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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 39 * $FreeBSD: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $ 40 * $DragonFly: src/sys/kern/kern_exit.c,v 1.91 2008/05/18 20:02:02 nth Exp $ 41 */ 42 43 #include "opt_compat.h" 44 #include "opt_ktrace.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/sysproto.h> 49 #include <sys/kernel.h> 50 #include <sys/malloc.h> 51 #include <sys/proc.h> 52 #include <sys/ktrace.h> 53 #include <sys/pioctl.h> 54 #include <sys/tty.h> 55 #include <sys/wait.h> 56 #include <sys/vnode.h> 57 #include <sys/resourcevar.h> 58 #include <sys/signalvar.h> 59 #include <sys/taskqueue.h> 60 #include <sys/ptrace.h> 61 #include <sys/acct.h> /* for acct_process() function prototype */ 62 #include <sys/filedesc.h> 63 #include <sys/shm.h> 64 #include <sys/sem.h> 65 #include <sys/aio.h> 66 #include <sys/jail.h> 67 #include <sys/kern_syscall.h> 68 #include <sys/upcall.h> 69 #include <sys/caps.h> 70 #include <sys/unistd.h> 71 #include <sys/eventhandler.h> 72 #include <sys/dsched.h> 73 74 #include <vm/vm.h> 75 #include <vm/vm_param.h> 76 #include <sys/lock.h> 77 #include <vm/pmap.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_extern.h> 80 #include <sys/user.h> 81 82 #include <sys/thread2.h> 83 #include <sys/sysref2.h> 84 #include <sys/mplock2.h> 85 86 static void reaplwps(void *context, int dummy); 87 static void reaplwp(struct lwp *lp); 88 static void killlwps(struct lwp *lp); 89 90 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 91 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status"); 92 93 static struct lwkt_token deadlwp_token = LWKT_TOKEN_INITIALIZER(deadlwp_token); 94 95 /* 96 * callout list for things to do at exit time 97 */ 98 struct exitlist { 99 exitlist_fn function; 100 TAILQ_ENTRY(exitlist) next; 101 }; 102 103 TAILQ_HEAD(exit_list_head, exitlist); 104 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 105 106 /* 107 * LWP reaper data 108 */ 109 struct task *deadlwp_task[MAXCPU]; 110 struct lwplist deadlwp_list[MAXCPU]; 111 112 /* 113 * exit -- 114 * Death of process. 115 * 116 * SYS_EXIT_ARGS(int rval) 117 */ 118 int 119 sys_exit(struct exit_args *uap) 120 { 121 exit1(W_EXITCODE(uap->rval, 0)); 122 /* NOTREACHED */ 123 } 124 125 /* 126 * Extended exit -- 127 * Death of a lwp or process with optional bells and whistles. 128 * 129 * MPALMOSTSAFE 130 */ 131 int 132 sys_extexit(struct extexit_args *uap) 133 { 134 struct proc *p = curproc; 135 int action, who; 136 int error; 137 138 action = EXTEXIT_ACTION(uap->how); 139 who = EXTEXIT_WHO(uap->how); 140 141 /* Check parameters before we might perform some action */ 142 switch (who) { 143 case EXTEXIT_PROC: 144 case EXTEXIT_LWP: 145 break; 146 default: 147 return (EINVAL); 148 } 149 150 switch (action) { 151 case EXTEXIT_SIMPLE: 152 break; 153 case EXTEXIT_SETINT: 154 error = copyout(&uap->status, uap->addr, sizeof(uap->status)); 155 if (error) 156 return (error); 157 break; 158 default: 159 return (EINVAL); 160 } 161 162 lwkt_gettoken(&p->p_token); 163 164 switch (who) { 165 case EXTEXIT_LWP: 166 /* 167 * Be sure only to perform a simple lwp exit if there is at 168 * least one more lwp in the proc, which will call exit1() 169 * later, otherwise the proc will be an UNDEAD and not even a 170 * SZOMB! 171 */ 172 if (p->p_nthreads > 1) { 173 lwp_exit(0); /* called w/ p_token held */ 174 /* NOT REACHED */ 175 } 176 /* else last lwp in proc: do the real thing */ 177 /* FALLTHROUGH */ 178 default: /* to help gcc */ 179 case EXTEXIT_PROC: 180 lwkt_reltoken(&p->p_token); 181 exit1(W_EXITCODE(uap->status, 0)); 182 /* NOTREACHED */ 183 } 184 185 /* NOTREACHED */ 186 lwkt_reltoken(&p->p_token); /* safety */ 187 } 188 189 /* 190 * Kill all lwps associated with the current process except the 191 * current lwp. Return an error if we race another thread trying to 192 * do the same thing and lose the race. 193 * 194 * If forexec is non-zero the current thread and process flags are 195 * cleaned up so they can be reused. 196 */ 197 int 198 killalllwps(int forexec) 199 { 200 struct lwp *lp = curthread->td_lwp; 201 struct proc *p = lp->lwp_proc; 202 203 /* 204 * Interlock against P_WEXIT. Only one of the process's thread 205 * is allowed to do the master exit. 206 */ 207 if (p->p_flag & P_WEXIT) 208 return (EALREADY); 209 p->p_flag |= P_WEXIT; 210 211 /* 212 * Interlock with LWP_WEXIT and kill any remaining LWPs 213 */ 214 lp->lwp_flag |= LWP_WEXIT; 215 if (p->p_nthreads > 1) 216 killlwps(lp); 217 218 /* 219 * If doing this for an exec, clean up the remaining thread 220 * (us) for continuing operation after all the other threads 221 * have been killed. 222 */ 223 if (forexec) { 224 lp->lwp_flag &= ~LWP_WEXIT; 225 p->p_flag &= ~P_WEXIT; 226 } 227 return(0); 228 } 229 230 /* 231 * Kill all LWPs except the current one. Do not try to signal 232 * LWPs which have exited on their own or have already been 233 * signaled. 234 */ 235 static void 236 killlwps(struct lwp *lp) 237 { 238 struct proc *p = lp->lwp_proc; 239 struct lwp *tlp; 240 241 /* 242 * Kill the remaining LWPs. We must send the signal before setting 243 * LWP_WEXIT. The setting of WEXIT is optional but helps reduce 244 * races. tlp must be held across the call as it might block and 245 * allow the target lwp to rip itself out from under our loop. 246 */ 247 FOREACH_LWP_IN_PROC(tlp, p) { 248 LWPHOLD(tlp); 249 if ((tlp->lwp_flag & LWP_WEXIT) == 0) { 250 lwpsignal(p, tlp, SIGKILL); 251 tlp->lwp_flag |= LWP_WEXIT; 252 } 253 LWPRELE(tlp); 254 } 255 256 /* 257 * Wait for everything to clear out. 258 */ 259 while (p->p_nthreads > 1) { 260 tsleep(&p->p_nthreads, 0, "killlwps", 0); 261 } 262 } 263 264 /* 265 * Exit: deallocate address space and other resources, change proc state 266 * to zombie, and unlink proc from allproc and parent's lists. Save exit 267 * status and rusage for wait(). Check for child processes and orphan them. 268 */ 269 void 270 exit1(int rv) 271 { 272 struct thread *td = curthread; 273 struct proc *p = td->td_proc; 274 struct lwp *lp = td->td_lwp; 275 struct proc *q, *nq; 276 struct vmspace *vm; 277 struct vnode *vtmp; 278 struct exitlist *ep; 279 int error; 280 281 lwkt_gettoken(&p->p_token); 282 283 if (p->p_pid == 1) { 284 kprintf("init died (signal %d, exit %d)\n", 285 WTERMSIG(rv), WEXITSTATUS(rv)); 286 panic("Going nowhere without my init!"); 287 } 288 varsymset_clean(&p->p_varsymset); 289 lockuninit(&p->p_varsymset.vx_lock); 290 /* 291 * Kill all lwps associated with the current process, return an 292 * error if we race another thread trying to do the same thing 293 * and lose the race. 294 */ 295 error = killalllwps(0); 296 if (error) { 297 lwp_exit(0); 298 /* NOT REACHED */ 299 } 300 301 caps_exit(lp->lwp_thread); 302 aio_proc_rundown(p); 303 304 /* are we a task leader? */ 305 if (p == p->p_leader) { 306 struct kill_args killArgs; 307 killArgs.signum = SIGKILL; 308 q = p->p_peers; 309 while(q) { 310 killArgs.pid = q->p_pid; 311 /* 312 * The interface for kill is better 313 * than the internal signal 314 */ 315 sys_kill(&killArgs); 316 nq = q; 317 q = q->p_peers; 318 } 319 while (p->p_peers) 320 tsleep((caddr_t)p, 0, "exit1", 0); 321 } 322 323 #ifdef PGINPROF 324 vmsizmon(); 325 #endif 326 STOPEVENT(p, S_EXIT, rv); 327 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */ 328 329 /* 330 * Check if any loadable modules need anything done at process exit. 331 * e.g. SYSV IPC stuff 332 * XXX what if one of these generates an error? 333 */ 334 p->p_xstat = rv; 335 EVENTHANDLER_INVOKE(process_exit, p); 336 337 /* 338 * XXX: imho, the eventhandler stuff is much cleaner than this. 339 * Maybe we should move everything to use eventhandler. 340 */ 341 TAILQ_FOREACH(ep, &exit_list, next) 342 (*ep->function)(td); 343 344 if (p->p_flag & P_PROFIL) 345 stopprofclock(p); 346 /* 347 * If parent is waiting for us to exit or exec, 348 * P_PPWAIT is set; we will wakeup the parent below. 349 */ 350 p->p_flag &= ~(P_TRACED | P_PPWAIT); 351 SIGEMPTYSET(p->p_siglist); 352 SIGEMPTYSET(lp->lwp_siglist); 353 if (timevalisset(&p->p_realtimer.it_value)) 354 callout_stop(&p->p_ithandle); 355 356 /* 357 * Reset any sigio structures pointing to us as a result of 358 * F_SETOWN with our pid. 359 */ 360 funsetownlst(&p->p_sigiolst); 361 362 /* 363 * Close open files and release open-file table. 364 * This may block! 365 */ 366 fdfree(p, NULL); 367 368 if(p->p_leader->p_peers) { 369 q = p->p_leader; 370 while(q->p_peers != p) 371 q = q->p_peers; 372 q->p_peers = p->p_peers; 373 wakeup((caddr_t)p->p_leader); 374 } 375 376 /* 377 * XXX Shutdown SYSV semaphores 378 */ 379 semexit(p); 380 381 KKASSERT(p->p_numposixlocks == 0); 382 383 /* The next two chunks should probably be moved to vmspace_exit. */ 384 vm = p->p_vmspace; 385 386 /* 387 * Release upcalls associated with this process 388 */ 389 if (vm->vm_upcalls) 390 upc_release(vm, lp); 391 392 /* 393 * Clean up data related to virtual kernel operation. Clean up 394 * any vkernel context related to the current lwp now so we can 395 * destroy p_vkernel. 396 */ 397 if (p->p_vkernel) { 398 vkernel_lwp_exit(lp); 399 vkernel_exit(p); 400 } 401 402 /* 403 * Release user portion of address space. 404 * This releases references to vnodes, 405 * which could cause I/O if the file has been unlinked. 406 * Need to do this early enough that we can still sleep. 407 * Can't free the entire vmspace as the kernel stack 408 * may be mapped within that space also. 409 * 410 * Processes sharing the same vmspace may exit in one order, and 411 * get cleaned up by vmspace_exit() in a different order. The 412 * last exiting process to reach this point releases as much of 413 * the environment as it can, and the last process cleaned up 414 * by vmspace_exit() (which decrements exitingcnt) cleans up the 415 * remainder. 416 */ 417 vmspace_exitbump(vm); 418 sysref_put(&vm->vm_sysref); 419 420 if (SESS_LEADER(p)) { 421 struct session *sp = p->p_session; 422 423 if (sp->s_ttyvp) { 424 /* 425 * We are the controlling process. Signal the 426 * foreground process group, drain the controlling 427 * terminal, and revoke access to the controlling 428 * terminal. 429 * 430 * NOTE: while waiting for the process group to exit 431 * it is possible that one of the processes in the 432 * group will revoke the tty, so the ttyclosesession() 433 * function will re-check sp->s_ttyvp. 434 */ 435 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 436 if (sp->s_ttyp->t_pgrp) 437 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 438 ttywait(sp->s_ttyp); 439 ttyclosesession(sp, 1); /* also revoke */ 440 } 441 /* 442 * Release the tty. If someone has it open via 443 * /dev/tty then close it (since they no longer can 444 * once we've NULL'd it out). 445 */ 446 ttyclosesession(sp, 0); 447 448 /* 449 * s_ttyp is not zero'd; we use this to indicate 450 * that the session once had a controlling terminal. 451 * (for logging and informational purposes) 452 */ 453 } 454 sp->s_leader = NULL; 455 } 456 fixjobc(p, p->p_pgrp, 0); 457 (void)acct_process(p); 458 #ifdef KTRACE 459 /* 460 * release trace file 461 */ 462 if (p->p_tracenode) 463 ktrdestroy(&p->p_tracenode); 464 p->p_traceflag = 0; 465 #endif 466 /* 467 * Release reference to text vnode 468 */ 469 if ((vtmp = p->p_textvp) != NULL) { 470 p->p_textvp = NULL; 471 vrele(vtmp); 472 } 473 474 /* Release namecache handle to text file */ 475 if (p->p_textnch.ncp) 476 cache_drop(&p->p_textnch); 477 478 /* 479 * Move the process to the zombie list. This will block 480 * until the process p_lock count reaches 0. The process will 481 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 482 * which is called from cpu_proc_exit(). 483 */ 484 proc_move_allproc_zombie(p); 485 486 q = LIST_FIRST(&p->p_children); 487 if (q) /* only need this if any child is S_ZOMB */ 488 wakeup((caddr_t) initproc); 489 for (; q != 0; q = nq) { 490 nq = LIST_NEXT(q, p_sibling); 491 LIST_REMOVE(q, p_sibling); 492 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling); 493 q->p_pptr = initproc; 494 q->p_sigparent = SIGCHLD; 495 /* 496 * Traced processes are killed 497 * since their existence means someone is screwing up. 498 */ 499 if (q->p_flag & P_TRACED) { 500 q->p_flag &= ~P_TRACED; 501 ksignal(q, SIGKILL); 502 } 503 } 504 505 /* 506 * Save exit status and final rusage info, adding in child rusage 507 * info and self times. 508 */ 509 calcru_proc(p, &p->p_ru); 510 ruadd(&p->p_ru, &p->p_cru); 511 512 /* 513 * notify interested parties of our demise. 514 */ 515 KNOTE(&p->p_klist, NOTE_EXIT); 516 517 /* 518 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 519 * flag set, notify process 1 instead (and hope it will handle 520 * this situation). 521 */ 522 if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) { 523 struct proc *pp = p->p_pptr; 524 525 PHOLD(pp); 526 lwkt_gettoken(&pp->p_token); 527 proc_reparent(p, initproc); 528 529 /* 530 * If this was the last child of our parent, notify 531 * parent, so in case he was wait(2)ing, he will 532 * continue. This function interlocks with pptr->p_token. 533 */ 534 if (LIST_EMPTY(&pp->p_children)) 535 wakeup((caddr_t)pp); 536 lwkt_reltoken(&pp->p_token); 537 PRELE(pp); 538 } 539 540 /* lwkt_gettoken(&proc_token); */ 541 q = p->p_pptr; 542 PHOLD(q); 543 if (p->p_sigparent && q != initproc) { 544 ksignal(q, p->p_sigparent); 545 } else { 546 ksignal(q, SIGCHLD); 547 } 548 wakeup(p->p_pptr); 549 PRELE(q); 550 /* lwkt_reltoken(&proc_token); */ 551 /* NOTE: p->p_pptr can get ripped out */ 552 /* 553 * cpu_exit is responsible for clearing curproc, since 554 * it is heavily integrated with the thread/switching sequence. 555 * 556 * Other substructures are freed from wait(). 557 */ 558 plimit_free(p); 559 560 /* 561 * Release the current user process designation on the process so 562 * the userland scheduler can work in someone else. 563 */ 564 p->p_usched->release_curproc(lp); 565 566 /* 567 * Finally, call machine-dependent code to release as many of the 568 * lwp's resources as we can and halt execution of this thread. 569 */ 570 lwp_exit(1); 571 } 572 573 /* 574 * Eventually called by every exiting LWP 575 * 576 * p->p_token must be held. mplock may be held and will be released. 577 */ 578 void 579 lwp_exit(int masterexit) 580 { 581 struct thread *td = curthread; 582 struct lwp *lp = td->td_lwp; 583 struct proc *p = lp->lwp_proc; 584 585 /* 586 * lwp_exit() may be called without setting LWP_WEXIT, so 587 * make sure it is set here. 588 */ 589 ASSERT_LWKT_TOKEN_HELD(&p->p_token); 590 lp->lwp_flag |= LWP_WEXIT; 591 592 /* 593 * Clean up any virtualization 594 */ 595 if (lp->lwp_vkernel) 596 vkernel_lwp_exit(lp); 597 598 /* 599 * Clean up select/poll support 600 */ 601 kqueue_terminate(&lp->lwp_kqueue); 602 603 /* 604 * Clean up any syscall-cached ucred 605 */ 606 if (td->td_ucred) { 607 crfree(td->td_ucred); 608 td->td_ucred = NULL; 609 } 610 611 /* 612 * Nobody actually wakes us when the lock 613 * count reaches zero, so just wait one tick. 614 */ 615 while (lp->lwp_lock > 0) 616 tsleep(lp, 0, "lwpexit", 1); 617 618 /* Hand down resource usage to our proc */ 619 ruadd(&p->p_ru, &lp->lwp_ru); 620 621 /* 622 * If we don't hold the process until the LWP is reaped wait*() 623 * may try to dispose of its vmspace before all the LWPs have 624 * actually terminated. 625 */ 626 PHOLD(p); 627 628 /* 629 * Do any remaining work that might block on us. We should be 630 * coded such that further blocking is ok after decrementing 631 * p_nthreads but don't take the chance. 632 */ 633 dsched_exit_thread(td); 634 biosched_done(curthread); 635 636 /* 637 * We have to use the reaper for all the LWPs except the one doing 638 * the master exit. The LWP doing the master exit can just be 639 * left on p_lwps and the process reaper will deal with it 640 * synchronously, which is much faster. 641 * 642 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0. 643 */ 644 if (masterexit == 0) { 645 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 646 --p->p_nthreads; 647 if (p->p_nthreads <= 1) 648 wakeup(&p->p_nthreads); 649 lwkt_gettoken(&deadlwp_token); 650 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry); 651 taskqueue_enqueue(taskqueue_thread[mycpuid], 652 deadlwp_task[mycpuid]); 653 lwkt_reltoken(&deadlwp_token); 654 } else { 655 --p->p_nthreads; 656 if (p->p_nthreads <= 1) 657 wakeup(&p->p_nthreads); 658 } 659 660 /* 661 * Release p_token. The mp_token may also be held and we depend on 662 * the lwkt_switch() code to clean it up. 663 */ 664 lwkt_reltoken(&p->p_token); 665 cpu_lwp_exit(); 666 } 667 668 /* 669 * Wait until a lwp is completely dead. 670 * 671 * If the thread is still executing, which can't be waited upon, 672 * return failure. The caller is responsible of waiting a little 673 * bit and checking again. 674 * 675 * Suggested use: 676 * while (!lwp_wait(lp)) 677 * tsleep(lp, 0, "lwpwait", 1); 678 */ 679 static int 680 lwp_wait(struct lwp *lp) 681 { 682 struct thread *td = lp->lwp_thread;; 683 684 KKASSERT(lwkt_preempted_proc() != lp); 685 686 while (lp->lwp_lock > 0) 687 tsleep(lp, 0, "lwpwait1", 1); 688 689 lwkt_wait_free(td); 690 691 /* 692 * The lwp's thread may still be in the middle 693 * of switching away, we can't rip its stack out from 694 * under it until TDF_EXITING is set and both 695 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 696 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 697 * will be cleared temporarily if a thread gets 698 * preempted. 699 * 700 * YYY no wakeup occurs, so we simply return failure 701 * and let the caller deal with sleeping and calling 702 * us again. 703 */ 704 if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK| 705 TDF_EXITING|TDF_RUNQ)) != TDF_EXITING) { 706 return (0); 707 } 708 KASSERT((td->td_flags & TDF_TSLEEPQ) == 0, 709 ("lwp_wait: td %p (%s) still on sleep queue", td, td->td_comm)); 710 return (1); 711 } 712 713 /* 714 * Release the resources associated with a lwp. 715 * The lwp must be completely dead. 716 */ 717 void 718 lwp_dispose(struct lwp *lp) 719 { 720 struct thread *td = lp->lwp_thread;; 721 722 KKASSERT(lwkt_preempted_proc() != lp); 723 KKASSERT(td->td_refs == 0); 724 KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) == 725 TDF_EXITING); 726 727 PRELE(lp->lwp_proc); 728 lp->lwp_proc = NULL; 729 if (td != NULL) { 730 td->td_proc = NULL; 731 td->td_lwp = NULL; 732 lp->lwp_thread = NULL; 733 lwkt_free_thread(td); 734 } 735 kfree(lp, M_LWP); 736 } 737 738 /* 739 * MPSAFE 740 */ 741 int 742 sys_wait4(struct wait_args *uap) 743 { 744 struct rusage rusage; 745 int error, status; 746 747 error = kern_wait(uap->pid, (uap->status ? &status : NULL), 748 uap->options, (uap->rusage ? &rusage : NULL), 749 &uap->sysmsg_result); 750 751 if (error == 0 && uap->status) 752 error = copyout(&status, uap->status, sizeof(*uap->status)); 753 if (error == 0 && uap->rusage) 754 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage)); 755 return (error); 756 } 757 758 /* 759 * wait1() 760 * 761 * wait_args(int pid, int *status, int options, struct rusage *rusage) 762 * 763 * MPALMOSTSAFE 764 */ 765 int 766 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res) 767 { 768 struct thread *td = curthread; 769 struct lwp *lp; 770 struct proc *q = td->td_proc; 771 struct proc *p, *t; 772 int nfound, error; 773 774 if (pid == 0) 775 pid = -q->p_pgid; 776 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE)) 777 return (EINVAL); 778 779 lwkt_gettoken(&q->p_token); 780 loop: 781 /* 782 * All sorts of things can change due to blocking so we have to loop 783 * all the way back up here. 784 * 785 * The problem is that if a process group is stopped and the parent 786 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 787 * of the child and then stop itself when it tries to return from the 788 * system call. When the process group is resumed the parent will 789 * then get the STOP status even though the child has now resumed 790 * (a followup wait*() will get the CONT status). 791 * 792 * Previously the CONT would overwrite the STOP because the tstop 793 * was handled within tsleep(), and the parent would only see 794 * the CONT when both are stopped and continued together. This little 795 * two-line hack restores this effect. 796 */ 797 while (q->p_stat == SSTOP) 798 tstop(); 799 800 nfound = 0; 801 802 LIST_FOREACH(p, &q->p_children, p_sibling) { 803 if (pid != WAIT_ANY && 804 p->p_pid != pid && p->p_pgid != -pid) { 805 continue; 806 } 807 808 /* 809 * This special case handles a kthread spawned by linux_clone 810 * (see linux_misc.c). The linux_wait4 and linux_waitpid 811 * functions need to be able to distinguish between waiting 812 * on a process and waiting on a thread. It is a thread if 813 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 814 * signifies we want to wait for threads and not processes. 815 */ 816 if ((p->p_sigparent != SIGCHLD) ^ 817 ((options & WLINUXCLONE) != 0)) { 818 continue; 819 } 820 821 nfound++; 822 if (p->p_stat == SZOMB) { 823 /* 824 * We may go into SZOMB with threads still present. 825 * We must wait for them to exit before we can reap 826 * the master thread, otherwise we may race reaping 827 * non-master threads. 828 */ 829 lwkt_gettoken(&p->p_token); 830 while (p->p_nthreads > 0) { 831 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 832 } 833 834 /* 835 * Reap any LWPs left in p->p_lwps. This is usually 836 * just the last LWP. This must be done before 837 * we loop on p_lock since the lwps hold a ref on 838 * it as a vmspace interlock. 839 * 840 * Once that is accomplished p_nthreads had better 841 * be zero. 842 */ 843 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 844 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 845 reaplwp(lp); 846 } 847 KKASSERT(p->p_nthreads == 0); 848 lwkt_reltoken(&p->p_token); 849 850 /* 851 * Don't do anything really bad until all references 852 * to the process go away. This may include other 853 * LWPs which are still in the process of being 854 * reaped. We can't just pull the rug out from under 855 * them because they may still be using the VM space. 856 * 857 * Certain kernel facilities such as /proc will also 858 * put a hold on the process for short periods of 859 * time. 860 */ 861 while (p->p_lock) 862 tsleep(p, 0, "reap3", hz); 863 864 /* Take care of our return values. */ 865 *res = p->p_pid; 866 if (status) 867 *status = p->p_xstat; 868 if (rusage) 869 *rusage = p->p_ru; 870 /* 871 * If we got the child via a ptrace 'attach', 872 * we need to give it back to the old parent. 873 */ 874 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 875 p->p_oppid = 0; 876 proc_reparent(p, t); 877 ksignal(t, SIGCHLD); 878 wakeup((caddr_t)t); 879 error = 0; 880 PRELE(t); 881 goto done; 882 } 883 884 /* 885 * Unlink the proc from its process group so that 886 * the following operations won't lead to an 887 * inconsistent state for processes running down 888 * the zombie list. 889 */ 890 KKASSERT(p->p_lock == 0); 891 proc_remove_zombie(p); 892 leavepgrp(p); 893 894 p->p_xstat = 0; 895 ruadd(&q->p_cru, &p->p_ru); 896 897 /* 898 * Decrement the count of procs running with this uid. 899 */ 900 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 901 902 /* 903 * Free up credentials. 904 */ 905 crfree(p->p_ucred); 906 p->p_ucred = NULL; 907 908 /* 909 * Remove unused arguments 910 */ 911 if (p->p_args && --p->p_args->ar_ref == 0) 912 FREE(p->p_args, M_PARGS); 913 914 if (--p->p_sigacts->ps_refcnt == 0) { 915 kfree(p->p_sigacts, M_SUBPROC); 916 p->p_sigacts = NULL; 917 } 918 919 vm_waitproc(p); 920 kfree(p, M_PROC); 921 nprocs--; 922 error = 0; 923 goto done; 924 } 925 if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 && 926 (p->p_flag & P_TRACED || options & WUNTRACED)) { 927 p->p_flag |= P_WAITED; 928 929 *res = p->p_pid; 930 if (status) 931 *status = W_STOPCODE(p->p_xstat); 932 /* Zero rusage so we get something consistent. */ 933 if (rusage) 934 bzero(rusage, sizeof(rusage)); 935 error = 0; 936 goto done; 937 } 938 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) { 939 *res = p->p_pid; 940 p->p_flag &= ~P_CONTINUED; 941 942 if (status) 943 *status = SIGCONT; 944 error = 0; 945 goto done; 946 } 947 } 948 if (nfound == 0) { 949 error = ECHILD; 950 goto done; 951 } 952 if (options & WNOHANG) { 953 *res = 0; 954 error = 0; 955 goto done; 956 } 957 958 /* 959 * Wait for signal - interlocked using q->p_token. 960 */ 961 error = tsleep(q, PCATCH, "wait", 0); 962 if (error) { 963 done: 964 lwkt_reltoken(&q->p_token); 965 return (error); 966 } 967 goto loop; 968 } 969 970 /* 971 * Make process 'parent' the new parent of process 'child'. 972 */ 973 void 974 proc_reparent(struct proc *child, struct proc *parent) 975 { 976 if (child->p_pptr == parent) 977 return; 978 PHOLD(parent); 979 lwkt_gettoken(&child->p_token); 980 lwkt_gettoken(&parent->p_token); 981 LIST_REMOVE(child, p_sibling); 982 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 983 child->p_pptr = parent; 984 lwkt_reltoken(&parent->p_token); 985 lwkt_reltoken(&child->p_token); 986 PRELE(parent); 987 } 988 989 /* 990 * The next two functions are to handle adding/deleting items on the 991 * exit callout list 992 * 993 * at_exit(): 994 * Take the arguments given and put them onto the exit callout list, 995 * However first make sure that it's not already there. 996 * returns 0 on success. 997 */ 998 999 int 1000 at_exit(exitlist_fn function) 1001 { 1002 struct exitlist *ep; 1003 1004 #ifdef INVARIANTS 1005 /* Be noisy if the programmer has lost track of things */ 1006 if (rm_at_exit(function)) 1007 kprintf("WARNING: exit callout entry (%p) already present\n", 1008 function); 1009 #endif 1010 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 1011 if (ep == NULL) 1012 return (ENOMEM); 1013 ep->function = function; 1014 TAILQ_INSERT_TAIL(&exit_list, ep, next); 1015 return (0); 1016 } 1017 1018 /* 1019 * Scan the exit callout list for the given item and remove it. 1020 * Returns the number of items removed (0 or 1) 1021 */ 1022 int 1023 rm_at_exit(exitlist_fn function) 1024 { 1025 struct exitlist *ep; 1026 1027 TAILQ_FOREACH(ep, &exit_list, next) { 1028 if (ep->function == function) { 1029 TAILQ_REMOVE(&exit_list, ep, next); 1030 kfree(ep, M_ATEXIT); 1031 return(1); 1032 } 1033 } 1034 return (0); 1035 } 1036 1037 /* 1038 * LWP reaper related code. 1039 */ 1040 static void 1041 reaplwps(void *context, int dummy) 1042 { 1043 struct lwplist *lwplist = context; 1044 struct lwp *lp; 1045 1046 lwkt_gettoken(&deadlwp_token); 1047 while ((lp = LIST_FIRST(lwplist))) { 1048 LIST_REMOVE(lp, u.lwp_reap_entry); 1049 reaplwp(lp); 1050 } 1051 lwkt_reltoken(&deadlwp_token); 1052 } 1053 1054 static void 1055 reaplwp(struct lwp *lp) 1056 { 1057 while (lwp_wait(lp) == 0) 1058 tsleep(lp, 0, "lwpreap", 1); 1059 lwp_dispose(lp); 1060 } 1061 1062 static void 1063 deadlwp_init(void) 1064 { 1065 int cpu; 1066 1067 for (cpu = 0; cpu < ncpus; cpu++) { 1068 LIST_INIT(&deadlwp_list[cpu]); 1069 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK); 1070 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1071 } 1072 } 1073 1074 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1075