1 /* 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 35 * $FreeBSD: src/sys/kern/kern_exit.c,v 1.92.2.11 2003/01/13 22:51:16 dillon Exp $ 36 */ 37 38 #include "opt_compat.h" 39 #include "opt_ktrace.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/sysproto.h> 44 #include <sys/kernel.h> 45 #include <sys/malloc.h> 46 #include <sys/proc.h> 47 #include <sys/ktrace.h> 48 #include <sys/pioctl.h> 49 #include <sys/tty.h> 50 #include <sys/wait.h> 51 #include <sys/vnode.h> 52 #include <sys/resourcevar.h> 53 #include <sys/signalvar.h> 54 #include <sys/taskqueue.h> 55 #include <sys/ptrace.h> 56 #include <sys/acct.h> /* for acct_process() function prototype */ 57 #include <sys/filedesc.h> 58 #include <sys/shm.h> 59 #include <sys/sem.h> 60 #include <sys/jail.h> 61 #include <sys/kern_syscall.h> 62 #include <sys/unistd.h> 63 #include <sys/eventhandler.h> 64 #include <sys/dsched.h> 65 66 #include <vm/vm.h> 67 #include <vm/vm_param.h> 68 #include <sys/lock.h> 69 #include <vm/pmap.h> 70 #include <vm/vm_map.h> 71 #include <vm/vm_extern.h> 72 #include <sys/user.h> 73 74 #include <sys/refcount.h> 75 #include <sys/thread2.h> 76 #include <sys/sysref2.h> 77 #include <sys/mplock2.h> 78 79 static void reaplwps(void *context, int dummy); 80 static void reaplwp(struct lwp *lp); 81 static void killlwps(struct lwp *lp); 82 83 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 84 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status"); 85 86 static struct lwkt_token deadlwp_token = LWKT_TOKEN_INITIALIZER(deadlwp_token); 87 88 /* 89 * callout list for things to do at exit time 90 */ 91 struct exitlist { 92 exitlist_fn function; 93 TAILQ_ENTRY(exitlist) next; 94 }; 95 96 TAILQ_HEAD(exit_list_head, exitlist); 97 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 98 99 /* 100 * LWP reaper data 101 */ 102 struct task *deadlwp_task[MAXCPU]; 103 struct lwplist deadlwp_list[MAXCPU]; 104 105 /* 106 * exit -- 107 * Death of process. 108 * 109 * SYS_EXIT_ARGS(int rval) 110 */ 111 int 112 sys_exit(struct exit_args *uap) 113 { 114 exit1(W_EXITCODE(uap->rval, 0)); 115 /* NOTREACHED */ 116 } 117 118 /* 119 * Extended exit -- 120 * Death of a lwp or process with optional bells and whistles. 121 * 122 * MPALMOSTSAFE 123 */ 124 int 125 sys_extexit(struct extexit_args *uap) 126 { 127 struct proc *p = curproc; 128 int action, who; 129 int error; 130 131 action = EXTEXIT_ACTION(uap->how); 132 who = EXTEXIT_WHO(uap->how); 133 134 /* Check parameters before we might perform some action */ 135 switch (who) { 136 case EXTEXIT_PROC: 137 case EXTEXIT_LWP: 138 break; 139 default: 140 return (EINVAL); 141 } 142 143 switch (action) { 144 case EXTEXIT_SIMPLE: 145 break; 146 case EXTEXIT_SETINT: 147 error = copyout(&uap->status, uap->addr, sizeof(uap->status)); 148 if (error) 149 return (error); 150 break; 151 default: 152 return (EINVAL); 153 } 154 155 lwkt_gettoken(&p->p_token); 156 157 switch (who) { 158 case EXTEXIT_LWP: 159 /* 160 * Be sure only to perform a simple lwp exit if there is at 161 * least one more lwp in the proc, which will call exit1() 162 * later, otherwise the proc will be an UNDEAD and not even a 163 * SZOMB! 164 */ 165 if (p->p_nthreads > 1) { 166 lwp_exit(0); /* called w/ p_token held */ 167 /* NOT REACHED */ 168 } 169 /* else last lwp in proc: do the real thing */ 170 /* FALLTHROUGH */ 171 default: /* to help gcc */ 172 case EXTEXIT_PROC: 173 lwkt_reltoken(&p->p_token); 174 exit1(W_EXITCODE(uap->status, 0)); 175 /* NOTREACHED */ 176 } 177 178 /* NOTREACHED */ 179 lwkt_reltoken(&p->p_token); /* safety */ 180 } 181 182 /* 183 * Kill all lwps associated with the current process except the 184 * current lwp. Return an error if we race another thread trying to 185 * do the same thing and lose the race. 186 * 187 * If forexec is non-zero the current thread and process flags are 188 * cleaned up so they can be reused. 189 * 190 * Caller must hold curproc->p_token 191 */ 192 int 193 killalllwps(int forexec) 194 { 195 struct lwp *lp = curthread->td_lwp; 196 struct proc *p = lp->lwp_proc; 197 198 /* 199 * Interlock against P_WEXIT. Only one of the process's thread 200 * is allowed to do the master exit. 201 */ 202 if (p->p_flags & P_WEXIT) 203 return (EALREADY); 204 p->p_flags |= P_WEXIT; 205 206 /* 207 * Interlock with LWP_MP_WEXIT and kill any remaining LWPs 208 */ 209 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 210 if (p->p_nthreads > 1) 211 killlwps(lp); 212 213 /* 214 * If doing this for an exec, clean up the remaining thread 215 * (us) for continuing operation after all the other threads 216 * have been killed. 217 */ 218 if (forexec) { 219 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 220 p->p_flags &= ~P_WEXIT; 221 } 222 return(0); 223 } 224 225 /* 226 * Kill all LWPs except the current one. Do not try to signal 227 * LWPs which have exited on their own or have already been 228 * signaled. 229 */ 230 static void 231 killlwps(struct lwp *lp) 232 { 233 struct proc *p = lp->lwp_proc; 234 struct lwp *tlp; 235 236 /* 237 * Kill the remaining LWPs. We must send the signal before setting 238 * LWP_MP_WEXIT. The setting of WEXIT is optional but helps reduce 239 * races. tlp must be held across the call as it might block and 240 * allow the target lwp to rip itself out from under our loop. 241 */ 242 FOREACH_LWP_IN_PROC(tlp, p) { 243 LWPHOLD(tlp); 244 lwkt_gettoken(&tlp->lwp_token); 245 if ((tlp->lwp_mpflags & LWP_MP_WEXIT) == 0) { 246 lwpsignal(p, tlp, SIGKILL); 247 atomic_set_int(&tlp->lwp_mpflags, LWP_MP_WEXIT); 248 } 249 lwkt_reltoken(&tlp->lwp_token); 250 LWPRELE(tlp); 251 } 252 253 /* 254 * Wait for everything to clear out. 255 */ 256 while (p->p_nthreads > 1) { 257 tsleep(&p->p_nthreads, 0, "killlwps", 0); 258 } 259 } 260 261 /* 262 * Exit: deallocate address space and other resources, change proc state 263 * to zombie, and unlink proc from allproc and parent's lists. Save exit 264 * status and rusage for wait(). Check for child processes and orphan them. 265 */ 266 void 267 exit1(int rv) 268 { 269 struct thread *td = curthread; 270 struct proc *p = td->td_proc; 271 struct lwp *lp = td->td_lwp; 272 struct proc *q; 273 struct vmspace *vm; 274 struct vnode *vtmp; 275 struct exitlist *ep; 276 int error; 277 278 lwkt_gettoken(&p->p_token); 279 280 if (p->p_pid == 1) { 281 kprintf("init died (signal %d, exit %d)\n", 282 WTERMSIG(rv), WEXITSTATUS(rv)); 283 panic("Going nowhere without my init!"); 284 } 285 varsymset_clean(&p->p_varsymset); 286 lockuninit(&p->p_varsymset.vx_lock); 287 288 /* 289 * Kill all lwps associated with the current process, return an 290 * error if we race another thread trying to do the same thing 291 * and lose the race. 292 */ 293 error = killalllwps(0); 294 if (error) { 295 lwp_exit(0); 296 /* NOT REACHED */ 297 } 298 299 /* are we a task leader? */ 300 if (p == p->p_leader) { 301 struct kill_args killArgs; 302 killArgs.signum = SIGKILL; 303 q = p->p_peers; 304 while(q) { 305 killArgs.pid = q->p_pid; 306 /* 307 * The interface for kill is better 308 * than the internal signal 309 */ 310 sys_kill(&killArgs); 311 q = q->p_peers; 312 } 313 while (p->p_peers) 314 tsleep((caddr_t)p, 0, "exit1", 0); 315 } 316 317 #ifdef PGINPROF 318 vmsizmon(); 319 #endif 320 STOPEVENT(p, S_EXIT, rv); 321 p->p_flags |= P_POSTEXIT; /* stop procfs stepping */ 322 323 /* 324 * Check if any loadable modules need anything done at process exit. 325 * e.g. SYSV IPC stuff 326 * XXX what if one of these generates an error? 327 */ 328 p->p_xstat = rv; 329 EVENTHANDLER_INVOKE(process_exit, p); 330 331 /* 332 * XXX: imho, the eventhandler stuff is much cleaner than this. 333 * Maybe we should move everything to use eventhandler. 334 */ 335 TAILQ_FOREACH(ep, &exit_list, next) 336 (*ep->function)(td); 337 338 if (p->p_flags & P_PROFIL) 339 stopprofclock(p); 340 341 SIGEMPTYSET(p->p_siglist); 342 SIGEMPTYSET(lp->lwp_siglist); 343 if (timevalisset(&p->p_realtimer.it_value)) 344 callout_stop_sync(&p->p_ithandle); 345 346 /* 347 * Reset any sigio structures pointing to us as a result of 348 * F_SETOWN with our pid. 349 */ 350 funsetownlst(&p->p_sigiolst); 351 352 /* 353 * Close open files and release open-file table. 354 * This may block! 355 */ 356 fdfree(p, NULL); 357 358 if(p->p_leader->p_peers) { 359 q = p->p_leader; 360 while(q->p_peers != p) 361 q = q->p_peers; 362 q->p_peers = p->p_peers; 363 wakeup((caddr_t)p->p_leader); 364 } 365 366 /* 367 * XXX Shutdown SYSV semaphores 368 */ 369 semexit(p); 370 371 KKASSERT(p->p_numposixlocks == 0); 372 373 /* The next two chunks should probably be moved to vmspace_exit. */ 374 vm = p->p_vmspace; 375 376 /* 377 * Clean up data related to virtual kernel operation. Clean up 378 * any vkernel context related to the current lwp now so we can 379 * destroy p_vkernel. 380 */ 381 if (p->p_vkernel) { 382 vkernel_lwp_exit(lp); 383 vkernel_exit(p); 384 } 385 386 /* 387 * Release user portion of address space. 388 * This releases references to vnodes, 389 * which could cause I/O if the file has been unlinked. 390 * Need to do this early enough that we can still sleep. 391 * Can't free the entire vmspace as the kernel stack 392 * may be mapped within that space also. 393 * 394 * Processes sharing the same vmspace may exit in one order, and 395 * get cleaned up by vmspace_exit() in a different order. The 396 * last exiting process to reach this point releases as much of 397 * the environment as it can, and the last process cleaned up 398 * by vmspace_exit() (which decrements exitingcnt) cleans up the 399 * remainder. 400 */ 401 vmspace_exitbump(vm); 402 sysref_put(&vm->vm_sysref); 403 404 if (SESS_LEADER(p)) { 405 struct session *sp = p->p_session; 406 407 if (sp->s_ttyvp) { 408 /* 409 * We are the controlling process. Signal the 410 * foreground process group, drain the controlling 411 * terminal, and revoke access to the controlling 412 * terminal. 413 * 414 * NOTE: while waiting for the process group to exit 415 * it is possible that one of the processes in the 416 * group will revoke the tty, so the ttyclosesession() 417 * function will re-check sp->s_ttyvp. 418 */ 419 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 420 if (sp->s_ttyp->t_pgrp) 421 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 422 ttywait(sp->s_ttyp); 423 ttyclosesession(sp, 1); /* also revoke */ 424 } 425 /* 426 * Release the tty. If someone has it open via 427 * /dev/tty then close it (since they no longer can 428 * once we've NULL'd it out). 429 */ 430 ttyclosesession(sp, 0); 431 432 /* 433 * s_ttyp is not zero'd; we use this to indicate 434 * that the session once had a controlling terminal. 435 * (for logging and informational purposes) 436 */ 437 } 438 sp->s_leader = NULL; 439 } 440 fixjobc(p, p->p_pgrp, 0); 441 (void)acct_process(p); 442 #ifdef KTRACE 443 /* 444 * release trace file 445 */ 446 if (p->p_tracenode) 447 ktrdestroy(&p->p_tracenode); 448 p->p_traceflag = 0; 449 #endif 450 /* 451 * Release reference to text vnode 452 */ 453 if ((vtmp = p->p_textvp) != NULL) { 454 p->p_textvp = NULL; 455 vrele(vtmp); 456 } 457 458 /* Release namecache handle to text file */ 459 if (p->p_textnch.ncp) 460 cache_drop(&p->p_textnch); 461 462 /* 463 * We have to handle PPWAIT here or proc_move_allproc_zombie() 464 * will block on the PHOLD() the parent is doing. 465 */ 466 if (p->p_flags & P_PPWAIT) { 467 p->p_flags &= ~P_PPWAIT; 468 wakeup(p->p_pptr); 469 } 470 471 /* 472 * Move the process to the zombie list. This will block 473 * until the process p_lock count reaches 0. The process will 474 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 475 * which is called from cpu_proc_exit(). 476 */ 477 proc_move_allproc_zombie(p); 478 479 /* 480 * Reparent all of this process's children to the init process. 481 * We must hold initproc->p_token in order to mess with 482 * initproc->p_children. We already hold p->p_token (to remove 483 * the children from our list). 484 */ 485 q = LIST_FIRST(&p->p_children); 486 if (q) { 487 lwkt_gettoken(&initproc->p_token); 488 while ((q = LIST_FIRST(&p->p_children)) != NULL) { 489 PHOLD(q); 490 lwkt_gettoken(&q->p_token); 491 if (q != LIST_FIRST(&p->p_children)) { 492 lwkt_reltoken(&q->p_token); 493 PRELE(q); 494 continue; 495 } 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 /* 502 * Traced processes are killed 503 * since their existence means someone is screwing up. 504 */ 505 if (q->p_flags & P_TRACED) { 506 q->p_flags &= ~P_TRACED; 507 ksignal(q, SIGKILL); 508 } 509 lwkt_reltoken(&q->p_token); 510 PRELE(q); 511 } 512 lwkt_reltoken(&initproc->p_token); 513 wakeup(initproc); 514 } 515 516 /* 517 * Save exit status and final rusage info, adding in child rusage 518 * info and self times. 519 */ 520 calcru_proc(p, &p->p_ru); 521 ruadd(&p->p_ru, &p->p_cru); 522 523 /* 524 * notify interested parties of our demise. 525 */ 526 KNOTE(&p->p_klist, NOTE_EXIT); 527 528 /* 529 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 530 * flag set, or if the handler is set to SIG_IGN, notify process 1 531 * instead (and hope it will handle this situation). 532 */ 533 if (p->p_pptr->p_sigacts->ps_flag & (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 534 proc_reparent(p, initproc); 535 } 536 537 /* lwkt_gettoken(&proc_token); */ 538 q = p->p_pptr; 539 PHOLD(q); 540 if (p->p_sigparent && q != initproc) { 541 ksignal(q, p->p_sigparent); 542 } else { 543 ksignal(q, SIGCHLD); 544 } 545 546 p->p_flags &= ~P_TRACED; 547 wakeup(p->p_pptr); 548 549 PRELE(q); 550 /* lwkt_reltoken(&proc_token); */ 551 /* NOTE: p->p_pptr can get ripped out */ 552 /* 553 * cpu_exit is responsible for clearing curproc, since 554 * it is heavily integrated with the thread/switching sequence. 555 * 556 * Other substructures are freed from wait(). 557 */ 558 plimit_free(p); 559 560 /* 561 * Release the current user process designation on the process so 562 * the userland scheduler can work in someone else. 563 */ 564 p->p_usched->release_curproc(lp); 565 566 /* 567 * Finally, call machine-dependent code to release as many of the 568 * lwp's resources as we can and halt execution of this thread. 569 */ 570 lwp_exit(1); 571 } 572 573 /* 574 * Eventually called by every exiting LWP 575 * 576 * p->p_token must be held. mplock may be held and will be released. 577 */ 578 void 579 lwp_exit(int masterexit) 580 { 581 struct thread *td = curthread; 582 struct lwp *lp = td->td_lwp; 583 struct proc *p = lp->lwp_proc; 584 int dowake = 0; 585 586 /* 587 * lwp_exit() may be called without setting LWP_MP_WEXIT, so 588 * make sure it is set here. 589 */ 590 ASSERT_LWKT_TOKEN_HELD(&p->p_token); 591 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WEXIT); 592 593 /* 594 * Clean up any virtualization 595 */ 596 if (lp->lwp_vkernel) 597 vkernel_lwp_exit(lp); 598 599 /* 600 * Clean up select/poll support 601 */ 602 kqueue_terminate(&lp->lwp_kqueue); 603 604 /* 605 * Clean up any syscall-cached ucred 606 */ 607 if (td->td_ucred) { 608 crfree(td->td_ucred); 609 td->td_ucred = NULL; 610 } 611 612 /* 613 * Nobody actually wakes us when the lock 614 * count reaches zero, so just wait one tick. 615 */ 616 while (lp->lwp_lock > 0) 617 tsleep(lp, 0, "lwpexit", 1); 618 619 /* Hand down resource usage to our proc */ 620 ruadd(&p->p_ru, &lp->lwp_ru); 621 622 /* 623 * If we don't hold the process until the LWP is reaped wait*() 624 * may try to dispose of its vmspace before all the LWPs have 625 * actually terminated. 626 */ 627 PHOLD(p); 628 629 /* 630 * Do any remaining work that might block on us. We should be 631 * coded such that further blocking is ok after decrementing 632 * p_nthreads but don't take the chance. 633 */ 634 dsched_exit_thread(td); 635 biosched_done(curthread); 636 637 /* 638 * We have to use the reaper for all the LWPs except the one doing 639 * the master exit. The LWP doing the master exit can just be 640 * left on p_lwps and the process reaper will deal with it 641 * synchronously, which is much faster. 642 * 643 * Wakeup anyone waiting on p_nthreads to drop to 1 or 0. 644 * 645 * The process is left held until the reaper calls lwp_dispose() on 646 * the lp (after calling lwp_wait()). 647 */ 648 if (masterexit == 0) { 649 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 650 --p->p_nthreads; 651 if (p->p_nthreads <= 1) 652 dowake = 1; 653 lwkt_gettoken(&deadlwp_token); 654 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry); 655 taskqueue_enqueue(taskqueue_thread[mycpuid], 656 deadlwp_task[mycpuid]); 657 lwkt_reltoken(&deadlwp_token); 658 } else { 659 --p->p_nthreads; 660 if (p->p_nthreads <= 1) 661 dowake = 1; 662 } 663 664 /* 665 * Release p_token. Issue the wakeup() on p_nthreads if necessary, 666 * as late as possible to give us a chance to actually deschedule and 667 * switch away before another cpu core hits reaplwp(). 668 */ 669 lwkt_reltoken(&p->p_token); 670 if (dowake) 671 wakeup(&p->p_nthreads); 672 673 /* 674 * Tell the userland scheduler that we are going away 675 */ 676 p->p_usched->heuristic_exiting(lp, p); 677 678 cpu_lwp_exit(); 679 } 680 681 /* 682 * Wait until a lwp is completely dead. The final interlock in this drama 683 * is when TDF_EXITING is set in cpu_thread_exit() just before the final 684 * switchout. 685 * 686 * At the point TDF_EXITING is set a complete exit is accomplished when 687 * TDF_RUNNING and TDF_PREEMPT_LOCK are both clear. td_mpflags has two 688 * post-switch interlock flags that can be used to wait for the TDF_ 689 * flags to clear. 690 * 691 * Returns non-zero on success, and zero if the caller needs to retry 692 * the lwp_wait(). 693 */ 694 static int 695 lwp_wait(struct lwp *lp) 696 { 697 struct thread *td = lp->lwp_thread; 698 u_int mpflags; 699 700 KKASSERT(lwkt_preempted_proc() != lp); 701 702 /* 703 * This bit of code uses the thread destruction interlock 704 * managed by lwkt_switch_return() to wait for the lwp's 705 * thread to completely disengage. 706 * 707 * It is possible for us to race another cpu core so we 708 * have to do this correctly. 709 */ 710 for (;;) { 711 mpflags = td->td_mpflags; 712 cpu_ccfence(); 713 if (mpflags & TDF_MP_EXITSIG) 714 break; 715 tsleep_interlock(td, 0); 716 if (atomic_cmpset_int(&td->td_mpflags, mpflags, 717 mpflags | TDF_MP_EXITWAIT)) { 718 tsleep(td, PINTERLOCKED, "lwpxt", 0); 719 } 720 } 721 722 /* 723 * We've already waited for the core exit but there can still 724 * be other refs from e.g. process scans and such. 725 */ 726 if (lp->lwp_lock > 0) { 727 tsleep(lp, 0, "lwpwait1", 1); 728 return(0); 729 } 730 if (td->td_refs) { 731 tsleep(td, 0, "lwpwait2", 1); 732 return(0); 733 } 734 735 /* 736 * Now that we have the thread destruction interlock these flags 737 * really should already be cleaned up, keep a check for safety. 738 * 739 * We can't rip its stack out from under it until TDF_EXITING is 740 * set and both TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 741 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 742 * will be cleared temporarily if a thread gets preempted. 743 */ 744 while ((td->td_flags & (TDF_RUNNING | 745 TDF_PREEMPT_LOCK | 746 TDF_EXITING)) != TDF_EXITING) { 747 tsleep(lp, 0, "lwpwait3", 1); 748 return (0); 749 } 750 751 KASSERT((td->td_flags & (TDF_RUNQ|TDF_TSLEEPQ)) == 0, 752 ("lwp_wait: td %p (%s) still on run or sleep queue", 753 td, td->td_comm)); 754 return (1); 755 } 756 757 /* 758 * Release the resources associated with a lwp. 759 * The lwp must be completely dead. 760 */ 761 void 762 lwp_dispose(struct lwp *lp) 763 { 764 struct thread *td = lp->lwp_thread; 765 766 KKASSERT(lwkt_preempted_proc() != lp); 767 KKASSERT(td->td_refs == 0); 768 KKASSERT((td->td_flags & (TDF_RUNNING | 769 TDF_PREEMPT_LOCK | 770 TDF_EXITING)) == TDF_EXITING); 771 772 PRELE(lp->lwp_proc); 773 lp->lwp_proc = NULL; 774 if (td != NULL) { 775 td->td_proc = NULL; 776 td->td_lwp = NULL; 777 lp->lwp_thread = NULL; 778 lwkt_free_thread(td); 779 } 780 kfree(lp, M_LWP); 781 } 782 783 /* 784 * MPSAFE 785 */ 786 int 787 sys_wait4(struct wait_args *uap) 788 { 789 struct rusage rusage; 790 int error, status; 791 792 error = kern_wait(uap->pid, (uap->status ? &status : NULL), 793 uap->options, (uap->rusage ? &rusage : NULL), 794 &uap->sysmsg_result); 795 796 if (error == 0 && uap->status) 797 error = copyout(&status, uap->status, sizeof(*uap->status)); 798 if (error == 0 && uap->rusage) 799 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage)); 800 return (error); 801 } 802 803 /* 804 * wait1() 805 * 806 * wait_args(int pid, int *status, int options, struct rusage *rusage) 807 * 808 * MPALMOSTSAFE 809 */ 810 int 811 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res) 812 { 813 struct thread *td = curthread; 814 struct lwp *lp; 815 struct proc *q = td->td_proc; 816 struct proc *p, *t; 817 struct pargs *pa; 818 struct sigacts *ps; 819 int nfound, error; 820 821 if (pid == 0) 822 pid = -q->p_pgid; 823 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE)) 824 return (EINVAL); 825 826 lwkt_gettoken(&q->p_token); 827 loop: 828 /* 829 * All sorts of things can change due to blocking so we have to loop 830 * all the way back up here. 831 * 832 * The problem is that if a process group is stopped and the parent 833 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 834 * of the child and then stop itself when it tries to return from the 835 * system call. When the process group is resumed the parent will 836 * then get the STOP status even though the child has now resumed 837 * (a followup wait*() will get the CONT status). 838 * 839 * Previously the CONT would overwrite the STOP because the tstop 840 * was handled within tsleep(), and the parent would only see 841 * the CONT when both are stopped and continued together. This little 842 * two-line hack restores this effect. 843 */ 844 while (q->p_stat == SSTOP) 845 tstop(); 846 847 nfound = 0; 848 849 /* 850 * Loop on children. 851 * 852 * NOTE: We don't want to break q's p_token in the loop for the 853 * case where no children are found or we risk breaking the 854 * interlock between child and parent. 855 */ 856 LIST_FOREACH(p, &q->p_children, p_sibling) { 857 if (pid != WAIT_ANY && 858 p->p_pid != pid && p->p_pgid != -pid) { 859 continue; 860 } 861 862 /* 863 * This special case handles a kthread spawned by linux_clone 864 * (see linux_misc.c). The linux_wait4 and linux_waitpid 865 * functions need to be able to distinguish between waiting 866 * on a process and waiting on a thread. It is a thread if 867 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 868 * signifies we want to wait for threads and not processes. 869 */ 870 if ((p->p_sigparent != SIGCHLD) ^ 871 ((options & WLINUXCLONE) != 0)) { 872 continue; 873 } 874 875 nfound++; 876 if (p->p_stat == SZOMB) { 877 /* 878 * We may go into SZOMB with threads still present. 879 * We must wait for them to exit before we can reap 880 * the master thread, otherwise we may race reaping 881 * non-master threads. 882 * 883 * Only this routine can remove a process from 884 * the zombie list and destroy it, use PACQUIREZOMB() 885 * to serialize us and loop if it blocks (interlocked 886 * by the parent's q->p_token). 887 * 888 * WARNING! (p) can be invalid when PHOLDZOMB(p) 889 * returns non-zero. Be sure not to 890 * mess with it. 891 */ 892 if (PHOLDZOMB(p)) 893 goto loop; 894 lwkt_gettoken(&p->p_token); 895 if (p->p_pptr != q) { 896 lwkt_reltoken(&p->p_token); 897 PRELEZOMB(p); 898 goto loop; 899 } 900 while (p->p_nthreads > 0) { 901 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 902 } 903 904 /* 905 * Reap any LWPs left in p->p_lwps. This is usually 906 * just the last LWP. This must be done before 907 * we loop on p_lock since the lwps hold a ref on 908 * it as a vmspace interlock. 909 * 910 * Once that is accomplished p_nthreads had better 911 * be zero. 912 */ 913 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 914 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 915 reaplwp(lp); 916 } 917 KKASSERT(p->p_nthreads == 0); 918 919 /* 920 * Don't do anything really bad until all references 921 * to the process go away. This may include other 922 * LWPs which are still in the process of being 923 * reaped. We can't just pull the rug out from under 924 * them because they may still be using the VM space. 925 * 926 * Certain kernel facilities such as /proc will also 927 * put a hold on the process for short periods of 928 * time. 929 */ 930 PRELE(p); 931 PSTALL(p, "reap3", 0); 932 933 /* Take care of our return values. */ 934 *res = p->p_pid; 935 936 if (status) 937 *status = p->p_xstat; 938 if (rusage) 939 *rusage = p->p_ru; 940 /* 941 * If we got the child via a ptrace 'attach', 942 * we need to give it back to the old parent. 943 */ 944 if (p->p_oppid && (t = pfind(p->p_oppid)) != NULL) { 945 PHOLD(p); 946 p->p_oppid = 0; 947 proc_reparent(p, t); 948 ksignal(t, SIGCHLD); 949 wakeup((caddr_t)t); 950 error = 0; 951 PRELE(t); 952 lwkt_reltoken(&p->p_token); 953 PRELEZOMB(p); 954 goto done; 955 } 956 957 /* 958 * Unlink the proc from its process group so that 959 * the following operations won't lead to an 960 * inconsistent state for processes running down 961 * the zombie list. 962 */ 963 proc_remove_zombie(p); 964 lwkt_reltoken(&p->p_token); 965 leavepgrp(p); 966 967 p->p_xstat = 0; 968 ruadd(&q->p_cru, &p->p_ru); 969 970 /* 971 * Decrement the count of procs running with this uid. 972 */ 973 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 974 975 /* 976 * Free up credentials. 977 */ 978 crfree(p->p_ucred); 979 p->p_ucred = NULL; 980 981 /* 982 * Remove unused arguments 983 */ 984 pa = p->p_args; 985 p->p_args = NULL; 986 if (pa && refcount_release(&pa->ar_ref)) { 987 kfree(pa, M_PARGS); 988 pa = NULL; 989 } 990 991 ps = p->p_sigacts; 992 p->p_sigacts = NULL; 993 if (ps && refcount_release(&ps->ps_refcnt)) { 994 kfree(ps, M_SUBPROC); 995 ps = NULL; 996 } 997 998 /* 999 * Our exitingcount was incremented when the process 1000 * became a zombie, now that the process has been 1001 * removed from (almost) all lists we should be able 1002 * to safely destroy its vmspace. Wait for any current 1003 * holders to go away (so the vmspace remains stable), 1004 * then scrap it. 1005 */ 1006 PSTALL(p, "reap4", 0); 1007 vmspace_exitfree(p); 1008 PSTALL(p, "reap5", 0); 1009 1010 /* 1011 * NOTE: We have to officially release ZOMB in order 1012 * to ensure that a racing thread in kern_wait() 1013 * which blocked on ZOMB is woken up. 1014 */ 1015 PHOLD(p); 1016 PRELEZOMB(p); 1017 kfree(p, M_PROC); 1018 atomic_add_int(&nprocs, -1); 1019 error = 0; 1020 goto done; 1021 } 1022 if (p->p_stat == SSTOP && (p->p_flags & P_WAITED) == 0 && 1023 ((p->p_flags & P_TRACED) || (options & WUNTRACED))) { 1024 PHOLD(p); 1025 lwkt_gettoken(&p->p_token); 1026 if (p->p_pptr != q) { 1027 lwkt_reltoken(&p->p_token); 1028 PRELE(p); 1029 goto loop; 1030 } 1031 if (p->p_stat != SSTOP || 1032 (p->p_flags & P_WAITED) != 0 || 1033 ((p->p_flags & P_TRACED) == 0 && 1034 (options & WUNTRACED) == 0)) { 1035 lwkt_reltoken(&p->p_token); 1036 PRELE(p); 1037 goto loop; 1038 } 1039 1040 p->p_flags |= P_WAITED; 1041 1042 *res = p->p_pid; 1043 if (status) 1044 *status = W_STOPCODE(p->p_xstat); 1045 /* Zero rusage so we get something consistent. */ 1046 if (rusage) 1047 bzero(rusage, sizeof(*rusage)); 1048 error = 0; 1049 lwkt_reltoken(&p->p_token); 1050 PRELE(p); 1051 goto done; 1052 } 1053 if ((options & WCONTINUED) && (p->p_flags & P_CONTINUED)) { 1054 PHOLD(p); 1055 lwkt_gettoken(&p->p_token); 1056 if (p->p_pptr != q) { 1057 lwkt_reltoken(&p->p_token); 1058 PRELE(p); 1059 goto loop; 1060 } 1061 if ((p->p_flags & P_CONTINUED) == 0) { 1062 lwkt_reltoken(&p->p_token); 1063 PRELE(p); 1064 goto loop; 1065 } 1066 1067 *res = p->p_pid; 1068 p->p_flags &= ~P_CONTINUED; 1069 1070 if (status) 1071 *status = SIGCONT; 1072 error = 0; 1073 lwkt_reltoken(&p->p_token); 1074 PRELE(p); 1075 goto done; 1076 } 1077 } 1078 if (nfound == 0) { 1079 error = ECHILD; 1080 goto done; 1081 } 1082 if (options & WNOHANG) { 1083 *res = 0; 1084 error = 0; 1085 goto done; 1086 } 1087 1088 /* 1089 * Wait for signal - interlocked using q->p_token. 1090 */ 1091 error = tsleep(q, PCATCH, "wait", 0); 1092 if (error) { 1093 done: 1094 lwkt_reltoken(&q->p_token); 1095 return (error); 1096 } 1097 goto loop; 1098 } 1099 1100 /* 1101 * Change child's parent process to parent. 1102 * 1103 * p_children/p_sibling requires the parent's token, and 1104 * changing pptr requires the child's token, so we have to 1105 * get three tokens to do this operation. We also need to 1106 * hold pointers that might get ripped out from under us to 1107 * preserve structural integrity. 1108 * 1109 * It is possible to race another reparent or disconnect or other 1110 * similar operation. We must retry when this situation occurs. 1111 * Once we successfully reparent the process we no longer care 1112 * about any races. 1113 */ 1114 void 1115 proc_reparent(struct proc *child, struct proc *parent) 1116 { 1117 struct proc *opp; 1118 1119 PHOLD(parent); 1120 while ((opp = child->p_pptr) != parent) { 1121 PHOLD(opp); 1122 lwkt_gettoken(&opp->p_token); 1123 lwkt_gettoken(&child->p_token); 1124 lwkt_gettoken(&parent->p_token); 1125 if (child->p_pptr != opp) { 1126 lwkt_reltoken(&parent->p_token); 1127 lwkt_reltoken(&child->p_token); 1128 lwkt_reltoken(&opp->p_token); 1129 PRELE(opp); 1130 continue; 1131 } 1132 LIST_REMOVE(child, p_sibling); 1133 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1134 child->p_pptr = parent; 1135 lwkt_reltoken(&parent->p_token); 1136 lwkt_reltoken(&child->p_token); 1137 lwkt_reltoken(&opp->p_token); 1138 if (LIST_EMPTY(&opp->p_children)) 1139 wakeup(opp); 1140 PRELE(opp); 1141 break; 1142 } 1143 PRELE(parent); 1144 } 1145 1146 /* 1147 * The next two functions are to handle adding/deleting items on the 1148 * exit callout list 1149 * 1150 * at_exit(): 1151 * Take the arguments given and put them onto the exit callout list, 1152 * However first make sure that it's not already there. 1153 * returns 0 on success. 1154 */ 1155 1156 int 1157 at_exit(exitlist_fn function) 1158 { 1159 struct exitlist *ep; 1160 1161 #ifdef INVARIANTS 1162 /* Be noisy if the programmer has lost track of things */ 1163 if (rm_at_exit(function)) 1164 kprintf("WARNING: exit callout entry (%p) already present\n", 1165 function); 1166 #endif 1167 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 1168 if (ep == NULL) 1169 return (ENOMEM); 1170 ep->function = function; 1171 TAILQ_INSERT_TAIL(&exit_list, ep, next); 1172 return (0); 1173 } 1174 1175 /* 1176 * Scan the exit callout list for the given item and remove it. 1177 * Returns the number of items removed (0 or 1) 1178 */ 1179 int 1180 rm_at_exit(exitlist_fn function) 1181 { 1182 struct exitlist *ep; 1183 1184 TAILQ_FOREACH(ep, &exit_list, next) { 1185 if (ep->function == function) { 1186 TAILQ_REMOVE(&exit_list, ep, next); 1187 kfree(ep, M_ATEXIT); 1188 return(1); 1189 } 1190 } 1191 return (0); 1192 } 1193 1194 /* 1195 * LWP reaper related code. 1196 */ 1197 static void 1198 reaplwps(void *context, int dummy) 1199 { 1200 struct lwplist *lwplist = context; 1201 struct lwp *lp; 1202 1203 lwkt_gettoken(&deadlwp_token); 1204 while ((lp = LIST_FIRST(lwplist))) { 1205 LIST_REMOVE(lp, u.lwp_reap_entry); 1206 reaplwp(lp); 1207 } 1208 lwkt_reltoken(&deadlwp_token); 1209 } 1210 1211 static void 1212 reaplwp(struct lwp *lp) 1213 { 1214 while (lwp_wait(lp) == 0) 1215 ; 1216 lwp_dispose(lp); 1217 } 1218 1219 static void 1220 deadlwp_init(void) 1221 { 1222 int cpu; 1223 1224 for (cpu = 0; cpu < ncpus; cpu++) { 1225 LIST_INIT(&deadlwp_list[cpu]); 1226 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), 1227 M_DEVBUF, M_WAITOK); 1228 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1229 } 1230 } 1231 1232 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1233