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/capsicum.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 struct proc * 101 proc_realparent(struct proc *child) 102 { 103 struct proc *p, *parent; 104 105 sx_assert(&proctree_lock, SX_LOCKED); 106 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 107 if (child->p_oppid == 0 || 108 child->p_pptr->p_pid == child->p_oppid) 109 parent = child->p_pptr; 110 else 111 parent = initproc; 112 return (parent); 113 } 114 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 115 /* Cannot use LIST_PREV(), since the list head is not known. */ 116 p = __containerof(p->p_orphan.le_prev, struct proc, 117 p_orphan.le_next); 118 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 119 ("missing P_ORPHAN %p", p)); 120 } 121 parent = __containerof(p->p_orphan.le_prev, struct proc, 122 p_orphans.lh_first); 123 return (parent); 124 } 125 126 static void 127 clear_orphan(struct proc *p) 128 { 129 struct proc *p1; 130 131 sx_assert(&proctree_lock, SA_XLOCKED); 132 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 133 return; 134 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 135 p1 = LIST_NEXT(p, p_orphan); 136 if (p1 != NULL) 137 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 138 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 139 } 140 LIST_REMOVE(p, p_orphan); 141 p->p_treeflag &= ~P_TREE_ORPHANED; 142 } 143 144 /* 145 * exit -- death of process. 146 */ 147 void 148 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 149 { 150 151 exit1(td, W_EXITCODE(uap->rval, 0)); 152 /* NOTREACHED */ 153 } 154 155 /* 156 * Exit: deallocate address space and other resources, change proc state to 157 * zombie, and unlink proc from allproc and parent's lists. Save exit status 158 * and rusage for wait(). Check for child processes and orphan them. 159 */ 160 void 161 exit1(struct thread *td, int rv) 162 { 163 struct proc *p, *nq, *q, *t; 164 struct thread *tdt; 165 struct vnode *ttyvp = NULL; 166 167 mtx_assert(&Giant, MA_NOTOWNED); 168 169 p = td->td_proc; 170 /* 171 * XXX in case we're rebooting we just let init die in order to 172 * work around an unsolved stack overflow seen very late during 173 * shutdown on sparc64 when the gmirror worker process exists. 174 */ 175 if (p == initproc && rebooting == 0) { 176 printf("init died (signal %d, exit %d)\n", 177 WTERMSIG(rv), WEXITSTATUS(rv)); 178 panic("Going nowhere without my init!"); 179 } 180 181 /* 182 * MUST abort all other threads before proceeding past here. 183 */ 184 PROC_LOCK(p); 185 while (p->p_flag & P_HADTHREADS) { 186 /* 187 * First check if some other thread got here before us. 188 * If so, act appropriately: exit or suspend. 189 */ 190 thread_suspend_check(0); 191 192 /* 193 * Kill off the other threads. This requires 194 * some co-operation from other parts of the kernel 195 * so it may not be instantaneous. With this state set 196 * any thread entering the kernel from userspace will 197 * thread_exit() in trap(). Any thread attempting to 198 * sleep will return immediately with EINTR or EWOULDBLOCK 199 * which will hopefully force them to back out to userland 200 * freeing resources as they go. Any thread attempting 201 * to return to userland will thread_exit() from userret(). 202 * thread_exit() will unsuspend us when the last of the 203 * other threads exits. 204 * If there is already a thread singler after resumption, 205 * calling thread_single will fail; in that case, we just 206 * re-check all suspension request, the thread should 207 * either be suspended there or exit. 208 */ 209 if (!thread_single(SINGLE_EXIT)) 210 break; 211 212 /* 213 * All other activity in this process is now stopped. 214 * Threading support has been turned off. 215 */ 216 } 217 KASSERT(p->p_numthreads == 1, 218 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 219 racct_sub(p, RACCT_NTHR, 1); 220 /* 221 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 222 * on our vmspace, so we should block below until they have 223 * released their reference to us. Note that if they have 224 * requested S_EXIT stops we will block here until they ack 225 * via PIOCCONT. 226 */ 227 _STOPEVENT(p, S_EXIT, rv); 228 229 /* 230 * Ignore any pending request to stop due to a stop signal. 231 * Once P_WEXIT is set, future requests will be ignored as 232 * well. 233 */ 234 p->p_flag &= ~P_STOPPED_SIG; 235 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 236 237 /* 238 * Note that we are exiting and do another wakeup of anyone in 239 * PIOCWAIT in case they aren't listening for S_EXIT stops or 240 * decided to wait again after we told them we are exiting. 241 */ 242 p->p_flag |= P_WEXIT; 243 wakeup(&p->p_stype); 244 245 /* 246 * Wait for any processes that have a hold on our vmspace to 247 * release their reference. 248 */ 249 while (p->p_lock > 0) 250 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 251 252 p->p_xstat = rv; /* Let event handler change exit status */ 253 PROC_UNLOCK(p); 254 /* Drain the limit callout while we don't have the proc locked */ 255 callout_drain(&p->p_limco); 256 257 #ifdef AUDIT 258 /* 259 * The Sun BSM exit token contains two components: an exit status as 260 * passed to exit(), and a return value to indicate what sort of exit 261 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 262 * what the return value is. 263 */ 264 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0); 265 AUDIT_SYSCALL_EXIT(0, td); 266 #endif 267 268 /* Are we a task leader with peers? */ 269 if (p->p_peers != NULL && p == p->p_leader) { 270 mtx_lock(&ppeers_lock); 271 q = p->p_peers; 272 while (q != NULL) { 273 PROC_LOCK(q); 274 kern_psignal(q, SIGKILL); 275 PROC_UNLOCK(q); 276 q = q->p_peers; 277 } 278 while (p->p_peers != NULL) 279 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 280 mtx_unlock(&ppeers_lock); 281 } 282 283 /* 284 * Check if any loadable modules need anything done at process exit. 285 * E.g. SYSV IPC stuff 286 * XXX what if one of these generates an error? 287 */ 288 EVENTHANDLER_INVOKE(process_exit, p); 289 290 /* 291 * If parent is waiting for us to exit or exec, 292 * P_PPWAIT is set; we will wakeup the parent below. 293 */ 294 PROC_LOCK(p); 295 rv = p->p_xstat; /* Event handler could change exit status */ 296 stopprofclock(p); 297 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 298 299 /* 300 * Stop the real interval timer. If the handler is currently 301 * executing, prevent it from rearming itself and let it finish. 302 */ 303 if (timevalisset(&p->p_realtimer.it_value) && 304 callout_stop(&p->p_itcallout) == 0) { 305 timevalclear(&p->p_realtimer.it_interval); 306 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 307 KASSERT(!timevalisset(&p->p_realtimer.it_value), 308 ("realtime timer is still armed")); 309 } 310 PROC_UNLOCK(p); 311 312 /* 313 * Reset any sigio structures pointing to us as a result of 314 * F_SETOWN with our pid. 315 */ 316 funsetownlst(&p->p_sigiolst); 317 318 /* 319 * If this process has an nlminfo data area (for lockd), release it 320 */ 321 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 322 (*nlminfo_release_p)(p); 323 324 /* 325 * Close open files and release open-file table. 326 * This may block! 327 */ 328 fdescfree(td); 329 330 /* 331 * If this thread tickled GEOM, we need to wait for the giggling to 332 * stop before we return to userland 333 */ 334 if (td->td_pflags & TDP_GEOM) 335 g_waitidle(); 336 337 /* 338 * Remove ourself from our leader's peer list and wake our leader. 339 */ 340 if (p->p_leader->p_peers != NULL) { 341 mtx_lock(&ppeers_lock); 342 if (p->p_leader->p_peers != NULL) { 343 q = p->p_leader; 344 while (q->p_peers != p) 345 q = q->p_peers; 346 q->p_peers = p->p_peers; 347 wakeup(p->p_leader); 348 } 349 mtx_unlock(&ppeers_lock); 350 } 351 352 vmspace_exit(td); 353 354 sx_xlock(&proctree_lock); 355 if (SESS_LEADER(p)) { 356 struct session *sp = p->p_session; 357 struct tty *tp; 358 359 /* 360 * s_ttyp is not zero'd; we use this to indicate that 361 * the session once had a controlling terminal. (for 362 * logging and informational purposes) 363 */ 364 SESS_LOCK(sp); 365 ttyvp = sp->s_ttyvp; 366 tp = sp->s_ttyp; 367 sp->s_ttyvp = NULL; 368 sp->s_ttydp = NULL; 369 sp->s_leader = NULL; 370 SESS_UNLOCK(sp); 371 372 /* 373 * Signal foreground pgrp and revoke access to 374 * controlling terminal if it has not been revoked 375 * already. 376 * 377 * Because the TTY may have been revoked in the mean 378 * time and could already have a new session associated 379 * with it, make sure we don't send a SIGHUP to a 380 * foreground process group that does not belong to this 381 * session. 382 */ 383 384 if (tp != NULL) { 385 tty_lock(tp); 386 if (tp->t_session == sp) 387 tty_signal_pgrp(tp, SIGHUP); 388 tty_unlock(tp); 389 } 390 391 if (ttyvp != NULL) { 392 sx_xunlock(&proctree_lock); 393 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 394 VOP_REVOKE(ttyvp, REVOKEALL); 395 VOP_UNLOCK(ttyvp, 0); 396 } 397 sx_xlock(&proctree_lock); 398 } 399 } 400 fixjobc(p, p->p_pgrp, 0); 401 sx_xunlock(&proctree_lock); 402 (void)acct_process(td); 403 404 /* Release the TTY now we've unlocked everything. */ 405 if (ttyvp != NULL) 406 vrele(ttyvp); 407 #ifdef KTRACE 408 ktrprocexit(td); 409 #endif 410 /* 411 * Release reference to text vnode 412 */ 413 if (p->p_textvp != NULL) { 414 vrele(p->p_textvp); 415 p->p_textvp = NULL; 416 } 417 418 /* 419 * Release our limits structure. 420 */ 421 lim_free(p->p_limit); 422 p->p_limit = NULL; 423 424 tidhash_remove(td); 425 426 /* 427 * Remove proc from allproc queue and pidhash chain. 428 * Place onto zombproc. Unlink from parent's child list. 429 */ 430 sx_xlock(&allproc_lock); 431 LIST_REMOVE(p, p_list); 432 LIST_INSERT_HEAD(&zombproc, p, p_list); 433 LIST_REMOVE(p, p_hash); 434 sx_xunlock(&allproc_lock); 435 436 /* 437 * Call machine-dependent code to release any 438 * machine-dependent resources other than the address space. 439 * The address space is released by "vmspace_exitfree(p)" in 440 * vm_waitproc(). 441 */ 442 cpu_exit(td); 443 444 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 445 446 /* 447 * Reparent all children processes: 448 * - traced ones to the original parent (or init if we are that parent) 449 * - the rest to init 450 */ 451 sx_xlock(&proctree_lock); 452 q = LIST_FIRST(&p->p_children); 453 if (q != NULL) /* only need this if any child is S_ZOMB */ 454 wakeup(initproc); 455 for (; q != NULL; q = nq) { 456 nq = LIST_NEXT(q, p_sibling); 457 PROC_LOCK(q); 458 q->p_sigparent = SIGCHLD; 459 460 if (!(q->p_flag & P_TRACED)) { 461 proc_reparent(q, initproc); 462 } else { 463 /* 464 * Traced processes are killed since their existence 465 * means someone is screwing up. 466 */ 467 t = proc_realparent(q); 468 if (t == p) { 469 proc_reparent(q, initproc); 470 } else { 471 PROC_LOCK(t); 472 proc_reparent(q, t); 473 PROC_UNLOCK(t); 474 } 475 /* 476 * Since q was found on our children list, the 477 * proc_reparent() call moved q to the orphan 478 * list due to present P_TRACED flag. Clear 479 * orphan link for q now while q is locked. 480 */ 481 clear_orphan(q); 482 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 483 FOREACH_THREAD_IN_PROC(q, tdt) 484 tdt->td_dbgflags &= ~TDB_SUSPEND; 485 kern_psignal(q, SIGKILL); 486 } 487 PROC_UNLOCK(q); 488 } 489 490 /* 491 * Also get rid of our orphans. 492 */ 493 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 494 PROC_LOCK(q); 495 clear_orphan(q); 496 PROC_UNLOCK(q); 497 } 498 499 /* Save exit status. */ 500 PROC_LOCK(p); 501 p->p_xthread = td; 502 503 /* Tell the prison that we are gone. */ 504 prison_proc_free(p->p_ucred->cr_prison); 505 506 #ifdef KDTRACE_HOOKS 507 /* 508 * Tell the DTrace fasttrap provider about the exit if it 509 * has declared an interest. 510 */ 511 if (dtrace_fasttrap_exit) 512 dtrace_fasttrap_exit(p); 513 #endif 514 515 /* 516 * Notify interested parties of our demise. 517 */ 518 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 519 520 #ifdef KDTRACE_HOOKS 521 int reason = CLD_EXITED; 522 if (WCOREDUMP(rv)) 523 reason = CLD_DUMPED; 524 else if (WIFSIGNALED(rv)) 525 reason = CLD_KILLED; 526 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 527 #endif 528 529 /* 530 * Just delete all entries in the p_klist. At this point we won't 531 * report any more events, and there are nasty race conditions that 532 * can beat us if we don't. 533 */ 534 knlist_clear(&p->p_klist, 1); 535 536 /* 537 * If this is a process with a descriptor, we may not need to deliver 538 * a signal to the parent. proctree_lock is held over 539 * procdesc_exit() to serialize concurrent calls to close() and 540 * exit(). 541 */ 542 if (p->p_procdesc == NULL || procdesc_exit(p)) { 543 /* 544 * Notify parent that we're gone. If parent has the 545 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 546 * notify process 1 instead (and hope it will handle this 547 * situation). 548 */ 549 PROC_LOCK(p->p_pptr); 550 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 551 if (p->p_pptr->p_sigacts->ps_flag & 552 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 553 struct proc *pp; 554 555 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 556 pp = p->p_pptr; 557 PROC_UNLOCK(pp); 558 proc_reparent(p, initproc); 559 p->p_sigparent = SIGCHLD; 560 PROC_LOCK(p->p_pptr); 561 562 /* 563 * Notify parent, so in case he was wait(2)ing or 564 * executing waitpid(2) with our pid, he will 565 * continue. 566 */ 567 wakeup(pp); 568 } else 569 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 570 571 if (p->p_pptr == initproc) 572 kern_psignal(p->p_pptr, SIGCHLD); 573 else if (p->p_sigparent != 0) { 574 if (p->p_sigparent == SIGCHLD) 575 childproc_exited(p); 576 else /* LINUX thread */ 577 kern_psignal(p->p_pptr, p->p_sigparent); 578 } 579 } else 580 PROC_LOCK(p->p_pptr); 581 sx_xunlock(&proctree_lock); 582 583 /* 584 * The state PRS_ZOMBIE prevents other proesses from sending 585 * signal to the process, to avoid memory leak, we free memory 586 * for signal queue at the time when the state is set. 587 */ 588 sigqueue_flush(&p->p_sigqueue); 589 sigqueue_flush(&td->td_sigqueue); 590 591 /* 592 * We have to wait until after acquiring all locks before 593 * changing p_state. We need to avoid all possible context 594 * switches (including ones from blocking on a mutex) while 595 * marked as a zombie. We also have to set the zombie state 596 * before we release the parent process' proc lock to avoid 597 * a lost wakeup. So, we first call wakeup, then we grab the 598 * sched lock, update the state, and release the parent process' 599 * proc lock. 600 */ 601 wakeup(p->p_pptr); 602 cv_broadcast(&p->p_pwait); 603 sched_exit(p->p_pptr, td); 604 PROC_SLOCK(p); 605 p->p_state = PRS_ZOMBIE; 606 PROC_UNLOCK(p->p_pptr); 607 608 /* 609 * Hopefully no one will try to deliver a signal to the process this 610 * late in the game. 611 */ 612 knlist_destroy(&p->p_klist); 613 614 /* 615 * Save our children's rusage information in our exit rusage. 616 */ 617 PROC_STATLOCK(p); 618 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 619 PROC_STATUNLOCK(p); 620 621 /* 622 * Make sure the scheduler takes this thread out of its tables etc. 623 * This will also release this thread's reference to the ucred. 624 * Other thread parts to release include pcb bits and such. 625 */ 626 thread_exit(); 627 } 628 629 630 #ifndef _SYS_SYSPROTO_H_ 631 struct abort2_args { 632 char *why; 633 int nargs; 634 void **args; 635 }; 636 #endif 637 638 int 639 sys_abort2(struct thread *td, struct abort2_args *uap) 640 { 641 struct proc *p = td->td_proc; 642 struct sbuf *sb; 643 void *uargs[16]; 644 int error, i, sig; 645 646 /* 647 * Do it right now so we can log either proper call of abort2(), or 648 * note, that invalid argument was passed. 512 is big enough to 649 * handle 16 arguments' descriptions with additional comments. 650 */ 651 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 652 sbuf_clear(sb); 653 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 654 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 655 /* 656 * Since we can't return from abort2(), send SIGKILL in cases, where 657 * abort2() was called improperly 658 */ 659 sig = SIGKILL; 660 /* Prevent from DoSes from user-space. */ 661 if (uap->nargs < 0 || uap->nargs > 16) 662 goto out; 663 if (uap->nargs > 0) { 664 if (uap->args == NULL) 665 goto out; 666 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 667 if (error != 0) 668 goto out; 669 } 670 /* 671 * Limit size of 'reason' string to 128. Will fit even when 672 * maximal number of arguments was chosen to be logged. 673 */ 674 if (uap->why != NULL) { 675 error = sbuf_copyin(sb, uap->why, 128); 676 if (error < 0) 677 goto out; 678 } else { 679 sbuf_printf(sb, "(null)"); 680 } 681 if (uap->nargs > 0) { 682 sbuf_printf(sb, "("); 683 for (i = 0;i < uap->nargs; i++) 684 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 685 sbuf_printf(sb, ")"); 686 } 687 /* 688 * Final stage: arguments were proper, string has been 689 * successfully copied from userspace, and copying pointers 690 * from user-space succeed. 691 */ 692 sig = SIGABRT; 693 out: 694 if (sig == SIGKILL) { 695 sbuf_trim(sb); 696 sbuf_printf(sb, " (Reason text inaccessible)"); 697 } 698 sbuf_cat(sb, "\n"); 699 sbuf_finish(sb); 700 log(LOG_INFO, "%s", sbuf_data(sb)); 701 sbuf_delete(sb); 702 exit1(td, W_EXITCODE(0, sig)); 703 return (0); 704 } 705 706 707 #ifdef COMPAT_43 708 /* 709 * The dirty work is handled by kern_wait(). 710 */ 711 int 712 owait(struct thread *td, struct owait_args *uap __unused) 713 { 714 int error, status; 715 716 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 717 if (error == 0) 718 td->td_retval[1] = status; 719 return (error); 720 } 721 #endif /* COMPAT_43 */ 722 723 /* 724 * The dirty work is handled by kern_wait(). 725 */ 726 int 727 sys_wait4(struct thread *td, struct wait4_args *uap) 728 { 729 struct rusage ru, *rup; 730 int error, status; 731 732 if (uap->rusage != NULL) 733 rup = &ru; 734 else 735 rup = NULL; 736 error = kern_wait(td, uap->pid, &status, uap->options, rup); 737 if (uap->status != NULL && error == 0) 738 error = copyout(&status, uap->status, sizeof(status)); 739 if (uap->rusage != NULL && error == 0) 740 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 741 return (error); 742 } 743 744 int 745 sys_wait6(struct thread *td, struct wait6_args *uap) 746 { 747 struct __wrusage wru, *wrup; 748 siginfo_t si, *sip; 749 idtype_t idtype; 750 id_t id; 751 int error, status; 752 753 idtype = uap->idtype; 754 id = uap->id; 755 756 if (uap->wrusage != NULL) 757 wrup = &wru; 758 else 759 wrup = NULL; 760 761 if (uap->info != NULL) { 762 sip = &si; 763 bzero(sip, sizeof(*sip)); 764 } else 765 sip = NULL; 766 767 /* 768 * We expect all callers of wait6() to know about WEXITED and 769 * WTRAPPED. 770 */ 771 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 772 773 if (uap->status != NULL && error == 0) 774 error = copyout(&status, uap->status, sizeof(status)); 775 if (uap->wrusage != NULL && error == 0) 776 error = copyout(&wru, uap->wrusage, sizeof(wru)); 777 if (uap->info != NULL && error == 0) 778 error = copyout(&si, uap->info, sizeof(si)); 779 return (error); 780 } 781 782 /* 783 * Reap the remains of a zombie process and optionally return status and 784 * rusage. Asserts and will release both the proctree_lock and the process 785 * lock as part of its work. 786 */ 787 void 788 proc_reap(struct thread *td, struct proc *p, int *status, int options) 789 { 790 struct proc *q, *t; 791 792 sx_assert(&proctree_lock, SA_XLOCKED); 793 PROC_LOCK_ASSERT(p, MA_OWNED); 794 PROC_SLOCK_ASSERT(p, MA_OWNED); 795 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 796 797 q = td->td_proc; 798 799 PROC_SUNLOCK(p); 800 td->td_retval[0] = p->p_pid; 801 if (status) 802 *status = p->p_xstat; /* convert to int */ 803 if (options & WNOWAIT) { 804 /* 805 * Only poll, returning the status. Caller does not wish to 806 * release the proc struct just yet. 807 */ 808 PROC_UNLOCK(p); 809 sx_xunlock(&proctree_lock); 810 return; 811 } 812 813 PROC_LOCK(q); 814 sigqueue_take(p->p_ksi); 815 PROC_UNLOCK(q); 816 PROC_UNLOCK(p); 817 818 /* 819 * If we got the child via a ptrace 'attach', we need to give it back 820 * to the old parent. 821 */ 822 if (p->p_oppid != 0) { 823 t = proc_realparent(p); 824 PROC_LOCK(t); 825 PROC_LOCK(p); 826 proc_reparent(p, t); 827 p->p_oppid = 0; 828 PROC_UNLOCK(p); 829 pksignal(t, SIGCHLD, p->p_ksi); 830 wakeup(t); 831 cv_broadcast(&p->p_pwait); 832 PROC_UNLOCK(t); 833 sx_xunlock(&proctree_lock); 834 return; 835 } 836 837 /* 838 * Remove other references to this process to ensure we have an 839 * exclusive reference. 840 */ 841 sx_xlock(&allproc_lock); 842 LIST_REMOVE(p, p_list); /* off zombproc */ 843 sx_xunlock(&allproc_lock); 844 LIST_REMOVE(p, p_sibling); 845 PROC_LOCK(p); 846 clear_orphan(p); 847 PROC_UNLOCK(p); 848 leavepgrp(p); 849 if (p->p_procdesc != NULL) 850 procdesc_reap(p); 851 sx_xunlock(&proctree_lock); 852 853 /* 854 * As a side effect of this lock, we know that all other writes to 855 * this proc are visible now, so no more locking is needed for p. 856 */ 857 PROC_LOCK(p); 858 p->p_xstat = 0; /* XXX: why? */ 859 PROC_UNLOCK(p); 860 PROC_LOCK(q); 861 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 862 PROC_UNLOCK(q); 863 864 /* 865 * Decrement the count of procs running with this uid. 866 */ 867 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 868 869 /* 870 * Destroy resource accounting information associated with the process. 871 */ 872 #ifdef RACCT 873 PROC_LOCK(p); 874 racct_sub(p, RACCT_NPROC, 1); 875 PROC_UNLOCK(p); 876 #endif 877 racct_proc_exit(p); 878 879 /* 880 * Free credentials, arguments, and sigacts. 881 */ 882 crfree(p->p_ucred); 883 p->p_ucred = NULL; 884 pargs_drop(p->p_args); 885 p->p_args = NULL; 886 sigacts_free(p->p_sigacts); 887 p->p_sigacts = NULL; 888 889 /* 890 * Do any thread-system specific cleanups. 891 */ 892 thread_wait(p); 893 894 /* 895 * Give vm and machine-dependent layer a chance to free anything that 896 * cpu_exit couldn't release while still running in process context. 897 */ 898 vm_waitproc(p); 899 #ifdef MAC 900 mac_proc_destroy(p); 901 #endif 902 KASSERT(FIRST_THREAD_IN_PROC(p), 903 ("proc_reap: no residual thread!")); 904 uma_zfree(proc_zone, p); 905 sx_xlock(&allproc_lock); 906 nprocs--; 907 sx_xunlock(&allproc_lock); 908 } 909 910 static int 911 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 912 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo) 913 { 914 struct proc *q; 915 struct rusage *rup; 916 917 sx_assert(&proctree_lock, SA_XLOCKED); 918 919 q = td->td_proc; 920 PROC_LOCK(p); 921 922 switch (idtype) { 923 case P_ALL: 924 break; 925 case P_PID: 926 if (p->p_pid != (pid_t)id) { 927 PROC_UNLOCK(p); 928 return (0); 929 } 930 break; 931 case P_PGID: 932 if (p->p_pgid != (pid_t)id) { 933 PROC_UNLOCK(p); 934 return (0); 935 } 936 break; 937 case P_SID: 938 if (p->p_session->s_sid != (pid_t)id) { 939 PROC_UNLOCK(p); 940 return (0); 941 } 942 break; 943 case P_UID: 944 if (p->p_ucred->cr_uid != (uid_t)id) { 945 PROC_UNLOCK(p); 946 return (0); 947 } 948 break; 949 case P_GID: 950 if (p->p_ucred->cr_gid != (gid_t)id) { 951 PROC_UNLOCK(p); 952 return (0); 953 } 954 break; 955 case P_JAILID: 956 if (p->p_ucred->cr_prison->pr_id != (int)id) { 957 PROC_UNLOCK(p); 958 return (0); 959 } 960 break; 961 /* 962 * It seems that the thread structures get zeroed out 963 * at process exit. This makes it impossible to 964 * support P_SETID, P_CID or P_CPUID. 965 */ 966 default: 967 PROC_UNLOCK(p); 968 return (0); 969 } 970 971 if (p_canwait(td, p)) { 972 PROC_UNLOCK(p); 973 return (0); 974 } 975 976 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 977 PROC_UNLOCK(p); 978 return (0); 979 } 980 981 /* 982 * This special case handles a kthread spawned by linux_clone 983 * (see linux_misc.c). The linux_wait4 and linux_waitpid 984 * functions need to be able to distinguish between waiting 985 * on a process and waiting on a thread. It is a thread if 986 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 987 * signifies we want to wait for threads and not processes. 988 */ 989 if ((p->p_sigparent != SIGCHLD) ^ 990 ((options & WLINUXCLONE) != 0)) { 991 PROC_UNLOCK(p); 992 return (0); 993 } 994 995 if (siginfo != NULL) { 996 bzero(siginfo, sizeof(*siginfo)); 997 siginfo->si_errno = 0; 998 999 /* 1000 * SUSv4 requires that the si_signo value is always 1001 * SIGCHLD. Obey it despite the rfork(2) interface 1002 * allows to request other signal for child exit 1003 * notification. 1004 */ 1005 siginfo->si_signo = SIGCHLD; 1006 1007 /* 1008 * This is still a rough estimate. We will fix the 1009 * cases TRAPPED, STOPPED, and CONTINUED later. 1010 */ 1011 if (WCOREDUMP(p->p_xstat)) { 1012 siginfo->si_code = CLD_DUMPED; 1013 siginfo->si_status = WTERMSIG(p->p_xstat); 1014 } else if (WIFSIGNALED(p->p_xstat)) { 1015 siginfo->si_code = CLD_KILLED; 1016 siginfo->si_status = WTERMSIG(p->p_xstat); 1017 } else { 1018 siginfo->si_code = CLD_EXITED; 1019 siginfo->si_status = WEXITSTATUS(p->p_xstat); 1020 } 1021 1022 siginfo->si_pid = p->p_pid; 1023 siginfo->si_uid = p->p_ucred->cr_uid; 1024 1025 /* 1026 * The si_addr field would be useful additional 1027 * detail, but apparently the PC value may be lost 1028 * when we reach this point. bzero() above sets 1029 * siginfo->si_addr to NULL. 1030 */ 1031 } 1032 1033 /* 1034 * There should be no reason to limit resources usage info to 1035 * exited processes only. A snapshot about any resources used 1036 * by a stopped process may be exactly what is needed. 1037 */ 1038 if (wrusage != NULL) { 1039 rup = &wrusage->wru_self; 1040 *rup = p->p_ru; 1041 PROC_STATLOCK(p); 1042 calcru(p, &rup->ru_utime, &rup->ru_stime); 1043 PROC_STATUNLOCK(p); 1044 1045 rup = &wrusage->wru_children; 1046 *rup = p->p_stats->p_cru; 1047 calccru(p, &rup->ru_utime, &rup->ru_stime); 1048 } 1049 1050 if (p->p_state == PRS_ZOMBIE) { 1051 PROC_SLOCK(p); 1052 proc_reap(td, p, status, options); 1053 return (-1); 1054 } 1055 PROC_UNLOCK(p); 1056 return (1); 1057 } 1058 1059 int 1060 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1061 struct rusage *rusage) 1062 { 1063 struct __wrusage wru, *wrup; 1064 idtype_t idtype; 1065 id_t id; 1066 int ret; 1067 1068 /* 1069 * Translate the special pid values into the (idtype, pid) 1070 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1071 * kern_wait6() on its own. 1072 */ 1073 if (pid == WAIT_ANY) { 1074 idtype = P_ALL; 1075 id = 0; 1076 } else if (pid < 0) { 1077 idtype = P_PGID; 1078 id = (id_t)-pid; 1079 } else { 1080 idtype = P_PID; 1081 id = (id_t)pid; 1082 } 1083 1084 if (rusage != NULL) 1085 wrup = &wru; 1086 else 1087 wrup = NULL; 1088 1089 /* 1090 * For backward compatibility we implicitly add flags WEXITED 1091 * and WTRAPPED here. 1092 */ 1093 options |= WEXITED | WTRAPPED; 1094 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1095 if (rusage != NULL) 1096 *rusage = wru.wru_self; 1097 return (ret); 1098 } 1099 1100 int 1101 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1102 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1103 { 1104 struct proc *p, *q; 1105 int error, nfound, ret; 1106 1107 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1108 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1109 AUDIT_ARG_VALUE(options); 1110 1111 q = td->td_proc; 1112 1113 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1114 PROC_LOCK(q); 1115 id = (id_t)q->p_pgid; 1116 PROC_UNLOCK(q); 1117 idtype = P_PGID; 1118 } 1119 1120 /* If we don't know the option, just return. */ 1121 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1122 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1123 return (EINVAL); 1124 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1125 /* 1126 * We will be unable to find any matching processes, 1127 * because there are no known events to look for. 1128 * Prefer to return error instead of blocking 1129 * indefinitely. 1130 */ 1131 return (EINVAL); 1132 } 1133 1134 loop: 1135 if (q->p_flag & P_STATCHILD) { 1136 PROC_LOCK(q); 1137 q->p_flag &= ~P_STATCHILD; 1138 PROC_UNLOCK(q); 1139 } 1140 nfound = 0; 1141 sx_xlock(&proctree_lock); 1142 LIST_FOREACH(p, &q->p_children, p_sibling) { 1143 ret = proc_to_reap(td, p, idtype, id, status, options, 1144 wrusage, siginfo); 1145 if (ret == 0) 1146 continue; 1147 else if (ret == 1) 1148 nfound++; 1149 else 1150 return (0); 1151 1152 PROC_LOCK(p); 1153 PROC_SLOCK(p); 1154 1155 if ((options & WTRAPPED) != 0 && 1156 (p->p_flag & P_TRACED) != 0 && 1157 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1158 (p->p_suspcount == p->p_numthreads) && 1159 ((p->p_flag & P_WAITED) == 0)) { 1160 PROC_SUNLOCK(p); 1161 if ((options & WNOWAIT) == 0) 1162 p->p_flag |= P_WAITED; 1163 sx_xunlock(&proctree_lock); 1164 td->td_retval[0] = p->p_pid; 1165 1166 if (status != NULL) 1167 *status = W_STOPCODE(p->p_xstat); 1168 if (siginfo != NULL) { 1169 siginfo->si_status = p->p_xstat; 1170 siginfo->si_code = CLD_TRAPPED; 1171 } 1172 if ((options & WNOWAIT) == 0) { 1173 PROC_LOCK(q); 1174 sigqueue_take(p->p_ksi); 1175 PROC_UNLOCK(q); 1176 } 1177 1178 PROC_UNLOCK(p); 1179 return (0); 1180 } 1181 if ((options & WUNTRACED) != 0 && 1182 (p->p_flag & P_STOPPED_SIG) != 0 && 1183 (p->p_suspcount == p->p_numthreads) && 1184 ((p->p_flag & P_WAITED) == 0)) { 1185 PROC_SUNLOCK(p); 1186 if ((options & WNOWAIT) == 0) 1187 p->p_flag |= P_WAITED; 1188 sx_xunlock(&proctree_lock); 1189 td->td_retval[0] = p->p_pid; 1190 1191 if (status != NULL) 1192 *status = W_STOPCODE(p->p_xstat); 1193 if (siginfo != NULL) { 1194 siginfo->si_status = p->p_xstat; 1195 siginfo->si_code = CLD_STOPPED; 1196 } 1197 if ((options & WNOWAIT) == 0) { 1198 PROC_LOCK(q); 1199 sigqueue_take(p->p_ksi); 1200 PROC_UNLOCK(q); 1201 } 1202 1203 PROC_UNLOCK(p); 1204 return (0); 1205 } 1206 PROC_SUNLOCK(p); 1207 if ((options & WCONTINUED) != 0 && 1208 (p->p_flag & P_CONTINUED) != 0) { 1209 sx_xunlock(&proctree_lock); 1210 td->td_retval[0] = p->p_pid; 1211 if ((options & WNOWAIT) == 0) { 1212 p->p_flag &= ~P_CONTINUED; 1213 PROC_LOCK(q); 1214 sigqueue_take(p->p_ksi); 1215 PROC_UNLOCK(q); 1216 } 1217 PROC_UNLOCK(p); 1218 1219 if (status != NULL) 1220 *status = SIGCONT; 1221 if (siginfo != NULL) { 1222 siginfo->si_status = SIGCONT; 1223 siginfo->si_code = CLD_CONTINUED; 1224 } 1225 return (0); 1226 } 1227 PROC_UNLOCK(p); 1228 } 1229 1230 /* 1231 * Look in the orphans list too, to allow the parent to 1232 * collect it's child exit status even if child is being 1233 * debugged. 1234 * 1235 * Debugger detaches from the parent upon successful 1236 * switch-over from parent to child. At this point due to 1237 * re-parenting the parent loses the child to debugger and a 1238 * wait4(2) call would report that it has no children to wait 1239 * for. By maintaining a list of orphans we allow the parent 1240 * to successfully wait until the child becomes a zombie. 1241 */ 1242 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1243 ret = proc_to_reap(td, p, idtype, id, status, options, 1244 wrusage, siginfo); 1245 if (ret == 0) 1246 continue; 1247 else if (ret == 1) 1248 nfound++; 1249 else 1250 return (0); 1251 } 1252 if (nfound == 0) { 1253 sx_xunlock(&proctree_lock); 1254 return (ECHILD); 1255 } 1256 if (options & WNOHANG) { 1257 sx_xunlock(&proctree_lock); 1258 td->td_retval[0] = 0; 1259 return (0); 1260 } 1261 PROC_LOCK(q); 1262 sx_xunlock(&proctree_lock); 1263 if (q->p_flag & P_STATCHILD) { 1264 q->p_flag &= ~P_STATCHILD; 1265 error = 0; 1266 } else 1267 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1268 PROC_UNLOCK(q); 1269 if (error) 1270 return (error); 1271 goto loop; 1272 } 1273 1274 /* 1275 * Make process 'parent' the new parent of process 'child'. 1276 * Must be called with an exclusive hold of proctree lock. 1277 */ 1278 void 1279 proc_reparent(struct proc *child, struct proc *parent) 1280 { 1281 1282 sx_assert(&proctree_lock, SX_XLOCKED); 1283 PROC_LOCK_ASSERT(child, MA_OWNED); 1284 if (child->p_pptr == parent) 1285 return; 1286 1287 PROC_LOCK(child->p_pptr); 1288 sigqueue_take(child->p_ksi); 1289 PROC_UNLOCK(child->p_pptr); 1290 LIST_REMOVE(child, p_sibling); 1291 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1292 1293 clear_orphan(child); 1294 if (child->p_flag & P_TRACED) { 1295 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1296 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1297 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1298 p_orphan); 1299 } else { 1300 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1301 child, p_orphan); 1302 } 1303 child->p_treeflag |= P_TREE_ORPHANED; 1304 } 1305 1306 child->p_pptr = parent; 1307 } 1308