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 1.67 2007/01/01 22:51:17 corecode 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 kprintf("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 /* 171 * If parent is waiting for us to exit or exec, 172 * P_PPWAIT is set; we will wakeup the parent below. 173 */ 174 p->p_flag &= ~(P_TRACED | P_PPWAIT); 175 p->p_flag |= P_WEXIT; 176 SIGEMPTYSET(p->p_siglist); 177 if (timevalisset(&p->p_realtimer.it_value)) 178 callout_stop(&p->p_ithandle); 179 180 /* 181 * Reset any sigio structures pointing to us as a result of 182 * F_SETOWN with our pid. 183 */ 184 funsetownlst(&p->p_sigiolst); 185 186 /* 187 * Close open files and release open-file table. 188 * This may block! 189 */ 190 fdfree(p); 191 p->p_fd = NULL; 192 193 if(p->p_leader->p_peers) { 194 q = p->p_leader; 195 while(q->p_peers != p) 196 q = q->p_peers; 197 q->p_peers = p->p_peers; 198 wakeup((caddr_t)p->p_leader); 199 } 200 201 /* 202 * XXX Shutdown SYSV semaphores 203 */ 204 semexit(p); 205 206 KKASSERT(p->p_numposixlocks == 0); 207 208 /* The next two chunks should probably be moved to vmspace_exit. */ 209 vm = p->p_vmspace; 210 211 /* 212 * Release upcalls associated with this process 213 */ 214 if (vm->vm_upcalls) 215 upc_release(vm, &p->p_lwp); 216 217 /* clean up data related to virtual kernel operation */ 218 if (p->p_vkernel) 219 vkernel_exit(p); 220 221 /* 222 * Release user portion of address space. 223 * This releases references to vnodes, 224 * which could cause I/O if the file has been unlinked. 225 * Need to do this early enough that we can still sleep. 226 * Can't free the entire vmspace as the kernel stack 227 * may be mapped within that space also. 228 * 229 * Processes sharing the same vmspace may exit in one order, and 230 * get cleaned up by vmspace_exit() in a different order. The 231 * last exiting process to reach this point releases as much of 232 * the environment as it can, and the last process cleaned up 233 * by vmspace_exit() (which decrements exitingcnt) cleans up the 234 * remainder. 235 */ 236 ++vm->vm_exitingcnt; 237 if (--vm->vm_refcnt == 0) { 238 shmexit(vm); 239 pmap_remove_pages(vmspace_pmap(vm), VM_MIN_USER_ADDRESS, 240 VM_MAX_USER_ADDRESS); 241 vm_map_remove(&vm->vm_map, VM_MIN_USER_ADDRESS, 242 VM_MAX_USER_ADDRESS); 243 } 244 245 if (SESS_LEADER(p)) { 246 struct session *sp = p->p_session; 247 struct vnode *vp; 248 249 if (sp->s_ttyvp) { 250 /* 251 * We are the controlling process. Signal the 252 * foreground process group, drain the controlling 253 * terminal, and revoke access to the controlling 254 * terminal. 255 * 256 * NOTE: while waiting for the process group to exit 257 * it is possible that one of the processes in the 258 * group will revoke the tty, so we have to recheck. 259 */ 260 if (sp->s_ttyp && (sp->s_ttyp->t_session == sp)) { 261 if (sp->s_ttyp->t_pgrp) 262 pgsignal(sp->s_ttyp->t_pgrp, SIGHUP, 1); 263 (void) ttywait(sp->s_ttyp); 264 /* 265 * The tty could have been revoked 266 * if we blocked. 267 */ 268 if ((vp = sp->s_ttyvp) != NULL) { 269 ttyclosesession(sp, 0); 270 vx_lock(vp); 271 VOP_REVOKE(vp, REVOKEALL); 272 vx_unlock(vp); 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 if (p->p_tracenode) 298 ktrdestroy(&p->p_tracenode); 299 p->p_traceflag = 0; 300 #endif 301 /* 302 * Release reference to text vnode 303 */ 304 if ((vtmp = p->p_textvp) != NULL) { 305 p->p_textvp = NULL; 306 vrele(vtmp); 307 } 308 309 /* 310 * Move the process to the zombie list. This will block 311 * until the process p_lock count reaches 0. The process will 312 * not be reaped until TDF_EXITING is set by cpu_thread_exit(), 313 * which is called from cpu_proc_exit(). 314 */ 315 proc_move_allproc_zombie(p); 316 317 q = LIST_FIRST(&p->p_children); 318 if (q) /* only need this if any child is S_ZOMB */ 319 wakeup((caddr_t) initproc); 320 for (; q != 0; q = nq) { 321 nq = LIST_NEXT(q, p_sibling); 322 LIST_REMOVE(q, p_sibling); 323 LIST_INSERT_HEAD(&initproc->p_children, q, p_sibling); 324 q->p_pptr = initproc; 325 q->p_sigparent = SIGCHLD; 326 /* 327 * Traced processes are killed 328 * since their existence means someone is screwing up. 329 */ 330 if (q->p_flag & P_TRACED) { 331 q->p_flag &= ~P_TRACED; 332 ksignal(q, SIGKILL); 333 } 334 } 335 336 /* 337 * Save exit status and final rusage info, adding in child rusage 338 * info and self times. 339 */ 340 p->p_xstat = rv; 341 calcru_proc(p, &p->p_ru); 342 ruadd(&p->p_ru, &p->p_cru); 343 344 /* 345 * notify interested parties of our demise. 346 */ 347 KNOTE(&p->p_klist, NOTE_EXIT); 348 349 /* 350 * Notify parent that we're gone. If parent has the PS_NOCLDWAIT 351 * flag set, notify process 1 instead (and hope it will handle 352 * this situation). 353 */ 354 if (p->p_pptr->p_procsig->ps_flag & PS_NOCLDWAIT) { 355 struct proc *pp = p->p_pptr; 356 proc_reparent(p, initproc); 357 /* 358 * If this was the last child of our parent, notify 359 * parent, so in case he was wait(2)ing, he will 360 * continue. 361 */ 362 if (LIST_EMPTY(&pp->p_children)) 363 wakeup((caddr_t)pp); 364 } 365 366 if (p->p_sigparent && p->p_pptr != initproc) { 367 ksignal(p->p_pptr, p->p_sigparent); 368 } else { 369 ksignal(p->p_pptr, SIGCHLD); 370 } 371 372 wakeup((caddr_t)p->p_pptr); 373 /* 374 * cpu_exit is responsible for clearing curproc, since 375 * it is heavily integrated with the thread/switching sequence. 376 * 377 * Other substructures are freed from wait(). 378 */ 379 plimit_free(&p->p_limit); 380 381 /* 382 * Release the current user process designation on the process so 383 * the userland scheduler can work in someone else. 384 */ 385 p->p_usched->release_curproc(lp); 386 387 /* 388 * Finally, call machine-dependent code to release the remaining 389 * resources including address space, the kernel stack and pcb. 390 * The address space is released by "vmspace_free(p->p_vmspace)"; 391 * This is machine-dependent, as we may have to change stacks 392 * or ensure that the current one isn't reallocated before we 393 * finish. cpu_exit will end with a call to cpu_switch(), finishing 394 * our execution (pun intended). 395 */ 396 cpu_proc_exit(); 397 } 398 399 int 400 sys_wait4(struct wait_args *uap) 401 { 402 struct rusage rusage; 403 int error, status; 404 405 error = kern_wait(uap->pid, uap->status ? &status : NULL, 406 uap->options, uap->rusage ? &rusage : NULL, &uap->sysmsg_fds[0]); 407 408 if (error == 0 && uap->status) 409 error = copyout(&status, uap->status, sizeof(*uap->status)); 410 if (error == 0 && uap->rusage) 411 error = copyout(&rusage, uap->rusage, sizeof(*uap->rusage)); 412 return (error); 413 } 414 415 /* 416 * wait1() 417 * 418 * wait_args(int pid, int *status, int options, struct rusage *rusage) 419 */ 420 int 421 kern_wait(pid_t pid, int *status, int options, struct rusage *rusage, int *res) 422 { 423 struct thread *td = curthread; 424 struct thread *deadtd; 425 struct proc *q = td->td_proc; 426 struct proc *p, *t; 427 int nfound, error; 428 429 if (pid == 0) 430 pid = -q->p_pgid; 431 if (options &~ (WUNTRACED|WNOHANG|WLINUXCLONE)) 432 return (EINVAL); 433 loop: 434 /* 435 * Hack for backwards compatibility with badly written user code. 436 * Or perhaps we have to do this anyway, it is unclear. XXX 437 * 438 * The problem is that if a process group is stopped and the parent 439 * is doing a wait*(..., WUNTRACED, ...), it will see the STOP 440 * of the child and then stop itself when it tries to return from the 441 * system call. When the process group is resumed the parent will 442 * then get the STOP status even though the child has now resumed 443 * (a followup wait*() will get the CONT status). 444 * 445 * Previously the CONT would overwrite the STOP because the tstop 446 * was handled within tsleep(), and the parent would only see 447 * the CONT when both are stopped and continued together. This litte 448 * two-line hack restores this effect. 449 */ 450 while (q->p_flag & P_STOPPED) 451 tstop(q); 452 453 nfound = 0; 454 LIST_FOREACH(p, &q->p_children, p_sibling) { 455 if (pid != WAIT_ANY && 456 p->p_pid != pid && p->p_pgid != -pid) 457 continue; 458 459 /* This special case handles a kthread spawned by linux_clone 460 * (see linux_misc.c). The linux_wait4 and linux_waitpid 461 * functions need to be able to distinguish between waiting 462 * on a process and waiting on a thread. It is a thread if 463 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 464 * signifies we want to wait for threads and not processes. 465 */ 466 if ((p->p_sigparent != SIGCHLD) ^ 467 ((options & WLINUXCLONE) != 0)) { 468 continue; 469 } 470 471 nfound++; 472 if (p->p_flag & P_ZOMBIE) { 473 KKASSERT((p->p_nthreads == 1)); 474 deadtd = LIST_FIRST(&p->p_lwps)->lwp_thread; 475 476 /* 477 * Other kernel threads may be in the middle of 478 * accessing the proc. For example, kern/kern_proc.c 479 * could be blocked writing proc data to a sysctl. 480 * At the moment, if this occurs, we are not woken 481 * up and rely on a one-second retry. 482 */ 483 if (p->p_lock) { 484 while (p->p_lock) 485 tsleep(p, 0, "reap3", hz); 486 } 487 lwkt_wait_free(deadtd); 488 489 /* 490 * The process's thread may still be in the middle 491 * of switching away, we can't rip its stack out from 492 * under it until TDF_EXITING is set and both 493 * TDF_RUNNING and TDF_PREEMPT_LOCK are clear. 494 * TDF_PREEMPT_LOCK must be checked because TDF_RUNNING 495 * will be cleared temporarily if a thread gets 496 * preempted. 497 * 498 * YYY no wakeup occurs so we depend on the timeout. 499 */ 500 if ((deadtd->td_flags & (TDF_RUNNING|TDF_PREEMPT_LOCK|TDF_EXITING)) != TDF_EXITING) { 501 tsleep(deadtd, 0, "reap2", 1); 502 goto loop; 503 } 504 505 /* scheduling hook for heuristic */ 506 p->p_usched->heuristic_exiting(td->td_lwp, deadtd->td_lwp); 507 508 /* Take care of our return values. */ 509 *res = p->p_pid; 510 if (status) 511 *status = p->p_xstat; 512 if (rusage) 513 *rusage = p->p_ru; 514 /* 515 * If we got the child via a ptrace 'attach', 516 * we need to give it back to the old parent. 517 */ 518 if (p->p_oppid && (t = pfind(p->p_oppid))) { 519 p->p_oppid = 0; 520 proc_reparent(p, t); 521 ksignal(t, SIGCHLD); 522 wakeup((caddr_t)t); 523 return (0); 524 } 525 p->p_xstat = 0; 526 ruadd(&q->p_cru, &p->p_ru); 527 528 /* 529 * Decrement the count of procs running with this uid. 530 */ 531 chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 532 533 /* 534 * Free up credentials. 535 */ 536 crfree(p->p_ucred); 537 p->p_ucred = NULL; 538 539 /* 540 * Remove unused arguments 541 */ 542 if (p->p_args && --p->p_args->ar_ref == 0) 543 FREE(p->p_args, M_PARGS); 544 545 /* 546 * Finally finished with old proc entry. 547 * Unlink it from its process group and free it. 548 */ 549 leavepgrp(p); 550 proc_remove_zombie(p); 551 552 if (--p->p_procsig->ps_refcnt == 0) { 553 if (p->p_sigacts != &p->p_addr->u_sigacts) 554 FREE(p->p_sigacts, M_SUBPROC); 555 FREE(p->p_procsig, M_SUBPROC); 556 p->p_procsig = NULL; 557 } 558 559 vm_waitproc(p); 560 zfree(proc_zone, p); 561 nprocs--; 562 return (0); 563 } 564 if ((p->p_flag & P_STOPPED) && (p->p_flag & P_WAITED) == 0 && 565 (p->p_flag & P_TRACED || options & WUNTRACED)) { 566 p->p_flag |= P_WAITED; 567 568 *res = p->p_pid; 569 if (status) 570 *status = W_STOPCODE(p->p_xstat); 571 /* Zero rusage so we get something consistent. */ 572 if (rusage) 573 bzero(rusage, sizeof(rusage)); 574 return (0); 575 } 576 } 577 if (nfound == 0) 578 return (ECHILD); 579 if (options & WNOHANG) { 580 *res = 0; 581 return (0); 582 } 583 error = tsleep((caddr_t)q, PCATCH, "wait", 0); 584 if (error) 585 return (error); 586 goto loop; 587 } 588 589 /* 590 * make process 'parent' the new parent of process 'child'. 591 */ 592 void 593 proc_reparent(struct proc *child, struct proc *parent) 594 { 595 596 if (child->p_pptr == parent) 597 return; 598 599 LIST_REMOVE(child, p_sibling); 600 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 601 child->p_pptr = parent; 602 } 603 604 /* 605 * The next two functions are to handle adding/deleting items on the 606 * exit callout list 607 * 608 * at_exit(): 609 * Take the arguments given and put them onto the exit callout list, 610 * However first make sure that it's not already there. 611 * returns 0 on success. 612 */ 613 614 int 615 at_exit(exitlist_fn function) 616 { 617 struct exitlist *ep; 618 619 #ifdef INVARIANTS 620 /* Be noisy if the programmer has lost track of things */ 621 if (rm_at_exit(function)) 622 kprintf("WARNING: exit callout entry (%p) already present\n", 623 function); 624 #endif 625 ep = kmalloc(sizeof(*ep), M_ATEXIT, M_NOWAIT); 626 if (ep == NULL) 627 return (ENOMEM); 628 ep->function = function; 629 TAILQ_INSERT_TAIL(&exit_list, ep, next); 630 return (0); 631 } 632 633 /* 634 * Scan the exit callout list for the given item and remove it. 635 * Returns the number of items removed (0 or 1) 636 */ 637 int 638 rm_at_exit(exitlist_fn function) 639 { 640 struct exitlist *ep; 641 642 TAILQ_FOREACH(ep, &exit_list, next) { 643 if (ep->function == function) { 644 TAILQ_REMOVE(&exit_list, ep, next); 645 kfree(ep, M_ATEXIT); 646 return(1); 647 } 648 } 649 return (0); 650 } 651 652 void 653 check_sigacts(void) 654 { 655 struct proc *p = curproc; 656 struct sigacts *pss; 657 658 if (p->p_procsig->ps_refcnt == 1 && 659 p->p_sigacts != &p->p_addr->u_sigacts) { 660 pss = p->p_sigacts; 661 crit_enter(); 662 p->p_addr->u_sigacts = *pss; 663 p->p_sigacts = &p->p_addr->u_sigacts; 664 crit_exit(); 665 FREE(pss, M_SUBPROC); 666 } 667 } 668 669