kern_exit.c revision 284021
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 284021 2015-06-05 08:36:25Z 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, 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 * MUST abort all other threads before proceeding past here. 211 */ 212 PROC_LOCK(p); 213 /* 214 * First check if some other thread or external request got 215 * here before us. If so, act appropriately: exit or suspend. 216 * We must ensure that stop requests are handled before we set 217 * P_WEXIT. 218 */ 219 thread_suspend_check(0); 220 while (p->p_flag & P_HADTHREADS) { 221 /* 222 * Kill off the other threads. This requires 223 * some co-operation from other parts of the kernel 224 * so it may not be instantaneous. With this state set 225 * any thread entering the kernel from userspace will 226 * thread_exit() in trap(). Any thread attempting to 227 * sleep will return immediately with EINTR or EWOULDBLOCK 228 * which will hopefully force them to back out to userland 229 * freeing resources as they go. Any thread attempting 230 * to return to userland will thread_exit() from userret(). 231 * thread_exit() will unsuspend us when the last of the 232 * other threads exits. 233 * If there is already a thread singler after resumption, 234 * calling thread_single will fail; in that case, we just 235 * re-check all suspension request, the thread should 236 * either be suspended there or exit. 237 */ 238 if (!thread_single(p, SINGLE_EXIT)) 239 /* 240 * All other activity in this process is now 241 * stopped. Threading support has been turned 242 * off. 243 */ 244 break; 245 /* 246 * Recheck for new stop or suspend requests which 247 * might appear while process lock was dropped in 248 * thread_single(). 249 */ 250 thread_suspend_check(0); 251 } 252 KASSERT(p->p_numthreads == 1, 253 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 254 racct_sub(p, RACCT_NTHR, 1); 255 /* 256 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 257 * on our vmspace, so we should block below until they have 258 * released their reference to us. Note that if they have 259 * requested S_EXIT stops we will block here until they ack 260 * via PIOCCONT. 261 */ 262 _STOPEVENT(p, S_EXIT, rv); 263 264 /* 265 * Ignore any pending request to stop due to a stop signal. 266 * Once P_WEXIT is set, future requests will be ignored as 267 * well. 268 */ 269 p->p_flag &= ~P_STOPPED_SIG; 270 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 271 272 /* 273 * Note that we are exiting and do another wakeup of anyone in 274 * PIOCWAIT in case they aren't listening for S_EXIT stops or 275 * decided to wait again after we told them we are exiting. 276 */ 277 p->p_flag |= P_WEXIT; 278 wakeup(&p->p_stype); 279 280 /* 281 * Wait for any processes that have a hold on our vmspace to 282 * release their reference. 283 */ 284 while (p->p_lock > 0) 285 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 286 287 p->p_xstat = rv; /* Let event handler change exit status */ 288 PROC_UNLOCK(p); 289 /* Drain the limit callout while we don't have the proc locked */ 290 callout_drain(&p->p_limco); 291 292#ifdef AUDIT 293 /* 294 * The Sun BSM exit token contains two components: an exit status as 295 * passed to exit(), and a return value to indicate what sort of exit 296 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 297 * what the return value is. 298 */ 299 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0); 300 AUDIT_SYSCALL_EXIT(0, td); 301#endif 302 303 /* Are we a task leader? */ 304 if (p == p->p_leader) { 305 mtx_lock(&ppeers_lock); 306 q = p->p_peers; 307 while (q != NULL) { 308 PROC_LOCK(q); 309 kern_psignal(q, SIGKILL); 310 PROC_UNLOCK(q); 311 q = q->p_peers; 312 } 313 while (p->p_peers != NULL) 314 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 315 mtx_unlock(&ppeers_lock); 316 } 317 318 /* 319 * Check if any loadable modules need anything done at process exit. 320 * E.g. SYSV IPC stuff 321 * XXX what if one of these generates an error? 322 */ 323 EVENTHANDLER_INVOKE(process_exit, p); 324 325 /* 326 * If parent is waiting for us to exit or exec, 327 * P_PPWAIT is set; we will wakeup the parent below. 328 */ 329 PROC_LOCK(p); 330 rv = p->p_xstat; /* Event handler could change exit status */ 331 stopprofclock(p); 332 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 333 334 /* 335 * Stop the real interval timer. If the handler is currently 336 * executing, prevent it from rearming itself and let it finish. 337 */ 338 if (timevalisset(&p->p_realtimer.it_value) && 339 callout_stop(&p->p_itcallout) == 0) { 340 timevalclear(&p->p_realtimer.it_interval); 341 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 342 KASSERT(!timevalisset(&p->p_realtimer.it_value), 343 ("realtime timer is still armed")); 344 } 345 PROC_UNLOCK(p); 346 347 /* 348 * Reset any sigio structures pointing to us as a result of 349 * F_SETOWN with our pid. 350 */ 351 funsetownlst(&p->p_sigiolst); 352 353 /* 354 * If this process has an nlminfo data area (for lockd), release it 355 */ 356 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 357 (*nlminfo_release_p)(p); 358 359 /* 360 * Close open files and release open-file table. 361 * This may block! 362 */ 363 fdescfree(td); 364 365 /* 366 * If this thread tickled GEOM, we need to wait for the giggling to 367 * stop before we return to userland 368 */ 369 if (td->td_pflags & TDP_GEOM) 370 g_waitidle(); 371 372 /* 373 * Remove ourself from our leader's peer list and wake our leader. 374 */ 375 mtx_lock(&ppeers_lock); 376 if (p->p_leader->p_peers) { 377 q = p->p_leader; 378 while (q->p_peers != p) 379 q = q->p_peers; 380 q->p_peers = p->p_peers; 381 wakeup(p->p_leader); 382 } 383 mtx_unlock(&ppeers_lock); 384 385 vmspace_exit(td); 386 387 sx_xlock(&proctree_lock); 388 if (SESS_LEADER(p)) { 389 struct session *sp = p->p_session; 390 struct tty *tp; 391 392 /* 393 * s_ttyp is not zero'd; we use this to indicate that 394 * the session once had a controlling terminal. (for 395 * logging and informational purposes) 396 */ 397 SESS_LOCK(sp); 398 ttyvp = sp->s_ttyvp; 399 tp = sp->s_ttyp; 400 sp->s_ttyvp = NULL; 401 sp->s_ttydp = NULL; 402 sp->s_leader = NULL; 403 SESS_UNLOCK(sp); 404 405 /* 406 * Signal foreground pgrp and revoke access to 407 * controlling terminal if it has not been revoked 408 * already. 409 * 410 * Because the TTY may have been revoked in the mean 411 * time and could already have a new session associated 412 * with it, make sure we don't send a SIGHUP to a 413 * foreground process group that does not belong to this 414 * session. 415 */ 416 417 if (tp != NULL) { 418 tty_lock(tp); 419 if (tp->t_session == sp) 420 tty_signal_pgrp(tp, SIGHUP); 421 tty_unlock(tp); 422 } 423 424 if (ttyvp != NULL) { 425 sx_xunlock(&proctree_lock); 426 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 427 VOP_REVOKE(ttyvp, REVOKEALL); 428 VOP_UNLOCK(ttyvp, 0); 429 } 430 sx_xlock(&proctree_lock); 431 } 432 } 433 fixjobc(p, p->p_pgrp, 0); 434 sx_xunlock(&proctree_lock); 435 (void)acct_process(td); 436 437 /* Release the TTY now we've unlocked everything. */ 438 if (ttyvp != NULL) 439 vrele(ttyvp); 440#ifdef KTRACE 441 ktrprocexit(td); 442#endif 443 /* 444 * Release reference to text vnode 445 */ 446 if (p->p_textvp != NULL) { 447 vrele(p->p_textvp); 448 p->p_textvp = NULL; 449 } 450 451 /* 452 * Release our limits structure. 453 */ 454 lim_free(p->p_limit); 455 p->p_limit = NULL; 456 457 tidhash_remove(td); 458 459 /* 460 * Remove proc from allproc queue and pidhash chain. 461 * Place onto zombproc. Unlink from parent's child list. 462 */ 463 sx_xlock(&allproc_lock); 464 LIST_REMOVE(p, p_list); 465 LIST_INSERT_HEAD(&zombproc, p, p_list); 466 LIST_REMOVE(p, p_hash); 467 sx_xunlock(&allproc_lock); 468 469 /* 470 * Call machine-dependent code to release any 471 * machine-dependent resources other than the address space. 472 * The address space is released by "vmspace_exitfree(p)" in 473 * vm_waitproc(). 474 */ 475 cpu_exit(td); 476 477 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 478 479 /* 480 * Reparent all children processes: 481 * - traced ones to the original parent (or init if we are that parent) 482 * - the rest to init 483 */ 484 sx_xlock(&proctree_lock); 485 q = LIST_FIRST(&p->p_children); 486 if (q != NULL) /* only need this if any child is S_ZOMB */ 487 wakeup(q->p_reaper); 488 for (; q != NULL; q = nq) { 489 nq = LIST_NEXT(q, p_sibling); 490 PROC_LOCK(q); 491 q->p_sigparent = SIGCHLD; 492 493 if (!(q->p_flag & P_TRACED)) { 494 proc_reparent(q, q->p_reaper); 495 } else { 496 /* 497 * Traced processes are killed since their existence 498 * means someone is screwing up. 499 */ 500 t = proc_realparent(q); 501 if (t == p) { 502 proc_reparent(q, q->p_reaper); 503 } else { 504 PROC_LOCK(t); 505 proc_reparent(q, t); 506 PROC_UNLOCK(t); 507 } 508 /* 509 * Since q was found on our children list, the 510 * proc_reparent() call moved q to the orphan 511 * list due to present P_TRACED flag. Clear 512 * orphan link for q now while q is locked. 513 */ 514 clear_orphan(q); 515 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 516 FOREACH_THREAD_IN_PROC(q, tdt) 517 tdt->td_dbgflags &= ~TDB_SUSPEND; 518 kern_psignal(q, SIGKILL); 519 } 520 PROC_UNLOCK(q); 521 } 522 523 /* 524 * Also get rid of our orphans. 525 */ 526 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 527 PROC_LOCK(q); 528 clear_orphan(q); 529 PROC_UNLOCK(q); 530 } 531 532 /* Save exit status. */ 533 PROC_LOCK(p); 534 p->p_xthread = td; 535 536 /* Tell the prison that we are gone. */ 537 prison_proc_free(p->p_ucred->cr_prison); 538 539#ifdef KDTRACE_HOOKS 540 /* 541 * Tell the DTrace fasttrap provider about the exit if it 542 * has declared an interest. 543 */ 544 if (dtrace_fasttrap_exit) 545 dtrace_fasttrap_exit(p); 546#endif 547 548 /* 549 * Notify interested parties of our demise. 550 */ 551 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 552 553#ifdef KDTRACE_HOOKS 554 int reason = CLD_EXITED; 555 if (WCOREDUMP(rv)) 556 reason = CLD_DUMPED; 557 else if (WIFSIGNALED(rv)) 558 reason = CLD_KILLED; 559 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 560#endif 561 562 /* 563 * Just delete all entries in the p_klist. At this point we won't 564 * report any more events, and there are nasty race conditions that 565 * can beat us if we don't. 566 */ 567 knlist_clear(&p->p_klist, 1); 568 569 /* 570 * If this is a process with a descriptor, we may not need to deliver 571 * a signal to the parent. proctree_lock is held over 572 * procdesc_exit() to serialize concurrent calls to close() and 573 * exit(). 574 */ 575#ifdef PROCDESC 576 if (p->p_procdesc == NULL || procdesc_exit(p)) { 577#endif 578 /* 579 * Notify parent that we're gone. If parent has the 580 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 581 * notify process 1 instead (and hope it will handle this 582 * situation). 583 */ 584 PROC_LOCK(p->p_pptr); 585 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 586 if (p->p_pptr->p_sigacts->ps_flag & 587 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 588 struct proc *pp; 589 590 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 591 pp = p->p_pptr; 592 PROC_UNLOCK(pp); 593 proc_reparent(p, p->p_reaper); 594 p->p_sigparent = SIGCHLD; 595 PROC_LOCK(p->p_pptr); 596 597 /* 598 * Notify parent, so in case he was wait(2)ing or 599 * executing waitpid(2) with our pid, he will 600 * continue. 601 */ 602 wakeup(pp); 603 } else 604 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 605 606 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 607 childproc_exited(p); 608 else if (p->p_sigparent != 0) { 609 if (p->p_sigparent == SIGCHLD) 610 childproc_exited(p); 611 else /* LINUX thread */ 612 kern_psignal(p->p_pptr, p->p_sigparent); 613 } 614#ifdef PROCDESC 615 } else 616 PROC_LOCK(p->p_pptr); 617#endif 618 sx_xunlock(&proctree_lock); 619 620 /* 621 * The state PRS_ZOMBIE prevents other proesses from sending 622 * signal to the process, to avoid memory leak, we free memory 623 * for signal queue at the time when the state is set. 624 */ 625 sigqueue_flush(&p->p_sigqueue); 626 sigqueue_flush(&td->td_sigqueue); 627 628 /* 629 * We have to wait until after acquiring all locks before 630 * changing p_state. We need to avoid all possible context 631 * switches (including ones from blocking on a mutex) while 632 * marked as a zombie. We also have to set the zombie state 633 * before we release the parent process' proc lock to avoid 634 * a lost wakeup. So, we first call wakeup, then we grab the 635 * sched lock, update the state, and release the parent process' 636 * proc lock. 637 */ 638 wakeup(p->p_pptr); 639 cv_broadcast(&p->p_pwait); 640 sched_exit(p->p_pptr, td); 641 umtx_thread_exit(td); 642 PROC_SLOCK(p); 643 p->p_state = PRS_ZOMBIE; 644 PROC_UNLOCK(p->p_pptr); 645 646 /* 647 * Hopefully no one will try to deliver a signal to the process this 648 * late in the game. 649 */ 650 knlist_destroy(&p->p_klist); 651 652 /* 653 * Save our children's rusage information in our exit rusage. 654 */ 655 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 656 657 /* 658 * Make sure the scheduler takes this thread out of its tables etc. 659 * This will also release this thread's reference to the ucred. 660 * Other thread parts to release include pcb bits and such. 661 */ 662 thread_exit(); 663} 664 665 666#ifndef _SYS_SYSPROTO_H_ 667struct abort2_args { 668 char *why; 669 int nargs; 670 void **args; 671}; 672#endif 673 674int 675sys_abort2(struct thread *td, struct abort2_args *uap) 676{ 677 struct proc *p = td->td_proc; 678 struct sbuf *sb; 679 void *uargs[16]; 680 int error, i, sig; 681 682 /* 683 * Do it right now so we can log either proper call of abort2(), or 684 * note, that invalid argument was passed. 512 is big enough to 685 * handle 16 arguments' descriptions with additional comments. 686 */ 687 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 688 sbuf_clear(sb); 689 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 690 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 691 /* 692 * Since we can't return from abort2(), send SIGKILL in cases, where 693 * abort2() was called improperly 694 */ 695 sig = SIGKILL; 696 /* Prevent from DoSes from user-space. */ 697 if (uap->nargs < 0 || uap->nargs > 16) 698 goto out; 699 if (uap->nargs > 0) { 700 if (uap->args == NULL) 701 goto out; 702 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 703 if (error != 0) 704 goto out; 705 } 706 /* 707 * Limit size of 'reason' string to 128. Will fit even when 708 * maximal number of arguments was chosen to be logged. 709 */ 710 if (uap->why != NULL) { 711 error = sbuf_copyin(sb, uap->why, 128); 712 if (error < 0) 713 goto out; 714 } else { 715 sbuf_printf(sb, "(null)"); 716 } 717 if (uap->nargs > 0) { 718 sbuf_printf(sb, "("); 719 for (i = 0;i < uap->nargs; i++) 720 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 721 sbuf_printf(sb, ")"); 722 } 723 /* 724 * Final stage: arguments were proper, string has been 725 * successfully copied from userspace, and copying pointers 726 * from user-space succeed. 727 */ 728 sig = SIGABRT; 729out: 730 if (sig == SIGKILL) { 731 sbuf_trim(sb); 732 sbuf_printf(sb, " (Reason text inaccessible)"); 733 } 734 sbuf_cat(sb, "\n"); 735 sbuf_finish(sb); 736 log(LOG_INFO, "%s", sbuf_data(sb)); 737 sbuf_delete(sb); 738 exit1(td, W_EXITCODE(0, sig)); 739 return (0); 740} 741 742 743#ifdef COMPAT_43 744/* 745 * The dirty work is handled by kern_wait(). 746 */ 747int 748owait(struct thread *td, struct owait_args *uap __unused) 749{ 750 int error, status; 751 752 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 753 if (error == 0) 754 td->td_retval[1] = status; 755 return (error); 756} 757#endif /* COMPAT_43 */ 758 759/* 760 * The dirty work is handled by kern_wait(). 761 */ 762int 763sys_wait4(struct thread *td, struct wait4_args *uap) 764{ 765 struct rusage ru, *rup; 766 int error, status; 767 768 if (uap->rusage != NULL) 769 rup = &ru; 770 else 771 rup = NULL; 772 error = kern_wait(td, uap->pid, &status, uap->options, rup); 773 if (uap->status != NULL && error == 0) 774 error = copyout(&status, uap->status, sizeof(status)); 775 if (uap->rusage != NULL && error == 0) 776 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 777 return (error); 778} 779 780int 781sys_wait6(struct thread *td, struct wait6_args *uap) 782{ 783 struct __wrusage wru, *wrup; 784 siginfo_t si, *sip; 785 idtype_t idtype; 786 id_t id; 787 int error, status; 788 789 idtype = uap->idtype; 790 id = uap->id; 791 792 if (uap->wrusage != NULL) 793 wrup = &wru; 794 else 795 wrup = NULL; 796 797 if (uap->info != NULL) { 798 sip = &si; 799 bzero(sip, sizeof(*sip)); 800 } else 801 sip = NULL; 802 803 /* 804 * We expect all callers of wait6() to know about WEXITED and 805 * WTRAPPED. 806 */ 807 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 808 809 if (uap->status != NULL && error == 0) 810 error = copyout(&status, uap->status, sizeof(status)); 811 if (uap->wrusage != NULL && error == 0) 812 error = copyout(&wru, uap->wrusage, sizeof(wru)); 813 if (uap->info != NULL && error == 0) 814 error = copyout(&si, uap->info, sizeof(si)); 815 return (error); 816} 817 818/* 819 * Reap the remains of a zombie process and optionally return status and 820 * rusage. Asserts and will release both the proctree_lock and the process 821 * lock as part of its work. 822 */ 823void 824proc_reap(struct thread *td, struct proc *p, int *status, int options) 825{ 826 struct proc *q, *t; 827 828 sx_assert(&proctree_lock, SA_XLOCKED); 829 PROC_LOCK_ASSERT(p, MA_OWNED); 830 PROC_SLOCK_ASSERT(p, MA_OWNED); 831 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 832 833 q = td->td_proc; 834 835 PROC_SUNLOCK(p); 836 td->td_retval[0] = p->p_pid; 837 if (status) 838 *status = p->p_xstat; /* convert to int */ 839 if (options & WNOWAIT) { 840 /* 841 * Only poll, returning the status. Caller does not wish to 842 * release the proc struct just yet. 843 */ 844 PROC_UNLOCK(p); 845 sx_xunlock(&proctree_lock); 846 return; 847 } 848 849 PROC_LOCK(q); 850 sigqueue_take(p->p_ksi); 851 PROC_UNLOCK(q); 852 PROC_UNLOCK(p); 853 854 /* 855 * If we got the child via a ptrace 'attach', we need to give it back 856 * to the old parent. 857 */ 858 if (p->p_oppid != 0) { 859 t = proc_realparent(p); 860 PROC_LOCK(t); 861 PROC_LOCK(p); 862 proc_reparent(p, t); 863 p->p_oppid = 0; 864 PROC_UNLOCK(p); 865 pksignal(t, SIGCHLD, p->p_ksi); 866 wakeup(t); 867 cv_broadcast(&p->p_pwait); 868 PROC_UNLOCK(t); 869 sx_xunlock(&proctree_lock); 870 return; 871 } 872 873 /* 874 * Remove other references to this process to ensure we have an 875 * exclusive reference. 876 */ 877 sx_xlock(&allproc_lock); 878 LIST_REMOVE(p, p_list); /* off zombproc */ 879 sx_xunlock(&allproc_lock); 880 LIST_REMOVE(p, p_sibling); 881 reaper_abandon_children(p, true); 882 LIST_REMOVE(p, p_reapsibling); 883 PROC_LOCK(p); 884 clear_orphan(p); 885 PROC_UNLOCK(p); 886 leavepgrp(p); 887#ifdef PROCDESC 888 if (p->p_procdesc != NULL) 889 procdesc_reap(p); 890#endif 891 sx_xunlock(&proctree_lock); 892 893 /* 894 * As a side effect of this lock, we know that all other writes to 895 * this proc are visible now, so no more locking is needed for p. 896 */ 897 PROC_LOCK(p); 898 p->p_xstat = 0; /* XXX: why? */ 899 PROC_UNLOCK(p); 900 PROC_LOCK(q); 901 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 902 PROC_UNLOCK(q); 903 904 /* 905 * Decrement the count of procs running with this uid. 906 */ 907 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 908 909 /* 910 * Destroy resource accounting information associated with the process. 911 */ 912#ifdef RACCT 913 PROC_LOCK(p); 914 racct_sub(p, RACCT_NPROC, 1); 915 PROC_UNLOCK(p); 916#endif 917 racct_proc_exit(p); 918 919 /* 920 * Free credentials, arguments, and sigacts. 921 */ 922 crfree(p->p_ucred); 923 p->p_ucred = NULL; 924 pargs_drop(p->p_args); 925 p->p_args = NULL; 926 sigacts_free(p->p_sigacts); 927 p->p_sigacts = NULL; 928 929 /* 930 * Do any thread-system specific cleanups. 931 */ 932 thread_wait(p); 933 934 /* 935 * Give vm and machine-dependent layer a chance to free anything that 936 * cpu_exit couldn't release while still running in process context. 937 */ 938 vm_waitproc(p); 939#ifdef MAC 940 mac_proc_destroy(p); 941#endif 942 KASSERT(FIRST_THREAD_IN_PROC(p), 943 ("proc_reap: no residual thread!")); 944 uma_zfree(proc_zone, p); 945 sx_xlock(&allproc_lock); 946 nprocs--; 947 sx_xunlock(&allproc_lock); 948} 949 950static int 951proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 952 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo) 953{ 954 struct rusage *rup; 955 956 sx_assert(&proctree_lock, SA_XLOCKED); 957 958 PROC_LOCK(p); 959 960 switch (idtype) { 961 case P_ALL: 962 break; 963 case P_PID: 964 if (p->p_pid != (pid_t)id) { 965 PROC_UNLOCK(p); 966 return (0); 967 } 968 break; 969 case P_PGID: 970 if (p->p_pgid != (pid_t)id) { 971 PROC_UNLOCK(p); 972 return (0); 973 } 974 break; 975 case P_SID: 976 if (p->p_session->s_sid != (pid_t)id) { 977 PROC_UNLOCK(p); 978 return (0); 979 } 980 break; 981 case P_UID: 982 if (p->p_ucred->cr_uid != (uid_t)id) { 983 PROC_UNLOCK(p); 984 return (0); 985 } 986 break; 987 case P_GID: 988 if (p->p_ucred->cr_gid != (gid_t)id) { 989 PROC_UNLOCK(p); 990 return (0); 991 } 992 break; 993 case P_JAILID: 994 if (p->p_ucred->cr_prison->pr_id != (int)id) { 995 PROC_UNLOCK(p); 996 return (0); 997 } 998 break; 999 /* 1000 * It seems that the thread structures get zeroed out 1001 * at process exit. This makes it impossible to 1002 * support P_SETID, P_CID or P_CPUID. 1003 */ 1004 default: 1005 PROC_UNLOCK(p); 1006 return (0); 1007 } 1008 1009 if (p_canwait(td, p)) { 1010 PROC_UNLOCK(p); 1011 return (0); 1012 } 1013 1014 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1015 PROC_UNLOCK(p); 1016 return (0); 1017 } 1018 1019 /* 1020 * This special case handles a kthread spawned by linux_clone 1021 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1022 * functions need to be able to distinguish between waiting 1023 * on a process and waiting on a thread. It is a thread if 1024 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1025 * signifies we want to wait for threads and not processes. 1026 */ 1027 if ((p->p_sigparent != SIGCHLD) ^ 1028 ((options & WLINUXCLONE) != 0)) { 1029 PROC_UNLOCK(p); 1030 return (0); 1031 } 1032 1033 PROC_SLOCK(p); 1034 1035 if (siginfo != NULL) { 1036 bzero(siginfo, sizeof(*siginfo)); 1037 siginfo->si_errno = 0; 1038 1039 /* 1040 * SUSv4 requires that the si_signo value is always 1041 * SIGCHLD. Obey it despite the rfork(2) interface 1042 * allows to request other signal for child exit 1043 * notification. 1044 */ 1045 siginfo->si_signo = SIGCHLD; 1046 1047 /* 1048 * This is still a rough estimate. We will fix the 1049 * cases TRAPPED, STOPPED, and CONTINUED later. 1050 */ 1051 if (WCOREDUMP(p->p_xstat)) { 1052 siginfo->si_code = CLD_DUMPED; 1053 siginfo->si_status = WTERMSIG(p->p_xstat); 1054 } else if (WIFSIGNALED(p->p_xstat)) { 1055 siginfo->si_code = CLD_KILLED; 1056 siginfo->si_status = WTERMSIG(p->p_xstat); 1057 } else { 1058 siginfo->si_code = CLD_EXITED; 1059 siginfo->si_status = WEXITSTATUS(p->p_xstat); 1060 } 1061 1062 siginfo->si_pid = p->p_pid; 1063 siginfo->si_uid = p->p_ucred->cr_uid; 1064 1065 /* 1066 * The si_addr field would be useful additional 1067 * detail, but apparently the PC value may be lost 1068 * when we reach this point. bzero() above sets 1069 * siginfo->si_addr to NULL. 1070 */ 1071 } 1072 1073 /* 1074 * There should be no reason to limit resources usage info to 1075 * exited processes only. A snapshot about any resources used 1076 * by a stopped process may be exactly what is needed. 1077 */ 1078 if (wrusage != NULL) { 1079 rup = &wrusage->wru_self; 1080 *rup = p->p_ru; 1081 calcru(p, &rup->ru_utime, &rup->ru_stime); 1082 1083 rup = &wrusage->wru_children; 1084 *rup = p->p_stats->p_cru; 1085 calccru(p, &rup->ru_utime, &rup->ru_stime); 1086 } 1087 1088 if (p->p_state == PRS_ZOMBIE) { 1089 proc_reap(td, p, status, options); 1090 return (-1); 1091 } 1092 PROC_SUNLOCK(p); 1093 PROC_UNLOCK(p); 1094 return (1); 1095} 1096 1097int 1098kern_wait(struct thread *td, pid_t pid, int *status, int options, 1099 struct rusage *rusage) 1100{ 1101 struct __wrusage wru, *wrup; 1102 idtype_t idtype; 1103 id_t id; 1104 int ret; 1105 1106 /* 1107 * Translate the special pid values into the (idtype, pid) 1108 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1109 * kern_wait6() on its own. 1110 */ 1111 if (pid == WAIT_ANY) { 1112 idtype = P_ALL; 1113 id = 0; 1114 } else if (pid < 0) { 1115 idtype = P_PGID; 1116 id = (id_t)-pid; 1117 } else { 1118 idtype = P_PID; 1119 id = (id_t)pid; 1120 } 1121 1122 if (rusage != NULL) 1123 wrup = &wru; 1124 else 1125 wrup = NULL; 1126 1127 /* 1128 * For backward compatibility we implicitly add flags WEXITED 1129 * and WTRAPPED here. 1130 */ 1131 options |= WEXITED | WTRAPPED; 1132 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1133 if (rusage != NULL) 1134 *rusage = wru.wru_self; 1135 return (ret); 1136} 1137 1138int 1139kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1140 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1141{ 1142 struct proc *p, *q; 1143 int error, nfound, ret; 1144 1145 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1146 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1147 AUDIT_ARG_VALUE(options); 1148 1149 q = td->td_proc; 1150 1151 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1152 PROC_LOCK(q); 1153 id = (id_t)q->p_pgid; 1154 PROC_UNLOCK(q); 1155 idtype = P_PGID; 1156 } 1157 1158 /* If we don't know the option, just return. */ 1159 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1160 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1161 return (EINVAL); 1162 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1163 /* 1164 * We will be unable to find any matching processes, 1165 * because there are no known events to look for. 1166 * Prefer to return error instead of blocking 1167 * indefinitely. 1168 */ 1169 return (EINVAL); 1170 } 1171 1172loop: 1173 if (q->p_flag & P_STATCHILD) { 1174 PROC_LOCK(q); 1175 q->p_flag &= ~P_STATCHILD; 1176 PROC_UNLOCK(q); 1177 } 1178 nfound = 0; 1179 sx_xlock(&proctree_lock); 1180 LIST_FOREACH(p, &q->p_children, p_sibling) { 1181 ret = proc_to_reap(td, p, idtype, id, status, options, 1182 wrusage, siginfo); 1183 if (ret == 0) 1184 continue; 1185 else if (ret == 1) 1186 nfound++; 1187 else 1188 return (0); 1189 1190 PROC_LOCK(p); 1191 PROC_SLOCK(p); 1192 1193 if ((options & WTRAPPED) != 0 && 1194 (p->p_flag & P_TRACED) != 0 && 1195 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1196 (p->p_suspcount == p->p_numthreads) && 1197 ((p->p_flag & P_WAITED) == 0)) { 1198 PROC_SUNLOCK(p); 1199 if ((options & WNOWAIT) == 0) 1200 p->p_flag |= P_WAITED; 1201 sx_xunlock(&proctree_lock); 1202 td->td_retval[0] = p->p_pid; 1203 1204 if (status != NULL) 1205 *status = W_STOPCODE(p->p_xstat); 1206 if (siginfo != NULL) { 1207 siginfo->si_status = p->p_xstat; 1208 siginfo->si_code = CLD_TRAPPED; 1209 } 1210 if ((options & WNOWAIT) == 0) { 1211 PROC_LOCK(q); 1212 sigqueue_take(p->p_ksi); 1213 PROC_UNLOCK(q); 1214 } 1215 1216 PROC_UNLOCK(p); 1217 return (0); 1218 } 1219 if ((options & WUNTRACED) != 0 && 1220 (p->p_flag & P_STOPPED_SIG) != 0 && 1221 (p->p_suspcount == p->p_numthreads) && 1222 ((p->p_flag & P_WAITED) == 0)) { 1223 PROC_SUNLOCK(p); 1224 if ((options & WNOWAIT) == 0) 1225 p->p_flag |= P_WAITED; 1226 sx_xunlock(&proctree_lock); 1227 td->td_retval[0] = p->p_pid; 1228 1229 if (status != NULL) 1230 *status = W_STOPCODE(p->p_xstat); 1231 if (siginfo != NULL) { 1232 siginfo->si_status = p->p_xstat; 1233 siginfo->si_code = CLD_STOPPED; 1234 } 1235 if ((options & WNOWAIT) == 0) { 1236 PROC_LOCK(q); 1237 sigqueue_take(p->p_ksi); 1238 PROC_UNLOCK(q); 1239 } 1240 1241 PROC_UNLOCK(p); 1242 return (0); 1243 } 1244 PROC_SUNLOCK(p); 1245 if ((options & WCONTINUED) != 0 && 1246 (p->p_flag & P_CONTINUED) != 0) { 1247 sx_xunlock(&proctree_lock); 1248 td->td_retval[0] = p->p_pid; 1249 if ((options & WNOWAIT) == 0) { 1250 p->p_flag &= ~P_CONTINUED; 1251 PROC_LOCK(q); 1252 sigqueue_take(p->p_ksi); 1253 PROC_UNLOCK(q); 1254 } 1255 PROC_UNLOCK(p); 1256 1257 if (status != NULL) 1258 *status = SIGCONT; 1259 if (siginfo != NULL) { 1260 siginfo->si_status = SIGCONT; 1261 siginfo->si_code = CLD_CONTINUED; 1262 } 1263 return (0); 1264 } 1265 PROC_UNLOCK(p); 1266 } 1267 1268 /* 1269 * Look in the orphans list too, to allow the parent to 1270 * collect it's child exit status even if child is being 1271 * debugged. 1272 * 1273 * Debugger detaches from the parent upon successful 1274 * switch-over from parent to child. At this point due to 1275 * re-parenting the parent loses the child to debugger and a 1276 * wait4(2) call would report that it has no children to wait 1277 * for. By maintaining a list of orphans we allow the parent 1278 * to successfully wait until the child becomes a zombie. 1279 */ 1280 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1281 ret = proc_to_reap(td, p, idtype, id, status, options, 1282 wrusage, siginfo); 1283 if (ret == 0) 1284 continue; 1285 else if (ret == 1) 1286 nfound++; 1287 else 1288 return (0); 1289 } 1290 if (nfound == 0) { 1291 sx_xunlock(&proctree_lock); 1292 return (ECHILD); 1293 } 1294 if (options & WNOHANG) { 1295 sx_xunlock(&proctree_lock); 1296 td->td_retval[0] = 0; 1297 return (0); 1298 } 1299 PROC_LOCK(q); 1300 sx_xunlock(&proctree_lock); 1301 if (q->p_flag & P_STATCHILD) { 1302 q->p_flag &= ~P_STATCHILD; 1303 error = 0; 1304 } else 1305 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1306 PROC_UNLOCK(q); 1307 if (error) 1308 return (error); 1309 goto loop; 1310} 1311 1312/* 1313 * Make process 'parent' the new parent of process 'child'. 1314 * Must be called with an exclusive hold of proctree lock. 1315 */ 1316void 1317proc_reparent(struct proc *child, struct proc *parent) 1318{ 1319 1320 sx_assert(&proctree_lock, SX_XLOCKED); 1321 PROC_LOCK_ASSERT(child, MA_OWNED); 1322 if (child->p_pptr == parent) 1323 return; 1324 1325 PROC_LOCK(child->p_pptr); 1326 sigqueue_take(child->p_ksi); 1327 PROC_UNLOCK(child->p_pptr); 1328 LIST_REMOVE(child, p_sibling); 1329 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1330 1331 clear_orphan(child); 1332 if (child->p_flag & P_TRACED) { 1333 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1334 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1335 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1336 p_orphan); 1337 } else { 1338 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1339 child, p_orphan); 1340 } 1341 child->p_treeflag |= P_TREE_ORPHANED; 1342 } 1343 1344 child->p_pptr = parent; 1345} 1346