kern_exit.c revision 280258
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 280258 2015-03-19 13:37:36Z rwatson $"); 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#ifdef KTRACE 77#include <sys/ktrace.h> 78#endif 79 80#include <security/audit/audit.h> 81#include <security/mac/mac_framework.h> 82 83#include <vm/vm.h> 84#include <vm/vm_extern.h> 85#include <vm/vm_param.h> 86#include <vm/pmap.h> 87#include <vm/vm_map.h> 88#include <vm/vm_page.h> 89#include <vm/uma.h> 90 91#ifdef KDTRACE_HOOKS 92#include <sys/dtrace_bsd.h> 93dtrace_execexit_func_t dtrace_fasttrap_exit; 94#endif 95 96SDT_PROVIDER_DECLARE(proc); 97SDT_PROBE_DEFINE1(proc, kernel, , exit, "int"); 98 99/* Hook for NFS teardown procedure. */ 100void (*nlminfo_release_p)(struct proc *p); 101 102struct proc * 103proc_realparent(struct proc *child) 104{ 105 struct proc *p, *parent; 106 107 sx_assert(&proctree_lock, SX_LOCKED); 108 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 109 if (child->p_oppid == 0 || 110 child->p_pptr->p_pid == child->p_oppid) 111 parent = child->p_pptr; 112 else 113 parent = initproc; 114 return (parent); 115 } 116 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 117 /* Cannot use LIST_PREV(), since the list head is not known. */ 118 p = __containerof(p->p_orphan.le_prev, struct proc, 119 p_orphan.le_next); 120 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 121 ("missing P_ORPHAN %p", p)); 122 } 123 parent = __containerof(p->p_orphan.le_prev, struct proc, 124 p_orphans.lh_first); 125 return (parent); 126} 127 128void 129reaper_abandon_children(struct proc *p, bool exiting) 130{ 131 struct proc *p1, *p2, *ptmp; 132 133 sx_assert(&proctree_lock, SX_LOCKED); 134 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 135 if ((p->p_treeflag & P_TREE_REAPER) == 0) 136 return; 137 p1 = p->p_reaper; 138 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 139 LIST_REMOVE(p2, p_reapsibling); 140 p2->p_reaper = p1; 141 p2->p_reapsubtree = p->p_reapsubtree; 142 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 143 if (exiting && p2->p_pptr == p) { 144 PROC_LOCK(p2); 145 proc_reparent(p2, p1); 146 PROC_UNLOCK(p2); 147 } 148 } 149 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 150 p->p_treeflag &= ~P_TREE_REAPER; 151} 152 153static void 154clear_orphan(struct proc *p) 155{ 156 struct proc *p1; 157 158 sx_assert(&proctree_lock, SA_XLOCKED); 159 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 160 return; 161 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 162 p1 = LIST_NEXT(p, p_orphan); 163 if (p1 != NULL) 164 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 165 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 166 } 167 LIST_REMOVE(p, p_orphan); 168 p->p_treeflag &= ~P_TREE_ORPHANED; 169} 170 171/* 172 * exit -- death of process. 173 */ 174void 175sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 176{ 177 178 exit1(td, W_EXITCODE(uap->rval, 0)); 179 /* NOTREACHED */ 180} 181 182/* 183 * Exit: deallocate address space and other resources, change proc state to 184 * zombie, and unlink proc from allproc and parent's lists. Save exit status 185 * and rusage for wait(). Check for child processes and orphan them. 186 */ 187void 188exit1(struct thread *td, int rv) 189{ 190 struct proc *p, *nq, *q, *t; 191 struct thread *tdt; 192 struct vnode *ttyvp = NULL; 193 194 mtx_assert(&Giant, MA_NOTOWNED); 195 196 p = td->td_proc; 197 /* 198 * XXX in case we're rebooting we just let init die in order to 199 * work around an unsolved stack overflow seen very late during 200 * shutdown on sparc64 when the gmirror worker process exists. 201 */ 202 if (p == initproc && rebooting == 0) { 203 printf("init died (signal %d, exit %d)\n", 204 WTERMSIG(rv), WEXITSTATUS(rv)); 205 panic("Going nowhere without my init!"); 206 } 207 208 /* 209 * MUST abort all other threads before proceeding past here. 210 */ 211 PROC_LOCK(p); 212 /* 213 * First check if some other thread or external request got 214 * here before us. If so, act appropriately: exit or suspend. 215 * We must ensure that stop requests are handled before we set 216 * P_WEXIT. 217 */ 218 thread_suspend_check(0); 219 while (p->p_flag & P_HADTHREADS) { 220 /* 221 * Kill off the other threads. This requires 222 * some co-operation from other parts of the kernel 223 * so it may not be instantaneous. With this state set 224 * any thread entering the kernel from userspace will 225 * thread_exit() in trap(). Any thread attempting to 226 * sleep will return immediately with EINTR or EWOULDBLOCK 227 * which will hopefully force them to back out to userland 228 * freeing resources as they go. Any thread attempting 229 * to return to userland will thread_exit() from userret(). 230 * thread_exit() will unsuspend us when the last of the 231 * other threads exits. 232 * If there is already a thread singler after resumption, 233 * calling thread_single will fail; in that case, we just 234 * re-check all suspension request, the thread should 235 * either be suspended there or exit. 236 */ 237 if (!thread_single(p, SINGLE_EXIT)) 238 /* 239 * All other activity in this process is now 240 * stopped. Threading support has been turned 241 * off. 242 */ 243 break; 244 /* 245 * Recheck for new stop or suspend requests which 246 * might appear while process lock was dropped in 247 * thread_single(). 248 */ 249 thread_suspend_check(0); 250 } 251 KASSERT(p->p_numthreads == 1, 252 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 253 racct_sub(p, RACCT_NTHR, 1); 254 /* 255 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 256 * on our vmspace, so we should block below until they have 257 * released their reference to us. Note that if they have 258 * requested S_EXIT stops we will block here until they ack 259 * via PIOCCONT. 260 */ 261 _STOPEVENT(p, S_EXIT, rv); 262 263 /* 264 * Ignore any pending request to stop due to a stop signal. 265 * Once P_WEXIT is set, future requests will be ignored as 266 * well. 267 */ 268 p->p_flag &= ~P_STOPPED_SIG; 269 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 270 271 /* 272 * Note that we are exiting and do another wakeup of anyone in 273 * PIOCWAIT in case they aren't listening for S_EXIT stops or 274 * decided to wait again after we told them we are exiting. 275 */ 276 p->p_flag |= P_WEXIT; 277 wakeup(&p->p_stype); 278 279 /* 280 * Wait for any processes that have a hold on our vmspace to 281 * release their reference. 282 */ 283 while (p->p_lock > 0) 284 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 285 286 p->p_xstat = rv; /* Let event handler change exit status */ 287 PROC_UNLOCK(p); 288 /* Drain the limit callout while we don't have the proc locked */ 289 callout_drain(&p->p_limco); 290 291#ifdef AUDIT 292 /* 293 * The Sun BSM exit token contains two components: an exit status as 294 * passed to exit(), and a return value to indicate what sort of exit 295 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 296 * what the return value is. 297 */ 298 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0); 299 AUDIT_SYSCALL_EXIT(0, td); 300#endif 301 302 /* Are we a task leader? */ 303 if (p == p->p_leader) { 304 mtx_lock(&ppeers_lock); 305 q = p->p_peers; 306 while (q != NULL) { 307 PROC_LOCK(q); 308 kern_psignal(q, SIGKILL); 309 PROC_UNLOCK(q); 310 q = q->p_peers; 311 } 312 while (p->p_peers != NULL) 313 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 314 mtx_unlock(&ppeers_lock); 315 } 316 317 /* 318 * Check if any loadable modules need anything done at process exit. 319 * E.g. SYSV IPC stuff 320 * XXX what if one of these generates an error? 321 */ 322 EVENTHANDLER_INVOKE(process_exit, p); 323 324 /* 325 * If parent is waiting for us to exit or exec, 326 * P_PPWAIT is set; we will wakeup the parent below. 327 */ 328 PROC_LOCK(p); 329 rv = p->p_xstat; /* Event handler could change exit status */ 330 stopprofclock(p); 331 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 332 333 /* 334 * Stop the real interval timer. If the handler is currently 335 * executing, prevent it from rearming itself and let it finish. 336 */ 337 if (timevalisset(&p->p_realtimer.it_value) && 338 callout_stop(&p->p_itcallout) == 0) { 339 timevalclear(&p->p_realtimer.it_interval); 340 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 341 KASSERT(!timevalisset(&p->p_realtimer.it_value), 342 ("realtime timer is still armed")); 343 } 344 PROC_UNLOCK(p); 345 346 /* 347 * Reset any sigio structures pointing to us as a result of 348 * F_SETOWN with our pid. 349 */ 350 funsetownlst(&p->p_sigiolst); 351 352 /* 353 * If this process has an nlminfo data area (for lockd), release it 354 */ 355 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 356 (*nlminfo_release_p)(p); 357 358 /* 359 * Close open files and release open-file table. 360 * This may block! 361 */ 362 fdescfree(td); 363 364 /* 365 * If this thread tickled GEOM, we need to wait for the giggling to 366 * stop before we return to userland 367 */ 368 if (td->td_pflags & TDP_GEOM) 369 g_waitidle(); 370 371 /* 372 * Remove ourself from our leader's peer list and wake our leader. 373 */ 374 mtx_lock(&ppeers_lock); 375 if (p->p_leader->p_peers) { 376 q = p->p_leader; 377 while (q->p_peers != p) 378 q = q->p_peers; 379 q->p_peers = p->p_peers; 380 wakeup(p->p_leader); 381 } 382 mtx_unlock(&ppeers_lock); 383 384 vmspace_exit(td); 385 386 sx_xlock(&proctree_lock); 387 if (SESS_LEADER(p)) { 388 struct session *sp = p->p_session; 389 struct tty *tp; 390 391 /* 392 * s_ttyp is not zero'd; we use this to indicate that 393 * the session once had a controlling terminal. (for 394 * logging and informational purposes) 395 */ 396 SESS_LOCK(sp); 397 ttyvp = sp->s_ttyvp; 398 tp = sp->s_ttyp; 399 sp->s_ttyvp = NULL; 400 sp->s_ttydp = NULL; 401 sp->s_leader = NULL; 402 SESS_UNLOCK(sp); 403 404 /* 405 * Signal foreground pgrp and revoke access to 406 * controlling terminal if it has not been revoked 407 * already. 408 * 409 * Because the TTY may have been revoked in the mean 410 * time and could already have a new session associated 411 * with it, make sure we don't send a SIGHUP to a 412 * foreground process group that does not belong to this 413 * session. 414 */ 415 416 if (tp != NULL) { 417 tty_lock(tp); 418 if (tp->t_session == sp) 419 tty_signal_pgrp(tp, SIGHUP); 420 tty_unlock(tp); 421 } 422 423 if (ttyvp != NULL) { 424 sx_xunlock(&proctree_lock); 425 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 426 VOP_REVOKE(ttyvp, REVOKEALL); 427 VOP_UNLOCK(ttyvp, 0); 428 } 429 sx_xlock(&proctree_lock); 430 } 431 } 432 fixjobc(p, p->p_pgrp, 0); 433 sx_xunlock(&proctree_lock); 434 (void)acct_process(td); 435 436 /* Release the TTY now we've unlocked everything. */ 437 if (ttyvp != NULL) 438 vrele(ttyvp); 439#ifdef KTRACE 440 ktrprocexit(td); 441#endif 442 /* 443 * Release reference to text vnode 444 */ 445 if (p->p_textvp != NULL) { 446 vrele(p->p_textvp); 447 p->p_textvp = NULL; 448 } 449 450 /* 451 * Release our limits structure. 452 */ 453 lim_free(p->p_limit); 454 p->p_limit = NULL; 455 456 tidhash_remove(td); 457 458 /* 459 * Remove proc from allproc queue and pidhash chain. 460 * Place onto zombproc. Unlink from parent's child list. 461 */ 462 sx_xlock(&allproc_lock); 463 LIST_REMOVE(p, p_list); 464 LIST_INSERT_HEAD(&zombproc, p, p_list); 465 LIST_REMOVE(p, p_hash); 466 sx_xunlock(&allproc_lock); 467 468 /* 469 * Call machine-dependent code to release any 470 * machine-dependent resources other than the address space. 471 * The address space is released by "vmspace_exitfree(p)" in 472 * vm_waitproc(). 473 */ 474 cpu_exit(td); 475 476 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 477 478 /* 479 * Reparent all children processes: 480 * - traced ones to the original parent (or init if we are that parent) 481 * - the rest to init 482 */ 483 sx_xlock(&proctree_lock); 484 q = LIST_FIRST(&p->p_children); 485 if (q != NULL) /* only need this if any child is S_ZOMB */ 486 wakeup(q->p_reaper); 487 for (; q != NULL; q = nq) { 488 nq = LIST_NEXT(q, p_sibling); 489 PROC_LOCK(q); 490 q->p_sigparent = SIGCHLD; 491 492 if (!(q->p_flag & P_TRACED)) { 493 proc_reparent(q, q->p_reaper); 494 } else { 495 /* 496 * Traced processes are killed since their existence 497 * means someone is screwing up. 498 */ 499 t = proc_realparent(q); 500 if (t == p) { 501 proc_reparent(q, q->p_reaper); 502 } else { 503 PROC_LOCK(t); 504 proc_reparent(q, t); 505 PROC_UNLOCK(t); 506 } 507 /* 508 * Since q was found on our children list, the 509 * proc_reparent() call moved q to the orphan 510 * list due to present P_TRACED flag. Clear 511 * orphan link for q now while q is locked. 512 */ 513 clear_orphan(q); 514 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 515 FOREACH_THREAD_IN_PROC(q, tdt) 516 tdt->td_dbgflags &= ~TDB_SUSPEND; 517 kern_psignal(q, SIGKILL); 518 } 519 PROC_UNLOCK(q); 520 } 521 522 /* 523 * Also get rid of our orphans. 524 */ 525 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 526 PROC_LOCK(q); 527 clear_orphan(q); 528 PROC_UNLOCK(q); 529 } 530 531 /* Save exit status. */ 532 PROC_LOCK(p); 533 p->p_xthread = td; 534 535 /* Tell the prison that we are gone. */ 536 prison_proc_free(p->p_ucred->cr_prison); 537 538#ifdef KDTRACE_HOOKS 539 /* 540 * Tell the DTrace fasttrap provider about the exit if it 541 * has declared an interest. 542 */ 543 if (dtrace_fasttrap_exit) 544 dtrace_fasttrap_exit(p); 545#endif 546 547 /* 548 * Notify interested parties of our demise. 549 */ 550 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 551 552#ifdef KDTRACE_HOOKS 553 int reason = CLD_EXITED; 554 if (WCOREDUMP(rv)) 555 reason = CLD_DUMPED; 556 else if (WIFSIGNALED(rv)) 557 reason = CLD_KILLED; 558 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 559#endif 560 561 /* 562 * Just delete all entries in the p_klist. At this point we won't 563 * report any more events, and there are nasty race conditions that 564 * can beat us if we don't. 565 */ 566 knlist_clear(&p->p_klist, 1); 567 568 /* 569 * If this is a process with a descriptor, we may not need to deliver 570 * a signal to the parent. proctree_lock is held over 571 * procdesc_exit() to serialize concurrent calls to close() and 572 * exit(). 573 */ 574#ifdef PROCDESC 575 if (p->p_procdesc == NULL || procdesc_exit(p)) { 576#endif 577 /* 578 * Notify parent that we're gone. If parent has the 579 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 580 * notify process 1 instead (and hope it will handle this 581 * situation). 582 */ 583 PROC_LOCK(p->p_pptr); 584 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 585 if (p->p_pptr->p_sigacts->ps_flag & 586 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 587 struct proc *pp; 588 589 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 590 pp = p->p_pptr; 591 PROC_UNLOCK(pp); 592 proc_reparent(p, p->p_reaper); 593 p->p_sigparent = SIGCHLD; 594 PROC_LOCK(p->p_pptr); 595 596 /* 597 * Notify parent, so in case he was wait(2)ing or 598 * executing waitpid(2) with our pid, he will 599 * continue. 600 */ 601 wakeup(pp); 602 } else 603 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 604 605 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 606 childproc_exited(p); 607 else if (p->p_sigparent != 0) { 608 if (p->p_sigparent == SIGCHLD) 609 childproc_exited(p); 610 else /* LINUX thread */ 611 kern_psignal(p->p_pptr, p->p_sigparent); 612 } 613#ifdef PROCDESC 614 } else 615 PROC_LOCK(p->p_pptr); 616#endif 617 sx_xunlock(&proctree_lock); 618 619 /* 620 * The state PRS_ZOMBIE prevents other proesses from sending 621 * signal to the process, to avoid memory leak, we free memory 622 * for signal queue at the time when the state is set. 623 */ 624 sigqueue_flush(&p->p_sigqueue); 625 sigqueue_flush(&td->td_sigqueue); 626 627 /* 628 * We have to wait until after acquiring all locks before 629 * changing p_state. We need to avoid all possible context 630 * switches (including ones from blocking on a mutex) while 631 * marked as a zombie. We also have to set the zombie state 632 * before we release the parent process' proc lock to avoid 633 * a lost wakeup. So, we first call wakeup, then we grab the 634 * sched lock, update the state, and release the parent process' 635 * proc lock. 636 */ 637 wakeup(p->p_pptr); 638 cv_broadcast(&p->p_pwait); 639 sched_exit(p->p_pptr, td); 640 PROC_SLOCK(p); 641 p->p_state = PRS_ZOMBIE; 642 PROC_UNLOCK(p->p_pptr); 643 644 /* 645 * Hopefully no one will try to deliver a signal to the process this 646 * late in the game. 647 */ 648 knlist_destroy(&p->p_klist); 649 650 /* 651 * Save our children's rusage information in our exit rusage. 652 */ 653 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 654 655 /* 656 * Make sure the scheduler takes this thread out of its tables etc. 657 * This will also release this thread's reference to the ucred. 658 * Other thread parts to release include pcb bits and such. 659 */ 660 thread_exit(); 661} 662 663 664#ifndef _SYS_SYSPROTO_H_ 665struct abort2_args { 666 char *why; 667 int nargs; 668 void **args; 669}; 670#endif 671 672int 673sys_abort2(struct thread *td, struct abort2_args *uap) 674{ 675 struct proc *p = td->td_proc; 676 struct sbuf *sb; 677 void *uargs[16]; 678 int error, i, sig; 679 680 /* 681 * Do it right now so we can log either proper call of abort2(), or 682 * note, that invalid argument was passed. 512 is big enough to 683 * handle 16 arguments' descriptions with additional comments. 684 */ 685 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 686 sbuf_clear(sb); 687 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 688 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 689 /* 690 * Since we can't return from abort2(), send SIGKILL in cases, where 691 * abort2() was called improperly 692 */ 693 sig = SIGKILL; 694 /* Prevent from DoSes from user-space. */ 695 if (uap->nargs < 0 || uap->nargs > 16) 696 goto out; 697 if (uap->nargs > 0) { 698 if (uap->args == NULL) 699 goto out; 700 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 701 if (error != 0) 702 goto out; 703 } 704 /* 705 * Limit size of 'reason' string to 128. Will fit even when 706 * maximal number of arguments was chosen to be logged. 707 */ 708 if (uap->why != NULL) { 709 error = sbuf_copyin(sb, uap->why, 128); 710 if (error < 0) 711 goto out; 712 } else { 713 sbuf_printf(sb, "(null)"); 714 } 715 if (uap->nargs > 0) { 716 sbuf_printf(sb, "("); 717 for (i = 0;i < uap->nargs; i++) 718 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 719 sbuf_printf(sb, ")"); 720 } 721 /* 722 * Final stage: arguments were proper, string has been 723 * successfully copied from userspace, and copying pointers 724 * from user-space succeed. 725 */ 726 sig = SIGABRT; 727out: 728 if (sig == SIGKILL) { 729 sbuf_trim(sb); 730 sbuf_printf(sb, " (Reason text inaccessible)"); 731 } 732 sbuf_cat(sb, "\n"); 733 sbuf_finish(sb); 734 log(LOG_INFO, "%s", sbuf_data(sb)); 735 sbuf_delete(sb); 736 exit1(td, W_EXITCODE(0, sig)); 737 return (0); 738} 739 740 741#ifdef COMPAT_43 742/* 743 * The dirty work is handled by kern_wait(). 744 */ 745int 746owait(struct thread *td, struct owait_args *uap __unused) 747{ 748 int error, status; 749 750 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 751 if (error == 0) 752 td->td_retval[1] = status; 753 return (error); 754} 755#endif /* COMPAT_43 */ 756 757/* 758 * The dirty work is handled by kern_wait(). 759 */ 760int 761sys_wait4(struct thread *td, struct wait4_args *uap) 762{ 763 struct rusage ru, *rup; 764 int error, status; 765 766 if (uap->rusage != NULL) 767 rup = &ru; 768 else 769 rup = NULL; 770 error = kern_wait(td, uap->pid, &status, uap->options, rup); 771 if (uap->status != NULL && error == 0) 772 error = copyout(&status, uap->status, sizeof(status)); 773 if (uap->rusage != NULL && error == 0) 774 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 775 return (error); 776} 777 778int 779sys_wait6(struct thread *td, struct wait6_args *uap) 780{ 781 struct __wrusage wru, *wrup; 782 siginfo_t si, *sip; 783 idtype_t idtype; 784 id_t id; 785 int error, status; 786 787 idtype = uap->idtype; 788 id = uap->id; 789 790 if (uap->wrusage != NULL) 791 wrup = &wru; 792 else 793 wrup = NULL; 794 795 if (uap->info != NULL) { 796 sip = &si; 797 bzero(sip, sizeof(*sip)); 798 } else 799 sip = NULL; 800 801 /* 802 * We expect all callers of wait6() to know about WEXITED and 803 * WTRAPPED. 804 */ 805 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 806 807 if (uap->status != NULL && error == 0) 808 error = copyout(&status, uap->status, sizeof(status)); 809 if (uap->wrusage != NULL && error == 0) 810 error = copyout(&wru, uap->wrusage, sizeof(wru)); 811 if (uap->info != NULL && error == 0) 812 error = copyout(&si, uap->info, sizeof(si)); 813 return (error); 814} 815 816/* 817 * Reap the remains of a zombie process and optionally return status and 818 * rusage. Asserts and will release both the proctree_lock and the process 819 * lock as part of its work. 820 */ 821void 822proc_reap(struct thread *td, struct proc *p, int *status, int options) 823{ 824 struct proc *q, *t; 825 826 sx_assert(&proctree_lock, SA_XLOCKED); 827 PROC_LOCK_ASSERT(p, MA_OWNED); 828 PROC_SLOCK_ASSERT(p, MA_OWNED); 829 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 830 831 q = td->td_proc; 832 833 PROC_SUNLOCK(p); 834 td->td_retval[0] = p->p_pid; 835 if (status) 836 *status = p->p_xstat; /* convert to int */ 837 if (options & WNOWAIT) { 838 /* 839 * Only poll, returning the status. Caller does not wish to 840 * release the proc struct just yet. 841 */ 842 PROC_UNLOCK(p); 843 sx_xunlock(&proctree_lock); 844 return; 845 } 846 847 PROC_LOCK(q); 848 sigqueue_take(p->p_ksi); 849 PROC_UNLOCK(q); 850 PROC_UNLOCK(p); 851 852 /* 853 * If we got the child via a ptrace 'attach', we need to give it back 854 * to the old parent. 855 */ 856 if (p->p_oppid != 0) { 857 t = proc_realparent(p); 858 PROC_LOCK(t); 859 PROC_LOCK(p); 860 proc_reparent(p, t); 861 p->p_oppid = 0; 862 PROC_UNLOCK(p); 863 pksignal(t, SIGCHLD, p->p_ksi); 864 wakeup(t); 865 cv_broadcast(&p->p_pwait); 866 PROC_UNLOCK(t); 867 sx_xunlock(&proctree_lock); 868 return; 869 } 870 871 /* 872 * Remove other references to this process to ensure we have an 873 * exclusive reference. 874 */ 875 sx_xlock(&allproc_lock); 876 LIST_REMOVE(p, p_list); /* off zombproc */ 877 sx_xunlock(&allproc_lock); 878 LIST_REMOVE(p, p_sibling); 879 reaper_abandon_children(p, true); 880 LIST_REMOVE(p, p_reapsibling); 881 PROC_LOCK(p); 882 clear_orphan(p); 883 PROC_UNLOCK(p); 884 leavepgrp(p); 885#ifdef PROCDESC 886 if (p->p_procdesc != NULL) 887 procdesc_reap(p); 888#endif 889 sx_xunlock(&proctree_lock); 890 891 /* 892 * As a side effect of this lock, we know that all other writes to 893 * this proc are visible now, so no more locking is needed for p. 894 */ 895 PROC_LOCK(p); 896 p->p_xstat = 0; /* XXX: why? */ 897 PROC_UNLOCK(p); 898 PROC_LOCK(q); 899 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 900 PROC_UNLOCK(q); 901 902 /* 903 * Decrement the count of procs running with this uid. 904 */ 905 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 906 907 /* 908 * Destroy resource accounting information associated with the process. 909 */ 910#ifdef RACCT 911 PROC_LOCK(p); 912 racct_sub(p, RACCT_NPROC, 1); 913 PROC_UNLOCK(p); 914#endif 915 racct_proc_exit(p); 916 917 /* 918 * Free credentials, arguments, and sigacts. 919 */ 920 crfree(p->p_ucred); 921 p->p_ucred = NULL; 922 pargs_drop(p->p_args); 923 p->p_args = NULL; 924 sigacts_free(p->p_sigacts); 925 p->p_sigacts = NULL; 926 927 /* 928 * Do any thread-system specific cleanups. 929 */ 930 thread_wait(p); 931 932 /* 933 * Give vm and machine-dependent layer a chance to free anything that 934 * cpu_exit couldn't release while still running in process context. 935 */ 936 vm_waitproc(p); 937#ifdef MAC 938 mac_proc_destroy(p); 939#endif 940 KASSERT(FIRST_THREAD_IN_PROC(p), 941 ("proc_reap: no residual thread!")); 942 uma_zfree(proc_zone, p); 943 sx_xlock(&allproc_lock); 944 nprocs--; 945 sx_xunlock(&allproc_lock); 946} 947 948static int 949proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 950 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo) 951{ 952 struct proc *q; 953 struct rusage *rup; 954 955 sx_assert(&proctree_lock, SA_XLOCKED); 956 957 q = td->td_proc; 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