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