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_ktrace.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/sysproto.h> 47 #include <sys/capsicum.h> 48 #include <sys/eventhandler.h> 49 #include <sys/kernel.h> 50 #include <sys/malloc.h> 51 #include <sys/lock.h> 52 #include <sys/mutex.h> 53 #include <sys/proc.h> 54 #include <sys/procdesc.h> 55 #include <sys/pioctl.h> 56 #include <sys/jail.h> 57 #include <sys/tty.h> 58 #include <sys/wait.h> 59 #include <sys/vmmeter.h> 60 #include <sys/vnode.h> 61 #include <sys/racct.h> 62 #include <sys/resourcevar.h> 63 #include <sys/sbuf.h> 64 #include <sys/signalvar.h> 65 #include <sys/sched.h> 66 #include <sys/sx.h> 67 #include <sys/syscallsubr.h> 68 #include <sys/syslog.h> 69 #include <sys/ptrace.h> 70 #include <sys/acct.h> /* for acct_process() function prototype */ 71 #include <sys/filedesc.h> 72 #include <sys/sdt.h> 73 #include <sys/shm.h> 74 #include <sys/sem.h> 75 #include <sys/umtx.h> 76 #ifdef KTRACE 77 #include <sys/ktrace.h> 78 #endif 79 80 #include <security/audit/audit.h> 81 #include <security/mac/mac_framework.h> 82 83 #include <vm/vm.h> 84 #include <vm/vm_extern.h> 85 #include <vm/vm_param.h> 86 #include <vm/pmap.h> 87 #include <vm/vm_map.h> 88 #include <vm/vm_page.h> 89 #include <vm/uma.h> 90 91 #ifdef KDTRACE_HOOKS 92 #include <sys/dtrace_bsd.h> 93 dtrace_execexit_func_t dtrace_fasttrap_exit; 94 #endif 95 96 SDT_PROVIDER_DECLARE(proc); 97 SDT_PROBE_DEFINE1(proc, , , exit, "int"); 98 99 /* Hook for NFS teardown procedure. */ 100 void (*nlminfo_release_p)(struct proc *p); 101 102 EVENTHANDLER_LIST_DECLARE(process_exit); 103 104 struct proc * 105 proc_realparent(struct proc *child) 106 { 107 struct proc *p, *parent; 108 109 sx_assert(&proctree_lock, SX_LOCKED); 110 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 111 if (child->p_oppid == 0 || 112 child->p_pptr->p_pid == child->p_oppid) 113 parent = child->p_pptr; 114 else 115 parent = initproc; 116 return (parent); 117 } 118 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 119 /* Cannot use LIST_PREV(), since the list head is not known. */ 120 p = __containerof(p->p_orphan.le_prev, struct proc, 121 p_orphan.le_next); 122 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 123 ("missing P_ORPHAN %p", p)); 124 } 125 parent = __containerof(p->p_orphan.le_prev, struct proc, 126 p_orphans.lh_first); 127 return (parent); 128 } 129 130 void 131 reaper_abandon_children(struct proc *p, bool exiting) 132 { 133 struct proc *p1, *p2, *ptmp; 134 135 sx_assert(&proctree_lock, SX_LOCKED); 136 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 137 if ((p->p_treeflag & P_TREE_REAPER) == 0) 138 return; 139 p1 = p->p_reaper; 140 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 141 LIST_REMOVE(p2, p_reapsibling); 142 p2->p_reaper = p1; 143 p2->p_reapsubtree = p->p_reapsubtree; 144 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 145 if (exiting && p2->p_pptr == p) { 146 PROC_LOCK(p2); 147 proc_reparent(p2, p1); 148 PROC_UNLOCK(p2); 149 } 150 } 151 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 152 p->p_treeflag &= ~P_TREE_REAPER; 153 } 154 155 static void 156 clear_orphan(struct proc *p) 157 { 158 struct proc *p1; 159 160 sx_assert(&proctree_lock, SA_XLOCKED); 161 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 162 return; 163 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 164 p1 = LIST_NEXT(p, p_orphan); 165 if (p1 != NULL) 166 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 167 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 168 } 169 LIST_REMOVE(p, p_orphan); 170 p->p_treeflag &= ~P_TREE_ORPHANED; 171 } 172 173 /* 174 * exit -- death of process. 175 */ 176 void 177 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 178 { 179 180 exit1(td, uap->rval, 0); 181 /* NOTREACHED */ 182 } 183 184 /* 185 * Exit: deallocate address space and other resources, change proc state to 186 * zombie, and unlink proc from allproc and parent's lists. Save exit status 187 * and rusage for wait(). Check for child processes and orphan them. 188 */ 189 void 190 exit1(struct thread *td, int rval, int signo) 191 { 192 struct proc *p, *nq, *q, *t; 193 struct thread *tdt; 194 ksiginfo_t *ksi, *ksi1; 195 int signal_parent; 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 * Call machine-dependent code to release any 427 * machine-dependent resources other than the address space. 428 * The address space is released by "vmspace_exitfree(p)" in 429 * vm_waitproc(). 430 */ 431 cpu_exit(td); 432 433 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 434 435 sx_xlock(&proctree_lock); 436 /* 437 * Remove proc from allproc queue and pidhash chain. 438 * Place onto zombproc. Unlink from parent's child list. 439 */ 440 sx_xlock(&allproc_lock); 441 LIST_REMOVE(p, p_list); 442 LIST_INSERT_HEAD(&zombproc, p, p_list); 443 LIST_REMOVE(p, p_hash); 444 sx_xunlock(&allproc_lock); 445 446 /* 447 * Reparent all children processes: 448 * - traced ones to the original parent (or init if we are that parent) 449 * - the rest to init 450 */ 451 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 } else if (q->p_pdeathsig > 0) { 484 /* 485 * The child asked to received a signal 486 * when we exit. 487 */ 488 kern_psignal(q, q->p_pdeathsig); 489 } 490 } else { 491 /* 492 * Traced processes are killed since their existence 493 * means someone is screwing up. 494 */ 495 t = proc_realparent(q); 496 if (t == p) { 497 proc_reparent(q, q->p_reaper); 498 } else { 499 PROC_LOCK(t); 500 proc_reparent(q, t); 501 PROC_UNLOCK(t); 502 } 503 /* 504 * Since q was found on our children list, the 505 * proc_reparent() call moved q to the orphan 506 * list due to present P_TRACED flag. Clear 507 * orphan link for q now while q is locked. 508 */ 509 clear_orphan(q); 510 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 511 q->p_flag2 &= ~P2_PTRACE_FSTP; 512 q->p_ptevents = 0; 513 FOREACH_THREAD_IN_PROC(q, tdt) { 514 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 515 TDB_FSTP); 516 } 517 kern_psignal(q, SIGKILL); 518 } 519 PROC_UNLOCK(q); 520 if (ksi != NULL) 521 ksiginfo_free(ksi); 522 } 523 524 /* 525 * Also get rid of our orphans. 526 */ 527 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 528 PROC_LOCK(q); 529 /* 530 * If we are the real parent of this process 531 * but it has been reparented to a debugger, then 532 * check if it asked for a signal when we exit. 533 */ 534 if (q->p_pdeathsig > 0) 535 kern_psignal(q, q->p_pdeathsig); 536 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 537 q->p_pid); 538 clear_orphan(q); 539 PROC_UNLOCK(q); 540 } 541 542 /* Save exit status. */ 543 PROC_LOCK(p); 544 p->p_xthread = td; 545 546 #ifdef KDTRACE_HOOKS 547 /* 548 * Tell the DTrace fasttrap provider about the exit if it 549 * has declared an interest. 550 */ 551 if (dtrace_fasttrap_exit) 552 dtrace_fasttrap_exit(p); 553 #endif 554 555 /* 556 * Notify interested parties of our demise. 557 */ 558 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 559 560 #ifdef KDTRACE_HOOKS 561 int reason = CLD_EXITED; 562 if (WCOREDUMP(signo)) 563 reason = CLD_DUMPED; 564 else if (WIFSIGNALED(signo)) 565 reason = CLD_KILLED; 566 SDT_PROBE1(proc, , , exit, reason); 567 #endif 568 569 /* 570 * If this is a process with a descriptor, we may not need to deliver 571 * a signal to the parent. proctree_lock is held over 572 * procdesc_exit() to serialize concurrent calls to close() and 573 * exit(). 574 */ 575 signal_parent = 0; 576 if (p->p_procdesc == NULL || procdesc_exit(p)) { 577 /* 578 * Notify parent that we're gone. If parent has the 579 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 580 * notify process 1 instead (and hope it will handle this 581 * situation). 582 */ 583 PROC_LOCK(p->p_pptr); 584 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 585 if (p->p_pptr->p_sigacts->ps_flag & 586 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 587 struct proc *pp; 588 589 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 590 pp = p->p_pptr; 591 PROC_UNLOCK(pp); 592 proc_reparent(p, p->p_reaper); 593 p->p_sigparent = SIGCHLD; 594 PROC_LOCK(p->p_pptr); 595 596 /* 597 * Notify parent, so in case he was wait(2)ing or 598 * executing waitpid(2) with our pid, he will 599 * continue. 600 */ 601 wakeup(pp); 602 } else 603 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 604 605 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) { 606 signal_parent = 1; 607 } else if (p->p_sigparent != 0) { 608 if (p->p_sigparent == SIGCHLD) { 609 signal_parent = 1; 610 } else { /* LINUX thread */ 611 signal_parent = 2; 612 } 613 } 614 } else 615 PROC_LOCK(p->p_pptr); 616 sx_xunlock(&proctree_lock); 617 618 if (signal_parent == 1) { 619 childproc_exited(p); 620 } else if (signal_parent == 2) { 621 kern_psignal(p->p_pptr, p->p_sigparent); 622 } 623 624 /* Tell the prison that we are gone. */ 625 prison_proc_free(p->p_ucred->cr_prison); 626 627 /* 628 * The state PRS_ZOMBIE prevents other proesses from sending 629 * signal to the process, to avoid memory leak, we free memory 630 * for signal queue at the time when the state is set. 631 */ 632 sigqueue_flush(&p->p_sigqueue); 633 sigqueue_flush(&td->td_sigqueue); 634 635 /* 636 * We have to wait until after acquiring all locks before 637 * changing p_state. We need to avoid all possible context 638 * switches (including ones from blocking on a mutex) while 639 * marked as a zombie. We also have to set the zombie state 640 * before we release the parent process' proc lock to avoid 641 * a lost wakeup. So, we first call wakeup, then we grab the 642 * sched lock, update the state, and release the parent process' 643 * proc lock. 644 */ 645 wakeup(p->p_pptr); 646 cv_broadcast(&p->p_pwait); 647 sched_exit(p->p_pptr, td); 648 PROC_SLOCK(p); 649 p->p_state = PRS_ZOMBIE; 650 PROC_UNLOCK(p->p_pptr); 651 652 /* 653 * Save our children's rusage information in our exit rusage. 654 */ 655 PROC_STATLOCK(p); 656 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 657 PROC_STATUNLOCK(p); 658 659 /* 660 * Make sure the scheduler takes this thread out of its tables etc. 661 * This will also release this thread's reference to the ucred. 662 * Other thread parts to release include pcb bits and such. 663 */ 664 thread_exit(); 665 } 666 667 668 #ifndef _SYS_SYSPROTO_H_ 669 struct abort2_args { 670 char *why; 671 int nargs; 672 void **args; 673 }; 674 #endif 675 676 int 677 sys_abort2(struct thread *td, struct abort2_args *uap) 678 { 679 struct proc *p = td->td_proc; 680 struct sbuf *sb; 681 void *uargs[16]; 682 int error, i, sig; 683 684 /* 685 * Do it right now so we can log either proper call of abort2(), or 686 * note, that invalid argument was passed. 512 is big enough to 687 * handle 16 arguments' descriptions with additional comments. 688 */ 689 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 690 sbuf_clear(sb); 691 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 692 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 693 /* 694 * Since we can't return from abort2(), send SIGKILL in cases, where 695 * abort2() was called improperly 696 */ 697 sig = SIGKILL; 698 /* Prevent from DoSes from user-space. */ 699 if (uap->nargs < 0 || uap->nargs > 16) 700 goto out; 701 if (uap->nargs > 0) { 702 if (uap->args == NULL) 703 goto out; 704 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 705 if (error != 0) 706 goto out; 707 } 708 /* 709 * Limit size of 'reason' string to 128. Will fit even when 710 * maximal number of arguments was chosen to be logged. 711 */ 712 if (uap->why != NULL) { 713 error = sbuf_copyin(sb, uap->why, 128); 714 if (error < 0) 715 goto out; 716 } else { 717 sbuf_printf(sb, "(null)"); 718 } 719 if (uap->nargs > 0) { 720 sbuf_printf(sb, "("); 721 for (i = 0;i < uap->nargs; i++) 722 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 723 sbuf_printf(sb, ")"); 724 } 725 /* 726 * Final stage: arguments were proper, string has been 727 * successfully copied from userspace, and copying pointers 728 * from user-space succeed. 729 */ 730 sig = SIGABRT; 731 out: 732 if (sig == SIGKILL) { 733 sbuf_trim(sb); 734 sbuf_printf(sb, " (Reason text inaccessible)"); 735 } 736 sbuf_cat(sb, "\n"); 737 sbuf_finish(sb); 738 log(LOG_INFO, "%s", sbuf_data(sb)); 739 sbuf_delete(sb); 740 exit1(td, 0, sig); 741 return (0); 742 } 743 744 745 #ifdef COMPAT_43 746 /* 747 * The dirty work is handled by kern_wait(). 748 */ 749 int 750 owait(struct thread *td, struct owait_args *uap __unused) 751 { 752 int error, status; 753 754 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 755 if (error == 0) 756 td->td_retval[1] = status; 757 return (error); 758 } 759 #endif /* COMPAT_43 */ 760 761 /* 762 * The dirty work is handled by kern_wait(). 763 */ 764 int 765 sys_wait4(struct thread *td, struct wait4_args *uap) 766 { 767 struct rusage ru, *rup; 768 int error, status; 769 770 if (uap->rusage != NULL) 771 rup = &ru; 772 else 773 rup = NULL; 774 error = kern_wait(td, uap->pid, &status, uap->options, rup); 775 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 776 error = copyout(&status, uap->status, sizeof(status)); 777 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 778 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 779 return (error); 780 } 781 782 int 783 sys_wait6(struct thread *td, struct wait6_args *uap) 784 { 785 struct __wrusage wru, *wrup; 786 siginfo_t si, *sip; 787 idtype_t idtype; 788 id_t id; 789 int error, status; 790 791 idtype = uap->idtype; 792 id = uap->id; 793 794 if (uap->wrusage != NULL) 795 wrup = &wru; 796 else 797 wrup = NULL; 798 799 if (uap->info != NULL) { 800 sip = &si; 801 bzero(sip, sizeof(*sip)); 802 } else 803 sip = NULL; 804 805 /* 806 * We expect all callers of wait6() to know about WEXITED and 807 * WTRAPPED. 808 */ 809 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 810 811 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 812 error = copyout(&status, uap->status, sizeof(status)); 813 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 814 error = copyout(&wru, uap->wrusage, sizeof(wru)); 815 if (uap->info != NULL && error == 0) 816 error = copyout(&si, uap->info, sizeof(si)); 817 return (error); 818 } 819 820 /* 821 * Reap the remains of a zombie process and optionally return status and 822 * rusage. Asserts and will release both the proctree_lock and the process 823 * lock as part of its work. 824 */ 825 void 826 proc_reap(struct thread *td, struct proc *p, int *status, int options) 827 { 828 struct proc *q, *t; 829 830 sx_assert(&proctree_lock, SA_XLOCKED); 831 PROC_LOCK_ASSERT(p, MA_OWNED); 832 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 833 834 mtx_spin_wait_unlocked(&p->p_slock); 835 836 q = td->td_proc; 837 838 if (status) 839 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 840 if (options & WNOWAIT) { 841 /* 842 * Only poll, returning the status. Caller does not wish to 843 * release the proc struct just yet. 844 */ 845 PROC_UNLOCK(p); 846 sx_xunlock(&proctree_lock); 847 return; 848 } 849 850 PROC_LOCK(q); 851 sigqueue_take(p->p_ksi); 852 PROC_UNLOCK(q); 853 854 /* 855 * If we got the child via a ptrace 'attach', we need to give it back 856 * to the old parent. 857 */ 858 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 859 PROC_UNLOCK(p); 860 t = proc_realparent(p); 861 PROC_LOCK(t); 862 PROC_LOCK(p); 863 CTR2(KTR_PTRACE, 864 "wait: traced child %d moved back to parent %d", p->p_pid, 865 t->p_pid); 866 proc_reparent(p, t); 867 p->p_oppid = 0; 868 PROC_UNLOCK(p); 869 pksignal(t, SIGCHLD, p->p_ksi); 870 wakeup(t); 871 cv_broadcast(&p->p_pwait); 872 PROC_UNLOCK(t); 873 sx_xunlock(&proctree_lock); 874 return; 875 } 876 p->p_oppid = 0; 877 PROC_UNLOCK(p); 878 879 /* 880 * Remove other references to this process to ensure we have an 881 * exclusive reference. 882 */ 883 sx_xlock(&allproc_lock); 884 LIST_REMOVE(p, p_list); /* off zombproc */ 885 sx_xunlock(&allproc_lock); 886 LIST_REMOVE(p, p_sibling); 887 reaper_abandon_children(p, true); 888 LIST_REMOVE(p, p_reapsibling); 889 PROC_LOCK(p); 890 clear_orphan(p); 891 PROC_UNLOCK(p); 892 leavepgrp(p); 893 if (p->p_procdesc != NULL) 894 procdesc_reap(p); 895 sx_xunlock(&proctree_lock); 896 897 PROC_LOCK(p); 898 knlist_detach(p->p_klist); 899 p->p_klist = NULL; 900 PROC_UNLOCK(p); 901 902 /* 903 * Removal from allproc list and process group list paired with 904 * PROC_LOCK which was executed during that time should guarantee 905 * nothing can reach this process anymore. As such further locking 906 * is unnecessary. 907 */ 908 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 909 910 PROC_LOCK(q); 911 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 912 PROC_UNLOCK(q); 913 914 /* 915 * Decrement the count of procs running with this uid. 916 */ 917 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 918 919 /* 920 * Destroy resource accounting information associated with the process. 921 */ 922 #ifdef RACCT 923 if (racct_enable) { 924 PROC_LOCK(p); 925 racct_sub(p, RACCT_NPROC, 1); 926 PROC_UNLOCK(p); 927 } 928 #endif 929 racct_proc_exit(p); 930 931 /* 932 * Free credentials, arguments, and sigacts. 933 */ 934 crfree(p->p_ucred); 935 proc_set_cred(p, NULL); 936 pargs_drop(p->p_args); 937 p->p_args = NULL; 938 sigacts_free(p->p_sigacts); 939 p->p_sigacts = NULL; 940 941 /* 942 * Do any thread-system specific cleanups. 943 */ 944 thread_wait(p); 945 946 /* 947 * Give vm and machine-dependent layer a chance to free anything that 948 * cpu_exit couldn't release while still running in process context. 949 */ 950 vm_waitproc(p); 951 #ifdef MAC 952 mac_proc_destroy(p); 953 #endif 954 955 KASSERT(FIRST_THREAD_IN_PROC(p), 956 ("proc_reap: no residual thread!")); 957 uma_zfree(proc_zone, p); 958 atomic_add_int(&nprocs, -1); 959 } 960 961 static int 962 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 963 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 964 int check_only) 965 { 966 struct rusage *rup; 967 968 sx_assert(&proctree_lock, SA_XLOCKED); 969 970 PROC_LOCK(p); 971 972 switch (idtype) { 973 case P_ALL: 974 if (p->p_procdesc != NULL) { 975 PROC_UNLOCK(p); 976 return (0); 977 } 978 break; 979 case P_PID: 980 if (p->p_pid != (pid_t)id) { 981 PROC_UNLOCK(p); 982 return (0); 983 } 984 break; 985 case P_PGID: 986 if (p->p_pgid != (pid_t)id) { 987 PROC_UNLOCK(p); 988 return (0); 989 } 990 break; 991 case P_SID: 992 if (p->p_session->s_sid != (pid_t)id) { 993 PROC_UNLOCK(p); 994 return (0); 995 } 996 break; 997 case P_UID: 998 if (p->p_ucred->cr_uid != (uid_t)id) { 999 PROC_UNLOCK(p); 1000 return (0); 1001 } 1002 break; 1003 case P_GID: 1004 if (p->p_ucred->cr_gid != (gid_t)id) { 1005 PROC_UNLOCK(p); 1006 return (0); 1007 } 1008 break; 1009 case P_JAILID: 1010 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1011 PROC_UNLOCK(p); 1012 return (0); 1013 } 1014 break; 1015 /* 1016 * It seems that the thread structures get zeroed out 1017 * at process exit. This makes it impossible to 1018 * support P_SETID, P_CID or P_CPUID. 1019 */ 1020 default: 1021 PROC_UNLOCK(p); 1022 return (0); 1023 } 1024 1025 if (p_canwait(td, p)) { 1026 PROC_UNLOCK(p); 1027 return (0); 1028 } 1029 1030 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1031 PROC_UNLOCK(p); 1032 return (0); 1033 } 1034 1035 /* 1036 * This special case handles a kthread spawned by linux_clone 1037 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1038 * functions need to be able to distinguish between waiting 1039 * on a process and waiting on a thread. It is a thread if 1040 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1041 * signifies we want to wait for threads and not processes. 1042 */ 1043 if ((p->p_sigparent != SIGCHLD) ^ 1044 ((options & WLINUXCLONE) != 0)) { 1045 PROC_UNLOCK(p); 1046 return (0); 1047 } 1048 1049 if (siginfo != NULL) { 1050 bzero(siginfo, sizeof(*siginfo)); 1051 siginfo->si_errno = 0; 1052 1053 /* 1054 * SUSv4 requires that the si_signo value is always 1055 * SIGCHLD. Obey it despite the rfork(2) interface 1056 * allows to request other signal for child exit 1057 * notification. 1058 */ 1059 siginfo->si_signo = SIGCHLD; 1060 1061 /* 1062 * This is still a rough estimate. We will fix the 1063 * cases TRAPPED, STOPPED, and CONTINUED later. 1064 */ 1065 if (WCOREDUMP(p->p_xsig)) { 1066 siginfo->si_code = CLD_DUMPED; 1067 siginfo->si_status = WTERMSIG(p->p_xsig); 1068 } else if (WIFSIGNALED(p->p_xsig)) { 1069 siginfo->si_code = CLD_KILLED; 1070 siginfo->si_status = WTERMSIG(p->p_xsig); 1071 } else { 1072 siginfo->si_code = CLD_EXITED; 1073 siginfo->si_status = p->p_xexit; 1074 } 1075 1076 siginfo->si_pid = p->p_pid; 1077 siginfo->si_uid = p->p_ucred->cr_uid; 1078 1079 /* 1080 * The si_addr field would be useful additional 1081 * detail, but apparently the PC value may be lost 1082 * when we reach this point. bzero() above sets 1083 * siginfo->si_addr to NULL. 1084 */ 1085 } 1086 1087 /* 1088 * There should be no reason to limit resources usage info to 1089 * exited processes only. A snapshot about any resources used 1090 * by a stopped process may be exactly what is needed. 1091 */ 1092 if (wrusage != NULL) { 1093 rup = &wrusage->wru_self; 1094 *rup = p->p_ru; 1095 PROC_STATLOCK(p); 1096 calcru(p, &rup->ru_utime, &rup->ru_stime); 1097 PROC_STATUNLOCK(p); 1098 1099 rup = &wrusage->wru_children; 1100 *rup = p->p_stats->p_cru; 1101 calccru(p, &rup->ru_utime, &rup->ru_stime); 1102 } 1103 1104 if (p->p_state == PRS_ZOMBIE && !check_only) { 1105 proc_reap(td, p, status, options); 1106 return (-1); 1107 } 1108 return (1); 1109 } 1110 1111 int 1112 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1113 struct rusage *rusage) 1114 { 1115 struct __wrusage wru, *wrup; 1116 idtype_t idtype; 1117 id_t id; 1118 int ret; 1119 1120 /* 1121 * Translate the special pid values into the (idtype, pid) 1122 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1123 * kern_wait6() on its own. 1124 */ 1125 if (pid == WAIT_ANY) { 1126 idtype = P_ALL; 1127 id = 0; 1128 } else if (pid < 0) { 1129 idtype = P_PGID; 1130 id = (id_t)-pid; 1131 } else { 1132 idtype = P_PID; 1133 id = (id_t)pid; 1134 } 1135 1136 if (rusage != NULL) 1137 wrup = &wru; 1138 else 1139 wrup = NULL; 1140 1141 /* 1142 * For backward compatibility we implicitly add flags WEXITED 1143 * and WTRAPPED here. 1144 */ 1145 options |= WEXITED | WTRAPPED; 1146 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1147 if (rusage != NULL) 1148 *rusage = wru.wru_self; 1149 return (ret); 1150 } 1151 1152 static void 1153 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1154 int *status, int options, int si_code) 1155 { 1156 bool cont; 1157 1158 PROC_LOCK_ASSERT(p, MA_OWNED); 1159 sx_assert(&proctree_lock, SA_XLOCKED); 1160 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1161 si_code == CLD_CONTINUED); 1162 1163 cont = si_code == CLD_CONTINUED; 1164 if ((options & WNOWAIT) == 0) { 1165 if (cont) 1166 p->p_flag &= ~P_CONTINUED; 1167 else 1168 p->p_flag |= P_WAITED; 1169 PROC_LOCK(td->td_proc); 1170 sigqueue_take(p->p_ksi); 1171 PROC_UNLOCK(td->td_proc); 1172 } 1173 sx_xunlock(&proctree_lock); 1174 if (siginfo != NULL) { 1175 siginfo->si_code = si_code; 1176 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1177 } 1178 if (status != NULL) 1179 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1180 PROC_UNLOCK(p); 1181 td->td_retval[0] = p->p_pid; 1182 } 1183 1184 int 1185 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1186 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1187 { 1188 struct proc *p, *q; 1189 pid_t pid; 1190 int error, nfound, ret; 1191 bool report; 1192 1193 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1194 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1195 AUDIT_ARG_VALUE(options); 1196 1197 q = td->td_proc; 1198 1199 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1200 PROC_LOCK(q); 1201 id = (id_t)q->p_pgid; 1202 PROC_UNLOCK(q); 1203 idtype = P_PGID; 1204 } 1205 1206 /* If we don't know the option, just return. */ 1207 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1208 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1209 return (EINVAL); 1210 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1211 /* 1212 * We will be unable to find any matching processes, 1213 * because there are no known events to look for. 1214 * Prefer to return error instead of blocking 1215 * indefinitely. 1216 */ 1217 return (EINVAL); 1218 } 1219 1220 loop: 1221 if (q->p_flag & P_STATCHILD) { 1222 PROC_LOCK(q); 1223 q->p_flag &= ~P_STATCHILD; 1224 PROC_UNLOCK(q); 1225 } 1226 sx_xlock(&proctree_lock); 1227 loop_locked: 1228 nfound = 0; 1229 LIST_FOREACH(p, &q->p_children, p_sibling) { 1230 pid = p->p_pid; 1231 ret = proc_to_reap(td, p, idtype, id, status, options, 1232 wrusage, siginfo, 0); 1233 if (ret == 0) 1234 continue; 1235 else if (ret != 1) { 1236 td->td_retval[0] = pid; 1237 return (0); 1238 } 1239 1240 nfound++; 1241 PROC_LOCK_ASSERT(p, MA_OWNED); 1242 1243 if ((options & WTRAPPED) != 0 && 1244 (p->p_flag & P_TRACED) != 0) { 1245 PROC_SLOCK(p); 1246 report = 1247 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) && 1248 p->p_suspcount == p->p_numthreads && 1249 (p->p_flag & P_WAITED) == 0); 1250 PROC_SUNLOCK(p); 1251 if (report) { 1252 CTR4(KTR_PTRACE, 1253 "wait: returning trapped pid %d status %#x " 1254 "(xstat %d) xthread %d", 1255 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1256 p->p_xthread != NULL ? 1257 p->p_xthread->td_tid : -1); 1258 report_alive_proc(td, p, siginfo, status, 1259 options, CLD_TRAPPED); 1260 return (0); 1261 } 1262 } 1263 if ((options & WUNTRACED) != 0 && 1264 (p->p_flag & P_STOPPED_SIG) != 0) { 1265 PROC_SLOCK(p); 1266 report = (p->p_suspcount == p->p_numthreads && 1267 ((p->p_flag & P_WAITED) == 0)); 1268 PROC_SUNLOCK(p); 1269 if (report) { 1270 report_alive_proc(td, p, siginfo, status, 1271 options, CLD_STOPPED); 1272 return (0); 1273 } 1274 } 1275 if ((options & WCONTINUED) != 0 && 1276 (p->p_flag & P_CONTINUED) != 0) { 1277 report_alive_proc(td, p, siginfo, status, options, 1278 CLD_CONTINUED); 1279 return (0); 1280 } 1281 PROC_UNLOCK(p); 1282 } 1283 1284 /* 1285 * Look in the orphans list too, to allow the parent to 1286 * collect it's child exit status even if child is being 1287 * debugged. 1288 * 1289 * Debugger detaches from the parent upon successful 1290 * switch-over from parent to child. At this point due to 1291 * re-parenting the parent loses the child to debugger and a 1292 * wait4(2) call would report that it has no children to wait 1293 * for. By maintaining a list of orphans we allow the parent 1294 * to successfully wait until the child becomes a zombie. 1295 */ 1296 if (nfound == 0) { 1297 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1298 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1299 NULL, NULL, 1); 1300 if (ret != 0) { 1301 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1302 (int)td->td_retval[0])); 1303 PROC_UNLOCK(p); 1304 nfound++; 1305 break; 1306 } 1307 } 1308 } 1309 if (nfound == 0) { 1310 sx_xunlock(&proctree_lock); 1311 return (ECHILD); 1312 } 1313 if (options & WNOHANG) { 1314 sx_xunlock(&proctree_lock); 1315 td->td_retval[0] = 0; 1316 return (0); 1317 } 1318 PROC_LOCK(q); 1319 if (q->p_flag & P_STATCHILD) { 1320 q->p_flag &= ~P_STATCHILD; 1321 PROC_UNLOCK(q); 1322 goto loop_locked; 1323 } 1324 sx_xunlock(&proctree_lock); 1325 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0); 1326 if (error) 1327 return (error); 1328 goto loop; 1329 } 1330 1331 /* 1332 * Make process 'parent' the new parent of process 'child'. 1333 * Must be called with an exclusive hold of proctree lock. 1334 */ 1335 void 1336 proc_reparent(struct proc *child, struct proc *parent) 1337 { 1338 1339 sx_assert(&proctree_lock, SX_XLOCKED); 1340 PROC_LOCK_ASSERT(child, MA_OWNED); 1341 if (child->p_pptr == parent) 1342 return; 1343 1344 PROC_LOCK(child->p_pptr); 1345 sigqueue_take(child->p_ksi); 1346 PROC_UNLOCK(child->p_pptr); 1347 LIST_REMOVE(child, p_sibling); 1348 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1349 1350 clear_orphan(child); 1351 if (child->p_flag & P_TRACED) { 1352 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1353 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1354 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1355 p_orphan); 1356 } else { 1357 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1358 child, p_orphan); 1359 } 1360 child->p_treeflag |= P_TREE_ORPHANED; 1361 } 1362 1363 child->p_pptr = parent; 1364 } 1365