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/sysmsg.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 static void reaplwps(void *context, int dummy); 76 static void reaplwp(struct lwp *lp); 77 static void killlwps(struct lwp *lp); 78 79 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 80 81 /* 82 * callout list for things to do at exit time 83 */ 84 struct exitlist { 85 exitlist_fn function; 86 TAILQ_ENTRY(exitlist) next; 87 }; 88 89 TAILQ_HEAD(exit_list_head, exitlist); 90 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 91 92 /* 93 * LWP reaper data 94 */ 95 static struct task *deadlwp_task[MAXCPU]; 96 static struct lwplist deadlwp_list[MAXCPU]; 97 static struct lwkt_token deadlwp_token[MAXCPU]; 98 99 void (*linux_task_drop_callback)(thread_t td); 100 void (*linux_proc_drop_callback)(struct proc *p); 101 102 /* 103 * exit -- 104 * Death of process. 105 * 106 * SYS_EXIT_ARGS(int rval) 107 */ 108 int 109 sys_exit(struct sysmsg *sysmsg, const 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 sysmsg *sysmsg, const 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 sysmsg sysmsg; 328 329 sysmsg.extargs.kill.signum = SIGKILL; 330 q = p->p_peers; 331 while(q) { 332 sysmsg.extargs.kill.pid = q->p_pid; 333 /* 334 * The interface for kill is better 335 * than the internal signal 336 */ 337 sys_kill(&sysmsg, &sysmsg.extargs.kill); 338 q = q->p_peers; 339 } 340 while (p->p_peers) 341 tsleep((caddr_t)p, 0, "exit1", 0); 342 } 343 344 #ifdef PGINPROF 345 vmsizmon(); 346 #endif 347 STOPEVENT(p, S_EXIT, rv); 348 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */ 349 350 /* 351 * Check if any loadable modules need anything done at process exit. 352 * e.g. SYSV IPC stuff 353 * XXX what if one of these generates an error? 354 */ 355 p->p_xstat = rv; 356 357 /* 358 * XXX: imho, the eventhandler stuff is much cleaner than this. 359 * Maybe we should move everything to use eventhandler. 360 */ 361 TAILQ_FOREACH(ep, &exit_list, next) 362 (*ep->function)(td); 363 364 if (p->p_flags & P_PROFIL) 365 stopprofclock(p); 366 367 SIGEMPTYSET(p->p_siglist); 368 SIGEMPTYSET(lp->lwp_siglist); 369 if (timevalisset(&p->p_realtimer.it_value)) 370 callout_terminate(&p->p_ithandle); 371 372 /* 373 * Reset any sigio structures pointing to us as a result of 374 * F_SETOWN with our pid. 375 */ 376 funsetownlst(&p->p_sigiolst); 377 378 /* 379 * Close open files and release open-file table. 380 * This may block! 381 */ 382 fdfree(p, NULL); 383 384 if (p->p_leader->p_peers) { 385 q = p->p_leader; 386 while(q->p_peers != p) 387 q = q->p_peers; 388 q->p_peers = p->p_peers; 389 wakeup((caddr_t)p->p_leader); 390 } 391 392 /* 393 * XXX Shutdown SYSV semaphores 394 */ 395 semexit(p); 396 397 /* The next two chunks should probably be moved to vmspace_exit. */ 398 vm = p->p_vmspace; 399 400 /* 401 * Clean up data related to virtual kernel operation. Clean up 402 * any vkernel context related to the current lwp now so we can 403 * destroy p_vkernel. 404 */ 405 if (p->p_vkernel) { 406 vkernel_lwp_exit(lp); 407 vkernel_exit(p); 408 } 409 410 /* 411 * Release the user portion of address space. The exitbump prevents 412 * the vmspace from being completely eradicated (using holdcnt). 413 * This releases references to vnodes, which could cause I/O if the 414 * file has been unlinked. We need to do this early enough that 415 * we can still sleep. 416 * 417 * We can't free the entire vmspace as the kernel stack may be mapped 418 * within that space also. 419 * 420 * Processes sharing the same vmspace may exit in one order, and 421 * get cleaned up by vmspace_exit() in a different order. The 422 * last exiting process to reach this point releases as much of 423 * the environment as it can, and the last process cleaned up 424 * by vmspace_exit() (which decrements exitingcnt) cleans up the 425 * remainder. 426 * 427 * NOTE: Releasing p_token around this call is helpful if the 428 * vmspace had a huge RSS. Otherwise some other process 429 * trying to do an allproc or other scan (like 'ps') may 430 * stall for a long time. 431 */ 432 lwkt_reltoken(&p->p_token); 433 vmspace_relexit(vm); 434 lwkt_gettoken(&p->p_token); 435 436 if (SESS_LEADER(p)) { 437 struct session *sp = p->p_session; 438 439 if (sp->s_ttyvp) { 440 /* 441 * We are the controlling process. Signal the 442 * foreground process group, drain the controlling 443 * terminal, and revoke access to the controlling 444 * terminal. 445 * 446 * NOTE: While waiting for the process group to exit 447 * it is possible that one of the processes in 448 * the group will revoke the tty, so the 449 * ttyclosesession() function will re-check 450 * sp->s_ttyvp. 451 * 452 * NOTE: Force a timeout of one second when draining 453 * the controlling terminal. PCATCH won't work 454 * in exit1(). 455 */ 456 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 457 if (sp->s_ttyp->t_pgrp) 458 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 459 sp->s_ttyp->t_timeout = hz; 460 ttywait(sp->s_ttyp); 461 ttyclosesession(sp, 1); /* also revoke */ 462 } 463 464 /* 465 * Release the tty. If someone has it open via 466 * /dev/tty then close it (since they no longer can 467 * once we've NULL'd it out). 468 */ 469 ttyclosesession(sp, 0); 470 471 /* 472 * s_ttyp is not zero'd; we use this to indicate 473 * that the session once had a controlling terminal. 474 * (for logging and informational purposes) 475 */ 476 } 477 sp->s_leader = NULL; 478 } 479 fixjobc(p, p->p_pgrp, 0); 480 (void)acct_process(p); 481 #ifdef KTRACE 482 /* 483 * release trace file 484 */ 485 if (p->p_tracenode) 486 ktrdestroy(&p->p_tracenode); 487 p->p_traceflag = 0; 488 #endif 489 /* 490 * Release reference to text vnode 491 */ 492 if ((vtmp = p->p_textvp) != NULL) { 493 p->p_textvp = NULL; 494 vrele(vtmp); 495 } 496 497 /* Release namecache handle to text file */ 498 if (p->p_textnch.ncp) 499 cache_drop(&p->p_textnch); 500 501 /* 502 * We have to handle PPWAIT here or proc_move_allproc_zombie() 503 * will block on the PHOLD() the parent is doing. 504 * 505 * We are using the flag as an interlock so an atomic op is 506 * necessary to synchronize with the parent's cpu. 507 */ 508 if (p->p_flags & P_PPWAIT) { 509 if (p->p_pptr && p->p_pptr->p_upmap) 510 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1); 511 atomic_clear_int(&p->p_flags, P_PPWAIT); 512 wakeup(p->p_pptr); 513 } 514 515 /* 516 * Move the process to the zombie list. This will block 517 * until the process p_lock count reaches 0. The process will 518 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 519 * which is called from cpu_proc_exit(). 520 * 521 * Interlock against waiters using p_waitgen. We increment 522 * p_waitgen after completing the move of our process to the 523 * zombie list. 524 * 525 * WARNING: pp becomes stale when we block, clear it now as a 526 * reminder. 527 */ 528 proc_move_allproc_zombie(p); 529 pp = p->p_pptr; 530 atomic_add_long(&pp->p_waitgen, 1); 531 pp = NULL; 532 533 /* 534 * release controlled reaper for exit if we own it and return the 535 * remaining reaper (the one for us), which we will drop after we 536 * are done. 537 */ 538 reap = reaper_exit(p); 539 540 /* 541 * Reparent all of this process's children to the init process or 542 * to the designated reaper. We must hold the reaper's p_token in 543 * order to safely mess with p_children. 544 * 545 * Issue the p_deathsig signal to children that request it. 546 * 547 * We already hold p->p_token (to remove the children from our list). 548 */ 549 reproc = NULL; 550 q = LIST_FIRST(&p->p_children); 551 if (q) { 552 reproc = reaper_get(reap); 553 lwkt_gettoken(&reproc->p_token); 554 while ((q = LIST_FIRST(&p->p_children)) != NULL) { 555 PHOLD(q); 556 lwkt_gettoken(&q->p_token); 557 if (q != LIST_FIRST(&p->p_children)) { 558 lwkt_reltoken(&q->p_token); 559 PRELE(q); 560 continue; 561 } 562 LIST_REMOVE(q, p_sibling); 563 LIST_INSERT_HEAD(&reproc->p_children, q, p_sibling); 564 q->p_pptr = reproc; 565 q->p_ppid = reproc->p_pid; 566 q->p_sigparent = SIGCHLD; 567 568 /* 569 * Traced processes are killed 570 * since their existence means someone is screwing up. 571 */ 572 if (q->p_flags & P_TRACED) { 573 q->p_flags &= ~P_TRACED; 574 ksignal(q, SIGKILL); 575 } 576 577 /* 578 * Issue p_deathsig to children that request it 579 */ 580 if (q->p_deathsig) 581 ksignal(q, q->p_deathsig); 582 lwkt_reltoken(&q->p_token); 583 PRELE(q); 584 } 585 lwkt_reltoken(&reproc->p_token); 586 wakeup(reproc); 587 } 588 589 /* 590 * Save exit status and final rusage info. We no longer add 591 * child rusage info into self times, wait4() and kern_wait() 592 * handles it in order to properly support wait6(). 593 */ 594 calcru_proc(p, &p->p_ru); 595 /*ruadd(&p->p_ru, &p->p_cru); REMOVED */ 596 597 /* 598 * notify interested parties of our demise. 599 */ 600 KNOTE(&p->p_klist, NOTE_EXIT); 601 602 /* 603 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 604 * flag set, or if the handler is set to SIG_IGN, notify the reaper 605 * instead (it will handle this situation). 606 * 607 * NOTE: The reaper can still be the parent process. 608 * 609 * (must reload pp) 610 */ 611 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 612 if (reproc == NULL) 613 reproc = reaper_get(reap); 614 proc_reparent(p, reproc); 615 } 616 if (reproc) 617 PRELE(reproc); 618 if (reap) 619 reaper_drop(reap); 620 621 /* 622 * Signal (possibly new) parent. 623 */ 624 pp = p->p_pptr; 625 PHOLD(pp); 626 if (p->p_sigparent && pp != initproc) { 627 int sig = p->p_sigparent; 628 629 if (sig != SIGUSR1 && sig != SIGCHLD) 630 sig = SIGCHLD; 631 ksignal(pp, sig); 632 } else { 633 ksignal(pp, SIGCHLD); 634 } 635 p->p_flags &= ~P_TRACED; 636 PRELE(pp); 637 638 /* 639 * cpu_exit is responsible for clearing curproc, since 640 * it is heavily integrated with the thread/switching sequence. 641 * 642 * Other substructures are freed from wait(). 643 */ 644 if (p->p_limit) { 645 struct plimit *rlimit; 646 647 rlimit = p->p_limit; 648 p->p_limit = NULL; 649 plimit_free(rlimit); 650 } 651 652 /* 653 * Finally, call machine-dependent code to release as many of the 654 * lwp's resources as we can and halt execution of this thread. 655 * 656 * pp is a wild pointer now but still the correct wakeup() target. 657 * lwp_exit() only uses it to send the wakeup() signal to the likely 658 * parent. Any reparenting race that occurs will get a signal 659 * automatically and not be an issue. 660 */ 661 lwp_exit(1, pp); 662 } 663 664 /* 665 * Eventually called by every exiting LWP 666 * 667 * p->p_token must be held. mplock may be held and will be released. 668 */ 669 void 670 lwp_exit(int masterexit, void *waddr) 671 { 672 struct thread *td = curthread; 673 struct lwp *lp = td->td_lwp; 674 struct proc *p = lp->lwp_proc; 675 int dowake = 0; 676 677 /* 678 * Release the current user process designation on the process so 679 * the userland scheduler can work in someone else. 680 */ 681 p->p_usched->release_curproc(lp); 682 683 /* 684 * Destroy the per-thread shared page and remove from any pmaps 685 * it resides in. 686 */ 687 lwp_userunmap(lp); 688 689 /* 690 * lwp_exit() may be called without setting LWP_MP_WEXIT, so 691 * make sure it is set here. 692 */ 693 ASSERT_LWKT_TOKEN_HELD(&p->p_token); 694 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 695 696 /* 697 * Clean up any virtualization 698 */ 699 if (lp->lwp_vkernel) 700 vkernel_lwp_exit(lp); 701 702 /* 703 * Clean up select/poll support 704 */ 705 kqueue_terminate(&lp->lwp_kqueue); 706 707 if (td->td_linux_task) 708 linux_task_drop_callback(td); 709 if (masterexit && p->p_linux_mm) 710 linux_proc_drop_callback(p); 711 712 /* 713 * Clean up any syscall-cached ucred or rlimit. 714 */ 715 if (td->td_ucred) { 716 crfree(td->td_ucred); 717 td->td_ucred = NULL; 718 } 719 if (td->td_limit) { 720 struct plimit *rlimit; 721 722 rlimit = td->td_limit; 723 td->td_limit = NULL; 724 plimit_free(rlimit); 725 } 726 727 /* 728 * Cleanup any cached descriptors for this thread 729 */ 730 if (p->p_fd) 731 fexitcache(td); 732 733 /* 734 * Nobody actually wakes us when the lock 735 * count reaches zero, so just wait one tick. 736 */ 737 while (lp->lwp_lock > 0) 738 tsleep(lp, 0, "lwpexit", 1); 739 740 /* Hand down resource usage to our proc */ 741 ruadd(&p->p_ru, &lp->lwp_ru); 742 743 /* 744 * If we don't hold the process until the LWP is reaped wait*() 745 * may try to dispose of its vmspace before all the LWPs have 746 * actually terminated. 747 */ 748 PHOLD(p); 749 750 /* 751 * Do any remaining work that might block on us. We should be 752 * coded such that further blocking is ok after decrementing 753 * p_nthreads but don't take the chance. 754 */ 755 dsched_exit_thread(td); 756 biosched_done(curthread); 757 758 /* 759 * We have to use the reaper for all the LWPs except the one doing 760 * the master exit. The LWP doing the master exit can just be 761 * left on p_lwps and the process reaper will deal with it 762 * synchronously, which is much faster. 763 * 764 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0. 765 * 766 * The process is left held until the reaper calls lwp_dispose() on 767 * the lp (after calling lwp_wait()). 768 */ 769 if (masterexit == 0) { 770 int cpu = mycpuid; 771 772 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 773 --p->p_nthreads; 774 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1) 775 dowake = 1; 776 lwkt_gettoken(&deadlwp_token[cpu]); 777 LIST_INSERT_HEAD(&deadlwp_list[cpu], lp, u.lwp_reap_entry); 778 taskqueue_enqueue(taskqueue_thread[cpu], deadlwp_task[cpu]); 779 lwkt_reltoken(&deadlwp_token[cpu]); 780 } else { 781 --p->p_nthreads; 782 if ((p->p_flags & P_MAYBETHREADED) && p->p_nthreads <= 1) 783 dowake = 1; 784 } 785 786 /* 787 * We no longer need p_token. 788 * 789 * Tell the userland scheduler that we are going away 790 */ 791 lwkt_reltoken(&p->p_token); 792 p->p_usched->heuristic_exiting(lp, p); 793 794 /* 795 * Issue late wakeups after releasing our token to give us a chance 796 * to deschedule and switch away before another cpu in a wait*() 797 * reaps us. This is done as late as possible to reduce contention. 798 */ 799 if (dowake) 800 wakeup(&p->p_nthreads); 801 if (waddr) 802 wakeup(waddr); 803 804 cpu_lwp_exit(); 805 } 806 807 /* 808 * Wait until a lwp is completely dead. The final interlock in this drama 809 * is when TDF_EXITING is set in cpu_thread_exit() just before the final 810 * switchout. 811 * 812 * At the point TDF_EXITING is set a complete exit is accomplished when 813 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two 814 * post-switch interlock flags that can be used to wait for the TDF_ 815 * flags to clear. 816 * 817 * Returns non-zero on success, and zero if the caller needs to retry 818 * the lwp_wait(). 819 */ 820 static int 821 lwp_wait(struct lwp *lp) 822 { 823 struct thread *td = lp->lwp_thread; 824 u_int mpflags; 825 826 KKASSERT(lwkt_preempted_proc() != lp); 827 828 /* 829 * This bit of code uses the thread destruction interlock 830 * managed by lwkt_switch_return() to wait for the lwp's 831 * thread to completely disengage. 832 * 833 * It is possible for us to race another cpu core so we 834 * have to do this correctly. 835 */ 836 for (;;) { 837 mpflags = td->td_mpflags; 838 cpu_ccfence(); 839 if (mpflags & TDF_MP_EXITSIG) 840 break; 841 tsleep_interlock(td, 0); 842 if (atomic_cmpset_int(&td->td_mpflags, mpflags, 843 mpflags | TDF_MP_EXITWAIT)) { 844 tsleep(td, PINTERLOCKED, "lwpxt", 0); 845 } 846 } 847 848 /* 849 * We've already waited for the core exit but there can still 850 * be other refs from e.g. process scans and such. 851 */ 852 if (lp->lwp_lock > 0) { 853 tsleep(lp, 0, "lwpwait1", 1); 854 return(0); 855 } 856 if (td->td_refs) { 857 tsleep(td, 0, "lwpwait2", 1); 858 return(0); 859 } 860 861 /* 862 * Now that we have the thread destruction interlock these flags 863 * really should already be cleaned up, keep a check for safety. 864 * 865 * We can't rip its stack out from under it until TDF_EXITING is 866 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 867 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 868 * will be cleared temporarily if a thread gets preempted. 869 */ 870 while ((td->td_flags & (TDF_RUNNING | 871 TDF_RUNQ | 872 TDF_PREEMPT_LOCK | 873 TDF_EXITING)) != TDF_EXITING) { 874 tsleep(lp, 0, "lwpwait3", 1); 875 return (0); 876 } 877 878 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0, 879 ("lwp_wait: td %p (%s) still on run or sleep queue", 880 td, td->td_comm)); 881 return (1); 882 } 883 884 /* 885 * Release the resources associated with a lwp. 886 * The lwp must be completely dead. 887 */ 888 void 889 lwp_dispose(struct lwp *lp) 890 { 891 struct thread *td = lp->lwp_thread; 892 893 KKASSERT(lwkt_preempted_proc() != lp); 894 KKASSERT(lp->lwp_lock == 0); 895 KKASSERT(td->td_refs == 0); 896 KKASSERT((td->td_flags & (TDF_RUNNING | 897 TDF_RUNQ | 898 TDF_PREEMPT_LOCK | 899 TDF_EXITING)) == TDF_EXITING); 900 901 PRELE(lp->lwp_proc); 902 lp->lwp_proc = NULL; 903 if (td != NULL) { 904 td->td_proc = NULL; 905 td->td_lwp = NULL; 906 lp->lwp_thread = NULL; 907 lwkt_free_thread(td); 908 } 909 kfree(lp, M_LWP); 910 } 911 912 int 913 sys_wait4(struct sysmsg *sysmsg, const struct wait_args *uap) 914 { 915 struct __wrusage wrusage; 916 int error; 917 int status; 918 int options; 919 id_t id; 920 idtype_t idtype; 921 922 options = uap->options | WEXITED | WTRAPPED; 923 id = uap->pid; 924 925 if (id == WAIT_ANY) { 926 idtype = P_ALL; 927 } else if (id == WAIT_MYPGRP) { 928 idtype = P_PGID; 929 id = curproc->p_pgid; 930 } else if (id < 0) { 931 idtype = P_PGID; 932 id = -id; 933 } else { 934 idtype = P_PID; 935 } 936 937 error = kern_wait(idtype, id, &status, options, &wrusage, 938 NULL, &sysmsg->sysmsg_result); 939 940 if (error == 0 && uap->status) 941 error = copyout(&status, uap->status, sizeof(*uap->status)); 942 if (error == 0 && uap->rusage) { 943 ruadd(&wrusage.wru_self, &wrusage.wru_children); 944 error = copyout(&wrusage.wru_self, uap->rusage, sizeof(*uap->rusage)); 945 } 946 return (error); 947 } 948 949 int 950 sys_wait6(struct sysmsg *sysmsg, const struct wait6_args *uap) 951 { 952 struct __wrusage wrusage; 953 siginfo_t info; 954 siginfo_t *infop; 955 int error; 956 int status; 957 int options; 958 id_t id; 959 idtype_t idtype; 960 961 /* 962 * NOTE: wait6() requires WEXITED and WTRAPPED to be specified if 963 * desired. 964 */ 965 options = uap->options; 966 idtype = uap->idtype; 967 id = uap->id; 968 infop = uap->info ? &info : NULL; 969 970 switch(idtype) { 971 case P_PID: 972 case P_PGID: 973 if (id == WAIT_MYPGRP) { 974 idtype = P_PGID; 975 id = curproc->p_pgid; 976 } 977 break; 978 default: 979 /* let kern_wait deal with the remainder */ 980 break; 981 } 982 983 error = kern_wait(idtype, id, &status, options, 984 &wrusage, infop, &sysmsg->sysmsg_result); 985 986 if (error == 0 && uap->status) 987 error = copyout(&status, uap->status, sizeof(*uap->status)); 988 if (error == 0 && uap->wrusage) 989 error = copyout(&wrusage, uap->wrusage, sizeof(*uap->wrusage)); 990 if (error == 0 && uap->info) 991 error = copyout(&info, uap->info, sizeof(*uap->info)); 992 return (error); 993 } 994 995 /* 996 * kernel wait*() system call support 997 */ 998 int 999 kern_wait(idtype_t idtype, id_t id, int *status, int options, 1000 struct __wrusage *wrusage, siginfo_t *info, int *res) 1001 { 1002 struct thread *td = curthread; 1003 struct lwp *lp; 1004 struct proc *q = td->td_proc; 1005 struct proc *p, *t; 1006 struct ucred *cr; 1007 struct pargs *pa; 1008 struct sigacts *ps; 1009 int nfound, error; 1010 long waitgen; 1011 1012 /* 1013 * Must not have extraneous options. Must have at least one 1014 * matchable option. 1015 */ 1016 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE|WSTOPPED| 1017 WEXITED|WTRAPPED|WNOWAIT)) { 1018 return (EINVAL); 1019 } 1020 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1021 return (EINVAL); 1022 } 1023 1024 /* 1025 * Protect the q->p_children list 1026 */ 1027 lwkt_gettoken(&q->p_token); 1028 loop: 1029 /* 1030 * All sorts of things can change due to blocking so we have to loop 1031 * all the way back up here. 1032 * 1033 * The problem is that if a process group is stopped and the parent 1034 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 1035 * of the child and then stop itself when it tries to return from the 1036 * system call. When the process group is resumed the parent will 1037 * then get the STOP status even though the child has now resumed 1038 * (a followup wait*() will get the CONT status). 1039 * 1040 * Previously the CONT would overwrite the STOP because the tstop 1041 * was handled within tsleep(), and the parent would only see 1042 * the CONT when both are stopped and continued together. This little 1043 * two-line hack restores this effect. 1044 * 1045 * No locks are held so we can safely block the process here. 1046 */ 1047 if (STOPLWP(q, td->td_lwp)) 1048 tstop(); 1049 1050 nfound = 0; 1051 1052 /* 1053 * Loop on children. 1054 * 1055 * NOTE: We don't want to break q's p_token in the loop for the 1056 * case where no children are found or we risk breaking the 1057 * interlock between child and parent. 1058 */ 1059 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000); 1060 LIST_FOREACH(p, &q->p_children, p_sibling) { 1061 /* 1062 * Skip children that another thread is already uninterruptably 1063 * reaping. 1064 */ 1065 if (PWAITRES_PENDING(p)) 1066 continue; 1067 1068 /* 1069 * Filter, (p) will be held on fall-through. Try to optimize 1070 * this to avoid the atomic op until we are pretty sure we 1071 * want this process. 1072 */ 1073 switch(idtype) { 1074 case P_ALL: 1075 PHOLD(p); 1076 break; 1077 case P_PID: 1078 if (p->p_pid != (pid_t)id) 1079 continue; 1080 PHOLD(p); 1081 break; 1082 case P_PGID: 1083 if (p->p_pgid != (pid_t)id) 1084 continue; 1085 PHOLD(p); 1086 break; 1087 case P_SID: 1088 PHOLD(p); 1089 if (p->p_session && p->p_session->s_sid != (pid_t)id) { 1090 PRELE(p); 1091 continue; 1092 } 1093 break; 1094 case P_UID: 1095 PHOLD(p); 1096 if (p->p_ucred->cr_uid != (uid_t)id) { 1097 PRELE(p); 1098 continue; 1099 } 1100 break; 1101 case P_GID: 1102 PHOLD(p); 1103 if (p->p_ucred->cr_gid != (gid_t)id) { 1104 PRELE(p); 1105 continue; 1106 } 1107 break; 1108 case P_JAILID: 1109 PHOLD(p); 1110 if (p->p_ucred->cr_prison && 1111 p->p_ucred->cr_prison->pr_id != (int)id) { 1112 PRELE(p); 1113 continue; 1114 } 1115 break; 1116 default: 1117 /* unsupported filter */ 1118 continue; 1119 } 1120 /* (p) is held at this point */ 1121 1122 /* 1123 * This special case handles a kthread spawned by linux_clone 1124 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1125 * functions need to be able to distinguish between waiting 1126 * on a process and waiting on a thread. It is a thread if 1127 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1128 * signifies we want to wait for threads and not processes. 1129 */ 1130 if ((p->p_sigparent != SIGCHLD) ^ 1131 ((options & WLINUXCLONE) != 0)) { 1132 PRELE(p); 1133 continue; 1134 } 1135 1136 nfound++; 1137 if (p->p_stat == SZOMB && (options & WEXITED)) { 1138 /* 1139 * We may go into SZOMB with threads still present. 1140 * We must wait for them to exit before we can reap 1141 * the master thread, otherwise we may race reaping 1142 * non-master threads. 1143 * 1144 * Only this routine can remove a process from 1145 * the zombie list and destroy it. 1146 * 1147 * This function will fail after sleeping if another 1148 * thread owns the zombie lock. This function will 1149 * fail immediately or after sleeping if another 1150 * thread owns or obtains ownership of the reap via 1151 * WAITRES. 1152 */ 1153 if (PHOLDZOMB(p)) { 1154 PRELE(p); 1155 goto loop; 1156 } 1157 lwkt_gettoken(&p->p_token); 1158 if (p->p_pptr != q) { 1159 lwkt_reltoken(&p->p_token); 1160 PRELE(p); 1161 PRELEZOMB(p); 1162 goto loop; 1163 } 1164 1165 /* 1166 * We are the reaper, from this point on the reap 1167 * cannot be aborted. 1168 */ 1169 PWAITRES_SET(p); 1170 while (p->p_nthreads > 0) { 1171 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 1172 } 1173 1174 /* 1175 * Reap any LWPs left in p->p_lwps. This is usually 1176 * just the last LWP. This must be done before 1177 * we loop on p_lock since the lwps hold a ref on 1178 * it as a vmspace interlock. 1179 * 1180 * Once that is accomplished p_nthreads had better 1181 * be zero. 1182 */ 1183 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 1184 /* 1185 * Make sure no one is using this lwp, before 1186 * it is removed from the tree. If we didn't 1187 * wait it here, lwp tree iteration with 1188 * blocking operation would be broken. 1189 */ 1190 while (lp->lwp_lock > 0) 1191 tsleep(lp, 0, "zomblwp", 1); 1192 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 1193 reaplwp(lp); 1194 } 1195 KKASSERT(p->p_nthreads == 0); 1196 1197 /* 1198 * Don't do anything really bad until all references 1199 * to the process go away. This may include other 1200 * LWPs which are still in the process of being 1201 * reaped. We can't just pull the rug out from under 1202 * them because they may still be using the VM space. 1203 * 1204 * Certain kernel facilities such as /proc will also 1205 * put a hold on the process for short periods of 1206 * time. 1207 */ 1208 PRELE(p); /* from top of loop */ 1209 PSTALL(p, "reap3", 1); /* 1 ref (for PZOMBHOLD) */ 1210 1211 /* Take care of our return values. */ 1212 *res = p->p_pid; 1213 1214 *status = p->p_xstat; 1215 wrusage->wru_self = p->p_ru; 1216 wrusage->wru_children = p->p_cru; 1217 1218 if (info) { 1219 bzero(info, sizeof(*info)); 1220 info->si_errno = 0; 1221 info->si_signo = SIGCHLD; 1222 if (WIFEXITED(p->p_xstat)) { 1223 info->si_code = CLD_EXITED; 1224 info->si_status = 1225 WEXITSTATUS(p->p_xstat); 1226 } else { 1227 info->si_code = CLD_KILLED; 1228 info->si_status = WTERMSIG(p->p_xstat); 1229 } 1230 info->si_pid = p->p_pid; 1231 info->si_uid = p->p_ucred->cr_uid; 1232 } 1233 1234 /* 1235 * WNOWAIT shortcuts to done here, leaving the 1236 * child on the zombie list. 1237 */ 1238 if (options & WNOWAIT) { 1239 lwkt_reltoken(&p->p_token); 1240 PRELEZOMB(p); 1241 error = 0; 1242 goto done; 1243 } 1244 1245 /* 1246 * If we got the child via a ptrace 'attach', 1247 * we need to give it back to the old parent. 1248 */ 1249 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 1250 p->p_oppid = 0; 1251 proc_reparent(p, t); 1252 ksignal(t, SIGCHLD); 1253 wakeup((caddr_t)t); 1254 PRELE(t); 1255 lwkt_reltoken(&p->p_token); 1256 PRELEZOMB(p); 1257 error = 0; 1258 goto done; 1259 } 1260 1261 /* 1262 * Unlink the proc from its process group so that 1263 * the following operations won't lead to an 1264 * inconsistent state for processes running down 1265 * the zombie list. 1266 */ 1267 proc_remove_zombie(p); 1268 proc_userunmap(p); 1269 lwkt_reltoken(&p->p_token); 1270 leavepgrp(p); 1271 1272 p->p_xstat = 0; 1273 ruadd(&q->p_cru, &p->p_ru); 1274 ruadd(&q->p_cru, &p->p_cru); 1275 1276 /* 1277 * Decrement the count of procs running with this uid. 1278 */ 1279 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 1280 1281 /* 1282 * Free up credentials. p_spin is required to 1283 * avoid races against allproc scans. 1284 */ 1285 spin_lock(&p->p_spin); 1286 cr = p->p_ucred; 1287 p->p_ucred = NULL; 1288 spin_unlock(&p->p_spin); 1289 crfree(cr); 1290 1291 /* 1292 * Remove unused arguments 1293 */ 1294 pa = p->p_args; 1295 p->p_args = NULL; 1296 if (pa && refcount_release(&pa->ar_ref)) { 1297 kfree(pa, M_PARGS); 1298 pa = NULL; 1299 } 1300 1301 ps = p->p_sigacts; 1302 p->p_sigacts = NULL; 1303 if (ps && refcount_release(&ps->ps_refcnt)) { 1304 kfree(ps, M_SUBPROC); 1305 ps = NULL; 1306 } 1307 1308 /* 1309 * Our exitingcount was incremented when the process 1310 * became a zombie, now that the process has been 1311 * removed from (almost) all lists we should be able 1312 * to safely destroy its vmspace. Wait for any current 1313 * holders to go away (so the vmspace remains stable), 1314 * then scrap it. 1315 * 1316 * NOTE: Releasing the parent process (q) p_token 1317 * across the vmspace_exitfree() call is 1318 * important here to reduce stalls on 1319 * interactions with (q) (such as 1320 * fork/exec/wait or 'ps'). 1321 */ 1322 PSTALL(p, "reap4", 1); 1323 lwkt_reltoken(&q->p_token); 1324 vmspace_exitfree(p); 1325 lwkt_gettoken(&q->p_token); 1326 PSTALL(p, "reap5", 1); 1327 1328 /* 1329 * NOTE: We have to officially release ZOMB in order 1330 * to ensure that a racing thread in kern_wait() 1331 * which blocked on ZOMB is woken up. 1332 */ 1333 PRELEZOMB(p); 1334 kfree(p->p_uidpcpu, M_SUBPROC); 1335 kfree(p, M_PROC); 1336 atomic_add_int(&nprocs, -1); 1337 error = 0; 1338 goto done; 1339 } 1340 1341 /* 1342 * Process has not yet exited 1343 */ 1344 if ((p->p_stat == SSTOP || p->p_stat == SCORE) && 1345 (p->p_flags & P_WAITED) == 0 && 1346 (((p->p_flags & P_TRACED) && (options & WTRAPPED)) || 1347 (options & WSTOPPED))) { 1348 lwkt_gettoken(&p->p_token); 1349 if (p->p_pptr != q) { 1350 lwkt_reltoken(&p->p_token); 1351 PRELE(p); 1352 goto loop; 1353 } 1354 if ((p->p_stat != SSTOP && p->p_stat != SCORE) || 1355 (p->p_flags & P_WAITED) != 0 || 1356 ((p->p_flags & P_TRACED) == 0 && 1357 (options & WUNTRACED) == 0)) { 1358 lwkt_reltoken(&p->p_token); 1359 PRELE(p); 1360 goto loop; 1361 } 1362 1363 /* 1364 * Don't set P_WAITED if WNOWAIT specified, leaving 1365 * the process in a waitable state. 1366 */ 1367 if ((options & WNOWAIT) == 0) 1368 p->p_flags |= P_WAITED; 1369 1370 *res = p->p_pid; 1371 *status = W_STOPCODE(p->p_xstat); 1372 /* Zero rusage so we get something consistent. */ 1373 bzero(wrusage, sizeof(*wrusage)); 1374 error = 0; 1375 if (info) { 1376 bzero(info, sizeof(*info)); 1377 if (p->p_flags & P_TRACED) 1378 info->si_code = CLD_TRAPPED; 1379 else 1380 info->si_code = CLD_STOPPED; 1381 info->si_status = WSTOPSIG(p->p_xstat); 1382 } 1383 lwkt_reltoken(&p->p_token); 1384 PRELE(p); 1385 goto done; 1386 } 1387 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) { 1388 lwkt_gettoken(&p->p_token); 1389 if (p->p_pptr != q) { 1390 lwkt_reltoken(&p->p_token); 1391 PRELE(p); 1392 goto loop; 1393 } 1394 if ((p->p_flags & P_CONTINUED) == 0) { 1395 lwkt_reltoken(&p->p_token); 1396 PRELE(p); 1397 goto loop; 1398 } 1399 1400 *res = p->p_pid; 1401 1402 /* 1403 * Don't set P_WAITED if WNOWAIT specified, leaving 1404 * the process in a waitable state. 1405 */ 1406 if ((options & WNOWAIT) == 0) 1407 p->p_flags &= ~P_CONTINUED; 1408 1409 *status = SIGCONT; 1410 error = 0; 1411 if (info) { 1412 bzero(info, sizeof(*info)); 1413 info->si_code = CLD_CONTINUED; 1414 info->si_status = WSTOPSIG(p->p_xstat); 1415 } 1416 lwkt_reltoken(&p->p_token); 1417 PRELE(p); 1418 goto done; 1419 } 1420 PRELE(p); 1421 } 1422 if (nfound == 0) { 1423 error = ECHILD; 1424 goto done; 1425 } 1426 if (options & WNOHANG) { 1427 *res = 0; 1428 error = 0; 1429 goto done; 1430 } 1431 1432 /* 1433 * Wait for signal - interlocked using q->p_waitgen. 1434 */ 1435 error = 0; 1436 while ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) { 1437 tsleep_interlock(q, PCATCH); 1438 waitgen = atomic_fetchadd_long(&q->p_waitgen, 0x80000000); 1439 if ((waitgen & 0x7FFFFFFF) == (q->p_waitgen & 0x7FFFFFFF)) { 1440 error = tsleep(q, PCATCH | PINTERLOCKED, "wait", 0); 1441 break; 1442 } 1443 } 1444 if (error) { 1445 done: 1446 lwkt_reltoken(&q->p_token); 1447 return (error); 1448 } 1449 goto loop; 1450 } 1451 1452 /* 1453 * Change child's parent process to parent. 1454 * 1455 * p_children/p_sibling requires the parent's token, and 1456 * changing pptr requires the child's token, so we have to 1457 * get three tokens to do this operation. We also need to 1458 * hold pointers that might get ripped out from under us to 1459 * preserve structural integrity. 1460 * 1461 * It is possible to race another reparent or disconnect or other 1462 * similar operation. We must retry when this situation occurs. 1463 * Once we successfully reparent the process we no longer care 1464 * about any races. 1465 */ 1466 void 1467 proc_reparent(struct proc *child, struct proc *parent) 1468 { 1469 struct proc *opp; 1470 1471 PHOLD(parent); 1472 while ((opp = child->p_pptr) != parent) { 1473 PHOLD(opp); 1474 lwkt_gettoken(&opp->p_token); 1475 lwkt_gettoken(&child->p_token); 1476 lwkt_gettoken(&parent->p_token); 1477 if (child->p_pptr != opp) { 1478 lwkt_reltoken(&parent->p_token); 1479 lwkt_reltoken(&child->p_token); 1480 lwkt_reltoken(&opp->p_token); 1481 PRELE(opp); 1482 continue; 1483 } 1484 LIST_REMOVE(child, p_sibling); 1485 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1486 child->p_pptr = parent; 1487 child->p_ppid = parent->p_pid; 1488 lwkt_reltoken(&parent->p_token); 1489 lwkt_reltoken(&child->p_token); 1490 lwkt_reltoken(&opp->p_token); 1491 if (LIST_EMPTY(&opp->p_children)) 1492 wakeup(opp); 1493 PRELE(opp); 1494 break; 1495 } 1496 PRELE(parent); 1497 } 1498 1499 /* 1500 * The next two functions are to handle adding/deleting items on the 1501 * exit callout list 1502 * 1503 * at_exit(): 1504 * Take the arguments given and put them onto the exit callout list, 1505 * However first make sure that it's not already there. 1506 * returns 0 on success. 1507 */ 1508 1509 int 1510 at_exit(exitlist_fn function) 1511 { 1512 struct exitlist *ep; 1513 1514 #ifdef INVARIANTS 1515 /* Be noisy if the programmer has lost track of things */ 1516 if (rm_at_exit(function)) 1517 kprintf("WARNING: exit callout entry (%p) already present\n", 1518 function); 1519 #endif 1520 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 1521 if (ep == NULL) 1522 return (ENOMEM); 1523 ep->function = function; 1524 TAILQ_INSERT_TAIL(&exit_list, ep, next); 1525 return (0); 1526 } 1527 1528 /* 1529 * Scan the exit callout list for the given item and remove it. 1530 * Returns the number of items removed (0 or 1) 1531 */ 1532 int 1533 rm_at_exit(exitlist_fn function) 1534 { 1535 struct exitlist *ep; 1536 1537 TAILQ_FOREACH(ep, &exit_list, next) { 1538 if (ep->function == function) { 1539 TAILQ_REMOVE(&exit_list, ep, next); 1540 kfree(ep, M_ATEXIT); 1541 return(1); 1542 } 1543 } 1544 return (0); 1545 } 1546 1547 /* 1548 * LWP reaper related code. 1549 */ 1550 static void 1551 reaplwps(void *context, int dummy) 1552 { 1553 struct lwplist *lwplist = context; 1554 struct lwp *lp; 1555 int cpu = mycpuid; 1556 1557 lwkt_gettoken(&deadlwp_token[cpu]); 1558 while ((lp = LIST_FIRST(lwplist))) { 1559 LIST_REMOVE(lp, u.lwp_reap_entry); 1560 reaplwp(lp); 1561 } 1562 lwkt_reltoken(&deadlwp_token[cpu]); 1563 } 1564 1565 static void 1566 reaplwp(struct lwp *lp) 1567 { 1568 while (lwp_wait(lp) == 0) 1569 ; 1570 lwp_dispose(lp); 1571 } 1572 1573 static void 1574 deadlwp_init(void) 1575 { 1576 int cpu; 1577 1578 for (cpu = 0; cpu < ncpus; cpu++) { 1579 lwkt_token_init(&deadlwp_token[cpu], "deadlwpl"); 1580 LIST_INIT(&deadlwp_list[cpu]); 1581 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), 1582 M_DEVBUF, M_WAITOK); 1583 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1584 } 1585 } 1586 1587 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1588