1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_compat.h" 43 #include "opt_ktrace.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/sysproto.h> 48 #include <sys/capsicum.h> 49 #include <sys/eventhandler.h> 50 #include <sys/kernel.h> 51 #include <sys/malloc.h> 52 #include <sys/lock.h> 53 #include <sys/mutex.h> 54 #include <sys/proc.h> 55 #include <sys/procdesc.h> 56 #include <sys/pioctl.h> 57 #include <sys/jail.h> 58 #include <sys/tty.h> 59 #include <sys/wait.h> 60 #include <sys/vmmeter.h> 61 #include <sys/vnode.h> 62 #include <sys/racct.h> 63 #include <sys/resourcevar.h> 64 #include <sys/sbuf.h> 65 #include <sys/signalvar.h> 66 #include <sys/sched.h> 67 #include <sys/sx.h> 68 #include <sys/syscallsubr.h> 69 #include <sys/syslog.h> 70 #include <sys/ptrace.h> 71 #include <sys/acct.h> /* for acct_process() function prototype */ 72 #include <sys/filedesc.h> 73 #include <sys/sdt.h> 74 #include <sys/shm.h> 75 #include <sys/sem.h> 76 #include <sys/umtx.h> 77 #ifdef KTRACE 78 #include <sys/ktrace.h> 79 #endif 80 81 #include <security/audit/audit.h> 82 #include <security/mac/mac_framework.h> 83 84 #include <vm/vm.h> 85 #include <vm/vm_extern.h> 86 #include <vm/vm_param.h> 87 #include <vm/pmap.h> 88 #include <vm/vm_map.h> 89 #include <vm/vm_page.h> 90 #include <vm/uma.h> 91 #include <vm/vm_domain.h> 92 93 #ifdef KDTRACE_HOOKS 94 #include <sys/dtrace_bsd.h> 95 dtrace_execexit_func_t dtrace_fasttrap_exit; 96 #endif 97 98 SDT_PROVIDER_DECLARE(proc); 99 SDT_PROBE_DEFINE1(proc, , , exit, "int"); 100 101 /* Hook for NFS teardown procedure. */ 102 void (*nlminfo_release_p)(struct proc *p); 103 104 EVENTHANDLER_LIST_DECLARE(process_exit); 105 106 struct proc * 107 proc_realparent(struct proc *child) 108 { 109 struct proc *p, *parent; 110 111 sx_assert(&proctree_lock, SX_LOCKED); 112 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 113 if (child->p_oppid == 0 || 114 child->p_pptr->p_pid == child->p_oppid) 115 parent = child->p_pptr; 116 else 117 parent = initproc; 118 return (parent); 119 } 120 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 121 /* Cannot use LIST_PREV(), since the list head is not known. */ 122 p = __containerof(p->p_orphan.le_prev, struct proc, 123 p_orphan.le_next); 124 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 125 ("missing P_ORPHAN %p", p)); 126 } 127 parent = __containerof(p->p_orphan.le_prev, struct proc, 128 p_orphans.lh_first); 129 return (parent); 130 } 131 132 void 133 reaper_abandon_children(struct proc *p, bool exiting) 134 { 135 struct proc *p1, *p2, *ptmp; 136 137 sx_assert(&proctree_lock, SX_LOCKED); 138 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 139 if ((p->p_treeflag & P_TREE_REAPER) == 0) 140 return; 141 p1 = p->p_reaper; 142 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 143 LIST_REMOVE(p2, p_reapsibling); 144 p2->p_reaper = p1; 145 p2->p_reapsubtree = p->p_reapsubtree; 146 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 147 if (exiting && p2->p_pptr == p) { 148 PROC_LOCK(p2); 149 proc_reparent(p2, p1); 150 PROC_UNLOCK(p2); 151 } 152 } 153 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 154 p->p_treeflag &= ~P_TREE_REAPER; 155 } 156 157 static void 158 clear_orphan(struct proc *p) 159 { 160 struct proc *p1; 161 162 sx_assert(&proctree_lock, SA_XLOCKED); 163 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 164 return; 165 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 166 p1 = LIST_NEXT(p, p_orphan); 167 if (p1 != NULL) 168 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 169 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 170 } 171 LIST_REMOVE(p, p_orphan); 172 p->p_treeflag &= ~P_TREE_ORPHANED; 173 } 174 175 /* 176 * exit -- death of process. 177 */ 178 void 179 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 180 { 181 182 exit1(td, uap->rval, 0); 183 /* NOTREACHED */ 184 } 185 186 /* 187 * Exit: deallocate address space and other resources, change proc state to 188 * zombie, and unlink proc from allproc and parent's lists. Save exit status 189 * and rusage for wait(). Check for child processes and orphan them. 190 */ 191 void 192 exit1(struct thread *td, int rval, int signo) 193 { 194 struct proc *p, *nq, *q, *t; 195 struct thread *tdt; 196 ksiginfo_t *ksi, *ksi1; 197 198 mtx_assert(&Giant, MA_NOTOWNED); 199 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 200 201 p = td->td_proc; 202 /* 203 * XXX in case we're rebooting we just let init die in order to 204 * work around an unsolved stack overflow seen very late during 205 * shutdown on sparc64 when the gmirror worker process exists. 206 */ 207 if (p == initproc && rebooting == 0) { 208 printf("init died (signal %d, exit %d)\n", signo, rval); 209 panic("Going nowhere without my init!"); 210 } 211 212 /* 213 * Deref SU mp, since the thread does not return to userspace. 214 */ 215 td_softdep_cleanup(td); 216 217 /* 218 * MUST abort all other threads before proceeding past here. 219 */ 220 PROC_LOCK(p); 221 /* 222 * First check if some other thread or external request got 223 * here before us. If so, act appropriately: exit or suspend. 224 * We must ensure that stop requests are handled before we set 225 * P_WEXIT. 226 */ 227 thread_suspend_check(0); 228 while (p->p_flag & P_HADTHREADS) { 229 /* 230 * Kill off the other threads. This requires 231 * some co-operation from other parts of the kernel 232 * so it may not be instantaneous. With this state set 233 * any thread entering the kernel from userspace will 234 * thread_exit() in trap(). Any thread attempting to 235 * sleep will return immediately with EINTR or EWOULDBLOCK 236 * which will hopefully force them to back out to userland 237 * freeing resources as they go. Any thread attempting 238 * to return to userland will thread_exit() from userret(). 239 * thread_exit() will unsuspend us when the last of the 240 * other threads exits. 241 * If there is already a thread singler after resumption, 242 * calling thread_single will fail; in that case, we just 243 * re-check all suspension request, the thread should 244 * either be suspended there or exit. 245 */ 246 if (!thread_single(p, SINGLE_EXIT)) 247 /* 248 * All other activity in this process is now 249 * stopped. Threading support has been turned 250 * off. 251 */ 252 break; 253 /* 254 * Recheck for new stop or suspend requests which 255 * might appear while process lock was dropped in 256 * thread_single(). 257 */ 258 thread_suspend_check(0); 259 } 260 KASSERT(p->p_numthreads == 1, 261 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 262 racct_sub(p, RACCT_NTHR, 1); 263 264 /* Let event handler change exit status */ 265 p->p_xexit = rval; 266 p->p_xsig = signo; 267 268 /* 269 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 270 * on our vmspace, so we should block below until they have 271 * released their reference to us. Note that if they have 272 * requested S_EXIT stops we will block here until they ack 273 * via PIOCCONT. 274 */ 275 _STOPEVENT(p, S_EXIT, 0); 276 277 /* 278 * Ignore any pending request to stop due to a stop signal. 279 * Once P_WEXIT is set, future requests will be ignored as 280 * well. 281 */ 282 p->p_flag &= ~P_STOPPED_SIG; 283 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 284 285 /* 286 * Note that we are exiting and do another wakeup of anyone in 287 * PIOCWAIT in case they aren't listening for S_EXIT stops or 288 * decided to wait again after we told them we are exiting. 289 */ 290 p->p_flag |= P_WEXIT; 291 wakeup(&p->p_stype); 292 293 /* 294 * Wait for any processes that have a hold on our vmspace to 295 * release their reference. 296 */ 297 while (p->p_lock > 0) 298 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 299 300 PROC_UNLOCK(p); 301 /* Drain the limit callout while we don't have the proc locked */ 302 callout_drain(&p->p_limco); 303 304 #ifdef AUDIT 305 /* 306 * The Sun BSM exit token contains two components: an exit status as 307 * passed to exit(), and a return value to indicate what sort of exit 308 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 309 * what the return value is. 310 */ 311 AUDIT_ARG_EXIT(rval, 0); 312 AUDIT_SYSCALL_EXIT(0, td); 313 #endif 314 315 /* Are we a task leader with peers? */ 316 if (p->p_peers != NULL && p == p->p_leader) { 317 mtx_lock(&ppeers_lock); 318 q = p->p_peers; 319 while (q != NULL) { 320 PROC_LOCK(q); 321 kern_psignal(q, SIGKILL); 322 PROC_UNLOCK(q); 323 q = q->p_peers; 324 } 325 while (p->p_peers != NULL) 326 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 327 mtx_unlock(&ppeers_lock); 328 } 329 330 /* 331 * Check if any loadable modules need anything done at process exit. 332 * E.g. SYSV IPC stuff. 333 * Event handler could change exit status. 334 * XXX what if one of these generates an error? 335 */ 336 EVENTHANDLER_DIRECT_INVOKE(process_exit, p); 337 338 /* 339 * If parent is waiting for us to exit or exec, 340 * P_PPWAIT is set; we will wakeup the parent below. 341 */ 342 PROC_LOCK(p); 343 stopprofclock(p); 344 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 345 p->p_ptevents = 0; 346 347 /* 348 * Stop the real interval timer. If the handler is currently 349 * executing, prevent it from rearming itself and let it finish. 350 */ 351 if (timevalisset(&p->p_realtimer.it_value) && 352 _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) { 353 timevalclear(&p->p_realtimer.it_interval); 354 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 355 KASSERT(!timevalisset(&p->p_realtimer.it_value), 356 ("realtime timer is still armed")); 357 } 358 359 PROC_UNLOCK(p); 360 361 umtx_thread_exit(td); 362 363 /* 364 * Reset any sigio structures pointing to us as a result of 365 * F_SETOWN with our pid. 366 */ 367 funsetownlst(&p->p_sigiolst); 368 369 /* 370 * If this process has an nlminfo data area (for lockd), release it 371 */ 372 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 373 (*nlminfo_release_p)(p); 374 375 /* 376 * Close open files and release open-file table. 377 * This may block! 378 */ 379 fdescfree(td); 380 381 /* 382 * If this thread tickled GEOM, we need to wait for the giggling to 383 * stop before we return to userland 384 */ 385 if (td->td_pflags & TDP_GEOM) 386 g_waitidle(); 387 388 /* 389 * Remove ourself from our leader's peer list and wake our leader. 390 */ 391 if (p->p_leader->p_peers != NULL) { 392 mtx_lock(&ppeers_lock); 393 if (p->p_leader->p_peers != NULL) { 394 q = p->p_leader; 395 while (q->p_peers != p) 396 q = q->p_peers; 397 q->p_peers = p->p_peers; 398 wakeup(p->p_leader); 399 } 400 mtx_unlock(&ppeers_lock); 401 } 402 403 vmspace_exit(td); 404 killjobc(); 405 (void)acct_process(td); 406 407 #ifdef KTRACE 408 ktrprocexit(td); 409 #endif 410 /* 411 * Release reference to text vnode 412 */ 413 if (p->p_textvp != NULL) { 414 vrele(p->p_textvp); 415 p->p_textvp = NULL; 416 } 417 418 /* 419 * Release our limits structure. 420 */ 421 lim_free(p->p_limit); 422 p->p_limit = NULL; 423 424 tidhash_remove(td); 425 426 /* 427 * Remove proc from allproc queue and pidhash chain. 428 * Place onto zombproc. Unlink from parent's child list. 429 */ 430 sx_xlock(&allproc_lock); 431 LIST_REMOVE(p, p_list); 432 LIST_INSERT_HEAD(&zombproc, p, p_list); 433 LIST_REMOVE(p, p_hash); 434 sx_xunlock(&allproc_lock); 435 436 /* 437 * Call machine-dependent code to release any 438 * machine-dependent resources other than the address space. 439 * The address space is released by "vmspace_exitfree(p)" in 440 * vm_waitproc(). 441 */ 442 cpu_exit(td); 443 444 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 445 446 /* 447 * Reparent all children processes: 448 * - traced ones to the original parent (or init if we are that parent) 449 * - the rest to init 450 */ 451 sx_xlock(&proctree_lock); 452 q = LIST_FIRST(&p->p_children); 453 if (q != NULL) /* only need this if any child is S_ZOMB */ 454 wakeup(q->p_reaper); 455 for (; q != NULL; q = nq) { 456 nq = LIST_NEXT(q, p_sibling); 457 ksi = ksiginfo_alloc(TRUE); 458 PROC_LOCK(q); 459 q->p_sigparent = SIGCHLD; 460 461 if (!(q->p_flag & P_TRACED)) { 462 proc_reparent(q, q->p_reaper); 463 if (q->p_state == PRS_ZOMBIE) { 464 /* 465 * Inform reaper about the reparented 466 * zombie, since wait(2) has something 467 * new to report. Guarantee queueing 468 * of the SIGCHLD signal, similar to 469 * the _exit() behaviour, by providing 470 * our ksiginfo. Ksi is freed by the 471 * signal delivery. 472 */ 473 if (q->p_ksi == NULL) { 474 ksi1 = NULL; 475 } else { 476 ksiginfo_copy(q->p_ksi, ksi); 477 ksi->ksi_flags |= KSI_INS; 478 ksi1 = ksi; 479 ksi = NULL; 480 } 481 PROC_LOCK(q->p_reaper); 482 pksignal(q->p_reaper, SIGCHLD, ksi1); 483 PROC_UNLOCK(q->p_reaper); 484 } 485 } else { 486 /* 487 * Traced processes are killed since their existence 488 * means someone is screwing up. 489 */ 490 t = proc_realparent(q); 491 if (t == p) { 492 proc_reparent(q, q->p_reaper); 493 } else { 494 PROC_LOCK(t); 495 proc_reparent(q, t); 496 PROC_UNLOCK(t); 497 } 498 /* 499 * Since q was found on our children list, the 500 * proc_reparent() call moved q to the orphan 501 * list due to present P_TRACED flag. Clear 502 * orphan link for q now while q is locked. 503 */ 504 clear_orphan(q); 505 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 506 q->p_flag2 &= ~P2_PTRACE_FSTP; 507 q->p_ptevents = 0; 508 FOREACH_THREAD_IN_PROC(q, tdt) { 509 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 510 TDB_FSTP); 511 } 512 kern_psignal(q, SIGKILL); 513 } 514 PROC_UNLOCK(q); 515 if (ksi != NULL) 516 ksiginfo_free(ksi); 517 } 518 519 /* 520 * Also get rid of our orphans. 521 */ 522 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 523 PROC_LOCK(q); 524 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 525 q->p_pid); 526 clear_orphan(q); 527 PROC_UNLOCK(q); 528 } 529 530 /* Save exit status. */ 531 PROC_LOCK(p); 532 p->p_xthread = td; 533 534 /* Tell the prison that we are gone. */ 535 prison_proc_free(p->p_ucred->cr_prison); 536 537 #ifdef KDTRACE_HOOKS 538 /* 539 * Tell the DTrace fasttrap provider about the exit if it 540 * has declared an interest. 541 */ 542 if (dtrace_fasttrap_exit) 543 dtrace_fasttrap_exit(p); 544 #endif 545 546 /* 547 * Notify interested parties of our demise. 548 */ 549 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 550 551 #ifdef KDTRACE_HOOKS 552 int reason = CLD_EXITED; 553 if (WCOREDUMP(signo)) 554 reason = CLD_DUMPED; 555 else if (WIFSIGNALED(signo)) 556 reason = CLD_KILLED; 557 SDT_PROBE1(proc, , , exit, reason); 558 #endif 559 560 /* 561 * If this is a process with a descriptor, we may not need to deliver 562 * a signal to the parent. proctree_lock is held over 563 * procdesc_exit() to serialize concurrent calls to close() and 564 * exit(). 565 */ 566 if (p->p_procdesc == NULL || procdesc_exit(p)) { 567 /* 568 * Notify parent that we're gone. If parent has the 569 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 570 * notify process 1 instead (and hope it will handle this 571 * situation). 572 */ 573 PROC_LOCK(p->p_pptr); 574 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 575 if (p->p_pptr->p_sigacts->ps_flag & 576 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 577 struct proc *pp; 578 579 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 580 pp = p->p_pptr; 581 PROC_UNLOCK(pp); 582 proc_reparent(p, p->p_reaper); 583 p->p_sigparent = SIGCHLD; 584 PROC_LOCK(p->p_pptr); 585 586 /* 587 * Notify parent, so in case he was wait(2)ing or 588 * executing waitpid(2) with our pid, he will 589 * continue. 590 */ 591 wakeup(pp); 592 } else 593 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 594 595 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 596 childproc_exited(p); 597 else if (p->p_sigparent != 0) { 598 if (p->p_sigparent == SIGCHLD) 599 childproc_exited(p); 600 else /* LINUX thread */ 601 kern_psignal(p->p_pptr, p->p_sigparent); 602 } 603 } else 604 PROC_LOCK(p->p_pptr); 605 sx_xunlock(&proctree_lock); 606 607 /* 608 * The state PRS_ZOMBIE prevents other proesses from sending 609 * signal to the process, to avoid memory leak, we free memory 610 * for signal queue at the time when the state is set. 611 */ 612 sigqueue_flush(&p->p_sigqueue); 613 sigqueue_flush(&td->td_sigqueue); 614 615 /* 616 * We have to wait until after acquiring all locks before 617 * changing p_state. We need to avoid all possible context 618 * switches (including ones from blocking on a mutex) while 619 * marked as a zombie. We also have to set the zombie state 620 * before we release the parent process' proc lock to avoid 621 * a lost wakeup. So, we first call wakeup, then we grab the 622 * sched lock, update the state, and release the parent process' 623 * proc lock. 624 */ 625 wakeup(p->p_pptr); 626 cv_broadcast(&p->p_pwait); 627 sched_exit(p->p_pptr, td); 628 PROC_SLOCK(p); 629 p->p_state = PRS_ZOMBIE; 630 PROC_UNLOCK(p->p_pptr); 631 632 /* 633 * Save our children's rusage information in our exit rusage. 634 */ 635 PROC_STATLOCK(p); 636 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 637 PROC_STATUNLOCK(p); 638 639 /* 640 * Make sure the scheduler takes this thread out of its tables etc. 641 * This will also release this thread's reference to the ucred. 642 * Other thread parts to release include pcb bits and such. 643 */ 644 thread_exit(); 645 } 646 647 648 #ifndef _SYS_SYSPROTO_H_ 649 struct abort2_args { 650 char *why; 651 int nargs; 652 void **args; 653 }; 654 #endif 655 656 int 657 sys_abort2(struct thread *td, struct abort2_args *uap) 658 { 659 struct proc *p = td->td_proc; 660 struct sbuf *sb; 661 void *uargs[16]; 662 int error, i, sig; 663 664 /* 665 * Do it right now so we can log either proper call of abort2(), or 666 * note, that invalid argument was passed. 512 is big enough to 667 * handle 16 arguments' descriptions with additional comments. 668 */ 669 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 670 sbuf_clear(sb); 671 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 672 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 673 /* 674 * Since we can't return from abort2(), send SIGKILL in cases, where 675 * abort2() was called improperly 676 */ 677 sig = SIGKILL; 678 /* Prevent from DoSes from user-space. */ 679 if (uap->nargs < 0 || uap->nargs > 16) 680 goto out; 681 if (uap->nargs > 0) { 682 if (uap->args == NULL) 683 goto out; 684 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 685 if (error != 0) 686 goto out; 687 } 688 /* 689 * Limit size of 'reason' string to 128. Will fit even when 690 * maximal number of arguments was chosen to be logged. 691 */ 692 if (uap->why != NULL) { 693 error = sbuf_copyin(sb, uap->why, 128); 694 if (error < 0) 695 goto out; 696 } else { 697 sbuf_printf(sb, "(null)"); 698 } 699 if (uap->nargs > 0) { 700 sbuf_printf(sb, "("); 701 for (i = 0;i < uap->nargs; i++) 702 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 703 sbuf_printf(sb, ")"); 704 } 705 /* 706 * Final stage: arguments were proper, string has been 707 * successfully copied from userspace, and copying pointers 708 * from user-space succeed. 709 */ 710 sig = SIGABRT; 711 out: 712 if (sig == SIGKILL) { 713 sbuf_trim(sb); 714 sbuf_printf(sb, " (Reason text inaccessible)"); 715 } 716 sbuf_cat(sb, "\n"); 717 sbuf_finish(sb); 718 log(LOG_INFO, "%s", sbuf_data(sb)); 719 sbuf_delete(sb); 720 exit1(td, 0, sig); 721 return (0); 722 } 723 724 725 #ifdef COMPAT_43 726 /* 727 * The dirty work is handled by kern_wait(). 728 */ 729 int 730 owait(struct thread *td, struct owait_args *uap __unused) 731 { 732 int error, status; 733 734 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 735 if (error == 0) 736 td->td_retval[1] = status; 737 return (error); 738 } 739 #endif /* COMPAT_43 */ 740 741 /* 742 * The dirty work is handled by kern_wait(). 743 */ 744 int 745 sys_wait4(struct thread *td, struct wait4_args *uap) 746 { 747 struct rusage ru, *rup; 748 int error, status; 749 750 if (uap->rusage != NULL) 751 rup = &ru; 752 else 753 rup = NULL; 754 error = kern_wait(td, uap->pid, &status, uap->options, rup); 755 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 756 error = copyout(&status, uap->status, sizeof(status)); 757 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 758 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 759 return (error); 760 } 761 762 int 763 sys_wait6(struct thread *td, struct wait6_args *uap) 764 { 765 struct __wrusage wru, *wrup; 766 siginfo_t si, *sip; 767 idtype_t idtype; 768 id_t id; 769 int error, status; 770 771 idtype = uap->idtype; 772 id = uap->id; 773 774 if (uap->wrusage != NULL) 775 wrup = &wru; 776 else 777 wrup = NULL; 778 779 if (uap->info != NULL) { 780 sip = &si; 781 bzero(sip, sizeof(*sip)); 782 } else 783 sip = NULL; 784 785 /* 786 * We expect all callers of wait6() to know about WEXITED and 787 * WTRAPPED. 788 */ 789 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 790 791 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 792 error = copyout(&status, uap->status, sizeof(status)); 793 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 794 error = copyout(&wru, uap->wrusage, sizeof(wru)); 795 if (uap->info != NULL && error == 0) 796 error = copyout(&si, uap->info, sizeof(si)); 797 return (error); 798 } 799 800 /* 801 * Reap the remains of a zombie process and optionally return status and 802 * rusage. Asserts and will release both the proctree_lock and the process 803 * lock as part of its work. 804 */ 805 void 806 proc_reap(struct thread *td, struct proc *p, int *status, int options) 807 { 808 struct proc *q, *t; 809 810 sx_assert(&proctree_lock, SA_XLOCKED); 811 PROC_LOCK_ASSERT(p, MA_OWNED); 812 PROC_SLOCK_ASSERT(p, MA_OWNED); 813 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 814 815 q = td->td_proc; 816 817 PROC_SUNLOCK(p); 818 if (status) 819 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 820 if (options & WNOWAIT) { 821 /* 822 * Only poll, returning the status. Caller does not wish to 823 * release the proc struct just yet. 824 */ 825 PROC_UNLOCK(p); 826 sx_xunlock(&proctree_lock); 827 return; 828 } 829 830 PROC_LOCK(q); 831 sigqueue_take(p->p_ksi); 832 PROC_UNLOCK(q); 833 834 /* 835 * If we got the child via a ptrace 'attach', we need to give it back 836 * to the old parent. 837 */ 838 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 839 PROC_UNLOCK(p); 840 t = proc_realparent(p); 841 PROC_LOCK(t); 842 PROC_LOCK(p); 843 CTR2(KTR_PTRACE, 844 "wait: traced child %d moved back to parent %d", p->p_pid, 845 t->p_pid); 846 proc_reparent(p, t); 847 p->p_oppid = 0; 848 PROC_UNLOCK(p); 849 pksignal(t, SIGCHLD, p->p_ksi); 850 wakeup(t); 851 cv_broadcast(&p->p_pwait); 852 PROC_UNLOCK(t); 853 sx_xunlock(&proctree_lock); 854 return; 855 } 856 p->p_oppid = 0; 857 PROC_UNLOCK(p); 858 859 /* 860 * Remove other references to this process to ensure we have an 861 * exclusive reference. 862 */ 863 sx_xlock(&allproc_lock); 864 LIST_REMOVE(p, p_list); /* off zombproc */ 865 sx_xunlock(&allproc_lock); 866 LIST_REMOVE(p, p_sibling); 867 reaper_abandon_children(p, true); 868 LIST_REMOVE(p, p_reapsibling); 869 PROC_LOCK(p); 870 clear_orphan(p); 871 PROC_UNLOCK(p); 872 leavepgrp(p); 873 if (p->p_procdesc != NULL) 874 procdesc_reap(p); 875 sx_xunlock(&proctree_lock); 876 877 PROC_LOCK(p); 878 knlist_detach(p->p_klist); 879 p->p_klist = NULL; 880 PROC_UNLOCK(p); 881 882 /* 883 * Removal from allproc list and process group list paired with 884 * PROC_LOCK which was executed during that time should guarantee 885 * nothing can reach this process anymore. As such further locking 886 * is unnecessary. 887 */ 888 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 889 890 PROC_LOCK(q); 891 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 892 PROC_UNLOCK(q); 893 894 /* 895 * Decrement the count of procs running with this uid. 896 */ 897 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 898 899 /* 900 * Destroy resource accounting information associated with the process. 901 */ 902 #ifdef RACCT 903 if (racct_enable) { 904 PROC_LOCK(p); 905 racct_sub(p, RACCT_NPROC, 1); 906 PROC_UNLOCK(p); 907 } 908 #endif 909 racct_proc_exit(p); 910 911 /* 912 * Free credentials, arguments, and sigacts. 913 */ 914 crfree(p->p_ucred); 915 proc_set_cred(p, NULL); 916 pargs_drop(p->p_args); 917 p->p_args = NULL; 918 sigacts_free(p->p_sigacts); 919 p->p_sigacts = NULL; 920 921 /* 922 * Do any thread-system specific cleanups. 923 */ 924 thread_wait(p); 925 926 /* 927 * Give vm and machine-dependent layer a chance to free anything that 928 * cpu_exit couldn't release while still running in process context. 929 */ 930 vm_waitproc(p); 931 #ifdef MAC 932 mac_proc_destroy(p); 933 #endif 934 /* 935 * Free any domain policy that's still hiding around. 936 */ 937 vm_domain_policy_cleanup(&p->p_vm_dom_policy); 938 939 KASSERT(FIRST_THREAD_IN_PROC(p), 940 ("proc_reap: no residual thread!")); 941 uma_zfree(proc_zone, p); 942 atomic_add_int(&nprocs, -1); 943 } 944 945 static int 946 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 947 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 948 int check_only) 949 { 950 struct rusage *rup; 951 952 sx_assert(&proctree_lock, SA_XLOCKED); 953 954 PROC_LOCK(p); 955 956 switch (idtype) { 957 case P_ALL: 958 if (p->p_procdesc != NULL) { 959 PROC_UNLOCK(p); 960 return (0); 961 } 962 break; 963 case P_PID: 964 if (p->p_pid != (pid_t)id) { 965 PROC_UNLOCK(p); 966 return (0); 967 } 968 break; 969 case P_PGID: 970 if (p->p_pgid != (pid_t)id) { 971 PROC_UNLOCK(p); 972 return (0); 973 } 974 break; 975 case P_SID: 976 if (p->p_session->s_sid != (pid_t)id) { 977 PROC_UNLOCK(p); 978 return (0); 979 } 980 break; 981 case P_UID: 982 if (p->p_ucred->cr_uid != (uid_t)id) { 983 PROC_UNLOCK(p); 984 return (0); 985 } 986 break; 987 case P_GID: 988 if (p->p_ucred->cr_gid != (gid_t)id) { 989 PROC_UNLOCK(p); 990 return (0); 991 } 992 break; 993 case P_JAILID: 994 if (p->p_ucred->cr_prison->pr_id != (int)id) { 995 PROC_UNLOCK(p); 996 return (0); 997 } 998 break; 999 /* 1000 * It seems that the thread structures get zeroed out 1001 * at process exit. This makes it impossible to 1002 * support P_SETID, P_CID or P_CPUID. 1003 */ 1004 default: 1005 PROC_UNLOCK(p); 1006 return (0); 1007 } 1008 1009 if (p_canwait(td, p)) { 1010 PROC_UNLOCK(p); 1011 return (0); 1012 } 1013 1014 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1015 PROC_UNLOCK(p); 1016 return (0); 1017 } 1018 1019 /* 1020 * This special case handles a kthread spawned by linux_clone 1021 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1022 * functions need to be able to distinguish between waiting 1023 * on a process and waiting on a thread. It is a thread if 1024 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1025 * signifies we want to wait for threads and not processes. 1026 */ 1027 if ((p->p_sigparent != SIGCHLD) ^ 1028 ((options & WLINUXCLONE) != 0)) { 1029 PROC_UNLOCK(p); 1030 return (0); 1031 } 1032 1033 if (siginfo != NULL) { 1034 bzero(siginfo, sizeof(*siginfo)); 1035 siginfo->si_errno = 0; 1036 1037 /* 1038 * SUSv4 requires that the si_signo value is always 1039 * SIGCHLD. Obey it despite the rfork(2) interface 1040 * allows to request other signal for child exit 1041 * notification. 1042 */ 1043 siginfo->si_signo = SIGCHLD; 1044 1045 /* 1046 * This is still a rough estimate. We will fix the 1047 * cases TRAPPED, STOPPED, and CONTINUED later. 1048 */ 1049 if (WCOREDUMP(p->p_xsig)) { 1050 siginfo->si_code = CLD_DUMPED; 1051 siginfo->si_status = WTERMSIG(p->p_xsig); 1052 } else if (WIFSIGNALED(p->p_xsig)) { 1053 siginfo->si_code = CLD_KILLED; 1054 siginfo->si_status = WTERMSIG(p->p_xsig); 1055 } else { 1056 siginfo->si_code = CLD_EXITED; 1057 siginfo->si_status = p->p_xexit; 1058 } 1059 1060 siginfo->si_pid = p->p_pid; 1061 siginfo->si_uid = p->p_ucred->cr_uid; 1062 1063 /* 1064 * The si_addr field would be useful additional 1065 * detail, but apparently the PC value may be lost 1066 * when we reach this point. bzero() above sets 1067 * siginfo->si_addr to NULL. 1068 */ 1069 } 1070 1071 /* 1072 * There should be no reason to limit resources usage info to 1073 * exited processes only. A snapshot about any resources used 1074 * by a stopped process may be exactly what is needed. 1075 */ 1076 if (wrusage != NULL) { 1077 rup = &wrusage->wru_self; 1078 *rup = p->p_ru; 1079 PROC_STATLOCK(p); 1080 calcru(p, &rup->ru_utime, &rup->ru_stime); 1081 PROC_STATUNLOCK(p); 1082 1083 rup = &wrusage->wru_children; 1084 *rup = p->p_stats->p_cru; 1085 calccru(p, &rup->ru_utime, &rup->ru_stime); 1086 } 1087 1088 if (p->p_state == PRS_ZOMBIE && !check_only) { 1089 PROC_SLOCK(p); 1090 proc_reap(td, p, status, options); 1091 return (-1); 1092 } 1093 return (1); 1094 } 1095 1096 int 1097 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1098 struct rusage *rusage) 1099 { 1100 struct __wrusage wru, *wrup; 1101 idtype_t idtype; 1102 id_t id; 1103 int ret; 1104 1105 /* 1106 * Translate the special pid values into the (idtype, pid) 1107 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1108 * kern_wait6() on its own. 1109 */ 1110 if (pid == WAIT_ANY) { 1111 idtype = P_ALL; 1112 id = 0; 1113 } else if (pid < 0) { 1114 idtype = P_PGID; 1115 id = (id_t)-pid; 1116 } else { 1117 idtype = P_PID; 1118 id = (id_t)pid; 1119 } 1120 1121 if (rusage != NULL) 1122 wrup = &wru; 1123 else 1124 wrup = NULL; 1125 1126 /* 1127 * For backward compatibility we implicitly add flags WEXITED 1128 * and WTRAPPED here. 1129 */ 1130 options |= WEXITED | WTRAPPED; 1131 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1132 if (rusage != NULL) 1133 *rusage = wru.wru_self; 1134 return (ret); 1135 } 1136 1137 static void 1138 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1139 int *status, int options, int si_code) 1140 { 1141 bool cont; 1142 1143 PROC_LOCK_ASSERT(p, MA_OWNED); 1144 sx_assert(&proctree_lock, SA_XLOCKED); 1145 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1146 si_code == CLD_CONTINUED); 1147 1148 cont = si_code == CLD_CONTINUED; 1149 if ((options & WNOWAIT) == 0) { 1150 if (cont) 1151 p->p_flag &= ~P_CONTINUED; 1152 else 1153 p->p_flag |= P_WAITED; 1154 PROC_LOCK(td->td_proc); 1155 sigqueue_take(p->p_ksi); 1156 PROC_UNLOCK(td->td_proc); 1157 } 1158 sx_xunlock(&proctree_lock); 1159 if (siginfo != NULL) { 1160 siginfo->si_code = si_code; 1161 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1162 } 1163 if (status != NULL) 1164 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1165 PROC_UNLOCK(p); 1166 td->td_retval[0] = p->p_pid; 1167 } 1168 1169 int 1170 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1171 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1172 { 1173 struct proc *p, *q; 1174 pid_t pid; 1175 int error, nfound, ret; 1176 1177 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1178 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1179 AUDIT_ARG_VALUE(options); 1180 1181 q = td->td_proc; 1182 1183 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1184 PROC_LOCK(q); 1185 id = (id_t)q->p_pgid; 1186 PROC_UNLOCK(q); 1187 idtype = P_PGID; 1188 } 1189 1190 /* If we don't know the option, just return. */ 1191 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1192 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1193 return (EINVAL); 1194 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1195 /* 1196 * We will be unable to find any matching processes, 1197 * because there are no known events to look for. 1198 * Prefer to return error instead of blocking 1199 * indefinitely. 1200 */ 1201 return (EINVAL); 1202 } 1203 1204 loop: 1205 if (q->p_flag & P_STATCHILD) { 1206 PROC_LOCK(q); 1207 q->p_flag &= ~P_STATCHILD; 1208 PROC_UNLOCK(q); 1209 } 1210 nfound = 0; 1211 sx_xlock(&proctree_lock); 1212 LIST_FOREACH(p, &q->p_children, p_sibling) { 1213 pid = p->p_pid; 1214 ret = proc_to_reap(td, p, idtype, id, status, options, 1215 wrusage, siginfo, 0); 1216 if (ret == 0) 1217 continue; 1218 else if (ret == 1) 1219 nfound++; 1220 else { 1221 td->td_retval[0] = pid; 1222 return (0); 1223 } 1224 1225 PROC_LOCK_ASSERT(p, MA_OWNED); 1226 1227 if ((options & (WTRAPPED | WUNTRACED)) != 0) 1228 PROC_SLOCK(p); 1229 1230 if ((options & WTRAPPED) != 0 && 1231 (p->p_flag & P_TRACED) != 0 && 1232 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1233 p->p_suspcount == p->p_numthreads && 1234 (p->p_flag & P_WAITED) == 0) { 1235 PROC_SUNLOCK(p); 1236 CTR4(KTR_PTRACE, 1237 "wait: returning trapped pid %d status %#x " 1238 "(xstat %d) xthread %d", 1239 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1240 p->p_xthread != NULL ? 1241 p->p_xthread->td_tid : -1); 1242 report_alive_proc(td, p, siginfo, status, options, 1243 CLD_TRAPPED); 1244 return (0); 1245 } 1246 if ((options & WUNTRACED) != 0 && 1247 (p->p_flag & P_STOPPED_SIG) != 0 && 1248 p->p_suspcount == p->p_numthreads && 1249 (p->p_flag & P_WAITED) == 0) { 1250 PROC_SUNLOCK(p); 1251 report_alive_proc(td, p, siginfo, status, options, 1252 CLD_STOPPED); 1253 return (0); 1254 } 1255 if ((options & (WTRAPPED | WUNTRACED)) != 0) 1256 PROC_SUNLOCK(p); 1257 if ((options & WCONTINUED) != 0 && 1258 (p->p_flag & P_CONTINUED) != 0) { 1259 report_alive_proc(td, p, siginfo, status, options, 1260 CLD_CONTINUED); 1261 return (0); 1262 } 1263 PROC_UNLOCK(p); 1264 } 1265 1266 /* 1267 * Look in the orphans list too, to allow the parent to 1268 * collect it's child exit status even if child is being 1269 * debugged. 1270 * 1271 * Debugger detaches from the parent upon successful 1272 * switch-over from parent to child. At this point due to 1273 * re-parenting the parent loses the child to debugger and a 1274 * wait4(2) call would report that it has no children to wait 1275 * for. By maintaining a list of orphans we allow the parent 1276 * to successfully wait until the child becomes a zombie. 1277 */ 1278 if (nfound == 0) { 1279 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1280 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1281 NULL, NULL, 1); 1282 if (ret != 0) { 1283 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1284 (int)td->td_retval[0])); 1285 PROC_UNLOCK(p); 1286 nfound++; 1287 break; 1288 } 1289 } 1290 } 1291 if (nfound == 0) { 1292 sx_xunlock(&proctree_lock); 1293 return (ECHILD); 1294 } 1295 if (options & WNOHANG) { 1296 sx_xunlock(&proctree_lock); 1297 td->td_retval[0] = 0; 1298 return (0); 1299 } 1300 PROC_LOCK(q); 1301 sx_xunlock(&proctree_lock); 1302 if (q->p_flag & P_STATCHILD) { 1303 q->p_flag &= ~P_STATCHILD; 1304 error = 0; 1305 } else 1306 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1307 PROC_UNLOCK(q); 1308 if (error) 1309 return (error); 1310 goto loop; 1311 } 1312 1313 /* 1314 * Make process 'parent' the new parent of process 'child'. 1315 * Must be called with an exclusive hold of proctree lock. 1316 */ 1317 void 1318 proc_reparent(struct proc *child, struct proc *parent) 1319 { 1320 1321 sx_assert(&proctree_lock, SX_XLOCKED); 1322 PROC_LOCK_ASSERT(child, MA_OWNED); 1323 if (child->p_pptr == parent) 1324 return; 1325 1326 PROC_LOCK(child->p_pptr); 1327 sigqueue_take(child->p_ksi); 1328 PROC_UNLOCK(child->p_pptr); 1329 LIST_REMOVE(child, p_sibling); 1330 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1331 1332 clear_orphan(child); 1333 if (child->p_flag & P_TRACED) { 1334 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1335 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1336 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1337 p_orphan); 1338 } else { 1339 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1340 child, p_orphan); 1341 } 1342 child->p_treeflag |= P_TREE_ORPHANED; 1343 } 1344 1345 child->p_pptr = parent; 1346 } 1347