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 #include <sys/umtx.h> 75 #ifdef KTRACE 76 #include <sys/ktrace.h> 77 #endif 78 79 #include <security/audit/audit.h> 80 #include <security/mac/mac_framework.h> 81 82 #include <vm/vm.h> 83 #include <vm/vm_extern.h> 84 #include <vm/vm_param.h> 85 #include <vm/pmap.h> 86 #include <vm/vm_map.h> 87 #include <vm/vm_page.h> 88 #include <vm/uma.h> 89 #include <vm/vm_domain.h> 90 91 #ifdef KDTRACE_HOOKS 92 #include <sys/dtrace_bsd.h> 93 dtrace_execexit_func_t dtrace_fasttrap_exit; 94 #endif 95 96 SDT_PROVIDER_DECLARE(proc); 97 SDT_PROBE_DEFINE1(proc, , , exit, "int"); 98 99 /* Hook for NFS teardown procedure. */ 100 void (*nlminfo_release_p)(struct proc *p); 101 102 struct proc * 103 proc_realparent(struct proc *child) 104 { 105 struct proc *p, *parent; 106 107 sx_assert(&proctree_lock, SX_LOCKED); 108 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 109 if (child->p_oppid == 0 || 110 child->p_pptr->p_pid == child->p_oppid) 111 parent = child->p_pptr; 112 else 113 parent = initproc; 114 return (parent); 115 } 116 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 117 /* Cannot use LIST_PREV(), since the list head is not known. */ 118 p = __containerof(p->p_orphan.le_prev, struct proc, 119 p_orphan.le_next); 120 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 121 ("missing P_ORPHAN %p", p)); 122 } 123 parent = __containerof(p->p_orphan.le_prev, struct proc, 124 p_orphans.lh_first); 125 return (parent); 126 } 127 128 void 129 reaper_abandon_children(struct proc *p, bool exiting) 130 { 131 struct proc *p1, *p2, *ptmp; 132 133 sx_assert(&proctree_lock, SX_LOCKED); 134 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 135 if ((p->p_treeflag & P_TREE_REAPER) == 0) 136 return; 137 p1 = p->p_reaper; 138 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 139 LIST_REMOVE(p2, p_reapsibling); 140 p2->p_reaper = p1; 141 p2->p_reapsubtree = p->p_reapsubtree; 142 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 143 if (exiting && p2->p_pptr == p) { 144 PROC_LOCK(p2); 145 proc_reparent(p2, p1); 146 PROC_UNLOCK(p2); 147 } 148 } 149 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 150 p->p_treeflag &= ~P_TREE_REAPER; 151 } 152 153 static void 154 clear_orphan(struct proc *p) 155 { 156 struct proc *p1; 157 158 sx_assert(&proctree_lock, SA_XLOCKED); 159 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 160 return; 161 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 162 p1 = LIST_NEXT(p, p_orphan); 163 if (p1 != NULL) 164 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 165 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 166 } 167 LIST_REMOVE(p, p_orphan); 168 p->p_treeflag &= ~P_TREE_ORPHANED; 169 } 170 171 /* 172 * exit -- death of process. 173 */ 174 void 175 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 176 { 177 178 exit1(td, uap->rval, 0); 179 /* NOTREACHED */ 180 } 181 182 /* 183 * Exit: deallocate address space and other resources, change proc state to 184 * zombie, and unlink proc from allproc and parent's lists. Save exit status 185 * and rusage for wait(). Check for child processes and orphan them. 186 */ 187 void 188 exit1(struct thread *td, int rval, int signo) 189 { 190 struct proc *p, *nq, *q, *t; 191 struct thread *tdt; 192 193 mtx_assert(&Giant, MA_NOTOWNED); 194 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 195 196 p = td->td_proc; 197 /* 198 * XXX in case we're rebooting we just let init die in order to 199 * work around an unsolved stack overflow seen very late during 200 * shutdown on sparc64 when the gmirror worker process exists. 201 */ 202 if (p == initproc && rebooting == 0) { 203 printf("init died (signal %d, exit %d)\n", signo, rval); 204 panic("Going nowhere without my init!"); 205 } 206 207 /* 208 * Deref SU mp, since the thread does not return to userspace. 209 */ 210 if (softdep_ast_cleanup != NULL) 211 softdep_ast_cleanup(); 212 213 /* 214 * MUST abort all other threads before proceeding past here. 215 */ 216 PROC_LOCK(p); 217 /* 218 * First check if some other thread or external request got 219 * here before us. If so, act appropriately: exit or suspend. 220 * We must ensure that stop requests are handled before we set 221 * P_WEXIT. 222 */ 223 thread_suspend_check(0); 224 while (p->p_flag & P_HADTHREADS) { 225 /* 226 * Kill off the other threads. This requires 227 * some co-operation from other parts of the kernel 228 * so it may not be instantaneous. With this state set 229 * any thread entering the kernel from userspace will 230 * thread_exit() in trap(). Any thread attempting to 231 * sleep will return immediately with EINTR or EWOULDBLOCK 232 * which will hopefully force them to back out to userland 233 * freeing resources as they go. Any thread attempting 234 * to return to userland will thread_exit() from userret(). 235 * thread_exit() will unsuspend us when the last of the 236 * other threads exits. 237 * If there is already a thread singler after resumption, 238 * calling thread_single will fail; in that case, we just 239 * re-check all suspension request, the thread should 240 * either be suspended there or exit. 241 */ 242 if (!thread_single(p, SINGLE_EXIT)) 243 /* 244 * All other activity in this process is now 245 * stopped. Threading support has been turned 246 * off. 247 */ 248 break; 249 /* 250 * Recheck for new stop or suspend requests which 251 * might appear while process lock was dropped in 252 * thread_single(). 253 */ 254 thread_suspend_check(0); 255 } 256 KASSERT(p->p_numthreads == 1, 257 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 258 racct_sub(p, RACCT_NTHR, 1); 259 260 /* Let event handler change exit status */ 261 p->p_xexit = rval; 262 p->p_xsig = signo; 263 264 /* 265 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 266 * on our vmspace, so we should block below until they have 267 * released their reference to us. Note that if they have 268 * requested S_EXIT stops we will block here until they ack 269 * via PIOCCONT. 270 */ 271 _STOPEVENT(p, S_EXIT, 0); 272 273 /* 274 * Ignore any pending request to stop due to a stop signal. 275 * Once P_WEXIT is set, future requests will be ignored as 276 * well. 277 */ 278 p->p_flag &= ~P_STOPPED_SIG; 279 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 280 281 /* 282 * Note that we are exiting and do another wakeup of anyone in 283 * PIOCWAIT in case they aren't listening for S_EXIT stops or 284 * decided to wait again after we told them we are exiting. 285 */ 286 p->p_flag |= P_WEXIT; 287 wakeup(&p->p_stype); 288 289 /* 290 * Wait for any processes that have a hold on our vmspace to 291 * release their reference. 292 */ 293 while (p->p_lock > 0) 294 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 295 296 PROC_UNLOCK(p); 297 /* Drain the limit callout while we don't have the proc locked */ 298 callout_drain(&p->p_limco); 299 300 #ifdef AUDIT 301 /* 302 * The Sun BSM exit token contains two components: an exit status as 303 * passed to exit(), and a return value to indicate what sort of exit 304 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 305 * what the return value is. 306 */ 307 AUDIT_ARG_EXIT(rval, 0); 308 AUDIT_SYSCALL_EXIT(0, td); 309 #endif 310 311 /* Are we a task leader with peers? */ 312 if (p->p_peers != NULL && p == p->p_leader) { 313 mtx_lock(&ppeers_lock); 314 q = p->p_peers; 315 while (q != NULL) { 316 PROC_LOCK(q); 317 kern_psignal(q, SIGKILL); 318 PROC_UNLOCK(q); 319 q = q->p_peers; 320 } 321 while (p->p_peers != NULL) 322 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 323 mtx_unlock(&ppeers_lock); 324 } 325 326 /* 327 * Check if any loadable modules need anything done at process exit. 328 * E.g. SYSV IPC stuff. 329 * Event handler could change exit status. 330 * XXX what if one of these generates an error? 331 */ 332 EVENTHANDLER_INVOKE(process_exit, p); 333 334 /* 335 * If parent is waiting for us to exit or exec, 336 * P_PPWAIT is set; we will wakeup the parent below. 337 */ 338 PROC_LOCK(p); 339 stopprofclock(p); 340 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 341 342 /* 343 * Stop the real interval timer. If the handler is currently 344 * executing, prevent it from rearming itself and let it finish. 345 */ 346 if (timevalisset(&p->p_realtimer.it_value) && 347 callout_stop(&p->p_itcallout) == 0) { 348 timevalclear(&p->p_realtimer.it_interval); 349 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 350 KASSERT(!timevalisset(&p->p_realtimer.it_value), 351 ("realtime timer is still armed")); 352 } 353 354 PROC_UNLOCK(p); 355 356 umtx_thread_exit(td); 357 358 /* 359 * Reset any sigio structures pointing to us as a result of 360 * F_SETOWN with our pid. 361 */ 362 funsetownlst(&p->p_sigiolst); 363 364 /* 365 * If this process has an nlminfo data area (for lockd), release it 366 */ 367 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 368 (*nlminfo_release_p)(p); 369 370 /* 371 * Close open files and release open-file table. 372 * This may block! 373 */ 374 fdescfree(td); 375 376 /* 377 * If this thread tickled GEOM, we need to wait for the giggling to 378 * stop before we return to userland 379 */ 380 if (td->td_pflags & TDP_GEOM) 381 g_waitidle(); 382 383 /* 384 * Remove ourself from our leader's peer list and wake our leader. 385 */ 386 if (p->p_leader->p_peers != NULL) { 387 mtx_lock(&ppeers_lock); 388 if (p->p_leader->p_peers != NULL) { 389 q = p->p_leader; 390 while (q->p_peers != p) 391 q = q->p_peers; 392 q->p_peers = p->p_peers; 393 wakeup(p->p_leader); 394 } 395 mtx_unlock(&ppeers_lock); 396 } 397 398 vmspace_exit(td); 399 killjobc(); 400 (void)acct_process(td); 401 402 #ifdef KTRACE 403 ktrprocexit(td); 404 #endif 405 /* 406 * Release reference to text vnode 407 */ 408 if (p->p_textvp != NULL) { 409 vrele(p->p_textvp); 410 p->p_textvp = NULL; 411 } 412 413 /* 414 * Release our limits structure. 415 */ 416 lim_free(p->p_limit); 417 p->p_limit = NULL; 418 419 tidhash_remove(td); 420 421 /* 422 * Remove proc from allproc queue and pidhash chain. 423 * Place onto zombproc. Unlink from parent's child list. 424 */ 425 sx_xlock(&allproc_lock); 426 LIST_REMOVE(p, p_list); 427 LIST_INSERT_HEAD(&zombproc, p, p_list); 428 LIST_REMOVE(p, p_hash); 429 sx_xunlock(&allproc_lock); 430 431 /* 432 * Call machine-dependent code to release any 433 * machine-dependent resources other than the address space. 434 * The address space is released by "vmspace_exitfree(p)" in 435 * vm_waitproc(). 436 */ 437 cpu_exit(td); 438 439 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 440 441 /* 442 * Reparent all children processes: 443 * - traced ones to the original parent (or init if we are that parent) 444 * - the rest to init 445 */ 446 sx_xlock(&proctree_lock); 447 q = LIST_FIRST(&p->p_children); 448 if (q != NULL) /* only need this if any child is S_ZOMB */ 449 wakeup(q->p_reaper); 450 for (; q != NULL; q = nq) { 451 nq = LIST_NEXT(q, p_sibling); 452 PROC_LOCK(q); 453 q->p_sigparent = SIGCHLD; 454 455 if (!(q->p_flag & P_TRACED)) { 456 proc_reparent(q, q->p_reaper); 457 } else { 458 /* 459 * Traced processes are killed since their existence 460 * means someone is screwing up. 461 */ 462 t = proc_realparent(q); 463 if (t == p) { 464 proc_reparent(q, q->p_reaper); 465 } else { 466 PROC_LOCK(t); 467 proc_reparent(q, t); 468 PROC_UNLOCK(t); 469 } 470 /* 471 * Since q was found on our children list, the 472 * proc_reparent() call moved q to the orphan 473 * list due to present P_TRACED flag. Clear 474 * orphan link for q now while q is locked. 475 */ 476 clear_orphan(q); 477 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 478 FOREACH_THREAD_IN_PROC(q, tdt) 479 tdt->td_dbgflags &= ~TDB_SUSPEND; 480 kern_psignal(q, SIGKILL); 481 } 482 PROC_UNLOCK(q); 483 } 484 485 /* 486 * Also get rid of our orphans. 487 */ 488 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 489 PROC_LOCK(q); 490 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 491 q->p_pid); 492 clear_orphan(q); 493 PROC_UNLOCK(q); 494 } 495 496 /* Save exit status. */ 497 PROC_LOCK(p); 498 p->p_xthread = td; 499 500 /* Tell the prison that we are gone. */ 501 prison_proc_free(p->p_ucred->cr_prison); 502 503 #ifdef KDTRACE_HOOKS 504 /* 505 * Tell the DTrace fasttrap provider about the exit if it 506 * has declared an interest. 507 */ 508 if (dtrace_fasttrap_exit) 509 dtrace_fasttrap_exit(p); 510 #endif 511 512 /* 513 * Notify interested parties of our demise. 514 */ 515 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 516 517 #ifdef KDTRACE_HOOKS 518 int reason = CLD_EXITED; 519 if (WCOREDUMP(signo)) 520 reason = CLD_DUMPED; 521 else if (WIFSIGNALED(signo)) 522 reason = CLD_KILLED; 523 SDT_PROBE1(proc, , , exit, reason); 524 #endif 525 526 /* 527 * Just delete all entries in the p_klist. At this point we won't 528 * report any more events, and there are nasty race conditions that 529 * can beat us if we don't. 530 */ 531 knlist_clear(&p->p_klist, 1); 532 533 /* 534 * If this is a process with a descriptor, we may not need to deliver 535 * a signal to the parent. proctree_lock is held over 536 * procdesc_exit() to serialize concurrent calls to close() and 537 * exit(). 538 */ 539 if (p->p_procdesc == NULL || procdesc_exit(p)) { 540 /* 541 * Notify parent that we're gone. If parent has the 542 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 543 * notify process 1 instead (and hope it will handle this 544 * situation). 545 */ 546 PROC_LOCK(p->p_pptr); 547 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 548 if (p->p_pptr->p_sigacts->ps_flag & 549 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 550 struct proc *pp; 551 552 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 553 pp = p->p_pptr; 554 PROC_UNLOCK(pp); 555 proc_reparent(p, p->p_reaper); 556 p->p_sigparent = SIGCHLD; 557 PROC_LOCK(p->p_pptr); 558 559 /* 560 * Notify parent, so in case he was wait(2)ing or 561 * executing waitpid(2) with our pid, he will 562 * continue. 563 */ 564 wakeup(pp); 565 } else 566 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 567 568 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 569 childproc_exited(p); 570 else if (p->p_sigparent != 0) { 571 if (p->p_sigparent == SIGCHLD) 572 childproc_exited(p); 573 else /* LINUX thread */ 574 kern_psignal(p->p_pptr, p->p_sigparent); 575 } 576 } else 577 PROC_LOCK(p->p_pptr); 578 sx_xunlock(&proctree_lock); 579 580 /* 581 * The state PRS_ZOMBIE prevents other proesses from sending 582 * signal to the process, to avoid memory leak, we free memory 583 * for signal queue at the time when the state is set. 584 */ 585 sigqueue_flush(&p->p_sigqueue); 586 sigqueue_flush(&td->td_sigqueue); 587 588 /* 589 * We have to wait until after acquiring all locks before 590 * changing p_state. We need to avoid all possible context 591 * switches (including ones from blocking on a mutex) while 592 * marked as a zombie. We also have to set the zombie state 593 * before we release the parent process' proc lock to avoid 594 * a lost wakeup. So, we first call wakeup, then we grab the 595 * sched lock, update the state, and release the parent process' 596 * proc lock. 597 */ 598 wakeup(p->p_pptr); 599 cv_broadcast(&p->p_pwait); 600 sched_exit(p->p_pptr, td); 601 PROC_SLOCK(p); 602 p->p_state = PRS_ZOMBIE; 603 PROC_UNLOCK(p->p_pptr); 604 605 /* 606 * Hopefully no one will try to deliver a signal to the process this 607 * late in the game. 608 */ 609 knlist_destroy(&p->p_klist); 610 611 /* 612 * Save our children's rusage information in our exit rusage. 613 */ 614 PROC_STATLOCK(p); 615 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 616 PROC_STATUNLOCK(p); 617 618 /* 619 * Make sure the scheduler takes this thread out of its tables etc. 620 * This will also release this thread's reference to the ucred. 621 * Other thread parts to release include pcb bits and such. 622 */ 623 thread_exit(); 624 } 625 626 627 #ifndef _SYS_SYSPROTO_H_ 628 struct abort2_args { 629 char *why; 630 int nargs; 631 void **args; 632 }; 633 #endif 634 635 int 636 sys_abort2(struct thread *td, struct abort2_args *uap) 637 { 638 struct proc *p = td->td_proc; 639 struct sbuf *sb; 640 void *uargs[16]; 641 int error, i, sig; 642 643 /* 644 * Do it right now so we can log either proper call of abort2(), or 645 * note, that invalid argument was passed. 512 is big enough to 646 * handle 16 arguments' descriptions with additional comments. 647 */ 648 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 649 sbuf_clear(sb); 650 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 651 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 652 /* 653 * Since we can't return from abort2(), send SIGKILL in cases, where 654 * abort2() was called improperly 655 */ 656 sig = SIGKILL; 657 /* Prevent from DoSes from user-space. */ 658 if (uap->nargs < 0 || uap->nargs > 16) 659 goto out; 660 if (uap->nargs > 0) { 661 if (uap->args == NULL) 662 goto out; 663 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 664 if (error != 0) 665 goto out; 666 } 667 /* 668 * Limit size of 'reason' string to 128. Will fit even when 669 * maximal number of arguments was chosen to be logged. 670 */ 671 if (uap->why != NULL) { 672 error = sbuf_copyin(sb, uap->why, 128); 673 if (error < 0) 674 goto out; 675 } else { 676 sbuf_printf(sb, "(null)"); 677 } 678 if (uap->nargs > 0) { 679 sbuf_printf(sb, "("); 680 for (i = 0;i < uap->nargs; i++) 681 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 682 sbuf_printf(sb, ")"); 683 } 684 /* 685 * Final stage: arguments were proper, string has been 686 * successfully copied from userspace, and copying pointers 687 * from user-space succeed. 688 */ 689 sig = SIGABRT; 690 out: 691 if (sig == SIGKILL) { 692 sbuf_trim(sb); 693 sbuf_printf(sb, " (Reason text inaccessible)"); 694 } 695 sbuf_cat(sb, "\n"); 696 sbuf_finish(sb); 697 log(LOG_INFO, "%s", sbuf_data(sb)); 698 sbuf_delete(sb); 699 exit1(td, 0, sig); 700 return (0); 701 } 702 703 704 #ifdef COMPAT_43 705 /* 706 * The dirty work is handled by kern_wait(). 707 */ 708 int 709 owait(struct thread *td, struct owait_args *uap __unused) 710 { 711 int error, status; 712 713 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 714 if (error == 0) 715 td->td_retval[1] = status; 716 return (error); 717 } 718 #endif /* COMPAT_43 */ 719 720 /* 721 * The dirty work is handled by kern_wait(). 722 */ 723 int 724 sys_wait4(struct thread *td, struct wait4_args *uap) 725 { 726 struct rusage ru, *rup; 727 int error, status; 728 729 if (uap->rusage != NULL) 730 rup = &ru; 731 else 732 rup = NULL; 733 error = kern_wait(td, uap->pid, &status, uap->options, rup); 734 if (uap->status != NULL && error == 0) 735 error = copyout(&status, uap->status, sizeof(status)); 736 if (uap->rusage != NULL && error == 0) 737 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 738 return (error); 739 } 740 741 int 742 sys_wait6(struct thread *td, struct wait6_args *uap) 743 { 744 struct __wrusage wru, *wrup; 745 siginfo_t si, *sip; 746 idtype_t idtype; 747 id_t id; 748 int error, status; 749 750 idtype = uap->idtype; 751 id = uap->id; 752 753 if (uap->wrusage != NULL) 754 wrup = &wru; 755 else 756 wrup = NULL; 757 758 if (uap->info != NULL) { 759 sip = &si; 760 bzero(sip, sizeof(*sip)); 761 } else 762 sip = NULL; 763 764 /* 765 * We expect all callers of wait6() to know about WEXITED and 766 * WTRAPPED. 767 */ 768 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 769 770 if (uap->status != NULL && error == 0) 771 error = copyout(&status, uap->status, sizeof(status)); 772 if (uap->wrusage != NULL && error == 0) 773 error = copyout(&wru, uap->wrusage, sizeof(wru)); 774 if (uap->info != NULL && error == 0) 775 error = copyout(&si, uap->info, sizeof(si)); 776 return (error); 777 } 778 779 /* 780 * Reap the remains of a zombie process and optionally return status and 781 * rusage. Asserts and will release both the proctree_lock and the process 782 * lock as part of its work. 783 */ 784 void 785 proc_reap(struct thread *td, struct proc *p, int *status, int options) 786 { 787 struct proc *q, *t; 788 789 sx_assert(&proctree_lock, SA_XLOCKED); 790 PROC_LOCK_ASSERT(p, MA_OWNED); 791 PROC_SLOCK_ASSERT(p, MA_OWNED); 792 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 793 794 q = td->td_proc; 795 796 PROC_SUNLOCK(p); 797 if (status) 798 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 799 if (options & WNOWAIT) { 800 /* 801 * Only poll, returning the status. Caller does not wish to 802 * release the proc struct just yet. 803 */ 804 PROC_UNLOCK(p); 805 sx_xunlock(&proctree_lock); 806 return; 807 } 808 809 PROC_LOCK(q); 810 sigqueue_take(p->p_ksi); 811 PROC_UNLOCK(q); 812 813 /* 814 * If we got the child via a ptrace 'attach', we need to give it back 815 * to the old parent. 816 */ 817 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 818 PROC_UNLOCK(p); 819 t = proc_realparent(p); 820 PROC_LOCK(t); 821 PROC_LOCK(p); 822 CTR2(KTR_PTRACE, 823 "wait: traced child %d moved back to parent %d", p->p_pid, 824 t->p_pid); 825 proc_reparent(p, t); 826 p->p_oppid = 0; 827 PROC_UNLOCK(p); 828 pksignal(t, SIGCHLD, p->p_ksi); 829 wakeup(t); 830 cv_broadcast(&p->p_pwait); 831 PROC_UNLOCK(t); 832 sx_xunlock(&proctree_lock); 833 return; 834 } 835 p->p_oppid = 0; 836 PROC_UNLOCK(p); 837 838 /* 839 * Remove other references to this process to ensure we have an 840 * exclusive reference. 841 */ 842 sx_xlock(&allproc_lock); 843 LIST_REMOVE(p, p_list); /* off zombproc */ 844 sx_xunlock(&allproc_lock); 845 LIST_REMOVE(p, p_sibling); 846 reaper_abandon_children(p, true); 847 LIST_REMOVE(p, p_reapsibling); 848 PROC_LOCK(p); 849 clear_orphan(p); 850 PROC_UNLOCK(p); 851 leavepgrp(p); 852 if (p->p_procdesc != NULL) 853 procdesc_reap(p); 854 sx_xunlock(&proctree_lock); 855 856 /* 857 * Removal from allproc list and process group list paired with 858 * PROC_LOCK which was executed during that time should guarantee 859 * nothing can reach this process anymore. As such further locking 860 * is unnecessary. 861 */ 862 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 863 864 PROC_LOCK(q); 865 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 866 PROC_UNLOCK(q); 867 868 /* 869 * Decrement the count of procs running with this uid. 870 */ 871 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 872 873 /* 874 * Destroy resource accounting information associated with the process. 875 */ 876 #ifdef RACCT 877 if (racct_enable) { 878 PROC_LOCK(p); 879 racct_sub(p, RACCT_NPROC, 1); 880 PROC_UNLOCK(p); 881 } 882 #endif 883 racct_proc_exit(p); 884 885 /* 886 * Free credentials, arguments, and sigacts. 887 */ 888 crfree(p->p_ucred); 889 proc_set_cred(p, NULL); 890 pargs_drop(p->p_args); 891 p->p_args = NULL; 892 sigacts_free(p->p_sigacts); 893 p->p_sigacts = NULL; 894 895 /* 896 * Do any thread-system specific cleanups. 897 */ 898 thread_wait(p); 899 900 /* 901 * Give vm and machine-dependent layer a chance to free anything that 902 * cpu_exit couldn't release while still running in process context. 903 */ 904 vm_waitproc(p); 905 #ifdef MAC 906 mac_proc_destroy(p); 907 #endif 908 /* 909 * Free any domain policy that's still hiding around. 910 */ 911 vm_domain_policy_cleanup(&p->p_vm_dom_policy); 912 913 KASSERT(FIRST_THREAD_IN_PROC(p), 914 ("proc_reap: no residual thread!")); 915 uma_zfree(proc_zone, p); 916 atomic_add_int(&nprocs, -1); 917 } 918 919 static int 920 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 921 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 922 int check_only) 923 { 924 struct rusage *rup; 925 926 sx_assert(&proctree_lock, SA_XLOCKED); 927 928 PROC_LOCK(p); 929 930 switch (idtype) { 931 case P_ALL: 932 if (p->p_procdesc != NULL) { 933 PROC_UNLOCK(p); 934 return (0); 935 } 936 break; 937 case P_PID: 938 if (p->p_pid != (pid_t)id) { 939 PROC_UNLOCK(p); 940 return (0); 941 } 942 break; 943 case P_PGID: 944 if (p->p_pgid != (pid_t)id) { 945 PROC_UNLOCK(p); 946 return (0); 947 } 948 break; 949 case P_SID: 950 if (p->p_session->s_sid != (pid_t)id) { 951 PROC_UNLOCK(p); 952 return (0); 953 } 954 break; 955 case P_UID: 956 if (p->p_ucred->cr_uid != (uid_t)id) { 957 PROC_UNLOCK(p); 958 return (0); 959 } 960 break; 961 case P_GID: 962 if (p->p_ucred->cr_gid != (gid_t)id) { 963 PROC_UNLOCK(p); 964 return (0); 965 } 966 break; 967 case P_JAILID: 968 if (p->p_ucred->cr_prison->pr_id != (int)id) { 969 PROC_UNLOCK(p); 970 return (0); 971 } 972 break; 973 /* 974 * It seems that the thread structures get zeroed out 975 * at process exit. This makes it impossible to 976 * support P_SETID, P_CID or P_CPUID. 977 */ 978 default: 979 PROC_UNLOCK(p); 980 return (0); 981 } 982 983 if (p_canwait(td, p)) { 984 PROC_UNLOCK(p); 985 return (0); 986 } 987 988 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 989 PROC_UNLOCK(p); 990 return (0); 991 } 992 993 /* 994 * This special case handles a kthread spawned by linux_clone 995 * (see linux_misc.c). The linux_wait4 and linux_waitpid 996 * functions need to be able to distinguish between waiting 997 * on a process and waiting on a thread. It is a thread if 998 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 999 * signifies we want to wait for threads and not processes. 1000 */ 1001 if ((p->p_sigparent != SIGCHLD) ^ 1002 ((options & WLINUXCLONE) != 0)) { 1003 PROC_UNLOCK(p); 1004 return (0); 1005 } 1006 1007 if (siginfo != NULL) { 1008 bzero(siginfo, sizeof(*siginfo)); 1009 siginfo->si_errno = 0; 1010 1011 /* 1012 * SUSv4 requires that the si_signo value is always 1013 * SIGCHLD. Obey it despite the rfork(2) interface 1014 * allows to request other signal for child exit 1015 * notification. 1016 */ 1017 siginfo->si_signo = SIGCHLD; 1018 1019 /* 1020 * This is still a rough estimate. We will fix the 1021 * cases TRAPPED, STOPPED, and CONTINUED later. 1022 */ 1023 if (WCOREDUMP(p->p_xsig)) { 1024 siginfo->si_code = CLD_DUMPED; 1025 siginfo->si_status = WTERMSIG(p->p_xsig); 1026 } else if (WIFSIGNALED(p->p_xsig)) { 1027 siginfo->si_code = CLD_KILLED; 1028 siginfo->si_status = WTERMSIG(p->p_xsig); 1029 } else { 1030 siginfo->si_code = CLD_EXITED; 1031 siginfo->si_status = p->p_xexit; 1032 } 1033 1034 siginfo->si_pid = p->p_pid; 1035 siginfo->si_uid = p->p_ucred->cr_uid; 1036 1037 /* 1038 * The si_addr field would be useful additional 1039 * detail, but apparently the PC value may be lost 1040 * when we reach this point. bzero() above sets 1041 * siginfo->si_addr to NULL. 1042 */ 1043 } 1044 1045 /* 1046 * There should be no reason to limit resources usage info to 1047 * exited processes only. A snapshot about any resources used 1048 * by a stopped process may be exactly what is needed. 1049 */ 1050 if (wrusage != NULL) { 1051 rup = &wrusage->wru_self; 1052 *rup = p->p_ru; 1053 PROC_STATLOCK(p); 1054 calcru(p, &rup->ru_utime, &rup->ru_stime); 1055 PROC_STATUNLOCK(p); 1056 1057 rup = &wrusage->wru_children; 1058 *rup = p->p_stats->p_cru; 1059 calccru(p, &rup->ru_utime, &rup->ru_stime); 1060 } 1061 1062 if (p->p_state == PRS_ZOMBIE && !check_only) { 1063 PROC_SLOCK(p); 1064 proc_reap(td, p, status, options); 1065 return (-1); 1066 } 1067 PROC_UNLOCK(p); 1068 return (1); 1069 } 1070 1071 int 1072 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1073 struct rusage *rusage) 1074 { 1075 struct __wrusage wru, *wrup; 1076 idtype_t idtype; 1077 id_t id; 1078 int ret; 1079 1080 /* 1081 * Translate the special pid values into the (idtype, pid) 1082 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1083 * kern_wait6() on its own. 1084 */ 1085 if (pid == WAIT_ANY) { 1086 idtype = P_ALL; 1087 id = 0; 1088 } else if (pid < 0) { 1089 idtype = P_PGID; 1090 id = (id_t)-pid; 1091 } else { 1092 idtype = P_PID; 1093 id = (id_t)pid; 1094 } 1095 1096 if (rusage != NULL) 1097 wrup = &wru; 1098 else 1099 wrup = NULL; 1100 1101 /* 1102 * For backward compatibility we implicitly add flags WEXITED 1103 * and WTRAPPED here. 1104 */ 1105 options |= WEXITED | WTRAPPED; 1106 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1107 if (rusage != NULL) 1108 *rusage = wru.wru_self; 1109 return (ret); 1110 } 1111 1112 int 1113 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1114 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1115 { 1116 struct proc *p, *q; 1117 pid_t pid; 1118 int error, nfound, ret; 1119 1120 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1121 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1122 AUDIT_ARG_VALUE(options); 1123 1124 q = td->td_proc; 1125 1126 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1127 PROC_LOCK(q); 1128 id = (id_t)q->p_pgid; 1129 PROC_UNLOCK(q); 1130 idtype = P_PGID; 1131 } 1132 1133 /* If we don't know the option, just return. */ 1134 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1135 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1136 return (EINVAL); 1137 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1138 /* 1139 * We will be unable to find any matching processes, 1140 * because there are no known events to look for. 1141 * Prefer to return error instead of blocking 1142 * indefinitely. 1143 */ 1144 return (EINVAL); 1145 } 1146 1147 loop: 1148 if (q->p_flag & P_STATCHILD) { 1149 PROC_LOCK(q); 1150 q->p_flag &= ~P_STATCHILD; 1151 PROC_UNLOCK(q); 1152 } 1153 nfound = 0; 1154 sx_xlock(&proctree_lock); 1155 LIST_FOREACH(p, &q->p_children, p_sibling) { 1156 pid = p->p_pid; 1157 ret = proc_to_reap(td, p, idtype, id, status, options, 1158 wrusage, siginfo, 0); 1159 if (ret == 0) 1160 continue; 1161 else if (ret == 1) 1162 nfound++; 1163 else { 1164 td->td_retval[0] = pid; 1165 return (0); 1166 } 1167 1168 PROC_LOCK(p); 1169 PROC_SLOCK(p); 1170 1171 if ((options & WTRAPPED) != 0 && 1172 (p->p_flag & P_TRACED) != 0 && 1173 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1174 (p->p_suspcount == p->p_numthreads) && 1175 ((p->p_flag & P_WAITED) == 0)) { 1176 PROC_SUNLOCK(p); 1177 if ((options & WNOWAIT) == 0) 1178 p->p_flag |= P_WAITED; 1179 sx_xunlock(&proctree_lock); 1180 1181 if (status != NULL) 1182 *status = W_STOPCODE(p->p_xsig); 1183 if (siginfo != NULL) { 1184 siginfo->si_status = p->p_xsig; 1185 siginfo->si_code = CLD_TRAPPED; 1186 } 1187 if ((options & WNOWAIT) == 0) { 1188 PROC_LOCK(q); 1189 sigqueue_take(p->p_ksi); 1190 PROC_UNLOCK(q); 1191 } 1192 1193 CTR4(KTR_PTRACE, 1194 "wait: returning trapped pid %d status %#x (xstat %d) xthread %d", 1195 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1196 p->p_xthread != NULL ? p->p_xthread->td_tid : -1); 1197 PROC_UNLOCK(p); 1198 td->td_retval[0] = pid; 1199 return (0); 1200 } 1201 if ((options & WUNTRACED) != 0 && 1202 (p->p_flag & P_STOPPED_SIG) != 0 && 1203 (p->p_suspcount == p->p_numthreads) && 1204 ((p->p_flag & P_WAITED) == 0)) { 1205 PROC_SUNLOCK(p); 1206 if ((options & WNOWAIT) == 0) 1207 p->p_flag |= P_WAITED; 1208 sx_xunlock(&proctree_lock); 1209 1210 if (status != NULL) 1211 *status = W_STOPCODE(p->p_xsig); 1212 if (siginfo != NULL) { 1213 siginfo->si_status = p->p_xsig; 1214 siginfo->si_code = CLD_STOPPED; 1215 } 1216 if ((options & WNOWAIT) == 0) { 1217 PROC_LOCK(q); 1218 sigqueue_take(p->p_ksi); 1219 PROC_UNLOCK(q); 1220 } 1221 1222 PROC_UNLOCK(p); 1223 td->td_retval[0] = pid; 1224 return (0); 1225 } 1226 PROC_SUNLOCK(p); 1227 if ((options & WCONTINUED) != 0 && 1228 (p->p_flag & P_CONTINUED) != 0) { 1229 sx_xunlock(&proctree_lock); 1230 if ((options & WNOWAIT) == 0) { 1231 p->p_flag &= ~P_CONTINUED; 1232 PROC_LOCK(q); 1233 sigqueue_take(p->p_ksi); 1234 PROC_UNLOCK(q); 1235 } 1236 PROC_UNLOCK(p); 1237 1238 if (status != NULL) 1239 *status = SIGCONT; 1240 if (siginfo != NULL) { 1241 siginfo->si_status = SIGCONT; 1242 siginfo->si_code = CLD_CONTINUED; 1243 } 1244 td->td_retval[0] = pid; 1245 return (0); 1246 } 1247 PROC_UNLOCK(p); 1248 } 1249 1250 /* 1251 * Look in the orphans list too, to allow the parent to 1252 * collect it's child exit status even if child is being 1253 * debugged. 1254 * 1255 * Debugger detaches from the parent upon successful 1256 * switch-over from parent to child. At this point due to 1257 * re-parenting the parent loses the child to debugger and a 1258 * wait4(2) call would report that it has no children to wait 1259 * for. By maintaining a list of orphans we allow the parent 1260 * to successfully wait until the child becomes a zombie. 1261 */ 1262 if (nfound == 0) { 1263 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1264 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1265 NULL, NULL, 1); 1266 if (ret != 0) { 1267 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1268 (int)td->td_retval[0])); 1269 nfound++; 1270 break; 1271 } 1272 } 1273 } 1274 if (nfound == 0) { 1275 sx_xunlock(&proctree_lock); 1276 return (ECHILD); 1277 } 1278 if (options & WNOHANG) { 1279 sx_xunlock(&proctree_lock); 1280 td->td_retval[0] = 0; 1281 return (0); 1282 } 1283 PROC_LOCK(q); 1284 sx_xunlock(&proctree_lock); 1285 if (q->p_flag & P_STATCHILD) { 1286 q->p_flag &= ~P_STATCHILD; 1287 error = 0; 1288 } else 1289 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1290 PROC_UNLOCK(q); 1291 if (error) 1292 return (error); 1293 goto loop; 1294 } 1295 1296 /* 1297 * Make process 'parent' the new parent of process 'child'. 1298 * Must be called with an exclusive hold of proctree lock. 1299 */ 1300 void 1301 proc_reparent(struct proc *child, struct proc *parent) 1302 { 1303 1304 sx_assert(&proctree_lock, SX_XLOCKED); 1305 PROC_LOCK_ASSERT(child, MA_OWNED); 1306 if (child->p_pptr == parent) 1307 return; 1308 1309 PROC_LOCK(child->p_pptr); 1310 sigqueue_take(child->p_ksi); 1311 PROC_UNLOCK(child->p_pptr); 1312 LIST_REMOVE(child, p_sibling); 1313 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1314 1315 clear_orphan(child); 1316 if (child->p_flag & P_TRACED) { 1317 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1318 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1319 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1320 p_orphan); 1321 } else { 1322 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1323 child, p_orphan); 1324 } 1325 child->p_treeflag |= P_TREE_ORPHANED; 1326 } 1327 1328 child->p_pptr = parent; 1329 } 1330