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 1.65 2006/11/07 17:51:23 dillon 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/ptrace.h> 60 #include <sys/acct.h> /* for acct_process() function prototype */ 61 #include <sys/filedesc.h> 62 #include <sys/shm.h> 63 #include <sys/sem.h> 64 #include <sys/aio.h> 65 #include <sys/jail.h> 66 #include <sys/kern_syscall.h> 67 #include <sys/upcall.h> 68 #include <sys/caps.h> 69 70 #include <vm/vm.h> 71 #include <vm/vm_param.h> 72 #include <sys/lock.h> 73 #include <vm/pmap.h> 74 #include <vm/vm_map.h> 75 #include <vm/vm_zone.h> 76 #include <vm/vm_extern.h> 77 #include <sys/user.h> 78 79 #include <sys/thread2.h> 80 81 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 82 static MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status"); 83 84 /* 85 * callout list for things to do at exit time 86 */ 87 struct exitlist { 88 exitlist_fn function; 89 TAILQ_ENTRY(exitlist) next; 90 }; 91 92 TAILQ_HEAD(exit_list_head, exitlist); 93 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 94 95 /* 96 * exit -- 97 * Death of process. 98 * 99 * SYS_EXIT_ARGS(int rval) 100 */ 101 int 102 sys_exit(struct exit_args *uap) 103 { 104 exit1(W_EXITCODE(uap->rval, 0)); 105 /* NOTREACHED */ 106 } 107 108 /* 109 * Exit: deallocate address space and other resources, change proc state 110 * to zombie, and unlink proc from allproc and parent's lists. Save exit 111 * status and rusage for wait(). Check for child processes and orphan them. 112 */ 113 void 114 exit1(int rv) 115 { 116 struct thread *td = curthread; 117 struct proc *p = td->td_proc; 118 struct lwp *lp = td->td_lwp; 119 struct proc *q, *nq; 120 struct vmspace *vm; 121 struct vnode *vtmp; 122 struct exitlist *ep; 123 124 if (p->p_pid == 1) { 125 printf("init died (signal %d, exit %d)\n", 126 WTERMSIG(rv), WEXITSTATUS(rv)); 127 panic("Going nowhere without my init!"); 128 } 129 130 /* XXX lwp kill other threads */ 131 132 caps_exit(lp->lwp_thread); 133 aio_proc_rundown(p); 134 135 /* are we a task leader? */ 136 if(p == p->p_leader) { 137 struct kill_args killArgs; 138 killArgs.signum = SIGKILL; 139 q = p->p_peers; 140 while(q) { 141 killArgs.pid = q->p_pid; 142 /* 143 * The interface for kill is better 144 * than the internal signal 145 */ 146 sys_kill(&killArgs); 147 nq = q; 148 q = q->p_peers; 149 } 150 while (p->p_peers) 151 tsleep((caddr_t)p, 0, "exit1", 0); 152 } 153 154 #ifdef PGINPROF 155 vmsizmon(); 156 #endif 157 STOPEVENT(p, S_EXIT, rv); 158 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */ 159 160 /* 161 * Check if any loadable modules need anything done at process exit. 162 * e.g. SYSV IPC stuff 163 * XXX what if one of these generates an error? 164 */ 165 TAILQ_FOREACH(ep, &exit_list, next) 166 (*ep->function)(td); 167 168 if (p->p_flag & P_PROFIL) 169 stopprofclock(p); 170 MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage), 171 M_ZOMBIE, M_WAITOK); 172 /* 173 * If parent is waiting for us to exit or exec, 174 * P_PPWAIT is set; we will wakeup the parent below. 175 */ 176 p->p_flag &= ~(P_TRACED | P_PPWAIT); 177 p->p_flag |= P_WEXIT; 178 SIGEMPTYSET(p->p_siglist); 179 if (timevalisset(&p->p_realtimer.it_value)) 180 callout_stop(&p->p_ithandle); 181 182 /* 183 * Reset any sigio structures pointing to us as a result of 184 * F_SETOWN with our pid. 185 */ 186 funsetownlst(&p->p_sigiolst); 187 188 /* 189 * Close open files and release open-file table. 190 * This may block! 191 */ 192 fdfree(p); 193 p->p_fd = NULL; 194 195 if(p->p_leader->p_peers) { 196 q = p->p_leader; 197 while(q->p_peers != p) 198 q = q->p_peers; 199 q->p_peers = p->p_peers; 200 wakeup((caddr_t)p->p_leader); 201 } 202 203 /* 204 * XXX Shutdown SYSV semaphores 205 */ 206 semexit(p); 207 208 KKASSERT(p->p_numposixlocks == 0); 209 210 /* The next two chunks should probably be moved to vmspace_exit. */ 211 vm = p->p_vmspace; 212 213 /* 214 * Release upcalls associated with this process 215 */ 216 if (vm->vm_upcalls) 217 upc_release(vm, &p->p_lwp); 218 219 /* clean up data related to virtual kernel operation */ 220 if (p->p_vkernel) 221 vkernel_exit(p); 222 223 /* 224 * Release user portion of address space. 225 * This releases references to vnodes, 226 * which could cause I/O if the file has been unlinked. 227 * Need to do this early enough that we can still sleep. 228 * Can't free the entire vmspace as the kernel stack 229 * may be mapped within that space also. 230 * 231 * Processes sharing the same vmspace may exit in one order, and 232 * get cleaned up by vmspace_exit() in a different order. The 233 * last exiting process to reach this point releases as much of 234 * the environment as it can, and the last process cleaned up 235 * by vmspace_exit() (which decrements exitingcnt) cleans up the 236 * remainder. 237 */ 238 ++vm->vm_exitingcnt; 239 if (--vm->vm_refcnt == 0) { 240 shmexit(vm); 241 pmap_remove_pages(vmspace_pmap(vm), VM_MIN_USER_ADDRESS, 242 VM_MAX_USER_ADDRESS); 243 vm_map_remove(&vm->vm_map, VM_MIN_USER_ADDRESS, 244 VM_MAX_USER_ADDRESS); 245 } 246 247 if (SESS_LEADER(p)) { 248 struct session *sp = p->p_session; 249 struct vnode *vp; 250 251 if (sp->s_ttyvp) { 252 /* 253 * We are the controlling process. Signal the 254 * foreground process group, drain the controlling 255 * terminal, and revoke access to the controlling 256 * terminal. 257 * 258 * NOTE: while waiting for the process group to exit 259 * it is possible that one of the processes in the 260 * group will revoke the tty, so we have to recheck. 261 */ 262 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 263 if (sp->s_ttyp->t_pgrp) 264 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 265 (void) ttywait(sp->s_ttyp); 266 /* 267 * The tty could have been revoked 268 * if we blocked. 269 */ 270 if ((vp = sp->s_ttyvp) != NULL) { 271 ttyclosesession(sp, 0); 272 vx_lock(vp); 273 VOP_REVOKE(vp, REVOKEALL); 274 vx_unlock(vp); 275 vrele(vp); /* s_ttyvp ref */ 276 } 277 } 278 /* 279 * Release the tty. If someone has it open via 280 * /dev/tty then close it (since they no longer can 281 * once we've NULL'd it out). 282 */ 283 if (sp->s_ttyvp) 284 ttyclosesession(sp, 1); 285 /* 286 * s_ttyp is not zero'd; we use this to indicate 287 * that the session once had a controlling terminal. 288 * (for logging and informational purposes) 289 */ 290 } 291 sp->s_leader = NULL; 292 } 293 fixjobc(p, p->p_pgrp, 0); 294 (void)acct_process(p); 295 #ifdef KTRACE 296 /* 297 * release trace file 298 */ 299 if (p->p_tracenode) 300 ktrdestroy(&p->p_tracenode); 301 p->p_traceflag = 0; 302 #endif 303 /* 304 * Release reference to text vnode 305 */ 306 if ((vtmp = p->p_textvp) != NULL) { 307 p->p_textvp = NULL; 308 vrele(vtmp); 309 } 310 311 /* 312 * Move the process to the zombie list. This will block 313 * until the process p_lock count reaches 0. The process will 314 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 315 * which is called from cpu_proc_exit(). 316 */ 317 proc_move_allproc_zombie(p); 318 319 q = LIST_FIRST(&p->p_children); 320 if (q) /* only need this if any child is S_ZOMB */ 321 wakeup((caddr_t) initproc); 322 for (; q != 0; q = nq) { 323 nq = LIST_NEXT(q, p_sibling); 324 LIST_REMOVE(q, p_sibling); 325 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling); 326 q->p_pptr = initproc; 327 q->p_sigparent = SIGCHLD; 328 /* 329 * Traced processes are killed 330 * since their existence means someone is screwing up. 331 */ 332 if (q->p_flag & P_TRACED) { 333 q->p_flag &= ~P_TRACED; 334 ksignal(q, SIGKILL); 335 } 336 } 337 338 /* 339 * Save exit status and final rusage info, adding in child rusage 340 * info and self times. 341 */ 342 p->p_xstat = rv; 343 *p->p_ru = p->p_stats->p_ru; 344 calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL); 345 ruadd(p->p_ru, &p->p_stats->p_cru); 346 347 /* 348 * notify interested parties of our demise. 349 */ 350 KNOTE(&p->p_klist, NOTE_EXIT); 351 352 /* 353 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 354 * flag set, notify process 1 instead (and hope it will handle 355 * this situation). 356 */ 357 if (p->p_pptr->p_procsig->ps_flag & PS_NOCLDWAIT) { 358 struct proc *pp = p->p_pptr; 359 proc_reparent(p, initproc); 360 /* 361 * If this was the last child of our parent, notify 362 * parent, so in case he was wait(2)ing, he will 363 * continue. 364 */ 365 if (LIST_EMPTY(&pp->p_children)) 366 wakeup((caddr_t)pp); 367 } 368 369 if (p->p_sigparent && p->p_pptr != initproc) { 370 ksignal(p->p_pptr, p->p_sigparent); 371 } else { 372 ksignal(p->p_pptr, SIGCHLD); 373 } 374 375 wakeup((caddr_t)p->p_pptr); 376 /* 377 * cpu_exit is responsible for clearing curproc, since 378 * it is heavily integrated with the thread/switching sequence. 379 * 380 * Other substructures are freed from wait(). 381 */ 382 plimit_free(&p->p_limit); 383 384 /* 385 * Release the current user process designation on the process so 386 * the userland scheduler can work in someone else. 387 */ 388 p->p_usched->release_curproc(lp); 389 390 /* 391 * Finally, call machine-dependent code to release the remaining 392 * resources including address space, the kernel stack and pcb. 393 * The address space is released by "vmspace_free(p->p_vmspace)"; 394 * This is machine-dependent, as we may have to change stacks 395 * or ensure that the current one isn't reallocated before we 396 * finish. cpu_exit will end with a call to cpu_switch(), finishing 397 * our execution (pun intended). 398 */ 399 cpu_proc_exit(); 400 } 401 402 int 403 sys_wait4(struct wait_args *uap) 404 { 405 struct rusage rusage; 406 int error, status; 407 408 error = kern_wait(uap->pid, uap->status ? &status : NULL, 409 uap->options, uap->rusage ? &rusage : NULL, &uap->sysmsg_fds[0]); 410 411 if (error == 0 && uap->status) 412 error = copyout(&status, uap->status, sizeof(*uap->status)); 413 if (error == 0 && uap->rusage) 414 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage)); 415 return (error); 416 } 417 418 /* 419 * wait1() 420 * 421 * wait_args(int pid, int *status, int options, struct rusage *rusage) 422 */ 423 int 424 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res) 425 { 426 struct thread *td = curthread; 427 struct thread *deadtd; 428 struct proc *q = td->td_proc; 429 struct proc *p, *t; 430 int nfound, error; 431 432 if (pid == 0) 433 pid = -q->p_pgid; 434 if (options &~ (WUNTRACED|WNOHANG|WLINUXCLONE)) 435 return (EINVAL); 436 loop: 437 /* 438 * Hack for backwards compatibility with badly written user code. 439 * Or perhaps we have to do this anyway, it is unclear. XXX 440 * 441 * The problem is that if a process group is stopped and the parent 442 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 443 * of the child and then stop itself when it tries to return from the 444 * system call. When the process group is resumed the parent will 445 * then get the STOP status even though the child has now resumed 446 * (a followup wait*() will get the CONT status). 447 * 448 * Previously the CONT would overwrite the STOP because the tstop 449 * was handled within tsleep(), and the parent would only see 450 * the CONT when both are stopped and continued together. This litte 451 * two-line hack restores this effect. 452 */ 453 while (q->p_flag & P_STOPPED) 454 tstop(q); 455 456 nfound = 0; 457 LIST_FOREACH(p, &q->p_children, p_sibling) { 458 if (pid != WAIT_ANY && 459 p->p_pid != pid && p->p_pgid != -pid) 460 continue; 461 462 /* This special case handles a kthread spawned by linux_clone 463 * (see linux_misc.c). The linux_wait4 and linux_waitpid 464 * functions need to be able to distinguish between waiting 465 * on a process and waiting on a thread. It is a thread if 466 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 467 * signifies we want to wait for threads and not processes. 468 */ 469 if ((p->p_sigparent != SIGCHLD) ^ 470 ((options & WLINUXCLONE) != 0)) { 471 continue; 472 } 473 474 nfound++; 475 if (p->p_flag & P_ZOMBIE) { 476 KKASSERT((p->p_nthreads == 1)); 477 deadtd = LIST_FIRST(&p->p_lwps)->lwp_thread; 478 479 /* 480 * Other kernel threads may be in the middle of 481 * accessing the proc. For example, kern/kern_proc.c 482 * could be blocked writing proc data to a sysctl. 483 * At the moment, if this occurs, we are not woken 484 * up and rely on a one-second retry. 485 */ 486 if (p->p_lock) { 487 while (p->p_lock) 488 tsleep(p, 0, "reap3", hz); 489 } 490 lwkt_wait_free(deadtd); 491 492 /* 493 * The process's thread may still be in the middle 494 * of switching away, we can't rip its stack out from 495 * under it until TDF_EXITING is set and both 496 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 497 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 498 * will be cleared temporarily if a thread gets 499 * preempted. 500 * 501 * YYY no wakeup occurs so we depend on the timeout. 502 */ 503 if ((deadtd->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) != TDF_EXITING) { 504 tsleep(deadtd, 0, "reap2", 1); 505 goto loop; 506 } 507 508 /* scheduling hook for heuristic */ 509 p->p_usched->heuristic_exiting(td->td_lwp, deadtd->td_lwp); 510 511 /* Take care of our return values. */ 512 *res = p->p_pid; 513 if (status) 514 *status = p->p_xstat; 515 if (rusage) 516 *rusage = *p->p_ru; 517 /* 518 * If we got the child via a ptrace 'attach', 519 * we need to give it back to the old parent. 520 */ 521 if (p->p_oppid && (t = pfind(p->p_oppid))) { 522 p->p_oppid = 0; 523 proc_reparent(p, t); 524 ksignal(t, SIGCHLD); 525 wakeup((caddr_t)t); 526 return (0); 527 } 528 p->p_xstat = 0; 529 ruadd(&q->p_stats->p_cru, p->p_ru); 530 FREE(p->p_ru, M_ZOMBIE); 531 p->p_ru = NULL; 532 533 /* 534 * Decrement the count of procs running with this uid. 535 */ 536 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 537 538 /* 539 * Free up credentials. 540 */ 541 crfree(p->p_ucred); 542 p->p_ucred = NULL; 543 544 /* 545 * Remove unused arguments 546 */ 547 if (p->p_args && --p->p_args->ar_ref == 0) 548 FREE(p->p_args, M_PARGS); 549 550 /* 551 * Finally finished with old proc entry. 552 * Unlink it from its process group and free it. 553 */ 554 leavepgrp(p); 555 proc_remove_zombie(p); 556 557 if (--p->p_procsig->ps_refcnt == 0) { 558 if (p->p_sigacts != &p->p_addr->u_sigacts) 559 FREE(p->p_sigacts, M_SUBPROC); 560 FREE(p->p_procsig, M_SUBPROC); 561 p->p_procsig = NULL; 562 } 563 564 vm_waitproc(p); 565 zfree(proc_zone, p); 566 nprocs--; 567 return (0); 568 } 569 if ((p->p_flag & P_STOPPED) && (p->p_flag & P_WAITED) == 0 && 570 (p->p_flag & P_TRACED || options & WUNTRACED)) { 571 p->p_flag |= P_WAITED; 572 573 *res = p->p_pid; 574 if (status) 575 *status = W_STOPCODE(p->p_xstat); 576 /* Zero rusage so we get something consistent. */ 577 if (rusage) 578 bzero(rusage, sizeof(rusage)); 579 return (0); 580 } 581 } 582 if (nfound == 0) 583 return (ECHILD); 584 if (options & WNOHANG) { 585 *res = 0; 586 return (0); 587 } 588 error = tsleep((caddr_t)q, PCATCH, "wait", 0); 589 if (error) 590 return (error); 591 goto loop; 592 } 593 594 /* 595 * make process 'parent' the new parent of process 'child'. 596 */ 597 void 598 proc_reparent(struct proc *child, struct proc *parent) 599 { 600 601 if (child->p_pptr == parent) 602 return; 603 604 LIST_REMOVE(child, p_sibling); 605 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 606 child->p_pptr = parent; 607 } 608 609 /* 610 * The next two functions are to handle adding/deleting items on the 611 * exit callout list 612 * 613 * at_exit(): 614 * Take the arguments given and put them onto the exit callout list, 615 * However first make sure that it's not already there. 616 * returns 0 on success. 617 */ 618 619 int 620 at_exit(exitlist_fn function) 621 { 622 struct exitlist *ep; 623 624 #ifdef INVARIANTS 625 /* Be noisy if the programmer has lost track of things */ 626 if (rm_at_exit(function)) 627 printf("WARNING: exit callout entry (%p) already present\n", 628 function); 629 #endif 630 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 631 if (ep == NULL) 632 return (ENOMEM); 633 ep->function = function; 634 TAILQ_INSERT_TAIL(&exit_list, ep, next); 635 return (0); 636 } 637 638 /* 639 * Scan the exit callout list for the given item and remove it. 640 * Returns the number of items removed (0 or 1) 641 */ 642 int 643 rm_at_exit(exitlist_fn function) 644 { 645 struct exitlist *ep; 646 647 TAILQ_FOREACH(ep, &exit_list, next) { 648 if (ep->function == function) { 649 TAILQ_REMOVE(&exit_list, ep, next); 650 kfree(ep, M_ATEXIT); 651 return(1); 652 } 653 } 654 return (0); 655 } 656 657 void 658 check_sigacts(void) 659 { 660 struct proc *p = curproc; 661 struct sigacts *pss; 662 663 if (p->p_procsig->ps_refcnt == 1 && 664 p->p_sigacts != &p->p_addr->u_sigacts) { 665 pss = p->p_sigacts; 666 crit_enter(); 667 p->p_addr->u_sigacts = *pss; 668 p->p_sigacts = &p->p_addr->u_sigacts; 669 crit_exit(); 670 FREE(pss, M_SUBPROC); 671 } 672 } 673 674