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_compat.h" 39 #include "opt_ktrace.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/sysproto.h> 44 #include <sys/kernel.h> 45 #include <sys/malloc.h> 46 #include <sys/proc.h> 47 #include <sys/ktrace.h> 48 #include <sys/pioctl.h> 49 #include <sys/tty.h> 50 #include <sys/wait.h> 51 #include <sys/vnode.h> 52 #include <sys/resourcevar.h> 53 #include <sys/signalvar.h> 54 #include <sys/taskqueue.h> 55 #include <sys/ptrace.h> 56 #include <sys/acct.h> /* for acct_process() function prototype */ 57 #include <sys/filedesc.h> 58 #include <sys/shm.h> 59 #include <sys/sem.h> 60 #include <sys/jail.h> 61 #include <sys/kern_syscall.h> 62 #include <sys/unistd.h> 63 #include <sys/eventhandler.h> 64 #include <sys/dsched.h> 65 66 #include <vm/vm.h> 67 #include <vm/vm_param.h> 68 #include <sys/lock.h> 69 #include <vm/pmap.h> 70 #include <vm/vm_map.h> 71 #include <vm/vm_extern.h> 72 #include <sys/user.h> 73 74 #include <sys/refcount.h> 75 #include <sys/thread2.h> 76 #include <sys/sysref2.h> 77 #include <sys/mplock2.h> 78 79 static void reaplwps(void *context, int dummy); 80 static void reaplwp(struct lwp *lp); 81 static void killlwps(struct lwp *lp); 82 83 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 84 85 static struct lwkt_token deadlwp_token = LWKT_TOKEN_INITIALIZER(deadlwp_token); 86 87 /* 88 * callout list for things to do at exit time 89 */ 90 struct exitlist { 91 exitlist_fn function; 92 TAILQ_ENTRY(exitlist) next; 93 }; 94 95 TAILQ_HEAD(exit_list_head, exitlist); 96 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 97 98 /* 99 * LWP reaper data 100 */ 101 struct task *deadlwp_task[MAXCPU]; 102 struct lwplist deadlwp_list[MAXCPU]; 103 104 /* 105 * exit -- 106 * Death of process. 107 * 108 * SYS_EXIT_ARGS(int rval) 109 */ 110 int 111 sys_exit(struct exit_args *uap) 112 { 113 exit1(W_EXITCODE(uap->rval, 0)); 114 /* NOTREACHED */ 115 } 116 117 /* 118 * Extended exit -- 119 * Death of a lwp or process with optional bells and whistles. 120 * 121 * MPALMOSTSAFE 122 */ 123 int 124 sys_extexit(struct extexit_args *uap) 125 { 126 struct proc *p = curproc; 127 int action, who; 128 int error; 129 130 action = EXTEXIT_ACTION(uap->how); 131 who = EXTEXIT_WHO(uap->how); 132 133 /* Check parameters before we might perform some action */ 134 switch (who) { 135 case EXTEXIT_PROC: 136 case EXTEXIT_LWP: 137 break; 138 default: 139 return (EINVAL); 140 } 141 142 switch (action) { 143 case EXTEXIT_SIMPLE: 144 break; 145 case EXTEXIT_SETINT: 146 error = copyout(&uap->status, uap->addr, sizeof(uap->status)); 147 if (error) 148 return (error); 149 break; 150 default: 151 return (EINVAL); 152 } 153 154 lwkt_gettoken(&p->p_token); 155 156 switch (who) { 157 case EXTEXIT_LWP: 158 /* 159 * Be sure only to perform a simple lwp exit if there is at 160 * least one more lwp in the proc, which will call exit1() 161 * later, otherwise the proc will be an UNDEAD and not even a 162 * SZOMB! 163 */ 164 if (p->p_nthreads > 1) { 165 lwp_exit(0); /* called w/ p_token held */ 166 /* NOT REACHED */ 167 } 168 /* else last lwp in proc: do the real thing */ 169 /* FALLTHROUGH */ 170 default: /* to help gcc */ 171 case EXTEXIT_PROC: 172 lwkt_reltoken(&p->p_token); 173 exit1(W_EXITCODE(uap->status, 0)); 174 /* NOTREACHED */ 175 } 176 177 /* NOTREACHED */ 178 lwkt_reltoken(&p->p_token); /* safety */ 179 } 180 181 /* 182 * Kill all lwps associated with the current process except the 183 * current lwp. Return an error if we race another thread trying to 184 * do the same thing and lose the race. 185 * 186 * If forexec is non-zero the current thread and process flags are 187 * cleaned up so they can be reused. 188 * 189 * Caller must hold curproc->p_token 190 */ 191 int 192 killalllwps(int forexec) 193 { 194 struct lwp *lp = curthread->td_lwp; 195 struct proc *p = lp->lwp_proc; 196 197 /* 198 * Interlock against P_WEXIT. Only one of the process's thread 199 * is allowed to do the master exit. 200 */ 201 if (p->p_flags & P_WEXIT) 202 return (EALREADY); 203 p->p_flags |= P_WEXIT; 204 205 /* 206 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs 207 */ 208 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 209 if (p->p_nthreads > 1) 210 killlwps(lp); 211 212 /* 213 * If doing this for an exec, clean up the remaining thread 214 * (us) for continuing operation after all the other threads 215 * have been killed. 216 */ 217 if (forexec) { 218 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 219 p->p_flags &= ~P_WEXIT; 220 } 221 return(0); 222 } 223 224 /* 225 * Kill all LWPs except the current one. Do not try to signal 226 * LWPs which have exited on their own or have already been 227 * signaled. 228 */ 229 static void 230 killlwps(struct lwp *lp) 231 { 232 struct proc *p = lp->lwp_proc; 233 struct lwp *tlp; 234 235 /* 236 * Kill the remaining LWPs. We must send the signal before setting 237 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce 238 * races. tlp must be held across the call as it might block and 239 * allow the target lwp to rip itself out from under our loop. 240 */ 241 FOREACH_LWP_IN_PROC(tlp, p) { 242 LWPHOLD(tlp); 243 lwkt_gettoken(&tlp->lwp_token); 244 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) { 245 lwpsignal(p, tlp, SIGKILL); 246 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT); 247 } 248 lwkt_reltoken(&tlp->lwp_token); 249 LWPRELE(tlp); 250 } 251 252 /* 253 * Wait for everything to clear out. 254 */ 255 while (p->p_nthreads > 1) { 256 tsleep(&p->p_nthreads, 0, "killlwps", 0); 257 } 258 } 259 260 /* 261 * Exit: deallocate address space and other resources, change proc state 262 * to zombie, and unlink proc from allproc and parent's lists. Save exit 263 * status and rusage for wait(). Check for child processes and orphan them. 264 */ 265 void 266 exit1(int rv) 267 { 268 struct thread *td = curthread; 269 struct proc *p = td->td_proc; 270 struct lwp *lp = td->td_lwp; 271 struct proc *q; 272 struct vmspace *vm; 273 struct vnode *vtmp; 274 struct exitlist *ep; 275 int error; 276 277 lwkt_gettoken(&p->p_token); 278 279 if (p->p_pid == 1) { 280 kprintf("init died (signal %d, exit %d)\n", 281 WTERMSIG(rv), WEXITSTATUS(rv)); 282 panic("Going nowhere without my init!"); 283 } 284 varsymset_clean(&p->p_varsymset); 285 lockuninit(&p->p_varsymset.vx_lock); 286 287 /* 288 * Kill all lwps associated with the current process, return an 289 * error if we race another thread trying to do the same thing 290 * and lose the race. 291 */ 292 error = killalllwps(0); 293 if (error) { 294 lwp_exit(0); 295 /* NOT REACHED */ 296 } 297 298 /* are we a task leader? */ 299 if (p == p->p_leader) { 300 struct kill_args killArgs; 301 killArgs.signum = SIGKILL; 302 q = p->p_peers; 303 while(q) { 304 killArgs.pid = q->p_pid; 305 /* 306 * The interface for kill is better 307 * than the internal signal 308 */ 309 sys_kill(&killArgs); 310 q = q->p_peers; 311 } 312 while (p->p_peers) 313 tsleep((caddr_t)p, 0, "exit1", 0); 314 } 315 316 #ifdef PGINPROF 317 vmsizmon(); 318 #endif 319 STOPEVENT(p, S_EXIT, rv); 320 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */ 321 322 /* 323 * Check if any loadable modules need anything done at process exit. 324 * e.g. SYSV IPC stuff 325 * XXX what if one of these generates an error? 326 */ 327 p->p_xstat = rv; 328 EVENTHANDLER_INVOKE(process_exit, p); 329 330 /* 331 * XXX: imho, the eventhandler stuff is much cleaner than this. 332 * Maybe we should move everything to use eventhandler. 333 */ 334 TAILQ_FOREACH(ep, &exit_list, next) 335 (*ep->function)(td); 336 337 if (p->p_flags & P_PROFIL) 338 stopprofclock(p); 339 340 SIGEMPTYSET(p->p_siglist); 341 SIGEMPTYSET(lp->lwp_siglist); 342 if (timevalisset(&p->p_realtimer.it_value)) 343 callout_stop_sync(&p->p_ithandle); 344 345 /* 346 * Reset any sigio structures pointing to us as a result of 347 * F_SETOWN with our pid. 348 */ 349 funsetownlst(&p->p_sigiolst); 350 351 /* 352 * Close open files and release open-file table. 353 * This may block! 354 */ 355 fdfree(p, NULL); 356 357 if(p->p_leader->p_peers) { 358 q = p->p_leader; 359 while(q->p_peers != p) 360 q = q->p_peers; 361 q->p_peers = p->p_peers; 362 wakeup((caddr_t)p->p_leader); 363 } 364 365 /* 366 * XXX Shutdown SYSV semaphores 367 */ 368 semexit(p); 369 370 KKASSERT(p->p_numposixlocks == 0); 371 372 /* The next two chunks should probably be moved to vmspace_exit. */ 373 vm = p->p_vmspace; 374 375 /* 376 * Clean up data related to virtual kernel operation. Clean up 377 * any vkernel context related to the current lwp now so we can 378 * destroy p_vkernel. 379 */ 380 if (p->p_vkernel) { 381 vkernel_lwp_exit(lp); 382 vkernel_exit(p); 383 } 384 385 /* 386 * Release user portion of address space. 387 * This releases references to vnodes, 388 * which could cause I/O if the file has been unlinked. 389 * Need to do this early enough that we can still sleep. 390 * Can't free the entire vmspace as the kernel stack 391 * may be mapped within that space also. 392 * 393 * Processes sharing the same vmspace may exit in one order, and 394 * get cleaned up by vmspace_exit() in a different order. The 395 * last exiting process to reach this point releases as much of 396 * the environment as it can, and the last process cleaned up 397 * by vmspace_exit() (which decrements exitingcnt) cleans up the 398 * remainder. 399 */ 400 vmspace_exitbump(vm); 401 sysref_put(&vm->vm_sysref); 402 403 if (SESS_LEADER(p)) { 404 struct session *sp = p->p_session; 405 406 if (sp->s_ttyvp) { 407 /* 408 * We are the controlling process. Signal the 409 * foreground process group, drain the controlling 410 * terminal, and revoke access to the controlling 411 * terminal. 412 * 413 * NOTE: while waiting for the process group to exit 414 * it is possible that one of the processes in the 415 * group will revoke the tty, so the ttyclosesession() 416 * function will re-check sp->s_ttyvp. 417 */ 418 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 419 if (sp->s_ttyp->t_pgrp) 420 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 421 ttywait(sp->s_ttyp); 422 ttyclosesession(sp, 1); /* also revoke */ 423 } 424 /* 425 * Release the tty. If someone has it open via 426 * /dev/tty then close it (since they no longer can 427 * once we've NULL'd it out). 428 */ 429 ttyclosesession(sp, 0); 430 431 /* 432 * s_ttyp is not zero'd; we use this to indicate 433 * that the session once had a controlling terminal. 434 * (for logging and informational purposes) 435 */ 436 } 437 sp->s_leader = NULL; 438 } 439 fixjobc(p, p->p_pgrp, 0); 440 (void)acct_process(p); 441 #ifdef KTRACE 442 /* 443 * release trace file 444 */ 445 if (p->p_tracenode) 446 ktrdestroy(&p->p_tracenode); 447 p->p_traceflag = 0; 448 #endif 449 /* 450 * Release reference to text vnode 451 */ 452 if ((vtmp = p->p_textvp) != NULL) { 453 p->p_textvp = NULL; 454 vrele(vtmp); 455 } 456 457 /* Release namecache handle to text file */ 458 if (p->p_textnch.ncp) 459 cache_drop(&p->p_textnch); 460 461 /* 462 * We have to handle PPWAIT here or proc_move_allproc_zombie() 463 * will block on the PHOLD() the parent is doing. 464 */ 465 if (p->p_flags & P_PPWAIT) { 466 p->p_flags &= ~P_PPWAIT; 467 wakeup(p->p_pptr); 468 } 469 470 /* 471 * Move the process to the zombie list. This will block 472 * until the process p_lock count reaches 0. The process will 473 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 474 * which is called from cpu_proc_exit(). 475 */ 476 proc_move_allproc_zombie(p); 477 478 /* 479 * Reparent all of this process's children to the init process. 480 * We must hold initproc->p_token in order to mess with 481 * initproc->p_children. We already hold p->p_token (to remove 482 * the children from our list). 483 */ 484 q = LIST_FIRST(&p->p_children); 485 if (q) { 486 lwkt_gettoken(&initproc->p_token); 487 while ((q = LIST_FIRST(&p->p_children)) != NULL) { 488 PHOLD(q); 489 lwkt_gettoken(&q->p_token); 490 if (q != LIST_FIRST(&p->p_children)) { 491 lwkt_reltoken(&q->p_token); 492 PRELE(q); 493 continue; 494 } 495 LIST_REMOVE(q, p_sibling); 496 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling); 497 q->p_pptr = initproc; 498 q->p_sigparent = SIGCHLD; 499 500 /* 501 * Traced processes are killed 502 * since their existence means someone is screwing up. 503 */ 504 if (q->p_flags & P_TRACED) { 505 q->p_flags &= ~P_TRACED; 506 ksignal(q, SIGKILL); 507 } 508 lwkt_reltoken(&q->p_token); 509 PRELE(q); 510 } 511 lwkt_reltoken(&initproc->p_token); 512 wakeup(initproc); 513 } 514 515 /* 516 * Save exit status and final rusage info, adding in child rusage 517 * info and self times. 518 */ 519 calcru_proc(p, &p->p_ru); 520 ruadd(&p->p_ru, &p->p_cru); 521 522 /* 523 * notify interested parties of our demise. 524 */ 525 KNOTE(&p->p_klist, NOTE_EXIT); 526 527 /* 528 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 529 * flag set, or if the handler is set to SIG_IGN, notify process 1 530 * instead (and hope it will handle this situation). 531 */ 532 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 533 proc_reparent(p, initproc); 534 } 535 536 /* lwkt_gettoken(&proc_token); */ 537 q = p->p_pptr; 538 PHOLD(q); 539 if (p->p_sigparent && q != initproc) { 540 ksignal(q, p->p_sigparent); 541 } else { 542 ksignal(q, SIGCHLD); 543 } 544 545 p->p_flags &= ~P_TRACED; 546 wakeup(p->p_pptr); 547 548 PRELE(q); 549 /* lwkt_reltoken(&proc_token); */ 550 /* NOTE: p->p_pptr can get ripped out */ 551 /* 552 * cpu_exit is responsible for clearing curproc, since 553 * it is heavily integrated with the thread/switching sequence. 554 * 555 * Other substructures are freed from wait(). 556 */ 557 plimit_free(p); 558 559 /* 560 * Release the current user process designation on the process so 561 * the userland scheduler can work in someone else. 562 */ 563 p->p_usched->release_curproc(lp); 564 565 /* 566 * Finally, call machine-dependent code to release as many of the 567 * lwp's resources as we can and halt execution of this thread. 568 */ 569 lwp_exit(1); 570 } 571 572 /* 573 * Eventually called by every exiting LWP 574 * 575 * p->p_token must be held. mplock may be held and will be released. 576 */ 577 void 578 lwp_exit(int masterexit) 579 { 580 struct thread *td = curthread; 581 struct lwp *lp = td->td_lwp; 582 struct proc *p = lp->lwp_proc; 583 int dowake = 0; 584 585 /* 586 * lwp_exit() may be called without setting LWP_MP_WEXIT, so 587 * make sure it is set here. 588 */ 589 ASSERT_LWKT_TOKEN_HELD(&p->p_token); 590 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 591 592 /* 593 * Clean up any virtualization 594 */ 595 if (lp->lwp_vkernel) 596 vkernel_lwp_exit(lp); 597 598 /* 599 * Clean up select/poll support 600 */ 601 kqueue_terminate(&lp->lwp_kqueue); 602 603 /* 604 * Clean up any syscall-cached ucred 605 */ 606 if (td->td_ucred) { 607 crfree(td->td_ucred); 608 td->td_ucred = NULL; 609 } 610 611 /* 612 * Nobody actually wakes us when the lock 613 * count reaches zero, so just wait one tick. 614 */ 615 while (lp->lwp_lock > 0) 616 tsleep(lp, 0, "lwpexit", 1); 617 618 /* Hand down resource usage to our proc */ 619 ruadd(&p->p_ru, &lp->lwp_ru); 620 621 /* 622 * If we don't hold the process until the LWP is reaped wait*() 623 * may try to dispose of its vmspace before all the LWPs have 624 * actually terminated. 625 */ 626 PHOLD(p); 627 628 /* 629 * Do any remaining work that might block on us. We should be 630 * coded such that further blocking is ok after decrementing 631 * p_nthreads but don't take the chance. 632 */ 633 dsched_exit_thread(td); 634 biosched_done(curthread); 635 636 /* 637 * We have to use the reaper for all the LWPs except the one doing 638 * the master exit. The LWP doing the master exit can just be 639 * left on p_lwps and the process reaper will deal with it 640 * synchronously, which is much faster. 641 * 642 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0. 643 * 644 * The process is left held until the reaper calls lwp_dispose() on 645 * the lp (after calling lwp_wait()). 646 */ 647 if (masterexit == 0) { 648 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 649 --p->p_nthreads; 650 if (p->p_nthreads <= 1) 651 dowake = 1; 652 lwkt_gettoken(&deadlwp_token); 653 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry); 654 taskqueue_enqueue(taskqueue_thread[mycpuid], 655 deadlwp_task[mycpuid]); 656 lwkt_reltoken(&deadlwp_token); 657 } else { 658 --p->p_nthreads; 659 if (p->p_nthreads <= 1) 660 dowake = 1; 661 } 662 663 /* 664 * Release p_token. Issue the wakeup() on p_nthreads if necessary, 665 * as late as possible to give us a chance to actually deschedule and 666 * switch away before another cpu core hits reaplwp(). 667 */ 668 lwkt_reltoken(&p->p_token); 669 if (dowake) 670 wakeup(&p->p_nthreads); 671 672 /* 673 * Tell the userland scheduler that we are going away 674 */ 675 p->p_usched->heuristic_exiting(lp, p); 676 677 cpu_lwp_exit(); 678 } 679 680 /* 681 * Wait until a lwp is completely dead. The final interlock in this drama 682 * is when TDF_EXITING is set in cpu_thread_exit() just before the final 683 * switchout. 684 * 685 * At the point TDF_EXITING is set a complete exit is accomplished when 686 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two 687 * post-switch interlock flags that can be used to wait for the TDF_ 688 * flags to clear. 689 * 690 * Returns non-zero on success, and zero if the caller needs to retry 691 * the lwp_wait(). 692 */ 693 static int 694 lwp_wait(struct lwp *lp) 695 { 696 struct thread *td = lp->lwp_thread; 697 u_int mpflags; 698 699 KKASSERT(lwkt_preempted_proc() != lp); 700 701 /* 702 * This bit of code uses the thread destruction interlock 703 * managed by lwkt_switch_return() to wait for the lwp's 704 * thread to completely disengage. 705 * 706 * It is possible for us to race another cpu core so we 707 * have to do this correctly. 708 */ 709 for (;;) { 710 mpflags = td->td_mpflags; 711 cpu_ccfence(); 712 if (mpflags & TDF_MP_EXITSIG) 713 break; 714 tsleep_interlock(td, 0); 715 if (atomic_cmpset_int(&td->td_mpflags, mpflags, 716 mpflags | TDF_MP_EXITWAIT)) { 717 tsleep(td, PINTERLOCKED, "lwpxt", 0); 718 } 719 } 720 721 /* 722 * We've already waited for the core exit but there can still 723 * be other refs from e.g. process scans and such. 724 */ 725 if (lp->lwp_lock > 0) { 726 tsleep(lp, 0, "lwpwait1", 1); 727 return(0); 728 } 729 if (td->td_refs) { 730 tsleep(td, 0, "lwpwait2", 1); 731 return(0); 732 } 733 734 /* 735 * Now that we have the thread destruction interlock these flags 736 * really should already be cleaned up, keep a check for safety. 737 * 738 * We can't rip its stack out from under it until TDF_EXITING is 739 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 740 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 741 * will be cleared temporarily if a thread gets preempted. 742 */ 743 while ((td->td_flags & (TDF_RUNNING | 744 TDF_PREEMPT_LOCK | 745 TDF_EXITING)) != TDF_EXITING) { 746 tsleep(lp, 0, "lwpwait3", 1); 747 return (0); 748 } 749 750 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0, 751 ("lwp_wait: td %p (%s) still on run or sleep queue", 752 td, td->td_comm)); 753 return (1); 754 } 755 756 /* 757 * Release the resources associated with a lwp. 758 * The lwp must be completely dead. 759 */ 760 void 761 lwp_dispose(struct lwp *lp) 762 { 763 struct thread *td = lp->lwp_thread; 764 765 KKASSERT(lwkt_preempted_proc() != lp); 766 KKASSERT(td->td_refs == 0); 767 KKASSERT((td->td_flags & (TDF_RUNNING | 768 TDF_PREEMPT_LOCK | 769 TDF_EXITING)) == TDF_EXITING); 770 771 PRELE(lp->lwp_proc); 772 lp->lwp_proc = NULL; 773 if (td != NULL) { 774 td->td_proc = NULL; 775 td->td_lwp = NULL; 776 lp->lwp_thread = NULL; 777 lwkt_free_thread(td); 778 } 779 kfree(lp, M_LWP); 780 } 781 782 /* 783 * MPSAFE 784 */ 785 int 786 sys_wait4(struct wait_args *uap) 787 { 788 struct rusage rusage; 789 int error, status; 790 791 error = kern_wait(uap->pid, (uap->status ? &status : NULL), 792 uap->options, (uap->rusage ? &rusage : NULL), 793 &uap->sysmsg_result); 794 795 if (error == 0 && uap->status) 796 error = copyout(&status, uap->status, sizeof(*uap->status)); 797 if (error == 0 && uap->rusage) 798 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage)); 799 return (error); 800 } 801 802 /* 803 * wait1() 804 * 805 * wait_args(int pid, int *status, int options, struct rusage *rusage) 806 * 807 * MPALMOSTSAFE 808 */ 809 int 810 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res) 811 { 812 struct thread *td = curthread; 813 struct lwp *lp; 814 struct proc *q = td->td_proc; 815 struct proc *p, *t; 816 struct pargs *pa; 817 struct sigacts *ps; 818 int nfound, error; 819 820 if (pid == 0) 821 pid = -q->p_pgid; 822 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE)) 823 return (EINVAL); 824 825 lwkt_gettoken(&q->p_token); 826 loop: 827 /* 828 * All sorts of things can change due to blocking so we have to loop 829 * all the way back up here. 830 * 831 * The problem is that if a process group is stopped and the parent 832 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 833 * of the child and then stop itself when it tries to return from the 834 * system call. When the process group is resumed the parent will 835 * then get the STOP status even though the child has now resumed 836 * (a followup wait*() will get the CONT status). 837 * 838 * Previously the CONT would overwrite the STOP because the tstop 839 * was handled within tsleep(), and the parent would only see 840 * the CONT when both are stopped and continued together. This little 841 * two-line hack restores this effect. 842 */ 843 while (q->p_stat == SSTOP) 844 tstop(); 845 846 nfound = 0; 847 848 /* 849 * Loop on children. 850 * 851 * NOTE: We don't want to break q's p_token in the loop for the 852 * case where no children are found or we risk breaking the 853 * interlock between child and parent. 854 */ 855 LIST_FOREACH(p, &q->p_children, p_sibling) { 856 if (pid != WAIT_ANY && 857 p->p_pid != pid && p->p_pgid != -pid) { 858 continue; 859 } 860 861 /* 862 * This special case handles a kthread spawned by linux_clone 863 * (see linux_misc.c). The linux_wait4 and linux_waitpid 864 * functions need to be able to distinguish between waiting 865 * on a process and waiting on a thread. It is a thread if 866 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 867 * signifies we want to wait for threads and not processes. 868 */ 869 if ((p->p_sigparent != SIGCHLD) ^ 870 ((options & WLINUXCLONE) != 0)) { 871 continue; 872 } 873 874 nfound++; 875 if (p->p_stat == SZOMB) { 876 /* 877 * We may go into SZOMB with threads still present. 878 * We must wait for them to exit before we can reap 879 * the master thread, otherwise we may race reaping 880 * non-master threads. 881 * 882 * Only this routine can remove a process from 883 * the zombie list and destroy it, use PACQUIREZOMB() 884 * to serialize us and loop if it blocks (interlocked 885 * by the parent's q->p_token). 886 * 887 * WARNING! (p) can be invalid when PHOLDZOMB(p) 888 * returns non-zero. Be sure not to 889 * mess with it. 890 */ 891 if (PHOLDZOMB(p)) 892 goto loop; 893 lwkt_gettoken(&p->p_token); 894 if (p->p_pptr != q) { 895 lwkt_reltoken(&p->p_token); 896 PRELEZOMB(p); 897 goto loop; 898 } 899 while (p->p_nthreads > 0) { 900 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 901 } 902 903 /* 904 * Reap any LWPs left in p->p_lwps. This is usually 905 * just the last LWP. This must be done before 906 * we loop on p_lock since the lwps hold a ref on 907 * it as a vmspace interlock. 908 * 909 * Once that is accomplished p_nthreads had better 910 * be zero. 911 */ 912 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 913 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 914 reaplwp(lp); 915 } 916 KKASSERT(p->p_nthreads == 0); 917 918 /* 919 * Don't do anything really bad until all references 920 * to the process go away. This may include other 921 * LWPs which are still in the process of being 922 * reaped. We can't just pull the rug out from under 923 * them because they may still be using the VM space. 924 * 925 * Certain kernel facilities such as /proc will also 926 * put a hold on the process for short periods of 927 * time. 928 */ 929 PRELE(p); 930 PSTALL(p, "reap3", 0); 931 932 /* Take care of our return values. */ 933 *res = p->p_pid; 934 935 if (status) 936 *status = p->p_xstat; 937 if (rusage) 938 *rusage = p->p_ru; 939 /* 940 * If we got the child via a ptrace 'attach', 941 * we need to give it back to the old parent. 942 */ 943 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 944 PHOLD(p); 945 p->p_oppid = 0; 946 proc_reparent(p, t); 947 ksignal(t, SIGCHLD); 948 wakeup((caddr_t)t); 949 error = 0; 950 PRELE(t); 951 lwkt_reltoken(&p->p_token); 952 PRELEZOMB(p); 953 goto done; 954 } 955 956 /* 957 * Unlink the proc from its process group so that 958 * the following operations won't lead to an 959 * inconsistent state for processes running down 960 * the zombie list. 961 */ 962 proc_remove_zombie(p); 963 lwkt_reltoken(&p->p_token); 964 leavepgrp(p); 965 966 p->p_xstat = 0; 967 ruadd(&q->p_cru, &p->p_ru); 968 969 /* 970 * Decrement the count of procs running with this uid. 971 */ 972 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 973 974 /* 975 * Free up credentials. 976 */ 977 crfree(p->p_ucred); 978 p->p_ucred = NULL; 979 980 /* 981 * Remove unused arguments 982 */ 983 pa = p->p_args; 984 p->p_args = NULL; 985 if (pa && refcount_release(&pa->ar_ref)) { 986 kfree(pa, M_PARGS); 987 pa = NULL; 988 } 989 990 ps = p->p_sigacts; 991 p->p_sigacts = NULL; 992 if (ps && refcount_release(&ps->ps_refcnt)) { 993 kfree(ps, M_SUBPROC); 994 ps = NULL; 995 } 996 997 /* 998 * Our exitingcount was incremented when the process 999 * became a zombie, now that the process has been 1000 * removed from (almost) all lists we should be able 1001 * to safely destroy its vmspace. Wait for any current 1002 * holders to go away (so the vmspace remains stable), 1003 * then scrap it. 1004 */ 1005 PSTALL(p, "reap4", 0); 1006 vmspace_exitfree(p); 1007 PSTALL(p, "reap5", 0); 1008 1009 /* 1010 * NOTE: We have to officially release ZOMB in order 1011 * to ensure that a racing thread in kern_wait() 1012 * which blocked on ZOMB is woken up. 1013 */ 1014 PHOLD(p); 1015 PRELEZOMB(p); 1016 kfree(p, M_PROC); 1017 atomic_add_int(&nprocs, -1); 1018 error = 0; 1019 goto done; 1020 } 1021 if (p->p_stat == SSTOP && (p->p_flags & P_WAITED) == 0 && 1022 ((p->p_flags & P_TRACED) || (options & WUNTRACED))) { 1023 PHOLD(p); 1024 lwkt_gettoken(&p->p_token); 1025 if (p->p_pptr != q) { 1026 lwkt_reltoken(&p->p_token); 1027 PRELE(p); 1028 goto loop; 1029 } 1030 if (p->p_stat != SSTOP || 1031 (p->p_flags & P_WAITED) != 0 || 1032 ((p->p_flags & P_TRACED) == 0 && 1033 (options & WUNTRACED) == 0)) { 1034 lwkt_reltoken(&p->p_token); 1035 PRELE(p); 1036 goto loop; 1037 } 1038 1039 p->p_flags |= P_WAITED; 1040 1041 *res = p->p_pid; 1042 if (status) 1043 *status = W_STOPCODE(p->p_xstat); 1044 /* Zero rusage so we get something consistent. */ 1045 if (rusage) 1046 bzero(rusage, sizeof(*rusage)); 1047 error = 0; 1048 lwkt_reltoken(&p->p_token); 1049 PRELE(p); 1050 goto done; 1051 } 1052 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) { 1053 PHOLD(p); 1054 lwkt_gettoken(&p->p_token); 1055 if (p->p_pptr != q) { 1056 lwkt_reltoken(&p->p_token); 1057 PRELE(p); 1058 goto loop; 1059 } 1060 if ((p->p_flags & P_CONTINUED) == 0) { 1061 lwkt_reltoken(&p->p_token); 1062 PRELE(p); 1063 goto loop; 1064 } 1065 1066 *res = p->p_pid; 1067 p->p_flags &= ~P_CONTINUED; 1068 1069 if (status) 1070 *status = SIGCONT; 1071 error = 0; 1072 lwkt_reltoken(&p->p_token); 1073 PRELE(p); 1074 goto done; 1075 } 1076 } 1077 if (nfound == 0) { 1078 error = ECHILD; 1079 goto done; 1080 } 1081 if (options & WNOHANG) { 1082 *res = 0; 1083 error = 0; 1084 goto done; 1085 } 1086 1087 /* 1088 * Wait for signal - interlocked using q->p_token. 1089 */ 1090 error = tsleep(q, PCATCH, "wait", 0); 1091 if (error) { 1092 done: 1093 lwkt_reltoken(&q->p_token); 1094 return (error); 1095 } 1096 goto loop; 1097 } 1098 1099 /* 1100 * Change child's parent process to parent. 1101 * 1102 * p_children/p_sibling requires the parent's token, and 1103 * changing pptr requires the child's token, so we have to 1104 * get three tokens to do this operation. We also need to 1105 * hold pointers that might get ripped out from under us to 1106 * preserve structural integrity. 1107 * 1108 * It is possible to race another reparent or disconnect or other 1109 * similar operation. We must retry when this situation occurs. 1110 * Once we successfully reparent the process we no longer care 1111 * about any races. 1112 */ 1113 void 1114 proc_reparent(struct proc *child, struct proc *parent) 1115 { 1116 struct proc *opp; 1117 1118 PHOLD(parent); 1119 while ((opp = child->p_pptr) != parent) { 1120 PHOLD(opp); 1121 lwkt_gettoken(&opp->p_token); 1122 lwkt_gettoken(&child->p_token); 1123 lwkt_gettoken(&parent->p_token); 1124 if (child->p_pptr != opp) { 1125 lwkt_reltoken(&parent->p_token); 1126 lwkt_reltoken(&child->p_token); 1127 lwkt_reltoken(&opp->p_token); 1128 PRELE(opp); 1129 continue; 1130 } 1131 LIST_REMOVE(child, p_sibling); 1132 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1133 child->p_pptr = parent; 1134 lwkt_reltoken(&parent->p_token); 1135 lwkt_reltoken(&child->p_token); 1136 lwkt_reltoken(&opp->p_token); 1137 if (LIST_EMPTY(&opp->p_children)) 1138 wakeup(opp); 1139 PRELE(opp); 1140 break; 1141 } 1142 PRELE(parent); 1143 } 1144 1145 /* 1146 * The next two functions are to handle adding/deleting items on the 1147 * exit callout list 1148 * 1149 * at_exit(): 1150 * Take the arguments given and put them onto the exit callout list, 1151 * However first make sure that it's not already there. 1152 * returns 0 on success. 1153 */ 1154 1155 int 1156 at_exit(exitlist_fn function) 1157 { 1158 struct exitlist *ep; 1159 1160 #ifdef INVARIANTS 1161 /* Be noisy if the programmer has lost track of things */ 1162 if (rm_at_exit(function)) 1163 kprintf("WARNING: exit callout entry (%p) already present\n", 1164 function); 1165 #endif 1166 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 1167 if (ep == NULL) 1168 return (ENOMEM); 1169 ep->function = function; 1170 TAILQ_INSERT_TAIL(&exit_list, ep, next); 1171 return (0); 1172 } 1173 1174 /* 1175 * Scan the exit callout list for the given item and remove it. 1176 * Returns the number of items removed (0 or 1) 1177 */ 1178 int 1179 rm_at_exit(exitlist_fn function) 1180 { 1181 struct exitlist *ep; 1182 1183 TAILQ_FOREACH(ep, &exit_list, next) { 1184 if (ep->function == function) { 1185 TAILQ_REMOVE(&exit_list, ep, next); 1186 kfree(ep, M_ATEXIT); 1187 return(1); 1188 } 1189 } 1190 return (0); 1191 } 1192 1193 /* 1194 * LWP reaper related code. 1195 */ 1196 static void 1197 reaplwps(void *context, int dummy) 1198 { 1199 struct lwplist *lwplist = context; 1200 struct lwp *lp; 1201 1202 lwkt_gettoken(&deadlwp_token); 1203 while ((lp = LIST_FIRST(lwplist))) { 1204 LIST_REMOVE(lp, u.lwp_reap_entry); 1205 reaplwp(lp); 1206 } 1207 lwkt_reltoken(&deadlwp_token); 1208 } 1209 1210 static void 1211 reaplwp(struct lwp *lp) 1212 { 1213 while (lwp_wait(lp) == 0) 1214 ; 1215 lwp_dispose(lp); 1216 } 1217 1218 static void 1219 deadlwp_init(void) 1220 { 1221 int cpu; 1222 1223 for (cpu = 0; cpu < ncpus; cpu++) { 1224 LIST_INIT(&deadlwp_list[cpu]); 1225 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), 1226 M_DEVBUF, M_WAITOK); 1227 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1228 } 1229 } 1230 1231 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1232