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 1.52 2005/12/28 19:13:34 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/pioctl.h> 53 #include <sys/tty.h> 54 #include <sys/wait.h> 55 #include <sys/vnode.h> 56 #include <sys/resourcevar.h> 57 #include <sys/signalvar.h> 58 #include <sys/ptrace.h> 59 #include <sys/acct.h> /* for acct_process() function prototype */ 60 #include <sys/filedesc.h> 61 #include <sys/shm.h> 62 #include <sys/sem.h> 63 #include <sys/aio.h> 64 #include <sys/jail.h> 65 #include <sys/kern_syscall.h> 66 #include <sys/upcall.h> 67 #include <sys/caps.h> 68 69 #include <vm/vm.h> 70 #include <vm/vm_param.h> 71 #include <sys/lock.h> 72 #include <vm/pmap.h> 73 #include <vm/vm_map.h> 74 #include <vm/vm_zone.h> 75 #include <vm/vm_extern.h> 76 #include <sys/user.h> 77 78 #include <sys/thread2.h> 79 80 /* Required to be non-static for SysVR4 emulator */ 81 MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status"); 82 83 static MALLOC_DEFINE(M_ATEXIT, "atexit", "atexit callback"); 84 85 /* 86 * callout list for things to do at exit time 87 */ 88 struct exitlist { 89 exitlist_fn function; 90 TAILQ_ENTRY(exitlist) next; 91 }; 92 93 TAILQ_HEAD(exit_list_head, exitlist); 94 static struct exit_list_head exit_list = TAILQ_HEAD_INITIALIZER(exit_list); 95 96 /* 97 * exit -- 98 * Death of process. 99 * 100 * SYS_EXIT_ARGS(int rval) 101 */ 102 void 103 sys_exit(struct sys_exit_args *uap) 104 { 105 exit1(W_EXITCODE(uap->rval, 0)); 106 /* NOTREACHED */ 107 } 108 109 /* 110 * Exit: deallocate address space and other resources, change proc state 111 * to zombie, and unlink proc from allproc and parent's lists. Save exit 112 * status and rusage for wait(). Check for child processes and orphan them. 113 */ 114 void 115 exit1(int rv) 116 { 117 struct proc *p = curproc; 118 struct lwp *lp; 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 lp = &p->p_lwp; /* XXX lwp kill other threads */ 131 132 sysmsg_rundown(lp, 1); 133 caps_exit(lp->lwp_thread); 134 aio_proc_rundown(p); 135 136 /* are we a task leader? */ 137 if(p == p->p_leader) { 138 struct kill_args killArgs; 139 killArgs.signum = SIGKILL; 140 q = p->p_peers; 141 while(q) { 142 killArgs.pid = q->p_pid; 143 /* 144 * The interface for kill is better 145 * than the internal signal 146 */ 147 kill(&killArgs); 148 nq = q; 149 q = q->p_peers; 150 } 151 while (p->p_peers) 152 tsleep((caddr_t)p, 0, "exit1", 0); 153 } 154 155 #ifdef PGINPROF 156 vmsizmon(); 157 #endif 158 STOPEVENT(p, S_EXIT, rv); 159 wakeup(&p->p_stype); /* Wakeup anyone in procfs' PIOCWAIT */ 160 161 /* 162 * Check if any loadable modules need anything done at process exit. 163 * e.g. SYSV IPC stuff 164 * XXX what if one of these generates an error? 165 */ 166 TAILQ_FOREACH(ep, &exit_list, next) 167 (*ep->function)(p->p_thread); 168 169 if (p->p_flag & P_PROFIL) 170 stopprofclock(p); 171 MALLOC(p->p_ru, struct rusage *, sizeof(struct rusage), 172 M_ZOMBIE, M_WAITOK); 173 /* 174 * If parent is waiting for us to exit or exec, 175 * P_PPWAIT is set; we will wakeup the parent below. 176 */ 177 p->p_flag &= ~(P_TRACED | P_PPWAIT); 178 p->p_flag |= P_WEXIT; 179 SIGEMPTYSET(p->p_siglist); 180 if (timevalisset(&p->p_realtimer.it_value)) 181 callout_stop(&p->p_ithandle); 182 183 /* 184 * Reset any sigio structures pointing to us as a result of 185 * F_SETOWN with our pid. 186 */ 187 funsetownlst(&p->p_sigiolst); 188 189 /* 190 * Close open files and release open-file table. 191 * This may block! 192 */ 193 fdfree(p); 194 p->p_fd = NULL; 195 196 if(p->p_leader->p_peers) { 197 q = p->p_leader; 198 while(q->p_peers != p) 199 q = q->p_peers; 200 q->p_peers = p->p_peers; 201 wakeup((caddr_t)p->p_leader); 202 } 203 204 /* 205 * XXX Shutdown SYSV semaphores 206 */ 207 semexit(p); 208 209 KKASSERT(p->p_numposixlocks == 0); 210 211 /* The next two chunks should probably be moved to vmspace_exit. */ 212 vm = p->p_vmspace; 213 214 /* 215 * Release upcalls associated with this process 216 */ 217 if (vm->vm_upcalls) 218 upc_release(vm, &p->p_lwp); 219 220 /* 221 * Release user portion of address space. 222 * This releases references to vnodes, 223 * which could cause I/O if the file has been unlinked. 224 * Need to do this early enough that we can still sleep. 225 * Can't free the entire vmspace as the kernel stack 226 * may be mapped within that space also. 227 * 228 * Processes sharing the same vmspace may exit in one order, and 229 * get cleaned up by vmspace_exit() in a different order. The 230 * last exiting process to reach this point releases as much of 231 * the environment as it can, and the last process cleaned up 232 * by vmspace_exit() (which decrements exitingcnt) cleans up the 233 * remainder. 234 */ 235 ++vm->vm_exitingcnt; 236 if (--vm->vm_refcnt == 0) { 237 shmexit(vm); 238 pmap_remove_pages(vmspace_pmap(vm), VM_MIN_ADDRESS, 239 VM_MAXUSER_ADDRESS); 240 (void) vm_map_remove(&vm->vm_map, VM_MIN_ADDRESS, 241 VM_MAXUSER_ADDRESS); 242 } 243 244 if (SESS_LEADER(p)) { 245 struct session *sp = p->p_session; 246 struct vnode *vp; 247 248 if (sp->s_ttyvp) { 249 /* 250 * We are the controlling process. Signal the 251 * foreground process group, drain the controlling 252 * terminal, and revoke access to the controlling 253 * terminal. 254 * 255 * NOTE: while waiting for the process group to exit 256 * it is possible that one of the processes in the 257 * group will revoke the tty, so we have to recheck. 258 */ 259 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 260 if (sp->s_ttyp->t_pgrp) 261 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 262 (void) ttywait(sp->s_ttyp); 263 /* 264 * The tty could have been revoked 265 * if we blocked. 266 */ 267 if ((vp = sp->s_ttyvp) != NULL) { 268 ttyclosesession(sp, 0); 269 if (vx_lock(vp) == 0) { 270 VOP_REVOKE(vp, REVOKEALL); 271 vx_unlock(vp); 272 } 273 vrele(vp); /* s_ttyvp ref */ 274 } 275 } 276 /* 277 * Release the tty. If someone has it open via 278 * /dev/tty then close it (since they no longer can 279 * once we've NULL'd it out). 280 */ 281 if (sp->s_ttyvp) 282 ttyclosesession(sp, 1); 283 /* 284 * s_ttyp is not zero'd; we use this to indicate 285 * that the session once had a controlling terminal. 286 * (for logging and informational purposes) 287 */ 288 } 289 sp->s_leader = NULL; 290 } 291 fixjobc(p, p->p_pgrp, 0); 292 (void)acct_process(p); 293 #ifdef KTRACE 294 /* 295 * release trace file 296 */ 297 p->p_traceflag = 0; /* don't trace the vrele() */ 298 if ((vtmp = p->p_tracep) != NULL) { 299 p->p_tracep = NULL; 300 vrele(vtmp); 301 } 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 * Once we set SZOMB the process can get reaped. The wait1 code 313 * will also wait for TDF_EXITING to be set and for both TDF_RUNNING 314 * and TDF_PREEMPT_LOCK to be cleared in the thread's flags, 315 * indicating that it has been completely switched out for the last 316 * time. 317 */ 318 319 /* 320 * Remove proc from allproc queue and pidhash chain. 321 * Place onto zombproc. Unlink from parent's child list. 322 * 323 * Interlock the SZOMB state with a tsleep against p_lock 324 * (PHOLD/PRELE) so allproc loops don't get confused. Get 325 * our own ref on p_lock to prevent us from getting reaped 326 * too early. 327 */ 328 LIST_REMOVE(p, p_list); 329 LIST_INSERT_HEAD(&zombproc, p, p_list); 330 p->p_flag |= P_ZOMBIE; 331 while (p->p_lock) 332 tsleep(p, 0, "reap1", hz / 10); 333 LIST_REMOVE(p, p_hash); 334 335 q = LIST_FIRST(&p->p_children); 336 if (q) /* only need this if any child is S_ZOMB */ 337 wakeup((caddr_t) initproc); 338 for (; q != 0; q = nq) { 339 nq = LIST_NEXT(q, p_sibling); 340 LIST_REMOVE(q, p_sibling); 341 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling); 342 q->p_pptr = initproc; 343 q->p_sigparent = SIGCHLD; 344 /* 345 * Traced processes are killed 346 * since their existence means someone is screwing up. 347 */ 348 if (q->p_flag & P_TRACED) { 349 q->p_flag &= ~P_TRACED; 350 psignal(q, SIGKILL); 351 } 352 } 353 354 /* 355 * Save exit status and final rusage info, adding in child rusage 356 * info and self times. 357 */ 358 p->p_xstat = rv; 359 *p->p_ru = p->p_stats->p_ru; 360 calcru(p, &p->p_ru->ru_utime, &p->p_ru->ru_stime, NULL); 361 ruadd(p->p_ru, &p->p_stats->p_cru); 362 363 /* 364 * notify interested parties of our demise. 365 */ 366 KNOTE(&p->p_klist, NOTE_EXIT); 367 368 /* 369 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 370 * flag set, notify process 1 instead (and hope it will handle 371 * this situation). 372 */ 373 if (p->p_pptr->p_procsig->ps_flag & PS_NOCLDWAIT) { 374 struct proc *pp = p->p_pptr; 375 proc_reparent(p, initproc); 376 /* 377 * If this was the last child of our parent, notify 378 * parent, so in case he was wait(2)ing, he will 379 * continue. 380 */ 381 if (LIST_EMPTY(&pp->p_children)) 382 wakeup((caddr_t)pp); 383 } 384 385 if (p->p_sigparent && p->p_pptr != initproc) { 386 psignal(p->p_pptr, p->p_sigparent); 387 } else { 388 psignal(p->p_pptr, SIGCHLD); 389 } 390 391 wakeup((caddr_t)p->p_pptr); 392 /* 393 * cpu_exit is responsible for clearing curproc, since 394 * it is heavily integrated with the thread/switching sequence. 395 * 396 * Other substructures are freed from wait(). 397 */ 398 if (--p->p_limit->p_refcnt == 0) { 399 FREE(p->p_limit, M_SUBPROC); 400 p->p_limit = NULL; 401 } 402 403 /* 404 * Release the current user process designation on the process so 405 * the userland scheduler can work in someone else. 406 */ 407 p->p_usched->release_curproc(lp); 408 409 /* 410 * Finally, call machine-dependent code to release the remaining 411 * resources including address space, the kernel stack and pcb. 412 * The address space is released by "vmspace_free(p->p_vmspace)"; 413 * This is machine-dependent, as we may have to change stacks 414 * or ensure that the current one isn't reallocated before we 415 * finish. cpu_exit will end with a call to cpu_switch(), finishing 416 * our execution (pun intended). 417 */ 418 cpu_proc_exit(); 419 } 420 421 int 422 wait4(struct wait_args *uap) 423 { 424 struct rusage rusage; 425 int error, status; 426 427 error = kern_wait(uap->pid, uap->status ? &status : NULL, 428 uap->options, uap->rusage ? &rusage : NULL, &uap->sysmsg_fds[0]); 429 430 if (error == 0 && uap->status) 431 error = copyout(&status, uap->status, sizeof(*uap->status)); 432 if (error == 0 && uap->rusage) 433 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage)); 434 return (error); 435 } 436 437 /* 438 * wait1() 439 * 440 * wait_args(int pid, int *status, int options, struct rusage *rusage) 441 */ 442 int 443 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res) 444 { 445 struct thread *td = curthread; 446 struct proc *q = td->td_proc; 447 struct proc *p, *t; 448 int nfound, error; 449 450 if (pid == 0) 451 pid = -q->p_pgid; 452 if (options &~ (WUNTRACED|WNOHANG|WLINUXCLONE)) 453 return (EINVAL); 454 loop: 455 /* 456 * Hack for backwards compatibility with badly written user code. 457 * Or perhaps we have to do this anyway, it is unclear. XXX 458 * 459 * The problem is that if a process group is stopped and the parent 460 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 461 * of the child and then stop itself when it tries to return from the 462 * system call. When the process group is resumed the parent will 463 * then get the STOP status even though the child has now resumed 464 * (a followup wait*() will get the CONT status). 465 * 466 * Previously the CONT would overwrite the STOP because the tstop 467 * was handled within tsleep(), and the parent would only see 468 * the CONT when both are stopped and continued together. This litte 469 * two-line hack restores this effect. 470 */ 471 while (q->p_flag & P_STOPPED) 472 tstop(q); 473 474 nfound = 0; 475 LIST_FOREACH(p, &q->p_children, p_sibling) { 476 if (pid != WAIT_ANY && 477 p->p_pid != pid && p->p_pgid != -pid) 478 continue; 479 480 /* This special case handles a kthread spawned by linux_clone 481 * (see linux_misc.c). The linux_wait4 and linux_waitpid 482 * functions need to be able to distinguish between waiting 483 * on a process and waiting on a thread. It is a thread if 484 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 485 * signifies we want to wait for threads and not processes. 486 */ 487 if ((p->p_sigparent != SIGCHLD) ^ 488 ((options & WLINUXCLONE) != 0)) { 489 continue; 490 } 491 492 nfound++; 493 if (p->p_flag & P_ZOMBIE) { 494 /* 495 * Other kernel threads may be in the middle of 496 * accessing the proc. For example, kern/kern_proc.c 497 * could be blocked writing proc data to a sysctl. 498 * At the moment, if this occurs, we are not woken 499 * up and rely on a one-second retry. 500 */ 501 if (p->p_lock) { 502 while (p->p_lock) 503 tsleep(p, 0, "reap3", hz); 504 } 505 lwkt_wait_free(p->p_thread); 506 507 /* 508 * The process's thread may still be in the middle 509 * of switching away, we can't rip its stack out from 510 * under it until TDF_EXITING is set and both 511 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 512 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 513 * will be cleared temporarily if a thread gets 514 * preempted. 515 * 516 * YYY no wakeup occurs so we depend on the timeout. 517 */ 518 if ((p->p_thread->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) != TDF_EXITING) { 519 tsleep(p->p_thread, 0, "reap2", 1); 520 goto loop; 521 } 522 523 /* scheduling hook for heuristic */ 524 p->p_usched->heuristic_exiting(td->td_lwp, &p->p_lwp); 525 526 /* Take care of our return values. */ 527 *res = p->p_pid; 528 if (status) 529 *status = p->p_xstat; 530 if (rusage) 531 *rusage = *p->p_ru; 532 /* 533 * If we got the child via a ptrace 'attach', 534 * we need to give it back to the old parent. 535 */ 536 if (p->p_oppid && (t = pfind(p->p_oppid))) { 537 p->p_oppid = 0; 538 proc_reparent(p, t); 539 psignal(t, SIGCHLD); 540 wakeup((caddr_t)t); 541 return (0); 542 } 543 p->p_xstat = 0; 544 ruadd(&q->p_stats->p_cru, p->p_ru); 545 FREE(p->p_ru, M_ZOMBIE); 546 p->p_ru = NULL; 547 548 /* 549 * Decrement the count of procs running with this uid. 550 */ 551 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 552 553 /* 554 * Free up credentials. 555 */ 556 crfree(p->p_ucred); 557 p->p_ucred = NULL; 558 559 /* 560 * Remove unused arguments 561 */ 562 if (p->p_args && --p->p_args->ar_ref == 0) 563 FREE(p->p_args, M_PARGS); 564 565 /* 566 * Finally finished with old proc entry. 567 * Unlink it from its process group and free it. 568 */ 569 leavepgrp(p); 570 LIST_REMOVE(p, p_list); /* off zombproc */ 571 LIST_REMOVE(p, p_sibling); 572 573 if (--p->p_procsig->ps_refcnt == 0) { 574 if (p->p_sigacts != &p->p_addr->u_sigacts) 575 FREE(p->p_sigacts, M_SUBPROC); 576 FREE(p->p_procsig, M_SUBPROC); 577 p->p_procsig = NULL; 578 } 579 580 vm_waitproc(p); 581 zfree(proc_zone, p); 582 nprocs--; 583 return (0); 584 } 585 if ((p->p_flag & P_STOPPED) && (p->p_flag & P_WAITED) == 0 && 586 (p->p_flag & P_TRACED || options & WUNTRACED)) { 587 p->p_flag |= P_WAITED; 588 589 *res = p->p_pid; 590 if (status) 591 *status = W_STOPCODE(p->p_xstat); 592 /* Zero rusage so we get something consistent. */ 593 if (rusage) 594 bzero(rusage, sizeof(rusage)); 595 return (0); 596 } 597 } 598 if (nfound == 0) 599 return (ECHILD); 600 if (options & WNOHANG) { 601 *res = 0; 602 return (0); 603 } 604 error = tsleep((caddr_t)q, PCATCH, "wait", 0); 605 if (error) 606 return (error); 607 goto loop; 608 } 609 610 /* 611 * make process 'parent' the new parent of process 'child'. 612 */ 613 void 614 proc_reparent(struct proc *child, struct proc *parent) 615 { 616 617 if (child->p_pptr == parent) 618 return; 619 620 LIST_REMOVE(child, p_sibling); 621 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 622 child->p_pptr = parent; 623 } 624 625 /* 626 * The next two functions are to handle adding/deleting items on the 627 * exit callout list 628 * 629 * at_exit(): 630 * Take the arguments given and put them onto the exit callout list, 631 * However first make sure that it's not already there. 632 * returns 0 on success. 633 */ 634 635 int 636 at_exit(exitlist_fn function) 637 { 638 struct exitlist *ep; 639 640 #ifdef INVARIANTS 641 /* Be noisy if the programmer has lost track of things */ 642 if (rm_at_exit(function)) 643 printf("WARNING: exit callout entry (%p) already present\n", 644 function); 645 #endif 646 ep = malloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 647 if (ep == NULL) 648 return (ENOMEM); 649 ep->function = function; 650 TAILQ_INSERT_TAIL(&exit_list, ep, next); 651 return (0); 652 } 653 654 /* 655 * Scan the exit callout list for the given item and remove it. 656 * Returns the number of items removed (0 or 1) 657 */ 658 int 659 rm_at_exit(exitlist_fn function) 660 { 661 struct exitlist *ep; 662 663 TAILQ_FOREACH(ep, &exit_list, next) { 664 if (ep->function == function) { 665 TAILQ_REMOVE(&exit_list, ep, next); 666 free(ep, M_ATEXIT); 667 return(1); 668 } 669 } 670 return (0); 671 } 672 673 void 674 check_sigacts(void) 675 { 676 struct proc *p = curproc; 677 struct sigacts *pss; 678 679 if (p->p_procsig->ps_refcnt == 1 && 680 p->p_sigacts != &p->p_addr->u_sigacts) { 681 pss = p->p_sigacts; 682 crit_enter(); 683 p->p_addr->u_sigacts = *pss; 684 p->p_sigacts = &p->p_addr->u_sigacts; 685 crit_exit(); 686 FREE(pss, M_SUBPROC); 687 } 688 } 689 690