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. 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 * $FreeBSD: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $ 36 */ 37 38 #include "opt_ktrace.h" 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/sysproto.h> 43 #include <sys/kernel.h> 44 #include <sys/malloc.h> 45 #include <sys/proc.h> 46 #include <sys/ktrace.h> 47 #include <sys/pioctl.h> 48 #include <sys/tty.h> 49 #include <sys/wait.h> 50 #include <sys/vnode.h> 51 #include <sys/resourcevar.h> 52 #include <sys/signalvar.h> 53 #include <sys/taskqueue.h> 54 #include <sys/ptrace.h> 55 #include <sys/acct.h> /* for acct_process() function prototype */ 56 #include <sys/filedesc.h> 57 #include <sys/shm.h> 58 #include <sys/sem.h> 59 #include <sys/jail.h> 60 #include <sys/kern_syscall.h> 61 #include <sys/unistd.h> 62 #include <sys/eventhandler.h> 63 #include <sys/dsched.h> 64 65 #include <vm/vm.h> 66 #include <vm/vm_param.h> 67 #include <sys/lock.h> 68 #include <vm/pmap.h> 69 #include <vm/vm_map.h> 70 #include <vm/vm_extern.h> 71 #include <sys/user.h> 72 73 #include <sys/refcount.h> 74 #include <sys/thread2.h> 75 #include <sys/spinlock2.h> 76 #include <sys/mplock2.h> 77 78 #include <machine/vmm.h> 79 80 static void reaplwps(void *context, int dummy); 81 static void reaplwp(struct lwp *lp); 82 static void killlwps(struct lwp *lp); 83 84 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 85 86 /* 87 * callout list for things to do at exit time 88 */ 89 struct exitlist { 90 exitlist_fn function; 91 TAILQ_ENTRY(exitlist) next; 92 }; 93 94 TAILQ_HEAD(exit_list_head, exitlist); 95 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 96 97 /* 98 * LWP reaper data 99 */ 100 static struct task *deadlwp_task[MAXCPU]; 101 static struct lwplist deadlwp_list[MAXCPU]; 102 static struct lwkt_token deadlwp_token[MAXCPU]; 103 104 /* 105 * exit -- 106 * Death of process. 107 * 108 * SYS_EXIT_ARGS(int rval) 109 */ 110 int 111 sys_exit(struct exit_args *uap) 112 { 113 exit1(W_EXITCODE(uap->rval, 0)); 114 /* NOTREACHED */ 115 } 116 117 /* 118 * Extended exit -- 119 * Death of a lwp or process with optional bells and whistles. 120 */ 121 int 122 sys_extexit(struct extexit_args *uap) 123 { 124 struct proc *p = curproc; 125 int action, who; 126 int error; 127 128 action = EXTEXIT_ACTION(uap->how); 129 who = EXTEXIT_WHO(uap->how); 130 131 /* Check parameters before we might perform some action */ 132 switch (who) { 133 case EXTEXIT_PROC: 134 case EXTEXIT_LWP: 135 break; 136 default: 137 return (EINVAL); 138 } 139 140 switch (action) { 141 case EXTEXIT_SIMPLE: 142 break; 143 case EXTEXIT_SETINT: 144 error = copyout(&uap->status, uap->addr, sizeof(uap->status)); 145 if (error) 146 return (error); 147 break; 148 default: 149 return (EINVAL); 150 } 151 152 lwkt_gettoken(&p->p_token); 153 154 switch (who) { 155 case EXTEXIT_LWP: 156 /* 157 * Be sure only to perform a simple lwp exit if there is at 158 * least one more lwp in the proc, which will call exit1() 159 * later, otherwise the proc will be an UNDEAD and not even a 160 * SZOMB! 161 */ 162 if (p->p_nthreads > 1) { 163 lwp_exit(0, NULL); /* called w/ p_token held */ 164 /* NOT REACHED */ 165 } 166 /* else last lwp in proc: do the real thing */ 167 /* FALLTHROUGH */ 168 default: /* to help gcc */ 169 case EXTEXIT_PROC: 170 lwkt_reltoken(&p->p_token); 171 exit1(W_EXITCODE(uap->status, 0)); 172 /* NOTREACHED */ 173 } 174 175 /* NOTREACHED */ 176 lwkt_reltoken(&p->p_token); /* safety */ 177 } 178 179 /* 180 * Kill all lwps associated with the current process except the 181 * current lwp. Return an error if we race another thread trying to 182 * do the same thing and lose the race. 183 * 184 * If forexec is non-zero the current thread and process flags are 185 * cleaned up so they can be reused. 186 * 187 * Caller must hold curproc->p_token 188 */ 189 int 190 killalllwps(int forexec) 191 { 192 struct lwp *lp = curthread->td_lwp; 193 struct proc *p = lp->lwp_proc; 194 int fakestop; 195 196 /* 197 * Interlock against P_WEXIT. Only one of the process's thread 198 * is allowed to do the master exit. 199 */ 200 if (p->p_flags & P_WEXIT) 201 return (EALREADY); 202 p->p_flags |= P_WEXIT; 203 204 /* 205 * Set temporary stopped state in case we are racing a coredump. 206 * Otherwise the coredump may hang forever. 207 */ 208 if (lp->lwp_mpflags & LWP_MP_WSTOP) { 209 fakestop = 0; 210 } else { 211 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP); 212 ++p->p_nstopped; 213 fakestop = 1; 214 wakeup(&p->p_nstopped); 215 } 216 217 /* 218 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs 219 */ 220 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 221 if (p->p_nthreads > 1) 222 killlwps(lp); 223 224 /* 225 * Undo temporary stopped state 226 */ 227 if (fakestop) { 228 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WSTOP); 229 --p->p_nstopped; 230 } 231 232 /* 233 * If doing this for an exec, clean up the remaining thread 234 * (us) for continuing operation after all the other threads 235 * have been killed. 236 */ 237 if (forexec) { 238 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 239 p->p_flags &= ~P_WEXIT; 240 } 241 return(0); 242 } 243 244 /* 245 * Kill all LWPs except the current one. Do not try to signal 246 * LWPs which have exited on their own or have already been 247 * signaled. 248 */ 249 static void 250 killlwps(struct lwp *lp) 251 { 252 struct proc *p = lp->lwp_proc; 253 struct lwp *tlp; 254 255 /* 256 * Kill the remaining LWPs. We must send the signal before setting 257 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce 258 * races. tlp must be held across the call as it might block and 259 * allow the target lwp to rip itself out from under our loop. 260 */ 261 FOREACH_LWP_IN_PROC(tlp, p) { 262 LWPHOLD(tlp); 263 lwkt_gettoken(&tlp->lwp_token); 264 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) { 265 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT); 266 lwpsignal(p, tlp, SIGKILL); 267 } 268 lwkt_reltoken(&tlp->lwp_token); 269 LWPRELE(tlp); 270 } 271 272 /* 273 * Wait for everything to clear out. Also make sure any tstop()s 274 * are signalled (we are holding p_token for the interlock). 275 */ 276 wakeup(p); 277 while (p->p_nthreads > 1) 278 tsleep(&p->p_nthreads, 0, "killlwps", 0); 279 } 280 281 /* 282 * Exit: deallocate address space and other resources, change proc state 283 * to zombie, and unlink proc from allproc and parent's lists. Save exit 284 * status and rusage for wait(). Check for child processes and orphan them. 285 */ 286 void 287 exit1(int rv) 288 { 289 struct thread *td = curthread; 290 struct proc *p = td->td_proc; 291 struct lwp *lp = td->td_lwp; 292 struct proc *q; 293 struct proc *pp; 294 struct proc *reproc; 295 struct sysreaper *reap; 296 struct vmspace *vm; 297 struct vnode *vtmp; 298 struct exitlist *ep; 299 int error; 300 301 lwkt_gettoken(&p->p_token); 302 303 if (p->p_pid == 1) { 304 kprintf("init died (signal %d, exit %d)\n", 305 WTERMSIG(rv), WEXITSTATUS(rv)); 306 panic("Going nowhere without my init!"); 307 } 308 varsymset_clean(&p->p_varsymset); 309 lockuninit(&p->p_varsymset.vx_lock); 310 311 /* 312 * Kill all lwps associated with the current process, return an 313 * error if we race another thread trying to do the same thing 314 * and lose the race. 315 */ 316 error = killalllwps(0); 317 if (error) { 318 lwp_exit(0, NULL); 319 /* NOT REACHED */ 320 } 321 322 /* are we a task leader? */ 323 if (p == p->p_leader) { 324 struct kill_args killArgs; 325 killArgs.signum = SIGKILL; 326 q = p->p_peers; 327 while(q) { 328 killArgs.pid = q->p_pid; 329 /* 330 * The interface for kill is better 331 * than the internal signal 332 */ 333 sys_kill(&killArgs); 334 q = q->p_peers; 335 } 336 while (p->p_peers) 337 tsleep((caddr_t)p, 0, "exit1", 0); 338 } 339 340 #ifdef PGINPROF 341 vmsizmon(); 342 #endif 343 STOPEVENT(p, S_EXIT, rv); 344 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */ 345 346 /* 347 * Check if any loadable modules need anything done at process exit. 348 * e.g. SYSV IPC stuff 349 * XXX what if one of these generates an error? 350 */ 351 p->p_xstat = rv; 352 353 /* 354 * XXX: imho, the eventhandler stuff is much cleaner than this. 355 * Maybe we should move everything to use eventhandler. 356 */ 357 TAILQ_FOREACH(ep, &exit_list, next) 358 (*ep->function)(td); 359 360 if (p->p_flags & P_PROFIL) 361 stopprofclock(p); 362 363 SIGEMPTYSET(p->p_siglist); 364 SIGEMPTYSET(lp->lwp_siglist); 365 if (timevalisset(&p->p_realtimer.it_value)) 366 callout_stop_sync(&p->p_ithandle); 367 368 /* 369 * Reset any sigio structures pointing to us as a result of 370 * F_SETOWN with our pid. 371 */ 372 funsetownlst(&p->p_sigiolst); 373 374 /* 375 * Close open files and release open-file table. 376 * This may block! 377 */ 378 fdfree(p, NULL); 379 380 if (p->p_leader->p_peers) { 381 q = p->p_leader; 382 while(q->p_peers != p) 383 q = q->p_peers; 384 q->p_peers = p->p_peers; 385 wakeup((caddr_t)p->p_leader); 386 } 387 388 /* 389 * XXX Shutdown SYSV semaphores 390 */ 391 semexit(p); 392 393 /* The next two chunks should probably be moved to vmspace_exit. */ 394 vm = p->p_vmspace; 395 396 /* 397 * Clean up data related to virtual kernel operation. Clean up 398 * any vkernel context related to the current lwp now so we can 399 * destroy p_vkernel. 400 */ 401 if (p->p_vkernel) { 402 vkernel_lwp_exit(lp); 403 vkernel_exit(p); 404 } 405 406 /* 407 * Release the user portion of address space. The exitbump prevents 408 * the vmspace from being completely eradicated (using holdcnt). 409 * This releases references to vnodes, which could cause I/O if the 410 * file has been unlinked. We need to do this early enough that 411 * we can still sleep. 412 * 413 * We can't free the entire vmspace as the kernel stack may be mapped 414 * within that space also. 415 * 416 * Processes sharing the same vmspace may exit in one order, and 417 * get cleaned up by vmspace_exit() in a different order. The 418 * last exiting process to reach this point releases as much of 419 * the environment as it can, and the last process cleaned up 420 * by vmspace_exit() (which decrements exitingcnt) cleans up the 421 * remainder. 422 * 423 * NOTE: Releasing p_token around this call is helpful if the 424 * vmspace had a huge RSS. Otherwise some other process 425 * trying to do an allproc or other scan (like 'ps') may 426 * stall for a long time. 427 */ 428 lwkt_reltoken(&p->p_token); 429 vmspace_relexit(vm); 430 lwkt_gettoken(&p->p_token); 431 432 if (SESS_LEADER(p)) { 433 struct session *sp = p->p_session; 434 435 if (sp->s_ttyvp) { 436 /* 437 * We are the controlling process. Signal the 438 * foreground process group, drain the controlling 439 * terminal, and revoke access to the controlling 440 * terminal. 441 * 442 * NOTE: while waiting for the process group to exit 443 * it is possible that one of the processes in the 444 * group will revoke the tty, so the ttyclosesession() 445 * function will re-check sp->s_ttyvp. 446 */ 447 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 448 if (sp->s_ttyp->t_pgrp) 449 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 450 ttywait(sp->s_ttyp); 451 ttyclosesession(sp, 1); /* also revoke */ 452 } 453 /* 454 * Release the tty. If someone has it open via 455 * /dev/tty then close it (since they no longer can 456 * once we've NULL'd it out). 457 */ 458 ttyclosesession(sp, 0); 459 460 /* 461 * s_ttyp is not zero'd; we use this to indicate 462 * that the session once had a controlling terminal. 463 * (for logging and informational purposes) 464 */ 465 } 466 sp->s_leader = NULL; 467 } 468 fixjobc(p, p->p_pgrp, 0); 469 (void)acct_process(p); 470 #ifdef KTRACE 471 /* 472 * release trace file 473 */ 474 if (p->p_tracenode) 475 ktrdestroy(&p->p_tracenode); 476 p->p_traceflag = 0; 477 #endif 478 /* 479 * Release reference to text vnode 480 */ 481 if ((vtmp = p->p_textvp) != NULL) { 482 p->p_textvp = NULL; 483 vrele(vtmp); 484 } 485 486 /* Release namecache handle to text file */ 487 if (p->p_textnch.ncp) 488 cache_drop(&p->p_textnch); 489 490 /* 491 * We have to handle PPWAIT here or proc_move_allproc_zombie() 492 * will block on the PHOLD() the parent is doing. 493 * 494 * We are using the flag as an interlock so an atomic op is 495 * necessary to synchronize with the parent's cpu. 496 */ 497 if (p->p_flags & P_PPWAIT) { 498 if (p->p_pptr && p->p_pptr->p_upmap) 499 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1); 500 atomic_clear_int(&p->p_flags, P_PPWAIT); 501 wakeup(p->p_pptr); 502 } 503 504 /* 505 * Move the process to the zombie list. This will block 506 * until the process p_lock count reaches 0. The process will 507 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 508 * which is called from cpu_proc_exit(). 509 * 510 * Interlock against waiters using p_waitgen. We increment 511 * p_waitgen after completing the move of our process to the 512 * zombie list. 513 * 514 * WARNING: pp becomes stale when we block, clear it now as a 515 * reminder. 516 */ 517 proc_move_allproc_zombie(p); 518 pp = p->p_pptr; 519 atomic_add_long(&pp->p_waitgen, 1); 520 pp = NULL; 521 522 /* 523 * release controlled reaper for exit if we own it and return the 524 * remaining reaper (the one for us), which we will drop after we 525 * are done. 526 */ 527 reap = reaper_exit(p); 528 529 /* 530 * Reparent all of this process's children to the init process or 531 * to the designated reaper. We must hold the reaper's p_token in 532 * order to safely mess with p_children. 533 * 534 * We already hold p->p_token (to remove the children from our list). 535 */ 536 reproc = NULL; 537 q = LIST_FIRST(&p->p_children); 538 if (q) { 539 reproc = reaper_get(reap); 540 lwkt_gettoken(&reproc->p_token); 541 while ((q = LIST_FIRST(&p->p_children)) != NULL) { 542 PHOLD(q); 543 lwkt_gettoken(&q->p_token); 544 if (q != LIST_FIRST(&p->p_children)) { 545 lwkt_reltoken(&q->p_token); 546 PRELE(q); 547 continue; 548 } 549 LIST_REMOVE(q, p_sibling); 550 LIST_INSERT_HEAD(&reproc->p_children, q, p_sibling); 551 q->p_pptr = reproc; 552 q->p_ppid = reproc->p_pid; 553 q->p_sigparent = SIGCHLD; 554 555 /* 556 * Traced processes are killed 557 * since their existence means someone is screwing up. 558 */ 559 if (q->p_flags & P_TRACED) { 560 q->p_flags &= ~P_TRACED; 561 ksignal(q, SIGKILL); 562 } 563 lwkt_reltoken(&q->p_token); 564 PRELE(q); 565 } 566 lwkt_reltoken(&reproc->p_token); 567 wakeup(reproc); 568 } 569 570 /* 571 * Save exit status and final rusage info. We no longer add 572 * child rusage info into self times, wait4() and kern_wait() 573 * handles it in order to properly support wait6(). 574 */ 575 calcru_proc(p, &p->p_ru); 576 /*ruadd(&p->p_ru, &p->p_cru); REMOVED */ 577 578 /* 579 * notify interested parties of our demise. 580 */ 581 KNOTE(&p->p_klist, NOTE_EXIT); 582 583 /* 584 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 585 * flag set, or if the handler is set to SIG_IGN, notify the reaper 586 * instead (it will handle this situation). 587 * 588 * NOTE: The reaper can still be the parent process. 589 * 590 * (must reload pp) 591 */ 592 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 593 if (reproc == NULL) 594 reproc = reaper_get(reap); 595 proc_reparent(p, reproc); 596 } 597 if (reproc) 598 PRELE(reproc); 599 if (reap) 600 reaper_drop(reap); 601 602 /* 603 * Signal (possibly new) parent. 604 */ 605 pp = p->p_pptr; 606 PHOLD(pp); 607 if (p->p_sigparent && pp != initproc) { 608 int sig = p->p_sigparent; 609 610 if (sig != SIGUSR1 && sig != SIGCHLD) 611 sig = SIGCHLD; 612 ksignal(pp, sig); 613 } else { 614 ksignal(pp, SIGCHLD); 615 } 616 p->p_flags &= ~P_TRACED; 617 PRELE(pp); 618 619 /* 620 * cpu_exit is responsible for clearing curproc, since 621 * it is heavily integrated with the thread/switching sequence. 622 * 623 * Other substructures are freed from wait(). 624 */ 625 if (p->p_limit) { 626 struct plimit *rlimit; 627 628 rlimit = p->p_limit; 629 p->p_limit = NULL; 630 plimit_free(rlimit); 631 } 632 633 /* 634 * Finally, call machine-dependent code to release as many of the 635 * lwp's resources as we can and halt execution of this thread. 636 * 637 * pp is a wild pointer now but still the correct wakeup() target. 638 * lwp_exit() only uses it to send the wakeup() signal to the likely 639 * parent. Any reparenting race that occurs will get a signal 640 * automatically and not be an issue. 641 */ 642 lwp_exit(1, pp); 643 } 644 645 /* 646 * Eventually called by every exiting LWP 647 * 648 * p->p_token must be held. mplock may be held and will be released. 649 */ 650 void 651 lwp_exit(int masterexit, void *waddr) 652 { 653 struct thread *td = curthread; 654 struct lwp *lp = td->td_lwp; 655 struct proc *p = lp->lwp_proc; 656 int dowake = 0; 657 658 /* 659 * Release the current user process designation on the process so 660 * the userland scheduler can work in someone else. 661 */ 662 p->p_usched->release_curproc(lp); 663 664 /* 665 * lwp_exit() may be called without setting LWP_MP_WEXIT, so 666 * make sure it is set here. 667 */ 668 ASSERT_LWKT_TOKEN_HELD(&p->p_token); 669 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 670 671 /* 672 * Clean up any virtualization 673 */ 674 if (lp->lwp_vkernel) 675 vkernel_lwp_exit(lp); 676 677 if (td->td_vmm) 678 vmm_vmdestroy(); 679 680 /* 681 * Clean up select/poll support 682 */ 683 kqueue_terminate(&lp->lwp_kqueue); 684 685 /* 686 * Clean up any syscall-cached ucred or rlimit. 687 */ 688 if (td->td_ucred) { 689 crfree(td->td_ucred); 690 td->td_ucred = NULL; 691 } 692 if (td->td_limit) { 693 struct plimit *rlimit; 694 695 rlimit = td->td_limit; 696 td->td_limit = NULL; 697 plimit_free(rlimit); 698 } 699 700 /* 701 * Cleanup any cached descriptors for this thread 702 */ 703 if (p->p_fd) 704 fexitcache(td); 705 706 /* 707 * Nobody actually wakes us when the lock 708 * count reaches zero, so just wait one tick. 709 */ 710 while (lp->lwp_lock > 0) 711 tsleep(lp, 0, "lwpexit", 1); 712 713 /* Hand down resource usage to our proc */ 714 ruadd(&p->p_ru, &lp->lwp_ru); 715 716 /* 717 * If we don't hold the process until the LWP is reaped wait*() 718 * may try to dispose of its vmspace before all the LWPs have 719 * actually terminated. 720 */ 721 PHOLD(p); 722 723 /* 724 * Do any remaining work that might block on us. We should be 725 * coded such that further blocking is ok after decrementing 726 * p_nthreads but don't take the chance. 727 */ 728 dsched_exit_thread(td); 729 biosched_done(curthread); 730 731 /* 732 * We have to use the reaper for all the LWPs except the one doing 733 * the master exit. The LWP doing the master exit can just be 734 * left on p_lwps and the process reaper will deal with it 735 * synchronously, which is much faster. 736 * 737 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0. 738 * 739 * The process is left held until the reaper calls lwp_dispose() on 740 * the lp (after calling lwp_wait()). 741 */ 742 if (masterexit == 0) { 743 int cpu = mycpuid; 744 745 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 746 --p->p_nthreads; 747 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1) 748 dowake = 1; 749 lwkt_gettoken(&deadlwp_token[cpu]); 750 LIST_INSERT_HEAD(&deadlwp_list[cpu], lp, u.lwp_reap_entry); 751 taskqueue_enqueue(taskqueue_thread[cpu], deadlwp_task[cpu]); 752 lwkt_reltoken(&deadlwp_token[cpu]); 753 } else { 754 --p->p_nthreads; 755 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1) 756 dowake = 1; 757 } 758 759 /* 760 * We no longer need p_token. 761 * 762 * Tell the userland scheduler that we are going away 763 */ 764 lwkt_reltoken(&p->p_token); 765 p->p_usched->heuristic_exiting(lp, p); 766 767 /* 768 * Issue late wakeups after releasing our token to give us a chance 769 * to deschedule and switch away before another cpu in a wait*() 770 * reaps us. This is done as late as possible to reduce contention. 771 */ 772 if (dowake) 773 wakeup(&p->p_nthreads); 774 if (waddr) 775 wakeup(waddr); 776 777 cpu_lwp_exit(); 778 } 779 780 /* 781 * Wait until a lwp is completely dead. The final interlock in this drama 782 * is when TDF_EXITING is set in cpu_thread_exit() just before the final 783 * switchout. 784 * 785 * At the point TDF_EXITING is set a complete exit is accomplished when 786 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two 787 * post-switch interlock flags that can be used to wait for the TDF_ 788 * flags to clear. 789 * 790 * Returns non-zero on success, and zero if the caller needs to retry 791 * the lwp_wait(). 792 */ 793 static int 794 lwp_wait(struct lwp *lp) 795 { 796 struct thread *td = lp->lwp_thread; 797 u_int mpflags; 798 799 KKASSERT(lwkt_preempted_proc() != lp); 800 801 /* 802 * This bit of code uses the thread destruction interlock 803 * managed by lwkt_switch_return() to wait for the lwp's 804 * thread to completely disengage. 805 * 806 * It is possible for us to race another cpu core so we 807 * have to do this correctly. 808 */ 809 for (;;) { 810 mpflags = td->td_mpflags; 811 cpu_ccfence(); 812 if (mpflags & TDF_MP_EXITSIG) 813 break; 814 tsleep_interlock(td, 0); 815 if (atomic_cmpset_int(&td->td_mpflags, mpflags, 816 mpflags | TDF_MP_EXITWAIT)) { 817 tsleep(td, PINTERLOCKED, "lwpxt", 0); 818 } 819 } 820 821 /* 822 * We've already waited for the core exit but there can still 823 * be other refs from e.g. process scans and such. 824 */ 825 if (lp->lwp_lock > 0) { 826 tsleep(lp, 0, "lwpwait1", 1); 827 return(0); 828 } 829 if (td->td_refs) { 830 tsleep(td, 0, "lwpwait2", 1); 831 return(0); 832 } 833 834 /* 835 * Now that we have the thread destruction interlock these flags 836 * really should already be cleaned up, keep a check for safety. 837 * 838 * We can't rip its stack out from under it until TDF_EXITING is 839 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 840 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 841 * will be cleared temporarily if a thread gets preempted. 842 */ 843 while ((td->td_flags & (TDF_RUNNING | 844 TDF_RUNQ | 845 TDF_PREEMPT_LOCK | 846 TDF_EXITING)) != TDF_EXITING) { 847 tsleep(lp, 0, "lwpwait3", 1); 848 return (0); 849 } 850 851 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0, 852 ("lwp_wait: td %p (%s) still on run or sleep queue", 853 td, td->td_comm)); 854 return (1); 855 } 856 857 /* 858 * Release the resources associated with a lwp. 859 * The lwp must be completely dead. 860 */ 861 void 862 lwp_dispose(struct lwp *lp) 863 { 864 struct thread *td = lp->lwp_thread; 865 866 KKASSERT(lwkt_preempted_proc() != lp); 867 KKASSERT(lp->lwp_lock == 0); 868 KKASSERT(td->td_refs == 0); 869 KKASSERT((td->td_flags & (TDF_RUNNING | 870 TDF_RUNQ | 871 TDF_PREEMPT_LOCK | 872 TDF_EXITING)) == TDF_EXITING); 873 874 PRELE(lp->lwp_proc); 875 lp->lwp_proc = NULL; 876 if (td != NULL) { 877 td->td_proc = NULL; 878 td->td_lwp = NULL; 879 lp->lwp_thread = NULL; 880 lwkt_free_thread(td); 881 } 882 kfree(lp, M_LWP); 883 } 884 885 int 886 sys_wait4(struct wait_args *uap) 887 { 888 struct __wrusage wrusage; 889 int error; 890 int status; 891 int options; 892 id_t id; 893 idtype_t idtype; 894 895 options = uap->options | WEXITED | WTRAPPED; 896 id = uap->pid; 897 898 if (id == WAIT_ANY) { 899 idtype = P_ALL; 900 } else if (id == WAIT_MYPGRP) { 901 idtype = P_PGID; 902 id = curproc->p_pgid; 903 } else if (id < 0) { 904 idtype = P_PGID; 905 id = -id; 906 } else { 907 idtype = P_PID; 908 } 909 910 error = kern_wait(idtype, id, &status, options, &wrusage, 911 NULL, &uap->sysmsg_result); 912 913 if (error == 0 && uap->status) 914 error = copyout(&status, uap->status, sizeof(*uap->status)); 915 if (error == 0 && uap->rusage) { 916 ruadd(&wrusage.wru_self, &wrusage.wru_children); 917 error = copyout(&wrusage.wru_self, uap->rusage, sizeof(*uap->rusage)); 918 } 919 return (error); 920 } 921 922 int 923 sys_wait6(struct wait6_args *uap) 924 { 925 struct __wrusage wrusage; 926 struct __siginfo info; 927 struct __siginfo *infop; 928 int error; 929 int status; 930 int options; 931 id_t id; 932 idtype_t idtype; 933 934 /* 935 * NOTE: wait6() requires WEXITED and WTRAPPED to be specified if 936 * desired. 937 */ 938 options = uap->options; 939 idtype = uap->idtype; 940 id = uap->id; 941 infop = uap->info ? &info : NULL; 942 943 switch(idtype) { 944 case P_PID: 945 case P_PGID: 946 if (id == WAIT_MYPGRP) { 947 idtype = P_PGID; 948 id = curproc->p_pgid; 949 } 950 break; 951 default: 952 /* let kern_wait deal with the remainder */ 953 break; 954 } 955 956 error = kern_wait(idtype, id, &status, options, 957 &wrusage, infop, &uap->sysmsg_result); 958 959 if (error == 0 && uap->status) 960 error = copyout(&status, uap->status, sizeof(*uap->status)); 961 if (error == 0 && uap->wrusage) 962 error = copyout(&wrusage, uap->wrusage, sizeof(*uap->wrusage)); 963 if (error == 0 && uap->info) 964 error = copyout(&info, uap->info, sizeof(*uap->info)); 965 return (error); 966 } 967 968 /* 969 * kernel wait*() system call support 970 */ 971 int 972 kern_wait(idtype_t idtype, id_t id, int *status, int options, 973 struct __wrusage *wrusage, struct __siginfo *info, int *res) 974 { 975 struct thread *td = curthread; 976 struct lwp *lp; 977 struct proc *q = td->td_proc; 978 struct proc *p, *t; 979 struct ucred *cr; 980 struct pargs *pa; 981 struct sigacts *ps; 982 int nfound, error; 983 long waitgen; 984 985 /* 986 * Must not have extraneous options. Must have at least one 987 * matchable option. 988 */ 989 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE|WSTOPPED| 990 WEXITED|WTRAPPED|WNOWAIT)) { 991 return (EINVAL); 992 } 993 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 994 return (EINVAL); 995 } 996 997 /* 998 * Protect the q->p_children list 999 */ 1000 lwkt_gettoken(&q->p_token); 1001 loop: 1002 /* 1003 * All sorts of things can change due to blocking so we have to loop 1004 * all the way back up here. 1005 * 1006 * The problem is that if a process group is stopped and the parent 1007 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 1008 * of the child and then stop itself when it tries to return from the 1009 * system call. When the process group is resumed the parent will 1010 * then get the STOP status even though the child has now resumed 1011 * (a followup wait*() will get the CONT status). 1012 * 1013 * Previously the CONT would overwrite the STOP because the tstop 1014 * was handled within tsleep(), and the parent would only see 1015 * the CONT when both are stopped and continued together. This little 1016 * two-line hack restores this effect. 1017 */ 1018 if (STOPLWP(q, td->td_lwp)) 1019 tstop(); 1020 1021 nfound = 0; 1022 1023 /* 1024 * Loop on children. 1025 * 1026 * NOTE: We don't want to break q's p_token in the loop for the 1027 * case where no children are found or we risk breaking the 1028 * interlock between child and parent. 1029 */ 1030 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000); 1031 LIST_FOREACH(p, &q->p_children, p_sibling) { 1032 /* 1033 * Filter, (p) will be held on fall-through. Try to optimize 1034 * this to avoid the atomic op until we are pretty sure we 1035 * want this process. 1036 */ 1037 switch(idtype) { 1038 case P_ALL: 1039 PHOLD(p); 1040 break; 1041 case P_PID: 1042 if (p->p_pid != (pid_t)id) 1043 continue; 1044 PHOLD(p); 1045 break; 1046 case P_PGID: 1047 if (p->p_pgid != (pid_t)id) 1048 continue; 1049 PHOLD(p); 1050 break; 1051 case P_SID: 1052 PHOLD(p); 1053 if (p->p_session && p->p_session->s_sid != (pid_t)id) { 1054 PRELE(p); 1055 continue; 1056 } 1057 break; 1058 case P_UID: 1059 PHOLD(p); 1060 if (p->p_ucred->cr_uid != (uid_t)id) { 1061 PRELE(p); 1062 continue; 1063 } 1064 break; 1065 case P_GID: 1066 PHOLD(p); 1067 if (p->p_ucred->cr_gid != (gid_t)id) { 1068 PRELE(p); 1069 continue; 1070 } 1071 break; 1072 case P_JAILID: 1073 PHOLD(p); 1074 if (p->p_ucred->cr_prison && 1075 p->p_ucred->cr_prison->pr_id != (int)id) { 1076 PRELE(p); 1077 continue; 1078 } 1079 break; 1080 default: 1081 /* unsupported filter */ 1082 continue; 1083 } 1084 /* (p) is held at this point */ 1085 1086 /* 1087 * This special case handles a kthread spawned by linux_clone 1088 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1089 * functions need to be able to distinguish between waiting 1090 * on a process and waiting on a thread. It is a thread if 1091 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1092 * signifies we want to wait for threads and not processes. 1093 */ 1094 if ((p->p_sigparent != SIGCHLD) ^ 1095 ((options & WLINUXCLONE) != 0)) { 1096 PRELE(p); 1097 continue; 1098 } 1099 1100 nfound++; 1101 if (p->p_stat == SZOMB && (options & WEXITED)) { 1102 /* 1103 * We may go into SZOMB with threads still present. 1104 * We must wait for them to exit before we can reap 1105 * the master thread, otherwise we may race reaping 1106 * non-master threads. 1107 * 1108 * Only this routine can remove a process from 1109 * the zombie list and destroy it. 1110 */ 1111 if (PHOLDZOMB(p)) { 1112 PRELE(p); 1113 goto loop; 1114 } 1115 lwkt_gettoken(&p->p_token); 1116 if (p->p_pptr != q) { 1117 lwkt_reltoken(&p->p_token); 1118 PRELE(p); 1119 PRELEZOMB(p); 1120 goto loop; 1121 } 1122 while (p->p_nthreads > 0) { 1123 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 1124 } 1125 1126 /* 1127 * Reap any LWPs left in p->p_lwps. This is usually 1128 * just the last LWP. This must be done before 1129 * we loop on p_lock since the lwps hold a ref on 1130 * it as a vmspace interlock. 1131 * 1132 * Once that is accomplished p_nthreads had better 1133 * be zero. 1134 */ 1135 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 1136 /* 1137 * Make sure no one is using this lwp, before 1138 * it is removed from the tree. If we didn't 1139 * wait it here, lwp tree iteration with 1140 * blocking operation would be broken. 1141 */ 1142 while (lp->lwp_lock > 0) 1143 tsleep(lp, 0, "zomblwp", 1); 1144 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 1145 reaplwp(lp); 1146 } 1147 KKASSERT(p->p_nthreads == 0); 1148 1149 /* 1150 * Don't do anything really bad until all references 1151 * to the process go away. This may include other 1152 * LWPs which are still in the process of being 1153 * reaped. We can't just pull the rug out from under 1154 * them because they may still be using the VM space. 1155 * 1156 * Certain kernel facilities such as /proc will also 1157 * put a hold on the process for short periods of 1158 * time. 1159 */ 1160 PRELE(p); /* from top of loop */ 1161 PSTALL(p, "reap3", 1); /* 1 ref (for PZOMBHOLD) */ 1162 1163 /* Take care of our return values. */ 1164 *res = p->p_pid; 1165 1166 *status = p->p_xstat; 1167 wrusage->wru_self = p->p_ru; 1168 wrusage->wru_children = p->p_cru; 1169 1170 if (info) { 1171 bzero(info, sizeof(*info)); 1172 info->si_errno = 0; 1173 info->si_signo = SIGCHLD; 1174 if (p->p_xstat) 1175 info->si_code = CLD_KILLED; 1176 else 1177 info->si_code = CLD_EXITED; 1178 info->si_status = p->p_xstat; 1179 info->si_pid = p->p_pid; 1180 info->si_uid = p->p_ucred->cr_uid; 1181 } 1182 1183 /* 1184 * WNOWAIT shortcuts to done here, leaving the 1185 * child on the zombie list. 1186 */ 1187 if (options & WNOWAIT) { 1188 lwkt_reltoken(&p->p_token); 1189 PRELEZOMB(p); 1190 error = 0; 1191 goto done; 1192 } 1193 1194 /* 1195 * If we got the child via a ptrace 'attach', 1196 * we need to give it back to the old parent. 1197 */ 1198 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 1199 p->p_oppid = 0; 1200 proc_reparent(p, t); 1201 ksignal(t, SIGCHLD); 1202 wakeup((caddr_t)t); 1203 PRELE(t); 1204 lwkt_reltoken(&p->p_token); 1205 PRELEZOMB(p); 1206 error = 0; 1207 goto done; 1208 } 1209 1210 /* 1211 * Unlink the proc from its process group so that 1212 * the following operations won't lead to an 1213 * inconsistent state for processes running down 1214 * the zombie list. 1215 */ 1216 proc_remove_zombie(p); 1217 proc_userunmap(p); 1218 lwkt_reltoken(&p->p_token); 1219 leavepgrp(p); 1220 1221 p->p_xstat = 0; 1222 ruadd(&q->p_cru, &p->p_ru); 1223 ruadd(&q->p_cru, &p->p_cru); 1224 1225 /* 1226 * Decrement the count of procs running with this uid. 1227 */ 1228 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 1229 1230 /* 1231 * Free up credentials. p_spin is required to 1232 * avoid races against allproc scans. 1233 */ 1234 spin_lock(&p->p_spin); 1235 cr = p->p_ucred; 1236 p->p_ucred = NULL; 1237 spin_unlock(&p->p_spin); 1238 crfree(cr); 1239 1240 /* 1241 * Remove unused arguments 1242 */ 1243 pa = p->p_args; 1244 p->p_args = NULL; 1245 if (pa && refcount_release(&pa->ar_ref)) { 1246 kfree(pa, M_PARGS); 1247 pa = NULL; 1248 } 1249 1250 ps = p->p_sigacts; 1251 p->p_sigacts = NULL; 1252 if (ps && refcount_release(&ps->ps_refcnt)) { 1253 kfree(ps, M_SUBPROC); 1254 ps = NULL; 1255 } 1256 1257 /* 1258 * Our exitingcount was incremented when the process 1259 * became a zombie, now that the process has been 1260 * removed from (almost) all lists we should be able 1261 * to safely destroy its vmspace. Wait for any current 1262 * holders to go away (so the vmspace remains stable), 1263 * then scrap it. 1264 * 1265 * NOTE: Releasing the parent process (q) p_token 1266 * across the vmspace_exitfree() call is 1267 * important here to reduce stalls on 1268 * interactions with (q) (such as 1269 * fork/exec/wait or 'ps'). 1270 */ 1271 PSTALL(p, "reap4", 1); 1272 lwkt_reltoken(&q->p_token); 1273 vmspace_exitfree(p); 1274 lwkt_gettoken(&q->p_token); 1275 PSTALL(p, "reap5", 1); 1276 1277 /* 1278 * NOTE: We have to officially release ZOMB in order 1279 * to ensure that a racing thread in kern_wait() 1280 * which blocked on ZOMB is woken up. 1281 */ 1282 PRELEZOMB(p); 1283 kfree(p->p_uidpcpu, M_SUBPROC); 1284 kfree(p, M_PROC); 1285 atomic_add_int(&nprocs, -1); 1286 error = 0; 1287 goto done; 1288 } 1289 1290 /* 1291 * Process has not yet exited 1292 */ 1293 if ((p->p_stat == SSTOP || p->p_stat == SCORE) && 1294 (p->p_flags & P_WAITED) == 0 && 1295 (((p->p_flags & P_TRACED) && (options & WTRAPPED)) || 1296 (options & WSTOPPED))) { 1297 lwkt_gettoken(&p->p_token); 1298 if (p->p_pptr != q) { 1299 lwkt_reltoken(&p->p_token); 1300 PRELE(p); 1301 goto loop; 1302 } 1303 if ((p->p_stat != SSTOP && p->p_stat != SCORE) || 1304 (p->p_flags & P_WAITED) != 0 || 1305 ((p->p_flags & P_TRACED) == 0 && 1306 (options & WUNTRACED) == 0)) { 1307 lwkt_reltoken(&p->p_token); 1308 PRELE(p); 1309 goto loop; 1310 } 1311 1312 /* 1313 * Don't set P_WAITED if WNOWAIT specified, leaving 1314 * the process in a waitable state. 1315 */ 1316 if ((options & WNOWAIT) == 0) 1317 p->p_flags |= P_WAITED; 1318 1319 *res = p->p_pid; 1320 *status = W_STOPCODE(p->p_xstat); 1321 /* Zero rusage so we get something consistent. */ 1322 bzero(wrusage, sizeof(*wrusage)); 1323 error = 0; 1324 if (info) { 1325 bzero(info, sizeof(*info)); 1326 if (p->p_flags & P_TRACED) 1327 info->si_code = CLD_TRAPPED; 1328 else 1329 info->si_code = CLD_STOPPED; 1330 info->si_status = p->p_xstat; 1331 } 1332 lwkt_reltoken(&p->p_token); 1333 PRELE(p); 1334 goto done; 1335 } 1336 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) { 1337 lwkt_gettoken(&p->p_token); 1338 if (p->p_pptr != q) { 1339 lwkt_reltoken(&p->p_token); 1340 PRELE(p); 1341 goto loop; 1342 } 1343 if ((p->p_flags & P_CONTINUED) == 0) { 1344 lwkt_reltoken(&p->p_token); 1345 PRELE(p); 1346 goto loop; 1347 } 1348 1349 *res = p->p_pid; 1350 1351 /* 1352 * Don't set P_WAITED if WNOWAIT specified, leaving 1353 * the process in a waitable state. 1354 */ 1355 if ((options & WNOWAIT) == 0) 1356 p->p_flags &= ~P_CONTINUED; 1357 1358 *status = SIGCONT; 1359 error = 0; 1360 if (info) { 1361 bzero(info, sizeof(*info)); 1362 info->si_code = CLD_CONTINUED; 1363 info->si_status = p->p_xstat; 1364 } 1365 lwkt_reltoken(&p->p_token); 1366 PRELE(p); 1367 goto done; 1368 } 1369 PRELE(p); 1370 } 1371 if (nfound == 0) { 1372 error = ECHILD; 1373 goto done; 1374 } 1375 if (options & WNOHANG) { 1376 *res = 0; 1377 error = 0; 1378 goto done; 1379 } 1380 1381 /* 1382 * Wait for signal - interlocked using q->p_waitgen. 1383 */ 1384 error = 0; 1385 while ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) { 1386 tsleep_interlock(q, PCATCH); 1387 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000); 1388 if ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) { 1389 error = tsleep(q, PCATCH | PINTERLOCKED, "wait", 0); 1390 break; 1391 } 1392 } 1393 if (error) { 1394 done: 1395 lwkt_reltoken(&q->p_token); 1396 return (error); 1397 } 1398 goto loop; 1399 } 1400 1401 /* 1402 * Change child's parent process to parent. 1403 * 1404 * p_children/p_sibling requires the parent's token, and 1405 * changing pptr requires the child's token, so we have to 1406 * get three tokens to do this operation. We also need to 1407 * hold pointers that might get ripped out from under us to 1408 * preserve structural integrity. 1409 * 1410 * It is possible to race another reparent or disconnect or other 1411 * similar operation. We must retry when this situation occurs. 1412 * Once we successfully reparent the process we no longer care 1413 * about any races. 1414 */ 1415 void 1416 proc_reparent(struct proc *child, struct proc *parent) 1417 { 1418 struct proc *opp; 1419 1420 PHOLD(parent); 1421 while ((opp = child->p_pptr) != parent) { 1422 PHOLD(opp); 1423 lwkt_gettoken(&opp->p_token); 1424 lwkt_gettoken(&child->p_token); 1425 lwkt_gettoken(&parent->p_token); 1426 if (child->p_pptr != opp) { 1427 lwkt_reltoken(&parent->p_token); 1428 lwkt_reltoken(&child->p_token); 1429 lwkt_reltoken(&opp->p_token); 1430 PRELE(opp); 1431 continue; 1432 } 1433 LIST_REMOVE(child, p_sibling); 1434 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1435 child->p_pptr = parent; 1436 child->p_ppid = parent->p_pid; 1437 lwkt_reltoken(&parent->p_token); 1438 lwkt_reltoken(&child->p_token); 1439 lwkt_reltoken(&opp->p_token); 1440 if (LIST_EMPTY(&opp->p_children)) 1441 wakeup(opp); 1442 PRELE(opp); 1443 break; 1444 } 1445 PRELE(parent); 1446 } 1447 1448 /* 1449 * The next two functions are to handle adding/deleting items on the 1450 * exit callout list 1451 * 1452 * at_exit(): 1453 * Take the arguments given and put them onto the exit callout list, 1454 * However first make sure that it's not already there. 1455 * returns 0 on success. 1456 */ 1457 1458 int 1459 at_exit(exitlist_fn function) 1460 { 1461 struct exitlist *ep; 1462 1463 #ifdef INVARIANTS 1464 /* Be noisy if the programmer has lost track of things */ 1465 if (rm_at_exit(function)) 1466 kprintf("WARNING: exit callout entry (%p) already present\n", 1467 function); 1468 #endif 1469 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 1470 if (ep == NULL) 1471 return (ENOMEM); 1472 ep->function = function; 1473 TAILQ_INSERT_TAIL(&exit_list, ep, next); 1474 return (0); 1475 } 1476 1477 /* 1478 * Scan the exit callout list for the given item and remove it. 1479 * Returns the number of items removed (0 or 1) 1480 */ 1481 int 1482 rm_at_exit(exitlist_fn function) 1483 { 1484 struct exitlist *ep; 1485 1486 TAILQ_FOREACH(ep, &exit_list, next) { 1487 if (ep->function == function) { 1488 TAILQ_REMOVE(&exit_list, ep, next); 1489 kfree(ep, M_ATEXIT); 1490 return(1); 1491 } 1492 } 1493 return (0); 1494 } 1495 1496 /* 1497 * LWP reaper related code. 1498 */ 1499 static void 1500 reaplwps(void *context, int dummy) 1501 { 1502 struct lwplist *lwplist = context; 1503 struct lwp *lp; 1504 int cpu = mycpuid; 1505 1506 lwkt_gettoken(&deadlwp_token[cpu]); 1507 while ((lp = LIST_FIRST(lwplist))) { 1508 LIST_REMOVE(lp, u.lwp_reap_entry); 1509 reaplwp(lp); 1510 } 1511 lwkt_reltoken(&deadlwp_token[cpu]); 1512 } 1513 1514 static void 1515 reaplwp(struct lwp *lp) 1516 { 1517 while (lwp_wait(lp) == 0) 1518 ; 1519 lwp_dispose(lp); 1520 } 1521 1522 static void 1523 deadlwp_init(void) 1524 { 1525 int cpu; 1526 1527 for (cpu = 0; cpu < ncpus; cpu++) { 1528 lwkt_token_init(&deadlwp_token[cpu], "deadlwpl"); 1529 LIST_INIT(&deadlwp_list[cpu]); 1530 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), 1531 M_DEVBUF, M_WAITOK); 1532 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1533 } 1534 } 1535 1536 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1537