NULL, NULL, 1); 1309 if (ret != 0) { 1310 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1311 (int)td->td_retval[0])); 1312 nfound++; 1313 break; 1314 } 1315 } 1316 } 1317 if (nfound == 0) { 1318 sx_xunlock(&proctree_lock); 1319 return (ECHILD); 1320 } 1321 if (options & WNOHANG) { 1322 sx_xunlock(&proctree_lock); 1323 td->td_retval[0] = 0; 1324 return (0); 1325 } 1326 PROC_LOCK(q); 1327 sx_xunlock(&proctree_lock); 1328 if (q->p_flag & P_STATCHILD) { 1329 q->p_flag &= ~P_STATCHILD; 1330 error = 0; 1331 } else 1332 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1333 PROC_UNLOCK(q); 1334 if (error) 1335 return (error); 1336 goto loop; 1337} 1338 1339/* 1340 * Make process 'parent' the new parent of process 'child'. 1341 * Must be called with an exclusive hold of proctree lock. 1342 */ 1343void 1344proc_reparent(struct proc *child, struct proc *parent) 1345{ 1346 1347 sx_assert(&proctree_lock, SX_XLOCKED); 1348 PROC_LOCK_ASSERT(child, MA_OWNED); 1349 if (child->p_pptr == parent) 1350 return; 1351 1352 PROC_LOCK(child->p_pptr); 1353 sigqueue_take(child->p_ksi); 1354 PROC_UNLOCK(child->p_pptr); 1355 LIST_REMOVE(child, p_sibling); 1356 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1357 1358 clear_orphan(child); 1359 if (child->p_flag & P_TRACED) { 1360 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1361 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1362 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1363 p_orphan); 1364 } else { 1365 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1366 child, p_orphan); 1367 } 1368 child->p_treeflag |= P_TREE_ORPHANED; 1369 } 1370 1371 child->p_pptr = parent; 1372} 1373