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("$FreeBSD$"); 39 40 #include "opt_compat.h" 41 #include "opt_ktrace.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/sysproto.h> 46 #include <sys/capability.h> 47 #include <sys/eventhandler.h> 48 #include <sys/kernel.h> 49 #include <sys/malloc.h> 50 #include <sys/lock.h> 51 #include <sys/mutex.h> 52 #include <sys/proc.h> 53 #include <sys/procdesc.h> 54 #include <sys/pioctl.h> 55 #include <sys/jail.h> 56 #include <sys/tty.h> 57 #include <sys/wait.h> 58 #include <sys/vmmeter.h> 59 #include <sys/vnode.h> 60 #include <sys/racct.h> 61 #include <sys/resourcevar.h> 62 #include <sys/sbuf.h> 63 #include <sys/signalvar.h> 64 #include <sys/sched.h> 65 #include <sys/sx.h> 66 #include <sys/syscallsubr.h> 67 #include <sys/syslog.h> 68 #include <sys/ptrace.h> 69 #include <sys/acct.h> /* for acct_process() function prototype */ 70 #include <sys/filedesc.h> 71 #include <sys/sdt.h> 72 #include <sys/shm.h> 73 #include <sys/sem.h> 74 #ifdef KTRACE 75 #include <sys/ktrace.h> 76 #endif 77 78 #include <security/audit/audit.h> 79 #include <security/mac/mac_framework.h> 80 81 #include <vm/vm.h> 82 #include <vm/vm_extern.h> 83 #include <vm/vm_param.h> 84 #include <vm/pmap.h> 85 #include <vm/vm_map.h> 86 #include <vm/vm_page.h> 87 #include <vm/uma.h> 88 89 #ifdef KDTRACE_HOOKS 90 #include <sys/dtrace_bsd.h> 91 dtrace_execexit_func_t dtrace_fasttrap_exit; 92 #endif 93 94 SDT_PROVIDER_DECLARE(proc); 95 SDT_PROBE_DEFINE1(proc, kernel, , exit, "int"); 96 97 /* Hook for NFS teardown procedure. */ 98 void (*nlminfo_release_p)(struct proc *p); 99 100 static void 101 clear_orphan(struct proc *p) 102 { 103 104 PROC_LOCK_ASSERT(p, MA_OWNED); 105 106 if (p->p_flag & P_ORPHAN) { 107 LIST_REMOVE(p, p_orphan); 108 p->p_flag &= ~P_ORPHAN; 109 } 110 } 111 112 /* 113 * exit -- death of process. 114 */ 115 void 116 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 117 { 118 119 exit1(td, W_EXITCODE(uap->rval, 0)); 120 /* NOTREACHED */ 121 } 122 123 /* 124 * Exit: deallocate address space and other resources, change proc state to 125 * zombie, and unlink proc from allproc and parent's lists. Save exit status 126 * and rusage for wait(). Check for child processes and orphan them. 127 */ 128 void 129 exit1(struct thread *td, int rv) 130 { 131 struct proc *p, *nq, *q; 132 struct vnode *vtmp; 133 struct vnode *ttyvp = NULL; 134 struct plimit *plim; 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 ((vtmp = p->p_textvp) != NULL) { 381 p->p_textvp = NULL; 382 vrele(vtmp); 383 } 384 385 /* 386 * Release our limits structure. 387 */ 388 PROC_LOCK(p); 389 plim = p->p_limit; 390 p->p_limit = NULL; 391 PROC_UNLOCK(p); 392 lim_free(plim); 393 394 tidhash_remove(td); 395 396 /* 397 * Remove proc from allproc queue and pidhash chain. 398 * Place onto zombproc. Unlink from parent's child list. 399 */ 400 sx_xlock(&allproc_lock); 401 LIST_REMOVE(p, p_list); 402 LIST_INSERT_HEAD(&zombproc, p, p_list); 403 LIST_REMOVE(p, p_hash); 404 sx_xunlock(&allproc_lock); 405 406 /* 407 * Call machine-dependent code to release any 408 * machine-dependent resources other than the address space. 409 * The address space is released by "vmspace_exitfree(p)" in 410 * vm_waitproc(). 411 */ 412 cpu_exit(td); 413 414 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 415 416 /* 417 * Reparent all of our children to init. 418 */ 419 sx_xlock(&proctree_lock); 420 q = LIST_FIRST(&p->p_children); 421 if (q != NULL) /* only need this if any child is S_ZOMB */ 422 wakeup(initproc); 423 for (; q != NULL; q = nq) { 424 nq = LIST_NEXT(q, p_sibling); 425 PROC_LOCK(q); 426 proc_reparent(q, initproc); 427 q->p_sigparent = SIGCHLD; 428 /* 429 * Traced processes are killed 430 * since their existence means someone is screwing up. 431 */ 432 if (q->p_flag & P_TRACED) { 433 struct thread *temp; 434 435 /* 436 * Since q was found on our children list, the 437 * proc_reparent() call moved q to the orphan 438 * list due to present P_TRACED flag. Clear 439 * orphan link for q now while q is locked. 440 */ 441 clear_orphan(q); 442 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 443 FOREACH_THREAD_IN_PROC(q, temp) 444 temp->td_dbgflags &= ~TDB_SUSPEND; 445 kern_psignal(q, SIGKILL); 446 } 447 PROC_UNLOCK(q); 448 } 449 450 /* 451 * Also get rid of our orphans. 452 */ 453 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 454 PROC_LOCK(q); 455 clear_orphan(q); 456 PROC_UNLOCK(q); 457 } 458 459 /* Save exit status. */ 460 PROC_LOCK(p); 461 p->p_xthread = td; 462 463 /* Tell the prison that we are gone. */ 464 prison_proc_free(p->p_ucred->cr_prison); 465 466 #ifdef KDTRACE_HOOKS 467 /* 468 * Tell the DTrace fasttrap provider about the exit if it 469 * has declared an interest. 470 */ 471 if (dtrace_fasttrap_exit) 472 dtrace_fasttrap_exit(p); 473 #endif 474 475 /* 476 * Notify interested parties of our demise. 477 */ 478 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 479 480 #ifdef KDTRACE_HOOKS 481 int reason = CLD_EXITED; 482 if (WCOREDUMP(rv)) 483 reason = CLD_DUMPED; 484 else if (WIFSIGNALED(rv)) 485 reason = CLD_KILLED; 486 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 487 #endif 488 489 /* 490 * Just delete all entries in the p_klist. At this point we won't 491 * report any more events, and there are nasty race conditions that 492 * can beat us if we don't. 493 */ 494 knlist_clear(&p->p_klist, 1); 495 496 /* 497 * If this is a process with a descriptor, we may not need to deliver 498 * a signal to the parent. proctree_lock is held over 499 * procdesc_exit() to serialize concurrent calls to close() and 500 * exit(). 501 */ 502 if (p->p_procdesc == NULL || procdesc_exit(p)) { 503 /* 504 * Notify parent that we're gone. If parent has the 505 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 506 * notify process 1 instead (and hope it will handle this 507 * situation). 508 */ 509 PROC_LOCK(p->p_pptr); 510 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 511 if (p->p_pptr->p_sigacts->ps_flag & 512 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 513 struct proc *pp; 514 515 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 516 pp = p->p_pptr; 517 PROC_UNLOCK(pp); 518 proc_reparent(p, initproc); 519 p->p_sigparent = SIGCHLD; 520 PROC_LOCK(p->p_pptr); 521 522 /* 523 * Notify parent, so in case he was wait(2)ing or 524 * executing waitpid(2) with our pid, he will 525 * continue. 526 */ 527 wakeup(pp); 528 } else 529 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 530 531 if (p->p_pptr == initproc) 532 kern_psignal(p->p_pptr, SIGCHLD); 533 else if (p->p_sigparent != 0) { 534 if (p->p_sigparent == SIGCHLD) 535 childproc_exited(p); 536 else /* LINUX thread */ 537 kern_psignal(p->p_pptr, p->p_sigparent); 538 } 539 } else 540 PROC_LOCK(p->p_pptr); 541 sx_xunlock(&proctree_lock); 542 543 /* 544 * The state PRS_ZOMBIE prevents other proesses from sending 545 * signal to the process, to avoid memory leak, we free memory 546 * for signal queue at the time when the state is set. 547 */ 548 sigqueue_flush(&p->p_sigqueue); 549 sigqueue_flush(&td->td_sigqueue); 550 551 /* 552 * We have to wait until after acquiring all locks before 553 * changing p_state. We need to avoid all possible context 554 * switches (including ones from blocking on a mutex) while 555 * marked as a zombie. We also have to set the zombie state 556 * before we release the parent process' proc lock to avoid 557 * a lost wakeup. So, we first call wakeup, then we grab the 558 * sched lock, update the state, and release the parent process' 559 * proc lock. 560 */ 561 wakeup(p->p_pptr); 562 cv_broadcast(&p->p_pwait); 563 sched_exit(p->p_pptr, td); 564 PROC_SLOCK(p); 565 p->p_state = PRS_ZOMBIE; 566 PROC_UNLOCK(p->p_pptr); 567 568 /* 569 * Hopefully no one will try to deliver a signal to the process this 570 * late in the game. 571 */ 572 knlist_destroy(&p->p_klist); 573 574 /* 575 * Save our children's rusage information in our exit rusage. 576 */ 577 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 578 579 /* 580 * Make sure the scheduler takes this thread out of its tables etc. 581 * This will also release this thread's reference to the ucred. 582 * Other thread parts to release include pcb bits and such. 583 */ 584 thread_exit(); 585 } 586 587 588 #ifndef _SYS_SYSPROTO_H_ 589 struct abort2_args { 590 char *why; 591 int nargs; 592 void **args; 593 }; 594 #endif 595 596 int 597 sys_abort2(struct thread *td, struct abort2_args *uap) 598 { 599 struct proc *p = td->td_proc; 600 struct sbuf *sb; 601 void *uargs[16]; 602 int error, i, sig; 603 604 /* 605 * Do it right now so we can log either proper call of abort2(), or 606 * note, that invalid argument was passed. 512 is big enough to 607 * handle 16 arguments' descriptions with additional comments. 608 */ 609 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 610 sbuf_clear(sb); 611 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 612 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 613 /* 614 * Since we can't return from abort2(), send SIGKILL in cases, where 615 * abort2() was called improperly 616 */ 617 sig = SIGKILL; 618 /* Prevent from DoSes from user-space. */ 619 if (uap->nargs < 0 || uap->nargs > 16) 620 goto out; 621 if (uap->nargs > 0) { 622 if (uap->args == NULL) 623 goto out; 624 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 625 if (error != 0) 626 goto out; 627 } 628 /* 629 * Limit size of 'reason' string to 128. Will fit even when 630 * maximal number of arguments was chosen to be logged. 631 */ 632 if (uap->why != NULL) { 633 error = sbuf_copyin(sb, uap->why, 128); 634 if (error < 0) 635 goto out; 636 } else { 637 sbuf_printf(sb, "(null)"); 638 } 639 if (uap->nargs > 0) { 640 sbuf_printf(sb, "("); 641 for (i = 0;i < uap->nargs; i++) 642 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 643 sbuf_printf(sb, ")"); 644 } 645 /* 646 * Final stage: arguments were proper, string has been 647 * successfully copied from userspace, and copying pointers 648 * from user-space succeed. 649 */ 650 sig = SIGABRT; 651 out: 652 if (sig == SIGKILL) { 653 sbuf_trim(sb); 654 sbuf_printf(sb, " (Reason text inaccessible)"); 655 } 656 sbuf_cat(sb, "\n"); 657 sbuf_finish(sb); 658 log(LOG_INFO, "%s", sbuf_data(sb)); 659 sbuf_delete(sb); 660 exit1(td, W_EXITCODE(0, sig)); 661 return (0); 662 } 663 664 665 #ifdef COMPAT_43 666 /* 667 * The dirty work is handled by kern_wait(). 668 */ 669 int 670 owait(struct thread *td, struct owait_args *uap __unused) 671 { 672 int error, status; 673 674 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 675 if (error == 0) 676 td->td_retval[1] = status; 677 return (error); 678 } 679 #endif /* COMPAT_43 */ 680 681 /* 682 * The dirty work is handled by kern_wait(). 683 */ 684 int 685 sys_wait4(struct thread *td, struct wait4_args *uap) 686 { 687 struct rusage ru, *rup; 688 int error, status; 689 690 if (uap->rusage != NULL) 691 rup = &ru; 692 else 693 rup = NULL; 694 error = kern_wait(td, uap->pid, &status, uap->options, rup); 695 if (uap->status != NULL && error == 0) 696 error = copyout(&status, uap->status, sizeof(status)); 697 if (uap->rusage != NULL && error == 0) 698 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 699 return (error); 700 } 701 702 int 703 sys_wait6(struct thread *td, struct wait6_args *uap) 704 { 705 struct __wrusage wru, *wrup; 706 siginfo_t si, *sip; 707 idtype_t idtype; 708 id_t id; 709 int error, status; 710 711 idtype = uap->idtype; 712 id = uap->id; 713 714 if (uap->wrusage != NULL) 715 wrup = &wru; 716 else 717 wrup = NULL; 718 719 if (uap->info != NULL) { 720 sip = &si; 721 bzero(sip, sizeof(*sip)); 722 } else 723 sip = NULL; 724 725 /* 726 * We expect all callers of wait6() to know about WEXITED and 727 * WTRAPPED. 728 */ 729 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 730 731 if (uap->status != NULL && error == 0) 732 error = copyout(&status, uap->status, sizeof(status)); 733 if (uap->wrusage != NULL && error == 0) 734 error = copyout(&wru, uap->wrusage, sizeof(wru)); 735 if (uap->info != NULL && error == 0) 736 error = copyout(&si, uap->info, sizeof(si)); 737 return (error); 738 } 739 740 /* 741 * Reap the remains of a zombie process and optionally return status and 742 * rusage. Asserts and will release both the proctree_lock and the process 743 * lock as part of its work. 744 */ 745 void 746 proc_reap(struct thread *td, struct proc *p, int *status, int options) 747 { 748 struct proc *q, *t; 749 750 sx_assert(&proctree_lock, SA_XLOCKED); 751 PROC_LOCK_ASSERT(p, MA_OWNED); 752 PROC_SLOCK_ASSERT(p, MA_OWNED); 753 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 754 755 q = td->td_proc; 756 757 PROC_SUNLOCK(p); 758 td->td_retval[0] = p->p_pid; 759 if (status) 760 *status = p->p_xstat; /* convert to int */ 761 if (options & WNOWAIT) { 762 /* 763 * Only poll, returning the status. Caller does not wish to 764 * release the proc struct just yet. 765 */ 766 PROC_UNLOCK(p); 767 sx_xunlock(&proctree_lock); 768 return; 769 } 770 771 PROC_LOCK(q); 772 sigqueue_take(p->p_ksi); 773 PROC_UNLOCK(q); 774 PROC_UNLOCK(p); 775 776 /* 777 * If we got the child via a ptrace 'attach', we need to give it back 778 * to the old parent. 779 */ 780 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 781 PROC_LOCK(p); 782 proc_reparent(p, t); 783 p->p_oppid = 0; 784 PROC_UNLOCK(p); 785 pksignal(t, SIGCHLD, p->p_ksi); 786 wakeup(t); 787 cv_broadcast(&p->p_pwait); 788 PROC_UNLOCK(t); 789 sx_xunlock(&proctree_lock); 790 return; 791 } 792 793 /* 794 * Remove other references to this process to ensure we have an 795 * exclusive reference. 796 */ 797 sx_xlock(&allproc_lock); 798 LIST_REMOVE(p, p_list); /* off zombproc */ 799 sx_xunlock(&allproc_lock); 800 LIST_REMOVE(p, p_sibling); 801 PROC_LOCK(p); 802 clear_orphan(p); 803 PROC_UNLOCK(p); 804 leavepgrp(p); 805 if (p->p_procdesc != NULL) 806 procdesc_reap(p); 807 sx_xunlock(&proctree_lock); 808 809 /* 810 * As a side effect of this lock, we know that all other writes to 811 * this proc are visible now, so no more locking is needed for p. 812 */ 813 PROC_LOCK(p); 814 p->p_xstat = 0; /* XXX: why? */ 815 PROC_UNLOCK(p); 816 PROC_LOCK(q); 817 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 818 PROC_UNLOCK(q); 819 820 /* 821 * Decrement the count of procs running with this uid. 822 */ 823 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 824 825 /* 826 * Destroy resource accounting information associated with the process. 827 */ 828 #ifdef RACCT 829 PROC_LOCK(p); 830 racct_sub(p, RACCT_NPROC, 1); 831 PROC_UNLOCK(p); 832 #endif 833 racct_proc_exit(p); 834 835 /* 836 * Free credentials, arguments, and sigacts. 837 */ 838 crfree(p->p_ucred); 839 p->p_ucred = NULL; 840 pargs_drop(p->p_args); 841 p->p_args = NULL; 842 sigacts_free(p->p_sigacts); 843 p->p_sigacts = NULL; 844 845 /* 846 * Do any thread-system specific cleanups. 847 */ 848 thread_wait(p); 849 850 /* 851 * Give vm and machine-dependent layer a chance to free anything that 852 * cpu_exit couldn't release while still running in process context. 853 */ 854 vm_waitproc(p); 855 #ifdef MAC 856 mac_proc_destroy(p); 857 #endif 858 KASSERT(FIRST_THREAD_IN_PROC(p), 859 ("proc_reap: no residual thread!")); 860 uma_zfree(proc_zone, p); 861 sx_xlock(&allproc_lock); 862 nprocs--; 863 sx_xunlock(&allproc_lock); 864 } 865 866 static int 867 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 868 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo) 869 { 870 struct proc *q; 871 struct rusage *rup; 872 873 sx_assert(&proctree_lock, SA_XLOCKED); 874 875 q = td->td_proc; 876 PROC_LOCK(p); 877 878 switch (idtype) { 879 case P_ALL: 880 break; 881 case P_PID: 882 if (p->p_pid != (pid_t)id) { 883 PROC_UNLOCK(p); 884 return (0); 885 } 886 break; 887 case P_PGID: 888 if (p->p_pgid != (pid_t)id) { 889 PROC_UNLOCK(p); 890 return (0); 891 } 892 break; 893 case P_SID: 894 if (p->p_session->s_sid != (pid_t)id) { 895 PROC_UNLOCK(p); 896 return (0); 897 } 898 break; 899 case P_UID: 900 if (p->p_ucred->cr_uid != (uid_t)id) { 901 PROC_UNLOCK(p); 902 return (0); 903 } 904 break; 905 case P_GID: 906 if (p->p_ucred->cr_gid != (gid_t)id) { 907 PROC_UNLOCK(p); 908 return (0); 909 } 910 break; 911 case P_JAILID: 912 if (p->p_ucred->cr_prison->pr_id != (int)id) { 913 PROC_UNLOCK(p); 914 return (0); 915 } 916 break; 917 /* 918 * It seems that the thread structures get zeroed out 919 * at process exit. This makes it impossible to 920 * support P_SETID, P_CID or P_CPUID. 921 */ 922 default: 923 PROC_UNLOCK(p); 924 return (0); 925 } 926 927 if (p_canwait(td, p)) { 928 PROC_UNLOCK(p); 929 return (0); 930 } 931 932 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 933 PROC_UNLOCK(p); 934 return (0); 935 } 936 937 /* 938 * This special case handles a kthread spawned by linux_clone 939 * (see linux_misc.c). The linux_wait4 and linux_waitpid 940 * functions need to be able to distinguish between waiting 941 * on a process and waiting on a thread. It is a thread if 942 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 943 * signifies we want to wait for threads and not processes. 944 */ 945 if ((p->p_sigparent != SIGCHLD) ^ 946 ((options & WLINUXCLONE) != 0)) { 947 PROC_UNLOCK(p); 948 return (0); 949 } 950 951 PROC_SLOCK(p); 952 953 if (siginfo != NULL) { 954 bzero(siginfo, sizeof(*siginfo)); 955 siginfo->si_errno = 0; 956 957 /* 958 * SUSv4 requires that the si_signo value is always 959 * SIGCHLD. Obey it despite the rfork(2) interface 960 * allows to request other signal for child exit 961 * notification. 962 */ 963 siginfo->si_signo = SIGCHLD; 964 965 /* 966 * This is still a rough estimate. We will fix the 967 * cases TRAPPED, STOPPED, and CONTINUED later. 968 */ 969 if (WCOREDUMP(p->p_xstat)) { 970 siginfo->si_code = CLD_DUMPED; 971 siginfo->si_status = WTERMSIG(p->p_xstat); 972 } else if (WIFSIGNALED(p->p_xstat)) { 973 siginfo->si_code = CLD_KILLED; 974 siginfo->si_status = WTERMSIG(p->p_xstat); 975 } else { 976 siginfo->si_code = CLD_EXITED; 977 siginfo->si_status = WEXITSTATUS(p->p_xstat); 978 } 979 980 siginfo->si_pid = p->p_pid; 981 siginfo->si_uid = p->p_ucred->cr_uid; 982 983 /* 984 * The si_addr field would be useful additional 985 * detail, but apparently the PC value may be lost 986 * when we reach this point. bzero() above sets 987 * siginfo->si_addr to NULL. 988 */ 989 } 990 991 /* 992 * There should be no reason to limit resources usage info to 993 * exited processes only. A snapshot about any resources used 994 * by a stopped process may be exactly what is needed. 995 */ 996 if (wrusage != NULL) { 997 rup = &wrusage->wru_self; 998 *rup = p->p_ru; 999 calcru(p, &rup->ru_utime, &rup->ru_stime); 1000 1001 rup = &wrusage->wru_children; 1002 *rup = p->p_stats->p_cru; 1003 calccru(p, &rup->ru_utime, &rup->ru_stime); 1004 } 1005 1006 if (p->p_state == PRS_ZOMBIE) { 1007 proc_reap(td, p, status, options); 1008 return (-1); 1009 } 1010 PROC_SUNLOCK(p); 1011 PROC_UNLOCK(p); 1012 return (1); 1013 } 1014 1015 int 1016 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1017 struct rusage *rusage) 1018 { 1019 struct __wrusage wru, *wrup; 1020 idtype_t idtype; 1021 id_t id; 1022 int ret; 1023 1024 /* 1025 * Translate the special pid values into the (idtype, pid) 1026 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1027 * kern_wait6() on its own. 1028 */ 1029 if (pid == WAIT_ANY) { 1030 idtype = P_ALL; 1031 id = 0; 1032 } else if (pid < 0) { 1033 idtype = P_PGID; 1034 id = (id_t)-pid; 1035 } else { 1036 idtype = P_PID; 1037 id = (id_t)pid; 1038 } 1039 1040 if (rusage != NULL) 1041 wrup = &wru; 1042 else 1043 wrup = NULL; 1044 1045 /* 1046 * For backward compatibility we implicitly add flags WEXITED 1047 * and WTRAPPED here. 1048 */ 1049 options |= WEXITED | WTRAPPED; 1050 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1051 if (rusage != NULL) 1052 *rusage = wru.wru_self; 1053 return (ret); 1054 } 1055 1056 int 1057 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1058 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1059 { 1060 struct proc *p, *q; 1061 int error, nfound, ret; 1062 1063 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1064 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1065 AUDIT_ARG_VALUE(options); 1066 1067 q = td->td_proc; 1068 1069 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1070 PROC_LOCK(q); 1071 id = (id_t)q->p_pgid; 1072 PROC_UNLOCK(q); 1073 idtype = P_PGID; 1074 } 1075 1076 /* If we don't know the option, just return. */ 1077 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1078 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1079 return (EINVAL); 1080 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1081 /* 1082 * We will be unable to find any matching processes, 1083 * because there are no known events to look for. 1084 * Prefer to return error instead of blocking 1085 * indefinitely. 1086 */ 1087 return (EINVAL); 1088 } 1089 1090 loop: 1091 if (q->p_flag & P_STATCHILD) { 1092 PROC_LOCK(q); 1093 q->p_flag &= ~P_STATCHILD; 1094 PROC_UNLOCK(q); 1095 } 1096 nfound = 0; 1097 sx_xlock(&proctree_lock); 1098 LIST_FOREACH(p, &q->p_children, p_sibling) { 1099 ret = proc_to_reap(td, p, idtype, id, status, options, 1100 wrusage, siginfo); 1101 if (ret == 0) 1102 continue; 1103 else if (ret == 1) 1104 nfound++; 1105 else 1106 return (0); 1107 1108 PROC_LOCK(p); 1109 PROC_SLOCK(p); 1110 1111 if ((options & WTRAPPED) != 0 && 1112 (p->p_flag & P_TRACED) != 0 && 1113 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1114 (p->p_suspcount == p->p_numthreads) && 1115 ((p->p_flag & P_WAITED) == 0)) { 1116 PROC_SUNLOCK(p); 1117 if ((options & WNOWAIT) == 0) 1118 p->p_flag |= P_WAITED; 1119 sx_xunlock(&proctree_lock); 1120 td->td_retval[0] = p->p_pid; 1121 1122 if (status != NULL) 1123 *status = W_STOPCODE(p->p_xstat); 1124 if (siginfo != NULL) { 1125 siginfo->si_status = p->p_xstat; 1126 siginfo->si_code = CLD_TRAPPED; 1127 } 1128 if ((options & WNOWAIT) == 0) { 1129 PROC_LOCK(q); 1130 sigqueue_take(p->p_ksi); 1131 PROC_UNLOCK(q); 1132 } 1133 1134 PROC_UNLOCK(p); 1135 return (0); 1136 } 1137 if ((options & WUNTRACED) != 0 && 1138 (p->p_flag & P_STOPPED_SIG) != 0 && 1139 (p->p_suspcount == p->p_numthreads) && 1140 ((p->p_flag & P_WAITED) == 0)) { 1141 PROC_SUNLOCK(p); 1142 if ((options & WNOWAIT) == 0) 1143 p->p_flag |= P_WAITED; 1144 sx_xunlock(&proctree_lock); 1145 td->td_retval[0] = p->p_pid; 1146 1147 if (status != NULL) 1148 *status = W_STOPCODE(p->p_xstat); 1149 if (siginfo != NULL) { 1150 siginfo->si_status = p->p_xstat; 1151 siginfo->si_code = CLD_STOPPED; 1152 } 1153 if ((options & WNOWAIT) == 0) { 1154 PROC_LOCK(q); 1155 sigqueue_take(p->p_ksi); 1156 PROC_UNLOCK(q); 1157 } 1158 1159 PROC_UNLOCK(p); 1160 return (0); 1161 } 1162 PROC_SUNLOCK(p); 1163 if ((options & WCONTINUED) != 0 && 1164 (p->p_flag & P_CONTINUED) != 0) { 1165 sx_xunlock(&proctree_lock); 1166 td->td_retval[0] = p->p_pid; 1167 if ((options & WNOWAIT) == 0) { 1168 p->p_flag &= ~P_CONTINUED; 1169 PROC_LOCK(q); 1170 sigqueue_take(p->p_ksi); 1171 PROC_UNLOCK(q); 1172 } 1173 PROC_UNLOCK(p); 1174 1175 if (status != NULL) 1176 *status = SIGCONT; 1177 if (siginfo != NULL) { 1178 siginfo->si_status = SIGCONT; 1179 siginfo->si_code = CLD_CONTINUED; 1180 } 1181 return (0); 1182 } 1183 PROC_UNLOCK(p); 1184 } 1185 1186 /* 1187 * Look in the orphans list too, to allow the parent to 1188 * collect it's child exit status even if child is being 1189 * debugged. 1190 * 1191 * Debugger detaches from the parent upon successful 1192 * switch-over from parent to child. At this point due to 1193 * re-parenting the parent loses the child to debugger and a 1194 * wait4(2) call would report that it has no children to wait 1195 * for. By maintaining a list of orphans we allow the parent 1196 * to successfully wait until the child becomes a zombie. 1197 */ 1198 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1199 ret = proc_to_reap(td, p, idtype, id, status, options, 1200 wrusage, siginfo); 1201 if (ret == 0) 1202 continue; 1203 else if (ret == 1) 1204 nfound++; 1205 else 1206 return (0); 1207 } 1208 if (nfound == 0) { 1209 sx_xunlock(&proctree_lock); 1210 return (ECHILD); 1211 } 1212 if (options & WNOHANG) { 1213 sx_xunlock(&proctree_lock); 1214 td->td_retval[0] = 0; 1215 return (0); 1216 } 1217 PROC_LOCK(q); 1218 sx_xunlock(&proctree_lock); 1219 if (q->p_flag & P_STATCHILD) { 1220 q->p_flag &= ~P_STATCHILD; 1221 error = 0; 1222 } else 1223 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1224 PROC_UNLOCK(q); 1225 if (error) 1226 return (error); 1227 goto loop; 1228 } 1229 1230 /* 1231 * Make process 'parent' the new parent of process 'child'. 1232 * Must be called with an exclusive hold of proctree lock. 1233 */ 1234 void 1235 proc_reparent(struct proc *child, struct proc *parent) 1236 { 1237 1238 sx_assert(&proctree_lock, SX_XLOCKED); 1239 PROC_LOCK_ASSERT(child, MA_OWNED); 1240 if (child->p_pptr == parent) 1241 return; 1242 1243 PROC_LOCK(child->p_pptr); 1244 sigqueue_take(child->p_ksi); 1245 PROC_UNLOCK(child->p_pptr); 1246 LIST_REMOVE(child, p_sibling); 1247 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1248 1249 clear_orphan(child); 1250 if (child->p_flag & P_TRACED) { 1251 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, p_orphan); 1252 child->p_flag |= P_ORPHAN; 1253 } 1254 1255 child->p_pptr = parent; 1256 } 1257