kern_exit.c revision 225617
11541Srgrimes/*- 21541Srgrimes * Copyright (c) 1982, 1986, 1989, 1991, 1993 31541Srgrimes * The Regents of the University of California. All rights reserved. 41541Srgrimes * (c) UNIX System Laboratories, Inc. 51541Srgrimes * All or some portions of this file are derived from material licensed 61541Srgrimes * to the University of California by American Telephone and Telegraph 71541Srgrimes * Co. or Unix System Laboratories, Inc. and are reproduced herein with 81541Srgrimes * the permission of UNIX System Laboratories, Inc. 91541Srgrimes * 101541Srgrimes * Redistribution and use in source and binary forms, with or without 111541Srgrimes * modification, are permitted provided that the following conditions 121541Srgrimes * are met: 131541Srgrimes * 1. Redistributions of source code must retain the above copyright 141541Srgrimes * notice, this list of conditions and the following disclaimer. 151541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 161541Srgrimes * notice, this list of conditions and the following disclaimer in the 171541Srgrimes * documentation and/or other materials provided with the distribution. 181541Srgrimes * 4. Neither the name of the University nor the names of its contributors 191541Srgrimes * may be used to endorse or promote products derived from this software 201541Srgrimes * without specific prior written permission. 211541Srgrimes * 221541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 231541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 241541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 251541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 261541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 271541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 281541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 291541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 301541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 311541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 321541Srgrimes * SUCH DAMAGE. 331541Srgrimes * 3450477Speter * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 351541Srgrimes */ 361541Srgrimes 371541Srgrimes#include <sys/cdefs.h> 381541Srgrimes__FBSDID("$FreeBSD: head/sys/kern/kern_exit.c 225617 2011-09-16 13:58:51Z kmacy $"); 391541Srgrimes 401541Srgrimes#include "opt_compat.h" 411541Srgrimes#include "opt_kdtrace.h" 421541Srgrimes#include "opt_ktrace.h" 4312577Sbde#include "opt_procdesc.h" 441541Srgrimes 451541Srgrimes#include <sys/param.h> 4624206Sbde#include <sys/systm.h> 4724206Sbde#include <sys/sysproto.h> 481541Srgrimes#include <sys/capability.h> 493308Sphk#include <sys/eventhandler.h> 5012517Sjulian#include <sys/kernel.h> 5129357Speter#include <sys/malloc.h> 5241086Struckman#include <sys/lock.h> 5370069Sjhb#include <sys/mutex.h> 5412517Sjulian#include <sys/proc.h> 551541Srgrimes#include <sys/procdesc.h> 561541Srgrimes#include <sys/pioctl.h> 571541Srgrimes#include <sys/jail.h> 581541Srgrimes#include <sys/tty.h> 591541Srgrimes#include <sys/wait.h> 6012675Sjulian#include <sys/vmmeter.h> 6112675Sjulian#include <sys/vnode.h> 6212675Sjulian#include <sys/racct.h> 6312675Sjulian#include <sys/resourcevar.h> 6429357Speter#include <sys/sbuf.h> 6512675Sjulian#include <sys/signalvar.h> 6670069Sjhb#include <sys/sched.h> 6770069Sjhb#include <sys/sx.h> 6812675Sjulian#include <sys/syscallsubr.h> 6947625Sphk#include <sys/syslog.h> 7047625Sphk#include <sys/ptrace.h> 7147625Sphk#include <sys/acct.h> /* for acct_process() function prototype */ 7247625Sphk#include <sys/filedesc.h> 7347625Sphk#include <sys/sdt.h> 7447625Sphk#include <sys/shm.h> 7547625Sphk#include <sys/sem.h> 7647625Sphk#ifdef KTRACE 7747625Sphk#include <sys/ktrace.h> 7847625Sphk#endif 7947625Sphk 8047625Sphk#include <security/audit/audit.h> 8147625Sphk#include <security/mac/mac_framework.h> 8247625Sphk 8347625Sphk#include <vm/vm.h> 8412675Sjulian#include <vm/vm_extern.h> 8512819Sphk#include <vm/vm_param.h> 861541Srgrimes#include <vm/pmap.h> 871541Srgrimes#include <vm/vm_map.h> 8841087Struckman#include <vm/vm_page.h> 8970069Sjhb#include <vm/uma.h> 901541Srgrimes 911541Srgrimes#ifdef KDTRACE_HOOKS 921541Srgrimes#include <sys/dtrace_bsd.h> 931541Srgrimesdtrace_execexit_func_t dtrace_fasttrap_exit; 9470069Sjhb#endif 9570069Sjhb 9670069SjhbSDT_PROVIDER_DECLARE(proc); 9770069SjhbSDT_PROBE_DEFINE(proc, kernel, , exit, exit); 9870069SjhbSDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int"); 991541Srgrimes 10012675Sjulian/* Hook for NFS teardown procedure. */ 10183366Sjulianvoid (*nlminfo_release_p)(struct proc *p); 1021541Srgrimes 1031541Srgrimes/* 1041541Srgrimes * exit -- death of process. 1051541Srgrimes */ 10670069Sjhbvoid 10783366Sjuliansys_sys_exit(struct thread *td, struct sys_exit_args *uap) 10870069Sjhb{ 10970069Sjhb 1101541Srgrimes exit1(td, W_EXITCODE(uap->rval, 0)); 1111541Srgrimes /* NOTREACHED */ 1121541Srgrimes} 1131541Srgrimes 11412675Sjulian/* 11583366Sjulian * Exit: deallocate address space and other resources, change proc state to 1161541Srgrimes * zombie, and unlink proc from allproc and parent's lists. Save exit status 1171541Srgrimes * and rusage for wait(). Check for child processes and orphan them. 11877057Sphk */ 11970069Sjhbvoid 1201541Srgrimesexit1(struct thread *td, int rv) 12196122Salfred{ 1221541Srgrimes struct proc *p, *nq, *q; 1231541Srgrimes struct vnode *vtmp; 1241541Srgrimes struct vnode *ttyvp = NULL; 1251541Srgrimes struct plimit *plim; 12612675Sjulian int locked; 12769741Sphk 1281541Srgrimes mtx_assert(&Giant, MA_NOTOWNED); 12969741Sphk 130106917Stmm p = td->td_proc; 1311541Srgrimes /* 1321541Srgrimes * XXX in case we're rebooting we just let init die in order to 1331541Srgrimes * work around an unsolved stack overflow seen very late during 1341541Srgrimes * shutdown on sparc64 when the gmirror worker process exists. 1351541Srgrimes */ 1361541Srgrimes if (p == initproc && rebooting == 0) { 1371541Srgrimes printf("init died (signal %d, exit %d)\n", 1381541Srgrimes WTERMSIG(rv), WEXITSTATUS(rv)); 13999012Salfred panic("Going nowhere without my init!"); 1401541Srgrimes } 1411541Srgrimes 1421541Srgrimes /* 1431541Srgrimes * MUST abort all other threads before proceeding past here. 1441541Srgrimes */ 1451541Srgrimes PROC_LOCK(p); 1461541Srgrimes while (p->p_flag & P_HADTHREADS) { 1471541Srgrimes /* 148106917Stmm * First check if some other thread got here before us.. 1491541Srgrimes * if so, act apropriatly, (exit or suspend); 15036179Sphk */ 151106917Stmm thread_suspend_check(0); 1521541Srgrimes 1531541Srgrimes /* 15436179Sphk * Kill off the other threads. This requires 155106917Stmm * some co-operation from other parts of the kernel 1561541Srgrimes * so it may not be instantaneous. With this state set 1571541Srgrimes * any thread entering the kernel from userspace will 1581541Srgrimes * thread_exit() in trap(). Any thread attempting to 15936179Sphk * sleep will return immediately with EINTR or EWOULDBLOCK 1601541Srgrimes * which will hopefully force them to back out to userland 1611541Srgrimes * freeing resources as they go. Any thread attempting 1621541Srgrimes * to return to userland will thread_exit() from userret(). 1631541Srgrimes * thread_exit() will unsuspend us when the last of the 1641541Srgrimes * other threads exits. 1651541Srgrimes * If there is already a thread singler after resumption, 16612675Sjulian * calling thread_single will fail; in that case, we just 16783366Sjulian * re-check all suspension request, the thread should 1681541Srgrimes * either be suspended there or exit. 16929357Speter */ 17029357Speter if (! thread_single(SINGLE_EXIT)) 1711541Srgrimes break; 17229357Speter 1731541Srgrimes /* 17446568Speter * All other activity in this process is now stopped. 17529357Speter * Threading support has been turned off. 17629357Speter */ 17729357Speter } 17883805Sjhb KASSERT(p->p_numthreads == 1, 17946568Speter ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 1801541Srgrimes racct_sub(p, RACCT_NTHR, 1); 18129357Speter /* 1821541Srgrimes * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 1831541Srgrimes * on our vmspace, so we should block below until they have 18470069Sjhb * released their reference to us. Note that if they have 18570069Sjhb * requested S_EXIT stops we will block here until they ack 18670069Sjhb * via PIOCCONT. 18770069Sjhb */ 18877057Sphk _STOPEVENT(p, S_EXIT, rv); 18970069Sjhb 19077057Sphk /* 19177057Sphk * Note that we are exiting and do another wakeup of anyone in 19277057Sphk * PIOCWAIT in case they aren't listening for S_EXIT stops or 19377057Sphk * decided to wait again after we told them we are exiting. 19477057Sphk */ 19570239Sphk p->p_flag |= P_WEXIT; 1961541Srgrimes wakeup(&p->p_stype); 19741086Struckman 19895883Salfred /* 1991541Srgrimes * Wait for any processes that have a hold on our vmspace to 2001541Srgrimes * release their reference. 2011541Srgrimes */ 2021541Srgrimes while (p->p_lock > 0) 20370069Sjhb msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 20470069Sjhb 2051541Srgrimes p->p_xstat = rv; /* Let event handler change exit status */ 2061541Srgrimes PROC_UNLOCK(p); 2071541Srgrimes /* Drain the limit callout while we don't have the proc locked */ 20812675Sjulian callout_drain(&p->p_limco); 20983366Sjulian 2101541Srgrimes#ifdef AUDIT 211106917Stmm /* 2121541Srgrimes * The Sun BSM exit token contains two components: an exit status as 2131541Srgrimes * passed to exit(), and a return value to indicate what sort of exit 2141541Srgrimes * it was. The exit status is WEXITSTATUS(rv), but it's not clear 2151541Srgrimes * what the return value is. 2161541Srgrimes */ 2171541Srgrimes AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0); 218106917Stmm AUDIT_SYSCALL_EXIT(0, td); 2191541Srgrimes#endif 2201541Srgrimes 22136179Sphk /* Are we a task leader? */ 2221541Srgrimes if (p == p->p_leader) { 2231541Srgrimes mtx_lock(&ppeers_lock); 2241541Srgrimes q = p->p_peers; 2251541Srgrimes while (q != NULL) { 2261541Srgrimes PROC_LOCK(q); 2271541Srgrimes kern_psignal(q, SIGKILL); 2281541Srgrimes PROC_UNLOCK(q); 2291541Srgrimes q = q->p_peers; 2301541Srgrimes } 2311541Srgrimes while (p->p_peers != NULL) 2321541Srgrimes msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 2331541Srgrimes mtx_unlock(&ppeers_lock); 2341541Srgrimes } 23541086Struckman 23641086Struckman /* 23741086Struckman * Check if any loadable modules need anything done at process exit. 23841086Struckman * E.g. SYSV IPC stuff 239104393Struckman * XXX what if one of these generates an error? 2401541Srgrimes */ 2411541Srgrimes EVENTHANDLER_INVOKE(process_exit, p); 24241086Struckman 24341086Struckman /* 24441086Struckman * If parent is waiting for us to exit or exec, 24541086Struckman * P_PPWAIT is set; we will wakeup the parent below. 24641086Struckman */ 2471541Srgrimes PROC_LOCK(p); 248104393Struckman rv = p->p_xstat; /* Event handler could change exit status */ 2491541Srgrimes stopprofclock(p); 2501541Srgrimes p->p_flag &= ~(P_TRACED | P_PPWAIT); 2511541Srgrimes 2528161Sjkh /* 2531541Srgrimes * Stop the real interval timer. If the handler is currently 2541541Srgrimes * executing, prevent it from rearming itself and let it finish. 2551541Srgrimes */ 25612517Sjulian if (timevalisset(&p->p_realtimer.it_value) && 25712675Sjulian callout_stop(&p->p_itcallout) == 0) { 25869741Sphk timevalclear(&p->p_realtimer.it_interval); 25912517Sjulian msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 26069741Sphk KASSERT(!timevalisset(&p->p_realtimer.it_value), 26150254Sphk ("realtime timer is still armed")); 26212517Sjulian } 26312517Sjulian PROC_UNLOCK(p); 26412517Sjulian 265 /* 266 * Reset any sigio structures pointing to us as a result of 267 * F_SETOWN with our pid. 268 */ 269 funsetownlst(&p->p_sigiolst); 270 271 /* 272 * If this process has an nlminfo data area (for lockd), release it 273 */ 274 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 275 (*nlminfo_release_p)(p); 276 277 /* 278 * Close open files and release open-file table. 279 * This may block! 280 */ 281 fdfree(td); 282 283 /* 284 * If this thread tickled GEOM, we need to wait for the giggling to 285 * stop before we return to userland 286 */ 287 if (td->td_pflags & TDP_GEOM) 288 g_waitidle(); 289 290 /* 291 * Remove ourself from our leader's peer list and wake our leader. 292 */ 293 mtx_lock(&ppeers_lock); 294 if (p->p_leader->p_peers) { 295 q = p->p_leader; 296 while (q->p_peers != p) 297 q = q->p_peers; 298 q->p_peers = p->p_peers; 299 wakeup(p->p_leader); 300 } 301 mtx_unlock(&ppeers_lock); 302 303 vmspace_exit(td); 304 305 sx_xlock(&proctree_lock); 306 if (SESS_LEADER(p)) { 307 struct session *sp = p->p_session; 308 struct tty *tp; 309 310 /* 311 * s_ttyp is not zero'd; we use this to indicate that 312 * the session once had a controlling terminal. (for 313 * logging and informational purposes) 314 */ 315 SESS_LOCK(sp); 316 ttyvp = sp->s_ttyvp; 317 tp = sp->s_ttyp; 318 sp->s_ttyvp = NULL; 319 sp->s_ttydp = NULL; 320 sp->s_leader = NULL; 321 SESS_UNLOCK(sp); 322 323 /* 324 * Signal foreground pgrp and revoke access to 325 * controlling terminal if it has not been revoked 326 * already. 327 * 328 * Because the TTY may have been revoked in the mean 329 * time and could already have a new session associated 330 * with it, make sure we don't send a SIGHUP to a 331 * foreground process group that does not belong to this 332 * session. 333 */ 334 335 if (tp != NULL) { 336 tty_lock(tp); 337 if (tp->t_session == sp) 338 tty_signal_pgrp(tp, SIGHUP); 339 tty_unlock(tp); 340 } 341 342 if (ttyvp != NULL) { 343 sx_xunlock(&proctree_lock); 344 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 345 VOP_REVOKE(ttyvp, REVOKEALL); 346 VOP_UNLOCK(ttyvp, 0); 347 } 348 sx_xlock(&proctree_lock); 349 } 350 } 351 fixjobc(p, p->p_pgrp, 0); 352 sx_xunlock(&proctree_lock); 353 (void)acct_process(td); 354 355 /* Release the TTY now we've unlocked everything. */ 356 if (ttyvp != NULL) 357 vrele(ttyvp); 358#ifdef KTRACE 359 ktrprocexit(td); 360#endif 361 /* 362 * Release reference to text vnode 363 */ 364 if ((vtmp = p->p_textvp) != NULL) { 365 p->p_textvp = NULL; 366 locked = VFS_LOCK_GIANT(vtmp->v_mount); 367 vrele(vtmp); 368 VFS_UNLOCK_GIANT(locked); 369 } 370 371 /* 372 * Release our limits structure. 373 */ 374 PROC_LOCK(p); 375 plim = p->p_limit; 376 p->p_limit = NULL; 377 PROC_UNLOCK(p); 378 lim_free(plim); 379 380 tidhash_remove(td); 381 382 /* 383 * Remove proc from allproc queue and pidhash chain. 384 * Place onto zombproc. Unlink from parent's child list. 385 */ 386 sx_xlock(&allproc_lock); 387 LIST_REMOVE(p, p_list); 388 LIST_INSERT_HEAD(&zombproc, p, p_list); 389 LIST_REMOVE(p, p_hash); 390 sx_xunlock(&allproc_lock); 391 392 /* 393 * Call machine-dependent code to release any 394 * machine-dependent resources other than the address space. 395 * The address space is released by "vmspace_exitfree(p)" in 396 * vm_waitproc(). 397 */ 398 cpu_exit(td); 399 400 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 401 402 /* 403 * Reparent all of our children to init. 404 */ 405 sx_xlock(&proctree_lock); 406 q = LIST_FIRST(&p->p_children); 407 if (q != NULL) /* only need this if any child is S_ZOMB */ 408 wakeup(initproc); 409 for (; q != NULL; q = nq) { 410 nq = LIST_NEXT(q, p_sibling); 411 PROC_LOCK(q); 412 proc_reparent(q, initproc); 413 q->p_sigparent = SIGCHLD; 414 /* 415 * Traced processes are killed 416 * since their existence means someone is screwing up. 417 */ 418 if (q->p_flag & P_TRACED) { 419 struct thread *temp; 420 421 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 422 FOREACH_THREAD_IN_PROC(q, temp) 423 temp->td_dbgflags &= ~TDB_SUSPEND; 424 kern_psignal(q, SIGKILL); 425 } 426 PROC_UNLOCK(q); 427 } 428 429 /* Save exit status. */ 430 PROC_LOCK(p); 431 p->p_xthread = td; 432 433 /* Tell the prison that we are gone. */ 434 prison_proc_free(p->p_ucred->cr_prison); 435 436#ifdef KDTRACE_HOOKS 437 /* 438 * Tell the DTrace fasttrap provider about the exit if it 439 * has declared an interest. 440 */ 441 if (dtrace_fasttrap_exit) 442 dtrace_fasttrap_exit(p); 443#endif 444 445 /* 446 * Notify interested parties of our demise. 447 */ 448 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 449 450#ifdef KDTRACE_HOOKS 451 int reason = CLD_EXITED; 452 if (WCOREDUMP(rv)) 453 reason = CLD_DUMPED; 454 else if (WIFSIGNALED(rv)) 455 reason = CLD_KILLED; 456 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 457#endif 458 459 /* 460 * Just delete all entries in the p_klist. At this point we won't 461 * report any more events, and there are nasty race conditions that 462 * can beat us if we don't. 463 */ 464 knlist_clear(&p->p_klist, 1); 465 466 /* 467 * If this is a process with a descriptor, we may not need to deliver 468 * a signal to the parent. proctree_lock is held over 469 * procdesc_exit() to serialize concurrent calls to close() and 470 * exit(). 471 */ 472#ifdef PROCDESC 473 if (p->p_procdesc == NULL || procdesc_exit(p)) { 474#endif 475 /* 476 * Notify parent that we're gone. If parent has the 477 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 478 * notify process 1 instead (and hope it will handle this 479 * situation). 480 */ 481 PROC_LOCK(p->p_pptr); 482 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 483 if (p->p_pptr->p_sigacts->ps_flag & 484 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 485 struct proc *pp; 486 487 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 488 pp = p->p_pptr; 489 PROC_UNLOCK(pp); 490 proc_reparent(p, initproc); 491 p->p_sigparent = SIGCHLD; 492 PROC_LOCK(p->p_pptr); 493 494 /* 495 * Notify parent, so in case he was wait(2)ing or 496 * executing waitpid(2) with our pid, he will 497 * continue. 498 */ 499 wakeup(pp); 500 } else 501 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 502 503 if (p->p_pptr == initproc) 504 kern_psignal(p->p_pptr, SIGCHLD); 505 else if (p->p_sigparent != 0) { 506 if (p->p_sigparent == SIGCHLD) 507 childproc_exited(p); 508 else /* LINUX thread */ 509 kern_psignal(p->p_pptr, p->p_sigparent); 510 } 511#ifdef PROCDESC 512 } else 513 PROC_LOCK(p->p_pptr); 514#endif 515 sx_xunlock(&proctree_lock); 516 517 /* 518 * The state PRS_ZOMBIE prevents other proesses from sending 519 * signal to the process, to avoid memory leak, we free memory 520 * for signal queue at the time when the state is set. 521 */ 522 sigqueue_flush(&p->p_sigqueue); 523 sigqueue_flush(&td->td_sigqueue); 524 525 /* 526 * We have to wait until after acquiring all locks before 527 * changing p_state. We need to avoid all possible context 528 * switches (including ones from blocking on a mutex) while 529 * marked as a zombie. We also have to set the zombie state 530 * before we release the parent process' proc lock to avoid 531 * a lost wakeup. So, we first call wakeup, then we grab the 532 * sched lock, update the state, and release the parent process' 533 * proc lock. 534 */ 535 wakeup(p->p_pptr); 536 cv_broadcast(&p->p_pwait); 537 sched_exit(p->p_pptr, td); 538 PROC_SLOCK(p); 539 p->p_state = PRS_ZOMBIE; 540 PROC_UNLOCK(p->p_pptr); 541 542 /* 543 * Hopefully no one will try to deliver a signal to the process this 544 * late in the game. 545 */ 546 knlist_destroy(&p->p_klist); 547 548 /* 549 * Save our children's rusage information in our exit rusage. 550 */ 551 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 552 553 /* 554 * Make sure the scheduler takes this thread out of its tables etc. 555 * This will also release this thread's reference to the ucred. 556 * Other thread parts to release include pcb bits and such. 557 */ 558 thread_exit(); 559} 560 561 562#ifndef _SYS_SYSPROTO_H_ 563struct abort2_args { 564 char *why; 565 int nargs; 566 void **args; 567}; 568#endif 569 570int 571sys_abort2(struct thread *td, struct abort2_args *uap) 572{ 573 struct proc *p = td->td_proc; 574 struct sbuf *sb; 575 void *uargs[16]; 576 int error, i, sig; 577 578 /* 579 * Do it right now so we can log either proper call of abort2(), or 580 * note, that invalid argument was passed. 512 is big enough to 581 * handle 16 arguments' descriptions with additional comments. 582 */ 583 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 584 sbuf_clear(sb); 585 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 586 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 587 /* 588 * Since we can't return from abort2(), send SIGKILL in cases, where 589 * abort2() was called improperly 590 */ 591 sig = SIGKILL; 592 /* Prevent from DoSes from user-space. */ 593 if (uap->nargs < 0 || uap->nargs > 16) 594 goto out; 595 if (uap->nargs > 0) { 596 if (uap->args == NULL) 597 goto out; 598 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 599 if (error != 0) 600 goto out; 601 } 602 /* 603 * Limit size of 'reason' string to 128. Will fit even when 604 * maximal number of arguments was chosen to be logged. 605 */ 606 if (uap->why != NULL) { 607 error = sbuf_copyin(sb, uap->why, 128); 608 if (error < 0) 609 goto out; 610 } else { 611 sbuf_printf(sb, "(null)"); 612 } 613 if (uap->nargs > 0) { 614 sbuf_printf(sb, "("); 615 for (i = 0;i < uap->nargs; i++) 616 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 617 sbuf_printf(sb, ")"); 618 } 619 /* 620 * Final stage: arguments were proper, string has been 621 * successfully copied from userspace, and copying pointers 622 * from user-space succeed. 623 */ 624 sig = SIGABRT; 625out: 626 if (sig == SIGKILL) { 627 sbuf_trim(sb); 628 sbuf_printf(sb, " (Reason text inaccessible)"); 629 } 630 sbuf_cat(sb, "\n"); 631 sbuf_finish(sb); 632 log(LOG_INFO, "%s", sbuf_data(sb)); 633 sbuf_delete(sb); 634 exit1(td, W_EXITCODE(0, sig)); 635 return (0); 636} 637 638 639#ifdef COMPAT_43 640/* 641 * The dirty work is handled by kern_wait(). 642 */ 643int 644owait(struct thread *td, struct owait_args *uap __unused) 645{ 646 int error, status; 647 648 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 649 if (error == 0) 650 td->td_retval[1] = status; 651 return (error); 652} 653#endif /* COMPAT_43 */ 654 655/* 656 * The dirty work is handled by kern_wait(). 657 */ 658int 659sys_wait4(struct thread *td, struct wait_args *uap) 660{ 661 struct rusage ru, *rup; 662 int error, status; 663 664 if (uap->rusage != NULL) 665 rup = &ru; 666 else 667 rup = NULL; 668 error = kern_wait(td, uap->pid, &status, uap->options, rup); 669 if (uap->status != NULL && error == 0) 670 error = copyout(&status, uap->status, sizeof(status)); 671 if (uap->rusage != NULL && error == 0) 672 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 673 return (error); 674} 675 676/* 677 * Reap the remains of a zombie process and optionally return status and 678 * rusage. Asserts and will release both the proctree_lock and the process 679 * lock as part of its work. 680 */ 681void 682proc_reap(struct thread *td, struct proc *p, int *status, int options, 683 struct rusage *rusage) 684{ 685 struct proc *q, *t; 686 687 sx_assert(&proctree_lock, SA_XLOCKED); 688 PROC_LOCK_ASSERT(p, MA_OWNED); 689 PROC_SLOCK_ASSERT(p, MA_OWNED); 690 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 691 692 q = td->td_proc; 693 if (rusage) { 694 *rusage = p->p_ru; 695 calcru(p, &rusage->ru_utime, &rusage->ru_stime); 696 } 697 PROC_SUNLOCK(p); 698 td->td_retval[0] = p->p_pid; 699 if (status) 700 *status = p->p_xstat; /* convert to int */ 701 if (options & WNOWAIT) { 702 /* 703 * Only poll, returning the status. Caller does not wish to 704 * release the proc struct just yet. 705 */ 706 PROC_UNLOCK(p); 707 sx_xunlock(&proctree_lock); 708 return; 709 } 710 711 PROC_LOCK(q); 712 sigqueue_take(p->p_ksi); 713 PROC_UNLOCK(q); 714 PROC_UNLOCK(p); 715 716 /* 717 * If we got the child via a ptrace 'attach', we need to give it back 718 * to the old parent. 719 */ 720 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 721 PROC_LOCK(p); 722 proc_reparent(p, t); 723 p->p_pptr->p_dbg_child--; 724 p->p_oppid = 0; 725 PROC_UNLOCK(p); 726 pksignal(t, SIGCHLD, p->p_ksi); 727 wakeup(t); 728 cv_broadcast(&p->p_pwait); 729 PROC_UNLOCK(t); 730 sx_xunlock(&proctree_lock); 731 return; 732 } 733 734 /* 735 * Remove other references to this process to ensure we have an 736 * exclusive reference. 737 */ 738 sx_xlock(&allproc_lock); 739 LIST_REMOVE(p, p_list); /* off zombproc */ 740 sx_xunlock(&allproc_lock); 741 LIST_REMOVE(p, p_sibling); 742 leavepgrp(p); 743#ifdef PROCDESC 744 if (p->p_procdesc != NULL) 745 procdesc_reap(p); 746#endif 747 sx_xunlock(&proctree_lock); 748 749 /* 750 * As a side effect of this lock, we know that all other writes to 751 * this proc are visible now, so no more locking is needed for p. 752 */ 753 PROC_LOCK(p); 754 p->p_xstat = 0; /* XXX: why? */ 755 PROC_UNLOCK(p); 756 PROC_LOCK(q); 757 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 758 PROC_UNLOCK(q); 759 760 /* 761 * Decrement the count of procs running with this uid. 762 */ 763 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 764 765 /* 766 * Destroy resource accounting information associated with the process. 767 */ 768 racct_proc_exit(p); 769#ifdef RACCT 770 PROC_LOCK(p->p_pptr); 771 racct_sub(p->p_pptr, RACCT_NPROC, 1); 772 PROC_UNLOCK(p->p_pptr); 773#endif 774 775 /* 776 * Free credentials, arguments, and sigacts. 777 */ 778 crfree(p->p_ucred); 779 p->p_ucred = NULL; 780 pargs_drop(p->p_args); 781 p->p_args = NULL; 782 sigacts_free(p->p_sigacts); 783 p->p_sigacts = NULL; 784 785 /* 786 * Do any thread-system specific cleanups. 787 */ 788 thread_wait(p); 789 790 /* 791 * Give vm and machine-dependent layer a chance to free anything that 792 * cpu_exit couldn't release while still running in process context. 793 */ 794 vm_waitproc(p); 795#ifdef MAC 796 mac_proc_destroy(p); 797#endif 798 KASSERT(FIRST_THREAD_IN_PROC(p), 799 ("proc_reap: no residual thread!")); 800 uma_zfree(proc_zone, p); 801 sx_xlock(&allproc_lock); 802 nprocs--; 803 sx_xunlock(&allproc_lock); 804} 805 806int 807kern_wait(struct thread *td, pid_t pid, int *status, int options, 808 struct rusage *rusage) 809{ 810 struct proc *p, *q; 811 int error, nfound; 812 813 AUDIT_ARG_PID(pid); 814 AUDIT_ARG_VALUE(options); 815 816 q = td->td_proc; 817 if (pid == 0) { 818 PROC_LOCK(q); 819 pid = -q->p_pgid; 820 PROC_UNLOCK(q); 821 } 822 /* If we don't know the option, just return. */ 823 if (options & ~(WUNTRACED|WNOHANG|WCONTINUED|WNOWAIT|WLINUXCLONE)) 824 return (EINVAL); 825loop: 826 if (q->p_flag & P_STATCHILD) { 827 PROC_LOCK(q); 828 q->p_flag &= ~P_STATCHILD; 829 PROC_UNLOCK(q); 830 } 831 nfound = 0; 832 sx_xlock(&proctree_lock); 833 LIST_FOREACH(p, &q->p_children, p_sibling) { 834 PROC_LOCK(p); 835 if (pid != WAIT_ANY && 836 p->p_pid != pid && p->p_pgid != -pid) { 837 PROC_UNLOCK(p); 838 continue; 839 } 840 if (p_canwait(td, p)) { 841 PROC_UNLOCK(p); 842 continue; 843 } 844 845 /* 846 * This special case handles a kthread spawned by linux_clone 847 * (see linux_misc.c). The linux_wait4 and linux_waitpid 848 * functions need to be able to distinguish between waiting 849 * on a process and waiting on a thread. It is a thread if 850 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 851 * signifies we want to wait for threads and not processes. 852 */ 853 if ((p->p_sigparent != SIGCHLD) ^ 854 ((options & WLINUXCLONE) != 0)) { 855 PROC_UNLOCK(p); 856 continue; 857 } 858 859 nfound++; 860 PROC_SLOCK(p); 861 if (p->p_state == PRS_ZOMBIE) { 862 proc_reap(td, p, status, options, rusage); 863 return (0); 864 } 865 if ((p->p_flag & P_STOPPED_SIG) && 866 (p->p_suspcount == p->p_numthreads) && 867 (p->p_flag & P_WAITED) == 0 && 868 (p->p_flag & P_TRACED || options & WUNTRACED)) { 869 PROC_SUNLOCK(p); 870 p->p_flag |= P_WAITED; 871 sx_xunlock(&proctree_lock); 872 td->td_retval[0] = p->p_pid; 873 if (status) 874 *status = W_STOPCODE(p->p_xstat); 875 876 PROC_LOCK(q); 877 sigqueue_take(p->p_ksi); 878 PROC_UNLOCK(q); 879 PROC_UNLOCK(p); 880 881 return (0); 882 } 883 PROC_SUNLOCK(p); 884 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) { 885 sx_xunlock(&proctree_lock); 886 td->td_retval[0] = p->p_pid; 887 p->p_flag &= ~P_CONTINUED; 888 889 PROC_LOCK(q); 890 sigqueue_take(p->p_ksi); 891 PROC_UNLOCK(q); 892 PROC_UNLOCK(p); 893 894 if (status) 895 *status = SIGCONT; 896 return (0); 897 } 898 PROC_UNLOCK(p); 899 } 900 if (nfound == 0) { 901 sx_xunlock(&proctree_lock); 902 if (td->td_proc->p_dbg_child) 903 return (0); 904 else 905 return (ECHILD); 906 } 907 if (options & WNOHANG) { 908 sx_xunlock(&proctree_lock); 909 td->td_retval[0] = 0; 910 return (0); 911 } 912 PROC_LOCK(q); 913 sx_xunlock(&proctree_lock); 914 if (q->p_flag & P_STATCHILD) { 915 q->p_flag &= ~P_STATCHILD; 916 error = 0; 917 } else 918 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 919 PROC_UNLOCK(q); 920 if (error) 921 return (error); 922 goto loop; 923} 924 925/* 926 * Make process 'parent' the new parent of process 'child'. 927 * Must be called with an exclusive hold of proctree lock. 928 */ 929void 930proc_reparent(struct proc *child, struct proc *parent) 931{ 932#ifdef RACCT 933 int locked; 934#endif 935 936 sx_assert(&proctree_lock, SX_XLOCKED); 937 PROC_LOCK_ASSERT(child, MA_OWNED); 938 if (child->p_pptr == parent) 939 return; 940 941#ifdef RACCT 942 locked = PROC_LOCKED(parent); 943 if (!locked) 944 PROC_LOCK(parent); 945 racct_add_force(parent, RACCT_NPROC, 1); 946 if (!locked) 947 PROC_UNLOCK(parent); 948#endif 949 PROC_LOCK(child->p_pptr); 950 racct_sub(child->p_pptr, RACCT_NPROC, 1); 951 sigqueue_take(child->p_ksi); 952 PROC_UNLOCK(child->p_pptr); 953 LIST_REMOVE(child, p_sibling); 954 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 955 child->p_pptr = parent; 956} 957