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/ktr.h> 51 #include <sys/malloc.h> 52 #include <sys/lock.h> 53 #include <sys/mutex.h> 54 #include <sys/proc.h> 55 #include <sys/procdesc.h> 56 #include <sys/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 struct proc * 100 proc_realparent(struct proc *child) 101 { 102 struct proc *p, *parent; 103 104 sx_assert(&proctree_lock, SX_LOCKED); 105 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) 106 return (child->p_pptr->p_pid == child->p_oppid ? 107 child->p_pptr : initproc); 108 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 109 /* Cannot use LIST_PREV(), since the list head is not known. */ 110 p = __containerof(p->p_orphan.le_prev, struct proc, 111 p_orphan.le_next); 112 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 113 ("missing P_ORPHAN %p", p)); 114 } 115 parent = __containerof(p->p_orphan.le_prev, struct proc, 116 p_orphans.lh_first); 117 return (parent); 118 } 119 120 void 121 reaper_abandon_children(struct proc *p, bool exiting) 122 { 123 struct proc *p1, *p2, *ptmp; 124 125 sx_assert(&proctree_lock, SX_LOCKED); 126 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 127 if ((p->p_treeflag & P_TREE_REAPER) == 0) 128 return; 129 p1 = p->p_reaper; 130 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 131 LIST_REMOVE(p2, p_reapsibling); 132 p2->p_reaper = p1; 133 p2->p_reapsubtree = p->p_reapsubtree; 134 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 135 if (exiting && p2->p_pptr == p) { 136 PROC_LOCK(p2); 137 proc_reparent(p2, p1, true); 138 PROC_UNLOCK(p2); 139 } 140 } 141 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 142 p->p_treeflag &= ~P_TREE_REAPER; 143 } 144 145 static void 146 reaper_clear(struct proc *p) 147 { 148 struct proc *p1; 149 bool clear; 150 151 sx_assert(&proctree_lock, SX_LOCKED); 152 LIST_REMOVE(p, p_reapsibling); 153 if (p->p_reapsubtree == 1) 154 return; 155 clear = true; 156 LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) { 157 if (p1->p_reapsubtree == p->p_reapsubtree) { 158 clear = false; 159 break; 160 } 161 } 162 if (clear) 163 proc_id_clear(PROC_ID_REAP, p->p_reapsubtree); 164 } 165 166 void 167 proc_clear_orphan(struct proc *p) 168 { 169 struct proc *p1; 170 171 sx_assert(&proctree_lock, SA_XLOCKED); 172 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 173 return; 174 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 175 p1 = LIST_NEXT(p, p_orphan); 176 if (p1 != NULL) 177 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 178 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 179 } 180 LIST_REMOVE(p, p_orphan); 181 p->p_treeflag &= ~P_TREE_ORPHANED; 182 } 183 184 /* 185 * exit -- death of process. 186 */ 187 void 188 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 189 { 190 191 exit1(td, uap->rval, 0); 192 /* NOTREACHED */ 193 } 194 195 /* 196 * Exit: deallocate address space and other resources, change proc state to 197 * zombie, and unlink proc from allproc and parent's lists. Save exit status 198 * and rusage for wait(). Check for child processes and orphan them. 199 */ 200 void 201 exit1(struct thread *td, int rval, int signo) 202 { 203 struct proc *p, *nq, *q, *t; 204 struct thread *tdt; 205 ksiginfo_t *ksi, *ksi1; 206 int signal_parent; 207 208 mtx_assert(&Giant, MA_NOTOWNED); 209 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 210 211 p = td->td_proc; 212 /* 213 * XXX in case we're rebooting we just let init die in order to 214 * work around an unsolved stack overflow seen very late during 215 * shutdown on sparc64 when the gmirror worker process exists. 216 * XXX what to do now that sparc64 is gone... remove if? 217 */ 218 if (p == initproc && rebooting == 0) { 219 printf("init died (signal %d, exit %d)\n", signo, rval); 220 panic("Going nowhere without my init!"); 221 } 222 223 /* 224 * Deref SU mp, since the thread does not return to userspace. 225 */ 226 td_softdep_cleanup(td); 227 228 /* 229 * MUST abort all other threads before proceeding past here. 230 */ 231 PROC_LOCK(p); 232 /* 233 * First check if some other thread or external request got 234 * here before us. If so, act appropriately: exit or suspend. 235 * We must ensure that stop requests are handled before we set 236 * P_WEXIT. 237 */ 238 thread_suspend_check(0); 239 while (p->p_flag & P_HADTHREADS) { 240 /* 241 * Kill off the other threads. This requires 242 * some co-operation from other parts of the kernel 243 * so it may not be instantaneous. With this state set 244 * any thread entering the kernel from userspace will 245 * thread_exit() in trap(). Any thread attempting to 246 * sleep will return immediately with EINTR or EWOULDBLOCK 247 * which will hopefully force them to back out to userland 248 * freeing resources as they go. Any thread attempting 249 * to return to userland will thread_exit() from userret(). 250 * thread_exit() will unsuspend us when the last of the 251 * other threads exits. 252 * If there is already a thread singler after resumption, 253 * calling thread_single will fail; in that case, we just 254 * re-check all suspension request, the thread should 255 * either be suspended there or exit. 256 */ 257 if (!thread_single(p, SINGLE_EXIT)) 258 /* 259 * All other activity in this process is now 260 * stopped. Threading support has been turned 261 * off. 262 */ 263 break; 264 /* 265 * Recheck for new stop or suspend requests which 266 * might appear while process lock was dropped in 267 * thread_single(). 268 */ 269 thread_suspend_check(0); 270 } 271 KASSERT(p->p_numthreads == 1, 272 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 273 racct_sub(p, RACCT_NTHR, 1); 274 275 /* Let event handler change exit status */ 276 p->p_xexit = rval; 277 p->p_xsig = signo; 278 279 /* 280 * Ignore any pending request to stop due to a stop signal. 281 * Once P_WEXIT is set, future requests will be ignored as 282 * well. 283 */ 284 p->p_flag &= ~P_STOPPED_SIG; 285 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 286 287 /* Note that we are exiting. */ 288 p->p_flag |= P_WEXIT; 289 290 /* 291 * Wait for any processes that have a hold on our vmspace to 292 * release their reference. 293 */ 294 while (p->p_lock > 0) 295 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 296 297 PROC_UNLOCK(p); 298 /* Drain the limit callout while we don't have the proc locked */ 299 callout_drain(&p->p_limco); 300 301 #ifdef AUDIT 302 /* 303 * The Sun BSM exit token contains two components: an exit status as 304 * passed to exit(), and a return value to indicate what sort of exit 305 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 306 * what the return value is. 307 */ 308 AUDIT_ARG_EXIT(rval, 0); 309 AUDIT_SYSCALL_EXIT(0, td); 310 #endif 311 312 /* Are we a task leader with peers? */ 313 if (p->p_peers != NULL && p == p->p_leader) { 314 mtx_lock(&ppeers_lock); 315 q = p->p_peers; 316 while (q != NULL) { 317 PROC_LOCK(q); 318 kern_psignal(q, SIGKILL); 319 PROC_UNLOCK(q); 320 q = q->p_peers; 321 } 322 while (p->p_peers != NULL) 323 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 324 mtx_unlock(&ppeers_lock); 325 } 326 327 /* 328 * Check if any loadable modules need anything done at process exit. 329 * E.g. SYSV IPC stuff. 330 * Event handler could change exit status. 331 * XXX what if one of these generates an error? 332 */ 333 EVENTHANDLER_DIRECT_INVOKE(process_exit, p); 334 335 /* 336 * If parent is waiting for us to exit or exec, 337 * P_PPWAIT is set; we will wakeup the parent below. 338 */ 339 PROC_LOCK(p); 340 stopprofclock(p); 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 * Close open files and release open-file table. 367 * This may block! 368 */ 369 fdescfree(td); 370 371 /* 372 * If this thread tickled GEOM, we need to wait for the giggling to 373 * stop before we return to userland 374 */ 375 if (td->td_pflags & TDP_GEOM) 376 g_waitidle(); 377 378 /* 379 * Remove ourself from our leader's peer list and wake our leader. 380 */ 381 if (p->p_leader->p_peers != NULL) { 382 mtx_lock(&ppeers_lock); 383 if (p->p_leader->p_peers != NULL) { 384 q = p->p_leader; 385 while (q->p_peers != p) 386 q = q->p_peers; 387 q->p_peers = p->p_peers; 388 wakeup(p->p_leader); 389 } 390 mtx_unlock(&ppeers_lock); 391 } 392 393 vmspace_exit(td); 394 (void)acct_process(td); 395 396 #ifdef KTRACE 397 ktrprocexit(td); 398 #endif 399 /* 400 * Release reference to text vnode 401 */ 402 if (p->p_textvp != NULL) { 403 vrele(p->p_textvp); 404 p->p_textvp = NULL; 405 } 406 407 /* 408 * Release our limits structure. 409 */ 410 lim_free(p->p_limit); 411 p->p_limit = NULL; 412 413 tidhash_remove(td); 414 415 /* 416 * Call machine-dependent code to release any 417 * machine-dependent resources other than the address space. 418 * The address space is released by "vmspace_exitfree(p)" in 419 * vm_waitproc(). 420 */ 421 cpu_exit(td); 422 423 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 424 425 /* 426 * Remove from allproc. It still sits in the hash. 427 */ 428 sx_xlock(&allproc_lock); 429 LIST_REMOVE(p, p_list); 430 sx_xunlock(&allproc_lock); 431 432 sx_xlock(&proctree_lock); 433 PROC_LOCK(p); 434 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 435 PROC_UNLOCK(p); 436 437 /* 438 * killjobc() might drop and re-acquire proctree_lock to 439 * revoke control tty if exiting process was a session leader. 440 */ 441 killjobc(); 442 443 /* 444 * Reparent all children processes: 445 * - traced ones to the original parent (or init if we are that parent) 446 * - the rest to init 447 */ 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) == 0) { 458 proc_reparent(q, q->p_reaper, true); 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 } else if (q->p_pdeathsig > 0) { 481 /* 482 * The child asked to received a signal 483 * when we exit. 484 */ 485 kern_psignal(q, q->p_pdeathsig); 486 } 487 } else { 488 /* 489 * Traced processes are killed since their existence 490 * means someone is screwing up. 491 */ 492 t = proc_realparent(q); 493 if (t == p) { 494 proc_reparent(q, q->p_reaper, true); 495 } else { 496 PROC_LOCK(t); 497 proc_reparent(q, t, true); 498 PROC_UNLOCK(t); 499 } 500 /* 501 * Since q was found on our children list, the 502 * proc_reparent() call moved q to the orphan 503 * list due to present P_TRACED flag. Clear 504 * orphan link for q now while q is locked. 505 */ 506 proc_clear_orphan(q); 507 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 508 q->p_flag2 &= ~P2_PTRACE_FSTP; 509 q->p_ptevents = 0; 510 FOREACH_THREAD_IN_PROC(q, tdt) { 511 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 512 TDB_FSTP); 513 } 514 kern_psignal(q, SIGKILL); 515 } 516 PROC_UNLOCK(q); 517 if (ksi != NULL) 518 ksiginfo_free(ksi); 519 } 520 521 /* 522 * Also get rid of our orphans. 523 */ 524 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 525 PROC_LOCK(q); 526 KASSERT(q->p_oppid == p->p_pid, 527 ("orphan %p of %p has unexpected oppid %d", q, p, 528 q->p_oppid)); 529 q->p_oppid = q->p_reaper->p_pid; 530 531 /* 532 * If we are the real parent of this process 533 * but it has been reparented to a debugger, then 534 * check if it asked for a signal when we exit. 535 */ 536 if (q->p_pdeathsig > 0) 537 kern_psignal(q, q->p_pdeathsig); 538 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 539 q->p_pid); 540 proc_clear_orphan(q); 541 PROC_UNLOCK(q); 542 } 543 544 #ifdef KDTRACE_HOOKS 545 if (SDT_PROBES_ENABLED()) { 546 int reason = CLD_EXITED; 547 if (WCOREDUMP(signo)) 548 reason = CLD_DUMPED; 549 else if (WIFSIGNALED(signo)) 550 reason = CLD_KILLED; 551 SDT_PROBE1(proc, , , exit, reason); 552 } 553 #endif 554 555 /* Save exit status. */ 556 PROC_LOCK(p); 557 p->p_xthread = td; 558 559 #ifdef KDTRACE_HOOKS 560 /* 561 * Tell the DTrace fasttrap provider about the exit if it 562 * has declared an interest. 563 */ 564 if (dtrace_fasttrap_exit) 565 dtrace_fasttrap_exit(p); 566 #endif 567 568 /* 569 * Notify interested parties of our demise. 570 */ 571 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 572 573 /* 574 * If this is a process with a descriptor, we may not need to deliver 575 * a signal to the parent. proctree_lock is held over 576 * procdesc_exit() to serialize concurrent calls to close() and 577 * exit(). 578 */ 579 signal_parent = 0; 580 if (p->p_procdesc == NULL || procdesc_exit(p)) { 581 /* 582 * Notify parent that we're gone. If parent has the 583 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 584 * notify process 1 instead (and hope it will handle this 585 * situation). 586 */ 587 PROC_LOCK(p->p_pptr); 588 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 589 if (p->p_pptr->p_sigacts->ps_flag & 590 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 591 struct proc *pp; 592 593 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 594 pp = p->p_pptr; 595 PROC_UNLOCK(pp); 596 proc_reparent(p, p->p_reaper, true); 597 p->p_sigparent = SIGCHLD; 598 PROC_LOCK(p->p_pptr); 599 600 /* 601 * Notify parent, so in case he was wait(2)ing or 602 * executing waitpid(2) with our pid, he will 603 * continue. 604 */ 605 wakeup(pp); 606 } else 607 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 608 609 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) { 610 signal_parent = 1; 611 } else if (p->p_sigparent != 0) { 612 if (p->p_sigparent == SIGCHLD) { 613 signal_parent = 1; 614 } else { /* LINUX thread */ 615 signal_parent = 2; 616 } 617 } 618 } else 619 PROC_LOCK(p->p_pptr); 620 sx_xunlock(&proctree_lock); 621 622 if (signal_parent == 1) { 623 childproc_exited(p); 624 } else if (signal_parent == 2) { 625 kern_psignal(p->p_pptr, p->p_sigparent); 626 } 627 628 /* Tell the prison that we are gone. */ 629 prison_proc_free(p->p_ucred->cr_prison); 630 631 /* 632 * The state PRS_ZOMBIE prevents other proesses from sending 633 * signal to the process, to avoid memory leak, we free memory 634 * for signal queue at the time when the state is set. 635 */ 636 sigqueue_flush(&p->p_sigqueue); 637 sigqueue_flush(&td->td_sigqueue); 638 639 /* 640 * We have to wait until after acquiring all locks before 641 * changing p_state. We need to avoid all possible context 642 * switches (including ones from blocking on a mutex) while 643 * marked as a zombie. We also have to set the zombie state 644 * before we release the parent process' proc lock to avoid 645 * a lost wakeup. So, we first call wakeup, then we grab the 646 * sched lock, update the state, and release the parent process' 647 * proc lock. 648 */ 649 wakeup(p->p_pptr); 650 cv_broadcast(&p->p_pwait); 651 sched_exit(p->p_pptr, td); 652 PROC_SLOCK(p); 653 p->p_state = PRS_ZOMBIE; 654 PROC_UNLOCK(p->p_pptr); 655 656 /* 657 * Save our children's rusage information in our exit rusage. 658 */ 659 PROC_STATLOCK(p); 660 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 661 PROC_STATUNLOCK(p); 662 663 /* 664 * Make sure the scheduler takes this thread out of its tables etc. 665 * This will also release this thread's reference to the ucred. 666 * Other thread parts to release include pcb bits and such. 667 */ 668 thread_exit(); 669 } 670 671 #ifndef _SYS_SYSPROTO_H_ 672 struct abort2_args { 673 char *why; 674 int nargs; 675 void **args; 676 }; 677 #endif 678 679 int 680 sys_abort2(struct thread *td, struct abort2_args *uap) 681 { 682 struct proc *p = td->td_proc; 683 struct sbuf *sb; 684 void *uargs[16]; 685 int error, i, sig; 686 687 /* 688 * Do it right now so we can log either proper call of abort2(), or 689 * note, that invalid argument was passed. 512 is big enough to 690 * handle 16 arguments' descriptions with additional comments. 691 */ 692 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 693 sbuf_clear(sb); 694 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 695 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 696 /* 697 * Since we can't return from abort2(), send SIGKILL in cases, where 698 * abort2() was called improperly 699 */ 700 sig = SIGKILL; 701 /* Prevent from DoSes from user-space. */ 702 if (uap->nargs < 0 || uap->nargs > 16) 703 goto out; 704 if (uap->nargs > 0) { 705 if (uap->args == NULL) 706 goto out; 707 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 708 if (error != 0) 709 goto out; 710 } 711 /* 712 * Limit size of 'reason' string to 128. Will fit even when 713 * maximal number of arguments was chosen to be logged. 714 */ 715 if (uap->why != NULL) { 716 error = sbuf_copyin(sb, uap->why, 128); 717 if (error < 0) 718 goto out; 719 } else { 720 sbuf_printf(sb, "(null)"); 721 } 722 if (uap->nargs > 0) { 723 sbuf_printf(sb, "("); 724 for (i = 0;i < uap->nargs; i++) 725 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 726 sbuf_printf(sb, ")"); 727 } 728 /* 729 * Final stage: arguments were proper, string has been 730 * successfully copied from userspace, and copying pointers 731 * from user-space succeed. 732 */ 733 sig = SIGABRT; 734 out: 735 if (sig == SIGKILL) { 736 sbuf_trim(sb); 737 sbuf_printf(sb, " (Reason text inaccessible)"); 738 } 739 sbuf_cat(sb, "\n"); 740 sbuf_finish(sb); 741 log(LOG_INFO, "%s", sbuf_data(sb)); 742 sbuf_delete(sb); 743 exit1(td, 0, sig); 744 return (0); 745 } 746 747 #ifdef COMPAT_43 748 /* 749 * The dirty work is handled by kern_wait(). 750 */ 751 int 752 owait(struct thread *td, struct owait_args *uap __unused) 753 { 754 int error, status; 755 756 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 757 if (error == 0) 758 td->td_retval[1] = status; 759 return (error); 760 } 761 #endif /* COMPAT_43 */ 762 763 /* 764 * The dirty work is handled by kern_wait(). 765 */ 766 int 767 sys_wait4(struct thread *td, struct wait4_args *uap) 768 { 769 struct rusage ru, *rup; 770 int error, status; 771 772 if (uap->rusage != NULL) 773 rup = &ru; 774 else 775 rup = NULL; 776 error = kern_wait(td, uap->pid, &status, uap->options, rup); 777 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 778 error = copyout(&status, uap->status, sizeof(status)); 779 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 780 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 781 return (error); 782 } 783 784 int 785 sys_wait6(struct thread *td, struct wait6_args *uap) 786 { 787 struct __wrusage wru, *wrup; 788 siginfo_t si, *sip; 789 idtype_t idtype; 790 id_t id; 791 int error, status; 792 793 idtype = uap->idtype; 794 id = uap->id; 795 796 if (uap->wrusage != NULL) 797 wrup = &wru; 798 else 799 wrup = NULL; 800 801 if (uap->info != NULL) { 802 sip = &si; 803 bzero(sip, sizeof(*sip)); 804 } else 805 sip = NULL; 806 807 /* 808 * We expect all callers of wait6() to know about WEXITED and 809 * WTRAPPED. 810 */ 811 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 812 813 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 814 error = copyout(&status, uap->status, sizeof(status)); 815 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 816 error = copyout(&wru, uap->wrusage, sizeof(wru)); 817 if (uap->info != NULL && error == 0) 818 error = copyout(&si, uap->info, sizeof(si)); 819 return (error); 820 } 821 822 /* 823 * Reap the remains of a zombie process and optionally return status and 824 * rusage. Asserts and will release both the proctree_lock and the process 825 * lock as part of its work. 826 */ 827 void 828 proc_reap(struct thread *td, struct proc *p, int *status, int options) 829 { 830 struct proc *q, *t; 831 832 sx_assert(&proctree_lock, SA_XLOCKED); 833 PROC_LOCK_ASSERT(p, MA_OWNED); 834 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 835 836 mtx_spin_wait_unlocked(&p->p_slock); 837 838 q = td->td_proc; 839 840 if (status) 841 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 842 if (options & WNOWAIT) { 843 /* 844 * Only poll, returning the status. Caller does not wish to 845 * release the proc struct just yet. 846 */ 847 PROC_UNLOCK(p); 848 sx_xunlock(&proctree_lock); 849 return; 850 } 851 852 PROC_LOCK(q); 853 sigqueue_take(p->p_ksi); 854 PROC_UNLOCK(q); 855 856 /* 857 * If we got the child via a ptrace 'attach', we need to give it back 858 * to the old parent. 859 */ 860 if (p->p_oppid != p->p_pptr->p_pid) { 861 PROC_UNLOCK(p); 862 t = proc_realparent(p); 863 PROC_LOCK(t); 864 PROC_LOCK(p); 865 CTR2(KTR_PTRACE, 866 "wait: traced child %d moved back to parent %d", p->p_pid, 867 t->p_pid); 868 proc_reparent(p, t, false); 869 PROC_UNLOCK(p); 870 pksignal(t, SIGCHLD, p->p_ksi); 871 wakeup(t); 872 cv_broadcast(&p->p_pwait); 873 PROC_UNLOCK(t); 874 sx_xunlock(&proctree_lock); 875 return; 876 } 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(PIDHASHLOCK(p->p_pid)); 884 LIST_REMOVE(p, p_hash); 885 sx_xunlock(PIDHASHLOCK(p->p_pid)); 886 LIST_REMOVE(p, p_sibling); 887 reaper_abandon_children(p, true); 888 reaper_clear(p); 889 PROC_LOCK(p); 890 proc_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_id_clear(PROC_ID_PID, p->p_pid); 898 899 PROC_LOCK(p); 900 knlist_detach(p->p_klist); 901 p->p_klist = NULL; 902 PROC_UNLOCK(p); 903 904 /* 905 * Removal from allproc list and process group list paired with 906 * PROC_LOCK which was executed during that time should guarantee 907 * nothing can reach this process anymore. As such further locking 908 * is unnecessary. 909 */ 910 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 911 912 PROC_LOCK(q); 913 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 914 PROC_UNLOCK(q); 915 916 /* 917 * Decrement the count of procs running with this uid. 918 */ 919 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 920 921 /* 922 * Destroy resource accounting information associated with the process. 923 */ 924 #ifdef RACCT 925 if (racct_enable) { 926 PROC_LOCK(p); 927 racct_sub(p, RACCT_NPROC, 1); 928 PROC_UNLOCK(p); 929 } 930 #endif 931 racct_proc_exit(p); 932 933 /* 934 * Free credentials, arguments, and sigacts. 935 */ 936 proc_unset_cred(p); 937 pargs_drop(p->p_args); 938 p->p_args = NULL; 939 sigacts_free(p->p_sigacts); 940 p->p_sigacts = NULL; 941 942 /* 943 * Do any thread-system specific cleanups. 944 */ 945 thread_wait(p); 946 947 /* 948 * Give vm and machine-dependent layer a chance to free anything that 949 * cpu_exit couldn't release while still running in process context. 950 */ 951 vm_waitproc(p); 952 #ifdef MAC 953 mac_proc_destroy(p); 954 #endif 955 956 KASSERT(FIRST_THREAD_IN_PROC(p), 957 ("proc_reap: no residual thread!")); 958 uma_zfree(proc_zone, p); 959 atomic_add_int(&nprocs, -1); 960 } 961 962 static int 963 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 964 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 965 int check_only) 966 { 967 struct rusage *rup; 968 969 sx_assert(&proctree_lock, SA_XLOCKED); 970 971 PROC_LOCK(p); 972 973 switch (idtype) { 974 case P_ALL: 975 if (p->p_procdesc == NULL || 976 (p->p_pptr == td->td_proc && 977 (p->p_flag & P_TRACED) != 0)) { 978 break; 979 } 980 981 PROC_UNLOCK(p); 982 return (0); 983 case P_PID: 984 if (p->p_pid != (pid_t)id) { 985 PROC_UNLOCK(p); 986 return (0); 987 } 988 break; 989 case P_PGID: 990 if (p->p_pgid != (pid_t)id) { 991 PROC_UNLOCK(p); 992 return (0); 993 } 994 break; 995 case P_SID: 996 if (p->p_session->s_sid != (pid_t)id) { 997 PROC_UNLOCK(p); 998 return (0); 999 } 1000 break; 1001 case P_UID: 1002 if (p->p_ucred->cr_uid != (uid_t)id) { 1003 PROC_UNLOCK(p); 1004 return (0); 1005 } 1006 break; 1007 case P_GID: 1008 if (p->p_ucred->cr_gid != (gid_t)id) { 1009 PROC_UNLOCK(p); 1010 return (0); 1011 } 1012 break; 1013 case P_JAILID: 1014 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1015 PROC_UNLOCK(p); 1016 return (0); 1017 } 1018 break; 1019 /* 1020 * It seems that the thread structures get zeroed out 1021 * at process exit. This makes it impossible to 1022 * support P_SETID, P_CID or P_CPUID. 1023 */ 1024 default: 1025 PROC_UNLOCK(p); 1026 return (0); 1027 } 1028 1029 if (p_canwait(td, p)) { 1030 PROC_UNLOCK(p); 1031 return (0); 1032 } 1033 1034 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1035 PROC_UNLOCK(p); 1036 return (0); 1037 } 1038 1039 /* 1040 * This special case handles a kthread spawned by linux_clone 1041 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1042 * functions need to be able to distinguish between waiting 1043 * on a process and waiting on a thread. It is a thread if 1044 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1045 * signifies we want to wait for threads and not processes. 1046 */ 1047 if ((p->p_sigparent != SIGCHLD) ^ 1048 ((options & WLINUXCLONE) != 0)) { 1049 PROC_UNLOCK(p); 1050 return (0); 1051 } 1052 1053 if (siginfo != NULL) { 1054 bzero(siginfo, sizeof(*siginfo)); 1055 siginfo->si_errno = 0; 1056 1057 /* 1058 * SUSv4 requires that the si_signo value is always 1059 * SIGCHLD. Obey it despite the rfork(2) interface 1060 * allows to request other signal for child exit 1061 * notification. 1062 */ 1063 siginfo->si_signo = SIGCHLD; 1064 1065 /* 1066 * This is still a rough estimate. We will fix the 1067 * cases TRAPPED, STOPPED, and CONTINUED later. 1068 */ 1069 if (WCOREDUMP(p->p_xsig)) { 1070 siginfo->si_code = CLD_DUMPED; 1071 siginfo->si_status = WTERMSIG(p->p_xsig); 1072 } else if (WIFSIGNALED(p->p_xsig)) { 1073 siginfo->si_code = CLD_KILLED; 1074 siginfo->si_status = WTERMSIG(p->p_xsig); 1075 } else { 1076 siginfo->si_code = CLD_EXITED; 1077 siginfo->si_status = p->p_xexit; 1078 } 1079 1080 siginfo->si_pid = p->p_pid; 1081 siginfo->si_uid = p->p_ucred->cr_uid; 1082 1083 /* 1084 * The si_addr field would be useful additional 1085 * detail, but apparently the PC value may be lost 1086 * when we reach this point. bzero() above sets 1087 * siginfo->si_addr to NULL. 1088 */ 1089 } 1090 1091 /* 1092 * There should be no reason to limit resources usage info to 1093 * exited processes only. A snapshot about any resources used 1094 * by a stopped process may be exactly what is needed. 1095 */ 1096 if (wrusage != NULL) { 1097 rup = &wrusage->wru_self; 1098 *rup = p->p_ru; 1099 PROC_STATLOCK(p); 1100 calcru(p, &rup->ru_utime, &rup->ru_stime); 1101 PROC_STATUNLOCK(p); 1102 1103 rup = &wrusage->wru_children; 1104 *rup = p->p_stats->p_cru; 1105 calccru(p, &rup->ru_utime, &rup->ru_stime); 1106 } 1107 1108 if (p->p_state == PRS_ZOMBIE && !check_only) { 1109 proc_reap(td, p, status, options); 1110 return (-1); 1111 } 1112 return (1); 1113 } 1114 1115 int 1116 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1117 struct rusage *rusage) 1118 { 1119 struct __wrusage wru, *wrup; 1120 idtype_t idtype; 1121 id_t id; 1122 int ret; 1123 1124 /* 1125 * Translate the special pid values into the (idtype, pid) 1126 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1127 * kern_wait6() on its own. 1128 */ 1129 if (pid == WAIT_ANY) { 1130 idtype = P_ALL; 1131 id = 0; 1132 } else if (pid < 0) { 1133 idtype = P_PGID; 1134 id = (id_t)-pid; 1135 } else { 1136 idtype = P_PID; 1137 id = (id_t)pid; 1138 } 1139 1140 if (rusage != NULL) 1141 wrup = &wru; 1142 else 1143 wrup = NULL; 1144 1145 /* 1146 * For backward compatibility we implicitly add flags WEXITED 1147 * and WTRAPPED here. 1148 */ 1149 options |= WEXITED | WTRAPPED; 1150 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1151 if (rusage != NULL) 1152 *rusage = wru.wru_self; 1153 return (ret); 1154 } 1155 1156 static void 1157 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1158 int *status, int options, int si_code) 1159 { 1160 bool cont; 1161 1162 PROC_LOCK_ASSERT(p, MA_OWNED); 1163 sx_assert(&proctree_lock, SA_XLOCKED); 1164 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1165 si_code == CLD_CONTINUED); 1166 1167 cont = si_code == CLD_CONTINUED; 1168 if ((options & WNOWAIT) == 0) { 1169 if (cont) 1170 p->p_flag &= ~P_CONTINUED; 1171 else 1172 p->p_flag |= P_WAITED; 1173 PROC_LOCK(td->td_proc); 1174 sigqueue_take(p->p_ksi); 1175 PROC_UNLOCK(td->td_proc); 1176 } 1177 sx_xunlock(&proctree_lock); 1178 if (siginfo != NULL) { 1179 siginfo->si_code = si_code; 1180 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1181 } 1182 if (status != NULL) 1183 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1184 PROC_UNLOCK(p); 1185 td->td_retval[0] = p->p_pid; 1186 } 1187 1188 int 1189 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1190 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1191 { 1192 struct proc *p, *q; 1193 pid_t pid; 1194 int error, nfound, ret; 1195 bool report; 1196 1197 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1198 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1199 AUDIT_ARG_VALUE(options); 1200 1201 q = td->td_proc; 1202 1203 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1204 PROC_LOCK(q); 1205 id = (id_t)q->p_pgid; 1206 PROC_UNLOCK(q); 1207 idtype = P_PGID; 1208 } 1209 1210 /* If we don't know the option, just return. */ 1211 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1212 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1213 return (EINVAL); 1214 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1215 /* 1216 * We will be unable to find any matching processes, 1217 * because there are no known events to look for. 1218 * Prefer to return error instead of blocking 1219 * indefinitely. 1220 */ 1221 return (EINVAL); 1222 } 1223 1224 loop: 1225 if (q->p_flag & P_STATCHILD) { 1226 PROC_LOCK(q); 1227 q->p_flag &= ~P_STATCHILD; 1228 PROC_UNLOCK(q); 1229 } 1230 sx_xlock(&proctree_lock); 1231 loop_locked: 1232 nfound = 0; 1233 LIST_FOREACH(p, &q->p_children, p_sibling) { 1234 pid = p->p_pid; 1235 ret = proc_to_reap(td, p, idtype, id, status, options, 1236 wrusage, siginfo, 0); 1237 if (ret == 0) 1238 continue; 1239 else if (ret != 1) { 1240 td->td_retval[0] = pid; 1241 return (0); 1242 } 1243 1244 nfound++; 1245 PROC_LOCK_ASSERT(p, MA_OWNED); 1246 1247 if ((options & WTRAPPED) != 0 && 1248 (p->p_flag & P_TRACED) != 0) { 1249 PROC_SLOCK(p); 1250 report = 1251 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) && 1252 p->p_suspcount == p->p_numthreads && 1253 (p->p_flag & P_WAITED) == 0); 1254 PROC_SUNLOCK(p); 1255 if (report) { 1256 CTR4(KTR_PTRACE, 1257 "wait: returning trapped pid %d status %#x " 1258 "(xstat %d) xthread %d", 1259 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1260 p->p_xthread != NULL ? 1261 p->p_xthread->td_tid : -1); 1262 report_alive_proc(td, p, siginfo, status, 1263 options, CLD_TRAPPED); 1264 return (0); 1265 } 1266 } 1267 if ((options & WUNTRACED) != 0 && 1268 (p->p_flag & P_STOPPED_SIG) != 0) { 1269 PROC_SLOCK(p); 1270 report = (p->p_suspcount == p->p_numthreads && 1271 ((p->p_flag & P_WAITED) == 0)); 1272 PROC_SUNLOCK(p); 1273 if (report) { 1274 report_alive_proc(td, p, siginfo, status, 1275 options, CLD_STOPPED); 1276 return (0); 1277 } 1278 } 1279 if ((options & WCONTINUED) != 0 && 1280 (p->p_flag & P_CONTINUED) != 0) { 1281 report_alive_proc(td, p, siginfo, status, options, 1282 CLD_CONTINUED); 1283 return (0); 1284 } 1285 PROC_UNLOCK(p); 1286 } 1287 1288 /* 1289 * Look in the orphans list too, to allow the parent to 1290 * collect it's child exit status even if child is being 1291 * debugged. 1292 * 1293 * Debugger detaches from the parent upon successful 1294 * switch-over from parent to child. At this point due to 1295 * re-parenting the parent loses the child to debugger and a 1296 * wait4(2) call would report that it has no children to wait 1297 * for. By maintaining a list of orphans we allow the parent 1298 * to successfully wait until the child becomes a zombie. 1299 */ 1300 if (nfound == 0) { 1301 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1302 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1303 NULL, NULL, 1); 1304 if (ret != 0) { 1305 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1306 (int)td->td_retval[0])); 1307 PROC_UNLOCK(p); 1308 nfound++; 1309 break; 1310 } 1311 } 1312 } 1313 if (nfound == 0) { 1314 sx_xunlock(&proctree_lock); 1315 return (ECHILD); 1316 } 1317 if (options & WNOHANG) { 1318 sx_xunlock(&proctree_lock); 1319 td->td_retval[0] = 0; 1320 return (0); 1321 } 1322 PROC_LOCK(q); 1323 if (q->p_flag & P_STATCHILD) { 1324 q->p_flag &= ~P_STATCHILD; 1325 PROC_UNLOCK(q); 1326 goto loop_locked; 1327 } 1328 sx_xunlock(&proctree_lock); 1329 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0); 1330 if (error) 1331 return (error); 1332 goto loop; 1333 } 1334 1335 void 1336 proc_add_orphan(struct proc *child, struct proc *parent) 1337 { 1338 1339 sx_assert(&proctree_lock, SX_XLOCKED); 1340 KASSERT((child->p_flag & P_TRACED) != 0, 1341 ("proc_add_orphan: not traced")); 1342 1343 if (LIST_EMPTY(&parent->p_orphans)) { 1344 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1345 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan); 1346 } else { 1347 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans), 1348 child, p_orphan); 1349 } 1350 child->p_treeflag |= P_TREE_ORPHANED; 1351 } 1352 1353 /* 1354 * Make process 'parent' the new parent of process 'child'. 1355 * Must be called with an exclusive hold of proctree lock. 1356 */ 1357 void 1358 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid) 1359 { 1360 1361 sx_assert(&proctree_lock, SX_XLOCKED); 1362 PROC_LOCK_ASSERT(child, MA_OWNED); 1363 if (child->p_pptr == parent) 1364 return; 1365 1366 PROC_LOCK(child->p_pptr); 1367 sigqueue_take(child->p_ksi); 1368 PROC_UNLOCK(child->p_pptr); 1369 LIST_REMOVE(child, p_sibling); 1370 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1371 1372 proc_clear_orphan(child); 1373 if ((child->p_flag & P_TRACED) != 0) { 1374 proc_add_orphan(child, child->p_pptr); 1375 } 1376 1377 child->p_pptr = parent; 1378 if (set_oppid) 1379 child->p_oppid = parent->p_pid; 1380 } 1381