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 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 /* 91 * callout list for things to do at exit time 92 */ 93 struct exitlist { 94 exitlist_fn function; 95 TAILQ_ENTRY(exitlist) next; 96 }; 97 98 TAILQ_HEAD(exit_list_head, exitlist); 99 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 100 101 /* 102 * LWP reaper data 103 */ 104 struct task *deadlwp_task[MAXCPU]; 105 struct lwplist deadlwp_list[MAXCPU]; 106 107 /* 108 * exit -- 109 * Death of process. 110 * 111 * SYS_EXIT_ARGS(int rval) 112 */ 113 int 114 sys_exit(struct exit_args *uap) 115 { 116 exit1(W_EXITCODE(uap->rval, 0)); 117 /* NOTREACHED */ 118 } 119 120 /* 121 * Extended exit -- 122 * Death of a lwp or process with optional bells and whistles. 123 */ 124 int 125 sys_extexit(struct extexit_args *uap) 126 { 127 int action, who; 128 int error; 129 130 action = EXTEXIT_ACTION(uap->how); 131 who = EXTEXIT_WHO(uap->how); 132 133 /* Check parameters before we might perform some action */ 134 switch (who) { 135 case EXTEXIT_PROC: 136 case EXTEXIT_LWP: 137 break; 138 139 default: 140 return (EINVAL); 141 } 142 143 switch (action) { 144 case EXTEXIT_SIMPLE: 145 break; 146 147 case EXTEXIT_SETINT: 148 error = copyout(&uap->status, uap->addr, sizeof(uap->status)); 149 if (error) 150 return (error); 151 break; 152 153 default: 154 return (EINVAL); 155 } 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 (curproc->p_nthreads > 1) { 166 lwp_exit(0); 167 /* NOT REACHED */ 168 } 169 /* else last lwp in proc: do the real thing */ 170 /* FALLTHROUGH */ 171 172 default: /* to help gcc */ 173 case EXTEXIT_PROC: 174 exit1(W_EXITCODE(uap->status, 0)); 175 /* NOTREACHED */ 176 } 177 178 /* NOTREACHED */ 179 } 180 181 /* 182 * Kill all lwps associated with the current process except the 183 * current lwp. Return an error if we race another thread trying to 184 * do the same thing and lose the race. 185 * 186 * If forexec is non-zero the current thread and process flags are 187 * cleaned up so they can be reused. 188 */ 189 int 190 killalllwps(int forexec) 191 { 192 struct lwp *lp = curthread->td_lwp; 193 struct proc *p = lp->lwp_proc; 194 195 /* 196 * Interlock against P_WEXIT. Only one of the process's thread 197 * is allowed to do the master exit. 198 */ 199 if (p->p_flag & P_WEXIT) 200 return (EALREADY); 201 p->p_flag |= P_WEXIT; 202 203 /* 204 * Interlock with LWP_WEXIT and kill any remaining LWPs 205 */ 206 lp->lwp_flag |= LWP_WEXIT; 207 if (p->p_nthreads > 1) 208 killlwps(lp); 209 210 /* 211 * If doing this for an exec, clean up the remaining thread 212 * (us) for continuing operation after all the other threads 213 * have been killed. 214 */ 215 if (forexec) { 216 lp->lwp_flag &= ~LWP_WEXIT; 217 p->p_flag &= ~P_WEXIT; 218 } 219 return(0); 220 } 221 222 /* 223 * Kill all LWPs except the current one. Do not try to signal 224 * LWPs which have exited on their own or have already been 225 * signaled. 226 */ 227 static void 228 killlwps(struct lwp *lp) 229 { 230 struct proc *p = lp->lwp_proc; 231 struct lwp *tlp; 232 233 /* 234 * Kill the remaining LWPs. We must send the signal before setting 235 * LWP_WEXIT. The setting of WEXIT is optional but helps reduce 236 * races. tlp must be held across the call as it might block and 237 * allow the target lwp to rip itself out from under our loop. 238 */ 239 FOREACH_LWP_IN_PROC(tlp, p) { 240 LWPHOLD(tlp); 241 if ((tlp->lwp_flag & LWP_WEXIT) == 0) { 242 lwpsignal(p, tlp, SIGKILL); 243 tlp->lwp_flag |= LWP_WEXIT; 244 } 245 LWPRELE(tlp); 246 } 247 248 /* 249 * Wait for everything to clear out. 250 */ 251 while (p->p_nthreads > 1) { 252 if (bootverbose) 253 kprintf("killlwps: waiting for %d lwps of pid " 254 "%d to die\n", 255 p->p_nthreads - 1, p->p_pid); 256 tsleep(&p->p_nthreads, 0, "killlwps", hz); 257 } 258 } 259 260 /* 261 * Exit: deallocate address space and other resources, change proc state 262 * to zombie, and unlink proc from allproc and parent's lists. Save exit 263 * status and rusage for wait(). Check for child processes and orphan them. 264 */ 265 void 266 exit1(int rv) 267 { 268 struct thread *td = curthread; 269 struct proc *p = td->td_proc; 270 struct lwp *lp = td->td_lwp; 271 struct proc *q, *nq; 272 struct vmspace *vm; 273 struct vnode *vtmp; 274 struct exitlist *ep; 275 int error; 276 277 if (p->p_pid == 1) { 278 kprintf("init died (signal %d, exit %d)\n", 279 WTERMSIG(rv), WEXITSTATUS(rv)); 280 panic("Going nowhere without my init!"); 281 } 282 283 /* 284 * Kill all lwps associated with the current process, return an 285 * error if we race another thread trying to do the same thing 286 * and lose the race. 287 */ 288 error = killalllwps(0); 289 if (error) { 290 lwp_exit(0); 291 /* NOT REACHED */ 292 } 293 294 caps_exit(lp->lwp_thread); 295 aio_proc_rundown(p); 296 297 /* are we a task leader? */ 298 if (p == p->p_leader) { 299 struct kill_args killArgs; 300 killArgs.signum = SIGKILL; 301 q = p->p_peers; 302 while(q) { 303 killArgs.pid = q->p_pid; 304 /* 305 * The interface for kill is better 306 * than the internal signal 307 */ 308 sys_kill(&killArgs); 309 nq = q; 310 q = q->p_peers; 311 } 312 while (p->p_peers) 313 tsleep((caddr_t)p, 0, "exit1", 0); 314 } 315 316 #ifdef PGINPROF 317 vmsizmon(); 318 #endif 319 STOPEVENT(p, S_EXIT, rv); 320 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */ 321 322 /* 323 * Check if any loadable modules need anything done at process exit. 324 * e.g. SYSV IPC stuff 325 * XXX what if one of these generates an error? 326 */ 327 TAILQ_FOREACH(ep, &exit_list, next) 328 (*ep->function)(td); 329 330 if (p->p_flag & P_PROFIL) 331 stopprofclock(p); 332 /* 333 * If parent is waiting for us to exit or exec, 334 * P_PPWAIT is set; we will wakeup the parent below. 335 */ 336 p->p_flag &= ~(P_TRACED | P_PPWAIT); 337 SIGEMPTYSET(p->p_siglist); 338 SIGEMPTYSET(lp->lwp_siglist); 339 if (timevalisset(&p->p_realtimer.it_value)) 340 callout_stop(&p->p_ithandle); 341 342 /* 343 * Reset any sigio structures pointing to us as a result of 344 * F_SETOWN with our pid. 345 */ 346 funsetownlst(&p->p_sigiolst); 347 348 /* 349 * Close open files and release open-file table. 350 * This may block! 351 */ 352 fdfree(p); 353 p->p_fd = NULL; 354 355 if(p->p_leader->p_peers) { 356 q = p->p_leader; 357 while(q->p_peers != p) 358 q = q->p_peers; 359 q->p_peers = p->p_peers; 360 wakeup((caddr_t)p->p_leader); 361 } 362 363 /* 364 * XXX Shutdown SYSV semaphores 365 */ 366 semexit(p); 367 368 KKASSERT(p->p_numposixlocks == 0); 369 370 /* The next two chunks should probably be moved to vmspace_exit. */ 371 vm = p->p_vmspace; 372 373 /* 374 * Release upcalls associated with this process 375 */ 376 if (vm->vm_upcalls) 377 upc_release(vm, lp); 378 379 /* 380 * Clean up data related to virtual kernel operation. Clean up 381 * any vkernel context related to the current lwp now so we can 382 * destroy p_vkernel. 383 */ 384 if (p->p_vkernel) { 385 vkernel_lwp_exit(lp); 386 vkernel_exit(p); 387 } 388 389 /* 390 * Release user portion of address space. 391 * This releases references to vnodes, 392 * which could cause I/O if the file has been unlinked. 393 * Need to do this early enough that we can still sleep. 394 * Can't free the entire vmspace as the kernel stack 395 * may be mapped within that space also. 396 * 397 * Processes sharing the same vmspace may exit in one order, and 398 * get cleaned up by vmspace_exit() in a different order. The 399 * last exiting process to reach this point releases as much of 400 * the environment as it can, and the last process cleaned up 401 * by vmspace_exit() (which decrements exitingcnt) cleans up the 402 * remainder. 403 */ 404 ++vm->vm_exitingcnt; 405 sysref_put(&vm->vm_sysref); 406 407 if (SESS_LEADER(p)) { 408 struct session *sp = p->p_session; 409 410 if (sp->s_ttyvp) { 411 /* 412 * We are the controlling process. Signal the 413 * foreground process group, drain the controlling 414 * terminal, and revoke access to the controlling 415 * terminal. 416 * 417 * NOTE: while waiting for the process group to exit 418 * it is possible that one of the processes in the 419 * group will revoke the tty, so the ttyclosesession() 420 * function will re-check sp->s_ttyvp. 421 */ 422 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 423 if (sp->s_ttyp->t_pgrp) 424 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 425 ttywait(sp->s_ttyp); 426 ttyclosesession(sp, 1); /* also revoke */ 427 } 428 /* 429 * Release the tty. If someone has it open via 430 * /dev/tty then close it (since they no longer can 431 * once we've NULL'd it out). 432 */ 433 ttyclosesession(sp, 0); 434 435 /* 436 * s_ttyp is not zero'd; we use this to indicate 437 * that the session once had a controlling terminal. 438 * (for logging and informational purposes) 439 */ 440 } 441 sp->s_leader = NULL; 442 } 443 fixjobc(p, p->p_pgrp, 0); 444 (void)acct_process(p); 445 #ifdef KTRACE 446 /* 447 * release trace file 448 */ 449 if (p->p_tracenode) 450 ktrdestroy(&p->p_tracenode); 451 p->p_traceflag = 0; 452 #endif 453 /* 454 * Release reference to text vnode 455 */ 456 if ((vtmp = p->p_textvp) != NULL) { 457 p->p_textvp = NULL; 458 vrele(vtmp); 459 } 460 461 /* 462 * Move the process to the zombie list. This will block 463 * until the process p_lock count reaches 0. The process will 464 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 465 * which is called from cpu_proc_exit(). 466 */ 467 proc_move_allproc_zombie(p); 468 469 q = LIST_FIRST(&p->p_children); 470 if (q) /* only need this if any child is S_ZOMB */ 471 wakeup((caddr_t) initproc); 472 for (; q != 0; q = nq) { 473 nq = LIST_NEXT(q, p_sibling); 474 LIST_REMOVE(q, p_sibling); 475 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling); 476 q->p_pptr = initproc; 477 q->p_sigparent = SIGCHLD; 478 /* 479 * Traced processes are killed 480 * since their existence means someone is screwing up. 481 */ 482 if (q->p_flag & P_TRACED) { 483 q->p_flag &= ~P_TRACED; 484 ksignal(q, SIGKILL); 485 } 486 } 487 488 /* 489 * Save exit status and final rusage info, adding in child rusage 490 * info and self times. 491 */ 492 p->p_xstat = rv; 493 calcru_proc(p, &p->p_ru); 494 ruadd(&p->p_ru, &p->p_cru); 495 496 /* 497 * notify interested parties of our demise. 498 */ 499 KNOTE(&p->p_klist, NOTE_EXIT); 500 501 /* 502 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 503 * flag set, notify process 1 instead (and hope it will handle 504 * this situation). 505 */ 506 if (p->p_pptr->p_sigacts->ps_flag & PS_NOCLDWAIT) { 507 struct proc *pp = p->p_pptr; 508 proc_reparent(p, initproc); 509 /* 510 * If this was the last child of our parent, notify 511 * parent, so in case he was wait(2)ing, he will 512 * continue. 513 */ 514 if (LIST_EMPTY(&pp->p_children)) 515 wakeup((caddr_t)pp); 516 } 517 518 if (p->p_sigparent && p->p_pptr != initproc) { 519 ksignal(p->p_pptr, p->p_sigparent); 520 } else { 521 ksignal(p->p_pptr, SIGCHLD); 522 } 523 524 wakeup((caddr_t)p->p_pptr); 525 /* 526 * cpu_exit is responsible for clearing curproc, since 527 * it is heavily integrated with the thread/switching sequence. 528 * 529 * Other substructures are freed from wait(). 530 */ 531 plimit_free(p); 532 533 /* 534 * Release the current user process designation on the process so 535 * the userland scheduler can work in someone else. 536 */ 537 p->p_usched->release_curproc(lp); 538 539 /* 540 * Finally, call machine-dependent code to release as many of the 541 * lwp's resources as we can and halt execution of this thread. 542 */ 543 lwp_exit(1); 544 } 545 546 void 547 lwp_exit(int masterexit) 548 { 549 struct lwp *lp = curthread->td_lwp; 550 struct proc *p = lp->lwp_proc; 551 552 /* 553 * lwp_exit() may be called without setting LWP_WEXIT, so 554 * make sure it is set here. 555 */ 556 lp->lwp_flag |= LWP_WEXIT; 557 558 /* 559 * Clean up any virtualization 560 */ 561 if (lp->lwp_vkernel) 562 vkernel_lwp_exit(lp); 563 564 /* 565 * Nobody actually wakes us when the lock 566 * count reaches zero, so just wait one tick. 567 */ 568 while (lp->lwp_lock > 0) 569 tsleep(lp, 0, "lwpexit", 1); 570 571 /* Hand down resource usage to our proc */ 572 ruadd(&p->p_ru, &lp->lwp_ru); 573 574 /* 575 * If we don't hold the process until the LWP is reaped wait*() 576 * may try to dispose of its vmspace before all the LWPs have 577 * actually terminated. 578 */ 579 PHOLD(p); 580 581 /* 582 * We have to use the reaper for all the LWPs except the one doing 583 * the master exit. The LWP doing the master exit can just be 584 * left on p_lwps and the process reaper will deal with it 585 * synchronously, which is much faster. 586 */ 587 if (masterexit == 0) { 588 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 589 --p->p_nthreads; 590 wakeup(&p->p_nthreads); 591 LIST_INSERT_HEAD(&deadlwp_list[mycpuid], lp, u.lwp_reap_entry); 592 taskqueue_enqueue(taskqueue_thread[mycpuid], deadlwp_task[mycpuid]); 593 } else { 594 --p->p_nthreads; 595 } 596 cpu_lwp_exit(); 597 } 598 599 /* 600 * Wait until a lwp is completely dead. 601 * 602 * If the thread is still executing, which can't be waited upon, 603 * return failure. The caller is responsible of waiting a little 604 * bit and checking again. 605 * 606 * Suggested use: 607 * while (!lwp_wait(lp)) 608 * tsleep(lp, 0, "lwpwait", 1); 609 */ 610 static int 611 lwp_wait(struct lwp *lp) 612 { 613 struct thread *td = lp->lwp_thread;; 614 615 KKASSERT(lwkt_preempted_proc() != lp); 616 617 while (lp->lwp_lock > 0) 618 tsleep(lp, 0, "lwpwait1", 1); 619 620 lwkt_wait_free(td); 621 622 /* 623 * The lwp's thread may still be in the middle 624 * of switching away, we can't rip its stack out from 625 * under it until TDF_EXITING is set and both 626 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 627 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 628 * will be cleared temporarily if a thread gets 629 * preempted. 630 * 631 * YYY no wakeup occurs, so we simply return failure 632 * and let the caller deal with sleeping and calling 633 * us again. 634 */ 635 if ((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) != 636 TDF_EXITING) 637 return (0); 638 639 return (1); 640 } 641 642 /* 643 * Release the resources associated with a lwp. 644 * The lwp must be completely dead. 645 */ 646 void 647 lwp_dispose(struct lwp *lp) 648 { 649 struct thread *td = lp->lwp_thread;; 650 651 KKASSERT(lwkt_preempted_proc() != lp); 652 KKASSERT(td->td_refs == 0); 653 KKASSERT((td->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) == 654 TDF_EXITING); 655 656 PRELE(lp->lwp_proc); 657 lp->lwp_proc = NULL; 658 if (td != NULL) { 659 td->td_proc = NULL; 660 td->td_lwp = NULL; 661 lp->lwp_thread = NULL; 662 lwkt_free_thread(td); 663 } 664 kfree(lp, M_LWP); 665 } 666 667 int 668 sys_wait4(struct wait_args *uap) 669 { 670 struct rusage rusage; 671 int error, status; 672 673 error = kern_wait(uap->pid, uap->status ? &status : NULL, 674 uap->options, uap->rusage ? &rusage : NULL, &uap->sysmsg_fds[0]); 675 676 if (error == 0 && uap->status) 677 error = copyout(&status, uap->status, sizeof(*uap->status)); 678 if (error == 0 && uap->rusage) 679 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage)); 680 return (error); 681 } 682 683 /* 684 * wait1() 685 * 686 * wait_args(int pid, int *status, int options, struct rusage *rusage) 687 */ 688 int 689 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res) 690 { 691 struct thread *td = curthread; 692 struct lwp *lp; 693 struct proc *q = td->td_proc; 694 struct proc *p, *t; 695 int nfound, error; 696 697 if (pid == 0) 698 pid = -q->p_pgid; 699 if (options &~ (WUNTRACED|WNOHANG|WCONTINUED|WLINUXCLONE)) 700 return (EINVAL); 701 loop: 702 /* 703 * Hack for backwards compatibility with badly written user code. 704 * Or perhaps we have to do this anyway, it is unclear. XXX 705 * 706 * The problem is that if a process group is stopped and the parent 707 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 708 * of the child and then stop itself when it tries to return from the 709 * system call. When the process group is resumed the parent will 710 * then get the STOP status even though the child has now resumed 711 * (a followup wait*() will get the CONT status). 712 * 713 * Previously the CONT would overwrite the STOP because the tstop 714 * was handled within tsleep(), and the parent would only see 715 * the CONT when both are stopped and continued together. This litte 716 * two-line hack restores this effect. 717 */ 718 while (q->p_stat == SSTOP) 719 tstop(); 720 721 nfound = 0; 722 LIST_FOREACH(p, &q->p_children, p_sibling) { 723 if (pid != WAIT_ANY && 724 p->p_pid != pid && p->p_pgid != -pid) 725 continue; 726 727 /* This special case handles a kthread spawned by linux_clone 728 * (see linux_misc.c). The linux_wait4 and linux_waitpid 729 * functions need to be able to distinguish between waiting 730 * on a process and waiting on a thread. It is a thread if 731 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 732 * signifies we want to wait for threads and not processes. 733 */ 734 if ((p->p_sigparent != SIGCHLD) ^ 735 ((options & WLINUXCLONE) != 0)) { 736 continue; 737 } 738 739 nfound++; 740 if (p->p_stat == SZOMB) { 741 /* 742 * We may go into SZOMB with threads still present. 743 * We must wait for them to exit before we can reap 744 * the master thread, otherwise we may race reaping 745 * non-master threads. 746 */ 747 while (p->p_nthreads > 0) { 748 tsleep(&p->p_nthreads, 0, "lwpzomb", hz); 749 } 750 751 /* 752 * Reap any LWPs left in p->p_lwps. This is usually 753 * just the last LWP. This must be done before 754 * we loop on p_lock since the lwps hold a ref on 755 * it as a vmspace interlock. 756 * 757 * Once that is accomplished p_nthreads had better 758 * be zero. 759 */ 760 while ((lp = RB_ROOT(&p->p_lwp_tree)) != NULL) { 761 lwp_rb_tree_RB_REMOVE(&p->p_lwp_tree, lp); 762 reaplwp(lp); 763 } 764 KKASSERT(p->p_nthreads == 0); 765 766 /* 767 * Don't do anything really bad until all references 768 * to the process go away. This may include other 769 * LWPs which are still in the process of being 770 * reaped. We can't just pull the rug out from under 771 * them because they may still be using the VM space. 772 * 773 * Certain kernel facilities such as /proc will also 774 * put a hold on the process for short periods of 775 * time. 776 */ 777 while (p->p_lock) 778 tsleep(p, 0, "reap3", hz); 779 780 /* scheduling hook for heuristic */ 781 /* XXX no lwp available, we need a different heuristic */ 782 /* 783 p->p_usched->heuristic_exiting(td->td_lwp, deadlp); 784 */ 785 786 /* Take care of our return values. */ 787 *res = p->p_pid; 788 if (status) 789 *status = p->p_xstat; 790 if (rusage) 791 *rusage = p->p_ru; 792 /* 793 * If we got the child via a ptrace 'attach', 794 * we need to give it back to the old parent. 795 */ 796 if (p->p_oppid && (t = pfind(p->p_oppid))) { 797 p->p_oppid = 0; 798 proc_reparent(p, t); 799 ksignal(t, SIGCHLD); 800 wakeup((caddr_t)t); 801 return (0); 802 } 803 p->p_xstat = 0; 804 ruadd(&q->p_cru, &p->p_ru); 805 806 /* 807 * Decrement the count of procs running with this uid. 808 */ 809 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 810 811 /* 812 * Free up credentials. 813 */ 814 crfree(p->p_ucred); 815 p->p_ucred = NULL; 816 817 /* 818 * Remove unused arguments 819 */ 820 if (p->p_args && --p->p_args->ar_ref == 0) 821 FREE(p->p_args, M_PARGS); 822 823 /* 824 * Finally finished with old proc entry. 825 * Unlink it from its process group and free it. 826 */ 827 proc_remove_zombie(p); 828 leavepgrp(p); 829 830 if (--p->p_sigacts->ps_refcnt == 0) { 831 kfree(p->p_sigacts, M_SUBPROC); 832 p->p_sigacts = NULL; 833 } 834 835 vm_waitproc(p); 836 kfree(p, M_PROC); 837 nprocs--; 838 return (0); 839 } 840 if (p->p_stat == SSTOP && (p->p_flag & P_WAITED) == 0 && 841 (p->p_flag & P_TRACED || options & WUNTRACED)) { 842 p->p_flag |= P_WAITED; 843 844 *res = p->p_pid; 845 if (status) 846 *status = W_STOPCODE(p->p_xstat); 847 /* Zero rusage so we get something consistent. */ 848 if (rusage) 849 bzero(rusage, sizeof(rusage)); 850 return (0); 851 } 852 if (options & WCONTINUED && (p->p_flag & P_CONTINUED)) { 853 *res = p->p_pid; 854 p->p_flag &= ~P_CONTINUED; 855 856 if (status) 857 *status = SIGCONT; 858 return (0); 859 } 860 } 861 if (nfound == 0) 862 return (ECHILD); 863 if (options & WNOHANG) { 864 *res = 0; 865 return (0); 866 } 867 error = tsleep((caddr_t)q, PCATCH, "wait", 0); 868 if (error) 869 return (error); 870 goto loop; 871 } 872 873 /* 874 * make process 'parent' the new parent of process 'child'. 875 */ 876 void 877 proc_reparent(struct proc *child, struct proc *parent) 878 { 879 880 if (child->p_pptr == parent) 881 return; 882 883 LIST_REMOVE(child, p_sibling); 884 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 885 child->p_pptr = parent; 886 } 887 888 /* 889 * The next two functions are to handle adding/deleting items on the 890 * exit callout list 891 * 892 * at_exit(): 893 * Take the arguments given and put them onto the exit callout list, 894 * However first make sure that it's not already there. 895 * returns 0 on success. 896 */ 897 898 int 899 at_exit(exitlist_fn function) 900 { 901 struct exitlist *ep; 902 903 #ifdef INVARIANTS 904 /* Be noisy if the programmer has lost track of things */ 905 if (rm_at_exit(function)) 906 kprintf("WARNING: exit callout entry (%p) already present\n", 907 function); 908 #endif 909 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 910 if (ep == NULL) 911 return (ENOMEM); 912 ep->function = function; 913 TAILQ_INSERT_TAIL(&exit_list, ep, next); 914 return (0); 915 } 916 917 /* 918 * Scan the exit callout list for the given item and remove it. 919 * Returns the number of items removed (0 or 1) 920 */ 921 int 922 rm_at_exit(exitlist_fn function) 923 { 924 struct exitlist *ep; 925 926 TAILQ_FOREACH(ep, &exit_list, next) { 927 if (ep->function == function) { 928 TAILQ_REMOVE(&exit_list, ep, next); 929 kfree(ep, M_ATEXIT); 930 return(1); 931 } 932 } 933 return (0); 934 } 935 936 /* 937 * LWP reaper related code. 938 */ 939 static void 940 reaplwps(void *context, int dummy) 941 { 942 struct lwplist *lwplist = context; 943 struct lwp *lp; 944 945 while ((lp = LIST_FIRST(lwplist))) { 946 LIST_REMOVE(lp, u.lwp_reap_entry); 947 reaplwp(lp); 948 } 949 } 950 951 static void 952 reaplwp(struct lwp *lp) 953 { 954 while (lwp_wait(lp) == 0) 955 tsleep(lp, 0, "lwpreap", 1); 956 lwp_dispose(lp); 957 } 958 959 static void 960 deadlwp_init(void) 961 { 962 int cpu; 963 964 for (cpu = 0; cpu < ncpus; cpu++) { 965 LIST_INIT(&deadlwp_list[cpu]); 966 deadlwp_task[cpu] = kmalloc(sizeof(*deadlwp_task[cpu]), M_DEVBUF, M_WAITOK); 967 TASK_INIT(deadlwp_task[cpu], 0, reaplwps, &deadlwp_list[cpu]); 968 } 969 } 970 971 SYSINIT(deadlwpinit, SI_SUB_CONFIGURE, SI_ORDER_ANY, deadlwp_init, NULL); 972