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 72 #include <sys/refcount.h> 73 #include <sys/spinlock2.h> 74 #include <sys/mplock2.h> 75 76 #include <machine/vmm.h> 77 78 static void reaplwps(void *context, int dummy); 79 static void reaplwp(struct lwp *lp); 80 static void killlwps(struct lwp *lp); 81 82 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 83 84 /* 85 * callout list for things to do at exit time 86 */ 87 struct exitlist { 88 exitlist_fn function; 89 TAILQ_ENTRY(exitlist) next; 90 }; 91 92 TAILQ_HEAD(exit_list_head, exitlist); 93 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 94 95 /* 96 * LWP reaper data 97 */ 98 static struct task *deadlwp_task[MAXCPU]; 99 static struct lwplist deadlwp_list[MAXCPU]; 100 static struct lwkt_token deadlwp_token[MAXCPU]; 101 102 /* 103 * exit -- 104 * Death of process. 105 * 106 * SYS_EXIT_ARGS(int rval) 107 */ 108 int 109 sys_exit(struct exit_args *uap) 110 { 111 exit1(W_EXITCODE(uap->rval, 0)); 112 /* NOTREACHED */ 113 } 114 115 /* 116 * Extended exit -- 117 * Death of a lwp or process with optional bells and whistles. 118 */ 119 int 120 sys_extexit(struct extexit_args *uap) 121 { 122 struct proc *p = curproc; 123 int action, who; 124 int error; 125 126 action = EXTEXIT_ACTION(uap->how); 127 who = EXTEXIT_WHO(uap->how); 128 129 /* Check parameters before we might perform some action */ 130 switch (who) { 131 case EXTEXIT_PROC: 132 case EXTEXIT_LWP: 133 break; 134 default: 135 return (EINVAL); 136 } 137 138 switch (action) { 139 case EXTEXIT_SIMPLE: 140 break; 141 case EXTEXIT_SETINT: 142 error = copyout(&uap->status, uap->addr, sizeof(uap->status)); 143 if (error) 144 return (error); 145 break; 146 default: 147 return (EINVAL); 148 } 149 150 lwkt_gettoken(&p->p_token); 151 152 switch (who) { 153 case EXTEXIT_LWP: 154 /* 155 * Be sure only to perform a simple lwp exit if there is at 156 * least one more lwp in the proc, which will call exit1() 157 * later, otherwise the proc will be an UNDEAD and not even a 158 * SZOMB! 159 */ 160 if (p->p_nthreads > 1) { 161 lwp_exit(0, NULL); /* called w/ p_token held */ 162 /* NOT REACHED */ 163 } 164 /* else last lwp in proc: do the real thing */ 165 /* FALLTHROUGH */ 166 default: /* to help gcc */ 167 case EXTEXIT_PROC: 168 lwkt_reltoken(&p->p_token); 169 exit1(W_EXITCODE(uap->status, 0)); 170 /* NOTREACHED */ 171 } 172 173 /* NOTREACHED */ 174 lwkt_reltoken(&p->p_token); /* safety */ 175 } 176 177 /* 178 * Kill all lwps associated with the current process except the 179 * current lwp. Return an error if we race another thread trying to 180 * do the same thing and lose the race. 181 * 182 * If forexec is non-zero the current thread and process flags are 183 * cleaned up so they can be reused. 184 */ 185 int 186 killalllwps(int forexec) 187 { 188 struct lwp *lp = curthread->td_lwp; 189 struct proc *p = lp->lwp_proc; 190 int fakestop; 191 192 /* 193 * Interlock against P_WEXIT. Only one of the process's thread 194 * is allowed to do the master exit. 195 */ 196 lwkt_gettoken(&p->p_token); 197 if (p->p_flags & P_WEXIT) { 198 lwkt_reltoken(&p->p_token); 199 return (EALREADY); 200 } 201 p->p_flags |= P_WEXIT; 202 lwkt_gettoken(&lp->lwp_token); 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 && (lp->lwp_mpflags & LWP_MP_WSTOP)) { 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 lwkt_reltoken(&lp->lwp_token); 242 lwkt_reltoken(&p->p_token); 243 244 return(0); 245 } 246 247 /* 248 * Kill all LWPs except the current one. Do not try to signal 249 * LWPs which have exited on their own or have already been 250 * signaled. 251 */ 252 static void 253 killlwps(struct lwp *lp) 254 { 255 struct proc *p = lp->lwp_proc; 256 struct lwp *tlp; 257 258 /* 259 * Kill the remaining LWPs. We must send the signal before setting 260 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce 261 * races. tlp must be held across the call as it might block and 262 * allow the target lwp to rip itself out from under our loop. 263 */ 264 FOREACH_LWP_IN_PROC(tlp, p) { 265 LWPHOLD(tlp); 266 lwkt_gettoken(&tlp->lwp_token); 267 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) { 268 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT); 269 lwpsignal(p, tlp, SIGKILL); 270 } 271 lwkt_reltoken(&tlp->lwp_token); 272 LWPRELE(tlp); 273 } 274 275 /* 276 * Wait for everything to clear out. Also make sure any tstop()s 277 * are signalled (we are holding p_token for the interlock). 278 */ 279 wakeup(p); 280 while (p->p_nthreads > 1) 281 tsleep(&p->p_nthreads, 0, "killlwps", 0); 282 } 283 284 /* 285 * Exit: deallocate address space and other resources, change proc state 286 * to zombie, and unlink proc from allproc and parent's lists. Save exit 287 * status and rusage for wait(). Check for child processes and orphan them. 288 */ 289 void 290 exit1(int rv) 291 { 292 struct thread *td = curthread; 293 struct proc *p = td->td_proc; 294 struct lwp *lp = td->td_lwp; 295 struct proc *q; 296 struct proc *pp; 297 struct proc *reproc; 298 struct sysreaper *reap; 299 struct vmspace *vm; 300 struct vnode *vtmp; 301 struct exitlist *ep; 302 int error; 303 304 lwkt_gettoken(&p->p_token); 305 306 if (p->p_pid == 1) { 307 kprintf("init died (signal %d, exit %d)\n", 308 WTERMSIG(rv), WEXITSTATUS(rv)); 309 panic("Going nowhere without my init!"); 310 } 311 varsymset_clean(&p->p_varsymset); 312 lockuninit(&p->p_varsymset.vx_lock); 313 314 /* 315 * Kill all lwps associated with the current process, return an 316 * error if we race another thread trying to do the same thing 317 * and lose the race. 318 */ 319 error = killalllwps(0); 320 if (error) { 321 lwp_exit(0, NULL); 322 /* NOT REACHED */ 323 } 324 325 /* are we a task leader? */ 326 if (p == p->p_leader) { 327 struct kill_args killArgs; 328 killArgs.signum = SIGKILL; 329 q = p->p_peers; 330 while(q) { 331 killArgs.pid = q->p_pid; 332 /* 333 * The interface for kill is better 334 * than the internal signal 335 */ 336 sys_kill(&killArgs); 337 q = q->p_peers; 338 } 339 while (p->p_peers) 340 tsleep((caddr_t)p, 0, "exit1", 0); 341 } 342 343 #ifdef PGINPROF 344 vmsizmon(); 345 #endif 346 STOPEVENT(p, S_EXIT, rv); 347 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */ 348 349 /* 350 * Check if any loadable modules need anything done at process exit. 351 * e.g. SYSV IPC stuff 352 * XXX what if one of these generates an error? 353 */ 354 p->p_xstat = rv; 355 356 /* 357 * XXX: imho, the eventhandler stuff is much cleaner than this. 358 * Maybe we should move everything to use eventhandler. 359 */ 360 TAILQ_FOREACH(ep, &exit_list, next) 361 (*ep->function)(td); 362 363 if (p->p_flags & P_PROFIL) 364 stopprofclock(p); 365 366 SIGEMPTYSET(p->p_siglist); 367 SIGEMPTYSET(lp->lwp_siglist); 368 if (timevalisset(&p->p_realtimer.it_value)) 369 callout_terminate(&p->p_ithandle); 370 371 /* 372 * Reset any sigio structures pointing to us as a result of 373 * F_SETOWN with our pid. 374 */ 375 funsetownlst(&p->p_sigiolst); 376 377 /* 378 * Close open files and release open-file table. 379 * This may block! 380 */ 381 fdfree(p, NULL); 382 383 if (p->p_leader->p_peers) { 384 q = p->p_leader; 385 while(q->p_peers != p) 386 q = q->p_peers; 387 q->p_peers = p->p_peers; 388 wakeup((caddr_t)p->p_leader); 389 } 390 391 /* 392 * XXX Shutdown SYSV semaphores 393 */ 394 semexit(p); 395 396 /* The next two chunks should probably be moved to vmspace_exit. */ 397 vm = p->p_vmspace; 398 399 /* 400 * Clean up data related to virtual kernel operation. Clean up 401 * any vkernel context related to the current lwp now so we can 402 * destroy p_vkernel. 403 */ 404 if (p->p_vkernel) { 405 vkernel_lwp_exit(lp); 406 vkernel_exit(p); 407 } 408 409 /* 410 * Release the user portion of address space. The exitbump prevents 411 * the vmspace from being completely eradicated (using holdcnt). 412 * This releases references to vnodes, which could cause I/O if the 413 * file has been unlinked. We need to do this early enough that 414 * we can still sleep. 415 * 416 * We can't free the entire vmspace as the kernel stack may be mapped 417 * within that space also. 418 * 419 * Processes sharing the same vmspace may exit in one order, and 420 * get cleaned up by vmspace_exit() in a different order. The 421 * last exiting process to reach this point releases as much of 422 * the environment as it can, and the last process cleaned up 423 * by vmspace_exit() (which decrements exitingcnt) cleans up the 424 * remainder. 425 * 426 * NOTE: Releasing p_token around this call is helpful if the 427 * vmspace had a huge RSS. Otherwise some other process 428 * trying to do an allproc or other scan (like 'ps') may 429 * stall for a long time. 430 */ 431 lwkt_reltoken(&p->p_token); 432 vmspace_relexit(vm); 433 lwkt_gettoken(&p->p_token); 434 435 if (SESS_LEADER(p)) { 436 struct session *sp = p->p_session; 437 438 if (sp->s_ttyvp) { 439 /* 440 * We are the controlling process. Signal the 441 * foreground process group, drain the controlling 442 * terminal, and revoke access to the controlling 443 * terminal. 444 * 445 * NOTE: while waiting for the process group to exit 446 * it is possible that one of the processes in the 447 * group will revoke the tty, so the ttyclosesession() 448 * function will re-check sp->s_ttyvp. 449 */ 450 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 451 if (sp->s_ttyp->t_pgrp) 452 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 453 ttywait(sp->s_ttyp); 454 ttyclosesession(sp, 1); /* also revoke */ 455 } 456 /* 457 * Release the tty. If someone has it open via 458 * /dev/tty then close it (since they no longer can 459 * once we've NULL'd it out). 460 */ 461 ttyclosesession(sp, 0); 462 463 /* 464 * s_ttyp is not zero'd; we use this to indicate 465 * that the session once had a controlling terminal. 466 * (for logging and informational purposes) 467 */ 468 } 469 sp->s_leader = NULL; 470 } 471 fixjobc(p, p->p_pgrp, 0); 472 (void)acct_process(p); 473 #ifdef KTRACE 474 /* 475 * release trace file 476 */ 477 if (p->p_tracenode) 478 ktrdestroy(&p->p_tracenode); 479 p->p_traceflag = 0; 480 #endif 481 /* 482 * Release reference to text vnode 483 */ 484 if ((vtmp = p->p_textvp) != NULL) { 485 p->p_textvp = NULL; 486 vrele(vtmp); 487 } 488 489 /* Release namecache handle to text file */ 490 if (p->p_textnch.ncp) 491 cache_drop(&p->p_textnch); 492 493 /* 494 * We have to handle PPWAIT here or proc_move_allproc_zombie() 495 * will block on the PHOLD() the parent is doing. 496 * 497 * We are using the flag as an interlock so an atomic op is 498 * necessary to synchronize with the parent's cpu. 499 */ 500 if (p->p_flags & P_PPWAIT) { 501 if (p->p_pptr && p->p_pptr->p_upmap) 502 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1); 503 atomic_clear_int(&p->p_flags, P_PPWAIT); 504 wakeup(p->p_pptr); 505 } 506 507 /* 508 * Move the process to the zombie list. This will block 509 * until the process p_lock count reaches 0. The process will 510 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 511 * which is called from cpu_proc_exit(). 512 * 513 * Interlock against waiters using p_waitgen. We increment 514 * p_waitgen after completing the move of our process to the 515 * zombie list. 516 * 517 * WARNING: pp becomes stale when we block, clear it now as a 518 * reminder. 519 */ 520 proc_move_allproc_zombie(p); 521 pp = p->p_pptr; 522 atomic_add_long(&pp->p_waitgen, 1); 523 pp = NULL; 524 525 /* 526 * release controlled reaper for exit if we own it and return the 527 * remaining reaper (the one for us), which we will drop after we 528 * are done. 529 */ 530 reap = reaper_exit(p); 531 532 /* 533 * Reparent all of this process's children to the init process or 534 * to the designated reaper. We must hold the reaper's p_token in 535 * order to safely mess with p_children. 536 * 537 * We already hold p->p_token (to remove the children from our list). 538 */ 539 reproc = NULL; 540 q = LIST_FIRST(&p->p_children); 541 if (q) { 542 reproc = reaper_get(reap); 543 lwkt_gettoken(&reproc->p_token); 544 while ((q = LIST_FIRST(&p->p_children)) != NULL) { 545 PHOLD(q); 546 lwkt_gettoken(&q->p_token); 547 if (q != LIST_FIRST(&p->p_children)) { 548 lwkt_reltoken(&q->p_token); 549 PRELE(q); 550 continue; 551 } 552 LIST_REMOVE(q, p_sibling); 553 LIST_INSERT_HEAD(&reproc->p_children, q, p_sibling); 554 q->p_pptr = reproc; 555 q->p_ppid = reproc->p_pid; 556 q->p_sigparent = SIGCHLD; 557 558 /* 559 * Traced processes are killed 560 * since their existence means someone is screwing up. 561 */ 562 if (q->p_flags & P_TRACED) { 563 q->p_flags &= ~P_TRACED; 564 ksignal(q, SIGKILL); 565 } 566 lwkt_reltoken(&q->p_token); 567 PRELE(q); 568 } 569 lwkt_reltoken(&reproc->p_token); 570 wakeup(reproc); 571 } 572 573 /* 574 * Save exit status and final rusage info. We no longer add 575 * child rusage info into self times, wait4() and kern_wait() 576 * handles it in order to properly support wait6(). 577 */ 578 calcru_proc(p, &p->p_ru); 579 /*ruadd(&p->p_ru, &p->p_cru); REMOVED */ 580 581 /* 582 * notify interested parties of our demise. 583 */ 584 KNOTE(&p->p_klist, NOTE_EXIT); 585 586 /* 587 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 588 * flag set, or if the handler is set to SIG_IGN, notify the reaper 589 * instead (it will handle this situation). 590 * 591 * NOTE: The reaper can still be the parent process. 592 * 593 * (must reload pp) 594 */ 595 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 596 if (reproc == NULL) 597 reproc = reaper_get(reap); 598 proc_reparent(p, reproc); 599 } 600 if (reproc) 601 PRELE(reproc); 602 if (reap) 603 reaper_drop(reap); 604 605 /* 606 * Signal (possibly new) parent. 607 */ 608 pp = p->p_pptr; 609 PHOLD(pp); 610 if (p->p_sigparent && pp != initproc) { 611 int sig = p->p_sigparent; 612 613 if (sig != SIGUSR1 && sig != SIGCHLD) 614 sig = SIGCHLD; 615 ksignal(pp, sig); 616 } else { 617 ksignal(pp, SIGCHLD); 618 } 619 p->p_flags &= ~P_TRACED; 620 PRELE(pp); 621 622 /* 623 * cpu_exit is responsible for clearing curproc, since 624 * it is heavily integrated with the thread/switching sequence. 625 * 626 * Other substructures are freed from wait(). 627 */ 628 if (p->p_limit) { 629 struct plimit *rlimit; 630 631 rlimit = p->p_limit; 632 p->p_limit = NULL; 633 plimit_free(rlimit); 634 } 635 636 /* 637 * Finally, call machine-dependent code to release as many of the 638 * lwp's resources as we can and halt execution of this thread. 639 * 640 * pp is a wild pointer now but still the correct wakeup() target. 641 * lwp_exit() only uses it to send the wakeup() signal to the likely 642 * parent. Any reparenting race that occurs will get a signal 643 * automatically and not be an issue. 644 */ 645 lwp_exit(1, pp); 646 } 647 648 /* 649 * Eventually called by every exiting LWP 650 * 651 * p->p_token must be held. mplock may be held and will be released. 652 */ 653 void 654 lwp_exit(int masterexit, void *waddr) 655 { 656 struct thread *td = curthread; 657 struct lwp *lp = td->td_lwp; 658 struct proc *p = lp->lwp_proc; 659 int dowake = 0; 660 661 /* 662 * Release the current user process designation on the process so 663 * the userland scheduler can work in someone else. 664 */ 665 p->p_usched->release_curproc(lp); 666 667 /* 668 * lwp_exit() may be called without setting LWP_MP_WEXIT, so 669 * make sure it is set here. 670 */ 671 ASSERT_LWKT_TOKEN_HELD(&p->p_token); 672 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 673 674 /* 675 * Clean up any virtualization 676 */ 677 if (lp->lwp_vkernel) 678 vkernel_lwp_exit(lp); 679 680 if (td->td_vmm) 681 vmm_vmdestroy(); 682 683 /* 684 * Clean up select/poll support 685 */ 686 kqueue_terminate(&lp->lwp_kqueue); 687 688 /* 689 * Clean up any syscall-cached ucred or rlimit. 690 */ 691 if (td->td_ucred) { 692 crfree(td->td_ucred); 693 td->td_ucred = NULL; 694 } 695 if (td->td_limit) { 696 struct plimit *rlimit; 697 698 rlimit = td->td_limit; 699 td->td_limit = NULL; 700 plimit_free(rlimit); 701 } 702 703 /* 704 * Cleanup any cached descriptors for this thread 705 */ 706 if (p->p_fd) 707 fexitcache(td); 708 709 /* 710 * Nobody actually wakes us when the lock 711 * count reaches zero, so just wait one tick. 712 */ 713 while (lp->lwp_lock > 0) 714 tsleep(lp, 0, "lwpexit", 1); 715 716 /* Hand down resource usage to our proc */ 717 ruadd(&p->p_ru, &lp->lwp_ru); 718 719 /* 720 * If we don't hold the process until the LWP is reaped wait*() 721 * may try to dispose of its vmspace before all the LWPs have 722 * actually terminated. 723 */ 724 PHOLD(p); 725 726 /* 727 * Do any remaining work that might block on us. We should be 728 * coded such that further blocking is ok after decrementing 729 * p_nthreads but don't take the chance. 730 */ 731 dsched_exit_thread(td); 732 biosched_done(curthread); 733 734 /* 735 * We have to use the reaper for all the LWPs except the one doing 736 * the master exit. The LWP doing the master exit can just be 737 * left on p_lwps and the process reaper will deal with it 738 * synchronously, which is much faster. 739 * 740 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0. 741 * 742 * The process is left held until the reaper calls lwp_dispose() on 743 * the lp (after calling lwp_wait()). 744 */ 745 if (masterexit == 0) { 746 int cpu = mycpuid; 747 748 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 749 --p->p_nthreads; 750 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1) 751 dowake = 1; 752 lwkt_gettoken(&deadlwp_token[cpu]); 753 LIST_INSERT_HEAD(&deadlwp_list[cpu], lp, u.lwp_reap_entry); 754 taskqueue_enqueue(taskqueue_thread[cpu], deadlwp_task[cpu]); 755 lwkt_reltoken(&deadlwp_token[cpu]); 756 } else { 757 --p->p_nthreads; 758 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1) 759 dowake = 1; 760 } 761 762 /* 763 * We no longer need p_token. 764 * 765 * Tell the userland scheduler that we are going away 766 */ 767 lwkt_reltoken(&p->p_token); 768 p->p_usched->heuristic_exiting(lp, p); 769 770 /* 771 * Issue late wakeups after releasing our token to give us a chance 772 * to deschedule and switch away before another cpu in a wait*() 773 * reaps us. This is done as late as possible to reduce contention. 774 */ 775 if (dowake) 776 wakeup(&p->p_nthreads); 777 if (waddr) 778 wakeup(waddr); 779 780 cpu_lwp_exit(); 781 } 782 783 /* 784 * Wait until a lwp is completely dead. The final interlock in this drama 785 * is when TDF_EXITING is set in cpu_thread_exit() just before the final 786 * switchout. 787 * 788 * At the point TDF_EXITING is set a complete exit is accomplished when 789 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two 790 * post-switch interlock flags that can be used to wait for the TDF_ 791 * flags to clear. 792 * 793 * Returns non-zero on success, and zero if the caller needs to retry 794 * the lwp_wait(). 795 */ 796 static int 797 lwp_wait(struct lwp *lp) 798 { 799 struct thread *td = lp->lwp_thread; 800 u_int mpflags; 801 802 KKASSERT(lwkt_preempted_proc() != lp); 803 804 /* 805 * This bit of code uses the thread destruction interlock 806 * managed by lwkt_switch_return() to wait for the lwp's 807 * thread to completely disengage. 808 * 809 * It is possible for us to race another cpu core so we 810 * have to do this correctly. 811 */ 812 for (;;) { 813 mpflags = td->td_mpflags; 814 cpu_ccfence(); 815 if (mpflags & TDF_MP_EXITSIG) 816 break; 817 tsleep_interlock(td, 0); 818 if (atomic_cmpset_int(&td->td_mpflags, mpflags, 819 mpflags | TDF_MP_EXITWAIT)) { 820 tsleep(td, PINTERLOCKED, "lwpxt", 0); 821 } 822 } 823 824 /* 825 * We've already waited for the core exit but there can still 826 * be other refs from e.g. process scans and such. 827 */ 828 if (lp->lwp_lock > 0) { 829 tsleep(lp, 0, "lwpwait1", 1); 830 return(0); 831 } 832 if (td->td_refs) { 833 tsleep(td, 0, "lwpwait2", 1); 834 return(0); 835 } 836 837 /* 838 * Now that we have the thread destruction interlock these flags 839 * really should already be cleaned up, keep a check for safety. 840 * 841 * We can't rip its stack out from under it until TDF_EXITING is 842 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 843 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 844 * will be cleared temporarily if a thread gets preempted. 845 */ 846 while ((td->td_flags & (TDF_RUNNING | 847 TDF_RUNQ | 848 TDF_PREEMPT_LOCK | 849 TDF_EXITING)) != TDF_EXITING) { 850 tsleep(lp, 0, "lwpwait3", 1); 851 return (0); 852 } 853 854 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0, 855 ("lwp_wait: td %p (%s) still on run or sleep queue", 856 td, td->td_comm)); 857 return (1); 858 } 859 860 /* 861 * Release the resources associated with a lwp. 862 * The lwp must be completely dead. 863 */ 864 void 865 lwp_dispose(struct lwp *lp) 866 { 867 struct thread *td = lp->lwp_thread; 868 869 KKASSERT(lwkt_preempted_proc() != lp); 870 KKASSERT(lp->lwp_lock == 0); 871 KKASSERT(td->td_refs == 0); 872 KKASSERT((td->td_flags & (TDF_RUNNING | 873 TDF_RUNQ | 874 TDF_PREEMPT_LOCK | 875 TDF_EXITING)) == TDF_EXITING); 876 877 PRELE(lp->lwp_proc); 878 lp->lwp_proc = NULL; 879 if (td != NULL) { 880 td->td_proc = NULL; 881 td->td_lwp = NULL; 882 lp->lwp_thread = NULL; 883 lwkt_free_thread(td); 884 } 885 kfree(lp, M_LWP); 886 } 887 888 int 889 sys_wait4(struct wait_args *uap) 890 { 891 struct __wrusage wrusage; 892 int error; 893 int status; 894 int options; 895 id_t id; 896 idtype_t idtype; 897 898 options = uap->options | WEXITED | WTRAPPED; 899 id = uap->pid; 900 901 if (id == WAIT_ANY) { 902 idtype = P_ALL; 903 } else if (id == WAIT_MYPGRP) { 904 idtype = P_PGID; 905 id = curproc->p_pgid; 906 } else if (id < 0) { 907 idtype = P_PGID; 908 id = -id; 909 } else { 910 idtype = P_PID; 911 } 912 913 error = kern_wait(idtype, id, &status, options, &wrusage, 914 NULL, &uap->sysmsg_result); 915 916 if (error == 0 && uap->status) 917 error = copyout(&status, uap->status, sizeof(*uap->status)); 918 if (error == 0 && uap->rusage) { 919 ruadd(&wrusage.wru_self, &wrusage.wru_children); 920 error = copyout(&wrusage.wru_self, uap->rusage, sizeof(*uap->rusage)); 921 } 922 return (error); 923 } 924 925 int 926 sys_wait6(struct wait6_args *uap) 927 { 928 struct __wrusage wrusage; 929 struct __siginfo info; 930 struct __siginfo *infop; 931 int error; 932 int status; 933 int options; 934 id_t id; 935 idtype_t idtype; 936 937 /* 938 * NOTE: wait6() requires WEXITED and WTRAPPED to be specified if 939 * desired. 940 */ 941 options = uap->options; 942 idtype = uap->idtype; 943 id = uap->id; 944 infop = uap->info ? &info : NULL; 945 946 switch(idtype) { 947 case P_PID: 948 case P_PGID: 949 if (id == WAIT_MYPGRP) { 950 idtype = P_PGID; 951 id = curproc->p_pgid; 952 } 953 break; 954 default: 955 /* let kern_wait deal with the remainder */ 956 break; 957 } 958 959 error = kern_wait(idtype, id, &status, options, 960 &wrusage, infop, &uap->sysmsg_result); 961 962 if (error == 0 && uap->status) 963 error = copyout(&status, uap->status, sizeof(*uap->status)); 964 if (error == 0 && uap->wrusage) 965 error = copyout(&wrusage, uap->wrusage, sizeof(*uap->wrusage)); 966 if (error == 0 && uap->info) 967 error = copyout(&info, uap->info, sizeof(*uap->info)); 968 return (error); 969 } 970 971 /* 972 * kernel wait*() system call support 973 */ 974 int 975 kern_wait(idtype_t idtype, id_t id, int *status, int options, 976 struct __wrusage *wrusage, struct __siginfo *info, int *res) 977 { 978 struct thread *td = curthread; 979 struct lwp *lp; 980 struct proc *q = td->td_proc; 981 struct proc *p, *t; 982 struct ucred *cr; 983 struct pargs *pa; 984 struct sigacts *ps; 985 int nfound, error; 986 long waitgen; 987 988 /* 989 * Must not have extraneous options. Must have at least one 990 * matchable option. 991 */ 992 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE|WSTOPPED| 993 WEXITED|WTRAPPED|WNOWAIT)) { 994 return (EINVAL); 995 } 996 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 997 return (EINVAL); 998 } 999 1000 /* 1001 * Protect the q->p_children list 1002 */ 1003 lwkt_gettoken(&q->p_token); 1004 loop: 1005 /* 1006 * All sorts of things can change due to blocking so we have to loop 1007 * all the way back up here. 1008 * 1009 * The problem is that if a process group is stopped and the parent 1010 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 1011 * of the child and then stop itself when it tries to return from the 1012 * system call. When the process group is resumed the parent will 1013 * then get the STOP status even though the child has now resumed 1014 * (a followup wait*() will get the CONT status). 1015 * 1016 * Previously the CONT would overwrite the STOP because the tstop 1017 * was handled within tsleep(), and the parent would only see 1018 * the CONT when both are stopped and continued together. This little 1019 * two-line hack restores this effect. 1020 */ 1021 if (STOPLWP(q, td->td_lwp)) 1022 tstop(); 1023 1024 nfound = 0; 1025 1026 /* 1027 * Loop on children. 1028 * 1029 * NOTE: We don't want to break q's p_token in the loop for the 1030 * case where no children are found or we risk breaking the 1031 * interlock between child and parent. 1032 */ 1033 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000); 1034 LIST_FOREACH(p, &q->p_children, p_sibling) { 1035 /* 1036 * Filter, (p) will be held on fall-through. Try to optimize 1037 * this to avoid the atomic op until we are pretty sure we 1038 * want this process. 1039 */ 1040 switch(idtype) { 1041 case P_ALL: 1042 PHOLD(p); 1043 break; 1044 case P_PID: 1045 if (p->p_pid != (pid_t)id) 1046 continue; 1047 PHOLD(p); 1048 break; 1049 case P_PGID: 1050 if (p->p_pgid != (pid_t)id) 1051 continue; 1052 PHOLD(p); 1053 break; 1054 case P_SID: 1055 PHOLD(p); 1056 if (p->p_session && p->p_session->s_sid != (pid_t)id) { 1057 PRELE(p); 1058 continue; 1059 } 1060 break; 1061 case P_UID: 1062 PHOLD(p); 1063 if (p->p_ucred->cr_uid != (uid_t)id) { 1064 PRELE(p); 1065 continue; 1066 } 1067 break; 1068 case P_GID: 1069 PHOLD(p); 1070 if (p->p_ucred->cr_gid != (gid_t)id) { 1071 PRELE(p); 1072 continue; 1073 } 1074 break; 1075 case P_JAILID: 1076 PHOLD(p); 1077 if (p->p_ucred->cr_prison && 1078 p->p_ucred->cr_prison->pr_id != (int)id) { 1079 PRELE(p); 1080 continue; 1081 } 1082 break; 1083 default: 1084 /* unsupported filter */ 1085 continue; 1086 } 1087 /* (p) is held at this point */ 1088 1089 /* 1090 * This special case handles a kthread spawned by linux_clone 1091 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1092 * functions need to be able to distinguish between waiting 1093 * on a process and waiting on a thread. It is a thread if 1094 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1095 * signifies we want to wait for threads and not processes. 1096 */ 1097 if ((p->p_sigparent != SIGCHLD) ^ 1098 ((options & WLINUXCLONE) != 0)) { 1099 PRELE(p); 1100 continue; 1101 } 1102 1103 nfound++; 1104 if (p->p_stat == SZOMB && (options & WEXITED)) { 1105 /* 1106 * We may go into SZOMB with threads still present. 1107 * We must wait for them to exit before we can reap 1108 * the master thread, otherwise we may race reaping 1109 * non-master threads. 1110 * 1111 * Only this routine can remove a process from 1112 * the zombie list and destroy it. 1113 */ 1114 if (PHOLDZOMB(p)) { 1115 PRELE(p); 1116 goto loop; 1117 } 1118 lwkt_gettoken(&p->p_token); 1119 if (p->p_pptr != q) { 1120 lwkt_reltoken(&p->p_token); 1121 PRELE(p); 1122 PRELEZOMB(p); 1123 goto loop; 1124 } 1125 while (p->p_nthreads > 0) { 1126 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 1127 } 1128 1129 /* 1130 * Reap any LWPs left in p->p_lwps. This is usually 1131 * just the last LWP. This must be done before 1132 * we loop on p_lock since the lwps hold a ref on 1133 * it as a vmspace interlock. 1134 * 1135 * Once that is accomplished p_nthreads had better 1136 * be zero. 1137 */ 1138 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 1139 /* 1140 * Make sure no one is using this lwp, before 1141 * it is removed from the tree. If we didn't 1142 * wait it here, lwp tree iteration with 1143 * blocking operation would be broken. 1144 */ 1145 while (lp->lwp_lock > 0) 1146 tsleep(lp, 0, "zomblwp", 1); 1147 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 1148 reaplwp(lp); 1149 } 1150 KKASSERT(p->p_nthreads == 0); 1151 1152 /* 1153 * Don't do anything really bad until all references 1154 * to the process go away. This may include other 1155 * LWPs which are still in the process of being 1156 * reaped. We can't just pull the rug out from under 1157 * them because they may still be using the VM space. 1158 * 1159 * Certain kernel facilities such as /proc will also 1160 * put a hold on the process for short periods of 1161 * time. 1162 */ 1163 PRELE(p); /* from top of loop */ 1164 PSTALL(p, "reap3", 1); /* 1 ref (for PZOMBHOLD) */ 1165 1166 /* Take care of our return values. */ 1167 *res = p->p_pid; 1168 1169 *status = p->p_xstat; 1170 wrusage->wru_self = p->p_ru; 1171 wrusage->wru_children = p->p_cru; 1172 1173 if (info) { 1174 bzero(info, sizeof(*info)); 1175 info->si_errno = 0; 1176 info->si_signo = SIGCHLD; 1177 if (WIFEXITED(p->p_xstat)) { 1178 info->si_code = CLD_EXITED; 1179 info->si_status = 1180 WEXITSTATUS(p->p_xstat); 1181 } else { 1182 info->si_code = CLD_KILLED; 1183 info->si_status = WTERMSIG(p->p_xstat); 1184 } 1185 info->si_pid = p->p_pid; 1186 info->si_uid = p->p_ucred->cr_uid; 1187 } 1188 1189 /* 1190 * WNOWAIT shortcuts to done here, leaving the 1191 * child on the zombie list. 1192 */ 1193 if (options & WNOWAIT) { 1194 lwkt_reltoken(&p->p_token); 1195 PRELEZOMB(p); 1196 error = 0; 1197 goto done; 1198 } 1199 1200 /* 1201 * If we got the child via a ptrace 'attach', 1202 * we need to give it back to the old parent. 1203 */ 1204 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 1205 p->p_oppid = 0; 1206 proc_reparent(p, t); 1207 ksignal(t, SIGCHLD); 1208 wakeup((caddr_t)t); 1209 PRELE(t); 1210 lwkt_reltoken(&p->p_token); 1211 PRELEZOMB(p); 1212 error = 0; 1213 goto done; 1214 } 1215 1216 /* 1217 * Unlink the proc from its process group so that 1218 * the following operations won't lead to an 1219 * inconsistent state for processes running down 1220 * the zombie list. 1221 */ 1222 proc_remove_zombie(p); 1223 proc_userunmap(p); 1224 lwkt_reltoken(&p->p_token); 1225 leavepgrp(p); 1226 1227 p->p_xstat = 0; 1228 ruadd(&q->p_cru, &p->p_ru); 1229 ruadd(&q->p_cru, &p->p_cru); 1230 1231 /* 1232 * Decrement the count of procs running with this uid. 1233 */ 1234 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 1235 1236 /* 1237 * Free up credentials. p_spin is required to 1238 * avoid races against allproc scans. 1239 */ 1240 spin_lock(&p->p_spin); 1241 cr = p->p_ucred; 1242 p->p_ucred = NULL; 1243 spin_unlock(&p->p_spin); 1244 crfree(cr); 1245 1246 /* 1247 * Remove unused arguments 1248 */ 1249 pa = p->p_args; 1250 p->p_args = NULL; 1251 if (pa && refcount_release(&pa->ar_ref)) { 1252 kfree(pa, M_PARGS); 1253 pa = NULL; 1254 } 1255 1256 ps = p->p_sigacts; 1257 p->p_sigacts = NULL; 1258 if (ps && refcount_release(&ps->ps_refcnt)) { 1259 kfree(ps, M_SUBPROC); 1260 ps = NULL; 1261 } 1262 1263 /* 1264 * Our exitingcount was incremented when the process 1265 * became a zombie, now that the process has been 1266 * removed from (almost) all lists we should be able 1267 * to safely destroy its vmspace. Wait for any current 1268 * holders to go away (so the vmspace remains stable), 1269 * then scrap it. 1270 * 1271 * NOTE: Releasing the parent process (q) p_token 1272 * across the vmspace_exitfree() call is 1273 * important here to reduce stalls on 1274 * interactions with (q) (such as 1275 * fork/exec/wait or 'ps'). 1276 */ 1277 PSTALL(p, "reap4", 1); 1278 lwkt_reltoken(&q->p_token); 1279 vmspace_exitfree(p); 1280 lwkt_gettoken(&q->p_token); 1281 PSTALL(p, "reap5", 1); 1282 1283 /* 1284 * NOTE: We have to officially release ZOMB in order 1285 * to ensure that a racing thread in kern_wait() 1286 * which blocked on ZOMB is woken up. 1287 */ 1288 PRELEZOMB(p); 1289 kfree(p->p_uidpcpu, M_SUBPROC); 1290 kfree(p, M_PROC); 1291 atomic_add_int(&nprocs, -1); 1292 error = 0; 1293 goto done; 1294 } 1295 1296 /* 1297 * Process has not yet exited 1298 */ 1299 if ((p->p_stat == SSTOP || p->p_stat == SCORE) && 1300 (p->p_flags & P_WAITED) == 0 && 1301 (((p->p_flags & P_TRACED) && (options & WTRAPPED)) || 1302 (options & WSTOPPED))) { 1303 lwkt_gettoken(&p->p_token); 1304 if (p->p_pptr != q) { 1305 lwkt_reltoken(&p->p_token); 1306 PRELE(p); 1307 goto loop; 1308 } 1309 if ((p->p_stat != SSTOP && p->p_stat != SCORE) || 1310 (p->p_flags & P_WAITED) != 0 || 1311 ((p->p_flags & P_TRACED) == 0 && 1312 (options & WUNTRACED) == 0)) { 1313 lwkt_reltoken(&p->p_token); 1314 PRELE(p); 1315 goto loop; 1316 } 1317 1318 /* 1319 * Don't set P_WAITED if WNOWAIT specified, leaving 1320 * the process in a waitable state. 1321 */ 1322 if ((options & WNOWAIT) == 0) 1323 p->p_flags |= P_WAITED; 1324 1325 *res = p->p_pid; 1326 *status = W_STOPCODE(p->p_xstat); 1327 /* Zero rusage so we get something consistent. */ 1328 bzero(wrusage, sizeof(*wrusage)); 1329 error = 0; 1330 if (info) { 1331 bzero(info, sizeof(*info)); 1332 if (p->p_flags & P_TRACED) 1333 info->si_code = CLD_TRAPPED; 1334 else 1335 info->si_code = CLD_STOPPED; 1336 info->si_status = WSTOPSIG(p->p_xstat); 1337 } 1338 lwkt_reltoken(&p->p_token); 1339 PRELE(p); 1340 goto done; 1341 } 1342 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) { 1343 lwkt_gettoken(&p->p_token); 1344 if (p->p_pptr != q) { 1345 lwkt_reltoken(&p->p_token); 1346 PRELE(p); 1347 goto loop; 1348 } 1349 if ((p->p_flags & P_CONTINUED) == 0) { 1350 lwkt_reltoken(&p->p_token); 1351 PRELE(p); 1352 goto loop; 1353 } 1354 1355 *res = p->p_pid; 1356 1357 /* 1358 * Don't set P_WAITED if WNOWAIT specified, leaving 1359 * the process in a waitable state. 1360 */ 1361 if ((options & WNOWAIT) == 0) 1362 p->p_flags &= ~P_CONTINUED; 1363 1364 *status = SIGCONT; 1365 error = 0; 1366 if (info) { 1367 bzero(info, sizeof(*info)); 1368 info->si_code = CLD_CONTINUED; 1369 info->si_status = WSTOPSIG(p->p_xstat); 1370 } 1371 lwkt_reltoken(&p->p_token); 1372 PRELE(p); 1373 goto done; 1374 } 1375 PRELE(p); 1376 } 1377 if (nfound == 0) { 1378 error = ECHILD; 1379 goto done; 1380 } 1381 if (options & WNOHANG) { 1382 *res = 0; 1383 error = 0; 1384 goto done; 1385 } 1386 1387 /* 1388 * Wait for signal - interlocked using q->p_waitgen. 1389 */ 1390 error = 0; 1391 while ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) { 1392 tsleep_interlock(q, PCATCH); 1393 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000); 1394 if ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) { 1395 error = tsleep(q, PCATCH | PINTERLOCKED, "wait", 0); 1396 break; 1397 } 1398 } 1399 if (error) { 1400 done: 1401 lwkt_reltoken(&q->p_token); 1402 return (error); 1403 } 1404 goto loop; 1405 } 1406 1407 /* 1408 * Change child's parent process to parent. 1409 * 1410 * p_children/p_sibling requires the parent's token, and 1411 * changing pptr requires the child's token, so we have to 1412 * get three tokens to do this operation. We also need to 1413 * hold pointers that might get ripped out from under us to 1414 * preserve structural integrity. 1415 * 1416 * It is possible to race another reparent or disconnect or other 1417 * similar operation. We must retry when this situation occurs. 1418 * Once we successfully reparent the process we no longer care 1419 * about any races. 1420 */ 1421 void 1422 proc_reparent(struct proc *child, struct proc *parent) 1423 { 1424 struct proc *opp; 1425 1426 PHOLD(parent); 1427 while ((opp = child->p_pptr) != parent) { 1428 PHOLD(opp); 1429 lwkt_gettoken(&opp->p_token); 1430 lwkt_gettoken(&child->p_token); 1431 lwkt_gettoken(&parent->p_token); 1432 if (child->p_pptr != opp) { 1433 lwkt_reltoken(&parent->p_token); 1434 lwkt_reltoken(&child->p_token); 1435 lwkt_reltoken(&opp->p_token); 1436 PRELE(opp); 1437 continue; 1438 } 1439 LIST_REMOVE(child, p_sibling); 1440 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1441 child->p_pptr = parent; 1442 child->p_ppid = parent->p_pid; 1443 lwkt_reltoken(&parent->p_token); 1444 lwkt_reltoken(&child->p_token); 1445 lwkt_reltoken(&opp->p_token); 1446 if (LIST_EMPTY(&opp->p_children)) 1447 wakeup(opp); 1448 PRELE(opp); 1449 break; 1450 } 1451 PRELE(parent); 1452 } 1453 1454 /* 1455 * The next two functions are to handle adding/deleting items on the 1456 * exit callout list 1457 * 1458 * at_exit(): 1459 * Take the arguments given and put them onto the exit callout list, 1460 * However first make sure that it's not already there. 1461 * returns 0 on success. 1462 */ 1463 1464 int 1465 at_exit(exitlist_fn function) 1466 { 1467 struct exitlist *ep; 1468 1469 #ifdef INVARIANTS 1470 /* Be noisy if the programmer has lost track of things */ 1471 if (rm_at_exit(function)) 1472 kprintf("WARNING: exit callout entry (%p) already present\n", 1473 function); 1474 #endif 1475 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 1476 if (ep == NULL) 1477 return (ENOMEM); 1478 ep->function = function; 1479 TAILQ_INSERT_TAIL(&exit_list, ep, next); 1480 return (0); 1481 } 1482 1483 /* 1484 * Scan the exit callout list for the given item and remove it. 1485 * Returns the number of items removed (0 or 1) 1486 */ 1487 int 1488 rm_at_exit(exitlist_fn function) 1489 { 1490 struct exitlist *ep; 1491 1492 TAILQ_FOREACH(ep, &exit_list, next) { 1493 if (ep->function == function) { 1494 TAILQ_REMOVE(&exit_list, ep, next); 1495 kfree(ep, M_ATEXIT); 1496 return(1); 1497 } 1498 } 1499 return (0); 1500 } 1501 1502 /* 1503 * LWP reaper related code. 1504 */ 1505 static void 1506 reaplwps(void *context, int dummy) 1507 { 1508 struct lwplist *lwplist = context; 1509 struct lwp *lp; 1510 int cpu = mycpuid; 1511 1512 lwkt_gettoken(&deadlwp_token[cpu]); 1513 while ((lp = LIST_FIRST(lwplist))) { 1514 LIST_REMOVE(lp, u.lwp_reap_entry); 1515 reaplwp(lp); 1516 } 1517 lwkt_reltoken(&deadlwp_token[cpu]); 1518 } 1519 1520 static void 1521 reaplwp(struct lwp *lp) 1522 { 1523 while (lwp_wait(lp) == 0) 1524 ; 1525 lwp_dispose(lp); 1526 } 1527 1528 static void 1529 deadlwp_init(void) 1530 { 1531 int cpu; 1532 1533 for (cpu = 0; cpu < ncpus; cpu++) { 1534 lwkt_token_init(&deadlwp_token[cpu], "deadlwpl"); 1535 LIST_INIT(&deadlwp_list[cpu]); 1536 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), 1537 M_DEVBUF, M_WAITOK); 1538 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1539 } 1540 } 1541 1542 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1543