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 * 3. 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 EVENTHANDLER_LIST_DECLARE(process_exit); 103 104 struct proc * 105 proc_realparent(struct proc *child) 106 { 107 struct proc *p, *parent; 108 109 sx_assert(&proctree_lock, SX_LOCKED); 110 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 111 if (child->p_oppid == 0 || 112 child->p_pptr->p_pid == child->p_oppid) 113 parent = child->p_pptr; 114 else 115 parent = initproc; 116 return (parent); 117 } 118 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 119 /* Cannot use LIST_PREV(), since the list head is not known. */ 120 p = __containerof(p->p_orphan.le_prev, struct proc, 121 p_orphan.le_next); 122 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 123 ("missing P_ORPHAN %p", p)); 124 } 125 parent = __containerof(p->p_orphan.le_prev, struct proc, 126 p_orphans.lh_first); 127 return (parent); 128 } 129 130 void 131 reaper_abandon_children(struct proc *p, bool exiting) 132 { 133 struct proc *p1, *p2, *ptmp; 134 135 sx_assert(&proctree_lock, SX_LOCKED); 136 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 137 if ((p->p_treeflag & P_TREE_REAPER) == 0) 138 return; 139 p1 = p->p_reaper; 140 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 141 LIST_REMOVE(p2, p_reapsibling); 142 p2->p_reaper = p1; 143 p2->p_reapsubtree = p->p_reapsubtree; 144 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 145 if (exiting && p2->p_pptr == p) { 146 PROC_LOCK(p2); 147 proc_reparent(p2, p1); 148 PROC_UNLOCK(p2); 149 } 150 } 151 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 152 p->p_treeflag &= ~P_TREE_REAPER; 153 } 154 155 static void 156 clear_orphan(struct proc *p) 157 { 158 struct proc *p1; 159 160 sx_assert(&proctree_lock, SA_XLOCKED); 161 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 162 return; 163 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 164 p1 = LIST_NEXT(p, p_orphan); 165 if (p1 != NULL) 166 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 167 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 168 } 169 LIST_REMOVE(p, p_orphan); 170 p->p_treeflag &= ~P_TREE_ORPHANED; 171 } 172 173 /* 174 * exit -- death of process. 175 */ 176 void 177 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 178 { 179 180 exit1(td, uap->rval, 0); 181 /* NOTREACHED */ 182 } 183 184 /* 185 * Exit: deallocate address space and other resources, change proc state to 186 * zombie, and unlink proc from allproc and parent's lists. Save exit status 187 * and rusage for wait(). Check for child processes and orphan them. 188 */ 189 void 190 exit1(struct thread *td, int rval, int signo) 191 { 192 struct proc *p, *nq, *q, *t; 193 struct thread *tdt; 194 ksiginfo_t *ksi, *ksi1; 195 196 mtx_assert(&Giant, MA_NOTOWNED); 197 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 198 199 p = td->td_proc; 200 /* 201 * XXX in case we're rebooting we just let init die in order to 202 * work around an unsolved stack overflow seen very late during 203 * shutdown on sparc64 when the gmirror worker process exists. 204 */ 205 if (p == initproc && rebooting == 0) { 206 printf("init died (signal %d, exit %d)\n", signo, rval); 207 panic("Going nowhere without my init!"); 208 } 209 210 /* 211 * Deref SU mp, since the thread does not return to userspace. 212 */ 213 td_softdep_cleanup(td); 214 215 /* 216 * MUST abort all other threads before proceeding past here. 217 */ 218 PROC_LOCK(p); 219 /* 220 * First check if some other thread or external request got 221 * here before us. If so, act appropriately: exit or suspend. 222 * We must ensure that stop requests are handled before we set 223 * P_WEXIT. 224 */ 225 thread_suspend_check(0); 226 while (p->p_flag & P_HADTHREADS) { 227 /* 228 * Kill off the other threads. This requires 229 * some co-operation from other parts of the kernel 230 * so it may not be instantaneous. With this state set 231 * any thread entering the kernel from userspace will 232 * thread_exit() in trap(). Any thread attempting to 233 * sleep will return immediately with EINTR or EWOULDBLOCK 234 * which will hopefully force them to back out to userland 235 * freeing resources as they go. Any thread attempting 236 * to return to userland will thread_exit() from userret(). 237 * thread_exit() will unsuspend us when the last of the 238 * other threads exits. 239 * If there is already a thread singler after resumption, 240 * calling thread_single will fail; in that case, we just 241 * re-check all suspension request, the thread should 242 * either be suspended there or exit. 243 */ 244 if (!thread_single(p, SINGLE_EXIT)) 245 /* 246 * All other activity in this process is now 247 * stopped. Threading support has been turned 248 * off. 249 */ 250 break; 251 /* 252 * Recheck for new stop or suspend requests which 253 * might appear while process lock was dropped in 254 * thread_single(). 255 */ 256 thread_suspend_check(0); 257 } 258 KASSERT(p->p_numthreads == 1, 259 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 260 racct_sub(p, RACCT_NTHR, 1); 261 262 /* Let event handler change exit status */ 263 p->p_xexit = rval; 264 p->p_xsig = signo; 265 266 /* 267 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 268 * on our vmspace, so we should block below until they have 269 * released their reference to us. Note that if they have 270 * requested S_EXIT stops we will block here until they ack 271 * via PIOCCONT. 272 */ 273 _STOPEVENT(p, S_EXIT, 0); 274 275 /* 276 * Ignore any pending request to stop due to a stop signal. 277 * Once P_WEXIT is set, future requests will be ignored as 278 * well. 279 */ 280 p->p_flag &= ~P_STOPPED_SIG; 281 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 282 283 /* 284 * Note that we are exiting and do another wakeup of anyone in 285 * PIOCWAIT in case they aren't listening for S_EXIT stops or 286 * decided to wait again after we told them we are exiting. 287 */ 288 p->p_flag |= P_WEXIT; 289 wakeup(&p->p_stype); 290 291 /* 292 * Wait for any processes that have a hold on our vmspace to 293 * release their reference. 294 */ 295 while (p->p_lock > 0) 296 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 297 298 PROC_UNLOCK(p); 299 /* Drain the limit callout while we don't have the proc locked */ 300 callout_drain(&p->p_limco); 301 302 #ifdef AUDIT 303 /* 304 * The Sun BSM exit token contains two components: an exit status as 305 * passed to exit(), and a return value to indicate what sort of exit 306 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 307 * what the return value is. 308 */ 309 AUDIT_ARG_EXIT(rval, 0); 310 AUDIT_SYSCALL_EXIT(0, td); 311 #endif 312 313 /* Are we a task leader with peers? */ 314 if (p->p_peers != NULL && p == p->p_leader) { 315 mtx_lock(&ppeers_lock); 316 q = p->p_peers; 317 while (q != NULL) { 318 PROC_LOCK(q); 319 kern_psignal(q, SIGKILL); 320 PROC_UNLOCK(q); 321 q = q->p_peers; 322 } 323 while (p->p_peers != NULL) 324 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 325 mtx_unlock(&ppeers_lock); 326 } 327 328 /* 329 * Check if any loadable modules need anything done at process exit. 330 * E.g. SYSV IPC stuff. 331 * Event handler could change exit status. 332 * XXX what if one of these generates an error? 333 */ 334 EVENTHANDLER_DIRECT_INVOKE(process_exit, p); 335 336 /* 337 * If parent is waiting for us to exit or exec, 338 * P_PPWAIT is set; we will wakeup the parent below. 339 */ 340 PROC_LOCK(p); 341 stopprofclock(p); 342 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 343 p->p_ptevents = 0; 344 345 /* 346 * Stop the real interval timer. If the handler is currently 347 * executing, prevent it from rearming itself and let it finish. 348 */ 349 if (timevalisset(&p->p_realtimer.it_value) && 350 _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) { 351 timevalclear(&p->p_realtimer.it_interval); 352 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 353 KASSERT(!timevalisset(&p->p_realtimer.it_value), 354 ("realtime timer is still armed")); 355 } 356 357 PROC_UNLOCK(p); 358 359 umtx_thread_exit(td); 360 361 /* 362 * Reset any sigio structures pointing to us as a result of 363 * F_SETOWN with our pid. 364 */ 365 funsetownlst(&p->p_sigiolst); 366 367 /* 368 * If this process has an nlminfo data area (for lockd), release it 369 */ 370 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 371 (*nlminfo_release_p)(p); 372 373 /* 374 * Close open files and release open-file table. 375 * This may block! 376 */ 377 fdescfree(td); 378 379 /* 380 * If this thread tickled GEOM, we need to wait for the giggling to 381 * stop before we return to userland 382 */ 383 if (td->td_pflags & TDP_GEOM) 384 g_waitidle(); 385 386 /* 387 * Remove ourself from our leader's peer list and wake our leader. 388 */ 389 if (p->p_leader->p_peers != NULL) { 390 mtx_lock(&ppeers_lock); 391 if (p->p_leader->p_peers != NULL) { 392 q = p->p_leader; 393 while (q->p_peers != p) 394 q = q->p_peers; 395 q->p_peers = p->p_peers; 396 wakeup(p->p_leader); 397 } 398 mtx_unlock(&ppeers_lock); 399 } 400 401 vmspace_exit(td); 402 killjobc(); 403 (void)acct_process(td); 404 405 #ifdef KTRACE 406 ktrprocexit(td); 407 #endif 408 /* 409 * Release reference to text vnode 410 */ 411 if (p->p_textvp != NULL) { 412 vrele(p->p_textvp); 413 p->p_textvp = NULL; 414 } 415 416 /* 417 * Release our limits structure. 418 */ 419 lim_free(p->p_limit); 420 p->p_limit = NULL; 421 422 tidhash_remove(td); 423 424 /* 425 * Remove proc from allproc queue and pidhash chain. 426 * Place onto zombproc. Unlink from parent's child list. 427 */ 428 sx_xlock(&allproc_lock); 429 LIST_REMOVE(p, p_list); 430 LIST_INSERT_HEAD(&zombproc, p, p_list); 431 LIST_REMOVE(p, p_hash); 432 sx_xunlock(&allproc_lock); 433 434 /* 435 * Call machine-dependent code to release any 436 * machine-dependent resources other than the address space. 437 * The address space is released by "vmspace_exitfree(p)" in 438 * vm_waitproc(). 439 */ 440 cpu_exit(td); 441 442 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 443 444 /* 445 * Reparent all children processes: 446 * - traced ones to the original parent (or init if we are that parent) 447 * - the rest to init 448 */ 449 sx_xlock(&proctree_lock); 450 q = LIST_FIRST(&p->p_children); 451 if (q != NULL) /* only need this if any child is S_ZOMB */ 452 wakeup(q->p_reaper); 453 for (; q != NULL; q = nq) { 454 nq = LIST_NEXT(q, p_sibling); 455 ksi = ksiginfo_alloc(TRUE); 456 PROC_LOCK(q); 457 q->p_sigparent = SIGCHLD; 458 459 if (!(q->p_flag & P_TRACED)) { 460 proc_reparent(q, q->p_reaper); 461 if (q->p_state == PRS_ZOMBIE) { 462 /* 463 * Inform reaper about the reparented 464 * zombie, since wait(2) has something 465 * new to report. Guarantee queueing 466 * of the SIGCHLD signal, similar to 467 * the _exit() behaviour, by providing 468 * our ksiginfo. Ksi is freed by the 469 * signal delivery. 470 */ 471 if (q->p_ksi == NULL) { 472 ksi1 = NULL; 473 } else { 474 ksiginfo_copy(q->p_ksi, ksi); 475 ksi->ksi_flags |= KSI_INS; 476 ksi1 = ksi; 477 ksi = NULL; 478 } 479 PROC_LOCK(q->p_reaper); 480 pksignal(q->p_reaper, SIGCHLD, ksi1); 481 PROC_UNLOCK(q->p_reaper); 482 } 483 } else { 484 /* 485 * Traced processes are killed since their existence 486 * means someone is screwing up. 487 */ 488 t = proc_realparent(q); 489 if (t == p) { 490 proc_reparent(q, q->p_reaper); 491 } else { 492 PROC_LOCK(t); 493 proc_reparent(q, t); 494 PROC_UNLOCK(t); 495 } 496 /* 497 * Since q was found on our children list, the 498 * proc_reparent() call moved q to the orphan 499 * list due to present P_TRACED flag. Clear 500 * orphan link for q now while q is locked. 501 */ 502 clear_orphan(q); 503 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 504 q->p_flag2 &= ~P2_PTRACE_FSTP; 505 q->p_ptevents = 0; 506 FOREACH_THREAD_IN_PROC(q, tdt) { 507 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 508 TDB_FSTP); 509 } 510 kern_psignal(q, SIGKILL); 511 } 512 PROC_UNLOCK(q); 513 if (ksi != NULL) 514 ksiginfo_free(ksi); 515 } 516 517 /* 518 * Also get rid of our orphans. 519 */ 520 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 521 PROC_LOCK(q); 522 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 523 q->p_pid); 524 clear_orphan(q); 525 PROC_UNLOCK(q); 526 } 527 528 /* Save exit status. */ 529 PROC_LOCK(p); 530 p->p_xthread = td; 531 532 /* Tell the prison that we are gone. */ 533 prison_proc_free(p->p_ucred->cr_prison); 534 535 #ifdef KDTRACE_HOOKS 536 /* 537 * Tell the DTrace fasttrap provider about the exit if it 538 * has declared an interest. 539 */ 540 if (dtrace_fasttrap_exit) 541 dtrace_fasttrap_exit(p); 542 #endif 543 544 /* 545 * Notify interested parties of our demise. 546 */ 547 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 548 549 #ifdef KDTRACE_HOOKS 550 int reason = CLD_EXITED; 551 if (WCOREDUMP(signo)) 552 reason = CLD_DUMPED; 553 else if (WIFSIGNALED(signo)) 554 reason = CLD_KILLED; 555 SDT_PROBE1(proc, , , exit, reason); 556 #endif 557 558 /* 559 * If this is a process with a descriptor, we may not need to deliver 560 * a signal to the parent. proctree_lock is held over 561 * procdesc_exit() to serialize concurrent calls to close() and 562 * exit(). 563 */ 564 if (p->p_procdesc == NULL || procdesc_exit(p)) { 565 /* 566 * Notify parent that we're gone. If parent has the 567 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 568 * notify process 1 instead (and hope it will handle this 569 * situation). 570 */ 571 PROC_LOCK(p->p_pptr); 572 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 573 if (p->p_pptr->p_sigacts->ps_flag & 574 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 575 struct proc *pp; 576 577 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 578 pp = p->p_pptr; 579 PROC_UNLOCK(pp); 580 proc_reparent(p, p->p_reaper); 581 p->p_sigparent = SIGCHLD; 582 PROC_LOCK(p->p_pptr); 583 584 /* 585 * Notify parent, so in case he was wait(2)ing or 586 * executing waitpid(2) with our pid, he will 587 * continue. 588 */ 589 wakeup(pp); 590 } else 591 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 592 593 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 594 childproc_exited(p); 595 else if (p->p_sigparent != 0) { 596 if (p->p_sigparent == SIGCHLD) 597 childproc_exited(p); 598 else /* LINUX thread */ 599 kern_psignal(p->p_pptr, p->p_sigparent); 600 } 601 } else 602 PROC_LOCK(p->p_pptr); 603 sx_xunlock(&proctree_lock); 604 605 /* 606 * The state PRS_ZOMBIE prevents other proesses from sending 607 * signal to the process, to avoid memory leak, we free memory 608 * for signal queue at the time when the state is set. 609 */ 610 sigqueue_flush(&p->p_sigqueue); 611 sigqueue_flush(&td->td_sigqueue); 612 613 /* 614 * We have to wait until after acquiring all locks before 615 * changing p_state. We need to avoid all possible context 616 * switches (including ones from blocking on a mutex) while 617 * marked as a zombie. We also have to set the zombie state 618 * before we release the parent process' proc lock to avoid 619 * a lost wakeup. So, we first call wakeup, then we grab the 620 * sched lock, update the state, and release the parent process' 621 * proc lock. 622 */ 623 wakeup(p->p_pptr); 624 cv_broadcast(&p->p_pwait); 625 sched_exit(p->p_pptr, td); 626 PROC_SLOCK(p); 627 p->p_state = PRS_ZOMBIE; 628 PROC_UNLOCK(p->p_pptr); 629 630 /* 631 * Save our children's rusage information in our exit rusage. 632 */ 633 PROC_STATLOCK(p); 634 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 635 PROC_STATUNLOCK(p); 636 637 /* 638 * Make sure the scheduler takes this thread out of its tables etc. 639 * This will also release this thread's reference to the ucred. 640 * Other thread parts to release include pcb bits and such. 641 */ 642 thread_exit(); 643 } 644 645 646 #ifndef _SYS_SYSPROTO_H_ 647 struct abort2_args { 648 char *why; 649 int nargs; 650 void **args; 651 }; 652 #endif 653 654 int 655 sys_abort2(struct thread *td, struct abort2_args *uap) 656 { 657 struct proc *p = td->td_proc; 658 struct sbuf *sb; 659 void *uargs[16]; 660 int error, i, sig; 661 662 /* 663 * Do it right now so we can log either proper call of abort2(), or 664 * note, that invalid argument was passed. 512 is big enough to 665 * handle 16 arguments' descriptions with additional comments. 666 */ 667 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 668 sbuf_clear(sb); 669 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 670 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 671 /* 672 * Since we can't return from abort2(), send SIGKILL in cases, where 673 * abort2() was called improperly 674 */ 675 sig = SIGKILL; 676 /* Prevent from DoSes from user-space. */ 677 if (uap->nargs < 0 || uap->nargs > 16) 678 goto out; 679 if (uap->nargs > 0) { 680 if (uap->args == NULL) 681 goto out; 682 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 683 if (error != 0) 684 goto out; 685 } 686 /* 687 * Limit size of 'reason' string to 128. Will fit even when 688 * maximal number of arguments was chosen to be logged. 689 */ 690 if (uap->why != NULL) { 691 error = sbuf_copyin(sb, uap->why, 128); 692 if (error < 0) 693 goto out; 694 } else { 695 sbuf_printf(sb, "(null)"); 696 } 697 if (uap->nargs > 0) { 698 sbuf_printf(sb, "("); 699 for (i = 0;i < uap->nargs; i++) 700 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 701 sbuf_printf(sb, ")"); 702 } 703 /* 704 * Final stage: arguments were proper, string has been 705 * successfully copied from userspace, and copying pointers 706 * from user-space succeed. 707 */ 708 sig = SIGABRT; 709 out: 710 if (sig == SIGKILL) { 711 sbuf_trim(sb); 712 sbuf_printf(sb, " (Reason text inaccessible)"); 713 } 714 sbuf_cat(sb, "\n"); 715 sbuf_finish(sb); 716 log(LOG_INFO, "%s", sbuf_data(sb)); 717 sbuf_delete(sb); 718 exit1(td, 0, sig); 719 return (0); 720 } 721 722 723 #ifdef COMPAT_43 724 /* 725 * The dirty work is handled by kern_wait(). 726 */ 727 int 728 owait(struct thread *td, struct owait_args *uap __unused) 729 { 730 int error, status; 731 732 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 733 if (error == 0) 734 td->td_retval[1] = status; 735 return (error); 736 } 737 #endif /* COMPAT_43 */ 738 739 /* 740 * The dirty work is handled by kern_wait(). 741 */ 742 int 743 sys_wait4(struct thread *td, struct wait4_args *uap) 744 { 745 struct rusage ru, *rup; 746 int error, status; 747 748 if (uap->rusage != NULL) 749 rup = &ru; 750 else 751 rup = NULL; 752 error = kern_wait(td, uap->pid, &status, uap->options, rup); 753 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 754 error = copyout(&status, uap->status, sizeof(status)); 755 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 756 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 757 return (error); 758 } 759 760 int 761 sys_wait6(struct thread *td, struct wait6_args *uap) 762 { 763 struct __wrusage wru, *wrup; 764 siginfo_t si, *sip; 765 idtype_t idtype; 766 id_t id; 767 int error, status; 768 769 idtype = uap->idtype; 770 id = uap->id; 771 772 if (uap->wrusage != NULL) 773 wrup = &wru; 774 else 775 wrup = NULL; 776 777 if (uap->info != NULL) { 778 sip = &si; 779 bzero(sip, sizeof(*sip)); 780 } else 781 sip = NULL; 782 783 /* 784 * We expect all callers of wait6() to know about WEXITED and 785 * WTRAPPED. 786 */ 787 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 788 789 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 790 error = copyout(&status, uap->status, sizeof(status)); 791 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 792 error = copyout(&wru, uap->wrusage, sizeof(wru)); 793 if (uap->info != NULL && error == 0) 794 error = copyout(&si, uap->info, sizeof(si)); 795 return (error); 796 } 797 798 /* 799 * Reap the remains of a zombie process and optionally return status and 800 * rusage. Asserts and will release both the proctree_lock and the process 801 * lock as part of its work. 802 */ 803 void 804 proc_reap(struct thread *td, struct proc *p, int *status, int options) 805 { 806 struct proc *q, *t; 807 808 sx_assert(&proctree_lock, SA_XLOCKED); 809 PROC_LOCK_ASSERT(p, MA_OWNED); 810 PROC_SLOCK_ASSERT(p, MA_OWNED); 811 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 812 813 q = td->td_proc; 814 815 PROC_SUNLOCK(p); 816 if (status) 817 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 818 if (options & WNOWAIT) { 819 /* 820 * Only poll, returning the status. Caller does not wish to 821 * release the proc struct just yet. 822 */ 823 PROC_UNLOCK(p); 824 sx_xunlock(&proctree_lock); 825 return; 826 } 827 828 PROC_LOCK(q); 829 sigqueue_take(p->p_ksi); 830 PROC_UNLOCK(q); 831 832 /* 833 * If we got the child via a ptrace 'attach', we need to give it back 834 * to the old parent. 835 */ 836 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 837 PROC_UNLOCK(p); 838 t = proc_realparent(p); 839 PROC_LOCK(t); 840 PROC_LOCK(p); 841 CTR2(KTR_PTRACE, 842 "wait: traced child %d moved back to parent %d", p->p_pid, 843 t->p_pid); 844 proc_reparent(p, t); 845 p->p_oppid = 0; 846 PROC_UNLOCK(p); 847 pksignal(t, SIGCHLD, p->p_ksi); 848 wakeup(t); 849 cv_broadcast(&p->p_pwait); 850 PROC_UNLOCK(t); 851 sx_xunlock(&proctree_lock); 852 return; 853 } 854 p->p_oppid = 0; 855 PROC_UNLOCK(p); 856 857 /* 858 * Remove other references to this process to ensure we have an 859 * exclusive reference. 860 */ 861 sx_xlock(&allproc_lock); 862 LIST_REMOVE(p, p_list); /* off zombproc */ 863 sx_xunlock(&allproc_lock); 864 LIST_REMOVE(p, p_sibling); 865 reaper_abandon_children(p, true); 866 LIST_REMOVE(p, p_reapsibling); 867 PROC_LOCK(p); 868 clear_orphan(p); 869 PROC_UNLOCK(p); 870 leavepgrp(p); 871 if (p->p_procdesc != NULL) 872 procdesc_reap(p); 873 sx_xunlock(&proctree_lock); 874 875 PROC_LOCK(p); 876 knlist_detach(p->p_klist); 877 p->p_klist = NULL; 878 PROC_UNLOCK(p); 879 880 /* 881 * Removal from allproc list and process group list paired with 882 * PROC_LOCK which was executed during that time should guarantee 883 * nothing can reach this process anymore. As such further locking 884 * is unnecessary. 885 */ 886 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 887 888 PROC_LOCK(q); 889 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 890 PROC_UNLOCK(q); 891 892 /* 893 * Decrement the count of procs running with this uid. 894 */ 895 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 896 897 /* 898 * Destroy resource accounting information associated with the process. 899 */ 900 #ifdef RACCT 901 if (racct_enable) { 902 PROC_LOCK(p); 903 racct_sub(p, RACCT_NPROC, 1); 904 PROC_UNLOCK(p); 905 } 906 #endif 907 racct_proc_exit(p); 908 909 /* 910 * Free credentials, arguments, and sigacts. 911 */ 912 crfree(p->p_ucred); 913 proc_set_cred(p, NULL); 914 pargs_drop(p->p_args); 915 p->p_args = NULL; 916 sigacts_free(p->p_sigacts); 917 p->p_sigacts = NULL; 918 919 /* 920 * Do any thread-system specific cleanups. 921 */ 922 thread_wait(p); 923 924 /* 925 * Give vm and machine-dependent layer a chance to free anything that 926 * cpu_exit couldn't release while still running in process context. 927 */ 928 vm_waitproc(p); 929 #ifdef MAC 930 mac_proc_destroy(p); 931 #endif 932 /* 933 * Free any domain policy that's still hiding around. 934 */ 935 vm_domain_policy_cleanup(&p->p_vm_dom_policy); 936 937 KASSERT(FIRST_THREAD_IN_PROC(p), 938 ("proc_reap: no residual thread!")); 939 uma_zfree(proc_zone, p); 940 atomic_add_int(&nprocs, -1); 941 } 942 943 static int 944 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 945 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 946 int check_only) 947 { 948 struct rusage *rup; 949 950 sx_assert(&proctree_lock, SA_XLOCKED); 951 952 PROC_LOCK(p); 953 954 switch (idtype) { 955 case P_ALL: 956 if (p->p_procdesc != NULL) { 957 PROC_UNLOCK(p); 958 return (0); 959 } 960 break; 961 case P_PID: 962 if (p->p_pid != (pid_t)id) { 963 PROC_UNLOCK(p); 964 return (0); 965 } 966 break; 967 case P_PGID: 968 if (p->p_pgid != (pid_t)id) { 969 PROC_UNLOCK(p); 970 return (0); 971 } 972 break; 973 case P_SID: 974 if (p->p_session->s_sid != (pid_t)id) { 975 PROC_UNLOCK(p); 976 return (0); 977 } 978 break; 979 case P_UID: 980 if (p->p_ucred->cr_uid != (uid_t)id) { 981 PROC_UNLOCK(p); 982 return (0); 983 } 984 break; 985 case P_GID: 986 if (p->p_ucred->cr_gid != (gid_t)id) { 987 PROC_UNLOCK(p); 988 return (0); 989 } 990 break; 991 case P_JAILID: 992 if (p->p_ucred->cr_prison->pr_id != (int)id) { 993 PROC_UNLOCK(p); 994 return (0); 995 } 996 break; 997 /* 998 * It seems that the thread structures get zeroed out 999 * at process exit. This makes it impossible to 1000 * support P_SETID, P_CID or P_CPUID. 1001 */ 1002 default: 1003 PROC_UNLOCK(p); 1004 return (0); 1005 } 1006 1007 if (p_canwait(td, p)) { 1008 PROC_UNLOCK(p); 1009 return (0); 1010 } 1011 1012 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1013 PROC_UNLOCK(p); 1014 return (0); 1015 } 1016 1017 /* 1018 * This special case handles a kthread spawned by linux_clone 1019 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1020 * functions need to be able to distinguish between waiting 1021 * on a process and waiting on a thread. It is a thread if 1022 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1023 * signifies we want to wait for threads and not processes. 1024 */ 1025 if ((p->p_sigparent != SIGCHLD) ^ 1026 ((options & WLINUXCLONE) != 0)) { 1027 PROC_UNLOCK(p); 1028 return (0); 1029 } 1030 1031 if (siginfo != NULL) { 1032 bzero(siginfo, sizeof(*siginfo)); 1033 siginfo->si_errno = 0; 1034 1035 /* 1036 * SUSv4 requires that the si_signo value is always 1037 * SIGCHLD. Obey it despite the rfork(2) interface 1038 * allows to request other signal for child exit 1039 * notification. 1040 */ 1041 siginfo->si_signo = SIGCHLD; 1042 1043 /* 1044 * This is still a rough estimate. We will fix the 1045 * cases TRAPPED, STOPPED, and CONTINUED later. 1046 */ 1047 if (WCOREDUMP(p->p_xsig)) { 1048 siginfo->si_code = CLD_DUMPED; 1049 siginfo->si_status = WTERMSIG(p->p_xsig); 1050 } else if (WIFSIGNALED(p->p_xsig)) { 1051 siginfo->si_code = CLD_KILLED; 1052 siginfo->si_status = WTERMSIG(p->p_xsig); 1053 } else { 1054 siginfo->si_code = CLD_EXITED; 1055 siginfo->si_status = p->p_xexit; 1056 } 1057 1058 siginfo->si_pid = p->p_pid; 1059 siginfo->si_uid = p->p_ucred->cr_uid; 1060 1061 /* 1062 * The si_addr field would be useful additional 1063 * detail, but apparently the PC value may be lost 1064 * when we reach this point. bzero() above sets 1065 * siginfo->si_addr to NULL. 1066 */ 1067 } 1068 1069 /* 1070 * There should be no reason to limit resources usage info to 1071 * exited processes only. A snapshot about any resources used 1072 * by a stopped process may be exactly what is needed. 1073 */ 1074 if (wrusage != NULL) { 1075 rup = &wrusage->wru_self; 1076 *rup = p->p_ru; 1077 PROC_STATLOCK(p); 1078 calcru(p, &rup->ru_utime, &rup->ru_stime); 1079 PROC_STATUNLOCK(p); 1080 1081 rup = &wrusage->wru_children; 1082 *rup = p->p_stats->p_cru; 1083 calccru(p, &rup->ru_utime, &rup->ru_stime); 1084 } 1085 1086 if (p->p_state == PRS_ZOMBIE && !check_only) { 1087 PROC_SLOCK(p); 1088 proc_reap(td, p, status, options); 1089 return (-1); 1090 } 1091 return (1); 1092 } 1093 1094 int 1095 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1096 struct rusage *rusage) 1097 { 1098 struct __wrusage wru, *wrup; 1099 idtype_t idtype; 1100 id_t id; 1101 int ret; 1102 1103 /* 1104 * Translate the special pid values into the (idtype, pid) 1105 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1106 * kern_wait6() on its own. 1107 */ 1108 if (pid == WAIT_ANY) { 1109 idtype = P_ALL; 1110 id = 0; 1111 } else if (pid < 0) { 1112 idtype = P_PGID; 1113 id = (id_t)-pid; 1114 } else { 1115 idtype = P_PID; 1116 id = (id_t)pid; 1117 } 1118 1119 if (rusage != NULL) 1120 wrup = &wru; 1121 else 1122 wrup = NULL; 1123 1124 /* 1125 * For backward compatibility we implicitly add flags WEXITED 1126 * and WTRAPPED here. 1127 */ 1128 options |= WEXITED | WTRAPPED; 1129 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1130 if (rusage != NULL) 1131 *rusage = wru.wru_self; 1132 return (ret); 1133 } 1134 1135 static void 1136 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1137 int *status, int options, int si_code) 1138 { 1139 bool cont; 1140 1141 PROC_LOCK_ASSERT(p, MA_OWNED); 1142 sx_assert(&proctree_lock, SA_XLOCKED); 1143 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1144 si_code == CLD_CONTINUED); 1145 1146 cont = si_code == CLD_CONTINUED; 1147 if ((options & WNOWAIT) == 0) { 1148 if (cont) 1149 p->p_flag &= ~P_CONTINUED; 1150 else 1151 p->p_flag |= P_WAITED; 1152 PROC_LOCK(td->td_proc); 1153 sigqueue_take(p->p_ksi); 1154 PROC_UNLOCK(td->td_proc); 1155 } 1156 sx_xunlock(&proctree_lock); 1157 if (siginfo != NULL) { 1158 siginfo->si_code = si_code; 1159 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1160 } 1161 if (status != NULL) 1162 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1163 PROC_UNLOCK(p); 1164 td->td_retval[0] = p->p_pid; 1165 } 1166 1167 int 1168 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1169 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1170 { 1171 struct proc *p, *q; 1172 pid_t pid; 1173 int error, nfound, ret; 1174 1175 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1176 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1177 AUDIT_ARG_VALUE(options); 1178 1179 q = td->td_proc; 1180 1181 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1182 PROC_LOCK(q); 1183 id = (id_t)q->p_pgid; 1184 PROC_UNLOCK(q); 1185 idtype = P_PGID; 1186 } 1187 1188 /* If we don't know the option, just return. */ 1189 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1190 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1191 return (EINVAL); 1192 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1193 /* 1194 * We will be unable to find any matching processes, 1195 * because there are no known events to look for. 1196 * Prefer to return error instead of blocking 1197 * indefinitely. 1198 */ 1199 return (EINVAL); 1200 } 1201 1202 loop: 1203 if (q->p_flag & P_STATCHILD) { 1204 PROC_LOCK(q); 1205 q->p_flag &= ~P_STATCHILD; 1206 PROC_UNLOCK(q); 1207 } 1208 nfound = 0; 1209 sx_xlock(&proctree_lock); 1210 LIST_FOREACH(p, &q->p_children, p_sibling) { 1211 pid = p->p_pid; 1212 ret = proc_to_reap(td, p, idtype, id, status, options, 1213 wrusage, siginfo, 0); 1214 if (ret == 0) 1215 continue; 1216 else if (ret == 1) 1217 nfound++; 1218 else { 1219 td->td_retval[0] = pid; 1220 return (0); 1221 } 1222 1223 PROC_LOCK_ASSERT(p, MA_OWNED); 1224 1225 if ((options & (WTRAPPED | WUNTRACED)) != 0) 1226 PROC_SLOCK(p); 1227 1228 if ((options & WTRAPPED) != 0 && 1229 (p->p_flag & P_TRACED) != 0 && 1230 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1231 p->p_suspcount == p->p_numthreads && 1232 (p->p_flag & P_WAITED) == 0) { 1233 PROC_SUNLOCK(p); 1234 CTR4(KTR_PTRACE, 1235 "wait: returning trapped pid %d status %#x " 1236 "(xstat %d) xthread %d", 1237 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1238 p->p_xthread != NULL ? 1239 p->p_xthread->td_tid : -1); 1240 report_alive_proc(td, p, siginfo, status, options, 1241 CLD_TRAPPED); 1242 return (0); 1243 } 1244 if ((options & WUNTRACED) != 0 && 1245 (p->p_flag & P_STOPPED_SIG) != 0 && 1246 p->p_suspcount == p->p_numthreads && 1247 (p->p_flag & P_WAITED) == 0) { 1248 PROC_SUNLOCK(p); 1249 report_alive_proc(td, p, siginfo, status, options, 1250 CLD_STOPPED); 1251 return (0); 1252 } 1253 if ((options & (WTRAPPED | WUNTRACED)) != 0) 1254 PROC_SUNLOCK(p); 1255 if ((options & WCONTINUED) != 0 && 1256 (p->p_flag & P_CONTINUED) != 0) { 1257 report_alive_proc(td, p, siginfo, status, options, 1258 CLD_CONTINUED); 1259 return (0); 1260 } 1261 PROC_UNLOCK(p); 1262 } 1263 1264 /* 1265 * Look in the orphans list too, to allow the parent to 1266 * collect it's child exit status even if child is being 1267 * debugged. 1268 * 1269 * Debugger detaches from the parent upon successful 1270 * switch-over from parent to child. At this point due to 1271 * re-parenting the parent loses the child to debugger and a 1272 * wait4(2) call would report that it has no children to wait 1273 * for. By maintaining a list of orphans we allow the parent 1274 * to successfully wait until the child becomes a zombie. 1275 */ 1276 if (nfound == 0) { 1277 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1278 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1279 NULL, NULL, 1); 1280 if (ret != 0) { 1281 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1282 (int)td->td_retval[0])); 1283 PROC_UNLOCK(p); 1284 nfound++; 1285 break; 1286 } 1287 } 1288 } 1289 if (nfound == 0) { 1290 sx_xunlock(&proctree_lock); 1291 return (ECHILD); 1292 } 1293 if (options & WNOHANG) { 1294 sx_xunlock(&proctree_lock); 1295 td->td_retval[0] = 0; 1296 return (0); 1297 } 1298 PROC_LOCK(q); 1299 sx_xunlock(&proctree_lock); 1300 if (q->p_flag & P_STATCHILD) { 1301 q->p_flag &= ~P_STATCHILD; 1302 error = 0; 1303 } else 1304 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1305 PROC_UNLOCK(q); 1306 if (error) 1307 return (error); 1308 goto loop; 1309 } 1310 1311 /* 1312 * Make process 'parent' the new parent of process 'child'. 1313 * Must be called with an exclusive hold of proctree lock. 1314 */ 1315 void 1316 proc_reparent(struct proc *child, struct proc *parent) 1317 { 1318 1319 sx_assert(&proctree_lock, SX_XLOCKED); 1320 PROC_LOCK_ASSERT(child, MA_OWNED); 1321 if (child->p_pptr == parent) 1322 return; 1323 1324 PROC_LOCK(child->p_pptr); 1325 sigqueue_take(child->p_ksi); 1326 PROC_UNLOCK(child->p_pptr); 1327 LIST_REMOVE(child, p_sibling); 1328 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1329 1330 clear_orphan(child); 1331 if (child->p_flag & P_TRACED) { 1332 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1333 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1334 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1335 p_orphan); 1336 } else { 1337 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1338 child, p_orphan); 1339 } 1340 child->p_treeflag |= P_TREE_ORPHANED; 1341 } 1342 1343 child->p_pptr = parent; 1344 } 1345