kern_exit.c revision 270264
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 270264 2014-08-21 10:46:19Z 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/capability.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 return (child->p_pptr->p_pid == child->p_oppid ? 110 child->p_pptr : initproc); 111 } 112 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 113 /* Cannot use LIST_PREV(), since the list head is not known. */ 114 p = __containerof(p->p_orphan.le_prev, struct proc, 115 p_orphan.le_next); 116 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 117 ("missing P_ORPHAN %p", p)); 118 } 119 parent = __containerof(p->p_orphan.le_prev, struct proc, 120 p_orphans.lh_first); 121 return (parent); 122} 123 124static void 125clear_orphan(struct proc *p) 126{ 127 struct proc *p1; 128 129 sx_assert(&proctree_lock, SA_XLOCKED); 130 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 131 return; 132 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 133 p1 = LIST_NEXT(p, p_orphan); 134 if (p1 != NULL) 135 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 136 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 137 } 138 LIST_REMOVE(p, p_orphan); 139 p->p_treeflag &= ~P_TREE_ORPHANED; 140} 141 142/* 143 * exit -- death of process. 144 */ 145void 146sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 147{ 148 149 exit1(td, W_EXITCODE(uap->rval, 0)); 150 /* NOTREACHED */ 151} 152 153/* 154 * Exit: deallocate address space and other resources, change proc state to 155 * zombie, and unlink proc from allproc and parent's lists. Save exit status 156 * and rusage for wait(). Check for child processes and orphan them. 157 */ 158void 159exit1(struct thread *td, int rv) 160{ 161 struct proc *p, *nq, *q; 162 struct vnode *ttyvp = NULL; 163 164 mtx_assert(&Giant, MA_NOTOWNED); 165 166 p = td->td_proc; 167 /* 168 * XXX in case we're rebooting we just let init die in order to 169 * work around an unsolved stack overflow seen very late during 170 * shutdown on sparc64 when the gmirror worker process exists. 171 */ 172 if (p == initproc && rebooting == 0) { 173 printf("init died (signal %d, exit %d)\n", 174 WTERMSIG(rv), WEXITSTATUS(rv)); 175 panic("Going nowhere without my init!"); 176 } 177 178 /* 179 * MUST abort all other threads before proceeding past here. 180 */ 181 PROC_LOCK(p); 182 while (p->p_flag & P_HADTHREADS) { 183 /* 184 * First check if some other thread got here before us. 185 * If so, act appropriately: exit or suspend. 186 */ 187 thread_suspend_check(0); 188 189 /* 190 * Kill off the other threads. This requires 191 * some co-operation from other parts of the kernel 192 * so it may not be instantaneous. With this state set 193 * any thread entering the kernel from userspace will 194 * thread_exit() in trap(). Any thread attempting to 195 * sleep will return immediately with EINTR or EWOULDBLOCK 196 * which will hopefully force them to back out to userland 197 * freeing resources as they go. Any thread attempting 198 * to return to userland will thread_exit() from userret(). 199 * thread_exit() will unsuspend us when the last of the 200 * other threads exits. 201 * If there is already a thread singler after resumption, 202 * calling thread_single will fail; in that case, we just 203 * re-check all suspension request, the thread should 204 * either be suspended there or exit. 205 */ 206 if (!thread_single(SINGLE_EXIT)) 207 break; 208 209 /* 210 * All other activity in this process is now stopped. 211 * Threading support has been turned off. 212 */ 213 } 214 KASSERT(p->p_numthreads == 1, 215 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 216 racct_sub(p, RACCT_NTHR, 1); 217 /* 218 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 219 * on our vmspace, so we should block below until they have 220 * released their reference to us. Note that if they have 221 * requested S_EXIT stops we will block here until they ack 222 * via PIOCCONT. 223 */ 224 _STOPEVENT(p, S_EXIT, rv); 225 226 /* 227 * Ignore any pending request to stop due to a stop signal. 228 * Once P_WEXIT is set, future requests will be ignored as 229 * well. 230 */ 231 p->p_flag &= ~P_STOPPED_SIG; 232 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 233 234 /* 235 * Note that we are exiting and do another wakeup of anyone in 236 * PIOCWAIT in case they aren't listening for S_EXIT stops or 237 * decided to wait again after we told them we are exiting. 238 */ 239 p->p_flag |= P_WEXIT; 240 wakeup(&p->p_stype); 241 242 /* 243 * Wait for any processes that have a hold on our vmspace to 244 * release their reference. 245 */ 246 while (p->p_lock > 0) 247 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 248 249 p->p_xstat = rv; /* Let event handler change exit status */ 250 PROC_UNLOCK(p); 251 /* Drain the limit callout while we don't have the proc locked */ 252 callout_drain(&p->p_limco); 253 254#ifdef AUDIT 255 /* 256 * The Sun BSM exit token contains two components: an exit status as 257 * passed to exit(), and a return value to indicate what sort of exit 258 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 259 * what the return value is. 260 */ 261 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0); 262 AUDIT_SYSCALL_EXIT(0, td); 263#endif 264 265 /* Are we a task leader? */ 266 if (p == p->p_leader) { 267 mtx_lock(&ppeers_lock); 268 q = p->p_peers; 269 while (q != NULL) { 270 PROC_LOCK(q); 271 kern_psignal(q, SIGKILL); 272 PROC_UNLOCK(q); 273 q = q->p_peers; 274 } 275 while (p->p_peers != NULL) 276 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 277 mtx_unlock(&ppeers_lock); 278 } 279 280 /* 281 * Check if any loadable modules need anything done at process exit. 282 * E.g. SYSV IPC stuff 283 * XXX what if one of these generates an error? 284 */ 285 EVENTHANDLER_INVOKE(process_exit, p); 286 287 /* 288 * If parent is waiting for us to exit or exec, 289 * P_PPWAIT is set; we will wakeup the parent below. 290 */ 291 PROC_LOCK(p); 292 rv = p->p_xstat; /* Event handler could change exit status */ 293 stopprofclock(p); 294 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 295 296 /* 297 * Stop the real interval timer. If the handler is currently 298 * executing, prevent it from rearming itself and let it finish. 299 */ 300 if (timevalisset(&p->p_realtimer.it_value) && 301 callout_stop(&p->p_itcallout) == 0) { 302 timevalclear(&p->p_realtimer.it_interval); 303 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 304 KASSERT(!timevalisset(&p->p_realtimer.it_value), 305 ("realtime timer is still armed")); 306 } 307 PROC_UNLOCK(p); 308 309 /* 310 * Reset any sigio structures pointing to us as a result of 311 * F_SETOWN with our pid. 312 */ 313 funsetownlst(&p->p_sigiolst); 314 315 /* 316 * If this process has an nlminfo data area (for lockd), release it 317 */ 318 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 319 (*nlminfo_release_p)(p); 320 321 /* 322 * Close open files and release open-file table. 323 * This may block! 324 */ 325 fdescfree(td); 326 327 /* 328 * If this thread tickled GEOM, we need to wait for the giggling to 329 * stop before we return to userland 330 */ 331 if (td->td_pflags & TDP_GEOM) 332 g_waitidle(); 333 334 /* 335 * Remove ourself from our leader's peer list and wake our leader. 336 */ 337 mtx_lock(&ppeers_lock); 338 if (p->p_leader->p_peers) { 339 q = p->p_leader; 340 while (q->p_peers != p) 341 q = q->p_peers; 342 q->p_peers = p->p_peers; 343 wakeup(p->p_leader); 344 } 345 mtx_unlock(&ppeers_lock); 346 347 vmspace_exit(td); 348 349 sx_xlock(&proctree_lock); 350 if (SESS_LEADER(p)) { 351 struct session *sp = p->p_session; 352 struct tty *tp; 353 354 /* 355 * s_ttyp is not zero'd; we use this to indicate that 356 * the session once had a controlling terminal. (for 357 * logging and informational purposes) 358 */ 359 SESS_LOCK(sp); 360 ttyvp = sp->s_ttyvp; 361 tp = sp->s_ttyp; 362 sp->s_ttyvp = NULL; 363 sp->s_ttydp = NULL; 364 sp->s_leader = NULL; 365 SESS_UNLOCK(sp); 366 367 /* 368 * Signal foreground pgrp and revoke access to 369 * controlling terminal if it has not been revoked 370 * already. 371 * 372 * Because the TTY may have been revoked in the mean 373 * time and could already have a new session associated 374 * with it, make sure we don't send a SIGHUP to a 375 * foreground process group that does not belong to this 376 * session. 377 */ 378 379 if (tp != NULL) { 380 tty_lock(tp); 381 if (tp->t_session == sp) 382 tty_signal_pgrp(tp, SIGHUP); 383 tty_unlock(tp); 384 } 385 386 if (ttyvp != NULL) { 387 sx_xunlock(&proctree_lock); 388 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 389 VOP_REVOKE(ttyvp, REVOKEALL); 390 VOP_UNLOCK(ttyvp, 0); 391 } 392 sx_xlock(&proctree_lock); 393 } 394 } 395 fixjobc(p, p->p_pgrp, 0); 396 sx_xunlock(&proctree_lock); 397 (void)acct_process(td); 398 399 /* Release the TTY now we've unlocked everything. */ 400 if (ttyvp != NULL) 401 vrele(ttyvp); 402#ifdef KTRACE 403 ktrprocexit(td); 404#endif 405 /* 406 * Release reference to text vnode 407 */ 408 if (p->p_textvp != NULL) { 409 vrele(p->p_textvp); 410 p->p_textvp = NULL; 411 } 412 413 /* 414 * Release our limits structure. 415 */ 416 lim_free(p->p_limit); 417 p->p_limit = NULL; 418 419 tidhash_remove(td); 420 421 /* 422 * Remove proc from allproc queue and pidhash chain. 423 * Place onto zombproc. Unlink from parent's child list. 424 */ 425 sx_xlock(&allproc_lock); 426 LIST_REMOVE(p, p_list); 427 LIST_INSERT_HEAD(&zombproc, p, p_list); 428 LIST_REMOVE(p, p_hash); 429 sx_xunlock(&allproc_lock); 430 431 /* 432 * Call machine-dependent code to release any 433 * machine-dependent resources other than the address space. 434 * The address space is released by "vmspace_exitfree(p)" in 435 * vm_waitproc(). 436 */ 437 cpu_exit(td); 438 439 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 440 441 /* 442 * Reparent all of our children to init. 443 */ 444 sx_xlock(&proctree_lock); 445 q = LIST_FIRST(&p->p_children); 446 if (q != NULL) /* only need this if any child is S_ZOMB */ 447 wakeup(initproc); 448 for (; q != NULL; q = nq) { 449 nq = LIST_NEXT(q, p_sibling); 450 PROC_LOCK(q); 451 proc_reparent(q, initproc); 452 q->p_sigparent = SIGCHLD; 453 /* 454 * Traced processes are killed 455 * since their existence means someone is screwing up. 456 */ 457 if (q->p_flag & P_TRACED) { 458 struct thread *temp; 459 460 /* 461 * Since q was found on our children list, the 462 * proc_reparent() call moved q to the orphan 463 * list due to present P_TRACED flag. Clear 464 * orphan link for q now while q is locked. 465 */ 466 clear_orphan(q); 467 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 468 FOREACH_THREAD_IN_PROC(q, temp) 469 temp->td_dbgflags &= ~TDB_SUSPEND; 470 kern_psignal(q, SIGKILL); 471 } 472 PROC_UNLOCK(q); 473 } 474 475 /* 476 * Also get rid of our orphans. 477 */ 478 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 479 PROC_LOCK(q); 480 clear_orphan(q); 481 PROC_UNLOCK(q); 482 } 483 484 /* Save exit status. */ 485 PROC_LOCK(p); 486 p->p_xthread = td; 487 488 /* Tell the prison that we are gone. */ 489 prison_proc_free(p->p_ucred->cr_prison); 490 491#ifdef KDTRACE_HOOKS 492 /* 493 * Tell the DTrace fasttrap provider about the exit if it 494 * has declared an interest. 495 */ 496 if (dtrace_fasttrap_exit) 497 dtrace_fasttrap_exit(p); 498#endif 499 500 /* 501 * Notify interested parties of our demise. 502 */ 503 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 504 505#ifdef KDTRACE_HOOKS 506 int reason = CLD_EXITED; 507 if (WCOREDUMP(rv)) 508 reason = CLD_DUMPED; 509 else if (WIFSIGNALED(rv)) 510 reason = CLD_KILLED; 511 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 512#endif 513 514 /* 515 * Just delete all entries in the p_klist. At this point we won't 516 * report any more events, and there are nasty race conditions that 517 * can beat us if we don't. 518 */ 519 knlist_clear(&p->p_klist, 1); 520 521 /* 522 * If this is a process with a descriptor, we may not need to deliver 523 * a signal to the parent. proctree_lock is held over 524 * procdesc_exit() to serialize concurrent calls to close() and 525 * exit(). 526 */ 527#ifdef PROCDESC 528 if (p->p_procdesc == NULL || procdesc_exit(p)) { 529#endif 530 /* 531 * Notify parent that we're gone. If parent has the 532 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 533 * notify process 1 instead (and hope it will handle this 534 * situation). 535 */ 536 PROC_LOCK(p->p_pptr); 537 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 538 if (p->p_pptr->p_sigacts->ps_flag & 539 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 540 struct proc *pp; 541 542 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 543 pp = p->p_pptr; 544 PROC_UNLOCK(pp); 545 proc_reparent(p, initproc); 546 p->p_sigparent = SIGCHLD; 547 PROC_LOCK(p->p_pptr); 548 549 /* 550 * Notify parent, so in case he was wait(2)ing or 551 * executing waitpid(2) with our pid, he will 552 * continue. 553 */ 554 wakeup(pp); 555 } else 556 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 557 558 if (p->p_pptr == initproc) 559 kern_psignal(p->p_pptr, SIGCHLD); 560 else if (p->p_sigparent != 0) { 561 if (p->p_sigparent == SIGCHLD) 562 childproc_exited(p); 563 else /* LINUX thread */ 564 kern_psignal(p->p_pptr, p->p_sigparent); 565 } 566#ifdef PROCDESC 567 } else 568 PROC_LOCK(p->p_pptr); 569#endif 570 sx_xunlock(&proctree_lock); 571 572 /* 573 * The state PRS_ZOMBIE prevents other proesses from sending 574 * signal to the process, to avoid memory leak, we free memory 575 * for signal queue at the time when the state is set. 576 */ 577 sigqueue_flush(&p->p_sigqueue); 578 sigqueue_flush(&td->td_sigqueue); 579 580 /* 581 * We have to wait until after acquiring all locks before 582 * changing p_state. We need to avoid all possible context 583 * switches (including ones from blocking on a mutex) while 584 * marked as a zombie. We also have to set the zombie state 585 * before we release the parent process' proc lock to avoid 586 * a lost wakeup. So, we first call wakeup, then we grab the 587 * sched lock, update the state, and release the parent process' 588 * proc lock. 589 */ 590 wakeup(p->p_pptr); 591 cv_broadcast(&p->p_pwait); 592 sched_exit(p->p_pptr, td); 593 PROC_SLOCK(p); 594 p->p_state = PRS_ZOMBIE; 595 PROC_UNLOCK(p->p_pptr); 596 597 /* 598 * Hopefully no one will try to deliver a signal to the process this 599 * late in the game. 600 */ 601 knlist_destroy(&p->p_klist); 602 603 /* 604 * Save our children's rusage information in our exit rusage. 605 */ 606 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 607 608 /* 609 * Make sure the scheduler takes this thread out of its tables etc. 610 * This will also release this thread's reference to the ucred. 611 * Other thread parts to release include pcb bits and such. 612 */ 613 thread_exit(); 614} 615 616 617#ifndef _SYS_SYSPROTO_H_ 618struct abort2_args { 619 char *why; 620 int nargs; 621 void **args; 622}; 623#endif 624 625int 626sys_abort2(struct thread *td, struct abort2_args *uap) 627{ 628 struct proc *p = td->td_proc; 629 struct sbuf *sb; 630 void *uargs[16]; 631 int error, i, sig; 632 633 /* 634 * Do it right now so we can log either proper call of abort2(), or 635 * note, that invalid argument was passed. 512 is big enough to 636 * handle 16 arguments' descriptions with additional comments. 637 */ 638 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 639 sbuf_clear(sb); 640 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 641 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 642 /* 643 * Since we can't return from abort2(), send SIGKILL in cases, where 644 * abort2() was called improperly 645 */ 646 sig = SIGKILL; 647 /* Prevent from DoSes from user-space. */ 648 if (uap->nargs < 0 || uap->nargs > 16) 649 goto out; 650 if (uap->nargs > 0) { 651 if (uap->args == NULL) 652 goto out; 653 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 654 if (error != 0) 655 goto out; 656 } 657 /* 658 * Limit size of 'reason' string to 128. Will fit even when 659 * maximal number of arguments was chosen to be logged. 660 */ 661 if (uap->why != NULL) { 662 error = sbuf_copyin(sb, uap->why, 128); 663 if (error < 0) 664 goto out; 665 } else { 666 sbuf_printf(sb, "(null)"); 667 } 668 if (uap->nargs > 0) { 669 sbuf_printf(sb, "("); 670 for (i = 0;i < uap->nargs; i++) 671 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 672 sbuf_printf(sb, ")"); 673 } 674 /* 675 * Final stage: arguments were proper, string has been 676 * successfully copied from userspace, and copying pointers 677 * from user-space succeed. 678 */ 679 sig = SIGABRT; 680out: 681 if (sig == SIGKILL) { 682 sbuf_trim(sb); 683 sbuf_printf(sb, " (Reason text inaccessible)"); 684 } 685 sbuf_cat(sb, "\n"); 686 sbuf_finish(sb); 687 log(LOG_INFO, "%s", sbuf_data(sb)); 688 sbuf_delete(sb); 689 exit1(td, W_EXITCODE(0, sig)); 690 return (0); 691} 692 693 694#ifdef COMPAT_43 695/* 696 * The dirty work is handled by kern_wait(). 697 */ 698int 699owait(struct thread *td, struct owait_args *uap __unused) 700{ 701 int error, status; 702 703 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 704 if (error == 0) 705 td->td_retval[1] = status; 706 return (error); 707} 708#endif /* COMPAT_43 */ 709 710/* 711 * The dirty work is handled by kern_wait(). 712 */ 713int 714sys_wait4(struct thread *td, struct wait4_args *uap) 715{ 716 struct rusage ru, *rup; 717 int error, status; 718 719 if (uap->rusage != NULL) 720 rup = &ru; 721 else 722 rup = NULL; 723 error = kern_wait(td, uap->pid, &status, uap->options, rup); 724 if (uap->status != NULL && error == 0) 725 error = copyout(&status, uap->status, sizeof(status)); 726 if (uap->rusage != NULL && error == 0) 727 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 728 return (error); 729} 730 731int 732sys_wait6(struct thread *td, struct wait6_args *uap) 733{ 734 struct __wrusage wru, *wrup; 735 siginfo_t si, *sip; 736 idtype_t idtype; 737 id_t id; 738 int error, status; 739 740 idtype = uap->idtype; 741 id = uap->id; 742 743 if (uap->wrusage != NULL) 744 wrup = &wru; 745 else 746 wrup = NULL; 747 748 if (uap->info != NULL) { 749 sip = &si; 750 bzero(sip, sizeof(*sip)); 751 } else 752 sip = NULL; 753 754 /* 755 * We expect all callers of wait6() to know about WEXITED and 756 * WTRAPPED. 757 */ 758 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 759 760 if (uap->status != NULL && error == 0) 761 error = copyout(&status, uap->status, sizeof(status)); 762 if (uap->wrusage != NULL && error == 0) 763 error = copyout(&wru, uap->wrusage, sizeof(wru)); 764 if (uap->info != NULL && error == 0) 765 error = copyout(&si, uap->info, sizeof(si)); 766 return (error); 767} 768 769/* 770 * Reap the remains of a zombie process and optionally return status and 771 * rusage. Asserts and will release both the proctree_lock and the process 772 * lock as part of its work. 773 */ 774void 775proc_reap(struct thread *td, struct proc *p, int *status, int options) 776{ 777 struct proc *q, *t; 778 779 sx_assert(&proctree_lock, SA_XLOCKED); 780 PROC_LOCK_ASSERT(p, MA_OWNED); 781 PROC_SLOCK_ASSERT(p, MA_OWNED); 782 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 783 784 q = td->td_proc; 785 786 PROC_SUNLOCK(p); 787 td->td_retval[0] = p->p_pid; 788 if (status) 789 *status = p->p_xstat; /* convert to int */ 790 if (options & WNOWAIT) { 791 /* 792 * Only poll, returning the status. Caller does not wish to 793 * release the proc struct just yet. 794 */ 795 PROC_UNLOCK(p); 796 sx_xunlock(&proctree_lock); 797 return; 798 } 799 800 PROC_LOCK(q); 801 sigqueue_take(p->p_ksi); 802 PROC_UNLOCK(q); 803 PROC_UNLOCK(p); 804 805 /* 806 * If we got the child via a ptrace 'attach', we need to give it back 807 * to the old parent. 808 */ 809 if (p->p_oppid != 0) { 810 t = proc_realparent(p); 811 PROC_LOCK(t); 812 PROC_LOCK(p); 813 proc_reparent(p, t); 814 p->p_oppid = 0; 815 PROC_UNLOCK(p); 816 pksignal(t, SIGCHLD, p->p_ksi); 817 wakeup(t); 818 cv_broadcast(&p->p_pwait); 819 PROC_UNLOCK(t); 820 sx_xunlock(&proctree_lock); 821 return; 822 } 823 824 /* 825 * Remove other references to this process to ensure we have an 826 * exclusive reference. 827 */ 828 sx_xlock(&allproc_lock); 829 LIST_REMOVE(p, p_list); /* off zombproc */ 830 sx_xunlock(&allproc_lock); 831 LIST_REMOVE(p, p_sibling); 832 PROC_LOCK(p); 833 clear_orphan(p); 834 PROC_UNLOCK(p); 835 leavepgrp(p); 836#ifdef PROCDESC 837 if (p->p_procdesc != NULL) 838 procdesc_reap(p); 839#endif 840 sx_xunlock(&proctree_lock); 841 842 /* 843 * As a side effect of this lock, we know that all other writes to 844 * this proc are visible now, so no more locking is needed for p. 845 */ 846 PROC_LOCK(p); 847 p->p_xstat = 0; /* XXX: why? */ 848 PROC_UNLOCK(p); 849 PROC_LOCK(q); 850 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 851 PROC_UNLOCK(q); 852 853 /* 854 * Decrement the count of procs running with this uid. 855 */ 856 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 857 858 /* 859 * Destroy resource accounting information associated with the process. 860 */ 861#ifdef RACCT 862 PROC_LOCK(p); 863 racct_sub(p, RACCT_NPROC, 1); 864 PROC_UNLOCK(p); 865#endif 866 racct_proc_exit(p); 867 868 /* 869 * Free credentials, arguments, and sigacts. 870 */ 871 crfree(p->p_ucred); 872 p->p_ucred = NULL; 873 pargs_drop(p->p_args); 874 p->p_args = NULL; 875 sigacts_free(p->p_sigacts); 876 p->p_sigacts = NULL; 877 878 /* 879 * Do any thread-system specific cleanups. 880 */ 881 thread_wait(p); 882 883 /* 884 * Give vm and machine-dependent layer a chance to free anything that 885 * cpu_exit couldn't release while still running in process context. 886 */ 887 vm_waitproc(p); 888#ifdef MAC 889 mac_proc_destroy(p); 890#endif 891 KASSERT(FIRST_THREAD_IN_PROC(p), 892 ("proc_reap: no residual thread!")); 893 uma_zfree(proc_zone, p); 894 sx_xlock(&allproc_lock); 895 nprocs--; 896 sx_xunlock(&allproc_lock); 897} 898 899static int 900proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 901 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo) 902{ 903 struct proc *q; 904 struct rusage *rup; 905 906 sx_assert(&proctree_lock, SA_XLOCKED); 907 908 q = td->td_proc; 909 PROC_LOCK(p); 910 911 switch (idtype) { 912 case P_ALL: 913 break; 914 case P_PID: 915 if (p->p_pid != (pid_t)id) { 916 PROC_UNLOCK(p); 917 return (0); 918 } 919 break; 920 case P_PGID: 921 if (p->p_pgid != (pid_t)id) { 922 PROC_UNLOCK(p); 923 return (0); 924 } 925 break; 926 case P_SID: 927 if (p->p_session->s_sid != (pid_t)id) { 928 PROC_UNLOCK(p); 929 return (0); 930 } 931 break; 932 case P_UID: 933 if (p->p_ucred->cr_uid != (uid_t)id) { 934 PROC_UNLOCK(p); 935 return (0); 936 } 937 break; 938 case P_GID: 939 if (p->p_ucred->cr_gid != (gid_t)id) { 940 PROC_UNLOCK(p); 941 return (0); 942 } 943 break; 944 case P_JAILID: 945 if (p->p_ucred->cr_prison->pr_id != (int)id) { 946 PROC_UNLOCK(p); 947 return (0); 948 } 949 break; 950 /* 951 * It seems that the thread structures get zeroed out 952 * at process exit. This makes it impossible to 953 * support P_SETID, P_CID or P_CPUID. 954 */ 955 default: 956 PROC_UNLOCK(p); 957 return (0); 958 } 959 960 if (p_canwait(td, p)) { 961 PROC_UNLOCK(p); 962 return (0); 963 } 964 965 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 966 PROC_UNLOCK(p); 967 return (0); 968 } 969 970 /* 971 * This special case handles a kthread spawned by linux_clone 972 * (see linux_misc.c). The linux_wait4 and linux_waitpid 973 * functions need to be able to distinguish between waiting 974 * on a process and waiting on a thread. It is a thread if 975 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 976 * signifies we want to wait for threads and not processes. 977 */ 978 if ((p->p_sigparent != SIGCHLD) ^ 979 ((options & WLINUXCLONE) != 0)) { 980 PROC_UNLOCK(p); 981 return (0); 982 } 983 984 PROC_SLOCK(p); 985 986 if (siginfo != NULL) { 987 bzero(siginfo, sizeof(*siginfo)); 988 siginfo->si_errno = 0; 989 990 /* 991 * SUSv4 requires that the si_signo value is always 992 * SIGCHLD. Obey it despite the rfork(2) interface 993 * allows to request other signal for child exit 994 * notification. 995 */ 996 siginfo->si_signo = SIGCHLD; 997 998 /* 999 * This is still a rough estimate. We will fix the 1000 * cases TRAPPED, STOPPED, and CONTINUED later. 1001 */ 1002 if (WCOREDUMP(p->p_xstat)) { 1003 siginfo->si_code = CLD_DUMPED; 1004 siginfo->si_status = WTERMSIG(p->p_xstat); 1005 } else if (WIFSIGNALED(p->p_xstat)) { 1006 siginfo->si_code = CLD_KILLED; 1007 siginfo->si_status = WTERMSIG(p->p_xstat); 1008 } else { 1009 siginfo->si_code = CLD_EXITED; 1010 siginfo->si_status = WEXITSTATUS(p->p_xstat); 1011 } 1012 1013 siginfo->si_pid = p->p_pid; 1014 siginfo->si_uid = p->p_ucred->cr_uid; 1015 1016 /* 1017 * The si_addr field would be useful additional 1018 * detail, but apparently the PC value may be lost 1019 * when we reach this point. bzero() above sets 1020 * siginfo->si_addr to NULL. 1021 */ 1022 } 1023 1024 /* 1025 * There should be no reason to limit resources usage info to 1026 * exited processes only. A snapshot about any resources used 1027 * by a stopped process may be exactly what is needed. 1028 */ 1029 if (wrusage != NULL) { 1030 rup = &wrusage->wru_self; 1031 *rup = p->p_ru; 1032 calcru(p, &rup->ru_utime, &rup->ru_stime); 1033 1034 rup = &wrusage->wru_children; 1035 *rup = p->p_stats->p_cru; 1036 calccru(p, &rup->ru_utime, &rup->ru_stime); 1037 } 1038 1039 if (p->p_state == PRS_ZOMBIE) { 1040 proc_reap(td, p, status, options); 1041 return (-1); 1042 } 1043 PROC_SUNLOCK(p); 1044 PROC_UNLOCK(p); 1045 return (1); 1046} 1047 1048int 1049kern_wait(struct thread *td, pid_t pid, int *status, int options, 1050 struct rusage *rusage) 1051{ 1052 struct __wrusage wru, *wrup; 1053 idtype_t idtype; 1054 id_t id; 1055 int ret; 1056 1057 /* 1058 * Translate the special pid values into the (idtype, pid) 1059 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1060 * kern_wait6() on its own. 1061 */ 1062 if (pid == WAIT_ANY) { 1063 idtype = P_ALL; 1064 id = 0; 1065 } else if (pid < 0) { 1066 idtype = P_PGID; 1067 id = (id_t)-pid; 1068 } else { 1069 idtype = P_PID; 1070 id = (id_t)pid; 1071 } 1072 1073 if (rusage != NULL) 1074 wrup = &wru; 1075 else 1076 wrup = NULL; 1077 1078 /* 1079 * For backward compatibility we implicitly add flags WEXITED 1080 * and WTRAPPED here. 1081 */ 1082 options |= WEXITED | WTRAPPED; 1083 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1084 if (rusage != NULL) 1085 *rusage = wru.wru_self; 1086 return (ret); 1087} 1088 1089int 1090kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1091 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1092{ 1093 struct proc *p, *q; 1094 int error, nfound, ret; 1095 1096 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1097 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1098 AUDIT_ARG_VALUE(options); 1099 1100 q = td->td_proc; 1101 1102 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1103 PROC_LOCK(q); 1104 id = (id_t)q->p_pgid; 1105 PROC_UNLOCK(q); 1106 idtype = P_PGID; 1107 } 1108 1109 /* If we don't know the option, just return. */ 1110 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1111 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1112 return (EINVAL); 1113 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1114 /* 1115 * We will be unable to find any matching processes, 1116 * because there are no known events to look for. 1117 * Prefer to return error instead of blocking 1118 * indefinitely. 1119 */ 1120 return (EINVAL); 1121 } 1122 1123loop: 1124 if (q->p_flag & P_STATCHILD) { 1125 PROC_LOCK(q); 1126 q->p_flag &= ~P_STATCHILD; 1127 PROC_UNLOCK(q); 1128 } 1129 nfound = 0; 1130 sx_xlock(&proctree_lock); 1131 LIST_FOREACH(p, &q->p_children, p_sibling) { 1132 ret = proc_to_reap(td, p, idtype, id, status, options, 1133 wrusage, siginfo); 1134 if (ret == 0) 1135 continue; 1136 else if (ret == 1) 1137 nfound++; 1138 else 1139 return (0); 1140 1141 PROC_LOCK(p); 1142 PROC_SLOCK(p); 1143 1144 if ((options & WTRAPPED) != 0 && 1145 (p->p_flag & P_TRACED) != 0 && 1146 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1147 (p->p_suspcount == p->p_numthreads) && 1148 ((p->p_flag & P_WAITED) == 0)) { 1149 PROC_SUNLOCK(p); 1150 if ((options & WNOWAIT) == 0) 1151 p->p_flag |= P_WAITED; 1152 sx_xunlock(&proctree_lock); 1153 td->td_retval[0] = p->p_pid; 1154 1155 if (status != NULL) 1156 *status = W_STOPCODE(p->p_xstat); 1157 if (siginfo != NULL) { 1158 siginfo->si_status = p->p_xstat; 1159 siginfo->si_code = CLD_TRAPPED; 1160 } 1161 if ((options & WNOWAIT) == 0) { 1162 PROC_LOCK(q); 1163 sigqueue_take(p->p_ksi); 1164 PROC_UNLOCK(q); 1165 } 1166 1167 PROC_UNLOCK(p); 1168 return (0); 1169 } 1170 if ((options & WUNTRACED) != 0 && 1171 (p->p_flag & P_STOPPED_SIG) != 0 && 1172 (p->p_suspcount == p->p_numthreads) && 1173 ((p->p_flag & P_WAITED) == 0)) { 1174 PROC_SUNLOCK(p); 1175 if ((options & WNOWAIT) == 0) 1176 p->p_flag |= P_WAITED; 1177 sx_xunlock(&proctree_lock); 1178 td->td_retval[0] = p->p_pid; 1179 1180 if (status != NULL) 1181 *status = W_STOPCODE(p->p_xstat); 1182 if (siginfo != NULL) { 1183 siginfo->si_status = p->p_xstat; 1184 siginfo->si_code = CLD_STOPPED; 1185 } 1186 if ((options & WNOWAIT) == 0) { 1187 PROC_LOCK(q); 1188 sigqueue_take(p->p_ksi); 1189 PROC_UNLOCK(q); 1190 } 1191 1192 PROC_UNLOCK(p); 1193 return (0); 1194 } 1195 PROC_SUNLOCK(p); 1196 if ((options & WCONTINUED) != 0 && 1197 (p->p_flag & P_CONTINUED) != 0) { 1198 sx_xunlock(&proctree_lock); 1199 td->td_retval[0] = p->p_pid; 1200 if ((options & WNOWAIT) == 0) { 1201 p->p_flag &= ~P_CONTINUED; 1202 PROC_LOCK(q); 1203 sigqueue_take(p->p_ksi); 1204 PROC_UNLOCK(q); 1205 } 1206 PROC_UNLOCK(p); 1207 1208 if (status != NULL) 1209 *status = SIGCONT; 1210 if (siginfo != NULL) { 1211 siginfo->si_status = SIGCONT; 1212 siginfo->si_code = CLD_CONTINUED; 1213 } 1214 return (0); 1215 } 1216 PROC_UNLOCK(p); 1217 } 1218 1219 /* 1220 * Look in the orphans list too, to allow the parent to 1221 * collect it's child exit status even if child is being 1222 * debugged. 1223 * 1224 * Debugger detaches from the parent upon successful 1225 * switch-over from parent to child. At this point due to 1226 * re-parenting the parent loses the child to debugger and a 1227 * wait4(2) call would report that it has no children to wait 1228 * for. By maintaining a list of orphans we allow the parent 1229 * to successfully wait until the child becomes a zombie. 1230 */ 1231 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1232 ret = proc_to_reap(td, p, idtype, id, status, options, 1233 wrusage, siginfo); 1234 if (ret == 0) 1235 continue; 1236 else if (ret == 1) 1237 nfound++; 1238 else 1239 return (0); 1240 } 1241 if (nfound == 0) { 1242 sx_xunlock(&proctree_lock); 1243 return (ECHILD); 1244 } 1245 if (options & WNOHANG) { 1246 sx_xunlock(&proctree_lock); 1247 td->td_retval[0] = 0; 1248 return (0); 1249 } 1250 PROC_LOCK(q); 1251 sx_xunlock(&proctree_lock); 1252 if (q->p_flag & P_STATCHILD) { 1253 q->p_flag &= ~P_STATCHILD; 1254 error = 0; 1255 } else 1256 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1257 PROC_UNLOCK(q); 1258 if (error) 1259 return (error); 1260 goto loop; 1261} 1262 1263/* 1264 * Make process 'parent' the new parent of process 'child'. 1265 * Must be called with an exclusive hold of proctree lock. 1266 */ 1267void 1268proc_reparent(struct proc *child, struct proc *parent) 1269{ 1270 1271 sx_assert(&proctree_lock, SX_XLOCKED); 1272 PROC_LOCK_ASSERT(child, MA_OWNED); 1273 if (child->p_pptr == parent) 1274 return; 1275 1276 PROC_LOCK(child->p_pptr); 1277 sigqueue_take(child->p_ksi); 1278 PROC_UNLOCK(child->p_pptr); 1279 LIST_REMOVE(child, p_sibling); 1280 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1281 1282 clear_orphan(child); 1283 if (child->p_flag & P_TRACED) { 1284 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1285 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1286 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1287 p_orphan); 1288 } else { 1289 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1290 child, p_orphan); 1291 } 1292 child->p_treeflag |= P_TREE_ORPHANED; 1293 } 1294 1295 child->p_pptr = parent; 1296} 1297