1 /*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_compat.h" 41 #include "opt_ktrace.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/sysproto.h> 46 #include <sys/capsicum.h> 47 #include <sys/eventhandler.h> 48 #include <sys/kernel.h> 49 #include <sys/malloc.h> 50 #include <sys/lock.h> 51 #include <sys/mutex.h> 52 #include <sys/proc.h> 53 #include <sys/procdesc.h> 54 #include <sys/pioctl.h> 55 #include <sys/jail.h> 56 #include <sys/tty.h> 57 #include <sys/wait.h> 58 #include <sys/vmmeter.h> 59 #include <sys/vnode.h> 60 #include <sys/racct.h> 61 #include <sys/resourcevar.h> 62 #include <sys/sbuf.h> 63 #include <sys/signalvar.h> 64 #include <sys/sched.h> 65 #include <sys/sx.h> 66 #include <sys/syscallsubr.h> 67 #include <sys/syslog.h> 68 #include <sys/ptrace.h> 69 #include <sys/acct.h> /* for acct_process() function prototype */ 70 #include <sys/filedesc.h> 71 #include <sys/sdt.h> 72 #include <sys/shm.h> 73 #include <sys/sem.h> 74 #include <sys/umtx.h> 75 #ifdef KTRACE 76 #include <sys/ktrace.h> 77 #endif 78 79 #include <security/audit/audit.h> 80 #include <security/mac/mac_framework.h> 81 82 #include <vm/vm.h> 83 #include <vm/vm_extern.h> 84 #include <vm/vm_param.h> 85 #include <vm/pmap.h> 86 #include <vm/vm_map.h> 87 #include <vm/vm_page.h> 88 #include <vm/uma.h> 89 90 #ifdef KDTRACE_HOOKS 91 #include <sys/dtrace_bsd.h> 92 dtrace_execexit_func_t dtrace_fasttrap_exit; 93 #endif 94 95 SDT_PROVIDER_DECLARE(proc); 96 SDT_PROBE_DEFINE1(proc, kernel, , exit, "int"); 97 98 /* Hook for NFS teardown procedure. */ 99 void (*nlminfo_release_p)(struct proc *p); 100 101 struct proc * 102 proc_realparent(struct proc *child) 103 { 104 struct proc *p, *parent; 105 106 sx_assert(&proctree_lock, SX_LOCKED); 107 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 108 if (child->p_oppid == 0 || 109 child->p_pptr->p_pid == child->p_oppid) 110 parent = child->p_pptr; 111 else 112 parent = initproc; 113 return (parent); 114 } 115 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 116 /* Cannot use LIST_PREV(), since the list head is not known. */ 117 p = __containerof(p->p_orphan.le_prev, struct proc, 118 p_orphan.le_next); 119 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 120 ("missing P_ORPHAN %p", p)); 121 } 122 parent = __containerof(p->p_orphan.le_prev, struct proc, 123 p_orphans.lh_first); 124 return (parent); 125 } 126 127 void 128 reaper_abandon_children(struct proc *p, bool exiting) 129 { 130 struct proc *p1, *p2, *ptmp; 131 132 sx_assert(&proctree_lock, SX_LOCKED); 133 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 134 if ((p->p_treeflag & P_TREE_REAPER) == 0) 135 return; 136 p1 = p->p_reaper; 137 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 138 LIST_REMOVE(p2, p_reapsibling); 139 p2->p_reaper = p1; 140 p2->p_reapsubtree = p->p_reapsubtree; 141 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 142 if (exiting && p2->p_pptr == p) { 143 PROC_LOCK(p2); 144 proc_reparent(p2, p1); 145 PROC_UNLOCK(p2); 146 } 147 } 148 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 149 p->p_treeflag &= ~P_TREE_REAPER; 150 } 151 152 static void 153 clear_orphan(struct proc *p) 154 { 155 struct proc *p1; 156 157 sx_assert(&proctree_lock, SA_XLOCKED); 158 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 159 return; 160 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 161 p1 = LIST_NEXT(p, p_orphan); 162 if (p1 != NULL) 163 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 164 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 165 } 166 LIST_REMOVE(p, p_orphan); 167 p->p_treeflag &= ~P_TREE_ORPHANED; 168 } 169 170 /* 171 * exit -- death of process. 172 */ 173 void 174 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 175 { 176 177 exit1(td, W_EXITCODE(uap->rval, 0)); 178 /* NOTREACHED */ 179 } 180 181 /* 182 * Exit: deallocate address space and other resources, change proc state to 183 * zombie, and unlink proc from allproc and parent's lists. Save exit status 184 * and rusage for wait(). Check for child processes and orphan them. 185 */ 186 void 187 exit1(struct thread *td, int rv) 188 { 189 struct proc *p, *nq, *q, *t; 190 struct thread *tdt; 191 struct vnode *ttyvp = NULL; 192 193 mtx_assert(&Giant, MA_NOTOWNED); 194 195 p = td->td_proc; 196 /* 197 * XXX in case we're rebooting we just let init die in order to 198 * work around an unsolved stack overflow seen very late during 199 * shutdown on sparc64 when the gmirror worker process exists. 200 */ 201 if (p == initproc && rebooting == 0) { 202 printf("init died (signal %d, exit %d)\n", 203 WTERMSIG(rv), WEXITSTATUS(rv)); 204 panic("Going nowhere without my init!"); 205 } 206 207 /* 208 * MUST abort all other threads before proceeding past here. 209 */ 210 PROC_LOCK(p); 211 /* 212 * First check if some other thread or external request got 213 * here before us. If so, act appropriately: exit or suspend. 214 * We must ensure that stop requests are handled before we set 215 * P_WEXIT. 216 */ 217 thread_suspend_check(0); 218 while (p->p_flag & P_HADTHREADS) { 219 /* 220 * Kill off the other threads. This requires 221 * some co-operation from other parts of the kernel 222 * so it may not be instantaneous. With this state set 223 * any thread entering the kernel from userspace will 224 * thread_exit() in trap(). Any thread attempting to 225 * sleep will return immediately with EINTR or EWOULDBLOCK 226 * which will hopefully force them to back out to userland 227 * freeing resources as they go. Any thread attempting 228 * to return to userland will thread_exit() from userret(). 229 * thread_exit() will unsuspend us when the last of the 230 * other threads exits. 231 * If there is already a thread singler after resumption, 232 * calling thread_single will fail; in that case, we just 233 * re-check all suspension request, the thread should 234 * either be suspended there or exit. 235 */ 236 if (!thread_single(p, SINGLE_EXIT)) 237 /* 238 * All other activity in this process is now 239 * stopped. Threading support has been turned 240 * off. 241 */ 242 break; 243 /* 244 * Recheck for new stop or suspend requests which 245 * might appear while process lock was dropped in 246 * thread_single(). 247 */ 248 thread_suspend_check(0); 249 } 250 KASSERT(p->p_numthreads == 1, 251 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 252 racct_sub(p, RACCT_NTHR, 1); 253 /* 254 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 255 * on our vmspace, so we should block below until they have 256 * released their reference to us. Note that if they have 257 * requested S_EXIT stops we will block here until they ack 258 * via PIOCCONT. 259 */ 260 _STOPEVENT(p, S_EXIT, rv); 261 262 /* 263 * Ignore any pending request to stop due to a stop signal. 264 * Once P_WEXIT is set, future requests will be ignored as 265 * well. 266 */ 267 p->p_flag &= ~P_STOPPED_SIG; 268 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 269 270 /* 271 * Note that we are exiting and do another wakeup of anyone in 272 * PIOCWAIT in case they aren't listening for S_EXIT stops or 273 * decided to wait again after we told them we are exiting. 274 */ 275 p->p_flag |= P_WEXIT; 276 wakeup(&p->p_stype); 277 278 /* 279 * Wait for any processes that have a hold on our vmspace to 280 * release their reference. 281 */ 282 while (p->p_lock > 0) 283 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 284 285 p->p_xstat = rv; /* Let event handler change exit status */ 286 PROC_UNLOCK(p); 287 /* Drain the limit callout while we don't have the proc locked */ 288 callout_drain(&p->p_limco); 289 290 #ifdef AUDIT 291 /* 292 * The Sun BSM exit token contains two components: an exit status as 293 * passed to exit(), and a return value to indicate what sort of exit 294 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 295 * what the return value is. 296 */ 297 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0); 298 AUDIT_SYSCALL_EXIT(0, td); 299 #endif 300 301 /* Are we a task leader with peers? */ 302 if (p->p_peers != NULL && p == p->p_leader) { 303 mtx_lock(&ppeers_lock); 304 q = p->p_peers; 305 while (q != NULL) { 306 PROC_LOCK(q); 307 kern_psignal(q, SIGKILL); 308 PROC_UNLOCK(q); 309 q = q->p_peers; 310 } 311 while (p->p_peers != NULL) 312 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 313 mtx_unlock(&ppeers_lock); 314 } 315 316 /* 317 * Check if any loadable modules need anything done at process exit. 318 * E.g. SYSV IPC stuff 319 * XXX what if one of these generates an error? 320 */ 321 EVENTHANDLER_INVOKE(process_exit, p); 322 323 /* 324 * If parent is waiting for us to exit or exec, 325 * P_PPWAIT is set; we will wakeup the parent below. 326 */ 327 PROC_LOCK(p); 328 rv = p->p_xstat; /* Event handler could change exit status */ 329 stopprofclock(p); 330 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 331 332 /* 333 * Stop the real interval timer. If the handler is currently 334 * executing, prevent it from rearming itself and let it finish. 335 */ 336 if (timevalisset(&p->p_realtimer.it_value) && 337 callout_stop(&p->p_itcallout) == 0) { 338 timevalclear(&p->p_realtimer.it_interval); 339 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 340 KASSERT(!timevalisset(&p->p_realtimer.it_value), 341 ("realtime timer is still armed")); 342 } 343 PROC_UNLOCK(p); 344 345 /* 346 * Reset any sigio structures pointing to us as a result of 347 * F_SETOWN with our pid. 348 */ 349 funsetownlst(&p->p_sigiolst); 350 351 /* 352 * If this process has an nlminfo data area (for lockd), release it 353 */ 354 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 355 (*nlminfo_release_p)(p); 356 357 /* 358 * Close open files and release open-file table. 359 * This may block! 360 */ 361 fdescfree(td); 362 363 /* 364 * If this thread tickled GEOM, we need to wait for the giggling to 365 * stop before we return to userland 366 */ 367 if (td->td_pflags & TDP_GEOM) 368 g_waitidle(); 369 370 /* 371 * Remove ourself from our leader's peer list and wake our leader. 372 */ 373 if (p->p_leader->p_peers != NULL) { 374 mtx_lock(&ppeers_lock); 375 if (p->p_leader->p_peers != NULL) { 376 q = p->p_leader; 377 while (q->p_peers != p) 378 q = q->p_peers; 379 q->p_peers = p->p_peers; 380 wakeup(p->p_leader); 381 } 382 mtx_unlock(&ppeers_lock); 383 } 384 385 vmspace_exit(td); 386 387 sx_xlock(&proctree_lock); 388 if (SESS_LEADER(p)) { 389 struct session *sp = p->p_session; 390 struct tty *tp; 391 392 /* 393 * s_ttyp is not zero'd; we use this to indicate that 394 * the session once had a controlling terminal. (for 395 * logging and informational purposes) 396 */ 397 SESS_LOCK(sp); 398 ttyvp = sp->s_ttyvp; 399 tp = sp->s_ttyp; 400 sp->s_ttyvp = NULL; 401 sp->s_ttydp = NULL; 402 sp->s_leader = NULL; 403 SESS_UNLOCK(sp); 404 405 /* 406 * Signal foreground pgrp and revoke access to 407 * controlling terminal if it has not been revoked 408 * already. 409 * 410 * Because the TTY may have been revoked in the mean 411 * time and could already have a new session associated 412 * with it, make sure we don't send a SIGHUP to a 413 * foreground process group that does not belong to this 414 * session. 415 */ 416 417 if (tp != NULL) { 418 tty_lock(tp); 419 if (tp->t_session == sp) 420 tty_signal_pgrp(tp, SIGHUP); 421 tty_unlock(tp); 422 } 423 424 if (ttyvp != NULL) { 425 sx_xunlock(&proctree_lock); 426 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 427 VOP_REVOKE(ttyvp, REVOKEALL); 428 VOP_UNLOCK(ttyvp, 0); 429 } 430 sx_xlock(&proctree_lock); 431 } 432 } 433 fixjobc(p, p->p_pgrp, 0); 434 sx_xunlock(&proctree_lock); 435 (void)acct_process(td); 436 437 /* Release the TTY now we've unlocked everything. */ 438 if (ttyvp != NULL) 439 vrele(ttyvp); 440 #ifdef KTRACE 441 ktrprocexit(td); 442 #endif 443 /* 444 * Release reference to text vnode 445 */ 446 if (p->p_textvp != NULL) { 447 vrele(p->p_textvp); 448 p->p_textvp = NULL; 449 } 450 451 /* 452 * Release our limits structure. 453 */ 454 lim_free(p->p_limit); 455 p->p_limit = NULL; 456 457 tidhash_remove(td); 458 459 /* 460 * Remove proc from allproc queue and pidhash chain. 461 * Place onto zombproc. Unlink from parent's child list. 462 */ 463 sx_xlock(&allproc_lock); 464 LIST_REMOVE(p, p_list); 465 LIST_INSERT_HEAD(&zombproc, p, p_list); 466 LIST_REMOVE(p, p_hash); 467 sx_xunlock(&allproc_lock); 468 469 /* 470 * Call machine-dependent code to release any 471 * machine-dependent resources other than the address space. 472 * The address space is released by "vmspace_exitfree(p)" in 473 * vm_waitproc(). 474 */ 475 cpu_exit(td); 476 477 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 478 479 /* 480 * Reparent all children processes: 481 * - traced ones to the original parent (or init if we are that parent) 482 * - the rest to init 483 */ 484 sx_xlock(&proctree_lock); 485 q = LIST_FIRST(&p->p_children); 486 if (q != NULL) /* only need this if any child is S_ZOMB */ 487 wakeup(q->p_reaper); 488 for (; q != NULL; q = nq) { 489 nq = LIST_NEXT(q, p_sibling); 490 PROC_LOCK(q); 491 q->p_sigparent = SIGCHLD; 492 493 if (!(q->p_flag & P_TRACED)) { 494 proc_reparent(q, q->p_reaper); 495 } else { 496 /* 497 * Traced processes are killed since their existence 498 * means someone is screwing up. 499 */ 500 t = proc_realparent(q); 501 if (t == p) { 502 proc_reparent(q, q->p_reaper); 503 } else { 504 PROC_LOCK(t); 505 proc_reparent(q, t); 506 PROC_UNLOCK(t); 507 } 508 /* 509 * Since q was found on our children list, the 510 * proc_reparent() call moved q to the orphan 511 * list due to present P_TRACED flag. Clear 512 * orphan link for q now while q is locked. 513 */ 514 clear_orphan(q); 515 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 516 FOREACH_THREAD_IN_PROC(q, tdt) 517 tdt->td_dbgflags &= ~TDB_SUSPEND; 518 kern_psignal(q, SIGKILL); 519 } 520 PROC_UNLOCK(q); 521 } 522 523 /* 524 * Also get rid of our orphans. 525 */ 526 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 527 PROC_LOCK(q); 528 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 529 q->p_pid); 530 clear_orphan(q); 531 PROC_UNLOCK(q); 532 } 533 534 /* Save exit status. */ 535 PROC_LOCK(p); 536 p->p_xthread = td; 537 538 /* Tell the prison that we are gone. */ 539 prison_proc_free(p->p_ucred->cr_prison); 540 541 #ifdef KDTRACE_HOOKS 542 /* 543 * Tell the DTrace fasttrap provider about the exit if it 544 * has declared an interest. 545 */ 546 if (dtrace_fasttrap_exit) 547 dtrace_fasttrap_exit(p); 548 #endif 549 550 /* 551 * Notify interested parties of our demise. 552 */ 553 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 554 555 #ifdef KDTRACE_HOOKS 556 int reason = CLD_EXITED; 557 if (WCOREDUMP(rv)) 558 reason = CLD_DUMPED; 559 else if (WIFSIGNALED(rv)) 560 reason = CLD_KILLED; 561 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 562 #endif 563 564 /* 565 * Just delete all entries in the p_klist. At this point we won't 566 * report any more events, and there are nasty race conditions that 567 * can beat us if we don't. 568 */ 569 knlist_clear(&p->p_klist, 1); 570 571 /* 572 * If this is a process with a descriptor, we may not need to deliver 573 * a signal to the parent. proctree_lock is held over 574 * procdesc_exit() to serialize concurrent calls to close() and 575 * exit(). 576 */ 577 if (p->p_procdesc == NULL || procdesc_exit(p)) { 578 /* 579 * Notify parent that we're gone. If parent has the 580 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 581 * notify process 1 instead (and hope it will handle this 582 * situation). 583 */ 584 PROC_LOCK(p->p_pptr); 585 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 586 if (p->p_pptr->p_sigacts->ps_flag & 587 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 588 struct proc *pp; 589 590 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 591 pp = p->p_pptr; 592 PROC_UNLOCK(pp); 593 proc_reparent(p, p->p_reaper); 594 p->p_sigparent = SIGCHLD; 595 PROC_LOCK(p->p_pptr); 596 597 /* 598 * Notify parent, so in case he was wait(2)ing or 599 * executing waitpid(2) with our pid, he will 600 * continue. 601 */ 602 wakeup(pp); 603 } else 604 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 605 606 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 607 childproc_exited(p); 608 else if (p->p_sigparent != 0) { 609 if (p->p_sigparent == SIGCHLD) 610 childproc_exited(p); 611 else /* LINUX thread */ 612 kern_psignal(p->p_pptr, p->p_sigparent); 613 } 614 } else 615 PROC_LOCK(p->p_pptr); 616 sx_xunlock(&proctree_lock); 617 618 /* 619 * The state PRS_ZOMBIE prevents other proesses from sending 620 * signal to the process, to avoid memory leak, we free memory 621 * for signal queue at the time when the state is set. 622 */ 623 sigqueue_flush(&p->p_sigqueue); 624 sigqueue_flush(&td->td_sigqueue); 625 626 /* 627 * We have to wait until after acquiring all locks before 628 * changing p_state. We need to avoid all possible context 629 * switches (including ones from blocking on a mutex) while 630 * marked as a zombie. We also have to set the zombie state 631 * before we release the parent process' proc lock to avoid 632 * a lost wakeup. So, we first call wakeup, then we grab the 633 * sched lock, update the state, and release the parent process' 634 * proc lock. 635 */ 636 wakeup(p->p_pptr); 637 cv_broadcast(&p->p_pwait); 638 sched_exit(p->p_pptr, td); 639 umtx_thread_exit(td); 640 PROC_SLOCK(p); 641 p->p_state = PRS_ZOMBIE; 642 PROC_UNLOCK(p->p_pptr); 643 644 /* 645 * Hopefully no one will try to deliver a signal to the process this 646 * late in the game. 647 */ 648 knlist_destroy(&p->p_klist); 649 650 /* 651 * Save our children's rusage information in our exit rusage. 652 */ 653 PROC_STATLOCK(p); 654 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 655 PROC_STATUNLOCK(p); 656 657 /* 658 * Make sure the scheduler takes this thread out of its tables etc. 659 * This will also release this thread's reference to the ucred. 660 * Other thread parts to release include pcb bits and such. 661 */ 662 thread_exit(); 663 } 664 665 666 #ifndef _SYS_SYSPROTO_H_ 667 struct abort2_args { 668 char *why; 669 int nargs; 670 void **args; 671 }; 672 #endif 673 674 int 675 sys_abort2(struct thread *td, struct abort2_args *uap) 676 { 677 struct proc *p = td->td_proc; 678 struct sbuf *sb; 679 void *uargs[16]; 680 int error, i, sig; 681 682 /* 683 * Do it right now so we can log either proper call of abort2(), or 684 * note, that invalid argument was passed. 512 is big enough to 685 * handle 16 arguments' descriptions with additional comments. 686 */ 687 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 688 sbuf_clear(sb); 689 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 690 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 691 /* 692 * Since we can't return from abort2(), send SIGKILL in cases, where 693 * abort2() was called improperly 694 */ 695 sig = SIGKILL; 696 /* Prevent from DoSes from user-space. */ 697 if (uap->nargs < 0 || uap->nargs > 16) 698 goto out; 699 if (uap->nargs > 0) { 700 if (uap->args == NULL) 701 goto out; 702 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 703 if (error != 0) 704 goto out; 705 } 706 /* 707 * Limit size of 'reason' string to 128. Will fit even when 708 * maximal number of arguments was chosen to be logged. 709 */ 710 if (uap->why != NULL) { 711 error = sbuf_copyin(sb, uap->why, 128); 712 if (error < 0) 713 goto out; 714 } else { 715 sbuf_printf(sb, "(null)"); 716 } 717 if (uap->nargs > 0) { 718 sbuf_printf(sb, "("); 719 for (i = 0;i < uap->nargs; i++) 720 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 721 sbuf_printf(sb, ")"); 722 } 723 /* 724 * Final stage: arguments were proper, string has been 725 * successfully copied from userspace, and copying pointers 726 * from user-space succeed. 727 */ 728 sig = SIGABRT; 729 out: 730 if (sig == SIGKILL) { 731 sbuf_trim(sb); 732 sbuf_printf(sb, " (Reason text inaccessible)"); 733 } 734 sbuf_cat(sb, "\n"); 735 sbuf_finish(sb); 736 log(LOG_INFO, "%s", sbuf_data(sb)); 737 sbuf_delete(sb); 738 exit1(td, W_EXITCODE(0, sig)); 739 return (0); 740 } 741 742 743 #ifdef COMPAT_43 744 /* 745 * The dirty work is handled by kern_wait(). 746 */ 747 int 748 owait(struct thread *td, struct owait_args *uap __unused) 749 { 750 int error, status; 751 752 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 753 if (error == 0) 754 td->td_retval[1] = status; 755 return (error); 756 } 757 #endif /* COMPAT_43 */ 758 759 /* 760 * The dirty work is handled by kern_wait(). 761 */ 762 int 763 sys_wait4(struct thread *td, struct wait4_args *uap) 764 { 765 struct rusage ru, *rup; 766 int error, status; 767 768 if (uap->rusage != NULL) 769 rup = &ru; 770 else 771 rup = NULL; 772 error = kern_wait(td, uap->pid, &status, uap->options, rup); 773 if (uap->status != NULL && error == 0) 774 error = copyout(&status, uap->status, sizeof(status)); 775 if (uap->rusage != NULL && error == 0) 776 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 777 return (error); 778 } 779 780 int 781 sys_wait6(struct thread *td, struct wait6_args *uap) 782 { 783 struct __wrusage wru, *wrup; 784 siginfo_t si, *sip; 785 idtype_t idtype; 786 id_t id; 787 int error, status; 788 789 idtype = uap->idtype; 790 id = uap->id; 791 792 if (uap->wrusage != NULL) 793 wrup = &wru; 794 else 795 wrup = NULL; 796 797 if (uap->info != NULL) { 798 sip = &si; 799 bzero(sip, sizeof(*sip)); 800 } else 801 sip = NULL; 802 803 /* 804 * We expect all callers of wait6() to know about WEXITED and 805 * WTRAPPED. 806 */ 807 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 808 809 if (uap->status != NULL && error == 0) 810 error = copyout(&status, uap->status, sizeof(status)); 811 if (uap->wrusage != NULL && error == 0) 812 error = copyout(&wru, uap->wrusage, sizeof(wru)); 813 if (uap->info != NULL && error == 0) 814 error = copyout(&si, uap->info, sizeof(si)); 815 return (error); 816 } 817 818 /* 819 * Reap the remains of a zombie process and optionally return status and 820 * rusage. Asserts and will release both the proctree_lock and the process 821 * lock as part of its work. 822 */ 823 void 824 proc_reap(struct thread *td, struct proc *p, int *status, int options) 825 { 826 struct proc *q, *t; 827 828 sx_assert(&proctree_lock, SA_XLOCKED); 829 PROC_LOCK_ASSERT(p, MA_OWNED); 830 PROC_SLOCK_ASSERT(p, MA_OWNED); 831 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 832 833 q = td->td_proc; 834 835 PROC_SUNLOCK(p); 836 td->td_retval[0] = p->p_pid; 837 if (status) 838 *status = p->p_xstat; /* convert to int */ 839 if (options & WNOWAIT) { 840 /* 841 * Only poll, returning the status. Caller does not wish to 842 * release the proc struct just yet. 843 */ 844 PROC_UNLOCK(p); 845 sx_xunlock(&proctree_lock); 846 return; 847 } 848 849 PROC_LOCK(q); 850 sigqueue_take(p->p_ksi); 851 PROC_UNLOCK(q); 852 853 /* 854 * If we got the child via a ptrace 'attach', we need to give it back 855 * to the old parent. 856 */ 857 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 858 PROC_UNLOCK(p); 859 t = proc_realparent(p); 860 PROC_LOCK(t); 861 PROC_LOCK(p); 862 CTR2(KTR_PTRACE, 863 "wait: traced child %d moved back to parent %d", p->p_pid, 864 t->p_pid); 865 proc_reparent(p, t); 866 p->p_oppid = 0; 867 PROC_UNLOCK(p); 868 pksignal(t, SIGCHLD, p->p_ksi); 869 wakeup(t); 870 cv_broadcast(&p->p_pwait); 871 PROC_UNLOCK(t); 872 sx_xunlock(&proctree_lock); 873 return; 874 } 875 p->p_oppid = 0; 876 PROC_UNLOCK(p); 877 878 /* 879 * Remove other references to this process to ensure we have an 880 * exclusive reference. 881 */ 882 sx_xlock(&allproc_lock); 883 LIST_REMOVE(p, p_list); /* off zombproc */ 884 sx_xunlock(&allproc_lock); 885 LIST_REMOVE(p, p_sibling); 886 reaper_abandon_children(p, true); 887 LIST_REMOVE(p, p_reapsibling); 888 PROC_LOCK(p); 889 clear_orphan(p); 890 PROC_UNLOCK(p); 891 leavepgrp(p); 892 if (p->p_procdesc != NULL) 893 procdesc_reap(p); 894 sx_xunlock(&proctree_lock); 895 896 /* 897 * Removal from allproc list and process group list paired with 898 * PROC_LOCK which was executed during that time should guarantee 899 * nothing can reach this process anymore. As such further locking 900 * is unnecessary. 901 */ 902 p->p_xstat = 0; /* XXX: why? */ 903 904 PROC_LOCK(q); 905 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 906 PROC_UNLOCK(q); 907 908 /* 909 * Decrement the count of procs running with this uid. 910 */ 911 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 912 913 /* 914 * Destroy resource accounting information associated with the process. 915 */ 916 #ifdef RACCT 917 if (racct_enable) { 918 PROC_LOCK(p); 919 racct_sub(p, RACCT_NPROC, 1); 920 PROC_UNLOCK(p); 921 } 922 #endif 923 racct_proc_exit(p); 924 925 /* 926 * Free credentials, arguments, and sigacts. 927 */ 928 crfree(p->p_ucred); 929 proc_set_cred(p, NULL); 930 pargs_drop(p->p_args); 931 p->p_args = NULL; 932 sigacts_free(p->p_sigacts); 933 p->p_sigacts = NULL; 934 935 /* 936 * Do any thread-system specific cleanups. 937 */ 938 thread_wait(p); 939 940 /* 941 * Give vm and machine-dependent layer a chance to free anything that 942 * cpu_exit couldn't release while still running in process context. 943 */ 944 vm_waitproc(p); 945 #ifdef MAC 946 mac_proc_destroy(p); 947 #endif 948 KASSERT(FIRST_THREAD_IN_PROC(p), 949 ("proc_reap: no residual thread!")); 950 uma_zfree(proc_zone, p); 951 sx_xlock(&allproc_lock); 952 nprocs--; 953 sx_xunlock(&allproc_lock); 954 } 955 956 static int 957 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 958 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 959 int check_only) 960 { 961 struct proc *q; 962 struct rusage *rup; 963 964 sx_assert(&proctree_lock, SA_XLOCKED); 965 966 q = td->td_proc; 967 PROC_LOCK(p); 968 969 switch (idtype) { 970 case P_ALL: 971 break; 972 case P_PID: 973 if (p->p_pid != (pid_t)id) { 974 PROC_UNLOCK(p); 975 return (0); 976 } 977 break; 978 case P_PGID: 979 if (p->p_pgid != (pid_t)id) { 980 PROC_UNLOCK(p); 981 return (0); 982 } 983 break; 984 case P_SID: 985 if (p->p_session->s_sid != (pid_t)id) { 986 PROC_UNLOCK(p); 987 return (0); 988 } 989 break; 990 case P_UID: 991 if (p->p_ucred->cr_uid != (uid_t)id) { 992 PROC_UNLOCK(p); 993 return (0); 994 } 995 break; 996 case P_GID: 997 if (p->p_ucred->cr_gid != (gid_t)id) { 998 PROC_UNLOCK(p); 999 return (0); 1000 } 1001 break; 1002 case P_JAILID: 1003 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1004 PROC_UNLOCK(p); 1005 return (0); 1006 } 1007 break; 1008 /* 1009 * It seems that the thread structures get zeroed out 1010 * at process exit. This makes it impossible to 1011 * support P_SETID, P_CID or P_CPUID. 1012 */ 1013 default: 1014 PROC_UNLOCK(p); 1015 return (0); 1016 } 1017 1018 if (p_canwait(td, p)) { 1019 PROC_UNLOCK(p); 1020 return (0); 1021 } 1022 1023 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1024 PROC_UNLOCK(p); 1025 return (0); 1026 } 1027 1028 /* 1029 * This special case handles a kthread spawned by linux_clone 1030 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1031 * functions need to be able to distinguish between waiting 1032 * on a process and waiting on a thread. It is a thread if 1033 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1034 * signifies we want to wait for threads and not processes. 1035 */ 1036 if ((p->p_sigparent != SIGCHLD) ^ 1037 ((options & WLINUXCLONE) != 0)) { 1038 PROC_UNLOCK(p); 1039 return (0); 1040 } 1041 1042 if (siginfo != NULL) { 1043 bzero(siginfo, sizeof(*siginfo)); 1044 siginfo->si_errno = 0; 1045 1046 /* 1047 * SUSv4 requires that the si_signo value is always 1048 * SIGCHLD. Obey it despite the rfork(2) interface 1049 * allows to request other signal for child exit 1050 * notification. 1051 */ 1052 siginfo->si_signo = SIGCHLD; 1053 1054 /* 1055 * This is still a rough estimate. We will fix the 1056 * cases TRAPPED, STOPPED, and CONTINUED later. 1057 */ 1058 if (WCOREDUMP(p->p_xstat)) { 1059 siginfo->si_code = CLD_DUMPED; 1060 siginfo->si_status = WTERMSIG(p->p_xstat); 1061 } else if (WIFSIGNALED(p->p_xstat)) { 1062 siginfo->si_code = CLD_KILLED; 1063 siginfo->si_status = WTERMSIG(p->p_xstat); 1064 } else { 1065 siginfo->si_code = CLD_EXITED; 1066 siginfo->si_status = WEXITSTATUS(p->p_xstat); 1067 } 1068 1069 siginfo->si_pid = p->p_pid; 1070 siginfo->si_uid = p->p_ucred->cr_uid; 1071 1072 /* 1073 * The si_addr field would be useful additional 1074 * detail, but apparently the PC value may be lost 1075 * when we reach this point. bzero() above sets 1076 * siginfo->si_addr to NULL. 1077 */ 1078 } 1079 1080 /* 1081 * There should be no reason to limit resources usage info to 1082 * exited processes only. A snapshot about any resources used 1083 * by a stopped process may be exactly what is needed. 1084 */ 1085 if (wrusage != NULL) { 1086 rup = &wrusage->wru_self; 1087 *rup = p->p_ru; 1088 PROC_STATLOCK(p); 1089 calcru(p, &rup->ru_utime, &rup->ru_stime); 1090 PROC_STATUNLOCK(p); 1091 1092 rup = &wrusage->wru_children; 1093 *rup = p->p_stats->p_cru; 1094 calccru(p, &rup->ru_utime, &rup->ru_stime); 1095 } 1096 1097 if (p->p_state == PRS_ZOMBIE && !check_only) { 1098 PROC_SLOCK(p); 1099 proc_reap(td, p, status, options); 1100 return (-1); 1101 } 1102 PROC_UNLOCK(p); 1103 return (1); 1104 } 1105 1106 int 1107 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1108 struct rusage *rusage) 1109 { 1110 struct __wrusage wru, *wrup; 1111 idtype_t idtype; 1112 id_t id; 1113 int ret; 1114 1115 /* 1116 * Translate the special pid values into the (idtype, pid) 1117 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1118 * kern_wait6() on its own. 1119 */ 1120 if (pid == WAIT_ANY) { 1121 idtype = P_ALL; 1122 id = 0; 1123 } else if (pid < 0) { 1124 idtype = P_PGID; 1125 id = (id_t)-pid; 1126 } else { 1127 idtype = P_PID; 1128 id = (id_t)pid; 1129 } 1130 1131 if (rusage != NULL) 1132 wrup = &wru; 1133 else 1134 wrup = NULL; 1135 1136 /* 1137 * For backward compatibility we implicitly add flags WEXITED 1138 * and WTRAPPED here. 1139 */ 1140 options |= WEXITED | WTRAPPED; 1141 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1142 if (rusage != NULL) 1143 *rusage = wru.wru_self; 1144 return (ret); 1145 } 1146 1147 int 1148 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1149 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1150 { 1151 struct proc *p, *q; 1152 int error, nfound, ret; 1153 1154 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1155 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1156 AUDIT_ARG_VALUE(options); 1157 1158 q = td->td_proc; 1159 1160 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1161 PROC_LOCK(q); 1162 id = (id_t)q->p_pgid; 1163 PROC_UNLOCK(q); 1164 idtype = P_PGID; 1165 } 1166 1167 /* If we don't know the option, just return. */ 1168 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1169 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1170 return (EINVAL); 1171 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1172 /* 1173 * We will be unable to find any matching processes, 1174 * because there are no known events to look for. 1175 * Prefer to return error instead of blocking 1176 * indefinitely. 1177 */ 1178 return (EINVAL); 1179 } 1180 1181 loop: 1182 if (q->p_flag & P_STATCHILD) { 1183 PROC_LOCK(q); 1184 q->p_flag &= ~P_STATCHILD; 1185 PROC_UNLOCK(q); 1186 } 1187 nfound = 0; 1188 sx_xlock(&proctree_lock); 1189 LIST_FOREACH(p, &q->p_children, p_sibling) { 1190 ret = proc_to_reap(td, p, idtype, id, status, options, 1191 wrusage, siginfo, 0); 1192 if (ret == 0) 1193 continue; 1194 else if (ret == 1) 1195 nfound++; 1196 else 1197 return (0); 1198 1199 PROC_LOCK(p); 1200 PROC_SLOCK(p); 1201 1202 if ((options & WTRAPPED) != 0 && 1203 (p->p_flag & P_TRACED) != 0 && 1204 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1205 (p->p_suspcount == p->p_numthreads) && 1206 ((p->p_flag & P_WAITED) == 0)) { 1207 PROC_SUNLOCK(p); 1208 if ((options & WNOWAIT) == 0) 1209 p->p_flag |= P_WAITED; 1210 sx_xunlock(&proctree_lock); 1211 td->td_retval[0] = p->p_pid; 1212 1213 if (status != NULL) 1214 *status = W_STOPCODE(p->p_xstat); 1215 if (siginfo != NULL) { 1216 siginfo->si_status = p->p_xstat; 1217 siginfo->si_code = CLD_TRAPPED; 1218 } 1219 if ((options & WNOWAIT) == 0) { 1220 PROC_LOCK(q); 1221 sigqueue_take(p->p_ksi); 1222 PROC_UNLOCK(q); 1223 } 1224 1225 CTR4(KTR_PTRACE, 1226 "wait: returning trapped pid %d status %#x (xstat %d) xthread %d", 1227 p->p_pid, W_STOPCODE(p->p_xstat), p->p_xstat, 1228 p->p_xthread != NULL ? p->p_xthread->td_tid : -1); 1229 PROC_UNLOCK(p); 1230 return (0); 1231 } 1232 if ((options & WUNTRACED) != 0 && 1233 (p->p_flag & P_STOPPED_SIG) != 0 && 1234 (p->p_suspcount == p->p_numthreads) && 1235 ((p->p_flag & P_WAITED) == 0)) { 1236 PROC_SUNLOCK(p); 1237 if ((options & WNOWAIT) == 0) 1238 p->p_flag |= P_WAITED; 1239 sx_xunlock(&proctree_lock); 1240 td->td_retval[0] = p->p_pid; 1241 1242 if (status != NULL) 1243 *status = W_STOPCODE(p->p_xstat); 1244 if (siginfo != NULL) { 1245 siginfo->si_status = p->p_xstat; 1246 siginfo->si_code = CLD_STOPPED; 1247 } 1248 if ((options & WNOWAIT) == 0) { 1249 PROC_LOCK(q); 1250 sigqueue_take(p->p_ksi); 1251 PROC_UNLOCK(q); 1252 } 1253 1254 PROC_UNLOCK(p); 1255 return (0); 1256 } 1257 PROC_SUNLOCK(p); 1258 if ((options & WCONTINUED) != 0 && 1259 (p->p_flag & P_CONTINUED) != 0) { 1260 sx_xunlock(&proctree_lock); 1261 td->td_retval[0] = p->p_pid; 1262 if ((options & WNOWAIT) == 0) { 1263 p->p_flag &= ~P_CONTINUED; 1264 PROC_LOCK(q); 1265 sigqueue_take(p->p_ksi); 1266 PROC_UNLOCK(q); 1267 } 1268 PROC_UNLOCK(p); 1269 1270 if (status != NULL) 1271 *status = SIGCONT; 1272 if (siginfo != NULL) { 1273 siginfo->si_status = SIGCONT; 1274 siginfo->si_code = CLD_CONTINUED; 1275 } 1276 return (0); 1277 } 1278 PROC_UNLOCK(p); 1279 } 1280 1281 /* 1282 * Look in the orphans list too, to allow the parent to 1283 * collect it's child exit status even if child is being 1284 * debugged. 1285 * 1286 * Debugger detaches from the parent upon successful 1287 * switch-over from parent to child. At this point due to 1288 * re-parenting the parent loses the child to debugger and a 1289 * wait4(2) call would report that it has no children to wait 1290 * for. By maintaining a list of orphans we allow the parent 1291 * to successfully wait until the child becomes a zombie. 1292 */ 1293 if (nfound == 0) { 1294 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1295 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1296 NULL, NULL, 1); 1297 if (ret != 0) { 1298 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1299 (int)td->td_retval[0])); 1300 nfound++; 1301 break; 1302 } 1303 } 1304 } 1305 if (nfound == 0) { 1306 sx_xunlock(&proctree_lock); 1307 return (ECHILD); 1308 } 1309 if (options & WNOHANG) { 1310 sx_xunlock(&proctree_lock); 1311 td->td_retval[0] = 0; 1312 return (0); 1313 } 1314 PROC_LOCK(q); 1315 sx_xunlock(&proctree_lock); 1316 if (q->p_flag & P_STATCHILD) { 1317 q->p_flag &= ~P_STATCHILD; 1318 error = 0; 1319 } else 1320 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1321 PROC_UNLOCK(q); 1322 if (error) 1323 return (error); 1324 goto loop; 1325 } 1326 1327 /* 1328 * Make process 'parent' the new parent of process 'child'. 1329 * Must be called with an exclusive hold of proctree lock. 1330 */ 1331 void 1332 proc_reparent(struct proc *child, struct proc *parent) 1333 { 1334 1335 sx_assert(&proctree_lock, SX_XLOCKED); 1336 PROC_LOCK_ASSERT(child, MA_OWNED); 1337 if (child->p_pptr == parent) 1338 return; 1339 1340 PROC_LOCK(child->p_pptr); 1341 sigqueue_take(child->p_ksi); 1342 PROC_UNLOCK(child->p_pptr); 1343 LIST_REMOVE(child, p_sibling); 1344 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1345 1346 clear_orphan(child); 1347 if (child->p_flag & P_TRACED) { 1348 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1349 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1350 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1351 p_orphan); 1352 } else { 1353 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1354 child, p_orphan); 1355 } 1356 child->p_treeflag |= P_TREE_ORPHANED; 1357 } 1358 1359 child->p_pptr = parent; 1360 } 1361