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 * If this is a process with a descriptor, we may not need to deliver 528 * a signal to the parent. proctree_lock is held over 529 * procdesc_exit() to serialize concurrent calls to close() and 530 * exit(). 531 */ 532 if (p->p_procdesc == NULL || procdesc_exit(p)) { 533 /* 534 * Notify parent that we're gone. If parent has the 535 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 536 * notify process 1 instead (and hope it will handle this 537 * situation). 538 */ 539 PROC_LOCK(p->p_pptr); 540 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 541 if (p->p_pptr->p_sigacts->ps_flag & 542 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 543 struct proc *pp; 544 545 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 546 pp = p->p_pptr; 547 PROC_UNLOCK(pp); 548 proc_reparent(p, p->p_reaper); 549 p->p_sigparent = SIGCHLD; 550 PROC_LOCK(p->p_pptr); 551 552 /* 553 * Notify parent, so in case he was wait(2)ing or 554 * executing waitpid(2) with our pid, he will 555 * continue. 556 */ 557 wakeup(pp); 558 } else 559 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 560 561 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 562 childproc_exited(p); 563 else if (p->p_sigparent != 0) { 564 if (p->p_sigparent == SIGCHLD) 565 childproc_exited(p); 566 else /* LINUX thread */ 567 kern_psignal(p->p_pptr, p->p_sigparent); 568 } 569 } else 570 PROC_LOCK(p->p_pptr); 571 sx_xunlock(&proctree_lock); 572 573 /* 574 * The state PRS_ZOMBIE prevents other proesses from sending 575 * signal to the process, to avoid memory leak, we free memory 576 * for signal queue at the time when the state is set. 577 */ 578 sigqueue_flush(&p->p_sigqueue); 579 sigqueue_flush(&td->td_sigqueue); 580 581 /* 582 * We have to wait until after acquiring all locks before 583 * changing p_state. We need to avoid all possible context 584 * switches (including ones from blocking on a mutex) while 585 * marked as a zombie. We also have to set the zombie state 586 * before we release the parent process' proc lock to avoid 587 * a lost wakeup. So, we first call wakeup, then we grab the 588 * sched lock, update the state, and release the parent process' 589 * proc lock. 590 */ 591 wakeup(p->p_pptr); 592 cv_broadcast(&p->p_pwait); 593 sched_exit(p->p_pptr, td); 594 PROC_SLOCK(p); 595 p->p_state = PRS_ZOMBIE; 596 PROC_UNLOCK(p->p_pptr); 597 598 /* 599 * Save our children's rusage information in our exit rusage. 600 */ 601 PROC_STATLOCK(p); 602 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 603 PROC_STATUNLOCK(p); 604 605 /* 606 * Make sure the scheduler takes this thread out of its tables etc. 607 * This will also release this thread's reference to the ucred. 608 * Other thread parts to release include pcb bits and such. 609 */ 610 thread_exit(); 611 } 612 613 614 #ifndef _SYS_SYSPROTO_H_ 615 struct abort2_args { 616 char *why; 617 int nargs; 618 void **args; 619 }; 620 #endif 621 622 int 623 sys_abort2(struct thread *td, struct abort2_args *uap) 624 { 625 struct proc *p = td->td_proc; 626 struct sbuf *sb; 627 void *uargs[16]; 628 int error, i, sig; 629 630 /* 631 * Do it right now so we can log either proper call of abort2(), or 632 * note, that invalid argument was passed. 512 is big enough to 633 * handle 16 arguments' descriptions with additional comments. 634 */ 635 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 636 sbuf_clear(sb); 637 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 638 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 639 /* 640 * Since we can't return from abort2(), send SIGKILL in cases, where 641 * abort2() was called improperly 642 */ 643 sig = SIGKILL; 644 /* Prevent from DoSes from user-space. */ 645 if (uap->nargs < 0 || uap->nargs > 16) 646 goto out; 647 if (uap->nargs > 0) { 648 if (uap->args == NULL) 649 goto out; 650 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 651 if (error != 0) 652 goto out; 653 } 654 /* 655 * Limit size of 'reason' string to 128. Will fit even when 656 * maximal number of arguments was chosen to be logged. 657 */ 658 if (uap->why != NULL) { 659 error = sbuf_copyin(sb, uap->why, 128); 660 if (error < 0) 661 goto out; 662 } else { 663 sbuf_printf(sb, "(null)"); 664 } 665 if (uap->nargs > 0) { 666 sbuf_printf(sb, "("); 667 for (i = 0;i < uap->nargs; i++) 668 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 669 sbuf_printf(sb, ")"); 670 } 671 /* 672 * Final stage: arguments were proper, string has been 673 * successfully copied from userspace, and copying pointers 674 * from user-space succeed. 675 */ 676 sig = SIGABRT; 677 out: 678 if (sig == SIGKILL) { 679 sbuf_trim(sb); 680 sbuf_printf(sb, " (Reason text inaccessible)"); 681 } 682 sbuf_cat(sb, "\n"); 683 sbuf_finish(sb); 684 log(LOG_INFO, "%s", sbuf_data(sb)); 685 sbuf_delete(sb); 686 exit1(td, 0, sig); 687 return (0); 688 } 689 690 691 #ifdef COMPAT_43 692 /* 693 * The dirty work is handled by kern_wait(). 694 */ 695 int 696 owait(struct thread *td, struct owait_args *uap __unused) 697 { 698 int error, status; 699 700 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 701 if (error == 0) 702 td->td_retval[1] = status; 703 return (error); 704 } 705 #endif /* COMPAT_43 */ 706 707 /* 708 * The dirty work is handled by kern_wait(). 709 */ 710 int 711 sys_wait4(struct thread *td, struct wait4_args *uap) 712 { 713 struct rusage ru, *rup; 714 int error, status; 715 716 if (uap->rusage != NULL) 717 rup = &ru; 718 else 719 rup = NULL; 720 error = kern_wait(td, uap->pid, &status, uap->options, rup); 721 if (uap->status != NULL && error == 0) 722 error = copyout(&status, uap->status, sizeof(status)); 723 if (uap->rusage != NULL && error == 0) 724 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 725 return (error); 726 } 727 728 int 729 sys_wait6(struct thread *td, struct wait6_args *uap) 730 { 731 struct __wrusage wru, *wrup; 732 siginfo_t si, *sip; 733 idtype_t idtype; 734 id_t id; 735 int error, status; 736 737 idtype = uap->idtype; 738 id = uap->id; 739 740 if (uap->wrusage != NULL) 741 wrup = &wru; 742 else 743 wrup = NULL; 744 745 if (uap->info != NULL) { 746 sip = &si; 747 bzero(sip, sizeof(*sip)); 748 } else 749 sip = NULL; 750 751 /* 752 * We expect all callers of wait6() to know about WEXITED and 753 * WTRAPPED. 754 */ 755 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 756 757 if (uap->status != NULL && error == 0) 758 error = copyout(&status, uap->status, sizeof(status)); 759 if (uap->wrusage != NULL && error == 0) 760 error = copyout(&wru, uap->wrusage, sizeof(wru)); 761 if (uap->info != NULL && error == 0) 762 error = copyout(&si, uap->info, sizeof(si)); 763 return (error); 764 } 765 766 /* 767 * Reap the remains of a zombie process and optionally return status and 768 * rusage. Asserts and will release both the proctree_lock and the process 769 * lock as part of its work. 770 */ 771 void 772 proc_reap(struct thread *td, struct proc *p, int *status, int options) 773 { 774 struct proc *q, *t; 775 776 sx_assert(&proctree_lock, SA_XLOCKED); 777 PROC_LOCK_ASSERT(p, MA_OWNED); 778 PROC_SLOCK_ASSERT(p, MA_OWNED); 779 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 780 781 q = td->td_proc; 782 783 PROC_SUNLOCK(p); 784 if (status) 785 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 786 if (options & WNOWAIT) { 787 /* 788 * Only poll, returning the status. Caller does not wish to 789 * release the proc struct just yet. 790 */ 791 PROC_UNLOCK(p); 792 sx_xunlock(&proctree_lock); 793 return; 794 } 795 796 PROC_LOCK(q); 797 sigqueue_take(p->p_ksi); 798 PROC_UNLOCK(q); 799 800 /* 801 * If we got the child via a ptrace 'attach', we need to give it back 802 * to the old parent. 803 */ 804 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 805 PROC_UNLOCK(p); 806 t = proc_realparent(p); 807 PROC_LOCK(t); 808 PROC_LOCK(p); 809 CTR2(KTR_PTRACE, 810 "wait: traced child %d moved back to parent %d", p->p_pid, 811 t->p_pid); 812 proc_reparent(p, t); 813 p->p_oppid = 0; 814 PROC_UNLOCK(p); 815 pksignal(t, SIGCHLD, p->p_ksi); 816 wakeup(t); 817 cv_broadcast(&p->p_pwait); 818 PROC_UNLOCK(t); 819 sx_xunlock(&proctree_lock); 820 return; 821 } 822 p->p_oppid = 0; 823 PROC_UNLOCK(p); 824 825 /* 826 * Remove other references to this process to ensure we have an 827 * exclusive reference. 828 */ 829 sx_xlock(&allproc_lock); 830 LIST_REMOVE(p, p_list); /* off zombproc */ 831 sx_xunlock(&allproc_lock); 832 LIST_REMOVE(p, p_sibling); 833 reaper_abandon_children(p, true); 834 LIST_REMOVE(p, p_reapsibling); 835 PROC_LOCK(p); 836 clear_orphan(p); 837 PROC_UNLOCK(p); 838 leavepgrp(p); 839 if (p->p_procdesc != NULL) 840 procdesc_reap(p); 841 sx_xunlock(&proctree_lock); 842 843 PROC_LOCK(p); 844 knlist_detach(p->p_klist); 845 p->p_klist = NULL; 846 PROC_UNLOCK(p); 847 848 /* 849 * Removal from allproc list and process group list paired with 850 * PROC_LOCK which was executed during that time should guarantee 851 * nothing can reach this process anymore. As such further locking 852 * is unnecessary. 853 */ 854 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 855 856 PROC_LOCK(q); 857 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 858 PROC_UNLOCK(q); 859 860 /* 861 * Decrement the count of procs running with this uid. 862 */ 863 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 864 865 /* 866 * Destroy resource accounting information associated with the process. 867 */ 868 #ifdef RACCT 869 if (racct_enable) { 870 PROC_LOCK(p); 871 racct_sub(p, RACCT_NPROC, 1); 872 PROC_UNLOCK(p); 873 } 874 #endif 875 racct_proc_exit(p); 876 877 /* 878 * Free credentials, arguments, and sigacts. 879 */ 880 crfree(p->p_ucred); 881 proc_set_cred(p, NULL); 882 pargs_drop(p->p_args); 883 p->p_args = NULL; 884 sigacts_free(p->p_sigacts); 885 p->p_sigacts = NULL; 886 887 /* 888 * Do any thread-system specific cleanups. 889 */ 890 thread_wait(p); 891 892 /* 893 * Give vm and machine-dependent layer a chance to free anything that 894 * cpu_exit couldn't release while still running in process context. 895 */ 896 vm_waitproc(p); 897 #ifdef MAC 898 mac_proc_destroy(p); 899 #endif 900 /* 901 * Free any domain policy that's still hiding around. 902 */ 903 vm_domain_policy_cleanup(&p->p_vm_dom_policy); 904 905 KASSERT(FIRST_THREAD_IN_PROC(p), 906 ("proc_reap: no residual thread!")); 907 uma_zfree(proc_zone, p); 908 atomic_add_int(&nprocs, -1); 909 } 910 911 static int 912 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 913 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 914 int check_only) 915 { 916 struct rusage *rup; 917 918 sx_assert(&proctree_lock, SA_XLOCKED); 919 920 PROC_LOCK(p); 921 922 switch (idtype) { 923 case P_ALL: 924 if (p->p_procdesc != NULL) { 925 PROC_UNLOCK(p); 926 return (0); 927 } 928 break; 929 case P_PID: 930 if (p->p_pid != (pid_t)id) { 931 PROC_UNLOCK(p); 932 return (0); 933 } 934 break; 935 case P_PGID: 936 if (p->p_pgid != (pid_t)id) { 937 PROC_UNLOCK(p); 938 return (0); 939 } 940 break; 941 case P_SID: 942 if (p->p_session->s_sid != (pid_t)id) { 943 PROC_UNLOCK(p); 944 return (0); 945 } 946 break; 947 case P_UID: 948 if (p->p_ucred->cr_uid != (uid_t)id) { 949 PROC_UNLOCK(p); 950 return (0); 951 } 952 break; 953 case P_GID: 954 if (p->p_ucred->cr_gid != (gid_t)id) { 955 PROC_UNLOCK(p); 956 return (0); 957 } 958 break; 959 case P_JAILID: 960 if (p->p_ucred->cr_prison->pr_id != (int)id) { 961 PROC_UNLOCK(p); 962 return (0); 963 } 964 break; 965 /* 966 * It seems that the thread structures get zeroed out 967 * at process exit. This makes it impossible to 968 * support P_SETID, P_CID or P_CPUID. 969 */ 970 default: 971 PROC_UNLOCK(p); 972 return (0); 973 } 974 975 if (p_canwait(td, p)) { 976 PROC_UNLOCK(p); 977 return (0); 978 } 979 980 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 981 PROC_UNLOCK(p); 982 return (0); 983 } 984 985 /* 986 * This special case handles a kthread spawned by linux_clone 987 * (see linux_misc.c). The linux_wait4 and linux_waitpid 988 * functions need to be able to distinguish between waiting 989 * on a process and waiting on a thread. It is a thread if 990 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 991 * signifies we want to wait for threads and not processes. 992 */ 993 if ((p->p_sigparent != SIGCHLD) ^ 994 ((options & WLINUXCLONE) != 0)) { 995 PROC_UNLOCK(p); 996 return (0); 997 } 998 999 if (siginfo != NULL) { 1000 bzero(siginfo, sizeof(*siginfo)); 1001 siginfo->si_errno = 0; 1002 1003 /* 1004 * SUSv4 requires that the si_signo value is always 1005 * SIGCHLD. Obey it despite the rfork(2) interface 1006 * allows to request other signal for child exit 1007 * notification. 1008 */ 1009 siginfo->si_signo = SIGCHLD; 1010 1011 /* 1012 * This is still a rough estimate. We will fix the 1013 * cases TRAPPED, STOPPED, and CONTINUED later. 1014 */ 1015 if (WCOREDUMP(p->p_xsig)) { 1016 siginfo->si_code = CLD_DUMPED; 1017 siginfo->si_status = WTERMSIG(p->p_xsig); 1018 } else if (WIFSIGNALED(p->p_xsig)) { 1019 siginfo->si_code = CLD_KILLED; 1020 siginfo->si_status = WTERMSIG(p->p_xsig); 1021 } else { 1022 siginfo->si_code = CLD_EXITED; 1023 siginfo->si_status = p->p_xexit; 1024 } 1025 1026 siginfo->si_pid = p->p_pid; 1027 siginfo->si_uid = p->p_ucred->cr_uid; 1028 1029 /* 1030 * The si_addr field would be useful additional 1031 * detail, but apparently the PC value may be lost 1032 * when we reach this point. bzero() above sets 1033 * siginfo->si_addr to NULL. 1034 */ 1035 } 1036 1037 /* 1038 * There should be no reason to limit resources usage info to 1039 * exited processes only. A snapshot about any resources used 1040 * by a stopped process may be exactly what is needed. 1041 */ 1042 if (wrusage != NULL) { 1043 rup = &wrusage->wru_self; 1044 *rup = p->p_ru; 1045 PROC_STATLOCK(p); 1046 calcru(p, &rup->ru_utime, &rup->ru_stime); 1047 PROC_STATUNLOCK(p); 1048 1049 rup = &wrusage->wru_children; 1050 *rup = p->p_stats->p_cru; 1051 calccru(p, &rup->ru_utime, &rup->ru_stime); 1052 } 1053 1054 if (p->p_state == PRS_ZOMBIE && !check_only) { 1055 PROC_SLOCK(p); 1056 proc_reap(td, p, status, options); 1057 return (-1); 1058 } 1059 PROC_UNLOCK(p); 1060 return (1); 1061 } 1062 1063 int 1064 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1065 struct rusage *rusage) 1066 { 1067 struct __wrusage wru, *wrup; 1068 idtype_t idtype; 1069 id_t id; 1070 int ret; 1071 1072 /* 1073 * Translate the special pid values into the (idtype, pid) 1074 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1075 * kern_wait6() on its own. 1076 */ 1077 if (pid == WAIT_ANY) { 1078 idtype = P_ALL; 1079 id = 0; 1080 } else if (pid < 0) { 1081 idtype = P_PGID; 1082 id = (id_t)-pid; 1083 } else { 1084 idtype = P_PID; 1085 id = (id_t)pid; 1086 } 1087 1088 if (rusage != NULL) 1089 wrup = &wru; 1090 else 1091 wrup = NULL; 1092 1093 /* 1094 * For backward compatibility we implicitly add flags WEXITED 1095 * and WTRAPPED here. 1096 */ 1097 options |= WEXITED | WTRAPPED; 1098 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1099 if (rusage != NULL) 1100 *rusage = wru.wru_self; 1101 return (ret); 1102 } 1103 1104 int 1105 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1106 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1107 { 1108 struct proc *p, *q; 1109 pid_t pid; 1110 int error, nfound, ret; 1111 1112 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1113 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1114 AUDIT_ARG_VALUE(options); 1115 1116 q = td->td_proc; 1117 1118 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1119 PROC_LOCK(q); 1120 id = (id_t)q->p_pgid; 1121 PROC_UNLOCK(q); 1122 idtype = P_PGID; 1123 } 1124 1125 /* If we don't know the option, just return. */ 1126 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1127 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1128 return (EINVAL); 1129 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1130 /* 1131 * We will be unable to find any matching processes, 1132 * because there are no known events to look for. 1133 * Prefer to return error instead of blocking 1134 * indefinitely. 1135 */ 1136 return (EINVAL); 1137 } 1138 1139 loop: 1140 if (q->p_flag & P_STATCHILD) { 1141 PROC_LOCK(q); 1142 q->p_flag &= ~P_STATCHILD; 1143 PROC_UNLOCK(q); 1144 } 1145 nfound = 0; 1146 sx_xlock(&proctree_lock); 1147 LIST_FOREACH(p, &q->p_children, p_sibling) { 1148 pid = p->p_pid; 1149 ret = proc_to_reap(td, p, idtype, id, status, options, 1150 wrusage, siginfo, 0); 1151 if (ret == 0) 1152 continue; 1153 else if (ret == 1) 1154 nfound++; 1155 else { 1156 td->td_retval[0] = pid; 1157 return (0); 1158 } 1159 1160 PROC_LOCK(p); 1161 PROC_SLOCK(p); 1162 1163 if ((options & WTRAPPED) != 0 && 1164 (p->p_flag & P_TRACED) != 0 && 1165 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1166 (p->p_suspcount == p->p_numthreads) && 1167 ((p->p_flag & P_WAITED) == 0)) { 1168 PROC_SUNLOCK(p); 1169 if ((options & WNOWAIT) == 0) 1170 p->p_flag |= P_WAITED; 1171 sx_xunlock(&proctree_lock); 1172 1173 if (status != NULL) 1174 *status = W_STOPCODE(p->p_xsig); 1175 if (siginfo != NULL) { 1176 siginfo->si_status = p->p_xsig; 1177 siginfo->si_code = CLD_TRAPPED; 1178 } 1179 if ((options & WNOWAIT) == 0) { 1180 PROC_LOCK(q); 1181 sigqueue_take(p->p_ksi); 1182 PROC_UNLOCK(q); 1183 } 1184 1185 CTR4(KTR_PTRACE, 1186 "wait: returning trapped pid %d status %#x (xstat %d) xthread %d", 1187 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1188 p->p_xthread != NULL ? p->p_xthread->td_tid : -1); 1189 PROC_UNLOCK(p); 1190 td->td_retval[0] = pid; 1191 return (0); 1192 } 1193 if ((options & WUNTRACED) != 0 && 1194 (p->p_flag & P_STOPPED_SIG) != 0 && 1195 (p->p_suspcount == p->p_numthreads) && 1196 ((p->p_flag & P_WAITED) == 0)) { 1197 PROC_SUNLOCK(p); 1198 if ((options & WNOWAIT) == 0) 1199 p->p_flag |= P_WAITED; 1200 sx_xunlock(&proctree_lock); 1201 1202 if (status != NULL) 1203 *status = W_STOPCODE(p->p_xsig); 1204 if (siginfo != NULL) { 1205 siginfo->si_status = p->p_xsig; 1206 siginfo->si_code = CLD_STOPPED; 1207 } 1208 if ((options & WNOWAIT) == 0) { 1209 PROC_LOCK(q); 1210 sigqueue_take(p->p_ksi); 1211 PROC_UNLOCK(q); 1212 } 1213 1214 PROC_UNLOCK(p); 1215 td->td_retval[0] = pid; 1216 return (0); 1217 } 1218 PROC_SUNLOCK(p); 1219 if ((options & WCONTINUED) != 0 && 1220 (p->p_flag & P_CONTINUED) != 0) { 1221 sx_xunlock(&proctree_lock); 1222 if ((options & WNOWAIT) == 0) { 1223 p->p_flag &= ~P_CONTINUED; 1224 PROC_LOCK(q); 1225 sigqueue_take(p->p_ksi); 1226 PROC_UNLOCK(q); 1227 } 1228 PROC_UNLOCK(p); 1229 1230 if (status != NULL) 1231 *status = SIGCONT; 1232 if (siginfo != NULL) { 1233 siginfo->si_status = SIGCONT; 1234 siginfo->si_code = CLD_CONTINUED; 1235 } 1236 td->td_retval[0] = pid; 1237 return (0); 1238 } 1239 PROC_UNLOCK(p); 1240 } 1241 1242 /* 1243 * Look in the orphans list too, to allow the parent to 1244 * collect it's child exit status even if child is being 1245 * debugged. 1246 * 1247 * Debugger detaches from the parent upon successful 1248 * switch-over from parent to child. At this point due to 1249 * re-parenting the parent loses the child to debugger and a 1250 * wait4(2) call would report that it has no children to wait 1251 * for. By maintaining a list of orphans we allow the parent 1252 * to successfully wait until the child becomes a zombie. 1253 */ 1254 if (nfound == 0) { 1255 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1256 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1257 NULL, NULL, 1); 1258 if (ret != 0) { 1259 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1260 (int)td->td_retval[0])); 1261 nfound++; 1262 break; 1263 } 1264 } 1265 } 1266 if (nfound == 0) { 1267 sx_xunlock(&proctree_lock); 1268 return (ECHILD); 1269 } 1270 if (options & WNOHANG) { 1271 sx_xunlock(&proctree_lock); 1272 td->td_retval[0] = 0; 1273 return (0); 1274 } 1275 PROC_LOCK(q); 1276 sx_xunlock(&proctree_lock); 1277 if (q->p_flag & P_STATCHILD) { 1278 q->p_flag &= ~P_STATCHILD; 1279 error = 0; 1280 } else 1281 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1282 PROC_UNLOCK(q); 1283 if (error) 1284 return (error); 1285 goto loop; 1286 } 1287 1288 /* 1289 * Make process 'parent' the new parent of process 'child'. 1290 * Must be called with an exclusive hold of proctree lock. 1291 */ 1292 void 1293 proc_reparent(struct proc *child, struct proc *parent) 1294 { 1295 1296 sx_assert(&proctree_lock, SX_XLOCKED); 1297 PROC_LOCK_ASSERT(child, MA_OWNED); 1298 if (child->p_pptr == parent) 1299 return; 1300 1301 PROC_LOCK(child->p_pptr); 1302 sigqueue_take(child->p_ksi); 1303 PROC_UNLOCK(child->p_pptr); 1304 LIST_REMOVE(child, p_sibling); 1305 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1306 1307 clear_orphan(child); 1308 if (child->p_flag & P_TRACED) { 1309 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1310 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1311 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1312 p_orphan); 1313 } else { 1314 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1315 child, p_orphan); 1316 } 1317 child->p_treeflag |= P_TREE_ORPHANED; 1318 } 1319 1320 child->p_pptr = parent; 1321 } 1322