kern_exit.c revision 333162
1/*- 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: stable/11/sys/kern/kern_exit.c 333162 2018-05-02 07:57:36Z kib $"); 39 40#include "opt_compat.h" 41#include "opt_ktrace.h" 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/sysproto.h> 46#include <sys/capsicum.h> 47#include <sys/eventhandler.h> 48#include <sys/kernel.h> 49#include <sys/malloc.h> 50#include <sys/lock.h> 51#include <sys/mutex.h> 52#include <sys/proc.h> 53#include <sys/procdesc.h> 54#include <sys/pioctl.h> 55#include <sys/jail.h> 56#include <sys/tty.h> 57#include <sys/wait.h> 58#include <sys/vmmeter.h> 59#include <sys/vnode.h> 60#include <sys/racct.h> 61#include <sys/resourcevar.h> 62#include <sys/sbuf.h> 63#include <sys/signalvar.h> 64#include <sys/sched.h> 65#include <sys/sx.h> 66#include <sys/syscallsubr.h> 67#include <sys/syslog.h> 68#include <sys/ptrace.h> 69#include <sys/acct.h> /* for acct_process() function prototype */ 70#include <sys/filedesc.h> 71#include <sys/sdt.h> 72#include <sys/shm.h> 73#include <sys/sem.h> 74#include <sys/umtx.h> 75#ifdef KTRACE 76#include <sys/ktrace.h> 77#endif 78 79#include <security/audit/audit.h> 80#include <security/mac/mac_framework.h> 81 82#include <vm/vm.h> 83#include <vm/vm_extern.h> 84#include <vm/vm_param.h> 85#include <vm/pmap.h> 86#include <vm/vm_map.h> 87#include <vm/vm_page.h> 88#include <vm/uma.h> 89#include <vm/vm_domain.h> 90 91#ifdef KDTRACE_HOOKS 92#include <sys/dtrace_bsd.h> 93dtrace_execexit_func_t dtrace_fasttrap_exit; 94#endif 95 96SDT_PROVIDER_DECLARE(proc); 97SDT_PROBE_DEFINE1(proc, , , exit, "int"); 98 99/* Hook for NFS teardown procedure. */ 100void (*nlminfo_release_p)(struct proc *p); 101 102EVENTHANDLER_LIST_DECLARE(process_exit); 103 104struct proc * 105proc_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 130void 131reaper_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 155static void 156clear_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 */ 176void 177sys_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 */ 189void 190exit1(struct thread *td, int rval, int signo) 191{ 192 struct proc *p, *nq, *q, *t; 193 struct thread *tdt; 194 ksiginfo_t *ksi, *ksi1; 195 196 mtx_assert(&Giant, MA_NOTOWNED); 197 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 198 199 p = td->td_proc; 200 /* 201 * XXX in case we're rebooting we just let init die in order to 202 * work around an unsolved stack overflow seen very late during 203 * shutdown on sparc64 when the gmirror worker process exists. 204 */ 205 if (p == initproc && rebooting == 0) { 206 printf("init died (signal %d, exit %d)\n", signo, rval); 207 panic("Going nowhere without my init!"); 208 } 209 210 /* 211 * Deref SU mp, since the thread does not return to userspace. 212 */ 213 td_softdep_cleanup(td); 214 215 /* 216 * MUST abort all other threads before proceeding past here. 217 */ 218 PROC_LOCK(p); 219 /* 220 * First check if some other thread or external request got 221 * here before us. If so, act appropriately: exit or suspend. 222 * We must ensure that stop requests are handled before we set 223 * P_WEXIT. 224 */ 225 thread_suspend_check(0); 226 while (p->p_flag & P_HADTHREADS) { 227 /* 228 * Kill off the other threads. This requires 229 * some co-operation from other parts of the kernel 230 * so it may not be instantaneous. With this state set 231 * any thread entering the kernel from userspace will 232 * thread_exit() in trap(). Any thread attempting to 233 * sleep will return immediately with EINTR or EWOULDBLOCK 234 * which will hopefully force them to back out to userland 235 * freeing resources as they go. Any thread attempting 236 * to return to userland will thread_exit() from userret(). 237 * thread_exit() will unsuspend us when the last of the 238 * other threads exits. 239 * If there is already a thread singler after resumption, 240 * calling thread_single will fail; in that case, we just 241 * re-check all suspension request, the thread should 242 * either be suspended there or exit. 243 */ 244 if (!thread_single(p, SINGLE_EXIT)) 245 /* 246 * All other activity in this process is now 247 * stopped. Threading support has been turned 248 * off. 249 */ 250 break; 251 /* 252 * Recheck for new stop or suspend requests which 253 * might appear while process lock was dropped in 254 * thread_single(). 255 */ 256 thread_suspend_check(0); 257 } 258 KASSERT(p->p_numthreads == 1, 259 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 260 racct_sub(p, RACCT_NTHR, 1); 261 262 /* Let event handler change exit status */ 263 p->p_xexit = rval; 264 p->p_xsig = signo; 265 266 /* 267 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 268 * on our vmspace, so we should block below until they have 269 * released their reference to us. Note that if they have 270 * requested S_EXIT stops we will block here until they ack 271 * via PIOCCONT. 272 */ 273 _STOPEVENT(p, S_EXIT, 0); 274 275 /* 276 * Ignore any pending request to stop due to a stop signal. 277 * Once P_WEXIT is set, future requests will be ignored as 278 * well. 279 */ 280 p->p_flag &= ~P_STOPPED_SIG; 281 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 282 283 /* 284 * Note that we are exiting and do another wakeup of anyone in 285 * PIOCWAIT in case they aren't listening for S_EXIT stops or 286 * decided to wait again after we told them we are exiting. 287 */ 288 p->p_flag |= P_WEXIT; 289 wakeup(&p->p_stype); 290 291 /* 292 * Wait for any processes that have a hold on our vmspace to 293 * release their reference. 294 */ 295 while (p->p_lock > 0) 296 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 297 298 PROC_UNLOCK(p); 299 /* Drain the limit callout while we don't have the proc locked */ 300 callout_drain(&p->p_limco); 301 302#ifdef AUDIT 303 /* 304 * The Sun BSM exit token contains two components: an exit status as 305 * passed to exit(), and a return value to indicate what sort of exit 306 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 307 * what the return value is. 308 */ 309 AUDIT_ARG_EXIT(rval, 0); 310 AUDIT_SYSCALL_EXIT(0, td); 311#endif 312 313 /* Are we a task leader with peers? */ 314 if (p->p_peers != NULL && p == p->p_leader) { 315 mtx_lock(&ppeers_lock); 316 q = p->p_peers; 317 while (q != NULL) { 318 PROC_LOCK(q); 319 kern_psignal(q, SIGKILL); 320 PROC_UNLOCK(q); 321 q = q->p_peers; 322 } 323 while (p->p_peers != NULL) 324 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 325 mtx_unlock(&ppeers_lock); 326 } 327 328 /* 329 * Check if any loadable modules need anything done at process exit. 330 * E.g. SYSV IPC stuff. 331 * Event handler could change exit status. 332 * XXX what if one of these generates an error? 333 */ 334 EVENTHANDLER_DIRECT_INVOKE(process_exit, p); 335 336 /* 337 * If parent is waiting for us to exit or exec, 338 * P_PPWAIT is set; we will wakeup the parent below. 339 */ 340 PROC_LOCK(p); 341 stopprofclock(p); 342 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 343 p->p_ptevents = 0; 344 345 /* 346 * Stop the real interval timer. If the handler is currently 347 * executing, prevent it from rearming itself and let it finish. 348 */ 349 if (timevalisset(&p->p_realtimer.it_value) && 350 _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) { 351 timevalclear(&p->p_realtimer.it_interval); 352 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 353 KASSERT(!timevalisset(&p->p_realtimer.it_value), 354 ("realtime timer is still armed")); 355 } 356 357 PROC_UNLOCK(p); 358 359 umtx_thread_exit(td); 360 361 /* 362 * Reset any sigio structures pointing to us as a result of 363 * F_SETOWN with our pid. 364 */ 365 funsetownlst(&p->p_sigiolst); 366 367 /* 368 * If this process has an nlminfo data area (for lockd), release it 369 */ 370 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 371 (*nlminfo_release_p)(p); 372 373 /* 374 * Close open files and release open-file table. 375 * This may block! 376 */ 377 fdescfree(td); 378 379 /* 380 * If this thread tickled GEOM, we need to wait for the giggling to 381 * stop before we return to userland 382 */ 383 if (td->td_pflags & TDP_GEOM) 384 g_waitidle(); 385 386 /* 387 * Remove ourself from our leader's peer list and wake our leader. 388 */ 389 if (p->p_leader->p_peers != NULL) { 390 mtx_lock(&ppeers_lock); 391 if (p->p_leader->p_peers != NULL) { 392 q = p->p_leader; 393 while (q->p_peers != p) 394 q = q->p_peers; 395 q->p_peers = p->p_peers; 396 wakeup(p->p_leader); 397 } 398 mtx_unlock(&ppeers_lock); 399 } 400 401 vmspace_exit(td); 402 killjobc(); 403 (void)acct_process(td); 404 405#ifdef KTRACE 406 ktrprocexit(td); 407#endif 408 /* 409 * Release reference to text vnode 410 */ 411 if (p->p_textvp != NULL) { 412 vrele(p->p_textvp); 413 p->p_textvp = NULL; 414 } 415 416 /* 417 * Release our limits structure. 418 */ 419 lim_free(p->p_limit); 420 p->p_limit = NULL; 421 422 tidhash_remove(td); 423 424 /* 425 * Remove proc from allproc queue and pidhash chain. 426 * Place onto zombproc. Unlink from parent's child list. 427 */ 428 sx_xlock(&allproc_lock); 429 LIST_REMOVE(p, p_list); 430 LIST_INSERT_HEAD(&zombproc, p, p_list); 431 LIST_REMOVE(p, p_hash); 432 sx_xunlock(&allproc_lock); 433 434 /* 435 * Call machine-dependent code to release any 436 * machine-dependent resources other than the address space. 437 * The address space is released by "vmspace_exitfree(p)" in 438 * vm_waitproc(). 439 */ 440 cpu_exit(td); 441 442 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 443 444 /* 445 * Reparent all children processes: 446 * - traced ones to the original parent (or init if we are that parent) 447 * - the rest to init 448 */ 449 sx_xlock(&proctree_lock); 450 q = LIST_FIRST(&p->p_children); 451 if (q != NULL) /* only need this if any child is S_ZOMB */ 452 wakeup(q->p_reaper); 453 for (; q != NULL; q = nq) { 454 nq = LIST_NEXT(q, p_sibling); 455 ksi = ksiginfo_alloc(TRUE); 456 PROC_LOCK(q); 457 q->p_sigparent = SIGCHLD; 458 459 if (!(q->p_flag & P_TRACED)) { 460 proc_reparent(q, q->p_reaper); 461 if (q->p_state == PRS_ZOMBIE) { 462 /* 463 * Inform reaper about the reparented 464 * zombie, since wait(2) has something 465 * new to report. Guarantee queueing 466 * of the SIGCHLD signal, similar to 467 * the _exit() behaviour, by providing 468 * our ksiginfo. Ksi is freed by the 469 * signal delivery. 470 */ 471 if (q->p_ksi == NULL) { 472 ksi1 = NULL; 473 } else { 474 ksiginfo_copy(q->p_ksi, ksi); 475 ksi->ksi_flags |= KSI_INS; 476 ksi1 = ksi; 477 ksi = NULL; 478 } 479 PROC_LOCK(q->p_reaper); 480 pksignal(q->p_reaper, SIGCHLD, ksi1); 481 PROC_UNLOCK(q->p_reaper); 482 } else if (q->p_pdeathsig > 0) { 483 /* 484 * The child asked to received a signal 485 * when we exit. 486 */ 487 kern_psignal(q, q->p_pdeathsig); 488 } 489 } else { 490 /* 491 * Traced processes are killed since their existence 492 * means someone is screwing up. 493 */ 494 t = proc_realparent(q); 495 if (t == p) { 496 proc_reparent(q, q->p_reaper); 497 } else { 498 PROC_LOCK(t); 499 proc_reparent(q, t); 500 PROC_UNLOCK(t); 501 } 502 /* 503 * Since q was found on our children list, the 504 * proc_reparent() call moved q to the orphan 505 * list due to present P_TRACED flag. Clear 506 * orphan link for q now while q is locked. 507 */ 508 clear_orphan(q); 509 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 510 q->p_flag2 &= ~P2_PTRACE_FSTP; 511 q->p_ptevents = 0; 512 FOREACH_THREAD_IN_PROC(q, tdt) { 513 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 514 TDB_FSTP); 515 } 516 kern_psignal(q, SIGKILL); 517 } 518 PROC_UNLOCK(q); 519 if (ksi != NULL) 520 ksiginfo_free(ksi); 521 } 522 523 /* 524 * Also get rid of our orphans. 525 */ 526 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 527 PROC_LOCK(q); 528 /* 529 * If we are the real parent of this process 530 * but it has been reparented to a debugger, then 531 * check if it asked for a signal when we exit. 532 */ 533 if (q->p_pdeathsig > 0) 534 kern_psignal(q, q->p_pdeathsig); 535 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 536 q->p_pid); 537 clear_orphan(q); 538 PROC_UNLOCK(q); 539 } 540 541 /* Save exit status. */ 542 PROC_LOCK(p); 543 p->p_xthread = td; 544 545 /* Tell the prison that we are gone. */ 546 prison_proc_free(p->p_ucred->cr_prison); 547 548#ifdef KDTRACE_HOOKS 549 /* 550 * Tell the DTrace fasttrap provider about the exit if it 551 * has declared an interest. 552 */ 553 if (dtrace_fasttrap_exit) 554 dtrace_fasttrap_exit(p); 555#endif 556 557 /* 558 * Notify interested parties of our demise. 559 */ 560 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 561 562#ifdef KDTRACE_HOOKS 563 int reason = CLD_EXITED; 564 if (WCOREDUMP(signo)) 565 reason = CLD_DUMPED; 566 else if (WIFSIGNALED(signo)) 567 reason = CLD_KILLED; 568 SDT_PROBE1(proc, , , exit, reason); 569#endif 570 571 /* 572 * If this is a process with a descriptor, we may not need to deliver 573 * a signal to the parent. proctree_lock is held over 574 * procdesc_exit() to serialize concurrent calls to close() and 575 * exit(). 576 */ 577 if (p->p_procdesc == NULL || procdesc_exit(p)) { 578 /* 579 * Notify parent that we're gone. If parent has the 580 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 581 * notify process 1 instead (and hope it will handle this 582 * situation). 583 */ 584 PROC_LOCK(p->p_pptr); 585 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 586 if (p->p_pptr->p_sigacts->ps_flag & 587 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 588 struct proc *pp; 589 590 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 591 pp = p->p_pptr; 592 PROC_UNLOCK(pp); 593 proc_reparent(p, p->p_reaper); 594 p->p_sigparent = SIGCHLD; 595 PROC_LOCK(p->p_pptr); 596 597 /* 598 * Notify parent, so in case he was wait(2)ing or 599 * executing waitpid(2) with our pid, he will 600 * continue. 601 */ 602 wakeup(pp); 603 } else 604 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 605 606 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 607 childproc_exited(p); 608 else if (p->p_sigparent != 0) { 609 if (p->p_sigparent == SIGCHLD) 610 childproc_exited(p); 611 else /* LINUX thread */ 612 kern_psignal(p->p_pptr, p->p_sigparent); 613 } 614 } else 615 PROC_LOCK(p->p_pptr); 616 sx_xunlock(&proctree_lock); 617 618 /* 619 * The state PRS_ZOMBIE prevents other proesses from sending 620 * signal to the process, to avoid memory leak, we free memory 621 * for signal queue at the time when the state is set. 622 */ 623 sigqueue_flush(&p->p_sigqueue); 624 sigqueue_flush(&td->td_sigqueue); 625 626 /* 627 * We have to wait until after acquiring all locks before 628 * changing p_state. We need to avoid all possible context 629 * switches (including ones from blocking on a mutex) while 630 * marked as a zombie. We also have to set the zombie state 631 * before we release the parent process' proc lock to avoid 632 * a lost wakeup. So, we first call wakeup, then we grab the 633 * sched lock, update the state, and release the parent process' 634 * proc lock. 635 */ 636 wakeup(p->p_pptr); 637 cv_broadcast(&p->p_pwait); 638 sched_exit(p->p_pptr, td); 639 PROC_SLOCK(p); 640 p->p_state = PRS_ZOMBIE; 641 PROC_UNLOCK(p->p_pptr); 642 643 /* 644 * Save our children's rusage information in our exit rusage. 645 */ 646 PROC_STATLOCK(p); 647 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 648 PROC_STATUNLOCK(p); 649 650 /* 651 * Make sure the scheduler takes this thread out of its tables etc. 652 * This will also release this thread's reference to the ucred. 653 * Other thread parts to release include pcb bits and such. 654 */ 655 thread_exit(); 656} 657 658 659#ifndef _SYS_SYSPROTO_H_ 660struct abort2_args { 661 char *why; 662 int nargs; 663 void **args; 664}; 665#endif 666 667int 668sys_abort2(struct thread *td, struct abort2_args *uap) 669{ 670 struct proc *p = td->td_proc; 671 struct sbuf *sb; 672 void *uargs[16]; 673 int error, i, sig; 674 675 /* 676 * Do it right now so we can log either proper call of abort2(), or 677 * note, that invalid argument was passed. 512 is big enough to 678 * handle 16 arguments' descriptions with additional comments. 679 */ 680 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 681 sbuf_clear(sb); 682 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 683 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 684 /* 685 * Since we can't return from abort2(), send SIGKILL in cases, where 686 * abort2() was called improperly 687 */ 688 sig = SIGKILL; 689 /* Prevent from DoSes from user-space. */ 690 if (uap->nargs < 0 || uap->nargs > 16) 691 goto out; 692 if (uap->nargs > 0) { 693 if (uap->args == NULL) 694 goto out; 695 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 696 if (error != 0) 697 goto out; 698 } 699 /* 700 * Limit size of 'reason' string to 128. Will fit even when 701 * maximal number of arguments was chosen to be logged. 702 */ 703 if (uap->why != NULL) { 704 error = sbuf_copyin(sb, uap->why, 128); 705 if (error < 0) 706 goto out; 707 } else { 708 sbuf_printf(sb, "(null)"); 709 } 710 if (uap->nargs > 0) { 711 sbuf_printf(sb, "("); 712 for (i = 0;i < uap->nargs; i++) 713 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 714 sbuf_printf(sb, ")"); 715 } 716 /* 717 * Final stage: arguments were proper, string has been 718 * successfully copied from userspace, and copying pointers 719 * from user-space succeed. 720 */ 721 sig = SIGABRT; 722out: 723 if (sig == SIGKILL) { 724 sbuf_trim(sb); 725 sbuf_printf(sb, " (Reason text inaccessible)"); 726 } 727 sbuf_cat(sb, "\n"); 728 sbuf_finish(sb); 729 log(LOG_INFO, "%s", sbuf_data(sb)); 730 sbuf_delete(sb); 731 exit1(td, 0, sig); 732 return (0); 733} 734 735 736#ifdef COMPAT_43 737/* 738 * The dirty work is handled by kern_wait(). 739 */ 740int 741owait(struct thread *td, struct owait_args *uap __unused) 742{ 743 int error, status; 744 745 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 746 if (error == 0) 747 td->td_retval[1] = status; 748 return (error); 749} 750#endif /* COMPAT_43 */ 751 752/* 753 * The dirty work is handled by kern_wait(). 754 */ 755int 756sys_wait4(struct thread *td, struct wait4_args *uap) 757{ 758 struct rusage ru, *rup; 759 int error, status; 760 761 if (uap->rusage != NULL) 762 rup = &ru; 763 else 764 rup = NULL; 765 error = kern_wait(td, uap->pid, &status, uap->options, rup); 766 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 767 error = copyout(&status, uap->status, sizeof(status)); 768 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 769 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 770 return (error); 771} 772 773int 774sys_wait6(struct thread *td, struct wait6_args *uap) 775{ 776 struct __wrusage wru, *wrup; 777 siginfo_t si, *sip; 778 idtype_t idtype; 779 id_t id; 780 int error, status; 781 782 idtype = uap->idtype; 783 id = uap->id; 784 785 if (uap->wrusage != NULL) 786 wrup = &wru; 787 else 788 wrup = NULL; 789 790 if (uap->info != NULL) { 791 sip = &si; 792 bzero(sip, sizeof(*sip)); 793 } else 794 sip = NULL; 795 796 /* 797 * We expect all callers of wait6() to know about WEXITED and 798 * WTRAPPED. 799 */ 800 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 801 802 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 803 error = copyout(&status, uap->status, sizeof(status)); 804 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 805 error = copyout(&wru, uap->wrusage, sizeof(wru)); 806 if (uap->info != NULL && error == 0) 807 error = copyout(&si, uap->info, sizeof(si)); 808 return (error); 809} 810 811/* 812 * Reap the remains of a zombie process and optionally return status and 813 * rusage. Asserts and will release both the proctree_lock and the process 814 * lock as part of its work. 815 */ 816void 817proc_reap(struct thread *td, struct proc *p, int *status, int options) 818{ 819 struct proc *q, *t; 820 821 sx_assert(&proctree_lock, SA_XLOCKED); 822 PROC_LOCK_ASSERT(p, MA_OWNED); 823 PROC_SLOCK_ASSERT(p, MA_OWNED); 824 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 825 826 q = td->td_proc; 827 828 PROC_SUNLOCK(p); 829 if (status) 830 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 831 if (options & WNOWAIT) { 832 /* 833 * Only poll, returning the status. Caller does not wish to 834 * release the proc struct just yet. 835 */ 836 PROC_UNLOCK(p); 837 sx_xunlock(&proctree_lock); 838 return; 839 } 840 841 PROC_LOCK(q); 842 sigqueue_take(p->p_ksi); 843 PROC_UNLOCK(q); 844 845 /* 846 * If we got the child via a ptrace 'attach', we need to give it back 847 * to the old parent. 848 */ 849 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 850 PROC_UNLOCK(p); 851 t = proc_realparent(p); 852 PROC_LOCK(t); 853 PROC_LOCK(p); 854 CTR2(KTR_PTRACE, 855 "wait: traced child %d moved back to parent %d", p->p_pid, 856 t->p_pid); 857 proc_reparent(p, t); 858 p->p_oppid = 0; 859 PROC_UNLOCK(p); 860 pksignal(t, SIGCHLD, p->p_ksi); 861 wakeup(t); 862 cv_broadcast(&p->p_pwait); 863 PROC_UNLOCK(t); 864 sx_xunlock(&proctree_lock); 865 return; 866 } 867 p->p_oppid = 0; 868 PROC_UNLOCK(p); 869 870 /* 871 * Remove other references to this process to ensure we have an 872 * exclusive reference. 873 */ 874 sx_xlock(&allproc_lock); 875 LIST_REMOVE(p, p_list); /* off zombproc */ 876 sx_xunlock(&allproc_lock); 877 LIST_REMOVE(p, p_sibling); 878 reaper_abandon_children(p, true); 879 LIST_REMOVE(p, p_reapsibling); 880 PROC_LOCK(p); 881 clear_orphan(p); 882 PROC_UNLOCK(p); 883 leavepgrp(p); 884 if (p->p_procdesc != NULL) 885 procdesc_reap(p); 886 sx_xunlock(&proctree_lock); 887 888 PROC_LOCK(p); 889 knlist_detach(p->p_klist); 890 p->p_klist = NULL; 891 PROC_UNLOCK(p); 892 893 /* 894 * Removal from allproc list and process group list paired with 895 * PROC_LOCK which was executed during that time should guarantee 896 * nothing can reach this process anymore. As such further locking 897 * is unnecessary. 898 */ 899 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 900 901 PROC_LOCK(q); 902 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 903 PROC_UNLOCK(q); 904 905 /* 906 * Decrement the count of procs running with this uid. 907 */ 908 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 909 910 /* 911 * Destroy resource accounting information associated with the process. 912 */ 913#ifdef RACCT 914 if (racct_enable) { 915 PROC_LOCK(p); 916 racct_sub(p, RACCT_NPROC, 1); 917 PROC_UNLOCK(p); 918 } 919#endif 920 racct_proc_exit(p); 921 922 /* 923 * Free credentials, arguments, and sigacts. 924 */ 925 crfree(p->p_ucred); 926 proc_set_cred(p, NULL); 927 pargs_drop(p->p_args); 928 p->p_args = NULL; 929 sigacts_free(p->p_sigacts); 930 p->p_sigacts = NULL; 931 932 /* 933 * Do any thread-system specific cleanups. 934 */ 935 thread_wait(p); 936 937 /* 938 * Give vm and machine-dependent layer a chance to free anything that 939 * cpu_exit couldn't release while still running in process context. 940 */ 941 vm_waitproc(p); 942#ifdef MAC 943 mac_proc_destroy(p); 944#endif 945 /* 946 * Free any domain policy that's still hiding around. 947 */ 948 vm_domain_policy_cleanup(&p->p_vm_dom_policy); 949 950 KASSERT(FIRST_THREAD_IN_PROC(p), 951 ("proc_reap: no residual thread!")); 952 uma_zfree(proc_zone, p); 953 atomic_add_int(&nprocs, -1); 954} 955 956static int 957proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 958 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 959 int check_only) 960{ 961 struct rusage *rup; 962 963 sx_assert(&proctree_lock, SA_XLOCKED); 964 965 PROC_LOCK(p); 966 967 switch (idtype) { 968 case P_ALL: 969 if (p->p_procdesc != NULL) { 970 PROC_UNLOCK(p); 971 return (0); 972 } 973 break; 974 case P_PID: 975 if (p->p_pid != (pid_t)id) { 976 PROC_UNLOCK(p); 977 return (0); 978 } 979 break; 980 case P_PGID: 981 if (p->p_pgid != (pid_t)id) { 982 PROC_UNLOCK(p); 983 return (0); 984 } 985 break; 986 case P_SID: 987 if (p->p_session->s_sid != (pid_t)id) { 988 PROC_UNLOCK(p); 989 return (0); 990 } 991 break; 992 case P_UID: 993 if (p->p_ucred->cr_uid != (uid_t)id) { 994 PROC_UNLOCK(p); 995 return (0); 996 } 997 break; 998 case P_GID: 999 if (p->p_ucred->cr_gid != (gid_t)id) { 1000 PROC_UNLOCK(p); 1001 return (0); 1002 } 1003 break; 1004 case P_JAILID: 1005 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1006 PROC_UNLOCK(p); 1007 return (0); 1008 } 1009 break; 1010 /* 1011 * It seems that the thread structures get zeroed out 1012 * at process exit. This makes it impossible to 1013 * support P_SETID, P_CID or P_CPUID. 1014 */ 1015 default: 1016 PROC_UNLOCK(p); 1017 return (0); 1018 } 1019 1020 if (p_canwait(td, p)) { 1021 PROC_UNLOCK(p); 1022 return (0); 1023 } 1024 1025 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1026 PROC_UNLOCK(p); 1027 return (0); 1028 } 1029 1030 /* 1031 * This special case handles a kthread spawned by linux_clone 1032 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1033 * functions need to be able to distinguish between waiting 1034 * on a process and waiting on a thread. It is a thread if 1035 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1036 * signifies we want to wait for threads and not processes. 1037 */ 1038 if ((p->p_sigparent != SIGCHLD) ^ 1039 ((options & WLINUXCLONE) != 0)) { 1040 PROC_UNLOCK(p); 1041 return (0); 1042 } 1043 1044 if (siginfo != NULL) { 1045 bzero(siginfo, sizeof(*siginfo)); 1046 siginfo->si_errno = 0; 1047 1048 /* 1049 * SUSv4 requires that the si_signo value is always 1050 * SIGCHLD. Obey it despite the rfork(2) interface 1051 * allows to request other signal for child exit 1052 * notification. 1053 */ 1054 siginfo->si_signo = SIGCHLD; 1055 1056 /* 1057 * This is still a rough estimate. We will fix the 1058 * cases TRAPPED, STOPPED, and CONTINUED later. 1059 */ 1060 if (WCOREDUMP(p->p_xsig)) { 1061 siginfo->si_code = CLD_DUMPED; 1062 siginfo->si_status = WTERMSIG(p->p_xsig); 1063 } else if (WIFSIGNALED(p->p_xsig)) { 1064 siginfo->si_code = CLD_KILLED; 1065 siginfo->si_status = WTERMSIG(p->p_xsig); 1066 } else { 1067 siginfo->si_code = CLD_EXITED; 1068 siginfo->si_status = p->p_xexit; 1069 } 1070 1071 siginfo->si_pid = p->p_pid; 1072 siginfo->si_uid = p->p_ucred->cr_uid; 1073 1074 /* 1075 * The si_addr field would be useful additional 1076 * detail, but apparently the PC value may be lost 1077 * when we reach this point. bzero() above sets 1078 * siginfo->si_addr to NULL. 1079 */ 1080 } 1081 1082 /* 1083 * There should be no reason to limit resources usage info to 1084 * exited processes only. A snapshot about any resources used 1085 * by a stopped process may be exactly what is needed. 1086 */ 1087 if (wrusage != NULL) { 1088 rup = &wrusage->wru_self; 1089 *rup = p->p_ru; 1090 PROC_STATLOCK(p); 1091 calcru(p, &rup->ru_utime, &rup->ru_stime); 1092 PROC_STATUNLOCK(p); 1093 1094 rup = &wrusage->wru_children; 1095 *rup = p->p_stats->p_cru; 1096 calccru(p, &rup->ru_utime, &rup->ru_stime); 1097 } 1098 1099 if (p->p_state == PRS_ZOMBIE && !check_only) { 1100 PROC_SLOCK(p); 1101 proc_reap(td, p, status, options); 1102 return (-1); 1103 } 1104 return (1); 1105} 1106 1107int 1108kern_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 1148static void 1149report_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 1180int 1181kern_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 1188 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1189 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1190 AUDIT_ARG_VALUE(options); 1191 1192 q = td->td_proc; 1193 1194 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1195 PROC_LOCK(q); 1196 id = (id_t)q->p_pgid; 1197 PROC_UNLOCK(q); 1198 idtype = P_PGID; 1199 } 1200 1201 /* If we don't know the option, just return. */ 1202 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1203 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1204 return (EINVAL); 1205 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1206 /* 1207 * We will be unable to find any matching processes, 1208 * because there are no known events to look for. 1209 * Prefer to return error instead of blocking 1210 * indefinitely. 1211 */ 1212 return (EINVAL); 1213 } 1214 1215loop: 1216 if (q->p_flag & P_STATCHILD) { 1217 PROC_LOCK(q); 1218 q->p_flag &= ~P_STATCHILD; 1219 PROC_UNLOCK(q); 1220 } 1221 nfound = 0; 1222 sx_xlock(&proctree_lock); 1223 LIST_FOREACH(p, &q->p_children, p_sibling) { 1224 pid = p->p_pid; 1225 ret = proc_to_reap(td, p, idtype, id, status, options, 1226 wrusage, siginfo, 0); 1227 if (ret == 0) 1228 continue; 1229 else if (ret == 1) 1230 nfound++; 1231 else { 1232 td->td_retval[0] = pid; 1233 return (0); 1234 } 1235 1236 PROC_LOCK_ASSERT(p, MA_OWNED); 1237 1238 if ((options & (WTRAPPED | WUNTRACED)) != 0) 1239 PROC_SLOCK(p); 1240 1241 if ((options & WTRAPPED) != 0 && 1242 (p->p_flag & P_TRACED) != 0 && 1243 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1244 p->p_suspcount == p->p_numthreads && 1245 (p->p_flag & P_WAITED) == 0) { 1246 PROC_SUNLOCK(p); 1247 CTR4(KTR_PTRACE, 1248 "wait: returning trapped pid %d status %#x " 1249 "(xstat %d) xthread %d", 1250 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1251 p->p_xthread != NULL ? 1252 p->p_xthread->td_tid : -1); 1253 report_alive_proc(td, p, siginfo, status, options, 1254 CLD_TRAPPED); 1255 return (0); 1256 } 1257 if ((options & WUNTRACED) != 0 && 1258 (p->p_flag & P_STOPPED_SIG) != 0 && 1259 p->p_suspcount == p->p_numthreads && 1260 (p->p_flag & P_WAITED) == 0) { 1261 PROC_SUNLOCK(p); 1262 report_alive_proc(td, p, siginfo, status, options, 1263 CLD_STOPPED); 1264 return (0); 1265 } 1266 if ((options & (WTRAPPED | WUNTRACED)) != 0) 1267 PROC_SUNLOCK(p); 1268 if ((options & WCONTINUED) != 0 && 1269 (p->p_flag & P_CONTINUED) != 0) { 1270 report_alive_proc(td, p, siginfo, status, options, 1271 CLD_CONTINUED); 1272 return (0); 1273 } 1274 PROC_UNLOCK(p); 1275 } 1276 1277 /* 1278 * Look in the orphans list too, to allow the parent to 1279 * collect it's child exit status even if child is being 1280 * debugged. 1281 * 1282 * Debugger detaches from the parent upon successful 1283 * switch-over from parent to child. At this point due to 1284 * re-parenting the parent loses the child to debugger and a 1285 * wait4(2) call would report that it has no children to wait 1286 * for. By maintaining a list of orphans we allow the parent 1287 * to successfully wait until the child becomes a zombie. 1288 */ 1289 if (nfound == 0) { 1290 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1291 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1292 NULL, NULL, 1); 1293 if (ret != 0) { 1294 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1295 (int)td->td_retval[0])); 1296 PROC_UNLOCK(p); 1297 nfound++; 1298 break; 1299 } 1300 } 1301 } 1302 if (nfound == 0) { 1303 sx_xunlock(&proctree_lock); 1304 return (ECHILD); 1305 } 1306 if (options & WNOHANG) { 1307 sx_xunlock(&proctree_lock); 1308 td->td_retval[0] = 0; 1309 return (0); 1310 } 1311 PROC_LOCK(q); 1312 sx_xunlock(&proctree_lock); 1313 if (q->p_flag & P_STATCHILD) { 1314 q->p_flag &= ~P_STATCHILD; 1315 error = 0; 1316 } else 1317 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1318 PROC_UNLOCK(q); 1319 if (error) 1320 return (error); 1321 goto loop; 1322} 1323 1324/* 1325 * Make process 'parent' the new parent of process 'child'. 1326 * Must be called with an exclusive hold of proctree lock. 1327 */ 1328void 1329proc_reparent(struct proc *child, struct proc *parent) 1330{ 1331 1332 sx_assert(&proctree_lock, SX_XLOCKED); 1333 PROC_LOCK_ASSERT(child, MA_OWNED); 1334 if (child->p_pptr == parent) 1335 return; 1336 1337 PROC_LOCK(child->p_pptr); 1338 sigqueue_take(child->p_ksi); 1339 PROC_UNLOCK(child->p_pptr); 1340 LIST_REMOVE(child, p_sibling); 1341 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1342 1343 clear_orphan(child); 1344 if (child->p_flag & P_TRACED) { 1345 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1346 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1347 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1348 p_orphan); 1349 } else { 1350 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1351 child, p_orphan); 1352 } 1353 child->p_treeflag |= P_TREE_ORPHANED; 1354 } 1355 1356 child->p_pptr = parent; 1357} 1358