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 75 #include <machine/vmm.h> 76 77 static void reaplwps(void *context, int dummy); 78 static void reaplwp(struct lwp *lp); 79 static void killlwps(struct lwp *lp); 80 81 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 82 83 /* 84 * callout list for things to do at exit time 85 */ 86 struct exitlist { 87 exitlist_fn function; 88 TAILQ_ENTRY(exitlist) next; 89 }; 90 91 TAILQ_HEAD(exit_list_head, exitlist); 92 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 93 94 /* 95 * LWP reaper data 96 */ 97 static struct task *deadlwp_task[MAXCPU]; 98 static struct lwplist deadlwp_list[MAXCPU]; 99 static struct lwkt_token deadlwp_token[MAXCPU]; 100 101 void (*linux_task_drop_callback)(thread_t td); 102 void (*linux_proc_drop_callback)(struct proc *p); 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 int 188 killalllwps(int forexec) 189 { 190 struct lwp *lp = curthread->td_lwp; 191 struct proc *p = lp->lwp_proc; 192 int fakestop; 193 194 /* 195 * Interlock against P_WEXIT. Only one of the process's thread 196 * is allowed to do the master exit. 197 */ 198 lwkt_gettoken(&p->p_token); 199 if (p->p_flags & P_WEXIT) { 200 lwkt_reltoken(&p->p_token); 201 return (EALREADY); 202 } 203 p->p_flags |= P_WEXIT; 204 lwkt_gettoken(&lp->lwp_token); 205 206 /* 207 * Set temporary stopped state in case we are racing a coredump. 208 * Otherwise the coredump may hang forever. 209 */ 210 if (lp->lwp_mpflags & LWP_MP_WSTOP) { 211 fakestop = 0; 212 } else { 213 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP); 214 ++p->p_nstopped; 215 fakestop = 1; 216 wakeup(&p->p_nstopped); 217 } 218 219 /* 220 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs 221 */ 222 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 223 if (p->p_nthreads > 1) 224 killlwps(lp); 225 226 /* 227 * Undo temporary stopped state 228 */ 229 if (fakestop && (lp->lwp_mpflags & LWP_MP_WSTOP)) { 230 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WSTOP); 231 --p->p_nstopped; 232 } 233 234 /* 235 * If doing this for an exec, clean up the remaining thread 236 * (us) for continuing operation after all the other threads 237 * have been killed. 238 */ 239 if (forexec) { 240 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 241 p->p_flags &= ~P_WEXIT; 242 } 243 lwkt_reltoken(&lp->lwp_token); 244 lwkt_reltoken(&p->p_token); 245 246 return(0); 247 } 248 249 /* 250 * Kill all LWPs except the current one. Do not try to signal 251 * LWPs which have exited on their own or have already been 252 * signaled. 253 */ 254 static void 255 killlwps(struct lwp *lp) 256 { 257 struct proc *p = lp->lwp_proc; 258 struct lwp *tlp; 259 260 /* 261 * Kill the remaining LWPs. We must send the signal before setting 262 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce 263 * races. tlp must be held across the call as it might block and 264 * allow the target lwp to rip itself out from under our loop. 265 */ 266 FOREACH_LWP_IN_PROC(tlp, p) { 267 LWPHOLD(tlp); 268 lwkt_gettoken(&tlp->lwp_token); 269 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) { 270 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT); 271 lwpsignal(p, tlp, SIGKILL); 272 } 273 lwkt_reltoken(&tlp->lwp_token); 274 LWPRELE(tlp); 275 } 276 277 /* 278 * Wait for everything to clear out. Also make sure any tstop()s 279 * are signalled (we are holding p_token for the interlock). 280 */ 281 wakeup(p); 282 while (p->p_nthreads > 1) 283 tsleep(&p->p_nthreads, 0, "killlwps", 0); 284 } 285 286 /* 287 * Exit: deallocate address space and other resources, change proc state 288 * to zombie, and unlink proc from allproc and parent's lists. Save exit 289 * status and rusage for wait(). Check for child processes and orphan them. 290 */ 291 void 292 exit1(int rv) 293 { 294 struct thread *td = curthread; 295 struct proc *p = td->td_proc; 296 struct lwp *lp = td->td_lwp; 297 struct proc *q; 298 struct proc *pp; 299 struct proc *reproc; 300 struct sysreaper *reap; 301 struct vmspace *vm; 302 struct vnode *vtmp; 303 struct exitlist *ep; 304 int error; 305 306 lwkt_gettoken(&p->p_token); 307 308 if (p->p_pid == 1) { 309 kprintf("init died (signal %d, exit %d)\n", 310 WTERMSIG(rv), WEXITSTATUS(rv)); 311 panic("Going nowhere without my init!"); 312 } 313 varsymset_clean(&p->p_varsymset); 314 lockuninit(&p->p_varsymset.vx_lock); 315 316 /* 317 * Kill all lwps associated with the current process, return an 318 * error if we race another thread trying to do the same thing 319 * and lose the race. 320 */ 321 error = killalllwps(0); 322 if (error) { 323 lwp_exit(0, NULL); 324 /* NOT REACHED */ 325 } 326 327 /* are we a task leader? */ 328 if (p == p->p_leader) { 329 struct kill_args killArgs; 330 killArgs.signum = SIGKILL; 331 q = p->p_peers; 332 while(q) { 333 killArgs.pid = q->p_pid; 334 /* 335 * The interface for kill is better 336 * than the internal signal 337 */ 338 sys_kill(&killArgs); 339 q = q->p_peers; 340 } 341 while (p->p_peers) 342 tsleep((caddr_t)p, 0, "exit1", 0); 343 } 344 345 #ifdef PGINPROF 346 vmsizmon(); 347 #endif 348 STOPEVENT(p, S_EXIT, rv); 349 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */ 350 351 /* 352 * Check if any loadable modules need anything done at process exit. 353 * e.g. SYSV IPC stuff 354 * XXX what if one of these generates an error? 355 */ 356 p->p_xstat = rv; 357 358 /* 359 * XXX: imho, the eventhandler stuff is much cleaner than this. 360 * Maybe we should move everything to use eventhandler. 361 */ 362 TAILQ_FOREACH(ep, &exit_list, next) 363 (*ep->function)(td); 364 365 if (p->p_flags & P_PROFIL) 366 stopprofclock(p); 367 368 SIGEMPTYSET(p->p_siglist); 369 SIGEMPTYSET(lp->lwp_siglist); 370 if (timevalisset(&p->p_realtimer.it_value)) 371 callout_terminate(&p->p_ithandle); 372 373 /* 374 * Reset any sigio structures pointing to us as a result of 375 * F_SETOWN with our pid. 376 */ 377 funsetownlst(&p->p_sigiolst); 378 379 /* 380 * Close open files and release open-file table. 381 * This may block! 382 */ 383 fdfree(p, NULL); 384 385 if (p->p_leader->p_peers) { 386 q = p->p_leader; 387 while(q->p_peers != p) 388 q = q->p_peers; 389 q->p_peers = p->p_peers; 390 wakeup((caddr_t)p->p_leader); 391 } 392 393 /* 394 * XXX Shutdown SYSV semaphores 395 */ 396 semexit(p); 397 398 /* The next two chunks should probably be moved to vmspace_exit. */ 399 vm = p->p_vmspace; 400 401 /* 402 * Clean up data related to virtual kernel operation. Clean up 403 * any vkernel context related to the current lwp now so we can 404 * destroy p_vkernel. 405 */ 406 if (p->p_vkernel) { 407 vkernel_lwp_exit(lp); 408 vkernel_exit(p); 409 } 410 411 /* 412 * Release the user portion of address space. The exitbump prevents 413 * the vmspace from being completely eradicated (using holdcnt). 414 * This releases references to vnodes, which could cause I/O if the 415 * file has been unlinked. We need to do this early enough that 416 * we can still sleep. 417 * 418 * We can't free the entire vmspace as the kernel stack may be mapped 419 * within that space also. 420 * 421 * Processes sharing the same vmspace may exit in one order, and 422 * get cleaned up by vmspace_exit() in a different order. The 423 * last exiting process to reach this point releases as much of 424 * the environment as it can, and the last process cleaned up 425 * by vmspace_exit() (which decrements exitingcnt) cleans up the 426 * remainder. 427 * 428 * NOTE: Releasing p_token around this call is helpful if the 429 * vmspace had a huge RSS. Otherwise some other process 430 * trying to do an allproc or other scan (like 'ps') may 431 * stall for a long time. 432 */ 433 lwkt_reltoken(&p->p_token); 434 vmspace_relexit(vm); 435 lwkt_gettoken(&p->p_token); 436 437 if (SESS_LEADER(p)) { 438 struct session *sp = p->p_session; 439 440 if (sp->s_ttyvp) { 441 /* 442 * We are the controlling process. Signal the 443 * foreground process group, drain the controlling 444 * terminal, and revoke access to the controlling 445 * terminal. 446 * 447 * NOTE: while waiting for the process group to exit 448 * it is possible that one of the processes in the 449 * group will revoke the tty, so the ttyclosesession() 450 * function will re-check sp->s_ttyvp. 451 */ 452 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 453 if (sp->s_ttyp->t_pgrp) 454 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 455 ttywait(sp->s_ttyp); 456 ttyclosesession(sp, 1); /* also revoke */ 457 } 458 /* 459 * Release the tty. If someone has it open via 460 * /dev/tty then close it (since they no longer can 461 * once we've NULL'd it out). 462 */ 463 ttyclosesession(sp, 0); 464 465 /* 466 * s_ttyp is not zero'd; we use this to indicate 467 * that the session once had a controlling terminal. 468 * (for logging and informational purposes) 469 */ 470 } 471 sp->s_leader = NULL; 472 } 473 fixjobc(p, p->p_pgrp, 0); 474 (void)acct_process(p); 475 #ifdef KTRACE 476 /* 477 * release trace file 478 */ 479 if (p->p_tracenode) 480 ktrdestroy(&p->p_tracenode); 481 p->p_traceflag = 0; 482 #endif 483 /* 484 * Release reference to text vnode 485 */ 486 if ((vtmp = p->p_textvp) != NULL) { 487 p->p_textvp = NULL; 488 vrele(vtmp); 489 } 490 491 /* Release namecache handle to text file */ 492 if (p->p_textnch.ncp) 493 cache_drop(&p->p_textnch); 494 495 /* 496 * We have to handle PPWAIT here or proc_move_allproc_zombie() 497 * will block on the PHOLD() the parent is doing. 498 * 499 * We are using the flag as an interlock so an atomic op is 500 * necessary to synchronize with the parent's cpu. 501 */ 502 if (p->p_flags & P_PPWAIT) { 503 if (p->p_pptr && p->p_pptr->p_upmap) 504 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1); 505 atomic_clear_int(&p->p_flags, P_PPWAIT); 506 wakeup(p->p_pptr); 507 } 508 509 /* 510 * Move the process to the zombie list. This will block 511 * until the process p_lock count reaches 0. The process will 512 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 513 * which is called from cpu_proc_exit(). 514 * 515 * Interlock against waiters using p_waitgen. We increment 516 * p_waitgen after completing the move of our process to the 517 * zombie list. 518 * 519 * WARNING: pp becomes stale when we block, clear it now as a 520 * reminder. 521 */ 522 proc_move_allproc_zombie(p); 523 pp = p->p_pptr; 524 atomic_add_long(&pp->p_waitgen, 1); 525 pp = NULL; 526 527 /* 528 * release controlled reaper for exit if we own it and return the 529 * remaining reaper (the one for us), which we will drop after we 530 * are done. 531 */ 532 reap = reaper_exit(p); 533 534 /* 535 * Reparent all of this process's children to the init process or 536 * to the designated reaper. We must hold the reaper's p_token in 537 * order to safely mess with p_children. 538 * 539 * We already hold p->p_token (to remove the children from our list). 540 */ 541 reproc = NULL; 542 q = LIST_FIRST(&p->p_children); 543 if (q) { 544 reproc = reaper_get(reap); 545 lwkt_gettoken(&reproc->p_token); 546 while ((q = LIST_FIRST(&p->p_children)) != NULL) { 547 PHOLD(q); 548 lwkt_gettoken(&q->p_token); 549 if (q != LIST_FIRST(&p->p_children)) { 550 lwkt_reltoken(&q->p_token); 551 PRELE(q); 552 continue; 553 } 554 LIST_REMOVE(q, p_sibling); 555 LIST_INSERT_HEAD(&reproc->p_children, q, p_sibling); 556 q->p_pptr = reproc; 557 q->p_ppid = reproc->p_pid; 558 q->p_sigparent = SIGCHLD; 559 560 /* 561 * Traced processes are killed 562 * since their existence means someone is screwing up. 563 */ 564 if (q->p_flags & P_TRACED) { 565 q->p_flags &= ~P_TRACED; 566 ksignal(q, SIGKILL); 567 } 568 lwkt_reltoken(&q->p_token); 569 PRELE(q); 570 } 571 lwkt_reltoken(&reproc->p_token); 572 wakeup(reproc); 573 } 574 575 /* 576 * Save exit status and final rusage info. We no longer add 577 * child rusage info into self times, wait4() and kern_wait() 578 * handles it in order to properly support wait6(). 579 */ 580 calcru_proc(p, &p->p_ru); 581 /*ruadd(&p->p_ru, &p->p_cru); REMOVED */ 582 583 /* 584 * notify interested parties of our demise. 585 */ 586 KNOTE(&p->p_klist, NOTE_EXIT); 587 588 /* 589 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 590 * flag set, or if the handler is set to SIG_IGN, notify the reaper 591 * instead (it will handle this situation). 592 * 593 * NOTE: The reaper can still be the parent process. 594 * 595 * (must reload pp) 596 */ 597 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 598 if (reproc == NULL) 599 reproc = reaper_get(reap); 600 proc_reparent(p, reproc); 601 } 602 if (reproc) 603 PRELE(reproc); 604 if (reap) 605 reaper_drop(reap); 606 607 /* 608 * Signal (possibly new) parent. 609 */ 610 pp = p->p_pptr; 611 PHOLD(pp); 612 if (p->p_sigparent && pp != initproc) { 613 int sig = p->p_sigparent; 614 615 if (sig != SIGUSR1 && sig != SIGCHLD) 616 sig = SIGCHLD; 617 ksignal(pp, sig); 618 } else { 619 ksignal(pp, SIGCHLD); 620 } 621 p->p_flags &= ~P_TRACED; 622 PRELE(pp); 623 624 /* 625 * cpu_exit is responsible for clearing curproc, since 626 * it is heavily integrated with the thread/switching sequence. 627 * 628 * Other substructures are freed from wait(). 629 */ 630 if (p->p_limit) { 631 struct plimit *rlimit; 632 633 rlimit = p->p_limit; 634 p->p_limit = NULL; 635 plimit_free(rlimit); 636 } 637 638 /* 639 * Finally, call machine-dependent code to release as many of the 640 * lwp's resources as we can and halt execution of this thread. 641 * 642 * pp is a wild pointer now but still the correct wakeup() target. 643 * lwp_exit() only uses it to send the wakeup() signal to the likely 644 * parent. Any reparenting race that occurs will get a signal 645 * automatically and not be an issue. 646 */ 647 lwp_exit(1, pp); 648 } 649 650 /* 651 * Eventually called by every exiting LWP 652 * 653 * p->p_token must be held. mplock may be held and will be released. 654 */ 655 void 656 lwp_exit(int masterexit, void *waddr) 657 { 658 struct thread *td = curthread; 659 struct lwp *lp = td->td_lwp; 660 struct proc *p = lp->lwp_proc; 661 int dowake = 0; 662 663 /* 664 * Release the current user process designation on the process so 665 * the userland scheduler can work in someone else. 666 */ 667 p->p_usched->release_curproc(lp); 668 669 /* 670 * Destroy the per-thread shared page and remove from any pmaps 671 * it resides in. 672 */ 673 lwp_userunmap(lp); 674 675 /* 676 * lwp_exit() may be called without setting LWP_MP_WEXIT, so 677 * make sure it is set here. 678 */ 679 ASSERT_LWKT_TOKEN_HELD(&p->p_token); 680 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 681 682 /* 683 * Clean up any virtualization 684 */ 685 if (lp->lwp_vkernel) 686 vkernel_lwp_exit(lp); 687 688 if (td->td_vmm) 689 vmm_vmdestroy(); 690 691 /* 692 * Clean up select/poll support 693 */ 694 kqueue_terminate(&lp->lwp_kqueue); 695 696 if (td->td_linux_task) 697 linux_task_drop_callback(td); 698 if (masterexit && p->p_linux_mm) 699 linux_proc_drop_callback(p); 700 701 /* 702 * Clean up any syscall-cached ucred or rlimit. 703 */ 704 if (td->td_ucred) { 705 crfree(td->td_ucred); 706 td->td_ucred = NULL; 707 } 708 if (td->td_limit) { 709 struct plimit *rlimit; 710 711 rlimit = td->td_limit; 712 td->td_limit = NULL; 713 plimit_free(rlimit); 714 } 715 716 /* 717 * Cleanup any cached descriptors for this thread 718 */ 719 if (p->p_fd) 720 fexitcache(td); 721 722 /* 723 * Nobody actually wakes us when the lock 724 * count reaches zero, so just wait one tick. 725 */ 726 while (lp->lwp_lock > 0) 727 tsleep(lp, 0, "lwpexit", 1); 728 729 /* Hand down resource usage to our proc */ 730 ruadd(&p->p_ru, &lp->lwp_ru); 731 732 /* 733 * If we don't hold the process until the LWP is reaped wait*() 734 * may try to dispose of its vmspace before all the LWPs have 735 * actually terminated. 736 */ 737 PHOLD(p); 738 739 /* 740 * Do any remaining work that might block on us. We should be 741 * coded such that further blocking is ok after decrementing 742 * p_nthreads but don't take the chance. 743 */ 744 dsched_exit_thread(td); 745 biosched_done(curthread); 746 747 /* 748 * We have to use the reaper for all the LWPs except the one doing 749 * the master exit. The LWP doing the master exit can just be 750 * left on p_lwps and the process reaper will deal with it 751 * synchronously, which is much faster. 752 * 753 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0. 754 * 755 * The process is left held until the reaper calls lwp_dispose() on 756 * the lp (after calling lwp_wait()). 757 */ 758 if (masterexit == 0) { 759 int cpu = mycpuid; 760 761 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 762 --p->p_nthreads; 763 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1) 764 dowake = 1; 765 lwkt_gettoken(&deadlwp_token[cpu]); 766 LIST_INSERT_HEAD(&deadlwp_list[cpu], lp, u.lwp_reap_entry); 767 taskqueue_enqueue(taskqueue_thread[cpu], deadlwp_task[cpu]); 768 lwkt_reltoken(&deadlwp_token[cpu]); 769 } else { 770 --p->p_nthreads; 771 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1) 772 dowake = 1; 773 } 774 775 /* 776 * We no longer need p_token. 777 * 778 * Tell the userland scheduler that we are going away 779 */ 780 lwkt_reltoken(&p->p_token); 781 p->p_usched->heuristic_exiting(lp, p); 782 783 /* 784 * Issue late wakeups after releasing our token to give us a chance 785 * to deschedule and switch away before another cpu in a wait*() 786 * reaps us. This is done as late as possible to reduce contention. 787 */ 788 if (dowake) 789 wakeup(&p->p_nthreads); 790 if (waddr) 791 wakeup(waddr); 792 793 cpu_lwp_exit(); 794 } 795 796 /* 797 * Wait until a lwp is completely dead. The final interlock in this drama 798 * is when TDF_EXITING is set in cpu_thread_exit() just before the final 799 * switchout. 800 * 801 * At the point TDF_EXITING is set a complete exit is accomplished when 802 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two 803 * post-switch interlock flags that can be used to wait for the TDF_ 804 * flags to clear. 805 * 806 * Returns non-zero on success, and zero if the caller needs to retry 807 * the lwp_wait(). 808 */ 809 static int 810 lwp_wait(struct lwp *lp) 811 { 812 struct thread *td = lp->lwp_thread; 813 u_int mpflags; 814 815 KKASSERT(lwkt_preempted_proc() != lp); 816 817 /* 818 * This bit of code uses the thread destruction interlock 819 * managed by lwkt_switch_return() to wait for the lwp's 820 * thread to completely disengage. 821 * 822 * It is possible for us to race another cpu core so we 823 * have to do this correctly. 824 */ 825 for (;;) { 826 mpflags = td->td_mpflags; 827 cpu_ccfence(); 828 if (mpflags & TDF_MP_EXITSIG) 829 break; 830 tsleep_interlock(td, 0); 831 if (atomic_cmpset_int(&td->td_mpflags, mpflags, 832 mpflags | TDF_MP_EXITWAIT)) { 833 tsleep(td, PINTERLOCKED, "lwpxt", 0); 834 } 835 } 836 837 /* 838 * We've already waited for the core exit but there can still 839 * be other refs from e.g. process scans and such. 840 */ 841 if (lp->lwp_lock > 0) { 842 tsleep(lp, 0, "lwpwait1", 1); 843 return(0); 844 } 845 if (td->td_refs) { 846 tsleep(td, 0, "lwpwait2", 1); 847 return(0); 848 } 849 850 /* 851 * Now that we have the thread destruction interlock these flags 852 * really should already be cleaned up, keep a check for safety. 853 * 854 * We can't rip its stack out from under it until TDF_EXITING is 855 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 856 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 857 * will be cleared temporarily if a thread gets preempted. 858 */ 859 while ((td->td_flags & (TDF_RUNNING | 860 TDF_RUNQ | 861 TDF_PREEMPT_LOCK | 862 TDF_EXITING)) != TDF_EXITING) { 863 tsleep(lp, 0, "lwpwait3", 1); 864 return (0); 865 } 866 867 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0, 868 ("lwp_wait: td %p (%s) still on run or sleep queue", 869 td, td->td_comm)); 870 return (1); 871 } 872 873 /* 874 * Release the resources associated with a lwp. 875 * The lwp must be completely dead. 876 */ 877 void 878 lwp_dispose(struct lwp *lp) 879 { 880 struct thread *td = lp->lwp_thread; 881 882 KKASSERT(lwkt_preempted_proc() != lp); 883 KKASSERT(lp->lwp_lock == 0); 884 KKASSERT(td->td_refs == 0); 885 KKASSERT((td->td_flags & (TDF_RUNNING | 886 TDF_RUNQ | 887 TDF_PREEMPT_LOCK | 888 TDF_EXITING)) == TDF_EXITING); 889 890 PRELE(lp->lwp_proc); 891 lp->lwp_proc = NULL; 892 if (td != NULL) { 893 td->td_proc = NULL; 894 td->td_lwp = NULL; 895 lp->lwp_thread = NULL; 896 lwkt_free_thread(td); 897 } 898 kfree(lp, M_LWP); 899 } 900 901 int 902 sys_wait4(struct wait_args *uap) 903 { 904 struct __wrusage wrusage; 905 int error; 906 int status; 907 int options; 908 id_t id; 909 idtype_t idtype; 910 911 options = uap->options | WEXITED | WTRAPPED; 912 id = uap->pid; 913 914 if (id == WAIT_ANY) { 915 idtype = P_ALL; 916 } else if (id == WAIT_MYPGRP) { 917 idtype = P_PGID; 918 id = curproc->p_pgid; 919 } else if (id < 0) { 920 idtype = P_PGID; 921 id = -id; 922 } else { 923 idtype = P_PID; 924 } 925 926 error = kern_wait(idtype, id, &status, options, &wrusage, 927 NULL, &uap->sysmsg_result); 928 929 if (error == 0 && uap->status) 930 error = copyout(&status, uap->status, sizeof(*uap->status)); 931 if (error == 0 && uap->rusage) { 932 ruadd(&wrusage.wru_self, &wrusage.wru_children); 933 error = copyout(&wrusage.wru_self, uap->rusage, sizeof(*uap->rusage)); 934 } 935 return (error); 936 } 937 938 int 939 sys_wait6(struct wait6_args *uap) 940 { 941 struct __wrusage wrusage; 942 siginfo_t info; 943 siginfo_t *infop; 944 int error; 945 int status; 946 int options; 947 id_t id; 948 idtype_t idtype; 949 950 /* 951 * NOTE: wait6() requires WEXITED and WTRAPPED to be specified if 952 * desired. 953 */ 954 options = uap->options; 955 idtype = uap->idtype; 956 id = uap->id; 957 infop = uap->info ? &info : NULL; 958 959 switch(idtype) { 960 case P_PID: 961 case P_PGID: 962 if (id == WAIT_MYPGRP) { 963 idtype = P_PGID; 964 id = curproc->p_pgid; 965 } 966 break; 967 default: 968 /* let kern_wait deal with the remainder */ 969 break; 970 } 971 972 error = kern_wait(idtype, id, &status, options, 973 &wrusage, infop, &uap->sysmsg_result); 974 975 if (error == 0 && uap->status) 976 error = copyout(&status, uap->status, sizeof(*uap->status)); 977 if (error == 0 && uap->wrusage) 978 error = copyout(&wrusage, uap->wrusage, sizeof(*uap->wrusage)); 979 if (error == 0 && uap->info) 980 error = copyout(&info, uap->info, sizeof(*uap->info)); 981 return (error); 982 } 983 984 /* 985 * kernel wait*() system call support 986 */ 987 int 988 kern_wait(idtype_t idtype, id_t id, int *status, int options, 989 struct __wrusage *wrusage, siginfo_t *info, int *res) 990 { 991 struct thread *td = curthread; 992 struct lwp *lp; 993 struct proc *q = td->td_proc; 994 struct proc *p, *t; 995 struct ucred *cr; 996 struct pargs *pa; 997 struct sigacts *ps; 998 int nfound, error; 999 long waitgen; 1000 1001 /* 1002 * Must not have extraneous options. Must have at least one 1003 * matchable option. 1004 */ 1005 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE|WSTOPPED| 1006 WEXITED|WTRAPPED|WNOWAIT)) { 1007 return (EINVAL); 1008 } 1009 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1010 return (EINVAL); 1011 } 1012 1013 /* 1014 * Protect the q->p_children list 1015 */ 1016 lwkt_gettoken(&q->p_token); 1017 loop: 1018 /* 1019 * All sorts of things can change due to blocking so we have to loop 1020 * all the way back up here. 1021 * 1022 * The problem is that if a process group is stopped and the parent 1023 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 1024 * of the child and then stop itself when it tries to return from the 1025 * system call. When the process group is resumed the parent will 1026 * then get the STOP status even though the child has now resumed 1027 * (a followup wait*() will get the CONT status). 1028 * 1029 * Previously the CONT would overwrite the STOP because the tstop 1030 * was handled within tsleep(), and the parent would only see 1031 * the CONT when both are stopped and continued together. This little 1032 * two-line hack restores this effect. 1033 * 1034 * No locks are held so we can safely block the process here. 1035 */ 1036 if (STOPLWP(q, td->td_lwp)) 1037 tstop(); 1038 1039 nfound = 0; 1040 1041 /* 1042 * Loop on children. 1043 * 1044 * NOTE: We don't want to break q's p_token in the loop for the 1045 * case where no children are found or we risk breaking the 1046 * interlock between child and parent. 1047 */ 1048 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000); 1049 LIST_FOREACH(p, &q->p_children, p_sibling) { 1050 /* 1051 * Skip children that another thread is already uninterruptably 1052 * reaping. 1053 */ 1054 if (PWAITRES_PENDING(p)) 1055 continue; 1056 1057 /* 1058 * Filter, (p) will be held on fall-through. Try to optimize 1059 * this to avoid the atomic op until we are pretty sure we 1060 * want this process. 1061 */ 1062 switch(idtype) { 1063 case P_ALL: 1064 PHOLD(p); 1065 break; 1066 case P_PID: 1067 if (p->p_pid != (pid_t)id) 1068 continue; 1069 PHOLD(p); 1070 break; 1071 case P_PGID: 1072 if (p->p_pgid != (pid_t)id) 1073 continue; 1074 PHOLD(p); 1075 break; 1076 case P_SID: 1077 PHOLD(p); 1078 if (p->p_session && p->p_session->s_sid != (pid_t)id) { 1079 PRELE(p); 1080 continue; 1081 } 1082 break; 1083 case P_UID: 1084 PHOLD(p); 1085 if (p->p_ucred->cr_uid != (uid_t)id) { 1086 PRELE(p); 1087 continue; 1088 } 1089 break; 1090 case P_GID: 1091 PHOLD(p); 1092 if (p->p_ucred->cr_gid != (gid_t)id) { 1093 PRELE(p); 1094 continue; 1095 } 1096 break; 1097 case P_JAILID: 1098 PHOLD(p); 1099 if (p->p_ucred->cr_prison && 1100 p->p_ucred->cr_prison->pr_id != (int)id) { 1101 PRELE(p); 1102 continue; 1103 } 1104 break; 1105 default: 1106 /* unsupported filter */ 1107 continue; 1108 } 1109 /* (p) is held at this point */ 1110 1111 /* 1112 * This special case handles a kthread spawned by linux_clone 1113 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1114 * functions need to be able to distinguish between waiting 1115 * on a process and waiting on a thread. It is a thread if 1116 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1117 * signifies we want to wait for threads and not processes. 1118 */ 1119 if ((p->p_sigparent != SIGCHLD) ^ 1120 ((options & WLINUXCLONE) != 0)) { 1121 PRELE(p); 1122 continue; 1123 } 1124 1125 nfound++; 1126 if (p->p_stat == SZOMB && (options & WEXITED)) { 1127 /* 1128 * We may go into SZOMB with threads still present. 1129 * We must wait for them to exit before we can reap 1130 * the master thread, otherwise we may race reaping 1131 * non-master threads. 1132 * 1133 * Only this routine can remove a process from 1134 * the zombie list and destroy it. 1135 * 1136 * This function will fail after sleeping if another 1137 * thread owns the zombie lock. This function will 1138 * fail immediately or after sleeping if another 1139 * thread owns or obtains ownership of the reap via 1140 * WAITRES. 1141 */ 1142 if (PHOLDZOMB(p)) { 1143 PRELE(p); 1144 goto loop; 1145 } 1146 lwkt_gettoken(&p->p_token); 1147 if (p->p_pptr != q) { 1148 lwkt_reltoken(&p->p_token); 1149 PRELE(p); 1150 PRELEZOMB(p); 1151 goto loop; 1152 } 1153 1154 /* 1155 * We are the reaper, from this point on the reap 1156 * cannot be aborted. 1157 */ 1158 PWAITRES_SET(p); 1159 while (p->p_nthreads > 0) { 1160 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 1161 } 1162 1163 /* 1164 * Reap any LWPs left in p->p_lwps. This is usually 1165 * just the last LWP. This must be done before 1166 * we loop on p_lock since the lwps hold a ref on 1167 * it as a vmspace interlock. 1168 * 1169 * Once that is accomplished p_nthreads had better 1170 * be zero. 1171 */ 1172 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 1173 /* 1174 * Make sure no one is using this lwp, before 1175 * it is removed from the tree. If we didn't 1176 * wait it here, lwp tree iteration with 1177 * blocking operation would be broken. 1178 */ 1179 while (lp->lwp_lock > 0) 1180 tsleep(lp, 0, "zomblwp", 1); 1181 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 1182 reaplwp(lp); 1183 } 1184 KKASSERT(p->p_nthreads == 0); 1185 1186 /* 1187 * Don't do anything really bad until all references 1188 * to the process go away. This may include other 1189 * LWPs which are still in the process of being 1190 * reaped. We can't just pull the rug out from under 1191 * them because they may still be using the VM space. 1192 * 1193 * Certain kernel facilities such as /proc will also 1194 * put a hold on the process for short periods of 1195 * time. 1196 */ 1197 PRELE(p); /* from top of loop */ 1198 PSTALL(p, "reap3", 1); /* 1 ref (for PZOMBHOLD) */ 1199 1200 /* Take care of our return values. */ 1201 *res = p->p_pid; 1202 1203 *status = p->p_xstat; 1204 wrusage->wru_self = p->p_ru; 1205 wrusage->wru_children = p->p_cru; 1206 1207 if (info) { 1208 bzero(info, sizeof(*info)); 1209 info->si_errno = 0; 1210 info->si_signo = SIGCHLD; 1211 if (WIFEXITED(p->p_xstat)) { 1212 info->si_code = CLD_EXITED; 1213 info->si_status = 1214 WEXITSTATUS(p->p_xstat); 1215 } else { 1216 info->si_code = CLD_KILLED; 1217 info->si_status = WTERMSIG(p->p_xstat); 1218 } 1219 info->si_pid = p->p_pid; 1220 info->si_uid = p->p_ucred->cr_uid; 1221 } 1222 1223 /* 1224 * WNOWAIT shortcuts to done here, leaving the 1225 * child on the zombie list. 1226 */ 1227 if (options & WNOWAIT) { 1228 lwkt_reltoken(&p->p_token); 1229 PRELEZOMB(p); 1230 error = 0; 1231 goto done; 1232 } 1233 1234 /* 1235 * If we got the child via a ptrace 'attach', 1236 * we need to give it back to the old parent. 1237 */ 1238 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 1239 p->p_oppid = 0; 1240 proc_reparent(p, t); 1241 ksignal(t, SIGCHLD); 1242 wakeup((caddr_t)t); 1243 PRELE(t); 1244 lwkt_reltoken(&p->p_token); 1245 PRELEZOMB(p); 1246 error = 0; 1247 goto done; 1248 } 1249 1250 /* 1251 * Unlink the proc from its process group so that 1252 * the following operations won't lead to an 1253 * inconsistent state for processes running down 1254 * the zombie list. 1255 */ 1256 proc_remove_zombie(p); 1257 proc_userunmap(p); 1258 lwkt_reltoken(&p->p_token); 1259 leavepgrp(p); 1260 1261 p->p_xstat = 0; 1262 ruadd(&q->p_cru, &p->p_ru); 1263 ruadd(&q->p_cru, &p->p_cru); 1264 1265 /* 1266 * Decrement the count of procs running with this uid. 1267 */ 1268 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 1269 1270 /* 1271 * Free up credentials. p_spin is required to 1272 * avoid races against allproc scans. 1273 */ 1274 spin_lock(&p->p_spin); 1275 cr = p->p_ucred; 1276 p->p_ucred = NULL; 1277 spin_unlock(&p->p_spin); 1278 crfree(cr); 1279 1280 /* 1281 * Remove unused arguments 1282 */ 1283 pa = p->p_args; 1284 p->p_args = NULL; 1285 if (pa && refcount_release(&pa->ar_ref)) { 1286 kfree(pa, M_PARGS); 1287 pa = NULL; 1288 } 1289 1290 ps = p->p_sigacts; 1291 p->p_sigacts = NULL; 1292 if (ps && refcount_release(&ps->ps_refcnt)) { 1293 kfree(ps, M_SUBPROC); 1294 ps = NULL; 1295 } 1296 1297 /* 1298 * Our exitingcount was incremented when the process 1299 * became a zombie, now that the process has been 1300 * removed from (almost) all lists we should be able 1301 * to safely destroy its vmspace. Wait for any current 1302 * holders to go away (so the vmspace remains stable), 1303 * then scrap it. 1304 * 1305 * NOTE: Releasing the parent process (q) p_token 1306 * across the vmspace_exitfree() call is 1307 * important here to reduce stalls on 1308 * interactions with (q) (such as 1309 * fork/exec/wait or 'ps'). 1310 */ 1311 PSTALL(p, "reap4", 1); 1312 lwkt_reltoken(&q->p_token); 1313 vmspace_exitfree(p); 1314 lwkt_gettoken(&q->p_token); 1315 PSTALL(p, "reap5", 1); 1316 1317 /* 1318 * NOTE: We have to officially release ZOMB in order 1319 * to ensure that a racing thread in kern_wait() 1320 * which blocked on ZOMB is woken up. 1321 */ 1322 PRELEZOMB(p); 1323 kfree(p->p_uidpcpu, M_SUBPROC); 1324 kfree(p, M_PROC); 1325 atomic_add_int(&nprocs, -1); 1326 error = 0; 1327 goto done; 1328 } 1329 1330 /* 1331 * Process has not yet exited 1332 */ 1333 if ((p->p_stat == SSTOP || p->p_stat == SCORE) && 1334 (p->p_flags & P_WAITED) == 0 && 1335 (((p->p_flags & P_TRACED) && (options & WTRAPPED)) || 1336 (options & WSTOPPED))) { 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_stat != SSTOP && p->p_stat != SCORE) || 1344 (p->p_flags & P_WAITED) != 0 || 1345 ((p->p_flags & P_TRACED) == 0 && 1346 (options & WUNTRACED) == 0)) { 1347 lwkt_reltoken(&p->p_token); 1348 PRELE(p); 1349 goto loop; 1350 } 1351 1352 /* 1353 * Don't set P_WAITED if WNOWAIT specified, leaving 1354 * the process in a waitable state. 1355 */ 1356 if ((options & WNOWAIT) == 0) 1357 p->p_flags |= P_WAITED; 1358 1359 *res = p->p_pid; 1360 *status = W_STOPCODE(p->p_xstat); 1361 /* Zero rusage so we get something consistent. */ 1362 bzero(wrusage, sizeof(*wrusage)); 1363 error = 0; 1364 if (info) { 1365 bzero(info, sizeof(*info)); 1366 if (p->p_flags & P_TRACED) 1367 info->si_code = CLD_TRAPPED; 1368 else 1369 info->si_code = CLD_STOPPED; 1370 info->si_status = WSTOPSIG(p->p_xstat); 1371 } 1372 lwkt_reltoken(&p->p_token); 1373 PRELE(p); 1374 goto done; 1375 } 1376 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) { 1377 lwkt_gettoken(&p->p_token); 1378 if (p->p_pptr != q) { 1379 lwkt_reltoken(&p->p_token); 1380 PRELE(p); 1381 goto loop; 1382 } 1383 if ((p->p_flags & P_CONTINUED) == 0) { 1384 lwkt_reltoken(&p->p_token); 1385 PRELE(p); 1386 goto loop; 1387 } 1388 1389 *res = p->p_pid; 1390 1391 /* 1392 * Don't set P_WAITED if WNOWAIT specified, leaving 1393 * the process in a waitable state. 1394 */ 1395 if ((options & WNOWAIT) == 0) 1396 p->p_flags &= ~P_CONTINUED; 1397 1398 *status = SIGCONT; 1399 error = 0; 1400 if (info) { 1401 bzero(info, sizeof(*info)); 1402 info->si_code = CLD_CONTINUED; 1403 info->si_status = WSTOPSIG(p->p_xstat); 1404 } 1405 lwkt_reltoken(&p->p_token); 1406 PRELE(p); 1407 goto done; 1408 } 1409 PRELE(p); 1410 } 1411 if (nfound == 0) { 1412 error = ECHILD; 1413 goto done; 1414 } 1415 if (options & WNOHANG) { 1416 *res = 0; 1417 error = 0; 1418 goto done; 1419 } 1420 1421 /* 1422 * Wait for signal - interlocked using q->p_waitgen. 1423 */ 1424 error = 0; 1425 while ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) { 1426 tsleep_interlock(q, PCATCH); 1427 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000); 1428 if ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) { 1429 error = tsleep(q, PCATCH | PINTERLOCKED, "wait", 0); 1430 break; 1431 } 1432 } 1433 if (error) { 1434 done: 1435 lwkt_reltoken(&q->p_token); 1436 return (error); 1437 } 1438 goto loop; 1439 } 1440 1441 /* 1442 * Change child's parent process to parent. 1443 * 1444 * p_children/p_sibling requires the parent's token, and 1445 * changing pptr requires the child's token, so we have to 1446 * get three tokens to do this operation. We also need to 1447 * hold pointers that might get ripped out from under us to 1448 * preserve structural integrity. 1449 * 1450 * It is possible to race another reparent or disconnect or other 1451 * similar operation. We must retry when this situation occurs. 1452 * Once we successfully reparent the process we no longer care 1453 * about any races. 1454 */ 1455 void 1456 proc_reparent(struct proc *child, struct proc *parent) 1457 { 1458 struct proc *opp; 1459 1460 PHOLD(parent); 1461 while ((opp = child->p_pptr) != parent) { 1462 PHOLD(opp); 1463 lwkt_gettoken(&opp->p_token); 1464 lwkt_gettoken(&child->p_token); 1465 lwkt_gettoken(&parent->p_token); 1466 if (child->p_pptr != opp) { 1467 lwkt_reltoken(&parent->p_token); 1468 lwkt_reltoken(&child->p_token); 1469 lwkt_reltoken(&opp->p_token); 1470 PRELE(opp); 1471 continue; 1472 } 1473 LIST_REMOVE(child, p_sibling); 1474 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1475 child->p_pptr = parent; 1476 child->p_ppid = parent->p_pid; 1477 lwkt_reltoken(&parent->p_token); 1478 lwkt_reltoken(&child->p_token); 1479 lwkt_reltoken(&opp->p_token); 1480 if (LIST_EMPTY(&opp->p_children)) 1481 wakeup(opp); 1482 PRELE(opp); 1483 break; 1484 } 1485 PRELE(parent); 1486 } 1487 1488 /* 1489 * The next two functions are to handle adding/deleting items on the 1490 * exit callout list 1491 * 1492 * at_exit(): 1493 * Take the arguments given and put them onto the exit callout list, 1494 * However first make sure that it's not already there. 1495 * returns 0 on success. 1496 */ 1497 1498 int 1499 at_exit(exitlist_fn function) 1500 { 1501 struct exitlist *ep; 1502 1503 #ifdef INVARIANTS 1504 /* Be noisy if the programmer has lost track of things */ 1505 if (rm_at_exit(function)) 1506 kprintf("WARNING: exit callout entry (%p) already present\n", 1507 function); 1508 #endif 1509 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 1510 if (ep == NULL) 1511 return (ENOMEM); 1512 ep->function = function; 1513 TAILQ_INSERT_TAIL(&exit_list, ep, next); 1514 return (0); 1515 } 1516 1517 /* 1518 * Scan the exit callout list for the given item and remove it. 1519 * Returns the number of items removed (0 or 1) 1520 */ 1521 int 1522 rm_at_exit(exitlist_fn function) 1523 { 1524 struct exitlist *ep; 1525 1526 TAILQ_FOREACH(ep, &exit_list, next) { 1527 if (ep->function == function) { 1528 TAILQ_REMOVE(&exit_list, ep, next); 1529 kfree(ep, M_ATEXIT); 1530 return(1); 1531 } 1532 } 1533 return (0); 1534 } 1535 1536 /* 1537 * LWP reaper related code. 1538 */ 1539 static void 1540 reaplwps(void *context, int dummy) 1541 { 1542 struct lwplist *lwplist = context; 1543 struct lwp *lp; 1544 int cpu = mycpuid; 1545 1546 lwkt_gettoken(&deadlwp_token[cpu]); 1547 while ((lp = LIST_FIRST(lwplist))) { 1548 LIST_REMOVE(lp, u.lwp_reap_entry); 1549 reaplwp(lp); 1550 } 1551 lwkt_reltoken(&deadlwp_token[cpu]); 1552 } 1553 1554 static void 1555 reaplwp(struct lwp *lp) 1556 { 1557 while (lwp_wait(lp) == 0) 1558 ; 1559 lwp_dispose(lp); 1560 } 1561 1562 static void 1563 deadlwp_init(void) 1564 { 1565 int cpu; 1566 1567 for (cpu = 0; cpu < ncpus; cpu++) { 1568 lwkt_token_init(&deadlwp_token[cpu], "deadlwpl"); 1569 LIST_INIT(&deadlwp_list[cpu]); 1570 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), 1571 M_DEVBUF, M_WAITOK); 1572 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1573 } 1574 } 1575 1576 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1577