1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_exit.c 8.7 (Berkeley) 2/12/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_ktrace.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/sysproto.h> 47 #include <sys/capsicum.h> 48 #include <sys/eventhandler.h> 49 #include <sys/kernel.h> 50 #include <sys/ktr.h> 51 #include <sys/malloc.h> 52 #include <sys/lock.h> 53 #include <sys/mutex.h> 54 #include <sys/proc.h> 55 #include <sys/procdesc.h> 56 #include <sys/jail.h> 57 #include <sys/tty.h> 58 #include <sys/wait.h> 59 #include <sys/vmmeter.h> 60 #include <sys/vnode.h> 61 #include <sys/racct.h> 62 #include <sys/resourcevar.h> 63 #include <sys/sbuf.h> 64 #include <sys/signalvar.h> 65 #include <sys/sched.h> 66 #include <sys/sx.h> 67 #include <sys/syscallsubr.h> 68 #include <sys/syslog.h> 69 #include <sys/ptrace.h> 70 #include <sys/acct.h> /* for acct_process() function prototype */ 71 #include <sys/filedesc.h> 72 #include <sys/sdt.h> 73 #include <sys/shm.h> 74 #include <sys/sem.h> 75 #include <sys/umtx.h> 76 #ifdef KTRACE 77 #include <sys/ktrace.h> 78 #endif 79 80 #include <security/audit/audit.h> 81 #include <security/mac/mac_framework.h> 82 83 #include <vm/vm.h> 84 #include <vm/vm_extern.h> 85 #include <vm/vm_param.h> 86 #include <vm/pmap.h> 87 #include <vm/vm_map.h> 88 #include <vm/vm_page.h> 89 #include <vm/uma.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 struct proc * 100 proc_realparent(struct proc *child) 101 { 102 struct proc *p, *parent; 103 104 sx_assert(&proctree_lock, SX_LOCKED); 105 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) 106 return (child->p_pptr->p_pid == child->p_oppid ? 107 child->p_pptr : initproc); 108 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 109 /* Cannot use LIST_PREV(), since the list head is not known. */ 110 p = __containerof(p->p_orphan.le_prev, struct proc, 111 p_orphan.le_next); 112 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 113 ("missing P_ORPHAN %p", p)); 114 } 115 parent = __containerof(p->p_orphan.le_prev, struct proc, 116 p_orphans.lh_first); 117 return (parent); 118 } 119 120 void 121 reaper_abandon_children(struct proc *p, bool exiting) 122 { 123 struct proc *p1, *p2, *ptmp; 124 125 sx_assert(&proctree_lock, SX_LOCKED); 126 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 127 if ((p->p_treeflag & P_TREE_REAPER) == 0) 128 return; 129 p1 = p->p_reaper; 130 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 131 LIST_REMOVE(p2, p_reapsibling); 132 p2->p_reaper = p1; 133 p2->p_reapsubtree = p->p_reapsubtree; 134 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 135 if (exiting && p2->p_pptr == p) { 136 PROC_LOCK(p2); 137 proc_reparent(p2, p1, true); 138 PROC_UNLOCK(p2); 139 } 140 } 141 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 142 p->p_treeflag &= ~P_TREE_REAPER; 143 } 144 145 static void 146 reaper_clear(struct proc *p) 147 { 148 struct proc *p1; 149 bool clear; 150 151 sx_assert(&proctree_lock, SX_LOCKED); 152 LIST_REMOVE(p, p_reapsibling); 153 if (p->p_reapsubtree == 1) 154 return; 155 clear = true; 156 LIST_FOREACH(p1, &p->p_reaper->p_reaplist, p_reapsibling) { 157 if (p1->p_reapsubtree == p->p_reapsubtree) { 158 clear = false; 159 break; 160 } 161 } 162 if (clear) 163 proc_id_clear(PROC_ID_REAP, p->p_reapsubtree); 164 } 165 166 void 167 proc_clear_orphan(struct proc *p) 168 { 169 struct proc *p1; 170 171 sx_assert(&proctree_lock, SA_XLOCKED); 172 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 173 return; 174 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 175 p1 = LIST_NEXT(p, p_orphan); 176 if (p1 != NULL) 177 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 178 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 179 } 180 LIST_REMOVE(p, p_orphan); 181 p->p_treeflag &= ~P_TREE_ORPHANED; 182 } 183 184 /* 185 * exit -- death of process. 186 */ 187 void 188 sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 189 { 190 191 exit1(td, uap->rval, 0); 192 /* NOTREACHED */ 193 } 194 195 /* 196 * Exit: deallocate address space and other resources, change proc state to 197 * zombie, and unlink proc from allproc and parent's lists. Save exit status 198 * and rusage for wait(). Check for child processes and orphan them. 199 */ 200 void 201 exit1(struct thread *td, int rval, int signo) 202 { 203 struct proc *p, *nq, *q, *t; 204 struct thread *tdt; 205 ksiginfo_t *ksi, *ksi1; 206 int signal_parent; 207 208 mtx_assert(&Giant, MA_NOTOWNED); 209 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 210 211 p = td->td_proc; 212 /* 213 * XXX in case we're rebooting we just let init die in order to 214 * work around an unsolved stack overflow seen very late during 215 * shutdown on sparc64 when the gmirror worker process exists. 216 * XXX what to do now that sparc64 is gone... remove if? 217 */ 218 if (p == initproc && rebooting == 0) { 219 printf("init died (signal %d, exit %d)\n", signo, rval); 220 panic("Going nowhere without my init!"); 221 } 222 223 /* 224 * Deref SU mp, since the thread does not return to userspace. 225 */ 226 td_softdep_cleanup(td); 227 228 /* 229 * MUST abort all other threads before proceeding past here. 230 */ 231 PROC_LOCK(p); 232 /* 233 * First check if some other thread or external request got 234 * here before us. If so, act appropriately: exit or suspend. 235 * We must ensure that stop requests are handled before we set 236 * P_WEXIT. 237 */ 238 thread_suspend_check(0); 239 while (p->p_flag & P_HADTHREADS) { 240 /* 241 * Kill off the other threads. This requires 242 * some co-operation from other parts of the kernel 243 * so it may not be instantaneous. With this state set 244 * any thread entering the kernel from userspace will 245 * thread_exit() in trap(). Any thread attempting to 246 * sleep will return immediately with EINTR or EWOULDBLOCK 247 * which will hopefully force them to back out to userland 248 * freeing resources as they go. Any thread attempting 249 * to return to userland will thread_exit() from userret(). 250 * thread_exit() will unsuspend us when the last of the 251 * other threads exits. 252 * If there is already a thread singler after resumption, 253 * calling thread_single will fail; in that case, we just 254 * re-check all suspension request, the thread should 255 * either be suspended there or exit. 256 */ 257 if (!thread_single(p, SINGLE_EXIT)) 258 /* 259 * All other activity in this process is now 260 * stopped. Threading support has been turned 261 * off. 262 */ 263 break; 264 /* 265 * Recheck for new stop or suspend requests which 266 * might appear while process lock was dropped in 267 * thread_single(). 268 */ 269 thread_suspend_check(0); 270 } 271 KASSERT(p->p_numthreads == 1, 272 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 273 racct_sub(p, RACCT_NTHR, 1); 274 275 /* Let event handler change exit status */ 276 p->p_xexit = rval; 277 p->p_xsig = signo; 278 279 /* 280 * Ignore any pending request to stop due to a stop signal. 281 * Once P_WEXIT is set, future requests will be ignored as 282 * well. 283 */ 284 p->p_flag &= ~P_STOPPED_SIG; 285 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 286 287 /* Note that we are exiting. */ 288 p->p_flag |= P_WEXIT; 289 290 /* 291 * Wait for any processes that have a hold on our vmspace to 292 * release their reference. 293 */ 294 while (p->p_lock > 0) 295 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 296 297 PROC_UNLOCK(p); 298 /* Drain the limit callout while we don't have the proc locked */ 299 callout_drain(&p->p_limco); 300 301 #ifdef AUDIT 302 /* 303 * The Sun BSM exit token contains two components: an exit status as 304 * passed to exit(), and a return value to indicate what sort of exit 305 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 306 * what the return value is. 307 */ 308 AUDIT_ARG_EXIT(rval, 0); 309 AUDIT_SYSCALL_EXIT(0, td); 310 #endif 311 312 /* Are we a task leader with peers? */ 313 if (p->p_peers != NULL && p == p->p_leader) { 314 mtx_lock(&ppeers_lock); 315 q = p->p_peers; 316 while (q != NULL) { 317 PROC_LOCK(q); 318 kern_psignal(q, SIGKILL); 319 PROC_UNLOCK(q); 320 q = q->p_peers; 321 } 322 while (p->p_peers != NULL) 323 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 324 mtx_unlock(&ppeers_lock); 325 } 326 327 /* 328 * Check if any loadable modules need anything done at process exit. 329 * E.g. SYSV IPC stuff. 330 * Event handler could change exit status. 331 * XXX what if one of these generates an error? 332 */ 333 EVENTHANDLER_DIRECT_INVOKE(process_exit, p); 334 335 /* 336 * If parent is waiting for us to exit or exec, 337 * P_PPWAIT is set; we will wakeup the parent below. 338 */ 339 PROC_LOCK(p); 340 stopprofclock(p); 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 * Close open files and release open-file table. 367 * This may block! 368 */ 369 fdescfree(td); 370 371 /* 372 * If this thread tickled GEOM, we need to wait for the giggling to 373 * stop before we return to userland 374 */ 375 if (td->td_pflags & TDP_GEOM) 376 g_waitidle(); 377 378 /* 379 * Remove ourself from our leader's peer list and wake our leader. 380 */ 381 if (p->p_leader->p_peers != NULL) { 382 mtx_lock(&ppeers_lock); 383 if (p->p_leader->p_peers != NULL) { 384 q = p->p_leader; 385 while (q->p_peers != p) 386 q = q->p_peers; 387 q->p_peers = p->p_peers; 388 wakeup(p->p_leader); 389 } 390 mtx_unlock(&ppeers_lock); 391 } 392 393 vmspace_exit(td); 394 killjobc(); 395 (void)acct_process(td); 396 397 #ifdef KTRACE 398 ktrprocexit(td); 399 #endif 400 /* 401 * Release reference to text vnode 402 */ 403 if (p->p_textvp != NULL) { 404 vrele(p->p_textvp); 405 p->p_textvp = NULL; 406 } 407 408 /* 409 * Release our limits structure. 410 */ 411 lim_free(p->p_limit); 412 p->p_limit = NULL; 413 414 tidhash_remove(td); 415 416 /* 417 * Call machine-dependent code to release any 418 * machine-dependent resources other than the address space. 419 * The address space is released by "vmspace_exitfree(p)" in 420 * vm_waitproc(). 421 */ 422 cpu_exit(td); 423 424 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 425 426 /* 427 * Remove from allproc. It still sits in the hash. 428 */ 429 sx_xlock(&allproc_lock); 430 LIST_REMOVE(p, p_list); 431 sx_xunlock(&allproc_lock); 432 433 sx_xlock(&proctree_lock); 434 PROC_LOCK(p); 435 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 436 PROC_UNLOCK(p); 437 438 /* 439 * Reparent all children processes: 440 * - traced ones to the original parent (or init if we are that parent) 441 * - the rest to init 442 */ 443 q = LIST_FIRST(&p->p_children); 444 if (q != NULL) /* only need this if any child is S_ZOMB */ 445 wakeup(q->p_reaper); 446 for (; q != NULL; q = nq) { 447 nq = LIST_NEXT(q, p_sibling); 448 ksi = ksiginfo_alloc(TRUE); 449 PROC_LOCK(q); 450 q->p_sigparent = SIGCHLD; 451 452 if ((q->p_flag & P_TRACED) == 0) { 453 proc_reparent(q, q->p_reaper, true); 454 if (q->p_state == PRS_ZOMBIE) { 455 /* 456 * Inform reaper about the reparented 457 * zombie, since wait(2) has something 458 * new to report. Guarantee queueing 459 * of the SIGCHLD signal, similar to 460 * the _exit() behaviour, by providing 461 * our ksiginfo. Ksi is freed by the 462 * signal delivery. 463 */ 464 if (q->p_ksi == NULL) { 465 ksi1 = NULL; 466 } else { 467 ksiginfo_copy(q->p_ksi, ksi); 468 ksi->ksi_flags |= KSI_INS; 469 ksi1 = ksi; 470 ksi = NULL; 471 } 472 PROC_LOCK(q->p_reaper); 473 pksignal(q->p_reaper, SIGCHLD, ksi1); 474 PROC_UNLOCK(q->p_reaper); 475 } else if (q->p_pdeathsig > 0) { 476 /* 477 * The child asked to received a signal 478 * when we exit. 479 */ 480 kern_psignal(q, q->p_pdeathsig); 481 } 482 } else { 483 /* 484 * Traced processes are killed since their existence 485 * means someone is screwing up. 486 */ 487 t = proc_realparent(q); 488 if (t == p) { 489 proc_reparent(q, q->p_reaper, true); 490 } else { 491 PROC_LOCK(t); 492 proc_reparent(q, t, true); 493 PROC_UNLOCK(t); 494 } 495 /* 496 * Since q was found on our children list, the 497 * proc_reparent() call moved q to the orphan 498 * list due to present P_TRACED flag. Clear 499 * orphan link for q now while q is locked. 500 */ 501 proc_clear_orphan(q); 502 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 503 q->p_flag2 &= ~P2_PTRACE_FSTP; 504 q->p_ptevents = 0; 505 FOREACH_THREAD_IN_PROC(q, tdt) { 506 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 507 TDB_FSTP); 508 } 509 kern_psignal(q, SIGKILL); 510 } 511 PROC_UNLOCK(q); 512 if (ksi != NULL) 513 ksiginfo_free(ksi); 514 } 515 516 /* 517 * Also get rid of our orphans. 518 */ 519 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 520 PROC_LOCK(q); 521 KASSERT(q->p_oppid == p->p_pid, 522 ("orphan %p of %p has unexpected oppid %d", q, p, 523 q->p_oppid)); 524 q->p_oppid = q->p_reaper->p_pid; 525 526 /* 527 * If we are the real parent of this process 528 * but it has been reparented to a debugger, then 529 * check if it asked for a signal when we exit. 530 */ 531 if (q->p_pdeathsig > 0) 532 kern_psignal(q, q->p_pdeathsig); 533 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 534 q->p_pid); 535 proc_clear_orphan(q); 536 PROC_UNLOCK(q); 537 } 538 539 #ifdef KDTRACE_HOOKS 540 if (SDT_PROBES_ENABLED()) { 541 int reason = CLD_EXITED; 542 if (WCOREDUMP(signo)) 543 reason = CLD_DUMPED; 544 else if (WIFSIGNALED(signo)) 545 reason = CLD_KILLED; 546 SDT_PROBE1(proc, , , exit, reason); 547 } 548 #endif 549 550 /* Save exit status. */ 551 PROC_LOCK(p); 552 p->p_xthread = td; 553 554 #ifdef KDTRACE_HOOKS 555 /* 556 * Tell the DTrace fasttrap provider about the exit if it 557 * has declared an interest. 558 */ 559 if (dtrace_fasttrap_exit) 560 dtrace_fasttrap_exit(p); 561 #endif 562 563 /* 564 * Notify interested parties of our demise. 565 */ 566 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 567 568 /* 569 * If this is a process with a descriptor, we may not need to deliver 570 * a signal to the parent. proctree_lock is held over 571 * procdesc_exit() to serialize concurrent calls to close() and 572 * exit(). 573 */ 574 signal_parent = 0; 575 if (p->p_procdesc == NULL || procdesc_exit(p)) { 576 /* 577 * Notify parent that we're gone. If parent has the 578 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 579 * notify process 1 instead (and hope it will handle this 580 * situation). 581 */ 582 PROC_LOCK(p->p_pptr); 583 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 584 if (p->p_pptr->p_sigacts->ps_flag & 585 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 586 struct proc *pp; 587 588 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 589 pp = p->p_pptr; 590 PROC_UNLOCK(pp); 591 proc_reparent(p, p->p_reaper, true); 592 p->p_sigparent = SIGCHLD; 593 PROC_LOCK(p->p_pptr); 594 595 /* 596 * Notify parent, so in case he was wait(2)ing or 597 * executing waitpid(2) with our pid, he will 598 * continue. 599 */ 600 wakeup(pp); 601 } else 602 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 603 604 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) { 605 signal_parent = 1; 606 } else if (p->p_sigparent != 0) { 607 if (p->p_sigparent == SIGCHLD) { 608 signal_parent = 1; 609 } else { /* LINUX thread */ 610 signal_parent = 2; 611 } 612 } 613 } else 614 PROC_LOCK(p->p_pptr); 615 sx_xunlock(&proctree_lock); 616 617 if (signal_parent == 1) { 618 childproc_exited(p); 619 } else if (signal_parent == 2) { 620 kern_psignal(p->p_pptr, p->p_sigparent); 621 } 622 623 /* Tell the prison that we are gone. */ 624 prison_proc_free(p->p_ucred->cr_prison); 625 626 /* 627 * The state PRS_ZOMBIE prevents other proesses from sending 628 * signal to the process, to avoid memory leak, we free memory 629 * for signal queue at the time when the state is set. 630 */ 631 sigqueue_flush(&p->p_sigqueue); 632 sigqueue_flush(&td->td_sigqueue); 633 634 /* 635 * We have to wait until after acquiring all locks before 636 * changing p_state. We need to avoid all possible context 637 * switches (including ones from blocking on a mutex) while 638 * marked as a zombie. We also have to set the zombie state 639 * before we release the parent process' proc lock to avoid 640 * a lost wakeup. So, we first call wakeup, then we grab the 641 * sched lock, update the state, and release the parent process' 642 * proc lock. 643 */ 644 wakeup(p->p_pptr); 645 cv_broadcast(&p->p_pwait); 646 sched_exit(p->p_pptr, td); 647 PROC_SLOCK(p); 648 p->p_state = PRS_ZOMBIE; 649 PROC_UNLOCK(p->p_pptr); 650 651 /* 652 * Save our children's rusage information in our exit rusage. 653 */ 654 PROC_STATLOCK(p); 655 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 656 PROC_STATUNLOCK(p); 657 658 /* 659 * Make sure the scheduler takes this thread out of its tables etc. 660 * This will also release this thread's reference to the ucred. 661 * Other thread parts to release include pcb bits and such. 662 */ 663 thread_exit(); 664 } 665 666 #ifndef _SYS_SYSPROTO_H_ 667 struct abort2_args { 668 char *why; 669 int nargs; 670 void **args; 671 }; 672 #endif 673 674 int 675 sys_abort2(struct thread *td, struct abort2_args *uap) 676 { 677 struct proc *p = td->td_proc; 678 struct sbuf *sb; 679 void *uargs[16]; 680 int error, i, sig; 681 682 /* 683 * Do it right now so we can log either proper call of abort2(), or 684 * note, that invalid argument was passed. 512 is big enough to 685 * handle 16 arguments' descriptions with additional comments. 686 */ 687 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 688 sbuf_clear(sb); 689 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 690 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 691 /* 692 * Since we can't return from abort2(), send SIGKILL in cases, where 693 * abort2() was called improperly 694 */ 695 sig = SIGKILL; 696 /* Prevent from DoSes from user-space. */ 697 if (uap->nargs < 0 || uap->nargs > 16) 698 goto out; 699 if (uap->nargs > 0) { 700 if (uap->args == NULL) 701 goto out; 702 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 703 if (error != 0) 704 goto out; 705 } 706 /* 707 * Limit size of 'reason' string to 128. Will fit even when 708 * maximal number of arguments was chosen to be logged. 709 */ 710 if (uap->why != NULL) { 711 error = sbuf_copyin(sb, uap->why, 128); 712 if (error < 0) 713 goto out; 714 } else { 715 sbuf_printf(sb, "(null)"); 716 } 717 if (uap->nargs > 0) { 718 sbuf_printf(sb, "("); 719 for (i = 0;i < uap->nargs; i++) 720 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 721 sbuf_printf(sb, ")"); 722 } 723 /* 724 * Final stage: arguments were proper, string has been 725 * successfully copied from userspace, and copying pointers 726 * from user-space succeed. 727 */ 728 sig = SIGABRT; 729 out: 730 if (sig == SIGKILL) { 731 sbuf_trim(sb); 732 sbuf_printf(sb, " (Reason text inaccessible)"); 733 } 734 sbuf_cat(sb, "\n"); 735 sbuf_finish(sb); 736 log(LOG_INFO, "%s", sbuf_data(sb)); 737 sbuf_delete(sb); 738 exit1(td, 0, sig); 739 return (0); 740 } 741 742 #ifdef COMPAT_43 743 /* 744 * The dirty work is handled by kern_wait(). 745 */ 746 int 747 owait(struct thread *td, struct owait_args *uap __unused) 748 { 749 int error, status; 750 751 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 752 if (error == 0) 753 td->td_retval[1] = status; 754 return (error); 755 } 756 #endif /* COMPAT_43 */ 757 758 /* 759 * The dirty work is handled by kern_wait(). 760 */ 761 int 762 sys_wait4(struct thread *td, struct wait4_args *uap) 763 { 764 struct rusage ru, *rup; 765 int error, status; 766 767 if (uap->rusage != NULL) 768 rup = &ru; 769 else 770 rup = NULL; 771 error = kern_wait(td, uap->pid, &status, uap->options, rup); 772 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 773 error = copyout(&status, uap->status, sizeof(status)); 774 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 775 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 776 return (error); 777 } 778 779 int 780 sys_wait6(struct thread *td, struct wait6_args *uap) 781 { 782 struct __wrusage wru, *wrup; 783 siginfo_t si, *sip; 784 idtype_t idtype; 785 id_t id; 786 int error, status; 787 788 idtype = uap->idtype; 789 id = uap->id; 790 791 if (uap->wrusage != NULL) 792 wrup = &wru; 793 else 794 wrup = NULL; 795 796 if (uap->info != NULL) { 797 sip = &si; 798 bzero(sip, sizeof(*sip)); 799 } else 800 sip = NULL; 801 802 /* 803 * We expect all callers of wait6() to know about WEXITED and 804 * WTRAPPED. 805 */ 806 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 807 808 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 809 error = copyout(&status, uap->status, sizeof(status)); 810 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 811 error = copyout(&wru, uap->wrusage, sizeof(wru)); 812 if (uap->info != NULL && error == 0) 813 error = copyout(&si, uap->info, sizeof(si)); 814 return (error); 815 } 816 817 /* 818 * Reap the remains of a zombie process and optionally return status and 819 * rusage. Asserts and will release both the proctree_lock and the process 820 * lock as part of its work. 821 */ 822 void 823 proc_reap(struct thread *td, struct proc *p, int *status, int options) 824 { 825 struct proc *q, *t; 826 827 sx_assert(&proctree_lock, SA_XLOCKED); 828 PROC_LOCK_ASSERT(p, MA_OWNED); 829 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 830 831 mtx_spin_wait_unlocked(&p->p_slock); 832 833 q = td->td_proc; 834 835 if (status) 836 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 837 if (options & WNOWAIT) { 838 /* 839 * Only poll, returning the status. Caller does not wish to 840 * release the proc struct just yet. 841 */ 842 PROC_UNLOCK(p); 843 sx_xunlock(&proctree_lock); 844 return; 845 } 846 847 PROC_LOCK(q); 848 sigqueue_take(p->p_ksi); 849 PROC_UNLOCK(q); 850 851 /* 852 * If we got the child via a ptrace 'attach', we need to give it back 853 * to the old parent. 854 */ 855 if (p->p_oppid != p->p_pptr->p_pid) { 856 PROC_UNLOCK(p); 857 t = proc_realparent(p); 858 PROC_LOCK(t); 859 PROC_LOCK(p); 860 CTR2(KTR_PTRACE, 861 "wait: traced child %d moved back to parent %d", p->p_pid, 862 t->p_pid); 863 proc_reparent(p, t, false); 864 PROC_UNLOCK(p); 865 pksignal(t, SIGCHLD, p->p_ksi); 866 wakeup(t); 867 cv_broadcast(&p->p_pwait); 868 PROC_UNLOCK(t); 869 sx_xunlock(&proctree_lock); 870 return; 871 } 872 PROC_UNLOCK(p); 873 874 /* 875 * Remove other references to this process to ensure we have an 876 * exclusive reference. 877 */ 878 sx_xlock(PIDHASHLOCK(p->p_pid)); 879 LIST_REMOVE(p, p_hash); 880 sx_xunlock(PIDHASHLOCK(p->p_pid)); 881 LIST_REMOVE(p, p_sibling); 882 reaper_abandon_children(p, true); 883 reaper_clear(p); 884 PROC_LOCK(p); 885 proc_clear_orphan(p); 886 PROC_UNLOCK(p); 887 leavepgrp(p); 888 if (p->p_procdesc != NULL) 889 procdesc_reap(p); 890 sx_xunlock(&proctree_lock); 891 892 proc_id_clear(PROC_ID_PID, p->p_pid); 893 894 PROC_LOCK(p); 895 knlist_detach(p->p_klist); 896 p->p_klist = NULL; 897 PROC_UNLOCK(p); 898 899 /* 900 * Removal from allproc list and process group list paired with 901 * PROC_LOCK which was executed during that time should guarantee 902 * nothing can reach this process anymore. As such further locking 903 * is unnecessary. 904 */ 905 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 906 907 PROC_LOCK(q); 908 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 909 PROC_UNLOCK(q); 910 911 /* 912 * Decrement the count of procs running with this uid. 913 */ 914 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 915 916 /* 917 * Destroy resource accounting information associated with the process. 918 */ 919 #ifdef RACCT 920 if (racct_enable) { 921 PROC_LOCK(p); 922 racct_sub(p, RACCT_NPROC, 1); 923 PROC_UNLOCK(p); 924 } 925 #endif 926 racct_proc_exit(p); 927 928 /* 929 * Free credentials, arguments, and sigacts. 930 */ 931 proc_unset_cred(p); 932 pargs_drop(p->p_args); 933 p->p_args = NULL; 934 sigacts_free(p->p_sigacts); 935 p->p_sigacts = NULL; 936 937 /* 938 * Do any thread-system specific cleanups. 939 */ 940 thread_wait(p); 941 942 /* 943 * Give vm and machine-dependent layer a chance to free anything that 944 * cpu_exit couldn't release while still running in process context. 945 */ 946 vm_waitproc(p); 947 #ifdef MAC 948 mac_proc_destroy(p); 949 #endif 950 951 KASSERT(FIRST_THREAD_IN_PROC(p), 952 ("proc_reap: no residual thread!")); 953 uma_zfree(proc_zone, p); 954 atomic_add_int(&nprocs, -1); 955 } 956 957 static int 958 proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 959 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 960 int check_only) 961 { 962 struct rusage *rup; 963 964 sx_assert(&proctree_lock, SA_XLOCKED); 965 966 PROC_LOCK(p); 967 968 switch (idtype) { 969 case P_ALL: 970 if (p->p_procdesc == NULL || 971 (p->p_pptr == td->td_proc && 972 (p->p_flag & P_TRACED) != 0)) { 973 break; 974 } 975 976 PROC_UNLOCK(p); 977 return (0); 978 case P_PID: 979 if (p->p_pid != (pid_t)id) { 980 PROC_UNLOCK(p); 981 return (0); 982 } 983 break; 984 case P_PGID: 985 if (p->p_pgid != (pid_t)id) { 986 PROC_UNLOCK(p); 987 return (0); 988 } 989 break; 990 case P_SID: 991 if (p->p_session->s_sid != (pid_t)id) { 992 PROC_UNLOCK(p); 993 return (0); 994 } 995 break; 996 case P_UID: 997 if (p->p_ucred->cr_uid != (uid_t)id) { 998 PROC_UNLOCK(p); 999 return (0); 1000 } 1001 break; 1002 case P_GID: 1003 if (p->p_ucred->cr_gid != (gid_t)id) { 1004 PROC_UNLOCK(p); 1005 return (0); 1006 } 1007 break; 1008 case P_JAILID: 1009 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1010 PROC_UNLOCK(p); 1011 return (0); 1012 } 1013 break; 1014 /* 1015 * It seems that the thread structures get zeroed out 1016 * at process exit. This makes it impossible to 1017 * support P_SETID, P_CID or P_CPUID. 1018 */ 1019 default: 1020 PROC_UNLOCK(p); 1021 return (0); 1022 } 1023 1024 if (p_canwait(td, p)) { 1025 PROC_UNLOCK(p); 1026 return (0); 1027 } 1028 1029 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1030 PROC_UNLOCK(p); 1031 return (0); 1032 } 1033 1034 /* 1035 * This special case handles a kthread spawned by linux_clone 1036 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1037 * functions need to be able to distinguish between waiting 1038 * on a process and waiting on a thread. It is a thread if 1039 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1040 * signifies we want to wait for threads and not processes. 1041 */ 1042 if ((p->p_sigparent != SIGCHLD) ^ 1043 ((options & WLINUXCLONE) != 0)) { 1044 PROC_UNLOCK(p); 1045 return (0); 1046 } 1047 1048 if (siginfo != NULL) { 1049 bzero(siginfo, sizeof(*siginfo)); 1050 siginfo->si_errno = 0; 1051 1052 /* 1053 * SUSv4 requires that the si_signo value is always 1054 * SIGCHLD. Obey it despite the rfork(2) interface 1055 * allows to request other signal for child exit 1056 * notification. 1057 */ 1058 siginfo->si_signo = SIGCHLD; 1059 1060 /* 1061 * This is still a rough estimate. We will fix the 1062 * cases TRAPPED, STOPPED, and CONTINUED later. 1063 */ 1064 if (WCOREDUMP(p->p_xsig)) { 1065 siginfo->si_code = CLD_DUMPED; 1066 siginfo->si_status = WTERMSIG(p->p_xsig); 1067 } else if (WIFSIGNALED(p->p_xsig)) { 1068 siginfo->si_code = CLD_KILLED; 1069 siginfo->si_status = WTERMSIG(p->p_xsig); 1070 } else { 1071 siginfo->si_code = CLD_EXITED; 1072 siginfo->si_status = p->p_xexit; 1073 } 1074 1075 siginfo->si_pid = p->p_pid; 1076 siginfo->si_uid = p->p_ucred->cr_uid; 1077 1078 /* 1079 * The si_addr field would be useful additional 1080 * detail, but apparently the PC value may be lost 1081 * when we reach this point. bzero() above sets 1082 * siginfo->si_addr to NULL. 1083 */ 1084 } 1085 1086 /* 1087 * There should be no reason to limit resources usage info to 1088 * exited processes only. A snapshot about any resources used 1089 * by a stopped process may be exactly what is needed. 1090 */ 1091 if (wrusage != NULL) { 1092 rup = &wrusage->wru_self; 1093 *rup = p->p_ru; 1094 PROC_STATLOCK(p); 1095 calcru(p, &rup->ru_utime, &rup->ru_stime); 1096 PROC_STATUNLOCK(p); 1097 1098 rup = &wrusage->wru_children; 1099 *rup = p->p_stats->p_cru; 1100 calccru(p, &rup->ru_utime, &rup->ru_stime); 1101 } 1102 1103 if (p->p_state == PRS_ZOMBIE && !check_only) { 1104 proc_reap(td, p, status, options); 1105 return (-1); 1106 } 1107 return (1); 1108 } 1109 1110 int 1111 kern_wait(struct thread *td, pid_t pid, int *status, int options, 1112 struct rusage *rusage) 1113 { 1114 struct __wrusage wru, *wrup; 1115 idtype_t idtype; 1116 id_t id; 1117 int ret; 1118 1119 /* 1120 * Translate the special pid values into the (idtype, pid) 1121 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1122 * kern_wait6() on its own. 1123 */ 1124 if (pid == WAIT_ANY) { 1125 idtype = P_ALL; 1126 id = 0; 1127 } else if (pid < 0) { 1128 idtype = P_PGID; 1129 id = (id_t)-pid; 1130 } else { 1131 idtype = P_PID; 1132 id = (id_t)pid; 1133 } 1134 1135 if (rusage != NULL) 1136 wrup = &wru; 1137 else 1138 wrup = NULL; 1139 1140 /* 1141 * For backward compatibility we implicitly add flags WEXITED 1142 * and WTRAPPED here. 1143 */ 1144 options |= WEXITED | WTRAPPED; 1145 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1146 if (rusage != NULL) 1147 *rusage = wru.wru_self; 1148 return (ret); 1149 } 1150 1151 static void 1152 report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1153 int *status, int options, int si_code) 1154 { 1155 bool cont; 1156 1157 PROC_LOCK_ASSERT(p, MA_OWNED); 1158 sx_assert(&proctree_lock, SA_XLOCKED); 1159 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1160 si_code == CLD_CONTINUED); 1161 1162 cont = si_code == CLD_CONTINUED; 1163 if ((options & WNOWAIT) == 0) { 1164 if (cont) 1165 p->p_flag &= ~P_CONTINUED; 1166 else 1167 p->p_flag |= P_WAITED; 1168 PROC_LOCK(td->td_proc); 1169 sigqueue_take(p->p_ksi); 1170 PROC_UNLOCK(td->td_proc); 1171 } 1172 sx_xunlock(&proctree_lock); 1173 if (siginfo != NULL) { 1174 siginfo->si_code = si_code; 1175 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1176 } 1177 if (status != NULL) 1178 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1179 PROC_UNLOCK(p); 1180 td->td_retval[0] = p->p_pid; 1181 } 1182 1183 int 1184 kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1185 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1186 { 1187 struct proc *p, *q; 1188 pid_t pid; 1189 int error, nfound, ret; 1190 bool report; 1191 1192 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1193 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1194 AUDIT_ARG_VALUE(options); 1195 1196 q = td->td_proc; 1197 1198 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1199 PROC_LOCK(q); 1200 id = (id_t)q->p_pgid; 1201 PROC_UNLOCK(q); 1202 idtype = P_PGID; 1203 } 1204 1205 /* If we don't know the option, just return. */ 1206 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1207 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1208 return (EINVAL); 1209 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1210 /* 1211 * We will be unable to find any matching processes, 1212 * because there are no known events to look for. 1213 * Prefer to return error instead of blocking 1214 * indefinitely. 1215 */ 1216 return (EINVAL); 1217 } 1218 1219 loop: 1220 if (q->p_flag & P_STATCHILD) { 1221 PROC_LOCK(q); 1222 q->p_flag &= ~P_STATCHILD; 1223 PROC_UNLOCK(q); 1224 } 1225 sx_xlock(&proctree_lock); 1226 loop_locked: 1227 nfound = 0; 1228 LIST_FOREACH(p, &q->p_children, p_sibling) { 1229 pid = p->p_pid; 1230 ret = proc_to_reap(td, p, idtype, id, status, options, 1231 wrusage, siginfo, 0); 1232 if (ret == 0) 1233 continue; 1234 else if (ret != 1) { 1235 td->td_retval[0] = pid; 1236 return (0); 1237 } 1238 1239 nfound++; 1240 PROC_LOCK_ASSERT(p, MA_OWNED); 1241 1242 if ((options & WTRAPPED) != 0 && 1243 (p->p_flag & P_TRACED) != 0) { 1244 PROC_SLOCK(p); 1245 report = 1246 ((p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) && 1247 p->p_suspcount == p->p_numthreads && 1248 (p->p_flag & P_WAITED) == 0); 1249 PROC_SUNLOCK(p); 1250 if (report) { 1251 CTR4(KTR_PTRACE, 1252 "wait: returning trapped pid %d status %#x " 1253 "(xstat %d) xthread %d", 1254 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1255 p->p_xthread != NULL ? 1256 p->p_xthread->td_tid : -1); 1257 report_alive_proc(td, p, siginfo, status, 1258 options, CLD_TRAPPED); 1259 return (0); 1260 } 1261 } 1262 if ((options & WUNTRACED) != 0 && 1263 (p->p_flag & P_STOPPED_SIG) != 0) { 1264 PROC_SLOCK(p); 1265 report = (p->p_suspcount == p->p_numthreads && 1266 ((p->p_flag & P_WAITED) == 0)); 1267 PROC_SUNLOCK(p); 1268 if (report) { 1269 report_alive_proc(td, p, siginfo, status, 1270 options, CLD_STOPPED); 1271 return (0); 1272 } 1273 } 1274 if ((options & WCONTINUED) != 0 && 1275 (p->p_flag & P_CONTINUED) != 0) { 1276 report_alive_proc(td, p, siginfo, status, options, 1277 CLD_CONTINUED); 1278 return (0); 1279 } 1280 PROC_UNLOCK(p); 1281 } 1282 1283 /* 1284 * Look in the orphans list too, to allow the parent to 1285 * collect it's child exit status even if child is being 1286 * debugged. 1287 * 1288 * Debugger detaches from the parent upon successful 1289 * switch-over from parent to child. At this point due to 1290 * re-parenting the parent loses the child to debugger and a 1291 * wait4(2) call would report that it has no children to wait 1292 * for. By maintaining a list of orphans we allow the parent 1293 * to successfully wait until the child becomes a zombie. 1294 */ 1295 if (nfound == 0) { 1296 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1297 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1298 NULL, NULL, 1); 1299 if (ret != 0) { 1300 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1301 (int)td->td_retval[0])); 1302 PROC_UNLOCK(p); 1303 nfound++; 1304 break; 1305 } 1306 } 1307 } 1308 if (nfound == 0) { 1309 sx_xunlock(&proctree_lock); 1310 return (ECHILD); 1311 } 1312 if (options & WNOHANG) { 1313 sx_xunlock(&proctree_lock); 1314 td->td_retval[0] = 0; 1315 return (0); 1316 } 1317 PROC_LOCK(q); 1318 if (q->p_flag & P_STATCHILD) { 1319 q->p_flag &= ~P_STATCHILD; 1320 PROC_UNLOCK(q); 1321 goto loop_locked; 1322 } 1323 sx_xunlock(&proctree_lock); 1324 error = msleep(q, &q->p_mtx, PWAIT | PCATCH | PDROP, "wait", 0); 1325 if (error) 1326 return (error); 1327 goto loop; 1328 } 1329 1330 void 1331 proc_add_orphan(struct proc *child, struct proc *parent) 1332 { 1333 1334 sx_assert(&proctree_lock, SX_XLOCKED); 1335 KASSERT((child->p_flag & P_TRACED) != 0, 1336 ("proc_add_orphan: not traced")); 1337 1338 if (LIST_EMPTY(&parent->p_orphans)) { 1339 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1340 LIST_INSERT_HEAD(&parent->p_orphans, child, p_orphan); 1341 } else { 1342 LIST_INSERT_AFTER(LIST_FIRST(&parent->p_orphans), 1343 child, p_orphan); 1344 } 1345 child->p_treeflag |= P_TREE_ORPHANED; 1346 } 1347 1348 /* 1349 * Make process 'parent' the new parent of process 'child'. 1350 * Must be called with an exclusive hold of proctree lock. 1351 */ 1352 void 1353 proc_reparent(struct proc *child, struct proc *parent, bool set_oppid) 1354 { 1355 1356 sx_assert(&proctree_lock, SX_XLOCKED); 1357 PROC_LOCK_ASSERT(child, MA_OWNED); 1358 if (child->p_pptr == parent) 1359 return; 1360 1361 PROC_LOCK(child->p_pptr); 1362 sigqueue_take(child->p_ksi); 1363 PROC_UNLOCK(child->p_pptr); 1364 LIST_REMOVE(child, p_sibling); 1365 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1366 1367 proc_clear_orphan(child); 1368 if ((child->p_flag & P_TRACED) != 0) { 1369 proc_add_orphan(child, child->p_pptr); 1370 } 1371 1372 child->p_pptr = parent; 1373 if (set_oppid) 1374 child->p_oppid = parent->p_pid; 1375 } 1376