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 void (*linux_task_drop_callback)(thread_t td); 103 void (*linux_proc_drop_callback)(struct proc *p); 104 105 /* 106 * exit -- 107 * Death of process. 108 * 109 * SYS_EXIT_ARGS(int rval) 110 */ 111 int 112 sys_exit(struct exit_args *uap) 113 { 114 exit1(W_EXITCODE(uap->rval, 0)); 115 /* NOTREACHED */ 116 } 117 118 /* 119 * Extended exit -- 120 * Death of a lwp or process with optional bells and whistles. 121 */ 122 int 123 sys_extexit(struct extexit_args *uap) 124 { 125 struct proc *p = curproc; 126 int action, who; 127 int error; 128 129 action = EXTEXIT_ACTION(uap->how); 130 who = EXTEXIT_WHO(uap->how); 131 132 /* Check parameters before we might perform some action */ 133 switch (who) { 134 case EXTEXIT_PROC: 135 case EXTEXIT_LWP: 136 break; 137 default: 138 return (EINVAL); 139 } 140 141 switch (action) { 142 case EXTEXIT_SIMPLE: 143 break; 144 case EXTEXIT_SETINT: 145 error = copyout(&uap->status, uap->addr, sizeof(uap->status)); 146 if (error) 147 return (error); 148 break; 149 default: 150 return (EINVAL); 151 } 152 153 lwkt_gettoken(&p->p_token); 154 155 switch (who) { 156 case EXTEXIT_LWP: 157 /* 158 * Be sure only to perform a simple lwp exit if there is at 159 * least one more lwp in the proc, which will call exit1() 160 * later, otherwise the proc will be an UNDEAD and not even a 161 * SZOMB! 162 */ 163 if (p->p_nthreads > 1) { 164 lwp_exit(0, NULL); /* called w/ p_token held */ 165 /* NOT REACHED */ 166 } 167 /* else last lwp in proc: do the real thing */ 168 /* FALLTHROUGH */ 169 default: /* to help gcc */ 170 case EXTEXIT_PROC: 171 lwkt_reltoken(&p->p_token); 172 exit1(W_EXITCODE(uap->status, 0)); 173 /* NOTREACHED */ 174 } 175 176 /* NOTREACHED */ 177 lwkt_reltoken(&p->p_token); /* safety */ 178 } 179 180 /* 181 * Kill all lwps associated with the current process except the 182 * current lwp. Return an error if we race another thread trying to 183 * do the same thing and lose the race. 184 * 185 * If forexec is non-zero the current thread and process flags are 186 * cleaned up so they can be reused. 187 */ 188 int 189 killalllwps(int forexec) 190 { 191 struct lwp *lp = curthread->td_lwp; 192 struct proc *p = lp->lwp_proc; 193 int fakestop; 194 195 /* 196 * Interlock against P_WEXIT. Only one of the process's thread 197 * is allowed to do the master exit. 198 */ 199 lwkt_gettoken(&p->p_token); 200 if (p->p_flags & P_WEXIT) { 201 lwkt_reltoken(&p->p_token); 202 return (EALREADY); 203 } 204 p->p_flags |= P_WEXIT; 205 lwkt_gettoken(&lp->lwp_token); 206 207 /* 208 * Set temporary stopped state in case we are racing a coredump. 209 * Otherwise the coredump may hang forever. 210 */ 211 if (lp->lwp_mpflags & LWP_MP_WSTOP) { 212 fakestop = 0; 213 } else { 214 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP); 215 ++p->p_nstopped; 216 fakestop = 1; 217 wakeup(&p->p_nstopped); 218 } 219 220 /* 221 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs 222 */ 223 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 224 if (p->p_nthreads > 1) 225 killlwps(lp); 226 227 /* 228 * Undo temporary stopped state 229 */ 230 if (fakestop && (lp->lwp_mpflags & LWP_MP_WSTOP)) { 231 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WSTOP); 232 --p->p_nstopped; 233 } 234 235 /* 236 * If doing this for an exec, clean up the remaining thread 237 * (us) for continuing operation after all the other threads 238 * have been killed. 239 */ 240 if (forexec) { 241 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 242 p->p_flags &= ~P_WEXIT; 243 } 244 lwkt_reltoken(&lp->lwp_token); 245 lwkt_reltoken(&p->p_token); 246 247 return(0); 248 } 249 250 /* 251 * Kill all LWPs except the current one. Do not try to signal 252 * LWPs which have exited on their own or have already been 253 * signaled. 254 */ 255 static void 256 killlwps(struct lwp *lp) 257 { 258 struct proc *p = lp->lwp_proc; 259 struct lwp *tlp; 260 261 /* 262 * Kill the remaining LWPs. We must send the signal before setting 263 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce 264 * races. tlp must be held across the call as it might block and 265 * allow the target lwp to rip itself out from under our loop. 266 */ 267 FOREACH_LWP_IN_PROC(tlp, p) { 268 LWPHOLD(tlp); 269 lwkt_gettoken(&tlp->lwp_token); 270 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) { 271 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT); 272 lwpsignal(p, tlp, SIGKILL); 273 } 274 lwkt_reltoken(&tlp->lwp_token); 275 LWPRELE(tlp); 276 } 277 278 /* 279 * Wait for everything to clear out. Also make sure any tstop()s 280 * are signalled (we are holding p_token for the interlock). 281 */ 282 wakeup(p); 283 while (p->p_nthreads > 1) 284 tsleep(&p->p_nthreads, 0, "killlwps", 0); 285 } 286 287 /* 288 * Exit: deallocate address space and other resources, change proc state 289 * to zombie, and unlink proc from allproc and parent's lists. Save exit 290 * status and rusage for wait(). Check for child processes and orphan them. 291 */ 292 void 293 exit1(int rv) 294 { 295 struct thread *td = curthread; 296 struct proc *p = td->td_proc; 297 struct lwp *lp = td->td_lwp; 298 struct proc *q; 299 struct proc *pp; 300 struct proc *reproc; 301 struct sysreaper *reap; 302 struct vmspace *vm; 303 struct vnode *vtmp; 304 struct exitlist *ep; 305 int error; 306 307 lwkt_gettoken(&p->p_token); 308 309 if (p->p_pid == 1) { 310 kprintf("init died (signal %d, exit %d)\n", 311 WTERMSIG(rv), WEXITSTATUS(rv)); 312 panic("Going nowhere without my init!"); 313 } 314 varsymset_clean(&p->p_varsymset); 315 lockuninit(&p->p_varsymset.vx_lock); 316 317 /* 318 * Kill all lwps associated with the current process, return an 319 * error if we race another thread trying to do the same thing 320 * and lose the race. 321 */ 322 error = killalllwps(0); 323 if (error) { 324 lwp_exit(0, NULL); 325 /* NOT REACHED */ 326 } 327 328 /* are we a task leader? */ 329 if (p == p->p_leader) { 330 struct kill_args killArgs; 331 killArgs.signum = SIGKILL; 332 q = p->p_peers; 333 while(q) { 334 killArgs.pid = q->p_pid; 335 /* 336 * The interface for kill is better 337 * than the internal signal 338 */ 339 sys_kill(&killArgs); 340 q = q->p_peers; 341 } 342 while (p->p_peers) 343 tsleep((caddr_t)p, 0, "exit1", 0); 344 } 345 346 #ifdef PGINPROF 347 vmsizmon(); 348 #endif 349 STOPEVENT(p, S_EXIT, rv); 350 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */ 351 352 /* 353 * Check if any loadable modules need anything done at process exit. 354 * e.g. SYSV IPC stuff 355 * XXX what if one of these generates an error? 356 */ 357 p->p_xstat = rv; 358 359 /* 360 * XXX: imho, the eventhandler stuff is much cleaner than this. 361 * Maybe we should move everything to use eventhandler. 362 */ 363 TAILQ_FOREACH(ep, &exit_list, next) 364 (*ep->function)(td); 365 366 if (p->p_flags & P_PROFIL) 367 stopprofclock(p); 368 369 SIGEMPTYSET(p->p_siglist); 370 SIGEMPTYSET(lp->lwp_siglist); 371 if (timevalisset(&p->p_realtimer.it_value)) 372 callout_terminate(&p->p_ithandle); 373 374 /* 375 * Reset any sigio structures pointing to us as a result of 376 * F_SETOWN with our pid. 377 */ 378 funsetownlst(&p->p_sigiolst); 379 380 /* 381 * Close open files and release open-file table. 382 * This may block! 383 */ 384 fdfree(p, NULL); 385 386 if (p->p_leader->p_peers) { 387 q = p->p_leader; 388 while(q->p_peers != p) 389 q = q->p_peers; 390 q->p_peers = p->p_peers; 391 wakeup((caddr_t)p->p_leader); 392 } 393 394 /* 395 * XXX Shutdown SYSV semaphores 396 */ 397 semexit(p); 398 399 /* The next two chunks should probably be moved to vmspace_exit. */ 400 vm = p->p_vmspace; 401 402 /* 403 * Clean up data related to virtual kernel operation. Clean up 404 * any vkernel context related to the current lwp now so we can 405 * destroy p_vkernel. 406 */ 407 if (p->p_vkernel) { 408 vkernel_lwp_exit(lp); 409 vkernel_exit(p); 410 } 411 412 /* 413 * Release the user portion of address space. The exitbump prevents 414 * the vmspace from being completely eradicated (using holdcnt). 415 * This releases references to vnodes, which could cause I/O if the 416 * file has been unlinked. We need to do this early enough that 417 * we can still sleep. 418 * 419 * We can't free the entire vmspace as the kernel stack may be mapped 420 * within that space also. 421 * 422 * Processes sharing the same vmspace may exit in one order, and 423 * get cleaned up by vmspace_exit() in a different order. The 424 * last exiting process to reach this point releases as much of 425 * the environment as it can, and the last process cleaned up 426 * by vmspace_exit() (which decrements exitingcnt) cleans up the 427 * remainder. 428 * 429 * NOTE: Releasing p_token around this call is helpful if the 430 * vmspace had a huge RSS. Otherwise some other process 431 * trying to do an allproc or other scan (like 'ps') may 432 * stall for a long time. 433 */ 434 lwkt_reltoken(&p->p_token); 435 vmspace_relexit(vm); 436 lwkt_gettoken(&p->p_token); 437 438 if (SESS_LEADER(p)) { 439 struct session *sp = p->p_session; 440 441 if (sp->s_ttyvp) { 442 /* 443 * We are the controlling process. Signal the 444 * foreground process group, drain the controlling 445 * terminal, and revoke access to the controlling 446 * terminal. 447 * 448 * NOTE: while waiting for the process group to exit 449 * it is possible that one of the processes in the 450 * group will revoke the tty, so the ttyclosesession() 451 * function will re-check sp->s_ttyvp. 452 */ 453 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 454 if (sp->s_ttyp->t_pgrp) 455 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 456 ttywait(sp->s_ttyp); 457 ttyclosesession(sp, 1); /* also revoke */ 458 } 459 /* 460 * Release the tty. If someone has it open via 461 * /dev/tty then close it (since they no longer can 462 * once we've NULL'd it out). 463 */ 464 ttyclosesession(sp, 0); 465 466 /* 467 * s_ttyp is not zero'd; we use this to indicate 468 * that the session once had a controlling terminal. 469 * (for logging and informational purposes) 470 */ 471 } 472 sp->s_leader = NULL; 473 } 474 fixjobc(p, p->p_pgrp, 0); 475 (void)acct_process(p); 476 #ifdef KTRACE 477 /* 478 * release trace file 479 */ 480 if (p->p_tracenode) 481 ktrdestroy(&p->p_tracenode); 482 p->p_traceflag = 0; 483 #endif 484 /* 485 * Release reference to text vnode 486 */ 487 if ((vtmp = p->p_textvp) != NULL) { 488 p->p_textvp = NULL; 489 vrele(vtmp); 490 } 491 492 /* Release namecache handle to text file */ 493 if (p->p_textnch.ncp) 494 cache_drop(&p->p_textnch); 495 496 /* 497 * We have to handle PPWAIT here or proc_move_allproc_zombie() 498 * will block on the PHOLD() the parent is doing. 499 * 500 * We are using the flag as an interlock so an atomic op is 501 * necessary to synchronize with the parent's cpu. 502 */ 503 if (p->p_flags & P_PPWAIT) { 504 if (p->p_pptr && p->p_pptr->p_upmap) 505 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1); 506 atomic_clear_int(&p->p_flags, P_PPWAIT); 507 wakeup(p->p_pptr); 508 } 509 510 /* 511 * Move the process to the zombie list. This will block 512 * until the process p_lock count reaches 0. The process will 513 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 514 * which is called from cpu_proc_exit(). 515 * 516 * Interlock against waiters using p_waitgen. We increment 517 * p_waitgen after completing the move of our process to the 518 * zombie list. 519 * 520 * WARNING: pp becomes stale when we block, clear it now as a 521 * reminder. 522 */ 523 proc_move_allproc_zombie(p); 524 pp = p->p_pptr; 525 atomic_add_long(&pp->p_waitgen, 1); 526 pp = NULL; 527 528 /* 529 * release controlled reaper for exit if we own it and return the 530 * remaining reaper (the one for us), which we will drop after we 531 * are done. 532 */ 533 reap = reaper_exit(p); 534 535 /* 536 * Reparent all of this process's children to the init process or 537 * to the designated reaper. We must hold the reaper's p_token in 538 * order to safely mess with p_children. 539 * 540 * We already hold p->p_token (to remove the children from our list). 541 */ 542 reproc = NULL; 543 q = LIST_FIRST(&p->p_children); 544 if (q) { 545 reproc = reaper_get(reap); 546 lwkt_gettoken(&reproc->p_token); 547 while ((q = LIST_FIRST(&p->p_children)) != NULL) { 548 PHOLD(q); 549 lwkt_gettoken(&q->p_token); 550 if (q != LIST_FIRST(&p->p_children)) { 551 lwkt_reltoken(&q->p_token); 552 PRELE(q); 553 continue; 554 } 555 LIST_REMOVE(q, p_sibling); 556 LIST_INSERT_HEAD(&reproc->p_children, q, p_sibling); 557 q->p_pptr = reproc; 558 q->p_ppid = reproc->p_pid; 559 q->p_sigparent = SIGCHLD; 560 561 /* 562 * Traced processes are killed 563 * since their existence means someone is screwing up. 564 */ 565 if (q->p_flags & P_TRACED) { 566 q->p_flags &= ~P_TRACED; 567 ksignal(q, SIGKILL); 568 } 569 lwkt_reltoken(&q->p_token); 570 PRELE(q); 571 } 572 lwkt_reltoken(&reproc->p_token); 573 wakeup(reproc); 574 } 575 576 /* 577 * Save exit status and final rusage info. We no longer add 578 * child rusage info into self times, wait4() and kern_wait() 579 * handles it in order to properly support wait6(). 580 */ 581 calcru_proc(p, &p->p_ru); 582 /*ruadd(&p->p_ru, &p->p_cru); REMOVED */ 583 584 /* 585 * notify interested parties of our demise. 586 */ 587 KNOTE(&p->p_klist, NOTE_EXIT); 588 589 /* 590 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 591 * flag set, or if the handler is set to SIG_IGN, notify the reaper 592 * instead (it will handle this situation). 593 * 594 * NOTE: The reaper can still be the parent process. 595 * 596 * (must reload pp) 597 */ 598 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 599 if (reproc == NULL) 600 reproc = reaper_get(reap); 601 proc_reparent(p, reproc); 602 } 603 if (reproc) 604 PRELE(reproc); 605 if (reap) 606 reaper_drop(reap); 607 608 /* 609 * Signal (possibly new) parent. 610 */ 611 pp = p->p_pptr; 612 PHOLD(pp); 613 if (p->p_sigparent && pp != initproc) { 614 int sig = p->p_sigparent; 615 616 if (sig != SIGUSR1 && sig != SIGCHLD) 617 sig = SIGCHLD; 618 ksignal(pp, sig); 619 } else { 620 ksignal(pp, SIGCHLD); 621 } 622 p->p_flags &= ~P_TRACED; 623 PRELE(pp); 624 625 /* 626 * cpu_exit is responsible for clearing curproc, since 627 * it is heavily integrated with the thread/switching sequence. 628 * 629 * Other substructures are freed from wait(). 630 */ 631 if (p->p_limit) { 632 struct plimit *rlimit; 633 634 rlimit = p->p_limit; 635 p->p_limit = NULL; 636 plimit_free(rlimit); 637 } 638 639 /* 640 * Finally, call machine-dependent code to release as many of the 641 * lwp's resources as we can and halt execution of this thread. 642 * 643 * pp is a wild pointer now but still the correct wakeup() target. 644 * lwp_exit() only uses it to send the wakeup() signal to the likely 645 * parent. Any reparenting race that occurs will get a signal 646 * automatically and not be an issue. 647 */ 648 lwp_exit(1, pp); 649 } 650 651 /* 652 * Eventually called by every exiting LWP 653 * 654 * p->p_token must be held. mplock may be held and will be released. 655 */ 656 void 657 lwp_exit(int masterexit, void *waddr) 658 { 659 struct thread *td = curthread; 660 struct lwp *lp = td->td_lwp; 661 struct proc *p = lp->lwp_proc; 662 int dowake = 0; 663 664 /* 665 * Release the current user process designation on the process so 666 * the userland scheduler can work in someone else. 667 */ 668 p->p_usched->release_curproc(lp); 669 670 /* 671 * Destroy the per-thread shared page and remove from any pmaps 672 * it resides in. 673 */ 674 lwp_userunmap(lp); 675 676 /* 677 * lwp_exit() may be called without setting LWP_MP_WEXIT, so 678 * make sure it is set here. 679 */ 680 ASSERT_LWKT_TOKEN_HELD(&p->p_token); 681 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 682 683 /* 684 * Clean up any virtualization 685 */ 686 if (lp->lwp_vkernel) 687 vkernel_lwp_exit(lp); 688 689 if (td->td_vmm) 690 vmm_vmdestroy(); 691 692 /* 693 * Clean up select/poll support 694 */ 695 kqueue_terminate(&lp->lwp_kqueue); 696 697 if (td->td_linux_task) 698 linux_task_drop_callback(td); 699 if (masterexit && p->p_linux_mm) 700 linux_proc_drop_callback(p); 701 702 /* 703 * Clean up any syscall-cached ucred or rlimit. 704 */ 705 if (td->td_ucred) { 706 crfree(td->td_ucred); 707 td->td_ucred = NULL; 708 } 709 if (td->td_limit) { 710 struct plimit *rlimit; 711 712 rlimit = td->td_limit; 713 td->td_limit = NULL; 714 plimit_free(rlimit); 715 } 716 717 /* 718 * Cleanup any cached descriptors for this thread 719 */ 720 if (p->p_fd) 721 fexitcache(td); 722 723 /* 724 * Nobody actually wakes us when the lock 725 * count reaches zero, so just wait one tick. 726 */ 727 while (lp->lwp_lock > 0) 728 tsleep(lp, 0, "lwpexit", 1); 729 730 /* Hand down resource usage to our proc */ 731 ruadd(&p->p_ru, &lp->lwp_ru); 732 733 /* 734 * If we don't hold the process until the LWP is reaped wait*() 735 * may try to dispose of its vmspace before all the LWPs have 736 * actually terminated. 737 */ 738 PHOLD(p); 739 740 /* 741 * Do any remaining work that might block on us. We should be 742 * coded such that further blocking is ok after decrementing 743 * p_nthreads but don't take the chance. 744 */ 745 dsched_exit_thread(td); 746 biosched_done(curthread); 747 748 /* 749 * We have to use the reaper for all the LWPs except the one doing 750 * the master exit. The LWP doing the master exit can just be 751 * left on p_lwps and the process reaper will deal with it 752 * synchronously, which is much faster. 753 * 754 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0. 755 * 756 * The process is left held until the reaper calls lwp_dispose() on 757 * the lp (after calling lwp_wait()). 758 */ 759 if (masterexit == 0) { 760 int cpu = mycpuid; 761 762 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 763 --p->p_nthreads; 764 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1) 765 dowake = 1; 766 lwkt_gettoken(&deadlwp_token[cpu]); 767 LIST_INSERT_HEAD(&deadlwp_list[cpu], lp, u.lwp_reap_entry); 768 taskqueue_enqueue(taskqueue_thread[cpu], deadlwp_task[cpu]); 769 lwkt_reltoken(&deadlwp_token[cpu]); 770 } else { 771 --p->p_nthreads; 772 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1) 773 dowake = 1; 774 } 775 776 /* 777 * We no longer need p_token. 778 * 779 * Tell the userland scheduler that we are going away 780 */ 781 lwkt_reltoken(&p->p_token); 782 p->p_usched->heuristic_exiting(lp, p); 783 784 /* 785 * Issue late wakeups after releasing our token to give us a chance 786 * to deschedule and switch away before another cpu in a wait*() 787 * reaps us. This is done as late as possible to reduce contention. 788 */ 789 if (dowake) 790 wakeup(&p->p_nthreads); 791 if (waddr) 792 wakeup(waddr); 793 794 cpu_lwp_exit(); 795 } 796 797 /* 798 * Wait until a lwp is completely dead. The final interlock in this drama 799 * is when TDF_EXITING is set in cpu_thread_exit() just before the final 800 * switchout. 801 * 802 * At the point TDF_EXITING is set a complete exit is accomplished when 803 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two 804 * post-switch interlock flags that can be used to wait for the TDF_ 805 * flags to clear. 806 * 807 * Returns non-zero on success, and zero if the caller needs to retry 808 * the lwp_wait(). 809 */ 810 static int 811 lwp_wait(struct lwp *lp) 812 { 813 struct thread *td = lp->lwp_thread; 814 u_int mpflags; 815 816 KKASSERT(lwkt_preempted_proc() != lp); 817 818 /* 819 * This bit of code uses the thread destruction interlock 820 * managed by lwkt_switch_return() to wait for the lwp's 821 * thread to completely disengage. 822 * 823 * It is possible for us to race another cpu core so we 824 * have to do this correctly. 825 */ 826 for (;;) { 827 mpflags = td->td_mpflags; 828 cpu_ccfence(); 829 if (mpflags & TDF_MP_EXITSIG) 830 break; 831 tsleep_interlock(td, 0); 832 if (atomic_cmpset_int(&td->td_mpflags, mpflags, 833 mpflags | TDF_MP_EXITWAIT)) { 834 tsleep(td, PINTERLOCKED, "lwpxt", 0); 835 } 836 } 837 838 /* 839 * We've already waited for the core exit but there can still 840 * be other refs from e.g. process scans and such. 841 */ 842 if (lp->lwp_lock > 0) { 843 tsleep(lp, 0, "lwpwait1", 1); 844 return(0); 845 } 846 if (td->td_refs) { 847 tsleep(td, 0, "lwpwait2", 1); 848 return(0); 849 } 850 851 /* 852 * Now that we have the thread destruction interlock these flags 853 * really should already be cleaned up, keep a check for safety. 854 * 855 * We can't rip its stack out from under it until TDF_EXITING is 856 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 857 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 858 * will be cleared temporarily if a thread gets preempted. 859 */ 860 while ((td->td_flags & (TDF_RUNNING | 861 TDF_RUNQ | 862 TDF_PREEMPT_LOCK | 863 TDF_EXITING)) != TDF_EXITING) { 864 tsleep(lp, 0, "lwpwait3", 1); 865 return (0); 866 } 867 868 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0, 869 ("lwp_wait: td %p (%s) still on run or sleep queue", 870 td, td->td_comm)); 871 return (1); 872 } 873 874 /* 875 * Release the resources associated with a lwp. 876 * The lwp must be completely dead. 877 */ 878 void 879 lwp_dispose(struct lwp *lp) 880 { 881 struct thread *td = lp->lwp_thread; 882 883 KKASSERT(lwkt_preempted_proc() != lp); 884 KKASSERT(lp->lwp_lock == 0); 885 KKASSERT(td->td_refs == 0); 886 KKASSERT((td->td_flags & (TDF_RUNNING | 887 TDF_RUNQ | 888 TDF_PREEMPT_LOCK | 889 TDF_EXITING)) == TDF_EXITING); 890 891 PRELE(lp->lwp_proc); 892 lp->lwp_proc = NULL; 893 if (td != NULL) { 894 td->td_proc = NULL; 895 td->td_lwp = NULL; 896 lp->lwp_thread = NULL; 897 lwkt_free_thread(td); 898 } 899 kfree(lp, M_LWP); 900 } 901 902 int 903 sys_wait4(struct wait_args *uap) 904 { 905 struct __wrusage wrusage; 906 int error; 907 int status; 908 int options; 909 id_t id; 910 idtype_t idtype; 911 912 options = uap->options | WEXITED | WTRAPPED; 913 id = uap->pid; 914 915 if (id == WAIT_ANY) { 916 idtype = P_ALL; 917 } else if (id == WAIT_MYPGRP) { 918 idtype = P_PGID; 919 id = curproc->p_pgid; 920 } else if (id < 0) { 921 idtype = P_PGID; 922 id = -id; 923 } else { 924 idtype = P_PID; 925 } 926 927 error = kern_wait(idtype, id, &status, options, &wrusage, 928 NULL, &uap->sysmsg_result); 929 930 if (error == 0 && uap->status) 931 error = copyout(&status, uap->status, sizeof(*uap->status)); 932 if (error == 0 && uap->rusage) { 933 ruadd(&wrusage.wru_self, &wrusage.wru_children); 934 error = copyout(&wrusage.wru_self, uap->rusage, sizeof(*uap->rusage)); 935 } 936 return (error); 937 } 938 939 int 940 sys_wait6(struct wait6_args *uap) 941 { 942 struct __wrusage wrusage; 943 siginfo_t info; 944 siginfo_t *infop; 945 int error; 946 int status; 947 int options; 948 id_t id; 949 idtype_t idtype; 950 951 /* 952 * NOTE: wait6() requires WEXITED and WTRAPPED to be specified if 953 * desired. 954 */ 955 options = uap->options; 956 idtype = uap->idtype; 957 id = uap->id; 958 infop = uap->info ? &info : NULL; 959 960 switch(idtype) { 961 case P_PID: 962 case P_PGID: 963 if (id == WAIT_MYPGRP) { 964 idtype = P_PGID; 965 id = curproc->p_pgid; 966 } 967 break; 968 default: 969 /* let kern_wait deal with the remainder */ 970 break; 971 } 972 973 error = kern_wait(idtype, id, &status, options, 974 &wrusage, infop, &uap->sysmsg_result); 975 976 if (error == 0 && uap->status) 977 error = copyout(&status, uap->status, sizeof(*uap->status)); 978 if (error == 0 && uap->wrusage) 979 error = copyout(&wrusage, uap->wrusage, sizeof(*uap->wrusage)); 980 if (error == 0 && uap->info) 981 error = copyout(&info, uap->info, sizeof(*uap->info)); 982 return (error); 983 } 984 985 /* 986 * kernel wait*() system call support 987 */ 988 int 989 kern_wait(idtype_t idtype, id_t id, int *status, int options, 990 struct __wrusage *wrusage, siginfo_t *info, int *res) 991 { 992 struct thread *td = curthread; 993 struct lwp *lp; 994 struct proc *q = td->td_proc; 995 struct proc *p, *t; 996 struct ucred *cr; 997 struct pargs *pa; 998 struct sigacts *ps; 999 int nfound, error; 1000 long waitgen; 1001 1002 /* 1003 * Must not have extraneous options. Must have at least one 1004 * matchable option. 1005 */ 1006 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE|WSTOPPED| 1007 WEXITED|WTRAPPED|WNOWAIT)) { 1008 return (EINVAL); 1009 } 1010 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1011 return (EINVAL); 1012 } 1013 1014 /* 1015 * Protect the q->p_children list 1016 */ 1017 lwkt_gettoken(&q->p_token); 1018 loop: 1019 /* 1020 * All sorts of things can change due to blocking so we have to loop 1021 * all the way back up here. 1022 * 1023 * The problem is that if a process group is stopped and the parent 1024 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 1025 * of the child and then stop itself when it tries to return from the 1026 * system call. When the process group is resumed the parent will 1027 * then get the STOP status even though the child has now resumed 1028 * (a followup wait*() will get the CONT status). 1029 * 1030 * Previously the CONT would overwrite the STOP because the tstop 1031 * was handled within tsleep(), and the parent would only see 1032 * the CONT when both are stopped and continued together. This little 1033 * two-line hack restores this effect. 1034 * 1035 * No locks are held so we can safely block the process here. 1036 */ 1037 if (STOPLWP(q, td->td_lwp)) 1038 tstop(); 1039 1040 nfound = 0; 1041 1042 /* 1043 * Loop on children. 1044 * 1045 * NOTE: We don't want to break q's p_token in the loop for the 1046 * case where no children are found or we risk breaking the 1047 * interlock between child and parent. 1048 */ 1049 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000); 1050 LIST_FOREACH(p, &q->p_children, p_sibling) { 1051 /* 1052 * Skip children that another thread is already uninterruptably 1053 * reaping. 1054 */ 1055 if (PWAITRES_PENDING(p)) 1056 continue; 1057 1058 /* 1059 * Filter, (p) will be held on fall-through. Try to optimize 1060 * this to avoid the atomic op until we are pretty sure we 1061 * want this process. 1062 */ 1063 switch(idtype) { 1064 case P_ALL: 1065 PHOLD(p); 1066 break; 1067 case P_PID: 1068 if (p->p_pid != (pid_t)id) 1069 continue; 1070 PHOLD(p); 1071 break; 1072 case P_PGID: 1073 if (p->p_pgid != (pid_t)id) 1074 continue; 1075 PHOLD(p); 1076 break; 1077 case P_SID: 1078 PHOLD(p); 1079 if (p->p_session && p->p_session->s_sid != (pid_t)id) { 1080 PRELE(p); 1081 continue; 1082 } 1083 break; 1084 case P_UID: 1085 PHOLD(p); 1086 if (p->p_ucred->cr_uid != (uid_t)id) { 1087 PRELE(p); 1088 continue; 1089 } 1090 break; 1091 case P_GID: 1092 PHOLD(p); 1093 if (p->p_ucred->cr_gid != (gid_t)id) { 1094 PRELE(p); 1095 continue; 1096 } 1097 break; 1098 case P_JAILID: 1099 PHOLD(p); 1100 if (p->p_ucred->cr_prison && 1101 p->p_ucred->cr_prison->pr_id != (int)id) { 1102 PRELE(p); 1103 continue; 1104 } 1105 break; 1106 default: 1107 /* unsupported filter */ 1108 continue; 1109 } 1110 /* (p) is held at this point */ 1111 1112 /* 1113 * This special case handles a kthread spawned by linux_clone 1114 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1115 * functions need to be able to distinguish between waiting 1116 * on a process and waiting on a thread. It is a thread if 1117 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1118 * signifies we want to wait for threads and not processes. 1119 */ 1120 if ((p->p_sigparent != SIGCHLD) ^ 1121 ((options & WLINUXCLONE) != 0)) { 1122 PRELE(p); 1123 continue; 1124 } 1125 1126 nfound++; 1127 if (p->p_stat == SZOMB && (options & WEXITED)) { 1128 /* 1129 * We may go into SZOMB with threads still present. 1130 * We must wait for them to exit before we can reap 1131 * the master thread, otherwise we may race reaping 1132 * non-master threads. 1133 * 1134 * Only this routine can remove a process from 1135 * the zombie list and destroy it. 1136 * 1137 * This function will fail after sleeping if another 1138 * thread owns the zombie lock. This function will 1139 * fail immediately or after sleeping if another 1140 * thread owns or obtains ownership of the reap via 1141 * WAITRES. 1142 */ 1143 if (PHOLDZOMB(p)) { 1144 PRELE(p); 1145 goto loop; 1146 } 1147 lwkt_gettoken(&p->p_token); 1148 if (p->p_pptr != q) { 1149 lwkt_reltoken(&p->p_token); 1150 PRELE(p); 1151 PRELEZOMB(p); 1152 goto loop; 1153 } 1154 1155 /* 1156 * We are the reaper, from this point on the reap 1157 * cannot be aborted. 1158 */ 1159 PWAITRES_SET(p); 1160 while (p->p_nthreads > 0) { 1161 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 1162 } 1163 1164 /* 1165 * Reap any LWPs left in p->p_lwps. This is usually 1166 * just the last LWP. This must be done before 1167 * we loop on p_lock since the lwps hold a ref on 1168 * it as a vmspace interlock. 1169 * 1170 * Once that is accomplished p_nthreads had better 1171 * be zero. 1172 */ 1173 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 1174 /* 1175 * Make sure no one is using this lwp, before 1176 * it is removed from the tree. If we didn't 1177 * wait it here, lwp tree iteration with 1178 * blocking operation would be broken. 1179 */ 1180 while (lp->lwp_lock > 0) 1181 tsleep(lp, 0, "zomblwp", 1); 1182 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 1183 reaplwp(lp); 1184 } 1185 KKASSERT(p->p_nthreads == 0); 1186 1187 /* 1188 * Don't do anything really bad until all references 1189 * to the process go away. This may include other 1190 * LWPs which are still in the process of being 1191 * reaped. We can't just pull the rug out from under 1192 * them because they may still be using the VM space. 1193 * 1194 * Certain kernel facilities such as /proc will also 1195 * put a hold on the process for short periods of 1196 * time. 1197 */ 1198 PRELE(p); /* from top of loop */ 1199 PSTALL(p, "reap3", 1); /* 1 ref (for PZOMBHOLD) */ 1200 1201 /* Take care of our return values. */ 1202 *res = p->p_pid; 1203 1204 *status = p->p_xstat; 1205 wrusage->wru_self = p->p_ru; 1206 wrusage->wru_children = p->p_cru; 1207 1208 if (info) { 1209 bzero(info, sizeof(*info)); 1210 info->si_errno = 0; 1211 info->si_signo = SIGCHLD; 1212 if (WIFEXITED(p->p_xstat)) { 1213 info->si_code = CLD_EXITED; 1214 info->si_status = 1215 WEXITSTATUS(p->p_xstat); 1216 } else { 1217 info->si_code = CLD_KILLED; 1218 info->si_status = WTERMSIG(p->p_xstat); 1219 } 1220 info->si_pid = p->p_pid; 1221 info->si_uid = p->p_ucred->cr_uid; 1222 } 1223 1224 /* 1225 * WNOWAIT shortcuts to done here, leaving the 1226 * child on the zombie list. 1227 */ 1228 if (options & WNOWAIT) { 1229 lwkt_reltoken(&p->p_token); 1230 PRELEZOMB(p); 1231 error = 0; 1232 goto done; 1233 } 1234 1235 /* 1236 * If we got the child via a ptrace 'attach', 1237 * we need to give it back to the old parent. 1238 */ 1239 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 1240 p->p_oppid = 0; 1241 proc_reparent(p, t); 1242 ksignal(t, SIGCHLD); 1243 wakeup((caddr_t)t); 1244 PRELE(t); 1245 lwkt_reltoken(&p->p_token); 1246 PRELEZOMB(p); 1247 error = 0; 1248 goto done; 1249 } 1250 1251 /* 1252 * Unlink the proc from its process group so that 1253 * the following operations won't lead to an 1254 * inconsistent state for processes running down 1255 * the zombie list. 1256 */ 1257 proc_remove_zombie(p); 1258 proc_userunmap(p); 1259 lwkt_reltoken(&p->p_token); 1260 leavepgrp(p); 1261 1262 p->p_xstat = 0; 1263 ruadd(&q->p_cru, &p->p_ru); 1264 ruadd(&q->p_cru, &p->p_cru); 1265 1266 /* 1267 * Decrement the count of procs running with this uid. 1268 */ 1269 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 1270 1271 /* 1272 * Free up credentials. p_spin is required to 1273 * avoid races against allproc scans. 1274 */ 1275 spin_lock(&p->p_spin); 1276 cr = p->p_ucred; 1277 p->p_ucred = NULL; 1278 spin_unlock(&p->p_spin); 1279 crfree(cr); 1280 1281 /* 1282 * Remove unused arguments 1283 */ 1284 pa = p->p_args; 1285 p->p_args = NULL; 1286 if (pa && refcount_release(&pa->ar_ref)) { 1287 kfree(pa, M_PARGS); 1288 pa = NULL; 1289 } 1290 1291 ps = p->p_sigacts; 1292 p->p_sigacts = NULL; 1293 if (ps && refcount_release(&ps->ps_refcnt)) { 1294 kfree(ps, M_SUBPROC); 1295 ps = NULL; 1296 } 1297 1298 /* 1299 * Our exitingcount was incremented when the process 1300 * became a zombie, now that the process has been 1301 * removed from (almost) all lists we should be able 1302 * to safely destroy its vmspace. Wait for any current 1303 * holders to go away (so the vmspace remains stable), 1304 * then scrap it. 1305 * 1306 * NOTE: Releasing the parent process (q) p_token 1307 * across the vmspace_exitfree() call is 1308 * important here to reduce stalls on 1309 * interactions with (q) (such as 1310 * fork/exec/wait or 'ps'). 1311 */ 1312 PSTALL(p, "reap4", 1); 1313 lwkt_reltoken(&q->p_token); 1314 vmspace_exitfree(p); 1315 lwkt_gettoken(&q->p_token); 1316 PSTALL(p, "reap5", 1); 1317 1318 /* 1319 * NOTE: We have to officially release ZOMB in order 1320 * to ensure that a racing thread in kern_wait() 1321 * which blocked on ZOMB is woken up. 1322 */ 1323 PRELEZOMB(p); 1324 kfree(p->p_uidpcpu, M_SUBPROC); 1325 kfree(p, M_PROC); 1326 atomic_add_int(&nprocs, -1); 1327 error = 0; 1328 goto done; 1329 } 1330 1331 /* 1332 * Process has not yet exited 1333 */ 1334 if ((p->p_stat == SSTOP || p->p_stat == SCORE) && 1335 (p->p_flags & P_WAITED) == 0 && 1336 (((p->p_flags & P_TRACED) && (options & WTRAPPED)) || 1337 (options & WSTOPPED))) { 1338 lwkt_gettoken(&p->p_token); 1339 if (p->p_pptr != q) { 1340 lwkt_reltoken(&p->p_token); 1341 PRELE(p); 1342 goto loop; 1343 } 1344 if ((p->p_stat != SSTOP && p->p_stat != SCORE) || 1345 (p->p_flags & P_WAITED) != 0 || 1346 ((p->p_flags & P_TRACED) == 0 && 1347 (options & WUNTRACED) == 0)) { 1348 lwkt_reltoken(&p->p_token); 1349 PRELE(p); 1350 goto loop; 1351 } 1352 1353 /* 1354 * Don't set P_WAITED if WNOWAIT specified, leaving 1355 * the process in a waitable state. 1356 */ 1357 if ((options & WNOWAIT) == 0) 1358 p->p_flags |= P_WAITED; 1359 1360 *res = p->p_pid; 1361 *status = W_STOPCODE(p->p_xstat); 1362 /* Zero rusage so we get something consistent. */ 1363 bzero(wrusage, sizeof(*wrusage)); 1364 error = 0; 1365 if (info) { 1366 bzero(info, sizeof(*info)); 1367 if (p->p_flags & P_TRACED) 1368 info->si_code = CLD_TRAPPED; 1369 else 1370 info->si_code = CLD_STOPPED; 1371 info->si_status = WSTOPSIG(p->p_xstat); 1372 } 1373 lwkt_reltoken(&p->p_token); 1374 PRELE(p); 1375 goto done; 1376 } 1377 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) { 1378 lwkt_gettoken(&p->p_token); 1379 if (p->p_pptr != q) { 1380 lwkt_reltoken(&p->p_token); 1381 PRELE(p); 1382 goto loop; 1383 } 1384 if ((p->p_flags & P_CONTINUED) == 0) { 1385 lwkt_reltoken(&p->p_token); 1386 PRELE(p); 1387 goto loop; 1388 } 1389 1390 *res = p->p_pid; 1391 1392 /* 1393 * Don't set P_WAITED if WNOWAIT specified, leaving 1394 * the process in a waitable state. 1395 */ 1396 if ((options & WNOWAIT) == 0) 1397 p->p_flags &= ~P_CONTINUED; 1398 1399 *status = SIGCONT; 1400 error = 0; 1401 if (info) { 1402 bzero(info, sizeof(*info)); 1403 info->si_code = CLD_CONTINUED; 1404 info->si_status = WSTOPSIG(p->p_xstat); 1405 } 1406 lwkt_reltoken(&p->p_token); 1407 PRELE(p); 1408 goto done; 1409 } 1410 PRELE(p); 1411 } 1412 if (nfound == 0) { 1413 error = ECHILD; 1414 goto done; 1415 } 1416 if (options & WNOHANG) { 1417 *res = 0; 1418 error = 0; 1419 goto done; 1420 } 1421 1422 /* 1423 * Wait for signal - interlocked using q->p_waitgen. 1424 */ 1425 error = 0; 1426 while ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) { 1427 tsleep_interlock(q, PCATCH); 1428 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000); 1429 if ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) { 1430 error = tsleep(q, PCATCH | PINTERLOCKED, "wait", 0); 1431 break; 1432 } 1433 } 1434 if (error) { 1435 done: 1436 lwkt_reltoken(&q->p_token); 1437 return (error); 1438 } 1439 goto loop; 1440 } 1441 1442 /* 1443 * Change child's parent process to parent. 1444 * 1445 * p_children/p_sibling requires the parent's token, and 1446 * changing pptr requires the child's token, so we have to 1447 * get three tokens to do this operation. We also need to 1448 * hold pointers that might get ripped out from under us to 1449 * preserve structural integrity. 1450 * 1451 * It is possible to race another reparent or disconnect or other 1452 * similar operation. We must retry when this situation occurs. 1453 * Once we successfully reparent the process we no longer care 1454 * about any races. 1455 */ 1456 void 1457 proc_reparent(struct proc *child, struct proc *parent) 1458 { 1459 struct proc *opp; 1460 1461 PHOLD(parent); 1462 while ((opp = child->p_pptr) != parent) { 1463 PHOLD(opp); 1464 lwkt_gettoken(&opp->p_token); 1465 lwkt_gettoken(&child->p_token); 1466 lwkt_gettoken(&parent->p_token); 1467 if (child->p_pptr != opp) { 1468 lwkt_reltoken(&parent->p_token); 1469 lwkt_reltoken(&child->p_token); 1470 lwkt_reltoken(&opp->p_token); 1471 PRELE(opp); 1472 continue; 1473 } 1474 LIST_REMOVE(child, p_sibling); 1475 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1476 child->p_pptr = parent; 1477 child->p_ppid = parent->p_pid; 1478 lwkt_reltoken(&parent->p_token); 1479 lwkt_reltoken(&child->p_token); 1480 lwkt_reltoken(&opp->p_token); 1481 if (LIST_EMPTY(&opp->p_children)) 1482 wakeup(opp); 1483 PRELE(opp); 1484 break; 1485 } 1486 PRELE(parent); 1487 } 1488 1489 /* 1490 * The next two functions are to handle adding/deleting items on the 1491 * exit callout list 1492 * 1493 * at_exit(): 1494 * Take the arguments given and put them onto the exit callout list, 1495 * However first make sure that it's not already there. 1496 * returns 0 on success. 1497 */ 1498 1499 int 1500 at_exit(exitlist_fn function) 1501 { 1502 struct exitlist *ep; 1503 1504 #ifdef INVARIANTS 1505 /* Be noisy if the programmer has lost track of things */ 1506 if (rm_at_exit(function)) 1507 kprintf("WARNING: exit callout entry (%p) already present\n", 1508 function); 1509 #endif 1510 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 1511 if (ep == NULL) 1512 return (ENOMEM); 1513 ep->function = function; 1514 TAILQ_INSERT_TAIL(&exit_list, ep, next); 1515 return (0); 1516 } 1517 1518 /* 1519 * Scan the exit callout list for the given item and remove it. 1520 * Returns the number of items removed (0 or 1) 1521 */ 1522 int 1523 rm_at_exit(exitlist_fn function) 1524 { 1525 struct exitlist *ep; 1526 1527 TAILQ_FOREACH(ep, &exit_list, next) { 1528 if (ep->function == function) { 1529 TAILQ_REMOVE(&exit_list, ep, next); 1530 kfree(ep, M_ATEXIT); 1531 return(1); 1532 } 1533 } 1534 return (0); 1535 } 1536 1537 /* 1538 * LWP reaper related code. 1539 */ 1540 static void 1541 reaplwps(void *context, int dummy) 1542 { 1543 struct lwplist *lwplist = context; 1544 struct lwp *lp; 1545 int cpu = mycpuid; 1546 1547 lwkt_gettoken(&deadlwp_token[cpu]); 1548 while ((lp = LIST_FIRST(lwplist))) { 1549 LIST_REMOVE(lp, u.lwp_reap_entry); 1550 reaplwp(lp); 1551 } 1552 lwkt_reltoken(&deadlwp_token[cpu]); 1553 } 1554 1555 static void 1556 reaplwp(struct lwp *lp) 1557 { 1558 while (lwp_wait(lp) == 0) 1559 ; 1560 lwp_dispose(lp); 1561 } 1562 1563 static void 1564 deadlwp_init(void) 1565 { 1566 int cpu; 1567 1568 for (cpu = 0; cpu < ncpus; cpu++) { 1569 lwkt_token_init(&deadlwp_token[cpu], "deadlwpl"); 1570 LIST_INIT(&deadlwp_list[cpu]); 1571 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), 1572 M_DEVBUF, M_WAITOK); 1573 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1574 } 1575 } 1576 1577 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1578