kern_exit.c revision 275793
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/10/sys/kern/kern_exit.c 275793 2014-12-15 10:29:02Z kib $"); 39 40#include "opt_compat.h" 41#include "opt_kdtrace.h" 42#include "opt_ktrace.h" 43#include "opt_procdesc.h" 44 45#include <sys/param.h> 46#include <sys/systm.h> 47#include <sys/sysproto.h> 48#include <sys/capability.h> 49#include <sys/eventhandler.h> 50#include <sys/kernel.h> 51#include <sys/malloc.h> 52#include <sys/lock.h> 53#include <sys/mutex.h> 54#include <sys/proc.h> 55#include <sys/procdesc.h> 56#include <sys/pioctl.h> 57#include <sys/jail.h> 58#include <sys/tty.h> 59#include <sys/wait.h> 60#include <sys/vmmeter.h> 61#include <sys/vnode.h> 62#include <sys/racct.h> 63#include <sys/resourcevar.h> 64#include <sys/sbuf.h> 65#include <sys/signalvar.h> 66#include <sys/sched.h> 67#include <sys/sx.h> 68#include <sys/syscallsubr.h> 69#include <sys/syslog.h> 70#include <sys/ptrace.h> 71#include <sys/acct.h> /* for acct_process() function prototype */ 72#include <sys/filedesc.h> 73#include <sys/sdt.h> 74#include <sys/shm.h> 75#include <sys/sem.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> 93dtrace_execexit_func_t dtrace_fasttrap_exit; 94#endif 95 96SDT_PROVIDER_DECLARE(proc); 97SDT_PROBE_DEFINE1(proc, kernel, , exit, "int"); 98 99/* Hook for NFS teardown procedure. */ 100void (*nlminfo_release_p)(struct proc *p); 101 102struct proc * 103proc_realparent(struct proc *child) 104{ 105 struct proc *p, *parent; 106 107 sx_assert(&proctree_lock, SX_LOCKED); 108 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 109 if (child->p_oppid == 0 || 110 child->p_pptr->p_pid == child->p_oppid) 111 parent = child->p_pptr; 112 else 113 parent = initproc; 114 return (parent); 115 } 116 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 117 /* Cannot use LIST_PREV(), since the list head is not known. */ 118 p = __containerof(p->p_orphan.le_prev, struct proc, 119 p_orphan.le_next); 120 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 121 ("missing P_ORPHAN %p", p)); 122 } 123 parent = __containerof(p->p_orphan.le_prev, struct proc, 124 p_orphans.lh_first); 125 return (parent); 126} 127 128static void 129clear_orphan(struct proc *p) 130{ 131 struct proc *p1; 132 133 sx_assert(&proctree_lock, SA_XLOCKED); 134 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 135 return; 136 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 137 p1 = LIST_NEXT(p, p_orphan); 138 if (p1 != NULL) 139 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 140 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 141 } 142 LIST_REMOVE(p, p_orphan); 143 p->p_treeflag &= ~P_TREE_ORPHANED; 144} 145 146/* 147 * exit -- death of process. 148 */ 149void 150sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 151{ 152 153 exit1(td, W_EXITCODE(uap->rval, 0)); 154 /* NOTREACHED */ 155} 156 157/* 158 * Exit: deallocate address space and other resources, change proc state to 159 * zombie, and unlink proc from allproc and parent's lists. Save exit status 160 * and rusage for wait(). Check for child processes and orphan them. 161 */ 162void 163exit1(struct thread *td, int rv) 164{ 165 struct proc *p, *nq, *q; 166 struct vnode *ttyvp = NULL; 167 168 mtx_assert(&Giant, MA_NOTOWNED); 169 170 p = td->td_proc; 171 /* 172 * XXX in case we're rebooting we just let init die in order to 173 * work around an unsolved stack overflow seen very late during 174 * shutdown on sparc64 when the gmirror worker process exists. 175 */ 176 if (p == initproc && rebooting == 0) { 177 printf("init died (signal %d, exit %d)\n", 178 WTERMSIG(rv), WEXITSTATUS(rv)); 179 panic("Going nowhere without my init!"); 180 } 181 182 /* 183 * MUST abort all other threads before proceeding past here. 184 */ 185 PROC_LOCK(p); 186 /* 187 * First check if some other thread or external request got 188 * here before us. If so, act appropriately: exit or suspend. 189 * We must ensure that stop requests are handled before we set 190 * P_WEXIT. 191 */ 192 thread_suspend_check(0); 193 while (p->p_flag & P_HADTHREADS) { 194 /* 195 * Kill off the other threads. This requires 196 * some co-operation from other parts of the kernel 197 * so it may not be instantaneous. With this state set 198 * any thread entering the kernel from userspace will 199 * thread_exit() in trap(). Any thread attempting to 200 * sleep will return immediately with EINTR or EWOULDBLOCK 201 * which will hopefully force them to back out to userland 202 * freeing resources as they go. Any thread attempting 203 * to return to userland will thread_exit() from userret(). 204 * thread_exit() will unsuspend us when the last of the 205 * other threads exits. 206 * If there is already a thread singler after resumption, 207 * calling thread_single will fail; in that case, we just 208 * re-check all suspension request, the thread should 209 * either be suspended there or exit. 210 */ 211 if (!thread_single(SINGLE_EXIT)) 212 /* 213 * All other activity in this process is now 214 * stopped. Threading support has been turned 215 * off. 216 */ 217 break; 218 /* 219 * Recheck for new stop or suspend requests which 220 * might appear while process lock was dropped in 221 * thread_single(). 222 */ 223 thread_suspend_check(0); 224 } 225 KASSERT(p->p_numthreads == 1, 226 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 227 racct_sub(p, RACCT_NTHR, 1); 228 /* 229 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 230 * on our vmspace, so we should block below until they have 231 * released their reference to us. Note that if they have 232 * requested S_EXIT stops we will block here until they ack 233 * via PIOCCONT. 234 */ 235 _STOPEVENT(p, S_EXIT, rv); 236 237 /* 238 * Ignore any pending request to stop due to a stop signal. 239 * Once P_WEXIT is set, future requests will be ignored as 240 * well. 241 */ 242 p->p_flag &= ~P_STOPPED_SIG; 243 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 244 245 /* 246 * Note that we are exiting and do another wakeup of anyone in 247 * PIOCWAIT in case they aren't listening for S_EXIT stops or 248 * decided to wait again after we told them we are exiting. 249 */ 250 p->p_flag |= P_WEXIT; 251 wakeup(&p->p_stype); 252 253 /* 254 * Wait for any processes that have a hold on our vmspace to 255 * release their reference. 256 */ 257 while (p->p_lock > 0) 258 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 259 260 p->p_xstat = rv; /* Let event handler change exit status */ 261 PROC_UNLOCK(p); 262 /* Drain the limit callout while we don't have the proc locked */ 263 callout_drain(&p->p_limco); 264 265#ifdef AUDIT 266 /* 267 * The Sun BSM exit token contains two components: an exit status as 268 * passed to exit(), and a return value to indicate what sort of exit 269 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 270 * what the return value is. 271 */ 272 AUDIT_ARG_EXIT(WEXITSTATUS(rv), 0); 273 AUDIT_SYSCALL_EXIT(0, td); 274#endif 275 276 /* Are we a task leader? */ 277 if (p == p->p_leader) { 278 mtx_lock(&ppeers_lock); 279 q = p->p_peers; 280 while (q != NULL) { 281 PROC_LOCK(q); 282 kern_psignal(q, SIGKILL); 283 PROC_UNLOCK(q); 284 q = q->p_peers; 285 } 286 while (p->p_peers != NULL) 287 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 288 mtx_unlock(&ppeers_lock); 289 } 290 291 /* 292 * Check if any loadable modules need anything done at process exit. 293 * E.g. SYSV IPC stuff 294 * XXX what if one of these generates an error? 295 */ 296 EVENTHANDLER_INVOKE(process_exit, p); 297 298 /* 299 * If parent is waiting for us to exit or exec, 300 * P_PPWAIT is set; we will wakeup the parent below. 301 */ 302 PROC_LOCK(p); 303 rv = p->p_xstat; /* Event handler could change exit status */ 304 stopprofclock(p); 305 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 306 307 /* 308 * Stop the real interval timer. If the handler is currently 309 * executing, prevent it from rearming itself and let it finish. 310 */ 311 if (timevalisset(&p->p_realtimer.it_value) && 312 callout_stop(&p->p_itcallout) == 0) { 313 timevalclear(&p->p_realtimer.it_interval); 314 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 315 KASSERT(!timevalisset(&p->p_realtimer.it_value), 316 ("realtime timer is still armed")); 317 } 318 PROC_UNLOCK(p); 319 320 /* 321 * Reset any sigio structures pointing to us as a result of 322 * F_SETOWN with our pid. 323 */ 324 funsetownlst(&p->p_sigiolst); 325 326 /* 327 * If this process has an nlminfo data area (for lockd), release it 328 */ 329 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 330 (*nlminfo_release_p)(p); 331 332 /* 333 * Close open files and release open-file table. 334 * This may block! 335 */ 336 fdescfree(td); 337 338 /* 339 * If this thread tickled GEOM, we need to wait for the giggling to 340 * stop before we return to userland 341 */ 342 if (td->td_pflags & TDP_GEOM) 343 g_waitidle(); 344 345 /* 346 * Remove ourself from our leader's peer list and wake our leader. 347 */ 348 mtx_lock(&ppeers_lock); 349 if (p->p_leader->p_peers) { 350 q = p->p_leader; 351 while (q->p_peers != p) 352 q = q->p_peers; 353 q->p_peers = p->p_peers; 354 wakeup(p->p_leader); 355 } 356 mtx_unlock(&ppeers_lock); 357 358 vmspace_exit(td); 359 360 sx_xlock(&proctree_lock); 361 if (SESS_LEADER(p)) { 362 struct session *sp = p->p_session; 363 struct tty *tp; 364 365 /* 366 * s_ttyp is not zero'd; we use this to indicate that 367 * the session once had a controlling terminal. (for 368 * logging and informational purposes) 369 */ 370 SESS_LOCK(sp); 371 ttyvp = sp->s_ttyvp; 372 tp = sp->s_ttyp; 373 sp->s_ttyvp = NULL; 374 sp->s_ttydp = NULL; 375 sp->s_leader = NULL; 376 SESS_UNLOCK(sp); 377 378 /* 379 * Signal foreground pgrp and revoke access to 380 * controlling terminal if it has not been revoked 381 * already. 382 * 383 * Because the TTY may have been revoked in the mean 384 * time and could already have a new session associated 385 * with it, make sure we don't send a SIGHUP to a 386 * foreground process group that does not belong to this 387 * session. 388 */ 389 390 if (tp != NULL) { 391 tty_lock(tp); 392 if (tp->t_session == sp) 393 tty_signal_pgrp(tp, SIGHUP); 394 tty_unlock(tp); 395 } 396 397 if (ttyvp != NULL) { 398 sx_xunlock(&proctree_lock); 399 if (vn_lock(ttyvp, LK_EXCLUSIVE) == 0) { 400 VOP_REVOKE(ttyvp, REVOKEALL); 401 VOP_UNLOCK(ttyvp, 0); 402 } 403 sx_xlock(&proctree_lock); 404 } 405 } 406 fixjobc(p, p->p_pgrp, 0); 407 sx_xunlock(&proctree_lock); 408 (void)acct_process(td); 409 410 /* Release the TTY now we've unlocked everything. */ 411 if (ttyvp != NULL) 412 vrele(ttyvp); 413#ifdef KTRACE 414 ktrprocexit(td); 415#endif 416 /* 417 * Release reference to text vnode 418 */ 419 if (p->p_textvp != NULL) { 420 vrele(p->p_textvp); 421 p->p_textvp = NULL; 422 } 423 424 /* 425 * Release our limits structure. 426 */ 427 lim_free(p->p_limit); 428 p->p_limit = NULL; 429 430 tidhash_remove(td); 431 432 /* 433 * Remove proc from allproc queue and pidhash chain. 434 * Place onto zombproc. Unlink from parent's child list. 435 */ 436 sx_xlock(&allproc_lock); 437 LIST_REMOVE(p, p_list); 438 LIST_INSERT_HEAD(&zombproc, p, p_list); 439 LIST_REMOVE(p, p_hash); 440 sx_xunlock(&allproc_lock); 441 442 /* 443 * Call machine-dependent code to release any 444 * machine-dependent resources other than the address space. 445 * The address space is released by "vmspace_exitfree(p)" in 446 * vm_waitproc(). 447 */ 448 cpu_exit(td); 449 450 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 451 452 /* 453 * Reparent all of our children to init. 454 */ 455 sx_xlock(&proctree_lock); 456 q = LIST_FIRST(&p->p_children); 457 if (q != NULL) /* only need this if any child is S_ZOMB */ 458 wakeup(initproc); 459 for (; q != NULL; q = nq) { 460 nq = LIST_NEXT(q, p_sibling); 461 PROC_LOCK(q); 462 proc_reparent(q, initproc); 463 q->p_sigparent = SIGCHLD; 464 /* 465 * Traced processes are killed 466 * since their existence means someone is screwing up. 467 */ 468 if (q->p_flag & P_TRACED) { 469 struct thread *temp; 470 471 /* 472 * Since q was found on our children list, the 473 * proc_reparent() call moved q to the orphan 474 * list due to present P_TRACED flag. Clear 475 * orphan link for q now while q is locked. 476 */ 477 clear_orphan(q); 478 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 479 FOREACH_THREAD_IN_PROC(q, temp) 480 temp->td_dbgflags &= ~TDB_SUSPEND; 481 kern_psignal(q, SIGKILL); 482 } 483 PROC_UNLOCK(q); 484 } 485 486 /* 487 * Also get rid of our orphans. 488 */ 489 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 490 PROC_LOCK(q); 491 clear_orphan(q); 492 PROC_UNLOCK(q); 493 } 494 495 /* Save exit status. */ 496 PROC_LOCK(p); 497 p->p_xthread = td; 498 499 /* Tell the prison that we are gone. */ 500 prison_proc_free(p->p_ucred->cr_prison); 501 502#ifdef KDTRACE_HOOKS 503 /* 504 * Tell the DTrace fasttrap provider about the exit if it 505 * has declared an interest. 506 */ 507 if (dtrace_fasttrap_exit) 508 dtrace_fasttrap_exit(p); 509#endif 510 511 /* 512 * Notify interested parties of our demise. 513 */ 514 KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); 515 516#ifdef KDTRACE_HOOKS 517 int reason = CLD_EXITED; 518 if (WCOREDUMP(rv)) 519 reason = CLD_DUMPED; 520 else if (WIFSIGNALED(rv)) 521 reason = CLD_KILLED; 522 SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); 523#endif 524 525 /* 526 * Just delete all entries in the p_klist. At this point we won't 527 * report any more events, and there are nasty race conditions that 528 * can beat us if we don't. 529 */ 530 knlist_clear(&p->p_klist, 1); 531 532 /* 533 * If this is a process with a descriptor, we may not need to deliver 534 * a signal to the parent. proctree_lock is held over 535 * procdesc_exit() to serialize concurrent calls to close() and 536 * exit(). 537 */ 538#ifdef PROCDESC 539 if (p->p_procdesc == NULL || procdesc_exit(p)) { 540#endif 541 /* 542 * Notify parent that we're gone. If parent has the 543 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 544 * notify process 1 instead (and hope it will handle this 545 * situation). 546 */ 547 PROC_LOCK(p->p_pptr); 548 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 549 if (p->p_pptr->p_sigacts->ps_flag & 550 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 551 struct proc *pp; 552 553 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 554 pp = p->p_pptr; 555 PROC_UNLOCK(pp); 556 proc_reparent(p, initproc); 557 p->p_sigparent = SIGCHLD; 558 PROC_LOCK(p->p_pptr); 559 560 /* 561 * Notify parent, so in case he was wait(2)ing or 562 * executing waitpid(2) with our pid, he will 563 * continue. 564 */ 565 wakeup(pp); 566 } else 567 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 568 569 if (p->p_pptr == initproc) 570 kern_psignal(p->p_pptr, SIGCHLD); 571 else if (p->p_sigparent != 0) { 572 if (p->p_sigparent == SIGCHLD) 573 childproc_exited(p); 574 else /* LINUX thread */ 575 kern_psignal(p->p_pptr, p->p_sigparent); 576 } 577#ifdef PROCDESC 578 } else 579 PROC_LOCK(p->p_pptr); 580#endif 581 sx_xunlock(&proctree_lock); 582 583 /* 584 * The state PRS_ZOMBIE prevents other proesses from sending 585 * signal to the process, to avoid memory leak, we free memory 586 * for signal queue at the time when the state is set. 587 */ 588 sigqueue_flush(&p->p_sigqueue); 589 sigqueue_flush(&td->td_sigqueue); 590 591 /* 592 * We have to wait until after acquiring all locks before 593 * changing p_state. We need to avoid all possible context 594 * switches (including ones from blocking on a mutex) while 595 * marked as a zombie. We also have to set the zombie state 596 * before we release the parent process' proc lock to avoid 597 * a lost wakeup. So, we first call wakeup, then we grab the 598 * sched lock, update the state, and release the parent process' 599 * proc lock. 600 */ 601 wakeup(p->p_pptr); 602 cv_broadcast(&p->p_pwait); 603 sched_exit(p->p_pptr, td); 604 PROC_SLOCK(p); 605 p->p_state = PRS_ZOMBIE; 606 PROC_UNLOCK(p->p_pptr); 607 608 /* 609 * Hopefully no one will try to deliver a signal to the process this 610 * late in the game. 611 */ 612 knlist_destroy(&p->p_klist); 613 614 /* 615 * Save our children's rusage information in our exit rusage. 616 */ 617 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 618 619 /* 620 * Make sure the scheduler takes this thread out of its tables etc. 621 * This will also release this thread's reference to the ucred. 622 * Other thread parts to release include pcb bits and such. 623 */ 624 thread_exit(); 625} 626 627 628#ifndef _SYS_SYSPROTO_H_ 629struct abort2_args { 630 char *why; 631 int nargs; 632 void **args; 633}; 634#endif 635 636int 637sys_abort2(struct thread *td, struct abort2_args *uap) 638{ 639 struct proc *p = td->td_proc; 640 struct sbuf *sb; 641 void *uargs[16]; 642 int error, i, sig; 643 644 /* 645 * Do it right now so we can log either proper call of abort2(), or 646 * note, that invalid argument was passed. 512 is big enough to 647 * handle 16 arguments' descriptions with additional comments. 648 */ 649 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 650 sbuf_clear(sb); 651 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 652 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 653 /* 654 * Since we can't return from abort2(), send SIGKILL in cases, where 655 * abort2() was called improperly 656 */ 657 sig = SIGKILL; 658 /* Prevent from DoSes from user-space. */ 659 if (uap->nargs < 0 || uap->nargs > 16) 660 goto out; 661 if (uap->nargs > 0) { 662 if (uap->args == NULL) 663 goto out; 664 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 665 if (error != 0) 666 goto out; 667 } 668 /* 669 * Limit size of 'reason' string to 128. Will fit even when 670 * maximal number of arguments was chosen to be logged. 671 */ 672 if (uap->why != NULL) { 673 error = sbuf_copyin(sb, uap->why, 128); 674 if (error < 0) 675 goto out; 676 } else { 677 sbuf_printf(sb, "(null)"); 678 } 679 if (uap->nargs > 0) { 680 sbuf_printf(sb, "("); 681 for (i = 0;i < uap->nargs; i++) 682 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 683 sbuf_printf(sb, ")"); 684 } 685 /* 686 * Final stage: arguments were proper, string has been 687 * successfully copied from userspace, and copying pointers 688 * from user-space succeed. 689 */ 690 sig = SIGABRT; 691out: 692 if (sig == SIGKILL) { 693 sbuf_trim(sb); 694 sbuf_printf(sb, " (Reason text inaccessible)"); 695 } 696 sbuf_cat(sb, "\n"); 697 sbuf_finish(sb); 698 log(LOG_INFO, "%s", sbuf_data(sb)); 699 sbuf_delete(sb); 700 exit1(td, W_EXITCODE(0, sig)); 701 return (0); 702} 703 704 705#ifdef COMPAT_43 706/* 707 * The dirty work is handled by kern_wait(). 708 */ 709int 710owait(struct thread *td, struct owait_args *uap __unused) 711{ 712 int error, status; 713 714 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 715 if (error == 0) 716 td->td_retval[1] = status; 717 return (error); 718} 719#endif /* COMPAT_43 */ 720 721/* 722 * The dirty work is handled by kern_wait(). 723 */ 724int 725sys_wait4(struct thread *td, struct wait4_args *uap) 726{ 727 struct rusage ru, *rup; 728 int error, status; 729 730 if (uap->rusage != NULL) 731 rup = &ru; 732 else 733 rup = NULL; 734 error = kern_wait(td, uap->pid, &status, uap->options, rup); 735 if (uap->status != NULL && error == 0) 736 error = copyout(&status, uap->status, sizeof(status)); 737 if (uap->rusage != NULL && error == 0) 738 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 739 return (error); 740} 741 742int 743sys_wait6(struct thread *td, struct wait6_args *uap) 744{ 745 struct __wrusage wru, *wrup; 746 siginfo_t si, *sip; 747 idtype_t idtype; 748 id_t id; 749 int error, status; 750 751 idtype = uap->idtype; 752 id = uap->id; 753 754 if (uap->wrusage != NULL) 755 wrup = &wru; 756 else 757 wrup = NULL; 758 759 if (uap->info != NULL) { 760 sip = &si; 761 bzero(sip, sizeof(*sip)); 762 } else 763 sip = NULL; 764 765 /* 766 * We expect all callers of wait6() to know about WEXITED and 767 * WTRAPPED. 768 */ 769 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 770 771 if (uap->status != NULL && error == 0) 772 error = copyout(&status, uap->status, sizeof(status)); 773 if (uap->wrusage != NULL && error == 0) 774 error = copyout(&wru, uap->wrusage, sizeof(wru)); 775 if (uap->info != NULL && error == 0) 776 error = copyout(&si, uap->info, sizeof(si)); 777 return (error); 778} 779 780/* 781 * Reap the remains of a zombie process and optionally return status and 782 * rusage. Asserts and will release both the proctree_lock and the process 783 * lock as part of its work. 784 */ 785void 786proc_reap(struct thread *td, struct proc *p, int *status, int options) 787{ 788 struct proc *q, *t; 789 790 sx_assert(&proctree_lock, SA_XLOCKED); 791 PROC_LOCK_ASSERT(p, MA_OWNED); 792 PROC_SLOCK_ASSERT(p, MA_OWNED); 793 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 794 795 q = td->td_proc; 796 797 PROC_SUNLOCK(p); 798 td->td_retval[0] = p->p_pid; 799 if (status) 800 *status = p->p_xstat; /* convert to int */ 801 if (options & WNOWAIT) { 802 /* 803 * Only poll, returning the status. Caller does not wish to 804 * release the proc struct just yet. 805 */ 806 PROC_UNLOCK(p); 807 sx_xunlock(&proctree_lock); 808 return; 809 } 810 811 PROC_LOCK(q); 812 sigqueue_take(p->p_ksi); 813 PROC_UNLOCK(q); 814 PROC_UNLOCK(p); 815 816 /* 817 * If we got the child via a ptrace 'attach', we need to give it back 818 * to the old parent. 819 */ 820 if (p->p_oppid != 0) { 821 t = proc_realparent(p); 822 PROC_LOCK(t); 823 PROC_LOCK(p); 824 proc_reparent(p, t); 825 p->p_oppid = 0; 826 PROC_UNLOCK(p); 827 pksignal(t, SIGCHLD, p->p_ksi); 828 wakeup(t); 829 cv_broadcast(&p->p_pwait); 830 PROC_UNLOCK(t); 831 sx_xunlock(&proctree_lock); 832 return; 833 } 834 835 /* 836 * Remove other references to this process to ensure we have an 837 * exclusive reference. 838 */ 839 sx_xlock(&allproc_lock); 840 LIST_REMOVE(p, p_list); /* off zombproc */ 841 sx_xunlock(&allproc_lock); 842 LIST_REMOVE(p, p_sibling); 843 PROC_LOCK(p); 844 clear_orphan(p); 845 PROC_UNLOCK(p); 846 leavepgrp(p); 847#ifdef PROCDESC 848 if (p->p_procdesc != NULL) 849 procdesc_reap(p); 850#endif 851 sx_xunlock(&proctree_lock); 852 853 /* 854 * As a side effect of this lock, we know that all other writes to 855 * this proc are visible now, so no more locking is needed for p. 856 */ 857 PROC_LOCK(p); 858 p->p_xstat = 0; /* XXX: why? */ 859 PROC_UNLOCK(p); 860 PROC_LOCK(q); 861 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 862 PROC_UNLOCK(q); 863 864 /* 865 * Decrement the count of procs running with this uid. 866 */ 867 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 868 869 /* 870 * Destroy resource accounting information associated with the process. 871 */ 872#ifdef RACCT 873 PROC_LOCK(p); 874 racct_sub(p, RACCT_NPROC, 1); 875 PROC_UNLOCK(p); 876#endif 877 racct_proc_exit(p); 878 879 /* 880 * Free credentials, arguments, and sigacts. 881 */ 882 crfree(p->p_ucred); 883 p->p_ucred = NULL; 884 pargs_drop(p->p_args); 885 p->p_args = NULL; 886 sigacts_free(p->p_sigacts); 887 p->p_sigacts = NULL; 888 889 /* 890 * Do any thread-system specific cleanups. 891 */ 892 thread_wait(p); 893 894 /* 895 * Give vm and machine-dependent layer a chance to free anything that 896 * cpu_exit couldn't release while still running in process context. 897 */ 898 vm_waitproc(p); 899#ifdef MAC 900 mac_proc_destroy(p); 901#endif 902 KASSERT(FIRST_THREAD_IN_PROC(p), 903 ("proc_reap: no residual thread!")); 904 uma_zfree(proc_zone, p); 905 sx_xlock(&allproc_lock); 906 nprocs--; 907 sx_xunlock(&allproc_lock); 908} 909 910static int 911proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 912 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo) 913{ 914 struct proc *q; 915 struct rusage *rup; 916 917 sx_assert(&proctree_lock, SA_XLOCKED); 918 919 q = td->td_proc; 920 PROC_LOCK(p); 921 922 switch (idtype) { 923 case P_ALL: 924 break; 925 case P_PID: 926 if (p->p_pid != (pid_t)id) { 927 PROC_UNLOCK(p); 928 return (0); 929 } 930 break; 931 case P_PGID: 932 if (p->p_pgid != (pid_t)id) { 933 PROC_UNLOCK(p); 934 return (0); 935 } 936 break; 937 case P_SID: 938 if (p->p_session->s_sid != (pid_t)id) { 939 PROC_UNLOCK(p); 940 return (0); 941 } 942 break; 943 case P_UID: 944 if (p->p_ucred->cr_uid != (uid_t)id) { 945 PROC_UNLOCK(p); 946 return (0); 947 } 948 break; 949 case P_GID: 950 if (p->p_ucred->cr_gid != (gid_t)id) { 951 PROC_UNLOCK(p); 952 return (0); 953 } 954 break; 955 case P_JAILID: 956 if (p->p_ucred->cr_prison->pr_id != (int)id) { 957 PROC_UNLOCK(p); 958 return (0); 959 } 960 break; 961 /* 962 * It seems that the thread structures get zeroed out 963 * at process exit. This makes it impossible to 964 * support P_SETID, P_CID or P_CPUID. 965 */ 966 default: 967 PROC_UNLOCK(p); 968 return (0); 969 } 970 971 if (p_canwait(td, p)) { 972 PROC_UNLOCK(p); 973 return (0); 974 } 975 976 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 977 PROC_UNLOCK(p); 978 return (0); 979 } 980 981 /* 982 * This special case handles a kthread spawned by linux_clone 983 * (see linux_misc.c). The linux_wait4 and linux_waitpid 984 * functions need to be able to distinguish between waiting 985 * on a process and waiting on a thread. It is a thread if 986 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 987 * signifies we want to wait for threads and not processes. 988 */ 989 if ((p->p_sigparent != SIGCHLD) ^ 990 ((options & WLINUXCLONE) != 0)) { 991 PROC_UNLOCK(p); 992 return (0); 993 } 994 995 PROC_SLOCK(p); 996 997 if (siginfo != NULL) { 998 bzero(siginfo, sizeof(*siginfo)); 999 siginfo->si_errno = 0; 1000 1001 /* 1002 * SUSv4 requires that the si_signo value is always 1003 * SIGCHLD. Obey it despite the rfork(2) interface 1004 * allows to request other signal for child exit 1005 * notification. 1006 */ 1007 siginfo->si_signo = SIGCHLD; 1008 1009 /* 1010 * This is still a rough estimate. We will fix the 1011 * cases TRAPPED, STOPPED, and CONTINUED later. 1012 */ 1013 if (WCOREDUMP(p->p_xstat)) { 1014 siginfo->si_code = CLD_DUMPED; 1015 siginfo->si_status = WTERMSIG(p->p_xstat); 1016 } else if (WIFSIGNALED(p->p_xstat)) { 1017 siginfo->si_code = CLD_KILLED; 1018 siginfo->si_status = WTERMSIG(p->p_xstat); 1019 } else { 1020 siginfo->si_code = CLD_EXITED; 1021 siginfo->si_status = WEXITSTATUS(p->p_xstat); 1022 } 1023 1024 siginfo->si_pid = p->p_pid; 1025 siginfo->si_uid = p->p_ucred->cr_uid; 1026 1027 /* 1028 * The si_addr field would be useful additional 1029 * detail, but apparently the PC value may be lost 1030 * when we reach this point. bzero() above sets 1031 * siginfo->si_addr to NULL. 1032 */ 1033 } 1034 1035 /* 1036 * There should be no reason to limit resources usage info to 1037 * exited processes only. A snapshot about any resources used 1038 * by a stopped process may be exactly what is needed. 1039 */ 1040 if (wrusage != NULL) { 1041 rup = &wrusage->wru_self; 1042 *rup = p->p_ru; 1043 calcru(p, &rup->ru_utime, &rup->ru_stime); 1044 1045 rup = &wrusage->wru_children; 1046 *rup = p->p_stats->p_cru; 1047 calccru(p, &rup->ru_utime, &rup->ru_stime); 1048 } 1049 1050 if (p->p_state == PRS_ZOMBIE) { 1051 proc_reap(td, p, status, options); 1052 return (-1); 1053 } 1054 PROC_SUNLOCK(p); 1055 PROC_UNLOCK(p); 1056 return (1); 1057} 1058 1059int 1060kern_wait(struct thread *td, pid_t pid, int *status, int options, 1061 struct rusage *rusage) 1062{ 1063 struct __wrusage wru, *wrup; 1064 idtype_t idtype; 1065 id_t id; 1066 int ret; 1067 1068 /* 1069 * Translate the special pid values into the (idtype, pid) 1070 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1071 * kern_wait6() on its own. 1072 */ 1073 if (pid == WAIT_ANY) { 1074 idtype = P_ALL; 1075 id = 0; 1076 } else if (pid < 0) { 1077 idtype = P_PGID; 1078 id = (id_t)-pid; 1079 } else { 1080 idtype = P_PID; 1081 id = (id_t)pid; 1082 } 1083 1084 if (rusage != NULL) 1085 wrup = &wru; 1086 else 1087 wrup = NULL; 1088 1089 /* 1090 * For backward compatibility we implicitly add flags WEXITED 1091 * and WTRAPPED here. 1092 */ 1093 options |= WEXITED | WTRAPPED; 1094 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1095 if (rusage != NULL) 1096 *rusage = wru.wru_self; 1097 return (ret); 1098} 1099 1100int 1101kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1102 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1103{ 1104 struct proc *p, *q; 1105 int error, nfound, ret; 1106 1107 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1108 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1109 AUDIT_ARG_VALUE(options); 1110 1111 q = td->td_proc; 1112 1113 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1114 PROC_LOCK(q); 1115 id = (id_t)q->p_pgid; 1116 PROC_UNLOCK(q); 1117 idtype = P_PGID; 1118 } 1119 1120 /* If we don't know the option, just return. */ 1121 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1122 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1123 return (EINVAL); 1124 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1125 /* 1126 * We will be unable to find any matching processes, 1127 * because there are no known events to look for. 1128 * Prefer to return error instead of blocking 1129 * indefinitely. 1130 */ 1131 return (EINVAL); 1132 } 1133 1134loop: 1135 if (q->p_flag & P_STATCHILD) { 1136 PROC_LOCK(q); 1137 q->p_flag &= ~P_STATCHILD; 1138 PROC_UNLOCK(q); 1139 } 1140 nfound = 0; 1141 sx_xlock(&proctree_lock); 1142 LIST_FOREACH(p, &q->p_children, p_sibling) { 1143 ret = proc_to_reap(td, p, idtype, id, status, options, 1144 wrusage, siginfo); 1145 if (ret == 0) 1146 continue; 1147 else if (ret == 1) 1148 nfound++; 1149 else 1150 return (0); 1151 1152 PROC_LOCK(p); 1153 PROC_SLOCK(p); 1154 1155 if ((options & WTRAPPED) != 0 && 1156 (p->p_flag & P_TRACED) != 0 && 1157 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1158 (p->p_suspcount == p->p_numthreads) && 1159 ((p->p_flag & P_WAITED) == 0)) { 1160 PROC_SUNLOCK(p); 1161 if ((options & WNOWAIT) == 0) 1162 p->p_flag |= P_WAITED; 1163 sx_xunlock(&proctree_lock); 1164 td->td_retval[0] = p->p_pid; 1165 1166 if (status != NULL) 1167 *status = W_STOPCODE(p->p_xstat); 1168 if (siginfo != NULL) { 1169 siginfo->si_status = p->p_xstat; 1170 siginfo->si_code = CLD_TRAPPED; 1171 } 1172 if ((options & WNOWAIT) == 0) { 1173 PROC_LOCK(q); 1174 sigqueue_take(p->p_ksi); 1175 PROC_UNLOCK(q); 1176 } 1177 1178 PROC_UNLOCK(p); 1179 return (0); 1180 } 1181 if ((options & WUNTRACED) != 0 && 1182 (p->p_flag & P_STOPPED_SIG) != 0 && 1183 (p->p_suspcount == p->p_numthreads) && 1184 ((p->p_flag & P_WAITED) == 0)) { 1185 PROC_SUNLOCK(p); 1186 if ((options & WNOWAIT) == 0) 1187 p->p_flag |= P_WAITED; 1188 sx_xunlock(&proctree_lock); 1189 td->td_retval[0] = p->p_pid; 1190 1191 if (status != NULL) 1192 *status = W_STOPCODE(p->p_xstat); 1193 if (siginfo != NULL) { 1194 siginfo->si_status = p->p_xstat; 1195 siginfo->si_code = CLD_STOPPED; 1196 } 1197 if ((options & WNOWAIT) == 0) { 1198 PROC_LOCK(q); 1199 sigqueue_take(p->p_ksi); 1200 PROC_UNLOCK(q); 1201 } 1202 1203 PROC_UNLOCK(p); 1204 return (0); 1205 } 1206 PROC_SUNLOCK(p); 1207 if ((options & WCONTINUED) != 0 && 1208 (p->p_flag & P_CONTINUED) != 0) { 1209 sx_xunlock(&proctree_lock); 1210 td->td_retval[0] = p->p_pid; 1211 if ((options & WNOWAIT) == 0) { 1212 p->p_flag &= ~P_CONTINUED; 1213 PROC_LOCK(q); 1214 sigqueue_take(p->p_ksi); 1215 PROC_UNLOCK(q); 1216 } 1217 PROC_UNLOCK(p); 1218 1219 if (status != NULL) 1220 *status = SIGCONT; 1221 if (siginfo != NULL) { 1222 siginfo->si_status = SIGCONT; 1223 siginfo->si_code = CLD_CONTINUED; 1224 } 1225 return (0); 1226 } 1227 PROC_UNLOCK(p); 1228 } 1229 1230 /* 1231 * Look in the orphans list too, to allow the parent to 1232 * collect it's child exit status even if child is being 1233 * debugged. 1234 * 1235 * Debugger detaches from the parent upon successful 1236 * switch-over from parent to child. At this point due to 1237 * re-parenting the parent loses the child to debugger and a 1238 * wait4(2) call would report that it has no children to wait 1239 * for. By maintaining a list of orphans we allow the parent 1240 * to successfully wait until the child becomes a zombie. 1241 */ 1242 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1243 ret = proc_to_reap(td, p, idtype, id, status, options, 1244 wrusage, siginfo); 1245 if (ret == 0) 1246 continue; 1247 else if (ret == 1) 1248 nfound++; 1249 else 1250 return (0); 1251 } 1252 if (nfound == 0) { 1253 sx_xunlock(&proctree_lock); 1254 return (ECHILD); 1255 } 1256 if (options & WNOHANG) { 1257 sx_xunlock(&proctree_lock); 1258 td->td_retval[0] = 0; 1259 return (0); 1260 } 1261 PROC_LOCK(q); 1262 sx_xunlock(&proctree_lock); 1263 if (q->p_flag & P_STATCHILD) { 1264 q->p_flag &= ~P_STATCHILD; 1265 error = 0; 1266 } else 1267 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1268 PROC_UNLOCK(q); 1269 if (error) 1270 return (error); 1271 goto loop; 1272} 1273 1274/* 1275 * Make process 'parent' the new parent of process 'child'. 1276 * Must be called with an exclusive hold of proctree lock. 1277 */ 1278void 1279proc_reparent(struct proc *child, struct proc *parent) 1280{ 1281 1282 sx_assert(&proctree_lock, SX_XLOCKED); 1283 PROC_LOCK_ASSERT(child, MA_OWNED); 1284 if (child->p_pptr == parent) 1285 return; 1286 1287 PROC_LOCK(child->p_pptr); 1288 sigqueue_take(child->p_ksi); 1289 PROC_UNLOCK(child->p_pptr); 1290 LIST_REMOVE(child, p_sibling); 1291 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1292 1293 clear_orphan(child); 1294 if (child->p_flag & P_TRACED) { 1295 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1296 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1297 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1298 p_orphan); 1299 } else { 1300 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1301 child, p_orphan); 1302 } 1303 child->p_treeflag |= P_TREE_ORPHANED; 1304 } 1305 1306 child->p_pptr = parent; 1307} 1308