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