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/aio.h> 66 #include <sys/jail.h> 67 #include <sys/kern_syscall.h> 68 #include <sys/upcall.h> 69 #include <sys/caps.h> 70 #include <sys/unistd.h> 71 72 #include <vm/vm.h> 73 #include <vm/vm_param.h> 74 #include <sys/lock.h> 75 #include <vm/pmap.h> 76 #include <vm/vm_map.h> 77 #include <vm/vm_extern.h> 78 #include <sys/user.h> 79 80 #include <sys/thread2.h> 81 #include <sys/sysref2.h> 82 #include <sys/mplock2.h> 83 84 static void reaplwps(void *context, int dummy); 85 static void reaplwp(struct lwp *lp); 86 static void killlwps(struct lwp *lp); 87 88 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 89 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status"); 90 91 /* 92 * callout list for things to do at exit time 93 */ 94 struct exitlist { 95 exitlist_fn function; 96 TAILQ_ENTRY(exitlist) next; 97 }; 98 99 TAILQ_HEAD(exit_list_head, exitlist); 100 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 101 102 /* 103 * LWP reaper data 104 */ 105 struct task *deadlwp_task[MAXCPU]; 106 struct lwplist deadlwp_list[MAXCPU]; 107 108 /* 109 * exit -- 110 * Death of process. 111 * 112 * SYS_EXIT_ARGS(int rval) 113 * 114 * MPALMOSTSAFE 115 */ 116 int 117 sys_exit(struct exit_args *uap) 118 { 119 get_mplock(); 120 exit1(W_EXITCODE(uap->rval, 0)); 121 /* NOTREACHED */ 122 rel_mplock(); 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 int action, who; 135 int error; 136 137 action = EXTEXIT_ACTION(uap->how); 138 who = EXTEXIT_WHO(uap->how); 139 140 /* Check parameters before we might perform some action */ 141 switch (who) { 142 case EXTEXIT_PROC: 143 case EXTEXIT_LWP: 144 break; 145 default: 146 return (EINVAL); 147 } 148 149 switch (action) { 150 case EXTEXIT_SIMPLE: 151 break; 152 case EXTEXIT_SETINT: 153 error = copyout(&uap->status, uap->addr, sizeof(uap->status)); 154 if (error) 155 return (error); 156 break; 157 default: 158 return (EINVAL); 159 } 160 161 get_mplock(); 162 163 switch (who) { 164 case EXTEXIT_LWP: 165 /* 166 * Be sure only to perform a simple lwp exit if there is at 167 * least one more lwp in the proc, which will call exit1() 168 * later, otherwise the proc will be an UNDEAD and not even a 169 * SZOMB! 170 */ 171 if (curproc->p_nthreads > 1) { 172 lwp_exit(0); 173 /* NOT REACHED */ 174 } 175 /* else last lwp in proc: do the real thing */ 176 /* FALLTHROUGH */ 177 default: /* to help gcc */ 178 case EXTEXIT_PROC: 179 exit1(W_EXITCODE(uap->status, 0)); 180 /* NOTREACHED */ 181 } 182 183 /* NOTREACHED */ 184 rel_mplock(); /* safety */ 185 } 186 187 /* 188 * Kill all lwps associated with the current process except the 189 * current lwp. Return an error if we race another thread trying to 190 * do the same thing and lose the race. 191 * 192 * If forexec is non-zero the current thread and process flags are 193 * cleaned up so they can be reused. 194 */ 195 int 196 killalllwps(int forexec) 197 { 198 struct lwp *lp = curthread->td_lwp; 199 struct proc *p = lp->lwp_proc; 200 201 /* 202 * Interlock against P_WEXIT. Only one of the process's thread 203 * is allowed to do the master exit. 204 */ 205 if (p->p_flag & P_WEXIT) 206 return (EALREADY); 207 p->p_flag |= P_WEXIT; 208 209 /* 210 * Interlock with LWP_WEXIT and kill any remaining LWPs 211 */ 212 lp->lwp_flag |= LWP_WEXIT; 213 if (p->p_nthreads > 1) 214 killlwps(lp); 215 216 /* 217 * If doing this for an exec, clean up the remaining thread 218 * (us) for continuing operation after all the other threads 219 * have been killed. 220 */ 221 if (forexec) { 222 lp->lwp_flag &= ~LWP_WEXIT; 223 p->p_flag &= ~P_WEXIT; 224 } 225 return(0); 226 } 227 228 /* 229 * Kill all LWPs except the current one. Do not try to signal 230 * LWPs which have exited on their own or have already been 231 * signaled. 232 */ 233 static void 234 killlwps(struct lwp *lp) 235 { 236 struct proc *p = lp->lwp_proc; 237 struct lwp *tlp; 238 239 /* 240 * Kill the remaining LWPs. We must send the signal before setting 241 * LWP_WEXIT. The setting of WEXIT is optional but helps reduce 242 * races. tlp must be held across the call as it might block and 243 * allow the target lwp to rip itself out from under our loop. 244 */ 245 FOREACH_LWP_IN_PROC(tlp, p) { 246 LWPHOLD(tlp); 247 if ((tlp->lwp_flag & LWP_WEXIT) == 0) { 248 lwpsignal(p, tlp, SIGKILL); 249 tlp->lwp_flag |= LWP_WEXIT; 250 } 251 LWPRELE(tlp); 252 } 253 254 /* 255 * Wait for everything to clear out. 256 */ 257 while (p->p_nthreads > 1) { 258 tsleep(&p->p_nthreads, 0, "killlwps", 0); 259 } 260 } 261 262 /* 263 * Exit: deallocate address space and other resources, change proc state 264 * to zombie, and unlink proc from allproc and parent's lists. Save exit 265 * status and rusage for wait(). Check for child processes and orphan them. 266 */ 267 void 268 exit1(int rv) 269 { 270 struct thread *td = curthread; 271 struct proc *p = td->td_proc; 272 struct lwp *lp = td->td_lwp; 273 struct proc *q, *nq; 274 struct vmspace *vm; 275 struct vnode *vtmp; 276 struct exitlist *ep; 277 int error; 278 279 if (p->p_pid == 1) { 280 kprintf("init died (signal %d, exit %d)\n", 281 WTERMSIG(rv), WEXITSTATUS(rv)); 282 panic("Going nowhere without my init!"); 283 } 284 285 varsymset_clean(&p->p_varsymset); 286 lockuninit(&p->p_varsymset.vx_lock); 287 /* 288 * Kill all lwps associated with the current process, return an 289 * error if we race another thread trying to do the same thing 290 * and lose the race. 291 */ 292 error = killalllwps(0); 293 if (error) { 294 lwp_exit(0); 295 /* NOT REACHED */ 296 } 297 298 caps_exit(lp->lwp_thread); 299 aio_proc_rundown(p); 300 301 /* are we a task leader? */ 302 if (p == p->p_leader) { 303 struct kill_args killArgs; 304 killArgs.signum = SIGKILL; 305 q = p->p_peers; 306 while(q) { 307 killArgs.pid = q->p_pid; 308 /* 309 * The interface for kill is better 310 * than the internal signal 311 */ 312 sys_kill(&killArgs); 313 nq = q; 314 q = q->p_peers; 315 } 316 while (p->p_peers) 317 tsleep((caddr_t)p, 0, "exit1", 0); 318 } 319 320 #ifdef PGINPROF 321 vmsizmon(); 322 #endif 323 STOPEVENT(p, S_EXIT, rv); 324 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */ 325 326 /* 327 * Check if any loadable modules need anything done at process exit. 328 * e.g. SYSV IPC stuff 329 * XXX what if one of these generates an error? 330 */ 331 TAILQ_FOREACH(ep, &exit_list, next) 332 (*ep->function)(td); 333 334 if (p->p_flag & P_PROFIL) 335 stopprofclock(p); 336 /* 337 * If parent is waiting for us to exit or exec, 338 * P_PPWAIT is set; we will wakeup the parent below. 339 */ 340 p->p_flag &= ~(P_TRACED | P_PPWAIT); 341 SIGEMPTYSET(p->p_siglist); 342 SIGEMPTYSET(lp->lwp_siglist); 343 if (timevalisset(&p->p_realtimer.it_value)) 344 callout_stop(&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 * Release upcalls associated with this process 378 */ 379 if (vm->vm_upcalls) 380 upc_release(vm, lp); 381 382 /* 383 * Clean up data related to virtual kernel operation. Clean up 384 * any vkernel context related to the current lwp now so we can 385 * destroy p_vkernel. 386 */ 387 if (p->p_vkernel) { 388 vkernel_lwp_exit(lp); 389 vkernel_exit(p); 390 } 391 392 /* 393 * Release user portion of address space. 394 * This releases references to vnodes, 395 * which could cause I/O if the file has been unlinked. 396 * Need to do this early enough that we can still sleep. 397 * Can't free the entire vmspace as the kernel stack 398 * may be mapped within that space also. 399 * 400 * Processes sharing the same vmspace may exit in one order, and 401 * get cleaned up by vmspace_exit() in a different order. The 402 * last exiting process to reach this point releases as much of 403 * the environment as it can, and the last process cleaned up 404 * by vmspace_exit() (which decrements exitingcnt) cleans up the 405 * remainder. 406 */ 407 ++vm->vm_exitingcnt; 408 sysref_put(&vm->vm_sysref); 409 410 if (SESS_LEADER(p)) { 411 struct session *sp = p->p_session; 412 413 if (sp->s_ttyvp) { 414 /* 415 * We are the controlling process. Signal the 416 * foreground process group, drain the controlling 417 * terminal, and revoke access to the controlling 418 * terminal. 419 * 420 * NOTE: while waiting for the process group to exit 421 * it is possible that one of the processes in the 422 * group will revoke the tty, so the ttyclosesession() 423 * function will re-check sp->s_ttyvp. 424 */ 425 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 426 if (sp->s_ttyp->t_pgrp) 427 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 428 ttywait(sp->s_ttyp); 429 ttyclosesession(sp, 1); /* also revoke */ 430 } 431 /* 432 * Release the tty. If someone has it open via 433 * /dev/tty then close it (since they no longer can 434 * once we've NULL'd it out). 435 */ 436 ttyclosesession(sp, 0); 437 438 /* 439 * s_ttyp is not zero'd; we use this to indicate 440 * that the session once had a controlling terminal. 441 * (for logging and informational purposes) 442 */ 443 } 444 sp->s_leader = NULL; 445 } 446 fixjobc(p, p->p_pgrp, 0); 447 (void)acct_process(p); 448 #ifdef KTRACE 449 /* 450 * release trace file 451 */ 452 if (p->p_tracenode) 453 ktrdestroy(&p->p_tracenode); 454 p->p_traceflag = 0; 455 #endif 456 /* 457 * Release reference to text vnode 458 */ 459 if ((vtmp = p->p_textvp) != NULL) { 460 p->p_textvp = NULL; 461 vrele(vtmp); 462 } 463 464 /* 465 * Move the process to the zombie list. This will block 466 * until the process p_lock count reaches 0. The process will 467 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 468 * which is called from cpu_proc_exit(). 469 */ 470 proc_move_allproc_zombie(p); 471 472 q = LIST_FIRST(&p->p_children); 473 if (q) /* only need this if any child is S_ZOMB */ 474 wakeup((caddr_t) initproc); 475 for (; q != 0; q = nq) { 476 nq = LIST_NEXT(q, p_sibling); 477 LIST_REMOVE(q, p_sibling); 478 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling); 479 q->p_pptr = initproc; 480 q->p_sigparent = SIGCHLD; 481 /* 482 * Traced processes are killed 483 * since their existence means someone is screwing up. 484 */ 485 if (q->p_flag & P_TRACED) { 486 q->p_flag &= ~P_TRACED; 487 ksignal(q, SIGKILL); 488 } 489 } 490 491 /* 492 * Save exit status and final rusage info, adding in child rusage 493 * info and self times. 494 */ 495 p->p_xstat = rv; 496 calcru_proc(p, &p->p_ru); 497 ruadd(&p->p_ru, &p->p_cru); 498 499 /* 500 * notify interested parties of our demise. 501 */ 502 KNOTE(&p->p_klist, NOTE_EXIT); 503 504 /* 505 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 506 * flag set, notify process 1 instead (and hope it will handle 507 * this situation). 508 */ 509 if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) { 510 struct proc *pp = p->p_pptr; 511 proc_reparent(p, initproc); 512 /* 513 * If this was the last child of our parent, notify 514 * parent, so in case he was wait(2)ing, he will 515 * continue. 516 */ 517 if (LIST_EMPTY(&pp->p_children)) 518 wakeup((caddr_t)pp); 519 } 520 521 if (p->p_sigparent && p->p_pptr != initproc) { 522 ksignal(p->p_pptr, p->p_sigparent); 523 } else { 524 ksignal(p->p_pptr, SIGCHLD); 525 } 526 527 wakeup((caddr_t)p->p_pptr); 528 /* 529 * cpu_exit is responsible for clearing curproc, since 530 * it is heavily integrated with the thread/switching sequence. 531 * 532 * Other substructures are freed from wait(). 533 */ 534 plimit_free(p); 535 536 /* 537 * Release the current user process designation on the process so 538 * the userland scheduler can work in someone else. 539 */ 540 p->p_usched->release_curproc(lp); 541 542 /* 543 * Finally, call machine-dependent code to release as many of the 544 * lwp's resources as we can and halt execution of this thread. 545 */ 546 lwp_exit(1); 547 } 548 549 /* 550 * Eventually called by every exiting LWP 551 */ 552 void 553 lwp_exit(int masterexit) 554 { 555 struct thread *td = curthread; 556 struct lwp *lp = td->td_lwp; 557 struct proc *p = lp->lwp_proc; 558 559 /* 560 * lwp_exit() may be called without setting LWP_WEXIT, so 561 * make sure it is set here. 562 */ 563 lp->lwp_flag |= LWP_WEXIT; 564 565 /* 566 * Clean up any virtualization 567 */ 568 if (lp->lwp_vkernel) 569 vkernel_lwp_exit(lp); 570 571 /* 572 * Clean up select/poll support 573 */ 574 kqueue_terminate(&lp->lwp_kqueue); 575 576 /* 577 * Clean up any syscall-cached ucred 578 */ 579 if (td->td_ucred) { 580 crfree(td->td_ucred); 581 td->td_ucred = NULL; 582 } 583 584 /* 585 * Nobody actually wakes us when the lock 586 * count reaches zero, so just wait one tick. 587 */ 588 while (lp->lwp_lock > 0) 589 tsleep(lp, 0, "lwpexit", 1); 590 591 /* Hand down resource usage to our proc */ 592 ruadd(&p->p_ru, &lp->lwp_ru); 593 594 /* 595 * If we don't hold the process until the LWP is reaped wait*() 596 * may try to dispose of its vmspace before all the LWPs have 597 * actually terminated. 598 */ 599 PHOLD(p); 600 601 /* 602 * We have to use the reaper for all the LWPs except the one doing 603 * the master exit. The LWP doing the master exit can just be 604 * left on p_lwps and the process reaper will deal with it 605 * synchronously, which is much faster. 606 */ 607 if (masterexit == 0) { 608 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 609 --p->p_nthreads; 610 wakeup(&p->p_nthreads); 611 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry); 612 taskqueue_enqueue(taskqueue_thread[mycpuid], deadlwp_task[mycpuid]); 613 } else { 614 --p->p_nthreads; 615 } 616 biosched_done(curthread); 617 cpu_lwp_exit(); 618 } 619 620 /* 621 * Wait until a lwp is completely dead. 622 * 623 * If the thread is still executing, which can't be waited upon, 624 * return failure. The caller is responsible of waiting a little 625 * bit and checking again. 626 * 627 * Suggested use: 628 * while (!lwp_wait(lp)) 629 * tsleep(lp, 0, "lwpwait", 1); 630 */ 631 static int 632 lwp_wait(struct lwp *lp) 633 { 634 struct thread *td = lp->lwp_thread;; 635 636 KKASSERT(lwkt_preempted_proc() != lp); 637 638 while (lp->lwp_lock > 0) 639 tsleep(lp, 0, "lwpwait1", 1); 640 641 lwkt_wait_free(td); 642 643 /* 644 * The lwp's thread may still be in the middle 645 * of switching away, we can't rip its stack out from 646 * under it until TDF_EXITING is set and both 647 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 648 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 649 * will be cleared temporarily if a thread gets 650 * preempted. 651 * 652 * YYY no wakeup occurs, so we simply return failure 653 * and let the caller deal with sleeping and calling 654 * us again. 655 */ 656 if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) != 657 TDF_EXITING) 658 return (0); 659 660 return (1); 661 } 662 663 /* 664 * Release the resources associated with a lwp. 665 * The lwp must be completely dead. 666 */ 667 void 668 lwp_dispose(struct lwp *lp) 669 { 670 struct thread *td = lp->lwp_thread;; 671 672 KKASSERT(lwkt_preempted_proc() != lp); 673 KKASSERT(td->td_refs == 0); 674 KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) == 675 TDF_EXITING); 676 677 PRELE(lp->lwp_proc); 678 lp->lwp_proc = NULL; 679 if (td != NULL) { 680 td->td_proc = NULL; 681 td->td_lwp = NULL; 682 lp->lwp_thread = NULL; 683 lwkt_free_thread(td); 684 } 685 kfree(lp, M_LWP); 686 } 687 688 /* 689 * MPSAFE 690 */ 691 int 692 sys_wait4(struct wait_args *uap) 693 { 694 struct rusage rusage; 695 int error, status; 696 697 error = kern_wait(uap->pid, (uap->status ? &status : NULL), 698 uap->options, (uap->rusage ? &rusage : NULL), 699 &uap->sysmsg_result); 700 701 if (error == 0 && uap->status) 702 error = copyout(&status, uap->status, sizeof(*uap->status)); 703 if (error == 0 && uap->rusage) 704 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage)); 705 return (error); 706 } 707 708 /* 709 * wait1() 710 * 711 * wait_args(int pid, int *status, int options, struct rusage *rusage) 712 * 713 * MPALMOSTSAFE 714 */ 715 int 716 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res) 717 { 718 struct thread *td = curthread; 719 struct lwp *lp; 720 struct proc *q = td->td_proc; 721 struct proc *p, *t; 722 int nfound, error; 723 724 if (pid == 0) 725 pid = -q->p_pgid; 726 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE)) 727 return (EINVAL); 728 get_mplock(); 729 loop: 730 /* 731 * Hack for backwards compatibility with badly written user code. 732 * Or perhaps we have to do this anyway, it is unclear. XXX 733 * 734 * The problem is that if a process group is stopped and the parent 735 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 736 * of the child and then stop itself when it tries to return from the 737 * system call. When the process group is resumed the parent will 738 * then get the STOP status even though the child has now resumed 739 * (a followup wait*() will get the CONT status). 740 * 741 * Previously the CONT would overwrite the STOP because the tstop 742 * was handled within tsleep(), and the parent would only see 743 * the CONT when both are stopped and continued together. This litte 744 * two-line hack restores this effect. 745 */ 746 while (q->p_stat == SSTOP) 747 tstop(); 748 749 nfound = 0; 750 LIST_FOREACH(p, &q->p_children, p_sibling) { 751 if (pid != WAIT_ANY && 752 p->p_pid != pid && p->p_pgid != -pid) 753 continue; 754 755 /* This special case handles a kthread spawned by linux_clone 756 * (see linux_misc.c). The linux_wait4 and linux_waitpid 757 * functions need to be able to distinguish between waiting 758 * on a process and waiting on a thread. It is a thread if 759 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 760 * signifies we want to wait for threads and not processes. 761 */ 762 if ((p->p_sigparent != SIGCHLD) ^ 763 ((options & WLINUXCLONE) != 0)) { 764 continue; 765 } 766 767 nfound++; 768 if (p->p_stat == SZOMB) { 769 /* 770 * We may go into SZOMB with threads still present. 771 * We must wait for them to exit before we can reap 772 * the master thread, otherwise we may race reaping 773 * non-master threads. 774 */ 775 while (p->p_nthreads > 0) { 776 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 777 } 778 779 /* 780 * Reap any LWPs left in p->p_lwps. This is usually 781 * just the last LWP. This must be done before 782 * we loop on p_lock since the lwps hold a ref on 783 * it as a vmspace interlock. 784 * 785 * Once that is accomplished p_nthreads had better 786 * be zero. 787 */ 788 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 789 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 790 reaplwp(lp); 791 } 792 KKASSERT(p->p_nthreads == 0); 793 794 /* 795 * Don't do anything really bad until all references 796 * to the process go away. This may include other 797 * LWPs which are still in the process of being 798 * reaped. We can't just pull the rug out from under 799 * them because they may still be using the VM space. 800 * 801 * Certain kernel facilities such as /proc will also 802 * put a hold on the process for short periods of 803 * time. 804 */ 805 while (p->p_lock) 806 tsleep(p, 0, "reap3", hz); 807 808 /* scheduling hook for heuristic */ 809 /* XXX no lwp available, we need a different heuristic */ 810 /* 811 p->p_usched->heuristic_exiting(td->td_lwp, deadlp); 812 */ 813 814 /* Take care of our return values. */ 815 *res = p->p_pid; 816 if (status) 817 *status = p->p_xstat; 818 if (rusage) 819 *rusage = p->p_ru; 820 /* 821 * If we got the child via a ptrace 'attach', 822 * we need to give it back to the old parent. 823 */ 824 if (p->p_oppid && (t = pfind(p->p_oppid))) { 825 p->p_oppid = 0; 826 proc_reparent(p, t); 827 ksignal(t, SIGCHLD); 828 wakeup((caddr_t)t); 829 error = 0; 830 goto done; 831 } 832 833 /* 834 * Unlink the proc from its process group so that 835 * the following operations won't lead to an 836 * inconsistent state for processes running down 837 * the zombie list. 838 */ 839 KKASSERT(p->p_lock == 0); 840 proc_remove_zombie(p); 841 leavepgrp(p); 842 843 p->p_xstat = 0; 844 ruadd(&q->p_cru, &p->p_ru); 845 846 /* 847 * Decrement the count of procs running with this uid. 848 */ 849 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 850 851 /* 852 * Free up credentials. 853 */ 854 crfree(p->p_ucred); 855 p->p_ucred = NULL; 856 857 /* 858 * Remove unused arguments 859 */ 860 if (p->p_args && --p->p_args->ar_ref == 0) 861 FREE(p->p_args, M_PARGS); 862 863 if (--p->p_sigacts->ps_refcnt == 0) { 864 kfree(p->p_sigacts, M_SUBPROC); 865 p->p_sigacts = NULL; 866 } 867 868 vm_waitproc(p); 869 kfree(p, M_PROC); 870 nprocs--; 871 error = 0; 872 goto done; 873 } 874 if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 && 875 (p->p_flag & P_TRACED || options & WUNTRACED)) { 876 p->p_flag |= P_WAITED; 877 878 *res = p->p_pid; 879 if (status) 880 *status = W_STOPCODE(p->p_xstat); 881 /* Zero rusage so we get something consistent. */ 882 if (rusage) 883 bzero(rusage, sizeof(rusage)); 884 error = 0; 885 goto done; 886 } 887 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) { 888 *res = p->p_pid; 889 p->p_flag &= ~P_CONTINUED; 890 891 if (status) 892 *status = SIGCONT; 893 error = 0; 894 goto done; 895 } 896 } 897 if (nfound == 0) { 898 error = ECHILD; 899 goto done; 900 } 901 if (options & WNOHANG) { 902 *res = 0; 903 error = 0; 904 goto done; 905 } 906 error = tsleep((caddr_t)q, PCATCH, "wait", 0); 907 if (error) { 908 done: 909 rel_mplock(); 910 return (error); 911 } 912 goto loop; 913 } 914 915 /* 916 * make process 'parent' the new parent of process 'child'. 917 */ 918 void 919 proc_reparent(struct proc *child, struct proc *parent) 920 { 921 922 if (child->p_pptr == parent) 923 return; 924 925 LIST_REMOVE(child, p_sibling); 926 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 927 child->p_pptr = parent; 928 } 929 930 /* 931 * The next two functions are to handle adding/deleting items on the 932 * exit callout list 933 * 934 * at_exit(): 935 * Take the arguments given and put them onto the exit callout list, 936 * However first make sure that it's not already there. 937 * returns 0 on success. 938 */ 939 940 int 941 at_exit(exitlist_fn function) 942 { 943 struct exitlist *ep; 944 945 #ifdef INVARIANTS 946 /* Be noisy if the programmer has lost track of things */ 947 if (rm_at_exit(function)) 948 kprintf("WARNING: exit callout entry (%p) already present\n", 949 function); 950 #endif 951 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 952 if (ep == NULL) 953 return (ENOMEM); 954 ep->function = function; 955 TAILQ_INSERT_TAIL(&exit_list, ep, next); 956 return (0); 957 } 958 959 /* 960 * Scan the exit callout list for the given item and remove it. 961 * Returns the number of items removed (0 or 1) 962 */ 963 int 964 rm_at_exit(exitlist_fn function) 965 { 966 struct exitlist *ep; 967 968 TAILQ_FOREACH(ep, &exit_list, next) { 969 if (ep->function == function) { 970 TAILQ_REMOVE(&exit_list, ep, next); 971 kfree(ep, M_ATEXIT); 972 return(1); 973 } 974 } 975 return (0); 976 } 977 978 /* 979 * LWP reaper related code. 980 */ 981 static void 982 reaplwps(void *context, int dummy) 983 { 984 struct lwplist *lwplist = context; 985 struct lwp *lp; 986 987 get_mplock(); 988 while ((lp = LIST_FIRST(lwplist))) { 989 LIST_REMOVE(lp, u.lwp_reap_entry); 990 reaplwp(lp); 991 } 992 rel_mplock(); 993 } 994 995 static void 996 reaplwp(struct lwp *lp) 997 { 998 while (lwp_wait(lp) == 0) 999 tsleep(lp, 0, "lwpreap", 1); 1000 lwp_dispose(lp); 1001 } 1002 1003 static void 1004 deadlwp_init(void) 1005 { 1006 int cpu; 1007 1008 for (cpu = 0; cpu < ncpus; cpu++) { 1009 LIST_INIT(&deadlwp_list[cpu]); 1010 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK); 1011 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 1012 } 1013 } 1014 1015 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 1016