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: stable/11/sys/kern/kern_exit.c 346541 2019-04-22 11:17:37Z markj $"); 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> 93dtrace_execexit_func_t dtrace_fasttrap_exit; 94#endif 95 96SDT_PROVIDER_DECLARE(proc); 97SDT_PROBE_DEFINE1(proc, , , exit, "int"); 98 99/* Hook for NFS teardown procedure. */ 100void (*nlminfo_release_p)(struct proc *p); 101 102EVENTHANDLER_LIST_DECLARE(process_exit); 103 104struct proc * 105proc_realparent(struct proc *child) 106{ 107 struct proc *p, *parent; 108 109 sx_assert(&proctree_lock, SX_LOCKED); 110 if ((child->p_treeflag & P_TREE_ORPHANED) == 0) { 111 if (child->p_oppid == 0 || 112 child->p_pptr->p_pid == child->p_oppid) 113 parent = child->p_pptr; 114 else 115 parent = initproc; 116 return (parent); 117 } 118 for (p = child; (p->p_treeflag & P_TREE_FIRST_ORPHAN) == 0;) { 119 /* Cannot use LIST_PREV(), since the list head is not known. */ 120 p = __containerof(p->p_orphan.le_prev, struct proc, 121 p_orphan.le_next); 122 KASSERT((p->p_treeflag & P_TREE_ORPHANED) != 0, 123 ("missing P_ORPHAN %p", p)); 124 } 125 parent = __containerof(p->p_orphan.le_prev, struct proc, 126 p_orphans.lh_first); 127 return (parent); 128} 129 130void 131reaper_abandon_children(struct proc *p, bool exiting) 132{ 133 struct proc *p1, *p2, *ptmp; 134 135 sx_assert(&proctree_lock, SX_LOCKED); 136 KASSERT(p != initproc, ("reaper_abandon_children for initproc")); 137 if ((p->p_treeflag & P_TREE_REAPER) == 0) 138 return; 139 p1 = p->p_reaper; 140 LIST_FOREACH_SAFE(p2, &p->p_reaplist, p_reapsibling, ptmp) { 141 LIST_REMOVE(p2, p_reapsibling); 142 p2->p_reaper = p1; 143 p2->p_reapsubtree = p->p_reapsubtree; 144 LIST_INSERT_HEAD(&p1->p_reaplist, p2, p_reapsibling); 145 if (exiting && p2->p_pptr == p) { 146 PROC_LOCK(p2); 147 proc_reparent(p2, p1); 148 PROC_UNLOCK(p2); 149 } 150 } 151 KASSERT(LIST_EMPTY(&p->p_reaplist), ("p_reaplist not empty")); 152 p->p_treeflag &= ~P_TREE_REAPER; 153} 154 155static void 156clear_orphan(struct proc *p) 157{ 158 struct proc *p1; 159 160 sx_assert(&proctree_lock, SA_XLOCKED); 161 if ((p->p_treeflag & P_TREE_ORPHANED) == 0) 162 return; 163 if ((p->p_treeflag & P_TREE_FIRST_ORPHAN) != 0) { 164 p1 = LIST_NEXT(p, p_orphan); 165 if (p1 != NULL) 166 p1->p_treeflag |= P_TREE_FIRST_ORPHAN; 167 p->p_treeflag &= ~P_TREE_FIRST_ORPHAN; 168 } 169 LIST_REMOVE(p, p_orphan); 170 p->p_treeflag &= ~P_TREE_ORPHANED; 171} 172 173/* 174 * exit -- death of process. 175 */ 176void 177sys_sys_exit(struct thread *td, struct sys_exit_args *uap) 178{ 179 180 exit1(td, uap->rval, 0); 181 /* NOTREACHED */ 182} 183 184/* 185 * Exit: deallocate address space and other resources, change proc state to 186 * zombie, and unlink proc from allproc and parent's lists. Save exit status 187 * and rusage for wait(). Check for child processes and orphan them. 188 */ 189void 190exit1(struct thread *td, int rval, int signo) 191{ 192 struct proc *p, *nq, *q, *t; 193 struct thread *tdt; 194 ksiginfo_t *ksi, *ksi1; 195 196 mtx_assert(&Giant, MA_NOTOWNED); 197 KASSERT(rval == 0 || signo == 0, ("exit1 rv %d sig %d", rval, signo)); 198 199 p = td->td_proc; 200 /* 201 * XXX in case we're rebooting we just let init die in order to 202 * work around an unsolved stack overflow seen very late during 203 * shutdown on sparc64 when the gmirror worker process exists. 204 */ 205 if (p == initproc && rebooting == 0) { 206 printf("init died (signal %d, exit %d)\n", signo, rval); 207 panic("Going nowhere without my init!"); 208 } 209 210 /* 211 * Deref SU mp, since the thread does not return to userspace. 212 */ 213 td_softdep_cleanup(td); 214 215 /* 216 * MUST abort all other threads before proceeding past here. 217 */ 218 PROC_LOCK(p); 219 /* 220 * First check if some other thread or external request got 221 * here before us. If so, act appropriately: exit or suspend. 222 * We must ensure that stop requests are handled before we set 223 * P_WEXIT. 224 */ 225 thread_suspend_check(0); 226 while (p->p_flag & P_HADTHREADS) { 227 /* 228 * Kill off the other threads. This requires 229 * some co-operation from other parts of the kernel 230 * so it may not be instantaneous. With this state set 231 * any thread entering the kernel from userspace will 232 * thread_exit() in trap(). Any thread attempting to 233 * sleep will return immediately with EINTR or EWOULDBLOCK 234 * which will hopefully force them to back out to userland 235 * freeing resources as they go. Any thread attempting 236 * to return to userland will thread_exit() from userret(). 237 * thread_exit() will unsuspend us when the last of the 238 * other threads exits. 239 * If there is already a thread singler after resumption, 240 * calling thread_single will fail; in that case, we just 241 * re-check all suspension request, the thread should 242 * either be suspended there or exit. 243 */ 244 if (!thread_single(p, SINGLE_EXIT)) 245 /* 246 * All other activity in this process is now 247 * stopped. Threading support has been turned 248 * off. 249 */ 250 break; 251 /* 252 * Recheck for new stop or suspend requests which 253 * might appear while process lock was dropped in 254 * thread_single(). 255 */ 256 thread_suspend_check(0); 257 } 258 KASSERT(p->p_numthreads == 1, 259 ("exit1: proc %p exiting with %d threads", p, p->p_numthreads)); 260 racct_sub(p, RACCT_NTHR, 1); 261 262 /* Let event handler change exit status */ 263 p->p_xexit = rval; 264 p->p_xsig = signo; 265 266 /* 267 * Wakeup anyone in procfs' PIOCWAIT. They should have a hold 268 * on our vmspace, so we should block below until they have 269 * released their reference to us. Note that if they have 270 * requested S_EXIT stops we will block here until they ack 271 * via PIOCCONT. 272 */ 273 _STOPEVENT(p, S_EXIT, 0); 274 275 /* 276 * Ignore any pending request to stop due to a stop signal. 277 * Once P_WEXIT is set, future requests will be ignored as 278 * well. 279 */ 280 p->p_flag &= ~P_STOPPED_SIG; 281 KASSERT(!P_SHOULDSTOP(p), ("exiting process is stopped")); 282 283 /* 284 * Note that we are exiting and do another wakeup of anyone in 285 * PIOCWAIT in case they aren't listening for S_EXIT stops or 286 * decided to wait again after we told them we are exiting. 287 */ 288 p->p_flag |= P_WEXIT; 289 wakeup(&p->p_stype); 290 291 /* 292 * Wait for any processes that have a hold on our vmspace to 293 * release their reference. 294 */ 295 while (p->p_lock > 0) 296 msleep(&p->p_lock, &p->p_mtx, PWAIT, "exithold", 0); 297 298 PROC_UNLOCK(p); 299 /* Drain the limit callout while we don't have the proc locked */ 300 callout_drain(&p->p_limco); 301 302#ifdef AUDIT 303 /* 304 * The Sun BSM exit token contains two components: an exit status as 305 * passed to exit(), and a return value to indicate what sort of exit 306 * it was. The exit status is WEXITSTATUS(rv), but it's not clear 307 * what the return value is. 308 */ 309 AUDIT_ARG_EXIT(rval, 0); 310 AUDIT_SYSCALL_EXIT(0, td); 311#endif 312 313 /* Are we a task leader with peers? */ 314 if (p->p_peers != NULL && p == p->p_leader) { 315 mtx_lock(&ppeers_lock); 316 q = p->p_peers; 317 while (q != NULL) { 318 PROC_LOCK(q); 319 kern_psignal(q, SIGKILL); 320 PROC_UNLOCK(q); 321 q = q->p_peers; 322 } 323 while (p->p_peers != NULL) 324 msleep(p, &ppeers_lock, PWAIT, "exit1", 0); 325 mtx_unlock(&ppeers_lock); 326 } 327 328 /* 329 * Check if any loadable modules need anything done at process exit. 330 * E.g. SYSV IPC stuff. 331 * Event handler could change exit status. 332 * XXX what if one of these generates an error? 333 */ 334 EVENTHANDLER_DIRECT_INVOKE(process_exit, p); 335 336 /* 337 * If parent is waiting for us to exit or exec, 338 * P_PPWAIT is set; we will wakeup the parent below. 339 */ 340 PROC_LOCK(p); 341 stopprofclock(p); 342 p->p_flag &= ~(P_TRACED | P_PPWAIT | P_PPTRACE); 343 p->p_ptevents = 0; 344 345 /* 346 * Stop the real interval timer. If the handler is currently 347 * executing, prevent it from rearming itself and let it finish. 348 */ 349 if (timevalisset(&p->p_realtimer.it_value) && 350 _callout_stop_safe(&p->p_itcallout, CS_EXECUTING, NULL) == 0) { 351 timevalclear(&p->p_realtimer.it_interval); 352 msleep(&p->p_itcallout, &p->p_mtx, PWAIT, "ritwait", 0); 353 KASSERT(!timevalisset(&p->p_realtimer.it_value), 354 ("realtime timer is still armed")); 355 } 356 357 PROC_UNLOCK(p); 358 359 umtx_thread_exit(td); 360 361 /* 362 * Reset any sigio structures pointing to us as a result of 363 * F_SETOWN with our pid. 364 */ 365 funsetownlst(&p->p_sigiolst); 366 367 /* 368 * If this process has an nlminfo data area (for lockd), release it 369 */ 370 if (nlminfo_release_p != NULL && p->p_nlminfo != NULL) 371 (*nlminfo_release_p)(p); 372 373 /* 374 * Close open files and release open-file table. 375 * This may block! 376 */ 377 fdescfree(td); 378 379 /* 380 * If this thread tickled GEOM, we need to wait for the giggling to 381 * stop before we return to userland 382 */ 383 if (td->td_pflags & TDP_GEOM) 384 g_waitidle(); 385 386 /* 387 * Remove ourself from our leader's peer list and wake our leader. 388 */ 389 if (p->p_leader->p_peers != NULL) { 390 mtx_lock(&ppeers_lock); 391 if (p->p_leader->p_peers != NULL) { 392 q = p->p_leader; 393 while (q->p_peers != p) 394 q = q->p_peers; 395 q->p_peers = p->p_peers; 396 wakeup(p->p_leader); 397 } 398 mtx_unlock(&ppeers_lock); 399 } 400 401 vmspace_exit(td); 402 killjobc(); 403 (void)acct_process(td); 404 405#ifdef KTRACE 406 ktrprocexit(td); 407#endif 408 /* 409 * Release reference to text vnode 410 */ 411 if (p->p_textvp != NULL) { 412 vrele(p->p_textvp); 413 p->p_textvp = NULL; 414 } 415 416 /* 417 * Release our limits structure. 418 */ 419 lim_free(p->p_limit); 420 p->p_limit = NULL; 421 422 tidhash_remove(td); 423 424 /* 425 * Remove proc from allproc queue and pidhash chain. 426 * Place onto zombproc. Unlink from parent's child list. 427 */ 428 sx_xlock(&allproc_lock); 429 LIST_REMOVE(p, p_list); 430 LIST_INSERT_HEAD(&zombproc, p, p_list); 431 LIST_REMOVE(p, p_hash); 432 sx_xunlock(&allproc_lock); 433 434 /* 435 * Call machine-dependent code to release any 436 * machine-dependent resources other than the address space. 437 * The address space is released by "vmspace_exitfree(p)" in 438 * vm_waitproc(). 439 */ 440 cpu_exit(td); 441 442 WITNESS_WARN(WARN_PANIC, NULL, "process (pid %d) exiting", p->p_pid); 443 444 /* 445 * Reparent all children processes: 446 * - traced ones to the original parent (or init if we are that parent) 447 * - the rest to init 448 */ 449 sx_xlock(&proctree_lock); 450 q = LIST_FIRST(&p->p_children); 451 if (q != NULL) /* only need this if any child is S_ZOMB */ 452 wakeup(q->p_reaper); 453 for (; q != NULL; q = nq) { 454 nq = LIST_NEXT(q, p_sibling); 455 ksi = ksiginfo_alloc(TRUE); 456 PROC_LOCK(q); 457 q->p_sigparent = SIGCHLD; 458 459 if (!(q->p_flag & P_TRACED)) { 460 proc_reparent(q, q->p_reaper); 461 if (q->p_state == PRS_ZOMBIE) { 462 /* 463 * Inform reaper about the reparented 464 * zombie, since wait(2) has something 465 * new to report. Guarantee queueing 466 * of the SIGCHLD signal, similar to 467 * the _exit() behaviour, by providing 468 * our ksiginfo. Ksi is freed by the 469 * signal delivery. 470 */ 471 if (q->p_ksi == NULL) { 472 ksi1 = NULL; 473 } else { 474 ksiginfo_copy(q->p_ksi, ksi); 475 ksi->ksi_flags |= KSI_INS; 476 ksi1 = ksi; 477 ksi = NULL; 478 } 479 PROC_LOCK(q->p_reaper); 480 pksignal(q->p_reaper, SIGCHLD, ksi1); 481 PROC_UNLOCK(q->p_reaper); 482 } else if (q->p_pdeathsig > 0) { 483 /* 484 * The child asked to received a signal 485 * when we exit. 486 */ 487 kern_psignal(q, q->p_pdeathsig); 488 } 489 } else { 490 /* 491 * Traced processes are killed since their existence 492 * means someone is screwing up. 493 */ 494 t = proc_realparent(q); 495 if (t == p) { 496 proc_reparent(q, q->p_reaper); 497 } else { 498 PROC_LOCK(t); 499 proc_reparent(q, t); 500 PROC_UNLOCK(t); 501 } 502 /* 503 * Since q was found on our children list, the 504 * proc_reparent() call moved q to the orphan 505 * list due to present P_TRACED flag. Clear 506 * orphan link for q now while q is locked. 507 */ 508 clear_orphan(q); 509 q->p_flag &= ~(P_TRACED | P_STOPPED_TRACE); 510 q->p_flag2 &= ~P2_PTRACE_FSTP; 511 q->p_ptevents = 0; 512 FOREACH_THREAD_IN_PROC(q, tdt) { 513 tdt->td_dbgflags &= ~(TDB_SUSPEND | TDB_XSIG | 514 TDB_FSTP); 515 } 516 kern_psignal(q, SIGKILL); 517 } 518 PROC_UNLOCK(q); 519 if (ksi != NULL) 520 ksiginfo_free(ksi); 521 } 522 523 /* 524 * Also get rid of our orphans. 525 */ 526 while ((q = LIST_FIRST(&p->p_orphans)) != NULL) { 527 PROC_LOCK(q); 528 KASSERT(q->p_oppid == p->p_pid, 529 ("orphan %p of %p has unexpected oppid %d", q, p, 530 q->p_oppid)); 531 q->p_oppid = q->p_reaper->p_pid; 532 533 /* 534 * If we are the real parent of this process 535 * but it has been reparented to a debugger, then 536 * check if it asked for a signal when we exit. 537 */ 538 if (q->p_pdeathsig > 0) 539 kern_psignal(q, q->p_pdeathsig); 540 CTR2(KTR_PTRACE, "exit: pid %d, clearing orphan %d", p->p_pid, 541 q->p_pid); 542 clear_orphan(q); 543 PROC_UNLOCK(q); 544 } 545 546 /* Save exit status. */ 547 PROC_LOCK(p); 548 p->p_xthread = td; 549 550 /* Tell the prison that we are gone. */ 551 prison_proc_free(p->p_ucred->cr_prison); 552 553#ifdef KDTRACE_HOOKS 554 /* 555 * Tell the DTrace fasttrap provider about the exit if it 556 * has declared an interest. 557 */ 558 if (dtrace_fasttrap_exit) 559 dtrace_fasttrap_exit(p); 560#endif 561 562 /* 563 * Notify interested parties of our demise. 564 */ 565 KNOTE_LOCKED(p->p_klist, NOTE_EXIT); 566 567#ifdef KDTRACE_HOOKS 568 int reason = CLD_EXITED; 569 if (WCOREDUMP(signo)) 570 reason = CLD_DUMPED; 571 else if (WIFSIGNALED(signo)) 572 reason = CLD_KILLED; 573 SDT_PROBE1(proc, , , exit, reason); 574#endif 575 576 /* 577 * If this is a process with a descriptor, we may not need to deliver 578 * a signal to the parent. proctree_lock is held over 579 * procdesc_exit() to serialize concurrent calls to close() and 580 * exit(). 581 */ 582 if (p->p_procdesc == NULL || procdesc_exit(p)) { 583 /* 584 * Notify parent that we're gone. If parent has the 585 * PS_NOCLDWAIT flag set, or if the handler is set to SIG_IGN, 586 * notify process 1 instead (and hope it will handle this 587 * situation). 588 */ 589 PROC_LOCK(p->p_pptr); 590 mtx_lock(&p->p_pptr->p_sigacts->ps_mtx); 591 if (p->p_pptr->p_sigacts->ps_flag & 592 (PS_NOCLDWAIT | PS_CLDSIGIGN)) { 593 struct proc *pp; 594 595 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 596 pp = p->p_pptr; 597 PROC_UNLOCK(pp); 598 proc_reparent(p, p->p_reaper); 599 p->p_sigparent = SIGCHLD; 600 PROC_LOCK(p->p_pptr); 601 602 /* 603 * Notify parent, so in case he was wait(2)ing or 604 * executing waitpid(2) with our pid, he will 605 * continue. 606 */ 607 wakeup(pp); 608 } else 609 mtx_unlock(&p->p_pptr->p_sigacts->ps_mtx); 610 611 if (p->p_pptr == p->p_reaper || p->p_pptr == initproc) 612 childproc_exited(p); 613 else if (p->p_sigparent != 0) { 614 if (p->p_sigparent == SIGCHLD) 615 childproc_exited(p); 616 else /* LINUX thread */ 617 kern_psignal(p->p_pptr, p->p_sigparent); 618 } 619 } else 620 PROC_LOCK(p->p_pptr); 621 sx_xunlock(&proctree_lock); 622 623 /* 624 * The state PRS_ZOMBIE prevents other proesses from sending 625 * signal to the process, to avoid memory leak, we free memory 626 * for signal queue at the time when the state is set. 627 */ 628 sigqueue_flush(&p->p_sigqueue); 629 sigqueue_flush(&td->td_sigqueue); 630 631 /* 632 * We have to wait until after acquiring all locks before 633 * changing p_state. We need to avoid all possible context 634 * switches (including ones from blocking on a mutex) while 635 * marked as a zombie. We also have to set the zombie state 636 * before we release the parent process' proc lock to avoid 637 * a lost wakeup. So, we first call wakeup, then we grab the 638 * sched lock, update the state, and release the parent process' 639 * proc lock. 640 */ 641 wakeup(p->p_pptr); 642 cv_broadcast(&p->p_pwait); 643 sched_exit(p->p_pptr, td); 644 PROC_SLOCK(p); 645 p->p_state = PRS_ZOMBIE; 646 PROC_UNLOCK(p->p_pptr); 647 648 /* 649 * Save our children's rusage information in our exit rusage. 650 */ 651 PROC_STATLOCK(p); 652 ruadd(&p->p_ru, &p->p_rux, &p->p_stats->p_cru, &p->p_crux); 653 PROC_STATUNLOCK(p); 654 655 /* 656 * Make sure the scheduler takes this thread out of its tables etc. 657 * This will also release this thread's reference to the ucred. 658 * Other thread parts to release include pcb bits and such. 659 */ 660 thread_exit(); 661} 662 663 664#ifndef _SYS_SYSPROTO_H_ 665struct abort2_args { 666 char *why; 667 int nargs; 668 void **args; 669}; 670#endif 671 672int 673sys_abort2(struct thread *td, struct abort2_args *uap) 674{ 675 struct proc *p = td->td_proc; 676 struct sbuf *sb; 677 void *uargs[16]; 678 int error, i, sig; 679 680 /* 681 * Do it right now so we can log either proper call of abort2(), or 682 * note, that invalid argument was passed. 512 is big enough to 683 * handle 16 arguments' descriptions with additional comments. 684 */ 685 sb = sbuf_new(NULL, NULL, 512, SBUF_FIXEDLEN); 686 sbuf_clear(sb); 687 sbuf_printf(sb, "%s(pid %d uid %d) aborted: ", 688 p->p_comm, p->p_pid, td->td_ucred->cr_uid); 689 /* 690 * Since we can't return from abort2(), send SIGKILL in cases, where 691 * abort2() was called improperly 692 */ 693 sig = SIGKILL; 694 /* Prevent from DoSes from user-space. */ 695 if (uap->nargs < 0 || uap->nargs > 16) 696 goto out; 697 if (uap->nargs > 0) { 698 if (uap->args == NULL) 699 goto out; 700 error = copyin(uap->args, uargs, uap->nargs * sizeof(void *)); 701 if (error != 0) 702 goto out; 703 } 704 /* 705 * Limit size of 'reason' string to 128. Will fit even when 706 * maximal number of arguments was chosen to be logged. 707 */ 708 if (uap->why != NULL) { 709 error = sbuf_copyin(sb, uap->why, 128); 710 if (error < 0) 711 goto out; 712 } else { 713 sbuf_printf(sb, "(null)"); 714 } 715 if (uap->nargs > 0) { 716 sbuf_printf(sb, "("); 717 for (i = 0;i < uap->nargs; i++) 718 sbuf_printf(sb, "%s%p", i == 0 ? "" : ", ", uargs[i]); 719 sbuf_printf(sb, ")"); 720 } 721 /* 722 * Final stage: arguments were proper, string has been 723 * successfully copied from userspace, and copying pointers 724 * from user-space succeed. 725 */ 726 sig = SIGABRT; 727out: 728 if (sig == SIGKILL) { 729 sbuf_trim(sb); 730 sbuf_printf(sb, " (Reason text inaccessible)"); 731 } 732 sbuf_cat(sb, "\n"); 733 sbuf_finish(sb); 734 log(LOG_INFO, "%s", sbuf_data(sb)); 735 sbuf_delete(sb); 736 exit1(td, 0, sig); 737 return (0); 738} 739 740 741#ifdef COMPAT_43 742/* 743 * The dirty work is handled by kern_wait(). 744 */ 745int 746owait(struct thread *td, struct owait_args *uap __unused) 747{ 748 int error, status; 749 750 error = kern_wait(td, WAIT_ANY, &status, 0, NULL); 751 if (error == 0) 752 td->td_retval[1] = status; 753 return (error); 754} 755#endif /* COMPAT_43 */ 756 757/* 758 * The dirty work is handled by kern_wait(). 759 */ 760int 761sys_wait4(struct thread *td, struct wait4_args *uap) 762{ 763 struct rusage ru, *rup; 764 int error, status; 765 766 if (uap->rusage != NULL) 767 rup = &ru; 768 else 769 rup = NULL; 770 error = kern_wait(td, uap->pid, &status, uap->options, rup); 771 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 772 error = copyout(&status, uap->status, sizeof(status)); 773 if (uap->rusage != NULL && error == 0 && td->td_retval[0] != 0) 774 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 775 return (error); 776} 777 778int 779sys_wait6(struct thread *td, struct wait6_args *uap) 780{ 781 struct __wrusage wru, *wrup; 782 siginfo_t si, *sip; 783 idtype_t idtype; 784 id_t id; 785 int error, status; 786 787 idtype = uap->idtype; 788 id = uap->id; 789 790 if (uap->wrusage != NULL) 791 wrup = &wru; 792 else 793 wrup = NULL; 794 795 if (uap->info != NULL) { 796 sip = &si; 797 bzero(sip, sizeof(*sip)); 798 } else 799 sip = NULL; 800 801 /* 802 * We expect all callers of wait6() to know about WEXITED and 803 * WTRAPPED. 804 */ 805 error = kern_wait6(td, idtype, id, &status, uap->options, wrup, sip); 806 807 if (uap->status != NULL && error == 0 && td->td_retval[0] != 0) 808 error = copyout(&status, uap->status, sizeof(status)); 809 if (uap->wrusage != NULL && error == 0 && td->td_retval[0] != 0) 810 error = copyout(&wru, uap->wrusage, sizeof(wru)); 811 if (uap->info != NULL && error == 0) 812 error = copyout(&si, uap->info, sizeof(si)); 813 return (error); 814} 815 816/* 817 * Reap the remains of a zombie process and optionally return status and 818 * rusage. Asserts and will release both the proctree_lock and the process 819 * lock as part of its work. 820 */ 821void 822proc_reap(struct thread *td, struct proc *p, int *status, int options) 823{ 824 struct proc *q, *t; 825 826 sx_assert(&proctree_lock, SA_XLOCKED); 827 PROC_LOCK_ASSERT(p, MA_OWNED); 828 PROC_SLOCK_ASSERT(p, MA_OWNED); 829 KASSERT(p->p_state == PRS_ZOMBIE, ("proc_reap: !PRS_ZOMBIE")); 830 831 q = td->td_proc; 832 833 PROC_SUNLOCK(p); 834 if (status) 835 *status = KW_EXITCODE(p->p_xexit, p->p_xsig); 836 if (options & WNOWAIT) { 837 /* 838 * Only poll, returning the status. Caller does not wish to 839 * release the proc struct just yet. 840 */ 841 PROC_UNLOCK(p); 842 sx_xunlock(&proctree_lock); 843 return; 844 } 845 846 PROC_LOCK(q); 847 sigqueue_take(p->p_ksi); 848 PROC_UNLOCK(q); 849 850 /* 851 * If we got the child via a ptrace 'attach', we need to give it back 852 * to the old parent. 853 */ 854 if (p->p_oppid != 0 && p->p_oppid != p->p_pptr->p_pid) { 855 PROC_UNLOCK(p); 856 t = proc_realparent(p); 857 PROC_LOCK(t); 858 PROC_LOCK(p); 859 CTR2(KTR_PTRACE, 860 "wait: traced child %d moved back to parent %d", p->p_pid, 861 t->p_pid); 862 proc_reparent(p, t); 863 p->p_oppid = 0; 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 p->p_oppid = 0; 873 PROC_UNLOCK(p); 874 875 /* 876 * Remove other references to this process to ensure we have an 877 * exclusive reference. 878 */ 879 sx_xlock(&allproc_lock); 880 LIST_REMOVE(p, p_list); /* off zombproc */ 881 sx_xunlock(&allproc_lock); 882 LIST_REMOVE(p, p_sibling); 883 reaper_abandon_children(p, true); 884 LIST_REMOVE(p, p_reapsibling); 885 PROC_LOCK(p); 886 clear_orphan(p); 887 PROC_UNLOCK(p); 888 leavepgrp(p); 889 if (p->p_procdesc != NULL) 890 procdesc_reap(p); 891 sx_xunlock(&proctree_lock); 892 893 PROC_LOCK(p); 894 knlist_detach(p->p_klist); 895 p->p_klist = NULL; 896 PROC_UNLOCK(p); 897 898 /* 899 * Removal from allproc list and process group list paired with 900 * PROC_LOCK which was executed during that time should guarantee 901 * nothing can reach this process anymore. As such further locking 902 * is unnecessary. 903 */ 904 p->p_xexit = p->p_xsig = 0; /* XXX: why? */ 905 906 PROC_LOCK(q); 907 ruadd(&q->p_stats->p_cru, &q->p_crux, &p->p_ru, &p->p_rux); 908 PROC_UNLOCK(q); 909 910 /* 911 * Decrement the count of procs running with this uid. 912 */ 913 (void)chgproccnt(p->p_ucred->cr_ruidinfo, -1, 0); 914 915 /* 916 * Destroy resource accounting information associated with the process. 917 */ 918#ifdef RACCT 919 if (racct_enable) { 920 PROC_LOCK(p); 921 racct_sub(p, RACCT_NPROC, 1); 922 PROC_UNLOCK(p); 923 } 924#endif 925 racct_proc_exit(p); 926 927 /* 928 * Free credentials, arguments, and sigacts. 929 */ 930 crfree(p->p_ucred); 931 proc_set_cred(p, NULL); 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 * Free any domain policy that's still hiding around. 952 */ 953 vm_domain_policy_cleanup(&p->p_vm_dom_policy); 954 955 KASSERT(FIRST_THREAD_IN_PROC(p), 956 ("proc_reap: no residual thread!")); 957 uma_zfree(proc_zone, p); 958 atomic_add_int(&nprocs, -1); 959} 960 961static int 962proc_to_reap(struct thread *td, struct proc *p, idtype_t idtype, id_t id, 963 int *status, int options, struct __wrusage *wrusage, siginfo_t *siginfo, 964 int check_only) 965{ 966 struct rusage *rup; 967 968 sx_assert(&proctree_lock, SA_XLOCKED); 969 970 PROC_LOCK(p); 971 972 switch (idtype) { 973 case P_ALL: 974 if (p->p_procdesc != NULL) { 975 PROC_UNLOCK(p); 976 return (0); 977 } 978 break; 979 case P_PID: 980 if (p->p_pid != (pid_t)id) { 981 PROC_UNLOCK(p); 982 return (0); 983 } 984 break; 985 case P_PGID: 986 if (p->p_pgid != (pid_t)id) { 987 PROC_UNLOCK(p); 988 return (0); 989 } 990 break; 991 case P_SID: 992 if (p->p_session->s_sid != (pid_t)id) { 993 PROC_UNLOCK(p); 994 return (0); 995 } 996 break; 997 case P_UID: 998 if (p->p_ucred->cr_uid != (uid_t)id) { 999 PROC_UNLOCK(p); 1000 return (0); 1001 } 1002 break; 1003 case P_GID: 1004 if (p->p_ucred->cr_gid != (gid_t)id) { 1005 PROC_UNLOCK(p); 1006 return (0); 1007 } 1008 break; 1009 case P_JAILID: 1010 if (p->p_ucred->cr_prison->pr_id != (int)id) { 1011 PROC_UNLOCK(p); 1012 return (0); 1013 } 1014 break; 1015 /* 1016 * It seems that the thread structures get zeroed out 1017 * at process exit. This makes it impossible to 1018 * support P_SETID, P_CID or P_CPUID. 1019 */ 1020 default: 1021 PROC_UNLOCK(p); 1022 return (0); 1023 } 1024 1025 if (p_canwait(td, p)) { 1026 PROC_UNLOCK(p); 1027 return (0); 1028 } 1029 1030 if (((options & WEXITED) == 0) && (p->p_state == PRS_ZOMBIE)) { 1031 PROC_UNLOCK(p); 1032 return (0); 1033 } 1034 1035 /* 1036 * This special case handles a kthread spawned by linux_clone 1037 * (see linux_misc.c). The linux_wait4 and linux_waitpid 1038 * functions need to be able to distinguish between waiting 1039 * on a process and waiting on a thread. It is a thread if 1040 * p_sigparent is not SIGCHLD, and the WLINUXCLONE option 1041 * signifies we want to wait for threads and not processes. 1042 */ 1043 if ((p->p_sigparent != SIGCHLD) ^ 1044 ((options & WLINUXCLONE) != 0)) { 1045 PROC_UNLOCK(p); 1046 return (0); 1047 } 1048 1049 if (siginfo != NULL) { 1050 bzero(siginfo, sizeof(*siginfo)); 1051 siginfo->si_errno = 0; 1052 1053 /* 1054 * SUSv4 requires that the si_signo value is always 1055 * SIGCHLD. Obey it despite the rfork(2) interface 1056 * allows to request other signal for child exit 1057 * notification. 1058 */ 1059 siginfo->si_signo = SIGCHLD; 1060 1061 /* 1062 * This is still a rough estimate. We will fix the 1063 * cases TRAPPED, STOPPED, and CONTINUED later. 1064 */ 1065 if (WCOREDUMP(p->p_xsig)) { 1066 siginfo->si_code = CLD_DUMPED; 1067 siginfo->si_status = WTERMSIG(p->p_xsig); 1068 } else if (WIFSIGNALED(p->p_xsig)) { 1069 siginfo->si_code = CLD_KILLED; 1070 siginfo->si_status = WTERMSIG(p->p_xsig); 1071 } else { 1072 siginfo->si_code = CLD_EXITED; 1073 siginfo->si_status = p->p_xexit; 1074 } 1075 1076 siginfo->si_pid = p->p_pid; 1077 siginfo->si_uid = p->p_ucred->cr_uid; 1078 1079 /* 1080 * The si_addr field would be useful additional 1081 * detail, but apparently the PC value may be lost 1082 * when we reach this point. bzero() above sets 1083 * siginfo->si_addr to NULL. 1084 */ 1085 } 1086 1087 /* 1088 * There should be no reason to limit resources usage info to 1089 * exited processes only. A snapshot about any resources used 1090 * by a stopped process may be exactly what is needed. 1091 */ 1092 if (wrusage != NULL) { 1093 rup = &wrusage->wru_self; 1094 *rup = p->p_ru; 1095 PROC_STATLOCK(p); 1096 calcru(p, &rup->ru_utime, &rup->ru_stime); 1097 PROC_STATUNLOCK(p); 1098 1099 rup = &wrusage->wru_children; 1100 *rup = p->p_stats->p_cru; 1101 calccru(p, &rup->ru_utime, &rup->ru_stime); 1102 } 1103 1104 if (p->p_state == PRS_ZOMBIE && !check_only) { 1105 PROC_SLOCK(p); 1106 proc_reap(td, p, status, options); 1107 return (-1); 1108 } 1109 return (1); 1110} 1111 1112int 1113kern_wait(struct thread *td, pid_t pid, int *status, int options, 1114 struct rusage *rusage) 1115{ 1116 struct __wrusage wru, *wrup; 1117 idtype_t idtype; 1118 id_t id; 1119 int ret; 1120 1121 /* 1122 * Translate the special pid values into the (idtype, pid) 1123 * pair for kern_wait6. The WAIT_MYPGRP case is handled by 1124 * kern_wait6() on its own. 1125 */ 1126 if (pid == WAIT_ANY) { 1127 idtype = P_ALL; 1128 id = 0; 1129 } else if (pid < 0) { 1130 idtype = P_PGID; 1131 id = (id_t)-pid; 1132 } else { 1133 idtype = P_PID; 1134 id = (id_t)pid; 1135 } 1136 1137 if (rusage != NULL) 1138 wrup = &wru; 1139 else 1140 wrup = NULL; 1141 1142 /* 1143 * For backward compatibility we implicitly add flags WEXITED 1144 * and WTRAPPED here. 1145 */ 1146 options |= WEXITED | WTRAPPED; 1147 ret = kern_wait6(td, idtype, id, status, options, wrup, NULL); 1148 if (rusage != NULL) 1149 *rusage = wru.wru_self; 1150 return (ret); 1151} 1152 1153static void 1154report_alive_proc(struct thread *td, struct proc *p, siginfo_t *siginfo, 1155 int *status, int options, int si_code) 1156{ 1157 bool cont; 1158 1159 PROC_LOCK_ASSERT(p, MA_OWNED); 1160 sx_assert(&proctree_lock, SA_XLOCKED); 1161 MPASS(si_code == CLD_TRAPPED || si_code == CLD_STOPPED || 1162 si_code == CLD_CONTINUED); 1163 1164 cont = si_code == CLD_CONTINUED; 1165 if ((options & WNOWAIT) == 0) { 1166 if (cont) 1167 p->p_flag &= ~P_CONTINUED; 1168 else 1169 p->p_flag |= P_WAITED; 1170 PROC_LOCK(td->td_proc); 1171 sigqueue_take(p->p_ksi); 1172 PROC_UNLOCK(td->td_proc); 1173 } 1174 sx_xunlock(&proctree_lock); 1175 if (siginfo != NULL) { 1176 siginfo->si_code = si_code; 1177 siginfo->si_status = cont ? SIGCONT : p->p_xsig; 1178 } 1179 if (status != NULL) 1180 *status = cont ? SIGCONT : W_STOPCODE(p->p_xsig); 1181 PROC_UNLOCK(p); 1182 td->td_retval[0] = p->p_pid; 1183} 1184 1185int 1186kern_wait6(struct thread *td, idtype_t idtype, id_t id, int *status, 1187 int options, struct __wrusage *wrusage, siginfo_t *siginfo) 1188{ 1189 struct proc *p, *q; 1190 pid_t pid; 1191 int error, nfound, ret; 1192 1193 AUDIT_ARG_VALUE((int)idtype); /* XXX - This is likely wrong! */ 1194 AUDIT_ARG_PID((pid_t)id); /* XXX - This may be wrong! */ 1195 AUDIT_ARG_VALUE(options); 1196 1197 q = td->td_proc; 1198 1199 if ((pid_t)id == WAIT_MYPGRP && (idtype == P_PID || idtype == P_PGID)) { 1200 PROC_LOCK(q); 1201 id = (id_t)q->p_pgid; 1202 PROC_UNLOCK(q); 1203 idtype = P_PGID; 1204 } 1205 1206 /* If we don't know the option, just return. */ 1207 if ((options & ~(WUNTRACED | WNOHANG | WCONTINUED | WNOWAIT | 1208 WEXITED | WTRAPPED | WLINUXCLONE)) != 0) 1209 return (EINVAL); 1210 if ((options & (WEXITED | WUNTRACED | WCONTINUED | WTRAPPED)) == 0) { 1211 /* 1212 * We will be unable to find any matching processes, 1213 * because there are no known events to look for. 1214 * Prefer to return error instead of blocking 1215 * indefinitely. 1216 */ 1217 return (EINVAL); 1218 } 1219 1220loop: 1221 if (q->p_flag & P_STATCHILD) { 1222 PROC_LOCK(q); 1223 q->p_flag &= ~P_STATCHILD; 1224 PROC_UNLOCK(q); 1225 } 1226 nfound = 0; 1227 sx_xlock(&proctree_lock); 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 nfound++; 1236 else { 1237 td->td_retval[0] = pid; 1238 return (0); 1239 } 1240 1241 PROC_LOCK_ASSERT(p, MA_OWNED); 1242 1243 if ((options & (WTRAPPED | WUNTRACED)) != 0) 1244 PROC_SLOCK(p); 1245 1246 if ((options & WTRAPPED) != 0 && 1247 (p->p_flag & P_TRACED) != 0 && 1248 (p->p_flag & (P_STOPPED_TRACE | P_STOPPED_SIG)) != 0 && 1249 p->p_suspcount == p->p_numthreads && 1250 (p->p_flag & P_WAITED) == 0) { 1251 PROC_SUNLOCK(p); 1252 CTR4(KTR_PTRACE, 1253 "wait: returning trapped pid %d status %#x " 1254 "(xstat %d) xthread %d", 1255 p->p_pid, W_STOPCODE(p->p_xsig), p->p_xsig, 1256 p->p_xthread != NULL ? 1257 p->p_xthread->td_tid : -1); 1258 report_alive_proc(td, p, siginfo, status, options, 1259 CLD_TRAPPED); 1260 return (0); 1261 } 1262 if ((options & WUNTRACED) != 0 && 1263 (p->p_flag & P_STOPPED_SIG) != 0 && 1264 p->p_suspcount == p->p_numthreads && 1265 (p->p_flag & P_WAITED) == 0) { 1266 PROC_SUNLOCK(p); 1267 report_alive_proc(td, p, siginfo, status, options, 1268 CLD_STOPPED); 1269 return (0); 1270 } 1271 if ((options & (WTRAPPED | WUNTRACED)) != 0) 1272 PROC_SUNLOCK(p); 1273 if ((options & WCONTINUED) != 0 && 1274 (p->p_flag & P_CONTINUED) != 0) { 1275 report_alive_proc(td, p, siginfo, status, options, 1276 CLD_CONTINUED); 1277 return (0); 1278 } 1279 PROC_UNLOCK(p); 1280 } 1281 1282 /* 1283 * Look in the orphans list too, to allow the parent to 1284 * collect it's child exit status even if child is being 1285 * debugged. 1286 * 1287 * Debugger detaches from the parent upon successful 1288 * switch-over from parent to child. At this point due to 1289 * re-parenting the parent loses the child to debugger and a 1290 * wait4(2) call would report that it has no children to wait 1291 * for. By maintaining a list of orphans we allow the parent 1292 * to successfully wait until the child becomes a zombie. 1293 */ 1294 if (nfound == 0) { 1295 LIST_FOREACH(p, &q->p_orphans, p_orphan) { 1296 ret = proc_to_reap(td, p, idtype, id, NULL, options, 1297 NULL, NULL, 1); 1298 if (ret != 0) { 1299 KASSERT(ret != -1, ("reaped an orphan (pid %d)", 1300 (int)td->td_retval[0])); 1301 PROC_UNLOCK(p); 1302 nfound++; 1303 break; 1304 } 1305 } 1306 } 1307 if (nfound == 0) { 1308 sx_xunlock(&proctree_lock); 1309 return (ECHILD); 1310 } 1311 if (options & WNOHANG) { 1312 sx_xunlock(&proctree_lock); 1313 td->td_retval[0] = 0; 1314 return (0); 1315 } 1316 PROC_LOCK(q); 1317 sx_xunlock(&proctree_lock); 1318 if (q->p_flag & P_STATCHILD) { 1319 q->p_flag &= ~P_STATCHILD; 1320 error = 0; 1321 } else 1322 error = msleep(q, &q->p_mtx, PWAIT | PCATCH, "wait", 0); 1323 PROC_UNLOCK(q); 1324 if (error) 1325 return (error); 1326 goto loop; 1327} 1328 1329/* 1330 * Make process 'parent' the new parent of process 'child'. 1331 * Must be called with an exclusive hold of proctree lock. 1332 */ 1333void 1334proc_reparent(struct proc *child, struct proc *parent) 1335{ 1336 1337 sx_assert(&proctree_lock, SX_XLOCKED); 1338 PROC_LOCK_ASSERT(child, MA_OWNED); 1339 if (child->p_pptr == parent) 1340 return; 1341 1342 PROC_LOCK(child->p_pptr); 1343 sigqueue_take(child->p_ksi); 1344 PROC_UNLOCK(child->p_pptr); 1345 LIST_REMOVE(child, p_sibling); 1346 LIST_INSERT_HEAD(&parent->p_children, child, p_sibling); 1347 1348 clear_orphan(child); 1349 if (child->p_flag & P_TRACED) { 1350 if (LIST_EMPTY(&child->p_pptr->p_orphans)) { 1351 child->p_treeflag |= P_TREE_FIRST_ORPHAN; 1352 LIST_INSERT_HEAD(&child->p_pptr->p_orphans, child, 1353 p_orphan); 1354 } else { 1355 LIST_INSERT_AFTER(LIST_FIRST(&child->p_pptr->p_orphans), 1356 child, p_orphan); 1357 } 1358 child->p_treeflag |= P_TREE_ORPHANED; 1359 } 1360 1361 child->p_pptr = parent; 1362} 1363