kern_exit.c revision 284199
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 284199 2015-06-10 02:04:02Z 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 * Deref SU mp, since the thread does not return to userspace. 211 */ 212 if (softdep_ast_cleanup != NULL) 213 softdep_ast_cleanup(); 214 215 /* 216 * MUST abort all other threads before proceeding past here. 217 */ 218 PROC_LOCK(p); 219 /* 220 * First check if some other thread or external request got 221 * here before us. If so, act appropriately: exit or suspend. 222 * We must ensure that stop requests are handled before we set 223 * P_WEXIT. 224 */ 225 thread_suspend_check(0); 226 while (p->p_flag & P_HADTHREADS) { 227 /* 228 * Kill off the other threads. This requires 229 * some co-operation from other parts of the kernel 230 * so it may not be instantaneous. With this state set 231 * any thread entering the kernel from userspace will 232 * thread_exit() in trap(). Any thread attempting to 233 * sleep will return immediately with EINTR or EWOULDBLOCK 234 * which will hopefully force them to back out to userland 235 * freeing resources as they go. Any thread attempting 236 * to return to userland will thread_exit() from userret(). 237 * thread_exit() will unsuspend us when the last of the 238 * other threads exits. 239 * If there is already a thread singler after resumption, 240 * calling thread_single will fail; in that case, we just 241 * re-check all suspension request, the thread should 242 * either be suspended there or exit. 243 */ 244 if (!thread_single(p, SINGLE_EXIT)) 245 /* 246 * All other activity in this process is now 247 * stopped. Threading support has been turned 248 * off. 249 */ 250 break; 251 /* 252 * Recheck for new stop or suspend requests which 253 * might appear while process lock was dropped in 254 * thread_single(). 255 */ 256 thread_suspend_check(0); 257 } 258 KASSERT(p->p_numthreads == 1, 259 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 260 racct_sub(p, RACCT_NTHR, 1); 261 /* 262 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 263 * on our vmspace, so we should block below until they have 264 * released their reference to us. Note that if they have 265 * requested S_EXIT stops we will block here until they ack 266 * via PIOCCONT. 267 */ 268 _STOPEVENT(p, S_EXIT, rv); 269 270 /* 271 * Ignore any pending request to stop due to a stop signal. 272 * Once P_WEXIT is set, future requests will be ignored as 273 * well. 274 */ 275 p->p_flag &= ~P_STOPPED_SIG; 276 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 277 278 /* 279 * Note that we are exiting and do another wakeup of anyone in 280 * PIOCWAIT in case they aren't listening for S_EXIT stops or 281 * decided to wait again after we told them we are exiting. 282 */ 283 p->p_flag |= P_WEXIT; 284 wakeup(&p->p_stype); 285 286 /* 287 * Wait for any processes that have a hold on our vmspace to 288 * release their reference. 289 */ 290 while (p->p_lock > 0) 291 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 292 293 p->p_xstat = rv; /* Let event handler change exit status */ 294 PROC_UNLOCK(p); 295 /* Drain the limit callout while we don't have the proc locked */ 296 callout_drain(&p->p_limco); 297 298#ifdef AUDIT 299 /* 300 * The Sun BSM exit token contains two components: an exit status as 301 * passed to exit(), and a return value to indicate what sort of exit 302 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 303 * what the return value is. 304 */ 305 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0); 306 AUDIT_SYSCALL_EXIT(0, td); 307#endif 308 309 /* Are we a task leader? */ 310 if (p == p->p_leader) { 311 mtx_lock(&ppeers_lock); 312 q = p->p_peers; 313 while (q != NULL) { 314 PROC_LOCK(q); 315 kern_psignal(q, SIGKILL); 316 PROC_UNLOCK(q); 317 q = q->p_peers; 318 } 319 while (p->p_peers != NULL) 320 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 321 mtx_unlock(&ppeers_lock); 322 } 323 324 /* 325 * Check if any loadable modules need anything done at process exit. 326 * E.g. SYSV IPC stuff 327 * XXX what if one of these generates an error? 328 */ 329 EVENTHANDLER_INVOKE(process_exit, p); 330 331 /* 332 * If parent is waiting for us to exit or exec, 333 * P_PPWAIT is set; we will wakeup the parent below. 334 */ 335 PROC_LOCK(p); 336 rv = p->p_xstat; /* Event handler could change exit status */ 337 stopprofclock(p); 338 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 339 340 /* 341 * Stop the real interval timer. If the handler is currently 342 * executing, prevent it from rearming itself and let it finish. 343 */ 344 if (timevalisset(&p->p_realtimer.it_value) && 345 callout_stop(&p->p_itcallout) == 0) { 346 timevalclear(&p->p_realtimer.it_interval); 347 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 348 KASSERT(!timevalisset(&p->p_realtimer.it_value), 349 ("realtime timer is still armed")); 350 } 351 PROC_UNLOCK(p); 352 353 /* 354 * Reset any sigio structures pointing to us as a result of 355 * F_SETOWN with our pid. 356 */ 357 funsetownlst(&p->p_sigiolst); 358 359 /* 360 * If this process has an nlminfo data area (for lockd), release it 361 */ 362 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 363 (*nlminfo_release_p)(p); 364 365 /* 366 * Close open files and release open-file table. 367 * This may block! 368 */ 369 fdescfree(td); 370 371 /* 372 * If this thread tickled GEOM, we need to wait for the giggling to 373 * stop before we return to userland 374 */ 375 if (td->td_pflags & TDP_GEOM) 376 g_waitidle(); 377 378 /* 379 * Remove ourself from our leader's peer list and wake our leader. 380 */ 381 mtx_lock(&ppeers_lock); 382 if (p->p_leader->p_peers) { 383 q = p->p_leader; 384 while (q->p_peers != p) 385 q = q->p_peers; 386 q->p_peers = p->p_peers; 387 wakeup(p->p_leader); 388 } 389 mtx_unlock(&ppeers_lock); 390 391 vmspace_exit(td); 392 393 sx_xlock(&proctree_lock); 394 if (SESS_LEADER(p)) { 395 struct session *sp = p->p_session; 396 struct tty *tp; 397 398 /* 399 * s_ttyp is not zero'd; we use this to indicate that 400 * the session once had a controlling terminal. (for 401 * logging and informational purposes) 402 */ 403 SESS_LOCK(sp); 404 ttyvp = sp->s_ttyvp; 405 tp = sp->s_ttyp; 406 sp->s_ttyvp = NULL; 407 sp->s_ttydp = NULL; 408 sp->s_leader = NULL; 409 SESS_UNLOCK(sp); 410 411 /* 412 * Signal foreground pgrp and revoke access to 413 * controlling terminal if it has not been revoked 414 * already. 415 * 416 * Because the TTY may have been revoked in the mean 417 * time and could already have a new session associated 418 * with it, make sure we don't send a SIGHUP to a 419 * foreground process group that does not belong to this 420 * session. 421 */ 422 423 if (tp != NULL) { 424 tty_lock(tp); 425 if (tp->t_session == sp) 426 tty_signal_pgrp(tp, SIGHUP); 427 tty_unlock(tp); 428 } 429 430 if (ttyvp != NULL) { 431 sx_xunlock(&proctree_lock); 432 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 433 VOP_REVOKE(ttyvp, REVOKEALL); 434 VOP_UNLOCK(ttyvp, 0); 435 } 436 sx_xlock(&proctree_lock); 437 } 438 } 439 fixjobc(p, p->p_pgrp, 0); 440 sx_xunlock(&proctree_lock); 441 (void)acct_process(td); 442 443 /* Release the TTY now we've unlocked everything. */ 444 if (ttyvp != NULL) 445 vrele(ttyvp); 446#ifdef KTRACE 447 ktrprocexit(td); 448#endif 449 /* 450 * Release reference to text vnode 451 */ 452 if (p->p_textvp != NULL) { 453 vrele(p->p_textvp); 454 p->p_textvp = NULL; 455 } 456 457 /* 458 * Release our limits structure. 459 */ 460 lim_free(p->p_limit); 461 p->p_limit = NULL; 462 463 tidhash_remove(td); 464 465 /* 466 * Remove proc from allproc queue and pidhash chain. 467 * Place onto zombproc. Unlink from parent's child list. 468 */ 469 sx_xlock(&allproc_lock); 470 LIST_REMOVE(p, p_list); 471 LIST_INSERT_HEAD(&zombproc, p, p_list); 472 LIST_REMOVE(p, p_hash); 473 sx_xunlock(&allproc_lock); 474 475 /* 476 * Call machine-dependent code to release any 477 * machine-dependent resources other than the address space. 478 * The address space is released by "vmspace_exitfree(p)" in 479 * vm_waitproc(). 480 */ 481 cpu_exit(td); 482 483 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 484 485 /* 486 * Reparent all children processes: 487 * - traced ones to the original parent (or init if we are that parent) 488 * - the rest to init 489 */ 490 sx_xlock(&proctree_lock); 491 q = LIST_FIRST(&p->p_children); 492 if (q != NULL) /* only need this if any child is S_ZOMB */ 493 wakeup(q->p_reaper); 494 for (; q != NULL; q = nq) { 495 nq = LIST_NEXT(q, p_sibling); 496 PROC_LOCK(q); 497 q->p_sigparent = SIGCHLD; 498 499 if (!(q->p_flag & P_TRACED)) { 500 proc_reparent(q, q->p_reaper); 501 } else { 502 /* 503 * Traced processes are killed since their existence 504 * means someone is screwing up. 505 */ 506 t = proc_realparent(q); 507 if (t == p) { 508 proc_reparent(q, q->p_reaper); 509 } else { 510 PROC_LOCK(t); 511 proc_reparent(q, t); 512 PROC_UNLOCK(t); 513 } 514 /* 515 * Since q was found on our children list, the 516 * proc_reparent() call moved q to the orphan 517 * list due to present P_TRACED flag. Clear 518 * orphan link for q now while q is locked. 519 */ 520 clear_orphan(q); 521 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 522 FOREACH_THREAD_IN_PROC(q, tdt) 523 tdt->td_dbgflags &= ~TDB_SUSPEND; 524 kern_psignal(q, SIGKILL); 525 } 526 PROC_UNLOCK(q); 527 } 528 529 /* 530 * Also get rid of our orphans. 531 */ 532 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 533 PROC_LOCK(q); 534 clear_orphan(q); 535 PROC_UNLOCK(q); 536 } 537 538 /* Save exit status. */ 539 PROC_LOCK(p); 540 p->p_xthread = td; 541 542 /* Tell the prison that we are gone. */ 543 prison_proc_free(p->p_ucred->cr_prison); 544 545#ifdef KDTRACE_HOOKS 546 /* 547 * Tell the DTrace fasttrap provider about the exit if it 548 * has declared an interest. 549 */ 550 if (dtrace_fasttrap_exit) 551 dtrace_fasttrap_exit(p); 552#endif 553 554 /* 555 * Notify interested parties of our demise. 556 */ 557 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 558 559#ifdef KDTRACE_HOOKS 560 int reason = CLD_EXITED; 561 if (WCOREDUMP(rv)) 562 reason = CLD_DUMPED; 563 else if (WIFSIGNALED(rv)) 564 reason = CLD_KILLED; 565 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 566#endif 567 568 /* 569 * Just delete all entries in the p_klist. At this point we won't 570 * report any more events, and there are nasty race conditions that 571 * can beat us if we don't. 572 */ 573 knlist_clear(&p->p_klist, 1); 574 575 /* 576 * If this is a process with a descriptor, we may not need to deliver 577 * a signal to the parent. proctree_lock is held over 578 * procdesc_exit() to serialize concurrent calls to close() and 579 * exit(). 580 */ 581#ifdef PROCDESC 582 if (p->p_procdesc == NULL || procdesc_exit(p)) { 583#endif 584 /* 585 * Notify parent that we're gone. If parent has the 586 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 587 * notify process 1 instead (and hope it will handle this 588 * situation). 589 */ 590 PROC_LOCK(p->p_pptr); 591 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 592 if (p->p_pptr->p_sigacts->ps_flag & 593 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 594 struct proc *pp; 595 596 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 597 pp = p->p_pptr; 598 PROC_UNLOCK(pp); 599 proc_reparent(p, p->p_reaper); 600 p->p_sigparent = SIGCHLD; 601 PROC_LOCK(p->p_pptr); 602 603 /* 604 * Notify parent, so in case he was wait(2)ing or 605 * executing waitpid(2) with our pid, he will 606 * continue. 607 */ 608 wakeup(pp); 609 } else 610 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 611 612 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 613 childproc_exited(p); 614 else if (p->p_sigparent != 0) { 615 if (p->p_sigparent == SIGCHLD) 616 childproc_exited(p); 617 else /* LINUX thread */ 618 kern_psignal(p->p_pptr, p->p_sigparent); 619 } 620#ifdef PROCDESC 621 } else 622 PROC_LOCK(p->p_pptr); 623#endif 624 sx_xunlock(&proctree_lock); 625 626 /* 627 * The state PRS_ZOMBIE prevents other proesses from sending 628 * signal to the process, to avoid memory leak, we free memory 629 * for signal queue at the time when the state is set. 630 */ 631 sigqueue_flush(&p->p_sigqueue); 632 sigqueue_flush(&td->td_sigqueue); 633 634 /* 635 * We have to wait until after acquiring all locks before 636 * changing p_state. We need to avoid all possible context 637 * switches (including ones from blocking on a mutex) while 638 * marked as a zombie. We also have to set the zombie state 639 * before we release the parent process' proc lock to avoid 640 * a lost wakeup. So, we first call wakeup, then we grab the 641 * sched lock, update the state, and release the parent process' 642 * proc lock. 643 */ 644 wakeup(p->p_pptr); 645 cv_broadcast(&p->p_pwait); 646 sched_exit(p->p_pptr, td); 647 umtx_thread_exit(td); 648 PROC_SLOCK(p); 649 p->p_state = PRS_ZOMBIE; 650 PROC_UNLOCK(p->p_pptr); 651 652 /* 653 * Hopefully no one will try to deliver a signal to the process this 654 * late in the game. 655 */ 656 knlist_destroy(&p->p_klist); 657 658 /* 659 * Save our children's rusage information in our exit rusage. 660 */ 661 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 662 663 /* 664 * Make sure the scheduler takes this thread out of its tables etc. 665 * This will also release this thread's reference to the ucred. 666 * Other thread parts to release include pcb bits and such. 667 */ 668 thread_exit(); 669} 670 671 672#ifndef _SYS_SYSPROTO_H_ 673struct abort2_args { 674 char *why; 675 int nargs; 676 void **args; 677}; 678#endif 679 680int 681sys_abort2(struct thread *td, struct abort2_args *uap) 682{ 683 struct proc *p = td->td_proc; 684 struct sbuf *sb; 685 void *uargs[16]; 686 int error, i, sig; 687 688 /* 689 * Do it right now so we can log either proper call of abort2(), or 690 * note, that invalid argument was passed. 512 is big enough to 691 * handle 16 arguments' descriptions with additional comments. 692 */ 693 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 694 sbuf_clear(sb); 695 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 696 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 697 /* 698 * Since we can't return from abort2(), send SIGKILL in cases, where 699 * abort2() was called improperly 700 */ 701 sig = SIGKILL; 702 /* Prevent from DoSes from user-space. */ 703 if (uap->nargs < 0 || uap->nargs > 16) 704 goto out; 705 if (uap->nargs > 0) { 706 if (uap->args == NULL) 707 goto out; 708 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 709 if (error != 0) 710 goto out; 711 } 712 /* 713 * Limit size of 'reason' string to 128. Will fit even when 714 * maximal number of arguments was chosen to be logged. 715 */ 716 if (uap->why != NULL) { 717 error = sbuf_copyin(sb, uap->why, 128); 718 if (error < 0) 719 goto out; 720 } else { 721 sbuf_printf(sb, "(null)"); 722 } 723 if (uap->nargs > 0) { 724 sbuf_printf(sb, "("); 725 for (i = 0;i < uap->nargs; i++) 726 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 727 sbuf_printf(sb, ")"); 728 } 729 /* 730 * Final stage: arguments were proper, string has been 731 * successfully copied from userspace, and copying pointers 732 * from user-space succeed. 733 */ 734 sig = SIGABRT; 735out: 736 if (sig == SIGKILL) { 737 sbuf_trim(sb); 738 sbuf_printf(sb, " (Reason text inaccessible)"); 739 } 740 sbuf_cat(sb, "\n"); 741 sbuf_finish(sb); 742 log(LOG_INFO, "%s", sbuf_data(sb)); 743 sbuf_delete(sb); 744 exit1(td, W_EXITCODE(0, sig)); 745 return (0); 746} 747 748 749#ifdef COMPAT_43 750/* 751 * The dirty work is handled by kern_wait(). 752 */ 753int 754owait(struct thread *td, struct owait_args *uap __unused) 755{ 756 int error, status; 757 758 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 759 if (error == 0) 760 td->td_retval[1] = status; 761 return (error); 762} 763#endif /* COMPAT_43 */ 764 765/* 766 * The dirty work is handled by kern_wait(). 767 */ 768int 769sys_wait4(struct thread *td, struct wait4_args *uap) 770{ 771 struct rusage ru, *rup; 772 int error, status; 773 774 if (uap->rusage != NULL) 775 rup = &ru; 776 else 777 rup = NULL; 778 error = kern_wait(td, uap->pid, &status, uap->options, rup); 779 if (uap->status != NULL && error == 0) 780 error = copyout(&status, uap->status, sizeof(status)); 781 if (uap->rusage != NULL && error == 0) 782 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 783 return (error); 784} 785 786int 787sys_wait6(struct thread *td, struct wait6_args *uap) 788{ 789 struct __wrusage wru, *wrup; 790 siginfo_t si, *sip; 791 idtype_t idtype; 792 id_t id; 793 int error, status; 794 795 idtype = uap->idtype; 796 id = uap->id; 797 798 if (uap->wrusage != NULL) 799 wrup = &wru; 800 else 801 wrup = NULL; 802 803 if (uap->info != NULL) { 804 sip = &si; 805 bzero(sip, sizeof(*sip)); 806 } else 807 sip = NULL; 808 809 /* 810 * We expect all callers of wait6() to know about WEXITED and 811 * WTRAPPED. 812 */ 813 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 814 815 if (uap->status != NULL && error == 0) 816 error = copyout(&status, uap->status, sizeof(status)); 817 if (uap->wrusage != NULL && error == 0) 818 error = copyout(&wru, uap->wrusage, sizeof(wru)); 819 if (uap->info != NULL && error == 0) 820 error = copyout(&si, uap->info, sizeof(si)); 821 return (error); 822} 823 824/* 825 * Reap the remains of a zombie process and optionally return status and 826 * rusage. Asserts and will release both the proctree_lock and the process 827 * lock as part of its work. 828 */ 829void 830proc_reap(struct thread *td, struct proc *p, int *status, int options) 831{ 832 struct proc *q, *t; 833 834 sx_assert(&proctree_lock, SA_XLOCKED); 835 PROC_LOCK_ASSERT(p, MA_OWNED); 836 PROC_SLOCK_ASSERT(p, MA_OWNED); 837 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 838 839 q = td->td_proc; 840 841 PROC_SUNLOCK(p); 842 td->td_retval[0] = p->p_pid; 843 if (status) 844 *status = p->p_xstat; /* convert to int */ 845 if (options & WNOWAIT) { 846 /* 847 * Only poll, returning the status. Caller does not wish to 848 * release the proc struct just yet. 849 */ 850 PROC_UNLOCK(p); 851 sx_xunlock(&proctree_lock); 852 return; 853 } 854 855 PROC_LOCK(q); 856 sigqueue_take(p->p_ksi); 857 PROC_UNLOCK(q); 858 PROC_UNLOCK(p); 859 860 /* 861 * If we got the child via a ptrace 'attach', we need to give it back 862 * to the old parent. 863 */ 864 if (p->p_oppid != 0) { 865 t = proc_realparent(p); 866 PROC_LOCK(t); 867 PROC_LOCK(p); 868 proc_reparent(p, t); 869 p->p_oppid = 0; 870 PROC_UNLOCK(p); 871 pksignal(t, SIGCHLD, p->p_ksi); 872 wakeup(t); 873 cv_broadcast(&p->p_pwait); 874 PROC_UNLOCK(t); 875 sx_xunlock(&proctree_lock); 876 return; 877 } 878 879 /* 880 * Remove other references to this process to ensure we have an 881 * exclusive reference. 882 */ 883 sx_xlock(&allproc_lock); 884 LIST_REMOVE(p, p_list); /* off zombproc */ 885 sx_xunlock(&allproc_lock); 886 LIST_REMOVE(p, p_sibling); 887 reaper_abandon_children(p, true); 888 LIST_REMOVE(p, p_reapsibling); 889 PROC_LOCK(p); 890 clear_orphan(p); 891 PROC_UNLOCK(p); 892 leavepgrp(p); 893#ifdef PROCDESC 894 if (p->p_procdesc != NULL) 895 procdesc_reap(p); 896#endif 897 sx_xunlock(&proctree_lock); 898 899 /* 900 * As a side effect of this lock, we know that all other writes to 901 * this proc are visible now, so no more locking is needed for p. 902 */ 903 PROC_LOCK(p); 904 p->p_xstat = 0; /* XXX: why? */ 905 PROC_UNLOCK(p); 906 PROC_LOCK(q); 907 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 908 PROC_UNLOCK(q); 909 910 /* 911 * Decrement the count of procs running with this uid. 912 */ 913 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 914 915 /* 916 * Destroy resource accounting information associated with the process. 917 */ 918#ifdef RACCT 919 PROC_LOCK(p); 920 racct_sub(p, RACCT_NPROC, 1); 921 PROC_UNLOCK(p); 922#endif 923 racct_proc_exit(p); 924 925 /* 926 * Free credentials, arguments, and sigacts. 927 */ 928 crfree(p->p_ucred); 929 p->p_ucred = NULL; 930 pargs_drop(p->p_args); 931 p->p_args = NULL; 932 sigacts_free(p->p_sigacts); 933 p->p_sigacts = NULL; 934 935 /* 936 * Do any thread-system specific cleanups. 937 */ 938 thread_wait(p); 939 940 /* 941 * Give vm and machine-dependent layer a chance to free anything that 942 * cpu_exit couldn't release while still running in process context. 943 */ 944 vm_waitproc(p); 945#ifdef MAC 946 mac_proc_destroy(p); 947#endif 948 KASSERT(FIRST_THREAD_IN_PROC(p), 949 ("proc_reap: no residual thread!")); 950 uma_zfree(proc_zone, p); 951 sx_xlock(&allproc_lock); 952 nprocs--; 953 sx_xunlock(&allproc_lock); 954} 955 956static int 957proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 958 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo) 959{ 960 struct rusage *rup; 961 962 sx_assert(&proctree_lock, SA_XLOCKED); 963 964 PROC_LOCK(p); 965 966 switch (idtype) { 967 case P_ALL: 968 break; 969 case P_PID: 970 if (p->p_pid != (pid_t)id) { 971 PROC_UNLOCK(p); 972 return (0); 973 } 974 break; 975 case P_PGID: 976 if (p->p_pgid != (pid_t)id) { 977 PROC_UNLOCK(p); 978 return (0); 979 } 980 break; 981 case P_SID: 982 if (p->p_session->s_sid != (pid_t)id) { 983 PROC_UNLOCK(p); 984 return (0); 985 } 986 break; 987 case P_UID: 988 if (p->p_ucred->cr_uid != (uid_t)id) { 989 PROC_UNLOCK(p); 990 return (0); 991 } 992 break; 993 case P_GID: 994 if (p->p_ucred->cr_gid != (gid_t)id) { 995 PROC_UNLOCK(p); 996 return (0); 997 } 998 break; 999 case P_JAILID: 1000 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1001 PROC_UNLOCK(p); 1002 return (0); 1003 } 1004 break; 1005 /* 1006 * It seems that the thread structures get zeroed out 1007 * at process exit. This makes it impossible to 1008 * support P_SETID, P_CID or P_CPUID. 1009 */ 1010 default: 1011 PROC_UNLOCK(p); 1012 return (0); 1013 } 1014 1015 if (p_canwait(td, p)) { 1016 PROC_UNLOCK(p); 1017 return (0); 1018 } 1019 1020 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1021 PROC_UNLOCK(p); 1022 return (0); 1023 } 1024 1025 /* 1026 * This special case handles a kthread spawned by linux_clone 1027 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1028 * functions need to be able to distinguish between waiting 1029 * on a process and waiting on a thread. It is a thread if 1030 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1031 * signifies we want to wait for threads and not processes. 1032 */ 1033 if ((p->p_sigparent != SIGCHLD) ^ 1034 ((options & WLINUXCLONE) != 0)) { 1035 PROC_UNLOCK(p); 1036 return (0); 1037 } 1038 1039 PROC_SLOCK(p); 1040 1041 if (siginfo != NULL) { 1042 bzero(siginfo, sizeof(*siginfo)); 1043 siginfo->si_errno = 0; 1044 1045 /* 1046 * SUSv4 requires that the si_signo value is always 1047 * SIGCHLD. Obey it despite the rfork(2) interface 1048 * allows to request other signal for child exit 1049 * notification. 1050 */ 1051 siginfo->si_signo = SIGCHLD; 1052 1053 /* 1054 * This is still a rough estimate. We will fix the 1055 * cases TRAPPED, STOPPED, and CONTINUED later. 1056 */ 1057 if (WCOREDUMP(p->p_xstat)) { 1058 siginfo->si_code = CLD_DUMPED; 1059 siginfo->si_status = WTERMSIG(p->p_xstat); 1060 } else if (WIFSIGNALED(p->p_xstat)) { 1061 siginfo->si_code = CLD_KILLED; 1062 siginfo->si_status = WTERMSIG(p->p_xstat); 1063 } else { 1064 siginfo->si_code = CLD_EXITED; 1065 siginfo->si_status = WEXITSTATUS(p->p_xstat); 1066 } 1067 1068 siginfo->si_pid = p->p_pid; 1069 siginfo->si_uid = p->p_ucred->cr_uid; 1070 1071 /* 1072 * The si_addr field would be useful additional 1073 * detail, but apparently the PC value may be lost 1074 * when we reach this point. bzero() above sets 1075 * siginfo->si_addr to NULL. 1076 */ 1077 } 1078 1079 /* 1080 * There should be no reason to limit resources usage info to 1081 * exited processes only. A snapshot about any resources used 1082 * by a stopped process may be exactly what is needed. 1083 */ 1084 if (wrusage != NULL) { 1085 rup = &wrusage->wru_self; 1086 *rup = p->p_ru; 1087 calcru(p, &rup->ru_utime, &rup->ru_stime); 1088 1089 rup = &wrusage->wru_children; 1090 *rup = p->p_stats->p_cru; 1091 calccru(p, &rup->ru_utime, &rup->ru_stime); 1092 } 1093 1094 if (p->p_state == PRS_ZOMBIE) { 1095 proc_reap(td, p, status, options); 1096 return (-1); 1097 } 1098 PROC_SUNLOCK(p); 1099 PROC_UNLOCK(p); 1100 return (1); 1101} 1102 1103int 1104kern_wait(struct thread *td, pid_t pid, int *status, int options, 1105 struct rusage *rusage) 1106{ 1107 struct __wrusage wru, *wrup; 1108 idtype_t idtype; 1109 id_t id; 1110 int ret; 1111 1112 /* 1113 * Translate the special pid values into the (idtype, pid) 1114 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1115 * kern_wait6() on its own. 1116 */ 1117 if (pid == WAIT_ANY) { 1118 idtype = P_ALL; 1119 id = 0; 1120 } else if (pid < 0) { 1121 idtype = P_PGID; 1122 id = (id_t)-pid; 1123 } else { 1124 idtype = P_PID; 1125 id = (id_t)pid; 1126 } 1127 1128 if (rusage != NULL) 1129 wrup = &wru; 1130 else 1131 wrup = NULL; 1132 1133 /* 1134 * For backward compatibility we implicitly add flags WEXITED 1135 * and WTRAPPED here. 1136 */ 1137 options |= WEXITED | WTRAPPED; 1138 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1139 if (rusage != NULL) 1140 *rusage = wru.wru_self; 1141 return (ret); 1142} 1143 1144int 1145kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1146 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1147{ 1148 struct proc *p, *q; 1149 int error, nfound, ret; 1150 1151 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1152 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1153 AUDIT_ARG_VALUE(options); 1154 1155 q = td->td_proc; 1156 1157 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1158 PROC_LOCK(q); 1159 id = (id_t)q->p_pgid; 1160 PROC_UNLOCK(q); 1161 idtype = P_PGID; 1162 } 1163 1164 /* If we don't know the option, just return. */ 1165 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1166 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1167 return (EINVAL); 1168 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1169 /* 1170 * We will be unable to find any matching processes, 1171 * because there are no known events to look for. 1172 * Prefer to return error instead of blocking 1173 * indefinitely. 1174 */ 1175 return (EINVAL); 1176 } 1177 1178loop: 1179 if (q->p_flag & P_STATCHILD) { 1180 PROC_LOCK(q); 1181 q->p_flag &= ~P_STATCHILD; 1182 PROC_UNLOCK(q); 1183 } 1184 nfound = 0; 1185 sx_xlock(&proctree_lock); 1186 LIST_FOREACH(p, &q->p_children, p_sibling) { 1187 ret = proc_to_reap(td, p, idtype, id, status, options, 1188 wrusage, siginfo); 1189 if (ret == 0) 1190 continue; 1191 else if (ret == 1) 1192 nfound++; 1193 else 1194 return (0); 1195 1196 PROC_LOCK(p); 1197 PROC_SLOCK(p); 1198 1199 if ((options & WTRAPPED) != 0 && 1200 (p->p_flag & P_TRACED) != 0 && 1201 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1202 (p->p_suspcount == p->p_numthreads) && 1203 ((p->p_flag & P_WAITED) == 0)) { 1204 PROC_SUNLOCK(p); 1205 if ((options & WNOWAIT) == 0) 1206 p->p_flag |= P_WAITED; 1207 sx_xunlock(&proctree_lock); 1208 td->td_retval[0] = p->p_pid; 1209 1210 if (status != NULL) 1211 *status = W_STOPCODE(p->p_xstat); 1212 if (siginfo != NULL) { 1213 siginfo->si_status = p->p_xstat; 1214 siginfo->si_code = CLD_TRAPPED; 1215 } 1216 if ((options & WNOWAIT) == 0) { 1217 PROC_LOCK(q); 1218 sigqueue_take(p->p_ksi); 1219 PROC_UNLOCK(q); 1220 } 1221 1222 PROC_UNLOCK(p); 1223 return (0); 1224 } 1225 if ((options & WUNTRACED) != 0 && 1226 (p->p_flag & P_STOPPED_SIG) != 0 && 1227 (p->p_suspcount == p->p_numthreads) && 1228 ((p->p_flag & P_WAITED) == 0)) { 1229 PROC_SUNLOCK(p); 1230 if ((options & WNOWAIT) == 0) 1231 p->p_flag |= P_WAITED; 1232 sx_xunlock(&proctree_lock); 1233 td->td_retval[0] = p->p_pid; 1234 1235 if (status != NULL) 1236 *status = W_STOPCODE(p->p_xstat); 1237 if (siginfo != NULL) { 1238 siginfo->si_status = p->p_xstat; 1239 siginfo->si_code = CLD_STOPPED; 1240 } 1241 if ((options & WNOWAIT) == 0) { 1242 PROC_LOCK(q); 1243 sigqueue_take(p->p_ksi); 1244 PROC_UNLOCK(q); 1245 } 1246 1247 PROC_UNLOCK(p); 1248 return (0); 1249 } 1250 PROC_SUNLOCK(p); 1251 if ((options & WCONTINUED) != 0 && 1252 (p->p_flag & P_CONTINUED) != 0) { 1253 sx_xunlock(&proctree_lock); 1254 td->td_retval[0] = p->p_pid; 1255 if ((options & WNOWAIT) == 0) { 1256 p->p_flag &= ~P_CONTINUED; 1257 PROC_LOCK(q); 1258 sigqueue_take(p->p_ksi); 1259 PROC_UNLOCK(q); 1260 } 1261 PROC_UNLOCK(p); 1262 1263 if (status != NULL) 1264 *status = SIGCONT; 1265 if (siginfo != NULL) { 1266 siginfo->si_status = SIGCONT; 1267 siginfo->si_code = CLD_CONTINUED; 1268 } 1269 return (0); 1270 } 1271 PROC_UNLOCK(p); 1272 } 1273 1274 /* 1275 * Look in the orphans list too, to allow the parent to 1276 * collect it's child exit status even if child is being 1277 * debugged. 1278 * 1279 * Debugger detaches from the parent upon successful 1280 * switch-over from parent to child. At this point due to 1281 * re-parenting the parent loses the child to debugger and a 1282 * wait4(2) call would report that it has no children to wait 1283 * for. By maintaining a list of orphans we allow the parent 1284 * to successfully wait until the child becomes a zombie. 1285 */ 1286 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1287 ret = proc_to_reap(td, p, idtype, id, status, options, 1288 wrusage, siginfo); 1289 if (ret == 0) 1290 continue; 1291 else if (ret == 1) 1292 nfound++; 1293 else 1294 return (0); 1295 } 1296 if (nfound == 0) { 1297 sx_xunlock(&proctree_lock); 1298 return (ECHILD); 1299 } 1300 if (options & WNOHANG) { 1301 sx_xunlock(&proctree_lock); 1302 td->td_retval[0] = 0; 1303 return (0); 1304 } 1305 PROC_LOCK(q); 1306 sx_xunlock(&proctree_lock); 1307 if (q->p_flag & P_STATCHILD) { 1308 q->p_flag &= ~P_STATCHILD; 1309 error = 0; 1310 } else 1311 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1312 PROC_UNLOCK(q); 1313 if (error) 1314 return (error); 1315 goto loop; 1316} 1317 1318/* 1319 * Make process 'parent' the new parent of process 'child'. 1320 * Must be called with an exclusive hold of proctree lock. 1321 */ 1322void 1323proc_reparent(struct proc *child, struct proc *parent) 1324{ 1325 1326 sx_assert(&proctree_lock, SX_XLOCKED); 1327 PROC_LOCK_ASSERT(child, MA_OWNED); 1328 if (child->p_pptr == parent) 1329 return; 1330 1331 PROC_LOCK(child->p_pptr); 1332 sigqueue_take(child->p_ksi); 1333 PROC_UNLOCK(child->p_pptr); 1334 LIST_REMOVE(child, p_sibling); 1335 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1336 1337 clear_orphan(child); 1338 if (child->p_flag & P_TRACED) { 1339 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1340 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1341 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1342 p_orphan); 1343 } else { 1344 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1345 child, p_orphan); 1346 } 1347 child->p_treeflag |= P_TREE_ORPHANED; 1348 } 1349 1350 child->p_pptr = parent; 1351} 1352