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