kern_exit.c revision 270090
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 270090 2014-08-17 07:06:55Z mjg $"); 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 102static void 103clear_orphan(struct proc *p) 104{ 105 106 PROC_LOCK_ASSERT(p, MA_OWNED); 107 108 if (p->p_flag & P_ORPHAN) { 109 LIST_REMOVE(p, p_orphan); 110 p->p_flag &= ~P_ORPHAN; 111 } 112} 113 114/* 115 * exit -- death of process. 116 */ 117void 118sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 119{ 120 121 exit1(td, W_EXITCODE(uap->rval, 0)); 122 /* NOTREACHED */ 123} 124 125/* 126 * Exit: deallocate address space and other resources, change proc state to 127 * zombie, and unlink proc from allproc and parent's lists. Save exit status 128 * and rusage for wait(). Check for child processes and orphan them. 129 */ 130void 131exit1(struct thread *td, int rv) 132{ 133 struct proc *p, *nq, *q; 134 struct vnode *ttyvp = NULL; 135 136 mtx_assert(&Giant, MA_NOTOWNED); 137 138 p = td->td_proc; 139 /* 140 * XXX in case we're rebooting we just let init die in order to 141 * work around an unsolved stack overflow seen very late during 142 * shutdown on sparc64 when the gmirror worker process exists. 143 */ 144 if (p == initproc && rebooting == 0) { 145 printf("init died (signal %d, exit %d)\n", 146 WTERMSIG(rv), WEXITSTATUS(rv)); 147 panic("Going nowhere without my init!"); 148 } 149 150 /* 151 * MUST abort all other threads before proceeding past here. 152 */ 153 PROC_LOCK(p); 154 while (p->p_flag & P_HADTHREADS) { 155 /* 156 * First check if some other thread got here before us. 157 * If so, act appropriately: exit or suspend. 158 */ 159 thread_suspend_check(0); 160 161 /* 162 * Kill off the other threads. This requires 163 * some co-operation from other parts of the kernel 164 * so it may not be instantaneous. With this state set 165 * any thread entering the kernel from userspace will 166 * thread_exit() in trap(). Any thread attempting to 167 * sleep will return immediately with EINTR or EWOULDBLOCK 168 * which will hopefully force them to back out to userland 169 * freeing resources as they go. Any thread attempting 170 * to return to userland will thread_exit() from userret(). 171 * thread_exit() will unsuspend us when the last of the 172 * other threads exits. 173 * If there is already a thread singler after resumption, 174 * calling thread_single will fail; in that case, we just 175 * re-check all suspension request, the thread should 176 * either be suspended there or exit. 177 */ 178 if (!thread_single(SINGLE_EXIT)) 179 break; 180 181 /* 182 * All other activity in this process is now stopped. 183 * Threading support has been turned off. 184 */ 185 } 186 KASSERT(p->p_numthreads == 1, 187 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 188 racct_sub(p, RACCT_NTHR, 1); 189 /* 190 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 191 * on our vmspace, so we should block below until they have 192 * released their reference to us. Note that if they have 193 * requested S_EXIT stops we will block here until they ack 194 * via PIOCCONT. 195 */ 196 _STOPEVENT(p, S_EXIT, rv); 197 198 /* 199 * Ignore any pending request to stop due to a stop signal. 200 * Once P_WEXIT is set, future requests will be ignored as 201 * well. 202 */ 203 p->p_flag &= ~P_STOPPED_SIG; 204 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 205 206 /* 207 * Note that we are exiting and do another wakeup of anyone in 208 * PIOCWAIT in case they aren't listening for S_EXIT stops or 209 * decided to wait again after we told them we are exiting. 210 */ 211 p->p_flag |= P_WEXIT; 212 wakeup(&p->p_stype); 213 214 /* 215 * Wait for any processes that have a hold on our vmspace to 216 * release their reference. 217 */ 218 while (p->p_lock > 0) 219 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 220 221 p->p_xstat = rv; /* Let event handler change exit status */ 222 PROC_UNLOCK(p); 223 /* Drain the limit callout while we don't have the proc locked */ 224 callout_drain(&p->p_limco); 225 226#ifdef AUDIT 227 /* 228 * The Sun BSM exit token contains two components: an exit status as 229 * passed to exit(), and a return value to indicate what sort of exit 230 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 231 * what the return value is. 232 */ 233 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0); 234 AUDIT_SYSCALL_EXIT(0, td); 235#endif 236 237 /* Are we a task leader? */ 238 if (p == p->p_leader) { 239 mtx_lock(&ppeers_lock); 240 q = p->p_peers; 241 while (q != NULL) { 242 PROC_LOCK(q); 243 kern_psignal(q, SIGKILL); 244 PROC_UNLOCK(q); 245 q = q->p_peers; 246 } 247 while (p->p_peers != NULL) 248 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 249 mtx_unlock(&ppeers_lock); 250 } 251 252 /* 253 * Check if any loadable modules need anything done at process exit. 254 * E.g. SYSV IPC stuff 255 * XXX what if one of these generates an error? 256 */ 257 EVENTHANDLER_INVOKE(process_exit, p); 258 259 /* 260 * If parent is waiting for us to exit or exec, 261 * P_PPWAIT is set; we will wakeup the parent below. 262 */ 263 PROC_LOCK(p); 264 rv = p->p_xstat; /* Event handler could change exit status */ 265 stopprofclock(p); 266 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 267 268 /* 269 * Stop the real interval timer. If the handler is currently 270 * executing, prevent it from rearming itself and let it finish. 271 */ 272 if (timevalisset(&p->p_realtimer.it_value) && 273 callout_stop(&p->p_itcallout) == 0) { 274 timevalclear(&p->p_realtimer.it_interval); 275 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 276 KASSERT(!timevalisset(&p->p_realtimer.it_value), 277 ("realtime timer is still armed")); 278 } 279 PROC_UNLOCK(p); 280 281 /* 282 * Reset any sigio structures pointing to us as a result of 283 * F_SETOWN with our pid. 284 */ 285 funsetownlst(&p->p_sigiolst); 286 287 /* 288 * If this process has an nlminfo data area (for lockd), release it 289 */ 290 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 291 (*nlminfo_release_p)(p); 292 293 /* 294 * Close open files and release open-file table. 295 * This may block! 296 */ 297 fdescfree(td); 298 299 /* 300 * If this thread tickled GEOM, we need to wait for the giggling to 301 * stop before we return to userland 302 */ 303 if (td->td_pflags & TDP_GEOM) 304 g_waitidle(); 305 306 /* 307 * Remove ourself from our leader's peer list and wake our leader. 308 */ 309 mtx_lock(&ppeers_lock); 310 if (p->p_leader->p_peers) { 311 q = p->p_leader; 312 while (q->p_peers != p) 313 q = q->p_peers; 314 q->p_peers = p->p_peers; 315 wakeup(p->p_leader); 316 } 317 mtx_unlock(&ppeers_lock); 318 319 vmspace_exit(td); 320 321 sx_xlock(&proctree_lock); 322 if (SESS_LEADER(p)) { 323 struct session *sp = p->p_session; 324 struct tty *tp; 325 326 /* 327 * s_ttyp is not zero'd; we use this to indicate that 328 * the session once had a controlling terminal. (for 329 * logging and informational purposes) 330 */ 331 SESS_LOCK(sp); 332 ttyvp = sp->s_ttyvp; 333 tp = sp->s_ttyp; 334 sp->s_ttyvp = NULL; 335 sp->s_ttydp = NULL; 336 sp->s_leader = NULL; 337 SESS_UNLOCK(sp); 338 339 /* 340 * Signal foreground pgrp and revoke access to 341 * controlling terminal if it has not been revoked 342 * already. 343 * 344 * Because the TTY may have been revoked in the mean 345 * time and could already have a new session associated 346 * with it, make sure we don't send a SIGHUP to a 347 * foreground process group that does not belong to this 348 * session. 349 */ 350 351 if (tp != NULL) { 352 tty_lock(tp); 353 if (tp->t_session == sp) 354 tty_signal_pgrp(tp, SIGHUP); 355 tty_unlock(tp); 356 } 357 358 if (ttyvp != NULL) { 359 sx_xunlock(&proctree_lock); 360 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 361 VOP_REVOKE(ttyvp, REVOKEALL); 362 VOP_UNLOCK(ttyvp, 0); 363 } 364 sx_xlock(&proctree_lock); 365 } 366 } 367 fixjobc(p, p->p_pgrp, 0); 368 sx_xunlock(&proctree_lock); 369 (void)acct_process(td); 370 371 /* Release the TTY now we've unlocked everything. */ 372 if (ttyvp != NULL) 373 vrele(ttyvp); 374#ifdef KTRACE 375 ktrprocexit(td); 376#endif 377 /* 378 * Release reference to text vnode 379 */ 380 if (p->p_textvp != NULL) { 381 vrele(p->p_textvp); 382 p->p_textvp = NULL; 383 } 384 385 /* 386 * Release our limits structure. 387 */ 388 lim_free(p->p_limit); 389 p->p_limit = NULL; 390 391 tidhash_remove(td); 392 393 /* 394 * Remove proc from allproc queue and pidhash chain. 395 * Place onto zombproc. Unlink from parent's child list. 396 */ 397 sx_xlock(&allproc_lock); 398 LIST_REMOVE(p, p_list); 399 LIST_INSERT_HEAD(&zombproc, p, p_list); 400 LIST_REMOVE(p, p_hash); 401 sx_xunlock(&allproc_lock); 402 403 /* 404 * Call machine-dependent code to release any 405 * machine-dependent resources other than the address space. 406 * The address space is released by "vmspace_exitfree(p)" in 407 * vm_waitproc(). 408 */ 409 cpu_exit(td); 410 411 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 412 413 /* 414 * Reparent all of our children to init. 415 */ 416 sx_xlock(&proctree_lock); 417 q = LIST_FIRST(&p->p_children); 418 if (q != NULL) /* only need this if any child is S_ZOMB */ 419 wakeup(initproc); 420 for (; q != NULL; q = nq) { 421 nq = LIST_NEXT(q, p_sibling); 422 PROC_LOCK(q); 423 proc_reparent(q, initproc); 424 q->p_sigparent = SIGCHLD; 425 /* 426 * Traced processes are killed 427 * since their existence means someone is screwing up. 428 */ 429 if (q->p_flag & P_TRACED) { 430 struct thread *temp; 431 432 /* 433 * Since q was found on our children list, the 434 * proc_reparent() call moved q to the orphan 435 * list due to present P_TRACED flag. Clear 436 * orphan link for q now while q is locked. 437 */ 438 clear_orphan(q); 439 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 440 FOREACH_THREAD_IN_PROC(q, temp) 441 temp->td_dbgflags &= ~TDB_SUSPEND; 442 kern_psignal(q, SIGKILL); 443 } 444 PROC_UNLOCK(q); 445 } 446 447 /* 448 * Also get rid of our orphans. 449 */ 450 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 451 PROC_LOCK(q); 452 clear_orphan(q); 453 PROC_UNLOCK(q); 454 } 455 456 /* Save exit status. */ 457 PROC_LOCK(p); 458 p->p_xthread = td; 459 460 /* Tell the prison that we are gone. */ 461 prison_proc_free(p->p_ucred->cr_prison); 462 463#ifdef KDTRACE_HOOKS 464 /* 465 * Tell the DTrace fasttrap provider about the exit if it 466 * has declared an interest. 467 */ 468 if (dtrace_fasttrap_exit) 469 dtrace_fasttrap_exit(p); 470#endif 471 472 /* 473 * Notify interested parties of our demise. 474 */ 475 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 476 477#ifdef KDTRACE_HOOKS 478 int reason = CLD_EXITED; 479 if (WCOREDUMP(rv)) 480 reason = CLD_DUMPED; 481 else if (WIFSIGNALED(rv)) 482 reason = CLD_KILLED; 483 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 484#endif 485 486 /* 487 * Just delete all entries in the p_klist. At this point we won't 488 * report any more events, and there are nasty race conditions that 489 * can beat us if we don't. 490 */ 491 knlist_clear(&p->p_klist, 1); 492 493 /* 494 * If this is a process with a descriptor, we may not need to deliver 495 * a signal to the parent. proctree_lock is held over 496 * procdesc_exit() to serialize concurrent calls to close() and 497 * exit(). 498 */ 499#ifdef PROCDESC 500 if (p->p_procdesc == NULL || procdesc_exit(p)) { 501#endif 502 /* 503 * Notify parent that we're gone. If parent has the 504 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 505 * notify process 1 instead (and hope it will handle this 506 * situation). 507 */ 508 PROC_LOCK(p->p_pptr); 509 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 510 if (p->p_pptr->p_sigacts->ps_flag & 511 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 512 struct proc *pp; 513 514 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 515 pp = p->p_pptr; 516 PROC_UNLOCK(pp); 517 proc_reparent(p, initproc); 518 p->p_sigparent = SIGCHLD; 519 PROC_LOCK(p->p_pptr); 520 521 /* 522 * Notify parent, so in case he was wait(2)ing or 523 * executing waitpid(2) with our pid, he will 524 * continue. 525 */ 526 wakeup(pp); 527 } else 528 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 529 530 if (p->p_pptr == initproc) 531 kern_psignal(p->p_pptr, SIGCHLD); 532 else if (p->p_sigparent != 0) { 533 if (p->p_sigparent == SIGCHLD) 534 childproc_exited(p); 535 else /* LINUX thread */ 536 kern_psignal(p->p_pptr, p->p_sigparent); 537 } 538#ifdef PROCDESC 539 } else 540 PROC_LOCK(p->p_pptr); 541#endif 542 sx_xunlock(&proctree_lock); 543 544 /* 545 * The state PRS_ZOMBIE prevents other proesses from sending 546 * signal to the process, to avoid memory leak, we free memory 547 * for signal queue at the time when the state is set. 548 */ 549 sigqueue_flush(&p->p_sigqueue); 550 sigqueue_flush(&td->td_sigqueue); 551 552 /* 553 * We have to wait until after acquiring all locks before 554 * changing p_state. We need to avoid all possible context 555 * switches (including ones from blocking on a mutex) while 556 * marked as a zombie. We also have to set the zombie state 557 * before we release the parent process' proc lock to avoid 558 * a lost wakeup. So, we first call wakeup, then we grab the 559 * sched lock, update the state, and release the parent process' 560 * proc lock. 561 */ 562 wakeup(p->p_pptr); 563 cv_broadcast(&p->p_pwait); 564 sched_exit(p->p_pptr, td); 565 PROC_SLOCK(p); 566 p->p_state = PRS_ZOMBIE; 567 PROC_UNLOCK(p->p_pptr); 568 569 /* 570 * Hopefully no one will try to deliver a signal to the process this 571 * late in the game. 572 */ 573 knlist_destroy(&p->p_klist); 574 575 /* 576 * Save our children's rusage information in our exit rusage. 577 */ 578 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 579 580 /* 581 * Make sure the scheduler takes this thread out of its tables etc. 582 * This will also release this thread's reference to the ucred. 583 * Other thread parts to release include pcb bits and such. 584 */ 585 thread_exit(); 586} 587 588 589#ifndef _SYS_SYSPROTO_H_ 590struct abort2_args { 591 char *why; 592 int nargs; 593 void **args; 594}; 595#endif 596 597int 598sys_abort2(struct thread *td, struct abort2_args *uap) 599{ 600 struct proc *p = td->td_proc; 601 struct sbuf *sb; 602 void *uargs[16]; 603 int error, i, sig; 604 605 /* 606 * Do it right now so we can log either proper call of abort2(), or 607 * note, that invalid argument was passed. 512 is big enough to 608 * handle 16 arguments' descriptions with additional comments. 609 */ 610 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 611 sbuf_clear(sb); 612 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 613 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 614 /* 615 * Since we can't return from abort2(), send SIGKILL in cases, where 616 * abort2() was called improperly 617 */ 618 sig = SIGKILL; 619 /* Prevent from DoSes from user-space. */ 620 if (uap->nargs < 0 || uap->nargs > 16) 621 goto out; 622 if (uap->nargs > 0) { 623 if (uap->args == NULL) 624 goto out; 625 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 626 if (error != 0) 627 goto out; 628 } 629 /* 630 * Limit size of 'reason' string to 128. Will fit even when 631 * maximal number of arguments was chosen to be logged. 632 */ 633 if (uap->why != NULL) { 634 error = sbuf_copyin(sb, uap->why, 128); 635 if (error < 0) 636 goto out; 637 } else { 638 sbuf_printf(sb, "(null)"); 639 } 640 if (uap->nargs > 0) { 641 sbuf_printf(sb, "("); 642 for (i = 0;i < uap->nargs; i++) 643 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 644 sbuf_printf(sb, ")"); 645 } 646 /* 647 * Final stage: arguments were proper, string has been 648 * successfully copied from userspace, and copying pointers 649 * from user-space succeed. 650 */ 651 sig = SIGABRT; 652out: 653 if (sig == SIGKILL) { 654 sbuf_trim(sb); 655 sbuf_printf(sb, " (Reason text inaccessible)"); 656 } 657 sbuf_cat(sb, "\n"); 658 sbuf_finish(sb); 659 log(LOG_INFO, "%s", sbuf_data(sb)); 660 sbuf_delete(sb); 661 exit1(td, W_EXITCODE(0, sig)); 662 return (0); 663} 664 665 666#ifdef COMPAT_43 667/* 668 * The dirty work is handled by kern_wait(). 669 */ 670int 671owait(struct thread *td, struct owait_args *uap __unused) 672{ 673 int error, status; 674 675 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 676 if (error == 0) 677 td->td_retval[1] = status; 678 return (error); 679} 680#endif /* COMPAT_43 */ 681 682/* 683 * The dirty work is handled by kern_wait(). 684 */ 685int 686sys_wait4(struct thread *td, struct wait4_args *uap) 687{ 688 struct rusage ru, *rup; 689 int error, status; 690 691 if (uap->rusage != NULL) 692 rup = &ru; 693 else 694 rup = NULL; 695 error = kern_wait(td, uap->pid, &status, uap->options, rup); 696 if (uap->status != NULL && error == 0) 697 error = copyout(&status, uap->status, sizeof(status)); 698 if (uap->rusage != NULL && error == 0) 699 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 700 return (error); 701} 702 703int 704sys_wait6(struct thread *td, struct wait6_args *uap) 705{ 706 struct __wrusage wru, *wrup; 707 siginfo_t si, *sip; 708 idtype_t idtype; 709 id_t id; 710 int error, status; 711 712 idtype = uap->idtype; 713 id = uap->id; 714 715 if (uap->wrusage != NULL) 716 wrup = &wru; 717 else 718 wrup = NULL; 719 720 if (uap->info != NULL) { 721 sip = &si; 722 bzero(sip, sizeof(*sip)); 723 } else 724 sip = NULL; 725 726 /* 727 * We expect all callers of wait6() to know about WEXITED and 728 * WTRAPPED. 729 */ 730 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 731 732 if (uap->status != NULL && error == 0) 733 error = copyout(&status, uap->status, sizeof(status)); 734 if (uap->wrusage != NULL && error == 0) 735 error = copyout(&wru, uap->wrusage, sizeof(wru)); 736 if (uap->info != NULL && error == 0) 737 error = copyout(&si, uap->info, sizeof(si)); 738 return (error); 739} 740 741/* 742 * Reap the remains of a zombie process and optionally return status and 743 * rusage. Asserts and will release both the proctree_lock and the process 744 * lock as part of its work. 745 */ 746void 747proc_reap(struct thread *td, struct proc *p, int *status, int options) 748{ 749 struct proc *q, *t; 750 751 sx_assert(&proctree_lock, SA_XLOCKED); 752 PROC_LOCK_ASSERT(p, MA_OWNED); 753 PROC_SLOCK_ASSERT(p, MA_OWNED); 754 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 755 756 q = td->td_proc; 757 758 PROC_SUNLOCK(p); 759 td->td_retval[0] = p->p_pid; 760 if (status) 761 *status = p->p_xstat; /* convert to int */ 762 if (options & WNOWAIT) { 763 /* 764 * Only poll, returning the status. Caller does not wish to 765 * release the proc struct just yet. 766 */ 767 PROC_UNLOCK(p); 768 sx_xunlock(&proctree_lock); 769 return; 770 } 771 772 PROC_LOCK(q); 773 sigqueue_take(p->p_ksi); 774 PROC_UNLOCK(q); 775 PROC_UNLOCK(p); 776 777 /* 778 * If we got the child via a ptrace 'attach', we need to give it back 779 * to the old parent. 780 */ 781 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 782 PROC_LOCK(p); 783 proc_reparent(p, t); 784 p->p_oppid = 0; 785 PROC_UNLOCK(p); 786 pksignal(t, SIGCHLD, p->p_ksi); 787 wakeup(t); 788 cv_broadcast(&p->p_pwait); 789 PROC_UNLOCK(t); 790 sx_xunlock(&proctree_lock); 791 return; 792 } 793 794 /* 795 * Remove other references to this process to ensure we have an 796 * exclusive reference. 797 */ 798 sx_xlock(&allproc_lock); 799 LIST_REMOVE(p, p_list); /* off zombproc */ 800 sx_xunlock(&allproc_lock); 801 LIST_REMOVE(p, p_sibling); 802 PROC_LOCK(p); 803 clear_orphan(p); 804 PROC_UNLOCK(p); 805 leavepgrp(p); 806#ifdef PROCDESC 807 if (p->p_procdesc != NULL) 808 procdesc_reap(p); 809#endif 810 sx_xunlock(&proctree_lock); 811 812 /* 813 * As a side effect of this lock, we know that all other writes to 814 * this proc are visible now, so no more locking is needed for p. 815 */ 816 PROC_LOCK(p); 817 p->p_xstat = 0; /* XXX: why? */ 818 PROC_UNLOCK(p); 819 PROC_LOCK(q); 820 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 821 PROC_UNLOCK(q); 822 823 /* 824 * Decrement the count of procs running with this uid. 825 */ 826 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 827 828 /* 829 * Destroy resource accounting information associated with the process. 830 */ 831#ifdef RACCT 832 PROC_LOCK(p); 833 racct_sub(p, RACCT_NPROC, 1); 834 PROC_UNLOCK(p); 835#endif 836 racct_proc_exit(p); 837 838 /* 839 * Free credentials, arguments, and sigacts. 840 */ 841 crfree(p->p_ucred); 842 p->p_ucred = NULL; 843 pargs_drop(p->p_args); 844 p->p_args = NULL; 845 sigacts_free(p->p_sigacts); 846 p->p_sigacts = NULL; 847 848 /* 849 * Do any thread-system specific cleanups. 850 */ 851 thread_wait(p); 852 853 /* 854 * Give vm and machine-dependent layer a chance to free anything that 855 * cpu_exit couldn't release while still running in process context. 856 */ 857 vm_waitproc(p); 858#ifdef MAC 859 mac_proc_destroy(p); 860#endif 861 KASSERT(FIRST_THREAD_IN_PROC(p), 862 ("proc_reap: no residual thread!")); 863 uma_zfree(proc_zone, p); 864 sx_xlock(&allproc_lock); 865 nprocs--; 866 sx_xunlock(&allproc_lock); 867} 868 869static int 870proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 871 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo) 872{ 873 struct proc *q; 874 struct rusage *rup; 875 876 sx_assert(&proctree_lock, SA_XLOCKED); 877 878 q = td->td_proc; 879 PROC_LOCK(p); 880 881 switch (idtype) { 882 case P_ALL: 883 break; 884 case P_PID: 885 if (p->p_pid != (pid_t)id) { 886 PROC_UNLOCK(p); 887 return (0); 888 } 889 break; 890 case P_PGID: 891 if (p->p_pgid != (pid_t)id) { 892 PROC_UNLOCK(p); 893 return (0); 894 } 895 break; 896 case P_SID: 897 if (p->p_session->s_sid != (pid_t)id) { 898 PROC_UNLOCK(p); 899 return (0); 900 } 901 break; 902 case P_UID: 903 if (p->p_ucred->cr_uid != (uid_t)id) { 904 PROC_UNLOCK(p); 905 return (0); 906 } 907 break; 908 case P_GID: 909 if (p->p_ucred->cr_gid != (gid_t)id) { 910 PROC_UNLOCK(p); 911 return (0); 912 } 913 break; 914 case P_JAILID: 915 if (p->p_ucred->cr_prison->pr_id != (int)id) { 916 PROC_UNLOCK(p); 917 return (0); 918 } 919 break; 920 /* 921 * It seems that the thread structures get zeroed out 922 * at process exit. This makes it impossible to 923 * support P_SETID, P_CID or P_CPUID. 924 */ 925 default: 926 PROC_UNLOCK(p); 927 return (0); 928 } 929 930 if (p_canwait(td, p)) { 931 PROC_UNLOCK(p); 932 return (0); 933 } 934 935 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 936 PROC_UNLOCK(p); 937 return (0); 938 } 939 940 /* 941 * This special case handles a kthread spawned by linux_clone 942 * (see linux_misc.c). The linux_wait4 and linux_waitpid 943 * functions need to be able to distinguish between waiting 944 * on a process and waiting on a thread. It is a thread if 945 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 946 * signifies we want to wait for threads and not processes. 947 */ 948 if ((p->p_sigparent != SIGCHLD) ^ 949 ((options & WLINUXCLONE) != 0)) { 950 PROC_UNLOCK(p); 951 return (0); 952 } 953 954 PROC_SLOCK(p); 955 956 if (siginfo != NULL) { 957 bzero(siginfo, sizeof(*siginfo)); 958 siginfo->si_errno = 0; 959 960 /* 961 * SUSv4 requires that the si_signo value is always 962 * SIGCHLD. Obey it despite the rfork(2) interface 963 * allows to request other signal for child exit 964 * notification. 965 */ 966 siginfo->si_signo = SIGCHLD; 967 968 /* 969 * This is still a rough estimate. We will fix the 970 * cases TRAPPED, STOPPED, and CONTINUED later. 971 */ 972 if (WCOREDUMP(p->p_xstat)) { 973 siginfo->si_code = CLD_DUMPED; 974 siginfo->si_status = WTERMSIG(p->p_xstat); 975 } else if (WIFSIGNALED(p->p_xstat)) { 976 siginfo->si_code = CLD_KILLED; 977 siginfo->si_status = WTERMSIG(p->p_xstat); 978 } else { 979 siginfo->si_code = CLD_EXITED; 980 siginfo->si_status = WEXITSTATUS(p->p_xstat); 981 } 982 983 siginfo->si_pid = p->p_pid; 984 siginfo->si_uid = p->p_ucred->cr_uid; 985 986 /* 987 * The si_addr field would be useful additional 988 * detail, but apparently the PC value may be lost 989 * when we reach this point. bzero() above sets 990 * siginfo->si_addr to NULL. 991 */ 992 } 993 994 /* 995 * There should be no reason to limit resources usage info to 996 * exited processes only. A snapshot about any resources used 997 * by a stopped process may be exactly what is needed. 998 */ 999 if (wrusage != NULL) { 1000 rup = &wrusage->wru_self; 1001 *rup = p->p_ru; 1002 calcru(p, &rup->ru_utime, &rup->ru_stime); 1003 1004 rup = &wrusage->wru_children; 1005 *rup = p->p_stats->p_cru; 1006 calccru(p, &rup->ru_utime, &rup->ru_stime); 1007 } 1008 1009 if (p->p_state == PRS_ZOMBIE) { 1010 proc_reap(td, p, status, options); 1011 return (-1); 1012 } 1013 PROC_SUNLOCK(p); 1014 PROC_UNLOCK(p); 1015 return (1); 1016} 1017 1018int 1019kern_wait(struct thread *td, pid_t pid, int *status, int options, 1020 struct rusage *rusage) 1021{ 1022 struct __wrusage wru, *wrup; 1023 idtype_t idtype; 1024 id_t id; 1025 int ret; 1026 1027 /* 1028 * Translate the special pid values into the (idtype, pid) 1029 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1030 * kern_wait6() on its own. 1031 */ 1032 if (pid == WAIT_ANY) { 1033 idtype = P_ALL; 1034 id = 0; 1035 } else if (pid < 0) { 1036 idtype = P_PGID; 1037 id = (id_t)-pid; 1038 } else { 1039 idtype = P_PID; 1040 id = (id_t)pid; 1041 } 1042 1043 if (rusage != NULL) 1044 wrup = &wru; 1045 else 1046 wrup = NULL; 1047 1048 /* 1049 * For backward compatibility we implicitly add flags WEXITED 1050 * and WTRAPPED here. 1051 */ 1052 options |= WEXITED | WTRAPPED; 1053 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1054 if (rusage != NULL) 1055 *rusage = wru.wru_self; 1056 return (ret); 1057} 1058 1059int 1060kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1061 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1062{ 1063 struct proc *p, *q; 1064 int error, nfound, ret; 1065 1066 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1067 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1068 AUDIT_ARG_VALUE(options); 1069 1070 q = td->td_proc; 1071 1072 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1073 PROC_LOCK(q); 1074 id = (id_t)q->p_pgid; 1075 PROC_UNLOCK(q); 1076 idtype = P_PGID; 1077 } 1078 1079 /* If we don't know the option, just return. */ 1080 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1081 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1082 return (EINVAL); 1083 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1084 /* 1085 * We will be unable to find any matching processes, 1086 * because there are no known events to look for. 1087 * Prefer to return error instead of blocking 1088 * indefinitely. 1089 */ 1090 return (EINVAL); 1091 } 1092 1093loop: 1094 if (q->p_flag & P_STATCHILD) { 1095 PROC_LOCK(q); 1096 q->p_flag &= ~P_STATCHILD; 1097 PROC_UNLOCK(q); 1098 } 1099 nfound = 0; 1100 sx_xlock(&proctree_lock); 1101 LIST_FOREACH(p, &q->p_children, p_sibling) { 1102 ret = proc_to_reap(td, p, idtype, id, status, options, 1103 wrusage, siginfo); 1104 if (ret == 0) 1105 continue; 1106 else if (ret == 1) 1107 nfound++; 1108 else 1109 return (0); 1110 1111 PROC_LOCK(p); 1112 PROC_SLOCK(p); 1113 1114 if ((options & WTRAPPED) != 0 && 1115 (p->p_flag & P_TRACED) != 0 && 1116 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1117 (p->p_suspcount == p->p_numthreads) && 1118 ((p->p_flag & P_WAITED) == 0)) { 1119 PROC_SUNLOCK(p); 1120 if ((options & WNOWAIT) == 0) 1121 p->p_flag |= P_WAITED; 1122 sx_xunlock(&proctree_lock); 1123 td->td_retval[0] = p->p_pid; 1124 1125 if (status != NULL) 1126 *status = W_STOPCODE(p->p_xstat); 1127 if (siginfo != NULL) { 1128 siginfo->si_status = p->p_xstat; 1129 siginfo->si_code = CLD_TRAPPED; 1130 } 1131 if ((options & WNOWAIT) == 0) { 1132 PROC_LOCK(q); 1133 sigqueue_take(p->p_ksi); 1134 PROC_UNLOCK(q); 1135 } 1136 1137 PROC_UNLOCK(p); 1138 return (0); 1139 } 1140 if ((options & WUNTRACED) != 0 && 1141 (p->p_flag & P_STOPPED_SIG) != 0 && 1142 (p->p_suspcount == p->p_numthreads) && 1143 ((p->p_flag & P_WAITED) == 0)) { 1144 PROC_SUNLOCK(p); 1145 if ((options & WNOWAIT) == 0) 1146 p->p_flag |= P_WAITED; 1147 sx_xunlock(&proctree_lock); 1148 td->td_retval[0] = p->p_pid; 1149 1150 if (status != NULL) 1151 *status = W_STOPCODE(p->p_xstat); 1152 if (siginfo != NULL) { 1153 siginfo->si_status = p->p_xstat; 1154 siginfo->si_code = CLD_STOPPED; 1155 } 1156 if ((options & WNOWAIT) == 0) { 1157 PROC_LOCK(q); 1158 sigqueue_take(p->p_ksi); 1159 PROC_UNLOCK(q); 1160 } 1161 1162 PROC_UNLOCK(p); 1163 return (0); 1164 } 1165 PROC_SUNLOCK(p); 1166 if ((options & WCONTINUED) != 0 && 1167 (p->p_flag & P_CONTINUED) != 0) { 1168 sx_xunlock(&proctree_lock); 1169 td->td_retval[0] = p->p_pid; 1170 if ((options & WNOWAIT) == 0) { 1171 p->p_flag &= ~P_CONTINUED; 1172 PROC_LOCK(q); 1173 sigqueue_take(p->p_ksi); 1174 PROC_UNLOCK(q); 1175 } 1176 PROC_UNLOCK(p); 1177 1178 if (status != NULL) 1179 *status = SIGCONT; 1180 if (siginfo != NULL) { 1181 siginfo->si_status = SIGCONT; 1182 siginfo->si_code = CLD_CONTINUED; 1183 } 1184 return (0); 1185 } 1186 PROC_UNLOCK(p); 1187 } 1188 1189 /* 1190 * Look in the orphans list too, to allow the parent to 1191 * collect it's child exit status even if child is being 1192 * debugged. 1193 * 1194 * Debugger detaches from the parent upon successful 1195 * switch-over from parent to child. At this point due to 1196 * re-parenting the parent loses the child to debugger and a 1197 * wait4(2) call would report that it has no children to wait 1198 * for. By maintaining a list of orphans we allow the parent 1199 * to successfully wait until the child becomes a zombie. 1200 */ 1201 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1202 ret = proc_to_reap(td, p, idtype, id, status, options, 1203 wrusage, siginfo); 1204 if (ret == 0) 1205 continue; 1206 else if (ret == 1) 1207 nfound++; 1208 else 1209 return (0); 1210 } 1211 if (nfound == 0) { 1212 sx_xunlock(&proctree_lock); 1213 return (ECHILD); 1214 } 1215 if (options & WNOHANG) { 1216 sx_xunlock(&proctree_lock); 1217 td->td_retval[0] = 0; 1218 return (0); 1219 } 1220 PROC_LOCK(q); 1221 sx_xunlock(&proctree_lock); 1222 if (q->p_flag & P_STATCHILD) { 1223 q->p_flag &= ~P_STATCHILD; 1224 error = 0; 1225 } else 1226 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1227 PROC_UNLOCK(q); 1228 if (error) 1229 return (error); 1230 goto loop; 1231} 1232 1233/* 1234 * Make process 'parent' the new parent of process 'child'. 1235 * Must be called with an exclusive hold of proctree lock. 1236 */ 1237void 1238proc_reparent(struct proc *child, struct proc *parent) 1239{ 1240 1241 sx_assert(&proctree_lock, SX_XLOCKED); 1242 PROC_LOCK_ASSERT(child, MA_OWNED); 1243 if (child->p_pptr == parent) 1244 return; 1245 1246 PROC_LOCK(child->p_pptr); 1247 sigqueue_take(child->p_ksi); 1248 PROC_UNLOCK(child->p_pptr); 1249 LIST_REMOVE(child, p_sibling); 1250 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1251 1252 clear_orphan(child); 1253 if (child->p_flag & P_TRACED) { 1254 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, p_orphan); 1255 child->p_flag |= P_ORPHAN; 1256 } 1257 1258 child->p_pptr = parent; 1259} 1260