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 return (child->p_pptr->p_pid == child->p_oppid ? 112 child->p_pptr : initproc); 113 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 114 /* Cannot use LIST_PREV(), since the list head is not known. */ 115 p = __containerof(p->p_orphan.le_prev, struct proc, 116 p_orphan.le_next); 117 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 118 ("missing P_ORPHAN %p", p)); 119 } 120 parent = __containerof(p->p_orphan.le_prev, struct proc, 121 p_orphans.lh_first); 122 return (parent); 123 } 124 125 void 126 reaper_abandon_children(struct proc *p, bool exiting) 127 { 128 struct proc *p1, *p2, *ptmp; 129 130 sx_assert(&proctree_lock, SX_LOCKED); 131 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 132 if ((p->p_treeflag & P_TREE_REAPER) == 0) 133 return; 134 p1 = p->p_reaper; 135 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 136 LIST_REMOVE(p2, p_reapsibling); 137 p2->p_reaper = p1; 138 p2->p_reapsubtree = p->p_reapsubtree; 139 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 140 if (exiting && p2->p_pptr == p) { 141 PROC_LOCK(p2); 142 proc_reparent(p2, p1, true); 143 PROC_UNLOCK(p2); 144 } 145 } 146 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 147 p->p_treeflag &= ~P_TREE_REAPER; 148 } 149 150 static void 151 clear_orphan(struct proc *p) 152 { 153 struct proc *p1; 154 155 sx_assert(&proctree_lock, SA_XLOCKED); 156 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 157 return; 158 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 159 p1 = LIST_NEXT(p, p_orphan); 160 if (p1 != NULL) 161 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 162 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 163 } 164 LIST_REMOVE(p, p_orphan); 165 p->p_treeflag &= ~P_TREE_ORPHANED; 166 } 167 168 /* 169 * exit -- death of process. 170 */ 171 void 172 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 173 { 174 175 exit1(td, uap->rval, 0); 176 /* NOTREACHED */ 177 } 178 179 /* 180 * Exit: deallocate address space and other resources, change proc state to 181 * zombie, and unlink proc from allproc and parent's lists. Save exit status 182 * and rusage for wait(). Check for child processes and orphan them. 183 */ 184 void 185 exit1(struct thread *td, int rval, int signo) 186 { 187 struct proc *p, *nq, *q, *t; 188 struct thread *tdt; 189 ksiginfo_t *ksi, *ksi1; 190 int signal_parent; 191 192 mtx_assert(&Giant, MA_NOTOWNED); 193 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 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", signo, rval); 203 panic("Going nowhere without my init!"); 204 } 205 206 /* 207 * Deref SU mp, since the thread does not return to userspace. 208 */ 209 td_softdep_cleanup(td); 210 211 /* 212 * MUST abort all other threads before proceeding past here. 213 */ 214 PROC_LOCK(p); 215 /* 216 * First check if some other thread or external request got 217 * here before us. If so, act appropriately: exit or suspend. 218 * We must ensure that stop requests are handled before we set 219 * P_WEXIT. 220 */ 221 thread_suspend_check(0); 222 while (p->p_flag & P_HADTHREADS) { 223 /* 224 * Kill off the other threads. This requires 225 * some co-operation from other parts of the kernel 226 * so it may not be instantaneous. With this state set 227 * any thread entering the kernel from userspace will 228 * thread_exit() in trap(). Any thread attempting to 229 * sleep will return immediately with EINTR or EWOULDBLOCK 230 * which will hopefully force them to back out to userland 231 * freeing resources as they go. Any thread attempting 232 * to return to userland will thread_exit() from userret(). 233 * thread_exit() will unsuspend us when the last of the 234 * other threads exits. 235 * If there is already a thread singler after resumption, 236 * calling thread_single will fail; in that case, we just 237 * re-check all suspension request, the thread should 238 * either be suspended there or exit. 239 */ 240 if (!thread_single(p, SINGLE_EXIT)) 241 /* 242 * All other activity in this process is now 243 * stopped. Threading support has been turned 244 * off. 245 */ 246 break; 247 /* 248 * Recheck for new stop or suspend requests which 249 * might appear while process lock was dropped in 250 * thread_single(). 251 */ 252 thread_suspend_check(0); 253 } 254 KASSERT(p->p_numthreads == 1, 255 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 256 racct_sub(p, RACCT_NTHR, 1); 257 258 /* Let event handler change exit status */ 259 p->p_xexit = rval; 260 p->p_xsig = signo; 261 262 /* 263 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 264 * on our vmspace, so we should block below until they have 265 * released their reference to us. Note that if they have 266 * requested S_EXIT stops we will block here until they ack 267 * via PIOCCONT. 268 */ 269 _STOPEVENT(p, S_EXIT, 0); 270 271 /* 272 * Ignore any pending request to stop due to a stop signal. 273 * Once P_WEXIT is set, future requests will be ignored as 274 * well. 275 */ 276 p->p_flag &= ~P_STOPPED_SIG; 277 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 278 279 /* 280 * Note that we are exiting and do another wakeup of anyone in 281 * PIOCWAIT in case they aren't listening for S_EXIT stops or 282 * decided to wait again after we told them we are exiting. 283 */ 284 p->p_flag |= P_WEXIT; 285 wakeup(&p->p_stype); 286 287 /* 288 * Wait for any processes that have a hold on our vmspace to 289 * release their reference. 290 */ 291 while (p->p_lock > 0) 292 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 293 294 PROC_UNLOCK(p); 295 /* Drain the limit callout while we don't have the proc locked */ 296 callout_drain(&p->p_limco); 297 298 #ifdef AUDIT 299 /* 300 * The Sun BSM exit token contains two components: an exit status as 301 * passed to exit(), and a return value to indicate what sort of exit 302 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 303 * what the return value is. 304 */ 305 AUDIT_ARG_EXIT(rval, 0); 306 AUDIT_SYSCALL_EXIT(0, td); 307 #endif 308 309 /* Are we a task leader with peers? */ 310 if (p->p_peers != NULL && p == p->p_leader) { 311 mtx_lock(&ppeers_lock); 312 q = p->p_peers; 313 while (q != NULL) { 314 PROC_LOCK(q); 315 kern_psignal(q, SIGKILL); 316 PROC_UNLOCK(q); 317 q = q->p_peers; 318 } 319 while (p->p_peers != NULL) 320 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 321 mtx_unlock(&ppeers_lock); 322 } 323 324 /* 325 * Check if any loadable modules need anything done at process exit. 326 * E.g. SYSV IPC stuff. 327 * Event handler could change exit status. 328 * XXX what if one of these generates an error? 329 */ 330 EVENTHANDLER_DIRECT_INVOKE(process_exit, p); 331 332 /* 333 * If parent is waiting for us to exit or exec, 334 * P_PPWAIT is set; we will wakeup the parent below. 335 */ 336 PROC_LOCK(p); 337 stopprofclock(p); 338 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 339 p->p_ptevents = 0; 340 341 /* 342 * Stop the real interval timer. If the handler is currently 343 * executing, prevent it from rearming itself and let it finish. 344 */ 345 if (timevalisset(&p->p_realtimer.it_value) && 346 _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) { 347 timevalclear(&p->p_realtimer.it_interval); 348 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 349 KASSERT(!timevalisset(&p->p_realtimer.it_value), 350 ("realtime timer is still armed")); 351 } 352 353 PROC_UNLOCK(p); 354 355 umtx_thread_exit(td); 356 357 /* 358 * Reset any sigio structures pointing to us as a result of 359 * F_SETOWN with our pid. 360 */ 361 funsetownlst(&p->p_sigiolst); 362 363 /* 364 * If this process has an nlminfo data area (for lockd), release it 365 */ 366 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 367 (*nlminfo_release_p)(p); 368 369 /* 370 * Close open files and release open-file table. 371 * This may block! 372 */ 373 fdescfree(td); 374 375 /* 376 * If this thread tickled GEOM, we need to wait for the giggling to 377 * stop before we return to userland 378 */ 379 if (td->td_pflags & TDP_GEOM) 380 g_waitidle(); 381 382 /* 383 * Remove ourself from our leader's peer list and wake our leader. 384 */ 385 if (p->p_leader->p_peers != NULL) { 386 mtx_lock(&ppeers_lock); 387 if (p->p_leader->p_peers != NULL) { 388 q = p->p_leader; 389 while (q->p_peers != p) 390 q = q->p_peers; 391 q->p_peers = p->p_peers; 392 wakeup(p->p_leader); 393 } 394 mtx_unlock(&ppeers_lock); 395 } 396 397 vmspace_exit(td); 398 killjobc(); 399 (void)acct_process(td); 400 401 #ifdef KTRACE 402 ktrprocexit(td); 403 #endif 404 /* 405 * Release reference to text vnode 406 */ 407 if (p->p_textvp != NULL) { 408 vrele(p->p_textvp); 409 p->p_textvp = NULL; 410 } 411 412 /* 413 * Release our limits structure. 414 */ 415 lim_free(p->p_limit); 416 p->p_limit = NULL; 417 418 tidhash_remove(td); 419 420 /* 421 * Call machine-dependent code to release any 422 * machine-dependent resources other than the address space. 423 * The address space is released by "vmspace_exitfree(p)" in 424 * vm_waitproc(). 425 */ 426 cpu_exit(td); 427 428 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 429 430 sx_xlock(&proctree_lock); 431 /* 432 * Move proc from allproc queue to zombproc. 433 */ 434 sx_xlock(&allproc_lock); 435 sx_xlock(&zombproc_lock); 436 LIST_REMOVE(p, p_list); 437 LIST_INSERT_HEAD(&zombproc, p, p_list); 438 sx_xunlock(&zombproc_lock); 439 sx_xunlock(&allproc_lock); 440 441 /* 442 * Reparent all children processes: 443 * - traced ones to the original parent (or init if we are that parent) 444 * - the rest to init 445 */ 446 q = LIST_FIRST(&p->p_children); 447 if (q != NULL) /* only need this if any child is S_ZOMB */ 448 wakeup(q->p_reaper); 449 for (; q != NULL; q = nq) { 450 nq = LIST_NEXT(q, p_sibling); 451 ksi = ksiginfo_alloc(TRUE); 452 PROC_LOCK(q); 453 q->p_sigparent = SIGCHLD; 454 455 if (!(q->p_flag & P_TRACED)) { 456 proc_reparent(q, q->p_reaper, true); 457 if (q->p_state == PRS_ZOMBIE) { 458 /* 459 * Inform reaper about the reparented 460 * zombie, since wait(2) has something 461 * new to report. Guarantee queueing 462 * of the SIGCHLD signal, similar to 463 * the _exit() behaviour, by providing 464 * our ksiginfo. Ksi is freed by the 465 * signal delivery. 466 */ 467 if (q->p_ksi == NULL) { 468 ksi1 = NULL; 469 } else { 470 ksiginfo_copy(q->p_ksi, ksi); 471 ksi->ksi_flags |= KSI_INS; 472 ksi1 = ksi; 473 ksi = NULL; 474 } 475 PROC_LOCK(q->p_reaper); 476 pksignal(q->p_reaper, SIGCHLD, ksi1); 477 PROC_UNLOCK(q->p_reaper); 478 } else if (q->p_pdeathsig > 0) { 479 /* 480 * The child asked to received a signal 481 * when we exit. 482 */ 483 kern_psignal(q, q->p_pdeathsig); 484 } 485 } else { 486 /* 487 * Traced processes are killed since their existence 488 * means someone is screwing up. 489 */ 490 t = proc_realparent(q); 491 if (t == p) { 492 proc_reparent(q, q->p_reaper, true); 493 } else { 494 PROC_LOCK(t); 495 proc_reparent(q, t, true); 496 PROC_UNLOCK(t); 497 } 498 /* 499 * Since q was found on our children list, the 500 * proc_reparent() call moved q to the orphan 501 * list due to present P_TRACED flag. Clear 502 * orphan link for q now while q is locked. 503 */ 504 clear_orphan(q); 505 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 506 q->p_flag2 &= ~P2_PTRACE_FSTP; 507 q->p_ptevents = 0; 508 FOREACH_THREAD_IN_PROC(q, tdt) { 509 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 510 TDB_FSTP); 511 } 512 kern_psignal(q, SIGKILL); 513 } 514 PROC_UNLOCK(q); 515 if (ksi != NULL) 516 ksiginfo_free(ksi); 517 } 518 519 /* 520 * Also get rid of our orphans. 521 */ 522 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 523 PROC_LOCK(q); 524 /* 525 * If we are the real parent of this process 526 * but it has been reparented to a debugger, then 527 * check if it asked for a signal when we exit. 528 */ 529 if (q->p_pdeathsig > 0) 530 kern_psignal(q, q->p_pdeathsig); 531 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 532 q->p_pid); 533 clear_orphan(q); 534 PROC_UNLOCK(q); 535 } 536 537 /* Save exit status. */ 538 PROC_LOCK(p); 539 p->p_xthread = td; 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(signo)) 558 reason = CLD_DUMPED; 559 else if (WIFSIGNALED(signo)) 560 reason = CLD_KILLED; 561 SDT_PROBE1(proc, , , exit, reason); 562 #endif 563 564 /* 565 * If this is a process with a descriptor, we may not need to deliver 566 * a signal to the parent. proctree_lock is held over 567 * procdesc_exit() to serialize concurrent calls to close() and 568 * exit(). 569 */ 570 signal_parent = 0; 571 if (p->p_procdesc == NULL || procdesc_exit(p)) { 572 /* 573 * Notify parent that we're gone. If parent has the 574 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 575 * notify process 1 instead (and hope it will handle this 576 * situation). 577 */ 578 PROC_LOCK(p->p_pptr); 579 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 580 if (p->p_pptr->p_sigacts->ps_flag & 581 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 582 struct proc *pp; 583 584 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 585 pp = p->p_pptr; 586 PROC_UNLOCK(pp); 587 proc_reparent(p, p->p_reaper, true); 588 p->p_sigparent = SIGCHLD; 589 PROC_LOCK(p->p_pptr); 590 591 /* 592 * Notify parent, so in case he was wait(2)ing or 593 * executing waitpid(2) with our pid, he will 594 * continue. 595 */ 596 wakeup(pp); 597 } else 598 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 599 600 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) { 601 signal_parent = 1; 602 } else if (p->p_sigparent != 0) { 603 if (p->p_sigparent == SIGCHLD) { 604 signal_parent = 1; 605 } else { /* LINUX thread */ 606 signal_parent = 2; 607 } 608 } 609 } else 610 PROC_LOCK(p->p_pptr); 611 sx_xunlock(&proctree_lock); 612 613 if (signal_parent == 1) { 614 childproc_exited(p); 615 } else if (signal_parent == 2) { 616 kern_psignal(p->p_pptr, p->p_sigparent); 617 } 618 619 /* Tell the prison that we are gone. */ 620 prison_proc_free(p->p_ucred->cr_prison); 621 622 /* 623 * The state PRS_ZOMBIE prevents other proesses from sending 624 * signal to the process, to avoid memory leak, we free memory 625 * for signal queue at the time when the state is set. 626 */ 627 sigqueue_flush(&p->p_sigqueue); 628 sigqueue_flush(&td->td_sigqueue); 629 630 /* 631 * We have to wait until after acquiring all locks before 632 * changing p_state. We need to avoid all possible context 633 * switches (including ones from blocking on a mutex) while 634 * marked as a zombie. We also have to set the zombie state 635 * before we release the parent process' proc lock to avoid 636 * a lost wakeup. So, we first call wakeup, then we grab the 637 * sched lock, update the state, and release the parent process' 638 * proc lock. 639 */ 640 wakeup(p->p_pptr); 641 cv_broadcast(&p->p_pwait); 642 sched_exit(p->p_pptr, td); 643 PROC_SLOCK(p); 644 p->p_state = PRS_ZOMBIE; 645 PROC_UNLOCK(p->p_pptr); 646 647 /* 648 * Save our children's rusage information in our exit rusage. 649 */ 650 PROC_STATLOCK(p); 651 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 652 PROC_STATUNLOCK(p); 653 654 /* 655 * Make sure the scheduler takes this thread out of its tables etc. 656 * This will also release this thread's reference to the ucred. 657 * Other thread parts to release include pcb bits and such. 658 */ 659 thread_exit(); 660 } 661 662 663 #ifndef _SYS_SYSPROTO_H_ 664 struct abort2_args { 665 char *why; 666 int nargs; 667 void **args; 668 }; 669 #endif 670 671 int 672 sys_abort2(struct thread *td, struct abort2_args *uap) 673 { 674 struct proc *p = td->td_proc; 675 struct sbuf *sb; 676 void *uargs[16]; 677 int error, i, sig; 678 679 /* 680 * Do it right now so we can log either proper call of abort2(), or 681 * note, that invalid argument was passed. 512 is big enough to 682 * handle 16 arguments' descriptions with additional comments. 683 */ 684 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 685 sbuf_clear(sb); 686 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 687 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 688 /* 689 * Since we can't return from abort2(), send SIGKILL in cases, where 690 * abort2() was called improperly 691 */ 692 sig = SIGKILL; 693 /* Prevent from DoSes from user-space. */ 694 if (uap->nargs < 0 || uap->nargs > 16) 695 goto out; 696 if (uap->nargs > 0) { 697 if (uap->args == NULL) 698 goto out; 699 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 700 if (error != 0) 701 goto out; 702 } 703 /* 704 * Limit size of 'reason' string to 128. Will fit even when 705 * maximal number of arguments was chosen to be logged. 706 */ 707 if (uap->why != NULL) { 708 error = sbuf_copyin(sb, uap->why, 128); 709 if (error < 0) 710 goto out; 711 } else { 712 sbuf_printf(sb, "(null)"); 713 } 714 if (uap->nargs > 0) { 715 sbuf_printf(sb, "("); 716 for (i = 0;i < uap->nargs; i++) 717 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 718 sbuf_printf(sb, ")"); 719 } 720 /* 721 * Final stage: arguments were proper, string has been 722 * successfully copied from userspace, and copying pointers 723 * from user-space succeed. 724 */ 725 sig = SIGABRT; 726 out: 727 if (sig == SIGKILL) { 728 sbuf_trim(sb); 729 sbuf_printf(sb, " (Reason text inaccessible)"); 730 } 731 sbuf_cat(sb, "\n"); 732 sbuf_finish(sb); 733 log(LOG_INFO, "%s", sbuf_data(sb)); 734 sbuf_delete(sb); 735 exit1(td, 0, sig); 736 return (0); 737 } 738 739 740 #ifdef COMPAT_43 741 /* 742 * The dirty work is handled by kern_wait(). 743 */ 744 int 745 owait(struct thread *td, struct owait_args *uap __unused) 746 { 747 int error, status; 748 749 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 750 if (error == 0) 751 td->td_retval[1] = status; 752 return (error); 753 } 754 #endif /* COMPAT_43 */ 755 756 /* 757 * The dirty work is handled by kern_wait(). 758 */ 759 int 760 sys_wait4(struct thread *td, struct wait4_args *uap) 761 { 762 struct rusage ru, *rup; 763 int error, status; 764 765 if (uap->rusage != NULL) 766 rup = &ru; 767 else 768 rup = NULL; 769 error = kern_wait(td, uap->pid, &status, uap->options, rup); 770 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 771 error = copyout(&status, uap->status, sizeof(status)); 772 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 773 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 774 return (error); 775 } 776 777 int 778 sys_wait6(struct thread *td, struct wait6_args *uap) 779 { 780 struct __wrusage wru, *wrup; 781 siginfo_t si, *sip; 782 idtype_t idtype; 783 id_t id; 784 int error, status; 785 786 idtype = uap->idtype; 787 id = uap->id; 788 789 if (uap->wrusage != NULL) 790 wrup = &wru; 791 else 792 wrup = NULL; 793 794 if (uap->info != NULL) { 795 sip = &si; 796 bzero(sip, sizeof(*sip)); 797 } else 798 sip = NULL; 799 800 /* 801 * We expect all callers of wait6() to know about WEXITED and 802 * WTRAPPED. 803 */ 804 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 805 806 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 807 error = copyout(&status, uap->status, sizeof(status)); 808 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 809 error = copyout(&wru, uap->wrusage, sizeof(wru)); 810 if (uap->info != NULL && error == 0) 811 error = copyout(&si, uap->info, sizeof(si)); 812 return (error); 813 } 814 815 /* 816 * Reap the remains of a zombie process and optionally return status and 817 * rusage. Asserts and will release both the proctree_lock and the process 818 * lock as part of its work. 819 */ 820 void 821 proc_reap(struct thread *td, struct proc *p, int *status, int options) 822 { 823 struct proc *q, *t; 824 825 sx_assert(&proctree_lock, SA_XLOCKED); 826 PROC_LOCK_ASSERT(p, MA_OWNED); 827 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 828 829 mtx_spin_wait_unlocked(&p->p_slock); 830 831 q = td->td_proc; 832 833 if (status) 834 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 835 if (options & WNOWAIT) { 836 /* 837 * Only poll, returning the status. Caller does not wish to 838 * release the proc struct just yet. 839 */ 840 PROC_UNLOCK(p); 841 sx_xunlock(&proctree_lock); 842 return; 843 } 844 845 PROC_LOCK(q); 846 sigqueue_take(p->p_ksi); 847 PROC_UNLOCK(q); 848 849 /* 850 * If we got the child via a ptrace 'attach', we need to give it back 851 * to the old parent. 852 */ 853 if (p->p_oppid != p->p_pptr->p_pid) { 854 PROC_UNLOCK(p); 855 t = proc_realparent(p); 856 PROC_LOCK(t); 857 PROC_LOCK(p); 858 CTR2(KTR_PTRACE, 859 "wait: traced child %d moved back to parent %d", p->p_pid, 860 t->p_pid); 861 proc_reparent(p, t, false); 862 PROC_UNLOCK(p); 863 pksignal(t, SIGCHLD, p->p_ksi); 864 wakeup(t); 865 cv_broadcast(&p->p_pwait); 866 PROC_UNLOCK(t); 867 sx_xunlock(&proctree_lock); 868 return; 869 } 870 PROC_UNLOCK(p); 871 872 /* 873 * Remove other references to this process to ensure we have an 874 * exclusive reference. 875 */ 876 sx_xlock(&zombproc_lock); 877 LIST_REMOVE(p, p_list); /* off zombproc */ 878 sx_xunlock(&zombproc_lock); 879 sx_xlock(PIDHASHLOCK(p->p_pid)); 880 LIST_REMOVE(p, p_hash); 881 sx_xunlock(PIDHASHLOCK(p->p_pid)); 882 LIST_REMOVE(p, p_sibling); 883 reaper_abandon_children(p, true); 884 LIST_REMOVE(p, p_reapsibling); 885 PROC_LOCK(p); 886 clear_orphan(p); 887 PROC_UNLOCK(p); 888 leavepgrp(p); 889 if (p->p_procdesc != NULL) 890 procdesc_reap(p); 891 sx_xunlock(&proctree_lock); 892 893 PROC_LOCK(p); 894 knlist_detach(p->p_klist); 895 p->p_klist = NULL; 896 PROC_UNLOCK(p); 897 898 /* 899 * Removal from allproc list and process group list paired with 900 * PROC_LOCK which was executed during that time should guarantee 901 * nothing can reach this process anymore. As such further locking 902 * is unnecessary. 903 */ 904 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 905 906 PROC_LOCK(q); 907 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 908 PROC_UNLOCK(q); 909 910 /* 911 * Decrement the count of procs running with this uid. 912 */ 913 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 914 915 /* 916 * Destroy resource accounting information associated with the process. 917 */ 918 #ifdef RACCT 919 if (racct_enable) { 920 PROC_LOCK(p); 921 racct_sub(p, RACCT_NPROC, 1); 922 PROC_UNLOCK(p); 923 } 924 #endif 925 racct_proc_exit(p); 926 927 /* 928 * Free credentials, arguments, and sigacts. 929 */ 930 crfree(p->p_ucred); 931 proc_set_cred(p, NULL); 932 pargs_drop(p->p_args); 933 p->p_args = NULL; 934 sigacts_free(p->p_sigacts); 935 p->p_sigacts = NULL; 936 937 /* 938 * Do any thread-system specific cleanups. 939 */ 940 thread_wait(p); 941 942 /* 943 * Give vm and machine-dependent layer a chance to free anything that 944 * cpu_exit couldn't release while still running in process context. 945 */ 946 vm_waitproc(p); 947 #ifdef MAC 948 mac_proc_destroy(p); 949 #endif 950 951 KASSERT(FIRST_THREAD_IN_PROC(p), 952 ("proc_reap: no residual thread!")); 953 uma_zfree(proc_zone, p); 954 atomic_add_int(&nprocs, -1); 955 } 956 957 static int 958 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 959 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 960 int check_only) 961 { 962 struct rusage *rup; 963 964 sx_assert(&proctree_lock, SA_XLOCKED); 965 966 PROC_LOCK(p); 967 968 switch (idtype) { 969 case P_ALL: 970 if (p->p_procdesc != NULL) { 971 PROC_UNLOCK(p); 972 return (0); 973 } 974 break; 975 case P_PID: 976 if (p->p_pid != (pid_t)id) { 977 PROC_UNLOCK(p); 978 return (0); 979 } 980 break; 981 case P_PGID: 982 if (p->p_pgid != (pid_t)id) { 983 PROC_UNLOCK(p); 984 return (0); 985 } 986 break; 987 case P_SID: 988 if (p->p_session->s_sid != (pid_t)id) { 989 PROC_UNLOCK(p); 990 return (0); 991 } 992 break; 993 case P_UID: 994 if (p->p_ucred->cr_uid != (uid_t)id) { 995 PROC_UNLOCK(p); 996 return (0); 997 } 998 break; 999 case P_GID: 1000 if (p->p_ucred->cr_gid != (gid_t)id) { 1001 PROC_UNLOCK(p); 1002 return (0); 1003 } 1004 break; 1005 case P_JAILID: 1006 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1007 PROC_UNLOCK(p); 1008 return (0); 1009 } 1010 break; 1011 /* 1012 * It seems that the thread structures get zeroed out 1013 * at process exit. This makes it impossible to 1014 * support P_SETID, P_CID or P_CPUID. 1015 */ 1016 default: 1017 PROC_UNLOCK(p); 1018 return (0); 1019 } 1020 1021 if (p_canwait(td, p)) { 1022 PROC_UNLOCK(p); 1023 return (0); 1024 } 1025 1026 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1027 PROC_UNLOCK(p); 1028 return (0); 1029 } 1030 1031 /* 1032 * This special case handles a kthread spawned by linux_clone 1033 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1034 * functions need to be able to distinguish between waiting 1035 * on a process and waiting on a thread. It is a thread if 1036 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1037 * signifies we want to wait for threads and not processes. 1038 */ 1039 if ((p->p_sigparent != SIGCHLD) ^ 1040 ((options & WLINUXCLONE) != 0)) { 1041 PROC_UNLOCK(p); 1042 return (0); 1043 } 1044 1045 if (siginfo != NULL) { 1046 bzero(siginfo, sizeof(*siginfo)); 1047 siginfo->si_errno = 0; 1048 1049 /* 1050 * SUSv4 requires that the si_signo value is always 1051 * SIGCHLD. Obey it despite the rfork(2) interface 1052 * allows to request other signal for child exit 1053 * notification. 1054 */ 1055 siginfo->si_signo = SIGCHLD; 1056 1057 /* 1058 * This is still a rough estimate. We will fix the 1059 * cases TRAPPED, STOPPED, and CONTINUED later. 1060 */ 1061 if (WCOREDUMP(p->p_xsig)) { 1062 siginfo->si_code = CLD_DUMPED; 1063 siginfo->si_status = WTERMSIG(p->p_xsig); 1064 } else if (WIFSIGNALED(p->p_xsig)) { 1065 siginfo->si_code = CLD_KILLED; 1066 siginfo->si_status = WTERMSIG(p->p_xsig); 1067 } else { 1068 siginfo->si_code = CLD_EXITED; 1069 siginfo->si_status = p->p_xexit; 1070 } 1071 1072 siginfo->si_pid = p->p_pid; 1073 siginfo->si_uid = p->p_ucred->cr_uid; 1074 1075 /* 1076 * The si_addr field would be useful additional 1077 * detail, but apparently the PC value may be lost 1078 * when we reach this point. bzero() above sets 1079 * siginfo->si_addr to NULL. 1080 */ 1081 } 1082 1083 /* 1084 * There should be no reason to limit resources usage info to 1085 * exited processes only. A snapshot about any resources used 1086 * by a stopped process may be exactly what is needed. 1087 */ 1088 if (wrusage != NULL) { 1089 rup = &wrusage->wru_self; 1090 *rup = p->p_ru; 1091 PROC_STATLOCK(p); 1092 calcru(p, &rup->ru_utime, &rup->ru_stime); 1093 PROC_STATUNLOCK(p); 1094 1095 rup = &wrusage->wru_children; 1096 *rup = p->p_stats->p_cru; 1097 calccru(p, &rup->ru_utime, &rup->ru_stime); 1098 } 1099 1100 if (p->p_state == PRS_ZOMBIE && !check_only) { 1101 proc_reap(td, p, status, options); 1102 return (-1); 1103 } 1104 return (1); 1105 } 1106 1107 int 1108 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1109 struct rusage *rusage) 1110 { 1111 struct __wrusage wru, *wrup; 1112 idtype_t idtype; 1113 id_t id; 1114 int ret; 1115 1116 /* 1117 * Translate the special pid values into the (idtype, pid) 1118 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1119 * kern_wait6() on its own. 1120 */ 1121 if (pid == WAIT_ANY) { 1122 idtype = P_ALL; 1123 id = 0; 1124 } else if (pid < 0) { 1125 idtype = P_PGID; 1126 id = (id_t)-pid; 1127 } else { 1128 idtype = P_PID; 1129 id = (id_t)pid; 1130 } 1131 1132 if (rusage != NULL) 1133 wrup = &wru; 1134 else 1135 wrup = NULL; 1136 1137 /* 1138 * For backward compatibility we implicitly add flags WEXITED 1139 * and WTRAPPED here. 1140 */ 1141 options |= WEXITED | WTRAPPED; 1142 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1143 if (rusage != NULL) 1144 *rusage = wru.wru_self; 1145 return (ret); 1146 } 1147 1148 static void 1149 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1150 int *status, int options, int si_code) 1151 { 1152 bool cont; 1153 1154 PROC_LOCK_ASSERT(p, MA_OWNED); 1155 sx_assert(&proctree_lock, SA_XLOCKED); 1156 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1157 si_code == CLD_CONTINUED); 1158 1159 cont = si_code == CLD_CONTINUED; 1160 if ((options & WNOWAIT) == 0) { 1161 if (cont) 1162 p->p_flag &= ~P_CONTINUED; 1163 else 1164 p->p_flag |= P_WAITED; 1165 PROC_LOCK(td->td_proc); 1166 sigqueue_take(p->p_ksi); 1167 PROC_UNLOCK(td->td_proc); 1168 } 1169 sx_xunlock(&proctree_lock); 1170 if (siginfo != NULL) { 1171 siginfo->si_code = si_code; 1172 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1173 } 1174 if (status != NULL) 1175 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1176 PROC_UNLOCK(p); 1177 td->td_retval[0] = p->p_pid; 1178 } 1179 1180 int 1181 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1182 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1183 { 1184 struct proc *p, *q; 1185 pid_t pid; 1186 int error, nfound, ret; 1187 bool report; 1188 1189 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1190 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1191 AUDIT_ARG_VALUE(options); 1192 1193 q = td->td_proc; 1194 1195 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1196 PROC_LOCK(q); 1197 id = (id_t)q->p_pgid; 1198 PROC_UNLOCK(q); 1199 idtype = P_PGID; 1200 } 1201 1202 /* If we don't know the option, just return. */ 1203 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1204 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1205 return (EINVAL); 1206 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1207 /* 1208 * We will be unable to find any matching processes, 1209 * because there are no known events to look for. 1210 * Prefer to return error instead of blocking 1211 * indefinitely. 1212 */ 1213 return (EINVAL); 1214 } 1215 1216 loop: 1217 if (q->p_flag & P_STATCHILD) { 1218 PROC_LOCK(q); 1219 q->p_flag &= ~P_STATCHILD; 1220 PROC_UNLOCK(q); 1221 } 1222 sx_xlock(&proctree_lock); 1223 loop_locked: 1224 nfound = 0; 1225 LIST_FOREACH(p, &q->p_children, p_sibling) { 1226 pid = p->p_pid; 1227 ret = proc_to_reap(td, p, idtype, id, status, options, 1228 wrusage, siginfo, 0); 1229 if (ret == 0) 1230 continue; 1231 else if (ret != 1) { 1232 td->td_retval[0] = pid; 1233 return (0); 1234 } 1235 1236 nfound++; 1237 PROC_LOCK_ASSERT(p, MA_OWNED); 1238 1239 if ((options & WTRAPPED) != 0 && 1240 (p->p_flag & P_TRACED) != 0) { 1241 PROC_SLOCK(p); 1242 report = 1243 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) && 1244 p->p_suspcount == p->p_numthreads && 1245 (p->p_flag & P_WAITED) == 0); 1246 PROC_SUNLOCK(p); 1247 if (report) { 1248 CTR4(KTR_PTRACE, 1249 "wait: returning trapped pid %d status %#x " 1250 "(xstat %d) xthread %d", 1251 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1252 p->p_xthread != NULL ? 1253 p->p_xthread->td_tid : -1); 1254 report_alive_proc(td, p, siginfo, status, 1255 options, CLD_TRAPPED); 1256 return (0); 1257 } 1258 } 1259 if ((options & WUNTRACED) != 0 && 1260 (p->p_flag & P_STOPPED_SIG) != 0) { 1261 PROC_SLOCK(p); 1262 report = (p->p_suspcount == p->p_numthreads && 1263 ((p->p_flag & P_WAITED) == 0)); 1264 PROC_SUNLOCK(p); 1265 if (report) { 1266 report_alive_proc(td, p, siginfo, status, 1267 options, CLD_STOPPED); 1268 return (0); 1269 } 1270 } 1271 if ((options & WCONTINUED) != 0 && 1272 (p->p_flag & P_CONTINUED) != 0) { 1273 report_alive_proc(td, p, siginfo, status, options, 1274 CLD_CONTINUED); 1275 return (0); 1276 } 1277 PROC_UNLOCK(p); 1278 } 1279 1280 /* 1281 * Look in the orphans list too, to allow the parent to 1282 * collect it's child exit status even if child is being 1283 * debugged. 1284 * 1285 * Debugger detaches from the parent upon successful 1286 * switch-over from parent to child. At this point due to 1287 * re-parenting the parent loses the child to debugger and a 1288 * wait4(2) call would report that it has no children to wait 1289 * for. By maintaining a list of orphans we allow the parent 1290 * to successfully wait until the child becomes a zombie. 1291 */ 1292 if (nfound == 0) { 1293 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1294 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1295 NULL, NULL, 1); 1296 if (ret != 0) { 1297 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1298 (int)td->td_retval[0])); 1299 PROC_UNLOCK(p); 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 if (q->p_flag & P_STATCHILD) { 1316 q->p_flag &= ~P_STATCHILD; 1317 PROC_UNLOCK(q); 1318 goto loop_locked; 1319 } 1320 sx_xunlock(&proctree_lock); 1321 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0); 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, bool set_oppid) 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 if (set_oppid) 1361 child->p_oppid = parent->p_pid; 1362 } 1363