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