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 * $DragonFly: src/sys/kern/kern_exit.c,v 1.91 2008/05/18 20:02:02 nth Exp $ 41 */ 42 43 #include "opt_compat.h" 44 #include "opt_ktrace.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/sysproto.h> 49 #include <sys/kernel.h> 50 #include <sys/malloc.h> 51 #include <sys/proc.h> 52 #include <sys/ktrace.h> 53 #include <sys/pioctl.h> 54 #include <sys/tty.h> 55 #include <sys/wait.h> 56 #include <sys/vnode.h> 57 #include <sys/resourcevar.h> 58 #include <sys/signalvar.h> 59 #include <sys/taskqueue.h> 60 #include <sys/ptrace.h> 61 #include <sys/acct.h> /* for acct_process() function prototype */ 62 #include <sys/filedesc.h> 63 #include <sys/shm.h> 64 #include <sys/sem.h> 65 #include <sys/jail.h> 66 #include <sys/kern_syscall.h> 67 #include <sys/upcall.h> 68 #include <sys/caps.h> 69 #include <sys/unistd.h> 70 #include <sys/eventhandler.h> 71 #include <sys/dsched.h> 72 73 #include <vm/vm.h> 74 #include <vm/vm_param.h> 75 #include <sys/lock.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_map.h> 78 #include <vm/vm_extern.h> 79 #include <sys/user.h> 80 81 #include <sys/refcount.h> 82 #include <sys/thread2.h> 83 #include <sys/sysref2.h> 84 #include <sys/mplock2.h> 85 86 static void reaplwps(void *context, int dummy); 87 static void reaplwp(struct lwp *lp); 88 static void killlwps(struct lwp *lp); 89 90 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 91 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status"); 92 93 static struct lwkt_token deadlwp_token = LWKT_TOKEN_INITIALIZER(deadlwp_token); 94 95 /* 96 * callout list for things to do at exit time 97 */ 98 struct exitlist { 99 exitlist_fn function; 100 TAILQ_ENTRY(exitlist) next; 101 }; 102 103 TAILQ_HEAD(exit_list_head, exitlist); 104 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 105 106 /* 107 * LWP reaper data 108 */ 109 struct task *deadlwp_task[MAXCPU]; 110 struct lwplist deadlwp_list[MAXCPU]; 111 112 /* 113 * exit -- 114 * Death of process. 115 * 116 * SYS_EXIT_ARGS(int rval) 117 */ 118 int 119 sys_exit(struct exit_args *uap) 120 { 121 exit1(W_EXITCODE(uap->rval, 0)); 122 /* NOTREACHED */ 123 } 124 125 /* 126 * Extended exit -- 127 * Death of a lwp or process with optional bells and whistles. 128 * 129 * MPALMOSTSAFE 130 */ 131 int 132 sys_extexit(struct extexit_args *uap) 133 { 134 struct proc *p = curproc; 135 int action, who; 136 int error; 137 138 action = EXTEXIT_ACTION(uap->how); 139 who = EXTEXIT_WHO(uap->how); 140 141 /* Check parameters before we might perform some action */ 142 switch (who) { 143 case EXTEXIT_PROC: 144 case EXTEXIT_LWP: 145 break; 146 default: 147 return (EINVAL); 148 } 149 150 switch (action) { 151 case EXTEXIT_SIMPLE: 152 break; 153 case EXTEXIT_SETINT: 154 error = copyout(&uap->status, uap->addr, sizeof(uap->status)); 155 if (error) 156 return (error); 157 break; 158 default: 159 return (EINVAL); 160 } 161 162 lwkt_gettoken(&p->p_token); 163 164 switch (who) { 165 case EXTEXIT_LWP: 166 /* 167 * Be sure only to perform a simple lwp exit if there is at 168 * least one more lwp in the proc, which will call exit1() 169 * later, otherwise the proc will be an UNDEAD and not even a 170 * SZOMB! 171 */ 172 if (p->p_nthreads > 1) { 173 lwp_exit(0); /* called w/ p_token held */ 174 /* NOT REACHED */ 175 } 176 /* else last lwp in proc: do the real thing */ 177 /* FALLTHROUGH */ 178 default: /* to help gcc */ 179 case EXTEXIT_PROC: 180 lwkt_reltoken(&p->p_token); 181 exit1(W_EXITCODE(uap->status, 0)); 182 /* NOTREACHED */ 183 } 184 185 /* NOTREACHED */ 186 lwkt_reltoken(&p->p_token); /* safety */ 187 } 188 189 /* 190 * Kill all lwps associated with the current process except the 191 * current lwp. Return an error if we race another thread trying to 192 * do the same thing and lose the race. 193 * 194 * If forexec is non-zero the current thread and process flags are 195 * cleaned up so they can be reused. 196 */ 197 int 198 killalllwps(int forexec) 199 { 200 struct lwp *lp = curthread->td_lwp; 201 struct proc *p = lp->lwp_proc; 202 203 /* 204 * Interlock against P_WEXIT. Only one of the process's thread 205 * is allowed to do the master exit. 206 */ 207 if (p->p_flag & P_WEXIT) 208 return (EALREADY); 209 p->p_flag |= P_WEXIT; 210 211 /* 212 * Interlock with LWP_WEXIT and kill any remaining LWPs 213 */ 214 lp->lwp_flag |= LWP_WEXIT; 215 if (p->p_nthreads > 1) 216 killlwps(lp); 217 218 /* 219 * If doing this for an exec, clean up the remaining thread 220 * (us) for continuing operation after all the other threads 221 * have been killed. 222 */ 223 if (forexec) { 224 lp->lwp_flag &= ~LWP_WEXIT; 225 p->p_flag &= ~P_WEXIT; 226 } 227 return(0); 228 } 229 230 /* 231 * Kill all LWPs except the current one. Do not try to signal 232 * LWPs which have exited on their own or have already been 233 * signaled. 234 */ 235 static void 236 killlwps(struct lwp *lp) 237 { 238 struct proc *p = lp->lwp_proc; 239 struct lwp *tlp; 240 241 /* 242 * Kill the remaining LWPs. We must send the signal before setting 243 * LWP_WEXIT. The setting of WEXIT is optional but helps reduce 244 * races. tlp must be held across the call as it might block and 245 * allow the target lwp to rip itself out from under our loop. 246 */ 247 FOREACH_LWP_IN_PROC(tlp, p) { 248 LWPHOLD(tlp); 249 if ((tlp->lwp_flag & LWP_WEXIT) == 0) { 250 lwpsignal(p, tlp, SIGKILL); 251 tlp->lwp_flag |= LWP_WEXIT; 252 } 253 LWPRELE(tlp); 254 } 255 256 /* 257 * Wait for everything to clear out. 258 */ 259 while (p->p_nthreads > 1) { 260 tsleep(&p->p_nthreads, 0, "killlwps", 0); 261 } 262 } 263 264 /* 265 * Exit: deallocate address space and other resources, change proc state 266 * to zombie, and unlink proc from allproc and parent's lists. Save exit 267 * status and rusage for wait(). Check for child processes and orphan them. 268 */ 269 void 270 exit1(int rv) 271 { 272 struct thread *td = curthread; 273 struct proc *p = td->td_proc; 274 struct lwp *lp = td->td_lwp; 275 struct proc *q, *nq; 276 struct vmspace *vm; 277 struct vnode *vtmp; 278 struct exitlist *ep; 279 int error; 280 281 lwkt_gettoken(&p->p_token); 282 283 if (p->p_pid == 1) { 284 kprintf("init died (signal %d, exit %d)\n", 285 WTERMSIG(rv), WEXITSTATUS(rv)); 286 panic("Going nowhere without my init!"); 287 } 288 varsymset_clean(&p->p_varsymset); 289 lockuninit(&p->p_varsymset.vx_lock); 290 /* 291 * Kill all lwps associated with the current process, return an 292 * error if we race another thread trying to do the same thing 293 * and lose the race. 294 */ 295 error = killalllwps(0); 296 if (error) { 297 lwp_exit(0); 298 /* NOT REACHED */ 299 } 300 301 caps_exit(lp->lwp_thread); 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 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */ 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_flag & P_PROFIL) 344 stopprofclock(p); 345 /* 346 * If parent is waiting for us to exit or exec, 347 * P_PPWAIT is set; we will wakeup the parent below. 348 */ 349 p->p_flag &= ~(P_TRACED | P_PPWAIT); 350 SIGEMPTYSET(p->p_siglist); 351 SIGEMPTYSET(lp->lwp_siglist); 352 if (timevalisset(&p->p_realtimer.it_value)) 353 callout_stop(&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 * Move the process to the zombie list. This will block 479 * until the process p_lock count reaches 0. The process will 480 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 481 * which is called from cpu_proc_exit(). 482 */ 483 proc_move_allproc_zombie(p); 484 485 /* 486 * Reparent all of this process's children to the init process. 487 * We must hold initproc->p_token in order to mess with 488 * initproc->p_children. We already hold p->p_token (to remove 489 * the children from our list). 490 */ 491 q = LIST_FIRST(&p->p_children); 492 if (q) { 493 lwkt_gettoken(&initproc->p_token); 494 while (q) { 495 nq = LIST_NEXT(q, p_sibling); 496 LIST_REMOVE(q, p_sibling); 497 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling); 498 q->p_pptr = initproc; 499 q->p_sigparent = SIGCHLD; 500 /* 501 * Traced processes are killed 502 * since their existence means someone is screwing up. 503 */ 504 if (q->p_flag & P_TRACED) { 505 q->p_flag &= ~P_TRACED; 506 ksignal(q, SIGKILL); 507 } 508 q = nq; 509 } 510 lwkt_reltoken(&initproc->p_token); 511 wakeup(initproc); 512 } 513 514 /* 515 * Save exit status and final rusage info, adding in child rusage 516 * info and self times. 517 */ 518 calcru_proc(p, &p->p_ru); 519 ruadd(&p->p_ru, &p->p_cru); 520 521 /* 522 * notify interested parties of our demise. 523 */ 524 KNOTE(&p->p_klist, NOTE_EXIT); 525 526 /* 527 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 528 * flag set, notify process 1 instead (and hope it will handle 529 * this situation). 530 */ 531 if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) { 532 struct proc *pp = p->p_pptr; 533 534 PHOLD(pp); 535 proc_reparent(p, initproc); 536 537 /* 538 * If this was the last child of our parent, notify 539 * parent, so in case he was wait(2)ing, he will 540 * continue. This function interlocks with pptr->p_token. 541 */ 542 if (LIST_EMPTY(&pp->p_children)) 543 wakeup((caddr_t)pp); 544 PRELE(pp); 545 } 546 547 /* lwkt_gettoken(&proc_token); */ 548 q = p->p_pptr; 549 PHOLD(q); 550 if (p->p_sigparent && q != initproc) { 551 ksignal(q, p->p_sigparent); 552 } else { 553 ksignal(q, SIGCHLD); 554 } 555 wakeup(p->p_pptr); 556 PRELE(q); 557 /* lwkt_reltoken(&proc_token); */ 558 /* NOTE: p->p_pptr can get ripped out */ 559 /* 560 * cpu_exit is responsible for clearing curproc, since 561 * it is heavily integrated with the thread/switching sequence. 562 * 563 * Other substructures are freed from wait(). 564 */ 565 plimit_free(p); 566 567 /* 568 * Release the current user process designation on the process so 569 * the userland scheduler can work in someone else. 570 */ 571 p->p_usched->release_curproc(lp); 572 573 /* 574 * Finally, call machine-dependent code to release as many of the 575 * lwp's resources as we can and halt execution of this thread. 576 */ 577 lwp_exit(1); 578 } 579 580 /* 581 * Eventually called by every exiting LWP 582 * 583 * p->p_token must be held. mplock may be held and will be released. 584 */ 585 void 586 lwp_exit(int masterexit) 587 { 588 struct thread *td = curthread; 589 struct lwp *lp = td->td_lwp; 590 struct proc *p = lp->lwp_proc; 591 592 /* 593 * lwp_exit() may be called without setting LWP_WEXIT, so 594 * make sure it is set here. 595 */ 596 ASSERT_LWKT_TOKEN_HELD(&p->p_token); 597 lp->lwp_flag |= LWP_WEXIT; 598 599 /* 600 * Clean up any virtualization 601 */ 602 if (lp->lwp_vkernel) 603 vkernel_lwp_exit(lp); 604 605 /* 606 * Clean up select/poll support 607 */ 608 kqueue_terminate(&lp->lwp_kqueue); 609 610 /* 611 * Clean up any syscall-cached ucred 612 */ 613 if (td->td_ucred) { 614 crfree(td->td_ucred); 615 td->td_ucred = NULL; 616 } 617 618 /* 619 * Nobody actually wakes us when the lock 620 * count reaches zero, so just wait one tick. 621 */ 622 while (lp->lwp_lock > 0) 623 tsleep(lp, 0, "lwpexit", 1); 624 625 /* Hand down resource usage to our proc */ 626 ruadd(&p->p_ru, &lp->lwp_ru); 627 628 /* 629 * If we don't hold the process until the LWP is reaped wait*() 630 * may try to dispose of its vmspace before all the LWPs have 631 * actually terminated. 632 */ 633 PHOLD(p); 634 635 /* 636 * Do any remaining work that might block on us. We should be 637 * coded such that further blocking is ok after decrementing 638 * p_nthreads but don't take the chance. 639 */ 640 dsched_exit_thread(td); 641 biosched_done(curthread); 642 643 /* 644 * We have to use the reaper for all the LWPs except the one doing 645 * the master exit. The LWP doing the master exit can just be 646 * left on p_lwps and the process reaper will deal with it 647 * synchronously, which is much faster. 648 * 649 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0. 650 */ 651 if (masterexit == 0) { 652 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 653 --p->p_nthreads; 654 if (p->p_nthreads <= 1) 655 wakeup(&p->p_nthreads); 656 lwkt_gettoken(&deadlwp_token); 657 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry); 658 taskqueue_enqueue(taskqueue_thread[mycpuid], 659 deadlwp_task[mycpuid]); 660 lwkt_reltoken(&deadlwp_token); 661 } else { 662 --p->p_nthreads; 663 if (p->p_nthreads <= 1) 664 wakeup(&p->p_nthreads); 665 } 666 667 /* 668 * Release p_token. The mp_token may also be held and we depend on 669 * the lwkt_switch() code to clean it up. 670 */ 671 lwkt_reltoken(&p->p_token); 672 cpu_lwp_exit(); 673 } 674 675 /* 676 * Wait until a lwp is completely dead. 677 * 678 * If the thread is still executing, which can't be waited upon, 679 * return failure. The caller is responsible of waiting a little 680 * bit and checking again. 681 * 682 * Suggested use: 683 * while (!lwp_wait(lp)) 684 * tsleep(lp, 0, "lwpwait", 1); 685 */ 686 static int 687 lwp_wait(struct lwp *lp) 688 { 689 struct thread *td = lp->lwp_thread;; 690 691 KKASSERT(lwkt_preempted_proc() != lp); 692 693 while (lp->lwp_lock > 0) 694 tsleep(lp, 0, "lwpwait1", 1); 695 696 lwkt_wait_free(td); 697 698 /* 699 * The lwp's thread may still be in the middle 700 * of switching away, we can't rip its stack out from 701 * under it until TDF_EXITING is set and both 702 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 703 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 704 * will be cleared temporarily if a thread gets 705 * preempted. 706 * 707 * YYY no wakeup occurs, so we simply return failure 708 * and let the caller deal with sleeping and calling 709 * us again. 710 */ 711 if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK| 712 TDF_EXITING|TDF_RUNQ)) != TDF_EXITING) { 713 return (0); 714 } 715 KASSERT((td->td_flags & TDF_TSLEEPQ) == 0, 716 ("lwp_wait: td %p (%s) still on sleep queue", td, td->td_comm)); 717 return (1); 718 } 719 720 /* 721 * Release the resources associated with a lwp. 722 * The lwp must be completely dead. 723 */ 724 void 725 lwp_dispose(struct lwp *lp) 726 { 727 struct thread *td = lp->lwp_thread;; 728 729 KKASSERT(lwkt_preempted_proc() != lp); 730 KKASSERT(td->td_refs == 0); 731 KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) == 732 TDF_EXITING); 733 734 PRELE(lp->lwp_proc); 735 lp->lwp_proc = NULL; 736 if (td != NULL) { 737 td->td_proc = NULL; 738 td->td_lwp = NULL; 739 lp->lwp_thread = NULL; 740 lwkt_free_thread(td); 741 } 742 kfree(lp, M_LWP); 743 } 744 745 /* 746 * MPSAFE 747 */ 748 int 749 sys_wait4(struct wait_args *uap) 750 { 751 struct rusage rusage; 752 int error, status; 753 754 error = kern_wait(uap->pid, (uap->status ? &status : NULL), 755 uap->options, (uap->rusage ? &rusage : NULL), 756 &uap->sysmsg_result); 757 758 if (error == 0 && uap->status) 759 error = copyout(&status, uap->status, sizeof(*uap->status)); 760 if (error == 0 && uap->rusage) 761 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage)); 762 return (error); 763 } 764 765 /* 766 * wait1() 767 * 768 * wait_args(int pid, int *status, int options, struct rusage *rusage) 769 * 770 * MPALMOSTSAFE 771 */ 772 int 773 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res) 774 { 775 struct thread *td = curthread; 776 struct lwp *lp; 777 struct proc *q = td->td_proc; 778 struct proc *p, *t; 779 struct pargs *pa; 780 struct sigacts *ps; 781 int nfound, error; 782 783 if (pid == 0) 784 pid = -q->p_pgid; 785 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE)) 786 return (EINVAL); 787 788 lwkt_gettoken(&q->p_token); 789 loop: 790 /* 791 * All sorts of things can change due to blocking so we have to loop 792 * all the way back up here. 793 * 794 * The problem is that if a process group is stopped and the parent 795 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 796 * of the child and then stop itself when it tries to return from the 797 * system call. When the process group is resumed the parent will 798 * then get the STOP status even though the child has now resumed 799 * (a followup wait*() will get the CONT status). 800 * 801 * Previously the CONT would overwrite the STOP because the tstop 802 * was handled within tsleep(), and the parent would only see 803 * the CONT when both are stopped and continued together. This little 804 * two-line hack restores this effect. 805 */ 806 while (q->p_stat == SSTOP) 807 tstop(); 808 809 nfound = 0; 810 811 LIST_FOREACH(p, &q->p_children, p_sibling) { 812 if (pid != WAIT_ANY && 813 p->p_pid != pid && p->p_pgid != -pid) { 814 continue; 815 } 816 817 /* 818 * This special case handles a kthread spawned by linux_clone 819 * (see linux_misc.c). The linux_wait4 and linux_waitpid 820 * functions need to be able to distinguish between waiting 821 * on a process and waiting on a thread. It is a thread if 822 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 823 * signifies we want to wait for threads and not processes. 824 */ 825 if ((p->p_sigparent != SIGCHLD) ^ 826 ((options & WLINUXCLONE) != 0)) { 827 continue; 828 } 829 830 nfound++; 831 if (p->p_stat == SZOMB) { 832 /* 833 * We may go into SZOMB with threads still present. 834 * We must wait for them to exit before we can reap 835 * the master thread, otherwise we may race reaping 836 * non-master threads. 837 */ 838 lwkt_gettoken(&p->p_token); 839 while (p->p_nthreads > 0) { 840 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 841 } 842 843 /* 844 * Reap any LWPs left in p->p_lwps. This is usually 845 * just the last LWP. This must be done before 846 * we loop on p_lock since the lwps hold a ref on 847 * it as a vmspace interlock. 848 * 849 * Once that is accomplished p_nthreads had better 850 * be zero. 851 */ 852 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 853 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 854 reaplwp(lp); 855 } 856 KKASSERT(p->p_nthreads == 0); 857 lwkt_reltoken(&p->p_token); 858 859 /* 860 * Don't do anything really bad until all references 861 * to the process go away. This may include other 862 * LWPs which are still in the process of being 863 * reaped. We can't just pull the rug out from under 864 * them because they may still be using the VM space. 865 * 866 * Certain kernel facilities such as /proc will also 867 * put a hold on the process for short periods of 868 * time. 869 */ 870 while (p->p_lock) 871 tsleep(p, 0, "reap3", hz); 872 873 /* Take care of our return values. */ 874 *res = p->p_pid; 875 p->p_usched->heuristic_exiting(td->td_lwp, p); 876 877 if (status) 878 *status = p->p_xstat; 879 if (rusage) 880 *rusage = p->p_ru; 881 /* 882 * If we got the child via a ptrace 'attach', 883 * we need to give it back to the old parent. 884 */ 885 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 886 p->p_oppid = 0; 887 proc_reparent(p, t); 888 ksignal(t, SIGCHLD); 889 wakeup((caddr_t)t); 890 error = 0; 891 PRELE(t); 892 goto done; 893 } 894 895 /* 896 * Unlink the proc from its process group so that 897 * the following operations won't lead to an 898 * inconsistent state for processes running down 899 * the zombie list. 900 */ 901 KKASSERT(p->p_lock == 0); 902 proc_remove_zombie(p); 903 leavepgrp(p); 904 905 p->p_xstat = 0; 906 ruadd(&q->p_cru, &p->p_ru); 907 908 /* 909 * Decrement the count of procs running with this uid. 910 */ 911 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 912 913 /* 914 * Free up credentials. 915 */ 916 crfree(p->p_ucred); 917 p->p_ucred = NULL; 918 919 /* 920 * Remove unused arguments 921 */ 922 pa = p->p_args; 923 p->p_args = NULL; 924 if (pa && refcount_release(&pa->ar_ref)) { 925 kfree(pa, M_PARGS); 926 pa = NULL; 927 } 928 929 ps = p->p_sigacts; 930 p->p_sigacts = NULL; 931 if (ps && refcount_release(&ps->ps_refcnt)) { 932 kfree(ps, M_SUBPROC); 933 ps = NULL; 934 } 935 936 vm_waitproc(p); 937 kfree(p, M_PROC); 938 nprocs--; 939 error = 0; 940 goto done; 941 } 942 if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 && 943 (p->p_flag & P_TRACED || options & WUNTRACED)) { 944 p->p_flag |= P_WAITED; 945 946 *res = p->p_pid; 947 p->p_usched->heuristic_exiting(td->td_lwp, p); 948 if (status) 949 *status = W_STOPCODE(p->p_xstat); 950 /* Zero rusage so we get something consistent. */ 951 if (rusage) 952 bzero(rusage, sizeof(rusage)); 953 error = 0; 954 goto done; 955 } 956 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) { 957 *res = p->p_pid; 958 p->p_usched->heuristic_exiting(td->td_lwp, p); 959 p->p_flag &= ~P_CONTINUED; 960 961 if (status) 962 *status = SIGCONT; 963 error = 0; 964 goto done; 965 } 966 } 967 if (nfound == 0) { 968 error = ECHILD; 969 goto done; 970 } 971 if (options & WNOHANG) { 972 *res = 0; 973 error = 0; 974 goto done; 975 } 976 977 /* 978 * Wait for signal - interlocked using q->p_token. 979 */ 980 error = tsleep(q, PCATCH, "wait", 0); 981 if (error) { 982 done: 983 lwkt_reltoken(&q->p_token); 984 return (error); 985 } 986 goto loop; 987 } 988 989 /* 990 * Make process 'parent' the new parent of process 'child'. 991 * 992 * p_children/p_sibling requires the parent's token, and 993 * changing pptr requires the child's token, so we have to 994 * get three tokens to do this operation. 995 */ 996 void 997 proc_reparent(struct proc *child, struct proc *parent) 998 { 999 struct proc *opp = child->p_pptr; 1000 1001 if (opp == parent) 1002 return; 1003 PHOLD(opp); 1004 PHOLD(parent); 1005 lwkt_gettoken(&opp->p_token); 1006 lwkt_gettoken(&child->p_token); 1007 lwkt_gettoken(&parent->p_token); 1008 KKASSERT(child->p_pptr == opp); 1009 LIST_REMOVE(child, p_sibling); 1010 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1011 child->p_pptr = parent; 1012 lwkt_reltoken(&parent->p_token); 1013 lwkt_reltoken(&child->p_token); 1014 lwkt_reltoken(&opp->p_token); 1015 PRELE(parent); 1016 PRELE(opp); 1017 } 1018 1019 /* 1020 * The next two functions are to handle adding/deleting items on the 1021 * exit callout list 1022 * 1023 * at_exit(): 1024 * Take the arguments given and put them onto the exit callout list, 1025 * However first make sure that it's not already there. 1026 * returns 0 on success. 1027 */ 1028 1029 int 1030 at_exit(exitlist_fn function) 1031 { 1032 struct exitlist *ep; 1033 1034 #ifdef INVARIANTS 1035 /* Be noisy if the programmer has lost track of things */ 1036 if (rm_at_exit(function)) 1037 kprintf("WARNING: exit callout entry (%p) already present\n", 1038 function); 1039 #endif 1040 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 1041 if (ep == NULL) 1042 return (ENOMEM); 1043 ep->function = function; 1044 TAILQ_INSERT_TAIL(&exit_list, ep, next); 1045 return (0); 1046 } 1047 1048 /* 1049 * Scan the exit callout list for the given item and remove it. 1050 * Returns the number of items removed (0 or 1) 1051 */ 1052 int 1053 rm_at_exit(exitlist_fn function) 1054 { 1055 struct exitlist *ep; 1056 1057 TAILQ_FOREACH(ep, &exit_list, next) { 1058 if (ep->function == function) { 1059 TAILQ_REMOVE(&exit_list, ep, next); 1060 kfree(ep, M_ATEXIT); 1061 return(1); 1062 } 1063 } 1064 return (0); 1065 } 1066 1067 /* 1068 * LWP reaper related code. 1069 */ 1070 static void 1071 reaplwps(void *context, int dummy) 1072 { 1073 struct lwplist *lwplist = context; 1074 struct lwp *lp; 1075 1076 lwkt_gettoken(&deadlwp_token); 1077 while ((lp = LIST_FIRST(lwplist))) { 1078 LIST_REMOVE(lp, u.lwp_reap_entry); 1079 reaplwp(lp); 1080 } 1081 lwkt_reltoken(&deadlwp_token); 1082 } 1083 1084 static void 1085 reaplwp(struct lwp *lp) 1086 { 1087 while (lwp_wait(lp) == 0) 1088 tsleep(lp, 0, "lwpreap", 1); 1089 lwp_dispose(lp); 1090 } 1091 1092 static void 1093 deadlwp_init(void) 1094 { 1095 int cpu; 1096 1097 for (cpu = 0; cpu < ncpus; cpu++) { 1098 LIST_INIT(&deadlwp_list[cpu]); 1099 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK); 1100 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1101 } 1102 } 1103 1104 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1105