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 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_compat.h" 41 #include "opt_ktrace.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/sysproto.h> 46 #include <sys/capsicum.h> 47 #include <sys/eventhandler.h> 48 #include <sys/kernel.h> 49 #include <sys/malloc.h> 50 #include <sys/lock.h> 51 #include <sys/mutex.h> 52 #include <sys/proc.h> 53 #include <sys/procdesc.h> 54 #include <sys/pioctl.h> 55 #include <sys/jail.h> 56 #include <sys/tty.h> 57 #include <sys/wait.h> 58 #include <sys/vmmeter.h> 59 #include <sys/vnode.h> 60 #include <sys/racct.h> 61 #include <sys/resourcevar.h> 62 #include <sys/sbuf.h> 63 #include <sys/signalvar.h> 64 #include <sys/sched.h> 65 #include <sys/sx.h> 66 #include <sys/syscallsubr.h> 67 #include <sys/syslog.h> 68 #include <sys/ptrace.h> 69 #include <sys/acct.h> /* for acct_process() function prototype */ 70 #include <sys/filedesc.h> 71 #include <sys/sdt.h> 72 #include <sys/shm.h> 73 #include <sys/sem.h> 74 #include <sys/umtx.h> 75 #ifdef KTRACE 76 #include <sys/ktrace.h> 77 #endif 78 79 #include <security/audit/audit.h> 80 #include <security/mac/mac_framework.h> 81 82 #include <vm/vm.h> 83 #include <vm/vm_extern.h> 84 #include <vm/vm_param.h> 85 #include <vm/pmap.h> 86 #include <vm/vm_map.h> 87 #include <vm/vm_page.h> 88 #include <vm/uma.h> 89 #include <vm/vm_domain.h> 90 91 #ifdef KDTRACE_HOOKS 92 #include <sys/dtrace_bsd.h> 93 dtrace_execexit_func_t dtrace_fasttrap_exit; 94 #endif 95 96 SDT_PROVIDER_DECLARE(proc); 97 SDT_PROBE_DEFINE1(proc, , , exit, "int"); 98 99 /* Hook for NFS teardown procedure. */ 100 void (*nlminfo_release_p)(struct proc *p); 101 102 struct proc * 103 proc_realparent(struct proc *child) 104 { 105 struct proc *p, *parent; 106 107 sx_assert(&proctree_lock, SX_LOCKED); 108 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 109 if (child->p_oppid == 0 || 110 child->p_pptr->p_pid == child->p_oppid) 111 parent = child->p_pptr; 112 else 113 parent = initproc; 114 return (parent); 115 } 116 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 117 /* Cannot use LIST_PREV(), since the list head is not known. */ 118 p = __containerof(p->p_orphan.le_prev, struct proc, 119 p_orphan.le_next); 120 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 121 ("missing P_ORPHAN %p", p)); 122 } 123 parent = __containerof(p->p_orphan.le_prev, struct proc, 124 p_orphans.lh_first); 125 return (parent); 126 } 127 128 void 129 reaper_abandon_children(struct proc *p, bool exiting) 130 { 131 struct proc *p1, *p2, *ptmp; 132 133 sx_assert(&proctree_lock, SX_LOCKED); 134 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 135 if ((p->p_treeflag & P_TREE_REAPER) == 0) 136 return; 137 p1 = p->p_reaper; 138 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 139 LIST_REMOVE(p2, p_reapsibling); 140 p2->p_reaper = p1; 141 p2->p_reapsubtree = p->p_reapsubtree; 142 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 143 if (exiting && p2->p_pptr == p) { 144 PROC_LOCK(p2); 145 proc_reparent(p2, p1); 146 PROC_UNLOCK(p2); 147 } 148 } 149 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 150 p->p_treeflag &= ~P_TREE_REAPER; 151 } 152 153 static void 154 clear_orphan(struct proc *p) 155 { 156 struct proc *p1; 157 158 sx_assert(&proctree_lock, SA_XLOCKED); 159 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 160 return; 161 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 162 p1 = LIST_NEXT(p, p_orphan); 163 if (p1 != NULL) 164 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 165 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 166 } 167 LIST_REMOVE(p, p_orphan); 168 p->p_treeflag &= ~P_TREE_ORPHANED; 169 } 170 171 /* 172 * exit -- death of process. 173 */ 174 void 175 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 176 { 177 178 exit1(td, uap->rval, 0); 179 /* NOTREACHED */ 180 } 181 182 /* 183 * Exit: deallocate address space and other resources, change proc state to 184 * zombie, and unlink proc from allproc and parent's lists. Save exit status 185 * and rusage for wait(). Check for child processes and orphan them. 186 */ 187 void 188 exit1(struct thread *td, int rval, int signo) 189 { 190 struct proc *p, *nq, *q, *t; 191 struct thread *tdt; 192 193 mtx_assert(&Giant, MA_NOTOWNED); 194 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 195 196 p = td->td_proc; 197 /* 198 * XXX in case we're rebooting we just let init die in order to 199 * work around an unsolved stack overflow seen very late during 200 * shutdown on sparc64 when the gmirror worker process exists. 201 */ 202 if (p == initproc && rebooting == 0) { 203 printf("init died (signal %d, exit %d)\n", signo, rval); 204 panic("Going nowhere without my init!"); 205 } 206 207 /* 208 * Deref SU mp, since the thread does not return to userspace. 209 */ 210 if (softdep_ast_cleanup != NULL) 211 softdep_ast_cleanup(); 212 213 /* 214 * MUST abort all other threads before proceeding past here. 215 */ 216 PROC_LOCK(p); 217 /* 218 * First check if some other thread or external request got 219 * here before us. If so, act appropriately: exit or suspend. 220 * We must ensure that stop requests are handled before we set 221 * P_WEXIT. 222 */ 223 thread_suspend_check(0); 224 while (p->p_flag & P_HADTHREADS) { 225 /* 226 * Kill off the other threads. This requires 227 * some co-operation from other parts of the kernel 228 * so it may not be instantaneous. With this state set 229 * any thread entering the kernel from userspace will 230 * thread_exit() in trap(). Any thread attempting to 231 * sleep will return immediately with EINTR or EWOULDBLOCK 232 * which will hopefully force them to back out to userland 233 * freeing resources as they go. Any thread attempting 234 * to return to userland will thread_exit() from userret(). 235 * thread_exit() will unsuspend us when the last of the 236 * other threads exits. 237 * If there is already a thread singler after resumption, 238 * calling thread_single will fail; in that case, we just 239 * re-check all suspension request, the thread should 240 * either be suspended there or exit. 241 */ 242 if (!thread_single(p, SINGLE_EXIT)) 243 /* 244 * All other activity in this process is now 245 * stopped. Threading support has been turned 246 * off. 247 */ 248 break; 249 /* 250 * Recheck for new stop or suspend requests which 251 * might appear while process lock was dropped in 252 * thread_single(). 253 */ 254 thread_suspend_check(0); 255 } 256 KASSERT(p->p_numthreads == 1, 257 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 258 racct_sub(p, RACCT_NTHR, 1); 259 260 /* Let event handler change exit status */ 261 p->p_xexit = rval; 262 p->p_xsig = signo; 263 264 /* 265 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 266 * on our vmspace, so we should block below until they have 267 * released their reference to us. Note that if they have 268 * requested S_EXIT stops we will block here until they ack 269 * via PIOCCONT. 270 */ 271 _STOPEVENT(p, S_EXIT, 0); 272 273 /* 274 * Ignore any pending request to stop due to a stop signal. 275 * Once P_WEXIT is set, future requests will be ignored as 276 * well. 277 */ 278 p->p_flag &= ~P_STOPPED_SIG; 279 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 280 281 /* 282 * Note that we are exiting and do another wakeup of anyone in 283 * PIOCWAIT in case they aren't listening for S_EXIT stops or 284 * decided to wait again after we told them we are exiting. 285 */ 286 p->p_flag |= P_WEXIT; 287 wakeup(&p->p_stype); 288 289 /* 290 * Wait for any processes that have a hold on our vmspace to 291 * release their reference. 292 */ 293 while (p->p_lock > 0) 294 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 295 296 PROC_UNLOCK(p); 297 /* Drain the limit callout while we don't have the proc locked */ 298 callout_drain(&p->p_limco); 299 300 #ifdef AUDIT 301 /* 302 * The Sun BSM exit token contains two components: an exit status as 303 * passed to exit(), and a return value to indicate what sort of exit 304 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 305 * what the return value is. 306 */ 307 AUDIT_ARG_EXIT(rval, 0); 308 AUDIT_SYSCALL_EXIT(0, td); 309 #endif 310 311 /* Are we a task leader with peers? */ 312 if (p->p_peers != NULL && p == p->p_leader) { 313 mtx_lock(&ppeers_lock); 314 q = p->p_peers; 315 while (q != NULL) { 316 PROC_LOCK(q); 317 kern_psignal(q, SIGKILL); 318 PROC_UNLOCK(q); 319 q = q->p_peers; 320 } 321 while (p->p_peers != NULL) 322 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 323 mtx_unlock(&ppeers_lock); 324 } 325 326 /* 327 * Check if any loadable modules need anything done at process exit. 328 * E.g. SYSV IPC stuff. 329 * Event handler could change exit status. 330 * XXX what if one of these generates an error? 331 */ 332 EVENTHANDLER_INVOKE(process_exit, p); 333 334 /* 335 * If parent is waiting for us to exit or exec, 336 * P_PPWAIT is set; we will wakeup the parent below. 337 */ 338 PROC_LOCK(p); 339 stopprofclock(p); 340 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 341 p->p_ptevents = 0; 342 343 /* 344 * Stop the real interval timer. If the handler is currently 345 * executing, prevent it from rearming itself and let it finish. 346 */ 347 if (timevalisset(&p->p_realtimer.it_value) && 348 _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) { 349 timevalclear(&p->p_realtimer.it_interval); 350 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 351 KASSERT(!timevalisset(&p->p_realtimer.it_value), 352 ("realtime timer is still armed")); 353 } 354 355 PROC_UNLOCK(p); 356 357 umtx_thread_exit(td); 358 359 /* 360 * Reset any sigio structures pointing to us as a result of 361 * F_SETOWN with our pid. 362 */ 363 funsetownlst(&p->p_sigiolst); 364 365 /* 366 * If this process has an nlminfo data area (for lockd), release it 367 */ 368 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 369 (*nlminfo_release_p)(p); 370 371 /* 372 * Close open files and release open-file table. 373 * This may block! 374 */ 375 fdescfree(td); 376 377 /* 378 * If this thread tickled GEOM, we need to wait for the giggling to 379 * stop before we return to userland 380 */ 381 if (td->td_pflags & TDP_GEOM) 382 g_waitidle(); 383 384 /* 385 * Remove ourself from our leader's peer list and wake our leader. 386 */ 387 if (p->p_leader->p_peers != NULL) { 388 mtx_lock(&ppeers_lock); 389 if (p->p_leader->p_peers != NULL) { 390 q = p->p_leader; 391 while (q->p_peers != p) 392 q = q->p_peers; 393 q->p_peers = p->p_peers; 394 wakeup(p->p_leader); 395 } 396 mtx_unlock(&ppeers_lock); 397 } 398 399 vmspace_exit(td); 400 killjobc(); 401 (void)acct_process(td); 402 403 #ifdef KTRACE 404 ktrprocexit(td); 405 #endif 406 /* 407 * Release reference to text vnode 408 */ 409 if (p->p_textvp != NULL) { 410 vrele(p->p_textvp); 411 p->p_textvp = NULL; 412 } 413 414 /* 415 * Release our limits structure. 416 */ 417 lim_free(p->p_limit); 418 p->p_limit = NULL; 419 420 tidhash_remove(td); 421 422 /* 423 * Remove proc from allproc queue and pidhash chain. 424 * Place onto zombproc. Unlink from parent's child list. 425 */ 426 sx_xlock(&allproc_lock); 427 LIST_REMOVE(p, p_list); 428 LIST_INSERT_HEAD(&zombproc, p, p_list); 429 LIST_REMOVE(p, p_hash); 430 sx_xunlock(&allproc_lock); 431 432 /* 433 * Call machine-dependent code to release any 434 * machine-dependent resources other than the address space. 435 * The address space is released by "vmspace_exitfree(p)" in 436 * vm_waitproc(). 437 */ 438 cpu_exit(td); 439 440 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 441 442 /* 443 * Reparent all children processes: 444 * - traced ones to the original parent (or init if we are that parent) 445 * - the rest to init 446 */ 447 sx_xlock(&proctree_lock); 448 q = LIST_FIRST(&p->p_children); 449 if (q != NULL) /* only need this if any child is S_ZOMB */ 450 wakeup(q->p_reaper); 451 for (; q != NULL; q = nq) { 452 nq = LIST_NEXT(q, p_sibling); 453 PROC_LOCK(q); 454 q->p_sigparent = SIGCHLD; 455 456 if (!(q->p_flag & P_TRACED)) { 457 proc_reparent(q, q->p_reaper); 458 } else { 459 /* 460 * Traced processes are killed since their existence 461 * means someone is screwing up. 462 */ 463 t = proc_realparent(q); 464 if (t == p) { 465 proc_reparent(q, q->p_reaper); 466 } else { 467 PROC_LOCK(t); 468 proc_reparent(q, t); 469 PROC_UNLOCK(t); 470 } 471 /* 472 * Since q was found on our children list, the 473 * proc_reparent() call moved q to the orphan 474 * list due to present P_TRACED flag. Clear 475 * orphan link for q now while q is locked. 476 */ 477 clear_orphan(q); 478 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 479 q->p_ptevents = 0; 480 FOREACH_THREAD_IN_PROC(q, tdt) 481 tdt->td_dbgflags &= ~TDB_SUSPEND; 482 kern_psignal(q, SIGKILL); 483 } 484 PROC_UNLOCK(q); 485 } 486 487 /* 488 * Also get rid of our orphans. 489 */ 490 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 491 PROC_LOCK(q); 492 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 493 q->p_pid); 494 clear_orphan(q); 495 PROC_UNLOCK(q); 496 } 497 498 /* Save exit status. */ 499 PROC_LOCK(p); 500 p->p_xthread = td; 501 502 /* Tell the prison that we are gone. */ 503 prison_proc_free(p->p_ucred->cr_prison); 504 505 #ifdef KDTRACE_HOOKS 506 /* 507 * Tell the DTrace fasttrap provider about the exit if it 508 * has declared an interest. 509 */ 510 if (dtrace_fasttrap_exit) 511 dtrace_fasttrap_exit(p); 512 #endif 513 514 /* 515 * Notify interested parties of our demise. 516 */ 517 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 518 519 #ifdef KDTRACE_HOOKS 520 int reason = CLD_EXITED; 521 if (WCOREDUMP(signo)) 522 reason = CLD_DUMPED; 523 else if (WIFSIGNALED(signo)) 524 reason = CLD_KILLED; 525 SDT_PROBE1(proc, , , exit, reason); 526 #endif 527 528 /* 529 * If this is a process with a descriptor, we may not need to deliver 530 * a signal to the parent. proctree_lock is held over 531 * procdesc_exit() to serialize concurrent calls to close() and 532 * exit(). 533 */ 534 if (p->p_procdesc == NULL || procdesc_exit(p)) { 535 /* 536 * Notify parent that we're gone. If parent has the 537 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 538 * notify process 1 instead (and hope it will handle this 539 * situation). 540 */ 541 PROC_LOCK(p->p_pptr); 542 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 543 if (p->p_pptr->p_sigacts->ps_flag & 544 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 545 struct proc *pp; 546 547 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 548 pp = p->p_pptr; 549 PROC_UNLOCK(pp); 550 proc_reparent(p, p->p_reaper); 551 p->p_sigparent = SIGCHLD; 552 PROC_LOCK(p->p_pptr); 553 554 /* 555 * Notify parent, so in case he was wait(2)ing or 556 * executing waitpid(2) with our pid, he will 557 * continue. 558 */ 559 wakeup(pp); 560 } else 561 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 562 563 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 564 childproc_exited(p); 565 else if (p->p_sigparent != 0) { 566 if (p->p_sigparent == SIGCHLD) 567 childproc_exited(p); 568 else /* LINUX thread */ 569 kern_psignal(p->p_pptr, p->p_sigparent); 570 } 571 } else 572 PROC_LOCK(p->p_pptr); 573 sx_xunlock(&proctree_lock); 574 575 /* 576 * The state PRS_ZOMBIE prevents other proesses from sending 577 * signal to the process, to avoid memory leak, we free memory 578 * for signal queue at the time when the state is set. 579 */ 580 sigqueue_flush(&p->p_sigqueue); 581 sigqueue_flush(&td->td_sigqueue); 582 583 /* 584 * We have to wait until after acquiring all locks before 585 * changing p_state. We need to avoid all possible context 586 * switches (including ones from blocking on a mutex) while 587 * marked as a zombie. We also have to set the zombie state 588 * before we release the parent process' proc lock to avoid 589 * a lost wakeup. So, we first call wakeup, then we grab the 590 * sched lock, update the state, and release the parent process' 591 * proc lock. 592 */ 593 wakeup(p->p_pptr); 594 cv_broadcast(&p->p_pwait); 595 sched_exit(p->p_pptr, td); 596 PROC_SLOCK(p); 597 p->p_state = PRS_ZOMBIE; 598 PROC_UNLOCK(p->p_pptr); 599 600 /* 601 * Save our children's rusage information in our exit rusage. 602 */ 603 PROC_STATLOCK(p); 604 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 605 PROC_STATUNLOCK(p); 606 607 /* 608 * Make sure the scheduler takes this thread out of its tables etc. 609 * This will also release this thread's reference to the ucred. 610 * Other thread parts to release include pcb bits and such. 611 */ 612 thread_exit(); 613 } 614 615 616 #ifndef _SYS_SYSPROTO_H_ 617 struct abort2_args { 618 char *why; 619 int nargs; 620 void **args; 621 }; 622 #endif 623 624 int 625 sys_abort2(struct thread *td, struct abort2_args *uap) 626 { 627 struct proc *p = td->td_proc; 628 struct sbuf *sb; 629 void *uargs[16]; 630 int error, i, sig; 631 632 /* 633 * Do it right now so we can log either proper call of abort2(), or 634 * note, that invalid argument was passed. 512 is big enough to 635 * handle 16 arguments' descriptions with additional comments. 636 */ 637 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 638 sbuf_clear(sb); 639 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 640 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 641 /* 642 * Since we can't return from abort2(), send SIGKILL in cases, where 643 * abort2() was called improperly 644 */ 645 sig = SIGKILL; 646 /* Prevent from DoSes from user-space. */ 647 if (uap->nargs < 0 || uap->nargs > 16) 648 goto out; 649 if (uap->nargs > 0) { 650 if (uap->args == NULL) 651 goto out; 652 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 653 if (error != 0) 654 goto out; 655 } 656 /* 657 * Limit size of 'reason' string to 128. Will fit even when 658 * maximal number of arguments was chosen to be logged. 659 */ 660 if (uap->why != NULL) { 661 error = sbuf_copyin(sb, uap->why, 128); 662 if (error < 0) 663 goto out; 664 } else { 665 sbuf_printf(sb, "(null)"); 666 } 667 if (uap->nargs > 0) { 668 sbuf_printf(sb, "("); 669 for (i = 0;i < uap->nargs; i++) 670 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 671 sbuf_printf(sb, ")"); 672 } 673 /* 674 * Final stage: arguments were proper, string has been 675 * successfully copied from userspace, and copying pointers 676 * from user-space succeed. 677 */ 678 sig = SIGABRT; 679 out: 680 if (sig == SIGKILL) { 681 sbuf_trim(sb); 682 sbuf_printf(sb, " (Reason text inaccessible)"); 683 } 684 sbuf_cat(sb, "\n"); 685 sbuf_finish(sb); 686 log(LOG_INFO, "%s", sbuf_data(sb)); 687 sbuf_delete(sb); 688 exit1(td, 0, sig); 689 return (0); 690 } 691 692 693 #ifdef COMPAT_43 694 /* 695 * The dirty work is handled by kern_wait(). 696 */ 697 int 698 owait(struct thread *td, struct owait_args *uap __unused) 699 { 700 int error, status; 701 702 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 703 if (error == 0) 704 td->td_retval[1] = status; 705 return (error); 706 } 707 #endif /* COMPAT_43 */ 708 709 /* 710 * The dirty work is handled by kern_wait(). 711 */ 712 int 713 sys_wait4(struct thread *td, struct wait4_args *uap) 714 { 715 struct rusage ru, *rup; 716 int error, status; 717 718 if (uap->rusage != NULL) 719 rup = &ru; 720 else 721 rup = NULL; 722 error = kern_wait(td, uap->pid, &status, uap->options, rup); 723 if (uap->status != NULL && error == 0) 724 error = copyout(&status, uap->status, sizeof(status)); 725 if (uap->rusage != NULL && error == 0) 726 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 727 return (error); 728 } 729 730 int 731 sys_wait6(struct thread *td, struct wait6_args *uap) 732 { 733 struct __wrusage wru, *wrup; 734 siginfo_t si, *sip; 735 idtype_t idtype; 736 id_t id; 737 int error, status; 738 739 idtype = uap->idtype; 740 id = uap->id; 741 742 if (uap->wrusage != NULL) 743 wrup = &wru; 744 else 745 wrup = NULL; 746 747 if (uap->info != NULL) { 748 sip = &si; 749 bzero(sip, sizeof(*sip)); 750 } else 751 sip = NULL; 752 753 /* 754 * We expect all callers of wait6() to know about WEXITED and 755 * WTRAPPED. 756 */ 757 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 758 759 if (uap->status != NULL && error == 0) 760 error = copyout(&status, uap->status, sizeof(status)); 761 if (uap->wrusage != NULL && error == 0) 762 error = copyout(&wru, uap->wrusage, sizeof(wru)); 763 if (uap->info != NULL && error == 0) 764 error = copyout(&si, uap->info, sizeof(si)); 765 return (error); 766 } 767 768 /* 769 * Reap the remains of a zombie process and optionally return status and 770 * rusage. Asserts and will release both the proctree_lock and the process 771 * lock as part of its work. 772 */ 773 void 774 proc_reap(struct thread *td, struct proc *p, int *status, int options) 775 { 776 struct proc *q, *t; 777 778 sx_assert(&proctree_lock, SA_XLOCKED); 779 PROC_LOCK_ASSERT(p, MA_OWNED); 780 PROC_SLOCK_ASSERT(p, MA_OWNED); 781 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 782 783 q = td->td_proc; 784 785 PROC_SUNLOCK(p); 786 if (status) 787 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 788 if (options & WNOWAIT) { 789 /* 790 * Only poll, returning the status. Caller does not wish to 791 * release the proc struct just yet. 792 */ 793 PROC_UNLOCK(p); 794 sx_xunlock(&proctree_lock); 795 return; 796 } 797 798 PROC_LOCK(q); 799 sigqueue_take(p->p_ksi); 800 PROC_UNLOCK(q); 801 802 /* 803 * If we got the child via a ptrace 'attach', we need to give it back 804 * to the old parent. 805 */ 806 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 807 PROC_UNLOCK(p); 808 t = proc_realparent(p); 809 PROC_LOCK(t); 810 PROC_LOCK(p); 811 CTR2(KTR_PTRACE, 812 "wait: traced child %d moved back to parent %d", p->p_pid, 813 t->p_pid); 814 proc_reparent(p, t); 815 p->p_oppid = 0; 816 PROC_UNLOCK(p); 817 pksignal(t, SIGCHLD, p->p_ksi); 818 wakeup(t); 819 cv_broadcast(&p->p_pwait); 820 PROC_UNLOCK(t); 821 sx_xunlock(&proctree_lock); 822 return; 823 } 824 p->p_oppid = 0; 825 PROC_UNLOCK(p); 826 827 /* 828 * Remove other references to this process to ensure we have an 829 * exclusive reference. 830 */ 831 sx_xlock(&allproc_lock); 832 LIST_REMOVE(p, p_list); /* off zombproc */ 833 sx_xunlock(&allproc_lock); 834 LIST_REMOVE(p, p_sibling); 835 reaper_abandon_children(p, true); 836 LIST_REMOVE(p, p_reapsibling); 837 PROC_LOCK(p); 838 clear_orphan(p); 839 PROC_UNLOCK(p); 840 leavepgrp(p); 841 if (p->p_procdesc != NULL) 842 procdesc_reap(p); 843 sx_xunlock(&proctree_lock); 844 845 PROC_LOCK(p); 846 knlist_detach(p->p_klist); 847 p->p_klist = NULL; 848 PROC_UNLOCK(p); 849 850 /* 851 * Removal from allproc list and process group list paired with 852 * PROC_LOCK which was executed during that time should guarantee 853 * nothing can reach this process anymore. As such further locking 854 * is unnecessary. 855 */ 856 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 857 858 PROC_LOCK(q); 859 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 860 PROC_UNLOCK(q); 861 862 /* 863 * Decrement the count of procs running with this uid. 864 */ 865 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 866 867 /* 868 * Destroy resource accounting information associated with the process. 869 */ 870 #ifdef RACCT 871 if (racct_enable) { 872 PROC_LOCK(p); 873 racct_sub(p, RACCT_NPROC, 1); 874 PROC_UNLOCK(p); 875 } 876 #endif 877 racct_proc_exit(p); 878 879 /* 880 * Free credentials, arguments, and sigacts. 881 */ 882 crfree(p->p_ucred); 883 proc_set_cred(p, NULL); 884 pargs_drop(p->p_args); 885 p->p_args = NULL; 886 sigacts_free(p->p_sigacts); 887 p->p_sigacts = NULL; 888 889 /* 890 * Do any thread-system specific cleanups. 891 */ 892 thread_wait(p); 893 894 /* 895 * Give vm and machine-dependent layer a chance to free anything that 896 * cpu_exit couldn't release while still running in process context. 897 */ 898 vm_waitproc(p); 899 #ifdef MAC 900 mac_proc_destroy(p); 901 #endif 902 /* 903 * Free any domain policy that's still hiding around. 904 */ 905 vm_domain_policy_cleanup(&p->p_vm_dom_policy); 906 907 KASSERT(FIRST_THREAD_IN_PROC(p), 908 ("proc_reap: no residual thread!")); 909 uma_zfree(proc_zone, p); 910 atomic_add_int(&nprocs, -1); 911 } 912 913 static int 914 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 915 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 916 int check_only) 917 { 918 struct rusage *rup; 919 920 sx_assert(&proctree_lock, SA_XLOCKED); 921 922 PROC_LOCK(p); 923 924 switch (idtype) { 925 case P_ALL: 926 if (p->p_procdesc != NULL) { 927 PROC_UNLOCK(p); 928 return (0); 929 } 930 break; 931 case P_PID: 932 if (p->p_pid != (pid_t)id) { 933 PROC_UNLOCK(p); 934 return (0); 935 } 936 break; 937 case P_PGID: 938 if (p->p_pgid != (pid_t)id) { 939 PROC_UNLOCK(p); 940 return (0); 941 } 942 break; 943 case P_SID: 944 if (p->p_session->s_sid != (pid_t)id) { 945 PROC_UNLOCK(p); 946 return (0); 947 } 948 break; 949 case P_UID: 950 if (p->p_ucred->cr_uid != (uid_t)id) { 951 PROC_UNLOCK(p); 952 return (0); 953 } 954 break; 955 case P_GID: 956 if (p->p_ucred->cr_gid != (gid_t)id) { 957 PROC_UNLOCK(p); 958 return (0); 959 } 960 break; 961 case P_JAILID: 962 if (p->p_ucred->cr_prison->pr_id != (int)id) { 963 PROC_UNLOCK(p); 964 return (0); 965 } 966 break; 967 /* 968 * It seems that the thread structures get zeroed out 969 * at process exit. This makes it impossible to 970 * support P_SETID, P_CID or P_CPUID. 971 */ 972 default: 973 PROC_UNLOCK(p); 974 return (0); 975 } 976 977 if (p_canwait(td, p)) { 978 PROC_UNLOCK(p); 979 return (0); 980 } 981 982 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 983 PROC_UNLOCK(p); 984 return (0); 985 } 986 987 /* 988 * This special case handles a kthread spawned by linux_clone 989 * (see linux_misc.c). The linux_wait4 and linux_waitpid 990 * functions need to be able to distinguish between waiting 991 * on a process and waiting on a thread. It is a thread if 992 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 993 * signifies we want to wait for threads and not processes. 994 */ 995 if ((p->p_sigparent != SIGCHLD) ^ 996 ((options & WLINUXCLONE) != 0)) { 997 PROC_UNLOCK(p); 998 return (0); 999 } 1000 1001 if (siginfo != NULL) { 1002 bzero(siginfo, sizeof(*siginfo)); 1003 siginfo->si_errno = 0; 1004 1005 /* 1006 * SUSv4 requires that the si_signo value is always 1007 * SIGCHLD. Obey it despite the rfork(2) interface 1008 * allows to request other signal for child exit 1009 * notification. 1010 */ 1011 siginfo->si_signo = SIGCHLD; 1012 1013 /* 1014 * This is still a rough estimate. We will fix the 1015 * cases TRAPPED, STOPPED, and CONTINUED later. 1016 */ 1017 if (WCOREDUMP(p->p_xsig)) { 1018 siginfo->si_code = CLD_DUMPED; 1019 siginfo->si_status = WTERMSIG(p->p_xsig); 1020 } else if (WIFSIGNALED(p->p_xsig)) { 1021 siginfo->si_code = CLD_KILLED; 1022 siginfo->si_status = WTERMSIG(p->p_xsig); 1023 } else { 1024 siginfo->si_code = CLD_EXITED; 1025 siginfo->si_status = p->p_xexit; 1026 } 1027 1028 siginfo->si_pid = p->p_pid; 1029 siginfo->si_uid = p->p_ucred->cr_uid; 1030 1031 /* 1032 * The si_addr field would be useful additional 1033 * detail, but apparently the PC value may be lost 1034 * when we reach this point. bzero() above sets 1035 * siginfo->si_addr to NULL. 1036 */ 1037 } 1038 1039 /* 1040 * There should be no reason to limit resources usage info to 1041 * exited processes only. A snapshot about any resources used 1042 * by a stopped process may be exactly what is needed. 1043 */ 1044 if (wrusage != NULL) { 1045 rup = &wrusage->wru_self; 1046 *rup = p->p_ru; 1047 PROC_STATLOCK(p); 1048 calcru(p, &rup->ru_utime, &rup->ru_stime); 1049 PROC_STATUNLOCK(p); 1050 1051 rup = &wrusage->wru_children; 1052 *rup = p->p_stats->p_cru; 1053 calccru(p, &rup->ru_utime, &rup->ru_stime); 1054 } 1055 1056 if (p->p_state == PRS_ZOMBIE && !check_only) { 1057 PROC_SLOCK(p); 1058 proc_reap(td, p, status, options); 1059 return (-1); 1060 } 1061 PROC_UNLOCK(p); 1062 return (1); 1063 } 1064 1065 int 1066 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1067 struct rusage *rusage) 1068 { 1069 struct __wrusage wru, *wrup; 1070 idtype_t idtype; 1071 id_t id; 1072 int ret; 1073 1074 /* 1075 * Translate the special pid values into the (idtype, pid) 1076 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1077 * kern_wait6() on its own. 1078 */ 1079 if (pid == WAIT_ANY) { 1080 idtype = P_ALL; 1081 id = 0; 1082 } else if (pid < 0) { 1083 idtype = P_PGID; 1084 id = (id_t)-pid; 1085 } else { 1086 idtype = P_PID; 1087 id = (id_t)pid; 1088 } 1089 1090 if (rusage != NULL) 1091 wrup = &wru; 1092 else 1093 wrup = NULL; 1094 1095 /* 1096 * For backward compatibility we implicitly add flags WEXITED 1097 * and WTRAPPED here. 1098 */ 1099 options |= WEXITED | WTRAPPED; 1100 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1101 if (rusage != NULL) 1102 *rusage = wru.wru_self; 1103 return (ret); 1104 } 1105 1106 int 1107 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1108 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1109 { 1110 struct proc *p, *q; 1111 pid_t pid; 1112 int error, nfound, ret; 1113 1114 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1115 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1116 AUDIT_ARG_VALUE(options); 1117 1118 q = td->td_proc; 1119 1120 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1121 PROC_LOCK(q); 1122 id = (id_t)q->p_pgid; 1123 PROC_UNLOCK(q); 1124 idtype = P_PGID; 1125 } 1126 1127 /* If we don't know the option, just return. */ 1128 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1129 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1130 return (EINVAL); 1131 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1132 /* 1133 * We will be unable to find any matching processes, 1134 * because there are no known events to look for. 1135 * Prefer to return error instead of blocking 1136 * indefinitely. 1137 */ 1138 return (EINVAL); 1139 } 1140 1141 loop: 1142 if (q->p_flag & P_STATCHILD) { 1143 PROC_LOCK(q); 1144 q->p_flag &= ~P_STATCHILD; 1145 PROC_UNLOCK(q); 1146 } 1147 nfound = 0; 1148 sx_xlock(&proctree_lock); 1149 LIST_FOREACH(p, &q->p_children, p_sibling) { 1150 pid = p->p_pid; 1151 ret = proc_to_reap(td, p, idtype, id, status, options, 1152 wrusage, siginfo, 0); 1153 if (ret == 0) 1154 continue; 1155 else if (ret == 1) 1156 nfound++; 1157 else { 1158 td->td_retval[0] = pid; 1159 return (0); 1160 } 1161 1162 PROC_LOCK(p); 1163 PROC_SLOCK(p); 1164 1165 if ((options & WTRAPPED) != 0 && 1166 (p->p_flag & P_TRACED) != 0 && 1167 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1168 (p->p_suspcount == p->p_numthreads) && 1169 ((p->p_flag & P_WAITED) == 0)) { 1170 PROC_SUNLOCK(p); 1171 if ((options & WNOWAIT) == 0) 1172 p->p_flag |= P_WAITED; 1173 sx_xunlock(&proctree_lock); 1174 1175 if (status != NULL) 1176 *status = W_STOPCODE(p->p_xsig); 1177 if (siginfo != NULL) { 1178 siginfo->si_status = p->p_xsig; 1179 siginfo->si_code = CLD_TRAPPED; 1180 } 1181 if ((options & WNOWAIT) == 0) { 1182 PROC_LOCK(q); 1183 sigqueue_take(p->p_ksi); 1184 PROC_UNLOCK(q); 1185 } 1186 1187 CTR4(KTR_PTRACE, 1188 "wait: returning trapped pid %d status %#x (xstat %d) xthread %d", 1189 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1190 p->p_xthread != NULL ? p->p_xthread->td_tid : -1); 1191 PROC_UNLOCK(p); 1192 td->td_retval[0] = pid; 1193 return (0); 1194 } 1195 if ((options & WUNTRACED) != 0 && 1196 (p->p_flag & P_STOPPED_SIG) != 0 && 1197 (p->p_suspcount == p->p_numthreads) && 1198 ((p->p_flag & P_WAITED) == 0)) { 1199 PROC_SUNLOCK(p); 1200 if ((options & WNOWAIT) == 0) 1201 p->p_flag |= P_WAITED; 1202 sx_xunlock(&proctree_lock); 1203 1204 if (status != NULL) 1205 *status = W_STOPCODE(p->p_xsig); 1206 if (siginfo != NULL) { 1207 siginfo->si_status = p->p_xsig; 1208 siginfo->si_code = CLD_STOPPED; 1209 } 1210 if ((options & WNOWAIT) == 0) { 1211 PROC_LOCK(q); 1212 sigqueue_take(p->p_ksi); 1213 PROC_UNLOCK(q); 1214 } 1215 1216 PROC_UNLOCK(p); 1217 td->td_retval[0] = pid; 1218 return (0); 1219 } 1220 PROC_SUNLOCK(p); 1221 if ((options & WCONTINUED) != 0 && 1222 (p->p_flag & P_CONTINUED) != 0) { 1223 sx_xunlock(&proctree_lock); 1224 if ((options & WNOWAIT) == 0) { 1225 p->p_flag &= ~P_CONTINUED; 1226 PROC_LOCK(q); 1227 sigqueue_take(p->p_ksi); 1228 PROC_UNLOCK(q); 1229 } 1230 PROC_UNLOCK(p); 1231 1232 if (status != NULL) 1233 *status = SIGCONT; 1234 if (siginfo != NULL) { 1235 siginfo->si_status = SIGCONT; 1236 siginfo->si_code = CLD_CONTINUED; 1237 } 1238 td->td_retval[0] = pid; 1239 return (0); 1240 } 1241 PROC_UNLOCK(p); 1242 } 1243 1244 /* 1245 * Look in the orphans list too, to allow the parent to 1246 * collect it's child exit status even if child is being 1247 * debugged. 1248 * 1249 * Debugger detaches from the parent upon successful 1250 * switch-over from parent to child. At this point due to 1251 * re-parenting the parent loses the child to debugger and a 1252 * wait4(2) call would report that it has no children to wait 1253 * for. By maintaining a list of orphans we allow the parent 1254 * to successfully wait until the child becomes a zombie. 1255 */ 1256 if (nfound == 0) { 1257 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1258 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1259 NULL, NULL, 1); 1260 if (ret != 0) { 1261 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1262 (int)td->td_retval[0])); 1263 nfound++; 1264 break; 1265 } 1266 } 1267 } 1268 if (nfound == 0) { 1269 sx_xunlock(&proctree_lock); 1270 return (ECHILD); 1271 } 1272 if (options & WNOHANG) { 1273 sx_xunlock(&proctree_lock); 1274 td->td_retval[0] = 0; 1275 return (0); 1276 } 1277 PROC_LOCK(q); 1278 sx_xunlock(&proctree_lock); 1279 if (q->p_flag & P_STATCHILD) { 1280 q->p_flag &= ~P_STATCHILD; 1281 error = 0; 1282 } else 1283 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1284 PROC_UNLOCK(q); 1285 if (error) 1286 return (error); 1287 goto loop; 1288 } 1289 1290 /* 1291 * Make process 'parent' the new parent of process 'child'. 1292 * Must be called with an exclusive hold of proctree lock. 1293 */ 1294 void 1295 proc_reparent(struct proc *child, struct proc *parent) 1296 { 1297 1298 sx_assert(&proctree_lock, SX_XLOCKED); 1299 PROC_LOCK_ASSERT(child, MA_OWNED); 1300 if (child->p_pptr == parent) 1301 return; 1302 1303 PROC_LOCK(child->p_pptr); 1304 sigqueue_take(child->p_ksi); 1305 PROC_UNLOCK(child->p_pptr); 1306 LIST_REMOVE(child, p_sibling); 1307 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1308 1309 clear_orphan(child); 1310 if (child->p_flag & P_TRACED) { 1311 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1312 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1313 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1314 p_orphan); 1315 } else { 1316 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1317 child, p_orphan); 1318 } 1319 child->p_treeflag |= P_TREE_ORPHANED; 1320 } 1321 1322 child->p_pptr = parent; 1323 } 1324