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. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 39 * $FreeBSD: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $ 40 */ 41 42 #include "opt_compat.h" 43 #include "opt_ktrace.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/sysproto.h> 48 #include <sys/kernel.h> 49 #include <sys/malloc.h> 50 #include <sys/proc.h> 51 #include <sys/ktrace.h> 52 #include <sys/pioctl.h> 53 #include <sys/tty.h> 54 #include <sys/wait.h> 55 #include <sys/vnode.h> 56 #include <sys/resourcevar.h> 57 #include <sys/signalvar.h> 58 #include <sys/taskqueue.h> 59 #include <sys/ptrace.h> 60 #include <sys/acct.h> /* for acct_process() function prototype */ 61 #include <sys/filedesc.h> 62 #include <sys/shm.h> 63 #include <sys/sem.h> 64 #include <sys/jail.h> 65 #include <sys/kern_syscall.h> 66 #include <sys/upcall.h> 67 #include <sys/caps.h> 68 #include <sys/unistd.h> 69 #include <sys/eventhandler.h> 70 #include <sys/dsched.h> 71 72 #include <vm/vm.h> 73 #include <vm/vm_param.h> 74 #include <sys/lock.h> 75 #include <vm/pmap.h> 76 #include <vm/vm_map.h> 77 #include <vm/vm_extern.h> 78 #include <sys/user.h> 79 80 #include <sys/refcount.h> 81 #include <sys/thread2.h> 82 #include <sys/sysref2.h> 83 #include <sys/mplock2.h> 84 85 static void reaplwps(void *context, int dummy); 86 static void reaplwp(struct lwp *lp); 87 static void killlwps(struct lwp *lp); 88 89 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 90 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status"); 91 92 static struct lwkt_token deadlwp_token = LWKT_TOKEN_INITIALIZER(deadlwp_token); 93 94 /* 95 * callout list for things to do at exit time 96 */ 97 struct exitlist { 98 exitlist_fn function; 99 TAILQ_ENTRY(exitlist) next; 100 }; 101 102 TAILQ_HEAD(exit_list_head, exitlist); 103 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 104 105 /* 106 * LWP reaper data 107 */ 108 struct task *deadlwp_task[MAXCPU]; 109 struct lwplist deadlwp_list[MAXCPU]; 110 111 /* 112 * exit -- 113 * Death of process. 114 * 115 * SYS_EXIT_ARGS(int rval) 116 */ 117 int 118 sys_exit(struct exit_args *uap) 119 { 120 exit1(W_EXITCODE(uap->rval, 0)); 121 /* NOTREACHED */ 122 } 123 124 /* 125 * Extended exit -- 126 * Death of a lwp or process with optional bells and whistles. 127 * 128 * MPALMOSTSAFE 129 */ 130 int 131 sys_extexit(struct extexit_args *uap) 132 { 133 struct proc *p = curproc; 134 int action, who; 135 int error; 136 137 action = EXTEXIT_ACTION(uap->how); 138 who = EXTEXIT_WHO(uap->how); 139 140 /* Check parameters before we might perform some action */ 141 switch (who) { 142 case EXTEXIT_PROC: 143 case EXTEXIT_LWP: 144 break; 145 default: 146 return (EINVAL); 147 } 148 149 switch (action) { 150 case EXTEXIT_SIMPLE: 151 break; 152 case EXTEXIT_SETINT: 153 error = copyout(&uap->status, uap->addr, sizeof(uap->status)); 154 if (error) 155 return (error); 156 break; 157 default: 158 return (EINVAL); 159 } 160 161 lwkt_gettoken(&p->p_token); 162 163 switch (who) { 164 case EXTEXIT_LWP: 165 /* 166 * Be sure only to perform a simple lwp exit if there is at 167 * least one more lwp in the proc, which will call exit1() 168 * later, otherwise the proc will be an UNDEAD and not even a 169 * SZOMB! 170 */ 171 if (p->p_nthreads > 1) { 172 lwp_exit(0); /* called w/ p_token held */ 173 /* NOT REACHED */ 174 } 175 /* else last lwp in proc: do the real thing */ 176 /* FALLTHROUGH */ 177 default: /* to help gcc */ 178 case EXTEXIT_PROC: 179 lwkt_reltoken(&p->p_token); 180 exit1(W_EXITCODE(uap->status, 0)); 181 /* NOTREACHED */ 182 } 183 184 /* NOTREACHED */ 185 lwkt_reltoken(&p->p_token); /* safety */ 186 } 187 188 /* 189 * Kill all lwps associated with the current process except the 190 * current lwp. Return an error if we race another thread trying to 191 * do the same thing and lose the race. 192 * 193 * If forexec is non-zero the current thread and process flags are 194 * cleaned up so they can be reused. 195 * 196 * Caller must hold curproc->p_token 197 */ 198 int 199 killalllwps(int forexec) 200 { 201 struct lwp *lp = curthread->td_lwp; 202 struct proc *p = lp->lwp_proc; 203 204 /* 205 * Interlock against P_WEXIT. Only one of the process's thread 206 * is allowed to do the master exit. 207 */ 208 if (p->p_flags & P_WEXIT) 209 return (EALREADY); 210 p->p_flags |= P_WEXIT; 211 212 /* 213 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs 214 */ 215 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 216 if (p->p_nthreads > 1) 217 killlwps(lp); 218 219 /* 220 * If doing this for an exec, clean up the remaining thread 221 * (us) for continuing operation after all the other threads 222 * have been killed. 223 */ 224 if (forexec) { 225 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 226 p->p_flags &= ~P_WEXIT; 227 } 228 return(0); 229 } 230 231 /* 232 * Kill all LWPs except the current one. Do not try to signal 233 * LWPs which have exited on their own or have already been 234 * signaled. 235 */ 236 static void 237 killlwps(struct lwp *lp) 238 { 239 struct proc *p = lp->lwp_proc; 240 struct lwp *tlp; 241 242 /* 243 * Kill the remaining LWPs. We must send the signal before setting 244 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce 245 * races. tlp must be held across the call as it might block and 246 * allow the target lwp to rip itself out from under our loop. 247 */ 248 FOREACH_LWP_IN_PROC(tlp, p) { 249 LWPHOLD(tlp); 250 lwkt_gettoken(&tlp->lwp_token); 251 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) { 252 lwpsignal(p, tlp, SIGKILL); 253 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT); 254 } 255 lwkt_reltoken(&tlp->lwp_token); 256 LWPRELE(tlp); 257 } 258 259 /* 260 * Wait for everything to clear out. 261 */ 262 while (p->p_nthreads > 1) { 263 tsleep(&p->p_nthreads, 0, "killlwps", 0); 264 } 265 } 266 267 /* 268 * Exit: deallocate address space and other resources, change proc state 269 * to zombie, and unlink proc from allproc and parent's lists. Save exit 270 * status and rusage for wait(). Check for child processes and orphan them. 271 */ 272 void 273 exit1(int rv) 274 { 275 struct thread *td = curthread; 276 struct proc *p = td->td_proc; 277 struct lwp *lp = td->td_lwp; 278 struct proc *q, *nq; 279 struct vmspace *vm; 280 struct vnode *vtmp; 281 struct exitlist *ep; 282 int error; 283 284 lwkt_gettoken(&p->p_token); 285 286 if (p->p_pid == 1) { 287 kprintf("init died (signal %d, exit %d)\n", 288 WTERMSIG(rv), WEXITSTATUS(rv)); 289 panic("Going nowhere without my init!"); 290 } 291 varsymset_clean(&p->p_varsymset); 292 lockuninit(&p->p_varsymset.vx_lock); 293 294 /* 295 * Kill all lwps associated with the current process, return an 296 * error if we race another thread trying to do the same thing 297 * and lose the race. 298 */ 299 error = killalllwps(0); 300 if (error) { 301 lwp_exit(0); 302 /* NOT REACHED */ 303 } 304 305 caps_exit(lp->lwp_thread); 306 307 /* are we a task leader? */ 308 if (p == p->p_leader) { 309 struct kill_args killArgs; 310 killArgs.signum = SIGKILL; 311 q = p->p_peers; 312 while(q) { 313 killArgs.pid = q->p_pid; 314 /* 315 * The interface for kill is better 316 * than the internal signal 317 */ 318 sys_kill(&killArgs); 319 nq = q; 320 q = q->p_peers; 321 } 322 while (p->p_peers) 323 tsleep((caddr_t)p, 0, "exit1", 0); 324 } 325 326 #ifdef PGINPROF 327 vmsizmon(); 328 #endif 329 STOPEVENT(p, S_EXIT, rv); 330 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */ 331 332 /* 333 * Check if any loadable modules need anything done at process exit. 334 * e.g. SYSV IPC stuff 335 * XXX what if one of these generates an error? 336 */ 337 p->p_xstat = rv; 338 EVENTHANDLER_INVOKE(process_exit, p); 339 340 /* 341 * XXX: imho, the eventhandler stuff is much cleaner than this. 342 * Maybe we should move everything to use eventhandler. 343 */ 344 TAILQ_FOREACH(ep, &exit_list, next) 345 (*ep->function)(td); 346 347 if (p->p_flags & P_PROFIL) 348 stopprofclock(p); 349 350 SIGEMPTYSET(p->p_siglist); 351 SIGEMPTYSET(lp->lwp_siglist); 352 if (timevalisset(&p->p_realtimer.it_value)) 353 callout_stop_sync(&p->p_ithandle); 354 355 /* 356 * Reset any sigio structures pointing to us as a result of 357 * F_SETOWN with our pid. 358 */ 359 funsetownlst(&p->p_sigiolst); 360 361 /* 362 * Close open files and release open-file table. 363 * This may block! 364 */ 365 fdfree(p, NULL); 366 367 if(p->p_leader->p_peers) { 368 q = p->p_leader; 369 while(q->p_peers != p) 370 q = q->p_peers; 371 q->p_peers = p->p_peers; 372 wakeup((caddr_t)p->p_leader); 373 } 374 375 /* 376 * XXX Shutdown SYSV semaphores 377 */ 378 semexit(p); 379 380 KKASSERT(p->p_numposixlocks == 0); 381 382 /* The next two chunks should probably be moved to vmspace_exit. */ 383 vm = p->p_vmspace; 384 385 /* 386 * Release upcalls associated with this process 387 */ 388 if (vm->vm_upcalls) 389 upc_release(vm, lp); 390 391 /* 392 * Clean up data related to virtual kernel operation. Clean up 393 * any vkernel context related to the current lwp now so we can 394 * destroy p_vkernel. 395 */ 396 if (p->p_vkernel) { 397 vkernel_lwp_exit(lp); 398 vkernel_exit(p); 399 } 400 401 /* 402 * Release user portion of address space. 403 * This releases references to vnodes, 404 * which could cause I/O if the file has been unlinked. 405 * Need to do this early enough that we can still sleep. 406 * Can't free the entire vmspace as the kernel stack 407 * may be mapped within that space also. 408 * 409 * Processes sharing the same vmspace may exit in one order, and 410 * get cleaned up by vmspace_exit() in a different order. The 411 * last exiting process to reach this point releases as much of 412 * the environment as it can, and the last process cleaned up 413 * by vmspace_exit() (which decrements exitingcnt) cleans up the 414 * remainder. 415 */ 416 vmspace_exitbump(vm); 417 sysref_put(&vm->vm_sysref); 418 419 if (SESS_LEADER(p)) { 420 struct session *sp = p->p_session; 421 422 if (sp->s_ttyvp) { 423 /* 424 * We are the controlling process. Signal the 425 * foreground process group, drain the controlling 426 * terminal, and revoke access to the controlling 427 * terminal. 428 * 429 * NOTE: while waiting for the process group to exit 430 * it is possible that one of the processes in the 431 * group will revoke the tty, so the ttyclosesession() 432 * function will re-check sp->s_ttyvp. 433 */ 434 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 435 if (sp->s_ttyp->t_pgrp) 436 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 437 ttywait(sp->s_ttyp); 438 ttyclosesession(sp, 1); /* also revoke */ 439 } 440 /* 441 * Release the tty. If someone has it open via 442 * /dev/tty then close it (since they no longer can 443 * once we've NULL'd it out). 444 */ 445 ttyclosesession(sp, 0); 446 447 /* 448 * s_ttyp is not zero'd; we use this to indicate 449 * that the session once had a controlling terminal. 450 * (for logging and informational purposes) 451 */ 452 } 453 sp->s_leader = NULL; 454 } 455 fixjobc(p, p->p_pgrp, 0); 456 (void)acct_process(p); 457 #ifdef KTRACE 458 /* 459 * release trace file 460 */ 461 if (p->p_tracenode) 462 ktrdestroy(&p->p_tracenode); 463 p->p_traceflag = 0; 464 #endif 465 /* 466 * Release reference to text vnode 467 */ 468 if ((vtmp = p->p_textvp) != NULL) { 469 p->p_textvp = NULL; 470 vrele(vtmp); 471 } 472 473 /* Release namecache handle to text file */ 474 if (p->p_textnch.ncp) 475 cache_drop(&p->p_textnch); 476 477 /* 478 * We have to handle PPWAIT here or proc_move_allproc_zombie() 479 * will block on the PHOLD() the parent is doing. 480 */ 481 if (p->p_flags & P_PPWAIT) { 482 p->p_flags &= ~P_PPWAIT; 483 wakeup(p->p_pptr); 484 } 485 486 /* 487 * Move the process to the zombie list. This will block 488 * until the process p_lock count reaches 0. The process will 489 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 490 * which is called from cpu_proc_exit(). 491 */ 492 proc_move_allproc_zombie(p); 493 494 /* 495 * Reparent all of this process's children to the init process. 496 * We must hold initproc->p_token in order to mess with 497 * initproc->p_children. We already hold p->p_token (to remove 498 * the children from our list). 499 */ 500 q = LIST_FIRST(&p->p_children); 501 if (q) { 502 lwkt_gettoken(&initproc->p_token); 503 while ((q = LIST_FIRST(&p->p_children)) != NULL) { 504 PHOLD(q); 505 lwkt_gettoken(&q->p_token); 506 if (q != LIST_FIRST(&p->p_children)) { 507 lwkt_reltoken(&q->p_token); 508 PRELE(q); 509 continue; 510 } 511 LIST_REMOVE(q, p_sibling); 512 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling); 513 q->p_pptr = initproc; 514 q->p_sigparent = SIGCHLD; 515 516 /* 517 * Traced processes are killed 518 * since their existence means someone is screwing up. 519 */ 520 if (q->p_flags & P_TRACED) { 521 q->p_flags &= ~P_TRACED; 522 ksignal(q, SIGKILL); 523 } 524 lwkt_reltoken(&q->p_token); 525 PRELE(q); 526 } 527 lwkt_reltoken(&initproc->p_token); 528 wakeup(initproc); 529 } 530 531 /* 532 * Save exit status and final rusage info, adding in child rusage 533 * info and self times. 534 */ 535 calcru_proc(p, &p->p_ru); 536 ruadd(&p->p_ru, &p->p_cru); 537 538 /* 539 * notify interested parties of our demise. 540 */ 541 KNOTE(&p->p_klist, NOTE_EXIT); 542 543 /* 544 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 545 * flag set, or if the handler is set to SIG_IGN, notify process 1 546 * instead (and hope it will handle this situation). 547 */ 548 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 549 struct proc *pp = p->p_pptr; 550 551 PHOLD(pp); 552 proc_reparent(p, initproc); 553 554 /* 555 * If this was the last child of our parent, notify 556 * parent, so in case he was wait(2)ing, he will 557 * continue. This function interlocks with pptr->p_token. 558 */ 559 if (LIST_EMPTY(&pp->p_children)) 560 wakeup((caddr_t)pp); 561 PRELE(pp); 562 } 563 564 /* lwkt_gettoken(&proc_token); */ 565 q = p->p_pptr; 566 PHOLD(q); 567 if (p->p_sigparent && q != initproc) { 568 ksignal(q, p->p_sigparent); 569 } else { 570 ksignal(q, SIGCHLD); 571 } 572 573 p->p_flags &= ~P_TRACED; 574 wakeup(p->p_pptr); 575 576 PRELE(q); 577 /* lwkt_reltoken(&proc_token); */ 578 /* NOTE: p->p_pptr can get ripped out */ 579 /* 580 * cpu_exit is responsible for clearing curproc, since 581 * it is heavily integrated with the thread/switching sequence. 582 * 583 * Other substructures are freed from wait(). 584 */ 585 plimit_free(p); 586 587 /* 588 * Release the current user process designation on the process so 589 * the userland scheduler can work in someone else. 590 */ 591 p->p_usched->release_curproc(lp); 592 593 /* 594 * Finally, call machine-dependent code to release as many of the 595 * lwp's resources as we can and halt execution of this thread. 596 */ 597 lwp_exit(1); 598 } 599 600 /* 601 * Eventually called by every exiting LWP 602 * 603 * p->p_token must be held. mplock may be held and will be released. 604 */ 605 void 606 lwp_exit(int masterexit) 607 { 608 struct thread *td = curthread; 609 struct lwp *lp = td->td_lwp; 610 struct proc *p = lp->lwp_proc; 611 int dowake = 0; 612 613 /* 614 * lwp_exit() may be called without setting LWP_MP_WEXIT, so 615 * make sure it is set here. 616 */ 617 ASSERT_LWKT_TOKEN_HELD(&p->p_token); 618 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 619 620 /* 621 * Clean up any virtualization 622 */ 623 if (lp->lwp_vkernel) 624 vkernel_lwp_exit(lp); 625 626 /* 627 * Clean up select/poll support 628 */ 629 kqueue_terminate(&lp->lwp_kqueue); 630 631 /* 632 * Clean up any syscall-cached ucred 633 */ 634 if (td->td_ucred) { 635 crfree(td->td_ucred); 636 td->td_ucred = NULL; 637 } 638 639 /* 640 * Nobody actually wakes us when the lock 641 * count reaches zero, so just wait one tick. 642 */ 643 while (lp->lwp_lock > 0) 644 tsleep(lp, 0, "lwpexit", 1); 645 646 /* Hand down resource usage to our proc */ 647 ruadd(&p->p_ru, &lp->lwp_ru); 648 649 /* 650 * If we don't hold the process until the LWP is reaped wait*() 651 * may try to dispose of its vmspace before all the LWPs have 652 * actually terminated. 653 */ 654 PHOLD(p); 655 656 /* 657 * Do any remaining work that might block on us. We should be 658 * coded such that further blocking is ok after decrementing 659 * p_nthreads but don't take the chance. 660 */ 661 dsched_exit_thread(td); 662 biosched_done(curthread); 663 664 /* 665 * We have to use the reaper for all the LWPs except the one doing 666 * the master exit. The LWP doing the master exit can just be 667 * left on p_lwps and the process reaper will deal with it 668 * synchronously, which is much faster. 669 * 670 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0. 671 * 672 * The process is left held until the reaper calls lwp_dispose() on 673 * the lp (after calling lwp_wait()). 674 */ 675 if (masterexit == 0) { 676 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 677 --p->p_nthreads; 678 if (p->p_nthreads <= 1) 679 dowake = 1; 680 lwkt_gettoken(&deadlwp_token); 681 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry); 682 taskqueue_enqueue(taskqueue_thread[mycpuid], 683 deadlwp_task[mycpuid]); 684 lwkt_reltoken(&deadlwp_token); 685 } else { 686 --p->p_nthreads; 687 if (p->p_nthreads <= 1) 688 dowake = 1; 689 } 690 691 /* 692 * Release p_token. Issue the wakeup() on p_nthreads if necessary, 693 * as late as possible to give us a chance to actually deschedule and 694 * switch away before another cpu core hits reaplwp(). 695 */ 696 lwkt_reltoken(&p->p_token); 697 if (dowake) 698 wakeup(&p->p_nthreads); 699 700 /* 701 * Tell the userland scheduler that we are going away 702 */ 703 p->p_usched->heuristic_exiting(lp, p); 704 705 cpu_lwp_exit(); 706 } 707 708 /* 709 * Wait until a lwp is completely dead. The final interlock in this drama 710 * is when TDF_EXITING is set in cpu_thread_exit() just before the final 711 * switchout. 712 * 713 * At the point TDF_EXITING is set a complete exit is accomplished when 714 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two 715 * post-switch interlock flags that can be used to wait for the TDF_ 716 * flags to clear. 717 * 718 * Returns non-zero on success, and zero if the caller needs to retry 719 * the lwp_wait(). 720 */ 721 static int 722 lwp_wait(struct lwp *lp) 723 { 724 struct thread *td = lp->lwp_thread;; 725 u_int mpflags; 726 727 KKASSERT(lwkt_preempted_proc() != lp); 728 729 /* 730 * This bit of code uses the thread destruction interlock 731 * managed by lwkt_switch_return() to wait for the lwp's 732 * thread to completely disengage. 733 * 734 * It is possible for us to race another cpu core so we 735 * have to do this correctly. 736 */ 737 for (;;) { 738 mpflags = td->td_mpflags; 739 cpu_ccfence(); 740 if (mpflags & TDF_MP_EXITSIG) 741 break; 742 tsleep_interlock(td, 0); 743 if (atomic_cmpset_int(&td->td_mpflags, mpflags, 744 mpflags | TDF_MP_EXITWAIT)) { 745 tsleep(td, PINTERLOCKED, "lwpxt", 0); 746 } 747 } 748 749 /* 750 * We've already waited for the core exit but there can still 751 * be other refs from e.g. process scans and such. 752 */ 753 if (lp->lwp_lock > 0) { 754 tsleep(lp, 0, "lwpwait1", 1); 755 return(0); 756 } 757 if (td->td_refs) { 758 tsleep(td, 0, "lwpwait2", 1); 759 return(0); 760 } 761 762 /* 763 * Now that we have the thread destruction interlock these flags 764 * really should already be cleaned up, keep a check for safety. 765 * 766 * We can't rip its stack out from under it until TDF_EXITING is 767 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 768 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 769 * will be cleared temporarily if a thread gets preempted. 770 */ 771 while ((td->td_flags & (TDF_RUNNING | 772 TDF_PREEMPT_LOCK | 773 TDF_EXITING)) != TDF_EXITING) { 774 tsleep(lp, 0, "lwpwait3", 1); 775 return (0); 776 } 777 778 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0, 779 ("lwp_wait: td %p (%s) still on run or sleep queue", 780 td, td->td_comm)); 781 return (1); 782 } 783 784 /* 785 * Release the resources associated with a lwp. 786 * The lwp must be completely dead. 787 */ 788 void 789 lwp_dispose(struct lwp *lp) 790 { 791 struct thread *td = lp->lwp_thread;; 792 793 KKASSERT(lwkt_preempted_proc() != lp); 794 KKASSERT(td->td_refs == 0); 795 KKASSERT((td->td_flags & (TDF_RUNNING | 796 TDF_PREEMPT_LOCK | 797 TDF_EXITING)) == TDF_EXITING); 798 799 PRELE(lp->lwp_proc); 800 lp->lwp_proc = NULL; 801 if (td != NULL) { 802 td->td_proc = NULL; 803 td->td_lwp = NULL; 804 lp->lwp_thread = NULL; 805 lwkt_free_thread(td); 806 } 807 kfree(lp, M_LWP); 808 } 809 810 /* 811 * MPSAFE 812 */ 813 int 814 sys_wait4(struct wait_args *uap) 815 { 816 struct rusage rusage; 817 int error, status; 818 819 error = kern_wait(uap->pid, (uap->status ? &status : NULL), 820 uap->options, (uap->rusage ? &rusage : NULL), 821 &uap->sysmsg_result); 822 823 if (error == 0 && uap->status) 824 error = copyout(&status, uap->status, sizeof(*uap->status)); 825 if (error == 0 && uap->rusage) 826 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage)); 827 return (error); 828 } 829 830 /* 831 * wait1() 832 * 833 * wait_args(int pid, int *status, int options, struct rusage *rusage) 834 * 835 * MPALMOSTSAFE 836 */ 837 int 838 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res) 839 { 840 struct thread *td = curthread; 841 struct lwp *lp; 842 struct proc *q = td->td_proc; 843 struct proc *p, *t; 844 struct pargs *pa; 845 struct sigacts *ps; 846 int nfound, error; 847 848 if (pid == 0) 849 pid = -q->p_pgid; 850 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE)) 851 return (EINVAL); 852 853 lwkt_gettoken(&q->p_token); 854 loop: 855 /* 856 * All sorts of things can change due to blocking so we have to loop 857 * all the way back up here. 858 * 859 * The problem is that if a process group is stopped and the parent 860 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 861 * of the child and then stop itself when it tries to return from the 862 * system call. When the process group is resumed the parent will 863 * then get the STOP status even though the child has now resumed 864 * (a followup wait*() will get the CONT status). 865 * 866 * Previously the CONT would overwrite the STOP because the tstop 867 * was handled within tsleep(), and the parent would only see 868 * the CONT when both are stopped and continued together. This little 869 * two-line hack restores this effect. 870 */ 871 while (q->p_stat == SSTOP) 872 tstop(); 873 874 nfound = 0; 875 876 /* 877 * Loop on children. 878 * 879 * NOTE: We don't want to break q's p_token in the loop for the 880 * case where no children are found or we risk breaking the 881 * interlock between child and parent. 882 */ 883 LIST_FOREACH(p, &q->p_children, p_sibling) { 884 if (pid != WAIT_ANY && 885 p->p_pid != pid && p->p_pgid != -pid) { 886 continue; 887 } 888 889 /* 890 * This special case handles a kthread spawned by linux_clone 891 * (see linux_misc.c). The linux_wait4 and linux_waitpid 892 * functions need to be able to distinguish between waiting 893 * on a process and waiting on a thread. It is a thread if 894 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 895 * signifies we want to wait for threads and not processes. 896 */ 897 if ((p->p_sigparent != SIGCHLD) ^ 898 ((options & WLINUXCLONE) != 0)) { 899 continue; 900 } 901 902 nfound++; 903 if (p->p_stat == SZOMB) { 904 /* 905 * We may go into SZOMB with threads still present. 906 * We must wait for them to exit before we can reap 907 * the master thread, otherwise we may race reaping 908 * non-master threads. 909 * 910 * Only this routine can remove a process from 911 * the zombie list and destroy it, use PACQUIREZOMB() 912 * to serialize us and loop if it blocks (interlocked 913 * by the parent's q->p_token). 914 * 915 * WARNING! (p) can be invalid when PHOLDZOMB(p) 916 * returns non-zero. Be sure not to 917 * mess with it. 918 */ 919 if (PHOLDZOMB(p)) 920 goto loop; 921 lwkt_gettoken(&p->p_token); 922 if (p->p_pptr != q) { 923 lwkt_reltoken(&p->p_token); 924 PRELEZOMB(p); 925 goto loop; 926 } 927 while (p->p_nthreads > 0) { 928 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 929 } 930 931 /* 932 * Reap any LWPs left in p->p_lwps. This is usually 933 * just the last LWP. This must be done before 934 * we loop on p_lock since the lwps hold a ref on 935 * it as a vmspace interlock. 936 * 937 * Once that is accomplished p_nthreads had better 938 * be zero. 939 */ 940 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 941 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 942 reaplwp(lp); 943 } 944 KKASSERT(p->p_nthreads == 0); 945 946 /* 947 * Don't do anything really bad until all references 948 * to the process go away. This may include other 949 * LWPs which are still in the process of being 950 * reaped. We can't just pull the rug out from under 951 * them because they may still be using the VM space. 952 * 953 * Certain kernel facilities such as /proc will also 954 * put a hold on the process for short periods of 955 * time. 956 */ 957 PRELE(p); 958 PSTALL(p, "reap3", 0); 959 960 /* Take care of our return values. */ 961 *res = p->p_pid; 962 963 if (status) 964 *status = p->p_xstat; 965 if (rusage) 966 *rusage = p->p_ru; 967 /* 968 * If we got the child via a ptrace 'attach', 969 * we need to give it back to the old parent. 970 */ 971 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 972 PHOLD(p); 973 p->p_oppid = 0; 974 proc_reparent(p, t); 975 ksignal(t, SIGCHLD); 976 wakeup((caddr_t)t); 977 error = 0; 978 PRELE(t); 979 lwkt_reltoken(&p->p_token); 980 PRELEZOMB(p); 981 goto done; 982 } 983 984 /* 985 * Unlink the proc from its process group so that 986 * the following operations won't lead to an 987 * inconsistent state for processes running down 988 * the zombie list. 989 */ 990 proc_remove_zombie(p); 991 lwkt_reltoken(&p->p_token); 992 leavepgrp(p); 993 994 p->p_xstat = 0; 995 ruadd(&q->p_cru, &p->p_ru); 996 997 /* 998 * Decrement the count of procs running with this uid. 999 */ 1000 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 1001 1002 /* 1003 * Free up credentials. 1004 */ 1005 crfree(p->p_ucred); 1006 p->p_ucred = NULL; 1007 1008 /* 1009 * Remove unused arguments 1010 */ 1011 pa = p->p_args; 1012 p->p_args = NULL; 1013 if (pa && refcount_release(&pa->ar_ref)) { 1014 kfree(pa, M_PARGS); 1015 pa = NULL; 1016 } 1017 1018 ps = p->p_sigacts; 1019 p->p_sigacts = NULL; 1020 if (ps && refcount_release(&ps->ps_refcnt)) { 1021 kfree(ps, M_SUBPROC); 1022 ps = NULL; 1023 } 1024 1025 /* 1026 * Our exitingcount was incremented when the process 1027 * became a zombie, now that the process has been 1028 * removed from (almost) all lists we should be able 1029 * to safely destroy its vmspace. Wait for any current 1030 * holders to go away (so the vmspace remains stable), 1031 * then scrap it. 1032 */ 1033 PSTALL(p, "reap4", 0); 1034 vmspace_exitfree(p); 1035 PSTALL(p, "reap5", 0); 1036 1037 /* 1038 * NOTE: We have to officially release ZOMB in order 1039 * to ensure that a racing thread in kern_wait() 1040 * which blocked on ZOMB is woken up. 1041 */ 1042 PHOLD(p); 1043 PRELEZOMB(p); 1044 kfree(p, M_PROC); 1045 atomic_add_int(&nprocs, -1); 1046 error = 0; 1047 goto done; 1048 } 1049 if (p->p_stat == SSTOP && (p->p_flags & P_WAITED) == 0 && 1050 ((p->p_flags & P_TRACED) || (options & WUNTRACED))) { 1051 PHOLD(p); 1052 lwkt_gettoken(&p->p_token); 1053 if (p->p_pptr != q) { 1054 lwkt_reltoken(&p->p_token); 1055 PRELE(p); 1056 goto loop; 1057 } 1058 if (p->p_stat != SSTOP || 1059 (p->p_flags & P_WAITED) != 0 || 1060 ((p->p_flags & P_TRACED) == 0 && 1061 (options & WUNTRACED) == 0)) { 1062 lwkt_reltoken(&p->p_token); 1063 PRELE(p); 1064 goto loop; 1065 } 1066 1067 p->p_flags |= P_WAITED; 1068 1069 *res = p->p_pid; 1070 if (status) 1071 *status = W_STOPCODE(p->p_xstat); 1072 /* Zero rusage so we get something consistent. */ 1073 if (rusage) 1074 bzero(rusage, sizeof(*rusage)); 1075 error = 0; 1076 lwkt_reltoken(&p->p_token); 1077 PRELE(p); 1078 goto done; 1079 } 1080 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) { 1081 PHOLD(p); 1082 lwkt_gettoken(&p->p_token); 1083 if (p->p_pptr != q) { 1084 lwkt_reltoken(&p->p_token); 1085 PRELE(p); 1086 goto loop; 1087 } 1088 if ((p->p_flags & P_CONTINUED) == 0) { 1089 lwkt_reltoken(&p->p_token); 1090 PRELE(p); 1091 goto loop; 1092 } 1093 1094 *res = p->p_pid; 1095 p->p_flags &= ~P_CONTINUED; 1096 1097 if (status) 1098 *status = SIGCONT; 1099 error = 0; 1100 lwkt_reltoken(&p->p_token); 1101 PRELE(p); 1102 goto done; 1103 } 1104 } 1105 if (nfound == 0) { 1106 error = ECHILD; 1107 goto done; 1108 } 1109 if (options & WNOHANG) { 1110 *res = 0; 1111 error = 0; 1112 goto done; 1113 } 1114 1115 /* 1116 * Wait for signal - interlocked using q->p_token. 1117 */ 1118 error = tsleep(q, PCATCH, "wait", 0); 1119 if (error) { 1120 done: 1121 lwkt_reltoken(&q->p_token); 1122 return (error); 1123 } 1124 goto loop; 1125 } 1126 1127 /* 1128 * Make process 'parent' the new parent of process 'child'. 1129 * 1130 * p_children/p_sibling requires the parent's token, and 1131 * changing pptr requires the child's token, so we have to 1132 * get three tokens to do this operation. 1133 */ 1134 void 1135 proc_reparent(struct proc *child, struct proc *parent) 1136 { 1137 struct proc *opp = child->p_pptr; 1138 1139 if (opp == parent) 1140 return; 1141 PHOLD(opp); 1142 PHOLD(parent); 1143 lwkt_gettoken(&opp->p_token); 1144 lwkt_gettoken(&child->p_token); 1145 lwkt_gettoken(&parent->p_token); 1146 KKASSERT(child->p_pptr == opp); 1147 LIST_REMOVE(child, p_sibling); 1148 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1149 child->p_pptr = parent; 1150 lwkt_reltoken(&parent->p_token); 1151 lwkt_reltoken(&child->p_token); 1152 lwkt_reltoken(&opp->p_token); 1153 PRELE(parent); 1154 PRELE(opp); 1155 } 1156 1157 /* 1158 * The next two functions are to handle adding/deleting items on the 1159 * exit callout list 1160 * 1161 * at_exit(): 1162 * Take the arguments given and put them onto the exit callout list, 1163 * However first make sure that it's not already there. 1164 * returns 0 on success. 1165 */ 1166 1167 int 1168 at_exit(exitlist_fn function) 1169 { 1170 struct exitlist *ep; 1171 1172 #ifdef INVARIANTS 1173 /* Be noisy if the programmer has lost track of things */ 1174 if (rm_at_exit(function)) 1175 kprintf("WARNING: exit callout entry (%p) already present\n", 1176 function); 1177 #endif 1178 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 1179 if (ep == NULL) 1180 return (ENOMEM); 1181 ep->function = function; 1182 TAILQ_INSERT_TAIL(&exit_list, ep, next); 1183 return (0); 1184 } 1185 1186 /* 1187 * Scan the exit callout list for the given item and remove it. 1188 * Returns the number of items removed (0 or 1) 1189 */ 1190 int 1191 rm_at_exit(exitlist_fn function) 1192 { 1193 struct exitlist *ep; 1194 1195 TAILQ_FOREACH(ep, &exit_list, next) { 1196 if (ep->function == function) { 1197 TAILQ_REMOVE(&exit_list, ep, next); 1198 kfree(ep, M_ATEXIT); 1199 return(1); 1200 } 1201 } 1202 return (0); 1203 } 1204 1205 /* 1206 * LWP reaper related code. 1207 */ 1208 static void 1209 reaplwps(void *context, int dummy) 1210 { 1211 struct lwplist *lwplist = context; 1212 struct lwp *lp; 1213 1214 lwkt_gettoken(&deadlwp_token); 1215 while ((lp = LIST_FIRST(lwplist))) { 1216 LIST_REMOVE(lp, u.lwp_reap_entry); 1217 reaplwp(lp); 1218 } 1219 lwkt_reltoken(&deadlwp_token); 1220 } 1221 1222 static void 1223 reaplwp(struct lwp *lp) 1224 { 1225 while (lwp_wait(lp) == 0) 1226 ; 1227 lwp_dispose(lp); 1228 } 1229 1230 static void 1231 deadlwp_init(void) 1232 { 1233 int cpu; 1234 1235 for (cpu = 0; cpu < ncpus; cpu++) { 1236 LIST_INIT(&deadlwp_list[cpu]); 1237 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), 1238 M_DEVBUF, M_WAITOK); 1239 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1240 } 1241 } 1242 1243 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1244